summaryrefslogtreecommitdiffstats
path: root/src/erasure-code/lrc/ErasureCodeLrc.cc
diff options
context:
space:
mode:
Diffstat (limited to '')
-rw-r--r--src/erasure-code/lrc/ErasureCodeLrc.cc862
1 files changed, 862 insertions, 0 deletions
diff --git a/src/erasure-code/lrc/ErasureCodeLrc.cc b/src/erasure-code/lrc/ErasureCodeLrc.cc
new file mode 100644
index 00000000..c882f868
--- /dev/null
+++ b/src/erasure-code/lrc/ErasureCodeLrc.cc
@@ -0,0 +1,862 @@
+// -*- 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) 2014 Cloudwatt <libre.licensing@cloudwatt.com>
+ * Copyright (C) 2014 Red Hat <contact@redhat.com>
+ *
+ * Author: Loic Dachary <loic@dachary.org>
+ *
+ * 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 <errno.h>
+#include <algorithm>
+
+#include "include/str_map.h"
+#include "common/debug.h"
+#include "crush/CrushWrapper.h"
+#include "osd/osd_types.h"
+#include "include/stringify.h"
+#include "erasure-code/ErasureCodePlugin.h"
+#include "json_spirit/json_spirit_writer.h"
+
+#include "ErasureCodeLrc.h"
+
+// re-include our assert to clobber boost's
+#include "include/ceph_assert.h"
+
+#define dout_context g_ceph_context
+#define dout_subsys ceph_subsys_osd
+#undef dout_prefix
+#define dout_prefix _prefix(_dout)
+
+using namespace std;
+
+static ostream& _prefix(std::ostream* _dout)
+{
+ return *_dout << "ErasureCodeLrc: ";
+}
+
+int ErasureCodeLrc::create_rule(const string &name,
+ CrushWrapper &crush,
+ ostream *ss) const
+{
+ if (crush.rule_exists(name)) {
+ *ss << "rule " << name << " exists";
+ return -EEXIST;
+ }
+ if (!crush.name_exists(rule_root)) {
+ *ss << "root item " << rule_root << " does not exist";
+ return -ENOENT;
+ }
+ int root = crush.get_item_id(rule_root);
+ if (rule_device_class.size()) {
+ if (!crush.class_exists(rule_device_class)) {
+ *ss << "device class " << rule_device_class << " does not exist";
+ return -ENOENT;
+ }
+ int c = crush.get_class_id(rule_device_class);
+ if (crush.class_bucket.count(root) == 0 ||
+ crush.class_bucket[root].count(c) == 0) {
+ *ss << "root item " << rule_root << " has no devices with class "
+ << rule_device_class;
+ return -EINVAL;
+ }
+ root = crush.class_bucket[root][c];
+ }
+
+ int rno = 0;
+ for (rno = 0; rno < crush.get_max_rules(); rno++) {
+ if (!crush.rule_exists(rno) && !crush.ruleset_exists(rno))
+ break;
+ }
+
+ int steps = 4 + rule_steps.size();
+ int min_rep = 3;
+ int max_rep = get_chunk_count();
+ int ret;
+ ret = crush.add_rule(rno, steps, pg_pool_t::TYPE_ERASURE,
+ min_rep, max_rep);
+ ceph_assert(ret == rno);
+ int step = 0;
+
+ ret = crush.set_rule_step(rno, step++, CRUSH_RULE_SET_CHOOSELEAF_TRIES, 5, 0);
+ ceph_assert(ret == 0);
+ ret = crush.set_rule_step(rno, step++, CRUSH_RULE_SET_CHOOSE_TRIES, 100, 0);
+ ceph_assert(ret == 0);
+ ret = crush.set_rule_step(rno, step++, CRUSH_RULE_TAKE, root, 0);
+ ceph_assert(ret == 0);
+ // [ [ "choose", "rack", 2 ],
+ // [ "chooseleaf", "host", 5 ] ]
+ for (vector<Step>::const_iterator i = rule_steps.begin();
+ i != rule_steps.end();
+ ++i) {
+ int op = i->op == "chooseleaf" ?
+ CRUSH_RULE_CHOOSELEAF_INDEP : CRUSH_RULE_CHOOSE_INDEP;
+ int type = crush.get_type_id(i->type);
+ if (type < 0) {
+ *ss << "unknown crush type " << i->type;
+ return -EINVAL;
+ }
+ ret = crush.set_rule_step(rno, step++, op, i->n, type);
+ ceph_assert(ret == 0);
+ }
+ ret = crush.set_rule_step(rno, step++, CRUSH_RULE_EMIT, 0, 0);
+ ceph_assert(ret == 0);
+ crush.set_rule_name(rno, name);
+ return rno;
+}
+
+int ErasureCodeLrc::layers_description(const ErasureCodeProfile &profile,
+ json_spirit::mArray *description,
+ ostream *ss) const
+{
+ if (profile.count("layers") == 0) {
+ *ss << "could not find 'layers' in " << profile << std::endl;
+ return ERROR_LRC_DESCRIPTION;
+ }
+ string str = profile.find("layers")->second;
+ try {
+ json_spirit::mValue json;
+ json_spirit::read_or_throw(str, json);
+
+ if (json.type() != json_spirit::array_type) {
+ *ss << "layers='" << str
+ << "' must be a JSON array but is of type "
+ << json.type() << " instead" << std::endl;
+ return ERROR_LRC_ARRAY;
+ }
+ *description = json.get_array();
+ } catch (json_spirit::Error_position &e) {
+ *ss << "failed to parse layers='" << str << "'"
+ << " at line " << e.line_ << ", column " << e.column_
+ << " : " << e.reason_ << std::endl;
+ return ERROR_LRC_PARSE_JSON;
+ }
+ return 0;
+}
+
+int ErasureCodeLrc::layers_parse(const string &description_string,
+ json_spirit::mArray description,
+ ostream *ss)
+{
+ int position = 0;
+ for (vector<json_spirit::mValue>::iterator i = description.begin();
+ i != description.end();
+ ++i, position++) {
+ if (i->type() != json_spirit::array_type) {
+ stringstream json_string;
+ json_spirit::write(*i, json_string);
+ *ss << "each element of the array "
+ << description_string << " must be a JSON array but "
+ << json_string.str() << " at position " << position
+ << " is of type " << i->type() << " instead" << std::endl;
+ return ERROR_LRC_ARRAY;
+ }
+ json_spirit::mArray layer_json = i->get_array();
+ ErasureCodeProfile profile;
+ int index = 0;
+ for (vector<json_spirit::mValue>::iterator j = layer_json.begin();
+ j != layer_json.end();
+ ++j, ++index) {
+ if (index == 0) {
+ if (j->type() != json_spirit::str_type) {
+ stringstream element;
+ json_spirit::write(*j, element);
+ *ss << "the first element of the entry "
+ << element.str() << " (first is zero) "
+ << position << " in " << description_string
+ << " is of type " << (*j).type() << " instead of string" << std::endl;
+ return ERROR_LRC_STR;
+ }
+ layers.push_back(Layer(j->get_str()));
+ Layer &layer = layers.back();
+ layer.chunks_map = j->get_str();
+ } else if(index == 1) {
+ Layer &layer = layers.back();
+ if (j->type() != json_spirit::str_type &&
+ j->type() != json_spirit::obj_type) {
+ stringstream element;
+ json_spirit::write(*j, element);
+ *ss << "the second element of the entry "
+ << element.str() << " (first is zero) "
+ << position << " in " << description_string
+ << " is of type " << (*j).type() << " instead of string or object"
+ << std::endl;
+ return ERROR_LRC_CONFIG_OPTIONS;
+ }
+ if (j->type() == json_spirit::str_type) {
+ int err = get_json_str_map(j->get_str(), *ss, &layer.profile);
+ if (err)
+ return err;
+ } else if (j->type() == json_spirit::obj_type) {
+ json_spirit::mObject o = j->get_obj();
+
+ for (map<string, json_spirit::mValue>::iterator i = o.begin();
+ i != o.end();
+ ++i) {
+ layer.profile[i->first] = i->second.get_str();
+ }
+ }
+ } else {
+ // ignore trailing elements
+ }
+ }
+ }
+ return 0;
+}
+
+int ErasureCodeLrc::layers_init(ostream *ss)
+{
+ ErasureCodePluginRegistry &registry = ErasureCodePluginRegistry::instance();
+ for (unsigned int i = 0; i < layers.size(); i++) {
+ Layer &layer = layers[i];
+ int position = 0;
+ for(std::string::iterator it = layer.chunks_map.begin();
+ it != layer.chunks_map.end();
+ ++it) {
+ if (*it == 'D')
+ layer.data.push_back(position);
+ if (*it == 'c')
+ layer.coding.push_back(position);
+ if (*it == 'c' || *it == 'D')
+ layer.chunks_as_set.insert(position);
+ position++;
+ }
+ layer.chunks = layer.data;
+ layer.chunks.insert(layer.chunks.end(),
+ layer.coding.begin(), layer.coding.end());
+ if (layer.profile.find("k") == layer.profile.end())
+ layer.profile["k"] = stringify(layer.data.size());
+ if (layer.profile.find("m") == layer.profile.end())
+ layer.profile["m"] = stringify(layer.coding.size());
+ if (layer.profile.find("plugin") == layer.profile.end())
+ layer.profile["plugin"] = "jerasure";
+ if (layer.profile.find("technique") == layer.profile.end())
+ layer.profile["technique"] = "reed_sol_van";
+ int err = registry.factory(layer.profile["plugin"],
+ directory,
+ layer.profile,
+ &layer.erasure_code,
+ ss);
+ if (err)
+ return err;
+ }
+ return 0;
+}
+
+int ErasureCodeLrc::layers_sanity_checks(const string &description_string,
+ ostream *ss) const
+{
+ int position = 0;
+
+ if (layers.size() < 1) {
+ *ss << "layers parameter has " << layers.size()
+ << " which is less than the minimum of one. "
+ << description_string << std::endl;
+ return ERROR_LRC_LAYERS_COUNT;
+ }
+ for (vector<Layer>::const_iterator layer = layers.begin();
+ layer != layers.end();
+ ++layer) {
+ if (chunk_count != layer->chunks_map.length()) {
+ *ss << "the first element of the array at position "
+ << position << " (starting from zero) "
+ << " is the string '" << layer->chunks_map
+ << " found in the layers parameter "
+ << description_string << ". It is expected to be "
+ << chunk_count << " characters long but is "
+ << layer->chunks_map.length() << " characters long instead "
+ << std::endl;
+ return ERROR_LRC_MAPPING_SIZE;
+ }
+ }
+ return 0;
+}
+
+int ErasureCodeLrc::parse(ErasureCodeProfile &profile,
+ ostream *ss)
+{
+ int r = ErasureCode::parse(profile, ss);
+ if (r)
+ return r;
+
+ return parse_rule(profile, ss);
+}
+
+const string ErasureCodeLrc::DEFAULT_KML("-1");
+
+int ErasureCodeLrc::parse_kml(ErasureCodeProfile &profile,
+ ostream *ss)
+{
+ int err = ErasureCode::parse(profile, ss);
+ const int DEFAULT_INT = -1;
+ int k, m, l;
+ err |= to_int("k", profile, &k, DEFAULT_KML, ss);
+ err |= to_int("m", profile, &m, DEFAULT_KML, ss);
+ err |= to_int("l", profile, &l, DEFAULT_KML, ss);
+
+ if (k == DEFAULT_INT && m == DEFAULT_INT && l == DEFAULT_INT)
+ return err;
+
+ if ((k != DEFAULT_INT || m != DEFAULT_INT || l != DEFAULT_INT) &&
+ (k == DEFAULT_INT || m == DEFAULT_INT || l == DEFAULT_INT)) {
+ *ss << "All of k, m, l must be set or none of them in "
+ << profile << std::endl;
+ return ERROR_LRC_ALL_OR_NOTHING;
+ }
+
+ const char *generated[] = { "mapping",
+ "layers",
+ "crush-steps" };
+
+ for (int i = 0; i < 3; i++) {
+ if (profile.count(generated[i])) {
+ *ss << "The " << generated[i] << " parameter cannot be set "
+ << "when k, m, l are set in " << profile << std::endl;
+ return ERROR_LRC_GENERATED;
+ }
+ }
+
+ if (l == 0 || (k + m) % l) {
+ *ss << "k + m must be a multiple of l in "
+ << profile << std::endl;
+ return ERROR_LRC_K_M_MODULO;
+ }
+
+ int local_group_count = (k + m) / l;
+
+ if (k % local_group_count) {
+ *ss << "k must be a multiple of (k + m) / l in "
+ << profile << std::endl;
+ return ERROR_LRC_K_MODULO;
+ }
+
+ if (m % local_group_count) {
+ *ss << "m must be a multiple of (k + m) / l in "
+ << profile << std::endl;
+ return ERROR_LRC_M_MODULO;
+ }
+
+ string mapping;
+ for (int i = 0; i < local_group_count; i++) {
+ mapping += string(k / local_group_count, 'D') +
+ string(m / local_group_count, '_') + "_";
+ }
+ profile["mapping"] = mapping;
+
+ string layers = "[ ";
+
+ // global layer
+ layers += " [ \"";
+ for (int i = 0; i < local_group_count; i++) {
+ layers += string(k / local_group_count, 'D') +
+ string(m / local_group_count, 'c') + "_";
+ }
+ layers += "\", \"\" ],";
+
+ // local layers
+ for (int i = 0; i < local_group_count; i++) {
+ layers += " [ \"";
+ for (int j = 0; j < local_group_count; j++) {
+ if (i == j)
+ layers += string(l, 'D') + "c";
+ else
+ layers += string(l + 1, '_');
+ }
+ layers += "\", \"\" ],";
+ }
+ profile["layers"] = layers + "]";
+
+ ErasureCodeProfile::const_iterator parameter;
+ string rule_locality;
+ parameter = profile.find("crush-locality");
+ if (parameter != profile.end())
+ rule_locality = parameter->second;
+ string rule_failure_domain = "host";
+ parameter = profile.find("crush-failure-domain");
+ if (parameter != profile.end())
+ rule_failure_domain = parameter->second;
+
+ if (rule_locality != "") {
+ rule_steps.clear();
+ rule_steps.push_back(Step("choose", rule_locality,
+ local_group_count));
+ rule_steps.push_back(Step("chooseleaf", rule_failure_domain,
+ l + 1));
+ } else if (rule_failure_domain != "") {
+ rule_steps.clear();
+ rule_steps.push_back(Step("chooseleaf", rule_failure_domain, 0));
+ }
+
+ return err;
+}
+
+int ErasureCodeLrc::parse_rule(ErasureCodeProfile &profile,
+ ostream *ss)
+{
+ int err = 0;
+ err |= to_string("crush-root", profile,
+ &rule_root,
+ "default", ss);
+ err |= to_string("crush-device-class", profile,
+ &rule_device_class,
+ "", ss);
+
+ if (profile.count("crush-steps") != 0) {
+ rule_steps.clear();
+ string str = profile.find("crush-steps")->second;
+ json_spirit::mArray description;
+ try {
+ json_spirit::mValue json;
+ json_spirit::read_or_throw(str, json);
+
+ if (json.type() != json_spirit::array_type) {
+ *ss << "crush-steps='" << str
+ << "' must be a JSON array but is of type "
+ << json.type() << " instead" << std::endl;
+ return ERROR_LRC_ARRAY;
+ }
+ description = json.get_array();
+ } catch (json_spirit::Error_position &e) {
+ *ss << "failed to parse crush-steps='" << str << "'"
+ << " at line " << e.line_ << ", column " << e.column_
+ << " : " << e.reason_ << std::endl;
+ return ERROR_LRC_PARSE_JSON;
+ }
+
+ int position = 0;
+ for (vector<json_spirit::mValue>::iterator i = description.begin();
+ i != description.end();
+ ++i, position++) {
+ if (i->type() != json_spirit::array_type) {
+ stringstream json_string;
+ json_spirit::write(*i, json_string);
+ *ss << "element of the array "
+ << str << " must be a JSON array but "
+ << json_string.str() << " at position " << position
+ << " is of type " << i->type() << " instead" << std::endl;
+ return ERROR_LRC_ARRAY;
+ }
+ int r = parse_rule_step(str, i->get_array(), ss);
+ if (r)
+ return r;
+ }
+ }
+ return 0;
+}
+
+int ErasureCodeLrc::parse_rule_step(const string &description_string,
+ json_spirit::mArray description,
+ ostream *ss)
+{
+ stringstream json_string;
+ json_spirit::write(description, json_string);
+ string op;
+ string type;
+ int n = 0;
+ int position = 0;
+ for (vector<json_spirit::mValue>::iterator i = description.begin();
+ i != description.end();
+ ++i, position++) {
+ if ((position == 0 || position == 1) &&
+ i->type() != json_spirit::str_type) {
+ *ss << "element " << position << " of the array "
+ << json_string.str() << " found in " << description_string
+ << " must be a JSON string but is of type "
+ << i->type() << " instead" << std::endl;
+ return position == 0 ? ERROR_LRC_RULE_OP : ERROR_LRC_RULE_TYPE;
+ }
+ if (position == 2 && i->type() != json_spirit::int_type) {
+ *ss << "element " << position << " of the array "
+ << json_string.str() << " found in " << description_string
+ << " must be a JSON int but is of type "
+ << i->type() << " instead" << std::endl;
+ return ERROR_LRC_RULE_N;
+ }
+
+ if (position == 0)
+ op = i->get_str();
+ if (position == 1)
+ type = i->get_str();
+ if (position == 2)
+ n = i->get_int();
+ }
+ rule_steps.push_back(Step(op, type, n));
+ return 0;
+}
+
+int ErasureCodeLrc::init(ErasureCodeProfile &profile,
+ ostream *ss)
+{
+ int r;
+
+ r = parse_kml(profile, ss);
+ if (r)
+ return r;
+
+ r = parse(profile, ss);
+ if (r)
+ return r;
+
+ json_spirit::mArray description;
+ r = layers_description(profile, &description, ss);
+ if (r)
+ return r;
+
+ string description_string = profile.find("layers")->second;
+
+ dout(10) << "init(" << description_string << ")" << dendl;
+
+ r = layers_parse(description_string, description, ss);
+ if (r)
+ return r;
+
+ r = layers_init(ss);
+ if (r)
+ return r;
+
+ if (profile.count("mapping") == 0) {
+ *ss << "the 'mapping' profile is missing from " << profile;
+ return ERROR_LRC_MAPPING;
+ }
+ string mapping = profile.find("mapping")->second;
+ data_chunk_count = count(begin(mapping), end(mapping), 'D');
+ chunk_count = mapping.length();
+
+ r = layers_sanity_checks(description_string, ss);
+ if (r)
+ return r;
+
+ //
+ // When initialized with kml, the profile parameters
+ // that were generated should not be stored because
+ // they would otherwise be exposed to the caller.
+ //
+ if (profile.find("l") != profile.end() &&
+ profile.find("l")->second != DEFAULT_KML) {
+ profile.erase("mapping");
+ profile.erase("layers");
+ }
+ ErasureCode::init(profile, ss);
+ return 0;
+}
+
+set<int> ErasureCodeLrc::get_erasures(const set<int> &want,
+ const set<int> &available) const
+{
+ set<int> result;
+ set_difference(want.begin(), want.end(),
+ available.begin(), available.end(),
+ inserter(result, result.end()));
+ return result;
+}
+
+unsigned int ErasureCodeLrc::get_chunk_size(unsigned int object_size) const
+{
+ return layers.front().erasure_code->get_chunk_size(object_size);
+}
+
+void p(const set<int> &s) { cerr << s; } // for gdb
+
+int ErasureCodeLrc::_minimum_to_decode(const set<int> &want_to_read,
+ const set<int> &available_chunks,
+ set<int> *minimum)
+{
+ dout(20) << __func__ << " want_to_read " << want_to_read
+ << " available_chunks " << available_chunks << dendl;
+ {
+ set<int> erasures_total;
+ set<int> erasures_not_recovered;
+ set<int> erasures_want;
+ for (unsigned int i = 0; i < get_chunk_count(); ++i) {
+ if (available_chunks.count(i) == 0) {
+ erasures_total.insert(i);
+ erasures_not_recovered.insert(i);
+ if (want_to_read.count(i) != 0)
+ erasures_want.insert(i);
+ }
+ }
+
+ //
+ // Case 1:
+ //
+ // When no chunk is missing there is no need to read more than what
+ // is wanted.
+ //
+ if (erasures_want.empty()) {
+ *minimum = want_to_read;
+ dout(20) << __func__ << " minimum == want_to_read == "
+ << want_to_read << dendl;
+ return 0;
+ }
+
+ //
+ // Case 2:
+ //
+ // Try to recover erasures with as few chunks as possible.
+ //
+ for (vector<Layer>::reverse_iterator i = layers.rbegin();
+ i != layers.rend();
+ ++i) {
+ //
+ // If this layer has no chunk that we want, skip it.
+ //
+ set<int> layer_want;
+ set_intersection(want_to_read.begin(), want_to_read.end(),
+ i->chunks_as_set.begin(), i->chunks_as_set.end(),
+ inserter(layer_want, layer_want.end()));
+ if (layer_want.empty())
+ continue;
+ //
+ // Are some of the chunks we want missing ?
+ //
+ set<int> layer_erasures;
+ set_intersection(layer_want.begin(), layer_want.end(),
+ erasures_want.begin(), erasures_want.end(),
+ inserter(layer_erasures, layer_erasures.end()));
+ set<int> layer_minimum;
+ if (layer_erasures.empty()) {
+ //
+ // The chunks we want are available, this is the minimum we need
+ // to read.
+ //
+ layer_minimum = layer_want;
+ } else {
+ set<int> erasures;
+ set_intersection(i->chunks_as_set.begin(), i->chunks_as_set.end(),
+ erasures_not_recovered.begin(), erasures_not_recovered.end(),
+ inserter(erasures, erasures.end()));
+
+ if (erasures.size() > i->erasure_code->get_coding_chunk_count()) {
+ //
+ // There are too many erasures for this layer to recover: skip
+ // it and hope that an upper layer will be do better.
+ //
+ continue;
+ } else {
+ //
+ // Get all available chunks in that layer to recover the
+ // missing one(s).
+ //
+ set_difference(i->chunks_as_set.begin(), i->chunks_as_set.end(),
+ erasures_not_recovered.begin(), erasures_not_recovered.end(),
+ inserter(layer_minimum, layer_minimum.end()));
+ //
+ // Chunks recovered by this layer are removed from the list of
+ // erasures so that upper levels do not attempt to recover
+ // them.
+ //
+ for (set<int>::const_iterator j = erasures.begin();
+ j != erasures.end();
+ ++j) {
+ erasures_not_recovered.erase(*j);
+ erasures_want.erase(*j);
+ }
+ }
+ }
+ minimum->insert(layer_minimum.begin(), layer_minimum.end());
+ }
+ if (erasures_want.empty()) {
+ minimum->insert(want_to_read.begin(), want_to_read.end());
+ for (set<int>::const_iterator i = erasures_total.begin();
+ i != erasures_total.end();
+ ++i) {
+ if (minimum->count(*i))
+ minimum->erase(*i);
+ }
+ dout(20) << __func__ << " minimum = " << *minimum << dendl;
+ return 0;
+ }
+ }
+
+ {
+ //
+ // Case 3:
+ //
+ // The previous strategy failed to recover from all erasures.
+ //
+ // Try to recover as many chunks as possible, even from layers
+ // that do not contain chunks that we want, in the hope that it
+ // will help the upper layers.
+ //
+ set<int> erasures_total;
+ for (unsigned int i = 0; i < get_chunk_count(); ++i) {
+ if (available_chunks.count(i) == 0)
+ erasures_total.insert(i);
+ }
+
+ for (vector<Layer>::reverse_iterator i = layers.rbegin();
+ i != layers.rend();
+ ++i) {
+ set<int> layer_erasures;
+ set_intersection(i->chunks_as_set.begin(), i->chunks_as_set.end(),
+ erasures_total.begin(), erasures_total.end(),
+ inserter(layer_erasures, layer_erasures.end()));
+ //
+ // If this layer has no erasure, skip it
+ //
+ if (layer_erasures.empty())
+ continue;
+
+ if (layer_erasures.size() > 0 &&
+ layer_erasures.size() <= i->erasure_code->get_coding_chunk_count()) {
+ //
+ // chunks recovered by this layer are removed from the list of
+ // erasures so that upper levels know they can rely on their
+ // availability
+ //
+ for (set<int>::const_iterator j = layer_erasures.begin();
+ j != layer_erasures.end();
+ ++j) {
+ erasures_total.erase(*j);
+ }
+ }
+ }
+ if (erasures_total.empty()) {
+ //
+ // Do not try to be smart about what chunks are necessary to
+ // recover, use all available chunks.
+ //
+ *minimum = available_chunks;
+ dout(20) << __func__ << " minimum == available_chunks == "
+ << available_chunks << dendl;
+ return 0;
+ }
+ }
+
+ derr << __func__ << " not enough chunks in " << available_chunks
+ << " to read " << want_to_read << dendl;
+ return -EIO;
+}
+
+int ErasureCodeLrc::encode_chunks(const set<int> &want_to_encode,
+ map<int, bufferlist> *encoded)
+{
+ unsigned int top = layers.size();
+ for (vector<Layer>::reverse_iterator i = layers.rbegin();
+ i != layers.rend();
+ ++i) {
+ --top;
+ if (includes(i->chunks_as_set.begin(), i->chunks_as_set.end(),
+ want_to_encode.begin(), want_to_encode.end()))
+ break;
+ }
+
+ for (unsigned int i = top; i < layers.size(); ++i) {
+ const Layer &layer = layers[i];
+ set<int> layer_want_to_encode;
+ map<int, bufferlist> layer_encoded;
+ int j = 0;
+ for (const auto& c : layer.chunks) {
+ std::swap(layer_encoded[j], (*encoded)[c]);
+ if (want_to_encode.find(c) != want_to_encode.end())
+ layer_want_to_encode.insert(j);
+ j++;
+ }
+ int err = layer.erasure_code->encode_chunks(layer_want_to_encode,
+ &layer_encoded);
+ j = 0;
+ for (const auto& c : layer.chunks) {
+ std::swap(layer_encoded[j++], (*encoded)[c]);
+ }
+ if (err) {
+ derr << __func__ << " layer " << layer.chunks_map
+ << " failed with " << err << " trying to encode "
+ << layer_want_to_encode << dendl;
+ return err;
+ }
+ }
+ return 0;
+}
+
+int ErasureCodeLrc::decode_chunks(const set<int> &want_to_read,
+ const map<int, bufferlist> &chunks,
+ map<int, bufferlist> *decoded)
+{
+ set<int> available_chunks;
+ set<int> erasures;
+ for (unsigned int i = 0; i < get_chunk_count(); ++i) {
+ if (chunks.count(i) != 0)
+ available_chunks.insert(i);
+ else
+ erasures.insert(i);
+ }
+
+ set<int> want_to_read_erasures;
+
+ for (vector<Layer>::reverse_iterator layer = layers.rbegin();
+ layer != layers.rend();
+ ++layer) {
+ set<int> layer_erasures;
+ set_intersection(layer->chunks_as_set.begin(), layer->chunks_as_set.end(),
+ erasures.begin(), erasures.end(),
+ inserter(layer_erasures, layer_erasures.end()));
+
+ if (layer_erasures.size() >
+ layer->erasure_code->get_coding_chunk_count()) {
+ // skip because there are too many erasures for this layer to recover
+ } else if(layer_erasures.size() == 0) {
+ // skip because all chunks are already available
+ } else {
+ set<int> layer_want_to_read;
+ map<int, bufferlist> layer_chunks;
+ map<int, bufferlist> layer_decoded;
+ int j = 0;
+ for (vector<int>::const_iterator c = layer->chunks.begin();
+ c != layer->chunks.end();
+ ++c) {
+ //
+ // Pick chunks from *decoded* instead of *chunks* to re-use
+ // chunks recovered by previous layers. In other words
+ // *chunks* does not change but *decoded* gradually improves
+ // as more layers recover from erasures.
+ //
+ if (erasures.count(*c) == 0)
+ layer_chunks[j] = (*decoded)[*c];
+ if (want_to_read.count(*c) != 0)
+ layer_want_to_read.insert(j);
+ layer_decoded[j] = (*decoded)[*c];
+ ++j;
+ }
+ int err = layer->erasure_code->decode_chunks(layer_want_to_read,
+ layer_chunks,
+ &layer_decoded);
+ if (err) {
+ derr << __func__ << " layer " << layer->chunks_map
+ << " failed with " << err << " trying to decode "
+ << layer_want_to_read << " with " << available_chunks << dendl;
+ return err;
+ }
+ j = 0;
+ for (vector<int>::const_iterator c = layer->chunks.begin();
+ c != layer->chunks.end();
+ ++c) {
+ (*decoded)[*c] = layer_decoded[j];
+ ++j;
+ erasures.erase(*c);
+ }
+ want_to_read_erasures.clear();
+ set_intersection(erasures.begin(), erasures.end(),
+ want_to_read.begin(), want_to_read.end(),
+ inserter(want_to_read_erasures, want_to_read_erasures.end()));
+ if (want_to_read_erasures.size() == 0)
+ break;
+ }
+ }
+
+ if (want_to_read_erasures.size() > 0) {
+ derr << __func__ << " want to read " << want_to_read
+ << " with available_chunks = " << available_chunks
+ << " end up being unable to read " << want_to_read_erasures << dendl;
+ return -EIO;
+ } else {
+ return 0;
+ }
+}