1 files changed, 282 insertions, 0 deletions
diff --git a/src/rgw/driver/rados/rgw_putobj_processor.h b/src/rgw/driver/rados/rgw_putobj_processor.h
new file mode 100644
index 000000000..fa9200f32
--- /dev/null
+++ b/src/rgw/driver/rados/rgw_putobj_processor.h
@@ -0,0 +1,282 @@
+// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*-
+// vim: ts=8 sw=2 smarttab ft=cpp
+
+/*
+ * Ceph - scalable distributed file system
+ *
+ * Copyright (C) 2018 Red Hat, Inc.
+ *
+ * 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.
+ *
+ */
+
+#pragma once
+
+#include <optional>
+
+#include "rgw_putobj.h"
+#include "services/svc_rados.h"
+#include "services/svc_tier_rados.h"
+#include "rgw_sal.h"
+#include "rgw_obj_manifest.h"
+
+namespace rgw {
+
+namespace sal {
+  class RadosStore;
+}
+
+class Aio;
+
+namespace putobj {
+
+// an object processor with special handling for the first chunk of the head.
+// the virtual process_first_chunk() function returns a processor to handle the
+// rest of the object
+class HeadObjectProcessor : public rgw::sal::ObjectProcessor {
+  uint64_t head_chunk_size;
+  // buffer to capture the first chunk of the head object
+  bufferlist head_data;
+  // initialized after process_first_chunk() to process everything else
+  rgw::sal::DataProcessor *processor = nullptr;
+  uint64_t data_offset = 0; // maximum offset of data written (ie compressed)
+ protected:
+  uint64_t get_actual_size() const { return data_offset; }
+
+  // process the first chunk of data and return a processor for the rest
+  virtual int process_first_chunk(bufferlist&& data,
+                                  rgw::sal::DataProcessor **processor) = 0;
+ public:
+  HeadObjectProcessor(uint64_t head_chunk_size)
+    : head_chunk_size(head_chunk_size)
+  {}
+
+  void set_head_chunk_size(uint64_t size) { head_chunk_size = size; }
+
+  // cache first chunk for process_first_chunk(), then forward everything else
+  // to the returned processor
+  int process(bufferlist&& data, uint64_t logical_offset) final override;
+};
+
+using RawObjSet = std::set<rgw_raw_obj>;
+
+// a data sink that writes to rados objects and deletes them on cancelation
+class RadosWriter : public rgw::sal::DataProcessor {
+  Aio *const aio;
+  RGWRados *const store;
+  const RGWBucketInfo& bucket_info;
+  RGWObjectCtx& obj_ctx;
+  const rgw_obj head_obj;
+  RGWSI_RADOS::Obj stripe_obj; // current stripe object
+  RawObjSet written; // set of written objects for deletion
+  const DoutPrefixProvider *dpp;
+  optional_yield y;
+
+ public:
+  RadosWriter(Aio *aio, RGWRados *store,
+              const RGWBucketInfo& bucket_info,
+              RGWObjectCtx& obj_ctx, const rgw_obj& _head_obj,
+              const DoutPrefixProvider *dpp, optional_yield y)
+    : aio(aio), store(store), bucket_info(bucket_info),
+      obj_ctx(obj_ctx), head_obj(_head_obj), dpp(dpp), y(y)
+  {}
+  ~RadosWriter();
+
+  // add alloc hint to osd
+  void add_write_hint(librados::ObjectWriteOperation& op);
+
+  // change the current stripe object
+  int set_stripe_obj(const rgw_raw_obj& obj);
+
+  // write the data at the given offset of the current stripe object
+  int process(bufferlist&& data, uint64_t stripe_offset) override;
+
+  // write the data as an exclusive create and wait for it to complete
+  int write_exclusive(const bufferlist& data);
+
+  int drain();
+
+  // when the operation completes successfully, clear the set of written objects
+  // so they aren't deleted on destruction
+  void clear_written() { written.clear(); }
+
+};
+
+
+// a rados object processor that stripes according to RGWObjManifest
+class ManifestObjectProcessor : public HeadObjectProcessor,
+                                public StripeGenerator {
+ protected:
+  RGWRados* const store;
+  RGWBucketInfo& bucket_info;
+  rgw_placement_rule tail_placement_rule;
+  rgw_user owner;
+  RGWObjectCtx& obj_ctx;
+  rgw_obj head_obj;
+
+  RadosWriter writer;
+  RGWObjManifest manifest;
+  RGWObjManifest::generator manifest_gen;
+  ChunkProcessor chunk;
+  StripeProcessor stripe;
+  const DoutPrefixProvider *dpp;
+
+  // implements StripeGenerator
+  int next(uint64_t offset, uint64_t *stripe_size) override;
+
+ public:
+  ManifestObjectProcessor(Aio *aio, RGWRados* store,
+                          RGWBucketInfo& bucket_info,
+                          const rgw_placement_rule *ptail_placement_rule,
+                          const rgw_user& owner, RGWObjectCtx& _obj_ctx,
+                          const rgw_obj& _head_obj,
+                          const DoutPrefixProvider* dpp, optional_yield y)
+    : HeadObjectProcessor(0),
+      store(store), bucket_info(bucket_info),
+      owner(owner),
+      obj_ctx(_obj_ctx), head_obj(_head_obj),
+      writer(aio, store, bucket_info, obj_ctx, head_obj, dpp, y),
+      chunk(&writer, 0), stripe(&chunk, this, 0), dpp(dpp) {
+        if (ptail_placement_rule) {
+          tail_placement_rule = *ptail_placement_rule;
+        }
+      }
+
+  void set_owner(const rgw_user& _owner) {
+    owner = _owner;
+  }
+
+  void set_tail_placement(const rgw_placement_rule& tpr) {
+    tail_placement_rule = tpr;
+  }
+  void set_tail_placement(const rgw_placement_rule&& tpr) {
+    tail_placement_rule = tpr;
+  }
+
+};
+
+
+// a processor that completes with an atomic write to the head object as part of
+// a bucket index transaction
+class AtomicObjectProcessor : public ManifestObjectProcessor {
+  const std::optional<uint64_t> olh_epoch;
+  const std::string unique_tag;
+  bufferlist first_chunk; // written with the head in complete()
+
+  int process_first_chunk(bufferlist&& data, rgw::sal::DataProcessor **processor) override;
+ public:
+  AtomicObjectProcessor(Aio *aio, RGWRados* store,
+                        RGWBucketInfo& bucket_info,
+                        const rgw_placement_rule *ptail_placement_rule,
+                        const rgw_user& owner,
+                        RGWObjectCtx& obj_ctx, const rgw_obj& _head_obj,
+                        std::optional<uint64_t> olh_epoch,
+                        const std::string& unique_tag,
+                        const DoutPrefixProvider *dpp, optional_yield y)
+    : ManifestObjectProcessor(aio, store, bucket_info, ptail_placement_rule,
+                              owner, obj_ctx, _head_obj, dpp, y),
+      olh_epoch(olh_epoch), unique_tag(unique_tag)
+  {}
+
+  // prepare a trivial manifest
+  int prepare(optional_yield y) override;
+  // write the head object atomically in a bucket index transaction
+  int complete(size_t accounted_size, const std::string& etag,
+               ceph::real_time *mtime, ceph::real_time set_mtime,
+               std::map<std::string, bufferlist>& attrs,
+               ceph::real_time delete_at,
+               const char *if_match, const char *if_nomatch,
+               const std::string *user_data,
+               rgw_zone_set *zones_trace, bool *canceled,
+               optional_yield y) override;
+
+};
+
+
+// a processor for multipart parts, which don't require atomic completion. the
+// part's head is written with an exclusive create to detect racing uploads of
+// the same part/upload id, which are restarted with a random oid prefix
+class MultipartObjectProcessor : public ManifestObjectProcessor {
+  const rgw_obj target_obj; // target multipart object
+  const std::string upload_id;
+  const int part_num;
+  const std::string part_num_str;
+  RGWMPObj mp;
+
+  // write the first chunk and wait on aio->drain() for its completion.
+  // on EEXIST, retry with random prefix
+  int process_first_chunk(bufferlist&& data, rgw::sal::DataProcessor **processor) override;
+  // prepare the head stripe and manifest
+  int prepare_head();
+ public:
+  MultipartObjectProcessor(Aio *aio, RGWRados* store,
+                           RGWBucketInfo& bucket_info,
+                           const rgw_placement_rule *ptail_placement_rule,
+                           const rgw_user& owner, RGWObjectCtx& obj_ctx,
+                           const rgw_obj& _head_obj,
+                           const std::string& upload_id, uint64_t part_num,
+                           const std::string& part_num_str,
+                           const DoutPrefixProvider *dpp, optional_yield y)
+    : ManifestObjectProcessor(aio, store, bucket_info, ptail_placement_rule,
+                              owner, obj_ctx, _head_obj, dpp, y),
+      target_obj(head_obj), upload_id(upload_id),
+      part_num(part_num), part_num_str(part_num_str),
+      mp(head_obj.key.name, upload_id)
+  {}
+
+  // prepare a multipart manifest
+  int prepare(optional_yield y) override;
+  // write the head object attributes in a bucket index transaction, then
+  // register the completed part with the multipart meta object
+  int complete(size_t accounted_size, const std::string& etag,
+               ceph::real_time *mtime, ceph::real_time set_mtime,
+               std::map<std::string, bufferlist>& attrs,
+               ceph::real_time delete_at,
+               const char *if_match, const char *if_nomatch,
+               const std::string *user_data,
+               rgw_zone_set *zones_trace, bool *canceled,
+               optional_yield y) override;
+
+};
+
+  class AppendObjectProcessor : public ManifestObjectProcessor {
+    uint64_t cur_part_num;
+    uint64_t position;
+    uint64_t cur_size;
+    uint64_t *cur_accounted_size;
+    std::string cur_etag;
+    const std::string unique_tag;
+
+    RGWObjManifest *cur_manifest;
+
+    int process_first_chunk(bufferlist&& data, rgw::sal::DataProcessor **processor) override;
+
+  public:
+    AppendObjectProcessor(Aio *aio, RGWRados* store,
+                          RGWBucketInfo& bucket_info,
+                          const rgw_placement_rule *ptail_placement_rule,
+                          const rgw_user& owner, RGWObjectCtx& obj_ctx,
+                          const rgw_obj& _head_obj,
+                          const std::string& unique_tag, uint64_t position,
+                          uint64_t *cur_accounted_size,
+                          const DoutPrefixProvider *dpp, optional_yield y)
+            : ManifestObjectProcessor(aio, store, bucket_info, ptail_placement_rule,
+                                      owner, obj_ctx, _head_obj, dpp, y),
+              position(position), cur_size(0), cur_accounted_size(cur_accounted_size),
+              unique_tag(unique_tag), cur_manifest(nullptr)
+    {}
+    int prepare(optional_yield y) override;
+    int complete(size_t accounted_size, const std::string& etag,
+                 ceph::real_time *mtime, ceph::real_time set_mtime,
+                 std::map<std::string, bufferlist>& attrs, ceph::real_time delete_at,
+                 const char *if_match, const char *if_nomatch, const std::string *user_data,
+                 rgw_zone_set *zones_trace, bool *canceled,
+                 optional_yield y) override;
+  };
+
+} // namespace putobj
+} // namespace rgw
+