[ceph.git] / ceph / src / erasure-code / ErasureCode.cc

// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
// vim: ts=8 sw=2 smarttab
/*
 * Ceph distributed storage 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 "ErasureCode.h"

#include "common/strtol.h"
#include "include/buffer.h"
#include "crush/CrushWrapper.h"
#include "osd/osd_types.h"

using namespace std;

const unsigned ErasureCode::SIMD_ALIGN = 32;

#define DEFAULT_RULE_ROOT "default"
#define DEFAULT_RULE_FAILURE_DOMAIN "host"

int ErasureCode::init(
  ErasureCodeProfile &profile,
  std::ostream *ss)
{
  int err = 0;
  err |= to_string("crush-root", profile,
		   &rule_root,
		   DEFAULT_RULE_ROOT, ss);
  err |= to_string("crush-failure-domain", profile,
		   &rule_failure_domain,
		   DEFAULT_RULE_FAILURE_DOMAIN, ss);
  err |= to_string("crush-device-class", profile,
		   &rule_device_class,
		   "", ss);
  if (err)
    return err;
  _profile = profile;
  return 0;
}

int ErasureCode::create_rule(
  const std::string &name,
  CrushWrapper &crush,
  std::ostream *ss) const
{
  int ruleid = crush.add_simple_rule(
    name,
    rule_root,
    rule_failure_domain,
    rule_device_class,
    "indep",
    pg_pool_t::TYPE_ERASURE,
    ss);

  if (ruleid < 0)
    return ruleid;

  crush.set_rule_mask_max_size(ruleid, get_chunk_count());
  return ruleid;
}

int ErasureCode::sanity_check_k(int k, ostream *ss)
{
  if (k < 2) {
    *ss << "k=" << k << " must be >= 2" << std::endl;
    return -EINVAL;
  } else {
    return 0;
  }
}

int ErasureCode::chunk_index(unsigned int i) const
{
  return chunk_mapping.size() > i ? chunk_mapping[i] : i;
}

int ErasureCode::minimum_to_decode(const set<int> &want_to_read,
                                   const set<int> &available_chunks,
                                   set<int> *minimum)
{
  if (includes(available_chunks.begin(), available_chunks.end(),
	       want_to_read.begin(), want_to_read.end())) {
    *minimum = want_to_read;
  } else {
    unsigned int k = get_data_chunk_count();
    if (available_chunks.size() < (unsigned)k)
      return -EIO;
    set<int>::iterator i;
    unsigned j;
    for (i = available_chunks.begin(), j = 0; j < (unsigned)k; ++i, j++)
      minimum->insert(*i);
  }
  return 0;
}

int ErasureCode::minimum_to_decode_with_cost(const set<int> &want_to_read,
                                             const map<int, int> &available,
                                             set<int> *minimum)
{
  set <int> available_chunks;
  for (map<int, int>::const_iterator i = available.begin();
       i != available.end();
       ++i)
    available_chunks.insert(i->first);
  return minimum_to_decode(want_to_read, available_chunks, minimum);
}

int ErasureCode::encode_prepare(const bufferlist &raw,
                                map<int, bufferlist> &encoded) const
{
  unsigned int k = get_data_chunk_count();
  unsigned int m = get_chunk_count() - k;
  unsigned blocksize = get_chunk_size(raw.length());
  unsigned padded_chunks = k - raw.length() / blocksize;
  bufferlist prepared = raw;

  for (unsigned int i = 0; i < k - padded_chunks; i++) {
    bufferlist &chunk = encoded[chunk_index(i)];
    chunk.substr_of(prepared, i * blocksize, blocksize);
    chunk.rebuild_aligned_size_and_memory(blocksize, SIMD_ALIGN);
    assert(chunk.is_contiguous());
  }
  if (padded_chunks) {
    unsigned remainder = raw.length() - (k - padded_chunks) * blocksize;
    bufferptr buf(buffer::create_aligned(blocksize, SIMD_ALIGN));

    raw.copy((k - padded_chunks) * blocksize, remainder, buf.c_str());
    buf.zero(remainder, blocksize - remainder);
    encoded[chunk_index(k-padded_chunks)].push_back(std::move(buf));

    for (unsigned int i = k - padded_chunks + 1; i < k; i++) {
      bufferptr buf(buffer::create_aligned(blocksize, SIMD_ALIGN));
      buf.zero();
      encoded[chunk_index(i)].push_back(std::move(buf));
    }
  }
  for (unsigned int i = k; i < k + m; i++) {
    bufferlist &chunk = encoded[chunk_index(i)];
    chunk.push_back(buffer::create_aligned(blocksize, SIMD_ALIGN));
  }

  return 0;
}

int ErasureCode::encode(const set<int> &want_to_encode,
                        const bufferlist &in,
                        map<int, bufferlist> *encoded)
{
  unsigned int k = get_data_chunk_count();
  unsigned int m = get_chunk_count() - k;
  bufferlist out;
  int err = encode_prepare(in, *encoded);
  if (err)
    return err;
  encode_chunks(want_to_encode, encoded);
  for (unsigned int i = 0; i < k + m; i++) {
    if (want_to_encode.count(i) == 0)
      encoded->erase(i);
  }
  return 0;
}

int ErasureCode::encode_chunks(const set<int> &want_to_encode,
                               map<int, bufferlist> *encoded)
{
  assert("ErasureCode::encode_chunks not implemented" == 0);
}
 
int ErasureCode::decode(const set<int> &want_to_read,
                        const map<int, bufferlist> &chunks,
                        map<int, bufferlist> *decoded)
{
  vector<int> have;
  have.reserve(chunks.size());
  for (map<int, bufferlist>::const_iterator i = chunks.begin();
       i != chunks.end();
       ++i) {
    have.push_back(i->first);
  }
  if (includes(
	have.begin(), have.end(), want_to_read.begin(), want_to_read.end())) {
    for (set<int>::iterator i = want_to_read.begin();
	 i != want_to_read.end();
	 ++i) {
      (*decoded)[*i] = chunks.find(*i)->second;
    }
    return 0;
  }
  unsigned int k = get_data_chunk_count();
  unsigned int m = get_chunk_count() - k;
  unsigned blocksize = (*chunks.begin()).second.length();
  for (unsigned int i =  0; i < k + m; i++) {
    if (chunks.find(i) == chunks.end()) {
      bufferptr ptr(buffer::create_aligned(blocksize, SIMD_ALIGN));
      (*decoded)[i].push_front(ptr);
    } else {
      (*decoded)[i] = chunks.find(i)->second;
      (*decoded)[i].rebuild_aligned(SIMD_ALIGN);
    }
  }
  return decode_chunks(want_to_read, chunks, decoded);
}

int ErasureCode::decode_chunks(const set<int> &want_to_read,
                               const map<int, bufferlist> &chunks,
                               map<int, bufferlist> *decoded)
{
  assert("ErasureCode::decode_chunks not implemented" == 0);
}

int ErasureCode::parse(const ErasureCodeProfile &profile,
		       ostream *ss)
{
  return to_mapping(profile, ss);
}

const vector<int> &ErasureCode::get_chunk_mapping() const {
  return chunk_mapping;
}

int ErasureCode::to_mapping(const ErasureCodeProfile &profile,
			    ostream *ss)
{
  if (profile.find("mapping") != profile.end()) {
    std::string mapping = profile.find("mapping")->second;
    int position = 0;
    vector<int> coding_chunk_mapping;
    for(std::string::iterator it = mapping.begin(); it != mapping.end(); ++it) {
      if (*it == 'D')
	chunk_mapping.push_back(position);
      else
	coding_chunk_mapping.push_back(position);
      position++;
    }
    chunk_mapping.insert(chunk_mapping.end(),
			 coding_chunk_mapping.begin(),
			 coding_chunk_mapping.end());
  }
  return 0;
}

int ErasureCode::to_int(const std::string &name,
			ErasureCodeProfile &profile,
			int *value,
			const std::string &default_value,
			ostream *ss)
{
  if (profile.find(name) == profile.end() ||
      profile.find(name)->second.size() == 0)
    profile[name] = default_value;
  std::string p = profile.find(name)->second;
  std::string err;
  int r = strict_strtol(p.c_str(), 10, &err);
  if (!err.empty()) {
    *ss << "could not convert " << name << "=" << p
	<< " to int because " << err
	<< ", set to default " << default_value << std::endl;
    *value = strict_strtol(default_value.c_str(), 10, &err);
    return -EINVAL;
  }
  *value = r;
  return 0;
}

int ErasureCode::to_bool(const std::string &name,
			 ErasureCodeProfile &profile,
			 bool *value,
			 const std::string &default_value,
			 ostream *ss)
{
  if (profile.find(name) == profile.end() ||
      profile.find(name)->second.size() == 0)
    profile[name] = default_value;
  const std::string p = profile.find(name)->second;
  *value = (p == "yes") || (p == "true");
  return 0;
}

int ErasureCode::to_string(const std::string &name,
			   ErasureCodeProfile &profile,
			   std::string *value,
			   const std::string &default_value,
			   ostream *ss)
{
  if (profile.find(name) == profile.end() ||
      profile.find(name)->second.size() == 0)
    profile[name] = default_value;
  *value = profile[name];
  return 0;
}

int ErasureCode::decode_concat(const map<int, bufferlist> &chunks,
			       bufferlist *decoded)
{
  set<int> want_to_read;

  for (unsigned int i = 0; i < get_data_chunk_count(); i++) {
    want_to_read.insert(chunk_index(i));
  }
  map<int, bufferlist> decoded_map;
  int r = decode(want_to_read, chunks, &decoded_map);
  if (r == 0) {
    for (unsigned int i = 0; i < get_data_chunk_count(); i++) {
      decoded->claim_append(decoded_map[chunk_index(i)]);
    }
  }
  return r;
}
Commit	Line	Data
7c673cae FG	1	// -- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t --
	2	// vim: ts=8 sw=2 smarttab
	3	/*
	4	* Ceph distributed storage system
	5	*
	6	* Copyright (C) 2014 Cloudwatt <libre.licensing@cloudwatt.com>
	7	* Copyright (C) 2014 Red Hat <contact@redhat.com>
	8	*
	9	* Author: Loic Dachary <loic@dachary.org>
	10	*
	11	* This library is free software; you can redistribute it and/or
	12	* modify it under the terms of the GNU Lesser General Public
	13	* License as published by the Free Software Foundation; either
	14	* version 2.1 of the License, or (at your option) any later version.
	15	*
	16	*/
	17
	18	#include <errno.h>
7c673cae	19	#include <algorithm>
7c673cae	20
7c673cae	21	#include "ErasureCode.h"
31f18b77 FG	22
31f18b77 FG	23	#include "common/strtol.h"
7c673cae	24	#include "include/buffer.h"
224ce89b WB	25	#include "crush/CrushWrapper.h"
224ce89b WB	26	#include "osd/osd_types.h"
7c673cae	27
31f18b77 FG	28	using namespace std;
31f18b77 FG	29
7c673cae FG	30	const unsigned ErasureCode::SIMD_ALIGN = 32;
7c673cae FG	31
224ce89b WB	32	#define DEFAULT_RULE_ROOT "default"
	33	#define DEFAULT_RULE_FAILURE_DOMAIN "host"
	34
	35	int ErasureCode::init(
	36	ErasureCodeProfile &profile,
	37	std::ostream *ss)
	38	{
	39	int err = 0;
	40	err \|= to_string("crush-root", profile,
	41	&rule_root,
	42	DEFAULT_RULE_ROOT, ss);
	43	err \|= to_string("crush-failure-domain", profile,
	44	&rule_failure_domain,
	45	DEFAULT_RULE_FAILURE_DOMAIN, ss);
	46	err \|= to_string("crush-device-class", profile,
	47	&rule_device_class,
	48	"", ss);
	49	if (err)
	50	return err;
	51	_profile = profile;
	52	return 0;
	53	}
	54
	55	int ErasureCode::create_rule(
	56	const std::string &name,
	57	CrushWrapper &crush,
	58	std::ostream *ss) const
	59	{
	60	int ruleid = crush.add_simple_rule(
	61	name,
	62	rule_root,
	63	rule_failure_domain,
	64	rule_device_class,
	65	"indep",
	66	pg_pool_t::TYPE_ERASURE,
	67	ss);
	68
	69	if (ruleid < 0)
	70	return ruleid;
	71
	72	crush.set_rule_mask_max_size(ruleid, get_chunk_count());
	73	return ruleid;
	74	}
	75
7c673cae FG	76	int ErasureCode::sanity_check_k(int k, ostream *ss)
	77	{
	78	if (k < 2) {
	79	*ss << "k=" << k << " must be >= 2" << std::endl;
	80	return -EINVAL;
	81	} else {
	82	return 0;
	83	}
	84	}
	85
	86	int ErasureCode::chunk_index(unsigned int i) const
	87	{
	88	return chunk_mapping.size() > i ? chunk_mapping[i] : i;
	89	}
	90
	91	int ErasureCode::minimum_to_decode(const set<int> &want_to_read,
	92	const set<int> &available_chunks,
	93	set<int> *minimum)
	94	{
	95	if (includes(available_chunks.begin(), available_chunks.end(),
	96	want_to_read.begin(), want_to_read.end())) {
	97	*minimum = want_to_read;
	98	} else {
	99	unsigned int k = get_data_chunk_count();
	100	if (available_chunks.size() < (unsigned)k)
	101	return -EIO;
	102	set<int>::iterator i;
	103	unsigned j;
	104	for (i = available_chunks.begin(), j = 0; j < (unsigned)k; ++i, j++)
	105	minimum->insert(*i);
	106	}
	107	return 0;
	108	}
	109
	110	int ErasureCode::minimum_to_decode_with_cost(const set<int> &want_to_read,
	111	const map<int, int> &available,
	112	set<int> *minimum)
	113	{
	114	set <int> available_chunks;
	115	for (map<int, int>::const_iterator i = available.begin();
	116	i != available.end();
	117	++i)
	118	available_chunks.insert(i->first);
	119	return minimum_to_decode(want_to_read, available_chunks, minimum);
	120	}
	121
	122	int ErasureCode::encode_prepare(const bufferlist &raw,
	123	map<int, bufferlist> &encoded) const
	124	{
	125	unsigned int k = get_data_chunk_count();
	126	unsigned int m = get_chunk_count() - k;
	127	unsigned blocksize = get_chunk_size(raw.length());
	128	unsigned padded_chunks = k - raw.length() / blocksize;
	129	bufferlist prepared = raw;
	130
	131	for (unsigned int i = 0; i < k - padded_chunks; i++) {
	132	bufferlist &chunk = encoded[chunk_index(i)];
	133	chunk.substr_of(prepared, i * blocksize, blocksize);
	134	chunk.rebuild_aligned_size_and_memory(blocksize, SIMD_ALIGN);
	135	assert(chunk.is_contiguous());
	136	}
	137	if (padded_chunks) {
	138	unsigned remainder = raw.length() - (k - padded_chunks) * blocksize;
	139	bufferptr buf(buffer::create_aligned(blocksize, SIMD_ALIGN));
140
141	raw.copy((k - padded_chunks) * blocksize, remainder, buf.c_str());
142	buf.zero(remainder, blocksize - remainder);
143	encoded[chunk_index(k-padded_chunks)].push_back(std::move(buf));
144
145	for (unsigned int i = k - padded_chunks + 1; i < k; i++) {
146	bufferptr buf(buffer::create_aligned(blocksize, SIMD_ALIGN));
147	buf.zero();
148	encoded[chunk_index(i)].push_back(std::move(buf));
149	}
150	}
151	for (unsigned int i = k; i < k + m; i++) {
152	bufferlist &chunk = encoded[chunk_index(i)];
153	chunk.push_back(buffer::create_aligned(blocksize, SIMD_ALIGN));
154	}
155
156	return 0;
157	}
158
159	int ErasureCode::encode(const set<int> &want_to_encode,
160	const bufferlist &in,
161	map<int, bufferlist> *encoded)
162	{
163	unsigned int k = get_data_chunk_count();
164	unsigned int m = get_chunk_count() - k;
165	bufferlist out;
166	int err = encode_prepare(in, *encoded);
167	if (err)
168	return err;
169	encode_chunks(want_to_encode, encoded);
170	for (unsigned int i = 0; i < k + m; i++) {
171	if (want_to_encode.count(i) == 0)
172	encoded->erase(i);
173	}
174	return 0;
175	}
176
177	int ErasureCode::encode_chunks(const set<int> &want_to_encode,
178	map<int, bufferlist> *encoded)
179	{
180	assert("ErasureCode::encode_chunks not implemented" == 0);
181	}
182
183	int ErasureCode::decode(const set<int> &want_to_read,
184	const map<int, bufferlist> &chunks,
185	map<int, bufferlist> *decoded)
186	{
187	vector<int> have;
188	have.reserve(chunks.size());
189	for (map<int, bufferlist>::const_iterator i = chunks.begin();
190	i != chunks.end();
191	++i) {
192	have.push_back(i->first);
193	}
194	if (includes(
195	have.begin(), have.end(), want_to_read.begin(), want_to_read.end())) {
196	for (set<int>::iterator i = want_to_read.begin();
197	i != want_to_read.end();
198	++i) {
199	(decoded)[i] = chunks.find(*i)->second;
200	}
201	return 0;
202	}
203	unsigned int k = get_data_chunk_count();
204	unsigned int m = get_chunk_count() - k;
205	unsigned blocksize = (*chunks.begin()).second.length();
206	for (unsigned int i = 0; i < k + m; i++) {
207	if (chunks.find(i) == chunks.end()) {
208	bufferptr ptr(buffer::create_aligned(blocksize, SIMD_ALIGN));
209	(*decoded)[i].push_front(ptr);
210	} else {
211	(*decoded)[i] = chunks.find(i)->second;
212	(*decoded)[i].rebuild_aligned(SIMD_ALIGN);
213	}
214	}
215	return decode_chunks(want_to_read, chunks, decoded);
216	}
217
218	int ErasureCode::decode_chunks(const set<int> &want_to_read,
219	const map<int, bufferlist> &chunks,
220	map<int, bufferlist> *decoded)
221	{
222	assert("ErasureCode::decode_chunks not implemented" == 0);
223	}
224
225	int ErasureCode::parse(const ErasureCodeProfile &profile,
226	ostream *ss)
227	{
228	return to_mapping(profile, ss);
229	}
230
231	const vector<int> &ErasureCode::get_chunk_mapping() const {
232	return chunk_mapping;
233	}
234
235	int ErasureCode::to_mapping(const ErasureCodeProfile &profile,
236	ostream *ss)
237	{
238	if (profile.find("mapping") != profile.end()) {
239	std::string mapping = profile.find("mapping")->second;
240	int position = 0;
241	vector<int> coding_chunk_mapping;
242	for(std::string::iterator it = mapping.begin(); it != mapping.end(); ++it) {
243	if (*it == 'D')
244	chunk_mapping.push_back(position);
245	else
246	coding_chunk_mapping.push_back(position);
247	position++;
248	}
249	chunk_mapping.insert(chunk_mapping.end(),
250	coding_chunk_mapping.begin(),
251	coding_chunk_mapping.end());
252	}
253	return 0;
254	}
255
256	int ErasureCode::to_int(const std::string &name,
257	ErasureCodeProfile &profile,
258	int *value,
259	const std::string &default_value,
260	ostream *ss)
261	{
262	if (profile.find(name) == profile.end() \|\|
263	profile.find(name)->second.size() == 0)
264	profile[name] = default_value;
265	std::string p = profile.find(name)->second;
266	std::string err;
267	int r = strict_strtol(p.c_str(), 10, &err);
268	if (!err.empty()) {
269	*ss << "could not convert " << name << "=" << p
270	<< " to int because " << err
271	<< ", set to default " << default_value << std::endl;
272	*value = strict_strtol(default_value.c_str(), 10, &err);
273	return -EINVAL;
274	}
275	*value = r;
276	return 0;
277	}
278
279	int ErasureCode::to_bool(const std::string &name,
280	ErasureCodeProfile &profile,
281	bool *value,
282	const std::string &default_value,
283	ostream *ss)
284	{
285	if (profile.find(name) == profile.end() \|\|
286	profile.find(name)->second.size() == 0)
287	profile[name] = default_value;
288	const std::string p = profile.find(name)->second;
289	*value = (p == "yes") \|\| (p == "true");
290	return 0;
291	}
292
293	int ErasureCode::to_string(const std::string &name,
294	ErasureCodeProfile &profile,
295	std::string *value,
296	const std::string &default_value,
297	ostream *ss)
298	{
299	if (profile.find(name) == profile.end() \|\|
300	profile.find(name)->second.size() == 0)
301	profile[name] = default_value;
302	*value = profile[name];
303	return 0;
304	}
305
306	int ErasureCode::decode_concat(const map<int, bufferlist> &chunks,
307	bufferlist *decoded)
308	{
309	set<int> want_to_read;
310
311	for (unsigned int i = 0; i < get_data_chunk_count(); i++) {
312	want_to_read.insert(chunk_index(i));
313	}
314	map<int, bufferlist> decoded_map;
315	int r = decode(want_to_read, chunks, &decoded_map);
316	if (r == 0) {
317	for (unsigned int i = 0; i < get_data_chunk_count(); i++) {
318	decoded->claim_append(decoded_map[chunk_index(i)]);
319	}
320	}
321	return r;
322	}