]>
Commit | Line | Data |
---|---|---|
11fdf7f2 | 1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
7c673cae FG |
2 | // vim: ts=8 sw=2 smarttab |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2014 Red Hat | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
1e59de90 | 15 | #include <bit> |
7c673cae FG |
16 | #include <unistd.h> |
17 | #include <stdlib.h> | |
18 | #include <sys/types.h> | |
19 | #include <sys/stat.h> | |
20 | #include <fcntl.h> | |
20effc67 | 21 | #include <algorithm> |
7c673cae | 22 | |
eafe8130 | 23 | #include <boost/container/flat_set.hpp> |
20effc67 | 24 | #include <boost/algorithm/string.hpp> |
39ae355f TL |
25 | #include <boost/random/mersenne_twister.hpp> |
26 | #include <boost/random/uniform_real.hpp> | |
eafe8130 | 27 | |
31f18b77 FG |
28 | #include "include/cpp-btree/btree_set.h" |
29 | ||
7c673cae | 30 | #include "BlueStore.h" |
f67539c2 | 31 | #include "bluestore_common.h" |
20effc67 | 32 | #include "simple_bitmap.h" |
7c673cae FG |
33 | #include "os/kv.h" |
34 | #include "include/compat.h" | |
35 | #include "include/intarith.h" | |
36 | #include "include/stringify.h" | |
11fdf7f2 TL |
37 | #include "include/str_map.h" |
38 | #include "include/util.h" | |
7c673cae FG |
39 | #include "common/errno.h" |
40 | #include "common/safe_io.h" | |
91327a77 | 41 | #include "common/PriorityCache.h" |
20effc67 | 42 | #include "common/url_escape.h" |
7c673cae FG |
43 | #include "Allocator.h" |
44 | #include "FreelistManager.h" | |
45 | #include "BlueFS.h" | |
46 | #include "BlueRocksEnv.h" | |
47 | #include "auth/Crypto.h" | |
48 | #include "common/EventTrace.h" | |
91327a77 | 49 | #include "perfglue/heap_profiler.h" |
11fdf7f2 TL |
50 | #include "common/blkdev.h" |
51 | #include "common/numa.h" | |
f67539c2 | 52 | #include "common/pretty_binary.h" |
20effc67 TL |
53 | #include "kv/KeyValueHistogram.h" |
54 | ||
55 | #ifdef HAVE_LIBZBD | |
56 | #include "ZonedAllocator.h" | |
57 | #include "ZonedFreelistManager.h" | |
58 | #endif | |
7c673cae | 59 | |
9f95a23c TL |
60 | #if defined(WITH_LTTNG) |
61 | #define TRACEPOINT_DEFINE | |
62 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
63 | #include "tracing/bluestore.h" | |
64 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
65 | #undef TRACEPOINT_DEFINE | |
66 | #else | |
67 | #define tracepoint(...) | |
68 | #endif | |
69 | ||
7c673cae FG |
70 | #define dout_context cct |
71 | #define dout_subsys ceph_subsys_bluestore | |
72 | ||
31f18b77 FG |
73 | using bid_t = decltype(BlueStore::Blob::id); |
74 | ||
75 | // bluestore_cache_onode | |
7c673cae | 76 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Onode, bluestore_onode, |
31f18b77 | 77 | bluestore_cache_onode); |
7c673cae | 78 | |
7c673cae | 79 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Buffer, bluestore_buffer, |
1e59de90 | 80 | bluestore_cache_buffer); |
7c673cae | 81 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Extent, bluestore_extent, |
1e59de90 | 82 | bluestore_extent); |
7c673cae | 83 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Blob, bluestore_blob, |
1e59de90 | 84 | bluestore_blob); |
7c673cae | 85 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::SharedBlob, bluestore_shared_blob, |
1e59de90 | 86 | bluestore_shared_blob); |
31f18b77 FG |
87 | |
88 | // bluestore_txc | |
89 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::TransContext, bluestore_transcontext, | |
90 | bluestore_txc); | |
20effc67 | 91 | using std::byte; |
f67539c2 TL |
92 | using std::deque; |
93 | using std::min; | |
94 | using std::make_pair; | |
95 | using std::numeric_limits; | |
96 | using std::pair; | |
20effc67 | 97 | using std::less; |
f67539c2 | 98 | using std::list; |
20effc67 | 99 | using std::make_unique; |
f67539c2 TL |
100 | using std::map; |
101 | using std::max; | |
102 | using std::ostream; | |
103 | using std::ostringstream; | |
104 | using std::set; | |
105 | using std::string; | |
106 | using std::stringstream; | |
20effc67 | 107 | using std::unique_ptr; |
f67539c2 TL |
108 | using std::vector; |
109 | ||
110 | using ceph::bufferlist; | |
111 | using ceph::bufferptr; | |
112 | using ceph::coarse_mono_clock; | |
113 | using ceph::decode; | |
114 | using ceph::encode; | |
115 | using ceph::Formatter; | |
116 | using ceph::JSONFormatter; | |
117 | using ceph::make_timespan; | |
118 | using ceph::mono_clock; | |
119 | using ceph::mono_time; | |
120 | using ceph::timespan_str; | |
7c673cae FG |
121 | |
122 | // kv store prefixes | |
11fdf7f2 TL |
123 | const string PREFIX_SUPER = "S"; // field -> value |
124 | const string PREFIX_STAT = "T"; // field -> value(int64 array) | |
125 | const string PREFIX_COLL = "C"; // collection name -> cnode_t | |
126 | const string PREFIX_OBJ = "O"; // object name -> onode_t | |
127 | const string PREFIX_OMAP = "M"; // u64 + keyname -> value | |
128 | const string PREFIX_PGMETA_OMAP = "P"; // u64 + keyname -> value(for meta coll) | |
9f95a23c | 129 | const string PREFIX_PERPOOL_OMAP = "m"; // s64 + u64 + keyname -> value |
f67539c2 | 130 | const string PREFIX_PERPG_OMAP = "p"; // u64(pool) + u32(hash) + u64(id) + keyname -> value |
11fdf7f2 TL |
131 | const string PREFIX_DEFERRED = "L"; // id -> deferred_transaction_t |
132 | const string PREFIX_ALLOC = "B"; // u64 offset -> u64 length (freelist) | |
133 | const string PREFIX_ALLOC_BITMAP = "b";// (see BitmapFreelistManager) | |
20effc67 TL |
134 | const string PREFIX_SHARED_BLOB = "X"; // u64 SB id -> shared_blob_t |
135 | ||
136 | #ifdef HAVE_LIBZBD | |
f67539c2 TL |
137 | const string PREFIX_ZONED_FM_META = "Z"; // (see ZonedFreelistManager) |
138 | const string PREFIX_ZONED_FM_INFO = "z"; // (see ZonedFreelistManager) | |
139 | const string PREFIX_ZONED_CL_INFO = "G"; // (per-zone cleaner metadata) | |
20effc67 | 140 | #endif |
7c673cae | 141 | |
11fdf7f2 TL |
142 | const string BLUESTORE_GLOBAL_STATFS_KEY = "bluestore_statfs"; |
143 | ||
7c673cae FG |
144 | // write a label in the first block. always use this size. note that |
145 | // bluefs makes a matching assumption about the location of its | |
146 | // superblock (always the second block of the device). | |
147 | #define BDEV_LABEL_BLOCK_SIZE 4096 | |
148 | ||
149 | // reserve: label (4k) + bluefs super (4k), which means we start at 8k. | |
150 | #define SUPER_RESERVED 8192 | |
151 | ||
152 | #define OBJECT_MAX_SIZE 0xffffffff // 32 bits | |
153 | ||
154 | ||
155 | /* | |
156 | * extent map blob encoding | |
157 | * | |
158 | * we use the low bits of the blobid field to indicate some common scenarios | |
159 | * and spanning vs local ids. See ExtentMap::{encode,decode}_some(). | |
160 | */ | |
161 | #define BLOBID_FLAG_CONTIGUOUS 0x1 // this extent starts at end of previous | |
162 | #define BLOBID_FLAG_ZEROOFFSET 0x2 // blob_offset is 0 | |
163 | #define BLOBID_FLAG_SAMELENGTH 0x4 // length matches previous extent | |
164 | #define BLOBID_FLAG_SPANNING 0x8 // has spanning blob id | |
165 | #define BLOBID_SHIFT_BITS 4 | |
166 | ||
167 | /* | |
168 | * object name key structure | |
169 | * | |
170 | * encoded u8: shard + 2^7 (so that it sorts properly) | |
171 | * encoded u64: poolid + 2^63 (so that it sorts properly) | |
172 | * encoded u32: hash (bit reversed) | |
173 | * | |
174 | * escaped string: namespace | |
175 | * | |
176 | * escaped string: key or object name | |
177 | * 1 char: '<', '=', or '>'. if =, then object key == object name, and | |
178 | * we are done. otherwise, we are followed by the object name. | |
179 | * escaped string: object name (unless '=' above) | |
180 | * | |
181 | * encoded u64: snap | |
182 | * encoded u64: generation | |
183 | * 'o' | |
184 | */ | |
185 | #define ONODE_KEY_SUFFIX 'o' | |
186 | ||
187 | /* | |
188 | * extent shard key | |
189 | * | |
190 | * object prefix key | |
191 | * u32 | |
192 | * 'x' | |
193 | */ | |
194 | #define EXTENT_SHARD_KEY_SUFFIX 'x' | |
195 | ||
196 | /* | |
197 | * string encoding in the key | |
198 | * | |
199 | * The key string needs to lexicographically sort the same way that | |
200 | * ghobject_t does. We do this by escaping anything <= to '#' with # | |
201 | * plus a 2 digit hex string, and anything >= '~' with ~ plus the two | |
202 | * hex digits. | |
203 | * | |
204 | * We use ! as a terminator for strings; this works because it is < # | |
205 | * and will get escaped if it is present in the string. | |
206 | * | |
f91f0fd5 TL |
207 | * NOTE: There is a bug in this implementation: due to implicit |
208 | * character type conversion in comparison it may produce unexpected | |
209 | * ordering. Unfortunately fixing the bug would mean invalidating the | |
210 | * keys in existing deployments. Instead we do additional sorting | |
211 | * where it is needed. | |
7c673cae FG |
212 | */ |
213 | template<typename S> | |
214 | static void append_escaped(const string &in, S *out) | |
215 | { | |
224ce89b WB |
216 | char hexbyte[in.length() * 3 + 1]; |
217 | char* ptr = &hexbyte[0]; | |
7c673cae | 218 | for (string::const_iterator i = in.begin(); i != in.end(); ++i) { |
f91f0fd5 | 219 | if (*i <= '#') { // bug: unexpected result for *i > 0x7f |
224ce89b WB |
220 | *ptr++ = '#'; |
221 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
222 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
f91f0fd5 | 223 | } else if (*i >= '~') { // bug: unexpected result for *i > 0x7f |
224ce89b WB |
224 | *ptr++ = '~'; |
225 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
226 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 227 | } else { |
224ce89b | 228 | *ptr++ = *i; |
7c673cae FG |
229 | } |
230 | } | |
224ce89b WB |
231 | *ptr++ = '!'; |
232 | out->append(hexbyte, ptr - &hexbyte[0]); | |
233 | } | |
234 | ||
235 | inline unsigned h2i(char c) | |
236 | { | |
237 | if ((c >= '0') && (c <= '9')) { | |
238 | return c - 0x30; | |
239 | } else if ((c >= 'a') && (c <= 'f')) { | |
240 | return c - 'a' + 10; | |
241 | } else if ((c >= 'A') && (c <= 'F')) { | |
242 | return c - 'A' + 10; | |
243 | } else { | |
244 | return 256; // make it always larger than 255 | |
245 | } | |
7c673cae FG |
246 | } |
247 | ||
248 | static int decode_escaped(const char *p, string *out) | |
249 | { | |
224ce89b WB |
250 | char buff[256]; |
251 | char* ptr = &buff[0]; | |
252 | char* max = &buff[252]; | |
7c673cae FG |
253 | const char *orig_p = p; |
254 | while (*p && *p != '!') { | |
255 | if (*p == '#' || *p == '~') { | |
224ce89b WB |
256 | unsigned hex = 0; |
257 | p++; | |
258 | hex = h2i(*p++) << 4; | |
259 | if (hex > 255) { | |
260 | return -EINVAL; | |
261 | } | |
262 | hex |= h2i(*p++); | |
263 | if (hex > 255) { | |
264 | return -EINVAL; | |
265 | } | |
266 | *ptr++ = hex; | |
7c673cae | 267 | } else { |
224ce89b WB |
268 | *ptr++ = *p++; |
269 | } | |
270 | if (ptr > max) { | |
271 | out->append(buff, ptr-buff); | |
272 | ptr = &buff[0]; | |
7c673cae FG |
273 | } |
274 | } | |
224ce89b WB |
275 | if (ptr != buff) { |
276 | out->append(buff, ptr-buff); | |
277 | } | |
7c673cae FG |
278 | return p - orig_p; |
279 | } | |
280 | ||
7c673cae FG |
281 | template<typename T> |
282 | static void _key_encode_shard(shard_id_t shard, T *key) | |
283 | { | |
284 | key->push_back((char)((uint8_t)shard.id + (uint8_t)0x80)); | |
285 | } | |
286 | ||
287 | static const char *_key_decode_shard(const char *key, shard_id_t *pshard) | |
288 | { | |
289 | pshard->id = (uint8_t)*key - (uint8_t)0x80; | |
290 | return key + 1; | |
291 | } | |
292 | ||
f91f0fd5 | 293 | static void get_coll_range(const coll_t& cid, int bits, |
f67539c2 | 294 | ghobject_t *temp_start, ghobject_t *temp_end, |
a4b75251 | 295 | ghobject_t *start, ghobject_t *end, bool legacy) |
7c673cae | 296 | { |
7c673cae | 297 | spg_t pgid; |
a4b75251 TL |
298 | constexpr uint32_t MAX_HASH = std::numeric_limits<uint32_t>::max(); |
299 | // use different nspaces due to we use different schemes when encoding | |
300 | // keys for listing objects | |
301 | const std::string_view MAX_NSPACE = legacy ? "\x7f" : "\xff"; | |
7c673cae | 302 | if (cid.is_pg(&pgid)) { |
f91f0fd5 | 303 | start->shard_id = pgid.shard; |
7c673cae FG |
304 | *temp_start = *start; |
305 | ||
f91f0fd5 TL |
306 | start->hobj.pool = pgid.pool(); |
307 | temp_start->hobj.pool = -2ll - pgid.pool(); | |
7c673cae FG |
308 | |
309 | *end = *start; | |
310 | *temp_end = *temp_start; | |
311 | ||
312 | uint32_t reverse_hash = hobject_t::_reverse_bits(pgid.ps()); | |
f91f0fd5 TL |
313 | start->hobj.set_bitwise_key_u32(reverse_hash); |
314 | temp_start->hobj.set_bitwise_key_u32(reverse_hash); | |
7c673cae FG |
315 | |
316 | uint64_t end_hash = reverse_hash + (1ull << (32 - bits)); | |
a4b75251 TL |
317 | if (end_hash > MAX_HASH) { |
318 | // make sure end hobj is even greater than the maximum possible hobj | |
319 | end->hobj.set_bitwise_key_u32(MAX_HASH); | |
320 | temp_end->hobj.set_bitwise_key_u32(MAX_HASH); | |
321 | end->hobj.nspace = MAX_NSPACE; | |
322 | } else { | |
323 | end->hobj.set_bitwise_key_u32(end_hash); | |
324 | temp_end->hobj.set_bitwise_key_u32(end_hash); | |
325 | } | |
7c673cae | 326 | } else { |
f91f0fd5 TL |
327 | start->shard_id = shard_id_t::NO_SHARD; |
328 | start->hobj.pool = -1ull; | |
329 | ||
7c673cae | 330 | *end = *start; |
f91f0fd5 | 331 | start->hobj.set_bitwise_key_u32(0); |
a4b75251 TL |
332 | end->hobj.set_bitwise_key_u32(MAX_HASH); |
333 | end->hobj.nspace = MAX_NSPACE; | |
7c673cae FG |
334 | // no separate temp section |
335 | *temp_start = *end; | |
336 | *temp_end = *end; | |
337 | } | |
f91f0fd5 TL |
338 | |
339 | start->generation = 0; | |
340 | end->generation = 0; | |
341 | temp_start->generation = 0; | |
342 | temp_end->generation = 0; | |
7c673cae FG |
343 | } |
344 | ||
345 | static void get_shared_blob_key(uint64_t sbid, string *key) | |
346 | { | |
347 | key->clear(); | |
348 | _key_encode_u64(sbid, key); | |
349 | } | |
350 | ||
351 | static int get_key_shared_blob(const string& key, uint64_t *sbid) | |
352 | { | |
353 | const char *p = key.c_str(); | |
354 | if (key.length() < sizeof(uint64_t)) | |
355 | return -1; | |
224ce89b | 356 | _key_decode_u64(p, sbid); |
7c673cae FG |
357 | return 0; |
358 | } | |
359 | ||
360 | template<typename S> | |
f91f0fd5 | 361 | static void _key_encode_prefix(const ghobject_t& oid, S *key) |
7c673cae | 362 | { |
f91f0fd5 TL |
363 | _key_encode_shard(oid.shard_id, key); |
364 | _key_encode_u64(oid.hobj.pool + 0x8000000000000000ull, key); | |
365 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), key); | |
366 | } | |
7c673cae | 367 | |
f91f0fd5 TL |
368 | static const char *_key_decode_prefix(const char *p, ghobject_t *oid) |
369 | { | |
7c673cae FG |
370 | p = _key_decode_shard(p, &oid->shard_id); |
371 | ||
372 | uint64_t pool; | |
373 | p = _key_decode_u64(p, &pool); | |
374 | oid->hobj.pool = pool - 0x8000000000000000ull; | |
375 | ||
376 | unsigned hash; | |
377 | p = _key_decode_u32(p, &hash); | |
378 | ||
379 | oid->hobj.set_bitwise_key_u32(hash); | |
380 | ||
f91f0fd5 TL |
381 | return p; |
382 | } | |
383 | ||
39ae355f | 384 | |
f91f0fd5 TL |
385 | #define ENCODED_KEY_PREFIX_LEN (1 + 8 + 4) |
386 | ||
20effc67 | 387 | static int _get_key_object(const char *p, ghobject_t *oid) |
f91f0fd5 TL |
388 | { |
389 | int r; | |
f91f0fd5 TL |
390 | |
391 | p = _key_decode_prefix(p, oid); | |
392 | ||
7c673cae FG |
393 | r = decode_escaped(p, &oid->hobj.nspace); |
394 | if (r < 0) | |
395 | return -2; | |
396 | p += r + 1; | |
397 | ||
398 | string k; | |
399 | r = decode_escaped(p, &k); | |
400 | if (r < 0) | |
401 | return -3; | |
402 | p += r + 1; | |
403 | if (*p == '=') { | |
404 | // no key | |
405 | ++p; | |
406 | oid->hobj.oid.name = k; | |
407 | } else if (*p == '<' || *p == '>') { | |
408 | // key + name | |
409 | ++p; | |
410 | r = decode_escaped(p, &oid->hobj.oid.name); | |
411 | if (r < 0) | |
412 | return -5; | |
413 | p += r + 1; | |
414 | oid->hobj.set_key(k); | |
415 | } else { | |
416 | // malformed | |
417 | return -6; | |
418 | } | |
419 | ||
420 | p = _key_decode_u64(p, &oid->hobj.snap.val); | |
421 | p = _key_decode_u64(p, &oid->generation); | |
422 | ||
423 | if (*p != ONODE_KEY_SUFFIX) { | |
424 | return -7; | |
425 | } | |
426 | p++; | |
427 | if (*p) { | |
428 | // if we get something other than a null terminator here, | |
429 | // something goes wrong. | |
430 | return -8; | |
431 | } | |
432 | ||
433 | return 0; | |
434 | } | |
435 | ||
436 | template<typename S> | |
20effc67 | 437 | static int get_key_object(const S& key, ghobject_t *oid) |
7c673cae | 438 | { |
20effc67 TL |
439 | if (key.length() < ENCODED_KEY_PREFIX_LEN) |
440 | return -1; | |
441 | if (key.length() == ENCODED_KEY_PREFIX_LEN) | |
442 | return -2; | |
443 | const char *p = key.c_str(); | |
444 | return _get_key_object(p, oid); | |
445 | } | |
7c673cae | 446 | |
20effc67 TL |
447 | template<typename S> |
448 | static void _get_object_key(const ghobject_t& oid, S *key) | |
449 | { | |
f91f0fd5 | 450 | size_t max_len = ENCODED_KEY_PREFIX_LEN + |
7c673cae FG |
451 | (oid.hobj.nspace.length() * 3 + 1) + |
452 | (oid.hobj.get_key().length() * 3 + 1) + | |
453 | 1 + // for '<', '=', or '>' | |
454 | (oid.hobj.oid.name.length() * 3 + 1) + | |
455 | 8 + 8 + 1; | |
456 | key->reserve(max_len); | |
457 | ||
f91f0fd5 | 458 | _key_encode_prefix(oid, key); |
7c673cae FG |
459 | |
460 | append_escaped(oid.hobj.nspace, key); | |
461 | ||
462 | if (oid.hobj.get_key().length()) { | |
463 | // is a key... could be < = or >. | |
464 | append_escaped(oid.hobj.get_key(), key); | |
465 | // (ASCII chars < = and > sort in that order, yay) | |
466 | int r = oid.hobj.get_key().compare(oid.hobj.oid.name); | |
467 | if (r) { | |
468 | key->append(r > 0 ? ">" : "<"); | |
469 | append_escaped(oid.hobj.oid.name, key); | |
470 | } else { | |
471 | // same as no key | |
472 | key->append("="); | |
473 | } | |
474 | } else { | |
475 | // no key | |
476 | append_escaped(oid.hobj.oid.name, key); | |
477 | key->append("="); | |
478 | } | |
479 | ||
480 | _key_encode_u64(oid.hobj.snap, key); | |
481 | _key_encode_u64(oid.generation, key); | |
482 | ||
483 | key->push_back(ONODE_KEY_SUFFIX); | |
20effc67 TL |
484 | } |
485 | ||
486 | template<typename S> | |
487 | static void get_object_key(CephContext *cct, const ghobject_t& oid, S *key) | |
488 | { | |
489 | key->clear(); | |
490 | _get_object_key(oid, key); | |
7c673cae FG |
491 | |
492 | // sanity check | |
493 | if (true) { | |
494 | ghobject_t t; | |
495 | int r = get_key_object(*key, &t); | |
496 | if (r || t != oid) { | |
497 | derr << " r " << r << dendl; | |
498 | derr << "key " << pretty_binary_string(*key) << dendl; | |
499 | derr << "oid " << oid << dendl; | |
500 | derr << " t " << t << dendl; | |
11fdf7f2 | 501 | ceph_assert(r == 0 && t == oid); |
7c673cae FG |
502 | } |
503 | } | |
504 | } | |
505 | ||
7c673cae FG |
506 | // extent shard keys are the onode key, plus a u32, plus 'x'. the trailing |
507 | // char lets us quickly test whether it is a shard key without decoding any | |
508 | // of the prefix bytes. | |
509 | template<typename S> | |
510 | static void get_extent_shard_key(const S& onode_key, uint32_t offset, | |
511 | string *key) | |
512 | { | |
513 | key->clear(); | |
514 | key->reserve(onode_key.length() + 4 + 1); | |
515 | key->append(onode_key.c_str(), onode_key.size()); | |
516 | _key_encode_u32(offset, key); | |
517 | key->push_back(EXTENT_SHARD_KEY_SUFFIX); | |
518 | } | |
519 | ||
520 | static void rewrite_extent_shard_key(uint32_t offset, string *key) | |
521 | { | |
11fdf7f2 TL |
522 | ceph_assert(key->size() > sizeof(uint32_t) + 1); |
523 | ceph_assert(*key->rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
524 | _key_encode_u32(offset, key->size() - sizeof(uint32_t) - 1, key); |
525 | } | |
526 | ||
527 | template<typename S> | |
528 | static void generate_extent_shard_key_and_apply( | |
529 | const S& onode_key, | |
530 | uint32_t offset, | |
531 | string *key, | |
532 | std::function<void(const string& final_key)> apply) | |
533 | { | |
534 | if (key->empty()) { // make full key | |
11fdf7f2 | 535 | ceph_assert(!onode_key.empty()); |
7c673cae FG |
536 | get_extent_shard_key(onode_key, offset, key); |
537 | } else { | |
538 | rewrite_extent_shard_key(offset, key); | |
539 | } | |
540 | apply(*key); | |
541 | } | |
542 | ||
543 | int get_key_extent_shard(const string& key, string *onode_key, uint32_t *offset) | |
544 | { | |
11fdf7f2 TL |
545 | ceph_assert(key.size() > sizeof(uint32_t) + 1); |
546 | ceph_assert(*key.rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
547 | int okey_len = key.size() - sizeof(uint32_t) - 1; |
548 | *onode_key = key.substr(0, okey_len); | |
549 | const char *p = key.data() + okey_len; | |
224ce89b | 550 | _key_decode_u32(p, offset); |
7c673cae FG |
551 | return 0; |
552 | } | |
553 | ||
554 | static bool is_extent_shard_key(const string& key) | |
555 | { | |
556 | return *key.rbegin() == EXTENT_SHARD_KEY_SUFFIX; | |
557 | } | |
558 | ||
7c673cae FG |
559 | static void get_deferred_key(uint64_t seq, string *out) |
560 | { | |
561 | _key_encode_u64(seq, out); | |
562 | } | |
563 | ||
11fdf7f2 TL |
564 | static void get_pool_stat_key(int64_t pool_id, string *key) |
565 | { | |
566 | key->clear(); | |
567 | _key_encode_u64(pool_id, key); | |
568 | } | |
569 | ||
570 | static int get_key_pool_stat(const string& key, uint64_t* pool_id) | |
571 | { | |
572 | const char *p = key.c_str(); | |
573 | if (key.length() < sizeof(uint64_t)) | |
574 | return -1; | |
575 | _key_decode_u64(p, pool_id); | |
576 | return 0; | |
577 | } | |
7c673cae | 578 | |
20effc67 TL |
579 | #ifdef HAVE_LIBZBD |
580 | static void get_zone_offset_object_key( | |
581 | uint32_t zone, | |
582 | uint64_t offset, | |
583 | ghobject_t oid, | |
584 | std::string *key) | |
585 | { | |
586 | key->clear(); | |
587 | _key_encode_u32(zone, key); | |
588 | _key_encode_u64(offset, key); | |
589 | _get_object_key(oid, key); | |
590 | } | |
591 | ||
592 | static int get_key_zone_offset_object( | |
593 | const string& key, | |
594 | uint32_t *zone, | |
595 | uint64_t *offset, | |
596 | ghobject_t *oid) | |
597 | { | |
598 | const char *p = key.c_str(); | |
599 | if (key.length() < sizeof(uint64_t) + sizeof(uint32_t) + ENCODED_KEY_PREFIX_LEN + 1) | |
600 | return -1; | |
601 | p = _key_decode_u32(p, zone); | |
602 | p = _key_decode_u64(p, offset); | |
603 | int r = _get_key_object(p, oid); | |
604 | if (r < 0) { | |
605 | return r; | |
606 | } | |
607 | return 0; | |
608 | } | |
609 | #endif | |
522d829b | 610 | |
81eedcae TL |
611 | template <int LogLevelV> |
612 | void _dump_extent_map(CephContext *cct, const BlueStore::ExtentMap &em) | |
613 | { | |
614 | uint64_t pos = 0; | |
615 | for (auto& s : em.shards) { | |
616 | dout(LogLevelV) << __func__ << " shard " << *s.shard_info | |
617 | << (s.loaded ? " (loaded)" : "") | |
618 | << (s.dirty ? " (dirty)" : "") | |
619 | << dendl; | |
620 | } | |
621 | for (auto& e : em.extent_map) { | |
622 | dout(LogLevelV) << __func__ << " " << e << dendl; | |
623 | ceph_assert(e.logical_offset >= pos); | |
624 | pos = e.logical_offset + e.length; | |
625 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
626 | if (blob.has_csum()) { | |
627 | vector<uint64_t> v; | |
628 | unsigned n = blob.get_csum_count(); | |
629 | for (unsigned i = 0; i < n; ++i) | |
630 | v.push_back(blob.get_csum_item(i)); | |
631 | dout(LogLevelV) << __func__ << " csum: " << std::hex << v << std::dec | |
632 | << dendl; | |
633 | } | |
634 | std::lock_guard l(e.blob->shared_blob->get_cache()->lock); | |
635 | for (auto& i : e.blob->shared_blob->bc.buffer_map) { | |
636 | dout(LogLevelV) << __func__ << " 0x" << std::hex << i.first | |
637 | << "~" << i.second->length << std::dec | |
638 | << " " << *i.second << dendl; | |
639 | } | |
640 | } | |
641 | } | |
642 | ||
643 | template <int LogLevelV> | |
644 | void _dump_onode(CephContext *cct, const BlueStore::Onode& o) | |
645 | { | |
646 | if (!cct->_conf->subsys.should_gather<ceph_subsys_bluestore, LogLevelV>()) | |
647 | return; | |
648 | dout(LogLevelV) << __func__ << " " << &o << " " << o.oid | |
649 | << " nid " << o.onode.nid | |
650 | << " size 0x" << std::hex << o.onode.size | |
651 | << " (" << std::dec << o.onode.size << ")" | |
652 | << " expected_object_size " << o.onode.expected_object_size | |
653 | << " expected_write_size " << o.onode.expected_write_size | |
654 | << " in " << o.onode.extent_map_shards.size() << " shards" | |
655 | << ", " << o.extent_map.spanning_blob_map.size() | |
656 | << " spanning blobs" | |
657 | << dendl; | |
20effc67 TL |
658 | for (auto& [zone, offset] : o.onode.zone_offset_refs) { |
659 | dout(LogLevelV) << __func__ << " zone ref 0x" << std::hex << zone | |
660 | << " offset 0x" << offset << std::dec << dendl; | |
661 | } | |
81eedcae TL |
662 | for (auto p = o.onode.attrs.begin(); |
663 | p != o.onode.attrs.end(); | |
664 | ++p) { | |
665 | dout(LogLevelV) << __func__ << " attr " << p->first | |
666 | << " len " << p->second.length() << dendl; | |
667 | } | |
668 | _dump_extent_map<LogLevelV>(cct, o.extent_map); | |
669 | } | |
670 | ||
671 | template <int LogLevelV> | |
672 | void _dump_transaction(CephContext *cct, ObjectStore::Transaction *t) | |
673 | { | |
674 | dout(LogLevelV) << __func__ << " transaction dump:\n"; | |
675 | JSONFormatter f(true); | |
676 | f.open_object_section("transaction"); | |
677 | t->dump(&f); | |
678 | f.close_section(); | |
679 | f.flush(*_dout); | |
680 | *_dout << dendl; | |
681 | } | |
682 | ||
7c673cae FG |
683 | // Buffer |
684 | ||
685 | ostream& operator<<(ostream& out, const BlueStore::Buffer& b) | |
686 | { | |
687 | out << "buffer(" << &b << " space " << b.space << " 0x" << std::hex | |
688 | << b.offset << "~" << b.length << std::dec | |
689 | << " " << BlueStore::Buffer::get_state_name(b.state); | |
690 | if (b.flags) | |
691 | out << " " << BlueStore::Buffer::get_flag_name(b.flags); | |
692 | return out << ")"; | |
693 | } | |
694 | ||
f91f0fd5 TL |
695 | namespace { |
696 | ||
697 | /* | |
698 | * Due to a bug in key string encoding (see a comment for append_escaped) | |
699 | * the KeyValueDB iterator does not lexicographically sort the same | |
700 | * way that ghobject_t does: objects with the same hash may have wrong order. | |
701 | * | |
702 | * This is the iterator wrapper that fixes the keys order. | |
703 | */ | |
704 | ||
705 | class CollectionListIterator { | |
706 | public: | |
707 | CollectionListIterator(const KeyValueDB::Iterator &it) | |
708 | : m_it(it) { | |
709 | } | |
710 | virtual ~CollectionListIterator() { | |
711 | } | |
712 | ||
713 | virtual bool valid() const = 0; | |
714 | virtual const ghobject_t &oid() const = 0; | |
715 | virtual void lower_bound(const ghobject_t &oid) = 0; | |
716 | virtual void upper_bound(const ghobject_t &oid) = 0; | |
717 | virtual void next() = 0; | |
718 | ||
adb31ebb TL |
719 | virtual int cmp(const ghobject_t &oid) const = 0; |
720 | ||
721 | bool is_ge(const ghobject_t &oid) const { | |
722 | return cmp(oid) >= 0; | |
723 | } | |
724 | ||
725 | bool is_lt(const ghobject_t &oid) const { | |
726 | return cmp(oid) < 0; | |
727 | } | |
728 | ||
f91f0fd5 TL |
729 | protected: |
730 | KeyValueDB::Iterator m_it; | |
731 | }; | |
732 | ||
733 | class SimpleCollectionListIterator : public CollectionListIterator { | |
734 | public: | |
735 | SimpleCollectionListIterator(CephContext *cct, const KeyValueDB::Iterator &it) | |
736 | : CollectionListIterator(it), m_cct(cct) { | |
737 | } | |
738 | ||
739 | bool valid() const override { | |
740 | return m_it->valid(); | |
741 | } | |
742 | ||
743 | const ghobject_t &oid() const override { | |
744 | ceph_assert(valid()); | |
745 | ||
746 | return m_oid; | |
747 | } | |
748 | ||
749 | void lower_bound(const ghobject_t &oid) override { | |
750 | string key; | |
751 | get_object_key(m_cct, oid, &key); | |
752 | ||
753 | m_it->lower_bound(key); | |
754 | get_oid(); | |
755 | } | |
756 | ||
757 | void upper_bound(const ghobject_t &oid) override { | |
758 | string key; | |
759 | get_object_key(m_cct, oid, &key); | |
760 | ||
761 | m_it->upper_bound(key); | |
762 | get_oid(); | |
763 | } | |
764 | ||
765 | void next() override { | |
766 | ceph_assert(valid()); | |
767 | ||
768 | m_it->next(); | |
769 | get_oid(); | |
770 | } | |
771 | ||
adb31ebb TL |
772 | int cmp(const ghobject_t &oid) const override { |
773 | ceph_assert(valid()); | |
774 | ||
775 | string key; | |
776 | get_object_key(m_cct, oid, &key); | |
777 | ||
778 | return m_it->key().compare(key); | |
779 | } | |
780 | ||
f91f0fd5 TL |
781 | private: |
782 | CephContext *m_cct; | |
783 | ghobject_t m_oid; | |
784 | ||
785 | void get_oid() { | |
f67539c2 TL |
786 | m_oid = ghobject_t(); |
787 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
788 | m_it->next(); | |
f91f0fd5 | 789 | } |
f67539c2 | 790 | if (!valid()) { |
f91f0fd5 TL |
791 | return; |
792 | } | |
793 | ||
f91f0fd5 TL |
794 | int r = get_key_object(m_it->key(), &m_oid); |
795 | ceph_assert(r == 0); | |
796 | } | |
797 | }; | |
798 | ||
799 | class SortedCollectionListIterator : public CollectionListIterator { | |
800 | public: | |
801 | SortedCollectionListIterator(const KeyValueDB::Iterator &it) | |
802 | : CollectionListIterator(it), m_chunk_iter(m_chunk.end()) { | |
803 | } | |
804 | ||
805 | bool valid() const override { | |
806 | return m_chunk_iter != m_chunk.end(); | |
807 | } | |
808 | ||
809 | const ghobject_t &oid() const override { | |
810 | ceph_assert(valid()); | |
811 | ||
812 | return m_chunk_iter->first; | |
813 | } | |
814 | ||
815 | void lower_bound(const ghobject_t &oid) override { | |
816 | std::string key; | |
817 | _key_encode_prefix(oid, &key); | |
818 | ||
819 | m_it->lower_bound(key); | |
820 | m_chunk_iter = m_chunk.end(); | |
821 | if (!get_next_chunk()) { | |
822 | return; | |
823 | } | |
824 | ||
825 | if (this->oid().shard_id != oid.shard_id || | |
826 | this->oid().hobj.pool != oid.hobj.pool || | |
827 | this->oid().hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
828 | return; | |
829 | } | |
830 | ||
831 | m_chunk_iter = m_chunk.lower_bound(oid); | |
832 | if (m_chunk_iter == m_chunk.end()) { | |
833 | get_next_chunk(); | |
834 | } | |
835 | } | |
836 | ||
837 | void upper_bound(const ghobject_t &oid) override { | |
838 | lower_bound(oid); | |
839 | ||
840 | if (valid() && this->oid() == oid) { | |
841 | next(); | |
842 | } | |
843 | } | |
844 | ||
845 | void next() override { | |
846 | ceph_assert(valid()); | |
847 | ||
848 | m_chunk_iter++; | |
849 | if (m_chunk_iter == m_chunk.end()) { | |
850 | get_next_chunk(); | |
851 | } | |
852 | } | |
853 | ||
adb31ebb TL |
854 | int cmp(const ghobject_t &oid) const override { |
855 | ceph_assert(valid()); | |
856 | ||
857 | if (this->oid() < oid) { | |
858 | return -1; | |
859 | } | |
860 | if (this->oid() > oid) { | |
861 | return 1; | |
862 | } | |
863 | return 0; | |
864 | } | |
865 | ||
f91f0fd5 TL |
866 | private: |
867 | std::map<ghobject_t, std::string> m_chunk; | |
868 | std::map<ghobject_t, std::string>::iterator m_chunk_iter; | |
869 | ||
870 | bool get_next_chunk() { | |
871 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
872 | m_it->next(); | |
873 | } | |
874 | ||
875 | if (!m_it->valid()) { | |
876 | return false; | |
877 | } | |
878 | ||
879 | ghobject_t oid; | |
880 | int r = get_key_object(m_it->key(), &oid); | |
881 | ceph_assert(r == 0); | |
882 | ||
883 | m_chunk.clear(); | |
884 | while (true) { | |
885 | m_chunk.insert({oid, m_it->key()}); | |
886 | ||
887 | do { | |
888 | m_it->next(); | |
889 | } while (m_it->valid() && is_extent_shard_key(m_it->key())); | |
890 | ||
891 | if (!m_it->valid()) { | |
892 | break; | |
893 | } | |
894 | ||
895 | ghobject_t next; | |
896 | r = get_key_object(m_it->key(), &next); | |
897 | ceph_assert(r == 0); | |
898 | if (next.shard_id != oid.shard_id || | |
899 | next.hobj.pool != oid.hobj.pool || | |
900 | next.hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
901 | break; | |
902 | } | |
903 | oid = next; | |
904 | } | |
905 | ||
906 | m_chunk_iter = m_chunk.begin(); | |
907 | return true; | |
908 | } | |
909 | }; | |
910 | ||
911 | } // anonymous namespace | |
912 | ||
7c673cae FG |
913 | // Garbage Collector |
914 | ||
915 | void BlueStore::GarbageCollector::process_protrusive_extents( | |
916 | const BlueStore::ExtentMap& extent_map, | |
917 | uint64_t start_offset, | |
918 | uint64_t end_offset, | |
919 | uint64_t start_touch_offset, | |
920 | uint64_t end_touch_offset, | |
921 | uint64_t min_alloc_size) | |
922 | { | |
11fdf7f2 | 923 | ceph_assert(start_offset <= start_touch_offset && end_offset>= end_touch_offset); |
7c673cae | 924 | |
11fdf7f2 TL |
925 | uint64_t lookup_start_offset = p2align(start_offset, min_alloc_size); |
926 | uint64_t lookup_end_offset = round_up_to(end_offset, min_alloc_size); | |
7c673cae FG |
927 | |
928 | dout(30) << __func__ << " (hex): [" << std::hex | |
929 | << lookup_start_offset << ", " << lookup_end_offset | |
930 | << ")" << std::dec << dendl; | |
931 | ||
932 | for (auto it = extent_map.seek_lextent(lookup_start_offset); | |
933 | it != extent_map.extent_map.end() && | |
934 | it->logical_offset < lookup_end_offset; | |
935 | ++it) { | |
936 | uint64_t alloc_unit_start = it->logical_offset / min_alloc_size; | |
937 | uint64_t alloc_unit_end = (it->logical_end() - 1) / min_alloc_size; | |
938 | ||
939 | dout(30) << __func__ << " " << *it | |
940 | << "alloc_units: " << alloc_unit_start << ".." << alloc_unit_end | |
941 | << dendl; | |
942 | ||
943 | Blob* b = it->blob.get(); | |
944 | ||
945 | if (it->logical_offset >=start_touch_offset && | |
946 | it->logical_end() <= end_touch_offset) { | |
947 | // Process extents within the range affected by | |
948 | // the current write request. | |
949 | // Need to take into account if existing extents | |
950 | // can be merged with them (uncompressed case) | |
951 | if (!b->get_blob().is_compressed()) { | |
952 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
953 | --blob_info_counted->expected_allocations; // don't need to allocate | |
954 | // new AU for compressed | |
955 | // data since another | |
956 | // collocated uncompressed | |
957 | // blob already exists | |
958 | dout(30) << __func__ << " --expected:" | |
959 | << alloc_unit_start << dendl; | |
960 | } | |
961 | used_alloc_unit = alloc_unit_end; | |
962 | blob_info_counted = nullptr; | |
963 | } | |
964 | } else if (b->get_blob().is_compressed()) { | |
965 | ||
966 | // additionally we take compressed blobs that were not impacted | |
967 | // by the write into account too | |
968 | BlobInfo& bi = | |
969 | affected_blobs.emplace( | |
970 | b, BlobInfo(b->get_referenced_bytes())).first->second; | |
971 | ||
972 | int adjust = | |
973 | (used_alloc_unit && used_alloc_unit == alloc_unit_start) ? 0 : 1; | |
974 | bi.expected_allocations += alloc_unit_end - alloc_unit_start + adjust; | |
975 | dout(30) << __func__ << " expected_allocations=" | |
976 | << bi.expected_allocations << " end_au:" | |
977 | << alloc_unit_end << dendl; | |
978 | ||
979 | blob_info_counted = &bi; | |
980 | used_alloc_unit = alloc_unit_end; | |
981 | ||
11fdf7f2 | 982 | ceph_assert(it->length <= bi.referenced_bytes); |
7c673cae FG |
983 | bi.referenced_bytes -= it->length; |
984 | dout(30) << __func__ << " affected_blob:" << *b | |
985 | << " unref 0x" << std::hex << it->length | |
986 | << " referenced = 0x" << bi.referenced_bytes | |
987 | << std::dec << dendl; | |
988 | // NOTE: we can't move specific blob to resulting GC list here | |
989 | // when reference counter == 0 since subsequent extents might | |
990 | // decrement its expected_allocation. | |
991 | // Hence need to enumerate all the extents first. | |
992 | if (!bi.collect_candidate) { | |
993 | bi.first_lextent = it; | |
994 | bi.collect_candidate = true; | |
995 | } | |
996 | bi.last_lextent = it; | |
997 | } else { | |
998 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
999 | // don't need to allocate new AU for compressed data since another | |
1000 | // collocated uncompressed blob already exists | |
1001 | --blob_info_counted->expected_allocations; | |
1002 | dout(30) << __func__ << " --expected_allocations:" | |
1003 | << alloc_unit_start << dendl; | |
1004 | } | |
1005 | used_alloc_unit = alloc_unit_end; | |
1006 | blob_info_counted = nullptr; | |
1007 | } | |
1008 | } | |
1009 | ||
1010 | for (auto b_it = affected_blobs.begin(); | |
1011 | b_it != affected_blobs.end(); | |
1012 | ++b_it) { | |
1013 | Blob* b = b_it->first; | |
1014 | BlobInfo& bi = b_it->second; | |
1015 | if (bi.referenced_bytes == 0) { | |
1016 | uint64_t len_on_disk = b_it->first->get_blob().get_ondisk_length(); | |
1017 | int64_t blob_expected_for_release = | |
11fdf7f2 | 1018 | round_up_to(len_on_disk, min_alloc_size) / min_alloc_size; |
7c673cae FG |
1019 | |
1020 | dout(30) << __func__ << " " << *(b_it->first) | |
1021 | << " expected4release=" << blob_expected_for_release | |
1022 | << " expected_allocations=" << bi.expected_allocations | |
1023 | << dendl; | |
1024 | int64_t benefit = blob_expected_for_release - bi.expected_allocations; | |
11fdf7f2 | 1025 | if (benefit >= g_conf()->bluestore_gc_enable_blob_threshold) { |
7c673cae FG |
1026 | if (bi.collect_candidate) { |
1027 | auto it = bi.first_lextent; | |
1028 | bool bExit = false; | |
1029 | do { | |
1030 | if (it->blob.get() == b) { | |
eafe8130 | 1031 | extents_to_collect.insert(it->logical_offset, it->length); |
7c673cae FG |
1032 | } |
1033 | bExit = it == bi.last_lextent; | |
1034 | ++it; | |
31f18b77 | 1035 | } while (!bExit); |
7c673cae FG |
1036 | } |
1037 | expected_for_release += blob_expected_for_release; | |
1038 | expected_allocations += bi.expected_allocations; | |
1039 | } | |
1040 | } | |
1041 | } | |
1042 | } | |
1043 | ||
1044 | int64_t BlueStore::GarbageCollector::estimate( | |
1045 | uint64_t start_offset, | |
1046 | uint64_t length, | |
1047 | const BlueStore::ExtentMap& extent_map, | |
1048 | const BlueStore::old_extent_map_t& old_extents, | |
1049 | uint64_t min_alloc_size) | |
1050 | { | |
1051 | ||
1052 | affected_blobs.clear(); | |
1053 | extents_to_collect.clear(); | |
1054 | used_alloc_unit = boost::optional<uint64_t >(); | |
1055 | blob_info_counted = nullptr; | |
1056 | ||
eafe8130 TL |
1057 | uint64_t gc_start_offset = start_offset; |
1058 | uint64_t gc_end_offset = start_offset + length; | |
7c673cae FG |
1059 | |
1060 | uint64_t end_offset = start_offset + length; | |
1061 | ||
1062 | for (auto it = old_extents.begin(); it != old_extents.end(); ++it) { | |
1063 | Blob* b = it->e.blob.get(); | |
1064 | if (b->get_blob().is_compressed()) { | |
1065 | ||
1066 | // update gc_start_offset/gc_end_offset if needed | |
1067 | gc_start_offset = min(gc_start_offset, (uint64_t)it->e.blob_start()); | |
11fdf7f2 | 1068 | gc_end_offset = std::max(gc_end_offset, (uint64_t)it->e.blob_end()); |
7c673cae FG |
1069 | |
1070 | auto o = it->e.logical_offset; | |
1071 | auto l = it->e.length; | |
1072 | ||
1073 | uint64_t ref_bytes = b->get_referenced_bytes(); | |
1074 | // micro optimization to bypass blobs that have no more references | |
1075 | if (ref_bytes != 0) { | |
1076 | dout(30) << __func__ << " affected_blob:" << *b | |
1077 | << " unref 0x" << std::hex << o << "~" << l | |
1078 | << std::dec << dendl; | |
1079 | affected_blobs.emplace(b, BlobInfo(ref_bytes)); | |
1080 | } | |
1081 | } | |
1082 | } | |
1083 | dout(30) << __func__ << " gc range(hex): [" << std::hex | |
1084 | << gc_start_offset << ", " << gc_end_offset | |
1085 | << ")" << std::dec << dendl; | |
1086 | ||
1087 | // enumerate preceeding extents to check if they reference affected blobs | |
1088 | if (gc_start_offset < start_offset || gc_end_offset > end_offset) { | |
1089 | process_protrusive_extents(extent_map, | |
1090 | gc_start_offset, | |
1091 | gc_end_offset, | |
1092 | start_offset, | |
1093 | end_offset, | |
1094 | min_alloc_size); | |
1095 | } | |
1096 | return expected_for_release - expected_allocations; | |
1097 | } | |
1098 | ||
9f95a23c TL |
1099 | // LruOnodeCacheShard |
1100 | struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard { | |
1101 | typedef boost::intrusive::list< | |
1102 | BlueStore::Onode, | |
1103 | boost::intrusive::member_hook< | |
1104 | BlueStore::Onode, | |
1105 | boost::intrusive::list_member_hook<>, | |
1106 | &BlueStore::Onode::lru_item> > list_t; | |
7c673cae | 1107 | |
9f95a23c | 1108 | list_t lru; |
7c673cae | 1109 | |
9f95a23c | 1110 | explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {} |
7c673cae | 1111 | |
f6b5b4d7 | 1112 | void _add(BlueStore::Onode* o, int level) override |
9f95a23c | 1113 | { |
39ae355f TL |
1114 | o->set_cached(); |
1115 | if (o->pin_nref == 1) { | |
9f95a23c | 1116 | (level > 0) ? lru.push_front(*o) : lru.push_back(*o); |
20effc67 TL |
1117 | o->cache_age_bin = age_bins.front(); |
1118 | *(o->cache_age_bin) += 1; | |
9f95a23c | 1119 | } |
f6b5b4d7 | 1120 | ++num; // we count both pinned and unpinned entries |
20effc67 TL |
1121 | dout(20) << __func__ << " " << this << " " << o->oid << " added, num=" |
1122 | << num << dendl; | |
eafe8130 | 1123 | } |
f6b5b4d7 | 1124 | void _rm(BlueStore::Onode* o) override |
9f95a23c | 1125 | { |
39ae355f TL |
1126 | o->clear_cached(); |
1127 | if (o->lru_item.is_linked()) { | |
20effc67 | 1128 | *(o->cache_age_bin) -= 1; |
9f95a23c TL |
1129 | lru.erase(lru.iterator_to(*o)); |
1130 | } | |
f6b5b4d7 TL |
1131 | ceph_assert(num); |
1132 | --num; | |
1133 | dout(20) << __func__ << " " << this << " " << " " << o->oid << " removed, num=" << num << dendl; | |
9f95a23c | 1134 | } |
39ae355f TL |
1135 | |
1136 | void maybe_unpin(BlueStore::Onode* o) override | |
adb31ebb | 1137 | { |
39ae355f TL |
1138 | OnodeCacheShard* ocs = this; |
1139 | ocs->lock.lock(); | |
1140 | // It is possible that during waiting split_cache moved us to different OnodeCacheShard. | |
1141 | while (ocs != o->c->get_onode_cache()) { | |
1142 | ocs->lock.unlock(); | |
1143 | ocs = o->c->get_onode_cache(); | |
1144 | ocs->lock.lock(); | |
1145 | } | |
1146 | if (o->is_cached() && o->pin_nref == 1) { | |
1147 | if(!o->lru_item.is_linked()) { | |
1148 | if (o->exists) { | |
1149 | lru.push_front(*o); | |
1150 | o->cache_age_bin = age_bins.front(); | |
1151 | *(o->cache_age_bin) += 1; | |
1152 | dout(20) << __func__ << " " << this << " " << o->oid << " unpinned" | |
1153 | << dendl; | |
1154 | } else { | |
1155 | ceph_assert(num); | |
1156 | --num; | |
1157 | o->clear_cached(); | |
1158 | dout(20) << __func__ << " " << this << " " << o->oid << " removed" | |
1159 | << dendl; | |
1160 | // remove will also decrement nref | |
1161 | o->c->onode_space._remove(o->oid); | |
1162 | } | |
1163 | } else if (o->exists) { | |
1164 | // move onode within LRU | |
1165 | lru.erase(lru.iterator_to(*o)); | |
1166 | lru.push_front(*o); | |
1167 | if (o->cache_age_bin != age_bins.front()) { | |
1168 | *(o->cache_age_bin) -= 1; | |
1169 | o->cache_age_bin = age_bins.front(); | |
1170 | *(o->cache_age_bin) += 1; | |
1171 | } | |
1172 | dout(20) << __func__ << " " << this << " " << o->oid << " touched" | |
1173 | << dendl; | |
1174 | } | |
1175 | } | |
1176 | ocs->lock.unlock(); | |
adb31ebb | 1177 | } |
39ae355f | 1178 | |
9f95a23c TL |
1179 | void _trim_to(uint64_t new_size) override |
1180 | { | |
1181 | if (new_size >= lru.size()) { | |
1182 | return; // don't even try | |
1183 | } | |
39ae355f TL |
1184 | uint64_t n = num - new_size; // note: we might get empty LRU |
1185 | // before n == 0 due to pinned | |
1186 | // entries. And hence being unable | |
1187 | // to reach new_size target. | |
1188 | while (n-- > 0 && lru.size() > 0) { | |
1189 | BlueStore::Onode *o = &lru.back(); | |
1190 | lru.pop_back(); | |
1191 | ||
f6b5b4d7 | 1192 | dout(20) << __func__ << " rm " << o->oid << " " |
39ae355f TL |
1193 | << o->nref << " " << o->cached << dendl; |
1194 | ||
1195 | *(o->cache_age_bin) -= 1; | |
1196 | if (o->pin_nref > 1) { | |
1197 | dout(20) << __func__ << " " << this << " " << " " << " " << o->oid << dendl; | |
9f95a23c | 1198 | } else { |
39ae355f TL |
1199 | ceph_assert(num); |
1200 | --num; | |
1201 | o->clear_cached(); | |
1202 | o->c->onode_space._remove(o->oid); | |
9f95a23c | 1203 | } |
9f95a23c | 1204 | } |
f6b5b4d7 | 1205 | } |
39ae355f | 1206 | void _move_pinned(OnodeCacheShard *to, BlueStore::Onode *o) override |
f6b5b4d7 TL |
1207 | { |
1208 | if (to == this) { | |
1209 | return; | |
1210 | } | |
39ae355f TL |
1211 | _rm(o); |
1212 | ceph_assert(o->nref > 1); | |
1213 | to->_add(o, 0); | |
9f95a23c TL |
1214 | } |
1215 | void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override | |
1216 | { | |
39ae355f | 1217 | std::lock_guard l(lock); |
f6b5b4d7 | 1218 | *onodes += num; |
39ae355f | 1219 | *pinned_onodes += num - lru.size(); |
9f95a23c TL |
1220 | } |
1221 | }; | |
7c673cae | 1222 | |
9f95a23c TL |
1223 | // OnodeCacheShard |
1224 | BlueStore::OnodeCacheShard *BlueStore::OnodeCacheShard::create( | |
1225 | CephContext* cct, | |
1226 | string type, | |
1227 | PerfCounters *logger) | |
7c673cae | 1228 | { |
9f95a23c TL |
1229 | BlueStore::OnodeCacheShard *c = nullptr; |
1230 | // Currently we only implement an LRU cache for onodes | |
1231 | c = new LruOnodeCacheShard(cct); | |
1232 | c->logger = logger; | |
1233 | return c; | |
7c673cae FG |
1234 | } |
1235 | ||
9f95a23c TL |
1236 | // LruBufferCacheShard |
1237 | struct LruBufferCacheShard : public BlueStore::BufferCacheShard { | |
1238 | typedef boost::intrusive::list< | |
1239 | BlueStore::Buffer, | |
1240 | boost::intrusive::member_hook< | |
1241 | BlueStore::Buffer, | |
1242 | boost::intrusive::list_member_hook<>, | |
1243 | &BlueStore::Buffer::lru_item> > list_t; | |
1244 | list_t lru; | |
1245 | ||
1246 | explicit LruBufferCacheShard(CephContext *cct) : BlueStore::BufferCacheShard(cct) {} | |
1247 | ||
1248 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override { | |
1249 | if (near) { | |
1250 | auto q = lru.iterator_to(*near); | |
1251 | lru.insert(q, *b); | |
1252 | } else if (level > 0) { | |
1253 | lru.push_front(*b); | |
1254 | } else { | |
1255 | lru.push_back(*b); | |
7c673cae | 1256 | } |
9f95a23c | 1257 | buffer_bytes += b->length; |
20effc67 TL |
1258 | b->cache_age_bin = age_bins.front(); |
1259 | *(b->cache_age_bin) += b->length; | |
9f95a23c TL |
1260 | num = lru.size(); |
1261 | } | |
1262 | void _rm(BlueStore::Buffer *b) override { | |
1263 | ceph_assert(buffer_bytes >= b->length); | |
1264 | buffer_bytes -= b->length; | |
20effc67 TL |
1265 | assert(*(b->cache_age_bin) >= b->length); |
1266 | *(b->cache_age_bin) -= b->length; | |
9f95a23c TL |
1267 | auto q = lru.iterator_to(*b); |
1268 | lru.erase(q); | |
1269 | num = lru.size(); | |
1270 | } | |
1271 | void _move(BlueStore::BufferCacheShard *src, BlueStore::Buffer *b) override { | |
1272 | src->_rm(b); | |
1273 | _add(b, 0, nullptr); | |
1274 | } | |
1275 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override { | |
1276 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1277 | buffer_bytes += delta; | |
20effc67 TL |
1278 | assert(*(b->cache_age_bin) + delta >= 0); |
1279 | *(b->cache_age_bin) += delta; | |
9f95a23c TL |
1280 | } |
1281 | void _touch(BlueStore::Buffer *b) override { | |
1282 | auto p = lru.iterator_to(*b); | |
1283 | lru.erase(p); | |
1284 | lru.push_front(*b); | |
20effc67 TL |
1285 | *(b->cache_age_bin) -= b->length; |
1286 | b->cache_age_bin = age_bins.front(); | |
1287 | *(b->cache_age_bin) += b->length; | |
9f95a23c TL |
1288 | num = lru.size(); |
1289 | _audit("_touch_buffer end"); | |
1290 | } | |
7c673cae | 1291 | |
9f95a23c TL |
1292 | void _trim_to(uint64_t max) override |
1293 | { | |
1294 | while (buffer_bytes > max) { | |
1295 | auto i = lru.rbegin(); | |
1296 | if (i == lru.rend()) { | |
1297 | // stop if lru is now empty | |
7c673cae FG |
1298 | break; |
1299 | } | |
1300 | ||
9f95a23c TL |
1301 | BlueStore::Buffer *b = &*i; |
1302 | ceph_assert(b->is_clean()); | |
1303 | dout(20) << __func__ << " rm " << *b << dendl; | |
20effc67 TL |
1304 | assert(*(b->cache_age_bin) >= b->length); |
1305 | *(b->cache_age_bin) -= b->length; | |
9f95a23c | 1306 | b->space->_rm_buffer(this, b); |
7c673cae | 1307 | } |
9f95a23c | 1308 | num = lru.size(); |
7c673cae | 1309 | } |
7c673cae | 1310 | |
9f95a23c TL |
1311 | void add_stats(uint64_t *extents, |
1312 | uint64_t *blobs, | |
1313 | uint64_t *buffers, | |
1314 | uint64_t *bytes) override { | |
1315 | *extents += num_extents; | |
1316 | *blobs += num_blobs; | |
1317 | *buffers += num; | |
1318 | *bytes += buffer_bytes; | |
7c673cae | 1319 | } |
9f95a23c TL |
1320 | #ifdef DEBUG_CACHE |
1321 | void _audit(const char *s) override | |
1322 | { | |
1323 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1324 | uint64_t s = 0; | |
1325 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1326 | s += i->length; | |
1327 | } | |
1328 | if (s != buffer_bytes) { | |
1329 | derr << __func__ << " buffer_size " << buffer_bytes << " actual " << s | |
1330 | << dendl; | |
1331 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1332 | derr << __func__ << " " << *i << dendl; | |
1333 | } | |
1334 | ceph_assert(s == buffer_bytes); | |
7c673cae | 1335 | } |
9f95a23c TL |
1336 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1337 | << " ok" << dendl; | |
7c673cae | 1338 | } |
7c673cae | 1339 | #endif |
9f95a23c | 1340 | }; |
7c673cae | 1341 | |
9f95a23c TL |
1342 | // TwoQBufferCacheShard |
1343 | ||
1344 | struct TwoQBufferCacheShard : public BlueStore::BufferCacheShard { | |
1345 | typedef boost::intrusive::list< | |
1346 | BlueStore::Buffer, | |
1347 | boost::intrusive::member_hook< | |
1348 | BlueStore::Buffer, | |
1349 | boost::intrusive::list_member_hook<>, | |
1350 | &BlueStore::Buffer::lru_item> > list_t; | |
1351 | list_t hot; ///< "Am" hot buffers | |
1352 | list_t warm_in; ///< "A1in" newly warm buffers | |
1353 | list_t warm_out; ///< "A1out" empty buffers we've evicted | |
9f95a23c TL |
1354 | |
1355 | enum { | |
1356 | BUFFER_NEW = 0, | |
1357 | BUFFER_WARM_IN, ///< in warm_in | |
1358 | BUFFER_WARM_OUT, ///< in warm_out | |
1359 | BUFFER_HOT, ///< in hot | |
1360 | BUFFER_TYPE_MAX | |
1361 | }; | |
7c673cae | 1362 | |
9f95a23c | 1363 | uint64_t list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type |
7c673cae | 1364 | |
9f95a23c TL |
1365 | public: |
1366 | explicit TwoQBufferCacheShard(CephContext *cct) : BufferCacheShard(cct) {} | |
7c673cae | 1367 | |
9f95a23c TL |
1368 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override |
1369 | { | |
1370 | dout(20) << __func__ << " level " << level << " near " << near | |
1371 | << " on " << *b | |
1372 | << " which has cache_private " << b->cache_private << dendl; | |
1373 | if (near) { | |
1374 | b->cache_private = near->cache_private; | |
1375 | switch (b->cache_private) { | |
1376 | case BUFFER_WARM_IN: | |
1377 | warm_in.insert(warm_in.iterator_to(*near), *b); | |
1378 | break; | |
1379 | case BUFFER_WARM_OUT: | |
1380 | ceph_assert(b->is_empty()); | |
1381 | warm_out.insert(warm_out.iterator_to(*near), *b); | |
1382 | break; | |
1383 | case BUFFER_HOT: | |
1384 | hot.insert(hot.iterator_to(*near), *b); | |
1385 | break; | |
1386 | default: | |
1387 | ceph_abort_msg("bad cache_private"); | |
1388 | } | |
1389 | } else if (b->cache_private == BUFFER_NEW) { | |
1390 | b->cache_private = BUFFER_WARM_IN; | |
1391 | if (level > 0) { | |
1392 | warm_in.push_front(*b); | |
1393 | } else { | |
1394 | // take caller hint to start at the back of the warm queue | |
1395 | warm_in.push_back(*b); | |
1396 | } | |
1397 | } else { | |
1398 | // we got a hint from discard | |
1399 | switch (b->cache_private) { | |
1400 | case BUFFER_WARM_IN: | |
1401 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1402 | // do this. | |
1403 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1404 | warm_in.push_front(*b); | |
1405 | break; | |
1406 | case BUFFER_WARM_OUT: | |
1407 | b->cache_private = BUFFER_HOT; | |
1408 | // move to hot. fall-thru | |
1409 | case BUFFER_HOT: | |
1410 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1411 | hot.push_front(*b); | |
1412 | break; | |
1413 | default: | |
1414 | ceph_abort_msg("bad cache_private"); | |
1415 | } | |
1416 | } | |
20effc67 | 1417 | b->cache_age_bin = age_bins.front(); |
9f95a23c TL |
1418 | if (!b->is_empty()) { |
1419 | buffer_bytes += b->length; | |
1420 | list_bytes[b->cache_private] += b->length; | |
20effc67 | 1421 | *(b->cache_age_bin) += b->length; |
9f95a23c TL |
1422 | } |
1423 | num = hot.size() + warm_in.size(); | |
1424 | } | |
1425 | ||
1426 | void _rm(BlueStore::Buffer *b) override | |
1427 | { | |
1428 | dout(20) << __func__ << " " << *b << dendl; | |
1429 | if (!b->is_empty()) { | |
1430 | ceph_assert(buffer_bytes >= b->length); | |
1431 | buffer_bytes -= b->length; | |
1432 | ceph_assert(list_bytes[b->cache_private] >= b->length); | |
1433 | list_bytes[b->cache_private] -= b->length; | |
20effc67 TL |
1434 | assert(*(b->cache_age_bin) >= b->length); |
1435 | *(b->cache_age_bin) -= b->length; | |
9f95a23c | 1436 | } |
7c673cae FG |
1437 | switch (b->cache_private) { |
1438 | case BUFFER_WARM_IN: | |
9f95a23c | 1439 | warm_in.erase(warm_in.iterator_to(*b)); |
7c673cae FG |
1440 | break; |
1441 | case BUFFER_WARM_OUT: | |
9f95a23c | 1442 | warm_out.erase(warm_out.iterator_to(*b)); |
7c673cae FG |
1443 | break; |
1444 | case BUFFER_HOT: | |
9f95a23c | 1445 | hot.erase(hot.iterator_to(*b)); |
7c673cae FG |
1446 | break; |
1447 | default: | |
11fdf7f2 | 1448 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1449 | } |
9f95a23c TL |
1450 | num = hot.size() + warm_in.size(); |
1451 | } | |
1452 | ||
1453 | void _move(BlueStore::BufferCacheShard *srcc, BlueStore::Buffer *b) override | |
1454 | { | |
1455 | TwoQBufferCacheShard *src = static_cast<TwoQBufferCacheShard*>(srcc); | |
1456 | src->_rm(b); | |
1457 | ||
1458 | // preserve which list we're on (even if we can't preserve the order!) | |
7c673cae FG |
1459 | switch (b->cache_private) { |
1460 | case BUFFER_WARM_IN: | |
9f95a23c TL |
1461 | ceph_assert(!b->is_empty()); |
1462 | warm_in.push_back(*b); | |
7c673cae FG |
1463 | break; |
1464 | case BUFFER_WARM_OUT: | |
9f95a23c TL |
1465 | ceph_assert(b->is_empty()); |
1466 | warm_out.push_back(*b); | |
1467 | break; | |
7c673cae | 1468 | case BUFFER_HOT: |
9f95a23c TL |
1469 | ceph_assert(!b->is_empty()); |
1470 | hot.push_back(*b); | |
7c673cae FG |
1471 | break; |
1472 | default: | |
11fdf7f2 | 1473 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1474 | } |
9f95a23c TL |
1475 | if (!b->is_empty()) { |
1476 | buffer_bytes += b->length; | |
1477 | list_bytes[b->cache_private] += b->length; | |
20effc67 | 1478 | *(b->cache_age_bin) += b->length; |
9f95a23c TL |
1479 | } |
1480 | num = hot.size() + warm_in.size(); | |
7c673cae | 1481 | } |
7c673cae | 1482 | |
9f95a23c TL |
1483 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override |
1484 | { | |
1485 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1486 | if (!b->is_empty()) { | |
1487 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1488 | buffer_bytes += delta; | |
1489 | ceph_assert((int64_t)list_bytes[b->cache_private] + delta >= 0); | |
1490 | list_bytes[b->cache_private] += delta; | |
20effc67 TL |
1491 | assert(*(b->cache_age_bin) + delta >= 0); |
1492 | *(b->cache_age_bin) += delta; | |
9f95a23c | 1493 | } |
7c673cae | 1494 | } |
7c673cae | 1495 | |
9f95a23c TL |
1496 | void _touch(BlueStore::Buffer *b) override { |
1497 | switch (b->cache_private) { | |
1498 | case BUFFER_WARM_IN: | |
1499 | // do nothing (somewhat counter-intuitively!) | |
1500 | break; | |
1501 | case BUFFER_WARM_OUT: | |
1502 | // move from warm_out to hot LRU | |
1503 | ceph_abort_msg("this happens via discard hint"); | |
1504 | break; | |
1505 | case BUFFER_HOT: | |
1506 | // move to front of hot LRU | |
1507 | hot.erase(hot.iterator_to(*b)); | |
1508 | hot.push_front(*b); | |
1509 | break; | |
1510 | } | |
20effc67 TL |
1511 | *(b->cache_age_bin) -= b->length; |
1512 | b->cache_age_bin = age_bins.front(); | |
1513 | *(b->cache_age_bin) += b->length; | |
9f95a23c TL |
1514 | num = hot.size() + warm_in.size(); |
1515 | _audit("_touch_buffer end"); | |
7c673cae | 1516 | } |
7c673cae | 1517 | |
9f95a23c TL |
1518 | void _trim_to(uint64_t max) override |
1519 | { | |
1520 | if (buffer_bytes > max) { | |
1521 | uint64_t kin = max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1522 | uint64_t khot = max - kin; | |
1523 | ||
1524 | // pre-calculate kout based on average buffer size too, | |
1525 | // which is typical(the warm_in and hot lists may change later) | |
1526 | uint64_t kout = 0; | |
1527 | uint64_t buffer_num = hot.size() + warm_in.size(); | |
1528 | if (buffer_num) { | |
1529 | uint64_t avg_size = buffer_bytes / buffer_num; | |
1530 | ceph_assert(avg_size); | |
1531 | uint64_t calculated_num = max / avg_size; | |
1532 | kout = calculated_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1533 | } | |
1534 | ||
1535 | if (list_bytes[BUFFER_HOT] < khot) { | |
1536 | // hot is small, give slack to warm_in | |
1537 | kin += khot - list_bytes[BUFFER_HOT]; | |
1538 | } else if (list_bytes[BUFFER_WARM_IN] < kin) { | |
1539 | // warm_in is small, give slack to hot | |
1540 | khot += kin - list_bytes[BUFFER_WARM_IN]; | |
1541 | } | |
1542 | ||
1543 | // adjust warm_in list | |
1544 | int64_t to_evict_bytes = list_bytes[BUFFER_WARM_IN] - kin; | |
1545 | uint64_t evicted = 0; | |
1546 | ||
1547 | while (to_evict_bytes > 0) { | |
1548 | auto p = warm_in.rbegin(); | |
1549 | if (p == warm_in.rend()) { | |
1550 | // stop if warm_in list is now empty | |
1551 | break; | |
1552 | } | |
7c673cae | 1553 | |
9f95a23c TL |
1554 | BlueStore::Buffer *b = &*p; |
1555 | ceph_assert(b->is_clean()); | |
1556 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; | |
1557 | ceph_assert(buffer_bytes >= b->length); | |
1558 | buffer_bytes -= b->length; | |
1559 | ceph_assert(list_bytes[BUFFER_WARM_IN] >= b->length); | |
1560 | list_bytes[BUFFER_WARM_IN] -= b->length; | |
20effc67 TL |
1561 | assert(*(b->cache_age_bin) >= b->length); |
1562 | *(b->cache_age_bin) -= b->length; | |
1563 | to_evict_bytes -= b->length; | |
9f95a23c TL |
1564 | evicted += b->length; |
1565 | b->state = BlueStore::Buffer::STATE_EMPTY; | |
1566 | b->data.clear(); | |
1567 | warm_in.erase(warm_in.iterator_to(*b)); | |
1568 | warm_out.push_front(*b); | |
1569 | b->cache_private = BUFFER_WARM_OUT; | |
1570 | } | |
1571 | ||
1572 | if (evicted > 0) { | |
1573 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1574 | << " from warm_in list, done evicting warm_in buffers" | |
1575 | << dendl; | |
1576 | } | |
7c673cae | 1577 | |
9f95a23c TL |
1578 | // adjust hot list |
1579 | to_evict_bytes = list_bytes[BUFFER_HOT] - khot; | |
1580 | evicted = 0; | |
7c673cae | 1581 | |
9f95a23c TL |
1582 | while (to_evict_bytes > 0) { |
1583 | auto p = hot.rbegin(); | |
1584 | if (p == hot.rend()) { | |
1585 | // stop if hot list is now empty | |
1586 | break; | |
1587 | } | |
7c673cae | 1588 | |
9f95a23c TL |
1589 | BlueStore::Buffer *b = &*p; |
1590 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
1591 | ceph_assert(b->is_clean()); | |
1592 | // adjust evict size before buffer goes invalid | |
1593 | to_evict_bytes -= b->length; | |
1594 | evicted += b->length; | |
1595 | b->space->_rm_buffer(this, b); | |
1596 | } | |
7c673cae | 1597 | |
9f95a23c TL |
1598 | if (evicted > 0) { |
1599 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1600 | << " from hot list, done evicting hot buffers" | |
1601 | << dendl; | |
7c673cae FG |
1602 | } |
1603 | ||
9f95a23c TL |
1604 | // adjust warm out list too, if necessary |
1605 | int64_t n = warm_out.size() - kout; | |
1606 | while (n-- > 0) { | |
1607 | BlueStore::Buffer *b = &*warm_out.rbegin(); | |
1608 | ceph_assert(b->is_empty()); | |
1609 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; | |
1610 | b->space->_rm_buffer(this, b); | |
1611 | } | |
7c673cae | 1612 | } |
9f95a23c TL |
1613 | num = hot.size() + warm_in.size(); |
1614 | } | |
7c673cae | 1615 | |
9f95a23c TL |
1616 | void add_stats(uint64_t *extents, |
1617 | uint64_t *blobs, | |
1618 | uint64_t *buffers, | |
1619 | uint64_t *bytes) override { | |
1620 | *extents += num_extents; | |
1621 | *blobs += num_blobs; | |
1622 | *buffers += num; | |
1623 | *bytes += buffer_bytes; | |
1624 | } | |
7c673cae | 1625 | |
9f95a23c TL |
1626 | #ifdef DEBUG_CACHE |
1627 | void _audit(const char *s) override | |
1628 | { | |
1629 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1630 | uint64_t s = 0; | |
1631 | for (auto i = hot.begin(); i != hot.end(); ++i) { | |
1632 | s += i->length; | |
7c673cae FG |
1633 | } |
1634 | ||
9f95a23c TL |
1635 | uint64_t hot_bytes = s; |
1636 | if (hot_bytes != list_bytes[BUFFER_HOT]) { | |
1637 | derr << __func__ << " hot_list_bytes " | |
1638 | << list_bytes[BUFFER_HOT] | |
1639 | << " != actual " << hot_bytes | |
1640 | << dendl; | |
1641 | ceph_assert(hot_bytes == list_bytes[BUFFER_HOT]); | |
7c673cae FG |
1642 | } |
1643 | ||
9f95a23c TL |
1644 | for (auto i = warm_in.begin(); i != warm_in.end(); ++i) { |
1645 | s += i->length; | |
7c673cae | 1646 | } |
7c673cae | 1647 | |
9f95a23c TL |
1648 | uint64_t warm_in_bytes = s - hot_bytes; |
1649 | if (warm_in_bytes != list_bytes[BUFFER_WARM_IN]) { | |
1650 | derr << __func__ << " warm_in_list_bytes " | |
1651 | << list_bytes[BUFFER_WARM_IN] | |
1652 | << " != actual " << warm_in_bytes | |
1653 | << dendl; | |
1654 | ceph_assert(warm_in_bytes == list_bytes[BUFFER_WARM_IN]); | |
7c673cae | 1655 | } |
7c673cae | 1656 | |
9f95a23c TL |
1657 | if (s != buffer_bytes) { |
1658 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1659 | << dendl; | |
1660 | ceph_assert(s == buffer_bytes); | |
1661 | } | |
7c673cae | 1662 | |
9f95a23c TL |
1663 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1664 | << " ok" << dendl; | |
7c673cae | 1665 | } |
9f95a23c TL |
1666 | #endif |
1667 | }; | |
7c673cae | 1668 | |
9f95a23c | 1669 | // BuferCacheShard |
7c673cae | 1670 | |
9f95a23c TL |
1671 | BlueStore::BufferCacheShard *BlueStore::BufferCacheShard::create( |
1672 | CephContext* cct, | |
1673 | string type, | |
1674 | PerfCounters *logger) | |
1675 | { | |
1676 | BufferCacheShard *c = nullptr; | |
1677 | if (type == "lru") | |
1678 | c = new LruBufferCacheShard(cct); | |
1679 | else if (type == "2q") | |
1680 | c = new TwoQBufferCacheShard(cct); | |
1681 | else | |
1682 | ceph_abort_msg("unrecognized cache type"); | |
1683 | c->logger = logger; | |
1684 | return c; | |
7c673cae | 1685 | } |
7c673cae FG |
1686 | |
1687 | // BufferSpace | |
1688 | ||
1689 | #undef dout_prefix | |
1690 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1691 | ||
9f95a23c | 1692 | void BlueStore::BufferSpace::_clear(BufferCacheShard* cache) |
7c673cae FG |
1693 | { |
1694 | // note: we already hold cache->lock | |
1695 | ldout(cache->cct, 20) << __func__ << dendl; | |
1696 | while (!buffer_map.empty()) { | |
1697 | _rm_buffer(cache, buffer_map.begin()); | |
1698 | } | |
1699 | } | |
1700 | ||
9f95a23c | 1701 | int BlueStore::BufferSpace::_discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) |
7c673cae FG |
1702 | { |
1703 | // note: we already hold cache->lock | |
1704 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1705 | << std::dec << dendl; | |
1706 | int cache_private = 0; | |
1707 | cache->_audit("discard start"); | |
1708 | auto i = _data_lower_bound(offset); | |
1709 | uint32_t end = offset + length; | |
1710 | while (i != buffer_map.end()) { | |
1711 | Buffer *b = i->second.get(); | |
1712 | if (b->offset >= end) { | |
1713 | break; | |
1714 | } | |
1715 | if (b->cache_private > cache_private) { | |
1716 | cache_private = b->cache_private; | |
1717 | } | |
1718 | if (b->offset < offset) { | |
1719 | int64_t front = offset - b->offset; | |
1720 | if (b->end() > end) { | |
1721 | // drop middle (split) | |
1722 | uint32_t tail = b->end() - end; | |
1723 | if (b->data.length()) { | |
1724 | bufferlist bl; | |
1725 | bl.substr_of(b->data, b->length - tail, tail); | |
f67539c2 | 1726 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl, b->flags); |
31f18b77 FG |
1727 | nb->maybe_rebuild(); |
1728 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1729 | } else { |
f67539c2 TL |
1730 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail, |
1731 | b->flags), | |
1732 | 0, b); | |
7c673cae FG |
1733 | } |
1734 | if (!b->is_writing()) { | |
9f95a23c | 1735 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1736 | } |
1737 | b->truncate(front); | |
31f18b77 | 1738 | b->maybe_rebuild(); |
7c673cae FG |
1739 | cache->_audit("discard end 1"); |
1740 | break; | |
1741 | } else { | |
1742 | // drop tail | |
1743 | if (!b->is_writing()) { | |
9f95a23c | 1744 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1745 | } |
1746 | b->truncate(front); | |
31f18b77 | 1747 | b->maybe_rebuild(); |
7c673cae FG |
1748 | ++i; |
1749 | continue; | |
1750 | } | |
1751 | } | |
1752 | if (b->end() <= end) { | |
1753 | // drop entire buffer | |
1754 | _rm_buffer(cache, i++); | |
1755 | continue; | |
1756 | } | |
1757 | // drop front | |
1758 | uint32_t keep = b->end() - end; | |
1759 | if (b->data.length()) { | |
1760 | bufferlist bl; | |
1761 | bl.substr_of(b->data, b->length - keep, keep); | |
f67539c2 | 1762 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl, b->flags); |
31f18b77 FG |
1763 | nb->maybe_rebuild(); |
1764 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1765 | } else { |
f67539c2 TL |
1766 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep, |
1767 | b->flags), | |
1768 | 0, b); | |
7c673cae FG |
1769 | } |
1770 | _rm_buffer(cache, i); | |
1771 | cache->_audit("discard end 2"); | |
1772 | break; | |
1773 | } | |
1774 | return cache_private; | |
1775 | } | |
1776 | ||
1777 | void BlueStore::BufferSpace::read( | |
9f95a23c | 1778 | BufferCacheShard* cache, |
224ce89b WB |
1779 | uint32_t offset, |
1780 | uint32_t length, | |
7c673cae | 1781 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1782 | interval_set<uint32_t>& res_intervals, |
1783 | int flags) | |
7c673cae | 1784 | { |
7c673cae FG |
1785 | res.clear(); |
1786 | res_intervals.clear(); | |
1787 | uint32_t want_bytes = length; | |
1788 | uint32_t end = offset + length; | |
224ce89b WB |
1789 | |
1790 | { | |
11fdf7f2 | 1791 | std::lock_guard l(cache->lock); |
224ce89b WB |
1792 | for (auto i = _data_lower_bound(offset); |
1793 | i != buffer_map.end() && offset < end && i->first < end; | |
1794 | ++i) { | |
1795 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1796 | ceph_assert(b->end() > offset); |
91327a77 AA |
1797 | |
1798 | bool val = false; | |
1799 | if (flags & BYPASS_CLEAN_CACHE) | |
1800 | val = b->is_writing(); | |
1801 | else | |
1802 | val = b->is_writing() || b->is_clean(); | |
1803 | if (val) { | |
224ce89b WB |
1804 | if (b->offset < offset) { |
1805 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1806 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1807 | res[offset].substr_of(b->data, skip, l); |
1808 | res_intervals.insert(offset, l); | |
1809 | offset += l; | |
1810 | length -= l; | |
1811 | if (!b->is_writing()) { | |
9f95a23c | 1812 | cache->_touch(b); |
f67539c2 | 1813 | } |
224ce89b WB |
1814 | continue; |
1815 | } | |
1816 | if (b->offset > offset) { | |
1817 | uint32_t gap = b->offset - offset; | |
1818 | if (length <= gap) { | |
1819 | break; | |
1820 | } | |
1821 | offset += gap; | |
1822 | length -= gap; | |
1823 | } | |
1824 | if (!b->is_writing()) { | |
9f95a23c | 1825 | cache->_touch(b); |
224ce89b WB |
1826 | } |
1827 | if (b->length > length) { | |
1828 | res[offset].substr_of(b->data, 0, length); | |
1829 | res_intervals.insert(offset, length); | |
7c673cae | 1830 | break; |
224ce89b WB |
1831 | } else { |
1832 | res[offset].append(b->data); | |
1833 | res_intervals.insert(offset, b->length); | |
1834 | if (b->length == length) | |
1835 | break; | |
1836 | offset += b->length; | |
1837 | length -= b->length; | |
1838 | } | |
7c673cae FG |
1839 | } |
1840 | } | |
1841 | } | |
1842 | ||
1843 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1844 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1845 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1846 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1847 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1848 | } | |
1849 | ||
9f95a23c | 1850 | void BlueStore::BufferSpace::_finish_write(BufferCacheShard* cache, uint64_t seq) |
7c673cae | 1851 | { |
7c673cae FG |
1852 | auto i = writing.begin(); |
1853 | while (i != writing.end()) { | |
1854 | if (i->seq > seq) { | |
1855 | break; | |
1856 | } | |
1857 | if (i->seq < seq) { | |
1858 | ++i; | |
1859 | continue; | |
1860 | } | |
1861 | ||
1862 | Buffer *b = &*i; | |
11fdf7f2 | 1863 | ceph_assert(b->is_writing()); |
7c673cae FG |
1864 | |
1865 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1866 | writing.erase(i++); | |
1867 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1868 | buffer_map.erase(b->offset); | |
1869 | } else { | |
1870 | b->state = Buffer::STATE_CLEAN; | |
1871 | writing.erase(i++); | |
31f18b77 FG |
1872 | b->maybe_rebuild(); |
1873 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
9f95a23c | 1874 | cache->_add(b, 1, nullptr); |
7c673cae FG |
1875 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; |
1876 | } | |
1877 | } | |
9f95a23c | 1878 | cache->_trim(); |
7c673cae FG |
1879 | cache->_audit("finish_write end"); |
1880 | } | |
1881 | ||
9f95a23c | 1882 | void BlueStore::BufferSpace::split(BufferCacheShard* cache, size_t pos, BlueStore::BufferSpace &r) |
7c673cae | 1883 | { |
11fdf7f2 | 1884 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1885 | if (buffer_map.empty()) |
1886 | return; | |
1887 | ||
1888 | auto p = --buffer_map.end(); | |
1889 | while (true) { | |
1890 | if (p->second->end() <= pos) | |
1891 | break; | |
1892 | ||
1893 | if (p->second->offset < pos) { | |
1894 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1895 | size_t left = pos - p->second->offset; | |
1896 | size_t right = p->second->length - left; | |
1897 | if (p->second->data.length()) { | |
1898 | bufferlist bl; | |
1899 | bl.substr_of(p->second->data, left, right); | |
f67539c2 TL |
1900 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, |
1901 | 0, bl, p->second->flags), | |
7c673cae FG |
1902 | 0, p->second.get()); |
1903 | } else { | |
f67539c2 TL |
1904 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, |
1905 | 0, right, p->second->flags), | |
7c673cae FG |
1906 | 0, p->second.get()); |
1907 | } | |
9f95a23c | 1908 | cache->_adjust_size(p->second.get(), -right); |
7c673cae FG |
1909 | p->second->truncate(left); |
1910 | break; | |
1911 | } | |
1912 | ||
11fdf7f2 | 1913 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1914 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1915 | if (p->second->data.length()) { | |
1916 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
f67539c2 | 1917 | p->second->offset - pos, p->second->data, p->second->flags), |
7c673cae FG |
1918 | 0, p->second.get()); |
1919 | } else { | |
1920 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
f67539c2 | 1921 | p->second->offset - pos, p->second->length, p->second->flags), |
7c673cae FG |
1922 | 0, p->second.get()); |
1923 | } | |
1924 | if (p == buffer_map.begin()) { | |
1925 | _rm_buffer(cache, p); | |
1926 | break; | |
1927 | } else { | |
1928 | _rm_buffer(cache, p--); | |
1929 | } | |
1930 | } | |
11fdf7f2 | 1931 | ceph_assert(writing.empty()); |
9f95a23c | 1932 | cache->_trim(); |
7c673cae FG |
1933 | } |
1934 | ||
1935 | // OnodeSpace | |
1936 | ||
1937 | #undef dout_prefix | |
1938 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1939 | ||
39ae355f | 1940 | BlueStore::OnodeRef BlueStore::OnodeSpace::add_onode(const ghobject_t& oid, |
f6b5b4d7 | 1941 | OnodeRef& o) |
7c673cae | 1942 | { |
11fdf7f2 | 1943 | std::lock_guard l(cache->lock); |
39ae355f TL |
1944 | // add entry or return existing one |
1945 | auto p = onode_map.emplace(oid, o); | |
1946 | if (!p.second) { | |
7c673cae | 1947 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o |
39ae355f | 1948 | << " raced, returning existing " << p.first->second |
7c673cae | 1949 | << dendl; |
39ae355f | 1950 | return p.first->second; |
7c673cae | 1951 | } |
f6b5b4d7 | 1952 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << o << dendl; |
f6b5b4d7 | 1953 | cache->_add(o.get(), 1); |
9f95a23c | 1954 | cache->_trim(); |
7c673cae FG |
1955 | return o; |
1956 | } | |
1957 | ||
f6b5b4d7 TL |
1958 | void BlueStore::OnodeSpace::_remove(const ghobject_t& oid) |
1959 | { | |
1960 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << dendl; | |
1961 | onode_map.erase(oid); | |
1962 | } | |
1963 | ||
7c673cae FG |
1964 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) |
1965 | { | |
7c673cae | 1966 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b | 1967 | OnodeRef o; |
224ce89b WB |
1968 | |
1969 | { | |
11fdf7f2 | 1970 | std::lock_guard l(cache->lock); |
224ce89b WB |
1971 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1972 | if (p == onode_map.end()) { | |
1973 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
39ae355f | 1974 | cache->logger->inc(l_bluestore_onode_misses); |
224ce89b WB |
1975 | } else { |
1976 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
f6b5b4d7 TL |
1977 | << " " << p->second->nref |
1978 | << " " << p->second->cached | |
224ce89b | 1979 | << dendl; |
f6b5b4d7 TL |
1980 | // This will pin onode and implicitly touch the cache when Onode |
1981 | // eventually will become unpinned | |
224ce89b | 1982 | o = p->second; |
f6b5b4d7 | 1983 | |
20effc67 | 1984 | cache->logger->inc(l_bluestore_onode_hits); |
224ce89b WB |
1985 | } |
1986 | } | |
1987 | ||
224ce89b | 1988 | return o; |
7c673cae FG |
1989 | } |
1990 | ||
1991 | void BlueStore::OnodeSpace::clear() | |
1992 | { | |
11fdf7f2 | 1993 | std::lock_guard l(cache->lock); |
f6b5b4d7 | 1994 | ldout(cache->cct, 10) << __func__ << " " << onode_map.size()<< dendl; |
7c673cae | 1995 | for (auto &p : onode_map) { |
f6b5b4d7 | 1996 | cache->_rm(p.second.get()); |
7c673cae FG |
1997 | } |
1998 | onode_map.clear(); | |
1999 | } | |
2000 | ||
2001 | bool BlueStore::OnodeSpace::empty() | |
2002 | { | |
11fdf7f2 | 2003 | std::lock_guard l(cache->lock); |
7c673cae FG |
2004 | return onode_map.empty(); |
2005 | } | |
2006 | ||
2007 | void BlueStore::OnodeSpace::rename( | |
2008 | OnodeRef& oldo, | |
2009 | const ghobject_t& old_oid, | |
2010 | const ghobject_t& new_oid, | |
f91f0fd5 | 2011 | const mempool::bluestore_cache_meta::string& new_okey) |
7c673cae | 2012 | { |
11fdf7f2 | 2013 | std::lock_guard l(cache->lock); |
7c673cae FG |
2014 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
2015 | << dendl; | |
2016 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
2017 | po = onode_map.find(old_oid); | |
2018 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 2019 | ceph_assert(po != pn); |
7c673cae | 2020 | |
11fdf7f2 | 2021 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
2022 | if (pn != onode_map.end()) { |
2023 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
2024 | << dendl; | |
f6b5b4d7 | 2025 | cache->_rm(pn->second.get()); |
7c673cae FG |
2026 | onode_map.erase(pn); |
2027 | } | |
2028 | OnodeRef o = po->second; | |
2029 | ||
2030 | // install a non-existent onode at old location | |
2031 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
2032 | po->second = oldo; | |
f6b5b4d7 TL |
2033 | cache->_add(oldo.get(), 1); |
2034 | // add at new position and fix oid, key. | |
2035 | // This will pin 'o' and implicitly touch cache | |
2036 | // when it will eventually become unpinned | |
7c673cae | 2037 | onode_map.insert(make_pair(new_oid, o)); |
f6b5b4d7 | 2038 | |
7c673cae FG |
2039 | o->oid = new_oid; |
2040 | o->key = new_okey; | |
9f95a23c | 2041 | cache->_trim(); |
7c673cae FG |
2042 | } |
2043 | ||
adb31ebb | 2044 | bool BlueStore::OnodeSpace::map_any(std::function<bool(Onode*)> f) |
7c673cae | 2045 | { |
11fdf7f2 | 2046 | std::lock_guard l(cache->lock); |
7c673cae FG |
2047 | ldout(cache->cct, 20) << __func__ << dendl; |
2048 | for (auto& i : onode_map) { | |
adb31ebb | 2049 | if (f(i.second.get())) { |
7c673cae FG |
2050 | return true; |
2051 | } | |
2052 | } | |
2053 | return false; | |
2054 | } | |
2055 | ||
11fdf7f2 TL |
2056 | template <int LogLevelV = 30> |
2057 | void BlueStore::OnodeSpace::dump(CephContext *cct) | |
3efd9988 FG |
2058 | { |
2059 | for (auto& i : onode_map) { | |
f6b5b4d7 TL |
2060 | ldout(cct, LogLevelV) << i.first << " : " << i.second |
2061 | << " " << i.second->nref | |
2062 | << " " << i.second->cached | |
f6b5b4d7 | 2063 | << dendl; |
3efd9988 FG |
2064 | } |
2065 | } | |
7c673cae FG |
2066 | |
2067 | // SharedBlob | |
2068 | ||
2069 | #undef dout_prefix | |
2070 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
9f95a23c TL |
2071 | #undef dout_context |
2072 | #define dout_context coll->store->cct | |
7c673cae | 2073 | |
9f95a23c | 2074 | void BlueStore::SharedBlob::dump(Formatter* f) const |
7c673cae | 2075 | { |
9f95a23c TL |
2076 | f->dump_bool("loaded", loaded); |
2077 | if (loaded) { | |
2078 | persistent->dump(f); | |
2079 | } else { | |
2080 | f->dump_unsigned("sbid_unloaded", sbid_unloaded); | |
2081 | } | |
2082 | } | |
2083 | ||
2084 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
2085 | { | |
2086 | out << "SharedBlob(" << &sb; | |
2087 | ||
7c673cae FG |
2088 | if (sb.loaded) { |
2089 | out << " loaded " << *sb.persistent; | |
2090 | } else { | |
2091 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
2092 | } | |
2093 | return out << ")"; | |
2094 | } | |
2095 | ||
2096 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
2097 | : coll(_coll), sbid_unloaded(i) | |
2098 | { | |
11fdf7f2 | 2099 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
2100 | if (get_cache()) { |
2101 | get_cache()->add_blob(); | |
2102 | } | |
2103 | } | |
2104 | ||
2105 | BlueStore::SharedBlob::~SharedBlob() | |
2106 | { | |
7c673cae FG |
2107 | if (loaded && persistent) { |
2108 | delete persistent; | |
2109 | } | |
2110 | } | |
2111 | ||
2112 | void BlueStore::SharedBlob::put() | |
2113 | { | |
2114 | if (--nref == 0) { | |
9f95a23c TL |
2115 | dout(20) << __func__ << " " << this |
2116 | << " removing self from set " << get_parent() | |
2117 | << dendl; | |
1adf2230 AA |
2118 | again: |
2119 | auto coll_snap = coll; | |
2120 | if (coll_snap) { | |
11fdf7f2 | 2121 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
2122 | if (coll_snap != coll) { |
2123 | goto again; | |
2124 | } | |
91327a77 AA |
2125 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
2126 | // race with lookup | |
2127 | return; | |
2128 | } | |
1adf2230 AA |
2129 | bc._clear(coll_snap->cache); |
2130 | coll_snap->cache->rm_blob(); | |
7c673cae | 2131 | } |
28e407b8 | 2132 | delete this; |
7c673cae FG |
2133 | } |
2134 | } | |
2135 | ||
2136 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
2137 | { | |
11fdf7f2 | 2138 | ceph_assert(persistent); |
7c673cae FG |
2139 | persistent->ref_map.get(offset, length); |
2140 | } | |
2141 | ||
2142 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 2143 | PExtentVector *r, |
11fdf7f2 | 2144 | bool *unshare) |
7c673cae | 2145 | { |
11fdf7f2 TL |
2146 | ceph_assert(persistent); |
2147 | persistent->ref_map.put(offset, length, r, | |
2148 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
2149 | } |
2150 | ||
f64942e4 AA |
2151 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
2152 | { | |
2153 | while (true) { | |
9f95a23c | 2154 | BufferCacheShard *cache = coll->cache; |
11fdf7f2 | 2155 | std::lock_guard l(cache->lock); |
f64942e4 | 2156 | if (coll->cache != cache) { |
9f95a23c TL |
2157 | dout(20) << __func__ |
2158 | << " raced with sb cache update, was " << cache | |
2159 | << ", now " << coll->cache << ", retrying" | |
2160 | << dendl; | |
f64942e4 AA |
2161 | continue; |
2162 | } | |
2163 | bc._finish_write(cache, seq); | |
2164 | break; | |
2165 | } | |
2166 | } | |
2167 | ||
3efd9988 FG |
2168 | // SharedBlobSet |
2169 | ||
2170 | #undef dout_prefix | |
2171 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
2172 | ||
11fdf7f2 TL |
2173 | template <int LogLevelV = 30> |
2174 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 2175 | { |
11fdf7f2 | 2176 | std::lock_guard l(lock); |
3efd9988 | 2177 | for (auto& i : sb_map) { |
11fdf7f2 | 2178 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
2179 | } |
2180 | } | |
2181 | ||
7c673cae FG |
2182 | // Blob |
2183 | ||
2184 | #undef dout_prefix | |
2185 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
2186 | ||
9f95a23c TL |
2187 | void BlueStore::Blob::dump(Formatter* f) const |
2188 | { | |
2189 | if (is_spanning()) { | |
2190 | f->dump_unsigned("spanning_id ", id); | |
2191 | } | |
2192 | blob.dump(f); | |
2193 | if (shared_blob) { | |
2194 | f->dump_object("shared", *shared_blob); | |
2195 | } | |
2196 | } | |
2197 | ||
7c673cae FG |
2198 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) |
2199 | { | |
2200 | out << "Blob(" << &b; | |
2201 | if (b.is_spanning()) { | |
2202 | out << " spanning " << b.id; | |
2203 | } | |
35e4c445 FG |
2204 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
2205 | if (b.shared_blob) { | |
2206 | out << " " << *b.shared_blob; | |
2207 | } else { | |
2208 | out << " (shared_blob=NULL)"; | |
2209 | } | |
2210 | out << ")"; | |
7c673cae FG |
2211 | return out; |
2212 | } | |
2213 | ||
2214 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
2215 | { | |
224ce89b | 2216 | if (get_blob().is_shared()) { |
7c673cae FG |
2217 | return; |
2218 | } | |
224ce89b | 2219 | if (get_blob().is_compressed()) { |
7c673cae FG |
2220 | bool discard = false; |
2221 | bool all_invalid = true; | |
224ce89b | 2222 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
2223 | if (!e.is_valid()) { |
2224 | discard = true; | |
2225 | } else { | |
2226 | all_invalid = false; | |
2227 | } | |
2228 | } | |
11fdf7f2 | 2229 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
2230 | // or none pextents are invalid. |
2231 | if (discard) { | |
224ce89b WB |
2232 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
2233 | get_blob().get_logical_length()); | |
7c673cae FG |
2234 | } |
2235 | } else { | |
2236 | size_t pos = 0; | |
224ce89b | 2237 | for (auto e : get_blob().get_extents()) { |
7c673cae | 2238 | if (!e.is_valid()) { |
9f95a23c TL |
2239 | dout(20) << __func__ << " 0x" << std::hex << pos |
2240 | << "~" << e.length | |
2241 | << std::dec << dendl; | |
7c673cae FG |
2242 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); |
2243 | } | |
2244 | pos += e.length; | |
2245 | } | |
224ce89b WB |
2246 | if (get_blob().can_prune_tail()) { |
2247 | dirty_blob().prune_tail(); | |
2248 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
224ce89b | 2249 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
2250 | } |
2251 | } | |
2252 | } | |
2253 | ||
2254 | void BlueStore::Blob::get_ref( | |
2255 | Collection *coll, | |
2256 | uint32_t offset, | |
2257 | uint32_t length) | |
2258 | { | |
2259 | // Caller has to initialize Blob's logical length prior to increment | |
2260 | // references. Otherwise one is neither unable to determine required | |
2261 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
2262 | // for single counter mode. | |
11fdf7f2 | 2263 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
2264 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2265 | << std::dec << " " << *this << dendl; | |
2266 | ||
2267 | if (used_in_blob.is_empty()) { | |
2268 | uint32_t min_release_size = | |
224ce89b WB |
2269 | get_blob().get_release_size(coll->store->min_alloc_size); |
2270 | uint64_t l = get_blob().get_logical_length(); | |
2271 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
2272 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
2273 | used_in_blob.init(l, min_release_size); |
2274 | } | |
2275 | used_in_blob.get( | |
2276 | offset, | |
2277 | length); | |
2278 | } | |
2279 | ||
2280 | bool BlueStore::Blob::put_ref( | |
2281 | Collection *coll, | |
2282 | uint32_t offset, | |
2283 | uint32_t length, | |
2284 | PExtentVector *r) | |
2285 | { | |
2286 | PExtentVector logical; | |
2287 | ||
7c673cae FG |
2288 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2289 | << std::dec << " " << *this << dendl; | |
2290 | ||
2291 | bool empty = used_in_blob.put( | |
2292 | offset, | |
2293 | length, | |
2294 | &logical); | |
2295 | r->clear(); | |
2296 | // nothing to release | |
2297 | if (!empty && logical.empty()) { | |
2298 | return false; | |
2299 | } | |
2300 | ||
2301 | bluestore_blob_t& b = dirty_blob(); | |
2302 | return b.release_extents(empty, logical, r); | |
2303 | } | |
2304 | ||
224ce89b | 2305 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
2306 | uint32_t target_blob_size, |
2307 | uint32_t b_offset, | |
2308 | uint32_t *length0) { | |
11fdf7f2 TL |
2309 | ceph_assert(min_alloc_size); |
2310 | ceph_assert(target_blob_size); | |
7c673cae FG |
2311 | if (!get_blob().is_mutable()) { |
2312 | return false; | |
2313 | } | |
2314 | ||
2315 | uint32_t length = *length0; | |
2316 | uint32_t end = b_offset + length; | |
2317 | ||
2318 | // Currently for the sake of simplicity we omit blob reuse if data is | |
2319 | // unaligned with csum chunk. Later we can perform padding if needed. | |
2320 | if (get_blob().has_csum() && | |
2321 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
2322 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
2323 | return false; | |
2324 | } | |
2325 | ||
2326 | auto blen = get_blob().get_logical_length(); | |
2327 | uint32_t new_blen = blen; | |
2328 | ||
2329 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 2330 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
2331 | |
2332 | if (b_offset >= blen) { | |
224ce89b WB |
2333 | // new data totally stands out of the existing blob |
2334 | new_blen = end; | |
7c673cae | 2335 | } else { |
224ce89b | 2336 | // new data overlaps with the existing blob |
11fdf7f2 | 2337 | new_blen = std::max(blen, end); |
224ce89b WB |
2338 | |
2339 | uint32_t overlap = 0; | |
2340 | if (new_blen > blen) { | |
2341 | overlap = blen - b_offset; | |
2342 | } else { | |
2343 | overlap = length; | |
2344 | } | |
2345 | ||
2346 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
2347 | // abort if any piece of the overlap has already been allocated | |
2348 | return false; | |
7c673cae FG |
2349 | } |
2350 | } | |
224ce89b | 2351 | |
7c673cae FG |
2352 | if (new_blen > blen) { |
2353 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
2354 | // Unable to decrease the provided length to fit into max_blob_size | |
2355 | if (overflow >= length) { | |
2356 | return false; | |
2357 | } | |
2358 | ||
2359 | // FIXME: in some cases we could reduce unused resolution | |
2360 | if (get_blob().has_unused()) { | |
2361 | return false; | |
2362 | } | |
2363 | ||
2364 | if (overflow > 0) { | |
2365 | new_blen -= overflow; | |
2366 | length -= overflow; | |
2367 | *length0 = length; | |
2368 | } | |
224ce89b | 2369 | |
7c673cae FG |
2370 | if (new_blen > blen) { |
2371 | dirty_blob().add_tail(new_blen); | |
2372 | used_in_blob.add_tail(new_blen, | |
224ce89b | 2373 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
2374 | } |
2375 | } | |
2376 | return true; | |
2377 | } | |
2378 | ||
2379 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
2380 | { | |
7c673cae FG |
2381 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec |
2382 | << " start " << *this << dendl; | |
11fdf7f2 TL |
2383 | ceph_assert(blob.can_split()); |
2384 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
2385 | bluestore_blob_t &lb = dirty_blob(); |
2386 | bluestore_blob_t &rb = r->dirty_blob(); | |
2387 | ||
2388 | used_in_blob.split( | |
2389 | blob_offset, | |
2390 | &(r->used_in_blob)); | |
2391 | ||
2392 | lb.split(blob_offset, rb); | |
2393 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2394 | ||
2395 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2396 | << " finish " << *this << dendl; | |
2397 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2398 | << " and " << *r << dendl; | |
2399 | } | |
2400 | ||
2401 | #ifndef CACHE_BLOB_BL | |
2402 | void BlueStore::Blob::decode( | |
11fdf7f2 | 2403 | bufferptr::const_iterator& p, |
7c673cae FG |
2404 | uint64_t struct_v, |
2405 | uint64_t* sbid, | |
39ae355f TL |
2406 | bool include_ref_map, |
2407 | Collection *coll) | |
7c673cae FG |
2408 | { |
2409 | denc(blob, p, struct_v); | |
2410 | if (blob.is_shared()) { | |
2411 | denc(*sbid, p); | |
2412 | } | |
2413 | if (include_ref_map) { | |
2414 | if (struct_v > 1) { | |
2415 | used_in_blob.decode(p); | |
2416 | } else { | |
2417 | used_in_blob.clear(); | |
2418 | bluestore_extent_ref_map_t legacy_ref_map; | |
2419 | legacy_ref_map.decode(p); | |
39ae355f TL |
2420 | if (coll) { |
2421 | for (auto r : legacy_ref_map.ref_map) { | |
2422 | get_ref( | |
2423 | coll, | |
2424 | r.first, | |
2425 | r.second.refs * r.second.length); | |
2426 | } | |
7c673cae FG |
2427 | } |
2428 | } | |
2429 | } | |
2430 | } | |
2431 | #endif | |
2432 | ||
2433 | // Extent | |
2434 | ||
9f95a23c TL |
2435 | void BlueStore::Extent::dump(Formatter* f) const |
2436 | { | |
2437 | f->dump_unsigned("logical_offset", logical_offset); | |
2438 | f->dump_unsigned("length", length); | |
2439 | f->dump_unsigned("blob_offset", blob_offset); | |
2440 | f->dump_object("blob", *blob); | |
2441 | } | |
2442 | ||
7c673cae FG |
2443 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) |
2444 | { | |
2445 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2446 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2447 | << " " << *e.blob; | |
2448 | } | |
2449 | ||
2450 | // OldExtent | |
2451 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2452 | uint32_t lo, | |
2453 | uint32_t o, | |
2454 | uint32_t l, | |
2455 | BlobRef& b) { | |
2456 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2457 | b->put_ref(c.get(), o, l, &(oe->r)); | |
adb31ebb | 2458 | oe->blob_empty = !b->is_referenced(); |
7c673cae FG |
2459 | return oe; |
2460 | } | |
2461 | ||
2462 | // ExtentMap | |
2463 | ||
2464 | #undef dout_prefix | |
2465 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
9f95a23c TL |
2466 | #undef dout_context |
2467 | #define dout_context onode->c->store->cct | |
7c673cae | 2468 | |
39ae355f | 2469 | BlueStore::ExtentMap::ExtentMap(Onode *o, size_t inline_shard_prealloc_size) |
7c673cae | 2470 | : onode(o), |
39ae355f | 2471 | inline_bl(inline_shard_prealloc_size) { |
7c673cae FG |
2472 | } |
2473 | ||
9f95a23c TL |
2474 | void BlueStore::ExtentMap::dump(Formatter* f) const |
2475 | { | |
2476 | f->open_array_section("extents"); | |
2477 | ||
2478 | for (auto& e : extent_map) { | |
2479 | f->dump_object("extent", e); | |
2480 | } | |
2481 | f->close_section(); | |
2482 | } | |
2483 | ||
11fdf7f2 TL |
2484 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2485 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2486 | uint64_t& length, uint64_t& dstoff) { | |
2487 | ||
2488 | auto cct = onode->c->store->cct; | |
2489 | bool inject_21040 = | |
2490 | cct->_conf->bluestore_debug_inject_bug21040; | |
2491 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2492 | for (auto& e : oldo->extent_map.extent_map) { | |
2493 | e.blob->last_encoded_id = -1; | |
2494 | } | |
2495 | ||
2496 | int n = 0; | |
2497 | uint64_t end = srcoff + length; | |
2498 | uint32_t dirty_range_begin = 0; | |
2499 | uint32_t dirty_range_end = 0; | |
2500 | bool src_dirty = false; | |
2501 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2502 | ep != oldo->extent_map.extent_map.end(); | |
2503 | ++ep) { | |
2504 | auto& e = *ep; | |
2505 | if (e.logical_offset >= end) { | |
2506 | break; | |
2507 | } | |
2508 | dout(20) << __func__ << " src " << e << dendl; | |
2509 | BlobRef cb; | |
2510 | bool blob_duped = true; | |
2511 | if (e.blob->last_encoded_id >= 0) { | |
2512 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2513 | blob_duped = false; | |
2514 | } else { | |
2515 | // dup the blob | |
2516 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2517 | // make sure it is shared | |
2518 | if (!blob.is_shared()) { | |
2519 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2520 | if (!inject_21040 && !src_dirty) { | |
2521 | src_dirty = true; | |
2522 | dirty_range_begin = e.logical_offset; | |
2523 | } else if (inject_21040 && | |
2524 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2525 | dirty_range_begin = e.logical_offset; | |
2526 | } | |
2527 | ceph_assert(e.logical_end() > 0); | |
2528 | // -1 to exclude next potential shard | |
2529 | dirty_range_end = e.logical_end() - 1; | |
2530 | } else { | |
2531 | c->load_shared_blob(e.blob->shared_blob); | |
2532 | } | |
2533 | cb = new Blob(); | |
2534 | e.blob->last_encoded_id = n; | |
2535 | id_to_blob[n] = cb; | |
2536 | e.blob->dup(*cb); | |
2537 | // bump the extent refs on the copied blob's extents | |
2538 | for (auto p : blob.get_extents()) { | |
2539 | if (p.is_valid()) { | |
2540 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2541 | } | |
2542 | } | |
2543 | txc->write_shared_blob(e.blob->shared_blob); | |
2544 | dout(20) << __func__ << " new " << *cb << dendl; | |
2545 | } | |
2546 | ||
2547 | int skip_front, skip_back; | |
2548 | if (e.logical_offset < srcoff) { | |
2549 | skip_front = srcoff - e.logical_offset; | |
2550 | } else { | |
2551 | skip_front = 0; | |
2552 | } | |
2553 | if (e.logical_end() > end) { | |
2554 | skip_back = e.logical_end() - end; | |
2555 | } else { | |
2556 | skip_back = 0; | |
2557 | } | |
2558 | ||
2559 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2560 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2561 | newo->extent_map.extent_map.insert(*ne); | |
2562 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2563 | // fixme: we may leave parts of new blob unreferenced that could | |
2564 | // be freed (relative to the shared_blob). | |
2565 | txc->statfs_delta.stored() += ne->length; | |
2566 | if (e.blob->get_blob().is_compressed()) { | |
2567 | txc->statfs_delta.compressed_original() += ne->length; | |
2568 | if (blob_duped) { | |
2569 | txc->statfs_delta.compressed() += | |
2570 | cb->get_blob().get_compressed_payload_length(); | |
2571 | } | |
2572 | } | |
2573 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2574 | ++n; | |
2575 | } | |
2576 | if ((!inject_21040 && src_dirty) || | |
2577 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2578 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2579 | dirty_range_end - dirty_range_begin); | |
2580 | txc->write_onode(oldo); | |
2581 | } | |
2582 | txc->write_onode(newo); | |
2583 | ||
2584 | if (dstoff + length > newo->onode.size) { | |
2585 | newo->onode.size = dstoff + length; | |
2586 | } | |
2587 | newo->extent_map.dirty_range(dstoff, length); | |
2588 | } | |
7c673cae FG |
2589 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2590 | bool force) | |
2591 | { | |
2592 | auto cct = onode->c->store->cct; //used by dout | |
2593 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2594 | if (onode->onode.extent_map_shards.empty()) { | |
2595 | if (inline_bl.length() == 0) { | |
2596 | unsigned n; | |
2597 | // we need to encode inline_bl to measure encoded length | |
2598 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
f91f0fd5 | 2599 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_inline_bl); |
11fdf7f2 | 2600 | ceph_assert(!never_happen); |
7c673cae FG |
2601 | size_t len = inline_bl.length(); |
2602 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2603 | << " extents" << dendl; | |
2604 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2605 | request_reshard(0, OBJECT_MAX_SIZE); | |
2606 | return; | |
2607 | } | |
2608 | } | |
2609 | // will persist in the onode key. | |
2610 | } else { | |
2611 | // pending shard update | |
2612 | struct dirty_shard_t { | |
2613 | Shard *shard; | |
2614 | bufferlist bl; | |
2615 | dirty_shard_t(Shard *s) : shard(s) {} | |
2616 | }; | |
2617 | vector<dirty_shard_t> encoded_shards; | |
2618 | // allocate slots for all shards in a single call instead of | |
2619 | // doing multiple allocations - one per each dirty shard | |
2620 | encoded_shards.reserve(shards.size()); | |
2621 | ||
2622 | auto p = shards.begin(); | |
2623 | auto prev_p = p; | |
2624 | while (p != shards.end()) { | |
11fdf7f2 | 2625 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2626 | auto n = p; |
2627 | ++n; | |
2628 | if (p->dirty) { | |
2629 | uint32_t endoff; | |
2630 | if (n == shards.end()) { | |
2631 | endoff = OBJECT_MAX_SIZE; | |
2632 | } else { | |
2633 | endoff = n->shard_info->offset; | |
2634 | } | |
2635 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2636 | bufferlist& bl = encoded_shards.back().bl; | |
2637 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2638 | bl, &p->extents)) { | |
2639 | if (force) { | |
2640 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2641 | ceph_assert(!force); |
7c673cae FG |
2642 | } |
2643 | } | |
2644 | size_t len = bl.length(); | |
2645 | ||
2646 | dout(20) << __func__ << " shard 0x" << std::hex | |
2647 | << p->shard_info->offset << std::dec << " is " << len | |
2648 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2649 | << p->extents << " extents" << dendl; | |
2650 | ||
2651 | if (!force) { | |
2652 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2653 | // we are big; reshard ourselves | |
2654 | request_reshard(p->shard_info->offset, endoff); | |
2655 | } | |
2656 | // avoid resharding the trailing shard, even if it is small | |
2657 | else if (n != shards.end() && | |
11fdf7f2 TL |
2658 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2659 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2660 | if (p == shards.begin()) { |
2661 | // we are the first shard, combine with next shard | |
7c673cae | 2662 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2663 | } else { |
31f18b77 FG |
2664 | // combine either with the previous shard or the next, |
2665 | // whichever is smaller | |
7c673cae FG |
2666 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2667 | request_reshard(p->shard_info->offset, endoff + 1); | |
2668 | } else { | |
2669 | request_reshard(prev_p->shard_info->offset, endoff); | |
2670 | } | |
2671 | } | |
2672 | } | |
2673 | } | |
2674 | } | |
2675 | prev_p = p; | |
2676 | p = n; | |
2677 | } | |
2678 | if (needs_reshard()) { | |
2679 | return; | |
2680 | } | |
2681 | ||
2682 | // schedule DB update for dirty shards | |
2683 | string key; | |
2684 | for (auto& it : encoded_shards) { | |
20effc67 TL |
2685 | dout(20) << __func__ << " encoding key for shard 0x" << std::hex |
2686 | << it.shard->shard_info->offset << std::dec << dendl; | |
7c673cae FG |
2687 | it.shard->dirty = false; |
2688 | it.shard->shard_info->bytes = it.bl.length(); | |
2689 | generate_extent_shard_key_and_apply( | |
2690 | onode->key, | |
2691 | it.shard->shard_info->offset, | |
2692 | &key, | |
2693 | [&](const string& final_key) { | |
2694 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2695 | } | |
2696 | ); | |
2697 | } | |
2698 | } | |
2699 | } | |
2700 | ||
31f18b77 FG |
2701 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2702 | { | |
2703 | if (spanning_blob_map.empty()) | |
2704 | return 0; | |
2705 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2706 | // bid is valid and available. | |
2707 | if (bid >= 0) | |
2708 | return bid; | |
2709 | // Find next unused bid; | |
2710 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2711 | const auto begin_bid = bid; | |
2712 | do { | |
2713 | if (!spanning_blob_map.count(bid)) | |
2714 | return bid; | |
2715 | else { | |
2716 | bid++; | |
2717 | if (bid < 0) bid = 0; | |
2718 | } | |
2719 | } while (bid != begin_bid); | |
81eedcae TL |
2720 | auto cct = onode->c->store->cct; // used by dout |
2721 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2722 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2723 | } |
2724 | ||
7c673cae FG |
2725 | void BlueStore::ExtentMap::reshard( |
2726 | KeyValueDB *db, | |
2727 | KeyValueDB::Transaction t) | |
2728 | { | |
2729 | auto cct = onode->c->store->cct; // used by dout | |
2730 | ||
2731 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2732 | << needs_reshard_end << ")" << std::dec | |
2733 | << " of " << onode->onode.extent_map_shards.size() | |
2734 | << " shards on " << onode->oid << dendl; | |
2735 | for (auto& p : spanning_blob_map) { | |
2736 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2737 | << dendl; | |
2738 | } | |
2739 | // determine shard index range | |
2740 | unsigned si_begin = 0, si_end = 0; | |
2741 | if (!shards.empty()) { | |
2742 | while (si_begin + 1 < shards.size() && | |
2743 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2744 | ++si_begin; | |
2745 | } | |
2746 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2747 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2748 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2749 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2750 | break; | |
2751 | } | |
2752 | } | |
2753 | if (si_end == shards.size()) { | |
2754 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2755 | } | |
2756 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2757 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2758 | << needs_reshard_end << ")" << std::dec << dendl; | |
2759 | } | |
2760 | ||
181888fb | 2761 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2762 | |
2763 | // we may need to fault in a larger interval later must have all | |
2764 | // referring extents for spanning blobs loaded in order to have | |
2765 | // accurate use_tracker values. | |
2766 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2767 | uint32_t spanning_scan_end = needs_reshard_end; | |
2768 | ||
2769 | // remove old keys | |
2770 | string key; | |
2771 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2772 | generate_extent_shard_key_and_apply( | |
2773 | onode->key, shards[i].shard_info->offset, &key, | |
2774 | [&](const string& final_key) { | |
2775 | t->rmkey(PREFIX_OBJ, final_key); | |
2776 | } | |
2777 | ); | |
2778 | } | |
2779 | ||
2780 | // calculate average extent size | |
2781 | unsigned bytes = 0; | |
2782 | unsigned extents = 0; | |
2783 | if (onode->onode.extent_map_shards.empty()) { | |
2784 | bytes = inline_bl.length(); | |
2785 | extents = extent_map.size(); | |
2786 | } else { | |
2787 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2788 | bytes += shards[i].shard_info->bytes; | |
2789 | extents += shards[i].extents; | |
2790 | } | |
2791 | } | |
2792 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2793 | unsigned slop = target * | |
2794 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2795 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2796 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2797 | << ", slop " << slop << dendl; | |
2798 | ||
2799 | // reshard | |
2800 | unsigned estimate = 0; | |
31f18b77 | 2801 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2802 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2803 | unsigned max_blob_end = 0; | |
2804 | Extent dummy(needs_reshard_begin); | |
2805 | for (auto e = extent_map.lower_bound(dummy); | |
2806 | e != extent_map.end(); | |
2807 | ++e) { | |
2808 | if (e->logical_offset >= needs_reshard_end) { | |
2809 | break; | |
2810 | } | |
2811 | dout(30) << " extent " << *e << dendl; | |
2812 | ||
2813 | // disfavor shard boundaries that span a blob | |
2814 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2815 | if (estimate && | |
2816 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2817 | // new shard | |
31f18b77 | 2818 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2819 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2820 | new_shard_info.back().offset = offset; | |
2821 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2822 | << std::dec << dendl; |
7c673cae FG |
2823 | } |
2824 | offset = e->logical_offset; | |
2825 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2826 | new_shard_info.back().offset = offset; | |
2827 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2828 | << std::dec << dendl; | |
2829 | estimate = 0; | |
2830 | } | |
2831 | estimate += extent_avg; | |
31f18b77 FG |
2832 | unsigned bs = e->blob_start(); |
2833 | if (bs < spanning_scan_begin) { | |
2834 | spanning_scan_begin = bs; | |
7c673cae FG |
2835 | } |
2836 | uint32_t be = e->blob_end(); | |
2837 | if (be > max_blob_end) { | |
2838 | max_blob_end = be; | |
2839 | } | |
2840 | if (be > spanning_scan_end) { | |
2841 | spanning_scan_end = be; | |
2842 | } | |
2843 | } | |
2844 | if (new_shard_info.empty() && (si_begin > 0 || | |
2845 | si_end < shards.size())) { | |
2846 | // we resharded a partial range; we must produce at least one output | |
2847 | // shard | |
2848 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2849 | new_shard_info.back().offset = needs_reshard_begin; | |
2850 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2851 | << std::dec << " (singleton degenerate case)" << dendl; | |
2852 | } | |
2853 | ||
2854 | auto& sv = onode->onode.extent_map_shards; | |
2855 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2856 | dout(20) << __func__ << " old " << sv << dendl; | |
2857 | if (sv.empty()) { | |
2858 | // no old shards to keep | |
2859 | sv.swap(new_shard_info); | |
2860 | init_shards(true, true); | |
2861 | } else { | |
2862 | // splice in new shards | |
2863 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2864 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2865 | sv.insert( | |
2866 | sv.begin() + si_begin, | |
2867 | new_shard_info.begin(), | |
2868 | new_shard_info.end()); | |
2869 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2870 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2871 | |
11fdf7f2 | 2872 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2873 | |
2874 | // note that we need to update every shard_info of shards here, | |
2875 | // as sv might have been totally re-allocated above | |
2876 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2877 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2878 | } |
2879 | ||
2880 | // mark newly added shards as dirty | |
2881 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2882 | shards[i].loaded = true; |
2883 | shards[i].dirty = true; | |
2884 | } | |
7c673cae FG |
2885 | } |
2886 | dout(20) << __func__ << " fin " << sv << dendl; | |
2887 | inline_bl.clear(); | |
2888 | ||
2889 | if (sv.empty()) { | |
2890 | // no more shards; unspan all previously spanning blobs | |
2891 | auto p = spanning_blob_map.begin(); | |
2892 | while (p != spanning_blob_map.end()) { | |
2893 | p->second->id = -1; | |
2894 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2895 | p = spanning_blob_map.erase(p); | |
2896 | } | |
2897 | } else { | |
2898 | // identify new spanning blobs | |
2899 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2900 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2901 | if (spanning_scan_begin < needs_reshard_begin) { | |
2902 | fault_range(db, spanning_scan_begin, | |
2903 | needs_reshard_begin - spanning_scan_begin); | |
2904 | } | |
2905 | if (spanning_scan_end > needs_reshard_end) { | |
2906 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2907 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2908 | } |
2909 | auto sp = sv.begin() + si_begin; | |
2910 | auto esp = sv.end(); | |
2911 | unsigned shard_start = sp->offset; | |
2912 | unsigned shard_end; | |
2913 | ++sp; | |
2914 | if (sp == esp) { | |
2915 | shard_end = OBJECT_MAX_SIZE; | |
2916 | } else { | |
2917 | shard_end = sp->offset; | |
2918 | } | |
7c673cae | 2919 | Extent dummy(needs_reshard_begin); |
9f95a23c TL |
2920 | |
2921 | bool was_too_many_blobs_check = false; | |
2922 | auto too_many_blobs_threshold = | |
2923 | g_conf()->bluestore_debug_too_many_blobs_threshold; | |
39ae355f TL |
2924 | auto& dumped_onodes = onode->c->onode_space.cache->dumped_onodes; |
2925 | decltype(onode->c->onode_space.cache->dumped_onodes)::value_type* oid_slot = nullptr; | |
2926 | decltype(onode->c->onode_space.cache->dumped_onodes)::value_type* oldest_slot = nullptr; | |
9f95a23c | 2927 | |
7c673cae FG |
2928 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { |
2929 | if (e->logical_offset >= needs_reshard_end) { | |
2930 | break; | |
2931 | } | |
2932 | dout(30) << " extent " << *e << dendl; | |
2933 | while (e->logical_offset >= shard_end) { | |
2934 | shard_start = shard_end; | |
11fdf7f2 | 2935 | ceph_assert(sp != esp); |
7c673cae FG |
2936 | ++sp; |
2937 | if (sp == esp) { | |
2938 | shard_end = OBJECT_MAX_SIZE; | |
2939 | } else { | |
2940 | shard_end = sp->offset; | |
2941 | } | |
2942 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2943 | << " to 0x" << shard_end << std::dec << dendl; | |
2944 | } | |
9f95a23c | 2945 | |
7c673cae FG |
2946 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { |
2947 | if (!e->blob->is_spanning()) { | |
2948 | // We have two options: (1) split the blob into pieces at the | |
2949 | // shard boundaries (and adjust extents accordingly), or (2) | |
2950 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2951 | // we may have to split it multiple times--potentially at every | |
2952 | // shard boundary. | |
2953 | bool must_span = false; | |
2954 | BlobRef b = e->blob; | |
2955 | if (b->can_split()) { | |
2956 | uint32_t bstart = e->blob_start(); | |
2957 | uint32_t bend = e->blob_end(); | |
2958 | for (const auto& sh : shards) { | |
2959 | if (bstart < sh.shard_info->offset && | |
2960 | bend > sh.shard_info->offset) { | |
2961 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2962 | if (b->can_split_at(blob_offset)) { | |
2963 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2964 | << std::hex << bstart << " blob_offset 0x" | |
2965 | << blob_offset << std::dec << " " << *b << dendl; | |
2966 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2967 | // switch b to the new right-hand side, in case it | |
2968 | // *also* has to get split. | |
2969 | bstart += blob_offset; | |
2970 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2971 | } else { | |
2972 | must_span = true; | |
2973 | break; | |
2974 | } | |
2975 | } | |
2976 | } | |
2977 | } else { | |
2978 | must_span = true; | |
2979 | } | |
2980 | if (must_span) { | |
31f18b77 FG |
2981 | auto bid = allocate_spanning_blob_id(); |
2982 | b->id = bid; | |
7c673cae FG |
2983 | spanning_blob_map[b->id] = b; |
2984 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
9f95a23c TL |
2985 | if (!was_too_many_blobs_check && |
2986 | too_many_blobs_threshold && | |
2987 | spanning_blob_map.size() >= size_t(too_many_blobs_threshold)) { | |
2988 | ||
2989 | was_too_many_blobs_check = true; | |
2990 | for (size_t i = 0; i < dumped_onodes.size(); ++i) { | |
2991 | if (dumped_onodes[i].first == onode->oid) { | |
2992 | oid_slot = &dumped_onodes[i]; | |
2993 | break; | |
2994 | } | |
2995 | if (!oldest_slot || (oldest_slot && | |
2996 | dumped_onodes[i].second < oldest_slot->second)) { | |
2997 | oldest_slot = &dumped_onodes[i]; | |
2998 | } | |
2999 | } | |
3000 | } | |
7c673cae FG |
3001 | } |
3002 | } | |
3003 | } else { | |
3004 | if (e->blob->is_spanning()) { | |
3005 | spanning_blob_map.erase(e->blob->id); | |
3006 | e->blob->id = -1; | |
3007 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
3008 | } | |
3009 | } | |
3010 | } | |
9f95a23c TL |
3011 | bool do_dump = (!oid_slot && was_too_many_blobs_check) || |
3012 | (oid_slot && | |
3013 | (mono_clock::now() - oid_slot->second >= make_timespan(5 * 60))); | |
3014 | if (do_dump) { | |
3015 | dout(0) << __func__ | |
3016 | << " spanning blob count exceeds threshold, " | |
3017 | << spanning_blob_map.size() << " spanning blobs" | |
3018 | << dendl; | |
3019 | _dump_onode<0>(cct, *onode); | |
3020 | if (oid_slot) { | |
3021 | oid_slot->second = mono_clock::now(); | |
3022 | } else { | |
3023 | ceph_assert(oldest_slot); | |
3024 | oldest_slot->first = onode->oid; | |
3025 | oldest_slot->second = mono_clock::now(); | |
3026 | } | |
3027 | } | |
7c673cae FG |
3028 | } |
3029 | ||
3030 | clear_needs_reshard(); | |
3031 | } | |
3032 | ||
3033 | bool BlueStore::ExtentMap::encode_some( | |
3034 | uint32_t offset, | |
3035 | uint32_t length, | |
3036 | bufferlist& bl, | |
3037 | unsigned *pn) | |
3038 | { | |
7c673cae FG |
3039 | Extent dummy(offset); |
3040 | auto start = extent_map.lower_bound(dummy); | |
3041 | uint32_t end = offset + length; | |
3042 | ||
3043 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
3044 | // serialization only. Hence there is no specific | |
3045 | // handling at ExtentMap level. | |
3046 | ||
3047 | unsigned n = 0; | |
3048 | size_t bound = 0; | |
7c673cae FG |
3049 | bool must_reshard = false; |
3050 | for (auto p = start; | |
3051 | p != extent_map.end() && p->logical_offset < end; | |
3052 | ++p, ++n) { | |
11fdf7f2 | 3053 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
3054 | p->blob->last_encoded_id = -1; |
3055 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
3056 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3057 | << std::dec << " hit new spanning blob " << *p << dendl; | |
3058 | request_reshard(p->blob_start(), p->blob_end()); | |
3059 | must_reshard = true; | |
3060 | } | |
31f18b77 FG |
3061 | if (!must_reshard) { |
3062 | denc_varint(0, bound); // blobid | |
3063 | denc_varint(0, bound); // logical_offset | |
3064 | denc_varint(0, bound); // len | |
3065 | denc_varint(0, bound); // blob_offset | |
7c673cae | 3066 | |
31f18b77 FG |
3067 | p->blob->bound_encode( |
3068 | bound, | |
3069 | struct_v, | |
3070 | p->blob->shared_blob->get_sbid(), | |
3071 | false); | |
3072 | } | |
7c673cae FG |
3073 | } |
3074 | if (must_reshard) { | |
3075 | return true; | |
3076 | } | |
3077 | ||
31f18b77 FG |
3078 | denc(struct_v, bound); |
3079 | denc_varint(0, bound); // number of extents | |
3080 | ||
7c673cae FG |
3081 | { |
3082 | auto app = bl.get_contiguous_appender(bound); | |
3083 | denc(struct_v, app); | |
3084 | denc_varint(n, app); | |
3085 | if (pn) { | |
3086 | *pn = n; | |
3087 | } | |
3088 | ||
3089 | n = 0; | |
3090 | uint64_t pos = 0; | |
3091 | uint64_t prev_len = 0; | |
3092 | for (auto p = start; | |
3093 | p != extent_map.end() && p->logical_offset < end; | |
3094 | ++p, ++n) { | |
3095 | unsigned blobid; | |
3096 | bool include_blob = false; | |
3097 | if (p->blob->is_spanning()) { | |
3098 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
3099 | blobid |= BLOBID_FLAG_SPANNING; | |
3100 | } else if (p->blob->last_encoded_id < 0) { | |
3101 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
3102 | include_blob = true; | |
3103 | blobid = 0; // the decoder will infer the id from n | |
3104 | } else { | |
3105 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
3106 | } | |
3107 | if (p->logical_offset == pos) { | |
3108 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
3109 | } | |
3110 | if (p->blob_offset == 0) { | |
3111 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
3112 | } | |
3113 | if (p->length == prev_len) { | |
3114 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
3115 | } else { | |
3116 | prev_len = p->length; | |
3117 | } | |
3118 | denc_varint(blobid, app); | |
3119 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3120 | denc_varint_lowz(p->logical_offset - pos, app); | |
3121 | } | |
3122 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3123 | denc_varint_lowz(p->blob_offset, app); | |
3124 | } | |
3125 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3126 | denc_varint_lowz(p->length, app); | |
3127 | } | |
3128 | pos = p->logical_end(); | |
3129 | if (include_blob) { | |
3130 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
3131 | } | |
3132 | } | |
3133 | } | |
3134 | /*derr << __func__ << bl << dendl; | |
3135 | derr << __func__ << ":"; | |
3136 | bl.hexdump(*_dout); | |
3137 | *_dout << dendl; | |
3138 | */ | |
3139 | return false; | |
3140 | } | |
3141 | ||
39ae355f TL |
3142 | /////////////////// BlueStore::ExtentMap::DecoderExtent /////////// |
3143 | void BlueStore::ExtentMap::ExtentDecoder::decode_extent( | |
3144 | Extent* le, | |
3145 | __u8 struct_v, | |
3146 | bptr_c_it_t& p, | |
3147 | Collection* c) | |
7c673cae | 3148 | { |
39ae355f TL |
3149 | uint64_t blobid; |
3150 | denc_varint(blobid, p); | |
3151 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3152 | uint64_t gap; | |
3153 | denc_varint_lowz(gap, p); | |
3154 | pos += gap; | |
3155 | } | |
3156 | le->logical_offset = pos; | |
3157 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3158 | denc_varint_lowz(le->blob_offset, p); | |
3159 | } else { | |
3160 | le->blob_offset = 0; | |
3161 | } | |
3162 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3163 | denc_varint_lowz(prev_len, p); | |
3164 | } | |
3165 | le->length = prev_len; | |
3166 | if (blobid & BLOBID_FLAG_SPANNING) { | |
3167 | consume_blobid(le, true, blobid >> BLOBID_SHIFT_BITS); | |
3168 | } else { | |
3169 | blobid >>= BLOBID_SHIFT_BITS; | |
3170 | if (blobid) { | |
3171 | consume_blobid(le, false, blobid - 1); | |
3172 | } else { | |
3173 | Blob *b = new Blob(); | |
3174 | uint64_t sbid = 0; | |
3175 | b->decode(p, struct_v, &sbid, false, c); | |
3176 | consume_blob(le, extent_pos, sbid, b); | |
3177 | } | |
3178 | } | |
3179 | pos += prev_len; | |
3180 | ++extent_pos; | |
3181 | } | |
3182 | ||
3183 | unsigned BlueStore::ExtentMap::ExtentDecoder::decode_some( | |
3184 | const bufferlist& bl, Collection* c) | |
3185 | { | |
3186 | __u8 struct_v; | |
3187 | uint32_t num; | |
7c673cae | 3188 | |
11fdf7f2 | 3189 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae | 3190 | auto p = bl.front().begin_deep(); |
7c673cae FG |
3191 | denc(struct_v, p); |
3192 | // Version 2 differs from v1 in blob's ref_map | |
3193 | // serialization only. Hence there is no specific | |
3194 | // handling at ExtentMap level below. | |
11fdf7f2 | 3195 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae | 3196 | denc_varint(num, p); |
7c673cae | 3197 | |
39ae355f | 3198 | extent_pos = 0; |
7c673cae | 3199 | while (!p.end()) { |
39ae355f TL |
3200 | Extent* le = get_next_extent(); |
3201 | decode_extent(le, struct_v, p, c); | |
3202 | add_extent(le); | |
3203 | } | |
3204 | ceph_assert(extent_pos == num); | |
3205 | return num; | |
3206 | } | |
7c673cae | 3207 | |
39ae355f TL |
3208 | void BlueStore::ExtentMap::ExtentDecoder::decode_spanning_blobs( |
3209 | bptr_c_it_t& p, Collection* c) | |
3210 | { | |
3211 | __u8 struct_v; | |
3212 | denc(struct_v, p); | |
3213 | // Version 2 differs from v1 in blob's ref_map | |
3214 | // serialization only. Hence there is no specific | |
3215 | // handling at ExtentMap level. | |
3216 | ceph_assert(struct_v == 1 || struct_v == 2); | |
3217 | ||
3218 | unsigned n; | |
3219 | denc_varint(n, p); | |
3220 | while (n--) { | |
3221 | BlueStore::BlobRef b(new Blob()); | |
3222 | denc_varint(b->id, p); | |
3223 | uint64_t sbid = 0; | |
3224 | b->decode(p, struct_v, &sbid, true, c); | |
3225 | consume_spanning_blob(sbid, b); | |
7c673cae | 3226 | } |
39ae355f | 3227 | } |
7c673cae | 3228 | |
39ae355f TL |
3229 | /////////////////// BlueStore::ExtentMap::DecoderExtentFull /////////// |
3230 | void BlueStore::ExtentMap::ExtentDecoderFull::consume_blobid( | |
3231 | BlueStore::Extent* le, bool spanning, uint64_t blobid) { | |
3232 | ceph_assert(le); | |
3233 | if (spanning) { | |
3234 | le->assign_blob(extent_map.get_spanning_blob(blobid)); | |
3235 | } else { | |
3236 | ceph_assert(blobid < blobs.size()); | |
3237 | le->assign_blob(blobs[blobid]); | |
3238 | // we build ref_map dynamically for non-spanning blobs | |
3239 | le->blob->get_ref( | |
3240 | extent_map.onode->c, | |
3241 | le->blob_offset, | |
3242 | le->length); | |
3243 | } | |
3244 | } | |
3245 | ||
3246 | void BlueStore::ExtentMap::ExtentDecoderFull::consume_blob( | |
3247 | BlueStore::Extent* le, uint64_t extent_no, uint64_t sbid, BlobRef b) { | |
3248 | ceph_assert(le); | |
3249 | blobs.resize(extent_no + 1); | |
3250 | blobs[extent_no] = b; | |
3251 | extent_map.onode->c->open_shared_blob(sbid, b); | |
3252 | le->assign_blob(b); | |
3253 | le->blob->get_ref( | |
3254 | extent_map.onode->c, | |
3255 | le->blob_offset, | |
3256 | le->length); | |
3257 | } | |
3258 | ||
3259 | void BlueStore::ExtentMap::ExtentDecoderFull::consume_spanning_blob( | |
3260 | uint64_t sbid, BlueStore::BlobRef b) { | |
3261 | extent_map.spanning_blob_map[b->id] = b; | |
3262 | extent_map.onode->c->open_shared_blob(sbid, b); | |
3263 | } | |
3264 | ||
3265 | BlueStore::Extent* BlueStore::ExtentMap::ExtentDecoderFull::get_next_extent() | |
3266 | { | |
3267 | return new Extent(); | |
3268 | } | |
3269 | ||
3270 | void BlueStore::ExtentMap::ExtentDecoderFull::add_extent(BlueStore::Extent* le) | |
3271 | { | |
3272 | extent_map.extent_map.insert(*le); | |
3273 | } | |
3274 | ||
3275 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
3276 | { | |
3277 | ExtentDecoderFull edecoder(*this); | |
3278 | unsigned n = edecoder.decode_some(bl, onode->c); | |
3279 | return n; | |
7c673cae FG |
3280 | } |
3281 | ||
3282 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
3283 | { | |
3284 | // Version 2 differs from v1 in blob's ref_map | |
3285 | // serialization only. Hence there is no specific | |
3286 | // handling at ExtentMap level. | |
3287 | __u8 struct_v = 2; | |
3288 | ||
3289 | denc(struct_v, p); | |
3290 | denc_varint((uint32_t)0, p); | |
3291 | size_t key_size = 0; | |
3292 | denc_varint((uint32_t)0, key_size); | |
3293 | p += spanning_blob_map.size() * key_size; | |
3294 | for (const auto& i : spanning_blob_map) { | |
3295 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3296 | } | |
3297 | } | |
3298 | ||
3299 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
3300 | bufferlist::contiguous_appender& p) | |
3301 | { | |
3302 | // Version 2 differs from v1 in blob's ref_map | |
3303 | // serialization only. Hence there is no specific | |
3304 | // handling at ExtentMap level. | |
3305 | __u8 struct_v = 2; | |
3306 | ||
3307 | denc(struct_v, p); | |
3308 | denc_varint(spanning_blob_map.size(), p); | |
3309 | for (auto& i : spanning_blob_map) { | |
3310 | denc_varint(i.second->id, p); | |
3311 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3312 | } | |
3313 | } | |
3314 | ||
7c673cae FG |
3315 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) |
3316 | { | |
3317 | shards.resize(onode->onode.extent_map_shards.size()); | |
3318 | unsigned i = 0; | |
3319 | for (auto &s : onode->onode.extent_map_shards) { | |
3320 | shards[i].shard_info = &s; | |
3321 | shards[i].loaded = loaded; | |
3322 | shards[i].dirty = dirty; | |
3323 | ++i; | |
3324 | } | |
3325 | } | |
3326 | ||
3327 | void BlueStore::ExtentMap::fault_range( | |
3328 | KeyValueDB *db, | |
3329 | uint32_t offset, | |
3330 | uint32_t length) | |
3331 | { | |
7c673cae FG |
3332 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3333 | << std::dec << dendl; | |
3334 | auto start = seek_shard(offset); | |
3335 | auto last = seek_shard(offset + length); | |
3336 | ||
3337 | if (start < 0) | |
3338 | return; | |
3339 | ||
11fdf7f2 | 3340 | ceph_assert(last >= start); |
7c673cae FG |
3341 | string key; |
3342 | while (start <= last) { | |
11fdf7f2 | 3343 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3344 | auto p = &shards[start]; |
3345 | if (!p->loaded) { | |
3346 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
3347 | << p->shard_info->offset << std::dec << dendl; | |
3348 | bufferlist v; | |
3349 | generate_extent_shard_key_and_apply( | |
3350 | onode->key, p->shard_info->offset, &key, | |
3351 | [&](const string& final_key) { | |
3352 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
3353 | if (r < 0) { | |
3354 | derr << __func__ << " missing shard 0x" << std::hex | |
3355 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
3356 | << dendl; | |
11fdf7f2 | 3357 | ceph_assert(r >= 0); |
7c673cae FG |
3358 | } |
3359 | } | |
3360 | ); | |
3361 | p->extents = decode_some(v); | |
3362 | p->loaded = true; | |
3363 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
3364 | << p->shard_info->offset |
3365 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 3366 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
3367 | ceph_assert(p->dirty == false); |
3368 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
3369 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
3370 | } else { | |
3371 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
3372 | } | |
3373 | ++start; | |
3374 | } | |
3375 | } | |
3376 | ||
3377 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
3378 | uint32_t offset, |
3379 | uint32_t length) | |
3380 | { | |
7c673cae FG |
3381 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3382 | << std::dec << dendl; | |
3383 | if (shards.empty()) { | |
3384 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
3385 | inline_bl.clear(); | |
3386 | return; | |
3387 | } | |
3388 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
3389 | if (length == 0) { |
3390 | length = 1; | |
3391 | } | |
3392 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
3393 | if (start < 0) |
3394 | return; | |
3395 | ||
11fdf7f2 | 3396 | ceph_assert(last >= start); |
7c673cae | 3397 | while (start <= last) { |
11fdf7f2 | 3398 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3399 | auto p = &shards[start]; |
3400 | if (!p->loaded) { | |
11fdf7f2 TL |
3401 | derr << __func__ << "on write 0x" << std::hex << offset |
3402 | << "~" << length << " shard 0x" << p->shard_info->offset | |
3403 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
3404 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
3405 | } |
3406 | if (!p->dirty) { | |
3407 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
3408 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
3409 | p->dirty = true; | |
3410 | } | |
3411 | ++start; | |
3412 | } | |
3413 | } | |
3414 | ||
3415 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
3416 | uint64_t offset) | |
3417 | { | |
3418 | Extent dummy(offset); | |
3419 | return extent_map.find(dummy); | |
3420 | } | |
3421 | ||
7c673cae FG |
3422 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
3423 | uint64_t offset) | |
3424 | { | |
3425 | Extent dummy(offset); | |
3426 | auto fp = extent_map.lower_bound(dummy); | |
3427 | if (fp != extent_map.begin()) { | |
3428 | --fp; | |
3429 | if (fp->logical_end() <= offset) { | |
3430 | ++fp; | |
3431 | } | |
3432 | } | |
3433 | return fp; | |
3434 | } | |
3435 | ||
3436 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
3437 | uint64_t offset) const | |
3438 | { | |
3439 | Extent dummy(offset); | |
3440 | auto fp = extent_map.lower_bound(dummy); | |
3441 | if (fp != extent_map.begin()) { | |
3442 | --fp; | |
3443 | if (fp->logical_end() <= offset) { | |
3444 | ++fp; | |
3445 | } | |
3446 | } | |
3447 | return fp; | |
3448 | } | |
3449 | ||
3450 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
3451 | { | |
3452 | auto fp = seek_lextent(offset); | |
3453 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
3454 | return false; | |
3455 | } | |
3456 | return true; | |
3457 | } | |
3458 | ||
3459 | int BlueStore::ExtentMap::compress_extent_map( | |
3460 | uint64_t offset, | |
3461 | uint64_t length) | |
3462 | { | |
7c673cae FG |
3463 | if (extent_map.empty()) |
3464 | return 0; | |
3465 | int removed = 0; | |
3466 | auto p = seek_lextent(offset); | |
3467 | if (p != extent_map.begin()) { | |
3468 | --p; // start to the left of offset | |
3469 | } | |
3470 | // the caller should have just written to this region | |
11fdf7f2 | 3471 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
3472 | |
3473 | // identify the *next* shard | |
3474 | auto pshard = shards.begin(); | |
3475 | while (pshard != shards.end() && | |
3476 | p->logical_offset >= pshard->shard_info->offset) { | |
3477 | ++pshard; | |
3478 | } | |
3479 | uint64_t shard_end; | |
3480 | if (pshard != shards.end()) { | |
3481 | shard_end = pshard->shard_info->offset; | |
3482 | } else { | |
3483 | shard_end = OBJECT_MAX_SIZE; | |
3484 | } | |
3485 | ||
3486 | auto n = p; | |
3487 | for (++n; n != extent_map.end(); p = n++) { | |
3488 | if (n->logical_offset > offset + length) { | |
3489 | break; // stop after end | |
3490 | } | |
3491 | while (n != extent_map.end() && | |
3492 | p->logical_end() == n->logical_offset && | |
3493 | p->blob == n->blob && | |
3494 | p->blob_offset + p->length == n->blob_offset && | |
3495 | n->logical_offset < shard_end) { | |
3496 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3497 | << " next shard 0x" << shard_end << std::dec | |
3498 | << " merging " << *p << " and " << *n << dendl; | |
3499 | p->length += n->length; | |
3500 | rm(n++); | |
3501 | ++removed; | |
3502 | } | |
3503 | if (n == extent_map.end()) { | |
3504 | break; | |
3505 | } | |
3506 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3507 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3508 | ++pshard; |
3509 | if (pshard != shards.end()) { | |
3510 | shard_end = pshard->shard_info->offset; | |
3511 | } else { | |
3512 | shard_end = OBJECT_MAX_SIZE; | |
3513 | } | |
3514 | } | |
3515 | } | |
11fdf7f2 | 3516 | if (removed) { |
7c673cae FG |
3517 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3518 | } | |
3519 | return removed; | |
3520 | } | |
3521 | ||
3522 | void BlueStore::ExtentMap::punch_hole( | |
3523 | CollectionRef &c, | |
3524 | uint64_t offset, | |
3525 | uint64_t length, | |
3526 | old_extent_map_t *old_extents) | |
3527 | { | |
3528 | auto p = seek_lextent(offset); | |
3529 | uint64_t end = offset + length; | |
3530 | while (p != extent_map.end()) { | |
3531 | if (p->logical_offset >= end) { | |
3532 | break; | |
3533 | } | |
3534 | if (p->logical_offset < offset) { | |
3535 | if (p->logical_end() > end) { | |
3536 | // split and deref middle | |
3537 | uint64_t front = offset - p->logical_offset; | |
3538 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3539 | length, p->blob); | |
3540 | old_extents->push_back(*oe); | |
3541 | add(end, | |
3542 | p->blob_offset + front + length, | |
3543 | p->length - front - length, | |
3544 | p->blob); | |
3545 | p->length = front; | |
3546 | break; | |
3547 | } else { | |
3548 | // deref tail | |
11fdf7f2 | 3549 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3550 | uint64_t keep = offset - p->logical_offset; |
3551 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3552 | p->length - keep, p->blob); | |
3553 | old_extents->push_back(*oe); | |
3554 | p->length = keep; | |
3555 | ++p; | |
3556 | continue; | |
3557 | } | |
3558 | } | |
3559 | if (p->logical_offset + p->length <= end) { | |
3560 | // deref whole lextent | |
3561 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3562 | p->length, p->blob); | |
3563 | old_extents->push_back(*oe); | |
3564 | rm(p++); | |
3565 | continue; | |
3566 | } | |
3567 | // deref head | |
3568 | uint64_t keep = p->logical_end() - end; | |
3569 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3570 | p->length - keep, p->blob); | |
3571 | old_extents->push_back(*oe); | |
3572 | ||
3573 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3574 | rm(p); | |
3575 | break; | |
3576 | } | |
3577 | } | |
3578 | ||
3579 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3580 | CollectionRef &c, | |
3581 | uint64_t logical_offset, | |
3582 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3583 | old_extent_map_t *old_extents) | |
3584 | { | |
3585 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3586 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3587 | |
3588 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3589 | // old_extents list if we overwre the blob totally | |
3590 | // This might happen during WAL overwrite. | |
3591 | b->get_ref(onode->c, blob_offset, length); | |
3592 | ||
3593 | if (old_extents) { | |
3594 | punch_hole(c, logical_offset, length, old_extents); | |
3595 | } | |
3596 | ||
3597 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3598 | extent_map.insert(*le); | |
3599 | if (spans_shard(logical_offset, length)) { | |
3600 | request_reshard(logical_offset, logical_offset + length); | |
3601 | } | |
3602 | return le; | |
3603 | } | |
3604 | ||
3605 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3606 | BlobRef lb, | |
3607 | uint32_t blob_offset, | |
3608 | uint32_t pos) | |
3609 | { | |
7c673cae FG |
3610 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; |
3611 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3612 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3613 | << dendl; | |
3614 | BlobRef rb = onode->c->new_blob(); | |
3615 | lb->split(onode->c, blob_offset, rb.get()); | |
3616 | ||
3617 | for (auto ep = seek_lextent(pos); | |
3618 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3619 | ++ep) { | |
3620 | if (ep->blob != lb) { | |
3621 | continue; | |
3622 | } | |
3623 | if (ep->logical_offset < pos) { | |
3624 | // split extent | |
3625 | size_t left = pos - ep->logical_offset; | |
3626 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3627 | extent_map.insert(*ne); | |
3628 | ep->length = left; | |
3629 | dout(30) << __func__ << " split " << *ep << dendl; | |
3630 | dout(30) << __func__ << " to " << *ne << dendl; | |
3631 | } else { | |
3632 | // switch blob | |
11fdf7f2 | 3633 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3634 | |
3635 | ep->blob = rb; | |
3636 | ep->blob_offset -= blob_offset; | |
3637 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3638 | } | |
3639 | } | |
3640 | return rb; | |
3641 | } | |
3642 | ||
3643 | // Onode | |
3644 | ||
3645 | #undef dout_prefix | |
3646 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3647 | ||
20effc67 TL |
3648 | const std::string& BlueStore::Onode::calc_omap_prefix(uint8_t flags) |
3649 | { | |
3650 | if (bluestore_onode_t::is_pgmeta_omap(flags)) { | |
3651 | return PREFIX_PGMETA_OMAP; | |
3652 | } | |
3653 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3654 | return PREFIX_PERPG_OMAP; | |
3655 | } | |
3656 | if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3657 | return PREFIX_PERPOOL_OMAP; | |
3658 | } | |
3659 | return PREFIX_OMAP; | |
3660 | } | |
3661 | ||
3662 | // '-' < '.' < '~' | |
3663 | void BlueStore::Onode::calc_omap_header( | |
3664 | uint8_t flags, | |
3665 | const Onode* o, | |
3666 | std::string* out) | |
3667 | { | |
3668 | if (!bluestore_onode_t::is_pgmeta_omap(flags)) { | |
3669 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3670 | _key_encode_u64(o->c->pool(), out); | |
3671 | _key_encode_u32(o->oid.hobj.get_bitwise_key_u32(), out); | |
3672 | } else if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3673 | _key_encode_u64(o->c->pool(), out); | |
3674 | } | |
3675 | } | |
3676 | _key_encode_u64(o->onode.nid, out); | |
3677 | out->push_back('-'); | |
3678 | } | |
3679 | ||
3680 | void BlueStore::Onode::calc_omap_key(uint8_t flags, | |
3681 | const Onode* o, | |
3682 | const std::string& key, | |
3683 | std::string* out) | |
3684 | { | |
3685 | if (!bluestore_onode_t::is_pgmeta_omap(flags)) { | |
3686 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3687 | _key_encode_u64(o->c->pool(), out); | |
3688 | _key_encode_u32(o->oid.hobj.get_bitwise_key_u32(), out); | |
3689 | } else if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3690 | _key_encode_u64(o->c->pool(), out); | |
3691 | } | |
3692 | } | |
3693 | _key_encode_u64(o->onode.nid, out); | |
3694 | out->push_back('.'); | |
3695 | out->append(key); | |
3696 | } | |
3697 | ||
3698 | void BlueStore::Onode::calc_omap_tail( | |
3699 | uint8_t flags, | |
3700 | const Onode* o, | |
3701 | std::string* out) | |
3702 | { | |
3703 | if (!bluestore_onode_t::is_pgmeta_omap(flags)) { | |
3704 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3705 | _key_encode_u64(o->c->pool(), out); | |
3706 | _key_encode_u32(o->oid.hobj.get_bitwise_key_u32(), out); | |
3707 | } else if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3708 | _key_encode_u64(o->c->pool(), out); | |
3709 | } | |
3710 | } | |
3711 | _key_encode_u64(o->onode.nid, out); | |
3712 | out->push_back('~'); | |
3713 | } | |
3714 | ||
39ae355f TL |
3715 | void BlueStore::Onode::get() |
3716 | { | |
3717 | ++nref; | |
3718 | ++pin_nref; | |
f6b5b4d7 | 3719 | } |
39ae355f TL |
3720 | void BlueStore::Onode::put() |
3721 | { | |
3722 | if (--pin_nref == 1) { | |
3723 | c->get_onode_cache()->maybe_unpin(this); | |
f6b5b4d7 | 3724 | } |
39ae355f | 3725 | if (--nref == 0) { |
f6b5b4d7 TL |
3726 | delete this; |
3727 | } | |
3728 | } | |
3729 | ||
39ae355f TL |
3730 | void BlueStore::Onode::decode_raw( |
3731 | BlueStore::Onode* on, | |
3732 | const bufferlist& v, | |
3733 | BlueStore::ExtentMap::ExtentDecoder& edecoder) | |
eafe8130 | 3734 | { |
eafe8130 TL |
3735 | on->exists = true; |
3736 | auto p = v.front().begin_deep(); | |
3737 | on->onode.decode(p); | |
eafe8130 TL |
3738 | |
3739 | // initialize extent_map | |
39ae355f | 3740 | edecoder.decode_spanning_blobs(p, on->c); |
eafe8130 TL |
3741 | if (on->onode.extent_map_shards.empty()) { |
3742 | denc(on->extent_map.inline_bl, p); | |
39ae355f | 3743 | edecoder.decode_some(on->extent_map.inline_bl, on->c); |
eafe8130 | 3744 | } |
39ae355f TL |
3745 | } |
3746 | ||
3747 | BlueStore::Onode* BlueStore::Onode::create_decode( | |
3748 | CollectionRef c, | |
3749 | const ghobject_t& oid, | |
3750 | const string& key, | |
3751 | const bufferlist& v, | |
3752 | bool allow_empty) | |
3753 | { | |
3754 | ceph_assert(v.length() || allow_empty); | |
3755 | Onode* on = new Onode(c.get(), oid, key); | |
3756 | ||
3757 | if (v.length()) { | |
3758 | ExtentMap::ExtentDecoderFull edecoder(on->extent_map); | |
3759 | decode_raw(on, v, edecoder); | |
3760 | ||
3761 | for (auto& i : on->onode.attrs) { | |
3762 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); | |
3763 | } | |
3764 | ||
3765 | // initialize extent_map | |
3766 | if (on->onode.extent_map_shards.empty()) { | |
3767 | on->extent_map.inline_bl.reassign_to_mempool( | |
3768 | mempool::mempool_bluestore_cache_data); | |
3769 | } else { | |
3770 | on->extent_map.init_shards(false, false); | |
3771 | } | |
eafe8130 TL |
3772 | } |
3773 | return on; | |
3774 | } | |
3775 | ||
7c673cae FG |
3776 | void BlueStore::Onode::flush() |
3777 | { | |
3778 | if (flushing_count.load()) { | |
3779 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
9f95a23c | 3780 | waiting_count++; |
11fdf7f2 | 3781 | std::unique_lock l(flush_lock); |
7c673cae FG |
3782 | while (flushing_count.load()) { |
3783 | flush_cond.wait(l); | |
3784 | } | |
9f95a23c | 3785 | waiting_count--; |
7c673cae FG |
3786 | } |
3787 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3788 | } | |
3789 | ||
9f95a23c TL |
3790 | void BlueStore::Onode::dump(Formatter* f) const |
3791 | { | |
3792 | onode.dump(f); | |
3793 | extent_map.dump(f); | |
3794 | } | |
3795 | ||
9f95a23c TL |
3796 | void BlueStore::Onode::rewrite_omap_key(const string& old, string *out) |
3797 | { | |
f67539c2 TL |
3798 | if (!onode.is_pgmeta_omap()) { |
3799 | if (onode.is_perpg_omap()) { | |
3800 | _key_encode_u64(c->pool(), out); | |
3801 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), out); | |
3802 | } else if (onode.is_perpool_omap()) { | |
3803 | _key_encode_u64(c->pool(), out); | |
3804 | } | |
9f95a23c TL |
3805 | } |
3806 | _key_encode_u64(onode.nid, out); | |
3807 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
3808 | } | |
3809 | ||
9f95a23c TL |
3810 | void BlueStore::Onode::decode_omap_key(const string& key, string *user_key) |
3811 | { | |
f67539c2 TL |
3812 | size_t pos = sizeof(uint64_t) + 1; |
3813 | if (!onode.is_pgmeta_omap()) { | |
3814 | if (onode.is_perpg_omap()) { | |
3815 | pos += sizeof(uint64_t) + sizeof(uint32_t); | |
3816 | } else if (onode.is_perpool_omap()) { | |
3817 | pos += sizeof(uint64_t); | |
3818 | } | |
9f95a23c | 3819 | } |
f67539c2 | 3820 | *user_key = key.substr(pos); |
9f95a23c TL |
3821 | } |
3822 | ||
7c673cae FG |
3823 | // ======================================================= |
3824 | // WriteContext | |
3825 | ||
3826 | /// Checks for writes to the same pextent within a blob | |
3827 | bool BlueStore::WriteContext::has_conflict( | |
3828 | BlobRef b, | |
3829 | uint64_t loffs, | |
3830 | uint64_t loffs_end, | |
3831 | uint64_t min_alloc_size) | |
3832 | { | |
11fdf7f2 TL |
3833 | ceph_assert((loffs % min_alloc_size) == 0); |
3834 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3835 | for (auto w : writes) { |
3836 | if (b == w.b) { | |
11fdf7f2 TL |
3837 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3838 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3839 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3840 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3841 | return true; |
3842 | } | |
3843 | } | |
3844 | } | |
3845 | return false; | |
3846 | } | |
3847 | ||
3848 | // ======================================================= | |
3849 | ||
3850 | // DeferredBatch | |
3851 | #undef dout_prefix | |
3852 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
9f95a23c TL |
3853 | #undef dout_context |
3854 | #define dout_context cct | |
7c673cae FG |
3855 | |
3856 | void BlueStore::DeferredBatch::prepare_write( | |
3857 | CephContext *cct, | |
3858 | uint64_t seq, uint64_t offset, uint64_t length, | |
3859 | bufferlist::const_iterator& blp) | |
3860 | { | |
3861 | _discard(cct, offset, length); | |
3862 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3863 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3864 | i.first->second.seq = seq; |
3865 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3866 | i.first->second.bl.reassign_to_mempool( |
3867 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3868 | dout(20) << __func__ << " seq " << seq |
3869 | << " 0x" << std::hex << offset << "~" << length | |
3870 | << " crc " << i.first->second.bl.crc32c(-1) | |
3871 | << std::dec << dendl; | |
3872 | seq_bytes[seq] += length; | |
3873 | #ifdef DEBUG_DEFERRED | |
3874 | _audit(cct); | |
3875 | #endif | |
3876 | } | |
3877 | ||
3878 | void BlueStore::DeferredBatch::_discard( | |
3879 | CephContext *cct, uint64_t offset, uint64_t length) | |
3880 | { | |
3881 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3882 | << std::dec << dendl; | |
3883 | auto p = iomap.lower_bound(offset); | |
3884 | if (p != iomap.begin()) { | |
3885 | --p; | |
3886 | auto end = p->first + p->second.bl.length(); | |
3887 | if (end > offset) { | |
3888 | bufferlist head; | |
3889 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3890 | dout(20) << __func__ << " keep head " << p->second.seq | |
3891 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3892 | << " -> 0x" << head.length() << std::dec << dendl; | |
3893 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3894 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3895 | if (end > offset + length) { |
3896 | bufferlist tail; | |
3897 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3898 | end - (offset + length)); | |
3899 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3900 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3901 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3902 | auto &n = iomap[offset + length]; | |
3903 | n.bl.swap(tail); | |
3904 | n.seq = p->second.seq; | |
3905 | i->second -= length; | |
3906 | } else { | |
3907 | i->second -= end - offset; | |
3908 | } | |
11fdf7f2 | 3909 | ceph_assert(i->second >= 0); |
7c673cae FG |
3910 | p->second.bl.swap(head); |
3911 | } | |
3912 | ++p; | |
3913 | } | |
3914 | while (p != iomap.end()) { | |
3915 | if (p->first >= offset + length) { | |
3916 | break; | |
3917 | } | |
3918 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3919 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3920 | auto end = p->first + p->second.bl.length(); |
3921 | if (end > offset + length) { | |
3922 | unsigned drop_front = offset + length - p->first; | |
3923 | unsigned keep_tail = end - (offset + length); | |
3924 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3925 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3926 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3927 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3928 | << std::dec << dendl; | |
3929 | auto &s = iomap[offset + length]; | |
3930 | s.seq = p->second.seq; | |
3931 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3932 | i->second -= drop_front; | |
3933 | } else { | |
3934 | dout(20) << __func__ << " drop " << p->second.seq | |
3935 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3936 | << std::dec << dendl; | |
3937 | i->second -= p->second.bl.length(); | |
3938 | } | |
11fdf7f2 | 3939 | ceph_assert(i->second >= 0); |
7c673cae FG |
3940 | p = iomap.erase(p); |
3941 | } | |
3942 | } | |
3943 | ||
3944 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3945 | { | |
3946 | map<uint64_t,int> sb; | |
3947 | for (auto p : seq_bytes) { | |
3948 | sb[p.first] = 0; // make sure we have the same set of keys | |
3949 | } | |
3950 | uint64_t pos = 0; | |
3951 | for (auto& p : iomap) { | |
11fdf7f2 | 3952 | ceph_assert(p.first >= pos); |
7c673cae FG |
3953 | sb[p.second.seq] += p.second.bl.length(); |
3954 | pos = p.first + p.second.bl.length(); | |
3955 | } | |
11fdf7f2 | 3956 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3957 | } |
3958 | ||
3959 | ||
3960 | // Collection | |
3961 | ||
3962 | #undef dout_prefix | |
3963 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3964 | ||
9f95a23c TL |
3965 | BlueStore::Collection::Collection(BlueStore *store_, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t cid) |
3966 | : CollectionImpl(store_->cct, cid), | |
11fdf7f2 | 3967 | store(store_), |
9f95a23c | 3968 | cache(bc), |
7c673cae | 3969 | exists(true), |
39ae355f | 3970 | onode_space(oc), |
11fdf7f2 TL |
3971 | commit_queue(nullptr) |
3972 | { | |
3973 | } | |
3974 | ||
3975 | bool BlueStore::Collection::flush_commit(Context *c) | |
3976 | { | |
3977 | return osr->flush_commit(c); | |
3978 | } | |
3979 | ||
3980 | void BlueStore::Collection::flush() | |
3981 | { | |
3982 | osr->flush(); | |
3983 | } | |
3984 | ||
3985 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3986 | { |
11fdf7f2 | 3987 | osr->flush_all_but_last(); |
7c673cae FG |
3988 | } |
3989 | ||
3990 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3991 | { | |
11fdf7f2 | 3992 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3993 | const bluestore_blob_t& blob = b->get_blob(); |
3994 | if (!blob.is_shared()) { | |
3995 | b->shared_blob = new SharedBlob(this); | |
3996 | return; | |
3997 | } | |
3998 | ||
3999 | b->shared_blob = shared_blob_set.lookup(sbid); | |
4000 | if (b->shared_blob) { | |
4001 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
4002 | << std::dec << " had " << *b->shared_blob << dendl; | |
4003 | } else { | |
4004 | b->shared_blob = new SharedBlob(sbid, this); | |
4005 | shared_blob_set.add(this, b->shared_blob.get()); | |
4006 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
4007 | << std::dec << " opened " << *b->shared_blob | |
4008 | << dendl; | |
4009 | } | |
4010 | } | |
4011 | ||
4012 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
4013 | { | |
4014 | if (!sb->is_loaded()) { | |
4015 | ||
4016 | bufferlist v; | |
4017 | string key; | |
4018 | auto sbid = sb->get_sbid(); | |
4019 | get_shared_blob_key(sbid, &key); | |
4020 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
4021 | if (r < 0) { | |
4022 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
4023 | << std::dec << " not found at key " | |
4024 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 4025 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
4026 | } |
4027 | ||
4028 | sb->loaded = true; | |
4029 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
4030 | auto p = v.cbegin(); |
4031 | decode(*(sb->persistent), p); | |
7c673cae FG |
4032 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
4033 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
4034 | } | |
4035 | } | |
4036 | ||
4037 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
4038 | { | |
7c673cae | 4039 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 4040 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
4041 | |
4042 | // update blob | |
31f18b77 | 4043 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 4044 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
4045 | |
4046 | // update shared blob | |
4047 | b->shared_blob->loaded = true; | |
4048 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
4049 | shared_blob_set.add(this, b->shared_blob.get()); | |
4050 | for (auto p : blob.get_extents()) { | |
4051 | if (p.is_valid()) { | |
4052 | b->shared_blob->get_ref( | |
4053 | p.offset, | |
4054 | p.length); | |
4055 | } | |
4056 | } | |
4057 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
4058 | } | |
4059 | ||
31f18b77 FG |
4060 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
4061 | { | |
4062 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 4063 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
4064 | |
4065 | uint64_t sbid = sb->get_sbid(); | |
4066 | shared_blob_set.remove(sb); | |
4067 | sb->loaded = false; | |
4068 | delete sb->persistent; | |
4069 | sb->sbid_unloaded = 0; | |
4070 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
4071 | return sbid; | |
4072 | } | |
4073 | ||
7c673cae FG |
4074 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
4075 | const ghobject_t& oid, | |
9f95a23c TL |
4076 | bool create, |
4077 | bool is_createop) | |
7c673cae | 4078 | { |
9f95a23c | 4079 | ceph_assert(create ? ceph_mutex_is_wlocked(lock) : ceph_mutex_is_locked(lock)); |
7c673cae FG |
4080 | |
4081 | spg_t pgid; | |
4082 | if (cid.is_pg(&pgid)) { | |
4083 | if (!oid.match(cnode.bits, pgid.ps())) { | |
4084 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
4085 | << pgid << " bits " << cnode.bits << dendl; | |
4086 | ceph_abort(); | |
4087 | } | |
4088 | } | |
4089 | ||
39ae355f | 4090 | OnodeRef o = onode_space.lookup(oid); |
7c673cae FG |
4091 | if (o) |
4092 | return o; | |
4093 | ||
eafe8130 | 4094 | string key; |
7c673cae FG |
4095 | get_object_key(store->cct, oid, &key); |
4096 | ||
4097 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
4098 | << pretty_binary_string(key) << dendl; | |
4099 | ||
4100 | bufferlist v; | |
9f95a23c | 4101 | int r = -ENOENT; |
7c673cae | 4102 | Onode *on; |
9f95a23c TL |
4103 | if (!is_createop) { |
4104 | r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
4105 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
4106 | } | |
7c673cae | 4107 | if (v.length() == 0) { |
11fdf7f2 | 4108 | ceph_assert(r == -ENOENT); |
f67539c2 | 4109 | if (!create) |
7c673cae | 4110 | return OnodeRef(); |
7c673cae | 4111 | } else { |
11fdf7f2 | 4112 | ceph_assert(r >= 0); |
7c673cae | 4113 | } |
39ae355f TL |
4114 | |
4115 | // new object, load onode if available | |
4116 | on = Onode::create_decode(this, oid, key, v, true); | |
7c673cae | 4117 | o.reset(on); |
39ae355f | 4118 | return onode_space.add_onode(oid, o); |
7c673cae FG |
4119 | } |
4120 | ||
4121 | void BlueStore::Collection::split_cache( | |
4122 | Collection *dest) | |
4123 | { | |
4124 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
4125 | ||
f67539c2 TL |
4126 | auto *ocache = get_onode_cache(); |
4127 | auto *ocache_dest = dest->get_onode_cache(); | |
4128 | ||
4129 | // lock cache shards | |
4130 | std::lock(ocache->lock, ocache_dest->lock, cache->lock, dest->cache->lock); | |
4131 | std::lock_guard l(ocache->lock, std::adopt_lock); | |
4132 | std::lock_guard l2(ocache_dest->lock, std::adopt_lock); | |
4133 | std::lock_guard l3(cache->lock, std::adopt_lock); | |
4134 | std::lock_guard l4(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
4135 | |
4136 | int destbits = dest->cnode.bits; | |
4137 | spg_t destpg; | |
4138 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 4139 | ceph_assert(is_pg); |
7c673cae | 4140 | |
39ae355f TL |
4141 | auto p = onode_space.onode_map.begin(); |
4142 | while (p != onode_space.onode_map.end()) { | |
11fdf7f2 | 4143 | OnodeRef o = p->second; |
7c673cae FG |
4144 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
4145 | // onode does not belong to this child | |
11fdf7f2 TL |
4146 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
4147 | << dendl; | |
7c673cae FG |
4148 | ++p; |
4149 | } else { | |
7c673cae FG |
4150 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
4151 | << dendl; | |
4152 | ||
39ae355f | 4153 | // ensuring that nref is always >= 2 and hence onode is pinned |
f6b5b4d7 | 4154 | OnodeRef o_pin = o; |
f6b5b4d7 | 4155 | |
39ae355f TL |
4156 | p = onode_space.onode_map.erase(p); |
4157 | dest->onode_space.onode_map[o->oid] = o; | |
adb31ebb | 4158 | if (o->cached) { |
39ae355f | 4159 | get_onode_cache()->_move_pinned(dest->get_onode_cache(), o.get()); |
9f95a23c | 4160 | } |
f6b5b4d7 | 4161 | o->c = dest; |
7c673cae FG |
4162 | |
4163 | // move over shared blobs and buffers. cover shared blobs from | |
4164 | // both extent map and spanning blob map (the full extent map | |
4165 | // may not be faulted in) | |
4166 | vector<SharedBlob*> sbvec; | |
4167 | for (auto& e : o->extent_map.extent_map) { | |
4168 | sbvec.push_back(e.blob->shared_blob.get()); | |
4169 | } | |
4170 | for (auto& b : o->extent_map.spanning_blob_map) { | |
4171 | sbvec.push_back(b.second->shared_blob.get()); | |
4172 | } | |
4173 | for (auto sb : sbvec) { | |
4174 | if (sb->coll == dest) { | |
4175 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
4176 | << dendl; | |
4177 | continue; | |
4178 | } | |
4179 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
4180 | if (sb->get_sbid()) { |
4181 | ldout(store->cct, 20) << __func__ | |
4182 | << " moving registration " << *sb << dendl; | |
4183 | shared_blob_set.remove(sb); | |
4184 | dest->shared_blob_set.add(dest, sb); | |
4185 | } | |
3efd9988 | 4186 | sb->coll = dest; |
7c673cae | 4187 | if (dest->cache != cache) { |
7c673cae FG |
4188 | for (auto& i : sb->bc.buffer_map) { |
4189 | if (!i.second->is_writing()) { | |
4190 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
4191 | << dendl; | |
9f95a23c | 4192 | dest->cache->_move(cache, i.second.get()); |
7c673cae FG |
4193 | } |
4194 | } | |
4195 | } | |
4196 | } | |
7c673cae FG |
4197 | } |
4198 | } | |
9f95a23c | 4199 | dest->cache->_trim(); |
7c673cae FG |
4200 | } |
4201 | ||
7c673cae FG |
4202 | // ======================================================= |
4203 | ||
91327a77 AA |
4204 | // MempoolThread |
4205 | ||
4206 | #undef dout_prefix | |
4207 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
9f95a23c TL |
4208 | #undef dout_context |
4209 | #define dout_context store->cct | |
91327a77 | 4210 | |
7c673cae FG |
4211 | void *BlueStore::MempoolThread::entry() |
4212 | { | |
9f95a23c | 4213 | std::unique_lock l{lock}; |
11fdf7f2 | 4214 | |
92f5a8d4 | 4215 | uint32_t prev_config_change = store->config_changed.load(); |
eafe8130 TL |
4216 | uint64_t base = store->osd_memory_base; |
4217 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4218 | uint64_t target = store->osd_memory_target; | |
4219 | uint64_t min = store->osd_memory_cache_min; | |
4220 | uint64_t max = min; | |
4221 | ||
4222 | // When setting the maximum amount of memory to use for cache, first | |
4223 | // assume some base amount of memory for the OSD and then fudge in | |
4224 | // some overhead for fragmentation that scales with cache usage. | |
4225 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4226 | if (ltarget > base + min) { | |
4227 | max = ltarget - base; | |
11fdf7f2 | 4228 | } |
31f18b77 | 4229 | |
eafe8130 | 4230 | binned_kv_cache = store->db->get_priority_cache(); |
f67539c2 | 4231 | binned_kv_onode_cache = store->db->get_priority_cache(PREFIX_OBJ); |
eafe8130 TL |
4232 | if (store->cache_autotune && binned_kv_cache != nullptr) { |
4233 | pcm = std::make_shared<PriorityCache::Manager>( | |
f67539c2 | 4234 | store->cct, min, max, target, true, "bluestore-pricache"); |
eafe8130 TL |
4235 | pcm->insert("kv", binned_kv_cache, true); |
4236 | pcm->insert("meta", meta_cache, true); | |
4237 | pcm->insert("data", data_cache, true); | |
f67539c2 TL |
4238 | if (binned_kv_onode_cache != nullptr) { |
4239 | pcm->insert("kv_onode", binned_kv_onode_cache, true); | |
4240 | } | |
eafe8130 | 4241 | } |
91327a77 AA |
4242 | |
4243 | utime_t next_balance = ceph_clock_now(); | |
4244 | utime_t next_resize = ceph_clock_now(); | |
20effc67 | 4245 | utime_t next_bin_rotation = ceph_clock_now(); |
9f95a23c TL |
4246 | utime_t next_deferred_force_submit = ceph_clock_now(); |
4247 | utime_t alloc_stats_dump_clock = ceph_clock_now(); | |
31f18b77 | 4248 | |
91327a77 | 4249 | bool interval_stats_trim = false; |
91327a77 | 4250 | while (!stop) { |
92f5a8d4 TL |
4251 | // Update pcm cache settings if related configuration was changed |
4252 | uint32_t cur_config_change = store->config_changed.load(); | |
4253 | if (cur_config_change != prev_config_change) { | |
4254 | _update_cache_settings(); | |
4255 | prev_config_change = cur_config_change; | |
4256 | } | |
4257 | ||
20effc67 TL |
4258 | // define various intervals for background work |
4259 | double age_bin_interval = store->cache_age_bin_interval; | |
91327a77 AA |
4260 | double autotune_interval = store->cache_autotune_interval; |
4261 | double resize_interval = store->osd_memory_cache_resize_interval; | |
9f95a23c | 4262 | double max_defer_interval = store->max_defer_interval; |
9f95a23c TL |
4263 | double alloc_stats_dump_interval = |
4264 | store->cct->_conf->bluestore_alloc_stats_dump_interval; | |
91327a77 | 4265 | |
20effc67 | 4266 | // alloc stats dump |
9f95a23c TL |
4267 | if (alloc_stats_dump_interval > 0 && |
4268 | alloc_stats_dump_clock + alloc_stats_dump_interval < ceph_clock_now()) { | |
4269 | store->_record_allocation_stats(); | |
4270 | alloc_stats_dump_clock = ceph_clock_now(); | |
4271 | } | |
20effc67 TL |
4272 | // cache age binning |
4273 | if (age_bin_interval > 0 && next_bin_rotation < ceph_clock_now()) { | |
4274 | if (binned_kv_cache != nullptr) { | |
4275 | binned_kv_cache->import_bins(store->kv_bins); | |
4276 | } | |
4277 | if (binned_kv_onode_cache != nullptr) { | |
4278 | binned_kv_onode_cache->import_bins(store->kv_onode_bins); | |
4279 | } | |
4280 | meta_cache->import_bins(store->meta_bins); | |
4281 | data_cache->import_bins(store->data_bins); | |
4282 | ||
4283 | if (pcm != nullptr) { | |
4284 | pcm->shift_bins(); | |
4285 | } | |
4286 | next_bin_rotation = ceph_clock_now(); | |
4287 | next_bin_rotation += age_bin_interval; | |
4288 | } | |
4289 | // cache balancing | |
91327a77 | 4290 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { |
20effc67 TL |
4291 | if (binned_kv_cache != nullptr) { |
4292 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
4293 | } | |
4294 | if (binned_kv_onode_cache != nullptr) { | |
4295 | binned_kv_onode_cache->set_cache_ratio(store->cache_kv_onode_ratio); | |
4296 | } | |
4297 | meta_cache->set_cache_ratio(store->cache_meta_ratio); | |
4298 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
11fdf7f2 | 4299 | |
91327a77 | 4300 | // Log events at 5 instead of 20 when balance happens. |
91327a77 | 4301 | interval_stats_trim = true; |
eafe8130 TL |
4302 | |
4303 | if (pcm != nullptr) { | |
4304 | pcm->balance(); | |
91327a77 | 4305 | } |
31f18b77 | 4306 | |
91327a77 AA |
4307 | next_balance = ceph_clock_now(); |
4308 | next_balance += autotune_interval; | |
4309 | } | |
20effc67 | 4310 | // memory resizing (ie autotuning) |
91327a77 | 4311 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { |
eafe8130 TL |
4312 | if (ceph_using_tcmalloc() && pcm != nullptr) { |
4313 | pcm->tune_memory(); | |
91327a77 AA |
4314 | } |
4315 | next_resize = ceph_clock_now(); | |
4316 | next_resize += resize_interval; | |
31f18b77 | 4317 | } |
20effc67 | 4318 | // deferred force submit |
9f95a23c TL |
4319 | if (max_defer_interval > 0 && |
4320 | next_deferred_force_submit < ceph_clock_now()) { | |
4321 | if (store->get_deferred_last_submitted() + max_defer_interval < | |
4322 | ceph_clock_now()) { | |
4323 | store->deferred_try_submit(); | |
4324 | } | |
4325 | next_deferred_force_submit = ceph_clock_now(); | |
4326 | next_deferred_force_submit += max_defer_interval/3; | |
4327 | } | |
4328 | ||
4329 | // Now Resize the shards | |
4330 | _resize_shards(interval_stats_trim); | |
91327a77 | 4331 | interval_stats_trim = false; |
31f18b77 | 4332 | |
1e59de90 | 4333 | store->_update_logger(); |
11fdf7f2 TL |
4334 | auto wait = ceph::make_timespan( |
4335 | store->cct->_conf->bluestore_cache_trim_interval); | |
4336 | cond.wait_for(l, wait); | |
7c673cae | 4337 | } |
9f95a23c TL |
4338 | // do final dump |
4339 | store->_record_allocation_stats(); | |
7c673cae | 4340 | stop = false; |
f67539c2 | 4341 | pcm = nullptr; |
7c673cae FG |
4342 | return NULL; |
4343 | } | |
4344 | ||
9f95a23c | 4345 | void BlueStore::MempoolThread::_resize_shards(bool interval_stats) |
91327a77 | 4346 | { |
9f95a23c TL |
4347 | size_t onode_shards = store->onode_cache_shards.size(); |
4348 | size_t buffer_shards = store->buffer_cache_shards.size(); | |
91327a77 | 4349 | int64_t kv_used = store->db->get_cache_usage(); |
f67539c2 | 4350 | int64_t kv_onode_used = store->db->get_cache_usage(PREFIX_OBJ); |
11fdf7f2 TL |
4351 | int64_t meta_used = meta_cache->_get_used_bytes(); |
4352 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
4353 | |
4354 | uint64_t cache_size = store->cache_size; | |
4355 | int64_t kv_alloc = | |
20effc67 | 4356 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
f67539c2 TL |
4357 | int64_t kv_onode_alloc = |
4358 | static_cast<int64_t>(store->cache_kv_onode_ratio * cache_size); | |
91327a77 | 4359 | int64_t meta_alloc = |
11fdf7f2 | 4360 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 4361 | int64_t data_alloc = |
11fdf7f2 | 4362 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 4363 | |
eafe8130 TL |
4364 | if (pcm != nullptr && binned_kv_cache != nullptr) { |
4365 | cache_size = pcm->get_tuned_mem(); | |
11fdf7f2 TL |
4366 | kv_alloc = binned_kv_cache->get_committed_size(); |
4367 | meta_alloc = meta_cache->get_committed_size(); | |
4368 | data_alloc = data_cache->get_committed_size(); | |
f67539c2 TL |
4369 | if (binned_kv_onode_cache != nullptr) { |
4370 | kv_onode_alloc = binned_kv_onode_cache->get_committed_size(); | |
4371 | } | |
91327a77 AA |
4372 | } |
4373 | ||
4374 | if (interval_stats) { | |
9f95a23c | 4375 | dout(5) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4376 | << " kv_alloc: " << kv_alloc |
4377 | << " kv_used: " << kv_used | |
f67539c2 TL |
4378 | << " kv_onode_alloc: " << kv_onode_alloc |
4379 | << " kv_onode_used: " << kv_onode_used | |
91327a77 AA |
4380 | << " meta_alloc: " << meta_alloc |
4381 | << " meta_used: " << meta_used | |
4382 | << " data_alloc: " << data_alloc | |
4383 | << " data_used: " << data_used << dendl; | |
4384 | } else { | |
9f95a23c | 4385 | dout(20) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4386 | << " kv_alloc: " << kv_alloc |
4387 | << " kv_used: " << kv_used | |
f67539c2 TL |
4388 | << " kv_onode_alloc: " << kv_onode_alloc |
4389 | << " kv_onode_used: " << kv_onode_used | |
91327a77 AA |
4390 | << " meta_alloc: " << meta_alloc |
4391 | << " meta_used: " << meta_used | |
4392 | << " data_alloc: " << data_alloc | |
4393 | << " data_used: " << data_used << dendl; | |
4394 | } | |
4395 | ||
4396 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
9f95a23c TL |
4397 | (meta_alloc / (double) onode_shards) / meta_cache->get_bytes_per_onode()); |
4398 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / buffer_shards); | |
91327a77 | 4399 | |
9f95a23c | 4400 | dout(30) << __func__ << " max_shard_onodes: " << max_shard_onodes |
91327a77 AA |
4401 | << " max_shard_buffer: " << max_shard_buffer << dendl; |
4402 | ||
9f95a23c TL |
4403 | for (auto i : store->onode_cache_shards) { |
4404 | i->set_max(max_shard_onodes); | |
4405 | } | |
4406 | for (auto i : store->buffer_cache_shards) { | |
4407 | i->set_max(max_shard_buffer); | |
91327a77 AA |
4408 | } |
4409 | } | |
4410 | ||
92f5a8d4 TL |
4411 | void BlueStore::MempoolThread::_update_cache_settings() |
4412 | { | |
4413 | // Nothing to do if pcm is not used. | |
4414 | if (pcm == nullptr) { | |
4415 | return; | |
4416 | } | |
4417 | ||
92f5a8d4 TL |
4418 | uint64_t target = store->osd_memory_target; |
4419 | uint64_t base = store->osd_memory_base; | |
4420 | uint64_t min = store->osd_memory_cache_min; | |
4421 | uint64_t max = min; | |
4422 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4423 | ||
4424 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4425 | if (ltarget > base + min) { | |
4426 | max = ltarget - base; | |
4427 | } | |
4428 | ||
4429 | // set pcm cache levels | |
4430 | pcm->set_target_memory(target); | |
4431 | pcm->set_min_memory(min); | |
4432 | pcm->set_max_memory(max); | |
4433 | ||
9f95a23c | 4434 | dout(5) << __func__ << " updated pcm target: " << target |
92f5a8d4 TL |
4435 | << " pcm min: " << min |
4436 | << " pcm max: " << max | |
4437 | << dendl; | |
4438 | } | |
4439 | ||
7c673cae FG |
4440 | // ======================================================= |
4441 | ||
31f18b77 FG |
4442 | // OmapIteratorImpl |
4443 | ||
4444 | #undef dout_prefix | |
4445 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
4446 | ||
4447 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
1e59de90 TL |
4448 | PerfCounters* _logger, CollectionRef c, OnodeRef& o, KeyValueDB::Iterator it) |
4449 | : logger(_logger), c(c), o(o), it(it) | |
31f18b77 | 4450 | { |
1e59de90 | 4451 | logger->inc(l_bluestore_omap_iterator_count); |
9f95a23c | 4452 | std::shared_lock l(c->lock); |
31f18b77 | 4453 | if (o->onode.has_omap()) { |
9f95a23c TL |
4454 | o->get_omap_key(string(), &head); |
4455 | o->get_omap_tail(&tail); | |
31f18b77 FG |
4456 | it->lower_bound(head); |
4457 | } | |
4458 | } | |
1e59de90 TL |
4459 | BlueStore::OmapIteratorImpl::~OmapIteratorImpl() |
4460 | { | |
4461 | logger->dec(l_bluestore_omap_iterator_count); | |
4462 | } | |
31f18b77 | 4463 | |
11fdf7f2 TL |
4464 | string BlueStore::OmapIteratorImpl::_stringify() const |
4465 | { | |
4466 | stringstream s; | |
4467 | s << " omap_iterator(cid = " << c->cid | |
4468 | <<", oid = " << o->oid << ")"; | |
4469 | return s.str(); | |
4470 | } | |
4471 | ||
31f18b77 FG |
4472 | int BlueStore::OmapIteratorImpl::seek_to_first() |
4473 | { | |
9f95a23c | 4474 | std::shared_lock l(c->lock); |
11fdf7f2 | 4475 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4476 | if (o->onode.has_omap()) { |
4477 | it->lower_bound(head); | |
4478 | } else { | |
4479 | it = KeyValueDB::Iterator(); | |
4480 | } | |
494da23a TL |
4481 | c->store->log_latency( |
4482 | __func__, | |
11fdf7f2 TL |
4483 | l_bluestore_omap_seek_to_first_lat, |
4484 | mono_clock::now() - start1, | |
494da23a | 4485 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 | 4486 | |
31f18b77 FG |
4487 | return 0; |
4488 | } | |
4489 | ||
4490 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
4491 | { | |
9f95a23c | 4492 | std::shared_lock l(c->lock); |
11fdf7f2 | 4493 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4494 | if (o->onode.has_omap()) { |
4495 | string key; | |
9f95a23c | 4496 | o->get_omap_key(after, &key); |
31f18b77 FG |
4497 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " |
4498 | << pretty_binary_string(key) << dendl; | |
4499 | it->upper_bound(key); | |
4500 | } else { | |
4501 | it = KeyValueDB::Iterator(); | |
4502 | } | |
11fdf7f2 | 4503 | c->store->log_latency_fn( |
494da23a | 4504 | __func__, |
11fdf7f2 TL |
4505 | l_bluestore_omap_upper_bound_lat, |
4506 | mono_clock::now() - start1, | |
494da23a | 4507 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4508 | [&] (const ceph::timespan& lat) { |
494da23a | 4509 | return ", after = " + after + |
11fdf7f2 TL |
4510 | _stringify(); |
4511 | } | |
4512 | ); | |
31f18b77 FG |
4513 | return 0; |
4514 | } | |
4515 | ||
4516 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
4517 | { | |
9f95a23c | 4518 | std::shared_lock l(c->lock); |
11fdf7f2 | 4519 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4520 | if (o->onode.has_omap()) { |
4521 | string key; | |
9f95a23c | 4522 | o->get_omap_key(to, &key); |
31f18b77 FG |
4523 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " |
4524 | << pretty_binary_string(key) << dendl; | |
4525 | it->lower_bound(key); | |
4526 | } else { | |
4527 | it = KeyValueDB::Iterator(); | |
4528 | } | |
11fdf7f2 | 4529 | c->store->log_latency_fn( |
494da23a | 4530 | __func__, |
11fdf7f2 TL |
4531 | l_bluestore_omap_lower_bound_lat, |
4532 | mono_clock::now() - start1, | |
494da23a | 4533 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4534 | [&] (const ceph::timespan& lat) { |
494da23a | 4535 | return ", to = " + to + |
11fdf7f2 TL |
4536 | _stringify(); |
4537 | } | |
4538 | ); | |
31f18b77 FG |
4539 | return 0; |
4540 | } | |
4541 | ||
4542 | bool BlueStore::OmapIteratorImpl::valid() | |
4543 | { | |
9f95a23c | 4544 | std::shared_lock l(c->lock); |
31f18b77 | 4545 | bool r = o->onode.has_omap() && it && it->valid() && |
494da23a | 4546 | it->raw_key().second < tail; |
31f18b77 FG |
4547 | if (it && it->valid()) { |
4548 | ldout(c->store->cct,20) << __func__ << " is at " | |
4549 | << pretty_binary_string(it->raw_key().second) | |
4550 | << dendl; | |
4551 | } | |
4552 | return r; | |
4553 | } | |
4554 | ||
11fdf7f2 | 4555 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 4556 | { |
11fdf7f2 | 4557 | int r = -1; |
9f95a23c | 4558 | std::shared_lock l(c->lock); |
11fdf7f2 | 4559 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4560 | if (o->onode.has_omap()) { |
4561 | it->next(); | |
11fdf7f2 | 4562 | r = 0; |
31f18b77 | 4563 | } |
494da23a TL |
4564 | c->store->log_latency( |
4565 | __func__, | |
11fdf7f2 TL |
4566 | l_bluestore_omap_next_lat, |
4567 | mono_clock::now() - start1, | |
494da23a | 4568 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 TL |
4569 | |
4570 | return r; | |
31f18b77 FG |
4571 | } |
4572 | ||
4573 | string BlueStore::OmapIteratorImpl::key() | |
4574 | { | |
9f95a23c | 4575 | std::shared_lock l(c->lock); |
11fdf7f2 | 4576 | ceph_assert(it->valid()); |
31f18b77 FG |
4577 | string db_key = it->raw_key().second; |
4578 | string user_key; | |
9f95a23c | 4579 | o->decode_omap_key(db_key, &user_key); |
494da23a | 4580 | |
31f18b77 FG |
4581 | return user_key; |
4582 | } | |
4583 | ||
4584 | bufferlist BlueStore::OmapIteratorImpl::value() | |
4585 | { | |
9f95a23c | 4586 | std::shared_lock l(c->lock); |
11fdf7f2 | 4587 | ceph_assert(it->valid()); |
31f18b77 FG |
4588 | return it->value(); |
4589 | } | |
4590 | ||
4591 | ||
4592 | // ===================================== | |
4593 | ||
7c673cae FG |
4594 | #undef dout_prefix |
4595 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
9f95a23c TL |
4596 | #undef dout_context |
4597 | #define dout_context cct | |
7c673cae FG |
4598 | |
4599 | ||
4600 | static void aio_cb(void *priv, void *priv2) | |
4601 | { | |
4602 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4603 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4604 | c->aio_finish(store); | |
4605 | } | |
4606 | ||
11fdf7f2 TL |
4607 | static void discard_cb(void *priv, void *priv2) |
4608 | { | |
4609 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4610 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4611 | store->handle_discard(*tmp); | |
4612 | } | |
4613 | ||
4614 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4615 | { | |
4616 | dout(10) << __func__ << dendl; | |
20effc67 TL |
4617 | ceph_assert(alloc); |
4618 | alloc->release(to_release); | |
11fdf7f2 TL |
4619 | } |
4620 | ||
7c673cae | 4621 | BlueStore::BlueStore(CephContext *cct, const string& path) |
9f95a23c | 4622 | : BlueStore(cct, path, 0) {} |
7c673cae FG |
4623 | |
4624 | BlueStore::BlueStore(CephContext *cct, | |
4625 | const string& path, | |
4626 | uint64_t _min_alloc_size) | |
4627 | : ObjectStore(cct, path), | |
9f95a23c | 4628 | throttle(cct), |
11fdf7f2 | 4629 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4630 | kv_sync_thread(this), |
31f18b77 | 4631 | kv_finalize_thread(this), |
20effc67 | 4632 | #ifdef HAVE_LIBZBD |
f67539c2 | 4633 | zoned_cleaner_thread(this), |
20effc67 | 4634 | #endif |
7c673cae | 4635 | min_alloc_size(_min_alloc_size), |
1e59de90 | 4636 | min_alloc_size_order(std::countr_zero(_min_alloc_size)), |
7c673cae FG |
4637 | mempool_thread(this) |
4638 | { | |
4639 | _init_logger(); | |
11fdf7f2 | 4640 | cct->_conf.add_observer(this); |
7c673cae | 4641 | set_cache_shards(1); |
7c673cae FG |
4642 | } |
4643 | ||
4644 | BlueStore::~BlueStore() | |
4645 | { | |
11fdf7f2 | 4646 | cct->_conf.remove_observer(this); |
7c673cae | 4647 | _shutdown_logger(); |
11fdf7f2 TL |
4648 | ceph_assert(!mounted); |
4649 | ceph_assert(db == NULL); | |
4650 | ceph_assert(bluefs == NULL); | |
4651 | ceph_assert(fsid_fd < 0); | |
4652 | ceph_assert(path_fd < 0); | |
9f95a23c TL |
4653 | for (auto i : onode_cache_shards) { |
4654 | delete i; | |
4655 | } | |
4656 | for (auto i : buffer_cache_shards) { | |
7c673cae FG |
4657 | delete i; |
4658 | } | |
9f95a23c TL |
4659 | onode_cache_shards.clear(); |
4660 | buffer_cache_shards.clear(); | |
7c673cae FG |
4661 | } |
4662 | ||
4663 | const char **BlueStore::get_tracked_conf_keys() const | |
4664 | { | |
4665 | static const char* KEYS[] = { | |
4666 | "bluestore_csum_type", | |
4667 | "bluestore_compression_mode", | |
4668 | "bluestore_compression_algorithm", | |
4669 | "bluestore_compression_min_blob_size", | |
4670 | "bluestore_compression_min_blob_size_ssd", | |
4671 | "bluestore_compression_min_blob_size_hdd", | |
4672 | "bluestore_compression_max_blob_size", | |
4673 | "bluestore_compression_max_blob_size_ssd", | |
4674 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4675 | "bluestore_compression_required_ratio", |
7c673cae FG |
4676 | "bluestore_max_alloc_size", |
4677 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4678 | "bluestore_prefer_deferred_size_hdd", |
4679 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4680 | "bluestore_deferred_batch_ops", |
4681 | "bluestore_deferred_batch_ops_hdd", | |
4682 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4683 | "bluestore_throttle_bytes", |
4684 | "bluestore_throttle_deferred_bytes", | |
4685 | "bluestore_throttle_cost_per_io_hdd", | |
4686 | "bluestore_throttle_cost_per_io_ssd", | |
4687 | "bluestore_throttle_cost_per_io", | |
4688 | "bluestore_max_blob_size", | |
4689 | "bluestore_max_blob_size_ssd", | |
4690 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4691 | "osd_memory_target", |
4692 | "osd_memory_target_cgroup_limit_ratio", | |
4693 | "osd_memory_base", | |
4694 | "osd_memory_cache_min", | |
92f5a8d4 | 4695 | "osd_memory_expected_fragmentation", |
11fdf7f2 TL |
4696 | "bluestore_cache_autotune", |
4697 | "bluestore_cache_autotune_interval", | |
20effc67 TL |
4698 | "bluestore_cache_age_bin_interval", |
4699 | "bluestore_cache_kv_age_bins", | |
4700 | "bluestore_cache_kv_onode_age_bins", | |
4701 | "bluestore_cache_meta_age_bins", | |
4702 | "bluestore_cache_data_age_bins", | |
81eedcae | 4703 | "bluestore_warn_on_legacy_statfs", |
9f95a23c | 4704 | "bluestore_warn_on_no_per_pool_omap", |
20effc67 | 4705 | "bluestore_warn_on_no_per_pg_omap", |
9f95a23c | 4706 | "bluestore_max_defer_interval", |
7c673cae FG |
4707 | NULL |
4708 | }; | |
4709 | return KEYS; | |
4710 | } | |
4711 | ||
11fdf7f2 | 4712 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4713 | const std::set<std::string> &changed) |
4714 | { | |
eafe8130 | 4715 | if (changed.count("bluestore_warn_on_legacy_statfs")) { |
81eedcae TL |
4716 | _check_legacy_statfs_alert(); |
4717 | } | |
f67539c2 TL |
4718 | if (changed.count("bluestore_warn_on_no_per_pool_omap") || |
4719 | changed.count("bluestore_warn_on_no_per_pg_omap")) { | |
4720 | _check_no_per_pg_or_pool_omap_alert(); | |
9f95a23c | 4721 | } |
81eedcae | 4722 | |
7c673cae FG |
4723 | if (changed.count("bluestore_csum_type")) { |
4724 | _set_csum(); | |
4725 | } | |
4726 | if (changed.count("bluestore_compression_mode") || | |
4727 | changed.count("bluestore_compression_algorithm") || | |
4728 | changed.count("bluestore_compression_min_blob_size") || | |
4729 | changed.count("bluestore_compression_max_blob_size")) { | |
4730 | if (bdev) { | |
4731 | _set_compression(); | |
4732 | } | |
4733 | } | |
4734 | if (changed.count("bluestore_max_blob_size") || | |
4735 | changed.count("bluestore_max_blob_size_ssd") || | |
4736 | changed.count("bluestore_max_blob_size_hdd")) { | |
4737 | if (bdev) { | |
4738 | // only after startup | |
4739 | _set_blob_size(); | |
4740 | } | |
4741 | } | |
4742 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4743 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4744 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4745 | changed.count("bluestore_max_alloc_size") || |
4746 | changed.count("bluestore_deferred_batch_ops") || | |
4747 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4748 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4749 | if (bdev) { | |
4750 | // only after startup | |
4751 | _set_alloc_sizes(); | |
4752 | } | |
4753 | } | |
4754 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4755 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4756 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4757 | if (bdev) { | |
4758 | _set_throttle_params(); | |
4759 | } | |
4760 | } | |
9f95a23c TL |
4761 | if (changed.count("bluestore_throttle_bytes") || |
4762 | changed.count("bluestore_throttle_deferred_bytes") || | |
4763 | changed.count("bluestore_throttle_trace_rate")) { | |
4764 | throttle.reset_throttle(conf); | |
7c673cae | 4765 | } |
9f95a23c TL |
4766 | if (changed.count("bluestore_max_defer_interval")) { |
4767 | if (bdev) { | |
4768 | _set_max_defer_interval(); | |
4769 | } | |
7c673cae | 4770 | } |
92f5a8d4 TL |
4771 | if (changed.count("osd_memory_target") || |
4772 | changed.count("osd_memory_base") || | |
4773 | changed.count("osd_memory_cache_min") || | |
4774 | changed.count("osd_memory_expected_fragmentation")) { | |
4775 | _update_osd_memory_options(); | |
4776 | } | |
7c673cae FG |
4777 | } |
4778 | ||
4779 | void BlueStore::_set_compression() | |
4780 | { | |
224ce89b WB |
4781 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4782 | if (m) { | |
11fdf7f2 | 4783 | _clear_compression_alert(); |
224ce89b WB |
4784 | comp_mode = *m; |
4785 | } else { | |
4786 | derr << __func__ << " unrecognized value '" | |
4787 | << cct->_conf->bluestore_compression_mode | |
4788 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4789 | << dendl; | |
4790 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4791 | string s("unknown mode: "); |
4792 | s += cct->_conf->bluestore_compression_mode; | |
4793 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4794 | } |
4795 | ||
4796 | compressor = nullptr; | |
4797 | ||
3efd9988 FG |
4798 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4799 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4800 | } else { |
11fdf7f2 | 4801 | ceph_assert(bdev); |
9f95a23c | 4802 | if (_use_rotational_settings()) { |
7c673cae FG |
4803 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; |
4804 | } else { | |
4805 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4806 | } | |
4807 | } | |
4808 | ||
4809 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4810 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4811 | } else { | |
11fdf7f2 | 4812 | ceph_assert(bdev); |
9f95a23c | 4813 | if (_use_rotational_settings()) { |
7c673cae FG |
4814 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; |
4815 | } else { | |
4816 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4817 | } | |
4818 | } | |
4819 | ||
7c673cae FG |
4820 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4821 | if (!alg_name.empty()) { | |
4822 | compressor = Compressor::create(cct, alg_name); | |
4823 | if (!compressor) { | |
4824 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4825 | << dendl; | |
11fdf7f2 | 4826 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4827 | } |
4828 | } | |
4829 | ||
4830 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4831 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4832 | << " min_blob " << comp_min_blob_size |
4833 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4834 | << dendl; |
4835 | } | |
4836 | ||
4837 | void BlueStore::_set_csum() | |
4838 | { | |
4839 | csum_type = Checksummer::CSUM_NONE; | |
4840 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4841 | if (t > Checksummer::CSUM_NONE) | |
4842 | csum_type = t; | |
4843 | ||
4844 | dout(10) << __func__ << " csum_type " | |
4845 | << Checksummer::get_csum_type_string(csum_type) | |
4846 | << dendl; | |
4847 | } | |
4848 | ||
4849 | void BlueStore::_set_throttle_params() | |
4850 | { | |
4851 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4852 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4853 | } else { | |
11fdf7f2 | 4854 | ceph_assert(bdev); |
9f95a23c | 4855 | if (_use_rotational_settings()) { |
7c673cae FG |
4856 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; |
4857 | } else { | |
4858 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4859 | } | |
4860 | } | |
4861 | ||
4862 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4863 | << dendl; | |
4864 | } | |
4865 | void BlueStore::_set_blob_size() | |
4866 | { | |
4867 | if (cct->_conf->bluestore_max_blob_size) { | |
4868 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4869 | } else { | |
11fdf7f2 | 4870 | ceph_assert(bdev); |
9f95a23c | 4871 | if (_use_rotational_settings()) { |
7c673cae FG |
4872 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; |
4873 | } else { | |
4874 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4875 | } | |
4876 | } | |
4877 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4878 | << std::dec << dendl; | |
4879 | } | |
4880 | ||
92f5a8d4 TL |
4881 | void BlueStore::_update_osd_memory_options() |
4882 | { | |
4883 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4884 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
4885 | osd_memory_expected_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
4886 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
4887 | config_changed++; | |
4888 | dout(10) << __func__ | |
4889 | << " osd_memory_target " << osd_memory_target | |
4890 | << " osd_memory_base " << osd_memory_base | |
4891 | << " osd_memory_expected_fragmentation " << osd_memory_expected_fragmentation | |
4892 | << " osd_memory_cache_min " << osd_memory_cache_min | |
4893 | << dendl; | |
4894 | } | |
4895 | ||
11fdf7f2 | 4896 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4897 | { |
11fdf7f2 TL |
4898 | ceph_assert(bdev); |
4899 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4900 | cache_autotune_interval = |
11fdf7f2 | 4901 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
20effc67 TL |
4902 | cache_age_bin_interval = |
4903 | cct->_conf.get_val<double>("bluestore_cache_age_bin_interval"); | |
4904 | auto _set_bin = [&](std::string conf_name, std::vector<uint64_t>* intervals) | |
4905 | { | |
4906 | std::string intervals_str = cct->_conf.get_val<std::string>(conf_name); | |
4907 | std::istringstream interval_stream(intervals_str); | |
4908 | std::copy( | |
4909 | std::istream_iterator<uint64_t>(interval_stream), | |
4910 | std::istream_iterator<uint64_t>(), | |
4911 | std::back_inserter(*intervals)); | |
4912 | }; | |
4913 | _set_bin("bluestore_cache_age_bins_kv", &kv_bins); | |
4914 | _set_bin("bluestore_cache_age_bins_kv_onode", &kv_onode_bins); | |
4915 | _set_bin("bluestore_cache_age_bins_meta", &meta_bins); | |
4916 | _set_bin("bluestore_cache_age_bins_data", &data_bins); | |
4917 | ||
11fdf7f2 TL |
4918 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); |
4919 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4920 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4921 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4922 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4923 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4924 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4925 | |
224ce89b WB |
4926 | if (cct->_conf->bluestore_cache_size) { |
4927 | cache_size = cct->_conf->bluestore_cache_size; | |
4928 | } else { | |
4929 | // choose global cache size based on backend type | |
9f95a23c | 4930 | if (_use_rotational_settings()) { |
224ce89b WB |
4931 | cache_size = cct->_conf->bluestore_cache_size_hdd; |
4932 | } else { | |
4933 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4934 | } | |
4935 | } | |
31f18b77 | 4936 | |
f67539c2 | 4937 | cache_meta_ratio = cct->_conf.get_val<double>("bluestore_cache_meta_ratio"); |
224ce89b | 4938 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4939 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4940 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4941 | return -EINVAL; |
4942 | } | |
91327a77 | 4943 | |
f67539c2 | 4944 | cache_kv_ratio = cct->_conf.get_val<double>("bluestore_cache_kv_ratio"); |
224ce89b | 4945 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4946 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4947 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4948 | return -EINVAL; |
4949 | } | |
91327a77 | 4950 | |
f67539c2 TL |
4951 | cache_kv_onode_ratio = cct->_conf.get_val<double>("bluestore_cache_kv_onode_ratio"); |
4952 | if (cache_kv_onode_ratio < 0 || cache_kv_onode_ratio > 1.0) { | |
4953 | derr << __func__ << " bluestore_cache_kv_onode_ratio (" << cache_kv_onode_ratio | |
4954 | << ") must be in range [0,1.0]" << dendl; | |
4955 | return -EINVAL; | |
4956 | } | |
4957 | ||
31f18b77 | 4958 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4959 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4960 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4961 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4962 | << dendl; | |
31f18b77 FG |
4963 | return -EINVAL; |
4964 | } | |
91327a77 | 4965 | |
f67539c2 TL |
4966 | cache_data_ratio = (double)1.0 - |
4967 | (double)cache_meta_ratio - | |
4968 | (double)cache_kv_ratio - | |
4969 | (double)cache_kv_onode_ratio; | |
31f18b77 FG |
4970 | if (cache_data_ratio < 0) { |
4971 | // deal with floating point imprecision | |
4972 | cache_data_ratio = 0; | |
4973 | } | |
91327a77 | 4974 | |
224ce89b WB |
4975 | dout(1) << __func__ << " cache_size " << cache_size |
4976 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4977 | << " kv " << cache_kv_ratio |
4978 | << " data " << cache_data_ratio | |
4979 | << dendl; | |
4980 | return 0; | |
4981 | } | |
4982 | ||
3efd9988 FG |
4983 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4984 | { | |
4985 | bluestore_bdev_label_t label; | |
4986 | string p = path + "/block"; | |
4987 | int r = _read_bdev_label(cct, p, &label); | |
4988 | if (r < 0) { | |
4989 | return ObjectStore::write_meta(key, value); | |
4990 | } | |
4991 | label.meta[key] = value; | |
4992 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4993 | ceph_assert(r == 0); |
3efd9988 FG |
4994 | return ObjectStore::write_meta(key, value); |
4995 | } | |
4996 | ||
4997 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4998 | { | |
4999 | bluestore_bdev_label_t label; | |
5000 | string p = path + "/block"; | |
5001 | int r = _read_bdev_label(cct, p, &label); | |
5002 | if (r < 0) { | |
5003 | return ObjectStore::read_meta(key, value); | |
5004 | } | |
5005 | auto i = label.meta.find(key); | |
5006 | if (i == label.meta.end()) { | |
5007 | return ObjectStore::read_meta(key, value); | |
5008 | } | |
5009 | *value = i->second; | |
5010 | return 0; | |
5011 | } | |
5012 | ||
7c673cae FG |
5013 | void BlueStore::_init_logger() |
5014 | { | |
5015 | PerfCountersBuilder b(cct, "bluestore", | |
5016 | l_bluestore_first, l_bluestore_last); | |
20effc67 TL |
5017 | |
5018 | // space utilization stats | |
5019 | //**************************************** | |
5020 | b.add_u64(l_bluestore_allocated, "allocated", | |
5021 | "Sum for allocated bytes", | |
5022 | "al_b", | |
5023 | PerfCountersBuilder::PRIO_CRITICAL, | |
5024 | unit_t(UNIT_BYTES)); | |
5025 | b.add_u64(l_bluestore_stored, "stored", | |
5026 | "Sum for stored bytes", | |
5027 | "st_b", | |
5028 | PerfCountersBuilder::PRIO_CRITICAL, | |
5029 | unit_t(UNIT_BYTES)); | |
5030 | b.add_u64(l_bluestore_fragmentation, "fragmentation_micros", | |
5031 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
5032 | b.add_u64(l_bluestore_alloc_unit, "alloc_unit", | |
5033 | "allocation unit size in bytes", | |
5034 | "au_b", | |
5035 | PerfCountersBuilder::PRIO_CRITICAL, | |
5036 | unit_t(UNIT_BYTES)); | |
5037 | //**************************************** | |
5038 | ||
5039 | // Update op processing state latencies | |
5040 | //**************************************** | |
7c673cae | 5041 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
20effc67 TL |
5042 | "Average prepare state latency", |
5043 | "sprl", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae FG |
5044 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", |
5045 | "Average aio_wait state latency", | |
20effc67 | 5046 | "sawl", PerfCountersBuilder::PRIO_INTERESTING); |
7c673cae | 5047 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", |
20effc67 TL |
5048 | "Average io_done state latency", |
5049 | "sidl", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5050 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", |
20effc67 TL |
5051 | "Average kv_queued state latency", |
5052 | "skql", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5053 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", |
20effc67 TL |
5054 | "Average kv_commiting state latency", |
5055 | "skcl", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5056 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", |
20effc67 TL |
5057 | "Average kv_done state latency", |
5058 | "skdl", PerfCountersBuilder::PRIO_USEFUL); | |
5059 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
5060 | "Average finishing state latency", | |
5061 | "sfnl", PerfCountersBuilder::PRIO_USEFUL); | |
5062 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
5063 | "Average done state latency", | |
5064 | "sdnl", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5065 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", |
20effc67 TL |
5066 | "Average deferred_queued state latency", |
5067 | "sdql", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5068 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", |
20effc67 TL |
5069 | "Average aio_wait state latency", |
5070 | "sdal", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5071 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", |
20effc67 TL |
5072 | "Average cleanup state latency", |
5073 | "sdcl", PerfCountersBuilder::PRIO_USEFUL); | |
5074 | //**************************************** | |
5075 | ||
5076 | // Update Transaction stats | |
5077 | //**************************************** | |
5078 | b.add_time_avg(l_bluestore_throttle_lat, "txc_throttle_lat", | |
7c673cae FG |
5079 | "Average submit throttle latency", |
5080 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
20effc67 | 5081 | b.add_time_avg(l_bluestore_submit_lat, "txc_submit_lat", |
7c673cae FG |
5082 | "Average submit latency", |
5083 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
20effc67 | 5084 | b.add_time_avg(l_bluestore_commit_lat, "txc_commit_lat", |
7c673cae FG |
5085 | "Average commit latency", |
5086 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
20effc67 TL |
5087 | b.add_u64_counter(l_bluestore_txc, "txc_count", "Transactions committed"); |
5088 | //**************************************** | |
5089 | ||
5090 | // Read op stats | |
5091 | //**************************************** | |
7c673cae | 5092 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", |
20effc67 TL |
5093 | "Average read onode metadata latency", |
5094 | "roml", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5095 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", |
20effc67 TL |
5096 | "Average read I/O waiting latency", |
5097 | "rwal", PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae | 5098 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", |
20effc67 TL |
5099 | "Average checksum latency", |
5100 | "csml", PerfCountersBuilder::PRIO_USEFUL); | |
5101 | b.add_u64_counter(l_bluestore_read_eio, "read_eio", | |
5102 | "Read EIO errors propagated to high level callers"); | |
5103 | b.add_u64_counter(l_bluestore_reads_with_retries, "reads_with_retries", | |
5104 | "Read operations that required at least one retry due to failed checksum validation", | |
5105 | "rd_r", PerfCountersBuilder::PRIO_USEFUL); | |
5106 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
5107 | "Average read latency", | |
5108 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
5109 | //**************************************** | |
5110 | ||
5111 | // kv_thread latencies | |
5112 | //**************************************** | |
5113 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
5114 | "Average kv_thread flush latency", | |
5115 | "kfsl", PerfCountersBuilder::PRIO_INTERESTING); | |
5116 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
5117 | "Average kv_thread commit latency", | |
5118 | "kcol", PerfCountersBuilder::PRIO_USEFUL); | |
5119 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", | |
5120 | "Average kv_sync thread latency", | |
5121 | "kscl", PerfCountersBuilder::PRIO_INTERESTING); | |
5122 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
5123 | "Average kv_finalize thread latency", | |
5124 | "kfll", PerfCountersBuilder::PRIO_INTERESTING); | |
5125 | //**************************************** | |
5126 | ||
5127 | // write op stats | |
5128 | //**************************************** | |
5129 | b.add_u64_counter(l_bluestore_write_big, "write_big", | |
7c673cae | 5130 | "Large aligned writes into fresh blobs"); |
20effc67 TL |
5131 | b.add_u64_counter(l_bluestore_write_big_bytes, "write_big_bytes", |
5132 | "Large aligned writes into fresh blobs (bytes)", | |
5133 | NULL, | |
5134 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5135 | unit_t(UNIT_BYTES)); | |
5136 | b.add_u64_counter(l_bluestore_write_big_blobs, "write_big_blobs", | |
7c673cae | 5137 | "Large aligned writes into fresh blobs (blobs)"); |
f67539c2 | 5138 | b.add_u64_counter(l_bluestore_write_big_deferred, |
20effc67 | 5139 | "write_big_deferred", |
f67539c2 | 5140 | "Big overwrites using deferred"); |
20effc67 TL |
5141 | |
5142 | b.add_u64_counter(l_bluestore_write_small, "write_small", | |
7c673cae | 5143 | "Small writes into existing or sparse small blobs"); |
20effc67 TL |
5144 | b.add_u64_counter(l_bluestore_write_small_bytes, "write_small_bytes", |
5145 | "Small writes into existing or sparse small blobs (bytes)", | |
5146 | NULL, | |
5147 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5148 | unit_t(UNIT_BYTES)); | |
7c673cae | 5149 | b.add_u64_counter(l_bluestore_write_small_unused, |
20effc67 | 5150 | "write_small_unused", |
7c673cae | 5151 | "Small writes into unused portion of existing blob"); |
7c673cae | 5152 | b.add_u64_counter(l_bluestore_write_small_pre_read, |
20effc67 | 5153 | "write_small_pre_read", |
7c673cae FG |
5154 | "Small writes that required we read some data (possibly " |
5155 | "cached) to fill out the block"); | |
7c673cae | 5156 | |
20effc67 TL |
5157 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", |
5158 | "Sum for write-op padded bytes", | |
5159 | NULL, | |
5160 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5161 | unit_t(UNIT_BYTES)); | |
5162 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", | |
5163 | "Sum for write penalty read ops"); | |
5164 | b.add_u64_counter(l_bluestore_write_new, "write_new", | |
5165 | "Write into new blob"); | |
5166 | ||
5167 | b.add_u64_counter(l_bluestore_issued_deferred_writes, | |
5168 | "issued_deferred_writes", | |
5169 | "Total deferred writes issued"); | |
5170 | b.add_u64_counter(l_bluestore_issued_deferred_write_bytes, | |
5171 | "issued_deferred_write_bytes", | |
5172 | "Total bytes in issued deferred writes", | |
5173 | NULL, | |
5174 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5175 | unit_t(UNIT_BYTES)); | |
5176 | b.add_u64_counter(l_bluestore_submitted_deferred_writes, | |
5177 | "submitted_deferred_writes", | |
5178 | "Total deferred writes submitted to disk"); | |
5179 | b.add_u64_counter(l_bluestore_submitted_deferred_write_bytes, | |
5180 | "submitted_deferred_write_bytes", | |
5181 | "Total bytes submitted to disk by deferred writes", | |
5182 | NULL, | |
5183 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5184 | unit_t(UNIT_BYTES)); | |
5185 | ||
5186 | b.add_u64_counter(l_bluestore_write_big_skipped_blobs, | |
5187 | "write_big_skipped_blobs", | |
5188 | "Large aligned writes into fresh blobs skipped due to zero detection (blobs)"); | |
5189 | b.add_u64_counter(l_bluestore_write_big_skipped_bytes, | |
5190 | "write_big_skipped_bytes", | |
5191 | "Large aligned writes into fresh blobs skipped due to zero detection (bytes)"); | |
5192 | b.add_u64_counter(l_bluestore_write_small_skipped, | |
5193 | "write_small_skipped", | |
5194 | "Small writes into existing or sparse small blobs skipped due to zero detection"); | |
5195 | b.add_u64_counter(l_bluestore_write_small_skipped_bytes, | |
5196 | "write_small_skipped_bytes", | |
5197 | "Small writes into existing or sparse small blobs skipped due to zero detection (bytes)"); | |
5198 | //**************************************** | |
5199 | ||
5200 | // compressions stats | |
5201 | //**************************************** | |
5202 | b.add_u64(l_bluestore_compressed, "compressed", | |
5203 | "Sum for stored compressed bytes", | |
5204 | "c", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
5205 | b.add_u64(l_bluestore_compressed_allocated, "compressed_allocated", | |
5206 | "Sum for bytes allocated for compressed data", | |
5207 | "c_a", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
5208 | b.add_u64(l_bluestore_compressed_original, "compressed_original", | |
5209 | "Sum for original bytes that were compressed", | |
5210 | "c_o", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
5211 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
5212 | "Average compress latency", | |
5213 | "_cpl", PerfCountersBuilder::PRIO_USEFUL); | |
5214 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
5215 | "Average decompress latency", | |
5216 | "dcpl", PerfCountersBuilder::PRIO_USEFUL); | |
5217 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
5218 | "Sum for beneficial compress ops"); | |
5219 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
5220 | "Sum for compress ops rejected due to low net gain of space"); | |
5221 | //**************************************** | |
5222 | ||
5223 | // onode cache stats | |
5224 | //**************************************** | |
5225 | b.add_u64(l_bluestore_onodes, "onodes", | |
5226 | "Number of onodes in cache"); | |
5227 | b.add_u64(l_bluestore_pinned_onodes, "onodes_pinned", | |
5228 | "Number of pinned onodes in cache"); | |
5229 | b.add_u64_counter(l_bluestore_onode_hits, "onode_hits", | |
5230 | "Count of onode cache lookup hits", | |
5231 | "o_ht", PerfCountersBuilder::PRIO_USEFUL); | |
5232 | b.add_u64_counter(l_bluestore_onode_misses, "onode_misses", | |
5233 | "Count of onode cache lookup misses", | |
5234 | "o_ms", PerfCountersBuilder::PRIO_USEFUL); | |
5235 | b.add_u64_counter(l_bluestore_onode_shard_hits, "onode_shard_hits", | |
5236 | "Count of onode shard cache lookups hits"); | |
5237 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
5238 | "onode_shard_misses", | |
5239 | "Count of onode shard cache lookups misses"); | |
5240 | b.add_u64(l_bluestore_extents, "onode_extents", | |
5241 | "Number of extents in cache"); | |
5242 | b.add_u64(l_bluestore_blobs, "onode_blobs", | |
5243 | "Number of blobs in cache"); | |
5244 | //**************************************** | |
5245 | ||
5246 | // buffer cache stats | |
5247 | //**************************************** | |
5248 | b.add_u64(l_bluestore_buffers, "buffers", | |
5249 | "Number of buffers in cache"); | |
5250 | b.add_u64(l_bluestore_buffer_bytes, "buffer_bytes", | |
5251 | "Number of buffer bytes in cache", | |
5252 | NULL, | |
5253 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5254 | unit_t(UNIT_BYTES)); | |
5255 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "buffer_hit_bytes", | |
5256 | "Sum for bytes of read hit in the cache", | |
5257 | NULL, | |
5258 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5259 | unit_t(UNIT_BYTES)); | |
5260 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "buffer_miss_bytes", | |
5261 | "Sum for bytes of read missed in the cache", | |
5262 | NULL, | |
5263 | PerfCountersBuilder::PRIO_DEBUGONLY, | |
5264 | unit_t(UNIT_BYTES)); | |
5265 | //**************************************** | |
5266 | ||
5267 | // internal stats | |
5268 | //**************************************** | |
5269 | b.add_u64_counter(l_bluestore_onode_reshard, "onode_reshard", | |
5270 | "Onode extent map reshard events"); | |
5271 | b.add_u64_counter(l_bluestore_blob_split, "blob_split", | |
7c673cae | 5272 | "Sum for blob splitting due to resharding"); |
20effc67 | 5273 | b.add_u64_counter(l_bluestore_extent_compress, "extent_compress", |
7c673cae | 5274 | "Sum for extents that have been removed due to compression"); |
20effc67 | 5275 | b.add_u64_counter(l_bluestore_gc_merged, "gc_merged", |
7c673cae FG |
5276 | "Sum for extents that have been merged due to garbage " |
5277 | "collection"); | |
20effc67 | 5278 | //**************************************** |
1e59de90 TL |
5279 | // misc |
5280 | //**************************************** | |
5281 | b.add_u64_counter(l_bluestore_omap_iterator_count, "omap_iterator_count", | |
5282 | "Open omap iterators count"); | |
5283 | b.add_u64_counter(l_bluestore_omap_rmkeys_count, "omap_rmkeys_count", | |
5284 | "amount of omap keys removed via rmkeys"); | |
5285 | b.add_u64_counter(l_bluestore_omap_rmkey_ranges_count, "omap_rmkey_range_count", | |
5286 | "amount of omap key ranges removed via rmkeys"); | |
5287 | //**************************************** | |
20effc67 TL |
5288 | // other client ops latencies |
5289 | //**************************************** | |
11fdf7f2 | 5290 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
20effc67 TL |
5291 | "Average omap iterator seek_to_first call latency", |
5292 | "osfl", PerfCountersBuilder::PRIO_USEFUL); | |
11fdf7f2 | 5293 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", |
20effc67 TL |
5294 | "Average omap iterator upper_bound call latency", |
5295 | "oubl", PerfCountersBuilder::PRIO_USEFUL); | |
11fdf7f2 | 5296 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", |
20effc67 TL |
5297 | "Average omap iterator lower_bound call latency", |
5298 | "olbl", PerfCountersBuilder::PRIO_USEFUL); | |
11fdf7f2 | 5299 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", |
20effc67 TL |
5300 | "Average omap iterator next call latency", |
5301 | "onxl", PerfCountersBuilder::PRIO_USEFUL); | |
adb31ebb | 5302 | b.add_time_avg(l_bluestore_omap_get_keys_lat, "omap_get_keys_lat", |
20effc67 TL |
5303 | "Average omap get_keys call latency", |
5304 | "ogkl", PerfCountersBuilder::PRIO_USEFUL); | |
adb31ebb | 5305 | b.add_time_avg(l_bluestore_omap_get_values_lat, "omap_get_values_lat", |
20effc67 TL |
5306 | "Average omap get_values call latency", |
5307 | "ogvl", PerfCountersBuilder::PRIO_USEFUL); | |
5308 | b.add_time_avg(l_bluestore_omap_clear_lat, "omap_clear_lat", | |
5309 | "Average omap clear call latency"); | |
494da23a | 5310 | b.add_time_avg(l_bluestore_clist_lat, "clist_lat", |
20effc67 TL |
5311 | "Average collection listing latency", |
5312 | "cl_l", PerfCountersBuilder::PRIO_USEFUL); | |
adb31ebb | 5313 | b.add_time_avg(l_bluestore_remove_lat, "remove_lat", |
20effc67 TL |
5314 | "Average removal latency", |
5315 | "rm_l", PerfCountersBuilder::PRIO_USEFUL); | |
5316 | b.add_time_avg(l_bluestore_truncate_lat, "truncate_lat", | |
5317 | "Average truncate latency", | |
5318 | "tr_l", PerfCountersBuilder::PRIO_USEFUL); | |
5319 | //**************************************** | |
5320 | ||
5321 | // Resulting size axis configuration for op histograms, values are in bytes | |
5322 | PerfHistogramCommon::axis_config_d alloc_hist_x_axis_config{ | |
5323 | "Given size (bytes)", | |
5324 | PerfHistogramCommon::SCALE_LOG2, ///< Request size in logarithmic scale | |
5325 | 0, ///< Start at 0 | |
5326 | 4096, ///< Quantization unit | |
5327 | 13, ///< Enough to cover 4+M requests | |
5328 | }; | |
5329 | // Req size axis configuration for op histograms, values are in bytes | |
5330 | PerfHistogramCommon::axis_config_d alloc_hist_y_axis_config{ | |
5331 | "Request size (bytes)", | |
5332 | PerfHistogramCommon::SCALE_LOG2, ///< Request size in logarithmic scale | |
5333 | 0, ///< Start at 0 | |
5334 | 4096, ///< Quantization unit | |
5335 | 13, ///< Enough to cover 4+M requests | |
5336 | }; | |
5337 | b.add_u64_counter_histogram( | |
5338 | l_bluestore_allocate_hist, "allocate_histogram", | |
5339 | alloc_hist_x_axis_config, alloc_hist_y_axis_config, | |
5340 | "Histogram of requested block allocations vs. given ones"); | |
adb31ebb | 5341 | |
7c673cae FG |
5342 | logger = b.create_perf_counters(); |
5343 | cct->get_perfcounters_collection()->add(logger); | |
5344 | } | |
5345 | ||
5346 | int BlueStore::_reload_logger() | |
5347 | { | |
5348 | struct store_statfs_t store_statfs; | |
7c673cae | 5349 | int r = statfs(&store_statfs); |
11fdf7f2 | 5350 | if (r >= 0) { |
7c673cae | 5351 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
5352 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
5353 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
5354 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
5355 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
5356 | } |
5357 | return r; | |
5358 | } | |
5359 | ||
5360 | void BlueStore::_shutdown_logger() | |
5361 | { | |
5362 | cct->get_perfcounters_collection()->remove(logger); | |
5363 | delete logger; | |
5364 | } | |
5365 | ||
5366 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
5367 | uuid_d *fsid) | |
5368 | { | |
5369 | bluestore_bdev_label_t label; | |
5370 | int r = _read_bdev_label(cct, path, &label); | |
5371 | if (r < 0) | |
5372 | return r; | |
5373 | *fsid = label.osd_uuid; | |
5374 | return 0; | |
5375 | } | |
5376 | ||
5377 | int BlueStore::_open_path() | |
5378 | { | |
b32b8144 | 5379 | // sanity check(s) |
11fdf7f2 | 5380 | ceph_assert(path_fd < 0); |
91327a77 | 5381 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
5382 | if (path_fd < 0) { |
5383 | int r = -errno; | |
5384 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
5385 | << dendl; | |
5386 | return r; | |
5387 | } | |
5388 | return 0; | |
5389 | } | |
5390 | ||
5391 | void BlueStore::_close_path() | |
5392 | { | |
5393 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
5394 | path_fd = -1; | |
5395 | } | |
5396 | ||
3efd9988 | 5397 | int BlueStore::_write_bdev_label(CephContext *cct, |
20effc67 | 5398 | const string &path, bluestore_bdev_label_t label) |
7c673cae FG |
5399 | { |
5400 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
5401 | bufferlist bl; | |
11fdf7f2 | 5402 | encode(label, bl); |
7c673cae | 5403 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
5404 | encode(crc, bl); |
5405 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
5406 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
5407 | z.zero(); | |
5408 | bl.append(std::move(z)); | |
5409 | ||
1e59de90 | 5410 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC|O_DIRECT)); |
7c673cae FG |
5411 | if (fd < 0) { |
5412 | fd = -errno; | |
5413 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5414 | << dendl; | |
5415 | return fd; | |
5416 | } | |
1e59de90 | 5417 | bl.rebuild_aligned_size_and_memory(BDEV_LABEL_BLOCK_SIZE, BDEV_LABEL_BLOCK_SIZE, IOV_MAX); |
7c673cae FG |
5418 | int r = bl.write_fd(fd); |
5419 | if (r < 0) { | |
5420 | derr << __func__ << " failed to write to " << path | |
5421 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 5422 | goto out; |
7c673cae | 5423 | } |
3efd9988 FG |
5424 | r = ::fsync(fd); |
5425 | if (r < 0) { | |
5426 | derr << __func__ << " failed to fsync " << path | |
5427 | << ": " << cpp_strerror(r) << dendl; | |
5428 | } | |
11fdf7f2 | 5429 | out: |
7c673cae FG |
5430 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
5431 | return r; | |
5432 | } | |
5433 | ||
20effc67 | 5434 | int BlueStore::_read_bdev_label(CephContext* cct, const string &path, |
7c673cae FG |
5435 | bluestore_bdev_label_t *label) |
5436 | { | |
5437 | dout(10) << __func__ << dendl; | |
91327a77 | 5438 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
5439 | if (fd < 0) { |
5440 | fd = -errno; | |
5441 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5442 | << dendl; | |
5443 | return fd; | |
5444 | } | |
5445 | bufferlist bl; | |
5446 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
5447 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5448 | if (r < 0) { | |
5449 | derr << __func__ << " failed to read from " << path | |
5450 | << ": " << cpp_strerror(r) << dendl; | |
5451 | return r; | |
5452 | } | |
5453 | ||
5454 | uint32_t crc, expected_crc; | |
11fdf7f2 | 5455 | auto p = bl.cbegin(); |
7c673cae | 5456 | try { |
11fdf7f2 | 5457 | decode(*label, p); |
7c673cae FG |
5458 | bufferlist t; |
5459 | t.substr_of(bl, 0, p.get_off()); | |
5460 | crc = t.crc32c(-1); | |
11fdf7f2 | 5461 | decode(expected_crc, p); |
7c673cae | 5462 | } |
f67539c2 | 5463 | catch (ceph::buffer::error& e) { |
1e59de90 | 5464 | derr << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
5465 | << ": " << e.what() |
5466 | << dendl; | |
b32b8144 | 5467 | return -ENOENT; |
7c673cae FG |
5468 | } |
5469 | if (crc != expected_crc) { | |
5470 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
5471 | << " != actual " << crc << dendl; | |
5472 | return -EIO; | |
5473 | } | |
5474 | dout(10) << __func__ << " got " << *label << dendl; | |
5475 | return 0; | |
5476 | } | |
5477 | ||
5478 | int BlueStore::_check_or_set_bdev_label( | |
5479 | string path, uint64_t size, string desc, bool create) | |
5480 | { | |
5481 | bluestore_bdev_label_t label; | |
5482 | if (create) { | |
5483 | label.osd_uuid = fsid; | |
5484 | label.size = size; | |
5485 | label.btime = ceph_clock_now(); | |
5486 | label.description = desc; | |
3efd9988 | 5487 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
5488 | if (r < 0) |
5489 | return r; | |
5490 | } else { | |
5491 | int r = _read_bdev_label(cct, path, &label); | |
5492 | if (r < 0) | |
5493 | return r; | |
31f18b77 FG |
5494 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
5495 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
5496 | << " and fsid " << fsid << " check bypassed" << dendl; | |
1911f103 | 5497 | } else if (label.osd_uuid != fsid) { |
7c673cae FG |
5498 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
5499 | << " does not match our fsid " << fsid << dendl; | |
5500 | return -EIO; | |
5501 | } | |
5502 | } | |
5503 | return 0; | |
5504 | } | |
5505 | ||
5506 | void BlueStore::_set_alloc_sizes(void) | |
5507 | { | |
7c673cae FG |
5508 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
5509 | ||
20effc67 TL |
5510 | #ifdef HAVE_LIBZBD |
5511 | ceph_assert(bdev); | |
5512 | if (bdev->is_smr()) { | |
5513 | prefer_deferred_size = 0; | |
5514 | } else | |
5515 | #endif | |
7c673cae FG |
5516 | if (cct->_conf->bluestore_prefer_deferred_size) { |
5517 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
5518 | } else { | |
9f95a23c | 5519 | if (_use_rotational_settings()) { |
7c673cae FG |
5520 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; |
5521 | } else { | |
5522 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
5523 | } | |
5524 | } | |
5525 | ||
5526 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
5527 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
5528 | } else { | |
9f95a23c | 5529 | if (_use_rotational_settings()) { |
7c673cae FG |
5530 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; |
5531 | } else { | |
5532 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
5533 | } | |
5534 | } | |
5535 | ||
5536 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 5537 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
5538 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
5539 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
5540 | << std::dec | |
5541 | << " deferred_batch_ops " << deferred_batch_ops | |
5542 | << dendl; | |
5543 | } | |
5544 | ||
5545 | int BlueStore::_open_bdev(bool create) | |
5546 | { | |
11fdf7f2 | 5547 | ceph_assert(bdev == NULL); |
7c673cae | 5548 | string p = path + "/block"; |
11fdf7f2 | 5549 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
5550 | int r = bdev->open(p); |
5551 | if (r < 0) | |
5552 | goto fail; | |
5553 | ||
11fdf7f2 | 5554 | if (create && cct->_conf->bdev_enable_discard) { |
1e59de90 TL |
5555 | interval_set<uint64_t> whole_device; |
5556 | whole_device.insert(0, bdev->get_size()); | |
5557 | bdev->try_discard(whole_device, false); | |
11fdf7f2 TL |
5558 | } |
5559 | ||
7c673cae FG |
5560 | if (bdev->supported_bdev_label()) { |
5561 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
5562 | if (r < 0) | |
5563 | goto fail_close; | |
5564 | } | |
5565 | ||
5566 | // initialize global block parameters | |
5567 | block_size = bdev->get_block_size(); | |
5568 | block_mask = ~(block_size - 1); | |
1e59de90 | 5569 | block_size_order = std::countr_zero(block_size); |
11fdf7f2 | 5570 | ceph_assert(block_size == 1u << block_size_order); |
9f95a23c | 5571 | _set_max_defer_interval(); |
224ce89b WB |
5572 | // and set cache_size based on device type |
5573 | r = _set_cache_sizes(); | |
5574 | if (r < 0) { | |
5575 | goto fail_close; | |
5576 | } | |
20effc67 TL |
5577 | // get block dev optimal io size |
5578 | optimal_io_size = bdev->get_optimal_io_size(); | |
f67539c2 | 5579 | |
7c673cae FG |
5580 | return 0; |
5581 | ||
5582 | fail_close: | |
5583 | bdev->close(); | |
5584 | fail: | |
5585 | delete bdev; | |
5586 | bdev = NULL; | |
5587 | return r; | |
5588 | } | |
5589 | ||
11fdf7f2 TL |
5590 | void BlueStore::_validate_bdev() |
5591 | { | |
5592 | ceph_assert(bdev); | |
11fdf7f2 | 5593 | uint64_t dev_size = bdev->get_size(); |
f67539c2 | 5594 | ceph_assert(dev_size > _get_ondisk_reserved()); |
11fdf7f2 TL |
5595 | } |
5596 | ||
7c673cae FG |
5597 | void BlueStore::_close_bdev() |
5598 | { | |
11fdf7f2 | 5599 | ceph_assert(bdev); |
7c673cae FG |
5600 | bdev->close(); |
5601 | delete bdev; | |
5602 | bdev = NULL; | |
5603 | } | |
5604 | ||
39ae355f TL |
5605 | int BlueStore::_open_fm(KeyValueDB::Transaction t, |
5606 | bool read_only, | |
5607 | bool db_avail, | |
5608 | bool fm_restore) | |
7c673cae | 5609 | { |
1911f103 | 5610 | int r; |
1911f103 | 5611 | |
20effc67 | 5612 | dout(5) << __func__ << "::NCB::freelist_type=" << freelist_type << dendl; |
11fdf7f2 | 5613 | ceph_assert(fm == NULL); |
20effc67 TL |
5614 | // fm_restore means we are transitioning from null-fm to bitmap-fm |
5615 | ceph_assert(!fm_restore || (freelist_type != "null")); | |
5616 | // fm restore must pass in a valid transaction | |
5617 | ceph_assert(!fm_restore || (t != nullptr)); | |
5618 | ||
39ae355f TL |
5619 | // when function is called in repair mode (to_repair=true) we skip db->open()/create() |
5620 | bool can_have_null_fm = !is_db_rotational() && | |
5621 | !read_only && | |
5622 | db_avail && | |
5623 | cct->_conf->bluestore_allocation_from_file && | |
5624 | !bdev->is_smr(); | |
5625 | ||
20effc67 | 5626 | // When allocation-info is stored in a single file we set freelist_type to "null" |
39ae355f TL |
5627 | if (can_have_null_fm) { |
5628 | freelist_type = "null"; | |
5629 | need_to_destage_allocation_file = true; | |
20effc67 | 5630 | } |
11fdf7f2 TL |
5631 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); |
5632 | ceph_assert(fm); | |
5633 | if (t) { | |
5634 | // create mode. initialize freespace | |
7c673cae | 5635 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
5636 | { |
5637 | bufferlist bl; | |
5638 | bl.append(freelist_type); | |
5639 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
5640 | } | |
b32b8144 FG |
5641 | // being able to allocate in units less than bdev block size |
5642 | // seems to be a bad idea. | |
20effc67 | 5643 | ceph_assert(cct->_conf->bdev_block_size <= min_alloc_size); |
f67539c2 TL |
5644 | |
5645 | uint64_t alloc_size = min_alloc_size; | |
39ae355f TL |
5646 | if (bdev->is_smr() && freelist_type != "zoned") { |
5647 | derr << "SMR device but freelist_type = " << freelist_type << " (not zoned)" | |
5648 | << dendl; | |
5649 | return -EINVAL; | |
5650 | } | |
5651 | if (!bdev->is_smr() && freelist_type == "zoned") { | |
20effc67 TL |
5652 | derr << "non-SMR device (or SMR support not built-in) but freelist_type = zoned" |
5653 | << dendl; | |
5654 | return -EINVAL; | |
f67539c2 TL |
5655 | } |
5656 | ||
20effc67 TL |
5657 | fm->create(bdev->get_size(), alloc_size, |
5658 | zone_size, first_sequential_zone, | |
5659 | t); | |
7c673cae FG |
5660 | |
5661 | // allocate superblock reserved space. note that we do not mark | |
5662 | // bluefs space as allocated in the freelist; we instead rely on | |
f67539c2 | 5663 | // bluefs doing that itself. |
11fdf7f2 | 5664 | auto reserved = _get_ondisk_reserved(); |
20effc67 TL |
5665 | if (fm_restore) { |
5666 | // we need to allocate the full space in restore case | |
5667 | // as later we will add free-space marked in the allocator file | |
5668 | fm->allocate(0, bdev->get_size(), t); | |
5669 | } else { | |
5670 | // allocate superblock reserved space. note that we do not mark | |
5671 | // bluefs space as allocated in the freelist; we instead rely on | |
5672 | // bluefs doing that itself. | |
5673 | fm->allocate(0, reserved, t); | |
5674 | } | |
5675 | // debug code - not needed for NULL FM | |
7c673cae FG |
5676 | if (cct->_conf->bluestore_debug_prefill > 0) { |
5677 | uint64_t end = bdev->get_size() - reserved; | |
5678 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
5679 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
5680 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 5681 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
5682 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
5683 | float r = cct->_conf->bluestore_debug_prefill; | |
5684 | r /= 1.0 - r; | |
5685 | bool stop = false; | |
5686 | ||
5687 | while (!stop && start < end) { | |
5688 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
5689 | if (start + l > end) { | |
5690 | l = end - start; | |
11fdf7f2 | 5691 | l = p2align(l, min_alloc_size); |
7c673cae | 5692 | } |
11fdf7f2 | 5693 | ceph_assert(start + l <= end); |
7c673cae FG |
5694 | |
5695 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 5696 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
5697 | if (start + l + u > end) { |
5698 | u = end - (start + l); | |
5699 | // trim to align so we don't overflow again | |
11fdf7f2 | 5700 | u = p2align(u, min_alloc_size); |
7c673cae FG |
5701 | stop = true; |
5702 | } | |
11fdf7f2 | 5703 | ceph_assert(start + l + u <= end); |
7c673cae | 5704 | |
11fdf7f2 | 5705 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
5706 | << " use 0x" << u << std::dec << dendl; |
5707 | ||
5708 | if (u == 0) { | |
5709 | // break if u has been trimmed to nothing | |
5710 | break; | |
5711 | } | |
5712 | ||
5713 | fm->allocate(start + l, u, t); | |
5714 | start += l + u; | |
5715 | } | |
5716 | } | |
f67539c2 | 5717 | r = _write_out_fm_meta(0); |
1911f103 TL |
5718 | ceph_assert(r == 0); |
5719 | } else { | |
39ae355f TL |
5720 | if (can_have_null_fm) { |
5721 | commit_to_null_manager(); | |
5722 | } | |
f67539c2 TL |
5723 | r = fm->init(db, read_only, |
5724 | [&](const std::string& key, std::string* result) { | |
5725 | return read_meta(key, result); | |
5726 | }); | |
1911f103 | 5727 | if (r < 0) { |
39ae355f | 5728 | derr << __func__ << " failed: " << cpp_strerror(r) << dendl; |
1911f103 TL |
5729 | delete fm; |
5730 | fm = NULL; | |
5731 | return r; | |
5732 | } | |
7c673cae | 5733 | } |
81eedcae TL |
5734 | // if space size tracked by free list manager is that higher than actual |
5735 | // dev size one can hit out-of-space allocation which will result | |
5736 | // in data loss and/or assertions | |
5737 | // Probably user altered the device size somehow. | |
5738 | // The only fix for now is to redeploy OSD. | |
5739 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
5740 | ostringstream ss; | |
5741 | ss << "slow device size mismatch detected, " | |
5742 | << " fm size(" << fm->get_size() | |
5743 | << ") > slow device size(" << bdev->get_size() | |
5744 | << "), Please stop using this OSD as it might cause data loss."; | |
5745 | _set_disk_size_mismatch_alert(ss.str()); | |
5746 | } | |
7c673cae FG |
5747 | return 0; |
5748 | } | |
5749 | ||
5750 | void BlueStore::_close_fm() | |
5751 | { | |
5752 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 5753 | ceph_assert(fm); |
7c673cae FG |
5754 | fm->shutdown(); |
5755 | delete fm; | |
5756 | fm = NULL; | |
5757 | } | |
5758 | ||
f67539c2 | 5759 | int BlueStore::_write_out_fm_meta(uint64_t target_size) |
1911f103 | 5760 | { |
f67539c2 | 5761 | int r = 0; |
1911f103 TL |
5762 | string p = path + "/block"; |
5763 | ||
5764 | std::vector<std::pair<string, string>> fm_meta; | |
5765 | fm->get_meta(target_size, &fm_meta); | |
5766 | ||
1911f103 | 5767 | for (auto& m : fm_meta) { |
f67539c2 TL |
5768 | r = write_meta(m.first, m.second); |
5769 | ceph_assert(r == 0); | |
1911f103 | 5770 | } |
1911f103 TL |
5771 | return r; |
5772 | } | |
5773 | ||
f67539c2 | 5774 | int BlueStore::_create_alloc() |
7c673cae | 5775 | { |
20effc67 | 5776 | ceph_assert(alloc == NULL); |
f67539c2 | 5777 | ceph_assert(shared_alloc.a == NULL); |
11fdf7f2 TL |
5778 | ceph_assert(bdev->get_size()); |
5779 | ||
f67539c2 | 5780 | uint64_t alloc_size = min_alloc_size; |
20effc67 TL |
5781 | |
5782 | std::string allocator_type = cct->_conf->bluestore_allocator; | |
5783 | ||
5784 | #ifdef HAVE_LIBZBD | |
5785 | if (freelist_type == "zoned") { | |
5786 | allocator_type = "zoned"; | |
11fdf7f2 | 5787 | } |
20effc67 | 5788 | #endif |
11fdf7f2 | 5789 | |
20effc67 TL |
5790 | alloc = Allocator::create( |
5791 | cct, allocator_type, | |
f67539c2 | 5792 | bdev->get_size(), |
20effc67 TL |
5793 | alloc_size, |
5794 | zone_size, | |
5795 | first_sequential_zone, | |
5796 | "block"); | |
5797 | if (!alloc) { | |
5798 | lderr(cct) << __func__ << " failed to create " << allocator_type << " allocator" | |
5799 | << dendl; | |
7c673cae FG |
5800 | return -EINVAL; |
5801 | } | |
20effc67 TL |
5802 | |
5803 | #ifdef HAVE_LIBZBD | |
5804 | if (freelist_type == "zoned") { | |
5805 | Allocator *a = Allocator::create( | |
5806 | cct, cct->_conf->bluestore_allocator, | |
5807 | bdev->get_conventional_region_size(), | |
5808 | alloc_size, | |
5809 | 0, 0, | |
5810 | "zoned_block"); | |
5811 | if (!a) { | |
5812 | lderr(cct) << __func__ << " failed to create " << cct->_conf->bluestore_allocator | |
5813 | << " allocator" << dendl; | |
5814 | delete alloc; | |
5815 | return -EINVAL; | |
5816 | } | |
39ae355f | 5817 | shared_alloc.set(a, alloc_size); |
20effc67 TL |
5818 | } else |
5819 | #endif | |
5820 | { | |
5821 | // BlueFS will share the same allocator | |
39ae355f | 5822 | shared_alloc.set(alloc, alloc_size); |
20effc67 TL |
5823 | } |
5824 | ||
f67539c2 TL |
5825 | return 0; |
5826 | } | |
5827 | ||
20effc67 | 5828 | int BlueStore::_init_alloc(std::map<uint64_t, uint64_t> *zone_adjustments) |
f67539c2 TL |
5829 | { |
5830 | int r = _create_alloc(); | |
5831 | if (r < 0) { | |
5832 | return r; | |
5833 | } | |
20effc67 | 5834 | ceph_assert(alloc != NULL); |
f67539c2 | 5835 | |
20effc67 | 5836 | #ifdef HAVE_LIBZBD |
f67539c2 | 5837 | if (bdev->is_smr()) { |
20effc67 TL |
5838 | auto a = dynamic_cast<ZonedAllocator*>(alloc); |
5839 | ceph_assert(a); | |
5840 | auto f = dynamic_cast<ZonedFreelistManager*>(fm); | |
5841 | ceph_assert(f); | |
5842 | vector<uint64_t> wp = bdev->get_zones(); | |
5843 | vector<zone_state_t> zones = f->get_zone_states(db); | |
5844 | ceph_assert(wp.size() == zones.size()); | |
5845 | ||
5846 | // reconcile zone state | |
5847 | auto num_zones = bdev->get_size() / zone_size; | |
5848 | for (unsigned i = first_sequential_zone; i < num_zones; ++i) { | |
5849 | ceph_assert(wp[i] >= i * zone_size); | |
5850 | ceph_assert(wp[i] <= (i + 1) * zone_size); // pos might be at start of next zone | |
5851 | uint64_t p = wp[i] - i * zone_size; | |
5852 | if (zones[i].write_pointer > p) { | |
5853 | derr << __func__ << " zone 0x" << std::hex << i | |
5854 | << " bluestore write pointer 0x" << zones[i].write_pointer | |
5855 | << " > device write pointer 0x" << p | |
5856 | << std::dec << " -- VERY SUSPICIOUS!" << dendl; | |
5857 | } else if (zones[i].write_pointer < p) { | |
5858 | // this is "normal" in that it can happen after any crash (if we have a | |
5859 | // write in flight but did not manage to commit the transaction) | |
5860 | auto delta = p - zones[i].write_pointer; | |
5861 | dout(1) << __func__ << " zone 0x" << std::hex << i | |
5862 | << " device write pointer 0x" << p | |
5863 | << " > bluestore pointer 0x" << zones[i].write_pointer | |
5864 | << ", advancing 0x" << delta << std::dec << dendl; | |
5865 | (*zone_adjustments)[zones[i].write_pointer] = delta; | |
5866 | zones[i].num_dead_bytes += delta; | |
5867 | zones[i].write_pointer = p; | |
5868 | } | |
5869 | } | |
5870 | ||
5871 | // start with conventional zone "free" (bluefs may adjust this when it starts up) | |
5872 | auto reserved = _get_ondisk_reserved(); | |
5873 | // for now we require a conventional zone | |
5874 | ceph_assert(bdev->get_conventional_region_size()); | |
5875 | ceph_assert(shared_alloc.a != alloc); // zoned allocator doesn't use conventional region | |
5876 | shared_alloc.a->init_add_free( | |
5877 | reserved, | |
5878 | p2align(bdev->get_conventional_region_size(), min_alloc_size) - reserved); | |
5879 | ||
5880 | // init sequential zone based on the device's write pointers | |
5881 | a->init_from_zone_pointers(std::move(zones)); | |
5882 | dout(1) << __func__ | |
5883 | << " loaded zone pointers: " | |
5884 | << std::hex | |
5885 | << ", allocator type " << alloc->get_type() | |
5886 | << ", capacity 0x" << alloc->get_capacity() | |
5887 | << ", block size 0x" << alloc->get_block_size() | |
5888 | << ", free 0x" << alloc->get_free() | |
5889 | << ", fragmentation " << alloc->get_fragmentation() | |
5890 | << std::dec << dendl; | |
5891 | ||
5892 | return 0; | |
f67539c2 | 5893 | } |
20effc67 | 5894 | #endif |
7c673cae FG |
5895 | |
5896 | uint64_t num = 0, bytes = 0; | |
20effc67 TL |
5897 | utime_t start_time = ceph_clock_now(); |
5898 | if (!fm->is_null_manager()) { | |
5899 | // This is the original path - loading allocation map from RocksDB and feeding into the allocator | |
5900 | dout(5) << __func__ << "::NCB::loading allocation from FM -> alloc" << dendl; | |
5901 | // initialize from freelist | |
5902 | fm->enumerate_reset(); | |
5903 | uint64_t offset, length; | |
5904 | while (fm->enumerate_next(db, &offset, &length)) { | |
5905 | alloc->init_add_free(offset, length); | |
5906 | ++num; | |
5907 | bytes += length; | |
5908 | } | |
5909 | fm->enumerate_reset(); | |
5910 | ||
5911 | utime_t duration = ceph_clock_now() - start_time; | |
5912 | dout(5) << __func__ << "::num_entries=" << num << " free_size=" << bytes << " alloc_size=" << | |
5913 | alloc->get_capacity() - bytes << " time=" << duration << " seconds" << dendl; | |
5914 | } else { | |
5915 | // This is the new path reading the allocation map from a flat bluefs file and feeding them into the allocator | |
7c673cae | 5916 | |
20effc67 TL |
5917 | if (!cct->_conf->bluestore_allocation_from_file) { |
5918 | derr << __func__ << "::NCB::cct->_conf->bluestore_allocation_from_file is set to FALSE with an active NULL-FM" << dendl; | |
5919 | derr << __func__ << "::NCB::Please change the value of bluestore_allocation_from_file to TRUE in your ceph.conf file" << dendl; | |
5920 | return -ENOTSUP; // Operation not supported | |
5921 | } | |
20effc67 TL |
5922 | if (restore_allocator(alloc, &num, &bytes) == 0) { |
5923 | dout(5) << __func__ << "::NCB::restore_allocator() completed successfully alloc=" << alloc << dendl; | |
5924 | } else { | |
5925 | // This must mean that we had an unplanned shutdown and didn't manage to destage the allocator | |
5926 | dout(0) << __func__ << "::NCB::restore_allocator() failed! Run Full Recovery from ONodes (might take a while) ..." << dendl; | |
5927 | // if failed must recover from on-disk ONode internal state | |
5928 | if (read_allocation_from_drive_on_startup() != 0) { | |
5929 | derr << __func__ << "::NCB::Failed Recovery" << dendl; | |
5930 | derr << __func__ << "::NCB::Ceph-OSD won't start, make sure your drives are connected and readable" << dendl; | |
5931 | derr << __func__ << "::NCB::If no HW fault is found, please report failure and consider redeploying OSD" << dendl; | |
5932 | return -ENOTRECOVERABLE; | |
5933 | } | |
5934 | } | |
5935 | } | |
f67539c2 TL |
5936 | dout(1) << __func__ |
5937 | << " loaded " << byte_u_t(bytes) << " in " << num << " extents" | |
5938 | << std::hex | |
20effc67 TL |
5939 | << ", allocator type " << alloc->get_type() |
5940 | << ", capacity 0x" << alloc->get_capacity() | |
5941 | << ", block size 0x" << alloc->get_block_size() | |
5942 | << ", free 0x" << alloc->get_free() | |
5943 | << ", fragmentation " << alloc->get_fragmentation() | |
f67539c2 | 5944 | << std::dec << dendl; |
1911f103 | 5945 | |
7c673cae FG |
5946 | return 0; |
5947 | } | |
5948 | ||
20effc67 TL |
5949 | void BlueStore::_post_init_alloc(const std::map<uint64_t, uint64_t>& zone_adjustments) |
5950 | { | |
39ae355f | 5951 | int r = 0; |
20effc67 | 5952 | #ifdef HAVE_LIBZBD |
39ae355f TL |
5953 | if (bdev->is_smr()) { |
5954 | if (zone_adjustments.empty()) { | |
5955 | return; | |
5956 | } | |
5957 | dout(1) << __func__ << " adjusting freelist based on device write pointers" << dendl; | |
5958 | auto f = dynamic_cast<ZonedFreelistManager*>(fm); | |
5959 | ceph_assert(f); | |
5960 | KeyValueDB::Transaction t = db->get_transaction(); | |
5961 | for (auto& i : zone_adjustments) { | |
5962 | // allocate AND release since this gap is now dead space | |
5963 | // note that the offset is imprecise, but only need to select the zone | |
5964 | f->allocate(i.first, i.second, t); | |
5965 | f->release(i.first, i.second, t); | |
5966 | } | |
5967 | r = db->submit_transaction_sync(t); | |
5968 | } else | |
20effc67 | 5969 | #endif |
39ae355f TL |
5970 | if (fm->is_null_manager()) { |
5971 | // Now that we load the allocation map we need to invalidate the file as new allocation won't be reflected | |
5972 | // Changes to the allocation map (alloc/release) are not updated inline and will only be stored on umount() | |
5973 | // This means that we should not use the existing file on failure case (unplanned shutdown) and must resort | |
5974 | // to recovery from RocksDB::ONodes | |
5975 | r = invalidate_allocation_file_on_bluefs(); | |
5976 | } | |
5977 | ceph_assert(r >= 0); | |
20effc67 TL |
5978 | } |
5979 | ||
7c673cae FG |
5980 | void BlueStore::_close_alloc() |
5981 | { | |
11fdf7f2 TL |
5982 | ceph_assert(bdev); |
5983 | bdev->discard_drain(); | |
5984 | ||
20effc67 TL |
5985 | ceph_assert(alloc); |
5986 | alloc->shutdown(); | |
5987 | delete alloc; | |
5988 | ||
f67539c2 | 5989 | ceph_assert(shared_alloc.a); |
20effc67 TL |
5990 | if (alloc != shared_alloc.a) { |
5991 | shared_alloc.a->shutdown(); | |
5992 | delete shared_alloc.a; | |
5993 | } | |
5994 | ||
f67539c2 | 5995 | shared_alloc.reset(); |
20effc67 | 5996 | alloc = nullptr; |
7c673cae FG |
5997 | } |
5998 | ||
5999 | int BlueStore::_open_fsid(bool create) | |
6000 | { | |
11fdf7f2 | 6001 | ceph_assert(fsid_fd < 0); |
91327a77 | 6002 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6003 | if (create) |
6004 | flags |= O_CREAT; | |
6005 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
6006 | if (fsid_fd < 0) { | |
6007 | int err = -errno; | |
6008 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
6009 | return err; | |
6010 | } | |
6011 | return 0; | |
6012 | } | |
6013 | ||
6014 | int BlueStore::_read_fsid(uuid_d *uuid) | |
6015 | { | |
6016 | char fsid_str[40]; | |
6017 | memset(fsid_str, 0, sizeof(fsid_str)); | |
6018 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
6019 | if (ret < 0) { | |
6020 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
6021 | return ret; | |
6022 | } | |
6023 | if (ret > 36) | |
6024 | fsid_str[36] = 0; | |
6025 | else | |
6026 | fsid_str[ret] = 0; | |
6027 | if (!uuid->parse(fsid_str)) { | |
6028 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
6029 | return -EINVAL; | |
6030 | } | |
6031 | return 0; | |
6032 | } | |
6033 | ||
6034 | int BlueStore::_write_fsid() | |
6035 | { | |
6036 | int r = ::ftruncate(fsid_fd, 0); | |
6037 | if (r < 0) { | |
6038 | r = -errno; | |
6039 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
6040 | return r; | |
6041 | } | |
6042 | string str = stringify(fsid) + "\n"; | |
6043 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
6044 | if (r < 0) { | |
6045 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
6046 | return r; | |
6047 | } | |
6048 | r = ::fsync(fsid_fd); | |
6049 | if (r < 0) { | |
6050 | r = -errno; | |
6051 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
6052 | return r; | |
6053 | } | |
6054 | return 0; | |
6055 | } | |
6056 | ||
6057 | void BlueStore::_close_fsid() | |
6058 | { | |
6059 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
6060 | fsid_fd = -1; | |
6061 | } | |
6062 | ||
6063 | int BlueStore::_lock_fsid() | |
6064 | { | |
6065 | struct flock l; | |
6066 | memset(&l, 0, sizeof(l)); | |
6067 | l.l_type = F_WRLCK; | |
6068 | l.l_whence = SEEK_SET; | |
6069 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
6070 | if (r < 0) { | |
6071 | int err = errno; | |
6072 | derr << __func__ << " failed to lock " << path << "/fsid" | |
6073 | << " (is another ceph-osd still running?)" | |
6074 | << cpp_strerror(err) << dendl; | |
6075 | return -err; | |
6076 | } | |
6077 | return 0; | |
6078 | } | |
6079 | ||
31f18b77 FG |
6080 | bool BlueStore::is_rotational() |
6081 | { | |
6082 | if (bdev) { | |
6083 | return bdev->is_rotational(); | |
6084 | } | |
6085 | ||
6086 | bool rotational = true; | |
6087 | int r = _open_path(); | |
6088 | if (r < 0) | |
6089 | goto out; | |
6090 | r = _open_fsid(false); | |
6091 | if (r < 0) | |
6092 | goto out_path; | |
6093 | r = _read_fsid(&fsid); | |
6094 | if (r < 0) | |
6095 | goto out_fsid; | |
6096 | r = _lock_fsid(); | |
6097 | if (r < 0) | |
6098 | goto out_fsid; | |
6099 | r = _open_bdev(false); | |
6100 | if (r < 0) | |
6101 | goto out_fsid; | |
6102 | rotational = bdev->is_rotational(); | |
6103 | _close_bdev(); | |
6104 | out_fsid: | |
6105 | _close_fsid(); | |
6106 | out_path: | |
6107 | _close_path(); | |
6108 | out: | |
6109 | return rotational; | |
6110 | } | |
6111 | ||
d2e6a577 FG |
6112 | bool BlueStore::is_journal_rotational() |
6113 | { | |
6114 | if (!bluefs) { | |
6115 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
6116 | << dendl; | |
6117 | return is_rotational(); | |
6118 | } | |
6119 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
6120 | return bluefs->wal_is_rotational(); | |
6121 | } | |
6122 | ||
1d09f67e TL |
6123 | bool BlueStore::is_db_rotational() |
6124 | { | |
6125 | if (!bluefs) { | |
6126 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
6127 | << dendl; | |
6128 | return is_rotational(); | |
6129 | } | |
6130 | dout(10) << __func__ << " " << (int)bluefs->db_is_rotational() << dendl; | |
6131 | return bluefs->db_is_rotational(); | |
6132 | } | |
6133 | ||
9f95a23c TL |
6134 | bool BlueStore::_use_rotational_settings() |
6135 | { | |
6136 | if (cct->_conf->bluestore_debug_enforce_settings == "hdd") { | |
6137 | return true; | |
6138 | } | |
6139 | if (cct->_conf->bluestore_debug_enforce_settings == "ssd") { | |
6140 | return false; | |
6141 | } | |
6142 | return bdev->is_rotational(); | |
6143 | } | |
6144 | ||
39ae355f TL |
6145 | bool BlueStore::is_statfs_recoverable() const |
6146 | { | |
6147 | // abuse fm for now | |
6148 | return has_null_manager(); | |
6149 | } | |
6150 | ||
7c673cae FG |
6151 | bool BlueStore::test_mount_in_use() |
6152 | { | |
6153 | // most error conditions mean the mount is not in use (e.g., because | |
6154 | // it doesn't exist). only if we fail to lock do we conclude it is | |
6155 | // in use. | |
6156 | bool ret = false; | |
6157 | int r = _open_path(); | |
6158 | if (r < 0) | |
6159 | return false; | |
6160 | r = _open_fsid(false); | |
6161 | if (r < 0) | |
6162 | goto out_path; | |
6163 | r = _lock_fsid(); | |
6164 | if (r < 0) | |
6165 | ret = true; // if we can't lock, it is in use | |
6166 | _close_fsid(); | |
6167 | out_path: | |
6168 | _close_path(); | |
6169 | return ret; | |
6170 | } | |
6171 | ||
11fdf7f2 | 6172 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
6173 | { |
6174 | int r; | |
11fdf7f2 | 6175 | bluefs = new BlueFS(cct); |
7c673cae | 6176 | |
11fdf7f2 TL |
6177 | string bfn; |
6178 | struct stat st; | |
6179 | ||
6180 | bfn = path + "/block.db"; | |
6181 | if (::stat(bfn.c_str(), &st) == 0) { | |
eafe8130 TL |
6182 | r = bluefs->add_block_device( |
6183 | BlueFS::BDEV_DB, bfn, | |
f67539c2 TL |
6184 | create && cct->_conf->bdev_enable_discard, |
6185 | SUPER_RESERVED); | |
7c673cae | 6186 | if (r < 0) { |
11fdf7f2 TL |
6187 | derr << __func__ << " add block device(" << bfn << ") returned: " |
6188 | << cpp_strerror(r) << dendl; | |
6189 | goto free_bluefs; | |
7c673cae | 6190 | } |
7c673cae | 6191 | |
11fdf7f2 TL |
6192 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
6193 | r = _check_or_set_bdev_label( | |
6194 | bfn, | |
6195 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
6196 | "bluefs db", create); | |
6197 | if (r < 0) { | |
6198 | derr << __func__ | |
6199 | << " check block device(" << bfn << ") label returned: " | |
6200 | << cpp_strerror(r) << dendl; | |
6201 | goto free_bluefs; | |
6202 | } | |
7c673cae | 6203 | } |
9f95a23c TL |
6204 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6205 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
6206 | } else { |
6207 | r = -errno; | |
6208 | if (::lstat(bfn.c_str(), &st) == -1) { | |
6209 | r = 0; | |
9f95a23c | 6210 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7c673cae | 6211 | } else { |
11fdf7f2 TL |
6212 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
6213 | << cpp_strerror(r) << dendl; | |
6214 | goto free_bluefs; | |
7c673cae FG |
6215 | } |
6216 | } | |
7c673cae | 6217 | |
11fdf7f2 TL |
6218 | // shared device |
6219 | bfn = path + "/block"; | |
6220 | // never trim here | |
9f95a23c | 6221 | r = bluefs->add_block_device(bluefs_layout.shared_bdev, bfn, false, |
f67539c2 TL |
6222 | 0, // no need to provide valid 'reserved' for shared dev |
6223 | &shared_alloc); | |
11fdf7f2 TL |
6224 | if (r < 0) { |
6225 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
6226 | << cpp_strerror(r) << dendl; | |
6227 | goto free_bluefs; | |
6228 | } | |
11fdf7f2 TL |
6229 | |
6230 | bfn = path + "/block.wal"; | |
6231 | if (::stat(bfn.c_str(), &st) == 0) { | |
6232 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
f67539c2 TL |
6233 | create && cct->_conf->bdev_enable_discard, |
6234 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
6235 | if (r < 0) { |
6236 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
6237 | << cpp_strerror(r) << dendl; | |
6238 | goto free_bluefs; | |
6239 | } | |
7c673cae | 6240 | |
11fdf7f2 TL |
6241 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
6242 | r = _check_or_set_bdev_label( | |
6243 | bfn, | |
6244 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
6245 | "bluefs wal", create); | |
7c673cae | 6246 | if (r < 0) { |
11fdf7f2 TL |
6247 | derr << __func__ << " check block device(" << bfn |
6248 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
6249 | goto free_bluefs; |
6250 | } | |
7c673cae FG |
6251 | } |
6252 | ||
9f95a23c | 6253 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6254 | } else { |
6255 | r = 0; | |
6256 | if (::lstat(bfn.c_str(), &st) != -1) { | |
6257 | r = -errno; | |
6258 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
6259 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
6260 | goto free_bluefs; |
6261 | } | |
11fdf7f2 TL |
6262 | } |
6263 | return 0; | |
7c673cae | 6264 | |
11fdf7f2 TL |
6265 | free_bluefs: |
6266 | ceph_assert(bluefs); | |
6267 | delete bluefs; | |
6268 | bluefs = NULL; | |
6269 | return r; | |
6270 | } | |
7c673cae | 6271 | |
f67539c2 | 6272 | int BlueStore::_open_bluefs(bool create, bool read_only) |
11fdf7f2 TL |
6273 | { |
6274 | int r = _minimal_open_bluefs(create); | |
6275 | if (r < 0) { | |
6276 | return r; | |
6277 | } | |
f67539c2 | 6278 | BlueFSVolumeSelector* vselector = nullptr; |
1e59de90 TL |
6279 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW || |
6280 | cct->_conf->bluestore_volume_selection_policy == "use_some_extra_enforced" || | |
6281 | cct->_conf->bluestore_volume_selection_policy == "fit_to_fast") { | |
9f95a23c TL |
6282 | |
6283 | string options = cct->_conf->bluestore_rocksdb_options; | |
cd265ab1 TL |
6284 | string options_annex = cct->_conf->bluestore_rocksdb_options_annex; |
6285 | if (!options_annex.empty()) { | |
6286 | if (!options.empty() && | |
6287 | *options.rbegin() != ',') { | |
6288 | options += ','; | |
6289 | } | |
6290 | options += options_annex; | |
6291 | } | |
9f95a23c TL |
6292 | |
6293 | rocksdb::Options rocks_opts; | |
f67539c2 | 6294 | r = RocksDBStore::ParseOptionsFromStringStatic( |
9f95a23c TL |
6295 | cct, |
6296 | options, | |
6297 | rocks_opts, | |
6298 | nullptr); | |
6299 | if (r < 0) { | |
6300 | return r; | |
6301 | } | |
f67539c2 TL |
6302 | if (cct->_conf->bluestore_volume_selection_policy == "fit_to_fast") { |
6303 | vselector = new FitToFastVolumeSelector( | |
9f95a23c TL |
6304 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, |
6305 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
f67539c2 TL |
6306 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100); |
6307 | } else { | |
6308 | double reserved_factor = cct->_conf->bluestore_volume_selection_reserved_factor; | |
6309 | vselector = | |
6310 | new RocksDBBlueFSVolumeSelector( | |
6311 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
6312 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
6313 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100, | |
6314 | 1024 * 1024 * 1024, //FIXME: set expected l0 size here | |
6315 | rocks_opts.max_bytes_for_level_base, | |
6316 | rocks_opts.max_bytes_for_level_multiplier, | |
6317 | reserved_factor, | |
6318 | cct->_conf->bluestore_volume_selection_reserved, | |
1e59de90 TL |
6319 | cct->_conf->bluestore_volume_selection_policy.find("use_some_extra") |
6320 | == 0); | |
f67539c2 | 6321 | } |
9f95a23c | 6322 | } |
11fdf7f2 | 6323 | if (create) { |
9f95a23c | 6324 | bluefs->mkfs(fsid, bluefs_layout); |
11fdf7f2 | 6325 | } |
9f95a23c | 6326 | bluefs->set_volume_selector(vselector); |
11fdf7f2 TL |
6327 | r = bluefs->mount(); |
6328 | if (r < 0) { | |
6329 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
6330 | } | |
9f95a23c | 6331 | ceph_assert_always(bluefs->maybe_verify_layout(bluefs_layout) == 0); |
11fdf7f2 TL |
6332 | return r; |
6333 | } | |
6334 | ||
20effc67 | 6335 | void BlueStore::_close_bluefs() |
11fdf7f2 | 6336 | { |
20effc67 | 6337 | bluefs->umount(db_was_opened_read_only); |
11fdf7f2 TL |
6338 | _minimal_close_bluefs(); |
6339 | } | |
6340 | ||
6341 | void BlueStore::_minimal_close_bluefs() | |
6342 | { | |
6343 | delete bluefs; | |
6344 | bluefs = NULL; | |
6345 | } | |
6346 | ||
6347 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
6348 | { | |
6349 | if (create) { | |
6350 | *ret = cct->_conf->bluestore_bluefs; | |
6351 | } else { | |
6352 | string s; | |
6353 | int r = read_meta("bluefs", &s); | |
6354 | if (r < 0) { | |
6355 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
6356 | return -EIO; | |
6357 | } | |
6358 | if (s == "1") { | |
6359 | *ret = true; | |
6360 | } else if (s == "0") { | |
6361 | *ret = false; | |
31f18b77 | 6362 | } else { |
11fdf7f2 TL |
6363 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
6364 | << dendl; | |
6365 | return -EIO; | |
6366 | } | |
6367 | } | |
6368 | return 0; | |
6369 | } | |
6370 | ||
6371 | /* | |
6372 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
6373 | * in the proper order | |
6374 | */ | |
f67539c2 | 6375 | int BlueStore::_open_db_and_around(bool read_only, bool to_repair) |
11fdf7f2 | 6376 | { |
20effc67 | 6377 | dout(5) << __func__ << "::NCB::read_only=" << read_only << ", to_repair=" << to_repair << dendl; |
f67539c2 TL |
6378 | { |
6379 | string type; | |
6380 | int r = read_meta("type", &type); | |
6381 | if (r < 0) { | |
6382 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
6383 | << dendl; | |
11fdf7f2 | 6384 | return r; |
f67539c2 | 6385 | } |
11fdf7f2 | 6386 | |
f67539c2 TL |
6387 | if (type != "bluestore") { |
6388 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6389 | return -EIO; | |
11fdf7f2 | 6390 | } |
f67539c2 | 6391 | } |
11fdf7f2 | 6392 | |
20effc67 TL |
6393 | // SMR devices may require a freelist adjustment, but that can only happen after |
6394 | // the db is read-write. we'll stash pending changes here. | |
6395 | std::map<uint64_t, uint64_t> zone_adjustments; | |
6396 | ||
f67539c2 TL |
6397 | int r = _open_path(); |
6398 | if (r < 0) | |
6399 | return r; | |
6400 | r = _open_fsid(false); | |
6401 | if (r < 0) | |
6402 | goto out_path; | |
11fdf7f2 | 6403 | |
f67539c2 TL |
6404 | r = _read_fsid(&fsid); |
6405 | if (r < 0) | |
6406 | goto out_fsid; | |
11fdf7f2 | 6407 | |
f67539c2 TL |
6408 | r = _lock_fsid(); |
6409 | if (r < 0) | |
6410 | goto out_fsid; | |
11fdf7f2 | 6411 | |
f67539c2 TL |
6412 | r = _open_bdev(false); |
6413 | if (r < 0) | |
6414 | goto out_fsid; | |
7c673cae | 6415 | |
20effc67 TL |
6416 | // GBH: can probably skip open_db step in REad-Only mode when operating in NULL-FM mode |
6417 | // (might need to open if failed to restore from file) | |
6418 | ||
f67539c2 TL |
6419 | // open in read-only first to read FM list and init allocator |
6420 | // as they might be needed for some BlueFS procedures | |
6421 | r = _open_db(false, false, true); | |
6422 | if (r < 0) | |
6423 | goto out_bdev; | |
11fdf7f2 | 6424 | |
f67539c2 TL |
6425 | r = _open_super_meta(); |
6426 | if (r < 0) { | |
6427 | goto out_db; | |
6428 | } | |
6429 | ||
39ae355f | 6430 | r = _open_fm(nullptr, true, false); |
f67539c2 TL |
6431 | if (r < 0) |
6432 | goto out_db; | |
6433 | ||
20effc67 | 6434 | r = _init_alloc(&zone_adjustments); |
f67539c2 TL |
6435 | if (r < 0) |
6436 | goto out_fm; | |
6437 | ||
6438 | // Re-open in the proper mode(s). | |
6439 | ||
6440 | // Can't simply bypass second open for read-only mode as we need to | |
6441 | // load allocated extents from bluefs into allocator. | |
6442 | // And now it's time to do that | |
6443 | // | |
20effc67 | 6444 | _close_db(); |
f67539c2 TL |
6445 | r = _open_db(false, to_repair, read_only); |
6446 | if (r < 0) { | |
6447 | goto out_alloc; | |
11fdf7f2 | 6448 | } |
20effc67 | 6449 | |
39ae355f | 6450 | if (!read_only) { |
20effc67 TL |
6451 | _post_init_alloc(zone_adjustments); |
6452 | } | |
6453 | ||
6454 | // when function is called in repair mode (to_repair=true) we skip db->open()/create() | |
6455 | // we can't change bluestore allocation so no need to invlidate allocation-file | |
6456 | if (fm->is_null_manager() && !read_only && !to_repair) { | |
6457 | // Now that we load the allocation map we need to invalidate the file as new allocation won't be reflected | |
6458 | // Changes to the allocation map (alloc/release) are not updated inline and will only be stored on umount() | |
6459 | // This means that we should not use the existing file on failure case (unplanned shutdown) and must resort | |
6460 | // to recovery from RocksDB::ONodes | |
6461 | r = invalidate_allocation_file_on_bluefs(); | |
6462 | if (r != 0) { | |
6463 | derr << __func__ << "::NCB::invalidate_allocation_file_on_bluefs() failed!" << dendl; | |
6464 | goto out_alloc; | |
6465 | } | |
6466 | } | |
6467 | ||
6468 | // when function is called in repair mode (to_repair=true) we skip db->open()/create() | |
1d09f67e | 6469 | if (!is_db_rotational() && !read_only && !to_repair && cct->_conf->bluestore_allocation_from_file |
20effc67 TL |
6470 | #ifdef HAVE_LIBZBD |
6471 | && !bdev->is_smr() | |
6472 | #endif | |
6473 | ) { | |
6474 | dout(5) << __func__ << "::NCB::Commit to Null-Manager" << dendl; | |
6475 | commit_to_null_manager(); | |
6476 | need_to_destage_allocation_file = true; | |
6477 | dout(10) << __func__ << "::NCB::need_to_destage_allocation_file was set" << dendl; | |
6478 | } | |
6479 | ||
11fdf7f2 TL |
6480 | return 0; |
6481 | ||
f67539c2 TL |
6482 | out_alloc: |
6483 | _close_alloc(); | |
6484 | out_fm: | |
11fdf7f2 TL |
6485 | _close_fm(); |
6486 | out_db: | |
20effc67 | 6487 | _close_db(); |
f67539c2 TL |
6488 | out_bdev: |
6489 | _close_bdev(); | |
6490 | out_fsid: | |
6491 | _close_fsid(); | |
6492 | out_path: | |
6493 | _close_path(); | |
11fdf7f2 TL |
6494 | return r; |
6495 | } | |
6496 | ||
20effc67 | 6497 | void BlueStore::_close_db_and_around() |
11fdf7f2 | 6498 | { |
20effc67 TL |
6499 | if (db) { |
6500 | _close_db(); | |
6501 | } | |
39ae355f TL |
6502 | _close_around_db(); |
6503 | } | |
6504 | ||
6505 | void BlueStore::_close_around_db() | |
6506 | { | |
20effc67 TL |
6507 | if (bluefs) { |
6508 | _close_bluefs(); | |
6509 | } | |
f67539c2 TL |
6510 | _close_fm(); |
6511 | _close_alloc(); | |
6512 | _close_bdev(); | |
6513 | _close_fsid(); | |
6514 | _close_path(); | |
6515 | } | |
6516 | ||
6517 | int BlueStore::open_db_environment(KeyValueDB **pdb, bool to_repair) | |
6518 | { | |
6519 | _kv_only = true; | |
6520 | int r = _open_db_and_around(false, to_repair); | |
6521 | if (r == 0) { | |
6522 | *pdb = db; | |
11fdf7f2 | 6523 | } else { |
f67539c2 | 6524 | *pdb = nullptr; |
11fdf7f2 | 6525 | } |
f67539c2 | 6526 | return r; |
11fdf7f2 TL |
6527 | } |
6528 | ||
f67539c2 | 6529 | int BlueStore::close_db_environment() |
11fdf7f2 | 6530 | { |
39ae355f TL |
6531 | if (db) { |
6532 | delete db; | |
6533 | db = nullptr; | |
6534 | } | |
6535 | _close_around_db(); | |
f67539c2 | 6536 | return 0; |
11fdf7f2 TL |
6537 | } |
6538 | ||
20effc67 TL |
6539 | /* gets access to bluefs supporting RocksDB */ |
6540 | BlueFS* BlueStore::get_bluefs() { | |
6541 | return bluefs; | |
6542 | } | |
6543 | ||
f67539c2 TL |
6544 | int BlueStore::_prepare_db_environment(bool create, bool read_only, |
6545 | std::string* _fn, std::string* _kv_backend) | |
11fdf7f2 TL |
6546 | { |
6547 | int r; | |
6548 | ceph_assert(!db); | |
f67539c2 TL |
6549 | std::string& fn=*_fn; |
6550 | std::string& kv_backend=*_kv_backend; | |
6551 | fn = path + "/db"; | |
11fdf7f2 TL |
6552 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); |
6553 | ||
11fdf7f2 TL |
6554 | if (create) { |
6555 | kv_backend = cct->_conf->bluestore_kvbackend; | |
6556 | } else { | |
6557 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 6558 | if (r < 0) { |
11fdf7f2 TL |
6559 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
6560 | return -EIO; | |
6561 | } | |
6562 | } | |
6563 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
6564 | ||
6565 | bool do_bluefs; | |
6566 | r = _is_bluefs(create, &do_bluefs); | |
6567 | if (r < 0) { | |
6568 | return r; | |
6569 | } | |
6570 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
6571 | ||
6572 | map<string,string> kv_options; | |
6573 | // force separate wal dir for all new deployments. | |
6574 | kv_options["separate_wal_dir"] = 1; | |
6575 | rocksdb::Env *env = NULL; | |
6576 | if (do_bluefs) { | |
6577 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
6578 | if (kv_backend != "rocksdb") { | |
6579 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
6580 | return -EINVAL; | |
7c673cae | 6581 | } |
11fdf7f2 | 6582 | |
f67539c2 | 6583 | r = _open_bluefs(create, read_only); |
11fdf7f2 TL |
6584 | if (r < 0) { |
6585 | return r; | |
6586 | } | |
11fdf7f2 | 6587 | |
7c673cae | 6588 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
9f95a23c TL |
6589 | rocksdb::Env* a = new BlueRocksEnv(bluefs); |
6590 | rocksdb::Env* b = rocksdb::Env::Default(); | |
7c673cae | 6591 | if (create) { |
9f95a23c TL |
6592 | string cmd = "rm -rf " + path + "/db " + |
6593 | path + "/db.slow " + | |
6594 | path + "/db.wal"; | |
6595 | int r = system(cmd.c_str()); | |
6596 | (void)r; | |
7c673cae FG |
6597 | } |
6598 | env = new rocksdb::EnvMirror(b, a, false, true); | |
1911f103 | 6599 | } else { |
7c673cae FG |
6600 | env = new BlueRocksEnv(bluefs); |
6601 | ||
6602 | // simplify the dir names, too, as "seen" by rocksdb | |
6603 | fn = "db"; | |
6604 | } | |
9f95a23c TL |
6605 | BlueFSVolumeSelector::paths paths; |
6606 | bluefs->get_vselector_paths(fn, paths); | |
7c673cae | 6607 | |
522d829b | 6608 | { |
7c673cae | 6609 | ostringstream db_paths; |
9f95a23c TL |
6610 | bool first = true; |
6611 | for (auto& p : paths) { | |
6612 | if (!first) { | |
6613 | db_paths << " "; | |
6614 | } | |
6615 | first = false; | |
6616 | db_paths << p.first << "," << p.second; | |
6617 | ||
6618 | } | |
11fdf7f2 | 6619 | kv_options["db_paths"] = db_paths.str(); |
9f95a23c | 6620 | dout(1) << __func__ << " set db_paths to " << db_paths.str() << dendl; |
7c673cae FG |
6621 | } |
6622 | ||
6623 | if (create) { | |
9f95a23c TL |
6624 | for (auto& p : paths) { |
6625 | env->CreateDir(p.first); | |
6626 | } | |
6627 | // Selectors don't provide wal path so far hence create explicitly | |
11fdf7f2 | 6628 | env->CreateDir(fn + ".wal"); |
11fdf7f2 TL |
6629 | } else { |
6630 | std::vector<std::string> res; | |
6631 | // check for dir presence | |
6632 | auto r = env->GetChildren(fn+".wal", &res); | |
6633 | if (r.IsNotFound()) { | |
6634 | kv_options.erase("separate_wal_dir"); | |
6635 | } | |
7c673cae | 6636 | } |
11fdf7f2 TL |
6637 | } else { |
6638 | string walfn = path + "/db.wal"; | |
7c673cae | 6639 | |
11fdf7f2 TL |
6640 | if (create) { |
6641 | int r = ::mkdir(fn.c_str(), 0755); | |
6642 | if (r < 0) | |
6643 | r = -errno; | |
6644 | if (r < 0 && r != -EEXIST) { | |
6645 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
6646 | << dendl; | |
6647 | return r; | |
6648 | } | |
6649 | ||
6650 | // wal_dir, too! | |
7c673cae FG |
6651 | r = ::mkdir(walfn.c_str(), 0755); |
6652 | if (r < 0) | |
6653 | r = -errno; | |
6654 | if (r < 0 && r != -EEXIST) { | |
6655 | derr << __func__ << " failed to create " << walfn | |
6656 | << ": " << cpp_strerror(r) | |
6657 | << dendl; | |
6658 | return r; | |
6659 | } | |
11fdf7f2 TL |
6660 | } else { |
6661 | struct stat st; | |
6662 | r = ::stat(walfn.c_str(), &st); | |
6663 | if (r < 0 && errno == ENOENT) { | |
6664 | kv_options.erase("separate_wal_dir"); | |
6665 | } | |
7c673cae FG |
6666 | } |
6667 | } | |
6668 | ||
91327a77 | 6669 | |
7c673cae FG |
6670 | db = KeyValueDB::create(cct, |
6671 | kv_backend, | |
6672 | fn, | |
11fdf7f2 | 6673 | kv_options, |
7c673cae FG |
6674 | static_cast<void*>(env)); |
6675 | if (!db) { | |
6676 | derr << __func__ << " error creating db" << dendl; | |
6677 | if (bluefs) { | |
20effc67 | 6678 | _close_bluefs(); |
7c673cae FG |
6679 | } |
6680 | // delete env manually here since we can't depend on db to do this | |
6681 | // under this case | |
6682 | delete env; | |
6683 | env = NULL; | |
6684 | return -EIO; | |
6685 | } | |
6686 | ||
f67539c2 | 6687 | FreelistManager::setup_merge_operators(db, freelist_type); |
7c673cae | 6688 | db->set_merge_operator(PREFIX_STAT, merge_op); |
91327a77 | 6689 | db->set_cache_size(cache_kv_ratio * cache_size); |
f67539c2 TL |
6690 | return 0; |
6691 | } | |
31f18b77 | 6692 | |
f67539c2 TL |
6693 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) |
6694 | { | |
6695 | int r; | |
6696 | ceph_assert(!(create && read_only)); | |
6697 | string options; | |
6698 | string options_annex; | |
6699 | stringstream err; | |
6700 | string kv_dir_fn; | |
6701 | string kv_backend; | |
6702 | std::string sharding_def; | |
20effc67 TL |
6703 | // prevent write attempts to BlueFS in case we failed before BlueFS was opened |
6704 | db_was_opened_read_only = true; | |
f67539c2 TL |
6705 | r = _prepare_db_environment(create, read_only, &kv_dir_fn, &kv_backend); |
6706 | if (r < 0) { | |
6707 | derr << __func__ << " failed to prepare db environment: " << err.str() << dendl; | |
6708 | return -EIO; | |
6709 | } | |
20effc67 TL |
6710 | // if reached here then BlueFS is already opened |
6711 | db_was_opened_read_only = read_only; | |
6712 | dout(10) << __func__ << "::db_was_opened_read_only was set to " << read_only << dendl; | |
11fdf7f2 | 6713 | if (kv_backend == "rocksdb") { |
7c673cae | 6714 | options = cct->_conf->bluestore_rocksdb_options; |
cd265ab1 TL |
6715 | options_annex = cct->_conf->bluestore_rocksdb_options_annex; |
6716 | if (!options_annex.empty()) { | |
6717 | if (!options.empty() && | |
6718 | *options.rbegin() != ',') { | |
6719 | options += ','; | |
6720 | } | |
6721 | options += options_annex; | |
6722 | } | |
11fdf7f2 | 6723 | |
f67539c2 TL |
6724 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { |
6725 | sharding_def = cct->_conf.get_val<std::string>("bluestore_rocksdb_cfs"); | |
11fdf7f2 TL |
6726 | } |
6727 | } | |
6728 | ||
7c673cae | 6729 | db->init(options); |
11fdf7f2 TL |
6730 | if (to_repair_db) |
6731 | return 0; | |
6732 | if (create) { | |
f67539c2 | 6733 | r = db->create_and_open(err, sharding_def); |
11fdf7f2 TL |
6734 | } else { |
6735 | // we pass in cf list here, but it is only used if the db already has | |
6736 | // column families created. | |
6737 | r = read_only ? | |
f67539c2 TL |
6738 | db->open_read_only(err, sharding_def) : |
6739 | db->open(err, sharding_def); | |
11fdf7f2 | 6740 | } |
7c673cae FG |
6741 | if (r) { |
6742 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
20effc67 | 6743 | _close_db(); |
7c673cae FG |
6744 | return -EIO; |
6745 | } | |
6746 | dout(1) << __func__ << " opened " << kv_backend | |
f67539c2 | 6747 | << " path " << kv_dir_fn << " options " << options << dendl; |
7c673cae | 6748 | return 0; |
7c673cae FG |
6749 | } |
6750 | ||
39ae355f | 6751 | void BlueStore::_close_db() |
7c673cae | 6752 | { |
39ae355f TL |
6753 | dout(10) << __func__ << ":read_only=" << db_was_opened_read_only |
6754 | << " fm=" << fm | |
6755 | << " destage_alloc_file=" << need_to_destage_allocation_file | |
6756 | << " per_pool=" << per_pool_stat_collection | |
6757 | << " pool stats=" << osd_pools.size() | |
6758 | << dendl; | |
6759 | bool do_destage = !db_was_opened_read_only && need_to_destage_allocation_file; | |
6760 | if (do_destage && is_statfs_recoverable()) { | |
6761 | auto t = db->get_transaction(); | |
6762 | store_statfs_t s; | |
6763 | if (per_pool_stat_collection) { | |
6764 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE); | |
6765 | uint64_t pool_id; | |
6766 | for (it->upper_bound(string()); it->valid(); it->next()) { | |
6767 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6768 | if (r >= 0) { | |
6769 | dout(10) << __func__ << " wiping statfs for: " << pool_id << dendl; | |
6770 | } else { | |
6771 | derr << __func__ << " wiping invalid statfs key: " << it->key() << dendl; | |
6772 | } | |
6773 | t->rmkey(PREFIX_STAT, it->key()); | |
6774 | } | |
6775 | ||
6776 | std::lock_guard l(vstatfs_lock); | |
6777 | for(auto &p : osd_pools) { | |
6778 | string key; | |
6779 | get_pool_stat_key(p.first, &key); | |
6780 | bufferlist bl; | |
6781 | if (!p.second.is_empty()) { | |
6782 | p.second.encode(bl); | |
6783 | p.second.publish(&s); | |
6784 | t->set(PREFIX_STAT, key, bl); | |
6785 | dout(10) << __func__ << " persisting: " | |
6786 | << p.first << "->" << s | |
6787 | << dendl; | |
6788 | } | |
6789 | } | |
6790 | } else { | |
6791 | bufferlist bl; | |
6792 | { | |
6793 | std::lock_guard l(vstatfs_lock); | |
6794 | vstatfs.encode(bl); | |
6795 | vstatfs.publish(&s); | |
6796 | } | |
6797 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
6798 | dout(10) << __func__ << "persisting: " << s << dendl; | |
6799 | } | |
6800 | int r = db->submit_transaction_sync(t); | |
6801 | dout(10) << __func__ << " statfs persisted." << dendl; | |
6802 | ceph_assert(r >= 0); | |
6803 | } | |
11fdf7f2 | 6804 | ceph_assert(db); |
7c673cae | 6805 | delete db; |
20effc67 | 6806 | db = nullptr; |
20effc67 | 6807 | |
39ae355f | 6808 | if (do_destage && fm && fm->is_null_manager()) { |
20effc67 TL |
6809 | int ret = store_allocator(alloc); |
6810 | if (ret != 0) { | |
6811 | derr << __func__ << "::NCB::store_allocator() failed (continue with bitmapFreelistManager)" << dendl; | |
6812 | } | |
6813 | } | |
6814 | ||
7c673cae | 6815 | if (bluefs) { |
20effc67 | 6816 | _close_bluefs(); |
7c673cae FG |
6817 | } |
6818 | } | |
6819 | ||
11fdf7f2 | 6820 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 6821 | { |
11fdf7f2 TL |
6822 | auto dump_interval = |
6823 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
6824 | if (dump_interval > 0 && | |
6825 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
f67539c2 | 6826 | shared_alloc.a->dump(); |
11fdf7f2 TL |
6827 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); |
6828 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 6829 | } |
11fdf7f2 | 6830 | } |
7c673cae | 6831 | |
eafe8130 | 6832 | int BlueStore::_open_collections() |
7c673cae | 6833 | { |
20effc67 TL |
6834 | if (!coll_map.empty()) { |
6835 | // could be opened from another path | |
6836 | dout(20) << __func__ << "::NCB::collections are already opened, nothing to do" << dendl; | |
6837 | return 0; | |
6838 | } | |
6839 | ||
28e407b8 | 6840 | dout(10) << __func__ << dendl; |
eafe8130 | 6841 | collections_had_errors = false; |
7c673cae | 6842 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
20effc67 | 6843 | size_t load_cnt = 0; |
7c673cae FG |
6844 | for (it->upper_bound(string()); |
6845 | it->valid(); | |
6846 | it->next()) { | |
6847 | coll_t cid; | |
6848 | if (cid.parse(it->key())) { | |
9f95a23c | 6849 | auto c = ceph::make_ref<Collection>( |
7c673cae | 6850 | this, |
9f95a23c TL |
6851 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
6852 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
6853 | cid); | |
7c673cae | 6854 | bufferlist bl = it->value(); |
11fdf7f2 | 6855 | auto p = bl.cbegin(); |
7c673cae | 6856 | try { |
11fdf7f2 | 6857 | decode(c->cnode, p); |
f67539c2 | 6858 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
6859 | derr << __func__ << " failed to decode cnode, key:" |
6860 | << pretty_binary_string(it->key()) << dendl; | |
6861 | return -EIO; | |
6862 | } | |
28e407b8 AA |
6863 | dout(20) << __func__ << " opened " << cid << " " << c |
6864 | << " " << c->cnode << dendl; | |
11fdf7f2 | 6865 | _osr_attach(c.get()); |
7c673cae | 6866 | coll_map[cid] = c; |
20effc67 | 6867 | load_cnt++; |
7c673cae FG |
6868 | } else { |
6869 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
eafe8130 | 6870 | collections_had_errors = true; |
7c673cae FG |
6871 | } |
6872 | } | |
20effc67 TL |
6873 | dout(10) << __func__ << " collections loaded: " << load_cnt |
6874 | << dendl; | |
7c673cae FG |
6875 | return 0; |
6876 | } | |
6877 | ||
eafe8130 TL |
6878 | void BlueStore::_fsck_collections(int64_t* errors) |
6879 | { | |
6880 | if (collections_had_errors) { | |
6881 | dout(10) << __func__ << dendl; | |
f67539c2 | 6882 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
6883 | for (it->upper_bound(string()); |
6884 | it->valid(); | |
6885 | it->next()) { | |
6886 | coll_t cid; | |
6887 | if (!cid.parse(it->key())) { | |
6888 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
6889 | if (errors) { | |
6890 | (*errors)++; | |
6891 | } | |
6892 | } | |
6893 | } | |
6894 | } | |
6895 | } | |
6896 | ||
9f95a23c TL |
6897 | void BlueStore::_set_per_pool_omap() |
6898 | { | |
f67539c2 | 6899 | per_pool_omap = OMAP_BULK; |
9f95a23c TL |
6900 | bufferlist bl; |
6901 | db->get(PREFIX_SUPER, "per_pool_omap", &bl); | |
6902 | if (bl.length()) { | |
f67539c2 TL |
6903 | auto s = bl.to_str(); |
6904 | if (s == stringify(OMAP_PER_POOL)) { | |
6905 | per_pool_omap = OMAP_PER_POOL; | |
a4b75251 | 6906 | } else if (s == stringify(OMAP_PER_PG)) { |
f67539c2 | 6907 | per_pool_omap = OMAP_PER_PG; |
a4b75251 TL |
6908 | } else { |
6909 | ceph_assert(s == stringify(OMAP_BULK)); | |
f67539c2 TL |
6910 | } |
6911 | dout(10) << __func__ << " per_pool_omap = " << per_pool_omap << dendl; | |
9f95a23c TL |
6912 | } else { |
6913 | dout(10) << __func__ << " per_pool_omap not present" << dendl; | |
6914 | } | |
f67539c2 | 6915 | _check_no_per_pg_or_pool_omap_alert(); |
9f95a23c TL |
6916 | } |
6917 | ||
224ce89b | 6918 | void BlueStore::_open_statfs() |
31f18b77 | 6919 | { |
11fdf7f2 TL |
6920 | osd_pools.clear(); |
6921 | vstatfs.reset(); | |
6922 | ||
31f18b77 | 6923 | bufferlist bl; |
11fdf7f2 | 6924 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 6925 | if (r >= 0) { |
11fdf7f2 | 6926 | per_pool_stat_collection = false; |
31f18b77 | 6927 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 6928 | auto it = bl.cbegin(); |
31f18b77 | 6929 | vstatfs.decode(it); |
11fdf7f2 | 6930 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 6931 | } else { |
31f18b77 FG |
6932 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
6933 | } | |
81eedcae | 6934 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
6935 | } else { |
6936 | per_pool_stat_collection = true; | |
6937 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
f67539c2 | 6938 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 TL |
6939 | for (it->upper_bound(string()); |
6940 | it->valid(); | |
6941 | it->next()) { | |
6942 | ||
6943 | uint64_t pool_id; | |
6944 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6945 | ceph_assert(r == 0); | |
6946 | ||
6947 | bufferlist bl; | |
6948 | bl = it->value(); | |
6949 | auto p = bl.cbegin(); | |
6950 | auto& st = osd_pools[pool_id]; | |
6951 | try { | |
6952 | st.decode(p); | |
6953 | vstatfs += st; | |
6954 | ||
39ae355f TL |
6955 | dout(10) << __func__ << " pool " << std::hex << pool_id |
6956 | << " statfs(hex) " << st | |
6957 | << std::dec << dendl; | |
f67539c2 | 6958 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
6959 | derr << __func__ << " failed to decode pool stats, key:" |
6960 | << pretty_binary_string(it->key()) << dendl; | |
6961 | } | |
6962 | } | |
31f18b77 | 6963 | } |
39ae355f TL |
6964 | dout(10) << __func__ << " statfs " << std::hex |
6965 | << vstatfs << std::dec << dendl; | |
11fdf7f2 | 6966 | |
31f18b77 FG |
6967 | } |
6968 | ||
7c673cae FG |
6969 | int BlueStore::_setup_block_symlink_or_file( |
6970 | string name, | |
6971 | string epath, | |
6972 | uint64_t size, | |
6973 | bool create) | |
6974 | { | |
6975 | dout(20) << __func__ << " name " << name << " path " << epath | |
6976 | << " size " << size << " create=" << (int)create << dendl; | |
6977 | int r = 0; | |
91327a77 | 6978 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6979 | if (create) |
6980 | flags |= O_CREAT; | |
6981 | if (epath.length()) { | |
6982 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
6983 | if (r < 0) { | |
6984 | r = -errno; | |
6985 | derr << __func__ << " failed to create " << name << " symlink to " | |
6986 | << epath << ": " << cpp_strerror(r) << dendl; | |
6987 | return r; | |
6988 | } | |
6989 | ||
6990 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
6991 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
6992 | if (fd < 0) { | |
6993 | r = -errno; | |
6994 | derr << __func__ << " failed to open " << epath << " file: " | |
6995 | << cpp_strerror(r) << dendl; | |
6996 | return r; | |
6997 | } | |
11fdf7f2 | 6998 | // write the Transport ID of the NVMe device |
1e59de90 | 6999 | // a transport id for PCIe looks like: "trtype:PCIe traddr:0000:02:00.0" |
11fdf7f2 TL |
7000 | // where "0000:02:00.0" is the selector of a PCI device, see |
7001 | // the first column of "lspci -mm -n -D" | |
1e59de90 TL |
7002 | // a transport id for tcp looks like: "trype:TCP adrfam:IPv4 traddr:172.31.89.152 trsvcid:4420" |
7003 | string trid = epath.substr(strlen(SPDK_PREFIX)); | |
11fdf7f2 TL |
7004 | r = ::write(fd, trid.c_str(), trid.size()); |
7005 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
7006 | dout(1) << __func__ << " created " << name << " symlink to " |
7007 | << epath << dendl; | |
7008 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
7009 | } | |
7010 | } | |
7011 | if (size) { | |
7012 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
7013 | if (fd >= 0) { | |
7014 | // block file is present | |
7015 | struct stat st; | |
7016 | int r = ::fstat(fd, &st); | |
7017 | if (r == 0 && | |
7018 | S_ISREG(st.st_mode) && // if it is a regular file | |
7019 | st.st_size == 0) { // and is 0 bytes | |
7020 | r = ::ftruncate(fd, size); | |
7021 | if (r < 0) { | |
7022 | r = -errno; | |
7023 | derr << __func__ << " failed to resize " << name << " file to " | |
7024 | << size << ": " << cpp_strerror(r) << dendl; | |
7025 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
7026 | return r; | |
7027 | } | |
7028 | ||
7029 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
7030 | r = ::ceph_posix_fallocate(fd, 0, size); |
7031 | if (r > 0) { | |
7c673cae FG |
7032 | derr << __func__ << " failed to prefallocate " << name << " file to " |
7033 | << size << ": " << cpp_strerror(r) << dendl; | |
7034 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
7035 | return -r; | |
7036 | } | |
7c673cae FG |
7037 | } |
7038 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 7039 | << byte_u_t(size) << dendl; |
7c673cae FG |
7040 | } |
7041 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
7042 | } else { | |
7043 | int r = -errno; | |
7044 | if (r != -ENOENT) { | |
7045 | derr << __func__ << " failed to open " << name << " file: " | |
7046 | << cpp_strerror(r) << dendl; | |
7047 | return r; | |
7048 | } | |
7049 | } | |
7050 | } | |
7051 | return 0; | |
7052 | } | |
7053 | ||
7054 | int BlueStore::mkfs() | |
7055 | { | |
7056 | dout(1) << __func__ << " path " << path << dendl; | |
7057 | int r; | |
7058 | uuid_d old_fsid; | |
f67539c2 | 7059 | uint64_t reserved; |
eafe8130 TL |
7060 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
7061 | derr << __func__ << " osd_max_object_size " | |
7062 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
7063 | << OBJECT_MAX_SIZE << dendl; | |
7064 | return -EINVAL; | |
7065 | } | |
7066 | ||
7c673cae FG |
7067 | { |
7068 | string done; | |
7069 | r = read_meta("mkfs_done", &done); | |
7070 | if (r == 0) { | |
7071 | dout(1) << __func__ << " already created" << dendl; | |
7072 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
7073 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
7074 | if (r < 0) { | |
7075 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
7076 | << dendl; | |
7077 | return r; | |
7078 | } | |
7079 | if (r > 0) { | |
7080 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
7081 | r = -EIO; | |
7082 | } | |
7083 | } | |
7084 | return r; // idempotent | |
7085 | } | |
7086 | } | |
7087 | ||
7088 | { | |
7089 | string type; | |
7090 | r = read_meta("type", &type); | |
7091 | if (r == 0) { | |
7092 | if (type != "bluestore") { | |
7093 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
7094 | return -EIO; | |
7095 | } | |
7096 | } else { | |
7097 | r = write_meta("type", "bluestore"); | |
7098 | if (r < 0) | |
7099 | return r; | |
7100 | } | |
7101 | } | |
7102 | ||
7c673cae FG |
7103 | r = _open_path(); |
7104 | if (r < 0) | |
7105 | return r; | |
7106 | ||
7107 | r = _open_fsid(true); | |
7108 | if (r < 0) | |
7109 | goto out_path_fd; | |
7110 | ||
7111 | r = _lock_fsid(); | |
7112 | if (r < 0) | |
7113 | goto out_close_fsid; | |
7114 | ||
7115 | r = _read_fsid(&old_fsid); | |
7116 | if (r < 0 || old_fsid.is_zero()) { | |
7117 | if (fsid.is_zero()) { | |
7118 | fsid.generate_random(); | |
7119 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
7120 | } else { | |
7121 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
7122 | } | |
7123 | // we'll write it later. | |
7124 | } else { | |
7125 | if (!fsid.is_zero() && fsid != old_fsid) { | |
7126 | derr << __func__ << " on-disk fsid " << old_fsid | |
7127 | << " != provided " << fsid << dendl; | |
7128 | r = -EINVAL; | |
7129 | goto out_close_fsid; | |
7130 | } | |
7131 | fsid = old_fsid; | |
7132 | } | |
7133 | ||
7134 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
7135 | cct->_conf->bluestore_block_size, | |
7136 | cct->_conf->bluestore_block_create); | |
7137 | if (r < 0) | |
7138 | goto out_close_fsid; | |
7139 | if (cct->_conf->bluestore_bluefs) { | |
7140 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
7141 | cct->_conf->bluestore_block_wal_size, | |
7142 | cct->_conf->bluestore_block_wal_create); | |
7143 | if (r < 0) | |
7144 | goto out_close_fsid; | |
7145 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
7146 | cct->_conf->bluestore_block_db_size, | |
7147 | cct->_conf->bluestore_block_db_create); | |
7148 | if (r < 0) | |
7149 | goto out_close_fsid; | |
7150 | } | |
7151 | ||
7152 | r = _open_bdev(true); | |
7153 | if (r < 0) | |
7154 | goto out_close_fsid; | |
7155 | ||
20effc67 TL |
7156 | // choose freelist manager |
7157 | #ifdef HAVE_LIBZBD | |
7158 | if (bdev->is_smr()) { | |
7159 | freelist_type = "zoned"; | |
7160 | zone_size = bdev->get_zone_size(); | |
7161 | first_sequential_zone = bdev->get_conventional_region_size() / zone_size; | |
7162 | bdev->reset_all_zones(); | |
7163 | } else | |
7164 | #endif | |
7165 | { | |
7166 | freelist_type = "bitmap"; | |
7167 | } | |
7168 | dout(10) << " freelist_type " << freelist_type << dendl; | |
7169 | ||
3efd9988 | 7170 | // choose min_alloc_size |
20effc67 TL |
7171 | dout(5) << __func__ << " optimal_io_size 0x" << std::hex << optimal_io_size |
7172 | << " block_size: 0x" << block_size << std::dec << dendl; | |
7173 | if ((cct->_conf->bluestore_use_optimal_io_size_for_min_alloc_size) && (optimal_io_size != 0)) { | |
7174 | dout(5) << __func__ << " optimal_io_size 0x" << std::hex << optimal_io_size | |
7175 | << " for min_alloc_size 0x" << min_alloc_size << std::dec << dendl; | |
7176 | min_alloc_size = optimal_io_size; | |
7177 | } | |
7178 | else if (cct->_conf->bluestore_min_alloc_size) { | |
3efd9988 FG |
7179 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; |
7180 | } else { | |
11fdf7f2 | 7181 | ceph_assert(bdev); |
f67539c2 | 7182 | if (_use_rotational_settings()) { |
3efd9988 FG |
7183 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; |
7184 | } else { | |
7185 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
7186 | } | |
7187 | } | |
11fdf7f2 | 7188 | _validate_bdev(); |
3efd9988 FG |
7189 | |
7190 | // make sure min_alloc_size is power of 2 aligned. | |
1e59de90 | 7191 | if (!std::has_single_bit(min_alloc_size)) { |
3efd9988 FG |
7192 | derr << __func__ << " min_alloc_size 0x" |
7193 | << std::hex << min_alloc_size << std::dec | |
7194 | << " is not power of 2 aligned!" | |
7195 | << dendl; | |
7196 | r = -EINVAL; | |
7197 | goto out_close_bdev; | |
7198 | } | |
7199 | ||
20effc67 TL |
7200 | // make sure min_alloc_size is >= and aligned with block size |
7201 | if (min_alloc_size % block_size != 0) { | |
7202 | derr << __func__ << " min_alloc_size 0x" | |
7203 | << std::hex << min_alloc_size | |
7204 | << " is less or not aligned with block_size: 0x" | |
7205 | << block_size << std::dec << dendl; | |
7206 | r = -EINVAL; | |
7207 | goto out_close_bdev; | |
7208 | } | |
7209 | ||
f67539c2 TL |
7210 | r = _create_alloc(); |
7211 | if (r < 0) { | |
7212 | goto out_close_bdev; | |
7213 | } | |
7214 | ||
7215 | reserved = _get_ondisk_reserved(); | |
20effc67 | 7216 | alloc->init_add_free(reserved, |
f67539c2 | 7217 | p2align(bdev->get_size(), min_alloc_size) - reserved); |
20effc67 TL |
7218 | #ifdef HAVE_LIBZBD |
7219 | if (bdev->is_smr() && alloc != shared_alloc.a) { | |
7220 | shared_alloc.a->init_add_free(reserved, | |
7221 | p2align(bdev->get_conventional_region_size(), | |
7222 | min_alloc_size) - reserved); | |
7223 | } | |
7224 | #endif | |
f67539c2 | 7225 | |
7c673cae FG |
7226 | r = _open_db(true); |
7227 | if (r < 0) | |
f67539c2 | 7228 | goto out_close_alloc; |
7c673cae | 7229 | |
7c673cae FG |
7230 | { |
7231 | KeyValueDB::Transaction t = db->get_transaction(); | |
39ae355f | 7232 | r = _open_fm(t, false, true); |
11fdf7f2 TL |
7233 | if (r < 0) |
7234 | goto out_close_db; | |
7c673cae FG |
7235 | { |
7236 | bufferlist bl; | |
11fdf7f2 | 7237 | encode((uint64_t)0, bl); |
7c673cae FG |
7238 | t->set(PREFIX_SUPER, "nid_max", bl); |
7239 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
7240 | } | |
7241 | ||
7c673cae FG |
7242 | { |
7243 | bufferlist bl; | |
11fdf7f2 | 7244 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
7245 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
7246 | } | |
9f95a23c TL |
7247 | { |
7248 | bufferlist bl; | |
a4b75251 TL |
7249 | if (cct->_conf.get_val<bool>("bluestore_debug_legacy_omap")) { |
7250 | bl.append(stringify(OMAP_BULK)); | |
7251 | } else { | |
7252 | bl.append(stringify(OMAP_PER_PG)); | |
7253 | } | |
9f95a23c TL |
7254 | t->set(PREFIX_SUPER, "per_pool_omap", bl); |
7255 | } | |
20effc67 TL |
7256 | |
7257 | #ifdef HAVE_LIBZBD | |
7258 | if (bdev->is_smr()) { | |
7259 | { | |
7260 | bufferlist bl; | |
7261 | encode((uint64_t)zone_size, bl); | |
7262 | t->set(PREFIX_SUPER, "zone_size", bl); | |
7263 | } | |
7264 | { | |
7265 | bufferlist bl; | |
7266 | encode((uint64_t)first_sequential_zone, bl); | |
7267 | t->set(PREFIX_SUPER, "first_sequential_zone", bl); | |
7268 | } | |
7269 | } | |
7270 | #endif | |
7271 | ||
7c673cae FG |
7272 | ondisk_format = latest_ondisk_format; |
7273 | _prepare_ondisk_format_super(t); | |
7274 | db->submit_transaction_sync(t); | |
7275 | } | |
7276 | ||
7c673cae FG |
7277 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
7278 | if (r < 0) | |
224ce89b WB |
7279 | goto out_close_fm; |
7280 | ||
3efd9988 | 7281 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 7282 | if (r < 0) |
224ce89b | 7283 | goto out_close_fm; |
7c673cae FG |
7284 | |
7285 | if (fsid != old_fsid) { | |
7286 | r = _write_fsid(); | |
7287 | if (r < 0) { | |
7288 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 7289 | goto out_close_fm; |
7c673cae FG |
7290 | } |
7291 | } | |
7292 | ||
7c673cae FG |
7293 | out_close_fm: |
7294 | _close_fm(); | |
7295 | out_close_db: | |
20effc67 | 7296 | _close_db(); |
f67539c2 TL |
7297 | out_close_alloc: |
7298 | _close_alloc(); | |
7c673cae FG |
7299 | out_close_bdev: |
7300 | _close_bdev(); | |
7301 | out_close_fsid: | |
7302 | _close_fsid(); | |
7303 | out_path_fd: | |
7304 | _close_path(); | |
7305 | ||
7306 | if (r == 0 && | |
7307 | cct->_conf->bluestore_fsck_on_mkfs) { | |
7308 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
7309 | if (rc < 0) | |
7310 | return rc; | |
7311 | if (rc > 0) { | |
7312 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7313 | r = -EIO; | |
7314 | } | |
11fdf7f2 TL |
7315 | } |
7316 | ||
7317 | if (r == 0) { | |
7318 | // indicate success by writing the 'mkfs_done' file | |
7319 | r = write_meta("mkfs_done", "yes"); | |
7320 | } | |
7321 | ||
7322 | if (r < 0) { | |
7323 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
7324 | } else { | |
7325 | dout(0) << __func__ << " success" << dendl; | |
7326 | } | |
7327 | return r; | |
7328 | } | |
7329 | ||
11fdf7f2 TL |
7330 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) |
7331 | { | |
7332 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
7333 | int r; | |
7334 | ceph_assert(path_fd < 0); | |
7335 | ||
7336 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
7337 | ||
7338 | if (!cct->_conf->bluestore_bluefs) { | |
7339 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
7340 | return -EIO; | |
7341 | } | |
20effc67 | 7342 | dout(5) << __func__ << "::NCB::calling open_db_and_around(read-only)" << dendl; |
f67539c2 | 7343 | r = _open_db_and_around(true); |
20effc67 TL |
7344 | if (r < 0) { |
7345 | return r; | |
7346 | } | |
11fdf7f2 | 7347 | |
11fdf7f2 TL |
7348 | if (id == BlueFS::BDEV_NEWWAL) { |
7349 | string p = path + "/block.wal"; | |
7350 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
7351 | cct->_conf->bluestore_block_wal_size, | |
7352 | true); | |
7353 | ceph_assert(r == 0); | |
7354 | ||
7355 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
f67539c2 TL |
7356 | cct->_conf->bdev_enable_discard, |
7357 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
7358 | ceph_assert(r == 0); |
7359 | ||
7360 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
7361 | r = _check_or_set_bdev_label( | |
7362 | p, | |
7363 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
7364 | "bluefs wal", | |
7365 | true); | |
7366 | ceph_assert(r == 0); | |
7367 | } | |
7368 | ||
9f95a23c | 7369 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
7370 | } else if (id == BlueFS::BDEV_NEWDB) { |
7371 | string p = path + "/block.db"; | |
7372 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
7373 | cct->_conf->bluestore_block_db_size, | |
7374 | true); | |
7375 | ceph_assert(r == 0); | |
7376 | ||
7377 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
f67539c2 TL |
7378 | cct->_conf->bdev_enable_discard, |
7379 | SUPER_RESERVED); | |
11fdf7f2 TL |
7380 | ceph_assert(r == 0); |
7381 | ||
7382 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
7383 | r = _check_or_set_bdev_label( | |
7384 | p, | |
7385 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
7386 | "bluefs db", | |
7387 | true); | |
7388 | ceph_assert(r == 0); | |
7389 | } | |
9f95a23c TL |
7390 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
7391 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 | 7392 | } |
11fdf7f2 TL |
7393 | bluefs->umount(); |
7394 | bluefs->mount(); | |
7395 | ||
9f95a23c | 7396 | r = bluefs->prepare_new_device(id, bluefs_layout); |
11fdf7f2 TL |
7397 | ceph_assert(r == 0); |
7398 | ||
7399 | if (r < 0) { | |
7400 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
7401 | } else { | |
7402 | dout(0) << __func__ << " success" << dendl; | |
7403 | } | |
7404 | ||
20effc67 | 7405 | _close_db_and_around(); |
11fdf7f2 TL |
7406 | return r; |
7407 | } | |
7408 | ||
7409 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
7410 | int id) | |
7411 | { | |
7412 | dout(10) << __func__ << " id:" << id << dendl; | |
7413 | ceph_assert(path_fd < 0); | |
7414 | ||
7415 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
7416 | ||
7417 | if (!cct->_conf->bluestore_bluefs) { | |
7418 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
7419 | return -EIO; | |
7420 | } | |
7421 | ||
f67539c2 | 7422 | int r = _open_db_and_around(true); |
20effc67 TL |
7423 | if (r < 0) { |
7424 | return r; | |
7425 | } | |
7426 | auto close_db = make_scope_guard([&] { | |
7427 | _close_db_and_around(); | |
7428 | }); | |
f67539c2 | 7429 | uint64_t used_space = 0; |
11fdf7f2 | 7430 | for(auto src_id : devs_source) { |
f67539c2 | 7431 | used_space += bluefs->get_used(src_id); |
11fdf7f2 TL |
7432 | } |
7433 | uint64_t target_free = bluefs->get_free(id); | |
f67539c2 | 7434 | if (target_free < used_space) { |
11fdf7f2 TL |
7435 | derr << __func__ |
7436 | << " can't migrate, free space at target: " << target_free | |
7437 | << " is less than required space: " << used_space | |
7438 | << dendl; | |
20effc67 | 7439 | return -ENOSPC; |
11fdf7f2 | 7440 | } |
9f95a23c TL |
7441 | if (devs_source.count(BlueFS::BDEV_DB)) { |
7442 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; | |
7443 | bluefs_layout.dedicated_db = false; | |
7444 | } | |
7445 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
7446 | bluefs_layout.dedicated_wal = false; | |
7447 | } | |
7448 | r = bluefs->device_migrate_to_existing(cct, devs_source, id, bluefs_layout); | |
11fdf7f2 TL |
7449 | if (r < 0) { |
7450 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
20effc67 | 7451 | return r; |
11fdf7f2 TL |
7452 | } |
7453 | ||
7454 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
7455 | r = unlink(string(path + "/block.db").c_str()); | |
7456 | ceph_assert(r == 0); | |
7457 | } | |
7458 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
7459 | r = unlink(string(path + "/block.wal").c_str()); | |
7460 | ceph_assert(r == 0); | |
7461 | } | |
11fdf7f2 TL |
7462 | return r; |
7463 | } | |
7464 | ||
7465 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
7466 | int id, | |
7467 | const string& dev_path) | |
7468 | { | |
7469 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
11fdf7f2 TL |
7470 | ceph_assert(path_fd < 0); |
7471 | ||
7472 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
7473 | ||
7474 | if (!cct->_conf->bluestore_bluefs) { | |
7475 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
7476 | return -EIO; | |
7477 | } | |
7478 | ||
20effc67 TL |
7479 | int r = _open_db_and_around(true); |
7480 | if (r < 0) { | |
7481 | return r; | |
7482 | } | |
7483 | auto close_db = make_scope_guard([&] { | |
7484 | _close_db_and_around(); | |
7485 | }); | |
11fdf7f2 | 7486 | |
11fdf7f2 TL |
7487 | string link_db; |
7488 | string link_wal; | |
7489 | if (devs_source.count(BlueFS::BDEV_DB) && | |
9f95a23c | 7490 | bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 7491 | link_db = path + "/block.db"; |
9f95a23c TL |
7492 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7493 | bluefs_layout.dedicated_db = false; | |
11fdf7f2 TL |
7494 | } |
7495 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
7496 | link_wal = path + "/block.wal"; | |
9f95a23c | 7497 | bluefs_layout.dedicated_wal = false; |
11fdf7f2 TL |
7498 | } |
7499 | ||
20effc67 | 7500 | size_t target_size = 0; |
11fdf7f2 TL |
7501 | string target_name; |
7502 | if (id == BlueFS::BDEV_NEWWAL) { | |
7503 | target_name = "block.wal"; | |
7504 | target_size = cct->_conf->bluestore_block_wal_size; | |
9f95a23c | 7505 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
7506 | |
7507 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
f67539c2 TL |
7508 | cct->_conf->bdev_enable_discard, |
7509 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
7510 | ceph_assert(r == 0); |
7511 | ||
7512 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
7513 | r = _check_or_set_bdev_label( | |
7514 | dev_path, | |
7515 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
7516 | "bluefs wal", | |
7517 | true); | |
7518 | ceph_assert(r == 0); | |
7519 | } | |
11fdf7f2 TL |
7520 | } else if (id == BlueFS::BDEV_NEWDB) { |
7521 | target_name = "block.db"; | |
7522 | target_size = cct->_conf->bluestore_block_db_size; | |
9f95a23c TL |
7523 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
7524 | bluefs_layout.dedicated_db = true; | |
31f18b77 | 7525 | |
11fdf7f2 | 7526 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
f67539c2 TL |
7527 | cct->_conf->bdev_enable_discard, |
7528 | SUPER_RESERVED); | |
11fdf7f2 TL |
7529 | ceph_assert(r == 0); |
7530 | ||
7531 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
7532 | r = _check_or_set_bdev_label( | |
7533 | dev_path, | |
7534 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
7535 | "bluefs db", | |
7536 | true); | |
7537 | ceph_assert(r == 0); | |
7538 | } | |
31f18b77 FG |
7539 | } |
7540 | ||
11fdf7f2 TL |
7541 | bluefs->umount(); |
7542 | bluefs->mount(); | |
7543 | ||
9f95a23c | 7544 | r = bluefs->device_migrate_to_new(cct, devs_source, id, bluefs_layout); |
11fdf7f2 | 7545 | |
7c673cae | 7546 | if (r < 0) { |
11fdf7f2 | 7547 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
20effc67 | 7548 | return r; |
11fdf7f2 TL |
7549 | } |
7550 | ||
7551 | if (!link_db.empty()) { | |
7552 | r = unlink(link_db.c_str()); | |
7553 | ceph_assert(r == 0); | |
7554 | } | |
7555 | if (!link_wal.empty()) { | |
7556 | r = unlink(link_wal.c_str()); | |
7557 | ceph_assert(r == 0); | |
7558 | } | |
7559 | r = _setup_block_symlink_or_file( | |
7560 | target_name, | |
7561 | dev_path, | |
7562 | target_size, | |
7563 | true); | |
7564 | ceph_assert(r == 0); | |
7565 | dout(0) << __func__ << " success" << dendl; | |
7566 | ||
11fdf7f2 TL |
7567 | return r; |
7568 | } | |
7569 | ||
7570 | string BlueStore::get_device_path(unsigned id) | |
7571 | { | |
7572 | string res; | |
7573 | if (id < BlueFS::MAX_BDEV) { | |
7574 | switch (id) { | |
7575 | case BlueFS::BDEV_WAL: | |
7576 | res = path + "/block.wal"; | |
7577 | break; | |
7578 | case BlueFS::BDEV_DB: | |
9f95a23c | 7579 | if (id == bluefs_layout.shared_bdev) { |
11fdf7f2 TL |
7580 | res = path + "/block"; |
7581 | } else { | |
7582 | res = path + "/block.db"; | |
7583 | } | |
7584 | break; | |
7585 | case BlueFS::BDEV_SLOW: | |
7586 | res = path + "/block"; | |
7587 | break; | |
7588 | } | |
7589 | } | |
7590 | return res; | |
7591 | } | |
7592 | ||
f67539c2 TL |
7593 | int BlueStore::_set_bdev_label_size(const string& path, uint64_t size) |
7594 | { | |
7595 | bluestore_bdev_label_t label; | |
7596 | int r = _read_bdev_label(cct, path, &label); | |
7597 | if (r < 0) { | |
7598 | derr << "unable to read label for " << path << ": " | |
7599 | << cpp_strerror(r) << dendl; | |
7600 | } else { | |
7601 | label.size = size; | |
7602 | r = _write_bdev_label(cct, path, label); | |
7603 | if (r < 0) { | |
7604 | derr << "unable to write label for " << path << ": " | |
7605 | << cpp_strerror(r) << dendl; | |
7606 | } | |
7607 | } | |
7608 | return r; | |
7609 | } | |
7610 | ||
11fdf7f2 TL |
7611 | int BlueStore::expand_devices(ostream& out) |
7612 | { | |
f67539c2 | 7613 | int r = _open_db_and_around(true); |
11fdf7f2 TL |
7614 | ceph_assert(r == 0); |
7615 | bluefs->dump_block_extents(out); | |
1911f103 | 7616 | out << "Expanding DB/WAL..." << std::endl; |
11fdf7f2 | 7617 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { |
9f95a23c | 7618 | if (devid == bluefs_layout.shared_bdev ) { |
11fdf7f2 TL |
7619 | continue; |
7620 | } | |
7621 | uint64_t size = bluefs->get_block_device_size(devid); | |
7622 | if (size == 0) { | |
7623 | // no bdev | |
7624 | continue; | |
7625 | } | |
7626 | ||
f67539c2 TL |
7627 | out << devid |
7628 | <<" : expanding " << " to 0x" << size << std::dec << std::endl; | |
7629 | string p = get_device_path(devid); | |
7630 | const char* path = p.c_str(); | |
7631 | if (path == nullptr) { | |
7632 | derr << devid | |
7633 | <<": can't find device path " << dendl; | |
7634 | continue; | |
7635 | } | |
7636 | if (bluefs->bdev_support_label(devid)) { | |
7637 | if (_set_bdev_label_size(p, size) >= 0) { | |
7638 | out << devid | |
7639 | << " : size label updated to " << size | |
7640 | << std::endl; | |
11fdf7f2 | 7641 | } |
11fdf7f2 TL |
7642 | } |
7643 | } | |
7644 | uint64_t size0 = fm->get_size(); | |
7645 | uint64_t size = bdev->get_size(); | |
7646 | if (size0 < size) { | |
9f95a23c | 7647 | out << bluefs_layout.shared_bdev |
1911f103 TL |
7648 | << " : expanding " << " from 0x" << std::hex |
7649 | << size0 << " to 0x" << size << std::dec << std::endl; | |
f67539c2 TL |
7650 | _write_out_fm_meta(size); |
7651 | if (bdev->supported_bdev_label()) { | |
7652 | if (_set_bdev_label_size(path, size) >= 0) { | |
7653 | out << bluefs_layout.shared_bdev | |
7654 | << " : size label updated to " << size | |
7655 | << std::endl; | |
7656 | } | |
7657 | } | |
20effc67 | 7658 | |
1d09f67e TL |
7659 | if (fm && fm->is_null_manager()) { |
7660 | // we grow the allocation range, must reflect it in the allocation file | |
7661 | alloc->init_add_free(size0, size - size0); | |
7662 | need_to_destage_allocation_file = true; | |
7663 | } | |
20effc67 | 7664 | _close_db_and_around(); |
1911f103 TL |
7665 | |
7666 | // mount in read/write to sync expansion changes | |
f67539c2 | 7667 | r = _mount(); |
11fdf7f2 | 7668 | ceph_assert(r == 0); |
1911f103 TL |
7669 | umount(); |
7670 | } else { | |
20effc67 | 7671 | _close_db_and_around(); |
7c673cae | 7672 | } |
1911f103 TL |
7673 | return r; |
7674 | } | |
7675 | ||
7676 | int BlueStore::dump_bluefs_sizes(ostream& out) | |
7677 | { | |
f67539c2 | 7678 | int r = _open_db_and_around(true); |
1911f103 TL |
7679 | ceph_assert(r == 0); |
7680 | bluefs->dump_block_extents(out); | |
20effc67 | 7681 | _close_db_and_around(); |
7c673cae FG |
7682 | return r; |
7683 | } | |
7684 | ||
7685 | void BlueStore::set_cache_shards(unsigned num) | |
7686 | { | |
7687 | dout(10) << __func__ << " " << num << dendl; | |
9f95a23c TL |
7688 | size_t oold = onode_cache_shards.size(); |
7689 | size_t bold = buffer_cache_shards.size(); | |
7690 | ceph_assert(num >= oold && num >= bold); | |
7691 | onode_cache_shards.resize(num); | |
7692 | buffer_cache_shards.resize(num); | |
7693 | for (unsigned i = oold; i < num; ++i) { | |
7694 | onode_cache_shards[i] = | |
7695 | OnodeCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
7696 | logger); | |
7697 | } | |
7698 | for (unsigned i = bold; i < num; ++i) { | |
7699 | buffer_cache_shards[i] = | |
7700 | BufferCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
7701 | logger); | |
7c673cae FG |
7702 | } |
7703 | } | |
7704 | ||
1d09f67e | 7705 | //--------------------------------------------- |
39ae355f | 7706 | bool BlueStore::has_null_manager() const |
1d09f67e TL |
7707 | { |
7708 | return (fm && fm->is_null_manager()); | |
7709 | } | |
7710 | ||
f67539c2 | 7711 | int BlueStore::_mount() |
7c673cae | 7712 | { |
20effc67 | 7713 | dout(5) << __func__ << "NCB:: path " << path << dendl; |
1d09f67e | 7714 | |
f67539c2 | 7715 | _kv_only = false; |
7c673cae | 7716 | if (cct->_conf->bluestore_fsck_on_mount) { |
20effc67 | 7717 | dout(5) << __func__ << "::NCB::calling fsck()" << dendl; |
7c673cae FG |
7718 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); |
7719 | if (rc < 0) | |
7720 | return rc; | |
7721 | if (rc > 0) { | |
7722 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7723 | return -EIO; | |
7724 | } | |
7725 | } | |
7726 | ||
eafe8130 TL |
7727 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
7728 | derr << __func__ << " osd_max_object_size " | |
7729 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
7730 | << OBJECT_MAX_SIZE << dendl; | |
7731 | return -EINVAL; | |
7732 | } | |
7733 | ||
20effc67 | 7734 | dout(5) << __func__ << "::NCB::calling open_db_and_around(read/write)" << dendl; |
f67539c2 | 7735 | int r = _open_db_and_around(false); |
9f95a23c | 7736 | if (r < 0) { |
f67539c2 | 7737 | return r; |
11fdf7f2 | 7738 | } |
20effc67 TL |
7739 | auto close_db = make_scope_guard([&] { |
7740 | if (!mounted) { | |
7741 | _close_db_and_around(); | |
7742 | } | |
7743 | }); | |
7c673cae | 7744 | |
11fdf7f2 TL |
7745 | r = _upgrade_super(); |
7746 | if (r < 0) { | |
20effc67 | 7747 | return r; |
11fdf7f2 | 7748 | } |
7c673cae | 7749 | |
20effc67 | 7750 | // The recovery process for allocation-map needs to open collection early |
7c673cae | 7751 | r = _open_collections(); |
20effc67 TL |
7752 | if (r < 0) { |
7753 | return r; | |
7754 | } | |
7755 | auto shutdown_cache = make_scope_guard([&] { | |
7756 | if (!mounted) { | |
7757 | _shutdown_cache(); | |
7758 | } | |
7759 | }); | |
7c673cae FG |
7760 | |
7761 | r = _reload_logger(); | |
20effc67 TL |
7762 | if (r < 0) { |
7763 | return r; | |
7764 | } | |
7c673cae | 7765 | |
31f18b77 | 7766 | _kv_start(); |
20effc67 TL |
7767 | auto stop_kv = make_scope_guard([&] { |
7768 | if (!mounted) { | |
7769 | _kv_stop(); | |
7770 | } | |
7771 | }); | |
7772 | ||
7773 | r = _deferred_replay(); | |
7774 | if (r < 0) { | |
7775 | return r; | |
7776 | } | |
7c673cae | 7777 | |
20effc67 | 7778 | #ifdef HAVE_LIBZBD |
f67539c2 TL |
7779 | if (bdev->is_smr()) { |
7780 | _zoned_cleaner_start(); | |
7781 | } | |
20effc67 | 7782 | #endif |
7c673cae FG |
7783 | |
7784 | mempool_thread.init(); | |
7785 | ||
f67539c2 | 7786 | if ((!per_pool_stat_collection || per_pool_omap != OMAP_PER_PG) && |
eafe8130 | 7787 | cct->_conf->bluestore_fsck_quick_fix_on_mount == true) { |
9f95a23c | 7788 | |
f67539c2 | 7789 | auto was_per_pool_omap = per_pool_omap; |
9f95a23c | 7790 | |
eafe8130 TL |
7791 | dout(1) << __func__ << " quick-fix on mount" << dendl; |
7792 | _fsck_on_open(FSCK_SHALLOW, true); | |
7793 | ||
9f95a23c | 7794 | //set again as hopefully it has been fixed |
f67539c2 | 7795 | if (was_per_pool_omap != OMAP_PER_PG) { |
9f95a23c TL |
7796 | _set_per_pool_omap(); |
7797 | } | |
eafe8130 TL |
7798 | } |
7799 | ||
7c673cae FG |
7800 | mounted = true; |
7801 | return 0; | |
7c673cae FG |
7802 | } |
7803 | ||
7804 | int BlueStore::umount() | |
7805 | { | |
11fdf7f2 | 7806 | ceph_assert(_kv_only || mounted); |
7c673cae | 7807 | _osr_drain_all(); |
7c673cae | 7808 | |
7c673cae | 7809 | mounted = false; |
20effc67 TL |
7810 | |
7811 | ceph_assert(alloc); | |
7812 | ||
3efd9988 FG |
7813 | if (!_kv_only) { |
7814 | mempool_thread.shutdown(); | |
20effc67 | 7815 | #ifdef HAVE_LIBZBD |
f67539c2 TL |
7816 | if (bdev->is_smr()) { |
7817 | dout(20) << __func__ << " stopping zone cleaner thread" << dendl; | |
7818 | _zoned_cleaner_stop(); | |
7819 | } | |
20effc67 | 7820 | #endif |
3efd9988 FG |
7821 | dout(20) << __func__ << " stopping kv thread" << dendl; |
7822 | _kv_stop(); | |
1d09f67e TL |
7823 | // skip cache cleanup step on fast shutdown |
7824 | if (likely(!m_fast_shutdown)) { | |
7825 | _shutdown_cache(); | |
7826 | } | |
3efd9988 | 7827 | dout(20) << __func__ << " closing" << dendl; |
3efd9988 | 7828 | } |
20effc67 | 7829 | _close_db_and_around(); |
1d09f67e TL |
7830 | // disable fsck on fast-shutdown |
7831 | if (cct->_conf->bluestore_fsck_on_umount && !m_fast_shutdown) { | |
7c673cae FG |
7832 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); |
7833 | if (rc < 0) | |
7834 | return rc; | |
7835 | if (rc > 0) { | |
7836 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7837 | return -EIO; | |
7838 | } | |
7839 | } | |
7840 | return 0; | |
7841 | } | |
7842 | ||
eafe8130 TL |
7843 | int BlueStore::cold_open() |
7844 | { | |
f67539c2 | 7845 | return _open_db_and_around(true); |
eafe8130 | 7846 | } |
f67539c2 | 7847 | |
eafe8130 TL |
7848 | int BlueStore::cold_close() |
7849 | { | |
20effc67 | 7850 | _close_db_and_around(); |
eafe8130 TL |
7851 | return 0; |
7852 | } | |
7853 | ||
9f95a23c TL |
7854 | // derr wrapper to limit enormous output and avoid log flooding. |
7855 | // Of limited use where such output is expected for now | |
7856 | #define fsck_derr(err_cnt, threshold) \ | |
7857 | if (err_cnt <= threshold) { \ | |
7858 | bool need_skip_print = err_cnt == threshold; \ | |
7859 | derr | |
7860 | ||
7861 | #define fsck_dendl \ | |
7862 | dendl; \ | |
7863 | if (need_skip_print) \ | |
7864 | derr << "more error lines skipped..." << dendl; \ | |
7c673cae | 7865 | } |
7c673cae | 7866 | |
eafe8130 TL |
7867 | int _fsck_sum_extents( |
7868 | const PExtentVector& extents, | |
7869 | bool compressed, | |
7870 | store_statfs_t& expected_statfs) | |
7871 | { | |
7872 | for (auto e : extents) { | |
7873 | if (!e.is_valid()) | |
7874 | continue; | |
7875 | expected_statfs.allocated += e.length; | |
7876 | if (compressed) { | |
7877 | expected_statfs.data_compressed_allocated += e.length; | |
7878 | } | |
7879 | } | |
7880 | return 0; | |
7881 | } | |
7882 | ||
7c673cae | 7883 | int BlueStore::_fsck_check_extents( |
20effc67 | 7884 | std::string_view ctx_descr, |
7c673cae FG |
7885 | const PExtentVector& extents, |
7886 | bool compressed, | |
7887 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 7888 | uint64_t granularity, |
11fdf7f2 | 7889 | BlueStoreRepairer* repairer, |
eafe8130 TL |
7890 | store_statfs_t& expected_statfs, |
7891 | FSCKDepth depth) | |
7c673cae | 7892 | { |
20effc67 | 7893 | dout(30) << __func__ << " " << ctx_descr << ", extents " << extents << dendl; |
7c673cae FG |
7894 | int errors = 0; |
7895 | for (auto e : extents) { | |
7896 | if (!e.is_valid()) | |
7897 | continue; | |
7898 | expected_statfs.allocated += e.length; | |
7899 | if (compressed) { | |
11fdf7f2 | 7900 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae | 7901 | } |
eafe8130 TL |
7902 | if (depth != FSCK_SHALLOW) { |
7903 | bool already = false; | |
9f95a23c | 7904 | apply_for_bitset_range( |
eafe8130 TL |
7905 | e.offset, e.length, granularity, used_blocks, |
7906 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
7907 | if (bs.test(pos)) { |
7908 | if (repairer) { | |
7909 | repairer->note_misreference( | |
7910 | pos * min_alloc_size, min_alloc_size, !already); | |
7911 | } | |
7912 | if (!already) { | |
20effc67 | 7913 | derr << __func__ << "::fsck error: " << ctx_descr << ", extent " << e |
eafe8130 TL |
7914 | << " or a subset is already allocated (misreferenced)" << dendl; |
7915 | ++errors; | |
7916 | already = true; | |
7917 | } | |
11fdf7f2 | 7918 | } |
eafe8130 TL |
7919 | else |
7920 | bs.set(pos); | |
7921 | }); | |
11fdf7f2 | 7922 | |
eafe8130 | 7923 | if (e.end() > bdev->get_size()) { |
20effc67 | 7924 | derr << "fsck error: " << ctx_descr << ", extent " << e |
eafe8130 TL |
7925 | << " past end of block device" << dendl; |
7926 | ++errors; | |
7927 | } | |
7c673cae FG |
7928 | } |
7929 | } | |
7930 | return errors; | |
7931 | } | |
7932 | ||
39ae355f TL |
7933 | void BlueStore::_fsck_check_statfs( |
7934 | const store_statfs_t& expected_statfs, | |
7935 | const per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
7936 | int64_t& errors, |
7937 | int64_t& warnings, | |
11fdf7f2 TL |
7938 | BlueStoreRepairer* repairer) |
7939 | { | |
39ae355f TL |
7940 | string key; |
7941 | store_statfs_t actual_statfs; | |
7942 | store_statfs_t s; | |
7943 | { | |
7944 | // make a copy | |
7945 | per_pool_statfs my_expected_pool_statfs(expected_pool_statfs); | |
7946 | auto op = osd_pools.begin(); | |
7947 | while (op != osd_pools.end()) { | |
7948 | get_pool_stat_key(op->first, &key); | |
7949 | op->second.publish(&s); | |
7950 | auto it_expected = my_expected_pool_statfs.find(op->first); | |
7951 | if (it_expected == my_expected_pool_statfs.end()) { | |
7952 | auto op0 = op++; | |
7953 | if (op0->second.is_empty()) { | |
7954 | // It's OK to lack relevant empty statfs record | |
7955 | continue; | |
7956 | } | |
7957 | derr << __func__ << "::fsck error: " << std::hex | |
7958 | << "pool " << op0->first << " has got no statfs to match against: " | |
7959 | << s | |
7960 | << std::dec << dendl; | |
7961 | ++errors; | |
11fdf7f2 | 7962 | if (repairer) { |
39ae355f | 7963 | osd_pools.erase(op0); |
11fdf7f2 TL |
7964 | repairer->remove_key(db, PREFIX_STAT, key); |
7965 | } | |
39ae355f TL |
7966 | } else { |
7967 | if (!(s == it_expected->second)) { | |
7968 | derr << "fsck error: actual " << s | |
7969 | << " != expected " << it_expected->second | |
7970 | << " for pool " | |
7971 | << std::hex << op->first << std::dec << dendl; | |
7972 | ++errors; | |
11fdf7f2 | 7973 | if (repairer) { |
39ae355f TL |
7974 | // repair in-memory in a hope this would be flushed properly on shutdown |
7975 | s = it_expected->second; | |
7976 | op->second = it_expected->second; | |
7977 | repairer->fix_statfs(db, key, it_expected->second); | |
11fdf7f2 | 7978 | } |
11fdf7f2 | 7979 | } |
39ae355f TL |
7980 | actual_statfs.add(s); |
7981 | my_expected_pool_statfs.erase(it_expected); | |
7982 | ++op; | |
11fdf7f2 | 7983 | } |
11fdf7f2 | 7984 | } |
39ae355f TL |
7985 | // check stats that lack matching entities in osd_pools |
7986 | for (auto &p : my_expected_pool_statfs) { | |
7987 | if (p.second.is_zero()) { | |
7988 | // It's OK to lack relevant empty statfs record | |
7989 | continue; | |
7990 | } | |
7991 | get_pool_stat_key(p.first, &key); | |
7992 | derr << __func__ << "::fsck error: " << std::hex | |
7993 | << "pool " << p.first << " has got no actual statfs: " | |
7994 | << std::dec << p.second | |
7995 | << dendl; | |
7996 | ++errors; | |
7997 | if (repairer) { | |
7998 | osd_pools[p.first] = p.second; | |
7999 | repairer->fix_statfs(db, key, p.second); | |
8000 | actual_statfs.add(p.second); | |
8001 | } | |
8002 | } | |
8003 | } | |
8004 | // process global statfs | |
8005 | if (repairer) { | |
8006 | if (!per_pool_stat_collection) { | |
8007 | // by virtue of running this method, we correct the top-level | |
8008 | // error of having global stats | |
8009 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
8010 | per_pool_stat_collection = true; | |
8011 | } | |
8012 | vstatfs = actual_statfs; | |
8013 | dout(20) << __func__ << " setting vstatfs to " << actual_statfs << dendl; | |
8014 | } else if (!per_pool_stat_collection) { | |
8015 | // check global stats only if fscking (not repairing) w/o per-pool stats | |
8016 | vstatfs.publish(&s); | |
8017 | if (!(s == expected_statfs)) { | |
8018 | derr << "fsck error: actual " << s | |
8019 | << " != expected " << expected_statfs << dendl; | |
8020 | ++errors; | |
11fdf7f2 | 8021 | } |
eafe8130 | 8022 | } |
11fdf7f2 TL |
8023 | } |
8024 | ||
20effc67 TL |
8025 | void BlueStore::_fsck_repair_shared_blobs( |
8026 | BlueStoreRepairer& repairer, | |
8027 | shared_blob_2hash_tracker_t& sb_ref_counts, | |
8028 | sb_info_space_efficient_map_t& sb_info) | |
8029 | { | |
8030 | auto sb_ref_mismatches = sb_ref_counts.count_non_zero(); | |
8031 | dout(1) << __func__ << " repairing shared_blobs, ref mismatch estimate: " | |
8032 | << sb_ref_mismatches << dendl; | |
8033 | if (!sb_ref_mismatches) // not expected to succeed, just in case | |
8034 | return; | |
8035 | ||
8036 | ||
8037 | auto foreach_shared_blob = [&](std::function< | |
8038 | void (coll_t, | |
8039 | ghobject_t, | |
8040 | uint64_t, | |
8041 | const bluestore_blob_t&)> cb) { | |
8042 | auto it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); | |
8043 | if (it) { | |
8044 | CollectionRef c; | |
8045 | spg_t pgid; | |
8046 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8047 | dout(30) << __func__ << " key " | |
8048 | << pretty_binary_string(it->key()) | |
8049 | << dendl; | |
8050 | if (is_extent_shard_key(it->key())) { | |
8051 | continue; | |
8052 | } | |
8053 | ||
8054 | ghobject_t oid; | |
8055 | int r = get_key_object(it->key(), &oid); | |
8056 | if (r < 0) { | |
8057 | continue; | |
8058 | } | |
8059 | ||
8060 | if (!c || | |
8061 | oid.shard_id != pgid.shard || | |
8062 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8063 | !c->contains(oid)) { | |
8064 | c = nullptr; | |
8065 | for (auto& p : coll_map) { | |
8066 | if (p.second->contains(oid)) { | |
8067 | c = p.second; | |
8068 | break; | |
8069 | } | |
8070 | } | |
8071 | if (!c) { | |
8072 | continue; | |
8073 | } | |
8074 | } | |
8075 | dout(20) << __func__ | |
8076 | << " inspecting shared blob refs for col:" << c->cid | |
8077 | << " obj:" << oid | |
8078 | << dendl; | |
8079 | ||
8080 | OnodeRef o; | |
39ae355f | 8081 | o.reset(Onode::create_decode(c, oid, it->key(), it->value())); |
20effc67 TL |
8082 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8083 | ||
8084 | _dump_onode<30>(cct, *o); | |
8085 | ||
8086 | mempool::bluestore_fsck::set<BlobRef> passed_sbs; | |
8087 | for (auto& e : o->extent_map.extent_map) { | |
8088 | auto& b = e.blob->get_blob(); | |
8089 | if (b.is_shared() && passed_sbs.count(e.blob) == 0) { | |
8090 | auto sbid = e.blob->shared_blob->get_sbid(); | |
8091 | cb(c->cid, oid, sbid, b); | |
8092 | passed_sbs.emplace(e.blob); | |
8093 | } | |
8094 | } // for ... extent_map | |
8095 | } // for ... it->valid | |
8096 | } //if (it(PREFIX_OBJ)) | |
8097 | }; //foreach_shared_blob fn declaration | |
8098 | ||
8099 | mempool::bluestore_fsck::map<uint64_t, bluestore_extent_ref_map_t> refs_map; | |
8100 | ||
8101 | // first iteration over objects to identify all the broken sbids | |
8102 | foreach_shared_blob( [&](coll_t cid, | |
8103 | ghobject_t oid, | |
8104 | uint64_t sbid, | |
8105 | const bluestore_blob_t& b) { | |
8106 | auto it = refs_map.lower_bound(sbid); | |
8107 | if(it != refs_map.end() && it->first == sbid) { | |
8108 | return; | |
8109 | } | |
8110 | for (auto& p : b.get_extents()) { | |
8111 | if (p.is_valid() && | |
8112 | !sb_ref_counts.test_all_zero_range(sbid, | |
8113 | p.offset, | |
8114 | p.length)) { | |
8115 | refs_map.emplace_hint(it, sbid, bluestore_extent_ref_map_t()); | |
8116 | dout(20) << __func__ | |
8117 | << " broken shared blob found for col:" << cid | |
8118 | << " obj:" << oid | |
8119 | << " sbid 0x " << std::hex << sbid << std::dec | |
8120 | << dendl; | |
8121 | break; | |
8122 | } | |
8123 | } | |
8124 | }); | |
8125 | ||
8126 | // second iteration over objects to build new ref map for the broken sbids | |
8127 | foreach_shared_blob( [&](coll_t cid, | |
8128 | ghobject_t oid, | |
8129 | uint64_t sbid, | |
8130 | const bluestore_blob_t& b) { | |
8131 | auto it = refs_map.find(sbid); | |
8132 | if(it == refs_map.end()) { | |
8133 | return; | |
8134 | } | |
8135 | for (auto& p : b.get_extents()) { | |
8136 | if (p.is_valid()) { | |
8137 | it->second.get(p.offset, p.length); | |
8138 | break; | |
8139 | } | |
8140 | } | |
8141 | }); | |
8142 | ||
8143 | // update shared blob records | |
8144 | auto ref_it = refs_map.begin(); | |
8145 | while (ref_it != refs_map.end()) { | |
8146 | size_t cnt = 0; | |
8147 | const size_t max_transactions = 4096; | |
8148 | KeyValueDB::Transaction txn = db->get_transaction(); | |
8149 | for (cnt = 0; | |
8150 | cnt < max_transactions && ref_it != refs_map.end(); | |
8151 | ref_it++) { | |
8152 | auto sbid = ref_it->first; | |
8153 | dout(20) << __func__ << " repaired shared_blob 0x" | |
8154 | << std::hex << sbid << std::dec | |
8155 | << ref_it->second << dendl; | |
8156 | repairer.fix_shared_blob(txn, sbid, &ref_it->second, 0); | |
8157 | cnt++; | |
8158 | } | |
8159 | if (cnt) { | |
8160 | db->submit_transaction_sync(txn); | |
8161 | cnt = 0; | |
8162 | } | |
8163 | } | |
8164 | // remove stray shared blob records | |
8165 | size_t cnt = 0; | |
8166 | const size_t max_transactions = 4096; | |
8167 | KeyValueDB::Transaction txn = db->get_transaction(); | |
8168 | sb_info.foreach_stray([&](const sb_info_t& sbi) { | |
8169 | auto sbid = sbi.get_sbid(); | |
8170 | dout(20) << __func__ << " removing stray shared_blob 0x" | |
8171 | << std::hex << sbid << std::dec | |
8172 | << dendl; | |
8173 | repairer.fix_shared_blob(txn, sbid, nullptr, 0); | |
8174 | cnt++; | |
8175 | if (cnt >= max_transactions) {} | |
8176 | db->submit_transaction_sync(txn); | |
8177 | txn = db->get_transaction(); | |
8178 | cnt = 0; | |
8179 | }); | |
8180 | if (cnt > 0) { | |
8181 | db->submit_transaction_sync(txn); | |
8182 | } | |
8183 | ||
8184 | // amount of repairs to report to be equal to previously | |
8185 | // determined error estimation, not the actual number of updated shared blobs | |
8186 | repairer.inc_repaired(sb_ref_mismatches); | |
8187 | } | |
8188 | ||
eafe8130 TL |
8189 | BlueStore::OnodeRef BlueStore::fsck_check_objects_shallow( |
8190 | BlueStore::FSCKDepth depth, | |
8191 | int64_t pool_id, | |
8192 | BlueStore::CollectionRef c, | |
8193 | const ghobject_t& oid, | |
8194 | const string& key, | |
8195 | const bufferlist& value, | |
9f95a23c | 8196 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
8197 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
8198 | const BlueStore::FSCK_ObjectCtx& ctx) | |
8199 | { | |
8200 | auto& errors = ctx.errors; | |
8201 | auto& num_objects = ctx.num_objects; | |
8202 | auto& num_extents = ctx.num_extents; | |
8203 | auto& num_blobs = ctx.num_blobs; | |
8204 | auto& num_sharded_objects = ctx.num_sharded_objects; | |
8205 | auto& num_spanning_blobs = ctx.num_spanning_blobs; | |
8206 | auto used_blocks = ctx.used_blocks; | |
8207 | auto sb_info_lock = ctx.sb_info_lock; | |
8208 | auto& sb_info = ctx.sb_info; | |
20effc67 | 8209 | auto& sb_ref_counts = ctx.sb_ref_counts; |
eafe8130 TL |
8210 | auto repairer = ctx.repairer; |
8211 | ||
8212 | store_statfs_t* res_statfs = (per_pool_stat_collection || repairer) ? | |
8213 | &ctx.expected_pool_statfs[pool_id] : | |
8214 | &ctx.expected_store_statfs; | |
8215 | ||
20effc67 TL |
8216 | map<uint32_t, uint64_t> zone_first_offsets; // for zoned/smr devices |
8217 | ||
eafe8130 TL |
8218 | dout(10) << __func__ << " " << oid << dendl; |
8219 | OnodeRef o; | |
39ae355f | 8220 | o.reset(Onode::create_decode(c, oid, key, value)); |
eafe8130 | 8221 | ++num_objects; |
7c673cae | 8222 | |
eafe8130 | 8223 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); |
7c673cae | 8224 | |
eafe8130 TL |
8225 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8226 | _dump_onode<30>(cct, *o); | |
8227 | // shards | |
8228 | if (!o->extent_map.shards.empty()) { | |
8229 | ++num_sharded_objects; | |
8230 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 8231 | ceph_assert(expecting_shards); |
eafe8130 TL |
8232 | for (auto& s : o->extent_map.shards) { |
8233 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
9f95a23c | 8234 | expecting_shards->push_back(string()); |
eafe8130 | 8235 | get_extent_shard_key(o->key, s.shard_info->offset, |
9f95a23c | 8236 | &expecting_shards->back()); |
eafe8130 TL |
8237 | if (s.shard_info->offset >= o->onode.size) { |
8238 | derr << "fsck error: " << oid << " shard 0x" << std::hex | |
8239 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size | |
8240 | << std::dec << dendl; | |
8241 | ++errors; | |
8242 | } | |
8243 | } | |
8244 | } | |
8245 | } | |
7c673cae | 8246 | |
eafe8130 TL |
8247 | // lextents |
8248 | uint64_t pos = 0; | |
8249 | mempool::bluestore_fsck::map<BlobRef, | |
8250 | bluestore_blob_use_tracker_t> ref_map; | |
8251 | for (auto& l : o->extent_map.extent_map) { | |
8252 | dout(20) << __func__ << " " << l << dendl; | |
8253 | if (l.logical_offset < pos) { | |
8254 | derr << "fsck error: " << oid << " lextent at 0x" | |
8255 | << std::hex << l.logical_offset | |
8256 | << " overlaps with the previous, which ends at 0x" << pos | |
8257 | << std::dec << dendl; | |
8258 | ++errors; | |
8259 | } | |
8260 | if (depth != FSCK_SHALLOW && | |
8261 | o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
8262 | derr << "fsck error: " << oid << " lextent at 0x" | |
8263 | << std::hex << l.logical_offset << "~" << l.length | |
8264 | << " spans a shard boundary" | |
8265 | << std::dec << dendl; | |
8266 | ++errors; | |
8267 | } | |
8268 | pos = l.logical_offset + l.length; | |
8269 | res_statfs->data_stored += l.length; | |
8270 | ceph_assert(l.blob); | |
8271 | const bluestore_blob_t& blob = l.blob->get_blob(); | |
8272 | ||
20effc67 TL |
8273 | #ifdef HAVE_LIBZBD |
8274 | if (bdev->is_smr() && depth != FSCK_SHALLOW) { | |
8275 | for (auto& e : blob.get_extents()) { | |
8276 | if (e.is_valid()) { | |
8277 | uint32_t zone = e.offset / zone_size; | |
8278 | uint64_t offset = e.offset % zone_size; | |
8279 | auto p = zone_first_offsets.find(zone); | |
8280 | if (p == zone_first_offsets.end() || p->second > offset) { | |
8281 | // FIXME: use interator for guided insert? | |
8282 | zone_first_offsets[zone] = offset; | |
8283 | } | |
8284 | } | |
8285 | } | |
8286 | } | |
8287 | #endif | |
8288 | ||
8289 | auto& ref = ref_map[l.blob]; | |
8290 | if (ref.is_empty()) { | |
8291 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
8292 | uint32_t l = blob.get_logical_length(); | |
8293 | ref.init(l, min_release_size); | |
eafe8130 TL |
8294 | } |
8295 | ref.get( | |
8296 | l.blob_offset, | |
8297 | l.length); | |
8298 | ++num_extents; | |
8299 | if (depth != FSCK_SHALLOW && | |
8300 | blob.has_unused()) { | |
8301 | ceph_assert(referenced); | |
8302 | auto p = referenced->find(l.blob); | |
8303 | bluestore_blob_t::unused_t* pu; | |
8304 | if (p == referenced->end()) { | |
8305 | pu = &(*referenced)[l.blob]; | |
8306 | } | |
8307 | else { | |
8308 | pu = &p->second; | |
8309 | } | |
8310 | uint64_t blob_len = blob.get_logical_length(); | |
8311 | ceph_assert((blob_len % (sizeof(*pu) * 8)) == 0); | |
8312 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
8313 | uint64_t chunk_size = blob_len / (sizeof(*pu) * 8); | |
8314 | uint64_t start = l.blob_offset / chunk_size; | |
8315 | uint64_t end = | |
8316 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; | |
8317 | for (auto i = start; i < end; ++i) { | |
8318 | (*pu) |= (1u << i); | |
8319 | } | |
8320 | } | |
8321 | } //for (auto& l : o->extent_map.extent_map) | |
8322 | ||
8323 | for (auto& i : ref_map) { | |
8324 | ++num_blobs; | |
8325 | const bluestore_blob_t& blob = i.first->get_blob(); | |
8326 | bool equal = | |
8327 | depth == FSCK_SHALLOW ? true : | |
8328 | i.first->get_blob_use_tracker().equal(i.second); | |
8329 | if (!equal) { | |
8330 | derr << "fsck error: " << oid << " blob " << *i.first | |
8331 | << " doesn't match expected ref_map " << i.second << dendl; | |
8332 | ++errors; | |
8333 | } | |
8334 | if (blob.is_compressed()) { | |
8335 | res_statfs->data_compressed += blob.get_compressed_payload_length(); | |
8336 | res_statfs->data_compressed_original += | |
8337 | i.first->get_referenced_bytes(); | |
8338 | } | |
20effc67 TL |
8339 | if (depth != FSCK_SHALLOW && repairer) { |
8340 | for (auto e : blob.get_extents()) { | |
8341 | if (!e.is_valid()) | |
8342 | continue; | |
8343 | repairer->set_space_used(e.offset, e.length, c->cid, oid); | |
8344 | } | |
8345 | } | |
eafe8130 TL |
8346 | if (blob.is_shared()) { |
8347 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
8348 | derr << "fsck error: " << oid << " blob " << blob | |
8349 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " | |
8350 | << blobid_max << dendl; | |
8351 | ++errors; | |
20effc67 | 8352 | } else if (i.first->shared_blob->get_sbid() == 0) { |
eafe8130 TL |
8353 | derr << "fsck error: " << oid << " blob " << blob |
8354 | << " marked as shared but has uninitialized sbid" | |
8355 | << dendl; | |
8356 | ++errors; | |
8357 | } | |
8358 | // the below lock is optional and provided in multithreading mode only | |
8359 | if (sb_info_lock) { | |
8360 | sb_info_lock->lock(); | |
8361 | } | |
20effc67 TL |
8362 | auto sbid = i.first->shared_blob->get_sbid(); |
8363 | sb_info_t& sbi = sb_info.add_or_adopt(i.first->shared_blob->get_sbid()); | |
8364 | ceph_assert(sbi.pool_id == sb_info_t::INVALID_POOL_ID || | |
eafe8130 | 8365 | sbi.pool_id == oid.hobj.get_logical_pool()); |
eafe8130 | 8366 | sbi.pool_id = oid.hobj.get_logical_pool(); |
20effc67 | 8367 | bool compressed = blob.is_compressed(); |
eafe8130 TL |
8368 | for (auto e : blob.get_extents()) { |
8369 | if (e.is_valid()) { | |
20effc67 TL |
8370 | if (compressed) { |
8371 | ceph_assert(sbi.allocated_chunks <= 0); | |
8372 | sbi.allocated_chunks -= (e.length >> min_alloc_size_order); | |
8373 | } else { | |
8374 | ceph_assert(sbi.allocated_chunks >= 0); | |
8375 | sbi.allocated_chunks += (e.length >> min_alloc_size_order); | |
8376 | } | |
8377 | sb_ref_counts.inc_range(sbid, e.offset, e.length, 1); | |
eafe8130 TL |
8378 | } |
8379 | } | |
8380 | if (sb_info_lock) { | |
8381 | sb_info_lock->unlock(); | |
8382 | } | |
8383 | } else if (depth != FSCK_SHALLOW) { | |
8384 | ceph_assert(used_blocks); | |
20effc67 TL |
8385 | string ctx_descr = " oid " + stringify(oid); |
8386 | errors += _fsck_check_extents(ctx_descr, | |
8387 | blob.get_extents(), | |
eafe8130 TL |
8388 | blob.is_compressed(), |
8389 | *used_blocks, | |
8390 | fm->get_alloc_size(), | |
20effc67 | 8391 | repairer, |
eafe8130 TL |
8392 | *res_statfs, |
8393 | depth); | |
8394 | } else { | |
8395 | errors += _fsck_sum_extents( | |
8396 | blob.get_extents(), | |
8397 | blob.is_compressed(), | |
8398 | *res_statfs); | |
8399 | } | |
8400 | } // for (auto& i : ref_map) | |
9f95a23c | 8401 | |
adb31ebb TL |
8402 | { |
8403 | auto &sbm = o->extent_map.spanning_blob_map; | |
8404 | size_t broken = 0; | |
8405 | BlobRef first_broken; | |
8406 | for (auto it = sbm.begin(); it != sbm.end();) { | |
8407 | auto it1 = it++; | |
8408 | if (ref_map.count(it1->second) == 0) { | |
8409 | if (!broken) { | |
8410 | first_broken = it1->second; | |
8411 | ++errors; | |
39ae355f TL |
8412 | derr << "fsck error:" << " stray spanning blob found:" << it1->first |
8413 | << dendl; | |
adb31ebb TL |
8414 | } |
8415 | broken++; | |
8416 | if (repairer) { | |
8417 | sbm.erase(it1); | |
8418 | } | |
8419 | } | |
8420 | } | |
20effc67 TL |
8421 | |
8422 | #ifdef HAVE_LIBZBD | |
8423 | if (bdev->is_smr() && depth != FSCK_SHALLOW) { | |
8424 | for (auto& [zone, first_offset] : zone_first_offsets) { | |
8425 | auto p = (*ctx.zone_refs)[zone].find(oid); | |
8426 | if (p != (*ctx.zone_refs)[zone].end()) { | |
8427 | if (first_offset < p->second) { | |
8428 | dout(20) << " slightly wonky zone ref 0x" << std::hex << zone | |
8429 | << " offset 0x" << p->second | |
8430 | << " but first offset is 0x" << first_offset | |
8431 | << "; this can happen due to clone_range" | |
8432 | << dendl; | |
8433 | } else { | |
8434 | dout(20) << " good zone ref 0x" << std::hex << zone << " offset 0x" << p->second | |
8435 | << " <= first offset 0x" << first_offset | |
8436 | << std::dec << dendl; | |
8437 | } | |
8438 | (*ctx.zone_refs)[zone].erase(p); | |
8439 | } else { | |
8440 | derr << "fsck error: " << oid << " references zone 0x" << std::hex << zone | |
8441 | << " but there is no zone ref" << std::dec << dendl; | |
8442 | // FIXME: add repair | |
8443 | ++errors; | |
8444 | } | |
8445 | } | |
8446 | } | |
8447 | #endif | |
8448 | ||
adb31ebb TL |
8449 | if (broken) { |
8450 | derr << "fsck error: " << oid << " - " << broken | |
8451 | << " zombie spanning blob(s) found, the first one: " | |
8452 | << *first_broken << dendl; | |
8453 | if(repairer) { | |
b3b6e05e TL |
8454 | repairer->fix_spanning_blobs( |
8455 | db, | |
8456 | [&](KeyValueDB::Transaction txn) { | |
8457 | _record_onode(o, txn); | |
8458 | }); | |
adb31ebb TL |
8459 | } |
8460 | } | |
8461 | } | |
8462 | ||
9f95a23c TL |
8463 | if (o->onode.has_omap()) { |
8464 | _fsck_check_object_omap(depth, o, ctx); | |
8465 | } | |
8466 | ||
eafe8130 TL |
8467 | return o; |
8468 | } | |
8469 | ||
8470 | #include "common/WorkQueue.h" | |
8471 | ||
8472 | class ShallowFSCKThreadPool : public ThreadPool | |
8473 | { | |
8474 | public: | |
8475 | ShallowFSCKThreadPool(CephContext* cct_, std::string nm, std::string tn, int n) : | |
8476 | ThreadPool(cct_, nm, tn, n) { | |
8477 | } | |
8478 | void worker(ThreadPool::WorkThread* wt) override { | |
8479 | int next_wq = 0; | |
8480 | while (!_stop) { | |
8481 | next_wq %= work_queues.size(); | |
8482 | WorkQueue_ *wq = work_queues[next_wq++]; | |
8483 | ||
8484 | void* item = wq->_void_dequeue(); | |
8485 | if (item) { | |
8486 | processing++; | |
8487 | TPHandle tp_handle(cct, nullptr, wq->timeout_interval, wq->suicide_interval); | |
8488 | wq->_void_process(item, tp_handle); | |
8489 | processing--; | |
8490 | } | |
8491 | } | |
8492 | } | |
8493 | template <size_t BatchLen> | |
8494 | struct FSCKWorkQueue : public ThreadPool::WorkQueue_ | |
8495 | { | |
8496 | struct Entry { | |
8497 | int64_t pool_id; | |
8498 | BlueStore::CollectionRef c; | |
8499 | ghobject_t oid; | |
8500 | string key; | |
8501 | bufferlist value; | |
8502 | }; | |
8503 | struct Batch { | |
8504 | std::atomic<size_t> running = { 0 }; | |
8505 | size_t entry_count = 0; | |
8506 | std::array<Entry, BatchLen> entries; | |
8507 | ||
8508 | int64_t errors = 0; | |
8509 | int64_t warnings = 0; | |
8510 | uint64_t num_objects = 0; | |
8511 | uint64_t num_extents = 0; | |
8512 | uint64_t num_blobs = 0; | |
8513 | uint64_t num_sharded_objects = 0; | |
8514 | uint64_t num_spanning_blobs = 0; | |
8515 | store_statfs_t expected_store_statfs; | |
8516 | BlueStore::per_pool_statfs expected_pool_statfs; | |
8517 | }; | |
8518 | ||
8519 | size_t batchCount; | |
8520 | BlueStore* store = nullptr; | |
8521 | ||
eafe8130 | 8522 | ceph::mutex* sb_info_lock = nullptr; |
20effc67 TL |
8523 | sb_info_space_efficient_map_t* sb_info = nullptr; |
8524 | shared_blob_2hash_tracker_t* sb_ref_counts = nullptr; | |
eafe8130 TL |
8525 | BlueStoreRepairer* repairer = nullptr; |
8526 | ||
8527 | Batch* batches = nullptr; | |
8528 | size_t last_batch_pos = 0; | |
8529 | bool batch_acquired = false; | |
8530 | ||
8531 | FSCKWorkQueue(std::string n, | |
8532 | size_t _batchCount, | |
8533 | BlueStore* _store, | |
eafe8130 | 8534 | ceph::mutex* _sb_info_lock, |
20effc67 TL |
8535 | sb_info_space_efficient_map_t& _sb_info, |
8536 | shared_blob_2hash_tracker_t& _sb_ref_counts, | |
eafe8130 | 8537 | BlueStoreRepairer* _repairer) : |
f67539c2 | 8538 | WorkQueue_(n, ceph::timespan::zero(), ceph::timespan::zero()), |
eafe8130 TL |
8539 | batchCount(_batchCount), |
8540 | store(_store), | |
eafe8130 TL |
8541 | sb_info_lock(_sb_info_lock), |
8542 | sb_info(&_sb_info), | |
20effc67 | 8543 | sb_ref_counts(&_sb_ref_counts), |
eafe8130 TL |
8544 | repairer(_repairer) |
8545 | { | |
8546 | batches = new Batch[batchCount]; | |
8547 | } | |
8548 | ~FSCKWorkQueue() { | |
8549 | delete[] batches; | |
8550 | } | |
8551 | ||
8552 | /// Remove all work items from the queue. | |
8553 | void _clear() override { | |
8554 | //do nothing | |
8555 | } | |
8556 | /// Check whether there is anything to do. | |
8557 | bool _empty() override { | |
8558 | ceph_assert(false); | |
8559 | } | |
8560 | ||
8561 | /// Get the next work item to process. | |
8562 | void* _void_dequeue() override { | |
8563 | size_t pos = rand() % batchCount; | |
8564 | size_t pos0 = pos; | |
8565 | do { | |
8566 | auto& batch = batches[pos]; | |
8567 | if (batch.running.fetch_add(1) == 0) { | |
8568 | if (batch.entry_count) { | |
8569 | return &batch; | |
8570 | } | |
8571 | } | |
8572 | batch.running--; | |
8573 | pos++; | |
8574 | pos %= batchCount; | |
8575 | } while (pos != pos0); | |
8576 | return nullptr; | |
8577 | } | |
8578 | /** @brief Process the work item. | |
8579 | * This function will be called several times in parallel | |
8580 | * and must therefore be thread-safe. */ | |
8581 | void _void_process(void* item, TPHandle& handle) override { | |
8582 | Batch* batch = (Batch*)item; | |
8583 | ||
8584 | BlueStore::FSCK_ObjectCtx ctx( | |
8585 | batch->errors, | |
8586 | batch->warnings, | |
8587 | batch->num_objects, | |
8588 | batch->num_extents, | |
8589 | batch->num_blobs, | |
8590 | batch->num_sharded_objects, | |
8591 | batch->num_spanning_blobs, | |
8592 | nullptr, // used_blocks | |
9f95a23c | 8593 | nullptr, //used_omap_head |
20effc67 | 8594 | nullptr, |
eafe8130 TL |
8595 | sb_info_lock, |
8596 | *sb_info, | |
20effc67 | 8597 | *sb_ref_counts, |
eafe8130 TL |
8598 | batch->expected_store_statfs, |
8599 | batch->expected_pool_statfs, | |
8600 | repairer); | |
8601 | ||
8602 | for (size_t i = 0; i < batch->entry_count; i++) { | |
8603 | auto& entry = batch->entries[i]; | |
8604 | ||
8605 | store->fsck_check_objects_shallow( | |
8606 | BlueStore::FSCK_SHALLOW, | |
8607 | entry.pool_id, | |
8608 | entry.c, | |
8609 | entry.oid, | |
8610 | entry.key, | |
8611 | entry.value, | |
9f95a23c | 8612 | nullptr, // expecting_shards - this will need a protection if passed |
eafe8130 TL |
8613 | nullptr, // referenced |
8614 | ctx); | |
8615 | } | |
eafe8130 TL |
8616 | batch->entry_count = 0; |
8617 | batch->running--; | |
8618 | } | |
8619 | /** @brief Synchronously finish processing a work item. | |
8620 | * This function is called after _void_process with the global thread pool lock held, | |
8621 | * so at most one copy will execute simultaneously for a given thread pool. | |
8622 | * It can be used for non-thread-safe finalization. */ | |
8623 | void _void_process_finish(void*) override { | |
8624 | ceph_assert(false); | |
8625 | } | |
8626 | ||
8627 | bool queue( | |
8628 | int64_t pool_id, | |
8629 | BlueStore::CollectionRef c, | |
8630 | const ghobject_t& oid, | |
8631 | const string& key, | |
8632 | const bufferlist& value) { | |
8633 | bool res = false; | |
8634 | size_t pos0 = last_batch_pos; | |
8635 | if (!batch_acquired) { | |
8636 | do { | |
8637 | auto& batch = batches[last_batch_pos]; | |
8638 | if (batch.running.fetch_add(1) == 0) { | |
8639 | if (batch.entry_count < BatchLen) { | |
8640 | batch_acquired = true; | |
8641 | break; | |
8642 | } | |
8643 | } | |
8644 | batch.running.fetch_sub(1); | |
8645 | last_batch_pos++; | |
8646 | last_batch_pos %= batchCount; | |
8647 | } while (last_batch_pos != pos0); | |
8648 | } | |
8649 | if (batch_acquired) { | |
8650 | auto& batch = batches[last_batch_pos]; | |
8651 | ceph_assert(batch.running); | |
8652 | ceph_assert(batch.entry_count < BatchLen); | |
8653 | ||
8654 | auto& entry = batch.entries[batch.entry_count]; | |
8655 | entry.pool_id = pool_id; | |
8656 | entry.c = c; | |
8657 | entry.oid = oid; | |
8658 | entry.key = key; | |
8659 | entry.value = value; | |
8660 | ||
8661 | ++batch.entry_count; | |
8662 | if (batch.entry_count == BatchLen) { | |
8663 | batch_acquired = false; | |
8664 | batch.running.fetch_sub(1); | |
8665 | last_batch_pos++; | |
8666 | last_batch_pos %= batchCount; | |
8667 | } | |
8668 | res = true; | |
8669 | } | |
8670 | return res; | |
8671 | } | |
8672 | ||
8673 | void finalize(ThreadPool& tp, | |
8674 | BlueStore::FSCK_ObjectCtx& ctx) { | |
8675 | if (batch_acquired) { | |
8676 | auto& batch = batches[last_batch_pos]; | |
8677 | ceph_assert(batch.running); | |
8678 | batch.running.fetch_sub(1); | |
8679 | } | |
8680 | tp.stop(); | |
8681 | ||
8682 | for (size_t i = 0; i < batchCount; i++) { | |
8683 | auto& batch = batches[i]; | |
8684 | ||
8685 | //process leftovers if any | |
8686 | if (batch.entry_count) { | |
8687 | TPHandle tp_handle(store->cct, | |
8688 | nullptr, | |
8689 | timeout_interval, | |
8690 | suicide_interval); | |
8691 | ceph_assert(batch.running == 0); | |
8692 | ||
8693 | batch.running++; // just to be on-par with the regular call | |
8694 | _void_process(&batch, tp_handle); | |
8695 | } | |
8696 | ceph_assert(batch.entry_count == 0); | |
8697 | ||
8698 | ctx.errors += batch.errors; | |
8699 | ctx.warnings += batch.warnings; | |
8700 | ctx.num_objects += batch.num_objects; | |
8701 | ctx.num_extents += batch.num_extents; | |
8702 | ctx.num_blobs += batch.num_blobs; | |
8703 | ctx.num_sharded_objects += batch.num_sharded_objects; | |
8704 | ctx.num_spanning_blobs += batch.num_spanning_blobs; | |
9f95a23c | 8705 | |
eafe8130 TL |
8706 | ctx.expected_store_statfs.add(batch.expected_store_statfs); |
8707 | ||
8708 | for (auto it = batch.expected_pool_statfs.begin(); | |
8709 | it != batch.expected_pool_statfs.end(); | |
8710 | it++) { | |
8711 | ctx.expected_pool_statfs[it->first].add(it->second); | |
8712 | } | |
8713 | } | |
8714 | } | |
8715 | }; | |
8716 | }; | |
8717 | ||
9f95a23c TL |
8718 | void BlueStore::_fsck_check_object_omap(FSCKDepth depth, |
8719 | OnodeRef& o, | |
8720 | const BlueStore::FSCK_ObjectCtx& ctx) | |
eafe8130 | 8721 | { |
9f95a23c TL |
8722 | auto& errors = ctx.errors; |
8723 | auto& warnings = ctx.warnings; | |
8724 | auto repairer = ctx.repairer; | |
8725 | ||
8726 | ceph_assert(o->onode.has_omap()); | |
8727 | if (!o->onode.is_perpool_omap() && !o->onode.is_pgmeta_omap()) { | |
f67539c2 | 8728 | if (per_pool_omap == OMAP_PER_POOL) { |
9f95a23c TL |
8729 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
8730 | << "fsck error: " << o->oid | |
8731 | << " has omap that is not per-pool or pgmeta" | |
8732 | << fsck_dendl; | |
8733 | ++errors; | |
8734 | } else { | |
8735 | const char* w; | |
8736 | int64_t num; | |
8737 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8738 | ++errors; | |
8739 | num = errors; | |
8740 | w = "error"; | |
8741 | } else { | |
8742 | ++warnings; | |
8743 | num = warnings; | |
8744 | w = "warning"; | |
8745 | } | |
8746 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
8747 | << "fsck " << w << ": " << o->oid | |
8748 | << " has omap that is not per-pool or pgmeta" | |
8749 | << fsck_dendl; | |
8750 | } | |
f67539c2 TL |
8751 | } else if (!o->onode.is_perpg_omap() && !o->onode.is_pgmeta_omap()) { |
8752 | if (per_pool_omap == OMAP_PER_PG) { | |
8753 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8754 | << "fsck error: " << o->oid | |
8755 | << " has omap that is not per-pg or pgmeta" | |
8756 | << fsck_dendl; | |
8757 | ++errors; | |
8758 | } else { | |
8759 | const char* w; | |
8760 | int64_t num; | |
8761 | if (cct->_conf->bluestore_fsck_error_on_no_per_pg_omap) { | |
8762 | ++errors; | |
8763 | num = errors; | |
8764 | w = "error"; | |
8765 | } else { | |
8766 | ++warnings; | |
8767 | num = warnings; | |
8768 | w = "warning"; | |
8769 | } | |
8770 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
8771 | << "fsck " << w << ": " << o->oid | |
8772 | << " has omap that is not per-pg or pgmeta" | |
8773 | << fsck_dendl; | |
8774 | } | |
9f95a23c TL |
8775 | } |
8776 | if (repairer && | |
f67539c2 | 8777 | !o->onode.is_perpg_omap() && |
9f95a23c | 8778 | !o->onode.is_pgmeta_omap()) { |
f67539c2 | 8779 | dout(10) << "fsck converting " << o->oid << " omap to per-pg" << dendl; |
522d829b | 8780 | bufferlist header; |
9f95a23c | 8781 | map<string, bufferlist> kv; |
522d829b TL |
8782 | { |
8783 | KeyValueDB::Transaction txn = db->get_transaction(); | |
8784 | uint64_t txn_cost = 0; | |
8785 | const string& prefix = Onode::calc_omap_prefix(o->onode.flags); | |
8786 | uint8_t new_flags = o->onode.flags | | |
8787 | bluestore_onode_t::FLAG_PERPOOL_OMAP | | |
8788 | bluestore_onode_t::FLAG_PERPG_OMAP; | |
8789 | const string& new_omap_prefix = Onode::calc_omap_prefix(new_flags); | |
8790 | ||
8791 | KeyValueDB::Iterator it = db->get_iterator(prefix); | |
8792 | string head, tail; | |
8793 | o->get_omap_header(&head); | |
8794 | o->get_omap_tail(&tail); | |
8795 | it->lower_bound(head); | |
8796 | // head | |
8797 | if (it->valid() && it->key() == head) { | |
8798 | dout(30) << __func__ << " got header" << dendl; | |
8799 | header = it->value(); | |
8800 | if (header.length()) { | |
8801 | string new_head; | |
8802 | Onode::calc_omap_header(new_flags, o.get(), &new_head); | |
8803 | txn->set(new_omap_prefix, new_head, header); | |
8804 | txn_cost += new_head.length() + header.length(); | |
8805 | } | |
a4b75251 | 8806 | it->next(); |
522d829b TL |
8807 | } |
8808 | // tail | |
8809 | { | |
8810 | string new_tail; | |
8811 | Onode::calc_omap_tail(new_flags, o.get(), &new_tail); | |
8812 | bufferlist empty; | |
8813 | txn->set(new_omap_prefix, new_tail, empty); | |
8814 | txn_cost += new_tail.length() + new_tail.length(); | |
8815 | } | |
8816 | // values | |
8817 | string final_key; | |
8818 | Onode::calc_omap_key(new_flags, o.get(), string(), &final_key); | |
8819 | size_t base_key_len = final_key.size(); | |
8820 | while (it->valid() && it->key() < tail) { | |
8821 | string user_key; | |
8822 | o->decode_omap_key(it->key(), &user_key); | |
8823 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
8824 | << " -> " << user_key << dendl; | |
8825 | ||
8826 | final_key.resize(base_key_len); | |
a4b75251 | 8827 | final_key += user_key; |
522d829b TL |
8828 | auto v = it->value(); |
8829 | txn->set(new_omap_prefix, final_key, v); | |
8830 | txn_cost += final_key.length() + v.length(); | |
8831 | ||
8832 | // submit a portion if cost exceeds 16MB | |
8833 | if (txn_cost >= 16 * (1 << 20) ) { | |
8834 | db->submit_transaction_sync(txn); | |
8835 | txn = db->get_transaction(); | |
8836 | txn_cost = 0; | |
8837 | } | |
8838 | it->next(); | |
8839 | } | |
8840 | if (txn_cost > 0) { | |
8841 | db->submit_transaction_sync(txn); | |
8842 | } | |
8843 | } | |
8844 | // finalize: remove legacy data | |
8845 | { | |
9f95a23c TL |
8846 | KeyValueDB::Transaction txn = db->get_transaction(); |
8847 | // remove old keys | |
8848 | const string& old_omap_prefix = o->get_omap_prefix(); | |
8849 | string old_head, old_tail; | |
8850 | o->get_omap_header(&old_head); | |
8851 | o->get_omap_tail(&old_tail); | |
8852 | txn->rm_range_keys(old_omap_prefix, old_head, old_tail); | |
8853 | txn->rmkey(old_omap_prefix, old_tail); | |
8854 | // set flag | |
f67539c2 | 8855 | o->onode.set_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP | bluestore_onode_t::FLAG_PERPG_OMAP); |
9f95a23c | 8856 | _record_onode(o, txn); |
9f95a23c TL |
8857 | db->submit_transaction_sync(txn); |
8858 | repairer->inc_repaired(); | |
522d829b | 8859 | repairer->request_compaction(); |
9f95a23c | 8860 | } |
eafe8130 | 8861 | } |
9f95a23c | 8862 | } |
eafe8130 | 8863 | |
20effc67 TL |
8864 | void BlueStore::_fsck_check_objects( |
8865 | FSCKDepth depth, | |
9f95a23c TL |
8866 | BlueStore::FSCK_ObjectCtx& ctx) |
8867 | { | |
eafe8130 | 8868 | auto& errors = ctx.errors; |
eafe8130 TL |
8869 | auto sb_info_lock = ctx.sb_info_lock; |
8870 | auto& sb_info = ctx.sb_info; | |
20effc67 | 8871 | auto& sb_ref_counts = ctx.sb_ref_counts; |
eafe8130 TL |
8872 | auto repairer = ctx.repairer; |
8873 | ||
8874 | uint64_t_btree_t used_nids; | |
8875 | ||
8876 | size_t processed_myself = 0; | |
8877 | ||
f67539c2 | 8878 | auto it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
8879 | mempool::bluestore_fsck::list<string> expecting_shards; |
8880 | if (it) { | |
8881 | const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads; | |
8882 | typedef ShallowFSCKThreadPool::FSCKWorkQueue<256> WQ; | |
8883 | std::unique_ptr<WQ> wq( | |
8884 | new WQ( | |
8885 | "FSCKWorkQueue", | |
8886 | (thread_count ? : 1) * 32, | |
8887 | this, | |
eafe8130 TL |
8888 | sb_info_lock, |
8889 | sb_info, | |
20effc67 | 8890 | sb_ref_counts, |
eafe8130 TL |
8891 | repairer)); |
8892 | ||
8893 | ShallowFSCKThreadPool thread_pool(cct, "ShallowFSCKThreadPool", "ShallowFSCK", thread_count); | |
8894 | ||
8895 | thread_pool.add_work_queue(wq.get()); | |
8896 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8897 | //not the best place but let's check anyway | |
8898 | ceph_assert(sb_info_lock); | |
8899 | thread_pool.start(); | |
8900 | } | |
8901 | ||
20effc67 | 8902 | // fill global if not overriden below |
eafe8130 TL |
8903 | CollectionRef c; |
8904 | int64_t pool_id = -1; | |
8905 | spg_t pgid; | |
8906 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8907 | dout(30) << __func__ << " key " | |
8908 | << pretty_binary_string(it->key()) << dendl; | |
8909 | if (is_extent_shard_key(it->key())) { | |
8910 | if (depth == FSCK_SHALLOW) { | |
8911 | continue; | |
8912 | } | |
8913 | while (!expecting_shards.empty() && | |
8914 | expecting_shards.front() < it->key()) { | |
8915 | derr << "fsck error: missing shard key " | |
8916 | << pretty_binary_string(expecting_shards.front()) | |
8917 | << dendl; | |
8918 | ++errors; | |
8919 | expecting_shards.pop_front(); | |
8920 | } | |
8921 | if (!expecting_shards.empty() && | |
8922 | expecting_shards.front() == it->key()) { | |
8923 | // all good | |
8924 | expecting_shards.pop_front(); | |
8925 | continue; | |
8926 | } | |
8927 | ||
8928 | uint32_t offset; | |
8929 | string okey; | |
8930 | get_key_extent_shard(it->key(), &okey, &offset); | |
8931 | derr << "fsck error: stray shard 0x" << std::hex << offset | |
8932 | << std::dec << dendl; | |
8933 | if (expecting_shards.empty()) { | |
8934 | derr << "fsck error: " << pretty_binary_string(it->key()) | |
8935 | << " is unexpected" << dendl; | |
8936 | ++errors; | |
8937 | continue; | |
8938 | } | |
8939 | while (expecting_shards.front() > it->key()) { | |
8940 | derr << "fsck error: saw " << pretty_binary_string(it->key()) | |
8941 | << dendl; | |
8942 | derr << "fsck error: exp " | |
8943 | << pretty_binary_string(expecting_shards.front()) << dendl; | |
8944 | ++errors; | |
8945 | expecting_shards.pop_front(); | |
8946 | if (expecting_shards.empty()) { | |
8947 | break; | |
8948 | } | |
8949 | } | |
8950 | continue; | |
8951 | } | |
8952 | ||
8953 | ghobject_t oid; | |
8954 | int r = get_key_object(it->key(), &oid); | |
8955 | if (r < 0) { | |
8956 | derr << "fsck error: bad object key " | |
8957 | << pretty_binary_string(it->key()) << dendl; | |
8958 | ++errors; | |
8959 | continue; | |
8960 | } | |
8961 | if (!c || | |
8962 | oid.shard_id != pgid.shard || | |
8963 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8964 | !c->contains(oid)) { | |
8965 | c = nullptr; | |
8966 | for (auto& p : coll_map) { | |
8967 | if (p.second->contains(oid)) { | |
8968 | c = p.second; | |
8969 | break; | |
8970 | } | |
8971 | } | |
8972 | if (!c) { | |
8973 | derr << "fsck error: stray object " << oid | |
8974 | << " not owned by any collection" << dendl; | |
8975 | ++errors; | |
8976 | continue; | |
8977 | } | |
8978 | pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
8979 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
8980 | << dendl; | |
8981 | } | |
8982 | ||
8983 | if (depth != FSCK_SHALLOW && | |
8984 | !expecting_shards.empty()) { | |
8985 | for (auto& k : expecting_shards) { | |
8986 | derr << "fsck error: missing shard key " | |
8987 | << pretty_binary_string(k) << dendl; | |
8988 | } | |
8989 | ++errors; | |
8990 | expecting_shards.clear(); | |
8991 | } | |
8992 | ||
8993 | bool queued = false; | |
8994 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8995 | queued = wq->queue( | |
8996 | pool_id, | |
8997 | c, | |
8998 | oid, | |
8999 | it->key(), | |
9000 | it->value()); | |
9001 | } | |
9002 | OnodeRef o; | |
9003 | map<BlobRef, bluestore_blob_t::unused_t> referenced; | |
9004 | ||
9005 | if (!queued) { | |
9006 | ++processed_myself; | |
eafe8130 TL |
9007 | o = fsck_check_objects_shallow( |
9008 | depth, | |
9009 | pool_id, | |
9010 | c, | |
9011 | oid, | |
9012 | it->key(), | |
9013 | it->value(), | |
9f95a23c | 9014 | &expecting_shards, |
eafe8130 TL |
9015 | &referenced, |
9016 | ctx); | |
9017 | } | |
9018 | ||
9019 | if (depth != FSCK_SHALLOW) { | |
9020 | ceph_assert(o != nullptr); | |
9021 | if (o->onode.nid) { | |
9022 | if (o->onode.nid > nid_max) { | |
9023 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
9024 | << " > nid_max " << nid_max << dendl; | |
9025 | ++errors; | |
9026 | } | |
9027 | if (used_nids.count(o->onode.nid)) { | |
9028 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
9029 | << " already in use" << dendl; | |
9030 | ++errors; | |
9031 | continue; // go for next object | |
9032 | } | |
9033 | used_nids.insert(o->onode.nid); | |
9034 | } | |
9035 | for (auto& i : referenced) { | |
9036 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
9037 | << std::dec << " for " << *i.first << dendl; | |
9038 | const bluestore_blob_t& blob = i.first->get_blob(); | |
9039 | if (i.second & blob.unused) { | |
9040 | derr << "fsck error: " << oid << " blob claims unused 0x" | |
9041 | << std::hex << blob.unused | |
9042 | << " but extents reference 0x" << i.second << std::dec | |
9043 | << " on blob " << *i.first << dendl; | |
9044 | ++errors; | |
9045 | } | |
9046 | if (blob.has_csum()) { | |
9047 | uint64_t blob_len = blob.get_logical_length(); | |
9048 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused) * 8); | |
9049 | unsigned csum_count = blob.get_csum_count(); | |
9050 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
9051 | for (unsigned p = 0; p < csum_count; ++p) { | |
9052 | unsigned pos = p * csum_chunk_size; | |
9053 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
9054 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
9055 | unsigned mask = 1u << firstbit; | |
9056 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
9057 | mask |= 1u << b; | |
9058 | } | |
9059 | if ((blob.unused & mask) == mask) { | |
9060 | // this csum chunk region is marked unused | |
9061 | if (blob.get_csum_item(p) != 0) { | |
9062 | derr << "fsck error: " << oid | |
9063 | << " blob claims csum chunk 0x" << std::hex << pos | |
9064 | << "~" << csum_chunk_size | |
9065 | << " is unused (mask 0x" << mask << " of unused 0x" | |
9066 | << blob.unused << ") but csum is non-zero 0x" | |
9067 | << blob.get_csum_item(p) << std::dec << " on blob " | |
9068 | << *i.first << dendl; | |
9069 | ++errors; | |
9070 | } | |
9071 | } | |
9072 | } | |
9073 | } | |
9074 | } | |
9075 | // omap | |
9076 | if (o->onode.has_omap()) { | |
9f95a23c TL |
9077 | ceph_assert(ctx.used_omap_head); |
9078 | if (ctx.used_omap_head->count(o->onode.nid)) { | |
9079 | derr << "fsck error: " << o->oid << " omap_head " << o->onode.nid | |
9080 | << " already in use" << dendl; | |
eafe8130 TL |
9081 | ++errors; |
9082 | } else { | |
9f95a23c | 9083 | ctx.used_omap_head->insert(o->onode.nid); |
eafe8130 | 9084 | } |
9f95a23c | 9085 | } // if (o->onode.has_omap()) |
eafe8130 TL |
9086 | if (depth == FSCK_DEEP) { |
9087 | bufferlist bl; | |
9088 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; | |
9089 | uint64_t offset = 0; | |
9090 | do { | |
9091 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
9092 | int r = _do_read(c.get(), o, offset, l, bl, | |
9093 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
9094 | if (r < 0) { | |
9095 | ++errors; | |
9096 | derr << "fsck error: " << oid << std::hex | |
9097 | << " error during read: " | |
9098 | << " " << offset << "~" << l | |
9099 | << " " << cpp_strerror(r) << std::dec | |
9100 | << dendl; | |
9101 | break; | |
9102 | } | |
9103 | offset += l; | |
9104 | } while (offset < o->onode.size); | |
9105 | } // deep | |
9106 | } //if (depth != FSCK_SHALLOW) | |
9107 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
9108 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
9109 | wq->finalize(thread_pool, ctx); | |
9110 | if (processed_myself) { | |
9111 | // may be needs more threads? | |
9112 | dout(0) << __func__ << " partial offload" | |
9113 | << ", done myself " << processed_myself | |
9114 | << " of " << ctx.num_objects | |
9115 | << "objects, threads " << thread_count | |
9116 | << dendl; | |
9117 | } | |
9118 | } | |
9119 | } // if (it) | |
9120 | } | |
9121 | /** | |
9122 | An overview for currently implemented repair logics | |
9123 | performed in fsck in two stages: detection(+preparation) and commit. | |
9124 | Detection stage (in processing order): | |
9125 | (Issue -> Repair action to schedule) | |
9126 | - Detect undecodable keys for Shared Blobs -> Remove | |
9127 | - Detect undecodable records for Shared Blobs -> Remove | |
9128 | (might trigger missed Shared Blob detection below) | |
9129 | - Detect stray records for Shared Blobs -> Remove | |
9130 | - Detect misreferenced pextents -> Fix | |
9131 | Prepare Bloom-like filter to track cid/oid -> pextent | |
9132 | Prepare list of extents that are improperly referenced | |
9133 | Enumerate Onode records that might use 'misreferenced' pextents | |
9134 | (Bloom-like filter applied to reduce computation) | |
9135 | Per each questinable Onode enumerate all blobs and identify broken ones | |
9136 | (i.e. blobs having 'misreferences') | |
9137 | Rewrite each broken blob data by allocating another extents and | |
9138 | copying data there | |
9139 | If blob is shared - unshare it and mark corresponding Shared Blob | |
9140 | for removal | |
9141 | Release previously allocated space | |
9142 | Update Extent Map | |
9143 | - Detect missed Shared Blobs -> Recreate | |
9144 | - Detect undecodable deferred transaction -> Remove | |
9145 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
9146 | - Detect Freelist Manager's leaked entries -> Mark as free | |
9147 | - Detect statfs inconsistency - Update | |
9148 | Commit stage (separate DB commit per each step): | |
9149 | - Apply leaked FM entries fix | |
9150 | - Apply 'false free' FM entries fix | |
9151 | - Apply 'Remove' actions | |
9152 | - Apply fix for misreference pextents | |
9153 | - Apply Shared Blob recreate | |
9154 | (can be merged with the step above if misreferences were dectected) | |
9155 | - Apply StatFS update | |
9156 | */ | |
9157 | int BlueStore::_fsck(BlueStore::FSCKDepth depth, bool repair) | |
9158 | { | |
20effc67 | 9159 | dout(5) << __func__ |
eafe8130 TL |
9160 | << (repair ? " repair" : " check") |
9161 | << (depth == FSCK_DEEP ? " (deep)" : | |
9162 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
9163 | << dendl; | |
9164 | ||
9165 | // in deep mode we need R/W write access to be able to replay deferred ops | |
20effc67 | 9166 | const bool read_only = !(repair || depth == FSCK_DEEP); |
f67539c2 | 9167 | int r = _open_db_and_around(read_only); |
20effc67 | 9168 | if (r < 0) { |
eafe8130 | 9169 | return r; |
20effc67 TL |
9170 | } |
9171 | auto close_db = make_scope_guard([&] { | |
9172 | _close_db_and_around(); | |
9173 | }); | |
7c673cae | 9174 | |
11fdf7f2 TL |
9175 | if (!read_only) { |
9176 | r = _upgrade_super(); | |
9177 | if (r < 0) { | |
20effc67 | 9178 | return r; |
11fdf7f2 TL |
9179 | } |
9180 | } | |
7c673cae | 9181 | |
20effc67 | 9182 | // NullFreelistManager needs to open collection early |
eafe8130 | 9183 | r = _open_collections(); |
20effc67 TL |
9184 | if (r < 0) { |
9185 | return r; | |
9186 | } | |
7c673cae FG |
9187 | |
9188 | mempool_thread.init(); | |
20effc67 TL |
9189 | auto stop_mempool = make_scope_guard([&] { |
9190 | mempool_thread.shutdown(); | |
9191 | _shutdown_cache(); | |
9192 | }); | |
11fdf7f2 TL |
9193 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
9194 | // enable in repair or deep mode modes only | |
9195 | if (!read_only) { | |
9196 | _kv_start(); | |
9197 | r = _deferred_replay(); | |
9198 | _kv_stop(); | |
9199 | } | |
eafe8130 | 9200 | |
20effc67 TL |
9201 | if (r < 0) { |
9202 | return r; | |
9203 | } | |
9204 | return _fsck_on_open(depth, repair); | |
eafe8130 TL |
9205 | } |
9206 | ||
9207 | int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair) | |
9208 | { | |
20effc67 TL |
9209 | uint64_t sb_hash_size = uint64_t( |
9210 | cct->_conf.get_val<Option::size_t>("osd_memory_target") * | |
9211 | cct->_conf.get_val<double>( | |
9212 | "bluestore_fsck_shared_blob_tracker_size")); | |
9213 | ||
eafe8130 TL |
9214 | dout(1) << __func__ |
9215 | << " <<<START>>>" | |
9216 | << (repair ? " repair" : " check") | |
9217 | << (depth == FSCK_DEEP ? " (deep)" : | |
9218 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
20effc67 TL |
9219 | << " start sb_tracker_hash_size:" << sb_hash_size |
9220 | << dendl; | |
eafe8130 TL |
9221 | int64_t errors = 0; |
9222 | int64_t warnings = 0; | |
9223 | unsigned repaired = 0; | |
9224 | ||
9225 | uint64_t_btree_t used_omap_head; | |
eafe8130 TL |
9226 | uint64_t_btree_t used_sbids; |
9227 | ||
f67539c2 | 9228 | mempool_dynamic_bitset used_blocks, bluefs_used_blocks; |
eafe8130 | 9229 | KeyValueDB::Iterator it; |
39ae355f | 9230 | store_statfs_t expected_store_statfs; |
eafe8130 TL |
9231 | per_pool_statfs expected_pool_statfs; |
9232 | ||
20effc67 TL |
9233 | sb_info_space_efficient_map_t sb_info; |
9234 | shared_blob_2hash_tracker_t sb_ref_counts( | |
9235 | sb_hash_size, | |
9236 | min_alloc_size); | |
9237 | size_t sb_ref_mismatches = 0; | |
9238 | ||
9239 | /// map of oid -> (first_)offset for each zone | |
9240 | std::vector<std::unordered_map<ghobject_t, uint64_t>> zone_refs; // FIXME: this may be a lot of RAM! | |
eafe8130 TL |
9241 | |
9242 | uint64_t num_objects = 0; | |
9243 | uint64_t num_extents = 0; | |
9244 | uint64_t num_blobs = 0; | |
9245 | uint64_t num_spanning_blobs = 0; | |
9246 | uint64_t num_shared_blobs = 0; | |
9247 | uint64_t num_sharded_objects = 0; | |
9248 | BlueStoreRepairer repairer; | |
9249 | ||
f67539c2 TL |
9250 | auto alloc_size = fm->get_alloc_size(); |
9251 | ||
eafe8130 TL |
9252 | utime_t start = ceph_clock_now(); |
9253 | ||
9254 | _fsck_collections(&errors); | |
b32b8144 | 9255 | used_blocks.resize(fm->get_alloc_units()); |
7c673cae FG |
9256 | |
9257 | if (bluefs) { | |
f67539c2 | 9258 | interval_set<uint64_t> bluefs_extents; |
11fdf7f2 | 9259 | |
1e59de90 TL |
9260 | bluefs->foreach_block_extents( |
9261 | bluefs_layout.shared_bdev, | |
9262 | [&](uint64_t start, uint32_t len) { | |
9263 | apply_for_bitset_range(start, len, alloc_size, used_blocks, | |
9264 | [&](uint64_t pos, mempool_dynamic_bitset& bs) { | |
9265 | ceph_assert(pos < bs.size()); | |
9266 | bs.set(pos); | |
9267 | } | |
9268 | ); | |
9269 | } | |
9270 | ); | |
f67539c2 TL |
9271 | } |
9272 | ||
9273 | bluefs_used_blocks = used_blocks; | |
9274 | ||
9275 | apply_for_bitset_range( | |
9276 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), alloc_size, used_blocks, | |
9277 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
9278 | bs.set(pos); | |
7c673cae | 9279 | } |
f67539c2 TL |
9280 | ); |
9281 | ||
9282 | ||
9283 | if (repair) { | |
b3b6e05e | 9284 | repairer.init_space_usage_tracker( |
f67539c2 TL |
9285 | bdev->get_size(), |
9286 | min_alloc_size); | |
9287 | } | |
9288 | ||
9289 | if (bluefs) { | |
eafe8130 | 9290 | int r = bluefs->fsck(); |
7c673cae | 9291 | if (r < 0) { |
eafe8130 | 9292 | return r; |
7c673cae FG |
9293 | } |
9294 | if (r > 0) | |
9295 | errors += r; | |
9296 | } | |
9297 | ||
eafe8130 TL |
9298 | if (!per_pool_stat_collection) { |
9299 | const char *w; | |
9300 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_stats) { | |
9301 | w = "error"; | |
9302 | ++errors; | |
9303 | } else { | |
9304 | w = "warning"; | |
9305 | ++warnings; | |
9306 | } | |
9307 | derr << "fsck " << w << ": store not yet converted to per-pool stats" | |
9308 | << dendl; | |
9309 | } | |
f67539c2 | 9310 | if (per_pool_omap != OMAP_PER_PG) { |
9f95a23c TL |
9311 | const char *w; |
9312 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
9313 | w = "error"; | |
9314 | ++errors; | |
9315 | } else { | |
9316 | w = "warning"; | |
9317 | ++warnings; | |
9318 | } | |
f67539c2 | 9319 | derr << "fsck " << w << ": store not yet converted to per-pg omap" |
9f95a23c TL |
9320 | << dendl; |
9321 | } | |
9322 | ||
eafe8130 TL |
9323 | if (g_conf()->bluestore_debug_fsck_abort) { |
9324 | dout(1) << __func__ << " debug abort" << dendl; | |
9325 | goto out_scan; | |
9326 | } | |
20effc67 TL |
9327 | |
9328 | #ifdef HAVE_LIBZBD | |
9329 | if (bdev->is_smr()) { | |
9330 | auto a = dynamic_cast<ZonedAllocator*>(alloc); | |
9331 | ceph_assert(a); | |
9332 | auto f = dynamic_cast<ZonedFreelistManager*>(fm); | |
9333 | ceph_assert(f); | |
9334 | vector<uint64_t> wp = bdev->get_zones(); | |
9335 | vector<zone_state_t> zones = f->get_zone_states(db); | |
9336 | ceph_assert(wp.size() == zones.size()); | |
9337 | auto num_zones = bdev->get_size() / zone_size; | |
9338 | for (unsigned i = first_sequential_zone; i < num_zones; ++i) { | |
9339 | uint64_t p = wp[i] == (i + 1) * zone_size ? zone_size : wp[i] % zone_size; | |
9340 | if (zones[i].write_pointer > p && | |
9341 | zones[i].num_dead_bytes < zones[i].write_pointer) { | |
9342 | derr << "fsck error: zone 0x" << std::hex << i | |
9343 | << " bluestore write pointer 0x" << zones[i].write_pointer | |
9344 | << " > device write pointer 0x" << p | |
9345 | << " (with only 0x" << zones[i].num_dead_bytes << " dead bytes)" | |
9346 | << std::dec << dendl; | |
9347 | ++errors; | |
9348 | } | |
9349 | } | |
9350 | ||
9351 | if (depth != FSCK_SHALLOW) { | |
9352 | // load zone refs | |
9353 | zone_refs.resize(bdev->get_size() / zone_size); | |
9354 | it = db->get_iterator(PREFIX_ZONED_CL_INFO, KeyValueDB::ITERATOR_NOCACHE); | |
9355 | if (it) { | |
9356 | for (it->lower_bound(string()); | |
9357 | it->valid(); | |
9358 | it->next()) { | |
9359 | uint32_t zone = 0; | |
9360 | uint64_t offset = 0; | |
9361 | ghobject_t oid; | |
9362 | string key = it->key(); | |
9363 | int r = get_key_zone_offset_object(key, &zone, &offset, &oid); | |
9364 | if (r < 0) { | |
9365 | derr << "fsck error: invalid zone ref key " << pretty_binary_string(key) | |
9366 | << dendl; | |
9367 | if (repair) { | |
9368 | repairer.remove_key(db, PREFIX_ZONED_CL_INFO, key); | |
9369 | } | |
9370 | ++errors; | |
9371 | continue; | |
9372 | } | |
9373 | dout(30) << " zone ref 0x" << std::hex << zone << " offset 0x" << offset | |
9374 | << " -> " << std::dec << oid << dendl; | |
9375 | if (zone_refs[zone].count(oid)) { | |
9376 | derr << "fsck error: second zone ref in zone 0x" << std::hex << zone | |
9377 | << " offset 0x" << offset << std::dec << " for " << oid << dendl; | |
9378 | if (repair) { | |
9379 | repairer.remove_key(db, PREFIX_ZONED_CL_INFO, key); | |
9380 | } | |
9381 | ++errors; | |
9382 | continue; | |
9383 | } | |
9384 | zone_refs[zone][oid] = offset; | |
9385 | } | |
9386 | } | |
9387 | } | |
9388 | } | |
9389 | #endif | |
9390 | ||
9391 | dout(1) << __func__ << " checking shared_blobs (phase 1)" << dendl; | |
9392 | it = db->get_iterator(PREFIX_SHARED_BLOB, KeyValueDB::ITERATOR_NOCACHE); | |
9393 | if (it) { | |
9394 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
9395 | string key = it->key(); | |
9396 | uint64_t sbid; | |
9397 | if (get_key_shared_blob(key, &sbid) < 0) { | |
9398 | // Failed to parse the key. | |
9399 | // This gonna to be handled at the second stage | |
9400 | continue; | |
9401 | } | |
9402 | bluestore_shared_blob_t shared_blob(sbid); | |
9403 | bufferlist bl = it->value(); | |
9404 | auto blp = bl.cbegin(); | |
9405 | try { | |
9406 | decode(shared_blob, blp); | |
9407 | } | |
9408 | catch (ceph::buffer::error& e) { | |
9409 | // this gonna to be handled at the second stage | |
9410 | continue; | |
9411 | } | |
9412 | dout(20) << __func__ << " " << shared_blob << dendl; | |
9413 | auto& sbi = sb_info.add_maybe_stray(sbid); | |
9414 | ||
9415 | // primarily to silent the 'unused' warning | |
9416 | ceph_assert(sbi.pool_id == sb_info_t::INVALID_POOL_ID); | |
9417 | ||
9418 | for (auto& r : shared_blob.ref_map.ref_map) { | |
9419 | sb_ref_counts.inc_range( | |
9420 | sbid, | |
9421 | r.first, | |
9422 | r.second.length, | |
9423 | -r.second.refs); | |
9424 | } | |
9425 | } | |
9426 | } // if (it) //checking shared_blobs (phase1) | |
9427 | ||
7c673cae | 9428 | // walk PREFIX_OBJ |
eafe8130 TL |
9429 | { |
9430 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
9431 | ceph::mutex sb_info_lock = ceph::make_mutex("BlueStore::fsck::sbinfo_lock"); | |
9432 | BlueStore::FSCK_ObjectCtx ctx( | |
9433 | errors, | |
9434 | warnings, | |
9435 | num_objects, | |
9436 | num_extents, | |
9437 | num_blobs, | |
9438 | num_sharded_objects, | |
9439 | num_spanning_blobs, | |
9440 | &used_blocks, | |
9441 | &used_omap_head, | |
20effc67 | 9442 | &zone_refs, |
9f95a23c TL |
9443 | //no need for the below lock when in non-shallow mode as |
9444 | // there is no multithreading in this case | |
9445 | depth == FSCK_SHALLOW ? &sb_info_lock : nullptr, | |
eafe8130 | 9446 | sb_info, |
20effc67 | 9447 | sb_ref_counts, |
eafe8130 TL |
9448 | expected_store_statfs, |
9449 | expected_pool_statfs, | |
9450 | repair ? &repairer : nullptr); | |
9f95a23c TL |
9451 | |
9452 | _fsck_check_objects(depth, ctx); | |
eafe8130 | 9453 | } |
11fdf7f2 | 9454 | |
20effc67 TL |
9455 | #ifdef HAVE_LIBZBD |
9456 | if (bdev->is_smr() && depth != FSCK_SHALLOW) { | |
9457 | dout(1) << __func__ << " checking for leaked zone refs" << dendl; | |
9458 | for (uint32_t zone = 0; zone < zone_refs.size(); ++zone) { | |
9459 | for (auto& [oid, offset] : zone_refs[zone]) { | |
9460 | derr << "fsck error: stray zone ref 0x" << std::hex << zone | |
9461 | << " offset 0x" << offset << " -> " << std::dec << oid << dendl; | |
9462 | // FIXME: add repair | |
9463 | ++errors; | |
9464 | } | |
9465 | } | |
9466 | } | |
9467 | #endif | |
9468 | ||
9469 | sb_ref_mismatches = sb_ref_counts.count_non_zero(); | |
9470 | if (sb_ref_mismatches != 0) { | |
39ae355f TL |
9471 | derr << "fsck error:" << "*" << sb_ref_mismatches |
9472 | << " shared blob references aren't matching, at least " | |
9473 | << sb_ref_mismatches << " found" << dendl; | |
20effc67 TL |
9474 | errors += sb_ref_mismatches; |
9475 | } | |
9476 | ||
9477 | if (depth != FSCK_SHALLOW && repair) { | |
9478 | _fsck_repair_shared_blobs(repairer, sb_ref_counts, sb_info); | |
9479 | } | |
9480 | dout(1) << __func__ << " checking shared_blobs (phase 2)" << dendl; | |
f67539c2 | 9481 | it = db->get_iterator(PREFIX_SHARED_BLOB, KeyValueDB::ITERATOR_NOCACHE); |
7c673cae | 9482 | if (it) { |
eafe8130 TL |
9483 | // FIXME minor: perhaps simplify for shallow mode? |
9484 | // fill global if not overriden below | |
9485 | auto expected_statfs = &expected_store_statfs; | |
7c673cae FG |
9486 | for (it->lower_bound(string()); it->valid(); it->next()) { |
9487 | string key = it->key(); | |
9488 | uint64_t sbid; | |
9489 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 9490 | derr << "fsck error: bad key '" << key |
20effc67 | 9491 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
9492 | if (repair) { |
9493 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
9494 | } | |
7c673cae FG |
9495 | ++errors; |
9496 | continue; | |
9497 | } | |
9498 | auto p = sb_info.find(sbid); | |
9499 | if (p == sb_info.end()) { | |
20effc67 TL |
9500 | if (sb_ref_mismatches > 0) { |
9501 | // highly likely this has been already reported before, ignoring... | |
9502 | dout(5) << __func__ << " found duplicate(?) stray shared blob data for sbid 0x" | |
9503 | << std::hex << sbid << std::dec << dendl; | |
9504 | } else { | |
9505 | derr<< "fsck error: found stray shared blob data for sbid 0x" | |
9506 | << std::hex << sbid << std::dec << dendl; | |
9507 | ++errors; | |
9508 | if (repair) { | |
9509 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
9510 | } | |
11fdf7f2 | 9511 | } |
7c673cae FG |
9512 | } else { |
9513 | ++num_shared_blobs; | |
20effc67 | 9514 | sb_info_t& sbi = *p; |
7c673cae FG |
9515 | bluestore_shared_blob_t shared_blob(sbid); |
9516 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
9517 | auto blp = bl.cbegin(); |
9518 | try { | |
20effc67 TL |
9519 | decode(shared_blob, blp); |
9520 | } | |
9521 | catch (ceph::buffer::error& e) { | |
7c673cae | 9522 | ++errors; |
20effc67 TL |
9523 | |
9524 | derr << "fsck error: failed to decode Shared Blob" | |
9525 | << pretty_binary_string(key) << dendl; | |
9526 | if (repair) { | |
9527 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
9528 | << pretty_binary_string(key) | |
9529 | << "', removing" << dendl; | |
9530 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
9531 | } | |
9532 | continue; | |
7c673cae | 9533 | } |
20effc67 | 9534 | dout(20) << __func__ << " " << shared_blob << dendl; |
7c673cae | 9535 | PExtentVector extents; |
20effc67 | 9536 | for (auto& r : shared_blob.ref_map.ref_map) { |
7c673cae FG |
9537 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); |
9538 | } | |
20effc67 TL |
9539 | if (sbi.pool_id != sb_info_t::INVALID_POOL_ID && |
9540 | (per_pool_stat_collection || repair)) { | |
11fdf7f2 TL |
9541 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; |
9542 | } | |
20effc67 TL |
9543 | std::stringstream ss; |
9544 | ss << "sbid 0x" << std::hex << sbid << std::dec; | |
9545 | errors += _fsck_check_extents(ss.str(), | |
9546 | extents, | |
9547 | sbi.allocated_chunks < 0, | |
9548 | used_blocks, | |
9549 | fm->get_alloc_size(), | |
9550 | repair ? &repairer : nullptr, | |
9551 | *expected_statfs, | |
9552 | depth); | |
11fdf7f2 TL |
9553 | } |
9554 | } | |
20effc67 | 9555 | } // if (it) /* checking shared_blobs (phase 2)*/ |
11fdf7f2 TL |
9556 | |
9557 | if (repair && repairer.preprocess_misreference(db)) { | |
9558 | ||
9559 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
11fdf7f2 TL |
9560 | auto& misref_extents = repairer.get_misreferences(); |
9561 | interval_set<uint64_t> to_release; | |
f67539c2 | 9562 | it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 | 9563 | if (it) { |
eafe8130 TL |
9564 | // fill global if not overriden below |
9565 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 TL |
9566 | |
9567 | CollectionRef c; | |
9568 | spg_t pgid; | |
9569 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
9570 | bool bypass_rest = false; | |
9571 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
9572 | it->next()) { | |
9573 | dout(30) << __func__ << " key " | |
9574 | << pretty_binary_string(it->key()) << dendl; | |
9575 | if (is_extent_shard_key(it->key())) { | |
9576 | continue; | |
9577 | } | |
9578 | ||
9579 | ghobject_t oid; | |
9580 | int r = get_key_object(it->key(), &oid); | |
b3b6e05e | 9581 | if (r < 0 || !repairer.is_used(oid)) { |
11fdf7f2 TL |
9582 | continue; |
9583 | } | |
9584 | ||
9585 | if (!c || | |
9586 | oid.shard_id != pgid.shard || | |
9587 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
9588 | !c->contains(oid)) { | |
9589 | c = nullptr; | |
9590 | for (auto& p : coll_map) { | |
9591 | if (p.second->contains(oid)) { | |
9592 | c = p.second; | |
9593 | break; | |
9594 | } | |
9595 | } | |
9596 | if (!c) { | |
9597 | continue; | |
9598 | } | |
eafe8130 TL |
9599 | if (per_pool_stat_collection || repair) { |
9600 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
11fdf7f2 TL |
9601 | expected_statfs = &expected_pool_statfs[pool_id]; |
9602 | } | |
9603 | } | |
b3b6e05e | 9604 | if (!repairer.is_used(c->cid)) { |
11fdf7f2 TL |
9605 | continue; |
9606 | } | |
9607 | ||
9608 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
9609 | << " obj:" << oid << dendl; | |
9610 | ||
eafe8130 | 9611 | OnodeRef o; |
39ae355f | 9612 | o.reset(Onode::create_decode(c, oid, it->key(), it->value())); |
11fdf7f2 TL |
9613 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
9614 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
9615 | ||
9616 | for (auto& e : o->extent_map.extent_map) { | |
9617 | blobs.insert(e.blob); | |
9618 | } | |
9619 | bool need_onode_update = false; | |
9620 | bool first_dump = true; | |
9621 | for(auto b : blobs) { | |
9622 | bool broken_blob = false; | |
9623 | auto& pextents = b->dirty_blob().dirty_extents(); | |
9624 | for (auto& e : pextents) { | |
9625 | if (!e.is_valid()) { | |
9626 | continue; | |
9627 | } | |
9628 | // for the sake of simplicity and proper shared blob handling | |
9629 | // always rewrite the whole blob even when it's partially | |
9630 | // misreferenced. | |
9631 | if (misref_extents.intersects(e.offset, e.length)) { | |
9632 | if (first_dump) { | |
9633 | first_dump = false; | |
81eedcae | 9634 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
9635 | } |
9636 | broken_blob = true; | |
9637 | break; | |
9638 | } | |
9639 | } | |
9640 | if (!broken_blob) | |
9641 | continue; | |
9642 | bool compressed = b->get_blob().is_compressed(); | |
9643 | need_onode_update = true; | |
9644 | dout(10) << __func__ | |
9645 | << " fix misreferences in oid:" << oid | |
9646 | << " " << *b << dendl; | |
9647 | uint64_t b_off = 0; | |
9648 | PExtentVector pext_to_release; | |
9649 | pext_to_release.reserve(pextents.size()); | |
9650 | // rewriting all valid pextents | |
9651 | for (auto e = pextents.begin(); e != pextents.end(); | |
a4b75251 TL |
9652 | e++) { |
9653 | auto b_off_cur = b_off; | |
9654 | b_off += e->length; | |
11fdf7f2 TL |
9655 | if (!e->is_valid()) { |
9656 | continue; | |
9657 | } | |
9658 | PExtentVector exts; | |
20effc67 | 9659 | dout(5) << __func__ << "::NCB::(F)alloc=" << alloc << ", length=" << e->length << dendl; |
f67539c2 | 9660 | int64_t alloc_len = |
20effc67 | 9661 | alloc->allocate(e->length, min_alloc_size, |
f67539c2 | 9662 | 0, 0, &exts); |
eafe8130 | 9663 | if (alloc_len < 0 || alloc_len < (int64_t)e->length) { |
11fdf7f2 TL |
9664 | derr << __func__ |
9665 | << " failed to allocate 0x" << std::hex << e->length | |
eafe8130 | 9666 | << " allocated 0x " << (alloc_len < 0 ? 0 : alloc_len) |
11fdf7f2 | 9667 | << " min_alloc_size 0x" << min_alloc_size |
20effc67 | 9668 | << " available 0x " << alloc->get_free() |
11fdf7f2 TL |
9669 | << std::dec << dendl; |
9670 | if (alloc_len > 0) { | |
20effc67 | 9671 | alloc->release(exts); |
11fdf7f2 TL |
9672 | } |
9673 | bypass_rest = true; | |
9674 | break; | |
9675 | } | |
9676 | expected_statfs->allocated += e->length; | |
9677 | if (compressed) { | |
9678 | expected_statfs->data_compressed_allocated += e->length; | |
9679 | } | |
9680 | ||
9681 | bufferlist bl; | |
20effc67 | 9682 | IOContext ioc(cct, NULL, !cct->_conf->bluestore_fail_eio); |
11fdf7f2 TL |
9683 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); |
9684 | if (r < 0) { | |
9685 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
9686 | <<"~" << e->length << std::dec << dendl; | |
9687 | ceph_abort_msg("read failed, wtf"); | |
9688 | } | |
9689 | pext_to_release.push_back(*e); | |
9690 | e = pextents.erase(e); | |
9691 | e = pextents.insert(e, exts.begin(), exts.end()); | |
9692 | b->get_blob().map_bl( | |
20effc67 | 9693 | b_off_cur, bl, |
11fdf7f2 TL |
9694 | [&](uint64_t offset, bufferlist& t) { |
9695 | int r = bdev->write(offset, t, false); | |
9696 | ceph_assert(r == 0); | |
9697 | }); | |
9698 | e += exts.size() - 1; | |
9699 | for (auto& p : exts) { | |
9700 | fm->allocate(p.offset, p.length, txn); | |
9701 | } | |
9702 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
9703 | ||
9704 | if (b->get_blob().is_shared()) { | |
9705 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
9706 | ||
20effc67 TL |
9707 | auto sbid = b->shared_blob->get_sbid(); |
9708 | auto sb_it = sb_info.find(sbid); | |
11fdf7f2 | 9709 | ceph_assert(sb_it != sb_info.end()); |
20effc67 TL |
9710 | sb_info_t& sbi = *sb_it; |
9711 | ||
9712 | if (sbi.allocated_chunks < 0) { | |
9713 | // NB: it's crucial to use compressed_allocated_chunks from sb_info_t | |
9714 | // as we originally used that value while accumulating | |
9715 | // expected_statfs | |
9716 | expected_statfs->allocated -= uint64_t(-sbi.allocated_chunks) << min_alloc_size_order; | |
9717 | expected_statfs->data_compressed_allocated -= | |
9718 | uint64_t(-sbi.allocated_chunks) << min_alloc_size_order; | |
9719 | } else { | |
9720 | expected_statfs->allocated -= uint64_t(sbi.allocated_chunks) << min_alloc_size_order; | |
11fdf7f2 | 9721 | } |
20effc67 TL |
9722 | sbi.allocated_chunks = 0; |
9723 | repairer.fix_shared_blob(txn, sbid, nullptr, 0); | |
9724 | ||
11fdf7f2 TL |
9725 | // relying on blob's pextents to decide what to release. |
9726 | for (auto& p : pext_to_release) { | |
9727 | to_release.union_insert(p.offset, p.length); | |
9728 | } | |
9729 | } else { | |
9730 | for (auto& p : pext_to_release) { | |
9731 | expected_statfs->allocated -= p.length; | |
9732 | if (compressed) { | |
9733 | expected_statfs->data_compressed_allocated -= p.length; | |
9734 | } | |
9735 | to_release.union_insert(p.offset, p.length); | |
9736 | } | |
9737 | } | |
9738 | if (bypass_rest) { | |
9739 | break; | |
9740 | } | |
9741 | } // for(auto b : blobs) | |
9742 | if (need_onode_update) { | |
9743 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
9744 | _record_onode(o, txn); | |
9745 | } | |
9746 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
9747 | ||
9748 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
9749 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
9750 | << "~" << it.get_len() << std::dec << dendl; | |
9751 | fm->release(it.get_start(), it.get_len(), txn); | |
9752 | } | |
20effc67 | 9753 | alloc->release(to_release); |
11fdf7f2 TL |
9754 | to_release.clear(); |
9755 | } // if (it) { | |
9756 | } //if (repair && repairer.preprocess_misreference()) { | |
11fdf7f2 | 9757 | sb_info.clear(); |
20effc67 | 9758 | sb_ref_counts.reset(); |
11fdf7f2 | 9759 | |
eafe8130 | 9760 | dout(1) << __func__ << " checking pool_statfs" << dendl; |
39ae355f TL |
9761 | _fsck_check_statfs(expected_store_statfs, expected_pool_statfs, |
9762 | errors, warnings, repair ? &repairer : nullptr); | |
eafe8130 | 9763 | if (depth != FSCK_SHALLOW) { |
9f95a23c | 9764 | dout(1) << __func__ << " checking for stray omap data " << dendl; |
f67539c2 | 9765 | it = db->get_iterator(PREFIX_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 | 9766 | if (it) { |
9f95a23c | 9767 | uint64_t last_omap_head = 0; |
eafe8130 TL |
9768 | for (it->lower_bound(string()); it->valid(); it->next()) { |
9769 | uint64_t omap_head; | |
f67539c2 | 9770 | |
eafe8130 | 9771 | _key_decode_u64(it->key().c_str(), &omap_head); |
f67539c2 | 9772 | |
9f95a23c | 9773 | if (used_omap_head.count(omap_head) == 0 && |
f67539c2 | 9774 | omap_head != last_omap_head) { |
20effc67 | 9775 | pair<string,string> rk = it->raw_key(); |
9f95a23c TL |
9776 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
9777 | << "fsck error: found stray omap data on omap_head " | |
20effc67 TL |
9778 | << omap_head << " " << last_omap_head |
9779 | << " prefix/key: " << url_escape(rk.first) | |
9780 | << " " << url_escape(rk.second) | |
9781 | << fsck_dendl; | |
f67539c2 TL |
9782 | ++errors; |
9783 | last_omap_head = omap_head; | |
eafe8130 | 9784 | } |
7c673cae FG |
9785 | } |
9786 | } | |
f67539c2 | 9787 | it = db->get_iterator(PREFIX_PGMETA_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 | 9788 | if (it) { |
9f95a23c | 9789 | uint64_t last_omap_head = 0; |
eafe8130 TL |
9790 | for (it->lower_bound(string()); it->valid(); it->next()) { |
9791 | uint64_t omap_head; | |
9792 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
9793 | if (used_omap_head.count(omap_head) == 0 && |
9794 | omap_head != last_omap_head) { | |
20effc67 | 9795 | pair<string,string> rk = it->raw_key(); |
9f95a23c TL |
9796 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
9797 | << "fsck error: found stray (pgmeta) omap data on omap_head " | |
20effc67 TL |
9798 | << omap_head << " " << last_omap_head |
9799 | << " prefix/key: " << url_escape(rk.first) | |
9800 | << " " << url_escape(rk.second) | |
9801 | << fsck_dendl; | |
9f95a23c | 9802 | last_omap_head = omap_head; |
eafe8130 TL |
9803 | ++errors; |
9804 | } | |
11fdf7f2 TL |
9805 | } |
9806 | } | |
f67539c2 | 9807 | it = db->get_iterator(PREFIX_PERPOOL_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
9f95a23c TL |
9808 | if (it) { |
9809 | uint64_t last_omap_head = 0; | |
9810 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
9811 | uint64_t pool; | |
9812 | uint64_t omap_head; | |
9813 | string k = it->key(); | |
9814 | const char *c = k.c_str(); | |
9815 | c = _key_decode_u64(c, &pool); | |
9816 | c = _key_decode_u64(c, &omap_head); | |
9817 | if (used_omap_head.count(omap_head) == 0 && | |
f67539c2 | 9818 | omap_head != last_omap_head) { |
20effc67 | 9819 | pair<string,string> rk = it->raw_key(); |
9f95a23c TL |
9820 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
9821 | << "fsck error: found stray (per-pool) omap data on omap_head " | |
20effc67 TL |
9822 | << omap_head << " " << last_omap_head |
9823 | << " prefix/key: " << url_escape(rk.first) | |
9824 | << " " << url_escape(rk.second) | |
9825 | << fsck_dendl; | |
9f95a23c | 9826 | ++errors; |
f67539c2 TL |
9827 | last_omap_head = omap_head; |
9828 | } | |
9829 | } | |
9830 | } | |
9831 | it = db->get_iterator(PREFIX_PERPG_OMAP, KeyValueDB::ITERATOR_NOCACHE); | |
9832 | if (it) { | |
9833 | uint64_t last_omap_head = 0; | |
9834 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
9835 | uint64_t pool; | |
9836 | uint32_t hash; | |
9837 | uint64_t omap_head; | |
9838 | string k = it->key(); | |
9839 | const char* c = k.c_str(); | |
9840 | c = _key_decode_u64(c, &pool); | |
9841 | c = _key_decode_u32(c, &hash); | |
9842 | c = _key_decode_u64(c, &omap_head); | |
9843 | if (used_omap_head.count(omap_head) == 0 && | |
9844 | omap_head != last_omap_head) { | |
9845 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
9846 | << "fsck error: found stray (per-pg) omap data on omap_head " | |
20effc67 | 9847 | << " key " << pretty_binary_string(it->key()) |
f67539c2 TL |
9848 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; |
9849 | ++errors; | |
9850 | last_omap_head = omap_head; | |
9f95a23c TL |
9851 | } |
9852 | } | |
9853 | } | |
eafe8130 | 9854 | dout(1) << __func__ << " checking deferred events" << dendl; |
f67539c2 | 9855 | it = db->get_iterator(PREFIX_DEFERRED, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
9856 | if (it) { |
9857 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
9858 | bufferlist bl = it->value(); | |
9859 | auto p = bl.cbegin(); | |
9860 | bluestore_deferred_transaction_t wt; | |
9861 | try { | |
9862 | decode(wt, p); | |
f67539c2 | 9863 | } catch (ceph::buffer::error& e) { |
eafe8130 TL |
9864 | derr << "fsck error: failed to decode deferred txn " |
9865 | << pretty_binary_string(it->key()) << dendl; | |
9866 | if (repair) { | |
9867 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
9868 | << pretty_binary_string(it->key()) | |
9869 | << "', removing" << dendl; | |
9870 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
9871 | } | |
9872 | continue; | |
9873 | } | |
9874 | dout(20) << __func__ << " deferred " << wt.seq | |
9875 | << " ops " << wt.ops.size() | |
9876 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
9877 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
9f95a23c | 9878 | apply_for_bitset_range( |
f67539c2 | 9879 | e.get_start(), e.get_len(), alloc_size, used_blocks, |
eafe8130 | 9880 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
eafe8130 TL |
9881 | bs.set(pos); |
9882 | } | |
9883 | ); | |
9884 | } | |
7c673cae | 9885 | } |
eafe8130 TL |
9886 | } |
9887 | ||
20effc67 TL |
9888 | // skip freelist vs allocated compare when we have Null fm |
9889 | if (!fm->is_null_manager()) { | |
9890 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
9891 | #ifdef HAVE_LIBZBD | |
9892 | if (freelist_type == "zoned") { | |
9893 | // verify per-zone state | |
9894 | // - verify no allocations beyond write pointer | |
9895 | // - verify num_dead_bytes count (neither allocated nor | |
9896 | // free space past the write pointer) | |
9897 | auto a = dynamic_cast<ZonedAllocator*>(alloc); | |
9898 | auto num_zones = bdev->get_size() / zone_size; | |
9899 | ||
9900 | // mark the free space past the write pointer | |
9901 | for (uint32_t zone = first_sequential_zone; zone < num_zones; ++zone) { | |
9902 | auto wp = a->get_write_pointer(zone); | |
9903 | uint64_t offset = zone_size * zone + wp; | |
9904 | uint64_t length = zone_size - wp; | |
9905 | if (!length) { | |
9906 | continue; | |
9907 | } | |
9908 | bool intersects = false; | |
9909 | dout(10) << " marking zone 0x" << std::hex << zone | |
9910 | << " region after wp 0x" << offset << "~" << length | |
9911 | << std::dec << dendl; | |
9912 | apply_for_bitset_range( | |
9913 | offset, length, alloc_size, used_blocks, | |
9914 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
9915 | if (bs.test(pos)) { | |
9916 | derr << "fsck error: zone 0x" << std::hex << zone | |
9917 | << " has used space at 0x" << pos * alloc_size | |
9918 | << " beyond write pointer 0x" << wp | |
9919 | << std::dec << dendl; | |
9920 | intersects = true; | |
eafe8130 | 9921 | } else { |
20effc67 | 9922 | bs.set(pos); |
11fdf7f2 | 9923 | } |
20effc67 TL |
9924 | } |
9925 | ); | |
9926 | if (intersects) { | |
9927 | ++errors; | |
9928 | } | |
9929 | } | |
9930 | ||
9931 | used_blocks.flip(); | |
9932 | ||
9933 | // skip conventional zones | |
9934 | uint64_t pos = (first_sequential_zone * zone_size) / min_alloc_size - 1; | |
9935 | pos = used_blocks.find_next(pos); | |
9936 | ||
9937 | uint64_t zone_dead = 0; | |
9938 | for (uint32_t zone = first_sequential_zone; | |
9939 | zone < num_zones; | |
9940 | ++zone, zone_dead = 0) { | |
9941 | while (pos != decltype(used_blocks)::npos && | |
9942 | (pos * min_alloc_size) / zone_size == zone) { | |
9943 | dout(40) << " zone 0x" << std::hex << zone | |
9944 | << " dead 0x" << (pos * min_alloc_size) << "~" << min_alloc_size | |
9945 | << std::dec << dendl; | |
9946 | zone_dead += min_alloc_size; | |
9947 | pos = used_blocks.find_next(pos); | |
9948 | } | |
9949 | dout(20) << " zone 0x" << std::hex << zone << " dead is 0x" << zone_dead | |
9950 | << std::dec << dendl; | |
9951 | // cross-check dead bytes against zone state | |
9952 | if (a->get_dead_bytes(zone) != zone_dead) { | |
9953 | derr << "fsck error: zone 0x" << std::hex << zone << " has 0x" << zone_dead | |
9954 | << " dead bytes but freelist says 0x" << a->get_dead_bytes(zone) | |
9955 | << dendl; | |
9956 | ++errors; | |
9957 | // TODO: repair | |
9958 | } | |
9959 | } | |
9960 | used_blocks.flip(); | |
9961 | } else | |
9962 | #endif | |
9963 | { | |
9964 | fm->enumerate_reset(); | |
9965 | uint64_t offset, length; | |
9966 | while (fm->enumerate_next(db, &offset, &length)) { | |
9967 | bool intersects = false; | |
9968 | apply_for_bitset_range( | |
9969 | offset, length, alloc_size, used_blocks, | |
9970 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
9971 | ceph_assert(pos < bs.size()); | |
9972 | if (bs.test(pos) && !bluefs_used_blocks.test(pos)) { | |
9973 | if (offset == SUPER_RESERVED && | |
9974 | length == min_alloc_size - SUPER_RESERVED) { | |
9975 | // this is due to the change just after luminous to min_alloc_size | |
9976 | // granularity allocations, and our baked in assumption at the top | |
9977 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
9978 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
9979 | // since we will never allocate this region below min_alloc_size. | |
9980 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
9981 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
9982 | << length << std::dec << dendl; | |
9983 | } else { | |
9984 | intersects = true; | |
9985 | if (repair) { | |
9986 | repairer.fix_false_free(db, fm, | |
9987 | pos * min_alloc_size, | |
9988 | min_alloc_size); | |
9989 | } | |
9990 | } | |
9991 | } else { | |
9992 | bs.set(pos); | |
eafe8130 | 9993 | } |
11fdf7f2 | 9994 | } |
20effc67 TL |
9995 | ); |
9996 | if (intersects) { | |
9997 | derr << "fsck error: free extent 0x" << std::hex << offset | |
9998 | << "~" << length << std::dec | |
9999 | << " intersects allocated blocks" << dendl; | |
10000 | ++errors; | |
b5b8bbf5 | 10001 | } |
20effc67 TL |
10002 | } |
10003 | fm->enumerate_reset(); | |
10004 | ||
10005 | // check for leaked extents | |
10006 | size_t count = used_blocks.count(); | |
10007 | if (used_blocks.size() != count) { | |
10008 | ceph_assert(used_blocks.size() > count); | |
10009 | used_blocks.flip(); | |
10010 | size_t start = used_blocks.find_first(); | |
10011 | while (start != decltype(used_blocks)::npos) { | |
10012 | size_t cur = start; | |
10013 | while (true) { | |
10014 | size_t next = used_blocks.find_next(cur); | |
10015 | if (next != cur + 1) { | |
10016 | ++errors; | |
10017 | derr << "fsck error: leaked extent 0x" << std::hex | |
10018 | << ((uint64_t)start * fm->get_alloc_size()) << "~" | |
10019 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
10020 | << dendl; | |
10021 | if (repair) { | |
10022 | repairer.fix_leaked(db, | |
10023 | fm, | |
10024 | start * min_alloc_size, | |
10025 | (cur + 1 - start) * min_alloc_size); | |
10026 | } | |
10027 | start = next; | |
10028 | break; | |
10029 | } | |
10030 | cur = next; | |
10031 | } | |
10032 | } | |
10033 | used_blocks.flip(); | |
10034 | } | |
b5b8bbf5 | 10035 | } |
7c673cae FG |
10036 | } |
10037 | } | |
11fdf7f2 | 10038 | if (repair) { |
f67539c2 TL |
10039 | if (per_pool_omap != OMAP_PER_PG) { |
10040 | dout(5) << __func__ << " fixing per_pg_omap" << dendl; | |
10041 | repairer.fix_per_pool_omap(db, OMAP_PER_PG); | |
9f95a23c TL |
10042 | } |
10043 | ||
11fdf7f2 TL |
10044 | dout(5) << __func__ << " applying repair results" << dendl; |
10045 | repaired = repairer.apply(db); | |
10046 | dout(5) << __func__ << " repair applied" << dendl; | |
10047 | } | |
7c673cae | 10048 | |
eafe8130 | 10049 | out_scan: |
7c673cae FG |
10050 | dout(2) << __func__ << " " << num_objects << " objects, " |
10051 | << num_sharded_objects << " of them sharded. " | |
10052 | << dendl; | |
10053 | dout(2) << __func__ << " " << num_extents << " extents to " | |
10054 | << num_blobs << " blobs, " | |
10055 | << num_spanning_blobs << " spanning, " | |
10056 | << num_shared_blobs << " shared." | |
10057 | << dendl; | |
10058 | ||
10059 | utime_t duration = ceph_clock_now() - start; | |
9f95a23c TL |
10060 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " |
10061 | << warnings << " warnings, " | |
10062 | << repaired << " repaired, " | |
10063 | << (errors + warnings - (int)repaired) << " remaining in " | |
7c673cae | 10064 | << duration << " seconds" << dendl; |
9f95a23c TL |
10065 | |
10066 | // In non-repair mode we should return error count only as | |
10067 | // it indicates if store status is OK. | |
10068 | // In repair mode both errors and warnings are taken into account | |
10069 | // since repaired counter relates to them both. | |
10070 | return repair ? errors + warnings - (int)repaired : errors; | |
11fdf7f2 TL |
10071 | } |
10072 | ||
10073 | /// methods to inject various errors fsck can repair | |
10074 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
10075 | const bufferlist& bl) | |
10076 | { | |
10077 | KeyValueDB::Transaction txn; | |
10078 | txn = db->get_transaction(); | |
10079 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
10080 | db->submit_transaction_sync(txn); | |
10081 | }; | |
10082 | ||
a4b75251 TL |
10083 | void BlueStore::inject_no_shared_blob_key() |
10084 | { | |
10085 | KeyValueDB::Transaction txn; | |
10086 | txn = db->get_transaction(); | |
10087 | ceph_assert(blobid_last > 0); | |
10088 | // kill the last used sbid, this can be broken due to blobid preallocation | |
10089 | // in rare cases, leaving as-is for the sake of simplicity | |
10090 | uint64_t sbid = blobid_last; | |
10091 | ||
10092 | string key; | |
10093 | dout(5) << __func__<< " " << sbid << dendl; | |
10094 | get_shared_blob_key(sbid, &key); | |
10095 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
10096 | db->submit_transaction_sync(txn); | |
10097 | }; | |
10098 | ||
20effc67 | 10099 | void BlueStore::inject_stray_shared_blob_key(uint64_t sbid) |
11fdf7f2 TL |
10100 | { |
10101 | KeyValueDB::Transaction txn; | |
10102 | txn = db->get_transaction(); | |
10103 | ||
20effc67 TL |
10104 | dout(5) << __func__ << " " << sbid << dendl; |
10105 | ||
10106 | string key; | |
10107 | get_shared_blob_key(sbid, &key); | |
10108 | bluestore_shared_blob_t persistent(sbid); | |
39ae355f | 10109 | persistent.ref_map.get(0xdead0000, min_alloc_size); |
20effc67 TL |
10110 | bufferlist bl; |
10111 | encode(persistent, bl); | |
10112 | dout(20) << __func__ << " sbid " << sbid | |
10113 | << " takes " << bl.length() << " bytes, updating" | |
10114 | << dendl; | |
10115 | ||
10116 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
10117 | db->submit_transaction_sync(txn); | |
10118 | }; | |
10119 | ||
10120 | ||
10121 | void BlueStore::inject_leaked(uint64_t len) | |
10122 | { | |
11fdf7f2 | 10123 | PExtentVector exts; |
20effc67 | 10124 | int64_t alloc_len = alloc->allocate(len, min_alloc_size, |
11fdf7f2 | 10125 | min_alloc_size * 256, 0, &exts); |
20effc67 TL |
10126 | |
10127 | if (fm->is_null_manager()) { | |
10128 | return; | |
10129 | } | |
10130 | ||
10131 | KeyValueDB::Transaction txn; | |
10132 | txn = db->get_transaction(); | |
10133 | ||
11fdf7f2 TL |
10134 | ceph_assert(alloc_len >= (int64_t)len); |
10135 | for (auto& p : exts) { | |
10136 | fm->allocate(p.offset, p.length, txn); | |
10137 | } | |
10138 | db->submit_transaction_sync(txn); | |
10139 | } | |
10140 | ||
10141 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
10142 | { | |
20effc67 TL |
10143 | ceph_assert(!fm->is_null_manager()); |
10144 | ||
11fdf7f2 TL |
10145 | KeyValueDB::Transaction txn; |
10146 | OnodeRef o; | |
10147 | CollectionRef c = _get_collection(cid); | |
10148 | ceph_assert(c); | |
10149 | { | |
9f95a23c | 10150 | std::unique_lock l{c->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
10151 | o = c->get_onode(oid, false); |
10152 | ceph_assert(o); | |
10153 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10154 | } | |
10155 | ||
10156 | bool injected = false; | |
10157 | txn = db->get_transaction(); | |
10158 | auto& em = o->extent_map.extent_map; | |
10159 | std::vector<const PExtentVector*> v; | |
10160 | if (em.size()) { | |
10161 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
10162 | } | |
10163 | if (em.size() > 1) { | |
10164 | auto it = em.end(); | |
10165 | --it; | |
10166 | v.push_back(&(it->blob->get_blob().get_extents())); | |
10167 | } | |
10168 | for (auto pext : v) { | |
10169 | if (pext->size()) { | |
10170 | auto p = pext->begin(); | |
10171 | while (p != pext->end()) { | |
10172 | if (p->is_valid()) { | |
10173 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
10174 | << "~" << p->length << std::dec << dendl; | |
10175 | fm->release(p->offset, p->length, txn); | |
10176 | injected = true; | |
10177 | break; | |
10178 | } | |
10179 | ++p; | |
10180 | } | |
10181 | } | |
10182 | } | |
10183 | ceph_assert(injected); | |
10184 | db->submit_transaction_sync(txn); | |
10185 | } | |
10186 | ||
9f95a23c TL |
10187 | void BlueStore::inject_legacy_omap() |
10188 | { | |
10189 | dout(1) << __func__ << dendl; | |
f67539c2 | 10190 | per_pool_omap = OMAP_BULK; |
9f95a23c TL |
10191 | KeyValueDB::Transaction txn; |
10192 | txn = db->get_transaction(); | |
10193 | txn->rmkey(PREFIX_SUPER, "per_pool_omap"); | |
10194 | db->submit_transaction_sync(txn); | |
10195 | } | |
10196 | ||
10197 | void BlueStore::inject_legacy_omap(coll_t cid, ghobject_t oid) | |
10198 | { | |
10199 | dout(1) << __func__ << " " | |
10200 | << cid << " " << oid | |
10201 | <<dendl; | |
10202 | KeyValueDB::Transaction txn; | |
10203 | OnodeRef o; | |
10204 | CollectionRef c = _get_collection(cid); | |
10205 | ceph_assert(c); | |
10206 | { | |
10207 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
10208 | o = c->get_onode(oid, false); | |
10209 | ceph_assert(o); | |
10210 | } | |
f67539c2 TL |
10211 | o->onode.clear_flag( |
10212 | bluestore_onode_t::FLAG_PERPG_OMAP | | |
10213 | bluestore_onode_t::FLAG_PERPOOL_OMAP | | |
10214 | bluestore_onode_t::FLAG_PGMETA_OMAP); | |
9f95a23c TL |
10215 | txn = db->get_transaction(); |
10216 | _record_onode(o, txn); | |
10217 | db->submit_transaction_sync(txn); | |
10218 | } | |
10219 | ||
20effc67 TL |
10220 | void BlueStore::inject_stray_omap(uint64_t head, const string& name) |
10221 | { | |
10222 | dout(1) << __func__ << dendl; | |
10223 | KeyValueDB::Transaction txn = db->get_transaction(); | |
10224 | ||
10225 | string key; | |
10226 | bufferlist bl; | |
10227 | _key_encode_u64(head, &key); | |
10228 | key.append(name); | |
10229 | txn->set(PREFIX_OMAP, key, bl); | |
10230 | ||
10231 | db->submit_transaction_sync(txn); | |
10232 | } | |
9f95a23c | 10233 | |
11fdf7f2 TL |
10234 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) |
10235 | { | |
10236 | BlueStoreRepairer repairer; | |
10237 | repairer.fix_statfs(db, key, new_statfs); | |
10238 | repairer.apply(db); | |
10239 | } | |
10240 | ||
eafe8130 TL |
10241 | void BlueStore::inject_global_statfs(const store_statfs_t& new_statfs) |
10242 | { | |
10243 | KeyValueDB::Transaction t = db->get_transaction(); | |
10244 | volatile_statfs v; | |
10245 | v = new_statfs; | |
10246 | bufferlist bl; | |
10247 | v.encode(bl); | |
10248 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
10249 | db->submit_transaction_sync(t); | |
10250 | } | |
10251 | ||
11fdf7f2 TL |
10252 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, |
10253 | coll_t cid2, ghobject_t oid2, | |
10254 | uint64_t offset) | |
10255 | { | |
10256 | OnodeRef o1; | |
10257 | CollectionRef c1 = _get_collection(cid1); | |
10258 | ceph_assert(c1); | |
10259 | { | |
9f95a23c | 10260 | std::unique_lock l{c1->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
10261 | o1 = c1->get_onode(oid1, false); |
10262 | ceph_assert(o1); | |
10263 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
10264 | } | |
10265 | OnodeRef o2; | |
10266 | CollectionRef c2 = _get_collection(cid2); | |
10267 | ceph_assert(c2); | |
10268 | { | |
9f95a23c | 10269 | std::unique_lock l{c2->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
10270 | o2 = c2->get_onode(oid2, false); |
10271 | ceph_assert(o2); | |
10272 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
10273 | } | |
10274 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
10275 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
10276 | ||
10277 | // require onode/extent layout to be the same (and simple) | |
10278 | // to make things easier | |
10279 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
10280 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
10281 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
10282 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
10283 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
10284 | ceph_assert(e1.length == e2.length); | |
10285 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
10286 | ||
10287 | KeyValueDB::Transaction txn; | |
10288 | txn = db->get_transaction(); | |
10289 | ||
10290 | // along with misreference error this will create space leaks errors | |
10291 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
10292 | o2->extent_map.dirty_range(offset, e2.length); | |
10293 | o2->extent_map.update(txn, false); | |
10294 | ||
10295 | _record_onode(o2, txn); | |
10296 | db->submit_transaction_sync(txn); | |
7c673cae FG |
10297 | } |
10298 | ||
adb31ebb TL |
10299 | void BlueStore::inject_zombie_spanning_blob(coll_t cid, ghobject_t oid, |
10300 | int16_t blob_id) | |
10301 | { | |
10302 | OnodeRef o; | |
10303 | CollectionRef c = _get_collection(cid); | |
10304 | ceph_assert(c); | |
10305 | { | |
10306 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
10307 | o = c->get_onode(oid, false); | |
10308 | ceph_assert(o); | |
10309 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10310 | } | |
10311 | ||
10312 | BlobRef b = c->new_blob(); | |
10313 | b->id = blob_id; | |
10314 | o->extent_map.spanning_blob_map[blob_id] = b; | |
10315 | ||
10316 | KeyValueDB::Transaction txn; | |
10317 | txn = db->get_transaction(); | |
10318 | ||
10319 | _record_onode(o, txn); | |
10320 | db->submit_transaction_sync(txn); | |
10321 | } | |
10322 | ||
a4b75251 TL |
10323 | void BlueStore::inject_bluefs_file(std::string_view dir, std::string_view name, size_t new_size) |
10324 | { | |
10325 | ceph_assert(bluefs); | |
10326 | ||
10327 | BlueFS::FileWriter* p_handle = nullptr; | |
10328 | auto ret = bluefs->open_for_write(dir, name, &p_handle, false); | |
10329 | ceph_assert(ret == 0); | |
10330 | ||
10331 | std::string s('0', new_size); | |
10332 | bufferlist bl; | |
10333 | bl.append(s); | |
10334 | p_handle->append(bl); | |
10335 | ||
10336 | bluefs->fsync(p_handle); | |
10337 | bluefs->close_writer(p_handle); | |
10338 | } | |
10339 | ||
7c673cae FG |
10340 | void BlueStore::collect_metadata(map<string,string> *pm) |
10341 | { | |
10342 | dout(10) << __func__ << dendl; | |
10343 | bdev->collect_metadata("bluestore_bdev_", pm); | |
10344 | if (bluefs) { | |
10345 | (*pm)["bluefs"] = "1"; | |
9f95a23c TL |
10346 | // this value is for backward compatibility only |
10347 | (*pm)["bluefs_single_shared_device"] = \ | |
10348 | stringify((int)bluefs_layout.single_shared_device()); | |
10349 | (*pm)["bluefs_dedicated_db"] = \ | |
10350 | stringify((int)bluefs_layout.dedicated_db); | |
10351 | (*pm)["bluefs_dedicated_wal"] = \ | |
10352 | stringify((int)bluefs_layout.dedicated_wal); | |
10353 | bluefs->collect_metadata(pm, bluefs_layout.shared_bdev); | |
7c673cae FG |
10354 | } else { |
10355 | (*pm)["bluefs"] = "0"; | |
10356 | } | |
11fdf7f2 TL |
10357 | |
10358 | // report numa mapping for underlying devices | |
10359 | int node = -1; | |
10360 | set<int> nodes; | |
10361 | set<string> failed; | |
10362 | int r = get_numa_node(&node, &nodes, &failed); | |
10363 | if (r >= 0) { | |
10364 | if (!failed.empty()) { | |
10365 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
10366 | } | |
10367 | if (!nodes.empty()) { | |
10368 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
10369 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
10370 | } | |
10371 | if (node >= 0) { | |
10372 | (*pm)["objectstore_numa_node"] = stringify(node); | |
10373 | } | |
10374 | } | |
05a536ef | 10375 | (*pm)["bluestore_min_alloc_size"] = stringify(min_alloc_size); |
11fdf7f2 TL |
10376 | } |
10377 | ||
10378 | int BlueStore::get_numa_node( | |
10379 | int *final_node, | |
10380 | set<int> *out_nodes, | |
10381 | set<string> *out_failed) | |
10382 | { | |
10383 | int node = -1; | |
10384 | set<string> devices; | |
10385 | get_devices(&devices); | |
10386 | set<int> nodes; | |
10387 | set<string> failed; | |
10388 | for (auto& devname : devices) { | |
10389 | int n; | |
10390 | BlkDev bdev(devname); | |
10391 | int r = bdev.get_numa_node(&n); | |
10392 | if (r < 0) { | |
10393 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
10394 | << dendl; | |
10395 | failed.insert(devname); | |
10396 | continue; | |
10397 | } | |
10398 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
10399 | << dendl; | |
10400 | nodes.insert(n); | |
10401 | if (node < 0) { | |
10402 | node = n; | |
10403 | } | |
10404 | } | |
10405 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
10406 | *final_node = node; | |
10407 | } | |
10408 | if (out_nodes) { | |
10409 | *out_nodes = nodes; | |
10410 | } | |
10411 | if (out_failed) { | |
10412 | *out_failed = failed; | |
10413 | } | |
10414 | return 0; | |
10415 | } | |
10416 | ||
1d09f67e TL |
10417 | void BlueStore::prepare_for_fast_shutdown() |
10418 | { | |
10419 | m_fast_shutdown = true; | |
10420 | } | |
10421 | ||
11fdf7f2 TL |
10422 | int BlueStore::get_devices(set<string> *ls) |
10423 | { | |
10424 | if (bdev) { | |
10425 | bdev->get_devices(ls); | |
10426 | if (bluefs) { | |
10427 | bluefs->get_devices(ls); | |
10428 | } | |
10429 | return 0; | |
10430 | } | |
20effc67 | 10431 | |
11fdf7f2 | 10432 | // grumble, we haven't started up yet. |
20effc67 TL |
10433 | if (int r = _open_path(); r < 0) { |
10434 | return r; | |
10435 | } | |
10436 | auto close_path = make_scope_guard([&] { | |
10437 | _close_path(); | |
10438 | }); | |
10439 | if (int r = _open_fsid(false); r < 0) { | |
10440 | return r; | |
10441 | } | |
10442 | auto close_fsid = make_scope_guard([&] { | |
10443 | _close_fsid(); | |
10444 | }); | |
10445 | if (int r = _read_fsid(&fsid); r < 0) { | |
10446 | return r; | |
10447 | } | |
10448 | if (int r = _lock_fsid(); r < 0) { | |
10449 | return r; | |
10450 | } | |
10451 | if (int r = _open_bdev(false); r < 0) { | |
10452 | return r; | |
10453 | } | |
10454 | auto close_bdev = make_scope_guard([&] { | |
10455 | _close_bdev(); | |
10456 | }); | |
10457 | if (int r = _minimal_open_bluefs(false); r < 0) { | |
10458 | return r; | |
10459 | } | |
11fdf7f2 TL |
10460 | bdev->get_devices(ls); |
10461 | if (bluefs) { | |
10462 | bluefs->get_devices(ls); | |
10463 | } | |
11fdf7f2 | 10464 | _minimal_close_bluefs(); |
20effc67 | 10465 | return 0; |
7c673cae FG |
10466 | } |
10467 | ||
11fdf7f2 | 10468 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
10469 | { |
10470 | buf->reset(); | |
11fdf7f2 | 10471 | |
f67539c2 TL |
10472 | auto prefix = per_pool_omap == OMAP_BULK ? |
10473 | PREFIX_OMAP : | |
10474 | per_pool_omap == OMAP_PER_POOL ? | |
10475 | PREFIX_PERPOOL_OMAP : | |
10476 | PREFIX_PERPG_OMAP; | |
9f95a23c | 10477 | buf->omap_allocated = |
f67539c2 | 10478 | db->estimate_prefix_size(prefix, string()); |
11fdf7f2 | 10479 | |
20effc67 | 10480 | uint64_t bfree = alloc->get_free(); |
7c673cae FG |
10481 | |
10482 | if (bluefs) { | |
f67539c2 | 10483 | buf->internally_reserved = 0; |
11fdf7f2 | 10484 | // include dedicated db, too, if that isn't the shared device. |
9f95a23c | 10485 | if (bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 10486 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); |
7c673cae | 10487 | } |
11fdf7f2 TL |
10488 | // call any non-omap bluefs space "internal metadata" |
10489 | buf->internal_metadata = | |
f67539c2 | 10490 | bluefs->get_used() |
11fdf7f2 | 10491 | - buf->omap_allocated; |
7c673cae FG |
10492 | } |
10493 | ||
1e59de90 TL |
10494 | ExtBlkDevState ebd_state; |
10495 | int rc = bdev->get_ebd_state(ebd_state); | |
10496 | if (rc == 0) { | |
10497 | buf->total += ebd_state.get_physical_total(); | |
11fdf7f2 TL |
10498 | |
10499 | // we are limited by both the size of the virtual device and the | |
10500 | // underlying physical device. | |
1e59de90 | 10501 | bfree = std::min(bfree, ebd_state.get_physical_avail()); |
11fdf7f2 | 10502 | |
1e59de90 | 10503 | buf->allocated = ebd_state.get_physical_total() - ebd_state.get_physical_avail();; |
11fdf7f2 TL |
10504 | } else { |
10505 | buf->total += bdev->get_size(); | |
10506 | } | |
10507 | buf->available = bfree; | |
10508 | } | |
10509 | ||
10510 | int BlueStore::statfs(struct store_statfs_t *buf, | |
10511 | osd_alert_list_t* alerts) | |
10512 | { | |
10513 | if (alerts) { | |
10514 | alerts->clear(); | |
10515 | _log_alerts(*alerts); | |
10516 | } | |
10517 | _get_statfs_overall(buf); | |
31f18b77 | 10518 | { |
11fdf7f2 | 10519 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 10520 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
10521 | buf->data_stored = vstatfs.stored(); |
10522 | buf->data_compressed = vstatfs.compressed(); | |
10523 | buf->data_compressed_original = vstatfs.compressed_original(); | |
10524 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
10525 | } | |
10526 | ||
10527 | dout(20) << __func__ << " " << *buf << dendl; | |
10528 | return 0; | |
10529 | } | |
10530 | ||
9f95a23c TL |
10531 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
10532 | bool *out_per_pool_omap) | |
11fdf7f2 TL |
10533 | { |
10534 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 10535 | |
11fdf7f2 TL |
10536 | if (!per_pool_stat_collection) { |
10537 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
10538 | return -ENOTSUP; | |
7c673cae | 10539 | } |
11fdf7f2 | 10540 | buf->reset(); |
7c673cae | 10541 | |
11fdf7f2 TL |
10542 | { |
10543 | std::lock_guard l(vstatfs_lock); | |
10544 | osd_pools[pool_id].publish(buf); | |
10545 | } | |
9f95a23c TL |
10546 | |
10547 | string key_prefix; | |
10548 | _key_encode_u64(pool_id, &key_prefix); | |
f67539c2 | 10549 | *out_per_pool_omap = per_pool_omap != OMAP_BULK; |
1d09f67e TL |
10550 | // stop calls after db was closed |
10551 | if (*out_per_pool_omap && db) { | |
f67539c2 TL |
10552 | auto prefix = per_pool_omap == OMAP_PER_POOL ? |
10553 | PREFIX_PERPOOL_OMAP : | |
10554 | PREFIX_PERPG_OMAP; | |
10555 | buf->omap_allocated = db->estimate_prefix_size(prefix, key_prefix); | |
10556 | } | |
9f95a23c | 10557 | |
11fdf7f2 | 10558 | dout(10) << __func__ << *buf << dendl; |
7c673cae FG |
10559 | return 0; |
10560 | } | |
10561 | ||
81eedcae TL |
10562 | void BlueStore::_check_legacy_statfs_alert() |
10563 | { | |
10564 | string s; | |
10565 | if (!per_pool_stat_collection && | |
eafe8130 | 10566 | cct->_conf->bluestore_warn_on_legacy_statfs) { |
81eedcae TL |
10567 | s = "legacy statfs reporting detected, " |
10568 | "suggest to run store repair to get consistent statistic reports"; | |
10569 | } | |
10570 | std::lock_guard l(qlock); | |
10571 | legacy_statfs_alert = s; | |
10572 | } | |
10573 | ||
f67539c2 | 10574 | void BlueStore::_check_no_per_pg_or_pool_omap_alert() |
9f95a23c | 10575 | { |
f67539c2 TL |
10576 | string per_pg, per_pool; |
10577 | if (per_pool_omap != OMAP_PER_PG) { | |
10578 | if (cct->_conf->bluestore_warn_on_no_per_pg_omap) { | |
10579 | per_pg = "legacy (not per-pg) omap detected, " | |
10580 | "suggest to run store repair to benefit from faster PG removal"; | |
10581 | } | |
10582 | if (per_pool_omap != OMAP_PER_POOL) { | |
10583 | if (cct->_conf->bluestore_warn_on_no_per_pool_omap) { | |
10584 | per_pool = "legacy (not per-pool) omap detected, " | |
10585 | "suggest to run store repair to benefit from per-pool omap usage statistics"; | |
10586 | } | |
10587 | } | |
9f95a23c TL |
10588 | } |
10589 | std::lock_guard l(qlock); | |
f67539c2 TL |
10590 | no_per_pg_omap_alert = per_pg; |
10591 | no_per_pool_omap_alert = per_pool; | |
9f95a23c TL |
10592 | } |
10593 | ||
7c673cae FG |
10594 | // --------------- |
10595 | // cache | |
10596 | ||
10597 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
10598 | { | |
9f95a23c | 10599 | std::shared_lock l(coll_lock); |
7c673cae FG |
10600 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); |
10601 | if (cp == coll_map.end()) | |
10602 | return CollectionRef(); | |
10603 | return cp->second; | |
10604 | } | |
10605 | ||
20effc67 TL |
10606 | BlueStore::CollectionRef BlueStore::_get_collection_by_oid(const ghobject_t& oid) |
10607 | { | |
10608 | std::shared_lock l(coll_lock); | |
10609 | ||
10610 | // FIXME: we must replace this with something more efficient | |
10611 | ||
10612 | for (auto& i : coll_map) { | |
10613 | spg_t spgid; | |
10614 | if (i.first.is_pg(&spgid) && | |
10615 | i.second->contains(oid)) { | |
10616 | return i.second; | |
10617 | } | |
10618 | } | |
10619 | return CollectionRef(); | |
10620 | } | |
10621 | ||
7c673cae FG |
10622 | void BlueStore::_queue_reap_collection(CollectionRef& c) |
10623 | { | |
10624 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
10625 | // _reap_collections and this in the same thread, |
10626 | // so no need a lock. | |
7c673cae FG |
10627 | removed_collections.push_back(c); |
10628 | } | |
10629 | ||
10630 | void BlueStore::_reap_collections() | |
10631 | { | |
94b18763 | 10632 | |
7c673cae FG |
10633 | list<CollectionRef> removed_colls; |
10634 | { | |
94b18763 FG |
10635 | // _queue_reap_collection and this in the same thread. |
10636 | // So no need a lock. | |
10637 | if (!removed_collections.empty()) | |
10638 | removed_colls.swap(removed_collections); | |
10639 | else | |
10640 | return; | |
7c673cae FG |
10641 | } |
10642 | ||
94b18763 FG |
10643 | list<CollectionRef>::iterator p = removed_colls.begin(); |
10644 | while (p != removed_colls.end()) { | |
7c673cae FG |
10645 | CollectionRef c = *p; |
10646 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
39ae355f | 10647 | if (c->onode_space.map_any([&](Onode* o) { |
11fdf7f2 | 10648 | ceph_assert(!o->exists); |
7c673cae FG |
10649 | if (o->flushing_count.load()) { |
10650 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
10651 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 10652 | return true; |
7c673cae | 10653 | } |
94b18763 | 10654 | return false; |
7c673cae | 10655 | })) { |
94b18763 | 10656 | ++p; |
7c673cae FG |
10657 | continue; |
10658 | } | |
39ae355f | 10659 | c->onode_space.clear(); |
94b18763 | 10660 | p = removed_colls.erase(p); |
7c673cae FG |
10661 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
10662 | } | |
94b18763 | 10663 | if (removed_colls.empty()) { |
7c673cae | 10664 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
10665 | } else { |
10666 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
10667 | } |
10668 | } | |
10669 | ||
1e59de90 | 10670 | void BlueStore::_update_logger() |
7c673cae FG |
10671 | { |
10672 | uint64_t num_onodes = 0; | |
9f95a23c | 10673 | uint64_t num_pinned_onodes = 0; |
7c673cae FG |
10674 | uint64_t num_extents = 0; |
10675 | uint64_t num_blobs = 0; | |
10676 | uint64_t num_buffers = 0; | |
10677 | uint64_t num_buffer_bytes = 0; | |
9f95a23c TL |
10678 | for (auto c : onode_cache_shards) { |
10679 | c->add_stats(&num_onodes, &num_pinned_onodes); | |
10680 | } | |
10681 | for (auto c : buffer_cache_shards) { | |
10682 | c->add_stats(&num_extents, &num_blobs, | |
10683 | &num_buffers, &num_buffer_bytes); | |
7c673cae FG |
10684 | } |
10685 | logger->set(l_bluestore_onodes, num_onodes); | |
9f95a23c | 10686 | logger->set(l_bluestore_pinned_onodes, num_pinned_onodes); |
7c673cae FG |
10687 | logger->set(l_bluestore_extents, num_extents); |
10688 | logger->set(l_bluestore_blobs, num_blobs); | |
10689 | logger->set(l_bluestore_buffers, num_buffers); | |
10690 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
10691 | } | |
10692 | ||
10693 | // --------------- | |
10694 | // read operations | |
10695 | ||
10696 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
10697 | { | |
10698 | return _get_collection(cid); | |
10699 | } | |
10700 | ||
11fdf7f2 TL |
10701 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
10702 | const coll_t& cid) | |
7c673cae | 10703 | { |
9f95a23c TL |
10704 | std::unique_lock l{coll_lock}; |
10705 | auto c = ceph::make_ref<Collection>( | |
11fdf7f2 | 10706 | this, |
9f95a23c TL |
10707 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
10708 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
11fdf7f2 TL |
10709 | cid); |
10710 | new_coll_map[cid] = c; | |
9f95a23c | 10711 | _osr_attach(c.get()); |
11fdf7f2 TL |
10712 | return c; |
10713 | } | |
10714 | ||
10715 | void BlueStore::set_collection_commit_queue( | |
10716 | const coll_t& cid, | |
10717 | ContextQueue *commit_queue) | |
10718 | { | |
10719 | if (commit_queue) { | |
9f95a23c | 10720 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
10721 | if (coll_map.count(cid)) { |
10722 | coll_map[cid]->commit_queue = commit_queue; | |
10723 | } else if (new_coll_map.count(cid)) { | |
10724 | new_coll_map[cid]->commit_queue = commit_queue; | |
10725 | } | |
10726 | } | |
7c673cae FG |
10727 | } |
10728 | ||
11fdf7f2 | 10729 | |
7c673cae FG |
10730 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
10731 | { | |
10732 | Collection *c = static_cast<Collection *>(c_.get()); | |
10733 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
10734 | if (!c->exists) | |
10735 | return false; | |
10736 | ||
10737 | bool r = true; | |
10738 | ||
10739 | { | |
9f95a23c | 10740 | std::shared_lock l(c->lock); |
7c673cae FG |
10741 | OnodeRef o = c->get_onode(oid, false); |
10742 | if (!o || !o->exists) | |
10743 | r = false; | |
10744 | } | |
10745 | ||
7c673cae FG |
10746 | return r; |
10747 | } | |
10748 | ||
7c673cae FG |
10749 | int BlueStore::stat( |
10750 | CollectionHandle &c_, | |
10751 | const ghobject_t& oid, | |
10752 | struct stat *st, | |
10753 | bool allow_eio) | |
10754 | { | |
10755 | Collection *c = static_cast<Collection *>(c_.get()); | |
10756 | if (!c->exists) | |
10757 | return -ENOENT; | |
10758 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
10759 | ||
10760 | { | |
9f95a23c | 10761 | std::shared_lock l(c->lock); |
7c673cae FG |
10762 | OnodeRef o = c->get_onode(oid, false); |
10763 | if (!o || !o->exists) | |
10764 | return -ENOENT; | |
10765 | st->st_size = o->onode.size; | |
10766 | st->st_blksize = 4096; | |
10767 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
10768 | st->st_nlink = 1; | |
10769 | } | |
10770 | ||
7c673cae FG |
10771 | int r = 0; |
10772 | if (_debug_mdata_eio(oid)) { | |
10773 | r = -EIO; | |
10774 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10775 | } | |
10776 | return r; | |
10777 | } | |
10778 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 10779 | CollectionHandle& ch, |
7c673cae FG |
10780 | const pool_opts_t& opts) |
10781 | { | |
7c673cae | 10782 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 10783 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
10784 | if (!c->exists) |
10785 | return -ENOENT; | |
9f95a23c | 10786 | std::unique_lock l{c->lock}; |
7c673cae FG |
10787 | c->pool_opts = opts; |
10788 | return 0; | |
10789 | } | |
10790 | ||
7c673cae FG |
10791 | int BlueStore::read( |
10792 | CollectionHandle &c_, | |
10793 | const ghobject_t& oid, | |
10794 | uint64_t offset, | |
10795 | size_t length, | |
10796 | bufferlist& bl, | |
224ce89b | 10797 | uint32_t op_flags) |
7c673cae | 10798 | { |
11fdf7f2 | 10799 | auto start = mono_clock::now(); |
7c673cae FG |
10800 | Collection *c = static_cast<Collection *>(c_.get()); |
10801 | const coll_t &cid = c->get_cid(); | |
10802 | dout(15) << __func__ << " " << cid << " " << oid | |
10803 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
10804 | << dendl; | |
10805 | if (!c->exists) | |
10806 | return -ENOENT; | |
10807 | ||
10808 | bl.clear(); | |
10809 | int r; | |
10810 | { | |
9f95a23c | 10811 | std::shared_lock l(c->lock); |
11fdf7f2 | 10812 | auto start1 = mono_clock::now(); |
7c673cae | 10813 | OnodeRef o = c->get_onode(oid, false); |
494da23a TL |
10814 | log_latency("get_onode@read", |
10815 | l_bluestore_read_onode_meta_lat, | |
10816 | mono_clock::now() - start1, | |
10817 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
10818 | if (!o || !o->exists) { |
10819 | r = -ENOENT; | |
10820 | goto out; | |
10821 | } | |
10822 | ||
10823 | if (offset == length && offset == 0) | |
10824 | length = o->onode.size; | |
10825 | ||
10826 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
10827 | if (r == -EIO) { |
10828 | logger->inc(l_bluestore_read_eio); | |
10829 | } | |
7c673cae FG |
10830 | } |
10831 | ||
10832 | out: | |
28e407b8 | 10833 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
10834 | r = -EIO; |
10835 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
10836 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
10837 | cct->_conf->bluestore_debug_random_read_err && | |
10838 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
10839 | 100.0)) == 0) { | |
224ce89b WB |
10840 | dout(0) << __func__ << ": inject random EIO" << dendl; |
10841 | r = -EIO; | |
7c673cae FG |
10842 | } |
10843 | dout(10) << __func__ << " " << cid << " " << oid | |
10844 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
10845 | << " = " << r << dendl; | |
494da23a TL |
10846 | log_latency(__func__, |
10847 | l_bluestore_read_lat, | |
10848 | mono_clock::now() - start, | |
10849 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
10850 | return r; |
10851 | } | |
10852 | ||
9f95a23c | 10853 | void BlueStore::_read_cache( |
39ae355f | 10854 | OnodeRef& o, |
7c673cae FG |
10855 | uint64_t offset, |
10856 | size_t length, | |
9f95a23c TL |
10857 | int read_cache_policy, |
10858 | ready_regions_t& ready_regions, | |
10859 | blobs2read_t& blobs2read) | |
7c673cae | 10860 | { |
7c673cae | 10861 | // build blob-wise list to of stuff read (that isn't cached) |
7c673cae FG |
10862 | unsigned left = length; |
10863 | uint64_t pos = offset; | |
7c673cae FG |
10864 | auto lp = o->extent_map.seek_lextent(offset); |
10865 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
10866 | if (pos < lp->logical_offset) { | |
10867 | unsigned hole = lp->logical_offset - pos; | |
10868 | if (hole >= left) { | |
9f95a23c | 10869 | break; |
7c673cae FG |
10870 | } |
10871 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
9f95a23c | 10872 | << std::dec << dendl; |
7c673cae FG |
10873 | pos += hole; |
10874 | left -= hole; | |
10875 | } | |
94b18763 | 10876 | BlobRef& bptr = lp->blob; |
7c673cae FG |
10877 | unsigned l_off = pos - lp->logical_offset; |
10878 | unsigned b_off = l_off + lp->blob_offset; | |
10879 | unsigned b_len = std::min(left, lp->length - l_off); | |
10880 | ||
10881 | ready_regions_t cache_res; | |
10882 | interval_set<uint32_t> cache_interval; | |
10883 | bptr->shared_blob->bc.read( | |
91327a77 AA |
10884 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
10885 | read_cache_policy); | |
7c673cae | 10886 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c TL |
10887 | << " need 0x" << b_off << "~" << b_len |
10888 | << " cache has 0x" << cache_interval | |
10889 | << std::dec << dendl; | |
7c673cae FG |
10890 | |
10891 | auto pc = cache_res.begin(); | |
11fdf7f2 | 10892 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
10893 | while (b_len > 0) { |
10894 | unsigned l; | |
10895 | if (pc != cache_res.end() && | |
9f95a23c TL |
10896 | pc->first == b_off) { |
10897 | l = pc->second.length(); | |
f67539c2 | 10898 | ready_regions[pos] = std::move(pc->second); |
9f95a23c TL |
10899 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" |
10900 | << b_off << "~" << l << std::dec << dendl; | |
10901 | ++pc; | |
7c673cae | 10902 | } else { |
9f95a23c TL |
10903 | l = b_len; |
10904 | if (pc != cache_res.end()) { | |
10905 | ceph_assert(pc->first > b_off); | |
10906 | l = pc->first - b_off; | |
10907 | } | |
10908 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
10909 | << b_off << "~" << l << std::dec << dendl; | |
10910 | // merge regions | |
10911 | { | |
10912 | uint64_t r_off = b_off; | |
10913 | uint64_t r_len = l; | |
10914 | uint64_t front = r_off % chunk_size; | |
10915 | if (front) { | |
10916 | r_off -= front; | |
10917 | r_len += front; | |
10918 | } | |
10919 | unsigned tail = r_len % chunk_size; | |
10920 | if (tail) { | |
10921 | r_len += chunk_size - tail; | |
10922 | } | |
10923 | bool merged = false; | |
10924 | regions2read_t& r2r = blobs2read[bptr]; | |
10925 | if (r2r.size()) { | |
10926 | read_req_t& pre = r2r.back(); | |
10927 | if (r_off <= (pre.r_off + pre.r_len)) { | |
10928 | front += (r_off - pre.r_off); | |
10929 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
10930 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
10931 | merged = true; | |
10932 | } | |
10933 | } | |
10934 | if (!merged) { | |
10935 | read_req_t req(r_off, r_len); | |
10936 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
10937 | r2r.emplace_back(std::move(req)); | |
10938 | } | |
10939 | } | |
7c673cae FG |
10940 | } |
10941 | pos += l; | |
10942 | b_off += l; | |
10943 | left -= l; | |
10944 | b_len -= l; | |
10945 | } | |
10946 | ++lp; | |
10947 | } | |
9f95a23c | 10948 | } |
7c673cae | 10949 | |
9f95a23c TL |
10950 | int BlueStore::_prepare_read_ioc( |
10951 | blobs2read_t& blobs2read, | |
10952 | vector<bufferlist>* compressed_blob_bls, | |
10953 | IOContext* ioc) | |
10954 | { | |
7c673cae | 10955 | for (auto& p : blobs2read) { |
94b18763 | 10956 | const BlobRef& bptr = p.first; |
11fdf7f2 | 10957 | regions2read_t& r2r = p.second; |
20effc67 TL |
10958 | dout(20) << __func__ << " blob " << *bptr << " need " |
10959 | << r2r << dendl; | |
7c673cae FG |
10960 | if (bptr->get_blob().is_compressed()) { |
10961 | // read the whole thing | |
9f95a23c TL |
10962 | if (compressed_blob_bls->empty()) { |
10963 | // ensure we avoid any reallocation on subsequent blobs | |
10964 | compressed_blob_bls->reserve(blobs2read.size()); | |
10965 | } | |
10966 | compressed_blob_bls->push_back(bufferlist()); | |
10967 | bufferlist& bl = compressed_blob_bls->back(); | |
10968 | auto r = bptr->get_blob().map( | |
10969 | 0, bptr->get_blob().get_ondisk_length(), | |
10970 | [&](uint64_t offset, uint64_t length) { | |
10971 | int r = bdev->aio_read(offset, length, &bl, ioc); | |
10972 | if (r < 0) | |
7c673cae FG |
10973 | return r; |
10974 | return 0; | |
9f95a23c | 10975 | }); |
b32b8144 FG |
10976 | if (r < 0) { |
10977 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
10978 | if (r == -EIO) { | |
10979 | // propagate EIO to caller | |
10980 | return r; | |
10981 | } | |
11fdf7f2 | 10982 | ceph_assert(r == 0); |
b32b8144 | 10983 | } |
7c673cae FG |
10984 | } else { |
10985 | // read the pieces | |
11fdf7f2 | 10986 | for (auto& req : r2r) { |
9f95a23c TL |
10987 | dout(20) << __func__ << " region 0x" << std::hex |
10988 | << req.regs.front().logical_offset | |
10989 | << ": 0x" << req.regs.front().blob_xoffset | |
10990 | << " reading 0x" << req.r_off | |
10991 | << "~" << req.r_len << std::dec | |
10992 | << dendl; | |
7c673cae | 10993 | |
9f95a23c TL |
10994 | // read it |
10995 | auto r = bptr->get_blob().map( | |
10996 | req.r_off, req.r_len, | |
10997 | [&](uint64_t offset, uint64_t length) { | |
10998 | int r = bdev->aio_read(offset, length, &req.bl, ioc); | |
10999 | if (r < 0) | |
7c673cae FG |
11000 | return r; |
11001 | return 0; | |
9f95a23c | 11002 | }); |
b32b8144 FG |
11003 | if (r < 0) { |
11004 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
11005 | << dendl; | |
11006 | if (r == -EIO) { | |
11007 | // propagate EIO to caller | |
11008 | return r; | |
11009 | } | |
11fdf7f2 | 11010 | ceph_assert(r == 0); |
b32b8144 | 11011 | } |
9f95a23c | 11012 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
11013 | } |
11014 | } | |
11015 | } | |
9f95a23c TL |
11016 | return 0; |
11017 | } | |
11fdf7f2 | 11018 | |
9f95a23c | 11019 | int BlueStore::_generate_read_result_bl( |
39ae355f | 11020 | OnodeRef& o, |
9f95a23c TL |
11021 | uint64_t offset, |
11022 | size_t length, | |
11023 | ready_regions_t& ready_regions, | |
11024 | vector<bufferlist>& compressed_blob_bls, | |
11025 | blobs2read_t& blobs2read, | |
11026 | bool buffered, | |
11027 | bool* csum_error, | |
11028 | bufferlist& bl) | |
11029 | { | |
11030 | // enumerate and decompress desired blobs | |
7c673cae FG |
11031 | auto p = compressed_blob_bls.begin(); |
11032 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
11033 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 11034 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 11035 | regions2read_t& r2r = b2r_it->second; |
20effc67 TL |
11036 | dout(20) << __func__ << " blob " << *bptr << " need " |
11037 | << r2r << dendl; | |
7c673cae | 11038 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 11039 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
11040 | bufferlist& compressed_bl = *p++; |
11041 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
9f95a23c TL |
11042 | r2r.front().regs.front().logical_offset) < 0) { |
11043 | *csum_error = true; | |
11044 | return -EIO; | |
7c673cae FG |
11045 | } |
11046 | bufferlist raw_bl; | |
9f95a23c | 11047 | auto r = _decompress(compressed_bl, &raw_bl); |
7c673cae | 11048 | if (r < 0) |
9f95a23c | 11049 | return r; |
7c673cae | 11050 | if (buffered) { |
9f95a23c TL |
11051 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, |
11052 | raw_bl); | |
7c673cae | 11053 | } |
11fdf7f2 TL |
11054 | for (auto& req : r2r) { |
11055 | for (auto& r : req.regs) { | |
11056 | ready_regions[r.logical_offset].substr_of( | |
11057 | raw_bl, r.blob_xoffset, r.length); | |
11058 | } | |
7c673cae FG |
11059 | } |
11060 | } else { | |
11fdf7f2 | 11061 | for (auto& req : r2r) { |
9f95a23c TL |
11062 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, |
11063 | req.regs.front().logical_offset) < 0) { | |
11064 | *csum_error = true; | |
11065 | return -EIO; | |
11066 | } | |
11067 | if (buffered) { | |
11068 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
11069 | req.r_off, req.bl); | |
11070 | } | |
7c673cae | 11071 | |
9f95a23c TL |
11072 | // prune and keep result |
11073 | for (const auto& r : req.regs) { | |
11074 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
11fdf7f2 | 11075 | } |
7c673cae FG |
11076 | } |
11077 | } | |
11078 | ++b2r_it; | |
11079 | } | |
11080 | ||
11081 | // generate a resulting buffer | |
11082 | auto pr = ready_regions.begin(); | |
11083 | auto pr_end = ready_regions.end(); | |
9f95a23c | 11084 | uint64_t pos = 0; |
7c673cae FG |
11085 | while (pos < length) { |
11086 | if (pr != pr_end && pr->first == pos + offset) { | |
11087 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
11088 | << ": data from 0x" << pr->first << "~" << pr->second.length() |
11089 | << std::dec << dendl; | |
7c673cae FG |
11090 | pos += pr->second.length(); |
11091 | bl.claim_append(pr->second); | |
11092 | ++pr; | |
11093 | } else { | |
11094 | uint64_t l = length - pos; | |
11095 | if (pr != pr_end) { | |
11fdf7f2 | 11096 | ceph_assert(pr->first > pos + offset); |
9f95a23c | 11097 | l = pr->first - (pos + offset); |
7c673cae FG |
11098 | } |
11099 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
11100 | << ": zeros for 0x" << (pos + offset) << "~" << l |
11101 | << std::dec << dendl; | |
7c673cae FG |
11102 | bl.append_zero(l); |
11103 | pos += l; | |
11104 | } | |
11105 | } | |
11fdf7f2 TL |
11106 | ceph_assert(bl.length() == length); |
11107 | ceph_assert(pos == length); | |
11108 | ceph_assert(pr == pr_end); | |
9f95a23c TL |
11109 | return 0; |
11110 | } | |
11111 | ||
11112 | int BlueStore::_do_read( | |
11113 | Collection *c, | |
39ae355f | 11114 | OnodeRef& o, |
9f95a23c TL |
11115 | uint64_t offset, |
11116 | size_t length, | |
11117 | bufferlist& bl, | |
11118 | uint32_t op_flags, | |
11119 | uint64_t retry_count) | |
11120 | { | |
11121 | FUNCTRACE(cct); | |
11122 | int r = 0; | |
11123 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
11124 | ||
11125 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
11126 | << " size 0x" << o->onode.size << " (" << std::dec | |
11127 | << o->onode.size << ")" << dendl; | |
11128 | bl.clear(); | |
11129 | ||
11130 | if (offset >= o->onode.size) { | |
11131 | return r; | |
11132 | } | |
11133 | ||
11134 | // generally, don't buffer anything, unless the client explicitly requests | |
11135 | // it. | |
11136 | bool buffered = false; | |
11137 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
11138 | dout(20) << __func__ << " will do buffered read" << dendl; | |
11139 | buffered = true; | |
11140 | } else if (cct->_conf->bluestore_default_buffered_read && | |
11141 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
11142 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
11143 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
11144 | buffered = true; | |
11145 | } | |
11146 | ||
11147 | if (offset + length > o->onode.size) { | |
11148 | length = o->onode.size - offset; | |
11149 | } | |
11150 | ||
11151 | auto start = mono_clock::now(); | |
11152 | o->extent_map.fault_range(db, offset, length); | |
11153 | log_latency(__func__, | |
11154 | l_bluestore_read_onode_meta_lat, | |
11155 | mono_clock::now() - start, | |
11156 | cct->_conf->bluestore_log_op_age); | |
11157 | _dump_onode<30>(cct, *o); | |
11158 | ||
11159 | // for deep-scrub, we only read dirty cache and bypass clean cache in | |
11160 | // order to read underlying block device in case there are silent disk errors. | |
11161 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
11162 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
11163 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
11164 | } | |
11165 | ||
11166 | // build blob-wise list to of stuff read (that isn't cached) | |
11167 | ready_regions_t ready_regions; | |
11168 | blobs2read_t blobs2read; | |
11169 | _read_cache(o, offset, length, read_cache_policy, ready_regions, blobs2read); | |
11170 | ||
11171 | ||
11172 | // read raw blob data. | |
11173 | start = mono_clock::now(); // for the sake of simplicity | |
11174 | // measure the whole block below. | |
11175 | // The error isn't that much... | |
11176 | vector<bufferlist> compressed_blob_bls; | |
20effc67 | 11177 | IOContext ioc(cct, NULL, !cct->_conf->bluestore_fail_eio); |
9f95a23c TL |
11178 | r = _prepare_read_ioc(blobs2read, &compressed_blob_bls, &ioc); |
11179 | // we always issue aio for reading, so errors other than EIO are not allowed | |
11180 | if (r < 0) | |
11181 | return r; | |
11182 | ||
f67539c2 | 11183 | int64_t num_ios = blobs2read.size(); |
9f95a23c | 11184 | if (ioc.has_pending_aios()) { |
f67539c2 | 11185 | num_ios = ioc.get_num_ios(); |
9f95a23c TL |
11186 | bdev->aio_submit(&ioc); |
11187 | dout(20) << __func__ << " waiting for aio" << dendl; | |
11188 | ioc.aio_wait(); | |
11189 | r = ioc.get_return_value(); | |
11190 | if (r < 0) { | |
11191 | ceph_assert(r == -EIO); // no other errors allowed | |
11192 | return -EIO; | |
11193 | } | |
11194 | } | |
11195 | log_latency_fn(__func__, | |
11196 | l_bluestore_read_wait_aio_lat, | |
11197 | mono_clock::now() - start, | |
11198 | cct->_conf->bluestore_log_op_age, | |
11199 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
11200 | ); | |
11201 | ||
11202 | bool csum_error = false; | |
11203 | r = _generate_read_result_bl(o, offset, length, ready_regions, | |
11204 | compressed_blob_bls, blobs2read, | |
20effc67 TL |
11205 | buffered && !ioc.skip_cache(), |
11206 | &csum_error, bl); | |
9f95a23c TL |
11207 | if (csum_error) { |
11208 | // Handles spurious read errors caused by a kernel bug. | |
11209 | // We sometimes get all-zero pages as a result of the read under | |
11210 | // high memory pressure. Retrying the failing read succeeds in most | |
11211 | // cases. | |
11212 | // See also: http://tracker.ceph.com/issues/22464 | |
11213 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
11214 | return -EIO; | |
11215 | } | |
11216 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
11217 | } | |
7c673cae | 11218 | r = bl.length(); |
f64942e4 AA |
11219 | if (retry_count) { |
11220 | logger->inc(l_bluestore_reads_with_retries); | |
11221 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
11222 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
f67539c2 TL |
11223 | stringstream s; |
11224 | s << " reads with retries: " << logger->get(l_bluestore_reads_with_retries); | |
11225 | _set_spurious_read_errors_alert(s.str()); | |
f64942e4 | 11226 | } |
7c673cae FG |
11227 | return r; |
11228 | } | |
11229 | ||
11230 | int BlueStore::_verify_csum(OnodeRef& o, | |
11231 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
11232 | const bufferlist& bl, | |
11233 | uint64_t logical_offset) const | |
11234 | { | |
11235 | int bad; | |
11236 | uint64_t bad_csum; | |
11fdf7f2 | 11237 | auto start = mono_clock::now(); |
7c673cae | 11238 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
11239 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
11240 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
11241 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
11242 | bad = blob_xoffset; | |
11243 | r = -1; | |
11244 | bad_csum = 0xDEADBEEF; | |
11245 | } | |
7c673cae FG |
11246 | if (r < 0) { |
11247 | if (r == -1) { | |
11248 | PExtentVector pex; | |
11249 | blob->map( | |
11250 | bad, | |
11251 | blob->get_csum_chunk_size(), | |
11252 | [&](uint64_t offset, uint64_t length) { | |
11253 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
11254 | return 0; | |
11255 | }); | |
11256 | derr << __func__ << " bad " | |
11257 | << Checksummer::get_csum_type_string(blob->csum_type) | |
11258 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
11259 | << " checksum at blob offset 0x" << bad | |
11260 | << ", got 0x" << bad_csum << ", expected 0x" | |
11261 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
11262 | << ", device location " << pex | |
11263 | << ", logical extent 0x" << std::hex | |
11264 | << (logical_offset + bad - blob_xoffset) << "~" | |
11265 | << blob->get_csum_chunk_size() << std::dec | |
11266 | << ", object " << o->oid | |
11267 | << dendl; | |
11268 | } else { | |
11269 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
11270 | } | |
11271 | } | |
494da23a TL |
11272 | log_latency(__func__, |
11273 | l_bluestore_csum_lat, | |
11274 | mono_clock::now() - start, | |
11275 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 TL |
11276 | if (cct->_conf->bluestore_ignore_data_csum) { |
11277 | return 0; | |
11278 | } | |
7c673cae FG |
11279 | return r; |
11280 | } | |
11281 | ||
11282 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
11283 | { | |
11284 | int r = 0; | |
11fdf7f2 TL |
11285 | auto start = mono_clock::now(); |
11286 | auto i = source.cbegin(); | |
7c673cae | 11287 | bluestore_compression_header_t chdr; |
11fdf7f2 | 11288 | decode(chdr, i); |
7c673cae FG |
11289 | int alg = int(chdr.type); |
11290 | CompressorRef cp = compressor; | |
11291 | if (!cp || (int)cp->get_type() != alg) { | |
11292 | cp = Compressor::create(cct, alg); | |
11293 | } | |
11294 | ||
11295 | if (!cp.get()) { | |
11296 | // if compressor isn't available - error, because cannot return | |
11297 | // decompressed data? | |
11fdf7f2 TL |
11298 | |
11299 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
11300 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
11301 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
11302 | r = -EIO; |
11303 | } else { | |
f67539c2 | 11304 | r = cp->decompress(i, chdr.length, *result, chdr.compressor_message); |
7c673cae FG |
11305 | if (r < 0) { |
11306 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
11307 | r = -EIO; | |
11308 | } | |
11309 | } | |
494da23a TL |
11310 | log_latency(__func__, |
11311 | l_bluestore_decompress_lat, | |
11312 | mono_clock::now() - start, | |
11313 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
11314 | return r; |
11315 | } | |
11316 | ||
11317 | // this stores fiemap into interval_set, other variations | |
11318 | // use it internally | |
11319 | int BlueStore::_fiemap( | |
11320 | CollectionHandle &c_, | |
11321 | const ghobject_t& oid, | |
11322 | uint64_t offset, | |
11323 | size_t length, | |
11324 | interval_set<uint64_t>& destset) | |
11325 | { | |
11326 | Collection *c = static_cast<Collection *>(c_.get()); | |
11327 | if (!c->exists) | |
11328 | return -ENOENT; | |
11329 | { | |
9f95a23c | 11330 | std::shared_lock l(c->lock); |
7c673cae FG |
11331 | |
11332 | OnodeRef o = c->get_onode(oid, false); | |
11333 | if (!o || !o->exists) { | |
11334 | return -ENOENT; | |
11335 | } | |
81eedcae | 11336 | _dump_onode<30>(cct, *o); |
7c673cae FG |
11337 | |
11338 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
11339 | << " size 0x" << o->onode.size << std::dec << dendl; | |
11340 | ||
11341 | boost::intrusive::set<Extent>::iterator ep, eend; | |
11342 | if (offset >= o->onode.size) | |
11343 | goto out; | |
11344 | ||
11345 | if (offset + length > o->onode.size) { | |
11346 | length = o->onode.size - offset; | |
11347 | } | |
11348 | ||
11349 | o->extent_map.fault_range(db, offset, length); | |
11350 | eend = o->extent_map.extent_map.end(); | |
11351 | ep = o->extent_map.seek_lextent(offset); | |
11352 | while (length > 0) { | |
11353 | dout(20) << __func__ << " offset " << offset << dendl; | |
11354 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
11355 | ++ep; | |
11356 | continue; | |
11357 | } | |
11358 | ||
11359 | uint64_t x_len = length; | |
11360 | if (ep != eend && ep->logical_offset <= offset) { | |
11361 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 11362 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
11363 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
11364 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
11365 | destset.insert(offset, x_len); | |
11366 | length -= x_len; | |
11367 | offset += x_len; | |
11368 | if (x_off + x_len == ep->length) | |
11369 | ++ep; | |
11370 | continue; | |
11371 | } | |
11372 | if (ep != eend && | |
11373 | ep->logical_offset > offset && | |
11374 | ep->logical_offset - offset < x_len) { | |
11375 | x_len = ep->logical_offset - offset; | |
11376 | } | |
11377 | offset += x_len; | |
11378 | length -= x_len; | |
11379 | } | |
11380 | } | |
9f95a23c TL |
11381 | |
11382 | out: | |
11383 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
11384 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
11385 | return 0; | |
11386 | } | |
11387 | ||
11388 | int BlueStore::fiemap( | |
11389 | CollectionHandle &c_, | |
11390 | const ghobject_t& oid, | |
11391 | uint64_t offset, | |
11392 | size_t length, | |
11393 | bufferlist& bl) | |
11394 | { | |
11395 | interval_set<uint64_t> m; | |
11396 | int r = _fiemap(c_, oid, offset, length, m); | |
11397 | if (r >= 0) { | |
11398 | encode(m, bl); | |
11399 | } | |
11400 | return r; | |
11401 | } | |
11402 | ||
11403 | int BlueStore::fiemap( | |
11404 | CollectionHandle &c_, | |
11405 | const ghobject_t& oid, | |
11406 | uint64_t offset, | |
11407 | size_t length, | |
11408 | map<uint64_t, uint64_t>& destmap) | |
11409 | { | |
11410 | interval_set<uint64_t> m; | |
11411 | int r = _fiemap(c_, oid, offset, length, m); | |
11412 | if (r >= 0) { | |
11413 | destmap = std::move(m).detach(); | |
11414 | } | |
11415 | return r; | |
11416 | } | |
11417 | ||
11418 | int BlueStore::readv( | |
11419 | CollectionHandle &c_, | |
11420 | const ghobject_t& oid, | |
11421 | interval_set<uint64_t>& m, | |
11422 | bufferlist& bl, | |
11423 | uint32_t op_flags) | |
11424 | { | |
11425 | auto start = mono_clock::now(); | |
11426 | Collection *c = static_cast<Collection *>(c_.get()); | |
11427 | const coll_t &cid = c->get_cid(); | |
11428 | dout(15) << __func__ << " " << cid << " " << oid | |
11429 | << " fiemap " << m | |
11430 | << dendl; | |
11431 | if (!c->exists) | |
11432 | return -ENOENT; | |
11433 | ||
11434 | bl.clear(); | |
11435 | int r; | |
11436 | { | |
11437 | std::shared_lock l(c->lock); | |
11438 | auto start1 = mono_clock::now(); | |
11439 | OnodeRef o = c->get_onode(oid, false); | |
11440 | log_latency("get_onode@read", | |
11441 | l_bluestore_read_onode_meta_lat, | |
11442 | mono_clock::now() - start1, | |
11443 | cct->_conf->bluestore_log_op_age); | |
11444 | if (!o || !o->exists) { | |
11445 | r = -ENOENT; | |
11446 | goto out; | |
11447 | } | |
11448 | ||
11449 | if (m.empty()) { | |
11450 | r = 0; | |
11451 | goto out; | |
11452 | } | |
11453 | ||
11454 | r = _do_readv(c, o, m, bl, op_flags); | |
11455 | if (r == -EIO) { | |
11456 | logger->inc(l_bluestore_read_eio); | |
11457 | } | |
11458 | } | |
11459 | ||
11460 | out: | |
11461 | if (r >= 0 && _debug_data_eio(oid)) { | |
11462 | r = -EIO; | |
11463 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11464 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ | |
11465 | cct->_conf->bluestore_debug_random_read_err && | |
11466 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
11467 | 100.0)) == 0) { | |
11468 | dout(0) << __func__ << ": inject random EIO" << dendl; | |
11469 | r = -EIO; | |
11470 | } | |
11471 | dout(10) << __func__ << " " << cid << " " << oid | |
11472 | << " fiemap " << m << std::dec | |
11473 | << " = " << r << dendl; | |
11474 | log_latency(__func__, | |
11475 | l_bluestore_read_lat, | |
11476 | mono_clock::now() - start, | |
11477 | cct->_conf->bluestore_log_op_age); | |
11478 | return r; | |
11479 | } | |
11480 | ||
11481 | int BlueStore::_do_readv( | |
11482 | Collection *c, | |
39ae355f | 11483 | OnodeRef& o, |
9f95a23c TL |
11484 | const interval_set<uint64_t>& m, |
11485 | bufferlist& bl, | |
11486 | uint32_t op_flags, | |
11487 | uint64_t retry_count) | |
11488 | { | |
11489 | FUNCTRACE(cct); | |
11490 | int r = 0; | |
11491 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
11492 | ||
11493 | dout(20) << __func__ << " fiemap " << m << std::hex | |
11494 | << " size 0x" << o->onode.size << " (" << std::dec | |
11495 | << o->onode.size << ")" << dendl; | |
11496 | ||
11497 | // generally, don't buffer anything, unless the client explicitly requests | |
11498 | // it. | |
11499 | bool buffered = false; | |
11500 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
11501 | dout(20) << __func__ << " will do buffered read" << dendl; | |
11502 | buffered = true; | |
11503 | } else if (cct->_conf->bluestore_default_buffered_read && | |
11504 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
11505 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
11506 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
11507 | buffered = true; | |
11508 | } | |
11509 | // this method must be idempotent since we may call it several times | |
11510 | // before we finally read the expected result. | |
11511 | bl.clear(); | |
11512 | ||
11513 | // call fiemap first! | |
11514 | ceph_assert(m.range_start() <= o->onode.size); | |
11515 | ceph_assert(m.range_end() <= o->onode.size); | |
11516 | auto start = mono_clock::now(); | |
11517 | o->extent_map.fault_range(db, m.range_start(), m.range_end() - m.range_start()); | |
11518 | log_latency(__func__, | |
11519 | l_bluestore_read_onode_meta_lat, | |
11520 | mono_clock::now() - start, | |
11521 | cct->_conf->bluestore_log_op_age); | |
11522 | _dump_onode<30>(cct, *o); | |
11523 | ||
20effc67 | 11524 | IOContext ioc(cct, NULL, !cct->_conf->bluestore_fail_eio); |
9f95a23c TL |
11525 | vector<std::tuple<ready_regions_t, vector<bufferlist>, blobs2read_t>> raw_results; |
11526 | raw_results.reserve(m.num_intervals()); | |
11527 | int i = 0; | |
11528 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
11529 | raw_results.push_back({}); | |
11530 | _read_cache(o, p.get_start(), p.get_len(), read_cache_policy, | |
11531 | std::get<0>(raw_results[i]), std::get<2>(raw_results[i])); | |
11532 | r = _prepare_read_ioc(std::get<2>(raw_results[i]), &std::get<1>(raw_results[i]), &ioc); | |
11533 | // we always issue aio for reading, so errors other than EIO are not allowed | |
11534 | if (r < 0) | |
11535 | return r; | |
11536 | } | |
11537 | ||
11538 | auto num_ios = m.size(); | |
11539 | if (ioc.has_pending_aios()) { | |
11540 | num_ios = ioc.get_num_ios(); | |
11541 | bdev->aio_submit(&ioc); | |
11542 | dout(20) << __func__ << " waiting for aio" << dendl; | |
11543 | ioc.aio_wait(); | |
11544 | r = ioc.get_return_value(); | |
11545 | if (r < 0) { | |
11546 | ceph_assert(r == -EIO); // no other errors allowed | |
11547 | return -EIO; | |
11548 | } | |
11549 | } | |
11550 | log_latency_fn(__func__, | |
11551 | l_bluestore_read_wait_aio_lat, | |
11552 | mono_clock::now() - start, | |
11553 | cct->_conf->bluestore_log_op_age, | |
11554 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
11555 | ); | |
11556 | ||
11557 | ceph_assert(raw_results.size() == (size_t)m.num_intervals()); | |
11558 | i = 0; | |
11559 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
11560 | bool csum_error = false; | |
11561 | bufferlist t; | |
11562 | r = _generate_read_result_bl(o, p.get_start(), p.get_len(), | |
11563 | std::get<0>(raw_results[i]), | |
11564 | std::get<1>(raw_results[i]), | |
11565 | std::get<2>(raw_results[i]), | |
11566 | buffered, &csum_error, t); | |
11567 | if (csum_error) { | |
11568 | // Handles spurious read errors caused by a kernel bug. | |
11569 | // We sometimes get all-zero pages as a result of the read under | |
11570 | // high memory pressure. Retrying the failing read succeeds in most | |
11571 | // cases. | |
11572 | // See also: http://tracker.ceph.com/issues/22464 | |
11573 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
11574 | return -EIO; | |
11575 | } | |
11576 | return _do_readv(c, o, m, bl, op_flags, retry_count + 1); | |
11577 | } | |
11578 | bl.claim_append(t); | |
11579 | } | |
11580 | if (retry_count) { | |
11581 | logger->inc(l_bluestore_reads_with_retries); | |
11582 | dout(5) << __func__ << " read fiemap " << m | |
11583 | << " failed " << retry_count << " times before succeeding" | |
11584 | << dendl; | |
11585 | } | |
11586 | return bl.length(); | |
7c673cae FG |
11587 | } |
11588 | ||
9f95a23c | 11589 | int BlueStore::dump_onode(CollectionHandle &c_, |
7c673cae | 11590 | const ghobject_t& oid, |
9f95a23c TL |
11591 | const string& section_name, |
11592 | Formatter *f) | |
7c673cae | 11593 | { |
9f95a23c TL |
11594 | Collection *c = static_cast<Collection *>(c_.get()); |
11595 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
11596 | if (!c->exists) | |
11597 | return -ENOENT; | |
7c673cae | 11598 | |
9f95a23c TL |
11599 | int r; |
11600 | { | |
11601 | std::shared_lock l(c->lock); | |
11602 | ||
11603 | OnodeRef o = c->get_onode(oid, false); | |
11604 | if (!o || !o->exists) { | |
11605 | r = -ENOENT; | |
11606 | goto out; | |
11607 | } | |
11608 | // FIXME minor: actually the next line isn't enough to | |
11609 | // load shared blobs. Leaving as is for now.. | |
11610 | // | |
11611 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
11612 | ||
11613 | _dump_onode<0>(cct, *o); | |
11614 | f->open_object_section(section_name.c_str()); | |
11615 | o->dump(f); | |
11616 | f->close_section(); | |
11617 | r = 0; | |
7c673cae | 11618 | } |
9f95a23c TL |
11619 | out: |
11620 | dout(10) << __func__ << " " << c->cid << " " << oid | |
11621 | << " = " << r << dendl; | |
7c673cae FG |
11622 | return r; |
11623 | } | |
11624 | ||
7c673cae FG |
11625 | int BlueStore::getattr( |
11626 | CollectionHandle &c_, | |
11627 | const ghobject_t& oid, | |
11628 | const char *name, | |
11629 | bufferptr& value) | |
11630 | { | |
11631 | Collection *c = static_cast<Collection *>(c_.get()); | |
11632 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
11633 | if (!c->exists) | |
11634 | return -ENOENT; | |
11635 | ||
11636 | int r; | |
11637 | { | |
9f95a23c | 11638 | std::shared_lock l(c->lock); |
f91f0fd5 | 11639 | mempool::bluestore_cache_meta::string k(name); |
7c673cae FG |
11640 | |
11641 | OnodeRef o = c->get_onode(oid, false); | |
11642 | if (!o || !o->exists) { | |
11643 | r = -ENOENT; | |
11644 | goto out; | |
11645 | } | |
11646 | ||
11647 | if (!o->onode.attrs.count(k)) { | |
11648 | r = -ENODATA; | |
11649 | goto out; | |
11650 | } | |
11651 | value = o->onode.attrs[k]; | |
11652 | r = 0; | |
11653 | } | |
11654 | out: | |
7c673cae FG |
11655 | if (r == 0 && _debug_mdata_eio(oid)) { |
11656 | r = -EIO; | |
11657 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11658 | } | |
11659 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
11660 | << " = " << r << dendl; | |
11661 | return r; | |
11662 | } | |
11663 | ||
7c673cae FG |
11664 | int BlueStore::getattrs( |
11665 | CollectionHandle &c_, | |
11666 | const ghobject_t& oid, | |
20effc67 | 11667 | map<string,bufferptr,less<>>& aset) |
7c673cae FG |
11668 | { |
11669 | Collection *c = static_cast<Collection *>(c_.get()); | |
11670 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
11671 | if (!c->exists) | |
11672 | return -ENOENT; | |
11673 | ||
11674 | int r; | |
11675 | { | |
9f95a23c | 11676 | std::shared_lock l(c->lock); |
7c673cae FG |
11677 | |
11678 | OnodeRef o = c->get_onode(oid, false); | |
11679 | if (!o || !o->exists) { | |
11680 | r = -ENOENT; | |
11681 | goto out; | |
11682 | } | |
11683 | for (auto& i : o->onode.attrs) { | |
11684 | aset.emplace(i.first.c_str(), i.second); | |
11685 | } | |
11686 | r = 0; | |
11687 | } | |
11688 | ||
11689 | out: | |
7c673cae FG |
11690 | if (r == 0 && _debug_mdata_eio(oid)) { |
11691 | r = -EIO; | |
11692 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11693 | } | |
11694 | dout(10) << __func__ << " " << c->cid << " " << oid | |
11695 | << " = " << r << dendl; | |
11696 | return r; | |
11697 | } | |
11698 | ||
11699 | int BlueStore::list_collections(vector<coll_t>& ls) | |
11700 | { | |
9f95a23c | 11701 | std::shared_lock l(coll_lock); |
11fdf7f2 | 11702 | ls.reserve(coll_map.size()); |
7c673cae FG |
11703 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
11704 | p != coll_map.end(); | |
11705 | ++p) | |
11706 | ls.push_back(p->first); | |
11707 | return 0; | |
11708 | } | |
11709 | ||
11710 | bool BlueStore::collection_exists(const coll_t& c) | |
11711 | { | |
9f95a23c | 11712 | std::shared_lock l(coll_lock); |
7c673cae FG |
11713 | return coll_map.count(c); |
11714 | } | |
11715 | ||
11fdf7f2 | 11716 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 11717 | { |
11fdf7f2 | 11718 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
11719 | vector<ghobject_t> ls; |
11720 | ghobject_t next; | |
11fdf7f2 | 11721 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
11722 | &ls, &next); |
11723 | if (r < 0) { | |
11724 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
11725 | << dendl; | |
11726 | return r; | |
11727 | } | |
11728 | *empty = ls.empty(); | |
11fdf7f2 | 11729 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
11730 | return 0; |
11731 | } | |
11732 | ||
11fdf7f2 | 11733 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 11734 | { |
11fdf7f2 TL |
11735 | dout(15) << __func__ << " " << ch->cid << dendl; |
11736 | Collection *c = static_cast<Collection*>(ch.get()); | |
9f95a23c | 11737 | std::shared_lock l(c->lock); |
11fdf7f2 | 11738 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
11739 | return c->cnode.bits; |
11740 | } | |
11741 | ||
7c673cae FG |
11742 | int BlueStore::collection_list( |
11743 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
11744 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
11745 | { | |
11746 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 11747 | c->flush(); |
7c673cae FG |
11748 | dout(15) << __func__ << " " << c->cid |
11749 | << " start " << start << " end " << end << " max " << max << dendl; | |
11750 | int r; | |
11751 | { | |
9f95a23c | 11752 | std::shared_lock l(c->lock); |
f91f0fd5 TL |
11753 | r = _collection_list(c, start, end, max, false, ls, pnext); |
11754 | } | |
11755 | ||
11756 | dout(10) << __func__ << " " << c->cid | |
11757 | << " start " << start << " end " << end << " max " << max | |
11758 | << " = " << r << ", ls.size() = " << ls->size() | |
11759 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
11760 | return r; | |
11761 | } | |
11762 | ||
11763 | int BlueStore::collection_list_legacy( | |
11764 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
11765 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
11766 | { | |
11767 | Collection *c = static_cast<Collection *>(c_.get()); | |
11768 | c->flush(); | |
11769 | dout(15) << __func__ << " " << c->cid | |
11770 | << " start " << start << " end " << end << " max " << max << dendl; | |
11771 | int r; | |
11772 | { | |
11773 | std::shared_lock l(c->lock); | |
11774 | r = _collection_list(c, start, end, max, true, ls, pnext); | |
7c673cae FG |
11775 | } |
11776 | ||
7c673cae FG |
11777 | dout(10) << __func__ << " " << c->cid |
11778 | << " start " << start << " end " << end << " max " << max | |
11779 | << " = " << r << ", ls.size() = " << ls->size() | |
11780 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
11781 | return r; | |
11782 | } | |
11783 | ||
11784 | int BlueStore::_collection_list( | |
11785 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
f91f0fd5 | 11786 | bool legacy, vector<ghobject_t> *ls, ghobject_t *pnext) |
7c673cae FG |
11787 | { |
11788 | ||
11789 | if (!c->exists) | |
11790 | return -ENOENT; | |
11791 | ||
7c673cae | 11792 | ghobject_t static_next; |
f91f0fd5 TL |
11793 | std::unique_ptr<CollectionListIterator> it; |
11794 | ghobject_t coll_range_temp_start, coll_range_temp_end; | |
11795 | ghobject_t coll_range_start, coll_range_end; | |
f91f0fd5 | 11796 | ghobject_t pend; |
7c673cae FG |
11797 | bool temp; |
11798 | ||
11799 | if (!pnext) | |
11800 | pnext = &static_next; | |
11801 | ||
a4b75251 TL |
11802 | auto log_latency = make_scope_guard( |
11803 | [&, start_time = mono_clock::now(), func_name = __func__] { | |
11804 | log_latency_fn( | |
11805 | func_name, | |
aee94f69 | 11806 | l_bluestore_clist_lat, |
a4b75251 TL |
11807 | mono_clock::now() - start_time, |
11808 | cct->_conf->bluestore_log_collection_list_age, | |
11809 | [&](const ceph::timespan& lat) { | |
11810 | ostringstream ostr; | |
11811 | ostr << ", lat = " << timespan_str(lat) | |
11812 | << " cid =" << c->cid | |
11813 | << " start " << start << " end " << end | |
11814 | << " max " << max; | |
11815 | return ostr.str(); | |
11816 | }); | |
11817 | }); | |
11818 | ||
11fdf7f2 | 11819 | if (start.is_max() || start.hobj.is_max()) { |
a4b75251 TL |
11820 | *pnext = ghobject_t::get_max(); |
11821 | return 0; | |
7c673cae | 11822 | } |
f91f0fd5 | 11823 | get_coll_range(c->cid, c->cnode.bits, &coll_range_temp_start, |
a4b75251 | 11824 | &coll_range_temp_end, &coll_range_start, &coll_range_end, legacy); |
7c673cae | 11825 | dout(20) << __func__ |
f91f0fd5 TL |
11826 | << " range " << coll_range_temp_start |
11827 | << " to " << coll_range_temp_end | |
11828 | << " and " << coll_range_start | |
11829 | << " to " << coll_range_end | |
7c673cae | 11830 | << " start " << start << dendl; |
f91f0fd5 TL |
11831 | if (legacy) { |
11832 | it = std::make_unique<SimpleCollectionListIterator>( | |
11833 | cct, db->get_iterator(PREFIX_OBJ)); | |
11834 | } else { | |
11835 | it = std::make_unique<SortedCollectionListIterator>( | |
11836 | db->get_iterator(PREFIX_OBJ)); | |
11837 | } | |
7c673cae FG |
11838 | if (start == ghobject_t() || |
11839 | start.hobj == hobject_t() || | |
11840 | start == c->cid.get_min_hobj()) { | |
f91f0fd5 | 11841 | it->upper_bound(coll_range_temp_start); |
7c673cae FG |
11842 | temp = true; |
11843 | } else { | |
7c673cae FG |
11844 | if (start.hobj.is_temp()) { |
11845 | temp = true; | |
f91f0fd5 | 11846 | ceph_assert(start >= coll_range_temp_start && start < coll_range_temp_end); |
7c673cae FG |
11847 | } else { |
11848 | temp = false; | |
f91f0fd5 | 11849 | ceph_assert(start >= coll_range_start && start < coll_range_end); |
7c673cae | 11850 | } |
f91f0fd5 TL |
11851 | dout(20) << __func__ << " temp=" << (int)temp << dendl; |
11852 | it->lower_bound(start); | |
7c673cae FG |
11853 | } |
11854 | if (end.hobj.is_max()) { | |
f91f0fd5 | 11855 | pend = temp ? coll_range_temp_end : coll_range_end; |
7c673cae | 11856 | } else { |
7c673cae | 11857 | if (end.hobj.is_temp()) { |
a4b75251 | 11858 | if (temp) { |
f91f0fd5 | 11859 | pend = end; |
a4b75251 TL |
11860 | } else { |
11861 | *pnext = ghobject_t::get_max(); | |
11862 | return 0; | |
11863 | } | |
7c673cae | 11864 | } else { |
f91f0fd5 | 11865 | pend = temp ? coll_range_temp_end : end; |
7c673cae FG |
11866 | } |
11867 | } | |
f91f0fd5 | 11868 | dout(20) << __func__ << " pend " << pend << dendl; |
7c673cae | 11869 | while (true) { |
adb31ebb | 11870 | if (!it->valid() || it->is_ge(pend)) { |
7c673cae FG |
11871 | if (!it->valid()) |
11872 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
11873 | else | |
f91f0fd5 | 11874 | dout(20) << __func__ << " oid " << it->oid() << " >= " << pend << dendl; |
7c673cae FG |
11875 | if (temp) { |
11876 | if (end.hobj.is_temp()) { | |
adb31ebb | 11877 | if (it->valid() && it->is_lt(coll_range_temp_end)) { |
f91f0fd5 | 11878 | *pnext = it->oid(); |
a4b75251 | 11879 | return 0; |
f91f0fd5 | 11880 | } |
7c673cae FG |
11881 | break; |
11882 | } | |
11883 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
11884 | temp = false; | |
f91f0fd5 TL |
11885 | it->upper_bound(coll_range_start); |
11886 | if (end.hobj.is_max()) | |
11887 | pend = coll_range_end; | |
11888 | else | |
11889 | pend = end; | |
11890 | dout(30) << __func__ << " pend " << pend << dendl; | |
7c673cae FG |
11891 | continue; |
11892 | } | |
adb31ebb | 11893 | if (it->valid() && it->is_lt(coll_range_end)) { |
f91f0fd5 | 11894 | *pnext = it->oid(); |
a4b75251 | 11895 | return 0; |
f91f0fd5 | 11896 | } |
7c673cae FG |
11897 | break; |
11898 | } | |
f91f0fd5 | 11899 | dout(20) << __func__ << " oid " << it->oid() << " end " << end << dendl; |
7c673cae FG |
11900 | if (ls->size() >= (unsigned)max) { |
11901 | dout(20) << __func__ << " reached max " << max << dendl; | |
f91f0fd5 | 11902 | *pnext = it->oid(); |
a4b75251 | 11903 | return 0; |
7c673cae | 11904 | } |
f91f0fd5 | 11905 | ls->push_back(it->oid()); |
7c673cae FG |
11906 | it->next(); |
11907 | } | |
a4b75251 TL |
11908 | *pnext = ghobject_t::get_max(); |
11909 | return 0; | |
7c673cae FG |
11910 | } |
11911 | ||
7c673cae FG |
11912 | int BlueStore::omap_get( |
11913 | CollectionHandle &c_, ///< [in] Collection containing oid | |
11914 | const ghobject_t &oid, ///< [in] Object containing omap | |
11915 | bufferlist *header, ///< [out] omap header | |
11916 | map<string, bufferlist> *out /// < [out] Key to value map | |
11917 | ) | |
11918 | { | |
11919 | Collection *c = static_cast<Collection *>(c_.get()); | |
9f95a23c TL |
11920 | return _omap_get(c, oid, header, out); |
11921 | } | |
11922 | ||
11923 | int BlueStore::_omap_get( | |
11924 | Collection *c, ///< [in] Collection containing oid | |
11925 | const ghobject_t &oid, ///< [in] Object containing omap | |
11926 | bufferlist *header, ///< [out] omap header | |
11927 | map<string, bufferlist> *out /// < [out] Key to value map | |
11928 | ) | |
11929 | { | |
7c673cae FG |
11930 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; |
11931 | if (!c->exists) | |
11932 | return -ENOENT; | |
9f95a23c | 11933 | std::shared_lock l(c->lock); |
7c673cae FG |
11934 | int r = 0; |
11935 | OnodeRef o = c->get_onode(oid, false); | |
11936 | if (!o || !o->exists) { | |
11937 | r = -ENOENT; | |
11938 | goto out; | |
11939 | } | |
9f95a23c TL |
11940 | r = _onode_omap_get(o, header, out); |
11941 | out: | |
11942 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
11943 | << dendl; | |
11944 | return r; | |
11945 | } | |
11946 | ||
11947 | int BlueStore::_onode_omap_get( | |
11948 | const OnodeRef &o, ///< [in] Object containing omap | |
11949 | bufferlist *header, ///< [out] omap header | |
11950 | map<string, bufferlist> *out /// < [out] Key to value map | |
11951 | ) | |
11952 | { | |
11953 | int r = 0; | |
11954 | if (!o || !o->exists) { | |
11955 | r = -ENOENT; | |
11956 | goto out; | |
11957 | } | |
7c673cae FG |
11958 | if (!o->onode.has_omap()) |
11959 | goto out; | |
11960 | o->flush(); | |
11961 | { | |
9f95a23c | 11962 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 11963 | string head, tail; |
9f95a23c TL |
11964 | o->get_omap_header(&head); |
11965 | o->get_omap_tail(&tail); | |
33c7a0ef | 11966 | KeyValueDB::Iterator it = db->get_iterator(prefix, 0, KeyValueDB::IteratorBounds{head, tail}); |
7c673cae FG |
11967 | it->lower_bound(head); |
11968 | while (it->valid()) { | |
11969 | if (it->key() == head) { | |
9f95a23c TL |
11970 | dout(30) << __func__ << " got header" << dendl; |
11971 | *header = it->value(); | |
7c673cae | 11972 | } else if (it->key() >= tail) { |
9f95a23c TL |
11973 | dout(30) << __func__ << " reached tail" << dendl; |
11974 | break; | |
7c673cae | 11975 | } else { |
9f95a23c TL |
11976 | string user_key; |
11977 | o->decode_omap_key(it->key(), &user_key); | |
11978 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
11979 | << " -> " << user_key << dendl; | |
11980 | (*out)[user_key] = it->value(); | |
7c673cae FG |
11981 | } |
11982 | it->next(); | |
11983 | } | |
11984 | } | |
9f95a23c | 11985 | out: |
7c673cae FG |
11986 | return r; |
11987 | } | |
11988 | ||
7c673cae FG |
11989 | int BlueStore::omap_get_header( |
11990 | CollectionHandle &c_, ///< [in] Collection containing oid | |
11991 | const ghobject_t &oid, ///< [in] Object containing omap | |
11992 | bufferlist *header, ///< [out] omap header | |
11993 | bool allow_eio ///< [in] don't assert on eio | |
11994 | ) | |
11995 | { | |
11996 | Collection *c = static_cast<Collection *>(c_.get()); | |
11997 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
11998 | if (!c->exists) | |
11999 | return -ENOENT; | |
9f95a23c | 12000 | std::shared_lock l(c->lock); |
7c673cae FG |
12001 | int r = 0; |
12002 | OnodeRef o = c->get_onode(oid, false); | |
12003 | if (!o || !o->exists) { | |
12004 | r = -ENOENT; | |
12005 | goto out; | |
12006 | } | |
12007 | if (!o->onode.has_omap()) | |
12008 | goto out; | |
12009 | o->flush(); | |
12010 | { | |
12011 | string head; | |
9f95a23c TL |
12012 | o->get_omap_header(&head); |
12013 | if (db->get(o->get_omap_prefix(), head, header) >= 0) { | |
7c673cae FG |
12014 | dout(30) << __func__ << " got header" << dendl; |
12015 | } else { | |
12016 | dout(30) << __func__ << " no header" << dendl; | |
12017 | } | |
12018 | } | |
12019 | out: | |
12020 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
12021 | << dendl; | |
12022 | return r; | |
12023 | } | |
12024 | ||
7c673cae FG |
12025 | int BlueStore::omap_get_keys( |
12026 | CollectionHandle &c_, ///< [in] Collection containing oid | |
12027 | const ghobject_t &oid, ///< [in] Object containing omap | |
12028 | set<string> *keys ///< [out] Keys defined on oid | |
12029 | ) | |
12030 | { | |
12031 | Collection *c = static_cast<Collection *>(c_.get()); | |
12032 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
12033 | if (!c->exists) | |
12034 | return -ENOENT; | |
adb31ebb | 12035 | auto start1 = mono_clock::now(); |
9f95a23c | 12036 | std::shared_lock l(c->lock); |
7c673cae FG |
12037 | int r = 0; |
12038 | OnodeRef o = c->get_onode(oid, false); | |
12039 | if (!o || !o->exists) { | |
12040 | r = -ENOENT; | |
12041 | goto out; | |
12042 | } | |
12043 | if (!o->onode.has_omap()) | |
12044 | goto out; | |
12045 | o->flush(); | |
12046 | { | |
9f95a23c | 12047 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 12048 | string head, tail; |
9f95a23c TL |
12049 | o->get_omap_key(string(), &head); |
12050 | o->get_omap_tail(&tail); | |
33c7a0ef | 12051 | KeyValueDB::Iterator it = db->get_iterator(prefix, 0, KeyValueDB::IteratorBounds{head, tail}); |
7c673cae FG |
12052 | it->lower_bound(head); |
12053 | while (it->valid()) { | |
12054 | if (it->key() >= tail) { | |
12055 | dout(30) << __func__ << " reached tail" << dendl; | |
12056 | break; | |
12057 | } | |
12058 | string user_key; | |
9f95a23c | 12059 | o->decode_omap_key(it->key(), &user_key); |
11fdf7f2 | 12060 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
12061 | << " -> " << user_key << dendl; |
12062 | keys->insert(user_key); | |
12063 | it->next(); | |
11fdf7f2 TL |
12064 | } |
12065 | } | |
12066 | out: | |
adb31ebb TL |
12067 | c->store->log_latency( |
12068 | __func__, | |
12069 | l_bluestore_omap_get_keys_lat, | |
12070 | mono_clock::now() - start1, | |
12071 | c->store->cct->_conf->bluestore_log_omap_iterator_age); | |
12072 | ||
11fdf7f2 TL |
12073 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r |
12074 | << dendl; | |
12075 | return r; | |
7c673cae FG |
12076 | } |
12077 | ||
12078 | int BlueStore::omap_get_values( | |
12079 | CollectionHandle &c_, ///< [in] Collection containing oid | |
12080 | const ghobject_t &oid, ///< [in] Object containing omap | |
12081 | const set<string> &keys, ///< [in] Keys to get | |
12082 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
12083 | ) | |
12084 | { | |
12085 | Collection *c = static_cast<Collection *>(c_.get()); | |
12086 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
12087 | if (!c->exists) | |
12088 | return -ENOENT; | |
9f95a23c | 12089 | std::shared_lock l(c->lock); |
adb31ebb | 12090 | auto start1 = mono_clock::now(); |
7c673cae FG |
12091 | int r = 0; |
12092 | string final_key; | |
12093 | OnodeRef o = c->get_onode(oid, false); | |
12094 | if (!o || !o->exists) { | |
12095 | r = -ENOENT; | |
12096 | goto out; | |
12097 | } | |
9f95a23c | 12098 | if (!o->onode.has_omap()) { |
7c673cae | 12099 | goto out; |
9f95a23c TL |
12100 | } |
12101 | o->flush(); | |
11fdf7f2 | 12102 | { |
9f95a23c TL |
12103 | const string& prefix = o->get_omap_prefix(); |
12104 | o->get_omap_key(string(), &final_key); | |
12105 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 12106 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 12107 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
12108 | final_key += *p; |
12109 | bufferlist val; | |
12110 | if (db->get(prefix, final_key, &val) >= 0) { | |
12111 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
12112 | << " -> " << *p << dendl; | |
12113 | out->insert(make_pair(*p, val)); | |
12114 | } | |
7c673cae FG |
12115 | } |
12116 | } | |
12117 | out: | |
adb31ebb TL |
12118 | c->store->log_latency( |
12119 | __func__, | |
12120 | l_bluestore_omap_get_values_lat, | |
12121 | mono_clock::now() - start1, | |
12122 | c->store->cct->_conf->bluestore_log_omap_iterator_age); | |
12123 | ||
7c673cae FG |
12124 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r |
12125 | << dendl; | |
12126 | return r; | |
12127 | } | |
12128 | ||
9f95a23c TL |
12129 | #ifdef WITH_SEASTAR |
12130 | int BlueStore::omap_get_values( | |
12131 | CollectionHandle &c_, ///< [in] Collection containing oid | |
12132 | const ghobject_t &oid, ///< [in] Object containing omap | |
12133 | const std::optional<string> &start_after, ///< [in] Keys to get | |
12134 | map<string, bufferlist> *output ///< [out] Returned keys and values | |
12135 | ) | |
12136 | { | |
12137 | Collection *c = static_cast<Collection *>(c_.get()); | |
12138 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
12139 | if (!c->exists) | |
12140 | return -ENOENT; | |
12141 | std::shared_lock l(c->lock); | |
12142 | int r = 0; | |
12143 | OnodeRef o = c->get_onode(oid, false); | |
12144 | if (!o || !o->exists) { | |
12145 | r = -ENOENT; | |
12146 | goto out; | |
12147 | } | |
12148 | if (!o->onode.has_omap()) { | |
12149 | goto out; | |
12150 | } | |
12151 | o->flush(); | |
12152 | { | |
12153 | ObjectMap::ObjectMapIterator iter = get_omap_iterator(c_, oid); | |
12154 | if (!iter) { | |
12155 | r = -ENOENT; | |
12156 | goto out; | |
12157 | } | |
12158 | iter->upper_bound(*start_after); | |
12159 | for (; iter->valid(); iter->next()) { | |
12160 | output->insert(make_pair(iter->key(), iter->value())); | |
12161 | } | |
12162 | } | |
12163 | ||
12164 | out: | |
12165 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
12166 | << dendl; | |
12167 | return r; | |
12168 | } | |
12169 | #endif | |
12170 | ||
7c673cae FG |
12171 | int BlueStore::omap_check_keys( |
12172 | CollectionHandle &c_, ///< [in] Collection containing oid | |
12173 | const ghobject_t &oid, ///< [in] Object containing omap | |
12174 | const set<string> &keys, ///< [in] Keys to check | |
12175 | set<string> *out ///< [out] Subset of keys defined on oid | |
12176 | ) | |
12177 | { | |
12178 | Collection *c = static_cast<Collection *>(c_.get()); | |
12179 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
12180 | if (!c->exists) | |
12181 | return -ENOENT; | |
9f95a23c | 12182 | std::shared_lock l(c->lock); |
7c673cae FG |
12183 | int r = 0; |
12184 | string final_key; | |
12185 | OnodeRef o = c->get_onode(oid, false); | |
12186 | if (!o || !o->exists) { | |
12187 | r = -ENOENT; | |
12188 | goto out; | |
12189 | } | |
9f95a23c | 12190 | if (!o->onode.has_omap()) { |
7c673cae | 12191 | goto out; |
9f95a23c TL |
12192 | } |
12193 | o->flush(); | |
11fdf7f2 | 12194 | { |
9f95a23c TL |
12195 | const string& prefix = o->get_omap_prefix(); |
12196 | o->get_omap_key(string(), &final_key); | |
12197 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 12198 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 12199 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
12200 | final_key += *p; |
12201 | bufferlist val; | |
12202 | if (db->get(prefix, final_key, &val) >= 0) { | |
12203 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
12204 | << " -> " << *p << dendl; | |
12205 | out->insert(*p); | |
12206 | } else { | |
12207 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
12208 | << " -> " << *p << dendl; | |
12209 | } | |
7c673cae FG |
12210 | } |
12211 | } | |
12212 | out: | |
12213 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
12214 | << dendl; | |
12215 | return r; | |
12216 | } | |
12217 | ||
7c673cae FG |
12218 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
12219 | CollectionHandle &c_, ///< [in] collection | |
12220 | const ghobject_t &oid ///< [in] object | |
12221 | ) | |
12222 | { | |
12223 | Collection *c = static_cast<Collection *>(c_.get()); | |
12224 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
12225 | if (!c->exists) { | |
12226 | return ObjectMap::ObjectMapIterator(); | |
12227 | } | |
9f95a23c | 12228 | std::shared_lock l(c->lock); |
7c673cae FG |
12229 | OnodeRef o = c->get_onode(oid, false); |
12230 | if (!o || !o->exists) { | |
12231 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
12232 | return ObjectMap::ObjectMapIterator(); | |
12233 | } | |
12234 | o->flush(); | |
12235 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
33c7a0ef TL |
12236 | auto bounds = KeyValueDB::IteratorBounds(); |
12237 | if (o->onode.has_omap()) { | |
12238 | std::string lower_bound, upper_bound; | |
12239 | o->get_omap_key(string(), &lower_bound); | |
12240 | o->get_omap_tail(&upper_bound); | |
12241 | bounds.lower_bound = std::move(lower_bound); | |
12242 | bounds.upper_bound = std::move(upper_bound); | |
12243 | } | |
12244 | KeyValueDB::Iterator it = db->get_iterator(o->get_omap_prefix(), 0, std::move(bounds)); | |
1e59de90 | 12245 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(logger,c, o, it)); |
7c673cae FG |
12246 | } |
12247 | ||
12248 | // ----------------- | |
12249 | // write helpers | |
12250 | ||
11fdf7f2 | 12251 | uint64_t BlueStore::_get_ondisk_reserved() const { |
f67539c2 | 12252 | ceph_assert(min_alloc_size); |
11fdf7f2 TL |
12253 | return round_up_to( |
12254 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
12255 | } | |
12256 | ||
7c673cae FG |
12257 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
12258 | { | |
12259 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
12260 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
12261 | << dendl; | |
11fdf7f2 | 12262 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
12263 | { |
12264 | bufferlist bl; | |
11fdf7f2 | 12265 | encode(ondisk_format, bl); |
7c673cae FG |
12266 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
12267 | } | |
12268 | { | |
12269 | bufferlist bl; | |
11fdf7f2 | 12270 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
12271 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
12272 | } | |
12273 | } | |
12274 | ||
12275 | int BlueStore::_open_super_meta() | |
12276 | { | |
12277 | // nid | |
12278 | { | |
12279 | nid_max = 0; | |
12280 | bufferlist bl; | |
12281 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 12282 | auto p = bl.cbegin(); |
7c673cae FG |
12283 | try { |
12284 | uint64_t v; | |
11fdf7f2 | 12285 | decode(v, p); |
7c673cae | 12286 | nid_max = v; |
f67539c2 | 12287 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12288 | derr << __func__ << " unable to read nid_max" << dendl; |
12289 | return -EIO; | |
12290 | } | |
f67539c2 | 12291 | dout(1) << __func__ << " old nid_max " << nid_max << dendl; |
7c673cae FG |
12292 | nid_last = nid_max.load(); |
12293 | } | |
12294 | ||
12295 | // blobid | |
12296 | { | |
12297 | blobid_max = 0; | |
12298 | bufferlist bl; | |
12299 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 12300 | auto p = bl.cbegin(); |
7c673cae FG |
12301 | try { |
12302 | uint64_t v; | |
11fdf7f2 | 12303 | decode(v, p); |
7c673cae | 12304 | blobid_max = v; |
f67539c2 | 12305 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12306 | derr << __func__ << " unable to read blobid_max" << dendl; |
12307 | return -EIO; | |
12308 | } | |
f67539c2 | 12309 | dout(1) << __func__ << " old blobid_max " << blobid_max << dendl; |
7c673cae FG |
12310 | blobid_last = blobid_max.load(); |
12311 | } | |
12312 | ||
12313 | // freelist | |
12314 | { | |
12315 | bufferlist bl; | |
12316 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
12317 | if (bl.length()) { | |
12318 | freelist_type = std::string(bl.c_str(), bl.length()); | |
7c673cae | 12319 | } else { |
11fdf7f2 | 12320 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 12321 | } |
20effc67 | 12322 | dout(5) << __func__ << "::NCB::freelist_type=" << freelist_type << dendl; |
7c673cae | 12323 | } |
7c673cae FG |
12324 | // ondisk format |
12325 | int32_t compat_ondisk_format = 0; | |
12326 | { | |
12327 | bufferlist bl; | |
12328 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
12329 | if (r < 0) { | |
12330 | // base case: kraken bluestore is v1 and readable by v1 | |
12331 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
12332 | << dendl; | |
12333 | ondisk_format = 1; | |
12334 | compat_ondisk_format = 1; | |
12335 | } else { | |
11fdf7f2 | 12336 | auto p = bl.cbegin(); |
7c673cae | 12337 | try { |
11fdf7f2 | 12338 | decode(ondisk_format, p); |
f67539c2 | 12339 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12340 | derr << __func__ << " unable to read ondisk_format" << dendl; |
12341 | return -EIO; | |
12342 | } | |
12343 | bl.clear(); | |
12344 | { | |
12345 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
12346 | ceph_assert(!r); |
12347 | auto p = bl.cbegin(); | |
7c673cae | 12348 | try { |
11fdf7f2 | 12349 | decode(compat_ondisk_format, p); |
f67539c2 | 12350 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12351 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; |
12352 | return -EIO; | |
12353 | } | |
12354 | } | |
12355 | } | |
f67539c2 | 12356 | dout(1) << __func__ << " ondisk_format " << ondisk_format |
7c673cae FG |
12357 | << " compat_ondisk_format " << compat_ondisk_format |
12358 | << dendl; | |
12359 | } | |
12360 | ||
12361 | if (latest_ondisk_format < compat_ondisk_format) { | |
12362 | derr << __func__ << " compat_ondisk_format is " | |
12363 | << compat_ondisk_format << " but we only understand version " | |
12364 | << latest_ondisk_format << dendl; | |
12365 | return -EPERM; | |
12366 | } | |
7c673cae FG |
12367 | |
12368 | { | |
12369 | bufferlist bl; | |
12370 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 12371 | auto p = bl.cbegin(); |
7c673cae FG |
12372 | try { |
12373 | uint64_t val; | |
11fdf7f2 | 12374 | decode(val, p); |
7c673cae | 12375 | min_alloc_size = val; |
1e59de90 | 12376 | min_alloc_size_order = std::countr_zero(val); |
20effc67 TL |
12377 | min_alloc_size_mask = min_alloc_size - 1; |
12378 | ||
11fdf7f2 | 12379 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
f67539c2 | 12380 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12381 | derr << __func__ << " unable to read min_alloc_size" << dendl; |
12382 | return -EIO; | |
12383 | } | |
f67539c2 | 12384 | dout(1) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size |
7c673cae | 12385 | << std::dec << dendl; |
20effc67 TL |
12386 | logger->set(l_bluestore_alloc_unit, min_alloc_size); |
12387 | } | |
12388 | ||
12389 | // smr fields | |
12390 | { | |
12391 | bufferlist bl; | |
12392 | int r = db->get(PREFIX_SUPER, "zone_size", &bl); | |
12393 | if (r >= 0) { | |
12394 | auto p = bl.cbegin(); | |
12395 | decode(zone_size, p); | |
12396 | dout(1) << __func__ << " zone_size 0x" << std::hex << zone_size << std::dec << dendl; | |
12397 | ceph_assert(bdev->is_smr()); | |
12398 | } else { | |
12399 | ceph_assert(!bdev->is_smr()); | |
12400 | } | |
12401 | } | |
12402 | { | |
12403 | bufferlist bl; | |
12404 | int r = db->get(PREFIX_SUPER, "first_sequential_zone", &bl); | |
12405 | if (r >= 0) { | |
12406 | auto p = bl.cbegin(); | |
12407 | decode(first_sequential_zone, p); | |
12408 | dout(1) << __func__ << " first_sequential_zone 0x" << std::hex | |
12409 | << first_sequential_zone << std::dec << dendl; | |
12410 | ceph_assert(bdev->is_smr()); | |
12411 | } else { | |
12412 | ceph_assert(!bdev->is_smr()); | |
12413 | } | |
7c673cae | 12414 | } |
9f95a23c TL |
12415 | |
12416 | _set_per_pool_omap(); | |
12417 | ||
224ce89b | 12418 | _open_statfs(); |
7c673cae FG |
12419 | _set_alloc_sizes(); |
12420 | _set_throttle_params(); | |
12421 | ||
12422 | _set_csum(); | |
12423 | _set_compression(); | |
12424 | _set_blob_size(); | |
12425 | ||
11fdf7f2 | 12426 | _validate_bdev(); |
7c673cae FG |
12427 | return 0; |
12428 | } | |
12429 | ||
12430 | int BlueStore::_upgrade_super() | |
12431 | { | |
12432 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
12433 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
12434 | if (ondisk_format < latest_ondisk_format) { |
12435 | ceph_assert(ondisk_format > 0); | |
12436 | ceph_assert(ondisk_format < latest_ondisk_format); | |
12437 | ||
1911f103 | 12438 | KeyValueDB::Transaction t = db->get_transaction(); |
11fdf7f2 TL |
12439 | if (ondisk_format == 1) { |
12440 | // changes: | |
12441 | // - super: added ondisk_format | |
12442 | // - super: added min_readable_ondisk_format | |
12443 | // - super: added min_compat_ondisk_format | |
12444 | // - super: added min_alloc_size | |
12445 | // - super: removed min_min_alloc_size | |
11fdf7f2 TL |
12446 | { |
12447 | bufferlist bl; | |
12448 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
12449 | auto p = bl.cbegin(); | |
12450 | try { | |
12451 | uint64_t val; | |
12452 | decode(val, p); | |
12453 | min_alloc_size = val; | |
f67539c2 | 12454 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
12455 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; |
12456 | return -EIO; | |
12457 | } | |
12458 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
12459 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 12460 | } |
11fdf7f2 | 12461 | ondisk_format = 2; |
7c673cae | 12462 | } |
9f95a23c TL |
12463 | if (ondisk_format == 2) { |
12464 | // changes: | |
f67539c2 TL |
12465 | // - onode has FLAG_PERPOOL_OMAP. Note that we do not know that *all* |
12466 | // oondes are using the per-pool prefix until a repair is run; at that | |
9f95a23c TL |
12467 | // point the per_pool_omap=1 key will be set. |
12468 | // - super: added per_pool_omap key, which indicates that *all* objects | |
12469 | // are using the new prefix and key format | |
12470 | ondisk_format = 3; | |
1911f103 TL |
12471 | } |
12472 | if (ondisk_format == 3) { | |
12473 | // changes: | |
12474 | // - FreelistManager keeps meta within bdev label | |
12475 | int r = _write_out_fm_meta(0); | |
9f95a23c | 12476 | ceph_assert(r == 0); |
1911f103 | 12477 | ondisk_format = 4; |
9f95a23c | 12478 | } |
1911f103 TL |
12479 | // This to be the last operation |
12480 | _prepare_ondisk_format_super(t); | |
12481 | int r = db->submit_transaction_sync(t); | |
12482 | ceph_assert(r == 0); | |
7c673cae | 12483 | } |
7c673cae FG |
12484 | // done |
12485 | dout(1) << __func__ << " done" << dendl; | |
12486 | return 0; | |
12487 | } | |
12488 | ||
39ae355f | 12489 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef& o) |
7c673cae | 12490 | { |
224ce89b | 12491 | if (o->onode.nid) { |
11fdf7f2 | 12492 | ceph_assert(o->exists); |
7c673cae | 12493 | return; |
224ce89b | 12494 | } |
7c673cae FG |
12495 | uint64_t nid = ++nid_last; |
12496 | dout(20) << __func__ << " " << nid << dendl; | |
12497 | o->onode.nid = nid; | |
12498 | txc->last_nid = nid; | |
224ce89b | 12499 | o->exists = true; |
7c673cae FG |
12500 | } |
12501 | ||
12502 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
12503 | { | |
12504 | uint64_t bid = ++blobid_last; | |
12505 | dout(20) << __func__ << " " << bid << dendl; | |
12506 | txc->last_blobid = bid; | |
12507 | return bid; | |
12508 | } | |
12509 | ||
12510 | void BlueStore::get_db_statistics(Formatter *f) | |
12511 | { | |
12512 | db->get_statistics(f); | |
12513 | } | |
12514 | ||
11fdf7f2 TL |
12515 | BlueStore::TransContext *BlueStore::_txc_create( |
12516 | Collection *c, OpSequencer *osr, | |
f67539c2 TL |
12517 | list<Context*> *on_commits, |
12518 | TrackedOpRef osd_op) | |
7c673cae | 12519 | { |
11fdf7f2 | 12520 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae | 12521 | txc->t = db->get_transaction(); |
f67539c2 TL |
12522 | |
12523 | #ifdef WITH_BLKIN | |
12524 | if (osd_op && osd_op->pg_trace) { | |
12525 | txc->trace.init("TransContext", &trace_endpoint, | |
12526 | &osd_op->pg_trace); | |
12527 | txc->trace.event("txc create"); | |
12528 | txc->trace.keyval("txc seq", txc->seq); | |
12529 | } | |
12530 | #endif | |
12531 | ||
7c673cae FG |
12532 | osr->queue_new(txc); |
12533 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
12534 | << " seq " << txc->seq << dendl; | |
12535 | return txc; | |
12536 | } | |
12537 | ||
12538 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
12539 | { | |
11fdf7f2 TL |
12540 | // one "io" for the kv commit |
12541 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
12542 | auto cost = throttle_cost_per_io.load(); |
12543 | txc->cost = ios * cost + txc->bytes; | |
9f95a23c | 12544 | txc->ios = ios; |
7c673cae FG |
12545 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" |
12546 | << ios << " ios * " << cost << " + " << txc->bytes | |
12547 | << " bytes)" << dendl; | |
12548 | } | |
12549 | ||
12550 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
12551 | { | |
12552 | if (txc->statfs_delta.is_empty()) | |
12553 | return; | |
12554 | ||
12555 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
12556 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
12557 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
12558 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
12559 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
12560 | ||
11fdf7f2 | 12561 | if (per_pool_stat_collection) { |
39ae355f TL |
12562 | if (!is_statfs_recoverable()) { |
12563 | bufferlist bl; | |
12564 | txc->statfs_delta.encode(bl); | |
12565 | string key; | |
12566 | get_pool_stat_key(txc->osd_pool_id, &key); | |
12567 | txc->t->merge(PREFIX_STAT, key, bl); | |
12568 | } | |
11fdf7f2 TL |
12569 | |
12570 | std::lock_guard l(vstatfs_lock); | |
12571 | auto& stats = osd_pools[txc->osd_pool_id]; | |
12572 | stats += txc->statfs_delta; | |
12573 | ||
12574 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
12575 | ||
12576 | } else { | |
39ae355f TL |
12577 | if (!is_statfs_recoverable()) { |
12578 | bufferlist bl; | |
12579 | txc->statfs_delta.encode(bl); | |
12580 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
12581 | } | |
7c673cae | 12582 | |
11fdf7f2 TL |
12583 | std::lock_guard l(vstatfs_lock); |
12584 | vstatfs += txc->statfs_delta; | |
12585 | } | |
7c673cae FG |
12586 | txc->statfs_delta.reset(); |
12587 | } | |
12588 | ||
12589 | void BlueStore::_txc_state_proc(TransContext *txc) | |
12590 | { | |
12591 | while (true) { | |
12592 | dout(10) << __func__ << " txc " << txc | |
12593 | << " " << txc->get_state_name() << dendl; | |
f67539c2 | 12594 | switch (txc->get_state()) { |
7c673cae | 12595 | case TransContext::STATE_PREPARE: |
9f95a23c | 12596 | throttle.log_state_latency(*txc, logger, l_bluestore_state_prepare_lat); |
7c673cae | 12597 | if (txc->ioc.has_pending_aios()) { |
f67539c2 TL |
12598 | txc->set_state(TransContext::STATE_AIO_WAIT); |
12599 | #ifdef WITH_BLKIN | |
12600 | if (txc->trace) { | |
12601 | txc->trace.keyval("pending aios", txc->ioc.num_pending.load()); | |
12602 | } | |
12603 | #endif | |
7c673cae FG |
12604 | txc->had_ios = true; |
12605 | _txc_aio_submit(txc); | |
12606 | return; | |
12607 | } | |
12608 | // ** fall-thru ** | |
12609 | ||
12610 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 | 12611 | { |
9f95a23c TL |
12612 | mono_clock::duration lat = throttle.log_state_latency( |
12613 | *txc, logger, l_bluestore_state_aio_wait_lat); | |
12614 | if (ceph::to_seconds<double>(lat) >= cct->_conf->bluestore_log_op_age) { | |
11fdf7f2 TL |
12615 | dout(0) << __func__ << " slow aio_wait, txc = " << txc |
12616 | << ", latency = " << lat | |
12617 | << dendl; | |
12618 | } | |
12619 | } | |
12620 | ||
7c673cae FG |
12621 | _txc_finish_io(txc); // may trigger blocked txc's too |
12622 | return; | |
12623 | ||
12624 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 12625 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
12626 | if (txc->had_ios) { |
12627 | ++txc->osr->txc_with_unstable_io; | |
12628 | } | |
9f95a23c | 12629 | throttle.log_state_latency(*txc, logger, l_bluestore_state_io_done_lat); |
f67539c2 | 12630 | txc->set_state(TransContext::STATE_KV_QUEUED); |
7c673cae FG |
12631 | if (cct->_conf->bluestore_sync_submit_transaction) { |
12632 | if (txc->last_nid >= nid_max || | |
12633 | txc->last_blobid >= blobid_max) { | |
12634 | dout(20) << __func__ | |
12635 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
12636 | << dendl; | |
12637 | } else if (txc->osr->kv_committing_serially) { | |
12638 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
12639 | << dendl; | |
12640 | // note: this is starvation-prone. once we have a txc in a busy | |
12641 | // sequencer that is committing serially it is possible to keep | |
12642 | // submitting new transactions fast enough that we get stuck doing | |
12643 | // so. the alternative is to block here... fixme? | |
12644 | } else if (txc->osr->txc_with_unstable_io) { | |
12645 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
12646 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
12647 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
12648 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
12649 | == 0) { | |
12650 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
12651 | << dendl; | |
12652 | } else { | |
9f95a23c | 12653 | _txc_apply_kv(txc, true); |
7c673cae FG |
12654 | } |
12655 | } | |
12656 | { | |
11fdf7f2 | 12657 | std::lock_guard l(kv_lock); |
7c673cae | 12658 | kv_queue.push_back(txc); |
9f95a23c TL |
12659 | if (!kv_sync_in_progress) { |
12660 | kv_sync_in_progress = true; | |
12661 | kv_cond.notify_one(); | |
12662 | } | |
f67539c2 | 12663 | if (txc->get_state() != TransContext::STATE_KV_SUBMITTED) { |
7c673cae FG |
12664 | kv_queue_unsubmitted.push_back(txc); |
12665 | ++txc->osr->kv_committing_serially; | |
12666 | } | |
31f18b77 FG |
12667 | if (txc->had_ios) |
12668 | kv_ios++; | |
12669 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
12670 | } |
12671 | return; | |
12672 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
12673 | _txc_committed_kv(txc); |
12674 | // ** fall-thru ** | |
12675 | ||
12676 | case TransContext::STATE_KV_DONE: | |
9f95a23c | 12677 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_done_lat); |
7c673cae | 12678 | if (txc->deferred_txn) { |
f67539c2 | 12679 | txc->set_state(TransContext::STATE_DEFERRED_QUEUED); |
7c673cae FG |
12680 | _deferred_queue(txc); |
12681 | return; | |
12682 | } | |
f67539c2 | 12683 | txc->set_state(TransContext::STATE_FINISHING); |
7c673cae FG |
12684 | break; |
12685 | ||
12686 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9f95a23c | 12687 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_cleanup_lat); |
f67539c2 | 12688 | txc->set_state(TransContext::STATE_FINISHING); |
7c673cae FG |
12689 | // ** fall-thru ** |
12690 | ||
12691 | case TransContext::STATE_FINISHING: | |
9f95a23c | 12692 | throttle.log_state_latency(*txc, logger, l_bluestore_state_finishing_lat); |
7c673cae FG |
12693 | _txc_finish(txc); |
12694 | return; | |
12695 | ||
12696 | default: | |
12697 | derr << __func__ << " unexpected txc " << txc | |
12698 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 12699 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
12700 | return; |
12701 | } | |
12702 | } | |
12703 | } | |
12704 | ||
12705 | void BlueStore::_txc_finish_io(TransContext *txc) | |
12706 | { | |
12707 | dout(20) << __func__ << " " << txc << dendl; | |
12708 | ||
12709 | /* | |
12710 | * we need to preserve the order of kv transactions, | |
12711 | * even though aio will complete in any order. | |
12712 | */ | |
12713 | ||
12714 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 12715 | std::lock_guard l(osr->qlock); |
f67539c2 | 12716 | txc->set_state(TransContext::STATE_IO_DONE); |
11fdf7f2 | 12717 | txc->ioc.release_running_aios(); |
7c673cae FG |
12718 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
12719 | while (p != osr->q.begin()) { | |
12720 | --p; | |
f67539c2 | 12721 | if (p->get_state() < TransContext::STATE_IO_DONE) { |
7c673cae FG |
12722 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " |
12723 | << p->get_state_name() << dendl; | |
12724 | return; | |
12725 | } | |
f67539c2 | 12726 | if (p->get_state() > TransContext::STATE_IO_DONE) { |
7c673cae FG |
12727 | ++p; |
12728 | break; | |
12729 | } | |
12730 | } | |
12731 | do { | |
12732 | _txc_state_proc(&*p++); | |
12733 | } while (p != osr->q.end() && | |
f67539c2 | 12734 | p->get_state() == TransContext::STATE_IO_DONE); |
7c673cae | 12735 | |
11fdf7f2 | 12736 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
12737 | osr->qcond.notify_all(); |
12738 | } | |
12739 | } | |
12740 | ||
12741 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
12742 | { | |
12743 | dout(20) << __func__ << " txc " << txc | |
12744 | << " onodes " << txc->onodes | |
12745 | << " shared_blobs " << txc->shared_blobs | |
12746 | << dendl; | |
12747 | ||
12748 | // finalize onodes | |
12749 | for (auto o : txc->onodes) { | |
11fdf7f2 | 12750 | _record_onode(o, t); |
7c673cae FG |
12751 | o->flushing_count++; |
12752 | } | |
12753 | ||
12754 | // objects we modified but didn't affect the onode | |
12755 | auto p = txc->modified_objects.begin(); | |
12756 | while (p != txc->modified_objects.end()) { | |
12757 | if (txc->onodes.count(*p) == 0) { | |
12758 | (*p)->flushing_count++; | |
12759 | ++p; | |
12760 | } else { | |
12761 | // remove dups with onodes list to avoid problems in _txc_finish | |
12762 | p = txc->modified_objects.erase(p); | |
12763 | } | |
12764 | } | |
12765 | ||
12766 | // finalize shared_blobs | |
12767 | for (auto sb : txc->shared_blobs) { | |
12768 | string key; | |
12769 | auto sbid = sb->get_sbid(); | |
12770 | get_shared_blob_key(sbid, &key); | |
12771 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
12772 | dout(20) << __func__ << " shared_blob 0x" |
12773 | << std::hex << sbid << std::dec | |
7c673cae FG |
12774 | << " is empty" << dendl; |
12775 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
12776 | } else { | |
12777 | bufferlist bl; | |
11fdf7f2 TL |
12778 | encode(*(sb->persistent), bl); |
12779 | dout(20) << __func__ << " shared_blob 0x" | |
12780 | << std::hex << sbid << std::dec | |
31f18b77 | 12781 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
12782 | t->set(PREFIX_SHARED_BLOB, key, bl); |
12783 | } | |
12784 | } | |
12785 | } | |
12786 | ||
12787 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
12788 | PerfCounters &logger) | |
12789 | { | |
11fdf7f2 TL |
12790 | os_commit_latency_ns.consume_next( |
12791 | logger.get_tavg_ns( | |
7c673cae | 12792 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
12793 | os_apply_latency_ns.consume_next( |
12794 | logger.get_tavg_ns( | |
7c673cae FG |
12795 | l_bluestore_commit_lat)); |
12796 | } | |
12797 | ||
12798 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) | |
12799 | { | |
12800 | dout(20) << __func__ << " txc " << txc << std::hex | |
12801 | << " allocated 0x" << txc->allocated | |
12802 | << " released 0x" << txc->released | |
12803 | << std::dec << dendl; | |
12804 | ||
20effc67 TL |
12805 | if (!fm->is_null_manager()) |
12806 | { | |
12807 | // We have to handle the case where we allocate *and* deallocate the | |
12808 | // same region in this transaction. The freelist doesn't like that. | |
12809 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
12810 | // debug check. But that's important.) | |
12811 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
12812 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
12813 | interval_set<uint64_t> *preleased = &txc->released; | |
12814 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
12815 | interval_set<uint64_t> overlap; | |
12816 | overlap.intersection_of(txc->allocated, txc->released); | |
12817 | if (!overlap.empty()) { | |
12818 | tmp_allocated = txc->allocated; | |
12819 | tmp_allocated.subtract(overlap); | |
12820 | tmp_released = txc->released; | |
12821 | tmp_released.subtract(overlap); | |
12822 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
12823 | << ", new allocated 0x" << tmp_allocated | |
12824 | << " released 0x" << tmp_released << std::dec | |
12825 | << dendl; | |
12826 | pallocated = &tmp_allocated; | |
12827 | preleased = &tmp_released; | |
12828 | } | |
7c673cae | 12829 | } |
7c673cae | 12830 | |
20effc67 TL |
12831 | // update freelist with non-overlap sets |
12832 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
12833 | p != pallocated->end(); | |
12834 | ++p) { | |
12835 | fm->allocate(p.get_start(), p.get_len(), t); | |
12836 | } | |
12837 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
12838 | p != preleased->end(); | |
12839 | ++p) { | |
12840 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
12841 | << "~" << p.get_len() << std::dec << dendl; | |
12842 | fm->release(p.get_start(), p.get_len(), t); | |
12843 | } | |
7c673cae FG |
12844 | } |
12845 | ||
20effc67 | 12846 | #ifdef HAVE_LIBZBD |
f67539c2 | 12847 | if (bdev->is_smr()) { |
20effc67 TL |
12848 | for (auto& i : txc->old_zone_offset_refs) { |
12849 | dout(20) << __func__ << " rm ref zone 0x" << std::hex << i.first.second | |
12850 | << " offset 0x" << i.second << std::dec | |
12851 | << " -> " << i.first.first->oid << dendl; | |
12852 | string key; | |
12853 | get_zone_offset_object_key(i.first.second, i.second, i.first.first->oid, &key); | |
12854 | txc->t->rmkey(PREFIX_ZONED_CL_INFO, key); | |
12855 | } | |
12856 | for (auto& i : txc->new_zone_offset_refs) { | |
12857 | // (zone, offset) -> oid | |
12858 | dout(20) << __func__ << " add ref zone 0x" << std::hex << i.first.second | |
12859 | << " offset 0x" << i.second << std::dec | |
12860 | << " -> " << i.first.first->oid << dendl; | |
12861 | string key; | |
12862 | get_zone_offset_object_key(i.first.second, i.second, i.first.first->oid, &key); | |
12863 | bufferlist v; | |
12864 | txc->t->set(PREFIX_ZONED_CL_INFO, key, v); | |
12865 | } | |
f67539c2 | 12866 | } |
20effc67 | 12867 | #endif |
f67539c2 | 12868 | |
7c673cae FG |
12869 | _txc_update_store_statfs(txc); |
12870 | } | |
12871 | ||
9f95a23c | 12872 | void BlueStore::_txc_apply_kv(TransContext *txc, bool sync_submit_transaction) |
7c673cae | 12873 | { |
f67539c2 | 12874 | ceph_assert(txc->get_state() == TransContext::STATE_KV_QUEUED); |
9f95a23c TL |
12875 | { |
12876 | #if defined(WITH_LTTNG) | |
12877 | auto start = mono_clock::now(); | |
12878 | #endif | |
12879 | ||
f67539c2 TL |
12880 | #ifdef WITH_BLKIN |
12881 | if (txc->trace) { | |
12882 | txc->trace.event("db async submit"); | |
12883 | } | |
12884 | #endif | |
12885 | ||
9f95a23c TL |
12886 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); |
12887 | ceph_assert(r == 0); | |
f67539c2 | 12888 | txc->set_state(TransContext::STATE_KV_SUBMITTED); |
9f95a23c TL |
12889 | if (txc->osr->kv_submitted_waiters) { |
12890 | std::lock_guard l(txc->osr->qlock); | |
12891 | txc->osr->qcond.notify_all(); | |
12892 | } | |
12893 | ||
12894 | #if defined(WITH_LTTNG) | |
12895 | if (txc->tracing) { | |
12896 | tracepoint( | |
12897 | bluestore, | |
12898 | transaction_kv_submit_latency, | |
12899 | txc->osr->get_sequencer_id(), | |
12900 | txc->seq, | |
12901 | sync_submit_transaction, | |
12902 | ceph::to_seconds<double>(mono_clock::now() - start)); | |
12903 | } | |
12904 | #endif | |
12905 | } | |
12906 | ||
7c673cae FG |
12907 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { |
12908 | for (auto& o : *ls) { | |
12909 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
12910 | << dendl; | |
9f95a23c | 12911 | if (--o->flushing_count == 0 && o->waiting_count.load()) { |
11fdf7f2 | 12912 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
12913 | o->flush_cond.notify_all(); |
12914 | } | |
12915 | } | |
12916 | } | |
12917 | } | |
12918 | ||
12919 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
12920 | { | |
12921 | dout(20) << __func__ << " txc " << txc << dendl; | |
9f95a23c | 12922 | throttle.complete_kv(*txc); |
1adf2230 | 12923 | { |
11fdf7f2 | 12924 | std::lock_guard l(txc->osr->qlock); |
f67539c2 | 12925 | txc->set_state(TransContext::STATE_KV_DONE); |
11fdf7f2 TL |
12926 | if (txc->ch->commit_queue) { |
12927 | txc->ch->commit_queue->queue(txc->oncommits); | |
12928 | } else { | |
12929 | finisher.queue(txc->oncommits); | |
1adf2230 | 12930 | } |
7c673cae | 12931 | } |
9f95a23c | 12932 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_committing_lat); |
494da23a TL |
12933 | log_latency_fn( |
12934 | __func__, | |
12935 | l_bluestore_commit_lat, | |
9f95a23c | 12936 | mono_clock::now() - txc->start, |
494da23a TL |
12937 | cct->_conf->bluestore_log_op_age, |
12938 | [&](auto lat) { | |
12939 | return ", txc = " + stringify(txc); | |
12940 | } | |
11fdf7f2 | 12941 | ); |
7c673cae FG |
12942 | } |
12943 | ||
12944 | void BlueStore::_txc_finish(TransContext *txc) | |
12945 | { | |
12946 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
f67539c2 | 12947 | ceph_assert(txc->get_state() == TransContext::STATE_FINISHING); |
7c673cae FG |
12948 | |
12949 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 12950 | sb->finish_write(txc->seq); |
7c673cae FG |
12951 | } |
12952 | txc->shared_blobs_written.clear(); | |
12953 | ||
12954 | while (!txc->removed_collections.empty()) { | |
12955 | _queue_reap_collection(txc->removed_collections.front()); | |
12956 | txc->removed_collections.pop_front(); | |
12957 | } | |
12958 | ||
12959 | OpSequencerRef osr = txc->osr; | |
7c673cae | 12960 | bool empty = false; |
31f18b77 | 12961 | bool submit_deferred = false; |
7c673cae FG |
12962 | OpSequencer::q_list_t releasing_txc; |
12963 | { | |
11fdf7f2 | 12964 | std::lock_guard l(osr->qlock); |
f67539c2 | 12965 | txc->set_state(TransContext::STATE_DONE); |
7c673cae FG |
12966 | bool notify = false; |
12967 | while (!osr->q.empty()) { | |
12968 | TransContext *txc = &osr->q.front(); | |
12969 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
12970 | << dendl; | |
f67539c2 TL |
12971 | if (txc->get_state() != TransContext::STATE_DONE) { |
12972 | if (txc->get_state() == TransContext::STATE_PREPARE && | |
7c673cae FG |
12973 | deferred_aggressive) { |
12974 | // for _osr_drain_preceding() | |
12975 | notify = true; | |
12976 | } | |
f67539c2 | 12977 | if (txc->get_state() == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 12978 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
12979 | submit_deferred = true; |
12980 | } | |
7c673cae FG |
12981 | break; |
12982 | } | |
12983 | ||
7c673cae FG |
12984 | osr->q.pop_front(); |
12985 | releasing_txc.push_back(*txc); | |
7c673cae | 12986 | } |
9f95a23c | 12987 | |
7c673cae FG |
12988 | if (osr->q.empty()) { |
12989 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
12990 | empty = true; | |
12991 | } | |
9f95a23c TL |
12992 | |
12993 | // only drain()/drain_preceding() need wakeup, | |
12994 | // other cases use kv_submitted_waiters | |
12995 | if (notify || empty) { | |
12996 | osr->qcond.notify_all(); | |
12997 | } | |
7c673cae | 12998 | } |
9f95a23c | 12999 | |
7c673cae FG |
13000 | while (!releasing_txc.empty()) { |
13001 | // release to allocator only after all preceding txc's have also | |
13002 | // finished any deferred writes that potentially land in these | |
13003 | // blocks | |
13004 | auto txc = &releasing_txc.front(); | |
13005 | _txc_release_alloc(txc); | |
13006 | releasing_txc.pop_front(); | |
9f95a23c TL |
13007 | throttle.log_state_latency(*txc, logger, l_bluestore_state_done_lat); |
13008 | throttle.complete(*txc); | |
7c673cae FG |
13009 | delete txc; |
13010 | } | |
13011 | ||
31f18b77 FG |
13012 | if (submit_deferred) { |
13013 | // we're pinning memory; flush! we could be more fine-grained here but | |
13014 | // i'm not sure it's worth the bother. | |
13015 | deferred_try_submit(); | |
7c673cae FG |
13016 | } |
13017 | ||
7c673cae | 13018 | if (empty && osr->zombie) { |
11fdf7f2 TL |
13019 | std::lock_guard l(zombie_osr_lock); |
13020 | if (zombie_osr_set.erase(osr->cid)) { | |
13021 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
13022 | } else { | |
13023 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
13024 | << dendl; | |
13025 | } | |
7c673cae | 13026 | } |
9f95a23c | 13027 | } |
7c673cae FG |
13028 | |
13029 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
13030 | { | |
1e59de90 | 13031 | bool discard_queued = false; |
a8e16298 | 13032 | // it's expected we're called with lazy_release_lock already taken! |
1e59de90 TL |
13033 | if (unlikely(cct->_conf->bluestore_debug_no_reuse_blocks)) { |
13034 | goto out; | |
13035 | } | |
13036 | discard_queued = bdev->try_discard(txc->released); | |
13037 | // if async discard succeeded, will do alloc->release when discard callback | |
13038 | // else we should release here | |
13039 | if (!discard_queued) { | |
13040 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
13041 | << txc->released << std::dec << dendl; | |
13042 | alloc->release(txc->released); | |
7c673cae FG |
13043 | } |
13044 | ||
11fdf7f2 | 13045 | out: |
7c673cae FG |
13046 | txc->allocated.clear(); |
13047 | txc->released.clear(); | |
13048 | } | |
13049 | ||
11fdf7f2 TL |
13050 | void BlueStore::_osr_attach(Collection *c) |
13051 | { | |
20effc67 | 13052 | // note: caller has coll_lock |
11fdf7f2 TL |
13053 | auto q = coll_map.find(c->cid); |
13054 | if (q != coll_map.end()) { | |
13055 | c->osr = q->second->osr; | |
13056 | ldout(cct, 10) << __func__ << " " << c->cid | |
13057 | << " reusing osr " << c->osr << " from existing coll " | |
13058 | << q->second << dendl; | |
13059 | } else { | |
13060 | std::lock_guard l(zombie_osr_lock); | |
13061 | auto p = zombie_osr_set.find(c->cid); | |
13062 | if (p == zombie_osr_set.end()) { | |
9f95a23c | 13063 | c->osr = ceph::make_ref<OpSequencer>(this, next_sequencer_id++, c->cid); |
11fdf7f2 TL |
13064 | ldout(cct, 10) << __func__ << " " << c->cid |
13065 | << " fresh osr " << c->osr << dendl; | |
13066 | } else { | |
13067 | c->osr = p->second; | |
13068 | zombie_osr_set.erase(p); | |
13069 | ldout(cct, 10) << __func__ << " " << c->cid | |
13070 | << " resurrecting zombie osr " << c->osr << dendl; | |
13071 | c->osr->zombie = false; | |
13072 | } | |
13073 | } | |
13074 | } | |
13075 | ||
13076 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
13077 | { | |
13078 | std::lock_guard l(zombie_osr_lock); | |
13079 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
13080 | osr->zombie = true; | |
13081 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
13082 | // this is either a new insertion or the same osr is already there | |
13083 | ceph_assert(i.second || i.first->second == osr); | |
13084 | } | |
13085 | ||
7c673cae FG |
13086 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
13087 | { | |
13088 | OpSequencer *osr = txc->osr.get(); | |
13089 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
13090 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
13091 | { | |
13092 | // submit anything pending | |
f67539c2 | 13093 | osr->deferred_lock.lock(); |
11fdf7f2 | 13094 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
13095 | _deferred_submit_unlock(osr); |
13096 | } else { | |
f67539c2 | 13097 | osr->deferred_lock.unlock(); |
7c673cae FG |
13098 | } |
13099 | } | |
13100 | { | |
13101 | // wake up any previously finished deferred events | |
11fdf7f2 | 13102 | std::lock_guard l(kv_lock); |
9f95a23c TL |
13103 | if (!kv_sync_in_progress) { |
13104 | kv_sync_in_progress = true; | |
13105 | kv_cond.notify_one(); | |
13106 | } | |
7c673cae FG |
13107 | } |
13108 | osr->drain_preceding(txc); | |
13109 | --deferred_aggressive; | |
13110 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
13111 | } | |
13112 | ||
11fdf7f2 TL |
13113 | void BlueStore::_osr_drain(OpSequencer *osr) |
13114 | { | |
13115 | dout(10) << __func__ << " " << osr << dendl; | |
13116 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
13117 | { | |
13118 | // submit anything pending | |
f67539c2 | 13119 | osr->deferred_lock.lock(); |
11fdf7f2 TL |
13120 | if (osr->deferred_pending && !osr->deferred_running) { |
13121 | _deferred_submit_unlock(osr); | |
13122 | } else { | |
f67539c2 | 13123 | osr->deferred_lock.unlock(); |
11fdf7f2 TL |
13124 | } |
13125 | } | |
13126 | { | |
13127 | // wake up any previously finished deferred events | |
13128 | std::lock_guard l(kv_lock); | |
9f95a23c TL |
13129 | if (!kv_sync_in_progress) { |
13130 | kv_sync_in_progress = true; | |
13131 | kv_cond.notify_one(); | |
13132 | } | |
11fdf7f2 TL |
13133 | } |
13134 | osr->drain(); | |
13135 | --deferred_aggressive; | |
13136 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
13137 | } | |
13138 | ||
7c673cae FG |
13139 | void BlueStore::_osr_drain_all() |
13140 | { | |
13141 | dout(10) << __func__ << dendl; | |
13142 | ||
13143 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
13144 | vector<OpSequencerRef> zombies; |
13145 | { | |
9f95a23c | 13146 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
13147 | for (auto& i : coll_map) { |
13148 | s.insert(i.second->osr); | |
13149 | } | |
13150 | } | |
7c673cae | 13151 | { |
11fdf7f2 TL |
13152 | std::lock_guard l(zombie_osr_lock); |
13153 | for (auto& i : zombie_osr_set) { | |
13154 | s.insert(i.second); | |
13155 | zombies.push_back(i.second); | |
13156 | } | |
7c673cae FG |
13157 | } |
13158 | dout(20) << __func__ << " osr_set " << s << dendl; | |
13159 | ||
13160 | ++deferred_aggressive; | |
13161 | { | |
13162 | // submit anything pending | |
224ce89b | 13163 | deferred_try_submit(); |
7c673cae FG |
13164 | } |
13165 | { | |
13166 | // wake up any previously finished deferred events | |
11fdf7f2 | 13167 | std::lock_guard l(kv_lock); |
7c673cae FG |
13168 | kv_cond.notify_one(); |
13169 | } | |
31f18b77 | 13170 | { |
11fdf7f2 | 13171 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
13172 | kv_finalize_cond.notify_one(); |
13173 | } | |
7c673cae FG |
13174 | for (auto osr : s) { |
13175 | dout(20) << __func__ << " drain " << osr << dendl; | |
13176 | osr->drain(); | |
13177 | } | |
13178 | --deferred_aggressive; | |
13179 | ||
7c673cae | 13180 | { |
11fdf7f2 TL |
13181 | std::lock_guard l(zombie_osr_lock); |
13182 | for (auto& osr : zombies) { | |
13183 | if (zombie_osr_set.erase(osr->cid)) { | |
13184 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
13185 | ceph_assert(osr->q.empty()); | |
13186 | } else if (osr->zombie) { | |
13187 | dout(10) << __func__ << " empty zombie osr " << osr | |
13188 | << " already reaped" << dendl; | |
13189 | ceph_assert(osr->q.empty()); | |
13190 | } else { | |
13191 | dout(10) << __func__ << " empty zombie osr " << osr | |
13192 | << " resurrected" << dendl; | |
13193 | } | |
7c673cae FG |
13194 | } |
13195 | } | |
11fdf7f2 TL |
13196 | |
13197 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
13198 | } |
13199 | ||
11fdf7f2 | 13200 | |
31f18b77 FG |
13201 | void BlueStore::_kv_start() |
13202 | { | |
13203 | dout(10) << __func__ << dendl; | |
13204 | ||
11fdf7f2 | 13205 | finisher.start(); |
31f18b77 FG |
13206 | kv_sync_thread.create("bstore_kv_sync"); |
13207 | kv_finalize_thread.create("bstore_kv_final"); | |
13208 | } | |
13209 | ||
13210 | void BlueStore::_kv_stop() | |
13211 | { | |
13212 | dout(10) << __func__ << dendl; | |
13213 | { | |
9f95a23c | 13214 | std::unique_lock l{kv_lock}; |
31f18b77 FG |
13215 | while (!kv_sync_started) { |
13216 | kv_cond.wait(l); | |
13217 | } | |
13218 | kv_stop = true; | |
13219 | kv_cond.notify_all(); | |
13220 | } | |
13221 | { | |
9f95a23c | 13222 | std::unique_lock l{kv_finalize_lock}; |
31f18b77 FG |
13223 | while (!kv_finalize_started) { |
13224 | kv_finalize_cond.wait(l); | |
13225 | } | |
13226 | kv_finalize_stop = true; | |
13227 | kv_finalize_cond.notify_all(); | |
13228 | } | |
13229 | kv_sync_thread.join(); | |
13230 | kv_finalize_thread.join(); | |
11fdf7f2 | 13231 | ceph_assert(removed_collections.empty()); |
31f18b77 | 13232 | { |
11fdf7f2 | 13233 | std::lock_guard l(kv_lock); |
31f18b77 FG |
13234 | kv_stop = false; |
13235 | } | |
13236 | { | |
11fdf7f2 | 13237 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
13238 | kv_finalize_stop = false; |
13239 | } | |
13240 | dout(10) << __func__ << " stopping finishers" << dendl; | |
11fdf7f2 TL |
13241 | finisher.wait_for_empty(); |
13242 | finisher.stop(); | |
31f18b77 FG |
13243 | dout(10) << __func__ << " stopped" << dendl; |
13244 | } | |
13245 | ||
7c673cae FG |
13246 | void BlueStore::_kv_sync_thread() |
13247 | { | |
13248 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 | 13249 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
9f95a23c | 13250 | std::unique_lock l{kv_lock}; |
11fdf7f2 | 13251 | ceph_assert(!kv_sync_started); |
31f18b77 FG |
13252 | kv_sync_started = true; |
13253 | kv_cond.notify_all(); | |
adb31ebb TL |
13254 | |
13255 | auto t0 = mono_clock::now(); | |
13256 | timespan twait = ceph::make_timespan(0); | |
13257 | size_t kv_submitted = 0; | |
13258 | ||
7c673cae | 13259 | while (true) { |
adb31ebb TL |
13260 | auto period = cct->_conf->bluestore_kv_sync_util_logging_s; |
13261 | auto observation_period = | |
13262 | ceph::make_timespan(period); | |
13263 | auto elapsed = mono_clock::now() - t0; | |
13264 | if (period && elapsed >= observation_period) { | |
13265 | dout(5) << __func__ << " utilization: idle " | |
13266 | << twait << " of " << elapsed | |
13267 | << ", submitted: " << kv_submitted | |
13268 | <<dendl; | |
13269 | t0 = mono_clock::now(); | |
13270 | twait = ceph::make_timespan(0); | |
13271 | kv_submitted = 0; | |
13272 | } | |
11fdf7f2 | 13273 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
13274 | if (kv_queue.empty() && |
13275 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 13276 | !deferred_aggressive)) { |
7c673cae FG |
13277 | if (kv_stop) |
13278 | break; | |
13279 | dout(20) << __func__ << " sleep" << dendl; | |
adb31ebb | 13280 | auto t = mono_clock::now(); |
9f95a23c | 13281 | kv_sync_in_progress = false; |
11fdf7f2 | 13282 | kv_cond.wait(l); |
adb31ebb TL |
13283 | twait += mono_clock::now() - t; |
13284 | ||
7c673cae FG |
13285 | dout(20) << __func__ << " wake" << dendl; |
13286 | } else { | |
13287 | deque<TransContext*> kv_submitting; | |
13288 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
13289 | uint64_t aios = 0, costs = 0; |
13290 | ||
7c673cae FG |
13291 | dout(20) << __func__ << " committing " << kv_queue.size() |
13292 | << " submitting " << kv_queue_unsubmitted.size() | |
13293 | << " deferred done " << deferred_done_queue.size() | |
13294 | << " stable " << deferred_stable_queue.size() | |
13295 | << dendl; | |
13296 | kv_committing.swap(kv_queue); | |
13297 | kv_submitting.swap(kv_queue_unsubmitted); | |
13298 | deferred_done.swap(deferred_done_queue); | |
13299 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
13300 | aios = kv_ios; |
13301 | costs = kv_throttle_costs; | |
13302 | kv_ios = 0; | |
13303 | kv_throttle_costs = 0; | |
7c673cae FG |
13304 | l.unlock(); |
13305 | ||
13306 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
13307 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
13308 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
13309 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
13310 | ||
11fdf7f2 TL |
13311 | auto start = mono_clock::now(); |
13312 | ||
7c673cae FG |
13313 | bool force_flush = false; |
13314 | // if bluefs is sharing the same device as data (only), then we | |
13315 | // can rely on the bluefs commit to flush the device and make | |
13316 | // deferred aios stable. that means that if we do have done deferred | |
13317 | // txcs AND we are not on a single device, we need to force a flush. | |
9f95a23c | 13318 | if (bluefs && bluefs_layout.single_shared_device()) { |
31f18b77 | 13319 | if (aios) { |
7c673cae | 13320 | force_flush = true; |
11fdf7f2 | 13321 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
13322 | force_flush = true; // there's nothing else to commit! |
13323 | } else if (deferred_aggressive) { | |
13324 | force_flush = true; | |
13325 | } | |
11fdf7f2 TL |
13326 | } else { |
13327 | if (aios || !deferred_done.empty()) { | |
13328 | force_flush = true; | |
13329 | } else { | |
13330 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
13331 | } | |
13332 | } | |
7c673cae FG |
13333 | |
13334 | if (force_flush) { | |
31f18b77 | 13335 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
13336 | << " force_flush=" << (int)force_flush |
13337 | << ", flushing, deferred done->stable" << dendl; | |
13338 | // flush/barrier on block device | |
13339 | bdev->flush(); | |
13340 | ||
1e59de90 TL |
13341 | // if we flush then deferred done are now deferred stable |
13342 | if (deferred_stable.empty()) { | |
13343 | deferred_stable.swap(deferred_done); | |
13344 | } else { | |
13345 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
13346 | deferred_done.end()); | |
13347 | deferred_done.clear(); | |
13348 | } | |
7c673cae | 13349 | } |
11fdf7f2 | 13350 | auto after_flush = mono_clock::now(); |
7c673cae FG |
13351 | |
13352 | // we will use one final transaction to force a sync | |
13353 | KeyValueDB::Transaction synct = db->get_transaction(); | |
13354 | ||
13355 | // increase {nid,blobid}_max? note that this covers both the | |
13356 | // case where we are approaching the max and the case we passed | |
13357 | // it. in either case, we increase the max in the earlier txn | |
13358 | // we submit. | |
13359 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
13360 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
13361 | KeyValueDB::Transaction t = | |
13362 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
13363 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
13364 | bufferlist bl; | |
11fdf7f2 | 13365 | encode(new_nid_max, bl); |
7c673cae FG |
13366 | t->set(PREFIX_SUPER, "nid_max", bl); |
13367 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
13368 | } | |
13369 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
13370 | KeyValueDB::Transaction t = | |
13371 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
13372 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
13373 | bufferlist bl; | |
11fdf7f2 | 13374 | encode(new_blobid_max, bl); |
7c673cae FG |
13375 | t->set(PREFIX_SUPER, "blobid_max", bl); |
13376 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
13377 | } | |
c07f9fc5 FG |
13378 | |
13379 | for (auto txc : kv_committing) { | |
9f95a23c | 13380 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_queued_lat); |
f67539c2 | 13381 | if (txc->get_state() == TransContext::STATE_KV_QUEUED) { |
adb31ebb | 13382 | ++kv_submitted; |
9f95a23c | 13383 | _txc_apply_kv(txc, false); |
c07f9fc5 | 13384 | --txc->osr->kv_committing_serially; |
c07f9fc5 | 13385 | } else { |
f67539c2 | 13386 | ceph_assert(txc->get_state() == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 13387 | } |
7c673cae FG |
13388 | if (txc->had_ios) { |
13389 | --txc->osr->txc_with_unstable_io; | |
13390 | } | |
7c673cae FG |
13391 | } |
13392 | ||
31f18b77 FG |
13393 | // release throttle *before* we commit. this allows new ops |
13394 | // to be prepared and enter pipeline while we are waiting on | |
13395 | // the kv commit sync/flush. then hopefully on the next | |
13396 | // iteration there will already be ops awake. otherwise, we | |
13397 | // end up going to sleep, and then wake up when the very first | |
13398 | // transaction is ready for commit. | |
9f95a23c | 13399 | throttle.release_kv_throttle(costs); |
31f18b77 | 13400 | |
7c673cae FG |
13401 | // cleanup sync deferred keys |
13402 | for (auto b : deferred_stable) { | |
13403 | for (auto& txc : b->txcs) { | |
13404 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 13405 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
13406 | string key; |
13407 | get_deferred_key(wt.seq, &key); | |
13408 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
13409 | } | |
13410 | } | |
13411 | ||
9f95a23c TL |
13412 | #if defined(WITH_LTTNG) |
13413 | auto sync_start = mono_clock::now(); | |
13414 | #endif | |
7c673cae | 13415 | // submit synct synchronously (block and wait for it to commit) |
31f18b77 | 13416 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
13417 | ceph_assert(r == 0); |
13418 | ||
f67539c2 TL |
13419 | #ifdef WITH_BLKIN |
13420 | for (auto txc : kv_committing) { | |
13421 | if (txc->trace) { | |
13422 | txc->trace.event("db sync submit"); | |
13423 | txc->trace.keyval("kv_committing size", kv_committing.size()); | |
13424 | } | |
13425 | } | |
13426 | #endif | |
13427 | ||
9f95a23c TL |
13428 | int committing_size = kv_committing.size(); |
13429 | int deferred_size = deferred_stable.size(); | |
13430 | ||
13431 | #if defined(WITH_LTTNG) | |
13432 | double sync_latency = ceph::to_seconds<double>(mono_clock::now() - sync_start); | |
13433 | for (auto txc: kv_committing) { | |
13434 | if (txc->tracing) { | |
13435 | tracepoint( | |
13436 | bluestore, | |
13437 | transaction_kv_sync_latency, | |
13438 | txc->osr->get_sequencer_id(), | |
13439 | txc->seq, | |
13440 | kv_committing.size(), | |
13441 | deferred_done.size(), | |
13442 | deferred_stable.size(), | |
13443 | sync_latency); | |
13444 | } | |
13445 | } | |
13446 | #endif | |
13447 | ||
11fdf7f2 | 13448 | { |
9f95a23c | 13449 | std::unique_lock m{kv_finalize_lock}; |
11fdf7f2 TL |
13450 | if (kv_committing_to_finalize.empty()) { |
13451 | kv_committing_to_finalize.swap(kv_committing); | |
13452 | } else { | |
13453 | kv_committing_to_finalize.insert( | |
13454 | kv_committing_to_finalize.end(), | |
13455 | kv_committing.begin(), | |
13456 | kv_committing.end()); | |
13457 | kv_committing.clear(); | |
13458 | } | |
13459 | if (deferred_stable_to_finalize.empty()) { | |
13460 | deferred_stable_to_finalize.swap(deferred_stable); | |
13461 | } else { | |
13462 | deferred_stable_to_finalize.insert( | |
13463 | deferred_stable_to_finalize.end(), | |
13464 | deferred_stable.begin(), | |
13465 | deferred_stable.end()); | |
13466 | deferred_stable.clear(); | |
13467 | } | |
9f95a23c TL |
13468 | if (!kv_finalize_in_progress) { |
13469 | kv_finalize_in_progress = true; | |
13470 | kv_finalize_cond.notify_one(); | |
13471 | } | |
11fdf7f2 | 13472 | } |
7c673cae FG |
13473 | |
13474 | if (new_nid_max) { | |
13475 | nid_max = new_nid_max; | |
13476 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
13477 | } | |
13478 | if (new_blobid_max) { | |
13479 | blobid_max = new_blobid_max; | |
13480 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
13481 | } | |
13482 | ||
224ce89b | 13483 | { |
11fdf7f2 TL |
13484 | auto finish = mono_clock::now(); |
13485 | ceph::timespan dur_flush = after_flush - start; | |
13486 | ceph::timespan dur_kv = finish - after_flush; | |
13487 | ceph::timespan dur = finish - start; | |
9f95a23c TL |
13488 | dout(20) << __func__ << " committed " << committing_size |
13489 | << " cleaned " << deferred_size | |
224ce89b WB |
13490 | << " in " << dur |
13491 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
13492 | << dendl; | |
494da23a TL |
13493 | log_latency("kv_flush", |
13494 | l_bluestore_kv_flush_lat, | |
13495 | dur_flush, | |
13496 | cct->_conf->bluestore_log_op_age); | |
13497 | log_latency("kv_commit", | |
13498 | l_bluestore_kv_commit_lat, | |
13499 | dur_kv, | |
13500 | cct->_conf->bluestore_log_op_age); | |
13501 | log_latency("kv_sync", | |
13502 | l_bluestore_kv_sync_lat, | |
13503 | dur, | |
13504 | cct->_conf->bluestore_log_op_age); | |
7c673cae | 13505 | } |
31f18b77 | 13506 | |
31f18b77 FG |
13507 | l.lock(); |
13508 | // previously deferred "done" are now "stable" by virtue of this | |
13509 | // commit cycle. | |
13510 | deferred_stable_queue.swap(deferred_done); | |
13511 | } | |
13512 | } | |
13513 | dout(10) << __func__ << " finish" << dendl; | |
13514 | kv_sync_started = false; | |
13515 | } | |
13516 | ||
13517 | void BlueStore::_kv_finalize_thread() | |
13518 | { | |
13519 | deque<TransContext*> kv_committed; | |
13520 | deque<DeferredBatch*> deferred_stable; | |
13521 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
13522 | std::unique_lock l(kv_finalize_lock); |
13523 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
13524 | kv_finalize_started = true; |
13525 | kv_finalize_cond.notify_all(); | |
13526 | while (true) { | |
11fdf7f2 TL |
13527 | ceph_assert(kv_committed.empty()); |
13528 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
13529 | if (kv_committing_to_finalize.empty() && |
13530 | deferred_stable_to_finalize.empty()) { | |
13531 | if (kv_finalize_stop) | |
13532 | break; | |
13533 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 13534 | kv_finalize_in_progress = false; |
31f18b77 FG |
13535 | kv_finalize_cond.wait(l); |
13536 | dout(20) << __func__ << " wake" << dendl; | |
13537 | } else { | |
13538 | kv_committed.swap(kv_committing_to_finalize); | |
13539 | deferred_stable.swap(deferred_stable_to_finalize); | |
13540 | l.unlock(); | |
13541 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
13542 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
13543 | ||
11fdf7f2 TL |
13544 | auto start = mono_clock::now(); |
13545 | ||
31f18b77 FG |
13546 | while (!kv_committed.empty()) { |
13547 | TransContext *txc = kv_committed.front(); | |
f67539c2 | 13548 | ceph_assert(txc->get_state() == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 13549 | _txc_state_proc(txc); |
31f18b77 | 13550 | kv_committed.pop_front(); |
7c673cae | 13551 | } |
31f18b77 | 13552 | |
7c673cae FG |
13553 | for (auto b : deferred_stable) { |
13554 | auto p = b->txcs.begin(); | |
13555 | while (p != b->txcs.end()) { | |
13556 | TransContext *txc = &*p; | |
13557 | p = b->txcs.erase(p); // unlink here because | |
13558 | _txc_state_proc(txc); // this may destroy txc | |
13559 | } | |
13560 | delete b; | |
13561 | } | |
31f18b77 | 13562 | deferred_stable.clear(); |
7c673cae FG |
13563 | |
13564 | if (!deferred_aggressive) { | |
31f18b77 | 13565 | if (deferred_queue_size >= deferred_batch_ops.load() || |
9f95a23c | 13566 | throttle.should_submit_deferred()) { |
224ce89b | 13567 | deferred_try_submit(); |
7c673cae FG |
13568 | } |
13569 | } | |
13570 | ||
13571 | // this is as good a place as any ... | |
13572 | _reap_collections(); | |
13573 | ||
11fdf7f2 | 13574 | logger->set(l_bluestore_fragmentation, |
20effc67 | 13575 | (uint64_t)(alloc->get_fragmentation() * 1000)); |
11fdf7f2 | 13576 | |
494da23a TL |
13577 | log_latency("kv_final", |
13578 | l_bluestore_kv_final_lat, | |
13579 | mono_clock::now() - start, | |
13580 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 | 13581 | |
7c673cae | 13582 | l.lock(); |
7c673cae FG |
13583 | } |
13584 | } | |
13585 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 13586 | kv_finalize_started = false; |
7c673cae FG |
13587 | } |
13588 | ||
20effc67 TL |
13589 | #ifdef HAVE_LIBZBD |
13590 | void BlueStore::_zoned_cleaner_start() | |
13591 | { | |
f67539c2 | 13592 | dout(10) << __func__ << dendl; |
f67539c2 TL |
13593 | zoned_cleaner_thread.create("bstore_zcleaner"); |
13594 | } | |
13595 | ||
20effc67 TL |
13596 | void BlueStore::_zoned_cleaner_stop() |
13597 | { | |
f67539c2 TL |
13598 | dout(10) << __func__ << dendl; |
13599 | { | |
13600 | std::unique_lock l{zoned_cleaner_lock}; | |
13601 | while (!zoned_cleaner_started) { | |
13602 | zoned_cleaner_cond.wait(l); | |
13603 | } | |
13604 | zoned_cleaner_stop = true; | |
13605 | zoned_cleaner_cond.notify_all(); | |
13606 | } | |
13607 | zoned_cleaner_thread.join(); | |
13608 | { | |
13609 | std::lock_guard l{zoned_cleaner_lock}; | |
13610 | zoned_cleaner_stop = false; | |
13611 | } | |
13612 | dout(10) << __func__ << " done" << dendl; | |
13613 | } | |
13614 | ||
20effc67 TL |
13615 | void BlueStore::_zoned_cleaner_thread() |
13616 | { | |
f67539c2 TL |
13617 | dout(10) << __func__ << " start" << dendl; |
13618 | std::unique_lock l{zoned_cleaner_lock}; | |
13619 | ceph_assert(!zoned_cleaner_started); | |
13620 | zoned_cleaner_started = true; | |
13621 | zoned_cleaner_cond.notify_all(); | |
20effc67 TL |
13622 | auto a = dynamic_cast<ZonedAllocator*>(alloc); |
13623 | ceph_assert(a); | |
13624 | auto f = dynamic_cast<ZonedFreelistManager*>(fm); | |
13625 | ceph_assert(f); | |
f67539c2 | 13626 | while (true) { |
20effc67 TL |
13627 | // thresholds to trigger cleaning |
13628 | // FIXME | |
13629 | float min_score = .05; // score: bytes saved / bytes moved | |
13630 | uint64_t min_saved = zone_size / 32; // min bytes saved to consider cleaning | |
13631 | auto zone_to_clean = a->pick_zone_to_clean(min_score, min_saved); | |
13632 | if (zone_to_clean < 0) { | |
f67539c2 TL |
13633 | if (zoned_cleaner_stop) { |
13634 | break; | |
13635 | } | |
20effc67 TL |
13636 | auto period = ceph::make_timespan(cct->_conf->bluestore_cleaner_sleep_interval); |
13637 | dout(20) << __func__ << " sleep for " << period << dendl; | |
13638 | zoned_cleaner_cond.wait_for(l, period); | |
f67539c2 TL |
13639 | dout(20) << __func__ << " wake" << dendl; |
13640 | } else { | |
f67539c2 | 13641 | l.unlock(); |
20effc67 TL |
13642 | a->set_cleaning_zone(zone_to_clean); |
13643 | _zoned_clean_zone(zone_to_clean, a, f); | |
13644 | a->clear_cleaning_zone(zone_to_clean); | |
f67539c2 TL |
13645 | l.lock(); |
13646 | } | |
13647 | } | |
13648 | dout(10) << __func__ << " finish" << dendl; | |
13649 | zoned_cleaner_started = false; | |
13650 | } | |
13651 | ||
20effc67 TL |
13652 | void BlueStore::_zoned_clean_zone( |
13653 | uint64_t zone, | |
13654 | ZonedAllocator *a, | |
13655 | ZonedFreelistManager *f | |
13656 | ) | |
13657 | { | |
13658 | dout(10) << __func__ << " cleaning zone 0x" << std::hex << zone << std::dec << dendl; | |
13659 | ||
13660 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_ZONED_CL_INFO); | |
13661 | std::string zone_start; | |
13662 | get_zone_offset_object_key(zone, 0, ghobject_t(), &zone_start); | |
13663 | for (it->lower_bound(zone_start); it->valid(); it->next()) { | |
13664 | uint32_t z; | |
13665 | uint64_t offset; | |
13666 | ghobject_t oid; | |
13667 | string k = it->key(); | |
13668 | int r = get_key_zone_offset_object(k, &z, &offset, &oid); | |
13669 | if (r < 0) { | |
13670 | derr << __func__ << " failed to decode zone ref " << pretty_binary_string(k) | |
13671 | << dendl; | |
13672 | continue; | |
13673 | } | |
13674 | if (zone != z) { | |
13675 | dout(10) << __func__ << " reached end of zone refs" << dendl; | |
13676 | break; | |
13677 | } | |
13678 | dout(10) << __func__ << " zone 0x" << std::hex << zone << " offset 0x" << offset | |
13679 | << std::dec << " " << oid << dendl; | |
13680 | _clean_some(oid, zone); | |
13681 | } | |
13682 | ||
13683 | if (a->get_live_bytes(zone) > 0) { | |
13684 | derr << "zone 0x" << std::hex << zone << " still has 0x" << a->get_live_bytes(zone) | |
13685 | << " live bytes" << std::dec << dendl; | |
13686 | // should we do something else here to avoid a live-lock in the event of a problem? | |
13687 | return; | |
13688 | } | |
13689 | ||
13690 | // make sure transactions flush/drain/commit (and data is all rewritten | |
13691 | // safely elsewhere) before we blow away the cleaned zone | |
13692 | _osr_drain_all(); | |
13693 | ||
13694 | // reset the device zone | |
13695 | dout(10) << __func__ << " resetting zone 0x" << std::hex << zone << std::dec << dendl; | |
13696 | bdev->reset_zone(zone); | |
13697 | ||
13698 | // record that we can now write there | |
13699 | f->mark_zone_to_clean_free(zone, db); | |
13700 | bdev->flush(); | |
13701 | ||
13702 | // then allow ourselves to start allocating there | |
13703 | dout(10) << __func__ << " done cleaning zone 0x" << std::hex << zone << std::dec | |
13704 | << dendl; | |
13705 | a->reset_zone(zone); | |
13706 | } | |
13707 | ||
13708 | void BlueStore::_clean_some(ghobject_t oid, uint32_t zone) | |
13709 | { | |
13710 | dout(10) << __func__ << " " << oid << " from zone 0x" << std::hex << zone << std::dec | |
13711 | << dendl; | |
13712 | ||
13713 | CollectionRef cref = _get_collection_by_oid(oid); | |
13714 | if (!cref) { | |
13715 | dout(10) << __func__ << " can't find collection for " << oid << dendl; | |
13716 | return; | |
13717 | } | |
13718 | Collection *c = cref.get(); | |
13719 | ||
13720 | // serialize io dispatch vs other transactions | |
13721 | std::lock_guard l(atomic_alloc_and_submit_lock); | |
13722 | std::unique_lock l2(c->lock); | |
13723 | ||
13724 | auto o = c->get_onode(oid, false); | |
13725 | if (!o) { | |
13726 | dout(10) << __func__ << " can't find " << oid << dendl; | |
13727 | return; | |
13728 | } | |
13729 | ||
13730 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
13731 | _dump_onode<30>(cct, *o); | |
13732 | ||
13733 | // NOTE: This is a naive rewrite strategy. If any blobs are | |
13734 | // shared, they will be duplicated for each object that references | |
13735 | // them. That means any cloned/snapshotted objects will explode | |
13736 | // their utilization. This won't matter for RGW workloads, but | |
13737 | // for RBD and CephFS it is completely unacceptable, and it's | |
13738 | // entirely reasonable to have "archival" data workloads on SMR | |
13739 | // for CephFS and (possibly/probably) RBD. | |
13740 | // | |
13741 | // At some point we need to replace this with something more | |
13742 | // sophisticated that ensures that a shared blob gets moved once | |
13743 | // and all referencing objects get updated to point to the new | |
13744 | // location. | |
13745 | ||
13746 | map<uint32_t, uint32_t> to_move; | |
13747 | for (auto& e : o->extent_map.extent_map) { | |
13748 | bool touches_zone = false; | |
13749 | for (auto& be : e.blob->get_blob().get_extents()) { | |
13750 | if (be.is_valid()) { | |
13751 | uint32_t z = be.offset / zone_size; | |
13752 | if (z == zone) { | |
13753 | touches_zone = true; | |
13754 | break; | |
13755 | } | |
13756 | } | |
13757 | } | |
13758 | if (touches_zone) { | |
13759 | to_move[e.logical_offset] = e.length; | |
13760 | } | |
13761 | } | |
13762 | if (to_move.empty()) { | |
13763 | dout(10) << __func__ << " no references to zone 0x" << std::hex << zone | |
13764 | << std::dec << " from " << oid << dendl; | |
13765 | return; | |
13766 | } | |
13767 | ||
13768 | dout(10) << __func__ << " rewriting object extents 0x" << std::hex << to_move | |
13769 | << std::dec << dendl; | |
13770 | OpSequencer *osr = c->osr.get(); | |
13771 | TransContext *txc = _txc_create(c, osr, nullptr); | |
13772 | ||
13773 | spg_t pgid; | |
13774 | if (c->cid.is_pg(&pgid)) { | |
13775 | txc->osd_pool_id = pgid.pool(); | |
13776 | } | |
13777 | ||
13778 | for (auto& [offset, length] : to_move) { | |
13779 | bufferlist bl; | |
13780 | int r = _do_read(c, o, offset, length, bl, 0); | |
13781 | ceph_assert(r == (int)length); | |
13782 | ||
13783 | r = _do_write(txc, cref, o, offset, length, bl, 0); | |
13784 | ceph_assert(r >= 0); | |
13785 | } | |
13786 | txc->write_onode(o); | |
13787 | ||
13788 | _txc_write_nodes(txc, txc->t); | |
13789 | _txc_finalize_kv(txc, txc->t); | |
13790 | _txc_state_proc(txc); | |
f67539c2 | 13791 | } |
20effc67 | 13792 | #endif |
f67539c2 | 13793 | |
7c673cae | 13794 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( |
522d829b | 13795 | TransContext *txc, uint64_t len) |
7c673cae FG |
13796 | { |
13797 | if (!txc->deferred_txn) { | |
13798 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
13799 | } | |
13800 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
20effc67 TL |
13801 | logger->inc(l_bluestore_issued_deferred_writes); |
13802 | logger->inc(l_bluestore_issued_deferred_write_bytes, len); | |
7c673cae FG |
13803 | return &txc->deferred_txn->ops.back(); |
13804 | } | |
13805 | ||
13806 | void BlueStore::_deferred_queue(TransContext *txc) | |
13807 | { | |
13808 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
f67539c2 TL |
13809 | |
13810 | DeferredBatch *tmp; | |
13811 | txc->osr->deferred_lock.lock(); | |
13812 | { | |
13813 | if (!txc->osr->deferred_pending) { | |
13814 | tmp = new DeferredBatch(cct, txc->osr.get()); | |
13815 | } else { | |
13816 | tmp = txc->osr->deferred_pending; | |
13817 | } | |
7c673cae | 13818 | } |
f67539c2 TL |
13819 | |
13820 | tmp->txcs.push_back(*txc); | |
7c673cae FG |
13821 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; |
13822 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
13823 | const auto& op = *opi; | |
11fdf7f2 | 13824 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
13825 | bufferlist::const_iterator p = op.data.begin(); |
13826 | for (auto e : op.extents) { | |
f67539c2 | 13827 | tmp->prepare_write(cct, wt.seq, e.offset, e.length, p); |
7c673cae FG |
13828 | } |
13829 | } | |
f67539c2 TL |
13830 | |
13831 | { | |
13832 | ++deferred_queue_size; | |
13833 | txc->osr->deferred_pending = tmp; | |
13834 | // condition "tmp->txcs.size() == 1" mean deferred_pending was originally empty. | |
13835 | // So we should add osr into deferred_queue. | |
13836 | if (!txc->osr->deferred_running && (tmp->txcs.size() == 1)) { | |
13837 | deferred_lock.lock(); | |
13838 | deferred_queue.push_back(*txc->osr); | |
13839 | deferred_lock.unlock(); | |
13840 | } | |
13841 | ||
13842 | if (deferred_aggressive && | |
13843 | !txc->osr->deferred_running) { | |
13844 | _deferred_submit_unlock(txc->osr.get()); | |
13845 | } else { | |
13846 | txc->osr->deferred_lock.unlock(); | |
13847 | } | |
7c673cae | 13848 | } |
f67539c2 | 13849 | } |
7c673cae | 13850 | |
224ce89b | 13851 | void BlueStore::deferred_try_submit() |
7c673cae FG |
13852 | { |
13853 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
13854 | << deferred_queue_size << " txcs" << dendl; | |
224ce89b | 13855 | vector<OpSequencerRef> osrs; |
f67539c2 TL |
13856 | |
13857 | { | |
13858 | std::lock_guard l(deferred_lock); | |
13859 | osrs.reserve(deferred_queue.size()); | |
13860 | for (auto& osr : deferred_queue) { | |
13861 | osrs.push_back(&osr); | |
13862 | } | |
224ce89b | 13863 | } |
f67539c2 | 13864 | |
224ce89b | 13865 | for (auto& osr : osrs) { |
f67539c2 | 13866 | osr->deferred_lock.lock(); |
181888fb FG |
13867 | if (osr->deferred_pending) { |
13868 | if (!osr->deferred_running) { | |
13869 | _deferred_submit_unlock(osr.get()); | |
181888fb | 13870 | } else { |
f67539c2 | 13871 | osr->deferred_lock.unlock(); |
181888fb FG |
13872 | dout(20) << __func__ << " osr " << osr << " already has running" |
13873 | << dendl; | |
13874 | } | |
13875 | } else { | |
f67539c2 | 13876 | osr->deferred_lock.unlock(); |
181888fb | 13877 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; |
7c673cae FG |
13878 | } |
13879 | } | |
9f95a23c | 13880 | |
f67539c2 TL |
13881 | { |
13882 | std::lock_guard l(deferred_lock); | |
13883 | deferred_last_submitted = ceph_clock_now(); | |
13884 | } | |
7c673cae FG |
13885 | } |
13886 | ||
224ce89b | 13887 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
13888 | { |
13889 | dout(10) << __func__ << " osr " << osr | |
13890 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
13891 | << dendl; | |
11fdf7f2 TL |
13892 | ceph_assert(osr->deferred_pending); |
13893 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
13894 | |
13895 | auto b = osr->deferred_pending; | |
13896 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 13897 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
13898 | |
13899 | osr->deferred_running = osr->deferred_pending; | |
13900 | osr->deferred_pending = nullptr; | |
13901 | ||
f67539c2 | 13902 | osr->deferred_lock.unlock(); |
11fdf7f2 TL |
13903 | |
13904 | for (auto& txc : b->txcs) { | |
9f95a23c | 13905 | throttle.log_state_latency(txc, logger, l_bluestore_state_deferred_queued_lat); |
11fdf7f2 | 13906 | } |
7c673cae FG |
13907 | uint64_t start = 0, pos = 0; |
13908 | bufferlist bl; | |
13909 | auto i = b->iomap.begin(); | |
13910 | while (true) { | |
13911 | if (i == b->iomap.end() || i->first != pos) { | |
13912 | if (bl.length()) { | |
13913 | dout(20) << __func__ << " write 0x" << std::hex | |
13914 | << start << "~" << bl.length() | |
13915 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 13916 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
20effc67 TL |
13917 | logger->inc(l_bluestore_submitted_deferred_writes); |
13918 | logger->inc(l_bluestore_submitted_deferred_write_bytes, bl.length()); | |
7c673cae | 13919 | int r = bdev->aio_write(start, bl, &b->ioc, false); |
11fdf7f2 | 13920 | ceph_assert(r == 0); |
7c673cae FG |
13921 | } |
13922 | } | |
13923 | if (i == b->iomap.end()) { | |
13924 | break; | |
13925 | } | |
13926 | start = 0; | |
13927 | pos = i->first; | |
13928 | bl.clear(); | |
13929 | } | |
13930 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
13931 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
13932 | << dendl; | |
13933 | if (!bl.length()) { | |
13934 | start = pos; | |
13935 | } | |
13936 | pos += i->second.bl.length(); | |
13937 | bl.claim_append(i->second.bl); | |
13938 | ++i; | |
13939 | } | |
224ce89b | 13940 | |
7c673cae FG |
13941 | bdev->aio_submit(&b->ioc); |
13942 | } | |
13943 | ||
3efd9988 FG |
13944 | struct C_DeferredTrySubmit : public Context { |
13945 | BlueStore *store; | |
13946 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
13947 | void finish(int r) { | |
13948 | store->deferred_try_submit(); | |
13949 | } | |
13950 | }; | |
13951 | ||
7c673cae FG |
13952 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
13953 | { | |
13954 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 13955 | ceph_assert(osr->deferred_running); |
7c673cae FG |
13956 | DeferredBatch *b = osr->deferred_running; |
13957 | ||
13958 | { | |
f67539c2 | 13959 | osr->deferred_lock.lock(); |
11fdf7f2 | 13960 | ceph_assert(osr->deferred_running == b); |
7c673cae FG |
13961 | osr->deferred_running = nullptr; |
13962 | if (!osr->deferred_pending) { | |
181888fb | 13963 | dout(20) << __func__ << " dequeueing" << dendl; |
f67539c2 TL |
13964 | { |
13965 | deferred_lock.lock(); | |
13966 | auto q = deferred_queue.iterator_to(*osr); | |
13967 | deferred_queue.erase(q); | |
13968 | deferred_lock.unlock(); | |
13969 | } | |
13970 | osr->deferred_lock.unlock(); | |
181888fb | 13971 | } else { |
f67539c2 | 13972 | osr->deferred_lock.unlock(); |
9f95a23c TL |
13973 | if (deferred_aggressive) { |
13974 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; | |
13975 | finisher.queue(new C_DeferredTrySubmit(this)); | |
13976 | } else { | |
13977 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
13978 | } | |
7c673cae FG |
13979 | } |
13980 | } | |
13981 | ||
13982 | { | |
31f18b77 | 13983 | uint64_t costs = 0; |
11fdf7f2 | 13984 | { |
11fdf7f2 TL |
13985 | for (auto& i : b->txcs) { |
13986 | TransContext *txc = &i; | |
9f95a23c | 13987 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_aio_wait_lat); |
f67539c2 | 13988 | txc->set_state(TransContext::STATE_DEFERRED_CLEANUP); |
11fdf7f2 TL |
13989 | costs += txc->cost; |
13990 | } | |
7c673cae | 13991 | } |
9f95a23c | 13992 | throttle.release_deferred_throttle(costs); |
7c673cae FG |
13993 | } |
13994 | ||
9f95a23c | 13995 | { |
11fdf7f2 | 13996 | std::lock_guard l(kv_lock); |
9f95a23c TL |
13997 | deferred_done_queue.emplace_back(b); |
13998 | ||
13999 | // in the normal case, do not bother waking up the kv thread; it will | |
14000 | // catch us on the next commit anyway. | |
14001 | if (deferred_aggressive && !kv_sync_in_progress) { | |
14002 | kv_sync_in_progress = true; | |
14003 | kv_cond.notify_one(); | |
14004 | } | |
7c673cae FG |
14005 | } |
14006 | } | |
14007 | ||
14008 | int BlueStore::_deferred_replay() | |
14009 | { | |
14010 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
14011 | int count = 0; |
14012 | int r = 0; | |
2a845540 TL |
14013 | interval_set<uint64_t> bluefs_extents; |
14014 | if (bluefs) { | |
1e59de90 TL |
14015 | bluefs->foreach_block_extents( |
14016 | bluefs_layout.shared_bdev, | |
14017 | [&] (uint64_t start, uint32_t len) { | |
14018 | bluefs_extents.insert(start, len); | |
14019 | } | |
14020 | ); | |
2a845540 | 14021 | } |
11fdf7f2 TL |
14022 | CollectionRef ch = _get_collection(coll_t::meta()); |
14023 | bool fake_ch = false; | |
14024 | if (!ch) { | |
14025 | // hmm, replaying initial mkfs? | |
14026 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
14027 | fake_ch = true; | |
14028 | } | |
14029 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
14030 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
14031 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
14032 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
14033 | << dendl; | |
14034 | bluestore_deferred_transaction_t *deferred_txn = | |
14035 | new bluestore_deferred_transaction_t; | |
14036 | bufferlist bl = it->value(); | |
11fdf7f2 | 14037 | auto p = bl.cbegin(); |
7c673cae | 14038 | try { |
11fdf7f2 | 14039 | decode(*deferred_txn, p); |
f67539c2 | 14040 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
14041 | derr << __func__ << " failed to decode deferred txn " |
14042 | << pretty_binary_string(it->key()) << dendl; | |
14043 | delete deferred_txn; | |
14044 | r = -EIO; | |
14045 | goto out; | |
14046 | } | |
2a845540 TL |
14047 | bool has_some = _eliminate_outdated_deferred(deferred_txn, bluefs_extents); |
14048 | if (has_some) { | |
14049 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); | |
14050 | txc->deferred_txn = deferred_txn; | |
14051 | txc->set_state(TransContext::STATE_KV_DONE); | |
14052 | _txc_state_proc(txc); | |
14053 | } else { | |
14054 | delete deferred_txn; | |
14055 | } | |
7c673cae FG |
14056 | } |
14057 | out: | |
14058 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 14059 | _osr_register_zombie(osr); |
7c673cae | 14060 | _osr_drain_all(); |
11fdf7f2 TL |
14061 | if (fake_ch) { |
14062 | new_coll_map.clear(); | |
14063 | } | |
7c673cae FG |
14064 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
14065 | return r; | |
14066 | } | |
14067 | ||
2a845540 TL |
14068 | bool BlueStore::_eliminate_outdated_deferred(bluestore_deferred_transaction_t* deferred_txn, |
14069 | interval_set<uint64_t>& bluefs_extents) | |
14070 | { | |
14071 | bool has_some = false; | |
14072 | dout(30) << __func__ << " bluefs_extents: " << std::hex << bluefs_extents << std::dec << dendl; | |
14073 | auto it = deferred_txn->ops.begin(); | |
14074 | while (it != deferred_txn->ops.end()) { | |
14075 | // We process a pair of _data_/_extents_ (here: it->data/it->extents) | |
14076 | // by eliminating _extents_ that belong to bluefs, removing relevant parts of _data_ | |
14077 | // example: | |
14078 | // +------------+---------------+---------------+---------------+ | |
14079 | // | data | aaaaaaaabbbbb | bbbbcccccdddd | ddddeeeeeefff | | |
14080 | // | extent | 40000 - 44000 | 50000 - 58000 | 58000 - 60000 | | |
14081 | // | in bluefs? | no | yes | no | | |
14082 | // +------------+---------------+---------------+---------------+ | |
14083 | // result: | |
14084 | // +------------+---------------+---------------+ | |
14085 | // | data | aaaaaaaabbbbb | ddddeeeeeefff | | |
14086 | // | extent | 40000 - 44000 | 58000 - 60000 | | |
14087 | // +------------+---------------+---------------+ | |
14088 | PExtentVector new_extents; | |
14089 | ceph::buffer::list new_data; | |
14090 | uint32_t data_offset = 0; // this tracks location of extent 'e' inside it->data | |
14091 | dout(30) << __func__ << " input extents: " << it->extents << dendl; | |
14092 | for (auto& e: it->extents) { | |
14093 | interval_set<uint64_t> region; | |
14094 | region.insert(e.offset, e.length); | |
14095 | ||
14096 | auto mi = bluefs_extents.lower_bound(e.offset); | |
14097 | if (mi != bluefs_extents.begin()) { | |
14098 | --mi; | |
14099 | if (mi.get_end() <= e.offset) { | |
14100 | ++mi; | |
14101 | } | |
14102 | } | |
14103 | while (mi != bluefs_extents.end() && mi.get_start() < e.offset + e.length) { | |
14104 | // The interval_set does not like (asserts) when we erase interval that does not exist. | |
14105 | // Hence we do we implement (region-mi) by ((region+mi)-mi). | |
14106 | region.union_insert(mi.get_start(), mi.get_len()); | |
14107 | region.erase(mi.get_start(), mi.get_len()); | |
14108 | ++mi; | |
14109 | } | |
14110 | // 'region' is now a subset of e, without parts used by bluefs | |
14111 | // we trim coresponding parts from it->data (actally constructing new_data / new_extents) | |
14112 | for (auto ki = region.begin(); ki != region.end(); ki++) { | |
14113 | ceph::buffer::list chunk; | |
14114 | // A chunk from it->data; data_offset is a an offset where 'e' was located; | |
14115 | // 'ki.get_start() - e.offset' is an offset of ki inside 'e'. | |
14116 | chunk.substr_of(it->data, data_offset + (ki.get_start() - e.offset), ki.get_len()); | |
14117 | new_data.claim_append(chunk); | |
14118 | new_extents.emplace_back(bluestore_pextent_t(ki.get_start(), ki.get_len())); | |
14119 | } | |
14120 | data_offset += e.length; | |
14121 | } | |
14122 | dout(30) << __func__ << " output extents: " << new_extents << dendl; | |
14123 | if (it->data.length() != new_data.length()) { | |
14124 | dout(10) << __func__ << " trimmed deferred extents: " << it->extents << "->" << new_extents << dendl; | |
14125 | } | |
14126 | if (new_extents.size() == 0) { | |
14127 | it = deferred_txn->ops.erase(it); | |
14128 | } else { | |
14129 | has_some = true; | |
14130 | std::swap(it->extents, new_extents); | |
14131 | std::swap(it->data, new_data); | |
14132 | ++it; | |
14133 | } | |
14134 | } | |
14135 | return has_some; | |
14136 | } | |
14137 | ||
7c673cae FG |
14138 | // --------------------------- |
14139 | // transactions | |
14140 | ||
14141 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
14142 | CollectionHandle& ch, |
14143 | vector<Transaction>& tls, | |
14144 | TrackedOpRef op, | |
14145 | ThreadPool::TPHandle *handle) | |
14146 | { | |
14147 | FUNCTRACE(cct); | |
14148 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 14149 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 14150 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae | 14151 | |
11fdf7f2 TL |
14152 | auto start = mono_clock::now(); |
14153 | ||
14154 | Collection *c = static_cast<Collection*>(ch.get()); | |
14155 | OpSequencer *osr = c->osr.get(); | |
14156 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae | 14157 | |
f67539c2 TL |
14158 | // With HM-SMR drives (and ZNS SSDs) we want the I/O allocation and I/O |
14159 | // submission to happen atomically because if I/O submission happens in a | |
14160 | // different order than I/O allocation, we end up issuing non-sequential | |
14161 | // writes to the drive. This is a temporary solution until ZONE APPEND | |
14162 | // support matures in the kernel. For more information please see: | |
14163 | // https://www.usenix.org/conference/vault20/presentation/bjorling | |
14164 | if (bdev->is_smr()) { | |
14165 | atomic_alloc_and_submit_lock.lock(); | |
14166 | } | |
20effc67 TL |
14167 | |
14168 | // prepare | |
14169 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, | |
14170 | &on_commit, op); | |
14171 | ||
7c673cae | 14172 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { |
7c673cae FG |
14173 | txc->bytes += (*p).get_num_bytes(); |
14174 | _txc_add_transaction(txc, &(*p)); | |
14175 | } | |
14176 | _txc_calc_cost(txc); | |
14177 | ||
14178 | _txc_write_nodes(txc, txc->t); | |
14179 | ||
14180 | // journal deferred items | |
14181 | if (txc->deferred_txn) { | |
14182 | txc->deferred_txn->seq = ++deferred_seq; | |
14183 | bufferlist bl; | |
11fdf7f2 | 14184 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
14185 | string key; |
14186 | get_deferred_key(txc->deferred_txn->seq, &key); | |
14187 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
14188 | } | |
14189 | ||
14190 | _txc_finalize_kv(txc, txc->t); | |
f67539c2 TL |
14191 | |
14192 | #ifdef WITH_BLKIN | |
14193 | if (txc->trace) { | |
14194 | txc->trace.event("txc encode finished"); | |
14195 | } | |
14196 | #endif | |
14197 | ||
7c673cae FG |
14198 | if (handle) |
14199 | handle->suspend_tp_timeout(); | |
14200 | ||
11fdf7f2 | 14201 | auto tstart = mono_clock::now(); |
9f95a23c TL |
14202 | |
14203 | if (!throttle.try_start_transaction( | |
14204 | *db, | |
14205 | *txc, | |
14206 | tstart)) { | |
7c673cae | 14207 | // ensure we do not block here because of deferred writes |
9f95a23c TL |
14208 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
14209 | << dendl; | |
14210 | ++deferred_aggressive; | |
14211 | deferred_try_submit(); | |
14212 | { | |
14213 | // wake up any previously finished deferred events | |
14214 | std::lock_guard l(kv_lock); | |
14215 | if (!kv_sync_in_progress) { | |
14216 | kv_sync_in_progress = true; | |
3efd9988 FG |
14217 | kv_cond.notify_one(); |
14218 | } | |
9f95a23c TL |
14219 | } |
14220 | throttle.finish_start_transaction(*db, *txc, tstart); | |
14221 | --deferred_aggressive; | |
7c673cae | 14222 | } |
11fdf7f2 | 14223 | auto tend = mono_clock::now(); |
7c673cae FG |
14224 | |
14225 | if (handle) | |
14226 | handle->reset_tp_timeout(); | |
14227 | ||
14228 | logger->inc(l_bluestore_txc); | |
14229 | ||
14230 | // execute (start) | |
14231 | _txc_state_proc(txc); | |
14232 | ||
f67539c2 TL |
14233 | if (bdev->is_smr()) { |
14234 | atomic_alloc_and_submit_lock.unlock(); | |
14235 | } | |
14236 | ||
11fdf7f2 TL |
14237 | // we're immediately readable (unlike FileStore) |
14238 | for (auto c : on_applied_sync) { | |
14239 | c->complete(0); | |
14240 | } | |
14241 | if (!on_applied.empty()) { | |
14242 | if (c->commit_queue) { | |
14243 | c->commit_queue->queue(on_applied); | |
14244 | } else { | |
14245 | finisher.queue(on_applied); | |
14246 | } | |
14247 | } | |
14248 | ||
f67539c2 TL |
14249 | #ifdef WITH_BLKIN |
14250 | if (txc->trace) { | |
14251 | txc->trace.event("txc applied"); | |
14252 | } | |
14253 | #endif | |
14254 | ||
494da23a TL |
14255 | log_latency("submit_transact", |
14256 | l_bluestore_submit_lat, | |
14257 | mono_clock::now() - start, | |
14258 | cct->_conf->bluestore_log_op_age); | |
14259 | log_latency("throttle_transact", | |
14260 | l_bluestore_throttle_lat, | |
14261 | tend - tstart, | |
14262 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
14263 | return 0; |
14264 | } | |
14265 | ||
14266 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
14267 | { | |
14268 | dout(10) << __func__ << " txc " << txc << dendl; | |
14269 | bdev->aio_submit(&txc->ioc); | |
14270 | } | |
14271 | ||
14272 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
14273 | { | |
14274 | Transaction::iterator i = t->begin(); | |
14275 | ||
81eedcae | 14276 | _dump_transaction<30>(cct, t); |
7c673cae FG |
14277 | |
14278 | vector<CollectionRef> cvec(i.colls.size()); | |
14279 | unsigned j = 0; | |
14280 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
14281 | ++p, ++j) { | |
14282 | cvec[j] = _get_collection(*p); | |
7c673cae | 14283 | } |
11fdf7f2 | 14284 | |
7c673cae FG |
14285 | vector<OnodeRef> ovec(i.objects.size()); |
14286 | ||
14287 | for (int pos = 0; i.have_op(); ++pos) { | |
14288 | Transaction::Op *op = i.decode_op(); | |
14289 | int r = 0; | |
14290 | ||
14291 | // no coll or obj | |
14292 | if (op->op == Transaction::OP_NOP) | |
14293 | continue; | |
14294 | ||
11fdf7f2 | 14295 | |
7c673cae FG |
14296 | // collection operations |
14297 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
14298 | |
14299 | // initialize osd_pool_id and do a smoke test that all collections belong | |
14300 | // to the same pool | |
14301 | spg_t pgid; | |
14302 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
14303 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
14304 | txc->osd_pool_id == pgid.pool()); | |
14305 | txc->osd_pool_id = pgid.pool(); | |
14306 | } | |
14307 | ||
7c673cae FG |
14308 | switch (op->op) { |
14309 | case Transaction::OP_RMCOLL: | |
14310 | { | |
14311 | const coll_t &cid = i.get_cid(op->cid); | |
14312 | r = _remove_collection(txc, cid, &c); | |
14313 | if (!r) | |
14314 | continue; | |
14315 | } | |
14316 | break; | |
14317 | ||
14318 | case Transaction::OP_MKCOLL: | |
14319 | { | |
11fdf7f2 | 14320 | ceph_assert(!c); |
7c673cae FG |
14321 | const coll_t &cid = i.get_cid(op->cid); |
14322 | r = _create_collection(txc, cid, op->split_bits, &c); | |
14323 | if (!r) | |
14324 | continue; | |
14325 | } | |
14326 | break; | |
14327 | ||
14328 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 14329 | ceph_abort_msg("deprecated"); |
7c673cae FG |
14330 | break; |
14331 | ||
14332 | case Transaction::OP_SPLIT_COLLECTION2: | |
14333 | { | |
14334 | uint32_t bits = op->split_bits; | |
14335 | uint32_t rem = op->split_rem; | |
14336 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
14337 | if (!r) | |
14338 | continue; | |
14339 | } | |
14340 | break; | |
14341 | ||
11fdf7f2 TL |
14342 | case Transaction::OP_MERGE_COLLECTION: |
14343 | { | |
14344 | uint32_t bits = op->split_bits; | |
14345 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
14346 | if (!r) | |
14347 | continue; | |
14348 | } | |
14349 | break; | |
14350 | ||
7c673cae FG |
14351 | case Transaction::OP_COLL_HINT: |
14352 | { | |
f67539c2 | 14353 | uint32_t type = op->hint; |
7c673cae FG |
14354 | bufferlist hint; |
14355 | i.decode_bl(hint); | |
11fdf7f2 | 14356 | auto hiter = hint.cbegin(); |
7c673cae FG |
14357 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
14358 | uint32_t pg_num; | |
14359 | uint64_t num_objs; | |
11fdf7f2 TL |
14360 | decode(pg_num, hiter); |
14361 | decode(num_objs, hiter); | |
7c673cae FG |
14362 | dout(10) << __func__ << " collection hint objects is a no-op, " |
14363 | << " pg_num " << pg_num << " num_objects " << num_objs | |
14364 | << dendl; | |
14365 | } else { | |
14366 | // Ignore the hint | |
14367 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
14368 | } | |
14369 | continue; | |
14370 | } | |
14371 | break; | |
14372 | ||
14373 | case Transaction::OP_COLL_SETATTR: | |
14374 | r = -EOPNOTSUPP; | |
14375 | break; | |
14376 | ||
14377 | case Transaction::OP_COLL_RMATTR: | |
14378 | r = -EOPNOTSUPP; | |
14379 | break; | |
14380 | ||
14381 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 14382 | ceph_abort_msg("not implemented"); |
7c673cae FG |
14383 | break; |
14384 | } | |
14385 | if (r < 0) { | |
14386 | derr << __func__ << " error " << cpp_strerror(r) | |
14387 | << " not handled on operation " << op->op | |
14388 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 14389 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 14390 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
14391 | } |
14392 | ||
14393 | // these operations implicity create the object | |
14394 | bool create = false; | |
14395 | if (op->op == Transaction::OP_TOUCH || | |
9f95a23c | 14396 | op->op == Transaction::OP_CREATE || |
7c673cae FG |
14397 | op->op == Transaction::OP_WRITE || |
14398 | op->op == Transaction::OP_ZERO) { | |
14399 | create = true; | |
14400 | } | |
14401 | ||
14402 | // object operations | |
9f95a23c | 14403 | std::unique_lock l(c->lock); |
7c673cae FG |
14404 | OnodeRef &o = ovec[op->oid]; |
14405 | if (!o) { | |
14406 | ghobject_t oid = i.get_oid(op->oid); | |
9f95a23c | 14407 | o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE); |
7c673cae FG |
14408 | } |
14409 | if (!create && (!o || !o->exists)) { | |
14410 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
14411 | << i.get_oid(op->oid) << dendl; | |
14412 | r = -ENOENT; | |
14413 | goto endop; | |
14414 | } | |
14415 | ||
14416 | switch (op->op) { | |
9f95a23c | 14417 | case Transaction::OP_CREATE: |
7c673cae FG |
14418 | case Transaction::OP_TOUCH: |
14419 | r = _touch(txc, c, o); | |
14420 | break; | |
14421 | ||
14422 | case Transaction::OP_WRITE: | |
14423 | { | |
14424 | uint64_t off = op->off; | |
14425 | uint64_t len = op->len; | |
14426 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
14427 | bufferlist bl; | |
14428 | i.decode_bl(bl); | |
14429 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
14430 | } | |
14431 | break; | |
14432 | ||
14433 | case Transaction::OP_ZERO: | |
14434 | { | |
14435 | uint64_t off = op->off; | |
14436 | uint64_t len = op->len; | |
14437 | r = _zero(txc, c, o, off, len); | |
14438 | } | |
14439 | break; | |
14440 | ||
14441 | case Transaction::OP_TRIMCACHE: | |
14442 | { | |
14443 | // deprecated, no-op | |
14444 | } | |
14445 | break; | |
14446 | ||
14447 | case Transaction::OP_TRUNCATE: | |
14448 | { | |
14449 | uint64_t off = op->off; | |
35e4c445 | 14450 | r = _truncate(txc, c, o, off); |
7c673cae FG |
14451 | } |
14452 | break; | |
14453 | ||
14454 | case Transaction::OP_REMOVE: | |
14455 | { | |
14456 | r = _remove(txc, c, o); | |
14457 | } | |
14458 | break; | |
14459 | ||
14460 | case Transaction::OP_SETATTR: | |
14461 | { | |
14462 | string name = i.decode_string(); | |
14463 | bufferptr bp; | |
14464 | i.decode_bp(bp); | |
14465 | r = _setattr(txc, c, o, name, bp); | |
14466 | } | |
14467 | break; | |
14468 | ||
14469 | case Transaction::OP_SETATTRS: | |
14470 | { | |
14471 | map<string, bufferptr> aset; | |
14472 | i.decode_attrset(aset); | |
14473 | r = _setattrs(txc, c, o, aset); | |
14474 | } | |
14475 | break; | |
14476 | ||
14477 | case Transaction::OP_RMATTR: | |
14478 | { | |
14479 | string name = i.decode_string(); | |
14480 | r = _rmattr(txc, c, o, name); | |
14481 | } | |
14482 | break; | |
14483 | ||
14484 | case Transaction::OP_RMATTRS: | |
14485 | { | |
14486 | r = _rmattrs(txc, c, o); | |
14487 | } | |
14488 | break; | |
14489 | ||
14490 | case Transaction::OP_CLONE: | |
14491 | { | |
14492 | OnodeRef& no = ovec[op->dest_oid]; | |
14493 | if (!no) { | |
14494 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
14495 | no = c->get_onode(noid, true); | |
14496 | } | |
14497 | r = _clone(txc, c, o, no); | |
14498 | } | |
14499 | break; | |
14500 | ||
14501 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 14502 | ceph_abort_msg("deprecated"); |
7c673cae FG |
14503 | break; |
14504 | ||
14505 | case Transaction::OP_CLONERANGE2: | |
14506 | { | |
14507 | OnodeRef& no = ovec[op->dest_oid]; | |
14508 | if (!no) { | |
14509 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
14510 | no = c->get_onode(noid, true); | |
14511 | } | |
14512 | uint64_t srcoff = op->off; | |
14513 | uint64_t len = op->len; | |
14514 | uint64_t dstoff = op->dest_off; | |
14515 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
14516 | } | |
14517 | break; | |
14518 | ||
14519 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 14520 | ceph_abort_msg("not implemented"); |
7c673cae FG |
14521 | break; |
14522 | ||
14523 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 14524 | ceph_abort_msg("not implemented"); |
7c673cae FG |
14525 | break; |
14526 | ||
14527 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 14528 | ceph_abort_msg("deprecated"); |
7c673cae FG |
14529 | break; |
14530 | ||
14531 | case Transaction::OP_COLL_MOVE_RENAME: | |
14532 | case Transaction::OP_TRY_RENAME: | |
14533 | { | |
11fdf7f2 | 14534 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
14535 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
14536 | OnodeRef& no = ovec[op->dest_oid]; | |
14537 | if (!no) { | |
14538 | no = c->get_onode(noid, false); | |
14539 | } | |
14540 | r = _rename(txc, c, o, no, noid); | |
14541 | } | |
14542 | break; | |
14543 | ||
14544 | case Transaction::OP_OMAP_CLEAR: | |
14545 | { | |
14546 | r = _omap_clear(txc, c, o); | |
14547 | } | |
14548 | break; | |
14549 | case Transaction::OP_OMAP_SETKEYS: | |
14550 | { | |
14551 | bufferlist aset_bl; | |
14552 | i.decode_attrset_bl(&aset_bl); | |
14553 | r = _omap_setkeys(txc, c, o, aset_bl); | |
14554 | } | |
14555 | break; | |
14556 | case Transaction::OP_OMAP_RMKEYS: | |
14557 | { | |
14558 | bufferlist keys_bl; | |
14559 | i.decode_keyset_bl(&keys_bl); | |
14560 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
14561 | } | |
14562 | break; | |
14563 | case Transaction::OP_OMAP_RMKEYRANGE: | |
14564 | { | |
14565 | string first, last; | |
14566 | first = i.decode_string(); | |
14567 | last = i.decode_string(); | |
14568 | r = _omap_rmkey_range(txc, c, o, first, last); | |
14569 | } | |
14570 | break; | |
14571 | case Transaction::OP_OMAP_SETHEADER: | |
14572 | { | |
14573 | bufferlist bl; | |
14574 | i.decode_bl(bl); | |
14575 | r = _omap_setheader(txc, c, o, bl); | |
14576 | } | |
14577 | break; | |
14578 | ||
14579 | case Transaction::OP_SETALLOCHINT: | |
14580 | { | |
14581 | r = _set_alloc_hint(txc, c, o, | |
14582 | op->expected_object_size, | |
14583 | op->expected_write_size, | |
f67539c2 | 14584 | op->hint); |
7c673cae FG |
14585 | } |
14586 | break; | |
14587 | ||
14588 | default: | |
11fdf7f2 | 14589 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
14590 | ceph_abort(); |
14591 | } | |
14592 | ||
14593 | endop: | |
14594 | if (r < 0) { | |
14595 | bool ok = false; | |
14596 | ||
14597 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
14598 | op->op == Transaction::OP_CLONE || | |
14599 | op->op == Transaction::OP_CLONERANGE2 || | |
14600 | op->op == Transaction::OP_COLL_ADD || | |
14601 | op->op == Transaction::OP_SETATTR || | |
14602 | op->op == Transaction::OP_SETATTRS || | |
14603 | op->op == Transaction::OP_RMATTR || | |
14604 | op->op == Transaction::OP_OMAP_SETKEYS || | |
14605 | op->op == Transaction::OP_OMAP_RMKEYS || | |
14606 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
14607 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
14608 | // -ENOENT is usually okay | |
14609 | ok = true; | |
14610 | if (r == -ENODATA) | |
14611 | ok = true; | |
14612 | ||
14613 | if (!ok) { | |
14614 | const char *msg = "unexpected error code"; | |
14615 | ||
14616 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
14617 | op->op == Transaction::OP_CLONE || | |
14618 | op->op == Transaction::OP_CLONERANGE2)) | |
14619 | msg = "ENOENT on clone suggests osd bug"; | |
14620 | ||
14621 | if (r == -ENOSPC) | |
14622 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
14623 | // by partially applying transactions. | |
14624 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
14625 | ||
14626 | if (r == -ENOTEMPTY) { | |
14627 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
14628 | } | |
14629 | ||
14630 | derr << __func__ << " error " << cpp_strerror(r) | |
14631 | << " not handled on operation " << op->op | |
14632 | << " (op " << pos << ", counting from 0)" | |
14633 | << dendl; | |
14634 | derr << msg << dendl; | |
81eedcae | 14635 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 14636 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
14637 | } |
14638 | } | |
14639 | } | |
14640 | } | |
14641 | ||
14642 | ||
14643 | ||
14644 | // ----------------- | |
14645 | // write operations | |
14646 | ||
14647 | int BlueStore::_touch(TransContext *txc, | |
14648 | CollectionRef& c, | |
39ae355f | 14649 | OnodeRef& o) |
7c673cae FG |
14650 | { |
14651 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14652 | int r = 0; | |
7c673cae FG |
14653 | _assign_nid(txc, o); |
14654 | txc->write_onode(o); | |
14655 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14656 | return r; | |
14657 | } | |
14658 | ||
7c673cae FG |
14659 | void BlueStore::_pad_zeros( |
14660 | bufferlist *bl, uint64_t *offset, | |
14661 | uint64_t chunk_size) | |
14662 | { | |
14663 | auto length = bl->length(); | |
14664 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
14665 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
14666 | dout(40) << "before:\n"; | |
14667 | bl->hexdump(*_dout); | |
14668 | *_dout << dendl; | |
14669 | // front | |
14670 | size_t front_pad = *offset % chunk_size; | |
14671 | size_t back_pad = 0; | |
14672 | size_t pad_count = 0; | |
14673 | if (front_pad) { | |
11fdf7f2 | 14674 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
f67539c2 | 14675 | bufferptr z = ceph::buffer::create_small_page_aligned(chunk_size); |
224ce89b | 14676 | z.zero(0, front_pad, false); |
7c673cae | 14677 | pad_count += front_pad; |
9f95a23c | 14678 | bl->begin().copy(front_copy, z.c_str() + front_pad); |
7c673cae FG |
14679 | if (front_copy + front_pad < chunk_size) { |
14680 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 14681 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
14682 | pad_count += back_pad; |
14683 | } | |
14684 | bufferlist old, t; | |
14685 | old.swap(*bl); | |
14686 | t.substr_of(old, front_copy, length - front_copy); | |
14687 | bl->append(z); | |
14688 | bl->claim_append(t); | |
14689 | *offset -= front_pad; | |
224ce89b | 14690 | length += pad_count; |
7c673cae FG |
14691 | } |
14692 | ||
14693 | // back | |
14694 | uint64_t end = *offset + length; | |
14695 | unsigned back_copy = end % chunk_size; | |
14696 | if (back_copy) { | |
11fdf7f2 | 14697 | ceph_assert(back_pad == 0); |
7c673cae | 14698 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 14699 | ceph_assert(back_copy <= length); |
7c673cae | 14700 | bufferptr tail(chunk_size); |
9f95a23c | 14701 | bl->begin(length - back_copy).copy(back_copy, tail.c_str()); |
224ce89b | 14702 | tail.zero(back_copy, back_pad, false); |
7c673cae FG |
14703 | bufferlist old; |
14704 | old.swap(*bl); | |
14705 | bl->substr_of(old, 0, length - back_copy); | |
14706 | bl->append(tail); | |
14707 | length += back_pad; | |
14708 | pad_count += back_pad; | |
14709 | } | |
14710 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
14711 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
14712 | << length << std::dec << dendl; | |
14713 | dout(40) << "after:\n"; | |
14714 | bl->hexdump(*_dout); | |
14715 | *_dout << dendl; | |
14716 | if (pad_count) | |
14717 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 14718 | ceph_assert(bl->length() == length); |
7c673cae FG |
14719 | } |
14720 | ||
14721 | void BlueStore::_do_write_small( | |
14722 | TransContext *txc, | |
14723 | CollectionRef &c, | |
39ae355f | 14724 | OnodeRef& o, |
7c673cae FG |
14725 | uint64_t offset, uint64_t length, |
14726 | bufferlist::iterator& blp, | |
14727 | WriteContext *wctx) | |
14728 | { | |
14729 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
14730 | << std::dec << dendl; | |
11fdf7f2 | 14731 | ceph_assert(length < min_alloc_size); |
f67539c2 | 14732 | |
7c673cae FG |
14733 | uint64_t end_offs = offset + length; |
14734 | ||
14735 | logger->inc(l_bluestore_write_small); | |
14736 | logger->inc(l_bluestore_write_small_bytes, length); | |
14737 | ||
14738 | bufferlist bl; | |
14739 | blp.copy(length, bl); | |
14740 | ||
81eedcae TL |
14741 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
14742 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
14743 | uint32_t alloc_len = min_alloc_size; | |
14744 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
14745 | ||
14746 | bool any_change; | |
14747 | ||
14748 | // search suitable extent in both forward and reverse direction in | |
14749 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
14750 | // then check if blob can be reused via can_reuse_blob func or apply | |
14751 | // direct/deferred write (the latter for extents including or higher | |
14752 | // than 'offset' only). | |
14753 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
14754 | ||
20effc67 | 14755 | #ifdef HAVE_LIBZBD |
f67539c2 TL |
14756 | // On zoned devices, the first goal is to support non-overwrite workloads, |
14757 | // such as RGW, with large, aligned objects. Therefore, for user writes | |
14758 | // _do_write_small should not trigger. OSDs, however, write and update a tiny | |
14759 | // amount of metadata, such as OSD maps, to disk. For those cases, we | |
14760 | // temporarily just pad them to min_alloc_size and write them to a new place | |
14761 | // on every update. | |
14762 | if (bdev->is_smr()) { | |
f67539c2 TL |
14763 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
14764 | uint64_t b_off0 = b_off; | |
f67539c2 | 14765 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); |
20effc67 TL |
14766 | |
14767 | // Zero detection -- small block | |
33c7a0ef | 14768 | if (!cct->_conf->bluestore_zero_block_detection || !bl.is_zero()) { |
20effc67 TL |
14769 | BlobRef b = c->new_blob(); |
14770 | _pad_zeros(&bl, &b_off0, min_alloc_size); | |
14771 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, false, true); | |
14772 | } else { // if (bl.is_zero()) | |
14773 | dout(20) << __func__ << " skip small zero block " << std::hex | |
14774 | << " (0x" << b_off0 << "~" << bl.length() << ")" | |
14775 | << " (0x" << b_off << "~" << length << ")" | |
14776 | << std::dec << dendl; | |
14777 | logger->inc(l_bluestore_write_small_skipped); | |
14778 | logger->inc(l_bluestore_write_small_skipped_bytes, length); | |
14779 | } | |
14780 | ||
f67539c2 TL |
14781 | return; |
14782 | } | |
20effc67 | 14783 | #endif |
f67539c2 | 14784 | |
7c673cae FG |
14785 | // Look for an existing mutable blob we can use. |
14786 | auto begin = o->extent_map.extent_map.begin(); | |
14787 | auto end = o->extent_map.extent_map.end(); | |
14788 | auto ep = o->extent_map.seek_lextent(offset); | |
14789 | if (ep != begin) { | |
14790 | --ep; | |
14791 | if (ep->blob_end() <= offset) { | |
14792 | ++ep; | |
14793 | } | |
14794 | } | |
f67539c2 TL |
14795 | auto prev_ep = end; |
14796 | if (ep != begin) { | |
14797 | prev_ep = ep; | |
7c673cae | 14798 | --prev_ep; |
7c673cae FG |
14799 | } |
14800 | ||
eafe8130 TL |
14801 | boost::container::flat_set<const bluestore_blob_t*> inspected_blobs; |
14802 | // We don't want to have more blobs than min alloc units fit | |
14803 | // into 2 max blobs | |
14804 | size_t blob_threshold = max_blob_size / min_alloc_size * 2 + 1; | |
14805 | bool above_blob_threshold = false; | |
14806 | ||
14807 | inspected_blobs.reserve(blob_threshold); | |
14808 | ||
14809 | uint64_t max_off = 0; | |
14810 | auto start_ep = ep; | |
14811 | auto end_ep = ep; // exclusively | |
7c673cae FG |
14812 | do { |
14813 | any_change = false; | |
14814 | ||
14815 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
14816 | BlobRef b = ep->blob; | |
eafe8130 TL |
14817 | if (!above_blob_threshold) { |
14818 | inspected_blobs.insert(&b->get_blob()); | |
14819 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
14820 | } | |
14821 | max_off = ep->logical_end(); | |
7c673cae | 14822 | auto bstart = ep->blob_start(); |
eafe8130 | 14823 | |
7c673cae FG |
14824 | dout(20) << __func__ << " considering " << *b |
14825 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
14826 | if (bstart >= end_offs) { | |
14827 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
14828 | } else if (!b->get_blob().is_mutable()) { | |
14829 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
14830 | } else if (ep->logical_offset % min_alloc_size != | |
14831 | ep->blob_offset % min_alloc_size) { | |
14832 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
14833 | } else { | |
14834 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
14835 | // can we pad our head/tail out with zeros? | |
14836 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
14837 | head_pad = p2phase(offset, chunk_size); |
14838 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
14839 | if (head_pad || tail_pad) { |
14840 | o->extent_map.fault_range(db, offset - head_pad, | |
14841 | end_offs - offset + head_pad + tail_pad); | |
14842 | } | |
14843 | if (head_pad && | |
a4b75251 | 14844 | o->extent_map.has_any_lextents(offset - head_pad, head_pad)) { |
7c673cae FG |
14845 | head_pad = 0; |
14846 | } | |
14847 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
14848 | tail_pad = 0; | |
14849 | } | |
14850 | ||
14851 | uint64_t b_off = offset - head_pad - bstart; | |
14852 | uint64_t b_len = length + head_pad + tail_pad; | |
14853 | ||
14854 | // direct write into unused blocks of an existing mutable blob? | |
14855 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
14856 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
14857 | b->get_blob().is_unused(b_off, b_len) && | |
14858 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 14859 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
14860 | |
14861 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
14862 | << b_off << "~" << b_len | |
14863 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
14864 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 14865 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
14866 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
14867 | ||
11fdf7f2 | 14868 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
522d829b | 14869 | if (b_len < prefer_deferred_size) { |
7c673cae FG |
14870 | dout(20) << __func__ << " deferring small 0x" << std::hex |
14871 | << b_len << std::dec << " unused write via deferred" << dendl; | |
522d829b | 14872 | bluestore_deferred_op_t *op = _get_deferred_op(txc, bl.length()); |
7c673cae FG |
14873 | op->op = bluestore_deferred_op_t::OP_WRITE; |
14874 | b->get_blob().map( | |
14875 | b_off, b_len, | |
14876 | [&](uint64_t offset, uint64_t length) { | |
14877 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
14878 | return 0; | |
14879 | }); | |
224ce89b | 14880 | op->data = bl; |
7c673cae FG |
14881 | } else { |
14882 | b->get_blob().map_bl( | |
224ce89b | 14883 | b_off, bl, |
7c673cae FG |
14884 | [&](uint64_t offset, bufferlist& t) { |
14885 | bdev->aio_write(offset, t, | |
14886 | &txc->ioc, wctx->buffered); | |
14887 | }); | |
14888 | } | |
14889 | } | |
224ce89b | 14890 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
14891 | dout(20) << __func__ << " lex old " << *ep << dendl; |
14892 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
14893 | b, | |
14894 | &wctx->old_extents); | |
14895 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
f67539c2 | 14896 | |
7c673cae FG |
14897 | txc->statfs_delta.stored() += le->length; |
14898 | dout(20) << __func__ << " lex " << *le << dendl; | |
14899 | logger->inc(l_bluestore_write_small_unused); | |
14900 | return; | |
14901 | } | |
14902 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
14903 | uint64_t head_read = p2phase(b_off, chunk_size); |
14904 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
14905 | if ((head_read || tail_read) && |
14906 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
14907 | head_read + tail_read < min_alloc_size) { | |
14908 | b_off -= head_read; | |
14909 | b_len += head_read + tail_read; | |
14910 | ||
14911 | } else { | |
14912 | head_read = tail_read = 0; | |
14913 | } | |
14914 | ||
14915 | // chunk-aligned deferred overwrite? | |
14916 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
14917 | b_off % chunk_size == 0 && | |
14918 | b_len % chunk_size == 0 && | |
14919 | b->get_blob().is_allocated(b_off, b_len)) { | |
14920 | ||
224ce89b | 14921 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
14922 | |
14923 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
14924 | << " and tail 0x" << tail_read << std::dec << dendl; | |
14925 | if (head_read) { | |
14926 | bufferlist head_bl; | |
14927 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
14928 | head_bl, 0); | |
11fdf7f2 | 14929 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
14930 | size_t zlen = head_read - r; |
14931 | if (zlen) { | |
14932 | head_bl.append_zero(zlen); | |
14933 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
14934 | } | |
11fdf7f2 TL |
14935 | head_bl.claim_append(bl); |
14936 | bl.swap(head_bl); | |
7c673cae FG |
14937 | logger->inc(l_bluestore_write_penalty_read_ops); |
14938 | } | |
14939 | if (tail_read) { | |
14940 | bufferlist tail_bl; | |
14941 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
14942 | tail_bl, 0); | |
11fdf7f2 | 14943 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
14944 | size_t zlen = tail_read - r; |
14945 | if (zlen) { | |
14946 | tail_bl.append_zero(zlen); | |
14947 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
14948 | } | |
224ce89b | 14949 | bl.claim_append(tail_bl); |
7c673cae FG |
14950 | logger->inc(l_bluestore_write_penalty_read_ops); |
14951 | } | |
f67539c2 | 14952 | logger->inc(l_bluestore_write_small_pre_read); |
7c673cae | 14953 | |
224ce89b | 14954 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
14955 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
14956 | ||
f67539c2 | 14957 | b->dirty_blob().calc_csum(b_off, bl); |
11fdf7f2 TL |
14958 | |
14959 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
522d829b | 14960 | bluestore_deferred_op_t *op = _get_deferred_op(txc, bl.length()); |
11fdf7f2 TL |
14961 | op->op = bluestore_deferred_op_t::OP_WRITE; |
14962 | int r = b->get_blob().map( | |
14963 | b_off, b_len, | |
14964 | [&](uint64_t offset, uint64_t length) { | |
14965 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
14966 | return 0; | |
14967 | }); | |
14968 | ceph_assert(r == 0); | |
f67539c2 | 14969 | op->data = std::move(bl); |
11fdf7f2 TL |
14970 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" |
14971 | << b_len << std::dec << " of mutable " << *b | |
14972 | << " at " << op->extents << dendl; | |
14973 | } | |
14974 | ||
7c673cae FG |
14975 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
14976 | b, &wctx->old_extents); | |
14977 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
14978 | txc->statfs_delta.stored() += le->length; | |
14979 | dout(20) << __func__ << " lex " << *le << dendl; | |
7c673cae FG |
14980 | return; |
14981 | } | |
224ce89b WB |
14982 | // try to reuse blob if we can |
14983 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
14984 | max_bsize, |
14985 | offset0 - bstart, | |
14986 | &alloc_len)) { | |
11fdf7f2 | 14987 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
14988 | // fit into reused blob |
14989 | // Need to check for pending writes desiring to | |
14990 | // reuse the same pextent. The rationale is that during GC two chunks | |
14991 | // from garbage blobs(compressed?) can share logical space within the same | |
14992 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
14993 | // Hence the second write will fail in an attempt to reuse blob at | |
14994 | // do_alloc_write(). | |
14995 | if (!wctx->has_conflict(b, | |
14996 | offset0, | |
14997 | offset0 + alloc_len, | |
14998 | min_alloc_size)) { | |
14999 | ||
15000 | // we can't reuse pad_head/pad_tail since they might be truncated | |
15001 | // due to existent extents | |
15002 | uint64_t b_off = offset - bstart; | |
15003 | uint64_t b_off0 = b_off; | |
20effc67 | 15004 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); |
7c673cae | 15005 | |
20effc67 | 15006 | // Zero detection -- small block |
33c7a0ef | 15007 | if (!cct->_conf->bluestore_zero_block_detection || !bl.is_zero()) { |
20effc67 TL |
15008 | _pad_zeros(&bl, &b_off0, chunk_size); |
15009 | ||
15010 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
15011 | << " (0x" << b_off0 << "~" << bl.length() << ")" | |
15012 | << " (0x" << b_off << "~" << length << ")" | |
15013 | << std::dec << dendl; | |
15014 | ||
15015 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
15016 | false, false); | |
15017 | logger->inc(l_bluestore_write_small_unused); | |
15018 | } else { // if (bl.is_zero()) | |
15019 | dout(20) << __func__ << " skip small zero block " << std::hex | |
15020 | << " (0x" << b_off0 << "~" << bl.length() << ")" | |
15021 | << " (0x" << b_off << "~" << length << ")" | |
15022 | << std::dec << dendl; | |
15023 | logger->inc(l_bluestore_write_small_skipped); | |
15024 | logger->inc(l_bluestore_write_small_skipped_bytes, length); | |
15025 | } | |
7c673cae | 15026 | |
7c673cae FG |
15027 | return; |
15028 | } | |
15029 | } | |
15030 | } | |
15031 | ++ep; | |
eafe8130 | 15032 | end_ep = ep; |
7c673cae FG |
15033 | any_change = true; |
15034 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
15035 | ||
15036 | // check extent for reuse in reverse order | |
15037 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
15038 | BlobRef b = prev_ep->blob; | |
eafe8130 TL |
15039 | if (!above_blob_threshold) { |
15040 | inspected_blobs.insert(&b->get_blob()); | |
15041 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
15042 | } | |
15043 | start_ep = prev_ep; | |
7c673cae FG |
15044 | auto bstart = prev_ep->blob_start(); |
15045 | dout(20) << __func__ << " considering " << *b | |
15046 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 15047 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
15048 | max_bsize, |
15049 | offset0 - bstart, | |
15050 | &alloc_len)) { | |
11fdf7f2 | 15051 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
15052 | // fit into reused blob |
15053 | // Need to check for pending writes desiring to | |
15054 | // reuse the same pextent. The rationale is that during GC two chunks | |
15055 | // from garbage blobs(compressed?) can share logical space within the same | |
15056 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
15057 | // Hence the second write will fail in an attempt to reuse blob at | |
15058 | // do_alloc_write(). | |
15059 | if (!wctx->has_conflict(b, | |
15060 | offset0, | |
15061 | offset0 + alloc_len, | |
15062 | min_alloc_size)) { | |
15063 | ||
7c673cae FG |
15064 | uint64_t b_off = offset - bstart; |
15065 | uint64_t b_off0 = b_off; | |
20effc67 | 15066 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); |
7c673cae | 15067 | |
20effc67 | 15068 | // Zero detection -- small block |
33c7a0ef | 15069 | if (!cct->_conf->bluestore_zero_block_detection || !bl.is_zero()) { |
20effc67 TL |
15070 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); |
15071 | _pad_zeros(&bl, &b_off0, chunk_size); | |
15072 | ||
15073 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
15074 | << " (0x" << b_off0 << "~" << bl.length() << ")" | |
15075 | << " (0x" << b_off << "~" << length << ")" | |
15076 | << std::dec << dendl; | |
15077 | ||
15078 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
15079 | false, false); | |
15080 | logger->inc(l_bluestore_write_small_unused); | |
15081 | } else { // if (bl.is_zero()) | |
15082 | dout(20) << __func__ << " skip small zero block " << std::hex | |
15083 | << " (0x" << b_off0 << "~" << bl.length() << ")" | |
15084 | << " (0x" << b_off << "~" << length << ")" | |
15085 | << std::dec << dendl; | |
15086 | logger->inc(l_bluestore_write_small_skipped); | |
15087 | logger->inc(l_bluestore_write_small_skipped_bytes, length); | |
15088 | } | |
7c673cae | 15089 | |
7c673cae FG |
15090 | return; |
15091 | } | |
15092 | } | |
15093 | if (prev_ep != begin) { | |
15094 | --prev_ep; | |
15095 | any_change = true; | |
15096 | } else { | |
15097 | prev_ep = end; // to avoid useless first extent re-check | |
15098 | } | |
15099 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
15100 | } while (any_change); | |
15101 | ||
eafe8130 TL |
15102 | if (above_blob_threshold) { |
15103 | dout(10) << __func__ << " request GC, blobs >= " << inspected_blobs.size() | |
15104 | << " " << std::hex << min_off << "~" << max_off << std::dec | |
15105 | << dendl; | |
15106 | ceph_assert(start_ep != end_ep); | |
15107 | for (auto ep = start_ep; ep != end_ep; ++ep) { | |
15108 | dout(20) << __func__ << " inserting for GC " | |
15109 | << std::hex << ep->logical_offset << "~" << ep->length | |
15110 | << std::dec << dendl; | |
15111 | ||
15112 | wctx->extents_to_gc.union_insert(ep->logical_offset, ep->length); | |
15113 | } | |
15114 | // insert newly written extent to GC | |
15115 | wctx->extents_to_gc.union_insert(offset, length); | |
15116 | dout(20) << __func__ << " inserting (last) for GC " | |
15117 | << std::hex << offset << "~" << length | |
15118 | << std::dec << dendl; | |
15119 | } | |
11fdf7f2 | 15120 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae | 15121 | uint64_t b_off0 = b_off; |
7c673cae | 15122 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); |
20effc67 TL |
15123 | |
15124 | // Zero detection -- small block | |
33c7a0ef | 15125 | if (!cct->_conf->bluestore_zero_block_detection || !bl.is_zero()) { |
20effc67 TL |
15126 | // new blob. |
15127 | BlobRef b = c->new_blob(); | |
15128 | _pad_zeros(&bl, &b_off0, block_size); | |
15129 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
15130 | min_alloc_size != block_size, // use 'unused' bitmap when alloc granularity | |
15131 | // doesn't match disk one only | |
15132 | true); | |
15133 | } else { // if (bl.is_zero()) | |
15134 | dout(20) << __func__ << " skip small zero block " << std::hex | |
15135 | << " (0x" << b_off0 << "~" << bl.length() << ")" | |
15136 | << " (0x" << b_off << "~" << length << ")" | |
15137 | << std::dec << dendl; | |
15138 | logger->inc(l_bluestore_write_small_skipped); | |
15139 | logger->inc(l_bluestore_write_small_skipped_bytes, length); | |
15140 | } | |
7c673cae FG |
15141 | |
15142 | return; | |
15143 | } | |
15144 | ||
f67539c2 TL |
15145 | bool BlueStore::BigDeferredWriteContext::can_defer( |
15146 | BlueStore::extent_map_t::iterator ep, | |
15147 | uint64_t prefer_deferred_size, | |
15148 | uint64_t block_size, | |
15149 | uint64_t offset, | |
15150 | uint64_t l) | |
15151 | { | |
15152 | bool res = false; | |
15153 | auto& blob = ep->blob->get_blob(); | |
15154 | if (offset >= ep->blob_start() && | |
15155 | blob.is_mutable()) { | |
15156 | off = offset; | |
15157 | b_off = offset - ep->blob_start(); | |
15158 | uint64_t chunk_size = blob.get_chunk_size(block_size); | |
15159 | uint64_t ondisk = blob.get_ondisk_length(); | |
15160 | used = std::min(l, ondisk - b_off); | |
15161 | ||
15162 | // will read some data to fill out the chunk? | |
15163 | head_read = p2phase<uint64_t>(b_off, chunk_size); | |
15164 | tail_read = p2nphase<uint64_t>(b_off + used, chunk_size); | |
15165 | b_off -= head_read; | |
15166 | ||
15167 | ceph_assert(b_off % chunk_size == 0); | |
15168 | ceph_assert(blob_aligned_len() % chunk_size == 0); | |
15169 | ||
522d829b | 15170 | res = blob_aligned_len() < prefer_deferred_size && |
f67539c2 TL |
15171 | blob_aligned_len() <= ondisk && |
15172 | blob.is_allocated(b_off, blob_aligned_len()); | |
15173 | if (res) { | |
15174 | blob_ref = ep->blob; | |
15175 | blob_start = ep->blob_start(); | |
15176 | } | |
15177 | } | |
15178 | return res; | |
15179 | } | |
15180 | ||
15181 | bool BlueStore::BigDeferredWriteContext::apply_defer() | |
15182 | { | |
15183 | int r = blob_ref->get_blob().map( | |
15184 | b_off, blob_aligned_len(), | |
15185 | [&](const bluestore_pextent_t& pext, | |
15186 | uint64_t offset, | |
15187 | uint64_t length) { | |
15188 | // apply deferred if overwrite breaks blob continuity only. | |
15189 | // if it totally overlaps some pextent - fallback to regular write | |
15190 | if (pext.offset < offset || | |
15191 | pext.end() > offset + length) { | |
15192 | res_extents.emplace_back(bluestore_pextent_t(offset, length)); | |
15193 | return 0; | |
15194 | } | |
15195 | return -1; | |
15196 | }); | |
15197 | return r >= 0; | |
15198 | } | |
15199 | ||
15200 | void BlueStore::_do_write_big_apply_deferred( | |
15201 | TransContext* txc, | |
15202 | CollectionRef& c, | |
39ae355f | 15203 | OnodeRef& o, |
f67539c2 TL |
15204 | BlueStore::BigDeferredWriteContext& dctx, |
15205 | bufferlist::iterator& blp, | |
15206 | WriteContext* wctx) | |
15207 | { | |
15208 | bufferlist bl; | |
15209 | dout(20) << __func__ << " reading head 0x" << std::hex << dctx.head_read | |
15210 | << " and tail 0x" << dctx.tail_read << std::dec << dendl; | |
15211 | if (dctx.head_read) { | |
15212 | int r = _do_read(c.get(), o, | |
15213 | dctx.off - dctx.head_read, | |
15214 | dctx.head_read, | |
15215 | bl, | |
15216 | 0); | |
15217 | ceph_assert(r >= 0 && r <= (int)dctx.head_read); | |
15218 | size_t zlen = dctx.head_read - r; | |
15219 | if (zlen) { | |
15220 | bl.append_zero(zlen); | |
15221 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
15222 | } | |
15223 | logger->inc(l_bluestore_write_penalty_read_ops); | |
15224 | } | |
15225 | blp.copy(dctx.used, bl); | |
15226 | ||
15227 | if (dctx.tail_read) { | |
15228 | bufferlist tail_bl; | |
15229 | int r = _do_read(c.get(), o, | |
15230 | dctx.off + dctx.used, dctx.tail_read, | |
15231 | tail_bl, 0); | |
15232 | ceph_assert(r >= 0 && r <= (int)dctx.tail_read); | |
15233 | size_t zlen = dctx.tail_read - r; | |
15234 | if (zlen) { | |
15235 | tail_bl.append_zero(zlen); | |
15236 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
15237 | } | |
15238 | bl.claim_append(tail_bl); | |
15239 | logger->inc(l_bluestore_write_penalty_read_ops); | |
15240 | } | |
15241 | auto& b0 = dctx.blob_ref; | |
15242 | _buffer_cache_write(txc, b0, dctx.b_off, bl, | |
15243 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
15244 | ||
15245 | b0->dirty_blob().calc_csum(dctx.b_off, bl); | |
15246 | ||
15247 | Extent* le = o->extent_map.set_lextent(c, dctx.off, | |
15248 | dctx.off - dctx.blob_start, dctx.used, b0, &wctx->old_extents); | |
15249 | ||
15250 | // in fact this is a no-op for big writes but left here to maintain | |
15251 | // uniformity and avoid missing after some refactor. | |
15252 | b0->dirty_blob().mark_used(le->blob_offset, le->length); | |
15253 | txc->statfs_delta.stored() += le->length; | |
15254 | ||
15255 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
522d829b | 15256 | bluestore_deferred_op_t* op = _get_deferred_op(txc, bl.length()); |
f67539c2 TL |
15257 | op->op = bluestore_deferred_op_t::OP_WRITE; |
15258 | op->extents.swap(dctx.res_extents); | |
15259 | op->data = std::move(bl); | |
15260 | } | |
15261 | } | |
15262 | ||
7c673cae FG |
15263 | void BlueStore::_do_write_big( |
15264 | TransContext *txc, | |
15265 | CollectionRef &c, | |
39ae355f | 15266 | OnodeRef& o, |
7c673cae FG |
15267 | uint64_t offset, uint64_t length, |
15268 | bufferlist::iterator& blp, | |
15269 | WriteContext *wctx) | |
15270 | { | |
15271 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
15272 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
15273 | << " compress " << (int)wctx->compress | |
15274 | << dendl; | |
15275 | logger->inc(l_bluestore_write_big); | |
15276 | logger->inc(l_bluestore_write_big_bytes, length); | |
11fdf7f2 | 15277 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
f67539c2 | 15278 | uint64_t prefer_deferred_size_snapshot = prefer_deferred_size.load(); |
7c673cae FG |
15279 | while (length > 0) { |
15280 | bool new_blob = false; | |
7c673cae FG |
15281 | BlobRef b; |
15282 | uint32_t b_off = 0; | |
522d829b | 15283 | uint32_t l = 0; |
7c673cae FG |
15284 | |
15285 | //attempting to reuse existing blob | |
15286 | if (!wctx->compress) { | |
522d829b TL |
15287 | // enforce target blob alignment with max_bsize |
15288 | l = max_bsize - p2phase(offset, max_bsize); | |
15289 | l = std::min(uint64_t(l), length); | |
15290 | ||
7c673cae | 15291 | auto end = o->extent_map.extent_map.end(); |
f67539c2 | 15292 | |
522d829b TL |
15293 | dout(20) << __func__ << " may be defer: 0x" << std::hex |
15294 | << offset << "~" << l | |
15295 | << std::dec << dendl; | |
15296 | ||
f67539c2 TL |
15297 | if (prefer_deferred_size_snapshot && |
15298 | l <= prefer_deferred_size_snapshot * 2) { | |
15299 | // Single write that spans two adjusted existing blobs can result | |
15300 | // in up to two deferred blocks of 'prefer_deferred_size' | |
15301 | // So we're trying to minimize the amount of resulting blobs | |
15302 | // and preserve 2 blobs rather than inserting one more in between | |
15303 | // E.g. write 0x10000~20000 over existing blobs | |
15304 | // (0x0~20000 and 0x20000~20000) is better (from subsequent reading | |
15305 | // performance point of view) to result in two deferred writes to | |
15306 | // existing blobs than having 3 blobs: 0x0~10000, 0x10000~20000, 0x30000~10000 | |
15307 | ||
15308 | // look for an existing mutable blob we can write into | |
15309 | auto ep = o->extent_map.seek_lextent(offset); | |
15310 | auto ep_next = end; | |
15311 | BigDeferredWriteContext head_info, tail_info; | |
15312 | ||
15313 | bool will_defer = ep != end ? | |
15314 | head_info.can_defer(ep, | |
15315 | prefer_deferred_size_snapshot, | |
15316 | block_size, | |
15317 | offset, | |
15318 | l) : | |
15319 | false; | |
15320 | auto offset_next = offset + head_info.used; | |
15321 | auto remaining = l - head_info.used; | |
15322 | if (will_defer && remaining) { | |
15323 | will_defer = false; | |
15324 | if (remaining <= prefer_deferred_size_snapshot) { | |
15325 | ep_next = o->extent_map.seek_lextent(offset_next); | |
15326 | // check if we can defer remaining totally | |
15327 | will_defer = ep_next == end ? | |
15328 | false : | |
15329 | tail_info.can_defer(ep_next, | |
15330 | prefer_deferred_size_snapshot, | |
15331 | block_size, | |
15332 | offset_next, | |
15333 | remaining); | |
15334 | will_defer = will_defer && remaining == tail_info.used; | |
15335 | } | |
15336 | } | |
15337 | if (will_defer) { | |
15338 | dout(20) << __func__ << " " << *(head_info.blob_ref) | |
15339 | << " deferring big " << std::hex | |
15340 | << " (0x" << head_info.b_off << "~" << head_info.blob_aligned_len() << ")" | |
15341 | << std::dec << " write via deferred" | |
15342 | << dendl; | |
15343 | if (remaining) { | |
15344 | dout(20) << __func__ << " " << *(tail_info.blob_ref) | |
15345 | << " deferring big " << std::hex | |
15346 | << " (0x" << tail_info.b_off << "~" << tail_info.blob_aligned_len() << ")" | |
15347 | << std::dec << " write via deferred" | |
15348 | << dendl; | |
15349 | } | |
15350 | ||
15351 | will_defer = head_info.apply_defer(); | |
15352 | if (!will_defer) { | |
15353 | dout(20) << __func__ | |
15354 | << " deferring big fell back, head isn't continuous" | |
15355 | << dendl; | |
15356 | } else if (remaining) { | |
15357 | will_defer = tail_info.apply_defer(); | |
15358 | if (!will_defer) { | |
15359 | dout(20) << __func__ | |
15360 | << " deferring big fell back, tail isn't continuous" | |
15361 | << dendl; | |
15362 | } | |
15363 | } | |
15364 | } | |
15365 | if (will_defer) { | |
15366 | _do_write_big_apply_deferred(txc, c, o, head_info, blp, wctx); | |
15367 | if (remaining) { | |
15368 | _do_write_big_apply_deferred(txc, c, o, tail_info, | |
15369 | blp, wctx); | |
15370 | } | |
522d829b TL |
15371 | dout(20) << __func__ << " defer big: 0x" << std::hex |
15372 | << offset << "~" << l | |
15373 | << std::dec << dendl; | |
f67539c2 TL |
15374 | offset += l; |
15375 | length -= l; | |
15376 | logger->inc(l_bluestore_write_big_blobs, remaining ? 2 : 1); | |
15377 | logger->inc(l_bluestore_write_big_deferred, remaining ? 2 : 1); | |
15378 | continue; | |
15379 | } | |
15380 | } | |
522d829b | 15381 | dout(20) << __func__ << " lookup for blocks to reuse..." << dendl; |
f67539c2 TL |
15382 | |
15383 | o->extent_map.punch_hole(c, offset, l, &wctx->old_extents); | |
15384 | ||
15385 | // seek again as punch_hole could invalidate ep | |
7c673cae | 15386 | auto ep = o->extent_map.seek_lextent(offset); |
f67539c2 TL |
15387 | auto begin = o->extent_map.extent_map.begin(); |
15388 | auto prev_ep = end; | |
15389 | if (ep != begin) { | |
15390 | prev_ep = ep; | |
7c673cae | 15391 | --prev_ep; |
7c673cae | 15392 | } |
f67539c2 | 15393 | |
7c673cae FG |
15394 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; |
15395 | // search suitable extent in both forward and reverse direction in | |
15396 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 15397 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
15398 | bool any_change; |
15399 | do { | |
15400 | any_change = false; | |
15401 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
522d829b TL |
15402 | dout(20) << __func__ << " considering " << *ep |
15403 | << " bstart 0x" << std::hex << ep->blob_start() << std::dec << dendl; | |
f67539c2 TL |
15404 | |
15405 | if (offset >= ep->blob_start() && | |
224ce89b | 15406 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
15407 | offset - ep->blob_start(), |
15408 | &l)) { | |
15409 | b = ep->blob; | |
f67539c2 | 15410 | b_off = offset - ep->blob_start(); |
7c673cae FG |
15411 | prev_ep = end; // to avoid check below |
15412 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 15413 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
15414 | } else { |
15415 | ++ep; | |
15416 | any_change = true; | |
15417 | } | |
15418 | } | |
15419 | ||
15420 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
522d829b TL |
15421 | dout(20) << __func__ << " considering rev " << *prev_ep |
15422 | << " bstart 0x" << std::hex << prev_ep->blob_start() << std::dec << dendl; | |
f67539c2 | 15423 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
15424 | offset - prev_ep->blob_start(), |
15425 | &l)) { | |
15426 | b = prev_ep->blob; | |
15427 | b_off = offset - prev_ep->blob_start(); | |
15428 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 15429 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
15430 | } else if (prev_ep != begin) { |
15431 | --prev_ep; | |
15432 | any_change = true; | |
15433 | } else { | |
15434 | prev_ep = end; // to avoid useless first extent re-check | |
15435 | } | |
15436 | } | |
15437 | } while (b == nullptr && any_change); | |
f67539c2 | 15438 | } else { |
522d829b TL |
15439 | // trying to utilize as longer chunk as permitted in case of compression. |
15440 | l = std::min(max_bsize, length); | |
f67539c2 TL |
15441 | o->extent_map.punch_hole(c, offset, l, &wctx->old_extents); |
15442 | } // if (!wctx->compress) | |
15443 | ||
7c673cae FG |
15444 | if (b == nullptr) { |
15445 | b = c->new_blob(); | |
15446 | b_off = 0; | |
15447 | new_blob = true; | |
15448 | } | |
7c673cae FG |
15449 | bufferlist t; |
15450 | blp.copy(l, t); | |
20effc67 TL |
15451 | |
15452 | // Zero detection -- big block | |
33c7a0ef | 15453 | if (!cct->_conf->bluestore_zero_block_detection || !t.is_zero()) { |
20effc67 TL |
15454 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); |
15455 | ||
15456 | dout(20) << __func__ << " schedule write big: 0x" | |
522d829b TL |
15457 | << std::hex << offset << "~" << l << std::dec |
15458 | << (new_blob ? " new " : " reuse ") | |
15459 | << *b << dendl; | |
20effc67 TL |
15460 | |
15461 | logger->inc(l_bluestore_write_big_blobs); | |
15462 | } else { // if (!t.is_zero()) | |
15463 | dout(20) << __func__ << " skip big zero block " << std::hex | |
15464 | << " (0x" << b_off << "~" << t.length() << ")" | |
15465 | << " (0x" << b_off << "~" << l << ")" | |
15466 | << std::dec << dendl; | |
15467 | logger->inc(l_bluestore_write_big_skipped_blobs); | |
15468 | logger->inc(l_bluestore_write_big_skipped_bytes, l); | |
15469 | } | |
15470 | ||
7c673cae FG |
15471 | offset += l; |
15472 | length -= l; | |
7c673cae FG |
15473 | } |
15474 | } | |
15475 | ||
15476 | int BlueStore::_do_alloc_write( | |
15477 | TransContext *txc, | |
15478 | CollectionRef coll, | |
39ae355f | 15479 | OnodeRef& o, |
7c673cae FG |
15480 | WriteContext *wctx) |
15481 | { | |
15482 | dout(20) << __func__ << " txc " << txc | |
15483 | << " " << wctx->writes.size() << " blobs" | |
15484 | << dendl; | |
3efd9988 FG |
15485 | if (wctx->writes.empty()) { |
15486 | return 0; | |
7c673cae FG |
15487 | } |
15488 | ||
7c673cae FG |
15489 | CompressorRef c; |
15490 | double crr = 0; | |
15491 | if (wctx->compress) { | |
15492 | c = select_option( | |
15493 | "compression_algorithm", | |
15494 | compressor, | |
15495 | [&]() { | |
15496 | string val; | |
15497 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
15498 | CompressorRef cp = compressor; | |
15499 | if (!cp || cp->get_type_name() != val) { | |
15500 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
15501 | if (!cp) { |
15502 | if (_set_compression_alert(false, val.c_str())) { | |
15503 | derr << __func__ << " unable to initialize " << val.c_str() | |
15504 | << " compressor" << dendl; | |
15505 | } | |
15506 | } | |
7c673cae | 15507 | } |
1e59de90 | 15508 | return std::optional<CompressorRef>(cp); |
7c673cae | 15509 | } |
1e59de90 | 15510 | return std::optional<CompressorRef>(); |
7c673cae FG |
15511 | } |
15512 | ); | |
15513 | ||
15514 | crr = select_option( | |
15515 | "compression_required_ratio", | |
15516 | cct->_conf->bluestore_compression_required_ratio, | |
15517 | [&]() { | |
15518 | double val; | |
3efd9988 | 15519 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
1e59de90 | 15520 | return std::optional<double>(val); |
7c673cae | 15521 | } |
1e59de90 | 15522 | return std::optional<double>(); |
7c673cae FG |
15523 | } |
15524 | ); | |
15525 | } | |
15526 | ||
15527 | // checksum | |
11fdf7f2 | 15528 | int64_t csum = csum_type.load(); |
7c673cae FG |
15529 | csum = select_option( |
15530 | "csum_type", | |
15531 | csum, | |
15532 | [&]() { | |
11fdf7f2 | 15533 | int64_t val; |
3efd9988 | 15534 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
1e59de90 | 15535 | return std::optional<int64_t>(val); |
7c673cae | 15536 | } |
1e59de90 | 15537 | return std::optional<int64_t>(); |
7c673cae FG |
15538 | } |
15539 | ); | |
15540 | ||
3efd9988 FG |
15541 | // compress (as needed) and calc needed space |
15542 | uint64_t need = 0; | |
39ae355f TL |
15543 | uint64_t data_size = 0; |
15544 | // 'need' is amount of space that must be provided by allocator. | |
15545 | // 'data_size' is a size of data that will be transferred to disk. | |
15546 | // Note that data_size is always <= need. This comes from: | |
15547 | // - write to blob was unaligned, and there is free space | |
15548 | // - data has been compressed | |
15549 | // | |
15550 | // We make one decision and apply it to all blobs. | |
15551 | // All blobs will be deferred or none will. | |
15552 | // We assume that allocator does its best to provide contiguous space, | |
15553 | // and the condition is : (data_size < deferred). | |
15554 | ||
11fdf7f2 | 15555 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 15556 | for (auto& wi : wctx->writes) { |
3efd9988 | 15557 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 15558 | auto start = mono_clock::now(); |
7c673cae FG |
15559 | |
15560 | // compress | |
11fdf7f2 TL |
15561 | ceph_assert(wi.b_off == 0); |
15562 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 15563 | |
7c673cae FG |
15564 | // FIXME: memory alignment here is bad |
15565 | bufferlist t; | |
1e59de90 | 15566 | std::optional<int32_t> compressor_message; |
f67539c2 | 15567 | int r = c->compress(wi.bl, t, compressor_message); |
3efd9988 | 15568 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 15569 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
15570 | bool rejected = false; |
15571 | uint64_t compressed_len = t.length(); | |
15572 | // do an approximate (fast) estimation for resulting blob size | |
15573 | // that doesn't take header overhead into account | |
11fdf7f2 | 15574 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
15575 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
15576 | bluestore_compression_header_t chdr; | |
15577 | chdr.type = c->get_type(); | |
15578 | chdr.length = t.length(); | |
f67539c2 | 15579 | chdr.compressor_message = compressor_message; |
a8e16298 TL |
15580 | encode(chdr, wi.compressed_bl); |
15581 | wi.compressed_bl.claim_append(t); | |
15582 | ||
15583 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 15584 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
15585 | if (result_len <= want_len && result_len < wi.blob_length) { |
15586 | // Cool. We compressed at least as much as we were hoping to. | |
15587 | // pad out to min_alloc_size | |
15588 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
15589 | wi.compressed_len = compressed_len; | |
15590 | wi.compressed = true; | |
15591 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
15592 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
15593 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
15594 | << " with " << c->get_type() | |
15595 | << std::dec << dendl; | |
15596 | txc->statfs_delta.compressed() += compressed_len; | |
15597 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
15598 | txc->statfs_delta.compressed_allocated() += result_len; | |
15599 | logger->inc(l_bluestore_compress_success_count); | |
15600 | need += result_len; | |
39ae355f | 15601 | data_size += result_len; |
a8e16298 TL |
15602 | } else { |
15603 | rejected = true; | |
15604 | } | |
15605 | } else if (r != 0) { | |
15606 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
15607 | << " bytes compressed using " << c->get_type_name() | |
15608 | << std::dec | |
15609 | << " failed with errcode = " << r | |
15610 | << ", leaving uncompressed" | |
15611 | << dendl; | |
15612 | logger->inc(l_bluestore_compress_rejected_count); | |
15613 | need += wi.blob_length; | |
39ae355f | 15614 | data_size += wi.bl.length(); |
7c673cae | 15615 | } else { |
a8e16298 TL |
15616 | rejected = true; |
15617 | } | |
15618 | ||
15619 | if (rejected) { | |
3efd9988 | 15620 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 15621 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
15622 | << " with " << c->get_type() |
15623 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 15624 | << " -> 0x" << want_len |
3efd9988 FG |
15625 | << ", leaving uncompressed" |
15626 | << std::dec << dendl; | |
15627 | logger->inc(l_bluestore_compress_rejected_count); | |
15628 | need += wi.blob_length; | |
39ae355f | 15629 | data_size += wi.bl.length(); |
7c673cae | 15630 | } |
494da23a TL |
15631 | log_latency("compress@_do_alloc_write", |
15632 | l_bluestore_compress_lat, | |
15633 | mono_clock::now() - start, | |
15634 | cct->_conf->bluestore_log_op_age ); | |
3efd9988 FG |
15635 | } else { |
15636 | need += wi.blob_length; | |
39ae355f | 15637 | data_size += wi.bl.length(); |
7c673cae | 15638 | } |
3efd9988 | 15639 | } |
a8e16298 | 15640 | PExtentVector prealloc; |
39ae355f | 15641 | prealloc.reserve(2 * wctx->writes.size()); |
11fdf7f2 | 15642 | int64_t prealloc_left = 0; |
20effc67 | 15643 | prealloc_left = alloc->allocate( |
3efd9988 FG |
15644 | need, min_alloc_size, need, |
15645 | 0, &prealloc); | |
eafe8130 | 15646 | if (prealloc_left < 0 || prealloc_left < (int64_t)need) { |
11fdf7f2 | 15647 | derr << __func__ << " failed to allocate 0x" << std::hex << need |
eafe8130 | 15648 | << " allocated 0x " << (prealloc_left < 0 ? 0 : prealloc_left) |
11fdf7f2 | 15649 | << " min_alloc_size 0x" << min_alloc_size |
20effc67 | 15650 | << " available 0x " << alloc->get_free() |
11fdf7f2 TL |
15651 | << std::dec << dendl; |
15652 | if (prealloc.size()) { | |
20effc67 | 15653 | alloc->release(prealloc); |
11fdf7f2 | 15654 | } |
a8e16298 TL |
15655 | return -ENOSPC; |
15656 | } | |
20effc67 | 15657 | _collect_allocation_stats(need, min_alloc_size, prealloc); |
f67539c2 | 15658 | |
39ae355f TL |
15659 | dout(20) << __func__ << std::hex << " need=0x" << need << " data=0x" << data_size |
15660 | << " prealloc " << prealloc << dendl; | |
3efd9988 | 15661 | auto prealloc_pos = prealloc.begin(); |
522d829b | 15662 | ceph_assert(prealloc_pos != prealloc.end()); |
3efd9988 FG |
15663 | |
15664 | for (auto& wi : wctx->writes) { | |
522d829b | 15665 | bluestore_blob_t& dblob = wi.b->dirty_blob(); |
3efd9988 FG |
15666 | uint64_t b_off = wi.b_off; |
15667 | bufferlist *l = &wi.bl; | |
15668 | uint64_t final_length = wi.blob_length; | |
15669 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
15670 | if (wi.compressed) { |
15671 | final_length = wi.compressed_bl.length(); | |
15672 | csum_length = final_length; | |
1e59de90 | 15673 | unsigned csum_order = std::countr_zero(csum_length); |
3efd9988 FG |
15674 | l = &wi.compressed_bl; |
15675 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
adb31ebb | 15676 | if (csum != Checksummer::CSUM_NONE) { |
522d829b TL |
15677 | dout(20) << __func__ |
15678 | << " initialize csum setting for compressed blob " << *wi.b | |
adb31ebb TL |
15679 | << " csum_type " << Checksummer::get_csum_type_string(csum) |
15680 | << " csum_order " << csum_order | |
15681 | << " csum_length 0x" << std::hex << csum_length | |
15682 | << " blob_length 0x" << wi.blob_length | |
15683 | << " compressed_length 0x" << wi.compressed_len << std::dec | |
15684 | << dendl; | |
15685 | dblob.init_csum(csum, csum_order, csum_length); | |
15686 | } | |
3efd9988 | 15687 | } else if (wi.new_blob) { |
adb31ebb | 15688 | unsigned csum_order; |
7c673cae | 15689 | // initialize newly created blob only |
11fdf7f2 | 15690 | ceph_assert(dblob.is_mutable()); |
7c673cae FG |
15691 | if (l->length() != wi.blob_length) { |
15692 | // hrm, maybe we could do better here, but let's not bother. | |
15693 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
15694 | << block_size_order << dendl; | |
31f18b77 | 15695 | csum_order = block_size_order; |
7c673cae | 15696 | } else { |
1e59de90 | 15697 | csum_order = std::min<unsigned>(wctx->csum_order, std::countr_zero(l->length())); |
7c673cae FG |
15698 | } |
15699 | // try to align blob with max_blob_size to improve | |
15700 | // its reuse ratio, e.g. in case of reverse write | |
15701 | uint32_t suggested_boff = | |
15702 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
15703 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
15704 | suggested_boff + final_length <= max_bsize && | |
15705 | suggested_boff > b_off) { | |
181888fb | 15706 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 15707 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 15708 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
15709 | csum_length += suggested_boff - b_off; |
15710 | b_off = suggested_boff; | |
15711 | } | |
181888fb | 15712 | if (csum != Checksummer::CSUM_NONE) { |
522d829b TL |
15713 | dout(20) << __func__ |
15714 | << " initialize csum setting for new blob " << *wi.b | |
181888fb FG |
15715 | << " csum_type " << Checksummer::get_csum_type_string(csum) |
15716 | << " csum_order " << csum_order | |
15717 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
15718 | << dendl; | |
15719 | dblob.init_csum(csum, csum_order, csum_length); | |
15720 | } | |
7c673cae FG |
15721 | } |
15722 | ||
a8e16298 | 15723 | PExtentVector extents; |
3efd9988 | 15724 | int64_t left = final_length; |
522d829b | 15725 | auto prefer_deferred_size_snapshot = prefer_deferred_size.load(); |
3efd9988 | 15726 | while (left > 0) { |
11fdf7f2 | 15727 | ceph_assert(prealloc_left > 0); |
3efd9988 FG |
15728 | if (prealloc_pos->length <= left) { |
15729 | prealloc_left -= prealloc_pos->length; | |
15730 | left -= prealloc_pos->length; | |
15731 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
15732 | extents.push_back(*prealloc_pos); | |
15733 | ++prealloc_pos; | |
15734 | } else { | |
15735 | extents.emplace_back(prealloc_pos->offset, left); | |
15736 | prealloc_pos->offset += left; | |
15737 | prealloc_pos->length -= left; | |
15738 | prealloc_left -= left; | |
15739 | txc->statfs_delta.allocated() += left; | |
15740 | left = 0; | |
15741 | break; | |
15742 | } | |
15743 | } | |
7c673cae | 15744 | for (auto& p : extents) { |
3efd9988 | 15745 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 15746 | } |
11fdf7f2 | 15747 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 15748 | |
522d829b | 15749 | dout(20) << __func__ << " blob " << *wi.b << dendl; |
181888fb | 15750 | if (dblob.has_csum()) { |
7c673cae FG |
15751 | dblob.calc_csum(b_off, *l); |
15752 | } | |
181888fb | 15753 | |
7c673cae | 15754 | if (wi.mark_unused) { |
1911f103 | 15755 | ceph_assert(!dblob.is_compressed()); |
7c673cae FG |
15756 | auto b_end = b_off + wi.bl.length(); |
15757 | if (b_off) { | |
15758 | dblob.add_unused(0, b_off); | |
15759 | } | |
1911f103 TL |
15760 | uint64_t llen = dblob.get_logical_length(); |
15761 | if (b_end < llen) { | |
15762 | dblob.add_unused(b_end, llen - b_end); | |
7c673cae FG |
15763 | } |
15764 | } | |
15765 | ||
15766 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
15767 | b_off + (wi.b_off0 - wi.b_off), | |
15768 | wi.length0, | |
15769 | wi.b, | |
15770 | nullptr); | |
15771 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
15772 | txc->statfs_delta.stored() += le->length; | |
15773 | dout(20) << __func__ << " lex " << *le << dendl; | |
15774 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
15775 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
15776 | ||
15777 | // queue io | |
11fdf7f2 | 15778 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
39ae355f | 15779 | if (data_size < prefer_deferred_size_snapshot) { |
f67539c2 | 15780 | dout(20) << __func__ << " deferring 0x" << std::hex |
7c673cae | 15781 | << l->length() << std::dec << " write via deferred" << dendl; |
522d829b | 15782 | bluestore_deferred_op_t *op = _get_deferred_op(txc, l->length()); |
7c673cae | 15783 | op->op = bluestore_deferred_op_t::OP_WRITE; |
522d829b | 15784 | int r = wi.b->get_blob().map( |
7c673cae FG |
15785 | b_off, l->length(), |
15786 | [&](uint64_t offset, uint64_t length) { | |
15787 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
15788 | return 0; | |
15789 | }); | |
11fdf7f2 | 15790 | ceph_assert(r == 0); |
7c673cae FG |
15791 | op->data = *l; |
15792 | } else { | |
522d829b | 15793 | wi.b->get_blob().map_bl( |
7c673cae FG |
15794 | b_off, *l, |
15795 | [&](uint64_t offset, bufferlist& t) { | |
15796 | bdev->aio_write(offset, t, &txc->ioc, false); | |
15797 | }); | |
f67539c2 | 15798 | logger->inc(l_bluestore_write_new); |
7c673cae FG |
15799 | } |
15800 | } | |
15801 | } | |
11fdf7f2 TL |
15802 | ceph_assert(prealloc_pos == prealloc.end()); |
15803 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
15804 | return 0; |
15805 | } | |
15806 | ||
15807 | void BlueStore::_wctx_finish( | |
15808 | TransContext *txc, | |
15809 | CollectionRef& c, | |
39ae355f | 15810 | OnodeRef& o, |
31f18b77 FG |
15811 | WriteContext *wctx, |
15812 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae | 15813 | { |
20effc67 TL |
15814 | #ifdef HAVE_LIBZBD |
15815 | if (bdev->is_smr()) { | |
15816 | for (auto& w : wctx->writes) { | |
15817 | for (auto& e : w.b->get_blob().get_extents()) { | |
15818 | if (!e.is_valid()) { | |
15819 | continue; | |
15820 | } | |
15821 | uint32_t zone = e.offset / zone_size; | |
15822 | if (!o->onode.zone_offset_refs.count(zone)) { | |
15823 | uint64_t zoff = e.offset % zone_size; | |
15824 | dout(20) << __func__ << " add ref zone 0x" << std::hex << zone | |
15825 | << " offset 0x" << zoff << std::dec << dendl; | |
15826 | txc->note_write_zone_offset(o, zone, zoff); | |
15827 | } | |
15828 | } | |
15829 | } | |
15830 | } | |
15831 | set<uint32_t> zones_with_releases; | |
15832 | #endif | |
15833 | ||
7c673cae FG |
15834 | auto oep = wctx->old_extents.begin(); |
15835 | while (oep != wctx->old_extents.end()) { | |
15836 | auto &lo = *oep; | |
15837 | oep = wctx->old_extents.erase(oep); | |
15838 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
15839 | BlobRef b = lo.e.blob; | |
15840 | const bluestore_blob_t& blob = b->get_blob(); | |
15841 | if (blob.is_compressed()) { | |
15842 | if (lo.blob_empty) { | |
15843 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
15844 | } | |
15845 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
15846 | } | |
15847 | auto& r = lo.r; | |
15848 | txc->statfs_delta.stored() -= lo.e.length; | |
15849 | if (!r.empty()) { | |
f67539c2 | 15850 | dout(20) << __func__ << " blob " << *b << " release " << r << dendl; |
7c673cae FG |
15851 | if (blob.is_shared()) { |
15852 | PExtentVector final; | |
15853 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
15854 | bool unshare = false; |
15855 | bool* unshare_ptr = | |
15856 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 15857 | for (auto e : r) { |
31f18b77 FG |
15858 | b->shared_blob->put_ref( |
15859 | e.offset, e.length, &final, | |
11fdf7f2 | 15860 | unshare_ptr); |
20effc67 TL |
15861 | #ifdef HAVE_LIBZBD |
15862 | // we also drop zone ref for shared blob extents | |
15863 | if (bdev->is_smr() && e.is_valid()) { | |
15864 | zones_with_releases.insert(e.offset / zone_size); | |
15865 | } | |
15866 | #endif | |
11fdf7f2 TL |
15867 | } |
15868 | if (unshare) { | |
15869 | ceph_assert(maybe_unshared_blobs); | |
15870 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
15871 | } |
15872 | dout(20) << __func__ << " shared_blob release " << final | |
15873 | << " from " << *b->shared_blob << dendl; | |
15874 | txc->write_shared_blob(b->shared_blob); | |
15875 | r.clear(); | |
15876 | r.swap(final); | |
15877 | } | |
15878 | } | |
15879 | // we can't invalidate our logical extents as we drop them because | |
15880 | // other lextents (either in our onode or others) may still | |
15881 | // reference them. but we can throw out anything that is no | |
15882 | // longer allocated. Note that this will leave behind edge bits | |
15883 | // that are no longer referenced but not deallocated (until they | |
15884 | // age out of the cache naturally). | |
15885 | b->discard_unallocated(c.get()); | |
15886 | for (auto e : r) { | |
15887 | dout(20) << __func__ << " release " << e << dendl; | |
15888 | txc->released.insert(e.offset, e.length); | |
15889 | txc->statfs_delta.allocated() -= e.length; | |
15890 | if (blob.is_compressed()) { | |
15891 | txc->statfs_delta.compressed_allocated() -= e.length; | |
15892 | } | |
20effc67 TL |
15893 | #ifdef HAVE_LIBZBD |
15894 | if (bdev->is_smr() && e.is_valid()) { | |
15895 | zones_with_releases.insert(e.offset / zone_size); | |
15896 | } | |
15897 | #endif | |
7c673cae | 15898 | } |
9f95a23c TL |
15899 | |
15900 | if (b->is_spanning() && !b->is_referenced() && lo.blob_empty) { | |
7c673cae FG |
15901 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b |
15902 | << dendl; | |
15903 | o->extent_map.spanning_blob_map.erase(b->id); | |
15904 | } | |
9f95a23c | 15905 | delete &lo; |
7c673cae | 15906 | } |
20effc67 TL |
15907 | |
15908 | #ifdef HAVE_LIBZBD | |
15909 | if (!zones_with_releases.empty()) { | |
15910 | // we need to fault the entire extent range in here to determinte if we've dropped | |
15911 | // all refs to a zone. | |
15912 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
15913 | for (auto& b : o->extent_map.extent_map) { | |
15914 | for (auto& e : b.blob->get_blob().get_extents()) { | |
15915 | if (e.is_valid()) { | |
15916 | zones_with_releases.erase(e.offset / zone_size); | |
15917 | } | |
15918 | } | |
15919 | } | |
15920 | for (auto zone : zones_with_releases) { | |
15921 | auto p = o->onode.zone_offset_refs.find(zone); | |
15922 | if (p != o->onode.zone_offset_refs.end()) { | |
15923 | dout(20) << __func__ << " rm ref zone 0x" << std::hex << zone | |
15924 | << " offset 0x" << p->second << std::dec << dendl; | |
15925 | txc->note_release_zone_offset(o, zone, p->second); | |
15926 | } | |
15927 | } | |
15928 | } | |
15929 | #endif | |
7c673cae FG |
15930 | } |
15931 | ||
15932 | void BlueStore::_do_write_data( | |
15933 | TransContext *txc, | |
15934 | CollectionRef& c, | |
39ae355f | 15935 | OnodeRef& o, |
7c673cae FG |
15936 | uint64_t offset, |
15937 | uint64_t length, | |
15938 | bufferlist& bl, | |
15939 | WriteContext *wctx) | |
15940 | { | |
15941 | uint64_t end = offset + length; | |
15942 | bufferlist::iterator p = bl.begin(); | |
15943 | ||
15944 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
15945 | (length != min_alloc_size)) { | |
15946 | // we fall within the same block | |
15947 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
15948 | } else { | |
15949 | uint64_t head_offset, head_length; | |
15950 | uint64_t middle_offset, middle_length; | |
15951 | uint64_t tail_offset, tail_length; | |
15952 | ||
15953 | head_offset = offset; | |
11fdf7f2 | 15954 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 15955 | |
11fdf7f2 TL |
15956 | tail_offset = p2align(end, min_alloc_size); |
15957 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
15958 | |
15959 | middle_offset = head_offset + head_length; | |
15960 | middle_length = length - head_length - tail_length; | |
15961 | ||
15962 | if (head_length) { | |
15963 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
15964 | } | |
15965 | ||
f67539c2 | 15966 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); |
7c673cae FG |
15967 | |
15968 | if (tail_length) { | |
15969 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
15970 | } | |
15971 | } | |
15972 | } | |
15973 | ||
31f18b77 FG |
15974 | void BlueStore::_choose_write_options( |
15975 | CollectionRef& c, | |
39ae355f | 15976 | OnodeRef& o, |
31f18b77 FG |
15977 | uint32_t fadvise_flags, |
15978 | WriteContext *wctx) | |
7c673cae | 15979 | { |
7c673cae FG |
15980 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
15981 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 15982 | wctx->buffered = true; |
7c673cae FG |
15983 | } else if (cct->_conf->bluestore_default_buffered_write && |
15984 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
15985 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
15986 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 15987 | wctx->buffered = true; |
7c673cae FG |
15988 | } |
15989 | ||
31f18b77 FG |
15990 | // apply basic csum block size |
15991 | wctx->csum_order = block_size_order; | |
7c673cae FG |
15992 | |
15993 | // compression parameters | |
15994 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
15995 | auto cm = select_option( | |
15996 | "compression_mode", | |
31f18b77 | 15997 | comp_mode.load(), |
7c673cae FG |
15998 | [&]() { |
15999 | string val; | |
11fdf7f2 | 16000 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
1e59de90 | 16001 | return std::optional<Compressor::CompressionMode>( |
31f18b77 | 16002 | Compressor::get_comp_mode_type(val)); |
7c673cae | 16003 | } |
1e59de90 | 16004 | return std::optional<Compressor::CompressionMode>(); |
7c673cae FG |
16005 | } |
16006 | ); | |
31f18b77 FG |
16007 | |
16008 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
16009 | ((cm == Compressor::COMP_FORCE) || |
16010 | (cm == Compressor::COMP_AGGRESSIVE && | |
16011 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
16012 | (cm == Compressor::COMP_PASSIVE && | |
16013 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
16014 | ||
16015 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
16016 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
16017 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
16018 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 16019 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 16020 | |
7c673cae | 16021 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 16022 | |
7c673cae | 16023 | if (o->onode.expected_write_size) { |
224ce89b | 16024 | wctx->csum_order = std::max(min_alloc_size_order, |
1e59de90 | 16025 | (uint8_t)std::countr_zero(o->onode.expected_write_size)); |
7c673cae | 16026 | } else { |
224ce89b | 16027 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
16028 | } |
16029 | ||
31f18b77 FG |
16030 | if (wctx->compress) { |
16031 | wctx->target_blob_size = select_option( | |
7c673cae | 16032 | "compression_max_blob_size", |
31f18b77 | 16033 | comp_max_blob_size.load(), |
7c673cae | 16034 | [&]() { |
11fdf7f2 TL |
16035 | int64_t val; |
16036 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
1e59de90 | 16037 | return std::optional<uint64_t>((uint64_t)val); |
7c673cae | 16038 | } |
1e59de90 | 16039 | return std::optional<uint64_t>(); |
7c673cae FG |
16040 | } |
16041 | ); | |
16042 | } | |
16043 | } else { | |
31f18b77 FG |
16044 | if (wctx->compress) { |
16045 | wctx->target_blob_size = select_option( | |
7c673cae | 16046 | "compression_min_blob_size", |
31f18b77 | 16047 | comp_min_blob_size.load(), |
7c673cae | 16048 | [&]() { |
11fdf7f2 TL |
16049 | int64_t val; |
16050 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
1e59de90 | 16051 | return std::optional<uint64_t>((uint64_t)val); |
7c673cae | 16052 | } |
1e59de90 | 16053 | return std::optional<uint64_t>(); |
7c673cae FG |
16054 | } |
16055 | ); | |
16056 | } | |
16057 | } | |
31f18b77 | 16058 | |
7c673cae | 16059 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
16060 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
16061 | wctx->target_blob_size = max_bsize; | |
7c673cae | 16062 | } |
31f18b77 | 16063 | |
7c673cae FG |
16064 | // set the min blob size floor at 2x the min_alloc_size, or else we |
16065 | // won't be able to allocate a smaller extent for the compressed | |
16066 | // data. | |
31f18b77 FG |
16067 | if (wctx->compress && |
16068 | wctx->target_blob_size < min_alloc_size * 2) { | |
16069 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 16070 | } |
31f18b77 FG |
16071 | |
16072 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
16073 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
16074 | << " compress=" << (int)wctx->compress |
16075 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
16076 | << std::dec << dendl; |
16077 | } | |
16078 | ||
16079 | int BlueStore::_do_gc( | |
16080 | TransContext *txc, | |
16081 | CollectionRef& c, | |
39ae355f | 16082 | OnodeRef& o, |
31f18b77 FG |
16083 | const WriteContext& wctx, |
16084 | uint64_t *dirty_start, | |
16085 | uint64_t *dirty_end) | |
16086 | { | |
31f18b77 | 16087 | |
1adf2230 | 16088 | bool dirty_range_updated = false; |
31f18b77 | 16089 | WriteContext wctx_gc; |
7c673cae | 16090 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 16091 | |
eafe8130 | 16092 | auto & extents_to_collect = wctx.extents_to_gc; |
31f18b77 FG |
16093 | for (auto it = extents_to_collect.begin(); |
16094 | it != extents_to_collect.end(); | |
16095 | ++it) { | |
16096 | bufferlist bl; | |
eafe8130 TL |
16097 | auto offset = (*it).first; |
16098 | auto length = (*it).second; | |
16099 | dout(20) << __func__ << " processing " << std::hex | |
16100 | << offset << "~" << length << std::dec | |
16101 | << dendl; | |
16102 | int r = _do_read(c.get(), o, offset, length, bl, 0); | |
16103 | ceph_assert(r == (int)length); | |
31f18b77 | 16104 | |
eafe8130 TL |
16105 | _do_write_data(txc, c, o, offset, length, bl, &wctx_gc); |
16106 | logger->inc(l_bluestore_gc_merged, length); | |
31f18b77 | 16107 | |
eafe8130 TL |
16108 | if (*dirty_start > offset) { |
16109 | *dirty_start = offset; | |
1adf2230 | 16110 | dirty_range_updated = true; |
31f18b77 FG |
16111 | } |
16112 | ||
eafe8130 TL |
16113 | if (*dirty_end < offset + length) { |
16114 | *dirty_end = offset + length; | |
1adf2230 | 16115 | dirty_range_updated = true; |
31f18b77 FG |
16116 | } |
16117 | } | |
1adf2230 AA |
16118 | if (dirty_range_updated) { |
16119 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
16120 | } | |
31f18b77 FG |
16121 | |
16122 | dout(30) << __func__ << " alloc write" << dendl; | |
16123 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
16124 | if (r < 0) { | |
16125 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
16126 | << dendl; | |
16127 | return r; | |
16128 | } | |
16129 | ||
16130 | _wctx_finish(txc, c, o, &wctx_gc); | |
16131 | return 0; | |
16132 | } | |
16133 | ||
16134 | int BlueStore::_do_write( | |
16135 | TransContext *txc, | |
16136 | CollectionRef& c, | |
39ae355f | 16137 | OnodeRef& o, |
31f18b77 FG |
16138 | uint64_t offset, |
16139 | uint64_t length, | |
16140 | bufferlist& bl, | |
16141 | uint32_t fadvise_flags) | |
16142 | { | |
16143 | int r = 0; | |
16144 | ||
16145 | dout(20) << __func__ | |
16146 | << " " << o->oid | |
16147 | << " 0x" << std::hex << offset << "~" << length | |
16148 | << " - have 0x" << o->onode.size | |
16149 | << " (" << std::dec << o->onode.size << ")" | |
f67539c2 TL |
16150 | << " bytes" << std::hex |
16151 | << " fadvise_flags 0x" << fadvise_flags | |
16152 | << " alloc_hint 0x" << o->onode.alloc_hint_flags | |
16153 | << " expected_object_size " << o->onode.expected_object_size | |
16154 | << " expected_write_size " << o->onode.expected_write_size | |
16155 | << std::dec | |
31f18b77 | 16156 | << dendl; |
81eedcae | 16157 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
16158 | |
16159 | if (length == 0) { | |
16160 | return 0; | |
16161 | } | |
16162 | ||
16163 | uint64_t end = offset + length; | |
16164 | ||
16165 | GarbageCollector gc(c->store->cct); | |
eafe8130 | 16166 | int64_t benefit = 0; |
31f18b77 FG |
16167 | auto dirty_start = offset; |
16168 | auto dirty_end = end; | |
16169 | ||
16170 | WriteContext wctx; | |
16171 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
16172 | o->extent_map.fault_range(db, offset, length); |
16173 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
16174 | r = _do_alloc_write(txc, c, o, &wctx); |
16175 | if (r < 0) { | |
16176 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
16177 | << dendl; | |
16178 | goto out; | |
16179 | } | |
16180 | ||
eafe8130 TL |
16181 | if (wctx.extents_to_gc.empty() || |
16182 | wctx.extents_to_gc.range_start() > offset || | |
16183 | wctx.extents_to_gc.range_end() < offset + length) { | |
16184 | benefit = gc.estimate(offset, | |
16185 | length, | |
16186 | o->extent_map, | |
16187 | wctx.old_extents, | |
16188 | min_alloc_size); | |
16189 | } | |
16190 | ||
31f18b77 FG |
16191 | // NB: _wctx_finish() will empty old_extents |
16192 | // so we must do gc estimation before that | |
7c673cae FG |
16193 | _wctx_finish(txc, c, o, &wctx); |
16194 | if (end > o->onode.size) { | |
16195 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 16196 | << std::dec << dendl; |
7c673cae FG |
16197 | o->onode.size = end; |
16198 | } | |
16199 | ||
11fdf7f2 | 16200 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
eafe8130 TL |
16201 | wctx.extents_to_gc.union_of(gc.get_extents_to_collect()); |
16202 | dout(20) << __func__ | |
16203 | << " perform garbage collection for compressed extents, " | |
16204 | << "expected benefit = " << benefit << " AUs" << dendl; | |
16205 | } | |
16206 | if (!wctx.extents_to_gc.empty()) { | |
16207 | dout(20) << __func__ << " perform garbage collection" << dendl; | |
16208 | ||
16209 | r = _do_gc(txc, c, o, | |
16210 | wctx, | |
16211 | &dirty_start, &dirty_end); | |
16212 | if (r < 0) { | |
16213 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
16214 | << dendl; | |
16215 | goto out; | |
7c673cae | 16216 | } |
eafe8130 TL |
16217 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
16218 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae | 16219 | } |
7c673cae | 16220 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
16221 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
16222 | ||
7c673cae FG |
16223 | r = 0; |
16224 | ||
16225 | out: | |
16226 | return r; | |
16227 | } | |
16228 | ||
16229 | int BlueStore::_write(TransContext *txc, | |
16230 | CollectionRef& c, | |
16231 | OnodeRef& o, | |
31f18b77 FG |
16232 | uint64_t offset, size_t length, |
16233 | bufferlist& bl, | |
16234 | uint32_t fadvise_flags) | |
7c673cae FG |
16235 | { |
16236 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16237 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
16238 | << dendl; | |
35e4c445 FG |
16239 | int r = 0; |
16240 | if (offset + length >= OBJECT_MAX_SIZE) { | |
16241 | r = -E2BIG; | |
16242 | } else { | |
16243 | _assign_nid(txc, o); | |
16244 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
16245 | txc->write_onode(o); | |
16246 | } | |
7c673cae FG |
16247 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
16248 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
16249 | << " = " << r << dendl; | |
16250 | return r; | |
16251 | } | |
16252 | ||
16253 | int BlueStore::_zero(TransContext *txc, | |
16254 | CollectionRef& c, | |
16255 | OnodeRef& o, | |
16256 | uint64_t offset, size_t length) | |
16257 | { | |
16258 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16259 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
16260 | << dendl; | |
35e4c445 FG |
16261 | int r = 0; |
16262 | if (offset + length >= OBJECT_MAX_SIZE) { | |
16263 | r = -E2BIG; | |
16264 | } else { | |
16265 | _assign_nid(txc, o); | |
16266 | r = _do_zero(txc, c, o, offset, length); | |
16267 | } | |
7c673cae FG |
16268 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
16269 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
16270 | << " = " << r << dendl; | |
16271 | return r; | |
16272 | } | |
16273 | ||
16274 | int BlueStore::_do_zero(TransContext *txc, | |
16275 | CollectionRef& c, | |
16276 | OnodeRef& o, | |
16277 | uint64_t offset, size_t length) | |
16278 | { | |
16279 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16280 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
16281 | << dendl; | |
16282 | int r = 0; | |
16283 | ||
81eedcae | 16284 | _dump_onode<30>(cct, *o); |
7c673cae FG |
16285 | |
16286 | WriteContext wctx; | |
16287 | o->extent_map.fault_range(db, offset, length); | |
16288 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 16289 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
16290 | _wctx_finish(txc, c, o, &wctx); |
16291 | ||
b32b8144 | 16292 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
16293 | o->onode.size = offset + length; |
16294 | dout(20) << __func__ << " extending size to " << offset + length | |
16295 | << dendl; | |
16296 | } | |
16297 | txc->write_onode(o); | |
16298 | ||
16299 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
16300 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
16301 | << " = " << r << dendl; | |
16302 | return r; | |
16303 | } | |
16304 | ||
16305 | void BlueStore::_do_truncate( | |
39ae355f | 16306 | TransContext *txc, CollectionRef& c, OnodeRef& o, uint64_t offset, |
31f18b77 | 16307 | set<SharedBlob*> *maybe_unshared_blobs) |
7c673cae FG |
16308 | { |
16309 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16310 | << " 0x" << std::hex << offset << std::dec << dendl; | |
16311 | ||
81eedcae | 16312 | _dump_onode<30>(cct, *o); |
7c673cae FG |
16313 | |
16314 | if (offset == o->onode.size) | |
31f18b77 | 16315 | return; |
7c673cae | 16316 | |
f67539c2 | 16317 | WriteContext wctx; |
7c673cae | 16318 | if (offset < o->onode.size) { |
7c673cae FG |
16319 | uint64_t length = o->onode.size - offset; |
16320 | o->extent_map.fault_range(db, offset, length); | |
16321 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 16322 | o->extent_map.dirty_range(offset, length); |
20effc67 | 16323 | |
31f18b77 | 16324 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); |
7c673cae FG |
16325 | |
16326 | // if we have shards past EOF, ask for a reshard | |
16327 | if (!o->onode.extent_map_shards.empty() && | |
16328 | o->onode.extent_map_shards.back().offset >= offset) { | |
16329 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
16330 | if (offset) { | |
16331 | o->extent_map.request_reshard(offset - 1, offset + length); | |
16332 | } else { | |
16333 | o->extent_map.request_reshard(0, length); | |
16334 | } | |
16335 | } | |
16336 | } | |
16337 | ||
16338 | o->onode.size = offset; | |
16339 | ||
16340 | txc->write_onode(o); | |
16341 | } | |
16342 | ||
35e4c445 | 16343 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
16344 | CollectionRef& c, |
16345 | OnodeRef& o, | |
16346 | uint64_t offset) | |
16347 | { | |
16348 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16349 | << " 0x" << std::hex << offset << std::dec | |
16350 | << dendl; | |
20effc67 TL |
16351 | |
16352 | auto start_time = mono_clock::now(); | |
35e4c445 FG |
16353 | int r = 0; |
16354 | if (offset >= OBJECT_MAX_SIZE) { | |
16355 | r = -E2BIG; | |
16356 | } else { | |
16357 | _do_truncate(txc, c, o, offset); | |
16358 | } | |
20effc67 TL |
16359 | log_latency_fn( |
16360 | __func__, | |
16361 | l_bluestore_truncate_lat, | |
16362 | mono_clock::now() - start_time, | |
16363 | cct->_conf->bluestore_log_op_age, | |
16364 | [&](const ceph::timespan& lat) { | |
16365 | ostringstream ostr; | |
16366 | ostr << ", lat = " << timespan_str(lat) | |
16367 | << " cid =" << c->cid | |
16368 | << " oid =" << o->oid; | |
16369 | return ostr.str(); | |
16370 | } | |
16371 | ); | |
35e4c445 FG |
16372 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
16373 | << " 0x" << std::hex << offset << std::dec | |
16374 | << " = " << r << dendl; | |
16375 | return r; | |
7c673cae FG |
16376 | } |
16377 | ||
16378 | int BlueStore::_do_remove( | |
16379 | TransContext *txc, | |
16380 | CollectionRef& c, | |
39ae355f | 16381 | OnodeRef& o) |
7c673cae | 16382 | { |
31f18b77 | 16383 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
16384 | bool is_gen = !o->oid.is_no_gen(); |
16385 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
16386 | if (o->onode.has_omap()) { |
16387 | o->flush(); | |
9f95a23c | 16388 | _do_omap_clear(txc, o); |
7c673cae FG |
16389 | } |
16390 | o->exists = false; | |
16391 | string key; | |
16392 | for (auto &s : o->extent_map.shards) { | |
16393 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
16394 | << s.shard_info->offset << std::dec << dendl; | |
16395 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
16396 | [&](const string& final_key) { | |
16397 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
16398 | } | |
16399 | ); | |
16400 | } | |
16401 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 16402 | txc->note_removed_object(o); |
7c673cae FG |
16403 | o->extent_map.clear(); |
16404 | o->onode = bluestore_onode_t(); | |
16405 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 16406 | |
224ce89b WB |
16407 | if (!is_gen || maybe_unshared_blobs.empty()) { |
16408 | return 0; | |
16409 | } | |
31f18b77 | 16410 | |
224ce89b WB |
16411 | // see if we can unshare blobs still referenced by the head |
16412 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
16413 | << maybe_unshared_blobs << dendl; | |
16414 | ghobject_t nogen = o->oid; | |
16415 | nogen.generation = ghobject_t::NO_GEN; | |
f67539c2 | 16416 | OnodeRef h = c->get_onode(nogen, false); |
224ce89b WB |
16417 | |
16418 | if (!h || !h->exists) { | |
16419 | return 0; | |
16420 | } | |
16421 | ||
16422 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
16423 | << " " << h->oid << dendl; | |
16424 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
16425 | for (auto& e : h->extent_map.extent_map) { | |
16426 | const bluestore_blob_t& b = e.blob->get_blob(); | |
16427 | SharedBlob *sb = e.blob->shared_blob.get(); | |
16428 | if (b.is_shared() && | |
16429 | sb->loaded && | |
16430 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
16431 | if (b.is_compressed()) { |
16432 | expect[sb].get(0, b.get_ondisk_length()); | |
16433 | } else { | |
16434 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
16435 | expect[sb].get(off, len); | |
16436 | return 0; | |
16437 | }); | |
16438 | } | |
224ce89b WB |
16439 | } |
16440 | } | |
31f18b77 | 16441 | |
224ce89b WB |
16442 | vector<SharedBlob*> unshared_blobs; |
16443 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
16444 | for (auto& p : expect) { | |
16445 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
16446 | if (p.first->persistent->ref_map == p.second) { | |
16447 | SharedBlob *sb = p.first; | |
16448 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
16449 | unshared_blobs.push_back(sb); | |
16450 | txc->unshare_blob(sb); | |
16451 | uint64_t sbid = c->make_blob_unshared(sb); | |
16452 | string key; | |
16453 | get_shared_blob_key(sbid, &key); | |
16454 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
16455 | } | |
16456 | } | |
16457 | ||
16458 | if (unshared_blobs.empty()) { | |
16459 | return 0; | |
16460 | } | |
16461 | ||
224ce89b WB |
16462 | for (auto& e : h->extent_map.extent_map) { |
16463 | const bluestore_blob_t& b = e.blob->get_blob(); | |
16464 | SharedBlob *sb = e.blob->shared_blob.get(); | |
16465 | if (b.is_shared() && | |
16466 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
16467 | sb) != unshared_blobs.end()) { | |
16468 | dout(20) << __func__ << " unsharing " << e << dendl; | |
16469 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
16470 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 16471 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
16472 | } |
16473 | } | |
224ce89b WB |
16474 | txc->write_onode(h); |
16475 | ||
7c673cae FG |
16476 | return 0; |
16477 | } | |
16478 | ||
16479 | int BlueStore::_remove(TransContext *txc, | |
16480 | CollectionRef& c, | |
39ae355f | 16481 | OnodeRef& o) |
7c673cae | 16482 | { |
11fdf7f2 TL |
16483 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
16484 | << " onode " << o.get() | |
16485 | << " txc "<< txc << dendl; | |
20effc67 | 16486 | auto start_time = mono_clock::now(); |
7c673cae | 16487 | int r = _do_remove(txc, c, o); |
20effc67 | 16488 | |
adb31ebb TL |
16489 | log_latency_fn( |
16490 | __func__, | |
16491 | l_bluestore_remove_lat, | |
16492 | mono_clock::now() - start_time, | |
16493 | cct->_conf->bluestore_log_op_age, | |
16494 | [&](const ceph::timespan& lat) { | |
16495 | ostringstream ostr; | |
16496 | ostr << ", lat = " << timespan_str(lat) | |
16497 | << " cid =" << c->cid | |
16498 | << " oid =" << o->oid; | |
16499 | return ostr.str(); | |
16500 | } | |
16501 | ); | |
16502 | ||
7c673cae FG |
16503 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; |
16504 | return r; | |
16505 | } | |
16506 | ||
16507 | int BlueStore::_setattr(TransContext *txc, | |
16508 | CollectionRef& c, | |
16509 | OnodeRef& o, | |
16510 | const string& name, | |
16511 | bufferptr& val) | |
16512 | { | |
16513 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16514 | << " " << name << " (" << val.length() << " bytes)" | |
16515 | << dendl; | |
16516 | int r = 0; | |
3efd9988 FG |
16517 | if (val.is_partial()) { |
16518 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
16519 | val.length()); | |
f91f0fd5 | 16520 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
16521 | } else { |
16522 | auto& b = o->onode.attrs[name.c_str()] = val; | |
f91f0fd5 | 16523 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 16524 | } |
7c673cae FG |
16525 | txc->write_onode(o); |
16526 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
16527 | << " " << name << " (" << val.length() << " bytes)" | |
16528 | << " = " << r << dendl; | |
16529 | return r; | |
16530 | } | |
16531 | ||
16532 | int BlueStore::_setattrs(TransContext *txc, | |
16533 | CollectionRef& c, | |
16534 | OnodeRef& o, | |
16535 | const map<string,bufferptr>& aset) | |
16536 | { | |
16537 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16538 | << " " << aset.size() << " keys" | |
16539 | << dendl; | |
16540 | int r = 0; | |
16541 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
16542 | p != aset.end(); ++p) { | |
3efd9988 FG |
16543 | if (p->second.is_partial()) { |
16544 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 16545 | bufferptr(p->second.c_str(), p->second.length()); |
f91f0fd5 | 16546 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
16547 | } else { |
16548 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
f91f0fd5 | 16549 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 16550 | } |
7c673cae FG |
16551 | } |
16552 | txc->write_onode(o); | |
16553 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
16554 | << " " << aset.size() << " keys" | |
16555 | << " = " << r << dendl; | |
16556 | return r; | |
16557 | } | |
16558 | ||
16559 | ||
16560 | int BlueStore::_rmattr(TransContext *txc, | |
16561 | CollectionRef& c, | |
16562 | OnodeRef& o, | |
16563 | const string& name) | |
16564 | { | |
16565 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16566 | << " " << name << dendl; | |
16567 | int r = 0; | |
16568 | auto it = o->onode.attrs.find(name.c_str()); | |
16569 | if (it == o->onode.attrs.end()) | |
16570 | goto out; | |
16571 | ||
16572 | o->onode.attrs.erase(it); | |
16573 | txc->write_onode(o); | |
16574 | ||
16575 | out: | |
16576 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
16577 | << " " << name << " = " << r << dendl; | |
16578 | return r; | |
16579 | } | |
16580 | ||
16581 | int BlueStore::_rmattrs(TransContext *txc, | |
16582 | CollectionRef& c, | |
16583 | OnodeRef& o) | |
16584 | { | |
16585 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
16586 | int r = 0; | |
16587 | ||
16588 | if (o->onode.attrs.empty()) | |
16589 | goto out; | |
16590 | ||
16591 | o->onode.attrs.clear(); | |
16592 | txc->write_onode(o); | |
16593 | ||
16594 | out: | |
16595 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
16596 | return r; | |
16597 | } | |
16598 | ||
9f95a23c | 16599 | void BlueStore::_do_omap_clear(TransContext *txc, OnodeRef& o) |
7c673cae | 16600 | { |
9f95a23c | 16601 | const string& omap_prefix = o->get_omap_prefix(); |
7c673cae | 16602 | string prefix, tail; |
9f95a23c TL |
16603 | o->get_omap_header(&prefix); |
16604 | o->get_omap_tail(&tail); | |
11fdf7f2 | 16605 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
494da23a | 16606 | txc->t->rmkey(omap_prefix, tail); |
20effc67 | 16607 | o->onode.clear_omap_flag(); |
11fdf7f2 TL |
16608 | dout(20) << __func__ << " remove range start: " |
16609 | << pretty_binary_string(prefix) << " end: " | |
16610 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
16611 | } |
16612 | ||
16613 | int BlueStore::_omap_clear(TransContext *txc, | |
16614 | CollectionRef& c, | |
16615 | OnodeRef& o) | |
16616 | { | |
16617 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
20effc67 TL |
16618 | auto t0 = mono_clock::now(); |
16619 | ||
7c673cae FG |
16620 | int r = 0; |
16621 | if (o->onode.has_omap()) { | |
16622 | o->flush(); | |
9f95a23c | 16623 | _do_omap_clear(txc, o); |
7c673cae FG |
16624 | txc->write_onode(o); |
16625 | } | |
20effc67 TL |
16626 | logger->tinc(l_bluestore_omap_clear_lat, mono_clock::now() - t0); |
16627 | ||
7c673cae FG |
16628 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; |
16629 | return r; | |
16630 | } | |
16631 | ||
16632 | int BlueStore::_omap_setkeys(TransContext *txc, | |
16633 | CollectionRef& c, | |
16634 | OnodeRef& o, | |
16635 | bufferlist &bl) | |
16636 | { | |
16637 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
16638 | int r; | |
11fdf7f2 | 16639 | auto p = bl.cbegin(); |
7c673cae FG |
16640 | __u32 num; |
16641 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 16642 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
16643 | o->onode.set_omap_flags_pgmeta(); |
16644 | } else { | |
522d829b | 16645 | o->onode.set_omap_flags(per_pool_omap == OMAP_BULK); |
11fdf7f2 | 16646 | } |
7c673cae | 16647 | txc->write_onode(o); |
494da23a | 16648 | |
9f95a23c | 16649 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
16650 | string key_tail; |
16651 | bufferlist tail; | |
9f95a23c | 16652 | o->get_omap_tail(&key_tail); |
494da23a | 16653 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
16654 | } else { |
16655 | txc->note_modified_object(o); | |
16656 | } | |
9f95a23c | 16657 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 16658 | string final_key; |
9f95a23c TL |
16659 | o->get_omap_key(string(), &final_key); |
16660 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 16661 | decode(num, p); |
7c673cae FG |
16662 | while (num--) { |
16663 | string key; | |
16664 | bufferlist value; | |
11fdf7f2 TL |
16665 | decode(key, p); |
16666 | decode(value, p); | |
9f95a23c | 16667 | final_key.resize(base_key_len); // keep prefix |
7c673cae | 16668 | final_key += key; |
11fdf7f2 | 16669 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 16670 | << " <- " << key << dendl; |
11fdf7f2 | 16671 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
16672 | } |
16673 | r = 0; | |
16674 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
16675 | return r; | |
16676 | } | |
16677 | ||
16678 | int BlueStore::_omap_setheader(TransContext *txc, | |
16679 | CollectionRef& c, | |
39ae355f | 16680 | OnodeRef& o, |
7c673cae FG |
16681 | bufferlist& bl) |
16682 | { | |
16683 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
16684 | int r; | |
16685 | string key; | |
16686 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 16687 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
16688 | o->onode.set_omap_flags_pgmeta(); |
16689 | } else { | |
522d829b | 16690 | o->onode.set_omap_flags(per_pool_omap == OMAP_BULK); |
11fdf7f2 | 16691 | } |
7c673cae | 16692 | txc->write_onode(o); |
494da23a | 16693 | |
9f95a23c | 16694 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
16695 | string key_tail; |
16696 | bufferlist tail; | |
9f95a23c | 16697 | o->get_omap_tail(&key_tail); |
494da23a | 16698 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
16699 | } else { |
16700 | txc->note_modified_object(o); | |
16701 | } | |
9f95a23c TL |
16702 | const string& prefix = o->get_omap_prefix(); |
16703 | o->get_omap_header(&key); | |
11fdf7f2 | 16704 | txc->t->set(prefix, key, bl); |
7c673cae FG |
16705 | r = 0; |
16706 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
16707 | return r; | |
16708 | } | |
16709 | ||
16710 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
16711 | CollectionRef& c, | |
16712 | OnodeRef& o, | |
16713 | bufferlist& bl) | |
16714 | { | |
16715 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
16716 | int r = 0; | |
11fdf7f2 | 16717 | auto p = bl.cbegin(); |
7c673cae FG |
16718 | __u32 num; |
16719 | string final_key; | |
7c673cae FG |
16720 | if (!o->onode.has_omap()) { |
16721 | goto out; | |
16722 | } | |
11fdf7f2 | 16723 | { |
9f95a23c TL |
16724 | const string& prefix = o->get_omap_prefix(); |
16725 | o->get_omap_key(string(), &final_key); | |
16726 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 16727 | decode(num, p); |
1e59de90 | 16728 | logger->inc(l_bluestore_omap_rmkeys_count, num); |
11fdf7f2 TL |
16729 | while (num--) { |
16730 | string key; | |
16731 | decode(key, p); | |
9f95a23c | 16732 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
16733 | final_key += key; |
16734 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
16735 | << " <- " << key << dendl; | |
16736 | txc->t->rmkey(prefix, final_key); | |
16737 | } | |
7c673cae FG |
16738 | } |
16739 | txc->note_modified_object(o); | |
16740 | ||
16741 | out: | |
16742 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
16743 | return r; | |
16744 | } | |
16745 | ||
16746 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
16747 | CollectionRef& c, | |
16748 | OnodeRef& o, | |
16749 | const string& first, const string& last) | |
16750 | { | |
16751 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
16752 | string key_first, key_last; |
16753 | int r = 0; | |
16754 | if (!o->onode.has_omap()) { | |
16755 | goto out; | |
16756 | } | |
11fdf7f2 | 16757 | { |
9f95a23c | 16758 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 16759 | o->flush(); |
9f95a23c TL |
16760 | o->get_omap_key(first, &key_first); |
16761 | o->get_omap_key(last, &key_last); | |
1e59de90 | 16762 | logger->inc(l_bluestore_omap_rmkey_ranges_count); |
11fdf7f2 TL |
16763 | txc->t->rm_range_keys(prefix, key_first, key_last); |
16764 | dout(20) << __func__ << " remove range start: " | |
16765 | << pretty_binary_string(key_first) << " end: " | |
16766 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
16767 | } |
16768 | txc->note_modified_object(o); | |
16769 | ||
16770 | out: | |
7c673cae FG |
16771 | return r; |
16772 | } | |
16773 | ||
16774 | int BlueStore::_set_alloc_hint( | |
16775 | TransContext *txc, | |
16776 | CollectionRef& c, | |
16777 | OnodeRef& o, | |
16778 | uint64_t expected_object_size, | |
16779 | uint64_t expected_write_size, | |
16780 | uint32_t flags) | |
16781 | { | |
16782 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
16783 | << " object_size " << expected_object_size | |
16784 | << " write_size " << expected_write_size | |
16785 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
16786 | << dendl; | |
16787 | int r = 0; | |
16788 | o->onode.expected_object_size = expected_object_size; | |
16789 | o->onode.expected_write_size = expected_write_size; | |
16790 | o->onode.alloc_hint_flags = flags; | |
16791 | txc->write_onode(o); | |
16792 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
16793 | << " object_size " << expected_object_size | |
16794 | << " write_size " << expected_write_size | |
16795 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
16796 | << " = " << r << dendl; | |
16797 | return r; | |
16798 | } | |
16799 | ||
16800 | int BlueStore::_clone(TransContext *txc, | |
16801 | CollectionRef& c, | |
16802 | OnodeRef& oldo, | |
16803 | OnodeRef& newo) | |
16804 | { | |
16805 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
16806 | << newo->oid << dendl; | |
16807 | int r = 0; | |
16808 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
16809 | derr << __func__ << " mismatched hash on " << oldo->oid | |
16810 | << " and " << newo->oid << dendl; | |
16811 | return -EINVAL; | |
16812 | } | |
16813 | ||
7c673cae FG |
16814 | _assign_nid(txc, newo); |
16815 | ||
16816 | // clone data | |
16817 | oldo->flush(); | |
16818 | _do_truncate(txc, c, newo, 0); | |
16819 | if (cct->_conf->bluestore_clone_cow) { | |
16820 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
16821 | } else { | |
16822 | bufferlist bl; | |
16823 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
16824 | if (r < 0) | |
16825 | goto out; | |
16826 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
16827 | if (r < 0) | |
16828 | goto out; | |
16829 | } | |
16830 | ||
16831 | // clone attrs | |
16832 | newo->onode.attrs = oldo->onode.attrs; | |
16833 | ||
16834 | // clone omap | |
16835 | if (newo->onode.has_omap()) { | |
16836 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
16837 | newo->flush(); | |
9f95a23c | 16838 | _do_omap_clear(txc, newo); |
7c673cae FG |
16839 | } |
16840 | if (oldo->onode.has_omap()) { | |
16841 | dout(20) << __func__ << " copying omap data" << dendl; | |
494da23a | 16842 | if (newo->oid.is_pgmeta()) { |
9f95a23c TL |
16843 | newo->onode.set_omap_flags_pgmeta(); |
16844 | } else { | |
522d829b | 16845 | newo->onode.set_omap_flags(per_pool_omap == OMAP_BULK); |
7c673cae | 16846 | } |
20effc67 TL |
16847 | // check if prefix for omap key is exactly the same size for both objects |
16848 | // otherwise rewrite_omap_key will corrupt data | |
16849 | ceph_assert(oldo->onode.flags == newo->onode.flags); | |
9f95a23c | 16850 | const string& prefix = newo->get_omap_prefix(); |
7c673cae | 16851 | string head, tail; |
9f95a23c TL |
16852 | oldo->get_omap_header(&head); |
16853 | oldo->get_omap_tail(&tail); | |
33c7a0ef | 16854 | KeyValueDB::Iterator it = db->get_iterator(prefix, 0, KeyValueDB::IteratorBounds{head, tail}); |
7c673cae FG |
16855 | it->lower_bound(head); |
16856 | while (it->valid()) { | |
16857 | if (it->key() >= tail) { | |
16858 | dout(30) << __func__ << " reached tail" << dendl; | |
16859 | break; | |
16860 | } else { | |
16861 | dout(30) << __func__ << " got header/data " | |
16862 | << pretty_binary_string(it->key()) << dendl; | |
16863 | string key; | |
9f95a23c | 16864 | newo->rewrite_omap_key(it->key(), &key); |
11fdf7f2 | 16865 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
16866 | } |
16867 | it->next(); | |
16868 | } | |
494da23a TL |
16869 | string new_tail; |
16870 | bufferlist new_tail_value; | |
9f95a23c | 16871 | newo->get_omap_tail(&new_tail); |
494da23a | 16872 | txc->t->set(prefix, new_tail, new_tail_value); |
7c673cae FG |
16873 | } |
16874 | ||
16875 | txc->write_onode(newo); | |
16876 | r = 0; | |
16877 | ||
16878 | out: | |
16879 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
16880 | << newo->oid << " = " << r << dendl; | |
16881 | return r; | |
16882 | } | |
16883 | ||
16884 | int BlueStore::_do_clone_range( | |
16885 | TransContext *txc, | |
16886 | CollectionRef& c, | |
16887 | OnodeRef& oldo, | |
16888 | OnodeRef& newo, | |
224ce89b WB |
16889 | uint64_t srcoff, |
16890 | uint64_t length, | |
16891 | uint64_t dstoff) | |
7c673cae FG |
16892 | { |
16893 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
16894 | << newo->oid | |
16895 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
16896 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
16897 | oldo->extent_map.fault_range(db, srcoff, length); | |
16898 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
16899 | _dump_onode<30>(cct, *oldo); |
16900 | _dump_onode<30>(cct, *newo); | |
7c673cae | 16901 | |
11fdf7f2 | 16902 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
7c673cae | 16903 | |
20effc67 TL |
16904 | #ifdef HAVE_LIBZBD |
16905 | if (bdev->is_smr()) { | |
16906 | // duplicate the refs for the shared region. | |
16907 | Extent dummy(dstoff); | |
16908 | for (auto e = newo->extent_map.extent_map.lower_bound(dummy); | |
16909 | e != newo->extent_map.extent_map.end(); | |
16910 | ++e) { | |
16911 | if (e->logical_offset >= dstoff + length) { | |
16912 | break; | |
16913 | } | |
16914 | for (auto& ex : e->blob->get_blob().get_extents()) { | |
16915 | // note that we may introduce a new extent reference that is | |
16916 | // earlier than the first zone ref. we allow this since it is | |
16917 | // a lot of work to avoid and has marginal impact on cleaning | |
16918 | // performance. | |
16919 | if (!ex.is_valid()) { | |
16920 | continue; | |
16921 | } | |
16922 | uint32_t zone = ex.offset / zone_size; | |
16923 | if (!newo->onode.zone_offset_refs.count(zone)) { | |
16924 | uint64_t zoff = ex.offset % zone_size; | |
16925 | dout(20) << __func__ << " add ref zone 0x" << std::hex << zone | |
16926 | << " offset 0x" << zoff << std::dec | |
16927 | << " -> " << newo->oid << dendl; | |
16928 | txc->note_write_zone_offset(newo, zone, zoff); | |
16929 | } | |
16930 | } | |
16931 | } | |
16932 | } | |
16933 | #endif | |
16934 | ||
16935 | _dump_onode<30>(cct, *oldo); | |
16936 | _dump_onode<30>(cct, *newo); | |
16937 | return 0; | |
16938 | } | |
16939 | ||
16940 | int BlueStore::_clone_range(TransContext *txc, | |
16941 | CollectionRef& c, | |
16942 | OnodeRef& oldo, | |
16943 | OnodeRef& newo, | |
16944 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
16945 | { | |
16946 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
7c673cae FG |
16947 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length |
16948 | << " to offset 0x" << dstoff << std::dec << dendl; | |
16949 | int r = 0; | |
16950 | ||
35e4c445 FG |
16951 | if (srcoff + length >= OBJECT_MAX_SIZE || |
16952 | dstoff + length >= OBJECT_MAX_SIZE) { | |
16953 | r = -E2BIG; | |
16954 | goto out; | |
16955 | } | |
7c673cae FG |
16956 | if (srcoff + length > oldo->onode.size) { |
16957 | r = -EINVAL; | |
16958 | goto out; | |
16959 | } | |
16960 | ||
7c673cae FG |
16961 | _assign_nid(txc, newo); |
16962 | ||
16963 | if (length > 0) { | |
16964 | if (cct->_conf->bluestore_clone_cow) { | |
16965 | _do_zero(txc, c, newo, dstoff, length); | |
16966 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
16967 | } else { | |
16968 | bufferlist bl; | |
16969 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
16970 | if (r < 0) | |
16971 | goto out; | |
16972 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
16973 | if (r < 0) | |
16974 | goto out; | |
16975 | } | |
16976 | } | |
16977 | ||
16978 | txc->write_onode(newo); | |
16979 | r = 0; | |
16980 | ||
16981 | out: | |
16982 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
16983 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
16984 | << " to offset 0x" << dstoff << std::dec | |
16985 | << " = " << r << dendl; | |
16986 | return r; | |
16987 | } | |
16988 | ||
16989 | int BlueStore::_rename(TransContext *txc, | |
16990 | CollectionRef& c, | |
16991 | OnodeRef& oldo, | |
16992 | OnodeRef& newo, | |
16993 | const ghobject_t& new_oid) | |
16994 | { | |
16995 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
16996 | << new_oid << dendl; | |
16997 | int r; | |
16998 | ghobject_t old_oid = oldo->oid; | |
f91f0fd5 | 16999 | mempool::bluestore_cache_meta::string new_okey; |
7c673cae FG |
17000 | |
17001 | if (newo) { | |
17002 | if (newo->exists) { | |
17003 | r = -EEXIST; | |
17004 | goto out; | |
17005 | } | |
11fdf7f2 | 17006 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
17007 | } |
17008 | ||
17009 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
17010 | ||
17011 | // rewrite shards | |
17012 | { | |
17013 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
17014 | get_object_key(cct, new_oid, &new_okey); | |
17015 | string key; | |
17016 | for (auto &s : oldo->extent_map.shards) { | |
17017 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
17018 | [&](const string& final_key) { | |
17019 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
17020 | } | |
17021 | ); | |
17022 | s.dirty = true; | |
17023 | } | |
17024 | } | |
17025 | ||
17026 | newo = oldo; | |
17027 | txc->write_onode(newo); | |
17028 | ||
17029 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
17030 | // Onode in the old slot | |
39ae355f | 17031 | c->onode_space.rename(oldo, old_oid, new_oid, new_okey); |
7c673cae FG |
17032 | r = 0; |
17033 | ||
f64942e4 AA |
17034 | // hold a ref to new Onode in old name position, to ensure we don't drop |
17035 | // it from the cache before this txc commits (or else someone may come along | |
17036 | // and read newo's metadata via the old name). | |
17037 | txc->note_modified_object(oldo); | |
17038 | ||
20effc67 TL |
17039 | #ifdef HAVE_LIBZBD |
17040 | if (bdev->is_smr()) { | |
17041 | // adjust zone refs | |
17042 | for (auto& [zone, offset] : newo->onode.zone_offset_refs) { | |
17043 | dout(20) << __func__ << " rm ref zone 0x" << std::hex << zone | |
17044 | << " offset 0x" << offset << std::dec | |
17045 | << " -> " << oldo->oid << dendl; | |
17046 | string key; | |
17047 | get_zone_offset_object_key(zone, offset, oldo->oid, &key); | |
17048 | txc->t->rmkey(PREFIX_ZONED_CL_INFO, key); | |
17049 | ||
17050 | dout(20) << __func__ << " add ref zone 0x" << std::hex << zone | |
17051 | << " offset 0x" << offset << std::dec | |
17052 | << " -> " << newo->oid << dendl; | |
17053 | get_zone_offset_object_key(zone, offset, newo->oid, &key); | |
17054 | bufferlist v; | |
17055 | txc->t->set(PREFIX_ZONED_CL_INFO, key, v); | |
17056 | } | |
17057 | } | |
17058 | #endif | |
17059 | ||
7c673cae FG |
17060 | out: |
17061 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
17062 | << new_oid << " = " << r << dendl; | |
17063 | return r; | |
17064 | } | |
17065 | ||
17066 | // collections | |
17067 | ||
17068 | int BlueStore::_create_collection( | |
17069 | TransContext *txc, | |
17070 | const coll_t &cid, | |
17071 | unsigned bits, | |
17072 | CollectionRef *c) | |
17073 | { | |
17074 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
17075 | int r; | |
17076 | bufferlist bl; | |
17077 | ||
17078 | { | |
9f95a23c | 17079 | std::unique_lock l(coll_lock); |
7c673cae FG |
17080 | if (*c) { |
17081 | r = -EEXIST; | |
17082 | goto out; | |
17083 | } | |
11fdf7f2 TL |
17084 | auto p = new_coll_map.find(cid); |
17085 | ceph_assert(p != new_coll_map.end()); | |
17086 | *c = p->second; | |
7c673cae FG |
17087 | (*c)->cnode.bits = bits; |
17088 | coll_map[cid] = *c; | |
11fdf7f2 | 17089 | new_coll_map.erase(p); |
7c673cae | 17090 | } |
11fdf7f2 | 17091 | encode((*c)->cnode, bl); |
7c673cae FG |
17092 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
17093 | r = 0; | |
17094 | ||
17095 | out: | |
17096 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
17097 | return r; | |
17098 | } | |
17099 | ||
17100 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
17101 | CollectionRef *c) | |
17102 | { | |
17103 | dout(15) << __func__ << " " << cid << dendl; | |
17104 | int r; | |
17105 | ||
11fdf7f2 | 17106 | (*c)->flush_all_but_last(); |
7c673cae | 17107 | { |
9f95a23c | 17108 | std::unique_lock l(coll_lock); |
7c673cae FG |
17109 | if (!*c) { |
17110 | r = -ENOENT; | |
17111 | goto out; | |
17112 | } | |
17113 | size_t nonexistent_count = 0; | |
11fdf7f2 | 17114 | ceph_assert((*c)->exists); |
39ae355f | 17115 | if ((*c)->onode_space.map_any([&](Onode* o) { |
f67539c2 TL |
17116 | if (o->exists) { |
17117 | dout(1) << __func__ << " " << o->oid << " " << o | |
17118 | << " exists in onode_map" << dendl; | |
7c673cae | 17119 | return true; |
f67539c2 TL |
17120 | } |
17121 | ++nonexistent_count; | |
17122 | return false; | |
17123 | })) { | |
7c673cae FG |
17124 | r = -ENOTEMPTY; |
17125 | goto out; | |
17126 | } | |
7c673cae FG |
17127 | vector<ghobject_t> ls; |
17128 | ghobject_t next; | |
17129 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
17130 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 17131 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae | 17132 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
f91f0fd5 | 17133 | nonexistent_count + 1, false, &ls, &next); |
7c673cae | 17134 | if (r >= 0) { |
11fdf7f2 TL |
17135 | // If true mean collecton has more objects than nonexistent_count, |
17136 | // so bypass check. | |
17137 | bool exists = (!next.is_max()); | |
7c673cae FG |
17138 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
17139 | dout(10) << __func__ << " oid " << *it << dendl; | |
39ae355f | 17140 | auto onode = (*c)->onode_space.lookup(*it); |
7c673cae FG |
17141 | exists = !onode || onode->exists; |
17142 | if (exists) { | |
494da23a | 17143 | dout(1) << __func__ << " " << *it |
f67539c2 TL |
17144 | << " exists in db, " |
17145 | << (!onode ? "not present in ram" : "present in ram") | |
17146 | << dendl; | |
7c673cae FG |
17147 | } |
17148 | } | |
17149 | if (!exists) { | |
f67539c2 | 17150 | _do_remove_collection(txc, c); |
7c673cae FG |
17151 | r = 0; |
17152 | } else { | |
17153 | dout(10) << __func__ << " " << cid | |
17154 | << " is non-empty" << dendl; | |
f67539c2 | 17155 | r = -ENOTEMPTY; |
7c673cae FG |
17156 | } |
17157 | } | |
17158 | } | |
f67539c2 | 17159 | out: |
7c673cae FG |
17160 | dout(10) << __func__ << " " << cid << " = " << r << dendl; |
17161 | return r; | |
17162 | } | |
17163 | ||
11fdf7f2 TL |
17164 | void BlueStore::_do_remove_collection(TransContext *txc, |
17165 | CollectionRef *c) | |
17166 | { | |
17167 | coll_map.erase((*c)->cid); | |
17168 | txc->removed_collections.push_back(*c); | |
17169 | (*c)->exists = false; | |
17170 | _osr_register_zombie((*c)->osr.get()); | |
17171 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
17172 | c->reset(); | |
17173 | } | |
17174 | ||
7c673cae FG |
17175 | int BlueStore::_split_collection(TransContext *txc, |
17176 | CollectionRef& c, | |
17177 | CollectionRef& d, | |
17178 | unsigned bits, int rem) | |
17179 | { | |
17180 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
17181 | << " bits " << bits << dendl; | |
9f95a23c TL |
17182 | std::unique_lock l(c->lock); |
17183 | std::unique_lock l2(d->lock); | |
7c673cae FG |
17184 | int r; |
17185 | ||
17186 | // flush all previous deferred writes on this sequencer. this is a bit | |
17187 | // heavyweight, but we need to make sure all deferred writes complete | |
17188 | // before we split as the new collection's sequencer may need to order | |
17189 | // this after those writes, and we don't bother with the complexity of | |
17190 | // moving those TransContexts over to the new osr. | |
17191 | _osr_drain_preceding(txc); | |
17192 | ||
17193 | // move any cached items (onodes and referenced shared blobs) that will | |
17194 | // belong to the child collection post-split. leave everything else behind. | |
17195 | // this may include things that don't strictly belong to the now-smaller | |
17196 | // parent split, but the OSD will always send us a split for every new | |
17197 | // child. | |
17198 | ||
17199 | spg_t pgid, dest_pgid; | |
17200 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 17201 | ceph_assert(is_pg); |
7c673cae | 17202 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 17203 | ceph_assert(is_pg); |
7c673cae FG |
17204 | |
17205 | // the destination should initially be empty. | |
39ae355f | 17206 | ceph_assert(d->onode_space.empty()); |
11fdf7f2 TL |
17207 | ceph_assert(d->shared_blob_set.empty()); |
17208 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
17209 | |
17210 | c->split_cache(d.get()); | |
17211 | ||
17212 | // adjust bits. note that this will be redundant for all but the first | |
17213 | // split call for this parent (first child). | |
17214 | c->cnode.bits = bits; | |
11fdf7f2 | 17215 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
17216 | r = 0; |
17217 | ||
17218 | bufferlist bl; | |
11fdf7f2 | 17219 | encode(c->cnode, bl); |
7c673cae FG |
17220 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
17221 | ||
17222 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
17223 | << " bits " << bits << " = " << r << dendl; | |
17224 | return r; | |
17225 | } | |
17226 | ||
11fdf7f2 TL |
17227 | int BlueStore::_merge_collection( |
17228 | TransContext *txc, | |
17229 | CollectionRef *c, | |
17230 | CollectionRef& d, | |
17231 | unsigned bits) | |
17232 | { | |
17233 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
17234 | << " bits " << bits << dendl; | |
9f95a23c TL |
17235 | std::unique_lock l((*c)->lock); |
17236 | std::unique_lock l2(d->lock); | |
11fdf7f2 TL |
17237 | int r; |
17238 | ||
17239 | coll_t cid = (*c)->cid; | |
17240 | ||
17241 | // flush all previous deferred writes on the source collection to ensure | |
17242 | // that all deferred writes complete before we merge as the target collection's | |
17243 | // sequencer may need to order new ops after those writes. | |
17244 | ||
17245 | _osr_drain((*c)->osr.get()); | |
17246 | ||
17247 | // move any cached items (onodes and referenced shared blobs) that will | |
17248 | // belong to the child collection post-split. leave everything else behind. | |
17249 | // this may include things that don't strictly belong to the now-smaller | |
17250 | // parent split, but the OSD will always send us a split for every new | |
17251 | // child. | |
17252 | ||
17253 | spg_t pgid, dest_pgid; | |
17254 | bool is_pg = cid.is_pg(&pgid); | |
17255 | ceph_assert(is_pg); | |
17256 | is_pg = d->cid.is_pg(&dest_pgid); | |
17257 | ceph_assert(is_pg); | |
17258 | ||
17259 | // adjust bits. note that this will be redundant for all but the first | |
17260 | // merge call for the parent/target. | |
17261 | d->cnode.bits = bits; | |
17262 | ||
17263 | // behavior depends on target (d) bits, so this after that is updated. | |
17264 | (*c)->split_cache(d.get()); | |
17265 | ||
17266 | // remove source collection | |
17267 | { | |
9f95a23c | 17268 | std::unique_lock l3(coll_lock); |
11fdf7f2 TL |
17269 | _do_remove_collection(txc, c); |
17270 | } | |
17271 | ||
17272 | r = 0; | |
17273 | ||
17274 | bufferlist bl; | |
17275 | encode(d->cnode, bl); | |
17276 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
17277 | ||
17278 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
17279 | << " bits " << bits << " = " << r << dendl; | |
17280 | return r; | |
17281 | } | |
17282 | ||
494da23a TL |
17283 | void BlueStore::log_latency( |
17284 | const char* name, | |
17285 | int idx, | |
17286 | const ceph::timespan& l, | |
17287 | double lat_threshold, | |
17288 | const char* info) const | |
17289 | { | |
17290 | logger->tinc(idx, l); | |
17291 | if (lat_threshold > 0.0 && | |
17292 | l >= make_timespan(lat_threshold)) { | |
17293 | dout(0) << __func__ << " slow operation observed for " << name | |
17294 | << ", latency = " << l | |
17295 | << info | |
17296 | << dendl; | |
17297 | } | |
17298 | } | |
17299 | ||
11fdf7f2 | 17300 | void BlueStore::log_latency_fn( |
494da23a | 17301 | const char* name, |
11fdf7f2 TL |
17302 | int idx, |
17303 | const ceph::timespan& l, | |
494da23a TL |
17304 | double lat_threshold, |
17305 | std::function<string (const ceph::timespan& lat)> fn) const | |
11fdf7f2 | 17306 | { |
494da23a TL |
17307 | logger->tinc(idx, l); |
17308 | if (lat_threshold > 0.0 && | |
17309 | l >= make_timespan(lat_threshold)) { | |
17310 | dout(0) << __func__ << " slow operation observed for " << name | |
17311 | << ", latency = " << l | |
17312 | << fn(l) | |
17313 | << dendl; | |
17314 | } | |
11fdf7f2 TL |
17315 | } |
17316 | ||
9f95a23c TL |
17317 | #if defined(WITH_LTTNG) |
17318 | void BlueStore::BlueStoreThrottle::emit_initial_tracepoint( | |
17319 | KeyValueDB &db, | |
17320 | TransContext &txc, | |
17321 | mono_clock::time_point start_throttle_acquire) | |
17322 | { | |
17323 | pending_kv_ios += txc.ios; | |
17324 | if (txc.deferred_txn) { | |
17325 | pending_deferred_ios += txc.ios; | |
17326 | } | |
17327 | ||
17328 | uint64_t started = 0; | |
17329 | uint64_t completed = 0; | |
17330 | if (should_trace(&started, &completed)) { | |
17331 | txc.tracing = true; | |
17332 | uint64_t rocksdb_base_level, | |
17333 | rocksdb_estimate_pending_compaction_bytes, | |
17334 | rocksdb_cur_size_all_mem_tables, | |
17335 | rocksdb_compaction_pending, | |
17336 | rocksdb_mem_table_flush_pending, | |
17337 | rocksdb_num_running_compactions, | |
17338 | rocksdb_num_running_flushes, | |
17339 | rocksdb_actual_delayed_write_rate; | |
17340 | db.get_property( | |
17341 | "rocksdb.base-level", | |
17342 | &rocksdb_base_level); | |
17343 | db.get_property( | |
17344 | "rocksdb.estimate-pending-compaction-bytes", | |
17345 | &rocksdb_estimate_pending_compaction_bytes); | |
17346 | db.get_property( | |
17347 | "rocksdb.cur-size-all-mem-tables", | |
17348 | &rocksdb_cur_size_all_mem_tables); | |
17349 | db.get_property( | |
17350 | "rocksdb.compaction-pending", | |
17351 | &rocksdb_compaction_pending); | |
17352 | db.get_property( | |
17353 | "rocksdb.mem-table-flush-pending", | |
17354 | &rocksdb_mem_table_flush_pending); | |
17355 | db.get_property( | |
17356 | "rocksdb.num-running-compactions", | |
17357 | &rocksdb_num_running_compactions); | |
17358 | db.get_property( | |
17359 | "rocksdb.num-running-flushes", | |
17360 | &rocksdb_num_running_flushes); | |
17361 | db.get_property( | |
17362 | "rocksdb.actual-delayed-write-rate", | |
17363 | &rocksdb_actual_delayed_write_rate); | |
17364 | ||
17365 | ||
17366 | tracepoint( | |
17367 | bluestore, | |
17368 | transaction_initial_state, | |
17369 | txc.osr->get_sequencer_id(), | |
17370 | txc.seq, | |
17371 | throttle_bytes.get_current(), | |
17372 | throttle_deferred_bytes.get_current(), | |
17373 | pending_kv_ios, | |
17374 | pending_deferred_ios, | |
17375 | started, | |
17376 | completed, | |
17377 | ceph::to_seconds<double>(mono_clock::now() - start_throttle_acquire)); | |
17378 | ||
17379 | tracepoint( | |
17380 | bluestore, | |
17381 | transaction_initial_state_rocksdb, | |
17382 | txc.osr->get_sequencer_id(), | |
17383 | txc.seq, | |
17384 | rocksdb_base_level, | |
17385 | rocksdb_estimate_pending_compaction_bytes, | |
17386 | rocksdb_cur_size_all_mem_tables, | |
17387 | rocksdb_compaction_pending, | |
17388 | rocksdb_mem_table_flush_pending, | |
17389 | rocksdb_num_running_compactions, | |
17390 | rocksdb_num_running_flushes, | |
17391 | rocksdb_actual_delayed_write_rate); | |
17392 | } | |
17393 | } | |
17394 | #endif | |
17395 | ||
17396 | mono_clock::duration BlueStore::BlueStoreThrottle::log_state_latency( | |
17397 | TransContext &txc, PerfCounters *logger, int state) | |
17398 | { | |
17399 | mono_clock::time_point now = mono_clock::now(); | |
17400 | mono_clock::duration lat = now - txc.last_stamp; | |
17401 | logger->tinc(state, lat); | |
17402 | #if defined(WITH_LTTNG) | |
17403 | if (txc.tracing && | |
17404 | state >= l_bluestore_state_prepare_lat && | |
17405 | state <= l_bluestore_state_done_lat) { | |
17406 | OID_ELAPSED("", lat.to_nsec() / 1000.0, txc.get_state_latency_name(state)); | |
17407 | tracepoint( | |
17408 | bluestore, | |
17409 | transaction_state_duration, | |
17410 | txc.osr->get_sequencer_id(), | |
17411 | txc.seq, | |
17412 | state, | |
17413 | ceph::to_seconds<double>(lat)); | |
17414 | } | |
17415 | #endif | |
17416 | txc.last_stamp = now; | |
17417 | return lat; | |
17418 | } | |
17419 | ||
17420 | bool BlueStore::BlueStoreThrottle::try_start_transaction( | |
17421 | KeyValueDB &db, | |
17422 | TransContext &txc, | |
17423 | mono_clock::time_point start_throttle_acquire) | |
17424 | { | |
17425 | throttle_bytes.get(txc.cost); | |
17426 | ||
17427 | if (!txc.deferred_txn || throttle_deferred_bytes.get_or_fail(txc.cost)) { | |
17428 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
17429 | return true; | |
17430 | } else { | |
17431 | return false; | |
17432 | } | |
17433 | } | |
17434 | ||
17435 | void BlueStore::BlueStoreThrottle::finish_start_transaction( | |
17436 | KeyValueDB &db, | |
17437 | TransContext &txc, | |
17438 | mono_clock::time_point start_throttle_acquire) | |
17439 | { | |
17440 | ceph_assert(txc.deferred_txn); | |
17441 | throttle_deferred_bytes.get(txc.cost); | |
17442 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
17443 | } | |
17444 | ||
17445 | #if defined(WITH_LTTNG) | |
17446 | void BlueStore::BlueStoreThrottle::complete_kv(TransContext &txc) | |
17447 | { | |
17448 | pending_kv_ios -= 1; | |
17449 | ios_completed_since_last_traced++; | |
17450 | if (txc.tracing) { | |
17451 | tracepoint( | |
17452 | bluestore, | |
17453 | transaction_commit_latency, | |
17454 | txc.osr->get_sequencer_id(), | |
17455 | txc.seq, | |
17456 | ceph::to_seconds<double>(mono_clock::now() - txc.start)); | |
17457 | } | |
17458 | } | |
17459 | #endif | |
17460 | ||
17461 | #if defined(WITH_LTTNG) | |
17462 | void BlueStore::BlueStoreThrottle::complete(TransContext &txc) | |
17463 | { | |
17464 | if (txc.deferred_txn) { | |
17465 | pending_deferred_ios -= 1; | |
17466 | } | |
17467 | if (txc.tracing) { | |
17468 | mono_clock::time_point now = mono_clock::now(); | |
17469 | mono_clock::duration lat = now - txc.start; | |
17470 | tracepoint( | |
17471 | bluestore, | |
17472 | transaction_total_duration, | |
17473 | txc.osr->get_sequencer_id(), | |
17474 | txc.seq, | |
17475 | ceph::to_seconds<double>(lat)); | |
17476 | } | |
17477 | } | |
17478 | #endif | |
11fdf7f2 | 17479 | |
7c673cae FG |
17480 | const string prefix_onode = "o"; |
17481 | const string prefix_onode_shard = "x"; | |
17482 | const string prefix_other = "Z"; | |
7c673cae FG |
17483 | //Itrerates through the db and collects the stats |
17484 | void BlueStore::generate_db_histogram(Formatter *f) | |
17485 | { | |
17486 | //globals | |
17487 | uint64_t num_onodes = 0; | |
17488 | uint64_t num_shards = 0; | |
17489 | uint64_t num_super = 0; | |
17490 | uint64_t num_coll = 0; | |
17491 | uint64_t num_omap = 0; | |
11fdf7f2 | 17492 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
17493 | uint64_t num_deferred = 0; |
17494 | uint64_t num_alloc = 0; | |
17495 | uint64_t num_stat = 0; | |
17496 | uint64_t num_others = 0; | |
17497 | uint64_t num_shared_shards = 0; | |
17498 | size_t max_key_size =0, max_value_size = 0; | |
17499 | uint64_t total_key_size = 0, total_value_size = 0; | |
17500 | size_t key_size = 0, value_size = 0; | |
20effc67 | 17501 | KeyValueHistogram hist; |
7c673cae | 17502 | |
11fdf7f2 | 17503 | auto start = coarse_mono_clock::now(); |
7c673cae | 17504 | |
11fdf7f2 | 17505 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
17506 | iter->seek_to_first(); |
17507 | while (iter->valid()) { | |
17508 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
17509 | key_size = iter->key_size(); | |
17510 | value_size = iter->value_size(); | |
17511 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
17512 | max_key_size = std::max(max_key_size, key_size); |
17513 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
17514 | total_key_size += key_size; |
17515 | total_value_size += value_size; | |
17516 | ||
17517 | pair<string,string> key(iter->raw_key()); | |
17518 | ||
17519 | if (key.first == PREFIX_SUPER) { | |
17520 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
17521 | num_super++; | |
17522 | } else if (key.first == PREFIX_STAT) { | |
17523 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
17524 | num_stat++; | |
17525 | } else if (key.first == PREFIX_COLL) { | |
17526 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
17527 | num_coll++; | |
17528 | } else if (key.first == PREFIX_OBJ) { | |
17529 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
17530 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
17531 | num_onodes++; | |
17532 | } else { | |
17533 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
17534 | num_shards++; | |
17535 | } | |
17536 | } else if (key.first == PREFIX_OMAP) { | |
17537 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
17538 | num_omap++; | |
f67539c2 TL |
17539 | } else if (key.first == PREFIX_PERPOOL_OMAP) { |
17540 | hist.update_hist_entry(hist.key_hist, PREFIX_PERPOOL_OMAP, key_size, value_size); | |
17541 | num_omap++; | |
17542 | } else if (key.first == PREFIX_PERPG_OMAP) { | |
17543 | hist.update_hist_entry(hist.key_hist, PREFIX_PERPG_OMAP, key_size, value_size); | |
17544 | num_omap++; | |
11fdf7f2 TL |
17545 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
17546 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
17547 | num_pgmeta_omap++; | |
7c673cae FG |
17548 | } else if (key.first == PREFIX_DEFERRED) { |
17549 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
17550 | num_deferred++; | |
11fdf7f2 | 17551 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
17552 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
17553 | num_alloc++; | |
17554 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
17555 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
17556 | num_shared_shards++; | |
17557 | } else { | |
17558 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
17559 | num_others++; | |
17560 | } | |
17561 | iter->next(); | |
17562 | } | |
17563 | ||
11fdf7f2 | 17564 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
17565 | f->open_object_section("rocksdb_key_value_stats"); |
17566 | f->dump_unsigned("num_onodes", num_onodes); | |
17567 | f->dump_unsigned("num_shards", num_shards); | |
17568 | f->dump_unsigned("num_super", num_super); | |
17569 | f->dump_unsigned("num_coll", num_coll); | |
17570 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 17571 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
17572 | f->dump_unsigned("num_deferred", num_deferred); |
17573 | f->dump_unsigned("num_alloc", num_alloc); | |
17574 | f->dump_unsigned("num_stat", num_stat); | |
17575 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
17576 | f->dump_unsigned("num_others", num_others); | |
17577 | f->dump_unsigned("max_key_size", max_key_size); | |
17578 | f->dump_unsigned("max_value_size", max_value_size); | |
17579 | f->dump_unsigned("total_key_size", total_key_size); | |
17580 | f->dump_unsigned("total_value_size", total_value_size); | |
17581 | f->close_section(); | |
17582 | ||
17583 | hist.dump(f); | |
17584 | ||
17585 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
17586 | ||
17587 | } | |
17588 | ||
f6b5b4d7 | 17589 | void BlueStore::_shutdown_cache() |
7c673cae FG |
17590 | { |
17591 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
17592 | for (auto i : buffer_cache_shards) { |
17593 | i->flush(); | |
11fdf7f2 | 17594 | ceph_assert(i->empty()); |
7c673cae FG |
17595 | } |
17596 | for (auto& p : coll_map) { | |
39ae355f | 17597 | p.second->onode_space.clear(); |
3efd9988 FG |
17598 | if (!p.second->shared_blob_set.empty()) { |
17599 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 17600 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 17601 | } |
39ae355f | 17602 | ceph_assert(p.second->onode_space.empty()); |
11fdf7f2 | 17603 | ceph_assert(p.second->shared_blob_set.empty()); |
7c673cae FG |
17604 | } |
17605 | coll_map.clear(); | |
f6b5b4d7 TL |
17606 | for (auto i : onode_cache_shards) { |
17607 | ceph_assert(i->empty()); | |
17608 | } | |
7c673cae FG |
17609 | } |
17610 | ||
31f18b77 FG |
17611 | // For external caller. |
17612 | // We use a best-effort policy instead, e.g., | |
17613 | // we don't care if there are still some pinned onodes/data in the cache | |
17614 | // after this command is completed. | |
11fdf7f2 | 17615 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
17616 | { |
17617 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
17618 | for (auto i : onode_cache_shards) { |
17619 | i->flush(); | |
17620 | } | |
17621 | for (auto i : buffer_cache_shards) { | |
17622 | i->flush(); | |
31f18b77 | 17623 | } |
11fdf7f2 TL |
17624 | |
17625 | return 0; | |
31f18b77 FG |
17626 | } |
17627 | ||
7c673cae FG |
17628 | void BlueStore::_apply_padding(uint64_t head_pad, |
17629 | uint64_t tail_pad, | |
7c673cae FG |
17630 | bufferlist& padded) |
17631 | { | |
7c673cae | 17632 | if (head_pad) { |
224ce89b | 17633 | padded.prepend_zero(head_pad); |
7c673cae FG |
17634 | } |
17635 | if (tail_pad) { | |
17636 | padded.append_zero(tail_pad); | |
17637 | } | |
17638 | if (head_pad || tail_pad) { | |
17639 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
17640 | << " tail 0x" << tail_pad << std::dec << dendl; | |
17641 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
17642 | } | |
17643 | } | |
17644 | ||
39ae355f | 17645 | void BlueStore::_record_onode(OnodeRef& o, KeyValueDB::Transaction &txn) |
11fdf7f2 TL |
17646 | { |
17647 | // finalize extent_map shards | |
17648 | o->extent_map.update(txn, false); | |
17649 | if (o->extent_map.needs_reshard()) { | |
17650 | o->extent_map.reshard(db, txn); | |
17651 | o->extent_map.update(txn, true); | |
17652 | if (o->extent_map.needs_reshard()) { | |
17653 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
17654 | << dendl; | |
17655 | o->extent_map.clear_needs_reshard(); | |
17656 | } | |
17657 | logger->inc(l_bluestore_onode_reshard); | |
17658 | } | |
17659 | ||
17660 | // bound encode | |
17661 | size_t bound = 0; | |
17662 | denc(o->onode, bound); | |
17663 | o->extent_map.bound_encode_spanning_blobs(bound); | |
17664 | if (o->onode.extent_map_shards.empty()) { | |
17665 | denc(o->extent_map.inline_bl, bound); | |
17666 | } | |
17667 | ||
17668 | // encode | |
17669 | bufferlist bl; | |
17670 | unsigned onode_part, blob_part, extent_part; | |
17671 | { | |
17672 | auto p = bl.get_contiguous_appender(bound, true); | |
17673 | denc(o->onode, p); | |
17674 | onode_part = p.get_logical_offset(); | |
17675 | o->extent_map.encode_spanning_blobs(p); | |
17676 | blob_part = p.get_logical_offset() - onode_part; | |
17677 | if (o->onode.extent_map_shards.empty()) { | |
17678 | denc(o->extent_map.inline_bl, p); | |
17679 | } | |
17680 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
17681 | } | |
17682 | ||
17683 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
17684 | << " (" << onode_part << " bytes onode + " | |
17685 | << blob_part << " bytes spanning blobs + " | |
17686 | << extent_part << " bytes inline extents)" | |
17687 | << dendl; | |
17688 | ||
17689 | ||
17690 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
17691 | } | |
17692 | ||
17693 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
17694 | { | |
17695 | std::lock_guard l(qlock); | |
1e59de90 TL |
17696 | size_t used = bluefs && bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW ? |
17697 | bluefs->get_used(BlueFS::BDEV_SLOW) : 0; | |
17698 | if (used > 0) { | |
17699 | auto db_used = bluefs->get_used(BlueFS::BDEV_DB); | |
17700 | auto db_total = bluefs->get_total(BlueFS::BDEV_DB); | |
17701 | ostringstream ss; | |
17702 | ss << "spilled over " << byte_u_t(used) | |
17703 | << " metadata from 'db' device (" << byte_u_t(db_used) | |
17704 | << " used of " << byte_u_t(db_total) << ") to slow device"; | |
17705 | spillover_alert = ss.str(); | |
17706 | } else if (!spillover_alert.empty()){ | |
17707 | spillover_alert.clear(); | |
17708 | } | |
11fdf7f2 | 17709 | |
522d829b TL |
17710 | if (!spurious_read_errors_alert.empty() && |
17711 | cct->_conf->bluestore_warn_on_spurious_read_errors) { | |
f67539c2 TL |
17712 | alerts.emplace( |
17713 | "BLUESTORE_SPURIOUS_READ_ERRORS", | |
17714 | spurious_read_errors_alert); | |
17715 | } | |
81eedcae TL |
17716 | if (!disk_size_mismatch_alert.empty()) { |
17717 | alerts.emplace( | |
17718 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
17719 | disk_size_mismatch_alert); | |
17720 | } | |
17721 | if (!legacy_statfs_alert.empty()) { | |
17722 | alerts.emplace( | |
17723 | "BLUESTORE_LEGACY_STATFS", | |
17724 | legacy_statfs_alert); | |
17725 | } | |
11fdf7f2 TL |
17726 | if (!spillover_alert.empty() && |
17727 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
17728 | alerts.emplace( | |
17729 | "BLUEFS_SPILLOVER", | |
17730 | spillover_alert); | |
17731 | } | |
f67539c2 TL |
17732 | if (!no_per_pg_omap_alert.empty()) { |
17733 | alerts.emplace( | |
17734 | "BLUESTORE_NO_PER_PG_OMAP", | |
17735 | no_per_pg_omap_alert); | |
17736 | } | |
9f95a23c TL |
17737 | if (!no_per_pool_omap_alert.empty()) { |
17738 | alerts.emplace( | |
17739 | "BLUESTORE_NO_PER_POOL_OMAP", | |
17740 | no_per_pool_omap_alert); | |
17741 | } | |
11fdf7f2 TL |
17742 | string s0(failed_cmode); |
17743 | ||
17744 | if (!failed_compressors.empty()) { | |
17745 | if (!s0.empty()) { | |
17746 | s0 += ", "; | |
17747 | } | |
17748 | s0 += "unable to load:"; | |
17749 | bool first = true; | |
17750 | for (auto& s : failed_compressors) { | |
17751 | if (first) { | |
17752 | first = false; | |
17753 | } else { | |
17754 | s0 += ", "; | |
17755 | } | |
17756 | s0 += s; | |
17757 | } | |
17758 | alerts.emplace( | |
17759 | "BLUESTORE_NO_COMPRESSION", | |
17760 | s0); | |
17761 | } | |
17762 | } | |
17763 | ||
9f95a23c | 17764 | void BlueStore::_collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
20effc67 | 17765 | const PExtentVector& extents) |
9f95a23c TL |
17766 | { |
17767 | alloc_stats_count++; | |
20effc67 | 17768 | alloc_stats_fragments += extents.size(); |
9f95a23c | 17769 | alloc_stats_size += need; |
20effc67 TL |
17770 | |
17771 | for (auto& e : extents) { | |
17772 | logger->hinc(l_bluestore_allocate_hist, e.length, need); | |
17773 | } | |
9f95a23c TL |
17774 | } |
17775 | ||
17776 | void BlueStore::_record_allocation_stats() | |
17777 | { | |
17778 | // don't care about data consistency, | |
17779 | // fields can be partially modified while making the tuple | |
17780 | auto t0 = std::make_tuple( | |
17781 | alloc_stats_count.exchange(0), | |
17782 | alloc_stats_fragments.exchange(0), | |
17783 | alloc_stats_size.exchange(0)); | |
17784 | ||
17785 | dout(0) << " allocation stats probe " | |
17786 | << probe_count << ":" | |
17787 | << " cnt: " << std::get<0>(t0) | |
17788 | << " frags: " << std::get<1>(t0) | |
17789 | << " size: " << std::get<2>(t0) | |
17790 | << dendl; | |
17791 | ||
17792 | ||
17793 | // | |
17794 | // Keep the history for probes from the power-of-two sequence: | |
17795 | // -1, -2, -4, -8, -16 | |
17796 | // | |
17797 | size_t base = 1; | |
17798 | for (auto& t : alloc_stats_history) { | |
17799 | dout(0) << " probe -" | |
17800 | << base + (probe_count % base) << ": " | |
17801 | << std::get<0>(t) | |
17802 | << ", " << std::get<1>(t) | |
17803 | << ", " << std::get<2>(t) | |
17804 | << dendl; | |
17805 | base <<= 1; | |
17806 | } | |
17807 | dout(0) << "------------" << dendl; | |
17808 | ||
f67539c2 | 17809 | ++ probe_count; |
9f95a23c | 17810 | |
f67539c2 TL |
17811 | for (ssize_t i = alloc_stats_history.size() - 1 ; i > 0 ; --i) { |
17812 | if ((probe_count % (1 << i)) == 0) { | |
17813 | alloc_stats_history[i] = alloc_stats_history[i - 1]; | |
17814 | } | |
9f95a23c TL |
17815 | } |
17816 | alloc_stats_history[0].swap(t0); | |
17817 | } | |
17818 | ||
7c673cae | 17819 | // =========================================== |
11fdf7f2 TL |
17820 | // BlueStoreRepairer |
17821 | ||
17822 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
17823 | const interval_set<uint64_t>& extents) | |
17824 | { | |
17825 | ceph_assert(granularity); // initialized | |
17826 | // can't call for the second time | |
17827 | ceph_assert(!was_filtered_out); | |
17828 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
17829 | ||
17830 | uint64_t prev_pos = 0; | |
17831 | uint64_t npos = collections_bfs.size(); | |
17832 | ||
17833 | bloom_vector collections_reduced; | |
17834 | bloom_vector objects_reduced; | |
17835 | ||
17836 | for (auto e : extents) { | |
17837 | if (e.second == 0) { | |
17838 | continue; | |
17839 | } | |
17840 | uint64_t pos = max(e.first / granularity, prev_pos); | |
17841 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
17842 | while (pos != npos && pos < end_pos) { | |
17843 | ceph_assert( collections_bfs[pos].element_count() == | |
17844 | objects_bfs[pos].element_count()); | |
17845 | if (collections_bfs[pos].element_count()) { | |
17846 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
17847 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
17848 | } | |
17849 | ++pos; | |
17850 | } | |
17851 | prev_pos = end_pos; | |
17852 | } | |
17853 | collections_reduced.swap(collections_bfs); | |
17854 | objects_reduced.swap(objects_bfs); | |
17855 | was_filtered_out = true; | |
17856 | return collections_bfs.size(); | |
17857 | } | |
17858 | ||
17859 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
17860 | const string& prefix, | |
17861 | const string& key) | |
17862 | { | |
b3b6e05e | 17863 | std::lock_guard l(lock); |
11fdf7f2 TL |
17864 | if (!remove_key_txn) { |
17865 | remove_key_txn = db->get_transaction(); | |
17866 | } | |
17867 | ++to_repair_cnt; | |
17868 | remove_key_txn->rmkey(prefix, key); | |
17869 | ||
17870 | return true; | |
17871 | } | |
17872 | ||
f67539c2 | 17873 | void BlueStoreRepairer::fix_per_pool_omap(KeyValueDB *db, int val) |
9f95a23c | 17874 | { |
b3b6e05e TL |
17875 | std::lock_guard l(lock); // possibly redundant |
17876 | ceph_assert(fix_per_pool_omap_txn == nullptr); | |
9f95a23c TL |
17877 | fix_per_pool_omap_txn = db->get_transaction(); |
17878 | ++to_repair_cnt; | |
17879 | bufferlist bl; | |
f67539c2 | 17880 | bl.append(stringify(val)); |
9f95a23c TL |
17881 | fix_per_pool_omap_txn->set(PREFIX_SUPER, "per_pool_omap", bl); |
17882 | } | |
17883 | ||
11fdf7f2 | 17884 | bool BlueStoreRepairer::fix_shared_blob( |
20effc67 | 17885 | KeyValueDB::Transaction txn, |
11fdf7f2 | 17886 | uint64_t sbid, |
20effc67 TL |
17887 | bluestore_extent_ref_map_t* ref_map, |
17888 | size_t repaired) | |
11fdf7f2 | 17889 | { |
11fdf7f2 TL |
17890 | string key; |
17891 | get_shared_blob_key(sbid, &key); | |
20effc67 TL |
17892 | if (ref_map) { |
17893 | bluestore_shared_blob_t persistent(sbid, std::move(*ref_map)); | |
17894 | bufferlist bl; | |
17895 | encode(persistent, bl); | |
17896 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
11fdf7f2 TL |
17897 | } else { |
17898 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
17899 | } | |
20effc67 | 17900 | to_repair_cnt += repaired; |
11fdf7f2 TL |
17901 | return true; |
17902 | } | |
17903 | ||
17904 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
17905 | const string& key, | |
17906 | const store_statfs_t& new_statfs) | |
17907 | { | |
b3b6e05e | 17908 | std::lock_guard l(lock); |
11fdf7f2 TL |
17909 | if (!fix_statfs_txn) { |
17910 | fix_statfs_txn = db->get_transaction(); | |
17911 | } | |
17912 | BlueStore::volatile_statfs vstatfs; | |
17913 | vstatfs = new_statfs; | |
17914 | bufferlist bl; | |
17915 | vstatfs.encode(bl); | |
17916 | ++to_repair_cnt; | |
17917 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
17918 | return true; | |
17919 | } | |
17920 | ||
17921 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
17922 | FreelistManager* fm, | |
17923 | uint64_t offset, uint64_t len) | |
17924 | { | |
b3b6e05e | 17925 | std::lock_guard l(lock); |
20effc67 TL |
17926 | ceph_assert(!fm->is_null_manager()); |
17927 | ||
11fdf7f2 TL |
17928 | if (!fix_fm_leaked_txn) { |
17929 | fix_fm_leaked_txn = db->get_transaction(); | |
17930 | } | |
17931 | ++to_repair_cnt; | |
17932 | fm->release(offset, len, fix_fm_leaked_txn); | |
17933 | return true; | |
17934 | } | |
17935 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
17936 | FreelistManager* fm, | |
17937 | uint64_t offset, uint64_t len) | |
17938 | { | |
b3b6e05e | 17939 | std::lock_guard l(lock); |
20effc67 TL |
17940 | ceph_assert(!fm->is_null_manager()); |
17941 | ||
11fdf7f2 TL |
17942 | if (!fix_fm_false_free_txn) { |
17943 | fix_fm_false_free_txn = db->get_transaction(); | |
17944 | } | |
17945 | ++to_repair_cnt; | |
17946 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
17947 | return true; | |
17948 | } | |
17949 | ||
b3b6e05e TL |
17950 | bool BlueStoreRepairer::fix_spanning_blobs( |
17951 | KeyValueDB* db, | |
17952 | std::function<void(KeyValueDB::Transaction)> f) | |
adb31ebb | 17953 | { |
b3b6e05e | 17954 | std::lock_guard l(lock); |
adb31ebb TL |
17955 | if (!fix_onode_txn) { |
17956 | fix_onode_txn = db->get_transaction(); | |
17957 | } | |
b3b6e05e | 17958 | f(fix_onode_txn); |
adb31ebb | 17959 | ++to_repair_cnt; |
b3b6e05e | 17960 | return true; |
adb31ebb TL |
17961 | } |
17962 | ||
11fdf7f2 TL |
17963 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) |
17964 | { | |
b3b6e05e | 17965 | //NB: not for use in multithreading mode!!! |
11fdf7f2 TL |
17966 | if (misreferenced_extents.size()) { |
17967 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
17968 | ceph_assert(n > 0); | |
17969 | if (!fix_misreferences_txn) { | |
17970 | fix_misreferences_txn = db->get_transaction(); | |
17971 | } | |
17972 | return true; | |
17973 | } | |
17974 | return false; | |
17975 | } | |
17976 | ||
17977 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
17978 | { | |
b3b6e05e | 17979 | //NB: not for use in multithreading mode!!! |
9f95a23c | 17980 | if (fix_per_pool_omap_txn) { |
20effc67 TL |
17981 | auto ok = db->submit_transaction_sync(fix_per_pool_omap_txn) == 0; |
17982 | ceph_assert(ok); | |
9f95a23c TL |
17983 | fix_per_pool_omap_txn = nullptr; |
17984 | } | |
11fdf7f2 | 17985 | if (fix_fm_leaked_txn) { |
20effc67 TL |
17986 | auto ok = db->submit_transaction_sync(fix_fm_leaked_txn) == 0; |
17987 | ceph_assert(ok); | |
11fdf7f2 TL |
17988 | fix_fm_leaked_txn = nullptr; |
17989 | } | |
17990 | if (fix_fm_false_free_txn) { | |
20effc67 TL |
17991 | auto ok = db->submit_transaction_sync(fix_fm_false_free_txn) == 0; |
17992 | ceph_assert(ok); | |
11fdf7f2 TL |
17993 | fix_fm_false_free_txn = nullptr; |
17994 | } | |
17995 | if (remove_key_txn) { | |
20effc67 TL |
17996 | auto ok = db->submit_transaction_sync(remove_key_txn) == 0; |
17997 | ceph_assert(ok); | |
11fdf7f2 TL |
17998 | remove_key_txn = nullptr; |
17999 | } | |
18000 | if (fix_misreferences_txn) { | |
20effc67 TL |
18001 | auto ok = db->submit_transaction_sync(fix_misreferences_txn) == 0; |
18002 | ceph_assert(ok); | |
11fdf7f2 TL |
18003 | fix_misreferences_txn = nullptr; |
18004 | } | |
adb31ebb | 18005 | if (fix_onode_txn) { |
20effc67 TL |
18006 | auto ok = db->submit_transaction_sync(fix_onode_txn) == 0; |
18007 | ceph_assert(ok); | |
adb31ebb TL |
18008 | fix_onode_txn = nullptr; |
18009 | } | |
11fdf7f2 | 18010 | if (fix_shared_blob_txn) { |
20effc67 TL |
18011 | auto ok = db->submit_transaction_sync(fix_shared_blob_txn) == 0; |
18012 | ceph_assert(ok); | |
11fdf7f2 TL |
18013 | fix_shared_blob_txn = nullptr; |
18014 | } | |
11fdf7f2 | 18015 | if (fix_statfs_txn) { |
20effc67 TL |
18016 | auto ok = db->submit_transaction_sync(fix_statfs_txn) == 0; |
18017 | ceph_assert(ok); | |
11fdf7f2 TL |
18018 | fix_statfs_txn = nullptr; |
18019 | } | |
522d829b TL |
18020 | if (need_compact) { |
18021 | db->compact(); | |
18022 | need_compact = false; | |
18023 | } | |
11fdf7f2 TL |
18024 | unsigned repaired = to_repair_cnt; |
18025 | to_repair_cnt = 0; | |
18026 | return repaired; | |
18027 | } | |
18028 | ||
18029 | // ======================================================= | |
9f95a23c TL |
18030 | // RocksDBBlueFSVolumeSelector |
18031 | ||
18032 | uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) { | |
18033 | ceph_assert(h != nullptr); | |
18034 | uint64_t hint = reinterpret_cast<uint64_t>(h); | |
18035 | uint8_t res; | |
18036 | switch (hint) { | |
18037 | case LEVEL_SLOW: | |
18038 | res = BlueFS::BDEV_SLOW; | |
18039 | if (db_avail4slow > 0) { | |
18040 | // considering statically available db space vs. | |
18041 | // - observed maximums on DB dev for DB/WAL/UNSORTED data | |
18042 | // - observed maximum spillovers | |
18043 | uint64_t max_db_use = 0; // max db usage we potentially observed | |
f6b5b4d7 | 18044 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_LOG - LEVEL_FIRST); |
9f95a23c TL |
18045 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_WAL - LEVEL_FIRST); |
18046 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_DB - LEVEL_FIRST); | |
18047 | // this could go to db hence using it in the estimation | |
18048 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_SLOW, LEVEL_DB - LEVEL_FIRST); | |
18049 | ||
18050 | auto db_total = l_totals[LEVEL_DB - LEVEL_FIRST]; | |
18051 | uint64_t avail = min( | |
18052 | db_avail4slow, | |
18053 | max_db_use < db_total ? db_total - max_db_use : 0); | |
18054 | ||
18055 | // considering current DB dev usage for SLOW data | |
18056 | if (avail > per_level_per_dev_usage.at(BlueFS::BDEV_DB, LEVEL_SLOW - LEVEL_FIRST)) { | |
18057 | res = BlueFS::BDEV_DB; | |
18058 | } | |
18059 | } | |
18060 | break; | |
f6b5b4d7 | 18061 | case LEVEL_LOG: |
9f95a23c TL |
18062 | case LEVEL_WAL: |
18063 | res = BlueFS::BDEV_WAL; | |
18064 | break; | |
18065 | case LEVEL_DB: | |
18066 | default: | |
18067 | res = BlueFS::BDEV_DB; | |
18068 | break; | |
18069 | } | |
18070 | return res; | |
18071 | } | |
18072 | ||
18073 | void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const | |
18074 | { | |
a4b75251 TL |
18075 | auto db_size = l_totals[LEVEL_DB - LEVEL_FIRST]; |
18076 | res.emplace_back(base, db_size); | |
18077 | auto slow_size = l_totals[LEVEL_SLOW - LEVEL_FIRST]; | |
18078 | if (slow_size == 0) { | |
18079 | slow_size = db_size; | |
18080 | } | |
18081 | res.emplace_back(base + ".slow", slow_size); | |
9f95a23c TL |
18082 | } |
18083 | ||
b3b6e05e | 18084 | void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(std::string_view dirname) const { |
9f95a23c TL |
18085 | uint8_t res = LEVEL_DB; |
18086 | if (dirname.length() > 5) { | |
18087 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
18088 | // match up with bluestore. the slow device is always the second | |
18089 | // one (when a dedicated block.db device is present and used at | |
18090 | // bdev 0). the wal device is always last. | |
18091 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
18092 | res = LEVEL_SLOW; | |
18093 | } | |
18094 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
18095 | res = LEVEL_WAL; | |
18096 | } | |
18097 | } | |
18098 | return reinterpret_cast<void*>(res); | |
18099 | } | |
18100 | ||
18101 | void RocksDBBlueFSVolumeSelector::dump(ostream& sout) { | |
18102 | auto max_x = per_level_per_dev_usage.get_max_x(); | |
18103 | auto max_y = per_level_per_dev_usage.get_max_y(); | |
1e59de90 TL |
18104 | |
18105 | sout << "RocksDBBlueFSVolumeSelector Usage Matrix:" << std::endl; | |
f6b5b4d7 | 18106 | constexpr std::array<const char*, 8> names{ { |
9f95a23c TL |
18107 | "DEV/LEV", |
18108 | "WAL", | |
18109 | "DB", | |
18110 | "SLOW", | |
18111 | "*", | |
18112 | "*", | |
f6b5b4d7 TL |
18113 | "REAL", |
18114 | "FILES", | |
9f95a23c TL |
18115 | } }; |
18116 | const size_t width = 12; | |
18117 | for (size_t i = 0; i < names.size(); ++i) { | |
18118 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18119 | sout.width(width); | |
18120 | sout << names[i]; | |
18121 | } | |
18122 | sout << std::endl; | |
18123 | for (size_t l = 0; l < max_y; l++) { | |
18124 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18125 | sout.width(width); | |
18126 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
18127 | case LEVEL_LOG: |
18128 | sout << "LOG"; break; | |
9f95a23c TL |
18129 | case LEVEL_WAL: |
18130 | sout << "WAL"; break; | |
18131 | case LEVEL_DB: | |
18132 | sout << "DB"; break; | |
18133 | case LEVEL_SLOW: | |
18134 | sout << "SLOW"; break; | |
18135 | case LEVEL_MAX: | |
1e59de90 | 18136 | sout << "TOTAL"; break; |
9f95a23c | 18137 | } |
f6b5b4d7 | 18138 | for (size_t d = 0; d < max_x; d++) { |
9f95a23c TL |
18139 | sout.setf(std::ios::left, std::ios::adjustfield); |
18140 | sout.width(width); | |
18141 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(d, l))); | |
18142 | } | |
18143 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18144 | sout.width(width); | |
f6b5b4d7 | 18145 | sout << stringify(per_level_files[l]) << std::endl; |
9f95a23c TL |
18146 | } |
18147 | ceph_assert(max_x == per_level_per_dev_max.get_max_x()); | |
18148 | ceph_assert(max_y == per_level_per_dev_max.get_max_y()); | |
18149 | sout << "MAXIMUMS:" << std::endl; | |
18150 | for (size_t l = 0; l < max_y; l++) { | |
18151 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18152 | sout.width(width); | |
18153 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
18154 | case LEVEL_LOG: |
18155 | sout << "LOG"; break; | |
9f95a23c TL |
18156 | case LEVEL_WAL: |
18157 | sout << "WAL"; break; | |
18158 | case LEVEL_DB: | |
18159 | sout << "DB"; break; | |
18160 | case LEVEL_SLOW: | |
18161 | sout << "SLOW"; break; | |
18162 | case LEVEL_MAX: | |
1e59de90 | 18163 | sout << "TOTAL"; break; |
9f95a23c TL |
18164 | } |
18165 | for (size_t d = 0; d < max_x - 1; d++) { | |
18166 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18167 | sout.width(width); | |
18168 | sout << stringify(byte_u_t(per_level_per_dev_max.at(d, l))); | |
18169 | } | |
18170 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18171 | sout.width(width); | |
18172 | sout << stringify(byte_u_t(per_level_per_dev_max.at(max_x - 1, l))); | |
1e59de90 | 18173 | sout << std::endl; |
9f95a23c | 18174 | } |
1e59de90 TL |
18175 | string sizes[] = { |
18176 | ">> SIZE <<", | |
18177 | stringify(byte_u_t(l_totals[LEVEL_WAL - LEVEL_FIRST])), | |
18178 | stringify(byte_u_t(l_totals[LEVEL_DB - LEVEL_FIRST])), | |
18179 | stringify(byte_u_t(l_totals[LEVEL_SLOW - LEVEL_FIRST])), | |
18180 | }; | |
18181 | for (size_t i = 0; i < (sizeof(sizes) / sizeof(sizes[0])); i++) { | |
18182 | sout.setf(std::ios::left, std::ios::adjustfield); | |
18183 | sout.width(width); | |
18184 | sout << sizes[i]; | |
18185 | } | |
18186 | sout << std::endl; | |
9f95a23c | 18187 | } |
11fdf7f2 | 18188 | |
20effc67 TL |
18189 | BlueFSVolumeSelector* RocksDBBlueFSVolumeSelector::clone_empty() const { |
18190 | RocksDBBlueFSVolumeSelector* ns = | |
18191 | new RocksDBBlueFSVolumeSelector(0, 0, 0, | |
18192 | 0, 0, 0, | |
18193 | 0, 0, false); | |
18194 | return ns; | |
18195 | } | |
18196 | ||
18197 | bool RocksDBBlueFSVolumeSelector::compare(BlueFSVolumeSelector* other) { | |
18198 | RocksDBBlueFSVolumeSelector* o = dynamic_cast<RocksDBBlueFSVolumeSelector*>(other); | |
18199 | ceph_assert(o); | |
18200 | bool equal = true; | |
18201 | for (size_t x = 0; x < BlueFS::MAX_BDEV + 1; x++) { | |
18202 | for (size_t y = 0; y <LEVEL_MAX - LEVEL_FIRST + 1; y++) { | |
18203 | equal &= (per_level_per_dev_usage.at(x, y) == o->per_level_per_dev_usage.at(x, y)); | |
18204 | } | |
18205 | } | |
18206 | for (size_t t = 0; t < LEVEL_MAX - LEVEL_FIRST + 1; t++) { | |
18207 | equal &= (per_level_files[t] == o->per_level_files[t]); | |
18208 | } | |
18209 | return equal; | |
18210 | } | |
18211 | ||
9f95a23c | 18212 | // ======================================================= |
20effc67 TL |
18213 | |
18214 | //================================================================================================================ | |
18215 | // BlueStore is committing all allocation information (alloc/release) into RocksDB before the client Write is performed. | |
18216 | // This cause a delay in write path and add significant load to the CPU/Memory/Disk. | |
18217 | // The reason for the RocksDB updates is that it allows Ceph to survive any failure without losing the allocation state. | |
18218 | // | |
18219 | // We changed the code skiping RocksDB updates on allocation time and instead performing a full desatge of the allocator object | |
18220 | // with all the OSD allocation state in a single step during umount(). | |
18221 | // This change leads to a 25% increase in IOPS and reduced latency in small random-write workload, but exposes the system | |
18222 | // to losing allocation info in failure cases where we don't call umount. | |
18223 | // We add code to perform a full allocation-map rebuild from information stored inside the ONode which is used in failure cases. | |
18224 | // When we perform a graceful shutdown there is no need for recovery and we simply read the allocation-map from a flat file | |
18225 | // where we store the allocation-map during umount(). | |
18226 | //================================================================================================================ | |
18227 | ||
18228 | #undef dout_prefix | |
18229 | #define dout_prefix *_dout << "bluestore::NCB::" << __func__ << "::" | |
18230 | ||
18231 | static const std::string allocator_dir = "ALLOCATOR_NCB_DIR"; | |
18232 | static const std::string allocator_file = "ALLOCATOR_NCB_FILE"; | |
18233 | static uint32_t s_format_version = 0x01; // support future changes to allocator-map file | |
18234 | static uint32_t s_serial = 0x01; | |
18235 | ||
18236 | #if 1 | |
18237 | #define CEPHTOH_32 le32toh | |
18238 | #define CEPHTOH_64 le64toh | |
18239 | #define HTOCEPH_32 htole32 | |
18240 | #define HTOCEPH_64 htole64 | |
18241 | #else | |
18242 | // help debug the encode/decode by forcing alien format | |
18243 | #define CEPHTOH_32 be32toh | |
18244 | #define CEPHTOH_64 be64toh | |
18245 | #define HTOCEPH_32 htobe32 | |
18246 | #define HTOCEPH_64 htobe64 | |
18247 | #endif | |
18248 | ||
18249 | // 48 Bytes header for on-disk alloator image | |
18250 | const uint64_t ALLOCATOR_IMAGE_VALID_SIGNATURE = 0x1FACE0FF; | |
18251 | struct allocator_image_header { | |
18252 | uint32_t format_version; // 0x00 | |
18253 | uint32_t valid_signature; // 0x04 | |
18254 | utime_t timestamp; // 0x08 | |
18255 | uint32_t serial; // 0x10 | |
18256 | uint32_t pad[0x7]; // 0x14 | |
18257 | ||
18258 | allocator_image_header() { | |
18259 | memset((char*)this, 0, sizeof(allocator_image_header)); | |
18260 | } | |
18261 | ||
18262 | // create header in CEPH format | |
18263 | allocator_image_header(utime_t timestamp, uint32_t format_version, uint32_t serial) { | |
18264 | this->format_version = format_version; | |
18265 | this->timestamp = timestamp; | |
18266 | this->valid_signature = ALLOCATOR_IMAGE_VALID_SIGNATURE; | |
18267 | this->serial = serial; | |
18268 | memset(this->pad, 0, sizeof(this->pad)); | |
18269 | } | |
18270 | ||
18271 | friend std::ostream& operator<<(std::ostream& out, const allocator_image_header& header) { | |
18272 | out << "format_version = " << header.format_version << std::endl; | |
18273 | out << "valid_signature = " << header.valid_signature << "/" << ALLOCATOR_IMAGE_VALID_SIGNATURE << std::endl; | |
18274 | out << "timestamp = " << header.timestamp << std::endl; | |
18275 | out << "serial = " << header.serial << std::endl; | |
18276 | for (unsigned i = 0; i < sizeof(header.pad)/sizeof(uint32_t); i++) { | |
18277 | if (header.pad[i]) { | |
18278 | out << "header.pad[" << i << "] = " << header.pad[i] << std::endl; | |
18279 | } | |
18280 | } | |
18281 | return out; | |
18282 | } | |
18283 | ||
18284 | DENC(allocator_image_header, v, p) { | |
18285 | denc(v.format_version, p); | |
18286 | denc(v.valid_signature, p); | |
18287 | denc(v.timestamp.tv.tv_sec, p); | |
18288 | denc(v.timestamp.tv.tv_nsec, p); | |
18289 | denc(v.serial, p); | |
18290 | for (auto& pad: v.pad) { | |
18291 | denc(pad, p); | |
18292 | } | |
18293 | } | |
18294 | ||
18295 | ||
18296 | int verify(CephContext* cct, const std::string &path) { | |
18297 | if (valid_signature == ALLOCATOR_IMAGE_VALID_SIGNATURE) { | |
18298 | for (unsigned i = 0; i < (sizeof(pad) / sizeof(uint32_t)); i++) { | |
18299 | if (this->pad[i]) { | |
18300 | derr << "Illegal Header - pad[" << i << "]="<< pad[i] << dendl; | |
18301 | return -1; | |
18302 | } | |
18303 | } | |
18304 | return 0; | |
18305 | } | |
18306 | else { | |
18307 | derr << "Illegal Header - signature="<< valid_signature << "(" << ALLOCATOR_IMAGE_VALID_SIGNATURE << ")" << dendl; | |
18308 | return -1; | |
18309 | } | |
18310 | } | |
18311 | }; | |
18312 | WRITE_CLASS_DENC(allocator_image_header) | |
18313 | ||
18314 | // 56 Bytes trailer for on-disk alloator image | |
18315 | struct allocator_image_trailer { | |
18316 | extent_t null_extent; // 0x00 | |
18317 | ||
18318 | uint32_t format_version; // 0x10 | |
18319 | uint32_t valid_signature; // 0x14 | |
18320 | ||
18321 | utime_t timestamp; // 0x18 | |
18322 | ||
18323 | uint32_t serial; // 0x20 | |
18324 | uint32_t pad; // 0x24 | |
18325 | uint64_t entries_count; // 0x28 | |
18326 | uint64_t allocation_size; // 0x30 | |
18327 | ||
18328 | // trailer is created in CEPH format | |
18329 | allocator_image_trailer(utime_t timestamp, uint32_t format_version, uint32_t serial, uint64_t entries_count, uint64_t allocation_size) { | |
18330 | memset((char*)&(this->null_extent), 0, sizeof(this->null_extent)); | |
18331 | this->format_version = format_version; | |
18332 | this->valid_signature = ALLOCATOR_IMAGE_VALID_SIGNATURE; | |
18333 | this->timestamp = timestamp; | |
18334 | this->serial = serial; | |
18335 | this->pad = 0; | |
18336 | this->entries_count = entries_count; | |
18337 | this->allocation_size = allocation_size; | |
18338 | } | |
18339 | ||
18340 | allocator_image_trailer() { | |
18341 | memset((char*)this, 0, sizeof(allocator_image_trailer)); | |
18342 | } | |
18343 | ||
18344 | friend std::ostream& operator<<(std::ostream& out, const allocator_image_trailer& trailer) { | |
18345 | if (trailer.null_extent.offset || trailer.null_extent.length) { | |
18346 | out << "trailer.null_extent.offset = " << trailer.null_extent.offset << std::endl; | |
18347 | out << "trailer.null_extent.length = " << trailer.null_extent.length << std::endl; | |
18348 | } | |
18349 | out << "format_version = " << trailer.format_version << std::endl; | |
18350 | out << "valid_signature = " << trailer.valid_signature << "/" << ALLOCATOR_IMAGE_VALID_SIGNATURE << std::endl; | |
18351 | out << "timestamp = " << trailer.timestamp << std::endl; | |
18352 | out << "serial = " << trailer.serial << std::endl; | |
18353 | if (trailer.pad) { | |
18354 | out << "trailer.pad= " << trailer.pad << std::endl; | |
18355 | } | |
18356 | out << "entries_count = " << trailer.entries_count << std::endl; | |
18357 | out << "allocation_size = " << trailer.allocation_size << std::endl; | |
18358 | return out; | |
18359 | } | |
18360 | ||
18361 | int verify(CephContext* cct, const std::string &path, const allocator_image_header *p_header, uint64_t entries_count, uint64_t allocation_size) { | |
18362 | if (valid_signature == ALLOCATOR_IMAGE_VALID_SIGNATURE) { | |
18363 | ||
18364 | // trailer must starts with null extents (both fields set to zero) [no need to convert formats for zero) | |
18365 | if (null_extent.offset || null_extent.length) { | |
18366 | derr << "illegal trailer - null_extent = [" << null_extent.offset << "," << null_extent.length << "]"<< dendl; | |
18367 | return -1; | |
18368 | } | |
18369 | ||
18370 | if (serial != p_header->serial) { | |
18371 | derr << "Illegal trailer: header->serial(" << p_header->serial << ") != trailer->serial(" << serial << ")" << dendl; | |
18372 | return -1; | |
18373 | } | |
18374 | ||
18375 | if (format_version != p_header->format_version) { | |
18376 | derr << "Illegal trailer: header->format_version(" << p_header->format_version | |
18377 | << ") != trailer->format_version(" << format_version << ")" << dendl; | |
18378 | return -1; | |
18379 | } | |
18380 | ||
18381 | if (timestamp != p_header->timestamp) { | |
18382 | derr << "Illegal trailer: header->timestamp(" << p_header->timestamp | |
18383 | << ") != trailer->timestamp(" << timestamp << ")" << dendl; | |
18384 | return -1; | |
18385 | } | |
18386 | ||
18387 | if (this->entries_count != entries_count) { | |
18388 | derr << "Illegal trailer: entries_count(" << entries_count << ") != trailer->entries_count(" | |
18389 | << this->entries_count << ")" << dendl; | |
18390 | return -1; | |
18391 | } | |
18392 | ||
18393 | if (this->allocation_size != allocation_size) { | |
18394 | derr << "Illegal trailer: allocation_size(" << allocation_size << ") != trailer->allocation_size(" | |
18395 | << this->allocation_size << ")" << dendl; | |
18396 | return -1; | |
18397 | } | |
18398 | ||
18399 | if (pad) { | |
18400 | derr << "Illegal Trailer - pad="<< pad << dendl; | |
18401 | return -1; | |
18402 | } | |
18403 | ||
18404 | // if arrived here -> trailer is valid !! | |
18405 | return 0; | |
18406 | } else { | |
18407 | derr << "Illegal Trailer - signature="<< valid_signature << "(" << ALLOCATOR_IMAGE_VALID_SIGNATURE << ")" << dendl; | |
18408 | return -1; | |
18409 | } | |
18410 | } | |
18411 | ||
18412 | DENC(allocator_image_trailer, v, p) { | |
18413 | denc(v.null_extent.offset, p); | |
18414 | denc(v.null_extent.length, p); | |
18415 | denc(v.format_version, p); | |
18416 | denc(v.valid_signature, p); | |
18417 | denc(v.timestamp.tv.tv_sec, p); | |
18418 | denc(v.timestamp.tv.tv_nsec, p); | |
18419 | denc(v.serial, p); | |
18420 | denc(v.pad, p); | |
18421 | denc(v.entries_count, p); | |
18422 | denc(v.allocation_size, p); | |
18423 | } | |
18424 | }; | |
18425 | WRITE_CLASS_DENC(allocator_image_trailer) | |
18426 | ||
18427 | ||
18428 | //------------------------------------------------------------------------------------- | |
18429 | // invalidate old allocation file if exists so will go directly to recovery after failure | |
18430 | // we can safely ignore non-existing file | |
18431 | int BlueStore::invalidate_allocation_file_on_bluefs() | |
18432 | { | |
18433 | // mark that allocation-file was invalidated and we should destage a new copy whne closing db | |
18434 | need_to_destage_allocation_file = true; | |
39ae355f | 18435 | dout(10) << __func__ << " need_to_destage_allocation_file was set" << dendl; |
20effc67 TL |
18436 | |
18437 | BlueFS::FileWriter *p_handle = nullptr; | |
18438 | if (!bluefs->dir_exists(allocator_dir)) { | |
18439 | dout(5) << "allocator_dir(" << allocator_dir << ") doesn't exist" << dendl; | |
18440 | // nothing to do -> return | |
18441 | return 0; | |
18442 | } | |
18443 | ||
18444 | int ret = bluefs->stat(allocator_dir, allocator_file, nullptr, nullptr); | |
18445 | if (ret != 0) { | |
39ae355f | 18446 | dout(5) << __func__ << " allocator_file(" << allocator_file << ") doesn't exist" << dendl; |
20effc67 TL |
18447 | // nothing to do -> return |
18448 | return 0; | |
18449 | } | |
18450 | ||
18451 | ||
18452 | ret = bluefs->open_for_write(allocator_dir, allocator_file, &p_handle, true); | |
18453 | if (ret != 0) { | |
39ae355f TL |
18454 | derr << __func__ << "::NCB:: Failed open_for_write with error-code " |
18455 | << ret << dendl; | |
20effc67 TL |
18456 | return -1; |
18457 | } | |
18458 | ||
18459 | dout(5) << "invalidate using bluefs->truncate(p_handle, 0)" << dendl; | |
18460 | ret = bluefs->truncate(p_handle, 0); | |
18461 | if (ret != 0) { | |
39ae355f TL |
18462 | derr << __func__ << "::NCB:: Failed truncaste with error-code " |
18463 | << ret << dendl; | |
20effc67 TL |
18464 | bluefs->close_writer(p_handle); |
18465 | return -1; | |
18466 | } | |
18467 | ||
18468 | bluefs->fsync(p_handle); | |
18469 | bluefs->close_writer(p_handle); | |
18470 | ||
18471 | return 0; | |
18472 | } | |
18473 | ||
20effc67 TL |
18474 | //----------------------------------------------------------------------------------- |
18475 | int BlueStore::copy_allocator(Allocator* src_alloc, Allocator* dest_alloc, uint64_t* p_num_entries) | |
18476 | { | |
18477 | *p_num_entries = 0; | |
18478 | auto count_entries = [&](uint64_t extent_offset, uint64_t extent_length) { | |
18479 | (*p_num_entries)++; | |
18480 | }; | |
1e59de90 | 18481 | src_alloc->foreach(count_entries); |
20effc67 TL |
18482 | |
18483 | dout(5) << "count num_entries=" << *p_num_entries << dendl; | |
18484 | ||
18485 | // add 16K extra entries in case new allocation happened | |
18486 | (*p_num_entries) += 16*1024; | |
18487 | unique_ptr<extent_t[]> arr; | |
18488 | try { | |
18489 | arr = make_unique<extent_t[]>(*p_num_entries); | |
18490 | } catch (std::bad_alloc&) { | |
18491 | derr << "****Failed dynamic allocation, num_entries=" << *p_num_entries << dendl; | |
18492 | return -1; | |
18493 | } | |
18494 | ||
18495 | uint64_t idx = 0; | |
18496 | auto copy_entries = [&](uint64_t extent_offset, uint64_t extent_length) { | |
18497 | if (extent_length > 0) { | |
18498 | if (idx < *p_num_entries) { | |
18499 | arr[idx] = {extent_offset, extent_length}; | |
18500 | } | |
18501 | idx++; | |
18502 | } | |
18503 | else { | |
18504 | derr << "zero length extent!!! offset=" << extent_offset << ", index=" << idx << dendl; | |
18505 | } | |
18506 | }; | |
1e59de90 | 18507 | src_alloc->foreach(copy_entries); |
20effc67 TL |
18508 | |
18509 | dout(5) << "copy num_entries=" << idx << dendl; | |
18510 | if (idx > *p_num_entries) { | |
18511 | derr << "****spillover, num_entries=" << *p_num_entries << ", spillover=" << (idx - *p_num_entries) << dendl; | |
18512 | ceph_assert(idx <= *p_num_entries); | |
18513 | } | |
18514 | ||
18515 | *p_num_entries = idx; | |
18516 | ||
18517 | for (idx = 0; idx < *p_num_entries; idx++) { | |
18518 | const extent_t *p_extent = &arr[idx]; | |
18519 | dest_alloc->init_add_free(p_extent->offset, p_extent->length); | |
18520 | } | |
18521 | ||
18522 | return 0; | |
18523 | } | |
18524 | ||
18525 | //----------------------------------------------------------------------------------- | |
18526 | static uint32_t flush_extent_buffer_with_crc(BlueFS::FileWriter *p_handle, const char* buffer, const char *p_curr, uint32_t crc) | |
18527 | { | |
18528 | std::ptrdiff_t length = p_curr - buffer; | |
18529 | p_handle->append(buffer, length); | |
18530 | ||
18531 | crc = ceph_crc32c(crc, (const uint8_t*)buffer, length); | |
18532 | uint32_t encoded_crc = HTOCEPH_32(crc); | |
18533 | p_handle->append((byte*)&encoded_crc, sizeof(encoded_crc)); | |
18534 | ||
18535 | return crc; | |
18536 | } | |
18537 | ||
18538 | const unsigned MAX_EXTENTS_IN_BUFFER = 4 * 1024; // 4K extents = 64KB of data | |
18539 | // write the allocator to a flat bluefs file - 4K extents at a time | |
18540 | //----------------------------------------------------------------------------------- | |
18541 | int BlueStore::store_allocator(Allocator* src_allocator) | |
18542 | { | |
18543 | // when storing allocations to file we must be sure there is no background compactions | |
18544 | // the easiest way to achieve it is to make sure db is closed | |
18545 | ceph_assert(db == nullptr); | |
18546 | utime_t start_time = ceph_clock_now(); | |
18547 | int ret = 0; | |
18548 | ||
18549 | // create dir if doesn't exist already | |
18550 | if (!bluefs->dir_exists(allocator_dir) ) { | |
18551 | ret = bluefs->mkdir(allocator_dir); | |
18552 | if (ret != 0) { | |
18553 | derr << "Failed mkdir with error-code " << ret << dendl; | |
18554 | return -1; | |
18555 | } | |
18556 | } | |
1d09f67e | 18557 | bluefs->compact_log(); |
20effc67 TL |
18558 | // reuse previous file-allocation if exists |
18559 | ret = bluefs->stat(allocator_dir, allocator_file, nullptr, nullptr); | |
18560 | bool overwrite_file = (ret == 0); | |
20effc67 TL |
18561 | BlueFS::FileWriter *p_handle = nullptr; |
18562 | ret = bluefs->open_for_write(allocator_dir, allocator_file, &p_handle, overwrite_file); | |
18563 | if (ret != 0) { | |
18564 | derr << __func__ << "Failed open_for_write with error-code " << ret << dendl; | |
18565 | return -1; | |
18566 | } | |
18567 | ||
18568 | uint64_t file_size = p_handle->file->fnode.size; | |
18569 | uint64_t allocated = p_handle->file->fnode.get_allocated(); | |
1d09f67e | 18570 | dout(10) << "file_size=" << file_size << ", allocated=" << allocated << dendl; |
20effc67 | 18571 | |
1d09f67e | 18572 | bluefs->sync_metadata(false); |
20effc67 TL |
18573 | unique_ptr<Allocator> allocator(clone_allocator_without_bluefs(src_allocator)); |
18574 | if (!allocator) { | |
18575 | bluefs->close_writer(p_handle); | |
18576 | return -1; | |
18577 | } | |
18578 | ||
18579 | // store all extents (except for the bluefs extents we removed) in a single flat file | |
18580 | utime_t timestamp = ceph_clock_now(); | |
18581 | uint32_t crc = -1; | |
18582 | { | |
18583 | allocator_image_header header(timestamp, s_format_version, s_serial); | |
18584 | bufferlist header_bl; | |
18585 | encode(header, header_bl); | |
18586 | crc = header_bl.crc32c(crc); | |
18587 | encode(crc, header_bl); | |
18588 | p_handle->append(header_bl); | |
18589 | } | |
18590 | ||
18591 | crc = -1; // reset crc | |
18592 | extent_t buffer[MAX_EXTENTS_IN_BUFFER]; // 64KB | |
18593 | extent_t *p_curr = buffer; | |
18594 | const extent_t *p_end = buffer + MAX_EXTENTS_IN_BUFFER; | |
18595 | uint64_t extent_count = 0; | |
18596 | uint64_t allocation_size = 0; | |
18597 | auto iterated_allocation = [&](uint64_t extent_offset, uint64_t extent_length) { | |
18598 | if (extent_length == 0) { | |
18599 | derr << __func__ << "" << extent_count << "::[" << extent_offset << "," << extent_length << "]" << dendl; | |
18600 | ret = -1; | |
18601 | return; | |
18602 | } | |
18603 | p_curr->offset = HTOCEPH_64(extent_offset); | |
18604 | p_curr->length = HTOCEPH_64(extent_length); | |
18605 | extent_count++; | |
18606 | allocation_size += extent_length; | |
18607 | p_curr++; | |
18608 | ||
18609 | if (p_curr == p_end) { | |
18610 | crc = flush_extent_buffer_with_crc(p_handle, (const char*)buffer, (const char*)p_curr, crc); | |
18611 | p_curr = buffer; // recycle the buffer | |
18612 | } | |
18613 | }; | |
1e59de90 | 18614 | allocator->foreach(iterated_allocation); |
20effc67 TL |
18615 | // if got null extent -> fail the operation |
18616 | if (ret != 0) { | |
18617 | derr << "Illegal extent, fail store operation" << dendl; | |
18618 | derr << "invalidate using bluefs->truncate(p_handle, 0)" << dendl; | |
18619 | bluefs->truncate(p_handle, 0); | |
18620 | bluefs->close_writer(p_handle); | |
18621 | return -1; | |
18622 | } | |
18623 | ||
18624 | // if we got any leftovers -> add crc and append to file | |
18625 | if (p_curr > buffer) { | |
18626 | crc = flush_extent_buffer_with_crc(p_handle, (const char*)buffer, (const char*)p_curr, crc); | |
18627 | } | |
18628 | ||
18629 | { | |
18630 | allocator_image_trailer trailer(timestamp, s_format_version, s_serial, extent_count, allocation_size); | |
18631 | bufferlist trailer_bl; | |
18632 | encode(trailer, trailer_bl); | |
18633 | uint32_t crc = -1; | |
18634 | crc = trailer_bl.crc32c(crc); | |
18635 | encode(crc, trailer_bl); | |
18636 | p_handle->append(trailer_bl); | |
18637 | } | |
18638 | ||
18639 | bluefs->fsync(p_handle); | |
18640 | bluefs->truncate(p_handle, p_handle->pos); | |
18641 | bluefs->fsync(p_handle); | |
18642 | ||
18643 | utime_t duration = ceph_clock_now() - start_time; | |
1d09f67e | 18644 | dout(5) <<"WRITE-extent_count=" << extent_count << ", allocation_size=" << allocation_size << ", serial=" << s_serial << dendl; |
20effc67 TL |
18645 | dout(5) <<"p_handle->pos=" << p_handle->pos << " WRITE-duration=" << duration << " seconds" << dendl; |
18646 | ||
18647 | bluefs->close_writer(p_handle); | |
18648 | need_to_destage_allocation_file = false; | |
20effc67 TL |
18649 | return 0; |
18650 | } | |
18651 | ||
18652 | //----------------------------------------------------------------------------------- | |
18653 | Allocator* BlueStore::create_bitmap_allocator(uint64_t bdev_size) { | |
18654 | // create allocator | |
18655 | uint64_t alloc_size = min_alloc_size; | |
18656 | Allocator* alloc = Allocator::create(cct, "bitmap", bdev_size, alloc_size, | |
18657 | zone_size, first_sequential_zone, | |
18658 | "recovery"); | |
18659 | if (alloc) { | |
18660 | return alloc; | |
18661 | } else { | |
18662 | derr << "Failed Allocator Creation" << dendl; | |
18663 | return nullptr; | |
18664 | } | |
18665 | } | |
18666 | ||
18667 | //----------------------------------------------------------------------------------- | |
18668 | size_t calc_allocator_image_header_size() | |
18669 | { | |
18670 | utime_t timestamp = ceph_clock_now(); | |
18671 | allocator_image_header header(timestamp, s_format_version, s_serial); | |
18672 | bufferlist header_bl; | |
18673 | encode(header, header_bl); | |
18674 | uint32_t crc = -1; | |
18675 | crc = header_bl.crc32c(crc); | |
18676 | encode(crc, header_bl); | |
18677 | ||
18678 | return header_bl.length(); | |
18679 | } | |
18680 | ||
18681 | //----------------------------------------------------------------------------------- | |
18682 | int calc_allocator_image_trailer_size() | |
18683 | { | |
18684 | utime_t timestamp = ceph_clock_now(); | |
18685 | uint64_t extent_count = -1; | |
18686 | uint64_t allocation_size = -1; | |
18687 | uint32_t crc = -1; | |
18688 | bufferlist trailer_bl; | |
18689 | allocator_image_trailer trailer(timestamp, s_format_version, s_serial, extent_count, allocation_size); | |
18690 | ||
18691 | encode(trailer, trailer_bl); | |
18692 | crc = trailer_bl.crc32c(crc); | |
18693 | encode(crc, trailer_bl); | |
18694 | return trailer_bl.length(); | |
18695 | } | |
18696 | ||
18697 | //----------------------------------------------------------------------------------- | |
18698 | int BlueStore::__restore_allocator(Allocator* allocator, uint64_t *num, uint64_t *bytes) | |
18699 | { | |
39ae355f TL |
18700 | if (cct->_conf->bluestore_debug_inject_allocation_from_file_failure > 0) { |
18701 | boost::mt11213b rng(time(NULL)); | |
18702 | boost::uniform_real<> ur(0, 1); | |
18703 | if (ur(rng) < cct->_conf->bluestore_debug_inject_allocation_from_file_failure) { | |
18704 | derr << __func__ << " failure injected." << dendl; | |
18705 | return -1; | |
18706 | } | |
18707 | } | |
20effc67 TL |
18708 | utime_t start_time = ceph_clock_now(); |
18709 | BlueFS::FileReader *p_temp_handle = nullptr; | |
18710 | int ret = bluefs->open_for_read(allocator_dir, allocator_file, &p_temp_handle, false); | |
18711 | if (ret != 0) { | |
1e59de90 | 18712 | dout(1) << "Failed open_for_read with error-code " << ret << dendl; |
20effc67 TL |
18713 | return -1; |
18714 | } | |
18715 | unique_ptr<BlueFS::FileReader> p_handle(p_temp_handle); | |
18716 | uint64_t read_alloc_size = 0; | |
18717 | uint64_t file_size = p_handle->file->fnode.size; | |
18718 | dout(5) << "file_size=" << file_size << ",sizeof(extent_t)=" << sizeof(extent_t) << dendl; | |
18719 | ||
18720 | // make sure we were able to store a valid copy | |
18721 | if (file_size == 0) { | |
1e59de90 | 18722 | dout(1) << "No Valid allocation info on disk (empty file)" << dendl; |
20effc67 TL |
18723 | return -1; |
18724 | } | |
18725 | ||
18726 | // first read the header | |
18727 | size_t offset = 0; | |
18728 | allocator_image_header header; | |
18729 | int header_size = calc_allocator_image_header_size(); | |
18730 | { | |
18731 | bufferlist header_bl,temp_bl; | |
18732 | int read_bytes = bluefs->read(p_handle.get(), offset, header_size, &temp_bl, nullptr); | |
18733 | if (read_bytes != header_size) { | |
18734 | derr << "Failed bluefs->read() for header::read_bytes=" << read_bytes << ", req_bytes=" << header_size << dendl; | |
18735 | return -1; | |
18736 | } | |
18737 | ||
18738 | offset += read_bytes; | |
18739 | ||
18740 | header_bl.claim_append(temp_bl); | |
18741 | auto p = header_bl.cbegin(); | |
18742 | decode(header, p); | |
18743 | if (header.verify(cct, path) != 0 ) { | |
18744 | derr << "header = \n" << header << dendl; | |
18745 | return -1; | |
18746 | } | |
18747 | ||
18748 | uint32_t crc_calc = -1, crc; | |
18749 | crc_calc = header_bl.cbegin().crc32c(p.get_off(), crc_calc); //crc from begin to current pos | |
18750 | decode(crc, p); | |
18751 | if (crc != crc_calc) { | |
18752 | derr << "crc mismatch!!! crc=" << crc << ", crc_calc=" << crc_calc << dendl; | |
18753 | derr << "header = \n" << header << dendl; | |
18754 | return -1; | |
18755 | } | |
18756 | ||
18757 | // increment version for next store | |
18758 | s_serial = header.serial + 1; | |
18759 | } | |
18760 | ||
18761 | // then read the payload (extents list) using a recycled buffer | |
18762 | extent_t buffer[MAX_EXTENTS_IN_BUFFER]; // 64KB | |
18763 | uint32_t crc = -1; | |
18764 | int trailer_size = calc_allocator_image_trailer_size(); | |
18765 | uint64_t extent_count = 0; | |
18766 | uint64_t extents_bytes_left = file_size - (header_size + trailer_size + sizeof(crc)); | |
18767 | while (extents_bytes_left) { | |
1e59de90 | 18768 | int req_bytes = std::min(extents_bytes_left, static_cast<uint64_t>(sizeof(buffer))); |
20effc67 TL |
18769 | int read_bytes = bluefs->read(p_handle.get(), offset, req_bytes, nullptr, (char*)buffer); |
18770 | if (read_bytes != req_bytes) { | |
18771 | derr << "Failed bluefs->read()::read_bytes=" << read_bytes << ", req_bytes=" << req_bytes << dendl; | |
18772 | return -1; | |
18773 | } | |
18774 | ||
18775 | offset += read_bytes; | |
18776 | extents_bytes_left -= read_bytes; | |
18777 | ||
18778 | const unsigned num_extent_in_buffer = read_bytes/sizeof(extent_t); | |
18779 | const extent_t *p_end = buffer + num_extent_in_buffer; | |
18780 | for (const extent_t *p_ext = buffer; p_ext < p_end; p_ext++) { | |
18781 | uint64_t offset = CEPHTOH_64(p_ext->offset); | |
18782 | uint64_t length = CEPHTOH_64(p_ext->length); | |
18783 | read_alloc_size += length; | |
18784 | ||
18785 | if (length > 0) { | |
18786 | allocator->init_add_free(offset, length); | |
18787 | extent_count ++; | |
18788 | } else { | |
18789 | derr << "extent with zero length at idx=" << extent_count << dendl; | |
18790 | return -1; | |
18791 | } | |
18792 | } | |
18793 | ||
18794 | uint32_t calc_crc = ceph_crc32c(crc, (const uint8_t*)buffer, read_bytes); | |
18795 | read_bytes = bluefs->read(p_handle.get(), offset, sizeof(crc), nullptr, (char*)&crc); | |
18796 | if (read_bytes == sizeof(crc) ) { | |
18797 | crc = CEPHTOH_32(crc); | |
18798 | if (crc != calc_crc) { | |
18799 | derr << "data crc mismatch!!! crc=" << crc << ", calc_crc=" << calc_crc << dendl; | |
18800 | derr << "extents_bytes_left=" << extents_bytes_left << ", offset=" << offset << ", extent_count=" << extent_count << dendl; | |
18801 | return -1; | |
18802 | } | |
18803 | ||
18804 | offset += read_bytes; | |
18805 | if (extents_bytes_left) { | |
18806 | extents_bytes_left -= read_bytes; | |
18807 | } | |
18808 | } else { | |
18809 | derr << "Failed bluefs->read() for crc::read_bytes=" << read_bytes << ", req_bytes=" << sizeof(crc) << dendl; | |
18810 | return -1; | |
18811 | } | |
18812 | ||
18813 | } | |
18814 | ||
1e59de90 | 18815 | // finally, read the trailer and verify it is in good shape and that we got all the extents |
20effc67 TL |
18816 | { |
18817 | bufferlist trailer_bl,temp_bl; | |
18818 | int read_bytes = bluefs->read(p_handle.get(), offset, trailer_size, &temp_bl, nullptr); | |
18819 | if (read_bytes != trailer_size) { | |
18820 | derr << "Failed bluefs->read() for trailer::read_bytes=" << read_bytes << ", req_bytes=" << trailer_size << dendl; | |
18821 | return -1; | |
18822 | } | |
18823 | offset += read_bytes; | |
18824 | ||
18825 | trailer_bl.claim_append(temp_bl); | |
18826 | uint32_t crc_calc = -1; | |
18827 | uint32_t crc; | |
18828 | allocator_image_trailer trailer; | |
18829 | auto p = trailer_bl.cbegin(); | |
18830 | decode(trailer, p); | |
18831 | if (trailer.verify(cct, path, &header, extent_count, read_alloc_size) != 0 ) { | |
18832 | derr << "trailer=\n" << trailer << dendl; | |
18833 | return -1; | |
18834 | } | |
18835 | ||
18836 | crc_calc = trailer_bl.cbegin().crc32c(p.get_off(), crc_calc); //crc from begin to current pos | |
18837 | decode(crc, p); | |
18838 | if (crc != crc_calc) { | |
18839 | derr << "trailer crc mismatch!::crc=" << crc << ", crc_calc=" << crc_calc << dendl; | |
18840 | derr << "trailer=\n" << trailer << dendl; | |
18841 | return -1; | |
18842 | } | |
18843 | } | |
18844 | ||
18845 | utime_t duration = ceph_clock_now() - start_time; | |
18846 | dout(5) << "READ--extent_count=" << extent_count << ", read_alloc_size= " | |
18847 | << read_alloc_size << ", file_size=" << file_size << dendl; | |
1d09f67e | 18848 | dout(5) << "READ duration=" << duration << " seconds, s_serial=" << header.serial << dendl; |
20effc67 TL |
18849 | *num = extent_count; |
18850 | *bytes = read_alloc_size; | |
18851 | return 0; | |
18852 | } | |
18853 | ||
18854 | //----------------------------------------------------------------------------------- | |
18855 | int BlueStore::restore_allocator(Allocator* dest_allocator, uint64_t *num, uint64_t *bytes) | |
18856 | { | |
18857 | utime_t start = ceph_clock_now(); | |
18858 | auto temp_allocator = unique_ptr<Allocator>(create_bitmap_allocator(bdev->get_size())); | |
18859 | int ret = __restore_allocator(temp_allocator.get(), num, bytes); | |
18860 | if (ret != 0) { | |
18861 | return ret; | |
18862 | } | |
18863 | ||
18864 | uint64_t num_entries = 0; | |
18865 | dout(5) << " calling copy_allocator(bitmap_allocator -> shared_alloc.a)" << dendl; | |
18866 | copy_allocator(temp_allocator.get(), dest_allocator, &num_entries); | |
18867 | utime_t duration = ceph_clock_now() - start; | |
18868 | dout(5) << "restored in " << duration << " seconds, num_entries=" << num_entries << dendl; | |
18869 | return ret; | |
18870 | } | |
18871 | ||
20effc67 TL |
18872 | //----------------------------------------------------------------------------------- |
18873 | void BlueStore::set_allocation_in_simple_bmap(SimpleBitmap* sbmap, uint64_t offset, uint64_t length) | |
18874 | { | |
39ae355f TL |
18875 | dout(30) << __func__ << " 0x" << std::hex |
18876 | << offset << "~" << length | |
18877 | << " " << min_alloc_size_mask | |
18878 | << dendl; | |
20effc67 TL |
18879 | ceph_assert((offset & min_alloc_size_mask) == 0); |
18880 | ceph_assert((length & min_alloc_size_mask) == 0); | |
18881 | sbmap->set(offset >> min_alloc_size_order, length >> min_alloc_size_order); | |
18882 | } | |
18883 | ||
39ae355f TL |
18884 | void BlueStore::ExtentDecoderPartial::_consume_new_blob(bool spanning, |
18885 | uint64_t extent_no, | |
18886 | uint64_t sbid, | |
18887 | BlobRef b) | |
18888 | { | |
18889 | [[maybe_unused]] auto cct = store.cct; | |
18890 | ceph_assert(per_pool_statfs); | |
18891 | ceph_assert(oid != ghobject_t()); | |
20effc67 | 18892 | |
39ae355f TL |
18893 | auto &blob = b->get_blob(); |
18894 | if(spanning) { | |
18895 | dout(20) << __func__ << " " << spanning << " " << b->id << dendl; | |
18896 | ceph_assert(b->id >= 0); | |
18897 | spanning_blobs[b->id] = b; | |
18898 | ++stats.spanning_blob_count; | |
18899 | } else { | |
18900 | dout(20) << __func__ << " " << spanning << " " << extent_no << dendl; | |
18901 | blobs[extent_no] = b; | |
18902 | } | |
18903 | bool compressed = blob.is_compressed(); | |
18904 | if (!blob.is_shared()) { | |
18905 | for (auto& pe : blob.get_extents()) { | |
18906 | if (pe.offset == bluestore_pextent_t::INVALID_OFFSET) { | |
18907 | ++stats.skipped_illegal_extent; | |
18908 | continue; | |
18909 | } | |
18910 | store.set_allocation_in_simple_bmap(&sbmap, pe.offset, pe.length); | |
20effc67 | 18911 | |
39ae355f TL |
18912 | per_pool_statfs->allocated() += pe.length; |
18913 | if (compressed) { | |
18914 | per_pool_statfs->compressed_allocated() += pe.length; | |
18915 | } | |
20effc67 | 18916 | } |
39ae355f TL |
18917 | if (compressed) { |
18918 | per_pool_statfs->compressed() += | |
18919 | blob.get_compressed_payload_length(); | |
18920 | ++stats.compressed_blob_count; | |
20effc67 | 18921 | } |
39ae355f TL |
18922 | } else { |
18923 | auto it = sb_info.find(sbid); | |
18924 | if (it == sb_info.end()) { | |
18925 | derr << __func__ << " shared blob not found:" << sbid | |
18926 | << dendl; | |
18927 | } | |
18928 | auto &sbi = *it; | |
18929 | auto pool_id = oid.hobj.get_logical_pool(); | |
18930 | if (sbi.pool_id == sb_info_t::INVALID_POOL_ID) { | |
18931 | sbi.pool_id = pool_id; | |
18932 | size_t alloc_delta = sbi.allocated_chunks << min_alloc_size_order; | |
18933 | per_pool_statfs->allocated() += alloc_delta; | |
18934 | if (compressed) { | |
18935 | per_pool_statfs->compressed_allocated() += alloc_delta; | |
18936 | ++stats.compressed_blob_count; | |
20effc67 | 18937 | } |
39ae355f TL |
18938 | } |
18939 | if (compressed) { | |
18940 | per_pool_statfs->compressed() += | |
18941 | blob.get_compressed_payload_length(); | |
18942 | } | |
18943 | } | |
18944 | } | |
20effc67 | 18945 | |
39ae355f TL |
18946 | void BlueStore::ExtentDecoderPartial::consume_blobid(Extent* le, |
18947 | bool spanning, | |
18948 | uint64_t blobid) | |
18949 | { | |
18950 | [[maybe_unused]] auto cct = store.cct; | |
18951 | dout(20) << __func__ << " " << spanning << " " << blobid << dendl; | |
18952 | auto &map = spanning ? spanning_blobs : blobs; | |
18953 | auto it = map.find(blobid); | |
18954 | ceph_assert(it != map.end()); | |
18955 | per_pool_statfs->stored() += le->length; | |
18956 | if (it->second->get_blob().is_compressed()) { | |
18957 | per_pool_statfs->compressed_original() += le->length; | |
18958 | } | |
18959 | } | |
20effc67 | 18960 | |
39ae355f TL |
18961 | void BlueStore::ExtentDecoderPartial::consume_blob(Extent* le, |
18962 | uint64_t extent_no, | |
18963 | uint64_t sbid, | |
18964 | BlobRef b) | |
18965 | { | |
18966 | _consume_new_blob(false, extent_no, sbid, b); | |
18967 | per_pool_statfs->stored() += le->length; | |
18968 | if (b->get_blob().is_compressed()) { | |
18969 | per_pool_statfs->compressed_original() += le->length; | |
18970 | } | |
18971 | } | |
20effc67 | 18972 | |
39ae355f TL |
18973 | void BlueStore::ExtentDecoderPartial::consume_spanning_blob(uint64_t sbid, |
18974 | BlobRef b) | |
18975 | { | |
18976 | _consume_new_blob(true, 0/*doesn't matter*/, sbid, b); | |
18977 | } | |
20effc67 | 18978 | |
39ae355f TL |
18979 | void BlueStore::ExtentDecoderPartial::reset(const ghobject_t _oid, |
18980 | volatile_statfs* _per_pool_statfs) | |
18981 | { | |
18982 | oid = _oid; | |
18983 | per_pool_statfs = _per_pool_statfs; | |
18984 | blob_map_t empty; | |
18985 | blob_map_t empty2; | |
18986 | std::swap(blobs, empty); | |
18987 | std::swap(spanning_blobs, empty2); | |
20effc67 TL |
18988 | } |
18989 | ||
20effc67 TL |
18990 | int BlueStore::read_allocation_from_onodes(SimpleBitmap *sbmap, read_alloc_stats_t& stats) |
18991 | { | |
39ae355f TL |
18992 | sb_info_space_efficient_map_t sb_info; |
18993 | // iterate over all shared blobs | |
18994 | auto it = db->get_iterator(PREFIX_SHARED_BLOB, KeyValueDB::ITERATOR_NOCACHE); | |
20effc67 | 18995 | if (!it) { |
39ae355f TL |
18996 | derr << "failed getting shared blob's iterator" << dendl; |
18997 | return -ENOENT; | |
18998 | } | |
18999 | if (it) { | |
19000 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
19001 | const auto& key = it->key(); | |
19002 | dout(20) << __func__ << " decode sb " << pretty_binary_string(key) << dendl; | |
19003 | uint64_t sbid = 0; | |
19004 | if (get_key_shared_blob(key, &sbid) != 0) { | |
19005 | derr << __func__ << " bad shared blob key '" << pretty_binary_string(key) | |
19006 | << "'" << dendl; | |
19007 | } | |
19008 | bluestore_shared_blob_t shared_blob(sbid); | |
19009 | bufferlist bl = it->value(); | |
19010 | auto blp = bl.cbegin(); | |
19011 | try { | |
19012 | decode(shared_blob, blp); | |
19013 | } | |
19014 | catch (ceph::buffer::error& e) { | |
19015 | derr << __func__ << " failed to decode Shared Blob" | |
19016 | << pretty_binary_string(key) << dendl; | |
19017 | continue; | |
19018 | } | |
19019 | dout(20) << __func__ << " " << shared_blob << dendl; | |
19020 | uint64_t allocated = 0; | |
19021 | for (auto& r : shared_blob.ref_map.ref_map) { | |
19022 | ceph_assert(r.first != bluestore_pextent_t::INVALID_OFFSET); | |
19023 | set_allocation_in_simple_bmap(sbmap, r.first, r.second.length); | |
19024 | allocated += r.second.length; | |
19025 | } | |
19026 | auto &sbi = sb_info.add_or_adopt(sbid); | |
19027 | ceph_assert(p2phase(allocated, min_alloc_size) == 0); | |
19028 | sbi.allocated_chunks += (allocated >> min_alloc_size_order); | |
19029 | ++stats.shared_blob_count; | |
19030 | } | |
19031 | } | |
19032 | ||
19033 | it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); | |
19034 | if (!it) { | |
19035 | derr << "failed getting onode's iterator" << dendl; | |
19036 | return -ENOENT; | |
20effc67 TL |
19037 | } |
19038 | ||
20effc67 TL |
19039 | uint64_t kv_count = 0; |
19040 | uint64_t count_interval = 1'000'000; | |
39ae355f TL |
19041 | ExtentDecoderPartial edecoder(*this, |
19042 | stats, | |
19043 | *sbmap, | |
19044 | sb_info, | |
19045 | min_alloc_size_order); | |
19046 | ||
20effc67 TL |
19047 | // iterate over all ONodes stored in RocksDB |
19048 | for (it->lower_bound(string()); it->valid(); it->next(), kv_count++) { | |
19049 | // trace an even after every million processed objects (typically every 5-10 seconds) | |
19050 | if (kv_count && (kv_count % count_interval == 0) ) { | |
39ae355f TL |
19051 | dout(5) << __func__ << " processed objects count = " << kv_count << dendl; |
19052 | } | |
20effc67 | 19053 | |
39ae355f TL |
19054 | auto key = it->key(); |
19055 | auto okey = key; | |
19056 | dout(20) << __func__ << " decode onode " << pretty_binary_string(key) << dendl; | |
19057 | ghobject_t oid; | |
19058 | if (!is_extent_shard_key(it->key())) { | |
19059 | int r = get_key_object(okey, &oid); | |
19060 | if (r != 0) { | |
19061 | derr << __func__ << " failed to decode onode key = " | |
19062 | << pretty_binary_string(okey) << dendl; | |
19063 | return -EIO; | |
20effc67 | 19064 | } |
39ae355f TL |
19065 | edecoder.reset(oid, |
19066 | &stats.actual_pool_vstatfs[oid.hobj.get_logical_pool()]); | |
19067 | Onode dummy_on(cct); | |
19068 | Onode::decode_raw(&dummy_on, | |
19069 | it->value(), | |
19070 | edecoder); | |
19071 | ++stats.onode_count; | |
19072 | } else { | |
19073 | uint32_t offset; | |
19074 | int r = get_key_extent_shard(key, &okey, &offset); | |
19075 | if (r != 0) { | |
19076 | derr << __func__ << " failed to decode onode extent key = " | |
19077 | << pretty_binary_string(key) << dendl; | |
19078 | return -EIO; | |
20effc67 | 19079 | } |
39ae355f TL |
19080 | r = get_key_object(okey, &oid); |
19081 | if (r != 0) { | |
19082 | derr << __func__ | |
19083 | << " failed to decode onode key= " << pretty_binary_string(okey) | |
19084 | << " from extent key= " << pretty_binary_string(key) | |
19085 | << dendl; | |
19086 | return -EIO; | |
20effc67 | 19087 | } |
39ae355f TL |
19088 | ceph_assert(oid == edecoder.get_oid()); |
19089 | edecoder.decode_some(it->value(), nullptr); | |
19090 | ++stats.shard_count; | |
20effc67 TL |
19091 | } |
19092 | } | |
19093 | ||
39ae355f TL |
19094 | std::lock_guard l(vstatfs_lock); |
19095 | store_statfs_t s; | |
19096 | osd_pools.clear(); | |
19097 | for (auto& p : stats.actual_pool_vstatfs) { | |
19098 | if (per_pool_stat_collection) { | |
19099 | osd_pools[p.first] = p.second; | |
19100 | } | |
19101 | stats.actual_store_vstatfs += p.second; | |
19102 | p.second.publish(&s); | |
19103 | dout(5) << __func__ << " recovered pool " | |
19104 | << std::hex | |
19105 | << p.first << "->" << s | |
19106 | << std::dec | |
19107 | << " per-pool:" << per_pool_stat_collection | |
19108 | << dendl; | |
20effc67 | 19109 | } |
39ae355f TL |
19110 | vstatfs = stats.actual_store_vstatfs; |
19111 | vstatfs.publish(&s); | |
19112 | dout(5) << __func__ << " recovered " << s | |
19113 | << dendl; | |
20effc67 TL |
19114 | return 0; |
19115 | } | |
19116 | ||
19117 | //--------------------------------------------------------- | |
19118 | int BlueStore::reconstruct_allocations(SimpleBitmap *sbmap, read_alloc_stats_t &stats) | |
19119 | { | |
19120 | // first set space used by superblock | |
19121 | auto super_length = std::max<uint64_t>(min_alloc_size, SUPER_RESERVED); | |
19122 | set_allocation_in_simple_bmap(sbmap, 0, super_length); | |
19123 | stats.extent_count++; | |
19124 | ||
19125 | // then set all space taken by Objects | |
19126 | int ret = read_allocation_from_onodes(sbmap, stats); | |
19127 | if (ret < 0) { | |
19128 | derr << "failed read_allocation_from_onodes()" << dendl; | |
19129 | return ret; | |
19130 | } | |
19131 | ||
19132 | return 0; | |
19133 | } | |
19134 | ||
19135 | //----------------------------------------------------------------------------------- | |
19136 | static void copy_simple_bitmap_to_allocator(SimpleBitmap* sbmap, Allocator* dest_alloc, uint64_t alloc_size) | |
19137 | { | |
1e59de90 | 19138 | int alloc_size_shift = std::countr_zero(alloc_size); |
20effc67 TL |
19139 | uint64_t offset = 0; |
19140 | extent_t ext = sbmap->get_next_clr_extent(offset); | |
19141 | while (ext.length != 0) { | |
19142 | dest_alloc->init_add_free(ext.offset << alloc_size_shift, ext.length << alloc_size_shift); | |
19143 | offset = ext.offset + ext.length; | |
19144 | ext = sbmap->get_next_clr_extent(offset); | |
19145 | } | |
19146 | } | |
19147 | ||
19148 | //--------------------------------------------------------- | |
19149 | int BlueStore::read_allocation_from_drive_on_startup() | |
19150 | { | |
19151 | int ret = 0; | |
19152 | ||
19153 | ret = _open_collections(); | |
19154 | if (ret < 0) { | |
19155 | return ret; | |
19156 | } | |
19157 | auto shutdown_cache = make_scope_guard([&] { | |
19158 | _shutdown_cache(); | |
19159 | }); | |
19160 | ||
19161 | utime_t start = ceph_clock_now(); | |
19162 | read_alloc_stats_t stats = {}; | |
1d09f67e | 19163 | SimpleBitmap sbmap(cct, (bdev->get_size()/ min_alloc_size)); |
20effc67 TL |
19164 | ret = reconstruct_allocations(&sbmap, stats); |
19165 | if (ret != 0) { | |
19166 | return ret; | |
19167 | } | |
19168 | ||
19169 | copy_simple_bitmap_to_allocator(&sbmap, alloc, min_alloc_size); | |
19170 | ||
19171 | utime_t duration = ceph_clock_now() - start; | |
19172 | dout(1) << "::Allocation Recovery was completed in " << duration << " seconds, extent_count=" << stats.extent_count << dendl; | |
19173 | return ret; | |
19174 | } | |
19175 | ||
19176 | ||
19177 | ||
19178 | ||
19179 | // Only used for debugging purposes - we build a secondary allocator from the Onodes and compare it to the existing one | |
19180 | // Not meant to be run by customers | |
19181 | #ifdef CEPH_BLUESTORE_TOOL_RESTORE_ALLOCATION | |
19182 | ||
19183 | #include <stdlib.h> | |
19184 | #include <algorithm> | |
19185 | //--------------------------------------------------------- | |
19186 | int cmpfunc (const void * a, const void * b) | |
19187 | { | |
19188 | if ( ((extent_t*)a)->offset > ((extent_t*)b)->offset ) { | |
19189 | return 1; | |
19190 | } | |
19191 | else if( ((extent_t*)a)->offset < ((extent_t*)b)->offset ) { | |
19192 | return -1; | |
19193 | } | |
19194 | else { | |
19195 | return 0; | |
19196 | } | |
19197 | } | |
19198 | ||
19199 | // compare the allocator built from Onodes with the system allocator (CF-B) | |
19200 | //--------------------------------------------------------- | |
19201 | int BlueStore::compare_allocators(Allocator* alloc1, Allocator* alloc2, uint64_t req_extent_count, uint64_t memory_target) | |
19202 | { | |
19203 | uint64_t allocation_size = std::min((req_extent_count) * sizeof(extent_t), memory_target / 3); | |
19204 | uint64_t extent_count = allocation_size/sizeof(extent_t); | |
19205 | dout(5) << "req_extent_count=" << req_extent_count << ", granted extent_count="<< extent_count << dendl; | |
19206 | ||
19207 | unique_ptr<extent_t[]> arr1; | |
19208 | unique_ptr<extent_t[]> arr2; | |
19209 | try { | |
19210 | arr1 = make_unique<extent_t[]>(extent_count); | |
19211 | arr2 = make_unique<extent_t[]>(extent_count); | |
19212 | } catch (std::bad_alloc&) { | |
19213 | derr << "****Failed dynamic allocation, extent_count=" << extent_count << dendl; | |
19214 | return -1; | |
19215 | } | |
19216 | ||
19217 | // copy the extents from the allocators into simple array and then compare them | |
19218 | uint64_t size1 = 0, size2 = 0; | |
19219 | uint64_t idx1 = 0, idx2 = 0; | |
19220 | auto iterated_mapper1 = [&](uint64_t offset, uint64_t length) { | |
19221 | size1 += length; | |
19222 | if (idx1 < extent_count) { | |
19223 | arr1[idx1++] = {offset, length}; | |
19224 | } | |
19225 | else if (idx1 == extent_count) { | |
19226 | derr << "(2)compare_allocators:: spillover" << dendl; | |
19227 | idx1 ++; | |
19228 | } | |
19229 | ||
19230 | }; | |
19231 | ||
19232 | auto iterated_mapper2 = [&](uint64_t offset, uint64_t length) { | |
19233 | size2 += length; | |
19234 | if (idx2 < extent_count) { | |
19235 | arr2[idx2++] = {offset, length}; | |
19236 | } | |
19237 | else if (idx2 == extent_count) { | |
19238 | derr << "(2)compare_allocators:: spillover" << dendl; | |
19239 | idx2 ++; | |
19240 | } | |
19241 | }; | |
19242 | ||
1e59de90 TL |
19243 | alloc1->foreach(iterated_mapper1); |
19244 | alloc2->foreach(iterated_mapper2); | |
20effc67 TL |
19245 | |
19246 | qsort(arr1.get(), std::min(idx1, extent_count), sizeof(extent_t), cmpfunc); | |
19247 | qsort(arr2.get(), std::min(idx2, extent_count), sizeof(extent_t), cmpfunc); | |
19248 | ||
19249 | if (idx1 == idx2) { | |
19250 | idx1 = idx2 = std::min(idx1, extent_count); | |
19251 | if (memcmp(arr1.get(), arr2.get(), sizeof(extent_t) * idx2) == 0) { | |
19252 | return 0; | |
19253 | } | |
19254 | derr << "Failed memcmp(arr1, arr2, sizeof(extent_t)*idx2)" << dendl; | |
19255 | for (uint64_t i = 0; i < idx1; i++) { | |
19256 | if (memcmp(arr1.get()+i, arr2.get()+i, sizeof(extent_t)) != 0) { | |
19257 | derr << "!!!![" << i << "] arr1::<" << arr1[i].offset << "," << arr1[i].length << ">" << dendl; | |
19258 | derr << "!!!![" << i << "] arr2::<" << arr2[i].offset << "," << arr2[i].length << ">" << dendl; | |
19259 | return -1; | |
19260 | } | |
19261 | } | |
19262 | return 0; | |
19263 | } else { | |
19264 | derr << "mismatch:: idx1=" << idx1 << " idx2=" << idx2 << dendl; | |
20effc67 TL |
19265 | return -1; |
19266 | } | |
19267 | } | |
19268 | ||
19269 | //--------------------------------------------------------- | |
19270 | int BlueStore::add_existing_bluefs_allocation(Allocator* allocator, read_alloc_stats_t &stats) | |
19271 | { | |
19272 | // then add space used by bluefs to store rocksdb | |
19273 | unsigned extent_count = 0; | |
19274 | if (bluefs) { | |
1e59de90 TL |
19275 | bluefs->foreach_block_extents( |
19276 | bluefs_layout.shared_bdev, | |
19277 | [&](uint64_t start, uint32_t len) { | |
19278 | allocator->init_rm_free(start, len); | |
19279 | stats.extent_count++; | |
19280 | } | |
19281 | ); | |
20effc67 TL |
19282 | } |
19283 | ||
19284 | dout(5) << "bluefs extent_count=" << extent_count << dendl; | |
19285 | return 0; | |
19286 | } | |
19287 | ||
19288 | //--------------------------------------------------------- | |
19289 | int BlueStore::read_allocation_from_drive_for_bluestore_tool() | |
19290 | { | |
19291 | dout(5) << __func__ << dendl; | |
19292 | int ret = 0; | |
19293 | uint64_t memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
19294 | ret = _open_db_and_around(true, false); | |
19295 | if (ret < 0) { | |
19296 | return ret; | |
19297 | } | |
19298 | ||
19299 | ret = _open_collections(); | |
19300 | if (ret < 0) { | |
19301 | _close_db_and_around(); | |
19302 | return ret; | |
19303 | } | |
19304 | ||
19305 | utime_t duration; | |
19306 | read_alloc_stats_t stats = {}; | |
19307 | utime_t start = ceph_clock_now(); | |
19308 | ||
19309 | auto shutdown_cache = make_scope_guard([&] { | |
1d09f67e TL |
19310 | dout(1) << "Allocation Recovery was completed in " << duration |
19311 | << " seconds; insert_count=" << stats.insert_count | |
19312 | << "; extent_count=" << stats.extent_count << dendl; | |
20effc67 TL |
19313 | _shutdown_cache(); |
19314 | _close_db_and_around(); | |
19315 | }); | |
19316 | ||
19317 | { | |
19318 | auto allocator = unique_ptr<Allocator>(create_bitmap_allocator(bdev->get_size())); | |
19319 | //reconstruct allocations into a temp simple-bitmap and copy into allocator | |
19320 | { | |
1d09f67e | 19321 | SimpleBitmap sbmap(cct, (bdev->get_size()/ min_alloc_size)); |
20effc67 TL |
19322 | ret = reconstruct_allocations(&sbmap, stats); |
19323 | if (ret != 0) { | |
19324 | return ret; | |
19325 | } | |
19326 | copy_simple_bitmap_to_allocator(&sbmap, allocator.get(), min_alloc_size); | |
19327 | } | |
19328 | ||
19329 | // add allocation space used by the bluefs itself | |
19330 | ret = add_existing_bluefs_allocation(allocator.get(), stats); | |
19331 | if (ret < 0) { | |
19332 | return ret; | |
19333 | } | |
19334 | ||
19335 | duration = ceph_clock_now() - start; | |
19336 | stats.insert_count = 0; | |
19337 | auto count_entries = [&](uint64_t extent_offset, uint64_t extent_length) { | |
19338 | stats.insert_count++; | |
19339 | }; | |
1e59de90 | 19340 | allocator->foreach(count_entries); |
20effc67 | 19341 | ret = compare_allocators(allocator.get(), alloc, stats.insert_count, memory_target); |
1d09f67e | 19342 | if (ret == 0) { |
20effc67 TL |
19343 | dout(5) << "Allocator drive - file integrity check OK" << dendl; |
19344 | } else { | |
19345 | derr << "FAILURE. Allocator from file and allocator from metadata differ::ret=" << ret << dendl; | |
19346 | } | |
19347 | } | |
19348 | ||
1d09f67e | 19349 | dout(1) << stats << dendl; |
20effc67 TL |
19350 | return ret; |
19351 | } | |
19352 | ||
19353 | //--------------------------------------------------------- | |
19354 | Allocator* BlueStore::clone_allocator_without_bluefs(Allocator *src_allocator) | |
19355 | { | |
19356 | uint64_t bdev_size = bdev->get_size(); | |
19357 | Allocator* allocator = create_bitmap_allocator(bdev_size); | |
19358 | if (allocator) { | |
19359 | dout(5) << "bitmap-allocator=" << allocator << dendl; | |
19360 | } else { | |
19361 | derr << "****failed create_bitmap_allocator()" << dendl; | |
19362 | return nullptr; | |
19363 | } | |
19364 | ||
19365 | uint64_t num_entries = 0; | |
19366 | copy_allocator(src_allocator, allocator, &num_entries); | |
19367 | ||
19368 | // BlueFS stores its internal allocation outside RocksDB (FM) so we should not destage them to the allcoator-file | |
19369 | // we are going to hide bluefs allocation during allocator-destage as they are stored elsewhere | |
19370 | { | |
1e59de90 TL |
19371 | bluefs->foreach_block_extents( |
19372 | bluefs_layout.shared_bdev, | |
19373 | [&] (uint64_t start, uint32_t len) { | |
19374 | allocator->init_add_free(start, len); | |
19375 | } | |
19376 | ); | |
20effc67 TL |
19377 | } |
19378 | ||
19379 | return allocator; | |
19380 | } | |
19381 | ||
19382 | //--------------------------------------------------------- | |
19383 | static void clear_allocation_objects_from_rocksdb(KeyValueDB *db, CephContext *cct, const std::string &path) | |
19384 | { | |
19385 | dout(5) << "t->rmkeys_by_prefix(PREFIX_ALLOC_BITMAP)" << dendl; | |
19386 | KeyValueDB::Transaction t = db->get_transaction(); | |
19387 | t->rmkeys_by_prefix(PREFIX_ALLOC_BITMAP); | |
19388 | db->submit_transaction_sync(t); | |
19389 | } | |
19390 | ||
19391 | //--------------------------------------------------------- | |
19392 | void BlueStore::copy_allocator_content_to_fm(Allocator *allocator, FreelistManager *real_fm) | |
19393 | { | |
19394 | unsigned max_txn = 1024; | |
19395 | dout(5) << "max_transaction_submit=" << max_txn << dendl; | |
19396 | uint64_t size = 0, idx = 0; | |
19397 | KeyValueDB::Transaction txn = db->get_transaction(); | |
19398 | auto iterated_insert = [&](uint64_t offset, uint64_t length) { | |
19399 | size += length; | |
19400 | real_fm->release(offset, length, txn); | |
19401 | if ((++idx % max_txn) == 0) { | |
19402 | db->submit_transaction_sync(txn); | |
19403 | txn = db->get_transaction(); | |
19404 | } | |
19405 | }; | |
1e59de90 | 19406 | allocator->foreach(iterated_insert); |
20effc67 TL |
19407 | if (idx % max_txn != 0) { |
19408 | db->submit_transaction_sync(txn); | |
19409 | } | |
19410 | dout(5) << "size=" << size << ", num extents=" << idx << dendl; | |
19411 | } | |
19412 | ||
19413 | //--------------------------------------------------------- | |
19414 | Allocator* BlueStore::initialize_allocator_from_freelist(FreelistManager *real_fm) | |
19415 | { | |
19416 | dout(5) << "real_fm->enumerate_next" << dendl; | |
19417 | Allocator* allocator2 = create_bitmap_allocator(bdev->get_size()); | |
19418 | if (allocator2) { | |
19419 | dout(5) << "bitmap-allocator=" << allocator2 << dendl; | |
19420 | } else { | |
19421 | return nullptr; | |
19422 | } | |
19423 | ||
19424 | uint64_t size2 = 0, idx2 = 0; | |
19425 | real_fm->enumerate_reset(); | |
19426 | uint64_t offset, length; | |
19427 | while (real_fm->enumerate_next(db, &offset, &length)) { | |
19428 | allocator2->init_add_free(offset, length); | |
19429 | ++idx2; | |
19430 | size2 += length; | |
19431 | } | |
19432 | real_fm->enumerate_reset(); | |
19433 | ||
19434 | dout(5) << "size2=" << size2 << ", num2=" << idx2 << dendl; | |
19435 | return allocator2; | |
19436 | } | |
19437 | ||
19438 | //--------------------------------------------------------- | |
19439 | // close the active fm and open it in a new mode like makefs() | |
19440 | // but make sure to mark the full device space as allocated | |
19441 | // later we will mark all exetents from the allocator as free | |
19442 | int BlueStore::reset_fm_for_restore() | |
19443 | { | |
19444 | dout(5) << "<<==>> fm->clear_null_manager()" << dendl; | |
19445 | fm->shutdown(); | |
19446 | delete fm; | |
19447 | fm = nullptr; | |
19448 | freelist_type = "bitmap"; | |
19449 | KeyValueDB::Transaction t = db->get_transaction(); | |
19450 | // call _open_fm() with fm_restore set to TRUE | |
19451 | // this will mark the full device space as allocated (and not just the reserved space) | |
39ae355f | 19452 | _open_fm(t, true, true, true); |
20effc67 TL |
19453 | if (fm == nullptr) { |
19454 | derr << "Failed _open_fm()" << dendl; | |
19455 | return -1; | |
19456 | } | |
19457 | db->submit_transaction_sync(t); | |
19458 | ceph_assert(!fm->is_null_manager()); | |
19459 | dout(5) << "fm was reactivated in full mode" << dendl; | |
19460 | return 0; | |
19461 | } | |
19462 | ||
19463 | ||
19464 | //--------------------------------------------------------- | |
19465 | // create a temp allocator filled with allocation state from the fm | |
19466 | // and compare it to the base allocator passed in | |
19467 | int BlueStore::verify_rocksdb_allocations(Allocator *allocator) | |
19468 | { | |
19469 | dout(5) << "verify that alloc content is identical to FM" << dendl; | |
19470 | // initialize from freelist | |
19471 | Allocator* temp_allocator = initialize_allocator_from_freelist(fm); | |
19472 | if (temp_allocator == nullptr) { | |
19473 | return -1; | |
19474 | } | |
19475 | ||
19476 | uint64_t insert_count = 0; | |
19477 | auto count_entries = [&](uint64_t extent_offset, uint64_t extent_length) { | |
19478 | insert_count++; | |
19479 | }; | |
1e59de90 | 19480 | temp_allocator->foreach(count_entries); |
20effc67 TL |
19481 | uint64_t memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); |
19482 | int ret = compare_allocators(allocator, temp_allocator, insert_count, memory_target); | |
19483 | ||
19484 | delete temp_allocator; | |
19485 | ||
19486 | if (ret == 0) { | |
19487 | dout(5) << "SUCCESS!!! compare(allocator, temp_allocator)" << dendl; | |
19488 | return 0; | |
19489 | } else { | |
19490 | derr << "**** FAILURE compare(allocator, temp_allocator)::ret=" << ret << dendl; | |
19491 | return -1; | |
19492 | } | |
19493 | } | |
19494 | ||
19495 | //--------------------------------------------------------- | |
19496 | int BlueStore::db_cleanup(int ret) | |
19497 | { | |
19498 | _shutdown_cache(); | |
19499 | _close_db_and_around(); | |
19500 | return ret; | |
19501 | } | |
19502 | ||
19503 | //--------------------------------------------------------- | |
19504 | // convert back the system from null-allocator to using rocksdb to store allocation | |
19505 | int BlueStore::push_allocation_to_rocksdb() | |
19506 | { | |
19507 | if (cct->_conf->bluestore_allocation_from_file) { | |
19508 | derr << "cct->_conf->bluestore_allocation_from_file must be cleared first" << dendl; | |
19509 | derr << "please change default to false in ceph.conf file>" << dendl; | |
19510 | return -1; | |
19511 | } | |
19512 | ||
19513 | dout(5) << "calling open_db_and_around() in read/write mode" << dendl; | |
19514 | int ret = _open_db_and_around(false); | |
19515 | if (ret < 0) { | |
19516 | return ret; | |
19517 | } | |
19518 | ||
19519 | if (!fm->is_null_manager()) { | |
19520 | derr << "This is not a NULL-MANAGER -> nothing to do..." << dendl; | |
19521 | return db_cleanup(0); | |
19522 | } | |
19523 | ||
19524 | // start by creating a clone copy of the shared-allocator | |
19525 | unique_ptr<Allocator> allocator(clone_allocator_without_bluefs(alloc)); | |
19526 | if (!allocator) { | |
19527 | return db_cleanup(-1); | |
19528 | } | |
19529 | ||
19530 | // remove all objects of PREFIX_ALLOC_BITMAP from RocksDB to guarantee a clean start | |
19531 | clear_allocation_objects_from_rocksdb(db, cct, path); | |
19532 | ||
19533 | // then open fm in new mode with the full devie marked as alloctaed | |
19534 | if (reset_fm_for_restore() != 0) { | |
19535 | return db_cleanup(-1); | |
19536 | } | |
19537 | ||
19538 | // push the free-space from the allocator (shared-alloc without bfs) to rocksdb | |
19539 | copy_allocator_content_to_fm(allocator.get(), fm); | |
19540 | ||
19541 | // compare the allocator info with the info stored in the fm/rocksdb | |
19542 | if (verify_rocksdb_allocations(allocator.get()) == 0) { | |
19543 | // all is good -> we can commit to rocksdb allocator | |
19544 | commit_to_real_manager(); | |
19545 | } else { | |
19546 | return db_cleanup(-1); | |
19547 | } | |
19548 | ||
19549 | // can't be too paranoid :-) | |
19550 | dout(5) << "Running full scale verification..." << dendl; | |
19551 | // close db/fm/allocator and start fresh | |
19552 | db_cleanup(0); | |
19553 | dout(5) << "calling open_db_and_around() in read-only mode" << dendl; | |
19554 | ret = _open_db_and_around(true); | |
19555 | if (ret < 0) { | |
19556 | return db_cleanup(ret); | |
19557 | } | |
19558 | ceph_assert(!fm->is_null_manager()); | |
19559 | ceph_assert(verify_rocksdb_allocations(allocator.get()) == 0); | |
19560 | ||
19561 | return db_cleanup(ret); | |
19562 | } | |
19563 | ||
19564 | #endif // CEPH_BLUESTORE_TOOL_RESTORE_ALLOCATION | |
19565 | ||
19566 | //------------------------------------------------------------------------------------- | |
39ae355f | 19567 | int BlueStore::commit_freelist_type() |
20effc67 TL |
19568 | { |
19569 | // When freelist_type to "bitmap" we will store allocation in RocksDB | |
19570 | // When allocation-info is stored in a single file we set freelist_type to "null" | |
19571 | // This will direct the startup code to read allocation from file and not RocksDB | |
19572 | KeyValueDB::Transaction t = db->get_transaction(); | |
19573 | if (t == nullptr) { | |
19574 | derr << "db->get_transaction() failed!!!" << dendl; | |
19575 | return -1; | |
19576 | } | |
19577 | ||
19578 | bufferlist bl; | |
19579 | bl.append(freelist_type); | |
19580 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
19581 | ||
19582 | int ret = db->submit_transaction_sync(t); | |
19583 | if (ret != 0) { | |
19584 | derr << "Failed db->submit_transaction_sync(t)" << dendl; | |
19585 | } | |
19586 | return ret; | |
19587 | } | |
19588 | ||
19589 | //------------------------------------------------------------------------------------- | |
19590 | int BlueStore::commit_to_null_manager() | |
19591 | { | |
39ae355f | 19592 | dout(5) << __func__ << " Set FreelistManager to NULL FM..." << dendl; |
20effc67 TL |
19593 | fm->set_null_manager(); |
19594 | freelist_type = "null"; | |
19595 | #if 1 | |
39ae355f | 19596 | return commit_freelist_type(); |
20effc67 TL |
19597 | #else |
19598 | // should check how long this step take on a big configuration as deletes are expensive | |
39ae355f | 19599 | if (commit_freelist_type() == 0) { |
20effc67 TL |
19600 | // remove all objects of PREFIX_ALLOC_BITMAP from RocksDB to guarantee a clean start |
19601 | clear_allocation_objects_from_rocksdb(db, cct, path); | |
19602 | } | |
19603 | #endif | |
19604 | } | |
19605 | ||
19606 | ||
19607 | //------------------------------------------------------------------------------------- | |
19608 | int BlueStore::commit_to_real_manager() | |
19609 | { | |
19610 | dout(5) << "Set FreelistManager to Real FM..." << dendl; | |
19611 | ceph_assert(!fm->is_null_manager()); | |
19612 | freelist_type = "bitmap"; | |
39ae355f | 19613 | int ret = commit_freelist_type(); |
20effc67 TL |
19614 | if (ret == 0) { |
19615 | //remove the allocation_file | |
19616 | invalidate_allocation_file_on_bluefs(); | |
19617 | ret = bluefs->unlink(allocator_dir, allocator_file); | |
19618 | bluefs->sync_metadata(false); | |
19619 | if (ret == 0) { | |
19620 | dout(5) << "Remove Allocation File successfully" << dendl; | |
19621 | } | |
19622 | else { | |
19623 | derr << "Remove Allocation File ret_code=" << ret << dendl; | |
19624 | } | |
19625 | } | |
19626 | ||
19627 | return ret; | |
19628 | } | |
19629 | ||
19630 | //================================================================================================================ | |
19631 | //================================================================================================================ |