Adding upstream version 14.2.21.upstream/14.2.21upstream

Signed-off-by: Daniel Baumann <daniel.baumann@progress-linux.org>

80 files changed, 35248 insertions, 0 deletions

diff --git a/src/cls/CMakeLists.txt b/src/cls/CMakeLists.txt
new file mode 100644
index 00000000..d62273d4
--- /dev/null
+++ b/src/cls/CMakeLists.txt
@@ -0,0 +1,264 @@
+## Rados object classes
+
+set(cls_dir ${CMAKE_INSTALL_LIBDIR}/rados-classes)
+
+# cls_sdk
+add_library(cls_sdk SHARED sdk/cls_sdk.cc)
+set_target_properties(cls_sdk PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_sdk DESTINATION ${cls_dir})
+
+# cls_hello
+set(cls_hello_srcs hello/cls_hello.cc)
+add_library(cls_hello SHARED ${cls_hello_srcs})
+set_target_properties(cls_hello PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_hello DESTINATION ${cls_dir})
+
+# cls_numops
+set(cls_numops_srcs numops/cls_numops.cc)
+add_library(cls_numops SHARED ${cls_numops_srcs})
+set_target_properties(cls_numops PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_numops DESTINATION ${cls_dir})
+
+set(cls_numops_client_srcs numops/cls_numops_client.cc)
+add_library(cls_numops_client STATIC ${cls_numops_client_srcs})
+
+
+# cls_rbd
+if (WITH_RBD)
+  set(cls_rbd_srcs rbd/cls_rbd.cc rbd/cls_rbd_types.cc)
+  add_library(cls_rbd SHARED ${cls_rbd_srcs})
+  set_target_properties(cls_rbd PROPERTIES
+    VERSION "1.0.0"
+    SOVERSION "1"
+    INSTALL_RPATH ""
+    CXX_VISIBILITY_PRESET hidden)
+  install(TARGETS cls_rbd DESTINATION ${cls_dir})
+
+  set(cls_rbd_client_srcs rbd/cls_rbd_client.cc rbd/cls_rbd_types.cc)
+  add_library(cls_rbd_client STATIC ${cls_rbd_client_srcs})
+  target_link_libraries(cls_rbd_client cls_lock_client)
+
+endif (WITH_RBD)
+
+# cls_lock
+set(cls_lock_srcs lock/cls_lock.cc)
+add_library(cls_lock SHARED ${cls_lock_srcs})
+set_target_properties(cls_lock PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_lock DESTINATION ${cls_dir})
+
+set(cls_lock_client_srcs
+  lock/cls_lock_client.cc
+  lock/cls_lock_types.cc
+  lock/cls_lock_ops.cc)
+add_library(cls_lock_client STATIC ${cls_lock_client_srcs})
+
+
+# cls_otp
+if (WITH_RADOSGW)
+  set(cls_otp_srcs otp/cls_otp.cc)
+  add_library(cls_otp SHARED ${cls_otp_srcs})
+  target_link_libraries(cls_otp OATH::OATH)
+  set_target_properties(cls_otp PROPERTIES
+    VERSION "1.0.0"
+    SOVERSION "1"
+    INSTALL_RPATH ""
+    CXX_VISIBILITY_PRESET hidden)
+  install(TARGETS cls_otp DESTINATION ${cls_dir})
+
+  set(cls_otp_client_srcs
+    otp/cls_otp_client.cc
+    otp/cls_otp_types.cc
+    )
+  add_library(cls_otp_client STATIC ${cls_otp_client_srcs})
+endif (WITH_RADOSGW)
+
+# cls_refcount
+set(cls_refcount_srcs
+  refcount/cls_refcount.cc
+  refcount/cls_refcount_ops.cc
+  ${CMAKE_SOURCE_DIR}/src/common/ceph_json.cc)
+add_library(cls_refcount SHARED ${cls_refcount_srcs})
+target_link_libraries(cls_refcount json_spirit)
+set_target_properties(cls_refcount PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_refcount DESTINATION ${cls_dir})
+
+set(cls_refcount_client_srcs
+  refcount/cls_refcount_client.cc
+  refcount/cls_refcount_ops.cc)
+add_library(cls_refcount_client STATIC ${cls_refcount_client_srcs})
+
+
+# cls_version
+set(cls_version_srcs version/cls_version.cc)
+add_library(cls_version SHARED ${cls_version_srcs})
+set_target_properties(cls_version PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_version DESTINATION ${cls_dir})
+
+set(cls_version_client_srcs
+  version/cls_version_client.cc
+  version/cls_version_types.cc)
+add_library(cls_version_client STATIC ${cls_version_client_srcs})
+
+
+# cls_log
+set(cls_log_srcs log/cls_log.cc)
+add_library(cls_log SHARED ${cls_log_srcs})
+set_target_properties(cls_log PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_log DESTINATION ${cls_dir})
+
+set(cls_log_client_srcs log/cls_log_client.cc)
+add_library(cls_log_client STATIC ${cls_log_client_srcs})
+
+
+# cls_timeindex
+set(cls_timeindex_srcs timeindex/cls_timeindex.cc)
+add_library(cls_timeindex SHARED ${cls_timeindex_srcs})
+set_target_properties(cls_timeindex PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_timeindex DESTINATION ${cls_dir})
+
+set(cls_timeindex_client_srcs timeindex/cls_timeindex_client.cc)
+add_library(cls_timeindex_client STATIC ${cls_timeindex_client_srcs})
+
+
+# cls_user
+set(cls_user_srcs user/cls_user.cc)
+add_library(cls_user SHARED ${cls_user_srcs})
+set_target_properties(cls_user PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_user DESTINATION ${cls_dir})
+
+set(cls_user_client_srcs
+  user/cls_user_client.cc
+  user/cls_user_types.cc
+  user/cls_user_ops.cc)
+add_library(cls_user_client STATIC ${cls_user_client_srcs})
+
+
+# cls_journal
+set(cls_journal_srcs
+  journal/cls_journal.cc
+  journal/cls_journal_types.cc)
+add_library(cls_journal SHARED ${cls_journal_srcs})
+set_target_properties(cls_journal PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_journal DESTINATION ${cls_dir})
+
+set(cls_journal_client_srcs
+  journal/cls_journal_client.cc
+  journal/cls_journal_types.cc)
+add_library(cls_journal_client STATIC ${cls_journal_client_srcs})
+
+
+# cls_rgw
+if (WITH_RADOSGW)
+  set(cls_rgw_srcs
+    rgw/cls_rgw.cc
+    rgw/cls_rgw_ops.cc
+    rgw/cls_rgw_types.cc
+    ${CMAKE_SOURCE_DIR}/src/common/ceph_json.cc)
+  add_library(cls_rgw SHARED ${cls_rgw_srcs})
+  target_link_libraries(cls_rgw json_spirit)
+  set_target_properties(cls_rgw PROPERTIES
+    VERSION "1.0.0"
+    SOVERSION "1"
+    INSTALL_RPATH ""
+    CXX_VISIBILITY_PRESET hidden)
+  install(TARGETS cls_rgw DESTINATION ${cls_dir})
+
+  set(cls_rgw_client_srcs
+    rgw/cls_rgw_client.cc
+    rgw/cls_rgw_types.cc
+    rgw/cls_rgw_ops.cc)
+  add_library(cls_rgw_client STATIC ${cls_rgw_client_srcs})
+
+endif (WITH_RADOSGW)
+
+# cls_cephfs
+if (WITH_CEPHFS)
+  set(cls_cephfs_srcs
+    cephfs/cls_cephfs.cc)
+  add_library(cls_cephfs SHARED ${cls_cephfs_srcs})
+  set_target_properties(cls_cephfs PROPERTIES
+    VERSION "1.0.0"
+    SOVERSION "1"
+    INSTALL_RPATH ""
+    CXX_VISIBILITY_PRESET hidden)
+  install(TARGETS cls_cephfs DESTINATION ${cls_dir})
+
+  set(cls_cephfs_client_srcs
+    cephfs/cls_cephfs_client.cc)
+  add_library(cls_cephfs_client STATIC ${cls_cephfs_client_srcs})
+
+endif (WITH_CEPHFS)
+
+# cls_lua
+set(cls_lua_srcs
+    lua/cls_lua.cc
+    lua/lua_bufferlist.cc)
+add_library(cls_lua SHARED ${cls_lua_srcs})
+set_target_properties(cls_lua PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_lua DESTINATION ${cls_dir})
+target_link_libraries(cls_lua
+    liblua
+    json_spirit)
+
+set(cls_lua_client_srcs
+    lua/cls_lua_client.cc)
+add_library(cls_lua_client STATIC ${cls_lua_client_srcs})
+
+# cls_cas
+set(cls_cas_srcs 
+  cas/cls_cas.cc)
+add_library(cls_cas SHARED ${cls_cas_srcs})
+set_target_properties(cls_cas PROPERTIES
+  VERSION "1.0.0"
+  SOVERSION "1"
+  INSTALL_RPATH ""
+  CXX_VISIBILITY_PRESET hidden)
+install(TARGETS cls_cas DESTINATION ${cls_dir})
+
+set(cls_cas_client_srcs
+  cas/cls_cas_client.cc)
+add_library(cls_cas_client STATIC ${cls_cas_client_srcs})
diff --git a/src/cls/cas/cls_cas.cc b/src/cls/cas/cls_cas.cc
new file mode 100644
index 00000000..c6a7b9b5
--- /dev/null
+++ b/src/cls/cas/cls_cas.cc
@@ -0,0 +1,265 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "objclass/objclass.h"
+#include "cls_cas_ops.h"
+
+#include "include/compat.h"
+#include "osd/osd_types.h"
+
+CLS_VER(1,0)
+CLS_NAME(cas)
+
+struct chunk_obj_refcount;
+
+static int chunk_read_refcount(cls_method_context_t hctx, chunk_obj_refcount *objr)
+{
+  bufferlist bl;
+  objr->refs.clear();
+  int ret = cls_cxx_getxattr(hctx, CHUNK_REFCOUNT_ATTR, &bl);
+  if (ret == -ENODATA) {
+    return 0;
+  }
+  if (ret < 0)
+    return ret;
+
+  try {
+    auto iter = bl.cbegin();
+    decode(*objr, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: chunk_read_refcount(): failed to decode refcount entry\n");
+    return -EIO;
+  }
+
+  return 0;
+}
+
+static int chunk_set_refcount(cls_method_context_t hctx, const struct chunk_obj_refcount& objr)
+{
+  bufferlist bl;
+
+  encode(objr, bl);
+
+  int ret = cls_cxx_setxattr(hctx, CHUNK_REFCOUNT_ATTR, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_chunk_refcount_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_chunk_refcount_get_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rc_refcount_get(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  chunk_obj_refcount objr;
+  int ret = chunk_read_refcount(hctx, &objr);
+  if (ret < 0)
+    return ret;
+
+  CLS_LOG(10, "cls_rc_chunk_refcount_get() oid=%s\n", op.source.oid.name.c_str());
+
+  objr.refs.insert(op.source);
+
+  ret = chunk_set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_chunk_refcount_put(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_chunk_refcount_put_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rc_chunk_refcount_put(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  chunk_obj_refcount objr;
+  int ret = chunk_read_refcount(hctx, &objr);
+  if (ret < 0)
+    return ret;
+
+  if (objr.refs.empty()) {// shouldn't happen!
+    CLS_LOG(0, "ERROR: cls_rc_chunk_refcount_put() was called without any references!\n");
+    return -EINVAL;
+  }
+
+  CLS_LOG(10, "cls_rc_chunk_refcount_put() oid=%s\n", op.source.oid.name.c_str());
+
+  bool found = false;
+  for (auto &p : objr.refs) {
+    if (p == op.source) {
+      found = true;
+      break;
+    }
+  }
+
+  if (!found) {
+    return 0;
+  }
+
+  auto p = objr.refs.find(op.source);
+  objr.refs.erase(p);
+
+  if (objr.refs.empty()) {
+    return cls_cxx_remove(hctx);
+  }
+
+  ret = chunk_set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_chunk_refcount_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_chunk_refcount_set_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_chunk_refcount_set(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  if (!op.refs.size()) {
+    return cls_cxx_remove(hctx);
+  }
+
+  chunk_obj_refcount objr;
+  objr.refs = op.refs;
+
+  int ret = chunk_set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_chunk_refcount_read(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  chunk_obj_refcount objr;
+
+  cls_chunk_refcount_read_ret read_ret;
+  int ret = chunk_read_refcount(hctx, &objr);
+  if (ret < 0)
+    return ret;
+
+  for (auto &p : objr.refs) {
+    read_ret.refs.insert(p);
+  }
+
+  encode(read_ret, *out);
+
+  return 0;
+}
+
+static int cls_rc_write_or_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+  hobject_t src_obj; 
+  bufferlist indata, outdata;
+  ceph_osd_op op;
+  try {
+    decode (op, in_iter);
+    decode(src_obj, in_iter);
+    in_iter.copy(op.extent.length, indata);
+  }
+  catch (buffer::error& e) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(10, " offset: %llu length: %llu \n",
+	  static_cast<long long unsigned>(op.extent.offset),
+	  static_cast<long long unsigned>(op.extent.length));
+  chunk_obj_refcount objr;
+  int ret = chunk_read_refcount(hctx, &objr);
+  if (ret == -ENOENT) {
+    objr.refs.insert(src_obj);
+    bufferlist set_bl;
+    encode(objr, set_bl);
+    ret = cls_cxx_chunk_write_and_set(hctx, op.extent.offset, op.extent.length, &indata, op.flags,
+				      &set_bl, set_bl.length());
+    if (ret < 0)
+      return ret;
+
+    return 0;
+  }
+
+  objr.refs.insert(src_obj);
+  ret = chunk_set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+
+static int cls_rc_has_chunk(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+  string fp_oid;
+  bufferlist indata, outdata;
+  try {
+    decode (fp_oid, in_iter);
+  }
+  catch (buffer::error& e) {
+    return -EINVAL;
+  }
+  CLS_LOG(10, " fp_oid: %s \n", fp_oid.c_str());
+
+  bool ret = cls_has_chunk(hctx, fp_oid);
+  if (ret) {
+    return 0;
+  }
+  return -ENOENT;
+}
+
+CLS_INIT(cas)
+{
+  CLS_LOG(1, "Loaded cas class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_cas_write_or_get;
+  cls_method_handle_t h_chunk_refcount_get;
+  cls_method_handle_t h_chunk_refcount_put;
+  cls_method_handle_t h_chunk_refcount_set;
+  cls_method_handle_t h_chunk_refcount_read;
+  cls_method_handle_t h_chunk_has_chunk;
+
+  cls_register("cas", &h_class);
+
+  /* chunk refcount */
+  cls_register_cxx_method(h_class, "chunk_get", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_chunk_refcount_get, 
+			  &h_chunk_refcount_get);
+  cls_register_cxx_method(h_class, "chunk_put", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_chunk_refcount_put, 
+			  &h_chunk_refcount_put);
+  cls_register_cxx_method(h_class, "chunk_set", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_chunk_refcount_set, 
+			  &h_chunk_refcount_set);
+  cls_register_cxx_method(h_class, "chunk_read", CLS_METHOD_RD, cls_rc_chunk_refcount_read, 
+			  &h_chunk_refcount_read);
+  cls_register_cxx_method(h_class, "cas_write_or_get", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_write_or_get, 
+			  &h_cas_write_or_get);
+  cls_register_cxx_method(h_class, "has_chunk", CLS_METHOD_RD, cls_rc_has_chunk, 
+	      &h_chunk_has_chunk);
+
+  return;
+}
+
diff --git a/src/cls/cas/cls_cas_client.cc b/src/cls/cas/cls_cas_client.cc
new file mode 100644
index 00000000..b041641f
--- /dev/null
+++ b/src/cls/cas/cls_cas_client.cc
@@ -0,0 +1,62 @@
+#include <errno.h>
+
+#include "cls/cas/cls_cas_client.h"
+#include "cls/cas/cls_cas_ops.h"
+#include "include/rados/librados.hpp"
+
+using namespace librados;
+
+void cls_chunk_refcount_get(librados::ObjectWriteOperation& op, const hobject_t& soid)
+{
+  bufferlist in;
+  cls_chunk_refcount_get_op call;
+  call.source = soid;
+  encode(call, in);
+  op.exec("cas", "chunk_get", in);
+}
+
+void cls_chunk_refcount_put(librados::ObjectWriteOperation& op, const hobject_t& soid)
+{
+  bufferlist in;
+  cls_chunk_refcount_put_op call;
+  call.source = soid;
+  encode(call, in);
+  op.exec("cas", "chunk_put", in);
+}
+
+void cls_chunk_refcount_set(librados::ObjectWriteOperation& op, set<hobject_t>& refs)
+{
+  bufferlist in;
+  cls_chunk_refcount_set_op call;
+  call.refs = refs;
+  encode(call, in);
+  op.exec("cas", "chunk_set", in);
+}
+
+int cls_chunk_refcount_read(librados::IoCtx& io_ctx, string& oid, set<hobject_t> *refs)
+{
+  bufferlist in, out;
+  int r = io_ctx.exec(oid, "cas", "chunk_read", in, out);
+  if (r < 0)
+    return r;
+
+  cls_chunk_refcount_read_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  *refs = ret.refs;
+
+  return r;
+}
+
+int cls_chunk_has_chunk(librados::IoCtx& io_ctx, string& oid, string& fp_oid)
+{
+  bufferlist in, out;
+  encode(fp_oid, in);
+  int r = io_ctx.exec(oid, "cas", "has_chunk", in, out);
+  return r;
+}
diff --git a/src/cls/cas/cls_cas_client.h b/src/cls/cas/cls_cas_client.h
new file mode 100644
index 00000000..323be9eb
--- /dev/null
+++ b/src/cls/cas/cls_cas_client.h
@@ -0,0 +1,13 @@
+#ifndef CEPH_CLS_CAS_CLIENT_H
+#define CEPH_CLS_CAS_CLIENT_H
+
+#include "include/types.h"
+#include "include/rados/librados_fwd.hpp"
+#include "common/hobject.h"
+
+void cls_chunk_refcount_get(librados::ObjectWriteOperation& op, const hobject_t& soid);
+void cls_chunk_refcount_put(librados::ObjectWriteOperation& op, const hobject_t& soid);
+void cls_chunk_refcount_set(librados::ObjectWriteOperation& op, set<hobject_t>& refs);
+int cls_chunk_refcount_read(librados::IoCtx& io_ctx, string& oid, set<hobject_t> *refs);
+int cls_chunk_has_chunk(librados::IoCtx& io_ctx, string& oid, string& fp_oid);
+#endif
diff --git a/src/cls/cas/cls_cas_ops.h b/src/cls/cas/cls_cas_ops.h
new file mode 100644
index 00000000..35bc90df
--- /dev/null
+++ b/src/cls/cas/cls_cas_ops.h
@@ -0,0 +1,142 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_CAS_OPS_H
+#define CEPH_CLS_CAS_OPS_H
+
+#include "include/types.h"
+#include "common/hobject.h"
+
+#define CHUNK_REFCOUNT_ATTR "chunk_refcount"
+
+struct cls_chunk_refcount_get_op {
+  hobject_t source;
+
+  cls_chunk_refcount_get_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(source, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(source, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_chunk_refcount_get_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_chunk_refcount_get_op)
+
+struct cls_chunk_refcount_put_op {
+  hobject_t source;
+
+  cls_chunk_refcount_put_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(source, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(source, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_chunk_refcount_put_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_chunk_refcount_put_op)
+
+struct cls_chunk_refcount_set_op {
+  set<hobject_t> refs;
+
+  cls_chunk_refcount_set_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(refs, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_chunk_refcount_set_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_chunk_refcount_set_op)
+
+struct cls_chunk_refcount_read_ret {
+  set<hobject_t> refs;
+
+  cls_chunk_refcount_read_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(refs, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_chunk_refcount_read_ret*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_chunk_refcount_read_ret)
+
+struct chunk_obj_refcount {
+  set<hobject_t> refs;
+
+  chunk_obj_refcount() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(refs, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(chunk_obj_refcount)
+
+struct obj_refcount {
+  map<string, bool> refs;
+  set<string> retired_refs;
+
+  obj_refcount() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(refs, bl);
+    encode(retired_refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(refs, bl);
+    if (struct_v >= 2) {
+      decode(retired_refs, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(obj_refcount)
+
+#endif
diff --git a/src/cls/cephfs/cls_cephfs.cc b/src/cls/cephfs/cls_cephfs.cc
new file mode 100644
index 00000000..7e3214b9
--- /dev/null
+++ b/src/cls/cephfs/cls_cephfs.cc
@@ -0,0 +1,210 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2015 Red Hat
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation.  See file COPYING.
+ *
+ */
+
+
+#include <string>
+#include <errno.h>
+
+#include "objclass/objclass.h"
+
+#include "cls_cephfs.h"
+
+CLS_VER(1,0)
+CLS_NAME(cephfs)
+
+
+std::ostream &operator<<(std::ostream &out, const ObjCeiling &in)
+{
+  out << "id: " << in.id << " size: " << in.size;
+  return out;
+}
+
+
+/**
+ * Set a named xattr to a given value, if and only if the xattr
+ * is not already set to a greater value.
+ *
+ * If the xattr is missing, then it is set to the input integer.
+ *
+ * @param xattr_name: name of xattr to compare against and set
+ * @param input_val: candidate new value, of encode()'able type
+ * @returns 0 on success (irrespective of whether our new value
+ *          was used) else an error code
+ */
+template <typename A>
+static int set_if_greater(cls_method_context_t hctx,
+    const std::string &xattr_name, const A input_val)
+{
+  bufferlist existing_val_bl;
+
+  bool set_val = false;
+  int r = cls_cxx_getxattr(hctx, xattr_name.c_str(), &existing_val_bl);
+  if (r == -ENOENT || existing_val_bl.length() == 0) {
+    set_val = true;
+  } else if (r >= 0) {
+    auto existing_p = existing_val_bl.cbegin();
+    try {
+      A existing_val;
+      decode(existing_val, existing_p);
+      if (!existing_p.end()) {
+        // Trailing junk?  Consider it invalid and overwrite
+        set_val = true;
+      } else {
+        // Valid existing value, do comparison
+        set_val = input_val > existing_val;
+      }
+    } catch (const buffer::error &err) {
+      // Corrupt or empty existing value, overwrite it
+      set_val = true;
+    }
+  } else {
+    return r;
+  }
+
+  // Conditionally set the new xattr
+  if (set_val) {
+    bufferlist set_bl;
+    encode(input_val, set_bl);
+    return cls_cxx_setxattr(hctx, xattr_name.c_str(), &set_bl);
+  } else {
+    return 0;
+  }
+}
+
+static int accumulate_inode_metadata(cls_method_context_t hctx,
+    bufferlist *in, bufferlist *out)
+{
+  ceph_assert(in != NULL);
+  ceph_assert(out != NULL);
+
+  int r = 0;
+
+  // Decode `in`
+  auto q = in->cbegin();
+  AccumulateArgs args;
+  try {
+    args.decode(q);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  ObjCeiling ceiling(args.obj_index, args.obj_size);
+  r = set_if_greater(hctx, args.obj_xattr_name, ceiling);
+  if (r < 0) {
+    return r;
+  }
+
+  r = set_if_greater(hctx, args.mtime_xattr_name, args.mtime);
+  if (r < 0) {
+    return r;
+  }
+
+  r = set_if_greater(hctx, args.obj_size_xattr_name, args.obj_size);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+// I want to select objects that have a name ending 00000000
+// and an xattr (scrub_tag) not equal to a specific value.
+// This is so special case that we can't really pretend it's
+// generic, so just fess up and call this the cephfs filter.
+class PGLSCephFSFilter : public PGLSFilter {
+protected:
+  std::string scrub_tag;
+public:
+  int init(bufferlist::const_iterator& params) override {
+    try {
+      InodeTagFilterArgs args;
+      args.decode(params);
+      scrub_tag = args.scrub_tag;
+    } catch (buffer::error &e) {
+      return -EINVAL;
+    }
+
+    if (scrub_tag.empty()) {
+      xattr = "";
+    } else {
+      xattr = "_scrub_tag";
+    }
+
+    return 0;
+  }
+
+  ~PGLSCephFSFilter() override {}
+  bool reject_empty_xattr() override { return false; }
+  bool filter(const hobject_t &obj, bufferlist& xattr_data,
+                      bufferlist& outdata) override;
+};
+
+bool PGLSCephFSFilter::filter(const hobject_t &obj,
+                             bufferlist& xattr_data, bufferlist& outdata)
+{
+  const std::string need_ending = ".00000000";
+  const std::string &obj_name = obj.oid.name;
+
+  if (obj_name.length() < need_ending.length()) {
+    return false;
+  }
+
+  const bool match = obj_name.compare (obj_name.length() - need_ending.length(), need_ending.length(), need_ending) == 0;
+  if (!match) {
+    return false;
+  }
+
+  if (!scrub_tag.empty() && xattr_data.length() > 0) {
+    std::string tag_ondisk;
+    auto q = xattr_data.cbegin();
+    try {
+      decode(tag_ondisk, q);
+      if (tag_ondisk == scrub_tag)
+	return false;
+    } catch (const buffer::error &err) {
+    }
+  }
+
+  return true;
+}
+
+PGLSFilter *inode_tag_filter()
+{
+  return new PGLSCephFSFilter();
+}
+
+/**
+ * initialize class
+ *
+ * We do two things here: we register the new class, and then register
+ * all of the class's methods.
+ */
+CLS_INIT(cephfs)
+{
+  // this log message, at level 0, will always appear in the ceph-osd
+  // log file.
+  CLS_LOG(0, "loading cephfs");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_accumulate_inode_metadata;
+
+  cls_register("cephfs", &h_class);
+  cls_register_cxx_method(h_class, "accumulate_inode_metadata",
+			  CLS_METHOD_WR | CLS_METHOD_RD,
+			  accumulate_inode_metadata, &h_accumulate_inode_metadata);
+
+  // A PGLS filter
+  cls_register_cxx_filter(h_class, "inode_tag", inode_tag_filter);
+}
+
diff --git a/src/cls/cephfs/cls_cephfs.h b/src/cls/cephfs/cls_cephfs.h
new file mode 100644
index 00000000..89f4dab4
--- /dev/null
+++ b/src/cls/cephfs/cls_cephfs.h
@@ -0,0 +1,147 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2015 Red Hat
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation.  See file COPYING.
+ *
+ */
+
+#include "include/encoding.h"
+
+/**
+ * Value class for the xattr we'll use to accumulate
+ * the highest object seen for a given inode
+ */
+class ObjCeiling {
+  public:
+    uint64_t id;
+    uint64_t size;
+
+    ObjCeiling()
+      : id(0), size(0)
+    {}
+
+    ObjCeiling(uint64_t id_, uint64_t size_)
+      : id(id_), size(size_)
+    {}
+
+    bool operator >(ObjCeiling const &rhs) const
+    {
+      return id > rhs.id;
+    }
+
+    void encode(bufferlist &bl) const
+    {
+      ENCODE_START(1, 1, bl);
+      encode(id, bl);
+      encode(size, bl);
+      ENCODE_FINISH(bl);
+    }
+
+    void decode(bufferlist::const_iterator &p)
+    {
+      DECODE_START(1, p);
+      decode(id, p);
+      decode(size, p);
+      DECODE_FINISH(p);
+    }
+};
+WRITE_CLASS_ENCODER(ObjCeiling)
+
+class AccumulateArgs
+{
+public:
+  uint64_t obj_index;
+  uint64_t obj_size;
+  int64_t mtime;
+  std::string obj_xattr_name;
+  std::string mtime_xattr_name;
+  std::string obj_size_xattr_name;
+
+  AccumulateArgs(
+      uint64_t obj_index_,
+      uint64_t obj_size_,
+      time_t mtime_,
+      const std::string &obj_xattr_name_,
+      const std::string &mtime_xattr_name_,
+      const std::string &obj_size_xattr_name_)
+   : obj_index(obj_index_),
+     obj_size(obj_size_),
+     mtime(mtime_),
+     obj_xattr_name(obj_xattr_name_),
+     mtime_xattr_name(mtime_xattr_name_),
+     obj_size_xattr_name(obj_size_xattr_name_)
+  {}
+
+  AccumulateArgs()
+    : obj_index(0), obj_size(0), mtime(0)
+  {}
+
+  void encode(bufferlist &bl) const
+  {
+    ENCODE_START(1, 1, bl);
+    encode(obj_xattr_name, bl);
+    encode(mtime_xattr_name, bl);
+    encode(obj_size_xattr_name, bl);
+    encode(obj_index, bl);
+    encode(obj_size, bl);
+    encode(mtime, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator &bl)
+  {
+    DECODE_START(1, bl);
+    decode(obj_xattr_name, bl);
+    decode(mtime_xattr_name, bl);
+    decode(obj_size_xattr_name, bl);
+    decode(obj_index, bl);
+    decode(obj_size, bl);
+    decode(mtime, bl);
+    DECODE_FINISH(bl);
+  }
+};
+
+class InodeTagFilterArgs
+{
+  public:
+    std::string scrub_tag;
+
+  void encode(bufferlist &bl) const
+  {
+    ENCODE_START(1, 1, bl);
+    encode(scrub_tag, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator &bl)
+  {
+    DECODE_START(1, bl);
+    decode(scrub_tag, bl);
+    DECODE_FINISH(bl);
+  }
+};
+
+class AccumulateResult
+{
+public:
+  // Index of the highest-indexed object seen
+  uint64_t ceiling_obj_index;
+  // Size of the highest-index object seen
+  uint64_t ceiling_obj_size;
+  // Largest object seen
+  uint64_t max_obj_size;
+  // Highest mtime seen
+  int64_t   max_mtime;
+
+  AccumulateResult()
+    : ceiling_obj_index(0), ceiling_obj_size(0), max_obj_size(0), max_mtime(0)
+  {}
+};
+
diff --git a/src/cls/cephfs/cls_cephfs_client.cc b/src/cls/cephfs/cls_cephfs_client.cc
new file mode 100644
index 00000000..988eab41
--- /dev/null
+++ b/src/cls/cephfs/cls_cephfs_client.cc
@@ -0,0 +1,177 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2015 Red Hat
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software
+ * Foundation.  See file COPYING.
+ *
+ */
+
+
+
+#include "include/rados/librados.hpp"
+#include "mds/CInode.h"
+
+#include "cls_cephfs_client.h"
+
+#define XATTR_CEILING "scan_ceiling"
+#define XATTR_MAX_MTIME "scan_max_mtime"
+#define XATTR_MAX_SIZE "scan_max_size"
+
+int ClsCephFSClient::accumulate_inode_metadata(
+  librados::IoCtx &ctx,
+  inodeno_t inode_no,
+  const uint64_t obj_index,
+  const uint64_t obj_size,
+  const time_t mtime)
+{
+  AccumulateArgs args(
+      obj_index,
+      obj_size,
+      mtime,
+      XATTR_CEILING,
+      XATTR_MAX_MTIME,
+      XATTR_MAX_SIZE);
+
+  // Generate 0th object name, where we will accumulate sizes/mtimes
+  object_t zeroth_object = InodeStore::get_object_name(inode_no, frag_t(), "");
+
+  // Construct a librados operation invoking our class method
+  librados::ObjectReadOperation op;
+  bufferlist inbl;
+  args.encode(inbl);
+  op.exec("cephfs", "accumulate_inode_metadata", inbl);
+
+  // Execute op
+  bufferlist outbl;
+  return ctx.operate(zeroth_object.name, &op, &outbl);
+}
+
+int ClsCephFSClient::delete_inode_accumulate_result(
+    librados::IoCtx &ctx,
+    const std::string &oid)
+{
+  librados::ObjectWriteOperation op;
+
+  // Remove xattrs from object
+  //
+  op.rmxattr(XATTR_CEILING);
+  op.rmxattr(XATTR_MAX_SIZE);
+  op.rmxattr(XATTR_MAX_MTIME);
+
+  return (ctx.operate(oid, &op));
+}
+
+int ClsCephFSClient::fetch_inode_accumulate_result(
+  librados::IoCtx &ctx,
+  const std::string &oid,
+  inode_backtrace_t *backtrace,
+  file_layout_t *layout,
+  AccumulateResult *result)
+{
+  ceph_assert(backtrace != NULL);
+  ceph_assert(result != NULL);
+
+  librados::ObjectReadOperation op;
+
+  int scan_ceiling_r = 0;
+  bufferlist scan_ceiling_bl;
+  op.getxattr(XATTR_CEILING, &scan_ceiling_bl, &scan_ceiling_r);
+
+  int scan_max_size_r = 0;
+  bufferlist scan_max_size_bl;
+  op.getxattr(XATTR_MAX_SIZE, &scan_max_size_bl, &scan_max_size_r);
+
+  int scan_max_mtime_r = 0;
+  bufferlist scan_max_mtime_bl;
+  op.getxattr(XATTR_MAX_MTIME, &scan_max_mtime_bl, &scan_max_mtime_r);
+
+  int parent_r = 0;
+  bufferlist parent_bl;
+  op.getxattr("parent", &parent_bl, &parent_r);
+  op.set_op_flags2(librados::OP_FAILOK);
+
+  int layout_r = 0;
+  bufferlist layout_bl;
+  op.getxattr("layout", &layout_bl, &layout_r);
+  op.set_op_flags2(librados::OP_FAILOK);
+
+  bufferlist op_bl;
+  int r = ctx.operate(oid, &op, &op_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  // Load scan_ceiling
+  try {
+    auto scan_ceiling_bl_iter = scan_ceiling_bl.cbegin();
+    ObjCeiling ceiling;
+    ceiling.decode(scan_ceiling_bl_iter);
+    result->ceiling_obj_index = ceiling.id;
+    result->ceiling_obj_size = ceiling.size;
+  } catch (const buffer::error &err) {
+    //dout(4) << "Invalid size attr on '" << oid << "'" << dendl;
+    return -EINVAL;
+  }
+
+  // Load scan_max_size
+  try {
+    auto scan_max_size_bl_iter = scan_max_size_bl.cbegin();
+    decode(result->max_obj_size, scan_max_size_bl_iter);
+  } catch (const buffer::error &err) {
+    //dout(4) << "Invalid size attr on '" << oid << "'" << dendl;
+    return -EINVAL;
+  }
+
+  // Load scan_max_mtime
+  try {
+    auto scan_max_mtime_bl_iter = scan_max_mtime_bl.cbegin();
+    decode(result->max_mtime, scan_max_mtime_bl_iter);
+  } catch (const buffer::error &err) {
+    //dout(4) << "Invalid size attr on '" << oid << "'" << dendl;
+    return -EINVAL;
+  }
+
+  // Deserialize backtrace
+  if (parent_bl.length()) {
+    try {
+      auto q = parent_bl.cbegin();
+      backtrace->decode(q);
+    } catch (buffer::error &e) {
+      //dout(4) << "Corrupt backtrace on '" << oid << "': " << e << dendl;
+      return -EINVAL;
+    }
+  }
+
+  // Deserialize layout
+  if (layout_bl.length()) {
+    try {
+      auto q = layout_bl.cbegin();
+      decode(*layout, q);
+    } catch (buffer::error &e) {
+      return -EINVAL;
+    }
+  }
+
+  return 0;
+}
+
+void ClsCephFSClient::build_tag_filter(
+          const std::string &scrub_tag,
+          bufferlist *out_bl)
+{
+  ceph_assert(out_bl != NULL);
+
+  // Leading part of bl is un-versioned string naming the filter
+  encode(std::string("cephfs.inode_tag"), *out_bl);
+
+  // Filter-specific part of the bl: in our case this is a versioned structure
+  InodeTagFilterArgs args;
+  args.scrub_tag = scrub_tag;
+  args.encode(*out_bl);
+}
diff --git a/src/cls/cephfs/cls_cephfs_client.h b/src/cls/cephfs/cls_cephfs_client.h
new file mode 100644
index 00000000..744c0aed
--- /dev/null
+++ b/src/cls/cephfs/cls_cephfs_client.h
@@ -0,0 +1,33 @@
+
+#include "include/rados/librados_fwd.hpp"
+#include "mds/mdstypes.h"
+#include "cls_cephfs.h"
+
+class AccumulateArgs;
+
+class ClsCephFSClient
+{
+  public:
+  static int accumulate_inode_metadata(
+      librados::IoCtx &ctx,
+      inodeno_t inode_no,
+      const uint64_t obj_index,
+      const uint64_t obj_size,
+      const time_t mtime);
+
+  static int fetch_inode_accumulate_result(
+      librados::IoCtx &ctx,
+      const std::string &oid,
+      inode_backtrace_t *backtrace,
+      file_layout_t *layout,
+      AccumulateResult *result);
+
+  static int delete_inode_accumulate_result(
+      librados::IoCtx &ctx,
+      const std::string &oid);
+
+  static void build_tag_filter(
+      const std::string &scrub_tag,
+      bufferlist *out_bl);
+};
+
diff --git a/src/cls/hello/cls_hello.cc b/src/cls/hello/cls_hello.cc
new file mode 100644
index 00000000..accca898
--- /dev/null
+++ b/src/cls/hello/cls_hello.cc
@@ -0,0 +1,344 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+/*
+ * This is a simple example RADOS class, designed to be usable as a
+ * template for implementing new methods.
+ *
+ * Our goal here is to illustrate the interface between the OSD and
+ * the class and demonstrate what kinds of things a class can do.
+ *
+ * Note that any *real* class will probably have a much more
+ * sophisticated protocol dealing with the in and out data buffers.
+ * For an example of the model that we've settled on for handling that
+ * in a clean way, please refer to cls_lock or cls_version for
+ * relatively simple examples of how the parameter encoding can be
+ * encoded in a way that allows for forward and backward compatibility
+ * between client vs class revisions.
+ */
+
+/*
+ * A quick note about bufferlists:
+ *
+ * The bufferlist class allows memory buffers to be concatenated,
+ * truncated, spliced, "copied," encoded/embedded, and decoded.  For
+ * most operations no actual data is ever copied, making bufferlists
+ * very convenient for efficiently passing data around.
+ *
+ * bufferlist is actually a typedef of buffer::list, and is defined in
+ * include/buffer.h (and implemented in common/buffer.cc).
+ */
+
+#include <algorithm>
+#include <string>
+#include <sstream>
+#include <errno.h>
+
+#include "objclass/objclass.h"
+
+CLS_VER(1,0)
+CLS_NAME(hello)
+
+/**
+ * say hello - a "read" method that does not depend on the object
+ *
+ * This is an example of a method that does some computation and
+ * returns data to the caller, without depending on the local object
+ * content.
+ */
+static int say_hello(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // see if the input data from the client matches what this method
+  // expects to receive.  your class can fill this buffer with what it
+  // wants.
+  if (in->length() > 100)
+    return -EINVAL;
+
+  // we generate our reply
+  out->append("Hello, ");
+  if (in->length() == 0)
+    out->append("world");
+  else
+    out->append(*in);
+  out->append("!");
+
+  // this return value will be returned back to the librados caller
+  return 0;
+}
+
+/**
+ * record hello - a "write" method that creates an object
+ *
+ * This method modifies a local object (in this case, by creating it
+ * if it doesn't exist).  We make multiple write calls (write,
+ * setxattr) which are accumulated and applied as an atomic
+ * transaction.
+ */
+static int record_hello(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // we can write arbitrary stuff to the ceph-osd debug log.  each log
+  // message is accompanied by an integer log level.  smaller is
+  // "louder".  how much of this makes it into the log is controlled
+  // by the debug_cls option on the ceph-osd, similar to how other log
+  // levels are controlled.  this message, at level 20, will generally
+  // not be seen by anyone unless debug_cls is set at 20 or higher.
+  CLS_LOG(20, "in record_hello");
+
+  // see if the input data from the client matches what this method
+  // expects to receive.  your class can fill this buffer with what it
+  // wants.
+  if (in->length() > 100)
+    return -EINVAL;
+
+  // only say hello to non-existent objects
+  if (cls_cxx_stat(hctx, NULL, NULL) == 0)
+    return -EEXIST;
+
+  bufferlist content;
+  content.append("Hello, ");
+  if (in->length() == 0)
+    content.append("world");
+  else
+    content.append(*in);
+  content.append("!");
+
+  // create/write the object
+  int r = cls_cxx_write_full(hctx, &content);
+  if (r < 0)
+    return r;
+
+  // also make note of who said it
+  entity_inst_t origin;
+  cls_get_request_origin(hctx, &origin);
+  ostringstream ss;
+  ss << origin;
+  bufferlist attrbl;
+  attrbl.append(ss.str());
+  r = cls_cxx_setxattr(hctx, "said_by", &attrbl);
+  if (r < 0)
+    return r;
+
+  // For write operations, there are two possible outcomes:
+  //
+  //  * For a failure, we return a negative error code.  The out
+  //    buffer can contain any data that we want, and that data will
+  //    be returned to the caller.  No change is made to the object.
+  //
+  //  * For a success, we must return 0 and *no* data in the out
+  //    buffer.  This is becaues the OSD does not log write result
+  //    codes or output buffers and we need a replayed/resent
+  //    operation (e.g., after a TCP disconnect) to be idempotent.
+  //
+  //    If a class returns a positive value or puts data in the out
+  //    buffer, the OSD code will ignore it and return 0 to the
+  //    client.
+  return 0;
+}
+
+static int writes_dont_return_data(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // make some change to the object
+  bufferlist attrbl;
+  attrbl.append("bar");
+  int r = cls_cxx_setxattr(hctx, "foo", &attrbl);
+  if (r < 0)
+    return r;
+
+  if (in->length() > 0) {
+    // note that if we return anything < 0 (an error), this
+    // operation/transaction will abort, and the setattr above will
+    // never happen.  however, we *can* return data on error.
+    out->append("too much input data!");
+    return -EINVAL;
+  }
+
+  // try to return some data.  note that this *won't* reach the
+  // client!  see the matching test case in test_cls_hello.cc.
+  out->append("you will never see this");
+
+  // if we try to return anything > 0 here the client will see 0.
+  return 42;
+}
+
+
+/**
+ * replay - a "read" method to get a previously recorded hello
+ *
+ * This is a read method that will retrieve a previously recorded
+ * hello statement.
+ */
+static int replay(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // read contents out of the on-disk object.  our behavior can be a
+  // function of either the request alone, or the request and the
+  // on-disk state, depending on whether the RD flag is specified when
+  // registering the method (see the __cls__init function below).
+  int r = cls_cxx_read(hctx, 0, 1100, out);
+  if (r < 0)
+    return r;
+
+  // note that our return value need not be the length of the returned
+  // data; it can be whatever value we want: positive, zero or
+  // negative (this is a read).
+  return 0;
+}
+
+/**
+ * turn_it_to_11 - a "write" method that mutates existing object data
+ *
+ * A write method can depend on previous object content (i.e., perform
+ * a read/modify/write operation).  This atomically transitions the
+ * object state from the old content to the new content.
+ */
+static int turn_it_to_11(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // see if the input data from the client matches what this method
+  // expects to receive.  your class can fill this buffer with what it
+  // wants.
+  if (in->length() != 0)
+    return -EINVAL;
+
+  bufferlist previous;
+  int r = cls_cxx_read(hctx, 0, 1100, &previous);
+  if (r < 0)
+    return r;
+
+  std::string str(previous.c_str(), previous.length());
+  std::transform(str.begin(), str.end(), str.begin(), ::toupper);
+  previous.clear();
+  previous.append(str);
+
+  // replace previous byte data content (write_full == truncate(0) + write)
+  r = cls_cxx_write_full(hctx, &previous);
+  if (r < 0)
+    return r;
+
+  // record who did it
+  entity_inst_t origin;
+  cls_get_request_origin(hctx, &origin);
+  ostringstream ss;
+  ss << origin;
+  bufferlist attrbl;
+  attrbl.append(ss.str());
+  r = cls_cxx_setxattr(hctx, "amplified_by", &attrbl);
+  if (r < 0)
+    return r;
+
+  // return value is 0 for success; out buffer is empty.
+  return 0;
+}
+
+/**
+ * example method that does not behave
+ *
+ * This method is registered as WR but tries to read
+ */
+static int bad_reader(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  return cls_cxx_read(hctx, 0, 100, out);
+}
+
+/**
+ * example method that does not behave
+ *
+ * This method is registered as RD but tries to write
+ */
+static int bad_writer(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  return cls_cxx_write_full(hctx, in);
+}
+
+
+class PGLSHelloFilter : public PGLSFilter {
+  string val;
+public:
+  int init(bufferlist::const_iterator& params) override {
+    try {
+      decode(xattr, params);
+      decode(val, params);
+    } catch (buffer::error &e) {
+      return -EINVAL;
+    }
+    return 0;
+  }
+
+  ~PGLSHelloFilter() override {}
+  bool filter(const hobject_t &obj, bufferlist& xattr_data,
+                      bufferlist& outdata) override
+  {
+    if (val.size() != xattr_data.length())
+      return false;
+
+    if (memcmp(val.c_str(), xattr_data.c_str(), val.size()))
+      return false;
+
+    return true;
+  }
+};
+
+
+PGLSFilter *hello_filter()
+{
+  return new PGLSHelloFilter();
+}
+
+
+/**
+ * initialize class
+ *
+ * We do two things here: we register the new class, and then register
+ * all of the class's methods.
+ */
+CLS_INIT(hello)
+{
+  // this log message, at level 0, will always appear in the ceph-osd
+  // log file.
+  CLS_LOG(0, "loading cls_hello");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_say_hello;
+  cls_method_handle_t h_record_hello;
+  cls_method_handle_t h_replay;
+  cls_method_handle_t h_writes_dont_return_data;
+  cls_method_handle_t h_turn_it_to_11;
+  cls_method_handle_t h_bad_reader;
+  cls_method_handle_t h_bad_writer;
+
+  cls_register("hello", &h_class);
+
+  // There are two flags we specify for methods:
+  //
+  //    RD : whether this method (may) read prior object state
+  //    WR : whether this method (may) write or update the object
+  //
+  // A method can be RD, WR, neither, or both.  If a method does
+  // neither, the data it returns to the caller is a function of the
+  // request and not the object contents.
+
+  cls_register_cxx_method(h_class, "say_hello",
+			  CLS_METHOD_RD,
+			  say_hello, &h_say_hello);
+  cls_register_cxx_method(h_class, "record_hello",
+			  CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+			  record_hello, &h_record_hello);
+  cls_register_cxx_method(h_class, "writes_dont_return_data",
+			  CLS_METHOD_WR,
+			  writes_dont_return_data, &h_writes_dont_return_data);
+  cls_register_cxx_method(h_class, "replay",
+			  CLS_METHOD_RD,
+			  replay, &h_replay);
+
+  // RD | WR is a read-modify-write method.
+  cls_register_cxx_method(h_class, "turn_it_to_11",
+			  CLS_METHOD_RD | CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+			  turn_it_to_11, &h_turn_it_to_11);
+
+  // counter-examples
+  cls_register_cxx_method(h_class, "bad_reader", CLS_METHOD_WR,
+			  bad_reader, &h_bad_reader);
+  cls_register_cxx_method(h_class, "bad_writer", CLS_METHOD_RD,
+			  bad_writer, &h_bad_writer);
+
+  // A PGLS filter
+  cls_register_cxx_filter(h_class, "hello", hello_filter);
+}
diff --git a/src/cls/journal/cls_journal.cc b/src/cls/journal/cls_journal.cc
new file mode 100644
index 00000000..ff909bea
--- /dev/null
+++ b/src/cls/journal/cls_journal.cc
@@ -0,0 +1,1230 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "include/int_types.h"
+#include "include/buffer.h"
+#include "include/encoding.h"
+#include "common/errno.h"
+#include "objclass/objclass.h"
+#include "cls/journal/cls_journal_types.h"
+#include <errno.h>
+#include <map>
+#include <string>
+#include <sstream>
+
+CLS_VER(1, 0)
+CLS_NAME(journal)
+
+namespace {
+
+static const uint64_t MAX_KEYS_READ = 64;
+
+static const std::string HEADER_KEY_ORDER          = "order";
+static const std::string HEADER_KEY_SPLAY_WIDTH    = "splay_width";
+static const std::string HEADER_KEY_POOL_ID        = "pool_id";
+static const std::string HEADER_KEY_MINIMUM_SET    = "minimum_set";
+static const std::string HEADER_KEY_ACTIVE_SET     = "active_set";
+static const std::string HEADER_KEY_NEXT_TAG_TID   = "next_tag_tid";
+static const std::string HEADER_KEY_NEXT_TAG_CLASS = "next_tag_class";
+static const std::string HEADER_KEY_CLIENT_PREFIX  = "client_";
+static const std::string HEADER_KEY_TAG_PREFIX     = "tag_";
+
+std::string to_hex(uint64_t value) {
+  std::ostringstream oss;
+  oss << std::setw(16) << std::setfill('0') << std::hex << value;
+  return oss.str();
+}
+
+std::string key_from_client_id(const std::string &client_id) {
+  return HEADER_KEY_CLIENT_PREFIX + client_id;
+}
+
+std::string key_from_tag_tid(uint64_t tag_tid) {
+  return HEADER_KEY_TAG_PREFIX + to_hex(tag_tid);
+}
+
+uint64_t tag_tid_from_key(const std::string &key) {
+  std::istringstream iss(key);
+  uint64_t id;
+  iss.ignore(HEADER_KEY_TAG_PREFIX.size()) >> std::hex >> id;
+  return id;
+}
+
+template <typename T>
+int read_key(cls_method_context_t hctx, const string &key, T *t,
+             bool ignore_enoent = false) {
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, key, &bl);
+  if (r == -ENOENT) {
+    if (ignore_enoent) {
+      r = 0;
+    }
+    return r;
+  } else if (r < 0) {
+    CLS_ERR("failed to get omap key: %s", key.c_str());
+    return r;
+  }
+
+  try {
+    auto iter = bl.cbegin();
+    decode(*t, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+  return 0;
+}
+
+template <typename T>
+int write_key(cls_method_context_t hctx, const string &key, const T &t) {
+  bufferlist bl;
+  encode(t, bl);
+
+  int r = cls_cxx_map_set_val(hctx, key, &bl);
+  if (r < 0) {
+    CLS_ERR("failed to set omap key: %s", key.c_str());
+    return r;
+  }
+  return 0;
+}
+
+int remove_key(cls_method_context_t hctx, const string &key) {
+  int r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0 && r != -ENOENT) {
+      CLS_ERR("failed to remove key: %s", key.c_str());
+      return r;
+  }
+  return 0;
+}
+
+int expire_tags(cls_method_context_t hctx, const std::string *skip_client_id) {
+
+  std::string skip_client_key;
+  if (skip_client_id != nullptr) {
+    skip_client_key = key_from_client_id(*skip_client_id);
+  }
+
+  uint64_t minimum_tag_tid = std::numeric_limits<uint64_t>::max();
+  std::string last_read = "";
+  bool more;
+  do {
+    std::map<std::string, bufferlist> vals;
+    int r = cls_cxx_map_get_vals(hctx, last_read, HEADER_KEY_CLIENT_PREFIX,
+                                 MAX_KEYS_READ, &vals, &more);
+    if (r < 0 && r != -ENOENT) {
+      CLS_ERR("failed to retrieve registered clients: %s",
+              cpp_strerror(r).c_str());
+      return r;
+    }
+
+    for (auto &val : vals) {
+      // if we are removing a client, skip its commit positions
+      if (val.first == skip_client_key) {
+        continue;
+      }
+
+      cls::journal::Client client;
+      auto iter = val.second.cbegin();
+      try {
+        decode(client, iter);
+      } catch (const buffer::error &err) {
+        CLS_ERR("error decoding registered client: %s",
+                val.first.c_str());
+        return -EIO;
+      }
+
+      if (client.state == cls::journal::CLIENT_STATE_DISCONNECTED) {
+        // don't allow a disconnected client to prevent pruning
+        continue;
+      } else if (client.commit_position.object_positions.empty()) {
+        // cannot prune if one or more clients has an empty commit history
+        return 0;
+      }
+
+      for (auto object_position : client.commit_position.object_positions) {
+        minimum_tag_tid = std::min(minimum_tag_tid, object_position.tag_tid);
+      }
+    }
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  } while (more);
+
+  // cannot expire tags if a client hasn't committed yet
+  if (minimum_tag_tid == std::numeric_limits<uint64_t>::max()) {
+    return 0;
+  }
+
+  // compute the minimum in-use tag for each class
+  std::map<uint64_t, uint64_t> minimum_tag_class_to_tids;
+  typedef enum { TAG_PASS_CALCULATE_MINIMUMS,
+                 TAG_PASS_SCRUB,
+                 TAG_PASS_DONE } TagPass;
+  int tag_pass = TAG_PASS_CALCULATE_MINIMUMS;
+  last_read = HEADER_KEY_TAG_PREFIX;
+  do {
+    std::map<std::string, bufferlist> vals;
+    int r = cls_cxx_map_get_vals(hctx, last_read, HEADER_KEY_TAG_PREFIX,
+                             MAX_KEYS_READ, &vals, &more);
+    if (r < 0 && r != -ENOENT) {
+      CLS_ERR("failed to retrieve tags: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+
+    for (auto &val : vals) {
+      cls::journal::Tag tag;
+      auto iter = val.second.cbegin();
+      try {
+        decode(tag, iter);
+      } catch (const buffer::error &err) {
+        CLS_ERR("error decoding tag: %s", val.first.c_str());
+        return -EIO;
+      }
+
+      if (tag.tid != tag_tid_from_key(val.first)) {
+        CLS_ERR("tag tid mismatched: %s", val.first.c_str());
+        return -EINVAL;
+      }
+
+      if (tag_pass == TAG_PASS_CALCULATE_MINIMUMS) {
+        minimum_tag_class_to_tids[tag.tag_class] = tag.tid;
+      } else if (tag_pass == TAG_PASS_SCRUB &&
+                 tag.tid < minimum_tag_class_to_tids[tag.tag_class]) {
+        r = remove_key(hctx, val.first);
+        if (r < 0) {
+          return r;
+        }
+      }
+
+      if (tag.tid >= minimum_tag_tid) {
+        // no need to check for tag classes beyond this point
+        vals.clear();
+        more = false;
+        break;
+      }
+    }
+
+    if (tag_pass != TAG_PASS_DONE && !more) {
+      last_read = HEADER_KEY_TAG_PREFIX;
+      ++tag_pass;
+    } else if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  } while (tag_pass != TAG_PASS_DONE);
+  return 0;
+}
+
+int get_client_list_range(cls_method_context_t hctx,
+                          std::set<cls::journal::Client> *clients,
+                          std::string start_after, uint64_t max_return) {
+  std::string last_read;
+  if (!start_after.empty()) {
+    last_read = key_from_client_id(start_after);
+  }
+
+  std::map<std::string, bufferlist> vals;
+  bool more;
+  int r = cls_cxx_map_get_vals(hctx, last_read, HEADER_KEY_CLIENT_PREFIX,
+                               max_return, &vals, &more);
+  if (r < 0) {
+    CLS_ERR("failed to retrieve omap values: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  for (std::map<std::string, bufferlist>::iterator it = vals.begin();
+       it != vals.end(); ++it) {
+    try {
+      auto iter = it->second.cbegin();
+
+      cls::journal::Client client;
+      decode(client, iter);
+      clients->insert(client);
+    } catch (const buffer::error &err) {
+      CLS_ERR("could not decode client '%s': %s", it->first.c_str(),
+              err.what());
+      return -EIO;
+    }
+  }
+
+  return 0;
+}
+
+int find_min_commit_position(cls_method_context_t hctx,
+                             cls::journal::ObjectSetPosition *minset) {
+  int r;
+  bool valid = false;
+  std::string start_after = "";
+  uint64_t tag_tid = 0, entry_tid = 0;
+
+  while (true) {
+    std::set<cls::journal::Client> batch;
+
+    r = get_client_list_range(hctx, &batch, start_after, cls::journal::JOURNAL_MAX_RETURN);
+    if ((r < 0) || batch.empty()) {
+      break;
+    }
+
+    start_after = batch.rbegin()->id;
+
+    // update the (minimum) commit position from this batch of clients
+    for(std::set<cls::journal::Client>::iterator it = batch.begin();
+        it != batch.end(); ++it) {
+      cls::journal::ObjectSetPosition object_set_position = (*it).commit_position;
+      if (object_set_position.object_positions.empty()) {
+	*minset = cls::journal::ObjectSetPosition();
+	break;
+      }
+      cls::journal::ObjectPosition first = object_set_position.object_positions.front();
+
+      // least tag_tid (or least entry_tid for matching tag_tid)
+      if (!valid || (tag_tid > first.tag_tid) || ((tag_tid == first.tag_tid) && (entry_tid > first.entry_tid))) {
+	tag_tid = first.tag_tid;
+	entry_tid = first.entry_tid;
+	*minset = cls::journal::ObjectSetPosition(object_set_position);
+	valid = true;
+      }
+    }
+
+    // got the last batch, we're done
+    if (batch.size() < cls::journal::JOURNAL_MAX_RETURN) {
+      break;
+    }
+  }
+
+  return r;
+}
+
+} // anonymous namespace
+
+/**
+ * Input:
+ * @param order (uint8_t) - bits to shift to compute the object max size
+ * @param splay width (uint8_t) - number of active journal objects
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_create(cls_method_context_t hctx, bufferlist *in, bufferlist *out) {
+  uint8_t order;
+  uint8_t splay_width;
+  int64_t pool_id;
+  try {
+    auto iter = in->cbegin();
+    decode(order, iter);
+    decode(splay_width, iter);
+    decode(pool_id, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  bufferlist stored_orderbl;
+  int r = cls_cxx_map_get_val(hctx, HEADER_KEY_ORDER, &stored_orderbl);
+  if (r >= 0) {
+    CLS_ERR("journal already exists");
+    return -EEXIST;
+  } else if (r != -ENOENT) {
+    return r;
+  }
+
+  r = write_key(hctx, HEADER_KEY_ORDER, order);
+  if (r < 0) {
+    return r;
+  }
+
+  r = write_key(hctx, HEADER_KEY_SPLAY_WIDTH, splay_width);
+  if (r < 0) {
+    return r;
+  }
+
+  r = write_key(hctx, HEADER_KEY_POOL_ID, pool_id);
+  if (r < 0) {
+    return r;
+  }
+
+  uint64_t object_set = 0;
+  r = write_key(hctx, HEADER_KEY_ACTIVE_SET, object_set);
+  if (r < 0) {
+    return r;
+  }
+
+  r = write_key(hctx, HEADER_KEY_MINIMUM_SET, object_set);
+  if (r < 0) {
+    return r;
+  }
+
+  uint64_t tag_id = 0;
+  r = write_key(hctx, HEADER_KEY_NEXT_TAG_TID, tag_id);
+  if (r < 0) {
+    return r;
+  }
+
+  r = write_key(hctx, HEADER_KEY_NEXT_TAG_CLASS, tag_id);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * order (uint8_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_order(cls_method_context_t hctx, bufferlist *in,
+                      bufferlist *out) {
+  uint8_t order;
+  int r = read_key(hctx, HEADER_KEY_ORDER, &order);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(order, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * splay_width (uint8_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_splay_width(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  uint8_t splay_width;
+  int r = read_key(hctx, HEADER_KEY_SPLAY_WIDTH, &splay_width);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(splay_width, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * pool_id (int64_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_pool_id(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  int64_t pool_id = 0;
+  int r = read_key(hctx, HEADER_KEY_POOL_ID, &pool_id);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(pool_id, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * object set (uint64_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_minimum_set(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  uint64_t minimum_set;
+  int r = read_key(hctx, HEADER_KEY_MINIMUM_SET, &minimum_set);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(minimum_set, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param object set (uint64_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_set_minimum_set(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  uint64_t object_set;
+  try {
+    auto iter = in->cbegin();
+    decode(object_set, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  uint64_t current_active_set;
+  int r = read_key(hctx, HEADER_KEY_ACTIVE_SET, &current_active_set);
+  if (r < 0) {
+    return r;
+  }
+
+  if (current_active_set < object_set) {
+    CLS_LOG(10, "active object set earlier than minimum: %" PRIu64
+            " < %" PRIu64, current_active_set, object_set);
+    return -EINVAL;
+  }
+
+  uint64_t current_minimum_set;
+  r = read_key(hctx, HEADER_KEY_MINIMUM_SET, &current_minimum_set);
+  if (r < 0) {
+    return r;
+  }
+
+  if (object_set == current_minimum_set) {
+    return 0;
+  } else if (object_set < current_minimum_set) {
+    CLS_ERR("object number earlier than current object: %" PRIu64 " < %" PRIu64,
+            object_set, current_minimum_set);
+    return -ESTALE;
+  }
+
+  r = write_key(hctx, HEADER_KEY_MINIMUM_SET, object_set);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * object set (uint64_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_active_set(cls_method_context_t hctx, bufferlist *in,
+                           bufferlist *out) {
+  uint64_t active_set;
+  int r = read_key(hctx, HEADER_KEY_ACTIVE_SET, &active_set);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(active_set, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param object set (uint64_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_set_active_set(cls_method_context_t hctx, bufferlist *in,
+                           bufferlist *out) {
+  uint64_t object_set;
+  try {
+    auto iter = in->cbegin();
+    decode(object_set, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  uint64_t current_minimum_set;
+  int r = read_key(hctx, HEADER_KEY_MINIMUM_SET, &current_minimum_set);
+  if (r < 0) {
+    return r;
+  }
+
+  if (current_minimum_set > object_set) {
+    CLS_ERR("minimum object set later than active: %" PRIu64
+            " > %" PRIu64, current_minimum_set, object_set);
+    return -EINVAL;
+  }
+
+  uint64_t current_active_set;
+  r = read_key(hctx, HEADER_KEY_ACTIVE_SET, &current_active_set);
+  if (r < 0) {
+    return r;
+  }
+
+  if (object_set == current_active_set) {
+    return 0;
+  } else if (object_set < current_active_set) {
+    CLS_ERR("object number earlier than current object: %" PRIu64 " < %" PRIu64,
+            object_set, current_active_set);
+    return -ESTALE;
+  }
+
+  r = write_key(hctx, HEADER_KEY_ACTIVE_SET, object_set);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param id (string) - unique client id
+ *
+ * Output:
+ * cls::journal::Client
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_client(cls_method_context_t hctx, bufferlist *in,
+                       bufferlist *out) {
+  std::string id;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::string key(key_from_client_id(id));
+  cls::journal::Client client;
+  int r = read_key(hctx, key, &client);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(client, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param id (string) - unique client id
+ * @param data (bufferlist) - opaque data associated to client
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_client_register(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  std::string id;
+  bufferlist data;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+    decode(data, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  uint8_t order;
+  int r = read_key(hctx, HEADER_KEY_ORDER, &order);
+  if (r < 0) {
+    return r;
+  }
+
+  std::string key(key_from_client_id(id));
+  bufferlist stored_clientbl;
+  r = cls_cxx_map_get_val(hctx, key, &stored_clientbl);
+  if (r >= 0) {
+    CLS_ERR("duplicate client id: %s", id.c_str());
+    return -EEXIST;
+  } else if (r != -ENOENT) {
+    return r;
+  }
+
+  cls::journal::ObjectSetPosition minset;
+  r = find_min_commit_position(hctx, &minset);
+  if (r < 0)
+    return r;
+
+  cls::journal::Client client(id, data, minset);
+  r = write_key(hctx, key, client);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param id (string) - unique client id
+ * @param data (bufferlist) - opaque data associated to client
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_client_update_data(cls_method_context_t hctx, bufferlist *in,
+                               bufferlist *out) {
+  std::string id;
+  bufferlist data;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+    decode(data, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::string key(key_from_client_id(id));
+  cls::journal::Client client;
+  int r = read_key(hctx, key, &client);
+  if (r < 0) {
+    return r;
+  }
+
+  client.data = data;
+  r = write_key(hctx, key, client);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param id (string) - unique client id
+ * @param state (uint8_t) - client state
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_client_update_state(cls_method_context_t hctx, bufferlist *in,
+                                bufferlist *out) {
+  std::string id;
+  cls::journal::ClientState state;
+  bufferlist data;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+    uint8_t state_raw;
+    decode(state_raw, iter);
+    state = static_cast<cls::journal::ClientState>(state_raw);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::string key(key_from_client_id(id));
+  cls::journal::Client client;
+  int r = read_key(hctx, key, &client);
+  if (r < 0) {
+    return r;
+  }
+
+  client.state = state;
+  r = write_key(hctx, key, client);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param id (string) - unique client id
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_client_unregister(cls_method_context_t hctx, bufferlist *in,
+                              bufferlist *out) {
+  std::string id;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::string key(key_from_client_id(id));
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, key, &bl);
+  if (r < 0) {
+    CLS_ERR("client is not registered: %s", id.c_str());
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0) {
+    CLS_ERR("failed to remove omap key: %s", key.c_str());
+    return r;
+  }
+
+  // prune expired tags
+  r = expire_tags(hctx, &id);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param client_id (uint64_t) - unique client id
+ * @param commit_position (ObjectSetPosition)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_client_commit(cls_method_context_t hctx, bufferlist *in,
+                          bufferlist *out) {
+  std::string id;
+  cls::journal::ObjectSetPosition commit_position;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+    decode(commit_position, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  uint8_t splay_width;
+  int r = read_key(hctx, HEADER_KEY_SPLAY_WIDTH, &splay_width);
+  if (r < 0) {
+    return r;
+  }
+  if (commit_position.object_positions.size() > splay_width) {
+    CLS_ERR("too many object positions");
+    return -EINVAL;
+  }
+
+  std::string key(key_from_client_id(id));
+  cls::journal::Client client;
+  r = read_key(hctx, key, &client);
+  if (r < 0) {
+    return r;
+  }
+
+  if (client.commit_position == commit_position) {
+    return 0;
+  }
+
+  client.commit_position = commit_position;
+  r = write_key(hctx, key, client);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param start_after (string)
+ * @param max_return (uint64_t)
+ *
+ * Output:
+ * clients (set<cls::journal::Client>) - collection of registered clients
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_client_list(cls_method_context_t hctx, bufferlist *in,
+                        bufferlist *out) {
+  std::string start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::set<cls::journal::Client> clients;
+  int r = get_client_list_range(hctx, &clients, start_after, max_return);
+  if (r < 0)
+    return r;
+
+  encode(clients, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_next_tag_tid(cls_method_context_t hctx, bufferlist *in,
+                             bufferlist *out) {
+  uint64_t tag_tid;
+  int r = read_key(hctx, HEADER_KEY_NEXT_TAG_TID, &tag_tid);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(tag_tid, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param tag_tid (uint64_t)
+ *
+ * Output:
+ * cls::journal::Tag
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_get_tag(cls_method_context_t hctx, bufferlist *in,
+                    bufferlist *out) {
+  uint64_t tag_tid;
+  try {
+    auto iter = in->cbegin();
+    decode(tag_tid, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::string key(key_from_tag_tid(tag_tid));
+  cls::journal::Tag tag;
+  int r = read_key(hctx, key, &tag);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(tag, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param tag_tid (uint64_t)
+ * @param tag_class (uint64_t)
+ * @param data (bufferlist)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_tag_create(cls_method_context_t hctx, bufferlist *in,
+                       bufferlist *out) {
+  uint64_t tag_tid;
+  uint64_t tag_class;
+  bufferlist data;
+  try {
+    auto iter = in->cbegin();
+    decode(tag_tid, iter);
+    decode(tag_class, iter);
+    decode(data, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  std::string key(key_from_tag_tid(tag_tid));
+  bufferlist stored_tag_bl;
+  int r = cls_cxx_map_get_val(hctx, key, &stored_tag_bl);
+  if (r >= 0) {
+    CLS_ERR("duplicate tag id: %" PRIu64, tag_tid);
+    return -EEXIST;
+  } else if (r != -ENOENT) {
+    return r;
+  }
+
+  // verify tag tid ordering
+  uint64_t next_tag_tid;
+  r = read_key(hctx, HEADER_KEY_NEXT_TAG_TID, &next_tag_tid);
+  if (r < 0) {
+    return r;
+  }
+  if (tag_tid != next_tag_tid) {
+    CLS_LOG(5, "out-of-order tag sequence: %" PRIu64, tag_tid);
+    return -ESTALE;
+  }
+
+  uint64_t next_tag_class;
+  r = read_key(hctx, HEADER_KEY_NEXT_TAG_CLASS, &next_tag_class);
+  if (r < 0) {
+    return r;
+  }
+
+  if (tag_class == cls::journal::Tag::TAG_CLASS_NEW) {
+    // allocate a new tag class
+    tag_class = next_tag_class;
+    r = write_key(hctx, HEADER_KEY_NEXT_TAG_CLASS, tag_class + 1);
+    if (r < 0) {
+      return r;
+    }
+  } else {
+    // verify tag class range
+    if (tag_class >= next_tag_class) {
+      CLS_ERR("out-of-sequence tag class: %" PRIu64, tag_class);
+      return -EINVAL;
+    }
+  }
+
+  // prune expired tags
+  r = expire_tags(hctx, nullptr);
+  if (r < 0) {
+    return r;
+  }
+
+  // update tag tid sequence
+  r = write_key(hctx, HEADER_KEY_NEXT_TAG_TID, tag_tid + 1);
+  if (r < 0) {
+    return r;
+  }
+
+  // write tag structure
+  cls::journal::Tag tag(tag_tid, tag_class, data);
+  key = key_from_tag_tid(tag_tid);
+  r = write_key(hctx, key, tag);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param start_after_tag_tid (uint64_t) - first tag tid
+ * @param max_return (uint64_t) - max tags to return
+ * @param client_id (std::string) - client id filter
+ * @param tag_class (boost::optional<uint64_t> - optional tag class filter
+ *
+ * Output:
+ * std::set<cls::journal::Tag> - collection of tags
+ * @returns 0 on success, negative error code on failure
+ */
+int journal_tag_list(cls_method_context_t hctx, bufferlist *in,
+                     bufferlist *out) {
+  uint64_t start_after_tag_tid;
+  uint64_t max_return;
+  std::string client_id;
+  boost::optional<uint64_t> tag_class(0);
+
+  // handle compiler false positive about use-before-init
+  tag_class = boost::none;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after_tag_tid, iter);
+    decode(max_return, iter);
+    decode(client_id, iter);
+    decode(tag_class, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  // calculate the minimum tag within client's commit position
+  uint64_t minimum_tag_tid = std::numeric_limits<uint64_t>::max();
+  cls::journal::Client client;
+  int r = read_key(hctx, key_from_client_id(client_id), &client);
+  if (r < 0) {
+    return r;
+  }
+
+  for (auto object_position : client.commit_position.object_positions) {
+    minimum_tag_tid = std::min(minimum_tag_tid, object_position.tag_tid);
+  }
+
+  // compute minimum tags in use per-class
+  std::set<cls::journal::Tag> tags;
+  std::map<uint64_t, uint64_t> minimum_tag_class_to_tids;
+  typedef enum { TAG_PASS_CALCULATE_MINIMUMS,
+                 TAG_PASS_LIST,
+                 TAG_PASS_DONE } TagPass;
+  int tag_pass = (minimum_tag_tid == std::numeric_limits<uint64_t>::max() ?
+    TAG_PASS_LIST : TAG_PASS_CALCULATE_MINIMUMS);
+  std::string last_read = HEADER_KEY_TAG_PREFIX;
+  do {
+    std::map<std::string, bufferlist> vals;
+    bool more;
+    r = cls_cxx_map_get_vals(hctx, last_read, HEADER_KEY_TAG_PREFIX,
+                             MAX_KEYS_READ, &vals, &more);
+    if (r < 0 && r != -ENOENT) {
+      CLS_ERR("failed to retrieve tags: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+
+    for (auto &val : vals) {
+      cls::journal::Tag tag;
+      auto iter = val.second.cbegin();
+      try {
+        decode(tag, iter);
+      } catch (const buffer::error &err) {
+        CLS_ERR("error decoding tag: %s", val.first.c_str());
+        return -EIO;
+      }
+
+      if (tag_pass == TAG_PASS_CALCULATE_MINIMUMS) {
+        minimum_tag_class_to_tids[tag.tag_class] = tag.tid;
+
+        // completed calculation of tag class minimums
+        if (tag.tid >= minimum_tag_tid) {
+          vals.clear();
+          more = false;
+          break;
+        }
+      } else if (tag_pass == TAG_PASS_LIST) {
+        if (start_after_tag_tid != 0 && tag.tid <= start_after_tag_tid) {
+          continue;
+        }
+
+        if (tag.tid >= minimum_tag_class_to_tids[tag.tag_class] &&
+            (!tag_class || *tag_class == tag.tag_class)) {
+          tags.insert(tag);
+        }
+        if (tags.size() >= max_return) {
+          tag_pass = TAG_PASS_DONE;
+        }
+      }
+    }
+
+    if (tag_pass != TAG_PASS_DONE && !more) {
+      last_read = HEADER_KEY_TAG_PREFIX;
+      ++tag_pass;
+    } else if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  } while (tag_pass != TAG_PASS_DONE);
+
+  encode(tags, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param soft_max_size (uint64_t)
+ *
+ * Output:
+ * @returns 0 if object size less than max, negative error code otherwise
+ */
+int journal_object_guard_append(cls_method_context_t hctx, bufferlist *in,
+                                bufferlist *out) {
+  uint64_t soft_max_size;
+  try {
+    auto iter = in->cbegin();
+    decode(soft_max_size, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode input parameters: %s", err.what());
+    return -EINVAL;
+  }
+
+  uint64_t size;
+  time_t mtime;
+  int r = cls_cxx_stat(hctx, &size, &mtime);
+  if (r == -ENOENT) {
+    return 0;
+  } else if (r < 0) {
+    CLS_ERR("failed to stat object: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (size >= soft_max_size) {
+    CLS_LOG(5, "journal object full: %" PRIu64 " >= %" PRIu64,
+            size, soft_max_size);
+    return -EOVERFLOW;
+  }
+  return 0;
+}
+
+CLS_INIT(journal)
+{
+  CLS_LOG(20, "Loaded journal class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_journal_create;
+  cls_method_handle_t h_journal_get_order;
+  cls_method_handle_t h_journal_get_splay_width;
+  cls_method_handle_t h_journal_get_pool_id;
+  cls_method_handle_t h_journal_get_minimum_set;
+  cls_method_handle_t h_journal_set_minimum_set;
+  cls_method_handle_t h_journal_get_active_set;
+  cls_method_handle_t h_journal_set_active_set;
+  cls_method_handle_t h_journal_get_client;
+  cls_method_handle_t h_journal_client_register;
+  cls_method_handle_t h_journal_client_update_data;
+  cls_method_handle_t h_journal_client_update_state;
+  cls_method_handle_t h_journal_client_unregister;
+  cls_method_handle_t h_journal_client_commit;
+  cls_method_handle_t h_journal_client_list;
+  cls_method_handle_t h_journal_get_next_tag_tid;
+  cls_method_handle_t h_journal_get_tag;
+  cls_method_handle_t h_journal_tag_create;
+  cls_method_handle_t h_journal_tag_list;
+  cls_method_handle_t h_journal_object_guard_append;
+
+  cls_register("journal", &h_class);
+
+  /// methods for journal.$journal_id objects
+  cls_register_cxx_method(h_class, "create",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_create, &h_journal_create);
+  cls_register_cxx_method(h_class, "get_order",
+                          CLS_METHOD_RD,
+                          journal_get_order, &h_journal_get_order);
+  cls_register_cxx_method(h_class, "get_splay_width",
+                          CLS_METHOD_RD,
+                          journal_get_splay_width, &h_journal_get_splay_width);
+  cls_register_cxx_method(h_class, "get_pool_id",
+                          CLS_METHOD_RD,
+                          journal_get_pool_id, &h_journal_get_pool_id);
+  cls_register_cxx_method(h_class, "get_minimum_set",
+                          CLS_METHOD_RD,
+                          journal_get_minimum_set,
+                          &h_journal_get_minimum_set);
+  cls_register_cxx_method(h_class, "set_minimum_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_set_minimum_set,
+                          &h_journal_set_minimum_set);
+  cls_register_cxx_method(h_class, "get_active_set",
+                          CLS_METHOD_RD,
+                          journal_get_active_set,
+                          &h_journal_get_active_set);
+  cls_register_cxx_method(h_class, "set_active_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_set_active_set,
+                          &h_journal_set_active_set);
+
+  cls_register_cxx_method(h_class, "get_client",
+                          CLS_METHOD_RD,
+                          journal_get_client, &h_journal_get_client);
+  cls_register_cxx_method(h_class, "client_register",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_client_register, &h_journal_client_register);
+  cls_register_cxx_method(h_class, "client_update_data",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_client_update_data,
+                          &h_journal_client_update_data);
+  cls_register_cxx_method(h_class, "client_update_state",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_client_update_state,
+                          &h_journal_client_update_state);
+  cls_register_cxx_method(h_class, "client_unregister",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_client_unregister,
+                          &h_journal_client_unregister);
+  cls_register_cxx_method(h_class, "client_commit",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_client_commit, &h_journal_client_commit);
+  cls_register_cxx_method(h_class, "client_list",
+                          CLS_METHOD_RD,
+                          journal_client_list, &h_journal_client_list);
+
+  cls_register_cxx_method(h_class, "get_next_tag_tid",
+                          CLS_METHOD_RD,
+                          journal_get_next_tag_tid,
+                          &h_journal_get_next_tag_tid);
+  cls_register_cxx_method(h_class, "get_tag",
+                          CLS_METHOD_RD,
+                          journal_get_tag, &h_journal_get_tag);
+  cls_register_cxx_method(h_class, "tag_create",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_tag_create, &h_journal_tag_create);
+  cls_register_cxx_method(h_class, "tag_list",
+                          CLS_METHOD_RD,
+                          journal_tag_list, &h_journal_tag_list);
+
+  /// methods for journal_data.$journal_id.$object_id objects
+  cls_register_cxx_method(h_class, "guard_append",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          journal_object_guard_append,
+                          &h_journal_object_guard_append);
+}
diff --git a/src/cls/journal/cls_journal_client.cc b/src/cls/journal/cls_journal_client.cc
new file mode 100644
index 00000000..c22a32cf
--- /dev/null
+++ b/src/cls/journal/cls_journal_client.cc
@@ -0,0 +1,498 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/journal/cls_journal_client.h"
+#include "include/rados/librados.hpp"
+#include "include/buffer.h"
+#include "include/Context.h"
+#include "common/Cond.h"
+#include <errno.h>
+
+namespace cls {
+namespace journal {
+namespace client {
+using ceph::encode;
+using ceph::decode;
+
+namespace {
+
+struct C_AioExec : public Context {
+  librados::IoCtx &ioctx;
+  std::string oid;
+
+  C_AioExec(librados::IoCtx &_ioctx, const std::string &_oid)
+    : ioctx(_ioctx), oid(_oid) {
+  }
+
+  static void rados_callback(rados_completion_t c, void *arg) {
+    Context *ctx = reinterpret_cast<Context *>(arg);
+    ctx->complete(rados_aio_get_return_value(c));
+  }
+};
+
+struct C_ClientList : public C_AioExec {
+  std::set<cls::journal::Client> *clients;
+  Context *on_finish;
+  bufferlist outbl;
+
+  C_ClientList(librados::IoCtx &_ioctx, const std::string &_oid,
+               std::set<cls::journal::Client> *_clients,
+               Context *_on_finish)
+    : C_AioExec(_ioctx, _oid), clients(_clients), on_finish(_on_finish) {}
+
+  void send(const std::string &start_after) {
+    bufferlist inbl;
+    encode(start_after, inbl);
+    encode(JOURNAL_MAX_RETURN, inbl);
+
+    librados::ObjectReadOperation op;
+    op.exec("journal", "client_list", inbl);
+
+    outbl.clear();
+    librados::AioCompletion *rados_completion =
+       librados::Rados::aio_create_completion(this, rados_callback, NULL);
+    int r = ioctx.aio_operate(oid, rados_completion, &op, &outbl);
+    ceph_assert(r == 0);
+    rados_completion->release();
+  }
+
+  void complete(int r) override {
+    if (r < 0) {
+      finish(r);
+      return;
+    }
+
+    try {
+      auto iter = outbl.cbegin();
+      std::set<cls::journal::Client> partial_clients;
+      decode(partial_clients, iter);
+
+      std::string start_after;
+      if (!partial_clients.empty()) {
+        start_after = partial_clients.rbegin()->id;
+        clients->insert(partial_clients.begin(), partial_clients.end());
+      }
+
+      if (partial_clients.size() < JOURNAL_MAX_RETURN) {
+        finish(0);
+      } else {
+        send(start_after);
+      }
+    } catch (const buffer::error &err) {
+      finish(-EBADMSG);
+    }
+  }
+
+  void finish(int r) override {
+    on_finish->complete(r);
+    delete this;
+  }
+};
+
+struct C_ImmutableMetadata : public C_AioExec {
+  uint8_t *order;
+  uint8_t *splay_width;
+  int64_t *pool_id;
+  Context *on_finish;
+  bufferlist outbl;
+
+  C_ImmutableMetadata(librados::IoCtx &_ioctx, const std::string &_oid,
+                      uint8_t *_order, uint8_t *_splay_width,
+		      int64_t *_pool_id, Context *_on_finish)
+    : C_AioExec(_ioctx, _oid), order(_order), splay_width(_splay_width),
+      pool_id(_pool_id), on_finish(_on_finish) {
+  }
+
+  void send() {
+    librados::ObjectReadOperation op;
+    bufferlist inbl;
+    op.exec("journal", "get_order", inbl);
+    op.exec("journal", "get_splay_width", inbl);
+    op.exec("journal", "get_pool_id", inbl);
+
+    librados::AioCompletion *rados_completion =
+      librados::Rados::aio_create_completion(this, rados_callback, NULL);
+    int r = ioctx.aio_operate(oid, rados_completion, &op, &outbl);
+    ceph_assert(r == 0);
+    rados_completion->release();
+  }
+
+  void finish(int r) override {
+    if (r == 0) {
+      try {
+        auto iter = outbl.cbegin();
+        decode(*order, iter);
+        decode(*splay_width, iter);
+        decode(*pool_id, iter);
+      } catch (const buffer::error &err) {
+        r = -EBADMSG;
+      }
+    }
+    on_finish->complete(r);
+  }
+};
+
+struct C_MutableMetadata : public C_AioExec {
+  uint64_t *minimum_set;
+  uint64_t *active_set;
+  C_ClientList *client_list;
+  bufferlist outbl;
+
+  C_MutableMetadata(librados::IoCtx &_ioctx, const std::string &_oid,
+                    uint64_t *_minimum_set, uint64_t *_active_set,
+                    C_ClientList *_client_list)
+    : C_AioExec(_ioctx, _oid), minimum_set(_minimum_set),
+      active_set(_active_set), client_list(_client_list) {}
+
+  void send() {
+    librados::ObjectReadOperation op;
+    bufferlist inbl;
+    op.exec("journal", "get_minimum_set", inbl);
+    op.exec("journal", "get_active_set", inbl);
+
+    librados::AioCompletion *rados_completion =
+      librados::Rados::aio_create_completion(this, rados_callback, NULL);
+    int r = ioctx.aio_operate(oid, rados_completion, &op, &outbl);
+    ceph_assert(r == 0);
+    rados_completion->release();
+  }
+
+  void finish(int r) override {
+    if (r == 0) {
+      try {
+        auto iter = outbl.cbegin();
+        decode(*minimum_set, iter);
+        decode(*active_set, iter);
+        client_list->send("");
+      } catch (const buffer::error &err) {
+        r = -EBADMSG;
+      }
+    }
+    if (r < 0) {
+      client_list->complete(r);
+    }
+  }
+};
+
+
+} // anonymous namespace
+
+void create(librados::ObjectWriteOperation *op,
+            uint8_t order, uint8_t splay, int64_t pool_id) {
+  bufferlist bl;
+  encode(order, bl);
+  encode(splay, bl);
+  encode(pool_id, bl);
+
+  op->exec("journal", "create", bl);
+}
+
+int create(librados::IoCtx &ioctx, const std::string &oid, uint8_t order,
+           uint8_t splay, int64_t pool_id) {
+  librados::ObjectWriteOperation op;
+  create(&op, order, splay, pool_id);
+
+  int r = ioctx.operate(oid, &op);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void get_immutable_metadata(librados::IoCtx &ioctx, const std::string &oid,
+                            uint8_t *order, uint8_t *splay_width,
+                            int64_t *pool_id, Context *on_finish) {
+  C_ImmutableMetadata *metadata = new C_ImmutableMetadata(ioctx, oid, order,
+                                                          splay_width, pool_id,
+                                                          on_finish);
+  metadata->send();
+}
+
+void get_mutable_metadata(librados::IoCtx &ioctx, const std::string &oid,
+                          uint64_t *minimum_set, uint64_t *active_set,
+                          std::set<cls::journal::Client> *clients,
+                          Context *on_finish) {
+  C_ClientList *client_list = new C_ClientList(ioctx, oid, clients, on_finish);
+  C_MutableMetadata *metadata = new C_MutableMetadata(
+    ioctx, oid, minimum_set, active_set, client_list);
+  metadata->send();
+}
+
+void set_minimum_set(librados::ObjectWriteOperation *op, uint64_t object_set) {
+  bufferlist bl;
+  encode(object_set, bl);
+  op->exec("journal", "set_minimum_set", bl);
+}
+
+void set_active_set(librados::ObjectWriteOperation *op, uint64_t object_set) {
+  bufferlist bl;
+  encode(object_set, bl);
+  op->exec("journal", "set_active_set", bl);
+}
+
+int get_client(librados::IoCtx &ioctx, const std::string &oid,
+               const std::string &id, cls::journal::Client *client) {
+  librados::ObjectReadOperation op;
+  get_client_start(&op, id);
+
+  bufferlist out_bl;
+  int r = ioctx.operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = get_client_finish(&iter, client);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void get_client_start(librados::ObjectReadOperation *op,
+                      const std::string &id) {
+  bufferlist bl;
+  encode(id, bl);
+  op->exec("journal", "get_client", bl);
+}
+
+int get_client_finish(bufferlist::const_iterator *iter,
+                      cls::journal::Client *client) {
+  try {
+    decode(*client, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int client_register(librados::IoCtx &ioctx, const std::string &oid,
+                    const std::string &id, const bufferlist &data) {
+  librados::ObjectWriteOperation op;
+  client_register(&op, id, data);
+  return ioctx.operate(oid, &op);
+}
+
+void client_register(librados::ObjectWriteOperation *op,
+                     const std::string &id, const bufferlist &data) {
+  bufferlist bl;
+  encode(id, bl);
+  encode(data, bl);
+  op->exec("journal", "client_register", bl);
+}
+
+int client_update_data(librados::IoCtx &ioctx, const std::string &oid,
+                       const std::string &id, const bufferlist &data) {
+  librados::ObjectWriteOperation op;
+  client_update_data(&op, id, data);
+  return ioctx.operate(oid, &op);
+}
+
+void client_update_data(librados::ObjectWriteOperation *op,
+                        const std::string &id, const bufferlist &data) {
+  bufferlist bl;
+  encode(id, bl);
+  encode(data, bl);
+  op->exec("journal", "client_update_data", bl);
+}
+
+int client_update_state(librados::IoCtx &ioctx, const std::string &oid,
+                        const std::string &id, cls::journal::ClientState state) {
+  librados::ObjectWriteOperation op;
+  client_update_state(&op, id, state);
+  return ioctx.operate(oid, &op);
+}
+
+void client_update_state(librados::ObjectWriteOperation *op,
+                         const std::string &id,
+                         cls::journal::ClientState state) {
+  bufferlist bl;
+  encode(id, bl);
+  encode(static_cast<uint8_t>(state), bl);
+  op->exec("journal", "client_update_state", bl);
+}
+
+int client_unregister(librados::IoCtx &ioctx, const std::string &oid,
+                       const std::string &id) {
+  librados::ObjectWriteOperation op;
+  client_unregister(&op, id);
+  return ioctx.operate(oid, &op);
+}
+
+void client_unregister(librados::ObjectWriteOperation *op,
+		       const std::string &id) {
+
+  bufferlist bl;
+  encode(id, bl);
+  op->exec("journal", "client_unregister", bl);
+}
+
+void client_commit(librados::ObjectWriteOperation *op, const std::string &id,
+                   const cls::journal::ObjectSetPosition &commit_position) {
+  bufferlist bl;
+  encode(id, bl);
+  encode(commit_position, bl);
+  op->exec("journal", "client_commit", bl);
+}
+
+int client_list(librados::IoCtx &ioctx, const std::string &oid,
+                std::set<cls::journal::Client> *clients) {
+  C_SaferCond cond;
+  client_list(ioctx, oid, clients, &cond);
+  return cond.wait();
+}
+
+void client_list(librados::IoCtx &ioctx, const std::string &oid,
+                 std::set<cls::journal::Client> *clients, Context *on_finish) {
+  C_ClientList *client_list = new C_ClientList(ioctx, oid, clients, on_finish);
+  client_list->send("");
+}
+
+int get_next_tag_tid(librados::IoCtx &ioctx, const std::string &oid,
+                     uint64_t *tag_tid) {
+  librados::ObjectReadOperation op;
+  get_next_tag_tid_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx.operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = get_next_tag_tid_finish(&iter, tag_tid);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void get_next_tag_tid_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("journal", "get_next_tag_tid", bl);
+}
+
+int get_next_tag_tid_finish(bufferlist::const_iterator *iter,
+                            uint64_t *tag_tid) {
+  try {
+    decode(*tag_tid, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_tag(librados::IoCtx &ioctx, const std::string &oid,
+            uint64_t tag_tid, cls::journal::Tag *tag) {
+  librados::ObjectReadOperation op;
+  get_tag_start(&op, tag_tid);
+
+  bufferlist out_bl;
+  int r = ioctx.operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = get_tag_finish(&iter, tag);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void get_tag_start(librados::ObjectReadOperation *op,
+                   uint64_t tag_tid) {
+  bufferlist bl;
+  encode(tag_tid, bl);
+  op->exec("journal", "get_tag", bl);
+}
+
+int get_tag_finish(bufferlist::const_iterator *iter, cls::journal::Tag *tag) {
+  try {
+    decode(*tag, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int tag_create(librados::IoCtx &ioctx, const std::string &oid,
+               uint64_t tag_tid, uint64_t tag_class,
+               const bufferlist &data) {
+  librados::ObjectWriteOperation op;
+  tag_create(&op, tag_tid, tag_class, data);
+  return ioctx.operate(oid, &op);
+}
+
+void tag_create(librados::ObjectWriteOperation *op, uint64_t tag_tid,
+                uint64_t tag_class, const bufferlist &data) {
+  bufferlist bl;
+  encode(tag_tid, bl);
+  encode(tag_class, bl);
+  encode(data, bl);
+  op->exec("journal", "tag_create", bl);
+}
+
+int tag_list(librados::IoCtx &ioctx, const std::string &oid,
+             const std::string &client_id, boost::optional<uint64_t> tag_class,
+             std::set<cls::journal::Tag> *tags) {
+  tags->clear();
+  uint64_t start_after_tag_tid = 0;
+  while (true) {
+    librados::ObjectReadOperation op;
+    tag_list_start(&op, start_after_tag_tid, JOURNAL_MAX_RETURN, client_id,
+                   tag_class);
+
+    bufferlist out_bl;
+    int r = ioctx.operate(oid, &op, &out_bl);
+    if (r < 0) {
+      return r;
+    }
+
+    auto iter = out_bl.cbegin();
+    std::set<cls::journal::Tag> decode_tags;
+    r = tag_list_finish(&iter, &decode_tags);
+    if (r < 0) {
+      return r;
+    }
+
+    tags->insert(decode_tags.begin(), decode_tags.end());
+    if (decode_tags.size() < JOURNAL_MAX_RETURN) {
+      break;
+    }
+  }
+  return 0;
+}
+
+void tag_list_start(librados::ObjectReadOperation *op,
+                    uint64_t start_after_tag_tid, uint64_t max_return,
+                    const std::string &client_id,
+                    boost::optional<uint64_t> tag_class) {
+  bufferlist bl;
+  encode(start_after_tag_tid, bl);
+  encode(max_return, bl);
+  encode(client_id, bl);
+  encode(tag_class, bl);
+  op->exec("journal", "tag_list", bl);
+}
+
+int tag_list_finish(bufferlist::const_iterator *iter,
+                    std::set<cls::journal::Tag> *tags) {
+  try {
+    decode(*tags, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+void guard_append(librados::ObjectWriteOperation *op, uint64_t soft_max_size) {
+  bufferlist bl;
+  encode(soft_max_size, bl);
+  op->exec("journal", "guard_append", bl);
+}
+
+} // namespace client
+} // namespace journal
+} // namespace cls
diff --git a/src/cls/journal/cls_journal_client.h b/src/cls/journal/cls_journal_client.h
new file mode 100644
index 00000000..50579cfc
--- /dev/null
+++ b/src/cls/journal/cls_journal_client.h
@@ -0,0 +1,107 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_JOURNAL_CLIENT_H
+#define CEPH_CLS_JOURNAL_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include "cls/journal/cls_journal_types.h"
+#include <set>
+#include <boost/optional.hpp>
+
+class Context;
+
+namespace cls {
+namespace journal {
+namespace client {
+
+void create(librados::ObjectWriteOperation *op,
+            uint8_t order, uint8_t splay, int64_t pool_id);
+int create(librados::IoCtx &ioctx, const std::string &oid, uint8_t order,
+           uint8_t splay, int64_t pool_id);
+
+void get_immutable_metadata(librados::IoCtx &ioctx, const std::string &oid,
+                            uint8_t *order, uint8_t *splay_width,
+			    int64_t *pool_id, Context *on_finish);
+void get_mutable_metadata(librados::IoCtx &ioctx, const std::string &oid,
+                          uint64_t *minimum_set, uint64_t *active_set,
+                          std::set<cls::journal::Client> *clients,
+                          Context *on_finish);
+
+void set_minimum_set(librados::ObjectWriteOperation *op, uint64_t object_set);
+void set_active_set(librados::ObjectWriteOperation *op, uint64_t object_set);
+
+// journal client helpers
+int get_client(librados::IoCtx &ioctx, const std::string &oid,
+               const std::string &id, cls::journal::Client *client);
+void get_client_start(librados::ObjectReadOperation *op,
+                      const std::string &id);
+int get_client_finish(bufferlist::const_iterator *iter,
+                      cls::journal::Client *client);
+
+int client_register(librados::IoCtx &ioctx, const std::string &oid,
+                    const std::string &id, const bufferlist &data);
+void client_register(librados::ObjectWriteOperation *op,
+                     const std::string &id, const bufferlist &data);
+
+int client_update_data(librados::IoCtx &ioctx, const std::string &oid,
+                       const std::string &id, const bufferlist &data);
+void client_update_data(librados::ObjectWriteOperation *op,
+                        const std::string &id, const bufferlist &data);
+int client_update_state(librados::IoCtx &ioctx, const std::string &oid,
+                        const std::string &id, cls::journal::ClientState state);
+void client_update_state(librados::ObjectWriteOperation *op,
+                         const std::string &id,
+                         cls::journal::ClientState state);
+
+int client_unregister(librados::IoCtx &ioctx, const std::string &oid,
+                      const std::string &id);
+void client_unregister(librados::ObjectWriteOperation *op,
+		       const std::string &id);
+
+void client_commit(librados::ObjectWriteOperation *op, const std::string &id,
+                   const cls::journal::ObjectSetPosition &commit_position);
+
+int client_list(librados::IoCtx &ioctx, const std::string &oid,
+                std::set<cls::journal::Client> *clients);
+void client_list(librados::IoCtx &ioctx, const std::string &oid,
+                 std::set<cls::journal::Client> *clients, Context *on_finish);
+
+// journal tag helpers
+int get_next_tag_tid(librados::IoCtx &ioctx, const std::string &oid,
+                     uint64_t *tag_tid);
+void get_next_tag_tid_start(librados::ObjectReadOperation *op);
+int get_next_tag_tid_finish(bufferlist::const_iterator *iter,
+                            uint64_t *tag_tid);
+
+int get_tag(librados::IoCtx &ioctx, const std::string &oid,
+            uint64_t tag_tid, cls::journal::Tag *tag);
+void get_tag_start(librados::ObjectReadOperation *op,
+                   uint64_t tag_tid);
+int get_tag_finish(bufferlist::const_iterator *iter, cls::journal::Tag *tag);
+
+int tag_create(librados::IoCtx &ioctx, const std::string &oid,
+               uint64_t tag_tid, uint64_t tag_class,
+               const bufferlist &data);
+void tag_create(librados::ObjectWriteOperation *op,
+                uint64_t tag_tid, uint64_t tag_class,
+                const bufferlist &data);
+
+int tag_list(librados::IoCtx &ioctx, const std::string &oid,
+             const std::string &client_id, boost::optional<uint64_t> tag_class,
+             std::set<cls::journal::Tag> *tags);
+void tag_list_start(librados::ObjectReadOperation *op,
+                    uint64_t start_after_tag_tid, uint64_t max_return,
+                    const std::string &client_id,
+                    boost::optional<uint64_t> tag_class);
+int tag_list_finish(bufferlist::const_iterator *iter,
+                    std::set<cls::journal::Tag> *tags);
+
+// journal entry helpers
+void guard_append(librados::ObjectWriteOperation *op, uint64_t soft_max_size);
+
+} // namespace client
+} // namespace journal
+} // namespace cls
+
+#endif // CEPH_CLS_JOURNAL_CLIENT_H
diff --git a/src/cls/journal/cls_journal_types.cc b/src/cls/journal/cls_journal_types.cc
new file mode 100644
index 00000000..6976304d
--- /dev/null
+++ b/src/cls/journal/cls_journal_types.cc
@@ -0,0 +1,196 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/journal/cls_journal_types.h"
+#include "include/stringify.h"
+#include "common/Formatter.h"
+
+namespace cls {
+namespace journal {
+
+void ObjectPosition::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(object_number, bl);
+  encode(tag_tid, bl);
+  encode(entry_tid, bl);
+  ENCODE_FINISH(bl);
+}
+
+void ObjectPosition::decode(bufferlist::const_iterator& iter) {
+  DECODE_START(1, iter);
+  decode(object_number, iter);
+  decode(tag_tid, iter);
+  decode(entry_tid, iter);
+  DECODE_FINISH(iter);
+}
+
+void ObjectPosition::dump(Formatter *f) const {
+  f->dump_unsigned("object_number", object_number);
+  f->dump_unsigned("tag_tid", tag_tid);
+  f->dump_unsigned("entry_tid", entry_tid);
+}
+
+void ObjectPosition::generate_test_instances(std::list<ObjectPosition *> &o) {
+  o.push_back(new ObjectPosition());
+  o.push_back(new ObjectPosition(1, 2, 3));
+}
+
+void ObjectSetPosition::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(object_positions, bl);
+  ENCODE_FINISH(bl);
+}
+
+void ObjectSetPosition::decode(bufferlist::const_iterator& iter) {
+  DECODE_START(1, iter);
+  decode(object_positions, iter);
+  DECODE_FINISH(iter);
+}
+
+void ObjectSetPosition::dump(Formatter *f) const {
+  f->open_array_section("object_positions");
+  for (auto &pos : object_positions) {
+    f->open_object_section("object_position");
+    pos.dump(f);
+    f->close_section();
+  }
+  f->close_section();
+}
+
+void ObjectSetPosition::generate_test_instances(
+    std::list<ObjectSetPosition *> &o) {
+  o.push_back(new ObjectSetPosition());
+  o.push_back(new ObjectSetPosition({{0, 1, 120}, {121, 2, 121}}));
+}
+
+void Client::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(id, bl);
+  encode(data, bl);
+  encode(commit_position, bl);
+  encode(static_cast<uint8_t>(state), bl);
+  ENCODE_FINISH(bl);
+}
+
+void Client::decode(bufferlist::const_iterator& iter) {
+  DECODE_START(1, iter);
+  decode(id, iter);
+  decode(data, iter);
+  decode(commit_position, iter);
+
+  uint8_t state_raw;
+  decode(state_raw, iter);
+  state = static_cast<ClientState>(state_raw);
+  DECODE_FINISH(iter);
+}
+
+void Client::dump(Formatter *f) const {
+  f->dump_string("id", id);
+
+  std::stringstream data_ss;
+  data.hexdump(data_ss);
+  f->dump_string("data", data_ss.str());
+
+  f->open_object_section("commit_position");
+  commit_position.dump(f);
+  f->close_section();
+
+  f->dump_string("state", stringify(state));
+}
+
+void Client::generate_test_instances(std::list<Client *> &o) {
+  bufferlist data;
+  data.append(std::string(128, '1'));
+
+  o.push_back(new Client());
+  o.push_back(new Client("id", data));
+  o.push_back(new Client("id", data, {{{1, 2, 120}, {2, 3, 121}}}));
+}
+
+void Tag::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(tid, bl);
+  encode(tag_class, bl);
+  encode(data, bl);
+  ENCODE_FINISH(bl);
+}
+
+void Tag::decode(bufferlist::const_iterator& iter) {
+  DECODE_START(1, iter);
+  decode(tid, iter);
+  decode(tag_class, iter);
+  decode(data, iter);
+  DECODE_FINISH(iter);
+}
+
+void Tag::dump(Formatter *f) const {
+  f->dump_unsigned("tid", tid);
+  f->dump_unsigned("tag_class", tag_class);
+
+  std::stringstream data_ss;
+  data.hexdump(data_ss);
+  f->dump_string("data", data_ss.str());
+}
+
+void Tag::generate_test_instances(std::list<Tag *> &o) {
+  o.push_back(new Tag());
+
+  bufferlist data;
+  data.append(std::string(128, '1'));
+  o.push_back(new Tag(123, 234, data));
+}
+
+std::ostream &operator<<(std::ostream &os, const ClientState &state) {
+  switch (state) {
+  case CLIENT_STATE_CONNECTED:
+    os << "connected";
+    break;
+  case CLIENT_STATE_DISCONNECTED:
+    os << "disconnected";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(state) << ")";
+    break;
+  }
+  return os;
+}
+
+std::ostream &operator<<(std::ostream &os,
+                         const ObjectPosition &object_position) {
+  os << "["
+     << "object_number=" << object_position.object_number << ", "
+     << "tag_tid=" << object_position.tag_tid << ", "
+     << "entry_tid=" << object_position.entry_tid << "]";
+  return os;
+}
+
+std::ostream &operator<<(std::ostream &os,
+                         const ObjectSetPosition &object_set_position) {
+  os << "[positions=[";
+  std::string delim;
+  for (auto &object_position : object_set_position.object_positions) {
+    os << delim << object_position;
+    delim = ", ";
+  }
+  os << "]]";
+  return os;
+}
+
+std::ostream &operator<<(std::ostream &os, const Client &client) {
+  os << "[id=" << client.id << ", "
+     << "commit_position=" << client.commit_position << ", "
+     << "state=" << client.state << "]";
+  return os;
+}
+
+std::ostream &operator<<(std::ostream &os, const Tag &tag) {
+  os << "[tid=" << tag.tid << ", "
+     << "tag_class=" << tag.tag_class << ", "
+     << "data=";
+  tag.data.hexdump(os);
+  os << "]";
+  return os;
+}
+
+} // namespace journal
+} // namespace cls
diff --git a/src/cls/journal/cls_journal_types.h b/src/cls/journal/cls_journal_types.h
new file mode 100644
index 00000000..2a617698
--- /dev/null
+++ b/src/cls/journal/cls_journal_types.h
@@ -0,0 +1,154 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_JOURNAL_TYPES_H
+#define CEPH_CLS_JOURNAL_TYPES_H
+
+#include "include/int_types.h"
+#include "include/buffer_fwd.h"
+#include "include/encoding.h"
+#include <iosfwd>
+#include <list>
+#include <string>
+
+namespace ceph {
+class Formatter;
+}
+
+namespace cls {
+namespace journal {
+
+static const uint64_t JOURNAL_MAX_RETURN = 256;
+
+struct ObjectPosition {
+  uint64_t object_number;
+  uint64_t tag_tid;
+  uint64_t entry_tid;
+
+  ObjectPosition() : object_number(0), tag_tid(0), entry_tid(0) {}
+  ObjectPosition(uint64_t _object_number, uint64_t _tag_tid,
+                 uint64_t _entry_tid)
+    : object_number(_object_number), tag_tid(_tag_tid), entry_tid(_entry_tid) {}
+
+  inline bool operator==(const ObjectPosition& rhs) const {
+    return (object_number == rhs.object_number &&
+            tag_tid == rhs.tag_tid &&
+            entry_tid == rhs.entry_tid);
+  }
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& iter);
+  void dump(Formatter *f) const;
+
+  inline bool operator<(const ObjectPosition &rhs) const {
+    if (object_number != rhs.object_number) {
+      return object_number < rhs.object_number;
+    } else if (tag_tid != rhs.tag_tid) {
+      return tag_tid < rhs.tag_tid;
+    }
+    return entry_tid < rhs.entry_tid;
+  }
+
+  static void generate_test_instances(std::list<ObjectPosition *> &o);
+};
+
+typedef std::list<ObjectPosition> ObjectPositions;
+
+struct ObjectSetPosition {
+  // stored in most-recent -> least recent committed entry order
+  ObjectPositions object_positions;
+
+  ObjectSetPosition() {}
+  ObjectSetPosition(const ObjectPositions &_object_positions)
+    : object_positions(_object_positions) {}
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& iter);
+  void dump(Formatter *f) const;
+
+  inline bool operator==(const ObjectSetPosition &rhs) const {
+    return (object_positions == rhs.object_positions);
+  }
+
+  static void generate_test_instances(std::list<ObjectSetPosition *> &o);
+};
+
+enum ClientState {
+  CLIENT_STATE_CONNECTED = 0,
+  CLIENT_STATE_DISCONNECTED = 1
+};
+
+struct Client {
+  std::string id;
+  bufferlist data;
+  ObjectSetPosition commit_position;
+  ClientState state;
+
+  Client() : state(CLIENT_STATE_CONNECTED) {}
+  Client(const std::string& _id, const bufferlist &_data,
+         const ObjectSetPosition &_commit_position = ObjectSetPosition(),
+         ClientState _state = CLIENT_STATE_CONNECTED)
+    : id(_id), data(_data), commit_position(_commit_position), state(_state) {}
+
+  inline bool operator==(const Client &rhs) const {
+    return (id == rhs.id &&
+            data.contents_equal(rhs.data) &&
+            commit_position == rhs.commit_position &&
+            state == rhs.state);
+  }
+  inline bool operator<(const Client &rhs) const {
+    return (id < rhs.id);
+  }
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& iter);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<Client *> &o);
+};
+
+struct Tag {
+  static const uint64_t TAG_CLASS_NEW = static_cast<uint64_t>(-1);
+
+  uint64_t tid;
+  uint64_t tag_class;
+  bufferlist data;
+
+  Tag() : tid(0), tag_class(0) {}
+  Tag(uint64_t tid, uint64_t tag_class, const bufferlist &data)
+    : tid(tid), tag_class(tag_class), data(data) {}
+
+  inline bool operator==(const Tag &rhs) const {
+    return (tid == rhs.tid &&
+            tag_class == rhs.tag_class &&
+            data.contents_equal(rhs.data));
+  }
+  inline bool operator<(const Tag &rhs) const {
+    return (tid < rhs.tid);
+  }
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& iter);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<Tag *> &o);
+};
+
+WRITE_CLASS_ENCODER(ObjectPosition);
+WRITE_CLASS_ENCODER(ObjectSetPosition);
+WRITE_CLASS_ENCODER(Client);
+WRITE_CLASS_ENCODER(Tag);
+
+std::ostream &operator<<(std::ostream &os, const ClientState &state);
+std::ostream &operator<<(std::ostream &os,
+                         const ObjectPosition &object_position);
+std::ostream &operator<<(std::ostream &os,
+                         const ObjectSetPosition &object_set_position);
+std::ostream &operator<<(std::ostream &os,
+			 const Client &client);
+std::ostream &operator<<(std::ostream &os, const Tag &tag);
+
+} // namespace journal
+} // namespace cls
+
+#endif // CEPH_CLS_JOURNAL_TYPES_H
diff --git a/src/cls/lock/cls_lock.cc b/src/cls/lock/cls_lock.cc
new file mode 100644
index 00000000..74880225
--- /dev/null
+++ b/src/cls/lock/cls_lock.cc
@@ -0,0 +1,647 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+/** \file
+ *
+ * This is an OSD class that implements methods for object
+ * advisory locking.
+ *
+ */
+
+#include <errno.h>
+#include <map>
+#include <sstream>
+
+#include "include/types.h"
+#include "include/utime.h"
+#include "objclass/objclass.h"
+
+#include "common/errno.h"
+#include "common/Clock.h"
+
+#include "cls/lock/cls_lock_types.h"
+#include "cls/lock/cls_lock_ops.h"
+
+#include "global/global_context.h"
+
+#include "include/compat.h"
+
+
+using namespace rados::cls::lock;
+
+
+CLS_VER(1,0)
+CLS_NAME(lock)
+
+#define LOCK_PREFIX    "lock."
+
+static int clean_lock(cls_method_context_t hctx)
+{
+  int r = cls_cxx_remove(hctx);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+static int read_lock(cls_method_context_t hctx,
+		     const string& name,
+		     lock_info_t *lock)
+{
+  bufferlist bl;
+  string key = LOCK_PREFIX;
+  key.append(name);
+ 
+  int r = cls_cxx_getxattr(hctx, key.c_str(), &bl);
+  if (r < 0) {
+    if (r ==  -ENODATA) {
+      *lock = lock_info_t();
+      return 0;
+    }
+    if (r != -ENOENT) {
+      CLS_ERR("error reading xattr %s: %d", key.c_str(), r);
+    }
+    return r;
+  }
+
+  try {
+    auto it = bl.cbegin();
+    decode(*lock, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("error decoding %s", key.c_str());
+    return -EIO;
+  }
+
+  /* now trim expired locks */
+
+  utime_t now = ceph_clock_now();
+
+  map<locker_id_t, locker_info_t>::iterator iter = lock->lockers.begin();
+
+  while (iter != lock->lockers.end()) {
+    struct locker_info_t& info = iter->second;
+    if (!info.expiration.is_zero() && info.expiration < now) {
+      CLS_LOG(20, "expiring locker");
+      iter = lock->lockers.erase(iter);
+    } else {
+      ++iter;
+    }
+  }
+
+  if (lock->lockers.empty() && cls_lock_is_ephemeral(lock->lock_type)) {
+    r = clean_lock(hctx);
+    if (r < 0) {
+      CLS_ERR("error, on read, cleaning lock object %s", cpp_strerror(r).c_str());
+    }
+  }
+
+  return 0;
+}
+
+static int write_lock(cls_method_context_t hctx, const string& name, const lock_info_t& lock)
+{
+  using ceph::encode;
+  string key = LOCK_PREFIX;
+  key.append(name);
+
+  bufferlist lock_bl;
+  encode(lock, lock_bl, cls_get_client_features(hctx));
+
+  int r = cls_cxx_setxattr(hctx, key.c_str(), &lock_bl);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+/**
+ * helper function to add a lock and update disk state.
+ *
+ * Input:
+ * @param name Lock name
+ * @param lock_type Type of lock (exclusive / shared)
+ * @param duration Duration of lock (in seconds). Zero means it doesn't expire.
+ * @param flags lock flags
+ * @param cookie The cookie to set in the lock
+ * @param tag The tag to match with the lock (can only lock with matching tags)
+ * @param lock_description The lock description to set (if not empty)
+ * @param locker_description The locker description
+ *
+ * @return 0 on success, or -errno on failure
+ */
+static int lock_obj(cls_method_context_t hctx,
+                    const string& name,
+                    ClsLockType lock_type,
+                    utime_t duration,
+                    const string& description,
+                    uint8_t flags,
+                    const string& cookie,
+                    const string& tag)
+{
+  bool exclusive = cls_lock_is_exclusive(lock_type);
+  lock_info_t linfo;
+  bool fail_if_exists = (flags & LOCK_FLAG_MAY_RENEW) == 0;
+  bool fail_if_does_not_exist = flags & LOCK_FLAG_MUST_RENEW;
+
+  CLS_LOG(20,
+	  "requested lock_type=%s fail_if_exists=%d fail_if_does_not_exist=%d",
+	  cls_lock_type_str(lock_type), fail_if_exists, fail_if_does_not_exist);
+  if (!cls_lock_is_valid(lock_type)) {
+    return -EINVAL;
+  }
+
+  if (name.empty())
+    return -EINVAL;
+
+  if (!fail_if_exists && fail_if_does_not_exist) {
+    // at most one of LOCK_FLAG_MAY_RENEW and LOCK_FLAG_MUST_RENEW may
+    // be set since they have different implications if the lock does
+    // not already exist
+    return -EINVAL;
+  }
+
+  // see if there's already a locker
+  int r = read_lock(hctx, name, &linfo);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("Could not read lock info: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  map<locker_id_t, locker_info_t>& lockers = linfo.lockers;
+  map<locker_id_t, locker_info_t>::iterator iter;
+
+  locker_id_t id;
+  id.cookie = cookie;
+  entity_inst_t inst;
+  r = cls_get_request_origin(hctx, &inst);
+  id.locker = inst.name;
+  ceph_assert(r == 0);
+
+  /* check this early, before we check fail_if_exists, otherwise we might
+   * remove the locker entry and not check it later */
+  if (lockers.size() && tag != linfo.tag) {
+    CLS_LOG(20, "cannot take lock on object, conflicting tag");
+    return -EBUSY;
+  }
+
+  ClsLockType existing_lock_type = linfo.lock_type;
+  CLS_LOG(20, "existing_lock_type=%s", cls_lock_type_str(existing_lock_type));
+  iter = lockers.find(id);
+  if (iter != lockers.end()) {
+    if (fail_if_exists && !fail_if_does_not_exist) {
+      return -EEXIST;
+    } else {
+      lockers.erase(iter); // remove old entry
+    }
+  } else if (fail_if_does_not_exist) {
+    return -ENOENT;
+  }
+
+  if (!lockers.empty()) {
+    if (exclusive) {
+      auto locker_lister =
+	[&lockers]() -> std::string {
+	  std::stringstream locker_list;
+	  locker_list << lockers;
+	  return locker_list.str();
+	};
+      CLS_LOG(20, "could not exclusive-lock object, already locked by %s",
+	      locker_lister().c_str());
+      return -EBUSY;
+    }
+
+    if (existing_lock_type != lock_type) {
+      CLS_LOG(20, "cannot take lock on object, conflicting lock type");
+      return -EBUSY;
+    }
+  }
+
+  linfo.lock_type = lock_type;
+  linfo.tag = tag;
+  utime_t expiration;
+  if (!duration.is_zero()) {
+    expiration = ceph_clock_now();
+    expiration += duration;
+
+  }
+  // make all addrs of type legacy, because v2 clients speak v2 or v1,
+  // even depending on which OSD they are talking to, and the type
+  // isn't what uniquely identifies them.  also, storing a v1 addr
+  // here means that old clients who get this locker_info won't see an
+  // old "msgr2:" prefix.
+  inst.addr.set_type(entity_addr_t::TYPE_LEGACY);
+
+  struct locker_info_t info(expiration, inst.addr, description);
+
+  linfo.lockers[id] = info;
+
+  r = write_lock(hctx, name, linfo);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+/**
+ * Set an exclusive lock on an object for the activating client, if possible.
+ *
+ * Input:
+ * @param cls_lock_lock_op request input
+ *
+ * @returns 0 on success, -EINVAL if it can't decode the lock_cookie,
+ * -EBUSY if the object is already locked, or -errno on (unexpected) failure.
+ */
+static int lock_op(cls_method_context_t hctx,
+                   bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "lock_op");
+  cls_lock_lock_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  return lock_obj(hctx,
+                  op.name, op.type, op.duration, op.description,
+                  op.flags, op.cookie, op.tag);
+}
+
+/**
+ *  helper function to remove a lock from on disk and clean up state.
+ *
+ *  @param name The lock name
+ *  @param locker The locker entity name
+ *  @param cookie The user-defined cookie associated with the lock.
+ *
+ *  @return 0 on success, -ENOENT if there is no such lock (either
+ *  entity or cookie is wrong), or -errno on other error.
+ */
+static int remove_lock(cls_method_context_t hctx,
+		       const string& name,
+		       entity_name_t& locker,
+		       const string& cookie)
+{
+  // get current lockers
+  lock_info_t linfo;
+  int r = read_lock(hctx, name, &linfo);
+  if (r < 0) {
+    CLS_ERR("Could not read list of current lockers off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  map<locker_id_t, locker_info_t>& lockers = linfo.lockers;
+  struct locker_id_t id(locker, cookie);
+
+  // remove named locker from set
+  map<locker_id_t, locker_info_t>::iterator iter = lockers.find(id);
+  if (iter == lockers.end()) { // no such key
+    return -ENOENT;
+  }
+  lockers.erase(iter);
+
+  if (cls_lock_is_ephemeral(linfo.lock_type)) {
+    ceph_assert(lockers.empty());
+    r = clean_lock(hctx);
+  } else {
+    r = write_lock(hctx, name, linfo);
+  }
+
+  return r;
+}
+
+/**
+ * Unlock an object which the activating client currently has locked.
+ *
+ * Input:
+ * @param cls_lock_unlock_op request input
+ *
+ * @return 0 on success, -EINVAL if it can't decode the cookie, -ENOENT
+ * if there is no such lock (either entity or cookie is wrong), or
+ * -errno on other (unexpected) error.
+ */
+static int unlock_op(cls_method_context_t hctx,
+                     bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "unlock_op");
+  cls_lock_unlock_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error& err) {
+    return -EINVAL;
+  }
+
+  entity_inst_t inst;
+  int r = cls_get_request_origin(hctx, &inst);
+  ceph_assert(r == 0);
+  return remove_lock(hctx, op.name, inst.name, op.cookie);
+}
+
+/**
+ * Break the lock on an object held by any client.
+ *
+ * Input:
+ * @param cls_lock_break_op request input
+ *
+ * @return 0 on success, -EINVAL if it can't decode the locker and
+ * cookie, -ENOENT if there is no such lock (either entity or cookie
+ * is wrong), or -errno on other (unexpected) error.
+ */
+static int break_lock(cls_method_context_t hctx,
+		      bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "break_lock");
+  cls_lock_break_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error& err) {
+    return -EINVAL;
+  }
+
+  return remove_lock(hctx, op.name, op.locker, op.cookie);
+}
+
+
+/**
+ * Retrieve lock info: lockers, tag, exclusive
+ *
+ * Input:
+ * @param cls_lock_list_lockers_op request input
+ *
+ * Output:
+ * @param cls_lock_list_lockers_reply result
+ *
+ * @return 0 on success, -errno on failure.
+ */
+static int get_info(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_info");
+  cls_lock_get_info_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error& err) {
+    return -EINVAL;
+  }
+
+  // get current lockers
+  lock_info_t linfo;
+  int r = read_lock(hctx, op.name, &linfo);
+  if (r < 0) {
+    CLS_ERR("Could not read lock info: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  struct cls_lock_get_info_reply ret;
+
+  map<locker_id_t, locker_info_t>::iterator iter;
+  for (iter = linfo.lockers.begin(); iter != linfo.lockers.end(); ++iter) {
+    ret.lockers[iter->first] = iter->second;
+  }
+  ret.lock_type = linfo.lock_type;
+  ret.tag = linfo.tag;
+
+  encode(ret, *out, cls_get_client_features(hctx));
+
+  return 0;
+}
+
+
+/**
+ * Retrieve a list of locks for this object
+ *
+ * Input:
+ * @param in is ignored.
+ *
+ * Output:
+ * @param out contains encoded cls_list_locks_reply
+ *
+ * @return 0 on success, -errno on failure.
+ */
+static int list_locks(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "list_locks");
+
+  map<string, bufferlist> attrs;
+
+  int r = cls_cxx_getxattrs(hctx, &attrs);
+  if (r < 0)
+    return r;
+
+  cls_lock_list_locks_reply ret;
+
+  map<string, bufferlist>::iterator iter;
+  size_t pos = sizeof(LOCK_PREFIX) - 1;
+  for (iter = attrs.begin(); iter != attrs.end(); ++iter) {
+    const string& attr = iter->first;
+    if (attr.substr(0, pos).compare(LOCK_PREFIX) == 0) {
+      ret.locks.push_back(attr.substr(pos));
+    }
+  }
+
+  encode(ret, *out);
+
+  return 0;
+}
+
+/**
+ * Assert that the object is currently locked
+ *
+ * Input:
+ * @param cls_lock_assert_op request input
+ *
+ * Output:
+ * @param none
+ *
+ * @return 0 on success, -errno on failure.
+ */
+int assert_locked(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "assert_locked");
+
+  cls_lock_assert_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error& err) {
+    return -EINVAL;
+  }
+
+  if (!cls_lock_is_valid(op.type)) {
+    return -EINVAL;
+  }
+
+  if (op.name.empty()) {
+    return -EINVAL;
+  }
+
+  // see if there's already a locker
+  lock_info_t linfo;
+  int r = read_lock(hctx, op.name, &linfo);
+  if (r < 0) {
+    CLS_ERR("Could not read lock info: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (linfo.lockers.empty()) {
+    CLS_LOG(20, "object not locked");
+    return -EBUSY;
+  }
+
+  if (linfo.lock_type != op.type) {
+    CLS_LOG(20, "lock type mismatch: current=%s, assert=%s",
+            cls_lock_type_str(linfo.lock_type), cls_lock_type_str(op.type));
+    return -EBUSY;
+  }
+
+  if (linfo.tag != op.tag) {
+    CLS_LOG(20, "lock tag mismatch: current=%s, assert=%s", linfo.tag.c_str(),
+            op.tag.c_str());
+    return -EBUSY;
+  }
+
+  entity_inst_t inst;
+  r = cls_get_request_origin(hctx, &inst);
+  ceph_assert(r == 0);
+
+  locker_id_t id;
+  id.cookie = op.cookie;
+  id.locker = inst.name;
+
+  map<locker_id_t, locker_info_t>::iterator iter = linfo.lockers.find(id);
+  if (iter == linfo.lockers.end()) {
+    CLS_LOG(20, "not locked by assert client");
+    return -EBUSY;
+  }
+  return 0;
+}
+
+/**
+ * Update the cookie associated with an object lock
+ *
+ * Input:
+ * @param cls_lock_set_cookie_op request input
+ *
+ * Output:
+ * @param none
+ *
+ * @return 0 on success, -errno on failure.
+ */
+int set_cookie(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "set_cookie");
+
+  cls_lock_set_cookie_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error& err) {
+    return -EINVAL;
+  }
+
+  if (!cls_lock_is_valid(op.type)) {
+    return -EINVAL;
+  }
+
+  if (op.name.empty()) {
+    return -EINVAL;
+  }
+
+  // see if there's already a locker
+  lock_info_t linfo;
+  int r = read_lock(hctx, op.name, &linfo);
+  if (r < 0) {
+    CLS_ERR("Could not read lock info: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (linfo.lockers.empty()) {
+    CLS_LOG(20, "object not locked");
+    return -EBUSY;
+  }
+
+  if (linfo.lock_type != op.type) {
+    CLS_LOG(20, "lock type mismatch: current=%s, assert=%s",
+            cls_lock_type_str(linfo.lock_type), cls_lock_type_str(op.type));
+    return -EBUSY;
+  }
+
+  if (linfo.tag != op.tag) {
+    CLS_LOG(20, "lock tag mismatch: current=%s, assert=%s", linfo.tag.c_str(),
+            op.tag.c_str());
+    return -EBUSY;
+  }
+
+  entity_inst_t inst;
+  r = cls_get_request_origin(hctx, &inst);
+  ceph_assert(r == 0);
+
+  locker_id_t id;
+  id.cookie = op.cookie;
+  id.locker = inst.name;
+
+  map<locker_id_t, locker_info_t>::iterator iter = linfo.lockers.find(id);
+  if (iter == linfo.lockers.end()) {
+    CLS_LOG(20, "not locked by client");
+    return -EBUSY;
+  }
+
+  id.cookie = op.new_cookie;
+  if (linfo.lockers.count(id) != 0) {
+    CLS_LOG(20, "lock cookie in-use");
+    return -EBUSY;
+  }
+
+  locker_info_t locker_info(iter->second);
+  linfo.lockers.erase(iter);
+
+  linfo.lockers[id] = locker_info;
+  r = write_lock(hctx, op.name, linfo);
+  if (r < 0) {
+    CLS_ERR("Could not update lock info: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+CLS_INIT(lock)
+{
+  CLS_LOG(20, "Loaded lock class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_lock_op;
+  cls_method_handle_t h_unlock_op;
+  cls_method_handle_t h_break_lock;
+  cls_method_handle_t h_get_info;
+  cls_method_handle_t h_list_locks;
+  cls_method_handle_t h_assert_locked;
+  cls_method_handle_t h_set_cookie;
+
+  cls_register("lock", &h_class);
+  cls_register_cxx_method(h_class, "lock",
+                          CLS_METHOD_RD | CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+                          lock_op, &h_lock_op);
+  cls_register_cxx_method(h_class, "unlock",
+                          CLS_METHOD_RD | CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+                          unlock_op, &h_unlock_op);
+  cls_register_cxx_method(h_class, "break_lock",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          break_lock, &h_break_lock);
+  cls_register_cxx_method(h_class, "get_info",
+                          CLS_METHOD_RD,
+                          get_info, &h_get_info);
+  cls_register_cxx_method(h_class, "list_locks",
+                          CLS_METHOD_RD,
+                          list_locks, &h_list_locks);
+  cls_register_cxx_method(h_class, "assert_locked",
+                          CLS_METHOD_RD | CLS_METHOD_PROMOTE,
+                          assert_locked, &h_assert_locked);
+  cls_register_cxx_method(h_class, "set_cookie",
+                          CLS_METHOD_RD | CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+                          set_cookie, &h_set_cookie);
+
+  return;
+}
diff --git a/src/cls/lock/cls_lock_client.cc b/src/cls/lock/cls_lock_client.cc
new file mode 100644
index 00000000..498d573f
--- /dev/null
+++ b/src/cls/lock/cls_lock_client.cc
@@ -0,0 +1,284 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software 
+ * Foundation.  See file COPYING.
+ * 
+ */
+
+#include "include/types.h"
+#include "msg/msg_types.h"
+#include "include/rados/librados.hpp"
+#include "include/utime.h"
+ 
+using namespace librados;
+
+#include "cls/lock/cls_lock_ops.h"
+#include "cls/lock/cls_lock_client.h"
+
+namespace rados {
+  namespace cls {
+    namespace lock {
+
+      void lock(ObjectWriteOperation *rados_op,
+                const string& name, ClsLockType type,
+                const string& cookie, const string& tag,
+                const string& description,
+                const utime_t& duration, uint8_t flags)
+      {
+        cls_lock_lock_op op;
+        op.name = name;
+        op.type = type;
+        op.cookie = cookie;
+        op.tag = tag;
+        op.description = description;
+        op.duration = duration;
+        op.flags = flags;
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("lock", "lock", in);
+      }
+
+      int lock(IoCtx *ioctx,
+               const string& oid,
+               const string& name, ClsLockType type,
+               const string& cookie, const string& tag,
+               const string& description, const utime_t& duration,
+	       uint8_t flags)
+      {
+        ObjectWriteOperation op;
+        lock(&op, name, type, cookie, tag, description, duration, flags);
+        return ioctx->operate(oid, &op);
+      }
+
+      void unlock(ObjectWriteOperation *rados_op,
+                  const string& name, const string& cookie)
+      {
+        cls_lock_unlock_op op;
+        op.name = name;
+        op.cookie = cookie;
+        bufferlist in;
+        encode(op, in);
+
+        rados_op->exec("lock", "unlock", in);
+      }
+
+      int unlock(IoCtx *ioctx, const string& oid,
+                 const string& name, const string& cookie)
+      {
+        ObjectWriteOperation op;
+        unlock(&op, name, cookie);
+        return ioctx->operate(oid, &op);
+      }
+
+      int aio_unlock(IoCtx *ioctx, const string& oid,
+		     const string& name, const string& cookie,
+		     librados::AioCompletion *completion)
+      {
+        ObjectWriteOperation op;
+        unlock(&op, name, cookie);
+        return ioctx->aio_operate(oid, completion, &op);
+      }
+
+      void break_lock(ObjectWriteOperation *rados_op,
+                      const string& name, const string& cookie,
+                      const entity_name_t& locker)
+      {
+        cls_lock_break_op op;
+        op.name = name;
+        op.cookie = cookie;
+        op.locker = locker;
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("lock", "break_lock", in);
+      }
+
+      int break_lock(IoCtx *ioctx, const string& oid,
+                     const string& name, const string& cookie,
+                     const entity_name_t& locker)
+      {
+        ObjectWriteOperation op;
+        break_lock(&op, name, cookie, locker);
+        return ioctx->operate(oid, &op);
+      }
+
+      int list_locks(IoCtx *ioctx, const string& oid, list<string> *locks)
+      {
+        bufferlist in, out;
+        int r = ioctx->exec(oid, "lock", "list_locks", in, out);
+        if (r < 0)
+          return r;
+
+        cls_lock_list_locks_reply ret;
+        auto iter = cbegin(out);
+        try {
+          decode(ret, iter);
+        } catch (buffer::error& err) {
+	  return -EBADMSG;
+        }
+
+        *locks = ret.locks;
+
+        return 0;
+      }
+
+      void get_lock_info_start(ObjectReadOperation *rados_op,
+			       const string& name)
+      {
+        bufferlist in;
+        cls_lock_get_info_op op;
+        op.name = name;
+        encode(op, in);
+        rados_op->exec("lock", "get_info", in);
+      }
+
+      int get_lock_info_finish(bufferlist::const_iterator *iter,
+			       map<locker_id_t, locker_info_t> *lockers,
+			       ClsLockType *type, string *tag)
+      {
+        cls_lock_get_info_reply ret;
+        try {
+          decode(ret, *iter);
+        } catch (buffer::error& err) {
+	  return -EBADMSG;
+        }
+
+        if (lockers) {
+          *lockers = ret.lockers;
+        }
+
+        if (type) {
+          *type = ret.lock_type;
+        }
+
+        if (tag) {
+          *tag = ret.tag;
+        }
+
+        return 0;
+      }
+
+      int get_lock_info(IoCtx *ioctx, const string& oid, const string& name,
+                        map<locker_id_t, locker_info_t> *lockers,
+                        ClsLockType *type, string *tag)
+      {
+        ObjectReadOperation op;
+        get_lock_info_start(&op, name);
+	bufferlist out;
+        int r = ioctx->operate(oid, &op, &out);
+	if (r < 0)
+	  return r;
+	auto it = std::cbegin(out);
+	return get_lock_info_finish(&it, lockers, type, tag);
+      }
+
+      void assert_locked(librados::ObjectOperation *rados_op,
+                         const std::string& name, ClsLockType type,
+                         const std::string& cookie, const std::string& tag)
+      {
+        cls_lock_assert_op op;
+        op.name = name;
+        op.type = type;
+        op.cookie = cookie;
+        op.tag = tag;
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("lock", "assert_locked", in);
+      }
+
+      void set_cookie(librados::ObjectWriteOperation *rados_op,
+                      const std::string& name, ClsLockType type,
+                      const std::string& cookie, const std::string& tag,
+                      const std::string& new_cookie)
+      {
+        cls_lock_set_cookie_op op;
+        op.name = name;
+        op.type = type;
+        op.cookie = cookie;
+        op.tag = tag;
+        op.new_cookie = new_cookie;
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("lock", "set_cookie", in);
+      }
+
+      void Lock::assert_locked_shared(ObjectOperation *op)
+      {
+        assert_locked(op, name, LOCK_SHARED, cookie, tag);
+      }
+
+      void Lock::assert_locked_exclusive(ObjectOperation *op)
+      {
+        assert_locked(op, name, LOCK_EXCLUSIVE, cookie, tag);
+      }
+
+      void Lock::assert_locked_exclusive_ephemeral(ObjectOperation *op)
+      {
+        assert_locked(op, name, LOCK_EXCLUSIVE_EPHEMERAL, cookie, tag);
+      }
+
+      void Lock::lock_shared(ObjectWriteOperation *op)
+      {
+        lock(op, name, LOCK_SHARED,
+             cookie, tag, description, duration, flags);
+      }
+
+      int Lock::lock_shared(IoCtx *ioctx, const string& oid)
+      {
+        return lock(ioctx, oid, name, LOCK_SHARED,
+                    cookie, tag, description, duration, flags);
+      }
+
+      void Lock::lock_exclusive(ObjectWriteOperation *op)
+      {
+        lock(op, name, LOCK_EXCLUSIVE,
+             cookie, tag, description, duration, flags);
+      }
+
+      int Lock::lock_exclusive(IoCtx *ioctx, const string& oid)
+      {
+        return lock(ioctx, oid, name, LOCK_EXCLUSIVE,
+                    cookie, tag, description, duration, flags);
+      }
+
+      void Lock::lock_exclusive_ephemeral(ObjectWriteOperation *op)
+      {
+        lock(op, name, LOCK_EXCLUSIVE_EPHEMERAL,
+             cookie, tag, description, duration, flags);
+      }
+
+      int Lock::lock_exclusive_ephemeral(IoCtx *ioctx, const string& oid)
+      {
+        return lock(ioctx, oid, name, LOCK_EXCLUSIVE_EPHEMERAL,
+                    cookie, tag, description, duration, flags);
+      }
+
+      void Lock::unlock(ObjectWriteOperation *op)
+      {
+	rados::cls::lock::unlock(op, name, cookie);
+      }
+
+      int Lock::unlock(IoCtx *ioctx, const string& oid)
+      {
+        return rados::cls::lock::unlock(ioctx, oid, name, cookie);
+      }
+
+      void Lock::break_lock(ObjectWriteOperation *op, const entity_name_t& locker)
+      {
+	rados::cls::lock::break_lock(op, name, cookie, locker);
+      }
+
+      int Lock::break_lock(IoCtx *ioctx, const string& oid, const entity_name_t& locker)
+      {
+          return rados::cls::lock::break_lock(ioctx, oid, name, cookie, locker);
+      }
+    } // namespace lock
+  } // namespace cls
+} // namespace rados
+
diff --git a/src/cls/lock/cls_lock_client.h b/src/cls/lock/cls_lock_client.h
new file mode 100644
index 00000000..11c0e658
--- /dev/null
+++ b/src/cls/lock/cls_lock_client.h
@@ -0,0 +1,141 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_LOCK_CLIENT_H
+#define CEPH_CLS_LOCK_CLIENT_H
+
+#include <chrono>
+
+#include "include/rados/librados_fwd.hpp"
+#include "cls/lock/cls_lock_types.h"
+
+namespace rados {
+  namespace cls {
+    namespace lock {
+      extern void lock(librados::ObjectWriteOperation *rados_op,
+		       const std::string& name, ClsLockType type,
+		       const std::string& cookie, const std::string& tag,
+		       const std::string& description, const utime_t& duration,
+		       uint8_t flags);
+
+      extern int lock(librados::IoCtx *ioctx,
+		      const std::string& oid,
+		      const std::string& name, ClsLockType type,
+		      const std::string& cookie, const std::string& tag,
+		      const std::string& description, const utime_t& duration,
+		      uint8_t flags);
+
+      extern void unlock(librados::ObjectWriteOperation *rados_op,
+			 const std::string& name, const std::string& cookie);
+
+      extern int unlock(librados::IoCtx *ioctx, const std::string& oid,
+			const std::string& name, const std::string& cookie);
+
+      extern int aio_unlock(librados::IoCtx *ioctx, const std::string& oid,
+			    const std::string& name, const std::string& cookie,
+			    librados::AioCompletion *completion);
+
+      extern void break_lock(librados::ObjectWriteOperation *op,
+			     const std::string& name, const std::string& cookie,
+			     const entity_name_t& locker);
+
+      extern int break_lock(librados::IoCtx *ioctx, const std::string& oid,
+			    const std::string& name, const std::string& cookie,
+			    const entity_name_t& locker);
+
+      extern int list_locks(librados::IoCtx *ioctx, const std::string& oid,
+			    list<std::string> *locks);
+      extern void get_lock_info_start(librados::ObjectReadOperation *rados_op,
+				      const std::string& name);
+      extern int get_lock_info_finish(ceph::bufferlist::const_iterator *out,
+				      map<locker_id_t, locker_info_t> *lockers,
+				      ClsLockType *type, std::string *tag);
+
+      extern int get_lock_info(librados::IoCtx *ioctx, const std::string& oid,
+			       const std::string& name,
+			       map<locker_id_t, locker_info_t> *lockers,
+			       ClsLockType *type, std::string *tag);
+
+      extern void assert_locked(librados::ObjectOperation *rados_op,
+                                const std::string& name, ClsLockType type,
+                                const std::string& cookie,
+                                const std::string& tag);
+
+      extern void set_cookie(librados::ObjectWriteOperation *rados_op,
+                             const std::string& name, ClsLockType type,
+                             const std::string& cookie, const std::string& tag,
+                             const std::string& new_cookie);
+
+      class Lock {
+	std::string name;
+	std::string cookie;
+	std::string tag;
+	std::string description;
+	utime_t duration;
+	uint8_t flags;
+
+      public:
+
+	Lock(const std::string& _n) : name(_n), flags(0) {}
+
+	void set_cookie(const std::string& c) { cookie = c; }
+	void set_tag(const std::string& t) { tag = t; }
+	void set_description(const std::string& desc) { description = desc; }
+	void set_duration(const utime_t& e) { duration = e; }
+	void set_duration(const ceph::timespan& d) {
+	  duration = utime_t(ceph::real_clock::zero() + d);
+	}
+
+	void set_may_renew(bool renew) {
+	  if (renew) {
+	    flags |= LOCK_FLAG_MAY_RENEW;
+	    flags &= ~LOCK_FLAG_MUST_RENEW; // if may then not must
+	  } else {
+	    flags &= ~LOCK_FLAG_MAY_RENEW;
+	  }
+	}
+
+	void set_must_renew(bool renew) {
+	  if (renew) {
+	    flags |= LOCK_FLAG_MUST_RENEW;
+	    flags &= ~LOCK_FLAG_MAY_RENEW; // if must then not may
+	  } else {
+	    flags &= ~LOCK_FLAG_MUST_RENEW;
+	  }
+	}
+
+        void assert_locked_shared(librados::ObjectOperation *rados_op);
+        void assert_locked_exclusive(librados::ObjectOperation *rados_op);
+        void assert_locked_exclusive_ephemeral(librados::ObjectOperation *rados_op);
+
+	/* ObjectWriteOperation */
+	void lock_shared(librados::ObjectWriteOperation *ioctx);
+	void lock_exclusive(librados::ObjectWriteOperation *ioctx);
+
+	// Be careful when using an exclusive ephemeral lock; it is
+	// intended strictly for cases when a lock object exists
+	// solely for a lock in a given process and the object is no
+	// longer needed when the lock is unlocked or expired, as the
+	// cls back-end will make an effort to delete it.
+	void lock_exclusive_ephemeral(librados::ObjectWriteOperation *ioctx);
+	void unlock(librados::ObjectWriteOperation *ioctx);
+	void break_lock(librados::ObjectWriteOperation *ioctx,
+			const entity_name_t& locker);
+
+	/* IoCtx */
+	int lock_shared(librados::IoCtx *ioctx, const std::string& oid);
+	int lock_exclusive(librados::IoCtx *ioctx, const std::string& oid);
+
+	// NB: see above comment on exclusive ephemeral locks
+	int lock_exclusive_ephemeral(librados::IoCtx *ioctx,
+				     const std::string& oid);
+	int unlock(librados::IoCtx *ioctx, const std::string& oid);
+	int break_lock(librados::IoCtx *ioctx, const std::string& oid,
+		       const entity_name_t& locker);
+      };
+
+    } // namespace lock
+  }  // namespace cls
+} // namespace rados
+
+#endif
diff --git a/src/cls/lock/cls_lock_ops.cc b/src/cls/lock/cls_lock_ops.cc
new file mode 100644
index 00000000..aa5f8245
--- /dev/null
+++ b/src/cls/lock/cls_lock_ops.cc
@@ -0,0 +1,210 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software 
+ * Foundation.  See file COPYING.
+ * 
+ */
+
+#include "msg/msg_types.h"
+#include "common/Formatter.h"
+
+#include "cls/lock/cls_lock_ops.h"
+
+using namespace rados::cls::lock;
+
+static void generate_lock_id(locker_id_t& i, int n, const string& cookie)
+{
+  i.locker = entity_name_t::CLIENT(n);
+  i.cookie = cookie;
+}
+
+void cls_lock_lock_op::dump(Formatter *f) const
+{
+  f->dump_string("name", name);
+  f->dump_string("type", cls_lock_type_str(type));
+  f->dump_string("cookie", cookie);
+  f->dump_string("tag", tag);
+  f->dump_string("description", description);
+  f->dump_stream("duration") << duration;
+  f->dump_int("flags", (int)flags);
+}
+
+void cls_lock_lock_op::generate_test_instances(list<cls_lock_lock_op*>& o)
+{
+  cls_lock_lock_op *i = new cls_lock_lock_op;
+  i->name = "name";
+  i->type = LOCK_SHARED;
+  i->cookie = "cookie";
+  i->tag = "tag";
+  i->description = "description";
+  i->duration = utime_t(5, 0);
+  i->flags = LOCK_FLAG_MAY_RENEW;
+  o.push_back(i);
+  o.push_back(new cls_lock_lock_op);
+}
+
+void cls_lock_unlock_op::dump(Formatter *f) const
+{
+  f->dump_string("name", name);
+  f->dump_string("cookie", cookie);
+}
+
+void cls_lock_unlock_op::generate_test_instances(list<cls_lock_unlock_op*>& o)
+{
+  cls_lock_unlock_op *i = new cls_lock_unlock_op;
+  i->name = "name";
+  i->cookie = "cookie";
+  o.push_back(i);
+  o.push_back(new cls_lock_unlock_op);
+}
+
+void cls_lock_break_op::dump(Formatter *f) const
+{
+  f->dump_string("name", name);
+  f->dump_string("cookie", cookie);
+  f->dump_stream("locker") << locker;
+}
+
+void cls_lock_break_op::generate_test_instances(list<cls_lock_break_op*>& o)
+{
+  cls_lock_break_op *i = new cls_lock_break_op;
+  i->name = "name";
+  i->cookie = "cookie";
+  i->locker =  entity_name_t::CLIENT(1);
+  o.push_back(i);
+  o.push_back(new cls_lock_break_op);
+}
+
+void cls_lock_get_info_op::dump(Formatter *f) const
+{
+  f->dump_string("name", name);
+}
+
+void cls_lock_get_info_op::generate_test_instances(list<cls_lock_get_info_op*>& o)
+{
+  cls_lock_get_info_op *i = new cls_lock_get_info_op;
+  i->name = "name";
+  o.push_back(i);
+  o.push_back(new cls_lock_get_info_op);
+}
+
+static void generate_test_addr(entity_addr_t& a, int nonce, int port)
+{
+  a.set_type(entity_addr_t::TYPE_LEGACY);
+  a.set_nonce(nonce);
+  a.set_family(AF_INET);
+  a.set_in4_quad(0, 127);
+  a.set_in4_quad(1, 0);
+  a.set_in4_quad(2, 1);
+  a.set_in4_quad(3, 2);
+  a.set_port(port);
+}
+
+void cls_lock_get_info_reply::dump(Formatter *f) const
+{
+  f->dump_string("lock_type", cls_lock_type_str(lock_type));
+  f->dump_string("tag", tag);
+  f->open_array_section("lockers");
+  map<locker_id_t, locker_info_t>::const_iterator iter;
+  for (iter = lockers.begin(); iter != lockers.end(); ++iter) {
+    const locker_id_t& id = iter->first;
+    const locker_info_t& info = iter->second;
+    f->open_object_section("object");
+    f->dump_stream("locker") << id.locker;
+    f->dump_string("description", info.description);
+    f->dump_string("cookie", id.cookie);
+    f->dump_stream("expiration") << info.expiration;
+    f->dump_string("addr", info.addr.get_legacy_str());
+    f->close_section();
+  }
+  f->close_section();
+}
+
+void cls_lock_get_info_reply::generate_test_instances(list<cls_lock_get_info_reply*>& o)
+{
+  cls_lock_get_info_reply *i = new cls_lock_get_info_reply;
+  i->lock_type = LOCK_SHARED;
+  i->tag = "tag";
+  locker_id_t id1, id2;
+  entity_addr_t addr1, addr2;
+  generate_lock_id(id1, 1, "cookie1");
+  generate_test_addr(addr1, 10, 20);
+  i->lockers[id1] = locker_info_t(utime_t(10, 0), addr1, "description1");
+  generate_lock_id(id2, 2, "cookie2");
+  generate_test_addr(addr2, 30, 40);
+  i->lockers[id2] = locker_info_t(utime_t(20, 0), addr2, "description2");
+
+  o.push_back(i);
+  o.push_back(new cls_lock_get_info_reply);
+}
+
+void cls_lock_list_locks_reply::dump(Formatter *f) const
+{
+  list<string>::const_iterator iter;
+  f->open_array_section("locks");
+  for (iter = locks.begin(); iter != locks.end(); ++iter) {
+    f->open_array_section("object");
+    f->dump_string("lock", *iter);
+    f->close_section();
+  }
+  f->close_section();
+}
+
+void cls_lock_list_locks_reply::generate_test_instances(list<cls_lock_list_locks_reply*>& o)
+{
+  cls_lock_list_locks_reply *i = new cls_lock_list_locks_reply;
+  i->locks.push_back("lock1");
+  i->locks.push_back("lock2");
+  i->locks.push_back("lock3");
+
+  o.push_back(i);
+  o.push_back(new cls_lock_list_locks_reply);
+}
+
+void cls_lock_assert_op::dump(Formatter *f) const
+{
+  f->dump_string("name", name);
+  f->dump_string("type", cls_lock_type_str(type));
+  f->dump_string("cookie", cookie);
+  f->dump_string("tag", tag);
+}
+
+void cls_lock_assert_op::generate_test_instances(list<cls_lock_assert_op*>& o)
+{
+  cls_lock_assert_op *i = new cls_lock_assert_op;
+  i->name = "name";
+  i->type = LOCK_SHARED;
+  i->cookie = "cookie";
+  i->tag = "tag";
+  o.push_back(i);
+  o.push_back(new cls_lock_assert_op);
+}
+
+void cls_lock_set_cookie_op::dump(Formatter *f) const
+{
+  f->dump_string("name", name);
+  f->dump_string("type", cls_lock_type_str(type));
+  f->dump_string("cookie", cookie);
+  f->dump_string("tag", tag);
+  f->dump_string("new_cookie", new_cookie);
+}
+
+void cls_lock_set_cookie_op::generate_test_instances(list<cls_lock_set_cookie_op*>& o)
+{
+  cls_lock_set_cookie_op *i = new cls_lock_set_cookie_op;
+  i->name = "name";
+  i->type = LOCK_SHARED;
+  i->cookie = "cookie";
+  i->tag = "tag";
+  i->new_cookie = "new cookie";
+  o.push_back(i);
+  o.push_back(new cls_lock_set_cookie_op);
+}
+
diff --git a/src/cls/lock/cls_lock_ops.h b/src/cls/lock/cls_lock_ops.h
new file mode 100644
index 00000000..5d22452b
--- /dev/null
+++ b/src/cls/lock/cls_lock_ops.h
@@ -0,0 +1,245 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_LOCK_OPS_H
+#define CEPH_CLS_LOCK_OPS_H
+
+#include "include/types.h"
+#include "include/utime.h"
+#include "cls/lock/cls_lock_types.h"
+
+struct cls_lock_lock_op
+{
+  string name;
+  ClsLockType type;
+  string cookie;
+  string tag;
+  string description;
+  utime_t duration;
+  uint8_t flags;
+
+  cls_lock_lock_op() : type(LOCK_NONE), flags(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    uint8_t t = (uint8_t)type;
+    encode(t, bl);
+    encode(cookie, bl);
+    encode(tag, bl);
+    encode(description, bl);
+    encode(duration, bl);
+    encode(flags, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(name, bl);
+    uint8_t t;
+    decode(t, bl);
+    type = (ClsLockType)t;
+    decode(cookie, bl);
+    decode(tag, bl);
+    decode(description, bl);
+    decode(duration, bl);
+    decode(flags, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_lock_op*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_lock_op)
+
+struct cls_lock_unlock_op
+{
+  string name;
+  string cookie;
+
+  cls_lock_unlock_op() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    encode(cookie, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(name, bl);
+    decode(cookie, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_unlock_op*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_unlock_op)
+
+struct cls_lock_break_op
+{
+  string name;
+  entity_name_t locker;
+  string cookie;
+
+  cls_lock_break_op() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    encode(locker, bl);
+    encode(cookie, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(name, bl);
+    decode(locker, bl);
+    decode(cookie, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_break_op*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_break_op)
+
+struct cls_lock_get_info_op
+{
+  string name;
+
+  cls_lock_get_info_op() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(name, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_get_info_op*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_get_info_op)
+
+struct cls_lock_get_info_reply
+{
+  map<rados::cls::lock::locker_id_t, rados::cls::lock::locker_info_t> lockers;
+  ClsLockType lock_type;
+  string tag;
+
+  cls_lock_get_info_reply() : lock_type(LOCK_NONE) {}
+
+  void encode(bufferlist &bl, uint64_t features) const {
+    ENCODE_START(1, 1, bl);
+    encode(lockers, bl, features);
+    uint8_t t = (uint8_t)lock_type;
+    encode(t, bl);
+    encode(tag, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(lockers, bl);
+    uint8_t t;
+    decode(t, bl);
+    lock_type = (ClsLockType)t; 
+    decode(tag, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_get_info_reply*>& o);
+};
+WRITE_CLASS_ENCODER_FEATURES(cls_lock_get_info_reply)
+
+struct cls_lock_list_locks_reply
+{
+  list<string> locks;
+
+  cls_lock_list_locks_reply() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(locks, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(locks, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_list_locks_reply*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_list_locks_reply)
+
+struct cls_lock_assert_op
+{
+  string name;
+  ClsLockType type;
+  string cookie;
+  string tag;
+
+  cls_lock_assert_op() : type(LOCK_NONE) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    uint8_t t = (uint8_t)type;
+    encode(t, bl);
+    encode(cookie, bl);
+    encode(tag, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(name, bl);
+    uint8_t t;
+    decode(t, bl);
+    type = (ClsLockType)t;
+    decode(cookie, bl);
+    decode(tag, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_assert_op*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_assert_op)
+
+struct cls_lock_set_cookie_op
+{
+  string name;
+  ClsLockType type;
+  string cookie;
+  string tag;
+  string new_cookie;
+
+  cls_lock_set_cookie_op() : type(LOCK_NONE) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    uint8_t t = (uint8_t)type;
+    encode(t, bl);
+    encode(cookie, bl);
+    encode(tag, bl);
+    encode(new_cookie, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+    decode(name, bl);
+    uint8_t t;
+    decode(t, bl);
+    type = (ClsLockType)t;
+    decode(cookie, bl);
+    decode(tag, bl);
+    decode(new_cookie, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_lock_set_cookie_op*>& o);
+};
+WRITE_CLASS_ENCODER(cls_lock_set_cookie_op)
+
+#endif
diff --git a/src/cls/lock/cls_lock_types.cc b/src/cls/lock/cls_lock_types.cc
new file mode 100644
index 00000000..d1aa6334
--- /dev/null
+++ b/src/cls/lock/cls_lock_types.cc
@@ -0,0 +1,98 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software 
+ * Foundation.  See file COPYING.
+ * 
+ */
+
+#include "common/Formatter.h"
+
+#include "cls/lock/cls_lock_types.h"
+
+using namespace rados::cls::lock;
+
+static void generate_lock_id(locker_id_t& i, int n, const string& cookie)
+{
+  i.locker = entity_name_t::CLIENT(n);
+  i.cookie = cookie;
+}
+
+void locker_id_t::dump(Formatter *f) const
+{
+  f->dump_stream("locker") << locker;
+  f->dump_string("cookie", cookie);
+}
+
+void locker_id_t::generate_test_instances(list<locker_id_t*>& o)
+{
+  locker_id_t *i = new locker_id_t;
+  generate_lock_id(*i, 1, "cookie");
+  o.push_back(i);
+  o.push_back(new locker_id_t);
+}
+
+void locker_info_t::dump(Formatter *f) const
+{
+  f->dump_stream("expiration") << expiration;
+  f->dump_string("addr", addr.get_legacy_str());
+  f->dump_string("description", description);
+}
+
+static void generate_test_addr(entity_addr_t& a, int nonce, int port)
+{
+  a.set_type(entity_addr_t::TYPE_LEGACY);
+  a.set_nonce(nonce);
+  a.set_family(AF_INET);
+  a.set_in4_quad(0, 127);
+  a.set_in4_quad(1, 0);
+  a.set_in4_quad(2, 1);
+  a.set_in4_quad(3, 2);
+  a.set_port(port);
+}
+
+void locker_info_t::generate_test_instances(list<locker_info_t*>& o)
+{
+  locker_info_t *i = new locker_info_t;
+  i->expiration = utime_t(5, 0);
+  generate_test_addr(i->addr, 1, 2);
+  i->description = "description";
+  o.push_back(i);
+  o.push_back(new locker_info_t);
+}
+
+void lock_info_t::dump(Formatter *f) const
+{
+  f->dump_int("lock_type", lock_type);
+  f->dump_string("tag", tag);
+  f->open_array_section("lockers");
+  for (auto &i : lockers) {
+    f->open_object_section("locker");
+    f->dump_object("id", i.first);
+    f->dump_object("info", i.second);
+    f->close_section();
+  }
+  f->close_section();
+}
+
+void lock_info_t::generate_test_instances(list<lock_info_t *>& o)
+{
+  lock_info_t *i = new lock_info_t;
+  locker_id_t id;
+  locker_info_t info;
+  generate_lock_id(id, 1, "cookie");
+  info.expiration = utime_t(5, 0);
+  generate_test_addr(info.addr, 1, 2);
+  info.description = "description";
+  i->lockers[id] = info;
+  i->lock_type = LOCK_EXCLUSIVE;
+  i->tag = "tag";
+  o.push_back(i);
+  o.push_back(new lock_info_t);
+}
diff --git a/src/cls/lock/cls_lock_types.h b/src/cls/lock/cls_lock_types.h
new file mode 100644
index 00000000..fbd5b571
--- /dev/null
+++ b/src/cls/lock/cls_lock_types.h
@@ -0,0 +1,172 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_LOCK_TYPES_H
+#define CEPH_CLS_LOCK_TYPES_H
+
+#include "include/encoding.h"
+#include "include/types.h"
+#include "include/utime.h"
+#include "msg/msg_types.h"
+
+/* lock flags */
+#define LOCK_FLAG_MAY_RENEW 0x1    /* idempotent lock acquire */
+#define LOCK_FLAG_MUST_RENEW 0x2   /* lock must already be acquired */
+
+enum ClsLockType {
+  LOCK_NONE                = 0,
+  LOCK_EXCLUSIVE           = 1,
+  LOCK_SHARED              = 2,
+  LOCK_EXCLUSIVE_EPHEMERAL = 3, /* lock object is removed @ unlock */
+};
+
+inline const char *cls_lock_type_str(ClsLockType type)
+{
+    switch (type) {
+      case LOCK_NONE:
+	return "none";
+      case LOCK_EXCLUSIVE:
+	return "exclusive";
+      case LOCK_SHARED:
+	return "shared";
+      case LOCK_EXCLUSIVE_EPHEMERAL:
+	return "exclusive-ephemeral";
+      default:
+	return "<unknown>";
+    }
+}
+
+inline bool cls_lock_is_exclusive(ClsLockType type) {
+  return LOCK_EXCLUSIVE == type || LOCK_EXCLUSIVE_EPHEMERAL == type;
+}
+
+inline bool cls_lock_is_ephemeral(ClsLockType type) {
+  return LOCK_EXCLUSIVE_EPHEMERAL == type;
+}
+
+inline bool cls_lock_is_valid(ClsLockType type) {
+  return LOCK_SHARED == type ||
+    LOCK_EXCLUSIVE == type ||
+    LOCK_EXCLUSIVE_EPHEMERAL == type;
+}
+
+namespace rados {
+  namespace cls {
+    namespace lock {
+
+      /*
+       * locker_id_t: the locker id, needs to be unique in a single lock
+       */
+      struct locker_id_t {
+        entity_name_t locker;   // locker's client name
+        string cookie;          // locker's cookie.
+
+        locker_id_t() {}
+        locker_id_t(entity_name_t& _n, const string& _c) : locker(_n), cookie(_c) {}
+
+        void encode(bufferlist &bl) const {
+          ENCODE_START(1, 1, bl);
+          encode(locker, bl);
+          encode(cookie, bl);
+          ENCODE_FINISH(bl);
+        }
+        void decode(bufferlist::const_iterator &bl) {
+          DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+          decode(locker, bl);
+          decode(cookie, bl);
+          DECODE_FINISH(bl);
+        }
+
+        bool operator<(const locker_id_t& rhs) const {
+          if (locker == rhs.locker)
+            return cookie.compare(rhs.cookie) < 0;
+          if (locker < rhs.locker)
+            return true;
+          return false;
+        }
+        void dump(Formatter *f) const;
+	friend std::ostream& operator<<(std::ostream& out,
+					const locker_id_t& data) {
+	  out << data.locker;
+	  return out;
+	}
+        static void generate_test_instances(list<locker_id_t*>& o);
+      };
+      WRITE_CLASS_ENCODER(locker_id_t)
+
+      struct locker_info_t
+      {
+        utime_t expiration;  // expiration: non-zero means epoch of locker expiration
+        entity_addr_t addr;  // addr: locker address
+        string description;  // description: locker description, may be empty
+
+        locker_info_t() {}
+        locker_info_t(const utime_t& _e, const entity_addr_t& _a,
+                      const string& _d) :  expiration(_e), addr(_a), description(_d) {}
+
+        void encode(bufferlist &bl, uint64_t features) const {
+          ENCODE_START(1, 1, bl);
+          encode(expiration, bl);
+          encode(addr, bl, features);
+          encode(description, bl);
+          ENCODE_FINISH(bl);
+        }
+        void decode(bufferlist::const_iterator &bl) {
+          DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+          decode(expiration, bl);
+          decode(addr, bl);
+          decode(description, bl);
+          DECODE_FINISH(bl);
+        }
+        void dump(Formatter *f) const;
+	friend std::ostream& operator<<(std::ostream& out,
+					const locker_info_t& data) {
+	  out << "{addr:" << data.addr << ", exp:";
+
+	  const auto& exp = data.expiration;
+	  if (exp.is_zero()) {
+	    out << "never}";
+	  } else {
+	    out << exp.to_real_time() << "}";
+	  }
+
+	  return out;
+	}
+        static void generate_test_instances(list<locker_info_t *>& o);
+      };
+      WRITE_CLASS_ENCODER_FEATURES(locker_info_t)
+
+      struct lock_info_t {
+        map<locker_id_t, locker_info_t> lockers; // map of lockers
+        ClsLockType lock_type;                   // lock type (exclusive / shared)
+        string tag;                              // tag: operations on lock can only succeed with this tag
+                                                 //      as long as set of non expired lockers
+                                                 //      is bigger than 0.
+
+        void encode(bufferlist &bl, uint64_t features) const {
+          ENCODE_START(1, 1, bl);
+          encode(lockers, bl, features);
+          uint8_t t = (uint8_t)lock_type;
+          encode(t, bl);
+          encode(tag, bl);
+          ENCODE_FINISH(bl);
+        }
+        void decode(bufferlist::const_iterator &bl) {
+          DECODE_START_LEGACY_COMPAT_LEN(1, 1, 1, bl);
+          decode(lockers, bl);
+          uint8_t t;
+          decode(t, bl);
+          lock_type = (ClsLockType)t;
+          decode(tag, bl);
+          DECODE_FINISH(bl);
+        }
+        lock_info_t() : lock_type(LOCK_NONE) {}
+        void dump(Formatter *f) const;
+        static void generate_test_instances(list<lock_info_t *>& o);
+      };
+      WRITE_CLASS_ENCODER_FEATURES(lock_info_t);
+    }
+  }
+}
+
+#endif
diff --git a/src/cls/log/cls_log.cc b/src/cls/log/cls_log.cc
new file mode 100644
index 00000000..3de35484
--- /dev/null
+++ b/src/cls/log/cls_log.cc
@@ -0,0 +1,317 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "include/types.h"
+#include "include/utime.h"
+#include "objclass/objclass.h"
+
+#include "cls_log_types.h"
+#include "cls_log_ops.h"
+
+#include "global/global_context.h"
+#include "include/compat.h"
+
+CLS_VER(1,0)
+CLS_NAME(log)
+
+static string log_index_prefix = "1_";
+
+
+static int write_log_entry(cls_method_context_t hctx, string& index, cls_log_entry& entry)
+{
+  bufferlist bl;
+  encode(entry, bl);
+
+  int ret = cls_cxx_map_set_val(hctx, index, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static void get_index_time_prefix(utime_t& ts, string& index)
+{
+  char buf[32];
+  snprintf(buf, sizeof(buf), "%010ld.%06ld_", (long)ts.sec(), (long)ts.usec());
+
+  index = log_index_prefix + buf;
+}
+
+static int read_header(cls_method_context_t hctx, cls_log_header& header)
+{
+  bufferlist header_bl;
+
+  int ret = cls_cxx_map_read_header(hctx, &header_bl);
+  if (ret < 0)
+    return ret;
+
+  if (header_bl.length() == 0) {
+    header = cls_log_header();
+    return 0;
+  }
+
+  auto iter = header_bl.cbegin();
+  try {
+    decode(header, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: read_header(): failed to decode header");
+  }
+
+  return 0;
+}
+
+static int write_header(cls_method_context_t hctx, cls_log_header& header)
+{
+  bufferlist header_bl;
+  encode(header, header_bl);
+
+  int ret = cls_cxx_map_write_header(hctx, &header_bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static void get_index(cls_method_context_t hctx, utime_t& ts, string& index)
+{
+  get_index_time_prefix(ts, index);
+
+  string unique_id;
+
+  cls_cxx_subop_version(hctx, &unique_id);
+
+  index.append(unique_id);
+}
+
+static int cls_log_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_log_add_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_log_add_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  cls_log_header header;
+
+  int ret = read_header(hctx, header);
+  if (ret < 0)
+    return ret;
+
+  for (list<cls_log_entry>::iterator iter = op.entries.begin();
+       iter != op.entries.end(); ++iter) {
+    cls_log_entry& entry = *iter;
+
+    string index;
+
+    utime_t timestamp = entry.timestamp;
+    if (op.monotonic_inc && timestamp < header.max_time)
+      timestamp = header.max_time;
+    else if (timestamp > header.max_time)
+      header.max_time = timestamp;
+
+    if (entry.id.empty()) {
+      get_index(hctx, timestamp, index);
+      entry.id = index;
+    } else {
+      index = entry.id;
+    }
+
+    CLS_LOG(20, "storing entry at %s", index.c_str());
+
+
+    if (index > header.max_marker)
+      header.max_marker = index;
+
+    ret = write_log_entry(hctx, index, entry);
+    if (ret < 0)
+      return ret;
+  }
+
+  ret = write_header(hctx, header);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_log_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_log_list_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_log_list_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> keys;
+
+  string from_index;
+  string to_index;
+
+  if (op.marker.empty()) {
+    get_index_time_prefix(op.from_time, from_index);
+  } else {
+    from_index = op.marker;
+  }
+  bool use_time_boundary = (!op.from_time.is_zero() && (op.to_time >= op.from_time));
+
+  if (use_time_boundary)
+    get_index_time_prefix(op.to_time, to_index);
+
+#define MAX_ENTRIES 1000
+  size_t max_entries = op.max_entries;
+  if (!max_entries || max_entries > MAX_ENTRIES)
+    max_entries = MAX_ENTRIES;
+
+  cls_log_list_ret ret;
+
+  int rc = cls_cxx_map_get_vals(hctx, from_index, log_index_prefix, max_entries, &keys, &ret.truncated);
+  if (rc < 0)
+    return rc;
+
+  list<cls_log_entry>& entries = ret.entries;
+  map<string, bufferlist>::iterator iter = keys.begin();
+
+  string marker;
+
+  for (; iter != keys.end(); ++iter) {
+    const string& index = iter->first;
+    marker = index;
+    if (use_time_boundary && index.compare(0, to_index.size(), to_index) >= 0) {
+      ret.truncated = false;
+      break;
+    }
+
+    bufferlist& bl = iter->second;
+    auto biter = bl.cbegin();
+    try {
+      cls_log_entry e;
+      decode(e, biter);
+      entries.push_back(e);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR: cls_log_list: could not decode entry, index=%s", index.c_str());
+    }
+  }
+
+  ret.marker = marker;
+
+  encode(ret, *out);
+
+  return 0;
+}
+
+
+static int cls_log_trim(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_log_trim_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: cls_log_list_op(): failed to decode entry");
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> keys;
+
+  string from_index;
+  string to_index;
+
+  if (op.from_marker.empty()) {
+    get_index_time_prefix(op.from_time, from_index);
+  } else {
+    from_index = op.from_marker;
+  }
+  if (op.to_marker.empty()) {
+    get_index_time_prefix(op.to_time, to_index);
+  } else {
+    to_index = op.to_marker;
+  }
+
+#define MAX_TRIM_ENTRIES 1000
+  size_t max_entries = MAX_TRIM_ENTRIES;
+  bool more;
+
+  int rc = cls_cxx_map_get_vals(hctx, from_index, log_index_prefix, max_entries, &keys, &more);
+  if (rc < 0)
+    return rc;
+
+  map<string, bufferlist>::iterator iter = keys.begin();
+
+  bool removed = false;
+  for (; iter != keys.end(); ++iter) {
+    const string& index = iter->first;
+
+    CLS_LOG(20, "index=%s to_index=%s", index.c_str(), to_index.c_str());
+
+    if (index.compare(0, to_index.size(), to_index) > 0)
+      break;
+
+    CLS_LOG(20, "removing key: index=%s", index.c_str());
+
+    int rc = cls_cxx_map_remove_key(hctx, index);
+    if (rc < 0) {
+      CLS_LOG(1, "ERROR: cls_cxx_map_remove_key failed rc=%d", rc);
+      return -EINVAL;
+    }
+    removed = true;
+  }
+
+  if (!removed)
+    return -ENODATA;
+
+  return 0;
+}
+
+static int cls_log_info(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_log_info_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_log_add_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  cls_log_info_ret ret;
+
+  int rc = read_header(hctx, ret.header);
+  if (rc < 0)
+    return rc;
+
+  encode(ret, *out);
+
+  return 0;
+}
+
+CLS_INIT(log)
+{
+  CLS_LOG(1, "Loaded log class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_log_add;
+  cls_method_handle_t h_log_list;
+  cls_method_handle_t h_log_trim;
+  cls_method_handle_t h_log_info;
+
+  cls_register("log", &h_class);
+
+  /* log */
+  cls_register_cxx_method(h_class, "add", CLS_METHOD_RD | CLS_METHOD_WR, cls_log_add, &h_log_add);
+  cls_register_cxx_method(h_class, "list", CLS_METHOD_RD, cls_log_list, &h_log_list);
+  cls_register_cxx_method(h_class, "trim", CLS_METHOD_RD | CLS_METHOD_WR, cls_log_trim, &h_log_trim);
+  cls_register_cxx_method(h_class, "info", CLS_METHOD_RD, cls_log_info, &h_log_info);
+
+  return;
+}
+
diff --git a/src/cls/log/cls_log_client.cc b/src/cls/log/cls_log_client.cc
new file mode 100644
index 00000000..accc9dd1
--- /dev/null
+++ b/src/cls/log/cls_log_client.cc
@@ -0,0 +1,154 @@
+#include <errno.h>
+
+#include "cls/log/cls_log_ops.h"
+#include "include/rados/librados.hpp"
+#include "include/compat.h"
+
+
+using namespace librados;
+
+
+
+void cls_log_add(librados::ObjectWriteOperation& op, list<cls_log_entry>& entries, bool monotonic_inc)
+{
+  bufferlist in;
+  cls_log_add_op call;
+  call.entries = entries;
+  encode(call, in);
+  op.exec("log", "add", in);
+}
+
+void cls_log_add(librados::ObjectWriteOperation& op, cls_log_entry& entry)
+{
+  bufferlist in;
+  cls_log_add_op call;
+  call.entries.push_back(entry);
+  encode(call, in);
+  op.exec("log", "add", in);
+}
+
+void cls_log_add_prepare_entry(cls_log_entry& entry, const utime_t& timestamp,
+                 const string& section, const string& name, bufferlist& bl)
+{
+  entry.timestamp = timestamp;
+  entry.section = section;
+  entry.name = name;
+  entry.data = bl;
+}
+
+void cls_log_add(librados::ObjectWriteOperation& op, const utime_t& timestamp,
+                 const string& section, const string& name, bufferlist& bl)
+{
+  cls_log_entry entry;
+
+  cls_log_add_prepare_entry(entry, timestamp, section, name, bl);
+  cls_log_add(op, entry);
+}
+
+void cls_log_trim(librados::ObjectWriteOperation& op, const utime_t& from_time, const utime_t& to_time,
+                  const string& from_marker, const string& to_marker)
+{
+  bufferlist in;
+  cls_log_trim_op call;
+  call.from_time = from_time;
+  call.to_time = to_time;
+  call.from_marker = from_marker;
+  call.to_marker = to_marker;
+  encode(call, in);
+  op.exec("log", "trim", in);
+}
+
+int cls_log_trim(librados::IoCtx& io_ctx, const string& oid, const utime_t& from_time, const utime_t& to_time,
+                 const string& from_marker, const string& to_marker)
+{
+  bool done = false;
+
+  do {
+    ObjectWriteOperation op;
+
+    cls_log_trim(op, from_time, to_time, from_marker, to_marker);
+
+    int r = io_ctx.operate(oid, &op);
+    if (r == -ENODATA)
+      done = true;
+    else if (r < 0)
+      return r;
+
+  } while (!done);
+
+
+  return 0;
+}
+
+class LogListCtx : public ObjectOperationCompletion {
+  list<cls_log_entry> *entries;
+  string *marker;
+  bool *truncated;
+public:
+  LogListCtx(list<cls_log_entry> *_entries, string *_marker, bool *_truncated) :
+                                      entries(_entries), marker(_marker), truncated(_truncated) {}
+  void handle_completion(int r, bufferlist& outbl) override {
+    if (r >= 0) {
+      cls_log_list_ret ret;
+      try {
+        auto iter = outbl.cbegin();
+        decode(ret, iter);
+        if (entries)
+          *entries = std::move(ret.entries);
+        if (truncated)
+          *truncated = ret.truncated;
+        if (marker)
+          *marker = std::move(ret.marker);
+      } catch (buffer::error& err) {
+        // nothing we can do about it atm
+      }
+    }
+  }
+};
+
+void cls_log_list(librados::ObjectReadOperation& op, utime_t& from, utime_t& to,
+                  const string& in_marker, int max_entries,
+		  list<cls_log_entry>& entries,
+                  string *out_marker, bool *truncated)
+{
+  bufferlist inbl;
+  cls_log_list_op call;
+  call.from_time = from;
+  call.to_time = to;
+  call.marker = in_marker;
+  call.max_entries = max_entries;
+
+  encode(call, inbl);
+
+  op.exec("log", "list", inbl, new LogListCtx(&entries, out_marker, truncated));
+}
+
+class LogInfoCtx : public ObjectOperationCompletion {
+  cls_log_header *header;
+public:
+  explicit LogInfoCtx(cls_log_header *_header) : header(_header) {}
+  void handle_completion(int r, bufferlist& outbl) override {
+    if (r >= 0) {
+      cls_log_info_ret ret;
+      try {
+        auto iter = outbl.cbegin();
+        decode(ret, iter);
+        if (header)
+	  *header = ret.header;
+      } catch (buffer::error& err) {
+        // nothing we can do about it atm
+      }
+    }
+  }
+};
+
+void cls_log_info(librados::ObjectReadOperation& op, cls_log_header *header)
+{
+  bufferlist inbl;
+  cls_log_info_op call;
+
+  encode(call, inbl);
+
+  op.exec("log", "info", inbl, new LogInfoCtx(header));
+}
+
diff --git a/src/cls/log/cls_log_client.h b/src/cls/log/cls_log_client.h
new file mode 100644
index 00000000..0e701ff9
--- /dev/null
+++ b/src/cls/log/cls_log_client.h
@@ -0,0 +1,31 @@
+#ifndef CEPH_CLS_LOG_CLIENT_H
+#define CEPH_CLS_LOG_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include "cls_log_types.h"
+
+/*
+ * log objclass
+ */
+
+void cls_log_add_prepare_entry(cls_log_entry& entry, const utime_t& timestamp,
+                 const string& section, const string& name, bufferlist& bl);
+
+void cls_log_add(librados::ObjectWriteOperation& op, list<cls_log_entry>& entries, bool monotonic_inc);
+void cls_log_add(librados::ObjectWriteOperation& op, cls_log_entry& entry);
+void cls_log_add(librados::ObjectWriteOperation& op, const utime_t& timestamp,
+                 const string& section, const string& name, bufferlist& bl);
+
+void cls_log_list(librados::ObjectReadOperation& op, utime_t& from, utime_t& to,
+                  const string& in_marker, int max_entries,
+		  list<cls_log_entry>& entries,
+                  string *out_marker, bool *truncated);
+
+void cls_log_trim(librados::ObjectWriteOperation& op, const utime_t& from_time, const utime_t& to_time,
+                  const string& from_marker, const string& to_marker);
+int cls_log_trim(librados::IoCtx& io_ctx, const string& oid, const utime_t& from_time, const utime_t& to_time,
+                 const string& from_marker, const string& to_marker);
+
+void cls_log_info(librados::ObjectReadOperation& op, cls_log_header *header);
+
+#endif
diff --git a/src/cls/log/cls_log_ops.h b/src/cls/log/cls_log_ops.h
new file mode 100644
index 00000000..2967da6e
--- /dev/null
+++ b/src/cls/log/cls_log_ops.h
@@ -0,0 +1,156 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_LOG_OPS_H
+#define CEPH_CLS_LOG_OPS_H
+
+#include "cls_log_types.h"
+
+struct cls_log_add_op {
+  list<cls_log_entry> entries;
+  bool monotonic_inc;
+
+  cls_log_add_op() : monotonic_inc(true) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(entries, bl);
+    encode(monotonic_inc, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(entries, bl);
+    if (struct_v >= 2) {
+      decode(monotonic_inc, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_add_op)
+
+struct cls_log_list_op {
+  utime_t from_time;
+  string marker; /* if not empty, overrides from_time */
+  utime_t to_time; /* not inclusive */
+  int max_entries; /* upperbound to returned num of entries
+                      might return less than that and still be truncated */
+
+  cls_log_list_op() : max_entries(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(from_time, bl);
+    encode(marker, bl);
+    encode(to_time, bl);
+    encode(max_entries, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(from_time, bl);
+    decode(marker, bl);
+    decode(to_time, bl);
+    decode(max_entries, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_list_op)
+
+struct cls_log_list_ret {
+  list<cls_log_entry> entries;
+  string marker;
+  bool truncated;
+
+  cls_log_list_ret() : truncated(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(marker, bl);
+    encode(truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(marker, bl);
+    decode(truncated, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_list_ret)
+
+
+/*
+ * operation will return 0 when successfully removed but not done. Will return
+ * -ENODATA when done, so caller needs to repeat sending request until that.
+ */
+struct cls_log_trim_op {
+  utime_t from_time;
+  utime_t to_time; /* inclusive */
+  string from_marker;
+  string to_marker;
+
+  cls_log_trim_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(from_time, bl);
+    encode(to_time, bl);
+    encode(from_marker, bl);
+    encode(to_marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(from_time, bl);
+    decode(to_time, bl);
+    if (struct_v >= 2) {
+    decode(from_marker, bl);
+    decode(to_marker, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_trim_op)
+
+struct cls_log_info_op {
+  cls_log_info_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    // currently empty request
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    // currently empty request
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_info_op)
+
+struct cls_log_info_ret {
+  cls_log_header header;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(header, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(header, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_info_ret)
+
+#endif
diff --git a/src/cls/log/cls_log_types.h b/src/cls/log/cls_log_types.h
new file mode 100644
index 00000000..5ec8628f
--- /dev/null
+++ b/src/cls/log/cls_log_types.h
@@ -0,0 +1,65 @@
+#ifndef CEPH_CLS_LOG_TYPES_H
+#define CEPH_CLS_LOG_TYPES_H
+
+#include "include/encoding.h"
+#include "include/types.h"
+
+#include "include/utime.h"
+
+class JSONObj;
+
+
+struct cls_log_entry {
+  string id;
+  string section;
+  string name;
+  utime_t timestamp;
+  bufferlist data;
+
+  cls_log_entry() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(section, bl);
+    encode(name, bl);
+    encode(timestamp, bl);
+    encode(data, bl);
+    encode(id, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(section, bl);
+    decode(name, bl);
+    decode(timestamp, bl);
+    decode(data, bl);
+    if (struct_v >= 2)
+      decode(id, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_entry)
+
+struct cls_log_header {
+  string max_marker;
+  utime_t max_time;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(max_marker, bl);
+    encode(max_time, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(max_marker, bl);
+    decode(max_time, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_log_header)
+
+
+#endif
diff --git a/src/cls/lua/cls_lua.cc b/src/cls/lua/cls_lua.cc
new file mode 100644
index 00000000..ecc5417f
--- /dev/null
+++ b/src/cls/lua/cls_lua.cc
@@ -0,0 +1,1052 @@
+/*
+ * Lua Bindings for RADOS Object Class
+ */
+#include <errno.h>
+#include <setjmp.h>
+#include <string>
+#include <sstream>
+#include <lua.hpp>
+#include "include/types.h"
+#include "objclass/objclass.h"
+#include "json_spirit/json_spirit.h"
+#include "cls_lua.h"
+#include "cls_lua_ops.h"
+
+CLS_VER(1,0)
+CLS_NAME(lua)
+
+/*
+ * Jump point for recovering from Lua panic.
+ */
+static jmp_buf cls_lua_panic_jump;
+
+/*
+ * Handle Lua panic.
+ */
+static int cls_lua_atpanic(lua_State *lua)
+{
+  CLS_ERR("error: Lua panic: %s", lua_tostring(lua, -1));
+  longjmp(cls_lua_panic_jump, 1);
+  return 0;
+}
+
+struct clslua_err {
+  bool error;
+  int ret;
+};
+
+/*
+ * Input parameter encoding.
+ */
+enum InputEncoding {
+  JSON_ENC,
+  BUFFERLIST_ENC,
+};
+
+struct clslua_hctx {
+  struct clslua_err error;
+  InputEncoding in_enc;
+  int ret;
+
+  cls_method_context_t *hctx;
+  bufferlist *inbl;   // raw cls input
+  bufferlist *outbl;  // raw cls output
+
+  string script;      // lua script
+  string handler;     // lua handler
+  bufferlist input;   // lua handler input
+};
+
+/* Lua registry key for method context */
+static char clslua_hctx_reg_key;
+
+/*
+ * Grabs the full method handler context
+ */
+static clslua_hctx *__clslua_get_hctx(lua_State *L)
+{
+  /* lookup registry value */
+  lua_pushlightuserdata(L, &clslua_hctx_reg_key);
+  lua_gettable(L, LUA_REGISTRYINDEX);
+
+  /* check cls_lua assumptions */
+  ceph_assert(!lua_isnil(L, -1));
+  ceph_assert(lua_type(L, -1) == LUA_TLIGHTUSERDATA);
+
+  /* cast and cleanup stack */
+  clslua_hctx *hctx = (struct clslua_hctx *)lua_touserdata(L, -1);
+  lua_pop(L, 1);
+
+  return hctx;
+}
+
+/*
+ * Get the method context out of the registry. This is called at the beginning
+ * of each clx_cxx_* wrapper, and must be set before there is any chance a Lua
+ * script calling a 'cls' module function that requires it.
+ */
+static cls_method_context_t clslua_get_hctx(lua_State *L)
+{
+  struct clslua_hctx *hctx = __clslua_get_hctx(L);
+  return *hctx->hctx;
+}
+
+/*
+ * Returns a reference to cls_lua error state from registry.
+ */
+struct clslua_err *clslua_checkerr(lua_State *L)
+{
+  struct clslua_hctx *hctx = __clslua_get_hctx(L);
+  struct clslua_err *err = &hctx->error;
+  return err;
+}
+
+
+/* Registry key for real `pcall` function */
+static char clslua_pcall_reg_key;
+
+/*
+ * Wrap Lua pcall to check for errors thrown by cls_lua (e.g. I/O errors or
+ * bufferlist decoding errors). The global error is cleared before returning
+ * to the caller.
+ */
+static int clslua_pcall(lua_State *L)
+{
+  int nargs = lua_gettop(L);
+  lua_pushlightuserdata(L, &clslua_pcall_reg_key);
+  lua_gettable(L, LUA_REGISTRYINDEX);
+  lua_insert(L, 1);
+  lua_call(L, nargs, LUA_MULTRET);
+  struct clslua_err *err = clslua_checkerr(L);
+  ceph_assert(err);
+  if (err->error) {
+    err->error = false;
+    lua_pushinteger(L, err->ret);
+    lua_insert(L, -2);
+  }
+  return lua_gettop(L);
+}
+
+
+/*
+ * cls_log
+ */
+static int clslua_log(lua_State *L)
+{
+  int nargs = lua_gettop(L);
+
+  if (!nargs)
+    return 0;
+
+  int loglevel = LOG_LEVEL_DEFAULT;
+  bool custom_ll = false;
+
+  /* check if first arg can be a log level */
+  if (nargs > 1 && lua_isnumber(L, 1)) {
+    int ll = (int)lua_tonumber(L, 1);
+    if (ll >= 0) {
+      loglevel = ll;
+      custom_ll = true;
+    }
+  }
+
+  /* check space for args and seperators (" ") */
+  int nelems = ((nargs - (custom_ll ? 1 : 0)) * 2) - 1;
+  luaL_checkstack(L, nelems, "rados.log(..)");
+
+  for (int i = custom_ll ? 2 : 1; i <= nargs; i++) {
+    const char *part = lua_tostring(L, i);
+    if (!part) {
+      if (lua_type(L, i) == LUA_TBOOLEAN)
+        part = lua_toboolean(L, i) ? "true" : "false";
+      else
+        part = luaL_typename(L, i);
+    }
+    lua_pushstring(L, part);
+    if ((i+1) <= nargs)
+      lua_pushstring(L, " ");
+  }
+
+  /* join string parts and send to Ceph/reply log */
+  lua_concat(L, nelems);
+  CLS_LOG(loglevel, "%s", lua_tostring(L, -1));
+
+  /* concat leaves result at top of stack */
+  return 1;
+}
+
+static char clslua_registered_handle_reg_key;
+
+/*
+ * Register a function to be used as a handler target
+ */
+static int clslua_register(lua_State *L)
+{
+  luaL_checktype(L, 1, LUA_TFUNCTION);
+
+  /* get table of registered handlers */
+  lua_pushlightuserdata(L, &clslua_registered_handle_reg_key);
+  lua_gettable(L, LUA_REGISTRYINDEX);
+  ceph_assert(lua_type(L, -1) == LUA_TTABLE);
+
+  /* lookup function argument */
+  lua_pushvalue(L, 1);
+  lua_gettable(L, -2);
+
+  if (lua_isnil(L, -1)) {
+    lua_pushvalue(L, 1);
+    lua_pushvalue(L, 1);
+    lua_settable(L, -4);
+  } else {
+    lua_pushstring(L, "Cannot register handler more than once");
+    return lua_error(L);
+  }
+
+  return 0;
+}
+
+/*
+ * Check if a function is registered as a handler
+ */
+static void clslua_check_registered_handler(lua_State *L)
+{
+  luaL_checktype(L, -1, LUA_TFUNCTION);
+
+  /* get table of registered handlers */
+  lua_pushlightuserdata(L, &clslua_registered_handle_reg_key);
+  lua_gettable(L, LUA_REGISTRYINDEX);
+  ceph_assert(lua_type(L, -1) == LUA_TTABLE);
+
+  /* lookup function argument */
+  lua_pushvalue(L, -2);
+  lua_gettable(L, -2);
+
+  if (!lua_rawequal(L, -1, -3)) {
+    lua_pushstring(L, "Handler is not registered");
+    lua_error(L);
+  }
+
+  lua_pop(L, 2);
+}
+
+/*
+ * Handle result of a cls_cxx_* call. If @ok is non-zero then we return with
+ * the number of Lua return arguments on the stack. Otherwise we save error
+ * information in the registry and throw a Lua error.
+ */
+static int clslua_opresult(lua_State *L, int ok, int ret, int nargs,
+    bool error_on_stack = false)
+{
+  struct clslua_err *err = clslua_checkerr(L);
+
+  ceph_assert(err);
+  if (err->error) {
+    CLS_ERR("error: cls_lua state machine: unexpected error");
+    ceph_abort();
+  }
+
+  /* everything is cherry */
+  if (ok)
+    return nargs;
+
+  /* set error in registry */
+  err->error = true;
+  err->ret = ret;
+
+  /* push error message */
+  if (!error_on_stack)
+    lua_pushfstring(L, "%s", strerror(-ret));
+
+  return lua_error(L);
+}
+
+/*
+ * cls_cxx_create
+ */
+static int clslua_create(lua_State *lua)
+{
+  cls_method_context_t hctx = clslua_get_hctx(lua);
+  int exclusive = lua_toboolean(lua, 1);
+
+  int ret = cls_cxx_create(hctx, exclusive);
+  return clslua_opresult(lua, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_remove
+ */
+static int clslua_remove(lua_State *lua)
+{
+  cls_method_context_t hctx = clslua_get_hctx(lua);
+
+  int ret = cls_cxx_remove(hctx);
+  return clslua_opresult(lua, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_stat
+ */
+static int clslua_stat(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+
+  uint64_t size;
+  time_t mtime;
+  int ret = cls_cxx_stat(hctx, &size, &mtime);
+  if (!ret) {
+    lua_pushinteger(L, size);
+    lua_pushinteger(L, mtime);
+  }
+  return clslua_opresult(L, (ret == 0), ret, 2);
+}
+
+/*
+ * cls_cxx_read
+ */
+static int clslua_read(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  int offset = luaL_checkinteger(L, 1);
+  int length = luaL_checkinteger(L, 2);
+  bufferlist *bl = clslua_pushbufferlist(L, NULL);
+  int ret = cls_cxx_read(hctx, offset, length, bl);
+  return clslua_opresult(L, (ret >= 0), ret, 1);
+}
+
+/*
+ * cls_cxx_write
+ */
+static int clslua_write(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  int offset = luaL_checkinteger(L, 1);
+  int length = luaL_checkinteger(L, 2);
+  bufferlist *bl = clslua_checkbufferlist(L, 3);
+  int ret = cls_cxx_write(hctx, offset, length, bl);
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_write_full
+ */
+static int clslua_write_full(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  bufferlist *bl = clslua_checkbufferlist(L, 1);
+  int ret = cls_cxx_write_full(hctx, bl);
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_getxattr
+ */
+static int clslua_getxattr(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *name = luaL_checkstring(L, 1);
+  bufferlist  *bl = clslua_pushbufferlist(L, NULL);
+  int ret = cls_cxx_getxattr(hctx, name, bl);
+  return clslua_opresult(L, (ret >= 0), ret, 1);
+}
+
+/*
+ * cls_cxx_getxattrs
+ */
+static int clslua_getxattrs(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+
+  map<string, bufferlist> attrs;
+  int ret = cls_cxx_getxattrs(hctx, &attrs);
+  if (ret < 0)
+    return clslua_opresult(L, 0, ret, 0);
+
+  lua_createtable(L, 0, attrs.size());
+
+  for (auto it = attrs.cbegin(); it != attrs.cend(); it++) {
+    lua_pushstring(L, it->first.c_str());
+    bufferlist  *bl = clslua_pushbufferlist(L, NULL);
+    *bl = it->second; // xfer ownership... will be GC'd
+    lua_settable(L, -3);
+  }
+
+  return clslua_opresult(L, 1, ret, 1);
+}
+
+/*
+ * cls_cxx_setxattr
+ */
+static int clslua_setxattr(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *name = luaL_checkstring(L, 1);
+  bufferlist *bl = clslua_checkbufferlist(L, 2);
+  int ret = cls_cxx_setxattr(hctx, name, bl);
+  return clslua_opresult(L, (ret == 0), ret, 1);
+}
+
+/*
+ * cls_cxx_map_get_val
+ */
+static int clslua_map_get_val(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *key = luaL_checkstring(L, 1);
+  bufferlist  *bl = clslua_pushbufferlist(L, NULL);
+  int ret = cls_cxx_map_get_val(hctx, key, bl);
+  return clslua_opresult(L, (ret == 0), ret, 1);
+}
+
+/*
+ * cls_cxx_map_set_val
+ */
+static int clslua_map_set_val(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *key = luaL_checkstring(L, 1);
+  bufferlist *val = clslua_checkbufferlist(L, 2);
+  int ret = cls_cxx_map_set_val(hctx, key, val);
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_map_clear
+ */
+static int clslua_map_clear(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  int ret = cls_cxx_map_clear(hctx);
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_map_get_keys
+ */
+static int clslua_map_get_keys(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *start_after = luaL_checkstring(L, 1);
+  int max_to_get = luaL_checkinteger(L, 2);
+
+  std::set<string> keys;
+  bool more;
+  int ret = cls_cxx_map_get_keys(hctx, start_after, max_to_get, &keys, &more);
+  if (ret < 0)
+    return clslua_opresult(L, 0, ret, 0);
+
+  lua_createtable(L, 0, keys.size());
+
+  for (auto it = keys.cbegin(); it != keys.cend(); it++) {
+    const std::string& key = *it;
+    lua_pushstring(L, key.c_str());
+    lua_pushboolean(L, 1);
+    lua_settable(L, -3);
+  }
+
+  return clslua_opresult(L, 1, ret, 1);
+}
+
+/*
+ * cls_cxx_map_get_vals
+ */
+static int clslua_map_get_vals(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *start_after = luaL_checkstring(L, 1);
+  const char *filter_prefix= luaL_checkstring(L, 2);
+  int max_to_get = luaL_checkinteger(L, 3);
+
+  map<string, bufferlist> kvpairs;
+  bool more;
+  int ret = cls_cxx_map_get_vals(hctx, start_after, filter_prefix,
+      max_to_get, &kvpairs, &more);
+  if (ret < 0)
+    return clslua_opresult(L, 0, ret, 0);
+
+  lua_createtable(L, 0, kvpairs.size());
+
+  for (auto it = kvpairs.cbegin(); it != kvpairs.cend(); it++) {
+    lua_pushstring(L, it->first.c_str());
+    bufferlist  *bl = clslua_pushbufferlist(L, NULL);
+    *bl = it->second; // xfer ownership... will be GC'd
+    lua_settable(L, -3);
+  }
+
+  return clslua_opresult(L, 1, ret, 1);
+}
+
+/*
+ * cls_cxx_map_read_header
+ */
+static int clslua_map_read_header(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  bufferlist *bl = clslua_pushbufferlist(L, NULL);
+  int ret = cls_cxx_map_read_header(hctx, bl);
+  return clslua_opresult(L, (ret >= 0), ret, 1);
+}
+
+/*
+ * cls_cxx_map_write_header
+ */
+static int clslua_map_write_header(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  bufferlist *bl = clslua_checkbufferlist(L, 1);
+  int ret = cls_cxx_map_write_header(hctx, bl);
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_map_set_vals
+ */
+static int clslua_map_set_vals(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  luaL_checktype(L, 1, LUA_TTABLE);
+
+  map<string, bufferlist> kvpairs;
+
+  for (lua_pushnil(L); lua_next(L, 1); lua_pop(L, 1)) {
+    /*
+     * In the case of a numeric key a copy is made on the stack because
+     * converting to a string would otherwise manipulate the original key and
+     * cause problems for iteration.
+     */
+    string key;
+    int type_code = lua_type(L, -2);
+    switch (type_code) {
+      case LUA_TSTRING:
+        key.assign(lua_tolstring(L, -2, NULL));
+        break;
+
+      case LUA_TNUMBER:
+        lua_pushvalue(L, -2);
+        key.assign(lua_tolstring(L, -1, NULL));
+        lua_pop(L, 1);
+        break;
+
+      default:
+        lua_pushfstring(L, "map_set_vals: invalid key type (%s)",
+            lua_typename(L, type_code));
+        return clslua_opresult(L, 0, -EINVAL, 0, true);
+    }
+
+    bufferlist val;
+    type_code = lua_type(L, -1);
+    switch (type_code) {
+      case LUA_TSTRING:
+        {
+          size_t len;
+          const char *data = lua_tolstring(L, -1, &len);
+          val.append(data, len);
+        }
+        break;
+
+      default:
+        lua_pushfstring(L, "map_set_vals: invalid val type (%s) for key (%s)",
+            lua_typename(L, type_code), key.c_str());
+        return clslua_opresult(L, 0, -EINVAL, 0, true);
+    }
+
+    kvpairs[key] = val;
+  }
+
+  int ret = cls_cxx_map_set_vals(hctx, &kvpairs);
+
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_cxx_map_remove_key
+ */
+static int clslua_map_remove_key(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  const char *key = luaL_checkstring(L, 1);
+  int ret = cls_cxx_map_remove_key(hctx, key);
+  return clslua_opresult(L, (ret == 0), ret, 0);
+}
+
+/*
+ * cls_current_version
+ */
+static int clslua_current_version(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  uint64_t version = cls_current_version(hctx);
+  lua_pushinteger(L, version);
+  return clslua_opresult(L, 1, 0, 1);
+}
+
+/*
+ * cls_current_subop_num
+ */
+static int clslua_current_subop_num(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  int num = cls_current_subop_num(hctx);
+  lua_pushinteger(L, num);
+  return clslua_opresult(L, 1, 0, 1);
+}
+
+/*
+ * cls_current_subop_version
+ */
+static int clslua_current_subop_version(lua_State *L)
+{
+  cls_method_context_t hctx = clslua_get_hctx(L);
+  string s;
+  cls_cxx_subop_version(hctx, &s);
+  lua_pushstring(L, s.c_str());
+  return clslua_opresult(L, 1, 0, 1);
+}
+
+/*
+ * Functions registered in the 'cls' module.
+ */
+static const luaL_Reg clslua_lib[] = {
+  // mgmt
+  {"register", clslua_register},
+  {"log", clslua_log},
+
+  // data
+  {"create", clslua_create},
+  {"remove", clslua_remove},
+  {"stat", clslua_stat},
+  {"read", clslua_read},
+  {"write", clslua_write},
+  {"write_full", clslua_write_full},
+
+  // xattr
+  {"getxattr", clslua_getxattr},
+  {"getxattrs", clslua_getxattrs},
+  {"setxattr", clslua_setxattr},
+
+  // omap
+  {"map_clear", clslua_map_clear},
+  {"map_get_keys", clslua_map_get_keys},
+  {"map_get_vals", clslua_map_get_vals},
+  {"map_read_header", clslua_map_read_header},
+  {"map_write_header", clslua_map_write_header},
+  {"map_get_val", clslua_map_get_val},
+  {"map_set_val", clslua_map_set_val},
+  {"map_set_vals", clslua_map_set_vals},
+  {"map_remove_key", clslua_map_remove_key},
+
+  // env
+  {"current_version", clslua_current_version},
+  {"current_subop_num", clslua_current_subop_num},
+  {"current_subop_version", clslua_current_subop_version},
+
+  {NULL, NULL}
+};
+
+/*
+ * Set const int in table at top of stack
+ */
+#define SET_INT_CONST(var) do { \
+  lua_pushinteger(L, var); \
+  lua_setfield(L, -2, #var); \
+} while (0)
+
+/*
+ *
+ */
+static int luaopen_objclass(lua_State *L)
+{
+  lua_newtable(L);
+
+  /*
+   * Register cls functions (cls.log, etc...)
+   */
+  luaL_setfuncs(L, clslua_lib, 0);
+
+  /*
+   * Register generic errno values under 'cls'
+   */
+  SET_INT_CONST(EPERM);
+  SET_INT_CONST(ENOENT);
+  SET_INT_CONST(ESRCH);
+  SET_INT_CONST(EINTR);
+  SET_INT_CONST(EIO);
+  SET_INT_CONST(ENXIO);
+  SET_INT_CONST(E2BIG);
+  SET_INT_CONST(ENOEXEC);
+  SET_INT_CONST(EBADF);
+  SET_INT_CONST(ECHILD);
+  SET_INT_CONST(EAGAIN);
+  SET_INT_CONST(ENOMEM);
+  SET_INT_CONST(EACCES);
+  SET_INT_CONST(EFAULT);
+  SET_INT_CONST(EBUSY);
+  SET_INT_CONST(EEXIST);
+  SET_INT_CONST(EXDEV);
+  SET_INT_CONST(ENODEV);
+  SET_INT_CONST(ENOTDIR);
+  SET_INT_CONST(EISDIR);
+  SET_INT_CONST(EINVAL);
+  SET_INT_CONST(ENFILE);
+  SET_INT_CONST(EMFILE);
+  SET_INT_CONST(ENOTTY);
+  SET_INT_CONST(EFBIG);
+  SET_INT_CONST(ENOSPC);
+  SET_INT_CONST(ESPIPE);
+  SET_INT_CONST(EROFS);
+  SET_INT_CONST(EMLINK);
+  SET_INT_CONST(EPIPE);
+  SET_INT_CONST(EDOM);
+  SET_INT_CONST(ERANGE);
+
+  return 1;
+}
+
+/*
+ * Setup the execution environment. Our sandbox currently is not
+ * sophisticated. With a new Lua state per-request we don't need to work about
+ * users stepping on each other, but we do rip out access to the local file
+ * system. All this will change when/if we decide to use some shared Lua
+ * states, most likely for performance reasons.
+ */
+static void clslua_setup_env(lua_State *L)
+{
+  luaL_requiref(L, "_G", luaopen_base, 1);
+  lua_pop(L, 1);
+
+  /*
+   * Wrap `pcall` to intercept errors. First save a reference to the default
+   * Lua `pcall` function, and then replace `pcall` with our version.
+   */
+  lua_pushlightuserdata(L, &clslua_pcall_reg_key);
+  lua_getglobal(L, "pcall");
+  lua_settable(L, LUA_REGISTRYINDEX);
+
+  lua_pushcfunction(L, clslua_pcall);
+  lua_setglobal(L, "pcall");
+
+  /* mask unsafe */
+  lua_pushnil(L);
+  lua_setglobal(L, "loadfile");
+
+  /* mask unsafe */
+  lua_pushnil(L);
+  lua_setglobal(L, "dofile");
+
+  /* not integrated into our error handling */
+  lua_pushnil(L);
+  lua_setglobal(L, "xpcall");
+
+  luaL_requiref(L, LUA_TABLIBNAME, luaopen_table, 1);
+  lua_pop(L, 1);
+
+  luaL_requiref(L, LUA_STRLIBNAME, luaopen_string, 1);
+  lua_pop(L, 1);
+
+  luaL_requiref(L, LUA_MATHLIBNAME, luaopen_math, 1);
+  lua_pop(L, 1);
+
+  luaL_requiref(L, "objclass", luaopen_objclass, 1);
+  lua_pop(L, 1);
+
+  luaL_requiref(L, "bufferlist", luaopen_bufferlist, 1);
+  lua_pop(L, 1);
+}
+
+/*
+ * Schema:
+ * {
+ *   "script": "...",
+ *   "handler": "...",
+ *   "input": "..." # optional
+ * }
+ */
+static int unpack_json_command(lua_State *L, struct clslua_hctx *ctx,
+    std::string& script, std::string& handler, std::string& input,
+    size_t *input_len)
+{
+  std::string json_input(ctx->inbl->c_str());
+  json_spirit::mValue value;
+
+  if (!json_spirit::read(json_input, value)) {
+    CLS_ERR("error: unparseable JSON");
+    ctx->ret = -EINVAL;
+    return 1;
+  }
+
+  if (value.type() != json_spirit::obj_type) {
+    CLS_ERR("error: input not a JSON object");
+    ctx->ret = -EINVAL;
+    return 1;
+  }
+  json_spirit::mObject obj = value.get_obj();
+
+  // grab the script
+  std::map<std::string, json_spirit::mValue>::const_iterator it = obj.find("script");
+  if (it == obj.end()) {
+    CLS_ERR("error: 'script' field found in JSON object");
+    ctx->ret = -EINVAL;
+    return 1;
+  }
+
+  if (it->second.type() != json_spirit::str_type) {
+    CLS_ERR("error: script is not a string");
+    ctx->ret = -EINVAL;
+    return 1;
+  }
+  script = it->second.get_str();
+
+  // grab the target function/handler name
+  it = obj.find("handler");
+  if (it == obj.end()) {
+    CLS_ERR("error: no target handler found in JSON object");
+    ctx->ret = -EINVAL;
+    return 1;
+  }
+
+  if (it->second.type() != json_spirit::str_type) {
+    CLS_ERR("error: target handler is not a string");
+    ctx->ret = -EINVAL;
+    return 1;
+  }
+  handler = it->second.get_str();
+
+  // grab the input (optional)
+  it = obj.find("input");
+  if (it != obj.end()) {
+    if (it->second.type() != json_spirit::str_type) {
+      CLS_ERR("error: handler input is not a string");
+      ctx->ret = -EINVAL;
+      return 1;
+    }
+    input = it->second.get_str();
+    *input_len = input.size();
+  }
+
+  return 0;
+}
+
+/*
+ * Runs the script, and calls handler.
+ */
+static int clslua_eval(lua_State *L)
+{
+  struct clslua_hctx *ctx = __clslua_get_hctx(L);
+  ctx->ret = -EIO; /* assume failure */
+
+  /*
+   * Load modules, errno value constants, and other environment goodies. Must
+   * be done before loading/compiling the chunk.
+   */
+  clslua_setup_env(L);
+
+  /*
+   * Deserialize the input that contains the script, the name of the handler
+   * to call, and the handler input.
+   */
+  switch (ctx->in_enc) {
+    case JSON_ENC:
+      {
+        std::string input_str;
+        size_t input_str_len = 0;
+
+        // if there is an error decoding json then ctx->ret will be set and we
+        // return normally from this function.
+        if (unpack_json_command(L, ctx, ctx->script, ctx->handler, input_str,
+              &input_str_len))
+          return 0;
+
+        bufferptr bp(input_str.c_str(), input_str_len);
+        ctx->input.push_back(bp);
+      }
+      break;
+
+    case BUFFERLIST_ENC:
+      {
+        cls_lua_eval_op op;
+
+        try {
+          auto it = ctx->inbl->cbegin();
+          decode(op, it);
+        } catch (const buffer::error &err) {
+          CLS_ERR("error: could not decode ceph encoded input");
+          ctx->ret = -EINVAL;
+          return 0;
+        }
+
+        ctx->script.swap(op.script);
+        ctx->handler.swap(op.handler);
+        ctx->input = op.input;
+      }
+      break;
+
+    default:
+      CLS_ERR("error: unknown encoding type");
+      ctx->ret = -EFAULT;
+      ceph_abort();
+      return 0;
+  }
+
+  /*
+   * Create table to hold registered (valid) handlers.
+   *
+   * Must be done before running the script for the first time because the
+   * script will immediately try to register one or more handlers using
+   * cls.register(function), which depends on this table.
+   */
+  lua_pushlightuserdata(L, &clslua_registered_handle_reg_key);
+  lua_newtable(L);
+  lua_settable(L, LUA_REGISTRYINDEX);
+
+  /* load and compile chunk */
+  if (luaL_loadstring(L, ctx->script.c_str()))
+    return lua_error(L);
+
+  /* execute chunk */
+  lua_call(L, 0, 0);
+
+  /* no error, but nothing left to do */
+  if (!ctx->handler.size()) {
+    CLS_LOG(10, "no handler name provided");
+    ctx->ret = 0; /* success */
+    return 0;
+  }
+
+  lua_getglobal(L, ctx->handler.c_str());
+  if (lua_type(L, -1) != LUA_TFUNCTION) {
+    CLS_ERR("error: unknown handler or not function: %s", ctx->handler.c_str());
+    ctx->ret = -EOPNOTSUPP;
+    return 0;
+  }
+
+  /* throw error if function is not registered */
+  clslua_check_registered_handler(L);
+
+  /* setup the input/output bufferlists */
+  clslua_pushbufferlist(L, &ctx->input);
+  clslua_pushbufferlist(L, ctx->outbl);
+
+  /*
+   * Call the target Lua object class handler. If the call is successful then
+   * we will examine the return value here and store it in the context. Errors
+   * that occur are handled in the top-level eval() function.
+   */
+  int top = lua_gettop(L);
+  lua_call(L, 2, LUA_MULTRET);
+
+  /* store return value in context */
+  if (!(lua_gettop(L) + 3 - top))
+    lua_pushinteger(L, 0);
+  ctx->ret = luaL_checkinteger(L, -1);
+
+  return 0;
+}
+
+/*
+ * Main handler. Proxies the Lua VM and the Lua-defined handler.
+ */
+static int eval_generic(cls_method_context_t hctx, bufferlist *in, bufferlist *out,
+    InputEncoding in_enc)
+{
+  struct clslua_hctx ctx;
+  lua_State *L = NULL;
+  int ret = -EIO;
+
+  /* stash context for use in Lua VM */
+  ctx.hctx = &hctx;
+  ctx.inbl = in;
+  ctx.in_enc = in_enc;
+  ctx.outbl = out;
+  ctx.error.error = false;
+
+  /* build lua vm state */
+  L = luaL_newstate();
+  if (!L) {
+    CLS_ERR("error creating new Lua state");
+    goto out;
+  }
+
+  /* panic handler for unhandled errors */
+  lua_atpanic(L, &cls_lua_atpanic);
+
+  if (setjmp(cls_lua_panic_jump) == 0) {
+
+    /*
+     * Stash the handler context in the register. It contains the objclass
+     * method context, global error state, and the command and reply structs.
+     */
+    lua_pushlightuserdata(L, &clslua_hctx_reg_key);
+    lua_pushlightuserdata(L, &ctx);
+    lua_settable(L, LUA_REGISTRYINDEX);
+
+    /* Process the input and run the script */
+    lua_pushcfunction(L, clslua_eval);
+    ret = lua_pcall(L, 0, 0, 0);
+
+    /* Encountered an error? */
+    if (ret) {
+      struct clslua_err *err = clslua_checkerr(L);
+      if (!err) {
+        CLS_ERR("error: cls_lua state machine: unexpected error");
+        ceph_abort();
+      }
+
+      /* Error origin a cls_cxx_* method? */
+      if (err->error) {
+        ret = err->ret; /* cls_cxx_* return value */
+
+        /* Errors always abort. Fix up ret and log error */
+        if (ret >= 0) {
+          CLS_ERR("error: unexpected handler return value");
+          ret = -EFAULT;
+        }
+
+      } else
+        ret = -EIO; /* Generic error code */
+
+      CLS_ERR("error: %s", lua_tostring(L, -1));
+
+    } else {
+      /*
+       * No Lua error encountered while running the script, but the handler
+       * may still have returned an error code (e.g. an errno value).
+       */
+      ret = ctx.ret;
+    }
+
+  } else {
+    CLS_ERR("error: recovering from Lua panic");
+    ret = -EFAULT;
+  }
+
+out:
+  if (L)
+    lua_close(L);
+  return ret;
+}
+
+static int eval_json(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  return eval_generic(hctx, in, out, JSON_ENC);
+}
+
+static int eval_bufferlist(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  return eval_generic(hctx, in, out, BUFFERLIST_ENC);
+}
+
+CLS_INIT(lua)
+{
+  CLS_LOG(20, "Loaded lua class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_eval_json;
+  cls_method_handle_t h_eval_bufferlist;
+
+  cls_register("lua", &h_class);
+
+  cls_register_cxx_method(h_class, "eval_json",
+      CLS_METHOD_RD | CLS_METHOD_WR, eval_json, &h_eval_json);
+
+  cls_register_cxx_method(h_class, "eval_bufferlist",
+      CLS_METHOD_RD | CLS_METHOD_WR, eval_bufferlist, &h_eval_bufferlist);
+}
diff --git a/src/cls/lua/cls_lua.h b/src/cls/lua/cls_lua.h
new file mode 100644
index 00000000..70ce9a92
--- /dev/null
+++ b/src/cls/lua/cls_lua.h
@@ -0,0 +1,14 @@
+#ifndef CEPH_CLS_LUA_H
+#define CEPH_CLS_LUA_H
+
+#include <lua.hpp>
+#include "include/types.h"
+
+#define LOG_LEVEL_DEFAULT 10
+
+int luaopen_bufferlist(lua_State *L);
+
+bufferlist *clslua_checkbufferlist(lua_State *L, int pos = 1);
+bufferlist *clslua_pushbufferlist(lua_State *L, bufferlist *set);
+
+#endif
diff --git a/src/cls/lua/cls_lua_client.cc b/src/cls/lua/cls_lua_client.cc
new file mode 100644
index 00000000..0e6544a2
--- /dev/null
+++ b/src/cls/lua/cls_lua_client.cc
@@ -0,0 +1,34 @@
+#include <string>
+#include <vector>
+#include "include/encoding.h"
+#include "include/rados/librados.hpp"  // for IoCtx
+#include "cls_lua_client.h"
+#include "cls_lua_ops.h"
+
+using std::string;
+using std::vector;
+using librados::IoCtx;
+using librados::bufferlist;
+
+namespace cls_lua_client {
+  /*
+   * Currently the return code and return bufferlist are not wrapped in a
+   * protocol that allows object class vs Lua to be distinguished. For
+   * instance, -EOPNOTSUPP might refer to cls_lua not being found, but would
+   * also be returned when cls_lua is found, but a Lua handler is not found.
+   */
+  int exec(IoCtx& ioctx, const string& oid, const string& script,
+      const string& handler, bufferlist& input, bufferlist& output)
+  {
+    cls_lua_eval_op op;
+
+    op.script = script;
+    op.handler = handler;
+    op.input = input;
+
+    bufferlist inbl;
+    encode(op, inbl);
+
+    return ioctx.exec(oid, "lua", "eval_bufferlist", inbl, output);
+  }
+}
diff --git a/src/cls/lua/cls_lua_client.h b/src/cls/lua/cls_lua_client.h
new file mode 100644
index 00000000..7e7e164b
--- /dev/null
+++ b/src/cls/lua/cls_lua_client.h
@@ -0,0 +1,13 @@
+#ifndef CLS_LUA_CLIENT_HPP
+#define CLS_LUA_CLIENT_HPP
+#include <string>
+
+#include "include/rados/librados.hpp"
+
+namespace cls_lua_client {
+  int exec(librados::IoCtx& ioctx, const std::string& oid,
+      const std::string& script, const std::string& handler,
+      librados::bufferlist& inbl, librados::bufferlist& outbl);
+}
+
+#endif
diff --git a/src/cls/lua/cls_lua_ops.h b/src/cls/lua/cls_lua_ops.h
new file mode 100644
index 00000000..c4afbd8a
--- /dev/null
+++ b/src/cls/lua/cls_lua_ops.h
@@ -0,0 +1,31 @@
+#ifndef CEPH_CLS_LUA_OPS_H
+#define CEPH_CLS_LUA_OPS_H
+
+#include <string>
+
+#include "include/encoding.h"
+
+struct cls_lua_eval_op {
+  std::string script;
+  std::string handler;
+  bufferlist input;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(script, bl);
+    encode(handler, bl);
+    encode(input, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(script, bl);
+    decode(handler, bl);
+    decode(input, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_lua_eval_op)
+
+#endif
diff --git a/src/cls/lua/lua_bufferlist.cc b/src/cls/lua/lua_bufferlist.cc
new file mode 100644
index 00000000..995b4082
--- /dev/null
+++ b/src/cls/lua/lua_bufferlist.cc
@@ -0,0 +1,180 @@
+/*
+ * Lua module wrapping librados::bufferlist
+ */
+#include <errno.h>
+#include <string>
+#include <sstream>
+#include <math.h>
+#include <lua.hpp>
+#include "include/types.h"
+#include "include/buffer.h"
+#include "objclass/objclass.h"
+#include "cls/lua/cls_lua.h"
+
+#define LUA_BUFFERLIST "ClsLua.Bufferlist"
+
+struct bufferlist_wrap {
+  bufferlist *bl;
+  int gc; /* do garbage collect? */
+};
+
+static inline struct bufferlist_wrap *to_blwrap(lua_State *L, int pos = 1)
+{
+  return (bufferlist_wrap *)luaL_checkudata(L, pos, LUA_BUFFERLIST);
+}
+
+bufferlist *clslua_checkbufferlist(lua_State *L, int pos)
+{
+  struct bufferlist_wrap *blw = to_blwrap(L, pos);
+  return blw->bl;
+}
+
+/*
+ * Pushes a new bufferlist userdata object onto the stack. If @set is non-null
+ * it is assumed to be a bufferlist that should not be garbage collected.
+ */
+bufferlist *clslua_pushbufferlist(lua_State *L, bufferlist *set)
+{
+  bufferlist_wrap *blw = static_cast<bufferlist_wrap *>(lua_newuserdata(L, sizeof(*blw)));
+  blw->bl = set ? set : new bufferlist();
+  blw->gc = set ? 0 : 1;
+  luaL_getmetatable(L, LUA_BUFFERLIST);
+  lua_setmetatable(L, -2);
+  return blw->bl;
+}
+
+/*
+ * Create a new bufferlist
+ */
+static int bl_new(lua_State *L)
+{
+  clslua_pushbufferlist(L, NULL);
+  return 1;
+}
+
+/*
+ * Convert bufferlist to Lua string
+ */
+static int bl_str(lua_State *L)
+{
+  bufferlist *bl = clslua_checkbufferlist(L);
+  lua_pushlstring(L, bl->c_str(), bl->length());
+  return 1;
+}
+
+/*
+ * Append a Lua string to bufferlist
+ */
+static int bl_append(lua_State *L)
+{
+  bufferlist *bl = clslua_checkbufferlist(L);
+  luaL_checktype(L, 2, LUA_TSTRING);
+
+  size_t len;
+  const char *data = lua_tolstring(L, 2, &len);
+  bl->append(data, len);
+
+  return 0;
+}
+
+/*
+ * Return the length in bytes of bufferlist
+ */
+static int bl_len(lua_State *L)
+{
+  bufferlist *bl = clslua_checkbufferlist(L);
+  lua_pushinteger(L, bl->length());
+  return 1;
+}
+
+/*
+ * Perform byte-for-byte bufferlist equality test
+ */
+static int bl_eq(lua_State *L)
+{
+  bufferlist *bl1 = clslua_checkbufferlist(L, 1);
+  bufferlist *bl2 = clslua_checkbufferlist(L, 2);
+  lua_pushboolean(L, *bl1 == *bl2 ? 1 : 0);
+  return 1;
+}
+
+/*
+ * Bufferlist < operator
+ */
+static int bl_lt(lua_State *L)
+{
+  bufferlist *bl1 = clslua_checkbufferlist(L, 1);
+  bufferlist *bl2 = clslua_checkbufferlist(L, 2);
+  lua_pushboolean(L, *bl1 < *bl2 ? 1 : 0);
+  return 1;
+}
+
+/*
+ * Bufferlist <= operator
+ */
+static int bl_le(lua_State *L)
+{
+  bufferlist *bl1 = clslua_checkbufferlist(L, 1);
+  bufferlist *bl2 = clslua_checkbufferlist(L, 2);
+  lua_pushboolean(L, *bl1 <= *bl2 ? 1 : 0);
+  return 1;
+}
+
+/*
+ * Bufferlist concatentation
+ */
+static int bl_concat(lua_State *L)
+{
+  bufferlist *bl1 = clslua_checkbufferlist(L, 1);
+  bufferlist *bl2 = clslua_checkbufferlist(L, 2);
+  bufferlist *ret = clslua_pushbufferlist(L, NULL);
+  ret->append(bl1->c_str(), bl1->length());
+  ret->append(bl2->c_str(), bl2->length());
+  return 1;
+}
+
+/*
+ * Garbage collect bufferlist
+ */
+static int bl_gc(lua_State *L)
+{
+  struct bufferlist_wrap *blw = to_blwrap(L);
+  ceph_assert(blw);
+  ceph_assert(blw->bl);
+  if (blw->gc)
+    delete blw->bl;
+  return 0;
+}
+
+static const struct luaL_Reg bufferlist_methods[] = {
+  {"str", bl_str},
+  {"append", bl_append},
+  {"__concat", bl_concat},
+  {"__len", bl_len},
+  {"__lt", bl_lt},
+  {"__le", bl_le},
+  {"__gc", bl_gc},
+  {"__eq", bl_eq},
+  {NULL, NULL}
+};
+
+static const struct luaL_Reg bllib_f[] = {
+  {"new", bl_new},
+  {NULL, NULL}
+};
+
+LUALIB_API int luaopen_bufferlist(lua_State *L)
+{
+  /* Setup bufferlist user-data type */
+  luaL_newmetatable(L, LUA_BUFFERLIST);
+  lua_pushvalue(L, -1);
+  lua_setfield(L, -2, "__index");
+
+  luaL_setfuncs(L, bufferlist_methods, 0);
+  lua_pop(L, 1);
+
+  lua_newtable(L);
+  luaL_setfuncs(L, bllib_f, 0);
+
+  return 1;
+}
diff --git a/src/cls/numops/cls_numops.cc b/src/cls/numops/cls_numops.cc
new file mode 100644
index 00000000..bccfd6d3
--- /dev/null
+++ b/src/cls/numops/cls_numops.cc
@@ -0,0 +1,161 @@
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2015 CERN
+ *
+ * Author: Joaquim Rocha <joaquim.rocha@cern.ch>
+ *
+ *  This library is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU Lesser General Public
+ *  License as published by the Free Software Foundation; either
+ *  version 2.1 of the License, or (at your option) any later version.
+ *
+ */
+
+/** \file
+ *
+ * This is an OSD class that implements methods for object numeric options on
+ * its omap values.
+ *
+ */
+
+#include "objclass/objclass.h"
+#include <errno.h>
+#include <string>
+#include <sstream>
+#include <cstdio>
+#include <include/compat.h>
+
+#define DECIMAL_PRECISION 10
+
+CLS_VER(1,0)
+CLS_NAME(numops)
+
+static int add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string key, diff_str;
+
+  auto iter = in->cbegin();
+  try {
+    decode(key, iter);
+    decode(diff_str, iter);
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "add: invalid decode of input");
+    return -EINVAL;
+  }
+
+  char *end_ptr = 0;
+  double difference = strtod(diff_str.c_str(), &end_ptr);
+
+  if (end_ptr && *end_ptr != '\0') {
+    CLS_ERR("add: invalid input value: %s", diff_str.c_str());
+    return -EINVAL;
+  }
+
+  bufferlist bl;
+  int ret = cls_cxx_map_get_val(hctx, key, &bl);
+
+  double value;
+
+  if (ret == -ENODATA || bl.length() == 0) {
+    value = 0;
+  } else if (ret < 0) {
+    if (ret != -ENOENT) {
+      CLS_ERR("add: error reading omap key %s: %d", key.c_str(), ret);
+    }
+    return ret;
+  } else {
+    std::string stored_value(bl.c_str(), bl.length());
+    end_ptr = 0;
+    value = strtod(stored_value.c_str(), &end_ptr);
+
+    if (end_ptr && *end_ptr != '\0') {
+      CLS_ERR("add: invalid stored value: %s", stored_value.c_str());
+      return -EBADMSG;
+    }
+  }
+
+  value += difference;
+
+  std::stringstream stream;
+  stream << std::setprecision(DECIMAL_PRECISION) << value;
+
+  bufferlist new_value;
+  new_value.append(stream.str());
+
+  return cls_cxx_map_set_val(hctx, key, &new_value);
+}
+
+static int mul(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string key, diff_str;
+
+  auto iter = in->cbegin();
+  try {
+    decode(key, iter);
+    decode(diff_str, iter);
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "mul: invalid decode of input");
+    return -EINVAL;
+  }
+
+  char *end_ptr = 0;
+  double difference = strtod(diff_str.c_str(), &end_ptr);
+
+  if (end_ptr && *end_ptr != '\0') {
+    CLS_ERR("mul: invalid input value: %s", diff_str.c_str());
+    return -EINVAL;
+  }
+
+  bufferlist bl;
+  int ret = cls_cxx_map_get_val(hctx, key, &bl);
+
+  double value;
+
+  if (ret == -ENODATA || bl.length() == 0) {
+    value = 0;
+  } else if (ret < 0) {
+    if (ret != -ENOENT) {
+      CLS_ERR("mul: error reading omap key %s: %d", key.c_str(), ret);
+    }
+    return ret;
+  } else {
+    std::string stored_value(bl.c_str(), bl.length());
+    end_ptr = 0;
+    value = strtod(stored_value.c_str(), &end_ptr);
+
+    if (end_ptr && *end_ptr != '\0') {
+      CLS_ERR("mul: invalid stored value: %s", stored_value.c_str());
+      return -EBADMSG;
+    }
+  }
+
+  value *= difference;
+
+  std::stringstream stream;
+  stream << std::setprecision(DECIMAL_PRECISION) << value;
+
+  bufferlist new_value;
+  new_value.append(stream.str());
+
+  return cls_cxx_map_set_val(hctx, key, &new_value);
+}
+
+CLS_INIT(numops)
+{
+  CLS_LOG(20, "loading cls_numops");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_add;
+  cls_method_handle_t h_mul;
+
+  cls_register("numops", &h_class);
+
+  cls_register_cxx_method(h_class, "add",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          add, &h_add);
+
+  cls_register_cxx_method(h_class, "mul",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mul, &h_mul);
+}
diff --git a/src/cls/numops/cls_numops_client.cc b/src/cls/numops/cls_numops_client.cc
new file mode 100644
index 00000000..fa1a69f2
--- /dev/null
+++ b/src/cls/numops/cls_numops_client.cc
@@ -0,0 +1,79 @@
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2015 CERN
+ *
+ * Author: Joaquim Rocha <joaquim.rocha@cern.ch>
+ *
+ *  This library is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU Lesser General Public
+ *  License as published by the Free Software Foundation; either
+ *  version 2.1 of the License, or (at your option) any later version.
+ *
+ */
+
+#include "cls/numops/cls_numops_client.h"
+#include "include/encoding.h"
+#include "include/rados/librados.hpp"
+
+#include <errno.h>
+#include <sstream>
+
+namespace rados {
+  namespace cls {
+    namespace numops {
+
+      int add(librados::IoCtx *ioctx,
+              const std::string& oid,
+              const std::string& key,
+              double value_to_add)
+      {
+        bufferlist in, out;
+        encode(key, in);
+
+        std::stringstream stream;
+        stream << value_to_add;
+
+        encode(stream.str(), in);
+
+        return ioctx->exec(oid, "numops", "add", in, out);
+      }
+
+      int sub(librados::IoCtx *ioctx,
+              const std::string& oid,
+              const std::string& key,
+              double value_to_subtract)
+      {
+        return add(ioctx, oid, key, -value_to_subtract);
+      }
+
+      int mul(librados::IoCtx *ioctx,
+              const std::string& oid,
+              const std::string& key,
+              double value_to_multiply)
+      {
+        bufferlist in, out;
+        encode(key, in);
+
+        std::stringstream stream;
+        stream << value_to_multiply;
+
+        encode(stream.str(), in);
+
+        return ioctx->exec(oid, "numops", "mul", in, out);
+      }
+
+      int div(librados::IoCtx *ioctx,
+              const std::string& oid,
+              const std::string& key,
+              double value_to_divide)
+      {
+        if (value_to_divide == 0)
+          return -EINVAL;
+
+        return mul(ioctx, oid, key, 1 / value_to_divide);
+      }
+
+    } // namespace numops
+  } // namespace cls
+} // namespace rados
diff --git a/src/cls/numops/cls_numops_client.h b/src/cls/numops/cls_numops_client.h
new file mode 100644
index 00000000..0b0ccbe5
--- /dev/null
+++ b/src/cls/numops/cls_numops_client.h
@@ -0,0 +1,50 @@
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2015 CERN
+ *
+ * Author: Joaquim Rocha <joaquim.rocha@cern.ch>
+ *
+ *  This library is free software; you can redistribute it and/or
+ *  modify it under the terms of the GNU Lesser General Public
+ *  License as published by the Free Software Foundation; either
+ *  version 2.1 of the License, or (at your option) any later version.
+ *
+ */
+
+#ifndef CEPH_LIBRBD_CLS_NUMOPS_CLIENT_H
+#define CEPH_LIBRBD_CLS_NUMOPS_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include <string>
+
+namespace rados {
+  namespace cls {
+    namespace numops {
+
+      extern int add(librados::IoCtx *ioctx,
+                     const std::string& oid,
+                     const std::string& key,
+                     double value_to_add);
+
+      extern int sub(librados::IoCtx *ioctx,
+                     const std::string& oid,
+                     const std::string& key,
+                     double value_to_subtract);
+
+      extern int mul(librados::IoCtx *ioctx,
+                     const std::string& oid,
+                     const std::string& key,
+                     double value_to_multiply);
+
+      extern int div(librados::IoCtx *ioctx,
+                     const std::string& oid,
+                     const std::string& key,
+                     double value_to_divide);
+
+    } // namespace numops
+  } // namespace cls
+} // namespace rados
+
+#endif // CEPH_LIBRBD_CLS_NUMOPS_CLIENT_H
+
diff --git a/src/cls/otp/cls_otp.cc b/src/cls/otp/cls_otp.cc
new file mode 100644
index 00000000..355e14da
--- /dev/null
+++ b/src/cls/otp/cls_otp.cc
@@ -0,0 +1,571 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+/** \file
+ *
+ * This is an OSD class that implements methods for management
+ * and use of otp (one time password).
+ *
+ */
+
+#include <errno.h>
+#include <map>
+#include <list>
+
+#include <boost/range/adaptor/reversed.hpp>
+
+#include <liboath/oath.h>
+
+#include "include/types.h"
+#include "include/utime.h"
+#include "objclass/objclass.h"
+
+#include "common/errno.h"
+#include "common/Clock.h"
+
+#include "cls/otp/cls_otp_ops.h"
+#include "cls/otp/cls_otp_types.h"
+
+
+using namespace rados::cls::otp;
+
+
+CLS_VER(1,0)
+CLS_NAME(otp)
+
+#define ATTEMPTS_PER_WINDOW 5
+
+static string otp_header_key = "header";
+static string otp_key_prefix = "otp/";
+
+struct otp_header {
+  set<string> ids;
+
+  otp_header() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(ids, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(ids, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(otp_header)
+
+struct otp_instance {
+  otp_info_t otp;
+
+  list<otp_check_t> last_checks; 
+  uint64_t last_success{0}; /* otp counter/step of last successful check */
+
+  otp_instance() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(otp, bl);
+    encode(last_checks, bl);
+    encode(last_success, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(otp, bl);
+    decode(last_checks, bl);
+    decode(last_success, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void trim_expired(const ceph::real_time& now);
+  void check(const string& token, const string& val, bool *update);
+  bool verify(const ceph::real_time& timestamp, const string& val);
+
+  void find(const string& token, otp_check_t *result);
+};
+WRITE_CLASS_ENCODER(otp_instance)
+
+
+void otp_instance::trim_expired(const ceph::real_time& now)
+{
+  ceph::real_time window_start = now - std::chrono::seconds(otp.step_size);
+
+  while (!last_checks.empty() &&
+         last_checks.front().timestamp < window_start) {
+    last_checks.pop_front();
+  }
+}
+
+void otp_instance::check(const string& token, const string& val, bool *update)
+{
+  ceph::real_time now = ceph::real_clock::now();
+  trim_expired(now);
+
+  if (last_checks.size() >= ATTEMPTS_PER_WINDOW) {
+    /* too many attempts */
+    *update = false;
+    return;
+  }
+
+  otp_check_t check;
+  check.token = token;
+  check.timestamp = now;
+  check.result = (verify(now, val) ? OTP_CHECK_SUCCESS : OTP_CHECK_FAIL);
+
+  last_checks.push_back(check);
+
+  *update = true;
+}
+
+bool otp_instance::verify(const ceph::real_time& timestamp, const string& val)
+{
+  uint64_t index;
+  uint32_t secs = (uint32_t)ceph::real_clock::to_time_t(timestamp);
+  int result = oath_totp_validate2(otp.seed_bin.c_str(), otp.seed_bin.length(),
+                                   secs, otp.step_size, otp.time_ofs, otp.window,
+                                   nullptr /* otp pos */,
+                                   val.c_str());
+  if (result == OATH_INVALID_OTP ||
+      result < 0) {
+    CLS_LOG(20, "otp check failed, result=%d", result);
+    return false;
+  }
+
+  index = result + (secs - otp.time_ofs) / otp.step_size;
+
+  if (index <= last_success) { /* already used value */
+    CLS_LOG(20, "otp, use of old token: index=%lld last_success=%lld", (long long)index, (long long)last_success);
+    return false;
+  }
+
+  last_success = index;
+
+  return true;
+}
+
+void otp_instance::find(const string& token, otp_check_t *result)
+{
+  ceph::real_time now = real_clock::now();
+  trim_expired(now);
+
+  for (auto& entry : boost::adaptors::reverse(last_checks)) {
+    if (entry.token == token) {
+      *result = entry;
+      return;
+    }
+  }
+  result->token = token;
+  result->result = OTP_CHECK_UNKNOWN;
+  result->timestamp = now;
+}
+
+static int get_otp_instance(cls_method_context_t hctx, const string& id, otp_instance *instance)
+{
+  bufferlist bl;
+  string key = otp_key_prefix + id;
+ 
+  int r = cls_cxx_map_get_val(hctx, key, &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading key %s: %d", key.c_str(), r);
+    }
+    return r;
+  }
+
+  try {
+    auto it = bl.cbegin();
+    decode(*instance, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: failed to decode %s", key.c_str());
+    return -EIO;
+  }
+
+  return 0;
+}
+
+static int write_otp_instance(cls_method_context_t hctx, const otp_instance& instance)
+{
+  string key = otp_key_prefix + instance.otp.id;
+
+  bufferlist bl;
+  encode(instance, bl);
+
+  int r = cls_cxx_map_set_val(hctx, key, &bl);
+  if (r < 0) {
+    CLS_ERR("ERROR: %s(): failed to store key (otp id=%s, r=%d)", __func__, instance.otp.id.c_str(), r);
+    return r;
+  }
+
+  return 0;
+}
+
+static int remove_otp_instance(cls_method_context_t hctx, const string& id)
+{
+  string key = otp_key_prefix + id;
+
+  int r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0) {
+    CLS_ERR("ERROR: %s(): failed to remove key (otp id=%s, r=%d)", __func__, id.c_str(), r);
+    return r;
+  }
+
+  return 0;
+}
+
+static int read_header(cls_method_context_t hctx, otp_header *h)
+{
+  bufferlist bl;
+  encode(h, bl);
+  int r = cls_cxx_map_get_val(hctx, otp_header_key, &bl);
+  if (r == -ENOENT || r == -ENODATA) {
+    *h = otp_header();
+    return 0;
+  }
+  if (r < 0) {
+    CLS_ERR("ERROR: %s(): failed to read header (r=%d)", __func__, r);
+    return r;
+  }
+
+  if (bl.length() == 0) {
+    *h = otp_header();
+    return 0;
+  }
+
+  auto iter = bl.cbegin();
+  try {
+    decode(*h, iter);
+  } catch (buffer::error& err) {
+    CLS_ERR("failed to decode otp_header");
+    return -EIO;
+  }
+
+  return 0;
+}
+
+static int write_header(cls_method_context_t hctx, const otp_header& h)
+{
+  bufferlist bl;
+  encode(h, bl);
+
+  int r = cls_cxx_map_set_val(hctx, otp_header_key, &bl);
+  if (r < 0) {
+    CLS_ERR("failed to store header (r=%d)", r);
+    return r;
+  }
+
+  return 0;
+}
+
+static int parse_seed(const string& seed, SeedType seed_type, bufferlist *seed_bin)
+{
+  size_t slen = seed.length();
+  char secret[seed.length()];
+  char *psecret = secret;
+  int result;
+  bool need_free = false;
+
+  seed_bin->clear();
+
+  switch (seed_type) {
+    case OTP_SEED_BASE32:
+      need_free = true; /* oath_base32_decode allocates dest buffer */
+      result = oath_base32_decode(seed.c_str(), seed.length(),
+                                  &psecret, &slen);
+      break;
+    default: /* just assume hex is the default */
+      result = oath_hex2bin(seed.c_str(), psecret, &slen);
+  }
+  if (result != OATH_OK) {
+    CLS_LOG(20, "failed to parse seed");
+    return -EINVAL;
+  }
+
+  seed_bin->append(psecret, slen);
+
+  if (need_free) {
+    free(psecret);
+  }
+
+  return 0;
+}
+
+static int otp_set_op(cls_method_context_t hctx,
+                       bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "%s", __func__);
+  cls_otp_set_otp_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  otp_header h;
+  int r = read_header(hctx, &h);
+  if (r < 0) {
+    return r;
+  }
+
+  for (auto entry : op.entries) {
+    otp_instance instance;
+    r = get_otp_instance(hctx, entry.id, &instance);
+    if (r < 0 &&
+        r != -ENOENT) {
+      return r;
+    }
+    instance.otp = entry;
+
+    r = parse_seed(instance.otp.seed, instance.otp.seed_type, &instance.otp.seed_bin);
+    if (r < 0) {
+      return r;
+    }
+
+    r = write_otp_instance(hctx, instance);
+    if (r < 0) {
+      return r;
+    }
+
+    h.ids.insert(entry.id);
+  }
+
+  r = write_header(hctx, h);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+static int otp_remove_op(cls_method_context_t hctx,
+                          bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "%s", __func__);
+  cls_otp_remove_otp_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  otp_header h;
+  bool removed_existing = false;
+  int r = read_header(hctx, &h);
+  if (r < 0) {
+    return r;
+  }
+
+  for (auto id : op.ids) {
+    bool existed = (h.ids.find(id) != h.ids.end());
+    removed_existing = (removed_existing || existed);
+
+    if (!existed) {
+      continue;
+    }
+
+    r = remove_otp_instance(hctx, id);
+    if (r < 0) {
+      return r;
+    }
+
+    h.ids.erase(id);
+  }
+
+  if (removed_existing) {
+    r = write_header(hctx, h);
+    if (r < 0) {
+      return r;
+    }
+  }
+
+  return 0;
+}
+
+static int otp_get_op(cls_method_context_t hctx,
+                          bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "%s", __func__);
+  cls_otp_get_otp_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  cls_otp_get_otp_reply result;
+
+  otp_header h;
+  int r;
+
+  r = read_header(hctx, &h);
+  if (r < 0) {
+    return r;
+  }
+
+  if (op.get_all) {
+    op.ids.clear();
+    for (auto id : h.ids) {
+      op.ids.push_back(id);
+    }
+  }
+
+  for (auto id : op.ids) {
+    bool exists = (h.ids.find(id) != h.ids.end());
+
+    if (!exists) {
+      continue;
+    }
+
+    otp_instance instance;
+    r = get_otp_instance(hctx, id, &instance);
+    if (r < 0) {
+      return r;
+    }
+
+    result.found_entries.push_back(instance.otp);
+  }
+
+  encode(result, *out);
+
+  return 0;
+}
+
+static int otp_check_op(cls_method_context_t hctx,
+                          bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "%s", __func__);
+  cls_otp_check_otp_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  otp_header h;
+  int r;
+
+  otp_instance instance;
+
+  r = get_otp_instance(hctx, op.id, &instance);
+  if (r < 0) {
+    return r;
+  }
+
+  bool update{false};
+  instance.check(op.token, op.val, &update);
+
+  if (update) {
+    r = write_otp_instance(hctx, instance);
+    if (r < 0) {
+      return r;
+    }
+  }
+
+  return 0;
+}
+
+static int otp_get_result(cls_method_context_t hctx,
+                          bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "%s", __func__);
+  cls_otp_check_otp_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  otp_header h;
+  int r;
+
+  otp_instance instance;
+
+  r = get_otp_instance(hctx, op.id, &instance);
+  if (r < 0) {
+    return r;
+  }
+
+  cls_otp_get_result_reply reply;
+  instance.find(op.token, &reply.result);
+  encode(reply, *out);
+
+  return 0;
+}
+
+static int otp_get_current_time_op(cls_method_context_t hctx,
+                        bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "%s", __func__);
+  cls_otp_get_current_time_op op;
+  try {
+    auto iter = in->cbegin();
+    decode(op, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  cls_otp_get_current_time_reply reply;
+  reply.time = real_clock::now();
+  encode(reply, *out);
+
+  return 0;
+}
+
+CLS_INIT(otp)
+{
+  CLS_LOG(20, "Loaded otp class!");
+
+  oath_init();
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_set_otp_op;
+  cls_method_handle_t h_get_otp_op;
+  cls_method_handle_t h_check_otp_op;
+  cls_method_handle_t h_get_result_op; /*
+                                        * need to check and get check result in two phases. The
+                                        * reason is that we need to update failure internally,
+                                        * however, there's no way to both return a failure and
+                                        * update, because a failure will cancel the operation,
+                                        * and write operations will not return a value. So
+                                        * we're returning a success, potentially updating the
+                                        * status internally, then a subsequent request can try
+                                        * to fetch the status. If it fails it means that failed
+                                        * to authenticate.
+                                        */
+  cls_method_handle_t h_remove_otp_op;
+  cls_method_handle_t h_get_current_time_op;
+
+  cls_register("otp", &h_class);
+  cls_register_cxx_method(h_class, "otp_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          otp_set_op, &h_set_otp_op);
+  cls_register_cxx_method(h_class, "otp_get",
+                          CLS_METHOD_RD,
+                          otp_get_op, &h_get_otp_op);
+  cls_register_cxx_method(h_class, "otp_check",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          otp_check_op, &h_check_otp_op);
+  cls_register_cxx_method(h_class, "otp_get_result",
+                          CLS_METHOD_RD,
+                          otp_get_result, &h_get_result_op);
+  cls_register_cxx_method(h_class, "otp_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          otp_remove_op, &h_remove_otp_op);
+  cls_register_cxx_method(h_class, "get_current_time",
+                          CLS_METHOD_RD,
+                          otp_get_current_time_op, &h_get_current_time_op);
+
+  return;
+}
diff --git a/src/cls/otp/cls_otp_client.cc b/src/cls/otp/cls_otp_client.cc
new file mode 100644
index 00000000..dc1efab4
--- /dev/null
+++ b/src/cls/otp/cls_otp_client.cc
@@ -0,0 +1,190 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software 
+ * Foundation.  See file COPYING.
+ * 
+ */
+
+#include "include/types.h"
+#include "msg/msg_types.h"
+#include "include/rados/librados.hpp"
+#include "include/utime.h"
+ 
+using namespace librados;
+
+#include "cls/otp/cls_otp_ops.h"
+#include "cls/otp/cls_otp_client.h"
+
+#include "common/random_string.h" /* for gen_rand_alphanumeric */
+
+namespace rados {
+  namespace cls {
+    namespace otp {
+
+      void OTP::create(librados::ObjectWriteOperation *rados_op,
+                       const otp_info_t& config) {
+        cls_otp_set_otp_op op;
+        op.entries.push_back(config);
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("otp", "otp_set", in);
+      }
+
+      void OTP::set(librados::ObjectWriteOperation *rados_op,
+                       const list<otp_info_t>& entries) {
+        cls_otp_set_otp_op op;
+        op.entries = entries;
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("otp", "otp_set", in);
+      }
+
+      void OTP::remove(librados::ObjectWriteOperation *rados_op,
+                       const string& id) {
+        cls_otp_remove_otp_op op;
+        op.ids.push_back(id);
+        bufferlist in;
+        encode(op, in);
+        rados_op->exec("otp", "otp_remove", in);
+      }
+
+      int OTP::check(CephContext *cct, librados::IoCtx& ioctx, const string& oid,
+                     const string& id, const string& val, otp_check_t *result) {
+        cls_otp_check_otp_op op;
+        op.id = id;
+        op.val = val;
+#define TOKEN_LEN 16
+        op.token = gen_rand_alphanumeric(cct, TOKEN_LEN);
+        
+        bufferlist in;
+        bufferlist out;
+        encode(op, in);
+        int r = ioctx.exec(oid, "otp", "otp_check", in, out);
+        if (r < 0) {
+          return r;
+        }
+
+        cls_otp_get_result_op op2;
+        op2.token = op.token;
+        bufferlist in2;
+        bufferlist out2;
+        encode(op2, in2);
+        r = ioctx.exec(oid, "otp", "otp_get_result", in, out);
+        if (r < 0) {
+          return r;
+        }
+
+        auto iter = out.cbegin();
+        cls_otp_get_result_reply ret;
+        try {
+          decode(ret, iter);
+        } catch (buffer::error& err) {
+	  return -EBADMSG;
+        }
+
+        *result = ret.result;
+
+        return 0;
+      }
+
+      int OTP::get(librados::ObjectReadOperation *rop,
+                   librados::IoCtx& ioctx, const string& oid,
+                   const list<string> *ids, bool get_all, list<otp_info_t> *result) {
+        librados::ObjectReadOperation _rop;
+        if (!rop) {
+          rop = &_rop;
+        }
+        cls_otp_get_otp_op op;
+        if (ids) {
+          op.ids = *ids;
+        }
+        op.get_all = get_all;
+        bufferlist in;
+        bufferlist out;
+        int op_ret;
+        encode(op, in);
+        rop->exec("otp", "otp_get", in, &out, &op_ret);
+        int r = ioctx.operate(oid, rop, nullptr);
+        if (r < 0) {
+          return r;
+        }
+        if (op_ret < 0) {
+          return op_ret;
+        }
+
+        cls_otp_get_otp_reply ret;
+        auto iter = out.cbegin();
+        try {
+          decode(ret, iter);
+        } catch (buffer::error& err) {
+	  return -EBADMSG;
+        }
+
+        *result = ret.found_entries;;
+
+        return 0;
+      }
+
+      int OTP::get(librados::ObjectReadOperation *op,
+                   librados::IoCtx& ioctx, const string& oid,
+                    const string& id, otp_info_t *result) {
+        list<string> ids{ id };
+        list<otp_info_t> ret;
+
+        int r = get(op, ioctx, oid, &ids, false, &ret);
+        if (r < 0) {
+          return r;
+        }
+        if (ret.empty()) {
+          return -ENOENT;
+        }
+        *result = ret.front();
+
+        return 0;
+      }
+
+      int OTP::get_all(librados::ObjectReadOperation *op, librados::IoCtx& ioctx, const string& oid,
+                       list<otp_info_t> *result) {
+        return get(op, ioctx, oid, nullptr, true, result);
+      }
+
+      int OTP::get_current_time(librados::IoCtx& ioctx, const string& oid,
+                                ceph::real_time *result) {
+        cls_otp_get_current_time_op op;
+        bufferlist in;
+        bufferlist out;
+        int op_ret;
+        encode(op, in);
+        ObjectReadOperation rop;
+        rop.exec("otp", "get_current_time", in, &out, &op_ret);
+        int r = ioctx.operate(oid, &rop, nullptr);
+        if (r < 0) {
+          return r;
+        }
+        if (op_ret < 0) {
+          return op_ret;
+        }
+
+        cls_otp_get_current_time_reply ret;
+        auto iter = out.cbegin();
+        try {
+          decode(ret, iter);
+        } catch (buffer::error& err) {
+	  return -EBADMSG;
+        }
+
+        *result = ret.time;
+
+        return 0;
+      }
+    } // namespace otp
+  } // namespace cls
+} // namespace rados
+
diff --git a/src/cls/otp/cls_otp_client.h b/src/cls/otp/cls_otp_client.h
new file mode 100644
index 00000000..f19c9459
--- /dev/null
+++ b/src/cls/otp/cls_otp_client.h
@@ -0,0 +1,56 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_OTP_CLIENT_H
+#define CEPH_CLS_OTP_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include "cls/otp/cls_otp_types.h"
+
+namespace rados {
+  namespace cls {
+    namespace otp {
+
+      class OTP {
+      public:
+        static void create(librados::ObjectWriteOperation *op, const otp_info_t& config);
+        static void set(librados::ObjectWriteOperation *op, const list<otp_info_t>& entries);
+        static void remove(librados::ObjectWriteOperation *op, const string& id);
+        static int get(librados::ObjectReadOperation *op,
+                       librados::IoCtx& ioctx, const string& oid,
+                       const list<string> *ids, bool get_all, list<otp_info_t> *result);
+        static int get(librados::ObjectReadOperation *op,
+                       librados::IoCtx& ioctx, const string& oid,
+                       const string& id, otp_info_t *result);
+        static int get_all(librados::ObjectReadOperation *op,
+                           librados::IoCtx& ioctx, const string& oid,
+                           list<otp_info_t> *result);
+        static int check(CephContext *cct, librados::IoCtx& ioctx, const string& oid,
+                         const string& id, const string& val, otp_check_t *result);
+        static int get_current_time(librados::IoCtx& ioctx, const string& oid,
+                                    ceph::real_time *result);
+      };
+
+      class TOTPConfig {
+        otp_info_t config;
+        public:
+          TOTPConfig(const string& id, const string& seed) {
+            config.type = OTP_TOTP;
+            config.id = id;
+            config.seed = seed;
+          }
+          void set_step_size(int step_size) {
+            config.step_size = step_size;
+          }
+          void set_window(int window) {
+            config.window = window;
+          }
+          void get_config(otp_info_t *conf) {
+            *conf = config;
+          }
+      };
+    } // namespace otp
+  }  // namespace cls
+} // namespace rados
+
+#endif
diff --git a/src/cls/otp/cls_otp_ops.h b/src/cls/otp/cls_otp_ops.h
new file mode 100644
index 00000000..51cec3eb
--- /dev/null
+++ b/src/cls/otp/cls_otp_ops.h
@@ -0,0 +1,166 @@
+#ifndef CEPH_CLS_OTP_OPS_H
+#define CEPH_CLS_OTP_OPS_H
+
+#include "include/types.h"
+#include "include/utime.h"
+#include "cls/otp/cls_otp_types.h"
+
+struct cls_otp_set_otp_op
+{
+  list<rados::cls::otp::otp_info_t> entries;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_set_otp_op)
+
+struct cls_otp_check_otp_op
+{
+  string id;
+  string val;
+  string token;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(id, bl);
+    encode(val, bl);
+    encode(token, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(id, bl);
+    decode(val, bl);
+    decode(token, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_check_otp_op)
+
+struct cls_otp_get_result_op
+{
+  string token;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(token, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(token, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_get_result_op)
+
+struct cls_otp_get_result_reply
+{
+  rados::cls::otp::otp_check_t result;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(result, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(result, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_get_result_reply)
+
+struct cls_otp_remove_otp_op
+{
+  list<string> ids;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(ids, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(ids, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_remove_otp_op)
+
+struct cls_otp_get_otp_op
+{
+  bool get_all{false};
+  list<string> ids;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(get_all, bl);
+    encode(ids, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(get_all, bl);
+    decode(ids, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_get_otp_op)
+
+struct cls_otp_get_otp_reply
+{
+  list<rados::cls::otp::otp_info_t> found_entries;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(found_entries, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(found_entries, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_get_otp_reply)
+
+struct cls_otp_get_current_time_op
+{
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_get_current_time_op)
+
+struct cls_otp_get_current_time_reply
+{
+  ceph::real_time time;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(time, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(time, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_otp_get_current_time_reply)
+
+#endif
diff --git a/src/cls/otp/cls_otp_types.cc b/src/cls/otp/cls_otp_types.cc
new file mode 100644
index 00000000..b8a6e012
--- /dev/null
+++ b/src/cls/otp/cls_otp_types.cc
@@ -0,0 +1,67 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
+// vim: ts=8 sw=2 smarttab
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
+ *
+ * This is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU Lesser General Public
+ * License version 2.1, as published by the Free Software 
+ * Foundation.  See file COPYING.
+ * 
+ */
+
+#include "objclass/objclass.h"
+#include "common/Formatter.h"
+#include "common/Clock.h"
+#include "common/ceph_json.h"
+
+#include "include/utime.h"
+
+#include "cls/otp/cls_otp_types.h"
+
+using namespace rados::cls::otp;
+
+void otp_info_t::dump(Formatter *f) const
+{
+  encode_json("type", (int)type, f);
+  encode_json("id", id, f);
+  encode_json("seed", seed, f);
+  string st;
+  switch (seed_type) {
+    case rados::cls::otp::OTP_SEED_HEX:
+      st = "hex";
+      break;
+    case rados::cls::otp::OTP_SEED_BASE32:
+      st = "base32";
+      break;
+    default:
+      st = "unknown";
+  }
+  encode_json("seed_type", st, f);
+  encode_json("time_ofs", time_ofs, f);
+  encode_json("step_size", step_size, f);
+  encode_json("window", window, f);
+}
+
+void otp_info_t::decode_json(JSONObj *obj)
+{
+  int t{-1};
+  JSONDecoder::decode_json("type", t, obj);
+  type = (OTPType)t;
+  JSONDecoder::decode_json("id", id, obj);
+  JSONDecoder::decode_json("seed", seed, obj);
+  string st;
+  JSONDecoder::decode_json("seed_type", st, obj);
+  if (st == "hex") {
+    seed_type = OTP_SEED_HEX;
+  } else if (st == "base32") {
+    seed_type = OTP_SEED_BASE32;
+  } else {
+    seed_type = OTP_SEED_UNKNOWN;
+  }
+  JSONDecoder::decode_json("time_ofs", time_ofs, obj);
+  JSONDecoder::decode_json("step_size", step_size, obj);
+  JSONDecoder::decode_json("window", window, obj);
+}
diff --git a/src/cls/otp/cls_otp_types.h b/src/cls/otp/cls_otp_types.h
new file mode 100644
index 00000000..b542b5cb
--- /dev/null
+++ b/src/cls/otp/cls_otp_types.h
@@ -0,0 +1,132 @@
+#ifndef CEPH_CLS_OTP_TYPES_H
+#define CEPH_CLS_OTP_TYPES_H
+
+#include "include/encoding.h"
+#include "include/types.h"
+
+
+#define CLS_OTP_MAX_REPO_SIZE 100
+
+class JSONObj;
+
+namespace rados {
+  namespace cls {
+    namespace otp {
+
+      enum OTPType {
+        OTP_UNKNOWN = 0,
+        OTP_HOTP = 1,  /* unsupported */
+        OTP_TOTP = 2,
+      };
+
+      enum SeedType {
+        OTP_SEED_UNKNOWN = 0,
+        OTP_SEED_HEX = 1,
+        OTP_SEED_BASE32 = 2,
+      };
+
+      struct otp_info_t {
+        OTPType type{OTP_TOTP};
+        string id;
+        string seed;
+        SeedType seed_type{OTP_SEED_UNKNOWN};
+        bufferlist seed_bin; /* parsed seed, built automatically by otp_set_op,
+                              * not being json encoded/decoded on purpose
+                              */
+        int32_t time_ofs{0};
+        uint32_t step_size{30}; /* num of seconds foreach otp to test */
+        uint32_t window{2}; /* num of otp after/before start otp to test */
+
+        otp_info_t() {}
+
+        void encode(bufferlist &bl) const {
+          ENCODE_START(1, 1, bl);
+          encode((uint8_t)type, bl);
+          /* if we ever implement anything other than TOTP
+           * then we'll need to branch here */
+          encode(id, bl);
+          encode(seed, bl);
+          encode((uint8_t)seed_type, bl);
+          encode(seed_bin, bl);
+          encode(time_ofs, bl);
+          encode(step_size, bl);
+          encode(window, bl);
+          ENCODE_FINISH(bl);
+        }
+        void decode(bufferlist::const_iterator &bl) {
+          DECODE_START(1, bl);
+          uint8_t t;
+          decode(t, bl);
+          type = (OTPType)t;
+          decode(id, bl);
+          decode(seed, bl);
+          uint8_t st;
+          decode(st, bl);
+          seed_type = (SeedType)st;
+          decode(seed_bin, bl);
+          decode(time_ofs, bl);
+          decode(step_size, bl);
+          decode(window, bl);
+          DECODE_FINISH(bl);
+        }
+        void dump(Formatter *f) const;
+        void decode_json(JSONObj *obj);
+      };
+      WRITE_CLASS_ENCODER(rados::cls::otp::otp_info_t)
+
+      enum OTPCheckResult {
+        OTP_CHECK_UNKNOWN = 0,
+        OTP_CHECK_SUCCESS = 1,
+        OTP_CHECK_FAIL = 2,
+      };
+
+      struct otp_check_t {
+        string token;
+        ceph::real_time timestamp;
+        OTPCheckResult result{OTP_CHECK_UNKNOWN};
+
+        void encode(bufferlist &bl) const {
+          ENCODE_START(1, 1, bl);
+          encode(token, bl);
+          encode(timestamp, bl);
+          encode((char)result, bl);
+          ENCODE_FINISH(bl);
+        }
+        void decode(bufferlist::const_iterator &bl) {
+          DECODE_START(1, bl);
+          decode(token, bl);
+          decode(timestamp, bl);
+          uint8_t t;
+          decode(t, bl);
+          result = (OTPCheckResult)t;
+          DECODE_FINISH(bl);
+        }
+      };
+      WRITE_CLASS_ENCODER(rados::cls::otp::otp_check_t)
+
+      struct otp_repo_t {
+        map<string, otp_info_t> entries;
+
+        otp_repo_t() {}
+
+        void encode(bufferlist &bl) const {
+          ENCODE_START(1, 1, bl);
+          encode(entries, bl);
+          ENCODE_FINISH(bl);
+        }
+        void decode(bufferlist::const_iterator &bl) {
+          DECODE_START(1, bl);
+          decode(entries, bl);
+          DECODE_FINISH(bl);
+        }
+      };
+      WRITE_CLASS_ENCODER(rados::cls::otp::otp_repo_t)
+    }
+  }
+}
+
+WRITE_CLASS_ENCODER(rados::cls::otp::otp_info_t)
+WRITE_CLASS_ENCODER(rados::cls::otp::otp_check_t)
+WRITE_CLASS_ENCODER(rados::cls::otp::otp_repo_t)
+
+#endif
diff --git a/src/cls/rbd/cls_rbd.cc b/src/cls/rbd/cls_rbd.cc
new file mode 100644
index 00000000..09f2e790
--- /dev/null
+++ b/src/cls/rbd/cls_rbd.cc
@@ -0,0 +1,7964 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+/** \file
+ *
+ * This is an OSD class that implements methods for
+ * use with rbd.
+ *
+ * Most of these deal with the rbd header object. Methods prefixed
+ * with old_ deal with the original rbd design, in which clients read
+ * and interpreted the header object directly.
+ *
+ * The new format is meant to be opaque to clients - all their
+ * interactions with non-data objects should go through this
+ * class. The OSD class interface leaves the class to implement its
+ * own argument and payload serialization/deserialization, so for ease
+ * of implementation we use the existing ceph encoding/decoding
+ * methods. Something like json might be preferable, but the rbd
+ * kernel module has to be able to understand format as well. The
+ * datatypes exposed to the clients are strings, unsigned integers,
+ * and vectors of those types. The on-wire format can be found in
+ * src/include/encoding.h.
+ *
+ * The methods for interacting with the new format document their
+ * parameters as the client sees them - it would be silly to mention
+ * in each one that they take an input and an output bufferlist.
+ */
+#include "include/types.h"
+
+#include <algorithm>
+#include <errno.h>
+#include <sstream>
+
+#include "include/uuid.h"
+#include "common/bit_vector.hpp"
+#include "common/errno.h"
+#include "objclass/objclass.h"
+#include "osd/osd_types.h"
+#include "include/rbd_types.h"
+#include "include/rbd/object_map_types.h"
+
+#include "cls/rbd/cls_rbd.h"
+#include "cls/rbd/cls_rbd_types.h"
+
+
+/*
+ * Object keys:
+ *
+ * <partial list>
+ *
+ * stripe_unit: size in bytes of the stripe unit.  if not present,
+ *   the stripe unit is assumed to match the object size (1 << order).
+ *
+ * stripe_count: number of objects to stripe over before looping back.
+ *   if not present or 1, striping is disabled.  this is the default.
+ *
+ */
+
+CLS_VER(2,0)
+CLS_NAME(rbd)
+
+#define RBD_MAX_KEYS_READ 64
+#define RBD_SNAP_KEY_PREFIX "snapshot_"
+#define RBD_SNAP_CHILDREN_KEY_PREFIX "snap_children_"
+#define RBD_DIR_ID_KEY_PREFIX "id_"
+#define RBD_DIR_NAME_KEY_PREFIX "name_"
+#define RBD_METADATA_KEY_PREFIX "metadata_"
+
+namespace {
+
+uint64_t get_encode_features(cls_method_context_t hctx) {
+  uint64_t features = 0;
+  int8_t require_osd_release = cls_get_required_osd_release(hctx);
+  if (require_osd_release >= CEPH_RELEASE_NAUTILUS) {
+    features |= CEPH_FEATURE_SERVER_NAUTILUS;
+  }
+  return features;
+}
+
+bool calc_sparse_extent(const bufferptr &bp, size_t sparse_size,
+                        uint64_t length, size_t *write_offset,
+                        size_t *write_length, size_t *offset) {
+  size_t extent_size;
+  if (*offset + sparse_size > length) {
+    extent_size = length - *offset;
+  } else {
+    extent_size = sparse_size;
+  }
+
+  bufferptr extent(bp, *offset, extent_size);
+  *offset += extent_size;
+
+  bool extent_is_zero = extent.is_zero();
+  if (!extent_is_zero) {
+    *write_length += extent_size;
+  }
+  if (extent_is_zero && *write_length == 0) {
+    *write_offset += extent_size;
+  }
+
+  if ((extent_is_zero || *offset == length) && *write_length != 0) {
+    return true;
+  }
+  return false;
+}
+
+} // anonymous namespace
+
+static int snap_read_header(cls_method_context_t hctx, bufferlist& bl)
+{
+  unsigned snap_count = 0;
+  uint64_t snap_names_len = 0;
+  struct rbd_obj_header_ondisk *header;
+
+  CLS_LOG(20, "snapshots_list");
+
+  while (1) {
+    int len = sizeof(*header) +
+      snap_count * sizeof(struct rbd_obj_snap_ondisk) +
+      snap_names_len;
+
+    int rc = cls_cxx_read(hctx, 0, len, &bl);
+    if (rc < 0)
+      return rc;
+
+    if (bl.length() < sizeof(*header))
+      return -EINVAL;
+
+    header = (struct rbd_obj_header_ondisk *)bl.c_str();
+    ceph_assert(header);
+
+    if ((snap_count != header->snap_count) ||
+        (snap_names_len != header->snap_names_len)) {
+      snap_count = header->snap_count;
+      snap_names_len = header->snap_names_len;
+      bl.clear();
+      continue;
+    }
+    break;
+  }
+
+  return 0;
+}
+
+static void key_from_snap_id(snapid_t snap_id, string *out)
+{
+  ostringstream oss;
+  oss << RBD_SNAP_KEY_PREFIX
+      << std::setw(16) << std::setfill('0') << std::hex << snap_id;
+  *out = oss.str();
+}
+
+static snapid_t snap_id_from_key(const string &key) {
+  istringstream iss(key);
+  uint64_t id;
+  iss.ignore(strlen(RBD_SNAP_KEY_PREFIX)) >> std::hex >> id;
+  return id;
+}
+
+template<typename T>
+static int read_key(cls_method_context_t hctx, const string &key, T *out)
+{
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, key, &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading omap key %s: %s", key.c_str(), cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  try {
+    auto it = bl.cbegin();
+    decode(*out, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("error decoding %s", key.c_str());
+    return -EIO;
+  }
+
+  return 0;
+}
+
+template <typename T>
+static int write_key(cls_method_context_t hctx, const string &key, const T &t) {
+  bufferlist bl;
+  encode(t, bl);
+
+  int r = cls_cxx_map_set_val(hctx, key, &bl);
+  if (r < 0) {
+    CLS_ERR("failed to set omap key: %s", key.c_str());
+    return r;
+  }
+  return 0;
+}
+
+template <typename T>
+static int write_key(cls_method_context_t hctx, const string &key, const T &t,
+                     uint64_t features) {
+  bufferlist bl;
+  encode(t, bl, features);
+
+  int r = cls_cxx_map_set_val(hctx, key, &bl);
+  if (r < 0) {
+    CLS_ERR("failed to set omap key: %s", key.c_str());
+    return r;
+  }
+  return 0;
+}
+
+static int remove_key(cls_method_context_t hctx, const string &key) {
+  int r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0 && r != -ENOENT) {
+      CLS_ERR("failed to remove key: %s", key.c_str());
+      return r;
+  }
+  return 0;
+}
+
+static bool is_valid_id(const string &id) {
+  if (!id.size())
+    return false;
+  for (size_t i = 0; i < id.size(); ++i) {
+    if (!isalnum(id[i])) {
+      return false;
+    }
+  }
+  return true;
+}
+
+/**
+ * verify that the header object exists
+ *
+ * @return 0 if the object exists, -ENOENT if it does not, or other error
+ */
+static int check_exists(cls_method_context_t hctx)
+{
+  uint64_t size;
+  time_t mtime;
+  return cls_cxx_stat(hctx, &size, &mtime);
+}
+
+namespace image {
+
+/**
+ * check that given feature(s) are set
+ *
+ * @param hctx context
+ * @param need features needed
+ * @return 0 if features are set, negative error (like ENOEXEC) otherwise
+ */
+int require_feature(cls_method_context_t hctx, uint64_t need)
+{
+  uint64_t features;
+  int r = read_key(hctx, "features", &features);
+  if (r == -ENOENT)   // this implies it's an old-style image with no features
+    return -ENOEXEC;
+  if (r < 0)
+    return r;
+  if ((features & need) != need) {
+    CLS_LOG(10, "require_feature missing feature %llx, have %llx",
+            (unsigned long long)need, (unsigned long long)features);
+    return -ENOEXEC;
+  }
+  return 0;
+}
+
+std::string snap_children_key_from_snap_id(snapid_t snap_id)
+{
+  ostringstream oss;
+  oss << RBD_SNAP_CHILDREN_KEY_PREFIX
+      << std::setw(16) << std::setfill('0') << std::hex << snap_id;
+  return oss.str();
+}
+
+int set_op_features(cls_method_context_t hctx, uint64_t op_features,
+                    uint64_t mask) {
+  uint64_t orig_features;
+  int r = read_key(hctx, "features", &orig_features);
+  if (r < 0) {
+    CLS_ERR("failed to read features off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  uint64_t orig_op_features = 0;
+  r = read_key(hctx, "op_features", &orig_op_features);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("Could not read op features off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  op_features = (orig_op_features & ~mask) | (op_features & mask);
+  CLS_LOG(10, "op_features=%" PRIu64 " orig_op_features=%" PRIu64,
+          op_features, orig_op_features);
+  if (op_features == orig_op_features) {
+    return 0;
+  }
+
+  uint64_t features = orig_features;
+  if (op_features == 0ULL) {
+    features &= ~RBD_FEATURE_OPERATIONS;
+
+    r = cls_cxx_map_remove_key(hctx, "op_features");
+    if (r == -ENOENT) {
+      r = 0;
+    }
+  } else {
+    features |= RBD_FEATURE_OPERATIONS;
+
+    bufferlist bl;
+    encode(op_features, bl);
+    r = cls_cxx_map_set_val(hctx, "op_features", &bl);
+  }
+
+  if (r < 0) {
+    CLS_ERR("error updating op features: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (features != orig_features) {
+    bufferlist bl;
+    encode(features, bl);
+    r = cls_cxx_map_set_val(hctx, "features", &bl);
+    if (r < 0) {
+      CLS_ERR("error updating features: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  }
+
+  return 0;
+}
+
+int set_migration(cls_method_context_t hctx,
+                  const cls::rbd::MigrationSpec &migration_spec, bool init) {
+  if (init) {
+    bufferlist bl;
+    int r = cls_cxx_map_get_val(hctx, "migration", &bl);
+    if (r != -ENOENT) {
+      if (r == 0) {
+        CLS_LOG(10, "migration already set");
+        return -EEXIST;
+      }
+      CLS_ERR("failed to read migration off disk: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+
+    uint64_t features = 0;
+    r = read_key(hctx, "features", &features);
+    if (r == -ENOENT) {
+      CLS_LOG(20, "no features, assuming v1 format");
+      bufferlist header;
+      r = cls_cxx_read(hctx, 0, sizeof(RBD_HEADER_TEXT), &header);
+      if (r < 0) {
+        CLS_ERR("failed to read v1 header: %s", cpp_strerror(r).c_str());
+        return r;
+      }
+      if (header.length() != sizeof(RBD_HEADER_TEXT)) {
+        CLS_ERR("unrecognized v1 header format");
+        return -ENXIO;
+      }
+      if (memcmp(RBD_HEADER_TEXT, header.c_str(), header.length()) != 0) {
+        if (memcmp(RBD_MIGRATE_HEADER_TEXT, header.c_str(),
+                   header.length()) == 0) {
+          CLS_LOG(10, "migration already set");
+          return -EEXIST;
+        } else {
+          CLS_ERR("unrecognized v1 header format");
+          return -ENXIO;
+        }
+      }
+      if (migration_spec.header_type != cls::rbd::MIGRATION_HEADER_TYPE_SRC) {
+        CLS_LOG(10, "v1 format image can only be migration source");
+        return -EINVAL;
+      }
+
+      header.clear();
+      header.append(RBD_MIGRATE_HEADER_TEXT);
+      r = cls_cxx_write(hctx, 0, header.length(), &header);
+      if (r < 0) {
+        CLS_ERR("error updating v1 header: %s", cpp_strerror(r).c_str());
+        return r;
+      }
+    } else if (r < 0) {
+      CLS_ERR("failed to read features off disk: %s", cpp_strerror(r).c_str());
+      return r;
+    } else if ((features & RBD_FEATURE_MIGRATING) != 0ULL) {
+      if (migration_spec.header_type != cls::rbd::MIGRATION_HEADER_TYPE_DST) {
+        CLS_LOG(10, "migrating feature already set");
+        return -EEXIST;
+      }
+    } else {
+      features |= RBD_FEATURE_MIGRATING;
+      bl.clear();
+      encode(features, bl);
+      r = cls_cxx_map_set_val(hctx, "features", &bl);
+      if (r < 0) {
+        CLS_ERR("error updating features: %s", cpp_strerror(r).c_str());
+        return r;
+      }
+    }
+  }
+
+  bufferlist bl;
+  encode(migration_spec, bl);
+  int r = cls_cxx_map_set_val(hctx, "migration", &bl);
+  if (r < 0) {
+    CLS_ERR("error setting migration: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+int read_migration(cls_method_context_t hctx,
+                   cls::rbd::MigrationSpec *migration_spec) {
+  uint64_t features = 0;
+  int r = read_key(hctx, "features", &features);
+  if (r == -ENOENT) {
+    CLS_LOG(20, "no features, assuming v1 format");
+    bufferlist header;
+    r = cls_cxx_read(hctx, 0, sizeof(RBD_HEADER_TEXT), &header);
+    if (r < 0) {
+      CLS_ERR("failed to read v1 header: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+    if (header.length() != sizeof(RBD_HEADER_TEXT)) {
+      CLS_ERR("unrecognized v1 header format");
+      return -ENXIO;
+    }
+    if (memcmp(RBD_MIGRATE_HEADER_TEXT, header.c_str(), header.length()) != 0) {
+      if (memcmp(RBD_HEADER_TEXT, header.c_str(), header.length()) == 0) {
+        CLS_LOG(10, "migration feature not set");
+        return -EINVAL;
+      } else {
+        CLS_ERR("unrecognized v1 header format");
+        return -ENXIO;
+      }
+    }
+    if (migration_spec->header_type != cls::rbd::MIGRATION_HEADER_TYPE_SRC) {
+      CLS_LOG(10, "v1 format image can only be migration source");
+      return -EINVAL;
+    }
+  } else if (r < 0) {
+    CLS_ERR("failed to read features off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  } else if ((features & RBD_FEATURE_MIGRATING) == 0ULL) {
+    CLS_LOG(10, "migration feature not set");
+    return -EINVAL;
+  }
+
+  r = read_key(hctx, "migration", migration_spec);
+  if (r < 0) {
+    CLS_ERR("failed to read migration off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+int remove_migration(cls_method_context_t hctx) {
+  int r = remove_key(hctx, "migration");
+  if (r < 0) {
+    return r;
+  }
+
+  uint64_t features = 0;
+  r = read_key(hctx, "features", &features);
+  if (r == -ENOENT) {
+    CLS_LOG(20, "no features, assuming v1 format");
+    bufferlist header;
+    r = cls_cxx_read(hctx, 0, sizeof(RBD_MIGRATE_HEADER_TEXT), &header);
+    if (header.length() != sizeof(RBD_MIGRATE_HEADER_TEXT)) {
+      CLS_ERR("unrecognized v1 header format");
+      return -ENXIO;
+    }
+    if (memcmp(RBD_MIGRATE_HEADER_TEXT, header.c_str(), header.length()) != 0) {
+      if (memcmp(RBD_HEADER_TEXT, header.c_str(), header.length()) == 0) {
+        CLS_LOG(10, "migration feature not set");
+        return -EINVAL;
+      } else {
+        CLS_ERR("unrecognized v1 header format");
+        return -ENXIO;
+      }
+    }
+    header.clear();
+    header.append(RBD_HEADER_TEXT);
+    r = cls_cxx_write(hctx, 0, header.length(), &header);
+    if (r < 0) {
+      CLS_ERR("error updating v1 header: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else if (r < 0) {
+    CLS_ERR("failed to read features off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  } else if ((features & RBD_FEATURE_MIGRATING) == 0ULL) {
+    CLS_LOG(10, "migrating feature not set");
+  } else {
+    features &= ~RBD_FEATURE_MIGRATING;
+    bufferlist bl;
+    encode(features, bl);
+    r = cls_cxx_map_set_val(hctx, "features", &bl);
+    if (r < 0) {
+      CLS_ERR("error updating features: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  }
+
+  return 0;
+}
+
+namespace snapshot {
+
+template<typename L>
+int iterate(cls_method_context_t hctx, L& lambda) {
+  int max_read = RBD_MAX_KEYS_READ;
+  string last_read = RBD_SNAP_KEY_PREFIX;
+  bool more = false;
+  do {
+    map<string, bufferlist> vals;
+    int r = cls_cxx_map_get_vals(hctx, last_read, RBD_SNAP_KEY_PREFIX,
+			         max_read, &vals, &more);
+    if (r < 0) {
+      return r;
+    }
+
+    cls_rbd_snap snap_meta;
+    for (auto& val : vals) {
+      auto iter = val.second.cbegin();
+      try {
+	decode(snap_meta, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("error decoding snapshot metadata for snap : %s",
+	        val.first.c_str());
+	return -EIO;
+      }
+
+      r = lambda(snap_meta);
+      if (r < 0) {
+        return r;
+      }
+    }
+
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  } while (more);
+
+  return 0;
+}
+
+int write(cls_method_context_t hctx, const std::string& snap_key,
+          cls_rbd_snap&& snap) {
+  int r;
+  uint64_t encode_features = get_encode_features(hctx);
+  if (snap.migrate_parent_format(encode_features)) {
+    // ensure the normalized parent link exists before removing it from the
+    // snapshot record
+    cls_rbd_parent on_disk_parent;
+    r = read_key(hctx, "parent", &on_disk_parent);
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    }
+
+    if (!on_disk_parent.exists()) {
+      on_disk_parent = snap.parent;
+      on_disk_parent.head_overlap = std::nullopt;
+
+      r = write_key(hctx, "parent", on_disk_parent, encode_features);
+      if (r < 0) {
+        return r;
+      }
+    }
+
+    // only store the parent overlap in the snapshot
+    snap.parent_overlap = snap.parent.head_overlap;
+    snap.parent = {};
+  }
+
+  r = write_key(hctx, snap_key, snap, encode_features);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+} // namespace snapshot
+
+namespace parent {
+
+int attach(cls_method_context_t hctx, cls_rbd_parent parent,
+           bool reattach) {
+  int r = check_exists(hctx);
+  if (r < 0) {
+    CLS_LOG(20, "cls_rbd::image::parent::attach: child doesn't exist");
+    return r;
+  }
+
+  r = image::require_feature(hctx, RBD_FEATURE_LAYERING);
+  if (r < 0) {
+    CLS_LOG(20, "cls_rbd::image::parent::attach: child does not support "
+                "layering");
+    return r;
+  }
+
+  CLS_LOG(20, "cls_rbd::image::parent::attach: pool=%" PRIi64 ", ns=%s, id=%s, "
+              "snapid=%" PRIu64 ", size=%" PRIu64,
+          parent.pool_id, parent.pool_namespace.c_str(),
+          parent.image_id.c_str(), parent.snap_id.val,
+          parent.head_overlap.value_or(0ULL));
+  if (!parent.exists() || parent.head_overlap.value_or(0ULL) == 0ULL) {
+    return -EINVAL;
+  }
+
+  // make sure there isn't already a parent
+  cls_rbd_parent on_disk_parent;
+  r = read_key(hctx, "parent", &on_disk_parent);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  auto on_disk_parent_without_overlap{on_disk_parent};
+  on_disk_parent_without_overlap.head_overlap = parent.head_overlap;
+
+  if (r == 0 &&
+      (on_disk_parent.head_overlap ||
+       on_disk_parent_without_overlap != parent) &&
+      !reattach) {
+    CLS_LOG(20, "cls_rbd::parent::attach: existing legacy parent "
+                "pool=%" PRIi64 ", ns=%s, id=%s, snapid=%" PRIu64 ", "
+                "overlap=%" PRIu64,
+            on_disk_parent.pool_id, on_disk_parent.pool_namespace.c_str(),
+            on_disk_parent.image_id.c_str(), on_disk_parent.snap_id.val,
+            on_disk_parent.head_overlap.value_or(0ULL));
+    return -EEXIST;
+  }
+
+  // our overlap is the min of our size and the parent's size.
+  uint64_t our_size;
+  r = read_key(hctx, "size", &our_size);
+  if (r < 0) {
+    return r;
+  }
+
+  parent.head_overlap = std::min(*parent.head_overlap, our_size);
+
+  r = write_key(hctx, "parent", parent, get_encode_features(hctx));
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+int detach(cls_method_context_t hctx, bool legacy_api) {
+  int r = check_exists(hctx);
+  if (r < 0) {
+    CLS_LOG(20, "cls_rbd::parent::detach: child doesn't exist");
+    return r;
+  }
+
+  uint64_t features;
+  r = read_key(hctx, "features", &features);
+  if (r == -ENOENT || ((features & RBD_FEATURE_LAYERING) == 0)) {
+    CLS_LOG(20, "cls_rbd::image::parent::detach: child does not support "
+                "layering");
+    return -ENOEXEC;
+  } else if (r < 0) {
+    return r;
+  }
+
+  cls_rbd_parent on_disk_parent;
+  r = read_key(hctx, "parent", &on_disk_parent);
+  if (r < 0) {
+    return r;
+  } else if (legacy_api && !on_disk_parent.pool_namespace.empty()) {
+    return -EXDEV;
+  } else if (!on_disk_parent.head_overlap) {
+    return -ENOENT;
+  }
+
+  auto detach_lambda = [hctx, features](const cls_rbd_snap& snap_meta) {
+    if (snap_meta.parent.pool_id != -1 || snap_meta.parent_overlap) {
+      if ((features & RBD_FEATURE_DEEP_FLATTEN) != 0ULL) {
+        // remove parent reference from snapshot
+        cls_rbd_snap snap_meta_copy = snap_meta;
+        snap_meta_copy.parent = {};
+        snap_meta_copy.parent_overlap = std::nullopt;
+
+        std::string snap_key;
+        key_from_snap_id(snap_meta_copy.id, &snap_key);
+        int r = snapshot::write(hctx, snap_key, std::move(snap_meta_copy));
+        if (r < 0) {
+          return r;
+        }
+      } else {
+        return -EEXIST;
+      }
+    }
+    return 0;
+  };
+
+  r = snapshot::iterate(hctx, detach_lambda);
+  bool has_child_snaps = (r == -EEXIST);
+  if (r < 0 && r != -EEXIST) {
+    return r;
+  }
+
+  int8_t require_osd_release = cls_get_required_osd_release(hctx);
+  if (has_child_snaps && require_osd_release >= CEPH_RELEASE_NAUTILUS) {
+    // remove overlap from HEAD revision but keep spec for snapshots
+    on_disk_parent.head_overlap = std::nullopt;
+    r = write_key(hctx, "parent", on_disk_parent, get_encode_features(hctx));
+    if (r < 0) {
+      return r;
+    }
+  } else {
+    r = remove_key(hctx, "parent");
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    }
+  }
+
+  if (!has_child_snaps) {
+    // disable clone child op feature if no longer associated
+    r = set_op_features(hctx, 0, RBD_OPERATION_FEATURE_CLONE_CHILD);
+    if (r < 0) {
+      return r;
+    }
+  }
+  return 0;
+}
+
+} // namespace parent
+} // namespace image
+
+/**
+ * Initialize the header with basic metadata.
+ * Extra features may initialize more fields in the future.
+ * Everything is stored as key/value pairs as omaps in the header object.
+ *
+ * If features the OSD does not understand are requested, -ENOSYS is
+ * returned.
+ *
+ * Input:
+ * @param size number of bytes in the image (uint64_t)
+ * @param order bits to shift to determine the size of data objects (uint8_t)
+ * @param features what optional things this image will use (uint64_t)
+ * @param object_prefix a prefix for all the data objects
+ * @param data_pool_id pool id where data objects is stored (int64_t)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int create(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string object_prefix;
+  uint64_t features, size;
+  uint8_t order;
+  int64_t data_pool_id = -1;
+
+  try {
+    auto iter = in->cbegin();
+    decode(size, iter);
+    decode(order, iter);
+    decode(features, iter);
+    decode(object_prefix, iter);
+    if (!iter.end()) {
+      decode(data_pool_id, iter);
+    }
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "create object_prefix=%s size=%llu order=%u features=%llu",
+	  object_prefix.c_str(), (unsigned long long)size, order,
+	  (unsigned long long)features);
+
+  if (features & ~RBD_FEATURES_ALL) {
+    return -ENOSYS;
+  }
+
+  if (!object_prefix.size()) {
+    return -EINVAL;
+  }
+
+  bufferlist stored_prefixbl;
+  int r = cls_cxx_map_get_val(hctx, "object_prefix", &stored_prefixbl);
+  if (r != -ENOENT) {
+    CLS_ERR("reading object_prefix returned %d", r);
+    return -EEXIST;
+  }
+
+  bufferlist sizebl;
+  bufferlist orderbl;
+  bufferlist featuresbl;
+  bufferlist object_prefixbl;
+  bufferlist snap_seqbl;
+  bufferlist timestampbl;
+  uint64_t snap_seq = 0;
+  utime_t timestamp = ceph_clock_now();
+  encode(size, sizebl);
+  encode(order, orderbl);
+  encode(features, featuresbl);
+  encode(object_prefix, object_prefixbl);
+  encode(snap_seq, snap_seqbl);
+  encode(timestamp, timestampbl);
+
+  map<string, bufferlist> omap_vals;
+  omap_vals["size"] = sizebl;
+  omap_vals["order"] = orderbl;
+  omap_vals["features"] = featuresbl;
+  omap_vals["object_prefix"] = object_prefixbl;
+  omap_vals["snap_seq"] = snap_seqbl;
+  omap_vals["create_timestamp"] = timestampbl;
+  omap_vals["access_timestamp"] = timestampbl;
+  omap_vals["modify_timestamp"] = timestampbl;
+
+  if ((features & RBD_FEATURE_OPERATIONS) != 0ULL) {
+    CLS_ERR("Attempting to set internal feature: operations");
+    return -EINVAL;
+  }
+
+  if (features & RBD_FEATURE_DATA_POOL) {
+    if (data_pool_id == -1) {
+      CLS_ERR("data pool not provided with feature enabled");
+      return -EINVAL;
+    }
+
+    bufferlist data_pool_id_bl;
+    encode(data_pool_id, data_pool_id_bl);
+    omap_vals["data_pool_id"] = data_pool_id_bl;
+  } else if (data_pool_id != -1) {
+    CLS_ERR("data pool provided with feature disabled");
+    return -EINVAL;
+  }
+
+  r = cls_cxx_map_set_vals(hctx, &omap_vals);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap_id which snapshot to query, or CEPH_NOSNAP (uint64_t) (deprecated)
+ * @param read_only true if the image will be used read-only (bool)
+ *
+ * Output:
+ * @param features list of enabled features for the given snapshot (uint64_t)
+ * @param incompatible incompatible feature bits
+ * @returns 0 on success, negative error code on failure
+ */
+int get_features(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bool read_only = false;
+
+  auto iter = in->cbegin();
+  try {
+    uint64_t snap_id;
+    decode(snap_id, iter);
+    if (!iter.end()) {
+      decode(read_only, iter);
+    }
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "get_features read_only=%d", read_only);
+
+  uint64_t features;
+  int r = read_key(hctx, "features", &features);
+  if (r < 0) {
+    CLS_ERR("failed to read features off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  uint64_t incompatible = (read_only ? features & RBD_FEATURES_INCOMPATIBLE :
+				       features & RBD_FEATURES_RW_INCOMPATIBLE);
+  encode(features, *out);
+  encode(incompatible, *out);
+  return 0;
+}
+
+/**
+ * set the image features
+ *
+ * Input:
+ * @param features image features
+ * @param mask image feature mask
+ *
+ * Output:
+ * none
+ *
+ * @returns 0 on success, negative error code upon failure
+ */
+int set_features(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t features;
+  uint64_t mask;
+  auto iter = in->cbegin();
+  try {
+    decode(features, iter);
+    decode(mask, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  // check that features exists to make sure this is a header object
+  // that was created correctly
+  uint64_t orig_features = 0;
+  int r = read_key(hctx, "features", &orig_features);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("Could not read image's features off disk: %s",
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if ((mask & RBD_FEATURES_INTERNAL) != 0ULL) {
+    CLS_ERR("Attempting to set internal feature: %" PRIu64,
+            static_cast<uint64_t>(mask & RBD_FEATURES_INTERNAL));
+    return -EINVAL;
+  }
+
+  // newer clients might attempt to mask off features we don't support
+  mask &= RBD_FEATURES_ALL;
+
+  uint64_t enabled_features = features & mask;
+  if ((enabled_features & RBD_FEATURES_MUTABLE) != enabled_features) {
+    CLS_ERR("Attempting to enable immutable feature: %" PRIu64,
+            static_cast<uint64_t>(enabled_features & ~RBD_FEATURES_MUTABLE));
+    return -EINVAL;
+  }
+
+  uint64_t disabled_features = ~features & mask;
+  uint64_t disable_mask = (RBD_FEATURES_MUTABLE | RBD_FEATURES_DISABLE_ONLY);
+  if ((disabled_features & disable_mask) != disabled_features) {
+       CLS_ERR("Attempting to disable immutable feature: %" PRIu64,
+               enabled_features & ~disable_mask);
+       return -EINVAL;
+  }
+
+  features = (orig_features & ~mask) | (features & mask);
+  CLS_LOG(10, "set_features features=%" PRIu64 " orig_features=%" PRIu64,
+          features, orig_features);
+
+  bufferlist bl;
+  encode(features, bl);
+  r = cls_cxx_map_set_val(hctx, "features", &bl);
+  if (r < 0) {
+    CLS_ERR("error updating features: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap_id which snapshot to query, or CEPH_NOSNAP (uint64_t)
+ *
+ * Output:
+ * @param order bits to shift to get the size of data objects (uint8_t)
+ * @param size size of the image in bytes for the given snapshot (uint64_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int get_size(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id, size;
+  uint8_t order;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "get_size snap_id=%llu", (unsigned long long)snap_id);
+
+  int r = read_key(hctx, "order", &order);
+  if (r < 0) {
+    CLS_ERR("failed to read the order off of disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (snap_id == CEPH_NOSNAP) {
+    r = read_key(hctx, "size", &size);
+    if (r < 0) {
+      CLS_ERR("failed to read the image's size off of disk: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    cls_rbd_snap snap;
+    string snapshot_key;
+    key_from_snap_id(snap_id, &snapshot_key);
+    int r = read_key(hctx, snapshot_key, &snap);
+    if (r < 0)
+      return r;
+
+    size = snap.image_size;
+  }
+
+  encode(order, *out);
+  encode(size, *out);
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param size new capacity of the image in bytes (uint64_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int set_size(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t size;
+
+  auto iter = in->cbegin();
+  try {
+    decode(size, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  // check that size exists to make sure this is a header object
+  // that was created correctly
+  uint64_t orig_size;
+  int r = read_key(hctx, "size", &orig_size);
+  if (r < 0) {
+    CLS_ERR("Could not read image's size off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  CLS_LOG(20, "set_size size=%llu orig_size=%llu", (unsigned long long)size,
+          (unsigned long long)orig_size);
+
+  bufferlist sizebl;
+  encode(size, sizebl);
+  r = cls_cxx_map_set_val(hctx, "size", &sizebl);
+  if (r < 0) {
+    CLS_ERR("error writing snapshot metadata: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  // if we are shrinking, and have a parent, shrink our overlap with
+  // the parent, too.
+  if (size < orig_size) {
+    cls_rbd_parent parent;
+    r = read_key(hctx, "parent", &parent);
+    if (r == -ENOENT)
+      r = 0;
+    if (r < 0)
+      return r;
+    if (parent.exists() && parent.head_overlap.value_or(0ULL) > size) {
+      parent.head_overlap = size;
+      r = write_key(hctx, "parent", parent, get_encode_features(hctx));
+      if (r < 0) {
+	return r;
+      }
+    }
+  }
+
+  return 0;
+}
+
+/**
+ * get the current protection status of the specified snapshot
+ *
+ * Input:
+ * @param snap_id (uint64_t) which snapshot to get the status of
+ *
+ * Output:
+ * @param status (uint8_t) one of:
+ * RBD_PROTECTION_STATUS_{PROTECTED, UNPROTECTED, UNPROTECTING}
+ *
+ * @returns 0 on success, negative error code on failure
+ * @returns -EINVAL if snapid is CEPH_NOSNAP
+ */
+int get_protection_status(cls_method_context_t hctx, bufferlist *in,
+			  bufferlist *out)
+{
+  snapid_t snap_id;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "get_protection_status: invalid decode");
+    return -EINVAL;
+  }
+
+  int r = check_exists(hctx);
+  if (r < 0)
+    return r;
+
+  CLS_LOG(20, "get_protection_status snap_id=%llu",
+         (unsigned long long)snap_id.val);
+
+  if (snap_id == CEPH_NOSNAP)
+    return -EINVAL;
+
+  cls_rbd_snap snap;
+  string snapshot_key;
+  key_from_snap_id(snap_id.val, &snapshot_key);
+  r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    CLS_ERR("could not read key for snapshot id %" PRIu64, snap_id.val);
+    return r;
+  }
+
+  if (snap.protection_status >= RBD_PROTECTION_STATUS_LAST) {
+    CLS_ERR("invalid protection status for snap id %llu: %u",
+	    (unsigned long long)snap_id.val, snap.protection_status);
+    return -EIO;
+  }
+
+  encode(snap.protection_status, *out);
+  return 0;
+}
+
+/**
+ * set the proctection status of a snapshot
+ *
+ * Input:
+ * @param snapid (uint64_t) which snapshot to set the status of
+ * @param status (uint8_t) one of:
+ * RBD_PROTECTION_STATUS_{PROTECTED, UNPROTECTED, UNPROTECTING}
+ *
+ * @returns 0 on success, negative error code on failure
+ * @returns -EINVAL if snapid is CEPH_NOSNAP
+ */
+int set_protection_status(cls_method_context_t hctx, bufferlist *in,
+			  bufferlist *out)
+{
+  snapid_t snap_id;
+  uint8_t status;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+    decode(status, iter);
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "set_protection_status: invalid decode");
+    return -EINVAL;
+  }
+
+  int r = check_exists(hctx);
+  if (r < 0)
+    return r;
+
+  r = image::require_feature(hctx, RBD_FEATURE_LAYERING);
+  if (r < 0) {
+    CLS_LOG(20, "image does not support layering");
+    return r;
+  }
+
+  CLS_LOG(20, "set_protection_status snapid=%llu status=%u",
+	  (unsigned long long)snap_id.val, status);
+
+  if (snap_id == CEPH_NOSNAP)
+    return -EINVAL;
+
+  if (status >= RBD_PROTECTION_STATUS_LAST) {
+    CLS_LOG(10, "invalid protection status for snap id %llu: %u",
+	    (unsigned long long)snap_id.val, status);
+    return -EINVAL;
+  }
+
+  cls_rbd_snap snap;
+  string snapshot_key;
+  key_from_snap_id(snap_id.val, &snapshot_key);
+  r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    CLS_ERR("could not read key for snapshot id %" PRIu64, snap_id.val);
+    return r;
+  }
+
+  snap.protection_status = status;
+  r = image::snapshot::write(hctx, snapshot_key, std::move(snap));
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * get striping parameters
+ *
+ * Input:
+ * none
+ *
+ * Output:
+ * @param stripe unit (bytes)
+ * @param stripe count (num objects)
+ *
+ * @returns 0 on success
+ */
+int get_stripe_unit_count(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r = check_exists(hctx);
+  if (r < 0)
+    return r;
+
+  CLS_LOG(20, "get_stripe_unit_count");
+
+  r = image::require_feature(hctx, RBD_FEATURE_STRIPINGV2);
+  if (r < 0)
+    return r;
+
+  uint64_t stripe_unit = 0, stripe_count = 0;
+  r = read_key(hctx, "stripe_unit", &stripe_unit);
+  if (r == -ENOENT) {
+    // default to object size
+    uint8_t order;
+    r = read_key(hctx, "order", &order);
+    if (r < 0) {
+      CLS_ERR("failed to read the order off of disk: %s", cpp_strerror(r).c_str());
+      return -EIO;
+    }
+    stripe_unit = 1ull << order;
+  }
+  if (r < 0)
+    return r;
+  r = read_key(hctx, "stripe_count", &stripe_count);
+  if (r == -ENOENT) {
+    // default to 1
+    stripe_count = 1;
+    r = 0;
+  }
+  if (r < 0)
+    return r;
+
+  encode(stripe_unit, *out);
+  encode(stripe_count, *out);
+  return 0;
+}
+
+/**
+ * set striping parameters
+ *
+ * Input:
+ * @param stripe unit (bytes)
+ * @param stripe count (num objects)
+ *
+ * @returns 0 on success
+ */
+int set_stripe_unit_count(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t stripe_unit, stripe_count;
+
+  auto iter = in->cbegin();
+  try {
+    decode(stripe_unit, iter);
+    decode(stripe_count, iter);
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "set_stripe_unit_count: invalid decode");
+    return -EINVAL;
+  }
+
+  if (!stripe_count || !stripe_unit)
+    return -EINVAL;
+
+  int r = check_exists(hctx);
+  if (r < 0)
+    return r;
+
+  CLS_LOG(20, "set_stripe_unit_count");
+
+  r = image::require_feature(hctx, RBD_FEATURE_STRIPINGV2);
+  if (r < 0)
+    return r;
+
+  uint8_t order;
+  r = read_key(hctx, "order", &order);
+  if (r < 0) {
+    CLS_ERR("failed to read the order off of disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  if ((1ull << order) % stripe_unit || stripe_unit > (1ull << order)) {
+    CLS_ERR("stripe unit %llu is not a factor of the object size %llu",
+            (unsigned long long)stripe_unit, 1ull << order);
+    return -EINVAL;
+  }
+
+  bufferlist bl, bl2;
+  encode(stripe_unit, bl);
+  r = cls_cxx_map_set_val(hctx, "stripe_unit", &bl);
+  if (r < 0) {
+    CLS_ERR("error writing stripe_unit metadata: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  encode(stripe_count, bl2);
+  r = cls_cxx_map_set_val(hctx, "stripe_count", &bl2);
+  if (r < 0) {
+    CLS_ERR("error writing stripe_count metadata: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+int get_create_timestamp(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_create_timestamp");
+
+  utime_t timestamp;
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, "create_timestamp", &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading create_timestamp: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    try {
+      auto it = bl.cbegin();
+      decode(timestamp, it);
+    } catch (const buffer::error &err) {
+      CLS_ERR("could not decode create_timestamp");
+      return -EIO;
+    }
+  }
+
+  encode(timestamp, *out);
+  return 0;
+}
+
+/**
+ * get the image access timestamp
+ *
+ * Input:
+ * @param none
+ *
+ * Output:
+ * @param timestamp the image access timestamp
+ *
+ * @returns 0 on success, negative error code upon failure
+ */
+int get_access_timestamp(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_access_timestamp");
+
+  utime_t timestamp;
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, "access_timestamp", &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading access_timestamp: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    try {
+      auto it = bl.cbegin();
+      decode(timestamp, it);
+    } catch (const buffer::error &err) {
+      CLS_ERR("could not decode access_timestamp");
+      return -EIO;
+    }
+  }
+
+  encode(timestamp, *out);
+  return 0;
+}
+
+/**
+ * get the image modify timestamp
+ *
+ * Input:
+ * @param none
+ *
+ * Output:
+ * @param timestamp the image modify timestamp
+ *
+ * @returns 0 on success, negative error code upon failure
+ */
+int get_modify_timestamp(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_modify_timestamp");
+
+  utime_t timestamp;
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, "modify_timestamp", &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading modify_timestamp: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    try {
+      auto it = bl.cbegin();
+      decode(timestamp, it);
+    } catch (const buffer::error &err) {
+      CLS_ERR("could not decode modify_timestamp");
+      return -EIO;
+    }
+  }
+
+  encode(timestamp, *out);
+  return 0;
+}
+
+
+/**
+ * get the image flags
+ *
+ * Input:
+ * @param snap_id which snapshot to query, to CEPH_NOSNAP (uint64_t)
+ *
+ * Output:
+ * @param flags image flags
+ *
+ * @returns 0 on success, negative error code upon failure
+ */
+int get_flags(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "get_flags snap_id=%llu", (unsigned long long)snap_id);
+
+  uint64_t flags = 0;
+  if (snap_id == CEPH_NOSNAP) {
+    int r = read_key(hctx, "flags", &flags);
+    if (r < 0 && r != -ENOENT) {
+      CLS_ERR("failed to read flags off disk: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    cls_rbd_snap snap;
+    string snapshot_key;
+    key_from_snap_id(snap_id, &snapshot_key);
+    int r = read_key(hctx, snapshot_key, &snap);
+    if (r < 0) {
+      return r;
+    }
+    flags = snap.flags;
+  }
+
+  encode(flags, *out);
+  return 0;
+}
+
+/**
+ * set the image flags
+ *
+ * Input:
+ * @param flags image flags
+ * @param mask image flag mask
+ * @param snap_id which snapshot to update, or CEPH_NOSNAP (uint64_t)
+ *
+ * Output:
+ * none
+ *
+ * @returns 0 on success, negative error code upon failure
+ */
+int set_flags(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t flags;
+  uint64_t mask;
+  uint64_t snap_id = CEPH_NOSNAP;
+  auto iter = in->cbegin();
+  try {
+    decode(flags, iter);
+    decode(mask, iter);
+    if (!iter.end()) {
+      decode(snap_id, iter);
+    }
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  // check that size exists to make sure this is a header object
+  // that was created correctly
+  int r;
+  uint64_t orig_flags = 0;
+  cls_rbd_snap snap_meta;
+  string snap_meta_key;
+  if (snap_id == CEPH_NOSNAP) {
+    r = read_key(hctx, "flags", &orig_flags);
+    if (r < 0 && r != -ENOENT) {
+      CLS_ERR("Could not read image's flags off disk: %s",
+              cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    key_from_snap_id(snap_id, &snap_meta_key);
+    r = read_key(hctx, snap_meta_key, &snap_meta);
+    if (r < 0) {
+      CLS_ERR("Could not read snapshot: snap_id=%" PRIu64 ": %s",
+              snap_id, cpp_strerror(r).c_str());
+      return r;
+    }
+    orig_flags = snap_meta.flags;
+  }
+
+  flags = (orig_flags & ~mask) | (flags & mask);
+  CLS_LOG(20, "set_flags snap_id=%" PRIu64 ", orig_flags=%" PRIu64 ", "
+              "new_flags=%" PRIu64 ", mask=%" PRIu64, snap_id, orig_flags,
+              flags, mask);
+
+  if (snap_id == CEPH_NOSNAP) {
+    r = write_key(hctx, "flags", flags);
+  } else {
+    snap_meta.flags = flags;
+    r = image::snapshot::write(hctx, snap_meta_key, std::move(snap_meta));
+  }
+
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Get the operation-based image features
+ *
+ * Input:
+ *
+ * Output:
+ * @param bitmask of enabled op features (uint64_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int op_features_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "op_features_get");
+
+  uint64_t op_features = 0;
+  int r = read_key(hctx, "op_features", &op_features);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("failed to read op features off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  encode(op_features, *out);
+  return 0;
+}
+
+/**
+ * Set the operation-based image features
+ *
+ * Input:
+ * @param op_features image op features
+ * @param mask image op feature mask
+ *
+ * Output:
+ * none
+ *
+ * @returns 0 on success, negative error code upon failure
+ */
+int op_features_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t op_features;
+  uint64_t mask;
+  auto iter = in->cbegin();
+  try {
+    decode(op_features, iter);
+    decode(mask, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  uint64_t unsupported_op_features = (mask & ~RBD_OPERATION_FEATURES_ALL);
+  if (unsupported_op_features != 0ULL) {
+    CLS_ERR("unsupported op features: %" PRIu64, unsupported_op_features);
+    return -EINVAL;
+  }
+
+  return image::set_op_features(hctx, op_features, mask);
+}
+
+/**
+ * get the current parent, if any
+ *
+ * Input:
+ * @param snap_id which snapshot to query, or CEPH_NOSNAP (uint64_t)
+ *
+ * Output:
+ * @param pool parent pool id (-1 if parent does not exist)
+ * @param image parent image id
+ * @param snapid parent snapid
+ * @param size portion of parent mapped under the child
+ *
+ * @returns 0 on success or parent does not exist, negative error code on failure
+ */
+int get_parent(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = check_exists(hctx);
+  if (r < 0) {
+    return r;
+  }
+
+  CLS_LOG(20, "get_parent snap_id=%" PRIu64, snap_id);
+
+  cls_rbd_parent parent;
+  r = image::require_feature(hctx, RBD_FEATURE_LAYERING);
+  if (r == 0) {
+    r = read_key(hctx, "parent", &parent);
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    } else if (!parent.pool_namespace.empty()) {
+      return -EXDEV;
+    }
+
+    if (snap_id != CEPH_NOSNAP) {
+      cls_rbd_snap snap;
+      std::string snapshot_key;
+      key_from_snap_id(snap_id, &snapshot_key);
+      r = read_key(hctx, snapshot_key, &snap);
+      if (r < 0 && r != -ENOENT) {
+	return r;
+      }
+
+      if (snap.parent.exists()) {
+        // legacy format where full parent spec is written within
+        // each snapshot record
+        parent = snap.parent;
+      } else if (snap.parent_overlap) {
+        // normalized parent reference
+        if (!parent.exists()) {
+          CLS_ERR("get_parent: snap_id=%" PRIu64 ": invalid parent spec",
+                  snap_id);
+          return -EINVAL;
+        }
+        parent.head_overlap = *snap.parent_overlap;
+      } else {
+        // snapshot doesn't have associated parent
+        parent = {};
+      }
+    }
+  }
+
+  encode(parent.pool_id, *out);
+  encode(parent.image_id, *out);
+  encode(parent.snap_id, *out);
+  encode(parent.head_overlap.value_or(0ULL), *out);
+  return 0;
+}
+
+/**
+ * set the image parent
+ *
+ * Input:
+ * @param pool parent pool
+ * @param id parent image id
+ * @param snapid parent snapid
+ * @param size parent size
+ *
+ * @returns 0 on success, or negative error code
+ */
+int set_parent(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  cls_rbd_parent parent;
+  auto iter = in->cbegin();
+  try {
+    decode(parent.pool_id, iter);
+    decode(parent.image_id, iter);
+    decode(parent.snap_id, iter);
+
+    uint64_t overlap;
+    decode(overlap, iter);
+    parent.head_overlap = overlap;
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "cls_rbd::set_parent: invalid decode");
+    return -EINVAL;
+  }
+
+  int r = image::parent::attach(hctx, parent, false);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+
+/**
+ * remove the parent pointer
+ *
+ * This can only happen on the head, not on a snapshot.  No arguments.
+ *
+ * @returns 0 on success, negative error code on failure.
+ */
+int remove_parent(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r = image::parent::detach(hctx, true);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param parent spec (cls::rbd::ParentImageSpec)
+ * @returns 0 on success, negative error code on failure
+ */
+int parent_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out) {
+  int r = check_exists(hctx);
+  if (r < 0) {
+    return r;
+  }
+
+  CLS_LOG(20, "parent_get");
+
+  cls_rbd_parent parent;
+  r = image::require_feature(hctx, RBD_FEATURE_LAYERING);
+  if (r == 0) {
+    r = read_key(hctx, "parent", &parent);
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    } else if (r == -ENOENT) {
+      // examine oldest snapshot to see if it has a denormalized parent
+      auto parent_lambda = [hctx, &parent](const cls_rbd_snap& snap_meta) {
+        if (snap_meta.parent.exists()) {
+          parent = snap_meta.parent;
+        }
+        return 0;
+      };
+
+      r = image::snapshot::iterate(hctx, parent_lambda);
+      if (r < 0) {
+        return r;
+      }
+    }
+  }
+
+  cls::rbd::ParentImageSpec parent_image_spec{
+    parent.pool_id, parent.pool_namespace, parent.image_id,
+    parent.snap_id};
+  encode(parent_image_spec, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap id (uint64_t) parent snapshot id
+ *
+ * Output:
+ * @param byte overlap of parent image (std::optional<uint64_t>)
+ * @returns 0 on success, negative error code on failure
+ */
+int parent_overlap_get(cls_method_context_t hctx, bufferlist *in,
+                       bufferlist *out) {
+  uint64_t snap_id;
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = check_exists(hctx);
+  CLS_LOG(20, "parent_overlap_get");
+
+  std::optional<uint64_t> parent_overlap = std::nullopt;
+  r = image::require_feature(hctx, RBD_FEATURE_LAYERING);
+  if (r == 0) {
+    if (snap_id == CEPH_NOSNAP) {
+      cls_rbd_parent parent;
+      r = read_key(hctx, "parent", &parent);
+      if (r < 0 && r != -ENOENT) {
+        return r;
+      } else if (r == 0) {
+        parent_overlap = parent.head_overlap;
+      }
+    } else {
+      cls_rbd_snap snap;
+      std::string snapshot_key;
+      key_from_snap_id(snap_id, &snapshot_key);
+      r = read_key(hctx, snapshot_key, &snap);
+      if (r < 0) {
+        return r;
+      }
+
+      if (snap.parent_overlap) {
+        parent_overlap = snap.parent_overlap;
+      } else if (snap.parent.exists()) {
+        // legacy format where full parent spec is written within
+        // each snapshot record
+        parent_overlap = snap.parent.head_overlap;
+      }
+    }
+  };
+
+  encode(parent_overlap, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param parent spec (cls::rbd::ParentImageSpec)
+ * @param size parent size (uint64_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int parent_attach(cls_method_context_t hctx, bufferlist *in, bufferlist *out) {
+  cls::rbd::ParentImageSpec parent_image_spec;
+  uint64_t parent_overlap;
+  bool reattach = false;
+
+  auto iter = in->cbegin();
+  try {
+    decode(parent_image_spec, iter);
+    decode(parent_overlap, iter);
+    if (!iter.end()) {
+      decode(reattach, iter);
+    }
+  } catch (const buffer::error &err) {
+    CLS_LOG(20, "cls_rbd::parent_attach: invalid decode");
+    return -EINVAL;
+  }
+
+  int r = image::parent::attach(hctx, {parent_image_spec, parent_overlap},
+                                reattach);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int parent_detach(cls_method_context_t hctx, bufferlist *in, bufferlist *out) {
+  int r = image::parent::detach(hctx, false);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+
+/**
+ * methods for dealing with rbd_children object
+ */
+
+static int decode_parent_common(bufferlist::const_iterator& it, uint64_t *pool_id,
+				string *image_id, snapid_t *snap_id)
+{
+  try {
+    decode(*pool_id, it);
+    decode(*image_id, it);
+    decode(*snap_id, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("error decoding parent spec");
+    return -EINVAL;
+  }
+  return 0;
+}
+
+static int decode_parent(bufferlist *in, uint64_t *pool_id,
+			 string *image_id, snapid_t *snap_id)
+{
+  auto it = in->cbegin();
+  return decode_parent_common(it, pool_id, image_id, snap_id);
+}
+
+static int decode_parent_and_child(bufferlist *in, uint64_t *pool_id,
+			           string *image_id, snapid_t *snap_id,
+				   string *c_image_id)
+{
+  auto it = in->cbegin();
+  int r = decode_parent_common(it, pool_id, image_id, snap_id);
+  if (r < 0)
+    return r;
+  try {
+    decode(*c_image_id, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("error decoding child image id");
+    return -EINVAL;
+  }
+  return 0;
+}
+
+static string parent_key(uint64_t pool_id, string image_id, snapid_t snap_id)
+{
+  bufferlist key_bl;
+  encode(pool_id, key_bl);
+  encode(image_id, key_bl);
+  encode(snap_id, key_bl);
+  return string(key_bl.c_str(), key_bl.length());
+}
+
+/**
+ * add child to rbd_children directory object
+ *
+ * rbd_children is a map of (p_pool_id, p_image_id, p_snap_id) to
+ * [c_image_id, [c_image_id ... ]]
+ *
+ * Input:
+ * @param p_pool_id parent pool id
+ * @param p_image_id parent image oid
+ * @param p_snap_id parent snapshot id
+ * @param c_image_id new child image oid to add
+ *
+ * @returns 0 on success, negative error on failure
+ */
+
+int add_child(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r;
+
+  uint64_t p_pool_id;
+  snapid_t p_snap_id;
+  string p_image_id, c_image_id;
+  // Use set for ease of erase() for remove_child()
+  std::set<string> children;
+
+  r = decode_parent_and_child(in, &p_pool_id, &p_image_id, &p_snap_id,
+			      &c_image_id);
+  if (r < 0)
+    return r;
+
+  CLS_LOG(20, "add_child %s to (%" PRIu64 ", %s, %" PRIu64 ")", c_image_id.c_str(),
+	  p_pool_id, p_image_id.c_str(), p_snap_id.val);
+
+  string key = parent_key(p_pool_id, p_image_id, p_snap_id);
+
+  // get current child list for parent, if any
+  r = read_key(hctx, key, &children);
+  if ((r < 0) && (r != -ENOENT)) {
+    CLS_LOG(20, "add_child: omap read failed: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (children.find(c_image_id) != children.end()) {
+    CLS_LOG(20, "add_child: child already exists: %s", c_image_id.c_str());
+    return -EEXIST;
+  }
+  // add new child
+  children.insert(c_image_id);
+
+  // write back
+  bufferlist childbl;
+  encode(children, childbl);
+  r = cls_cxx_map_set_val(hctx, key, &childbl);
+  if (r < 0)
+    CLS_LOG(20, "add_child: omap write failed: %s", cpp_strerror(r).c_str());
+  return r;
+}
+
+/**
+ * remove child from rbd_children directory object
+ *
+ * Input:
+ * @param p_pool_id parent pool id
+ * @param p_image_id parent image oid
+ * @param p_snap_id parent snapshot id
+ * @param c_image_id new child image oid to add
+ *
+ * @returns 0 on success, negative error on failure
+ */
+
+int remove_child(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r;
+
+  uint64_t p_pool_id;
+  snapid_t p_snap_id;
+  string p_image_id, c_image_id;
+  std::set<string> children;
+
+  r = decode_parent_and_child(in, &p_pool_id, &p_image_id, &p_snap_id,
+			      &c_image_id);
+  if (r < 0)
+    return r;
+
+  CLS_LOG(20, "remove_child %s from (%" PRIu64 ", %s, %" PRIu64 ")",
+	       c_image_id.c_str(), p_pool_id, p_image_id.c_str(),
+	       p_snap_id.val);
+
+  string key = parent_key(p_pool_id, p_image_id, p_snap_id);
+
+  // get current child list for parent.  Unlike add_child(), an empty list
+  // is an error (how can we remove something that doesn't exist?)
+  r = read_key(hctx, key, &children);
+  if (r < 0) {
+    CLS_LOG(20, "remove_child: read omap failed: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (children.find(c_image_id) == children.end()) {
+    CLS_LOG(20, "remove_child: child not found: %s", c_image_id.c_str());
+    return -ENOENT;
+  }
+  // find and remove child
+  children.erase(c_image_id);
+
+  // now empty?  remove key altogether
+  if (children.empty()) {
+    r = cls_cxx_map_remove_key(hctx, key);
+    if (r < 0)
+      CLS_LOG(20, "remove_child: remove key failed: %s", cpp_strerror(r).c_str());
+  } else {
+    // write back shortened children list
+    bufferlist childbl;
+    encode(children, childbl);
+    r = cls_cxx_map_set_val(hctx, key, &childbl);
+    if (r < 0)
+      CLS_LOG(20, "remove_child: write omap failed: %s", cpp_strerror(r).c_str());
+  }
+  return r;
+}
+
+/**
+ * Input:
+ * @param p_pool_id parent pool id
+ * @param p_image_id parent image oid
+ * @param p_snap_id parent snapshot id
+ * @param c_image_id new child image oid to add
+ *
+ * Output:
+ * @param children set<string> of children
+ *
+ * @returns 0 on success, negative error on failure
+ */
+int get_children(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r;
+  uint64_t p_pool_id;
+  snapid_t p_snap_id;
+  string p_image_id;
+  std::set<string> children;
+
+  r = decode_parent(in, &p_pool_id, &p_image_id, &p_snap_id);
+  if (r < 0)
+    return r;
+
+  CLS_LOG(20, "get_children of (%" PRIu64 ", %s, %" PRIu64 ")",
+	  p_pool_id, p_image_id.c_str(), p_snap_id.val);
+
+  string key = parent_key(p_pool_id, p_image_id, p_snap_id);
+
+  r = read_key(hctx, key, &children);
+  if (r < 0) {
+    if (r != -ENOENT)
+      CLS_LOG(20, "get_children: read omap failed: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  encode(children, *out);
+  return 0;
+}
+
+
+/**
+ * Get the information needed to create a rados snap context for doing
+ * I/O to the data objects. This must include all snapshots.
+ *
+ * Output:
+ * @param snap_seq the highest snapshot id ever associated with the image (uint64_t)
+ * @param snap_ids existing snapshot ids in descending order (vector<uint64_t>)
+ * @returns 0 on success, negative error code on failure
+ */
+int get_snapcontext(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_snapcontext");
+
+  int r;
+  int max_read = RBD_MAX_KEYS_READ;
+  vector<snapid_t> snap_ids;
+  string last_read = RBD_SNAP_KEY_PREFIX;
+  bool more;
+
+  do {
+    set<string> keys;
+    r = cls_cxx_map_get_keys(hctx, last_read, max_read, &keys, &more);
+    if (r < 0)
+      return r;
+
+    for (set<string>::const_iterator it = keys.begin();
+	 it != keys.end(); ++it) {
+      if ((*it).find(RBD_SNAP_KEY_PREFIX) != 0)
+	break;
+      snapid_t snap_id = snap_id_from_key(*it);
+      snap_ids.push_back(snap_id);
+    }
+    if (!keys.empty())
+      last_read = *(keys.rbegin());
+  } while (more);
+
+  uint64_t snap_seq;
+  r = read_key(hctx, "snap_seq", &snap_seq);
+  if (r < 0) {
+    CLS_ERR("could not read the image's snap_seq off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  // snap_ids must be descending in a snap context
+  std::reverse(snap_ids.begin(), snap_ids.end());
+
+  encode(snap_seq, *out);
+  encode(snap_ids, *out);
+
+  return 0;
+}
+
+/**
+ * Output:
+ * @param object_prefix prefix for data object names (string)
+ * @returns 0 on success, negative error code on failure
+ */
+int get_object_prefix(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_object_prefix");
+
+  string object_prefix;
+  int r = read_key(hctx, "object_prefix", &object_prefix);
+  if (r < 0) {
+    CLS_ERR("failed to read the image's object prefix off of disk: %s",
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  encode(object_prefix, *out);
+
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param pool_id (int64_t) of data pool or -1 if none
+ * @returns 0 on success, negative error code on failure
+ */
+int get_data_pool(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "get_data_pool");
+
+  int64_t data_pool_id = -1;
+  int r = read_key(hctx, "data_pool_id", &data_pool_id);
+  if (r == -ENOENT) {
+    data_pool_id = -1;
+  } else if (r < 0) {
+    CLS_ERR("error reading image data pool id: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  encode(data_pool_id, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap_id which snapshot to query
+ *
+ * Output:
+ * @param name (string) of the snapshot
+ * @returns 0 on success, negative error code on failure
+ */
+int get_snapshot_name(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "get_snapshot_name snap_id=%llu", (unsigned long long)snap_id);
+
+  if (snap_id == CEPH_NOSNAP)
+    return -EINVAL;
+
+  cls_rbd_snap snap;
+  string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0)
+    return r;
+
+  encode(snap.name, *out);
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap_id which snapshot to query
+ *
+ * Output:
+ * @param timestamp (utime_t) of the snapshot
+ * @returns 0 on success, negative error code on failure
+ *
+ * NOTE: deprecated - remove this method after Luminous is unsupported
+ */
+int get_snapshot_timestamp(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "get_snapshot_timestamp snap_id=%llu", (unsigned long long)snap_id);
+
+  if (snap_id == CEPH_NOSNAP) {
+    return -EINVAL;
+  }
+
+  cls_rbd_snap snap;
+  string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(snap.timestamp, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap_id which snapshot to query
+ *
+ * Output:
+ * @param snapshot (cls::rbd::SnapshotInfo)
+ * @returns 0 on success, negative error code on failure
+ */
+int snapshot_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+
+  auto iter = in->cbegin();
+  try {
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "snapshot_get snap_id=%llu", (unsigned long long)snap_id);
+  if (snap_id == CEPH_NOSNAP) {
+    return -EINVAL;
+  }
+
+  cls_rbd_snap snap;
+  string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    return r;
+  }
+
+  cls::rbd::SnapshotInfo snapshot_info{snap.id, snap.snapshot_namespace,
+                                       snap.name, snap.image_size,
+                                       snap.timestamp, snap.child_count};
+  encode(snapshot_info, *out);
+  return 0;
+}
+
+/**
+ * Adds a snapshot to an rbd header. Ensures the id and name are unique.
+ *
+ * Input:
+ * @param snap_name name of the snapshot (string)
+ * @param snap_id id of the snapshot (uint64_t)
+ * @param snap_namespace namespace of the snapshot (cls::rbd::SnapshotNamespace)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure.
+ * @returns -ESTALE if the input snap_id is less than the image's snap_seq
+ * @returns -EEXIST if the id or name are already used by another snapshot
+ */
+int snapshot_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist snap_namebl, snap_idbl;
+  cls_rbd_snap snap_meta;
+  uint64_t snap_limit;
+
+  try {
+    auto iter = in->cbegin();
+    decode(snap_meta.name, iter);
+    decode(snap_meta.id, iter);
+    if (!iter.end()) {
+      decode(snap_meta.snapshot_namespace, iter);
+    }
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (boost::get<cls::rbd::UnknownSnapshotNamespace>(
+        &snap_meta.snapshot_namespace) != nullptr) {
+    CLS_ERR("Unknown snapshot namespace provided");
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "snapshot_add name=%s id=%llu", snap_meta.name.c_str(),
+	 (unsigned long long)snap_meta.id.val);
+
+  if (snap_meta.id > CEPH_MAXSNAP)
+    return -EINVAL;
+
+  uint64_t cur_snap_seq;
+  int r = read_key(hctx, "snap_seq", &cur_snap_seq);
+  if (r < 0) {
+    CLS_ERR("Could not read image's snap_seq off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  // client lost a race with another snapshot creation.
+  // snap_seq must be monotonically increasing.
+  if (snap_meta.id < cur_snap_seq)
+    return -ESTALE;
+
+  r = read_key(hctx, "size", &snap_meta.image_size);
+  if (r < 0) {
+    CLS_ERR("Could not read image's size off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  r = read_key(hctx, "flags", &snap_meta.flags);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("Could not read image's flags off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  r = read_key(hctx, "snap_limit", &snap_limit);
+  if (r == -ENOENT) {
+    snap_limit = UINT64_MAX;
+  } else if (r < 0) {
+    CLS_ERR("Could not read snapshot limit off disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  snap_meta.timestamp = ceph_clock_now();
+
+  uint64_t total_read = 0;
+  auto pre_check_lambda =
+    [&snap_meta, &total_read, snap_limit](const cls_rbd_snap& old_meta) {
+      ++total_read;
+      if (total_read >= snap_limit) {
+        CLS_ERR("Attempt to create snapshot over limit of %" PRIu64,
+                snap_limit);
+        return -EDQUOT;
+      }
+
+      if ((snap_meta.name == old_meta.name &&
+	    snap_meta.snapshot_namespace == old_meta.snapshot_namespace) ||
+	  snap_meta.id == old_meta.id) {
+	CLS_LOG(20, "snap_name %s or snap_id %" PRIu64 " matches existing snap "
+                "%s %" PRIu64, snap_meta.name.c_str(), snap_meta.id.val,
+		old_meta.name.c_str(), old_meta.id.val);
+	return -EEXIST;
+      }
+      return 0;
+    };
+
+  r = image::snapshot::iterate(hctx, pre_check_lambda);
+  if (r < 0) {
+    return r;
+  }
+
+  // snapshot inherits parent, if any
+  cls_rbd_parent parent;
+  r = read_key(hctx, "parent", &parent);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+  if (r == 0) {
+    // write helper method will convert to normalized format if required
+    snap_meta.parent = parent;
+  }
+
+  if (cls::rbd::get_snap_namespace_type(snap_meta.snapshot_namespace) ==
+        cls::rbd::SNAPSHOT_NAMESPACE_TYPE_TRASH) {
+    // add snap_trash feature bit if not already enabled
+    r = image::set_op_features(hctx, RBD_OPERATION_FEATURE_SNAP_TRASH,
+                               RBD_OPERATION_FEATURE_SNAP_TRASH);
+    if (r < 0) {
+      return r;
+    }
+  }
+
+  r = write_key(hctx, "snap_seq", snap_meta.id);
+  if (r < 0) {
+    return r;
+  }
+
+  std::string snapshot_key;
+  key_from_snap_id(snap_meta.id, &snapshot_key);
+  r = image::snapshot::write(hctx, snapshot_key, std::move(snap_meta));
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * rename snapshot .
+ *
+ * Input:
+ * @param src_snap_id old snap id of the snapshot (snapid_t)
+ * @param dst_snap_name new name of the snapshot (string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure.
+ */
+int snapshot_rename(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist snap_namebl, snap_idbl;
+  snapid_t src_snap_id;
+  string dst_snap_name;
+  cls_rbd_snap snap_meta;
+  int r;
+
+  try {
+    auto iter = in->cbegin();
+    decode(src_snap_id, iter);
+    decode(dst_snap_name, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "snapshot_rename id=%" PRIu64 ", dst_name=%s",
+          src_snap_id.val, dst_snap_name.c_str());
+
+  auto duplicate_name_lambda = [&dst_snap_name](const cls_rbd_snap& snap_meta) {
+    if (cls::rbd::get_snap_namespace_type(snap_meta.snapshot_namespace) ==
+          cls::rbd::SNAPSHOT_NAMESPACE_TYPE_USER &&
+        snap_meta.name == dst_snap_name) {
+      CLS_LOG(20, "snap_name %s matches existing snap with snap id %" PRIu64,
+              dst_snap_name.c_str(), snap_meta.id.val);
+      return -EEXIST;
+    }
+    return 0;
+  };
+  r = image::snapshot::iterate(hctx, duplicate_name_lambda);
+  if (r < 0) {
+    return r;
+  }
+
+  std::string src_snap_key;
+  key_from_snap_id(src_snap_id, &src_snap_key);
+  r = read_key(hctx, src_snap_key, &snap_meta);
+  if (r == -ENOENT) {
+    CLS_LOG(20, "cannot find existing snap with snap id = %" PRIu64,
+            src_snap_id.val);
+    return r;
+  }
+
+  if (cls::rbd::get_snap_namespace_type(snap_meta.snapshot_namespace) !=
+        cls::rbd::SNAPSHOT_NAMESPACE_TYPE_USER) {
+    // can only rename user snapshots
+    return -EINVAL;
+  }
+
+  snap_meta.name = dst_snap_name;
+  r = image::snapshot::write(hctx, src_snap_key, std::move(snap_meta));
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Removes a snapshot from an rbd header.
+ *
+ * Input:
+ * @param snap_id the id of the snapshot to remove (uint64_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int snapshot_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  snapid_t snap_id;
+
+  try {
+    auto iter = in->cbegin();
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "snapshot_remove id=%llu", (unsigned long long)snap_id.val);
+
+  // check if the key exists. we can't rely on remove_key doing this for
+  // us, since OMAPRMKEYS returns success if the key is not there.
+  // bug or feature? sounds like a bug, since tmap did not have this
+  // behavior, but cls_rgw may rely on it...
+  cls_rbd_snap snap;
+  string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r == -ENOENT) {
+    return -ENOENT;
+  }
+
+  if (snap.protection_status != RBD_PROTECTION_STATUS_UNPROTECTED) {
+    return -EBUSY;
+  }
+
+  // snapshot is in-use by clone v2 child
+  if (snap.child_count > 0) {
+    return -EBUSY;
+  }
+
+  r = remove_key(hctx, snapshot_key);
+  if (r < 0) {
+    return r;
+  }
+
+  bool has_child_snaps = false;
+  bool has_trash_snaps = false;
+  auto remove_lambda = [snap_id, &has_child_snaps, &has_trash_snaps](
+      const cls_rbd_snap& snap_meta) {
+    if (snap_meta.id != snap_id) {
+      if (snap_meta.parent.pool_id != -1 || snap_meta.parent_overlap) {
+        has_child_snaps = true;
+      }
+
+      if (cls::rbd::get_snap_namespace_type(snap_meta.snapshot_namespace) ==
+            cls::rbd::SNAPSHOT_NAMESPACE_TYPE_TRASH) {
+        has_trash_snaps = true;
+      }
+    }
+    return 0;
+  };
+
+  r = image::snapshot::iterate(hctx, remove_lambda);
+  if (r < 0) {
+    return r;
+  }
+
+  cls_rbd_parent parent;
+  r = read_key(hctx, "parent", &parent);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  bool has_parent = (r >= 0 && parent.exists());
+  bool is_head_child = (has_parent && parent.head_overlap);
+  int8_t require_osd_release = cls_get_required_osd_release(hctx);
+  if (has_parent && !is_head_child && !has_child_snaps &&
+      require_osd_release >= CEPH_RELEASE_NAUTILUS) {
+    // remove the unused parent image spec
+    r = remove_key(hctx, "parent");
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    }
+  }
+
+  uint64_t op_features_mask = 0ULL;
+  if (!has_child_snaps && !is_head_child) {
+    // disable clone child op feature if no longer associated
+    op_features_mask |= RBD_OPERATION_FEATURE_CLONE_CHILD;
+  }
+  if (!has_trash_snaps) {
+    // remove the snap_trash op feature if not in-use by any other snapshots
+    op_features_mask |= RBD_OPERATION_FEATURE_SNAP_TRASH;
+  }
+
+  if (op_features_mask != 0ULL) {
+    r = image::set_op_features(hctx, 0, op_features_mask);
+    if (r < 0) {
+      return r;
+    }
+  }
+
+  return 0;
+}
+
+/**
+ * Moves a snapshot to the trash namespace.
+ *
+ * Input:
+ * @param snap_id the id of the snapshot to move to the trash (uint64_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int snapshot_trash_add(cls_method_context_t hctx, bufferlist *in,
+                       bufferlist *out)
+{
+  snapid_t snap_id;
+
+  try {
+    auto iter = in->cbegin();
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "snapshot_trash_add id=%" PRIu64, snap_id.val);
+
+  cls_rbd_snap snap;
+  std::string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r == -ENOENT) {
+    return r;
+  }
+
+  if (snap.protection_status != RBD_PROTECTION_STATUS_UNPROTECTED) {
+    return -EBUSY;
+  }
+
+  auto snap_type = cls::rbd::get_snap_namespace_type(snap.snapshot_namespace);
+  if (snap_type == cls::rbd::SNAPSHOT_NAMESPACE_TYPE_TRASH) {
+    return -EEXIST;
+  }
+
+  // add snap_trash feature bit if not already enabled
+  r = image::set_op_features(hctx, RBD_OPERATION_FEATURE_SNAP_TRASH,
+                             RBD_OPERATION_FEATURE_SNAP_TRASH);
+  if (r < 0) {
+    return r;
+  }
+
+  snap.snapshot_namespace = cls::rbd::TrashSnapshotNamespace{snap_type,
+                                                             snap.name};
+  uuid_d uuid_gen;
+  uuid_gen.generate_random();
+  snap.name = uuid_gen.to_string();
+
+  r = image::snapshot::write(hctx, snapshot_key, std::move(snap));
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Returns a uint64_t of all the features supported by this class.
+ */
+int get_all_features(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t all_features = RBD_FEATURES_ALL;
+  encode(all_features, *out);
+  return 0;
+}
+
+/**
+ * "Copy up" data from the parent of a clone to the clone's object(s).
+ * Used for implementing copy-on-write for a clone image.  Client
+ * will pass down a chunk of data that fits completely within one
+ * clone block (one object), and is aligned (starts at beginning of block),
+ * but may be shorter (for non-full parent blocks).  The class method
+ * can't know the object size to validate the requested length,
+ * so it just writes the data as given if the child object doesn't
+ * already exist, and returns success if it does.
+ *
+ * Input:
+ * @param in bufferlist of data to write
+ *
+ * Output:
+ * @returns 0 on success, or if block already exists in child
+ *  negative error code on other error
+ */
+
+int copyup(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // check for existence; if child object exists, just return success
+  if (cls_cxx_stat(hctx, NULL, NULL) == 0)
+    return 0;
+  CLS_LOG(20, "copyup: writing length %d\n", in->length());
+  return cls_cxx_write(hctx, 0, in->length(), in);
+}
+
+
+/************************ rbd_id object methods **************************/
+
+/**
+ * Input:
+ * @param in ignored
+ *
+ * Output:
+ * @param id the id stored in the object
+ * @returns 0 on success, negative error code on failure
+ */
+int get_id(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t size;
+  int r = cls_cxx_stat(hctx, &size, NULL);
+  if (r < 0)
+    return r;
+
+  if (size == 0)
+    return -ENOENT;
+
+  bufferlist read_bl;
+  r = cls_cxx_read(hctx, 0, size, &read_bl);
+  if (r < 0) {
+    CLS_ERR("get_id: could not read id: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  string id;
+  try {
+    auto iter = read_bl.cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EIO;
+  }
+
+  encode(id, *out);
+  return 0;
+}
+
+/**
+ * Set the id of an image. The object must already exist.
+ *
+ * Input:
+ * @param id the id of the image, as an alpha-numeric string
+ *
+ * Output:
+ * @returns 0 on success, -EEXIST if the atomic create fails,
+ *          negative error code on other error
+ */
+int set_id(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r = check_exists(hctx);
+  if (r < 0)
+    return r;
+
+  string id;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (!is_valid_id(id)) {
+    CLS_ERR("set_id: invalid id '%s'", id.c_str());
+    return -EINVAL;
+  }
+
+  uint64_t size;
+  r = cls_cxx_stat(hctx, &size, NULL);
+  if (r < 0)
+    return r;
+  if (size != 0)
+    return -EEXIST;
+
+  CLS_LOG(20, "set_id: id=%s", id.c_str());
+
+  bufferlist write_bl;
+  encode(id, write_bl);
+  return cls_cxx_write(hctx, 0, write_bl.length(), &write_bl);
+}
+
+/**
+ * Update the access timestamp of an image
+ *
+ * Input:
+ * @param none
+ *
+ * Output:
+ * @returns 0 on success, negative error code on other error
+ */
+int set_access_timestamp(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+   int r = check_exists(hctx);
+   if(r < 0)
+     return r;
+   
+   utime_t timestamp = ceph_clock_now();
+   r = write_key(hctx, "access_timestamp", timestamp);
+   if(r < 0) {
+     CLS_ERR("error setting access_timestamp");
+     return r;
+   }
+
+   return 0;
+}
+
+/**
+ * Update the modify timestamp of an image
+ *
+ * Input:
+ * @param none
+ *
+ * Output:
+ * @returns 0 on success, negative error code on other error
+ */
+
+int set_modify_timestamp(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+   int r = check_exists(hctx);
+   if(r < 0)
+     return r;
+   
+   utime_t timestamp = ceph_clock_now();
+   r = write_key(hctx, "modify_timestamp", timestamp);
+   if(r < 0) {
+     CLS_ERR("error setting modify_timestamp");
+     return r;
+   }
+
+   return 0;
+}
+
+
+
+/*********************** methods for rbd_directory ***********************/
+
+static const string dir_key_for_id(const string &id)
+{
+  return RBD_DIR_ID_KEY_PREFIX + id;
+}
+
+static const string dir_key_for_name(const string &name)
+{
+  return RBD_DIR_NAME_KEY_PREFIX + name;
+}
+
+static const string dir_name_from_key(const string &key)
+{
+  return key.substr(strlen(RBD_DIR_NAME_KEY_PREFIX));
+}
+
+static int dir_add_image_helper(cls_method_context_t hctx,
+				const string &name, const string &id,
+				bool check_for_unique_id)
+{
+  if (!name.size() || !is_valid_id(id)) {
+    CLS_ERR("dir_add_image_helper: invalid name '%s' or id '%s'",
+	    name.c_str(), id.c_str());
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "dir_add_image_helper name=%s id=%s", name.c_str(), id.c_str());
+
+  string tmp;
+  string name_key = dir_key_for_name(name);
+  string id_key = dir_key_for_id(id);
+  int r = read_key(hctx, name_key, &tmp);
+  if (r != -ENOENT) {
+    CLS_LOG(10, "name already exists");
+    return -EEXIST;
+  }
+  r = read_key(hctx, id_key, &tmp);
+  if (r != -ENOENT && check_for_unique_id) {
+    CLS_LOG(10, "id already exists");
+    return -EBADF;
+  }
+  bufferlist id_bl, name_bl;
+  encode(id, id_bl);
+  encode(name, name_bl);
+  map<string, bufferlist> omap_vals;
+  omap_vals[name_key] = id_bl;
+  omap_vals[id_key] = name_bl;
+  return cls_cxx_map_set_vals(hctx, &omap_vals);
+}
+
+static int dir_remove_image_helper(cls_method_context_t hctx,
+				   const string &name, const string &id)
+{
+  CLS_LOG(20, "dir_remove_image_helper name=%s id=%s",
+	  name.c_str(), id.c_str());
+
+  string stored_name, stored_id;
+  string name_key = dir_key_for_name(name);
+  string id_key = dir_key_for_id(id);
+  int r = read_key(hctx, name_key, &stored_id);
+  if (r < 0) {
+    if (r != -ENOENT)
+      CLS_ERR("error reading name to id mapping: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  r = read_key(hctx, id_key, &stored_name);
+  if (r < 0) {
+    CLS_ERR("error reading id to name mapping: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  // check if this op raced with a rename
+  if (stored_name != name || stored_id != id) {
+    CLS_ERR("stored name '%s' and id '%s' do not match args '%s' and '%s'",
+	    stored_name.c_str(), stored_id.c_str(), name.c_str(), id.c_str());
+    return -ESTALE;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, name_key);
+  if (r < 0) {
+    CLS_ERR("error removing name: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, id_key);
+  if (r < 0) {
+    CLS_ERR("error removing id: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Rename an image in the directory, updating both indexes
+ * atomically. This can't be done from the client calling
+ * dir_add_image and dir_remove_image in one transaction because the
+ * results of the first method are not visibale to later steps.
+ *
+ * Input:
+ * @param src original name of the image
+ * @param dest new name of the image
+ * @param id the id of the image
+ *
+ * Output:
+ * @returns -ESTALE if src and id do not map to each other
+ * @returns -ENOENT if src or id are not in the directory
+ * @returns -EEXIST if dest already exists
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_rename_image(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string src, dest, id;
+  try {
+    auto iter = in->cbegin();
+    decode(src, iter);
+    decode(dest, iter);
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = dir_remove_image_helper(hctx, src, id);
+  if (r < 0)
+    return r;
+  // ignore duplicate id because the result of
+  // remove_image_helper is not visible yet
+  return dir_add_image_helper(hctx, dest, id, false);
+}
+
+/**
+ * Get the id of an image given its name.
+ *
+ * Input:
+ * @param name the name of the image
+ *
+ * Output:
+ * @param id the id of the image
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_get_id(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string name;
+
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "dir_get_id: name=%s", name.c_str());
+
+  string id;
+  int r = read_key(hctx, dir_key_for_name(name), &id);
+  if (r < 0) {
+    if (r != -ENOENT)
+      CLS_ERR("error reading id for name '%s': %s", name.c_str(), cpp_strerror(r).c_str());
+    return r;
+  }
+  encode(id, *out);
+  return 0;
+}
+
+/**
+ * Get the name of an image given its id.
+ *
+ * Input:
+ * @param id the id of the image
+ *
+ * Output:
+ * @param name the name of the image
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_get_name(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string id;
+
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "dir_get_name: id=%s", id.c_str());
+
+  string name;
+  int r = read_key(hctx, dir_key_for_id(id), &name);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading name for id '%s': %s", id.c_str(),
+              cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+  encode(name, *out);
+  return 0;
+}
+
+/**
+ * List the names and ids of the images in the directory, sorted by
+ * name.
+ *
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param images map from name to id of up to max_return images
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string start_after;
+  uint64_t max_return;
+
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int max_read = RBD_MAX_KEYS_READ;
+  map<string, string> images;
+  string last_read = dir_key_for_name(start_after);
+  bool more = true;
+
+  while (more && images.size() < max_return) {
+    map<string, bufferlist> vals;
+    CLS_LOG(20, "last_read = '%s'", last_read.c_str());
+    int r = cls_cxx_map_get_vals(hctx, last_read, RBD_DIR_NAME_KEY_PREFIX,
+                                 max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading directory by name: %s", cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (map<string, bufferlist>::iterator it = vals.begin();
+	 it != vals.end(); ++it) {
+      string id;
+      auto iter = it->second.cbegin();
+      try {
+	decode(id, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode id of image '%s'", it->first.c_str());
+	return -EIO;
+      }
+      CLS_LOG(20, "adding '%s' -> '%s'", dir_name_from_key(it->first).c_str(), id.c_str());
+      images[dir_name_from_key(it->first)] = id;
+      if (images.size() >= max_return)
+	break;
+    }
+    if (!vals.empty()) {
+      last_read = dir_key_for_name(images.rbegin()->first);
+    }
+  }
+
+  encode(images, *out);
+
+  return 0;
+}
+
+/**
+ * Add an image to the rbd directory. Creates the directory object if
+ * needed, and updates the index from id to name and name to id.
+ *
+ * Input:
+ * @param name the name of the image
+ * @param id the id of the image
+ *
+ * Output:
+ * @returns -EEXIST if the image name is already in the directory
+ * @returns -EBADF if the image id is already in the directory
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_add_image(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r = cls_cxx_create(hctx, false);
+  if (r < 0) {
+    CLS_ERR("could not create directory: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  string name, id;
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  return dir_add_image_helper(hctx, name, id, true);
+}
+
+/**
+ * Remove an image from the rbd directory.
+ *
+ * Input:
+ * @param name the name of the image
+ * @param id the id of the image
+ *
+ * Output:
+ * @returns -ESTALE if the name and id do not map to each other
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_remove_image(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string name, id;
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  return dir_remove_image_helper(hctx, name, id);
+}
+
+/**
+ * Verify the current state of the directory
+ *
+ * Input:
+ * @param state the DirectoryState of the directory
+ *
+ * Output:
+ * @returns -ENOENT if the state does not match
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_state_assert(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  cls::rbd::DirectoryState directory_state = cls::rbd::DIRECTORY_STATE_READY;
+  try {
+    auto iter = in->cbegin();
+    decode(directory_state, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  cls::rbd::DirectoryState on_disk_directory_state = directory_state;
+  int r = read_key(hctx, "state", &on_disk_directory_state);
+  if (r < 0) {
+    return r;
+  }
+
+  if (directory_state != on_disk_directory_state) {
+    return -ENOENT;
+  }
+  return 0;
+}
+
+/**
+ * Set the current state of the directory
+ *
+ * Input:
+ * @param state the DirectoryState of the directory
+ *
+ * Output:
+ * @returns -ENOENT if the state does not match
+ * @returns 0 on success, negative error code on failure
+ */
+int dir_state_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  cls::rbd::DirectoryState directory_state;
+  try {
+    auto iter = in->cbegin();
+    decode(directory_state, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = check_exists(hctx);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  switch (directory_state) {
+  case cls::rbd::DIRECTORY_STATE_READY:
+    break;
+  case cls::rbd::DIRECTORY_STATE_ADD_DISABLED:
+    {
+      if (r == -ENOENT) {
+        return r;
+      }
+
+      // verify that the directory is empty
+      std::map<std::string, bufferlist> vals;
+      bool more;
+      r = cls_cxx_map_get_vals(hctx, RBD_DIR_NAME_KEY_PREFIX,
+                               RBD_DIR_NAME_KEY_PREFIX, 1, &vals, &more);
+      if (r < 0) {
+        return r;
+      } else if (!vals.empty()) {
+        return -EBUSY;
+      }
+    }
+    break;
+  default:
+    return -EINVAL;
+  }
+
+  r = write_key(hctx, "state", directory_state);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+int object_map_read(cls_method_context_t hctx, BitVector<2> &object_map)
+{
+  uint64_t size;
+  int r = cls_cxx_stat(hctx, &size, NULL);
+  if (r < 0) {
+    return r;
+  }
+  if (size == 0) {
+    return -ENOENT;
+  }
+
+  bufferlist bl;
+  r = cls_cxx_read(hctx, 0, size, &bl);
+  if (r < 0) {
+   return r;
+  }
+
+  try {
+    auto iter = bl.cbegin();
+    decode(object_map, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode object map: %s", err.what());
+    return -EINVAL;
+  }
+  return 0;
+}
+
+/**
+ * Load an rbd image's object map
+ *
+ * Input:
+ * none
+ *
+ * Output:
+ * @param object map bit vector
+ * @returns 0 on success, negative error code on failure
+ */
+int object_map_load(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  BitVector<2> object_map;
+  int r = object_map_read(hctx, object_map);
+  if (r < 0) {
+    return r;
+  }
+
+  object_map.set_crc_enabled(false);
+  encode(object_map, *out);
+  return 0;
+}
+
+/**
+ * Save an rbd image's object map
+ *
+ * Input:
+ * @param object map bit vector
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int object_map_save(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  BitVector<2> object_map;
+  try {
+    auto iter = in->cbegin();
+    decode(object_map, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  object_map.set_crc_enabled(true);
+
+  bufferlist bl;
+  encode(object_map, bl);
+  CLS_LOG(20, "object_map_save: object size=%" PRIu64 ", byte size=%u",
+	  object_map.size(), bl.length());
+  return cls_cxx_write_full(hctx, &bl);
+}
+
+/**
+ * Resize an rbd image's object map
+ *
+ * Input:
+ * @param object_count the max number of objects in the image
+ * @param default_state the default state of newly created objects
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int object_map_resize(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t object_count;
+  uint8_t default_state;
+  try {
+    auto iter = in->cbegin();
+    decode(object_count, iter);
+    decode(default_state, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  // protect against excessive memory requirements
+  if (object_count > cls::rbd::MAX_OBJECT_MAP_OBJECT_COUNT) {
+    CLS_ERR("object map too large: %" PRIu64, object_count);
+    return -EINVAL;
+  }
+
+  BitVector<2> object_map;
+  int r = object_map_read(hctx, object_map);
+  if ((r < 0) && (r != -ENOENT)) {
+    return r;
+  }
+
+  size_t orig_object_map_size = object_map.size();
+  if (object_count < orig_object_map_size) {
+    auto it = object_map.begin() + object_count;
+    auto end_it = object_map.end() ;
+    uint64_t i = object_count;
+    for (; it != end_it; ++it, ++i) {
+      if (*it != default_state) {
+	CLS_ERR("object map indicates object still exists: %" PRIu64, i);
+	return -ESTALE;
+      }
+    }
+    object_map.resize(object_count);
+  } else if (object_count > orig_object_map_size) {
+    object_map.resize(object_count);
+    auto it = object_map.begin() + orig_object_map_size;
+    auto end_it = object_map.end();
+    for (; it != end_it; ++it) {
+      *it = default_state;
+    }
+  }
+
+  bufferlist map;
+  encode(object_map, map);
+  CLS_LOG(20, "object_map_resize: object size=%" PRIu64 ", byte size=%u",
+	  object_count, map.length());
+  return cls_cxx_write_full(hctx, &map);
+}
+
+/**
+ * Update an rbd image's object map
+ *
+ * Input:
+ * @param start_object_no the start object iterator
+ * @param end_object_no the end object iterator
+ * @param new_object_state the new object state
+ * @param current_object_state optional current object state filter
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int object_map_update(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t start_object_no;
+  uint64_t end_object_no;
+  uint8_t new_object_state;
+  boost::optional<uint8_t> current_object_state;
+  try {
+    auto iter = in->cbegin();
+    decode(start_object_no, iter);
+    decode(end_object_no, iter);
+    decode(new_object_state, iter);
+    decode(current_object_state, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode message");
+    return -EINVAL;
+  }
+
+  uint64_t size;
+  int r = cls_cxx_stat(hctx, &size, NULL);
+  if (r < 0) {
+    return r;
+  }
+
+  BitVector<2> object_map;
+  bufferlist header_bl;
+  r = cls_cxx_read2(hctx, 0, object_map.get_header_length(), &header_bl,
+                    CEPH_OSD_OP_FLAG_FADVISE_WILLNEED);
+  if (r < 0) {
+    CLS_ERR("object map header read failed");
+    return r;
+  }
+
+  try {
+    auto it = header_bl.cbegin();
+    object_map.decode_header(it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode object map header: %s", err.what());
+    return -EINVAL;
+  }
+
+  uint64_t object_byte_offset;
+  uint64_t byte_length;
+  object_map.get_header_crc_extents(&object_byte_offset, &byte_length);
+
+  bufferlist footer_bl;
+  r = cls_cxx_read2(hctx, object_byte_offset, byte_length, &footer_bl,
+                    CEPH_OSD_OP_FLAG_FADVISE_WILLNEED);
+  if (r < 0) {
+    CLS_ERR("object map footer read header CRC failed");
+    return r;
+  }
+
+  try {
+    auto it = footer_bl.cbegin();
+    object_map.decode_header_crc(it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode object map header CRC: %s", err.what());
+  }
+
+  if (start_object_no >= end_object_no || end_object_no > object_map.size()) {
+    return -ERANGE;
+  }
+
+  uint64_t object_count = end_object_no - start_object_no;
+  object_map.get_data_crcs_extents(start_object_no, object_count,
+                                   &object_byte_offset, &byte_length);
+  const auto footer_object_offset = object_byte_offset;
+
+  footer_bl.clear();
+  r = cls_cxx_read2(hctx, object_byte_offset, byte_length, &footer_bl,
+                    CEPH_OSD_OP_FLAG_FADVISE_WILLNEED);
+  if (r < 0) {
+    CLS_ERR("object map footer read data CRCs failed");
+    return r;
+  }
+
+  try {
+    auto it = footer_bl.cbegin();
+    object_map.decode_data_crcs(it, start_object_no);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode object map data CRCs: %s", err.what());
+  }
+
+  uint64_t data_byte_offset;
+  object_map.get_data_extents(start_object_no, object_count,
+                              &data_byte_offset, &object_byte_offset,
+                              &byte_length);
+
+  bufferlist data_bl;
+  r = cls_cxx_read2(hctx, object_byte_offset, byte_length, &data_bl,
+                    CEPH_OSD_OP_FLAG_FADVISE_WILLNEED);
+  if (r < 0) {
+    CLS_ERR("object map data read failed");
+    return r;
+  }
+
+  try {
+    auto it = data_bl.cbegin();
+    object_map.decode_data(it, data_byte_offset);
+  } catch (const buffer::error &err) {
+    CLS_ERR("failed to decode data chunk [%" PRIu64 "]: %s",
+	    data_byte_offset, err.what());
+    return -EINVAL;
+  }
+
+  bool updated = false;
+  auto it = object_map.begin() + start_object_no;
+  auto end_it = object_map.begin() + end_object_no;
+  for (; it != end_it; ++it) {
+    uint8_t state = *it;
+    if ((!current_object_state || state == *current_object_state ||
+        (*current_object_state == OBJECT_EXISTS &&
+         state == OBJECT_EXISTS_CLEAN)) && state != new_object_state) {
+      *it = new_object_state;
+      updated = true;
+    }
+  }
+
+  if (updated) {
+    CLS_LOG(20, "object_map_update: %" PRIu64 "~%" PRIu64 " -> %" PRIu64,
+	    data_byte_offset, byte_length, object_byte_offset);
+
+    bufferlist data_bl;
+    object_map.encode_data(data_bl, data_byte_offset, byte_length);
+    r = cls_cxx_write2(hctx, object_byte_offset, data_bl.length(), &data_bl,
+                       CEPH_OSD_OP_FLAG_FADVISE_WILLNEED);
+    if (r < 0) {
+      CLS_ERR("failed to write object map header: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+
+    footer_bl.clear();
+    object_map.encode_data_crcs(footer_bl, start_object_no, object_count);
+    r = cls_cxx_write2(hctx, footer_object_offset, footer_bl.length(),
+		       &footer_bl, CEPH_OSD_OP_FLAG_FADVISE_WILLNEED);
+    if (r < 0) {
+      CLS_ERR("failed to write object map footer: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    CLS_LOG(20, "object_map_update: no update necessary");
+  }
+
+  return 0;
+}
+
+/**
+ * Mark all _EXISTS objects as _EXISTS_CLEAN so future writes to the
+ * image HEAD can be tracked.
+ *
+ * Input:
+ * none
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int object_map_snap_add(cls_method_context_t hctx, bufferlist *in,
+                        bufferlist *out)
+{
+  BitVector<2> object_map;
+  int r = object_map_read(hctx, object_map);
+  if (r < 0) {
+    return r;
+  }
+
+  bool updated = false;
+  auto it = object_map.begin();
+  auto end_it = object_map.end();
+  for (; it != end_it; ++it) {
+    if (*it == OBJECT_EXISTS) {
+      *it = OBJECT_EXISTS_CLEAN;
+      updated = true;
+    }
+  }
+
+  if (updated) {
+    bufferlist bl;
+    encode(object_map, bl);
+    r = cls_cxx_write_full(hctx, &bl);
+  }
+  return r;
+}
+
+/**
+ * Mark all _EXISTS_CLEAN objects as _EXISTS in the current object map
+ * if the provided snapshot object map object is marked as _EXISTS.
+ *
+ * Input:
+ * @param snapshot object map bit vector
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int object_map_snap_remove(cls_method_context_t hctx, bufferlist *in,
+                           bufferlist *out)
+{
+  BitVector<2> src_object_map;
+  try {
+    auto iter = in->cbegin();
+    decode(src_object_map, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  BitVector<2> dst_object_map;
+  int r = object_map_read(hctx, dst_object_map);
+  if (r < 0) {
+    return r;
+  }
+
+  bool updated = false;
+  auto src_it = src_object_map.begin();
+  auto dst_it = dst_object_map.begin();
+  auto dst_it_end = dst_object_map.end();
+  uint64_t i = 0;
+  for (; dst_it != dst_it_end; ++dst_it) {
+    if (*dst_it == OBJECT_EXISTS_CLEAN &&
+        (i >= src_object_map.size() || *src_it == OBJECT_EXISTS)) {
+      *dst_it = OBJECT_EXISTS;
+      updated = true;
+    }
+    if (i < src_object_map.size())
+      ++src_it;
+    ++i;
+  }
+
+  if (updated) {
+    bufferlist bl;
+    encode(dst_object_map, bl);
+    r = cls_cxx_write_full(hctx, &bl);
+  }
+  return r;
+}
+
+static const string metadata_key_for_name(const string &name)
+{
+  return RBD_METADATA_KEY_PREFIX + name;
+}
+
+static const string metadata_name_from_key(const string &key)
+{
+  return key.substr(strlen(RBD_METADATA_KEY_PREFIX));
+}
+
+/**
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+
+ * Output:
+ * @param value
+ * @returns 0 on success, negative error code on failure
+ */
+int metadata_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string start_after;
+  uint64_t max_return;
+
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  // TODO remove implicit support for zero during the N-release
+  if (max_return == 0) {
+    max_return = RBD_MAX_KEYS_READ;
+  }
+
+  map<string, bufferlist> data;
+  string last_read = metadata_key_for_name(start_after);
+  bool more = true;
+
+  while (more && data.size() < max_return) {
+    map<string, bufferlist> raw_data;
+    int max_read = std::min<uint64_t>(RBD_MAX_KEYS_READ, max_return - data.size());
+    int r = cls_cxx_map_get_vals(hctx, last_read, RBD_METADATA_KEY_PREFIX,
+                                 max_read, &raw_data, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("failed to read the vals off of disk: %s",
+                cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto& kv : raw_data) {
+      data[metadata_name_from_key(kv.first)].swap(kv.second);
+    }
+
+    if (!raw_data.empty()) {
+      last_read = raw_data.rbegin()->first;
+    }
+  }
+
+  encode(data, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param data <map(key, value)>
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int metadata_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  map<string, bufferlist> data, raw_data;
+
+  auto iter = in->cbegin();
+  try {
+    decode(data, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  for (map<string, bufferlist>::iterator it = data.begin();
+       it != data.end(); ++it) {
+    CLS_LOG(20, "metadata_set key=%s value=%.*s", it->first.c_str(),
+	    it->second.length(), it->second.c_str());
+    raw_data[metadata_key_for_name(it->first)].swap(it->second);
+  }
+  int r = cls_cxx_map_set_vals(hctx, &raw_data);
+  if (r < 0) {
+    CLS_ERR("error writing metadata: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param key
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int metadata_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string key;
+
+  auto iter = in->cbegin();
+  try {
+    decode(key, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "metadata_remove key=%s", key.c_str());
+
+  int r = cls_cxx_map_remove_key(hctx, metadata_key_for_name(key));
+  if (r < 0) {
+    CLS_ERR("error removing metadata: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param key
+ *
+ * Output:
+ * @param metadata value associated with the key
+ * @returns 0 on success, negative error code on failure
+ */
+int metadata_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string key;
+  bufferlist value;
+
+  auto iter = in->cbegin();
+  try {
+    decode(key, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "metadata_get key=%s", key.c_str());
+
+  int r = cls_cxx_map_get_val(hctx, metadata_key_for_name(key), &value);
+  if (r < 0) {
+    if (r != -ENOENT)
+      CLS_ERR("error getting metadata: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  encode(value, *out);
+  return 0;
+}
+
+int snapshot_get_limit(cls_method_context_t hctx, bufferlist *in,
+		       bufferlist *out)
+{
+  uint64_t snap_limit;
+  int r = read_key(hctx, "snap_limit", &snap_limit);
+  if (r == -ENOENT) {
+    snap_limit = UINT64_MAX;
+  } else if (r < 0) {
+    CLS_ERR("error retrieving snapshot limit: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  CLS_LOG(20, "read snapshot limit %" PRIu64, snap_limit);
+  encode(snap_limit, *out);
+
+  return 0;
+}
+
+int snapshot_set_limit(cls_method_context_t hctx, bufferlist *in,
+		       bufferlist *out)
+{
+  int rc;
+  uint64_t new_limit;
+  bufferlist bl;
+  size_t snap_count = 0;
+
+  try {
+    auto iter = in->cbegin();
+    decode(new_limit, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (new_limit == UINT64_MAX) {
+    CLS_LOG(20, "remove snapshot limit\n");
+    rc = cls_cxx_map_remove_key(hctx, "snap_limit");
+    return rc;
+  }
+
+  //try to read header as v1 format
+  rc = snap_read_header(hctx, bl);
+
+  // error when reading header
+  if (rc < 0 && rc != -EINVAL) {
+    return rc;
+  } else if (rc >= 0) {
+    // success, the image is v1 format
+    struct rbd_obj_header_ondisk *header;
+    header = (struct rbd_obj_header_ondisk *)bl.c_str();
+    snap_count = header->snap_count;
+  } else {
+    // else, the image is v2 format
+    int max_read = RBD_MAX_KEYS_READ;
+    string last_read = RBD_SNAP_KEY_PREFIX;
+    bool more;
+
+    do {
+      set<string> keys;
+      rc = cls_cxx_map_get_keys(hctx, last_read, max_read, &keys, &more);
+      if (rc < 0) {
+        CLS_ERR("error retrieving snapshots: %s", cpp_strerror(rc).c_str());
+        return rc;
+      }
+      for (auto& key : keys) {
+        if (key.find(RBD_SNAP_KEY_PREFIX) != 0)
+          break;
+        snap_count++;
+      }
+      if (!keys.empty())
+        last_read = *(keys.rbegin());
+    } while (more);
+  }
+
+  if (new_limit < snap_count) {
+    rc = -ERANGE;
+    CLS_LOG(10, "snapshot limit is less than the number of snapshots.\n");
+  } else {
+    CLS_LOG(20, "set snapshot limit to %" PRIu64 "\n", new_limit);
+    bl.clear();
+    encode(new_limit, bl);
+    rc = cls_cxx_map_set_val(hctx, "snap_limit", &bl);
+  }
+
+  return rc;
+}
+
+
+/**
+ * Input:
+ * @param snap id (uint64_t) parent snapshot id
+ * @param child spec (cls::rbd::ChildImageSpec) child image
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int child_attach(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+  cls::rbd::ChildImageSpec child_image;
+  try {
+    auto it = in->cbegin();
+    decode(snap_id, it);
+    decode(child_image, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "child_attach snap_id=%" PRIu64 ", child_pool_id=%" PRIi64 ", "
+              "child_image_id=%s", snap_id, child_image.pool_id,
+               child_image.image_id.c_str());
+
+  cls_rbd_snap snap;
+  std::string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    return r;
+  }
+
+  if (cls::rbd::get_snap_namespace_type(snap.snapshot_namespace) ==
+        cls::rbd::SNAPSHOT_NAMESPACE_TYPE_TRASH) {
+    // cannot attach to a deleted snapshot
+    return -ENOENT;
+  }
+
+  auto children_key = image::snap_children_key_from_snap_id(snap_id);
+  cls::rbd::ChildImageSpecs child_images;
+  r = read_key(hctx, children_key, &child_images);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error reading snapshot children: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  auto it = child_images.insert(child_image);
+  if (!it.second) {
+    // child already attached to the snapshot
+    return -EEXIST;
+  }
+
+  r = write_key(hctx, children_key, child_images);
+  if (r < 0) {
+    CLS_ERR("error writing snapshot children: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  ++snap.child_count;
+  r = image::snapshot::write(hctx, snapshot_key, std::move(snap));
+  if (r < 0) {
+    return r;
+  }
+
+  r = image::set_op_features(hctx, RBD_OPERATION_FEATURE_CLONE_PARENT,
+                             RBD_OPERATION_FEATURE_CLONE_PARENT);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap id (uint64_t) parent snapshot id
+ * @param child spec (cls::rbd::ChildImageSpec) child image
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int child_detach(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+  cls::rbd::ChildImageSpec child_image;
+  try {
+    auto it = in->cbegin();
+    decode(snap_id, it);
+    decode(child_image, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "child_detach snap_id=%" PRIu64 ", child_pool_id=%" PRIi64 ", "
+              "child_image_id=%s", snap_id, child_image.pool_id,
+               child_image.image_id.c_str());
+
+  cls_rbd_snap snap;
+  std::string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    return r;
+  }
+
+  auto children_key = image::snap_children_key_from_snap_id(snap_id);
+  cls::rbd::ChildImageSpecs child_images;
+  r = read_key(hctx, children_key, &child_images);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error reading snapshot children: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (snap.child_count != child_images.size()) {
+    // children and reference count don't match
+    CLS_ERR("children reference count mismatch: %" PRIu64, snap_id);
+    return -EINVAL;
+  }
+
+  if (child_images.erase(child_image) == 0) {
+    // child not attached to the snapshot
+    return -ENOENT;
+  }
+
+  if (child_images.empty()) {
+    r = remove_key(hctx, children_key);
+  } else {
+    r = write_key(hctx, children_key, child_images);
+    if (r < 0) {
+      CLS_ERR("error writing snapshot children: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  }
+
+  --snap.child_count;
+  r = image::snapshot::write(hctx, snapshot_key, std::move(snap));
+  if (r < 0) {
+    return r;
+  }
+
+  if (snap.child_count == 0) {
+    auto clone_in_use_lambda = [snap_id](const cls_rbd_snap& snap_meta) {
+      if (snap_meta.id != snap_id && snap_meta.child_count > 0) {
+        return -EEXIST;
+      }
+      return 0;
+    };
+
+    r = image::snapshot::iterate(hctx, clone_in_use_lambda);
+    if (r < 0 && r != -EEXIST) {
+      return r;
+    }
+
+    if (r != -EEXIST) {
+      // remove the clone_v2 op feature if not in-use by any other snapshots
+      r = image::set_op_features(hctx, 0, RBD_OPERATION_FEATURE_CLONE_PARENT);
+      if (r < 0) {
+        return r;
+      }
+    }
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param snap id (uint64_t) parent snapshot id
+ *
+ * Output:
+ * @param (cls::rbd::ChildImageSpecs) child images
+ * @returns 0 on success, negative error code on failure
+ */
+int children_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  uint64_t snap_id;
+  try {
+    auto it = in->cbegin();
+    decode(snap_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "child_detach snap_id=%" PRIu64, snap_id);
+
+  cls_rbd_snap snap;
+  std::string snapshot_key;
+  key_from_snap_id(snap_id, &snapshot_key);
+  int r = read_key(hctx, snapshot_key, &snap);
+  if (r < 0) {
+    return r;
+  }
+
+  auto children_key = image::snap_children_key_from_snap_id(snap_id);
+  cls::rbd::ChildImageSpecs child_images;
+  r = read_key(hctx, children_key, &child_images);
+  if (r == -ENOENT) {
+    return r;
+  } else if (r < 0) {
+    CLS_ERR("error reading snapshot children: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  encode(child_images, *out);
+  return 0;
+}
+
+/**
+ * Set image migration.
+ *
+ * Input:
+ * @param migration_spec (cls::rbd::MigrationSpec) image migration spec
+ *
+ * Output:
+ *
+ * @returns 0 on success, negative error code on failure
+ */
+int migration_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out) {
+  cls::rbd::MigrationSpec migration_spec;
+  try {
+    auto it = in->cbegin();
+    decode(migration_spec, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = image::set_migration(hctx, migration_spec, true);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Set image migration state.
+ *
+ * Input:
+ * @param state (cls::rbd::MigrationState) migration state
+ * @param description (std::string) migration state description
+ *
+ * Output:
+ *
+ * @returns 0 on success, negative error code on failure
+ */
+int migration_set_state(cls_method_context_t hctx, bufferlist *in,
+                        bufferlist *out) {
+  cls::rbd::MigrationState state;
+  std::string description;
+  try {
+    auto it = in->cbegin();
+    decode(state, it);
+    decode(description, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  cls::rbd::MigrationSpec migration_spec;
+  int r = image::read_migration(hctx, &migration_spec);
+  if (r < 0) {
+    return r;
+  }
+
+  migration_spec.state = state;
+  migration_spec.state_description = description;
+
+  r = image::set_migration(hctx, migration_spec, false);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Get image migration spec.
+ *
+ * Input:
+ *
+ * Output:
+ * @param migration_spec (cls::rbd::MigrationSpec) image migration spec
+ *
+ * @returns 0 on success, negative error code on failure
+ */
+int migration_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out) {
+  cls::rbd::MigrationSpec migration_spec;
+  int r = image::read_migration(hctx, &migration_spec);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(migration_spec, *out);
+
+  return 0;
+}
+
+/**
+ * Remove image migration spec.
+ *
+ * Input:
+ *
+ * Output:
+ *
+ * @returns 0 on success, negative error code on failure
+ */
+int migration_remove(cls_method_context_t hctx, bufferlist *in,
+                     bufferlist *out) {
+  int r = image::remove_migration(hctx);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Ensure writer snapc state
+ *
+ * Input:
+ * @param snap id (uint64_t) snap context sequence id
+ * @param state (cls::rbd::AssertSnapcSeqState) snap context state
+ *
+ * Output:
+ * @returns -ERANGE if assertion fails
+ * @returns 0 on success, negative error code on failure
+ */
+int assert_snapc_seq(cls_method_context_t hctx, bufferlist *in,
+                     bufferlist *out)
+{
+  uint64_t snapc_seq;
+  cls::rbd::AssertSnapcSeqState state;
+  try {
+    auto it = in->cbegin();
+    decode(snapc_seq, it);
+    decode(state, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  uint64_t snapset_seq;
+  int r = cls_get_snapset_seq(hctx, &snapset_seq);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  switch (state) {
+  case cls::rbd::ASSERT_SNAPC_SEQ_GT_SNAPSET_SEQ:
+    return (r == -ENOENT || snapc_seq > snapset_seq) ? 0 : -ERANGE;
+  case cls::rbd::ASSERT_SNAPC_SEQ_LE_SNAPSET_SEQ:
+    return (r == -ENOENT || snapc_seq > snapset_seq) ? -ERANGE : 0;
+  default:
+    return -EOPNOTSUPP;
+  }
+}
+
+/****************************** Old format *******************************/
+
+int old_snapshots_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist bl;
+  struct rbd_obj_header_ondisk *header;
+  int rc = snap_read_header(hctx, bl);
+  if (rc < 0)
+    return rc;
+
+  header = (struct rbd_obj_header_ondisk *)bl.c_str();
+  bufferptr p(header->snap_names_len);
+  char *buf = (char *)header;
+  char *name = buf + sizeof(*header) + header->snap_count * sizeof(struct rbd_obj_snap_ondisk);
+  char *end = name + header->snap_names_len;
+  memcpy(p.c_str(),
+         buf + sizeof(*header) + header->snap_count * sizeof(struct rbd_obj_snap_ondisk),
+         header->snap_names_len);
+
+  encode(header->snap_seq, *out);
+  encode(header->snap_count, *out);
+
+  for (unsigned i = 0; i < header->snap_count; i++) {
+    string s = name;
+    encode(header->snaps[i].id, *out);
+    encode(header->snaps[i].image_size, *out);
+    encode(s, *out);
+
+    name += strlen(name) + 1;
+    if (name > end)
+      return -EIO;
+  }
+
+  return 0;
+}
+
+int old_snapshot_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist bl;
+  struct rbd_obj_header_ondisk *header;
+  bufferlist newbl;
+  bufferptr header_bp(sizeof(*header));
+  struct rbd_obj_snap_ondisk *new_snaps;
+
+  int rc = snap_read_header(hctx, bl);
+  if (rc < 0)
+    return rc;
+
+  header = (struct rbd_obj_header_ondisk *)bl.c_str();
+
+  int snaps_id_ofs = sizeof(*header);
+  int names_ofs = snaps_id_ofs + sizeof(*new_snaps) * header->snap_count;
+  const char *snap_name;
+  const char *snap_names = ((char *)header) + names_ofs;
+  const char *end = snap_names + header->snap_names_len;
+  auto iter = in->cbegin();
+  string s;
+  uint64_t snap_id;
+
+  try {
+    decode(s, iter);
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+  snap_name = s.c_str();
+
+  if (header->snap_seq > snap_id)
+    return -ESTALE;
+
+  uint64_t snap_limit;
+  rc = read_key(hctx, "snap_limit", &snap_limit);
+  if (rc == -ENOENT) {
+    snap_limit = UINT64_MAX;
+  } else if (rc < 0) {
+    return rc;
+  }
+
+  if (header->snap_count >= snap_limit)
+    return -EDQUOT;
+
+  const char *cur_snap_name;
+  for (cur_snap_name = snap_names; cur_snap_name < end; cur_snap_name += strlen(cur_snap_name) + 1) {
+    if (strncmp(cur_snap_name, snap_name, end - cur_snap_name) == 0)
+      return -EEXIST;
+  }
+  if (cur_snap_name > end)
+    return -EIO;
+
+  int snap_name_len = strlen(snap_name);
+
+  bufferptr new_names_bp(header->snap_names_len + snap_name_len + 1);
+  bufferptr new_snaps_bp(sizeof(*new_snaps) * (header->snap_count + 1));
+
+  /* copy snap names and append to new snap name */
+  char *new_snap_names = new_names_bp.c_str();
+  strcpy(new_snap_names, snap_name);
+  memcpy(new_snap_names + snap_name_len + 1, snap_names, header->snap_names_len);
+
+  /* append new snap id */
+  new_snaps = (struct rbd_obj_snap_ondisk *)new_snaps_bp.c_str();
+  memcpy(new_snaps + 1, header->snaps, sizeof(*new_snaps) * header->snap_count);
+
+  header->snap_count = header->snap_count + 1;
+  header->snap_names_len = header->snap_names_len + snap_name_len + 1;
+  header->snap_seq = snap_id;
+
+  new_snaps[0].id = snap_id;
+  new_snaps[0].image_size = header->image_size;
+
+  memcpy(header_bp.c_str(), header, sizeof(*header));
+
+  newbl.push_back(header_bp);
+  newbl.push_back(new_snaps_bp);
+  newbl.push_back(new_names_bp);
+
+  rc = cls_cxx_write_full(hctx, &newbl);
+  if (rc < 0)
+    return rc;
+
+  return 0;
+}
+
+int old_snapshot_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist bl;
+  struct rbd_obj_header_ondisk *header;
+  bufferlist newbl;
+  bufferptr header_bp(sizeof(*header));
+
+  int rc = snap_read_header(hctx, bl);
+  if (rc < 0)
+    return rc;
+
+  header = (struct rbd_obj_header_ondisk *)bl.c_str();
+
+  int snaps_id_ofs = sizeof(*header);
+  int names_ofs = snaps_id_ofs + sizeof(struct rbd_obj_snap_ondisk) * header->snap_count;
+  const char *snap_name;
+  const char *snap_names = ((char *)header) + names_ofs;
+  const char *orig_names = snap_names;
+  const char *end = snap_names + header->snap_names_len;
+  auto iter = in->cbegin();
+  string s;
+  unsigned i;
+  bool found = false;
+  struct rbd_obj_snap_ondisk snap;
+
+  try {
+    decode(s, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+  snap_name = s.c_str();
+
+  for (i = 0; snap_names < end; i++) {
+    if (strcmp(snap_names, snap_name) == 0) {
+      snap = header->snaps[i];
+      found = true;
+      break;
+    }
+    snap_names += strlen(snap_names) + 1;
+  }
+  if (!found) {
+    CLS_ERR("couldn't find snap %s\n", snap_name);
+    return -ENOENT;
+  }
+
+  header->snap_names_len  = header->snap_names_len - (s.length() + 1);
+  header->snap_count = header->snap_count - 1;
+
+  bufferptr new_names_bp(header->snap_names_len);
+  bufferptr new_snaps_bp(sizeof(header->snaps[0]) * header->snap_count);
+
+  memcpy(header_bp.c_str(), header, sizeof(*header));
+  newbl.push_back(header_bp);
+
+  if (header->snap_count) {
+    int snaps_len = 0;
+    int names_len = 0;
+    CLS_LOG(20, "i=%u\n", i);
+    if (i > 0) {
+      snaps_len = sizeof(header->snaps[0]) * i;
+      names_len =  snap_names - orig_names;
+      memcpy(new_snaps_bp.c_str(), header->snaps, snaps_len);
+      memcpy(new_names_bp.c_str(), orig_names, names_len);
+    }
+    snap_names += s.length() + 1;
+
+    if (i < header->snap_count) {
+      memcpy(new_snaps_bp.c_str() + snaps_len,
+             header->snaps + i + 1,
+             sizeof(header->snaps[0]) * (header->snap_count - i));
+      memcpy(new_names_bp.c_str() + names_len, snap_names , end - snap_names);
+    }
+    newbl.push_back(new_snaps_bp);
+    newbl.push_back(new_names_bp);
+  }
+
+  rc = cls_cxx_write_full(hctx, &newbl);
+  if (rc < 0)
+    return rc;
+
+  return 0;
+}
+
+/**
+ * rename snapshot of old format.
+ *
+ * Input:
+ * @param src_snap_id old snap id of the snapshot (snapid_t)
+ * @param dst_snap_name new name of the snapshot (string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure.
+*/
+int old_snapshot_rename(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist bl;
+  struct rbd_obj_header_ondisk *header;
+  bufferlist newbl;
+  bufferptr header_bp(sizeof(*header));
+  snapid_t src_snap_id;
+  const char *dst_snap_name;
+  string dst;
+
+  int rc = snap_read_header(hctx, bl);
+  if (rc < 0)
+    return rc;
+
+  header = (struct rbd_obj_header_ondisk *)bl.c_str();
+
+  int snaps_id_ofs = sizeof(*header);
+  int names_ofs = snaps_id_ofs + sizeof(rbd_obj_snap_ondisk) * header->snap_count;
+  const char *snap_names = ((char *)header) + names_ofs;
+  const char *orig_names = snap_names;
+  const char *end = snap_names + header->snap_names_len;
+  auto iter = in->cbegin();
+  unsigned i;
+  bool found = false;
+
+  try {
+    decode(src_snap_id, iter);
+    decode(dst, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+  dst_snap_name = dst.c_str();
+
+  const char *cur_snap_name;
+  for (cur_snap_name = snap_names; cur_snap_name < end; 
+    cur_snap_name += strlen(cur_snap_name) + 1) {
+    if (strcmp(cur_snap_name, dst_snap_name) == 0)
+      return -EEXIST;
+  }
+  if (cur_snap_name > end)
+    return -EIO;
+  for (i = 0; i < header->snap_count; i++) {
+    if (src_snap_id == header->snaps[i].id) {
+      found = true;
+      break;
+    }
+    snap_names += strlen(snap_names) + 1;
+  }
+  if (!found) {
+    CLS_ERR("couldn't find snap %llu\n", (unsigned long long)src_snap_id.val);
+    return -ENOENT;
+  }
+  
+  CLS_LOG(20, "rename snap with snap id %llu to dest name %s", (unsigned long long)src_snap_id.val, dst_snap_name);
+  header->snap_names_len  = header->snap_names_len - strlen(snap_names) + dst.length();
+
+  bufferptr new_names_bp(header->snap_names_len);
+  bufferptr new_snaps_bp(sizeof(header->snaps[0]) * header->snap_count);
+
+  if (header->snap_count) {
+    int names_len = 0;
+    CLS_LOG(20, "i=%u\n", i);
+    if (i > 0) {
+      names_len =  snap_names - orig_names;
+      memcpy(new_names_bp.c_str(), orig_names, names_len);
+    }
+    strcpy(new_names_bp.c_str() + names_len, dst_snap_name);
+    names_len += strlen(dst_snap_name) + 1;
+    snap_names += strlen(snap_names) + 1;
+    if (i < header->snap_count) {
+      memcpy(new_names_bp.c_str() + names_len, snap_names , end - snap_names);
+    }
+    memcpy(new_snaps_bp.c_str(), header->snaps, sizeof(header->snaps[0]) * header->snap_count);
+  }
+
+  memcpy(header_bp.c_str(), header, sizeof(*header));
+  newbl.push_back(header_bp);
+  newbl.push_back(new_snaps_bp);
+  newbl.push_back(new_names_bp);
+
+  rc = cls_cxx_write_full(hctx, &newbl);
+  if (rc < 0)
+    return rc;
+  return 0;
+}
+
+
+namespace mirror {
+
+static const std::string UUID("mirror_uuid");
+static const std::string MODE("mirror_mode");
+static const std::string PEER_KEY_PREFIX("mirror_peer_");
+static const std::string IMAGE_KEY_PREFIX("image_");
+static const std::string GLOBAL_KEY_PREFIX("global_");
+static const std::string STATUS_GLOBAL_KEY_PREFIX("status_global_");
+static const std::string INSTANCE_KEY_PREFIX("instance_");
+static const std::string MIRROR_IMAGE_MAP_KEY_PREFIX("image_map_");
+
+std::string peer_key(const std::string &uuid) {
+  return PEER_KEY_PREFIX + uuid;
+}
+
+std::string image_key(const string &image_id) {
+  return IMAGE_KEY_PREFIX + image_id;
+}
+
+std::string global_key(const string &global_id) {
+  return GLOBAL_KEY_PREFIX + global_id;
+}
+
+std::string status_global_key(const string &global_id) {
+  return STATUS_GLOBAL_KEY_PREFIX + global_id;
+}
+
+std::string instance_key(const string &instance_id) {
+  return INSTANCE_KEY_PREFIX + instance_id;
+}
+
+std::string mirror_image_map_key(const string& global_image_id) {
+  return MIRROR_IMAGE_MAP_KEY_PREFIX + global_image_id;
+}
+
+int uuid_get(cls_method_context_t hctx, std::string *mirror_uuid) {
+  bufferlist mirror_uuid_bl;
+  int r = cls_cxx_map_get_val(hctx, mirror::UUID, &mirror_uuid_bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading mirror uuid: %s", cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  *mirror_uuid = std::string(mirror_uuid_bl.c_str(), mirror_uuid_bl.length());
+  return 0;
+}
+
+void sanitize_entity_inst(entity_inst_t* entity_inst) {
+  // make all addrs of type ANY because the type isn't what uniquely
+  // identifies them and clients and on-disk formats can be encoded
+  // with different backwards compatibility settings.
+  entity_inst->addr.set_type(entity_addr_t::TYPE_ANY);
+}
+
+int list_watchers(cls_method_context_t hctx,
+                  std::set<entity_inst_t> *entities) {
+  obj_list_watch_response_t watchers;
+  int r = cls_cxx_list_watchers(hctx, &watchers);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error listing watchers: '%s'", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  entities->clear();
+  for (auto &w : watchers.entries) {
+    entity_inst_t entity_inst{w.name, w.addr};
+    sanitize_entity_inst(&entity_inst);
+
+    entities->insert(entity_inst);
+  }
+  return 0;
+}
+
+int read_peers(cls_method_context_t hctx,
+               std::vector<cls::rbd::MirrorPeer> *peers) {
+  std::string last_read = PEER_KEY_PREFIX;
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+  while (more) {
+    std::map<std::string, bufferlist> vals;
+    int r = cls_cxx_map_get_vals(hctx, last_read, PEER_KEY_PREFIX.c_str(),
+                                 max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading peers: %s", cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto &it : vals) {
+      try {
+        auto bl_it = it.second.cbegin();
+        cls::rbd::MirrorPeer peer;
+	decode(peer, bl_it);
+        peers->push_back(peer);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode peer '%s'", it.first.c_str());
+	return -EIO;
+      }
+    }
+
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  }
+  return 0;
+}
+
+int read_peer(cls_method_context_t hctx, const std::string &id,
+              cls::rbd::MirrorPeer *peer) {
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, peer_key(id), &bl);
+  if (r < 0) {
+    CLS_ERR("error reading peer '%s': %s", id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  try {
+    auto bl_it = bl.cbegin();
+    decode(*peer, bl_it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("could not decode peer '%s'", id.c_str());
+    return -EIO;
+  }
+  return 0;
+}
+
+int write_peer(cls_method_context_t hctx, const std::string &id,
+               const cls::rbd::MirrorPeer &peer) {
+  bufferlist bl;
+  encode(peer, bl);
+
+  int r = cls_cxx_map_set_val(hctx, peer_key(id), &bl);
+  if (r < 0) {
+    CLS_ERR("error writing peer '%s': %s", id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+int image_get(cls_method_context_t hctx, const string &image_id,
+	      cls::rbd::MirrorImage *mirror_image) {
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, image_key(image_id), &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading mirrored image '%s': '%s'", image_id.c_str(),
+	      cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  try {
+    auto it = bl.cbegin();
+    decode(*mirror_image, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("could not decode mirrored image '%s'", image_id.c_str());
+    return -EIO;
+  }
+
+  return 0;
+}
+
+int image_set(cls_method_context_t hctx, const string &image_id,
+	      const cls::rbd::MirrorImage &mirror_image) {
+  bufferlist bl;
+  encode(mirror_image, bl);
+
+  cls::rbd::MirrorImage existing_mirror_image;
+  int r = image_get(hctx, image_id, &existing_mirror_image);
+  if (r == -ENOENT) {
+    // make sure global id doesn't already exist
+    std::string global_id_key = global_key(mirror_image.global_image_id);
+    std::string image_id;
+    r = read_key(hctx, global_id_key, &image_id);
+    if (r >= 0) {
+      return -EEXIST;
+    } else if (r != -ENOENT) {
+      CLS_ERR("error reading global image id: '%s': '%s'", image_id.c_str(),
+              cpp_strerror(r).c_str());
+      return r;
+    }
+
+    // make sure this was not a race for disabling
+    if (mirror_image.state == cls::rbd::MIRROR_IMAGE_STATE_DISABLING) {
+      CLS_ERR("image '%s' is already disabled", image_id.c_str());
+      return r;
+    }
+  } else if (r < 0) {
+    CLS_ERR("error reading mirrored image '%s': '%s'", image_id.c_str(),
+	    cpp_strerror(r).c_str());
+    return r;
+  } else if (existing_mirror_image.global_image_id !=
+                mirror_image.global_image_id) {
+    // cannot change the global id
+    return -EINVAL;
+  }
+
+  r = cls_cxx_map_set_val(hctx, image_key(image_id), &bl);
+  if (r < 0) {
+    CLS_ERR("error adding mirrored image '%s': %s", image_id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  bufferlist image_id_bl;
+  encode(image_id, image_id_bl);
+  r = cls_cxx_map_set_val(hctx, global_key(mirror_image.global_image_id),
+                          &image_id_bl);
+  if (r < 0) {
+    CLS_ERR("error adding global id for image '%s': %s", image_id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+int image_remove(cls_method_context_t hctx, const string &image_id) {
+  bufferlist bl;
+  cls::rbd::MirrorImage mirror_image;
+  int r = image_get(hctx, image_id, &mirror_image);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading mirrored image '%s': '%s'", image_id.c_str(),
+	      cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  if (mirror_image.state != cls::rbd::MIRROR_IMAGE_STATE_DISABLING) {
+    return -EBUSY;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, image_key(image_id));
+  if (r < 0) {
+    CLS_ERR("error removing mirrored image '%s': %s", image_id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, global_key(mirror_image.global_image_id));
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error removing global id for image '%s': %s", image_id.c_str(),
+           cpp_strerror(r).c_str());
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx,
+                             status_global_key(mirror_image.global_image_id));
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error removing global status for image '%s': %s", image_id.c_str(),
+           cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+struct MirrorImageStatusOnDisk : cls::rbd::MirrorImageStatus {
+  entity_inst_t origin;
+
+  MirrorImageStatusOnDisk() {
+  }
+  MirrorImageStatusOnDisk(const cls::rbd::MirrorImageStatus &status) :
+    cls::rbd::MirrorImageStatus(status) {
+  }
+
+  void encode_meta(bufferlist &bl, uint64_t features) const {
+    ENCODE_START(1, 1, bl);
+    auto sanitized_origin = origin;
+    sanitize_entity_inst(&sanitized_origin);
+    encode(sanitized_origin, bl, features);
+    ENCODE_FINISH(bl);
+  }
+
+  void encode(bufferlist &bl, uint64_t features) const {
+    encode_meta(bl, features);
+    cls::rbd::MirrorImageStatus::encode(bl);
+  }
+
+  void decode_meta(bufferlist::const_iterator &it) {
+    DECODE_START(1, it);
+    decode(origin, it);
+    sanitize_entity_inst(&origin);
+    DECODE_FINISH(it);
+  }
+
+  void decode(bufferlist::const_iterator &it) {
+    decode_meta(it);
+    cls::rbd::MirrorImageStatus::decode(it);
+  }
+};
+WRITE_CLASS_ENCODER_FEATURES(MirrorImageStatusOnDisk)
+
+int image_status_set(cls_method_context_t hctx, const string &global_image_id,
+		     const cls::rbd::MirrorImageStatus &status) {
+  MirrorImageStatusOnDisk ondisk_status(status);
+  ondisk_status.up = false;
+  ondisk_status.last_update = ceph_clock_now();
+
+  int r = cls_get_request_origin(hctx, &ondisk_status.origin);
+  ceph_assert(r == 0);
+
+  bufferlist bl;
+  encode(ondisk_status, bl, cls_get_features(hctx));
+
+  r = cls_cxx_map_set_val(hctx, status_global_key(global_image_id), &bl);
+  if (r < 0) {
+    CLS_ERR("error setting status for mirrored image, global id '%s': %s",
+	    global_image_id.c_str(), cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+int image_status_remove(cls_method_context_t hctx,
+			const string &global_image_id) {
+
+  int r = cls_cxx_map_remove_key(hctx, status_global_key(global_image_id));
+  if (r < 0) {
+    CLS_ERR("error removing status for mirrored image, global id '%s': %s",
+	    global_image_id.c_str(), cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+int image_status_get(cls_method_context_t hctx, const string &global_image_id,
+                     const std::set<entity_inst_t> &watchers,
+		     cls::rbd::MirrorImageStatus *status) {
+
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, status_global_key(global_image_id), &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading status for mirrored image, global id '%s': '%s'",
+	      global_image_id.c_str(), cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  MirrorImageStatusOnDisk ondisk_status;
+  try {
+    auto it = bl.cbegin();
+    decode(ondisk_status, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("could not decode status for mirrored image, global id '%s'",
+	    global_image_id.c_str());
+    return -EIO;
+  }
+
+
+  *status = static_cast<cls::rbd::MirrorImageStatus>(ondisk_status);
+  status->up = (watchers.find(ondisk_status.origin) != watchers.end());
+  return 0;
+}
+
+int image_status_list(cls_method_context_t hctx,
+	const std::string &start_after, uint64_t max_return,
+	map<std::string, cls::rbd::MirrorImage> *mirror_images,
+        map<std::string, cls::rbd::MirrorImageStatus> *mirror_statuses) {
+  std::string last_read = image_key(start_after);
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+
+  std::set<entity_inst_t> watchers;
+  int r = list_watchers(hctx, &watchers);
+  if (r < 0) {
+    return r;
+  }
+
+  while (more && mirror_images->size() < max_return) {
+    std::map<std::string, bufferlist> vals;
+    CLS_LOG(20, "last_read = '%s'", last_read.c_str());
+    r = cls_cxx_map_get_vals(hctx, last_read, IMAGE_KEY_PREFIX, max_read, &vals,
+                             &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading mirror image directory by name: %s",
+                cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto it = vals.begin(); it != vals.end() &&
+	   mirror_images->size() < max_return; ++it) {
+      const std::string &image_id = it->first.substr(IMAGE_KEY_PREFIX.size());
+      cls::rbd::MirrorImage mirror_image;
+      auto iter = it->second.cbegin();
+      try {
+	decode(mirror_image, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode mirror image payload of image '%s'",
+                image_id.c_str());
+	return -EIO;
+      }
+
+      (*mirror_images)[image_id] = mirror_image;
+
+      cls::rbd::MirrorImageStatus status;
+      int r1 = image_status_get(hctx, mirror_image.global_image_id, watchers,
+                                &status);
+      if (r1 < 0) {
+	continue;
+      }
+
+      (*mirror_statuses)[image_id] = status;
+    }
+    if (!vals.empty()) {
+      last_read = image_key(mirror_images->rbegin()->first);
+    }
+  }
+
+  return 0;
+}
+
+int image_status_get_summary(
+    cls_method_context_t hctx,
+    std::map<cls::rbd::MirrorImageStatusState, int> *states) {
+  std::set<entity_inst_t> watchers;
+  int r = list_watchers(hctx, &watchers);
+  if (r < 0) {
+    return r;
+  }
+
+  states->clear();
+
+  string last_read = IMAGE_KEY_PREFIX;
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+  while (more) {
+    map<string, bufferlist> vals;
+    r = cls_cxx_map_get_vals(hctx, last_read, IMAGE_KEY_PREFIX,
+			     max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading mirrored images: %s", cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto &list_it : vals) {
+      const string &key = list_it.first;
+
+      if (0 != key.compare(0, IMAGE_KEY_PREFIX.size(), IMAGE_KEY_PREFIX)) {
+	break;
+      }
+
+      cls::rbd::MirrorImage mirror_image;
+      auto iter = list_it.second.cbegin();
+      try {
+	decode(mirror_image, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode mirror image payload for key '%s'",
+                key.c_str());
+	return -EIO;
+      }
+
+      cls::rbd::MirrorImageStatus status;
+      image_status_get(hctx, mirror_image.global_image_id, watchers, &status);
+
+      cls::rbd::MirrorImageStatusState state = status.up ? status.state :
+	cls::rbd::MIRROR_IMAGE_STATUS_STATE_UNKNOWN;
+      (*states)[state]++;
+    }
+
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  }
+
+  return 0;
+}
+
+int image_status_remove_down(cls_method_context_t hctx) {
+  std::set<entity_inst_t> watchers;
+  int r = list_watchers(hctx, &watchers);
+  if (r < 0) {
+    return r;
+  }
+
+  string last_read = STATUS_GLOBAL_KEY_PREFIX;
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+  while (more) {
+    map<string, bufferlist> vals;
+    r = cls_cxx_map_get_vals(hctx, last_read, STATUS_GLOBAL_KEY_PREFIX,
+			     max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading mirrored images: %s", cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto &list_it : vals) {
+      const string &key = list_it.first;
+
+      if (0 != key.compare(0, STATUS_GLOBAL_KEY_PREFIX.size(),
+			   STATUS_GLOBAL_KEY_PREFIX)) {
+	break;
+      }
+
+      MirrorImageStatusOnDisk status;
+      try {
+	auto it = list_it.second.cbegin();
+	status.decode_meta(it);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode status metadata for mirrored image '%s'",
+		key.c_str());
+	return -EIO;
+      }
+
+      if (watchers.find(status.origin) == watchers.end()) {
+	CLS_LOG(20, "removing stale status object for key %s",
+		key.c_str());
+	int r1 = cls_cxx_map_remove_key(hctx, key);
+	if (r1 < 0) {
+	  CLS_ERR("error removing stale status for key '%s': %s",
+		  key.c_str(), cpp_strerror(r1).c_str());
+	  return r1;
+	}
+      }
+    }
+
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  }
+
+  return 0;
+}
+
+int image_instance_get(cls_method_context_t hctx,
+                       const string &global_image_id,
+                       const std::set<entity_inst_t> &watchers,
+                       entity_inst_t *instance) {
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, status_global_key(global_image_id), &bl);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading status for mirrored image, global id '%s': '%s'",
+              global_image_id.c_str(), cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  MirrorImageStatusOnDisk ondisk_status;
+  try {
+    auto it = bl.cbegin();
+    decode(ondisk_status, it);
+  } catch (const buffer::error &err) {
+    CLS_ERR("could not decode status for mirrored image, global id '%s'",
+            global_image_id.c_str());
+    return -EIO;
+  }
+
+  if (watchers.find(ondisk_status.origin) == watchers.end()) {
+    return -ESTALE;
+  }
+
+  *instance = ondisk_status.origin;
+  return 0;
+}
+
+int image_instance_list(cls_method_context_t hctx,
+                        const std::string &start_after,
+                        uint64_t max_return,
+                        map<std::string, entity_inst_t> *instances) {
+  std::string last_read = image_key(start_after);
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+
+  std::set<entity_inst_t> watchers;
+  int r = list_watchers(hctx, &watchers);
+  if (r < 0) {
+    return r;
+  }
+
+  while (more && instances->size() < max_return) {
+    std::map<std::string, bufferlist> vals;
+    CLS_LOG(20, "last_read = '%s'", last_read.c_str());
+    r = cls_cxx_map_get_vals(hctx, last_read, IMAGE_KEY_PREFIX, max_read, &vals,
+                             &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading mirror image directory by name: %s",
+                cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto it = vals.begin(); it != vals.end() &&
+           instances->size() < max_return; ++it) {
+      const std::string &image_id = it->first.substr(IMAGE_KEY_PREFIX.size());
+      cls::rbd::MirrorImage mirror_image;
+      auto iter = it->second.cbegin();
+      try {
+        decode(mirror_image, iter);
+      } catch (const buffer::error &err) {
+        CLS_ERR("could not decode mirror image payload of image '%s'",
+                image_id.c_str());
+        return -EIO;
+      }
+
+      entity_inst_t instance;
+      r = image_instance_get(hctx, mirror_image.global_image_id, watchers,
+                             &instance);
+      if (r < 0) {
+        continue;
+      }
+
+      (*instances)[image_id] = instance;
+    }
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  }
+
+  return 0;
+}
+
+int instances_list(cls_method_context_t hctx,
+                   std::vector<std::string> *instance_ids) {
+  std::string last_read = INSTANCE_KEY_PREFIX;
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+  while (more) {
+    std::map<std::string, bufferlist> vals;
+    int r = cls_cxx_map_get_vals(hctx, last_read, INSTANCE_KEY_PREFIX.c_str(),
+                                 max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+	CLS_ERR("error reading mirror instances: %s", cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto &it : vals) {
+      instance_ids->push_back(it.first.substr(INSTANCE_KEY_PREFIX.size()));
+    }
+
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  }
+  return 0;
+}
+
+int instances_add(cls_method_context_t hctx, const string &instance_id) {
+  bufferlist bl;
+
+  int r = cls_cxx_map_set_val(hctx, instance_key(instance_id), &bl);
+  if (r < 0) {
+    CLS_ERR("error setting mirror instance %s: %s", instance_id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+int instances_remove(cls_method_context_t hctx, const string &instance_id) {
+
+  int r = cls_cxx_map_remove_key(hctx, instance_key(instance_id));
+  if (r < 0) {
+    CLS_ERR("error removing mirror instance %s: %s", instance_id.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+int mirror_image_map_list(cls_method_context_t hctx,
+                          const std::string &start_after,
+                          uint64_t max_return,
+                          std::map<std::string, cls::rbd::MirrorImageMap> *image_mapping) {
+  bool more = true;
+  std::string last_read = mirror_image_map_key(start_after);
+
+  while (more && image_mapping->size() < max_return) {
+    std::map<std::string, bufferlist> vals;
+    CLS_LOG(20, "last read: '%s'", last_read.c_str());
+
+    int max_read = std::min<uint64_t>(RBD_MAX_KEYS_READ, max_return - image_mapping->size());
+    int r = cls_cxx_map_get_vals(hctx, last_read, MIRROR_IMAGE_MAP_KEY_PREFIX,
+                                 max_read, &vals, &more);
+    if (r < 0) {
+      CLS_ERR("error reading image map: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+
+    if (vals.empty()) {
+      return 0;
+    }
+
+    for (auto it = vals.begin(); it != vals.end(); ++it) {
+      const std::string &global_image_id =
+        it->first.substr(MIRROR_IMAGE_MAP_KEY_PREFIX.size());
+
+      cls::rbd::MirrorImageMap mirror_image_map;
+      auto iter = it->second.cbegin();
+      try {
+        decode(mirror_image_map, iter);
+      } catch (const buffer::error &err) {
+        CLS_ERR("could not decode image map payload: %s",
+                cpp_strerror(r).c_str());
+        return -EINVAL;
+      }
+
+      image_mapping->insert(std::make_pair(global_image_id, mirror_image_map));
+    }
+
+    if (!vals.empty()) {
+      last_read = vals.rbegin()->first;
+    }
+  }
+
+  return 0;
+}
+
+} // namespace mirror
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param uuid (std::string)
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_uuid_get(cls_method_context_t hctx, bufferlist *in,
+                    bufferlist *out) {
+  std::string mirror_uuid;
+  int r = mirror::uuid_get(hctx, &mirror_uuid);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(mirror_uuid, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param mirror_uuid (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_uuid_set(cls_method_context_t hctx, bufferlist *in,
+                    bufferlist *out) {
+  std::string mirror_uuid;
+  try {
+    auto bl_it = in->cbegin();
+    decode(mirror_uuid, bl_it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (mirror_uuid.empty()) {
+    CLS_ERR("cannot set empty mirror uuid");
+    return -EINVAL;
+  }
+
+  uint32_t mirror_mode;
+  int r = read_key(hctx, mirror::MODE, &mirror_mode);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  } else if (r == 0 && mirror_mode != cls::rbd::MIRROR_MODE_DISABLED) {
+    CLS_ERR("cannot set mirror uuid while mirroring enabled");
+    return -EINVAL;
+  }
+
+  bufferlist mirror_uuid_bl;
+  mirror_uuid_bl.append(mirror_uuid);
+  r = cls_cxx_map_set_val(hctx, mirror::UUID, &mirror_uuid_bl);
+  if (r < 0) {
+    CLS_ERR("failed to set mirror uuid");
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param cls::rbd::MirrorMode (uint32_t)
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_mode_get(cls_method_context_t hctx, bufferlist *in,
+                    bufferlist *out) {
+  uint32_t mirror_mode_decode;
+  int r = read_key(hctx, mirror::MODE, &mirror_mode_decode);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(mirror_mode_decode, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param mirror_mode (cls::rbd::MirrorMode) (uint32_t)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_mode_set(cls_method_context_t hctx, bufferlist *in,
+                    bufferlist *out) {
+  uint32_t mirror_mode_decode;
+  try {
+    auto bl_it = in->cbegin();
+    decode(mirror_mode_decode, bl_it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  bool enabled;
+  switch (static_cast<cls::rbd::MirrorMode>(mirror_mode_decode)) {
+  case cls::rbd::MIRROR_MODE_DISABLED:
+    enabled = false;
+    break;
+  case cls::rbd::MIRROR_MODE_IMAGE:
+  case cls::rbd::MIRROR_MODE_POOL:
+    enabled = true;
+    break;
+  default:
+    CLS_ERR("invalid mirror mode: %d", mirror_mode_decode);
+    return -EINVAL;
+  }
+
+  int r;
+  if (enabled) {
+    std::string mirror_uuid;
+    r = mirror::uuid_get(hctx, &mirror_uuid);
+    if (r == -ENOENT) {
+      return -EINVAL;
+    } else if (r < 0) {
+      return r;
+    }
+
+    bufferlist bl;
+    encode(mirror_mode_decode, bl);
+
+    r = cls_cxx_map_set_val(hctx, mirror::MODE, &bl);
+    if (r < 0) {
+      CLS_ERR("error enabling mirroring: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+  } else {
+    std::vector<cls::rbd::MirrorPeer> peers;
+    r = mirror::read_peers(hctx, &peers);
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    }
+
+    if (!peers.empty()) {
+      CLS_ERR("mirroring peers still registered");
+      return -EBUSY;
+    }
+
+    r = remove_key(hctx, mirror::MODE);
+    if (r < 0) {
+      return r;
+    }
+
+    r = remove_key(hctx, mirror::UUID);
+    if (r < 0) {
+      return r;
+    }
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param std::vector<cls::rbd::MirrorPeer>: collection of peers
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_peer_list(cls_method_context_t hctx, bufferlist *in,
+                     bufferlist *out) {
+  std::vector<cls::rbd::MirrorPeer> peers;
+  int r = mirror::read_peers(hctx, &peers);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  encode(peers, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param mirror_peer (cls::rbd::MirrorPeer)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_peer_add(cls_method_context_t hctx, bufferlist *in,
+                    bufferlist *out) {
+  cls::rbd::MirrorPeer mirror_peer;
+  try {
+    auto it = in->cbegin();
+    decode(mirror_peer, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  uint32_t mirror_mode_decode;
+  int r = read_key(hctx, mirror::MODE, &mirror_mode_decode);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  } else if (r == -ENOENT ||
+             mirror_mode_decode == cls::rbd::MIRROR_MODE_DISABLED) {
+    CLS_ERR("mirroring must be enabled on the pool");
+    return -EINVAL;
+  } else if (!mirror_peer.is_valid()) {
+    CLS_ERR("mirror peer is not valid");
+    return -EINVAL;
+  }
+
+  std::string mirror_uuid;
+  r = mirror::uuid_get(hctx, &mirror_uuid);
+  if (r < 0) {
+    CLS_ERR("error retrieving mirroring uuid: %s", cpp_strerror(r).c_str());
+    return r;
+  } else if (mirror_peer.uuid == mirror_uuid) {
+    CLS_ERR("peer uuid '%s' matches pool mirroring uuid",
+            mirror_uuid.c_str());
+    return -EINVAL;
+  }
+
+  std::vector<cls::rbd::MirrorPeer> peers;
+  r = mirror::read_peers(hctx, &peers);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  for (auto const &peer : peers) {
+    if (peer.uuid == mirror_peer.uuid) {
+      CLS_ERR("peer uuid '%s' already exists",
+              peer.uuid.c_str());
+      return -ESTALE;
+    } else if (peer.cluster_name == mirror_peer.cluster_name &&
+               (peer.pool_id == -1 || mirror_peer.pool_id == -1 ||
+                peer.pool_id == mirror_peer.pool_id)) {
+      CLS_ERR("peer cluster name '%s' already exists",
+              peer.cluster_name.c_str());
+      return -EEXIST;
+    }
+  }
+
+  bufferlist bl;
+  encode(mirror_peer, bl);
+  r = cls_cxx_map_set_val(hctx, mirror::peer_key(mirror_peer.uuid),
+                          &bl);
+  if (r < 0) {
+    CLS_ERR("error adding peer: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param uuid (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_peer_remove(cls_method_context_t hctx, bufferlist *in,
+                       bufferlist *out) {
+  std::string uuid;
+  try {
+    auto it = in->cbegin();
+    decode(uuid, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = cls_cxx_map_remove_key(hctx, mirror::peer_key(uuid));
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error removing peer: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param uuid (std::string)
+ * @param client_name (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_peer_set_client(cls_method_context_t hctx, bufferlist *in,
+                           bufferlist *out) {
+  std::string uuid;
+  std::string client_name;
+  try {
+    auto it = in->cbegin();
+    decode(uuid, it);
+    decode(client_name, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  cls::rbd::MirrorPeer peer;
+  int r = mirror::read_peer(hctx, uuid, &peer);
+  if (r < 0) {
+    return r;
+  }
+
+  peer.client_name = client_name;
+  r = mirror::write_peer(hctx, uuid, peer);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param uuid (std::string)
+ * @param cluster_name (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_peer_set_cluster(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  std::string uuid;
+  std::string cluster_name;
+  try {
+    auto it = in->cbegin();
+    decode(uuid, it);
+    decode(cluster_name, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  cls::rbd::MirrorPeer peer;
+  int r = mirror::read_peer(hctx, uuid, &peer);
+  if (r < 0) {
+    return r;
+  }
+
+  peer.cluster_name = cluster_name;
+  r = mirror::write_peer(hctx, uuid, peer);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param std::map<std::string, std::string>: local id to global id map
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_list(cls_method_context_t hctx, bufferlist *in,
+		     bufferlist *out) {
+  std::string start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+  std::map<std::string, std::string> mirror_images;
+  std::string last_read = mirror::image_key(start_after);
+
+  while (more && mirror_images.size() < max_return) {
+    std::map<std::string, bufferlist> vals;
+    CLS_LOG(20, "last_read = '%s'", last_read.c_str());
+    int r = cls_cxx_map_get_vals(hctx, last_read, mirror::IMAGE_KEY_PREFIX,
+                                 max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading mirror image directory by name: %s",
+                cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto it = vals.begin(); it != vals.end(); ++it) {
+      const std::string &image_id =
+        it->first.substr(mirror::IMAGE_KEY_PREFIX.size());
+      cls::rbd::MirrorImage mirror_image;
+      auto iter = it->second.cbegin();
+      try {
+	decode(mirror_image, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode mirror image payload of image '%s'",
+                image_id.c_str());
+	return -EIO;
+      }
+
+      mirror_images[image_id] = mirror_image.global_image_id;
+      if (mirror_images.size() >= max_return) {
+	break;
+      }
+    }
+    if (!vals.empty()) {
+      last_read = mirror::image_key(mirror_images.rbegin()->first);
+    }
+  }
+
+  encode(mirror_images, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_id (std::string)
+ *
+ * Output:
+ * @param std::string - image id
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_get_image_id(cls_method_context_t hctx, bufferlist *in,
+                              bufferlist *out) {
+  std::string global_id;
+  try {
+    auto it = in->cbegin();
+    decode(global_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::string image_id;
+  int r = read_key(hctx, mirror::global_key(global_id), &image_id);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error retrieving image id for global id '%s': %s",
+              global_id.c_str(), cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  encode(image_id, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param image_id (std::string)
+ *
+ * Output:
+ * @param cls::rbd::MirrorImage - metadata associated with the image_id
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_get(cls_method_context_t hctx, bufferlist *in,
+		     bufferlist *out) {
+  string image_id;
+  try {
+    auto it = in->cbegin();
+    decode(image_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  cls::rbd::MirrorImage mirror_image;
+  int r = mirror::image_get(hctx, image_id, &mirror_image);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(mirror_image, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param image_id (std::string)
+ * @param mirror_image (cls::rbd::MirrorImage)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ * @returns -EEXIST if there's an existing image_id with a different global_image_id
+ */
+int mirror_image_set(cls_method_context_t hctx, bufferlist *in,
+		     bufferlist *out) {
+  string image_id;
+  cls::rbd::MirrorImage mirror_image;
+  try {
+    auto it = in->cbegin();
+    decode(image_id, it);
+    decode(mirror_image, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = mirror::image_set(hctx, image_id, mirror_image);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param image_id (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_remove(cls_method_context_t hctx, bufferlist *in,
+			bufferlist *out) {
+  string image_id;
+  try {
+    auto it = in->cbegin();
+    decode(image_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = mirror::image_remove(hctx, image_id);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_image_id (std::string)
+ * @param status (cls::rbd::MirrorImageStatus)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_status_set(cls_method_context_t hctx, bufferlist *in,
+			    bufferlist *out) {
+  string global_image_id;
+  cls::rbd::MirrorImageStatus status;
+  try {
+    auto it = in->cbegin();
+    decode(global_image_id, it);
+    decode(status, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = mirror::image_status_set(hctx, global_image_id, status);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_image_id (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_status_remove(cls_method_context_t hctx, bufferlist *in,
+			       bufferlist *out) {
+  string global_image_id;
+  try {
+    auto it = in->cbegin();
+    decode(global_image_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = mirror::image_status_remove(hctx, global_image_id);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_image_id (std::string)
+ *
+ * Output:
+ * @param cls::rbd::MirrorImageStatus - metadata associated with the global_image_id
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_status_get(cls_method_context_t hctx, bufferlist *in,
+			    bufferlist *out) {
+  string global_image_id;
+  try {
+    auto it = in->cbegin();
+    decode(global_image_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::set<entity_inst_t> watchers;
+  int r = mirror::list_watchers(hctx, &watchers);
+  if (r < 0) {
+    return r;
+  }
+
+  cls::rbd::MirrorImageStatus status;
+  r = mirror::image_status_get(hctx, global_image_id, watchers, &status);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(status, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param std::map<std::string, cls::rbd::MirrorImage>: image id to image map
+ * @param std::map<std::string, cls::rbd::MirrorImageStatus>: image it to status map
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_status_list(cls_method_context_t hctx, bufferlist *in,
+			     bufferlist *out) {
+  std::string start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  map<std::string, cls::rbd::MirrorImage> images;
+  map<std::string, cls::rbd::MirrorImageStatus> statuses;
+  int r = mirror::image_status_list(hctx, start_after, max_return, &images,
+				    &statuses);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(images, *out);
+  encode(statuses, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param std::map<cls::rbd::MirrorImageStatusState, int>: states counts
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_status_get_summary(cls_method_context_t hctx, bufferlist *in,
+				    bufferlist *out) {
+  std::map<cls::rbd::MirrorImageStatusState, int> states;
+
+  int r = mirror::image_status_get_summary(hctx, &states);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(states, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_status_remove_down(cls_method_context_t hctx, bufferlist *in,
+				    bufferlist *out) {
+  int r = mirror::image_status_remove_down(hctx);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_image_id (std::string)
+ *
+ * Output:
+ * @param entity_inst_t - instance
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_instance_get(cls_method_context_t hctx, bufferlist *in,
+                              bufferlist *out) {
+  string global_image_id;
+  try {
+    auto it = in->cbegin();
+    decode(global_image_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::set<entity_inst_t> watchers;
+  int r = mirror::list_watchers(hctx, &watchers);
+  if (r < 0) {
+    return r;
+  }
+
+  entity_inst_t instance;
+  r = mirror::image_instance_get(hctx, global_image_id, watchers, &instance);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(instance, *out, cls_get_features(hctx));
+  return 0;
+}
+
+/**
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param std::map<std::string, entity_inst_t>: image id to instance map
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_instance_list(cls_method_context_t hctx, bufferlist *in,
+                               bufferlist *out) {
+  std::string start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  map<std::string, entity_inst_t> instances;
+  int r = mirror::image_instance_list(hctx, start_after, max_return,
+                                      &instances);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(instances, *out, cls_get_features(hctx));
+  return 0;
+}
+
+/**
+ * Input:
+ * none
+ *
+ * Output:
+ * @param std::vector<std::string>: instance ids
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_instances_list(cls_method_context_t hctx, bufferlist *in,
+                          bufferlist *out) {
+  std::vector<std::string> instance_ids;
+
+  int r = mirror::instances_list(hctx, &instance_ids);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(instance_ids, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param instance_id (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_instances_add(cls_method_context_t hctx, bufferlist *in,
+                         bufferlist *out) {
+  std::string instance_id;
+  try {
+    auto iter = in->cbegin();
+    decode(instance_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = mirror::instances_add(hctx, instance_id);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param instance_id (std::string)
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_instances_remove(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  std::string instance_id;
+  try {
+    auto iter = in->cbegin();
+    decode(instance_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = mirror::instances_remove(hctx, instance_id);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+/**
+ * Input:
+ * @param start_after: key to start after
+ * @param max_return: max return items
+ *
+ * Output:
+ * @param std::map<std::string, cls::rbd::MirrorImageMap>: image mapping
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_map_list(cls_method_context_t hctx, bufferlist *in,
+                          bufferlist *out) {
+  std::string start_after;
+  uint64_t max_return;
+  try {
+    auto it = in->cbegin();
+    decode(start_after, it);
+    decode(max_return, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::map<std::string, cls::rbd::MirrorImageMap> image_mapping;
+  int r = mirror::mirror_image_map_list(hctx, start_after, max_return, &image_mapping);
+  if (r < 0) {
+    return r;
+  }
+
+  encode(image_mapping, *out);
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_image_id: global image id
+ * @param image_map: image map
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_map_update(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  std::string global_image_id;
+  cls::rbd::MirrorImageMap image_map;
+
+  try {
+    auto it = in->cbegin();
+    decode(global_image_id, it);
+    decode(image_map, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  bufferlist bl;
+  encode(image_map, bl);
+
+  const std::string key = mirror::mirror_image_map_key(global_image_id);
+  int r = cls_cxx_map_set_val(hctx, key, &bl);
+  if (r < 0) {
+    CLS_ERR("error updating image map %s: %s", key.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Input:
+ * @param global_image_id: global image id
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int mirror_image_map_remove(cls_method_context_t hctx, bufferlist *in,
+                            bufferlist *out) {
+  std::string global_image_id;
+
+  try {
+    auto it = in->cbegin();
+    decode(global_image_id, it);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  const std::string key = mirror::mirror_image_map_key(global_image_id);
+  int r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error removing image map %s: %s", key.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+namespace group {
+
+/********************** methods for rbd_group_directory ***********************/
+
+int dir_add(cls_method_context_t hctx,
+            const string &name, const string &id,
+            bool check_for_unique_id)
+{
+  if (!name.size() || !is_valid_id(id)) {
+    CLS_ERR("invalid group name '%s' or id '%s'",
+            name.c_str(), id.c_str());
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "dir_add name=%s id=%s", name.c_str(), id.c_str());
+
+  string name_key = dir_key_for_name(name);
+  string id_key = dir_key_for_id(id);
+  string tmp;
+  int r = read_key(hctx, name_key, &tmp);
+  if (r != -ENOENT) {
+    CLS_LOG(10, "name already exists");
+    return -EEXIST;
+  }
+  r = read_key(hctx, id_key, &tmp);
+  if (r != -ENOENT && check_for_unique_id) {
+    CLS_LOG(10, "id already exists");
+    return -EBADF;
+  }
+  bufferlist id_bl, name_bl;
+  encode(id, id_bl);
+  encode(name, name_bl);
+  map<string, bufferlist> omap_vals;
+  omap_vals[name_key] = id_bl;
+  omap_vals[id_key] = name_bl;
+  return cls_cxx_map_set_vals(hctx, &omap_vals);
+}
+
+int dir_remove(cls_method_context_t hctx,
+               const string &name, const string &id)
+{
+  CLS_LOG(20, "dir_remove name=%s id=%s", name.c_str(), id.c_str());
+
+  string name_key = dir_key_for_name(name);
+  string id_key = dir_key_for_id(id);
+  string stored_name, stored_id;
+
+  int r = read_key(hctx, name_key, &stored_id);
+  if (r < 0) {
+    if (r != -ENOENT)
+      CLS_ERR("error reading name to id mapping: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+  r = read_key(hctx, id_key, &stored_name);
+  if (r < 0) {
+    if (r != -ENOENT)
+      CLS_ERR("error reading id to name mapping: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  // check if this op raced with a rename
+  if (stored_name != name || stored_id != id) {
+    CLS_ERR("stored name '%s' and id '%s' do not match args '%s' and '%s'",
+            stored_name.c_str(), stored_id.c_str(), name.c_str(), id.c_str());
+    return -ESTALE;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, name_key);
+  if (r < 0) {
+    CLS_ERR("error removing name: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, id_key);
+  if (r < 0) {
+    CLS_ERR("error removing id: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+static const string RBD_GROUP_SNAP_KEY_PREFIX = "snapshot_";
+
+std::string snap_key(const std::string &snap_id) {
+  ostringstream oss;
+  oss << RBD_GROUP_SNAP_KEY_PREFIX << snap_id;
+  return oss.str();
+}
+
+int snap_list(cls_method_context_t hctx, cls::rbd::GroupSnapshot start_after,
+              uint64_t max_return,
+              std::vector<cls::rbd::GroupSnapshot> *group_snaps)
+{
+  int max_read = RBD_MAX_KEYS_READ;
+  std::map<string, bufferlist> vals;
+  string last_read = snap_key(start_after.id);
+
+  group_snaps->clear();
+
+  bool more;
+  do {
+    int r = cls_cxx_map_get_vals(hctx, last_read,
+				 RBD_GROUP_SNAP_KEY_PREFIX,
+				 max_read, &vals, &more);
+    if (r < 0)
+      return r;
+
+    for (map<string, bufferlist>::iterator it = vals.begin();
+	 it != vals.end() && group_snaps->size() < max_return; ++it) {
+
+      auto iter = it->second.cbegin();
+      cls::rbd::GroupSnapshot snap;
+      try {
+	decode(snap, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("error decoding snapshot: %s", it->first.c_str());
+	return -EIO;
+      }
+      CLS_LOG(20, "Discovered snapshot %s %s",
+	      snap.name.c_str(),
+	      snap.id.c_str());
+      group_snaps->push_back(snap);
+    }
+
+  } while (more && (group_snaps->size() < max_return));
+
+  return 0;
+}
+
+static int check_duplicate_snap_name(cls_method_context_t hctx,
+				     const std::string &snap_name,
+				     const std::string &snap_id)
+{
+  const int max_read = 1024;
+  cls::rbd::GroupSnapshot snap_last;
+  std::vector<cls::rbd::GroupSnapshot> page;
+
+  for (;;) {
+    int r = snap_list(hctx, snap_last, max_read, &page);
+    if (r < 0) {
+      return r;
+    }
+    for (auto& snap: page) {
+      if (snap.name == snap_name && snap.id != snap_id) {
+	return -EEXIST;
+      }
+    }
+
+    if (page.size() < max_read) {
+      break;
+    }
+
+    snap_last = *page.rbegin();
+  }
+
+  return 0;
+}
+
+} // namespace group
+
+/**
+ * List groups from the directory.
+ *
+ * Input:
+ * @param start_after (std::string)
+ * @param max_return (int64_t)
+ *
+ * Output:
+ * @param map of groups (name, id)
+ * @return 0 on success, negative error code on failure
+ */
+int group_dir_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string start_after;
+  uint64_t max_return;
+
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int max_read = RBD_MAX_KEYS_READ;
+  bool more = true;
+  map<string, string> groups;
+  string last_read = dir_key_for_name(start_after);
+
+  while (more && groups.size() < max_return) {
+    map<string, bufferlist> vals;
+    CLS_LOG(20, "last_read = '%s'", last_read.c_str());
+    int r = cls_cxx_map_get_vals(hctx, last_read, RBD_DIR_NAME_KEY_PREFIX,
+                                 max_read, &vals, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("error reading directory by name: %s", cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (pair<string, bufferlist> val: vals) {
+      string id;
+      auto iter = val.second.cbegin();
+      try {
+	decode(id, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("could not decode id of group '%s'", val.first.c_str());
+	return -EIO;
+      }
+      CLS_LOG(20, "adding '%s' -> '%s'", dir_name_from_key(val.first).c_str(), id.c_str());
+      groups[dir_name_from_key(val.first)] = id;
+      if (groups.size() >= max_return)
+	break;
+    }
+    if (!vals.empty()) {
+      last_read = dir_key_for_name(groups.rbegin()->first);
+    }
+  }
+
+  encode(groups, *out);
+
+  return 0;
+}
+
+/**
+ * Add a group to the directory.
+ *
+ * Input:
+ * @param name (std::string)
+ * @param id (std::string)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_dir_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r = cls_cxx_create(hctx, false);
+
+  if (r < 0) {
+    CLS_ERR("could not create group directory: %s",
+	    cpp_strerror(r).c_str());
+    return r;
+  }
+
+  string name, id;
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  return group::dir_add(hctx, name, id, true);
+}
+
+/**
+ * Rename a group to the directory.
+ *
+ * Input:
+ * @param src original name of the group (std::string)
+ * @param dest new name of the group (std::string)
+ * @param id the id of the group (std::string)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_dir_rename(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string src, dest, id;
+  try {
+    auto iter = in->cbegin();
+    decode(src, iter);
+    decode(dest, iter);
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = group::dir_remove(hctx, src, id);
+  if (r < 0)
+    return r;
+
+  return group::dir_add(hctx, dest, id, false);
+}
+
+/**
+ * Remove a group from the directory.
+ *
+ * Input:
+ * @param name (std::string)
+ * @param id (std::string)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_dir_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string name, id;
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  return group::dir_remove(hctx, name, id);
+}
+
+/**
+ * Set state of an image in the group.
+ *
+ * Input:
+ * @param image_status (cls::rbd::GroupImageStatus)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_image_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_image_set");
+
+  cls::rbd::GroupImageStatus st;
+  try {
+    auto iter = in->cbegin();
+    decode(st, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  string image_key = st.spec.image_key();
+
+  bufferlist image_val_bl;
+  encode(st.state, image_val_bl);
+  int r = cls_cxx_map_set_val(hctx, image_key, &image_val_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Remove reference to an image from the group.
+ *
+ * Input:
+ * @param spec (cls::rbd::GroupImageSpec)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_image_remove(cls_method_context_t hctx,
+                       bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_image_remove");
+  cls::rbd::GroupImageSpec spec;
+  try {
+    auto iter = in->cbegin();
+    decode(spec, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  string image_key = spec.image_key();
+
+  int r = cls_cxx_map_remove_key(hctx, image_key);
+  if (r < 0) {
+    CLS_ERR("error removing image from group: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/*
+ * List images in the group.
+ *
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param tuples of descriptions of the images: image_id, pool_id, image reference state.
+ * @return 0 on success, negative error code on failure
+ */
+int group_image_list(cls_method_context_t hctx,
+                     bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_image_list");
+  cls::rbd::GroupImageSpec start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int max_read = RBD_MAX_KEYS_READ;
+  std::map<string, bufferlist> vals;
+  string last_read = start_after.image_key();
+  std::vector<cls::rbd::GroupImageStatus> res;
+  bool more;
+  do {
+    int r = cls_cxx_map_get_vals(hctx, last_read,
+				 cls::rbd::RBD_GROUP_IMAGE_KEY_PREFIX,
+				 max_read, &vals, &more);
+    if (r < 0)
+      return r;
+
+    for (map<string, bufferlist>::iterator it = vals.begin();
+	 it != vals.end() && res.size() < max_return; ++it) {
+
+      auto iter = it->second.cbegin();
+      cls::rbd::GroupImageLinkState state;
+      try {
+	decode(state, iter);
+      } catch (const buffer::error &err) {
+	CLS_ERR("error decoding state for image: %s", it->first.c_str());
+	return -EIO;
+      }
+      cls::rbd::GroupImageSpec spec;
+      int r = cls::rbd::GroupImageSpec::from_key(it->first, &spec);
+      if (r < 0)
+	return r;
+
+      CLS_LOG(20, "Discovered image %s %" PRId64 " %d", spec.image_id.c_str(),
+	                                         spec.pool_id,
+					         (int)state);
+      res.push_back(cls::rbd::GroupImageStatus(spec, state));
+    }
+    if (res.size() > 0) {
+      last_read = res.rbegin()->spec.image_key();
+    }
+
+  } while (more && (res.size() < max_return));
+  encode(res, *out);
+
+  return 0;
+}
+
+/**
+ * Reference the group this image belongs to.
+ *
+ * Input:
+ * @param group_id (std::string)
+ * @param pool_id (int64_t)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int image_group_add(cls_method_context_t hctx,
+		    bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "image_group_add");
+  cls::rbd::GroupSpec new_group;
+  try {
+    auto iter = in->cbegin();
+    decode(new_group, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  bufferlist existing_refbl;
+
+  int r = cls_cxx_map_get_val(hctx, RBD_GROUP_REF, &existing_refbl);
+  if (r == 0) {
+    // If we are trying to link this image to the same group then return
+    // success. If this image already belongs to another group then abort.
+    cls::rbd::GroupSpec old_group;
+    try {
+      auto iter = existing_refbl.cbegin();
+      decode(old_group, iter);
+    } catch (const buffer::error &err) {
+      return -EINVAL;
+    }
+
+    if ((old_group.group_id != new_group.group_id) ||
+        (old_group.pool_id != new_group.pool_id)) {
+      return -EEXIST;
+    } else {
+      return 0; // In this case the values are already correct
+    }
+  } else if (r < 0 && r != -ENOENT) {
+    // No entry means this image is not a member of any group.
+    return r;
+  }
+
+  r = image::set_op_features(hctx, RBD_OPERATION_FEATURE_GROUP,
+                             RBD_OPERATION_FEATURE_GROUP);
+  if (r < 0) {
+    return r;
+  }
+
+  bufferlist refbl;
+  encode(new_group, refbl);
+  r = cls_cxx_map_set_val(hctx, RBD_GROUP_REF, &refbl);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Remove image's pointer to the group.
+ *
+ * Input:
+ * @param cg_id (std::string)
+ * @param pool_id (int64_t)
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int image_group_remove(cls_method_context_t hctx,
+		       bufferlist *in,
+		       bufferlist *out)
+{
+  CLS_LOG(20, "image_group_remove");
+  cls::rbd::GroupSpec spec;
+  try {
+    auto iter = in->cbegin();
+    decode(spec, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  bufferlist refbl;
+  int r = cls_cxx_map_get_val(hctx, RBD_GROUP_REF, &refbl);
+  if (r < 0) {
+    return r;
+  }
+
+  cls::rbd::GroupSpec ref_spec;
+  auto iter = refbl.cbegin();
+  try {
+    decode(ref_spec, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (ref_spec.pool_id != spec.pool_id || ref_spec.group_id != spec.group_id) {
+    return -EBADF;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, RBD_GROUP_REF);
+  if (r < 0) {
+    return r;
+  }
+
+  r = image::set_op_features(hctx, 0, RBD_OPERATION_FEATURE_GROUP);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Retrieve the id and pool of the group this image belongs to.
+ *
+ * Input:
+ * none
+ *
+ * Output:
+ * @param GroupSpec
+ * @return 0 on success, negative error code on failure
+ */
+int image_group_get(cls_method_context_t hctx,
+		    bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "image_group_get");
+  bufferlist refbl;
+  int r = cls_cxx_map_get_val(hctx, RBD_GROUP_REF, &refbl);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  }
+
+  cls::rbd::GroupSpec spec;
+
+  if (r != -ENOENT) {
+    auto iter = refbl.cbegin();
+    try {
+      decode(spec, iter);
+    } catch (const buffer::error &err) {
+      return -EINVAL;
+    }
+  }
+
+  encode(spec, *out);
+  return 0;
+}
+
+/**
+ * Save initial snapshot record.
+ *
+ * Input:
+ * @param GroupSnapshot
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_snap_set(cls_method_context_t hctx,
+		   bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_snap_set");
+  cls::rbd::GroupSnapshot group_snap;
+  try {
+    auto iter = in->cbegin();
+    decode(group_snap, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (group_snap.name.empty()) {
+    CLS_ERR("group snapshot name is empty");
+    return -EINVAL;
+  }
+  if (group_snap.id.empty()) {
+    CLS_ERR("group snapshot id is empty");
+    return -EINVAL;
+  }
+
+  int r = group::check_duplicate_snap_name(hctx, group_snap.name,
+                                           group_snap.id);
+  if (r < 0) {
+    return r;
+  }
+
+  std::string key = group::snap_key(group_snap.id);
+  if (group_snap.state == cls::rbd::GROUP_SNAPSHOT_STATE_INCOMPLETE) {
+    bufferlist snap_bl;
+    r = cls_cxx_map_get_val(hctx, key, &snap_bl);
+    if (r < 0 && r != -ENOENT) {
+      return r;
+    } else if (r >= 0) {
+      return -EEXIST;
+    }
+  }
+
+  bufferlist obl;
+  encode(group_snap, obl);
+  r = cls_cxx_map_set_val(hctx, key, &obl);
+  return r;
+}
+
+/**
+ * Remove snapshot record.
+ *
+ * Input:
+ * @param id Snapshot id
+ *
+ * Output:
+ * @return 0 on success, negative error code on failure
+ */
+int group_snap_remove(cls_method_context_t hctx,
+		      bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_snap_remove");
+  std::string snap_id;
+  try {
+    auto iter = in->cbegin();
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::string snap_key = group::snap_key(snap_id);
+
+  CLS_LOG(20, "removing snapshot with key %s", snap_key.c_str());
+  int r = cls_cxx_map_remove_key(hctx, snap_key);
+  return r;
+}
+
+/**
+ * Get group's snapshot by id.
+ *
+ * Input:
+ * @param snapshot_id the id of the snapshot to look for.
+ *
+ * Output:
+ * @param GroupSnapshot the requested snapshot
+ * @return 0 on success, negative error code on failure
+ */
+int group_snap_get_by_id(cls_method_context_t hctx,
+			 bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_snap_get_by_id");
+
+  std::string snap_id;
+  try {
+    auto iter = in->cbegin();
+    decode(snap_id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  bufferlist snapbl;
+
+  int r = cls_cxx_map_get_val(hctx, group::snap_key(snap_id), &snapbl);
+  if (r < 0) {
+    return r;
+  }
+
+  cls::rbd::GroupSnapshot group_snap;
+  auto iter = snapbl.cbegin();
+  try {
+    decode(group_snap, iter);
+  } catch (const buffer::error &err) {
+    CLS_ERR("error decoding snapshot: %s", snap_id.c_str());
+    return -EIO;
+  }
+
+  encode(group_snap, *out);
+
+  return 0;
+}
+
+/**
+ * List group's snapshots.
+ *
+ * Input:
+ * @param start_after which name to begin listing after
+ * 	  (use the empty string to start at the beginning)
+ * @param max_return the maximum number of snapshots to list
+ *
+ * Output:
+ * @param list of snapshots
+ * @return 0 on success, negative error code on failure
+ */
+int group_snap_list(cls_method_context_t hctx,
+		    bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(20, "group_snap_list");
+
+  cls::rbd::GroupSnapshot start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+  std::vector<cls::rbd::GroupSnapshot> group_snaps;
+  group::snap_list(hctx, start_after, max_return, &group_snaps);
+
+  encode(group_snaps, *out);
+
+  return 0;
+}
+
+namespace trash {
+
+static const std::string IMAGE_KEY_PREFIX("id_");
+
+std::string image_key(const std::string &image_id) {
+  return IMAGE_KEY_PREFIX + image_id;
+}
+
+std::string image_id_from_key(const std::string &key) {
+  return key.substr(IMAGE_KEY_PREFIX.size());
+}
+
+} // namespace trash
+
+/**
+ * Add an image entry to the rbd trash. Creates the trash object if
+ * needed, and stores the trash spec information of the deleted image.
+ *
+ * Input:
+ * @param id the id of the image
+ * @param trash_spec the spec info of the deleted image
+ *
+ * Output:
+ * @returns -EEXIST if the image id is already in the trash
+ * @returns 0 on success, negative error code on failure
+ */
+int trash_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  int r = cls_cxx_create(hctx, false);
+  if (r < 0) {
+    CLS_ERR("could not create trash: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  string id;
+  cls::rbd::TrashImageSpec trash_spec;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+    decode(trash_spec, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  if (!is_valid_id(id)) {
+    CLS_ERR("trash_add: invalid id '%s'", id.c_str());
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "trash_add id=%s", id.c_str());
+
+  string key = trash::image_key(id);
+  cls::rbd::TrashImageSpec tmp;
+  r = read_key(hctx, key, &tmp);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("could not read key %s entry from trash: %s", key.c_str(),
+            cpp_strerror(r).c_str());
+    return r;
+  } else if (r == 0) {
+    CLS_LOG(10, "id already exists");
+    return -EEXIST;
+  }
+
+  map<string, bufferlist> omap_vals;
+  encode(trash_spec, omap_vals[key]);
+  return cls_cxx_map_set_vals(hctx, &omap_vals);
+}
+
+/**
+ * Removes an image entry from the rbd trash object.
+ * image.
+ *
+ * Input:
+ * @param id the id of the image
+ *
+ * Output:
+ * @returns -ENOENT if the image id does not exist in the trash
+ * @returns 0 on success, negative error code on failure
+ */
+int trash_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string id;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "trash_remove id=%s", id.c_str());
+
+  string key = trash::image_key(id);
+  bufferlist tmp;
+  int r = cls_cxx_map_get_val(hctx, key, &tmp);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("error reading entry key %s: %s", key.c_str(), cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0) {
+    CLS_ERR("error removing entry: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Returns the list of trash spec entries registered in the rbd_trash
+ * object.
+ *
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param data the map between image id and trash spec info
+ *
+ * @returns 0 on success, negative error code on failure
+ */
+int trash_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string start_after;
+  uint64_t max_return;
+
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  map<string, cls::rbd::TrashImageSpec> data;
+  string last_read = trash::image_key(start_after);
+  bool more = true;
+
+  CLS_LOG(20, "trash_get_images");
+  while (data.size() < max_return) {
+    map<string, bufferlist> raw_data;
+    int max_read = std::min<int32_t>(RBD_MAX_KEYS_READ,
+                                     max_return - data.size());
+    int r = cls_cxx_map_get_vals(hctx, last_read, trash::IMAGE_KEY_PREFIX,
+                                 max_read, &raw_data, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("failed to read the vals off of disk: %s",
+                cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+    if (raw_data.empty()) {
+      break;
+    }
+
+    map<string, bufferlist>::iterator it = raw_data.begin();
+    for (; it != raw_data.end(); ++it) {
+      decode(data[trash::image_id_from_key(it->first)], it->second);
+    }
+
+    if (!more) {
+      break;
+    }
+
+    last_read = raw_data.rbegin()->first;
+  }
+
+  encode(data, *out);
+  return 0;
+}
+
+/**
+ * Returns the trash spec entry of an image registered in the rbd_trash
+ * object.
+ *
+ * Input:
+ * @param id the id of the image
+ *
+ * Output:
+ * @param out the trash spec entry
+ *
+ * @returns 0 on success, negative error code on failure
+ */
+int trash_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string id;
+  try {
+    auto iter = in->cbegin();
+    decode(id, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "trash_get_image id=%s", id.c_str());
+
+
+  string key = trash::image_key(id);
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, key, out);
+  if (r < 0 && r != -ENOENT) {
+    CLS_ERR("error reading image from trash '%s': '%s'", id.c_str(),
+            cpp_strerror(r).c_str());
+  }
+  return r;
+}
+
+/**
+ * Set state of an image in the rbd_trash object.
+ *
+ * Input:
+ * @param id the id of the image
+ * @param trash_state the state of the image to be set
+ * @param expect_state the expected state of the image
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int trash_state_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string id;
+  cls::rbd::TrashImageState trash_state;
+  cls::rbd::TrashImageState expect_state;
+  try {
+    bufferlist::const_iterator iter = in->begin();
+    decode(id, iter);
+    decode(trash_state, iter);
+    decode(expect_state, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  CLS_LOG(20, "trash_state_set id=%s", id.c_str());
+
+  string key = trash::image_key(id);
+  cls::rbd::TrashImageSpec trash_spec;
+  int r = read_key(hctx, key, &trash_spec);
+  if (r < 0) {
+    if (r != -ENOENT) {
+      CLS_ERR("Could not read trash image spec off disk: %s",
+              cpp_strerror(r).c_str());
+    }
+    return r;
+  }
+
+  if (trash_spec.state == expect_state) {
+    trash_spec.state = trash_state;
+    r = write_key(hctx, key, trash_spec);
+    if (r < 0) {
+      CLS_ERR("error setting trash image state: %s", cpp_strerror(r).c_str());
+      return r;
+    }
+
+    return 0;
+  } else if (trash_spec.state == trash_state) {
+    return 0;
+  } else {
+    CLS_ERR("Current trash state: %d do not match expected: %d or set: %d",
+            trash_spec.state, expect_state, trash_state);
+    return -ESTALE;
+  }
+}
+
+namespace nspace {
+
+const std::string NAME_KEY_PREFIX("name_");
+
+std::string key_for_name(const std::string& name) {
+  return NAME_KEY_PREFIX + name;
+}
+
+std::string name_from_key(const std::string &key) {
+  return key.substr(NAME_KEY_PREFIX.size());
+}
+
+} // namespace nspace
+
+/**
+ * Add a namespace to the namespace directory.
+ *
+ * Input:
+ * @param name the name of the namespace
+ *
+ * Output:
+ * @returns -EEXIST if the namespace is already exists
+ * @returns 0 on success, negative error code on failure
+ */
+int namespace_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  std::string name;
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::string key(nspace::key_for_name(name));
+  bufferlist value;
+  int r = cls_cxx_map_get_val(hctx, key, &value);
+  if (r < 0 && r != -ENOENT) {
+    return r;
+  } else if (r == 0) {
+    return -EEXIST;
+  }
+
+  r = cls_cxx_map_set_val(hctx, key, &value);
+  if (r < 0) {
+    CLS_ERR("failed to set omap key: %s", key.c_str());
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Remove a namespace from the namespace directory.
+ *
+ * Input:
+ * @param name the name of the namespace
+ *
+ * Output:
+ * @returns 0 on success, negative error code on failure
+ */
+int namespace_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  std::string name;
+  try {
+    auto iter = in->cbegin();
+    decode(name, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::string key(nspace::key_for_name(name));
+  bufferlist bl;
+  int r = cls_cxx_map_get_val(hctx, key, &bl);
+  if (r < 0) {
+    return r;
+  }
+
+  r = cls_cxx_map_remove_key(hctx, key);
+  if (r < 0) {
+    return r;
+  }
+
+  return 0;
+}
+
+/**
+ * Returns the list of namespaces in the rbd_namespace object
+ *
+ * Input:
+ * @param start_after which name to begin listing after
+ *        (use the empty string to start at the beginning)
+ * @param max_return the maximum number of names to list
+ *
+ * Output:
+ * @param data list of namespace names
+ * @returns 0 on success, negative error code on failure
+ */
+int namespace_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string start_after;
+  uint64_t max_return;
+  try {
+    auto iter = in->cbegin();
+    decode(start_after, iter);
+    decode(max_return, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  std::list<std::string> data;
+  std::string last_read = nspace::key_for_name(start_after);
+  bool more = true;
+
+  CLS_LOG(20, "namespace_list");
+  while (data.size() < max_return) {
+    std::map<std::string, bufferlist> raw_data;
+    int max_read = std::min<int32_t>(RBD_MAX_KEYS_READ,
+                                     max_return - data.size());
+    int r = cls_cxx_map_get_vals(hctx, last_read, nspace::NAME_KEY_PREFIX,
+                                 max_read, &raw_data, &more);
+    if (r < 0) {
+      if (r != -ENOENT) {
+        CLS_ERR("failed to read the vals off of disk: %s",
+                cpp_strerror(r).c_str());
+      }
+      return r;
+    }
+
+    for (auto& it : raw_data) {
+      data.push_back(nspace::name_from_key(it.first));
+    }
+
+    if (raw_data.empty() || !more) {
+      break;
+    }
+
+    last_read = raw_data.rbegin()->first;
+  }
+
+  encode(data, *out);
+  return 0;
+}
+
+/**
+ *  Reclaim space for zeroed extents
+ *
+ * Input:
+ * @param sparse_size minimal zeroed block to sparse
+ * @param remove_empty boolean, true if the object should be removed if empty
+ *
+ * Output:
+ * @returns -ENOENT if the object does not exist or has been removed
+ * @returns 0 on success, negative error code on failure
+ */
+int sparsify(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  size_t sparse_size;
+  bool remove_empty;
+  try {
+    auto iter = in->cbegin();
+    decode(sparse_size, iter);
+    decode(remove_empty, iter);
+  } catch (const buffer::error &err) {
+    return -EINVAL;
+  }
+
+  int r = check_exists(hctx);
+  if (r < 0) {
+    return r;
+  }
+
+  bufferlist bl;
+  r = cls_cxx_read(hctx, 0, 0, &bl);
+  if (r < 0) {
+    CLS_ERR("failed to read data off of disk: %s", cpp_strerror(r).c_str());
+    return r;
+  }
+
+  if (bl.is_zero()) {
+    if (remove_empty) {
+      CLS_LOG(20, "remove");
+      r = cls_cxx_remove(hctx);
+      if (r < 0) {
+        CLS_ERR("remove failed: %s", cpp_strerror(r).c_str());
+        return r;
+      }
+    } else if (bl.length() > 0) {
+      CLS_LOG(20, "truncate");
+      bufferlist write_bl;
+      r = cls_cxx_replace(hctx, 0, 0, &write_bl);
+      if (r < 0) {
+        CLS_ERR("truncate failed: %s", cpp_strerror(r).c_str());
+        return r;
+      }
+    } else {
+      CLS_LOG(20, "skip empty");
+    }
+    return 0;
+  }
+
+  bl.rebuild(buffer::ptr_node::create(bl.length()));
+  size_t write_offset = 0;
+  size_t write_length = 0;
+  size_t offset = 0;
+  size_t length = bl.length();
+  const auto& ptr = bl.front();
+  bool replace = true;
+  while (offset < length) {
+    if (calc_sparse_extent(ptr, sparse_size, length, &write_offset,
+                           &write_length, &offset)) {
+      if (write_offset == 0 && write_length == length) {
+        CLS_LOG(20, "nothing to do");
+        return 0;
+      }
+      CLS_LOG(20, "write%s %" PRIu64 "~%" PRIu64, (replace ? "(replace)" : ""),
+              write_offset, write_length);
+      bufferlist write_bl;
+      write_bl.push_back(buffer::ptr_node::create(ptr, write_offset,
+                                                  write_length));
+      if (replace) {
+        r = cls_cxx_replace(hctx, write_offset, write_length, &write_bl);
+        replace = false;
+      } else {
+        r = cls_cxx_write(hctx, write_offset, write_length, &write_bl);
+      }
+      if (r < 0) {
+        CLS_ERR("write failed: %s", cpp_strerror(r).c_str());
+        return r;
+      }
+      write_offset = offset;
+      write_length = 0;
+    }
+  }
+
+  return 0;
+}
+
+CLS_INIT(rbd)
+{
+  CLS_LOG(20, "Loaded rbd class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_create;
+  cls_method_handle_t h_get_features;
+  cls_method_handle_t h_set_features;
+  cls_method_handle_t h_get_size;
+  cls_method_handle_t h_set_size;
+  cls_method_handle_t h_get_parent;
+  cls_method_handle_t h_set_parent;
+  cls_method_handle_t h_remove_parent;
+  cls_method_handle_t h_parent_get;
+  cls_method_handle_t h_parent_overlap_get;
+  cls_method_handle_t h_parent_attach;
+  cls_method_handle_t h_parent_detach;
+  cls_method_handle_t h_get_protection_status;
+  cls_method_handle_t h_set_protection_status;
+  cls_method_handle_t h_get_stripe_unit_count;
+  cls_method_handle_t h_set_stripe_unit_count;
+  cls_method_handle_t h_get_create_timestamp;
+  cls_method_handle_t h_get_access_timestamp;
+  cls_method_handle_t h_get_modify_timestamp;
+  cls_method_handle_t h_get_flags;
+  cls_method_handle_t h_set_flags;
+  cls_method_handle_t h_op_features_get;
+  cls_method_handle_t h_op_features_set;
+  cls_method_handle_t h_add_child;
+  cls_method_handle_t h_remove_child;
+  cls_method_handle_t h_get_children;
+  cls_method_handle_t h_get_snapcontext;
+  cls_method_handle_t h_get_object_prefix;
+  cls_method_handle_t h_get_data_pool;
+  cls_method_handle_t h_get_snapshot_name;
+  cls_method_handle_t h_get_snapshot_timestamp;
+  cls_method_handle_t h_snapshot_get;
+  cls_method_handle_t h_snapshot_add;
+  cls_method_handle_t h_snapshot_remove;
+  cls_method_handle_t h_snapshot_rename;
+  cls_method_handle_t h_snapshot_trash_add;
+  cls_method_handle_t h_get_all_features;
+  cls_method_handle_t h_get_id;
+  cls_method_handle_t h_set_id;
+  cls_method_handle_t h_set_modify_timestamp;
+  cls_method_handle_t h_set_access_timestamp;
+  cls_method_handle_t h_dir_get_id;
+  cls_method_handle_t h_dir_get_name;
+  cls_method_handle_t h_dir_list;
+  cls_method_handle_t h_dir_add_image;
+  cls_method_handle_t h_dir_remove_image;
+  cls_method_handle_t h_dir_rename_image;
+  cls_method_handle_t h_dir_state_assert;
+  cls_method_handle_t h_dir_state_set;
+  cls_method_handle_t h_object_map_load;
+  cls_method_handle_t h_object_map_save;
+  cls_method_handle_t h_object_map_resize;
+  cls_method_handle_t h_object_map_update;
+  cls_method_handle_t h_object_map_snap_add;
+  cls_method_handle_t h_object_map_snap_remove;
+  cls_method_handle_t h_metadata_set;
+  cls_method_handle_t h_metadata_remove;
+  cls_method_handle_t h_metadata_list;
+  cls_method_handle_t h_metadata_get;
+  cls_method_handle_t h_snapshot_get_limit;
+  cls_method_handle_t h_snapshot_set_limit;
+  cls_method_handle_t h_child_attach;
+  cls_method_handle_t h_child_detach;
+  cls_method_handle_t h_children_list;
+  cls_method_handle_t h_migration_set;
+  cls_method_handle_t h_migration_set_state;
+  cls_method_handle_t h_migration_get;
+  cls_method_handle_t h_migration_remove;
+  cls_method_handle_t h_old_snapshots_list;
+  cls_method_handle_t h_old_snapshot_add;
+  cls_method_handle_t h_old_snapshot_remove;
+  cls_method_handle_t h_old_snapshot_rename;
+  cls_method_handle_t h_mirror_uuid_get;
+  cls_method_handle_t h_mirror_uuid_set;
+  cls_method_handle_t h_mirror_mode_get;
+  cls_method_handle_t h_mirror_mode_set;
+  cls_method_handle_t h_mirror_peer_list;
+  cls_method_handle_t h_mirror_peer_add;
+  cls_method_handle_t h_mirror_peer_remove;
+  cls_method_handle_t h_mirror_peer_set_client;
+  cls_method_handle_t h_mirror_peer_set_cluster;
+  cls_method_handle_t h_mirror_image_list;
+  cls_method_handle_t h_mirror_image_get_image_id;
+  cls_method_handle_t h_mirror_image_get;
+  cls_method_handle_t h_mirror_image_set;
+  cls_method_handle_t h_mirror_image_remove;
+  cls_method_handle_t h_mirror_image_status_set;
+  cls_method_handle_t h_mirror_image_status_remove;
+  cls_method_handle_t h_mirror_image_status_get;
+  cls_method_handle_t h_mirror_image_status_list;
+  cls_method_handle_t h_mirror_image_status_get_summary;
+  cls_method_handle_t h_mirror_image_status_remove_down;
+  cls_method_handle_t h_mirror_image_instance_get;
+  cls_method_handle_t h_mirror_image_instance_list;
+  cls_method_handle_t h_mirror_instances_list;
+  cls_method_handle_t h_mirror_instances_add;
+  cls_method_handle_t h_mirror_instances_remove;
+  cls_method_handle_t h_mirror_image_map_list;
+  cls_method_handle_t h_mirror_image_map_update;
+  cls_method_handle_t h_mirror_image_map_remove;
+  cls_method_handle_t h_group_dir_list;
+  cls_method_handle_t h_group_dir_add;
+  cls_method_handle_t h_group_dir_remove;
+  cls_method_handle_t h_group_dir_rename;
+  cls_method_handle_t h_group_image_remove;
+  cls_method_handle_t h_group_image_list;
+  cls_method_handle_t h_group_image_set;
+  cls_method_handle_t h_image_group_add;
+  cls_method_handle_t h_image_group_remove;
+  cls_method_handle_t h_image_group_get;
+  cls_method_handle_t h_group_snap_set;
+  cls_method_handle_t h_group_snap_remove;
+  cls_method_handle_t h_group_snap_get_by_id;
+  cls_method_handle_t h_group_snap_list;
+  cls_method_handle_t h_trash_add;
+  cls_method_handle_t h_trash_remove;
+  cls_method_handle_t h_trash_list;
+  cls_method_handle_t h_trash_get;
+  cls_method_handle_t h_trash_state_set;
+  cls_method_handle_t h_namespace_add;
+  cls_method_handle_t h_namespace_remove;
+  cls_method_handle_t h_namespace_list;
+  cls_method_handle_t h_copyup;
+  cls_method_handle_t h_assert_snapc_seq;
+  cls_method_handle_t h_sparsify;
+
+  cls_register("rbd", &h_class);
+  cls_register_cxx_method(h_class, "create",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  create, &h_create);
+  cls_register_cxx_method(h_class, "get_features",
+			  CLS_METHOD_RD,
+			  get_features, &h_get_features);
+  cls_register_cxx_method(h_class, "set_features",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  set_features, &h_set_features);
+  cls_register_cxx_method(h_class, "get_size",
+			  CLS_METHOD_RD,
+			  get_size, &h_get_size);
+  cls_register_cxx_method(h_class, "set_size",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  set_size, &h_set_size);
+  cls_register_cxx_method(h_class, "get_snapcontext",
+			  CLS_METHOD_RD,
+			  get_snapcontext, &h_get_snapcontext);
+  cls_register_cxx_method(h_class, "get_object_prefix",
+			  CLS_METHOD_RD,
+			  get_object_prefix, &h_get_object_prefix);
+  cls_register_cxx_method(h_class, "get_data_pool", CLS_METHOD_RD,
+                          get_data_pool, &h_get_data_pool);
+  cls_register_cxx_method(h_class, "get_snapshot_name",
+			  CLS_METHOD_RD,
+			  get_snapshot_name, &h_get_snapshot_name);
+  cls_register_cxx_method(h_class, "get_snapshot_timestamp",
+			  CLS_METHOD_RD,
+			  get_snapshot_timestamp, &h_get_snapshot_timestamp);
+  cls_register_cxx_method(h_class, "snapshot_get",
+                          CLS_METHOD_RD,
+                          snapshot_get, &h_snapshot_get);
+  cls_register_cxx_method(h_class, "snapshot_add",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  snapshot_add, &h_snapshot_add);
+  cls_register_cxx_method(h_class, "snapshot_remove",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  snapshot_remove, &h_snapshot_remove);
+  cls_register_cxx_method(h_class, "snapshot_rename",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  snapshot_rename, &h_snapshot_rename);
+  cls_register_cxx_method(h_class, "snapshot_trash_add",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          snapshot_trash_add, &h_snapshot_trash_add);
+  cls_register_cxx_method(h_class, "get_all_features",
+			  CLS_METHOD_RD,
+			  get_all_features, &h_get_all_features);
+
+  // NOTE: deprecate v1 parent APIs after mimic EOLed
+  cls_register_cxx_method(h_class, "get_parent",
+			  CLS_METHOD_RD,
+			  get_parent, &h_get_parent);
+  cls_register_cxx_method(h_class, "set_parent",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  set_parent, &h_set_parent);
+  cls_register_cxx_method(h_class, "remove_parent",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  remove_parent, &h_remove_parent);
+
+  cls_register_cxx_method(h_class, "parent_get",
+                          CLS_METHOD_RD, parent_get, &h_parent_get);
+  cls_register_cxx_method(h_class, "parent_overlap_get",
+                          CLS_METHOD_RD, parent_overlap_get,
+                          &h_parent_overlap_get);
+  cls_register_cxx_method(h_class, "parent_attach",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          parent_attach, &h_parent_attach);
+  cls_register_cxx_method(h_class, "parent_detach",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          parent_detach, &h_parent_detach);
+
+  cls_register_cxx_method(h_class, "set_protection_status",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  set_protection_status, &h_set_protection_status);
+  cls_register_cxx_method(h_class, "get_protection_status",
+			  CLS_METHOD_RD,
+			  get_protection_status, &h_get_protection_status);
+  cls_register_cxx_method(h_class, "get_stripe_unit_count",
+			  CLS_METHOD_RD,
+			  get_stripe_unit_count, &h_get_stripe_unit_count);
+  cls_register_cxx_method(h_class, "set_stripe_unit_count",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  set_stripe_unit_count, &h_set_stripe_unit_count);
+  cls_register_cxx_method(h_class, "get_create_timestamp",
+                          CLS_METHOD_RD,
+                          get_create_timestamp, &h_get_create_timestamp);
+  cls_register_cxx_method(h_class, "get_access_timestamp",
+                          CLS_METHOD_RD,
+                          get_access_timestamp, &h_get_access_timestamp);
+  cls_register_cxx_method(h_class, "get_modify_timestamp",
+                          CLS_METHOD_RD,
+                          get_modify_timestamp, &h_get_modify_timestamp);
+  cls_register_cxx_method(h_class, "get_flags",
+                          CLS_METHOD_RD,
+                          get_flags, &h_get_flags);
+  cls_register_cxx_method(h_class, "set_flags",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          set_flags, &h_set_flags);
+  cls_register_cxx_method(h_class, "op_features_get", CLS_METHOD_RD,
+                          op_features_get, &h_op_features_get);
+  cls_register_cxx_method(h_class, "op_features_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          op_features_set, &h_op_features_set);
+  cls_register_cxx_method(h_class, "metadata_list",
+                          CLS_METHOD_RD,
+			  metadata_list, &h_metadata_list);
+  cls_register_cxx_method(h_class, "metadata_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  metadata_set, &h_metadata_set);
+  cls_register_cxx_method(h_class, "metadata_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  metadata_remove, &h_metadata_remove);
+  cls_register_cxx_method(h_class, "metadata_get",
+                          CLS_METHOD_RD,
+			  metadata_get, &h_metadata_get);
+  cls_register_cxx_method(h_class, "snapshot_get_limit",
+			  CLS_METHOD_RD,
+			  snapshot_get_limit, &h_snapshot_get_limit);
+  cls_register_cxx_method(h_class, "snapshot_set_limit",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  snapshot_set_limit, &h_snapshot_set_limit);
+  cls_register_cxx_method(h_class, "child_attach",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          child_attach, &h_child_attach);
+  cls_register_cxx_method(h_class, "child_detach",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          child_detach, &h_child_detach);
+  cls_register_cxx_method(h_class, "children_list",
+                          CLS_METHOD_RD,
+                          children_list, &h_children_list);
+  cls_register_cxx_method(h_class, "migration_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          migration_set, &h_migration_set);
+  cls_register_cxx_method(h_class, "migration_set_state",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          migration_set_state, &h_migration_set_state);
+  cls_register_cxx_method(h_class, "migration_get",
+                          CLS_METHOD_RD,
+                          migration_get, &h_migration_get);
+  cls_register_cxx_method(h_class, "migration_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          migration_remove, &h_migration_remove);
+
+  cls_register_cxx_method(h_class, "set_modify_timestamp",
+	            	  CLS_METHOD_RD | CLS_METHOD_WR,
+                          set_modify_timestamp, &h_set_modify_timestamp);
+
+  cls_register_cxx_method(h_class, "set_access_timestamp",
+	            	  CLS_METHOD_RD | CLS_METHOD_WR,
+                          set_access_timestamp, &h_set_access_timestamp);
+
+  /* methods for the rbd_children object */
+  cls_register_cxx_method(h_class, "add_child",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  add_child, &h_add_child);
+  cls_register_cxx_method(h_class, "remove_child",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  remove_child, &h_remove_child);
+  cls_register_cxx_method(h_class, "get_children",
+			  CLS_METHOD_RD,
+			  get_children, &h_get_children);
+
+  /* methods for the rbd_id.$image_name objects */
+  cls_register_cxx_method(h_class, "get_id",
+			  CLS_METHOD_RD,
+			  get_id, &h_get_id);
+  cls_register_cxx_method(h_class, "set_id",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  set_id, &h_set_id);
+
+  /* methods for the rbd_directory object */
+  cls_register_cxx_method(h_class, "dir_get_id",
+			  CLS_METHOD_RD,
+			  dir_get_id, &h_dir_get_id);
+  cls_register_cxx_method(h_class, "dir_get_name",
+			  CLS_METHOD_RD,
+			  dir_get_name, &h_dir_get_name);
+  cls_register_cxx_method(h_class, "dir_list",
+			  CLS_METHOD_RD,
+			  dir_list, &h_dir_list);
+  cls_register_cxx_method(h_class, "dir_add_image",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  dir_add_image, &h_dir_add_image);
+  cls_register_cxx_method(h_class, "dir_remove_image",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  dir_remove_image, &h_dir_remove_image);
+  cls_register_cxx_method(h_class, "dir_rename_image",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  dir_rename_image, &h_dir_rename_image);
+  cls_register_cxx_method(h_class, "dir_state_assert", CLS_METHOD_RD,
+                          dir_state_assert, &h_dir_state_assert);
+  cls_register_cxx_method(h_class, "dir_state_set",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+                          dir_state_set, &h_dir_state_set);
+
+  /* methods for the rbd_object_map.$image_id object */
+  cls_register_cxx_method(h_class, "object_map_load",
+                          CLS_METHOD_RD,
+			  object_map_load, &h_object_map_load);
+  cls_register_cxx_method(h_class, "object_map_save",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  object_map_save, &h_object_map_save);
+  cls_register_cxx_method(h_class, "object_map_resize",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  object_map_resize, &h_object_map_resize);
+  cls_register_cxx_method(h_class, "object_map_update",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  object_map_update, &h_object_map_update);
+  cls_register_cxx_method(h_class, "object_map_snap_add",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  object_map_snap_add, &h_object_map_snap_add);
+  cls_register_cxx_method(h_class, "object_map_snap_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+			  object_map_snap_remove, &h_object_map_snap_remove);
+
+ /* methods for the old format */
+  cls_register_cxx_method(h_class, "snap_list",
+			  CLS_METHOD_RD,
+			  old_snapshots_list, &h_old_snapshots_list);
+  cls_register_cxx_method(h_class, "snap_add",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  old_snapshot_add, &h_old_snapshot_add);
+  cls_register_cxx_method(h_class, "snap_remove",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  old_snapshot_remove, &h_old_snapshot_remove);
+  cls_register_cxx_method(h_class, "snap_rename",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  old_snapshot_rename, &h_old_snapshot_rename);
+
+  /* methods for the rbd_mirroring object */
+  cls_register_cxx_method(h_class, "mirror_uuid_get", CLS_METHOD_RD,
+                          mirror_uuid_get, &h_mirror_uuid_get);
+  cls_register_cxx_method(h_class, "mirror_uuid_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_uuid_set, &h_mirror_uuid_set);
+  cls_register_cxx_method(h_class, "mirror_mode_get", CLS_METHOD_RD,
+                          mirror_mode_get, &h_mirror_mode_get);
+  cls_register_cxx_method(h_class, "mirror_mode_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_mode_set, &h_mirror_mode_set);
+  cls_register_cxx_method(h_class, "mirror_peer_list", CLS_METHOD_RD,
+                          mirror_peer_list, &h_mirror_peer_list);
+  cls_register_cxx_method(h_class, "mirror_peer_add",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_peer_add, &h_mirror_peer_add);
+  cls_register_cxx_method(h_class, "mirror_peer_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_peer_remove, &h_mirror_peer_remove);
+  cls_register_cxx_method(h_class, "mirror_peer_set_client",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_peer_set_client, &h_mirror_peer_set_client);
+  cls_register_cxx_method(h_class, "mirror_peer_set_cluster",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_peer_set_cluster, &h_mirror_peer_set_cluster);
+  cls_register_cxx_method(h_class, "mirror_image_list", CLS_METHOD_RD,
+                          mirror_image_list, &h_mirror_image_list);
+  cls_register_cxx_method(h_class, "mirror_image_get_image_id", CLS_METHOD_RD,
+                          mirror_image_get_image_id,
+                          &h_mirror_image_get_image_id);
+  cls_register_cxx_method(h_class, "mirror_image_get", CLS_METHOD_RD,
+                          mirror_image_get, &h_mirror_image_get);
+  cls_register_cxx_method(h_class, "mirror_image_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_image_set, &h_mirror_image_set);
+  cls_register_cxx_method(h_class, "mirror_image_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_image_remove, &h_mirror_image_remove);
+  cls_register_cxx_method(h_class, "mirror_image_status_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+                          mirror_image_status_set, &h_mirror_image_status_set);
+  cls_register_cxx_method(h_class, "mirror_image_status_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_image_status_remove,
+			  &h_mirror_image_status_remove);
+  cls_register_cxx_method(h_class, "mirror_image_status_get", CLS_METHOD_RD,
+                          mirror_image_status_get, &h_mirror_image_status_get);
+  cls_register_cxx_method(h_class, "mirror_image_status_list", CLS_METHOD_RD,
+                          mirror_image_status_list,
+			  &h_mirror_image_status_list);
+  cls_register_cxx_method(h_class, "mirror_image_status_get_summary",
+			  CLS_METHOD_RD, mirror_image_status_get_summary,
+			  &h_mirror_image_status_get_summary);
+  cls_register_cxx_method(h_class, "mirror_image_status_remove_down",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_image_status_remove_down,
+			  &h_mirror_image_status_remove_down);
+  cls_register_cxx_method(h_class, "mirror_image_instance_get", CLS_METHOD_RD,
+                          mirror_image_instance_get,
+                          &h_mirror_image_instance_get);
+  cls_register_cxx_method(h_class, "mirror_image_instance_list", CLS_METHOD_RD,
+                          mirror_image_instance_list,
+                          &h_mirror_image_instance_list);
+  cls_register_cxx_method(h_class, "mirror_instances_list", CLS_METHOD_RD,
+                          mirror_instances_list, &h_mirror_instances_list);
+  cls_register_cxx_method(h_class, "mirror_instances_add",
+                          CLS_METHOD_RD | CLS_METHOD_WR | CLS_METHOD_PROMOTE,
+                          mirror_instances_add, &h_mirror_instances_add);
+  cls_register_cxx_method(h_class, "mirror_instances_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          mirror_instances_remove,
+                          &h_mirror_instances_remove);
+  cls_register_cxx_method(h_class, "mirror_image_map_list",
+                          CLS_METHOD_RD, mirror_image_map_list,
+                          &h_mirror_image_map_list);
+  cls_register_cxx_method(h_class, "mirror_image_map_update",
+                          CLS_METHOD_WR, mirror_image_map_update,
+                          &h_mirror_image_map_update);
+  cls_register_cxx_method(h_class, "mirror_image_map_remove",
+                          CLS_METHOD_WR, mirror_image_map_remove,
+                          &h_mirror_image_map_remove);
+
+  /* methods for the groups feature */
+  cls_register_cxx_method(h_class, "group_dir_list",
+			  CLS_METHOD_RD,
+			  group_dir_list, &h_group_dir_list);
+  cls_register_cxx_method(h_class, "group_dir_add",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  group_dir_add, &h_group_dir_add);
+  cls_register_cxx_method(h_class, "group_dir_remove",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  group_dir_remove, &h_group_dir_remove);
+  cls_register_cxx_method(h_class, "group_dir_rename",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          group_dir_rename, &h_group_dir_rename);
+  cls_register_cxx_method(h_class, "group_image_remove",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  group_image_remove, &h_group_image_remove);
+  cls_register_cxx_method(h_class, "group_image_list",
+			  CLS_METHOD_RD,
+			  group_image_list, &h_group_image_list);
+  cls_register_cxx_method(h_class, "group_image_set",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  group_image_set, &h_group_image_set);
+  cls_register_cxx_method(h_class, "image_group_add",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  image_group_add, &h_image_group_add);
+  cls_register_cxx_method(h_class, "image_group_remove",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  image_group_remove, &h_image_group_remove);
+  cls_register_cxx_method(h_class, "image_group_get",
+			  CLS_METHOD_RD,
+			  image_group_get, &h_image_group_get);
+  cls_register_cxx_method(h_class, "group_snap_set",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  group_snap_set, &h_group_snap_set);
+  cls_register_cxx_method(h_class, "group_snap_remove",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  group_snap_remove, &h_group_snap_remove);
+  cls_register_cxx_method(h_class, "group_snap_get_by_id",
+			  CLS_METHOD_RD,
+			  group_snap_get_by_id, &h_group_snap_get_by_id);
+  cls_register_cxx_method(h_class, "group_snap_list",
+			  CLS_METHOD_RD,
+			  group_snap_list, &h_group_snap_list);
+
+  /* rbd_trash object methods */
+  cls_register_cxx_method(h_class, "trash_add",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          trash_add, &h_trash_add);
+  cls_register_cxx_method(h_class, "trash_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          trash_remove, &h_trash_remove);
+  cls_register_cxx_method(h_class, "trash_list",
+                          CLS_METHOD_RD,
+                          trash_list, &h_trash_list);
+  cls_register_cxx_method(h_class, "trash_get",
+                          CLS_METHOD_RD,
+                          trash_get, &h_trash_get);
+  cls_register_cxx_method(h_class, "trash_state_set",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          trash_state_set, &h_trash_state_set);
+
+  /* rbd_namespace object methods */
+  cls_register_cxx_method(h_class, "namespace_add",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          namespace_add, &h_namespace_add);
+  cls_register_cxx_method(h_class, "namespace_remove",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          namespace_remove, &h_namespace_remove);
+  cls_register_cxx_method(h_class, "namespace_list", CLS_METHOD_RD,
+                          namespace_list, &h_namespace_list);
+
+  /* data object methods */
+  cls_register_cxx_method(h_class, "copyup",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  copyup, &h_copyup);
+  cls_register_cxx_method(h_class, "assert_snapc_seq",
+                          CLS_METHOD_RD | CLS_METHOD_WR,
+                          assert_snapc_seq,
+                          &h_assert_snapc_seq);
+  cls_register_cxx_method(h_class, "sparsify",
+			  CLS_METHOD_RD | CLS_METHOD_WR,
+			  sparsify, &h_sparsify);
+}
diff --git a/src/cls/rbd/cls_rbd.h b/src/cls/rbd/cls_rbd.h
new file mode 100644
index 00000000..0dc0a9b9
--- /dev/null
+++ b/src/cls/rbd/cls_rbd.h
@@ -0,0 +1,243 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef __CEPH_CLS_RBD_H
+#define __CEPH_CLS_RBD_H
+
+#include "include/types.h"
+#include "include/buffer_fwd.h"
+#include "include/rbd_types.h"
+#include "common/Formatter.h"
+#include "cls/rbd/cls_rbd_types.h"
+
+/// information about our parent image, if any
+struct cls_rbd_parent {
+  int64_t pool_id = -1;
+  std::string pool_namespace;
+  std::string image_id;
+  snapid_t snap_id = CEPH_NOSNAP;
+  std::optional<uint64_t> head_overlap = std::nullopt;
+
+  cls_rbd_parent() {
+  }
+  cls_rbd_parent(const cls::rbd::ParentImageSpec& parent_image_spec,
+                 const std::optional<uint64_t>& head_overlap)
+    : pool_id(parent_image_spec.pool_id),
+      pool_namespace(parent_image_spec.pool_namespace),
+      image_id(parent_image_spec.image_id), snap_id(parent_image_spec.snap_id),
+      head_overlap(head_overlap) {
+  }
+
+  inline bool exists() const {
+    return (pool_id >= 0 && !image_id.empty() && snap_id != CEPH_NOSNAP);
+  }
+
+  inline bool operator==(const cls_rbd_parent& rhs) const {
+    return (pool_id == rhs.pool_id &&
+            pool_namespace == rhs.pool_namespace &&
+            image_id == rhs.image_id &&
+            snap_id == rhs.snap_id);
+  }
+  inline bool operator!=(const cls_rbd_parent& rhs) const {
+    return !(*this == rhs);
+  }
+
+  void encode(bufferlist& bl, uint64_t features) const {
+    // NOTE: remove support for version 1 after Nautilus EOLed
+    uint8_t version = 1;
+    if ((features & CEPH_FEATURE_SERVER_NAUTILUS) != 0ULL) {
+      // break backwards compatability when using nautilus or later OSDs
+      version = 2;
+    }
+
+    ENCODE_START(version, version, bl);
+    encode(pool_id, bl);
+    if (version >= 2) {
+      encode(pool_namespace, bl);
+    }
+    encode(image_id, bl);
+    encode(snap_id, bl);
+    if (version == 1) {
+      encode(head_overlap.value_or(0ULL), bl);
+    } else {
+      encode(head_overlap, bl);
+    }
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(pool_id, bl);
+    if (struct_v >= 2) {
+      decode(pool_namespace, bl);
+    }
+    decode(image_id, bl);
+    decode(snap_id, bl);
+    if (struct_v == 1) {
+      uint64_t overlap;
+      decode(overlap, bl);
+      head_overlap = overlap;
+    } else {
+      decode(head_overlap, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const {
+    f->dump_int("pool_id", pool_id);
+    f->dump_string("pool_namespace", pool_namespace);
+    f->dump_string("image_id", image_id);
+    f->dump_unsigned("snap_id", snap_id);
+    if (head_overlap) {
+      f->dump_unsigned("head_overlap", *head_overlap);
+    }
+  }
+
+  static void generate_test_instances(list<cls_rbd_parent*>& o) {
+    o.push_back(new cls_rbd_parent{});
+    o.push_back(new cls_rbd_parent{{1, "", "image id", 234}, {}});
+    o.push_back(new cls_rbd_parent{{1, "", "image id", 234}, {123}});
+    o.push_back(new cls_rbd_parent{{1, "ns", "image id", 234}, {123}});
+  }
+};
+WRITE_CLASS_ENCODER_FEATURES(cls_rbd_parent)
+
+struct cls_rbd_snap {
+  snapid_t id = CEPH_NOSNAP;
+  std::string name;
+  uint64_t image_size = 0;
+  uint8_t protection_status = RBD_PROTECTION_STATUS_UNPROTECTED;
+  cls_rbd_parent parent;
+  uint64_t flags = 0;
+  utime_t timestamp;
+  cls::rbd::SnapshotNamespace snapshot_namespace = {
+    cls::rbd::UserSnapshotNamespace{}};
+  uint32_t child_count = 0;
+  std::optional<uint64_t> parent_overlap = std::nullopt;
+
+  cls_rbd_snap() {
+  }
+  cls_rbd_snap(snapid_t id, const std::string& name, uint64_t image_size,
+               uint8_t protection_status, const cls_rbd_parent& parent,
+               uint64_t flags, utime_t timestamp,
+               const cls::rbd::SnapshotNamespace& snapshot_namespace,
+               uint32_t child_count,
+               const std::optional<uint64_t>& parent_overlap)
+    : id(id), name(name), image_size(image_size),
+      protection_status(protection_status), parent(parent), flags(flags),
+      timestamp(timestamp), snapshot_namespace(snapshot_namespace),
+      child_count(child_count), parent_overlap(parent_overlap) {
+  }
+
+  bool migrate_parent_format(uint64_t features) const {
+    return (((features & CEPH_FEATURE_SERVER_NAUTILUS) != 0) &&
+            (parent.exists()));
+  }
+
+  void encode(bufferlist& bl, uint64_t features) const {
+    // NOTE: remove support for versions < 8 after Nautilus EOLed
+    uint8_t min_version = 1;
+    if ((features & CEPH_FEATURE_SERVER_NAUTILUS) != 0ULL) {
+      // break backwards compatability when using nautilus or later OSDs
+      min_version = 8;
+    }
+
+    ENCODE_START(8, min_version, bl);
+    encode(id, bl);
+    encode(name, bl);
+    encode(image_size, bl);
+    if (min_version < 8) {
+      uint64_t image_features = 0;
+      encode(image_features, bl); // unused -- preserve ABI
+      encode(parent, bl, features);
+    }
+    encode(protection_status, bl);
+    encode(flags, bl);
+    encode(snapshot_namespace, bl);
+    encode(timestamp, bl);
+    encode(child_count, bl);
+    encode(parent_overlap, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& p) {
+    DECODE_START(8, p);
+    decode(id, p);
+    decode(name, p);
+    decode(image_size, p);
+    if (struct_compat < 8) {
+      uint64_t features;
+      decode(features, p); // unused -- preserve ABI
+    }
+    if (struct_v >= 2 && struct_compat < 8) {
+      decode(parent, p);
+    }
+    if (struct_v >= 3) {
+      decode(protection_status, p);
+    }
+    if (struct_v >= 4) {
+      decode(flags, p);
+    }
+    if (struct_v >= 5) {
+      decode(snapshot_namespace, p);
+    }
+    if (struct_v >= 6) {
+      decode(timestamp, p);
+    }
+    if (struct_v >= 7) {
+      decode(child_count, p);
+    }
+    if (struct_v >= 8) {
+      decode(parent_overlap, p);
+    }
+    DECODE_FINISH(p);
+  }
+
+  void dump(Formatter *f) const {
+    f->dump_unsigned("id", id);
+    f->dump_string("name", name);
+    f->dump_unsigned("image_size", image_size);
+    if (parent.exists()) {
+      f->open_object_section("parent");
+      parent.dump(f);
+      f->close_section();
+    }
+    switch (protection_status) {
+    case RBD_PROTECTION_STATUS_UNPROTECTED:
+      f->dump_string("protection_status", "unprotected");
+      break;
+    case RBD_PROTECTION_STATUS_UNPROTECTING:
+      f->dump_string("protection_status", "unprotecting");
+      break;
+    case RBD_PROTECTION_STATUS_PROTECTED:
+      f->dump_string("protection_status", "protected");
+      break;
+    default:
+      ceph_abort();
+    }
+    f->dump_unsigned("child_count", child_count);
+    if (parent_overlap) {
+      f->dump_unsigned("parent_overlap", *parent_overlap);
+    }
+  }
+
+  static void generate_test_instances(list<cls_rbd_snap*>& o) {
+    o.push_back(new cls_rbd_snap{});
+    o.push_back(new cls_rbd_snap{1, "snap", 123456,
+                                 RBD_PROTECTION_STATUS_PROTECTED,
+                                 {{1, "", "image", 123}, 234}, 31, {},
+                                 cls::rbd::UserSnapshotNamespace{}, 543, {}});
+    o.push_back(new cls_rbd_snap{1, "snap", 123456,
+                                 RBD_PROTECTION_STATUS_PROTECTED,
+                                 {{1, "", "image", 123}, 234}, 31, {},
+                                 cls::rbd::UserSnapshotNamespace{}, 543, {0}});
+    o.push_back(new cls_rbd_snap{1, "snap", 123456,
+                                 RBD_PROTECTION_STATUS_PROTECTED,
+                                 {{1, "ns", "image", 123}, 234}, 31, {},
+                                 cls::rbd::UserSnapshotNamespace{}, 543,
+                                 {123}});
+  }
+};
+WRITE_CLASS_ENCODER_FEATURES(cls_rbd_snap)
+
+#endif // __CEPH_CLS_RBD_H
diff --git a/src/cls/rbd/cls_rbd_client.cc b/src/cls/rbd/cls_rbd_client.cc
new file mode 100644
index 00000000..9b868c55
--- /dev/null
+++ b/src/cls/rbd/cls_rbd_client.cc
@@ -0,0 +1,2807 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/rbd/cls_rbd_client.h"
+#include "cls/lock/cls_lock_client.h"
+#include "include/buffer.h"
+#include "include/encoding.h"
+#include "include/rbd_types.h"
+#include "include/rados/librados.hpp"
+#include "common/bit_vector.hpp"
+
+#include <errno.h>
+
+namespace librbd {
+namespace cls_client {
+
+void create_image(librados::ObjectWriteOperation *op, uint64_t size,
+                  uint8_t order, uint64_t features,
+                  const std::string &object_prefix, int64_t data_pool_id)
+{
+  bufferlist bl;
+  encode(size, bl);
+  encode(order, bl);
+  encode(features, bl);
+  encode(object_prefix, bl);
+  encode(data_pool_id, bl);
+
+  op->exec("rbd", "create", bl);
+}
+
+int create_image(librados::IoCtx *ioctx, const std::string &oid,
+                 uint64_t size, uint8_t order, uint64_t features,
+                 const std::string &object_prefix, int64_t data_pool_id)
+{
+  librados::ObjectWriteOperation op;
+  create_image(&op, size, order, features, object_prefix, data_pool_id);
+
+  return ioctx->operate(oid, &op);
+}
+
+void get_features_start(librados::ObjectReadOperation *op, bool read_only)
+{
+  bufferlist bl;
+  encode(static_cast<uint64_t>(CEPH_NOSNAP), bl);
+  encode(read_only, bl);
+  op->exec("rbd", "get_features", bl);
+}
+
+int get_features_finish(bufferlist::const_iterator *it, uint64_t *features,
+                        uint64_t *incompatible_features)
+{
+  try {
+    decode(*features, *it);
+    decode(*incompatible_features, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int get_features(librados::IoCtx *ioctx, const std::string &oid,
+                 bool read_only, uint64_t *features,
+                 uint64_t *incompatible_features)
+{
+  librados::ObjectReadOperation op;
+  get_features_start(&op, read_only);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_features_finish(&it, features, incompatible_features);
+}
+
+void set_features(librados::ObjectWriteOperation *op, uint64_t features,
+                  uint64_t mask)
+{
+  bufferlist bl;
+  encode(features, bl);
+  encode(mask, bl);
+
+  op->exec("rbd", "set_features", bl);
+}
+
+int set_features(librados::IoCtx *ioctx, const std::string &oid,
+                  uint64_t features, uint64_t mask)
+{
+  librados::ObjectWriteOperation op;
+  set_features(&op, features, mask);
+
+  return ioctx->operate(oid, &op);
+}
+
+void get_object_prefix_start(librados::ObjectReadOperation *op)
+{
+  bufferlist bl;
+  op->exec("rbd", "get_object_prefix", bl);
+}
+
+int get_object_prefix_finish(bufferlist::const_iterator *it,
+                             std::string *object_prefix)
+{
+  try {
+    decode(*object_prefix, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_object_prefix(librados::IoCtx *ioctx, const std::string &oid,
+                      std::string *object_prefix)
+{
+  librados::ObjectReadOperation op;
+  get_object_prefix_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_object_prefix_finish(&it, object_prefix);
+}
+
+void get_data_pool_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "get_data_pool", bl);
+}
+
+int get_data_pool_finish(bufferlist::const_iterator *it, int64_t *data_pool_id) {
+  try {
+    decode(*data_pool_id, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_data_pool(librados::IoCtx *ioctx, const std::string &oid,
+                  int64_t *data_pool_id) {
+  librados::ObjectReadOperation op;
+  get_data_pool_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_data_pool_finish(&it, data_pool_id);
+}
+
+void get_size_start(librados::ObjectReadOperation *op, snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "get_size", bl);
+}
+
+int get_size_finish(bufferlist::const_iterator *it, uint64_t *size,
+                    uint8_t *order)
+{
+  try {
+    decode(*order, *it);
+    decode(*size, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_size(librados::IoCtx *ioctx, const std::string &oid,
+             snapid_t snap_id, uint64_t *size, uint8_t *order)
+{
+  librados::ObjectReadOperation op;
+  get_size_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_size_finish(&it, size, order);
+}
+
+int set_size(librados::IoCtx *ioctx, const std::string &oid,
+             uint64_t size)
+{
+  librados::ObjectWriteOperation op;
+  set_size(&op, size);
+  return ioctx->operate(oid, &op);
+}
+
+void set_size(librados::ObjectWriteOperation *op, uint64_t size)
+{
+  bufferlist bl;
+  encode(size, bl);
+  op->exec("rbd", "set_size", bl);
+}
+
+void get_flags_start(librados::ObjectReadOperation *op, snapid_t snap_id) {
+  bufferlist in_bl;
+  encode(static_cast<snapid_t>(snap_id), in_bl);
+  op->exec("rbd", "get_flags", in_bl);
+}
+
+int get_flags_finish(bufferlist::const_iterator *it, uint64_t *flags) {
+  try {
+    decode(*flags, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_flags(librados::IoCtx *ioctx, const std::string &oid,
+              snapid_t snap_id, uint64_t *flags)
+{
+  librados::ObjectReadOperation op;
+  get_flags_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_flags_finish(&it, flags);
+}
+
+void set_flags(librados::ObjectWriteOperation *op, snapid_t snap_id,
+               uint64_t flags, uint64_t mask)
+{
+  bufferlist inbl;
+  encode(flags, inbl);
+  encode(mask, inbl);
+  encode(snap_id, inbl);
+  op->exec("rbd", "set_flags", inbl);
+}
+
+void op_features_get_start(librados::ObjectReadOperation *op)
+{
+  bufferlist in_bl;
+  op->exec("rbd", "op_features_get", in_bl);
+}
+
+int op_features_get_finish(bufferlist::const_iterator *it, uint64_t *op_features)
+{
+  try {
+    decode(*op_features, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int op_features_get(librados::IoCtx *ioctx, const std::string &oid,
+                    uint64_t *op_features)
+{
+  librados::ObjectReadOperation op;
+  op_features_get_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return op_features_get_finish(&it, op_features);
+}
+
+void op_features_set(librados::ObjectWriteOperation *op,
+                     uint64_t op_features, uint64_t mask)
+{
+  bufferlist inbl;
+  encode(op_features, inbl);
+  encode(mask, inbl);
+  op->exec("rbd", "op_features_set", inbl);
+}
+
+int op_features_set(librados::IoCtx *ioctx, const std::string &oid,
+                    uint64_t op_features, uint64_t mask)
+{
+  librados::ObjectWriteOperation op;
+  op_features_set(&op, op_features, mask);
+
+  return ioctx->operate(oid, &op);
+}
+
+void get_parent_start(librados::ObjectReadOperation *op, snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "get_parent", bl);
+}
+
+int get_parent_finish(bufferlist::const_iterator *it,
+                      cls::rbd::ParentImageSpec *pspec,
+                      uint64_t *parent_overlap)
+{
+  *pspec = {};
+  try {
+    decode(pspec->pool_id, *it);
+    decode(pspec->image_id, *it);
+    decode(pspec->snap_id, *it);
+    decode(*parent_overlap, *it);
+  } catch (const buffer::error &) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_parent(librados::IoCtx *ioctx, const std::string &oid,
+               snapid_t snap_id, cls::rbd::ParentImageSpec *pspec,
+               uint64_t *parent_overlap)
+{
+  librados::ObjectReadOperation op;
+  get_parent_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_parent_finish(&it, pspec, parent_overlap);
+}
+
+int set_parent(librados::IoCtx *ioctx, const std::string &oid,
+               const cls::rbd::ParentImageSpec &pspec, uint64_t parent_overlap)
+{
+  librados::ObjectWriteOperation op;
+  set_parent(&op, pspec, parent_overlap);
+  return ioctx->operate(oid, &op);
+}
+
+void set_parent(librados::ObjectWriteOperation *op,
+                const cls::rbd::ParentImageSpec &pspec,
+                uint64_t parent_overlap) {
+  assert(pspec.pool_namespace.empty());
+
+  bufferlist in_bl;
+  encode(pspec.pool_id, in_bl);
+  encode(pspec.image_id, in_bl);
+  encode(pspec.snap_id, in_bl);
+  encode(parent_overlap, in_bl);
+
+  op->exec("rbd", "set_parent", in_bl);
+}
+
+int remove_parent(librados::IoCtx *ioctx, const std::string &oid)
+{
+  librados::ObjectWriteOperation op;
+  remove_parent(&op);
+  return ioctx->operate(oid, &op);
+}
+
+void remove_parent(librados::ObjectWriteOperation *op)
+{
+  bufferlist inbl;
+  op->exec("rbd", "remove_parent", inbl);
+}
+
+void parent_get_start(librados::ObjectReadOperation* op) {
+  bufferlist in_bl;
+  op->exec("rbd", "parent_get", in_bl);
+}
+
+int parent_get_finish(bufferlist::const_iterator* it,
+                      cls::rbd::ParentImageSpec* parent_image_spec) {
+  try {
+    decode(*parent_image_spec, *it);
+  } catch (const buffer::error &) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int parent_get(librados::IoCtx* ioctx, const std::string &oid,
+               cls::rbd::ParentImageSpec* parent_image_spec) {
+  librados::ObjectReadOperation op;
+  parent_get_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  r = parent_get_finish(&it, parent_image_spec);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void parent_overlap_get_start(librados::ObjectReadOperation* op,
+                              snapid_t snap_id) {
+  bufferlist in_bl;
+  encode(snap_id, in_bl);
+  op->exec("rbd", "parent_overlap_get", in_bl);
+}
+
+int parent_overlap_get_finish(bufferlist::const_iterator* it,
+                              std::optional<uint64_t>* parent_overlap) {
+  try {
+    decode(*parent_overlap, *it);
+  } catch (const buffer::error &) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int parent_overlap_get(librados::IoCtx* ioctx, const std::string &oid,
+                       snapid_t snap_id,
+                       std::optional<uint64_t>* parent_overlap) {
+  librados::ObjectReadOperation op;
+  parent_overlap_get_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  r = parent_overlap_get_finish(&it, parent_overlap);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void parent_attach(librados::ObjectWriteOperation* op,
+                   const cls::rbd::ParentImageSpec& parent_image_spec,
+                   uint64_t parent_overlap, bool reattach) {
+  bufferlist in_bl;
+  encode(parent_image_spec, in_bl);
+  encode(parent_overlap, in_bl);
+  encode(reattach, in_bl);
+  op->exec("rbd", "parent_attach", in_bl);
+}
+
+int parent_attach(librados::IoCtx *ioctx, const std::string &oid,
+                  const cls::rbd::ParentImageSpec& parent_image_spec,
+                  uint64_t parent_overlap, bool reattach) {
+  librados::ObjectWriteOperation op;
+  parent_attach(&op, parent_image_spec, parent_overlap, reattach);
+  return ioctx->operate(oid, &op);
+}
+
+void parent_detach(librados::ObjectWriteOperation* op) {
+  bufferlist in_bl;
+  op->exec("rbd", "parent_detach", in_bl);
+}
+
+int parent_detach(librados::IoCtx *ioctx, const std::string &oid) {
+  librados::ObjectWriteOperation op;
+  parent_detach(&op);
+  return ioctx->operate(oid, &op);
+}
+
+int add_child(librados::IoCtx *ioctx, const std::string &oid,
+              const cls::rbd::ParentImageSpec &pspec,
+              const std::string &c_imageid)
+{
+  librados::ObjectWriteOperation op;
+  add_child(&op, pspec, c_imageid);
+  return ioctx->operate(oid, &op);
+}
+
+void add_child(librados::ObjectWriteOperation *op,
+               const cls::rbd::ParentImageSpec& pspec,
+               const std::string &c_imageid)
+{
+  assert(pspec.pool_namespace.empty());
+
+  bufferlist in;
+  encode(pspec.pool_id, in);
+  encode(pspec.image_id, in);
+  encode(pspec.snap_id, in);
+  encode(c_imageid, in);
+
+  op->exec("rbd", "add_child", in);
+}
+
+void remove_child(librados::ObjectWriteOperation *op,
+                  const cls::rbd::ParentImageSpec &pspec,
+                  const std::string &c_imageid)
+{
+  assert(pspec.pool_namespace.empty());
+
+  bufferlist in;
+  encode(pspec.pool_id, in);
+  encode(pspec.image_id, in);
+  encode(pspec.snap_id, in);
+  encode(c_imageid, in);
+  op->exec("rbd", "remove_child", in);
+}
+
+int remove_child(librados::IoCtx *ioctx, const std::string &oid,
+                 const cls::rbd::ParentImageSpec &pspec,
+                 const std::string &c_imageid)
+{
+  librados::ObjectWriteOperation op;
+  remove_child(&op, pspec, c_imageid);
+  return ioctx->operate(oid, &op);
+}
+
+void get_children_start(librados::ObjectReadOperation *op,
+                        const cls::rbd::ParentImageSpec &pspec) {
+  bufferlist in_bl;
+  encode(pspec.pool_id, in_bl);
+  encode(pspec.image_id, in_bl);
+  encode(pspec.snap_id, in_bl);
+  op->exec("rbd", "get_children", in_bl);
+}
+
+int get_children_finish(bufferlist::const_iterator *it,
+                        std::set<std::string>* children) {
+  try {
+    decode(*children, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_children(librados::IoCtx *ioctx, const std::string &oid,
+                 const cls::rbd::ParentImageSpec &pspec, set<string>& children)
+{
+  librados::ObjectReadOperation op;
+  get_children_start(&op, pspec);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_children_finish(&it, &children);
+}
+
+void snapshot_get_start(librados::ObjectReadOperation *op, snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "snapshot_get", bl);
+}
+
+int snapshot_get_finish(bufferlist::const_iterator* it,
+                        cls::rbd::SnapshotInfo* snap_info)
+{
+  try {
+    decode(*snap_info, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int snapshot_get(librados::IoCtx *ioctx, const std::string &oid,
+                 snapid_t snap_id, cls::rbd::SnapshotInfo* snap_info)
+{
+  librados::ObjectReadOperation op;
+  snapshot_get_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return snapshot_get_finish(&it, snap_info);
+}
+
+void snapshot_add(librados::ObjectWriteOperation *op, snapid_t snap_id,
+                  const std::string &snap_name,
+                  const cls::rbd::SnapshotNamespace &snap_namespace)
+{
+  bufferlist bl;
+  encode(snap_name, bl);
+  encode(snap_id, bl);
+  encode(snap_namespace, bl);
+  op->exec("rbd", "snapshot_add", bl);
+}
+
+void snapshot_remove(librados::ObjectWriteOperation *op, snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "snapshot_remove", bl);
+}
+
+void snapshot_rename(librados::ObjectWriteOperation *op,
+                     snapid_t src_snap_id,
+                     const std::string &dst_name)
+{
+  bufferlist bl;
+  encode(src_snap_id, bl);
+  encode(dst_name, bl);
+  op->exec("rbd", "snapshot_rename", bl);
+}
+
+void snapshot_trash_add(librados::ObjectWriteOperation *op,
+                        snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "snapshot_trash_add", bl);
+}
+
+void get_snapcontext_start(librados::ObjectReadOperation *op)
+{
+  bufferlist bl;
+  op->exec("rbd", "get_snapcontext", bl);
+}
+
+int get_snapcontext_finish(bufferlist::const_iterator *it,
+                           ::SnapContext *snapc)
+{
+  try {
+    decode(*snapc, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  if (!snapc->is_valid()) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_snapcontext(librados::IoCtx *ioctx, const std::string &oid,
+                    ::SnapContext *snapc)
+{
+  librados::ObjectReadOperation op;
+  get_snapcontext_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto bl_it = out_bl.cbegin();
+  return get_snapcontext_finish(&bl_it, snapc);
+}
+
+void get_snapshot_name_start(librados::ObjectReadOperation *op,
+                             snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "get_snapshot_name", bl);
+}
+
+int get_snapshot_name_finish(bufferlist::const_iterator *it,
+                             std::string *name)
+{
+  try {
+    decode(*name, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_snapshot_name(librados::IoCtx *ioctx, const std::string &oid,
+                      snapid_t snap_id, std::string *name)
+{
+  librados::ObjectReadOperation op;
+  get_snapshot_name_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_snapshot_name_finish(&it, name);
+}
+
+void get_snapshot_timestamp_start(librados::ObjectReadOperation *op,
+                                  snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "get_snapshot_timestamp", bl);
+}
+
+int get_snapshot_timestamp_finish(bufferlist::const_iterator *it,
+                                  utime_t *timestamp)
+{
+  try {
+    decode(*timestamp, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_snapshot_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                           snapid_t snap_id, utime_t *timestamp)
+{
+  librados::ObjectReadOperation op;
+  get_snapshot_timestamp_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_snapshot_timestamp_finish(&it, timestamp);
+}
+
+void old_snapshot_add(librados::ObjectWriteOperation *op,
+                      snapid_t snap_id, const std::string &snap_name)
+{
+  bufferlist bl;
+  encode(snap_name, bl);
+  encode(snap_id, bl);
+  op->exec("rbd", "snap_add", bl);
+}
+
+void old_snapshot_remove(librados::ObjectWriteOperation *op,
+                         const std::string &snap_name)
+{
+  bufferlist bl;
+  encode(snap_name, bl);
+  op->exec("rbd", "snap_remove", bl);
+}
+
+void old_snapshot_rename(librados::ObjectWriteOperation *op,
+                         snapid_t src_snap_id, const std::string &dst_name)
+{
+  bufferlist bl;
+  encode(src_snap_id, bl);
+  encode(dst_name, bl);
+  op->exec("rbd", "snap_rename", bl);
+}
+
+void old_snapshot_list_start(librados::ObjectReadOperation *op) {
+  bufferlist in_bl;
+  op->exec("rbd", "snap_list", in_bl);
+}
+
+int old_snapshot_list_finish(bufferlist::const_iterator *it,
+                             std::vector<string> *names,
+                             std::vector<uint64_t> *sizes,
+                             ::SnapContext *snapc) {
+  try {
+    uint32_t num_snaps;
+    decode(snapc->seq, *it);
+    decode(num_snaps, *it);
+
+    names->resize(num_snaps);
+    sizes->resize(num_snaps);
+    snapc->snaps.resize(num_snaps);
+    for (uint32_t i = 0; i < num_snaps; ++i) {
+      decode(snapc->snaps[i], *it);
+      decode((*sizes)[i], *it);
+      decode((*names)[i], *it);
+    }
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int old_snapshot_list(librados::IoCtx *ioctx, const std::string &oid,
+                      std::vector<string> *names,
+                      std::vector<uint64_t> *sizes,
+                      ::SnapContext *snapc)
+{
+  librados::ObjectReadOperation op;
+  old_snapshot_list_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return old_snapshot_list_finish(&it, names, sizes, snapc);
+}
+
+void get_all_features_start(librados::ObjectReadOperation *op) {
+  bufferlist in;
+  op->exec("rbd", "get_all_features", in);
+}
+
+int get_all_features_finish(bufferlist::const_iterator *it,
+                            uint64_t *all_features) {
+  try {
+    decode(*all_features, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_all_features(librados::IoCtx *ioctx, const std::string &oid,
+                     uint64_t *all_features) {
+  librados::ObjectReadOperation op;
+  get_all_features_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_all_features_finish(&it, all_features);
+}
+
+int copyup(librados::IoCtx *ioctx, const std::string &oid,
+           bufferlist data) {
+  bufferlist out;
+  return ioctx->exec(oid, "rbd", "copyup", data, out);
+}
+
+void get_protection_status_start(librados::ObjectReadOperation *op,
+                                 snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "get_protection_status", bl);
+}
+
+int get_protection_status_finish(bufferlist::const_iterator *it,
+                                 uint8_t *protection_status)
+{
+  try {
+    decode(*protection_status, *it);
+  } catch (const buffer::error &) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_protection_status(librados::IoCtx *ioctx, const std::string &oid,
+                          snapid_t snap_id, uint8_t *protection_status)
+{
+  librados::ObjectReadOperation op;
+  get_protection_status_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_protection_status_finish(&it, protection_status);
+}
+
+int set_protection_status(librados::IoCtx *ioctx, const std::string &oid,
+                          snapid_t snap_id, uint8_t protection_status)
+{
+  // TODO remove
+  librados::ObjectWriteOperation op;
+  set_protection_status(&op, snap_id, protection_status);
+  return ioctx->operate(oid, &op);
+}
+
+void set_protection_status(librados::ObjectWriteOperation *op,
+                           snapid_t snap_id, uint8_t protection_status)
+{
+  bufferlist in;
+  encode(snap_id, in);
+  encode(protection_status, in);
+  op->exec("rbd", "set_protection_status", in);
+}
+
+int snapshot_get_limit(librados::IoCtx *ioctx, const std::string &oid,
+                       uint64_t *limit)
+{
+  bufferlist in, out;
+  int r =  ioctx->exec(oid, "rbd", "snapshot_get_limit", in, out);
+
+  if (r < 0) {
+    return r;
+  }
+
+  try {
+    auto iter = out.cbegin();
+    decode(*limit, iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+void snapshot_set_limit(librados::ObjectWriteOperation *op, uint64_t limit)
+{
+  bufferlist in;
+  encode(limit, in);
+  op->exec("rbd", "snapshot_set_limit", in);
+}
+
+void get_stripe_unit_count_start(librados::ObjectReadOperation *op) {
+  bufferlist empty_bl;
+  op->exec("rbd", "get_stripe_unit_count", empty_bl);
+}
+
+int get_stripe_unit_count_finish(bufferlist::const_iterator *it,
+                                 uint64_t *stripe_unit,
+                                 uint64_t *stripe_count) {
+  ceph_assert(stripe_unit);
+  ceph_assert(stripe_count);
+
+  try {
+    decode(*stripe_unit, *it);
+    decode(*stripe_count, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_stripe_unit_count(librados::IoCtx *ioctx, const std::string &oid,
+                          uint64_t *stripe_unit, uint64_t *stripe_count)
+{
+  librados::ObjectReadOperation op;
+  get_stripe_unit_count_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_stripe_unit_count_finish(&it, stripe_unit, stripe_count);
+}
+
+void set_stripe_unit_count(librados::ObjectWriteOperation *op,
+                           uint64_t stripe_unit, uint64_t stripe_count)
+{
+  bufferlist bl;
+  encode(stripe_unit, bl);
+  encode(stripe_count, bl);
+
+  op->exec("rbd", "set_stripe_unit_count", bl);
+}
+
+int set_stripe_unit_count(librados::IoCtx *ioctx, const std::string &oid,
+                          uint64_t stripe_unit, uint64_t stripe_count)
+{
+  librados::ObjectWriteOperation op;
+  set_stripe_unit_count(&op, stripe_unit, stripe_count);
+
+  return ioctx->operate(oid, &op);
+}
+
+void get_create_timestamp_start(librados::ObjectReadOperation *op) {
+  bufferlist empty_bl;
+  op->exec("rbd", "get_create_timestamp", empty_bl);
+}
+
+int get_create_timestamp_finish(bufferlist::const_iterator *it,
+                                utime_t *timestamp) {
+  ceph_assert(timestamp);
+
+  try {
+    decode(*timestamp, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_create_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                         utime_t *timestamp)
+{
+  librados::ObjectReadOperation op;
+  get_create_timestamp_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_create_timestamp_finish(&it, timestamp);
+}
+
+void get_access_timestamp_start(librados::ObjectReadOperation *op) {
+  bufferlist empty_bl;
+  op->exec("rbd", "get_access_timestamp", empty_bl);
+}
+
+int get_access_timestamp_finish(bufferlist::const_iterator *it,
+                                utime_t *timestamp) {
+  ceph_assert(timestamp);
+
+  try {
+    decode(*timestamp, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_access_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                         utime_t *timestamp)
+{
+  librados::ObjectReadOperation op;
+  get_access_timestamp_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_access_timestamp_finish(&it, timestamp);
+}
+
+void set_access_timestamp(librados::ObjectWriteOperation *op)
+{
+    bufferlist empty_bl;
+    op->exec("rbd","set_access_timestamp",empty_bl);
+}
+
+int set_access_timestamp(librados::IoCtx *ioctx, const std::string &oid)
+{
+    librados::ObjectWriteOperation op;
+    set_access_timestamp(&op);
+    return ioctx->operate(oid, &op);
+}
+
+void get_modify_timestamp_start(librados::ObjectReadOperation *op) {
+  bufferlist empty_bl;
+  op->exec("rbd", "get_modify_timestamp", empty_bl);
+}
+
+int get_modify_timestamp_finish(bufferlist::const_iterator *it,
+                                  utime_t *timestamp) {
+  ceph_assert(timestamp);
+
+  try {
+    decode(*timestamp, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_modify_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                           utime_t *timestamp)
+{
+  librados::ObjectReadOperation op;
+  get_modify_timestamp_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_modify_timestamp_finish(&it, timestamp);
+}
+
+void set_modify_timestamp(librados::ObjectWriteOperation *op)
+{
+    bufferlist empty_bl;
+    op->exec("rbd","set_modify_timestamp",empty_bl);
+}
+
+int set_modify_timestamp(librados::IoCtx *ioctx, const std::string &oid)
+{
+    librados::ObjectWriteOperation op;
+    set_modify_timestamp(&op);
+    return ioctx->operate(oid, &op);
+}
+
+
+/************************ rbd_id object methods ************************/
+
+void get_id_start(librados::ObjectReadOperation *op) {
+  bufferlist empty_bl;
+  op->exec("rbd", "get_id", empty_bl);
+}
+
+int get_id_finish(bufferlist::const_iterator *it, std::string *id) {
+  try {
+    decode(*id, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int get_id(librados::IoCtx *ioctx, const std::string &oid, std::string *id)
+{
+  librados::ObjectReadOperation op;
+  get_id_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return get_id_finish(&it, id);
+}
+
+void set_id(librados::ObjectWriteOperation *op, const std::string &id)
+{
+  bufferlist bl;
+  encode(id, bl);
+  op->exec("rbd", "set_id", bl);
+}
+
+int set_id(librados::IoCtx *ioctx, const std::string &oid, const std::string &id)
+{
+  librados::ObjectWriteOperation op;
+  set_id(&op, id);
+
+  return ioctx->operate(oid, &op);
+}
+
+/******************** rbd_directory object methods ********************/
+
+void dir_get_id_start(librados::ObjectReadOperation *op,
+                      const std::string &image_name) {
+  bufferlist bl;
+  encode(image_name, bl);
+
+  op->exec("rbd", "dir_get_id", bl);
+}
+
+int dir_get_id_finish(bufferlist::const_iterator *iter, std::string *image_id) {
+  try {
+    decode(*image_id, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int dir_get_id(librados::IoCtx *ioctx, const std::string &oid,
+               const std::string &name, std::string *id) {
+  librados::ObjectReadOperation op;
+  dir_get_id_start(&op, name);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  return dir_get_id_finish(&iter, id);
+}
+
+void dir_get_name_start(librados::ObjectReadOperation *op,
+                        const std::string &id) {
+  bufferlist in_bl;
+  encode(id, in_bl);
+  op->exec("rbd", "dir_get_name", in_bl);
+}
+
+int dir_get_name_finish(bufferlist::const_iterator *it, std::string *name) {
+  try {
+    decode(*name, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int dir_get_name(librados::IoCtx *ioctx, const std::string &oid,
+                 const std::string &id, std::string *name) {
+  librados::ObjectReadOperation op;
+  dir_get_name_start(&op, id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return dir_get_name_finish(&it, name);
+}
+
+void dir_list_start(librados::ObjectReadOperation *op,
+                    const std::string &start, uint64_t max_return)
+{
+  bufferlist in_bl;
+  encode(start, in_bl);
+  encode(max_return, in_bl);
+
+  op->exec("rbd", "dir_list", in_bl);
+}
+
+int dir_list_finish(bufferlist::const_iterator *it, map<string, string> *images)
+{
+  try {
+    decode(*images, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int dir_list(librados::IoCtx *ioctx, const std::string &oid,
+             const std::string &start, uint64_t max_return,
+             map<string, string> *images)
+{
+  librados::ObjectReadOperation op;
+  dir_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  return dir_list_finish(&iter, images);
+}
+
+void dir_add_image(librados::ObjectWriteOperation *op,
+                   const std::string &name, const std::string &id)
+{
+  bufferlist bl;
+  encode(name, bl);
+  encode(id, bl);
+  op->exec("rbd", "dir_add_image", bl);
+}
+
+int dir_add_image(librados::IoCtx *ioctx, const std::string &oid,
+                  const std::string &name, const std::string &id)
+{
+  librados::ObjectWriteOperation op;
+  dir_add_image(&op, name, id);
+
+  return ioctx->operate(oid, &op);
+}
+
+int dir_remove_image(librados::IoCtx *ioctx, const std::string &oid,
+                     const std::string &name, const std::string &id)
+{
+  librados::ObjectWriteOperation op;
+  dir_remove_image(&op, name, id);
+
+  return ioctx->operate(oid, &op);
+}
+
+void dir_state_assert(librados::ObjectOperation *op,
+                      cls::rbd::DirectoryState directory_state)
+{
+  bufferlist bl;
+  encode(directory_state, bl);
+  op->exec("rbd", "dir_state_assert", bl);
+}
+
+int dir_state_assert(librados::IoCtx *ioctx, const std::string &oid,
+                     cls::rbd::DirectoryState directory_state)
+{
+  librados::ObjectWriteOperation op;
+  dir_state_assert(&op, directory_state);
+
+  return ioctx->operate(oid, &op);
+}
+
+void dir_state_set(librados::ObjectWriteOperation *op,
+                   cls::rbd::DirectoryState directory_state)
+{
+  bufferlist bl;
+  encode(directory_state, bl);
+  op->exec("rbd", "dir_state_set", bl);
+}
+
+int dir_state_set(librados::IoCtx *ioctx, const std::string &oid,
+                  cls::rbd::DirectoryState directory_state)
+{
+  librados::ObjectWriteOperation op;
+  dir_state_set(&op, directory_state);
+
+  return ioctx->operate(oid, &op);
+}
+
+void dir_remove_image(librados::ObjectWriteOperation *op,
+                      const std::string &name, const std::string &id)
+{
+  bufferlist bl;
+  encode(name, bl);
+  encode(id, bl);
+
+  op->exec("rbd", "dir_remove_image", bl);
+}
+
+void dir_rename_image(librados::ObjectWriteOperation *op,
+                      const std::string &src, const std::string &dest,
+                      const std::string &id)
+{
+  bufferlist in;
+  encode(src, in);
+  encode(dest, in);
+  encode(id, in);
+  op->exec("rbd", "dir_rename_image", in);
+}
+
+void object_map_load_start(librados::ObjectReadOperation *op) {
+  bufferlist in_bl;
+  op->exec("rbd", "object_map_load", in_bl);
+}
+
+int object_map_load_finish(bufferlist::const_iterator *it,
+                           ceph::BitVector<2> *object_map) {
+  try {
+    decode(*object_map, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int object_map_load(librados::IoCtx *ioctx, const std::string &oid,
+                    ceph::BitVector<2> *object_map)
+{
+  librados::ObjectReadOperation op;
+  object_map_load_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return object_map_load_finish(&it, object_map);
+}
+
+void object_map_save(librados::ObjectWriteOperation *rados_op,
+                     const ceph::BitVector<2> &object_map)
+{
+  ceph::BitVector<2> object_map_copy(object_map);
+  object_map_copy.set_crc_enabled(false);
+
+  bufferlist in;
+  encode(object_map_copy, in);
+  rados_op->exec("rbd", "object_map_save", in);
+}
+
+void object_map_resize(librados::ObjectWriteOperation *rados_op,
+                       uint64_t object_count, uint8_t default_state)
+{
+  bufferlist in;
+  encode(object_count, in);
+  encode(default_state, in);
+  rados_op->exec("rbd", "object_map_resize", in);
+}
+
+void object_map_update(librados::ObjectWriteOperation *rados_op,
+                       uint64_t start_object_no, uint64_t end_object_no,
+                       uint8_t new_object_state,
+                       const boost::optional<uint8_t> &current_object_state)
+{
+  bufferlist in;
+  encode(start_object_no, in);
+  encode(end_object_no, in);
+  encode(new_object_state, in);
+  encode(current_object_state, in);
+  rados_op->exec("rbd", "object_map_update", in);
+}
+
+void object_map_snap_add(librados::ObjectWriteOperation *rados_op)
+{
+  bufferlist in;
+  rados_op->exec("rbd", "object_map_snap_add", in);
+}
+
+void object_map_snap_remove(librados::ObjectWriteOperation *rados_op,
+                            const ceph::BitVector<2> &object_map)
+{
+  ceph::BitVector<2> object_map_copy(object_map);
+  object_map_copy.set_crc_enabled(false);
+
+  bufferlist in;
+  encode(object_map_copy, in);
+  rados_op->exec("rbd", "object_map_snap_remove", in);
+}
+
+void metadata_set(librados::ObjectWriteOperation *op,
+                  const map<string, bufferlist> &data)
+{
+  bufferlist bl;
+  encode(data, bl);
+
+  op->exec("rbd", "metadata_set", bl);
+}
+
+int metadata_set(librados::IoCtx *ioctx, const std::string &oid,
+                 const map<string, bufferlist> &data)
+{
+  librados::ObjectWriteOperation op;
+  metadata_set(&op, data);
+
+  return ioctx->operate(oid, &op);
+}
+
+void metadata_remove(librados::ObjectWriteOperation *op,
+                     const std::string &key)
+{
+  bufferlist bl;
+  encode(key, bl);
+
+  op->exec("rbd", "metadata_remove", bl);
+}
+
+int metadata_remove(librados::IoCtx *ioctx, const std::string &oid,
+                    const std::string &key)
+{
+  librados::ObjectWriteOperation op;
+  metadata_remove(&op, key);
+
+  return ioctx->operate(oid, &op);
+}
+
+int metadata_list(librados::IoCtx *ioctx, const std::string &oid,
+                  const std::string &start, uint64_t max_return,
+                  map<string, bufferlist> *pairs)
+{
+  librados::ObjectReadOperation op;
+  metadata_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return metadata_list_finish(&it, pairs);
+}
+
+void metadata_list_start(librados::ObjectReadOperation *op,
+                         const std::string &start, uint64_t max_return)
+{
+  bufferlist in_bl;
+  encode(start, in_bl);
+  encode(max_return, in_bl);
+  op->exec("rbd", "metadata_list", in_bl);
+}
+
+int metadata_list_finish(bufferlist::const_iterator *it,
+                         std::map<std::string, bufferlist> *pairs)
+{
+  ceph_assert(pairs);
+  try {
+    decode(*pairs, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int metadata_get(librados::IoCtx *ioctx, const std::string &oid,
+                 const std::string &key, string *s)
+{
+  ceph_assert(s);
+  bufferlist in, out;
+  encode(key, in);
+  int r = ioctx->exec(oid, "rbd", "metadata_get", in, out);
+  if (r < 0)
+    return r;
+
+  auto iter = out.cbegin();
+  try {
+    decode(*s, iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+void child_attach(librados::ObjectWriteOperation *op, snapid_t snap_id,
+                  const cls::rbd::ChildImageSpec& child_image)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  encode(child_image, bl);
+  op->exec("rbd", "child_attach", bl);
+}
+
+int child_attach(librados::IoCtx *ioctx, const std::string &oid,
+                 snapid_t snap_id,
+                 const cls::rbd::ChildImageSpec& child_image)
+{
+  librados::ObjectWriteOperation op;
+  child_attach(&op, snap_id, child_image);
+
+  int r = ioctx->operate(oid, &op);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void child_detach(librados::ObjectWriteOperation *op, snapid_t snap_id,
+                  const cls::rbd::ChildImageSpec& child_image)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  encode(child_image, bl);
+  op->exec("rbd", "child_detach", bl);
+}
+
+int child_detach(librados::IoCtx *ioctx, const std::string &oid,
+                 snapid_t snap_id,
+                 const cls::rbd::ChildImageSpec& child_image)
+{
+  librados::ObjectWriteOperation op;
+  child_detach(&op, snap_id, child_image);
+
+  int r = ioctx->operate(oid, &op);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void children_list_start(librados::ObjectReadOperation *op,
+                         snapid_t snap_id)
+{
+  bufferlist bl;
+  encode(snap_id, bl);
+  op->exec("rbd", "children_list", bl);
+}
+
+int children_list_finish(bufferlist::const_iterator *it,
+                         cls::rbd::ChildImageSpecs *child_images)
+{
+  child_images->clear();
+  try {
+    decode(*child_images, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int children_list(librados::IoCtx *ioctx, const std::string &oid,
+                  snapid_t snap_id,
+                  cls::rbd::ChildImageSpecs *child_images)
+{
+  librados::ObjectReadOperation op;
+  children_list_start(&op, snap_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  r = children_list_finish(&it, child_images);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int migration_set(librados::IoCtx *ioctx, const std::string &oid,
+                  const cls::rbd::MigrationSpec &migration_spec) {
+  librados::ObjectWriteOperation op;
+  migration_set(&op, migration_spec);
+  return ioctx->operate(oid, &op);
+}
+
+void migration_set(librados::ObjectWriteOperation *op,
+                   const cls::rbd::MigrationSpec &migration_spec) {
+  bufferlist bl;
+  encode(migration_spec, bl);
+  op->exec("rbd", "migration_set", bl);
+}
+
+int migration_set_state(librados::IoCtx *ioctx, const std::string &oid,
+                        cls::rbd::MigrationState state,
+                        const std::string &description) {
+  librados::ObjectWriteOperation op;
+  migration_set_state(&op, state, description);
+  return ioctx->operate(oid, &op);
+}
+
+void migration_set_state(librados::ObjectWriteOperation *op,
+                         cls::rbd::MigrationState state,
+                         const std::string &description) {
+  bufferlist bl;
+  encode(state, bl);
+  encode(description, bl);
+  op->exec("rbd", "migration_set_state", bl);
+}
+
+void migration_get_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "migration_get", bl);
+}
+
+int migration_get_finish(bufferlist::const_iterator *it,
+                         cls::rbd::MigrationSpec *migration_spec) {
+  try {
+    decode(*migration_spec, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int migration_get(librados::IoCtx *ioctx, const std::string &oid,
+                  cls::rbd::MigrationSpec *migration_spec) {
+  librados::ObjectReadOperation op;
+  migration_get_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = migration_get_finish(&iter, migration_spec);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int migration_remove(librados::IoCtx *ioctx, const std::string &oid) {
+  librados::ObjectWriteOperation op;
+  migration_remove(&op);
+  return ioctx->operate(oid, &op);
+}
+
+void migration_remove(librados::ObjectWriteOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "migration_remove", bl);
+}
+
+int assert_snapc_seq(librados::IoCtx *ioctx, const std::string &oid,
+                     uint64_t snapc_seq,
+                     cls::rbd::AssertSnapcSeqState state) {
+  librados::ObjectWriteOperation op;
+  assert_snapc_seq(&op, snapc_seq, state);
+  return ioctx->operate(oid, &op);
+}
+
+void assert_snapc_seq(librados::ObjectWriteOperation *op,
+                      uint64_t snapc_seq,
+                      cls::rbd::AssertSnapcSeqState state) {
+  bufferlist bl;
+  encode(snapc_seq, bl);
+  encode(state, bl);
+  op->exec("rbd", "assert_snapc_seq", bl);
+}
+
+void mirror_uuid_get_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "mirror_uuid_get", bl);
+}
+
+int mirror_uuid_get_finish(bufferlist::const_iterator *it,
+                           std::string *uuid) {
+  try {
+    decode(*uuid, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_uuid_get(librados::IoCtx *ioctx, std::string *uuid) {
+  librados::ObjectReadOperation op;
+  mirror_uuid_get_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  r = mirror_uuid_get_finish(&it, uuid);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_uuid_set(librados::IoCtx *ioctx, const std::string &uuid) {
+  bufferlist in_bl;
+  encode(uuid, in_bl);
+
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_uuid_set", in_bl,
+                      out_bl);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_mode_get_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "mirror_mode_get", bl);
+}
+
+int mirror_mode_get_finish(bufferlist::const_iterator *it,
+                           cls::rbd::MirrorMode *mirror_mode) {
+  try {
+    uint32_t mirror_mode_decode;
+    decode(mirror_mode_decode, *it);
+    *mirror_mode = static_cast<cls::rbd::MirrorMode>(mirror_mode_decode);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int mirror_mode_get(librados::IoCtx *ioctx,
+                    cls::rbd::MirrorMode *mirror_mode) {
+  librados::ObjectReadOperation op;
+  mirror_mode_get_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r == -ENOENT) {
+    *mirror_mode = cls::rbd::MIRROR_MODE_DISABLED;
+    return 0;
+  } else if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  r = mirror_mode_get_finish(&it, mirror_mode);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_mode_set(librados::IoCtx *ioctx,
+                    cls::rbd::MirrorMode mirror_mode) {
+  bufferlist in_bl;
+  encode(static_cast<uint32_t>(mirror_mode), in_bl);
+
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_mode_set", in_bl,
+                      out_bl);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_peer_list(librados::IoCtx *ioctx,
+                     std::vector<cls::rbd::MirrorPeer> *peers) {
+  bufferlist in_bl;
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_peer_list", in_bl,
+                      out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  peers->clear();
+  try {
+    auto bl_it = out_bl.cbegin();
+    decode(*peers, bl_it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_peer_add(librados::IoCtx *ioctx, const std::string &uuid,
+                    const std::string &cluster_name,
+                    const std::string &client_name) {
+  cls::rbd::MirrorPeer peer(uuid, cluster_name, client_name, -1);
+  bufferlist in_bl;
+  encode(peer, in_bl);
+
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_peer_add", in_bl,
+                      out_bl);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_peer_remove(librados::IoCtx *ioctx,
+                       const std::string &uuid) {
+  bufferlist in_bl;
+  encode(uuid, in_bl);
+
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_peer_remove", in_bl,
+                      out_bl);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_peer_set_client(librados::IoCtx *ioctx,
+                           const std::string &uuid,
+                           const std::string &client_name) {
+  bufferlist in_bl;
+  encode(uuid, in_bl);
+  encode(client_name, in_bl);
+
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_peer_set_client",
+                      in_bl, out_bl);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_peer_set_cluster(librados::IoCtx *ioctx,
+                            const std::string &uuid,
+                            const std::string &cluster_name) {
+  bufferlist in_bl;
+  encode(uuid, in_bl);
+  encode(cluster_name, in_bl);
+
+  bufferlist out_bl;
+  int r = ioctx->exec(RBD_MIRRORING, "rbd", "mirror_peer_set_cluster",
+                      in_bl, out_bl);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_list_start(librados::ObjectReadOperation *op,
+                             const std::string &start, uint64_t max_return)
+{
+  bufferlist in_bl;
+  encode(start, in_bl);
+  encode(max_return, in_bl);
+  op->exec("rbd", "mirror_image_list", in_bl);
+}
+
+int mirror_image_list_finish(bufferlist::const_iterator *it,
+                             std::map<string, string> *mirror_image_ids)
+{
+  try {
+    decode(*mirror_image_ids, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_list(librados::IoCtx *ioctx,
+                      const std::string &start, uint64_t max_return,
+                      std::map<std::string, std::string> *mirror_image_ids) {
+  librados::ObjectReadOperation op;
+  mirror_image_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto bl_it = out_bl.cbegin();
+  return mirror_image_list_finish(&bl_it, mirror_image_ids);
+}
+
+void mirror_image_get_image_id_start(librados::ObjectReadOperation *op,
+                                     const std::string &global_image_id) {
+  bufferlist in_bl;
+  encode(global_image_id, in_bl);
+  op->exec( "rbd", "mirror_image_get_image_id", in_bl);
+}
+
+int mirror_image_get_image_id_finish(bufferlist::const_iterator *it,
+                                     std::string *image_id) {
+  try {
+    decode(*image_id, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_get_image_id(librados::IoCtx *ioctx,
+                              const std::string &global_image_id,
+                              std::string *image_id) {
+  librados::ObjectReadOperation op;
+  mirror_image_get_image_id_start(&op, global_image_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return mirror_image_get_image_id_finish(&it, image_id);
+}
+
+int mirror_image_get(librados::IoCtx *ioctx, const std::string &image_id,
+                     cls::rbd::MirrorImage *mirror_image) {
+  librados::ObjectReadOperation op;
+  mirror_image_get_start(&op, image_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_image_get_finish(&iter, mirror_image);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_get_start(librados::ObjectReadOperation *op,
+                            const std::string &image_id) {
+  bufferlist in_bl;
+  encode(image_id, in_bl);
+
+  op->exec("rbd", "mirror_image_get", in_bl);
+}
+
+int mirror_image_get_finish(bufferlist::const_iterator *iter,
+                            cls::rbd::MirrorImage *mirror_image) {
+  try {
+    decode(*mirror_image, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+void mirror_image_set(librados::ObjectWriteOperation *op,
+                      const std::string &image_id,
+                      const cls::rbd::MirrorImage &mirror_image) {
+  bufferlist bl;
+  encode(image_id, bl);
+  encode(mirror_image, bl);
+
+  op->exec("rbd", "mirror_image_set", bl);
+}
+
+int mirror_image_set(librados::IoCtx *ioctx, const std::string &image_id,
+                     const cls::rbd::MirrorImage &mirror_image) {
+  librados::ObjectWriteOperation op;
+  mirror_image_set(&op, image_id, mirror_image);
+
+  int r = ioctx->operate(RBD_MIRRORING, &op);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_remove(librados::ObjectWriteOperation *op,
+                         const std::string &image_id) {
+  bufferlist bl;
+  encode(image_id, bl);
+
+  op->exec("rbd", "mirror_image_remove", bl);
+}
+
+int mirror_image_remove(librados::IoCtx *ioctx, const std::string &image_id) {
+  librados::ObjectWriteOperation op;
+  mirror_image_remove(&op, image_id);
+
+  int r = ioctx->operate(RBD_MIRRORING, &op);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+int mirror_image_status_set(librados::IoCtx *ioctx,
+                            const std::string &global_image_id,
+                            const cls::rbd::MirrorImageStatus &status) {
+  librados::ObjectWriteOperation op;
+  mirror_image_status_set(&op, global_image_id, status);
+  return ioctx->operate(RBD_MIRRORING, &op);
+}
+
+void mirror_image_status_set(librados::ObjectWriteOperation *op,
+                             const std::string &global_image_id,
+                             const cls::rbd::MirrorImageStatus &status) {
+  bufferlist bl;
+  encode(global_image_id, bl);
+  encode(status, bl);
+  op->exec("rbd", "mirror_image_status_set", bl);
+}
+
+int mirror_image_status_remove(librados::IoCtx *ioctx,
+                               const std::string &global_image_id) {
+  librados::ObjectWriteOperation op;
+  mirror_image_status_remove(&op, global_image_id);
+  return ioctx->operate(RBD_MIRRORING, &op);
+}
+
+void mirror_image_status_remove(librados::ObjectWriteOperation *op,
+                                const std::string &global_image_id) {
+  bufferlist bl;
+  encode(global_image_id, bl);
+  op->exec("rbd", "mirror_image_status_remove", bl);
+}
+
+int mirror_image_status_get(librados::IoCtx *ioctx,
+                            const std::string &global_image_id,
+                            cls::rbd::MirrorImageStatus *status) {
+  librados::ObjectReadOperation op;
+  mirror_image_status_get_start(&op, global_image_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_image_status_get_finish(&iter, status);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_status_get_start(librados::ObjectReadOperation *op,
+                                   const std::string &global_image_id) {
+  bufferlist bl;
+  encode(global_image_id, bl);
+  op->exec("rbd", "mirror_image_status_get", bl);
+}
+
+int mirror_image_status_get_finish(bufferlist::const_iterator *iter,
+                                   cls::rbd::MirrorImageStatus *status) {
+  try {
+    decode(*status, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_status_list(librados::IoCtx *ioctx,
+                             const std::string &start, uint64_t max_return,
+                             std::map<std::string, cls::rbd::MirrorImage> *images,
+                             std::map<std::string, cls::rbd::MirrorImageStatus> *statuses) {
+  librados::ObjectReadOperation op;
+  mirror_image_status_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_image_status_list_finish(&iter, images, statuses);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_status_list_start(librados::ObjectReadOperation *op,
+                                    const std::string &start,
+                                    uint64_t max_return) {
+  bufferlist bl;
+  encode(start, bl);
+  encode(max_return, bl);
+  op->exec("rbd", "mirror_image_status_list", bl);
+}
+
+int mirror_image_status_list_finish(bufferlist::const_iterator *iter,
+                                    std::map<std::string, cls::rbd::MirrorImage> *images,
+                                    std::map<std::string, cls::rbd::MirrorImageStatus> *statuses) {
+  images->clear();
+  statuses->clear();
+  try {
+    decode(*images, *iter);
+    decode(*statuses, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_status_get_summary(librados::IoCtx *ioctx,
+                                    std::map<cls::rbd::MirrorImageStatusState, int> *states) {
+  librados::ObjectReadOperation op;
+  mirror_image_status_get_summary_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_image_status_get_summary_finish(&iter, states);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_status_get_summary_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "mirror_image_status_get_summary", bl);
+}
+
+int mirror_image_status_get_summary_finish(bufferlist::const_iterator *iter,
+                                           std::map<cls::rbd::MirrorImageStatusState, int> *states) {
+  try {
+    decode(*states, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_status_remove_down(librados::IoCtx *ioctx) {
+  librados::ObjectWriteOperation op;
+  mirror_image_status_remove_down(&op);
+  return ioctx->operate(RBD_MIRRORING, &op);
+}
+
+void mirror_image_status_remove_down(librados::ObjectWriteOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "mirror_image_status_remove_down", bl);
+}
+
+int mirror_image_instance_get(librados::IoCtx *ioctx,
+                              const std::string &global_image_id,
+                              entity_inst_t *instance) {
+  librados::ObjectReadOperation op;
+  mirror_image_instance_get_start(&op, global_image_id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_image_instance_get_finish(&iter, instance);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_instance_get_start(librados::ObjectReadOperation *op,
+                                     const std::string &global_image_id) {
+  bufferlist bl;
+  encode(global_image_id, bl);
+  op->exec("rbd", "mirror_image_instance_get", bl);
+}
+
+int mirror_image_instance_get_finish(bufferlist::const_iterator *iter,
+                                     entity_inst_t *instance) {
+  try {
+    decode(*instance, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_instance_list(
+    librados::IoCtx *ioctx, const std::string &start, uint64_t max_return,
+    std::map<std::string, entity_inst_t> *instances) {
+  librados::ObjectReadOperation op;
+  mirror_image_instance_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_image_instance_list_finish(&iter, instances);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_image_instance_list_start(librados::ObjectReadOperation *op,
+                                      const std::string &start,
+                                      uint64_t max_return) {
+  bufferlist bl;
+  encode(start, bl);
+  encode(max_return, bl);
+  op->exec("rbd", "mirror_image_instance_list", bl);
+}
+
+int mirror_image_instance_list_finish(
+    bufferlist::const_iterator *iter,
+    std::map<std::string, entity_inst_t> *instances) {
+  instances->clear();
+  try {
+    decode(*instances, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+void mirror_instances_list_start(librados::ObjectReadOperation *op) {
+  bufferlist bl;
+  op->exec("rbd", "mirror_instances_list", bl);
+}
+
+int mirror_instances_list_finish(bufferlist::const_iterator *iter,
+                                 std::vector<std::string> *instance_ids) {
+  instance_ids->clear();
+  try {
+    decode(*instance_ids, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_instances_list(librados::IoCtx *ioctx,
+                          std::vector<std::string> *instance_ids) {
+  librados::ObjectReadOperation op;
+  mirror_instances_list_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRROR_LEADER, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  r = mirror_instances_list_finish(&iter, instance_ids);
+  if (r < 0) {
+    return r;
+  }
+  return 0;
+}
+
+void mirror_instances_add(librados::ObjectWriteOperation *op,
+                          const std::string &instance_id) {
+  bufferlist bl;
+  encode(instance_id, bl);
+  op->exec("rbd", "mirror_instances_add", bl);
+}
+
+int mirror_instances_add(librados::IoCtx *ioctx,
+                         const std::string &instance_id) {
+  librados::ObjectWriteOperation op;
+  mirror_instances_add(&op, instance_id);
+  return ioctx->operate(RBD_MIRROR_LEADER, &op);
+}
+
+void mirror_instances_remove(librados::ObjectWriteOperation *op,
+                             const std::string &instance_id) {
+  bufferlist bl;
+  encode(instance_id, bl);
+  op->exec("rbd", "mirror_instances_remove", bl);
+}
+
+int mirror_instances_remove(librados::IoCtx *ioctx,
+                            const std::string &instance_id) {
+  librados::ObjectWriteOperation op;
+  mirror_instances_remove(&op, instance_id);
+  return ioctx->operate(RBD_MIRROR_LEADER, &op);
+}
+
+void mirror_image_map_list_start(librados::ObjectReadOperation *op,
+                                 const std::string &start_after,
+                                 uint64_t max_read) {
+  bufferlist bl;
+  encode(start_after, bl);
+  encode(max_read, bl);
+
+  op->exec("rbd", "mirror_image_map_list", bl);
+}
+
+int mirror_image_map_list_finish(bufferlist::const_iterator *iter,
+                                 std::map<std::string, cls::rbd::MirrorImageMap> *image_mapping) {
+  try {
+    decode(*image_mapping, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int mirror_image_map_list(
+    librados::IoCtx *ioctx, const std::string &start_after,
+    uint64_t max_read,
+    std::map<std::string, cls::rbd::MirrorImageMap> *image_mapping) {
+  librados::ObjectReadOperation op;
+  mirror_image_map_list_start(&op, start_after, max_read);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_MIRRORING, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  return mirror_image_map_list_finish(&iter, image_mapping);
+}
+
+void mirror_image_map_update(librados::ObjectWriteOperation *op,
+                             const std::string &global_image_id,
+                             const cls::rbd::MirrorImageMap &image_map) {
+  bufferlist bl;
+  encode(global_image_id, bl);
+  encode(image_map, bl);
+
+  op->exec("rbd", "mirror_image_map_update", bl);
+}
+
+void mirror_image_map_remove(librados::ObjectWriteOperation *op,
+                             const std::string &global_image_id) {
+  bufferlist bl;
+  encode(global_image_id, bl);
+
+  op->exec("rbd", "mirror_image_map_remove", bl);
+}
+
+// Groups functions
+int group_dir_list(librados::IoCtx *ioctx, const std::string &oid,
+                   const std::string &start, uint64_t max_return,
+                   map<string, string> *cgs)
+{
+  bufferlist in, out;
+  encode(start, in);
+  encode(max_return, in);
+  int r = ioctx->exec(oid, "rbd", "group_dir_list", in, out);
+  if (r < 0)
+    return r;
+
+  auto iter = out.cbegin();
+  try {
+    decode(*cgs, iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int group_dir_add(librados::IoCtx *ioctx, const std::string &oid,
+                  const std::string &name, const std::string &id)
+{
+  bufferlist in, out;
+  encode(name, in);
+  encode(id, in);
+  return ioctx->exec(oid, "rbd", "group_dir_add", in, out);
+}
+
+int group_dir_rename(librados::IoCtx *ioctx, const std::string &oid,
+                     const std::string &src, const std::string &dest,
+                     const std::string &id)
+{
+  bufferlist in, out;
+  encode(src, in);
+  encode(dest, in);
+  encode(id, in);
+  return ioctx->exec(oid, "rbd", "group_dir_rename", in, out);
+}
+
+int group_dir_remove(librados::IoCtx *ioctx, const std::string &oid,
+                     const std::string &name, const std::string &id)
+{
+  bufferlist in, out;
+  encode(name, in);
+  encode(id, in);
+  return ioctx->exec(oid, "rbd", "group_dir_remove", in, out);
+}
+
+int group_image_remove(librados::IoCtx *ioctx, const std::string &oid,
+                       const cls::rbd::GroupImageSpec &spec)
+{
+  bufferlist bl, bl2;
+  encode(spec, bl);
+
+  return ioctx->exec(oid, "rbd", "group_image_remove", bl, bl2);
+}
+
+int group_image_list(librados::IoCtx *ioctx,
+                     const std::string &oid,
+                     const cls::rbd::GroupImageSpec &start,
+                     uint64_t max_return,
+                     std::vector<cls::rbd::GroupImageStatus> *images)
+{
+  bufferlist bl, bl2;
+  encode(start, bl);
+  encode(max_return, bl);
+
+  int r = ioctx->exec(oid, "rbd", "group_image_list", bl, bl2);
+  if (r < 0)
+    return r;
+
+  auto iter = bl2.cbegin();
+  try {
+    decode(*images, iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int group_image_set(librados::IoCtx *ioctx, const std::string &oid,
+                    const cls::rbd::GroupImageStatus &st)
+{
+  bufferlist bl, bl2;
+  encode(st, bl);
+
+  return ioctx->exec(oid, "rbd", "group_image_set", bl, bl2);
+}
+
+int image_group_add(librados::IoCtx *ioctx, const std::string &oid,
+                    const cls::rbd::GroupSpec &group_spec)
+{
+  bufferlist bl, bl2;
+  encode(group_spec, bl);
+
+  return ioctx->exec(oid, "rbd", "image_group_add", bl, bl2);
+}
+
+int image_group_remove(librados::IoCtx *ioctx, const std::string &oid,
+                       const cls::rbd::GroupSpec &group_spec)
+{
+  bufferlist bl, bl2;
+  encode(group_spec, bl);
+
+  return ioctx->exec(oid, "rbd", "image_group_remove", bl, bl2);
+}
+
+void image_group_get_start(librados::ObjectReadOperation *op)
+{
+  bufferlist in_bl;
+  op->exec("rbd", "image_group_get", in_bl);
+}
+
+int image_group_get_finish(bufferlist::const_iterator *iter,
+                           cls::rbd::GroupSpec *group_spec)
+{
+  try {
+    decode(*group_spec, *iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+  return 0;
+}
+
+int image_group_get(librados::IoCtx *ioctx, const std::string &oid,
+                    cls::rbd::GroupSpec *group_spec)
+{
+  librados::ObjectReadOperation op;
+  image_group_get_start(&op);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(oid, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  return image_group_get_finish(&iter, group_spec);
+}
+
+int group_snap_set(librados::IoCtx *ioctx, const std::string &oid,
+                   const cls::rbd::GroupSnapshot &snapshot)
+{
+  using ceph::encode;
+  bufferlist inbl, outbl;
+  encode(snapshot, inbl);
+  int r = ioctx->exec(oid, "rbd", "group_snap_set", inbl, outbl);
+  return r;
+}
+
+int group_snap_remove(librados::IoCtx *ioctx, const std::string &oid,
+                      const std::string &snap_id)
+{
+  using ceph::encode;
+  bufferlist inbl, outbl;
+  encode(snap_id, inbl);
+  return ioctx->exec(oid, "rbd", "group_snap_remove", inbl, outbl);
+}
+
+int group_snap_get_by_id(librados::IoCtx *ioctx, const std::string &oid,
+                         const std::string &snap_id,
+                         cls::rbd::GroupSnapshot *snapshot)
+{
+  using ceph::encode;
+  using ceph::decode;
+  bufferlist inbl, outbl;
+
+  encode(snap_id, inbl);
+  int r = ioctx->exec(oid, "rbd", "group_snap_get_by_id", inbl, outbl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = outbl.cbegin();
+  try {
+    decode(*snapshot, iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+int group_snap_list(librados::IoCtx *ioctx, const std::string &oid,
+                    const cls::rbd::GroupSnapshot &start,
+                    uint64_t max_return,
+                    std::vector<cls::rbd::GroupSnapshot> *snapshots)
+{
+  using ceph::encode;
+  using ceph::decode;
+  bufferlist inbl, outbl;
+  encode(start, inbl);
+  encode(max_return, inbl);
+
+  int r = ioctx->exec(oid, "rbd", "group_snap_list", inbl, outbl);
+  if (r < 0) {
+    return r;
+  }
+  auto iter = outbl.cbegin();
+  try {
+    decode(*snapshots, iter);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+// rbd_trash functions
+void trash_add(librados::ObjectWriteOperation *op,
+               const std::string &id,
+               const cls::rbd::TrashImageSpec &trash_spec)
+{
+  bufferlist bl;
+  encode(id, bl);
+  encode(trash_spec, bl);
+  op->exec("rbd", "trash_add", bl);
+}
+
+int trash_add(librados::IoCtx *ioctx, const std::string &id,
+              const cls::rbd::TrashImageSpec &trash_spec)
+{
+  librados::ObjectWriteOperation op;
+  trash_add(&op, id, trash_spec);
+
+  return ioctx->operate(RBD_TRASH, &op);
+}
+
+void trash_remove(librados::ObjectWriteOperation *op,
+                  const std::string &id)
+{
+  bufferlist bl;
+  encode(id, bl);
+  op->exec("rbd", "trash_remove", bl);
+}
+
+int trash_remove(librados::IoCtx *ioctx, const std::string &id)
+{
+  librados::ObjectWriteOperation op;
+  trash_remove(&op, id);
+
+  return ioctx->operate(RBD_TRASH, &op);
+}
+
+void trash_list_start(librados::ObjectReadOperation *op,
+                      const std::string &start, uint64_t max_return)
+{
+  bufferlist bl;
+  encode(start, bl);
+  encode(max_return, bl);
+  op->exec("rbd", "trash_list", bl);
+}
+
+int trash_list_finish(bufferlist::const_iterator *it,
+                      map<string, cls::rbd::TrashImageSpec> *entries)
+{
+  ceph_assert(entries);
+
+  try {
+    decode(*entries, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int trash_list(librados::IoCtx *ioctx,
+               const std::string &start, uint64_t max_return,
+               map<string, cls::rbd::TrashImageSpec> *entries)
+{
+  librados::ObjectReadOperation op;
+  trash_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_TRASH, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  return trash_list_finish(&iter, entries);
+}
+
+void trash_get_start(librados::ObjectReadOperation *op,
+                     const std::string &id)
+{
+  bufferlist bl;
+  encode(id, bl);
+  op->exec("rbd", "trash_get", bl);
+}
+
+int trash_get_finish(bufferlist::const_iterator *it,
+                     cls::rbd::TrashImageSpec *trash_spec) {
+  ceph_assert(trash_spec);
+  try {
+    decode(*trash_spec, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int trash_get(librados::IoCtx *ioctx, const std::string &id,
+              cls::rbd::TrashImageSpec *trash_spec)
+{
+  librados::ObjectReadOperation op;
+  trash_get_start(&op, id);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_TRASH, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto it = out_bl.cbegin();
+  return trash_get_finish(&it, trash_spec);
+}
+
+void trash_state_set(librados::ObjectWriteOperation *op,
+                     const std::string &id,
+                     const cls::rbd::TrashImageState &trash_state,
+                     const cls::rbd::TrashImageState &expect_state)
+{
+  bufferlist bl;
+  encode(id, bl);
+  encode(trash_state, bl);
+  encode(expect_state, bl);
+  op->exec("rbd", "trash_state_set", bl);
+}
+
+int trash_state_set(librados::IoCtx *ioctx, const std::string &id,
+                    const cls::rbd::TrashImageState &trash_state,
+                    const cls::rbd::TrashImageState &expect_state)
+{
+  librados::ObjectWriteOperation op;
+  trash_state_set(&op, id, trash_state, expect_state);
+
+  return ioctx->operate(RBD_TRASH, &op);
+}
+
+void namespace_add(librados::ObjectWriteOperation *op,
+                   const std::string &name)
+{
+  bufferlist bl;
+  encode(name, bl);
+  op->exec("rbd", "namespace_add", bl);
+}
+
+int namespace_add(librados::IoCtx *ioctx, const std::string &name)
+{
+  librados::ObjectWriteOperation op;
+  namespace_add(&op, name);
+
+  return ioctx->operate(RBD_NAMESPACE, &op);
+}
+
+void namespace_remove(librados::ObjectWriteOperation *op,
+                      const std::string &name)
+{
+  bufferlist bl;
+  encode(name, bl);
+  op->exec("rbd", "namespace_remove", bl);
+}
+
+int namespace_remove(librados::IoCtx *ioctx, const std::string &name)
+{
+  librados::ObjectWriteOperation op;
+  namespace_remove(&op, name);
+
+  return ioctx->operate(RBD_NAMESPACE, &op);
+}
+
+void namespace_list_start(librados::ObjectReadOperation *op,
+                          const std::string &start, uint64_t max_return)
+{
+  bufferlist bl;
+  encode(start, bl);
+  encode(max_return, bl);
+  op->exec("rbd", "namespace_list", bl);
+}
+
+int namespace_list_finish(bufferlist::const_iterator *it,
+                          std::list<std::string> *entries)
+{
+  ceph_assert(entries);
+
+  try {
+    decode(*entries, *it);
+  } catch (const buffer::error &err) {
+    return -EBADMSG;
+  }
+
+  return 0;
+}
+
+int namespace_list(librados::IoCtx *ioctx,
+                   const std::string &start, uint64_t max_return,
+                   std::list<std::string> *entries)
+{
+  librados::ObjectReadOperation op;
+  namespace_list_start(&op, start, max_return);
+
+  bufferlist out_bl;
+  int r = ioctx->operate(RBD_NAMESPACE, &op, &out_bl);
+  if (r < 0) {
+    return r;
+  }
+
+  auto iter = out_bl.cbegin();
+  return namespace_list_finish(&iter, entries);
+}
+
+void sparsify(librados::ObjectWriteOperation *op, size_t sparse_size,
+              bool remove_empty)
+{
+  bufferlist bl;
+  encode(sparse_size, bl);
+  encode(remove_empty, bl);
+  op->exec("rbd", "sparsify", bl);
+}
+
+int sparsify(librados::IoCtx *ioctx, const std::string &oid, size_t sparse_size,
+             bool remove_empty)
+{
+  librados::ObjectWriteOperation op;
+  sparsify(&op, sparse_size, remove_empty);
+
+  return ioctx->operate(oid, &op);
+}
+
+} // namespace cls_client
+} // namespace librbd
diff --git a/src/cls/rbd/cls_rbd_client.h b/src/cls/rbd/cls_rbd_client.h
new file mode 100644
index 00000000..27a64cc2
--- /dev/null
+++ b/src/cls/rbd/cls_rbd_client.h
@@ -0,0 +1,610 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_LIBRBD_CLS_RBD_CLIENT_H
+#define CEPH_LIBRBD_CLS_RBD_CLIENT_H
+
+#include "cls/lock/cls_lock_types.h"
+#include "cls/rbd/cls_rbd_types.h"
+#include "common/snap_types.h"
+#include "include/types.h"
+#include "include/rados/librados_fwd.hpp"
+
+class Context;
+namespace ceph { template <uint8_t> class BitVector; }
+
+namespace librbd {
+namespace cls_client {
+
+// low-level interface (mainly for testing)
+void create_image(librados::ObjectWriteOperation *op, uint64_t size,
+                  uint8_t order, uint64_t features,
+                  const std::string &object_prefix, int64_t data_pool_id);
+int create_image(librados::IoCtx *ioctx, const std::string &oid,
+                 uint64_t size, uint8_t order, uint64_t features,
+                 const std::string &object_prefix, int64_t data_pool_id);
+
+void get_features_start(librados::ObjectReadOperation *op, bool read_only);
+int get_features_finish(bufferlist::const_iterator *it, uint64_t *features,
+                        uint64_t *incompatible_features);
+int get_features(librados::IoCtx *ioctx, const std::string &oid,
+                 bool read_only, uint64_t *features,
+                 uint64_t *incompatible_features);
+void set_features(librados::ObjectWriteOperation *op, uint64_t features,
+                  uint64_t mask);
+int set_features(librados::IoCtx *ioctx, const std::string &oid,
+                 uint64_t features, uint64_t mask);
+
+void get_object_prefix_start(librados::ObjectReadOperation *op);
+int get_object_prefix_finish(bufferlist::const_iterator *it,
+                             std::string *object_prefix);
+int get_object_prefix(librados::IoCtx *ioctx, const std::string &oid,
+                      std::string *object_prefix);
+
+void get_data_pool_start(librados::ObjectReadOperation *op);
+int get_data_pool_finish(bufferlist::const_iterator *it, int64_t *data_pool_id);
+int get_data_pool(librados::IoCtx *ioctx, const std::string &oid,
+                  int64_t *data_pool_id);
+
+void get_size_start(librados::ObjectReadOperation *op, snapid_t snap_id);
+int get_size_finish(bufferlist::const_iterator *it, uint64_t *size,
+                    uint8_t *order);
+int get_size(librados::IoCtx *ioctx, const std::string &oid,
+             snapid_t snap_id, uint64_t *size, uint8_t *order);
+int set_size(librados::IoCtx *ioctx, const std::string &oid,
+             uint64_t size);
+void set_size(librados::ObjectWriteOperation *op, uint64_t size);
+
+void get_flags_start(librados::ObjectReadOperation *op, snapid_t snap_id);
+int get_flags_finish(bufferlist::const_iterator *it, uint64_t *flags);
+int get_flags(librados::IoCtx *ioctx, const std::string &oid,
+              snapid_t snap_id, uint64_t *flags);
+
+void set_flags(librados::ObjectWriteOperation *op, snapid_t snap_id,
+               uint64_t flags, uint64_t mask);
+
+void op_features_get_start(librados::ObjectReadOperation *op);
+int op_features_get_finish(bufferlist::const_iterator *it,
+                           uint64_t *op_features);
+int op_features_get(librados::IoCtx *ioctx, const std::string &oid,
+                    uint64_t *op_features);
+void op_features_set(librados::ObjectWriteOperation *op,
+                     uint64_t op_features, uint64_t mask);
+int op_features_set(librados::IoCtx *ioctx, const std::string &oid,
+                    uint64_t op_features, uint64_t mask);
+
+// NOTE: deprecate v1 parent APIs after mimic EOLed
+void get_parent_start(librados::ObjectReadOperation *op, snapid_t snap_id);
+int get_parent_finish(bufferlist::const_iterator *it,
+                      cls::rbd::ParentImageSpec *pspec,
+                      uint64_t *parent_overlap);
+int get_parent(librados::IoCtx *ioctx, const std::string &oid,
+               snapid_t snap_id, cls::rbd::ParentImageSpec *pspec,
+               uint64_t *parent_overlap);
+int set_parent(librados::IoCtx *ioctx, const std::string &oid,
+               const cls::rbd::ParentImageSpec &pspec, uint64_t parent_overlap);
+void set_parent(librados::ObjectWriteOperation *op,
+                const cls::rbd::ParentImageSpec &pspec,
+                uint64_t parent_overlap);
+int remove_parent(librados::IoCtx *ioctx, const std::string &oid);
+void remove_parent(librados::ObjectWriteOperation *op);
+
+// v2 parent APIs
+void parent_get_start(librados::ObjectReadOperation* op);
+int parent_get_finish(bufferlist::const_iterator* it,
+                      cls::rbd::ParentImageSpec* parent_image_spec);
+int parent_get(librados::IoCtx* ioctx, const std::string &oid,
+               cls::rbd::ParentImageSpec* parent_image_spec);
+
+void parent_overlap_get_start(librados::ObjectReadOperation* op,
+                              snapid_t snap_id);
+int parent_overlap_get_finish(bufferlist::const_iterator* it,
+                              std::optional<uint64_t>* parent_overlap);
+int parent_overlap_get(librados::IoCtx* ioctx, const std::string &oid,
+                       snapid_t snap_id,
+                       std::optional<uint64_t>* parent_overlap);
+
+void parent_attach(librados::ObjectWriteOperation* op,
+                   const cls::rbd::ParentImageSpec& parent_image_spec,
+                   uint64_t parent_overlap, bool reattach);
+int parent_attach(librados::IoCtx *ioctx, const std::string &oid,
+                  const cls::rbd::ParentImageSpec& parent_image_spec,
+                  uint64_t parent_overlap, bool reattach);
+
+void parent_detach(librados::ObjectWriteOperation* op);
+int parent_detach(librados::IoCtx *ioctx, const std::string &oid);
+
+int add_child(librados::IoCtx *ioctx, const std::string &oid,
+              const cls::rbd::ParentImageSpec &pspec,
+              const std::string &c_imageid);
+void add_child(librados::ObjectWriteOperation *op,
+               const cls::rbd::ParentImageSpec& pspec,
+               const std::string &c_imageid);
+void remove_child(librados::ObjectWriteOperation *op,
+                  const cls::rbd::ParentImageSpec &pspec,
+                  const std::string &c_imageid);
+int remove_child(librados::IoCtx *ioctx, const std::string &oid,
+                 const cls::rbd::ParentImageSpec &pspec,
+                 const std::string &c_imageid);
+void get_children_start(librados::ObjectReadOperation *op,
+                        const cls::rbd::ParentImageSpec &pspec);
+int get_children_finish(bufferlist::const_iterator *it,
+                        std::set<string> *children);
+int get_children(librados::IoCtx *ioctx, const std::string &oid,
+                 const cls::rbd::ParentImageSpec& pspec, set<string>& children);
+
+void snapshot_get_start(librados::ObjectReadOperation* op,
+                        snapid_t snap_id);
+int snapshot_get_finish(bufferlist::const_iterator* it,
+                        cls::rbd::SnapshotInfo* snap_info);
+int snapshot_get(librados::IoCtx *ioctx, const std::string &oid,
+                 snapid_t snap_id, cls::rbd::SnapshotInfo* snap_info);
+
+void snapshot_add(librados::ObjectWriteOperation *op, snapid_t snap_id,
+                  const std::string &snap_name,
+                  const cls::rbd::SnapshotNamespace &snap_namespace);
+void snapshot_remove(librados::ObjectWriteOperation *op, snapid_t snap_id);
+void snapshot_rename(librados::ObjectWriteOperation *op,
+                     snapid_t src_snap_id,
+                     const std::string &dst_name);
+void snapshot_trash_add(librados::ObjectWriteOperation *op,
+                        snapid_t snap_id);
+
+void get_snapcontext_start(librados::ObjectReadOperation *op);
+int get_snapcontext_finish(bufferlist::const_iterator *it,
+                           ::SnapContext *snapc);
+int get_snapcontext(librados::IoCtx *ioctx, const std::string &oid,
+                    ::SnapContext *snapc);
+
+/// NOTE: remove after Luminous is retired
+void get_snapshot_name_start(librados::ObjectReadOperation *op,
+                             snapid_t snap_id);
+int get_snapshot_name_finish(bufferlist::const_iterator *it,
+                             std::string *name);
+int get_snapshot_name(librados::IoCtx *ioctx, const std::string &oid,
+                      snapid_t snap_id, std::string *name);
+
+/// NOTE: remove after Luminous is retired
+void get_snapshot_timestamp_start(librados::ObjectReadOperation *op,
+                                  snapid_t snap_id);
+int get_snapshot_timestamp_finish(bufferlist::const_iterator *it,
+                                  utime_t *timestamp);
+int get_snapshot_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                           snapid_t snap_id, utime_t *timestamp);
+
+void get_all_features_start(librados::ObjectReadOperation *op);
+int get_all_features_finish(bufferlist::const_iterator *it,
+                            uint64_t *all_features);
+int get_all_features(librados::IoCtx *ioctx, const std::string &oid,
+                     uint64_t *all_features);
+
+/// NOTE: remove protection after clone v1 is retired
+void get_protection_status_start(librados::ObjectReadOperation *op,
+                                 snapid_t snap_id);
+int get_protection_status_finish(bufferlist::const_iterator *it,
+                                 uint8_t *protection_status);
+int get_protection_status(librados::IoCtx *ioctx, const std::string &oid,
+                          snapid_t snap_id, uint8_t *protection_status);
+
+int set_protection_status(librados::IoCtx *ioctx, const std::string &oid,
+                          snapid_t snap_id, uint8_t protection_status);
+void set_protection_status(librados::ObjectWriteOperation *op,
+                           snapid_t snap_id, uint8_t protection_status);
+
+int snapshot_get_limit(librados::IoCtx *ioctx, const std::string &oid,
+                       uint64_t *limit);
+void snapshot_set_limit(librados::ObjectWriteOperation *op,
+                        uint64_t limit);
+
+void get_stripe_unit_count_start(librados::ObjectReadOperation *op);
+int get_stripe_unit_count_finish(bufferlist::const_iterator *it,
+                                 uint64_t *stripe_unit,
+                                 uint64_t *stripe_count);
+int get_stripe_unit_count(librados::IoCtx *ioctx, const std::string &oid,
+                          uint64_t *stripe_unit, uint64_t *stripe_count);
+
+void set_stripe_unit_count(librados::ObjectWriteOperation *op,
+                           uint64_t stripe_unit, uint64_t stripe_count);
+int set_stripe_unit_count(librados::IoCtx *ioctx, const std::string &oid,
+                          uint64_t stripe_unit, uint64_t stripe_count);
+
+void get_create_timestamp_start(librados::ObjectReadOperation *op);
+int get_create_timestamp_finish(bufferlist::const_iterator *it,
+                                utime_t *timestamp);
+int get_create_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                         utime_t *timestamp);
+
+void get_access_timestamp_start(librados::ObjectReadOperation *op);
+int get_access_timestamp_finish(bufferlist::const_iterator *it,
+                                utime_t *timestamp);
+int get_access_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                         utime_t *timestamp);
+
+void set_access_timestamp(librados::ObjectWriteOperation *op);
+int set_access_timestamp(librados::IoCtx *ioctx, const std::string &oid);
+
+void get_modify_timestamp_start(librados::ObjectReadOperation *op);
+int get_modify_timestamp_finish(bufferlist::const_iterator *it,
+                                utime_t *timestamp);
+int get_modify_timestamp(librados::IoCtx *ioctx, const std::string &oid,
+                         utime_t *timestamp);
+
+void set_modify_timestamp(librados::ObjectWriteOperation *op);
+int set_modify_timestamp(librados::IoCtx *ioctx, const std::string &oid);
+
+int metadata_list(librados::IoCtx *ioctx, const std::string &oid,
+                  const std::string &start, uint64_t max_return,
+                  map<string, bufferlist> *pairs);
+void metadata_list_start(librados::ObjectReadOperation *op,
+                         const std::string &start, uint64_t max_return);
+int metadata_list_finish(bufferlist::const_iterator *it,
+                         std::map<std::string, bufferlist> *pairs);
+void metadata_set(librados::ObjectWriteOperation *op,
+                  const map<std::string, bufferlist> &data);
+int metadata_set(librados::IoCtx *ioctx, const std::string &oid,
+                 const map<std::string, bufferlist> &data);
+void metadata_remove(librados::ObjectWriteOperation *op,
+                     const std::string &key);
+int metadata_remove(librados::IoCtx *ioctx, const std::string &oid,
+                    const std::string &key);
+int metadata_get(librados::IoCtx *ioctx, const std::string &oid,
+                 const std::string &key, string *v);
+
+void child_attach(librados::ObjectWriteOperation *op, snapid_t snap_id,
+                  const cls::rbd::ChildImageSpec& child_image);
+int child_attach(librados::IoCtx *ioctx, const std::string &oid,
+                 snapid_t snap_id,
+                 const cls::rbd::ChildImageSpec& child_image);
+void child_detach(librados::ObjectWriteOperation *op, snapid_t snap_id,
+                  const cls::rbd::ChildImageSpec& child_image);
+int child_detach(librados::IoCtx *ioctx, const std::string &oid,
+                 snapid_t snap_id,
+                 const cls::rbd::ChildImageSpec& child_image);
+void children_list_start(librados::ObjectReadOperation *op,
+                         snapid_t snap_id);
+int children_list_finish(bufferlist::const_iterator *it,
+                         cls::rbd::ChildImageSpecs *child_images);
+int children_list(librados::IoCtx *ioctx, const std::string &oid,
+                  snapid_t snap_id,
+                  cls::rbd::ChildImageSpecs *child_images);
+int migration_set(librados::IoCtx *ioctx, const std::string &oid,
+                  const cls::rbd::MigrationSpec &migration_spec);
+void migration_set(librados::ObjectWriteOperation *op,
+                   const cls::rbd::MigrationSpec &migration_spec);
+int migration_set_state(librados::IoCtx *ioctx, const std::string &oid,
+                        cls::rbd::MigrationState state,
+                        const std::string &description);
+void migration_set_state(librados::ObjectWriteOperation *op,
+                         cls::rbd::MigrationState state,
+                         const std::string &description);
+void migration_get_start(librados::ObjectReadOperation *op);
+int migration_get_finish(bufferlist::const_iterator *it,
+                         cls::rbd::MigrationSpec *migration_spec);
+int migration_get(librados::IoCtx *ioctx, const std::string &oid,
+                  cls::rbd::MigrationSpec *migration_spec);
+int migration_remove(librados::IoCtx *ioctx, const std::string &oid);
+void migration_remove(librados::ObjectWriteOperation *op);
+
+// operations on rbd_id objects
+void get_id_start(librados::ObjectReadOperation *op);
+int get_id_finish(bufferlist::const_iterator *it, std::string *id);
+int get_id(librados::IoCtx *ioctx, const std::string &oid, std::string *id);
+
+void set_id(librados::ObjectWriteOperation *op, const std::string &id);
+int set_id(librados::IoCtx *ioctx, const std::string &oid, const std::string &id);
+
+// operations on rbd_directory objects
+int dir_get_id(librados::IoCtx *ioctx, const std::string &oid,
+               const std::string &name, std::string *id);
+void dir_get_id_start(librados::ObjectReadOperation *op,
+                      const std::string &image_name);
+int dir_get_id_finish(bufferlist::const_iterator *iter, std::string *image_id);
+void dir_get_name_start(librados::ObjectReadOperation *op,
+                        const std::string &id);
+int dir_get_name_finish(bufferlist::const_iterator *it, std::string *name);
+int dir_get_name(librados::IoCtx *ioctx, const std::string &oid,
+                 const std::string &id, std::string *name);
+void dir_list_start(librados::ObjectReadOperation *op,
+                    const std::string &start, uint64_t max_return);
+int dir_list_finish(bufferlist::const_iterator *it, map<string, string> *images);
+int dir_list(librados::IoCtx *ioctx, const std::string &oid,
+             const std::string &start, uint64_t max_return,
+             map<string, string> *images);
+void dir_add_image(librados::ObjectWriteOperation *op,
+                   const std::string &name, const std::string &id);
+int dir_add_image(librados::IoCtx *ioctx, const std::string &oid,
+                  const std::string &name, const std::string &id);
+int dir_remove_image(librados::IoCtx *ioctx, const std::string &oid,
+                     const std::string &name, const std::string &id);
+void dir_remove_image(librados::ObjectWriteOperation *op,
+                      const std::string &name, const std::string &id);
+// atomic remove and add
+void dir_rename_image(librados::ObjectWriteOperation *op,
+                      const std::string &src, const std::string &dest,
+                      const std::string &id);
+void dir_state_assert(librados::ObjectOperation *op,
+                      cls::rbd::DirectoryState directory_state);
+int dir_state_assert(librados::IoCtx *ioctx, const std::string &oid,
+                     cls::rbd::DirectoryState directory_state);
+void dir_state_set(librados::ObjectWriteOperation *op,
+                   cls::rbd::DirectoryState directory_state);
+int dir_state_set(librados::IoCtx *ioctx, const std::string &oid,
+                  cls::rbd::DirectoryState directory_state);
+
+// operations on the rbd_object_map.$image_id object
+void object_map_load_start(librados::ObjectReadOperation *op);
+int object_map_load_finish(bufferlist::const_iterator *it,
+                           ceph::BitVector<2> *object_map);
+int object_map_load(librados::IoCtx *ioctx, const std::string &oid,
+                    ceph::BitVector<2> *object_map);
+void object_map_save(librados::ObjectWriteOperation *rados_op,
+                     const ceph::BitVector<2> &object_map);
+void object_map_resize(librados::ObjectWriteOperation *rados_op,
+                       uint64_t object_count, uint8_t default_state);
+void object_map_update(librados::ObjectWriteOperation *rados_op,
+                       uint64_t start_object_no, uint64_t end_object_no,
+                       uint8_t new_object_state,
+                       const boost::optional<uint8_t> &current_object_state);
+void object_map_snap_add(librados::ObjectWriteOperation *rados_op);
+void object_map_snap_remove(librados::ObjectWriteOperation *rados_op,
+                            const ceph::BitVector<2> &object_map);
+
+// class operations on the old format, kept for
+// backwards compatibility
+void old_snapshot_add(librados::ObjectWriteOperation *rados_op,
+                      snapid_t snap_id, const std::string &snap_name);
+void old_snapshot_remove(librados::ObjectWriteOperation *rados_op,
+                         const std::string &snap_name);
+void old_snapshot_rename(librados::ObjectWriteOperation *rados_op,
+                         snapid_t src_snap_id, const std::string &dst_name);
+
+void old_snapshot_list_start(librados::ObjectReadOperation *op);
+int old_snapshot_list_finish(bufferlist::const_iterator *it,
+                             std::vector<string> *names,
+                             std::vector<uint64_t> *sizes,
+                             ::SnapContext *snapc);
+int old_snapshot_list(librados::IoCtx *ioctx, const std::string &oid,
+                      std::vector<string> *names,
+                      std::vector<uint64_t> *sizes,
+                      ::SnapContext *snapc);
+
+// operations on the rbd_mirroring object
+void mirror_uuid_get_start(librados::ObjectReadOperation *op);
+int mirror_uuid_get_finish(bufferlist::const_iterator *it,
+                           std::string *uuid);
+int mirror_uuid_get(librados::IoCtx *ioctx, std::string *uuid);
+int mirror_uuid_set(librados::IoCtx *ioctx, const std::string &uuid);
+void mirror_mode_get_start(librados::ObjectReadOperation *op);
+int mirror_mode_get_finish(bufferlist::const_iterator *it,
+                           cls::rbd::MirrorMode *mirror_mode);
+int mirror_mode_get(librados::IoCtx *ioctx,
+                    cls::rbd::MirrorMode *mirror_mode);
+int mirror_mode_set(librados::IoCtx *ioctx,
+                    cls::rbd::MirrorMode mirror_mode);
+int mirror_peer_list(librados::IoCtx *ioctx,
+                     std::vector<cls::rbd::MirrorPeer> *peers);
+int mirror_peer_add(librados::IoCtx *ioctx, const std::string &uuid,
+                    const std::string &cluster_name,
+                    const std::string &client_name);
+int mirror_peer_remove(librados::IoCtx *ioctx,
+                       const std::string &uuid);
+int mirror_peer_set_client(librados::IoCtx *ioctx,
+                           const std::string &uuid,
+                           const std::string &client_name);
+int mirror_peer_set_cluster(librados::IoCtx *ioctx,
+                            const std::string &uuid,
+                            const std::string &cluster_name);
+void mirror_image_list_start(librados::ObjectReadOperation *op,
+                             const std::string &start, uint64_t max_return);
+int mirror_image_list_finish(bufferlist::const_iterator *it,
+                             std::map<string, string> *mirror_image_ids);
+int mirror_image_list(librados::IoCtx *ioctx,
+                      const std::string &start, uint64_t max_return,
+                      std::map<std::string, std::string> *mirror_image_ids);
+void mirror_image_get_image_id_start(librados::ObjectReadOperation *op,
+                                     const std::string &global_image_id);
+int mirror_image_get_image_id_finish(bufferlist::const_iterator *it,
+                                     std::string *image_id);
+int mirror_image_get_image_id(librados::IoCtx *ioctx,
+                              const std::string &global_image_id,
+                              std::string *image_id);
+int mirror_image_get(librados::IoCtx *ioctx, const std::string &image_id,
+                     cls::rbd::MirrorImage *mirror_image);
+void mirror_image_get_start(librados::ObjectReadOperation *op,
+                            const std::string &image_id);
+int mirror_image_get_finish(bufferlist::const_iterator *iter,
+                            cls::rbd::MirrorImage *mirror_image);
+void mirror_image_set(librados::ObjectWriteOperation *op,
+                      const std::string &image_id,
+                      const cls::rbd::MirrorImage &mirror_image);
+int mirror_image_set(librados::IoCtx *ioctx, const std::string &image_id,
+                     const cls::rbd::MirrorImage &mirror_image);
+void mirror_image_remove(librados::ObjectWriteOperation *op,
+                         const std::string &image_id);
+int mirror_image_remove(librados::IoCtx *ioctx,
+                        const std::string &image_id);
+int mirror_image_status_set(librados::IoCtx *ioctx,
+                            const std::string &global_image_id,
+                            const cls::rbd::MirrorImageStatus &status);
+void mirror_image_status_set(librados::ObjectWriteOperation *op,
+                             const std::string &global_image_id,
+                             const cls::rbd::MirrorImageStatus &status);
+int mirror_image_status_remove(librados::IoCtx *ioctx,
+                               const std::string &global_image_id);
+void mirror_image_status_remove(librados::ObjectWriteOperation *op,
+                                const std::string &global_image_id);
+int mirror_image_status_get(librados::IoCtx *ioctx,
+                            const std::string &global_image_id,
+                            cls::rbd::MirrorImageStatus *status);
+void mirror_image_status_get_start(librados::ObjectReadOperation *op,
+                                   const std::string &global_image_id);
+int mirror_image_status_get_finish(bufferlist::const_iterator *iter,
+                                   cls::rbd::MirrorImageStatus *status);
+int mirror_image_status_list(librados::IoCtx *ioctx,
+                             const std::string &start, uint64_t max_return,
+                             std::map<std::string, cls::rbd::MirrorImage> *images,
+                             std::map<std::string, cls::rbd::MirrorImageStatus> *statuses);
+void mirror_image_status_list_start(librados::ObjectReadOperation *op,
+                                    const std::string &start,
+                                    uint64_t max_return);
+int mirror_image_status_list_finish(bufferlist::const_iterator *iter,
+                                    std::map<std::string, cls::rbd::MirrorImage> *images,
+                                    std::map<std::string, cls::rbd::MirrorImageStatus> *statuses);
+int mirror_image_status_get_summary(librados::IoCtx *ioctx,
+                                    std::map<cls::rbd::MirrorImageStatusState, int> *states);
+void mirror_image_status_get_summary_start(librados::ObjectReadOperation *op);
+int mirror_image_status_get_summary_finish(bufferlist::const_iterator *iter,
+                                           std::map<cls::rbd::MirrorImageStatusState, int> *states);
+int mirror_image_status_remove_down(librados::IoCtx *ioctx);
+void mirror_image_status_remove_down(librados::ObjectWriteOperation *op);
+
+int mirror_image_instance_get(librados::IoCtx *ioctx,
+                              const std::string &global_image_id,
+                              entity_inst_t *instance);
+void mirror_image_instance_get_start(librados::ObjectReadOperation *op,
+                                     const std::string &global_image_id);
+int mirror_image_instance_get_finish(bufferlist::const_iterator *iter,
+                                     entity_inst_t *instance);
+int mirror_image_instance_list(librados::IoCtx *ioctx,
+                               const std::string &start, uint64_t max_return,
+                               std::map<std::string, entity_inst_t> *instances);
+void mirror_image_instance_list_start(librados::ObjectReadOperation *op,
+                                      const std::string &start,
+                                      uint64_t max_return);
+int mirror_image_instance_list_finish(bufferlist::const_iterator *iter,
+                                      std::map<std::string, entity_inst_t> *instances);
+
+void mirror_instances_list_start(librados::ObjectReadOperation *op);
+int mirror_instances_list_finish(bufferlist::const_iterator *iter,
+                                 std::vector<std::string> *instance_ids);
+int mirror_instances_list(librados::IoCtx *ioctx,
+                          std::vector<std::string> *instance_ids);
+void mirror_instances_add(librados::ObjectWriteOperation *op,
+                          const std::string &instance_id);
+int mirror_instances_add(librados::IoCtx *ioctx,
+                         const std::string &instance_id);
+void mirror_instances_remove(librados::ObjectWriteOperation *op,
+                             const std::string &instance_id);
+int mirror_instances_remove(librados::IoCtx *ioctx,
+                            const std::string &instance_id);
+
+// image mapping related routines
+void mirror_image_map_list_start(librados::ObjectReadOperation *op,
+                                 const std::string &start_after,
+                                 uint64_t max_read);
+int mirror_image_map_list_finish(bufferlist::const_iterator *iter,
+                                 std::map<std::string, cls::rbd::MirrorImageMap> *image_mapping);
+int mirror_image_map_list(librados::IoCtx *ioctx,
+                          const std::string &start_after, uint64_t max_read,
+                          std::map<std::string, cls::rbd::MirrorImageMap> *image_mapping);
+void mirror_image_map_update(librados::ObjectWriteOperation *op,
+                             const std::string &global_image_id,
+                             const cls::rbd::MirrorImageMap &image_map);
+void mirror_image_map_remove(librados::ObjectWriteOperation *op,
+                             const std::string &global_image_id);
+
+// Groups functions
+int group_dir_list(librados::IoCtx *ioctx, const std::string &oid,
+                   const std::string &start, uint64_t max_return,
+                   map<string, string> *groups);
+int group_dir_add(librados::IoCtx *ioctx, const std::string &oid,
+                  const std::string &name, const std::string &id);
+int group_dir_rename(librados::IoCtx *ioctx, const std::string &oid,
+                     const std::string &src, const std::string &dest,
+                     const std::string &id);
+int group_dir_remove(librados::IoCtx *ioctx, const std::string &oid,
+                     const std::string &name, const std::string &id);
+int group_image_remove(librados::IoCtx *ioctx, const std::string &oid,
+                       const cls::rbd::GroupImageSpec &spec);
+int group_image_list(librados::IoCtx *ioctx, const std::string &oid,
+                     const cls::rbd::GroupImageSpec &start,
+                     uint64_t max_return,
+                     std::vector<cls::rbd::GroupImageStatus> *images);
+int group_image_set(librados::IoCtx *ioctx, const std::string &oid,
+                    const cls::rbd::GroupImageStatus &st);
+int image_group_add(librados::IoCtx *ioctx, const std::string &oid,
+                    const cls::rbd::GroupSpec &group_spec);
+int image_group_remove(librados::IoCtx *ioctx, const std::string &oid,
+                       const cls::rbd::GroupSpec &group_spec);
+void image_group_get_start(librados::ObjectReadOperation *op);
+int image_group_get_finish(bufferlist::const_iterator *iter,
+                           cls::rbd::GroupSpec *group_spec);
+int image_group_get(librados::IoCtx *ioctx, const std::string &oid,
+                    cls::rbd::GroupSpec *group_spec);
+int group_snap_set(librados::IoCtx *ioctx, const std::string &oid,
+                   const cls::rbd::GroupSnapshot &snapshot);
+int group_snap_remove(librados::IoCtx *ioctx, const std::string &oid,
+                      const std::string &snap_id);
+int group_snap_get_by_id(librados::IoCtx *ioctx, const std::string &oid,
+                         const std::string &snap_id,
+                         cls::rbd::GroupSnapshot *snapshot);
+int group_snap_list(librados::IoCtx *ioctx, const std::string &oid,
+                    const cls::rbd::GroupSnapshot &start,
+                    uint64_t max_return,
+                    std::vector<cls::rbd::GroupSnapshot> *snapshots);
+
+// operations on rbd_trash object
+void trash_add(librados::ObjectWriteOperation *op,
+               const std::string &id,
+               const cls::rbd::TrashImageSpec &trash_spec);
+int trash_add(librados::IoCtx *ioctx, const std::string &id,
+              const cls::rbd::TrashImageSpec &trash_spec);
+void trash_remove(librados::ObjectWriteOperation *op,
+                  const std::string &id);
+int trash_remove(librados::IoCtx *ioctx, const std::string &id);
+void trash_list_start(librados::ObjectReadOperation *op,
+                      const std::string &start, uint64_t max_return);
+int trash_list_finish(bufferlist::const_iterator *it,
+                      map<string, cls::rbd::TrashImageSpec> *entries);
+int trash_list(librados::IoCtx *ioctx,
+               const std::string &start, uint64_t max_return,
+               map<string, cls::rbd::TrashImageSpec> *entries);
+void trash_get_start(librados::ObjectReadOperation *op,
+                     const std::string &id);
+int trash_get_finish(bufferlist::const_iterator *it,
+                     cls::rbd::TrashImageSpec *trash_spec);
+int trash_get(librados::IoCtx *ioctx, const std::string &id,
+              cls::rbd::TrashImageSpec *trash_spec);
+void trash_state_set(librados::ObjectWriteOperation *op,
+                     const std::string &id,
+                     const cls::rbd::TrashImageState &trash_state,
+                     const cls::rbd::TrashImageState &expect_state);
+int trash_state_set(librados::IoCtx *ioctx, const std::string &id,
+                    const cls::rbd::TrashImageState &trash_state,
+                    const cls::rbd::TrashImageState &expect_state);
+
+// operations on rbd_namespace object
+void namespace_add(librados::ObjectWriteOperation *op,
+                   const std::string &name);
+int namespace_add(librados::IoCtx *ioctx, const std::string &name);
+void namespace_remove(librados::ObjectWriteOperation *op,
+                      const std::string &name);
+int namespace_remove(librados::IoCtx *ioctx, const std::string &name);
+void namespace_list_start(librados::ObjectReadOperation *op,
+                          const std::string &start, uint64_t max_return);
+int namespace_list_finish(bufferlist::const_iterator *it,
+                          std::list<std::string> *entries);
+int namespace_list(librados::IoCtx *ioctx,
+                   const std::string &start, uint64_t max_return,
+                   std::list<std::string> *entries);
+
+// operations on data objects
+int assert_snapc_seq(librados::IoCtx *ioctx, const std::string &oid,
+                     uint64_t snapc_seq,
+                     cls::rbd::AssertSnapcSeqState state);
+void assert_snapc_seq(librados::ObjectWriteOperation *op,
+                      uint64_t snapc_seq,
+                      cls::rbd::AssertSnapcSeqState state);
+
+int copyup(librados::IoCtx *ioctx, const std::string &oid,
+           bufferlist data);
+
+void sparsify(librados::ObjectWriteOperation *op, size_t sparse_size,
+              bool remove_empty);
+int sparsify(librados::IoCtx *ioctx, const std::string &oid, size_t sparse_size,
+             bool remove_empty);
+
+} // namespace cls_client
+} // namespace librbd
+
+#endif // CEPH_LIBRBD_CLS_RBD_CLIENT_H
diff --git a/src/cls/rbd/cls_rbd_types.cc b/src/cls/rbd/cls_rbd_types.cc
new file mode 100644
index 00000000..0d2c9c75
--- /dev/null
+++ b/src/cls/rbd/cls_rbd_types.cc
@@ -0,0 +1,909 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <boost/variant.hpp>
+#include "cls/rbd/cls_rbd_types.h"
+#include "common/Formatter.h"
+
+namespace cls {
+namespace rbd {
+
+void MirrorPeer::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(uuid, bl);
+  encode(cluster_name, bl);
+  encode(client_name, bl);
+  encode(pool_id, bl);
+  ENCODE_FINISH(bl);
+}
+
+void MirrorPeer::decode(bufferlist::const_iterator &it) {
+  DECODE_START(1, it);
+  decode(uuid, it);
+  decode(cluster_name, it);
+  decode(client_name, it);
+  decode(pool_id, it);
+  DECODE_FINISH(it);
+}
+
+void MirrorPeer::dump(Formatter *f) const {
+  f->dump_string("uuid", uuid);
+  f->dump_string("cluster_name", cluster_name);
+  f->dump_string("client_name", client_name);
+  f->dump_int("pool_id", pool_id);
+}
+
+void MirrorPeer::generate_test_instances(std::list<MirrorPeer*> &o) {
+  o.push_back(new MirrorPeer());
+  o.push_back(new MirrorPeer("uuid-123", "cluster name", "client name", 123));
+}
+
+bool MirrorPeer::operator==(const MirrorPeer &rhs) const {
+  return (uuid == rhs.uuid &&
+          cluster_name == rhs.cluster_name &&
+          client_name == rhs.client_name &&
+          pool_id == rhs.pool_id);
+}
+
+std::ostream& operator<<(std::ostream& os, const MirrorMode& mirror_mode) {
+  switch (mirror_mode) {
+  case MIRROR_MODE_DISABLED:
+    os << "disabled";
+    break;
+  case MIRROR_MODE_IMAGE:
+    os << "image";
+    break;
+  case MIRROR_MODE_POOL:
+    os << "pool";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(mirror_mode) << ")";
+    break;
+  }
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const MirrorPeer& peer) {
+  os << "["
+     << "uuid=" << peer.uuid << ", "
+     << "cluster_name=" << peer.cluster_name << ", "
+     << "client_name=" << peer.client_name;
+  if (peer.pool_id != -1) {
+    os << ", pool_id=" << peer.pool_id;
+  }
+  os << "]";
+  return os;
+}
+
+void MirrorImage::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(global_image_id, bl);
+  encode(static_cast<uint8_t>(state), bl);
+  ENCODE_FINISH(bl);
+}
+
+void MirrorImage::decode(bufferlist::const_iterator &it) {
+  uint8_t int_state;
+  DECODE_START(1, it);
+  decode(global_image_id, it);
+  decode(int_state, it);
+  state = static_cast<MirrorImageState>(int_state);
+  DECODE_FINISH(it);
+}
+
+void MirrorImage::dump(Formatter *f) const {
+  f->dump_string("global_image_id", global_image_id);
+  f->dump_int("state", state);
+}
+
+void MirrorImage::generate_test_instances(std::list<MirrorImage*> &o) {
+  o.push_back(new MirrorImage());
+  o.push_back(new MirrorImage("uuid-123", MIRROR_IMAGE_STATE_ENABLED));
+  o.push_back(new MirrorImage("uuid-abc", MIRROR_IMAGE_STATE_DISABLING));
+}
+
+bool MirrorImage::operator==(const MirrorImage &rhs) const {
+  return global_image_id == rhs.global_image_id && state == rhs.state;
+}
+
+bool MirrorImage::operator<(const MirrorImage &rhs) const {
+  return global_image_id < rhs.global_image_id ||
+	(global_image_id == rhs.global_image_id  && state < rhs.state);
+}
+
+std::ostream& operator<<(std::ostream& os, const MirrorImageState& mirror_state) {
+  switch (mirror_state) {
+  case MIRROR_IMAGE_STATE_DISABLING:
+    os << "disabling";
+    break;
+  case MIRROR_IMAGE_STATE_ENABLED:
+    os << "enabled";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(mirror_state) << ")";
+    break;
+  }
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const MirrorImage& mirror_image) {
+  os << "["
+     << "global_image_id=" << mirror_image.global_image_id << ", "
+     << "state=" << mirror_image.state << "]";
+  return os;
+}
+
+void MirrorImageStatus::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(state, bl);
+  encode(description, bl);
+  encode(last_update, bl);
+  encode(up, bl);
+  ENCODE_FINISH(bl);
+}
+
+void MirrorImageStatus::decode(bufferlist::const_iterator &it) {
+  DECODE_START(1, it);
+  decode(state, it);
+  decode(description, it);
+  decode(last_update, it);
+  decode(up, it);
+  DECODE_FINISH(it);
+}
+
+void MirrorImageStatus::dump(Formatter *f) const {
+  f->dump_string("state", state_to_string());
+  f->dump_string("description", description);
+  f->dump_stream("last_update") << last_update;
+}
+
+std::string MirrorImageStatus::state_to_string() const {
+  std::stringstream ss;
+  ss << (up ? "up+" : "down+") << state;
+  return ss.str();
+}
+
+void MirrorImageStatus::generate_test_instances(
+  std::list<MirrorImageStatus*> &o) {
+  o.push_back(new MirrorImageStatus());
+  o.push_back(new MirrorImageStatus(MIRROR_IMAGE_STATUS_STATE_REPLAYING));
+  o.push_back(new MirrorImageStatus(MIRROR_IMAGE_STATUS_STATE_ERROR, "error"));
+}
+
+bool MirrorImageStatus::operator==(const MirrorImageStatus &rhs) const {
+  return state == rhs.state && description == rhs.description && up == rhs.up;
+}
+
+std::ostream& operator<<(std::ostream& os, const MirrorImageStatusState& state) {
+  switch (state) {
+  case MIRROR_IMAGE_STATUS_STATE_UNKNOWN:
+    os << "unknown";
+    break;
+  case MIRROR_IMAGE_STATUS_STATE_ERROR:
+    os << "error";
+    break;
+  case MIRROR_IMAGE_STATUS_STATE_SYNCING:
+    os << "syncing";
+    break;
+  case MIRROR_IMAGE_STATUS_STATE_STARTING_REPLAY:
+    os << "starting_replay";
+    break;
+  case MIRROR_IMAGE_STATUS_STATE_REPLAYING:
+    os << "replaying";
+    break;
+  case MIRROR_IMAGE_STATUS_STATE_STOPPING_REPLAY:
+    os << "stopping_replay";
+    break;
+  case MIRROR_IMAGE_STATUS_STATE_STOPPED:
+    os << "stopped";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(state) << ")";
+    break;
+  }
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const MirrorImageStatus& status) {
+  os << "["
+     << "state=" << status.state_to_string() << ", "
+     << "description=" << status.description << ", "
+     << "last_update=" << status.last_update << "]";
+  return os;
+}
+
+void ParentImageSpec::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(pool_id, bl);
+  encode(pool_namespace, bl);
+  encode(image_id, bl);
+  encode(snap_id, bl);
+  ENCODE_FINISH(bl);
+}
+
+void ParentImageSpec::decode(bufferlist::const_iterator& bl) {
+  DECODE_START(1, bl);
+  decode(pool_id, bl);
+  decode(pool_namespace, bl);
+  decode(image_id, bl);
+  decode(snap_id, bl);
+  DECODE_FINISH(bl);
+}
+
+void ParentImageSpec::dump(Formatter *f) const {
+  f->dump_int("pool_id", pool_id);
+  f->dump_string("pool_namespace", pool_namespace);
+  f->dump_string("image_id", image_id);
+  f->dump_unsigned("snap_id", snap_id);
+}
+
+void ParentImageSpec::generate_test_instances(std::list<ParentImageSpec*>& o) {
+  o.push_back(new ParentImageSpec{});
+  o.push_back(new ParentImageSpec{1, "", "foo", 3});
+  o.push_back(new ParentImageSpec{1, "ns", "foo", 3});
+}
+
+void ChildImageSpec::encode(bufferlist &bl) const {
+  ENCODE_START(2, 1, bl);
+  encode(pool_id, bl);
+  encode(image_id, bl);
+  encode(pool_namespace, bl);
+  ENCODE_FINISH(bl);
+}
+
+void ChildImageSpec::decode(bufferlist::const_iterator &it) {
+  DECODE_START(2, it);
+  decode(pool_id, it);
+  decode(image_id, it);
+  if (struct_v >= 2) {
+    decode(pool_namespace, it);
+  }
+  DECODE_FINISH(it);
+}
+
+void ChildImageSpec::dump(Formatter *f) const {
+  f->dump_int("pool_id", pool_id);
+  f->dump_string("pool_namespace", pool_namespace);
+  f->dump_string("image_id", image_id);
+}
+
+void ChildImageSpec::generate_test_instances(std::list<ChildImageSpec*> &o) {
+  o.push_back(new ChildImageSpec());
+  o.push_back(new ChildImageSpec(123, "", "abc"));
+  o.push_back(new ChildImageSpec(123, "ns", "abc"));
+}
+
+void GroupImageSpec::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(image_id, bl);
+  encode(pool_id, bl);
+  ENCODE_FINISH(bl);
+}
+
+void GroupImageSpec::decode(bufferlist::const_iterator &it) {
+  DECODE_START(1, it);
+  decode(image_id, it);
+  decode(pool_id, it);
+  DECODE_FINISH(it);
+}
+
+void GroupImageSpec::dump(Formatter *f) const {
+  f->dump_string("image_id", image_id);
+  f->dump_int("pool_id", pool_id);
+}
+
+int GroupImageSpec::from_key(const std::string &image_key,
+				    GroupImageSpec *spec) {
+  if (nullptr == spec) return -EINVAL;
+  int prefix_len = cls::rbd::RBD_GROUP_IMAGE_KEY_PREFIX.size();
+  std::string data_string = image_key.substr(prefix_len,
+					     image_key.size() - prefix_len);
+  size_t p = data_string.find("_");
+  if (std::string::npos == p) {
+    return -EIO;
+  }
+  data_string[p] = ' ';
+
+  istringstream iss(data_string);
+  uint64_t pool_id;
+  string image_id;
+  iss >> std::hex >> pool_id >> image_id;
+
+  spec->image_id = image_id;
+  spec->pool_id = pool_id;
+  return 0;
+}
+
+std::string GroupImageSpec::image_key() {
+  if (-1 == pool_id)
+    return "";
+  else {
+    ostringstream oss;
+    oss << RBD_GROUP_IMAGE_KEY_PREFIX << std::setw(16)
+	<< std::setfill('0') << std::hex << pool_id << "_" << image_id;
+    return oss.str();
+  }
+}
+
+void GroupImageSpec::generate_test_instances(std::list<GroupImageSpec*> &o) {
+  o.push_back(new GroupImageSpec("10152ae8944a", 0));
+  o.push_back(new GroupImageSpec("1018643c9869", 3));
+}
+
+void GroupImageStatus::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(spec, bl);
+  encode(state, bl);
+  ENCODE_FINISH(bl);
+}
+
+void GroupImageStatus::decode(bufferlist::const_iterator &it) {
+  DECODE_START(1, it);
+  decode(spec, it);
+  decode(state, it);
+  DECODE_FINISH(it);
+}
+
+std::string GroupImageStatus::state_to_string() const {
+  std::stringstream ss;
+  if (state == GROUP_IMAGE_LINK_STATE_INCOMPLETE) {
+    ss << "incomplete";
+  }
+  if (state == GROUP_IMAGE_LINK_STATE_ATTACHED) {
+    ss << "attached";
+  }
+  return ss.str();
+}
+
+void GroupImageStatus::dump(Formatter *f) const {
+  spec.dump(f);
+  f->dump_string("state", state_to_string());
+}
+
+void GroupImageStatus::generate_test_instances(std::list<GroupImageStatus*> &o) {
+  o.push_back(new GroupImageStatus(GroupImageSpec("10152ae8944a", 0), GROUP_IMAGE_LINK_STATE_ATTACHED));
+  o.push_back(new GroupImageStatus(GroupImageSpec("1018643c9869", 3), GROUP_IMAGE_LINK_STATE_ATTACHED));
+  o.push_back(new GroupImageStatus(GroupImageSpec("10152ae8944a", 0), GROUP_IMAGE_LINK_STATE_INCOMPLETE));
+  o.push_back(new GroupImageStatus(GroupImageSpec("1018643c9869", 3), GROUP_IMAGE_LINK_STATE_INCOMPLETE));
+}
+
+
+void GroupSpec::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(pool_id, bl);
+  encode(group_id, bl);
+  ENCODE_FINISH(bl);
+}
+
+void GroupSpec::decode(bufferlist::const_iterator &it) {
+  DECODE_START(1, it);
+  decode(pool_id, it);
+  decode(group_id, it);
+  DECODE_FINISH(it);
+}
+
+void GroupSpec::dump(Formatter *f) const {
+  f->dump_string("group_id", group_id);
+  f->dump_int("pool_id", pool_id);
+}
+
+bool GroupSpec::is_valid() const {
+  return (!group_id.empty()) && (pool_id != -1);
+}
+
+void GroupSpec::generate_test_instances(std::list<GroupSpec *> &o) {
+  o.push_back(new GroupSpec("10152ae8944a", 0));
+  o.push_back(new GroupSpec("1018643c9869", 3));
+}
+
+void GroupSnapshotNamespace::encode(bufferlist& bl) const {
+  using ceph::encode;
+  encode(group_pool, bl);
+  encode(group_id, bl);
+  encode(group_snapshot_id, bl);
+}
+
+void GroupSnapshotNamespace::decode(bufferlist::const_iterator& it) {
+  using ceph::decode;
+  decode(group_pool, it);
+  decode(group_id, it);
+  decode(group_snapshot_id, it);
+}
+
+void GroupSnapshotNamespace::dump(Formatter *f) const {
+  f->dump_int("group_pool", group_pool);
+  f->dump_string("group_id", group_id);
+  f->dump_string("group_snapshot_id", group_snapshot_id);
+}
+
+void TrashSnapshotNamespace::encode(bufferlist& bl) const {
+  using ceph::encode;
+  encode(original_name, bl);
+  encode(static_cast<uint32_t>(original_snapshot_namespace_type), bl);
+}
+
+void TrashSnapshotNamespace::decode(bufferlist::const_iterator& it) {
+  using ceph::decode;
+  decode(original_name, it);
+  uint32_t snap_type;
+  decode(snap_type, it);
+  original_snapshot_namespace_type = static_cast<SnapshotNamespaceType>(
+    snap_type);
+}
+
+void TrashSnapshotNamespace::dump(Formatter *f) const {
+  f->dump_string("original_name", original_name);
+  f->dump_stream("original_snapshot_namespace")
+    << original_snapshot_namespace_type;
+}
+
+class EncodeSnapshotNamespaceVisitor : public boost::static_visitor<void> {
+public:
+  explicit EncodeSnapshotNamespaceVisitor(bufferlist &bl) : m_bl(bl) {
+  }
+
+  template <typename T>
+  inline void operator()(const T& t) const {
+    using ceph::encode;
+    encode(static_cast<uint32_t>(T::SNAPSHOT_NAMESPACE_TYPE), m_bl);
+    t.encode(m_bl);
+  }
+
+private:
+  bufferlist &m_bl;
+};
+
+class DecodeSnapshotNamespaceVisitor : public boost::static_visitor<void> {
+public:
+  DecodeSnapshotNamespaceVisitor(bufferlist::const_iterator &iter)
+    : m_iter(iter) {
+  }
+
+  template <typename T>
+  inline void operator()(T& t) const {
+    t.decode(m_iter);
+  }
+private:
+  bufferlist::const_iterator &m_iter;
+};
+
+class DumpSnapshotNamespaceVisitor : public boost::static_visitor<void> {
+public:
+  explicit DumpSnapshotNamespaceVisitor(Formatter *formatter, const std::string &key)
+    : m_formatter(formatter), m_key(key) {}
+
+  template <typename T>
+  inline void operator()(const T& t) const {
+    auto type = T::SNAPSHOT_NAMESPACE_TYPE;
+    m_formatter->dump_string(m_key.c_str(), stringify(type));
+    t.dump(m_formatter);
+  }
+private:
+  ceph::Formatter *m_formatter;
+  std::string m_key;
+};
+
+class GetTypeVisitor : public boost::static_visitor<SnapshotNamespaceType> {
+public:
+  template <typename T>
+  inline SnapshotNamespaceType operator()(const T&) const {
+    return static_cast<SnapshotNamespaceType>(T::SNAPSHOT_NAMESPACE_TYPE);
+  }
+};
+
+SnapshotNamespaceType get_snap_namespace_type(
+    const SnapshotNamespace& snapshot_namespace) {
+  return static_cast<SnapshotNamespaceType>(boost::apply_visitor(
+    GetTypeVisitor(), snapshot_namespace));
+}
+
+void SnapshotInfo::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(id, bl);
+  encode(snapshot_namespace, bl);
+  encode(name, bl);
+  encode(image_size, bl);
+  encode(timestamp, bl);
+  encode(child_count, bl);
+  ENCODE_FINISH(bl);
+}
+
+void SnapshotInfo::decode(bufferlist::const_iterator& it) {
+  DECODE_START(1, it);
+  decode(id, it);
+  decode(snapshot_namespace, it);
+  decode(name, it);
+  decode(image_size, it);
+  decode(timestamp, it);
+  decode(child_count, it);
+  DECODE_FINISH(it);
+}
+
+void SnapshotInfo::dump(Formatter *f) const {
+  f->dump_unsigned("id", id);
+  f->open_object_section("namespace");
+  boost::apply_visitor(DumpSnapshotNamespaceVisitor(f, "type"),
+                       snapshot_namespace);
+  f->close_section();
+  f->dump_string("name", name);
+  f->dump_unsigned("image_size", image_size);
+  f->dump_stream("timestamp") << timestamp;
+}
+
+void SnapshotInfo::generate_test_instances(std::list<SnapshotInfo*> &o) {
+  o.push_back(new SnapshotInfo(1ULL, UserSnapshotNamespace{}, "snap1", 123,
+                               {123456, 0}, 12));
+  o.push_back(new SnapshotInfo(2ULL,
+                               GroupSnapshotNamespace{567, "group1", "snap1"},
+                               "snap1", 123, {123456, 0}, 987));
+  o.push_back(new SnapshotInfo(3ULL,
+                               TrashSnapshotNamespace{
+                                 SNAPSHOT_NAMESPACE_TYPE_USER, "snap1"},
+                               "12345", 123, {123456, 0}, 429));
+}
+
+void SnapshotNamespace::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  boost::apply_visitor(EncodeSnapshotNamespaceVisitor(bl), *this);
+  ENCODE_FINISH(bl);
+}
+
+void SnapshotNamespace::decode(bufferlist::const_iterator &p)
+{
+  DECODE_START(1, p);
+  uint32_t snap_type;
+  decode(snap_type, p);
+  switch (snap_type) {
+    case cls::rbd::SNAPSHOT_NAMESPACE_TYPE_USER:
+      *this = UserSnapshotNamespace();
+      break;
+    case cls::rbd::SNAPSHOT_NAMESPACE_TYPE_GROUP:
+      *this = GroupSnapshotNamespace();
+      break;
+    case cls::rbd::SNAPSHOT_NAMESPACE_TYPE_TRASH:
+      *this = TrashSnapshotNamespace();
+      break;
+    default:
+      *this = UnknownSnapshotNamespace();
+      break;
+  }
+  boost::apply_visitor(DecodeSnapshotNamespaceVisitor(p), *this);
+  DECODE_FINISH(p);
+}
+
+void SnapshotNamespace::dump(Formatter *f) const {
+  boost::apply_visitor(
+    DumpSnapshotNamespaceVisitor(f, "snapshot_namespace_type"), *this);
+}
+
+void SnapshotNamespace::generate_test_instances(std::list<SnapshotNamespace*> &o) {
+  o.push_back(new SnapshotNamespace(UserSnapshotNamespace()));
+  o.push_back(new SnapshotNamespace(GroupSnapshotNamespace(0, "10152ae8944a",
+                                                           "2118643c9732")));
+  o.push_back(new SnapshotNamespace(GroupSnapshotNamespace(5, "1018643c9869",
+                                                           "33352be8933c")));
+  o.push_back(new SnapshotNamespace(TrashSnapshotNamespace()));
+}
+
+std::ostream& operator<<(std::ostream& os, const SnapshotNamespaceType& type) {
+  switch (type) {
+  case SNAPSHOT_NAMESPACE_TYPE_USER:
+    os << "user";
+    break;
+  case SNAPSHOT_NAMESPACE_TYPE_GROUP:
+    os << "group";
+    break;
+  case SNAPSHOT_NAMESPACE_TYPE_TRASH:
+    os << "trash";
+    break;
+  default:
+    os << "unknown";
+    break;
+  }
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const UserSnapshotNamespace& ns) {
+  os << "[" << SNAPSHOT_NAMESPACE_TYPE_USER << "]";
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const GroupSnapshotNamespace& ns) {
+  os << "[" << SNAPSHOT_NAMESPACE_TYPE_GROUP << " "
+     << "group_pool=" << ns.group_pool << ", "
+     << "group_id=" << ns.group_id << ", "
+     << "group_snapshot_id=" << ns.group_snapshot_id << "]";
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const TrashSnapshotNamespace& ns) {
+  os << "[" << SNAPSHOT_NAMESPACE_TYPE_TRASH << " "
+     << "original_name=" << ns.original_name << ", "
+     << "original_snapshot_namespace=" << ns.original_snapshot_namespace_type
+     << "]";
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const UnknownSnapshotNamespace& ns) {
+  os << "[unknown]";
+  return os;
+}
+
+void ImageSnapshotSpec::encode(bufferlist& bl) const {
+  using ceph::encode;
+  ENCODE_START(1, 1, bl);
+  encode(pool, bl);
+  encode(image_id, bl);
+  encode(snap_id, bl);
+  ENCODE_FINISH(bl);
+}
+
+void ImageSnapshotSpec::decode(bufferlist::const_iterator& it) {
+  using ceph::decode;
+  DECODE_START(1, it);
+  decode(pool, it);
+  decode(image_id, it);
+  decode(snap_id, it);
+  DECODE_FINISH(it);
+}
+
+void ImageSnapshotSpec::dump(Formatter *f) const {
+  f->dump_int("pool", pool);
+  f->dump_string("image_id", image_id);
+  f->dump_int("snap_id", snap_id);
+}
+
+void ImageSnapshotSpec::generate_test_instances(std::list<ImageSnapshotSpec *> &o) {
+  o.push_back(new ImageSnapshotSpec(0, "myimage", 2));
+  o.push_back(new ImageSnapshotSpec(1, "testimage", 7));
+}
+
+void GroupSnapshot::encode(bufferlist& bl) const {
+  using ceph::encode;
+  ENCODE_START(1, 1, bl);
+  encode(id, bl);
+  encode(name, bl);
+  encode(state, bl);
+  encode(snaps, bl);
+  ENCODE_FINISH(bl);
+}
+
+void GroupSnapshot::decode(bufferlist::const_iterator& it) {
+  using ceph::decode;
+  DECODE_START(1, it);
+  decode(id, it);
+  decode(name, it);
+  decode(state, it);
+  decode(snaps, it);
+  DECODE_FINISH(it);
+}
+
+void GroupSnapshot::dump(Formatter *f) const {
+  f->dump_string("id", id);
+  f->dump_string("name", name);
+  f->dump_int("state", state);
+}
+
+void GroupSnapshot::generate_test_instances(std::list<GroupSnapshot *> &o) {
+  o.push_back(new GroupSnapshot("10152ae8944a", "groupsnapshot1", GROUP_SNAPSHOT_STATE_INCOMPLETE));
+  o.push_back(new GroupSnapshot("1018643c9869", "groupsnapshot2", GROUP_SNAPSHOT_STATE_COMPLETE));
+}
+void TrashImageSpec::encode(bufferlist& bl) const {
+  ENCODE_START(2, 1, bl);
+  encode(source, bl);
+  encode(name, bl);
+  encode(deletion_time, bl);
+  encode(deferment_end_time, bl);
+  encode(state, bl);
+  ENCODE_FINISH(bl);
+}
+
+void TrashImageSpec::decode(bufferlist::const_iterator &it) {
+  DECODE_START(2, it);
+  decode(source, it);
+  decode(name, it);
+  decode(deletion_time, it);
+  decode(deferment_end_time, it);
+  if (struct_v >= 2) {
+    decode(state, it);
+  }
+  DECODE_FINISH(it);
+}
+
+void TrashImageSpec::dump(Formatter *f) const {
+  f->dump_stream("source") << source;
+  f->dump_string("name", name);
+  f->dump_unsigned("deletion_time", deletion_time);
+  f->dump_unsigned("deferment_end_time", deferment_end_time);
+}
+
+void MirrorImageMap::encode(bufferlist &bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(instance_id, bl);
+  encode(mapped_time, bl);
+  encode(data, bl);
+  ENCODE_FINISH(bl);
+}
+
+void MirrorImageMap::decode(bufferlist::const_iterator &it) {
+  DECODE_START(1, it);
+  decode(instance_id, it);
+  decode(mapped_time, it);
+  decode(data, it);
+  DECODE_FINISH(it);
+}
+
+void MirrorImageMap::dump(Formatter *f) const {
+  f->dump_string("instance_id", instance_id);
+  f->dump_stream("mapped_time") << mapped_time;
+
+  std::stringstream data_ss;
+  data.hexdump(data_ss);
+  f->dump_string("data", data_ss.str());
+}
+
+void MirrorImageMap::generate_test_instances(
+  std::list<MirrorImageMap*> &o) {
+  bufferlist data;
+  data.append(std::string(128, '1'));
+
+  o.push_back(new MirrorImageMap("uuid-123", utime_t(), data));
+  o.push_back(new MirrorImageMap("uuid-abc", utime_t(), data));
+}
+
+bool MirrorImageMap::operator==(const MirrorImageMap &rhs) const {
+  return instance_id == rhs.instance_id && mapped_time == rhs.mapped_time &&
+    data.contents_equal(rhs.data);
+}
+
+bool MirrorImageMap::operator<(const MirrorImageMap &rhs) const {
+  return instance_id < rhs.instance_id ||
+        (instance_id == rhs.instance_id && mapped_time < rhs.mapped_time);
+}
+
+std::ostream& operator<<(std::ostream& os,
+                         const MirrorImageMap &image_map) {
+  return os << "[" << "instance_id=" << image_map.instance_id << ", mapped_time="
+            << image_map.mapped_time << "]";
+}
+
+std::ostream& operator<<(std::ostream& os,
+                         const MigrationHeaderType& type) {
+  switch (type) {
+  case MIGRATION_HEADER_TYPE_SRC:
+    os << "source";
+    break;
+  case MIGRATION_HEADER_TYPE_DST:
+    os << "destination";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(type) << ")";
+    break;
+  }
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os,
+                         const MigrationState& migration_state) {
+  switch (migration_state) {
+  case MIGRATION_STATE_ERROR:
+    os << "error";
+    break;
+  case MIGRATION_STATE_PREPARING:
+    os << "preparing";
+    break;
+  case MIGRATION_STATE_PREPARED:
+    os << "prepared";
+    break;
+  case MIGRATION_STATE_EXECUTING:
+    os << "executing";
+    break;
+  case MIGRATION_STATE_EXECUTED:
+    os << "executed";
+    break;
+  case MIGRATION_STATE_ABORTING:
+    os << "aborting";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(migration_state) << ")";
+    break;
+  }
+  return os;
+}
+
+void MigrationSpec::encode(bufferlist& bl) const {
+  ENCODE_START(1, 1, bl);
+  encode(header_type, bl);
+  encode(pool_id, bl);
+  encode(pool_namespace, bl);
+  encode(image_name, bl);
+  encode(image_id, bl);
+  encode(snap_seqs, bl);
+  encode(overlap, bl);
+  encode(flatten, bl);
+  encode(mirroring, bl);
+  encode(state, bl);
+  encode(state_description, bl);
+  ENCODE_FINISH(bl);
+}
+
+void MigrationSpec::decode(bufferlist::const_iterator& bl) {
+  DECODE_START(1, bl);
+  decode(header_type, bl);
+  decode(pool_id, bl);
+  decode(pool_namespace, bl);
+  decode(image_name, bl);
+  decode(image_id, bl);
+  decode(snap_seqs, bl);
+  decode(overlap, bl);
+  decode(flatten, bl);
+  decode(mirroring, bl);
+  decode(state, bl);
+  decode(state_description, bl);
+  DECODE_FINISH(bl);
+}
+
+std::ostream& operator<<(std::ostream& os,
+                         const std::map<uint64_t, uint64_t>& snap_seqs) {
+  os << "{";
+  size_t count = 0;
+  for (auto &it : snap_seqs) {
+    os << (count++ > 0 ? ", " : "") << "(" << it.first << ", " << it.second
+       << ")";
+  }
+  os << "}";
+  return os;
+}
+
+void MigrationSpec::dump(Formatter *f) const {
+  f->dump_stream("header_type") << header_type;
+  f->dump_int("pool_id", pool_id);
+  f->dump_string("pool_namespace", pool_namespace);
+  f->dump_string("image_name", image_name);
+  f->dump_string("image_id", image_id);
+  f->dump_stream("snap_seqs") << snap_seqs;
+  f->dump_unsigned("overlap", overlap);
+  f->dump_bool("mirroring", mirroring);
+}
+
+void MigrationSpec::generate_test_instances(std::list<MigrationSpec*> &o) {
+  o.push_back(new MigrationSpec());
+  o.push_back(new MigrationSpec(MIGRATION_HEADER_TYPE_SRC, 1, "ns",
+                                "image_name", "image_id", {{1, 2}}, 123, true,
+                                true, MIGRATION_STATE_PREPARED, "description"));
+}
+
+std::ostream& operator<<(std::ostream& os,
+                         const MigrationSpec& migration_spec) {
+  os << "["
+     << "header_type=" << migration_spec.header_type << ", "
+     << "pool_id=" << migration_spec.pool_id << ", "
+     << "pool_namespace=" << migration_spec.pool_namespace << ", "
+     << "image_name=" << migration_spec.image_name << ", "
+     << "image_id=" << migration_spec.image_id << ", "
+     << "snap_seqs=" << migration_spec.snap_seqs << ", "
+     << "overlap=" << migration_spec.overlap << ", "
+     << "flatten=" << migration_spec.flatten << ", "
+     << "mirroring=" << migration_spec.mirroring << ", "
+     << "state=" << migration_spec.state << ", "
+     << "state_description=" << migration_spec.state_description << "]";
+  return os;
+}
+
+std::ostream& operator<<(std::ostream& os, const AssertSnapcSeqState& state) {
+  switch (state) {
+  case ASSERT_SNAPC_SEQ_GT_SNAPSET_SEQ:
+    os << "gt";
+    break;
+  case ASSERT_SNAPC_SEQ_LE_SNAPSET_SEQ:
+    os << "le";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(state) << ")";
+    break;
+  }
+  return os;
+}
+
+} // namespace rbd
+} // namespace cls
diff --git a/src/cls/rbd/cls_rbd_types.h b/src/cls/rbd/cls_rbd_types.h
new file mode 100644
index 00000000..073006b3
--- /dev/null
+++ b/src/cls/rbd/cls_rbd_types.h
@@ -0,0 +1,791 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_RBD_TYPES_H
+#define CEPH_CLS_RBD_TYPES_H
+
+#include <boost/variant.hpp>
+#include "include/int_types.h"
+#include "include/buffer.h"
+#include "include/encoding.h"
+#include "include/stringify.h"
+#include "include/utime.h"
+#include <iosfwd>
+#include <string>
+#include <set>
+
+#define RBD_GROUP_REF "rbd_group_ref"
+
+namespace ceph { class Formatter; }
+
+namespace cls {
+namespace rbd {
+
+static const uint32_t MAX_OBJECT_MAP_OBJECT_COUNT = 256000000;
+static const string RBD_GROUP_IMAGE_KEY_PREFIX = "image_";
+
+enum DirectoryState {
+  DIRECTORY_STATE_READY         = 0,
+  DIRECTORY_STATE_ADD_DISABLED  = 1
+};
+
+inline void encode(DirectoryState state, bufferlist& bl,
+		   uint64_t features=0)
+{
+  ceph::encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(DirectoryState &state, bufferlist::const_iterator& it)
+{
+  uint8_t int_state;
+  ceph::decode(int_state, it);
+  state = static_cast<DirectoryState>(int_state);
+}
+
+enum MirrorMode {
+  MIRROR_MODE_DISABLED = 0,
+  MIRROR_MODE_IMAGE    = 1,
+  MIRROR_MODE_POOL     = 2
+};
+
+enum GroupImageLinkState {
+  GROUP_IMAGE_LINK_STATE_ATTACHED,
+  GROUP_IMAGE_LINK_STATE_INCOMPLETE
+};
+
+inline void encode(const GroupImageLinkState &state, bufferlist& bl,
+		   uint64_t features=0)
+{
+  using ceph::encode;
+  encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(GroupImageLinkState &state, bufferlist::const_iterator& it)
+{
+  uint8_t int_state;
+  using ceph::decode;
+  decode(int_state, it);
+  state = static_cast<GroupImageLinkState>(int_state);
+}
+
+struct MirrorPeer {
+  MirrorPeer() {
+  }
+  MirrorPeer(const std::string &uuid, const std::string &cluster_name,
+             const std::string &client_name, int64_t pool_id)
+    : uuid(uuid), cluster_name(cluster_name), client_name(client_name),
+      pool_id(pool_id) {
+  }
+
+  std::string uuid;
+  std::string cluster_name;
+  std::string client_name;
+  int64_t pool_id = -1;
+
+  inline bool is_valid() const {
+    return (!uuid.empty() && !cluster_name.empty() && !client_name.empty());
+  }
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<MirrorPeer*> &o);
+
+  bool operator==(const MirrorPeer &rhs) const;
+};
+
+std::ostream& operator<<(std::ostream& os, const MirrorMode& mirror_mode);
+std::ostream& operator<<(std::ostream& os, const MirrorPeer& peer);
+
+WRITE_CLASS_ENCODER(MirrorPeer);
+
+enum MirrorImageState {
+  MIRROR_IMAGE_STATE_DISABLING = 0,
+  MIRROR_IMAGE_STATE_ENABLED   = 1,
+  MIRROR_IMAGE_STATE_DISABLED  = 2,
+};
+
+struct MirrorImage {
+  MirrorImage() {}
+  MirrorImage(const std::string &global_image_id, MirrorImageState state)
+    : global_image_id(global_image_id), state(state) {}
+
+  std::string global_image_id;
+  MirrorImageState state = MIRROR_IMAGE_STATE_DISABLING;
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<MirrorImage*> &o);
+
+  bool operator==(const MirrorImage &rhs) const;
+  bool operator<(const MirrorImage &rhs) const;
+};
+
+std::ostream& operator<<(std::ostream& os, const MirrorImageState& mirror_state);
+std::ostream& operator<<(std::ostream& os, const MirrorImage& mirror_image);
+
+WRITE_CLASS_ENCODER(MirrorImage);
+
+enum MirrorImageStatusState {
+  MIRROR_IMAGE_STATUS_STATE_UNKNOWN         = 0,
+  MIRROR_IMAGE_STATUS_STATE_ERROR           = 1,
+  MIRROR_IMAGE_STATUS_STATE_SYNCING         = 2,
+  MIRROR_IMAGE_STATUS_STATE_STARTING_REPLAY = 3,
+  MIRROR_IMAGE_STATUS_STATE_REPLAYING       = 4,
+  MIRROR_IMAGE_STATUS_STATE_STOPPING_REPLAY = 5,
+  MIRROR_IMAGE_STATUS_STATE_STOPPED         = 6,
+};
+
+inline void encode(const MirrorImageStatusState &state, bufferlist& bl,
+		   uint64_t features=0)
+{
+  using ceph::encode;
+  encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(MirrorImageStatusState &state, bufferlist::const_iterator& it)
+{
+  uint8_t int_state;
+  using ceph::decode;
+  decode(int_state, it);
+  state = static_cast<MirrorImageStatusState>(int_state);
+}
+
+struct MirrorImageStatus {
+  MirrorImageStatus() {}
+  MirrorImageStatus(MirrorImageStatusState state,
+		    const std::string &description = "")
+    : state(state), description(description) {}
+
+  MirrorImageStatusState state = MIRROR_IMAGE_STATUS_STATE_UNKNOWN;
+  std::string description;
+  utime_t last_update;
+  bool up = false;
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  std::string state_to_string() const;
+
+  static void generate_test_instances(std::list<MirrorImageStatus*> &o);
+
+  bool operator==(const MirrorImageStatus &rhs) const;
+};
+
+std::ostream& operator<<(std::ostream& os, const MirrorImageStatus& status);
+std::ostream& operator<<(std::ostream& os, const MirrorImageStatusState& state);
+
+WRITE_CLASS_ENCODER(MirrorImageStatus);
+
+struct ParentImageSpec {
+  int64_t pool_id = -1;
+  std::string pool_namespace;
+  std::string image_id;
+  snapid_t snap_id = CEPH_NOSNAP;
+
+  ParentImageSpec() {
+  }
+  ParentImageSpec(int64_t pool_id, const std::string& pool_namespace,
+                  const std::string& image_id, snapid_t snap_id)
+    : pool_id(pool_id), pool_namespace(pool_namespace), image_id(image_id),
+      snap_id(snap_id) {
+  }
+
+  bool exists() const {
+    return (pool_id >= 0 && !image_id.empty() && snap_id != CEPH_NOSNAP);
+  }
+
+  bool operator==(const ParentImageSpec& rhs) const {
+    return ((pool_id == rhs.pool_id) &&
+            (pool_namespace == rhs.pool_namespace) &&
+            (image_id == rhs.image_id) &&
+            (snap_id == rhs.snap_id));
+  }
+
+  bool operator!=(const ParentImageSpec& rhs) const {
+    return !(*this == rhs);
+  }
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<ParentImageSpec*> &o);
+};
+
+WRITE_CLASS_ENCODER(ParentImageSpec);
+
+struct ChildImageSpec {
+  int64_t pool_id = -1;
+  std::string pool_namespace;
+  std::string image_id;
+
+  ChildImageSpec() {}
+  ChildImageSpec(int64_t pool_id, const std::string& pool_namespace,
+                 const std::string& image_id)
+    : pool_id(pool_id), pool_namespace(pool_namespace), image_id(image_id) {
+  }
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<ChildImageSpec*> &o);
+
+  inline bool operator==(const ChildImageSpec& rhs) const {
+    return (pool_id == rhs.pool_id &&
+            pool_namespace == rhs.pool_namespace &&
+            image_id == rhs.image_id);
+  }
+  inline bool operator<(const ChildImageSpec& rhs) const {
+    if (pool_id != rhs.pool_id) {
+      return pool_id < rhs.pool_id;
+    }
+    if (pool_namespace != rhs.pool_namespace) {
+      return pool_namespace < rhs.pool_namespace;
+    }
+    return image_id < rhs.image_id;
+  }
+};
+WRITE_CLASS_ENCODER(ChildImageSpec);
+
+typedef std::set<ChildImageSpec> ChildImageSpecs;
+
+struct GroupImageSpec {
+  GroupImageSpec() {}
+
+  GroupImageSpec(const std::string &image_id, int64_t pool_id)
+    : image_id(image_id), pool_id(pool_id) {}
+
+  static int from_key(const std::string &image_key, GroupImageSpec *spec);
+
+  std::string image_id;
+  int64_t pool_id = -1;
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<GroupImageSpec*> &o);
+
+  std::string image_key();
+
+};
+WRITE_CLASS_ENCODER(GroupImageSpec);
+
+struct GroupImageStatus {
+  GroupImageStatus() {}
+  GroupImageStatus(const std::string &image_id,
+		   int64_t pool_id,
+		   GroupImageLinkState state)
+    : spec(image_id, pool_id), state(state) {}
+
+  GroupImageStatus(GroupImageSpec spec,
+		   GroupImageLinkState state)
+    : spec(spec), state(state) {}
+
+  GroupImageSpec spec;
+  GroupImageLinkState state = GROUP_IMAGE_LINK_STATE_INCOMPLETE;
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<GroupImageStatus*> &o);
+
+  std::string state_to_string() const;
+};
+
+WRITE_CLASS_ENCODER(GroupImageStatus);
+
+struct GroupSpec {
+  GroupSpec() {}
+  GroupSpec(const std::string &group_id, int64_t pool_id)
+    : group_id(group_id), pool_id(pool_id) {}
+
+  std::string group_id;
+  int64_t pool_id = -1;
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+  bool is_valid() const;
+
+  static void generate_test_instances(std::list<GroupSpec *> &o);
+};
+
+WRITE_CLASS_ENCODER(GroupSpec);
+
+enum SnapshotNamespaceType {
+  SNAPSHOT_NAMESPACE_TYPE_USER  = 0,
+  SNAPSHOT_NAMESPACE_TYPE_GROUP = 1,
+  SNAPSHOT_NAMESPACE_TYPE_TRASH = 2
+};
+
+struct UserSnapshotNamespace {
+  static const SnapshotNamespaceType SNAPSHOT_NAMESPACE_TYPE =
+    SNAPSHOT_NAMESPACE_TYPE_USER;
+
+  UserSnapshotNamespace() {}
+
+  void encode(bufferlist& bl) const {}
+  void decode(bufferlist::const_iterator& it) {}
+
+  void dump(Formatter *f) const {}
+
+  inline bool operator==(const UserSnapshotNamespace& usn) const {
+    return true;
+  }
+
+  inline bool operator<(const UserSnapshotNamespace& usn) const {
+    return false;
+  }
+};
+
+struct GroupSnapshotNamespace {
+  static const SnapshotNamespaceType SNAPSHOT_NAMESPACE_TYPE =
+    SNAPSHOT_NAMESPACE_TYPE_GROUP;
+
+  GroupSnapshotNamespace() {}
+
+  GroupSnapshotNamespace(int64_t _group_pool,
+			 const string &_group_id,
+			 const string &_group_snapshot_id)
+    : group_id(_group_id), group_pool(_group_pool),
+      group_snapshot_id(_group_snapshot_id) {}
+
+  string group_id;
+  int64_t group_pool = 0;
+  string group_snapshot_id;
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& it);
+
+  void dump(Formatter *f) const;
+
+  inline bool operator==(const GroupSnapshotNamespace& gsn) const {
+    return group_pool == gsn.group_pool &&
+	   group_id == gsn.group_id &&
+	   group_snapshot_id == gsn.group_snapshot_id;
+  }
+
+  inline bool operator<(const GroupSnapshotNamespace& gsn) const {
+    if (group_pool < gsn.group_pool) {
+      return true;
+    } else if (group_id < gsn.group_id) {
+      return true;
+    } else {
+      return (group_snapshot_id < gsn.group_snapshot_id);
+    }
+    return false;
+  }
+};
+
+struct TrashSnapshotNamespace {
+  static const SnapshotNamespaceType SNAPSHOT_NAMESPACE_TYPE =
+    SNAPSHOT_NAMESPACE_TYPE_TRASH;
+
+  std::string original_name;
+  SnapshotNamespaceType original_snapshot_namespace_type =
+    SNAPSHOT_NAMESPACE_TYPE_USER;
+
+  TrashSnapshotNamespace() {}
+  TrashSnapshotNamespace(SnapshotNamespaceType original_snapshot_namespace_type,
+                         const std::string& original_name)
+    : original_name(original_name),
+      original_snapshot_namespace_type(original_snapshot_namespace_type) {}
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& it);
+  void dump(Formatter *f) const;
+
+  inline bool operator==(const TrashSnapshotNamespace& usn) const {
+    return true;
+  }
+  inline bool operator<(const TrashSnapshotNamespace& usn) const {
+    return false;
+  }
+};
+
+struct UnknownSnapshotNamespace {
+  static const SnapshotNamespaceType SNAPSHOT_NAMESPACE_TYPE =
+    static_cast<SnapshotNamespaceType>(-1);
+
+  UnknownSnapshotNamespace() {}
+
+  void encode(bufferlist& bl) const {}
+  void decode(bufferlist::const_iterator& it) {}
+  void dump(Formatter *f) const {}
+
+  inline bool operator==(const UnknownSnapshotNamespace& gsn) const {
+    return true;
+  }
+
+  inline bool operator<(const UnknownSnapshotNamespace& gsn) const {
+    return false;
+  }
+};
+
+std::ostream& operator<<(std::ostream& os, const SnapshotNamespaceType& type);
+std::ostream& operator<<(std::ostream& os, const UserSnapshotNamespace& ns);
+std::ostream& operator<<(std::ostream& os, const GroupSnapshotNamespace& ns);
+std::ostream& operator<<(std::ostream& os, const TrashSnapshotNamespace& ns);
+std::ostream& operator<<(std::ostream& os, const UnknownSnapshotNamespace& ns);
+
+typedef boost::variant<UserSnapshotNamespace,
+                       GroupSnapshotNamespace,
+                       TrashSnapshotNamespace,
+                       UnknownSnapshotNamespace> SnapshotNamespaceVariant;
+
+struct SnapshotNamespace : public SnapshotNamespaceVariant {
+  SnapshotNamespace() {
+  }
+
+  template <typename T>
+  SnapshotNamespace(T&& t) : SnapshotNamespaceVariant(std::forward<T>(t)) {
+  }
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<SnapshotNamespace*> &o);
+
+  inline bool operator==(const SnapshotNamespaceVariant& sn) const {
+    return static_cast<const SnapshotNamespaceVariant&>(*this) == sn;
+  }
+  inline bool operator<(const SnapshotNamespaceVariant& sn) const {
+    return static_cast<const SnapshotNamespaceVariant&>(*this) < sn;
+  }
+};
+WRITE_CLASS_ENCODER(SnapshotNamespace);
+
+SnapshotNamespaceType get_snap_namespace_type(
+    const SnapshotNamespace& snapshot_namespace);
+
+struct SnapshotInfo {
+  snapid_t id = CEPH_NOSNAP;
+  cls::rbd::SnapshotNamespace snapshot_namespace = {UserSnapshotNamespace{}};
+  std::string name;
+  uint64_t image_size = 0;
+  utime_t timestamp;
+  uint32_t child_count = 0;
+
+  SnapshotInfo() {
+  }
+  SnapshotInfo(snapid_t id,
+               const cls::rbd::SnapshotNamespace& snapshot_namespace,
+               const std::string& name, uint64_t image_size,
+               const utime_t& timestamp, uint32_t child_count)
+    : id(id), snapshot_namespace(snapshot_namespace),
+      name(name), image_size(image_size), timestamp(timestamp),
+      child_count(child_count) {
+  }
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<SnapshotInfo*> &o);
+};
+WRITE_CLASS_ENCODER(SnapshotInfo);
+
+enum GroupSnapshotState {
+  GROUP_SNAPSHOT_STATE_INCOMPLETE = 0,
+  GROUP_SNAPSHOT_STATE_COMPLETE = 1,
+};
+
+inline void encode(const GroupSnapshotState &state, bufferlist& bl, uint64_t features=0)
+{
+  using ceph::encode;
+  encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(GroupSnapshotState &state, bufferlist::const_iterator& it)
+{
+  using ceph::decode;
+  uint8_t int_state;
+  decode(int_state, it);
+  state = static_cast<GroupSnapshotState>(int_state);
+}
+
+struct ImageSnapshotSpec {
+  int64_t pool;
+  string image_id;
+  snapid_t snap_id;
+
+  ImageSnapshotSpec() {}
+  ImageSnapshotSpec(int64_t _pool,
+		    string _image_id,
+		    snapid_t _snap_id) : pool(_pool),
+					 image_id(_image_id),
+					 snap_id(_snap_id) {}
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& it);
+
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<ImageSnapshotSpec *> &o);
+};
+WRITE_CLASS_ENCODER(ImageSnapshotSpec);
+
+struct GroupSnapshot {
+  std::string id;
+  std::string name;
+  GroupSnapshotState state = GROUP_SNAPSHOT_STATE_INCOMPLETE;
+
+  GroupSnapshot() {}
+  GroupSnapshot(std::string _id,
+		std::string _name,
+		GroupSnapshotState _state) : id(_id),
+					     name(_name),
+					     state(_state) {}
+
+  vector<ImageSnapshotSpec> snaps;
+
+  void encode(bufferlist& bl) const;
+  void decode(bufferlist::const_iterator& it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<GroupSnapshot *> &o);
+};
+WRITE_CLASS_ENCODER(GroupSnapshot);
+enum TrashImageSource {
+  TRASH_IMAGE_SOURCE_USER = 0,
+  TRASH_IMAGE_SOURCE_MIRRORING = 1,
+  TRASH_IMAGE_SOURCE_MIGRATION = 2,
+  TRASH_IMAGE_SOURCE_REMOVING = 3,
+};
+
+inline std::ostream& operator<<(std::ostream& os,
+                                const TrashImageSource& source) {
+  switch (source) {
+  case TRASH_IMAGE_SOURCE_USER:
+    os << "user";
+    break;
+  case TRASH_IMAGE_SOURCE_MIRRORING:
+    os << "mirroring";
+    break;
+  case TRASH_IMAGE_SOURCE_MIGRATION:
+    os << "migration";
+    break;
+  case TRASH_IMAGE_SOURCE_REMOVING:
+    os << "removing";
+    break;
+  default:
+    os << "unknown (" << static_cast<uint32_t>(source) << ")";
+    break;
+  }
+  return os;
+}
+
+inline void encode(const TrashImageSource &source, bufferlist& bl,
+		   uint64_t features=0)
+{
+  using ceph::encode;
+  encode(static_cast<uint8_t>(source), bl);
+}
+
+inline void decode(TrashImageSource &source, bufferlist::const_iterator& it)
+{
+  uint8_t int_source;
+  using ceph::decode;
+  decode(int_source, it);
+  source = static_cast<TrashImageSource>(int_source);
+}
+
+enum TrashImageState {
+  TRASH_IMAGE_STATE_NORMAL    = 0,
+  TRASH_IMAGE_STATE_MOVING    = 1,
+  TRASH_IMAGE_STATE_REMOVING  = 2,
+  TRASH_IMAGE_STATE_RESTORING = 3
+};
+
+inline void encode(const TrashImageState &state, bufferlist &bl)
+{
+  using ceph::encode;
+  encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(TrashImageState &state, bufferlist::const_iterator &it)
+{
+  uint8_t int_state;
+  using ceph::decode;
+  decode(int_state, it);
+  state = static_cast<TrashImageState>(int_state);
+}
+
+struct TrashImageSpec {
+  TrashImageSource source = TRASH_IMAGE_SOURCE_USER;
+  std::string name;
+  utime_t deletion_time; // time of deletion
+  utime_t deferment_end_time;
+  TrashImageState state = TRASH_IMAGE_STATE_NORMAL;
+
+  TrashImageSpec() {}
+  TrashImageSpec(TrashImageSource source, const std::string &name,
+                 const utime_t& deletion_time,
+                 const utime_t& deferment_end_time)
+    : source(source), name(name), deletion_time(deletion_time),
+      deferment_end_time(deferment_end_time) {
+  }
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator& it);
+  void dump(Formatter *f) const;
+
+  inline bool operator==(const TrashImageSpec& rhs) const {
+    return (source == rhs.source &&
+            name == rhs.name &&
+            deletion_time == rhs.deletion_time &&
+            deferment_end_time == rhs.deferment_end_time);
+  }
+};
+WRITE_CLASS_ENCODER(TrashImageSpec);
+
+struct MirrorImageMap {
+  MirrorImageMap() {
+  }
+
+  MirrorImageMap(const std::string &instance_id, utime_t mapped_time,
+                 const bufferlist &data)
+    : instance_id(instance_id),
+      mapped_time(mapped_time),
+      data(data) {
+  }
+
+  std::string instance_id;
+  utime_t mapped_time;
+  bufferlist data;
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator &it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<MirrorImageMap*> &o);
+
+  bool operator==(const MirrorImageMap &rhs) const;
+  bool operator<(const MirrorImageMap &rhs) const;
+};
+
+std::ostream& operator<<(std::ostream& os, const MirrorImageMap &image_map);
+
+WRITE_CLASS_ENCODER(MirrorImageMap);
+
+enum MigrationHeaderType {
+  MIGRATION_HEADER_TYPE_SRC = 1,
+  MIGRATION_HEADER_TYPE_DST = 2,
+};
+
+inline void encode(const MigrationHeaderType &type, bufferlist& bl) {
+  using ceph::encode;
+  encode(static_cast<uint8_t>(type), bl);
+}
+
+inline void decode(MigrationHeaderType &type, bufferlist::const_iterator& it) {
+  uint8_t int_type;
+  using ceph::decode;
+  decode(int_type, it);
+  type = static_cast<MigrationHeaderType>(int_type);
+}
+
+enum MigrationState {
+  MIGRATION_STATE_ERROR = 0,
+  MIGRATION_STATE_PREPARING = 1,
+  MIGRATION_STATE_PREPARED = 2,
+  MIGRATION_STATE_EXECUTING = 3,
+  MIGRATION_STATE_EXECUTED = 4,
+  MIGRATION_STATE_ABORTING = 5,
+};
+
+inline void encode(const MigrationState &state, bufferlist& bl) {
+  using ceph::encode;
+  encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(MigrationState &state, bufferlist::const_iterator& it) {
+  uint8_t int_state;
+  using ceph::decode;
+  decode(int_state, it);
+  state = static_cast<MigrationState>(int_state);
+}
+
+std::ostream& operator<<(std::ostream& os,
+                         const MigrationState& migration_state);
+
+struct MigrationSpec {
+  MigrationHeaderType header_type = MIGRATION_HEADER_TYPE_SRC;
+  int64_t pool_id = -1;
+  std::string pool_namespace;
+  std::string image_name;
+  std::string image_id;
+  std::map<uint64_t, uint64_t> snap_seqs;
+  uint64_t overlap = 0;
+  bool flatten = false;
+  bool mirroring = false;
+  MigrationState state = MIGRATION_STATE_ERROR;
+  std::string state_description;
+
+  MigrationSpec() {
+  }
+  MigrationSpec(MigrationHeaderType header_type, int64_t pool_id,
+                const std::string& pool_namespace,
+                const std::string &image_name, const std::string &image_id,
+                const std::map<uint64_t, uint64_t> &snap_seqs, uint64_t overlap,
+                bool mirroring, bool flatten, MigrationState state,
+                const std::string &state_description)
+    : header_type(header_type), pool_id(pool_id),
+      pool_namespace(pool_namespace), image_name(image_name),
+      image_id(image_id), snap_seqs(snap_seqs), overlap(overlap),
+      flatten(flatten), mirroring(mirroring), state(state),
+      state_description(state_description) {
+  }
+
+  void encode(bufferlist &bl) const;
+  void decode(bufferlist::const_iterator& it);
+  void dump(Formatter *f) const;
+
+  static void generate_test_instances(std::list<MigrationSpec*> &o);
+
+  inline bool operator==(const MigrationSpec& ms) const {
+    return header_type == ms.header_type && pool_id == ms.pool_id &&
+      pool_namespace == ms.pool_namespace && image_name == ms.image_name &&
+      image_id == ms.image_id && snap_seqs == ms.snap_seqs &&
+      overlap == ms.overlap && flatten == ms.flatten &&
+      mirroring == ms.mirroring && state == ms.state &&
+      state_description == ms.state_description;
+  }
+};
+
+std::ostream& operator<<(std::ostream& os, const MigrationSpec& migration_spec);
+
+WRITE_CLASS_ENCODER(MigrationSpec);
+
+enum AssertSnapcSeqState {
+  ASSERT_SNAPC_SEQ_GT_SNAPSET_SEQ = 0,
+  ASSERT_SNAPC_SEQ_LE_SNAPSET_SEQ = 1,
+};
+
+inline void encode(const AssertSnapcSeqState &state, bufferlist& bl) {
+  using ceph::encode;
+  encode(static_cast<uint8_t>(state), bl);
+}
+
+inline void decode(AssertSnapcSeqState &state, bufferlist::const_iterator& it) {
+  uint8_t int_state;
+  using ceph::decode;
+  decode(int_state, it);
+  state = static_cast<AssertSnapcSeqState>(int_state);
+}
+
+std::ostream& operator<<(std::ostream& os, const AssertSnapcSeqState& state);
+
+} // namespace rbd
+} // namespace cls
+
+#endif // CEPH_CLS_RBD_TYPES_H
diff --git a/src/cls/refcount/cls_refcount.cc b/src/cls/refcount/cls_refcount.cc
new file mode 100644
index 00000000..73cc3b85
--- /dev/null
+++ b/src/cls/refcount/cls_refcount.cc
@@ -0,0 +1,216 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "objclass/objclass.h"
+#include "cls/refcount/cls_refcount_ops.h"
+
+#include "include/compat.h"
+
+CLS_VER(1,0)
+CLS_NAME(refcount)
+
+
+#define REFCOUNT_ATTR "refcount"
+
+static string wildcard_tag;
+
+static int read_refcount(cls_method_context_t hctx, bool implicit_ref, obj_refcount *objr)
+{
+  bufferlist bl;
+  objr->refs.clear();
+  int ret = cls_cxx_getxattr(hctx, REFCOUNT_ATTR, &bl);
+  if (ret == -ENODATA) {
+    if (implicit_ref) {
+      objr->refs[wildcard_tag] = true;
+    }
+    return 0;
+  }
+  if (ret < 0)
+    return ret;
+
+  try {
+    auto iter = bl.cbegin();
+    decode(*objr, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: read_refcount(): failed to decode refcount entry\n");
+    return -EIO;
+  }
+
+  return 0;
+}
+
+static int set_refcount(cls_method_context_t hctx, const struct obj_refcount& objr)
+{
+  bufferlist bl;
+
+  encode(objr, bl);
+
+  int ret = cls_cxx_setxattr(hctx, REFCOUNT_ATTR, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_refcount_get(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_refcount_get_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rc_refcount_get(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  obj_refcount objr;
+  int ret = read_refcount(hctx, op.implicit_ref, &objr);
+  if (ret < 0)
+    return ret;
+
+  CLS_LOG(10, "cls_rc_refcount_get() tag=%s\n", op.tag.c_str());
+
+  objr.refs[op.tag] = true;
+
+  ret = set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_refcount_put(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_refcount_put_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rc_refcount_put(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  obj_refcount objr;
+  int ret = read_refcount(hctx, op.implicit_ref, &objr);
+  if (ret < 0)
+    return ret;
+
+  if (objr.refs.empty()) {// shouldn't happen!
+    CLS_LOG(0, "ERROR: cls_rc_refcount_put() was called without any references!\n");
+    return -EINVAL;
+  }
+
+  CLS_LOG(10, "cls_rc_refcount_put() tag=%s\n", op.tag.c_str());
+
+  bool found = false;
+  map<string, bool>::iterator iter = objr.refs.find(op.tag);
+  if (iter != objr.refs.end()) {
+    found = true;
+  } else if (op.implicit_ref) {
+    iter = objr.refs.find(wildcard_tag);
+    if (iter != objr.refs.end()) {
+      found = true;
+    }
+  }
+
+  if (!found ||
+      objr.retired_refs.find(op.tag) != objr.retired_refs.end())
+    return 0;
+
+  objr.retired_refs.insert(op.tag);
+  objr.refs.erase(iter);
+
+  if (objr.refs.empty()) {
+    return cls_cxx_remove(hctx);
+  }
+
+  ret = set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_refcount_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_refcount_set_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_refcount_set(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  if (!op.refs.size()) {
+    return cls_cxx_remove(hctx);
+  }
+
+  obj_refcount objr;
+  list<string>::iterator iter;
+  for (iter = op.refs.begin(); iter != op.refs.end(); ++iter) {
+    objr.refs[*iter] = true;
+  }
+
+  int ret = set_refcount(hctx, objr);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_rc_refcount_read(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_refcount_read_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rc_refcount_read(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  obj_refcount objr;
+
+  cls_refcount_read_ret read_ret;
+  int ret = read_refcount(hctx, op.implicit_ref, &objr);
+  if (ret < 0)
+    return ret;
+
+  map<string, bool>::iterator iter;
+  for (iter = objr.refs.begin(); iter != objr.refs.end(); ++iter) {
+    read_ret.refs.push_back(iter->first);
+  }
+
+  encode(read_ret, *out);
+
+  return 0;
+}
+
+CLS_INIT(refcount)
+{
+  CLS_LOG(1, "Loaded refcount class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_refcount_get;
+  cls_method_handle_t h_refcount_put;
+  cls_method_handle_t h_refcount_set;
+  cls_method_handle_t h_refcount_read;
+
+  cls_register("refcount", &h_class);
+
+  /* refcount */
+  cls_register_cxx_method(h_class, "get", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_refcount_get, &h_refcount_get);
+  cls_register_cxx_method(h_class, "put", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_refcount_put, &h_refcount_put);
+  cls_register_cxx_method(h_class, "set", CLS_METHOD_RD | CLS_METHOD_WR, cls_rc_refcount_set, &h_refcount_set);
+  cls_register_cxx_method(h_class, "read", CLS_METHOD_RD, cls_rc_refcount_read, &h_refcount_read);
+
+  return;
+}
+
diff --git a/src/cls/refcount/cls_refcount_client.cc b/src/cls/refcount/cls_refcount_client.cc
new file mode 100644
index 00000000..9d5210c8
--- /dev/null
+++ b/src/cls/refcount/cls_refcount_client.cc
@@ -0,0 +1,61 @@
+#include <errno.h>
+
+#include "cls/refcount/cls_refcount_client.h"
+#include "cls/refcount/cls_refcount_ops.h"
+#include "include/rados/librados.hpp"
+
+using namespace librados;
+
+
+void cls_refcount_get(librados::ObjectWriteOperation& op, const string& tag, bool implicit_ref)
+{
+  bufferlist in;
+  cls_refcount_get_op call;
+  call.tag = tag;
+  call.implicit_ref = implicit_ref;
+  encode(call, in);
+  op.exec("refcount", "get", in);
+}
+
+void cls_refcount_put(librados::ObjectWriteOperation& op, const string& tag, bool implicit_ref)
+{
+  bufferlist in;
+  cls_refcount_put_op call;
+  call.tag = tag;
+  call.implicit_ref = implicit_ref;
+  encode(call, in);
+  op.exec("refcount", "put", in);
+}
+
+void cls_refcount_set(librados::ObjectWriteOperation& op, list<string>& refs)
+{
+  bufferlist in;
+  cls_refcount_set_op call;
+  call.refs = refs;
+  encode(call, in);
+  op.exec("refcount", "set", in);
+}
+
+int cls_refcount_read(librados::IoCtx& io_ctx, string& oid, list<string> *refs, bool implicit_ref)
+{
+  bufferlist in, out;
+  cls_refcount_read_op call;
+  call.implicit_ref = implicit_ref;
+  encode(call, in);
+  int r = io_ctx.exec(oid, "refcount", "read", in, out);
+  if (r < 0)
+    return r;
+
+  cls_refcount_read_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  *refs = ret.refs;
+
+  return r;
+}
+
diff --git a/src/cls/refcount/cls_refcount_client.h b/src/cls/refcount/cls_refcount_client.h
new file mode 100644
index 00000000..c41a4f6c
--- /dev/null
+++ b/src/cls/refcount/cls_refcount_client.h
@@ -0,0 +1,34 @@
+#ifndef CEPH_CLS_REFCOUNT_CLIENT_H
+#define CEPH_CLS_REFCOUNT_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include "include/types.h"
+
+/*
+ * refcount objclass
+ *
+ * The refcount objclass implements a refcounting scheme that allows having multiple references
+ * to a single rados object. The canonical way to use it is to add a reference and to remove a
+ * reference using a specific tag. This way we ensure that refcounting operations are idempotent,
+ * that is, a single client can only increase/decrease the refcount once using a single tag, so
+ * any replay of operations (implicit or explicit) is possible.
+ *
+ * So, the regular usage would be to create an object, to increase the refcount. Then, when
+ * wanting to have another reference to it, increase the refcount using a different tag. When
+ * removing a reference it is required to drop the refcount (using the same tag that was used
+ * for that reference). When the refcount drops to zero, the object is removed automaticfally.
+ *
+ * In order to maintain backwards compatibility with objects that were created without having
+ * their refcount increased, the implicit_ref was added. Any object that was created without
+ * having it's refcount increased (explicitly) is having an implicit refcount of 1. Since
+ * we don't have a tag for this refcount, we consider this tag as a wildcard. So if the refcount
+ * is being decreased by an unknown tag and we still have one wildcard tag, we'll accept it
+ * as the relevant tag, and the refcount will be decreased.
+ */
+
+void cls_refcount_get(librados::ObjectWriteOperation& op, const string& tag, bool implicit_ref = false);
+void cls_refcount_put(librados::ObjectWriteOperation& op, const string& tag, bool implicit_ref = false);
+void cls_refcount_set(librados::ObjectWriteOperation& op, list<string>& refs);
+int cls_refcount_read(librados::IoCtx& io_ctx, string& oid, list<string> *refs, bool implicit_ref = false);
+
+#endif
diff --git a/src/cls/refcount/cls_refcount_ops.cc b/src/cls/refcount/cls_refcount_ops.cc
new file mode 100644
index 00000000..ca7f93ed
--- /dev/null
+++ b/src/cls/refcount/cls_refcount_ops.cc
@@ -0,0 +1,104 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls_refcount_ops.h"
+#include "common/Formatter.h"
+#include "common/ceph_json.h"
+
+void cls_refcount_get_op::dump(ceph::Formatter *f) const
+{
+  f->dump_string("tag", tag);
+  f->dump_int("implicit_ref", (int)implicit_ref);
+}
+
+void cls_refcount_get_op::generate_test_instances(list<cls_refcount_get_op*>& ls)
+{
+  ls.push_back(new cls_refcount_get_op);
+  ls.push_back(new cls_refcount_get_op);
+  ls.back()->tag = "foo";
+  ls.back()->implicit_ref = true;
+}
+
+
+void cls_refcount_put_op::dump(ceph::Formatter *f) const
+{
+  f->dump_string("tag", tag);
+  f->dump_int("implicit_ref", (int)implicit_ref);
+}
+
+void cls_refcount_put_op::generate_test_instances(list<cls_refcount_put_op*>& ls)
+{
+  ls.push_back(new cls_refcount_put_op);
+  ls.push_back(new cls_refcount_put_op);
+  ls.back()->tag = "foo";
+  ls.back()->implicit_ref = true;
+}
+
+
+
+void cls_refcount_set_op::dump(ceph::Formatter *f) const
+{
+  encode_json("refs", refs, f);
+}
+
+void cls_refcount_set_op::generate_test_instances(list<cls_refcount_set_op*>& ls)
+{
+  ls.push_back(new cls_refcount_set_op);
+  ls.push_back(new cls_refcount_set_op);
+  ls.back()->refs.push_back("foo");
+  ls.back()->refs.push_back("bar");
+}
+
+
+void cls_refcount_read_op::dump(ceph::Formatter *f) const
+{
+  f->dump_int("implicit_ref", (int)implicit_ref);
+}
+
+void cls_refcount_read_op::generate_test_instances(list<cls_refcount_read_op*>& ls)
+{
+  ls.push_back(new cls_refcount_read_op);
+  ls.push_back(new cls_refcount_read_op);
+  ls.back()->implicit_ref = true;
+}
+
+
+void cls_refcount_read_ret::dump(ceph::Formatter *f) const
+{
+  f->open_array_section("refs");
+  for (list<string>::const_iterator p = refs.begin(); p != refs.end(); ++p)
+    f->dump_string("ref", *p);
+  f->close_section();
+}
+
+void cls_refcount_read_ret::generate_test_instances(list<cls_refcount_read_ret*>& ls)
+{
+  ls.push_back(new cls_refcount_read_ret);
+  ls.push_back(new cls_refcount_read_ret);
+  ls.back()->refs.push_back("foo");
+  ls.back()->refs.push_back("bar");
+}
+
+void obj_refcount::dump(ceph::Formatter *f) const
+{
+  f->open_array_section("refs");
+  for (const auto &kv: refs) {
+    f->open_object_section("ref");
+    f->dump_string("oid", kv.first.c_str());
+    f->dump_bool("active",kv.second);
+    f->close_section();
+  }
+  f->close_section();
+
+  f->open_array_section("retired_refs");
+  for (const auto& it: retired_refs)
+    f->dump_string("ref", it.c_str());
+  f->close_section();
+}
+
+void obj_refcount::generate_test_instances(list<obj_refcount*>& ls)
+{
+  ls.push_back(new obj_refcount);
+  ls.back()->refs.emplace("foo",true);
+  ls.back()->retired_refs.emplace("bar");
+}
diff --git a/src/cls/refcount/cls_refcount_ops.h b/src/cls/refcount/cls_refcount_ops.h
new file mode 100644
index 00000000..946c11aa
--- /dev/null
+++ b/src/cls/refcount/cls_refcount_ops.h
@@ -0,0 +1,154 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_REFCOUNT_OPS_H
+#define CEPH_CLS_REFCOUNT_OPS_H
+
+#include "include/types.h"
+#include "common/hobject.h"
+
+struct cls_refcount_get_op {
+  string tag;
+  bool implicit_ref;
+
+  cls_refcount_get_op() : implicit_ref(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(tag, bl);
+    encode(implicit_ref, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(tag, bl);
+    decode(implicit_ref, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_refcount_get_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_refcount_get_op)
+
+struct cls_refcount_put_op {
+  string tag;
+  bool implicit_ref; // assume wildcard reference for
+                          // objects without a set ref
+
+  cls_refcount_put_op() : implicit_ref(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(tag, bl);
+    encode(implicit_ref, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(tag, bl);
+    decode(implicit_ref, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_refcount_put_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_refcount_put_op)
+
+struct cls_refcount_set_op {
+  list<string> refs;
+
+  cls_refcount_set_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(refs, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_refcount_set_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_refcount_set_op)
+
+struct cls_refcount_read_op {
+  bool implicit_ref; // assume wildcard reference for
+                          // objects without a set ref
+
+  cls_refcount_read_op() : implicit_ref(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(implicit_ref, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(implicit_ref, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_refcount_read_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_refcount_read_op)
+
+struct cls_refcount_read_ret {
+  list<string> refs;
+
+  cls_refcount_read_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(refs, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<cls_refcount_read_ret*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_refcount_read_ret)
+
+struct obj_refcount {
+  map<string, bool> refs;
+  set<string> retired_refs;
+
+  obj_refcount() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(refs, bl);
+    encode(retired_refs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(refs, bl);
+    if (struct_v >= 2) {
+      decode(retired_refs, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  void dump(ceph::Formatter *f) const;
+  static void generate_test_instances(list<obj_refcount*>& ls);
+};
+WRITE_CLASS_ENCODER(obj_refcount)
+
+#endif
diff --git a/src/cls/rgw/cls_rgw.cc b/src/cls/rgw/cls_rgw.cc
new file mode 100644
index 00000000..741d8c57
--- /dev/null
+++ b/src/cls/rgw/cls_rgw.cc
@@ -0,0 +1,4031 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "include/types.h"
+
+#include <errno.h>
+
+#include "objclass/objclass.h"
+#include "cls/rgw/cls_rgw_ops.h"
+#include "cls/rgw/cls_rgw_const.h"
+#include "common/Clock.h"
+#include "common/strtol.h"
+#include "common/escape.h"
+
+#include "include/compat.h"
+
+CLS_VER(1,0)
+CLS_NAME(rgw)
+
+
+#define BI_PREFIX_CHAR 0x80
+
+#define BI_BUCKET_OBJS_INDEX          0
+#define BI_BUCKET_LOG_INDEX           1
+#define BI_BUCKET_OBJ_INSTANCE_INDEX  2
+#define BI_BUCKET_OLH_DATA_INDEX      3
+
+#define BI_BUCKET_LAST_INDEX          4
+
+static std::string bucket_index_prefixes[] = { "", /* special handling for the objs list index */
+                                          "0_",     /* bucket log index */
+                                          "1000_",  /* obj instance index */
+                                          "1001_",  /* olh data index */
+
+                                          /* this must be the last index */
+                                          "9999_",};
+
+static bool bi_is_objs_index(const string& s) {
+  return ((unsigned char)s[0] != BI_PREFIX_CHAR);
+}
+
+int bi_entry_type(const string& s)
+{
+  if (bi_is_objs_index(s)) {
+    return BI_BUCKET_OBJS_INDEX;
+  }
+
+  for (size_t i = 1;
+       i < sizeof(bucket_index_prefixes) / sizeof(bucket_index_prefixes[0]);
+       ++i) {
+    const string& t = bucket_index_prefixes[i];
+
+    if (s.compare(1, t.size(), t) == 0) {
+      return i;
+    }
+  }
+
+  return -EINVAL;
+}
+
+static bool bi_entry_gt(const string& first, const string& second)
+{
+  int fi = bi_entry_type(first);
+  int si = bi_entry_type(second);
+
+  if (fi > si) {
+    return true;
+  } else if (fi < si) {
+    return false;
+  }
+
+  return first > second;
+}
+
+static void get_time_key(real_time& ut, string *key)
+{
+  char buf[32];
+  ceph_timespec ts = ceph::real_clock::to_ceph_timespec(ut);
+  snprintf(buf, 32, "%011llu.%09u", (unsigned long long)ts.tv_sec, (unsigned int)ts.tv_nsec);
+  *key = buf;
+}
+
+static void get_index_ver_key(cls_method_context_t hctx, uint64_t index_ver, string *key)
+{
+  char buf[48];
+  snprintf(buf, sizeof(buf), "%011llu.%llu.%d", (unsigned long long)index_ver,
+           (unsigned long long)cls_current_version(hctx),
+           cls_current_subop_num(hctx));
+  *key = buf;
+}
+
+static void bi_log_prefix(string& key)
+{
+  key = BI_PREFIX_CHAR;
+  key.append(bucket_index_prefixes[BI_BUCKET_LOG_INDEX]);
+}
+
+static void bi_log_index_key(cls_method_context_t hctx, string& key, string& id, uint64_t index_ver)
+{
+  bi_log_prefix(key);
+  get_index_ver_key(hctx, index_ver, &id);
+  key.append(id);
+}
+
+static int log_index_operation(cls_method_context_t hctx, cls_rgw_obj_key& obj_key, RGWModifyOp op,
+                               string& tag, real_time& timestamp,
+                               rgw_bucket_entry_ver& ver, RGWPendingState state, uint64_t index_ver,
+                               string& max_marker, uint16_t bilog_flags, string *owner, string *owner_display_name, rgw_zone_set *zones_trace)
+{
+  bufferlist bl;
+
+  rgw_bi_log_entry entry;
+
+  entry.object = obj_key.name;
+  entry.instance = obj_key.instance;
+  entry.timestamp = timestamp;
+  entry.op = op;
+  entry.ver = ver;
+  entry.state = state;
+  entry.index_ver = index_ver;
+  entry.tag = tag;
+  entry.bilog_flags = bilog_flags;
+  if (owner) {
+    entry.owner = *owner;
+  }
+  if (owner_display_name) {
+    entry.owner_display_name = *owner_display_name;
+  }
+  if (zones_trace) {
+    entry.zones_trace = std::move(*zones_trace);
+  }
+
+  string key;
+  bi_log_index_key(hctx, key, entry.id, index_ver);
+
+  encode(entry, bl);
+
+  if (entry.id > max_marker)
+    max_marker = entry.id;
+
+  return cls_cxx_map_set_val(hctx, key, &bl);
+}
+
+/*
+ * read list of objects, skips objects in the ugly namespace
+ */
+static int get_obj_vals(cls_method_context_t hctx, const string& start, const string& filter_prefix,
+                        int num_entries, map<string, bufferlist> *pkeys, bool *pmore)
+{
+  int ret = cls_cxx_map_get_vals(hctx, start, filter_prefix, num_entries, pkeys, pmore);
+  if (ret < 0)
+    return ret;
+
+  if (pkeys->empty())
+    return 0;
+
+  auto last_element = pkeys->rbegin();
+  if ((unsigned char)last_element->first[0] < BI_PREFIX_CHAR) {
+    /* nothing to see here, move along */
+    return 0;
+  }
+
+  auto first_element = pkeys->begin();
+  if ((unsigned char)first_element->first[0] > BI_PREFIX_CHAR) {
+    return 0;
+  }
+
+  /* let's rebuild the list, only keep entries we're interested in */
+  auto comp = [](const pair<string, bufferlist>& l, const string &r) { return l.first < r; };
+  string new_start = {static_cast<char>(BI_PREFIX_CHAR + 1)};
+
+  auto lower = pkeys->lower_bound(string{static_cast<char>(BI_PREFIX_CHAR)});
+  auto upper = std::lower_bound(lower, pkeys->end(), new_start, comp);
+  pkeys->erase(lower, upper);
+
+  if (num_entries == (int)pkeys->size() || !(*pmore))
+    return 0;
+
+  if (pkeys->size() && new_start < pkeys->rbegin()->first) {
+    new_start = pkeys->rbegin()->first;
+  }
+
+  map<string, bufferlist> new_keys;
+
+  /* now get some more keys */
+  ret = cls_cxx_map_get_vals(hctx, new_start, filter_prefix, num_entries - pkeys->size(), &new_keys, pmore);
+  if (ret < 0)
+    return ret;
+
+  pkeys->insert(std::make_move_iterator(new_keys.begin()),
+                std::make_move_iterator(new_keys.end()));
+  return 0;
+}
+
+/*
+ * get a monotonically decreasing string representation.
+ * For num = x, num = y, where x > y, str(x) < str(y)
+ * Another property is that string size starts short and grows as num increases
+ */
+static void decreasing_str(uint64_t num, string *str)
+{
+  char buf[32];
+  if (num < 0x10) { /* 16 */
+    snprintf(buf, sizeof(buf), "9%02lld", 15 - (long long)num);
+  } else if (num < 0x100) { /* 256 */
+    snprintf(buf, sizeof(buf), "8%03lld", 255 - (long long)num);
+  } else if (num < 0x1000) /* 4096 */ {
+    snprintf(buf, sizeof(buf), "7%04lld", 4095 - (long long)num);
+  } else if (num < 0x10000) /* 65536 */ {
+    snprintf(buf, sizeof(buf), "6%05lld", 65535 - (long long)num);
+  } else if (num < 0x100000000) /* 4G */ {
+    snprintf(buf, sizeof(buf), "5%010lld", 0xFFFFFFFF - (long long)num);
+  } else {
+    snprintf(buf, sizeof(buf), "4%020lld",  (long long)-num);
+  }
+
+  *str = buf;
+}
+
+/*
+ * we now hold two different indexes for objects. The first one holds the list of objects in the
+ * order that we want them to be listed. The second one only holds the objects instances (for
+ * versioned objects), and they're not arranged in any particular order.
+ * When listing objects we'll use the first index, when doing operations on the objects themselves
+ * we'll use the second index. Note that regular objects only map to the first index anyway
+ */
+
+static void get_list_index_key(rgw_bucket_dir_entry& entry, string *index_key)
+{
+  *index_key = entry.key.name;
+
+  string ver_str;
+  decreasing_str(entry.versioned_epoch, &ver_str);
+  string instance_delim("\0i", 2);
+  string ver_delim("\0v", 2);
+
+  index_key->append(ver_delim);
+  index_key->append(ver_str);
+  index_key->append(instance_delim);
+  index_key->append(entry.key.instance);
+}
+
+static void encode_obj_versioned_data_key(const cls_rgw_obj_key& key, string *index_key, bool append_delete_marker_suffix = false)
+{
+  *index_key = BI_PREFIX_CHAR;
+  index_key->append(bucket_index_prefixes[BI_BUCKET_OBJ_INSTANCE_INDEX]);
+  index_key->append(key.name);
+  string delim("\0i", 2);
+  index_key->append(delim);
+  index_key->append(key.instance);
+  if (append_delete_marker_suffix) {
+    string dm("\0d", 2);
+    index_key->append(dm);
+  }
+}
+
+static void encode_obj_index_key(const cls_rgw_obj_key& key, string *index_key)
+{
+  if (key.instance.empty()) {
+    *index_key = key.name;
+  } else {
+    encode_obj_versioned_data_key(key, index_key);
+  }
+}
+
+static void encode_olh_data_key(const cls_rgw_obj_key& key, string *index_key)
+{
+  *index_key = BI_PREFIX_CHAR;
+  index_key->append(bucket_index_prefixes[BI_BUCKET_OLH_DATA_INDEX]);
+  index_key->append(key.name);
+}
+
+template <class T>
+static int read_index_entry(cls_method_context_t hctx, string& name, T *entry);
+
+static int encode_list_index_key(cls_method_context_t hctx, const cls_rgw_obj_key& key, string *index_key)
+{
+  if (key.instance.empty()) {
+    *index_key = key.name;
+    return 0;
+  }
+
+  string obj_index_key;
+  cls_rgw_obj_key tmp_key(key);
+  if (tmp_key.instance == "null") {
+    tmp_key.instance.clear();
+  }
+  encode_obj_versioned_data_key(tmp_key, &obj_index_key);
+
+  rgw_bucket_dir_entry entry;
+
+  int ret = read_index_entry(hctx, obj_index_key, &entry);
+  if (ret == -ENOENT) {
+   /* couldn't find the entry, set key value after the current object */
+    char buf[2] = { 0x1, 0 };
+    string s(buf);
+    *index_key  = key.name + s;
+    return 0;
+  }
+  if (ret < 0) {
+    CLS_LOG(1, "ERROR: encode_list_index_key(): cls_cxx_map_get_val returned %d\n", ret);
+    return ret;
+  }
+
+  get_list_index_key(entry, index_key);
+
+  return 0;
+}
+
+static void split_key(const string& key, list<string>& vals)
+{
+  size_t pos = 0;
+  const char *p = key.c_str();
+  while (pos < key.size()) {
+    size_t len = strlen(p);
+    vals.push_back(p);
+    pos += len + 1;
+    p += len + 1;
+  }
+}
+
+static string escape_str(const string& s)
+{
+  int len = escape_json_attr_len(s.c_str(), s.size());
+  std::string escaped(len, 0);
+  escape_json_attr(s.c_str(), s.size(), escaped.data());
+  return escaped;
+}
+
+/*
+ * list index key structure:
+ *
+ * <obj name>\0[v<ver>\0i<instance id>]
+ */
+static int decode_list_index_key(const string& index_key, cls_rgw_obj_key *key, uint64_t *ver)
+{
+  size_t len = strlen(index_key.c_str());
+
+  key->instance.clear();
+  *ver = 0;
+
+  if (len == index_key.size()) {
+    key->name = index_key;
+    return 0;
+  }
+
+  list<string> vals;
+  split_key(index_key, vals);
+
+  if (vals.empty()) {
+    CLS_LOG(0, "ERROR: %s(): bad index_key (%s): split_key() returned empty vals", __func__, escape_str(index_key).c_str());
+    return -EIO;
+  }
+
+  list<string>::iterator iter = vals.begin();
+  key->name = *iter;
+  ++iter;
+
+  if (iter == vals.end()) {
+    CLS_LOG(0, "ERROR: %s(): bad index_key (%s): no vals", __func__, escape_str(index_key).c_str());
+    return -EIO;
+  }
+
+  for (; iter != vals.end(); ++iter) {
+    string& val = *iter;
+    if (val[0] == 'i') {
+      key->instance = val.substr(1);
+    } else if (val[0] == 'v') {
+      string err;
+      const char *s = val.c_str() + 1;
+      *ver = strict_strtoll(s, 10, &err);
+      if (!err.empty()) {
+        CLS_LOG(0, "ERROR: %s(): bad index_key (%s): could not parse val (v=%s)", __func__, escape_str(index_key).c_str(), s);
+        return -EIO;
+      }
+    }
+  }
+
+  return 0;
+}
+
+static int read_bucket_header(cls_method_context_t hctx,
+			      rgw_bucket_dir_header *header)
+{
+  bufferlist bl;
+  int rc = cls_cxx_map_read_header(hctx, &bl);
+  if (rc < 0)
+    return rc;
+
+  if (bl.length() == 0) {
+      *header = rgw_bucket_dir_header();
+      return 0;
+  }
+  auto iter = bl.cbegin();
+  try {
+    decode(*header, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: read_bucket_header(): failed to decode header\n");
+    return -EIO;
+  }
+
+  return 0;
+}
+
+int rgw_bucket_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto iter = in->cbegin();
+
+  rgw_cls_list_op op;
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_bucket_list(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  rgw_cls_list_ret ret;
+  rgw_bucket_dir& new_dir = ret.dir;
+  int rc = read_bucket_header(hctx, &new_dir.header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_list(): failed to read header\n");
+    return rc;
+  }
+
+  map<string, bufferlist> keys;
+  std::map<string, bufferlist>::iterator kiter;
+  string start_key;
+  encode_list_index_key(hctx, op.start_obj, &start_key);
+  bool done = false;
+  uint32_t left_to_read = op.num_entries;
+  bool more;
+
+  do {
+    rc = get_obj_vals(hctx, start_key, op.filter_prefix, left_to_read, &keys, &more);
+    if (rc < 0)
+      return rc;
+
+    std::map<string, rgw_bucket_dir_entry>& m = new_dir.m;
+
+    done = keys.empty();
+
+    for (kiter = keys.begin(); kiter != keys.end(); ++kiter) {
+      rgw_bucket_dir_entry entry;
+
+      if (!bi_is_objs_index(kiter->first)) {
+        done = true;
+        break;
+      }
+
+      bufferlist& entrybl = kiter->second;
+      auto eiter = entrybl.cbegin();
+      try {
+        decode(entry, eiter);
+      } catch (buffer::error& err) {
+        CLS_LOG(1, "ERROR: rgw_bucket_list(): failed to decode entry, key=%s\n", kiter->first.c_str());
+        return -EINVAL;
+      }
+
+      cls_rgw_obj_key key;
+      uint64_t ver;
+
+      start_key = kiter->first;
+      CLS_LOG(20, "start_key=%s len=%zu", start_key.c_str(), start_key.size());
+
+      int ret = decode_list_index_key(kiter->first, &key, &ver);
+      if (ret < 0) {
+        CLS_LOG(0, "ERROR: failed to decode list index key (%s)\n", escape_str(kiter->first).c_str());
+        continue;
+      }
+
+      if (!entry.is_valid()) {
+        CLS_LOG(20, "entry %s[%s] is not valid\n", key.name.c_str(), key.instance.c_str());
+        continue;
+      }
+      
+      // filter out noncurrent versions, delete markers, and initial marker
+      if (!op.list_versions && (!entry.is_visible() || op.start_obj.name == key.name)) {
+        CLS_LOG(20, "entry %s[%s] is not visible\n", key.name.c_str(), key.instance.c_str());
+        continue;
+      }
+      if (m.size() < op.num_entries) {
+        m[kiter->first] = entry;
+      }
+      left_to_read--;
+
+      CLS_LOG(20, "got entry %s[%s] m.size()=%d\n", key.name.c_str(), key.instance.c_str(), (int)m.size());
+    }
+  } while (left_to_read > 0 && !done);
+
+  ret.is_truncated = more && !done;
+
+  encode(ret, *out);
+  return 0;
+}
+
+static int check_index(cls_method_context_t hctx,
+		       rgw_bucket_dir_header *existing_header,
+		       rgw_bucket_dir_header *calc_header)
+{
+  int rc = read_bucket_header(hctx, existing_header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: check_index(): failed to read header\n");
+    return rc;
+  }
+
+  calc_header->tag_timeout = existing_header->tag_timeout;
+  calc_header->ver = existing_header->ver;
+  calc_header->syncstopped = existing_header->syncstopped;
+
+  map<string, bufferlist> keys;
+  string start_obj;
+  string filter_prefix;
+
+#define CHECK_CHUNK_SIZE 1000
+  bool done = false;
+  bool more;
+
+  do {
+    rc = get_obj_vals(hctx, start_obj, filter_prefix, CHECK_CHUNK_SIZE, &keys, &more);
+    if (rc < 0)
+      return rc;
+
+    std::map<string, bufferlist>::iterator kiter = keys.begin();
+    for (; kiter != keys.end(); ++kiter) {
+      if (!bi_is_objs_index(kiter->first)) {
+        done = true;
+        break;
+      }
+
+      rgw_bucket_dir_entry entry;
+      auto eiter = kiter->second.cbegin();
+      try {
+        decode(entry, eiter);
+      } catch (buffer::error& err) {
+        CLS_LOG(1, "ERROR: rgw_bucket_list(): failed to decode entry, key=%s\n", kiter->first.c_str());
+        return -EIO;
+      }
+      rgw_bucket_category_stats& stats = calc_header->stats[entry.meta.category];
+      stats.num_entries++;
+      stats.total_size += entry.meta.accounted_size;
+      stats.total_size_rounded += cls_rgw_get_rounded_size(entry.meta.accounted_size);
+      stats.actual_size += entry.meta.size;
+
+      start_obj = kiter->first;
+    }
+  } while (keys.size() == CHECK_CHUNK_SIZE && !done);
+
+  return 0;
+}
+
+int rgw_bucket_check_index(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  rgw_cls_check_index_ret ret;
+
+  int rc = check_index(hctx, &ret.existing_header, &ret.calculated_header);
+  if (rc < 0)
+    return rc;
+
+  encode(ret, *out);
+
+  return 0;
+}
+
+static int write_bucket_header(cls_method_context_t hctx, rgw_bucket_dir_header *header)
+{
+  header->ver++;
+
+  bufferlist header_bl;
+  encode(*header, header_bl);
+  return cls_cxx_map_write_header(hctx, &header_bl);
+}
+
+
+int rgw_bucket_rebuild_index(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  rgw_bucket_dir_header existing_header;
+  rgw_bucket_dir_header calc_header;
+  int rc = check_index(hctx, &existing_header, &calc_header);
+  if (rc < 0)
+    return rc;
+
+  return write_bucket_header(hctx, &calc_header);
+}
+
+int rgw_bucket_update_stats(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_bucket_update_stats_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: %s(): failed to decode request\n", __func__);
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: %s(): failed to read header\n", __func__);
+    return rc;
+  }
+
+  for (auto& s : op.stats) {
+    auto& dest = header.stats[s.first];
+    if (op.absolute) {
+      dest = s.second;
+    } else {
+      dest.total_size += s.second.total_size;
+      dest.total_size_rounded += s.second.total_size_rounded;
+      dest.num_entries += s.second.num_entries;
+      dest.actual_size += s.second.actual_size;
+    }
+  }
+
+  return write_bucket_header(hctx, &header);
+}
+
+int rgw_bucket_init_index(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  bufferlist header_bl;
+  int rc = cls_cxx_map_read_header(hctx, &header_bl);
+  if (rc < 0) {
+    switch (rc) {
+    case -ENODATA:
+    case -ENOENT:
+      break;
+    default:
+      return rc;
+    }
+  }
+
+  if (header_bl.length() != 0) {
+    CLS_LOG(1, "ERROR: index already initialized\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir dir;
+
+  return write_bucket_header(hctx, &dir.header);
+}
+
+int rgw_bucket_set_tag_timeout(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_tag_timeout_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_bucket_set_tag_timeout(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_set_tag_timeout(): failed to read header\n");
+    return rc;
+  }
+
+  header.tag_timeout = op.tag_timeout;
+
+  return write_bucket_header(hctx, &header);
+}
+
+static int read_key_entry(cls_method_context_t hctx, cls_rgw_obj_key& key,
+			  string *idx, rgw_bucket_dir_entry *entry,
+                          bool special_delete_marker_name = false);
+
+int rgw_bucket_prepare_op(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_obj_prepare_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_bucket_prepare_op(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  if (op.tag.empty()) {
+    CLS_LOG(1, "ERROR: tag is empty\n");
+    return -EINVAL;
+  }
+
+  CLS_LOG(1, "rgw_bucket_prepare_op(): request: op=%d name=%s instance=%s tag=%s\n",
+          op.op, op.key.name.c_str(), op.key.instance.c_str(), op.tag.c_str());
+
+  // get on-disk state
+  string idx;
+
+  rgw_bucket_dir_entry entry;
+  int rc = read_key_entry(hctx, op.key, &idx, &entry);
+  if (rc < 0 && rc != -ENOENT)
+    return rc;
+
+  bool noent = (rc == -ENOENT);
+
+  rc = 0;
+
+  if (noent) { // no entry, initialize fields
+    entry.key = op.key;
+    entry.ver = rgw_bucket_entry_ver();
+    entry.exists = false;
+    entry.locator = op.locator;
+  }
+
+  // fill in proper state
+  rgw_bucket_pending_info info;
+  info.timestamp = real_clock::now();
+  info.state = CLS_RGW_STATE_PENDING_MODIFY;
+  info.op = op.op;
+  entry.pending_map.insert(pair<string, rgw_bucket_pending_info>(op.tag, info));
+
+  rgw_bucket_dir_header header;
+  rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_prepare_op(): failed to read header\n");
+    return rc;
+  }
+
+  if (op.log_op && !header.syncstopped) {
+    rc = log_index_operation(hctx, op.key, op.op, op.tag, entry.meta.mtime,
+                             entry.ver, info.state, header.ver, header.max_marker, op.bilog_flags, NULL, NULL, &op.zones_trace);
+    if (rc < 0)
+      return rc;
+  }
+
+  // write out new key to disk
+  bufferlist info_bl;
+  encode(entry, info_bl);
+  rc = cls_cxx_map_set_val(hctx, idx, &info_bl);
+  if (rc < 0)
+    return rc;
+
+  if (op.log_op && !header.syncstopped)
+    return write_bucket_header(hctx, &header);
+  return 0;
+}
+
+static void unaccount_entry(rgw_bucket_dir_header& header,
+			    rgw_bucket_dir_entry& entry)
+{
+  rgw_bucket_category_stats& stats = header.stats[entry.meta.category];
+  stats.num_entries--;
+  stats.total_size -= entry.meta.accounted_size;
+  stats.total_size_rounded -= cls_rgw_get_rounded_size(entry.meta.accounted_size);
+  stats.actual_size -= entry.meta.size;
+}
+
+static void log_entry(const char *func, const char *str, rgw_bucket_dir_entry *entry)
+{
+  CLS_LOG(1, "%s(): %s: ver=%ld:%llu name=%s instance=%s locator=%s\n", func, str,
+          (long)entry->ver.pool, (unsigned long long)entry->ver.epoch,
+          entry->key.name.c_str(), entry->key.instance.c_str(), entry->locator.c_str());
+}
+
+static void log_entry(const char *func, const char *str, rgw_bucket_olh_entry *entry)
+{
+  CLS_LOG(1, "%s(): %s: epoch=%llu name=%s instance=%s tag=%s\n", func, str,
+          (unsigned long long)entry->epoch, entry->key.name.c_str(), entry->key.instance.c_str(),
+          entry->tag.c_str());
+}
+
+template <class T>
+static int read_omap_entry(cls_method_context_t hctx, const std::string& name,
+                           T* entry)
+{
+  bufferlist current_entry;
+  int rc = cls_cxx_map_get_val(hctx, name, &current_entry);
+  if (rc < 0) {
+    return rc;
+  }
+
+  auto cur_iter = current_entry.cbegin();
+  try {
+    decode(*entry, cur_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: %s(): failed to decode entry\n", __func__);
+    return -EIO;
+  }
+  return 0;
+}
+
+template <class T>
+static int read_index_entry(cls_method_context_t hctx, string& name, T* entry)
+{
+  int ret = read_omap_entry(hctx, name, entry);
+  if (ret < 0) {
+    return ret;
+  }
+
+  log_entry(__func__, "existing entry", entry);
+  return 0;
+}
+
+static int read_key_entry(cls_method_context_t hctx, cls_rgw_obj_key& key,
+			  string *idx, rgw_bucket_dir_entry *entry,
+                          bool special_delete_marker_name)
+{
+  encode_obj_index_key(key, idx);
+  int rc = read_index_entry(hctx, *idx, entry);
+  if (rc < 0) {
+    return rc;
+  }
+
+  if (key.instance.empty() &&
+      entry->flags & RGW_BUCKET_DIRENT_FLAG_VER_MARKER) {
+    /* we only do it where key.instance is empty. In this case the delete marker will have a
+     * separate entry in the index to avoid collisions with the actual object, as it's mutable
+     */
+    if (special_delete_marker_name) {
+      encode_obj_versioned_data_key(key, idx, true);
+      rc = read_index_entry(hctx, *idx, entry);
+      if (rc == 0) {
+        return 0;
+      }
+    }
+    encode_obj_versioned_data_key(key, idx);
+    rc = read_index_entry(hctx, *idx, entry);
+    if (rc < 0) {
+      *entry = rgw_bucket_dir_entry(); /* need to reset entry because we initialized it earlier */
+      return rc;
+    }
+  }
+
+  return 0;
+}
+
+int rgw_bucket_complete_op(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_obj_complete_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_bucket_complete_op(): failed to decode request\n");
+    return -EINVAL;
+  }
+  CLS_LOG(1, "rgw_bucket_complete_op(): request: op=%d name=%s instance=%s ver=%lu:%llu tag=%s\n",
+          op.op, op.key.name.c_str(), op.key.instance.c_str(),
+          (unsigned long)op.ver.pool, (unsigned long long)op.ver.epoch,
+          op.tag.c_str());
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_complete_op(): failed to read header\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_entry entry;
+  bool ondisk = true;
+
+  string idx;
+  rc = read_key_entry(hctx, op.key, &idx, &entry);
+  if (rc == -ENOENT) {
+    entry.key = op.key;
+    entry.ver = op.ver;
+    entry.meta = op.meta;
+    entry.locator = op.locator;
+    ondisk = false;
+  } else if (rc < 0) {
+    return rc;
+  }
+
+  entry.index_ver = header.ver;
+  /* resetting entry flags, entry might have been previously a delete
+   * marker */
+  entry.flags &= RGW_BUCKET_DIRENT_FLAG_VER;
+
+  if (op.tag.size()) {
+    map<string, rgw_bucket_pending_info>::iterator pinter = entry.pending_map.find(op.tag);
+    if (pinter == entry.pending_map.end()) {
+      CLS_LOG(1, "ERROR: couldn't find tag for pending operation\n");
+      return -EINVAL;
+    }
+    entry.pending_map.erase(pinter);
+  }
+
+  bool cancel = false;
+  bufferlist update_bl;
+
+  if (op.tag.size() && op.op == CLS_RGW_OP_CANCEL) {
+    CLS_LOG(1, "rgw_bucket_complete_op(): cancel requested\n");
+    cancel = true;
+  } else if (op.ver.pool == entry.ver.pool &&
+             op.ver.epoch && op.ver.epoch <= entry.ver.epoch) {
+    CLS_LOG(1, "rgw_bucket_complete_op(): skipping request, old epoch\n");
+    cancel = true;
+  }
+
+  bufferlist op_bl;
+  if (cancel) {
+    if (op.log_op && !header.syncstopped) {
+      rc = log_index_operation(hctx, op.key, op.op, op.tag, entry.meta.mtime, entry.ver,
+                               CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker, op.bilog_flags, NULL, NULL, &op.zones_trace);
+      if (rc < 0)
+        return rc;
+    }
+
+    if (op.tag.size()) {
+      bufferlist new_key_bl;
+      encode(entry, new_key_bl);
+      rc = cls_cxx_map_set_val(hctx, idx, &new_key_bl);
+      if (rc < 0)
+        return rc;
+    }
+
+    if (op.log_op && !header.syncstopped) {
+      return write_bucket_header(hctx, &header);
+    }
+    return 0;
+  }
+
+  if (entry.exists) {
+    unaccount_entry(header, entry);
+  }
+
+  entry.ver = op.ver;
+  switch ((int)op.op) {
+  case CLS_RGW_OP_DEL:
+    entry.meta = op.meta;
+    if (ondisk) {
+      if (!entry.pending_map.size()) {
+	int ret = cls_cxx_map_remove_key(hctx, idx);
+	if (ret < 0)
+	  return ret;
+      } else {
+        entry.exists = false;
+        bufferlist new_key_bl;
+        encode(entry, new_key_bl);
+	int ret = cls_cxx_map_set_val(hctx, idx, &new_key_bl);
+	if (ret < 0)
+	  return ret;
+      }
+    } else {
+      return -ENOENT;
+    }
+    break;
+  case CLS_RGW_OP_ADD:
+    {
+      rgw_bucket_dir_entry_meta& meta = op.meta;
+      rgw_bucket_category_stats& stats = header.stats[meta.category];
+      entry.meta = meta;
+      entry.key = op.key;
+      entry.exists = true;
+      entry.tag = op.tag;
+      stats.num_entries++;
+      stats.total_size += meta.accounted_size;
+      stats.total_size_rounded += cls_rgw_get_rounded_size(meta.accounted_size);
+      stats.actual_size += meta.size;
+      bufferlist new_key_bl;
+      encode(entry, new_key_bl);
+      int ret = cls_cxx_map_set_val(hctx, idx, &new_key_bl);
+      if (ret < 0)
+	return ret;
+    }
+    break;
+  }
+
+  if (op.log_op && !header.syncstopped) {
+    rc = log_index_operation(hctx, op.key, op.op, op.tag, entry.meta.mtime, entry.ver,
+                             CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker, op.bilog_flags, NULL, NULL, &op.zones_trace);
+    if (rc < 0)
+      return rc;
+  }
+
+  list<cls_rgw_obj_key>::iterator remove_iter;
+  CLS_LOG(20, "rgw_bucket_complete_op(): remove_objs.size()=%d\n", (int)op.remove_objs.size());
+  for (remove_iter = op.remove_objs.begin(); remove_iter != op.remove_objs.end(); ++remove_iter) {
+    cls_rgw_obj_key& remove_key = *remove_iter;
+    CLS_LOG(1, "rgw_bucket_complete_op(): removing entries, read_index_entry name=%s instance=%s\n",
+            remove_key.name.c_str(), remove_key.instance.c_str());
+    rgw_bucket_dir_entry remove_entry;
+    string k;
+    int ret = read_key_entry(hctx, remove_key, &k, &remove_entry);
+    if (ret < 0) {
+      CLS_LOG(1, "rgw_bucket_complete_op(): removing entries, read_index_entry name=%s instance=%s ret=%d\n",
+            remove_key.name.c_str(), remove_key.instance.c_str(), ret);
+      continue;
+    }
+    CLS_LOG(0,
+	    "rgw_bucket_complete_op(): entry.name=%s entry.instance=%s entry.meta.category=%d\n",
+            remove_entry.key.name.c_str(),
+	    remove_entry.key.instance.c_str(),
+	    int(remove_entry.meta.category));
+    unaccount_entry(header, remove_entry);
+
+    if (op.log_op && !header.syncstopped) {
+      ++header.ver; // increment index version, or we'll overwrite keys previously written
+      rc = log_index_operation(hctx, remove_key, CLS_RGW_OP_DEL, op.tag, remove_entry.meta.mtime,
+                               remove_entry.ver, CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker, op.bilog_flags, NULL, NULL, &op.zones_trace);
+      if (rc < 0)
+        continue;
+    }
+
+    ret = cls_cxx_map_remove_key(hctx, k);
+    if (ret < 0) {
+      CLS_LOG(1, "rgw_bucket_complete_op(): cls_cxx_map_remove_key, failed to remove entry, name=%s instance=%s read_index_entry ret=%d\n", remove_key.name.c_str(), remove_key.instance.c_str(), rc);
+      continue;
+    }
+  }
+
+  return write_bucket_header(hctx, &header);
+}
+
+template <class T>
+static int write_entry(cls_method_context_t hctx, T& entry, const string& key)
+{
+  bufferlist bl;
+  encode(entry, bl);
+  return cls_cxx_map_set_val(hctx, key, &bl);
+}
+
+static int read_olh(cls_method_context_t hctx,cls_rgw_obj_key& obj_key, rgw_bucket_olh_entry *olh_data_entry, string *index_key, bool *found)
+{
+  cls_rgw_obj_key olh_key;
+  olh_key.name = obj_key.name;
+
+  encode_olh_data_key(olh_key, index_key);
+  int ret = read_index_entry(hctx, *index_key, olh_data_entry);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: read_index_entry() olh_key=%s ret=%d", olh_key.name.c_str(), ret);
+    return ret;
+  }
+  if (found) {
+    *found = (ret != -ENOENT);
+  }
+  return 0;
+}
+
+static void update_olh_log(rgw_bucket_olh_entry& olh_data_entry, OLHLogOp op, const string& op_tag,
+                           cls_rgw_obj_key& key, bool delete_marker, uint64_t epoch)
+{
+  vector<rgw_bucket_olh_log_entry>& log = olh_data_entry.pending_log[olh_data_entry.epoch];
+  rgw_bucket_olh_log_entry log_entry;
+  log_entry.epoch = epoch;
+  log_entry.op = op;
+  log_entry.op_tag = op_tag;
+  log_entry.key = key;
+  log_entry.delete_marker = delete_marker;
+  log.push_back(log_entry);
+}
+
+static int write_obj_instance_entry(cls_method_context_t hctx, rgw_bucket_dir_entry& instance_entry, const string& instance_idx)
+{
+  CLS_LOG(20, "write_entry() instance=%s idx=%s flags=%d", escape_str(instance_entry.key.instance).c_str(), instance_idx.c_str(), instance_entry.flags);
+  /* write the instance entry */
+  int ret = write_entry(hctx, instance_entry, instance_idx);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: write_entry() instance_key=%s ret=%d", escape_str(instance_idx).c_str(), ret);
+    return ret;
+  }
+  return 0;
+}
+
+/*
+ * write object instance entry, and if needed also the list entry
+ */
+static int write_obj_entries(cls_method_context_t hctx, rgw_bucket_dir_entry& instance_entry, const string& instance_idx)
+{
+  int ret = write_obj_instance_entry(hctx, instance_entry, instance_idx);
+  if (ret < 0) {
+    return ret;
+  }
+  string instance_list_idx;
+  get_list_index_key(instance_entry, &instance_list_idx);
+
+  if (instance_idx != instance_list_idx) {
+    CLS_LOG(20, "write_entry() idx=%s flags=%d", escape_str(instance_list_idx).c_str(), instance_entry.flags);
+    /* write a new list entry for the object instance */
+    ret = write_entry(hctx, instance_entry, instance_list_idx);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: write_entry() instance=%s instance_list_idx=%s ret=%d", instance_entry.key.instance.c_str(), instance_list_idx.c_str(), ret);
+      return ret;
+    }
+  }
+  return 0;
+}
+
+
+class BIVerObjEntry {
+  cls_method_context_t hctx;
+  cls_rgw_obj_key key;
+  string instance_idx;
+
+  rgw_bucket_dir_entry instance_entry;
+
+  bool initialized;
+
+public:
+  BIVerObjEntry(cls_method_context_t& _hctx, const cls_rgw_obj_key& _key) : hctx(_hctx), key(_key), initialized(false) {
+  }
+
+  int init(bool check_delete_marker = true) {
+    int ret = read_key_entry(hctx, key, &instance_idx, &instance_entry,
+                             check_delete_marker && key.instance.empty()); /* this is potentially a delete marker, for null objects we
+                                                                              keep separate instance entry for the delete markers */
+
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: read_key_entry() idx=%s ret=%d", instance_idx.c_str(), ret);
+      return ret;
+    }
+    initialized = true;
+    CLS_LOG(20, "read instance_entry key.name=%s key.instance=%s flags=%d", instance_entry.key.name.c_str(), instance_entry.key.instance.c_str(), instance_entry.flags);
+    return 0;
+  }
+
+  rgw_bucket_dir_entry& get_dir_entry() {
+    return instance_entry;
+  }
+
+  void init_as_delete_marker(rgw_bucket_dir_entry_meta& meta) {
+    /* a deletion marker, need to initialize it, there's no instance entry for it yet */
+    instance_entry.key = key;
+    instance_entry.flags = RGW_BUCKET_DIRENT_FLAG_DELETE_MARKER;
+    instance_entry.meta = meta;
+    instance_entry.tag = "delete-marker";
+
+    initialized = true;
+  }
+
+  void set_epoch(uint64_t epoch) {
+    instance_entry.versioned_epoch = epoch;
+  }
+
+  int unlink_list_entry() {
+    string list_idx;
+    /* this instance has a previous list entry, remove that entry */
+    get_list_index_key(instance_entry, &list_idx);
+    CLS_LOG(20, "unlink_list_entry() list_idx=%s", escape_str(list_idx).c_str());
+    int ret = cls_cxx_map_remove_key(hctx, list_idx);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: cls_cxx_map_remove_key() list_idx=%s ret=%d", list_idx.c_str(), ret);
+      return ret;
+    }
+    return 0;
+  }
+
+  int unlink() {
+    /* remove the instance entry */
+    CLS_LOG(20, "unlink() idx=%s", escape_str(instance_idx).c_str());
+    int ret = cls_cxx_map_remove_key(hctx, instance_idx);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: cls_cxx_map_remove_key() instance_idx=%s ret=%d", instance_idx.c_str(), ret);
+      return ret;
+    }
+    return 0;
+  }
+
+  int write_entries(uint64_t flags_set, uint64_t flags_reset) {
+    if (!initialized) {
+      int ret = init();
+      if (ret < 0) {
+        return ret;
+      }
+    }
+    instance_entry.flags &= ~flags_reset;
+    instance_entry.flags |= flags_set;
+
+    /* write the instance and list entries */
+    bool special_delete_marker_key = (instance_entry.is_delete_marker() && instance_entry.key.instance.empty());
+    encode_obj_versioned_data_key(key, &instance_idx, special_delete_marker_key);
+    int ret = write_obj_entries(hctx, instance_entry, instance_idx);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: write_obj_entries() instance_idx=%s ret=%d", instance_idx.c_str(), ret);
+      return ret;
+    }
+
+    return 0;
+  }
+
+  int write(uint64_t epoch, bool current) {
+    if (instance_entry.versioned_epoch > 0) {
+      CLS_LOG(20, "%s(): instance_entry.versioned_epoch=%d epoch=%d", __func__, (int)instance_entry.versioned_epoch, (int)epoch);
+      /* this instance has a previous list entry, remove that entry */
+      int ret = unlink_list_entry();
+      if (ret < 0) {
+        return ret;
+      }
+    }
+
+    uint64_t flags = RGW_BUCKET_DIRENT_FLAG_VER;
+    if (current) {
+      flags |= RGW_BUCKET_DIRENT_FLAG_CURRENT;
+    }
+
+    instance_entry.versioned_epoch = epoch;
+    return write_entries(flags, 0);
+  }
+
+  int demote_current() {
+    return write_entries(0, RGW_BUCKET_DIRENT_FLAG_CURRENT);
+  }
+
+  bool is_delete_marker() {
+    return instance_entry.is_delete_marker();
+  }
+
+  int find_next_key(cls_rgw_obj_key *next_key, bool *found) {
+    string list_idx;
+    /* this instance has a previous list entry, remove that entry */
+    get_list_index_key(instance_entry, &list_idx);
+    /* this is the current head, need to update! */
+    map<string, bufferlist> keys;
+    bool more;
+    string filter = key.name; /* list key starts with key name, filter it to avoid a case where we cross to
+                                 different namespace */
+    int ret = cls_cxx_map_get_vals(hctx, list_idx, filter, 1, &keys, &more);
+    if (ret < 0) {
+      return ret;
+    }
+
+    if (keys.size() < 1) {
+      *found = false;
+      return 0;
+    }
+
+    rgw_bucket_dir_entry next_entry;
+
+    map<string, bufferlist>::reverse_iterator last = keys.rbegin();
+    try {
+      auto iter = last->second.cbegin();
+      decode(next_entry, iter);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR; failed to decode entry: %s", last->first.c_str());
+      return -EIO;
+    }
+
+    *found = (key.name == next_entry.key.name);
+    if (*found) {
+      *next_key = next_entry.key;
+    }
+
+    return 0;
+  }
+
+  real_time mtime() {
+    return instance_entry.meta.mtime;
+  }
+};
+
+
+class BIOLHEntry {
+  cls_method_context_t hctx;
+  cls_rgw_obj_key key;
+
+  string olh_data_idx;
+  rgw_bucket_olh_entry olh_data_entry;
+
+  bool initialized;
+public:
+  BIOLHEntry(cls_method_context_t& _hctx, const cls_rgw_obj_key& _key) : hctx(_hctx), key(_key), initialized(false) { }
+
+  int init(bool *exists) {
+    /* read olh */
+    int ret = read_olh(hctx, key, &olh_data_entry, &olh_data_idx, exists);
+    if (ret < 0) {
+      return ret;
+    }
+
+    initialized = true;
+    return 0;
+  }
+
+  bool start_modify(uint64_t candidate_epoch) {
+    if (candidate_epoch) {
+      if (candidate_epoch < olh_data_entry.epoch) {
+        return false; /* olh cannot be modified, old epoch */
+      }
+      olh_data_entry.epoch = candidate_epoch;
+    } else {
+      if (olh_data_entry.epoch == 0) {
+        olh_data_entry.epoch = 2; /* versioned epoch should start with 2, 1 is reserved to converted plain entries */
+      } else {
+        olh_data_entry.epoch++;
+      }
+    }
+    return true;
+  }
+
+  uint64_t get_epoch() {
+    return olh_data_entry.epoch;
+  }
+
+  rgw_bucket_olh_entry& get_entry() {
+    return olh_data_entry;
+  }
+
+  void update(cls_rgw_obj_key& key, bool delete_marker) {
+    olh_data_entry.delete_marker = delete_marker;
+    olh_data_entry.key = key;
+  }
+
+  int write() {
+    /* write the olh data entry */
+    int ret = write_entry(hctx, olh_data_entry, olh_data_idx);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: write_entry() olh_key=%s ret=%d", olh_data_idx.c_str(), ret);
+      return ret;
+    }
+
+    return 0;
+  }
+
+  void update_log(OLHLogOp op, const string& op_tag, cls_rgw_obj_key& key, bool delete_marker, uint64_t epoch = 0) {
+    if (epoch == 0) {
+      epoch = olh_data_entry.epoch;
+    }
+    update_olh_log(olh_data_entry, op, op_tag, key, delete_marker, epoch);
+  }
+
+  bool exists() { return olh_data_entry.exists; }
+
+  void set_exists(bool exists) {
+    olh_data_entry.exists = exists;
+  }
+
+  bool pending_removal() { return olh_data_entry.pending_removal; }
+
+  void set_pending_removal(bool pending_removal) {
+    olh_data_entry.pending_removal = pending_removal;
+  }
+
+  const string& get_tag() { return olh_data_entry.tag; }
+  void set_tag(const string& tag) {
+    olh_data_entry.tag = tag;
+  }
+};
+
+static int write_version_marker(cls_method_context_t hctx, cls_rgw_obj_key& key)
+{
+  rgw_bucket_dir_entry entry;
+  entry.key = key;
+  entry.flags = RGW_BUCKET_DIRENT_FLAG_VER_MARKER;
+  int ret = write_entry(hctx, entry, key.name);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: write_entry returned ret=%d", ret);
+    return ret;
+  }
+  return 0;
+}
+
+/*
+ * plain entries are the ones who were created when bucket was not versioned,
+ * if we override these objects, we need to convert these to versioned entries -- ones that have
+ * both data entry, and listing key. Their version is going to be empty though
+ */
+static int convert_plain_entry_to_versioned(cls_method_context_t hctx, cls_rgw_obj_key& key, bool demote_current, bool instance_only)
+{
+  if (!key.instance.empty()) {
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_entry entry;
+
+  string orig_idx;
+  int ret = read_key_entry(hctx, key, &orig_idx, &entry);
+  if (ret != -ENOENT) {
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: read_key_entry() returned ret=%d", ret);
+      return ret;
+    }
+
+    entry.versioned_epoch = 1; /* converted entries are always 1 */
+    entry.flags |= RGW_BUCKET_DIRENT_FLAG_VER;
+
+    if (demote_current) {
+      entry.flags &= ~RGW_BUCKET_DIRENT_FLAG_CURRENT;
+    }
+
+    string new_idx;
+    encode_obj_versioned_data_key(key, &new_idx);
+
+    if (instance_only) {
+      ret = write_obj_instance_entry(hctx, entry, new_idx);
+    } else {
+      ret = write_obj_entries(hctx, entry, new_idx);
+    }
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: write_obj_entries new_idx=%s returned %d", new_idx.c_str(), ret);
+      return ret;
+    }
+  }
+
+  ret = write_version_marker(hctx, key);
+  if (ret < 0) {
+    return ret;
+  }
+
+  return 0;
+}
+
+/*
+ * link an object version to an olh, update the relevant index entries. It will also handle the
+ * deletion marker case. We have a few entries that we need to take care of. For object 'foo',
+ * instance BAR, we'd update the following (not actual encoding):
+ *  - olh data: [BI_BUCKET_OLH_DATA_INDEX]foo
+ *  - object instance data: [BI_BUCKET_OBJ_INSTANCE_INDEX]foo,BAR
+ *  - object instance list entry: foo,123,BAR
+ *
+ *  The instance list entry needs to be ordered by newer to older, so we generate an appropriate
+ *  number string that follows the name.
+ *  The top instance for each object is marked appropriately.
+ *  We generate instance entry for deletion markers here, as they are not created prior.
+ */
+static int rgw_bucket_link_olh(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string olh_data_idx;
+  string instance_idx;
+
+  // decode request
+  rgw_cls_link_olh_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: rgw_bucket_link_olh_op(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  BIVerObjEntry obj(hctx, op.key);
+  BIOLHEntry olh(hctx, op.key);
+
+  /* read instance entry */
+  int ret = obj.init(op.delete_marker);
+  bool existed = (ret == 0);
+  if (ret == -ENOENT && op.delete_marker) {
+    ret = 0;
+  }
+  if (ret < 0) {
+    return ret;
+  }
+
+  if (existed && !real_clock::is_zero(op.unmod_since)) {
+    timespec mtime = ceph::real_clock::to_timespec(obj.mtime());
+    timespec unmod = ceph::real_clock::to_timespec(op.unmod_since);
+    if (!op.high_precision_time) {
+      mtime.tv_nsec = 0;
+      unmod.tv_nsec = 0;
+    }
+    if (mtime >= unmod) {
+      return 0; /* no need to set error, we just return 0 and avoid writing to the bi log */
+    }
+  }
+
+  bool removing;
+
+  /*
+   * Special handling for null instance object / delete-marker. For these objects we're going to
+   * have separate instances for a data object vs. delete-marker to avoid collisions. We now check
+   * if we got to overwrite a previous entry, and in that case we'll remove its list entry.
+   */
+  if (op.key.instance.empty()) {
+    BIVerObjEntry other_obj(hctx, op.key);
+    ret = other_obj.init(!op.delete_marker); /* try reading the other null versioned entry */
+    existed = (ret >= 0 && !other_obj.is_delete_marker());
+    if (ret >= 0 && other_obj.is_delete_marker() != op.delete_marker) {
+      ret = other_obj.unlink_list_entry();
+      if (ret < 0) {
+        return ret;
+      }
+    }
+
+    removing = existed && op.delete_marker;
+    if (!removing) {
+      ret = other_obj.unlink();
+      if (ret < 0) {
+        return ret;
+      }
+    }
+  } else {
+    removing = (existed && !obj.is_delete_marker() && op.delete_marker);
+  }
+
+  if (op.delete_marker) {
+    /* a deletion marker, need to initialize entry as such */
+    obj.init_as_delete_marker(op.meta);
+  }
+
+  /* read olh */
+  bool olh_found;
+  ret = olh.init(&olh_found);
+  if (ret < 0) {
+    return ret;
+  }
+  const uint64_t prev_epoch = olh.get_epoch();
+
+  if (!olh.start_modify(op.olh_epoch)) {
+    ret = obj.write(op.olh_epoch, false);
+    if (ret < 0) {
+      return ret;
+    }
+    if (removing) {
+      olh.update_log(CLS_RGW_OLH_OP_REMOVE_INSTANCE, op.op_tag, op.key, false, op.olh_epoch);
+    }
+    return 0;
+  }
+
+  // promote this version to current if it's a newer epoch, or if it matches the
+  // current epoch and sorts after the current instance
+  const bool promote = (olh.get_epoch() > prev_epoch) ||
+      (olh.get_epoch() == prev_epoch &&
+       olh.get_entry().key.instance >= op.key.instance);
+
+  if (olh_found) {
+    const string& olh_tag = olh.get_tag();
+    if (op.olh_tag != olh_tag) {
+      if (!olh.pending_removal()) {
+        CLS_LOG(5, "NOTICE: op.olh_tag (%s) != olh.tag (%s)", op.olh_tag.c_str(), olh_tag.c_str());
+        return -ECANCELED;
+      }
+      /* if pending removal, this is a new olh instance */
+      olh.set_tag(op.olh_tag);
+    }
+    if (promote && olh.exists()) {
+      rgw_bucket_olh_entry& olh_entry = olh.get_entry();
+      /* found olh, previous instance is no longer the latest, need to update */
+      if (!(olh_entry.key == op.key)) {
+        BIVerObjEntry old_obj(hctx, olh_entry.key);
+
+        ret = old_obj.demote_current();
+        if (ret < 0) {
+          CLS_LOG(0, "ERROR: could not demote current on previous key ret=%d", ret);
+          return ret;
+        }
+      }
+    }
+    olh.set_pending_removal(false);
+  } else {
+    bool instance_only = (op.key.instance.empty() && op.delete_marker);
+    cls_rgw_obj_key key(op.key.name);
+    ret = convert_plain_entry_to_versioned(hctx, key, promote, instance_only);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: convert_plain_entry_to_versioned ret=%d", ret);
+      return ret;
+    }
+    olh.set_tag(op.olh_tag);
+  }
+
+  /* update the olh log */
+  olh.update_log(CLS_RGW_OLH_OP_LINK_OLH, op.op_tag, op.key, op.delete_marker);
+  if (removing) {
+    olh.update_log(CLS_RGW_OLH_OP_REMOVE_INSTANCE, op.op_tag, op.key, false);
+  }
+
+  if (promote) {
+    olh.update(op.key, op.delete_marker);
+  }
+  olh.set_exists(true);
+
+  ret = olh.write();
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: failed to update olh ret=%d", ret);
+    return ret;
+  }
+
+  /* write the instance and list entries */
+  ret = obj.write(olh.get_epoch(), promote);
+  if (ret < 0) {
+    return ret;
+  }
+
+  rgw_bucket_dir_header header;
+  ret = read_bucket_header(hctx, &header);
+  if (ret < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_link_olh(): failed to read header\n");
+    return ret;
+  }
+
+  if (op.log_op && !header.syncstopped) {
+    rgw_bucket_dir_entry& entry = obj.get_dir_entry();
+
+    rgw_bucket_entry_ver ver;
+    ver.epoch = (op.olh_epoch ? op.olh_epoch : olh.get_epoch());
+
+    string *powner = NULL;
+    string *powner_display_name = NULL;
+
+    if (op.delete_marker) {
+      powner = &entry.meta.owner;
+      powner_display_name = &entry.meta.owner_display_name;
+    }
+
+    RGWModifyOp operation = (op.delete_marker ? CLS_RGW_OP_LINK_OLH_DM : CLS_RGW_OP_LINK_OLH);
+    ret = log_index_operation(hctx, op.key, operation, op.op_tag,
+                              entry.meta.mtime, ver,
+                              CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker, op.bilog_flags | RGW_BILOG_FLAG_VERSIONED_OP,
+                              powner, powner_display_name, &op.zones_trace);
+    if (ret < 0)
+      return ret;
+
+    return write_bucket_header(hctx, &header); /* updates header version */
+  }
+
+  return 0;
+}
+
+static int rgw_bucket_unlink_instance(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  string olh_data_idx;
+  string instance_idx;
+
+  // decode request
+  rgw_cls_unlink_instance_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: rgw_bucket_rm_obj_instance_op(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  cls_rgw_obj_key dest_key = op.key;
+  if (dest_key.instance == "null") {
+    dest_key.instance.clear();
+  }
+
+  BIVerObjEntry obj(hctx, dest_key);
+  BIOLHEntry olh(hctx, dest_key);
+
+  int ret = obj.init();
+  if (ret == -ENOENT) {
+    return 0; /* already removed */
+  }
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: obj.init() returned ret=%d", ret);
+    return ret;
+  }
+
+  bool olh_found;
+  ret = olh.init(&olh_found);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: olh.init() returned ret=%d", ret);
+    return ret;
+  }
+
+  if (!olh_found) {
+    bool instance_only = false;
+    cls_rgw_obj_key key(dest_key.name);
+    ret = convert_plain_entry_to_versioned(hctx, key, true, instance_only);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: convert_plain_entry_to_versioned ret=%d", ret);
+      return ret;
+    }
+    olh.update(dest_key, false);
+    olh.set_tag(op.olh_tag);
+
+    obj.set_epoch(1);
+  }
+
+  if (!olh.start_modify(op.olh_epoch)) {
+    ret = obj.unlink_list_entry();
+    if (ret < 0) {
+      return ret;
+    }
+
+    if (!obj.is_delete_marker()) {
+      olh.update_log(CLS_RGW_OLH_OP_REMOVE_INSTANCE, op.op_tag, op.key, false, op.olh_epoch);
+    }
+
+    return 0;
+  }
+
+  rgw_bucket_olh_entry& olh_entry = olh.get_entry();
+  cls_rgw_obj_key& olh_key = olh_entry.key;
+  CLS_LOG(20, "%s(): updating olh log: existing olh entry: %s[%s] (delete_marker=%d)", __func__,
+             olh_key.name.c_str(), olh_key.instance.c_str(), olh_entry.delete_marker);
+
+  if (olh_key == dest_key) {
+    /* this is the current head, need to update! */
+    cls_rgw_obj_key next_key;
+    bool found = false;
+    ret = obj.find_next_key(&next_key, &found);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: obj.find_next_key() returned ret=%d", ret);
+      return ret;
+    }
+
+    if (found) {
+      BIVerObjEntry next(hctx, next_key);
+      ret = next.write(olh.get_epoch(), true);
+      if (ret < 0) {
+        CLS_LOG(0, "ERROR: next.write() returned ret=%d", ret);
+        return ret;
+      }
+
+      CLS_LOG(20, "%s(): updating olh log: link olh -> %s[%s] (is_delete=%d)", __func__,
+              next_key.name.c_str(), next_key.instance.c_str(), (int)next.is_delete_marker());
+
+      olh.update(next_key, next.is_delete_marker());
+      olh.update_log(CLS_RGW_OLH_OP_LINK_OLH, op.op_tag, next_key, next.is_delete_marker());
+    } else {
+      // next_key is empty, but we need to preserve its name in case this entry
+      // gets resharded, because this key is used for hash placement
+      next_key.name = dest_key.name;
+      olh.update(next_key, false);
+      olh.update_log(CLS_RGW_OLH_OP_UNLINK_OLH, op.op_tag, next_key, false);
+      olh.set_exists(false);
+      olh.set_pending_removal(true);
+    }
+  }
+
+  if (!obj.is_delete_marker()) {
+    olh.update_log(CLS_RGW_OLH_OP_REMOVE_INSTANCE, op.op_tag, op.key, false);
+  } else {
+    /* this is a delete marker, it's our responsibility to remove its instance entry */
+    ret = obj.unlink();
+    if (ret < 0) {
+      return ret;
+    }
+  }
+
+  ret = obj.unlink_list_entry();
+  if (ret < 0) {
+    return ret;
+  }
+
+  ret = olh.write();
+  if (ret < 0) {
+    return ret;
+  }
+
+  rgw_bucket_dir_header header;
+  ret = read_bucket_header(hctx, &header);
+  if (ret < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_unlink_instance(): failed to read header\n");
+    return ret;
+  }
+
+  if (op.log_op && !header.syncstopped) {
+    rgw_bucket_entry_ver ver;
+    ver.epoch = (op.olh_epoch ? op.olh_epoch : olh.get_epoch());
+
+    real_time mtime = obj.mtime(); /* mtime has no real meaning in instance removal context */
+    ret = log_index_operation(hctx, op.key, CLS_RGW_OP_UNLINK_INSTANCE, op.op_tag,
+                              mtime, ver,
+                              CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker,
+                              op.bilog_flags | RGW_BILOG_FLAG_VERSIONED_OP, NULL, NULL, &op.zones_trace);
+    if (ret < 0)
+      return ret;
+
+    return write_bucket_header(hctx, &header); /* updates header version */
+  }
+
+  return 0;
+}
+
+static int rgw_bucket_read_olh_log(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_read_olh_log_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: rgw_bucket_read_olh_log(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  if (!op.olh.instance.empty()) {
+    CLS_LOG(1, "bad key passed in (non empty instance)");
+    return -EINVAL;
+  }
+
+  rgw_bucket_olh_entry olh_data_entry;
+  string olh_data_key;
+  encode_olh_data_key(op.olh, &olh_data_key);
+  int ret = read_index_entry(hctx, olh_data_key, &olh_data_entry);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: read_index_entry() olh_key=%s ret=%d", olh_data_key.c_str(), ret);
+    return ret;
+  }
+
+  if (olh_data_entry.tag != op.olh_tag) {
+    CLS_LOG(1, "NOTICE: %s(): olh_tag_mismatch olh_data_entry.tag=%s op.olh_tag=%s", __func__, olh_data_entry.tag.c_str(), op.olh_tag.c_str());
+    return -ECANCELED;
+  }
+
+  rgw_cls_read_olh_log_ret op_ret;
+
+#define MAX_OLH_LOG_ENTRIES 1000
+  map<uint64_t, vector<rgw_bucket_olh_log_entry> >& log = olh_data_entry.pending_log;
+
+  if (log.begin()->first > op.ver_marker && log.size() <= MAX_OLH_LOG_ENTRIES) {
+    op_ret.log = log;
+    op_ret.is_truncated = false;
+  } else {
+    map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator iter = log.upper_bound(op.ver_marker);
+
+    for (int i = 0; i < MAX_OLH_LOG_ENTRIES && iter != log.end(); ++i, ++iter) {
+      op_ret.log[iter->first] = iter->second;
+    }
+    op_ret.is_truncated = (iter != log.end());
+  }
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+static int rgw_bucket_trim_olh_log(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_trim_olh_log_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: rgw_bucket_trim_olh_log(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  if (!op.olh.instance.empty()) {
+    CLS_LOG(1, "bad key passed in (non empty instance)");
+    return -EINVAL;
+  }
+
+  /* read olh entry */
+  rgw_bucket_olh_entry olh_data_entry;
+  string olh_data_key;
+  encode_olh_data_key(op.olh, &olh_data_key);
+  int ret = read_index_entry(hctx, olh_data_key, &olh_data_entry);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: read_index_entry() olh_key=%s ret=%d", olh_data_key.c_str(), ret);
+    return ret;
+  }
+
+  if (olh_data_entry.tag != op.olh_tag) {
+    CLS_LOG(1, "NOTICE: %s(): olh_tag_mismatch olh_data_entry.tag=%s op.olh_tag=%s", __func__, olh_data_entry.tag.c_str(), op.olh_tag.c_str());
+    return -ECANCELED;
+  }
+
+  /* remove all versions up to and including ver from the pending map */
+  map<uint64_t, vector<rgw_bucket_olh_log_entry> >& log = olh_data_entry.pending_log;
+  map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator liter = log.begin();
+  while (liter != log.end() && liter->first <= op.ver) {
+    map<uint64_t, vector<rgw_bucket_olh_log_entry> >::iterator rm_iter = liter;
+    ++liter;
+    log.erase(rm_iter);
+  }
+
+  /* write the olh data entry */
+  ret = write_entry(hctx, olh_data_entry, olh_data_key);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: write_entry() olh_key=%s ret=%d", olh_data_key.c_str(), ret);
+    return ret;
+  }
+
+  return 0;
+}
+
+static int rgw_bucket_clear_olh(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_bucket_clear_olh_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: rgw_bucket_clear_olh(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  if (!op.key.instance.empty()) {
+    CLS_LOG(1, "bad key passed in (non empty instance)");
+    return -EINVAL;
+  }
+
+  /* read olh entry */
+  rgw_bucket_olh_entry olh_data_entry;
+  string olh_data_key;
+  encode_olh_data_key(op.key, &olh_data_key);
+  int ret = read_index_entry(hctx, olh_data_key, &olh_data_entry);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: read_index_entry() olh_key=%s ret=%d", olh_data_key.c_str(), ret);
+    return ret;
+  }
+
+  if (olh_data_entry.tag != op.olh_tag) {
+    CLS_LOG(1, "NOTICE: %s(): olh_tag_mismatch olh_data_entry.tag=%s op.olh_tag=%s", __func__, olh_data_entry.tag.c_str(), op.olh_tag.c_str());
+    return -ECANCELED;
+  }
+
+  ret = cls_cxx_map_remove_key(hctx, olh_data_key);
+  if (ret < 0) {
+    CLS_LOG(1, "NOTICE: %s(): can't remove key %s ret=%d", __func__, olh_data_key.c_str(), ret);
+    return ret;
+  }
+
+  rgw_bucket_dir_entry plain_entry;
+
+  /* read plain entry, make sure it's a versioned place holder */
+  ret = read_index_entry(hctx, op.key.name, &plain_entry);
+  if (ret == -ENOENT) {
+    /* we're done, no entry existing */
+    return 0;
+  }
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: read_index_entry key=%s ret=%d", op.key.name.c_str(), ret);
+    return ret;
+  }
+
+  if ((plain_entry.flags & RGW_BUCKET_DIRENT_FLAG_VER_MARKER) == 0) {
+    /* it's not a version marker, don't remove it */
+    return 0;
+  }
+
+  ret = cls_cxx_map_remove_key(hctx, op.key.name);
+  if (ret < 0) {
+    CLS_LOG(1, "NOTICE: %s(): can't remove key %s ret=%d", __func__, op.key.name.c_str(), ret);
+    return ret;
+  }
+
+  return 0;
+}
+
+int rgw_dir_suggest_changes(cls_method_context_t hctx,
+			    bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(1, "rgw_dir_suggest_changes()");
+
+  bufferlist header_bl;
+  rgw_bucket_dir_header header;
+  bool header_changed = false;
+
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_dir_suggest_changes(): failed to read header\n");
+    return rc;
+  }
+
+  timespan tag_timeout(
+    std::chrono::seconds(
+      header.tag_timeout ? header.tag_timeout : CEPH_RGW_TAG_TIMEOUT));
+
+  auto in_iter = in->cbegin();
+
+  while (!in_iter.end()) {
+    __u8 op;
+    rgw_bucket_dir_entry cur_change;
+    rgw_bucket_dir_entry cur_disk;
+    try {
+      decode(op, in_iter);
+      decode(cur_change, in_iter);
+    } catch (buffer::error& err) {
+      CLS_LOG(1, "ERROR: rgw_dir_suggest_changes(): failed to decode request\n");
+      return -EINVAL;
+    }
+
+    bufferlist cur_disk_bl;
+    string cur_change_key;
+    encode_obj_index_key(cur_change.key, &cur_change_key);
+    int ret = cls_cxx_map_get_val(hctx, cur_change_key, &cur_disk_bl);
+    if (ret < 0 && ret != -ENOENT)
+      return -EINVAL;
+
+    if (ret == -ENOENT) {
+      continue;
+    }
+
+    if (cur_disk_bl.length()) {
+      auto cur_disk_iter = cur_disk_bl.cbegin();
+      try {
+        decode(cur_disk, cur_disk_iter);
+      } catch (buffer::error& error) {
+        CLS_LOG(1, "ERROR: rgw_dir_suggest_changes(): failed to decode cur_disk\n");
+        return -EINVAL;
+      }
+
+      real_time cur_time = real_clock::now();
+      map<string, rgw_bucket_pending_info>::iterator iter =
+                cur_disk.pending_map.begin();
+      while(iter != cur_disk.pending_map.end()) {
+        map<string, rgw_bucket_pending_info>::iterator cur_iter=iter++;
+        if (cur_time > (cur_iter->second.timestamp + timespan(tag_timeout))) {
+          cur_disk.pending_map.erase(cur_iter);
+        }
+      }
+    }
+
+    CLS_LOG(20, "cur_disk.pending_map.empty()=%d op=%d cur_disk.exists=%d cur_change.pending_map.size()=%d cur_change.exists=%d\n",
+	    cur_disk.pending_map.empty(), (int)op, cur_disk.exists,
+	    (int)cur_change.pending_map.size(), cur_change.exists);
+
+    if (cur_disk.pending_map.empty()) {
+      if (cur_disk.exists) {
+        rgw_bucket_category_stats& old_stats = header.stats[cur_disk.meta.category];
+        CLS_LOG(10, "total_entries: %" PRId64 " -> %" PRId64 "\n", old_stats.num_entries, old_stats.num_entries - 1);
+        old_stats.num_entries--;
+        old_stats.total_size -= cur_disk.meta.accounted_size;
+        old_stats.total_size_rounded -= cls_rgw_get_rounded_size(cur_disk.meta.accounted_size);
+        old_stats.actual_size -= cur_disk.meta.size;
+        header_changed = true;
+      }
+      rgw_bucket_category_stats& stats = header.stats[cur_change.meta.category];
+      bool log_op = (op & CEPH_RGW_DIR_SUGGEST_LOG_OP) != 0;
+      op &= CEPH_RGW_DIR_SUGGEST_OP_MASK;
+      switch(op) {
+      case CEPH_RGW_REMOVE:
+        CLS_LOG(10, "CEPH_RGW_REMOVE name=%s instance=%s\n", cur_change.key.name.c_str(), cur_change.key.instance.c_str());
+	ret = cls_cxx_map_remove_key(hctx, cur_change_key);
+	if (ret < 0)
+	  return ret;
+        if (log_op && cur_disk.exists && !header.syncstopped) {
+          ret = log_index_operation(hctx, cur_disk.key, CLS_RGW_OP_DEL, cur_disk.tag, cur_disk.meta.mtime,
+                                    cur_disk.ver, CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker, 0, NULL, NULL, NULL);
+          if (ret < 0) {
+            CLS_LOG(0, "ERROR: %s(): failed to log operation ret=%d", __func__, ret);
+            return ret;
+          }
+        }
+        break;
+      case CEPH_RGW_UPDATE:
+        CLS_LOG(10, "CEPH_RGW_UPDATE name=%s instance=%s total_entries: %" PRId64 " -> %" PRId64 "\n",
+                cur_change.key.name.c_str(), cur_change.key.instance.c_str(), stats.num_entries, stats.num_entries + 1);
+
+        stats.num_entries++;
+        stats.total_size += cur_change.meta.accounted_size;
+        stats.total_size_rounded += cls_rgw_get_rounded_size(cur_change.meta.accounted_size);
+        stats.actual_size += cur_change.meta.size;
+        header_changed = true;
+        cur_change.index_ver = header.ver;
+        bufferlist cur_state_bl;
+        encode(cur_change, cur_state_bl);
+        ret = cls_cxx_map_set_val(hctx, cur_change_key, &cur_state_bl);
+        if (ret < 0)
+	  return ret;
+        if (log_op && !header.syncstopped) {
+          ret = log_index_operation(hctx, cur_change.key, CLS_RGW_OP_ADD, cur_change.tag, cur_change.meta.mtime,
+                                    cur_change.ver, CLS_RGW_STATE_COMPLETE, header.ver, header.max_marker, 0, NULL, NULL, NULL);
+          if (ret < 0) {
+            CLS_LOG(0, "ERROR: %s(): failed to log operation ret=%d", __func__, ret);
+            return ret;
+          }
+        }
+        break;
+      } // switch(op)
+    } // if (cur_disk.pending_map.empty())
+  } // while (!in_iter.end())
+
+  if (header_changed) {
+    return write_bucket_header(hctx, &header);
+  }
+  return 0;
+}
+
+static int rgw_obj_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_obj_remove_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  if (op.keep_attr_prefixes.empty()) {
+    return cls_cxx_remove(hctx);
+  }
+
+  map<string, bufferlist> attrset;
+  int ret = cls_cxx_getxattrs(hctx, &attrset);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_getxattrs() returned %d", __func__, ret);
+    return ret;
+  }
+
+  map<string, bufferlist> new_attrs;
+  for (list<string>::iterator iter = op.keep_attr_prefixes.begin();
+       iter != op.keep_attr_prefixes.end(); ++iter) {
+    string& check_prefix = *iter;
+
+    for (map<string, bufferlist>::iterator aiter = attrset.lower_bound(check_prefix);
+         aiter != attrset.end(); ++aiter) {
+      const string& attr = aiter->first;
+
+      if (attr.substr(0, check_prefix.size()) > check_prefix) {
+        break;
+      }
+
+      new_attrs[attr] = aiter->second;
+    }
+  }
+
+  CLS_LOG(20, "%s(): removing object", __func__);
+  ret = cls_cxx_remove(hctx);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_remove returned %d", __func__, ret);
+    return ret;
+  }
+
+  if (new_attrs.empty()) {
+    /* no data to keep */
+    return 0;
+  }
+
+  ret = cls_cxx_create(hctx, false);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_create returned %d", __func__, ret);
+    return ret;
+  }
+
+  for (map<string, bufferlist>::iterator aiter = new_attrs.begin();
+       aiter != new_attrs.end(); ++aiter) {
+    const string& attr = aiter->first;
+
+    ret = cls_cxx_setxattr(hctx, attr.c_str(), &aiter->second);
+    CLS_LOG(20, "%s(): setting attr: %s", __func__, attr.c_str());
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: %s(): cls_cxx_setxattr (attr=%s) returned %d", __func__, attr.c_str(), ret);
+      return ret;
+    }
+  }
+
+  return 0;
+}
+
+static int rgw_obj_store_pg_ver(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_obj_store_pg_ver_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  bufferlist bl;
+  uint64_t ver = cls_current_version(hctx);
+  encode(ver, bl);
+  int ret = cls_cxx_setxattr(hctx, op.attr.c_str(), &bl);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_setxattr (attr=%s) returned %d", __func__, op.attr.c_str(), ret);
+    return ret;
+  }
+
+  return 0;
+}
+
+static int rgw_obj_check_attrs_prefix(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_obj_check_attrs_prefix op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  if (op.check_prefix.empty()) {
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> attrset;
+  int ret = cls_cxx_getxattrs(hctx, &attrset);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_getxattrs() returned %d", __func__, ret);
+    return ret;
+  }
+
+  bool exist = false;
+
+  for (map<string, bufferlist>::iterator aiter = attrset.lower_bound(op.check_prefix);
+       aiter != attrset.end(); ++aiter) {
+    const string& attr = aiter->first;
+
+    if (attr.substr(0, op.check_prefix.size()) > op.check_prefix) {
+      break;
+    }
+
+    exist = true;
+  }
+
+  if (exist == op.fail_if_exist) {
+    return -ECANCELED;
+  }
+
+  return 0;
+}
+
+static int rgw_obj_check_mtime(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_obj_check_mtime op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  real_time obj_ut;
+  int ret = cls_cxx_stat2(hctx, NULL, &obj_ut);
+  if (ret < 0 && ret != -ENOENT) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_stat() returned %d", __func__, ret);
+    return ret;
+  }
+  if (ret == -ENOENT) {
+    CLS_LOG(10, "object does not exist, skipping check");
+  }
+
+  ceph_timespec obj_ts = ceph::real_clock::to_ceph_timespec(obj_ut);
+  ceph_timespec op_ts = ceph::real_clock::to_ceph_timespec(op.mtime);
+
+  if (!op.high_precision_time) {
+    obj_ts.tv_nsec = 0;
+    op_ts.tv_nsec = 0;
+  }
+
+  CLS_LOG(10, "%s: obj_ut=%lld.%06lld op.mtime=%lld.%06lld", __func__,
+          (long long)obj_ts.tv_sec, (long long)obj_ts.tv_nsec,
+          (long long)op_ts.tv_sec, (long long)op_ts.tv_nsec);
+
+  bool check;
+
+  switch (op.type) {
+  case CLS_RGW_CHECK_TIME_MTIME_EQ:
+    check = (obj_ts == op_ts);
+    break;
+  case CLS_RGW_CHECK_TIME_MTIME_LT:
+    check = (obj_ts < op_ts);
+    break;
+  case CLS_RGW_CHECK_TIME_MTIME_LE:
+    check = (obj_ts <= op_ts);
+    break;
+  case CLS_RGW_CHECK_TIME_MTIME_GT:
+    check = (obj_ts > op_ts);
+    break;
+  case CLS_RGW_CHECK_TIME_MTIME_GE:
+    check = (obj_ts >= op_ts);
+    break;
+  default:
+    return -EINVAL;
+  };
+
+  if (!check) {
+    return -ECANCELED;
+  }
+
+  return 0;
+}
+
+static int rgw_bi_get_op(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_bi_get_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  string idx;
+
+  switch (op.type) {
+    case BIIndexType::Plain:
+      idx = op.key.name;
+      break;
+    case BIIndexType::Instance:
+      encode_obj_index_key(op.key, &idx);
+      break;
+    case BIIndexType::OLH:
+      encode_olh_data_key(op.key, &idx);
+      break;
+    default:
+      CLS_LOG(10, "%s(): invalid key type encoding: %d",
+	      __func__, int(op.type));
+      return -EINVAL;
+  }
+
+  rgw_cls_bi_get_ret op_ret;
+
+  rgw_cls_bi_entry& entry = op_ret.entry;
+
+  entry.type = op.type;
+  entry.idx = idx;
+
+  int r = cls_cxx_map_get_val(hctx, idx, &entry.data);
+  if (r < 0) {
+      CLS_LOG(10, "%s(): cls_cxx_map_get_val() returned %d", __func__, r);
+      return r;
+  }
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+static int rgw_bi_put_op(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_bi_put_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  rgw_cls_bi_entry& entry = op.entry;
+
+  int r = cls_cxx_map_set_val(hctx, entry.idx, &entry.data);
+  if (r < 0) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_map_set_val() returned r=%d", __func__, r);
+  }
+
+  return 0;
+}
+
+static int list_plain_entries(cls_method_context_t hctx, const string& name, const string& marker, uint32_t max,
+                              list<rgw_cls_bi_entry> *entries, bool *pmore)
+{
+  string filter = name;
+  string start_key = marker;
+
+  string end_key; // stop listing at bi_log_prefix
+  bi_log_prefix(end_key);
+
+  int count = 0;
+  map<string, bufferlist> keys;
+  int ret = cls_cxx_map_get_vals(hctx, start_key, filter, max, &keys, pmore);
+  if (ret < 0) {
+    return ret;
+  }
+
+  map<string, bufferlist>::iterator iter;
+  for (iter = keys.begin(); iter != keys.end(); ++iter) {
+    if (iter->first >= end_key) {
+      /* past the end of plain namespace */
+      if (pmore) {
+	*pmore = false;
+      }
+      return count;
+    }
+
+    rgw_cls_bi_entry entry;
+    entry.type = BIIndexType::Plain;
+    entry.idx = iter->first;
+    entry.data = iter->second;
+
+    auto biter = entry.data.cbegin();
+
+    rgw_bucket_dir_entry e;
+    try {
+      decode(e, biter);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR: %s(): failed to decode buffer", __func__);
+      return -EIO;
+    }
+
+    CLS_LOG(20, "%s(): entry.idx=%s e.key.name=%s", __func__, escape_str(entry.idx).c_str(), escape_str(e.key.name).c_str());
+
+    if (!name.empty() && e.key.name != name) {
+      /* we are skipping the rest of the entries */
+      if (pmore) {
+	*pmore = false;
+      }
+      return count;
+    }
+
+    entries->push_back(entry);
+    count++;
+    if (count >= (int)max) {
+      return count;
+    }
+    start_key = entry.idx;
+  }
+
+  return count;
+}
+
+static int list_instance_entries(cls_method_context_t hctx, const string& name, const string& marker, uint32_t max,
+                                 list<rgw_cls_bi_entry> *entries, bool *pmore)
+{
+  cls_rgw_obj_key key(name);
+  string first_instance_idx;
+  encode_obj_versioned_data_key(key, &first_instance_idx);
+  string start_key;
+
+  if (!name.empty()) {
+    start_key = first_instance_idx;
+  } else {
+    start_key = BI_PREFIX_CHAR;
+    start_key.append(bucket_index_prefixes[BI_BUCKET_OBJ_INSTANCE_INDEX]);
+  }
+  string filter = start_key;
+  if (bi_entry_gt(marker, start_key)) {
+    start_key = marker;
+  }
+  int count = 0;
+  map<string, bufferlist> keys;
+  bufferlist k;
+  int ret = cls_cxx_map_get_val(hctx, start_key, &k);
+  if (ret < 0 && ret != -ENOENT) {
+    return ret;
+  }
+  bool found_first = (ret == 0);
+  if (found_first) {
+    --max;
+  }
+  if (max > 0) {
+    ret = cls_cxx_map_get_vals(hctx, start_key, string(), max, &keys, pmore);
+    CLS_LOG(20, "%s(): start_key=%s first_instance_idx=%s keys.size()=%d", __func__, escape_str(start_key).c_str(), escape_str(first_instance_idx).c_str(), (int)keys.size());
+    if (ret < 0) {
+      return ret;
+    }
+  }
+  if (found_first) {
+    keys[start_key].claim(k);
+  }
+
+  map<string, bufferlist>::iterator iter;
+  for (iter = keys.begin(); iter != keys.end(); ++iter) {
+    rgw_cls_bi_entry entry;
+    entry.type = BIIndexType::Instance;
+    entry.idx = iter->first;
+    entry.data = iter->second;
+
+    if (!filter.empty() && entry.idx.compare(0, filter.size(), filter) != 0) {
+      /* we are skipping the rest of the entries */
+      if (pmore) {
+	*pmore = false;
+      }
+      return count;
+    }
+
+    CLS_LOG(20, "%s(): entry.idx=%s", __func__, escape_str(entry.idx).c_str());
+
+    auto biter = entry.data.cbegin();
+
+    rgw_bucket_dir_entry e;
+    try {
+      decode(e, biter);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR: %s(): failed to decode buffer (size=%d)", __func__, entry.data.length());
+      return -EIO;
+    }
+
+    if (!name.empty() && e.key.name != name) {
+      /* we are skipping the rest of the entries */
+      if (pmore) {
+	*pmore = false;
+      }
+      return count;
+    }
+
+    entries->push_back(entry);
+    count++;
+    start_key = entry.idx;
+  }
+
+  return count;
+}
+
+static int list_olh_entries(cls_method_context_t hctx, const string& name, const string& marker, uint32_t max,
+                            list<rgw_cls_bi_entry> *entries, bool *pmore)
+{
+  cls_rgw_obj_key key(name);
+  string first_instance_idx;
+  encode_olh_data_key(key, &first_instance_idx);
+  string start_key;
+
+  if (!name.empty()) {
+    start_key = first_instance_idx;
+  } else {
+    start_key = BI_PREFIX_CHAR;
+    start_key.append(bucket_index_prefixes[BI_BUCKET_OLH_DATA_INDEX]);
+  }
+  string filter = start_key;
+  if (bi_entry_gt(marker, start_key)) {
+    start_key = marker;
+  }
+  int count = 0;
+  map<string, bufferlist> keys;
+  int ret;
+  bufferlist k;
+  ret = cls_cxx_map_get_val(hctx, start_key, &k);
+  if (ret < 0 && ret != -ENOENT) {
+    return ret;
+  }
+  bool found_first = (ret == 0);
+  if (found_first) {
+    --max;
+  }
+  if (max > 0) {
+    ret = cls_cxx_map_get_vals(hctx, start_key, string(), max, &keys, pmore);
+    CLS_LOG(20, "%s(): start_key=%s first_instance_idx=%s keys.size()=%d", __func__, escape_str(start_key).c_str(), escape_str(first_instance_idx).c_str(), (int)keys.size());
+    if (ret < 0) {
+      return ret;
+    }
+  }
+
+  if (found_first) {
+    keys[start_key].claim(k);
+  }
+
+  map<string, bufferlist>::iterator iter;
+  for (iter = keys.begin(); iter != keys.end(); ++iter) {
+    rgw_cls_bi_entry entry;
+    entry.type = BIIndexType::OLH;
+    entry.idx = iter->first;
+    entry.data = iter->second;
+
+    if (!filter.empty() && entry.idx.compare(0, filter.size(), filter) != 0) {
+      /* we are skipping the rest of the entries */
+      if (pmore) {
+	*pmore = false;
+      }
+      return count;
+    }
+
+    CLS_LOG(20, "%s(): entry.idx=%s", __func__, escape_str(entry.idx).c_str());
+
+    auto biter = entry.data.cbegin();
+
+    rgw_bucket_olh_entry e;
+    try {
+      decode(e, biter);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR: %s(): failed to decode buffer (size=%d)", __func__, entry.data.length());
+      return -EIO;
+    }
+
+    if (!name.empty() && e.key.name != name) {
+      /* we are skipping the rest of the entries */
+      if (pmore) {
+	*pmore = false;
+      }
+      return count;
+    }
+
+    entries->push_back(entry);
+    count++;
+    start_key = entry.idx;
+  }
+
+  return count;
+}
+
+static int rgw_bi_list_op(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // decode request
+  rgw_cls_bi_list_op op;
+  auto iter = in->cbegin();
+  try {
+    decode(op, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s(): failed to decode request", __func__);
+    return -EINVAL;
+  }
+
+  rgw_cls_bi_list_ret op_ret;
+
+  string filter = op.name;
+#define MAX_BI_LIST_ENTRIES 1000
+  int32_t max = (op.max < MAX_BI_LIST_ENTRIES ? op.max : MAX_BI_LIST_ENTRIES);
+  string start_key = op.marker;
+  bool more;
+  int ret = list_plain_entries(hctx, op.name, op.marker, max, &op_ret.entries, &more);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: %s(): list_plain_entries returned ret=%d", __func__, ret);
+    return ret;
+  }
+  int count = ret;
+
+  CLS_LOG(20, "found %d plain entries", count);
+
+  if (!more) {
+    ret = list_instance_entries(hctx, op.name, op.marker, max - count, &op_ret.entries, &more);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: %s(): list_instance_entries returned ret=%d", __func__, ret);
+      return ret;
+    }
+
+    count += ret;
+  }
+
+  if (!more) {
+    ret = list_olh_entries(hctx, op.name, op.marker, max - count, &op_ret.entries, &more);
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: %s(): list_olh_entries returned ret=%d", __func__, ret);
+      return ret;
+    }
+
+    count += ret;
+  }
+
+  op_ret.is_truncated = (count >= max) || more;
+  while (count > max) {
+    op_ret.entries.pop_back();
+    count--;
+  }
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+int bi_log_record_decode(bufferlist& bl, rgw_bi_log_entry& e)
+{
+  auto iter = bl.cbegin();
+  try {
+    decode(e, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: failed to decode rgw_bi_log_entry");
+    return -EIO;
+  }
+  return 0;
+}
+
+static int bi_log_iterate_entries(cls_method_context_t hctx, const string& marker, const string& end_marker,
+                              string& key_iter, uint32_t max_entries, bool *truncated,
+                              int (*cb)(cls_method_context_t, const string&, rgw_bi_log_entry&, void *),
+                              void *param)
+{
+  CLS_LOG(10, "bi_log_iterate_range");
+
+  map<string, bufferlist> keys;
+  string filter_prefix, end_key;
+  uint32_t i = 0;
+  string key;
+
+  if (truncated)
+    *truncated = false;
+
+  string start_key;
+  if (key_iter.empty()) {
+    key = BI_PREFIX_CHAR;
+    key.append(bucket_index_prefixes[BI_BUCKET_LOG_INDEX]);
+    key.append(marker);
+
+    start_key = key;
+  } else {
+    start_key = key_iter;
+  }
+
+  if (end_marker.empty()) {
+    end_key = BI_PREFIX_CHAR;
+    end_key.append(bucket_index_prefixes[BI_BUCKET_LOG_INDEX + 1]);
+  } else {
+    end_key = BI_PREFIX_CHAR;
+    end_key.append(bucket_index_prefixes[BI_BUCKET_LOG_INDEX]);
+    end_key.append(end_marker);
+  }
+
+  CLS_LOG(10, "bi_log_iterate_entries start_key=%s end_key=%s\n", start_key.c_str(), end_key.c_str());
+
+  string filter;
+
+  int ret = cls_cxx_map_get_vals(hctx, start_key, filter, max_entries, &keys, truncated);
+  if (ret < 0)
+    return ret;
+
+  map<string, bufferlist>::iterator iter = keys.begin();
+  if (iter == keys.end())
+    return 0;
+
+  uint32_t num_keys = keys.size();
+
+  for (; iter != keys.end(); ++iter,++i) {
+    const string& key = iter->first;
+    rgw_bi_log_entry e;
+
+    CLS_LOG(10, "bi_log_iterate_entries key=%s bl.length=%d\n", key.c_str(), (int)iter->second.length());
+
+    if (key.compare(end_key) > 0) {
+      key_iter = key;
+      if (truncated) {
+        *truncated = false;
+      }
+      return 0;
+    }
+
+    ret = bi_log_record_decode(iter->second, e);
+    if (ret < 0)
+      return ret;
+
+    ret = cb(hctx, key, e, param);
+    if (ret < 0)
+      return ret;
+
+    if (i == num_keys - 1) {
+      key_iter = key;
+    }
+  }
+
+  return 0;
+}
+
+static int bi_log_list_cb(cls_method_context_t hctx, const string& key, rgw_bi_log_entry& info, void *param)
+{
+  list<rgw_bi_log_entry> *l = (list<rgw_bi_log_entry> *)param;
+  l->push_back(info);
+  return 0;
+}
+
+static int bi_log_list_entries(cls_method_context_t hctx, const string& marker,
+			   uint32_t max, list<rgw_bi_log_entry>& entries, bool *truncated)
+{
+  string key_iter;
+  string end_marker;
+  int ret = bi_log_iterate_entries(hctx, marker, end_marker,
+                              key_iter, max, truncated,
+                              bi_log_list_cb, &entries);
+  return ret;
+}
+
+static int rgw_bi_log_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_bi_log_list_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_bi_log_list(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  cls_rgw_bi_log_list_ret op_ret;
+  int ret = bi_log_list_entries(hctx, op.marker, op.max, op_ret.entries, &op_ret.truncated);
+  if (ret < 0)
+    return ret;
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+static int bi_log_list_trim_cb(cls_method_context_t hctx, const string& key, rgw_bi_log_entry& info, void *param)
+{
+  list<rgw_bi_log_entry> *entries = (list<rgw_bi_log_entry> *)param;
+
+  entries->push_back(info);
+  return 0;
+}
+
+static int bi_log_remove_entry(cls_method_context_t hctx, rgw_bi_log_entry& entry)
+{
+  string key;
+  key = BI_PREFIX_CHAR;
+  key.append(bucket_index_prefixes[BI_BUCKET_LOG_INDEX]);
+  key.append(entry.id);
+  return cls_cxx_map_remove_key(hctx, key);
+}
+
+static int bi_log_list_trim_entries(cls_method_context_t hctx,
+                                    const string& start_marker, const string& end_marker,
+			            list<rgw_bi_log_entry>& entries, bool *truncated)
+{
+  string key_iter;
+#define MAX_TRIM_ENTRIES 1000 /* max entries to trim in a single operation */
+  int ret = bi_log_iterate_entries(hctx, start_marker, end_marker,
+                              key_iter, MAX_TRIM_ENTRIES, truncated,
+                              bi_log_list_trim_cb, &entries);
+  return ret;
+}
+
+static int rgw_bi_log_trim(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_bi_log_trim_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_bi_log_list(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  cls_rgw_bi_log_list_ret op_ret;
+  list<rgw_bi_log_entry> entries;
+#define MAX_TRIM_ENTRIES 1000 /* don't do more than that in a single operation */
+  bool truncated;
+  int ret = bi_log_list_trim_entries(hctx, op.start_marker, op.end_marker, entries, &truncated);
+  if (ret < 0)
+    return ret;
+
+  if (entries.empty())
+    return -ENODATA;
+
+  list<rgw_bi_log_entry>::iterator iter;
+  for (iter = entries.begin(); iter != entries.end(); ++iter) {
+    rgw_bi_log_entry& entry = *iter;
+
+    ret = bi_log_remove_entry(hctx, entry);
+    if (ret < 0)
+      return ret;
+  }
+
+  return 0;
+}
+
+static int rgw_bi_log_resync(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_complete_op(): failed to read header\n");
+    return rc;
+  }
+
+  bufferlist bl;
+
+  rgw_bi_log_entry entry;
+
+  entry.timestamp = real_clock::now();
+  entry.op = RGWModifyOp::CLS_RGW_OP_RESYNC;
+  entry.state = RGWPendingState::CLS_RGW_STATE_COMPLETE;
+
+  string key;
+  bi_log_index_key(hctx, key, entry.id, header.ver);
+
+  encode(entry, bl);
+
+  if (entry.id > header.max_marker)
+    header.max_marker = entry.id;
+
+  header.syncstopped = false;
+
+  rc = cls_cxx_map_set_val(hctx, key, &bl);
+  if (rc < 0)
+    return rc;
+
+  return write_bucket_header(hctx, &header);
+}
+
+static int rgw_bi_log_stop(cls_method_context_t hctx, bufferlist *in, bufferlist *out) 
+{
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: rgw_bucket_complete_op(): failed to read header\n");
+    return rc;
+  }
+
+  bufferlist bl;
+
+  rgw_bi_log_entry entry;
+
+  entry.timestamp = real_clock::now();
+  entry.op = RGWModifyOp::CLS_RGW_OP_SYNCSTOP;
+  entry.state = RGWPendingState::CLS_RGW_STATE_COMPLETE;
+
+  string key;
+  bi_log_index_key(hctx, key, entry.id, header.ver);
+
+  encode(entry, bl);
+
+  if (entry.id > header.max_marker)
+    header.max_marker = entry.id;
+  header.syncstopped = true;
+
+  rc = cls_cxx_map_set_val(hctx, key, &bl);
+  if (rc < 0)
+    return rc;
+
+  return write_bucket_header(hctx, &header);
+}
+
+
+static void usage_record_prefix_by_time(uint64_t epoch, string& key)
+{
+  char buf[32];
+  snprintf(buf, sizeof(buf), "%011llu", (long long unsigned)epoch);
+  key = buf;
+}
+
+static void usage_record_prefix_by_user(const string& user, uint64_t epoch, string& key)
+{
+  char buf[user.size() + 32];
+  snprintf(buf, sizeof(buf), "%s_%011llu_", user.c_str(), (long long unsigned)epoch);
+  key = buf;
+}
+
+static void usage_record_name_by_time(uint64_t epoch, const string& user, const string& bucket, string& key)
+{
+  char buf[32 + user.size() + bucket.size()];
+  snprintf(buf, sizeof(buf), "%011llu_%s_%s", (long long unsigned)epoch, user.c_str(), bucket.c_str());
+  key = buf;
+}
+
+static void usage_record_name_by_user(const string& user, uint64_t epoch, const string& bucket, string& key)
+{
+  char buf[32 + user.size() + bucket.size()];
+  snprintf(buf, sizeof(buf), "%s_%011llu_%s", user.c_str(), (long long unsigned)epoch, bucket.c_str());
+  key = buf;
+}
+
+static int usage_record_decode(bufferlist& record_bl, rgw_usage_log_entry& e)
+{
+  auto kiter = record_bl.cbegin();
+  try {
+    decode(e, kiter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: usage_record_decode(): failed to decode record_bl\n");
+    return -EINVAL;
+  }
+
+  return 0;
+}
+
+int rgw_user_usage_log_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(10, "rgw_user_usage_log_add()");
+
+  auto in_iter = in->cbegin();
+  rgw_cls_usage_log_add_op op;
+
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_user_usage_log_add(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  rgw_usage_log_info& info = op.info;
+  vector<rgw_usage_log_entry>::iterator iter;
+
+  for (iter = info.entries.begin(); iter != info.entries.end(); ++iter) {
+    rgw_usage_log_entry& entry = *iter;
+    string key_by_time;
+
+    rgw_user *puser = (entry.payer.empty() ? &entry.owner : &entry.payer);
+
+    usage_record_name_by_time(entry.epoch, puser->to_str(), entry.bucket, key_by_time);
+
+    CLS_LOG(10, "rgw_user_usage_log_add user=%s bucket=%s\n", puser->to_str().c_str(), entry.bucket.c_str());
+
+    bufferlist record_bl;
+    int ret = cls_cxx_map_get_val(hctx, key_by_time, &record_bl);
+    if (ret < 0 && ret != -ENOENT) {
+      CLS_LOG(1, "ERROR: rgw_user_usage_log_add(): cls_cxx_map_read_key returned %d\n", ret);
+      return -EINVAL;
+    }
+    if (ret >= 0) {
+      rgw_usage_log_entry e;
+      ret = usage_record_decode(record_bl, e);
+      if (ret < 0)
+        return ret;
+      CLS_LOG(10, "rgw_user_usage_log_add aggregating existing bucket\n");
+      entry.aggregate(e);
+    }
+
+    bufferlist new_record_bl;
+    encode(entry, new_record_bl);
+    ret = cls_cxx_map_set_val(hctx, key_by_time, &new_record_bl);
+    if (ret < 0)
+      return ret;
+
+    string key_by_user;
+    usage_record_name_by_user(puser->to_str(), entry.epoch, entry.bucket, key_by_user);
+    ret = cls_cxx_map_set_val(hctx, key_by_user, &new_record_bl);
+    if (ret < 0)
+      return ret;
+  }
+
+  return 0;
+}
+
+static int usage_iterate_range(cls_method_context_t hctx, uint64_t start, uint64_t end, const string& user,
+                            const string& bucket, string& key_iter, uint32_t max_entries, bool *truncated,
+                            int (*cb)(cls_method_context_t, const string&, rgw_usage_log_entry&, void *),
+                            void *param)
+{
+  CLS_LOG(10, "usage_iterate_range");
+
+  map<string, bufferlist> keys;
+  string filter_prefix;
+  string start_key, end_key;
+  bool by_user = !user.empty();
+  string user_key;
+  bool truncated_status = false;
+
+  ceph_assert(truncated != nullptr);
+
+  if (!by_user) {
+    usage_record_prefix_by_time(end, end_key);
+  } else {
+    user_key = user;
+    user_key.append("_");
+  }
+
+  if (key_iter.empty()) {
+    if (by_user) {
+      usage_record_prefix_by_user(user, start, start_key);
+    } else {
+      usage_record_prefix_by_time(start, start_key);
+    }
+  } else {
+    start_key = key_iter;
+  }
+
+  CLS_LOG(20, "usage_iterate_range start_key=%s", start_key.c_str());
+  int ret = cls_cxx_map_get_vals(hctx, start_key, filter_prefix, max_entries, &keys, &truncated_status);
+  if (ret < 0)
+    return ret;
+
+  *truncated = truncated_status;
+      
+  map<string, bufferlist>::iterator iter = keys.begin();
+  if (iter == keys.end())
+    return 0;
+
+  for (; iter != keys.end(); ++iter) {
+    const string& key = iter->first;
+    rgw_usage_log_entry e;
+
+    key_iter = key;
+    if (!by_user && key.compare(end_key) >= 0) {
+      CLS_LOG(20, "usage_iterate_range reached key=%s, done", key.c_str());
+      *truncated = false;
+      key_iter = key;
+      return 0;
+    }
+
+    if (by_user && key.compare(0, user_key.size(), user_key) != 0) {
+      CLS_LOG(20, "usage_iterate_range reached key=%s, done", key.c_str());
+      *truncated = false;
+      key_iter = key;
+      return 0;
+    }
+
+    ret = usage_record_decode(iter->second, e);
+    if (ret < 0)
+      return ret;
+
+    if (!bucket.empty() && bucket.compare(e.bucket))
+      continue;
+
+    if (e.epoch < start)
+      continue;
+
+    /* keys are sorted by epoch, so once we're past end we're done */
+    if (e.epoch >= end) {
+      *truncated = false;
+      return 0;
+    }
+
+    ret = cb(hctx, key, e, param);
+    if (ret < 0)
+      return ret;
+  }
+  return 0;
+}
+
+static int usage_log_read_cb(cls_method_context_t hctx, const string& key, rgw_usage_log_entry& entry, void *param)
+{
+  map<rgw_user_bucket, rgw_usage_log_entry> *usage = (map<rgw_user_bucket, rgw_usage_log_entry> *)param;
+  rgw_user *puser;
+  if (!entry.payer.empty()) {
+    puser = &entry.payer;
+  } else {
+    puser = &entry.owner;
+  }
+  rgw_user_bucket ub(puser->to_str(), entry.bucket);
+  rgw_usage_log_entry& le = (*usage)[ub];
+  le.aggregate(entry);
+
+  return 0;
+}
+
+int rgw_user_usage_log_read(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(10, "rgw_user_usage_log_read()");
+
+  auto in_iter = in->cbegin();
+  rgw_cls_usage_log_read_op op;
+
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_user_usage_log_read(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  rgw_cls_usage_log_read_ret ret_info;
+  map<rgw_user_bucket, rgw_usage_log_entry> *usage = &ret_info.usage;
+  string iter = op.iter;
+#define MAX_ENTRIES 1000
+  uint32_t max_entries = (op.max_entries ? op.max_entries : MAX_ENTRIES);
+  int ret = usage_iterate_range(hctx, op.start_epoch, op.end_epoch, op.owner, op.bucket, iter, max_entries, &ret_info.truncated, usage_log_read_cb, (void *)usage);
+  if (ret < 0)
+    return ret;
+
+  if (ret_info.truncated)
+    ret_info.next_iter = iter;
+
+  encode(ret_info, *out);
+  return 0;
+}
+
+static int usage_log_trim_cb(cls_method_context_t hctx, const string& key, rgw_usage_log_entry& entry, void *param)
+{
+  bool *found = (bool *)param;
+  if (found) {
+    *found = true;
+  }
+  string key_by_time;
+  string key_by_user;
+
+  string o = entry.owner.to_str();
+  usage_record_name_by_time(entry.epoch, o, entry.bucket, key_by_time);
+  usage_record_name_by_user(o, entry.epoch, entry.bucket, key_by_user);
+
+  int ret = cls_cxx_map_remove_key(hctx, key_by_time);
+  if (ret < 0)
+    return ret;
+
+  return cls_cxx_map_remove_key(hctx, key_by_user);
+}
+
+int rgw_user_usage_log_trim(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(10, "rgw_user_usage_log_trim()");
+
+  /* only continue if object exists! */
+  int ret = cls_cxx_stat(hctx, NULL, NULL);
+  if (ret < 0)
+    return ret;
+
+  auto in_iter = in->cbegin();
+  rgw_cls_usage_log_trim_op op;
+
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_user_log_usage_log_trim(): failed to decode request\n");
+    return -EINVAL;
+  }
+
+  string iter;
+  bool more;
+  bool found = false;
+#define MAX_USAGE_TRIM_ENTRIES 128
+  ret = usage_iterate_range(hctx, op.start_epoch, op.end_epoch, op.user, op.bucket, iter, MAX_USAGE_TRIM_ENTRIES, &more, usage_log_trim_cb, (void *)&found);
+  if (ret < 0)
+    return ret;
+
+  if (!more && !found)
+    return -ENODATA;
+
+  return 0;
+}
+
+int rgw_usage_log_clear(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(10,"%s", __func__);
+
+  int ret = cls_cxx_map_clear(hctx);
+  /* if object doesn't exist all the logs are cleared anyway */
+  if (ret == -ENOENT)
+    ret = 0;
+
+  return ret;
+}
+/*
+ * We hold the garbage collection chain data under two different indexes: the first 'name' index
+ * keeps them under a unique tag that represents the chains, and a second 'time' index keeps
+ * them by their expiration timestamp
+ */
+#define GC_OBJ_NAME_INDEX 0
+#define GC_OBJ_TIME_INDEX 1
+
+static string gc_index_prefixes[] = { "0_",
+                                      "1_" };
+
+static void prepend_index_prefix(const string& src, int index, string *dest)
+{
+  *dest = gc_index_prefixes[index];
+  dest->append(src);
+}
+
+static int gc_omap_get(cls_method_context_t hctx, int type, const string& key, cls_rgw_gc_obj_info *info)
+{
+  string index;
+  prepend_index_prefix(key, type, &index);
+
+  int ret = read_omap_entry(hctx, index, info);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int gc_omap_set(cls_method_context_t hctx, int type, const string& key, const cls_rgw_gc_obj_info *info)
+{
+  bufferlist bl;
+  encode(*info, bl);
+
+  string index = gc_index_prefixes[type];
+  index.append(key);
+
+  int ret = cls_cxx_map_set_val(hctx, index, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int gc_omap_remove(cls_method_context_t hctx, int type, const string& key)
+{
+  string index = gc_index_prefixes[type];
+  index.append(key);
+
+  int ret = cls_cxx_map_remove_key(hctx, index);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static bool key_in_index(const string& key, int index_type)
+{
+  const string& prefix = gc_index_prefixes[index_type];
+  return (key.compare(0, prefix.size(), prefix) == 0);
+}
+
+
+static int gc_update_entry(cls_method_context_t hctx, uint32_t expiration_secs,
+                           cls_rgw_gc_obj_info& info)
+{
+  cls_rgw_gc_obj_info old_info;
+  int ret = gc_omap_get(hctx, GC_OBJ_NAME_INDEX, info.tag, &old_info);
+  if (ret == 0) {
+    string key;
+    get_time_key(old_info.time, &key);
+    ret = gc_omap_remove(hctx, GC_OBJ_TIME_INDEX, key);
+    if (ret < 0 && ret != -ENOENT) {
+      CLS_LOG(0, "ERROR: failed to remove key=%s\n", key.c_str());
+      return ret;
+    }
+  }
+
+  // calculate time and time key
+  info.time = ceph::real_clock::now();
+  info.time += make_timespan(expiration_secs);
+  string time_key;
+  get_time_key(info.time, &time_key);
+
+  if (info.chain.objs.empty()) {
+    CLS_LOG(0,
+	    "WARNING: %s setting GC log entry with zero-length chain, "
+	    "tag='%s', timekey='%s'",
+	    __func__, info.tag.c_str(), time_key.c_str());
+  }
+
+  ret = gc_omap_set(hctx, GC_OBJ_NAME_INDEX, info.tag, &info);
+  if (ret < 0)
+    return ret;
+
+  ret = gc_omap_set(hctx, GC_OBJ_TIME_INDEX, time_key, &info);
+  if (ret < 0)
+    goto done_err;
+
+  return 0;
+
+done_err:
+
+  CLS_LOG(0, "ERROR: gc_set_entry error info.tag=%s, ret=%d\n",
+	  info.tag.c_str(), ret);
+  gc_omap_remove(hctx, GC_OBJ_NAME_INDEX, info.tag);
+
+  return ret;
+}
+
+static int gc_defer_entry(cls_method_context_t hctx, const string& tag, uint32_t expiration_secs)
+{
+  cls_rgw_gc_obj_info info;
+  int ret = gc_omap_get(hctx, GC_OBJ_NAME_INDEX, tag, &info);
+  if (ret == -ENOENT)
+    return 0;
+  if (ret < 0)
+    return ret;
+  return gc_update_entry(hctx, expiration_secs, info);
+}
+
+int gc_record_decode(bufferlist& bl, cls_rgw_gc_obj_info& e)
+{
+  auto iter = bl.cbegin();
+  try {
+    decode(e, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: failed to decode cls_rgw_gc_obj_info");
+    return -EIO;
+  }
+  return 0;
+}
+
+static int rgw_cls_gc_set_entry(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_gc_set_entry_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_gc_set_entry(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  return gc_update_entry(hctx, op.expiration_secs, op.info);
+}
+
+static int rgw_cls_gc_defer_entry(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_gc_defer_entry_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_gc_defer_entry(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  return gc_defer_entry(hctx, op.tag, op.expiration_secs);
+}
+
+static int gc_iterate_entries(cls_method_context_t hctx,
+			      const string& marker,
+			      bool expired_only,
+                              string& out_marker,
+			      uint32_t max_entries,
+			      bool *truncated,
+                              int (*cb)(cls_method_context_t,
+					const string&,
+					cls_rgw_gc_obj_info&,
+					void *),
+                              void *param)
+{
+  CLS_LOG(10, "gc_iterate_entries");
+
+  map<string, bufferlist> keys;
+  string filter_prefix, end_key;
+  string key;
+
+  if (truncated)
+    *truncated = false;
+
+  string start_key;
+  if (marker.empty()) {
+    prepend_index_prefix(marker, GC_OBJ_TIME_INDEX, &start_key);
+  } else {
+    start_key = marker;
+  }
+
+  if (expired_only) {
+    real_time now = ceph::real_clock::now();
+    string now_str;
+    get_time_key(now, &now_str);
+    prepend_index_prefix(now_str, GC_OBJ_TIME_INDEX, &end_key);
+
+    CLS_LOG(10, "gc_iterate_entries end_key=%s\n", end_key.c_str());
+  }
+
+  string filter;
+
+  int ret = cls_cxx_map_get_vals(hctx, start_key, filter, max_entries,
+				 &keys, truncated);
+  if (ret < 0)
+    return ret;
+
+  map<string, bufferlist>::iterator iter = keys.begin();
+  if (iter == keys.end()) {
+    // if keys empty must not come back as truncated
+    ceph_assert(!truncated || !(*truncated));
+    return 0;
+  }
+
+  const string* last_key = nullptr; // last key processed, for end-marker
+  for (; iter != keys.end(); ++iter) {
+    const string& key = iter->first;
+    cls_rgw_gc_obj_info e;
+
+    CLS_LOG(10, "gc_iterate_entries key=%s\n", key.c_str());
+
+    if (!end_key.empty() && key.compare(end_key) >= 0) {
+      if (truncated)
+        *truncated = false;
+      return 0;
+    }
+
+    if (!key_in_index(key, GC_OBJ_TIME_INDEX)) {
+      if (truncated)
+	*truncated = false;
+      return 0;
+    }
+
+    ret = gc_record_decode(iter->second, e);
+    if (ret < 0)
+      return ret;
+
+    ret = cb(hctx, key, e, param);
+    if (ret < 0)
+      return ret;
+    last_key = &(iter->first); // update when callback successful
+  }
+
+  // set the out marker if either caller does not capture truncated or
+  // if they do capture and we are truncated
+  if (!truncated || *truncated) {
+    assert(last_key);
+    out_marker = *last_key;
+  }
+
+  return 0;
+}
+
+static int gc_list_cb(cls_method_context_t hctx, const string& key, cls_rgw_gc_obj_info& info, void *param)
+{
+  list<cls_rgw_gc_obj_info> *l = (list<cls_rgw_gc_obj_info> *)param;
+  l->push_back(info);
+  return 0;
+}
+
+static int gc_list_entries(cls_method_context_t hctx, const string& marker,
+			   uint32_t max, bool expired_only,
+                           list<cls_rgw_gc_obj_info>& entries, bool *truncated, string& next_marker)
+{
+  int ret = gc_iterate_entries(hctx, marker, expired_only,
+                              next_marker, max, truncated,
+                              gc_list_cb, &entries);
+  return ret;
+}
+
+static int rgw_cls_gc_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_gc_list_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_gc_list(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  cls_rgw_gc_list_ret op_ret;
+#define GC_LIST_ENTRIES_DEFAULT 128
+  int ret = gc_list_entries(hctx, op.marker, (op.max ? op.max : GC_LIST_ENTRIES_DEFAULT), op.expired_only, 
+   op_ret.entries, &op_ret.truncated, op_ret.next_marker);
+  if (ret < 0)
+    return ret;
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+static int gc_remove(cls_method_context_t hctx, vector<string>& tags)
+{
+  for (auto iter = tags.begin(); iter != tags.end(); ++iter) {
+    string& tag = *iter;
+    cls_rgw_gc_obj_info info;
+    int ret = gc_omap_get(hctx, GC_OBJ_NAME_INDEX, tag, &info);
+    if (ret == -ENOENT) {
+      CLS_LOG(0, "couldn't find tag in name index tag=%s\n", tag.c_str());
+      continue;
+    }
+
+    if (ret < 0)
+      return ret;
+
+    string time_key;
+    get_time_key(info.time, &time_key);
+    ret = gc_omap_remove(hctx, GC_OBJ_TIME_INDEX, time_key);
+    if (ret < 0 && ret != -ENOENT)
+      return ret;
+    if (ret == -ENOENT) {
+      CLS_LOG(0, "couldn't find key in time index key=%s\n", time_key.c_str());
+    }
+
+    ret = gc_omap_remove(hctx, GC_OBJ_NAME_INDEX, tag);
+    if (ret < 0 && ret != -ENOENT)
+      return ret;
+  }
+
+  return 0;
+}
+
+static int rgw_cls_gc_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_gc_remove_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_gc_remove(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  return gc_remove(hctx, op.tags);
+}
+
+static int rgw_cls_lc_get_entry(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_lc_get_entry_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_set_entry(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  rgw_lc_entry_t lc_entry;
+  int ret = read_omap_entry(hctx, op.marker, &lc_entry);
+  if (ret < 0)
+    return ret;
+
+  cls_rgw_lc_get_entry_ret op_ret(std::move(lc_entry));
+  encode(op_ret, *out);
+  return 0;
+}
+
+
+static int rgw_cls_lc_set_entry(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_lc_set_entry_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_set_entry(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  bufferlist bl;
+  encode(op.entry, bl);
+
+  int ret = cls_cxx_map_set_val(hctx, op.entry.first, &bl);
+  return ret;
+}
+
+static int rgw_cls_lc_rm_entry(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_lc_rm_entry_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_rm_entry(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  int ret = cls_cxx_map_remove_key(hctx, op.entry.first);
+  return ret;
+}
+
+static int rgw_cls_lc_get_next_entry(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+  cls_rgw_lc_get_next_entry_ret op_ret;
+  cls_rgw_lc_get_next_entry_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_get_next_entry: failed to decode op\n");
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> vals;
+  string filter_prefix;
+  bool more;
+  int ret = cls_cxx_map_get_vals(hctx, op.marker, filter_prefix, 1, &vals, &more);
+  if (ret < 0)
+    return ret;
+  map<string, bufferlist>::iterator it;
+  pair<string, int> entry;
+  if (!vals.empty()) {
+    it=vals.begin();
+    in_iter = it->second.begin();
+    try {
+      decode(entry, in_iter);
+    } catch (buffer::error& err) {
+      CLS_LOG(1, "ERROR: rgw_cls_lc_get_next_entry(): failed to decode entry\n");
+      return -EIO;
+    }
+  }
+  op_ret.entry = entry;
+  encode(op_ret, *out);
+  return 0;
+}
+
+static int rgw_cls_lc_list_entries(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  cls_rgw_lc_list_entries_op op;
+  auto in_iter = in->cbegin();
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_list_entries(): failed to decode op\n");
+    return -EINVAL;
+  }
+
+  cls_rgw_lc_list_entries_ret op_ret;
+  bufferlist::const_iterator iter;
+  map<string, bufferlist> vals;
+  string filter_prefix;
+  int ret = cls_cxx_map_get_vals(hctx, op.marker, filter_prefix, op.max_entries, &vals, &op_ret.is_truncated);
+  if (ret < 0)
+    return ret;
+  map<string, bufferlist>::iterator it;
+  pair<string, int> entry;
+  for (it = vals.begin(); it != vals.end(); ++it) {
+    iter = it->second.cbegin();
+    try {
+    decode(entry, iter);
+    } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_list_entries(): failed to decode entry\n");
+    return -EIO;
+   }
+   op_ret.entries.insert(entry);
+  }
+  encode(op_ret, *out);
+  return 0;
+}
+
+static int rgw_cls_lc_put_head(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_lc_put_head_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_lc_put_head(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  bufferlist bl;
+  encode(op.head, bl);
+  int ret = cls_cxx_map_write_header(hctx,&bl);
+  return ret;
+}
+
+static int rgw_cls_lc_get_head(cls_method_context_t hctx, bufferlist *in,  bufferlist *out)
+{
+  bufferlist bl;
+  int ret = cls_cxx_map_read_header(hctx, &bl);
+  if (ret < 0)
+    return ret;
+  cls_rgw_lc_obj_head head;
+  if (bl.length() != 0) {
+    auto iter = bl.cbegin();
+    try {
+      decode(head, iter);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR: rgw_cls_lc_get_head(): failed to decode entry %s\n",err.what());
+      return -EINVAL;
+    }
+  } else {
+    head.start_date = 0;
+    head.marker.clear();
+  }
+  cls_rgw_lc_get_head_ret op_ret;
+  op_ret.head = head;
+  encode(op_ret, *out);
+  return 0;
+}
+
+static int rgw_reshard_add(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_reshard_add_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_reshard_add: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+
+  string key;
+  op.entry.get_key(&key);
+
+  bufferlist bl;
+  encode(op.entry, bl);
+  int ret = cls_cxx_map_set_val(hctx, key, &bl);
+  if (ret < 0) {
+    CLS_ERR("error adding reshard job for bucket %s with key %s",op.entry.bucket_name.c_str(), key.c_str());
+    return ret;
+  }
+
+  return ret;
+}
+
+static int rgw_reshard_list(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  cls_rgw_reshard_list_op op;
+  auto in_iter = in->cbegin();
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_rehard_list(): failed to decode entry\n");
+    return -EINVAL;
+  }
+  cls_rgw_reshard_list_ret op_ret;
+  bufferlist::const_iterator iter;
+  map<string, bufferlist> vals;
+  string filter_prefix;
+#define MAX_RESHARD_LIST_ENTRIES 1000
+  /* one extra entry for identifying truncation */
+  int32_t max = (op.max && (op.max < MAX_RESHARD_LIST_ENTRIES) ? op.max : MAX_RESHARD_LIST_ENTRIES);
+  int ret = cls_cxx_map_get_vals(hctx, op.marker, filter_prefix, max, &vals, &op_ret.is_truncated);
+  if (ret < 0)
+    return ret;
+  map<string, bufferlist>::iterator it;
+  cls_rgw_reshard_entry entry;
+  int i = 0;
+  for (it = vals.begin(); i < (int)op.max && it != vals.end(); ++it, ++i) {
+    iter = it->second.cbegin();
+    try {
+      decode(entry, iter);
+    } catch (buffer::error& err) {
+      CLS_LOG(1, "ERROR: rgw_cls_rehard_list(): failed to decode entry\n");
+      return -EIO;
+   }
+    op_ret.entries.push_back(entry);
+  }
+  encode(op_ret, *out);
+  return 0;
+}
+
+static int rgw_reshard_get(cls_method_context_t hctx, bufferlist *in,  bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_reshard_get_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_reshard_get: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  string key;
+  cls_rgw_reshard_entry  entry;
+  op.entry.get_key(&key);
+  int ret = read_omap_entry(hctx, key, &entry);
+  if (ret < 0) {
+    return ret;
+  }
+
+  cls_rgw_reshard_get_ret op_ret;
+  op_ret.entry = entry;
+  encode(op_ret, *out);
+  return 0;
+}
+
+static int rgw_reshard_remove(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_rgw_reshard_remove_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: rgw_cls_rehard_remove: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  string key;
+  cls_rgw_reshard_entry  entry;
+  cls_rgw_reshard_entry::generate_key(op.tenant, op.bucket_name, &key);
+  int ret = read_omap_entry(hctx, key, &entry);
+  if (ret < 0) {
+    return ret;
+  }
+
+  if (!op.bucket_id.empty() &&
+      entry.bucket_id != op.bucket_id) {
+    return 0;
+  }
+
+  ret = cls_cxx_map_remove_key(hctx, key);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: failed to remove key: key=%s ret=%d", key.c_str(), ret);
+    return 0;
+  }
+  return ret;
+}
+
+static int rgw_set_bucket_resharding(cls_method_context_t hctx, bufferlist *in,  bufferlist *out)
+{
+  cls_rgw_set_bucket_resharding_op op;
+
+  auto in_iter = in->cbegin();
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rgw_set_bucket_resharding: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: %s(): failed to read header\n", __func__);
+    return rc;
+  }
+
+  header.new_instance.set_status(op.entry.new_bucket_instance_id, op.entry.num_shards, op.entry.reshard_status);
+
+  return write_bucket_header(hctx, &header);
+}
+
+static int rgw_clear_bucket_resharding(cls_method_context_t hctx, bufferlist *in,  bufferlist *out)
+{
+  cls_rgw_clear_bucket_resharding_op op;
+
+  auto in_iter = in->cbegin();
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rgw_clear_bucket_resharding: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: %s(): failed to read header\n", __func__);
+    return rc;
+  }
+  header.new_instance.clear();
+
+  return write_bucket_header(hctx, &header);
+}
+
+static int rgw_guard_bucket_resharding(cls_method_context_t hctx, bufferlist *in,  bufferlist *out)
+{
+  cls_rgw_guard_bucket_resharding_op op;
+
+  auto in_iter = in->cbegin();
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rgw_clear_bucket_resharding: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: %s(): failed to read header\n", __func__);
+    return rc;
+  }
+
+  if (header.resharding()) {
+    return op.ret_err;
+  }
+
+  return 0;
+}
+
+static int rgw_get_bucket_resharding(cls_method_context_t hctx,
+				     bufferlist *in, bufferlist *out)
+{
+  cls_rgw_get_bucket_resharding_op op;
+
+  auto in_iter = in->cbegin();
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_rgw_clear_bucket_resharding: failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  rgw_bucket_dir_header header;
+  int rc = read_bucket_header(hctx, &header);
+  if (rc < 0) {
+    CLS_LOG(1, "ERROR: %s(): failed to read header\n", __func__);
+    return rc;
+  }
+
+  cls_rgw_get_bucket_resharding_ret op_ret;
+  op_ret.new_instance = header.new_instance;
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+CLS_INIT(rgw)
+{
+  CLS_LOG(1, "Loaded rgw class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_rgw_bucket_init_index;
+  cls_method_handle_t h_rgw_bucket_set_tag_timeout;
+  cls_method_handle_t h_rgw_bucket_list;
+  cls_method_handle_t h_rgw_bucket_check_index;
+  cls_method_handle_t h_rgw_bucket_rebuild_index;
+  cls_method_handle_t h_rgw_bucket_update_stats;
+  cls_method_handle_t h_rgw_bucket_prepare_op;
+  cls_method_handle_t h_rgw_bucket_complete_op;
+  cls_method_handle_t h_rgw_bucket_link_olh;
+  cls_method_handle_t h_rgw_bucket_unlink_instance_op;
+  cls_method_handle_t h_rgw_bucket_read_olh_log;
+  cls_method_handle_t h_rgw_bucket_trim_olh_log;
+  cls_method_handle_t h_rgw_bucket_clear_olh;
+  cls_method_handle_t h_rgw_obj_remove;
+  cls_method_handle_t h_rgw_obj_store_pg_ver;
+  cls_method_handle_t h_rgw_obj_check_attrs_prefix;
+  cls_method_handle_t h_rgw_obj_check_mtime;
+  cls_method_handle_t h_rgw_bi_get_op;
+  cls_method_handle_t h_rgw_bi_put_op;
+  cls_method_handle_t h_rgw_bi_list_op;
+  cls_method_handle_t h_rgw_bi_log_list_op;
+  cls_method_handle_t h_rgw_bi_log_resync_op;
+  cls_method_handle_t h_rgw_bi_log_stop_op;
+  cls_method_handle_t h_rgw_dir_suggest_changes;
+  cls_method_handle_t h_rgw_user_usage_log_add;
+  cls_method_handle_t h_rgw_user_usage_log_read;
+  cls_method_handle_t h_rgw_user_usage_log_trim;
+  cls_method_handle_t h_rgw_usage_log_clear;
+  cls_method_handle_t h_rgw_gc_set_entry;
+  cls_method_handle_t h_rgw_gc_list;
+  cls_method_handle_t h_rgw_gc_remove;
+  cls_method_handle_t h_rgw_lc_get_entry;
+  cls_method_handle_t h_rgw_lc_set_entry;
+  cls_method_handle_t h_rgw_lc_rm_entry;
+  cls_method_handle_t h_rgw_lc_get_next_entry;
+  cls_method_handle_t h_rgw_lc_put_head;
+  cls_method_handle_t h_rgw_lc_get_head;
+  cls_method_handle_t h_rgw_lc_list_entries;
+  cls_method_handle_t h_rgw_reshard_add;
+  cls_method_handle_t h_rgw_reshard_list;
+  cls_method_handle_t h_rgw_reshard_get;
+  cls_method_handle_t h_rgw_reshard_remove;
+  cls_method_handle_t h_rgw_set_bucket_resharding;
+  cls_method_handle_t h_rgw_clear_bucket_resharding;
+  cls_method_handle_t h_rgw_guard_bucket_resharding;
+  cls_method_handle_t h_rgw_get_bucket_resharding;
+
+
+  cls_register(RGW_CLASS, &h_class);
+
+  /* bucket index */
+  cls_register_cxx_method(h_class, RGW_BUCKET_INIT_INDEX, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_init_index, &h_rgw_bucket_init_index);
+  cls_register_cxx_method(h_class, RGW_BUCKET_SET_TAG_TIMEOUT, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_set_tag_timeout, &h_rgw_bucket_set_tag_timeout);
+  cls_register_cxx_method(h_class, RGW_BUCKET_LIST, CLS_METHOD_RD, rgw_bucket_list, &h_rgw_bucket_list);
+  cls_register_cxx_method(h_class, RGW_BUCKET_CHECK_INDEX, CLS_METHOD_RD, rgw_bucket_check_index, &h_rgw_bucket_check_index);
+  cls_register_cxx_method(h_class, RGW_BUCKET_REBUILD_INDEX, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_rebuild_index, &h_rgw_bucket_rebuild_index);
+  cls_register_cxx_method(h_class, RGW_BUCKET_UPDATE_STATS, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_update_stats, &h_rgw_bucket_update_stats);
+  cls_register_cxx_method(h_class, RGW_BUCKET_PREPARE_OP, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_prepare_op, &h_rgw_bucket_prepare_op);
+  cls_register_cxx_method(h_class, RGW_BUCKET_COMPLETE_OP, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_complete_op, &h_rgw_bucket_complete_op);
+  cls_register_cxx_method(h_class, RGW_BUCKET_LINK_OLH, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_link_olh, &h_rgw_bucket_link_olh);
+  cls_register_cxx_method(h_class, RGW_BUCKET_UNLINK_INSTANCE, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_unlink_instance, &h_rgw_bucket_unlink_instance_op);
+  cls_register_cxx_method(h_class, RGW_BUCKET_READ_OLH_LOG, CLS_METHOD_RD, rgw_bucket_read_olh_log, &h_rgw_bucket_read_olh_log);
+  cls_register_cxx_method(h_class, RGW_BUCKET_TRIM_OLH_LOG, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_trim_olh_log, &h_rgw_bucket_trim_olh_log);
+  cls_register_cxx_method(h_class, RGW_BUCKET_CLEAR_OLH, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bucket_clear_olh, &h_rgw_bucket_clear_olh);
+
+  cls_register_cxx_method(h_class, RGW_OBJ_REMOVE, CLS_METHOD_RD | CLS_METHOD_WR, rgw_obj_remove, &h_rgw_obj_remove);
+  cls_register_cxx_method(h_class, RGW_OBJ_STORE_PG_VER, CLS_METHOD_WR, rgw_obj_store_pg_ver, &h_rgw_obj_store_pg_ver);
+  cls_register_cxx_method(h_class, RGW_OBJ_CHECK_ATTRS_PREFIX, CLS_METHOD_RD, rgw_obj_check_attrs_prefix, &h_rgw_obj_check_attrs_prefix);
+  cls_register_cxx_method(h_class, RGW_OBJ_CHECK_MTIME, CLS_METHOD_RD, rgw_obj_check_mtime, &h_rgw_obj_check_mtime);
+
+  cls_register_cxx_method(h_class, RGW_BI_GET, CLS_METHOD_RD, rgw_bi_get_op, &h_rgw_bi_get_op);
+  cls_register_cxx_method(h_class, RGW_BI_PUT, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bi_put_op, &h_rgw_bi_put_op);
+  cls_register_cxx_method(h_class, RGW_BI_LIST, CLS_METHOD_RD, rgw_bi_list_op, &h_rgw_bi_list_op);
+
+  cls_register_cxx_method(h_class, RGW_BI_LOG_LIST, CLS_METHOD_RD, rgw_bi_log_list, &h_rgw_bi_log_list_op);
+  cls_register_cxx_method(h_class, RGW_BI_LOG_TRIM, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bi_log_trim, &h_rgw_bi_log_list_op);
+  cls_register_cxx_method(h_class, RGW_DIR_SUGGEST_CHANGES, CLS_METHOD_RD | CLS_METHOD_WR, rgw_dir_suggest_changes, &h_rgw_dir_suggest_changes);
+
+  cls_register_cxx_method(h_class, RGW_BI_LOG_RESYNC, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bi_log_resync, &h_rgw_bi_log_resync_op);
+  cls_register_cxx_method(h_class, RGW_BI_LOG_STOP, CLS_METHOD_RD | CLS_METHOD_WR, rgw_bi_log_stop, &h_rgw_bi_log_stop_op);
+
+  /* usage logging */
+  cls_register_cxx_method(h_class, RGW_USER_USAGE_LOG_ADD, CLS_METHOD_RD | CLS_METHOD_WR, rgw_user_usage_log_add, &h_rgw_user_usage_log_add);
+  cls_register_cxx_method(h_class, RGW_USER_USAGE_LOG_READ, CLS_METHOD_RD, rgw_user_usage_log_read, &h_rgw_user_usage_log_read);
+  cls_register_cxx_method(h_class, RGW_USER_USAGE_LOG_TRIM, CLS_METHOD_RD | CLS_METHOD_WR, rgw_user_usage_log_trim, &h_rgw_user_usage_log_trim);
+  cls_register_cxx_method(h_class, RGW_USAGE_LOG_CLEAR, CLS_METHOD_WR, rgw_usage_log_clear, &h_rgw_usage_log_clear);
+
+  /* garbage collection */
+  cls_register_cxx_method(h_class, RGW_GC_SET_ENTRY, CLS_METHOD_RD | CLS_METHOD_WR, rgw_cls_gc_set_entry, &h_rgw_gc_set_entry);
+  cls_register_cxx_method(h_class, RGW_GC_DEFER_ENTRY, CLS_METHOD_RD | CLS_METHOD_WR, rgw_cls_gc_defer_entry, &h_rgw_gc_set_entry);
+  cls_register_cxx_method(h_class, RGW_GC_LIST, CLS_METHOD_RD, rgw_cls_gc_list, &h_rgw_gc_list);
+  cls_register_cxx_method(h_class, RGW_GC_REMOVE, CLS_METHOD_RD | CLS_METHOD_WR, rgw_cls_gc_remove, &h_rgw_gc_remove);
+
+  /* lifecycle bucket list */
+  cls_register_cxx_method(h_class, RGW_LC_GET_ENTRY, CLS_METHOD_RD, rgw_cls_lc_get_entry, &h_rgw_lc_get_entry);
+  cls_register_cxx_method(h_class, RGW_LC_SET_ENTRY, CLS_METHOD_RD | CLS_METHOD_WR, rgw_cls_lc_set_entry, &h_rgw_lc_set_entry);
+  cls_register_cxx_method(h_class, RGW_LC_RM_ENTRY, CLS_METHOD_RD | CLS_METHOD_WR, rgw_cls_lc_rm_entry, &h_rgw_lc_rm_entry);
+  cls_register_cxx_method(h_class, RGW_LC_GET_NEXT_ENTRY, CLS_METHOD_RD, rgw_cls_lc_get_next_entry, &h_rgw_lc_get_next_entry);
+  cls_register_cxx_method(h_class, RGW_LC_PUT_HEAD, CLS_METHOD_RD| CLS_METHOD_WR, rgw_cls_lc_put_head, &h_rgw_lc_put_head);
+  cls_register_cxx_method(h_class, RGW_LC_GET_HEAD, CLS_METHOD_RD, rgw_cls_lc_get_head, &h_rgw_lc_get_head);
+  cls_register_cxx_method(h_class, RGW_LC_LIST_ENTRIES, CLS_METHOD_RD, rgw_cls_lc_list_entries, &h_rgw_lc_list_entries);
+
+  /* resharding */
+  cls_register_cxx_method(h_class, RGW_RESHARD_ADD, CLS_METHOD_RD | CLS_METHOD_WR, rgw_reshard_add, &h_rgw_reshard_add);
+  cls_register_cxx_method(h_class, RGW_RESHARD_LIST, CLS_METHOD_RD, rgw_reshard_list, &h_rgw_reshard_list);
+  cls_register_cxx_method(h_class, RGW_RESHARD_GET, CLS_METHOD_RD,rgw_reshard_get, &h_rgw_reshard_get);
+  cls_register_cxx_method(h_class, RGW_RESHARD_REMOVE, CLS_METHOD_RD | CLS_METHOD_WR, rgw_reshard_remove, &h_rgw_reshard_remove);
+
+  /* resharding attribute  */
+  cls_register_cxx_method(h_class, RGW_SET_BUCKET_RESHARDING, CLS_METHOD_RD | CLS_METHOD_WR,
+			  rgw_set_bucket_resharding, &h_rgw_set_bucket_resharding);
+  cls_register_cxx_method(h_class, RGW_CLEAR_BUCKET_RESHARDING, CLS_METHOD_RD | CLS_METHOD_WR,
+			  rgw_clear_bucket_resharding, &h_rgw_clear_bucket_resharding);
+  cls_register_cxx_method(h_class, RGW_GUARD_BUCKET_RESHARDING, CLS_METHOD_RD ,
+			  rgw_guard_bucket_resharding, &h_rgw_guard_bucket_resharding);
+  cls_register_cxx_method(h_class, RGW_GET_BUCKET_RESHARDING, CLS_METHOD_RD ,
+			  rgw_get_bucket_resharding, &h_rgw_get_bucket_resharding);
+
+  return;
+}
+
diff --git a/src/cls/rgw/cls_rgw_client.cc b/src/cls/rgw/cls_rgw_client.cc
new file mode 100644
index 00000000..97426ed7
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_client.cc
@@ -0,0 +1,1018 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "cls/rgw/cls_rgw_const.h"
+#include "cls/rgw/cls_rgw_client.h"
+
+#include "common/debug.h"
+
+using namespace librados;
+
+const string BucketIndexShardsManager::KEY_VALUE_SEPARATOR = "#";
+const string BucketIndexShardsManager::SHARDS_SEPARATOR = ",";
+
+/**
+ * This class represents the bucket index object operation callback context.
+ */
+template <typename T>
+class ClsBucketIndexOpCtx : public ObjectOperationCompletion {
+private:
+  T *data;
+  int *ret_code;
+public:
+  ClsBucketIndexOpCtx(T* _data, int *_ret_code) : data(_data), ret_code(_ret_code) { ceph_assert(data); }
+  ~ClsBucketIndexOpCtx() override {}
+  void handle_completion(int r, bufferlist& outbl) override {
+    if (r >= 0) {
+      try {
+        auto iter = outbl.cbegin();
+        decode((*data), iter);
+      } catch (buffer::error& err) {
+        r = -EIO;
+      }
+    }
+    if (ret_code) {
+      *ret_code = r;
+    }
+  }
+};
+
+void BucketIndexAioManager::do_completion(int id) {
+  Mutex::Locker l(lock);
+
+  map<int, librados::AioCompletion*>::iterator iter = pendings.find(id);
+  ceph_assert(iter != pendings.end());
+  completions[id] = iter->second;
+  pendings.erase(iter);
+
+  // If the caller needs a list of finished objects, store them
+  // for further processing
+  map<int, string>::iterator miter = pending_objs.find(id);
+  if (miter != pending_objs.end()) {
+    completion_objs[id] = miter->second;
+    pending_objs.erase(miter);
+  }
+
+  cond.Signal();
+}
+
+bool BucketIndexAioManager::wait_for_completions(int valid_ret_code,
+    int *num_completions, int *ret_code, map<int, string> *objs) {
+  lock.Lock();
+  if (pendings.empty() && completions.empty()) {
+    lock.Unlock();
+    return false;
+  }
+
+  if (completions.empty()) {
+    // Wait for AIO completion
+    cond.Wait(lock);
+  }
+
+  // Clear the completed AIOs
+  map<int, librados::AioCompletion*>::iterator iter = completions.begin();
+  for (; iter != completions.end(); ++iter) {
+    int r = iter->second->get_return_value();
+    if (objs && r == 0) { /* update list of successfully completed objs */
+      map<int, string>::iterator liter = completion_objs.find(iter->first);
+      if (liter != completion_objs.end()) {
+        (*objs)[liter->first] = liter->second;
+      }
+    }
+    if (ret_code && (r < 0 && r != valid_ret_code))
+      (*ret_code) = r;
+    iter->second->release();
+  }
+  if (num_completions)
+    (*num_completions) = completions.size();
+  completions.clear();
+  lock.Unlock();
+
+  return true;
+}
+
+// note: currently only called by tesing code
+void cls_rgw_bucket_init_index(ObjectWriteOperation& o)
+{
+  bufferlist in;
+  o.exec(RGW_CLASS, RGW_BUCKET_INIT_INDEX, in);
+}
+
+static bool issue_bucket_index_init_op(librados::IoCtx& io_ctx,
+				       const string& oid,
+				       BucketIndexAioManager *manager) {
+  bufferlist in;
+  librados::ObjectWriteOperation op;
+  op.create(true);
+  op.exec(RGW_CLASS, RGW_BUCKET_INIT_INDEX, in);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+static bool issue_bucket_index_clean_op(librados::IoCtx& io_ctx,
+					const string& oid,
+					BucketIndexAioManager *manager) {
+  bufferlist in;
+  librados::ObjectWriteOperation op;
+  op.remove();
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+static bool issue_bucket_set_tag_timeout_op(librados::IoCtx& io_ctx,
+    const string& oid, uint64_t timeout, BucketIndexAioManager *manager) {
+  bufferlist in;
+  rgw_cls_tag_timeout_op call;
+  call.tag_timeout = timeout;
+  encode(call, in);
+  ObjectWriteOperation op;
+  op.exec(RGW_CLASS, RGW_BUCKET_SET_TAG_TIMEOUT, in);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueBucketIndexInit::issue_op(int shard_id, const string& oid)
+{
+  return issue_bucket_index_init_op(io_ctx, oid, &manager);
+}
+
+void CLSRGWIssueBucketIndexInit::cleanup()
+{
+  // Do best effort removal
+  for (auto citer = objs_container.begin(); citer != iter; ++citer) {
+    io_ctx.remove(citer->second);
+  }
+}
+
+int CLSRGWIssueBucketIndexClean::issue_op(int shard_id, const string& oid)
+{
+  return issue_bucket_index_clean_op(io_ctx, oid, &manager);
+}
+
+int CLSRGWIssueSetTagTimeout::issue_op(int shard_id, const string& oid)
+{
+  return issue_bucket_set_tag_timeout_op(io_ctx, oid, tag_timeout, &manager);
+}
+
+void cls_rgw_bucket_update_stats(librados::ObjectWriteOperation& o,
+				 bool absolute,
+                                 const map<RGWObjCategory, rgw_bucket_category_stats>& stats)
+{
+  rgw_cls_bucket_update_stats_op call;
+  call.absolute = absolute;
+  call.stats = stats;
+  bufferlist in;
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_BUCKET_UPDATE_STATS, in);
+}
+
+void cls_rgw_bucket_prepare_op(ObjectWriteOperation& o, RGWModifyOp op, string& tag,
+                               const cls_rgw_obj_key& key, const string& locator, bool log_op,
+                               uint16_t bilog_flags, rgw_zone_set& zones_trace)
+{
+  rgw_cls_obj_prepare_op call;
+  call.op = op;
+  call.tag = tag;
+  call.key = key;
+  call.locator = locator;
+  call.log_op = log_op;
+  call.bilog_flags = bilog_flags;
+  call.zones_trace = zones_trace;
+  bufferlist in;
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_BUCKET_PREPARE_OP, in);
+}
+
+void cls_rgw_bucket_complete_op(ObjectWriteOperation& o, RGWModifyOp op, string& tag,
+                                rgw_bucket_entry_ver& ver,
+                                const cls_rgw_obj_key& key,
+                                rgw_bucket_dir_entry_meta& dir_meta,
+				list<cls_rgw_obj_key> *remove_objs, bool log_op,
+                                uint16_t bilog_flags,
+                                rgw_zone_set *zones_trace)
+{
+
+  bufferlist in;
+  rgw_cls_obj_complete_op call;
+  call.op = op;
+  call.tag = tag;
+  call.key = key;
+  call.ver = ver;
+  call.meta = dir_meta;
+  call.log_op = log_op;
+  call.bilog_flags = bilog_flags;
+  if (remove_objs)
+    call.remove_objs = *remove_objs;
+  if (zones_trace) {
+    call.zones_trace = *zones_trace;
+  }
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_BUCKET_COMPLETE_OP, in);
+}
+
+void cls_rgw_bucket_list_op(librados::ObjectReadOperation& op,
+                            const cls_rgw_obj_key& start_obj,
+                            const std::string& filter_prefix,
+                            uint32_t num_entries,
+                            bool list_versions,
+                            rgw_cls_list_ret* result)
+{
+  bufferlist in;
+  rgw_cls_list_op call;
+  call.start_obj = start_obj;
+  call.filter_prefix = filter_prefix;
+  call.num_entries = num_entries;
+  call.list_versions = list_versions;
+  encode(call, in);
+
+  op.exec(RGW_CLASS, RGW_BUCKET_LIST, in, new ClsBucketIndexOpCtx<rgw_cls_list_ret>(result, NULL));
+}
+
+static bool issue_bucket_list_op(librados::IoCtx& io_ctx, const string& oid,
+				 const cls_rgw_obj_key& start_obj,
+				 const string& filter_prefix,
+				 uint32_t num_entries, bool list_versions,
+				 BucketIndexAioManager *manager,
+				 rgw_cls_list_ret *pdata) {
+  librados::ObjectReadOperation op;
+  cls_rgw_bucket_list_op(op, start_obj, filter_prefix,
+                         num_entries, list_versions, pdata);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueBucketList::issue_op(int shard_id, const string& oid)
+{
+  return issue_bucket_list_op(io_ctx, oid, start_obj, filter_prefix, num_entries, list_versions, &manager, &result[shard_id]);
+}
+
+void cls_rgw_remove_obj(librados::ObjectWriteOperation& o, list<string>& keep_attr_prefixes)
+{
+  bufferlist in;
+  rgw_cls_obj_remove_op call;
+  call.keep_attr_prefixes = keep_attr_prefixes;
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_OBJ_REMOVE, in);
+}
+
+void cls_rgw_obj_store_pg_ver(librados::ObjectWriteOperation& o, const string& attr)
+{
+  bufferlist in;
+  rgw_cls_obj_store_pg_ver_op call;
+  call.attr = attr;
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_OBJ_STORE_PG_VER, in);
+}
+
+void cls_rgw_obj_check_attrs_prefix(librados::ObjectOperation& o, const string& prefix, bool fail_if_exist)
+{
+  bufferlist in;
+  rgw_cls_obj_check_attrs_prefix call;
+  call.check_prefix = prefix;
+  call.fail_if_exist = fail_if_exist;
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_OBJ_CHECK_ATTRS_PREFIX, in);
+}
+
+void cls_rgw_obj_check_mtime(librados::ObjectOperation& o, const real_time& mtime, bool high_precision_time, RGWCheckMTimeType type)
+{
+  bufferlist in;
+  rgw_cls_obj_check_mtime call;
+  call.mtime = mtime;
+  call.high_precision_time = high_precision_time;
+  call.type = type;
+  encode(call, in);
+  o.exec(RGW_CLASS, RGW_OBJ_CHECK_MTIME, in);
+}
+
+int cls_rgw_bi_get(librados::IoCtx& io_ctx, const string oid,
+                   BIIndexType index_type, cls_rgw_obj_key& key,
+                   rgw_cls_bi_entry *entry)
+{
+  bufferlist in, out;
+  rgw_cls_bi_get_op call;
+  call.key = key;
+  call.type = index_type;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_BI_GET, in, out);
+  if (r < 0)
+    return r;
+
+  rgw_cls_bi_get_ret op_ret;
+  auto iter = out.cbegin();
+  try {
+    decode(op_ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  *entry = op_ret.entry;
+
+  return 0;
+}
+
+int cls_rgw_bi_put(librados::IoCtx& io_ctx, const string oid, rgw_cls_bi_entry& entry)
+{
+  bufferlist in, out;
+  rgw_cls_bi_put_op call;
+  call.entry = entry;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_BI_PUT, in, out);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+void cls_rgw_bi_put(ObjectWriteOperation& op, const string oid, rgw_cls_bi_entry& entry)
+{
+  bufferlist in, out;
+  rgw_cls_bi_put_op call;
+  call.entry = entry;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_BI_PUT, in);
+}
+
+int cls_rgw_bi_list(librados::IoCtx& io_ctx, const string oid,
+                   const string& name, const string& marker, uint32_t max,
+                   list<rgw_cls_bi_entry> *entries, bool *is_truncated)
+{
+  bufferlist in, out;
+  rgw_cls_bi_list_op call;
+  call.name = name;
+  call.marker = marker;
+  call.max = max;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_BI_LIST, in, out);
+  if (r < 0)
+    return r;
+
+  rgw_cls_bi_list_ret op_ret;
+  auto iter = out.cbegin();
+  try {
+    decode(op_ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  entries->swap(op_ret.entries);
+  *is_truncated = op_ret.is_truncated;
+
+  return 0;
+}
+
+int cls_rgw_bucket_link_olh(librados::IoCtx& io_ctx, librados::ObjectWriteOperation& op,
+                            const string& oid, const cls_rgw_obj_key& key, bufferlist& olh_tag,
+                            bool delete_marker, const string& op_tag, rgw_bucket_dir_entry_meta *meta,
+                            uint64_t olh_epoch, ceph::real_time unmod_since, bool high_precision_time, bool log_op, rgw_zone_set& zones_trace)
+{
+  bufferlist in, out;
+  rgw_cls_link_olh_op call;
+  call.key = key;
+  call.olh_tag = string(olh_tag.c_str(), olh_tag.length());
+  call.op_tag = op_tag;
+  call.delete_marker = delete_marker;
+  if (meta) {
+    call.meta = *meta;
+  }
+  call.olh_epoch = olh_epoch;
+  call.log_op = log_op;
+  call.unmod_since = unmod_since;
+  call.high_precision_time = high_precision_time;
+  call.zones_trace = zones_trace;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_BUCKET_LINK_OLH, in);
+  int r = io_ctx.operate(oid, &op);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+int cls_rgw_bucket_unlink_instance(librados::IoCtx& io_ctx, librados::ObjectWriteOperation& op,
+                                   const string& oid,
+                                   const cls_rgw_obj_key& key, const string& op_tag,
+                                   const string& olh_tag, uint64_t olh_epoch, bool log_op, rgw_zone_set& zones_trace)
+{
+  bufferlist in, out;
+  rgw_cls_unlink_instance_op call;
+  call.key = key;
+  call.op_tag = op_tag;
+  call.olh_epoch = olh_epoch;
+  call.olh_tag = olh_tag;
+  call.log_op = log_op;
+  call.zones_trace = zones_trace;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_BUCKET_UNLINK_INSTANCE, in);
+  int r = io_ctx.operate(oid, &op);
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+int cls_rgw_get_olh_log(IoCtx& io_ctx, string& oid, librados::ObjectReadOperation& op, const cls_rgw_obj_key& olh, uint64_t ver_marker,
+                        const string& olh_tag,
+                        map<uint64_t, vector<rgw_bucket_olh_log_entry> > *log, bool *is_truncated)
+{
+  bufferlist in, out;
+  rgw_cls_read_olh_log_op call;
+  call.olh = olh;
+  call.ver_marker = ver_marker;
+  call.olh_tag = olh_tag;
+  encode(call, in);
+  int op_ret;
+  op.exec(RGW_CLASS, RGW_BUCKET_READ_OLH_LOG, in, &out, &op_ret);
+  int r = io_ctx.operate(oid, &op, NULL);
+  if (r < 0) {
+    return r;
+  }
+  if (op_ret < 0) {
+    return op_ret;
+  }
+
+  rgw_cls_read_olh_log_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  if (log) {
+    *log = ret.log;
+  }
+  if (is_truncated) {
+    *is_truncated = ret.is_truncated;
+  }
+
+ return r;
+}
+
+void cls_rgw_trim_olh_log(librados::ObjectWriteOperation& op, const cls_rgw_obj_key& olh, uint64_t ver, const string& olh_tag)
+{
+  bufferlist in;
+  rgw_cls_trim_olh_log_op call;
+  call.olh = olh;
+  call.ver = ver;
+  call.olh_tag = olh_tag;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_BUCKET_TRIM_OLH_LOG, in);
+}
+
+int cls_rgw_clear_olh(IoCtx& io_ctx, librados::ObjectWriteOperation& op, string& oid, const cls_rgw_obj_key& olh, const string& olh_tag)
+{
+  bufferlist in, out;
+  rgw_cls_bucket_clear_olh_op call;
+  call.key = olh;
+  call.olh_tag = olh_tag;
+  encode(call, in);
+  int op_ret;
+  op.exec(RGW_CLASS, RGW_BUCKET_CLEAR_OLH, in, &out, &op_ret);
+  int r = io_ctx.operate(oid, &op);
+  if (r < 0) {
+    return r;
+  }
+  return op_ret;
+}
+
+static bool issue_bi_log_list_op(librados::IoCtx& io_ctx, const string& oid, int shard_id,
+                                 BucketIndexShardsManager& marker_mgr, uint32_t max, BucketIndexAioManager *manager,
+    cls_rgw_bi_log_list_ret *pdata) {
+  bufferlist in;
+  cls_rgw_bi_log_list_op call;
+  call.marker = marker_mgr.get(shard_id, "");
+  call.max = max;
+  encode(call, in);
+
+  librados::ObjectReadOperation op;
+  op.exec(RGW_CLASS, RGW_BI_LOG_LIST, in, new ClsBucketIndexOpCtx<cls_rgw_bi_log_list_ret>(pdata, NULL));
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueBILogList::issue_op(int shard_id, const string& oid)
+{
+  return issue_bi_log_list_op(io_ctx, oid, shard_id, marker_mgr, max, &manager, &result[shard_id]);
+}
+
+static bool issue_bi_log_trim(librados::IoCtx& io_ctx, const string& oid, int shard_id,
+                              BucketIndexShardsManager& start_marker_mgr,
+                              BucketIndexShardsManager& end_marker_mgr, BucketIndexAioManager *manager) {
+  bufferlist in;
+  cls_rgw_bi_log_trim_op call;
+  call.start_marker = start_marker_mgr.get(shard_id, "");
+  call.end_marker = end_marker_mgr.get(shard_id, "");
+  encode(call, in);
+  ObjectWriteOperation op;
+  op.exec(RGW_CLASS, RGW_BI_LOG_TRIM, in);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueBILogTrim::issue_op(int shard_id, const string& oid)
+{
+  return issue_bi_log_trim(io_ctx, oid, shard_id, start_marker_mgr, end_marker_mgr, &manager);
+}
+
+static bool issue_bucket_check_index_op(IoCtx& io_ctx, const string& oid, BucketIndexAioManager *manager,
+    rgw_cls_check_index_ret *pdata) {
+  bufferlist in;
+  librados::ObjectReadOperation op;
+  op.exec(RGW_CLASS, RGW_BUCKET_CHECK_INDEX, in, new ClsBucketIndexOpCtx<rgw_cls_check_index_ret>(
+        pdata, NULL));
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueBucketCheck::issue_op(int shard_id, const string& oid)
+{
+  return issue_bucket_check_index_op(io_ctx, oid, &manager, &result[shard_id]);
+}
+
+static bool issue_bucket_rebuild_index_op(IoCtx& io_ctx, const string& oid,
+    BucketIndexAioManager *manager) {
+  bufferlist in;
+  librados::ObjectWriteOperation op;
+  op.exec(RGW_CLASS, RGW_BUCKET_REBUILD_INDEX, in);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueBucketRebuild::issue_op(int shard_id, const string& oid)
+{
+  return issue_bucket_rebuild_index_op(io_ctx, oid, &manager);
+}
+
+void cls_rgw_encode_suggestion(char op, rgw_bucket_dir_entry& dirent, bufferlist& updates)
+{
+  updates.append(op);
+  encode(dirent, updates);
+}
+
+void cls_rgw_suggest_changes(ObjectWriteOperation& o, bufferlist& updates)
+{
+  o.exec(RGW_CLASS, RGW_DIR_SUGGEST_CHANGES, updates);
+}
+
+int CLSRGWIssueGetDirHeader::issue_op(int shard_id, const string& oid)
+{
+  cls_rgw_obj_key nokey;
+  return issue_bucket_list_op(io_ctx, oid, nokey, "", 0, false, &manager, &result[shard_id]);
+}
+
+static bool issue_resync_bi_log(librados::IoCtx& io_ctx, const string& oid, BucketIndexAioManager *manager)
+{
+  bufferlist in;
+  librados::ObjectWriteOperation op;
+  op.exec("rgw", "bi_log_resync", in);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueResyncBucketBILog::issue_op(int shard_id, const string& oid)
+{
+  return issue_resync_bi_log(io_ctx, oid, &manager);
+}
+
+static bool issue_bi_log_stop(librados::IoCtx& io_ctx, const string& oid, BucketIndexAioManager *manager)
+{
+  bufferlist in;
+  librados::ObjectWriteOperation op;
+  op.exec("rgw", "bi_log_stop", in);
+  return manager->aio_operate(io_ctx, oid, &op); 
+}
+
+int CLSRGWIssueBucketBILogStop::issue_op(int shard_id, const string& oid)
+{
+  return issue_bi_log_stop(io_ctx, oid, &manager);
+}
+
+class GetDirHeaderCompletion : public ObjectOperationCompletion {
+  RGWGetDirHeader_CB *ret_ctx;
+public:
+  explicit GetDirHeaderCompletion(RGWGetDirHeader_CB *_ctx) : ret_ctx(_ctx) {}
+  ~GetDirHeaderCompletion() override {
+    ret_ctx->put();
+  }
+  void handle_completion(int r, bufferlist& outbl) override {
+    rgw_cls_list_ret ret;
+    try {
+      auto iter = outbl.cbegin();
+      decode(ret, iter);
+    } catch (buffer::error& err) {
+      r = -EIO;
+    }
+
+    ret_ctx->handle_response(r, ret.dir.header);
+  }
+};
+
+int cls_rgw_get_dir_header_async(IoCtx& io_ctx, string& oid, RGWGetDirHeader_CB *ctx)
+{
+  bufferlist in, out;
+  rgw_cls_list_op call;
+  call.num_entries = 0;
+  encode(call, in);
+  ObjectReadOperation op;
+  GetDirHeaderCompletion *cb = new GetDirHeaderCompletion(ctx);
+  op.exec(RGW_CLASS, RGW_BUCKET_LIST, in, cb);
+  AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL);
+  int r = io_ctx.aio_operate(oid, c, &op, NULL);
+  c->release();
+  if (r < 0)
+    return r;
+
+  return 0;
+}
+
+int cls_rgw_usage_log_read(IoCtx& io_ctx, const string& oid, const string& user, const string& bucket,
+                           uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries,
+                           string& read_iter, map<rgw_user_bucket, rgw_usage_log_entry>& usage,
+                           bool *is_truncated)
+{
+  if (is_truncated)
+    *is_truncated = false;
+
+  bufferlist in, out;
+  rgw_cls_usage_log_read_op call;
+  call.start_epoch = start_epoch;
+  call.end_epoch = end_epoch;
+  call.owner = user;
+  call.max_entries = max_entries;
+  call.bucket = bucket;
+  call.iter = read_iter;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_USER_USAGE_LOG_READ, in, out);
+  if (r < 0)
+    return r;
+
+  try {
+    rgw_cls_usage_log_read_ret result;
+    auto iter = out.cbegin();
+    decode(result, iter);
+    read_iter = result.next_iter;
+    if (is_truncated)
+      *is_truncated = result.truncated;
+
+    usage = result.usage;
+  } catch (buffer::error& e) {
+    return -EINVAL;
+  }
+
+  return 0;
+}
+
+int cls_rgw_usage_log_trim(IoCtx& io_ctx, const string& oid, const string& user, const string& bucket,
+			   uint64_t start_epoch, uint64_t end_epoch)
+{
+  bufferlist in;
+  rgw_cls_usage_log_trim_op call;
+  call.start_epoch = start_epoch;
+  call.end_epoch = end_epoch;
+  call.user = user;
+  call.bucket = bucket;
+  encode(call, in);
+
+  bool done = false;
+  do {
+    ObjectWriteOperation op;
+    op.exec(RGW_CLASS, RGW_USER_USAGE_LOG_TRIM, in);
+    int r = io_ctx.operate(oid, &op);
+    if (r == -ENODATA)
+      done = true;
+    else if (r < 0)
+      return r;
+  } while (!done);
+
+  return 0;
+}
+
+void cls_rgw_usage_log_clear(ObjectWriteOperation& op)
+{
+  bufferlist in;
+  op.exec(RGW_CLASS, RGW_USAGE_LOG_CLEAR, in);
+}
+
+void cls_rgw_usage_log_add(ObjectWriteOperation& op, rgw_usage_log_info& info)
+{
+  bufferlist in;
+  rgw_cls_usage_log_add_op call;
+  call.info = info;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_USER_USAGE_LOG_ADD, in);
+}
+
+/* garbage collection */
+
+void cls_rgw_gc_set_entry(ObjectWriteOperation& op, uint32_t expiration_secs, cls_rgw_gc_obj_info& info)
+{
+  bufferlist in;
+  cls_rgw_gc_set_entry_op call;
+  call.expiration_secs = expiration_secs;
+  call.info = info;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_GC_SET_ENTRY, in);
+}
+
+void cls_rgw_gc_defer_entry(ObjectWriteOperation& op, uint32_t expiration_secs, const string& tag)
+{
+  bufferlist in;
+  cls_rgw_gc_defer_entry_op call;
+  call.expiration_secs = expiration_secs;
+  call.tag = tag;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_GC_DEFER_ENTRY, in);
+}
+
+int cls_rgw_gc_list(IoCtx& io_ctx, string& oid, string& marker, uint32_t max, bool expired_only,
+                    list<cls_rgw_gc_obj_info>& entries, bool *truncated, string& next_marker)
+{
+  bufferlist in, out;
+  cls_rgw_gc_list_op call;
+  call.marker = marker;
+  call.max = max;
+  call.expired_only = expired_only;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_GC_LIST, in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_gc_list_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  entries.swap(ret.entries);
+
+  if (truncated)
+    *truncated = ret.truncated;
+  next_marker = std::move(ret.next_marker);
+  return r;
+}
+
+void cls_rgw_gc_remove(librados::ObjectWriteOperation& op, const vector<string>& tags)
+{
+  bufferlist in;
+  cls_rgw_gc_remove_op call;
+  call.tags = tags;
+  encode(call, in);
+  op.exec(RGW_CLASS, RGW_GC_REMOVE, in);
+}
+
+int cls_rgw_lc_get_head(IoCtx& io_ctx, const string& oid, cls_rgw_lc_obj_head& head)
+{
+  bufferlist in, out;
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_GET_HEAD, in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_lc_get_head_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+  head = ret.head;
+
+ return r;
+}
+
+int cls_rgw_lc_put_head(IoCtx& io_ctx, const string& oid, cls_rgw_lc_obj_head& head)
+{
+  bufferlist in, out;
+  cls_rgw_lc_put_head_op call;
+  call.head = head;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_PUT_HEAD, in, out);
+  return r;
+}
+
+int cls_rgw_lc_get_next_entry(IoCtx& io_ctx, const string& oid, string& marker, pair<string, int>& entry)
+{
+  bufferlist in, out;
+  cls_rgw_lc_get_next_entry_op call;
+  call.marker = marker;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_GET_NEXT_ENTRY, in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_lc_get_next_entry_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+  entry = ret.entry;
+
+ return r;
+}
+
+int cls_rgw_lc_rm_entry(IoCtx& io_ctx, const string& oid, const pair<string, int>& entry)
+{
+  bufferlist in, out;
+  cls_rgw_lc_rm_entry_op call;
+  call.entry = entry;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_RM_ENTRY, in, out);
+ return r;
+}
+
+int cls_rgw_lc_set_entry(IoCtx& io_ctx, const string& oid, const pair<string, int>& entry)
+{
+  bufferlist in, out;
+  cls_rgw_lc_set_entry_op call;
+  call.entry = entry;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_SET_ENTRY, in, out);
+  return r;
+}
+
+int cls_rgw_lc_get_entry(IoCtx& io_ctx, const string& oid, const std::string& marker, rgw_lc_entry_t& entry)
+{
+  bufferlist in, out;
+  cls_rgw_lc_get_entry_op call{marker};;
+  encode(call, in);
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_GET_ENTRY, in, out);
+
+  if (r < 0) {
+    return r;
+  }
+
+  cls_rgw_lc_get_entry_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  entry = std::move(ret.entry);
+  return r;
+}
+
+int cls_rgw_lc_list(IoCtx& io_ctx, const string& oid,
+                    const string& marker,
+                    uint32_t max_entries,
+                    map<string, int>& entries)
+{
+  bufferlist in, out;
+  cls_rgw_lc_list_entries_op op;
+
+  entries.clear();
+
+  op.marker = marker;
+  op.max_entries = max_entries;
+
+  encode(op, in);
+
+  int r = io_ctx.exec(oid, RGW_CLASS, RGW_LC_LIST_ENTRIES, in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_lc_list_entries_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+  entries.insert(ret.entries.begin(),ret.entries.end());
+
+ return r;
+}
+
+void cls_rgw_reshard_add(librados::ObjectWriteOperation& op, const cls_rgw_reshard_entry& entry)
+{
+  bufferlist in;
+  cls_rgw_reshard_add_op call;
+  call.entry = entry;
+  encode(call, in);
+  op.exec("rgw", "reshard_add", in);
+}
+
+int cls_rgw_reshard_list(librados::IoCtx& io_ctx, const string& oid, string& marker, uint32_t max,
+                         list<cls_rgw_reshard_entry>& entries, bool* is_truncated)
+{
+  bufferlist in, out;
+  cls_rgw_reshard_list_op call;
+  call.marker = marker;
+  call.max = max;
+  encode(call, in);
+  int r = io_ctx.exec(oid, "rgw", "reshard_list", in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_reshard_list_ret op_ret;
+  auto iter = out.cbegin();
+  try {
+    decode(op_ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  entries.swap(op_ret.entries);
+  *is_truncated = op_ret.is_truncated;
+
+  return 0;
+}
+
+int cls_rgw_reshard_get(librados::IoCtx& io_ctx, const string& oid, cls_rgw_reshard_entry& entry)
+{
+  bufferlist in, out;
+  cls_rgw_reshard_get_op call;
+  call.entry = entry;
+  encode(call, in);
+  int r = io_ctx.exec(oid, "rgw", "reshard_get", in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_reshard_get_ret op_ret;
+  auto iter = out.cbegin();
+  try {
+    decode(op_ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  entry = op_ret.entry;
+
+  return 0;
+}
+
+void cls_rgw_reshard_remove(librados::ObjectWriteOperation& op, const cls_rgw_reshard_entry& entry)
+{
+  bufferlist in;
+  cls_rgw_reshard_remove_op call;
+  call.tenant = entry.tenant;
+  call.bucket_name = entry.bucket_name;
+  call.bucket_id = entry.bucket_id;
+  encode(call, in);
+  op.exec("rgw", "reshard_remove", in);
+}
+
+int cls_rgw_set_bucket_resharding(librados::IoCtx& io_ctx, const string& oid,
+				  const cls_rgw_bucket_instance_entry& entry)
+{
+  bufferlist in, out;
+  cls_rgw_set_bucket_resharding_op call;
+  call.entry = entry;
+  encode(call, in);
+  return io_ctx.exec(oid, "rgw", "set_bucket_resharding", in, out);
+}
+
+int cls_rgw_clear_bucket_resharding(librados::IoCtx& io_ctx, const string& oid)
+{
+  bufferlist in, out;
+  cls_rgw_clear_bucket_resharding_op call;
+  encode(call, in);
+  return io_ctx.exec(oid, "rgw", "clear_bucket_resharding", in, out);
+}
+
+int cls_rgw_get_bucket_resharding(librados::IoCtx& io_ctx, const string& oid,
+				  cls_rgw_bucket_instance_entry *entry)
+{
+  bufferlist in, out;
+  cls_rgw_get_bucket_resharding_op call;
+  encode(call, in);
+  int r= io_ctx.exec(oid, "rgw", "get_bucket_resharding", in, out);
+  if (r < 0)
+    return r;
+
+  cls_rgw_get_bucket_resharding_ret op_ret;
+  auto iter = out.cbegin();
+  try {
+    decode(op_ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  *entry = op_ret.new_instance;
+
+  return 0;
+}
+
+void cls_rgw_guard_bucket_resharding(librados::ObjectOperation& op, int ret_err)
+{
+  bufferlist in, out;
+  cls_rgw_guard_bucket_resharding_op call;
+  call.ret_err = ret_err;
+  encode(call, in);
+  op.exec("rgw", "guard_bucket_resharding", in);
+}
+
+static bool issue_set_bucket_resharding(librados::IoCtx& io_ctx, const string& oid,
+                                        const cls_rgw_bucket_instance_entry& entry,
+                                        BucketIndexAioManager *manager) {
+  bufferlist in;
+  cls_rgw_set_bucket_resharding_op call;
+  call.entry = entry;
+  encode(call, in);
+  librados::ObjectWriteOperation op;
+  op.exec("rgw", "set_bucket_resharding", in);
+  return manager->aio_operate(io_ctx, oid, &op);
+}
+
+int CLSRGWIssueSetBucketResharding::issue_op(int shard_id, const string& oid)
+{
+  return issue_set_bucket_resharding(io_ctx, oid, entry, &manager);
+}
diff --git a/src/cls/rgw/cls_rgw_client.h b/src/cls/rgw/cls_rgw_client.h
new file mode 100644
index 00000000..b090fbd7
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_client.h
@@ -0,0 +1,593 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_RGW_CLIENT_H
+#define CEPH_CLS_RGW_CLIENT_H
+
+#include "include/str_list.h"
+#include "include/rados/librados.hpp"
+#include "cls_rgw_ops.h"
+#include "cls_rgw_const.h"
+#include "common/RefCountedObj.h"
+#include "include/compat.h"
+#include "common/ceph_time.h"
+#include "common/Mutex.h"
+#include "common/Cond.h"
+
+// Forward declaration
+class BucketIndexAioManager;
+/*
+ * Bucket index AIO request argument, this is used to pass a argument
+ * to callback.
+ */
+struct BucketIndexAioArg : public RefCountedObject {
+  BucketIndexAioArg(int _id, BucketIndexAioManager* _manager) :
+    id(_id), manager(_manager) {}
+  int id;
+  BucketIndexAioManager* manager;
+};
+
+/*
+ * This class manages AIO completions. This class is not completely thread-safe,
+ * methods like *get_next* is not thread-safe and is expected to be called from
+ * within one thread.
+ */
+class BucketIndexAioManager {
+private:
+  map<int, librados::AioCompletion*> pendings;
+  map<int, librados::AioCompletion*> completions;
+  map<int, string> pending_objs;
+  map<int, string> completion_objs;
+  int next;
+  Mutex lock;
+  Cond cond;
+  /*
+   * Callback implementation for AIO request.
+   */
+  static void bucket_index_op_completion_cb(void* cb, void* arg) {
+    BucketIndexAioArg* cb_arg = (BucketIndexAioArg*) arg;
+    cb_arg->manager->do_completion(cb_arg->id);
+    cb_arg->put();
+  }
+
+  /*
+   * Get next request ID. This method is not thread-safe.
+   *
+   * Return next request ID.
+   */
+  int get_next() { return next++; }
+    
+  /*
+   * Add a new pending AIO completion instance.
+   *
+   * @param id         - the request ID.
+   * @param completion - the AIO completion instance.
+   * @param oid        - the object id associated with the object, if it is NULL, we don't
+   *                     track the object id per callback.
+   */
+  void add_pending(int id, librados::AioCompletion* completion, const string& oid) {
+    pendings[id] = completion;
+    pending_objs[id] = oid;
+  }
+public:
+  /*
+   * Create a new instance.
+   */
+  BucketIndexAioManager() : next(0), lock("BucketIndexAioManager::lock") {}
+
+
+  /*
+   * Do completion for the given AIO request.
+   */
+  void do_completion(int id);
+
+  /*
+   * Wait for AIO completions.
+   *
+   * valid_ret_code  - valid AIO return code.
+   * num_completions - number of completions.
+   * ret_code        - return code of failed AIO.
+   * objs            - a list of objects that has been finished the AIO.
+   *
+   * Return false if there is no pending AIO, true otherwise.
+   */
+  bool wait_for_completions(int valid_ret_code, int *num_completions, int *ret_code,
+      map<int, string> *objs);
+
+  /**
+   * Do aio read operation.
+   */
+  bool aio_operate(librados::IoCtx& io_ctx, const string& oid, librados::ObjectReadOperation *op) {
+    Mutex::Locker l(lock);
+    BucketIndexAioArg *arg = new BucketIndexAioArg(get_next(), this);
+    librados::AioCompletion *c = librados::Rados::aio_create_completion((void*)arg, NULL, bucket_index_op_completion_cb);
+    int r = io_ctx.aio_operate(oid, c, (librados::ObjectReadOperation*)op, NULL);
+    if (r >= 0) {
+      add_pending(arg->id, c, oid);
+    } else {
+      c->release();
+    }
+    return r;
+  }
+
+  /**
+   * Do aio write operation.
+   */
+  bool aio_operate(librados::IoCtx& io_ctx, const string& oid, librados::ObjectWriteOperation *op) {
+    Mutex::Locker l(lock);
+    BucketIndexAioArg *arg = new BucketIndexAioArg(get_next(), this);
+    librados::AioCompletion *c = librados::Rados::aio_create_completion((void*)arg, NULL, bucket_index_op_completion_cb);
+    int r = io_ctx.aio_operate(oid, c, (librados::ObjectWriteOperation*)op);
+    if (r >= 0) {
+      add_pending(arg->id, c, oid);
+    } else {
+      c->release();
+    }
+    return r;
+  }
+};
+
+class RGWGetDirHeader_CB : public RefCountedObject {
+public:
+  ~RGWGetDirHeader_CB() override {}
+  virtual void handle_response(int r, rgw_bucket_dir_header& header) = 0;
+};
+
+class BucketIndexShardsManager {
+private:
+  // Per shard setting manager, for example, marker.
+  map<int, string> value_by_shards;
+public:
+  const static string KEY_VALUE_SEPARATOR;
+  const static string SHARDS_SEPARATOR;
+
+  void add(int shard, const string& value) {
+    value_by_shards[shard] = value;
+  }
+
+  const string& get(int shard, const string& default_value) {
+    map<int, string>::iterator iter = value_by_shards.find(shard);
+    return (iter == value_by_shards.end() ? default_value : iter->second);
+  }
+
+  map<int, string>& get() {
+    return value_by_shards;
+  }
+
+  bool empty() {
+    return value_by_shards.empty();
+  }
+
+  void to_string(string *out) const {
+    if (!out) {
+      return;
+    }
+    out->clear();
+    map<int, string>::const_iterator iter = value_by_shards.begin();
+    for (; iter != value_by_shards.end(); ++iter) {
+      if (out->length()) {
+        // Not the first item, append a separator first
+        out->append(SHARDS_SEPARATOR);
+      }
+      char buf[16];
+      snprintf(buf, sizeof(buf), "%d", iter->first);
+      out->append(buf);
+      out->append(KEY_VALUE_SEPARATOR);
+      out->append(iter->second);
+    }
+  }
+
+  static bool is_shards_marker(const string& marker) {
+    return marker.find(KEY_VALUE_SEPARATOR) != string::npos;
+  }
+
+  /*
+   * convert from string. There are two options of how the string looks like:
+   *
+   * 1. Single shard, no shard id specified, e.g. 000001.23.1
+   *
+   * for this case, if passed shard_id >= 0, use this shard id, otherwise assume that it's a
+   * bucket with no shards.
+   *
+   * 2. One or more shards, shard id specified for each shard, e.g., 0#00002.12,1#00003.23.2
+   *
+   */
+  int from_string(const string& composed_marker, int shard_id) {
+    value_by_shards.clear();
+    vector<string> shards;
+    get_str_vec(composed_marker, SHARDS_SEPARATOR.c_str(), shards);
+    if (shards.size() > 1 && shard_id >= 0) {
+      return -EINVAL;
+    }
+    vector<string>::const_iterator iter = shards.begin();
+    for (; iter != shards.end(); ++iter) {
+      size_t pos = iter->find(KEY_VALUE_SEPARATOR);
+      if (pos == string::npos) {
+        if (!value_by_shards.empty()) {
+          return -EINVAL;
+        }
+        if (shard_id < 0) {
+          add(0, *iter);
+        } else {
+          add(shard_id, *iter);
+        }
+        return 0;
+      }
+      string shard_str = iter->substr(0, pos);
+      string err;
+      int shard = (int)strict_strtol(shard_str.c_str(), 10, &err);
+      if (!err.empty()) {
+        return -EINVAL;
+      }
+      add(shard, iter->substr(pos + 1));
+    }
+    return 0;
+  }
+
+  // trim the '<shard-id>#' prefix from a single shard marker if present
+  static std::string get_shard_marker(const std::string& marker) {
+    auto p = marker.find(KEY_VALUE_SEPARATOR);
+    if (p == marker.npos) {
+      return marker;
+    }
+    return marker.substr(p + 1);
+  }
+};
+
+/* bucket index */
+void cls_rgw_bucket_init_index(librados::ObjectWriteOperation& o);
+
+class CLSRGWConcurrentIO {
+protected:
+  librados::IoCtx& io_ctx;
+  map<int, string>& objs_container;
+  map<int, string>::iterator iter;
+  uint32_t max_aio;
+  BucketIndexAioManager manager;
+
+  virtual int issue_op(int shard_id, const string& oid) = 0;
+
+  virtual void cleanup() {}
+  virtual int valid_ret_code() { return 0; }
+  // Return true if multiple rounds of OPs might be needed, this happens when
+  // OP needs to be re-send until a certain code is returned.
+  virtual bool need_multiple_rounds() { return false; }
+  // Add a new object to the end of the container.
+  virtual void add_object(int shard, const string& oid) {}
+  virtual void reset_container(map<int, string>& objs) {}
+
+public:
+
+  CLSRGWConcurrentIO(librados::IoCtx& ioc,
+		     map<int, string>& _objs_container,
+		     uint32_t _max_aio) :
+  io_ctx(ioc), objs_container(_objs_container), max_aio(_max_aio)
+  {}
+
+  virtual ~CLSRGWConcurrentIO()
+  {}
+
+  int operator()() {
+    int ret = 0;
+    iter = objs_container.begin();
+    for (; iter != objs_container.end() && max_aio-- > 0; ++iter) {
+      ret = issue_op(iter->first, iter->second);
+      if (ret < 0)
+        break;
+    }
+
+    int num_completions = 0, r = 0;
+    map<int, string> objs;
+    map<int, string> *pobjs = (need_multiple_rounds() ? &objs : NULL);
+    while (manager.wait_for_completions(valid_ret_code(), &num_completions, &r, pobjs)) {
+      if (r >= 0 && ret >= 0) {
+        for (; num_completions && iter != objs_container.end(); --num_completions, ++iter) {
+          int issue_ret = issue_op(iter->first, iter->second);
+          if (issue_ret < 0) {
+            ret = issue_ret;
+            break;
+          }
+        }
+      } else if (ret >= 0) {
+        ret = r;
+      }
+      if (need_multiple_rounds() && iter == objs_container.end() && !objs.empty()) {
+        // For those objects which need another round, use them to reset
+        // the container
+        reset_container(objs);
+        iter = objs_container.begin();
+        for (; num_completions && iter != objs_container.end(); --num_completions, ++iter) {
+          int issue_ret = issue_op(iter->first, iter->second);
+          if (issue_ret < 0) {
+            ret = issue_ret;
+            break;
+          }
+        }
+      }
+    }
+
+    if (ret < 0) {
+      cleanup();
+    }
+    return ret;
+  }
+};
+
+class CLSRGWIssueBucketIndexInit : public CLSRGWConcurrentIO {
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+  int valid_ret_code() override { return -EEXIST; }
+  void cleanup() override;
+public:
+  CLSRGWIssueBucketIndexInit(librados::IoCtx& ioc, map<int, string>& _bucket_objs,
+                     uint32_t _max_aio) :
+    CLSRGWConcurrentIO(ioc, _bucket_objs, _max_aio) {}
+};
+
+
+class CLSRGWIssueBucketIndexClean : public CLSRGWConcurrentIO {
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+  int valid_ret_code() override {
+    return -ENOENT;
+  }
+
+public:
+  CLSRGWIssueBucketIndexClean(librados::IoCtx& ioc,
+			      map<int, string>& _bucket_objs,
+			      uint32_t _max_aio) :
+  CLSRGWConcurrentIO(ioc, _bucket_objs, _max_aio)
+  {}
+};
+
+
+class CLSRGWIssueSetTagTimeout : public CLSRGWConcurrentIO {
+  uint64_t tag_timeout;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueSetTagTimeout(librados::IoCtx& ioc, map<int, string>& _bucket_objs,
+                     uint32_t _max_aio, uint64_t _tag_timeout) :
+    CLSRGWConcurrentIO(ioc, _bucket_objs, _max_aio), tag_timeout(_tag_timeout) {}
+};
+
+void cls_rgw_bucket_update_stats(librados::ObjectWriteOperation& o,
+                                 bool absolute,
+                                 const map<RGWObjCategory, rgw_bucket_category_stats>& stats);
+
+void cls_rgw_bucket_prepare_op(librados::ObjectWriteOperation& o, RGWModifyOp op, string& tag,
+                               const cls_rgw_obj_key& key, const string& locator, bool log_op,
+                               uint16_t bilog_op, rgw_zone_set& zones_trace);
+
+void cls_rgw_bucket_complete_op(librados::ObjectWriteOperation& o, RGWModifyOp op, string& tag,
+                                rgw_bucket_entry_ver& ver,
+                                const cls_rgw_obj_key& key,
+                                rgw_bucket_dir_entry_meta& dir_meta,
+				list<cls_rgw_obj_key> *remove_objs, bool log_op,
+                                uint16_t bilog_op, rgw_zone_set *zones_trace);
+
+void cls_rgw_remove_obj(librados::ObjectWriteOperation& o, list<string>& keep_attr_prefixes);
+void cls_rgw_obj_store_pg_ver(librados::ObjectWriteOperation& o, const string& attr);
+void cls_rgw_obj_check_attrs_prefix(librados::ObjectOperation& o, const string& prefix, bool fail_if_exist);
+void cls_rgw_obj_check_mtime(librados::ObjectOperation& o, const ceph::real_time& mtime, bool high_precision_time, RGWCheckMTimeType type);
+
+int cls_rgw_bi_get(librados::IoCtx& io_ctx, const string oid,
+                   BIIndexType index_type, cls_rgw_obj_key& key,
+                   rgw_cls_bi_entry *entry);
+int cls_rgw_bi_put(librados::IoCtx& io_ctx, const string oid, rgw_cls_bi_entry& entry);
+void cls_rgw_bi_put(librados::ObjectWriteOperation& op, const string oid, rgw_cls_bi_entry& entry);
+int cls_rgw_bi_list(librados::IoCtx& io_ctx, const string oid,
+                   const string& name, const string& marker, uint32_t max,
+                   list<rgw_cls_bi_entry> *entries, bool *is_truncated);
+
+
+int cls_rgw_bucket_link_olh(librados::IoCtx& io_ctx, librados::ObjectWriteOperation& op,
+                            const string& oid, const cls_rgw_obj_key& key, bufferlist& olh_tag,
+                            bool delete_marker, const string& op_tag, rgw_bucket_dir_entry_meta *meta,
+                            uint64_t olh_epoch, ceph::real_time unmod_since, bool high_precision_time, bool log_op, rgw_zone_set& zones_trace);
+int cls_rgw_bucket_unlink_instance(librados::IoCtx& io_ctx, librados::ObjectWriteOperation& op,
+                                   const string& oid, const cls_rgw_obj_key& key, const string& op_tag,
+                                   const string& olh_tag, uint64_t olh_epoch, bool log_op, rgw_zone_set& zones_trace);
+int cls_rgw_get_olh_log(librados::IoCtx& io_ctx, string& oid, librados::ObjectReadOperation& op, const cls_rgw_obj_key& olh, uint64_t ver_marker,
+                        const string& olh_tag,
+                        map<uint64_t, vector<rgw_bucket_olh_log_entry> > *log, bool *is_truncated);
+void cls_rgw_trim_olh_log(librados::ObjectWriteOperation& op, const cls_rgw_obj_key& olh, uint64_t ver, const string& olh_tag);
+int cls_rgw_clear_olh(librados::IoCtx& io_ctx, librados::ObjectWriteOperation& op, string& oid, const cls_rgw_obj_key& olh, const string& olh_tag);
+
+/**
+ * List the bucket with the starting object and filter prefix.
+ * NOTE: this method do listing requests for each bucket index shards identified by
+ *       the keys of the *list_results* map, which means the map should be popludated
+ *       by the caller to fill with each bucket index object id.
+ *
+ * io_ctx        - IO context for rados.
+ * start_obj     - marker for the listing.
+ * filter_prefix - filter prefix.
+ * num_entries   - number of entries to request for each object (note the total
+ *                 amount of entries returned depends on the number of shardings).
+ * list_results  - the list results keyed by bucket index object id.
+ * max_aio       - the maximum number of AIO (for throttling).
+ *
+ * Return 0 on success, a failure code otherwise.
+*/
+
+class CLSRGWIssueBucketList : public CLSRGWConcurrentIO {
+  cls_rgw_obj_key start_obj;
+  string filter_prefix;
+  uint32_t num_entries;
+  bool list_versions;
+  map<int, rgw_cls_list_ret>& result;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueBucketList(librados::IoCtx& io_ctx, const cls_rgw_obj_key& _start_obj,
+                        const string& _filter_prefix, uint32_t _num_entries,
+                        bool _list_versions,
+                        map<int, string>& oids,
+                        map<int, rgw_cls_list_ret>& list_results,
+                        uint32_t max_aio) :
+  CLSRGWConcurrentIO(io_ctx, oids, max_aio),
+  start_obj(_start_obj), filter_prefix(_filter_prefix), num_entries(_num_entries), list_versions(_list_versions), result(list_results) {}
+};
+
+void cls_rgw_bucket_list_op(librados::ObjectReadOperation& op,
+                            const cls_rgw_obj_key& start_obj,
+                            const std::string& filter_prefix,
+                            uint32_t num_entries,
+                            bool list_versions,
+                            rgw_cls_list_ret* result);
+
+class CLSRGWIssueBILogList : public CLSRGWConcurrentIO {
+  map<int, cls_rgw_bi_log_list_ret>& result;
+  BucketIndexShardsManager& marker_mgr;
+  uint32_t max;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueBILogList(librados::IoCtx& io_ctx, BucketIndexShardsManager& _marker_mgr, uint32_t _max,
+                       map<int, string>& oids,
+                       map<int, cls_rgw_bi_log_list_ret>& bi_log_lists, uint32_t max_aio) :
+    CLSRGWConcurrentIO(io_ctx, oids, max_aio), result(bi_log_lists),
+    marker_mgr(_marker_mgr), max(_max) {}
+};
+
+class CLSRGWIssueBILogTrim : public CLSRGWConcurrentIO {
+  BucketIndexShardsManager& start_marker_mgr;
+  BucketIndexShardsManager& end_marker_mgr;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+  // Trim until -ENODATA is returned.
+  int valid_ret_code() override { return -ENODATA; }
+  bool need_multiple_rounds() override { return true; }
+  void add_object(int shard, const string& oid) override { objs_container[shard] = oid; }
+  void reset_container(map<int, string>& objs) override {
+    objs_container.swap(objs);
+    iter = objs_container.begin();
+    objs.clear();
+  }
+public:
+  CLSRGWIssueBILogTrim(librados::IoCtx& io_ctx, BucketIndexShardsManager& _start_marker_mgr,
+      BucketIndexShardsManager& _end_marker_mgr, map<int, string>& _bucket_objs, uint32_t max_aio) :
+    CLSRGWConcurrentIO(io_ctx, _bucket_objs, max_aio),
+    start_marker_mgr(_start_marker_mgr), end_marker_mgr(_end_marker_mgr) {}
+};
+
+/**
+ * Check the bucket index.
+ *
+ * io_ctx          - IO context for rados.
+ * bucket_objs_ret - check result for all shards.
+ * max_aio         - the maximum number of AIO (for throttling).
+ *
+ * Return 0 on success, a failure code otherwise.
+ */
+class CLSRGWIssueBucketCheck : public CLSRGWConcurrentIO /*<map<string, rgw_cls_check_index_ret> >*/ {
+  map<int, rgw_cls_check_index_ret>& result;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueBucketCheck(librados::IoCtx& ioc, map<int, string>& oids,
+			 map<int, rgw_cls_check_index_ret>& bucket_objs_ret,
+			 uint32_t _max_aio) :
+    CLSRGWConcurrentIO(ioc, oids, _max_aio), result(bucket_objs_ret) {}
+};
+
+class CLSRGWIssueBucketRebuild : public CLSRGWConcurrentIO {
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueBucketRebuild(librados::IoCtx& io_ctx, map<int, string>& bucket_objs,
+                           uint32_t max_aio) : CLSRGWConcurrentIO(io_ctx, bucket_objs, max_aio) {}
+};
+
+class CLSRGWIssueGetDirHeader : public CLSRGWConcurrentIO {
+  map<int, rgw_cls_list_ret>& result;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueGetDirHeader(librados::IoCtx& io_ctx, map<int, string>& oids, map<int, rgw_cls_list_ret>& dir_headers,
+                          uint32_t max_aio) :
+    CLSRGWConcurrentIO(io_ctx, oids, max_aio), result(dir_headers) {}
+};
+
+class CLSRGWIssueSetBucketResharding : public CLSRGWConcurrentIO {
+  cls_rgw_bucket_instance_entry entry;
+protected:
+  int issue_op(int shard_id, const string& oid) override;
+public:
+  CLSRGWIssueSetBucketResharding(librados::IoCtx& ioc, map<int, string>& _bucket_objs,
+                                 const cls_rgw_bucket_instance_entry& _entry,
+                                 uint32_t _max_aio) : CLSRGWConcurrentIO(ioc, _bucket_objs, _max_aio), entry(_entry) {}
+};
+
+class CLSRGWIssueResyncBucketBILog : public CLSRGWConcurrentIO {
+protected:
+  int issue_op(int shard_id, const string& oid);
+public:
+  CLSRGWIssueResyncBucketBILog(librados::IoCtx& io_ctx, map<int, string>& _bucket_objs, uint32_t max_aio) :
+    CLSRGWConcurrentIO(io_ctx, _bucket_objs, max_aio) {}
+};
+
+class CLSRGWIssueBucketBILogStop : public CLSRGWConcurrentIO {
+protected:
+  int issue_op(int shard_id, const string& oid);
+public:
+  CLSRGWIssueBucketBILogStop(librados::IoCtx& io_ctx, map<int, string>& _bucket_objs, uint32_t max_aio) :
+    CLSRGWConcurrentIO(io_ctx, _bucket_objs, max_aio) {}
+};
+
+int cls_rgw_get_dir_header_async(librados::IoCtx& io_ctx, string& oid, RGWGetDirHeader_CB *ctx);
+
+void cls_rgw_encode_suggestion(char op, rgw_bucket_dir_entry& dirent, bufferlist& updates);
+
+void cls_rgw_suggest_changes(librados::ObjectWriteOperation& o, bufferlist& updates);
+
+/* usage logging */
+int cls_rgw_usage_log_read(librados::IoCtx& io_ctx, const string& oid, const string& user, const string& bucket,
+                           uint64_t start_epoch, uint64_t end_epoch, uint32_t max_entries, string& read_iter,
+			   map<rgw_user_bucket, rgw_usage_log_entry>& usage, bool *is_truncated);
+
+int cls_rgw_usage_log_trim(librados::IoCtx& io_ctx, const string& oid, const string& user, const string& bucket,
+                           uint64_t start_epoch, uint64_t end_epoch);
+
+void cls_rgw_usage_log_clear(librados::ObjectWriteOperation& op);
+void cls_rgw_usage_log_add(librados::ObjectWriteOperation& op, rgw_usage_log_info& info);
+
+/* garbage collection */
+void cls_rgw_gc_set_entry(librados::ObjectWriteOperation& op, uint32_t expiration_secs, cls_rgw_gc_obj_info& info);
+void cls_rgw_gc_defer_entry(librados::ObjectWriteOperation& op, uint32_t expiration_secs, const string& tag);
+
+int cls_rgw_gc_list(librados::IoCtx& io_ctx, string& oid, string& marker, uint32_t max, bool expired_only,
+                    list<cls_rgw_gc_obj_info>& entries, bool *truncated, string& next_marker);
+
+void cls_rgw_gc_remove(librados::ObjectWriteOperation& op, const vector<string>& tags);
+
+/* lifecycle */
+int cls_rgw_lc_get_head(librados::IoCtx& io_ctx, const string& oid, cls_rgw_lc_obj_head& head);
+int cls_rgw_lc_put_head(librados::IoCtx& io_ctx, const string& oid, cls_rgw_lc_obj_head& head);
+int cls_rgw_lc_get_next_entry(librados::IoCtx& io_ctx, const string& oid, string& marker, pair<string, int>& entry);
+int cls_rgw_lc_rm_entry(librados::IoCtx& io_ctx, const string& oid, const pair<string, int>& entry);
+int cls_rgw_lc_set_entry(librados::IoCtx& io_ctx, const string& oid, const pair<string, int>& entry);
+int cls_rgw_lc_get_entry(librados::IoCtx& io_ctx, const string& oid, const std::string& marker, rgw_lc_entry_t& entry);
+int cls_rgw_lc_list(librados::IoCtx& io_ctx, const string& oid,
+                    const string& marker,
+                    uint32_t max_entries,
+                    map<string, int>& entries);
+
+/* resharding */
+void cls_rgw_reshard_add(librados::ObjectWriteOperation& op, const cls_rgw_reshard_entry& entry);
+int cls_rgw_reshard_list(librados::IoCtx& io_ctx, const string& oid, string& marker, uint32_t max,
+                         list<cls_rgw_reshard_entry>& entries, bool* is_truncated);
+int cls_rgw_reshard_get(librados::IoCtx& io_ctx, const string& oid, cls_rgw_reshard_entry& entry);
+int cls_rgw_reshard_get_head(librados::IoCtx& io_ctx, const string& oid, cls_rgw_reshard_entry& entry);
+void cls_rgw_reshard_remove(librados::ObjectWriteOperation& op, const cls_rgw_reshard_entry& entry);
+
+/* resharding attribute on bucket index shard headers */
+int cls_rgw_set_bucket_resharding(librados::IoCtx& io_ctx, const string& oid,
+                                  const cls_rgw_bucket_instance_entry& entry);
+int cls_rgw_clear_bucket_resharding(librados::IoCtx& io_ctx, const string& oid);
+void cls_rgw_guard_bucket_resharding(librados::ObjectOperation& op, int ret_err);
+int cls_rgw_get_bucket_resharding(librados::IoCtx& io_ctx, const string& oid,
+                                  cls_rgw_bucket_instance_entry *entry);
+
+#endif
diff --git a/src/cls/rgw/cls_rgw_const.h b/src/cls/rgw/cls_rgw_const.h
new file mode 100644
index 00000000..fc3537ea
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_const.h
@@ -0,0 +1,76 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_RGW_CONST_H
+#define CEPH_CLS_RGW_CONST_H
+
+#define RGW_CLASS "rgw"
+
+/* bucket index */
+#define RGW_BUCKET_INIT_INDEX "bucket_init_index"
+
+
+#define RGW_BUCKET_SET_TAG_TIMEOUT "bucket_set_tag_timeout"
+#define RGW_BUCKET_LIST "bucket_list"
+#define RGW_BUCKET_CHECK_INDEX "bucket_check_index"
+#define RGW_BUCKET_REBUILD_INDEX "bucket_rebuild_index"
+#define RGW_BUCKET_UPDATE_STATS "bucket_update_stats"
+#define RGW_BUCKET_PREPARE_OP "bucket_prepare_op"
+#define RGW_BUCKET_COMPLETE_OP "bucket_complete_op"
+#define RGW_BUCKET_LINK_OLH "bucket_link_olh"
+#define RGW_BUCKET_UNLINK_INSTANCE "bucket_unlink_instance"
+#define RGW_BUCKET_READ_OLH_LOG "bucket_read_olh_log"
+#define RGW_BUCKET_TRIM_OLH_LOG "bucket_trim_olh_log"
+#define RGW_BUCKET_CLEAR_OLH "bucket_clear_olh"
+
+#define RGW_OBJ_REMOVE "obj_remove"
+#define RGW_OBJ_STORE_PG_VER "obj_store_pg_ver"
+#define RGW_OBJ_CHECK_ATTRS_PREFIX "obj_check_attrs_prefix"
+#define RGW_OBJ_CHECK_MTIME "obj_check_mtime"
+
+#define RGW_BI_GET "bi_get"
+#define RGW_BI_PUT "bi_put"
+#define RGW_BI_LIST "bi_list"
+
+#define RGW_BI_LOG_LIST "bi_log_list"
+#define RGW_BI_LOG_TRIM "bi_log_trim"
+#define RGW_DIR_SUGGEST_CHANGES "dir_suggest_changes"
+
+#define RGW_BI_LOG_RESYNC "bi_log_resync"
+#define RGW_BI_LOG_STOP "bi_log_stop"
+
+/* usage logging */
+#define RGW_USER_USAGE_LOG_ADD "user_usage_log_add"
+#define RGW_USER_USAGE_LOG_READ "user_usage_log_read"
+#define RGW_USER_USAGE_LOG_TRIM "user_usage_log_trim"
+#define RGW_USAGE_LOG_CLEAR "usage_log_clear"
+
+/* garbage collection */
+#define RGW_GC_SET_ENTRY "gc_set_entry"
+#define RGW_GC_DEFER_ENTRY "gc_defer_entry"
+#define RGW_GC_LIST "gc_list"
+#define RGW_GC_REMOVE "gc_remove"
+
+/* lifecycle bucket list */
+#define RGW_LC_GET_ENTRY "lc_get_entry"
+#define RGW_LC_SET_ENTRY "lc_set_entry"
+#define RGW_LC_RM_ENTRY "lc_rm_entry"
+#define RGW_LC_GET_NEXT_ENTRY "lc_get_next_entry"
+#define RGW_LC_PUT_HEAD "lc_put_head"
+#define RGW_LC_GET_HEAD "lc_get_head"
+#define RGW_LC_LIST_ENTRIES "lc_list_entries"
+
+/* resharding */
+#define RGW_RESHARD_ADD "reshard_add"
+#define RGW_RESHARD_LIST "reshard_list"
+#define RGW_RESHARD_GET "reshard_get"
+#define RGW_RESHARD_REMOVE "reshard_remove"
+
+/* resharding attribute  */
+#define RGW_SET_BUCKET_RESHARDING "set_bucket_resharding"
+#define RGW_CLEAR_BUCKET_RESHARDING "clear_bucket_resharding"
+#define RGW_GUARD_BUCKET_RESHARDING "guard_bucket_resharding"
+#define RGW_GET_BUCKET_RESHARDING "get_bucket_resharding"
+
+
+#endif
diff --git a/src/cls/rgw/cls_rgw_ops.cc b/src/cls/rgw/cls_rgw_ops.cc
new file mode 100644
index 00000000..713cee41
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_ops.cc
@@ -0,0 +1,548 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/rgw/cls_rgw_ops.h"
+
+#include "common/Formatter.h"
+#include "common/ceph_json.h"
+#include "include/utime.h"
+
+void rgw_cls_tag_timeout_op::dump(Formatter *f) const
+{
+  f->dump_int("tag_timeout", tag_timeout);
+}
+
+void rgw_cls_tag_timeout_op::generate_test_instances(list<rgw_cls_tag_timeout_op*>& ls)
+{
+  ls.push_back(new rgw_cls_tag_timeout_op);
+  ls.push_back(new rgw_cls_tag_timeout_op);
+  ls.back()->tag_timeout = 23323;
+}
+
+void cls_rgw_gc_set_entry_op::dump(Formatter *f) const
+{
+  f->dump_unsigned("expiration_secs", expiration_secs);
+  f->open_object_section("obj_info");
+  info.dump(f);
+  f->close_section();
+}
+
+void cls_rgw_gc_set_entry_op::generate_test_instances(list<cls_rgw_gc_set_entry_op*>& ls)
+{
+  ls.push_back(new cls_rgw_gc_set_entry_op);
+  ls.push_back(new cls_rgw_gc_set_entry_op);
+  ls.back()->expiration_secs = 123;
+}
+
+void cls_rgw_gc_defer_entry_op::dump(Formatter *f) const
+{
+  f->dump_unsigned("expiration_secs", expiration_secs);
+  f->dump_string("tag", tag);
+}
+
+void cls_rgw_gc_defer_entry_op::generate_test_instances(list<cls_rgw_gc_defer_entry_op*>& ls)
+{
+  ls.push_back(new cls_rgw_gc_defer_entry_op);
+  ls.push_back(new cls_rgw_gc_defer_entry_op);
+  ls.back()->expiration_secs = 123;
+  ls.back()->tag = "footag";
+}
+
+void cls_rgw_gc_list_op::dump(Formatter *f) const
+{
+  f->dump_string("marker", marker);
+  f->dump_unsigned("max", max);
+  f->dump_bool("expired_only", expired_only);
+}
+
+void cls_rgw_gc_list_op::generate_test_instances(list<cls_rgw_gc_list_op*>& ls)
+{
+  ls.push_back(new cls_rgw_gc_list_op);
+  ls.push_back(new cls_rgw_gc_list_op);
+  ls.back()->marker = "mymarker";
+  ls.back()->max = 2312;
+}
+
+void cls_rgw_gc_list_ret::dump(Formatter *f) const
+{
+  encode_json("entries", entries, f);
+  f->dump_string("next_marker", next_marker);
+  f->dump_int("truncated", (int)truncated);
+}
+
+void cls_rgw_gc_list_ret::generate_test_instances(list<cls_rgw_gc_list_ret*>& ls)
+{
+  ls.push_back(new cls_rgw_gc_list_ret);
+  ls.push_back(new cls_rgw_gc_list_ret);
+  ls.back()->entries.push_back(cls_rgw_gc_obj_info());
+  ls.back()->truncated = true;
+}
+
+
+void cls_rgw_gc_remove_op::dump(Formatter *f) const
+{
+  encode_json("tags", tags, f);
+}
+
+void cls_rgw_gc_remove_op::generate_test_instances(list<cls_rgw_gc_remove_op*>& ls)
+{
+  ls.push_back(new cls_rgw_gc_remove_op);
+  ls.push_back(new cls_rgw_gc_remove_op);
+  ls.back()->tags.push_back("tag1");
+  ls.back()->tags.push_back("tag2");
+}
+
+void rgw_cls_obj_prepare_op::generate_test_instances(list<rgw_cls_obj_prepare_op*>& o)
+{
+  rgw_cls_obj_prepare_op *op = new rgw_cls_obj_prepare_op;
+  op->op = CLS_RGW_OP_ADD;
+  op->key.name = "name";
+  op->tag = "tag";
+  op->locator = "locator";
+  o.push_back(op);
+  o.push_back(new rgw_cls_obj_prepare_op);
+}
+
+void rgw_cls_obj_prepare_op::dump(Formatter *f) const
+{
+  f->dump_int("op", op);
+  f->dump_string("name", key.name);
+  f->dump_string("tag", tag);
+  f->dump_string("locator", locator);
+  f->dump_bool("log_op", log_op);
+  f->dump_int("bilog_flags", bilog_flags);
+  ::encode_json("zones_trace", zones_trace, f);
+}
+
+void rgw_cls_obj_complete_op::generate_test_instances(list<rgw_cls_obj_complete_op*>& o)
+{
+  rgw_cls_obj_complete_op *op = new rgw_cls_obj_complete_op;
+  op->op = CLS_RGW_OP_DEL;
+  op->key.name = "name";
+  op->locator = "locator";
+  op->ver.pool = 2;
+  op->ver.epoch = 100;
+  op->tag = "tag";
+
+  list<rgw_bucket_dir_entry_meta *> l;
+  rgw_bucket_dir_entry_meta::generate_test_instances(l);
+  list<rgw_bucket_dir_entry_meta *>::iterator iter = l.begin();
+  op->meta = *(*iter);
+
+  o.push_back(op);
+
+  o.push_back(new rgw_cls_obj_complete_op);
+}
+
+void rgw_cls_obj_complete_op::dump(Formatter *f) const
+{
+  f->dump_int("op", (int)op);
+  f->dump_string("name", key.name);
+  f->dump_string("instance", key.instance);
+  f->dump_string("locator", locator);
+  f->open_object_section("ver");
+  ver.dump(f);
+  f->close_section();
+  f->open_object_section("meta");
+  meta.dump(f);
+  f->close_section();
+  f->dump_string("tag", tag);
+  f->dump_bool("log_op", log_op);
+  f->dump_int("bilog_flags", bilog_flags);
+  ::encode_json("zones_trace", zones_trace, f);
+}
+
+void rgw_cls_link_olh_op::generate_test_instances(list<rgw_cls_link_olh_op*>& o)
+{
+  rgw_cls_link_olh_op *op = new rgw_cls_link_olh_op;
+  op->key.name = "name";
+  op->olh_tag = "olh_tag";
+  op->delete_marker = true;
+  op->op_tag = "op_tag";
+  op->olh_epoch = 123;
+  list<rgw_bucket_dir_entry_meta *> l;
+  rgw_bucket_dir_entry_meta::generate_test_instances(l);
+  list<rgw_bucket_dir_entry_meta *>::iterator iter = l.begin();
+  op->meta = *(*iter);
+  op->log_op = true;
+
+  o.push_back(op);
+
+  o.push_back(new rgw_cls_link_olh_op);
+}
+
+void rgw_cls_link_olh_op::dump(Formatter *f) const
+{
+  ::encode_json("key", key, f);
+  ::encode_json("olh_tag", olh_tag, f);
+  ::encode_json("delete_marker", delete_marker, f);
+  ::encode_json("op_tag", op_tag, f);
+  ::encode_json("meta", meta, f);
+  ::encode_json("olh_epoch", olh_epoch, f);
+  ::encode_json("log_op", log_op, f);
+  ::encode_json("bilog_flags", (uint32_t)bilog_flags, f);
+  utime_t ut(unmod_since);
+  ::encode_json("unmod_since", ut, f);
+  ::encode_json("high_precision_time", high_precision_time, f);
+  ::encode_json("zones_trace", zones_trace, f);
+}
+
+void rgw_cls_unlink_instance_op::generate_test_instances(list<rgw_cls_unlink_instance_op*>& o)
+{
+  rgw_cls_unlink_instance_op *op = new rgw_cls_unlink_instance_op;
+  op->key.name = "name";
+  op->op_tag = "op_tag";
+  op->olh_epoch = 124;
+  op->log_op = true;
+
+  o.push_back(op);
+
+  o.push_back(new rgw_cls_unlink_instance_op);
+}
+
+void rgw_cls_unlink_instance_op::dump(Formatter *f) const
+{
+  ::encode_json("key", key, f);
+  ::encode_json("op_tag", op_tag, f);
+  ::encode_json("olh_epoch", olh_epoch, f);
+  ::encode_json("log_op", log_op, f);
+  ::encode_json("bilog_flags", (uint32_t)bilog_flags, f);
+  ::encode_json("zones_trace", zones_trace, f);
+}
+
+void rgw_cls_read_olh_log_op::generate_test_instances(list<rgw_cls_read_olh_log_op*>& o)
+{
+  rgw_cls_read_olh_log_op *op = new rgw_cls_read_olh_log_op;
+  op->olh.name = "name";
+  op->ver_marker = 123;
+  op->olh_tag = "olh_tag";
+
+  o.push_back(op);
+
+  o.push_back(new rgw_cls_read_olh_log_op);
+}
+
+void rgw_cls_read_olh_log_op::dump(Formatter *f) const
+{
+  ::encode_json("olh", olh, f);
+  ::encode_json("ver_marker", ver_marker, f);
+  ::encode_json("olh_tag", olh_tag, f);
+}
+
+void rgw_cls_read_olh_log_ret::generate_test_instances(list<rgw_cls_read_olh_log_ret*>& o)
+{
+  rgw_cls_read_olh_log_ret *r = new rgw_cls_read_olh_log_ret;
+  r->is_truncated = true;
+  list<rgw_bucket_olh_log_entry *> l;
+  rgw_bucket_olh_log_entry::generate_test_instances(l);
+  list<rgw_bucket_olh_log_entry *>::iterator iter = l.begin();
+  r->log[1].push_back(*(*iter));
+
+  o.push_back(r);
+
+  o.push_back(new rgw_cls_read_olh_log_ret);
+}
+
+void rgw_cls_read_olh_log_ret::dump(Formatter *f) const
+{
+  ::encode_json("log", log, f);
+  ::encode_json("is_truncated", is_truncated, f);
+}
+
+void rgw_cls_trim_olh_log_op::generate_test_instances(list<rgw_cls_trim_olh_log_op*>& o)
+{
+  rgw_cls_trim_olh_log_op *op = new rgw_cls_trim_olh_log_op;
+  op->olh.name = "olh.name";
+  op->ver = 100;
+  op->olh_tag = "olh_tag";
+
+  o.push_back(op);
+
+  o.push_back(new rgw_cls_trim_olh_log_op);
+}
+
+void rgw_cls_trim_olh_log_op::dump(Formatter *f) const
+{
+  ::encode_json("olh", olh, f);
+  ::encode_json("ver", ver, f);
+  ::encode_json("olh_tag", olh_tag, f);
+}
+
+void rgw_cls_bucket_clear_olh_op::generate_test_instances(list<rgw_cls_bucket_clear_olh_op *>& o)
+{
+
+  rgw_cls_bucket_clear_olh_op *op = new rgw_cls_bucket_clear_olh_op;
+  op->key.name = "key.name";
+  op->olh_tag = "olh_tag";
+
+  o.push_back(op);
+  o.push_back(new rgw_cls_bucket_clear_olh_op);
+}
+
+void rgw_cls_bucket_clear_olh_op::dump(Formatter *f) const
+{
+  ::encode_json("key", key, f);
+  ::encode_json("olh_tag", olh_tag, f);
+}
+
+void rgw_cls_list_op::generate_test_instances(list<rgw_cls_list_op*>& o)
+{
+  rgw_cls_list_op *op = new rgw_cls_list_op;
+  op->start_obj.name = "start_obj";
+  op->num_entries = 100;
+  op->filter_prefix = "filter_prefix";
+  o.push_back(op);
+  o.push_back(new rgw_cls_list_op);
+}
+
+void rgw_cls_list_op::dump(Formatter *f) const
+{
+  f->dump_string("start_obj", start_obj.name);
+  f->dump_unsigned("num_entries", num_entries);
+}
+
+void rgw_cls_list_ret::generate_test_instances(list<rgw_cls_list_ret*>& o)
+{
+ list<rgw_bucket_dir *> l;
+  rgw_bucket_dir::generate_test_instances(l);
+  list<rgw_bucket_dir *>::iterator iter;
+  for (iter = l.begin(); iter != l.end(); ++iter) {
+    rgw_bucket_dir *d = *iter;
+
+    rgw_cls_list_ret *ret = new rgw_cls_list_ret;
+    ret->dir = *d;
+    ret->is_truncated = true;
+
+    o.push_back(ret);
+
+    delete d;
+  }
+
+  o.push_back(new rgw_cls_list_ret);
+}
+
+void rgw_cls_list_ret::dump(Formatter *f) const
+{
+  f->open_object_section("dir");
+  dir.dump(f);
+  f->close_section();
+  f->dump_int("is_truncated", (int)is_truncated);
+}
+
+void rgw_cls_check_index_ret::generate_test_instances(list<rgw_cls_check_index_ret*>& o)
+{
+  list<rgw_bucket_dir_header *> h;
+  rgw_bucket_dir_header::generate_test_instances(h);
+  rgw_cls_check_index_ret *r = new rgw_cls_check_index_ret;
+  r->existing_header = *(h.front());
+  r->calculated_header = *(h.front());
+  o.push_back(r);
+
+  for (list<rgw_bucket_dir_header *>::iterator iter = h.begin(); iter != h.end(); ++iter) {
+    delete *iter;
+  }
+  o.push_back(new rgw_cls_check_index_ret);
+}
+
+void rgw_cls_check_index_ret::dump(Formatter *f) const
+{
+  ::encode_json("existing_header", existing_header, f);
+  ::encode_json("calculated_header", calculated_header, f);
+}
+
+void rgw_cls_bucket_update_stats_op::generate_test_instances(list<rgw_cls_bucket_update_stats_op*>& o)
+{
+  rgw_cls_bucket_update_stats_op *r = new rgw_cls_bucket_update_stats_op;
+  r->absolute = true;
+  rgw_bucket_category_stats& s = r->stats[RGWObjCategory::None];
+  s.total_size = 1;
+  s.total_size_rounded = 4096;
+  s.num_entries = 1;
+  o.push_back(r);
+
+  o.push_back(new rgw_cls_bucket_update_stats_op);
+}
+
+void rgw_cls_bucket_update_stats_op::dump(Formatter *f) const
+{
+  ::encode_json("absolute", absolute, f);
+  map<int, rgw_bucket_category_stats> s;
+  for (auto& entry : stats) {
+    s[(int)entry.first] = entry.second;
+  }
+  ::encode_json("stats", s, f);
+}
+
+void cls_rgw_bi_log_list_op::dump(Formatter *f) const
+{
+  f->dump_string("marker", marker);
+  f->dump_unsigned("max", max);
+}
+
+void cls_rgw_bi_log_list_op::generate_test_instances(list<cls_rgw_bi_log_list_op*>& ls)
+{
+  ls.push_back(new cls_rgw_bi_log_list_op);
+  ls.push_back(new cls_rgw_bi_log_list_op);
+  ls.back()->marker = "mark";
+  ls.back()->max = 123;
+}
+
+void cls_rgw_bi_log_trim_op::dump(Formatter *f) const
+{
+  f->dump_string("start_marker", start_marker);
+  f->dump_string("end_marker", end_marker);
+}
+
+void cls_rgw_bi_log_trim_op::generate_test_instances(list<cls_rgw_bi_log_trim_op*>& ls)
+{
+  ls.push_back(new cls_rgw_bi_log_trim_op);
+  ls.push_back(new cls_rgw_bi_log_trim_op);
+  ls.back()->start_marker = "foo";
+  ls.back()->end_marker = "bar";
+}
+
+void cls_rgw_bi_log_list_ret::dump(Formatter *f) const
+{
+  encode_json("entries", entries, f);
+  f->dump_unsigned("truncated", (int)truncated);
+}
+
+void cls_rgw_bi_log_list_ret::generate_test_instances(list<cls_rgw_bi_log_list_ret*>& ls)
+{
+  ls.push_back(new cls_rgw_bi_log_list_ret);
+  ls.push_back(new cls_rgw_bi_log_list_ret);
+  ls.back()->entries.push_back(rgw_bi_log_entry());
+  ls.back()->truncated = true;
+}
+
+void cls_rgw_reshard_add_op::generate_test_instances(list<cls_rgw_reshard_add_op*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_add_op);
+  ls.push_back(new cls_rgw_reshard_add_op);
+  list<cls_rgw_reshard_entry *> l;
+  cls_rgw_reshard_entry::generate_test_instances(l);
+  list<cls_rgw_reshard_entry *>::iterator iter = l.begin();
+  ls.back()->entry = *(*iter);
+}
+
+void cls_rgw_reshard_add_op::dump(Formatter *f) const
+{
+  ::encode_json("entry", entry, f);
+}
+
+void cls_rgw_reshard_list_op::generate_test_instances(list<cls_rgw_reshard_list_op*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_list_op);
+  ls.push_back(new cls_rgw_reshard_list_op);
+  ls.back()->max = 1000;
+  ls.back()->marker = "foo";
+}
+
+void cls_rgw_reshard_list_op::dump(Formatter *f) const
+{
+  ::encode_json("max", max, f);
+  ::encode_json("marker", marker, f);
+}
+
+void cls_rgw_reshard_list_ret::generate_test_instances(list<cls_rgw_reshard_list_ret*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_list_ret);
+  ls.push_back(new cls_rgw_reshard_list_ret);
+  ls.back()->entries.push_back(cls_rgw_reshard_entry());
+  ls.back()->is_truncated = true;
+}
+
+void cls_rgw_reshard_list_ret::dump(Formatter *f) const
+{
+  ::encode_json("entries", entries, f);
+  ::encode_json("is_truncated", is_truncated, f);
+}
+
+void cls_rgw_reshard_get_op::generate_test_instances(list<cls_rgw_reshard_get_op*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_get_op);
+  ls.push_back(new cls_rgw_reshard_get_op);
+}
+
+void cls_rgw_reshard_get_op::dump(Formatter *f) const
+{
+  ::encode_json("entry", entry, f);
+}
+
+void cls_rgw_reshard_get_ret::generate_test_instances(list<cls_rgw_reshard_get_ret*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_get_ret);
+  ls.push_back(new cls_rgw_reshard_get_ret);
+}
+
+void cls_rgw_reshard_get_ret::dump(Formatter *f) const
+{
+  ::encode_json("entry", entry, f);
+}
+
+void cls_rgw_reshard_remove_op::generate_test_instances(list<cls_rgw_reshard_remove_op*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_remove_op);
+  ls.push_back(new cls_rgw_reshard_remove_op);
+  ls.back()->bucket_name = "foo";
+  ls.back()->bucket_id = "bucket_id";
+}
+
+void cls_rgw_reshard_remove_op::dump(Formatter *f) const
+{
+  ::encode_json("bucket_name", bucket_name, f);
+  ::encode_json("bucket_id", bucket_name, f);
+}
+
+
+void cls_rgw_set_bucket_resharding_op::generate_test_instances(
+  list<cls_rgw_set_bucket_resharding_op*>& ls)
+{
+  ls.push_back(new cls_rgw_set_bucket_resharding_op);
+  ls.push_back(new cls_rgw_set_bucket_resharding_op);
+}
+
+void cls_rgw_set_bucket_resharding_op::dump(Formatter *f) const
+{
+  ::encode_json("entry", entry, f);
+}
+
+void cls_rgw_clear_bucket_resharding_op::generate_test_instances(
+  list<cls_rgw_clear_bucket_resharding_op*>& ls)
+{
+  ls.push_back(new cls_rgw_clear_bucket_resharding_op);
+  ls.push_back(new cls_rgw_clear_bucket_resharding_op);
+}
+
+void cls_rgw_clear_bucket_resharding_op::dump(Formatter *f) const
+{
+}
+
+void cls_rgw_guard_bucket_resharding_op::generate_test_instances(
+  list<cls_rgw_guard_bucket_resharding_op*>& ls)
+{
+  ls.push_back(new cls_rgw_guard_bucket_resharding_op);
+  ls.push_back(new cls_rgw_guard_bucket_resharding_op);
+}
+
+void cls_rgw_guard_bucket_resharding_op::dump(Formatter *f) const
+{
+  ::encode_json("ret_err", ret_err, f);
+}
+
+
+void cls_rgw_get_bucket_resharding_op::generate_test_instances(
+  list<cls_rgw_get_bucket_resharding_op*>& ls)
+{
+  ls.push_back(new cls_rgw_get_bucket_resharding_op);
+  ls.push_back(new cls_rgw_get_bucket_resharding_op);
+}
+
+void cls_rgw_get_bucket_resharding_op::dump(Formatter *f) const
+{
+}
+
+
+
+
+
diff --git a/src/cls/rgw/cls_rgw_ops.h b/src/cls/rgw/cls_rgw_ops.h
new file mode 100644
index 00000000..86f08bcf
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_ops.h
@@ -0,0 +1,1429 @@
+#ifndef CEPH_CLS_RGW_OPS_H
+#define CEPH_CLS_RGW_OPS_H
+
+#include "cls/rgw/cls_rgw_types.h"
+
+struct rgw_cls_tag_timeout_op
+{
+  uint64_t tag_timeout;
+
+  rgw_cls_tag_timeout_op() : tag_timeout(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(tag_timeout, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(tag_timeout, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_tag_timeout_op*>& ls);
+};
+WRITE_CLASS_ENCODER(rgw_cls_tag_timeout_op)
+
+struct rgw_cls_obj_prepare_op
+{
+  RGWModifyOp op;
+  cls_rgw_obj_key key;
+  string tag;
+  string locator;
+  bool log_op;
+  uint16_t bilog_flags;
+  rgw_zone_set zones_trace;
+
+  rgw_cls_obj_prepare_op() : op(CLS_RGW_OP_UNKNOWN), log_op(false), bilog_flags(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(7, 5, bl);
+    uint8_t c = (uint8_t)op;
+    encode(c, bl);
+    encode(tag, bl);
+    encode(locator, bl);
+    encode(log_op, bl);
+    encode(key, bl);
+    encode(bilog_flags, bl);
+    encode(zones_trace, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(7, 3, 3, bl);
+    uint8_t c;
+    decode(c, bl);
+    op = (RGWModifyOp)c;
+    if (struct_v < 5) {
+      decode(key.name, bl);
+    }
+    decode(tag, bl);
+    if (struct_v >= 2) {
+      decode(locator, bl);
+    }
+    if (struct_v >= 4) {
+      decode(log_op, bl);
+    }
+    if (struct_v >= 5) {
+      decode(key, bl);
+    }
+    if (struct_v >= 6) {
+      decode(bilog_flags, bl);
+    }
+    if (struct_v >= 7) {
+      decode(zones_trace, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_obj_prepare_op*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_cls_obj_prepare_op)
+
+struct rgw_cls_obj_complete_op
+{
+  RGWModifyOp op;
+  cls_rgw_obj_key key;
+  string locator;
+  rgw_bucket_entry_ver ver;
+  rgw_bucket_dir_entry_meta meta;
+  string tag;
+  bool log_op;
+  uint16_t bilog_flags;
+
+  list<cls_rgw_obj_key> remove_objs;
+  rgw_zone_set zones_trace;
+
+  rgw_cls_obj_complete_op() : op(CLS_RGW_OP_ADD), log_op(false), bilog_flags(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(9, 7, bl);
+    uint8_t c = (uint8_t)op;
+    encode(c, bl);
+    encode(ver.epoch, bl);
+    encode(meta, bl);
+    encode(tag, bl);
+    encode(locator, bl);
+    encode(remove_objs, bl);
+    encode(ver, bl);
+    encode(log_op, bl);
+    encode(key, bl);
+    encode(bilog_flags, bl);
+    encode(zones_trace, bl);
+    ENCODE_FINISH(bl);
+ }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(9, 3, 3, bl);
+    uint8_t c;
+    decode(c, bl);
+    op = (RGWModifyOp)c;
+    if (struct_v < 7) {
+      decode(key.name, bl);
+    }
+    decode(ver.epoch, bl);
+    decode(meta, bl);
+    decode(tag, bl);
+    if (struct_v >= 2) {
+      decode(locator, bl);
+    }
+    if (struct_v >= 4 && struct_v < 7) {
+      list<string> old_remove_objs;
+      decode(old_remove_objs, bl);
+
+      for (list<string>::iterator iter = old_remove_objs.begin();
+           iter != old_remove_objs.end(); ++iter) {
+        cls_rgw_obj_key k;
+        k.name = *iter;
+        remove_objs.push_back(k);
+      }
+    } else {
+      decode(remove_objs, bl);
+    }
+    if (struct_v >= 5) {
+      decode(ver, bl);
+    } else {
+      ver.pool = -1;
+    }
+    if (struct_v >= 6) {
+      decode(log_op, bl);
+    }
+    if (struct_v >= 7) {
+      decode(key, bl);
+    }
+    if (struct_v >= 8) {
+      decode(bilog_flags, bl);
+    }
+    if (struct_v >= 9) {
+      decode(zones_trace, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_obj_complete_op*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_cls_obj_complete_op)
+
+struct rgw_cls_link_olh_op {
+  cls_rgw_obj_key key;
+  string olh_tag;
+  bool delete_marker;
+  string op_tag;
+  rgw_bucket_dir_entry_meta meta;
+  uint64_t olh_epoch;
+  bool log_op;
+  uint16_t bilog_flags;
+  real_time unmod_since; /* only create delete marker if newer then this */
+  bool high_precision_time;
+  rgw_zone_set zones_trace;
+
+  rgw_cls_link_olh_op() : delete_marker(false), olh_epoch(0), log_op(false), bilog_flags(0), high_precision_time(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(5, 1, bl);
+    encode(key, bl);
+    encode(olh_tag, bl);
+    encode(delete_marker, bl);
+    encode(op_tag, bl);
+    encode(meta, bl);
+    encode(olh_epoch, bl);
+    encode(log_op, bl);
+    encode(bilog_flags, bl);
+    uint64_t t = ceph::real_clock::to_time_t(unmod_since);
+    encode(t, bl);
+    encode(unmod_since, bl);
+    encode(high_precision_time, bl);
+    encode(zones_trace, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(5, bl);
+    decode(key, bl);
+    decode(olh_tag, bl);
+    decode(delete_marker, bl);
+    decode(op_tag, bl);
+    decode(meta, bl);
+    decode(olh_epoch, bl);
+    decode(log_op, bl);
+    decode(bilog_flags, bl);
+    if (struct_v == 2) {
+      uint64_t t;
+      decode(t, bl);
+      unmod_since = ceph::real_clock::from_time_t(static_cast<time_t>(t));
+    }
+    if (struct_v >= 3) {
+      uint64_t t;
+      decode(t, bl);
+      decode(unmod_since, bl);
+    }
+    if (struct_v >= 4) {
+      decode(high_precision_time, bl);
+    }
+    if (struct_v >= 5) {
+      decode(zones_trace, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  static void generate_test_instances(list<rgw_cls_link_olh_op *>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(rgw_cls_link_olh_op)
+
+struct rgw_cls_unlink_instance_op {
+  cls_rgw_obj_key key;
+  string op_tag;
+  uint64_t olh_epoch;
+  bool log_op;
+  uint16_t bilog_flags;
+  string olh_tag;
+  rgw_zone_set zones_trace;
+
+  rgw_cls_unlink_instance_op() : olh_epoch(0), log_op(false), bilog_flags(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(3, 1, bl);
+    encode(key, bl);
+    encode(op_tag, bl);
+    encode(olh_epoch, bl);
+    encode(log_op, bl);
+    encode(bilog_flags, bl);
+    encode(olh_tag, bl);
+    encode(zones_trace, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(3, bl);
+    decode(key, bl);
+    decode(op_tag, bl);
+    decode(olh_epoch, bl);
+    decode(log_op, bl);
+    decode(bilog_flags, bl);
+    if (struct_v >= 2) {
+      decode(olh_tag, bl);
+    }
+    if (struct_v >= 3) {
+      decode(zones_trace, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  static void generate_test_instances(list<rgw_cls_unlink_instance_op *>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(rgw_cls_unlink_instance_op)
+
+struct rgw_cls_read_olh_log_op
+{
+  cls_rgw_obj_key olh;
+  uint64_t ver_marker;
+  string olh_tag;
+
+  rgw_cls_read_olh_log_op() : ver_marker(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(olh, bl);
+    encode(ver_marker, bl);
+    encode(olh_tag, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(olh, bl);
+    decode(ver_marker, bl);
+    decode(olh_tag, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<rgw_cls_read_olh_log_op *>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(rgw_cls_read_olh_log_op)
+
+
+struct rgw_cls_read_olh_log_ret
+{
+  map<uint64_t, vector<rgw_bucket_olh_log_entry> > log;
+  bool is_truncated;
+
+  rgw_cls_read_olh_log_ret() : is_truncated(false) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(log, bl);
+    encode(is_truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(log, bl);
+    decode(is_truncated, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<rgw_cls_read_olh_log_ret *>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(rgw_cls_read_olh_log_ret)
+
+struct rgw_cls_trim_olh_log_op
+{
+  cls_rgw_obj_key olh;
+  uint64_t ver;
+  string olh_tag;
+
+  rgw_cls_trim_olh_log_op() : ver(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(olh, bl);
+    encode(ver, bl);
+    encode(olh_tag, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(olh, bl);
+    decode(ver, bl);
+    decode(olh_tag, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<rgw_cls_trim_olh_log_op *>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(rgw_cls_trim_olh_log_op)
+
+struct rgw_cls_bucket_clear_olh_op {
+  cls_rgw_obj_key key;
+  string olh_tag;
+
+  rgw_cls_bucket_clear_olh_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(key, bl);
+    encode(olh_tag, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(key, bl);
+    decode(olh_tag, bl);
+    DECODE_FINISH(bl);
+  }
+
+  static void generate_test_instances(list<rgw_cls_bucket_clear_olh_op *>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(rgw_cls_bucket_clear_olh_op)
+
+struct rgw_cls_list_op
+{
+  cls_rgw_obj_key start_obj;
+  uint32_t num_entries;
+  string filter_prefix;
+  bool list_versions;
+
+  rgw_cls_list_op() : num_entries(0), list_versions(false) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(5, 4, bl);
+    encode(num_entries, bl);
+    encode(filter_prefix, bl);
+    encode(start_obj, bl);
+    encode(list_versions, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(5, 2, 2, bl);
+    if (struct_v < 4) {
+      decode(start_obj.name, bl);
+    }
+    decode(num_entries, bl);
+    if (struct_v >= 3)
+      decode(filter_prefix, bl);
+    if (struct_v >= 4)
+      decode(start_obj, bl);
+    if (struct_v >= 5)
+      decode(list_versions, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_list_op*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_cls_list_op)
+
+struct rgw_cls_list_ret {
+  rgw_bucket_dir dir;
+  bool is_truncated;
+
+  rgw_cls_list_ret() : is_truncated(false) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(2, 2, bl);
+    encode(dir, bl);
+    encode(is_truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+    decode(dir, bl);
+    decode(is_truncated, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_list_ret*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_cls_list_ret)
+
+struct rgw_cls_check_index_ret
+{
+  rgw_bucket_dir_header existing_header;
+  rgw_bucket_dir_header calculated_header;
+
+  rgw_cls_check_index_ret() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(existing_header, bl);
+    encode(calculated_header, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(existing_header, bl);
+    decode(calculated_header, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_check_index_ret *>& o);
+};
+WRITE_CLASS_ENCODER(rgw_cls_check_index_ret)
+
+struct rgw_cls_bucket_update_stats_op
+{
+  bool absolute{false};
+  map<RGWObjCategory, rgw_bucket_category_stats> stats;
+
+  rgw_cls_bucket_update_stats_op() {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(absolute, bl);
+    encode(stats, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(absolute, bl);
+    decode(stats, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_cls_bucket_update_stats_op *>& o);
+};
+WRITE_CLASS_ENCODER(rgw_cls_bucket_update_stats_op)
+
+struct rgw_cls_obj_remove_op {
+  list<string> keep_attr_prefixes;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(keep_attr_prefixes, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(keep_attr_prefixes, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_obj_remove_op)
+
+struct rgw_cls_obj_store_pg_ver_op {
+  string attr;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(attr, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(attr, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_obj_store_pg_ver_op)
+
+struct rgw_cls_obj_check_attrs_prefix {
+  string check_prefix;
+  bool fail_if_exist;
+
+  rgw_cls_obj_check_attrs_prefix() : fail_if_exist(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(check_prefix, bl);
+    encode(fail_if_exist, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(check_prefix, bl);
+    decode(fail_if_exist, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_obj_check_attrs_prefix)
+
+struct rgw_cls_obj_check_mtime {
+  ceph::real_time mtime;
+  RGWCheckMTimeType type;
+  bool high_precision_time;
+
+  rgw_cls_obj_check_mtime() : type(CLS_RGW_CHECK_TIME_MTIME_EQ), high_precision_time(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(mtime, bl);
+    encode((uint8_t)type, bl);
+    encode(high_precision_time, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(mtime, bl);
+    uint8_t c;
+    decode(c, bl);
+    type = (RGWCheckMTimeType)c;
+    if (struct_v >= 2) {
+      decode(high_precision_time, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_obj_check_mtime)
+
+struct rgw_cls_usage_log_add_op {
+  rgw_usage_log_info info;
+  rgw_user user;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(info, bl);
+    encode(user.to_str(), bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(info, bl);
+    if (struct_v >= 2) {
+      string s;
+      decode(s, bl);
+      user.from_str(s);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_usage_log_add_op)
+
+struct rgw_cls_bi_get_op {
+  cls_rgw_obj_key key;
+  BIIndexType type; /* namespace: plain, instance, olh */
+
+  rgw_cls_bi_get_op() : type(BIIndexType::Plain) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(key, bl);
+    encode((uint8_t)type, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(key, bl);
+    uint8_t c;
+    decode(c, bl);
+    type = (BIIndexType)c;
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_bi_get_op)
+
+struct rgw_cls_bi_get_ret {
+  rgw_cls_bi_entry entry;
+
+  rgw_cls_bi_get_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_bi_get_ret)
+
+struct rgw_cls_bi_put_op {
+  rgw_cls_bi_entry entry;
+
+  rgw_cls_bi_put_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_bi_put_op)
+
+struct rgw_cls_bi_list_op {
+  uint32_t max;
+  string name;
+  string marker;
+
+  rgw_cls_bi_list_op() : max(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(max, bl);
+    encode(name, bl);
+    encode(marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(max, bl);
+    decode(name, bl);
+    decode(marker, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_bi_list_op)
+
+struct rgw_cls_bi_list_ret {
+  list<rgw_cls_bi_entry> entries;
+  bool is_truncated;
+
+  rgw_cls_bi_list_ret() : is_truncated(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(is_truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(is_truncated, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_bi_list_ret)
+
+struct rgw_cls_usage_log_read_op {
+  uint64_t start_epoch;
+  uint64_t end_epoch;
+  string owner;
+  string bucket;
+
+  string iter;  // should be empty for the first call, non empty for subsequent calls
+  uint32_t max_entries;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(start_epoch, bl);
+    encode(end_epoch, bl);
+    encode(owner, bl);
+    encode(iter, bl);
+    encode(max_entries, bl);
+    encode(bucket, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(start_epoch, bl);
+    decode(end_epoch, bl);
+    decode(owner, bl);
+    decode(iter, bl);
+    decode(max_entries, bl);
+    if (struct_v >= 2) {
+      decode(bucket, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_usage_log_read_op)
+
+struct rgw_cls_usage_log_read_ret {
+  map<rgw_user_bucket, rgw_usage_log_entry> usage;
+  bool truncated;
+  string next_iter;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(usage, bl);
+    encode(truncated, bl);
+    encode(next_iter, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(usage, bl);
+    decode(truncated, bl);
+    decode(next_iter, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_usage_log_read_ret)
+
+struct rgw_cls_usage_log_trim_op {
+  uint64_t start_epoch;
+  uint64_t end_epoch;
+  string user;
+  string bucket;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(3, 2, bl);
+    encode(start_epoch, bl);
+    encode(end_epoch, bl);
+    encode(user, bl);
+    encode(bucket, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(3, bl);
+    decode(start_epoch, bl);
+    decode(end_epoch, bl);
+    decode(user, bl);
+    if (struct_v >= 3) {
+      decode(bucket, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_cls_usage_log_trim_op)
+
+struct cls_rgw_gc_set_entry_op {
+  uint32_t expiration_secs;
+  cls_rgw_gc_obj_info info;
+  cls_rgw_gc_set_entry_op() : expiration_secs(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(expiration_secs, bl);
+    encode(info, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(expiration_secs, bl);
+    decode(info, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_gc_set_entry_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_gc_set_entry_op)
+
+struct cls_rgw_gc_defer_entry_op {
+  uint32_t expiration_secs;
+  string tag;
+  cls_rgw_gc_defer_entry_op() : expiration_secs(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(expiration_secs, bl);
+    encode(tag, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(expiration_secs, bl);
+    decode(tag, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_gc_defer_entry_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_gc_defer_entry_op)
+
+struct cls_rgw_gc_list_op {
+  string marker;
+  uint32_t max;
+  bool expired_only;
+
+  cls_rgw_gc_list_op() : max(0), expired_only(true) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(marker, bl);
+    encode(max, bl);
+    encode(expired_only, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(marker, bl);
+    decode(max, bl);
+    if (struct_v >= 2) {
+      decode(expired_only, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_gc_list_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_gc_list_op)
+
+struct cls_rgw_gc_list_ret {
+  list<cls_rgw_gc_obj_info> entries;
+  string next_marker;
+  bool truncated;
+
+  cls_rgw_gc_list_ret() : truncated(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(entries, bl);
+    encode(next_marker, bl);
+    encode(truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(entries, bl);
+    if (struct_v >= 2)
+      decode(next_marker, bl);
+    decode(truncated, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_gc_list_ret*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_gc_list_ret)
+
+struct cls_rgw_gc_remove_op {
+  vector<string> tags;
+
+  cls_rgw_gc_remove_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(tags, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(tags, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_gc_remove_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_gc_remove_op)
+
+struct cls_rgw_bi_log_list_op {
+  string marker;
+  uint32_t max;
+
+  cls_rgw_bi_log_list_op() : max(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(marker, bl);
+    encode(max, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(marker, bl);
+    decode(max, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_bi_log_list_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_bi_log_list_op)
+
+struct cls_rgw_bi_log_trim_op {
+  string start_marker;
+  string end_marker;
+
+  cls_rgw_bi_log_trim_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(start_marker, bl);
+    encode(end_marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(start_marker, bl);
+    decode(end_marker, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_bi_log_trim_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_bi_log_trim_op)
+
+struct cls_rgw_bi_log_list_ret {
+  list<rgw_bi_log_entry> entries;
+  bool truncated;
+
+  cls_rgw_bi_log_list_ret() : truncated(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(truncated, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_bi_log_list_ret*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_bi_log_list_ret)
+
+struct cls_rgw_lc_get_next_entry_op {
+  string marker;
+  cls_rgw_lc_get_next_entry_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(marker, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_get_next_entry_op)
+
+using rgw_lc_entry_t = std::pair<std::string, int>;
+
+struct cls_rgw_lc_get_next_entry_ret {
+  rgw_lc_entry_t entry;
+  cls_rgw_lc_get_next_entry_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_get_next_entry_ret)
+
+struct cls_rgw_lc_get_entry_op {
+  string marker;
+  cls_rgw_lc_get_entry_op() {}
+  cls_rgw_lc_get_entry_op(const std::string& _marker) : marker(_marker) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(marker, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_get_entry_op)
+
+struct cls_rgw_lc_get_entry_ret {
+  rgw_lc_entry_t entry;
+  cls_rgw_lc_get_entry_ret() {}
+  cls_rgw_lc_get_entry_ret(rgw_lc_entry_t&& _entry) : entry(std::move(_entry)) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_get_entry_ret)
+
+
+struct cls_rgw_lc_rm_entry_op {
+  rgw_lc_entry_t entry;
+  cls_rgw_lc_rm_entry_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_rm_entry_op)
+
+struct cls_rgw_lc_set_entry_op {
+  rgw_lc_entry_t entry;
+  cls_rgw_lc_set_entry_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_set_entry_op)
+
+struct cls_rgw_lc_put_head_op {
+  cls_rgw_lc_obj_head head;
+
+
+  cls_rgw_lc_put_head_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(head, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(head, bl);
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_put_head_op)
+
+struct cls_rgw_lc_get_head_ret {
+  cls_rgw_lc_obj_head head;
+
+  cls_rgw_lc_get_head_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(head, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(head, bl);
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_get_head_ret)
+
+struct cls_rgw_lc_list_entries_op {
+  string marker;
+  uint32_t max_entries = 0;
+
+  cls_rgw_lc_list_entries_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(marker, bl);
+    encode(max_entries, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(marker, bl);
+    decode(max_entries, bl);
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_list_entries_op)
+
+struct cls_rgw_lc_list_entries_ret {
+  map<string, int> entries;
+  bool is_truncated{false};
+
+  cls_rgw_lc_list_entries_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(entries, bl);
+    encode(is_truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(entries, bl);
+    if (struct_v >= 2) {
+      decode(is_truncated, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_list_entries_ret)
+
+struct cls_rgw_reshard_add_op {
+ cls_rgw_reshard_entry entry;
+
+  cls_rgw_reshard_add_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_reshard_add_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_add_op)
+
+struct cls_rgw_reshard_list_op {
+  uint32_t max{0};
+  string marker;
+
+  cls_rgw_reshard_list_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(max, bl);
+    encode(marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(max, bl);
+    decode(marker, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_reshard_list_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_list_op)
+
+
+struct cls_rgw_reshard_list_ret {
+  list<cls_rgw_reshard_entry> entries;
+  bool is_truncated{false};
+
+  cls_rgw_reshard_list_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(is_truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(is_truncated, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_reshard_list_ret*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_list_ret)
+
+struct cls_rgw_reshard_get_op {
+  cls_rgw_reshard_entry entry;
+
+  cls_rgw_reshard_get_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_reshard_get_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_get_op)
+
+struct cls_rgw_reshard_get_ret {
+  cls_rgw_reshard_entry entry;
+
+  cls_rgw_reshard_get_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_reshard_get_ret*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_get_ret)
+
+struct cls_rgw_reshard_remove_op {
+  string tenant;
+  string bucket_name;
+  string bucket_id;
+
+  cls_rgw_reshard_remove_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(tenant, bl);
+    encode(bucket_name, bl);
+    encode(bucket_id, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(tenant, bl);
+    decode(bucket_name, bl);
+    decode(bucket_id, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_reshard_remove_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_remove_op)
+
+struct cls_rgw_set_bucket_resharding_op  {
+  cls_rgw_bucket_instance_entry entry;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entry, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entry, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_set_bucket_resharding_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_set_bucket_resharding_op)
+
+struct cls_rgw_clear_bucket_resharding_op {
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<cls_rgw_clear_bucket_resharding_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_clear_bucket_resharding_op)
+
+struct cls_rgw_guard_bucket_resharding_op  {
+  int ret_err{0};
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(ret_err, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(ret_err, bl);
+    DECODE_FINISH(bl);
+  }
+
+  static void generate_test_instances(list<cls_rgw_guard_bucket_resharding_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_guard_bucket_resharding_op)
+
+struct cls_rgw_get_bucket_resharding_op  {
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    DECODE_FINISH(bl);
+  }
+
+  static void generate_test_instances(list<cls_rgw_get_bucket_resharding_op*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_get_bucket_resharding_op)
+
+struct cls_rgw_get_bucket_resharding_ret  {
+  cls_rgw_bucket_instance_entry new_instance;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(new_instance, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(new_instance, bl);
+    DECODE_FINISH(bl);
+  }
+
+  static void generate_test_instances(list<cls_rgw_get_bucket_resharding_ret*>& o);
+  void dump(Formatter *f) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_get_bucket_resharding_ret)
+
+#endif /* CEPH_CLS_RGW_OPS_H */
diff --git a/src/cls/rgw/cls_rgw_types.cc b/src/cls/rgw/cls_rgw_types.cc
new file mode 100644
index 00000000..b533ec05
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_types.cc
@@ -0,0 +1,700 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/rgw/cls_rgw_types.h"
+#include "common/ceph_json.h"
+#include "include/utime.h"
+
+
+void rgw_bucket_pending_info::generate_test_instances(list<rgw_bucket_pending_info*>& o)
+{
+  rgw_bucket_pending_info *i = new rgw_bucket_pending_info;
+  i->state = CLS_RGW_STATE_COMPLETE;
+  i->op = CLS_RGW_OP_DEL;
+  o.push_back(i);
+  o.push_back(new rgw_bucket_pending_info);
+}
+
+void rgw_bucket_pending_info::dump(Formatter *f) const
+{
+  encode_json("state", (int)state, f);
+  utime_t ut(timestamp);
+  encode_json("timestamp", ut, f);
+  encode_json("op", (int)op, f);
+}
+
+void rgw_bucket_pending_info::decode_json(JSONObj *obj) {
+  int val;
+  JSONDecoder::decode_json("state", val, obj);
+  state = (RGWPendingState)val;
+  utime_t ut(timestamp);
+  JSONDecoder::decode_json("timestamp", ut, obj);
+  JSONDecoder::decode_json("op", val, obj);
+  op = (uint8_t)val;
+}
+
+void cls_rgw_obj_key::decode_json(JSONObj *obj) {
+  JSONDecoder::decode_json("name", name, obj);
+  JSONDecoder::decode_json("instance", instance, obj);
+}
+
+void rgw_bucket_dir_entry_meta::generate_test_instances(list<rgw_bucket_dir_entry_meta*>& o)
+{
+  rgw_bucket_dir_entry_meta *m = new rgw_bucket_dir_entry_meta;
+  m->category = RGWObjCategory::Main;
+  m->size = 100;
+  m->etag = "etag";
+  m->owner = "owner";
+  m->owner_display_name = "display name";
+  m->content_type = "content/type";
+  o.push_back(m);
+  o.push_back(new rgw_bucket_dir_entry_meta);
+}
+
+void rgw_bucket_dir_entry_meta::dump(Formatter *f) const
+{
+  encode_json("category", (int)category, f);
+  encode_json("size", size, f);
+  utime_t ut(mtime);
+  encode_json("mtime", ut, f);
+  encode_json("etag", etag, f);
+  encode_json("storage_class", storage_class, f);
+  encode_json("owner", owner, f);
+  encode_json("owner_display_name", owner_display_name, f);
+  encode_json("content_type", content_type, f);
+  encode_json("accounted_size", accounted_size, f);
+  encode_json("user_data", user_data, f);
+  encode_json("appendable", appendable, f);
+}
+
+void rgw_bucket_dir_entry_meta::decode_json(JSONObj *obj) {
+  int val;
+  JSONDecoder::decode_json("category", val, obj);
+  category = static_cast<RGWObjCategory>(val);
+  JSONDecoder::decode_json("size", size, obj);
+  utime_t ut;
+  JSONDecoder::decode_json("mtime", ut, obj);
+  mtime = ut.to_real_time();
+  JSONDecoder::decode_json("etag", etag, obj);
+  JSONDecoder::decode_json("storage_class", storage_class, obj);
+  JSONDecoder::decode_json("owner", owner, obj);
+  JSONDecoder::decode_json("owner_display_name", owner_display_name, obj);
+  JSONDecoder::decode_json("content_type", content_type, obj);
+  JSONDecoder::decode_json("accounted_size", accounted_size, obj);
+  JSONDecoder::decode_json("user_data", user_data, obj);
+  JSONDecoder::decode_json("appendable", appendable, obj);
+}
+
+void rgw_bucket_dir_entry::generate_test_instances(list<rgw_bucket_dir_entry*>& o)
+{
+  list<rgw_bucket_dir_entry_meta *> l;
+  rgw_bucket_dir_entry_meta::generate_test_instances(l);
+
+  list<rgw_bucket_dir_entry_meta *>::iterator iter;
+  for (iter = l.begin(); iter != l.end(); ++iter) {
+    rgw_bucket_dir_entry_meta *m = *iter;
+    rgw_bucket_dir_entry *e = new rgw_bucket_dir_entry;
+    e->key.name = "name";
+    e->ver.pool = 1;
+    e->ver.epoch = 1234;
+    e->locator = "locator";
+    e->exists = true;
+    e->meta = *m;
+    e->tag = "tag";
+
+    o.push_back(e);
+
+    delete m;
+  }
+  o.push_back(new rgw_bucket_dir_entry);
+}
+
+void rgw_bucket_entry_ver::dump(Formatter *f) const
+{
+  encode_json("pool", pool, f);
+  encode_json("epoch", epoch, f);
+}
+
+void rgw_bucket_entry_ver::decode_json(JSONObj *obj) {
+  JSONDecoder::decode_json("pool", pool, obj);
+  JSONDecoder::decode_json("epoch", epoch, obj);
+}
+
+void rgw_bucket_entry_ver::generate_test_instances(list<rgw_bucket_entry_ver*>& ls)
+{
+  ls.push_back(new rgw_bucket_entry_ver);
+  ls.push_back(new rgw_bucket_entry_ver);
+  ls.back()->pool = 123;
+  ls.back()->epoch = 12322;
+}
+
+
+void rgw_bucket_dir_entry::dump(Formatter *f) const
+{
+  encode_json("name", key.name, f);
+  encode_json("instance", key.instance , f);
+  encode_json("ver", ver , f);
+  encode_json("locator", locator , f);
+  encode_json("exists", exists , f);
+  encode_json("meta", meta , f);
+  encode_json("tag", tag , f);
+  encode_json("flags", (int)flags , f);
+  encode_json("pending_map", pending_map, f);
+  encode_json("versioned_epoch", versioned_epoch , f);
+}
+
+void rgw_bucket_dir_entry::decode_json(JSONObj *obj) {
+  JSONDecoder::decode_json("name", key.name, obj);
+  JSONDecoder::decode_json("instance", key.instance , obj);
+  JSONDecoder::decode_json("ver", ver , obj);
+  JSONDecoder::decode_json("locator", locator , obj);
+  JSONDecoder::decode_json("exists", exists , obj);
+  JSONDecoder::decode_json("meta", meta , obj);
+  JSONDecoder::decode_json("tag", tag , obj);
+  int val;
+  JSONDecoder::decode_json("flags", val , obj);
+  flags = (uint16_t)val;
+  JSONDecoder::decode_json("pending_map", pending_map, obj);
+  JSONDecoder::decode_json("versioned_epoch", versioned_epoch, obj);
+}
+
+static void dump_bi_entry(bufferlist bl, BIIndexType index_type, Formatter *formatter)
+{
+  auto iter = bl.cbegin();
+  switch (index_type) {
+    case BIIndexType::Plain:
+    case BIIndexType::Instance:
+      {
+        rgw_bucket_dir_entry entry;
+        decode(entry, iter);
+        encode_json("entry", entry, formatter);
+      }
+      break;
+    case BIIndexType::OLH:
+      {
+        rgw_bucket_olh_entry entry;
+        decode(entry, iter);
+        encode_json("entry", entry, formatter);
+      }
+      break;
+    default:
+      break;
+  }
+}
+
+void rgw_cls_bi_entry::decode_json(JSONObj *obj, cls_rgw_obj_key *effective_key) {
+  JSONDecoder::decode_json("idx", idx, obj);
+  string s;
+  JSONDecoder::decode_json("type", s, obj);
+  if (s == "plain") {
+    type = BIIndexType::Plain;
+  } else if (s == "instance") {
+    type = BIIndexType::Instance;
+  } else if (s == "olh") {
+    type = BIIndexType::OLH;
+  } else {
+    type = BIIndexType::Invalid;
+  }
+  using ceph::encode;
+  switch (type) {
+    case BIIndexType::Plain:
+    case BIIndexType::Instance:
+      {
+        rgw_bucket_dir_entry entry;
+        JSONDecoder::decode_json("entry", entry, obj);
+        encode(entry, data);
+
+        if (effective_key) {
+          *effective_key = entry.key;
+        }
+      }
+      break;
+    case BIIndexType::OLH:
+      {
+        rgw_bucket_olh_entry entry;
+        JSONDecoder::decode_json("entry", entry, obj);
+        encode(entry, data);
+
+        if (effective_key) {
+          *effective_key = entry.key;
+        }
+      }
+      break;
+    default:
+      break;
+  }
+}
+
+void rgw_cls_bi_entry::dump(Formatter *f) const
+{
+  string type_str;
+  switch (type) {
+  case BIIndexType::Plain:
+    type_str = "plain";
+    break;
+  case BIIndexType::Instance:
+    type_str = "instance";
+    break;
+  case BIIndexType::OLH:
+    type_str = "olh";
+    break;
+  default:
+    type_str = "invalid";
+  }
+  encode_json("type", type_str, f);
+  encode_json("idx", idx, f);
+  dump_bi_entry(data, type, f);
+}
+
+bool rgw_cls_bi_entry::get_info(cls_rgw_obj_key *key,
+                                RGWObjCategory *category,
+                                rgw_bucket_category_stats *accounted_stats)
+{
+  bool account = false;
+  auto iter = data.cbegin();
+  using ceph::decode;
+  switch (type) {
+    case BIIndexType::Plain:
+        account = true;
+        // NO BREAK; falls through to case InstanceIdx:
+    case BIIndexType::Instance:
+      {
+        rgw_bucket_dir_entry entry;
+        decode(entry, iter);
+        account = (account && entry.exists);
+        *key = entry.key;
+        *category = entry.meta.category;
+        accounted_stats->num_entries++;
+        accounted_stats->total_size += entry.meta.accounted_size;
+        accounted_stats->total_size_rounded += cls_rgw_get_rounded_size(entry.meta.accounted_size);
+        accounted_stats->actual_size += entry.meta.size;
+      }
+      break;
+    case BIIndexType::OLH:
+      {
+        rgw_bucket_olh_entry entry;
+        decode(entry, iter);
+        *key = entry.key;
+      }
+      break;
+    default:
+      break;
+  }
+
+  return account;
+}
+
+void rgw_bucket_olh_entry::dump(Formatter *f) const
+{
+  encode_json("key", key, f);
+  encode_json("delete_marker", delete_marker, f);
+  encode_json("epoch", epoch, f);
+  encode_json("pending_log", pending_log, f);
+  encode_json("tag", tag, f);
+  encode_json("exists", exists, f);
+  encode_json("pending_removal", pending_removal, f);
+}
+
+void rgw_bucket_olh_entry::decode_json(JSONObj *obj)
+{
+  JSONDecoder::decode_json("key", key, obj);
+  JSONDecoder::decode_json("delete_marker", delete_marker, obj);
+  JSONDecoder::decode_json("epoch", epoch, obj);
+  JSONDecoder::decode_json("pending_log", pending_log, obj);
+  JSONDecoder::decode_json("tag", tag, obj);
+  JSONDecoder::decode_json("exists", exists, obj);
+  JSONDecoder::decode_json("pending_removal", pending_removal, obj);
+}
+
+void rgw_bucket_olh_log_entry::generate_test_instances(list<rgw_bucket_olh_log_entry*>& o)
+{
+  rgw_bucket_olh_log_entry *entry = new rgw_bucket_olh_log_entry;
+  entry->epoch = 1234;
+  entry->op = CLS_RGW_OLH_OP_LINK_OLH;
+  entry->op_tag = "op_tag";
+  entry->key.name = "key.name";
+  entry->key.instance = "key.instance";
+  entry->delete_marker = true;
+  o.push_back(entry);
+  o.push_back(new rgw_bucket_olh_log_entry);
+}
+
+void rgw_bucket_olh_log_entry::dump(Formatter *f) const
+{
+  encode_json("epoch", epoch, f);
+  const char *op_str;
+  switch (op) {
+    case CLS_RGW_OLH_OP_LINK_OLH:
+      op_str = "link_olh";
+      break;
+    case CLS_RGW_OLH_OP_UNLINK_OLH:
+      op_str = "unlink_olh";
+      break;
+    case CLS_RGW_OLH_OP_REMOVE_INSTANCE:
+      op_str = "remove_instance";
+      break;
+    default:
+      op_str = "unknown";
+  }
+  encode_json("op", op_str, f);
+  encode_json("op_tag", op_tag, f);
+  encode_json("key", key, f);
+  encode_json("delete_marker", delete_marker, f);
+}
+
+void rgw_bucket_olh_log_entry::decode_json(JSONObj *obj)
+{
+  JSONDecoder::decode_json("epoch", epoch, obj);
+  string op_str;
+  JSONDecoder::decode_json("op", op_str, obj);
+  if (op_str == "link_olh") {
+    op = CLS_RGW_OLH_OP_LINK_OLH;
+  } else if (op_str == "unlink_olh") {
+    op = CLS_RGW_OLH_OP_UNLINK_OLH;
+  } else if (op_str == "remove_instance") {
+    op = CLS_RGW_OLH_OP_REMOVE_INSTANCE;
+  } else {
+    op = CLS_RGW_OLH_OP_UNKNOWN;
+  }
+  JSONDecoder::decode_json("op_tag", op_tag, obj);
+  JSONDecoder::decode_json("key", key, obj);
+  JSONDecoder::decode_json("delete_marker", delete_marker, obj);
+}
+void rgw_bi_log_entry::decode_json(JSONObj *obj)
+{
+  JSONDecoder::decode_json("op_id", id, obj);
+  JSONDecoder::decode_json("op_tag", tag, obj);
+  string op_str;
+  JSONDecoder::decode_json("op", op_str, obj);
+  if (op_str == "write") {
+    op = CLS_RGW_OP_ADD;
+  } else if (op_str == "del") {
+    op = CLS_RGW_OP_DEL;
+  } else if (op_str == "cancel") {
+    op = CLS_RGW_OP_CANCEL;
+  } else if (op_str == "unknown") {
+    op = CLS_RGW_OP_UNKNOWN;
+  } else if (op_str == "link_olh") {
+    op = CLS_RGW_OP_LINK_OLH;
+  } else if (op_str == "link_olh_del") {
+    op = CLS_RGW_OP_LINK_OLH_DM;
+  } else if (op_str == "unlink_instance") {
+    op = CLS_RGW_OP_UNLINK_INSTANCE;
+  } else if (op_str == "syncstop") {
+    op = CLS_RGW_OP_SYNCSTOP;
+  } else if (op_str == "resync") {
+    op = CLS_RGW_OP_RESYNC;
+  } else {
+    op = CLS_RGW_OP_UNKNOWN;
+  }
+  JSONDecoder::decode_json("object", object, obj);
+  JSONDecoder::decode_json("instance", instance, obj);
+  string state_str;
+  JSONDecoder::decode_json("state", state_str, obj);
+  if (state_str == "pending") {
+    state = CLS_RGW_STATE_PENDING_MODIFY;
+  } else if (state_str == "complete") {
+    state = CLS_RGW_STATE_COMPLETE;
+  } else {
+    state = CLS_RGW_STATE_UNKNOWN;
+  }
+  JSONDecoder::decode_json("index_ver", index_ver, obj);
+  utime_t ut;
+  JSONDecoder::decode_json("timestamp", ut, obj);
+  timestamp = ut.to_real_time();
+  uint32_t f;
+  JSONDecoder::decode_json("bilog_flags", f, obj);
+  JSONDecoder::decode_json("ver", ver, obj);
+  bilog_flags = (uint16_t)f;
+  JSONDecoder::decode_json("owner", owner, obj);
+  JSONDecoder::decode_json("owner_display_name", owner_display_name, obj);
+  JSONDecoder::decode_json("zones_trace", zones_trace, obj);
+}
+
+void rgw_bi_log_entry::dump(Formatter *f) const
+{
+  f->dump_string("op_id", id);
+  f->dump_string("op_tag", tag);
+  switch (op) {
+    case CLS_RGW_OP_ADD:
+      f->dump_string("op", "write");
+      break;
+    case CLS_RGW_OP_DEL:
+      f->dump_string("op", "del");
+      break;
+    case CLS_RGW_OP_CANCEL:
+      f->dump_string("op", "cancel");
+      break;
+    case CLS_RGW_OP_UNKNOWN:
+      f->dump_string("op", "unknown");
+      break;
+    case CLS_RGW_OP_LINK_OLH:
+      f->dump_string("op", "link_olh");
+      break;
+    case CLS_RGW_OP_LINK_OLH_DM:
+      f->dump_string("op", "link_olh_del");
+      break;
+    case CLS_RGW_OP_UNLINK_INSTANCE:
+      f->dump_string("op", "unlink_instance");
+      break;
+    case CLS_RGW_OP_SYNCSTOP:
+      f->dump_string("op", "syncstop");
+      break;
+    case CLS_RGW_OP_RESYNC:
+      f->dump_string("op", "resync");
+      break;
+    default:
+      f->dump_string("op", "invalid");
+      break;
+  }
+
+  f->dump_string("object", object);
+  f->dump_string("instance", instance);
+
+  switch (state) {
+    case CLS_RGW_STATE_PENDING_MODIFY:
+      f->dump_string("state", "pending");
+      break;
+    case CLS_RGW_STATE_COMPLETE:
+      f->dump_string("state", "complete");
+      break;
+    default:
+      f->dump_string("state", "invalid");
+      break;
+  }
+
+  f->dump_int("index_ver", index_ver);
+  utime_t ut(timestamp);
+  ut.gmtime_nsec(f->dump_stream("timestamp"));
+  f->open_object_section("ver");
+  ver.dump(f);
+  f->close_section();
+  f->dump_int("bilog_flags", bilog_flags);
+  f->dump_bool("versioned", (bilog_flags & RGW_BILOG_FLAG_VERSIONED_OP) != 0);
+  f->dump_string("owner", owner);
+  f->dump_string("owner_display_name", owner_display_name);
+  encode_json("zones_trace", zones_trace, f);
+}
+
+void rgw_bi_log_entry::generate_test_instances(list<rgw_bi_log_entry*>& ls)
+{
+  ls.push_back(new rgw_bi_log_entry);
+  ls.push_back(new rgw_bi_log_entry);
+  ls.back()->id = "midf";
+  ls.back()->object = "obj";
+  ls.back()->timestamp = ceph::real_clock::from_ceph_timespec({init_le32(2), init_le32(3)});
+  ls.back()->index_ver = 4323;
+  ls.back()->tag = "tagasdfds";
+  ls.back()->op = CLS_RGW_OP_DEL;
+  ls.back()->state = CLS_RGW_STATE_PENDING_MODIFY;
+}
+
+void rgw_bucket_category_stats::generate_test_instances(list<rgw_bucket_category_stats*>& o)
+{
+  rgw_bucket_category_stats *s = new rgw_bucket_category_stats;
+  s->total_size = 1024;
+  s->total_size_rounded = 4096;
+  s->num_entries = 2;
+  s->actual_size = 1024;
+  o.push_back(s);
+  o.push_back(new rgw_bucket_category_stats);
+}
+
+void rgw_bucket_category_stats::dump(Formatter *f) const
+{
+  f->dump_unsigned("total_size", total_size);
+  f->dump_unsigned("total_size_rounded", total_size_rounded);
+  f->dump_unsigned("num_entries", num_entries);
+  f->dump_unsigned("actual_size", actual_size);
+}
+
+void rgw_bucket_dir_header::generate_test_instances(list<rgw_bucket_dir_header*>& o)
+{
+  list<rgw_bucket_category_stats *> l;
+  list<rgw_bucket_category_stats *>::iterator iter;
+  rgw_bucket_category_stats::generate_test_instances(l);
+
+  uint8_t i;
+  for (i = 0, iter = l.begin(); iter != l.end(); ++iter, ++i) {
+    RGWObjCategory c = static_cast<RGWObjCategory>(i);
+    rgw_bucket_dir_header *h = new rgw_bucket_dir_header;
+    rgw_bucket_category_stats *s = *iter;
+    h->stats[c] = *s;
+
+    o.push_back(h);
+
+    delete s;
+  }
+
+  o.push_back(new rgw_bucket_dir_header);
+}
+
+void rgw_bucket_dir_header::dump(Formatter *f) const
+{
+  f->dump_int("ver", ver);
+  f->dump_int("master_ver", master_ver);
+  f->open_array_section("stats");
+  for (auto iter = stats.begin(); iter != stats.end(); ++iter) {
+    f->dump_int("category", int(iter->first));
+    f->open_object_section("category_stats");
+    iter->second.dump(f);
+    f->close_section();
+  }
+  f->close_section();
+  ::encode_json("new_instance", new_instance, f);
+}
+
+void rgw_bucket_dir::generate_test_instances(list<rgw_bucket_dir*>& o)
+{
+  list<rgw_bucket_dir_header *> l;
+  list<rgw_bucket_dir_header *>::iterator iter;
+  rgw_bucket_dir_header::generate_test_instances(l);
+
+  uint8_t i;
+  for (i = 0, iter = l.begin(); iter != l.end(); ++iter, ++i) {
+    rgw_bucket_dir *d = new rgw_bucket_dir;
+    rgw_bucket_dir_header *h = *iter;
+    d->header = *h;
+
+    list<rgw_bucket_dir_entry *> el;
+    list<rgw_bucket_dir_entry *>::iterator eiter;
+    for (eiter = el.begin(); eiter != el.end(); ++eiter) {
+      rgw_bucket_dir_entry *e = *eiter;
+      d->m[e->key.name] = *e;
+
+      delete e;
+    }
+
+    o.push_back(d);
+
+    delete h;
+  }
+
+  o.push_back(new rgw_bucket_dir);
+}
+
+void rgw_bucket_dir::dump(Formatter *f) const
+{
+  f->open_object_section("header");
+  header.dump(f);
+  f->close_section();
+  map<string, rgw_bucket_dir_entry>::const_iterator iter = m.begin();
+  f->open_array_section("map");
+  for (; iter != m.end(); ++iter) {
+    f->dump_string("key", iter->first);
+    f->open_object_section("dir_entry");
+    iter->second.dump(f);
+    f->close_section();
+  }
+  f->close_section();
+}
+
+void rgw_usage_log_entry::dump(Formatter *f) const
+{
+  f->dump_string("owner", owner.to_str());
+  f->dump_string("payer", payer.to_str());
+  f->dump_string("bucket", bucket);
+  f->dump_unsigned("epoch", epoch);
+
+  f->open_object_section("total_usage");
+  f->dump_unsigned("bytes_sent", total_usage.bytes_sent);
+  f->dump_unsigned("bytes_received", total_usage.bytes_received);
+  f->dump_unsigned("ops", total_usage.ops);
+  f->dump_unsigned("successful_ops", total_usage.successful_ops);
+  f->close_section();
+
+  f->open_array_section("categories");
+  if (usage_map.size() > 0) {
+    map<string, rgw_usage_data>::const_iterator it;
+    for (it = usage_map.begin(); it != usage_map.end(); it++) {
+      const rgw_usage_data& total_usage = it->second;
+      f->open_object_section("entry");
+      f->dump_string("category", it->first.c_str());
+      f->dump_unsigned("bytes_sent", total_usage.bytes_sent);
+      f->dump_unsigned("bytes_received", total_usage.bytes_received);
+      f->dump_unsigned("ops", total_usage.ops);
+      f->dump_unsigned("successful_ops", total_usage.successful_ops);
+      f->close_section();
+    }
+  }
+  f->close_section();
+}
+
+void rgw_usage_log_entry::generate_test_instances(list<rgw_usage_log_entry *> &o)
+{
+  rgw_usage_log_entry *entry = new rgw_usage_log_entry;
+  rgw_usage_data usage_data{1024, 2048};
+  entry->owner = rgw_user("owner");
+  entry->payer = rgw_user("payer");
+  entry->bucket = "bucket";
+  entry->epoch = 1234;
+  entry->total_usage.bytes_sent = usage_data.bytes_sent;
+  entry->total_usage.bytes_received = usage_data.bytes_received;
+  entry->total_usage.ops = usage_data.ops;
+  entry->total_usage.successful_ops = usage_data.successful_ops;
+  entry->usage_map["get_obj"] = usage_data;
+  o.push_back(entry);
+  o.push_back(new rgw_usage_log_entry);
+}
+
+void cls_rgw_reshard_entry::generate_key(const string& tenant, const string& bucket_name, string *key)
+{
+  *key = tenant + ":" + bucket_name;
+}
+
+void cls_rgw_reshard_entry::get_key(string *key) const
+{
+  generate_key(tenant, bucket_name, key);
+}
+
+void cls_rgw_reshard_entry::dump(Formatter *f) const
+{
+  utime_t ut(time);
+  encode_json("time",ut, f);
+  encode_json("tenant", tenant, f);
+  encode_json("bucket_name", bucket_name, f);
+  encode_json("bucket_id", bucket_id, f);
+  encode_json("new_instance_id", new_instance_id, f);
+  encode_json("old_num_shards", old_num_shards, f);
+  encode_json("new_num_shards", new_num_shards, f);
+
+}
+
+void cls_rgw_reshard_entry::generate_test_instances(list<cls_rgw_reshard_entry*>& ls)
+{
+  ls.push_back(new cls_rgw_reshard_entry);
+  ls.push_back(new cls_rgw_reshard_entry);
+  ls.back()->time = ceph::real_clock::from_ceph_timespec({init_le32(2), init_le32(3)});
+  ls.back()->tenant = "tenant";
+  ls.back()->bucket_name = "bucket1""";
+  ls.back()->bucket_id = "bucket_id";
+  ls.back()->new_instance_id = "new_instance_id";
+  ls.back()->old_num_shards = 8;
+  ls.back()->new_num_shards = 64;
+}
+
+void cls_rgw_bucket_instance_entry::dump(Formatter *f) const
+{
+  encode_json("reshard_status", to_string(reshard_status), f);
+  encode_json("new_bucket_instance_id", new_bucket_instance_id, f);
+  encode_json("num_shards", num_shards, f);
+
+}
+
+void cls_rgw_bucket_instance_entry::generate_test_instances(list<cls_rgw_bucket_instance_entry*>& ls)
+{
+  ls.push_back(new cls_rgw_bucket_instance_entry);
+  ls.push_back(new cls_rgw_bucket_instance_entry);
+  ls.back()->reshard_status = CLS_RGW_RESHARD_IN_PROGRESS;
+  ls.back()->new_bucket_instance_id = "new_instance_id";
+}
+  
+void cls_rgw_lc_obj_head::dump(Formatter *f) const 
+{
+  encode_json("start_date", start_date, f);
+  encode_json("marker", marker, f);
+}
+
+void cls_rgw_lc_obj_head::generate_test_instances(list<cls_rgw_lc_obj_head*>& ls)
+{
+}
diff --git a/src/cls/rgw/cls_rgw_types.h b/src/cls/rgw/cls_rgw_types.h
new file mode 100644
index 00000000..01dbe192
--- /dev/null
+++ b/src/cls/rgw/cls_rgw_types.h
@@ -0,0 +1,1183 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_RGW_TYPES_H
+#define CEPH_CLS_RGW_TYPES_H
+
+#include "common/ceph_time.h"
+#include "common/Formatter.h"
+
+#include "rgw/rgw_basic_types.h"
+
+#define CEPH_RGW_REMOVE 'r'
+#define CEPH_RGW_UPDATE 'u'
+#define CEPH_RGW_TAG_TIMEOUT 120
+#define CEPH_RGW_DIR_SUGGEST_LOG_OP  0x80
+#define CEPH_RGW_DIR_SUGGEST_OP_MASK 0x7f
+
+class JSONObj;
+
+namespace ceph {
+  class Formatter;
+}
+using ceph::operator <<;
+
+using rgw_zone_set = std::set<std::string>;
+
+enum RGWPendingState {
+  CLS_RGW_STATE_PENDING_MODIFY = 0,
+  CLS_RGW_STATE_COMPLETE       = 1,
+  CLS_RGW_STATE_UNKNOWN        = 2,
+};
+
+enum RGWModifyOp {
+  CLS_RGW_OP_ADD     = 0,
+  CLS_RGW_OP_DEL     = 1,
+  CLS_RGW_OP_CANCEL  = 2,
+  CLS_RGW_OP_UNKNOWN = 3,
+  CLS_RGW_OP_LINK_OLH        = 4,
+  CLS_RGW_OP_LINK_OLH_DM     = 5, /* creation of delete marker */
+  CLS_RGW_OP_UNLINK_INSTANCE = 6,
+  CLS_RGW_OP_SYNCSTOP        = 7,
+  CLS_RGW_OP_RESYNC          = 8,
+};
+
+enum RGWBILogFlags {
+  RGW_BILOG_FLAG_VERSIONED_OP = 0x1,
+};
+
+enum RGWCheckMTimeType {
+  CLS_RGW_CHECK_TIME_MTIME_EQ = 0,
+  CLS_RGW_CHECK_TIME_MTIME_LT = 1,
+  CLS_RGW_CHECK_TIME_MTIME_LE = 2,
+  CLS_RGW_CHECK_TIME_MTIME_GT = 3,
+  CLS_RGW_CHECK_TIME_MTIME_GE = 4,
+};
+
+#define ROUND_BLOCK_SIZE 4096
+
+static inline uint64_t cls_rgw_get_rounded_size(uint64_t size)
+{
+  return (size + ROUND_BLOCK_SIZE - 1) & ~(ROUND_BLOCK_SIZE - 1);
+}
+
+struct rgw_bucket_pending_info {
+  RGWPendingState state;
+  ceph::real_time timestamp;
+  uint8_t op;
+
+  rgw_bucket_pending_info() : state(CLS_RGW_STATE_PENDING_MODIFY), op(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(2, 2, bl);
+    uint8_t s = (uint8_t)state;
+    encode(s, bl);
+    encode(timestamp, bl);
+    encode(op, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+    uint8_t s;
+    decode(s, bl);
+    state = (RGWPendingState)s;
+    decode(timestamp, bl);
+    decode(op, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+  static void generate_test_instances(list<rgw_bucket_pending_info*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_pending_info)
+
+
+// categories of objects stored in a bucket index (b-i) and used to
+// differentiate their associated statistics (bucket stats, and in
+// some cases user stats)
+enum class RGWObjCategory : uint8_t {
+  None      = 0,  // b-i entries for delete markers; also used in
+                  // testing and for default values in default
+                  // constructors
+
+  Main      = 1,  // b-i entries for standard objs
+
+  Shadow    = 2,  // presumfably intended for multipart shadow
+                  // uploads; not currently used in the codebase
+
+  MultiMeta = 3,  // b-i entries for multipart upload metadata objs
+};
+
+
+struct rgw_bucket_dir_entry_meta {
+  RGWObjCategory category;
+  uint64_t size;
+  ceph::real_time mtime;
+  string etag;
+  string owner;
+  string owner_display_name;
+  string content_type;
+  uint64_t accounted_size;
+  string user_data;
+  string storage_class;
+  bool appendable;
+
+  rgw_bucket_dir_entry_meta() :
+    category(RGWObjCategory::None), size(0), accounted_size(0), appendable(false) { }
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(7, 3, bl);
+    encode(category, bl);
+    encode(size, bl);
+    encode(mtime, bl);
+    encode(etag, bl);
+    encode(owner, bl);
+    encode(owner_display_name, bl);
+    encode(content_type, bl);
+    encode(accounted_size, bl);
+    encode(user_data, bl);
+    encode(storage_class, bl);
+    encode(appendable, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(6, 3, 3, bl);
+    decode(category, bl);
+    decode(size, bl);
+    decode(mtime, bl);
+    decode(etag, bl);
+    decode(owner, bl);
+    decode(owner_display_name, bl);
+    if (struct_v >= 2)
+      decode(content_type, bl);
+    if (struct_v >= 4)
+      decode(accounted_size, bl);
+    else
+      accounted_size = size;
+    if (struct_v >= 5)
+      decode(user_data, bl);
+    if (struct_v >= 6)
+      decode(storage_class, bl);
+    if (struct_v >= 7)
+      decode(appendable, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+  static void generate_test_instances(list<rgw_bucket_dir_entry_meta*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_dir_entry_meta)
+
+template<class T>
+void encode_packed_val(T val, bufferlist& bl)
+{
+  using ceph::encode;
+  if ((uint64_t)val < 0x80) {
+    encode((uint8_t)val, bl);
+  } else {
+    unsigned char c = 0x80;
+
+    if ((uint64_t)val < 0x100) {
+      c |= 1;
+      encode(c, bl);
+      encode((uint8_t)val, bl);
+    } else if ((uint64_t)val <= 0x10000) {
+      c |= 2;
+      encode(c, bl);
+      encode((uint16_t)val, bl);
+    } else if ((uint64_t)val <= 0x1000000) {
+      c |= 4;
+      encode(c, bl);
+      encode((uint32_t)val, bl);
+    } else {
+      c |= 8;
+      encode(c, bl);
+      encode((uint64_t)val, bl);
+    }
+  }
+}
+
+template<class T>
+void decode_packed_val(T& val, bufferlist::const_iterator& bl)
+{
+  using ceph::decode;
+  unsigned char c;
+  decode(c, bl);
+  if (c < 0x80) {
+    val = c;
+    return;
+  }
+
+  c &= ~0x80;
+
+  switch (c) {
+    case 1:
+      {
+        uint8_t v;
+        decode(v, bl);
+        val = v;
+      }
+      break;
+    case 2:
+      {
+        uint16_t v;
+        decode(v, bl);
+        val = v;
+      }
+      break;
+    case 4:
+      {
+        uint32_t v;
+        decode(v, bl);
+        val = v;
+      }
+      break;
+    case 8:
+      {
+        uint64_t v;
+        decode(v, bl);
+        val = v;
+      }
+      break;
+    default:
+      throw buffer::error();
+  }
+}
+
+struct rgw_bucket_entry_ver {
+  int64_t pool;
+  uint64_t epoch;
+
+  rgw_bucket_entry_ver() : pool(-1), epoch(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode_packed_val(pool, bl);
+    encode_packed_val(epoch, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode_packed_val(pool, bl);
+    decode_packed_val(epoch, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+  static void generate_test_instances(list<rgw_bucket_entry_ver*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_entry_ver)
+
+struct cls_rgw_obj_key {
+  string name;
+  string instance;
+
+  cls_rgw_obj_key() {}
+  cls_rgw_obj_key(const string &_name) : name(_name) {}
+  cls_rgw_obj_key(const string& n, const string& i) : name(n), instance(i) {}
+
+  void set(const string& _name) {
+    name = _name;
+  }
+
+  bool operator==(const cls_rgw_obj_key& k) const {
+    return (name.compare(k.name) == 0) &&
+           (instance.compare(k.instance) == 0);
+  }
+  bool operator<(const cls_rgw_obj_key& k) const {
+    int r = name.compare(k.name);
+    if (r == 0) {
+      r = instance.compare(k.instance);
+    }
+    return (r < 0);
+  }
+  bool operator>(const cls_rgw_obj_key& k) const {
+    return k < *this;
+  }
+  bool operator<=(const cls_rgw_obj_key& k) const {
+    return !(k < *this);
+  }
+  bool operator>=(const cls_rgw_obj_key& k) const {
+    return !(k > *this);
+  }
+  bool empty() const {
+    return name.empty();
+  }
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(name, bl);
+    encode(instance, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(name, bl);
+    decode(instance, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const {
+    f->dump_string("name", name);
+    f->dump_string("instance", instance);
+  }
+  void decode_json(JSONObj *obj);
+  static void generate_test_instances(list<cls_rgw_obj_key*>& ls) {
+    ls.push_back(new cls_rgw_obj_key);
+    ls.push_back(new cls_rgw_obj_key);
+    ls.back()->name = "name";
+    ls.back()->instance = "instance";
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_obj_key)
+
+
+#define RGW_BUCKET_DIRENT_FLAG_VER           0x1    /* a versioned object instance */
+#define RGW_BUCKET_DIRENT_FLAG_CURRENT       0x2    /* the last object instance of a versioned object */
+#define RGW_BUCKET_DIRENT_FLAG_DELETE_MARKER 0x4    /* delete marker */
+#define RGW_BUCKET_DIRENT_FLAG_VER_MARKER    0x8    /* object is versioned, a placeholder for the plain entry */
+
+struct rgw_bucket_dir_entry {
+  cls_rgw_obj_key key;
+  rgw_bucket_entry_ver ver;
+  std::string locator;
+  bool exists;
+  rgw_bucket_dir_entry_meta meta;
+  multimap<string, rgw_bucket_pending_info> pending_map;
+  uint64_t index_ver;
+  string tag;
+  uint16_t flags;
+  uint64_t versioned_epoch;
+
+  rgw_bucket_dir_entry() :
+    exists(false), index_ver(0), flags(0), versioned_epoch(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(8, 3, bl);
+    encode(key.name, bl);
+    encode(ver.epoch, bl);
+    encode(exists, bl);
+    encode(meta, bl);
+    encode(pending_map, bl);
+    encode(locator, bl);
+    encode(ver, bl);
+    encode_packed_val(index_ver, bl);
+    encode(tag, bl);
+    encode(key.instance, bl);
+    encode(flags, bl);
+    encode(versioned_epoch, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(8, 3, 3, bl);
+    decode(key.name, bl);
+    decode(ver.epoch, bl);
+    decode(exists, bl);
+    decode(meta, bl);
+    decode(pending_map, bl);
+    if (struct_v >= 2) {
+      decode(locator, bl);
+    }
+    if (struct_v >= 4) {
+      decode(ver, bl);
+    } else {
+      ver.pool = -1;
+    }
+    if (struct_v >= 5) {
+      decode_packed_val(index_ver, bl);
+      decode(tag, bl);
+    }
+    if (struct_v >= 6) {
+      decode(key.instance, bl);
+    }
+    if (struct_v >= 7) {
+      decode(flags, bl);
+    }
+    if (struct_v >= 8) {
+      decode(versioned_epoch, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  bool is_current() const {
+    int test_flags = RGW_BUCKET_DIRENT_FLAG_VER | RGW_BUCKET_DIRENT_FLAG_CURRENT;
+    return (flags & RGW_BUCKET_DIRENT_FLAG_VER) == 0 ||
+           (flags & test_flags) == test_flags;
+  }
+  bool is_delete_marker() const {
+    return (flags & RGW_BUCKET_DIRENT_FLAG_DELETE_MARKER) != 0;
+  }
+  bool is_visible() const {
+    return is_current() && !is_delete_marker();
+  }
+  bool is_valid() const { return (flags & RGW_BUCKET_DIRENT_FLAG_VER_MARKER) == 0; }
+
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+  static void generate_test_instances(list<rgw_bucket_dir_entry*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_dir_entry)
+
+enum class BIIndexType : uint8_t {
+  Invalid    = 0,
+  Plain      = 1,
+  Instance   = 2,
+  OLH        = 3,
+};
+
+struct rgw_bucket_category_stats;
+
+struct rgw_cls_bi_entry {
+  BIIndexType type;
+  string idx;
+  bufferlist data;
+
+  rgw_cls_bi_entry() : type(BIIndexType::Invalid) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(type, bl);
+    encode(idx, bl);
+    encode(data, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    uint8_t c;
+    decode(c, bl);
+    type = (BIIndexType)c;
+    decode(idx, bl);
+    decode(data, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj, cls_rgw_obj_key *effective_key = NULL);
+
+  bool get_info(cls_rgw_obj_key *key, RGWObjCategory *category,
+		rgw_bucket_category_stats *accounted_stats);
+};
+WRITE_CLASS_ENCODER(rgw_cls_bi_entry)
+
+enum OLHLogOp {
+  CLS_RGW_OLH_OP_UNKNOWN         = 0,
+  CLS_RGW_OLH_OP_LINK_OLH        = 1,
+  CLS_RGW_OLH_OP_UNLINK_OLH      = 2, /* object does not exist */
+  CLS_RGW_OLH_OP_REMOVE_INSTANCE = 3,
+};
+
+struct rgw_bucket_olh_log_entry {
+  uint64_t epoch;
+  OLHLogOp op;
+  string op_tag;
+  cls_rgw_obj_key key;
+  bool delete_marker;
+
+  rgw_bucket_olh_log_entry() : epoch(0), op(CLS_RGW_OLH_OP_UNKNOWN), delete_marker(false) {}
+
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(epoch, bl);
+    encode((__u8)op, bl);
+    encode(op_tag, bl);
+    encode(key, bl);
+    encode(delete_marker, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(epoch, bl);
+    uint8_t c;
+    decode(c, bl);
+    op = (OLHLogOp)c;
+    decode(op_tag, bl);
+    decode(key, bl);
+    decode(delete_marker, bl);
+    DECODE_FINISH(bl);
+  }
+  static void generate_test_instances(list<rgw_bucket_olh_log_entry*>& o);
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_olh_log_entry)
+
+struct rgw_bucket_olh_entry {
+  cls_rgw_obj_key key;
+  bool delete_marker;
+  uint64_t epoch;
+  map<uint64_t, vector<struct rgw_bucket_olh_log_entry> > pending_log;
+  string tag;
+  bool exists;
+  bool pending_removal;
+
+  rgw_bucket_olh_entry() : delete_marker(false), epoch(0), exists(false), pending_removal(false) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(key, bl);
+    encode(delete_marker, bl);
+    encode(epoch, bl);
+    encode(pending_log, bl);
+    encode(tag, bl);
+    encode(exists, bl);
+    encode(pending_removal, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(1, bl);
+    decode(key, bl);
+    decode(delete_marker, bl);
+    decode(epoch, bl);
+    decode(pending_log, bl);
+    decode(tag, bl);
+    decode(exists, bl);
+    decode(pending_removal, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_olh_entry)
+
+struct rgw_bi_log_entry {
+  string id;
+  string object;
+  string instance;
+  ceph::real_time timestamp;
+  rgw_bucket_entry_ver ver;
+  RGWModifyOp op;
+  RGWPendingState state;
+  uint64_t index_ver;
+  string tag;
+  uint16_t bilog_flags;
+  string owner; /* only being set if it's a delete marker */
+  string owner_display_name; /* only being set if it's a delete marker */
+  rgw_zone_set zones_trace;
+
+  rgw_bi_log_entry() : op(CLS_RGW_OP_UNKNOWN), state(CLS_RGW_STATE_PENDING_MODIFY), index_ver(0), bilog_flags(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(4, 1, bl);
+    encode(id, bl);
+    encode(object, bl);
+    encode(timestamp, bl);
+    encode(ver, bl);
+    encode(tag, bl);
+    uint8_t c = (uint8_t)op;
+    encode(c, bl);
+    c = (uint8_t)state;
+    encode(c, bl);
+    encode_packed_val(index_ver, bl);
+    encode(instance, bl);
+    encode(bilog_flags, bl);
+    encode(owner, bl);
+    encode(owner_display_name, bl);
+    encode(zones_trace, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START(4, bl);
+    decode(id, bl);
+    decode(object, bl);
+    decode(timestamp, bl);
+    decode(ver, bl);
+    decode(tag, bl);
+    uint8_t c;
+    decode(c, bl);
+    op = (RGWModifyOp)c;
+    decode(c, bl);
+    state = (RGWPendingState)c;
+    decode_packed_val(index_ver, bl);
+    if (struct_v >= 2) {
+      decode(instance, bl);
+      decode(bilog_flags, bl);
+    }
+    if (struct_v >= 3) {
+      decode(owner, bl);
+      decode(owner_display_name, bl);
+    }
+    if (struct_v >= 4) {
+      decode(zones_trace, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+  static void generate_test_instances(list<rgw_bi_log_entry*>& o);
+
+  bool is_versioned() {
+    return ((bilog_flags & RGW_BILOG_FLAG_VERSIONED_OP) != 0);
+  }
+};
+WRITE_CLASS_ENCODER(rgw_bi_log_entry)
+
+struct rgw_bucket_category_stats {
+  uint64_t total_size;
+  uint64_t total_size_rounded;
+  uint64_t num_entries;
+  uint64_t actual_size{0}; //< account for compression, encryption
+
+  rgw_bucket_category_stats() : total_size(0), total_size_rounded(0), num_entries(0) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(3, 2, bl);
+    encode(total_size, bl);
+    encode(total_size_rounded, bl);
+    encode(num_entries, bl);
+    encode(actual_size, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(3, 2, 2, bl);
+    decode(total_size, bl);
+    decode(total_size_rounded, bl);
+    decode(num_entries, bl);
+    if (struct_v >= 3) {
+      decode(actual_size, bl);
+    } else {
+      actual_size = total_size;
+    }
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_bucket_category_stats*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_category_stats)
+
+enum cls_rgw_reshard_status {
+  CLS_RGW_RESHARD_NOT_RESHARDING  = 0,
+  CLS_RGW_RESHARD_IN_PROGRESS     = 1,
+  CLS_RGW_RESHARD_DONE            = 2,
+};
+
+static inline std::string to_string(const enum cls_rgw_reshard_status status)
+{
+  switch (status) {
+  case CLS_RGW_RESHARD_NOT_RESHARDING:
+    return "not-resharding";
+    break;
+  case CLS_RGW_RESHARD_IN_PROGRESS:
+    return "in-progress";
+    break;
+  case CLS_RGW_RESHARD_DONE:
+    return "done";
+    break;
+  default:
+    break;
+  };
+  return "Unknown reshard status";
+}
+
+struct cls_rgw_bucket_instance_entry {
+  cls_rgw_reshard_status reshard_status{CLS_RGW_RESHARD_NOT_RESHARDING};
+  string new_bucket_instance_id;
+  int32_t num_shards{-1};
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode((uint8_t)reshard_status, bl);
+    encode(new_bucket_instance_id, bl);
+    encode(num_shards, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    uint8_t s;
+    decode(s, bl);
+    reshard_status = (cls_rgw_reshard_status)s;
+    decode(new_bucket_instance_id, bl);
+    decode(num_shards, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_bucket_instance_entry*>& o);
+
+  void clear() {
+    reshard_status = CLS_RGW_RESHARD_NOT_RESHARDING;
+    new_bucket_instance_id.clear();
+  }
+
+  void set_status(const string& new_instance_id, int32_t new_num_shards, cls_rgw_reshard_status s) {
+    reshard_status = s;
+    new_bucket_instance_id = new_instance_id;
+    num_shards = new_num_shards;
+  }
+
+  bool resharding() const {
+    return reshard_status != CLS_RGW_RESHARD_NOT_RESHARDING;
+  }
+  bool resharding_in_progress() const {
+    return reshard_status == CLS_RGW_RESHARD_IN_PROGRESS;
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_bucket_instance_entry)
+
+struct rgw_bucket_dir_header {
+  map<RGWObjCategory, rgw_bucket_category_stats> stats;
+  uint64_t tag_timeout;
+  uint64_t ver;
+  uint64_t master_ver;
+  string max_marker;
+  cls_rgw_bucket_instance_entry new_instance;
+  bool syncstopped;
+
+  rgw_bucket_dir_header() : tag_timeout(0), ver(0), master_ver(0), syncstopped(false) {}
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(7, 2, bl);
+    encode(stats, bl);
+    encode(tag_timeout, bl);
+    encode(ver, bl);
+    encode(master_ver, bl);
+    encode(max_marker, bl);
+    encode(new_instance, bl);
+    encode(syncstopped,bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(6, 2, 2, bl);
+    decode(stats, bl);
+    if (struct_v > 2) {
+      decode(tag_timeout, bl);
+    } else {
+      tag_timeout = 0;
+    }
+    if (struct_v >= 4) {
+      decode(ver, bl);
+      decode(master_ver, bl);
+    } else {
+      ver = 0;
+    }
+    if (struct_v >= 5) {
+      decode(max_marker, bl);
+    }
+    if (struct_v >= 6) {
+      decode(new_instance, bl);
+    } else {
+      new_instance = cls_rgw_bucket_instance_entry();
+    }
+    if (struct_v >= 7) {
+      decode(syncstopped,bl);
+    }
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_bucket_dir_header*>& o);
+
+  bool resharding() const {
+    return new_instance.resharding();
+  }
+  bool resharding_in_progress() const {
+    return new_instance.resharding_in_progress();
+  }
+};
+WRITE_CLASS_ENCODER(rgw_bucket_dir_header)
+
+struct rgw_bucket_dir {
+  rgw_bucket_dir_header header;
+  std::map<string, rgw_bucket_dir_entry> m;
+
+  void encode(bufferlist &bl) const {
+    ENCODE_START(2, 2, bl);
+    encode(header, bl);
+    encode(m, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator &bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+    decode(header, bl);
+    decode(m, bl);
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<rgw_bucket_dir*>& o);
+};
+WRITE_CLASS_ENCODER(rgw_bucket_dir)
+
+struct rgw_usage_data {
+  uint64_t bytes_sent;
+  uint64_t bytes_received;
+  uint64_t ops;
+  uint64_t successful_ops;
+
+  rgw_usage_data() : bytes_sent(0), bytes_received(0), ops(0), successful_ops(0) {}
+  rgw_usage_data(uint64_t sent, uint64_t received) : bytes_sent(sent), bytes_received(received), ops(0), successful_ops(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(bytes_sent, bl);
+    encode(bytes_received, bl);
+    encode(ops, bl);
+    encode(successful_ops, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(bytes_sent, bl);
+    decode(bytes_received, bl);
+    decode(ops, bl);
+    decode(successful_ops, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void aggregate(const rgw_usage_data& usage) {
+    bytes_sent += usage.bytes_sent;
+    bytes_received += usage.bytes_received;
+    ops += usage.ops;
+    successful_ops += usage.successful_ops;
+  }
+};
+WRITE_CLASS_ENCODER(rgw_usage_data)
+
+
+struct rgw_usage_log_entry {
+  rgw_user owner;
+  rgw_user payer; /* if empty, same as owner */
+  string bucket;
+  uint64_t epoch;
+  rgw_usage_data total_usage; /* this one is kept for backwards compatibility */
+  map<string, rgw_usage_data> usage_map;
+
+  rgw_usage_log_entry() : epoch(0) {}
+  rgw_usage_log_entry(string& o, string& b) : owner(o), bucket(b), epoch(0) {}
+  rgw_usage_log_entry(string& o, string& p, string& b) : owner(o), payer(p), bucket(b), epoch(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(3, 1, bl);
+    encode(owner.to_str(), bl);
+    encode(bucket, bl);
+    encode(epoch, bl);
+    encode(total_usage.bytes_sent, bl);
+    encode(total_usage.bytes_received, bl);
+    encode(total_usage.ops, bl);
+    encode(total_usage.successful_ops, bl);
+    encode(usage_map, bl);
+    encode(payer.to_str(), bl);
+    ENCODE_FINISH(bl);
+  }
+
+
+   void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(3, bl);
+    string s;
+    decode(s, bl);
+    owner.from_str(s);
+    decode(bucket, bl);
+    decode(epoch, bl);
+    decode(total_usage.bytes_sent, bl);
+    decode(total_usage.bytes_received, bl);
+    decode(total_usage.ops, bl);
+    decode(total_usage.successful_ops, bl);
+    if (struct_v < 2) {
+      usage_map[""] = total_usage;
+    } else {
+      decode(usage_map, bl);
+    }
+    if (struct_v >= 3) {
+      string p;
+      decode(p, bl);
+      payer.from_str(p);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  void aggregate(const rgw_usage_log_entry& e, map<string, bool> *categories = NULL) {
+    if (owner.empty()) {
+      owner = e.owner;
+      bucket = e.bucket;
+      epoch = e.epoch;
+      payer = e.payer;
+    }
+
+    map<string, rgw_usage_data>::const_iterator iter;
+    for (iter = e.usage_map.begin(); iter != e.usage_map.end(); ++iter) {
+      if (!categories || !categories->size() || categories->count(iter->first)) {
+        add(iter->first, iter->second);
+      }
+    }
+  }
+
+  void sum(rgw_usage_data& usage, map<string, bool>& categories) const {
+    usage = rgw_usage_data();
+    for (map<string, rgw_usage_data>::const_iterator iter = usage_map.begin(); iter != usage_map.end(); ++iter) {
+      if (!categories.size() || categories.count(iter->first)) {
+        usage.aggregate(iter->second);
+      }
+    }
+  }
+
+  void add(const string& category, const rgw_usage_data& data) {
+    usage_map[category].aggregate(data);
+    total_usage.aggregate(data);
+  }
+
+  void dump(Formatter* f) const;
+  static void generate_test_instances(list<rgw_usage_log_entry*>& o);
+
+};
+WRITE_CLASS_ENCODER(rgw_usage_log_entry)
+
+struct rgw_usage_log_info {
+  vector<rgw_usage_log_entry> entries;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    DECODE_FINISH(bl);
+  }
+
+  rgw_usage_log_info() {}
+};
+WRITE_CLASS_ENCODER(rgw_usage_log_info)
+
+struct rgw_user_bucket {
+  string user;
+  string bucket;
+
+  rgw_user_bucket() {}
+  rgw_user_bucket(const string& u, const string& b) : user(u), bucket(b) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(user, bl);
+    encode(bucket, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(user, bl);
+    decode(bucket, bl);
+    DECODE_FINISH(bl);
+  }
+
+  bool operator<(const rgw_user_bucket& ub2) const {
+    int comp = user.compare(ub2.user);
+    if (comp < 0)
+      return true;
+    else if (!comp)
+      return bucket.compare(ub2.bucket) < 0;
+
+    return false;
+  }
+};
+WRITE_CLASS_ENCODER(rgw_user_bucket)
+
+enum cls_rgw_gc_op {
+  CLS_RGW_GC_DEL_OBJ,
+  CLS_RGW_GC_DEL_BUCKET,
+};
+
+struct cls_rgw_obj {
+  string pool;
+  cls_rgw_obj_key key;
+  string loc;
+
+  cls_rgw_obj() {}
+  cls_rgw_obj(string& _p, cls_rgw_obj_key& _k) : pool(_p), key(_k) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(pool, bl);
+    encode(key.name, bl);
+    encode(loc, bl);
+    encode(key, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(pool, bl);
+    decode(key.name, bl);
+    decode(loc, bl);
+    if (struct_v >= 2) {
+      decode(key, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const {
+    f->dump_string("pool", pool);
+    f->dump_string("oid", key.name);
+    f->dump_string("key", loc);
+    f->dump_string("instance", key.instance);
+  }
+  static void generate_test_instances(list<cls_rgw_obj*>& ls) {
+    ls.push_back(new cls_rgw_obj);
+    ls.push_back(new cls_rgw_obj);
+    ls.back()->pool = "mypool";
+    ls.back()->key.name = "myoid";
+    ls.back()->loc = "mykey";
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_obj)
+
+struct cls_rgw_obj_chain {
+  list<cls_rgw_obj> objs;
+
+  cls_rgw_obj_chain() {}
+
+  void push_obj(const string& pool, const cls_rgw_obj_key& key, const string& loc) {
+    cls_rgw_obj obj;
+    obj.pool = pool;
+    obj.key = key;
+    obj.loc = loc;
+    objs.push_back(obj);
+  }
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(objs, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(objs, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const {
+    f->open_array_section("objs");
+    for (list<cls_rgw_obj>::const_iterator p = objs.begin(); p != objs.end(); ++p) {
+      f->open_object_section("obj");
+      p->dump(f);
+      f->close_section();
+    }
+    f->close_section();
+  }
+  static void generate_test_instances(list<cls_rgw_obj_chain*>& ls) {
+    ls.push_back(new cls_rgw_obj_chain);
+  }
+
+  bool empty() {
+    return objs.empty();
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_obj_chain)
+
+struct cls_rgw_gc_obj_info
+{
+  string tag;
+  cls_rgw_obj_chain chain;
+  ceph::real_time time;
+
+  cls_rgw_gc_obj_info() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(tag, bl);
+    encode(chain, bl);
+    encode(time, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(tag, bl);
+    decode(chain, bl);
+    decode(time, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const {
+    f->dump_string("tag", tag);
+    f->open_object_section("chain");
+    chain.dump(f);
+    f->close_section();
+    f->dump_stream("time") << time;
+  }
+  static void generate_test_instances(list<cls_rgw_gc_obj_info*>& ls) {
+    ls.push_back(new cls_rgw_gc_obj_info);
+    ls.push_back(new cls_rgw_gc_obj_info);
+    ls.back()->tag = "footag";
+    ceph_timespec ts{init_le32(21), init_le32(32)};
+    ls.back()->time = ceph::real_clock::from_ceph_timespec(ts);
+  }
+};
+WRITE_CLASS_ENCODER(cls_rgw_gc_obj_info)
+
+struct cls_rgw_lc_obj_head
+{
+  time_t start_date = 0;
+  string marker;
+
+  cls_rgw_lc_obj_head() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    uint64_t t = start_date;
+    encode(t, bl);
+    encode(marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    uint64_t t;
+    decode(t, bl);
+    start_date = static_cast<time_t>(t);
+    decode(marker, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_lc_obj_head*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_rgw_lc_obj_head)
+
+struct cls_rgw_reshard_entry
+{
+  ceph::real_time time;
+  string tenant;
+  string bucket_name;
+  string bucket_id;
+  string new_instance_id;
+  uint32_t old_num_shards{0};
+  uint32_t new_num_shards{0};
+
+  cls_rgw_reshard_entry() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(time, bl);
+    encode(tenant, bl);
+    encode(bucket_name, bl);
+    encode(bucket_id, bl);
+    encode(new_instance_id, bl);
+    encode(old_num_shards, bl);
+    encode(new_num_shards, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(time, bl);
+    decode(tenant, bl);
+    decode(bucket_name, bl);
+    decode(bucket_id, bl);
+    decode(new_instance_id, bl);
+    decode(old_num_shards, bl);
+    decode(new_num_shards, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_rgw_reshard_entry*>& o);
+
+  static void generate_key(const string& tenant, const string& bucket_name, string *key);
+  void get_key(string *key) const;
+};
+WRITE_CLASS_ENCODER(cls_rgw_reshard_entry)
+
+#endif
diff --git a/src/cls/sdk/cls_sdk.cc b/src/cls/sdk/cls_sdk.cc
new file mode 100644
index 00000000..95b5096a
--- /dev/null
+++ b/src/cls/sdk/cls_sdk.cc
@@ -0,0 +1,131 @@
+/*
+ * This is an example RADOS object class built using only the Ceph SDK interface.
+ */
+#include "include/rados/objclass.h"
+
+CLS_VER(1,0)
+CLS_NAME(sdk)
+
+cls_handle_t h_class;
+cls_method_handle_t h_test_coverage_write;
+cls_method_handle_t h_test_coverage_replay;
+
+/**
+ * test_coverage_write - a "write" method that creates an object
+ *
+ * This method modifies the object by making multiple write calls (write,
+ * setxattr and set_val).
+ */
+static int test_coverage_write(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  // create the object
+  int ret = cls_cxx_create(hctx, false);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: %s(): cls_cxx_create returned %d", __func__, ret);	  
+    return ret;
+  }
+
+  uint64_t size;
+  // get the size of the object
+  ret = cls_cxx_stat(hctx, &size, NULL);
+  if (ret < 0)
+    return ret;
+  
+  std::string c = "test";
+  bufferlist bl;
+  bl.append(c);
+
+  // write to the object
+  ret = cls_cxx_write(hctx, 0, bl.length(), &bl);
+  if (ret < 0)
+    return ret;
+  
+  uint64_t new_size;
+  // get the new size of the object
+  ret = cls_cxx_stat(hctx, &new_size, NULL);
+  if (ret < 0)
+    return ret;
+
+  // make some change to the xattr
+  ret = cls_cxx_setxattr(hctx, "foo", &bl);
+  if (ret < 0)
+    return ret;
+
+  // make some change to the omap
+  ret = cls_cxx_map_set_val(hctx, "foo", &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+/**
+ * test_coverage_replay - a "read" method to retrieve previously written data
+ *
+ * This method reads the object by making multiple read calls (read, getxattr 
+ * and get_val). It also removes the object after reading.
+ */
+
+static int test_coverage_replay(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  CLS_LOG(0, "reading already written object");
+  uint64_t size;
+  // get the size of the object
+  int ret = cls_cxx_stat(hctx, &size, NULL);
+  if (ret < 0)
+    return ret;
+  
+  bufferlist bl;
+  // read the object entry
+  ret = cls_cxx_read(hctx, 0, size, &bl);
+  if (ret < 0)
+    return ret;
+
+  // if the size is incorrect 
+  if (bl.length() != size)
+    return -EIO;
+  
+  bl.clear();
+
+  // read xattr entry
+  ret = cls_cxx_getxattr(hctx, "foo", &bl);
+  if (ret < 0)
+    return ret;
+
+  // if the size is incorrect
+  if (bl.length() != size)
+    return -EIO;
+  
+  bl.clear();
+
+  // read omap entry
+  ret = cls_cxx_map_get_val(hctx, "foo", &bl);
+  if (ret < 0)
+    return ret;
+
+  // if the size is incorrect
+  if (bl.length() != size)
+    return -EIO;
+
+  // remove the object
+  ret = cls_cxx_remove(hctx);
+  if (ret < 0)
+    return ret;
+  
+  return 0;
+}
+
+CLS_INIT(sdk)
+{
+  CLS_LOG(0, "loading cls_sdk");
+
+  cls_register("sdk", &h_class);
+
+  cls_register_cxx_method(h_class, "test_coverage_write",
+      CLS_METHOD_RD|CLS_METHOD_WR,
+      test_coverage_write, &h_test_coverage_write);
+  
+  cls_register_cxx_method(h_class, "test_coverage_replay",
+      CLS_METHOD_RD|CLS_METHOD_WR,
+      test_coverage_replay, &h_test_coverage_replay);
+}
diff --git a/src/cls/timeindex/cls_timeindex.cc b/src/cls/timeindex/cls_timeindex.cc
new file mode 100644
index 00000000..92ea15be
--- /dev/null
+++ b/src/cls/timeindex/cls_timeindex.cc
@@ -0,0 +1,261 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "objclass/objclass.h"
+
+#include "cls_timeindex_ops.h"
+
+#include "include/compat.h"
+
+CLS_VER(1,0)
+CLS_NAME(timeindex)
+
+static const size_t MAX_LIST_ENTRIES = 1000;
+static const size_t MAX_TRIM_ENTRIES = 1000;
+
+static const string TIMEINDEX_PREFIX = "1_";
+
+static void get_index_time_prefix(const utime_t& ts,
+                                  string& index)
+{
+  char buf[32];
+
+  snprintf(buf, sizeof(buf), "%s%010ld.%06ld_", TIMEINDEX_PREFIX.c_str(),
+          (long)ts.sec(), (long)ts.usec());
+  buf[sizeof(buf) - 1] = '\0';
+
+  index = buf;
+}
+
+static void get_index(cls_method_context_t hctx,
+                      const utime_t& key_ts,
+                      const string& key_ext,
+                      string& index)
+{
+  get_index_time_prefix(key_ts, index);
+  index.append(key_ext);
+}
+
+static int parse_index(const string& index,
+                       utime_t& key_ts,
+                       string& key_ext)
+{
+  int sec, usec;
+  char keyext[256];
+
+  int ret = sscanf(index.c_str(), "1_%d.%d_%255s", &sec, &usec, keyext);
+
+  key_ts  = utime_t(sec, usec);
+  key_ext = string(keyext);
+  return ret;
+}
+
+static int cls_timeindex_add(cls_method_context_t hctx,
+                             bufferlist * const in,
+                             bufferlist * const out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_timeindex_add_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_timeindex_add_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  for (list<cls_timeindex_entry>::iterator iter = op.entries.begin();
+       iter != op.entries.end();
+       ++iter) {
+    cls_timeindex_entry& entry = *iter;
+
+    string index;
+    get_index(hctx, entry.key_ts, entry.key_ext, index);
+
+    CLS_LOG(20, "storing entry at %s", index.c_str());
+
+    int ret = cls_cxx_map_set_val(hctx, index, &entry.value);
+    if (ret < 0) {
+      return ret;
+    }
+  }
+
+  return 0;
+}
+
+static int cls_timeindex_list(cls_method_context_t hctx,
+                              bufferlist * const in,
+                              bufferlist * const out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_timeindex_list_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_timeindex_list_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> keys;
+
+  string from_index;
+  string to_index;
+
+  if (op.marker.empty()) {
+    get_index_time_prefix(op.from_time, from_index);
+  } else {
+    from_index = op.marker;
+  }
+  const bool use_time_boundary = (op.to_time >= op.from_time);
+
+  if (use_time_boundary) {
+    get_index_time_prefix(op.to_time, to_index);
+  }
+
+  size_t max_entries = op.max_entries;
+  if (max_entries > MAX_LIST_ENTRIES) {
+    max_entries = MAX_LIST_ENTRIES;
+  }
+
+  cls_timeindex_list_ret ret;
+
+  int rc = cls_cxx_map_get_vals(hctx, from_index, TIMEINDEX_PREFIX,
+          max_entries, &keys, &ret.truncated);
+  if (rc < 0) {
+    return rc;
+  }
+
+  list<cls_timeindex_entry>& entries = ret.entries;
+  map<string, bufferlist>::iterator iter = keys.begin();
+
+  string marker;
+
+  for (; iter != keys.end(); ++iter) {
+    const string& index = iter->first;
+    bufferlist& bl = iter->second;
+
+    if (use_time_boundary && index.compare(0, to_index.size(), to_index) >= 0) {
+      CLS_LOG(20, "DEBUG: cls_timeindex_list: finishing on to_index=%s",
+              to_index.c_str());
+      ret.truncated = false;
+      break;
+    }
+
+    cls_timeindex_entry e;
+
+    if (parse_index(index, e.key_ts, e.key_ext) < 0) {
+      CLS_LOG(0, "ERROR: cls_timeindex_list: could not parse index=%s",
+              index.c_str());
+    } else {
+      CLS_LOG(20, "DEBUG: cls_timeindex_list: index=%s, key_ext=%s, bl.len = %d",
+              index.c_str(), e.key_ext.c_str(), bl.length());
+      e.value = bl;
+      entries.push_back(e);
+    }
+    marker = index;
+  }
+
+  ret.marker = marker;
+
+  encode(ret, *out);
+
+  return 0;
+}
+
+
+static int cls_timeindex_trim(cls_method_context_t hctx,
+                              bufferlist * const in,
+                              bufferlist * const out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_timeindex_trim_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_timeindex_trim: failed to decode entry");
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> keys;
+
+  string from_index;
+  string to_index;
+
+  if (op.from_marker.empty()) {
+    get_index_time_prefix(op.from_time, from_index);
+  } else {
+    from_index = op.from_marker;
+  }
+
+  if (op.to_marker.empty()) {
+    get_index_time_prefix(op.to_time, to_index);
+  } else {
+    to_index = op.to_marker;
+  }
+
+  bool more;
+
+  int rc = cls_cxx_map_get_vals(hctx, from_index, TIMEINDEX_PREFIX,
+          MAX_TRIM_ENTRIES, &keys, &more);
+  if (rc < 0) {
+    return rc;
+  }
+
+  map<string, bufferlist>::iterator iter = keys.begin();
+
+  bool removed = false;
+  for (; iter != keys.end(); ++iter) {
+    const string& index = iter->first;
+
+    CLS_LOG(20, "index=%s to_index=%s", index.c_str(), to_index.c_str());
+
+    if (index.compare(0, to_index.size(), to_index) > 0) {
+      CLS_LOG(20, "DEBUG: cls_timeindex_trim: finishing on to_index=%s",
+              to_index.c_str());
+      break;
+    }
+
+    CLS_LOG(20, "removing key: index=%s", index.c_str());
+
+    int rc = cls_cxx_map_remove_key(hctx, index);
+    if (rc < 0) {
+      CLS_LOG(1, "ERROR: cls_cxx_map_remove_key failed rc=%d", rc);
+      return rc;
+    }
+
+    removed = true;
+  }
+
+  if (!removed) {
+    return -ENODATA;
+  }
+
+  return 0;
+}
+
+CLS_INIT(timeindex)
+{
+  CLS_LOG(1, "Loaded timeindex class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_timeindex_add;
+  cls_method_handle_t h_timeindex_list;
+  cls_method_handle_t h_timeindex_trim;
+
+  cls_register("timeindex", &h_class);
+
+  /* timeindex */
+  cls_register_cxx_method(h_class, "add", CLS_METHOD_RD | CLS_METHOD_WR,
+          cls_timeindex_add, &h_timeindex_add);
+  cls_register_cxx_method(h_class, "list", CLS_METHOD_RD,
+          cls_timeindex_list, &h_timeindex_list);
+  cls_register_cxx_method(h_class, "trim", CLS_METHOD_RD | CLS_METHOD_WR,
+          cls_timeindex_trim, &h_timeindex_trim);
+
+  return;
+}
+
diff --git a/src/cls/timeindex/cls_timeindex_client.cc b/src/cls/timeindex/cls_timeindex_client.cc
new file mode 100644
index 00000000..7a38ff5f
--- /dev/null
+++ b/src/cls/timeindex/cls_timeindex_client.cc
@@ -0,0 +1,120 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "cls/timeindex/cls_timeindex_ops.h"
+#include "cls/timeindex/cls_timeindex_client.h"
+#include "include/compat.h"
+
+void cls_timeindex_add(
+  librados::ObjectWriteOperation& op,
+  std::list<cls_timeindex_entry>& entries)
+{
+  librados::bufferlist in;
+  cls_timeindex_add_op call;
+  call.entries = entries;
+
+  encode(call, in);
+  op.exec("timeindex", "add", in);
+}
+
+void cls_timeindex_add(
+  librados::ObjectWriteOperation& op,
+  cls_timeindex_entry& entry)
+{
+  librados::bufferlist in;
+  cls_timeindex_add_op call;
+  call.entries.push_back(entry);
+
+  encode(call, in);
+  op.exec("timeindex", "add", in);
+}
+
+void cls_timeindex_add_prepare_entry(
+  cls_timeindex_entry& entry,
+  const utime_t& key_timestamp,
+  const std::string& key_ext,
+  const librados::bufferlist& bl)
+{
+  entry.key_ts = key_timestamp;
+  entry.key_ext = key_ext;
+  entry.value = bl;
+}
+
+void cls_timeindex_add(
+  librados::ObjectWriteOperation& op,
+  const utime_t& key_timestamp,
+  const std::string& key_ext,
+  const librados::bufferlist& bl)
+{
+  cls_timeindex_entry entry;
+  cls_timeindex_add_prepare_entry(entry, key_timestamp, key_ext, bl);
+  cls_timeindex_add(op, entry);
+}
+
+void cls_timeindex_trim(
+  librados::ObjectWriteOperation& op,
+  const utime_t& from_time,
+  const utime_t& to_time,
+  const std::string& from_marker,
+  const std::string& to_marker)
+{
+  librados::bufferlist in;
+  cls_timeindex_trim_op call;
+  call.from_time = from_time;
+  call.to_time = to_time;
+  call.from_marker = from_marker;
+  call.to_marker = to_marker;
+
+  encode(call, in);
+
+  op.exec("timeindex", "trim", in);
+}
+
+int cls_timeindex_trim(
+  librados::IoCtx& io_ctx,
+  const std::string& oid,
+  const utime_t& from_time,
+  const utime_t& to_time,
+  const std::string& from_marker,
+  const std::string& to_marker)
+{
+  bool done = false;
+
+  do {
+    librados::ObjectWriteOperation op;
+    cls_timeindex_trim(op, from_time, to_time, from_marker, to_marker);
+    int r = io_ctx.operate(oid, &op);
+
+    if (r == -ENODATA)
+      done = true;
+    else if (r < 0)
+      return r;
+  } while (!done);
+
+  return 0;
+}
+
+void cls_timeindex_list(
+  librados::ObjectReadOperation& op,
+  const utime_t& from,
+  const utime_t& to,
+  const std::string& in_marker,
+  const int max_entries,
+  std::list<cls_timeindex_entry>& entries,
+  std::string *out_marker,
+  bool *truncated)
+{
+  librados::bufferlist in;
+  cls_timeindex_list_op call;
+  call.from_time = from;
+  call.to_time = to;
+  call.marker = in_marker;
+  call.max_entries = max_entries;
+
+  encode(call, in);
+
+  op.exec("timeindex", "list", in,
+          new TimeindexListCtx(&entries, out_marker, truncated));
+}
diff --git a/src/cls/timeindex/cls_timeindex_client.h b/src/cls/timeindex/cls_timeindex_client.h
new file mode 100644
index 00000000..fc57a160
--- /dev/null
+++ b/src/cls/timeindex/cls_timeindex_client.h
@@ -0,0 +1,93 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_TIMEINDEX_CLIENT_H
+#define CEPH_CLS_TIMEINDEX_CLIENT_H
+
+#include "include/rados/librados.hpp"
+
+#include "cls_timeindex_ops.h"
+
+/**
+ * timeindex objclass
+ */
+class TimeindexListCtx : public librados::ObjectOperationCompletion {
+  std::list<cls_timeindex_entry> *entries;
+  std::string *marker;
+  bool *truncated;
+
+public:
+  ///* ctor
+  TimeindexListCtx(
+    std::list<cls_timeindex_entry> *_entries,
+    std::string *_marker,
+    bool *_truncated)
+    : entries(_entries), marker(_marker), truncated(_truncated) {}
+
+  ///* dtor
+  ~TimeindexListCtx() {}
+
+  void handle_completion(int r, bufferlist& bl) override {
+    if (r >= 0) {
+      cls_timeindex_list_ret ret;
+      try {
+        auto iter = bl.cbegin();
+        decode(ret, iter);
+        if (entries)
+          *entries = ret.entries;
+        if (truncated)
+          *truncated = ret.truncated;
+        if (marker)
+          *marker = ret.marker;
+      } catch (buffer::error& err) {
+        // nothing we can do about it atm
+      }
+    }
+  }
+};
+
+void cls_timeindex_add_prepare_entry(
+  cls_timeindex_entry& entry,
+  const utime_t& key_timestamp,
+  const std::string& key_ext,
+  bufferlist& bl);
+
+void cls_timeindex_add(
+  librados::ObjectWriteOperation& op,
+  const std::list<cls_timeindex_entry>& entry);
+
+void cls_timeindex_add(
+  librados::ObjectWriteOperation& op,
+  const cls_timeindex_entry& entry);
+
+void cls_timeindex_add(
+  librados::ObjectWriteOperation& op,
+  const utime_t& timestamp,
+  const std::string& name,
+  const bufferlist& bl);
+
+void cls_timeindex_list(
+  librados::ObjectReadOperation& op,
+  const utime_t& from,
+  const utime_t& to,
+  const std::string& in_marker,
+  const int max_entries,
+  std::list<cls_timeindex_entry>& entries,
+  std::string *out_marker,
+  bool *truncated);
+
+void cls_timeindex_trim(
+  librados::ObjectWriteOperation& op,
+  const utime_t& from_time,
+  const utime_t& to_time,
+  const std::string& from_marker = std::string(),
+  const std::string& to_marker = std::string());
+
+int cls_timeindex_trim(
+  librados::IoCtx& io_ctx,
+  const std::string& oid,
+  const utime_t& from_time,
+  const utime_t& to_time,
+  const std::string& from_marker = std::string(),
+  const std::string& to_marker = std::string());
+#endif
diff --git a/src/cls/timeindex/cls_timeindex_ops.h b/src/cls/timeindex/cls_timeindex_ops.h
new file mode 100644
index 00000000..05aa8b19
--- /dev/null
+++ b/src/cls/timeindex/cls_timeindex_ops.h
@@ -0,0 +1,115 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_TIMEINDEX_OPS_H
+#define CEPH_CLS_TIMEINDEX_OPS_H
+
+#include "cls_timeindex_types.h"
+
+struct cls_timeindex_add_op {
+  list<cls_timeindex_entry> entries;
+
+  cls_timeindex_add_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_timeindex_add_op)
+
+struct cls_timeindex_list_op {
+  utime_t from_time;
+  string marker; /* if not empty, overrides from_time */
+  utime_t to_time; /* not inclusive */
+  int max_entries; /* upperbound to returned num of entries
+                      might return less than that and still be truncated */
+
+  cls_timeindex_list_op() : max_entries(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(from_time, bl);
+    encode(marker, bl);
+    encode(to_time, bl);
+    encode(max_entries, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(from_time, bl);
+    decode(marker, bl);
+    decode(to_time, bl);
+    decode(max_entries, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_timeindex_list_op)
+
+struct cls_timeindex_list_ret {
+  list<cls_timeindex_entry> entries;
+  string marker;
+  bool truncated;
+
+  cls_timeindex_list_ret() : truncated(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(marker, bl);
+    encode(truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(marker, bl);
+    decode(truncated, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_timeindex_list_ret)
+
+
+/*
+ * operation will return 0 when successfully removed but not done. Will return
+ * -ENODATA when done, so caller needs to repeat sending request until that.
+ */
+struct cls_timeindex_trim_op {
+  utime_t from_time;
+  utime_t to_time; /* inclusive */
+  string from_marker;
+  string to_marker;
+
+  cls_timeindex_trim_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(from_time, bl);
+    encode(to_time, bl);
+    encode(from_marker, bl);
+    encode(to_marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(from_time, bl);
+    decode(to_time, bl);
+    decode(from_marker, bl);
+    decode(to_marker, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_timeindex_trim_op)
+
+#endif /* CEPH_CLS_TIMEINDEX_OPS_H */
diff --git a/src/cls/timeindex/cls_timeindex_types.h b/src/cls/timeindex/cls_timeindex_types.h
new file mode 100644
index 00000000..2b381767
--- /dev/null
+++ b/src/cls/timeindex/cls_timeindex_types.h
@@ -0,0 +1,43 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_TIMEINDEX_TYPES_H
+#define CEPH_CLS_TIMEINDEX_TYPES_H
+
+#include "include/encoding.h"
+#include "include/types.h"
+
+#include "include/utime.h"
+
+class JSONObj;
+
+struct cls_timeindex_entry {
+  /* Mandatory timestamp. Will be part of the key. */
+  utime_t key_ts;
+  /* Not mandatory. The name_ext field, if not empty, will form second
+   * part of the key. */
+  string key_ext;
+  /* Become value of OMAP-based mapping. */
+  bufferlist value;
+
+  cls_timeindex_entry() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(key_ts, bl);
+    encode(key_ext, bl);
+    encode(value, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(key_ts, bl);
+    decode(key_ext, bl);
+    decode(value, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_timeindex_entry)
+
+#endif /* CEPH_CLS_TIMEINDEX_TYPES_H */
diff --git a/src/cls/user/cls_user.cc b/src/cls/user/cls_user.cc
new file mode 100644
index 00000000..d3015a12
--- /dev/null
+++ b/src/cls/user/cls_user.cc
@@ -0,0 +1,455 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "include/utime.h"
+#include "objclass/objclass.h"
+
+#include "cls_user_ops.h"
+
+CLS_VER(1,0)
+CLS_NAME(user)
+
+static int write_entry(cls_method_context_t hctx, const string& key, const cls_user_bucket_entry& entry)
+{
+  bufferlist bl;
+  encode(entry, bl);
+
+  int ret = cls_cxx_map_set_val(hctx, key, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int remove_entry(cls_method_context_t hctx, const string& key)
+{
+  int ret = cls_cxx_map_remove_key(hctx, key);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static void get_key_by_bucket_name(const string& bucket_name, string *key)
+{
+  *key = bucket_name;
+}
+
+static int get_existing_bucket_entry(cls_method_context_t hctx, const string& bucket_name,
+                                     cls_user_bucket_entry& entry)
+{
+  if (bucket_name.empty()) {
+    return -EINVAL;
+  }
+
+  string key;
+  get_key_by_bucket_name(bucket_name, &key);
+
+  bufferlist bl;
+  int rc = cls_cxx_map_get_val(hctx, key, &bl);
+  if (rc < 0) {
+    CLS_LOG(10, "could not read entry %s", key.c_str());
+    return rc;
+  }
+  try {
+    auto iter = bl.cbegin();
+    decode(entry, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: failed to decode entry %s", key.c_str());
+    return -EIO;
+  }
+
+  return 0;
+}
+
+static int read_header(cls_method_context_t hctx, cls_user_header *header)
+{
+  bufferlist bl;
+
+  int ret = cls_cxx_map_read_header(hctx, &bl);
+  if (ret < 0)
+    return ret;
+
+  if (bl.length() == 0) {
+    *header = cls_user_header();
+    return 0;
+  }
+
+  try {
+    decode(*header, bl);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: failed to decode user header");
+    return -EIO;
+  }
+
+  return 0;
+}
+
+static void add_header_stats(cls_user_stats *stats, cls_user_bucket_entry& entry)
+{
+  stats->total_entries += entry.count;
+  stats->total_bytes += entry.size;
+  stats->total_bytes_rounded += entry.size_rounded;
+}
+
+static void dec_header_stats(cls_user_stats *stats, cls_user_bucket_entry& entry)
+{
+  stats->total_bytes -= entry.size;
+  stats->total_bytes_rounded -= entry.size_rounded;
+  stats->total_entries -= entry.count;
+}
+
+static void apply_entry_stats(const cls_user_bucket_entry& src_entry, cls_user_bucket_entry *target_entry)
+{
+  target_entry->size = src_entry.size;
+  target_entry->size_rounded = src_entry.size_rounded;
+  target_entry->count = src_entry.count;
+}
+
+static int cls_user_set_buckets_info(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_user_set_buckets_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_user_add_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  cls_user_header header;
+  int ret = read_header(hctx, &header);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: failed to read user info header ret=%d", ret);
+    return ret;
+  }
+
+  for (list<cls_user_bucket_entry>::iterator iter = op.entries.begin();
+       iter != op.entries.end(); ++iter) {
+    cls_user_bucket_entry& update_entry = *iter;
+
+    string key;
+
+    get_key_by_bucket_name(update_entry.bucket.name, &key);
+
+    cls_user_bucket_entry entry;
+    ret = get_existing_bucket_entry(hctx, key, entry);
+
+    if (ret == -ENOENT) {
+     if (!op.add)
+      continue; /* racing bucket removal */
+
+     entry = update_entry;
+
+     ret = 0;
+    } else if (op.add) {
+      // bucket id may have changed (ie reshard)
+      entry.bucket.bucket_id = update_entry.bucket.bucket_id;
+      // creation date may have changed (ie delete/recreate bucket)
+      entry.creation_time = update_entry.creation_time;
+    }
+
+    if (ret < 0) {
+      CLS_LOG(0, "ERROR: get_existing_bucket_entry() key=%s returned %d", key.c_str(), ret);
+      return ret;
+    } else if (ret >= 0 && entry.user_stats_sync) {
+      dec_header_stats(&header.stats, entry);
+    }
+
+    CLS_LOG(20, "storing entry for key=%s size=%lld count=%lld",
+            key.c_str(), (long long)update_entry.size, (long long)update_entry.count);
+
+    // sync entry stats when not an op.add, as when the case is op.add if its a
+    // new entry we already have copied update_entry earlier, OTOH, for an existing entry
+    // we end up clobbering the existing stats for the bucket
+    if (!op.add){
+      apply_entry_stats(update_entry, &entry);
+    }
+    entry.user_stats_sync = true;
+
+    ret = write_entry(hctx, key, entry);
+    if (ret < 0)
+      return ret;
+
+    add_header_stats(&header.stats, entry);
+  }
+
+  bufferlist bl;
+
+  CLS_LOG(20, "header: total bytes=%lld entries=%lld", (long long)header.stats.total_bytes, (long long)header.stats.total_entries);
+
+  if (header.last_stats_update < op.time)
+    header.last_stats_update = op.time;
+
+  encode(header, bl);
+  
+  ret = cls_cxx_map_write_header(hctx, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_user_complete_stats_sync(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_user_complete_stats_sync_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_user_add_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  cls_user_header header;
+  int ret = read_header(hctx, &header);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: failed to read user info header ret=%d", ret);
+    return ret;
+  }
+
+  if (header.last_stats_sync < op.time)
+    header.last_stats_sync = op.time;
+
+  bufferlist bl;
+
+  encode(header, bl);
+
+  ret = cls_cxx_map_write_header(hctx, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_user_remove_bucket(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_user_remove_bucket_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_user_add_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  cls_user_header header;
+  int ret = read_header(hctx, &header);
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: failed to read user info header ret=%d", ret);
+    return ret;
+  }
+
+  string key;
+
+  get_key_by_bucket_name(op.bucket.name, &key);
+
+  cls_user_bucket_entry entry;
+  ret = get_existing_bucket_entry(hctx, key, entry);
+  if (ret == -ENOENT) {
+    return 0; /* idempotent removal */
+  }
+  if (ret < 0) {
+    CLS_LOG(0, "ERROR: get existing bucket entry, key=%s ret=%d", key.c_str(), ret);
+    return ret;
+  }
+
+  CLS_LOG(20, "removing entry at %s", key.c_str());
+
+  ret = remove_entry(hctx, key);
+  if (ret < 0)
+    return ret;
+
+  if (!entry.user_stats_sync) {
+    return 0;
+  }
+
+  dec_header_stats(&header.stats, entry);
+
+  CLS_LOG(20, "header: total bytes=%lld entries=%lld", (long long)header.stats.total_bytes, (long long)header.stats.total_entries);
+
+  bufferlist bl;
+  encode(header, bl);
+  return cls_cxx_map_write_header(hctx, &bl);
+}
+
+static int cls_user_list_buckets(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_user_list_buckets_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_user_list_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  map<string, bufferlist> keys;
+
+  const string& from_index = op.marker;
+  const string& to_index = op.end_marker;
+  const bool to_index_valid = !to_index.empty();
+
+#define MAX_ENTRIES 1000
+  size_t max_entries = op.max_entries;
+  if (max_entries > MAX_ENTRIES)
+    max_entries = MAX_ENTRIES;
+
+  string match_prefix;
+  cls_user_list_buckets_ret ret;
+
+  int rc = cls_cxx_map_get_vals(hctx, from_index, match_prefix, max_entries, &keys, &ret.truncated);
+  if (rc < 0)
+    return rc;
+
+  CLS_LOG(20, "from_index=%s to_index=%s match_prefix=%s",
+          from_index.c_str(),
+          to_index.c_str(),
+          match_prefix.c_str());
+
+  list<cls_user_bucket_entry>& entries = ret.entries;
+  map<string, bufferlist>::iterator iter = keys.begin();
+
+  string marker;
+
+  for (; iter != keys.end(); ++iter) {
+    const string& index = iter->first;
+    marker = index;
+
+    if (to_index_valid && to_index.compare(index) <= 0) {
+      ret.truncated = false;
+      break;
+    }
+
+    bufferlist& bl = iter->second;
+    auto biter = bl.cbegin();
+    try {
+      cls_user_bucket_entry e;
+      decode(e, biter);
+      entries.push_back(e);
+    } catch (buffer::error& err) {
+      CLS_LOG(0, "ERROR: cls_user_list: could not decode entry, index=%s", index.c_str());
+    }
+  }
+
+  if (ret.truncated) {
+    ret.marker = marker;
+  }
+
+  encode(ret, *out);
+
+  return 0;
+}
+
+static int cls_user_get_header(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_user_get_header_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_user_get_header_op(): failed to decode op");
+    return -EINVAL;
+  }
+
+  cls_user_get_header_ret op_ret;
+
+  int ret = read_header(hctx, &op_ret.header);
+  if (ret < 0)
+    return ret;
+
+  encode(op_ret, *out);
+
+  return 0;
+}
+
+/// A method to reset the user.buckets header stats in accordance to
+/// the values seen in the user.buckets omap keys. This is not be
+/// equivalent to --sync-stats which also re-calculates the stats for
+/// each bucket.
+static int cls_user_reset_stats(cls_method_context_t hctx,
+				bufferlist *in, bufferlist *out /*ignore*/)
+{
+  cls_user_reset_stats_op op;
+
+  try {
+    auto bliter = in->cbegin();
+    decode(op, bliter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: %s failed to decode op", __func__);
+    return -EINVAL;
+  }
+
+  cls_user_header header;
+  bool truncated = false;
+  string from_index, prefix;
+  do {
+    map<string, bufferlist> keys;
+    int rc = cls_cxx_map_get_vals(hctx, from_index, prefix, MAX_ENTRIES,
+				  &keys, &truncated);
+    if (rc < 0) {
+      CLS_LOG(0, "ERROR: %s failed to retrieve omap key-values", __func__);
+      return rc;
+    }
+    CLS_LOG(20, "%s: read %lu key-values, truncated=%d",
+	    __func__, keys.size(), truncated);
+
+    for (const auto& kv : keys) {
+      cls_user_bucket_entry e;
+      try {
+	auto bl = kv.second;
+	auto bliter = bl.cbegin();
+	decode(e, bliter);
+      } catch (buffer::error& err) {
+	CLS_LOG(0, "ERROR: %s failed to decode bucket entry for %s",
+		__func__, kv.first.c_str());
+	return -EIO;
+      }
+      add_header_stats(&header.stats, e);
+    }
+    if (!keys.empty()) {
+      from_index = keys.rbegin()->first;
+    }
+  } while (truncated);
+
+  bufferlist bl;
+  header.last_stats_update = op.time;
+  encode(header, bl);
+
+  CLS_LOG(20, "%s: updating header", __func__);
+  return cls_cxx_map_write_header(hctx, &bl);
+}
+
+CLS_INIT(user)
+{
+  CLS_LOG(1, "Loaded user class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_user_set_buckets_info;
+  cls_method_handle_t h_user_complete_stats_sync;
+  cls_method_handle_t h_user_remove_bucket;
+  cls_method_handle_t h_user_list_buckets;
+  cls_method_handle_t h_user_get_header;
+  cls_method_handle_t h_user_reset_stats;
+
+  cls_register("user", &h_class);
+
+  /* log */
+  cls_register_cxx_method(h_class, "set_buckets_info", CLS_METHOD_RD | CLS_METHOD_WR,
+                          cls_user_set_buckets_info, &h_user_set_buckets_info);
+  cls_register_cxx_method(h_class, "complete_stats_sync", CLS_METHOD_RD | CLS_METHOD_WR,
+                          cls_user_complete_stats_sync, &h_user_complete_stats_sync);
+  cls_register_cxx_method(h_class, "remove_bucket", CLS_METHOD_RD | CLS_METHOD_WR, cls_user_remove_bucket, &h_user_remove_bucket);
+  cls_register_cxx_method(h_class, "list_buckets", CLS_METHOD_RD, cls_user_list_buckets, &h_user_list_buckets);
+  cls_register_cxx_method(h_class, "get_header", CLS_METHOD_RD, cls_user_get_header, &h_user_get_header);
+  cls_register_cxx_method(h_class, "reset_user_stats", CLS_METHOD_RD | CLS_METHOD_WR, cls_user_reset_stats, &h_user_reset_stats);
+  return;
+}
+
diff --git a/src/cls/user/cls_user_client.cc b/src/cls/user/cls_user_client.cc
new file mode 100644
index 00000000..c18be4a2
--- /dev/null
+++ b/src/cls/user/cls_user_client.cc
@@ -0,0 +1,158 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "cls/user/cls_user_client.h"
+#include "include/rados/librados.hpp"
+
+
+using namespace librados;
+
+
+void cls_user_set_buckets(librados::ObjectWriteOperation& op, list<cls_user_bucket_entry>& entries, bool add)
+{
+  bufferlist in;
+  cls_user_set_buckets_op call;
+  call.entries = entries;
+  call.add = add;
+  call.time = real_clock::now();
+  encode(call, in);
+  op.exec("user", "set_buckets_info", in);
+}
+
+void cls_user_complete_stats_sync(librados::ObjectWriteOperation& op)
+{
+  bufferlist in;
+  cls_user_complete_stats_sync_op call;
+  call.time = real_clock::now();
+  encode(call, in);
+  op.exec("user", "complete_stats_sync", in);
+}
+
+void cls_user_remove_bucket(librados::ObjectWriteOperation& op, const cls_user_bucket& bucket)
+{
+  bufferlist in;
+  cls_user_remove_bucket_op call;
+  call.bucket = bucket;
+  encode(call, in);
+  op.exec("user", "remove_bucket", in);
+}
+
+class ClsUserListCtx : public ObjectOperationCompletion {
+  list<cls_user_bucket_entry> *entries;
+  string *marker;
+  bool *truncated;
+  int *pret;
+public:
+  ClsUserListCtx(list<cls_user_bucket_entry> *_entries, string *_marker, bool *_truncated, int *_pret) :
+                                      entries(_entries), marker(_marker), truncated(_truncated), pret(_pret) {}
+  void handle_completion(int r, bufferlist& outbl) override {
+    if (r >= 0) {
+      cls_user_list_buckets_ret ret;
+      try {
+        auto iter = outbl.cbegin();
+        decode(ret, iter);
+        if (entries)
+	  *entries = ret.entries;
+        if (truncated)
+          *truncated = ret.truncated;
+        if (marker)
+          *marker = ret.marker;
+      } catch (buffer::error& err) {
+        r = -EIO;
+      }
+    }
+    if (pret) {
+      *pret = r;
+    }
+  }
+};
+
+void cls_user_bucket_list(librados::ObjectReadOperation& op,
+                          const string& in_marker,
+                          const string& end_marker,
+                          int max_entries,
+                          list<cls_user_bucket_entry>& entries,
+                          string *out_marker,
+                          bool *truncated,
+                          int *pret)
+{
+  bufferlist inbl;
+  cls_user_list_buckets_op call;
+  call.marker = in_marker;
+  call.end_marker = end_marker;
+  call.max_entries = max_entries;
+
+  encode(call, inbl);
+
+  op.exec("user", "list_buckets", inbl, new ClsUserListCtx(&entries, out_marker, truncated, pret));
+}
+
+class ClsUserGetHeaderCtx : public ObjectOperationCompletion {
+  cls_user_header *header;
+  RGWGetUserHeader_CB *ret_ctx;
+  int *pret;
+public:
+  ClsUserGetHeaderCtx(cls_user_header *_h, RGWGetUserHeader_CB *_ctx, int *_pret) : header(_h), ret_ctx(_ctx), pret(_pret) {}
+  ~ClsUserGetHeaderCtx() override {
+    if (ret_ctx) {
+      ret_ctx->put();
+    }
+  }
+  void handle_completion(int r, bufferlist& outbl) override {
+    if (r >= 0) {
+      cls_user_get_header_ret ret;
+      try {
+        auto iter = outbl.cbegin();
+        decode(ret, iter);
+        if (header)
+	  *header = ret.header;
+      } catch (buffer::error& err) {
+        r = -EIO;
+      }
+      if (ret_ctx) {
+        ret_ctx->handle_response(r, ret.header);
+      }
+    }
+    if (pret) {
+      *pret = r;
+    }
+  }
+};
+
+void cls_user_get_header(librados::ObjectReadOperation& op,
+                       cls_user_header *header, int *pret)
+{
+  bufferlist inbl;
+  cls_user_get_header_op call;
+
+  encode(call, inbl);
+
+  op.exec("user", "get_header", inbl, new ClsUserGetHeaderCtx(header, NULL, pret));
+}
+
+void cls_user_reset_stats(librados::ObjectWriteOperation &op)
+{
+  bufferlist inbl;
+  cls_user_reset_stats_op call;
+  call.time = real_clock::now();
+  encode(call, inbl);
+  op.exec("user", "reset_user_stats", inbl);
+}
+
+int cls_user_get_header_async(IoCtx& io_ctx, string& oid, RGWGetUserHeader_CB *ctx)
+{
+  bufferlist in, out;
+  cls_user_get_header_op call;
+  encode(call, in);
+  ObjectReadOperation op;
+  op.exec("user", "get_header", in, new ClsUserGetHeaderCtx(NULL, ctx, NULL)); /* no need to pass pret, as we'll call ctx->handle_response() with correct error */
+  AioCompletion *c = librados::Rados::aio_create_completion(NULL, NULL, NULL);
+  int r = io_ctx.aio_operate(oid, c, &op, NULL);
+  c->release();
+  if (r < 0)
+    return r;
+
+  return 0;
+}
diff --git a/src/cls/user/cls_user_client.h b/src/cls/user/cls_user_client.h
new file mode 100644
index 00000000..077c15dc
--- /dev/null
+++ b/src/cls/user/cls_user_client.h
@@ -0,0 +1,36 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_USER_CLIENT_H
+#define CEPH_CLS_USER_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include "cls_user_ops.h"
+#include "common/RefCountedObj.h"
+
+class RGWGetUserHeader_CB : public RefCountedObject {
+public:
+  ~RGWGetUserHeader_CB() override {}
+  virtual void handle_response(int r, cls_user_header& header) = 0;
+};
+
+/*
+ * user objclass
+ */
+
+void cls_user_set_buckets(librados::ObjectWriteOperation& op, list<cls_user_bucket_entry>& entries, bool add);
+void cls_user_complete_stats_sync(librados::ObjectWriteOperation& op);
+void cls_user_remove_bucket(librados::ObjectWriteOperation& op,  const cls_user_bucket& bucket);
+void cls_user_bucket_list(librados::ObjectReadOperation& op,
+                       const string& in_marker,
+                       const string& end_marker,
+                       int max_entries,
+                       list<cls_user_bucket_entry>& entries,
+                       string *out_marker,
+                       bool *truncated,
+                       int *pret);
+void cls_user_get_header(librados::ObjectReadOperation& op, cls_user_header *header, int *pret);
+int cls_user_get_header_async(librados::IoCtx& io_ctx, string& oid, RGWGetUserHeader_CB *ctx);
+void cls_user_reset_stats(librados::ObjectWriteOperation& op);
+
+#endif
diff --git a/src/cls/user/cls_user_ops.cc b/src/cls/user/cls_user_ops.cc
new file mode 100644
index 00000000..65d889bb
--- /dev/null
+++ b/src/cls/user/cls_user_ops.cc
@@ -0,0 +1,114 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/user/cls_user_ops.h"
+#include "common/Formatter.h"
+#include "common/ceph_json.h"
+
+void cls_user_set_buckets_op::dump(Formatter *f) const
+{
+  encode_json("entries", entries, f);
+  encode_json("add", add, f);
+  encode_json("time", utime_t(time), f);
+}
+
+void cls_user_set_buckets_op::generate_test_instances(list<cls_user_set_buckets_op*>& ls)
+{
+  ls.push_back(new cls_user_set_buckets_op);
+  cls_user_set_buckets_op *op = new cls_user_set_buckets_op;
+  for (int i = 0; i < 3; i++) {
+    cls_user_bucket_entry e;
+    cls_user_gen_test_bucket_entry(&e, i);
+    op->entries.push_back(e);
+  }
+  op->add = true;
+  op->time = utime_t(1, 0).to_real_time();
+  ls.push_back(op);
+}
+
+void cls_user_remove_bucket_op::dump(Formatter *f) const
+{
+  encode_json("bucket", bucket, f);
+}
+
+void cls_user_remove_bucket_op::generate_test_instances(list<cls_user_remove_bucket_op*>& ls)
+{
+  ls.push_back(new cls_user_remove_bucket_op);
+  cls_user_remove_bucket_op *op = new cls_user_remove_bucket_op;
+  cls_user_gen_test_bucket(&op->bucket, 0);
+  ls.push_back(op);
+}
+
+void cls_user_list_buckets_op::dump(Formatter *f) const
+{
+  encode_json("marker", marker, f);
+  encode_json("max_entries", max_entries, f);
+}
+
+void cls_user_list_buckets_op::generate_test_instances(list<cls_user_list_buckets_op*>& ls)
+{
+  ls.push_back(new cls_user_list_buckets_op);
+  cls_user_list_buckets_op *op = new cls_user_list_buckets_op;
+  op->marker = "marker";
+  op->max_entries = 1000;
+  ls.push_back(op);
+}
+
+void cls_user_list_buckets_ret::dump(Formatter *f) const
+{
+  encode_json("entries", entries, f);
+  encode_json("marker", marker, f);
+  encode_json("truncated", truncated, f);
+}
+
+void cls_user_list_buckets_ret::generate_test_instances(list<cls_user_list_buckets_ret*>& ls)
+{
+  ls.push_back(new cls_user_list_buckets_ret);
+  cls_user_list_buckets_ret *ret = new cls_user_list_buckets_ret;
+  for (int i = 0; i < 3; i++) {
+    cls_user_bucket_entry e;
+    cls_user_gen_test_bucket_entry(&e, i);
+    ret->entries.push_back(e);
+  }
+  ret->marker = "123";
+  ret->truncated = true;
+  ls.push_back(ret);
+}
+
+void cls_user_get_header_op::dump(Formatter *f) const
+{
+  // empty!
+}
+
+void cls_user_get_header_op::generate_test_instances(list<cls_user_get_header_op*>& ls)
+{
+  ls.push_back(new cls_user_get_header_op);
+}
+
+void cls_user_get_header_ret::dump(Formatter *f) const
+{
+  encode_json("header", header, f);
+}
+
+void cls_user_get_header_ret::generate_test_instances(list<cls_user_get_header_ret*>& ls)
+{
+  ls.push_back(new cls_user_get_header_ret);
+  cls_user_get_header_ret *ret = new cls_user_get_header_ret;
+  cls_user_gen_test_header(&ret->header);
+  ls.push_back(ret);
+}
+
+void cls_user_complete_stats_sync_op::dump(Formatter *f) const
+{
+  encode_json("time", utime_t(time), f);
+}
+
+void cls_user_complete_stats_sync_op::generate_test_instances(list<cls_user_complete_stats_sync_op*>& ls)
+{
+  ls.push_back(new cls_user_complete_stats_sync_op);
+  cls_user_complete_stats_sync_op *op = new cls_user_complete_stats_sync_op;
+  op->time = utime_t(12345, 0).to_real_time();
+  ls.push_back(op);
+}
+
+
diff --git a/src/cls/user/cls_user_ops.h b/src/cls/user/cls_user_ops.h
new file mode 100644
index 00000000..fa4b1f31
--- /dev/null
+++ b/src/cls/user/cls_user_ops.h
@@ -0,0 +1,204 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_USER_OPS_H
+#define CEPH_CLS_USER_OPS_H
+
+#include "cls_user_types.h"
+
+struct cls_user_set_buckets_op {
+  list<cls_user_bucket_entry> entries;
+  bool add;
+  real_time time; /* op time */
+
+  cls_user_set_buckets_op() : add(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(add, bl);
+    encode(time, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(add, bl);
+    decode(time, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_set_buckets_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_set_buckets_op)
+
+struct cls_user_remove_bucket_op {
+  cls_user_bucket bucket;
+
+  cls_user_remove_bucket_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(bucket, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(bucket, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_remove_bucket_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_remove_bucket_op)
+
+struct cls_user_list_buckets_op {
+  string marker;
+  string end_marker;
+  int max_entries; /* upperbound to returned num of entries
+                      might return less than that and still be truncated */
+
+  cls_user_list_buckets_op()
+    : max_entries(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(2, 1, bl);
+    encode(marker, bl);
+    encode(max_entries, bl);
+    encode(end_marker, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(2, bl);
+    decode(marker, bl);
+    decode(max_entries, bl);
+    if (struct_v >= 2) {
+      decode(end_marker, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_list_buckets_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_list_buckets_op)
+
+struct cls_user_list_buckets_ret {
+  list<cls_user_bucket_entry> entries;
+  string marker;
+  bool truncated;
+
+  cls_user_list_buckets_ret() : truncated(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(entries, bl);
+    encode(marker, bl);
+    encode(truncated, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(entries, bl);
+    decode(marker, bl);
+    decode(truncated, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_list_buckets_ret*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_list_buckets_ret)
+
+
+struct cls_user_get_header_op {
+  cls_user_get_header_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_get_header_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_get_header_op)
+
+struct cls_user_reset_stats_op {
+  real_time time;
+  cls_user_reset_stats_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(time, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(time, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_reset_stats_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_reset_stats_op);
+
+struct cls_user_get_header_ret {
+  cls_user_header header;
+
+  cls_user_get_header_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(header, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(header, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_get_header_ret*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_get_header_ret)
+
+struct cls_user_complete_stats_sync_op {
+  real_time time;
+
+  cls_user_complete_stats_sync_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(time, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(time, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_complete_stats_sync_op*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_complete_stats_sync_op)
+
+
+#endif
diff --git a/src/cls/user/cls_user_types.cc b/src/cls/user/cls_user_types.cc
new file mode 100644
index 00000000..be3280ca
--- /dev/null
+++ b/src/cls/user/cls_user_types.cc
@@ -0,0 +1,104 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include "cls/user/cls_user_types.h"
+#include "common/Formatter.h"
+#include "common/ceph_json.h"
+#include "include/utime.h"
+
+void cls_user_gen_test_bucket(cls_user_bucket *bucket, int i)
+{
+  char buf[16];
+  snprintf(buf, sizeof(buf), ".%d", i);
+
+  bucket->name = string("buck") + buf;
+  bucket->marker = string("mark") + buf;
+  bucket->bucket_id = string("bucket.id") + buf;
+}
+
+void cls_user_bucket::dump(Formatter *f) const
+{
+  encode_json("name", name, f);
+  encode_json("marker", marker,f);
+  encode_json("bucket_id", bucket_id,f);
+}
+
+void cls_user_bucket::generate_test_instances(list<cls_user_bucket*>& ls)
+{
+  ls.push_back(new cls_user_bucket);
+  cls_user_bucket *b = new cls_user_bucket;
+  cls_user_gen_test_bucket(b, 0);
+  ls.push_back(b);
+}
+
+void cls_user_bucket_entry::dump(Formatter *f) const
+{
+  encode_json("bucket", bucket, f);
+  encode_json("size", size, f);
+  encode_json("size_rounded", size_rounded, f);
+  encode_json("creation_time", utime_t(creation_time), f);
+  encode_json("count", count, f);
+  encode_json("user_stats_sync", user_stats_sync, f);
+}
+
+void cls_user_gen_test_bucket_entry(cls_user_bucket_entry *entry, int i)
+{
+  cls_user_gen_test_bucket(&entry->bucket, i);
+  entry->size = i + 1;
+  entry->size_rounded = i + 2;
+  entry->creation_time = real_clock::from_time_t(i + 3);
+  entry->count = i + 4;
+  entry->user_stats_sync = true;
+}
+
+void cls_user_bucket_entry::generate_test_instances(list<cls_user_bucket_entry*>& ls)
+{
+  ls.push_back(new cls_user_bucket_entry);
+  cls_user_bucket_entry *entry = new cls_user_bucket_entry;
+  cls_user_gen_test_bucket_entry(entry, 0);
+  ls.push_back(entry);
+}
+
+void cls_user_gen_test_stats(cls_user_stats *s)
+{
+  s->total_entries = 1;
+  s->total_bytes = 2;
+  s->total_bytes_rounded = 3;
+}
+
+void cls_user_stats::dump(Formatter *f) const
+{
+  f->dump_int("total_entries", total_entries);
+  f->dump_int("total_bytes", total_bytes);
+  f->dump_int("total_bytes_rounded", total_bytes_rounded);
+}
+
+void cls_user_stats::generate_test_instances(list<cls_user_stats*>& ls)
+{
+  ls.push_back(new cls_user_stats);
+  cls_user_stats *s = new cls_user_stats;
+  cls_user_gen_test_stats(s);
+  ls.push_back(s);
+}
+
+void cls_user_gen_test_header(cls_user_header *h)
+{
+  cls_user_gen_test_stats(&h->stats);
+  h->last_stats_sync = utime_t(1, 0).to_real_time();
+  h->last_stats_update = utime_t(2, 0).to_real_time();
+}
+  
+void cls_user_header::dump(Formatter *f) const
+{
+  encode_json("stats", stats, f);
+  encode_json("last_stats_sync", utime_t(last_stats_sync), f);
+  encode_json("last_stats_update", utime_t(last_stats_update), f);
+}
+
+void cls_user_header::generate_test_instances(list<cls_user_header*>& ls)
+{
+  ls.push_back(new cls_user_header);
+  cls_user_header *h = new cls_user_header;
+  cls_user_gen_test_header(h);
+  ls.push_back(h);
+}
diff --git a/src/cls/user/cls_user_types.h b/src/cls/user/cls_user_types.h
new file mode 100644
index 00000000..8efd8bc5
--- /dev/null
+++ b/src/cls/user/cls_user_types.h
@@ -0,0 +1,226 @@
+// -*- mode:C; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_USER_TYPES_H
+#define CEPH_CLS_USER_TYPES_H
+
+#include "include/encoding.h"
+#include "include/types.h"
+#include "include/utime.h"
+#include "common/ceph_time.h"
+
+/*
+ * this needs to be compatible with rgw_bucket, as it replaces it
+ */
+struct cls_user_bucket {
+  std::string name;
+  std::string marker;
+  std::string bucket_id;
+  std::string placement_id;
+  struct {
+    std::string data_pool;
+    std::string index_pool;
+    std::string data_extra_pool;
+  } explicit_placement;
+
+  void encode(bufferlist& bl) const {
+    /* since new version of this structure is not backward compatible,
+     * we have older rgw running against newer osd if we encode it
+     * in the new way. Only encode newer version if placement_id is
+     * not empty, otherwise keep handling it as before
+     */
+    if (!placement_id.empty()) {
+      ENCODE_START(9, 8, bl);
+      encode(name, bl);
+      encode(marker, bl);
+      encode(bucket_id, bl);
+      encode(placement_id, bl);
+      ENCODE_FINISH(bl);
+    } else {
+      ENCODE_START(7, 3, bl);
+      encode(name, bl);
+      encode(explicit_placement.data_pool, bl);
+      encode(marker, bl);
+      encode(bucket_id, bl);
+      encode(explicit_placement.index_pool, bl);
+      encode(explicit_placement.data_extra_pool, bl);
+      ENCODE_FINISH(bl);
+    }
+  }
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(8, 3, 3, bl);
+    decode(name, bl);
+    if (struct_v < 8) {
+      decode(explicit_placement.data_pool, bl);
+    }
+    if (struct_v >= 2) {
+      decode(marker, bl);
+      if (struct_v <= 3) {
+        uint64_t id;
+        decode(id, bl);
+        char buf[16];
+        snprintf(buf, sizeof(buf), "%llu", (long long)id);
+        bucket_id = buf;
+      } else {
+        decode(bucket_id, bl);
+      }
+    }
+    if (struct_v < 8) {
+      if (struct_v >= 5) {
+        decode(explicit_placement.index_pool, bl);
+      } else {
+        explicit_placement.index_pool = explicit_placement.data_pool;
+      }
+      if (struct_v >= 7) {
+        decode(explicit_placement.data_extra_pool, bl);
+      }
+    } else {
+      decode(placement_id, bl);
+      if (struct_v == 8 && placement_id.empty()) {
+        decode(explicit_placement.data_pool, bl);
+        decode(explicit_placement.index_pool, bl);
+        decode(explicit_placement.data_extra_pool, bl);
+      }
+    }
+    DECODE_FINISH(bl);
+  }
+
+  bool operator<(const cls_user_bucket& b) const {
+    return name.compare(b.name) < 0;
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_bucket*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_bucket)
+
+/*
+ * this structure overrides RGWBucketEnt
+ */
+struct cls_user_bucket_entry {
+  cls_user_bucket bucket;
+  size_t size;
+  size_t size_rounded;
+  ceph::real_time creation_time;
+  uint64_t count;
+  bool user_stats_sync;
+
+  cls_user_bucket_entry() : size(0), size_rounded(0), count(0), user_stats_sync(false) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(9, 5, bl);
+    uint64_t s = size;
+    __u32 mt = ceph::real_clock::to_time_t(creation_time);
+    string empty_str;  // originally had the bucket name here, but we encode bucket later
+    encode(empty_str, bl);
+    encode(s, bl);
+    encode(mt, bl);
+    encode(count, bl);
+    encode(bucket, bl);
+    s = size_rounded;
+    encode(s, bl);
+    encode(user_stats_sync, bl);
+    encode(creation_time, bl);
+    //::encode(placement_rule, bl); removed in v9
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START_LEGACY_COMPAT_LEN(9, 5, 5, bl);
+    __u32 mt;
+    uint64_t s;
+    string empty_str;  // backward compatibility
+    decode(empty_str, bl);
+    decode(s, bl);
+    decode(mt, bl);
+    size = s;
+    if (struct_v < 7) {
+      creation_time = ceph::real_clock::from_time_t(mt);
+    }
+    if (struct_v >= 2)
+      decode(count, bl);
+    if (struct_v >= 3)
+      decode(bucket, bl);
+    if (struct_v >= 4)
+      decode(s, bl);
+    size_rounded = s;
+    if (struct_v >= 6)
+      decode(user_stats_sync, bl);
+    if (struct_v >= 7)
+      decode(creation_time, bl);
+    if (struct_v == 8) { // added in v8, removed in v9
+      std::string placement_rule;
+      decode(placement_rule, bl);
+    }
+    DECODE_FINISH(bl);
+  }
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_bucket_entry*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_bucket_entry)
+
+struct cls_user_stats {
+  uint64_t total_entries;
+  uint64_t total_bytes;
+  uint64_t total_bytes_rounded;
+
+  cls_user_stats()
+    : total_entries(0),
+      total_bytes(0),
+      total_bytes_rounded(0) {}
+
+  void encode(bufferlist& bl) const {
+     ENCODE_START(1, 1, bl);
+    encode(total_entries, bl);
+    encode(total_bytes, bl);
+    encode(total_bytes_rounded, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(total_entries, bl);
+    decode(total_bytes, bl);
+    decode(total_bytes_rounded, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_stats*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_stats)
+
+/*
+ * this needs to be compatible with rgw_bucket, as it replaces it
+ */
+struct cls_user_header {
+  cls_user_stats stats;
+  ceph::real_time last_stats_sync;     /* last time a full stats sync completed */
+  ceph::real_time last_stats_update;   /* last time a stats update was done */
+
+  void encode(bufferlist& bl) const {
+     ENCODE_START(1, 1, bl);
+    encode(stats, bl);
+    encode(last_stats_sync, bl);
+    encode(last_stats_update, bl);
+    ENCODE_FINISH(bl);
+  }
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(stats, bl);
+    decode(last_stats_sync, bl);
+    decode(last_stats_update, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void dump(Formatter *f) const;
+  static void generate_test_instances(list<cls_user_header*>& ls);
+};
+WRITE_CLASS_ENCODER(cls_user_header)
+
+void cls_user_gen_test_bucket(cls_user_bucket *bucket, int i);
+void cls_user_gen_test_bucket_entry(cls_user_bucket_entry *entry, int i);
+void cls_user_gen_test_stats(cls_user_stats *stats);
+void cls_user_gen_test_header(cls_user_header *h);
+
+#endif
+
+
diff --git a/src/cls/version/cls_version.cc b/src/cls/version/cls_version.cc
new file mode 100644
index 00000000..61f9fec0
--- /dev/null
+++ b/src/cls/version/cls_version.cc
@@ -0,0 +1,234 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#include <errno.h>
+
+#include "objclass/objclass.h"
+
+#include "cls/version/cls_version_ops.h"
+
+#include "include/compat.h"
+
+CLS_VER(1,0)
+CLS_NAME(version)
+
+
+#define VERSION_ATTR "ceph.objclass.version"
+
+static int set_version(cls_method_context_t hctx, struct obj_version *objv)
+{
+  bufferlist bl;
+
+  encode(*objv, bl);
+
+  CLS_LOG(20, "cls_version: set_version %s:%d", objv->tag.c_str(), (int)objv->ver);
+
+  int ret = cls_cxx_setxattr(hctx, VERSION_ATTR, &bl);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int init_version(cls_method_context_t hctx, struct obj_version *objv)
+{
+#define TAG_LEN 24
+  char buf[TAG_LEN + 1];
+
+  int ret = cls_gen_rand_base64(buf, sizeof(buf));
+  if (ret < 0)
+    return ret;
+
+  objv->ver = 1;
+  objv->tag = buf;
+
+  CLS_LOG(20, "cls_version: init_version %s:%d", objv->tag.c_str(), (int)objv->ver);
+
+  return set_version(hctx, objv);
+}
+
+/* implicit create should be true only if called from a write operation (set, inc), never from a read operation (read, check) */
+static int read_version(cls_method_context_t hctx, obj_version *objv, bool implicit_create)
+{
+  bufferlist bl;
+  int ret = cls_cxx_getxattr(hctx, VERSION_ATTR, &bl);
+  if (ret == -ENOENT || ret == -ENODATA) {
+    objv->ver = 0;
+
+    if (implicit_create) {
+      return init_version(hctx, objv);
+    }
+    return 0;
+  }
+  if (ret < 0)
+    return ret;
+
+  try {
+    auto iter = bl.cbegin();
+    decode(*objv, iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(0, "ERROR: read_version(): failed to decode version entry\n");
+    return -EIO;
+  }
+  CLS_LOG(20, "cls_version: read_version %s:%d", objv->tag.c_str(), (int)objv->ver);
+
+  return 0;
+}
+
+static int cls_version_set(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_version_set_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_version_get(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  int ret = set_version(hctx, &op.objv);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static bool check_conds(list<obj_version_cond>& conds, obj_version& objv)
+{
+  if (conds.empty())
+    return true;
+
+  for (list<obj_version_cond>::iterator iter = conds.begin(); iter != conds.end(); ++iter) {
+    obj_version_cond& cond = *iter;
+    obj_version& v = cond.ver;
+    CLS_LOG(20, "cls_version: check_version %s:%d (cond=%d)", v.tag.c_str(), (int)v.ver, (int)cond.cond);
+
+    switch (cond.cond) {
+      case VER_COND_NONE:
+	break;
+      case VER_COND_EQ:
+	if (!objv.compare(&v))
+	  return false;
+	break;
+      case VER_COND_GT:
+	if (!(objv.ver > v.ver))
+	  return false;
+	break;
+      case VER_COND_GE:
+	if (!(objv.ver >= v.ver))
+	  return false;
+	break;
+      case VER_COND_LT:
+	if (!(objv.ver < v.ver))
+	  return false;
+	break;
+      case VER_COND_LE:
+	if (!(objv.ver <= v.ver))
+	  return false;
+	break;
+      case VER_COND_TAG_EQ:
+	if (objv.tag.compare(v.tag) != 0)
+	  return false;
+	break;
+      case VER_COND_TAG_NE:
+	if (objv.tag.compare(v.tag) == 0)
+	  return false;
+	break;
+    }
+  }
+
+  return true;
+}
+
+static int cls_version_inc(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_version_inc_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_version_get(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  obj_version objv;
+  int ret = read_version(hctx, &objv, true);
+  if (ret < 0)
+    return ret;
+  
+  if (!check_conds(op.conds, objv)) {
+    return -ECANCELED;
+  }
+  objv.inc();
+
+  ret = set_version(hctx, &objv);
+  if (ret < 0)
+    return ret;
+
+  return 0;
+}
+
+static int cls_version_check(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  auto in_iter = in->cbegin();
+
+  cls_version_check_op op;
+  try {
+    decode(op, in_iter);
+  } catch (buffer::error& err) {
+    CLS_LOG(1, "ERROR: cls_version_get(): failed to decode entry\n");
+    return -EINVAL;
+  }
+
+  obj_version objv;
+  int ret = read_version(hctx, &objv, false);
+  if (ret < 0)
+    return ret;
+  
+  if (!check_conds(op.conds, objv)) {
+    CLS_LOG(20, "cls_version: failed condition check");
+    return -ECANCELED;
+  }
+
+  return 0;
+}
+
+static int cls_version_read(cls_method_context_t hctx, bufferlist *in, bufferlist *out)
+{
+  obj_version objv;
+
+  cls_version_read_ret read_ret;
+  int ret = read_version(hctx, &read_ret.objv, false);
+  if (ret < 0)
+    return ret;
+
+  encode(read_ret, *out);
+
+  return 0;
+}
+
+CLS_INIT(version)
+{
+  CLS_LOG(1, "Loaded version class!");
+
+  cls_handle_t h_class;
+  cls_method_handle_t h_version_set;
+  cls_method_handle_t h_version_inc;
+  cls_method_handle_t h_version_inc_conds;
+  cls_method_handle_t h_version_read;
+  cls_method_handle_t h_version_check_conds;
+
+  cls_register("version", &h_class);
+
+  /* version */
+  cls_register_cxx_method(h_class, "set", CLS_METHOD_RD | CLS_METHOD_WR, cls_version_set, &h_version_set);
+  cls_register_cxx_method(h_class, "inc", CLS_METHOD_RD | CLS_METHOD_WR, cls_version_inc, &h_version_inc);
+  cls_register_cxx_method(h_class, "inc_conds", CLS_METHOD_RD | CLS_METHOD_WR, cls_version_inc, &h_version_inc_conds);
+  cls_register_cxx_method(h_class, "read", CLS_METHOD_RD, cls_version_read, &h_version_read);
+  cls_register_cxx_method(h_class, "check_conds", CLS_METHOD_RD, cls_version_check, &h_version_check_conds);
+
+  return;
+}
+
diff --git a/src/cls/version/cls_version_client.cc b/src/cls/version/cls_version_client.cc
new file mode 100644
index 00000000..3c1792aa
--- /dev/null
+++ b/src/cls/version/cls_version_client.cc
@@ -0,0 +1,101 @@
+#include <errno.h>
+
+#include "cls/version/cls_version_client.h"
+#include "include/rados/librados.hpp"
+
+
+using namespace librados;
+
+
+void cls_version_set(librados::ObjectWriteOperation& op, obj_version& objv)
+{
+  bufferlist in;
+  cls_version_set_op call;
+  call.objv = objv;
+  encode(call, in);
+  op.exec("version", "set", in);
+}
+
+void cls_version_inc(librados::ObjectWriteOperation& op)
+{
+  bufferlist in;
+  cls_version_inc_op call;
+  encode(call, in);
+  op.exec("version", "inc", in);
+}
+
+void cls_version_inc(librados::ObjectWriteOperation& op, obj_version& objv, VersionCond cond)
+{
+  bufferlist in;
+  cls_version_inc_op call;
+  call.objv = objv;
+  
+  obj_version_cond c;
+  c.cond = cond;
+  c.ver = objv;
+
+  call.conds.push_back(c);
+
+  encode(call, in);
+  op.exec("version", "inc_conds", in);
+}
+
+void cls_version_check(librados::ObjectOperation& op, obj_version& objv, VersionCond cond)
+{
+  bufferlist in;
+  cls_version_check_op call;
+  call.objv = objv;
+
+  obj_version_cond c;
+  c.cond = cond;
+  c.ver = objv;
+
+  call.conds.push_back(c);
+
+  encode(call, in);
+  op.exec("version", "check_conds", in);
+}
+
+class VersionReadCtx : public ObjectOperationCompletion {
+  obj_version *objv;
+public:
+  explicit VersionReadCtx(obj_version *_objv) : objv(_objv) {}
+  void handle_completion(int r, bufferlist& outbl) override {
+    if (r >= 0) {
+      cls_version_read_ret ret;
+      try {
+        auto iter = outbl.cbegin();
+        decode(ret, iter);
+	*objv = ret.objv;
+      } catch (buffer::error& err) {
+        // nothing we can do about it atm
+      }
+    }
+  }
+};
+
+void cls_version_read(librados::ObjectReadOperation& op, obj_version *objv)
+{
+  bufferlist inbl;
+  op.exec("version", "read", inbl, new VersionReadCtx(objv));
+}
+
+int cls_version_read(librados::IoCtx& io_ctx, string& oid, obj_version *ver)
+{
+  bufferlist in, out;
+  int r = io_ctx.exec(oid, "version", "read", in, out);
+  if (r < 0)
+    return r;
+
+  cls_version_read_ret ret;
+  try {
+    auto iter = out.cbegin();
+    decode(ret, iter);
+  } catch (buffer::error& err) {
+    return -EIO;
+  }
+
+  *ver = ret.objv;
+
+  return r;
+}
diff --git a/src/cls/version/cls_version_client.h b/src/cls/version/cls_version_client.h
new file mode 100644
index 00000000..01e7d52c
--- /dev/null
+++ b/src/cls/version/cls_version_client.h
@@ -0,0 +1,25 @@
+#ifndef CEPH_CLS_VERSION_CLIENT_H
+#define CEPH_CLS_VERSION_CLIENT_H
+
+#include "include/rados/librados_fwd.hpp"
+#include "cls_version_ops.h"
+
+/*
+ * version objclass
+ */
+
+void cls_version_set(librados::ObjectWriteOperation& op, obj_version& ver);
+
+/* increase anyway */
+void cls_version_inc(librados::ObjectWriteOperation& op);
+
+/* conditional increase, return -EAGAIN if condition fails */
+void cls_version_inc(librados::ObjectWriteOperation& op, obj_version& ver, VersionCond cond);
+
+void cls_version_read(librados::ObjectReadOperation& op, obj_version *objv);
+
+int cls_version_read(librados::IoCtx& io_ctx, string& oid, obj_version *ver);
+
+void cls_version_check(librados::ObjectOperation& op, obj_version& ver, VersionCond cond);
+
+#endif
diff --git a/src/cls/version/cls_version_ops.h b/src/cls/version/cls_version_ops.h
new file mode 100644
index 00000000..36592fa6
--- /dev/null
+++ b/src/cls/version/cls_version_ops.h
@@ -0,0 +1,92 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab
+
+#ifndef CEPH_CLS_VERSION_OPS_H
+#define CEPH_CLS_VERSION_OPS_H
+
+#include "cls_version_types.h"
+
+struct cls_version_set_op {
+  obj_version objv;
+
+  cls_version_set_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(objv, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(objv, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_version_set_op)
+
+struct cls_version_inc_op {
+  obj_version objv;
+  list<obj_version_cond> conds;
+
+  cls_version_inc_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(objv, bl);
+    encode(conds, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(objv, bl);
+    decode(conds, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_version_inc_op)
+
+struct cls_version_check_op {
+  obj_version objv;
+  list<obj_version_cond> conds;
+
+  cls_version_check_op() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(objv, bl);
+    encode(conds, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(objv, bl);
+    decode(conds, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_version_check_op)
+
+struct cls_version_read_ret {
+  obj_version objv;
+
+  cls_version_read_ret() {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(objv, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(objv, bl);
+    DECODE_FINISH(bl);
+  }
+};
+WRITE_CLASS_ENCODER(cls_version_read_ret)
+
+
+#endif
diff --git a/src/cls/version/cls_version_types.cc b/src/cls/version/cls_version_types.cc
new file mode 100644
index 00000000..4ec67867
--- /dev/null
+++ b/src/cls/version/cls_version_types.cc
@@ -0,0 +1,18 @@
+
+#include "cls/version/cls_version_types.h"
+#include "common/Formatter.h"
+#include "common/ceph_json.h"
+
+
+void obj_version::dump(Formatter *f) const
+{
+  f->dump_int("ver", ver);
+  f->dump_string("tag", tag);
+}
+
+void obj_version::decode_json(JSONObj *obj)
+{
+  JSONDecoder::decode_json("ver", ver, obj);
+  JSONDecoder::decode_json("tag", tag, obj);
+}
+
diff --git a/src/cls/version/cls_version_types.h b/src/cls/version/cls_version_types.h
new file mode 100644
index 00000000..15433b46
--- /dev/null
+++ b/src/cls/version/cls_version_types.h
@@ -0,0 +1,89 @@
+#ifndef CEPH_CLS_VERSION_TYPES_H
+#define CEPH_CLS_VERSION_TYPES_H
+
+#include "include/encoding.h"
+#include "include/types.h"
+
+class JSONObj;
+
+
+struct obj_version {
+  uint64_t ver;
+  string tag;
+
+  obj_version() : ver(0) {}
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(ver, bl);
+    encode(tag, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(ver, bl);
+    decode(tag, bl);
+    DECODE_FINISH(bl);
+  }
+
+  void inc() {
+    ver++;
+  }
+
+  void clear() {
+    ver = 0;
+    tag.clear();
+  }
+
+  bool empty() {
+    return tag.empty();
+  }
+
+  bool compare(struct obj_version *v) {
+    return (ver == v->ver &&
+            tag.compare(v->tag) == 0);
+  }
+
+  void dump(Formatter *f) const;
+  void decode_json(JSONObj *obj);
+};
+WRITE_CLASS_ENCODER(obj_version)
+
+enum VersionCond {
+  VER_COND_NONE =      0,
+  VER_COND_EQ,  /* equal */
+  VER_COND_GT,  /* greater than */
+  VER_COND_GE,  /* greater or equal */
+  VER_COND_LT,  /* less than */
+  VER_COND_LE,  /* less or equal */
+  VER_COND_TAG_EQ,
+  VER_COND_TAG_NE,
+};
+
+struct obj_version_cond {
+  struct obj_version ver;
+  VersionCond cond;
+
+  void encode(bufferlist& bl) const {
+    ENCODE_START(1, 1, bl);
+    encode(ver, bl);
+    uint32_t c = (uint32_t)cond;
+    encode(c, bl);
+    ENCODE_FINISH(bl);
+  }
+
+  void decode(bufferlist::const_iterator& bl) {
+    DECODE_START(1, bl);
+    decode(ver, bl);
+    uint32_t c;
+    decode(c, bl);
+    cond = (VersionCond)c;
+    DECODE_FINISH(bl);
+  }
+
+};
+WRITE_CLASS_ENCODER(obj_version_cond)
+
+
+#endif