]>
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 | ||
15 | #include <unistd.h> | |
16 | #include <stdlib.h> | |
17 | #include <sys/types.h> | |
18 | #include <sys/stat.h> | |
19 | #include <fcntl.h> | |
20 | ||
eafe8130 | 21 | #include <boost/container/flat_set.hpp> |
9f95a23c | 22 | #include "boost/algorithm/string.hpp" |
eafe8130 | 23 | |
31f18b77 FG |
24 | #include "include/cpp-btree/btree_set.h" |
25 | ||
7c673cae | 26 | #include "BlueStore.h" |
f67539c2 | 27 | #include "bluestore_common.h" |
7c673cae FG |
28 | #include "os/kv.h" |
29 | #include "include/compat.h" | |
30 | #include "include/intarith.h" | |
31 | #include "include/stringify.h" | |
11fdf7f2 TL |
32 | #include "include/str_map.h" |
33 | #include "include/util.h" | |
7c673cae FG |
34 | #include "common/errno.h" |
35 | #include "common/safe_io.h" | |
91327a77 | 36 | #include "common/PriorityCache.h" |
9f95a23c | 37 | #include "common/RWLock.h" |
7c673cae FG |
38 | #include "Allocator.h" |
39 | #include "FreelistManager.h" | |
40 | #include "BlueFS.h" | |
41 | #include "BlueRocksEnv.h" | |
42 | #include "auth/Crypto.h" | |
43 | #include "common/EventTrace.h" | |
91327a77 | 44 | #include "perfglue/heap_profiler.h" |
11fdf7f2 TL |
45 | #include "common/blkdev.h" |
46 | #include "common/numa.h" | |
f67539c2 | 47 | #include "common/pretty_binary.h" |
7c673cae | 48 | |
9f95a23c TL |
49 | #if defined(WITH_LTTNG) |
50 | #define TRACEPOINT_DEFINE | |
51 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
52 | #include "tracing/bluestore.h" | |
53 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
54 | #undef TRACEPOINT_DEFINE | |
55 | #else | |
56 | #define tracepoint(...) | |
57 | #endif | |
58 | ||
7c673cae FG |
59 | #define dout_context cct |
60 | #define dout_subsys ceph_subsys_bluestore | |
61 | ||
31f18b77 FG |
62 | using bid_t = decltype(BlueStore::Blob::id); |
63 | ||
64 | // bluestore_cache_onode | |
7c673cae | 65 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Onode, bluestore_onode, |
31f18b77 | 66 | bluestore_cache_onode); |
7c673cae | 67 | |
31f18b77 | 68 | // bluestore_cache_other |
7c673cae | 69 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Buffer, bluestore_buffer, |
f91f0fd5 | 70 | bluestore_Buffer); |
7c673cae | 71 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Extent, bluestore_extent, |
f91f0fd5 | 72 | bluestore_Extent); |
7c673cae | 73 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Blob, bluestore_blob, |
f91f0fd5 | 74 | bluestore_Blob); |
7c673cae | 75 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::SharedBlob, bluestore_shared_blob, |
f91f0fd5 | 76 | bluestore_SharedBlob); |
31f18b77 FG |
77 | |
78 | // bluestore_txc | |
79 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::TransContext, bluestore_transcontext, | |
80 | bluestore_txc); | |
f67539c2 TL |
81 | using std::deque; |
82 | using std::min; | |
83 | using std::make_pair; | |
84 | using std::numeric_limits; | |
85 | using std::pair; | |
86 | using std::list; | |
87 | using std::map; | |
88 | using std::max; | |
89 | using std::ostream; | |
90 | using std::ostringstream; | |
91 | using std::set; | |
92 | using std::string; | |
93 | using std::stringstream; | |
94 | using std::vector; | |
95 | ||
96 | using ceph::bufferlist; | |
97 | using ceph::bufferptr; | |
98 | using ceph::coarse_mono_clock; | |
99 | using ceph::decode; | |
100 | using ceph::encode; | |
101 | using ceph::Formatter; | |
102 | using ceph::JSONFormatter; | |
103 | using ceph::make_timespan; | |
104 | using ceph::mono_clock; | |
105 | using ceph::mono_time; | |
106 | using ceph::timespan_str; | |
7c673cae FG |
107 | |
108 | // kv store prefixes | |
11fdf7f2 TL |
109 | const string PREFIX_SUPER = "S"; // field -> value |
110 | const string PREFIX_STAT = "T"; // field -> value(int64 array) | |
111 | const string PREFIX_COLL = "C"; // collection name -> cnode_t | |
112 | const string PREFIX_OBJ = "O"; // object name -> onode_t | |
113 | const string PREFIX_OMAP = "M"; // u64 + keyname -> value | |
114 | const string PREFIX_PGMETA_OMAP = "P"; // u64 + keyname -> value(for meta coll) | |
9f95a23c | 115 | const string PREFIX_PERPOOL_OMAP = "m"; // s64 + u64 + keyname -> value |
f67539c2 | 116 | const string PREFIX_PERPG_OMAP = "p"; // u64(pool) + u32(hash) + u64(id) + keyname -> value |
11fdf7f2 TL |
117 | const string PREFIX_DEFERRED = "L"; // id -> deferred_transaction_t |
118 | const string PREFIX_ALLOC = "B"; // u64 offset -> u64 length (freelist) | |
119 | const string PREFIX_ALLOC_BITMAP = "b";// (see BitmapFreelistManager) | |
7c673cae | 120 | const string PREFIX_SHARED_BLOB = "X"; // u64 offset -> shared_blob_t |
f67539c2 TL |
121 | const string PREFIX_ZONED_FM_META = "Z"; // (see ZonedFreelistManager) |
122 | const string PREFIX_ZONED_FM_INFO = "z"; // (see ZonedFreelistManager) | |
123 | const string PREFIX_ZONED_CL_INFO = "G"; // (per-zone cleaner metadata) | |
7c673cae | 124 | |
11fdf7f2 TL |
125 | const string BLUESTORE_GLOBAL_STATFS_KEY = "bluestore_statfs"; |
126 | ||
7c673cae FG |
127 | // write a label in the first block. always use this size. note that |
128 | // bluefs makes a matching assumption about the location of its | |
129 | // superblock (always the second block of the device). | |
130 | #define BDEV_LABEL_BLOCK_SIZE 4096 | |
131 | ||
132 | // reserve: label (4k) + bluefs super (4k), which means we start at 8k. | |
133 | #define SUPER_RESERVED 8192 | |
134 | ||
135 | #define OBJECT_MAX_SIZE 0xffffffff // 32 bits | |
136 | ||
137 | ||
138 | /* | |
139 | * extent map blob encoding | |
140 | * | |
141 | * we use the low bits of the blobid field to indicate some common scenarios | |
142 | * and spanning vs local ids. See ExtentMap::{encode,decode}_some(). | |
143 | */ | |
144 | #define BLOBID_FLAG_CONTIGUOUS 0x1 // this extent starts at end of previous | |
145 | #define BLOBID_FLAG_ZEROOFFSET 0x2 // blob_offset is 0 | |
146 | #define BLOBID_FLAG_SAMELENGTH 0x4 // length matches previous extent | |
147 | #define BLOBID_FLAG_SPANNING 0x8 // has spanning blob id | |
148 | #define BLOBID_SHIFT_BITS 4 | |
149 | ||
150 | /* | |
151 | * object name key structure | |
152 | * | |
153 | * encoded u8: shard + 2^7 (so that it sorts properly) | |
154 | * encoded u64: poolid + 2^63 (so that it sorts properly) | |
155 | * encoded u32: hash (bit reversed) | |
156 | * | |
157 | * escaped string: namespace | |
158 | * | |
159 | * escaped string: key or object name | |
160 | * 1 char: '<', '=', or '>'. if =, then object key == object name, and | |
161 | * we are done. otherwise, we are followed by the object name. | |
162 | * escaped string: object name (unless '=' above) | |
163 | * | |
164 | * encoded u64: snap | |
165 | * encoded u64: generation | |
166 | * 'o' | |
167 | */ | |
168 | #define ONODE_KEY_SUFFIX 'o' | |
169 | ||
170 | /* | |
171 | * extent shard key | |
172 | * | |
173 | * object prefix key | |
174 | * u32 | |
175 | * 'x' | |
176 | */ | |
177 | #define EXTENT_SHARD_KEY_SUFFIX 'x' | |
178 | ||
179 | /* | |
180 | * string encoding in the key | |
181 | * | |
182 | * The key string needs to lexicographically sort the same way that | |
183 | * ghobject_t does. We do this by escaping anything <= to '#' with # | |
184 | * plus a 2 digit hex string, and anything >= '~' with ~ plus the two | |
185 | * hex digits. | |
186 | * | |
187 | * We use ! as a terminator for strings; this works because it is < # | |
188 | * and will get escaped if it is present in the string. | |
189 | * | |
f91f0fd5 TL |
190 | * NOTE: There is a bug in this implementation: due to implicit |
191 | * character type conversion in comparison it may produce unexpected | |
192 | * ordering. Unfortunately fixing the bug would mean invalidating the | |
193 | * keys in existing deployments. Instead we do additional sorting | |
194 | * where it is needed. | |
7c673cae FG |
195 | */ |
196 | template<typename S> | |
197 | static void append_escaped(const string &in, S *out) | |
198 | { | |
224ce89b WB |
199 | char hexbyte[in.length() * 3 + 1]; |
200 | char* ptr = &hexbyte[0]; | |
7c673cae | 201 | for (string::const_iterator i = in.begin(); i != in.end(); ++i) { |
f91f0fd5 | 202 | if (*i <= '#') { // bug: unexpected result for *i > 0x7f |
224ce89b WB |
203 | *ptr++ = '#'; |
204 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
205 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
f91f0fd5 | 206 | } else if (*i >= '~') { // bug: unexpected result for *i > 0x7f |
224ce89b WB |
207 | *ptr++ = '~'; |
208 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
209 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 210 | } else { |
224ce89b | 211 | *ptr++ = *i; |
7c673cae FG |
212 | } |
213 | } | |
224ce89b WB |
214 | *ptr++ = '!'; |
215 | out->append(hexbyte, ptr - &hexbyte[0]); | |
216 | } | |
217 | ||
218 | inline unsigned h2i(char c) | |
219 | { | |
220 | if ((c >= '0') && (c <= '9')) { | |
221 | return c - 0x30; | |
222 | } else if ((c >= 'a') && (c <= 'f')) { | |
223 | return c - 'a' + 10; | |
224 | } else if ((c >= 'A') && (c <= 'F')) { | |
225 | return c - 'A' + 10; | |
226 | } else { | |
227 | return 256; // make it always larger than 255 | |
228 | } | |
7c673cae FG |
229 | } |
230 | ||
231 | static int decode_escaped(const char *p, string *out) | |
232 | { | |
224ce89b WB |
233 | char buff[256]; |
234 | char* ptr = &buff[0]; | |
235 | char* max = &buff[252]; | |
7c673cae FG |
236 | const char *orig_p = p; |
237 | while (*p && *p != '!') { | |
238 | if (*p == '#' || *p == '~') { | |
224ce89b WB |
239 | unsigned hex = 0; |
240 | p++; | |
241 | hex = h2i(*p++) << 4; | |
242 | if (hex > 255) { | |
243 | return -EINVAL; | |
244 | } | |
245 | hex |= h2i(*p++); | |
246 | if (hex > 255) { | |
247 | return -EINVAL; | |
248 | } | |
249 | *ptr++ = hex; | |
7c673cae | 250 | } else { |
224ce89b WB |
251 | *ptr++ = *p++; |
252 | } | |
253 | if (ptr > max) { | |
254 | out->append(buff, ptr-buff); | |
255 | ptr = &buff[0]; | |
7c673cae FG |
256 | } |
257 | } | |
224ce89b WB |
258 | if (ptr != buff) { |
259 | out->append(buff, ptr-buff); | |
260 | } | |
7c673cae FG |
261 | return p - orig_p; |
262 | } | |
263 | ||
7c673cae FG |
264 | template<typename T> |
265 | static void _key_encode_shard(shard_id_t shard, T *key) | |
266 | { | |
267 | key->push_back((char)((uint8_t)shard.id + (uint8_t)0x80)); | |
268 | } | |
269 | ||
270 | static const char *_key_decode_shard(const char *key, shard_id_t *pshard) | |
271 | { | |
272 | pshard->id = (uint8_t)*key - (uint8_t)0x80; | |
273 | return key + 1; | |
274 | } | |
275 | ||
f91f0fd5 | 276 | static void get_coll_range(const coll_t& cid, int bits, |
f67539c2 TL |
277 | ghobject_t *temp_start, ghobject_t *temp_end, |
278 | ghobject_t *start, ghobject_t *end) | |
7c673cae | 279 | { |
7c673cae FG |
280 | spg_t pgid; |
281 | if (cid.is_pg(&pgid)) { | |
f91f0fd5 | 282 | start->shard_id = pgid.shard; |
7c673cae FG |
283 | *temp_start = *start; |
284 | ||
f91f0fd5 TL |
285 | start->hobj.pool = pgid.pool(); |
286 | temp_start->hobj.pool = -2ll - pgid.pool(); | |
7c673cae FG |
287 | |
288 | *end = *start; | |
289 | *temp_end = *temp_start; | |
290 | ||
291 | uint32_t reverse_hash = hobject_t::_reverse_bits(pgid.ps()); | |
f91f0fd5 TL |
292 | start->hobj.set_bitwise_key_u32(reverse_hash); |
293 | temp_start->hobj.set_bitwise_key_u32(reverse_hash); | |
7c673cae FG |
294 | |
295 | uint64_t end_hash = reverse_hash + (1ull << (32 - bits)); | |
296 | if (end_hash > 0xffffffffull) | |
297 | end_hash = 0xffffffffull; | |
298 | ||
f91f0fd5 TL |
299 | end->hobj.set_bitwise_key_u32(end_hash); |
300 | temp_end->hobj.set_bitwise_key_u32(end_hash); | |
7c673cae | 301 | } else { |
f91f0fd5 TL |
302 | start->shard_id = shard_id_t::NO_SHARD; |
303 | start->hobj.pool = -1ull; | |
304 | ||
7c673cae | 305 | *end = *start; |
f91f0fd5 TL |
306 | start->hobj.set_bitwise_key_u32(0); |
307 | end->hobj.set_bitwise_key_u32(0xffffffff); | |
7c673cae FG |
308 | |
309 | // no separate temp section | |
310 | *temp_start = *end; | |
311 | *temp_end = *end; | |
312 | } | |
f91f0fd5 TL |
313 | |
314 | start->generation = 0; | |
315 | end->generation = 0; | |
316 | temp_start->generation = 0; | |
317 | temp_end->generation = 0; | |
7c673cae FG |
318 | } |
319 | ||
320 | static void get_shared_blob_key(uint64_t sbid, string *key) | |
321 | { | |
322 | key->clear(); | |
323 | _key_encode_u64(sbid, key); | |
324 | } | |
325 | ||
326 | static int get_key_shared_blob(const string& key, uint64_t *sbid) | |
327 | { | |
328 | const char *p = key.c_str(); | |
329 | if (key.length() < sizeof(uint64_t)) | |
330 | return -1; | |
224ce89b | 331 | _key_decode_u64(p, sbid); |
7c673cae FG |
332 | return 0; |
333 | } | |
334 | ||
335 | template<typename S> | |
f91f0fd5 | 336 | static void _key_encode_prefix(const ghobject_t& oid, S *key) |
7c673cae | 337 | { |
f91f0fd5 TL |
338 | _key_encode_shard(oid.shard_id, key); |
339 | _key_encode_u64(oid.hobj.pool + 0x8000000000000000ull, key); | |
340 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), key); | |
341 | } | |
7c673cae | 342 | |
f91f0fd5 TL |
343 | static const char *_key_decode_prefix(const char *p, ghobject_t *oid) |
344 | { | |
7c673cae FG |
345 | p = _key_decode_shard(p, &oid->shard_id); |
346 | ||
347 | uint64_t pool; | |
348 | p = _key_decode_u64(p, &pool); | |
349 | oid->hobj.pool = pool - 0x8000000000000000ull; | |
350 | ||
351 | unsigned hash; | |
352 | p = _key_decode_u32(p, &hash); | |
353 | ||
354 | oid->hobj.set_bitwise_key_u32(hash); | |
355 | ||
f91f0fd5 TL |
356 | return p; |
357 | } | |
358 | ||
359 | #define ENCODED_KEY_PREFIX_LEN (1 + 8 + 4) | |
360 | ||
361 | template<typename S> | |
362 | static int get_key_object(const S& key, ghobject_t *oid) | |
363 | { | |
364 | int r; | |
365 | const char *p = key.c_str(); | |
366 | ||
367 | if (key.length() < ENCODED_KEY_PREFIX_LEN) | |
368 | return -1; | |
369 | ||
370 | p = _key_decode_prefix(p, oid); | |
371 | ||
372 | if (key.length() == ENCODED_KEY_PREFIX_LEN) | |
373 | return -2; | |
374 | ||
7c673cae FG |
375 | r = decode_escaped(p, &oid->hobj.nspace); |
376 | if (r < 0) | |
377 | return -2; | |
378 | p += r + 1; | |
379 | ||
380 | string k; | |
381 | r = decode_escaped(p, &k); | |
382 | if (r < 0) | |
383 | return -3; | |
384 | p += r + 1; | |
385 | if (*p == '=') { | |
386 | // no key | |
387 | ++p; | |
388 | oid->hobj.oid.name = k; | |
389 | } else if (*p == '<' || *p == '>') { | |
390 | // key + name | |
391 | ++p; | |
392 | r = decode_escaped(p, &oid->hobj.oid.name); | |
393 | if (r < 0) | |
394 | return -5; | |
395 | p += r + 1; | |
396 | oid->hobj.set_key(k); | |
397 | } else { | |
398 | // malformed | |
399 | return -6; | |
400 | } | |
401 | ||
402 | p = _key_decode_u64(p, &oid->hobj.snap.val); | |
403 | p = _key_decode_u64(p, &oid->generation); | |
404 | ||
405 | if (*p != ONODE_KEY_SUFFIX) { | |
406 | return -7; | |
407 | } | |
408 | p++; | |
409 | if (*p) { | |
410 | // if we get something other than a null terminator here, | |
411 | // something goes wrong. | |
412 | return -8; | |
413 | } | |
414 | ||
415 | return 0; | |
416 | } | |
417 | ||
418 | template<typename S> | |
419 | static void get_object_key(CephContext *cct, const ghobject_t& oid, S *key) | |
420 | { | |
421 | key->clear(); | |
422 | ||
f91f0fd5 | 423 | size_t max_len = ENCODED_KEY_PREFIX_LEN + |
7c673cae FG |
424 | (oid.hobj.nspace.length() * 3 + 1) + |
425 | (oid.hobj.get_key().length() * 3 + 1) + | |
426 | 1 + // for '<', '=', or '>' | |
427 | (oid.hobj.oid.name.length() * 3 + 1) + | |
428 | 8 + 8 + 1; | |
429 | key->reserve(max_len); | |
430 | ||
f91f0fd5 | 431 | _key_encode_prefix(oid, key); |
7c673cae FG |
432 | |
433 | append_escaped(oid.hobj.nspace, key); | |
434 | ||
435 | if (oid.hobj.get_key().length()) { | |
436 | // is a key... could be < = or >. | |
437 | append_escaped(oid.hobj.get_key(), key); | |
438 | // (ASCII chars < = and > sort in that order, yay) | |
439 | int r = oid.hobj.get_key().compare(oid.hobj.oid.name); | |
440 | if (r) { | |
441 | key->append(r > 0 ? ">" : "<"); | |
442 | append_escaped(oid.hobj.oid.name, key); | |
443 | } else { | |
444 | // same as no key | |
445 | key->append("="); | |
446 | } | |
447 | } else { | |
448 | // no key | |
449 | append_escaped(oid.hobj.oid.name, key); | |
450 | key->append("="); | |
451 | } | |
452 | ||
453 | _key_encode_u64(oid.hobj.snap, key); | |
454 | _key_encode_u64(oid.generation, key); | |
455 | ||
456 | key->push_back(ONODE_KEY_SUFFIX); | |
457 | ||
458 | // sanity check | |
459 | if (true) { | |
460 | ghobject_t t; | |
461 | int r = get_key_object(*key, &t); | |
462 | if (r || t != oid) { | |
463 | derr << " r " << r << dendl; | |
464 | derr << "key " << pretty_binary_string(*key) << dendl; | |
465 | derr << "oid " << oid << dendl; | |
466 | derr << " t " << t << dendl; | |
11fdf7f2 | 467 | ceph_assert(r == 0 && t == oid); |
7c673cae FG |
468 | } |
469 | } | |
470 | } | |
471 | ||
7c673cae FG |
472 | // extent shard keys are the onode key, plus a u32, plus 'x'. the trailing |
473 | // char lets us quickly test whether it is a shard key without decoding any | |
474 | // of the prefix bytes. | |
475 | template<typename S> | |
476 | static void get_extent_shard_key(const S& onode_key, uint32_t offset, | |
477 | string *key) | |
478 | { | |
479 | key->clear(); | |
480 | key->reserve(onode_key.length() + 4 + 1); | |
481 | key->append(onode_key.c_str(), onode_key.size()); | |
482 | _key_encode_u32(offset, key); | |
483 | key->push_back(EXTENT_SHARD_KEY_SUFFIX); | |
484 | } | |
485 | ||
486 | static void rewrite_extent_shard_key(uint32_t offset, string *key) | |
487 | { | |
11fdf7f2 TL |
488 | ceph_assert(key->size() > sizeof(uint32_t) + 1); |
489 | ceph_assert(*key->rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
490 | _key_encode_u32(offset, key->size() - sizeof(uint32_t) - 1, key); |
491 | } | |
492 | ||
493 | template<typename S> | |
494 | static void generate_extent_shard_key_and_apply( | |
495 | const S& onode_key, | |
496 | uint32_t offset, | |
497 | string *key, | |
498 | std::function<void(const string& final_key)> apply) | |
499 | { | |
500 | if (key->empty()) { // make full key | |
11fdf7f2 | 501 | ceph_assert(!onode_key.empty()); |
7c673cae FG |
502 | get_extent_shard_key(onode_key, offset, key); |
503 | } else { | |
504 | rewrite_extent_shard_key(offset, key); | |
505 | } | |
506 | apply(*key); | |
507 | } | |
508 | ||
509 | int get_key_extent_shard(const string& key, string *onode_key, uint32_t *offset) | |
510 | { | |
11fdf7f2 TL |
511 | ceph_assert(key.size() > sizeof(uint32_t) + 1); |
512 | ceph_assert(*key.rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
513 | int okey_len = key.size() - sizeof(uint32_t) - 1; |
514 | *onode_key = key.substr(0, okey_len); | |
515 | const char *p = key.data() + okey_len; | |
224ce89b | 516 | _key_decode_u32(p, offset); |
7c673cae FG |
517 | return 0; |
518 | } | |
519 | ||
520 | static bool is_extent_shard_key(const string& key) | |
521 | { | |
522 | return *key.rbegin() == EXTENT_SHARD_KEY_SUFFIX; | |
523 | } | |
524 | ||
7c673cae FG |
525 | static void get_deferred_key(uint64_t seq, string *out) |
526 | { | |
527 | _key_encode_u64(seq, out); | |
528 | } | |
529 | ||
11fdf7f2 TL |
530 | static void get_pool_stat_key(int64_t pool_id, string *key) |
531 | { | |
532 | key->clear(); | |
533 | _key_encode_u64(pool_id, key); | |
534 | } | |
535 | ||
536 | static int get_key_pool_stat(const string& key, uint64_t* pool_id) | |
537 | { | |
538 | const char *p = key.c_str(); | |
539 | if (key.length() < sizeof(uint64_t)) | |
540 | return -1; | |
541 | _key_decode_u64(p, pool_id); | |
542 | return 0; | |
543 | } | |
7c673cae | 544 | |
81eedcae TL |
545 | template <int LogLevelV> |
546 | void _dump_extent_map(CephContext *cct, const BlueStore::ExtentMap &em) | |
547 | { | |
548 | uint64_t pos = 0; | |
549 | for (auto& s : em.shards) { | |
550 | dout(LogLevelV) << __func__ << " shard " << *s.shard_info | |
551 | << (s.loaded ? " (loaded)" : "") | |
552 | << (s.dirty ? " (dirty)" : "") | |
553 | << dendl; | |
554 | } | |
555 | for (auto& e : em.extent_map) { | |
556 | dout(LogLevelV) << __func__ << " " << e << dendl; | |
557 | ceph_assert(e.logical_offset >= pos); | |
558 | pos = e.logical_offset + e.length; | |
559 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
560 | if (blob.has_csum()) { | |
561 | vector<uint64_t> v; | |
562 | unsigned n = blob.get_csum_count(); | |
563 | for (unsigned i = 0; i < n; ++i) | |
564 | v.push_back(blob.get_csum_item(i)); | |
565 | dout(LogLevelV) << __func__ << " csum: " << std::hex << v << std::dec | |
566 | << dendl; | |
567 | } | |
568 | std::lock_guard l(e.blob->shared_blob->get_cache()->lock); | |
569 | for (auto& i : e.blob->shared_blob->bc.buffer_map) { | |
570 | dout(LogLevelV) << __func__ << " 0x" << std::hex << i.first | |
571 | << "~" << i.second->length << std::dec | |
572 | << " " << *i.second << dendl; | |
573 | } | |
574 | } | |
575 | } | |
576 | ||
577 | template <int LogLevelV> | |
578 | void _dump_onode(CephContext *cct, const BlueStore::Onode& o) | |
579 | { | |
580 | if (!cct->_conf->subsys.should_gather<ceph_subsys_bluestore, LogLevelV>()) | |
581 | return; | |
582 | dout(LogLevelV) << __func__ << " " << &o << " " << o.oid | |
583 | << " nid " << o.onode.nid | |
584 | << " size 0x" << std::hex << o.onode.size | |
585 | << " (" << std::dec << o.onode.size << ")" | |
586 | << " expected_object_size " << o.onode.expected_object_size | |
587 | << " expected_write_size " << o.onode.expected_write_size | |
588 | << " in " << o.onode.extent_map_shards.size() << " shards" | |
589 | << ", " << o.extent_map.spanning_blob_map.size() | |
590 | << " spanning blobs" | |
591 | << dendl; | |
592 | for (auto p = o.onode.attrs.begin(); | |
593 | p != o.onode.attrs.end(); | |
594 | ++p) { | |
595 | dout(LogLevelV) << __func__ << " attr " << p->first | |
596 | << " len " << p->second.length() << dendl; | |
597 | } | |
598 | _dump_extent_map<LogLevelV>(cct, o.extent_map); | |
599 | } | |
600 | ||
601 | template <int LogLevelV> | |
602 | void _dump_transaction(CephContext *cct, ObjectStore::Transaction *t) | |
603 | { | |
604 | dout(LogLevelV) << __func__ << " transaction dump:\n"; | |
605 | JSONFormatter f(true); | |
606 | f.open_object_section("transaction"); | |
607 | t->dump(&f); | |
608 | f.close_section(); | |
609 | f.flush(*_dout); | |
610 | *_dout << dendl; | |
611 | } | |
612 | ||
7c673cae FG |
613 | // Buffer |
614 | ||
615 | ostream& operator<<(ostream& out, const BlueStore::Buffer& b) | |
616 | { | |
617 | out << "buffer(" << &b << " space " << b.space << " 0x" << std::hex | |
618 | << b.offset << "~" << b.length << std::dec | |
619 | << " " << BlueStore::Buffer::get_state_name(b.state); | |
620 | if (b.flags) | |
621 | out << " " << BlueStore::Buffer::get_flag_name(b.flags); | |
622 | return out << ")"; | |
623 | } | |
624 | ||
f91f0fd5 TL |
625 | namespace { |
626 | ||
627 | /* | |
628 | * Due to a bug in key string encoding (see a comment for append_escaped) | |
629 | * the KeyValueDB iterator does not lexicographically sort the same | |
630 | * way that ghobject_t does: objects with the same hash may have wrong order. | |
631 | * | |
632 | * This is the iterator wrapper that fixes the keys order. | |
633 | */ | |
634 | ||
635 | class CollectionListIterator { | |
636 | public: | |
637 | CollectionListIterator(const KeyValueDB::Iterator &it) | |
638 | : m_it(it) { | |
639 | } | |
640 | virtual ~CollectionListIterator() { | |
641 | } | |
642 | ||
643 | virtual bool valid() const = 0; | |
644 | virtual const ghobject_t &oid() const = 0; | |
645 | virtual void lower_bound(const ghobject_t &oid) = 0; | |
646 | virtual void upper_bound(const ghobject_t &oid) = 0; | |
647 | virtual void next() = 0; | |
648 | ||
adb31ebb TL |
649 | virtual int cmp(const ghobject_t &oid) const = 0; |
650 | ||
651 | bool is_ge(const ghobject_t &oid) const { | |
652 | return cmp(oid) >= 0; | |
653 | } | |
654 | ||
655 | bool is_lt(const ghobject_t &oid) const { | |
656 | return cmp(oid) < 0; | |
657 | } | |
658 | ||
f91f0fd5 TL |
659 | protected: |
660 | KeyValueDB::Iterator m_it; | |
661 | }; | |
662 | ||
663 | class SimpleCollectionListIterator : public CollectionListIterator { | |
664 | public: | |
665 | SimpleCollectionListIterator(CephContext *cct, const KeyValueDB::Iterator &it) | |
666 | : CollectionListIterator(it), m_cct(cct) { | |
667 | } | |
668 | ||
669 | bool valid() const override { | |
670 | return m_it->valid(); | |
671 | } | |
672 | ||
673 | const ghobject_t &oid() const override { | |
674 | ceph_assert(valid()); | |
675 | ||
676 | return m_oid; | |
677 | } | |
678 | ||
679 | void lower_bound(const ghobject_t &oid) override { | |
680 | string key; | |
681 | get_object_key(m_cct, oid, &key); | |
682 | ||
683 | m_it->lower_bound(key); | |
684 | get_oid(); | |
685 | } | |
686 | ||
687 | void upper_bound(const ghobject_t &oid) override { | |
688 | string key; | |
689 | get_object_key(m_cct, oid, &key); | |
690 | ||
691 | m_it->upper_bound(key); | |
692 | get_oid(); | |
693 | } | |
694 | ||
695 | void next() override { | |
696 | ceph_assert(valid()); | |
697 | ||
698 | m_it->next(); | |
699 | get_oid(); | |
700 | } | |
701 | ||
adb31ebb TL |
702 | int cmp(const ghobject_t &oid) const override { |
703 | ceph_assert(valid()); | |
704 | ||
705 | string key; | |
706 | get_object_key(m_cct, oid, &key); | |
707 | ||
708 | return m_it->key().compare(key); | |
709 | } | |
710 | ||
f91f0fd5 TL |
711 | private: |
712 | CephContext *m_cct; | |
713 | ghobject_t m_oid; | |
714 | ||
715 | void get_oid() { | |
f67539c2 TL |
716 | m_oid = ghobject_t(); |
717 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
718 | m_it->next(); | |
f91f0fd5 | 719 | } |
f67539c2 | 720 | if (!valid()) { |
f91f0fd5 TL |
721 | return; |
722 | } | |
723 | ||
f91f0fd5 TL |
724 | int r = get_key_object(m_it->key(), &m_oid); |
725 | ceph_assert(r == 0); | |
726 | } | |
727 | }; | |
728 | ||
729 | class SortedCollectionListIterator : public CollectionListIterator { | |
730 | public: | |
731 | SortedCollectionListIterator(const KeyValueDB::Iterator &it) | |
732 | : CollectionListIterator(it), m_chunk_iter(m_chunk.end()) { | |
733 | } | |
734 | ||
735 | bool valid() const override { | |
736 | return m_chunk_iter != m_chunk.end(); | |
737 | } | |
738 | ||
739 | const ghobject_t &oid() const override { | |
740 | ceph_assert(valid()); | |
741 | ||
742 | return m_chunk_iter->first; | |
743 | } | |
744 | ||
745 | void lower_bound(const ghobject_t &oid) override { | |
746 | std::string key; | |
747 | _key_encode_prefix(oid, &key); | |
748 | ||
749 | m_it->lower_bound(key); | |
750 | m_chunk_iter = m_chunk.end(); | |
751 | if (!get_next_chunk()) { | |
752 | return; | |
753 | } | |
754 | ||
755 | if (this->oid().shard_id != oid.shard_id || | |
756 | this->oid().hobj.pool != oid.hobj.pool || | |
757 | this->oid().hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
758 | return; | |
759 | } | |
760 | ||
761 | m_chunk_iter = m_chunk.lower_bound(oid); | |
762 | if (m_chunk_iter == m_chunk.end()) { | |
763 | get_next_chunk(); | |
764 | } | |
765 | } | |
766 | ||
767 | void upper_bound(const ghobject_t &oid) override { | |
768 | lower_bound(oid); | |
769 | ||
770 | if (valid() && this->oid() == oid) { | |
771 | next(); | |
772 | } | |
773 | } | |
774 | ||
775 | void next() override { | |
776 | ceph_assert(valid()); | |
777 | ||
778 | m_chunk_iter++; | |
779 | if (m_chunk_iter == m_chunk.end()) { | |
780 | get_next_chunk(); | |
781 | } | |
782 | } | |
783 | ||
adb31ebb TL |
784 | int cmp(const ghobject_t &oid) const override { |
785 | ceph_assert(valid()); | |
786 | ||
787 | if (this->oid() < oid) { | |
788 | return -1; | |
789 | } | |
790 | if (this->oid() > oid) { | |
791 | return 1; | |
792 | } | |
793 | return 0; | |
794 | } | |
795 | ||
f91f0fd5 TL |
796 | private: |
797 | std::map<ghobject_t, std::string> m_chunk; | |
798 | std::map<ghobject_t, std::string>::iterator m_chunk_iter; | |
799 | ||
800 | bool get_next_chunk() { | |
801 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
802 | m_it->next(); | |
803 | } | |
804 | ||
805 | if (!m_it->valid()) { | |
806 | return false; | |
807 | } | |
808 | ||
809 | ghobject_t oid; | |
810 | int r = get_key_object(m_it->key(), &oid); | |
811 | ceph_assert(r == 0); | |
812 | ||
813 | m_chunk.clear(); | |
814 | while (true) { | |
815 | m_chunk.insert({oid, m_it->key()}); | |
816 | ||
817 | do { | |
818 | m_it->next(); | |
819 | } while (m_it->valid() && is_extent_shard_key(m_it->key())); | |
820 | ||
821 | if (!m_it->valid()) { | |
822 | break; | |
823 | } | |
824 | ||
825 | ghobject_t next; | |
826 | r = get_key_object(m_it->key(), &next); | |
827 | ceph_assert(r == 0); | |
828 | if (next.shard_id != oid.shard_id || | |
829 | next.hobj.pool != oid.hobj.pool || | |
830 | next.hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
831 | break; | |
832 | } | |
833 | oid = next; | |
834 | } | |
835 | ||
836 | m_chunk_iter = m_chunk.begin(); | |
837 | return true; | |
838 | } | |
839 | }; | |
840 | ||
841 | } // anonymous namespace | |
842 | ||
7c673cae FG |
843 | // Garbage Collector |
844 | ||
845 | void BlueStore::GarbageCollector::process_protrusive_extents( | |
846 | const BlueStore::ExtentMap& extent_map, | |
847 | uint64_t start_offset, | |
848 | uint64_t end_offset, | |
849 | uint64_t start_touch_offset, | |
850 | uint64_t end_touch_offset, | |
851 | uint64_t min_alloc_size) | |
852 | { | |
11fdf7f2 | 853 | ceph_assert(start_offset <= start_touch_offset && end_offset>= end_touch_offset); |
7c673cae | 854 | |
11fdf7f2 TL |
855 | uint64_t lookup_start_offset = p2align(start_offset, min_alloc_size); |
856 | uint64_t lookup_end_offset = round_up_to(end_offset, min_alloc_size); | |
7c673cae FG |
857 | |
858 | dout(30) << __func__ << " (hex): [" << std::hex | |
859 | << lookup_start_offset << ", " << lookup_end_offset | |
860 | << ")" << std::dec << dendl; | |
861 | ||
862 | for (auto it = extent_map.seek_lextent(lookup_start_offset); | |
863 | it != extent_map.extent_map.end() && | |
864 | it->logical_offset < lookup_end_offset; | |
865 | ++it) { | |
866 | uint64_t alloc_unit_start = it->logical_offset / min_alloc_size; | |
867 | uint64_t alloc_unit_end = (it->logical_end() - 1) / min_alloc_size; | |
868 | ||
869 | dout(30) << __func__ << " " << *it | |
870 | << "alloc_units: " << alloc_unit_start << ".." << alloc_unit_end | |
871 | << dendl; | |
872 | ||
873 | Blob* b = it->blob.get(); | |
874 | ||
875 | if (it->logical_offset >=start_touch_offset && | |
876 | it->logical_end() <= end_touch_offset) { | |
877 | // Process extents within the range affected by | |
878 | // the current write request. | |
879 | // Need to take into account if existing extents | |
880 | // can be merged with them (uncompressed case) | |
881 | if (!b->get_blob().is_compressed()) { | |
882 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
883 | --blob_info_counted->expected_allocations; // don't need to allocate | |
884 | // new AU for compressed | |
885 | // data since another | |
886 | // collocated uncompressed | |
887 | // blob already exists | |
888 | dout(30) << __func__ << " --expected:" | |
889 | << alloc_unit_start << dendl; | |
890 | } | |
891 | used_alloc_unit = alloc_unit_end; | |
892 | blob_info_counted = nullptr; | |
893 | } | |
894 | } else if (b->get_blob().is_compressed()) { | |
895 | ||
896 | // additionally we take compressed blobs that were not impacted | |
897 | // by the write into account too | |
898 | BlobInfo& bi = | |
899 | affected_blobs.emplace( | |
900 | b, BlobInfo(b->get_referenced_bytes())).first->second; | |
901 | ||
902 | int adjust = | |
903 | (used_alloc_unit && used_alloc_unit == alloc_unit_start) ? 0 : 1; | |
904 | bi.expected_allocations += alloc_unit_end - alloc_unit_start + adjust; | |
905 | dout(30) << __func__ << " expected_allocations=" | |
906 | << bi.expected_allocations << " end_au:" | |
907 | << alloc_unit_end << dendl; | |
908 | ||
909 | blob_info_counted = &bi; | |
910 | used_alloc_unit = alloc_unit_end; | |
911 | ||
11fdf7f2 | 912 | ceph_assert(it->length <= bi.referenced_bytes); |
7c673cae FG |
913 | bi.referenced_bytes -= it->length; |
914 | dout(30) << __func__ << " affected_blob:" << *b | |
915 | << " unref 0x" << std::hex << it->length | |
916 | << " referenced = 0x" << bi.referenced_bytes | |
917 | << std::dec << dendl; | |
918 | // NOTE: we can't move specific blob to resulting GC list here | |
919 | // when reference counter == 0 since subsequent extents might | |
920 | // decrement its expected_allocation. | |
921 | // Hence need to enumerate all the extents first. | |
922 | if (!bi.collect_candidate) { | |
923 | bi.first_lextent = it; | |
924 | bi.collect_candidate = true; | |
925 | } | |
926 | bi.last_lextent = it; | |
927 | } else { | |
928 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
929 | // don't need to allocate new AU for compressed data since another | |
930 | // collocated uncompressed blob already exists | |
931 | --blob_info_counted->expected_allocations; | |
932 | dout(30) << __func__ << " --expected_allocations:" | |
933 | << alloc_unit_start << dendl; | |
934 | } | |
935 | used_alloc_unit = alloc_unit_end; | |
936 | blob_info_counted = nullptr; | |
937 | } | |
938 | } | |
939 | ||
940 | for (auto b_it = affected_blobs.begin(); | |
941 | b_it != affected_blobs.end(); | |
942 | ++b_it) { | |
943 | Blob* b = b_it->first; | |
944 | BlobInfo& bi = b_it->second; | |
945 | if (bi.referenced_bytes == 0) { | |
946 | uint64_t len_on_disk = b_it->first->get_blob().get_ondisk_length(); | |
947 | int64_t blob_expected_for_release = | |
11fdf7f2 | 948 | round_up_to(len_on_disk, min_alloc_size) / min_alloc_size; |
7c673cae FG |
949 | |
950 | dout(30) << __func__ << " " << *(b_it->first) | |
951 | << " expected4release=" << blob_expected_for_release | |
952 | << " expected_allocations=" << bi.expected_allocations | |
953 | << dendl; | |
954 | int64_t benefit = blob_expected_for_release - bi.expected_allocations; | |
11fdf7f2 | 955 | if (benefit >= g_conf()->bluestore_gc_enable_blob_threshold) { |
7c673cae FG |
956 | if (bi.collect_candidate) { |
957 | auto it = bi.first_lextent; | |
958 | bool bExit = false; | |
959 | do { | |
960 | if (it->blob.get() == b) { | |
eafe8130 | 961 | extents_to_collect.insert(it->logical_offset, it->length); |
7c673cae FG |
962 | } |
963 | bExit = it == bi.last_lextent; | |
964 | ++it; | |
31f18b77 | 965 | } while (!bExit); |
7c673cae FG |
966 | } |
967 | expected_for_release += blob_expected_for_release; | |
968 | expected_allocations += bi.expected_allocations; | |
969 | } | |
970 | } | |
971 | } | |
972 | } | |
973 | ||
974 | int64_t BlueStore::GarbageCollector::estimate( | |
975 | uint64_t start_offset, | |
976 | uint64_t length, | |
977 | const BlueStore::ExtentMap& extent_map, | |
978 | const BlueStore::old_extent_map_t& old_extents, | |
979 | uint64_t min_alloc_size) | |
980 | { | |
981 | ||
982 | affected_blobs.clear(); | |
983 | extents_to_collect.clear(); | |
984 | used_alloc_unit = boost::optional<uint64_t >(); | |
985 | blob_info_counted = nullptr; | |
986 | ||
eafe8130 TL |
987 | uint64_t gc_start_offset = start_offset; |
988 | uint64_t gc_end_offset = start_offset + length; | |
7c673cae FG |
989 | |
990 | uint64_t end_offset = start_offset + length; | |
991 | ||
992 | for (auto it = old_extents.begin(); it != old_extents.end(); ++it) { | |
993 | Blob* b = it->e.blob.get(); | |
994 | if (b->get_blob().is_compressed()) { | |
995 | ||
996 | // update gc_start_offset/gc_end_offset if needed | |
997 | gc_start_offset = min(gc_start_offset, (uint64_t)it->e.blob_start()); | |
11fdf7f2 | 998 | gc_end_offset = std::max(gc_end_offset, (uint64_t)it->e.blob_end()); |
7c673cae FG |
999 | |
1000 | auto o = it->e.logical_offset; | |
1001 | auto l = it->e.length; | |
1002 | ||
1003 | uint64_t ref_bytes = b->get_referenced_bytes(); | |
1004 | // micro optimization to bypass blobs that have no more references | |
1005 | if (ref_bytes != 0) { | |
1006 | dout(30) << __func__ << " affected_blob:" << *b | |
1007 | << " unref 0x" << std::hex << o << "~" << l | |
1008 | << std::dec << dendl; | |
1009 | affected_blobs.emplace(b, BlobInfo(ref_bytes)); | |
1010 | } | |
1011 | } | |
1012 | } | |
1013 | dout(30) << __func__ << " gc range(hex): [" << std::hex | |
1014 | << gc_start_offset << ", " << gc_end_offset | |
1015 | << ")" << std::dec << dendl; | |
1016 | ||
1017 | // enumerate preceeding extents to check if they reference affected blobs | |
1018 | if (gc_start_offset < start_offset || gc_end_offset > end_offset) { | |
1019 | process_protrusive_extents(extent_map, | |
1020 | gc_start_offset, | |
1021 | gc_end_offset, | |
1022 | start_offset, | |
1023 | end_offset, | |
1024 | min_alloc_size); | |
1025 | } | |
1026 | return expected_for_release - expected_allocations; | |
1027 | } | |
1028 | ||
9f95a23c TL |
1029 | // LruOnodeCacheShard |
1030 | struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard { | |
1031 | typedef boost::intrusive::list< | |
1032 | BlueStore::Onode, | |
1033 | boost::intrusive::member_hook< | |
1034 | BlueStore::Onode, | |
1035 | boost::intrusive::list_member_hook<>, | |
1036 | &BlueStore::Onode::lru_item> > list_t; | |
7c673cae | 1037 | |
9f95a23c | 1038 | list_t lru; |
7c673cae | 1039 | |
9f95a23c | 1040 | explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {} |
7c673cae | 1041 | |
f6b5b4d7 | 1042 | void _add(BlueStore::Onode* o, int level) override |
9f95a23c | 1043 | { |
f6b5b4d7 | 1044 | if (o->put_cache()) { |
9f95a23c | 1045 | (level > 0) ? lru.push_front(*o) : lru.push_back(*o); |
f6b5b4d7 TL |
1046 | } else { |
1047 | ++num_pinned; | |
9f95a23c | 1048 | } |
f6b5b4d7 TL |
1049 | ++num; // we count both pinned and unpinned entries |
1050 | dout(20) << __func__ << " " << this << " " << o->oid << " added, num=" << num << dendl; | |
eafe8130 | 1051 | } |
f6b5b4d7 | 1052 | void _rm(BlueStore::Onode* o) override |
9f95a23c | 1053 | { |
f6b5b4d7 | 1054 | if (o->pop_cache()) { |
9f95a23c | 1055 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
1056 | } else { |
1057 | ceph_assert(num_pinned); | |
1058 | --num_pinned; | |
9f95a23c | 1059 | } |
f6b5b4d7 TL |
1060 | ceph_assert(num); |
1061 | --num; | |
1062 | dout(20) << __func__ << " " << this << " " << " " << o->oid << " removed, num=" << num << dendl; | |
9f95a23c | 1063 | } |
f6b5b4d7 | 1064 | void _pin(BlueStore::Onode* o) override |
9f95a23c | 1065 | { |
9f95a23c | 1066 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
1067 | ++num_pinned; |
1068 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " pinned" << dendl; | |
9f95a23c | 1069 | } |
f6b5b4d7 | 1070 | void _unpin(BlueStore::Onode* o) override |
9f95a23c | 1071 | { |
f6b5b4d7 TL |
1072 | lru.push_front(*o); |
1073 | ceph_assert(num_pinned); | |
1074 | --num_pinned; | |
1075 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " unpinned" << dendl; | |
9f95a23c | 1076 | } |
adb31ebb TL |
1077 | void _unpin_and_rm(BlueStore::Onode* o) override |
1078 | { | |
1079 | o->pop_cache(); | |
1080 | ceph_assert(num_pinned); | |
1081 | --num_pinned; | |
1082 | ceph_assert(num); | |
1083 | --num; | |
1084 | } | |
9f95a23c TL |
1085 | void _trim_to(uint64_t new_size) override |
1086 | { | |
1087 | if (new_size >= lru.size()) { | |
1088 | return; // don't even try | |
1089 | } | |
1090 | uint64_t n = lru.size() - new_size; | |
1091 | auto p = lru.end(); | |
1092 | ceph_assert(p != lru.begin()); | |
1093 | --p; | |
f6b5b4d7 TL |
1094 | ceph_assert(num >= n); |
1095 | num -= n; | |
1096 | while (n-- > 0) { | |
9f95a23c | 1097 | BlueStore::Onode *o = &*p; |
f6b5b4d7 TL |
1098 | dout(20) << __func__ << " rm " << o->oid << " " |
1099 | << o->nref << " " << o->cached << " " << o->pinned << dendl; | |
9f95a23c TL |
1100 | if (p != lru.begin()) { |
1101 | lru.erase(p--); | |
1102 | } else { | |
f6b5b4d7 | 1103 | ceph_assert(n == 0); |
9f95a23c | 1104 | lru.erase(p); |
9f95a23c | 1105 | } |
f6b5b4d7 TL |
1106 | auto pinned = !o->pop_cache(); |
1107 | ceph_assert(!pinned); | |
1108 | o->c->onode_map._remove(o->oid); | |
9f95a23c | 1109 | } |
f6b5b4d7 TL |
1110 | } |
1111 | void move_pinned(OnodeCacheShard *to, BlueStore::Onode *o) override | |
1112 | { | |
1113 | if (to == this) { | |
1114 | return; | |
1115 | } | |
1116 | ceph_assert(o->cached); | |
1117 | ceph_assert(o->pinned); | |
1118 | ceph_assert(num); | |
1119 | ceph_assert(num_pinned); | |
1120 | --num_pinned; | |
1121 | --num; | |
1122 | ++to->num_pinned; | |
1123 | ++to->num; | |
9f95a23c TL |
1124 | } |
1125 | void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override | |
1126 | { | |
f6b5b4d7 | 1127 | *onodes += num; |
9f95a23c TL |
1128 | *pinned_onodes += num_pinned; |
1129 | } | |
1130 | }; | |
7c673cae | 1131 | |
9f95a23c TL |
1132 | // OnodeCacheShard |
1133 | BlueStore::OnodeCacheShard *BlueStore::OnodeCacheShard::create( | |
1134 | CephContext* cct, | |
1135 | string type, | |
1136 | PerfCounters *logger) | |
7c673cae | 1137 | { |
9f95a23c TL |
1138 | BlueStore::OnodeCacheShard *c = nullptr; |
1139 | // Currently we only implement an LRU cache for onodes | |
1140 | c = new LruOnodeCacheShard(cct); | |
1141 | c->logger = logger; | |
1142 | return c; | |
7c673cae FG |
1143 | } |
1144 | ||
9f95a23c TL |
1145 | // LruBufferCacheShard |
1146 | struct LruBufferCacheShard : public BlueStore::BufferCacheShard { | |
1147 | typedef boost::intrusive::list< | |
1148 | BlueStore::Buffer, | |
1149 | boost::intrusive::member_hook< | |
1150 | BlueStore::Buffer, | |
1151 | boost::intrusive::list_member_hook<>, | |
1152 | &BlueStore::Buffer::lru_item> > list_t; | |
1153 | list_t lru; | |
1154 | ||
1155 | explicit LruBufferCacheShard(CephContext *cct) : BlueStore::BufferCacheShard(cct) {} | |
1156 | ||
1157 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override { | |
1158 | if (near) { | |
1159 | auto q = lru.iterator_to(*near); | |
1160 | lru.insert(q, *b); | |
1161 | } else if (level > 0) { | |
1162 | lru.push_front(*b); | |
1163 | } else { | |
1164 | lru.push_back(*b); | |
7c673cae | 1165 | } |
9f95a23c TL |
1166 | buffer_bytes += b->length; |
1167 | num = lru.size(); | |
1168 | } | |
1169 | void _rm(BlueStore::Buffer *b) override { | |
1170 | ceph_assert(buffer_bytes >= b->length); | |
1171 | buffer_bytes -= b->length; | |
1172 | auto q = lru.iterator_to(*b); | |
1173 | lru.erase(q); | |
1174 | num = lru.size(); | |
1175 | } | |
1176 | void _move(BlueStore::BufferCacheShard *src, BlueStore::Buffer *b) override { | |
1177 | src->_rm(b); | |
1178 | _add(b, 0, nullptr); | |
1179 | } | |
1180 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override { | |
1181 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1182 | buffer_bytes += delta; | |
1183 | } | |
1184 | void _touch(BlueStore::Buffer *b) override { | |
1185 | auto p = lru.iterator_to(*b); | |
1186 | lru.erase(p); | |
1187 | lru.push_front(*b); | |
1188 | num = lru.size(); | |
1189 | _audit("_touch_buffer end"); | |
1190 | } | |
7c673cae | 1191 | |
9f95a23c TL |
1192 | void _trim_to(uint64_t max) override |
1193 | { | |
1194 | while (buffer_bytes > max) { | |
1195 | auto i = lru.rbegin(); | |
1196 | if (i == lru.rend()) { | |
1197 | // stop if lru is now empty | |
7c673cae FG |
1198 | break; |
1199 | } | |
1200 | ||
9f95a23c TL |
1201 | BlueStore::Buffer *b = &*i; |
1202 | ceph_assert(b->is_clean()); | |
1203 | dout(20) << __func__ << " rm " << *b << dendl; | |
1204 | b->space->_rm_buffer(this, b); | |
7c673cae | 1205 | } |
9f95a23c | 1206 | num = lru.size(); |
7c673cae | 1207 | } |
7c673cae | 1208 | |
9f95a23c TL |
1209 | void add_stats(uint64_t *extents, |
1210 | uint64_t *blobs, | |
1211 | uint64_t *buffers, | |
1212 | uint64_t *bytes) override { | |
1213 | *extents += num_extents; | |
1214 | *blobs += num_blobs; | |
1215 | *buffers += num; | |
1216 | *bytes += buffer_bytes; | |
7c673cae | 1217 | } |
9f95a23c TL |
1218 | #ifdef DEBUG_CACHE |
1219 | void _audit(const char *s) override | |
1220 | { | |
1221 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1222 | uint64_t s = 0; | |
1223 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1224 | s += i->length; | |
1225 | } | |
1226 | if (s != buffer_bytes) { | |
1227 | derr << __func__ << " buffer_size " << buffer_bytes << " actual " << s | |
1228 | << dendl; | |
1229 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1230 | derr << __func__ << " " << *i << dendl; | |
1231 | } | |
1232 | ceph_assert(s == buffer_bytes); | |
7c673cae | 1233 | } |
9f95a23c TL |
1234 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1235 | << " ok" << dendl; | |
7c673cae | 1236 | } |
7c673cae | 1237 | #endif |
9f95a23c | 1238 | }; |
7c673cae | 1239 | |
9f95a23c TL |
1240 | // TwoQBufferCacheShard |
1241 | ||
1242 | struct TwoQBufferCacheShard : public BlueStore::BufferCacheShard { | |
1243 | typedef boost::intrusive::list< | |
1244 | BlueStore::Buffer, | |
1245 | boost::intrusive::member_hook< | |
1246 | BlueStore::Buffer, | |
1247 | boost::intrusive::list_member_hook<>, | |
1248 | &BlueStore::Buffer::lru_item> > list_t; | |
1249 | list_t hot; ///< "Am" hot buffers | |
1250 | list_t warm_in; ///< "A1in" newly warm buffers | |
1251 | list_t warm_out; ///< "A1out" empty buffers we've evicted | |
9f95a23c TL |
1252 | |
1253 | enum { | |
1254 | BUFFER_NEW = 0, | |
1255 | BUFFER_WARM_IN, ///< in warm_in | |
1256 | BUFFER_WARM_OUT, ///< in warm_out | |
1257 | BUFFER_HOT, ///< in hot | |
1258 | BUFFER_TYPE_MAX | |
1259 | }; | |
7c673cae | 1260 | |
9f95a23c | 1261 | uint64_t list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type |
7c673cae | 1262 | |
9f95a23c TL |
1263 | public: |
1264 | explicit TwoQBufferCacheShard(CephContext *cct) : BufferCacheShard(cct) {} | |
7c673cae | 1265 | |
9f95a23c TL |
1266 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override |
1267 | { | |
1268 | dout(20) << __func__ << " level " << level << " near " << near | |
1269 | << " on " << *b | |
1270 | << " which has cache_private " << b->cache_private << dendl; | |
1271 | if (near) { | |
1272 | b->cache_private = near->cache_private; | |
1273 | switch (b->cache_private) { | |
1274 | case BUFFER_WARM_IN: | |
1275 | warm_in.insert(warm_in.iterator_to(*near), *b); | |
1276 | break; | |
1277 | case BUFFER_WARM_OUT: | |
1278 | ceph_assert(b->is_empty()); | |
1279 | warm_out.insert(warm_out.iterator_to(*near), *b); | |
1280 | break; | |
1281 | case BUFFER_HOT: | |
1282 | hot.insert(hot.iterator_to(*near), *b); | |
1283 | break; | |
1284 | default: | |
1285 | ceph_abort_msg("bad cache_private"); | |
1286 | } | |
1287 | } else if (b->cache_private == BUFFER_NEW) { | |
1288 | b->cache_private = BUFFER_WARM_IN; | |
1289 | if (level > 0) { | |
1290 | warm_in.push_front(*b); | |
1291 | } else { | |
1292 | // take caller hint to start at the back of the warm queue | |
1293 | warm_in.push_back(*b); | |
1294 | } | |
1295 | } else { | |
1296 | // we got a hint from discard | |
1297 | switch (b->cache_private) { | |
1298 | case BUFFER_WARM_IN: | |
1299 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1300 | // do this. | |
1301 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1302 | warm_in.push_front(*b); | |
1303 | break; | |
1304 | case BUFFER_WARM_OUT: | |
1305 | b->cache_private = BUFFER_HOT; | |
1306 | // move to hot. fall-thru | |
1307 | case BUFFER_HOT: | |
1308 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1309 | hot.push_front(*b); | |
1310 | break; | |
1311 | default: | |
1312 | ceph_abort_msg("bad cache_private"); | |
1313 | } | |
1314 | } | |
1315 | if (!b->is_empty()) { | |
1316 | buffer_bytes += b->length; | |
1317 | list_bytes[b->cache_private] += b->length; | |
1318 | } | |
1319 | num = hot.size() + warm_in.size(); | |
1320 | } | |
1321 | ||
1322 | void _rm(BlueStore::Buffer *b) override | |
1323 | { | |
1324 | dout(20) << __func__ << " " << *b << dendl; | |
1325 | if (!b->is_empty()) { | |
1326 | ceph_assert(buffer_bytes >= b->length); | |
1327 | buffer_bytes -= b->length; | |
1328 | ceph_assert(list_bytes[b->cache_private] >= b->length); | |
1329 | list_bytes[b->cache_private] -= b->length; | |
1330 | } | |
7c673cae FG |
1331 | switch (b->cache_private) { |
1332 | case BUFFER_WARM_IN: | |
9f95a23c | 1333 | warm_in.erase(warm_in.iterator_to(*b)); |
7c673cae FG |
1334 | break; |
1335 | case BUFFER_WARM_OUT: | |
9f95a23c | 1336 | warm_out.erase(warm_out.iterator_to(*b)); |
7c673cae FG |
1337 | break; |
1338 | case BUFFER_HOT: | |
9f95a23c | 1339 | hot.erase(hot.iterator_to(*b)); |
7c673cae FG |
1340 | break; |
1341 | default: | |
11fdf7f2 | 1342 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1343 | } |
9f95a23c TL |
1344 | num = hot.size() + warm_in.size(); |
1345 | } | |
1346 | ||
1347 | void _move(BlueStore::BufferCacheShard *srcc, BlueStore::Buffer *b) override | |
1348 | { | |
1349 | TwoQBufferCacheShard *src = static_cast<TwoQBufferCacheShard*>(srcc); | |
1350 | src->_rm(b); | |
1351 | ||
1352 | // preserve which list we're on (even if we can't preserve the order!) | |
7c673cae FG |
1353 | switch (b->cache_private) { |
1354 | case BUFFER_WARM_IN: | |
9f95a23c TL |
1355 | ceph_assert(!b->is_empty()); |
1356 | warm_in.push_back(*b); | |
7c673cae FG |
1357 | break; |
1358 | case BUFFER_WARM_OUT: | |
9f95a23c TL |
1359 | ceph_assert(b->is_empty()); |
1360 | warm_out.push_back(*b); | |
1361 | break; | |
7c673cae | 1362 | case BUFFER_HOT: |
9f95a23c TL |
1363 | ceph_assert(!b->is_empty()); |
1364 | hot.push_back(*b); | |
7c673cae FG |
1365 | break; |
1366 | default: | |
11fdf7f2 | 1367 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1368 | } |
9f95a23c TL |
1369 | if (!b->is_empty()) { |
1370 | buffer_bytes += b->length; | |
1371 | list_bytes[b->cache_private] += b->length; | |
1372 | } | |
1373 | num = hot.size() + warm_in.size(); | |
7c673cae | 1374 | } |
7c673cae | 1375 | |
9f95a23c TL |
1376 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override |
1377 | { | |
1378 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1379 | if (!b->is_empty()) { | |
1380 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1381 | buffer_bytes += delta; | |
1382 | ceph_assert((int64_t)list_bytes[b->cache_private] + delta >= 0); | |
1383 | list_bytes[b->cache_private] += delta; | |
1384 | } | |
7c673cae | 1385 | } |
7c673cae | 1386 | |
9f95a23c TL |
1387 | void _touch(BlueStore::Buffer *b) override { |
1388 | switch (b->cache_private) { | |
1389 | case BUFFER_WARM_IN: | |
1390 | // do nothing (somewhat counter-intuitively!) | |
1391 | break; | |
1392 | case BUFFER_WARM_OUT: | |
1393 | // move from warm_out to hot LRU | |
1394 | ceph_abort_msg("this happens via discard hint"); | |
1395 | break; | |
1396 | case BUFFER_HOT: | |
1397 | // move to front of hot LRU | |
1398 | hot.erase(hot.iterator_to(*b)); | |
1399 | hot.push_front(*b); | |
1400 | break; | |
1401 | } | |
1402 | num = hot.size() + warm_in.size(); | |
1403 | _audit("_touch_buffer end"); | |
7c673cae | 1404 | } |
7c673cae | 1405 | |
9f95a23c TL |
1406 | void _trim_to(uint64_t max) override |
1407 | { | |
1408 | if (buffer_bytes > max) { | |
1409 | uint64_t kin = max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1410 | uint64_t khot = max - kin; | |
1411 | ||
1412 | // pre-calculate kout based on average buffer size too, | |
1413 | // which is typical(the warm_in and hot lists may change later) | |
1414 | uint64_t kout = 0; | |
1415 | uint64_t buffer_num = hot.size() + warm_in.size(); | |
1416 | if (buffer_num) { | |
1417 | uint64_t avg_size = buffer_bytes / buffer_num; | |
1418 | ceph_assert(avg_size); | |
1419 | uint64_t calculated_num = max / avg_size; | |
1420 | kout = calculated_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1421 | } | |
1422 | ||
1423 | if (list_bytes[BUFFER_HOT] < khot) { | |
1424 | // hot is small, give slack to warm_in | |
1425 | kin += khot - list_bytes[BUFFER_HOT]; | |
1426 | } else if (list_bytes[BUFFER_WARM_IN] < kin) { | |
1427 | // warm_in is small, give slack to hot | |
1428 | khot += kin - list_bytes[BUFFER_WARM_IN]; | |
1429 | } | |
1430 | ||
1431 | // adjust warm_in list | |
1432 | int64_t to_evict_bytes = list_bytes[BUFFER_WARM_IN] - kin; | |
1433 | uint64_t evicted = 0; | |
1434 | ||
1435 | while (to_evict_bytes > 0) { | |
1436 | auto p = warm_in.rbegin(); | |
1437 | if (p == warm_in.rend()) { | |
1438 | // stop if warm_in list is now empty | |
1439 | break; | |
1440 | } | |
7c673cae | 1441 | |
9f95a23c TL |
1442 | BlueStore::Buffer *b = &*p; |
1443 | ceph_assert(b->is_clean()); | |
1444 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; | |
1445 | ceph_assert(buffer_bytes >= b->length); | |
1446 | buffer_bytes -= b->length; | |
1447 | ceph_assert(list_bytes[BUFFER_WARM_IN] >= b->length); | |
1448 | list_bytes[BUFFER_WARM_IN] -= b->length; | |
1449 | to_evict_bytes -= b->length; | |
1450 | evicted += b->length; | |
1451 | b->state = BlueStore::Buffer::STATE_EMPTY; | |
1452 | b->data.clear(); | |
1453 | warm_in.erase(warm_in.iterator_to(*b)); | |
1454 | warm_out.push_front(*b); | |
1455 | b->cache_private = BUFFER_WARM_OUT; | |
1456 | } | |
1457 | ||
1458 | if (evicted > 0) { | |
1459 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1460 | << " from warm_in list, done evicting warm_in buffers" | |
1461 | << dendl; | |
1462 | } | |
7c673cae | 1463 | |
9f95a23c TL |
1464 | // adjust hot list |
1465 | to_evict_bytes = list_bytes[BUFFER_HOT] - khot; | |
1466 | evicted = 0; | |
7c673cae | 1467 | |
9f95a23c TL |
1468 | while (to_evict_bytes > 0) { |
1469 | auto p = hot.rbegin(); | |
1470 | if (p == hot.rend()) { | |
1471 | // stop if hot list is now empty | |
1472 | break; | |
1473 | } | |
7c673cae | 1474 | |
9f95a23c TL |
1475 | BlueStore::Buffer *b = &*p; |
1476 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
1477 | ceph_assert(b->is_clean()); | |
1478 | // adjust evict size before buffer goes invalid | |
1479 | to_evict_bytes -= b->length; | |
1480 | evicted += b->length; | |
1481 | b->space->_rm_buffer(this, b); | |
1482 | } | |
7c673cae | 1483 | |
9f95a23c TL |
1484 | if (evicted > 0) { |
1485 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1486 | << " from hot list, done evicting hot buffers" | |
1487 | << dendl; | |
7c673cae FG |
1488 | } |
1489 | ||
9f95a23c TL |
1490 | // adjust warm out list too, if necessary |
1491 | int64_t n = warm_out.size() - kout; | |
1492 | while (n-- > 0) { | |
1493 | BlueStore::Buffer *b = &*warm_out.rbegin(); | |
1494 | ceph_assert(b->is_empty()); | |
1495 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; | |
1496 | b->space->_rm_buffer(this, b); | |
1497 | } | |
7c673cae | 1498 | } |
9f95a23c TL |
1499 | num = hot.size() + warm_in.size(); |
1500 | } | |
7c673cae | 1501 | |
9f95a23c TL |
1502 | void add_stats(uint64_t *extents, |
1503 | uint64_t *blobs, | |
1504 | uint64_t *buffers, | |
1505 | uint64_t *bytes) override { | |
1506 | *extents += num_extents; | |
1507 | *blobs += num_blobs; | |
1508 | *buffers += num; | |
1509 | *bytes += buffer_bytes; | |
1510 | } | |
7c673cae | 1511 | |
9f95a23c TL |
1512 | #ifdef DEBUG_CACHE |
1513 | void _audit(const char *s) override | |
1514 | { | |
1515 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1516 | uint64_t s = 0; | |
1517 | for (auto i = hot.begin(); i != hot.end(); ++i) { | |
1518 | s += i->length; | |
7c673cae FG |
1519 | } |
1520 | ||
9f95a23c TL |
1521 | uint64_t hot_bytes = s; |
1522 | if (hot_bytes != list_bytes[BUFFER_HOT]) { | |
1523 | derr << __func__ << " hot_list_bytes " | |
1524 | << list_bytes[BUFFER_HOT] | |
1525 | << " != actual " << hot_bytes | |
1526 | << dendl; | |
1527 | ceph_assert(hot_bytes == list_bytes[BUFFER_HOT]); | |
7c673cae FG |
1528 | } |
1529 | ||
9f95a23c TL |
1530 | for (auto i = warm_in.begin(); i != warm_in.end(); ++i) { |
1531 | s += i->length; | |
7c673cae | 1532 | } |
7c673cae | 1533 | |
9f95a23c TL |
1534 | uint64_t warm_in_bytes = s - hot_bytes; |
1535 | if (warm_in_bytes != list_bytes[BUFFER_WARM_IN]) { | |
1536 | derr << __func__ << " warm_in_list_bytes " | |
1537 | << list_bytes[BUFFER_WARM_IN] | |
1538 | << " != actual " << warm_in_bytes | |
1539 | << dendl; | |
1540 | ceph_assert(warm_in_bytes == list_bytes[BUFFER_WARM_IN]); | |
7c673cae | 1541 | } |
7c673cae | 1542 | |
9f95a23c TL |
1543 | if (s != buffer_bytes) { |
1544 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1545 | << dendl; | |
1546 | ceph_assert(s == buffer_bytes); | |
1547 | } | |
7c673cae | 1548 | |
9f95a23c TL |
1549 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1550 | << " ok" << dendl; | |
7c673cae | 1551 | } |
9f95a23c TL |
1552 | #endif |
1553 | }; | |
7c673cae | 1554 | |
9f95a23c | 1555 | // BuferCacheShard |
7c673cae | 1556 | |
9f95a23c TL |
1557 | BlueStore::BufferCacheShard *BlueStore::BufferCacheShard::create( |
1558 | CephContext* cct, | |
1559 | string type, | |
1560 | PerfCounters *logger) | |
1561 | { | |
1562 | BufferCacheShard *c = nullptr; | |
1563 | if (type == "lru") | |
1564 | c = new LruBufferCacheShard(cct); | |
1565 | else if (type == "2q") | |
1566 | c = new TwoQBufferCacheShard(cct); | |
1567 | else | |
1568 | ceph_abort_msg("unrecognized cache type"); | |
1569 | c->logger = logger; | |
1570 | return c; | |
7c673cae | 1571 | } |
7c673cae FG |
1572 | |
1573 | // BufferSpace | |
1574 | ||
1575 | #undef dout_prefix | |
1576 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1577 | ||
9f95a23c | 1578 | void BlueStore::BufferSpace::_clear(BufferCacheShard* cache) |
7c673cae FG |
1579 | { |
1580 | // note: we already hold cache->lock | |
1581 | ldout(cache->cct, 20) << __func__ << dendl; | |
1582 | while (!buffer_map.empty()) { | |
1583 | _rm_buffer(cache, buffer_map.begin()); | |
1584 | } | |
1585 | } | |
1586 | ||
9f95a23c | 1587 | int BlueStore::BufferSpace::_discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) |
7c673cae FG |
1588 | { |
1589 | // note: we already hold cache->lock | |
1590 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1591 | << std::dec << dendl; | |
1592 | int cache_private = 0; | |
1593 | cache->_audit("discard start"); | |
1594 | auto i = _data_lower_bound(offset); | |
1595 | uint32_t end = offset + length; | |
1596 | while (i != buffer_map.end()) { | |
1597 | Buffer *b = i->second.get(); | |
1598 | if (b->offset >= end) { | |
1599 | break; | |
1600 | } | |
1601 | if (b->cache_private > cache_private) { | |
1602 | cache_private = b->cache_private; | |
1603 | } | |
1604 | if (b->offset < offset) { | |
1605 | int64_t front = offset - b->offset; | |
1606 | if (b->end() > end) { | |
1607 | // drop middle (split) | |
1608 | uint32_t tail = b->end() - end; | |
1609 | if (b->data.length()) { | |
1610 | bufferlist bl; | |
1611 | bl.substr_of(b->data, b->length - tail, tail); | |
f67539c2 | 1612 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl, b->flags); |
31f18b77 FG |
1613 | nb->maybe_rebuild(); |
1614 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1615 | } else { |
f67539c2 TL |
1616 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail, |
1617 | b->flags), | |
1618 | 0, b); | |
7c673cae FG |
1619 | } |
1620 | if (!b->is_writing()) { | |
9f95a23c | 1621 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1622 | } |
1623 | b->truncate(front); | |
31f18b77 | 1624 | b->maybe_rebuild(); |
7c673cae FG |
1625 | cache->_audit("discard end 1"); |
1626 | break; | |
1627 | } else { | |
1628 | // drop tail | |
1629 | if (!b->is_writing()) { | |
9f95a23c | 1630 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1631 | } |
1632 | b->truncate(front); | |
31f18b77 | 1633 | b->maybe_rebuild(); |
7c673cae FG |
1634 | ++i; |
1635 | continue; | |
1636 | } | |
1637 | } | |
1638 | if (b->end() <= end) { | |
1639 | // drop entire buffer | |
1640 | _rm_buffer(cache, i++); | |
1641 | continue; | |
1642 | } | |
1643 | // drop front | |
1644 | uint32_t keep = b->end() - end; | |
1645 | if (b->data.length()) { | |
1646 | bufferlist bl; | |
1647 | bl.substr_of(b->data, b->length - keep, keep); | |
f67539c2 | 1648 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl, b->flags); |
31f18b77 FG |
1649 | nb->maybe_rebuild(); |
1650 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1651 | } else { |
f67539c2 TL |
1652 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep, |
1653 | b->flags), | |
1654 | 0, b); | |
7c673cae FG |
1655 | } |
1656 | _rm_buffer(cache, i); | |
1657 | cache->_audit("discard end 2"); | |
1658 | break; | |
1659 | } | |
1660 | return cache_private; | |
1661 | } | |
1662 | ||
1663 | void BlueStore::BufferSpace::read( | |
9f95a23c | 1664 | BufferCacheShard* cache, |
224ce89b WB |
1665 | uint32_t offset, |
1666 | uint32_t length, | |
7c673cae | 1667 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1668 | interval_set<uint32_t>& res_intervals, |
1669 | int flags) | |
7c673cae | 1670 | { |
7c673cae FG |
1671 | res.clear(); |
1672 | res_intervals.clear(); | |
1673 | uint32_t want_bytes = length; | |
1674 | uint32_t end = offset + length; | |
224ce89b WB |
1675 | |
1676 | { | |
11fdf7f2 | 1677 | std::lock_guard l(cache->lock); |
224ce89b WB |
1678 | for (auto i = _data_lower_bound(offset); |
1679 | i != buffer_map.end() && offset < end && i->first < end; | |
1680 | ++i) { | |
1681 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1682 | ceph_assert(b->end() > offset); |
91327a77 AA |
1683 | |
1684 | bool val = false; | |
1685 | if (flags & BYPASS_CLEAN_CACHE) | |
1686 | val = b->is_writing(); | |
1687 | else | |
1688 | val = b->is_writing() || b->is_clean(); | |
1689 | if (val) { | |
224ce89b WB |
1690 | if (b->offset < offset) { |
1691 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1692 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1693 | res[offset].substr_of(b->data, skip, l); |
1694 | res_intervals.insert(offset, l); | |
1695 | offset += l; | |
1696 | length -= l; | |
1697 | if (!b->is_writing()) { | |
9f95a23c | 1698 | cache->_touch(b); |
f67539c2 | 1699 | } |
224ce89b WB |
1700 | continue; |
1701 | } | |
1702 | if (b->offset > offset) { | |
1703 | uint32_t gap = b->offset - offset; | |
1704 | if (length <= gap) { | |
1705 | break; | |
1706 | } | |
1707 | offset += gap; | |
1708 | length -= gap; | |
1709 | } | |
1710 | if (!b->is_writing()) { | |
9f95a23c | 1711 | cache->_touch(b); |
224ce89b WB |
1712 | } |
1713 | if (b->length > length) { | |
1714 | res[offset].substr_of(b->data, 0, length); | |
1715 | res_intervals.insert(offset, length); | |
7c673cae | 1716 | break; |
224ce89b WB |
1717 | } else { |
1718 | res[offset].append(b->data); | |
1719 | res_intervals.insert(offset, b->length); | |
1720 | if (b->length == length) | |
1721 | break; | |
1722 | offset += b->length; | |
1723 | length -= b->length; | |
1724 | } | |
7c673cae FG |
1725 | } |
1726 | } | |
1727 | } | |
1728 | ||
1729 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1730 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1731 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1732 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1733 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1734 | } | |
1735 | ||
9f95a23c | 1736 | void BlueStore::BufferSpace::_finish_write(BufferCacheShard* cache, uint64_t seq) |
7c673cae | 1737 | { |
7c673cae FG |
1738 | auto i = writing.begin(); |
1739 | while (i != writing.end()) { | |
1740 | if (i->seq > seq) { | |
1741 | break; | |
1742 | } | |
1743 | if (i->seq < seq) { | |
1744 | ++i; | |
1745 | continue; | |
1746 | } | |
1747 | ||
1748 | Buffer *b = &*i; | |
11fdf7f2 | 1749 | ceph_assert(b->is_writing()); |
7c673cae FG |
1750 | |
1751 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1752 | writing.erase(i++); | |
1753 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1754 | buffer_map.erase(b->offset); | |
1755 | } else { | |
1756 | b->state = Buffer::STATE_CLEAN; | |
1757 | writing.erase(i++); | |
31f18b77 FG |
1758 | b->maybe_rebuild(); |
1759 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
9f95a23c | 1760 | cache->_add(b, 1, nullptr); |
7c673cae FG |
1761 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; |
1762 | } | |
1763 | } | |
9f95a23c | 1764 | cache->_trim(); |
7c673cae FG |
1765 | cache->_audit("finish_write end"); |
1766 | } | |
1767 | ||
9f95a23c | 1768 | void BlueStore::BufferSpace::split(BufferCacheShard* cache, size_t pos, BlueStore::BufferSpace &r) |
7c673cae | 1769 | { |
11fdf7f2 | 1770 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1771 | if (buffer_map.empty()) |
1772 | return; | |
1773 | ||
1774 | auto p = --buffer_map.end(); | |
1775 | while (true) { | |
1776 | if (p->second->end() <= pos) | |
1777 | break; | |
1778 | ||
1779 | if (p->second->offset < pos) { | |
1780 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1781 | size_t left = pos - p->second->offset; | |
1782 | size_t right = p->second->length - left; | |
1783 | if (p->second->data.length()) { | |
1784 | bufferlist bl; | |
1785 | bl.substr_of(p->second->data, left, right); | |
f67539c2 TL |
1786 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, |
1787 | 0, bl, p->second->flags), | |
7c673cae FG |
1788 | 0, p->second.get()); |
1789 | } else { | |
f67539c2 TL |
1790 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, |
1791 | 0, right, p->second->flags), | |
7c673cae FG |
1792 | 0, p->second.get()); |
1793 | } | |
9f95a23c | 1794 | cache->_adjust_size(p->second.get(), -right); |
7c673cae FG |
1795 | p->second->truncate(left); |
1796 | break; | |
1797 | } | |
1798 | ||
11fdf7f2 | 1799 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1800 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1801 | if (p->second->data.length()) { | |
1802 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
f67539c2 | 1803 | p->second->offset - pos, p->second->data, p->second->flags), |
7c673cae FG |
1804 | 0, p->second.get()); |
1805 | } else { | |
1806 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
f67539c2 | 1807 | p->second->offset - pos, p->second->length, p->second->flags), |
7c673cae FG |
1808 | 0, p->second.get()); |
1809 | } | |
1810 | if (p == buffer_map.begin()) { | |
1811 | _rm_buffer(cache, p); | |
1812 | break; | |
1813 | } else { | |
1814 | _rm_buffer(cache, p--); | |
1815 | } | |
1816 | } | |
11fdf7f2 | 1817 | ceph_assert(writing.empty()); |
9f95a23c | 1818 | cache->_trim(); |
7c673cae FG |
1819 | } |
1820 | ||
1821 | // OnodeSpace | |
1822 | ||
1823 | #undef dout_prefix | |
1824 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1825 | ||
f6b5b4d7 TL |
1826 | BlueStore::OnodeRef BlueStore::OnodeSpace::add(const ghobject_t& oid, |
1827 | OnodeRef& o) | |
7c673cae | 1828 | { |
11fdf7f2 | 1829 | std::lock_guard l(cache->lock); |
7c673cae FG |
1830 | auto p = onode_map.find(oid); |
1831 | if (p != onode_map.end()) { | |
1832 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o | |
1833 | << " raced, returning existing " << p->second | |
1834 | << dendl; | |
1835 | return p->second; | |
1836 | } | |
f6b5b4d7 | 1837 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << o << dendl; |
7c673cae | 1838 | onode_map[oid] = o; |
f6b5b4d7 | 1839 | cache->_add(o.get(), 1); |
9f95a23c | 1840 | cache->_trim(); |
7c673cae FG |
1841 | return o; |
1842 | } | |
1843 | ||
f6b5b4d7 TL |
1844 | void BlueStore::OnodeSpace::_remove(const ghobject_t& oid) |
1845 | { | |
1846 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << dendl; | |
1847 | onode_map.erase(oid); | |
1848 | } | |
1849 | ||
7c673cae FG |
1850 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) |
1851 | { | |
7c673cae | 1852 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b WB |
1853 | OnodeRef o; |
1854 | bool hit = false; | |
1855 | ||
1856 | { | |
11fdf7f2 | 1857 | std::lock_guard l(cache->lock); |
224ce89b WB |
1858 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1859 | if (p == onode_map.end()) { | |
1860 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
1861 | } else { | |
1862 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
f6b5b4d7 TL |
1863 | << " " << p->second->nref |
1864 | << " " << p->second->cached | |
1865 | << " " << p->second->pinned | |
224ce89b | 1866 | << dendl; |
f6b5b4d7 TL |
1867 | // This will pin onode and implicitly touch the cache when Onode |
1868 | // eventually will become unpinned | |
224ce89b | 1869 | o = p->second; |
f6b5b4d7 TL |
1870 | ceph_assert(!o->cached || o->pinned); |
1871 | ||
1872 | hit = true; | |
224ce89b WB |
1873 | } |
1874 | } | |
1875 | ||
1876 | if (hit) { | |
1877 | cache->logger->inc(l_bluestore_onode_hits); | |
1878 | } else { | |
7c673cae | 1879 | cache->logger->inc(l_bluestore_onode_misses); |
7c673cae | 1880 | } |
224ce89b | 1881 | return o; |
7c673cae FG |
1882 | } |
1883 | ||
1884 | void BlueStore::OnodeSpace::clear() | |
1885 | { | |
11fdf7f2 | 1886 | std::lock_guard l(cache->lock); |
f6b5b4d7 | 1887 | ldout(cache->cct, 10) << __func__ << " " << onode_map.size()<< dendl; |
7c673cae | 1888 | for (auto &p : onode_map) { |
f6b5b4d7 | 1889 | cache->_rm(p.second.get()); |
7c673cae FG |
1890 | } |
1891 | onode_map.clear(); | |
1892 | } | |
1893 | ||
1894 | bool BlueStore::OnodeSpace::empty() | |
1895 | { | |
11fdf7f2 | 1896 | std::lock_guard l(cache->lock); |
7c673cae FG |
1897 | return onode_map.empty(); |
1898 | } | |
1899 | ||
1900 | void BlueStore::OnodeSpace::rename( | |
1901 | OnodeRef& oldo, | |
1902 | const ghobject_t& old_oid, | |
1903 | const ghobject_t& new_oid, | |
f91f0fd5 | 1904 | const mempool::bluestore_cache_meta::string& new_okey) |
7c673cae | 1905 | { |
11fdf7f2 | 1906 | std::lock_guard l(cache->lock); |
7c673cae FG |
1907 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
1908 | << dendl; | |
1909 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
1910 | po = onode_map.find(old_oid); | |
1911 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 1912 | ceph_assert(po != pn); |
7c673cae | 1913 | |
11fdf7f2 | 1914 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
1915 | if (pn != onode_map.end()) { |
1916 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
1917 | << dendl; | |
f6b5b4d7 | 1918 | cache->_rm(pn->second.get()); |
7c673cae FG |
1919 | onode_map.erase(pn); |
1920 | } | |
1921 | OnodeRef o = po->second; | |
1922 | ||
1923 | // install a non-existent onode at old location | |
1924 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
1925 | po->second = oldo; | |
f6b5b4d7 TL |
1926 | cache->_add(oldo.get(), 1); |
1927 | // add at new position and fix oid, key. | |
1928 | // This will pin 'o' and implicitly touch cache | |
1929 | // when it will eventually become unpinned | |
7c673cae | 1930 | onode_map.insert(make_pair(new_oid, o)); |
f6b5b4d7 TL |
1931 | ceph_assert(o->pinned); |
1932 | ||
7c673cae FG |
1933 | o->oid = new_oid; |
1934 | o->key = new_okey; | |
9f95a23c | 1935 | cache->_trim(); |
7c673cae FG |
1936 | } |
1937 | ||
adb31ebb | 1938 | bool BlueStore::OnodeSpace::map_any(std::function<bool(Onode*)> f) |
7c673cae | 1939 | { |
11fdf7f2 | 1940 | std::lock_guard l(cache->lock); |
7c673cae FG |
1941 | ldout(cache->cct, 20) << __func__ << dendl; |
1942 | for (auto& i : onode_map) { | |
adb31ebb | 1943 | if (f(i.second.get())) { |
7c673cae FG |
1944 | return true; |
1945 | } | |
1946 | } | |
1947 | return false; | |
1948 | } | |
1949 | ||
11fdf7f2 TL |
1950 | template <int LogLevelV = 30> |
1951 | void BlueStore::OnodeSpace::dump(CephContext *cct) | |
3efd9988 FG |
1952 | { |
1953 | for (auto& i : onode_map) { | |
f6b5b4d7 TL |
1954 | ldout(cct, LogLevelV) << i.first << " : " << i.second |
1955 | << " " << i.second->nref | |
1956 | << " " << i.second->cached | |
1957 | << " " << i.second->pinned | |
1958 | << dendl; | |
3efd9988 FG |
1959 | } |
1960 | } | |
7c673cae FG |
1961 | |
1962 | // SharedBlob | |
1963 | ||
1964 | #undef dout_prefix | |
1965 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
9f95a23c TL |
1966 | #undef dout_context |
1967 | #define dout_context coll->store->cct | |
7c673cae | 1968 | |
9f95a23c | 1969 | void BlueStore::SharedBlob::dump(Formatter* f) const |
7c673cae | 1970 | { |
9f95a23c TL |
1971 | f->dump_bool("loaded", loaded); |
1972 | if (loaded) { | |
1973 | persistent->dump(f); | |
1974 | } else { | |
1975 | f->dump_unsigned("sbid_unloaded", sbid_unloaded); | |
1976 | } | |
1977 | } | |
1978 | ||
1979 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
1980 | { | |
1981 | out << "SharedBlob(" << &sb; | |
1982 | ||
7c673cae FG |
1983 | if (sb.loaded) { |
1984 | out << " loaded " << *sb.persistent; | |
1985 | } else { | |
1986 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
1987 | } | |
1988 | return out << ")"; | |
1989 | } | |
1990 | ||
1991 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
1992 | : coll(_coll), sbid_unloaded(i) | |
1993 | { | |
11fdf7f2 | 1994 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
1995 | if (get_cache()) { |
1996 | get_cache()->add_blob(); | |
1997 | } | |
1998 | } | |
1999 | ||
2000 | BlueStore::SharedBlob::~SharedBlob() | |
2001 | { | |
7c673cae FG |
2002 | if (loaded && persistent) { |
2003 | delete persistent; | |
2004 | } | |
2005 | } | |
2006 | ||
2007 | void BlueStore::SharedBlob::put() | |
2008 | { | |
2009 | if (--nref == 0) { | |
9f95a23c TL |
2010 | dout(20) << __func__ << " " << this |
2011 | << " removing self from set " << get_parent() | |
2012 | << dendl; | |
1adf2230 AA |
2013 | again: |
2014 | auto coll_snap = coll; | |
2015 | if (coll_snap) { | |
11fdf7f2 | 2016 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
2017 | if (coll_snap != coll) { |
2018 | goto again; | |
2019 | } | |
91327a77 AA |
2020 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
2021 | // race with lookup | |
2022 | return; | |
2023 | } | |
1adf2230 AA |
2024 | bc._clear(coll_snap->cache); |
2025 | coll_snap->cache->rm_blob(); | |
7c673cae | 2026 | } |
28e407b8 | 2027 | delete this; |
7c673cae FG |
2028 | } |
2029 | } | |
2030 | ||
2031 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
2032 | { | |
11fdf7f2 | 2033 | ceph_assert(persistent); |
7c673cae FG |
2034 | persistent->ref_map.get(offset, length); |
2035 | } | |
2036 | ||
2037 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 2038 | PExtentVector *r, |
11fdf7f2 | 2039 | bool *unshare) |
7c673cae | 2040 | { |
11fdf7f2 TL |
2041 | ceph_assert(persistent); |
2042 | persistent->ref_map.put(offset, length, r, | |
2043 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
2044 | } |
2045 | ||
f64942e4 AA |
2046 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
2047 | { | |
2048 | while (true) { | |
9f95a23c | 2049 | BufferCacheShard *cache = coll->cache; |
11fdf7f2 | 2050 | std::lock_guard l(cache->lock); |
f64942e4 | 2051 | if (coll->cache != cache) { |
9f95a23c TL |
2052 | dout(20) << __func__ |
2053 | << " raced with sb cache update, was " << cache | |
2054 | << ", now " << coll->cache << ", retrying" | |
2055 | << dendl; | |
f64942e4 AA |
2056 | continue; |
2057 | } | |
2058 | bc._finish_write(cache, seq); | |
2059 | break; | |
2060 | } | |
2061 | } | |
2062 | ||
3efd9988 FG |
2063 | // SharedBlobSet |
2064 | ||
2065 | #undef dout_prefix | |
2066 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
2067 | ||
11fdf7f2 TL |
2068 | template <int LogLevelV = 30> |
2069 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 2070 | { |
11fdf7f2 | 2071 | std::lock_guard l(lock); |
3efd9988 | 2072 | for (auto& i : sb_map) { |
11fdf7f2 | 2073 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
2074 | } |
2075 | } | |
2076 | ||
7c673cae FG |
2077 | // Blob |
2078 | ||
2079 | #undef dout_prefix | |
2080 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
2081 | ||
9f95a23c TL |
2082 | void BlueStore::Blob::dump(Formatter* f) const |
2083 | { | |
2084 | if (is_spanning()) { | |
2085 | f->dump_unsigned("spanning_id ", id); | |
2086 | } | |
2087 | blob.dump(f); | |
2088 | if (shared_blob) { | |
2089 | f->dump_object("shared", *shared_blob); | |
2090 | } | |
2091 | } | |
2092 | ||
7c673cae FG |
2093 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) |
2094 | { | |
2095 | out << "Blob(" << &b; | |
2096 | if (b.is_spanning()) { | |
2097 | out << " spanning " << b.id; | |
2098 | } | |
35e4c445 FG |
2099 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
2100 | if (b.shared_blob) { | |
2101 | out << " " << *b.shared_blob; | |
2102 | } else { | |
2103 | out << " (shared_blob=NULL)"; | |
2104 | } | |
2105 | out << ")"; | |
7c673cae FG |
2106 | return out; |
2107 | } | |
2108 | ||
2109 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
2110 | { | |
224ce89b | 2111 | if (get_blob().is_shared()) { |
7c673cae FG |
2112 | return; |
2113 | } | |
224ce89b | 2114 | if (get_blob().is_compressed()) { |
7c673cae FG |
2115 | bool discard = false; |
2116 | bool all_invalid = true; | |
224ce89b | 2117 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
2118 | if (!e.is_valid()) { |
2119 | discard = true; | |
2120 | } else { | |
2121 | all_invalid = false; | |
2122 | } | |
2123 | } | |
11fdf7f2 | 2124 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
2125 | // or none pextents are invalid. |
2126 | if (discard) { | |
224ce89b WB |
2127 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
2128 | get_blob().get_logical_length()); | |
7c673cae FG |
2129 | } |
2130 | } else { | |
2131 | size_t pos = 0; | |
224ce89b | 2132 | for (auto e : get_blob().get_extents()) { |
7c673cae | 2133 | if (!e.is_valid()) { |
9f95a23c TL |
2134 | dout(20) << __func__ << " 0x" << std::hex << pos |
2135 | << "~" << e.length | |
2136 | << std::dec << dendl; | |
7c673cae FG |
2137 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); |
2138 | } | |
2139 | pos += e.length; | |
2140 | } | |
224ce89b WB |
2141 | if (get_blob().can_prune_tail()) { |
2142 | dirty_blob().prune_tail(); | |
2143 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
224ce89b | 2144 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
2145 | } |
2146 | } | |
2147 | } | |
2148 | ||
2149 | void BlueStore::Blob::get_ref( | |
2150 | Collection *coll, | |
2151 | uint32_t offset, | |
2152 | uint32_t length) | |
2153 | { | |
2154 | // Caller has to initialize Blob's logical length prior to increment | |
2155 | // references. Otherwise one is neither unable to determine required | |
2156 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
2157 | // for single counter mode. | |
11fdf7f2 | 2158 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
2159 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2160 | << std::dec << " " << *this << dendl; | |
2161 | ||
2162 | if (used_in_blob.is_empty()) { | |
2163 | uint32_t min_release_size = | |
224ce89b WB |
2164 | get_blob().get_release_size(coll->store->min_alloc_size); |
2165 | uint64_t l = get_blob().get_logical_length(); | |
2166 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
2167 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
2168 | used_in_blob.init(l, min_release_size); |
2169 | } | |
2170 | used_in_blob.get( | |
2171 | offset, | |
2172 | length); | |
2173 | } | |
2174 | ||
2175 | bool BlueStore::Blob::put_ref( | |
2176 | Collection *coll, | |
2177 | uint32_t offset, | |
2178 | uint32_t length, | |
2179 | PExtentVector *r) | |
2180 | { | |
2181 | PExtentVector logical; | |
2182 | ||
7c673cae FG |
2183 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2184 | << std::dec << " " << *this << dendl; | |
2185 | ||
2186 | bool empty = used_in_blob.put( | |
2187 | offset, | |
2188 | length, | |
2189 | &logical); | |
2190 | r->clear(); | |
2191 | // nothing to release | |
2192 | if (!empty && logical.empty()) { | |
2193 | return false; | |
2194 | } | |
2195 | ||
2196 | bluestore_blob_t& b = dirty_blob(); | |
2197 | return b.release_extents(empty, logical, r); | |
2198 | } | |
2199 | ||
224ce89b | 2200 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
2201 | uint32_t target_blob_size, |
2202 | uint32_t b_offset, | |
2203 | uint32_t *length0) { | |
11fdf7f2 TL |
2204 | ceph_assert(min_alloc_size); |
2205 | ceph_assert(target_blob_size); | |
7c673cae FG |
2206 | if (!get_blob().is_mutable()) { |
2207 | return false; | |
2208 | } | |
2209 | ||
2210 | uint32_t length = *length0; | |
2211 | uint32_t end = b_offset + length; | |
2212 | ||
2213 | // Currently for the sake of simplicity we omit blob reuse if data is | |
2214 | // unaligned with csum chunk. Later we can perform padding if needed. | |
2215 | if (get_blob().has_csum() && | |
2216 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
2217 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
2218 | return false; | |
2219 | } | |
2220 | ||
2221 | auto blen = get_blob().get_logical_length(); | |
2222 | uint32_t new_blen = blen; | |
2223 | ||
2224 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 2225 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
2226 | |
2227 | if (b_offset >= blen) { | |
224ce89b WB |
2228 | // new data totally stands out of the existing blob |
2229 | new_blen = end; | |
7c673cae | 2230 | } else { |
224ce89b | 2231 | // new data overlaps with the existing blob |
11fdf7f2 | 2232 | new_blen = std::max(blen, end); |
224ce89b WB |
2233 | |
2234 | uint32_t overlap = 0; | |
2235 | if (new_blen > blen) { | |
2236 | overlap = blen - b_offset; | |
2237 | } else { | |
2238 | overlap = length; | |
2239 | } | |
2240 | ||
2241 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
2242 | // abort if any piece of the overlap has already been allocated | |
2243 | return false; | |
7c673cae FG |
2244 | } |
2245 | } | |
224ce89b | 2246 | |
7c673cae FG |
2247 | if (new_blen > blen) { |
2248 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
2249 | // Unable to decrease the provided length to fit into max_blob_size | |
2250 | if (overflow >= length) { | |
2251 | return false; | |
2252 | } | |
2253 | ||
2254 | // FIXME: in some cases we could reduce unused resolution | |
2255 | if (get_blob().has_unused()) { | |
2256 | return false; | |
2257 | } | |
2258 | ||
2259 | if (overflow > 0) { | |
2260 | new_blen -= overflow; | |
2261 | length -= overflow; | |
2262 | *length0 = length; | |
2263 | } | |
224ce89b | 2264 | |
7c673cae FG |
2265 | if (new_blen > blen) { |
2266 | dirty_blob().add_tail(new_blen); | |
2267 | used_in_blob.add_tail(new_blen, | |
224ce89b | 2268 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
2269 | } |
2270 | } | |
2271 | return true; | |
2272 | } | |
2273 | ||
2274 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
2275 | { | |
7c673cae FG |
2276 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec |
2277 | << " start " << *this << dendl; | |
11fdf7f2 TL |
2278 | ceph_assert(blob.can_split()); |
2279 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
2280 | bluestore_blob_t &lb = dirty_blob(); |
2281 | bluestore_blob_t &rb = r->dirty_blob(); | |
2282 | ||
2283 | used_in_blob.split( | |
2284 | blob_offset, | |
2285 | &(r->used_in_blob)); | |
2286 | ||
2287 | lb.split(blob_offset, rb); | |
2288 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2289 | ||
2290 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2291 | << " finish " << *this << dendl; | |
2292 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2293 | << " and " << *r << dendl; | |
2294 | } | |
2295 | ||
2296 | #ifndef CACHE_BLOB_BL | |
2297 | void BlueStore::Blob::decode( | |
2298 | Collection *coll, | |
11fdf7f2 | 2299 | bufferptr::const_iterator& p, |
7c673cae FG |
2300 | uint64_t struct_v, |
2301 | uint64_t* sbid, | |
2302 | bool include_ref_map) | |
2303 | { | |
2304 | denc(blob, p, struct_v); | |
2305 | if (blob.is_shared()) { | |
2306 | denc(*sbid, p); | |
2307 | } | |
2308 | if (include_ref_map) { | |
2309 | if (struct_v > 1) { | |
2310 | used_in_blob.decode(p); | |
2311 | } else { | |
2312 | used_in_blob.clear(); | |
2313 | bluestore_extent_ref_map_t legacy_ref_map; | |
2314 | legacy_ref_map.decode(p); | |
2315 | for (auto r : legacy_ref_map.ref_map) { | |
2316 | get_ref( | |
2317 | coll, | |
2318 | r.first, | |
2319 | r.second.refs * r.second.length); | |
2320 | } | |
2321 | } | |
2322 | } | |
2323 | } | |
2324 | #endif | |
2325 | ||
2326 | // Extent | |
2327 | ||
9f95a23c TL |
2328 | void BlueStore::Extent::dump(Formatter* f) const |
2329 | { | |
2330 | f->dump_unsigned("logical_offset", logical_offset); | |
2331 | f->dump_unsigned("length", length); | |
2332 | f->dump_unsigned("blob_offset", blob_offset); | |
2333 | f->dump_object("blob", *blob); | |
2334 | } | |
2335 | ||
7c673cae FG |
2336 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) |
2337 | { | |
2338 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2339 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2340 | << " " << *e.blob; | |
2341 | } | |
2342 | ||
2343 | // OldExtent | |
2344 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2345 | uint32_t lo, | |
2346 | uint32_t o, | |
2347 | uint32_t l, | |
2348 | BlobRef& b) { | |
2349 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2350 | b->put_ref(c.get(), o, l, &(oe->r)); | |
adb31ebb | 2351 | oe->blob_empty = !b->is_referenced(); |
7c673cae FG |
2352 | return oe; |
2353 | } | |
2354 | ||
2355 | // ExtentMap | |
2356 | ||
2357 | #undef dout_prefix | |
2358 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
9f95a23c TL |
2359 | #undef dout_context |
2360 | #define dout_context onode->c->store->cct | |
7c673cae FG |
2361 | |
2362 | BlueStore::ExtentMap::ExtentMap(Onode *o) | |
2363 | : onode(o), | |
2364 | inline_bl( | |
2365 | o->c->store->cct->_conf->bluestore_extent_map_inline_shard_prealloc_size) { | |
2366 | } | |
2367 | ||
9f95a23c TL |
2368 | void BlueStore::ExtentMap::dump(Formatter* f) const |
2369 | { | |
2370 | f->open_array_section("extents"); | |
2371 | ||
2372 | for (auto& e : extent_map) { | |
2373 | f->dump_object("extent", e); | |
2374 | } | |
2375 | f->close_section(); | |
2376 | } | |
2377 | ||
11fdf7f2 TL |
2378 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2379 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2380 | uint64_t& length, uint64_t& dstoff) { | |
2381 | ||
2382 | auto cct = onode->c->store->cct; | |
2383 | bool inject_21040 = | |
2384 | cct->_conf->bluestore_debug_inject_bug21040; | |
2385 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2386 | for (auto& e : oldo->extent_map.extent_map) { | |
2387 | e.blob->last_encoded_id = -1; | |
2388 | } | |
2389 | ||
2390 | int n = 0; | |
2391 | uint64_t end = srcoff + length; | |
2392 | uint32_t dirty_range_begin = 0; | |
2393 | uint32_t dirty_range_end = 0; | |
2394 | bool src_dirty = false; | |
2395 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2396 | ep != oldo->extent_map.extent_map.end(); | |
2397 | ++ep) { | |
2398 | auto& e = *ep; | |
2399 | if (e.logical_offset >= end) { | |
2400 | break; | |
2401 | } | |
2402 | dout(20) << __func__ << " src " << e << dendl; | |
2403 | BlobRef cb; | |
2404 | bool blob_duped = true; | |
2405 | if (e.blob->last_encoded_id >= 0) { | |
2406 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2407 | blob_duped = false; | |
2408 | } else { | |
2409 | // dup the blob | |
2410 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2411 | // make sure it is shared | |
2412 | if (!blob.is_shared()) { | |
2413 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2414 | if (!inject_21040 && !src_dirty) { | |
2415 | src_dirty = true; | |
2416 | dirty_range_begin = e.logical_offset; | |
2417 | } else if (inject_21040 && | |
2418 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2419 | dirty_range_begin = e.logical_offset; | |
2420 | } | |
2421 | ceph_assert(e.logical_end() > 0); | |
2422 | // -1 to exclude next potential shard | |
2423 | dirty_range_end = e.logical_end() - 1; | |
2424 | } else { | |
2425 | c->load_shared_blob(e.blob->shared_blob); | |
2426 | } | |
2427 | cb = new Blob(); | |
2428 | e.blob->last_encoded_id = n; | |
2429 | id_to_blob[n] = cb; | |
2430 | e.blob->dup(*cb); | |
2431 | // bump the extent refs on the copied blob's extents | |
2432 | for (auto p : blob.get_extents()) { | |
2433 | if (p.is_valid()) { | |
2434 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2435 | } | |
2436 | } | |
2437 | txc->write_shared_blob(e.blob->shared_blob); | |
2438 | dout(20) << __func__ << " new " << *cb << dendl; | |
2439 | } | |
2440 | ||
2441 | int skip_front, skip_back; | |
2442 | if (e.logical_offset < srcoff) { | |
2443 | skip_front = srcoff - e.logical_offset; | |
2444 | } else { | |
2445 | skip_front = 0; | |
2446 | } | |
2447 | if (e.logical_end() > end) { | |
2448 | skip_back = e.logical_end() - end; | |
2449 | } else { | |
2450 | skip_back = 0; | |
2451 | } | |
2452 | ||
2453 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2454 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2455 | newo->extent_map.extent_map.insert(*ne); | |
2456 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2457 | // fixme: we may leave parts of new blob unreferenced that could | |
2458 | // be freed (relative to the shared_blob). | |
2459 | txc->statfs_delta.stored() += ne->length; | |
2460 | if (e.blob->get_blob().is_compressed()) { | |
2461 | txc->statfs_delta.compressed_original() += ne->length; | |
2462 | if (blob_duped) { | |
2463 | txc->statfs_delta.compressed() += | |
2464 | cb->get_blob().get_compressed_payload_length(); | |
2465 | } | |
2466 | } | |
2467 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2468 | ++n; | |
2469 | } | |
2470 | if ((!inject_21040 && src_dirty) || | |
2471 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2472 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2473 | dirty_range_end - dirty_range_begin); | |
2474 | txc->write_onode(oldo); | |
2475 | } | |
2476 | txc->write_onode(newo); | |
2477 | ||
2478 | if (dstoff + length > newo->onode.size) { | |
2479 | newo->onode.size = dstoff + length; | |
2480 | } | |
2481 | newo->extent_map.dirty_range(dstoff, length); | |
2482 | } | |
7c673cae FG |
2483 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2484 | bool force) | |
2485 | { | |
2486 | auto cct = onode->c->store->cct; //used by dout | |
2487 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2488 | if (onode->onode.extent_map_shards.empty()) { | |
2489 | if (inline_bl.length() == 0) { | |
2490 | unsigned n; | |
2491 | // we need to encode inline_bl to measure encoded length | |
2492 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
f91f0fd5 | 2493 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_inline_bl); |
11fdf7f2 | 2494 | ceph_assert(!never_happen); |
7c673cae FG |
2495 | size_t len = inline_bl.length(); |
2496 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2497 | << " extents" << dendl; | |
2498 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2499 | request_reshard(0, OBJECT_MAX_SIZE); | |
2500 | return; | |
2501 | } | |
2502 | } | |
2503 | // will persist in the onode key. | |
2504 | } else { | |
2505 | // pending shard update | |
2506 | struct dirty_shard_t { | |
2507 | Shard *shard; | |
2508 | bufferlist bl; | |
2509 | dirty_shard_t(Shard *s) : shard(s) {} | |
2510 | }; | |
2511 | vector<dirty_shard_t> encoded_shards; | |
2512 | // allocate slots for all shards in a single call instead of | |
2513 | // doing multiple allocations - one per each dirty shard | |
2514 | encoded_shards.reserve(shards.size()); | |
2515 | ||
2516 | auto p = shards.begin(); | |
2517 | auto prev_p = p; | |
2518 | while (p != shards.end()) { | |
11fdf7f2 | 2519 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2520 | auto n = p; |
2521 | ++n; | |
2522 | if (p->dirty) { | |
2523 | uint32_t endoff; | |
2524 | if (n == shards.end()) { | |
2525 | endoff = OBJECT_MAX_SIZE; | |
2526 | } else { | |
2527 | endoff = n->shard_info->offset; | |
2528 | } | |
2529 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2530 | bufferlist& bl = encoded_shards.back().bl; | |
2531 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2532 | bl, &p->extents)) { | |
2533 | if (force) { | |
2534 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2535 | ceph_assert(!force); |
7c673cae FG |
2536 | } |
2537 | } | |
2538 | size_t len = bl.length(); | |
2539 | ||
2540 | dout(20) << __func__ << " shard 0x" << std::hex | |
2541 | << p->shard_info->offset << std::dec << " is " << len | |
2542 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2543 | << p->extents << " extents" << dendl; | |
2544 | ||
2545 | if (!force) { | |
2546 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2547 | // we are big; reshard ourselves | |
2548 | request_reshard(p->shard_info->offset, endoff); | |
2549 | } | |
2550 | // avoid resharding the trailing shard, even if it is small | |
2551 | else if (n != shards.end() && | |
11fdf7f2 TL |
2552 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2553 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2554 | if (p == shards.begin()) { |
2555 | // we are the first shard, combine with next shard | |
7c673cae | 2556 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2557 | } else { |
31f18b77 FG |
2558 | // combine either with the previous shard or the next, |
2559 | // whichever is smaller | |
7c673cae FG |
2560 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2561 | request_reshard(p->shard_info->offset, endoff + 1); | |
2562 | } else { | |
2563 | request_reshard(prev_p->shard_info->offset, endoff); | |
2564 | } | |
2565 | } | |
2566 | } | |
2567 | } | |
2568 | } | |
2569 | prev_p = p; | |
2570 | p = n; | |
2571 | } | |
2572 | if (needs_reshard()) { | |
2573 | return; | |
2574 | } | |
2575 | ||
2576 | // schedule DB update for dirty shards | |
2577 | string key; | |
2578 | for (auto& it : encoded_shards) { | |
2579 | it.shard->dirty = false; | |
2580 | it.shard->shard_info->bytes = it.bl.length(); | |
2581 | generate_extent_shard_key_and_apply( | |
2582 | onode->key, | |
2583 | it.shard->shard_info->offset, | |
2584 | &key, | |
2585 | [&](const string& final_key) { | |
2586 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2587 | } | |
2588 | ); | |
2589 | } | |
2590 | } | |
2591 | } | |
2592 | ||
31f18b77 FG |
2593 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2594 | { | |
2595 | if (spanning_blob_map.empty()) | |
2596 | return 0; | |
2597 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2598 | // bid is valid and available. | |
2599 | if (bid >= 0) | |
2600 | return bid; | |
2601 | // Find next unused bid; | |
2602 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2603 | const auto begin_bid = bid; | |
2604 | do { | |
2605 | if (!spanning_blob_map.count(bid)) | |
2606 | return bid; | |
2607 | else { | |
2608 | bid++; | |
2609 | if (bid < 0) bid = 0; | |
2610 | } | |
2611 | } while (bid != begin_bid); | |
81eedcae TL |
2612 | auto cct = onode->c->store->cct; // used by dout |
2613 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2614 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2615 | } |
2616 | ||
7c673cae FG |
2617 | void BlueStore::ExtentMap::reshard( |
2618 | KeyValueDB *db, | |
2619 | KeyValueDB::Transaction t) | |
2620 | { | |
2621 | auto cct = onode->c->store->cct; // used by dout | |
2622 | ||
2623 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2624 | << needs_reshard_end << ")" << std::dec | |
2625 | << " of " << onode->onode.extent_map_shards.size() | |
2626 | << " shards on " << onode->oid << dendl; | |
2627 | for (auto& p : spanning_blob_map) { | |
2628 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2629 | << dendl; | |
2630 | } | |
2631 | // determine shard index range | |
2632 | unsigned si_begin = 0, si_end = 0; | |
2633 | if (!shards.empty()) { | |
2634 | while (si_begin + 1 < shards.size() && | |
2635 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2636 | ++si_begin; | |
2637 | } | |
2638 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2639 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2640 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2641 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2642 | break; | |
2643 | } | |
2644 | } | |
2645 | if (si_end == shards.size()) { | |
2646 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2647 | } | |
2648 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2649 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2650 | << needs_reshard_end << ")" << std::dec << dendl; | |
2651 | } | |
2652 | ||
181888fb | 2653 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2654 | |
2655 | // we may need to fault in a larger interval later must have all | |
2656 | // referring extents for spanning blobs loaded in order to have | |
2657 | // accurate use_tracker values. | |
2658 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2659 | uint32_t spanning_scan_end = needs_reshard_end; | |
2660 | ||
2661 | // remove old keys | |
2662 | string key; | |
2663 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2664 | generate_extent_shard_key_and_apply( | |
2665 | onode->key, shards[i].shard_info->offset, &key, | |
2666 | [&](const string& final_key) { | |
2667 | t->rmkey(PREFIX_OBJ, final_key); | |
2668 | } | |
2669 | ); | |
2670 | } | |
2671 | ||
2672 | // calculate average extent size | |
2673 | unsigned bytes = 0; | |
2674 | unsigned extents = 0; | |
2675 | if (onode->onode.extent_map_shards.empty()) { | |
2676 | bytes = inline_bl.length(); | |
2677 | extents = extent_map.size(); | |
2678 | } else { | |
2679 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2680 | bytes += shards[i].shard_info->bytes; | |
2681 | extents += shards[i].extents; | |
2682 | } | |
2683 | } | |
2684 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2685 | unsigned slop = target * | |
2686 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2687 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2688 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2689 | << ", slop " << slop << dendl; | |
2690 | ||
2691 | // reshard | |
2692 | unsigned estimate = 0; | |
31f18b77 | 2693 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2694 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2695 | unsigned max_blob_end = 0; | |
2696 | Extent dummy(needs_reshard_begin); | |
2697 | for (auto e = extent_map.lower_bound(dummy); | |
2698 | e != extent_map.end(); | |
2699 | ++e) { | |
2700 | if (e->logical_offset >= needs_reshard_end) { | |
2701 | break; | |
2702 | } | |
2703 | dout(30) << " extent " << *e << dendl; | |
2704 | ||
2705 | // disfavor shard boundaries that span a blob | |
2706 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2707 | if (estimate && | |
2708 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2709 | // new shard | |
31f18b77 | 2710 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2711 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2712 | new_shard_info.back().offset = offset; | |
2713 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2714 | << std::dec << dendl; |
7c673cae FG |
2715 | } |
2716 | offset = e->logical_offset; | |
2717 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2718 | new_shard_info.back().offset = offset; | |
2719 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2720 | << std::dec << dendl; | |
2721 | estimate = 0; | |
2722 | } | |
2723 | estimate += extent_avg; | |
31f18b77 FG |
2724 | unsigned bs = e->blob_start(); |
2725 | if (bs < spanning_scan_begin) { | |
2726 | spanning_scan_begin = bs; | |
7c673cae FG |
2727 | } |
2728 | uint32_t be = e->blob_end(); | |
2729 | if (be > max_blob_end) { | |
2730 | max_blob_end = be; | |
2731 | } | |
2732 | if (be > spanning_scan_end) { | |
2733 | spanning_scan_end = be; | |
2734 | } | |
2735 | } | |
2736 | if (new_shard_info.empty() && (si_begin > 0 || | |
2737 | si_end < shards.size())) { | |
2738 | // we resharded a partial range; we must produce at least one output | |
2739 | // shard | |
2740 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2741 | new_shard_info.back().offset = needs_reshard_begin; | |
2742 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2743 | << std::dec << " (singleton degenerate case)" << dendl; | |
2744 | } | |
2745 | ||
2746 | auto& sv = onode->onode.extent_map_shards; | |
2747 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2748 | dout(20) << __func__ << " old " << sv << dendl; | |
2749 | if (sv.empty()) { | |
2750 | // no old shards to keep | |
2751 | sv.swap(new_shard_info); | |
2752 | init_shards(true, true); | |
2753 | } else { | |
2754 | // splice in new shards | |
2755 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2756 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2757 | sv.insert( | |
2758 | sv.begin() + si_begin, | |
2759 | new_shard_info.begin(), | |
2760 | new_shard_info.end()); | |
2761 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2762 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2763 | |
11fdf7f2 | 2764 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2765 | |
2766 | // note that we need to update every shard_info of shards here, | |
2767 | // as sv might have been totally re-allocated above | |
2768 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2769 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2770 | } |
2771 | ||
2772 | // mark newly added shards as dirty | |
2773 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2774 | shards[i].loaded = true; |
2775 | shards[i].dirty = true; | |
2776 | } | |
7c673cae FG |
2777 | } |
2778 | dout(20) << __func__ << " fin " << sv << dendl; | |
2779 | inline_bl.clear(); | |
2780 | ||
2781 | if (sv.empty()) { | |
2782 | // no more shards; unspan all previously spanning blobs | |
2783 | auto p = spanning_blob_map.begin(); | |
2784 | while (p != spanning_blob_map.end()) { | |
2785 | p->second->id = -1; | |
2786 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2787 | p = spanning_blob_map.erase(p); | |
2788 | } | |
2789 | } else { | |
2790 | // identify new spanning blobs | |
2791 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2792 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2793 | if (spanning_scan_begin < needs_reshard_begin) { | |
2794 | fault_range(db, spanning_scan_begin, | |
2795 | needs_reshard_begin - spanning_scan_begin); | |
2796 | } | |
2797 | if (spanning_scan_end > needs_reshard_end) { | |
2798 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2799 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2800 | } |
2801 | auto sp = sv.begin() + si_begin; | |
2802 | auto esp = sv.end(); | |
2803 | unsigned shard_start = sp->offset; | |
2804 | unsigned shard_end; | |
2805 | ++sp; | |
2806 | if (sp == esp) { | |
2807 | shard_end = OBJECT_MAX_SIZE; | |
2808 | } else { | |
2809 | shard_end = sp->offset; | |
2810 | } | |
7c673cae | 2811 | Extent dummy(needs_reshard_begin); |
9f95a23c TL |
2812 | |
2813 | bool was_too_many_blobs_check = false; | |
2814 | auto too_many_blobs_threshold = | |
2815 | g_conf()->bluestore_debug_too_many_blobs_threshold; | |
2816 | auto& dumped_onodes = onode->c->onode_map.cache->dumped_onodes; | |
2817 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oid_slot = nullptr; | |
2818 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oldest_slot = nullptr; | |
2819 | ||
7c673cae FG |
2820 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { |
2821 | if (e->logical_offset >= needs_reshard_end) { | |
2822 | break; | |
2823 | } | |
2824 | dout(30) << " extent " << *e << dendl; | |
2825 | while (e->logical_offset >= shard_end) { | |
2826 | shard_start = shard_end; | |
11fdf7f2 | 2827 | ceph_assert(sp != esp); |
7c673cae FG |
2828 | ++sp; |
2829 | if (sp == esp) { | |
2830 | shard_end = OBJECT_MAX_SIZE; | |
2831 | } else { | |
2832 | shard_end = sp->offset; | |
2833 | } | |
2834 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2835 | << " to 0x" << shard_end << std::dec << dendl; | |
2836 | } | |
9f95a23c | 2837 | |
7c673cae FG |
2838 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { |
2839 | if (!e->blob->is_spanning()) { | |
2840 | // We have two options: (1) split the blob into pieces at the | |
2841 | // shard boundaries (and adjust extents accordingly), or (2) | |
2842 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2843 | // we may have to split it multiple times--potentially at every | |
2844 | // shard boundary. | |
2845 | bool must_span = false; | |
2846 | BlobRef b = e->blob; | |
2847 | if (b->can_split()) { | |
2848 | uint32_t bstart = e->blob_start(); | |
2849 | uint32_t bend = e->blob_end(); | |
2850 | for (const auto& sh : shards) { | |
2851 | if (bstart < sh.shard_info->offset && | |
2852 | bend > sh.shard_info->offset) { | |
2853 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2854 | if (b->can_split_at(blob_offset)) { | |
2855 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2856 | << std::hex << bstart << " blob_offset 0x" | |
2857 | << blob_offset << std::dec << " " << *b << dendl; | |
2858 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2859 | // switch b to the new right-hand side, in case it | |
2860 | // *also* has to get split. | |
2861 | bstart += blob_offset; | |
2862 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2863 | } else { | |
2864 | must_span = true; | |
2865 | break; | |
2866 | } | |
2867 | } | |
2868 | } | |
2869 | } else { | |
2870 | must_span = true; | |
2871 | } | |
2872 | if (must_span) { | |
31f18b77 FG |
2873 | auto bid = allocate_spanning_blob_id(); |
2874 | b->id = bid; | |
7c673cae FG |
2875 | spanning_blob_map[b->id] = b; |
2876 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
9f95a23c TL |
2877 | if (!was_too_many_blobs_check && |
2878 | too_many_blobs_threshold && | |
2879 | spanning_blob_map.size() >= size_t(too_many_blobs_threshold)) { | |
2880 | ||
2881 | was_too_many_blobs_check = true; | |
2882 | for (size_t i = 0; i < dumped_onodes.size(); ++i) { | |
2883 | if (dumped_onodes[i].first == onode->oid) { | |
2884 | oid_slot = &dumped_onodes[i]; | |
2885 | break; | |
2886 | } | |
2887 | if (!oldest_slot || (oldest_slot && | |
2888 | dumped_onodes[i].second < oldest_slot->second)) { | |
2889 | oldest_slot = &dumped_onodes[i]; | |
2890 | } | |
2891 | } | |
2892 | } | |
7c673cae FG |
2893 | } |
2894 | } | |
2895 | } else { | |
2896 | if (e->blob->is_spanning()) { | |
2897 | spanning_blob_map.erase(e->blob->id); | |
2898 | e->blob->id = -1; | |
2899 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
2900 | } | |
2901 | } | |
2902 | } | |
9f95a23c TL |
2903 | bool do_dump = (!oid_slot && was_too_many_blobs_check) || |
2904 | (oid_slot && | |
2905 | (mono_clock::now() - oid_slot->second >= make_timespan(5 * 60))); | |
2906 | if (do_dump) { | |
2907 | dout(0) << __func__ | |
2908 | << " spanning blob count exceeds threshold, " | |
2909 | << spanning_blob_map.size() << " spanning blobs" | |
2910 | << dendl; | |
2911 | _dump_onode<0>(cct, *onode); | |
2912 | if (oid_slot) { | |
2913 | oid_slot->second = mono_clock::now(); | |
2914 | } else { | |
2915 | ceph_assert(oldest_slot); | |
2916 | oldest_slot->first = onode->oid; | |
2917 | oldest_slot->second = mono_clock::now(); | |
2918 | } | |
2919 | } | |
7c673cae FG |
2920 | } |
2921 | ||
2922 | clear_needs_reshard(); | |
2923 | } | |
2924 | ||
2925 | bool BlueStore::ExtentMap::encode_some( | |
2926 | uint32_t offset, | |
2927 | uint32_t length, | |
2928 | bufferlist& bl, | |
2929 | unsigned *pn) | |
2930 | { | |
7c673cae FG |
2931 | Extent dummy(offset); |
2932 | auto start = extent_map.lower_bound(dummy); | |
2933 | uint32_t end = offset + length; | |
2934 | ||
2935 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
2936 | // serialization only. Hence there is no specific | |
2937 | // handling at ExtentMap level. | |
2938 | ||
2939 | unsigned n = 0; | |
2940 | size_t bound = 0; | |
7c673cae FG |
2941 | bool must_reshard = false; |
2942 | for (auto p = start; | |
2943 | p != extent_map.end() && p->logical_offset < end; | |
2944 | ++p, ++n) { | |
11fdf7f2 | 2945 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
2946 | p->blob->last_encoded_id = -1; |
2947 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
2948 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2949 | << std::dec << " hit new spanning blob " << *p << dendl; | |
2950 | request_reshard(p->blob_start(), p->blob_end()); | |
2951 | must_reshard = true; | |
2952 | } | |
31f18b77 FG |
2953 | if (!must_reshard) { |
2954 | denc_varint(0, bound); // blobid | |
2955 | denc_varint(0, bound); // logical_offset | |
2956 | denc_varint(0, bound); // len | |
2957 | denc_varint(0, bound); // blob_offset | |
7c673cae | 2958 | |
31f18b77 FG |
2959 | p->blob->bound_encode( |
2960 | bound, | |
2961 | struct_v, | |
2962 | p->blob->shared_blob->get_sbid(), | |
2963 | false); | |
2964 | } | |
7c673cae FG |
2965 | } |
2966 | if (must_reshard) { | |
2967 | return true; | |
2968 | } | |
2969 | ||
31f18b77 FG |
2970 | denc(struct_v, bound); |
2971 | denc_varint(0, bound); // number of extents | |
2972 | ||
7c673cae FG |
2973 | { |
2974 | auto app = bl.get_contiguous_appender(bound); | |
2975 | denc(struct_v, app); | |
2976 | denc_varint(n, app); | |
2977 | if (pn) { | |
2978 | *pn = n; | |
2979 | } | |
2980 | ||
2981 | n = 0; | |
2982 | uint64_t pos = 0; | |
2983 | uint64_t prev_len = 0; | |
2984 | for (auto p = start; | |
2985 | p != extent_map.end() && p->logical_offset < end; | |
2986 | ++p, ++n) { | |
2987 | unsigned blobid; | |
2988 | bool include_blob = false; | |
2989 | if (p->blob->is_spanning()) { | |
2990 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
2991 | blobid |= BLOBID_FLAG_SPANNING; | |
2992 | } else if (p->blob->last_encoded_id < 0) { | |
2993 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
2994 | include_blob = true; | |
2995 | blobid = 0; // the decoder will infer the id from n | |
2996 | } else { | |
2997 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
2998 | } | |
2999 | if (p->logical_offset == pos) { | |
3000 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
3001 | } | |
3002 | if (p->blob_offset == 0) { | |
3003 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
3004 | } | |
3005 | if (p->length == prev_len) { | |
3006 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
3007 | } else { | |
3008 | prev_len = p->length; | |
3009 | } | |
3010 | denc_varint(blobid, app); | |
3011 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3012 | denc_varint_lowz(p->logical_offset - pos, app); | |
3013 | } | |
3014 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3015 | denc_varint_lowz(p->blob_offset, app); | |
3016 | } | |
3017 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3018 | denc_varint_lowz(p->length, app); | |
3019 | } | |
3020 | pos = p->logical_end(); | |
3021 | if (include_blob) { | |
3022 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
3023 | } | |
3024 | } | |
3025 | } | |
3026 | /*derr << __func__ << bl << dendl; | |
3027 | derr << __func__ << ":"; | |
3028 | bl.hexdump(*_dout); | |
3029 | *_dout << dendl; | |
3030 | */ | |
3031 | return false; | |
3032 | } | |
3033 | ||
3034 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
3035 | { | |
7c673cae FG |
3036 | /* |
3037 | derr << __func__ << ":"; | |
3038 | bl.hexdump(*_dout); | |
3039 | *_dout << dendl; | |
3040 | */ | |
3041 | ||
11fdf7f2 | 3042 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae FG |
3043 | auto p = bl.front().begin_deep(); |
3044 | __u8 struct_v; | |
3045 | denc(struct_v, p); | |
3046 | // Version 2 differs from v1 in blob's ref_map | |
3047 | // serialization only. Hence there is no specific | |
3048 | // handling at ExtentMap level below. | |
11fdf7f2 | 3049 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
3050 | |
3051 | uint32_t num; | |
3052 | denc_varint(num, p); | |
3053 | vector<BlobRef> blobs(num); | |
3054 | uint64_t pos = 0; | |
3055 | uint64_t prev_len = 0; | |
3056 | unsigned n = 0; | |
3057 | ||
3058 | while (!p.end()) { | |
3059 | Extent *le = new Extent(); | |
3060 | uint64_t blobid; | |
3061 | denc_varint(blobid, p); | |
3062 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3063 | uint64_t gap; | |
3064 | denc_varint_lowz(gap, p); | |
3065 | pos += gap; | |
3066 | } | |
3067 | le->logical_offset = pos; | |
3068 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3069 | denc_varint_lowz(le->blob_offset, p); | |
3070 | } else { | |
3071 | le->blob_offset = 0; | |
3072 | } | |
3073 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3074 | denc_varint_lowz(prev_len, p); | |
3075 | } | |
3076 | le->length = prev_len; | |
3077 | ||
3078 | if (blobid & BLOBID_FLAG_SPANNING) { | |
3079 | dout(30) << __func__ << " getting spanning blob " | |
3080 | << (blobid >> BLOBID_SHIFT_BITS) << dendl; | |
3081 | le->assign_blob(get_spanning_blob(blobid >> BLOBID_SHIFT_BITS)); | |
3082 | } else { | |
3083 | blobid >>= BLOBID_SHIFT_BITS; | |
3084 | if (blobid) { | |
3085 | le->assign_blob(blobs[blobid - 1]); | |
11fdf7f2 | 3086 | ceph_assert(le->blob); |
7c673cae FG |
3087 | } else { |
3088 | Blob *b = new Blob(); | |
3089 | uint64_t sbid = 0; | |
3090 | b->decode(onode->c, p, struct_v, &sbid, false); | |
3091 | blobs[n] = b; | |
3092 | onode->c->open_shared_blob(sbid, b); | |
3093 | le->assign_blob(b); | |
3094 | } | |
3095 | // we build ref_map dynamically for non-spanning blobs | |
3096 | le->blob->get_ref( | |
3097 | onode->c, | |
3098 | le->blob_offset, | |
3099 | le->length); | |
3100 | } | |
3101 | pos += prev_len; | |
3102 | ++n; | |
3103 | extent_map.insert(*le); | |
3104 | } | |
3105 | ||
11fdf7f2 | 3106 | ceph_assert(n == num); |
7c673cae FG |
3107 | return num; |
3108 | } | |
3109 | ||
3110 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
3111 | { | |
3112 | // Version 2 differs from v1 in blob's ref_map | |
3113 | // serialization only. Hence there is no specific | |
3114 | // handling at ExtentMap level. | |
3115 | __u8 struct_v = 2; | |
3116 | ||
3117 | denc(struct_v, p); | |
3118 | denc_varint((uint32_t)0, p); | |
3119 | size_t key_size = 0; | |
3120 | denc_varint((uint32_t)0, key_size); | |
3121 | p += spanning_blob_map.size() * key_size; | |
3122 | for (const auto& i : spanning_blob_map) { | |
3123 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3124 | } | |
3125 | } | |
3126 | ||
3127 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
3128 | bufferlist::contiguous_appender& p) | |
3129 | { | |
3130 | // Version 2 differs from v1 in blob's ref_map | |
3131 | // serialization only. Hence there is no specific | |
3132 | // handling at ExtentMap level. | |
3133 | __u8 struct_v = 2; | |
3134 | ||
3135 | denc(struct_v, p); | |
3136 | denc_varint(spanning_blob_map.size(), p); | |
3137 | for (auto& i : spanning_blob_map) { | |
3138 | denc_varint(i.second->id, p); | |
3139 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3140 | } | |
3141 | } | |
3142 | ||
3143 | void BlueStore::ExtentMap::decode_spanning_blobs( | |
11fdf7f2 | 3144 | bufferptr::const_iterator& p) |
7c673cae FG |
3145 | { |
3146 | __u8 struct_v; | |
3147 | denc(struct_v, p); | |
3148 | // Version 2 differs from v1 in blob's ref_map | |
3149 | // serialization only. Hence there is no specific | |
3150 | // handling at ExtentMap level. | |
11fdf7f2 | 3151 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
3152 | |
3153 | unsigned n; | |
3154 | denc_varint(n, p); | |
3155 | while (n--) { | |
3156 | BlobRef b(new Blob()); | |
3157 | denc_varint(b->id, p); | |
3158 | spanning_blob_map[b->id] = b; | |
3159 | uint64_t sbid = 0; | |
3160 | b->decode(onode->c, p, struct_v, &sbid, true); | |
3161 | onode->c->open_shared_blob(sbid, b); | |
3162 | } | |
3163 | } | |
3164 | ||
3165 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) | |
3166 | { | |
3167 | shards.resize(onode->onode.extent_map_shards.size()); | |
3168 | unsigned i = 0; | |
3169 | for (auto &s : onode->onode.extent_map_shards) { | |
3170 | shards[i].shard_info = &s; | |
3171 | shards[i].loaded = loaded; | |
3172 | shards[i].dirty = dirty; | |
3173 | ++i; | |
3174 | } | |
3175 | } | |
3176 | ||
3177 | void BlueStore::ExtentMap::fault_range( | |
3178 | KeyValueDB *db, | |
3179 | uint32_t offset, | |
3180 | uint32_t length) | |
3181 | { | |
7c673cae FG |
3182 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3183 | << std::dec << dendl; | |
3184 | auto start = seek_shard(offset); | |
3185 | auto last = seek_shard(offset + length); | |
3186 | ||
3187 | if (start < 0) | |
3188 | return; | |
3189 | ||
11fdf7f2 | 3190 | ceph_assert(last >= start); |
7c673cae FG |
3191 | string key; |
3192 | while (start <= last) { | |
11fdf7f2 | 3193 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3194 | auto p = &shards[start]; |
3195 | if (!p->loaded) { | |
3196 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
3197 | << p->shard_info->offset << std::dec << dendl; | |
3198 | bufferlist v; | |
3199 | generate_extent_shard_key_and_apply( | |
3200 | onode->key, p->shard_info->offset, &key, | |
3201 | [&](const string& final_key) { | |
3202 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
3203 | if (r < 0) { | |
3204 | derr << __func__ << " missing shard 0x" << std::hex | |
3205 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
3206 | << dendl; | |
11fdf7f2 | 3207 | ceph_assert(r >= 0); |
7c673cae FG |
3208 | } |
3209 | } | |
3210 | ); | |
3211 | p->extents = decode_some(v); | |
3212 | p->loaded = true; | |
3213 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
3214 | << p->shard_info->offset |
3215 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 3216 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
3217 | ceph_assert(p->dirty == false); |
3218 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
3219 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
3220 | } else { | |
3221 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
3222 | } | |
3223 | ++start; | |
3224 | } | |
3225 | } | |
3226 | ||
3227 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
3228 | uint32_t offset, |
3229 | uint32_t length) | |
3230 | { | |
7c673cae FG |
3231 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3232 | << std::dec << dendl; | |
3233 | if (shards.empty()) { | |
3234 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
3235 | inline_bl.clear(); | |
3236 | return; | |
3237 | } | |
3238 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
3239 | if (length == 0) { |
3240 | length = 1; | |
3241 | } | |
3242 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
3243 | if (start < 0) |
3244 | return; | |
3245 | ||
11fdf7f2 | 3246 | ceph_assert(last >= start); |
7c673cae | 3247 | while (start <= last) { |
11fdf7f2 | 3248 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3249 | auto p = &shards[start]; |
3250 | if (!p->loaded) { | |
11fdf7f2 TL |
3251 | derr << __func__ << "on write 0x" << std::hex << offset |
3252 | << "~" << length << " shard 0x" << p->shard_info->offset | |
3253 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
3254 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
3255 | } |
3256 | if (!p->dirty) { | |
3257 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
3258 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
3259 | p->dirty = true; | |
3260 | } | |
3261 | ++start; | |
3262 | } | |
3263 | } | |
3264 | ||
3265 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
3266 | uint64_t offset) | |
3267 | { | |
3268 | Extent dummy(offset); | |
3269 | return extent_map.find(dummy); | |
3270 | } | |
3271 | ||
7c673cae FG |
3272 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
3273 | uint64_t offset) | |
3274 | { | |
3275 | Extent dummy(offset); | |
3276 | auto fp = extent_map.lower_bound(dummy); | |
3277 | if (fp != extent_map.begin()) { | |
3278 | --fp; | |
3279 | if (fp->logical_end() <= offset) { | |
3280 | ++fp; | |
3281 | } | |
3282 | } | |
3283 | return fp; | |
3284 | } | |
3285 | ||
3286 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
3287 | uint64_t offset) const | |
3288 | { | |
3289 | Extent dummy(offset); | |
3290 | auto fp = extent_map.lower_bound(dummy); | |
3291 | if (fp != extent_map.begin()) { | |
3292 | --fp; | |
3293 | if (fp->logical_end() <= offset) { | |
3294 | ++fp; | |
3295 | } | |
3296 | } | |
3297 | return fp; | |
3298 | } | |
3299 | ||
3300 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
3301 | { | |
3302 | auto fp = seek_lextent(offset); | |
3303 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
3304 | return false; | |
3305 | } | |
3306 | return true; | |
3307 | } | |
3308 | ||
3309 | int BlueStore::ExtentMap::compress_extent_map( | |
3310 | uint64_t offset, | |
3311 | uint64_t length) | |
3312 | { | |
7c673cae FG |
3313 | if (extent_map.empty()) |
3314 | return 0; | |
3315 | int removed = 0; | |
3316 | auto p = seek_lextent(offset); | |
3317 | if (p != extent_map.begin()) { | |
3318 | --p; // start to the left of offset | |
3319 | } | |
3320 | // the caller should have just written to this region | |
11fdf7f2 | 3321 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
3322 | |
3323 | // identify the *next* shard | |
3324 | auto pshard = shards.begin(); | |
3325 | while (pshard != shards.end() && | |
3326 | p->logical_offset >= pshard->shard_info->offset) { | |
3327 | ++pshard; | |
3328 | } | |
3329 | uint64_t shard_end; | |
3330 | if (pshard != shards.end()) { | |
3331 | shard_end = pshard->shard_info->offset; | |
3332 | } else { | |
3333 | shard_end = OBJECT_MAX_SIZE; | |
3334 | } | |
3335 | ||
3336 | auto n = p; | |
3337 | for (++n; n != extent_map.end(); p = n++) { | |
3338 | if (n->logical_offset > offset + length) { | |
3339 | break; // stop after end | |
3340 | } | |
3341 | while (n != extent_map.end() && | |
3342 | p->logical_end() == n->logical_offset && | |
3343 | p->blob == n->blob && | |
3344 | p->blob_offset + p->length == n->blob_offset && | |
3345 | n->logical_offset < shard_end) { | |
3346 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3347 | << " next shard 0x" << shard_end << std::dec | |
3348 | << " merging " << *p << " and " << *n << dendl; | |
3349 | p->length += n->length; | |
3350 | rm(n++); | |
3351 | ++removed; | |
3352 | } | |
3353 | if (n == extent_map.end()) { | |
3354 | break; | |
3355 | } | |
3356 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3357 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3358 | ++pshard; |
3359 | if (pshard != shards.end()) { | |
3360 | shard_end = pshard->shard_info->offset; | |
3361 | } else { | |
3362 | shard_end = OBJECT_MAX_SIZE; | |
3363 | } | |
3364 | } | |
3365 | } | |
11fdf7f2 | 3366 | if (removed) { |
7c673cae FG |
3367 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3368 | } | |
3369 | return removed; | |
3370 | } | |
3371 | ||
3372 | void BlueStore::ExtentMap::punch_hole( | |
3373 | CollectionRef &c, | |
3374 | uint64_t offset, | |
3375 | uint64_t length, | |
3376 | old_extent_map_t *old_extents) | |
3377 | { | |
3378 | auto p = seek_lextent(offset); | |
3379 | uint64_t end = offset + length; | |
3380 | while (p != extent_map.end()) { | |
3381 | if (p->logical_offset >= end) { | |
3382 | break; | |
3383 | } | |
3384 | if (p->logical_offset < offset) { | |
3385 | if (p->logical_end() > end) { | |
3386 | // split and deref middle | |
3387 | uint64_t front = offset - p->logical_offset; | |
3388 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3389 | length, p->blob); | |
3390 | old_extents->push_back(*oe); | |
3391 | add(end, | |
3392 | p->blob_offset + front + length, | |
3393 | p->length - front - length, | |
3394 | p->blob); | |
3395 | p->length = front; | |
3396 | break; | |
3397 | } else { | |
3398 | // deref tail | |
11fdf7f2 | 3399 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3400 | uint64_t keep = offset - p->logical_offset; |
3401 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3402 | p->length - keep, p->blob); | |
3403 | old_extents->push_back(*oe); | |
3404 | p->length = keep; | |
3405 | ++p; | |
3406 | continue; | |
3407 | } | |
3408 | } | |
3409 | if (p->logical_offset + p->length <= end) { | |
3410 | // deref whole lextent | |
3411 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3412 | p->length, p->blob); | |
3413 | old_extents->push_back(*oe); | |
3414 | rm(p++); | |
3415 | continue; | |
3416 | } | |
3417 | // deref head | |
3418 | uint64_t keep = p->logical_end() - end; | |
3419 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3420 | p->length - keep, p->blob); | |
3421 | old_extents->push_back(*oe); | |
3422 | ||
3423 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3424 | rm(p); | |
3425 | break; | |
3426 | } | |
3427 | } | |
3428 | ||
3429 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3430 | CollectionRef &c, | |
3431 | uint64_t logical_offset, | |
3432 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3433 | old_extent_map_t *old_extents) | |
3434 | { | |
3435 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3436 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3437 | |
3438 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3439 | // old_extents list if we overwre the blob totally | |
3440 | // This might happen during WAL overwrite. | |
3441 | b->get_ref(onode->c, blob_offset, length); | |
3442 | ||
3443 | if (old_extents) { | |
3444 | punch_hole(c, logical_offset, length, old_extents); | |
3445 | } | |
3446 | ||
3447 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3448 | extent_map.insert(*le); | |
3449 | if (spans_shard(logical_offset, length)) { | |
3450 | request_reshard(logical_offset, logical_offset + length); | |
3451 | } | |
3452 | return le; | |
3453 | } | |
3454 | ||
3455 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3456 | BlobRef lb, | |
3457 | uint32_t blob_offset, | |
3458 | uint32_t pos) | |
3459 | { | |
7c673cae FG |
3460 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; |
3461 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3462 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3463 | << dendl; | |
3464 | BlobRef rb = onode->c->new_blob(); | |
3465 | lb->split(onode->c, blob_offset, rb.get()); | |
3466 | ||
3467 | for (auto ep = seek_lextent(pos); | |
3468 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3469 | ++ep) { | |
3470 | if (ep->blob != lb) { | |
3471 | continue; | |
3472 | } | |
3473 | if (ep->logical_offset < pos) { | |
3474 | // split extent | |
3475 | size_t left = pos - ep->logical_offset; | |
3476 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3477 | extent_map.insert(*ne); | |
3478 | ep->length = left; | |
3479 | dout(30) << __func__ << " split " << *ep << dendl; | |
3480 | dout(30) << __func__ << " to " << *ne << dendl; | |
3481 | } else { | |
3482 | // switch blob | |
11fdf7f2 | 3483 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3484 | |
3485 | ep->blob = rb; | |
3486 | ep->blob_offset -= blob_offset; | |
3487 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3488 | } | |
3489 | } | |
3490 | return rb; | |
3491 | } | |
3492 | ||
3493 | // Onode | |
3494 | ||
3495 | #undef dout_prefix | |
3496 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3497 | ||
f6b5b4d7 TL |
3498 | // |
3499 | // A tricky thing about Onode's ref counter is that we do an additional | |
3500 | // increment when newly pinned instance is detected. And -1 on unpin. | |
3501 | // This prevents from a conflict with a delete call (when nref == 0). | |
3502 | // The latter might happen while the thread is in unpin() function | |
3503 | // (and e.g. waiting for lock acquisition) since nref is already | |
3504 | // decremented. And another 'putting' thread on the instance will release it. | |
3505 | // | |
3506 | void BlueStore::Onode::get() { | |
adb31ebb TL |
3507 | if (++nref >= 2 && !pinned) { |
3508 | OnodeCacheShard* ocs = c->get_onode_cache(); | |
f67539c2 TL |
3509 | ocs->lock.lock(); |
3510 | // It is possible that during waiting split_cache moved us to different OnodeCacheShard. | |
3511 | while (ocs != c->get_onode_cache()) { | |
3512 | ocs->lock.unlock(); | |
3513 | ocs = c->get_onode_cache(); | |
3514 | ocs->lock.lock(); | |
3515 | } | |
adb31ebb TL |
3516 | bool was_pinned = pinned; |
3517 | pinned = nref >= 2; | |
3518 | // additional increment for newly pinned instance | |
3519 | bool r = !was_pinned && pinned; | |
3520 | if (r) { | |
3521 | ++nref; | |
3522 | } | |
3523 | if (cached && r) { | |
3524 | ocs->_pin(this); | |
3525 | } | |
f67539c2 | 3526 | ocs->lock.unlock(); |
f6b5b4d7 TL |
3527 | } |
3528 | } | |
3529 | void BlueStore::Onode::put() { | |
adb31ebb TL |
3530 | int n = --nref; |
3531 | if (n == 2) { | |
3532 | OnodeCacheShard* ocs = c->get_onode_cache(); | |
f67539c2 TL |
3533 | ocs->lock.lock(); |
3534 | // It is possible that during waiting split_cache moved us to different OnodeCacheShard. | |
3535 | while (ocs != c->get_onode_cache()) { | |
3536 | ocs->lock.unlock(); | |
3537 | ocs = c->get_onode_cache(); | |
3538 | ocs->lock.lock(); | |
3539 | } | |
adb31ebb TL |
3540 | bool need_unpin = pinned; |
3541 | pinned = pinned && nref > 2; // intentionally use > not >= as we have | |
3542 | // +1 due to pinned state | |
3543 | need_unpin = need_unpin && !pinned; | |
3544 | if (cached && need_unpin) { | |
3545 | if (exists) { | |
3546 | ocs->_unpin(this); | |
3547 | } else { | |
3548 | ocs->_unpin_and_rm(this); | |
3549 | // remove will also decrement nref and delete Onode | |
3550 | c->onode_map._remove(oid); | |
3551 | } | |
3552 | } | |
3553 | // additional decrement for newly unpinned instance | |
3554 | // should be the last action since Onode can be released | |
3555 | // at any point after this decrement | |
3556 | if (need_unpin) { | |
3557 | n = --nref; | |
3558 | } | |
f67539c2 | 3559 | ocs->lock.unlock(); |
f6b5b4d7 | 3560 | } |
adb31ebb | 3561 | if (n == 0) { |
f6b5b4d7 TL |
3562 | delete this; |
3563 | } | |
3564 | } | |
3565 | ||
eafe8130 TL |
3566 | BlueStore::Onode* BlueStore::Onode::decode( |
3567 | CollectionRef c, | |
3568 | const ghobject_t& oid, | |
3569 | const string& key, | |
3570 | const bufferlist& v) | |
3571 | { | |
3572 | Onode* on = new Onode(c.get(), oid, key); | |
3573 | on->exists = true; | |
3574 | auto p = v.front().begin_deep(); | |
3575 | on->onode.decode(p); | |
3576 | for (auto& i : on->onode.attrs) { | |
f91f0fd5 | 3577 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
eafe8130 TL |
3578 | } |
3579 | ||
3580 | // initialize extent_map | |
3581 | on->extent_map.decode_spanning_blobs(p); | |
3582 | if (on->onode.extent_map_shards.empty()) { | |
3583 | denc(on->extent_map.inline_bl, p); | |
3584 | on->extent_map.decode_some(on->extent_map.inline_bl); | |
3585 | on->extent_map.inline_bl.reassign_to_mempool( | |
f91f0fd5 | 3586 | mempool::mempool_bluestore_cache_data); |
eafe8130 TL |
3587 | } |
3588 | else { | |
3589 | on->extent_map.init_shards(false, false); | |
3590 | } | |
3591 | return on; | |
3592 | } | |
3593 | ||
7c673cae FG |
3594 | void BlueStore::Onode::flush() |
3595 | { | |
3596 | if (flushing_count.load()) { | |
3597 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
9f95a23c | 3598 | waiting_count++; |
11fdf7f2 | 3599 | std::unique_lock l(flush_lock); |
7c673cae FG |
3600 | while (flushing_count.load()) { |
3601 | flush_cond.wait(l); | |
3602 | } | |
9f95a23c | 3603 | waiting_count--; |
7c673cae FG |
3604 | } |
3605 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3606 | } | |
3607 | ||
9f95a23c TL |
3608 | void BlueStore::Onode::dump(Formatter* f) const |
3609 | { | |
3610 | onode.dump(f); | |
3611 | extent_map.dump(f); | |
3612 | } | |
3613 | ||
3614 | ||
3615 | const string& BlueStore::Onode::get_omap_prefix() | |
3616 | { | |
3617 | if (onode.is_pgmeta_omap()) { | |
3618 | return PREFIX_PGMETA_OMAP; | |
3619 | } | |
f67539c2 TL |
3620 | if (onode.is_perpg_omap()) { |
3621 | return PREFIX_PERPG_OMAP; | |
3622 | } | |
9f95a23c TL |
3623 | if (onode.is_perpool_omap()) { |
3624 | return PREFIX_PERPOOL_OMAP; | |
3625 | } | |
3626 | return PREFIX_OMAP; | |
3627 | } | |
3628 | ||
3629 | // '-' < '.' < '~' | |
3630 | ||
3631 | void BlueStore::Onode::get_omap_header(string *out) | |
3632 | { | |
f67539c2 TL |
3633 | if (!onode.is_pgmeta_omap()) { |
3634 | if (onode.is_perpg_omap()) { | |
3635 | _key_encode_u64(c->pool(), out); | |
3636 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), out); | |
3637 | } else if (onode.is_perpool_omap()) { | |
3638 | _key_encode_u64(c->pool(), out); | |
3639 | } | |
9f95a23c TL |
3640 | } |
3641 | _key_encode_u64(onode.nid, out); | |
3642 | out->push_back('-'); | |
3643 | } | |
3644 | ||
3645 | void BlueStore::Onode::get_omap_key(const string& key, string *out) | |
3646 | { | |
f67539c2 TL |
3647 | if (!onode.is_pgmeta_omap()) { |
3648 | if (onode.is_perpg_omap()) { | |
3649 | _key_encode_u64(c->pool(), out); | |
3650 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), out); | |
3651 | } else if (onode.is_perpool_omap()) { | |
3652 | _key_encode_u64(c->pool(), out); | |
3653 | } | |
9f95a23c TL |
3654 | } |
3655 | _key_encode_u64(onode.nid, out); | |
3656 | out->push_back('.'); | |
3657 | out->append(key); | |
3658 | } | |
3659 | ||
3660 | void BlueStore::Onode::rewrite_omap_key(const string& old, string *out) | |
3661 | { | |
f67539c2 TL |
3662 | if (!onode.is_pgmeta_omap()) { |
3663 | if (onode.is_perpg_omap()) { | |
3664 | _key_encode_u64(c->pool(), out); | |
3665 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), out); | |
3666 | } else if (onode.is_perpool_omap()) { | |
3667 | _key_encode_u64(c->pool(), out); | |
3668 | } | |
9f95a23c TL |
3669 | } |
3670 | _key_encode_u64(onode.nid, out); | |
3671 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
3672 | } | |
3673 | ||
3674 | void BlueStore::Onode::get_omap_tail(string *out) | |
3675 | { | |
f67539c2 TL |
3676 | if (!onode.is_pgmeta_omap()) { |
3677 | if (onode.is_perpg_omap()) { | |
3678 | _key_encode_u64(c->pool(), out); | |
3679 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), out); | |
3680 | } else if (onode.is_perpool_omap()) { | |
3681 | _key_encode_u64(c->pool(), out); | |
3682 | } | |
9f95a23c TL |
3683 | } |
3684 | _key_encode_u64(onode.nid, out); | |
3685 | out->push_back('~'); | |
3686 | } | |
3687 | ||
3688 | void BlueStore::Onode::decode_omap_key(const string& key, string *user_key) | |
3689 | { | |
f67539c2 TL |
3690 | size_t pos = sizeof(uint64_t) + 1; |
3691 | if (!onode.is_pgmeta_omap()) { | |
3692 | if (onode.is_perpg_omap()) { | |
3693 | pos += sizeof(uint64_t) + sizeof(uint32_t); | |
3694 | } else if (onode.is_perpool_omap()) { | |
3695 | pos += sizeof(uint64_t); | |
3696 | } | |
9f95a23c | 3697 | } |
f67539c2 | 3698 | *user_key = key.substr(pos); |
9f95a23c TL |
3699 | } |
3700 | ||
3701 | ||
7c673cae FG |
3702 | // ======================================================= |
3703 | // WriteContext | |
3704 | ||
3705 | /// Checks for writes to the same pextent within a blob | |
3706 | bool BlueStore::WriteContext::has_conflict( | |
3707 | BlobRef b, | |
3708 | uint64_t loffs, | |
3709 | uint64_t loffs_end, | |
3710 | uint64_t min_alloc_size) | |
3711 | { | |
11fdf7f2 TL |
3712 | ceph_assert((loffs % min_alloc_size) == 0); |
3713 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3714 | for (auto w : writes) { |
3715 | if (b == w.b) { | |
11fdf7f2 TL |
3716 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3717 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3718 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3719 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3720 | return true; |
3721 | } | |
3722 | } | |
3723 | } | |
3724 | return false; | |
3725 | } | |
3726 | ||
3727 | // ======================================================= | |
3728 | ||
3729 | // DeferredBatch | |
3730 | #undef dout_prefix | |
3731 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
9f95a23c TL |
3732 | #undef dout_context |
3733 | #define dout_context cct | |
7c673cae FG |
3734 | |
3735 | void BlueStore::DeferredBatch::prepare_write( | |
3736 | CephContext *cct, | |
3737 | uint64_t seq, uint64_t offset, uint64_t length, | |
3738 | bufferlist::const_iterator& blp) | |
3739 | { | |
3740 | _discard(cct, offset, length); | |
3741 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3742 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3743 | i.first->second.seq = seq; |
3744 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3745 | i.first->second.bl.reassign_to_mempool( |
3746 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3747 | dout(20) << __func__ << " seq " << seq |
3748 | << " 0x" << std::hex << offset << "~" << length | |
3749 | << " crc " << i.first->second.bl.crc32c(-1) | |
3750 | << std::dec << dendl; | |
3751 | seq_bytes[seq] += length; | |
3752 | #ifdef DEBUG_DEFERRED | |
3753 | _audit(cct); | |
3754 | #endif | |
3755 | } | |
3756 | ||
3757 | void BlueStore::DeferredBatch::_discard( | |
3758 | CephContext *cct, uint64_t offset, uint64_t length) | |
3759 | { | |
3760 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3761 | << std::dec << dendl; | |
3762 | auto p = iomap.lower_bound(offset); | |
3763 | if (p != iomap.begin()) { | |
3764 | --p; | |
3765 | auto end = p->first + p->second.bl.length(); | |
3766 | if (end > offset) { | |
3767 | bufferlist head; | |
3768 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3769 | dout(20) << __func__ << " keep head " << p->second.seq | |
3770 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3771 | << " -> 0x" << head.length() << std::dec << dendl; | |
3772 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3773 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3774 | if (end > offset + length) { |
3775 | bufferlist tail; | |
3776 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3777 | end - (offset + length)); | |
3778 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3779 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3780 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3781 | auto &n = iomap[offset + length]; | |
3782 | n.bl.swap(tail); | |
3783 | n.seq = p->second.seq; | |
3784 | i->second -= length; | |
3785 | } else { | |
3786 | i->second -= end - offset; | |
3787 | } | |
11fdf7f2 | 3788 | ceph_assert(i->second >= 0); |
7c673cae FG |
3789 | p->second.bl.swap(head); |
3790 | } | |
3791 | ++p; | |
3792 | } | |
3793 | while (p != iomap.end()) { | |
3794 | if (p->first >= offset + length) { | |
3795 | break; | |
3796 | } | |
3797 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3798 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3799 | auto end = p->first + p->second.bl.length(); |
3800 | if (end > offset + length) { | |
3801 | unsigned drop_front = offset + length - p->first; | |
3802 | unsigned keep_tail = end - (offset + length); | |
3803 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3804 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3805 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3806 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3807 | << std::dec << dendl; | |
3808 | auto &s = iomap[offset + length]; | |
3809 | s.seq = p->second.seq; | |
3810 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3811 | i->second -= drop_front; | |
3812 | } else { | |
3813 | dout(20) << __func__ << " drop " << p->second.seq | |
3814 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3815 | << std::dec << dendl; | |
3816 | i->second -= p->second.bl.length(); | |
3817 | } | |
11fdf7f2 | 3818 | ceph_assert(i->second >= 0); |
7c673cae FG |
3819 | p = iomap.erase(p); |
3820 | } | |
3821 | } | |
3822 | ||
3823 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3824 | { | |
3825 | map<uint64_t,int> sb; | |
3826 | for (auto p : seq_bytes) { | |
3827 | sb[p.first] = 0; // make sure we have the same set of keys | |
3828 | } | |
3829 | uint64_t pos = 0; | |
3830 | for (auto& p : iomap) { | |
11fdf7f2 | 3831 | ceph_assert(p.first >= pos); |
7c673cae FG |
3832 | sb[p.second.seq] += p.second.bl.length(); |
3833 | pos = p.first + p.second.bl.length(); | |
3834 | } | |
11fdf7f2 | 3835 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3836 | } |
3837 | ||
3838 | ||
3839 | // Collection | |
3840 | ||
3841 | #undef dout_prefix | |
3842 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3843 | ||
9f95a23c TL |
3844 | BlueStore::Collection::Collection(BlueStore *store_, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t cid) |
3845 | : CollectionImpl(store_->cct, cid), | |
11fdf7f2 | 3846 | store(store_), |
9f95a23c | 3847 | cache(bc), |
7c673cae | 3848 | exists(true), |
9f95a23c | 3849 | onode_map(oc), |
11fdf7f2 TL |
3850 | commit_queue(nullptr) |
3851 | { | |
3852 | } | |
3853 | ||
3854 | bool BlueStore::Collection::flush_commit(Context *c) | |
3855 | { | |
3856 | return osr->flush_commit(c); | |
3857 | } | |
3858 | ||
3859 | void BlueStore::Collection::flush() | |
3860 | { | |
3861 | osr->flush(); | |
3862 | } | |
3863 | ||
3864 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3865 | { |
11fdf7f2 | 3866 | osr->flush_all_but_last(); |
7c673cae FG |
3867 | } |
3868 | ||
3869 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3870 | { | |
11fdf7f2 | 3871 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3872 | const bluestore_blob_t& blob = b->get_blob(); |
3873 | if (!blob.is_shared()) { | |
3874 | b->shared_blob = new SharedBlob(this); | |
3875 | return; | |
3876 | } | |
3877 | ||
3878 | b->shared_blob = shared_blob_set.lookup(sbid); | |
3879 | if (b->shared_blob) { | |
3880 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3881 | << std::dec << " had " << *b->shared_blob << dendl; | |
3882 | } else { | |
3883 | b->shared_blob = new SharedBlob(sbid, this); | |
3884 | shared_blob_set.add(this, b->shared_blob.get()); | |
3885 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3886 | << std::dec << " opened " << *b->shared_blob | |
3887 | << dendl; | |
3888 | } | |
3889 | } | |
3890 | ||
3891 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
3892 | { | |
3893 | if (!sb->is_loaded()) { | |
3894 | ||
3895 | bufferlist v; | |
3896 | string key; | |
3897 | auto sbid = sb->get_sbid(); | |
3898 | get_shared_blob_key(sbid, &key); | |
3899 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
3900 | if (r < 0) { | |
3901 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
3902 | << std::dec << " not found at key " | |
3903 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 3904 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
3905 | } |
3906 | ||
3907 | sb->loaded = true; | |
3908 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
3909 | auto p = v.cbegin(); |
3910 | decode(*(sb->persistent), p); | |
7c673cae FG |
3911 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
3912 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
3913 | } | |
3914 | } | |
3915 | ||
3916 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
3917 | { | |
7c673cae | 3918 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 3919 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
3920 | |
3921 | // update blob | |
31f18b77 | 3922 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 3923 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
3924 | |
3925 | // update shared blob | |
3926 | b->shared_blob->loaded = true; | |
3927 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
3928 | shared_blob_set.add(this, b->shared_blob.get()); | |
3929 | for (auto p : blob.get_extents()) { | |
3930 | if (p.is_valid()) { | |
3931 | b->shared_blob->get_ref( | |
3932 | p.offset, | |
3933 | p.length); | |
3934 | } | |
3935 | } | |
3936 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
3937 | } | |
3938 | ||
31f18b77 FG |
3939 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
3940 | { | |
3941 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 3942 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
3943 | |
3944 | uint64_t sbid = sb->get_sbid(); | |
3945 | shared_blob_set.remove(sb); | |
3946 | sb->loaded = false; | |
3947 | delete sb->persistent; | |
3948 | sb->sbid_unloaded = 0; | |
3949 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
3950 | return sbid; | |
3951 | } | |
3952 | ||
7c673cae FG |
3953 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
3954 | const ghobject_t& oid, | |
9f95a23c TL |
3955 | bool create, |
3956 | bool is_createop) | |
7c673cae | 3957 | { |
9f95a23c | 3958 | ceph_assert(create ? ceph_mutex_is_wlocked(lock) : ceph_mutex_is_locked(lock)); |
7c673cae FG |
3959 | |
3960 | spg_t pgid; | |
3961 | if (cid.is_pg(&pgid)) { | |
3962 | if (!oid.match(cnode.bits, pgid.ps())) { | |
3963 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
3964 | << pgid << " bits " << cnode.bits << dendl; | |
3965 | ceph_abort(); | |
3966 | } | |
3967 | } | |
3968 | ||
3969 | OnodeRef o = onode_map.lookup(oid); | |
3970 | if (o) | |
3971 | return o; | |
3972 | ||
eafe8130 | 3973 | string key; |
7c673cae FG |
3974 | get_object_key(store->cct, oid, &key); |
3975 | ||
3976 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
3977 | << pretty_binary_string(key) << dendl; | |
3978 | ||
3979 | bufferlist v; | |
9f95a23c | 3980 | int r = -ENOENT; |
7c673cae | 3981 | Onode *on; |
9f95a23c TL |
3982 | if (!is_createop) { |
3983 | r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
3984 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
3985 | } | |
7c673cae | 3986 | if (v.length() == 0) { |
11fdf7f2 | 3987 | ceph_assert(r == -ENOENT); |
f67539c2 | 3988 | if (!create) |
7c673cae FG |
3989 | return OnodeRef(); |
3990 | ||
3991 | // new object, new onode | |
3992 | on = new Onode(this, oid, key); | |
3993 | } else { | |
3994 | // loaded | |
11fdf7f2 | 3995 | ceph_assert(r >= 0); |
eafe8130 | 3996 | on = Onode::decode(this, oid, key, v); |
7c673cae FG |
3997 | } |
3998 | o.reset(on); | |
3999 | return onode_map.add(oid, o); | |
4000 | } | |
4001 | ||
4002 | void BlueStore::Collection::split_cache( | |
4003 | Collection *dest) | |
4004 | { | |
4005 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
4006 | ||
f67539c2 TL |
4007 | auto *ocache = get_onode_cache(); |
4008 | auto *ocache_dest = dest->get_onode_cache(); | |
4009 | ||
4010 | // lock cache shards | |
4011 | std::lock(ocache->lock, ocache_dest->lock, cache->lock, dest->cache->lock); | |
4012 | std::lock_guard l(ocache->lock, std::adopt_lock); | |
4013 | std::lock_guard l2(ocache_dest->lock, std::adopt_lock); | |
4014 | std::lock_guard l3(cache->lock, std::adopt_lock); | |
4015 | std::lock_guard l4(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
4016 | |
4017 | int destbits = dest->cnode.bits; | |
4018 | spg_t destpg; | |
4019 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 4020 | ceph_assert(is_pg); |
7c673cae FG |
4021 | |
4022 | auto p = onode_map.onode_map.begin(); | |
4023 | while (p != onode_map.onode_map.end()) { | |
11fdf7f2 | 4024 | OnodeRef o = p->second; |
7c673cae FG |
4025 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
4026 | // onode does not belong to this child | |
11fdf7f2 TL |
4027 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
4028 | << dendl; | |
7c673cae FG |
4029 | ++p; |
4030 | } else { | |
7c673cae FG |
4031 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
4032 | << dendl; | |
4033 | ||
f6b5b4d7 TL |
4034 | // ensuring that nref is always >= 2 and hence onode is pinned and |
4035 | // physically out of cache during the transition | |
4036 | OnodeRef o_pin = o; | |
4037 | ceph_assert(o->pinned); | |
4038 | ||
7c673cae | 4039 | p = onode_map.onode_map.erase(p); |
7c673cae | 4040 | dest->onode_map.onode_map[o->oid] = o; |
adb31ebb | 4041 | if (o->cached) { |
f6b5b4d7 | 4042 | get_onode_cache()->move_pinned(dest->get_onode_cache(), o.get()); |
9f95a23c | 4043 | } |
f6b5b4d7 | 4044 | o->c = dest; |
7c673cae FG |
4045 | |
4046 | // move over shared blobs and buffers. cover shared blobs from | |
4047 | // both extent map and spanning blob map (the full extent map | |
4048 | // may not be faulted in) | |
4049 | vector<SharedBlob*> sbvec; | |
4050 | for (auto& e : o->extent_map.extent_map) { | |
4051 | sbvec.push_back(e.blob->shared_blob.get()); | |
4052 | } | |
4053 | for (auto& b : o->extent_map.spanning_blob_map) { | |
4054 | sbvec.push_back(b.second->shared_blob.get()); | |
4055 | } | |
4056 | for (auto sb : sbvec) { | |
4057 | if (sb->coll == dest) { | |
4058 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
4059 | << dendl; | |
4060 | continue; | |
4061 | } | |
4062 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
4063 | if (sb->get_sbid()) { |
4064 | ldout(store->cct, 20) << __func__ | |
4065 | << " moving registration " << *sb << dendl; | |
4066 | shared_blob_set.remove(sb); | |
4067 | dest->shared_blob_set.add(dest, sb); | |
4068 | } | |
3efd9988 | 4069 | sb->coll = dest; |
7c673cae | 4070 | if (dest->cache != cache) { |
7c673cae FG |
4071 | for (auto& i : sb->bc.buffer_map) { |
4072 | if (!i.second->is_writing()) { | |
4073 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
4074 | << dendl; | |
9f95a23c | 4075 | dest->cache->_move(cache, i.second.get()); |
7c673cae FG |
4076 | } |
4077 | } | |
4078 | } | |
4079 | } | |
7c673cae FG |
4080 | } |
4081 | } | |
9f95a23c | 4082 | dest->cache->_trim(); |
7c673cae FG |
4083 | } |
4084 | ||
7c673cae FG |
4085 | // ======================================================= |
4086 | ||
91327a77 AA |
4087 | // MempoolThread |
4088 | ||
4089 | #undef dout_prefix | |
4090 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
9f95a23c TL |
4091 | #undef dout_context |
4092 | #define dout_context store->cct | |
91327a77 | 4093 | |
7c673cae FG |
4094 | void *BlueStore::MempoolThread::entry() |
4095 | { | |
9f95a23c | 4096 | std::unique_lock l{lock}; |
11fdf7f2 | 4097 | |
92f5a8d4 | 4098 | uint32_t prev_config_change = store->config_changed.load(); |
eafe8130 TL |
4099 | uint64_t base = store->osd_memory_base; |
4100 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4101 | uint64_t target = store->osd_memory_target; | |
4102 | uint64_t min = store->osd_memory_cache_min; | |
4103 | uint64_t max = min; | |
4104 | ||
4105 | // When setting the maximum amount of memory to use for cache, first | |
4106 | // assume some base amount of memory for the OSD and then fudge in | |
4107 | // some overhead for fragmentation that scales with cache usage. | |
4108 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4109 | if (ltarget > base + min) { | |
4110 | max = ltarget - base; | |
11fdf7f2 | 4111 | } |
31f18b77 | 4112 | |
eafe8130 | 4113 | binned_kv_cache = store->db->get_priority_cache(); |
f67539c2 | 4114 | binned_kv_onode_cache = store->db->get_priority_cache(PREFIX_OBJ); |
eafe8130 TL |
4115 | if (store->cache_autotune && binned_kv_cache != nullptr) { |
4116 | pcm = std::make_shared<PriorityCache::Manager>( | |
f67539c2 | 4117 | store->cct, min, max, target, true, "bluestore-pricache"); |
eafe8130 TL |
4118 | pcm->insert("kv", binned_kv_cache, true); |
4119 | pcm->insert("meta", meta_cache, true); | |
4120 | pcm->insert("data", data_cache, true); | |
f67539c2 TL |
4121 | if (binned_kv_onode_cache != nullptr) { |
4122 | pcm->insert("kv_onode", binned_kv_onode_cache, true); | |
4123 | } | |
eafe8130 | 4124 | } |
91327a77 AA |
4125 | |
4126 | utime_t next_balance = ceph_clock_now(); | |
4127 | utime_t next_resize = ceph_clock_now(); | |
9f95a23c TL |
4128 | utime_t next_deferred_force_submit = ceph_clock_now(); |
4129 | utime_t alloc_stats_dump_clock = ceph_clock_now(); | |
31f18b77 | 4130 | |
91327a77 | 4131 | bool interval_stats_trim = false; |
91327a77 | 4132 | while (!stop) { |
92f5a8d4 TL |
4133 | // Update pcm cache settings if related configuration was changed |
4134 | uint32_t cur_config_change = store->config_changed.load(); | |
4135 | if (cur_config_change != prev_config_change) { | |
4136 | _update_cache_settings(); | |
4137 | prev_config_change = cur_config_change; | |
4138 | } | |
4139 | ||
91327a77 AA |
4140 | // Before we trim, check and see if it's time to rebalance/resize. |
4141 | double autotune_interval = store->cache_autotune_interval; | |
4142 | double resize_interval = store->osd_memory_cache_resize_interval; | |
9f95a23c TL |
4143 | double max_defer_interval = store->max_defer_interval; |
4144 | ||
4145 | double alloc_stats_dump_interval = | |
4146 | store->cct->_conf->bluestore_alloc_stats_dump_interval; | |
91327a77 | 4147 | |
9f95a23c TL |
4148 | if (alloc_stats_dump_interval > 0 && |
4149 | alloc_stats_dump_clock + alloc_stats_dump_interval < ceph_clock_now()) { | |
4150 | store->_record_allocation_stats(); | |
4151 | alloc_stats_dump_clock = ceph_clock_now(); | |
4152 | } | |
91327a77 | 4153 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { |
11fdf7f2 TL |
4154 | _adjust_cache_settings(); |
4155 | ||
91327a77 | 4156 | // Log events at 5 instead of 20 when balance happens. |
91327a77 | 4157 | interval_stats_trim = true; |
eafe8130 TL |
4158 | |
4159 | if (pcm != nullptr) { | |
4160 | pcm->balance(); | |
91327a77 | 4161 | } |
31f18b77 | 4162 | |
91327a77 AA |
4163 | next_balance = ceph_clock_now(); |
4164 | next_balance += autotune_interval; | |
4165 | } | |
4166 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { | |
eafe8130 TL |
4167 | if (ceph_using_tcmalloc() && pcm != nullptr) { |
4168 | pcm->tune_memory(); | |
91327a77 AA |
4169 | } |
4170 | next_resize = ceph_clock_now(); | |
4171 | next_resize += resize_interval; | |
31f18b77 FG |
4172 | } |
4173 | ||
9f95a23c TL |
4174 | if (max_defer_interval > 0 && |
4175 | next_deferred_force_submit < ceph_clock_now()) { | |
4176 | if (store->get_deferred_last_submitted() + max_defer_interval < | |
4177 | ceph_clock_now()) { | |
4178 | store->deferred_try_submit(); | |
4179 | } | |
4180 | next_deferred_force_submit = ceph_clock_now(); | |
4181 | next_deferred_force_submit += max_defer_interval/3; | |
4182 | } | |
4183 | ||
4184 | // Now Resize the shards | |
4185 | _resize_shards(interval_stats_trim); | |
91327a77 | 4186 | interval_stats_trim = false; |
31f18b77 | 4187 | |
91327a77 | 4188 | store->_update_cache_logger(); |
11fdf7f2 TL |
4189 | auto wait = ceph::make_timespan( |
4190 | store->cct->_conf->bluestore_cache_trim_interval); | |
4191 | cond.wait_for(l, wait); | |
7c673cae | 4192 | } |
9f95a23c TL |
4193 | // do final dump |
4194 | store->_record_allocation_stats(); | |
7c673cae | 4195 | stop = false; |
f67539c2 | 4196 | pcm = nullptr; |
7c673cae FG |
4197 | return NULL; |
4198 | } | |
4199 | ||
91327a77 AA |
4200 | void BlueStore::MempoolThread::_adjust_cache_settings() |
4201 | { | |
11fdf7f2 TL |
4202 | if (binned_kv_cache != nullptr) { |
4203 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
4204 | } | |
f67539c2 TL |
4205 | if (binned_kv_onode_cache != nullptr) { |
4206 | binned_kv_onode_cache->set_cache_ratio(store->cache_kv_onode_ratio); | |
4207 | } | |
11fdf7f2 TL |
4208 | meta_cache->set_cache_ratio(store->cache_meta_ratio); |
4209 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
91327a77 AA |
4210 | } |
4211 | ||
9f95a23c | 4212 | void BlueStore::MempoolThread::_resize_shards(bool interval_stats) |
91327a77 | 4213 | { |
9f95a23c TL |
4214 | size_t onode_shards = store->onode_cache_shards.size(); |
4215 | size_t buffer_shards = store->buffer_cache_shards.size(); | |
91327a77 | 4216 | int64_t kv_used = store->db->get_cache_usage(); |
f67539c2 | 4217 | int64_t kv_onode_used = store->db->get_cache_usage(PREFIX_OBJ); |
11fdf7f2 TL |
4218 | int64_t meta_used = meta_cache->_get_used_bytes(); |
4219 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
4220 | |
4221 | uint64_t cache_size = store->cache_size; | |
4222 | int64_t kv_alloc = | |
11fdf7f2 | 4223 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
f67539c2 TL |
4224 | int64_t kv_onode_alloc = |
4225 | static_cast<int64_t>(store->cache_kv_onode_ratio * cache_size); | |
91327a77 | 4226 | int64_t meta_alloc = |
11fdf7f2 | 4227 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 4228 | int64_t data_alloc = |
11fdf7f2 | 4229 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 4230 | |
eafe8130 TL |
4231 | if (pcm != nullptr && binned_kv_cache != nullptr) { |
4232 | cache_size = pcm->get_tuned_mem(); | |
11fdf7f2 TL |
4233 | kv_alloc = binned_kv_cache->get_committed_size(); |
4234 | meta_alloc = meta_cache->get_committed_size(); | |
4235 | data_alloc = data_cache->get_committed_size(); | |
f67539c2 TL |
4236 | if (binned_kv_onode_cache != nullptr) { |
4237 | kv_onode_alloc = binned_kv_onode_cache->get_committed_size(); | |
4238 | } | |
91327a77 AA |
4239 | } |
4240 | ||
4241 | if (interval_stats) { | |
9f95a23c | 4242 | dout(5) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4243 | << " kv_alloc: " << kv_alloc |
4244 | << " kv_used: " << kv_used | |
f67539c2 TL |
4245 | << " kv_onode_alloc: " << kv_onode_alloc |
4246 | << " kv_onode_used: " << kv_onode_used | |
91327a77 AA |
4247 | << " meta_alloc: " << meta_alloc |
4248 | << " meta_used: " << meta_used | |
4249 | << " data_alloc: " << data_alloc | |
4250 | << " data_used: " << data_used << dendl; | |
4251 | } else { | |
9f95a23c | 4252 | dout(20) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4253 | << " kv_alloc: " << kv_alloc |
4254 | << " kv_used: " << kv_used | |
f67539c2 TL |
4255 | << " kv_onode_alloc: " << kv_onode_alloc |
4256 | << " kv_onode_used: " << kv_onode_used | |
91327a77 AA |
4257 | << " meta_alloc: " << meta_alloc |
4258 | << " meta_used: " << meta_used | |
4259 | << " data_alloc: " << data_alloc | |
4260 | << " data_used: " << data_used << dendl; | |
4261 | } | |
4262 | ||
4263 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
9f95a23c TL |
4264 | (meta_alloc / (double) onode_shards) / meta_cache->get_bytes_per_onode()); |
4265 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / buffer_shards); | |
91327a77 | 4266 | |
9f95a23c | 4267 | dout(30) << __func__ << " max_shard_onodes: " << max_shard_onodes |
91327a77 AA |
4268 | << " max_shard_buffer: " << max_shard_buffer << dendl; |
4269 | ||
9f95a23c TL |
4270 | for (auto i : store->onode_cache_shards) { |
4271 | i->set_max(max_shard_onodes); | |
4272 | } | |
4273 | for (auto i : store->buffer_cache_shards) { | |
4274 | i->set_max(max_shard_buffer); | |
91327a77 AA |
4275 | } |
4276 | } | |
4277 | ||
92f5a8d4 TL |
4278 | void BlueStore::MempoolThread::_update_cache_settings() |
4279 | { | |
4280 | // Nothing to do if pcm is not used. | |
4281 | if (pcm == nullptr) { | |
4282 | return; | |
4283 | } | |
4284 | ||
92f5a8d4 TL |
4285 | uint64_t target = store->osd_memory_target; |
4286 | uint64_t base = store->osd_memory_base; | |
4287 | uint64_t min = store->osd_memory_cache_min; | |
4288 | uint64_t max = min; | |
4289 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4290 | ||
4291 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4292 | if (ltarget > base + min) { | |
4293 | max = ltarget - base; | |
4294 | } | |
4295 | ||
4296 | // set pcm cache levels | |
4297 | pcm->set_target_memory(target); | |
4298 | pcm->set_min_memory(min); | |
4299 | pcm->set_max_memory(max); | |
4300 | ||
9f95a23c | 4301 | dout(5) << __func__ << " updated pcm target: " << target |
92f5a8d4 TL |
4302 | << " pcm min: " << min |
4303 | << " pcm max: " << max | |
4304 | << dendl; | |
4305 | } | |
4306 | ||
7c673cae FG |
4307 | // ======================================================= |
4308 | ||
31f18b77 FG |
4309 | // OmapIteratorImpl |
4310 | ||
4311 | #undef dout_prefix | |
4312 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
4313 | ||
4314 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
4315 | CollectionRef c, OnodeRef o, KeyValueDB::Iterator it) | |
4316 | : c(c), o(o), it(it) | |
4317 | { | |
9f95a23c | 4318 | std::shared_lock l(c->lock); |
31f18b77 | 4319 | if (o->onode.has_omap()) { |
9f95a23c TL |
4320 | o->get_omap_key(string(), &head); |
4321 | o->get_omap_tail(&tail); | |
31f18b77 FG |
4322 | it->lower_bound(head); |
4323 | } | |
4324 | } | |
4325 | ||
11fdf7f2 TL |
4326 | string BlueStore::OmapIteratorImpl::_stringify() const |
4327 | { | |
4328 | stringstream s; | |
4329 | s << " omap_iterator(cid = " << c->cid | |
4330 | <<", oid = " << o->oid << ")"; | |
4331 | return s.str(); | |
4332 | } | |
4333 | ||
31f18b77 FG |
4334 | int BlueStore::OmapIteratorImpl::seek_to_first() |
4335 | { | |
9f95a23c | 4336 | std::shared_lock l(c->lock); |
11fdf7f2 | 4337 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4338 | if (o->onode.has_omap()) { |
4339 | it->lower_bound(head); | |
4340 | } else { | |
4341 | it = KeyValueDB::Iterator(); | |
4342 | } | |
494da23a TL |
4343 | c->store->log_latency( |
4344 | __func__, | |
11fdf7f2 TL |
4345 | l_bluestore_omap_seek_to_first_lat, |
4346 | mono_clock::now() - start1, | |
494da23a | 4347 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 | 4348 | |
31f18b77 FG |
4349 | return 0; |
4350 | } | |
4351 | ||
4352 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
4353 | { | |
9f95a23c | 4354 | std::shared_lock l(c->lock); |
11fdf7f2 | 4355 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4356 | if (o->onode.has_omap()) { |
4357 | string key; | |
9f95a23c | 4358 | o->get_omap_key(after, &key); |
31f18b77 FG |
4359 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " |
4360 | << pretty_binary_string(key) << dendl; | |
4361 | it->upper_bound(key); | |
4362 | } else { | |
4363 | it = KeyValueDB::Iterator(); | |
4364 | } | |
11fdf7f2 | 4365 | c->store->log_latency_fn( |
494da23a | 4366 | __func__, |
11fdf7f2 TL |
4367 | l_bluestore_omap_upper_bound_lat, |
4368 | mono_clock::now() - start1, | |
494da23a | 4369 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4370 | [&] (const ceph::timespan& lat) { |
494da23a | 4371 | return ", after = " + after + |
11fdf7f2 TL |
4372 | _stringify(); |
4373 | } | |
4374 | ); | |
31f18b77 FG |
4375 | return 0; |
4376 | } | |
4377 | ||
4378 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
4379 | { | |
9f95a23c | 4380 | std::shared_lock l(c->lock); |
11fdf7f2 | 4381 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4382 | if (o->onode.has_omap()) { |
4383 | string key; | |
9f95a23c | 4384 | o->get_omap_key(to, &key); |
31f18b77 FG |
4385 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " |
4386 | << pretty_binary_string(key) << dendl; | |
4387 | it->lower_bound(key); | |
4388 | } else { | |
4389 | it = KeyValueDB::Iterator(); | |
4390 | } | |
11fdf7f2 | 4391 | c->store->log_latency_fn( |
494da23a | 4392 | __func__, |
11fdf7f2 TL |
4393 | l_bluestore_omap_lower_bound_lat, |
4394 | mono_clock::now() - start1, | |
494da23a | 4395 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4396 | [&] (const ceph::timespan& lat) { |
494da23a | 4397 | return ", to = " + to + |
11fdf7f2 TL |
4398 | _stringify(); |
4399 | } | |
4400 | ); | |
31f18b77 FG |
4401 | return 0; |
4402 | } | |
4403 | ||
4404 | bool BlueStore::OmapIteratorImpl::valid() | |
4405 | { | |
9f95a23c | 4406 | std::shared_lock l(c->lock); |
31f18b77 | 4407 | bool r = o->onode.has_omap() && it && it->valid() && |
494da23a | 4408 | it->raw_key().second < tail; |
31f18b77 FG |
4409 | if (it && it->valid()) { |
4410 | ldout(c->store->cct,20) << __func__ << " is at " | |
4411 | << pretty_binary_string(it->raw_key().second) | |
4412 | << dendl; | |
4413 | } | |
4414 | return r; | |
4415 | } | |
4416 | ||
11fdf7f2 | 4417 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 4418 | { |
11fdf7f2 | 4419 | int r = -1; |
9f95a23c | 4420 | std::shared_lock l(c->lock); |
11fdf7f2 | 4421 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4422 | if (o->onode.has_omap()) { |
4423 | it->next(); | |
11fdf7f2 | 4424 | r = 0; |
31f18b77 | 4425 | } |
494da23a TL |
4426 | c->store->log_latency( |
4427 | __func__, | |
11fdf7f2 TL |
4428 | l_bluestore_omap_next_lat, |
4429 | mono_clock::now() - start1, | |
494da23a | 4430 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 TL |
4431 | |
4432 | return r; | |
31f18b77 FG |
4433 | } |
4434 | ||
4435 | string BlueStore::OmapIteratorImpl::key() | |
4436 | { | |
9f95a23c | 4437 | std::shared_lock l(c->lock); |
11fdf7f2 | 4438 | ceph_assert(it->valid()); |
31f18b77 FG |
4439 | string db_key = it->raw_key().second; |
4440 | string user_key; | |
9f95a23c | 4441 | o->decode_omap_key(db_key, &user_key); |
494da23a | 4442 | |
31f18b77 FG |
4443 | return user_key; |
4444 | } | |
4445 | ||
4446 | bufferlist BlueStore::OmapIteratorImpl::value() | |
4447 | { | |
9f95a23c | 4448 | std::shared_lock l(c->lock); |
11fdf7f2 | 4449 | ceph_assert(it->valid()); |
31f18b77 FG |
4450 | return it->value(); |
4451 | } | |
4452 | ||
4453 | ||
4454 | // ===================================== | |
4455 | ||
7c673cae FG |
4456 | #undef dout_prefix |
4457 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
9f95a23c TL |
4458 | #undef dout_context |
4459 | #define dout_context cct | |
7c673cae FG |
4460 | |
4461 | ||
4462 | static void aio_cb(void *priv, void *priv2) | |
4463 | { | |
4464 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4465 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4466 | c->aio_finish(store); | |
4467 | } | |
4468 | ||
11fdf7f2 TL |
4469 | static void discard_cb(void *priv, void *priv2) |
4470 | { | |
4471 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4472 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4473 | store->handle_discard(*tmp); | |
4474 | } | |
4475 | ||
4476 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4477 | { | |
4478 | dout(10) << __func__ << dendl; | |
f67539c2 TL |
4479 | ceph_assert(shared_alloc.a); |
4480 | shared_alloc.a->release(to_release); | |
11fdf7f2 TL |
4481 | } |
4482 | ||
7c673cae | 4483 | BlueStore::BlueStore(CephContext *cct, const string& path) |
9f95a23c | 4484 | : BlueStore(cct, path, 0) {} |
7c673cae FG |
4485 | |
4486 | BlueStore::BlueStore(CephContext *cct, | |
4487 | const string& path, | |
4488 | uint64_t _min_alloc_size) | |
4489 | : ObjectStore(cct, path), | |
9f95a23c | 4490 | throttle(cct), |
11fdf7f2 | 4491 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4492 | kv_sync_thread(this), |
31f18b77 | 4493 | kv_finalize_thread(this), |
f67539c2 | 4494 | zoned_cleaner_thread(this), |
7c673cae FG |
4495 | min_alloc_size(_min_alloc_size), |
4496 | min_alloc_size_order(ctz(_min_alloc_size)), | |
4497 | mempool_thread(this) | |
4498 | { | |
4499 | _init_logger(); | |
11fdf7f2 | 4500 | cct->_conf.add_observer(this); |
7c673cae | 4501 | set_cache_shards(1); |
7c673cae FG |
4502 | } |
4503 | ||
4504 | BlueStore::~BlueStore() | |
4505 | { | |
11fdf7f2 | 4506 | cct->_conf.remove_observer(this); |
7c673cae | 4507 | _shutdown_logger(); |
11fdf7f2 TL |
4508 | ceph_assert(!mounted); |
4509 | ceph_assert(db == NULL); | |
4510 | ceph_assert(bluefs == NULL); | |
4511 | ceph_assert(fsid_fd < 0); | |
4512 | ceph_assert(path_fd < 0); | |
9f95a23c TL |
4513 | for (auto i : onode_cache_shards) { |
4514 | delete i; | |
4515 | } | |
4516 | for (auto i : buffer_cache_shards) { | |
7c673cae FG |
4517 | delete i; |
4518 | } | |
9f95a23c TL |
4519 | onode_cache_shards.clear(); |
4520 | buffer_cache_shards.clear(); | |
7c673cae FG |
4521 | } |
4522 | ||
4523 | const char **BlueStore::get_tracked_conf_keys() const | |
4524 | { | |
4525 | static const char* KEYS[] = { | |
4526 | "bluestore_csum_type", | |
4527 | "bluestore_compression_mode", | |
4528 | "bluestore_compression_algorithm", | |
4529 | "bluestore_compression_min_blob_size", | |
4530 | "bluestore_compression_min_blob_size_ssd", | |
4531 | "bluestore_compression_min_blob_size_hdd", | |
4532 | "bluestore_compression_max_blob_size", | |
4533 | "bluestore_compression_max_blob_size_ssd", | |
4534 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4535 | "bluestore_compression_required_ratio", |
7c673cae FG |
4536 | "bluestore_max_alloc_size", |
4537 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4538 | "bluestore_prefer_deferred_size_hdd", |
4539 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4540 | "bluestore_deferred_batch_ops", |
4541 | "bluestore_deferred_batch_ops_hdd", | |
4542 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4543 | "bluestore_throttle_bytes", |
4544 | "bluestore_throttle_deferred_bytes", | |
4545 | "bluestore_throttle_cost_per_io_hdd", | |
4546 | "bluestore_throttle_cost_per_io_ssd", | |
4547 | "bluestore_throttle_cost_per_io", | |
4548 | "bluestore_max_blob_size", | |
4549 | "bluestore_max_blob_size_ssd", | |
4550 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4551 | "osd_memory_target", |
4552 | "osd_memory_target_cgroup_limit_ratio", | |
4553 | "osd_memory_base", | |
4554 | "osd_memory_cache_min", | |
92f5a8d4 | 4555 | "osd_memory_expected_fragmentation", |
11fdf7f2 TL |
4556 | "bluestore_cache_autotune", |
4557 | "bluestore_cache_autotune_interval", | |
81eedcae | 4558 | "bluestore_warn_on_legacy_statfs", |
9f95a23c TL |
4559 | "bluestore_warn_on_no_per_pool_omap", |
4560 | "bluestore_max_defer_interval", | |
7c673cae FG |
4561 | NULL |
4562 | }; | |
4563 | return KEYS; | |
4564 | } | |
4565 | ||
11fdf7f2 | 4566 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4567 | const std::set<std::string> &changed) |
4568 | { | |
eafe8130 | 4569 | if (changed.count("bluestore_warn_on_legacy_statfs")) { |
81eedcae TL |
4570 | _check_legacy_statfs_alert(); |
4571 | } | |
f67539c2 TL |
4572 | if (changed.count("bluestore_warn_on_no_per_pool_omap") || |
4573 | changed.count("bluestore_warn_on_no_per_pg_omap")) { | |
4574 | _check_no_per_pg_or_pool_omap_alert(); | |
9f95a23c | 4575 | } |
81eedcae | 4576 | |
7c673cae FG |
4577 | if (changed.count("bluestore_csum_type")) { |
4578 | _set_csum(); | |
4579 | } | |
4580 | if (changed.count("bluestore_compression_mode") || | |
4581 | changed.count("bluestore_compression_algorithm") || | |
4582 | changed.count("bluestore_compression_min_blob_size") || | |
4583 | changed.count("bluestore_compression_max_blob_size")) { | |
4584 | if (bdev) { | |
4585 | _set_compression(); | |
4586 | } | |
4587 | } | |
4588 | if (changed.count("bluestore_max_blob_size") || | |
4589 | changed.count("bluestore_max_blob_size_ssd") || | |
4590 | changed.count("bluestore_max_blob_size_hdd")) { | |
4591 | if (bdev) { | |
4592 | // only after startup | |
4593 | _set_blob_size(); | |
4594 | } | |
4595 | } | |
4596 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4597 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4598 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4599 | changed.count("bluestore_max_alloc_size") || |
4600 | changed.count("bluestore_deferred_batch_ops") || | |
4601 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4602 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4603 | if (bdev) { | |
4604 | // only after startup | |
4605 | _set_alloc_sizes(); | |
4606 | } | |
4607 | } | |
4608 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4609 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4610 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4611 | if (bdev) { | |
4612 | _set_throttle_params(); | |
4613 | } | |
4614 | } | |
9f95a23c TL |
4615 | if (changed.count("bluestore_throttle_bytes") || |
4616 | changed.count("bluestore_throttle_deferred_bytes") || | |
4617 | changed.count("bluestore_throttle_trace_rate")) { | |
4618 | throttle.reset_throttle(conf); | |
7c673cae | 4619 | } |
9f95a23c TL |
4620 | if (changed.count("bluestore_max_defer_interval")) { |
4621 | if (bdev) { | |
4622 | _set_max_defer_interval(); | |
4623 | } | |
7c673cae | 4624 | } |
92f5a8d4 TL |
4625 | if (changed.count("osd_memory_target") || |
4626 | changed.count("osd_memory_base") || | |
4627 | changed.count("osd_memory_cache_min") || | |
4628 | changed.count("osd_memory_expected_fragmentation")) { | |
4629 | _update_osd_memory_options(); | |
4630 | } | |
7c673cae FG |
4631 | } |
4632 | ||
4633 | void BlueStore::_set_compression() | |
4634 | { | |
224ce89b WB |
4635 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4636 | if (m) { | |
11fdf7f2 | 4637 | _clear_compression_alert(); |
224ce89b WB |
4638 | comp_mode = *m; |
4639 | } else { | |
4640 | derr << __func__ << " unrecognized value '" | |
4641 | << cct->_conf->bluestore_compression_mode | |
4642 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4643 | << dendl; | |
4644 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4645 | string s("unknown mode: "); |
4646 | s += cct->_conf->bluestore_compression_mode; | |
4647 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4648 | } |
4649 | ||
4650 | compressor = nullptr; | |
4651 | ||
3efd9988 FG |
4652 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4653 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4654 | } else { |
11fdf7f2 | 4655 | ceph_assert(bdev); |
9f95a23c | 4656 | if (_use_rotational_settings()) { |
7c673cae FG |
4657 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; |
4658 | } else { | |
4659 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4660 | } | |
4661 | } | |
4662 | ||
4663 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4664 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4665 | } else { | |
11fdf7f2 | 4666 | ceph_assert(bdev); |
9f95a23c | 4667 | if (_use_rotational_settings()) { |
7c673cae FG |
4668 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; |
4669 | } else { | |
4670 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4671 | } | |
4672 | } | |
4673 | ||
7c673cae FG |
4674 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4675 | if (!alg_name.empty()) { | |
4676 | compressor = Compressor::create(cct, alg_name); | |
4677 | if (!compressor) { | |
4678 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4679 | << dendl; | |
11fdf7f2 | 4680 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4681 | } |
4682 | } | |
4683 | ||
4684 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4685 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4686 | << " min_blob " << comp_min_blob_size |
4687 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4688 | << dendl; |
4689 | } | |
4690 | ||
4691 | void BlueStore::_set_csum() | |
4692 | { | |
4693 | csum_type = Checksummer::CSUM_NONE; | |
4694 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4695 | if (t > Checksummer::CSUM_NONE) | |
4696 | csum_type = t; | |
4697 | ||
4698 | dout(10) << __func__ << " csum_type " | |
4699 | << Checksummer::get_csum_type_string(csum_type) | |
4700 | << dendl; | |
4701 | } | |
4702 | ||
4703 | void BlueStore::_set_throttle_params() | |
4704 | { | |
4705 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4706 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4707 | } else { | |
11fdf7f2 | 4708 | ceph_assert(bdev); |
9f95a23c | 4709 | if (_use_rotational_settings()) { |
7c673cae FG |
4710 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; |
4711 | } else { | |
4712 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4713 | } | |
4714 | } | |
4715 | ||
4716 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4717 | << dendl; | |
4718 | } | |
4719 | void BlueStore::_set_blob_size() | |
4720 | { | |
4721 | if (cct->_conf->bluestore_max_blob_size) { | |
4722 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4723 | } else { | |
11fdf7f2 | 4724 | ceph_assert(bdev); |
9f95a23c | 4725 | if (_use_rotational_settings()) { |
7c673cae FG |
4726 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; |
4727 | } else { | |
4728 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4729 | } | |
4730 | } | |
4731 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4732 | << std::dec << dendl; | |
4733 | } | |
4734 | ||
92f5a8d4 TL |
4735 | void BlueStore::_update_osd_memory_options() |
4736 | { | |
4737 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4738 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
4739 | osd_memory_expected_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
4740 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
4741 | config_changed++; | |
4742 | dout(10) << __func__ | |
4743 | << " osd_memory_target " << osd_memory_target | |
4744 | << " osd_memory_base " << osd_memory_base | |
4745 | << " osd_memory_expected_fragmentation " << osd_memory_expected_fragmentation | |
4746 | << " osd_memory_cache_min " << osd_memory_cache_min | |
4747 | << dendl; | |
4748 | } | |
4749 | ||
11fdf7f2 | 4750 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4751 | { |
11fdf7f2 TL |
4752 | ceph_assert(bdev); |
4753 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4754 | cache_autotune_interval = |
11fdf7f2 TL |
4755 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
4756 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4757 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4758 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4759 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4760 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4761 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4762 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4763 | |
224ce89b WB |
4764 | if (cct->_conf->bluestore_cache_size) { |
4765 | cache_size = cct->_conf->bluestore_cache_size; | |
4766 | } else { | |
4767 | // choose global cache size based on backend type | |
9f95a23c | 4768 | if (_use_rotational_settings()) { |
224ce89b WB |
4769 | cache_size = cct->_conf->bluestore_cache_size_hdd; |
4770 | } else { | |
4771 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4772 | } | |
4773 | } | |
31f18b77 | 4774 | |
f67539c2 | 4775 | cache_meta_ratio = cct->_conf.get_val<double>("bluestore_cache_meta_ratio"); |
224ce89b | 4776 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4777 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4778 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4779 | return -EINVAL; |
4780 | } | |
91327a77 | 4781 | |
f67539c2 | 4782 | cache_kv_ratio = cct->_conf.get_val<double>("bluestore_cache_kv_ratio"); |
224ce89b | 4783 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4784 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4785 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4786 | return -EINVAL; |
4787 | } | |
91327a77 | 4788 | |
f67539c2 TL |
4789 | cache_kv_onode_ratio = cct->_conf.get_val<double>("bluestore_cache_kv_onode_ratio"); |
4790 | if (cache_kv_onode_ratio < 0 || cache_kv_onode_ratio > 1.0) { | |
4791 | derr << __func__ << " bluestore_cache_kv_onode_ratio (" << cache_kv_onode_ratio | |
4792 | << ") must be in range [0,1.0]" << dendl; | |
4793 | return -EINVAL; | |
4794 | } | |
4795 | ||
31f18b77 | 4796 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4797 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4798 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4799 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4800 | << dendl; | |
31f18b77 FG |
4801 | return -EINVAL; |
4802 | } | |
91327a77 | 4803 | |
f67539c2 TL |
4804 | cache_data_ratio = (double)1.0 - |
4805 | (double)cache_meta_ratio - | |
4806 | (double)cache_kv_ratio - | |
4807 | (double)cache_kv_onode_ratio; | |
31f18b77 FG |
4808 | if (cache_data_ratio < 0) { |
4809 | // deal with floating point imprecision | |
4810 | cache_data_ratio = 0; | |
4811 | } | |
91327a77 | 4812 | |
224ce89b WB |
4813 | dout(1) << __func__ << " cache_size " << cache_size |
4814 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4815 | << " kv " << cache_kv_ratio |
4816 | << " data " << cache_data_ratio | |
4817 | << dendl; | |
4818 | return 0; | |
4819 | } | |
4820 | ||
3efd9988 FG |
4821 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4822 | { | |
4823 | bluestore_bdev_label_t label; | |
4824 | string p = path + "/block"; | |
4825 | int r = _read_bdev_label(cct, p, &label); | |
4826 | if (r < 0) { | |
4827 | return ObjectStore::write_meta(key, value); | |
4828 | } | |
4829 | label.meta[key] = value; | |
4830 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4831 | ceph_assert(r == 0); |
3efd9988 FG |
4832 | return ObjectStore::write_meta(key, value); |
4833 | } | |
4834 | ||
4835 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4836 | { | |
4837 | bluestore_bdev_label_t label; | |
4838 | string p = path + "/block"; | |
4839 | int r = _read_bdev_label(cct, p, &label); | |
4840 | if (r < 0) { | |
4841 | return ObjectStore::read_meta(key, value); | |
4842 | } | |
4843 | auto i = label.meta.find(key); | |
4844 | if (i == label.meta.end()) { | |
4845 | return ObjectStore::read_meta(key, value); | |
4846 | } | |
4847 | *value = i->second; | |
4848 | return 0; | |
4849 | } | |
4850 | ||
7c673cae FG |
4851 | void BlueStore::_init_logger() |
4852 | { | |
4853 | PerfCountersBuilder b(cct, "bluestore", | |
4854 | l_bluestore_first, l_bluestore_last); | |
4855 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
4856 | "Average kv_thread flush latency", | |
4857 | "fl_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4858 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
4859 | "Average kv_thread commit latency"); | |
11fdf7f2 TL |
4860 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", |
4861 | "Average kv_sync thread latency", | |
4862 | "ks_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4863 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
4864 | "Average kv_finalize thread latency", | |
4865 | "kf_l", PerfCountersBuilder::PRIO_INTERESTING); | |
7c673cae FG |
4866 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
4867 | "Average prepare state latency"); | |
4868 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", | |
4869 | "Average aio_wait state latency", | |
4870 | "io_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4871 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", | |
4872 | "Average io_done state latency"); | |
4873 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", | |
4874 | "Average kv_queued state latency"); | |
4875 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", | |
4876 | "Average kv_commiting state latency"); | |
4877 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", | |
4878 | "Average kv_done state latency"); | |
4879 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", | |
4880 | "Average deferred_queued state latency"); | |
4881 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", | |
4882 | "Average aio_wait state latency"); | |
4883 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", | |
4884 | "Average cleanup state latency"); | |
4885 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
4886 | "Average finishing state latency"); | |
4887 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
4888 | "Average done state latency"); | |
4889 | b.add_time_avg(l_bluestore_throttle_lat, "throttle_lat", | |
4890 | "Average submit throttle latency", | |
4891 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4892 | b.add_time_avg(l_bluestore_submit_lat, "submit_lat", | |
4893 | "Average submit latency", | |
4894 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4895 | b.add_time_avg(l_bluestore_commit_lat, "commit_lat", | |
4896 | "Average commit latency", | |
4897 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4898 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
4899 | "Average read latency", | |
4900 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4901 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", | |
4902 | "Average read onode metadata latency"); | |
4903 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", | |
4904 | "Average read latency"); | |
4905 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
4906 | "Average compress latency"); | |
4907 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
4908 | "Average decompress latency"); | |
4909 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", | |
4910 | "Average checksum latency"); | |
4911 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
4912 | "Sum for beneficial compress ops"); | |
4913 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
4914 | "Sum for compress ops rejected due to low net gain of space"); | |
4915 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", | |
11fdf7f2 | 4916 | "Sum for write-op padded bytes", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4917 | b.add_u64_counter(l_bluestore_deferred_write_ops, "deferred_write_ops", |
4918 | "Sum for deferred write op"); | |
4919 | b.add_u64_counter(l_bluestore_deferred_write_bytes, "deferred_write_bytes", | |
11fdf7f2 | 4920 | "Sum for deferred write bytes", "def", 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4921 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", |
4922 | "Sum for write penalty read ops"); | |
4923 | b.add_u64(l_bluestore_allocated, "bluestore_allocated", | |
4924 | "Sum for allocated bytes"); | |
4925 | b.add_u64(l_bluestore_stored, "bluestore_stored", | |
4926 | "Sum for stored bytes"); | |
4927 | b.add_u64(l_bluestore_compressed, "bluestore_compressed", | |
92f5a8d4 TL |
4928 | "Sum for stored compressed bytes", |
4929 | "c", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4930 | b.add_u64(l_bluestore_compressed_allocated, "bluestore_compressed_allocated", |
92f5a8d4 TL |
4931 | "Sum for bytes allocated for compressed data", |
4932 | "c_a", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4933 | b.add_u64(l_bluestore_compressed_original, "bluestore_compressed_original", |
92f5a8d4 TL |
4934 | "Sum for original bytes that were compressed", |
4935 | "c_o", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae FG |
4936 | b.add_u64(l_bluestore_onodes, "bluestore_onodes", |
4937 | "Number of onodes in cache"); | |
9f95a23c TL |
4938 | b.add_u64(l_bluestore_pinned_onodes, "bluestore_pinned_onodes", |
4939 | "Number of pinned onodes in cache"); | |
7c673cae FG |
4940 | b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits", |
4941 | "Sum for onode-lookups hit in the cache"); | |
4942 | b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses", | |
4943 | "Sum for onode-lookups missed in the cache"); | |
4944 | b.add_u64_counter(l_bluestore_onode_shard_hits, "bluestore_onode_shard_hits", | |
4945 | "Sum for onode-shard lookups hit in the cache"); | |
4946 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
4947 | "bluestore_onode_shard_misses", | |
4948 | "Sum for onode-shard lookups missed in the cache"); | |
4949 | b.add_u64(l_bluestore_extents, "bluestore_extents", | |
4950 | "Number of extents in cache"); | |
4951 | b.add_u64(l_bluestore_blobs, "bluestore_blobs", | |
4952 | "Number of blobs in cache"); | |
4953 | b.add_u64(l_bluestore_buffers, "bluestore_buffers", | |
4954 | "Number of buffers in cache"); | |
4955 | b.add_u64(l_bluestore_buffer_bytes, "bluestore_buffer_bytes", | |
11fdf7f2 | 4956 | "Number of buffer bytes in cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4957 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "bluestore_buffer_hit_bytes", |
11fdf7f2 | 4958 | "Sum for bytes of read hit in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4959 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "bluestore_buffer_miss_bytes", |
11fdf7f2 | 4960 | "Sum for bytes of read missed in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4961 | |
4962 | b.add_u64_counter(l_bluestore_write_big, "bluestore_write_big", | |
4963 | "Large aligned writes into fresh blobs"); | |
4964 | b.add_u64_counter(l_bluestore_write_big_bytes, "bluestore_write_big_bytes", | |
11fdf7f2 | 4965 | "Large aligned writes into fresh blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4966 | b.add_u64_counter(l_bluestore_write_big_blobs, "bluestore_write_big_blobs", |
4967 | "Large aligned writes into fresh blobs (blobs)"); | |
f67539c2 TL |
4968 | b.add_u64_counter(l_bluestore_write_big_deferred, |
4969 | "bluestore_write_big_deferred", | |
4970 | "Big overwrites using deferred"); | |
7c673cae FG |
4971 | b.add_u64_counter(l_bluestore_write_small, "bluestore_write_small", |
4972 | "Small writes into existing or sparse small blobs"); | |
4973 | b.add_u64_counter(l_bluestore_write_small_bytes, "bluestore_write_small_bytes", | |
11fdf7f2 | 4974 | "Small writes into existing or sparse small blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4975 | b.add_u64_counter(l_bluestore_write_small_unused, |
4976 | "bluestore_write_small_unused", | |
4977 | "Small writes into unused portion of existing blob"); | |
f67539c2 TL |
4978 | b.add_u64_counter(l_bluestore_write_deferred, |
4979 | "bluestore_write_deferred", | |
4980 | "Overwrites using deferred"); | |
7c673cae FG |
4981 | b.add_u64_counter(l_bluestore_write_small_pre_read, |
4982 | "bluestore_write_small_pre_read", | |
4983 | "Small writes that required we read some data (possibly " | |
4984 | "cached) to fill out the block"); | |
f67539c2 TL |
4985 | b.add_u64_counter(l_bluestore_write_new, "bluestore_write_new", |
4986 | "Write into new blob"); | |
7c673cae FG |
4987 | |
4988 | b.add_u64_counter(l_bluestore_txc, "bluestore_txc", "Transactions committed"); | |
4989 | b.add_u64_counter(l_bluestore_onode_reshard, "bluestore_onode_reshard", | |
4990 | "Onode extent map reshard events"); | |
4991 | b.add_u64_counter(l_bluestore_blob_split, "bluestore_blob_split", | |
4992 | "Sum for blob splitting due to resharding"); | |
4993 | b.add_u64_counter(l_bluestore_extent_compress, "bluestore_extent_compress", | |
4994 | "Sum for extents that have been removed due to compression"); | |
4995 | b.add_u64_counter(l_bluestore_gc_merged, "bluestore_gc_merged", | |
4996 | "Sum for extents that have been merged due to garbage " | |
4997 | "collection"); | |
b32b8144 FG |
4998 | b.add_u64_counter(l_bluestore_read_eio, "bluestore_read_eio", |
4999 | "Read EIO errors propagated to high level callers"); | |
f64942e4 AA |
5000 | b.add_u64_counter(l_bluestore_reads_with_retries, "bluestore_reads_with_retries", |
5001 | "Read operations that required at least one retry due to failed checksum validation"); | |
a8e16298 TL |
5002 | b.add_u64(l_bluestore_fragmentation, "bluestore_fragmentation_micros", |
5003 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
11fdf7f2 TL |
5004 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
5005 | "Average omap iterator seek_to_first call latency"); | |
5006 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", | |
5007 | "Average omap iterator upper_bound call latency"); | |
5008 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", | |
5009 | "Average omap iterator lower_bound call latency"); | |
5010 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", | |
5011 | "Average omap iterator next call latency"); | |
adb31ebb TL |
5012 | b.add_time_avg(l_bluestore_omap_get_keys_lat, "omap_get_keys_lat", |
5013 | "Average omap get_keys call latency"); | |
5014 | b.add_time_avg(l_bluestore_omap_get_values_lat, "omap_get_values_lat", | |
5015 | "Average omap get_values call latency"); | |
494da23a TL |
5016 | b.add_time_avg(l_bluestore_clist_lat, "clist_lat", |
5017 | "Average collection listing latency"); | |
adb31ebb TL |
5018 | b.add_time_avg(l_bluestore_remove_lat, "remove_lat", |
5019 | "Average removal latency"); | |
5020 | ||
7c673cae FG |
5021 | logger = b.create_perf_counters(); |
5022 | cct->get_perfcounters_collection()->add(logger); | |
5023 | } | |
5024 | ||
5025 | int BlueStore::_reload_logger() | |
5026 | { | |
5027 | struct store_statfs_t store_statfs; | |
7c673cae | 5028 | int r = statfs(&store_statfs); |
11fdf7f2 | 5029 | if (r >= 0) { |
7c673cae | 5030 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
5031 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
5032 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
5033 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
5034 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
5035 | } |
5036 | return r; | |
5037 | } | |
5038 | ||
5039 | void BlueStore::_shutdown_logger() | |
5040 | { | |
5041 | cct->get_perfcounters_collection()->remove(logger); | |
5042 | delete logger; | |
5043 | } | |
5044 | ||
5045 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
5046 | uuid_d *fsid) | |
5047 | { | |
5048 | bluestore_bdev_label_t label; | |
5049 | int r = _read_bdev_label(cct, path, &label); | |
5050 | if (r < 0) | |
5051 | return r; | |
5052 | *fsid = label.osd_uuid; | |
5053 | return 0; | |
5054 | } | |
5055 | ||
5056 | int BlueStore::_open_path() | |
5057 | { | |
b32b8144 | 5058 | // sanity check(s) |
11fdf7f2 | 5059 | ceph_assert(path_fd < 0); |
91327a77 | 5060 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
5061 | if (path_fd < 0) { |
5062 | int r = -errno; | |
5063 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
5064 | << dendl; | |
5065 | return r; | |
5066 | } | |
5067 | return 0; | |
5068 | } | |
5069 | ||
5070 | void BlueStore::_close_path() | |
5071 | { | |
5072 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
5073 | path_fd = -1; | |
5074 | } | |
5075 | ||
3efd9988 FG |
5076 | int BlueStore::_write_bdev_label(CephContext *cct, |
5077 | string path, bluestore_bdev_label_t label) | |
7c673cae FG |
5078 | { |
5079 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
5080 | bufferlist bl; | |
11fdf7f2 | 5081 | encode(label, bl); |
7c673cae | 5082 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
5083 | encode(crc, bl); |
5084 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
5085 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
5086 | z.zero(); | |
5087 | bl.append(std::move(z)); | |
5088 | ||
91327a77 | 5089 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC)); |
7c673cae FG |
5090 | if (fd < 0) { |
5091 | fd = -errno; | |
5092 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5093 | << dendl; | |
5094 | return fd; | |
5095 | } | |
5096 | int r = bl.write_fd(fd); | |
5097 | if (r < 0) { | |
5098 | derr << __func__ << " failed to write to " << path | |
5099 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 5100 | goto out; |
7c673cae | 5101 | } |
3efd9988 FG |
5102 | r = ::fsync(fd); |
5103 | if (r < 0) { | |
5104 | derr << __func__ << " failed to fsync " << path | |
5105 | << ": " << cpp_strerror(r) << dendl; | |
5106 | } | |
11fdf7f2 | 5107 | out: |
7c673cae FG |
5108 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
5109 | return r; | |
5110 | } | |
5111 | ||
5112 | int BlueStore::_read_bdev_label(CephContext* cct, string path, | |
5113 | bluestore_bdev_label_t *label) | |
5114 | { | |
5115 | dout(10) << __func__ << dendl; | |
91327a77 | 5116 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
5117 | if (fd < 0) { |
5118 | fd = -errno; | |
5119 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5120 | << dendl; | |
5121 | return fd; | |
5122 | } | |
5123 | bufferlist bl; | |
5124 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
5125 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5126 | if (r < 0) { | |
5127 | derr << __func__ << " failed to read from " << path | |
5128 | << ": " << cpp_strerror(r) << dendl; | |
5129 | return r; | |
5130 | } | |
5131 | ||
5132 | uint32_t crc, expected_crc; | |
11fdf7f2 | 5133 | auto p = bl.cbegin(); |
7c673cae | 5134 | try { |
11fdf7f2 | 5135 | decode(*label, p); |
7c673cae FG |
5136 | bufferlist t; |
5137 | t.substr_of(bl, 0, p.get_off()); | |
5138 | crc = t.crc32c(-1); | |
11fdf7f2 | 5139 | decode(expected_crc, p); |
7c673cae | 5140 | } |
f67539c2 | 5141 | catch (ceph::buffer::error& e) { |
b32b8144 | 5142 | dout(2) << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
5143 | << ": " << e.what() |
5144 | << dendl; | |
b32b8144 | 5145 | return -ENOENT; |
7c673cae FG |
5146 | } |
5147 | if (crc != expected_crc) { | |
5148 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
5149 | << " != actual " << crc << dendl; | |
5150 | return -EIO; | |
5151 | } | |
5152 | dout(10) << __func__ << " got " << *label << dendl; | |
5153 | return 0; | |
5154 | } | |
5155 | ||
5156 | int BlueStore::_check_or_set_bdev_label( | |
5157 | string path, uint64_t size, string desc, bool create) | |
5158 | { | |
5159 | bluestore_bdev_label_t label; | |
5160 | if (create) { | |
5161 | label.osd_uuid = fsid; | |
5162 | label.size = size; | |
5163 | label.btime = ceph_clock_now(); | |
5164 | label.description = desc; | |
3efd9988 | 5165 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
5166 | if (r < 0) |
5167 | return r; | |
5168 | } else { | |
5169 | int r = _read_bdev_label(cct, path, &label); | |
5170 | if (r < 0) | |
5171 | return r; | |
31f18b77 FG |
5172 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
5173 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
5174 | << " and fsid " << fsid << " check bypassed" << dendl; | |
1911f103 | 5175 | } else if (label.osd_uuid != fsid) { |
7c673cae FG |
5176 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
5177 | << " does not match our fsid " << fsid << dendl; | |
5178 | return -EIO; | |
5179 | } | |
5180 | } | |
5181 | return 0; | |
5182 | } | |
5183 | ||
5184 | void BlueStore::_set_alloc_sizes(void) | |
5185 | { | |
7c673cae FG |
5186 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
5187 | ||
5188 | if (cct->_conf->bluestore_prefer_deferred_size) { | |
5189 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
5190 | } else { | |
11fdf7f2 | 5191 | ceph_assert(bdev); |
9f95a23c | 5192 | if (_use_rotational_settings()) { |
7c673cae FG |
5193 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; |
5194 | } else { | |
5195 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
5196 | } | |
5197 | } | |
5198 | ||
5199 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
5200 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
5201 | } else { | |
11fdf7f2 | 5202 | ceph_assert(bdev); |
9f95a23c | 5203 | if (_use_rotational_settings()) { |
7c673cae FG |
5204 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; |
5205 | } else { | |
5206 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
5207 | } | |
5208 | } | |
5209 | ||
5210 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 5211 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
5212 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
5213 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
5214 | << std::dec | |
5215 | << " deferred_batch_ops " << deferred_batch_ops | |
5216 | << dendl; | |
5217 | } | |
5218 | ||
5219 | int BlueStore::_open_bdev(bool create) | |
5220 | { | |
11fdf7f2 | 5221 | ceph_assert(bdev == NULL); |
7c673cae | 5222 | string p = path + "/block"; |
11fdf7f2 | 5223 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
5224 | int r = bdev->open(p); |
5225 | if (r < 0) | |
5226 | goto fail; | |
5227 | ||
11fdf7f2 TL |
5228 | if (create && cct->_conf->bdev_enable_discard) { |
5229 | bdev->discard(0, bdev->get_size()); | |
5230 | } | |
5231 | ||
7c673cae FG |
5232 | if (bdev->supported_bdev_label()) { |
5233 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
5234 | if (r < 0) | |
5235 | goto fail_close; | |
5236 | } | |
5237 | ||
5238 | // initialize global block parameters | |
5239 | block_size = bdev->get_block_size(); | |
5240 | block_mask = ~(block_size - 1); | |
5241 | block_size_order = ctz(block_size); | |
11fdf7f2 | 5242 | ceph_assert(block_size == 1u << block_size_order); |
9f95a23c | 5243 | _set_max_defer_interval(); |
224ce89b WB |
5244 | // and set cache_size based on device type |
5245 | r = _set_cache_sizes(); | |
5246 | if (r < 0) { | |
5247 | goto fail_close; | |
5248 | } | |
f67539c2 TL |
5249 | |
5250 | if (bdev->is_smr()) { | |
5251 | freelist_type = "zoned"; | |
5252 | } | |
7c673cae FG |
5253 | return 0; |
5254 | ||
5255 | fail_close: | |
5256 | bdev->close(); | |
5257 | fail: | |
5258 | delete bdev; | |
5259 | bdev = NULL; | |
5260 | return r; | |
5261 | } | |
5262 | ||
11fdf7f2 TL |
5263 | void BlueStore::_validate_bdev() |
5264 | { | |
5265 | ceph_assert(bdev); | |
11fdf7f2 | 5266 | uint64_t dev_size = bdev->get_size(); |
f67539c2 | 5267 | ceph_assert(dev_size > _get_ondisk_reserved()); |
11fdf7f2 TL |
5268 | } |
5269 | ||
7c673cae FG |
5270 | void BlueStore::_close_bdev() |
5271 | { | |
11fdf7f2 | 5272 | ceph_assert(bdev); |
7c673cae FG |
5273 | bdev->close(); |
5274 | delete bdev; | |
5275 | bdev = NULL; | |
5276 | } | |
5277 | ||
1911f103 | 5278 | int BlueStore::_open_fm(KeyValueDB::Transaction t, bool read_only) |
7c673cae | 5279 | { |
1911f103 | 5280 | int r; |
1911f103 | 5281 | |
11fdf7f2 TL |
5282 | ceph_assert(fm == NULL); |
5283 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); | |
5284 | ceph_assert(fm); | |
5285 | if (t) { | |
5286 | // create mode. initialize freespace | |
7c673cae | 5287 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
5288 | { |
5289 | bufferlist bl; | |
5290 | bl.append(freelist_type); | |
5291 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
5292 | } | |
b32b8144 FG |
5293 | // being able to allocate in units less than bdev block size |
5294 | // seems to be a bad idea. | |
11fdf7f2 | 5295 | ceph_assert( cct->_conf->bdev_block_size <= (int64_t)min_alloc_size); |
f67539c2 TL |
5296 | |
5297 | uint64_t alloc_size = min_alloc_size; | |
5298 | if (bdev->is_smr()) { | |
5299 | alloc_size = _zoned_piggyback_device_parameters_onto(alloc_size); | |
5300 | } | |
5301 | ||
5302 | fm->create(bdev->get_size(), alloc_size, t); | |
7c673cae FG |
5303 | |
5304 | // allocate superblock reserved space. note that we do not mark | |
5305 | // bluefs space as allocated in the freelist; we instead rely on | |
f67539c2 | 5306 | // bluefs doing that itself. |
11fdf7f2 | 5307 | auto reserved = _get_ondisk_reserved(); |
3efd9988 | 5308 | fm->allocate(0, reserved, t); |
7c673cae | 5309 | |
7c673cae FG |
5310 | if (cct->_conf->bluestore_debug_prefill > 0) { |
5311 | uint64_t end = bdev->get_size() - reserved; | |
5312 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
5313 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
5314 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 5315 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
5316 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
5317 | float r = cct->_conf->bluestore_debug_prefill; | |
5318 | r /= 1.0 - r; | |
5319 | bool stop = false; | |
5320 | ||
5321 | while (!stop && start < end) { | |
5322 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
5323 | if (start + l > end) { | |
5324 | l = end - start; | |
11fdf7f2 | 5325 | l = p2align(l, min_alloc_size); |
7c673cae | 5326 | } |
11fdf7f2 | 5327 | ceph_assert(start + l <= end); |
7c673cae FG |
5328 | |
5329 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 5330 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
5331 | if (start + l + u > end) { |
5332 | u = end - (start + l); | |
5333 | // trim to align so we don't overflow again | |
11fdf7f2 | 5334 | u = p2align(u, min_alloc_size); |
7c673cae FG |
5335 | stop = true; |
5336 | } | |
11fdf7f2 | 5337 | ceph_assert(start + l + u <= end); |
7c673cae | 5338 | |
11fdf7f2 | 5339 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
5340 | << " use 0x" << u << std::dec << dendl; |
5341 | ||
5342 | if (u == 0) { | |
5343 | // break if u has been trimmed to nothing | |
5344 | break; | |
5345 | } | |
5346 | ||
5347 | fm->allocate(start + l, u, t); | |
5348 | start += l + u; | |
5349 | } | |
5350 | } | |
f67539c2 | 5351 | r = _write_out_fm_meta(0); |
1911f103 TL |
5352 | ceph_assert(r == 0); |
5353 | } else { | |
f67539c2 TL |
5354 | r = fm->init(db, read_only, |
5355 | [&](const std::string& key, std::string* result) { | |
5356 | return read_meta(key, result); | |
5357 | }); | |
1911f103 | 5358 | if (r < 0) { |
f67539c2 | 5359 | derr << __func__ << " freelist init failed: " << cpp_strerror(r) << dendl; |
1911f103 TL |
5360 | delete fm; |
5361 | fm = NULL; | |
5362 | return r; | |
5363 | } | |
7c673cae | 5364 | } |
81eedcae TL |
5365 | // if space size tracked by free list manager is that higher than actual |
5366 | // dev size one can hit out-of-space allocation which will result | |
5367 | // in data loss and/or assertions | |
5368 | // Probably user altered the device size somehow. | |
5369 | // The only fix for now is to redeploy OSD. | |
5370 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
5371 | ostringstream ss; | |
5372 | ss << "slow device size mismatch detected, " | |
5373 | << " fm size(" << fm->get_size() | |
5374 | << ") > slow device size(" << bdev->get_size() | |
5375 | << "), Please stop using this OSD as it might cause data loss."; | |
5376 | _set_disk_size_mismatch_alert(ss.str()); | |
5377 | } | |
7c673cae FG |
5378 | return 0; |
5379 | } | |
5380 | ||
5381 | void BlueStore::_close_fm() | |
5382 | { | |
5383 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 5384 | ceph_assert(fm); |
7c673cae FG |
5385 | fm->shutdown(); |
5386 | delete fm; | |
5387 | fm = NULL; | |
5388 | } | |
5389 | ||
f67539c2 | 5390 | int BlueStore::_write_out_fm_meta(uint64_t target_size) |
1911f103 | 5391 | { |
f67539c2 | 5392 | int r = 0; |
1911f103 TL |
5393 | string p = path + "/block"; |
5394 | ||
5395 | std::vector<std::pair<string, string>> fm_meta; | |
5396 | fm->get_meta(target_size, &fm_meta); | |
5397 | ||
1911f103 | 5398 | for (auto& m : fm_meta) { |
f67539c2 TL |
5399 | r = write_meta(m.first, m.second); |
5400 | ceph_assert(r == 0); | |
1911f103 | 5401 | } |
1911f103 TL |
5402 | return r; |
5403 | } | |
5404 | ||
f67539c2 | 5405 | int BlueStore::_create_alloc() |
7c673cae | 5406 | { |
f67539c2 | 5407 | ceph_assert(shared_alloc.a == NULL); |
11fdf7f2 TL |
5408 | ceph_assert(bdev->get_size()); |
5409 | ||
f67539c2 TL |
5410 | uint64_t alloc_size = min_alloc_size; |
5411 | if (bdev->is_smr()) { | |
5412 | int r = _zoned_check_config_settings(); | |
5413 | if (r < 0) | |
11fdf7f2 | 5414 | return r; |
f67539c2 | 5415 | alloc_size = _zoned_piggyback_device_parameters_onto(alloc_size); |
11fdf7f2 TL |
5416 | } |
5417 | ||
f67539c2 TL |
5418 | shared_alloc.set(Allocator::create(cct, cct->_conf->bluestore_allocator, |
5419 | bdev->get_size(), | |
5420 | alloc_size, "block")); | |
5421 | ||
5422 | if (!shared_alloc.a) { | |
5423 | lderr(cct) << __func__ << "Failed to create allocator:: " | |
5424 | << cct->_conf->bluestore_allocator | |
5425 | << dendl; | |
7c673cae FG |
5426 | return -EINVAL; |
5427 | } | |
f67539c2 TL |
5428 | return 0; |
5429 | } | |
5430 | ||
5431 | int BlueStore::_init_alloc() | |
5432 | { | |
5433 | int r = _create_alloc(); | |
5434 | if (r < 0) { | |
5435 | return r; | |
5436 | } | |
5437 | ceph_assert(shared_alloc.a != NULL); | |
5438 | ||
5439 | if (bdev->is_smr()) { | |
5440 | shared_alloc.a->zoned_set_zone_states(fm->get_zone_states(db)); | |
5441 | } | |
7c673cae FG |
5442 | |
5443 | uint64_t num = 0, bytes = 0; | |
5444 | ||
5445 | dout(1) << __func__ << " opening allocation metadata" << dendl; | |
5446 | // initialize from freelist | |
5447 | fm->enumerate_reset(); | |
5448 | uint64_t offset, length; | |
11fdf7f2 | 5449 | while (fm->enumerate_next(db, &offset, &length)) { |
f67539c2 | 5450 | shared_alloc.a->init_add_free(offset, length); |
7c673cae FG |
5451 | ++num; |
5452 | bytes += length; | |
5453 | } | |
224ce89b | 5454 | fm->enumerate_reset(); |
7c673cae | 5455 | |
f67539c2 TL |
5456 | dout(1) << __func__ |
5457 | << " loaded " << byte_u_t(bytes) << " in " << num << " extents" | |
5458 | << std::hex | |
5459 | << ", allocator type " << shared_alloc.a->get_type() | |
5460 | << ", capacity 0x" << shared_alloc.a->get_capacity() | |
5461 | << ", block size 0x" << shared_alloc.a->get_block_size() | |
5462 | << ", free 0x" << shared_alloc.a->get_free() | |
5463 | << ", fragmentation " << shared_alloc.a->get_fragmentation() | |
5464 | << std::dec << dendl; | |
1911f103 | 5465 | |
7c673cae FG |
5466 | return 0; |
5467 | } | |
5468 | ||
5469 | void BlueStore::_close_alloc() | |
5470 | { | |
11fdf7f2 TL |
5471 | ceph_assert(bdev); |
5472 | bdev->discard_drain(); | |
5473 | ||
f67539c2 TL |
5474 | ceph_assert(shared_alloc.a); |
5475 | shared_alloc.a->shutdown(); | |
5476 | delete shared_alloc.a; | |
5477 | shared_alloc.reset(); | |
7c673cae FG |
5478 | } |
5479 | ||
5480 | int BlueStore::_open_fsid(bool create) | |
5481 | { | |
11fdf7f2 | 5482 | ceph_assert(fsid_fd < 0); |
91327a77 | 5483 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
5484 | if (create) |
5485 | flags |= O_CREAT; | |
5486 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
5487 | if (fsid_fd < 0) { | |
5488 | int err = -errno; | |
5489 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
5490 | return err; | |
5491 | } | |
5492 | return 0; | |
5493 | } | |
5494 | ||
5495 | int BlueStore::_read_fsid(uuid_d *uuid) | |
5496 | { | |
5497 | char fsid_str[40]; | |
5498 | memset(fsid_str, 0, sizeof(fsid_str)); | |
5499 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
5500 | if (ret < 0) { | |
5501 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
5502 | return ret; | |
5503 | } | |
5504 | if (ret > 36) | |
5505 | fsid_str[36] = 0; | |
5506 | else | |
5507 | fsid_str[ret] = 0; | |
5508 | if (!uuid->parse(fsid_str)) { | |
5509 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
5510 | return -EINVAL; | |
5511 | } | |
5512 | return 0; | |
5513 | } | |
5514 | ||
5515 | int BlueStore::_write_fsid() | |
5516 | { | |
5517 | int r = ::ftruncate(fsid_fd, 0); | |
5518 | if (r < 0) { | |
5519 | r = -errno; | |
5520 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
5521 | return r; | |
5522 | } | |
5523 | string str = stringify(fsid) + "\n"; | |
5524 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
5525 | if (r < 0) { | |
5526 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
5527 | return r; | |
5528 | } | |
5529 | r = ::fsync(fsid_fd); | |
5530 | if (r < 0) { | |
5531 | r = -errno; | |
5532 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
5533 | return r; | |
5534 | } | |
5535 | return 0; | |
5536 | } | |
5537 | ||
5538 | void BlueStore::_close_fsid() | |
5539 | { | |
5540 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
5541 | fsid_fd = -1; | |
5542 | } | |
5543 | ||
5544 | int BlueStore::_lock_fsid() | |
5545 | { | |
5546 | struct flock l; | |
5547 | memset(&l, 0, sizeof(l)); | |
5548 | l.l_type = F_WRLCK; | |
5549 | l.l_whence = SEEK_SET; | |
5550 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
5551 | if (r < 0) { | |
5552 | int err = errno; | |
5553 | derr << __func__ << " failed to lock " << path << "/fsid" | |
5554 | << " (is another ceph-osd still running?)" | |
5555 | << cpp_strerror(err) << dendl; | |
5556 | return -err; | |
5557 | } | |
5558 | return 0; | |
5559 | } | |
5560 | ||
31f18b77 FG |
5561 | bool BlueStore::is_rotational() |
5562 | { | |
5563 | if (bdev) { | |
5564 | return bdev->is_rotational(); | |
5565 | } | |
5566 | ||
5567 | bool rotational = true; | |
5568 | int r = _open_path(); | |
5569 | if (r < 0) | |
5570 | goto out; | |
5571 | r = _open_fsid(false); | |
5572 | if (r < 0) | |
5573 | goto out_path; | |
5574 | r = _read_fsid(&fsid); | |
5575 | if (r < 0) | |
5576 | goto out_fsid; | |
5577 | r = _lock_fsid(); | |
5578 | if (r < 0) | |
5579 | goto out_fsid; | |
5580 | r = _open_bdev(false); | |
5581 | if (r < 0) | |
5582 | goto out_fsid; | |
5583 | rotational = bdev->is_rotational(); | |
5584 | _close_bdev(); | |
5585 | out_fsid: | |
5586 | _close_fsid(); | |
5587 | out_path: | |
5588 | _close_path(); | |
5589 | out: | |
5590 | return rotational; | |
5591 | } | |
5592 | ||
d2e6a577 FG |
5593 | bool BlueStore::is_journal_rotational() |
5594 | { | |
5595 | if (!bluefs) { | |
5596 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
5597 | << dendl; | |
5598 | return is_rotational(); | |
5599 | } | |
5600 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
5601 | return bluefs->wal_is_rotational(); | |
5602 | } | |
5603 | ||
9f95a23c TL |
5604 | bool BlueStore::_use_rotational_settings() |
5605 | { | |
5606 | if (cct->_conf->bluestore_debug_enforce_settings == "hdd") { | |
5607 | return true; | |
5608 | } | |
5609 | if (cct->_conf->bluestore_debug_enforce_settings == "ssd") { | |
5610 | return false; | |
5611 | } | |
5612 | return bdev->is_rotational(); | |
5613 | } | |
5614 | ||
7c673cae FG |
5615 | bool BlueStore::test_mount_in_use() |
5616 | { | |
5617 | // most error conditions mean the mount is not in use (e.g., because | |
5618 | // it doesn't exist). only if we fail to lock do we conclude it is | |
5619 | // in use. | |
5620 | bool ret = false; | |
5621 | int r = _open_path(); | |
5622 | if (r < 0) | |
5623 | return false; | |
5624 | r = _open_fsid(false); | |
5625 | if (r < 0) | |
5626 | goto out_path; | |
5627 | r = _lock_fsid(); | |
5628 | if (r < 0) | |
5629 | ret = true; // if we can't lock, it is in use | |
5630 | _close_fsid(); | |
5631 | out_path: | |
5632 | _close_path(); | |
5633 | return ret; | |
5634 | } | |
5635 | ||
11fdf7f2 | 5636 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
5637 | { |
5638 | int r; | |
11fdf7f2 | 5639 | bluefs = new BlueFS(cct); |
7c673cae | 5640 | |
11fdf7f2 TL |
5641 | string bfn; |
5642 | struct stat st; | |
5643 | ||
5644 | bfn = path + "/block.db"; | |
5645 | if (::stat(bfn.c_str(), &st) == 0) { | |
eafe8130 TL |
5646 | r = bluefs->add_block_device( |
5647 | BlueFS::BDEV_DB, bfn, | |
f67539c2 TL |
5648 | create && cct->_conf->bdev_enable_discard, |
5649 | SUPER_RESERVED); | |
7c673cae | 5650 | if (r < 0) { |
11fdf7f2 TL |
5651 | derr << __func__ << " add block device(" << bfn << ") returned: " |
5652 | << cpp_strerror(r) << dendl; | |
5653 | goto free_bluefs; | |
7c673cae | 5654 | } |
7c673cae | 5655 | |
11fdf7f2 TL |
5656 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
5657 | r = _check_or_set_bdev_label( | |
5658 | bfn, | |
5659 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
5660 | "bluefs db", create); | |
5661 | if (r < 0) { | |
5662 | derr << __func__ | |
5663 | << " check block device(" << bfn << ") label returned: " | |
5664 | << cpp_strerror(r) << dendl; | |
5665 | goto free_bluefs; | |
5666 | } | |
7c673cae | 5667 | } |
9f95a23c TL |
5668 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
5669 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
5670 | } else { |
5671 | r = -errno; | |
5672 | if (::lstat(bfn.c_str(), &st) == -1) { | |
5673 | r = 0; | |
9f95a23c | 5674 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7c673cae | 5675 | } else { |
11fdf7f2 TL |
5676 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
5677 | << cpp_strerror(r) << dendl; | |
5678 | goto free_bluefs; | |
7c673cae FG |
5679 | } |
5680 | } | |
7c673cae | 5681 | |
11fdf7f2 TL |
5682 | // shared device |
5683 | bfn = path + "/block"; | |
5684 | // never trim here | |
9f95a23c | 5685 | r = bluefs->add_block_device(bluefs_layout.shared_bdev, bfn, false, |
f67539c2 TL |
5686 | 0, // no need to provide valid 'reserved' for shared dev |
5687 | &shared_alloc); | |
11fdf7f2 TL |
5688 | if (r < 0) { |
5689 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5690 | << cpp_strerror(r) << dendl; | |
5691 | goto free_bluefs; | |
5692 | } | |
11fdf7f2 TL |
5693 | |
5694 | bfn = path + "/block.wal"; | |
5695 | if (::stat(bfn.c_str(), &st) == 0) { | |
5696 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
f67539c2 TL |
5697 | create && cct->_conf->bdev_enable_discard, |
5698 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
5699 | if (r < 0) { |
5700 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5701 | << cpp_strerror(r) << dendl; | |
5702 | goto free_bluefs; | |
5703 | } | |
7c673cae | 5704 | |
11fdf7f2 TL |
5705 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
5706 | r = _check_or_set_bdev_label( | |
5707 | bfn, | |
5708 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
5709 | "bluefs wal", create); | |
7c673cae | 5710 | if (r < 0) { |
11fdf7f2 TL |
5711 | derr << __func__ << " check block device(" << bfn |
5712 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
5713 | goto free_bluefs; |
5714 | } | |
7c673cae FG |
5715 | } |
5716 | ||
9f95a23c | 5717 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
5718 | } else { |
5719 | r = 0; | |
5720 | if (::lstat(bfn.c_str(), &st) != -1) { | |
5721 | r = -errno; | |
5722 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
5723 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
5724 | goto free_bluefs; |
5725 | } | |
11fdf7f2 TL |
5726 | } |
5727 | return 0; | |
7c673cae | 5728 | |
11fdf7f2 TL |
5729 | free_bluefs: |
5730 | ceph_assert(bluefs); | |
5731 | delete bluefs; | |
5732 | bluefs = NULL; | |
5733 | return r; | |
5734 | } | |
7c673cae | 5735 | |
f67539c2 | 5736 | int BlueStore::_open_bluefs(bool create, bool read_only) |
11fdf7f2 TL |
5737 | { |
5738 | int r = _minimal_open_bluefs(create); | |
5739 | if (r < 0) { | |
5740 | return r; | |
5741 | } | |
f67539c2 | 5742 | BlueFSVolumeSelector* vselector = nullptr; |
9f95a23c TL |
5743 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
5744 | ||
5745 | string options = cct->_conf->bluestore_rocksdb_options; | |
cd265ab1 TL |
5746 | string options_annex = cct->_conf->bluestore_rocksdb_options_annex; |
5747 | if (!options_annex.empty()) { | |
5748 | if (!options.empty() && | |
5749 | *options.rbegin() != ',') { | |
5750 | options += ','; | |
5751 | } | |
5752 | options += options_annex; | |
5753 | } | |
9f95a23c TL |
5754 | |
5755 | rocksdb::Options rocks_opts; | |
f67539c2 | 5756 | r = RocksDBStore::ParseOptionsFromStringStatic( |
9f95a23c TL |
5757 | cct, |
5758 | options, | |
5759 | rocks_opts, | |
5760 | nullptr); | |
5761 | if (r < 0) { | |
5762 | return r; | |
5763 | } | |
f67539c2 TL |
5764 | if (cct->_conf->bluestore_volume_selection_policy == "fit_to_fast") { |
5765 | vselector = new FitToFastVolumeSelector( | |
9f95a23c TL |
5766 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, |
5767 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
f67539c2 TL |
5768 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100); |
5769 | } else { | |
5770 | double reserved_factor = cct->_conf->bluestore_volume_selection_reserved_factor; | |
5771 | vselector = | |
5772 | new RocksDBBlueFSVolumeSelector( | |
5773 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
5774 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
5775 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100, | |
5776 | 1024 * 1024 * 1024, //FIXME: set expected l0 size here | |
5777 | rocks_opts.max_bytes_for_level_base, | |
5778 | rocks_opts.max_bytes_for_level_multiplier, | |
5779 | reserved_factor, | |
5780 | cct->_conf->bluestore_volume_selection_reserved, | |
5781 | cct->_conf->bluestore_volume_selection_policy == "use_some_extra"); | |
5782 | } | |
9f95a23c | 5783 | } |
11fdf7f2 | 5784 | if (create) { |
9f95a23c | 5785 | bluefs->mkfs(fsid, bluefs_layout); |
11fdf7f2 | 5786 | } |
9f95a23c | 5787 | bluefs->set_volume_selector(vselector); |
11fdf7f2 TL |
5788 | r = bluefs->mount(); |
5789 | if (r < 0) { | |
5790 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
5791 | } | |
9f95a23c | 5792 | ceph_assert_always(bluefs->maybe_verify_layout(bluefs_layout) == 0); |
11fdf7f2 TL |
5793 | return r; |
5794 | } | |
5795 | ||
1911f103 | 5796 | void BlueStore::_close_bluefs(bool cold_close) |
11fdf7f2 | 5797 | { |
1911f103 | 5798 | bluefs->umount(cold_close); |
11fdf7f2 TL |
5799 | _minimal_close_bluefs(); |
5800 | } | |
5801 | ||
5802 | void BlueStore::_minimal_close_bluefs() | |
5803 | { | |
5804 | delete bluefs; | |
5805 | bluefs = NULL; | |
5806 | } | |
5807 | ||
5808 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
5809 | { | |
5810 | if (create) { | |
5811 | *ret = cct->_conf->bluestore_bluefs; | |
5812 | } else { | |
5813 | string s; | |
5814 | int r = read_meta("bluefs", &s); | |
5815 | if (r < 0) { | |
5816 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
5817 | return -EIO; | |
5818 | } | |
5819 | if (s == "1") { | |
5820 | *ret = true; | |
5821 | } else if (s == "0") { | |
5822 | *ret = false; | |
31f18b77 | 5823 | } else { |
11fdf7f2 TL |
5824 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
5825 | << dendl; | |
5826 | return -EIO; | |
5827 | } | |
5828 | } | |
5829 | return 0; | |
5830 | } | |
5831 | ||
5832 | /* | |
5833 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
5834 | * in the proper order | |
5835 | */ | |
f67539c2 | 5836 | int BlueStore::_open_db_and_around(bool read_only, bool to_repair) |
11fdf7f2 | 5837 | { |
f67539c2 TL |
5838 | dout(0) << __func__ << " read-only:" << read_only |
5839 | << " repair:" << to_repair << dendl; | |
5840 | { | |
5841 | string type; | |
5842 | int r = read_meta("type", &type); | |
5843 | if (r < 0) { | |
5844 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
5845 | << dendl; | |
11fdf7f2 | 5846 | return r; |
f67539c2 | 5847 | } |
11fdf7f2 | 5848 | |
f67539c2 TL |
5849 | if (type != "bluestore") { |
5850 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
5851 | return -EIO; | |
11fdf7f2 | 5852 | } |
f67539c2 | 5853 | } |
11fdf7f2 | 5854 | |
f67539c2 TL |
5855 | int r = _open_path(); |
5856 | if (r < 0) | |
5857 | return r; | |
5858 | r = _open_fsid(false); | |
5859 | if (r < 0) | |
5860 | goto out_path; | |
11fdf7f2 | 5861 | |
f67539c2 TL |
5862 | r = _read_fsid(&fsid); |
5863 | if (r < 0) | |
5864 | goto out_fsid; | |
11fdf7f2 | 5865 | |
f67539c2 TL |
5866 | r = _lock_fsid(); |
5867 | if (r < 0) | |
5868 | goto out_fsid; | |
11fdf7f2 | 5869 | |
f67539c2 TL |
5870 | r = _open_bdev(false); |
5871 | if (r < 0) | |
5872 | goto out_fsid; | |
7c673cae | 5873 | |
f67539c2 TL |
5874 | // open in read-only first to read FM list and init allocator |
5875 | // as they might be needed for some BlueFS procedures | |
5876 | r = _open_db(false, false, true); | |
5877 | if (r < 0) | |
5878 | goto out_bdev; | |
11fdf7f2 | 5879 | |
f67539c2 TL |
5880 | r = _open_super_meta(); |
5881 | if (r < 0) { | |
5882 | goto out_db; | |
5883 | } | |
5884 | ||
5885 | r = _open_fm(nullptr, true); | |
5886 | if (r < 0) | |
5887 | goto out_db; | |
5888 | ||
5889 | r = _init_alloc(); | |
5890 | if (r < 0) | |
5891 | goto out_fm; | |
5892 | ||
5893 | // Re-open in the proper mode(s). | |
5894 | ||
5895 | // Can't simply bypass second open for read-only mode as we need to | |
5896 | // load allocated extents from bluefs into allocator. | |
5897 | // And now it's time to do that | |
5898 | // | |
5899 | _close_db(true); | |
5900 | ||
5901 | r = _open_db(false, to_repair, read_only); | |
5902 | if (r < 0) { | |
5903 | goto out_alloc; | |
11fdf7f2 TL |
5904 | } |
5905 | return 0; | |
5906 | ||
f67539c2 TL |
5907 | out_alloc: |
5908 | _close_alloc(); | |
5909 | out_fm: | |
11fdf7f2 TL |
5910 | _close_fm(); |
5911 | out_db: | |
1911f103 | 5912 | _close_db(read_only); |
f67539c2 TL |
5913 | out_bdev: |
5914 | _close_bdev(); | |
5915 | out_fsid: | |
5916 | _close_fsid(); | |
5917 | out_path: | |
5918 | _close_path(); | |
11fdf7f2 TL |
5919 | return r; |
5920 | } | |
5921 | ||
1911f103 | 5922 | void BlueStore::_close_db_and_around(bool read_only) |
11fdf7f2 | 5923 | { |
f67539c2 TL |
5924 | _close_db(read_only); |
5925 | _close_fm(); | |
5926 | _close_alloc(); | |
5927 | _close_bdev(); | |
5928 | _close_fsid(); | |
5929 | _close_path(); | |
5930 | } | |
5931 | ||
5932 | int BlueStore::open_db_environment(KeyValueDB **pdb, bool to_repair) | |
5933 | { | |
5934 | _kv_only = true; | |
5935 | int r = _open_db_and_around(false, to_repair); | |
5936 | if (r == 0) { | |
5937 | *pdb = db; | |
11fdf7f2 | 5938 | } else { |
f67539c2 | 5939 | *pdb = nullptr; |
11fdf7f2 | 5940 | } |
f67539c2 | 5941 | return r; |
11fdf7f2 TL |
5942 | } |
5943 | ||
f67539c2 | 5944 | int BlueStore::close_db_environment() |
11fdf7f2 | 5945 | { |
f67539c2 TL |
5946 | _close_db_and_around(false); |
5947 | return 0; | |
11fdf7f2 TL |
5948 | } |
5949 | ||
f67539c2 TL |
5950 | int BlueStore::_prepare_db_environment(bool create, bool read_only, |
5951 | std::string* _fn, std::string* _kv_backend) | |
11fdf7f2 TL |
5952 | { |
5953 | int r; | |
5954 | ceph_assert(!db); | |
f67539c2 TL |
5955 | std::string& fn=*_fn; |
5956 | std::string& kv_backend=*_kv_backend; | |
5957 | fn = path + "/db"; | |
11fdf7f2 TL |
5958 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); |
5959 | ||
11fdf7f2 TL |
5960 | if (create) { |
5961 | kv_backend = cct->_conf->bluestore_kvbackend; | |
5962 | } else { | |
5963 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 5964 | if (r < 0) { |
11fdf7f2 TL |
5965 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
5966 | return -EIO; | |
5967 | } | |
5968 | } | |
5969 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
5970 | ||
5971 | bool do_bluefs; | |
5972 | r = _is_bluefs(create, &do_bluefs); | |
5973 | if (r < 0) { | |
5974 | return r; | |
5975 | } | |
5976 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
5977 | ||
5978 | map<string,string> kv_options; | |
5979 | // force separate wal dir for all new deployments. | |
5980 | kv_options["separate_wal_dir"] = 1; | |
5981 | rocksdb::Env *env = NULL; | |
5982 | if (do_bluefs) { | |
5983 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
5984 | if (kv_backend != "rocksdb") { | |
5985 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
5986 | return -EINVAL; | |
7c673cae | 5987 | } |
11fdf7f2 | 5988 | |
f67539c2 | 5989 | r = _open_bluefs(create, read_only); |
11fdf7f2 TL |
5990 | if (r < 0) { |
5991 | return r; | |
5992 | } | |
11fdf7f2 | 5993 | |
7c673cae | 5994 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
9f95a23c TL |
5995 | rocksdb::Env* a = new BlueRocksEnv(bluefs); |
5996 | rocksdb::Env* b = rocksdb::Env::Default(); | |
7c673cae | 5997 | if (create) { |
9f95a23c TL |
5998 | string cmd = "rm -rf " + path + "/db " + |
5999 | path + "/db.slow " + | |
6000 | path + "/db.wal"; | |
6001 | int r = system(cmd.c_str()); | |
6002 | (void)r; | |
7c673cae FG |
6003 | } |
6004 | env = new rocksdb::EnvMirror(b, a, false, true); | |
1911f103 | 6005 | } else { |
7c673cae FG |
6006 | env = new BlueRocksEnv(bluefs); |
6007 | ||
6008 | // simplify the dir names, too, as "seen" by rocksdb | |
6009 | fn = "db"; | |
6010 | } | |
9f95a23c TL |
6011 | BlueFSVolumeSelector::paths paths; |
6012 | bluefs->get_vselector_paths(fn, paths); | |
7c673cae | 6013 | |
9f95a23c | 6014 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
7c673cae FG |
6015 | // we have both block.db and block; tell rocksdb! |
6016 | // note: the second (last) size value doesn't really matter | |
6017 | ostringstream db_paths; | |
9f95a23c TL |
6018 | bool first = true; |
6019 | for (auto& p : paths) { | |
6020 | if (!first) { | |
6021 | db_paths << " "; | |
6022 | } | |
6023 | first = false; | |
6024 | db_paths << p.first << "," << p.second; | |
6025 | ||
6026 | } | |
11fdf7f2 | 6027 | kv_options["db_paths"] = db_paths.str(); |
9f95a23c | 6028 | dout(1) << __func__ << " set db_paths to " << db_paths.str() << dendl; |
7c673cae FG |
6029 | } |
6030 | ||
6031 | if (create) { | |
9f95a23c TL |
6032 | for (auto& p : paths) { |
6033 | env->CreateDir(p.first); | |
6034 | } | |
6035 | // Selectors don't provide wal path so far hence create explicitly | |
11fdf7f2 | 6036 | env->CreateDir(fn + ".wal"); |
11fdf7f2 TL |
6037 | } else { |
6038 | std::vector<std::string> res; | |
6039 | // check for dir presence | |
6040 | auto r = env->GetChildren(fn+".wal", &res); | |
6041 | if (r.IsNotFound()) { | |
6042 | kv_options.erase("separate_wal_dir"); | |
6043 | } | |
7c673cae | 6044 | } |
11fdf7f2 TL |
6045 | } else { |
6046 | string walfn = path + "/db.wal"; | |
7c673cae | 6047 | |
11fdf7f2 TL |
6048 | if (create) { |
6049 | int r = ::mkdir(fn.c_str(), 0755); | |
6050 | if (r < 0) | |
6051 | r = -errno; | |
6052 | if (r < 0 && r != -EEXIST) { | |
6053 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
6054 | << dendl; | |
6055 | return r; | |
6056 | } | |
6057 | ||
6058 | // wal_dir, too! | |
7c673cae FG |
6059 | r = ::mkdir(walfn.c_str(), 0755); |
6060 | if (r < 0) | |
6061 | r = -errno; | |
6062 | if (r < 0 && r != -EEXIST) { | |
6063 | derr << __func__ << " failed to create " << walfn | |
6064 | << ": " << cpp_strerror(r) | |
6065 | << dendl; | |
6066 | return r; | |
6067 | } | |
11fdf7f2 TL |
6068 | } else { |
6069 | struct stat st; | |
6070 | r = ::stat(walfn.c_str(), &st); | |
6071 | if (r < 0 && errno == ENOENT) { | |
6072 | kv_options.erase("separate_wal_dir"); | |
6073 | } | |
7c673cae FG |
6074 | } |
6075 | } | |
6076 | ||
91327a77 | 6077 | |
7c673cae FG |
6078 | db = KeyValueDB::create(cct, |
6079 | kv_backend, | |
6080 | fn, | |
11fdf7f2 | 6081 | kv_options, |
7c673cae FG |
6082 | static_cast<void*>(env)); |
6083 | if (!db) { | |
6084 | derr << __func__ << " error creating db" << dendl; | |
6085 | if (bluefs) { | |
1911f103 | 6086 | _close_bluefs(read_only); |
7c673cae FG |
6087 | } |
6088 | // delete env manually here since we can't depend on db to do this | |
6089 | // under this case | |
6090 | delete env; | |
6091 | env = NULL; | |
6092 | return -EIO; | |
6093 | } | |
6094 | ||
f67539c2 | 6095 | FreelistManager::setup_merge_operators(db, freelist_type); |
7c673cae | 6096 | db->set_merge_operator(PREFIX_STAT, merge_op); |
91327a77 | 6097 | db->set_cache_size(cache_kv_ratio * cache_size); |
f67539c2 TL |
6098 | return 0; |
6099 | } | |
31f18b77 | 6100 | |
f67539c2 TL |
6101 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) |
6102 | { | |
6103 | int r; | |
6104 | ceph_assert(!(create && read_only)); | |
6105 | string options; | |
6106 | string options_annex; | |
6107 | stringstream err; | |
6108 | string kv_dir_fn; | |
6109 | string kv_backend; | |
6110 | std::string sharding_def; | |
6111 | r = _prepare_db_environment(create, read_only, &kv_dir_fn, &kv_backend); | |
6112 | if (r < 0) { | |
6113 | derr << __func__ << " failed to prepare db environment: " << err.str() << dendl; | |
6114 | return -EIO; | |
6115 | } | |
11fdf7f2 | 6116 | if (kv_backend == "rocksdb") { |
7c673cae | 6117 | options = cct->_conf->bluestore_rocksdb_options; |
cd265ab1 TL |
6118 | options_annex = cct->_conf->bluestore_rocksdb_options_annex; |
6119 | if (!options_annex.empty()) { | |
6120 | if (!options.empty() && | |
6121 | *options.rbegin() != ',') { | |
6122 | options += ','; | |
6123 | } | |
6124 | options += options_annex; | |
6125 | } | |
11fdf7f2 | 6126 | |
f67539c2 TL |
6127 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { |
6128 | sharding_def = cct->_conf.get_val<std::string>("bluestore_rocksdb_cfs"); | |
11fdf7f2 TL |
6129 | } |
6130 | } | |
6131 | ||
7c673cae | 6132 | db->init(options); |
11fdf7f2 TL |
6133 | if (to_repair_db) |
6134 | return 0; | |
6135 | if (create) { | |
f67539c2 | 6136 | r = db->create_and_open(err, sharding_def); |
11fdf7f2 TL |
6137 | } else { |
6138 | // we pass in cf list here, but it is only used if the db already has | |
6139 | // column families created. | |
6140 | r = read_only ? | |
f67539c2 TL |
6141 | db->open_read_only(err, sharding_def) : |
6142 | db->open(err, sharding_def); | |
11fdf7f2 | 6143 | } |
7c673cae FG |
6144 | if (r) { |
6145 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
1911f103 | 6146 | _close_db(read_only); |
7c673cae FG |
6147 | return -EIO; |
6148 | } | |
6149 | dout(1) << __func__ << " opened " << kv_backend | |
f67539c2 | 6150 | << " path " << kv_dir_fn << " options " << options << dendl; |
7c673cae | 6151 | return 0; |
7c673cae FG |
6152 | } |
6153 | ||
1911f103 | 6154 | void BlueStore::_close_db(bool cold_close) |
7c673cae | 6155 | { |
11fdf7f2 | 6156 | ceph_assert(db); |
7c673cae FG |
6157 | delete db; |
6158 | db = NULL; | |
6159 | if (bluefs) { | |
1911f103 | 6160 | _close_bluefs(cold_close); |
7c673cae FG |
6161 | } |
6162 | } | |
6163 | ||
11fdf7f2 | 6164 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 6165 | { |
11fdf7f2 TL |
6166 | auto dump_interval = |
6167 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
6168 | if (dump_interval > 0 && | |
6169 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
f67539c2 | 6170 | shared_alloc.a->dump(); |
11fdf7f2 TL |
6171 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); |
6172 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 6173 | } |
11fdf7f2 | 6174 | } |
7c673cae | 6175 | |
eafe8130 | 6176 | int BlueStore::_open_collections() |
7c673cae | 6177 | { |
28e407b8 | 6178 | dout(10) << __func__ << dendl; |
eafe8130 | 6179 | collections_had_errors = false; |
11fdf7f2 | 6180 | ceph_assert(coll_map.empty()); |
7c673cae FG |
6181 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
6182 | for (it->upper_bound(string()); | |
6183 | it->valid(); | |
6184 | it->next()) { | |
6185 | coll_t cid; | |
6186 | if (cid.parse(it->key())) { | |
9f95a23c | 6187 | auto c = ceph::make_ref<Collection>( |
7c673cae | 6188 | this, |
9f95a23c TL |
6189 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
6190 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
6191 | cid); | |
7c673cae | 6192 | bufferlist bl = it->value(); |
11fdf7f2 | 6193 | auto p = bl.cbegin(); |
7c673cae | 6194 | try { |
11fdf7f2 | 6195 | decode(c->cnode, p); |
f67539c2 | 6196 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
6197 | derr << __func__ << " failed to decode cnode, key:" |
6198 | << pretty_binary_string(it->key()) << dendl; | |
6199 | return -EIO; | |
6200 | } | |
28e407b8 AA |
6201 | dout(20) << __func__ << " opened " << cid << " " << c |
6202 | << " " << c->cnode << dendl; | |
11fdf7f2 | 6203 | _osr_attach(c.get()); |
7c673cae | 6204 | coll_map[cid] = c; |
11fdf7f2 | 6205 | |
7c673cae FG |
6206 | } else { |
6207 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
eafe8130 | 6208 | collections_had_errors = true; |
7c673cae FG |
6209 | } |
6210 | } | |
6211 | return 0; | |
6212 | } | |
6213 | ||
eafe8130 TL |
6214 | void BlueStore::_fsck_collections(int64_t* errors) |
6215 | { | |
6216 | if (collections_had_errors) { | |
6217 | dout(10) << __func__ << dendl; | |
f67539c2 | 6218 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
6219 | for (it->upper_bound(string()); |
6220 | it->valid(); | |
6221 | it->next()) { | |
6222 | coll_t cid; | |
6223 | if (!cid.parse(it->key())) { | |
6224 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
6225 | if (errors) { | |
6226 | (*errors)++; | |
6227 | } | |
6228 | } | |
6229 | } | |
6230 | } | |
6231 | } | |
6232 | ||
9f95a23c TL |
6233 | void BlueStore::_set_per_pool_omap() |
6234 | { | |
f67539c2 | 6235 | per_pool_omap = OMAP_BULK; |
9f95a23c TL |
6236 | bufferlist bl; |
6237 | db->get(PREFIX_SUPER, "per_pool_omap", &bl); | |
6238 | if (bl.length()) { | |
f67539c2 TL |
6239 | auto s = bl.to_str(); |
6240 | if (s == stringify(OMAP_PER_POOL)) { | |
6241 | per_pool_omap = OMAP_PER_POOL; | |
6242 | } else { | |
6243 | ceph_assert(s == stringify(OMAP_PER_PG)); | |
6244 | per_pool_omap = OMAP_PER_PG; | |
6245 | } | |
6246 | dout(10) << __func__ << " per_pool_omap = " << per_pool_omap << dendl; | |
9f95a23c TL |
6247 | } else { |
6248 | dout(10) << __func__ << " per_pool_omap not present" << dendl; | |
6249 | } | |
f67539c2 | 6250 | _check_no_per_pg_or_pool_omap_alert(); |
9f95a23c TL |
6251 | } |
6252 | ||
224ce89b | 6253 | void BlueStore::_open_statfs() |
31f18b77 | 6254 | { |
11fdf7f2 TL |
6255 | osd_pools.clear(); |
6256 | vstatfs.reset(); | |
6257 | ||
31f18b77 | 6258 | bufferlist bl; |
11fdf7f2 | 6259 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 6260 | if (r >= 0) { |
11fdf7f2 | 6261 | per_pool_stat_collection = false; |
31f18b77 | 6262 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 6263 | auto it = bl.cbegin(); |
31f18b77 | 6264 | vstatfs.decode(it); |
11fdf7f2 | 6265 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 6266 | } else { |
31f18b77 FG |
6267 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
6268 | } | |
81eedcae | 6269 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
6270 | } else { |
6271 | per_pool_stat_collection = true; | |
6272 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
f67539c2 | 6273 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 TL |
6274 | for (it->upper_bound(string()); |
6275 | it->valid(); | |
6276 | it->next()) { | |
6277 | ||
6278 | uint64_t pool_id; | |
6279 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6280 | ceph_assert(r == 0); | |
6281 | ||
6282 | bufferlist bl; | |
6283 | bl = it->value(); | |
6284 | auto p = bl.cbegin(); | |
6285 | auto& st = osd_pools[pool_id]; | |
6286 | try { | |
6287 | st.decode(p); | |
6288 | vstatfs += st; | |
6289 | ||
6290 | dout(30) << __func__ << " pool " << pool_id | |
6291 | << " statfs " << st << dendl; | |
f67539c2 | 6292 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
6293 | derr << __func__ << " failed to decode pool stats, key:" |
6294 | << pretty_binary_string(it->key()) << dendl; | |
6295 | } | |
6296 | } | |
31f18b77 | 6297 | } |
11fdf7f2 TL |
6298 | dout(30) << __func__ << " statfs " << vstatfs << dendl; |
6299 | ||
31f18b77 FG |
6300 | } |
6301 | ||
7c673cae FG |
6302 | int BlueStore::_setup_block_symlink_or_file( |
6303 | string name, | |
6304 | string epath, | |
6305 | uint64_t size, | |
6306 | bool create) | |
6307 | { | |
6308 | dout(20) << __func__ << " name " << name << " path " << epath | |
6309 | << " size " << size << " create=" << (int)create << dendl; | |
6310 | int r = 0; | |
91327a77 | 6311 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6312 | if (create) |
6313 | flags |= O_CREAT; | |
6314 | if (epath.length()) { | |
6315 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
6316 | if (r < 0) { | |
6317 | r = -errno; | |
6318 | derr << __func__ << " failed to create " << name << " symlink to " | |
6319 | << epath << ": " << cpp_strerror(r) << dendl; | |
6320 | return r; | |
6321 | } | |
6322 | ||
6323 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
6324 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
6325 | if (fd < 0) { | |
6326 | r = -errno; | |
6327 | derr << __func__ << " failed to open " << epath << " file: " | |
6328 | << cpp_strerror(r) << dendl; | |
6329 | return r; | |
6330 | } | |
11fdf7f2 TL |
6331 | // write the Transport ID of the NVMe device |
6332 | // a transport id looks like: "trtype:PCIe traddr:0000:02:00.0" | |
6333 | // where "0000:02:00.0" is the selector of a PCI device, see | |
6334 | // the first column of "lspci -mm -n -D" | |
6335 | string trid{"trtype:PCIe "}; | |
6336 | trid += "traddr:"; | |
6337 | trid += epath.substr(strlen(SPDK_PREFIX)); | |
6338 | r = ::write(fd, trid.c_str(), trid.size()); | |
6339 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
6340 | dout(1) << __func__ << " created " << name << " symlink to " |
6341 | << epath << dendl; | |
6342 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6343 | } | |
6344 | } | |
6345 | if (size) { | |
6346 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
6347 | if (fd >= 0) { | |
6348 | // block file is present | |
6349 | struct stat st; | |
6350 | int r = ::fstat(fd, &st); | |
6351 | if (r == 0 && | |
6352 | S_ISREG(st.st_mode) && // if it is a regular file | |
6353 | st.st_size == 0) { // and is 0 bytes | |
6354 | r = ::ftruncate(fd, size); | |
6355 | if (r < 0) { | |
6356 | r = -errno; | |
6357 | derr << __func__ << " failed to resize " << name << " file to " | |
6358 | << size << ": " << cpp_strerror(r) << dendl; | |
6359 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6360 | return r; | |
6361 | } | |
6362 | ||
6363 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
6364 | r = ::ceph_posix_fallocate(fd, 0, size); |
6365 | if (r > 0) { | |
7c673cae FG |
6366 | derr << __func__ << " failed to prefallocate " << name << " file to " |
6367 | << size << ": " << cpp_strerror(r) << dendl; | |
6368 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6369 | return -r; | |
6370 | } | |
7c673cae FG |
6371 | } |
6372 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 6373 | << byte_u_t(size) << dendl; |
7c673cae FG |
6374 | } |
6375 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6376 | } else { | |
6377 | int r = -errno; | |
6378 | if (r != -ENOENT) { | |
6379 | derr << __func__ << " failed to open " << name << " file: " | |
6380 | << cpp_strerror(r) << dendl; | |
6381 | return r; | |
6382 | } | |
6383 | } | |
6384 | } | |
6385 | return 0; | |
6386 | } | |
6387 | ||
6388 | int BlueStore::mkfs() | |
6389 | { | |
6390 | dout(1) << __func__ << " path " << path << dendl; | |
6391 | int r; | |
6392 | uuid_d old_fsid; | |
f67539c2 | 6393 | uint64_t reserved; |
eafe8130 TL |
6394 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6395 | derr << __func__ << " osd_max_object_size " | |
6396 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6397 | << OBJECT_MAX_SIZE << dendl; | |
6398 | return -EINVAL; | |
6399 | } | |
6400 | ||
7c673cae FG |
6401 | { |
6402 | string done; | |
6403 | r = read_meta("mkfs_done", &done); | |
6404 | if (r == 0) { | |
6405 | dout(1) << __func__ << " already created" << dendl; | |
6406 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
6407 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6408 | if (r < 0) { | |
6409 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
6410 | << dendl; | |
6411 | return r; | |
6412 | } | |
6413 | if (r > 0) { | |
6414 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
6415 | r = -EIO; | |
6416 | } | |
6417 | } | |
6418 | return r; // idempotent | |
6419 | } | |
6420 | } | |
6421 | ||
6422 | { | |
6423 | string type; | |
6424 | r = read_meta("type", &type); | |
6425 | if (r == 0) { | |
6426 | if (type != "bluestore") { | |
6427 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6428 | return -EIO; | |
6429 | } | |
6430 | } else { | |
6431 | r = write_meta("type", "bluestore"); | |
6432 | if (r < 0) | |
6433 | return r; | |
6434 | } | |
6435 | } | |
6436 | ||
6437 | freelist_type = "bitmap"; | |
6438 | ||
6439 | r = _open_path(); | |
6440 | if (r < 0) | |
6441 | return r; | |
6442 | ||
6443 | r = _open_fsid(true); | |
6444 | if (r < 0) | |
6445 | goto out_path_fd; | |
6446 | ||
6447 | r = _lock_fsid(); | |
6448 | if (r < 0) | |
6449 | goto out_close_fsid; | |
6450 | ||
6451 | r = _read_fsid(&old_fsid); | |
6452 | if (r < 0 || old_fsid.is_zero()) { | |
6453 | if (fsid.is_zero()) { | |
6454 | fsid.generate_random(); | |
6455 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
6456 | } else { | |
6457 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
6458 | } | |
6459 | // we'll write it later. | |
6460 | } else { | |
6461 | if (!fsid.is_zero() && fsid != old_fsid) { | |
6462 | derr << __func__ << " on-disk fsid " << old_fsid | |
6463 | << " != provided " << fsid << dendl; | |
6464 | r = -EINVAL; | |
6465 | goto out_close_fsid; | |
6466 | } | |
6467 | fsid = old_fsid; | |
6468 | } | |
6469 | ||
6470 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
6471 | cct->_conf->bluestore_block_size, | |
6472 | cct->_conf->bluestore_block_create); | |
6473 | if (r < 0) | |
6474 | goto out_close_fsid; | |
6475 | if (cct->_conf->bluestore_bluefs) { | |
6476 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
6477 | cct->_conf->bluestore_block_wal_size, | |
6478 | cct->_conf->bluestore_block_wal_create); | |
6479 | if (r < 0) | |
6480 | goto out_close_fsid; | |
6481 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
6482 | cct->_conf->bluestore_block_db_size, | |
6483 | cct->_conf->bluestore_block_db_create); | |
6484 | if (r < 0) | |
6485 | goto out_close_fsid; | |
6486 | } | |
6487 | ||
6488 | r = _open_bdev(true); | |
6489 | if (r < 0) | |
6490 | goto out_close_fsid; | |
6491 | ||
3efd9988 FG |
6492 | // choose min_alloc_size |
6493 | if (cct->_conf->bluestore_min_alloc_size) { | |
6494 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; | |
6495 | } else { | |
11fdf7f2 | 6496 | ceph_assert(bdev); |
f67539c2 | 6497 | if (_use_rotational_settings()) { |
3efd9988 FG |
6498 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; |
6499 | } else { | |
6500 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
6501 | } | |
6502 | } | |
11fdf7f2 | 6503 | _validate_bdev(); |
3efd9988 FG |
6504 | |
6505 | // make sure min_alloc_size is power of 2 aligned. | |
11fdf7f2 | 6506 | if (!isp2(min_alloc_size)) { |
3efd9988 FG |
6507 | derr << __func__ << " min_alloc_size 0x" |
6508 | << std::hex << min_alloc_size << std::dec | |
6509 | << " is not power of 2 aligned!" | |
6510 | << dendl; | |
6511 | r = -EINVAL; | |
6512 | goto out_close_bdev; | |
6513 | } | |
6514 | ||
f67539c2 TL |
6515 | r = _create_alloc(); |
6516 | if (r < 0) { | |
6517 | goto out_close_bdev; | |
6518 | } | |
6519 | ||
6520 | reserved = _get_ondisk_reserved(); | |
6521 | shared_alloc.a->init_add_free(reserved, | |
6522 | p2align(bdev->get_size(), min_alloc_size) - reserved); | |
6523 | ||
7c673cae FG |
6524 | r = _open_db(true); |
6525 | if (r < 0) | |
f67539c2 | 6526 | goto out_close_alloc; |
7c673cae | 6527 | |
7c673cae FG |
6528 | { |
6529 | KeyValueDB::Transaction t = db->get_transaction(); | |
1911f103 | 6530 | r = _open_fm(t, true); |
11fdf7f2 TL |
6531 | if (r < 0) |
6532 | goto out_close_db; | |
7c673cae FG |
6533 | { |
6534 | bufferlist bl; | |
11fdf7f2 | 6535 | encode((uint64_t)0, bl); |
7c673cae FG |
6536 | t->set(PREFIX_SUPER, "nid_max", bl); |
6537 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
6538 | } | |
6539 | ||
7c673cae FG |
6540 | { |
6541 | bufferlist bl; | |
11fdf7f2 | 6542 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
6543 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
6544 | } | |
9f95a23c TL |
6545 | { |
6546 | bufferlist bl; | |
f67539c2 | 6547 | bl.append(stringify(OMAP_PER_PG)); |
9f95a23c TL |
6548 | t->set(PREFIX_SUPER, "per_pool_omap", bl); |
6549 | } | |
7c673cae FG |
6550 | ondisk_format = latest_ondisk_format; |
6551 | _prepare_ondisk_format_super(t); | |
6552 | db->submit_transaction_sync(t); | |
6553 | } | |
6554 | ||
7c673cae FG |
6555 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
6556 | if (r < 0) | |
224ce89b WB |
6557 | goto out_close_fm; |
6558 | ||
3efd9988 | 6559 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 6560 | if (r < 0) |
224ce89b | 6561 | goto out_close_fm; |
7c673cae FG |
6562 | |
6563 | if (fsid != old_fsid) { | |
6564 | r = _write_fsid(); | |
6565 | if (r < 0) { | |
6566 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 6567 | goto out_close_fm; |
7c673cae FG |
6568 | } |
6569 | } | |
6570 | ||
7c673cae FG |
6571 | out_close_fm: |
6572 | _close_fm(); | |
6573 | out_close_db: | |
1911f103 | 6574 | _close_db(false); |
f67539c2 TL |
6575 | out_close_alloc: |
6576 | _close_alloc(); | |
7c673cae FG |
6577 | out_close_bdev: |
6578 | _close_bdev(); | |
6579 | out_close_fsid: | |
6580 | _close_fsid(); | |
6581 | out_path_fd: | |
6582 | _close_path(); | |
6583 | ||
6584 | if (r == 0 && | |
6585 | cct->_conf->bluestore_fsck_on_mkfs) { | |
6586 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6587 | if (rc < 0) | |
6588 | return rc; | |
6589 | if (rc > 0) { | |
6590 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6591 | r = -EIO; | |
6592 | } | |
11fdf7f2 TL |
6593 | } |
6594 | ||
6595 | if (r == 0) { | |
6596 | // indicate success by writing the 'mkfs_done' file | |
6597 | r = write_meta("mkfs_done", "yes"); | |
6598 | } | |
6599 | ||
6600 | if (r < 0) { | |
6601 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6602 | } else { | |
6603 | dout(0) << __func__ << " success" << dendl; | |
6604 | } | |
6605 | return r; | |
6606 | } | |
6607 | ||
11fdf7f2 TL |
6608 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) |
6609 | { | |
6610 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6611 | int r; | |
6612 | ceph_assert(path_fd < 0); | |
6613 | ||
6614 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6615 | ||
6616 | if (!cct->_conf->bluestore_bluefs) { | |
6617 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6618 | return -EIO; | |
6619 | } | |
6620 | ||
f67539c2 | 6621 | r = _open_db_and_around(true); |
11fdf7f2 | 6622 | |
11fdf7f2 TL |
6623 | if (id == BlueFS::BDEV_NEWWAL) { |
6624 | string p = path + "/block.wal"; | |
6625 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
6626 | cct->_conf->bluestore_block_wal_size, | |
6627 | true); | |
6628 | ceph_assert(r == 0); | |
6629 | ||
6630 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
f67539c2 TL |
6631 | cct->_conf->bdev_enable_discard, |
6632 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
6633 | ceph_assert(r == 0); |
6634 | ||
6635 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6636 | r = _check_or_set_bdev_label( | |
6637 | p, | |
6638 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6639 | "bluefs wal", | |
6640 | true); | |
6641 | ceph_assert(r == 0); | |
6642 | } | |
6643 | ||
9f95a23c | 6644 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6645 | } else if (id == BlueFS::BDEV_NEWDB) { |
6646 | string p = path + "/block.db"; | |
6647 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
6648 | cct->_conf->bluestore_block_db_size, | |
6649 | true); | |
6650 | ceph_assert(r == 0); | |
6651 | ||
6652 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
f67539c2 TL |
6653 | cct->_conf->bdev_enable_discard, |
6654 | SUPER_RESERVED); | |
11fdf7f2 TL |
6655 | ceph_assert(r == 0); |
6656 | ||
6657 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6658 | r = _check_or_set_bdev_label( | |
6659 | p, | |
6660 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6661 | "bluefs db", | |
6662 | true); | |
6663 | ceph_assert(r == 0); | |
6664 | } | |
9f95a23c TL |
6665 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6666 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
6667 | } |
6668 | ||
6669 | bluefs->umount(); | |
6670 | bluefs->mount(); | |
6671 | ||
9f95a23c | 6672 | r = bluefs->prepare_new_device(id, bluefs_layout); |
11fdf7f2 TL |
6673 | ceph_assert(r == 0); |
6674 | ||
6675 | if (r < 0) { | |
6676 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6677 | } else { | |
6678 | dout(0) << __func__ << " success" << dendl; | |
6679 | } | |
6680 | ||
f67539c2 | 6681 | _close_db_and_around(true); |
11fdf7f2 TL |
6682 | return r; |
6683 | } | |
6684 | ||
6685 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
6686 | int id) | |
6687 | { | |
6688 | dout(10) << __func__ << " id:" << id << dendl; | |
6689 | ceph_assert(path_fd < 0); | |
6690 | ||
6691 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
6692 | ||
6693 | if (!cct->_conf->bluestore_bluefs) { | |
6694 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6695 | return -EIO; | |
6696 | } | |
6697 | ||
f67539c2 | 6698 | int r = _open_db_and_around(true); |
11fdf7f2 | 6699 | |
f67539c2 | 6700 | uint64_t used_space = 0; |
11fdf7f2 | 6701 | for(auto src_id : devs_source) { |
f67539c2 | 6702 | used_space += bluefs->get_used(src_id); |
11fdf7f2 TL |
6703 | } |
6704 | uint64_t target_free = bluefs->get_free(id); | |
f67539c2 | 6705 | if (target_free < used_space) { |
11fdf7f2 TL |
6706 | derr << __func__ |
6707 | << " can't migrate, free space at target: " << target_free | |
6708 | << " is less than required space: " << used_space | |
6709 | << dendl; | |
f67539c2 TL |
6710 | r = -ENOSPC; |
6711 | goto shutdown; | |
11fdf7f2 | 6712 | } |
9f95a23c TL |
6713 | if (devs_source.count(BlueFS::BDEV_DB)) { |
6714 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; | |
6715 | bluefs_layout.dedicated_db = false; | |
6716 | } | |
6717 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6718 | bluefs_layout.dedicated_wal = false; | |
6719 | } | |
6720 | r = bluefs->device_migrate_to_existing(cct, devs_source, id, bluefs_layout); | |
11fdf7f2 TL |
6721 | if (r < 0) { |
6722 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
6723 | goto shutdown; | |
6724 | } | |
6725 | ||
6726 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
6727 | r = unlink(string(path + "/block.db").c_str()); | |
6728 | ceph_assert(r == 0); | |
6729 | } | |
6730 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6731 | r = unlink(string(path + "/block.wal").c_str()); | |
6732 | ceph_assert(r == 0); | |
6733 | } | |
6734 | ||
6735 | shutdown: | |
f67539c2 | 6736 | _close_db_and_around(true); |
11fdf7f2 TL |
6737 | return r; |
6738 | } | |
6739 | ||
6740 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
6741 | int id, | |
6742 | const string& dev_path) | |
6743 | { | |
6744 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6745 | int r; | |
6746 | ceph_assert(path_fd < 0); | |
6747 | ||
6748 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6749 | ||
6750 | if (!cct->_conf->bluestore_bluefs) { | |
6751 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6752 | return -EIO; | |
6753 | } | |
6754 | ||
f67539c2 | 6755 | r = _open_db_and_around(true); |
11fdf7f2 | 6756 | |
11fdf7f2 TL |
6757 | string link_db; |
6758 | string link_wal; | |
6759 | if (devs_source.count(BlueFS::BDEV_DB) && | |
9f95a23c | 6760 | bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 6761 | link_db = path + "/block.db"; |
9f95a23c TL |
6762 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
6763 | bluefs_layout.dedicated_db = false; | |
11fdf7f2 TL |
6764 | } |
6765 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6766 | link_wal = path + "/block.wal"; | |
9f95a23c | 6767 | bluefs_layout.dedicated_wal = false; |
11fdf7f2 TL |
6768 | } |
6769 | ||
6770 | size_t target_size; | |
6771 | string target_name; | |
6772 | if (id == BlueFS::BDEV_NEWWAL) { | |
6773 | target_name = "block.wal"; | |
6774 | target_size = cct->_conf->bluestore_block_wal_size; | |
9f95a23c | 6775 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6776 | |
6777 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
f67539c2 TL |
6778 | cct->_conf->bdev_enable_discard, |
6779 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
6780 | ceph_assert(r == 0); |
6781 | ||
6782 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6783 | r = _check_or_set_bdev_label( | |
6784 | dev_path, | |
6785 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6786 | "bluefs wal", | |
6787 | true); | |
6788 | ceph_assert(r == 0); | |
6789 | } | |
11fdf7f2 TL |
6790 | } else if (id == BlueFS::BDEV_NEWDB) { |
6791 | target_name = "block.db"; | |
6792 | target_size = cct->_conf->bluestore_block_db_size; | |
9f95a23c TL |
6793 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6794 | bluefs_layout.dedicated_db = true; | |
31f18b77 | 6795 | |
11fdf7f2 | 6796 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
f67539c2 TL |
6797 | cct->_conf->bdev_enable_discard, |
6798 | SUPER_RESERVED); | |
11fdf7f2 TL |
6799 | ceph_assert(r == 0); |
6800 | ||
6801 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6802 | r = _check_or_set_bdev_label( | |
6803 | dev_path, | |
6804 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6805 | "bluefs db", | |
6806 | true); | |
6807 | ceph_assert(r == 0); | |
6808 | } | |
31f18b77 FG |
6809 | } |
6810 | ||
11fdf7f2 TL |
6811 | bluefs->umount(); |
6812 | bluefs->mount(); | |
6813 | ||
9f95a23c | 6814 | r = bluefs->device_migrate_to_new(cct, devs_source, id, bluefs_layout); |
11fdf7f2 | 6815 | |
7c673cae | 6816 | if (r < 0) { |
11fdf7f2 TL |
6817 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
6818 | goto shutdown; | |
6819 | } | |
6820 | ||
6821 | if (!link_db.empty()) { | |
6822 | r = unlink(link_db.c_str()); | |
6823 | ceph_assert(r == 0); | |
6824 | } | |
6825 | if (!link_wal.empty()) { | |
6826 | r = unlink(link_wal.c_str()); | |
6827 | ceph_assert(r == 0); | |
6828 | } | |
6829 | r = _setup_block_symlink_or_file( | |
6830 | target_name, | |
6831 | dev_path, | |
6832 | target_size, | |
6833 | true); | |
6834 | ceph_assert(r == 0); | |
6835 | dout(0) << __func__ << " success" << dendl; | |
6836 | ||
6837 | shutdown: | |
f67539c2 TL |
6838 | _close_db_and_around(true); |
6839 | ||
11fdf7f2 TL |
6840 | return r; |
6841 | } | |
6842 | ||
6843 | string BlueStore::get_device_path(unsigned id) | |
6844 | { | |
6845 | string res; | |
6846 | if (id < BlueFS::MAX_BDEV) { | |
6847 | switch (id) { | |
6848 | case BlueFS::BDEV_WAL: | |
6849 | res = path + "/block.wal"; | |
6850 | break; | |
6851 | case BlueFS::BDEV_DB: | |
9f95a23c | 6852 | if (id == bluefs_layout.shared_bdev) { |
11fdf7f2 TL |
6853 | res = path + "/block"; |
6854 | } else { | |
6855 | res = path + "/block.db"; | |
6856 | } | |
6857 | break; | |
6858 | case BlueFS::BDEV_SLOW: | |
6859 | res = path + "/block"; | |
6860 | break; | |
6861 | } | |
6862 | } | |
6863 | return res; | |
6864 | } | |
6865 | ||
f67539c2 TL |
6866 | int BlueStore::_set_bdev_label_size(const string& path, uint64_t size) |
6867 | { | |
6868 | bluestore_bdev_label_t label; | |
6869 | int r = _read_bdev_label(cct, path, &label); | |
6870 | if (r < 0) { | |
6871 | derr << "unable to read label for " << path << ": " | |
6872 | << cpp_strerror(r) << dendl; | |
6873 | } else { | |
6874 | label.size = size; | |
6875 | r = _write_bdev_label(cct, path, label); | |
6876 | if (r < 0) { | |
6877 | derr << "unable to write label for " << path << ": " | |
6878 | << cpp_strerror(r) << dendl; | |
6879 | } | |
6880 | } | |
6881 | return r; | |
6882 | } | |
6883 | ||
11fdf7f2 TL |
6884 | int BlueStore::expand_devices(ostream& out) |
6885 | { | |
f67539c2 | 6886 | int r = _open_db_and_around(true); |
11fdf7f2 TL |
6887 | ceph_assert(r == 0); |
6888 | bluefs->dump_block_extents(out); | |
1911f103 | 6889 | out << "Expanding DB/WAL..." << std::endl; |
11fdf7f2 | 6890 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { |
9f95a23c | 6891 | if (devid == bluefs_layout.shared_bdev ) { |
11fdf7f2 TL |
6892 | continue; |
6893 | } | |
6894 | uint64_t size = bluefs->get_block_device_size(devid); | |
6895 | if (size == 0) { | |
6896 | // no bdev | |
6897 | continue; | |
6898 | } | |
6899 | ||
f67539c2 TL |
6900 | out << devid |
6901 | <<" : expanding " << " to 0x" << size << std::dec << std::endl; | |
6902 | string p = get_device_path(devid); | |
6903 | const char* path = p.c_str(); | |
6904 | if (path == nullptr) { | |
6905 | derr << devid | |
6906 | <<": can't find device path " << dendl; | |
6907 | continue; | |
6908 | } | |
6909 | if (bluefs->bdev_support_label(devid)) { | |
6910 | if (_set_bdev_label_size(p, size) >= 0) { | |
6911 | out << devid | |
6912 | << " : size label updated to " << size | |
6913 | << std::endl; | |
11fdf7f2 | 6914 | } |
11fdf7f2 TL |
6915 | } |
6916 | } | |
6917 | uint64_t size0 = fm->get_size(); | |
6918 | uint64_t size = bdev->get_size(); | |
6919 | if (size0 < size) { | |
9f95a23c | 6920 | out << bluefs_layout.shared_bdev |
1911f103 TL |
6921 | << " : expanding " << " from 0x" << std::hex |
6922 | << size0 << " to 0x" << size << std::dec << std::endl; | |
f67539c2 TL |
6923 | _write_out_fm_meta(size); |
6924 | if (bdev->supported_bdev_label()) { | |
6925 | if (_set_bdev_label_size(path, size) >= 0) { | |
6926 | out << bluefs_layout.shared_bdev | |
6927 | << " : size label updated to " << size | |
6928 | << std::endl; | |
6929 | } | |
6930 | } | |
6931 | _close_db_and_around(true); | |
1911f103 TL |
6932 | |
6933 | // mount in read/write to sync expansion changes | |
f67539c2 | 6934 | r = _mount(); |
11fdf7f2 | 6935 | ceph_assert(r == 0); |
1911f103 TL |
6936 | umount(); |
6937 | } else { | |
f67539c2 | 6938 | _close_db_and_around(true); |
7c673cae | 6939 | } |
1911f103 TL |
6940 | return r; |
6941 | } | |
6942 | ||
6943 | int BlueStore::dump_bluefs_sizes(ostream& out) | |
6944 | { | |
f67539c2 | 6945 | int r = _open_db_and_around(true); |
1911f103 TL |
6946 | ceph_assert(r == 0); |
6947 | bluefs->dump_block_extents(out); | |
f67539c2 | 6948 | _close_db_and_around(true); |
7c673cae FG |
6949 | return r; |
6950 | } | |
6951 | ||
6952 | void BlueStore::set_cache_shards(unsigned num) | |
6953 | { | |
6954 | dout(10) << __func__ << " " << num << dendl; | |
9f95a23c TL |
6955 | size_t oold = onode_cache_shards.size(); |
6956 | size_t bold = buffer_cache_shards.size(); | |
6957 | ceph_assert(num >= oold && num >= bold); | |
6958 | onode_cache_shards.resize(num); | |
6959 | buffer_cache_shards.resize(num); | |
6960 | for (unsigned i = oold; i < num; ++i) { | |
6961 | onode_cache_shards[i] = | |
6962 | OnodeCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6963 | logger); | |
6964 | } | |
6965 | for (unsigned i = bold; i < num; ++i) { | |
6966 | buffer_cache_shards[i] = | |
6967 | BufferCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6968 | logger); | |
7c673cae FG |
6969 | } |
6970 | } | |
6971 | ||
f67539c2 | 6972 | int BlueStore::_mount() |
7c673cae FG |
6973 | { |
6974 | dout(1) << __func__ << " path " << path << dendl; | |
6975 | ||
f67539c2 | 6976 | _kv_only = false; |
7c673cae FG |
6977 | if (cct->_conf->bluestore_fsck_on_mount) { |
6978 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); | |
6979 | if (rc < 0) | |
6980 | return rc; | |
6981 | if (rc > 0) { | |
6982 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6983 | return -EIO; | |
6984 | } | |
6985 | } | |
6986 | ||
eafe8130 TL |
6987 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6988 | derr << __func__ << " osd_max_object_size " | |
6989 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6990 | << OBJECT_MAX_SIZE << dendl; | |
6991 | return -EINVAL; | |
6992 | } | |
6993 | ||
f67539c2 | 6994 | int r = _open_db_and_around(false); |
9f95a23c | 6995 | if (r < 0) { |
f67539c2 | 6996 | return r; |
11fdf7f2 | 6997 | } |
7c673cae | 6998 | |
11fdf7f2 TL |
6999 | r = _upgrade_super(); |
7000 | if (r < 0) { | |
7c673cae | 7001 | goto out_db; |
11fdf7f2 | 7002 | } |
7c673cae FG |
7003 | |
7004 | r = _open_collections(); | |
7005 | if (r < 0) | |
11fdf7f2 | 7006 | goto out_db; |
7c673cae FG |
7007 | |
7008 | r = _reload_logger(); | |
7009 | if (r < 0) | |
7010 | goto out_coll; | |
7011 | ||
31f18b77 | 7012 | _kv_start(); |
7c673cae | 7013 | |
f67539c2 TL |
7014 | if (bdev->is_smr()) { |
7015 | _zoned_cleaner_start(); | |
7016 | } | |
7017 | ||
7c673cae FG |
7018 | r = _deferred_replay(); |
7019 | if (r < 0) | |
7020 | goto out_stop; | |
7021 | ||
7022 | mempool_thread.init(); | |
7023 | ||
f67539c2 | 7024 | if ((!per_pool_stat_collection || per_pool_omap != OMAP_PER_PG) && |
eafe8130 | 7025 | cct->_conf->bluestore_fsck_quick_fix_on_mount == true) { |
9f95a23c | 7026 | |
f67539c2 | 7027 | auto was_per_pool_omap = per_pool_omap; |
9f95a23c | 7028 | |
eafe8130 TL |
7029 | dout(1) << __func__ << " quick-fix on mount" << dendl; |
7030 | _fsck_on_open(FSCK_SHALLOW, true); | |
7031 | ||
7032 | //reread statfs | |
7033 | //FIXME minor: replace with actual open/close? | |
7034 | _open_statfs(); | |
eafe8130 | 7035 | _check_legacy_statfs_alert(); |
9f95a23c TL |
7036 | |
7037 | //set again as hopefully it has been fixed | |
f67539c2 | 7038 | if (was_per_pool_omap != OMAP_PER_PG) { |
9f95a23c TL |
7039 | _set_per_pool_omap(); |
7040 | } | |
eafe8130 TL |
7041 | } |
7042 | ||
7c673cae FG |
7043 | mounted = true; |
7044 | return 0; | |
7045 | ||
7046 | out_stop: | |
f67539c2 TL |
7047 | if (bdev->is_smr()) { |
7048 | _zoned_cleaner_stop(); | |
7049 | } | |
7c673cae | 7050 | _kv_stop(); |
7c673cae | 7051 | out_coll: |
f6b5b4d7 | 7052 | _shutdown_cache(); |
7c673cae | 7053 | out_db: |
1911f103 | 7054 | _close_db_and_around(false); |
7c673cae FG |
7055 | return r; |
7056 | } | |
7057 | ||
7058 | int BlueStore::umount() | |
7059 | { | |
11fdf7f2 | 7060 | ceph_assert(_kv_only || mounted); |
7c673cae FG |
7061 | dout(1) << __func__ << dendl; |
7062 | ||
7063 | _osr_drain_all(); | |
7c673cae | 7064 | |
7c673cae | 7065 | mounted = false; |
3efd9988 FG |
7066 | if (!_kv_only) { |
7067 | mempool_thread.shutdown(); | |
f67539c2 TL |
7068 | if (bdev->is_smr()) { |
7069 | dout(20) << __func__ << " stopping zone cleaner thread" << dendl; | |
7070 | _zoned_cleaner_stop(); | |
7071 | } | |
3efd9988 FG |
7072 | dout(20) << __func__ << " stopping kv thread" << dendl; |
7073 | _kv_stop(); | |
f6b5b4d7 | 7074 | _shutdown_cache(); |
3efd9988 FG |
7075 | dout(20) << __func__ << " closing" << dendl; |
7076 | ||
3efd9988 | 7077 | } |
1911f103 | 7078 | _close_db_and_around(false); |
7c673cae FG |
7079 | |
7080 | if (cct->_conf->bluestore_fsck_on_umount) { | |
7081 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); | |
7082 | if (rc < 0) | |
7083 | return rc; | |
7084 | if (rc > 0) { | |
7085 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7086 | return -EIO; | |
7087 | } | |
7088 | } | |
7089 | return 0; | |
7090 | } | |
7091 | ||
eafe8130 TL |
7092 | int BlueStore::cold_open() |
7093 | { | |
f67539c2 | 7094 | return _open_db_and_around(true); |
eafe8130 | 7095 | } |
f67539c2 | 7096 | |
eafe8130 TL |
7097 | int BlueStore::cold_close() |
7098 | { | |
1911f103 | 7099 | _close_db_and_around(true); |
eafe8130 TL |
7100 | return 0; |
7101 | } | |
7102 | ||
9f95a23c TL |
7103 | // derr wrapper to limit enormous output and avoid log flooding. |
7104 | // Of limited use where such output is expected for now | |
7105 | #define fsck_derr(err_cnt, threshold) \ | |
7106 | if (err_cnt <= threshold) { \ | |
7107 | bool need_skip_print = err_cnt == threshold; \ | |
7108 | derr | |
7109 | ||
7110 | #define fsck_dendl \ | |
7111 | dendl; \ | |
7112 | if (need_skip_print) \ | |
7113 | derr << "more error lines skipped..." << dendl; \ | |
7c673cae | 7114 | } |
7c673cae | 7115 | |
eafe8130 TL |
7116 | int _fsck_sum_extents( |
7117 | const PExtentVector& extents, | |
7118 | bool compressed, | |
7119 | store_statfs_t& expected_statfs) | |
7120 | { | |
7121 | for (auto e : extents) { | |
7122 | if (!e.is_valid()) | |
7123 | continue; | |
7124 | expected_statfs.allocated += e.length; | |
7125 | if (compressed) { | |
7126 | expected_statfs.data_compressed_allocated += e.length; | |
7127 | } | |
7128 | } | |
7129 | return 0; | |
7130 | } | |
7131 | ||
7c673cae | 7132 | int BlueStore::_fsck_check_extents( |
11fdf7f2 | 7133 | const coll_t& cid, |
7c673cae FG |
7134 | const ghobject_t& oid, |
7135 | const PExtentVector& extents, | |
7136 | bool compressed, | |
7137 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 7138 | uint64_t granularity, |
11fdf7f2 | 7139 | BlueStoreRepairer* repairer, |
eafe8130 TL |
7140 | store_statfs_t& expected_statfs, |
7141 | FSCKDepth depth) | |
7c673cae FG |
7142 | { |
7143 | dout(30) << __func__ << " oid " << oid << " extents " << extents << dendl; | |
7144 | int errors = 0; | |
7145 | for (auto e : extents) { | |
7146 | if (!e.is_valid()) | |
7147 | continue; | |
7148 | expected_statfs.allocated += e.length; | |
7149 | if (compressed) { | |
11fdf7f2 | 7150 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae | 7151 | } |
eafe8130 TL |
7152 | if (depth != FSCK_SHALLOW) { |
7153 | bool already = false; | |
9f95a23c | 7154 | apply_for_bitset_range( |
eafe8130 TL |
7155 | e.offset, e.length, granularity, used_blocks, |
7156 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
7157 | if (bs.test(pos)) { |
7158 | if (repairer) { | |
7159 | repairer->note_misreference( | |
7160 | pos * min_alloc_size, min_alloc_size, !already); | |
7161 | } | |
7162 | if (!already) { | |
7163 | derr << "fsck error: " << oid << " extent " << e | |
7164 | << " or a subset is already allocated (misreferenced)" << dendl; | |
7165 | ++errors; | |
7166 | already = true; | |
7167 | } | |
11fdf7f2 | 7168 | } |
eafe8130 TL |
7169 | else |
7170 | bs.set(pos); | |
7171 | }); | |
7172 | if (repairer) { | |
b3b6e05e | 7173 | repairer->set_space_used(e.offset, e.length, cid, oid); |
eafe8130 | 7174 | } |
11fdf7f2 | 7175 | |
eafe8130 TL |
7176 | if (e.end() > bdev->get_size()) { |
7177 | derr << "fsck error: " << oid << " extent " << e | |
7178 | << " past end of block device" << dendl; | |
7179 | ++errors; | |
7180 | } | |
7c673cae FG |
7181 | } |
7182 | } | |
7183 | return errors; | |
7184 | } | |
7185 | ||
11fdf7f2 TL |
7186 | void BlueStore::_fsck_check_pool_statfs( |
7187 | BlueStore::per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
7188 | int64_t& errors, |
7189 | int64_t& warnings, | |
11fdf7f2 TL |
7190 | BlueStoreRepairer* repairer) |
7191 | { | |
f67539c2 | 7192 | auto it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 TL |
7193 | if (it) { |
7194 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7195 | string key = it->key(); | |
7196 | if (key == BLUESTORE_GLOBAL_STATFS_KEY) { | |
7197 | if (repairer) { | |
eafe8130 TL |
7198 | ++errors; |
7199 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
7200 | derr << "fsck error: " << "legacy statfs record found, removing" | |
11fdf7f2 TL |
7201 | << dendl; |
7202 | } | |
7203 | continue; | |
7204 | } | |
11fdf7f2 TL |
7205 | uint64_t pool_id; |
7206 | if (get_key_pool_stat(key, &pool_id) < 0) { | |
7207 | derr << "fsck error: bad key " << key | |
7208 | << "in statfs namespece" << dendl; | |
7209 | if (repairer) { | |
7210 | repairer->remove_key(db, PREFIX_STAT, key); | |
7211 | } | |
7212 | ++errors; | |
7213 | continue; | |
7214 | } | |
7215 | ||
7216 | volatile_statfs vstatfs; | |
7217 | bufferlist bl = it->value(); | |
7218 | auto blp = bl.cbegin(); | |
7219 | try { | |
7220 | vstatfs.decode(blp); | |
f67539c2 | 7221 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
7222 | derr << "fsck error: failed to decode Pool StatFS record" |
7223 | << pretty_binary_string(key) << dendl; | |
7224 | if (repairer) { | |
7225 | dout(20) << __func__ << " undecodable Pool StatFS record, key:'" | |
7226 | << pretty_binary_string(key) | |
7227 | << "', removing" << dendl; | |
7228 | repairer->remove_key(db, PREFIX_STAT, key); | |
7229 | } | |
7230 | ++errors; | |
7231 | vstatfs.reset(); | |
7232 | } | |
7233 | auto stat_it = expected_pool_statfs.find(pool_id); | |
7234 | if (stat_it == expected_pool_statfs.end()) { | |
7235 | if (vstatfs.is_empty()) { | |
7236 | // we don't consider that as an error since empty pool statfs | |
7237 | // are left in DB for now | |
7238 | dout(20) << "fsck inf: found empty stray Pool StatFS record for pool id 0x" | |
7239 | << std::hex << pool_id << std::dec << dendl; | |
7240 | if (repairer) { | |
7241 | // but we need to increment error count in case of repair | |
7242 | // to have proper counters at the end | |
7243 | // (as repairer increments recovery counter anyway). | |
7244 | ++errors; | |
7245 | } | |
7246 | } else { | |
7247 | derr << "fsck error: found stray Pool StatFS record for pool id 0x" | |
7248 | << std::hex << pool_id << std::dec << dendl; | |
7249 | ++errors; | |
7250 | } | |
7251 | if (repairer) { | |
7252 | repairer->remove_key(db, PREFIX_SHARED_BLOB, key); | |
7253 | } | |
7254 | continue; | |
7255 | } | |
7256 | store_statfs_t statfs; | |
7257 | vstatfs.publish(&statfs); | |
7258 | if (!(stat_it->second == statfs)) { | |
7259 | derr << "fsck error: actual " << statfs | |
7260 | << " != expected " << stat_it->second | |
7261 | << " for pool " | |
7262 | << std::hex << pool_id << std::dec << dendl; | |
7263 | if (repairer) { | |
7264 | repairer->fix_statfs(db, key, stat_it->second); | |
7265 | } | |
7266 | ++errors; | |
7267 | } | |
7268 | expected_pool_statfs.erase(stat_it); | |
7269 | } | |
7270 | } // if (it) | |
eafe8130 TL |
7271 | for (auto& s : expected_pool_statfs) { |
7272 | if (s.second.is_zero()) { | |
11fdf7f2 TL |
7273 | // we might lack empty statfs recs in DB |
7274 | continue; | |
7275 | } | |
7276 | derr << "fsck error: missing Pool StatFS record for pool " | |
eafe8130 | 7277 | << std::hex << s.first << std::dec << dendl; |
11fdf7f2 TL |
7278 | if (repairer) { |
7279 | string key; | |
eafe8130 TL |
7280 | get_pool_stat_key(s.first, &key); |
7281 | repairer->fix_statfs(db, key, s.second); | |
11fdf7f2 TL |
7282 | } |
7283 | ++errors; | |
7284 | } | |
eafe8130 | 7285 | if (!per_pool_stat_collection && |
eafe8130 TL |
7286 | repairer) { |
7287 | // by virtue of running this method, we correct the top-level | |
7288 | // error of having global stats | |
7289 | repairer->inc_repaired(); | |
7290 | } | |
11fdf7f2 TL |
7291 | } |
7292 | ||
eafe8130 TL |
7293 | BlueStore::OnodeRef BlueStore::fsck_check_objects_shallow( |
7294 | BlueStore::FSCKDepth depth, | |
7295 | int64_t pool_id, | |
7296 | BlueStore::CollectionRef c, | |
7297 | const ghobject_t& oid, | |
7298 | const string& key, | |
7299 | const bufferlist& value, | |
9f95a23c | 7300 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
7301 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
7302 | const BlueStore::FSCK_ObjectCtx& ctx) | |
7303 | { | |
7304 | auto& errors = ctx.errors; | |
7305 | auto& num_objects = ctx.num_objects; | |
7306 | auto& num_extents = ctx.num_extents; | |
7307 | auto& num_blobs = ctx.num_blobs; | |
7308 | auto& num_sharded_objects = ctx.num_sharded_objects; | |
7309 | auto& num_spanning_blobs = ctx.num_spanning_blobs; | |
7310 | auto used_blocks = ctx.used_blocks; | |
7311 | auto sb_info_lock = ctx.sb_info_lock; | |
7312 | auto& sb_info = ctx.sb_info; | |
7313 | auto repairer = ctx.repairer; | |
7314 | ||
7315 | store_statfs_t* res_statfs = (per_pool_stat_collection || repairer) ? | |
7316 | &ctx.expected_pool_statfs[pool_id] : | |
7317 | &ctx.expected_store_statfs; | |
7318 | ||
7319 | dout(10) << __func__ << " " << oid << dendl; | |
7320 | OnodeRef o; | |
7321 | o.reset(Onode::decode(c, oid, key, value)); | |
7322 | ++num_objects; | |
7c673cae | 7323 | |
eafe8130 | 7324 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); |
7c673cae | 7325 | |
eafe8130 TL |
7326 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
7327 | _dump_onode<30>(cct, *o); | |
7328 | // shards | |
7329 | if (!o->extent_map.shards.empty()) { | |
7330 | ++num_sharded_objects; | |
7331 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 7332 | ceph_assert(expecting_shards); |
eafe8130 TL |
7333 | for (auto& s : o->extent_map.shards) { |
7334 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
9f95a23c | 7335 | expecting_shards->push_back(string()); |
eafe8130 | 7336 | get_extent_shard_key(o->key, s.shard_info->offset, |
9f95a23c | 7337 | &expecting_shards->back()); |
eafe8130 TL |
7338 | if (s.shard_info->offset >= o->onode.size) { |
7339 | derr << "fsck error: " << oid << " shard 0x" << std::hex | |
7340 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size | |
7341 | << std::dec << dendl; | |
7342 | ++errors; | |
7343 | } | |
7344 | } | |
7345 | } | |
7346 | } | |
7c673cae | 7347 | |
eafe8130 TL |
7348 | // lextents |
7349 | uint64_t pos = 0; | |
7350 | mempool::bluestore_fsck::map<BlobRef, | |
7351 | bluestore_blob_use_tracker_t> ref_map; | |
7352 | for (auto& l : o->extent_map.extent_map) { | |
7353 | dout(20) << __func__ << " " << l << dendl; | |
7354 | if (l.logical_offset < pos) { | |
7355 | derr << "fsck error: " << oid << " lextent at 0x" | |
7356 | << std::hex << l.logical_offset | |
7357 | << " overlaps with the previous, which ends at 0x" << pos | |
7358 | << std::dec << dendl; | |
7359 | ++errors; | |
7360 | } | |
7361 | if (depth != FSCK_SHALLOW && | |
7362 | o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
7363 | derr << "fsck error: " << oid << " lextent at 0x" | |
7364 | << std::hex << l.logical_offset << "~" << l.length | |
7365 | << " spans a shard boundary" | |
7366 | << std::dec << dendl; | |
7367 | ++errors; | |
7368 | } | |
7369 | pos = l.logical_offset + l.length; | |
7370 | res_statfs->data_stored += l.length; | |
7371 | ceph_assert(l.blob); | |
7372 | const bluestore_blob_t& blob = l.blob->get_blob(); | |
7373 | ||
7374 | auto& ref = ref_map[l.blob]; | |
7375 | if (ref.is_empty()) { | |
7376 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
7377 | uint32_t l = blob.get_logical_length(); | |
7378 | ref.init(l, min_release_size); | |
7379 | } | |
7380 | ref.get( | |
7381 | l.blob_offset, | |
7382 | l.length); | |
7383 | ++num_extents; | |
7384 | if (depth != FSCK_SHALLOW && | |
7385 | blob.has_unused()) { | |
7386 | ceph_assert(referenced); | |
7387 | auto p = referenced->find(l.blob); | |
7388 | bluestore_blob_t::unused_t* pu; | |
7389 | if (p == referenced->end()) { | |
7390 | pu = &(*referenced)[l.blob]; | |
7391 | } | |
7392 | else { | |
7393 | pu = &p->second; | |
7394 | } | |
7395 | uint64_t blob_len = blob.get_logical_length(); | |
7396 | ceph_assert((blob_len % (sizeof(*pu) * 8)) == 0); | |
7397 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
7398 | uint64_t chunk_size = blob_len / (sizeof(*pu) * 8); | |
7399 | uint64_t start = l.blob_offset / chunk_size; | |
7400 | uint64_t end = | |
7401 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; | |
7402 | for (auto i = start; i < end; ++i) { | |
7403 | (*pu) |= (1u << i); | |
7404 | } | |
7405 | } | |
7406 | } //for (auto& l : o->extent_map.extent_map) | |
7407 | ||
7408 | for (auto& i : ref_map) { | |
7409 | ++num_blobs; | |
7410 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7411 | bool equal = | |
7412 | depth == FSCK_SHALLOW ? true : | |
7413 | i.first->get_blob_use_tracker().equal(i.second); | |
7414 | if (!equal) { | |
7415 | derr << "fsck error: " << oid << " blob " << *i.first | |
7416 | << " doesn't match expected ref_map " << i.second << dendl; | |
7417 | ++errors; | |
7418 | } | |
7419 | if (blob.is_compressed()) { | |
7420 | res_statfs->data_compressed += blob.get_compressed_payload_length(); | |
7421 | res_statfs->data_compressed_original += | |
7422 | i.first->get_referenced_bytes(); | |
7423 | } | |
7424 | if (blob.is_shared()) { | |
7425 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
7426 | derr << "fsck error: " << oid << " blob " << blob | |
7427 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " | |
7428 | << blobid_max << dendl; | |
7429 | ++errors; | |
7430 | } | |
7431 | else if (i.first->shared_blob->get_sbid() == 0) { | |
7432 | derr << "fsck error: " << oid << " blob " << blob | |
7433 | << " marked as shared but has uninitialized sbid" | |
7434 | << dendl; | |
7435 | ++errors; | |
7436 | } | |
7437 | // the below lock is optional and provided in multithreading mode only | |
7438 | if (sb_info_lock) { | |
7439 | sb_info_lock->lock(); | |
7440 | } | |
7441 | sb_info_t& sbi = sb_info[i.first->shared_blob->get_sbid()]; | |
7442 | ceph_assert(sbi.cid == coll_t() || sbi.cid == c->cid); | |
7443 | ceph_assert(sbi.pool_id == INT64_MIN || | |
7444 | sbi.pool_id == oid.hobj.get_logical_pool()); | |
7445 | sbi.cid = c->cid; | |
7446 | sbi.pool_id = oid.hobj.get_logical_pool(); | |
7447 | sbi.sb = i.first->shared_blob; | |
7448 | sbi.oids.push_back(oid); | |
7449 | sbi.compressed = blob.is_compressed(); | |
7450 | for (auto e : blob.get_extents()) { | |
7451 | if (e.is_valid()) { | |
7452 | sbi.ref_map.get(e.offset, e.length); | |
7453 | } | |
7454 | } | |
7455 | if (sb_info_lock) { | |
7456 | sb_info_lock->unlock(); | |
7457 | } | |
7458 | } else if (depth != FSCK_SHALLOW) { | |
7459 | ceph_assert(used_blocks); | |
7460 | errors += _fsck_check_extents(c->cid, oid, blob.get_extents(), | |
7461 | blob.is_compressed(), | |
7462 | *used_blocks, | |
7463 | fm->get_alloc_size(), | |
7464 | repairer, | |
7465 | *res_statfs, | |
7466 | depth); | |
7467 | } else { | |
7468 | errors += _fsck_sum_extents( | |
7469 | blob.get_extents(), | |
7470 | blob.is_compressed(), | |
7471 | *res_statfs); | |
7472 | } | |
7473 | } // for (auto& i : ref_map) | |
9f95a23c | 7474 | |
adb31ebb TL |
7475 | { |
7476 | auto &sbm = o->extent_map.spanning_blob_map; | |
7477 | size_t broken = 0; | |
7478 | BlobRef first_broken; | |
7479 | for (auto it = sbm.begin(); it != sbm.end();) { | |
7480 | auto it1 = it++; | |
7481 | if (ref_map.count(it1->second) == 0) { | |
7482 | if (!broken) { | |
7483 | first_broken = it1->second; | |
7484 | ++errors; | |
7485 | } | |
7486 | broken++; | |
7487 | if (repairer) { | |
7488 | sbm.erase(it1); | |
7489 | } | |
7490 | } | |
7491 | } | |
7492 | if (broken) { | |
7493 | derr << "fsck error: " << oid << " - " << broken | |
7494 | << " zombie spanning blob(s) found, the first one: " | |
7495 | << *first_broken << dendl; | |
7496 | if(repairer) { | |
b3b6e05e TL |
7497 | repairer->fix_spanning_blobs( |
7498 | db, | |
7499 | [&](KeyValueDB::Transaction txn) { | |
7500 | _record_onode(o, txn); | |
7501 | }); | |
adb31ebb TL |
7502 | } |
7503 | } | |
7504 | } | |
7505 | ||
9f95a23c TL |
7506 | if (o->onode.has_omap()) { |
7507 | _fsck_check_object_omap(depth, o, ctx); | |
7508 | } | |
7509 | ||
eafe8130 TL |
7510 | return o; |
7511 | } | |
7512 | ||
7513 | #include "common/WorkQueue.h" | |
7514 | ||
7515 | class ShallowFSCKThreadPool : public ThreadPool | |
7516 | { | |
7517 | public: | |
7518 | ShallowFSCKThreadPool(CephContext* cct_, std::string nm, std::string tn, int n) : | |
7519 | ThreadPool(cct_, nm, tn, n) { | |
7520 | } | |
7521 | void worker(ThreadPool::WorkThread* wt) override { | |
7522 | int next_wq = 0; | |
7523 | while (!_stop) { | |
7524 | next_wq %= work_queues.size(); | |
7525 | WorkQueue_ *wq = work_queues[next_wq++]; | |
7526 | ||
7527 | void* item = wq->_void_dequeue(); | |
7528 | if (item) { | |
7529 | processing++; | |
7530 | TPHandle tp_handle(cct, nullptr, wq->timeout_interval, wq->suicide_interval); | |
7531 | wq->_void_process(item, tp_handle); | |
7532 | processing--; | |
7533 | } | |
7534 | } | |
7535 | } | |
7536 | template <size_t BatchLen> | |
7537 | struct FSCKWorkQueue : public ThreadPool::WorkQueue_ | |
7538 | { | |
7539 | struct Entry { | |
7540 | int64_t pool_id; | |
7541 | BlueStore::CollectionRef c; | |
7542 | ghobject_t oid; | |
7543 | string key; | |
7544 | bufferlist value; | |
7545 | }; | |
7546 | struct Batch { | |
7547 | std::atomic<size_t> running = { 0 }; | |
7548 | size_t entry_count = 0; | |
7549 | std::array<Entry, BatchLen> entries; | |
7550 | ||
7551 | int64_t errors = 0; | |
7552 | int64_t warnings = 0; | |
7553 | uint64_t num_objects = 0; | |
7554 | uint64_t num_extents = 0; | |
7555 | uint64_t num_blobs = 0; | |
7556 | uint64_t num_sharded_objects = 0; | |
7557 | uint64_t num_spanning_blobs = 0; | |
7558 | store_statfs_t expected_store_statfs; | |
7559 | BlueStore::per_pool_statfs expected_pool_statfs; | |
7560 | }; | |
7561 | ||
7562 | size_t batchCount; | |
7563 | BlueStore* store = nullptr; | |
7564 | ||
eafe8130 TL |
7565 | ceph::mutex* sb_info_lock = nullptr; |
7566 | BlueStore::sb_info_map_t* sb_info = nullptr; | |
7567 | BlueStoreRepairer* repairer = nullptr; | |
7568 | ||
7569 | Batch* batches = nullptr; | |
7570 | size_t last_batch_pos = 0; | |
7571 | bool batch_acquired = false; | |
7572 | ||
7573 | FSCKWorkQueue(std::string n, | |
7574 | size_t _batchCount, | |
7575 | BlueStore* _store, | |
eafe8130 TL |
7576 | ceph::mutex* _sb_info_lock, |
7577 | BlueStore::sb_info_map_t& _sb_info, | |
7578 | BlueStoreRepairer* _repairer) : | |
f67539c2 | 7579 | WorkQueue_(n, ceph::timespan::zero(), ceph::timespan::zero()), |
eafe8130 TL |
7580 | batchCount(_batchCount), |
7581 | store(_store), | |
eafe8130 TL |
7582 | sb_info_lock(_sb_info_lock), |
7583 | sb_info(&_sb_info), | |
7584 | repairer(_repairer) | |
7585 | { | |
7586 | batches = new Batch[batchCount]; | |
7587 | } | |
7588 | ~FSCKWorkQueue() { | |
7589 | delete[] batches; | |
7590 | } | |
7591 | ||
7592 | /// Remove all work items from the queue. | |
7593 | void _clear() override { | |
7594 | //do nothing | |
7595 | } | |
7596 | /// Check whether there is anything to do. | |
7597 | bool _empty() override { | |
7598 | ceph_assert(false); | |
7599 | } | |
7600 | ||
7601 | /// Get the next work item to process. | |
7602 | void* _void_dequeue() override { | |
7603 | size_t pos = rand() % batchCount; | |
7604 | size_t pos0 = pos; | |
7605 | do { | |
7606 | auto& batch = batches[pos]; | |
7607 | if (batch.running.fetch_add(1) == 0) { | |
7608 | if (batch.entry_count) { | |
7609 | return &batch; | |
7610 | } | |
7611 | } | |
7612 | batch.running--; | |
7613 | pos++; | |
7614 | pos %= batchCount; | |
7615 | } while (pos != pos0); | |
7616 | return nullptr; | |
7617 | } | |
7618 | /** @brief Process the work item. | |
7619 | * This function will be called several times in parallel | |
7620 | * and must therefore be thread-safe. */ | |
7621 | void _void_process(void* item, TPHandle& handle) override { | |
7622 | Batch* batch = (Batch*)item; | |
7623 | ||
7624 | BlueStore::FSCK_ObjectCtx ctx( | |
7625 | batch->errors, | |
7626 | batch->warnings, | |
7627 | batch->num_objects, | |
7628 | batch->num_extents, | |
7629 | batch->num_blobs, | |
7630 | batch->num_sharded_objects, | |
7631 | batch->num_spanning_blobs, | |
7632 | nullptr, // used_blocks | |
9f95a23c | 7633 | nullptr, //used_omap_head |
eafe8130 TL |
7634 | sb_info_lock, |
7635 | *sb_info, | |
7636 | batch->expected_store_statfs, | |
7637 | batch->expected_pool_statfs, | |
7638 | repairer); | |
7639 | ||
7640 | for (size_t i = 0; i < batch->entry_count; i++) { | |
7641 | auto& entry = batch->entries[i]; | |
7642 | ||
7643 | store->fsck_check_objects_shallow( | |
7644 | BlueStore::FSCK_SHALLOW, | |
7645 | entry.pool_id, | |
7646 | entry.c, | |
7647 | entry.oid, | |
7648 | entry.key, | |
7649 | entry.value, | |
9f95a23c | 7650 | nullptr, // expecting_shards - this will need a protection if passed |
eafe8130 TL |
7651 | nullptr, // referenced |
7652 | ctx); | |
7653 | } | |
7654 | //std::cout << "processed " << batch << std::endl; | |
7655 | batch->entry_count = 0; | |
7656 | batch->running--; | |
7657 | } | |
7658 | /** @brief Synchronously finish processing a work item. | |
7659 | * This function is called after _void_process with the global thread pool lock held, | |
7660 | * so at most one copy will execute simultaneously for a given thread pool. | |
7661 | * It can be used for non-thread-safe finalization. */ | |
7662 | void _void_process_finish(void*) override { | |
7663 | ceph_assert(false); | |
7664 | } | |
7665 | ||
7666 | bool queue( | |
7667 | int64_t pool_id, | |
7668 | BlueStore::CollectionRef c, | |
7669 | const ghobject_t& oid, | |
7670 | const string& key, | |
7671 | const bufferlist& value) { | |
7672 | bool res = false; | |
7673 | size_t pos0 = last_batch_pos; | |
7674 | if (!batch_acquired) { | |
7675 | do { | |
7676 | auto& batch = batches[last_batch_pos]; | |
7677 | if (batch.running.fetch_add(1) == 0) { | |
7678 | if (batch.entry_count < BatchLen) { | |
7679 | batch_acquired = true; | |
7680 | break; | |
7681 | } | |
7682 | } | |
7683 | batch.running.fetch_sub(1); | |
7684 | last_batch_pos++; | |
7685 | last_batch_pos %= batchCount; | |
7686 | } while (last_batch_pos != pos0); | |
7687 | } | |
7688 | if (batch_acquired) { | |
7689 | auto& batch = batches[last_batch_pos]; | |
7690 | ceph_assert(batch.running); | |
7691 | ceph_assert(batch.entry_count < BatchLen); | |
7692 | ||
7693 | auto& entry = batch.entries[batch.entry_count]; | |
7694 | entry.pool_id = pool_id; | |
7695 | entry.c = c; | |
7696 | entry.oid = oid; | |
7697 | entry.key = key; | |
7698 | entry.value = value; | |
7699 | ||
7700 | ++batch.entry_count; | |
7701 | if (batch.entry_count == BatchLen) { | |
7702 | batch_acquired = false; | |
7703 | batch.running.fetch_sub(1); | |
7704 | last_batch_pos++; | |
7705 | last_batch_pos %= batchCount; | |
7706 | } | |
7707 | res = true; | |
7708 | } | |
7709 | return res; | |
7710 | } | |
7711 | ||
7712 | void finalize(ThreadPool& tp, | |
7713 | BlueStore::FSCK_ObjectCtx& ctx) { | |
7714 | if (batch_acquired) { | |
7715 | auto& batch = batches[last_batch_pos]; | |
7716 | ceph_assert(batch.running); | |
7717 | batch.running.fetch_sub(1); | |
7718 | } | |
7719 | tp.stop(); | |
7720 | ||
7721 | for (size_t i = 0; i < batchCount; i++) { | |
7722 | auto& batch = batches[i]; | |
7723 | ||
7724 | //process leftovers if any | |
7725 | if (batch.entry_count) { | |
7726 | TPHandle tp_handle(store->cct, | |
7727 | nullptr, | |
7728 | timeout_interval, | |
7729 | suicide_interval); | |
7730 | ceph_assert(batch.running == 0); | |
7731 | ||
7732 | batch.running++; // just to be on-par with the regular call | |
7733 | _void_process(&batch, tp_handle); | |
7734 | } | |
7735 | ceph_assert(batch.entry_count == 0); | |
7736 | ||
7737 | ctx.errors += batch.errors; | |
7738 | ctx.warnings += batch.warnings; | |
7739 | ctx.num_objects += batch.num_objects; | |
7740 | ctx.num_extents += batch.num_extents; | |
7741 | ctx.num_blobs += batch.num_blobs; | |
7742 | ctx.num_sharded_objects += batch.num_sharded_objects; | |
7743 | ctx.num_spanning_blobs += batch.num_spanning_blobs; | |
9f95a23c | 7744 | |
eafe8130 TL |
7745 | ctx.expected_store_statfs.add(batch.expected_store_statfs); |
7746 | ||
7747 | for (auto it = batch.expected_pool_statfs.begin(); | |
7748 | it != batch.expected_pool_statfs.end(); | |
7749 | it++) { | |
7750 | ctx.expected_pool_statfs[it->first].add(it->second); | |
7751 | } | |
7752 | } | |
7753 | } | |
7754 | }; | |
7755 | }; | |
7756 | ||
9f95a23c TL |
7757 | void BlueStore::_fsck_check_object_omap(FSCKDepth depth, |
7758 | OnodeRef& o, | |
7759 | const BlueStore::FSCK_ObjectCtx& ctx) | |
eafe8130 | 7760 | { |
9f95a23c TL |
7761 | auto& errors = ctx.errors; |
7762 | auto& warnings = ctx.warnings; | |
7763 | auto repairer = ctx.repairer; | |
7764 | ||
7765 | ceph_assert(o->onode.has_omap()); | |
7766 | if (!o->onode.is_perpool_omap() && !o->onode.is_pgmeta_omap()) { | |
f67539c2 | 7767 | if (per_pool_omap == OMAP_PER_POOL) { |
9f95a23c TL |
7768 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
7769 | << "fsck error: " << o->oid | |
7770 | << " has omap that is not per-pool or pgmeta" | |
7771 | << fsck_dendl; | |
7772 | ++errors; | |
7773 | } else { | |
7774 | const char* w; | |
7775 | int64_t num; | |
7776 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
7777 | ++errors; | |
7778 | num = errors; | |
7779 | w = "error"; | |
7780 | } else { | |
7781 | ++warnings; | |
7782 | num = warnings; | |
7783 | w = "warning"; | |
7784 | } | |
7785 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
7786 | << "fsck " << w << ": " << o->oid | |
7787 | << " has omap that is not per-pool or pgmeta" | |
7788 | << fsck_dendl; | |
7789 | } | |
f67539c2 TL |
7790 | } else if (!o->onode.is_perpg_omap() && !o->onode.is_pgmeta_omap()) { |
7791 | if (per_pool_omap == OMAP_PER_PG) { | |
7792 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
7793 | << "fsck error: " << o->oid | |
7794 | << " has omap that is not per-pg or pgmeta" | |
7795 | << fsck_dendl; | |
7796 | ++errors; | |
7797 | } else { | |
7798 | const char* w; | |
7799 | int64_t num; | |
7800 | if (cct->_conf->bluestore_fsck_error_on_no_per_pg_omap) { | |
7801 | ++errors; | |
7802 | num = errors; | |
7803 | w = "error"; | |
7804 | } else { | |
7805 | ++warnings; | |
7806 | num = warnings; | |
7807 | w = "warning"; | |
7808 | } | |
7809 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
7810 | << "fsck " << w << ": " << o->oid | |
7811 | << " has omap that is not per-pg or pgmeta" | |
7812 | << fsck_dendl; | |
7813 | } | |
9f95a23c TL |
7814 | } |
7815 | if (repairer && | |
f67539c2 | 7816 | !o->onode.is_perpg_omap() && |
9f95a23c | 7817 | !o->onode.is_pgmeta_omap()) { |
f67539c2 | 7818 | dout(10) << "fsck converting " << o->oid << " omap to per-pg" << dendl; |
9f95a23c TL |
7819 | bufferlist h; |
7820 | map<string, bufferlist> kv; | |
7821 | int r = _onode_omap_get(o, &h, &kv); | |
7822 | if (r < 0) { | |
7823 | derr << " got " << r << " " << cpp_strerror(r) << dendl; | |
7824 | } else { | |
7825 | KeyValueDB::Transaction txn = db->get_transaction(); | |
7826 | // remove old keys | |
7827 | const string& old_omap_prefix = o->get_omap_prefix(); | |
7828 | string old_head, old_tail; | |
7829 | o->get_omap_header(&old_head); | |
7830 | o->get_omap_tail(&old_tail); | |
7831 | txn->rm_range_keys(old_omap_prefix, old_head, old_tail); | |
7832 | txn->rmkey(old_omap_prefix, old_tail); | |
7833 | // set flag | |
f67539c2 | 7834 | o->onode.set_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP | bluestore_onode_t::FLAG_PERPG_OMAP); |
9f95a23c TL |
7835 | _record_onode(o, txn); |
7836 | const string& new_omap_prefix = o->get_omap_prefix(); | |
7837 | // head | |
7838 | if (h.length()) { | |
7839 | string new_head; | |
7840 | o->get_omap_header(&new_head); | |
7841 | txn->set(new_omap_prefix, new_head, h); | |
7842 | } | |
7843 | // tail | |
7844 | string new_tail; | |
7845 | o->get_omap_tail(&new_tail); | |
7846 | bufferlist empty; | |
7847 | txn->set(new_omap_prefix, new_tail, empty); | |
7848 | // values | |
7849 | string final_key; | |
7850 | o->get_omap_key(string(), &final_key); | |
7851 | size_t base_key_len = final_key.size(); | |
7852 | for (auto& i : kv) { | |
7853 | final_key.resize(base_key_len); | |
7854 | final_key += i.first; | |
7855 | txn->set(new_omap_prefix, final_key, i.second); | |
7856 | } | |
7857 | db->submit_transaction_sync(txn); | |
7858 | repairer->inc_repaired(); | |
7859 | } | |
eafe8130 | 7860 | } |
9f95a23c | 7861 | } |
eafe8130 | 7862 | |
9f95a23c TL |
7863 | void BlueStore::_fsck_check_objects(FSCKDepth depth, |
7864 | BlueStore::FSCK_ObjectCtx& ctx) | |
7865 | { | |
eafe8130 | 7866 | auto& errors = ctx.errors; |
eafe8130 TL |
7867 | auto sb_info_lock = ctx.sb_info_lock; |
7868 | auto& sb_info = ctx.sb_info; | |
7869 | auto repairer = ctx.repairer; | |
7870 | ||
7871 | uint64_t_btree_t used_nids; | |
7872 | ||
7873 | size_t processed_myself = 0; | |
7874 | ||
f67539c2 | 7875 | auto it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
7876 | mempool::bluestore_fsck::list<string> expecting_shards; |
7877 | if (it) { | |
7878 | const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads; | |
7879 | typedef ShallowFSCKThreadPool::FSCKWorkQueue<256> WQ; | |
7880 | std::unique_ptr<WQ> wq( | |
7881 | new WQ( | |
7882 | "FSCKWorkQueue", | |
7883 | (thread_count ? : 1) * 32, | |
7884 | this, | |
eafe8130 TL |
7885 | sb_info_lock, |
7886 | sb_info, | |
7887 | repairer)); | |
7888 | ||
7889 | ShallowFSCKThreadPool thread_pool(cct, "ShallowFSCKThreadPool", "ShallowFSCK", thread_count); | |
7890 | ||
7891 | thread_pool.add_work_queue(wq.get()); | |
7892 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
7893 | //not the best place but let's check anyway | |
7894 | ceph_assert(sb_info_lock); | |
7895 | thread_pool.start(); | |
7896 | } | |
7897 | ||
7898 | //fill global if not overriden below | |
7899 | CollectionRef c; | |
7900 | int64_t pool_id = -1; | |
7901 | spg_t pgid; | |
7902 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7903 | dout(30) << __func__ << " key " | |
7904 | << pretty_binary_string(it->key()) << dendl; | |
7905 | if (is_extent_shard_key(it->key())) { | |
7906 | if (depth == FSCK_SHALLOW) { | |
7907 | continue; | |
7908 | } | |
7909 | while (!expecting_shards.empty() && | |
7910 | expecting_shards.front() < it->key()) { | |
7911 | derr << "fsck error: missing shard key " | |
7912 | << pretty_binary_string(expecting_shards.front()) | |
7913 | << dendl; | |
7914 | ++errors; | |
7915 | expecting_shards.pop_front(); | |
7916 | } | |
7917 | if (!expecting_shards.empty() && | |
7918 | expecting_shards.front() == it->key()) { | |
7919 | // all good | |
7920 | expecting_shards.pop_front(); | |
7921 | continue; | |
7922 | } | |
7923 | ||
7924 | uint32_t offset; | |
7925 | string okey; | |
7926 | get_key_extent_shard(it->key(), &okey, &offset); | |
7927 | derr << "fsck error: stray shard 0x" << std::hex << offset | |
7928 | << std::dec << dendl; | |
7929 | if (expecting_shards.empty()) { | |
7930 | derr << "fsck error: " << pretty_binary_string(it->key()) | |
7931 | << " is unexpected" << dendl; | |
7932 | ++errors; | |
7933 | continue; | |
7934 | } | |
7935 | while (expecting_shards.front() > it->key()) { | |
7936 | derr << "fsck error: saw " << pretty_binary_string(it->key()) | |
7937 | << dendl; | |
7938 | derr << "fsck error: exp " | |
7939 | << pretty_binary_string(expecting_shards.front()) << dendl; | |
7940 | ++errors; | |
7941 | expecting_shards.pop_front(); | |
7942 | if (expecting_shards.empty()) { | |
7943 | break; | |
7944 | } | |
7945 | } | |
7946 | continue; | |
7947 | } | |
7948 | ||
7949 | ghobject_t oid; | |
7950 | int r = get_key_object(it->key(), &oid); | |
7951 | if (r < 0) { | |
7952 | derr << "fsck error: bad object key " | |
7953 | << pretty_binary_string(it->key()) << dendl; | |
7954 | ++errors; | |
7955 | continue; | |
7956 | } | |
7957 | if (!c || | |
7958 | oid.shard_id != pgid.shard || | |
7959 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
7960 | !c->contains(oid)) { | |
7961 | c = nullptr; | |
7962 | for (auto& p : coll_map) { | |
7963 | if (p.second->contains(oid)) { | |
7964 | c = p.second; | |
7965 | break; | |
7966 | } | |
7967 | } | |
7968 | if (!c) { | |
7969 | derr << "fsck error: stray object " << oid | |
7970 | << " not owned by any collection" << dendl; | |
7971 | ++errors; | |
7972 | continue; | |
7973 | } | |
7974 | pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
7975 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
7976 | << dendl; | |
7977 | } | |
7978 | ||
7979 | if (depth != FSCK_SHALLOW && | |
7980 | !expecting_shards.empty()) { | |
7981 | for (auto& k : expecting_shards) { | |
7982 | derr << "fsck error: missing shard key " | |
7983 | << pretty_binary_string(k) << dendl; | |
7984 | } | |
7985 | ++errors; | |
7986 | expecting_shards.clear(); | |
7987 | } | |
7988 | ||
7989 | bool queued = false; | |
7990 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
7991 | queued = wq->queue( | |
7992 | pool_id, | |
7993 | c, | |
7994 | oid, | |
7995 | it->key(), | |
7996 | it->value()); | |
7997 | } | |
7998 | OnodeRef o; | |
7999 | map<BlobRef, bluestore_blob_t::unused_t> referenced; | |
8000 | ||
8001 | if (!queued) { | |
8002 | ++processed_myself; | |
8003 | ||
8004 | o = fsck_check_objects_shallow( | |
8005 | depth, | |
8006 | pool_id, | |
8007 | c, | |
8008 | oid, | |
8009 | it->key(), | |
8010 | it->value(), | |
9f95a23c | 8011 | &expecting_shards, |
eafe8130 TL |
8012 | &referenced, |
8013 | ctx); | |
8014 | } | |
8015 | ||
8016 | if (depth != FSCK_SHALLOW) { | |
8017 | ceph_assert(o != nullptr); | |
8018 | if (o->onode.nid) { | |
8019 | if (o->onode.nid > nid_max) { | |
8020 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8021 | << " > nid_max " << nid_max << dendl; | |
8022 | ++errors; | |
8023 | } | |
8024 | if (used_nids.count(o->onode.nid)) { | |
8025 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8026 | << " already in use" << dendl; | |
8027 | ++errors; | |
8028 | continue; // go for next object | |
8029 | } | |
8030 | used_nids.insert(o->onode.nid); | |
8031 | } | |
8032 | for (auto& i : referenced) { | |
8033 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
8034 | << std::dec << " for " << *i.first << dendl; | |
8035 | const bluestore_blob_t& blob = i.first->get_blob(); | |
8036 | if (i.second & blob.unused) { | |
8037 | derr << "fsck error: " << oid << " blob claims unused 0x" | |
8038 | << std::hex << blob.unused | |
8039 | << " but extents reference 0x" << i.second << std::dec | |
8040 | << " on blob " << *i.first << dendl; | |
8041 | ++errors; | |
8042 | } | |
8043 | if (blob.has_csum()) { | |
8044 | uint64_t blob_len = blob.get_logical_length(); | |
8045 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused) * 8); | |
8046 | unsigned csum_count = blob.get_csum_count(); | |
8047 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
8048 | for (unsigned p = 0; p < csum_count; ++p) { | |
8049 | unsigned pos = p * csum_chunk_size; | |
8050 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
8051 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
8052 | unsigned mask = 1u << firstbit; | |
8053 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
8054 | mask |= 1u << b; | |
8055 | } | |
8056 | if ((blob.unused & mask) == mask) { | |
8057 | // this csum chunk region is marked unused | |
8058 | if (blob.get_csum_item(p) != 0) { | |
8059 | derr << "fsck error: " << oid | |
8060 | << " blob claims csum chunk 0x" << std::hex << pos | |
8061 | << "~" << csum_chunk_size | |
8062 | << " is unused (mask 0x" << mask << " of unused 0x" | |
8063 | << blob.unused << ") but csum is non-zero 0x" | |
8064 | << blob.get_csum_item(p) << std::dec << " on blob " | |
8065 | << *i.first << dendl; | |
8066 | ++errors; | |
8067 | } | |
8068 | } | |
8069 | } | |
8070 | } | |
8071 | } | |
8072 | // omap | |
8073 | if (o->onode.has_omap()) { | |
9f95a23c TL |
8074 | ceph_assert(ctx.used_omap_head); |
8075 | if (ctx.used_omap_head->count(o->onode.nid)) { | |
8076 | derr << "fsck error: " << o->oid << " omap_head " << o->onode.nid | |
8077 | << " already in use" << dendl; | |
eafe8130 TL |
8078 | ++errors; |
8079 | } else { | |
9f95a23c | 8080 | ctx.used_omap_head->insert(o->onode.nid); |
eafe8130 | 8081 | } |
9f95a23c | 8082 | } // if (o->onode.has_omap()) |
eafe8130 TL |
8083 | if (depth == FSCK_DEEP) { |
8084 | bufferlist bl; | |
8085 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; | |
8086 | uint64_t offset = 0; | |
8087 | do { | |
8088 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
8089 | int r = _do_read(c.get(), o, offset, l, bl, | |
8090 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
8091 | if (r < 0) { | |
8092 | ++errors; | |
8093 | derr << "fsck error: " << oid << std::hex | |
8094 | << " error during read: " | |
8095 | << " " << offset << "~" << l | |
8096 | << " " << cpp_strerror(r) << std::dec | |
8097 | << dendl; | |
8098 | break; | |
8099 | } | |
8100 | offset += l; | |
8101 | } while (offset < o->onode.size); | |
8102 | } // deep | |
8103 | } //if (depth != FSCK_SHALLOW) | |
8104 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8105 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8106 | wq->finalize(thread_pool, ctx); | |
8107 | if (processed_myself) { | |
8108 | // may be needs more threads? | |
8109 | dout(0) << __func__ << " partial offload" | |
8110 | << ", done myself " << processed_myself | |
8111 | << " of " << ctx.num_objects | |
8112 | << "objects, threads " << thread_count | |
8113 | << dendl; | |
8114 | } | |
8115 | } | |
8116 | } // if (it) | |
8117 | } | |
8118 | /** | |
8119 | An overview for currently implemented repair logics | |
8120 | performed in fsck in two stages: detection(+preparation) and commit. | |
8121 | Detection stage (in processing order): | |
8122 | (Issue -> Repair action to schedule) | |
8123 | - Detect undecodable keys for Shared Blobs -> Remove | |
8124 | - Detect undecodable records for Shared Blobs -> Remove | |
8125 | (might trigger missed Shared Blob detection below) | |
8126 | - Detect stray records for Shared Blobs -> Remove | |
8127 | - Detect misreferenced pextents -> Fix | |
8128 | Prepare Bloom-like filter to track cid/oid -> pextent | |
8129 | Prepare list of extents that are improperly referenced | |
8130 | Enumerate Onode records that might use 'misreferenced' pextents | |
8131 | (Bloom-like filter applied to reduce computation) | |
8132 | Per each questinable Onode enumerate all blobs and identify broken ones | |
8133 | (i.e. blobs having 'misreferences') | |
8134 | Rewrite each broken blob data by allocating another extents and | |
8135 | copying data there | |
8136 | If blob is shared - unshare it and mark corresponding Shared Blob | |
8137 | for removal | |
8138 | Release previously allocated space | |
8139 | Update Extent Map | |
8140 | - Detect missed Shared Blobs -> Recreate | |
8141 | - Detect undecodable deferred transaction -> Remove | |
8142 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
8143 | - Detect Freelist Manager's leaked entries -> Mark as free | |
8144 | - Detect statfs inconsistency - Update | |
8145 | Commit stage (separate DB commit per each step): | |
8146 | - Apply leaked FM entries fix | |
8147 | - Apply 'false free' FM entries fix | |
8148 | - Apply 'Remove' actions | |
8149 | - Apply fix for misreference pextents | |
8150 | - Apply Shared Blob recreate | |
8151 | (can be merged with the step above if misreferences were dectected) | |
8152 | - Apply StatFS update | |
8153 | */ | |
8154 | int BlueStore::_fsck(BlueStore::FSCKDepth depth, bool repair) | |
8155 | { | |
8156 | dout(1) << __func__ | |
8157 | << (repair ? " repair" : " check") | |
8158 | << (depth == FSCK_DEEP ? " (deep)" : | |
8159 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8160 | << dendl; | |
8161 | ||
8162 | // in deep mode we need R/W write access to be able to replay deferred ops | |
8163 | bool read_only = !(repair || depth == FSCK_DEEP); | |
8164 | ||
f67539c2 | 8165 | int r = _open_db_and_around(read_only); |
eafe8130 TL |
8166 | if (r < 0) |
8167 | return r; | |
7c673cae | 8168 | |
11fdf7f2 TL |
8169 | if (!read_only) { |
8170 | r = _upgrade_super(); | |
8171 | if (r < 0) { | |
8172 | goto out_db; | |
8173 | } | |
8174 | } | |
7c673cae | 8175 | |
eafe8130 | 8176 | r = _open_collections(); |
7c673cae | 8177 | if (r < 0) |
11fdf7f2 | 8178 | goto out_db; |
7c673cae FG |
8179 | |
8180 | mempool_thread.init(); | |
8181 | ||
11fdf7f2 TL |
8182 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
8183 | // enable in repair or deep mode modes only | |
8184 | if (!read_only) { | |
8185 | _kv_start(); | |
8186 | r = _deferred_replay(); | |
8187 | _kv_stop(); | |
8188 | } | |
7c673cae FG |
8189 | if (r < 0) |
8190 | goto out_scan; | |
8191 | ||
eafe8130 TL |
8192 | r = _fsck_on_open(depth, repair); |
8193 | ||
8194 | out_scan: | |
8195 | mempool_thread.shutdown(); | |
f6b5b4d7 | 8196 | _shutdown_cache(); |
eafe8130 | 8197 | out_db: |
1911f103 | 8198 | _close_db_and_around(false); |
eafe8130 TL |
8199 | |
8200 | return r; | |
8201 | } | |
8202 | ||
8203 | int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair) | |
8204 | { | |
8205 | dout(1) << __func__ | |
8206 | << " <<<START>>>" | |
8207 | << (repair ? " repair" : " check") | |
8208 | << (depth == FSCK_DEEP ? " (deep)" : | |
8209 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8210 | << " start" << dendl; | |
8211 | int64_t errors = 0; | |
8212 | int64_t warnings = 0; | |
8213 | unsigned repaired = 0; | |
8214 | ||
8215 | uint64_t_btree_t used_omap_head; | |
eafe8130 TL |
8216 | uint64_t_btree_t used_sbids; |
8217 | ||
f67539c2 | 8218 | mempool_dynamic_bitset used_blocks, bluefs_used_blocks; |
eafe8130 TL |
8219 | KeyValueDB::Iterator it; |
8220 | store_statfs_t expected_store_statfs, actual_statfs; | |
8221 | per_pool_statfs expected_pool_statfs; | |
8222 | ||
8223 | sb_info_map_t sb_info; | |
8224 | ||
8225 | uint64_t num_objects = 0; | |
8226 | uint64_t num_extents = 0; | |
8227 | uint64_t num_blobs = 0; | |
8228 | uint64_t num_spanning_blobs = 0; | |
8229 | uint64_t num_shared_blobs = 0; | |
8230 | uint64_t num_sharded_objects = 0; | |
8231 | BlueStoreRepairer repairer; | |
8232 | ||
f67539c2 TL |
8233 | auto alloc_size = fm->get_alloc_size(); |
8234 | ||
eafe8130 TL |
8235 | utime_t start = ceph_clock_now(); |
8236 | ||
8237 | _fsck_collections(&errors); | |
b32b8144 | 8238 | used_blocks.resize(fm->get_alloc_units()); |
7c673cae FG |
8239 | |
8240 | if (bluefs) { | |
f67539c2 | 8241 | interval_set<uint64_t> bluefs_extents; |
11fdf7f2 | 8242 | |
f67539c2 TL |
8243 | int r = bluefs->get_block_extents(bluefs_layout.shared_bdev, &bluefs_extents); |
8244 | ceph_assert(r == 0); | |
8245 | for (auto [start, len] : bluefs_extents) { | |
8246 | apply_for_bitset_range(start, len, alloc_size, used_blocks, | |
8247 | [&](uint64_t pos, mempool_dynamic_bitset& bs) { | |
8248 | ceph_assert(pos < bs.size()); | |
7c673cae | 8249 | bs.set(pos); |
f67539c2 TL |
8250 | } |
8251 | ); | |
8252 | } | |
8253 | } | |
8254 | ||
8255 | bluefs_used_blocks = used_blocks; | |
8256 | ||
8257 | apply_for_bitset_range( | |
8258 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), alloc_size, used_blocks, | |
8259 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
8260 | bs.set(pos); | |
7c673cae | 8261 | } |
f67539c2 TL |
8262 | ); |
8263 | ||
8264 | ||
8265 | if (repair) { | |
b3b6e05e | 8266 | repairer.init_space_usage_tracker( |
f67539c2 TL |
8267 | bdev->get_size(), |
8268 | min_alloc_size); | |
8269 | } | |
8270 | ||
8271 | if (bluefs) { | |
eafe8130 | 8272 | int r = bluefs->fsck(); |
7c673cae | 8273 | if (r < 0) { |
eafe8130 | 8274 | return r; |
7c673cae FG |
8275 | } |
8276 | if (r > 0) | |
8277 | errors += r; | |
8278 | } | |
8279 | ||
eafe8130 TL |
8280 | if (!per_pool_stat_collection) { |
8281 | const char *w; | |
8282 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_stats) { | |
8283 | w = "error"; | |
8284 | ++errors; | |
8285 | } else { | |
8286 | w = "warning"; | |
8287 | ++warnings; | |
8288 | } | |
8289 | derr << "fsck " << w << ": store not yet converted to per-pool stats" | |
8290 | << dendl; | |
8291 | } | |
f67539c2 | 8292 | if (per_pool_omap != OMAP_PER_PG) { |
9f95a23c TL |
8293 | const char *w; |
8294 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8295 | w = "error"; | |
8296 | ++errors; | |
8297 | } else { | |
8298 | w = "warning"; | |
8299 | ++warnings; | |
8300 | } | |
f67539c2 | 8301 | derr << "fsck " << w << ": store not yet converted to per-pg omap" |
9f95a23c TL |
8302 | << dendl; |
8303 | } | |
8304 | ||
11fdf7f2 | 8305 | // get expected statfs; reset unaffected fields to be able to compare |
7c673cae FG |
8306 | // structs |
8307 | statfs(&actual_statfs); | |
11fdf7f2 TL |
8308 | actual_statfs.total = 0; |
8309 | actual_statfs.internally_reserved = 0; | |
8310 | actual_statfs.available = 0; | |
8311 | actual_statfs.internal_metadata = 0; | |
8312 | actual_statfs.omap_allocated = 0; | |
8313 | ||
eafe8130 TL |
8314 | if (g_conf()->bluestore_debug_fsck_abort) { |
8315 | dout(1) << __func__ << " debug abort" << dendl; | |
8316 | goto out_scan; | |
8317 | } | |
7c673cae | 8318 | // walk PREFIX_OBJ |
eafe8130 TL |
8319 | { |
8320 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
8321 | ceph::mutex sb_info_lock = ceph::make_mutex("BlueStore::fsck::sbinfo_lock"); | |
8322 | BlueStore::FSCK_ObjectCtx ctx( | |
8323 | errors, | |
8324 | warnings, | |
8325 | num_objects, | |
8326 | num_extents, | |
8327 | num_blobs, | |
8328 | num_sharded_objects, | |
8329 | num_spanning_blobs, | |
8330 | &used_blocks, | |
8331 | &used_omap_head, | |
9f95a23c TL |
8332 | //no need for the below lock when in non-shallow mode as |
8333 | // there is no multithreading in this case | |
8334 | depth == FSCK_SHALLOW ? &sb_info_lock : nullptr, | |
eafe8130 TL |
8335 | sb_info, |
8336 | expected_store_statfs, | |
8337 | expected_pool_statfs, | |
8338 | repair ? &repairer : nullptr); | |
9f95a23c TL |
8339 | |
8340 | _fsck_check_objects(depth, ctx); | |
eafe8130 | 8341 | } |
11fdf7f2 | 8342 | |
7c673cae | 8343 | dout(1) << __func__ << " checking shared_blobs" << dendl; |
f67539c2 | 8344 | it = db->get_iterator(PREFIX_SHARED_BLOB, KeyValueDB::ITERATOR_NOCACHE); |
7c673cae | 8345 | if (it) { |
eafe8130 TL |
8346 | // FIXME minor: perhaps simplify for shallow mode? |
8347 | // fill global if not overriden below | |
8348 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 | 8349 | |
7c673cae FG |
8350 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8351 | string key = it->key(); | |
8352 | uint64_t sbid; | |
8353 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 8354 | derr << "fsck error: bad key '" << key |
7c673cae | 8355 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
8356 | if (repair) { |
8357 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8358 | } | |
7c673cae FG |
8359 | ++errors; |
8360 | continue; | |
8361 | } | |
8362 | auto p = sb_info.find(sbid); | |
8363 | if (p == sb_info.end()) { | |
3efd9988 | 8364 | derr << "fsck error: found stray shared blob data for sbid 0x" |
7c673cae | 8365 | << std::hex << sbid << std::dec << dendl; |
11fdf7f2 TL |
8366 | if (repair) { |
8367 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8368 | } | |
7c673cae FG |
8369 | ++errors; |
8370 | } else { | |
8371 | ++num_shared_blobs; | |
8372 | sb_info_t& sbi = p->second; | |
8373 | bluestore_shared_blob_t shared_blob(sbid); | |
8374 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
8375 | auto blp = bl.cbegin(); |
8376 | try { | |
8377 | decode(shared_blob, blp); | |
f67539c2 | 8378 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
8379 | ++errors; |
8380 | // Force update and don't report as missing | |
8381 | sbi.updated = sbi.passed = true; | |
8382 | ||
8383 | derr << "fsck error: failed to decode Shared Blob" | |
8384 | << pretty_binary_string(it->key()) << dendl; | |
8385 | if (repair) { | |
8386 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
8387 | << pretty_binary_string(it->key()) | |
8388 | << "', removing" << dendl; | |
8389 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8390 | } | |
8391 | continue; | |
8392 | } | |
7c673cae FG |
8393 | dout(20) << __func__ << " " << *sbi.sb << " " << shared_blob << dendl; |
8394 | if (shared_blob.ref_map != sbi.ref_map) { | |
3efd9988 | 8395 | derr << "fsck error: shared blob 0x" << std::hex << sbid |
11fdf7f2 TL |
8396 | << std::dec << " ref_map " << shared_blob.ref_map |
8397 | << " != expected " << sbi.ref_map << dendl; | |
8398 | sbi.updated = true; // will update later in repair mode only! | |
7c673cae FG |
8399 | ++errors; |
8400 | } | |
8401 | PExtentVector extents; | |
8402 | for (auto &r : shared_blob.ref_map.ref_map) { | |
8403 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
8404 | } | |
eafe8130 | 8405 | if (per_pool_stat_collection || repair) { |
11fdf7f2 TL |
8406 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; |
8407 | } | |
8408 | errors += _fsck_check_extents(sbi.cid, | |
8409 | p->second.oids.front(), | |
7c673cae FG |
8410 | extents, |
8411 | p->second.compressed, | |
b32b8144 FG |
8412 | used_blocks, |
8413 | fm->get_alloc_size(), | |
11fdf7f2 | 8414 | repair ? &repairer : nullptr, |
eafe8130 TL |
8415 | *expected_statfs, |
8416 | depth); | |
11fdf7f2 TL |
8417 | sbi.passed = true; |
8418 | } | |
8419 | } | |
8420 | } // if (it) | |
8421 | ||
8422 | if (repair && repairer.preprocess_misreference(db)) { | |
8423 | ||
8424 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
11fdf7f2 TL |
8425 | auto& misref_extents = repairer.get_misreferences(); |
8426 | interval_set<uint64_t> to_release; | |
f67539c2 | 8427 | it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 | 8428 | if (it) { |
eafe8130 TL |
8429 | // fill global if not overriden below |
8430 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 TL |
8431 | |
8432 | CollectionRef c; | |
8433 | spg_t pgid; | |
8434 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
8435 | bool bypass_rest = false; | |
8436 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
8437 | it->next()) { | |
8438 | dout(30) << __func__ << " key " | |
8439 | << pretty_binary_string(it->key()) << dendl; | |
8440 | if (is_extent_shard_key(it->key())) { | |
8441 | continue; | |
8442 | } | |
8443 | ||
8444 | ghobject_t oid; | |
8445 | int r = get_key_object(it->key(), &oid); | |
b3b6e05e | 8446 | if (r < 0 || !repairer.is_used(oid)) { |
11fdf7f2 TL |
8447 | continue; |
8448 | } | |
8449 | ||
8450 | if (!c || | |
8451 | oid.shard_id != pgid.shard || | |
8452 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8453 | !c->contains(oid)) { | |
8454 | c = nullptr; | |
8455 | for (auto& p : coll_map) { | |
8456 | if (p.second->contains(oid)) { | |
8457 | c = p.second; | |
8458 | break; | |
8459 | } | |
8460 | } | |
8461 | if (!c) { | |
8462 | continue; | |
8463 | } | |
eafe8130 TL |
8464 | if (per_pool_stat_collection || repair) { |
8465 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
11fdf7f2 TL |
8466 | expected_statfs = &expected_pool_statfs[pool_id]; |
8467 | } | |
8468 | } | |
b3b6e05e | 8469 | if (!repairer.is_used(c->cid)) { |
11fdf7f2 TL |
8470 | continue; |
8471 | } | |
8472 | ||
8473 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
8474 | << " obj:" << oid << dendl; | |
8475 | ||
eafe8130 TL |
8476 | OnodeRef o; |
8477 | o.reset(Onode::decode(c, oid, it->key(), it->value())); | |
11fdf7f2 TL |
8478 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8479 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
8480 | ||
8481 | for (auto& e : o->extent_map.extent_map) { | |
8482 | blobs.insert(e.blob); | |
8483 | } | |
8484 | bool need_onode_update = false; | |
8485 | bool first_dump = true; | |
8486 | for(auto b : blobs) { | |
8487 | bool broken_blob = false; | |
8488 | auto& pextents = b->dirty_blob().dirty_extents(); | |
8489 | for (auto& e : pextents) { | |
8490 | if (!e.is_valid()) { | |
8491 | continue; | |
8492 | } | |
8493 | // for the sake of simplicity and proper shared blob handling | |
8494 | // always rewrite the whole blob even when it's partially | |
8495 | // misreferenced. | |
8496 | if (misref_extents.intersects(e.offset, e.length)) { | |
8497 | if (first_dump) { | |
8498 | first_dump = false; | |
81eedcae | 8499 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
8500 | } |
8501 | broken_blob = true; | |
8502 | break; | |
8503 | } | |
8504 | } | |
8505 | if (!broken_blob) | |
8506 | continue; | |
8507 | bool compressed = b->get_blob().is_compressed(); | |
8508 | need_onode_update = true; | |
8509 | dout(10) << __func__ | |
8510 | << " fix misreferences in oid:" << oid | |
8511 | << " " << *b << dendl; | |
8512 | uint64_t b_off = 0; | |
8513 | PExtentVector pext_to_release; | |
8514 | pext_to_release.reserve(pextents.size()); | |
8515 | // rewriting all valid pextents | |
8516 | for (auto e = pextents.begin(); e != pextents.end(); | |
8517 | b_off += e->length, e++) { | |
8518 | if (!e->is_valid()) { | |
8519 | continue; | |
8520 | } | |
8521 | PExtentVector exts; | |
f67539c2 TL |
8522 | int64_t alloc_len = |
8523 | shared_alloc.a->allocate(e->length, min_alloc_size, | |
8524 | 0, 0, &exts); | |
eafe8130 | 8525 | if (alloc_len < 0 || alloc_len < (int64_t)e->length) { |
11fdf7f2 TL |
8526 | derr << __func__ |
8527 | << " failed to allocate 0x" << std::hex << e->length | |
eafe8130 | 8528 | << " allocated 0x " << (alloc_len < 0 ? 0 : alloc_len) |
11fdf7f2 | 8529 | << " min_alloc_size 0x" << min_alloc_size |
f67539c2 | 8530 | << " available 0x " << shared_alloc.a->get_free() |
11fdf7f2 TL |
8531 | << std::dec << dendl; |
8532 | if (alloc_len > 0) { | |
f67539c2 | 8533 | shared_alloc.a->release(exts); |
11fdf7f2 TL |
8534 | } |
8535 | bypass_rest = true; | |
8536 | break; | |
8537 | } | |
8538 | expected_statfs->allocated += e->length; | |
8539 | if (compressed) { | |
8540 | expected_statfs->data_compressed_allocated += e->length; | |
8541 | } | |
8542 | ||
8543 | bufferlist bl; | |
8544 | IOContext ioc(cct, NULL, true); // allow EIO | |
8545 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); | |
8546 | if (r < 0) { | |
8547 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
8548 | <<"~" << e->length << std::dec << dendl; | |
8549 | ceph_abort_msg("read failed, wtf"); | |
8550 | } | |
8551 | pext_to_release.push_back(*e); | |
8552 | e = pextents.erase(e); | |
8553 | e = pextents.insert(e, exts.begin(), exts.end()); | |
8554 | b->get_blob().map_bl( | |
8555 | b_off, bl, | |
8556 | [&](uint64_t offset, bufferlist& t) { | |
8557 | int r = bdev->write(offset, t, false); | |
8558 | ceph_assert(r == 0); | |
8559 | }); | |
8560 | e += exts.size() - 1; | |
8561 | for (auto& p : exts) { | |
8562 | fm->allocate(p.offset, p.length, txn); | |
8563 | } | |
8564 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
8565 | ||
8566 | if (b->get_blob().is_shared()) { | |
8567 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
8568 | ||
8569 | auto sb_it = sb_info.find(b->shared_blob->get_sbid()); | |
8570 | ceph_assert(sb_it != sb_info.end()); | |
8571 | sb_info_t& sbi = sb_it->second; | |
8572 | ||
8573 | for (auto& r : sbi.ref_map.ref_map) { | |
8574 | expected_statfs->allocated -= r.second.length; | |
8575 | if (sbi.compressed) { | |
8576 | // NB: it's crucial to use compressed flag from sb_info_t | |
8577 | // as we originally used that value while accumulating | |
8578 | // expected_statfs | |
8579 | expected_statfs->data_compressed_allocated -= r.second.length; | |
8580 | } | |
8581 | } | |
8582 | sbi.updated = sbi.passed = true; | |
8583 | sbi.ref_map.clear(); | |
8584 | ||
8585 | // relying on blob's pextents to decide what to release. | |
8586 | for (auto& p : pext_to_release) { | |
8587 | to_release.union_insert(p.offset, p.length); | |
8588 | } | |
8589 | } else { | |
8590 | for (auto& p : pext_to_release) { | |
8591 | expected_statfs->allocated -= p.length; | |
8592 | if (compressed) { | |
8593 | expected_statfs->data_compressed_allocated -= p.length; | |
8594 | } | |
8595 | to_release.union_insert(p.offset, p.length); | |
8596 | } | |
8597 | } | |
8598 | if (bypass_rest) { | |
8599 | break; | |
8600 | } | |
8601 | } // for(auto b : blobs) | |
8602 | if (need_onode_update) { | |
8603 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
8604 | _record_onode(o, txn); | |
8605 | } | |
8606 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8607 | ||
8608 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
8609 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
8610 | << "~" << it.get_len() << std::dec << dendl; | |
8611 | fm->release(it.get_start(), it.get_len(), txn); | |
8612 | } | |
f67539c2 | 8613 | shared_alloc.a->release(to_release); |
11fdf7f2 TL |
8614 | to_release.clear(); |
8615 | } // if (it) { | |
8616 | } //if (repair && repairer.preprocess_misreference()) { | |
8617 | ||
eafe8130 TL |
8618 | if (depth != FSCK_SHALLOW) { |
8619 | for (auto &p : sb_info) { | |
8620 | sb_info_t& sbi = p.second; | |
8621 | if (!sbi.passed) { | |
8622 | derr << "fsck error: missing " << *sbi.sb << dendl; | |
8623 | ++errors; | |
8624 | } | |
8625 | if (repair && (!sbi.passed || sbi.updated)) { | |
8626 | auto sbid = p.first; | |
8627 | if (sbi.ref_map.empty()) { | |
8628 | ceph_assert(sbi.passed); | |
8629 | dout(20) << __func__ << " " << *sbi.sb | |
8630 | << " is empty, removing" << dendl; | |
8631 | repairer.fix_shared_blob(db, sbid, nullptr); | |
8632 | } else { | |
8633 | bufferlist bl; | |
8634 | bluestore_shared_blob_t persistent(sbid, std::move(sbi.ref_map)); | |
8635 | encode(persistent, bl); | |
8636 | dout(20) << __func__ << " " << *sbi.sb | |
8637 | << " is " << bl.length() << " bytes, updating" << dendl; | |
11fdf7f2 | 8638 | |
eafe8130 TL |
8639 | repairer.fix_shared_blob(db, sbid, &bl); |
8640 | } | |
7c673cae FG |
8641 | } |
8642 | } | |
8643 | } | |
11fdf7f2 TL |
8644 | sb_info.clear(); |
8645 | ||
eafe8130 TL |
8646 | // check global stats only if fscking (not repairing) w/o per-pool stats |
8647 | if (!per_pool_stat_collection && | |
8648 | !repair && | |
8649 | !(actual_statfs == expected_store_statfs)) { | |
8650 | derr << "fsck error: actual " << actual_statfs | |
8651 | << " != expected " << expected_store_statfs << dendl; | |
8652 | if (repair) { | |
8653 | repairer.fix_statfs(db, BLUESTORE_GLOBAL_STATFS_KEY, | |
8654 | expected_store_statfs); | |
11fdf7f2 | 8655 | } |
eafe8130 | 8656 | ++errors; |
7c673cae FG |
8657 | } |
8658 | ||
eafe8130 TL |
8659 | dout(1) << __func__ << " checking pool_statfs" << dendl; |
8660 | _fsck_check_pool_statfs(expected_pool_statfs, | |
8661 | errors, warnings, repair ? &repairer : nullptr); | |
8662 | ||
8663 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 8664 | dout(1) << __func__ << " checking for stray omap data " << dendl; |
f67539c2 | 8665 | it = db->get_iterator(PREFIX_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 | 8666 | if (it) { |
9f95a23c | 8667 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8668 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8669 | uint64_t omap_head; | |
f67539c2 | 8670 | |
eafe8130 | 8671 | _key_decode_u64(it->key().c_str(), &omap_head); |
f67539c2 | 8672 | |
9f95a23c | 8673 | if (used_omap_head.count(omap_head) == 0 && |
f67539c2 | 8674 | omap_head != last_omap_head) { |
9f95a23c TL |
8675 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
8676 | << "fsck error: found stray omap data on omap_head " | |
f67539c2 TL |
8677 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; |
8678 | ++errors; | |
8679 | last_omap_head = omap_head; | |
eafe8130 | 8680 | } |
7c673cae FG |
8681 | } |
8682 | } | |
f67539c2 | 8683 | it = db->get_iterator(PREFIX_PGMETA_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 | 8684 | if (it) { |
9f95a23c | 8685 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8686 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8687 | uint64_t omap_head; | |
8688 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8689 | if (used_omap_head.count(omap_head) == 0 && |
8690 | omap_head != last_omap_head) { | |
8691 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8692 | << "fsck error: found stray (pgmeta) omap data on omap_head " | |
8693 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8694 | last_omap_head = omap_head; | |
eafe8130 TL |
8695 | ++errors; |
8696 | } | |
11fdf7f2 TL |
8697 | } |
8698 | } | |
f67539c2 | 8699 | it = db->get_iterator(PREFIX_PERPOOL_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
9f95a23c TL |
8700 | if (it) { |
8701 | uint64_t last_omap_head = 0; | |
8702 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8703 | uint64_t pool; | |
8704 | uint64_t omap_head; | |
8705 | string k = it->key(); | |
8706 | const char *c = k.c_str(); | |
8707 | c = _key_decode_u64(c, &pool); | |
8708 | c = _key_decode_u64(c, &omap_head); | |
8709 | if (used_omap_head.count(omap_head) == 0 && | |
f67539c2 | 8710 | omap_head != last_omap_head) { |
9f95a23c TL |
8711 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
8712 | << "fsck error: found stray (per-pool) omap data on omap_head " | |
8713 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8714 | ++errors; | |
f67539c2 TL |
8715 | last_omap_head = omap_head; |
8716 | } | |
8717 | } | |
8718 | } | |
8719 | it = db->get_iterator(PREFIX_PERPG_OMAP, KeyValueDB::ITERATOR_NOCACHE); | |
8720 | if (it) { | |
8721 | uint64_t last_omap_head = 0; | |
8722 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8723 | uint64_t pool; | |
8724 | uint32_t hash; | |
8725 | uint64_t omap_head; | |
8726 | string k = it->key(); | |
8727 | const char* c = k.c_str(); | |
8728 | c = _key_decode_u64(c, &pool); | |
8729 | c = _key_decode_u32(c, &hash); | |
8730 | c = _key_decode_u64(c, &omap_head); | |
8731 | if (used_omap_head.count(omap_head) == 0 && | |
8732 | omap_head != last_omap_head) { | |
8733 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8734 | << "fsck error: found stray (per-pg) omap data on omap_head " | |
8735 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8736 | ++errors; | |
8737 | last_omap_head = omap_head; | |
9f95a23c TL |
8738 | } |
8739 | } | |
8740 | } | |
eafe8130 | 8741 | dout(1) << __func__ << " checking deferred events" << dendl; |
f67539c2 | 8742 | it = db->get_iterator(PREFIX_DEFERRED, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
8743 | if (it) { |
8744 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8745 | bufferlist bl = it->value(); | |
8746 | auto p = bl.cbegin(); | |
8747 | bluestore_deferred_transaction_t wt; | |
8748 | try { | |
8749 | decode(wt, p); | |
f67539c2 | 8750 | } catch (ceph::buffer::error& e) { |
eafe8130 TL |
8751 | derr << "fsck error: failed to decode deferred txn " |
8752 | << pretty_binary_string(it->key()) << dendl; | |
8753 | if (repair) { | |
8754 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
8755 | << pretty_binary_string(it->key()) | |
8756 | << "', removing" << dendl; | |
8757 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8758 | } | |
8759 | continue; | |
8760 | } | |
8761 | dout(20) << __func__ << " deferred " << wt.seq | |
8762 | << " ops " << wt.ops.size() | |
8763 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
8764 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
9f95a23c | 8765 | apply_for_bitset_range( |
f67539c2 | 8766 | e.get_start(), e.get_len(), alloc_size, used_blocks, |
eafe8130 | 8767 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
eafe8130 TL |
8768 | bs.set(pos); |
8769 | } | |
8770 | ); | |
8771 | } | |
7c673cae | 8772 | } |
eafe8130 TL |
8773 | } |
8774 | ||
8775 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
8776 | { | |
eafe8130 TL |
8777 | fm->enumerate_reset(); |
8778 | uint64_t offset, length; | |
8779 | while (fm->enumerate_next(db, &offset, &length)) { | |
8780 | bool intersects = false; | |
9f95a23c | 8781 | apply_for_bitset_range( |
f67539c2 | 8782 | offset, length, alloc_size, used_blocks, |
eafe8130 | 8783 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
f67539c2 TL |
8784 | ceph_assert(pos < bs.size()); |
8785 | if (bs.test(pos) && !bluefs_used_blocks.test(pos)) { | |
eafe8130 TL |
8786 | if (offset == SUPER_RESERVED && |
8787 | length == min_alloc_size - SUPER_RESERVED) { | |
8788 | // this is due to the change just after luminous to min_alloc_size | |
8789 | // granularity allocations, and our baked in assumption at the top | |
8790 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
8791 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
8792 | // since we will never allocate this region below min_alloc_size. | |
8793 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
8794 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
8795 | << length << std::dec << dendl; | |
8796 | } else { | |
8797 | intersects = true; | |
8798 | if (repair) { | |
8799 | repairer.fix_false_free(db, fm, | |
8800 | pos * min_alloc_size, | |
8801 | min_alloc_size); | |
8802 | } | |
11fdf7f2 | 8803 | } |
eafe8130 TL |
8804 | } else { |
8805 | bs.set(pos); | |
8806 | } | |
7c673cae | 8807 | } |
eafe8130 TL |
8808 | ); |
8809 | if (intersects) { | |
8810 | derr << "fsck error: free extent 0x" << std::hex << offset | |
8811 | << "~" << length << std::dec | |
8812 | << " intersects allocated blocks" << dendl; | |
8813 | ++errors; | |
7c673cae | 8814 | } |
b5b8bbf5 | 8815 | } |
eafe8130 TL |
8816 | fm->enumerate_reset(); |
8817 | size_t count = used_blocks.count(); | |
8818 | if (used_blocks.size() != count) { | |
8819 | ceph_assert(used_blocks.size() > count); | |
8820 | used_blocks.flip(); | |
8821 | size_t start = used_blocks.find_first(); | |
8822 | while (start != decltype(used_blocks)::npos) { | |
8823 | size_t cur = start; | |
8824 | while (true) { | |
8825 | size_t next = used_blocks.find_next(cur); | |
8826 | if (next != cur + 1) { | |
8827 | ++errors; | |
8828 | derr << "fsck error: leaked extent 0x" << std::hex | |
8829 | << ((uint64_t)start * fm->get_alloc_size()) << "~" | |
8830 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
8831 | << dendl; | |
8832 | if (repair) { | |
8833 | repairer.fix_leaked(db, | |
8834 | fm, | |
8835 | start * min_alloc_size, | |
8836 | (cur + 1 - start) * min_alloc_size); | |
8837 | } | |
8838 | start = next; | |
8839 | break; | |
11fdf7f2 | 8840 | } |
eafe8130 | 8841 | cur = next; |
b5b8bbf5 | 8842 | } |
eafe8130 TL |
8843 | } |
8844 | used_blocks.flip(); | |
b5b8bbf5 | 8845 | } |
7c673cae FG |
8846 | } |
8847 | } | |
11fdf7f2 | 8848 | if (repair) { |
f67539c2 TL |
8849 | if (per_pool_omap != OMAP_PER_PG) { |
8850 | dout(5) << __func__ << " fixing per_pg_omap" << dendl; | |
8851 | repairer.fix_per_pool_omap(db, OMAP_PER_PG); | |
9f95a23c TL |
8852 | } |
8853 | ||
11fdf7f2 TL |
8854 | dout(5) << __func__ << " applying repair results" << dendl; |
8855 | repaired = repairer.apply(db); | |
8856 | dout(5) << __func__ << " repair applied" << dendl; | |
8857 | } | |
7c673cae | 8858 | |
eafe8130 | 8859 | out_scan: |
7c673cae FG |
8860 | dout(2) << __func__ << " " << num_objects << " objects, " |
8861 | << num_sharded_objects << " of them sharded. " | |
8862 | << dendl; | |
8863 | dout(2) << __func__ << " " << num_extents << " extents to " | |
8864 | << num_blobs << " blobs, " | |
8865 | << num_spanning_blobs << " spanning, " | |
8866 | << num_shared_blobs << " shared." | |
8867 | << dendl; | |
8868 | ||
8869 | utime_t duration = ceph_clock_now() - start; | |
9f95a23c TL |
8870 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " |
8871 | << warnings << " warnings, " | |
8872 | << repaired << " repaired, " | |
8873 | << (errors + warnings - (int)repaired) << " remaining in " | |
7c673cae | 8874 | << duration << " seconds" << dendl; |
9f95a23c TL |
8875 | |
8876 | // In non-repair mode we should return error count only as | |
8877 | // it indicates if store status is OK. | |
8878 | // In repair mode both errors and warnings are taken into account | |
8879 | // since repaired counter relates to them both. | |
8880 | return repair ? errors + warnings - (int)repaired : errors; | |
11fdf7f2 TL |
8881 | } |
8882 | ||
8883 | /// methods to inject various errors fsck can repair | |
8884 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
8885 | const bufferlist& bl) | |
8886 | { | |
8887 | KeyValueDB::Transaction txn; | |
8888 | txn = db->get_transaction(); | |
8889 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
8890 | db->submit_transaction_sync(txn); | |
8891 | }; | |
8892 | ||
8893 | void BlueStore::inject_leaked(uint64_t len) | |
8894 | { | |
8895 | KeyValueDB::Transaction txn; | |
8896 | txn = db->get_transaction(); | |
8897 | ||
8898 | PExtentVector exts; | |
f67539c2 | 8899 | int64_t alloc_len = shared_alloc.a->allocate(len, min_alloc_size, |
11fdf7f2 TL |
8900 | min_alloc_size * 256, 0, &exts); |
8901 | ceph_assert(alloc_len >= (int64_t)len); | |
8902 | for (auto& p : exts) { | |
8903 | fm->allocate(p.offset, p.length, txn); | |
8904 | } | |
8905 | db->submit_transaction_sync(txn); | |
8906 | } | |
8907 | ||
8908 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
8909 | { | |
8910 | KeyValueDB::Transaction txn; | |
8911 | OnodeRef o; | |
8912 | CollectionRef c = _get_collection(cid); | |
8913 | ceph_assert(c); | |
8914 | { | |
9f95a23c | 8915 | std::unique_lock l{c->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
8916 | o = c->get_onode(oid, false); |
8917 | ceph_assert(o); | |
8918 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
8919 | } | |
8920 | ||
8921 | bool injected = false; | |
8922 | txn = db->get_transaction(); | |
8923 | auto& em = o->extent_map.extent_map; | |
8924 | std::vector<const PExtentVector*> v; | |
8925 | if (em.size()) { | |
8926 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
8927 | } | |
8928 | if (em.size() > 1) { | |
8929 | auto it = em.end(); | |
8930 | --it; | |
8931 | v.push_back(&(it->blob->get_blob().get_extents())); | |
8932 | } | |
8933 | for (auto pext : v) { | |
8934 | if (pext->size()) { | |
8935 | auto p = pext->begin(); | |
8936 | while (p != pext->end()) { | |
8937 | if (p->is_valid()) { | |
8938 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
8939 | << "~" << p->length << std::dec << dendl; | |
8940 | fm->release(p->offset, p->length, txn); | |
8941 | injected = true; | |
8942 | break; | |
8943 | } | |
8944 | ++p; | |
8945 | } | |
8946 | } | |
8947 | } | |
8948 | ceph_assert(injected); | |
8949 | db->submit_transaction_sync(txn); | |
8950 | } | |
8951 | ||
9f95a23c TL |
8952 | void BlueStore::inject_legacy_omap() |
8953 | { | |
8954 | dout(1) << __func__ << dendl; | |
f67539c2 | 8955 | per_pool_omap = OMAP_BULK; |
9f95a23c TL |
8956 | KeyValueDB::Transaction txn; |
8957 | txn = db->get_transaction(); | |
8958 | txn->rmkey(PREFIX_SUPER, "per_pool_omap"); | |
8959 | db->submit_transaction_sync(txn); | |
8960 | } | |
8961 | ||
8962 | void BlueStore::inject_legacy_omap(coll_t cid, ghobject_t oid) | |
8963 | { | |
8964 | dout(1) << __func__ << " " | |
8965 | << cid << " " << oid | |
8966 | <<dendl; | |
8967 | KeyValueDB::Transaction txn; | |
8968 | OnodeRef o; | |
8969 | CollectionRef c = _get_collection(cid); | |
8970 | ceph_assert(c); | |
8971 | { | |
8972 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
8973 | o = c->get_onode(oid, false); | |
8974 | ceph_assert(o); | |
8975 | } | |
f67539c2 TL |
8976 | o->onode.clear_flag( |
8977 | bluestore_onode_t::FLAG_PERPG_OMAP | | |
8978 | bluestore_onode_t::FLAG_PERPOOL_OMAP | | |
8979 | bluestore_onode_t::FLAG_PGMETA_OMAP); | |
9f95a23c TL |
8980 | txn = db->get_transaction(); |
8981 | _record_onode(o, txn); | |
8982 | db->submit_transaction_sync(txn); | |
8983 | } | |
8984 | ||
8985 | ||
11fdf7f2 TL |
8986 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) |
8987 | { | |
8988 | BlueStoreRepairer repairer; | |
8989 | repairer.fix_statfs(db, key, new_statfs); | |
8990 | repairer.apply(db); | |
8991 | } | |
8992 | ||
eafe8130 TL |
8993 | void BlueStore::inject_global_statfs(const store_statfs_t& new_statfs) |
8994 | { | |
8995 | KeyValueDB::Transaction t = db->get_transaction(); | |
8996 | volatile_statfs v; | |
8997 | v = new_statfs; | |
8998 | bufferlist bl; | |
8999 | v.encode(bl); | |
9000 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
9001 | db->submit_transaction_sync(t); | |
9002 | } | |
9003 | ||
11fdf7f2 TL |
9004 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, |
9005 | coll_t cid2, ghobject_t oid2, | |
9006 | uint64_t offset) | |
9007 | { | |
9008 | OnodeRef o1; | |
9009 | CollectionRef c1 = _get_collection(cid1); | |
9010 | ceph_assert(c1); | |
9011 | { | |
9f95a23c | 9012 | std::unique_lock l{c1->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9013 | o1 = c1->get_onode(oid1, false); |
9014 | ceph_assert(o1); | |
9015 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9016 | } | |
9017 | OnodeRef o2; | |
9018 | CollectionRef c2 = _get_collection(cid2); | |
9019 | ceph_assert(c2); | |
9020 | { | |
9f95a23c | 9021 | std::unique_lock l{c2->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9022 | o2 = c2->get_onode(oid2, false); |
9023 | ceph_assert(o2); | |
9024 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9025 | } | |
9026 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
9027 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
9028 | ||
9029 | // require onode/extent layout to be the same (and simple) | |
9030 | // to make things easier | |
9031 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
9032 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
9033 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
9034 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
9035 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
9036 | ceph_assert(e1.length == e2.length); | |
9037 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
9038 | ||
9039 | KeyValueDB::Transaction txn; | |
9040 | txn = db->get_transaction(); | |
9041 | ||
9042 | // along with misreference error this will create space leaks errors | |
9043 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
9044 | o2->extent_map.dirty_range(offset, e2.length); | |
9045 | o2->extent_map.update(txn, false); | |
9046 | ||
9047 | _record_onode(o2, txn); | |
9048 | db->submit_transaction_sync(txn); | |
7c673cae FG |
9049 | } |
9050 | ||
adb31ebb TL |
9051 | void BlueStore::inject_zombie_spanning_blob(coll_t cid, ghobject_t oid, |
9052 | int16_t blob_id) | |
9053 | { | |
9054 | OnodeRef o; | |
9055 | CollectionRef c = _get_collection(cid); | |
9056 | ceph_assert(c); | |
9057 | { | |
9058 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
9059 | o = c->get_onode(oid, false); | |
9060 | ceph_assert(o); | |
9061 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
9062 | } | |
9063 | ||
9064 | BlobRef b = c->new_blob(); | |
9065 | b->id = blob_id; | |
9066 | o->extent_map.spanning_blob_map[blob_id] = b; | |
9067 | ||
9068 | KeyValueDB::Transaction txn; | |
9069 | txn = db->get_transaction(); | |
9070 | ||
9071 | _record_onode(o, txn); | |
9072 | db->submit_transaction_sync(txn); | |
9073 | } | |
9074 | ||
7c673cae FG |
9075 | void BlueStore::collect_metadata(map<string,string> *pm) |
9076 | { | |
9077 | dout(10) << __func__ << dendl; | |
9078 | bdev->collect_metadata("bluestore_bdev_", pm); | |
9079 | if (bluefs) { | |
9080 | (*pm)["bluefs"] = "1"; | |
9f95a23c TL |
9081 | // this value is for backward compatibility only |
9082 | (*pm)["bluefs_single_shared_device"] = \ | |
9083 | stringify((int)bluefs_layout.single_shared_device()); | |
9084 | (*pm)["bluefs_dedicated_db"] = \ | |
9085 | stringify((int)bluefs_layout.dedicated_db); | |
9086 | (*pm)["bluefs_dedicated_wal"] = \ | |
9087 | stringify((int)bluefs_layout.dedicated_wal); | |
9088 | bluefs->collect_metadata(pm, bluefs_layout.shared_bdev); | |
7c673cae FG |
9089 | } else { |
9090 | (*pm)["bluefs"] = "0"; | |
9091 | } | |
11fdf7f2 TL |
9092 | |
9093 | // report numa mapping for underlying devices | |
9094 | int node = -1; | |
9095 | set<int> nodes; | |
9096 | set<string> failed; | |
9097 | int r = get_numa_node(&node, &nodes, &failed); | |
9098 | if (r >= 0) { | |
9099 | if (!failed.empty()) { | |
9100 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
9101 | } | |
9102 | if (!nodes.empty()) { | |
9103 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
9104 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
9105 | } | |
9106 | if (node >= 0) { | |
9107 | (*pm)["objectstore_numa_node"] = stringify(node); | |
9108 | } | |
9109 | } | |
9110 | } | |
9111 | ||
9112 | int BlueStore::get_numa_node( | |
9113 | int *final_node, | |
9114 | set<int> *out_nodes, | |
9115 | set<string> *out_failed) | |
9116 | { | |
9117 | int node = -1; | |
9118 | set<string> devices; | |
9119 | get_devices(&devices); | |
9120 | set<int> nodes; | |
9121 | set<string> failed; | |
9122 | for (auto& devname : devices) { | |
9123 | int n; | |
9124 | BlkDev bdev(devname); | |
9125 | int r = bdev.get_numa_node(&n); | |
9126 | if (r < 0) { | |
9127 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
9128 | << dendl; | |
9129 | failed.insert(devname); | |
9130 | continue; | |
9131 | } | |
9132 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
9133 | << dendl; | |
9134 | nodes.insert(n); | |
9135 | if (node < 0) { | |
9136 | node = n; | |
9137 | } | |
9138 | } | |
9139 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
9140 | *final_node = node; | |
9141 | } | |
9142 | if (out_nodes) { | |
9143 | *out_nodes = nodes; | |
9144 | } | |
9145 | if (out_failed) { | |
9146 | *out_failed = failed; | |
9147 | } | |
9148 | return 0; | |
9149 | } | |
9150 | ||
9151 | int BlueStore::get_devices(set<string> *ls) | |
9152 | { | |
9153 | if (bdev) { | |
9154 | bdev->get_devices(ls); | |
9155 | if (bluefs) { | |
9156 | bluefs->get_devices(ls); | |
9157 | } | |
9158 | return 0; | |
9159 | } | |
9160 | ||
9161 | // grumble, we haven't started up yet. | |
9162 | int r = _open_path(); | |
9163 | if (r < 0) | |
9164 | goto out; | |
9165 | r = _open_fsid(false); | |
9166 | if (r < 0) | |
9167 | goto out_path; | |
9168 | r = _read_fsid(&fsid); | |
9169 | if (r < 0) | |
9170 | goto out_fsid; | |
9171 | r = _lock_fsid(); | |
9172 | if (r < 0) | |
9173 | goto out_fsid; | |
9174 | r = _open_bdev(false); | |
9175 | if (r < 0) | |
9176 | goto out_fsid; | |
9177 | r = _minimal_open_bluefs(false); | |
9178 | if (r < 0) | |
9179 | goto out_bdev; | |
9180 | bdev->get_devices(ls); | |
9181 | if (bluefs) { | |
9182 | bluefs->get_devices(ls); | |
9183 | } | |
9184 | r = 0; | |
9185 | _minimal_close_bluefs(); | |
9186 | out_bdev: | |
9187 | _close_bdev(); | |
9188 | out_fsid: | |
9189 | _close_fsid(); | |
9190 | out_path: | |
9191 | _close_path(); | |
9192 | out: | |
9193 | return r; | |
7c673cae FG |
9194 | } |
9195 | ||
11fdf7f2 | 9196 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
9197 | { |
9198 | buf->reset(); | |
11fdf7f2 | 9199 | |
f67539c2 TL |
9200 | auto prefix = per_pool_omap == OMAP_BULK ? |
9201 | PREFIX_OMAP : | |
9202 | per_pool_omap == OMAP_PER_POOL ? | |
9203 | PREFIX_PERPOOL_OMAP : | |
9204 | PREFIX_PERPG_OMAP; | |
9f95a23c | 9205 | buf->omap_allocated = |
f67539c2 | 9206 | db->estimate_prefix_size(prefix, string()); |
11fdf7f2 | 9207 | |
f67539c2 | 9208 | uint64_t bfree = shared_alloc.a->get_free(); |
7c673cae FG |
9209 | |
9210 | if (bluefs) { | |
f67539c2 | 9211 | buf->internally_reserved = 0; |
11fdf7f2 | 9212 | // include dedicated db, too, if that isn't the shared device. |
9f95a23c | 9213 | if (bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 9214 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); |
7c673cae | 9215 | } |
11fdf7f2 TL |
9216 | // call any non-omap bluefs space "internal metadata" |
9217 | buf->internal_metadata = | |
f67539c2 | 9218 | bluefs->get_used() |
11fdf7f2 | 9219 | - buf->omap_allocated; |
7c673cae FG |
9220 | } |
9221 | ||
11fdf7f2 TL |
9222 | uint64_t thin_total, thin_avail; |
9223 | if (bdev->get_thin_utilization(&thin_total, &thin_avail)) { | |
9224 | buf->total += thin_total; | |
9225 | ||
9226 | // we are limited by both the size of the virtual device and the | |
9227 | // underlying physical device. | |
9228 | bfree = std::min(bfree, thin_avail); | |
9229 | ||
9230 | buf->allocated = thin_total - thin_avail; | |
9231 | } else { | |
9232 | buf->total += bdev->get_size(); | |
9233 | } | |
9234 | buf->available = bfree; | |
9235 | } | |
9236 | ||
9237 | int BlueStore::statfs(struct store_statfs_t *buf, | |
9238 | osd_alert_list_t* alerts) | |
9239 | { | |
9240 | if (alerts) { | |
9241 | alerts->clear(); | |
9242 | _log_alerts(*alerts); | |
9243 | } | |
9244 | _get_statfs_overall(buf); | |
31f18b77 | 9245 | { |
11fdf7f2 | 9246 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 9247 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
9248 | buf->data_stored = vstatfs.stored(); |
9249 | buf->data_compressed = vstatfs.compressed(); | |
9250 | buf->data_compressed_original = vstatfs.compressed_original(); | |
9251 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
9252 | } | |
9253 | ||
9254 | dout(20) << __func__ << " " << *buf << dendl; | |
9255 | return 0; | |
9256 | } | |
9257 | ||
9f95a23c TL |
9258 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
9259 | bool *out_per_pool_omap) | |
11fdf7f2 TL |
9260 | { |
9261 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 9262 | |
11fdf7f2 TL |
9263 | if (!per_pool_stat_collection) { |
9264 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
9265 | return -ENOTSUP; | |
7c673cae | 9266 | } |
11fdf7f2 | 9267 | buf->reset(); |
7c673cae | 9268 | |
11fdf7f2 TL |
9269 | { |
9270 | std::lock_guard l(vstatfs_lock); | |
9271 | osd_pools[pool_id].publish(buf); | |
9272 | } | |
9f95a23c TL |
9273 | |
9274 | string key_prefix; | |
9275 | _key_encode_u64(pool_id, &key_prefix); | |
f67539c2 TL |
9276 | *out_per_pool_omap = per_pool_omap != OMAP_BULK; |
9277 | if (*out_per_pool_omap) { | |
9278 | auto prefix = per_pool_omap == OMAP_PER_POOL ? | |
9279 | PREFIX_PERPOOL_OMAP : | |
9280 | PREFIX_PERPG_OMAP; | |
9281 | buf->omap_allocated = db->estimate_prefix_size(prefix, key_prefix); | |
9282 | } | |
9f95a23c | 9283 | |
11fdf7f2 | 9284 | dout(10) << __func__ << *buf << dendl; |
7c673cae FG |
9285 | return 0; |
9286 | } | |
9287 | ||
81eedcae TL |
9288 | void BlueStore::_check_legacy_statfs_alert() |
9289 | { | |
9290 | string s; | |
9291 | if (!per_pool_stat_collection && | |
eafe8130 | 9292 | cct->_conf->bluestore_warn_on_legacy_statfs) { |
81eedcae TL |
9293 | s = "legacy statfs reporting detected, " |
9294 | "suggest to run store repair to get consistent statistic reports"; | |
9295 | } | |
9296 | std::lock_guard l(qlock); | |
9297 | legacy_statfs_alert = s; | |
9298 | } | |
9299 | ||
f67539c2 | 9300 | void BlueStore::_check_no_per_pg_or_pool_omap_alert() |
9f95a23c | 9301 | { |
f67539c2 TL |
9302 | string per_pg, per_pool; |
9303 | if (per_pool_omap != OMAP_PER_PG) { | |
9304 | if (cct->_conf->bluestore_warn_on_no_per_pg_omap) { | |
9305 | per_pg = "legacy (not per-pg) omap detected, " | |
9306 | "suggest to run store repair to benefit from faster PG removal"; | |
9307 | } | |
9308 | if (per_pool_omap != OMAP_PER_POOL) { | |
9309 | if (cct->_conf->bluestore_warn_on_no_per_pool_omap) { | |
9310 | per_pool = "legacy (not per-pool) omap detected, " | |
9311 | "suggest to run store repair to benefit from per-pool omap usage statistics"; | |
9312 | } | |
9313 | } | |
9f95a23c TL |
9314 | } |
9315 | std::lock_guard l(qlock); | |
f67539c2 TL |
9316 | no_per_pg_omap_alert = per_pg; |
9317 | no_per_pool_omap_alert = per_pool; | |
9f95a23c TL |
9318 | } |
9319 | ||
7c673cae FG |
9320 | // --------------- |
9321 | // cache | |
9322 | ||
9323 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
9324 | { | |
9f95a23c | 9325 | std::shared_lock l(coll_lock); |
7c673cae FG |
9326 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); |
9327 | if (cp == coll_map.end()) | |
9328 | return CollectionRef(); | |
9329 | return cp->second; | |
9330 | } | |
9331 | ||
9332 | void BlueStore::_queue_reap_collection(CollectionRef& c) | |
9333 | { | |
9334 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
9335 | // _reap_collections and this in the same thread, |
9336 | // so no need a lock. | |
7c673cae FG |
9337 | removed_collections.push_back(c); |
9338 | } | |
9339 | ||
9340 | void BlueStore::_reap_collections() | |
9341 | { | |
94b18763 | 9342 | |
7c673cae FG |
9343 | list<CollectionRef> removed_colls; |
9344 | { | |
94b18763 FG |
9345 | // _queue_reap_collection and this in the same thread. |
9346 | // So no need a lock. | |
9347 | if (!removed_collections.empty()) | |
9348 | removed_colls.swap(removed_collections); | |
9349 | else | |
9350 | return; | |
7c673cae FG |
9351 | } |
9352 | ||
94b18763 FG |
9353 | list<CollectionRef>::iterator p = removed_colls.begin(); |
9354 | while (p != removed_colls.end()) { | |
7c673cae FG |
9355 | CollectionRef c = *p; |
9356 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
adb31ebb | 9357 | if (c->onode_map.map_any([&](Onode* o) { |
11fdf7f2 | 9358 | ceph_assert(!o->exists); |
7c673cae FG |
9359 | if (o->flushing_count.load()) { |
9360 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
9361 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 9362 | return true; |
7c673cae | 9363 | } |
94b18763 | 9364 | return false; |
7c673cae | 9365 | })) { |
94b18763 | 9366 | ++p; |
7c673cae FG |
9367 | continue; |
9368 | } | |
9369 | c->onode_map.clear(); | |
94b18763 | 9370 | p = removed_colls.erase(p); |
7c673cae FG |
9371 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
9372 | } | |
94b18763 | 9373 | if (removed_colls.empty()) { |
7c673cae | 9374 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
9375 | } else { |
9376 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
9377 | } |
9378 | } | |
9379 | ||
9380 | void BlueStore::_update_cache_logger() | |
9381 | { | |
9382 | uint64_t num_onodes = 0; | |
9f95a23c | 9383 | uint64_t num_pinned_onodes = 0; |
7c673cae FG |
9384 | uint64_t num_extents = 0; |
9385 | uint64_t num_blobs = 0; | |
9386 | uint64_t num_buffers = 0; | |
9387 | uint64_t num_buffer_bytes = 0; | |
9f95a23c TL |
9388 | for (auto c : onode_cache_shards) { |
9389 | c->add_stats(&num_onodes, &num_pinned_onodes); | |
9390 | } | |
9391 | for (auto c : buffer_cache_shards) { | |
9392 | c->add_stats(&num_extents, &num_blobs, | |
9393 | &num_buffers, &num_buffer_bytes); | |
7c673cae FG |
9394 | } |
9395 | logger->set(l_bluestore_onodes, num_onodes); | |
9f95a23c | 9396 | logger->set(l_bluestore_pinned_onodes, num_pinned_onodes); |
7c673cae FG |
9397 | logger->set(l_bluestore_extents, num_extents); |
9398 | logger->set(l_bluestore_blobs, num_blobs); | |
9399 | logger->set(l_bluestore_buffers, num_buffers); | |
9400 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
9401 | } | |
9402 | ||
9403 | // --------------- | |
9404 | // read operations | |
9405 | ||
9406 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
9407 | { | |
9408 | return _get_collection(cid); | |
9409 | } | |
9410 | ||
11fdf7f2 TL |
9411 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
9412 | const coll_t& cid) | |
7c673cae | 9413 | { |
9f95a23c TL |
9414 | std::unique_lock l{coll_lock}; |
9415 | auto c = ceph::make_ref<Collection>( | |
11fdf7f2 | 9416 | this, |
9f95a23c TL |
9417 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
9418 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
11fdf7f2 TL |
9419 | cid); |
9420 | new_coll_map[cid] = c; | |
9f95a23c | 9421 | _osr_attach(c.get()); |
11fdf7f2 TL |
9422 | return c; |
9423 | } | |
9424 | ||
9425 | void BlueStore::set_collection_commit_queue( | |
9426 | const coll_t& cid, | |
9427 | ContextQueue *commit_queue) | |
9428 | { | |
9429 | if (commit_queue) { | |
9f95a23c | 9430 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
9431 | if (coll_map.count(cid)) { |
9432 | coll_map[cid]->commit_queue = commit_queue; | |
9433 | } else if (new_coll_map.count(cid)) { | |
9434 | new_coll_map[cid]->commit_queue = commit_queue; | |
9435 | } | |
9436 | } | |
7c673cae FG |
9437 | } |
9438 | ||
11fdf7f2 | 9439 | |
7c673cae FG |
9440 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
9441 | { | |
9442 | Collection *c = static_cast<Collection *>(c_.get()); | |
9443 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
9444 | if (!c->exists) | |
9445 | return false; | |
9446 | ||
9447 | bool r = true; | |
9448 | ||
9449 | { | |
9f95a23c | 9450 | std::shared_lock l(c->lock); |
7c673cae FG |
9451 | OnodeRef o = c->get_onode(oid, false); |
9452 | if (!o || !o->exists) | |
9453 | r = false; | |
9454 | } | |
9455 | ||
7c673cae FG |
9456 | return r; |
9457 | } | |
9458 | ||
7c673cae FG |
9459 | int BlueStore::stat( |
9460 | CollectionHandle &c_, | |
9461 | const ghobject_t& oid, | |
9462 | struct stat *st, | |
9463 | bool allow_eio) | |
9464 | { | |
9465 | Collection *c = static_cast<Collection *>(c_.get()); | |
9466 | if (!c->exists) | |
9467 | return -ENOENT; | |
9468 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
9469 | ||
9470 | { | |
9f95a23c | 9471 | std::shared_lock l(c->lock); |
7c673cae FG |
9472 | OnodeRef o = c->get_onode(oid, false); |
9473 | if (!o || !o->exists) | |
9474 | return -ENOENT; | |
9475 | st->st_size = o->onode.size; | |
9476 | st->st_blksize = 4096; | |
9477 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
9478 | st->st_nlink = 1; | |
9479 | } | |
9480 | ||
7c673cae FG |
9481 | int r = 0; |
9482 | if (_debug_mdata_eio(oid)) { | |
9483 | r = -EIO; | |
9484 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9485 | } | |
9486 | return r; | |
9487 | } | |
9488 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 9489 | CollectionHandle& ch, |
7c673cae FG |
9490 | const pool_opts_t& opts) |
9491 | { | |
7c673cae | 9492 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 9493 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
9494 | if (!c->exists) |
9495 | return -ENOENT; | |
9f95a23c | 9496 | std::unique_lock l{c->lock}; |
7c673cae FG |
9497 | c->pool_opts = opts; |
9498 | return 0; | |
9499 | } | |
9500 | ||
7c673cae FG |
9501 | int BlueStore::read( |
9502 | CollectionHandle &c_, | |
9503 | const ghobject_t& oid, | |
9504 | uint64_t offset, | |
9505 | size_t length, | |
9506 | bufferlist& bl, | |
224ce89b | 9507 | uint32_t op_flags) |
7c673cae | 9508 | { |
11fdf7f2 | 9509 | auto start = mono_clock::now(); |
7c673cae FG |
9510 | Collection *c = static_cast<Collection *>(c_.get()); |
9511 | const coll_t &cid = c->get_cid(); | |
9512 | dout(15) << __func__ << " " << cid << " " << oid | |
9513 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9514 | << dendl; | |
9515 | if (!c->exists) | |
9516 | return -ENOENT; | |
9517 | ||
9518 | bl.clear(); | |
9519 | int r; | |
9520 | { | |
9f95a23c | 9521 | std::shared_lock l(c->lock); |
11fdf7f2 | 9522 | auto start1 = mono_clock::now(); |
7c673cae | 9523 | OnodeRef o = c->get_onode(oid, false); |
494da23a TL |
9524 | log_latency("get_onode@read", |
9525 | l_bluestore_read_onode_meta_lat, | |
9526 | mono_clock::now() - start1, | |
9527 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9528 | if (!o || !o->exists) { |
9529 | r = -ENOENT; | |
9530 | goto out; | |
9531 | } | |
9532 | ||
9533 | if (offset == length && offset == 0) | |
9534 | length = o->onode.size; | |
9535 | ||
9536 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
9537 | if (r == -EIO) { |
9538 | logger->inc(l_bluestore_read_eio); | |
9539 | } | |
7c673cae FG |
9540 | } |
9541 | ||
9542 | out: | |
28e407b8 | 9543 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
9544 | r = -EIO; |
9545 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
9546 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
9547 | cct->_conf->bluestore_debug_random_read_err && | |
9548 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
9549 | 100.0)) == 0) { | |
224ce89b WB |
9550 | dout(0) << __func__ << ": inject random EIO" << dendl; |
9551 | r = -EIO; | |
7c673cae FG |
9552 | } |
9553 | dout(10) << __func__ << " " << cid << " " << oid | |
9554 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9555 | << " = " << r << dendl; | |
494da23a TL |
9556 | log_latency(__func__, |
9557 | l_bluestore_read_lat, | |
9558 | mono_clock::now() - start, | |
9559 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9560 | return r; |
9561 | } | |
9562 | ||
9f95a23c | 9563 | void BlueStore::_read_cache( |
7c673cae FG |
9564 | OnodeRef o, |
9565 | uint64_t offset, | |
9566 | size_t length, | |
9f95a23c TL |
9567 | int read_cache_policy, |
9568 | ready_regions_t& ready_regions, | |
9569 | blobs2read_t& blobs2read) | |
7c673cae | 9570 | { |
7c673cae | 9571 | // build blob-wise list to of stuff read (that isn't cached) |
7c673cae FG |
9572 | unsigned left = length; |
9573 | uint64_t pos = offset; | |
7c673cae FG |
9574 | auto lp = o->extent_map.seek_lextent(offset); |
9575 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
9576 | if (pos < lp->logical_offset) { | |
9577 | unsigned hole = lp->logical_offset - pos; | |
9578 | if (hole >= left) { | |
9f95a23c | 9579 | break; |
7c673cae FG |
9580 | } |
9581 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
9f95a23c | 9582 | << std::dec << dendl; |
7c673cae FG |
9583 | pos += hole; |
9584 | left -= hole; | |
9585 | } | |
94b18763 | 9586 | BlobRef& bptr = lp->blob; |
7c673cae FG |
9587 | unsigned l_off = pos - lp->logical_offset; |
9588 | unsigned b_off = l_off + lp->blob_offset; | |
9589 | unsigned b_len = std::min(left, lp->length - l_off); | |
9590 | ||
9591 | ready_regions_t cache_res; | |
9592 | interval_set<uint32_t> cache_interval; | |
9593 | bptr->shared_blob->bc.read( | |
91327a77 AA |
9594 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
9595 | read_cache_policy); | |
7c673cae | 9596 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c TL |
9597 | << " need 0x" << b_off << "~" << b_len |
9598 | << " cache has 0x" << cache_interval | |
9599 | << std::dec << dendl; | |
7c673cae FG |
9600 | |
9601 | auto pc = cache_res.begin(); | |
11fdf7f2 | 9602 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
9603 | while (b_len > 0) { |
9604 | unsigned l; | |
9605 | if (pc != cache_res.end() && | |
9f95a23c TL |
9606 | pc->first == b_off) { |
9607 | l = pc->second.length(); | |
f67539c2 | 9608 | ready_regions[pos] = std::move(pc->second); |
9f95a23c TL |
9609 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" |
9610 | << b_off << "~" << l << std::dec << dendl; | |
9611 | ++pc; | |
7c673cae | 9612 | } else { |
9f95a23c TL |
9613 | l = b_len; |
9614 | if (pc != cache_res.end()) { | |
9615 | ceph_assert(pc->first > b_off); | |
9616 | l = pc->first - b_off; | |
9617 | } | |
9618 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
9619 | << b_off << "~" << l << std::dec << dendl; | |
9620 | // merge regions | |
9621 | { | |
9622 | uint64_t r_off = b_off; | |
9623 | uint64_t r_len = l; | |
9624 | uint64_t front = r_off % chunk_size; | |
9625 | if (front) { | |
9626 | r_off -= front; | |
9627 | r_len += front; | |
9628 | } | |
9629 | unsigned tail = r_len % chunk_size; | |
9630 | if (tail) { | |
9631 | r_len += chunk_size - tail; | |
9632 | } | |
9633 | bool merged = false; | |
9634 | regions2read_t& r2r = blobs2read[bptr]; | |
9635 | if (r2r.size()) { | |
9636 | read_req_t& pre = r2r.back(); | |
9637 | if (r_off <= (pre.r_off + pre.r_len)) { | |
9638 | front += (r_off - pre.r_off); | |
9639 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
9640 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9641 | merged = true; | |
9642 | } | |
9643 | } | |
9644 | if (!merged) { | |
9645 | read_req_t req(r_off, r_len); | |
9646 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9647 | r2r.emplace_back(std::move(req)); | |
9648 | } | |
9649 | } | |
7c673cae FG |
9650 | } |
9651 | pos += l; | |
9652 | b_off += l; | |
9653 | left -= l; | |
9654 | b_len -= l; | |
9655 | } | |
9656 | ++lp; | |
9657 | } | |
9f95a23c | 9658 | } |
7c673cae | 9659 | |
9f95a23c TL |
9660 | int BlueStore::_prepare_read_ioc( |
9661 | blobs2read_t& blobs2read, | |
9662 | vector<bufferlist>* compressed_blob_bls, | |
9663 | IOContext* ioc) | |
9664 | { | |
7c673cae | 9665 | for (auto& p : blobs2read) { |
94b18763 | 9666 | const BlobRef& bptr = p.first; |
11fdf7f2 | 9667 | regions2read_t& r2r = p.second; |
7c673cae | 9668 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9669 | << " need " << r2r << std::dec << dendl; |
7c673cae FG |
9670 | if (bptr->get_blob().is_compressed()) { |
9671 | // read the whole thing | |
9f95a23c TL |
9672 | if (compressed_blob_bls->empty()) { |
9673 | // ensure we avoid any reallocation on subsequent blobs | |
9674 | compressed_blob_bls->reserve(blobs2read.size()); | |
9675 | } | |
9676 | compressed_blob_bls->push_back(bufferlist()); | |
9677 | bufferlist& bl = compressed_blob_bls->back(); | |
9678 | auto r = bptr->get_blob().map( | |
9679 | 0, bptr->get_blob().get_ondisk_length(), | |
9680 | [&](uint64_t offset, uint64_t length) { | |
9681 | int r = bdev->aio_read(offset, length, &bl, ioc); | |
9682 | if (r < 0) | |
7c673cae FG |
9683 | return r; |
9684 | return 0; | |
9f95a23c | 9685 | }); |
b32b8144 FG |
9686 | if (r < 0) { |
9687 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
9688 | if (r == -EIO) { | |
9689 | // propagate EIO to caller | |
9690 | return r; | |
9691 | } | |
11fdf7f2 | 9692 | ceph_assert(r == 0); |
b32b8144 | 9693 | } |
7c673cae FG |
9694 | } else { |
9695 | // read the pieces | |
11fdf7f2 | 9696 | for (auto& req : r2r) { |
9f95a23c TL |
9697 | dout(20) << __func__ << " region 0x" << std::hex |
9698 | << req.regs.front().logical_offset | |
9699 | << ": 0x" << req.regs.front().blob_xoffset | |
9700 | << " reading 0x" << req.r_off | |
9701 | << "~" << req.r_len << std::dec | |
9702 | << dendl; | |
7c673cae | 9703 | |
9f95a23c TL |
9704 | // read it |
9705 | auto r = bptr->get_blob().map( | |
9706 | req.r_off, req.r_len, | |
9707 | [&](uint64_t offset, uint64_t length) { | |
9708 | int r = bdev->aio_read(offset, length, &req.bl, ioc); | |
9709 | if (r < 0) | |
7c673cae FG |
9710 | return r; |
9711 | return 0; | |
9f95a23c | 9712 | }); |
b32b8144 FG |
9713 | if (r < 0) { |
9714 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
9715 | << dendl; | |
9716 | if (r == -EIO) { | |
9717 | // propagate EIO to caller | |
9718 | return r; | |
9719 | } | |
11fdf7f2 | 9720 | ceph_assert(r == 0); |
b32b8144 | 9721 | } |
9f95a23c | 9722 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
9723 | } |
9724 | } | |
9725 | } | |
9f95a23c TL |
9726 | return 0; |
9727 | } | |
11fdf7f2 | 9728 | |
9f95a23c TL |
9729 | int BlueStore::_generate_read_result_bl( |
9730 | OnodeRef o, | |
9731 | uint64_t offset, | |
9732 | size_t length, | |
9733 | ready_regions_t& ready_regions, | |
9734 | vector<bufferlist>& compressed_blob_bls, | |
9735 | blobs2read_t& blobs2read, | |
9736 | bool buffered, | |
9737 | bool* csum_error, | |
9738 | bufferlist& bl) | |
9739 | { | |
9740 | // enumerate and decompress desired blobs | |
7c673cae FG |
9741 | auto p = compressed_blob_bls.begin(); |
9742 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
9743 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 9744 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 9745 | regions2read_t& r2r = b2r_it->second; |
7c673cae | 9746 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9747 | << " need 0x" << r2r << std::dec << dendl; |
7c673cae | 9748 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 9749 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
9750 | bufferlist& compressed_bl = *p++; |
9751 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
9f95a23c TL |
9752 | r2r.front().regs.front().logical_offset) < 0) { |
9753 | *csum_error = true; | |
9754 | return -EIO; | |
7c673cae FG |
9755 | } |
9756 | bufferlist raw_bl; | |
9f95a23c | 9757 | auto r = _decompress(compressed_bl, &raw_bl); |
7c673cae | 9758 | if (r < 0) |
9f95a23c | 9759 | return r; |
7c673cae | 9760 | if (buffered) { |
9f95a23c TL |
9761 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, |
9762 | raw_bl); | |
7c673cae | 9763 | } |
11fdf7f2 TL |
9764 | for (auto& req : r2r) { |
9765 | for (auto& r : req.regs) { | |
9766 | ready_regions[r.logical_offset].substr_of( | |
9767 | raw_bl, r.blob_xoffset, r.length); | |
9768 | } | |
7c673cae FG |
9769 | } |
9770 | } else { | |
11fdf7f2 | 9771 | for (auto& req : r2r) { |
9f95a23c TL |
9772 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, |
9773 | req.regs.front().logical_offset) < 0) { | |
9774 | *csum_error = true; | |
9775 | return -EIO; | |
9776 | } | |
9777 | if (buffered) { | |
9778 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
9779 | req.r_off, req.bl); | |
9780 | } | |
7c673cae | 9781 | |
9f95a23c TL |
9782 | // prune and keep result |
9783 | for (const auto& r : req.regs) { | |
9784 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
11fdf7f2 | 9785 | } |
7c673cae FG |
9786 | } |
9787 | } | |
9788 | ++b2r_it; | |
9789 | } | |
9790 | ||
9791 | // generate a resulting buffer | |
9792 | auto pr = ready_regions.begin(); | |
9793 | auto pr_end = ready_regions.end(); | |
9f95a23c | 9794 | uint64_t pos = 0; |
7c673cae FG |
9795 | while (pos < length) { |
9796 | if (pr != pr_end && pr->first == pos + offset) { | |
9797 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9798 | << ": data from 0x" << pr->first << "~" << pr->second.length() |
9799 | << std::dec << dendl; | |
7c673cae FG |
9800 | pos += pr->second.length(); |
9801 | bl.claim_append(pr->second); | |
9802 | ++pr; | |
9803 | } else { | |
9804 | uint64_t l = length - pos; | |
9805 | if (pr != pr_end) { | |
11fdf7f2 | 9806 | ceph_assert(pr->first > pos + offset); |
9f95a23c | 9807 | l = pr->first - (pos + offset); |
7c673cae FG |
9808 | } |
9809 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9810 | << ": zeros for 0x" << (pos + offset) << "~" << l |
9811 | << std::dec << dendl; | |
7c673cae FG |
9812 | bl.append_zero(l); |
9813 | pos += l; | |
9814 | } | |
9815 | } | |
11fdf7f2 TL |
9816 | ceph_assert(bl.length() == length); |
9817 | ceph_assert(pos == length); | |
9818 | ceph_assert(pr == pr_end); | |
9f95a23c TL |
9819 | return 0; |
9820 | } | |
9821 | ||
9822 | int BlueStore::_do_read( | |
9823 | Collection *c, | |
9824 | OnodeRef o, | |
9825 | uint64_t offset, | |
9826 | size_t length, | |
9827 | bufferlist& bl, | |
9828 | uint32_t op_flags, | |
9829 | uint64_t retry_count) | |
9830 | { | |
9831 | FUNCTRACE(cct); | |
9832 | int r = 0; | |
9833 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
9834 | ||
9835 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
9836 | << " size 0x" << o->onode.size << " (" << std::dec | |
9837 | << o->onode.size << ")" << dendl; | |
9838 | bl.clear(); | |
9839 | ||
9840 | if (offset >= o->onode.size) { | |
9841 | return r; | |
9842 | } | |
9843 | ||
9844 | // generally, don't buffer anything, unless the client explicitly requests | |
9845 | // it. | |
9846 | bool buffered = false; | |
9847 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
9848 | dout(20) << __func__ << " will do buffered read" << dendl; | |
9849 | buffered = true; | |
9850 | } else if (cct->_conf->bluestore_default_buffered_read && | |
9851 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
9852 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
9853 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
9854 | buffered = true; | |
9855 | } | |
9856 | ||
9857 | if (offset + length > o->onode.size) { | |
9858 | length = o->onode.size - offset; | |
9859 | } | |
9860 | ||
9861 | auto start = mono_clock::now(); | |
9862 | o->extent_map.fault_range(db, offset, length); | |
9863 | log_latency(__func__, | |
9864 | l_bluestore_read_onode_meta_lat, | |
9865 | mono_clock::now() - start, | |
9866 | cct->_conf->bluestore_log_op_age); | |
9867 | _dump_onode<30>(cct, *o); | |
9868 | ||
9869 | // for deep-scrub, we only read dirty cache and bypass clean cache in | |
9870 | // order to read underlying block device in case there are silent disk errors. | |
9871 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
9872 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
9873 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
9874 | } | |
9875 | ||
9876 | // build blob-wise list to of stuff read (that isn't cached) | |
9877 | ready_regions_t ready_regions; | |
9878 | blobs2read_t blobs2read; | |
9879 | _read_cache(o, offset, length, read_cache_policy, ready_regions, blobs2read); | |
9880 | ||
9881 | ||
9882 | // read raw blob data. | |
9883 | start = mono_clock::now(); // for the sake of simplicity | |
9884 | // measure the whole block below. | |
9885 | // The error isn't that much... | |
9886 | vector<bufferlist> compressed_blob_bls; | |
9887 | IOContext ioc(cct, NULL, true); // allow EIO | |
9888 | r = _prepare_read_ioc(blobs2read, &compressed_blob_bls, &ioc); | |
9889 | // we always issue aio for reading, so errors other than EIO are not allowed | |
9890 | if (r < 0) | |
9891 | return r; | |
9892 | ||
f67539c2 | 9893 | int64_t num_ios = blobs2read.size(); |
9f95a23c | 9894 | if (ioc.has_pending_aios()) { |
f67539c2 | 9895 | num_ios = ioc.get_num_ios(); |
9f95a23c TL |
9896 | bdev->aio_submit(&ioc); |
9897 | dout(20) << __func__ << " waiting for aio" << dendl; | |
9898 | ioc.aio_wait(); | |
9899 | r = ioc.get_return_value(); | |
9900 | if (r < 0) { | |
9901 | ceph_assert(r == -EIO); // no other errors allowed | |
9902 | return -EIO; | |
9903 | } | |
9904 | } | |
9905 | log_latency_fn(__func__, | |
9906 | l_bluestore_read_wait_aio_lat, | |
9907 | mono_clock::now() - start, | |
9908 | cct->_conf->bluestore_log_op_age, | |
9909 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
9910 | ); | |
9911 | ||
9912 | bool csum_error = false; | |
9913 | r = _generate_read_result_bl(o, offset, length, ready_regions, | |
9914 | compressed_blob_bls, blobs2read, | |
9915 | buffered, &csum_error, bl); | |
9916 | if (csum_error) { | |
9917 | // Handles spurious read errors caused by a kernel bug. | |
9918 | // We sometimes get all-zero pages as a result of the read under | |
9919 | // high memory pressure. Retrying the failing read succeeds in most | |
9920 | // cases. | |
9921 | // See also: http://tracker.ceph.com/issues/22464 | |
9922 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
9923 | return -EIO; | |
9924 | } | |
9925 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
9926 | } | |
7c673cae | 9927 | r = bl.length(); |
f64942e4 AA |
9928 | if (retry_count) { |
9929 | logger->inc(l_bluestore_reads_with_retries); | |
9930 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
9931 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
f67539c2 TL |
9932 | stringstream s; |
9933 | s << " reads with retries: " << logger->get(l_bluestore_reads_with_retries); | |
9934 | _set_spurious_read_errors_alert(s.str()); | |
f64942e4 | 9935 | } |
7c673cae FG |
9936 | return r; |
9937 | } | |
9938 | ||
9939 | int BlueStore::_verify_csum(OnodeRef& o, | |
9940 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
9941 | const bufferlist& bl, | |
9942 | uint64_t logical_offset) const | |
9943 | { | |
9944 | int bad; | |
9945 | uint64_t bad_csum; | |
11fdf7f2 | 9946 | auto start = mono_clock::now(); |
7c673cae | 9947 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
9948 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
9949 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
9950 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
9951 | bad = blob_xoffset; | |
9952 | r = -1; | |
9953 | bad_csum = 0xDEADBEEF; | |
9954 | } | |
7c673cae FG |
9955 | if (r < 0) { |
9956 | if (r == -1) { | |
9957 | PExtentVector pex; | |
9958 | blob->map( | |
9959 | bad, | |
9960 | blob->get_csum_chunk_size(), | |
9961 | [&](uint64_t offset, uint64_t length) { | |
9962 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
9963 | return 0; | |
9964 | }); | |
9965 | derr << __func__ << " bad " | |
9966 | << Checksummer::get_csum_type_string(blob->csum_type) | |
9967 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
9968 | << " checksum at blob offset 0x" << bad | |
9969 | << ", got 0x" << bad_csum << ", expected 0x" | |
9970 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
9971 | << ", device location " << pex | |
9972 | << ", logical extent 0x" << std::hex | |
9973 | << (logical_offset + bad - blob_xoffset) << "~" | |
9974 | << blob->get_csum_chunk_size() << std::dec | |
9975 | << ", object " << o->oid | |
9976 | << dendl; | |
9977 | } else { | |
9978 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
9979 | } | |
9980 | } | |
494da23a TL |
9981 | log_latency(__func__, |
9982 | l_bluestore_csum_lat, | |
9983 | mono_clock::now() - start, | |
9984 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 TL |
9985 | if (cct->_conf->bluestore_ignore_data_csum) { |
9986 | return 0; | |
9987 | } | |
7c673cae FG |
9988 | return r; |
9989 | } | |
9990 | ||
9991 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
9992 | { | |
9993 | int r = 0; | |
11fdf7f2 TL |
9994 | auto start = mono_clock::now(); |
9995 | auto i = source.cbegin(); | |
7c673cae | 9996 | bluestore_compression_header_t chdr; |
11fdf7f2 | 9997 | decode(chdr, i); |
7c673cae FG |
9998 | int alg = int(chdr.type); |
9999 | CompressorRef cp = compressor; | |
10000 | if (!cp || (int)cp->get_type() != alg) { | |
10001 | cp = Compressor::create(cct, alg); | |
10002 | } | |
10003 | ||
10004 | if (!cp.get()) { | |
10005 | // if compressor isn't available - error, because cannot return | |
10006 | // decompressed data? | |
11fdf7f2 TL |
10007 | |
10008 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
10009 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
10010 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
10011 | r = -EIO; |
10012 | } else { | |
f67539c2 | 10013 | r = cp->decompress(i, chdr.length, *result, chdr.compressor_message); |
7c673cae FG |
10014 | if (r < 0) { |
10015 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
10016 | r = -EIO; | |
10017 | } | |
10018 | } | |
494da23a TL |
10019 | log_latency(__func__, |
10020 | l_bluestore_decompress_lat, | |
10021 | mono_clock::now() - start, | |
10022 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
10023 | return r; |
10024 | } | |
10025 | ||
10026 | // this stores fiemap into interval_set, other variations | |
10027 | // use it internally | |
10028 | int BlueStore::_fiemap( | |
10029 | CollectionHandle &c_, | |
10030 | const ghobject_t& oid, | |
10031 | uint64_t offset, | |
10032 | size_t length, | |
10033 | interval_set<uint64_t>& destset) | |
10034 | { | |
10035 | Collection *c = static_cast<Collection *>(c_.get()); | |
10036 | if (!c->exists) | |
10037 | return -ENOENT; | |
10038 | { | |
9f95a23c | 10039 | std::shared_lock l(c->lock); |
7c673cae FG |
10040 | |
10041 | OnodeRef o = c->get_onode(oid, false); | |
10042 | if (!o || !o->exists) { | |
10043 | return -ENOENT; | |
10044 | } | |
81eedcae | 10045 | _dump_onode<30>(cct, *o); |
7c673cae FG |
10046 | |
10047 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10048 | << " size 0x" << o->onode.size << std::dec << dendl; | |
10049 | ||
10050 | boost::intrusive::set<Extent>::iterator ep, eend; | |
10051 | if (offset >= o->onode.size) | |
10052 | goto out; | |
10053 | ||
10054 | if (offset + length > o->onode.size) { | |
10055 | length = o->onode.size - offset; | |
10056 | } | |
10057 | ||
10058 | o->extent_map.fault_range(db, offset, length); | |
10059 | eend = o->extent_map.extent_map.end(); | |
10060 | ep = o->extent_map.seek_lextent(offset); | |
10061 | while (length > 0) { | |
10062 | dout(20) << __func__ << " offset " << offset << dendl; | |
10063 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
10064 | ++ep; | |
10065 | continue; | |
10066 | } | |
10067 | ||
10068 | uint64_t x_len = length; | |
10069 | if (ep != eend && ep->logical_offset <= offset) { | |
10070 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 10071 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
10072 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
10073 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
10074 | destset.insert(offset, x_len); | |
10075 | length -= x_len; | |
10076 | offset += x_len; | |
10077 | if (x_off + x_len == ep->length) | |
10078 | ++ep; | |
10079 | continue; | |
10080 | } | |
10081 | if (ep != eend && | |
10082 | ep->logical_offset > offset && | |
10083 | ep->logical_offset - offset < x_len) { | |
10084 | x_len = ep->logical_offset - offset; | |
10085 | } | |
10086 | offset += x_len; | |
10087 | length -= x_len; | |
10088 | } | |
10089 | } | |
9f95a23c TL |
10090 | |
10091 | out: | |
10092 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10093 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
10094 | return 0; | |
10095 | } | |
10096 | ||
10097 | int BlueStore::fiemap( | |
10098 | CollectionHandle &c_, | |
10099 | const ghobject_t& oid, | |
10100 | uint64_t offset, | |
10101 | size_t length, | |
10102 | bufferlist& bl) | |
10103 | { | |
10104 | interval_set<uint64_t> m; | |
10105 | int r = _fiemap(c_, oid, offset, length, m); | |
10106 | if (r >= 0) { | |
10107 | encode(m, bl); | |
10108 | } | |
10109 | return r; | |
10110 | } | |
10111 | ||
10112 | int BlueStore::fiemap( | |
10113 | CollectionHandle &c_, | |
10114 | const ghobject_t& oid, | |
10115 | uint64_t offset, | |
10116 | size_t length, | |
10117 | map<uint64_t, uint64_t>& destmap) | |
10118 | { | |
10119 | interval_set<uint64_t> m; | |
10120 | int r = _fiemap(c_, oid, offset, length, m); | |
10121 | if (r >= 0) { | |
10122 | destmap = std::move(m).detach(); | |
10123 | } | |
10124 | return r; | |
10125 | } | |
10126 | ||
10127 | int BlueStore::readv( | |
10128 | CollectionHandle &c_, | |
10129 | const ghobject_t& oid, | |
10130 | interval_set<uint64_t>& m, | |
10131 | bufferlist& bl, | |
10132 | uint32_t op_flags) | |
10133 | { | |
10134 | auto start = mono_clock::now(); | |
10135 | Collection *c = static_cast<Collection *>(c_.get()); | |
10136 | const coll_t &cid = c->get_cid(); | |
10137 | dout(15) << __func__ << " " << cid << " " << oid | |
10138 | << " fiemap " << m | |
10139 | << dendl; | |
10140 | if (!c->exists) | |
10141 | return -ENOENT; | |
10142 | ||
10143 | bl.clear(); | |
10144 | int r; | |
10145 | { | |
10146 | std::shared_lock l(c->lock); | |
10147 | auto start1 = mono_clock::now(); | |
10148 | OnodeRef o = c->get_onode(oid, false); | |
10149 | log_latency("get_onode@read", | |
10150 | l_bluestore_read_onode_meta_lat, | |
10151 | mono_clock::now() - start1, | |
10152 | cct->_conf->bluestore_log_op_age); | |
10153 | if (!o || !o->exists) { | |
10154 | r = -ENOENT; | |
10155 | goto out; | |
10156 | } | |
10157 | ||
10158 | if (m.empty()) { | |
10159 | r = 0; | |
10160 | goto out; | |
10161 | } | |
10162 | ||
10163 | r = _do_readv(c, o, m, bl, op_flags); | |
10164 | if (r == -EIO) { | |
10165 | logger->inc(l_bluestore_read_eio); | |
10166 | } | |
10167 | } | |
10168 | ||
10169 | out: | |
10170 | if (r >= 0 && _debug_data_eio(oid)) { | |
10171 | r = -EIO; | |
10172 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10173 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ | |
10174 | cct->_conf->bluestore_debug_random_read_err && | |
10175 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
10176 | 100.0)) == 0) { | |
10177 | dout(0) << __func__ << ": inject random EIO" << dendl; | |
10178 | r = -EIO; | |
10179 | } | |
10180 | dout(10) << __func__ << " " << cid << " " << oid | |
10181 | << " fiemap " << m << std::dec | |
10182 | << " = " << r << dendl; | |
10183 | log_latency(__func__, | |
10184 | l_bluestore_read_lat, | |
10185 | mono_clock::now() - start, | |
10186 | cct->_conf->bluestore_log_op_age); | |
10187 | return r; | |
10188 | } | |
10189 | ||
10190 | int BlueStore::_do_readv( | |
10191 | Collection *c, | |
10192 | OnodeRef o, | |
10193 | const interval_set<uint64_t>& m, | |
10194 | bufferlist& bl, | |
10195 | uint32_t op_flags, | |
10196 | uint64_t retry_count) | |
10197 | { | |
10198 | FUNCTRACE(cct); | |
10199 | int r = 0; | |
10200 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
10201 | ||
10202 | dout(20) << __func__ << " fiemap " << m << std::hex | |
10203 | << " size 0x" << o->onode.size << " (" << std::dec | |
10204 | << o->onode.size << ")" << dendl; | |
10205 | ||
10206 | // generally, don't buffer anything, unless the client explicitly requests | |
10207 | // it. | |
10208 | bool buffered = false; | |
10209 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
10210 | dout(20) << __func__ << " will do buffered read" << dendl; | |
10211 | buffered = true; | |
10212 | } else if (cct->_conf->bluestore_default_buffered_read && | |
10213 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
10214 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
10215 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
10216 | buffered = true; | |
10217 | } | |
10218 | // this method must be idempotent since we may call it several times | |
10219 | // before we finally read the expected result. | |
10220 | bl.clear(); | |
10221 | ||
10222 | // call fiemap first! | |
10223 | ceph_assert(m.range_start() <= o->onode.size); | |
10224 | ceph_assert(m.range_end() <= o->onode.size); | |
10225 | auto start = mono_clock::now(); | |
10226 | o->extent_map.fault_range(db, m.range_start(), m.range_end() - m.range_start()); | |
10227 | log_latency(__func__, | |
10228 | l_bluestore_read_onode_meta_lat, | |
10229 | mono_clock::now() - start, | |
10230 | cct->_conf->bluestore_log_op_age); | |
10231 | _dump_onode<30>(cct, *o); | |
10232 | ||
10233 | IOContext ioc(cct, NULL, true); // allow EIO | |
10234 | vector<std::tuple<ready_regions_t, vector<bufferlist>, blobs2read_t>> raw_results; | |
10235 | raw_results.reserve(m.num_intervals()); | |
10236 | int i = 0; | |
10237 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10238 | raw_results.push_back({}); | |
10239 | _read_cache(o, p.get_start(), p.get_len(), read_cache_policy, | |
10240 | std::get<0>(raw_results[i]), std::get<2>(raw_results[i])); | |
10241 | r = _prepare_read_ioc(std::get<2>(raw_results[i]), &std::get<1>(raw_results[i]), &ioc); | |
10242 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10243 | if (r < 0) | |
10244 | return r; | |
10245 | } | |
10246 | ||
10247 | auto num_ios = m.size(); | |
10248 | if (ioc.has_pending_aios()) { | |
10249 | num_ios = ioc.get_num_ios(); | |
10250 | bdev->aio_submit(&ioc); | |
10251 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10252 | ioc.aio_wait(); | |
10253 | r = ioc.get_return_value(); | |
10254 | if (r < 0) { | |
10255 | ceph_assert(r == -EIO); // no other errors allowed | |
10256 | return -EIO; | |
10257 | } | |
10258 | } | |
10259 | log_latency_fn(__func__, | |
10260 | l_bluestore_read_wait_aio_lat, | |
10261 | mono_clock::now() - start, | |
10262 | cct->_conf->bluestore_log_op_age, | |
10263 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10264 | ); | |
10265 | ||
10266 | ceph_assert(raw_results.size() == (size_t)m.num_intervals()); | |
10267 | i = 0; | |
10268 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10269 | bool csum_error = false; | |
10270 | bufferlist t; | |
10271 | r = _generate_read_result_bl(o, p.get_start(), p.get_len(), | |
10272 | std::get<0>(raw_results[i]), | |
10273 | std::get<1>(raw_results[i]), | |
10274 | std::get<2>(raw_results[i]), | |
10275 | buffered, &csum_error, t); | |
10276 | if (csum_error) { | |
10277 | // Handles spurious read errors caused by a kernel bug. | |
10278 | // We sometimes get all-zero pages as a result of the read under | |
10279 | // high memory pressure. Retrying the failing read succeeds in most | |
10280 | // cases. | |
10281 | // See also: http://tracker.ceph.com/issues/22464 | |
10282 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10283 | return -EIO; | |
10284 | } | |
10285 | return _do_readv(c, o, m, bl, op_flags, retry_count + 1); | |
10286 | } | |
10287 | bl.claim_append(t); | |
10288 | } | |
10289 | if (retry_count) { | |
10290 | logger->inc(l_bluestore_reads_with_retries); | |
10291 | dout(5) << __func__ << " read fiemap " << m | |
10292 | << " failed " << retry_count << " times before succeeding" | |
10293 | << dendl; | |
10294 | } | |
10295 | return bl.length(); | |
7c673cae FG |
10296 | } |
10297 | ||
9f95a23c | 10298 | int BlueStore::dump_onode(CollectionHandle &c_, |
7c673cae | 10299 | const ghobject_t& oid, |
9f95a23c TL |
10300 | const string& section_name, |
10301 | Formatter *f) | |
7c673cae | 10302 | { |
9f95a23c TL |
10303 | Collection *c = static_cast<Collection *>(c_.get()); |
10304 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10305 | if (!c->exists) | |
10306 | return -ENOENT; | |
7c673cae | 10307 | |
9f95a23c TL |
10308 | int r; |
10309 | { | |
10310 | std::shared_lock l(c->lock); | |
10311 | ||
10312 | OnodeRef o = c->get_onode(oid, false); | |
10313 | if (!o || !o->exists) { | |
10314 | r = -ENOENT; | |
10315 | goto out; | |
10316 | } | |
10317 | // FIXME minor: actually the next line isn't enough to | |
10318 | // load shared blobs. Leaving as is for now.. | |
10319 | // | |
10320 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10321 | ||
10322 | _dump_onode<0>(cct, *o); | |
10323 | f->open_object_section(section_name.c_str()); | |
10324 | o->dump(f); | |
10325 | f->close_section(); | |
10326 | r = 0; | |
7c673cae | 10327 | } |
9f95a23c TL |
10328 | out: |
10329 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10330 | << " = " << r << dendl; | |
7c673cae FG |
10331 | return r; |
10332 | } | |
10333 | ||
7c673cae FG |
10334 | int BlueStore::getattr( |
10335 | CollectionHandle &c_, | |
10336 | const ghobject_t& oid, | |
10337 | const char *name, | |
10338 | bufferptr& value) | |
10339 | { | |
10340 | Collection *c = static_cast<Collection *>(c_.get()); | |
10341 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
10342 | if (!c->exists) | |
10343 | return -ENOENT; | |
10344 | ||
10345 | int r; | |
10346 | { | |
9f95a23c | 10347 | std::shared_lock l(c->lock); |
f91f0fd5 | 10348 | mempool::bluestore_cache_meta::string k(name); |
7c673cae FG |
10349 | |
10350 | OnodeRef o = c->get_onode(oid, false); | |
10351 | if (!o || !o->exists) { | |
10352 | r = -ENOENT; | |
10353 | goto out; | |
10354 | } | |
10355 | ||
10356 | if (!o->onode.attrs.count(k)) { | |
10357 | r = -ENODATA; | |
10358 | goto out; | |
10359 | } | |
10360 | value = o->onode.attrs[k]; | |
10361 | r = 0; | |
10362 | } | |
10363 | out: | |
7c673cae FG |
10364 | if (r == 0 && _debug_mdata_eio(oid)) { |
10365 | r = -EIO; | |
10366 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10367 | } | |
10368 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
10369 | << " = " << r << dendl; | |
10370 | return r; | |
10371 | } | |
10372 | ||
7c673cae FG |
10373 | int BlueStore::getattrs( |
10374 | CollectionHandle &c_, | |
10375 | const ghobject_t& oid, | |
10376 | map<string,bufferptr>& aset) | |
10377 | { | |
10378 | Collection *c = static_cast<Collection *>(c_.get()); | |
10379 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10380 | if (!c->exists) | |
10381 | return -ENOENT; | |
10382 | ||
10383 | int r; | |
10384 | { | |
9f95a23c | 10385 | std::shared_lock l(c->lock); |
7c673cae FG |
10386 | |
10387 | OnodeRef o = c->get_onode(oid, false); | |
10388 | if (!o || !o->exists) { | |
10389 | r = -ENOENT; | |
10390 | goto out; | |
10391 | } | |
10392 | for (auto& i : o->onode.attrs) { | |
10393 | aset.emplace(i.first.c_str(), i.second); | |
10394 | } | |
10395 | r = 0; | |
10396 | } | |
10397 | ||
10398 | out: | |
7c673cae FG |
10399 | if (r == 0 && _debug_mdata_eio(oid)) { |
10400 | r = -EIO; | |
10401 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10402 | } | |
10403 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10404 | << " = " << r << dendl; | |
10405 | return r; | |
10406 | } | |
10407 | ||
10408 | int BlueStore::list_collections(vector<coll_t>& ls) | |
10409 | { | |
9f95a23c | 10410 | std::shared_lock l(coll_lock); |
11fdf7f2 | 10411 | ls.reserve(coll_map.size()); |
7c673cae FG |
10412 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
10413 | p != coll_map.end(); | |
10414 | ++p) | |
10415 | ls.push_back(p->first); | |
10416 | return 0; | |
10417 | } | |
10418 | ||
10419 | bool BlueStore::collection_exists(const coll_t& c) | |
10420 | { | |
9f95a23c | 10421 | std::shared_lock l(coll_lock); |
7c673cae FG |
10422 | return coll_map.count(c); |
10423 | } | |
10424 | ||
11fdf7f2 | 10425 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 10426 | { |
11fdf7f2 | 10427 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
10428 | vector<ghobject_t> ls; |
10429 | ghobject_t next; | |
11fdf7f2 | 10430 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
10431 | &ls, &next); |
10432 | if (r < 0) { | |
10433 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
10434 | << dendl; | |
10435 | return r; | |
10436 | } | |
10437 | *empty = ls.empty(); | |
11fdf7f2 | 10438 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
10439 | return 0; |
10440 | } | |
10441 | ||
11fdf7f2 | 10442 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 10443 | { |
11fdf7f2 TL |
10444 | dout(15) << __func__ << " " << ch->cid << dendl; |
10445 | Collection *c = static_cast<Collection*>(ch.get()); | |
9f95a23c | 10446 | std::shared_lock l(c->lock); |
11fdf7f2 | 10447 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
10448 | return c->cnode.bits; |
10449 | } | |
10450 | ||
7c673cae FG |
10451 | int BlueStore::collection_list( |
10452 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10453 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10454 | { | |
10455 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 10456 | c->flush(); |
7c673cae FG |
10457 | dout(15) << __func__ << " " << c->cid |
10458 | << " start " << start << " end " << end << " max " << max << dendl; | |
10459 | int r; | |
10460 | { | |
9f95a23c | 10461 | std::shared_lock l(c->lock); |
f91f0fd5 TL |
10462 | r = _collection_list(c, start, end, max, false, ls, pnext); |
10463 | } | |
10464 | ||
10465 | dout(10) << __func__ << " " << c->cid | |
10466 | << " start " << start << " end " << end << " max " << max | |
10467 | << " = " << r << ", ls.size() = " << ls->size() | |
10468 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10469 | return r; | |
10470 | } | |
10471 | ||
10472 | int BlueStore::collection_list_legacy( | |
10473 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10474 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10475 | { | |
10476 | Collection *c = static_cast<Collection *>(c_.get()); | |
10477 | c->flush(); | |
10478 | dout(15) << __func__ << " " << c->cid | |
10479 | << " start " << start << " end " << end << " max " << max << dendl; | |
10480 | int r; | |
10481 | { | |
10482 | std::shared_lock l(c->lock); | |
10483 | r = _collection_list(c, start, end, max, true, ls, pnext); | |
7c673cae FG |
10484 | } |
10485 | ||
7c673cae FG |
10486 | dout(10) << __func__ << " " << c->cid |
10487 | << " start " << start << " end " << end << " max " << max | |
10488 | << " = " << r << ", ls.size() = " << ls->size() | |
10489 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10490 | return r; | |
10491 | } | |
10492 | ||
10493 | int BlueStore::_collection_list( | |
10494 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
f91f0fd5 | 10495 | bool legacy, vector<ghobject_t> *ls, ghobject_t *pnext) |
7c673cae FG |
10496 | { |
10497 | ||
10498 | if (!c->exists) | |
10499 | return -ENOENT; | |
10500 | ||
494da23a | 10501 | auto start_time = mono_clock::now(); |
7c673cae FG |
10502 | int r = 0; |
10503 | ghobject_t static_next; | |
f91f0fd5 TL |
10504 | std::unique_ptr<CollectionListIterator> it; |
10505 | ghobject_t coll_range_temp_start, coll_range_temp_end; | |
10506 | ghobject_t coll_range_start, coll_range_end; | |
7c673cae | 10507 | bool set_next = false; |
f91f0fd5 | 10508 | ghobject_t pend; |
7c673cae FG |
10509 | bool temp; |
10510 | ||
10511 | if (!pnext) | |
10512 | pnext = &static_next; | |
10513 | ||
11fdf7f2 | 10514 | if (start.is_max() || start.hobj.is_max()) { |
7c673cae FG |
10515 | goto out; |
10516 | } | |
f91f0fd5 TL |
10517 | get_coll_range(c->cid, c->cnode.bits, &coll_range_temp_start, |
10518 | &coll_range_temp_end, &coll_range_start, &coll_range_end); | |
7c673cae | 10519 | dout(20) << __func__ |
f91f0fd5 TL |
10520 | << " range " << coll_range_temp_start |
10521 | << " to " << coll_range_temp_end | |
10522 | << " and " << coll_range_start | |
10523 | << " to " << coll_range_end | |
7c673cae | 10524 | << " start " << start << dendl; |
f91f0fd5 TL |
10525 | if (legacy) { |
10526 | it = std::make_unique<SimpleCollectionListIterator>( | |
10527 | cct, db->get_iterator(PREFIX_OBJ)); | |
10528 | } else { | |
10529 | it = std::make_unique<SortedCollectionListIterator>( | |
10530 | db->get_iterator(PREFIX_OBJ)); | |
10531 | } | |
7c673cae FG |
10532 | if (start == ghobject_t() || |
10533 | start.hobj == hobject_t() || | |
10534 | start == c->cid.get_min_hobj()) { | |
f91f0fd5 | 10535 | it->upper_bound(coll_range_temp_start); |
7c673cae FG |
10536 | temp = true; |
10537 | } else { | |
7c673cae FG |
10538 | if (start.hobj.is_temp()) { |
10539 | temp = true; | |
f91f0fd5 | 10540 | ceph_assert(start >= coll_range_temp_start && start < coll_range_temp_end); |
7c673cae FG |
10541 | } else { |
10542 | temp = false; | |
f91f0fd5 | 10543 | ceph_assert(start >= coll_range_start && start < coll_range_end); |
7c673cae | 10544 | } |
f91f0fd5 TL |
10545 | dout(20) << __func__ << " temp=" << (int)temp << dendl; |
10546 | it->lower_bound(start); | |
7c673cae FG |
10547 | } |
10548 | if (end.hobj.is_max()) { | |
f91f0fd5 | 10549 | pend = temp ? coll_range_temp_end : coll_range_end; |
7c673cae | 10550 | } else { |
7c673cae FG |
10551 | if (end.hobj.is_temp()) { |
10552 | if (temp) | |
f91f0fd5 | 10553 | pend = end; |
7c673cae | 10554 | else |
f91f0fd5 | 10555 | goto out; |
7c673cae | 10556 | } else { |
f91f0fd5 | 10557 | pend = temp ? coll_range_temp_end : end; |
7c673cae FG |
10558 | } |
10559 | } | |
f91f0fd5 | 10560 | dout(20) << __func__ << " pend " << pend << dendl; |
7c673cae | 10561 | while (true) { |
adb31ebb | 10562 | if (!it->valid() || it->is_ge(pend)) { |
7c673cae FG |
10563 | if (!it->valid()) |
10564 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
10565 | else | |
f91f0fd5 | 10566 | dout(20) << __func__ << " oid " << it->oid() << " >= " << pend << dendl; |
7c673cae FG |
10567 | if (temp) { |
10568 | if (end.hobj.is_temp()) { | |
adb31ebb | 10569 | if (it->valid() && it->is_lt(coll_range_temp_end)) { |
f91f0fd5 TL |
10570 | *pnext = it->oid(); |
10571 | set_next = true; | |
10572 | } | |
7c673cae FG |
10573 | break; |
10574 | } | |
10575 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
10576 | temp = false; | |
f91f0fd5 TL |
10577 | it->upper_bound(coll_range_start); |
10578 | if (end.hobj.is_max()) | |
10579 | pend = coll_range_end; | |
10580 | else | |
10581 | pend = end; | |
10582 | dout(30) << __func__ << " pend " << pend << dendl; | |
7c673cae FG |
10583 | continue; |
10584 | } | |
adb31ebb | 10585 | if (it->valid() && it->is_lt(coll_range_end)) { |
f91f0fd5 TL |
10586 | *pnext = it->oid(); |
10587 | set_next = true; | |
10588 | } | |
7c673cae FG |
10589 | break; |
10590 | } | |
f91f0fd5 | 10591 | dout(20) << __func__ << " oid " << it->oid() << " end " << end << dendl; |
7c673cae FG |
10592 | if (ls->size() >= (unsigned)max) { |
10593 | dout(20) << __func__ << " reached max " << max << dendl; | |
f91f0fd5 | 10594 | *pnext = it->oid(); |
7c673cae FG |
10595 | set_next = true; |
10596 | break; | |
10597 | } | |
f91f0fd5 | 10598 | ls->push_back(it->oid()); |
7c673cae FG |
10599 | it->next(); |
10600 | } | |
10601 | out: | |
10602 | if (!set_next) { | |
10603 | *pnext = ghobject_t::get_max(); | |
10604 | } | |
494da23a TL |
10605 | log_latency_fn( |
10606 | __func__, | |
10607 | l_bluestore_clist_lat, | |
10608 | mono_clock::now() - start_time, | |
10609 | cct->_conf->bluestore_log_collection_list_age, | |
10610 | [&] (const ceph::timespan& lat) { | |
10611 | ostringstream ostr; | |
10612 | ostr << ", lat = " << timespan_str(lat) | |
10613 | << " cid =" << c->cid | |
10614 | << " start " << start << " end " << end | |
10615 | << " max " << max; | |
10616 | return ostr.str(); | |
10617 | } | |
10618 | ); | |
7c673cae FG |
10619 | return r; |
10620 | } | |
10621 | ||
7c673cae FG |
10622 | int BlueStore::omap_get( |
10623 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10624 | const ghobject_t &oid, ///< [in] Object containing omap | |
10625 | bufferlist *header, ///< [out] omap header | |
10626 | map<string, bufferlist> *out /// < [out] Key to value map | |
10627 | ) | |
10628 | { | |
10629 | Collection *c = static_cast<Collection *>(c_.get()); | |
9f95a23c TL |
10630 | return _omap_get(c, oid, header, out); |
10631 | } | |
10632 | ||
10633 | int BlueStore::_omap_get( | |
10634 | Collection *c, ///< [in] Collection containing oid | |
10635 | const ghobject_t &oid, ///< [in] Object containing omap | |
10636 | bufferlist *header, ///< [out] omap header | |
10637 | map<string, bufferlist> *out /// < [out] Key to value map | |
10638 | ) | |
10639 | { | |
7c673cae FG |
10640 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; |
10641 | if (!c->exists) | |
10642 | return -ENOENT; | |
9f95a23c | 10643 | std::shared_lock l(c->lock); |
7c673cae FG |
10644 | int r = 0; |
10645 | OnodeRef o = c->get_onode(oid, false); | |
10646 | if (!o || !o->exists) { | |
10647 | r = -ENOENT; | |
10648 | goto out; | |
10649 | } | |
9f95a23c TL |
10650 | r = _onode_omap_get(o, header, out); |
10651 | out: | |
10652 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10653 | << dendl; | |
10654 | return r; | |
10655 | } | |
10656 | ||
10657 | int BlueStore::_onode_omap_get( | |
10658 | const OnodeRef &o, ///< [in] Object containing omap | |
10659 | bufferlist *header, ///< [out] omap header | |
10660 | map<string, bufferlist> *out /// < [out] Key to value map | |
10661 | ) | |
10662 | { | |
10663 | int r = 0; | |
10664 | if (!o || !o->exists) { | |
10665 | r = -ENOENT; | |
10666 | goto out; | |
10667 | } | |
7c673cae FG |
10668 | if (!o->onode.has_omap()) |
10669 | goto out; | |
10670 | o->flush(); | |
10671 | { | |
9f95a23c | 10672 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10673 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10674 | string head, tail; |
9f95a23c TL |
10675 | o->get_omap_header(&head); |
10676 | o->get_omap_tail(&tail); | |
7c673cae FG |
10677 | it->lower_bound(head); |
10678 | while (it->valid()) { | |
10679 | if (it->key() == head) { | |
9f95a23c TL |
10680 | dout(30) << __func__ << " got header" << dendl; |
10681 | *header = it->value(); | |
7c673cae | 10682 | } else if (it->key() >= tail) { |
9f95a23c TL |
10683 | dout(30) << __func__ << " reached tail" << dendl; |
10684 | break; | |
7c673cae | 10685 | } else { |
9f95a23c TL |
10686 | string user_key; |
10687 | o->decode_omap_key(it->key(), &user_key); | |
10688 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
10689 | << " -> " << user_key << dendl; | |
10690 | (*out)[user_key] = it->value(); | |
7c673cae FG |
10691 | } |
10692 | it->next(); | |
10693 | } | |
10694 | } | |
9f95a23c | 10695 | out: |
7c673cae FG |
10696 | return r; |
10697 | } | |
10698 | ||
7c673cae FG |
10699 | int BlueStore::omap_get_header( |
10700 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10701 | const ghobject_t &oid, ///< [in] Object containing omap | |
10702 | bufferlist *header, ///< [out] omap header | |
10703 | bool allow_eio ///< [in] don't assert on eio | |
10704 | ) | |
10705 | { | |
10706 | Collection *c = static_cast<Collection *>(c_.get()); | |
10707 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10708 | if (!c->exists) | |
10709 | return -ENOENT; | |
9f95a23c | 10710 | std::shared_lock l(c->lock); |
7c673cae FG |
10711 | int r = 0; |
10712 | OnodeRef o = c->get_onode(oid, false); | |
10713 | if (!o || !o->exists) { | |
10714 | r = -ENOENT; | |
10715 | goto out; | |
10716 | } | |
10717 | if (!o->onode.has_omap()) | |
10718 | goto out; | |
10719 | o->flush(); | |
10720 | { | |
10721 | string head; | |
9f95a23c TL |
10722 | o->get_omap_header(&head); |
10723 | if (db->get(o->get_omap_prefix(), head, header) >= 0) { | |
7c673cae FG |
10724 | dout(30) << __func__ << " got header" << dendl; |
10725 | } else { | |
10726 | dout(30) << __func__ << " no header" << dendl; | |
10727 | } | |
10728 | } | |
10729 | out: | |
10730 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10731 | << dendl; | |
10732 | return r; | |
10733 | } | |
10734 | ||
7c673cae FG |
10735 | int BlueStore::omap_get_keys( |
10736 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10737 | const ghobject_t &oid, ///< [in] Object containing omap | |
10738 | set<string> *keys ///< [out] Keys defined on oid | |
10739 | ) | |
10740 | { | |
10741 | Collection *c = static_cast<Collection *>(c_.get()); | |
10742 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10743 | if (!c->exists) | |
10744 | return -ENOENT; | |
adb31ebb | 10745 | auto start1 = mono_clock::now(); |
9f95a23c | 10746 | std::shared_lock l(c->lock); |
7c673cae FG |
10747 | int r = 0; |
10748 | OnodeRef o = c->get_onode(oid, false); | |
10749 | if (!o || !o->exists) { | |
10750 | r = -ENOENT; | |
10751 | goto out; | |
10752 | } | |
10753 | if (!o->onode.has_omap()) | |
10754 | goto out; | |
10755 | o->flush(); | |
10756 | { | |
9f95a23c | 10757 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10758 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10759 | string head, tail; |
9f95a23c TL |
10760 | o->get_omap_key(string(), &head); |
10761 | o->get_omap_tail(&tail); | |
7c673cae FG |
10762 | it->lower_bound(head); |
10763 | while (it->valid()) { | |
10764 | if (it->key() >= tail) { | |
10765 | dout(30) << __func__ << " reached tail" << dendl; | |
10766 | break; | |
10767 | } | |
10768 | string user_key; | |
9f95a23c | 10769 | o->decode_omap_key(it->key(), &user_key); |
11fdf7f2 | 10770 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
10771 | << " -> " << user_key << dendl; |
10772 | keys->insert(user_key); | |
10773 | it->next(); | |
11fdf7f2 TL |
10774 | } |
10775 | } | |
10776 | out: | |
adb31ebb TL |
10777 | c->store->log_latency( |
10778 | __func__, | |
10779 | l_bluestore_omap_get_keys_lat, | |
10780 | mono_clock::now() - start1, | |
10781 | c->store->cct->_conf->bluestore_log_omap_iterator_age); | |
10782 | ||
11fdf7f2 TL |
10783 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r |
10784 | << dendl; | |
10785 | return r; | |
7c673cae FG |
10786 | } |
10787 | ||
10788 | int BlueStore::omap_get_values( | |
10789 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10790 | const ghobject_t &oid, ///< [in] Object containing omap | |
10791 | const set<string> &keys, ///< [in] Keys to get | |
10792 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
10793 | ) | |
10794 | { | |
10795 | Collection *c = static_cast<Collection *>(c_.get()); | |
10796 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10797 | if (!c->exists) | |
10798 | return -ENOENT; | |
9f95a23c | 10799 | std::shared_lock l(c->lock); |
adb31ebb | 10800 | auto start1 = mono_clock::now(); |
7c673cae FG |
10801 | int r = 0; |
10802 | string final_key; | |
10803 | OnodeRef o = c->get_onode(oid, false); | |
10804 | if (!o || !o->exists) { | |
10805 | r = -ENOENT; | |
10806 | goto out; | |
10807 | } | |
9f95a23c | 10808 | if (!o->onode.has_omap()) { |
7c673cae | 10809 | goto out; |
9f95a23c TL |
10810 | } |
10811 | o->flush(); | |
11fdf7f2 | 10812 | { |
9f95a23c TL |
10813 | const string& prefix = o->get_omap_prefix(); |
10814 | o->get_omap_key(string(), &final_key); | |
10815 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10816 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10817 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10818 | final_key += *p; |
10819 | bufferlist val; | |
10820 | if (db->get(prefix, final_key, &val) >= 0) { | |
10821 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
10822 | << " -> " << *p << dendl; | |
10823 | out->insert(make_pair(*p, val)); | |
10824 | } | |
7c673cae FG |
10825 | } |
10826 | } | |
10827 | out: | |
adb31ebb TL |
10828 | c->store->log_latency( |
10829 | __func__, | |
10830 | l_bluestore_omap_get_values_lat, | |
10831 | mono_clock::now() - start1, | |
10832 | c->store->cct->_conf->bluestore_log_omap_iterator_age); | |
10833 | ||
7c673cae FG |
10834 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r |
10835 | << dendl; | |
10836 | return r; | |
10837 | } | |
10838 | ||
9f95a23c TL |
10839 | #ifdef WITH_SEASTAR |
10840 | int BlueStore::omap_get_values( | |
10841 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10842 | const ghobject_t &oid, ///< [in] Object containing omap | |
10843 | const std::optional<string> &start_after, ///< [in] Keys to get | |
10844 | map<string, bufferlist> *output ///< [out] Returned keys and values | |
10845 | ) | |
10846 | { | |
10847 | Collection *c = static_cast<Collection *>(c_.get()); | |
10848 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10849 | if (!c->exists) | |
10850 | return -ENOENT; | |
10851 | std::shared_lock l(c->lock); | |
10852 | int r = 0; | |
10853 | OnodeRef o = c->get_onode(oid, false); | |
10854 | if (!o || !o->exists) { | |
10855 | r = -ENOENT; | |
10856 | goto out; | |
10857 | } | |
10858 | if (!o->onode.has_omap()) { | |
10859 | goto out; | |
10860 | } | |
10861 | o->flush(); | |
10862 | { | |
10863 | ObjectMap::ObjectMapIterator iter = get_omap_iterator(c_, oid); | |
10864 | if (!iter) { | |
10865 | r = -ENOENT; | |
10866 | goto out; | |
10867 | } | |
10868 | iter->upper_bound(*start_after); | |
10869 | for (; iter->valid(); iter->next()) { | |
10870 | output->insert(make_pair(iter->key(), iter->value())); | |
10871 | } | |
10872 | } | |
10873 | ||
10874 | out: | |
10875 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10876 | << dendl; | |
10877 | return r; | |
10878 | } | |
10879 | #endif | |
10880 | ||
7c673cae FG |
10881 | int BlueStore::omap_check_keys( |
10882 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10883 | const ghobject_t &oid, ///< [in] Object containing omap | |
10884 | const set<string> &keys, ///< [in] Keys to check | |
10885 | set<string> *out ///< [out] Subset of keys defined on oid | |
10886 | ) | |
10887 | { | |
10888 | Collection *c = static_cast<Collection *>(c_.get()); | |
10889 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10890 | if (!c->exists) | |
10891 | return -ENOENT; | |
9f95a23c | 10892 | std::shared_lock l(c->lock); |
7c673cae FG |
10893 | int r = 0; |
10894 | string final_key; | |
10895 | OnodeRef o = c->get_onode(oid, false); | |
10896 | if (!o || !o->exists) { | |
10897 | r = -ENOENT; | |
10898 | goto out; | |
10899 | } | |
9f95a23c | 10900 | if (!o->onode.has_omap()) { |
7c673cae | 10901 | goto out; |
9f95a23c TL |
10902 | } |
10903 | o->flush(); | |
11fdf7f2 | 10904 | { |
9f95a23c TL |
10905 | const string& prefix = o->get_omap_prefix(); |
10906 | o->get_omap_key(string(), &final_key); | |
10907 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10908 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10909 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10910 | final_key += *p; |
10911 | bufferlist val; | |
10912 | if (db->get(prefix, final_key, &val) >= 0) { | |
10913 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
10914 | << " -> " << *p << dendl; | |
10915 | out->insert(*p); | |
10916 | } else { | |
10917 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
10918 | << " -> " << *p << dendl; | |
10919 | } | |
7c673cae FG |
10920 | } |
10921 | } | |
10922 | out: | |
10923 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10924 | << dendl; | |
10925 | return r; | |
10926 | } | |
10927 | ||
7c673cae FG |
10928 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
10929 | CollectionHandle &c_, ///< [in] collection | |
10930 | const ghobject_t &oid ///< [in] object | |
10931 | ) | |
10932 | { | |
10933 | Collection *c = static_cast<Collection *>(c_.get()); | |
10934 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
10935 | if (!c->exists) { | |
10936 | return ObjectMap::ObjectMapIterator(); | |
10937 | } | |
9f95a23c | 10938 | std::shared_lock l(c->lock); |
7c673cae FG |
10939 | OnodeRef o = c->get_onode(oid, false); |
10940 | if (!o || !o->exists) { | |
10941 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
10942 | return ObjectMap::ObjectMapIterator(); | |
10943 | } | |
10944 | o->flush(); | |
10945 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
9f95a23c | 10946 | KeyValueDB::Iterator it = db->get_iterator(o->get_omap_prefix()); |
7c673cae FG |
10947 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(c, o, it)); |
10948 | } | |
10949 | ||
10950 | // ----------------- | |
10951 | // write helpers | |
10952 | ||
11fdf7f2 | 10953 | uint64_t BlueStore::_get_ondisk_reserved() const { |
f67539c2 | 10954 | ceph_assert(min_alloc_size); |
11fdf7f2 TL |
10955 | return round_up_to( |
10956 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
10957 | } | |
10958 | ||
7c673cae FG |
10959 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
10960 | { | |
10961 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
10962 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
10963 | << dendl; | |
11fdf7f2 | 10964 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
10965 | { |
10966 | bufferlist bl; | |
11fdf7f2 | 10967 | encode(ondisk_format, bl); |
7c673cae FG |
10968 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
10969 | } | |
10970 | { | |
10971 | bufferlist bl; | |
11fdf7f2 | 10972 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
10973 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
10974 | } | |
10975 | } | |
10976 | ||
10977 | int BlueStore::_open_super_meta() | |
10978 | { | |
10979 | // nid | |
10980 | { | |
10981 | nid_max = 0; | |
10982 | bufferlist bl; | |
10983 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 10984 | auto p = bl.cbegin(); |
7c673cae FG |
10985 | try { |
10986 | uint64_t v; | |
11fdf7f2 | 10987 | decode(v, p); |
7c673cae | 10988 | nid_max = v; |
f67539c2 | 10989 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
10990 | derr << __func__ << " unable to read nid_max" << dendl; |
10991 | return -EIO; | |
10992 | } | |
f67539c2 | 10993 | dout(1) << __func__ << " old nid_max " << nid_max << dendl; |
7c673cae FG |
10994 | nid_last = nid_max.load(); |
10995 | } | |
10996 | ||
10997 | // blobid | |
10998 | { | |
10999 | blobid_max = 0; | |
11000 | bufferlist bl; | |
11001 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 11002 | auto p = bl.cbegin(); |
7c673cae FG |
11003 | try { |
11004 | uint64_t v; | |
11fdf7f2 | 11005 | decode(v, p); |
7c673cae | 11006 | blobid_max = v; |
f67539c2 | 11007 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11008 | derr << __func__ << " unable to read blobid_max" << dendl; |
11009 | return -EIO; | |
11010 | } | |
f67539c2 | 11011 | dout(1) << __func__ << " old blobid_max " << blobid_max << dendl; |
7c673cae FG |
11012 | blobid_last = blobid_max.load(); |
11013 | } | |
11014 | ||
11015 | // freelist | |
11016 | { | |
11017 | bufferlist bl; | |
11018 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
11019 | if (bl.length()) { | |
11020 | freelist_type = std::string(bl.c_str(), bl.length()); | |
f67539c2 | 11021 | dout(1) << __func__ << " freelist_type " << freelist_type << dendl; |
7c673cae | 11022 | } else { |
11fdf7f2 | 11023 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 11024 | } |
7c673cae FG |
11025 | } |
11026 | ||
11027 | // ondisk format | |
11028 | int32_t compat_ondisk_format = 0; | |
11029 | { | |
11030 | bufferlist bl; | |
11031 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
11032 | if (r < 0) { | |
11033 | // base case: kraken bluestore is v1 and readable by v1 | |
11034 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
11035 | << dendl; | |
11036 | ondisk_format = 1; | |
11037 | compat_ondisk_format = 1; | |
11038 | } else { | |
11fdf7f2 | 11039 | auto p = bl.cbegin(); |
7c673cae | 11040 | try { |
11fdf7f2 | 11041 | decode(ondisk_format, p); |
f67539c2 | 11042 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11043 | derr << __func__ << " unable to read ondisk_format" << dendl; |
11044 | return -EIO; | |
11045 | } | |
11046 | bl.clear(); | |
11047 | { | |
11048 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
11049 | ceph_assert(!r); |
11050 | auto p = bl.cbegin(); | |
7c673cae | 11051 | try { |
11fdf7f2 | 11052 | decode(compat_ondisk_format, p); |
f67539c2 | 11053 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11054 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; |
11055 | return -EIO; | |
11056 | } | |
11057 | } | |
11058 | } | |
f67539c2 | 11059 | dout(1) << __func__ << " ondisk_format " << ondisk_format |
7c673cae FG |
11060 | << " compat_ondisk_format " << compat_ondisk_format |
11061 | << dendl; | |
11062 | } | |
11063 | ||
11064 | if (latest_ondisk_format < compat_ondisk_format) { | |
11065 | derr << __func__ << " compat_ondisk_format is " | |
11066 | << compat_ondisk_format << " but we only understand version " | |
11067 | << latest_ondisk_format << dendl; | |
11068 | return -EPERM; | |
11069 | } | |
7c673cae FG |
11070 | |
11071 | { | |
11072 | bufferlist bl; | |
11073 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 11074 | auto p = bl.cbegin(); |
7c673cae FG |
11075 | try { |
11076 | uint64_t val; | |
11fdf7f2 | 11077 | decode(val, p); |
7c673cae | 11078 | min_alloc_size = val; |
224ce89b | 11079 | min_alloc_size_order = ctz(val); |
11fdf7f2 | 11080 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
f67539c2 | 11081 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11082 | derr << __func__ << " unable to read min_alloc_size" << dendl; |
11083 | return -EIO; | |
11084 | } | |
f67539c2 | 11085 | dout(1) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size |
7c673cae FG |
11086 | << std::dec << dendl; |
11087 | } | |
9f95a23c TL |
11088 | |
11089 | _set_per_pool_omap(); | |
11090 | ||
224ce89b | 11091 | _open_statfs(); |
7c673cae FG |
11092 | _set_alloc_sizes(); |
11093 | _set_throttle_params(); | |
11094 | ||
11095 | _set_csum(); | |
11096 | _set_compression(); | |
11097 | _set_blob_size(); | |
11098 | ||
11fdf7f2 | 11099 | _validate_bdev(); |
7c673cae FG |
11100 | return 0; |
11101 | } | |
11102 | ||
11103 | int BlueStore::_upgrade_super() | |
11104 | { | |
11105 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
11106 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
11107 | if (ondisk_format < latest_ondisk_format) { |
11108 | ceph_assert(ondisk_format > 0); | |
11109 | ceph_assert(ondisk_format < latest_ondisk_format); | |
11110 | ||
1911f103 | 11111 | KeyValueDB::Transaction t = db->get_transaction(); |
11fdf7f2 TL |
11112 | if (ondisk_format == 1) { |
11113 | // changes: | |
11114 | // - super: added ondisk_format | |
11115 | // - super: added min_readable_ondisk_format | |
11116 | // - super: added min_compat_ondisk_format | |
11117 | // - super: added min_alloc_size | |
11118 | // - super: removed min_min_alloc_size | |
11fdf7f2 TL |
11119 | { |
11120 | bufferlist bl; | |
11121 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
11122 | auto p = bl.cbegin(); | |
11123 | try { | |
11124 | uint64_t val; | |
11125 | decode(val, p); | |
11126 | min_alloc_size = val; | |
f67539c2 | 11127 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
11128 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; |
11129 | return -EIO; | |
11130 | } | |
11131 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
11132 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 11133 | } |
11fdf7f2 | 11134 | ondisk_format = 2; |
7c673cae | 11135 | } |
9f95a23c TL |
11136 | if (ondisk_format == 2) { |
11137 | // changes: | |
f67539c2 TL |
11138 | // - onode has FLAG_PERPOOL_OMAP. Note that we do not know that *all* |
11139 | // oondes are using the per-pool prefix until a repair is run; at that | |
9f95a23c TL |
11140 | // point the per_pool_omap=1 key will be set. |
11141 | // - super: added per_pool_omap key, which indicates that *all* objects | |
11142 | // are using the new prefix and key format | |
11143 | ondisk_format = 3; | |
1911f103 TL |
11144 | } |
11145 | if (ondisk_format == 3) { | |
11146 | // changes: | |
11147 | // - FreelistManager keeps meta within bdev label | |
11148 | int r = _write_out_fm_meta(0); | |
9f95a23c | 11149 | ceph_assert(r == 0); |
1911f103 | 11150 | ondisk_format = 4; |
9f95a23c | 11151 | } |
1911f103 TL |
11152 | // This to be the last operation |
11153 | _prepare_ondisk_format_super(t); | |
11154 | int r = db->submit_transaction_sync(t); | |
11155 | ceph_assert(r == 0); | |
7c673cae | 11156 | } |
7c673cae FG |
11157 | // done |
11158 | dout(1) << __func__ << " done" << dendl; | |
11159 | return 0; | |
11160 | } | |
11161 | ||
11162 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef o) | |
11163 | { | |
224ce89b | 11164 | if (o->onode.nid) { |
11fdf7f2 | 11165 | ceph_assert(o->exists); |
7c673cae | 11166 | return; |
224ce89b | 11167 | } |
7c673cae FG |
11168 | uint64_t nid = ++nid_last; |
11169 | dout(20) << __func__ << " " << nid << dendl; | |
11170 | o->onode.nid = nid; | |
11171 | txc->last_nid = nid; | |
224ce89b | 11172 | o->exists = true; |
7c673cae FG |
11173 | } |
11174 | ||
11175 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
11176 | { | |
11177 | uint64_t bid = ++blobid_last; | |
11178 | dout(20) << __func__ << " " << bid << dendl; | |
11179 | txc->last_blobid = bid; | |
11180 | return bid; | |
11181 | } | |
11182 | ||
11183 | void BlueStore::get_db_statistics(Formatter *f) | |
11184 | { | |
11185 | db->get_statistics(f); | |
11186 | } | |
11187 | ||
11fdf7f2 TL |
11188 | BlueStore::TransContext *BlueStore::_txc_create( |
11189 | Collection *c, OpSequencer *osr, | |
f67539c2 TL |
11190 | list<Context*> *on_commits, |
11191 | TrackedOpRef osd_op) | |
7c673cae | 11192 | { |
11fdf7f2 | 11193 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae | 11194 | txc->t = db->get_transaction(); |
f67539c2 TL |
11195 | |
11196 | #ifdef WITH_BLKIN | |
11197 | if (osd_op && osd_op->pg_trace) { | |
11198 | txc->trace.init("TransContext", &trace_endpoint, | |
11199 | &osd_op->pg_trace); | |
11200 | txc->trace.event("txc create"); | |
11201 | txc->trace.keyval("txc seq", txc->seq); | |
11202 | } | |
11203 | #endif | |
11204 | ||
7c673cae FG |
11205 | osr->queue_new(txc); |
11206 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
11207 | << " seq " << txc->seq << dendl; | |
11208 | return txc; | |
11209 | } | |
11210 | ||
11211 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
11212 | { | |
11fdf7f2 TL |
11213 | // one "io" for the kv commit |
11214 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
11215 | auto cost = throttle_cost_per_io.load(); |
11216 | txc->cost = ios * cost + txc->bytes; | |
9f95a23c | 11217 | txc->ios = ios; |
7c673cae FG |
11218 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" |
11219 | << ios << " ios * " << cost << " + " << txc->bytes | |
11220 | << " bytes)" << dendl; | |
11221 | } | |
11222 | ||
11223 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
11224 | { | |
11225 | if (txc->statfs_delta.is_empty()) | |
11226 | return; | |
11227 | ||
11228 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
11229 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
11230 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
11231 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
11232 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
11233 | ||
11234 | bufferlist bl; | |
11235 | txc->statfs_delta.encode(bl); | |
11fdf7f2 TL |
11236 | if (per_pool_stat_collection) { |
11237 | string key; | |
11238 | get_pool_stat_key(txc->osd_pool_id, &key); | |
11239 | txc->t->merge(PREFIX_STAT, key, bl); | |
11240 | ||
11241 | std::lock_guard l(vstatfs_lock); | |
11242 | auto& stats = osd_pools[txc->osd_pool_id]; | |
11243 | stats += txc->statfs_delta; | |
11244 | ||
11245 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
11246 | ||
11247 | } else { | |
11248 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
7c673cae | 11249 | |
11fdf7f2 TL |
11250 | std::lock_guard l(vstatfs_lock); |
11251 | vstatfs += txc->statfs_delta; | |
11252 | } | |
7c673cae FG |
11253 | txc->statfs_delta.reset(); |
11254 | } | |
11255 | ||
11256 | void BlueStore::_txc_state_proc(TransContext *txc) | |
11257 | { | |
11258 | while (true) { | |
11259 | dout(10) << __func__ << " txc " << txc | |
11260 | << " " << txc->get_state_name() << dendl; | |
f67539c2 | 11261 | switch (txc->get_state()) { |
7c673cae | 11262 | case TransContext::STATE_PREPARE: |
9f95a23c | 11263 | throttle.log_state_latency(*txc, logger, l_bluestore_state_prepare_lat); |
7c673cae | 11264 | if (txc->ioc.has_pending_aios()) { |
f67539c2 TL |
11265 | txc->set_state(TransContext::STATE_AIO_WAIT); |
11266 | #ifdef WITH_BLKIN | |
11267 | if (txc->trace) { | |
11268 | txc->trace.keyval("pending aios", txc->ioc.num_pending.load()); | |
11269 | } | |
11270 | #endif | |
7c673cae FG |
11271 | txc->had_ios = true; |
11272 | _txc_aio_submit(txc); | |
11273 | return; | |
11274 | } | |
11275 | // ** fall-thru ** | |
11276 | ||
11277 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 | 11278 | { |
9f95a23c TL |
11279 | mono_clock::duration lat = throttle.log_state_latency( |
11280 | *txc, logger, l_bluestore_state_aio_wait_lat); | |
11281 | if (ceph::to_seconds<double>(lat) >= cct->_conf->bluestore_log_op_age) { | |
11fdf7f2 TL |
11282 | dout(0) << __func__ << " slow aio_wait, txc = " << txc |
11283 | << ", latency = " << lat | |
11284 | << dendl; | |
11285 | } | |
11286 | } | |
11287 | ||
7c673cae FG |
11288 | _txc_finish_io(txc); // may trigger blocked txc's too |
11289 | return; | |
11290 | ||
11291 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 11292 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
11293 | if (txc->had_ios) { |
11294 | ++txc->osr->txc_with_unstable_io; | |
11295 | } | |
9f95a23c | 11296 | throttle.log_state_latency(*txc, logger, l_bluestore_state_io_done_lat); |
f67539c2 | 11297 | txc->set_state(TransContext::STATE_KV_QUEUED); |
7c673cae FG |
11298 | if (cct->_conf->bluestore_sync_submit_transaction) { |
11299 | if (txc->last_nid >= nid_max || | |
11300 | txc->last_blobid >= blobid_max) { | |
11301 | dout(20) << __func__ | |
11302 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
11303 | << dendl; | |
11304 | } else if (txc->osr->kv_committing_serially) { | |
11305 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
11306 | << dendl; | |
11307 | // note: this is starvation-prone. once we have a txc in a busy | |
11308 | // sequencer that is committing serially it is possible to keep | |
11309 | // submitting new transactions fast enough that we get stuck doing | |
11310 | // so. the alternative is to block here... fixme? | |
11311 | } else if (txc->osr->txc_with_unstable_io) { | |
11312 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
11313 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
11314 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
11315 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
11316 | == 0) { | |
11317 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
11318 | << dendl; | |
11319 | } else { | |
9f95a23c | 11320 | _txc_apply_kv(txc, true); |
7c673cae FG |
11321 | } |
11322 | } | |
11323 | { | |
11fdf7f2 | 11324 | std::lock_guard l(kv_lock); |
7c673cae | 11325 | kv_queue.push_back(txc); |
9f95a23c TL |
11326 | if (!kv_sync_in_progress) { |
11327 | kv_sync_in_progress = true; | |
11328 | kv_cond.notify_one(); | |
11329 | } | |
f67539c2 | 11330 | if (txc->get_state() != TransContext::STATE_KV_SUBMITTED) { |
7c673cae FG |
11331 | kv_queue_unsubmitted.push_back(txc); |
11332 | ++txc->osr->kv_committing_serially; | |
11333 | } | |
31f18b77 FG |
11334 | if (txc->had_ios) |
11335 | kv_ios++; | |
11336 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
11337 | } |
11338 | return; | |
11339 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
11340 | _txc_committed_kv(txc); |
11341 | // ** fall-thru ** | |
11342 | ||
11343 | case TransContext::STATE_KV_DONE: | |
9f95a23c | 11344 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_done_lat); |
7c673cae | 11345 | if (txc->deferred_txn) { |
f67539c2 | 11346 | txc->set_state(TransContext::STATE_DEFERRED_QUEUED); |
7c673cae FG |
11347 | _deferred_queue(txc); |
11348 | return; | |
11349 | } | |
f67539c2 | 11350 | txc->set_state(TransContext::STATE_FINISHING); |
7c673cae FG |
11351 | break; |
11352 | ||
11353 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9f95a23c | 11354 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_cleanup_lat); |
f67539c2 | 11355 | txc->set_state(TransContext::STATE_FINISHING); |
7c673cae FG |
11356 | // ** fall-thru ** |
11357 | ||
11358 | case TransContext::STATE_FINISHING: | |
9f95a23c | 11359 | throttle.log_state_latency(*txc, logger, l_bluestore_state_finishing_lat); |
7c673cae FG |
11360 | _txc_finish(txc); |
11361 | return; | |
11362 | ||
11363 | default: | |
11364 | derr << __func__ << " unexpected txc " << txc | |
11365 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 11366 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
11367 | return; |
11368 | } | |
11369 | } | |
11370 | } | |
11371 | ||
11372 | void BlueStore::_txc_finish_io(TransContext *txc) | |
11373 | { | |
11374 | dout(20) << __func__ << " " << txc << dendl; | |
11375 | ||
11376 | /* | |
11377 | * we need to preserve the order of kv transactions, | |
11378 | * even though aio will complete in any order. | |
11379 | */ | |
11380 | ||
11381 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 11382 | std::lock_guard l(osr->qlock); |
f67539c2 | 11383 | txc->set_state(TransContext::STATE_IO_DONE); |
11fdf7f2 | 11384 | txc->ioc.release_running_aios(); |
7c673cae FG |
11385 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
11386 | while (p != osr->q.begin()) { | |
11387 | --p; | |
f67539c2 | 11388 | if (p->get_state() < TransContext::STATE_IO_DONE) { |
7c673cae FG |
11389 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " |
11390 | << p->get_state_name() << dendl; | |
11391 | return; | |
11392 | } | |
f67539c2 | 11393 | if (p->get_state() > TransContext::STATE_IO_DONE) { |
7c673cae FG |
11394 | ++p; |
11395 | break; | |
11396 | } | |
11397 | } | |
11398 | do { | |
11399 | _txc_state_proc(&*p++); | |
11400 | } while (p != osr->q.end() && | |
f67539c2 | 11401 | p->get_state() == TransContext::STATE_IO_DONE); |
7c673cae | 11402 | |
11fdf7f2 | 11403 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
11404 | osr->qcond.notify_all(); |
11405 | } | |
11406 | } | |
11407 | ||
11408 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
11409 | { | |
11410 | dout(20) << __func__ << " txc " << txc | |
11411 | << " onodes " << txc->onodes | |
11412 | << " shared_blobs " << txc->shared_blobs | |
11413 | << dendl; | |
11414 | ||
11415 | // finalize onodes | |
11416 | for (auto o : txc->onodes) { | |
11fdf7f2 | 11417 | _record_onode(o, t); |
7c673cae FG |
11418 | o->flushing_count++; |
11419 | } | |
11420 | ||
11421 | // objects we modified but didn't affect the onode | |
11422 | auto p = txc->modified_objects.begin(); | |
11423 | while (p != txc->modified_objects.end()) { | |
11424 | if (txc->onodes.count(*p) == 0) { | |
11425 | (*p)->flushing_count++; | |
11426 | ++p; | |
11427 | } else { | |
11428 | // remove dups with onodes list to avoid problems in _txc_finish | |
11429 | p = txc->modified_objects.erase(p); | |
11430 | } | |
11431 | } | |
11432 | ||
11433 | // finalize shared_blobs | |
11434 | for (auto sb : txc->shared_blobs) { | |
11435 | string key; | |
11436 | auto sbid = sb->get_sbid(); | |
11437 | get_shared_blob_key(sbid, &key); | |
11438 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
11439 | dout(20) << __func__ << " shared_blob 0x" |
11440 | << std::hex << sbid << std::dec | |
7c673cae FG |
11441 | << " is empty" << dendl; |
11442 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
11443 | } else { | |
11444 | bufferlist bl; | |
11fdf7f2 TL |
11445 | encode(*(sb->persistent), bl); |
11446 | dout(20) << __func__ << " shared_blob 0x" | |
11447 | << std::hex << sbid << std::dec | |
31f18b77 | 11448 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
11449 | t->set(PREFIX_SHARED_BLOB, key, bl); |
11450 | } | |
11451 | } | |
11452 | } | |
11453 | ||
11454 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
11455 | PerfCounters &logger) | |
11456 | { | |
11fdf7f2 TL |
11457 | os_commit_latency_ns.consume_next( |
11458 | logger.get_tavg_ns( | |
7c673cae | 11459 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
11460 | os_apply_latency_ns.consume_next( |
11461 | logger.get_tavg_ns( | |
7c673cae FG |
11462 | l_bluestore_commit_lat)); |
11463 | } | |
11464 | ||
f67539c2 TL |
11465 | // For every object we maintain <zone_num+oid, offset> tuple in the key-value |
11466 | // store. When a new object written to a zone, we insert the corresponding | |
11467 | // tuple to the database. When an object is truncated, we remove the | |
11468 | // corresponding tuple. When an object is overwritten, we remove the old tuple | |
11469 | // and insert a new tuple corresponding to the new location of the object. The | |
11470 | // cleaner can now identify live objects within the zone <zone_num> by | |
11471 | // enumerating all the keys starting with <zone_num> prefix. | |
11472 | void BlueStore::_zoned_update_cleaning_metadata(TransContext *txc) { | |
11473 | for (const auto &[o, offsets] : txc->zoned_onode_to_offset_map) { | |
11474 | std::string key; | |
11475 | get_object_key(cct, o->oid, &key); | |
11476 | for (auto offset : offsets) { | |
11477 | if (offset > 0) { | |
11478 | bufferlist offset_bl; | |
11479 | encode(offset, offset_bl); | |
11480 | txc->t->set(_zoned_get_prefix(offset), key, offset_bl); | |
11481 | } else { | |
11482 | txc->t->rmkey(_zoned_get_prefix(-offset), key); | |
11483 | } | |
11484 | } | |
11485 | } | |
11486 | } | |
11487 | ||
11488 | std::string BlueStore::_zoned_get_prefix(uint64_t offset) { | |
11489 | uint64_t zone_num = offset / bdev->get_zone_size(); | |
11490 | std::string zone_key; | |
11491 | _key_encode_u64(zone_num, &zone_key); | |
11492 | return PREFIX_ZONED_CL_INFO + zone_key; | |
11493 | } | |
11494 | ||
11495 | // For now, to avoid interface changes we piggyback zone_size (in MiB) and the | |
11496 | // first sequential zone number onto min_alloc_size and pass it to functions | |
11497 | // Allocator::create and FreelistManager::create. | |
11498 | uint64_t BlueStore::_zoned_piggyback_device_parameters_onto(uint64_t min_alloc_size) { | |
11499 | uint64_t zone_size = bdev->get_zone_size(); | |
11500 | uint64_t zone_size_mb = zone_size / (1024 * 1024); | |
11501 | uint64_t first_seq_zone = bdev->get_conventional_region_size() / zone_size; | |
11502 | min_alloc_size |= (zone_size_mb << 32); | |
11503 | min_alloc_size |= (first_seq_zone << 48); | |
11504 | return min_alloc_size; | |
11505 | } | |
11506 | ||
11507 | int BlueStore::_zoned_check_config_settings() { | |
11508 | if (cct->_conf->bluestore_allocator != "zoned") { | |
11509 | dout(1) << __func__ << " The drive is HM-SMR but " | |
11510 | << cct->_conf->bluestore_allocator << " allocator is specified. " | |
11511 | << "Only zoned allocator can be used with HM-SMR drive." << dendl; | |
11512 | return -EINVAL; | |
11513 | } | |
11514 | ||
11515 | // At least for now we want to use large min_alloc_size with HM-SMR drives. | |
11516 | // Populating used_blocks bitset on a debug build of ceph-osd takes about 5 | |
11517 | // minutes with a 14 TB HM-SMR drive and 4 KiB min_alloc_size. | |
11518 | if (min_alloc_size < 64 * 1024) { | |
11519 | dout(1) << __func__ << " The drive is HM-SMR but min_alloc_size is " | |
11520 | << min_alloc_size << ". " | |
11521 | << "Please set to at least 64 KiB." << dendl; | |
11522 | return -EINVAL; | |
11523 | } | |
11524 | ||
11525 | // We don't want to defer writes with HM-SMR because it violates sequential | |
11526 | // write requirement. | |
11527 | if (prefer_deferred_size) { | |
11528 | dout(1) << __func__ << " The drive is HM-SMR but prefer_deferred_size is " | |
11529 | << prefer_deferred_size << ". " | |
11530 | << "Please set to 0." << dendl; | |
11531 | return -EINVAL; | |
11532 | } | |
11533 | return 0; | |
11534 | } | |
11535 | ||
7c673cae FG |
11536 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) |
11537 | { | |
11538 | dout(20) << __func__ << " txc " << txc << std::hex | |
11539 | << " allocated 0x" << txc->allocated | |
11540 | << " released 0x" << txc->released | |
11541 | << std::dec << dendl; | |
11542 | ||
11543 | // We have to handle the case where we allocate *and* deallocate the | |
11544 | // same region in this transaction. The freelist doesn't like that. | |
11545 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
11546 | // debug check. But that's important.) | |
11547 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
11548 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
11549 | interval_set<uint64_t> *preleased = &txc->released; | |
11550 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
11551 | interval_set<uint64_t> overlap; | |
11552 | overlap.intersection_of(txc->allocated, txc->released); | |
11553 | if (!overlap.empty()) { | |
11554 | tmp_allocated = txc->allocated; | |
11555 | tmp_allocated.subtract(overlap); | |
11556 | tmp_released = txc->released; | |
11557 | tmp_released.subtract(overlap); | |
11558 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
11559 | << ", new allocated 0x" << tmp_allocated | |
11560 | << " released 0x" << tmp_released << std::dec | |
11561 | << dendl; | |
11562 | pallocated = &tmp_allocated; | |
11563 | preleased = &tmp_released; | |
11564 | } | |
11565 | } | |
11566 | ||
11567 | // update freelist with non-overlap sets | |
11568 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
11569 | p != pallocated->end(); | |
11570 | ++p) { | |
11571 | fm->allocate(p.get_start(), p.get_len(), t); | |
11572 | } | |
11573 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
11574 | p != preleased->end(); | |
11575 | ++p) { | |
11576 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
11577 | << "~" << p.get_len() << std::dec << dendl; | |
11578 | fm->release(p.get_start(), p.get_len(), t); | |
11579 | } | |
11580 | ||
f67539c2 TL |
11581 | if (bdev->is_smr()) { |
11582 | _zoned_update_cleaning_metadata(txc); | |
11583 | } | |
11584 | ||
7c673cae FG |
11585 | _txc_update_store_statfs(txc); |
11586 | } | |
11587 | ||
9f95a23c | 11588 | void BlueStore::_txc_apply_kv(TransContext *txc, bool sync_submit_transaction) |
7c673cae | 11589 | { |
f67539c2 | 11590 | ceph_assert(txc->get_state() == TransContext::STATE_KV_QUEUED); |
9f95a23c TL |
11591 | { |
11592 | #if defined(WITH_LTTNG) | |
11593 | auto start = mono_clock::now(); | |
11594 | #endif | |
11595 | ||
f67539c2 TL |
11596 | #ifdef WITH_BLKIN |
11597 | if (txc->trace) { | |
11598 | txc->trace.event("db async submit"); | |
11599 | } | |
11600 | #endif | |
11601 | ||
9f95a23c TL |
11602 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); |
11603 | ceph_assert(r == 0); | |
f67539c2 | 11604 | txc->set_state(TransContext::STATE_KV_SUBMITTED); |
9f95a23c TL |
11605 | if (txc->osr->kv_submitted_waiters) { |
11606 | std::lock_guard l(txc->osr->qlock); | |
11607 | txc->osr->qcond.notify_all(); | |
11608 | } | |
11609 | ||
11610 | #if defined(WITH_LTTNG) | |
11611 | if (txc->tracing) { | |
11612 | tracepoint( | |
11613 | bluestore, | |
11614 | transaction_kv_submit_latency, | |
11615 | txc->osr->get_sequencer_id(), | |
11616 | txc->seq, | |
11617 | sync_submit_transaction, | |
11618 | ceph::to_seconds<double>(mono_clock::now() - start)); | |
11619 | } | |
11620 | #endif | |
11621 | } | |
11622 | ||
7c673cae FG |
11623 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { |
11624 | for (auto& o : *ls) { | |
11625 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
11626 | << dendl; | |
9f95a23c | 11627 | if (--o->flushing_count == 0 && o->waiting_count.load()) { |
11fdf7f2 | 11628 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
11629 | o->flush_cond.notify_all(); |
11630 | } | |
11631 | } | |
11632 | } | |
11633 | } | |
11634 | ||
11635 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
11636 | { | |
11637 | dout(20) << __func__ << " txc " << txc << dendl; | |
9f95a23c | 11638 | throttle.complete_kv(*txc); |
1adf2230 | 11639 | { |
11fdf7f2 | 11640 | std::lock_guard l(txc->osr->qlock); |
f67539c2 | 11641 | txc->set_state(TransContext::STATE_KV_DONE); |
11fdf7f2 TL |
11642 | if (txc->ch->commit_queue) { |
11643 | txc->ch->commit_queue->queue(txc->oncommits); | |
11644 | } else { | |
11645 | finisher.queue(txc->oncommits); | |
1adf2230 | 11646 | } |
7c673cae | 11647 | } |
9f95a23c | 11648 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_committing_lat); |
494da23a TL |
11649 | log_latency_fn( |
11650 | __func__, | |
11651 | l_bluestore_commit_lat, | |
9f95a23c | 11652 | mono_clock::now() - txc->start, |
494da23a TL |
11653 | cct->_conf->bluestore_log_op_age, |
11654 | [&](auto lat) { | |
11655 | return ", txc = " + stringify(txc); | |
11656 | } | |
11fdf7f2 | 11657 | ); |
7c673cae FG |
11658 | } |
11659 | ||
11660 | void BlueStore::_txc_finish(TransContext *txc) | |
11661 | { | |
11662 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
f67539c2 | 11663 | ceph_assert(txc->get_state() == TransContext::STATE_FINISHING); |
7c673cae FG |
11664 | |
11665 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 11666 | sb->finish_write(txc->seq); |
7c673cae FG |
11667 | } |
11668 | txc->shared_blobs_written.clear(); | |
11669 | ||
11670 | while (!txc->removed_collections.empty()) { | |
11671 | _queue_reap_collection(txc->removed_collections.front()); | |
11672 | txc->removed_collections.pop_front(); | |
11673 | } | |
11674 | ||
11675 | OpSequencerRef osr = txc->osr; | |
7c673cae | 11676 | bool empty = false; |
31f18b77 | 11677 | bool submit_deferred = false; |
7c673cae FG |
11678 | OpSequencer::q_list_t releasing_txc; |
11679 | { | |
11fdf7f2 | 11680 | std::lock_guard l(osr->qlock); |
f67539c2 | 11681 | txc->set_state(TransContext::STATE_DONE); |
7c673cae FG |
11682 | bool notify = false; |
11683 | while (!osr->q.empty()) { | |
11684 | TransContext *txc = &osr->q.front(); | |
11685 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
11686 | << dendl; | |
f67539c2 TL |
11687 | if (txc->get_state() != TransContext::STATE_DONE) { |
11688 | if (txc->get_state() == TransContext::STATE_PREPARE && | |
7c673cae FG |
11689 | deferred_aggressive) { |
11690 | // for _osr_drain_preceding() | |
11691 | notify = true; | |
11692 | } | |
f67539c2 | 11693 | if (txc->get_state() == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 11694 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
11695 | submit_deferred = true; |
11696 | } | |
7c673cae FG |
11697 | break; |
11698 | } | |
11699 | ||
7c673cae FG |
11700 | osr->q.pop_front(); |
11701 | releasing_txc.push_back(*txc); | |
7c673cae | 11702 | } |
9f95a23c | 11703 | |
7c673cae FG |
11704 | if (osr->q.empty()) { |
11705 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
11706 | empty = true; | |
11707 | } | |
9f95a23c TL |
11708 | |
11709 | // only drain()/drain_preceding() need wakeup, | |
11710 | // other cases use kv_submitted_waiters | |
11711 | if (notify || empty) { | |
11712 | osr->qcond.notify_all(); | |
11713 | } | |
7c673cae | 11714 | } |
9f95a23c | 11715 | |
7c673cae FG |
11716 | while (!releasing_txc.empty()) { |
11717 | // release to allocator only after all preceding txc's have also | |
11718 | // finished any deferred writes that potentially land in these | |
11719 | // blocks | |
11720 | auto txc = &releasing_txc.front(); | |
11721 | _txc_release_alloc(txc); | |
11722 | releasing_txc.pop_front(); | |
9f95a23c TL |
11723 | throttle.log_state_latency(*txc, logger, l_bluestore_state_done_lat); |
11724 | throttle.complete(*txc); | |
7c673cae FG |
11725 | delete txc; |
11726 | } | |
11727 | ||
31f18b77 FG |
11728 | if (submit_deferred) { |
11729 | // we're pinning memory; flush! we could be more fine-grained here but | |
11730 | // i'm not sure it's worth the bother. | |
11731 | deferred_try_submit(); | |
7c673cae FG |
11732 | } |
11733 | ||
7c673cae | 11734 | if (empty && osr->zombie) { |
11fdf7f2 TL |
11735 | std::lock_guard l(zombie_osr_lock); |
11736 | if (zombie_osr_set.erase(osr->cid)) { | |
11737 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11738 | } else { | |
11739 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
11740 | << dendl; | |
11741 | } | |
7c673cae | 11742 | } |
9f95a23c | 11743 | } |
7c673cae FG |
11744 | |
11745 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
11746 | { | |
a8e16298 | 11747 | // it's expected we're called with lazy_release_lock already taken! |
11fdf7f2 TL |
11748 | if (likely(!cct->_conf->bluestore_debug_no_reuse_blocks)) { |
11749 | int r = 0; | |
11750 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
11751 | r = bdev->queue_discard(txc->released); | |
11752 | if (r == 0) { | |
11753 | dout(10) << __func__ << "(queued) " << txc << " " << std::hex | |
11754 | << txc->released << std::dec << dendl; | |
11755 | goto out; | |
11756 | } | |
11757 | } else if (cct->_conf->bdev_enable_discard) { | |
11758 | for (auto p = txc->released.begin(); p != txc->released.end(); ++p) { | |
11759 | bdev->discard(p.get_start(), p.get_len()); | |
11760 | } | |
11761 | } | |
11762 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
94b18763 | 11763 | << txc->released << std::dec << dendl; |
f67539c2 | 11764 | shared_alloc.a->release(txc->released); |
7c673cae FG |
11765 | } |
11766 | ||
11fdf7f2 | 11767 | out: |
7c673cae FG |
11768 | txc->allocated.clear(); |
11769 | txc->released.clear(); | |
11770 | } | |
11771 | ||
11fdf7f2 TL |
11772 | void BlueStore::_osr_attach(Collection *c) |
11773 | { | |
11774 | // note: caller has RWLock on coll_map | |
11775 | auto q = coll_map.find(c->cid); | |
11776 | if (q != coll_map.end()) { | |
11777 | c->osr = q->second->osr; | |
11778 | ldout(cct, 10) << __func__ << " " << c->cid | |
11779 | << " reusing osr " << c->osr << " from existing coll " | |
11780 | << q->second << dendl; | |
11781 | } else { | |
11782 | std::lock_guard l(zombie_osr_lock); | |
11783 | auto p = zombie_osr_set.find(c->cid); | |
11784 | if (p == zombie_osr_set.end()) { | |
9f95a23c | 11785 | c->osr = ceph::make_ref<OpSequencer>(this, next_sequencer_id++, c->cid); |
11fdf7f2 TL |
11786 | ldout(cct, 10) << __func__ << " " << c->cid |
11787 | << " fresh osr " << c->osr << dendl; | |
11788 | } else { | |
11789 | c->osr = p->second; | |
11790 | zombie_osr_set.erase(p); | |
11791 | ldout(cct, 10) << __func__ << " " << c->cid | |
11792 | << " resurrecting zombie osr " << c->osr << dendl; | |
11793 | c->osr->zombie = false; | |
11794 | } | |
11795 | } | |
11796 | } | |
11797 | ||
11798 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
11799 | { | |
11800 | std::lock_guard l(zombie_osr_lock); | |
11801 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
11802 | osr->zombie = true; | |
11803 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
11804 | // this is either a new insertion or the same osr is already there | |
11805 | ceph_assert(i.second || i.first->second == osr); | |
11806 | } | |
11807 | ||
7c673cae FG |
11808 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
11809 | { | |
11810 | OpSequencer *osr = txc->osr.get(); | |
11811 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
11812 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11813 | { | |
11814 | // submit anything pending | |
f67539c2 | 11815 | osr->deferred_lock.lock(); |
11fdf7f2 | 11816 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
11817 | _deferred_submit_unlock(osr); |
11818 | } else { | |
f67539c2 | 11819 | osr->deferred_lock.unlock(); |
7c673cae FG |
11820 | } |
11821 | } | |
11822 | { | |
11823 | // wake up any previously finished deferred events | |
11fdf7f2 | 11824 | std::lock_guard l(kv_lock); |
9f95a23c TL |
11825 | if (!kv_sync_in_progress) { |
11826 | kv_sync_in_progress = true; | |
11827 | kv_cond.notify_one(); | |
11828 | } | |
7c673cae FG |
11829 | } |
11830 | osr->drain_preceding(txc); | |
11831 | --deferred_aggressive; | |
11832 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11833 | } | |
11834 | ||
11fdf7f2 TL |
11835 | void BlueStore::_osr_drain(OpSequencer *osr) |
11836 | { | |
11837 | dout(10) << __func__ << " " << osr << dendl; | |
11838 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11839 | { | |
11840 | // submit anything pending | |
f67539c2 | 11841 | osr->deferred_lock.lock(); |
11fdf7f2 TL |
11842 | if (osr->deferred_pending && !osr->deferred_running) { |
11843 | _deferred_submit_unlock(osr); | |
11844 | } else { | |
f67539c2 | 11845 | osr->deferred_lock.unlock(); |
11fdf7f2 TL |
11846 | } |
11847 | } | |
11848 | { | |
11849 | // wake up any previously finished deferred events | |
11850 | std::lock_guard l(kv_lock); | |
9f95a23c TL |
11851 | if (!kv_sync_in_progress) { |
11852 | kv_sync_in_progress = true; | |
11853 | kv_cond.notify_one(); | |
11854 | } | |
11fdf7f2 TL |
11855 | } |
11856 | osr->drain(); | |
11857 | --deferred_aggressive; | |
11858 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11859 | } | |
11860 | ||
7c673cae FG |
11861 | void BlueStore::_osr_drain_all() |
11862 | { | |
11863 | dout(10) << __func__ << dendl; | |
11864 | ||
11865 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
11866 | vector<OpSequencerRef> zombies; |
11867 | { | |
9f95a23c | 11868 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
11869 | for (auto& i : coll_map) { |
11870 | s.insert(i.second->osr); | |
11871 | } | |
11872 | } | |
7c673cae | 11873 | { |
11fdf7f2 TL |
11874 | std::lock_guard l(zombie_osr_lock); |
11875 | for (auto& i : zombie_osr_set) { | |
11876 | s.insert(i.second); | |
11877 | zombies.push_back(i.second); | |
11878 | } | |
7c673cae FG |
11879 | } |
11880 | dout(20) << __func__ << " osr_set " << s << dendl; | |
11881 | ||
11882 | ++deferred_aggressive; | |
11883 | { | |
11884 | // submit anything pending | |
224ce89b | 11885 | deferred_try_submit(); |
7c673cae FG |
11886 | } |
11887 | { | |
11888 | // wake up any previously finished deferred events | |
11fdf7f2 | 11889 | std::lock_guard l(kv_lock); |
7c673cae FG |
11890 | kv_cond.notify_one(); |
11891 | } | |
31f18b77 | 11892 | { |
11fdf7f2 | 11893 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11894 | kv_finalize_cond.notify_one(); |
11895 | } | |
7c673cae FG |
11896 | for (auto osr : s) { |
11897 | dout(20) << __func__ << " drain " << osr << dendl; | |
11898 | osr->drain(); | |
11899 | } | |
11900 | --deferred_aggressive; | |
11901 | ||
7c673cae | 11902 | { |
11fdf7f2 TL |
11903 | std::lock_guard l(zombie_osr_lock); |
11904 | for (auto& osr : zombies) { | |
11905 | if (zombie_osr_set.erase(osr->cid)) { | |
11906 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11907 | ceph_assert(osr->q.empty()); | |
11908 | } else if (osr->zombie) { | |
11909 | dout(10) << __func__ << " empty zombie osr " << osr | |
11910 | << " already reaped" << dendl; | |
11911 | ceph_assert(osr->q.empty()); | |
11912 | } else { | |
11913 | dout(10) << __func__ << " empty zombie osr " << osr | |
11914 | << " resurrected" << dendl; | |
11915 | } | |
7c673cae FG |
11916 | } |
11917 | } | |
11fdf7f2 TL |
11918 | |
11919 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
11920 | } |
11921 | ||
11fdf7f2 | 11922 | |
31f18b77 FG |
11923 | void BlueStore::_kv_start() |
11924 | { | |
11925 | dout(10) << __func__ << dendl; | |
11926 | ||
11fdf7f2 | 11927 | finisher.start(); |
31f18b77 FG |
11928 | kv_sync_thread.create("bstore_kv_sync"); |
11929 | kv_finalize_thread.create("bstore_kv_final"); | |
11930 | } | |
11931 | ||
11932 | void BlueStore::_kv_stop() | |
11933 | { | |
11934 | dout(10) << __func__ << dendl; | |
11935 | { | |
9f95a23c | 11936 | std::unique_lock l{kv_lock}; |
31f18b77 FG |
11937 | while (!kv_sync_started) { |
11938 | kv_cond.wait(l); | |
11939 | } | |
11940 | kv_stop = true; | |
11941 | kv_cond.notify_all(); | |
11942 | } | |
11943 | { | |
9f95a23c | 11944 | std::unique_lock l{kv_finalize_lock}; |
31f18b77 FG |
11945 | while (!kv_finalize_started) { |
11946 | kv_finalize_cond.wait(l); | |
11947 | } | |
11948 | kv_finalize_stop = true; | |
11949 | kv_finalize_cond.notify_all(); | |
11950 | } | |
11951 | kv_sync_thread.join(); | |
11952 | kv_finalize_thread.join(); | |
11fdf7f2 | 11953 | ceph_assert(removed_collections.empty()); |
31f18b77 | 11954 | { |
11fdf7f2 | 11955 | std::lock_guard l(kv_lock); |
31f18b77 FG |
11956 | kv_stop = false; |
11957 | } | |
11958 | { | |
11fdf7f2 | 11959 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11960 | kv_finalize_stop = false; |
11961 | } | |
11962 | dout(10) << __func__ << " stopping finishers" << dendl; | |
11fdf7f2 TL |
11963 | finisher.wait_for_empty(); |
11964 | finisher.stop(); | |
31f18b77 FG |
11965 | dout(10) << __func__ << " stopped" << dendl; |
11966 | } | |
11967 | ||
7c673cae FG |
11968 | void BlueStore::_kv_sync_thread() |
11969 | { | |
11970 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 | 11971 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
9f95a23c | 11972 | std::unique_lock l{kv_lock}; |
11fdf7f2 | 11973 | ceph_assert(!kv_sync_started); |
31f18b77 FG |
11974 | kv_sync_started = true; |
11975 | kv_cond.notify_all(); | |
adb31ebb TL |
11976 | |
11977 | auto t0 = mono_clock::now(); | |
11978 | timespan twait = ceph::make_timespan(0); | |
11979 | size_t kv_submitted = 0; | |
11980 | ||
7c673cae | 11981 | while (true) { |
adb31ebb TL |
11982 | auto period = cct->_conf->bluestore_kv_sync_util_logging_s; |
11983 | auto observation_period = | |
11984 | ceph::make_timespan(period); | |
11985 | auto elapsed = mono_clock::now() - t0; | |
11986 | if (period && elapsed >= observation_period) { | |
11987 | dout(5) << __func__ << " utilization: idle " | |
11988 | << twait << " of " << elapsed | |
11989 | << ", submitted: " << kv_submitted | |
11990 | <<dendl; | |
11991 | t0 = mono_clock::now(); | |
11992 | twait = ceph::make_timespan(0); | |
11993 | kv_submitted = 0; | |
11994 | } | |
11fdf7f2 | 11995 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
11996 | if (kv_queue.empty() && |
11997 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 11998 | !deferred_aggressive)) { |
7c673cae FG |
11999 | if (kv_stop) |
12000 | break; | |
12001 | dout(20) << __func__ << " sleep" << dendl; | |
adb31ebb | 12002 | auto t = mono_clock::now(); |
9f95a23c | 12003 | kv_sync_in_progress = false; |
11fdf7f2 | 12004 | kv_cond.wait(l); |
adb31ebb TL |
12005 | twait += mono_clock::now() - t; |
12006 | ||
7c673cae FG |
12007 | dout(20) << __func__ << " wake" << dendl; |
12008 | } else { | |
12009 | deque<TransContext*> kv_submitting; | |
12010 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
12011 | uint64_t aios = 0, costs = 0; |
12012 | ||
7c673cae FG |
12013 | dout(20) << __func__ << " committing " << kv_queue.size() |
12014 | << " submitting " << kv_queue_unsubmitted.size() | |
12015 | << " deferred done " << deferred_done_queue.size() | |
12016 | << " stable " << deferred_stable_queue.size() | |
12017 | << dendl; | |
12018 | kv_committing.swap(kv_queue); | |
12019 | kv_submitting.swap(kv_queue_unsubmitted); | |
12020 | deferred_done.swap(deferred_done_queue); | |
12021 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
12022 | aios = kv_ios; |
12023 | costs = kv_throttle_costs; | |
12024 | kv_ios = 0; | |
12025 | kv_throttle_costs = 0; | |
7c673cae FG |
12026 | l.unlock(); |
12027 | ||
12028 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
12029 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
12030 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
12031 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12032 | ||
11fdf7f2 TL |
12033 | auto start = mono_clock::now(); |
12034 | ||
7c673cae FG |
12035 | bool force_flush = false; |
12036 | // if bluefs is sharing the same device as data (only), then we | |
12037 | // can rely on the bluefs commit to flush the device and make | |
12038 | // deferred aios stable. that means that if we do have done deferred | |
12039 | // txcs AND we are not on a single device, we need to force a flush. | |
9f95a23c | 12040 | if (bluefs && bluefs_layout.single_shared_device()) { |
31f18b77 | 12041 | if (aios) { |
7c673cae | 12042 | force_flush = true; |
11fdf7f2 | 12043 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
12044 | force_flush = true; // there's nothing else to commit! |
12045 | } else if (deferred_aggressive) { | |
12046 | force_flush = true; | |
12047 | } | |
11fdf7f2 TL |
12048 | } else { |
12049 | if (aios || !deferred_done.empty()) { | |
12050 | force_flush = true; | |
12051 | } else { | |
12052 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
12053 | } | |
12054 | } | |
7c673cae FG |
12055 | |
12056 | if (force_flush) { | |
31f18b77 | 12057 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
12058 | << " force_flush=" << (int)force_flush |
12059 | << ", flushing, deferred done->stable" << dendl; | |
12060 | // flush/barrier on block device | |
12061 | bdev->flush(); | |
12062 | ||
12063 | // if we flush then deferred done are now deferred stable | |
12064 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
12065 | deferred_done.end()); | |
12066 | deferred_done.clear(); | |
12067 | } | |
11fdf7f2 | 12068 | auto after_flush = mono_clock::now(); |
7c673cae FG |
12069 | |
12070 | // we will use one final transaction to force a sync | |
12071 | KeyValueDB::Transaction synct = db->get_transaction(); | |
12072 | ||
12073 | // increase {nid,blobid}_max? note that this covers both the | |
12074 | // case where we are approaching the max and the case we passed | |
12075 | // it. in either case, we increase the max in the earlier txn | |
12076 | // we submit. | |
12077 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
12078 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
12079 | KeyValueDB::Transaction t = | |
12080 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
12081 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
12082 | bufferlist bl; | |
11fdf7f2 | 12083 | encode(new_nid_max, bl); |
7c673cae FG |
12084 | t->set(PREFIX_SUPER, "nid_max", bl); |
12085 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
12086 | } | |
12087 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
12088 | KeyValueDB::Transaction t = | |
12089 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
12090 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
12091 | bufferlist bl; | |
11fdf7f2 | 12092 | encode(new_blobid_max, bl); |
7c673cae FG |
12093 | t->set(PREFIX_SUPER, "blobid_max", bl); |
12094 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
12095 | } | |
c07f9fc5 FG |
12096 | |
12097 | for (auto txc : kv_committing) { | |
9f95a23c | 12098 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_queued_lat); |
f67539c2 | 12099 | if (txc->get_state() == TransContext::STATE_KV_QUEUED) { |
adb31ebb | 12100 | ++kv_submitted; |
9f95a23c | 12101 | _txc_apply_kv(txc, false); |
c07f9fc5 | 12102 | --txc->osr->kv_committing_serially; |
c07f9fc5 | 12103 | } else { |
f67539c2 | 12104 | ceph_assert(txc->get_state() == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12105 | } |
7c673cae FG |
12106 | if (txc->had_ios) { |
12107 | --txc->osr->txc_with_unstable_io; | |
12108 | } | |
7c673cae FG |
12109 | } |
12110 | ||
31f18b77 FG |
12111 | // release throttle *before* we commit. this allows new ops |
12112 | // to be prepared and enter pipeline while we are waiting on | |
12113 | // the kv commit sync/flush. then hopefully on the next | |
12114 | // iteration there will already be ops awake. otherwise, we | |
12115 | // end up going to sleep, and then wake up when the very first | |
12116 | // transaction is ready for commit. | |
9f95a23c | 12117 | throttle.release_kv_throttle(costs); |
31f18b77 | 12118 | |
7c673cae FG |
12119 | // cleanup sync deferred keys |
12120 | for (auto b : deferred_stable) { | |
12121 | for (auto& txc : b->txcs) { | |
12122 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 12123 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
12124 | string key; |
12125 | get_deferred_key(wt.seq, &key); | |
12126 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
12127 | } | |
12128 | } | |
12129 | ||
9f95a23c TL |
12130 | #if defined(WITH_LTTNG) |
12131 | auto sync_start = mono_clock::now(); | |
12132 | #endif | |
7c673cae | 12133 | // submit synct synchronously (block and wait for it to commit) |
31f18b77 | 12134 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
12135 | ceph_assert(r == 0); |
12136 | ||
f67539c2 TL |
12137 | #ifdef WITH_BLKIN |
12138 | for (auto txc : kv_committing) { | |
12139 | if (txc->trace) { | |
12140 | txc->trace.event("db sync submit"); | |
12141 | txc->trace.keyval("kv_committing size", kv_committing.size()); | |
12142 | } | |
12143 | } | |
12144 | #endif | |
12145 | ||
9f95a23c TL |
12146 | int committing_size = kv_committing.size(); |
12147 | int deferred_size = deferred_stable.size(); | |
12148 | ||
12149 | #if defined(WITH_LTTNG) | |
12150 | double sync_latency = ceph::to_seconds<double>(mono_clock::now() - sync_start); | |
12151 | for (auto txc: kv_committing) { | |
12152 | if (txc->tracing) { | |
12153 | tracepoint( | |
12154 | bluestore, | |
12155 | transaction_kv_sync_latency, | |
12156 | txc->osr->get_sequencer_id(), | |
12157 | txc->seq, | |
12158 | kv_committing.size(), | |
12159 | deferred_done.size(), | |
12160 | deferred_stable.size(), | |
12161 | sync_latency); | |
12162 | } | |
12163 | } | |
12164 | #endif | |
12165 | ||
11fdf7f2 | 12166 | { |
9f95a23c | 12167 | std::unique_lock m{kv_finalize_lock}; |
11fdf7f2 TL |
12168 | if (kv_committing_to_finalize.empty()) { |
12169 | kv_committing_to_finalize.swap(kv_committing); | |
12170 | } else { | |
12171 | kv_committing_to_finalize.insert( | |
12172 | kv_committing_to_finalize.end(), | |
12173 | kv_committing.begin(), | |
12174 | kv_committing.end()); | |
12175 | kv_committing.clear(); | |
12176 | } | |
12177 | if (deferred_stable_to_finalize.empty()) { | |
12178 | deferred_stable_to_finalize.swap(deferred_stable); | |
12179 | } else { | |
12180 | deferred_stable_to_finalize.insert( | |
12181 | deferred_stable_to_finalize.end(), | |
12182 | deferred_stable.begin(), | |
12183 | deferred_stable.end()); | |
12184 | deferred_stable.clear(); | |
12185 | } | |
9f95a23c TL |
12186 | if (!kv_finalize_in_progress) { |
12187 | kv_finalize_in_progress = true; | |
12188 | kv_finalize_cond.notify_one(); | |
12189 | } | |
11fdf7f2 | 12190 | } |
7c673cae FG |
12191 | |
12192 | if (new_nid_max) { | |
12193 | nid_max = new_nid_max; | |
12194 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
12195 | } | |
12196 | if (new_blobid_max) { | |
12197 | blobid_max = new_blobid_max; | |
12198 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
12199 | } | |
12200 | ||
224ce89b | 12201 | { |
11fdf7f2 TL |
12202 | auto finish = mono_clock::now(); |
12203 | ceph::timespan dur_flush = after_flush - start; | |
12204 | ceph::timespan dur_kv = finish - after_flush; | |
12205 | ceph::timespan dur = finish - start; | |
9f95a23c TL |
12206 | dout(20) << __func__ << " committed " << committing_size |
12207 | << " cleaned " << deferred_size | |
224ce89b WB |
12208 | << " in " << dur |
12209 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
12210 | << dendl; | |
494da23a TL |
12211 | log_latency("kv_flush", |
12212 | l_bluestore_kv_flush_lat, | |
12213 | dur_flush, | |
12214 | cct->_conf->bluestore_log_op_age); | |
12215 | log_latency("kv_commit", | |
12216 | l_bluestore_kv_commit_lat, | |
12217 | dur_kv, | |
12218 | cct->_conf->bluestore_log_op_age); | |
12219 | log_latency("kv_sync", | |
12220 | l_bluestore_kv_sync_lat, | |
12221 | dur, | |
12222 | cct->_conf->bluestore_log_op_age); | |
7c673cae | 12223 | } |
31f18b77 | 12224 | |
31f18b77 FG |
12225 | l.lock(); |
12226 | // previously deferred "done" are now "stable" by virtue of this | |
12227 | // commit cycle. | |
12228 | deferred_stable_queue.swap(deferred_done); | |
12229 | } | |
12230 | } | |
12231 | dout(10) << __func__ << " finish" << dendl; | |
12232 | kv_sync_started = false; | |
12233 | } | |
12234 | ||
12235 | void BlueStore::_kv_finalize_thread() | |
12236 | { | |
12237 | deque<TransContext*> kv_committed; | |
12238 | deque<DeferredBatch*> deferred_stable; | |
12239 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
12240 | std::unique_lock l(kv_finalize_lock); |
12241 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
12242 | kv_finalize_started = true; |
12243 | kv_finalize_cond.notify_all(); | |
12244 | while (true) { | |
11fdf7f2 TL |
12245 | ceph_assert(kv_committed.empty()); |
12246 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
12247 | if (kv_committing_to_finalize.empty() && |
12248 | deferred_stable_to_finalize.empty()) { | |
12249 | if (kv_finalize_stop) | |
12250 | break; | |
12251 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 12252 | kv_finalize_in_progress = false; |
31f18b77 FG |
12253 | kv_finalize_cond.wait(l); |
12254 | dout(20) << __func__ << " wake" << dendl; | |
12255 | } else { | |
12256 | kv_committed.swap(kv_committing_to_finalize); | |
12257 | deferred_stable.swap(deferred_stable_to_finalize); | |
12258 | l.unlock(); | |
12259 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
12260 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12261 | ||
11fdf7f2 TL |
12262 | auto start = mono_clock::now(); |
12263 | ||
31f18b77 FG |
12264 | while (!kv_committed.empty()) { |
12265 | TransContext *txc = kv_committed.front(); | |
f67539c2 | 12266 | ceph_assert(txc->get_state() == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12267 | _txc_state_proc(txc); |
31f18b77 | 12268 | kv_committed.pop_front(); |
7c673cae | 12269 | } |
31f18b77 | 12270 | |
7c673cae FG |
12271 | for (auto b : deferred_stable) { |
12272 | auto p = b->txcs.begin(); | |
12273 | while (p != b->txcs.end()) { | |
12274 | TransContext *txc = &*p; | |
12275 | p = b->txcs.erase(p); // unlink here because | |
12276 | _txc_state_proc(txc); // this may destroy txc | |
12277 | } | |
12278 | delete b; | |
12279 | } | |
31f18b77 | 12280 | deferred_stable.clear(); |
7c673cae FG |
12281 | |
12282 | if (!deferred_aggressive) { | |
31f18b77 | 12283 | if (deferred_queue_size >= deferred_batch_ops.load() || |
9f95a23c | 12284 | throttle.should_submit_deferred()) { |
224ce89b | 12285 | deferred_try_submit(); |
7c673cae FG |
12286 | } |
12287 | } | |
12288 | ||
12289 | // this is as good a place as any ... | |
12290 | _reap_collections(); | |
12291 | ||
11fdf7f2 | 12292 | logger->set(l_bluestore_fragmentation, |
f67539c2 | 12293 | (uint64_t)(shared_alloc.a->get_fragmentation() * 1000)); |
11fdf7f2 | 12294 | |
494da23a TL |
12295 | log_latency("kv_final", |
12296 | l_bluestore_kv_final_lat, | |
12297 | mono_clock::now() - start, | |
12298 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 | 12299 | |
7c673cae | 12300 | l.lock(); |
7c673cae FG |
12301 | } |
12302 | } | |
12303 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 12304 | kv_finalize_started = false; |
7c673cae FG |
12305 | } |
12306 | ||
f67539c2 TL |
12307 | void BlueStore::_zoned_cleaner_start() { |
12308 | dout(10) << __func__ << dendl; | |
12309 | ||
12310 | zoned_cleaner_thread.create("bstore_zcleaner"); | |
12311 | } | |
12312 | ||
12313 | void BlueStore::_zoned_cleaner_stop() { | |
12314 | dout(10) << __func__ << dendl; | |
12315 | { | |
12316 | std::unique_lock l{zoned_cleaner_lock}; | |
12317 | while (!zoned_cleaner_started) { | |
12318 | zoned_cleaner_cond.wait(l); | |
12319 | } | |
12320 | zoned_cleaner_stop = true; | |
12321 | zoned_cleaner_cond.notify_all(); | |
12322 | } | |
12323 | zoned_cleaner_thread.join(); | |
12324 | { | |
12325 | std::lock_guard l{zoned_cleaner_lock}; | |
12326 | zoned_cleaner_stop = false; | |
12327 | } | |
12328 | dout(10) << __func__ << " done" << dendl; | |
12329 | } | |
12330 | ||
12331 | void BlueStore::_zoned_cleaner_thread() { | |
12332 | dout(10) << __func__ << " start" << dendl; | |
12333 | std::unique_lock l{zoned_cleaner_lock}; | |
12334 | ceph_assert(!zoned_cleaner_started); | |
12335 | zoned_cleaner_started = true; | |
12336 | zoned_cleaner_cond.notify_all(); | |
12337 | std::deque<uint64_t> zones_to_clean; | |
12338 | while (true) { | |
12339 | if (zoned_cleaner_queue.empty()) { | |
12340 | if (zoned_cleaner_stop) { | |
12341 | break; | |
12342 | } | |
12343 | dout(20) << __func__ << " sleep" << dendl; | |
12344 | zoned_cleaner_cond.wait(l); | |
12345 | dout(20) << __func__ << " wake" << dendl; | |
12346 | } else { | |
12347 | zones_to_clean.swap(zoned_cleaner_queue); | |
12348 | l.unlock(); | |
12349 | while (!zones_to_clean.empty()) { | |
12350 | _zoned_clean_zone(zones_to_clean.front()); | |
12351 | zones_to_clean.pop_front(); | |
12352 | } | |
12353 | l.lock(); | |
12354 | } | |
12355 | } | |
12356 | dout(10) << __func__ << " finish" << dendl; | |
12357 | zoned_cleaner_started = false; | |
12358 | } | |
12359 | ||
12360 | void BlueStore::_zoned_clean_zone(uint64_t zone_num) { | |
12361 | dout(10) << __func__ << " cleaning zone " << zone_num << dendl; | |
12362 | } | |
12363 | ||
7c673cae | 12364 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( |
9f95a23c | 12365 | TransContext *txc) |
7c673cae FG |
12366 | { |
12367 | if (!txc->deferred_txn) { | |
12368 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
12369 | } | |
12370 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
12371 | return &txc->deferred_txn->ops.back(); | |
12372 | } | |
12373 | ||
12374 | void BlueStore::_deferred_queue(TransContext *txc) | |
12375 | { | |
12376 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
f67539c2 TL |
12377 | |
12378 | DeferredBatch *tmp; | |
12379 | txc->osr->deferred_lock.lock(); | |
12380 | { | |
12381 | if (!txc->osr->deferred_pending) { | |
12382 | tmp = new DeferredBatch(cct, txc->osr.get()); | |
12383 | } else { | |
12384 | tmp = txc->osr->deferred_pending; | |
12385 | } | |
7c673cae | 12386 | } |
f67539c2 TL |
12387 | |
12388 | tmp->txcs.push_back(*txc); | |
7c673cae FG |
12389 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; |
12390 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
12391 | const auto& op = *opi; | |
11fdf7f2 | 12392 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
12393 | bufferlist::const_iterator p = op.data.begin(); |
12394 | for (auto e : op.extents) { | |
f67539c2 | 12395 | tmp->prepare_write(cct, wt.seq, e.offset, e.length, p); |
7c673cae FG |
12396 | } |
12397 | } | |
f67539c2 TL |
12398 | |
12399 | { | |
12400 | ++deferred_queue_size; | |
12401 | txc->osr->deferred_pending = tmp; | |
12402 | // condition "tmp->txcs.size() == 1" mean deferred_pending was originally empty. | |
12403 | // So we should add osr into deferred_queue. | |
12404 | if (!txc->osr->deferred_running && (tmp->txcs.size() == 1)) { | |
12405 | deferred_lock.lock(); | |
12406 | deferred_queue.push_back(*txc->osr); | |
12407 | deferred_lock.unlock(); | |
12408 | } | |
12409 | ||
12410 | if (deferred_aggressive && | |
12411 | !txc->osr->deferred_running) { | |
12412 | _deferred_submit_unlock(txc->osr.get()); | |
12413 | } else { | |
12414 | txc->osr->deferred_lock.unlock(); | |
12415 | } | |
7c673cae | 12416 | } |
f67539c2 | 12417 | } |
7c673cae | 12418 | |
224ce89b | 12419 | void BlueStore::deferred_try_submit() |
7c673cae FG |
12420 | { |
12421 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
12422 | << deferred_queue_size << " txcs" << dendl; | |
224ce89b | 12423 | vector<OpSequencerRef> osrs; |
f67539c2 TL |
12424 | |
12425 | { | |
12426 | std::lock_guard l(deferred_lock); | |
12427 | osrs.reserve(deferred_queue.size()); | |
12428 | for (auto& osr : deferred_queue) { | |
12429 | osrs.push_back(&osr); | |
12430 | } | |
224ce89b | 12431 | } |
f67539c2 | 12432 | |
224ce89b | 12433 | for (auto& osr : osrs) { |
f67539c2 | 12434 | osr->deferred_lock.lock(); |
181888fb FG |
12435 | if (osr->deferred_pending) { |
12436 | if (!osr->deferred_running) { | |
12437 | _deferred_submit_unlock(osr.get()); | |
181888fb | 12438 | } else { |
f67539c2 | 12439 | osr->deferred_lock.unlock(); |
181888fb FG |
12440 | dout(20) << __func__ << " osr " << osr << " already has running" |
12441 | << dendl; | |
12442 | } | |
12443 | } else { | |
f67539c2 | 12444 | osr->deferred_lock.unlock(); |
181888fb | 12445 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; |
7c673cae FG |
12446 | } |
12447 | } | |
9f95a23c | 12448 | |
f67539c2 TL |
12449 | { |
12450 | std::lock_guard l(deferred_lock); | |
12451 | deferred_last_submitted = ceph_clock_now(); | |
12452 | } | |
7c673cae FG |
12453 | } |
12454 | ||
224ce89b | 12455 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
12456 | { |
12457 | dout(10) << __func__ << " osr " << osr | |
12458 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
12459 | << dendl; | |
11fdf7f2 TL |
12460 | ceph_assert(osr->deferred_pending); |
12461 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
12462 | |
12463 | auto b = osr->deferred_pending; | |
12464 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 12465 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
12466 | |
12467 | osr->deferred_running = osr->deferred_pending; | |
12468 | osr->deferred_pending = nullptr; | |
12469 | ||
f67539c2 | 12470 | osr->deferred_lock.unlock(); |
11fdf7f2 TL |
12471 | |
12472 | for (auto& txc : b->txcs) { | |
9f95a23c | 12473 | throttle.log_state_latency(txc, logger, l_bluestore_state_deferred_queued_lat); |
11fdf7f2 | 12474 | } |
7c673cae FG |
12475 | uint64_t start = 0, pos = 0; |
12476 | bufferlist bl; | |
12477 | auto i = b->iomap.begin(); | |
12478 | while (true) { | |
12479 | if (i == b->iomap.end() || i->first != pos) { | |
12480 | if (bl.length()) { | |
12481 | dout(20) << __func__ << " write 0x" << std::hex | |
12482 | << start << "~" << bl.length() | |
12483 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 12484 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
12485 | logger->inc(l_bluestore_deferred_write_ops); |
12486 | logger->inc(l_bluestore_deferred_write_bytes, bl.length()); | |
12487 | int r = bdev->aio_write(start, bl, &b->ioc, false); | |
11fdf7f2 | 12488 | ceph_assert(r == 0); |
7c673cae FG |
12489 | } |
12490 | } | |
12491 | if (i == b->iomap.end()) { | |
12492 | break; | |
12493 | } | |
12494 | start = 0; | |
12495 | pos = i->first; | |
12496 | bl.clear(); | |
12497 | } | |
12498 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
12499 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
12500 | << dendl; | |
12501 | if (!bl.length()) { | |
12502 | start = pos; | |
12503 | } | |
12504 | pos += i->second.bl.length(); | |
12505 | bl.claim_append(i->second.bl); | |
12506 | ++i; | |
12507 | } | |
224ce89b | 12508 | |
7c673cae FG |
12509 | bdev->aio_submit(&b->ioc); |
12510 | } | |
12511 | ||
3efd9988 FG |
12512 | struct C_DeferredTrySubmit : public Context { |
12513 | BlueStore *store; | |
12514 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
12515 | void finish(int r) { | |
12516 | store->deferred_try_submit(); | |
12517 | } | |
12518 | }; | |
12519 | ||
7c673cae FG |
12520 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
12521 | { | |
12522 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 12523 | ceph_assert(osr->deferred_running); |
7c673cae FG |
12524 | DeferredBatch *b = osr->deferred_running; |
12525 | ||
12526 | { | |
f67539c2 | 12527 | osr->deferred_lock.lock(); |
11fdf7f2 | 12528 | ceph_assert(osr->deferred_running == b); |
7c673cae FG |
12529 | osr->deferred_running = nullptr; |
12530 | if (!osr->deferred_pending) { | |
181888fb | 12531 | dout(20) << __func__ << " dequeueing" << dendl; |
f67539c2 TL |
12532 | { |
12533 | deferred_lock.lock(); | |
12534 | auto q = deferred_queue.iterator_to(*osr); | |
12535 | deferred_queue.erase(q); | |
12536 | deferred_lock.unlock(); | |
12537 | } | |
12538 | osr->deferred_lock.unlock(); | |
181888fb | 12539 | } else { |
f67539c2 | 12540 | osr->deferred_lock.unlock(); |
9f95a23c TL |
12541 | if (deferred_aggressive) { |
12542 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; | |
12543 | finisher.queue(new C_DeferredTrySubmit(this)); | |
12544 | } else { | |
12545 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
12546 | } | |
7c673cae FG |
12547 | } |
12548 | } | |
12549 | ||
12550 | { | |
31f18b77 | 12551 | uint64_t costs = 0; |
11fdf7f2 | 12552 | { |
11fdf7f2 TL |
12553 | for (auto& i : b->txcs) { |
12554 | TransContext *txc = &i; | |
9f95a23c | 12555 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_aio_wait_lat); |
f67539c2 | 12556 | txc->set_state(TransContext::STATE_DEFERRED_CLEANUP); |
11fdf7f2 TL |
12557 | costs += txc->cost; |
12558 | } | |
7c673cae | 12559 | } |
9f95a23c | 12560 | throttle.release_deferred_throttle(costs); |
7c673cae FG |
12561 | } |
12562 | ||
9f95a23c | 12563 | { |
11fdf7f2 | 12564 | std::lock_guard l(kv_lock); |
9f95a23c TL |
12565 | deferred_done_queue.emplace_back(b); |
12566 | ||
12567 | // in the normal case, do not bother waking up the kv thread; it will | |
12568 | // catch us on the next commit anyway. | |
12569 | if (deferred_aggressive && !kv_sync_in_progress) { | |
12570 | kv_sync_in_progress = true; | |
12571 | kv_cond.notify_one(); | |
12572 | } | |
7c673cae FG |
12573 | } |
12574 | } | |
12575 | ||
12576 | int BlueStore::_deferred_replay() | |
12577 | { | |
12578 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
12579 | int count = 0; |
12580 | int r = 0; | |
11fdf7f2 TL |
12581 | CollectionRef ch = _get_collection(coll_t::meta()); |
12582 | bool fake_ch = false; | |
12583 | if (!ch) { | |
12584 | // hmm, replaying initial mkfs? | |
12585 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
12586 | fake_ch = true; | |
12587 | } | |
12588 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
12589 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
12590 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
12591 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
12592 | << dendl; | |
12593 | bluestore_deferred_transaction_t *deferred_txn = | |
12594 | new bluestore_deferred_transaction_t; | |
12595 | bufferlist bl = it->value(); | |
11fdf7f2 | 12596 | auto p = bl.cbegin(); |
7c673cae | 12597 | try { |
11fdf7f2 | 12598 | decode(*deferred_txn, p); |
f67539c2 | 12599 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12600 | derr << __func__ << " failed to decode deferred txn " |
12601 | << pretty_binary_string(it->key()) << dendl; | |
12602 | delete deferred_txn; | |
12603 | r = -EIO; | |
12604 | goto out; | |
12605 | } | |
11fdf7f2 | 12606 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); |
7c673cae | 12607 | txc->deferred_txn = deferred_txn; |
f67539c2 | 12608 | txc->set_state(TransContext::STATE_KV_DONE); |
7c673cae FG |
12609 | _txc_state_proc(txc); |
12610 | } | |
12611 | out: | |
12612 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 12613 | _osr_register_zombie(osr); |
7c673cae | 12614 | _osr_drain_all(); |
11fdf7f2 TL |
12615 | if (fake_ch) { |
12616 | new_coll_map.clear(); | |
12617 | } | |
7c673cae FG |
12618 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
12619 | return r; | |
12620 | } | |
12621 | ||
12622 | // --------------------------- | |
12623 | // transactions | |
12624 | ||
12625 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
12626 | CollectionHandle& ch, |
12627 | vector<Transaction>& tls, | |
12628 | TrackedOpRef op, | |
12629 | ThreadPool::TPHandle *handle) | |
12630 | { | |
12631 | FUNCTRACE(cct); | |
12632 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 12633 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 12634 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae | 12635 | |
11fdf7f2 TL |
12636 | auto start = mono_clock::now(); |
12637 | ||
12638 | Collection *c = static_cast<Collection*>(ch.get()); | |
12639 | OpSequencer *osr = c->osr.get(); | |
12640 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae FG |
12641 | |
12642 | // prepare | |
11fdf7f2 | 12643 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, |
f67539c2 | 12644 | &on_commit, op); |
7c673cae | 12645 | |
f67539c2 TL |
12646 | // With HM-SMR drives (and ZNS SSDs) we want the I/O allocation and I/O |
12647 | // submission to happen atomically because if I/O submission happens in a | |
12648 | // different order than I/O allocation, we end up issuing non-sequential | |
12649 | // writes to the drive. This is a temporary solution until ZONE APPEND | |
12650 | // support matures in the kernel. For more information please see: | |
12651 | // https://www.usenix.org/conference/vault20/presentation/bjorling | |
12652 | if (bdev->is_smr()) { | |
12653 | atomic_alloc_and_submit_lock.lock(); | |
12654 | } | |
7c673cae | 12655 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { |
7c673cae FG |
12656 | txc->bytes += (*p).get_num_bytes(); |
12657 | _txc_add_transaction(txc, &(*p)); | |
12658 | } | |
12659 | _txc_calc_cost(txc); | |
12660 | ||
12661 | _txc_write_nodes(txc, txc->t); | |
12662 | ||
12663 | // journal deferred items | |
12664 | if (txc->deferred_txn) { | |
12665 | txc->deferred_txn->seq = ++deferred_seq; | |
12666 | bufferlist bl; | |
11fdf7f2 | 12667 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
12668 | string key; |
12669 | get_deferred_key(txc->deferred_txn->seq, &key); | |
12670 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
12671 | } | |
12672 | ||
12673 | _txc_finalize_kv(txc, txc->t); | |
f67539c2 TL |
12674 | |
12675 | #ifdef WITH_BLKIN | |
12676 | if (txc->trace) { | |
12677 | txc->trace.event("txc encode finished"); | |
12678 | } | |
12679 | #endif | |
12680 | ||
7c673cae FG |
12681 | if (handle) |
12682 | handle->suspend_tp_timeout(); | |
12683 | ||
11fdf7f2 | 12684 | auto tstart = mono_clock::now(); |
9f95a23c TL |
12685 | |
12686 | if (!throttle.try_start_transaction( | |
12687 | *db, | |
12688 | *txc, | |
12689 | tstart)) { | |
7c673cae | 12690 | // ensure we do not block here because of deferred writes |
9f95a23c TL |
12691 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
12692 | << dendl; | |
12693 | ++deferred_aggressive; | |
12694 | deferred_try_submit(); | |
12695 | { | |
12696 | // wake up any previously finished deferred events | |
12697 | std::lock_guard l(kv_lock); | |
12698 | if (!kv_sync_in_progress) { | |
12699 | kv_sync_in_progress = true; | |
3efd9988 FG |
12700 | kv_cond.notify_one(); |
12701 | } | |
9f95a23c TL |
12702 | } |
12703 | throttle.finish_start_transaction(*db, *txc, tstart); | |
12704 | --deferred_aggressive; | |
7c673cae | 12705 | } |
11fdf7f2 | 12706 | auto tend = mono_clock::now(); |
7c673cae FG |
12707 | |
12708 | if (handle) | |
12709 | handle->reset_tp_timeout(); | |
12710 | ||
12711 | logger->inc(l_bluestore_txc); | |
12712 | ||
12713 | // execute (start) | |
12714 | _txc_state_proc(txc); | |
12715 | ||
f67539c2 TL |
12716 | if (bdev->is_smr()) { |
12717 | atomic_alloc_and_submit_lock.unlock(); | |
12718 | } | |
12719 | ||
11fdf7f2 TL |
12720 | // we're immediately readable (unlike FileStore) |
12721 | for (auto c : on_applied_sync) { | |
12722 | c->complete(0); | |
12723 | } | |
12724 | if (!on_applied.empty()) { | |
12725 | if (c->commit_queue) { | |
12726 | c->commit_queue->queue(on_applied); | |
12727 | } else { | |
12728 | finisher.queue(on_applied); | |
12729 | } | |
12730 | } | |
12731 | ||
f67539c2 TL |
12732 | #ifdef WITH_BLKIN |
12733 | if (txc->trace) { | |
12734 | txc->trace.event("txc applied"); | |
12735 | } | |
12736 | #endif | |
12737 | ||
494da23a TL |
12738 | log_latency("submit_transact", |
12739 | l_bluestore_submit_lat, | |
12740 | mono_clock::now() - start, | |
12741 | cct->_conf->bluestore_log_op_age); | |
12742 | log_latency("throttle_transact", | |
12743 | l_bluestore_throttle_lat, | |
12744 | tend - tstart, | |
12745 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
12746 | return 0; |
12747 | } | |
12748 | ||
12749 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
12750 | { | |
12751 | dout(10) << __func__ << " txc " << txc << dendl; | |
12752 | bdev->aio_submit(&txc->ioc); | |
12753 | } | |
12754 | ||
12755 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
12756 | { | |
12757 | Transaction::iterator i = t->begin(); | |
12758 | ||
81eedcae | 12759 | _dump_transaction<30>(cct, t); |
7c673cae FG |
12760 | |
12761 | vector<CollectionRef> cvec(i.colls.size()); | |
12762 | unsigned j = 0; | |
12763 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
12764 | ++p, ++j) { | |
12765 | cvec[j] = _get_collection(*p); | |
7c673cae | 12766 | } |
11fdf7f2 | 12767 | |
7c673cae FG |
12768 | vector<OnodeRef> ovec(i.objects.size()); |
12769 | ||
12770 | for (int pos = 0; i.have_op(); ++pos) { | |
12771 | Transaction::Op *op = i.decode_op(); | |
12772 | int r = 0; | |
12773 | ||
12774 | // no coll or obj | |
12775 | if (op->op == Transaction::OP_NOP) | |
12776 | continue; | |
12777 | ||
11fdf7f2 | 12778 | |
7c673cae FG |
12779 | // collection operations |
12780 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
12781 | |
12782 | // initialize osd_pool_id and do a smoke test that all collections belong | |
12783 | // to the same pool | |
12784 | spg_t pgid; | |
12785 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
12786 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
12787 | txc->osd_pool_id == pgid.pool()); | |
12788 | txc->osd_pool_id = pgid.pool(); | |
12789 | } | |
12790 | ||
7c673cae FG |
12791 | switch (op->op) { |
12792 | case Transaction::OP_RMCOLL: | |
12793 | { | |
12794 | const coll_t &cid = i.get_cid(op->cid); | |
12795 | r = _remove_collection(txc, cid, &c); | |
12796 | if (!r) | |
12797 | continue; | |
12798 | } | |
12799 | break; | |
12800 | ||
12801 | case Transaction::OP_MKCOLL: | |
12802 | { | |
11fdf7f2 | 12803 | ceph_assert(!c); |
7c673cae FG |
12804 | const coll_t &cid = i.get_cid(op->cid); |
12805 | r = _create_collection(txc, cid, op->split_bits, &c); | |
12806 | if (!r) | |
12807 | continue; | |
12808 | } | |
12809 | break; | |
12810 | ||
12811 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 12812 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12813 | break; |
12814 | ||
12815 | case Transaction::OP_SPLIT_COLLECTION2: | |
12816 | { | |
12817 | uint32_t bits = op->split_bits; | |
12818 | uint32_t rem = op->split_rem; | |
12819 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
12820 | if (!r) | |
12821 | continue; | |
12822 | } | |
12823 | break; | |
12824 | ||
11fdf7f2 TL |
12825 | case Transaction::OP_MERGE_COLLECTION: |
12826 | { | |
12827 | uint32_t bits = op->split_bits; | |
12828 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
12829 | if (!r) | |
12830 | continue; | |
12831 | } | |
12832 | break; | |
12833 | ||
7c673cae FG |
12834 | case Transaction::OP_COLL_HINT: |
12835 | { | |
f67539c2 | 12836 | uint32_t type = op->hint; |
7c673cae FG |
12837 | bufferlist hint; |
12838 | i.decode_bl(hint); | |
11fdf7f2 | 12839 | auto hiter = hint.cbegin(); |
7c673cae FG |
12840 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
12841 | uint32_t pg_num; | |
12842 | uint64_t num_objs; | |
11fdf7f2 TL |
12843 | decode(pg_num, hiter); |
12844 | decode(num_objs, hiter); | |
7c673cae FG |
12845 | dout(10) << __func__ << " collection hint objects is a no-op, " |
12846 | << " pg_num " << pg_num << " num_objects " << num_objs | |
12847 | << dendl; | |
12848 | } else { | |
12849 | // Ignore the hint | |
12850 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
12851 | } | |
12852 | continue; | |
12853 | } | |
12854 | break; | |
12855 | ||
12856 | case Transaction::OP_COLL_SETATTR: | |
12857 | r = -EOPNOTSUPP; | |
12858 | break; | |
12859 | ||
12860 | case Transaction::OP_COLL_RMATTR: | |
12861 | r = -EOPNOTSUPP; | |
12862 | break; | |
12863 | ||
12864 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 12865 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12866 | break; |
12867 | } | |
12868 | if (r < 0) { | |
12869 | derr << __func__ << " error " << cpp_strerror(r) | |
12870 | << " not handled on operation " << op->op | |
12871 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 12872 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12873 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12874 | } |
12875 | ||
12876 | // these operations implicity create the object | |
12877 | bool create = false; | |
12878 | if (op->op == Transaction::OP_TOUCH || | |
9f95a23c | 12879 | op->op == Transaction::OP_CREATE || |
7c673cae FG |
12880 | op->op == Transaction::OP_WRITE || |
12881 | op->op == Transaction::OP_ZERO) { | |
12882 | create = true; | |
12883 | } | |
12884 | ||
12885 | // object operations | |
9f95a23c | 12886 | std::unique_lock l(c->lock); |
7c673cae FG |
12887 | OnodeRef &o = ovec[op->oid]; |
12888 | if (!o) { | |
12889 | ghobject_t oid = i.get_oid(op->oid); | |
9f95a23c | 12890 | o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE); |
7c673cae FG |
12891 | } |
12892 | if (!create && (!o || !o->exists)) { | |
12893 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
12894 | << i.get_oid(op->oid) << dendl; | |
12895 | r = -ENOENT; | |
12896 | goto endop; | |
12897 | } | |
12898 | ||
12899 | switch (op->op) { | |
9f95a23c | 12900 | case Transaction::OP_CREATE: |
7c673cae FG |
12901 | case Transaction::OP_TOUCH: |
12902 | r = _touch(txc, c, o); | |
12903 | break; | |
12904 | ||
12905 | case Transaction::OP_WRITE: | |
12906 | { | |
12907 | uint64_t off = op->off; | |
12908 | uint64_t len = op->len; | |
12909 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
12910 | bufferlist bl; | |
12911 | i.decode_bl(bl); | |
12912 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
12913 | } | |
12914 | break; | |
12915 | ||
12916 | case Transaction::OP_ZERO: | |
12917 | { | |
12918 | uint64_t off = op->off; | |
12919 | uint64_t len = op->len; | |
12920 | r = _zero(txc, c, o, off, len); | |
12921 | } | |
12922 | break; | |
12923 | ||
12924 | case Transaction::OP_TRIMCACHE: | |
12925 | { | |
12926 | // deprecated, no-op | |
12927 | } | |
12928 | break; | |
12929 | ||
12930 | case Transaction::OP_TRUNCATE: | |
12931 | { | |
12932 | uint64_t off = op->off; | |
35e4c445 | 12933 | r = _truncate(txc, c, o, off); |
7c673cae FG |
12934 | } |
12935 | break; | |
12936 | ||
12937 | case Transaction::OP_REMOVE: | |
12938 | { | |
12939 | r = _remove(txc, c, o); | |
12940 | } | |
12941 | break; | |
12942 | ||
12943 | case Transaction::OP_SETATTR: | |
12944 | { | |
12945 | string name = i.decode_string(); | |
12946 | bufferptr bp; | |
12947 | i.decode_bp(bp); | |
12948 | r = _setattr(txc, c, o, name, bp); | |
12949 | } | |
12950 | break; | |
12951 | ||
12952 | case Transaction::OP_SETATTRS: | |
12953 | { | |
12954 | map<string, bufferptr> aset; | |
12955 | i.decode_attrset(aset); | |
12956 | r = _setattrs(txc, c, o, aset); | |
12957 | } | |
12958 | break; | |
12959 | ||
12960 | case Transaction::OP_RMATTR: | |
12961 | { | |
12962 | string name = i.decode_string(); | |
12963 | r = _rmattr(txc, c, o, name); | |
12964 | } | |
12965 | break; | |
12966 | ||
12967 | case Transaction::OP_RMATTRS: | |
12968 | { | |
12969 | r = _rmattrs(txc, c, o); | |
12970 | } | |
12971 | break; | |
12972 | ||
12973 | case Transaction::OP_CLONE: | |
12974 | { | |
12975 | OnodeRef& no = ovec[op->dest_oid]; | |
12976 | if (!no) { | |
12977 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12978 | no = c->get_onode(noid, true); | |
12979 | } | |
12980 | r = _clone(txc, c, o, no); | |
12981 | } | |
12982 | break; | |
12983 | ||
12984 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 12985 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12986 | break; |
12987 | ||
12988 | case Transaction::OP_CLONERANGE2: | |
12989 | { | |
12990 | OnodeRef& no = ovec[op->dest_oid]; | |
12991 | if (!no) { | |
12992 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12993 | no = c->get_onode(noid, true); | |
12994 | } | |
12995 | uint64_t srcoff = op->off; | |
12996 | uint64_t len = op->len; | |
12997 | uint64_t dstoff = op->dest_off; | |
12998 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
12999 | } | |
13000 | break; | |
13001 | ||
13002 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 13003 | ceph_abort_msg("not implemented"); |
7c673cae FG |
13004 | break; |
13005 | ||
13006 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 13007 | ceph_abort_msg("not implemented"); |
7c673cae FG |
13008 | break; |
13009 | ||
13010 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 13011 | ceph_abort_msg("deprecated"); |
7c673cae FG |
13012 | break; |
13013 | ||
13014 | case Transaction::OP_COLL_MOVE_RENAME: | |
13015 | case Transaction::OP_TRY_RENAME: | |
13016 | { | |
11fdf7f2 | 13017 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
13018 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
13019 | OnodeRef& no = ovec[op->dest_oid]; | |
13020 | if (!no) { | |
13021 | no = c->get_onode(noid, false); | |
13022 | } | |
13023 | r = _rename(txc, c, o, no, noid); | |
13024 | } | |
13025 | break; | |
13026 | ||
13027 | case Transaction::OP_OMAP_CLEAR: | |
13028 | { | |
13029 | r = _omap_clear(txc, c, o); | |
13030 | } | |
13031 | break; | |
13032 | case Transaction::OP_OMAP_SETKEYS: | |
13033 | { | |
13034 | bufferlist aset_bl; | |
13035 | i.decode_attrset_bl(&aset_bl); | |
13036 | r = _omap_setkeys(txc, c, o, aset_bl); | |
13037 | } | |
13038 | break; | |
13039 | case Transaction::OP_OMAP_RMKEYS: | |
13040 | { | |
13041 | bufferlist keys_bl; | |
13042 | i.decode_keyset_bl(&keys_bl); | |
13043 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
13044 | } | |
13045 | break; | |
13046 | case Transaction::OP_OMAP_RMKEYRANGE: | |
13047 | { | |
13048 | string first, last; | |
13049 | first = i.decode_string(); | |
13050 | last = i.decode_string(); | |
13051 | r = _omap_rmkey_range(txc, c, o, first, last); | |
13052 | } | |
13053 | break; | |
13054 | case Transaction::OP_OMAP_SETHEADER: | |
13055 | { | |
13056 | bufferlist bl; | |
13057 | i.decode_bl(bl); | |
13058 | r = _omap_setheader(txc, c, o, bl); | |
13059 | } | |
13060 | break; | |
13061 | ||
13062 | case Transaction::OP_SETALLOCHINT: | |
13063 | { | |
13064 | r = _set_alloc_hint(txc, c, o, | |
13065 | op->expected_object_size, | |
13066 | op->expected_write_size, | |
f67539c2 | 13067 | op->hint); |
7c673cae FG |
13068 | } |
13069 | break; | |
13070 | ||
13071 | default: | |
11fdf7f2 | 13072 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
13073 | ceph_abort(); |
13074 | } | |
13075 | ||
13076 | endop: | |
13077 | if (r < 0) { | |
13078 | bool ok = false; | |
13079 | ||
13080 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
13081 | op->op == Transaction::OP_CLONE || | |
13082 | op->op == Transaction::OP_CLONERANGE2 || | |
13083 | op->op == Transaction::OP_COLL_ADD || | |
13084 | op->op == Transaction::OP_SETATTR || | |
13085 | op->op == Transaction::OP_SETATTRS || | |
13086 | op->op == Transaction::OP_RMATTR || | |
13087 | op->op == Transaction::OP_OMAP_SETKEYS || | |
13088 | op->op == Transaction::OP_OMAP_RMKEYS || | |
13089 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
13090 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
13091 | // -ENOENT is usually okay | |
13092 | ok = true; | |
13093 | if (r == -ENODATA) | |
13094 | ok = true; | |
13095 | ||
13096 | if (!ok) { | |
13097 | const char *msg = "unexpected error code"; | |
13098 | ||
13099 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
13100 | op->op == Transaction::OP_CLONE || | |
13101 | op->op == Transaction::OP_CLONERANGE2)) | |
13102 | msg = "ENOENT on clone suggests osd bug"; | |
13103 | ||
13104 | if (r == -ENOSPC) | |
13105 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
13106 | // by partially applying transactions. | |
13107 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
13108 | ||
13109 | if (r == -ENOTEMPTY) { | |
13110 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
13111 | } | |
13112 | ||
13113 | derr << __func__ << " error " << cpp_strerror(r) | |
13114 | << " not handled on operation " << op->op | |
13115 | << " (op " << pos << ", counting from 0)" | |
13116 | << dendl; | |
13117 | derr << msg << dendl; | |
81eedcae | 13118 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 13119 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
13120 | } |
13121 | } | |
13122 | } | |
13123 | } | |
13124 | ||
13125 | ||
13126 | ||
13127 | // ----------------- | |
13128 | // write operations | |
13129 | ||
13130 | int BlueStore::_touch(TransContext *txc, | |
13131 | CollectionRef& c, | |
13132 | OnodeRef &o) | |
13133 | { | |
13134 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
13135 | int r = 0; | |
7c673cae FG |
13136 | _assign_nid(txc, o); |
13137 | txc->write_onode(o); | |
13138 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
13139 | return r; | |
13140 | } | |
13141 | ||
7c673cae FG |
13142 | void BlueStore::_pad_zeros( |
13143 | bufferlist *bl, uint64_t *offset, | |
13144 | uint64_t chunk_size) | |
13145 | { | |
13146 | auto length = bl->length(); | |
13147 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
13148 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
13149 | dout(40) << "before:\n"; | |
13150 | bl->hexdump(*_dout); | |
13151 | *_dout << dendl; | |
13152 | // front | |
13153 | size_t front_pad = *offset % chunk_size; | |
13154 | size_t back_pad = 0; | |
13155 | size_t pad_count = 0; | |
13156 | if (front_pad) { | |
11fdf7f2 | 13157 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
f67539c2 | 13158 | bufferptr z = ceph::buffer::create_small_page_aligned(chunk_size); |
224ce89b | 13159 | z.zero(0, front_pad, false); |
7c673cae | 13160 | pad_count += front_pad; |
9f95a23c | 13161 | bl->begin().copy(front_copy, z.c_str() + front_pad); |
7c673cae FG |
13162 | if (front_copy + front_pad < chunk_size) { |
13163 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 13164 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
13165 | pad_count += back_pad; |
13166 | } | |
13167 | bufferlist old, t; | |
13168 | old.swap(*bl); | |
13169 | t.substr_of(old, front_copy, length - front_copy); | |
13170 | bl->append(z); | |
13171 | bl->claim_append(t); | |
13172 | *offset -= front_pad; | |
224ce89b | 13173 | length += pad_count; |
7c673cae FG |
13174 | } |
13175 | ||
13176 | // back | |
13177 | uint64_t end = *offset + length; | |
13178 | unsigned back_copy = end % chunk_size; | |
13179 | if (back_copy) { | |
11fdf7f2 | 13180 | ceph_assert(back_pad == 0); |
7c673cae | 13181 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 13182 | ceph_assert(back_copy <= length); |
7c673cae | 13183 | bufferptr tail(chunk_size); |
9f95a23c | 13184 | bl->begin(length - back_copy).copy(back_copy, tail.c_str()); |
224ce89b | 13185 | tail.zero(back_copy, back_pad, false); |
7c673cae FG |
13186 | bufferlist old; |
13187 | old.swap(*bl); | |
13188 | bl->substr_of(old, 0, length - back_copy); | |
13189 | bl->append(tail); | |
13190 | length += back_pad; | |
13191 | pad_count += back_pad; | |
13192 | } | |
13193 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
13194 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
13195 | << length << std::dec << dendl; | |
13196 | dout(40) << "after:\n"; | |
13197 | bl->hexdump(*_dout); | |
13198 | *_dout << dendl; | |
13199 | if (pad_count) | |
13200 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 13201 | ceph_assert(bl->length() == length); |
7c673cae FG |
13202 | } |
13203 | ||
13204 | void BlueStore::_do_write_small( | |
13205 | TransContext *txc, | |
13206 | CollectionRef &c, | |
13207 | OnodeRef o, | |
13208 | uint64_t offset, uint64_t length, | |
13209 | bufferlist::iterator& blp, | |
13210 | WriteContext *wctx) | |
13211 | { | |
13212 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13213 | << std::dec << dendl; | |
11fdf7f2 | 13214 | ceph_assert(length < min_alloc_size); |
f67539c2 | 13215 | |
7c673cae FG |
13216 | uint64_t end_offs = offset + length; |
13217 | ||
13218 | logger->inc(l_bluestore_write_small); | |
13219 | logger->inc(l_bluestore_write_small_bytes, length); | |
13220 | ||
13221 | bufferlist bl; | |
13222 | blp.copy(length, bl); | |
13223 | ||
81eedcae TL |
13224 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
13225 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
13226 | uint32_t alloc_len = min_alloc_size; | |
13227 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
13228 | ||
13229 | bool any_change; | |
13230 | ||
13231 | // search suitable extent in both forward and reverse direction in | |
13232 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
13233 | // then check if blob can be reused via can_reuse_blob func or apply | |
13234 | // direct/deferred write (the latter for extents including or higher | |
13235 | // than 'offset' only). | |
13236 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
13237 | ||
f67539c2 TL |
13238 | // On zoned devices, the first goal is to support non-overwrite workloads, |
13239 | // such as RGW, with large, aligned objects. Therefore, for user writes | |
13240 | // _do_write_small should not trigger. OSDs, however, write and update a tiny | |
13241 | // amount of metadata, such as OSD maps, to disk. For those cases, we | |
13242 | // temporarily just pad them to min_alloc_size and write them to a new place | |
13243 | // on every update. | |
13244 | if (bdev->is_smr()) { | |
13245 | BlobRef b = c->new_blob(); | |
13246 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); | |
13247 | uint64_t b_off0 = b_off; | |
13248 | _pad_zeros(&bl, &b_off0, min_alloc_size); | |
13249 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13250 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, false, true); | |
13251 | return; | |
13252 | } | |
13253 | ||
7c673cae FG |
13254 | // Look for an existing mutable blob we can use. |
13255 | auto begin = o->extent_map.extent_map.begin(); | |
13256 | auto end = o->extent_map.extent_map.end(); | |
13257 | auto ep = o->extent_map.seek_lextent(offset); | |
13258 | if (ep != begin) { | |
13259 | --ep; | |
13260 | if (ep->blob_end() <= offset) { | |
13261 | ++ep; | |
13262 | } | |
13263 | } | |
f67539c2 TL |
13264 | auto prev_ep = end; |
13265 | if (ep != begin) { | |
13266 | prev_ep = ep; | |
7c673cae | 13267 | --prev_ep; |
7c673cae FG |
13268 | } |
13269 | ||
eafe8130 TL |
13270 | boost::container::flat_set<const bluestore_blob_t*> inspected_blobs; |
13271 | // We don't want to have more blobs than min alloc units fit | |
13272 | // into 2 max blobs | |
13273 | size_t blob_threshold = max_blob_size / min_alloc_size * 2 + 1; | |
13274 | bool above_blob_threshold = false; | |
13275 | ||
13276 | inspected_blobs.reserve(blob_threshold); | |
13277 | ||
13278 | uint64_t max_off = 0; | |
13279 | auto start_ep = ep; | |
13280 | auto end_ep = ep; // exclusively | |
7c673cae FG |
13281 | do { |
13282 | any_change = false; | |
13283 | ||
13284 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13285 | BlobRef b = ep->blob; | |
eafe8130 TL |
13286 | if (!above_blob_threshold) { |
13287 | inspected_blobs.insert(&b->get_blob()); | |
13288 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13289 | } | |
13290 | max_off = ep->logical_end(); | |
7c673cae | 13291 | auto bstart = ep->blob_start(); |
eafe8130 | 13292 | |
7c673cae FG |
13293 | dout(20) << __func__ << " considering " << *b |
13294 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
13295 | if (bstart >= end_offs) { | |
13296 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
13297 | } else if (!b->get_blob().is_mutable()) { | |
13298 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
13299 | } else if (ep->logical_offset % min_alloc_size != | |
13300 | ep->blob_offset % min_alloc_size) { | |
13301 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
13302 | } else { | |
13303 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13304 | // can we pad our head/tail out with zeros? | |
13305 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
13306 | head_pad = p2phase(offset, chunk_size); |
13307 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
13308 | if (head_pad || tail_pad) { |
13309 | o->extent_map.fault_range(db, offset - head_pad, | |
13310 | end_offs - offset + head_pad + tail_pad); | |
13311 | } | |
13312 | if (head_pad && | |
13313 | o->extent_map.has_any_lextents(offset - head_pad, chunk_size)) { | |
13314 | head_pad = 0; | |
13315 | } | |
13316 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
13317 | tail_pad = 0; | |
13318 | } | |
13319 | ||
13320 | uint64_t b_off = offset - head_pad - bstart; | |
13321 | uint64_t b_len = length + head_pad + tail_pad; | |
13322 | ||
13323 | // direct write into unused blocks of an existing mutable blob? | |
13324 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
13325 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13326 | b->get_blob().is_unused(b_off, b_len) && | |
13327 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 13328 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13329 | |
13330 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
13331 | << b_off << "~" << b_len | |
13332 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
13333 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 13334 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13335 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13336 | ||
11fdf7f2 | 13337 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13338 | if (b_len <= prefer_deferred_size) { |
13339 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13340 | << b_len << std::dec << " unused write via deferred" << dendl; | |
9f95a23c | 13341 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13342 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13343 | b->get_blob().map( | |
13344 | b_off, b_len, | |
13345 | [&](uint64_t offset, uint64_t length) { | |
13346 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13347 | return 0; | |
13348 | }); | |
224ce89b | 13349 | op->data = bl; |
7c673cae FG |
13350 | } else { |
13351 | b->get_blob().map_bl( | |
224ce89b | 13352 | b_off, bl, |
7c673cae FG |
13353 | [&](uint64_t offset, bufferlist& t) { |
13354 | bdev->aio_write(offset, t, | |
13355 | &txc->ioc, wctx->buffered); | |
13356 | }); | |
13357 | } | |
13358 | } | |
224ce89b | 13359 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
13360 | dout(20) << __func__ << " lex old " << *ep << dendl; |
13361 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
13362 | b, | |
13363 | &wctx->old_extents); | |
13364 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
f67539c2 | 13365 | |
7c673cae FG |
13366 | txc->statfs_delta.stored() += le->length; |
13367 | dout(20) << __func__ << " lex " << *le << dendl; | |
13368 | logger->inc(l_bluestore_write_small_unused); | |
13369 | return; | |
13370 | } | |
13371 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
13372 | uint64_t head_read = p2phase(b_off, chunk_size); |
13373 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
13374 | if ((head_read || tail_read) && |
13375 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
13376 | head_read + tail_read < min_alloc_size) { | |
13377 | b_off -= head_read; | |
13378 | b_len += head_read + tail_read; | |
13379 | ||
13380 | } else { | |
13381 | head_read = tail_read = 0; | |
13382 | } | |
13383 | ||
13384 | // chunk-aligned deferred overwrite? | |
13385 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13386 | b_off % chunk_size == 0 && | |
13387 | b_len % chunk_size == 0 && | |
13388 | b->get_blob().is_allocated(b_off, b_len)) { | |
13389 | ||
224ce89b | 13390 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13391 | |
13392 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
13393 | << " and tail 0x" << tail_read << std::dec << dendl; | |
13394 | if (head_read) { | |
13395 | bufferlist head_bl; | |
13396 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
13397 | head_bl, 0); | |
11fdf7f2 | 13398 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
13399 | size_t zlen = head_read - r; |
13400 | if (zlen) { | |
13401 | head_bl.append_zero(zlen); | |
13402 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13403 | } | |
11fdf7f2 TL |
13404 | head_bl.claim_append(bl); |
13405 | bl.swap(head_bl); | |
7c673cae FG |
13406 | logger->inc(l_bluestore_write_penalty_read_ops); |
13407 | } | |
13408 | if (tail_read) { | |
13409 | bufferlist tail_bl; | |
13410 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
13411 | tail_bl, 0); | |
11fdf7f2 | 13412 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
13413 | size_t zlen = tail_read - r; |
13414 | if (zlen) { | |
13415 | tail_bl.append_zero(zlen); | |
13416 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13417 | } | |
224ce89b | 13418 | bl.claim_append(tail_bl); |
7c673cae FG |
13419 | logger->inc(l_bluestore_write_penalty_read_ops); |
13420 | } | |
f67539c2 | 13421 | logger->inc(l_bluestore_write_small_pre_read); |
7c673cae | 13422 | |
224ce89b | 13423 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13424 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13425 | ||
f67539c2 | 13426 | b->dirty_blob().calc_csum(b_off, bl); |
11fdf7f2 TL |
13427 | |
13428 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
9f95a23c | 13429 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
11fdf7f2 TL |
13430 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13431 | int r = b->get_blob().map( | |
13432 | b_off, b_len, | |
13433 | [&](uint64_t offset, uint64_t length) { | |
13434 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13435 | return 0; | |
13436 | }); | |
13437 | ceph_assert(r == 0); | |
f67539c2 | 13438 | op->data = std::move(bl); |
11fdf7f2 TL |
13439 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" |
13440 | << b_len << std::dec << " of mutable " << *b | |
13441 | << " at " << op->extents << dendl; | |
13442 | } | |
13443 | ||
7c673cae FG |
13444 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
13445 | b, &wctx->old_extents); | |
13446 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13447 | txc->statfs_delta.stored() += le->length; | |
13448 | dout(20) << __func__ << " lex " << *le << dendl; | |
f67539c2 | 13449 | logger->inc(l_bluestore_write_deferred); |
7c673cae FG |
13450 | return; |
13451 | } | |
224ce89b WB |
13452 | // try to reuse blob if we can |
13453 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
13454 | max_bsize, |
13455 | offset0 - bstart, | |
13456 | &alloc_len)) { | |
11fdf7f2 | 13457 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13458 | // fit into reused blob |
13459 | // Need to check for pending writes desiring to | |
13460 | // reuse the same pextent. The rationale is that during GC two chunks | |
13461 | // from garbage blobs(compressed?) can share logical space within the same | |
13462 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13463 | // Hence the second write will fail in an attempt to reuse blob at | |
13464 | // do_alloc_write(). | |
13465 | if (!wctx->has_conflict(b, | |
13466 | offset0, | |
13467 | offset0 + alloc_len, | |
13468 | min_alloc_size)) { | |
13469 | ||
13470 | // we can't reuse pad_head/pad_tail since they might be truncated | |
13471 | // due to existent extents | |
13472 | uint64_t b_off = offset - bstart; | |
13473 | uint64_t b_off0 = b_off; | |
13474 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13475 | ||
13476 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13477 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13478 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13479 | << std::dec << dendl; |
13480 | ||
13481 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13482 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13483 | false, false); | |
13484 | logger->inc(l_bluestore_write_small_unused); | |
13485 | return; | |
13486 | } | |
13487 | } | |
13488 | } | |
13489 | ++ep; | |
eafe8130 | 13490 | end_ep = ep; |
7c673cae FG |
13491 | any_change = true; |
13492 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
13493 | ||
13494 | // check extent for reuse in reverse order | |
13495 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
13496 | BlobRef b = prev_ep->blob; | |
eafe8130 TL |
13497 | if (!above_blob_threshold) { |
13498 | inspected_blobs.insert(&b->get_blob()); | |
13499 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13500 | } | |
13501 | start_ep = prev_ep; | |
7c673cae FG |
13502 | auto bstart = prev_ep->blob_start(); |
13503 | dout(20) << __func__ << " considering " << *b | |
13504 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 13505 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
13506 | max_bsize, |
13507 | offset0 - bstart, | |
13508 | &alloc_len)) { | |
11fdf7f2 | 13509 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13510 | // fit into reused blob |
13511 | // Need to check for pending writes desiring to | |
13512 | // reuse the same pextent. The rationale is that during GC two chunks | |
13513 | // from garbage blobs(compressed?) can share logical space within the same | |
13514 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13515 | // Hence the second write will fail in an attempt to reuse blob at | |
13516 | // do_alloc_write(). | |
13517 | if (!wctx->has_conflict(b, | |
13518 | offset0, | |
13519 | offset0 + alloc_len, | |
13520 | min_alloc_size)) { | |
13521 | ||
13522 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13523 | uint64_t b_off = offset - bstart; | |
13524 | uint64_t b_off0 = b_off; | |
13525 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13526 | ||
13527 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13528 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13529 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13530 | << std::dec << dendl; |
13531 | ||
13532 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13533 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13534 | false, false); | |
13535 | logger->inc(l_bluestore_write_small_unused); | |
13536 | return; | |
13537 | } | |
13538 | } | |
13539 | if (prev_ep != begin) { | |
13540 | --prev_ep; | |
13541 | any_change = true; | |
13542 | } else { | |
13543 | prev_ep = end; // to avoid useless first extent re-check | |
13544 | } | |
13545 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
13546 | } while (any_change); | |
13547 | ||
eafe8130 TL |
13548 | if (above_blob_threshold) { |
13549 | dout(10) << __func__ << " request GC, blobs >= " << inspected_blobs.size() | |
13550 | << " " << std::hex << min_off << "~" << max_off << std::dec | |
13551 | << dendl; | |
13552 | ceph_assert(start_ep != end_ep); | |
13553 | for (auto ep = start_ep; ep != end_ep; ++ep) { | |
13554 | dout(20) << __func__ << " inserting for GC " | |
13555 | << std::hex << ep->logical_offset << "~" << ep->length | |
13556 | << std::dec << dendl; | |
13557 | ||
13558 | wctx->extents_to_gc.union_insert(ep->logical_offset, ep->length); | |
13559 | } | |
13560 | // insert newly written extent to GC | |
13561 | wctx->extents_to_gc.union_insert(offset, length); | |
13562 | dout(20) << __func__ << " inserting (last) for GC " | |
13563 | << std::hex << offset << "~" << length | |
13564 | << std::dec << dendl; | |
13565 | } | |
7c673cae | 13566 | // new blob. |
7c673cae | 13567 | BlobRef b = c->new_blob(); |
11fdf7f2 | 13568 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae FG |
13569 | uint64_t b_off0 = b_off; |
13570 | _pad_zeros(&bl, &b_off0, block_size); | |
13571 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
1911f103 TL |
13572 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, |
13573 | min_alloc_size != block_size, // use 'unused' bitmap when alloc granularity | |
13574 | // doesn't match disk one only | |
13575 | true); | |
7c673cae FG |
13576 | |
13577 | return; | |
13578 | } | |
13579 | ||
f67539c2 TL |
13580 | bool BlueStore::BigDeferredWriteContext::can_defer( |
13581 | BlueStore::extent_map_t::iterator ep, | |
13582 | uint64_t prefer_deferred_size, | |
13583 | uint64_t block_size, | |
13584 | uint64_t offset, | |
13585 | uint64_t l) | |
13586 | { | |
13587 | bool res = false; | |
13588 | auto& blob = ep->blob->get_blob(); | |
13589 | if (offset >= ep->blob_start() && | |
13590 | blob.is_mutable()) { | |
13591 | off = offset; | |
13592 | b_off = offset - ep->blob_start(); | |
13593 | uint64_t chunk_size = blob.get_chunk_size(block_size); | |
13594 | uint64_t ondisk = blob.get_ondisk_length(); | |
13595 | used = std::min(l, ondisk - b_off); | |
13596 | ||
13597 | // will read some data to fill out the chunk? | |
13598 | head_read = p2phase<uint64_t>(b_off, chunk_size); | |
13599 | tail_read = p2nphase<uint64_t>(b_off + used, chunk_size); | |
13600 | b_off -= head_read; | |
13601 | ||
13602 | ceph_assert(b_off % chunk_size == 0); | |
13603 | ceph_assert(blob_aligned_len() % chunk_size == 0); | |
13604 | ||
13605 | res = blob_aligned_len() <= prefer_deferred_size && | |
13606 | blob_aligned_len() <= ondisk && | |
13607 | blob.is_allocated(b_off, blob_aligned_len()); | |
13608 | if (res) { | |
13609 | blob_ref = ep->blob; | |
13610 | blob_start = ep->blob_start(); | |
13611 | } | |
13612 | } | |
13613 | return res; | |
13614 | } | |
13615 | ||
13616 | bool BlueStore::BigDeferredWriteContext::apply_defer() | |
13617 | { | |
13618 | int r = blob_ref->get_blob().map( | |
13619 | b_off, blob_aligned_len(), | |
13620 | [&](const bluestore_pextent_t& pext, | |
13621 | uint64_t offset, | |
13622 | uint64_t length) { | |
13623 | // apply deferred if overwrite breaks blob continuity only. | |
13624 | // if it totally overlaps some pextent - fallback to regular write | |
13625 | if (pext.offset < offset || | |
13626 | pext.end() > offset + length) { | |
13627 | res_extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13628 | return 0; | |
13629 | } | |
13630 | return -1; | |
13631 | }); | |
13632 | return r >= 0; | |
13633 | } | |
13634 | ||
13635 | void BlueStore::_do_write_big_apply_deferred( | |
13636 | TransContext* txc, | |
13637 | CollectionRef& c, | |
13638 | OnodeRef o, | |
13639 | BlueStore::BigDeferredWriteContext& dctx, | |
13640 | bufferlist::iterator& blp, | |
13641 | WriteContext* wctx) | |
13642 | { | |
13643 | bufferlist bl; | |
13644 | dout(20) << __func__ << " reading head 0x" << std::hex << dctx.head_read | |
13645 | << " and tail 0x" << dctx.tail_read << std::dec << dendl; | |
13646 | if (dctx.head_read) { | |
13647 | int r = _do_read(c.get(), o, | |
13648 | dctx.off - dctx.head_read, | |
13649 | dctx.head_read, | |
13650 | bl, | |
13651 | 0); | |
13652 | ceph_assert(r >= 0 && r <= (int)dctx.head_read); | |
13653 | size_t zlen = dctx.head_read - r; | |
13654 | if (zlen) { | |
13655 | bl.append_zero(zlen); | |
13656 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13657 | } | |
13658 | logger->inc(l_bluestore_write_penalty_read_ops); | |
13659 | } | |
13660 | blp.copy(dctx.used, bl); | |
13661 | ||
13662 | if (dctx.tail_read) { | |
13663 | bufferlist tail_bl; | |
13664 | int r = _do_read(c.get(), o, | |
13665 | dctx.off + dctx.used, dctx.tail_read, | |
13666 | tail_bl, 0); | |
13667 | ceph_assert(r >= 0 && r <= (int)dctx.tail_read); | |
13668 | size_t zlen = dctx.tail_read - r; | |
13669 | if (zlen) { | |
13670 | tail_bl.append_zero(zlen); | |
13671 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13672 | } | |
13673 | bl.claim_append(tail_bl); | |
13674 | logger->inc(l_bluestore_write_penalty_read_ops); | |
13675 | } | |
13676 | auto& b0 = dctx.blob_ref; | |
13677 | _buffer_cache_write(txc, b0, dctx.b_off, bl, | |
13678 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
13679 | ||
13680 | b0->dirty_blob().calc_csum(dctx.b_off, bl); | |
13681 | ||
13682 | Extent* le = o->extent_map.set_lextent(c, dctx.off, | |
13683 | dctx.off - dctx.blob_start, dctx.used, b0, &wctx->old_extents); | |
13684 | ||
13685 | // in fact this is a no-op for big writes but left here to maintain | |
13686 | // uniformity and avoid missing after some refactor. | |
13687 | b0->dirty_blob().mark_used(le->blob_offset, le->length); | |
13688 | txc->statfs_delta.stored() += le->length; | |
13689 | ||
13690 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
13691 | bluestore_deferred_op_t* op = _get_deferred_op(txc); | |
13692 | op->op = bluestore_deferred_op_t::OP_WRITE; | |
13693 | op->extents.swap(dctx.res_extents); | |
13694 | op->data = std::move(bl); | |
13695 | } | |
13696 | } | |
13697 | ||
7c673cae FG |
13698 | void BlueStore::_do_write_big( |
13699 | TransContext *txc, | |
13700 | CollectionRef &c, | |
13701 | OnodeRef o, | |
13702 | uint64_t offset, uint64_t length, | |
13703 | bufferlist::iterator& blp, | |
13704 | WriteContext *wctx) | |
13705 | { | |
13706 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13707 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
13708 | << " compress " << (int)wctx->compress | |
13709 | << dendl; | |
13710 | logger->inc(l_bluestore_write_big); | |
13711 | logger->inc(l_bluestore_write_big_bytes, length); | |
11fdf7f2 | 13712 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
f67539c2 | 13713 | uint64_t prefer_deferred_size_snapshot = prefer_deferred_size.load(); |
7c673cae FG |
13714 | while (length > 0) { |
13715 | bool new_blob = false; | |
11fdf7f2 | 13716 | uint32_t l = std::min(max_bsize, length); |
7c673cae FG |
13717 | BlobRef b; |
13718 | uint32_t b_off = 0; | |
13719 | ||
13720 | //attempting to reuse existing blob | |
13721 | if (!wctx->compress) { | |
7c673cae | 13722 | auto end = o->extent_map.extent_map.end(); |
f67539c2 TL |
13723 | |
13724 | if (prefer_deferred_size_snapshot && | |
13725 | l <= prefer_deferred_size_snapshot * 2) { | |
13726 | // Single write that spans two adjusted existing blobs can result | |
13727 | // in up to two deferred blocks of 'prefer_deferred_size' | |
13728 | // So we're trying to minimize the amount of resulting blobs | |
13729 | // and preserve 2 blobs rather than inserting one more in between | |
13730 | // E.g. write 0x10000~20000 over existing blobs | |
13731 | // (0x0~20000 and 0x20000~20000) is better (from subsequent reading | |
13732 | // performance point of view) to result in two deferred writes to | |
13733 | // existing blobs than having 3 blobs: 0x0~10000, 0x10000~20000, 0x30000~10000 | |
13734 | ||
13735 | // look for an existing mutable blob we can write into | |
13736 | auto ep = o->extent_map.seek_lextent(offset); | |
13737 | auto ep_next = end; | |
13738 | BigDeferredWriteContext head_info, tail_info; | |
13739 | ||
13740 | bool will_defer = ep != end ? | |
13741 | head_info.can_defer(ep, | |
13742 | prefer_deferred_size_snapshot, | |
13743 | block_size, | |
13744 | offset, | |
13745 | l) : | |
13746 | false; | |
13747 | auto offset_next = offset + head_info.used; | |
13748 | auto remaining = l - head_info.used; | |
13749 | if (will_defer && remaining) { | |
13750 | will_defer = false; | |
13751 | if (remaining <= prefer_deferred_size_snapshot) { | |
13752 | ep_next = o->extent_map.seek_lextent(offset_next); | |
13753 | // check if we can defer remaining totally | |
13754 | will_defer = ep_next == end ? | |
13755 | false : | |
13756 | tail_info.can_defer(ep_next, | |
13757 | prefer_deferred_size_snapshot, | |
13758 | block_size, | |
13759 | offset_next, | |
13760 | remaining); | |
13761 | will_defer = will_defer && remaining == tail_info.used; | |
13762 | } | |
13763 | } | |
13764 | if (will_defer) { | |
13765 | dout(20) << __func__ << " " << *(head_info.blob_ref) | |
13766 | << " deferring big " << std::hex | |
13767 | << " (0x" << head_info.b_off << "~" << head_info.blob_aligned_len() << ")" | |
13768 | << std::dec << " write via deferred" | |
13769 | << dendl; | |
13770 | if (remaining) { | |
13771 | dout(20) << __func__ << " " << *(tail_info.blob_ref) | |
13772 | << " deferring big " << std::hex | |
13773 | << " (0x" << tail_info.b_off << "~" << tail_info.blob_aligned_len() << ")" | |
13774 | << std::dec << " write via deferred" | |
13775 | << dendl; | |
13776 | } | |
13777 | ||
13778 | will_defer = head_info.apply_defer(); | |
13779 | if (!will_defer) { | |
13780 | dout(20) << __func__ | |
13781 | << " deferring big fell back, head isn't continuous" | |
13782 | << dendl; | |
13783 | } else if (remaining) { | |
13784 | will_defer = tail_info.apply_defer(); | |
13785 | if (!will_defer) { | |
13786 | dout(20) << __func__ | |
13787 | << " deferring big fell back, tail isn't continuous" | |
13788 | << dendl; | |
13789 | } | |
13790 | } | |
13791 | } | |
13792 | if (will_defer) { | |
13793 | _do_write_big_apply_deferred(txc, c, o, head_info, blp, wctx); | |
13794 | if (remaining) { | |
13795 | _do_write_big_apply_deferred(txc, c, o, tail_info, | |
13796 | blp, wctx); | |
13797 | } | |
13798 | offset += l; | |
13799 | length -= l; | |
13800 | logger->inc(l_bluestore_write_big_blobs, remaining ? 2 : 1); | |
13801 | logger->inc(l_bluestore_write_big_deferred, remaining ? 2 : 1); | |
13802 | continue; | |
13803 | } | |
13804 | } | |
13805 | ||
13806 | o->extent_map.punch_hole(c, offset, l, &wctx->old_extents); | |
13807 | ||
13808 | // seek again as punch_hole could invalidate ep | |
7c673cae | 13809 | auto ep = o->extent_map.seek_lextent(offset); |
f67539c2 TL |
13810 | auto begin = o->extent_map.extent_map.begin(); |
13811 | auto prev_ep = end; | |
13812 | if (ep != begin) { | |
13813 | prev_ep = ep; | |
7c673cae | 13814 | --prev_ep; |
7c673cae | 13815 | } |
f67539c2 TL |
13816 | dout(20) << __func__ << " no deferred" << dendl; |
13817 | ||
7c673cae FG |
13818 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; |
13819 | // search suitable extent in both forward and reverse direction in | |
13820 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 13821 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
13822 | bool any_change; |
13823 | do { | |
13824 | any_change = false; | |
13825 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
f67539c2 TL |
13826 | dout(20) << __func__ << " considering " << *ep << dendl; |
13827 | dout(20) << __func__ << " considering " << *(ep->blob) | |
13828 | << " bstart 0x" << std::hex << ep->blob_start() << std::dec << dendl; | |
13829 | ||
13830 | if (offset >= ep->blob_start() && | |
224ce89b | 13831 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13832 | offset - ep->blob_start(), |
13833 | &l)) { | |
13834 | b = ep->blob; | |
f67539c2 | 13835 | b_off = offset - ep->blob_start(); |
7c673cae FG |
13836 | prev_ep = end; // to avoid check below |
13837 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13838 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13839 | } else { |
13840 | ++ep; | |
13841 | any_change = true; | |
13842 | } | |
13843 | } | |
13844 | ||
13845 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
f67539c2 TL |
13846 | dout(20) << __func__ << " considering rev " << *prev_ep << dendl; |
13847 | dout(20) << __func__ << " considering reverse " << *(prev_ep->blob) | |
13848 | << " bstart 0x" << std::hex << prev_ep->blob_start() << std::dec << dendl; | |
13849 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, | |
7c673cae FG |
13850 | offset - prev_ep->blob_start(), |
13851 | &l)) { | |
13852 | b = prev_ep->blob; | |
13853 | b_off = offset - prev_ep->blob_start(); | |
13854 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13855 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13856 | } else if (prev_ep != begin) { |
13857 | --prev_ep; | |
13858 | any_change = true; | |
13859 | } else { | |
13860 | prev_ep = end; // to avoid useless first extent re-check | |
13861 | } | |
13862 | } | |
13863 | } while (b == nullptr && any_change); | |
f67539c2 TL |
13864 | } else { |
13865 | o->extent_map.punch_hole(c, offset, l, &wctx->old_extents); | |
13866 | } // if (!wctx->compress) | |
13867 | ||
7c673cae FG |
13868 | if (b == nullptr) { |
13869 | b = c->new_blob(); | |
13870 | b_off = 0; | |
13871 | new_blob = true; | |
13872 | } | |
7c673cae FG |
13873 | bufferlist t; |
13874 | blp.copy(l, t); | |
13875 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); | |
13876 | offset += l; | |
13877 | length -= l; | |
13878 | logger->inc(l_bluestore_write_big_blobs); | |
13879 | } | |
13880 | } | |
13881 | ||
13882 | int BlueStore::_do_alloc_write( | |
13883 | TransContext *txc, | |
13884 | CollectionRef coll, | |
13885 | OnodeRef o, | |
13886 | WriteContext *wctx) | |
13887 | { | |
13888 | dout(20) << __func__ << " txc " << txc | |
13889 | << " " << wctx->writes.size() << " blobs" | |
13890 | << dendl; | |
3efd9988 FG |
13891 | if (wctx->writes.empty()) { |
13892 | return 0; | |
7c673cae FG |
13893 | } |
13894 | ||
7c673cae FG |
13895 | CompressorRef c; |
13896 | double crr = 0; | |
13897 | if (wctx->compress) { | |
13898 | c = select_option( | |
13899 | "compression_algorithm", | |
13900 | compressor, | |
13901 | [&]() { | |
13902 | string val; | |
13903 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
13904 | CompressorRef cp = compressor; | |
13905 | if (!cp || cp->get_type_name() != val) { | |
13906 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
13907 | if (!cp) { |
13908 | if (_set_compression_alert(false, val.c_str())) { | |
13909 | derr << __func__ << " unable to initialize " << val.c_str() | |
13910 | << " compressor" << dendl; | |
13911 | } | |
13912 | } | |
7c673cae FG |
13913 | } |
13914 | return boost::optional<CompressorRef>(cp); | |
13915 | } | |
13916 | return boost::optional<CompressorRef>(); | |
13917 | } | |
13918 | ); | |
13919 | ||
13920 | crr = select_option( | |
13921 | "compression_required_ratio", | |
13922 | cct->_conf->bluestore_compression_required_ratio, | |
13923 | [&]() { | |
13924 | double val; | |
3efd9988 | 13925 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
7c673cae FG |
13926 | return boost::optional<double>(val); |
13927 | } | |
13928 | return boost::optional<double>(); | |
13929 | } | |
13930 | ); | |
13931 | } | |
13932 | ||
13933 | // checksum | |
11fdf7f2 | 13934 | int64_t csum = csum_type.load(); |
7c673cae FG |
13935 | csum = select_option( |
13936 | "csum_type", | |
13937 | csum, | |
13938 | [&]() { | |
11fdf7f2 | 13939 | int64_t val; |
3efd9988 | 13940 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
11fdf7f2 | 13941 | return boost::optional<int64_t>(val); |
7c673cae | 13942 | } |
11fdf7f2 | 13943 | return boost::optional<int64_t>(); |
7c673cae FG |
13944 | } |
13945 | ); | |
13946 | ||
3efd9988 FG |
13947 | // compress (as needed) and calc needed space |
13948 | uint64_t need = 0; | |
11fdf7f2 | 13949 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 13950 | for (auto& wi : wctx->writes) { |
3efd9988 | 13951 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 13952 | auto start = mono_clock::now(); |
7c673cae FG |
13953 | |
13954 | // compress | |
11fdf7f2 TL |
13955 | ceph_assert(wi.b_off == 0); |
13956 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 13957 | |
7c673cae FG |
13958 | // FIXME: memory alignment here is bad |
13959 | bufferlist t; | |
f67539c2 TL |
13960 | boost::optional<int32_t> compressor_message; |
13961 | int r = c->compress(wi.bl, t, compressor_message); | |
3efd9988 | 13962 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 13963 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
13964 | bool rejected = false; |
13965 | uint64_t compressed_len = t.length(); | |
13966 | // do an approximate (fast) estimation for resulting blob size | |
13967 | // that doesn't take header overhead into account | |
11fdf7f2 | 13968 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13969 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
13970 | bluestore_compression_header_t chdr; | |
13971 | chdr.type = c->get_type(); | |
13972 | chdr.length = t.length(); | |
f67539c2 | 13973 | chdr.compressor_message = compressor_message; |
a8e16298 TL |
13974 | encode(chdr, wi.compressed_bl); |
13975 | wi.compressed_bl.claim_append(t); | |
13976 | ||
13977 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 13978 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13979 | if (result_len <= want_len && result_len < wi.blob_length) { |
13980 | // Cool. We compressed at least as much as we were hoping to. | |
13981 | // pad out to min_alloc_size | |
13982 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
13983 | wi.compressed_len = compressed_len; | |
13984 | wi.compressed = true; | |
13985 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
13986 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
13987 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
13988 | << " with " << c->get_type() | |
13989 | << std::dec << dendl; | |
13990 | txc->statfs_delta.compressed() += compressed_len; | |
13991 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
13992 | txc->statfs_delta.compressed_allocated() += result_len; | |
13993 | logger->inc(l_bluestore_compress_success_count); | |
13994 | need += result_len; | |
13995 | } else { | |
13996 | rejected = true; | |
13997 | } | |
13998 | } else if (r != 0) { | |
13999 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
14000 | << " bytes compressed using " << c->get_type_name() | |
14001 | << std::dec | |
14002 | << " failed with errcode = " << r | |
14003 | << ", leaving uncompressed" | |
14004 | << dendl; | |
14005 | logger->inc(l_bluestore_compress_rejected_count); | |
14006 | need += wi.blob_length; | |
7c673cae | 14007 | } else { |
a8e16298 TL |
14008 | rejected = true; |
14009 | } | |
14010 | ||
14011 | if (rejected) { | |
3efd9988 | 14012 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 14013 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
14014 | << " with " << c->get_type() |
14015 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 14016 | << " -> 0x" << want_len |
3efd9988 FG |
14017 | << ", leaving uncompressed" |
14018 | << std::dec << dendl; | |
14019 | logger->inc(l_bluestore_compress_rejected_count); | |
14020 | need += wi.blob_length; | |
7c673cae | 14021 | } |
494da23a TL |
14022 | log_latency("compress@_do_alloc_write", |
14023 | l_bluestore_compress_lat, | |
14024 | mono_clock::now() - start, | |
14025 | cct->_conf->bluestore_log_op_age ); | |
3efd9988 FG |
14026 | } else { |
14027 | need += wi.blob_length; | |
7c673cae | 14028 | } |
3efd9988 | 14029 | } |
a8e16298 | 14030 | PExtentVector prealloc; |
3efd9988 | 14031 | prealloc.reserve(2 * wctx->writes.size());; |
11fdf7f2 | 14032 | int64_t prealloc_left = 0; |
f67539c2 | 14033 | prealloc_left = shared_alloc.a->allocate( |
3efd9988 FG |
14034 | need, min_alloc_size, need, |
14035 | 0, &prealloc); | |
eafe8130 | 14036 | if (prealloc_left < 0 || prealloc_left < (int64_t)need) { |
11fdf7f2 | 14037 | derr << __func__ << " failed to allocate 0x" << std::hex << need |
eafe8130 | 14038 | << " allocated 0x " << (prealloc_left < 0 ? 0 : prealloc_left) |
11fdf7f2 | 14039 | << " min_alloc_size 0x" << min_alloc_size |
f67539c2 | 14040 | << " available 0x " << shared_alloc.a->get_free() |
11fdf7f2 TL |
14041 | << std::dec << dendl; |
14042 | if (prealloc.size()) { | |
f67539c2 | 14043 | shared_alloc.a->release(prealloc); |
11fdf7f2 | 14044 | } |
a8e16298 TL |
14045 | return -ENOSPC; |
14046 | } | |
9f95a23c | 14047 | _collect_allocation_stats(need, min_alloc_size, prealloc.size()); |
a8e16298 | 14048 | |
f67539c2 TL |
14049 | if (bdev->is_smr()) { |
14050 | std::deque<uint64_t> zones_to_clean; | |
14051 | if (shared_alloc.a->zoned_get_zones_to_clean(&zones_to_clean)) { | |
14052 | std::lock_guard l{zoned_cleaner_lock}; | |
14053 | zoned_cleaner_queue.swap(zones_to_clean); | |
14054 | zoned_cleaner_cond.notify_one(); | |
14055 | } | |
14056 | } | |
14057 | ||
3efd9988 FG |
14058 | dout(20) << __func__ << " prealloc " << prealloc << dendl; |
14059 | auto prealloc_pos = prealloc.begin(); | |
14060 | ||
14061 | for (auto& wi : wctx->writes) { | |
14062 | BlobRef b = wi.b; | |
14063 | bluestore_blob_t& dblob = b->dirty_blob(); | |
14064 | uint64_t b_off = wi.b_off; | |
14065 | bufferlist *l = &wi.bl; | |
14066 | uint64_t final_length = wi.blob_length; | |
14067 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
14068 | if (wi.compressed) { |
14069 | final_length = wi.compressed_bl.length(); | |
14070 | csum_length = final_length; | |
adb31ebb | 14071 | unsigned csum_order = ctz(csum_length); |
3efd9988 FG |
14072 | l = &wi.compressed_bl; |
14073 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
adb31ebb TL |
14074 | if (csum != Checksummer::CSUM_NONE) { |
14075 | dout(20) << __func__ << " initialize csum setting for compressed blob " << *b | |
14076 | << " csum_type " << Checksummer::get_csum_type_string(csum) | |
14077 | << " csum_order " << csum_order | |
14078 | << " csum_length 0x" << std::hex << csum_length | |
14079 | << " blob_length 0x" << wi.blob_length | |
14080 | << " compressed_length 0x" << wi.compressed_len << std::dec | |
14081 | << dendl; | |
14082 | dblob.init_csum(csum, csum_order, csum_length); | |
14083 | } | |
3efd9988 | 14084 | } else if (wi.new_blob) { |
adb31ebb | 14085 | unsigned csum_order; |
7c673cae | 14086 | // initialize newly created blob only |
11fdf7f2 | 14087 | ceph_assert(dblob.is_mutable()); |
7c673cae FG |
14088 | if (l->length() != wi.blob_length) { |
14089 | // hrm, maybe we could do better here, but let's not bother. | |
14090 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
14091 | << block_size_order << dendl; | |
31f18b77 | 14092 | csum_order = block_size_order; |
7c673cae FG |
14093 | } else { |
14094 | csum_order = std::min(wctx->csum_order, ctz(l->length())); | |
14095 | } | |
14096 | // try to align blob with max_blob_size to improve | |
14097 | // its reuse ratio, e.g. in case of reverse write | |
14098 | uint32_t suggested_boff = | |
14099 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
14100 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
14101 | suggested_boff + final_length <= max_bsize && | |
14102 | suggested_boff > b_off) { | |
181888fb | 14103 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 14104 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 14105 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
14106 | csum_length += suggested_boff - b_off; |
14107 | b_off = suggested_boff; | |
14108 | } | |
181888fb FG |
14109 | if (csum != Checksummer::CSUM_NONE) { |
14110 | dout(20) << __func__ << " initialize csum setting for new blob " << *b | |
14111 | << " csum_type " << Checksummer::get_csum_type_string(csum) | |
14112 | << " csum_order " << csum_order | |
14113 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
14114 | << dendl; | |
14115 | dblob.init_csum(csum, csum_order, csum_length); | |
14116 | } | |
7c673cae FG |
14117 | } |
14118 | ||
a8e16298 | 14119 | PExtentVector extents; |
3efd9988 FG |
14120 | int64_t left = final_length; |
14121 | while (left > 0) { | |
11fdf7f2 | 14122 | ceph_assert(prealloc_left > 0); |
3efd9988 FG |
14123 | if (prealloc_pos->length <= left) { |
14124 | prealloc_left -= prealloc_pos->length; | |
14125 | left -= prealloc_pos->length; | |
14126 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
14127 | extents.push_back(*prealloc_pos); | |
14128 | ++prealloc_pos; | |
14129 | } else { | |
14130 | extents.emplace_back(prealloc_pos->offset, left); | |
14131 | prealloc_pos->offset += left; | |
14132 | prealloc_pos->length -= left; | |
14133 | prealloc_left -= left; | |
14134 | txc->statfs_delta.allocated() += left; | |
14135 | left = 0; | |
14136 | break; | |
14137 | } | |
14138 | } | |
7c673cae | 14139 | for (auto& p : extents) { |
3efd9988 | 14140 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 14141 | } |
11fdf7f2 | 14142 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 14143 | |
181888fb FG |
14144 | dout(20) << __func__ << " blob " << *b << dendl; |
14145 | if (dblob.has_csum()) { | |
7c673cae FG |
14146 | dblob.calc_csum(b_off, *l); |
14147 | } | |
181888fb | 14148 | |
7c673cae | 14149 | if (wi.mark_unused) { |
1911f103 | 14150 | ceph_assert(!dblob.is_compressed()); |
7c673cae FG |
14151 | auto b_end = b_off + wi.bl.length(); |
14152 | if (b_off) { | |
14153 | dblob.add_unused(0, b_off); | |
14154 | } | |
1911f103 TL |
14155 | uint64_t llen = dblob.get_logical_length(); |
14156 | if (b_end < llen) { | |
14157 | dblob.add_unused(b_end, llen - b_end); | |
7c673cae FG |
14158 | } |
14159 | } | |
14160 | ||
14161 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
14162 | b_off + (wi.b_off0 - wi.b_off), | |
14163 | wi.length0, | |
14164 | wi.b, | |
14165 | nullptr); | |
14166 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
14167 | txc->statfs_delta.stored() += le->length; | |
14168 | dout(20) << __func__ << " lex " << *le << dendl; | |
14169 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
14170 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
14171 | ||
14172 | // queue io | |
11fdf7f2 | 14173 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae | 14174 | if (l->length() <= prefer_deferred_size.load()) { |
f67539c2 | 14175 | dout(20) << __func__ << " deferring 0x" << std::hex |
7c673cae | 14176 | << l->length() << std::dec << " write via deferred" << dendl; |
9f95a23c | 14177 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
14178 | op->op = bluestore_deferred_op_t::OP_WRITE; |
14179 | int r = b->get_blob().map( | |
14180 | b_off, l->length(), | |
14181 | [&](uint64_t offset, uint64_t length) { | |
14182 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
14183 | return 0; | |
14184 | }); | |
11fdf7f2 | 14185 | ceph_assert(r == 0); |
7c673cae | 14186 | op->data = *l; |
f67539c2 | 14187 | logger->inc(l_bluestore_write_deferred); |
7c673cae FG |
14188 | } else { |
14189 | b->get_blob().map_bl( | |
14190 | b_off, *l, | |
14191 | [&](uint64_t offset, bufferlist& t) { | |
14192 | bdev->aio_write(offset, t, &txc->ioc, false); | |
14193 | }); | |
f67539c2 | 14194 | logger->inc(l_bluestore_write_new); |
7c673cae FG |
14195 | } |
14196 | } | |
14197 | } | |
11fdf7f2 TL |
14198 | ceph_assert(prealloc_pos == prealloc.end()); |
14199 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
14200 | return 0; |
14201 | } | |
14202 | ||
14203 | void BlueStore::_wctx_finish( | |
14204 | TransContext *txc, | |
14205 | CollectionRef& c, | |
14206 | OnodeRef o, | |
31f18b77 FG |
14207 | WriteContext *wctx, |
14208 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14209 | { |
14210 | auto oep = wctx->old_extents.begin(); | |
14211 | while (oep != wctx->old_extents.end()) { | |
14212 | auto &lo = *oep; | |
14213 | oep = wctx->old_extents.erase(oep); | |
14214 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
14215 | BlobRef b = lo.e.blob; | |
14216 | const bluestore_blob_t& blob = b->get_blob(); | |
14217 | if (blob.is_compressed()) { | |
14218 | if (lo.blob_empty) { | |
14219 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
14220 | } | |
14221 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
14222 | } | |
14223 | auto& r = lo.r; | |
14224 | txc->statfs_delta.stored() -= lo.e.length; | |
14225 | if (!r.empty()) { | |
f67539c2 | 14226 | dout(20) << __func__ << " blob " << *b << " release " << r << dendl; |
7c673cae FG |
14227 | if (blob.is_shared()) { |
14228 | PExtentVector final; | |
14229 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
14230 | bool unshare = false; |
14231 | bool* unshare_ptr = | |
14232 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 14233 | for (auto e : r) { |
31f18b77 FG |
14234 | b->shared_blob->put_ref( |
14235 | e.offset, e.length, &final, | |
11fdf7f2 TL |
14236 | unshare_ptr); |
14237 | } | |
14238 | if (unshare) { | |
14239 | ceph_assert(maybe_unshared_blobs); | |
14240 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
14241 | } |
14242 | dout(20) << __func__ << " shared_blob release " << final | |
14243 | << " from " << *b->shared_blob << dendl; | |
14244 | txc->write_shared_blob(b->shared_blob); | |
14245 | r.clear(); | |
14246 | r.swap(final); | |
14247 | } | |
14248 | } | |
14249 | // we can't invalidate our logical extents as we drop them because | |
14250 | // other lextents (either in our onode or others) may still | |
14251 | // reference them. but we can throw out anything that is no | |
14252 | // longer allocated. Note that this will leave behind edge bits | |
14253 | // that are no longer referenced but not deallocated (until they | |
14254 | // age out of the cache naturally). | |
14255 | b->discard_unallocated(c.get()); | |
14256 | for (auto e : r) { | |
14257 | dout(20) << __func__ << " release " << e << dendl; | |
14258 | txc->released.insert(e.offset, e.length); | |
14259 | txc->statfs_delta.allocated() -= e.length; | |
14260 | if (blob.is_compressed()) { | |
14261 | txc->statfs_delta.compressed_allocated() -= e.length; | |
14262 | } | |
14263 | } | |
9f95a23c TL |
14264 | |
14265 | if (b->is_spanning() && !b->is_referenced() && lo.blob_empty) { | |
7c673cae FG |
14266 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b |
14267 | << dendl; | |
14268 | o->extent_map.spanning_blob_map.erase(b->id); | |
14269 | } | |
9f95a23c | 14270 | delete &lo; |
7c673cae FG |
14271 | } |
14272 | } | |
14273 | ||
14274 | void BlueStore::_do_write_data( | |
14275 | TransContext *txc, | |
14276 | CollectionRef& c, | |
14277 | OnodeRef o, | |
14278 | uint64_t offset, | |
14279 | uint64_t length, | |
14280 | bufferlist& bl, | |
14281 | WriteContext *wctx) | |
14282 | { | |
14283 | uint64_t end = offset + length; | |
14284 | bufferlist::iterator p = bl.begin(); | |
14285 | ||
14286 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
14287 | (length != min_alloc_size)) { | |
14288 | // we fall within the same block | |
14289 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
14290 | } else { | |
14291 | uint64_t head_offset, head_length; | |
14292 | uint64_t middle_offset, middle_length; | |
14293 | uint64_t tail_offset, tail_length; | |
14294 | ||
14295 | head_offset = offset; | |
11fdf7f2 | 14296 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 14297 | |
11fdf7f2 TL |
14298 | tail_offset = p2align(end, min_alloc_size); |
14299 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
14300 | |
14301 | middle_offset = head_offset + head_length; | |
14302 | middle_length = length - head_length - tail_length; | |
14303 | ||
14304 | if (head_length) { | |
14305 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
14306 | } | |
14307 | ||
f67539c2 | 14308 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); |
7c673cae FG |
14309 | |
14310 | if (tail_length) { | |
14311 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
14312 | } | |
14313 | } | |
14314 | } | |
14315 | ||
31f18b77 FG |
14316 | void BlueStore::_choose_write_options( |
14317 | CollectionRef& c, | |
14318 | OnodeRef o, | |
14319 | uint32_t fadvise_flags, | |
14320 | WriteContext *wctx) | |
7c673cae | 14321 | { |
7c673cae FG |
14322 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
14323 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 14324 | wctx->buffered = true; |
7c673cae FG |
14325 | } else if (cct->_conf->bluestore_default_buffered_write && |
14326 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
14327 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
14328 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 14329 | wctx->buffered = true; |
7c673cae FG |
14330 | } |
14331 | ||
31f18b77 FG |
14332 | // apply basic csum block size |
14333 | wctx->csum_order = block_size_order; | |
7c673cae FG |
14334 | |
14335 | // compression parameters | |
14336 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
14337 | auto cm = select_option( | |
14338 | "compression_mode", | |
31f18b77 | 14339 | comp_mode.load(), |
7c673cae FG |
14340 | [&]() { |
14341 | string val; | |
11fdf7f2 | 14342 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
31f18b77 FG |
14343 | return boost::optional<Compressor::CompressionMode>( |
14344 | Compressor::get_comp_mode_type(val)); | |
7c673cae FG |
14345 | } |
14346 | return boost::optional<Compressor::CompressionMode>(); | |
14347 | } | |
14348 | ); | |
31f18b77 FG |
14349 | |
14350 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
14351 | ((cm == Compressor::COMP_FORCE) || |
14352 | (cm == Compressor::COMP_AGGRESSIVE && | |
14353 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
14354 | (cm == Compressor::COMP_PASSIVE && | |
14355 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
14356 | ||
14357 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
14358 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
14359 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
14360 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 14361 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 14362 | |
7c673cae | 14363 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 14364 | |
7c673cae | 14365 | if (o->onode.expected_write_size) { |
224ce89b | 14366 | wctx->csum_order = std::max(min_alloc_size_order, |
31f18b77 | 14367 | (uint8_t)ctz(o->onode.expected_write_size)); |
7c673cae | 14368 | } else { |
224ce89b | 14369 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
14370 | } |
14371 | ||
31f18b77 FG |
14372 | if (wctx->compress) { |
14373 | wctx->target_blob_size = select_option( | |
7c673cae | 14374 | "compression_max_blob_size", |
31f18b77 | 14375 | comp_max_blob_size.load(), |
7c673cae | 14376 | [&]() { |
11fdf7f2 TL |
14377 | int64_t val; |
14378 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
7c673cae FG |
14379 | return boost::optional<uint64_t>((uint64_t)val); |
14380 | } | |
14381 | return boost::optional<uint64_t>(); | |
14382 | } | |
14383 | ); | |
14384 | } | |
14385 | } else { | |
31f18b77 FG |
14386 | if (wctx->compress) { |
14387 | wctx->target_blob_size = select_option( | |
7c673cae | 14388 | "compression_min_blob_size", |
31f18b77 | 14389 | comp_min_blob_size.load(), |
7c673cae | 14390 | [&]() { |
11fdf7f2 TL |
14391 | int64_t val; |
14392 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
7c673cae FG |
14393 | return boost::optional<uint64_t>((uint64_t)val); |
14394 | } | |
14395 | return boost::optional<uint64_t>(); | |
14396 | } | |
14397 | ); | |
14398 | } | |
14399 | } | |
31f18b77 | 14400 | |
7c673cae | 14401 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
14402 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
14403 | wctx->target_blob_size = max_bsize; | |
7c673cae | 14404 | } |
31f18b77 | 14405 | |
7c673cae FG |
14406 | // set the min blob size floor at 2x the min_alloc_size, or else we |
14407 | // won't be able to allocate a smaller extent for the compressed | |
14408 | // data. | |
31f18b77 FG |
14409 | if (wctx->compress && |
14410 | wctx->target_blob_size < min_alloc_size * 2) { | |
14411 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 14412 | } |
31f18b77 FG |
14413 | |
14414 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
14415 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
14416 | << " compress=" << (int)wctx->compress |
14417 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
14418 | << std::dec << dendl; |
14419 | } | |
14420 | ||
14421 | int BlueStore::_do_gc( | |
14422 | TransContext *txc, | |
14423 | CollectionRef& c, | |
14424 | OnodeRef o, | |
31f18b77 FG |
14425 | const WriteContext& wctx, |
14426 | uint64_t *dirty_start, | |
14427 | uint64_t *dirty_end) | |
14428 | { | |
31f18b77 | 14429 | |
1adf2230 | 14430 | bool dirty_range_updated = false; |
31f18b77 | 14431 | WriteContext wctx_gc; |
7c673cae | 14432 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 14433 | |
eafe8130 | 14434 | auto & extents_to_collect = wctx.extents_to_gc; |
31f18b77 FG |
14435 | for (auto it = extents_to_collect.begin(); |
14436 | it != extents_to_collect.end(); | |
14437 | ++it) { | |
14438 | bufferlist bl; | |
eafe8130 TL |
14439 | auto offset = (*it).first; |
14440 | auto length = (*it).second; | |
14441 | dout(20) << __func__ << " processing " << std::hex | |
14442 | << offset << "~" << length << std::dec | |
14443 | << dendl; | |
14444 | int r = _do_read(c.get(), o, offset, length, bl, 0); | |
14445 | ceph_assert(r == (int)length); | |
31f18b77 | 14446 | |
eafe8130 TL |
14447 | _do_write_data(txc, c, o, offset, length, bl, &wctx_gc); |
14448 | logger->inc(l_bluestore_gc_merged, length); | |
31f18b77 | 14449 | |
eafe8130 TL |
14450 | if (*dirty_start > offset) { |
14451 | *dirty_start = offset; | |
1adf2230 | 14452 | dirty_range_updated = true; |
31f18b77 FG |
14453 | } |
14454 | ||
eafe8130 TL |
14455 | if (*dirty_end < offset + length) { |
14456 | *dirty_end = offset + length; | |
1adf2230 | 14457 | dirty_range_updated = true; |
31f18b77 FG |
14458 | } |
14459 | } | |
1adf2230 AA |
14460 | if (dirty_range_updated) { |
14461 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
14462 | } | |
31f18b77 FG |
14463 | |
14464 | dout(30) << __func__ << " alloc write" << dendl; | |
14465 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
14466 | if (r < 0) { | |
14467 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14468 | << dendl; | |
14469 | return r; | |
14470 | } | |
14471 | ||
14472 | _wctx_finish(txc, c, o, &wctx_gc); | |
14473 | return 0; | |
14474 | } | |
14475 | ||
14476 | int BlueStore::_do_write( | |
14477 | TransContext *txc, | |
14478 | CollectionRef& c, | |
14479 | OnodeRef o, | |
14480 | uint64_t offset, | |
14481 | uint64_t length, | |
14482 | bufferlist& bl, | |
14483 | uint32_t fadvise_flags) | |
14484 | { | |
14485 | int r = 0; | |
14486 | ||
14487 | dout(20) << __func__ | |
14488 | << " " << o->oid | |
14489 | << " 0x" << std::hex << offset << "~" << length | |
14490 | << " - have 0x" << o->onode.size | |
14491 | << " (" << std::dec << o->onode.size << ")" | |
f67539c2 TL |
14492 | << " bytes" << std::hex |
14493 | << " fadvise_flags 0x" << fadvise_flags | |
14494 | << " alloc_hint 0x" << o->onode.alloc_hint_flags | |
14495 | << " expected_object_size " << o->onode.expected_object_size | |
14496 | << " expected_write_size " << o->onode.expected_write_size | |
14497 | << std::dec | |
31f18b77 | 14498 | << dendl; |
81eedcae | 14499 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
14500 | |
14501 | if (length == 0) { | |
14502 | return 0; | |
14503 | } | |
14504 | ||
14505 | uint64_t end = offset + length; | |
14506 | ||
14507 | GarbageCollector gc(c->store->cct); | |
eafe8130 | 14508 | int64_t benefit = 0; |
31f18b77 FG |
14509 | auto dirty_start = offset; |
14510 | auto dirty_end = end; | |
14511 | ||
14512 | WriteContext wctx; | |
14513 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
14514 | o->extent_map.fault_range(db, offset, length); |
14515 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
14516 | r = _do_alloc_write(txc, c, o, &wctx); |
14517 | if (r < 0) { | |
14518 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14519 | << dendl; | |
14520 | goto out; | |
14521 | } | |
14522 | ||
eafe8130 TL |
14523 | if (wctx.extents_to_gc.empty() || |
14524 | wctx.extents_to_gc.range_start() > offset || | |
14525 | wctx.extents_to_gc.range_end() < offset + length) { | |
14526 | benefit = gc.estimate(offset, | |
14527 | length, | |
14528 | o->extent_map, | |
14529 | wctx.old_extents, | |
14530 | min_alloc_size); | |
14531 | } | |
14532 | ||
f67539c2 TL |
14533 | if (bdev->is_smr()) { |
14534 | if (wctx.old_extents.empty()) { | |
14535 | txc->zoned_note_new_object(o); | |
14536 | } else { | |
14537 | int64_t old_ondisk_offset = wctx.old_extents.begin()->r.begin()->offset; | |
14538 | txc->zoned_note_updated_object(o, old_ondisk_offset); | |
14539 | } | |
14540 | } | |
14541 | ||
31f18b77 FG |
14542 | // NB: _wctx_finish() will empty old_extents |
14543 | // so we must do gc estimation before that | |
7c673cae FG |
14544 | _wctx_finish(txc, c, o, &wctx); |
14545 | if (end > o->onode.size) { | |
14546 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 14547 | << std::dec << dendl; |
7c673cae FG |
14548 | o->onode.size = end; |
14549 | } | |
14550 | ||
11fdf7f2 | 14551 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
eafe8130 TL |
14552 | wctx.extents_to_gc.union_of(gc.get_extents_to_collect()); |
14553 | dout(20) << __func__ | |
14554 | << " perform garbage collection for compressed extents, " | |
14555 | << "expected benefit = " << benefit << " AUs" << dendl; | |
14556 | } | |
14557 | if (!wctx.extents_to_gc.empty()) { | |
14558 | dout(20) << __func__ << " perform garbage collection" << dendl; | |
14559 | ||
14560 | r = _do_gc(txc, c, o, | |
14561 | wctx, | |
14562 | &dirty_start, &dirty_end); | |
14563 | if (r < 0) { | |
14564 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
14565 | << dendl; | |
14566 | goto out; | |
7c673cae | 14567 | } |
eafe8130 TL |
14568 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
14569 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae | 14570 | } |
7c673cae | 14571 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
14572 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
14573 | ||
7c673cae FG |
14574 | r = 0; |
14575 | ||
14576 | out: | |
14577 | return r; | |
14578 | } | |
14579 | ||
14580 | int BlueStore::_write(TransContext *txc, | |
14581 | CollectionRef& c, | |
14582 | OnodeRef& o, | |
31f18b77 FG |
14583 | uint64_t offset, size_t length, |
14584 | bufferlist& bl, | |
14585 | uint32_t fadvise_flags) | |
7c673cae FG |
14586 | { |
14587 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14588 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14589 | << dendl; | |
35e4c445 FG |
14590 | int r = 0; |
14591 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14592 | r = -E2BIG; | |
14593 | } else { | |
14594 | _assign_nid(txc, o); | |
14595 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
14596 | txc->write_onode(o); | |
14597 | } | |
7c673cae FG |
14598 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14599 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14600 | << " = " << r << dendl; | |
14601 | return r; | |
14602 | } | |
14603 | ||
14604 | int BlueStore::_zero(TransContext *txc, | |
14605 | CollectionRef& c, | |
14606 | OnodeRef& o, | |
14607 | uint64_t offset, size_t length) | |
14608 | { | |
14609 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14610 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14611 | << dendl; | |
35e4c445 FG |
14612 | int r = 0; |
14613 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14614 | r = -E2BIG; | |
14615 | } else { | |
14616 | _assign_nid(txc, o); | |
14617 | r = _do_zero(txc, c, o, offset, length); | |
14618 | } | |
7c673cae FG |
14619 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14620 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14621 | << " = " << r << dendl; | |
14622 | return r; | |
14623 | } | |
14624 | ||
14625 | int BlueStore::_do_zero(TransContext *txc, | |
14626 | CollectionRef& c, | |
14627 | OnodeRef& o, | |
14628 | uint64_t offset, size_t length) | |
14629 | { | |
14630 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14631 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14632 | << dendl; | |
14633 | int r = 0; | |
14634 | ||
81eedcae | 14635 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14636 | |
14637 | WriteContext wctx; | |
14638 | o->extent_map.fault_range(db, offset, length); | |
14639 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 14640 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
14641 | _wctx_finish(txc, c, o, &wctx); |
14642 | ||
b32b8144 | 14643 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
14644 | o->onode.size = offset + length; |
14645 | dout(20) << __func__ << " extending size to " << offset + length | |
14646 | << dendl; | |
14647 | } | |
14648 | txc->write_onode(o); | |
14649 | ||
14650 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14651 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14652 | << " = " << r << dendl; | |
14653 | return r; | |
14654 | } | |
14655 | ||
14656 | void BlueStore::_do_truncate( | |
31f18b77 FG |
14657 | TransContext *txc, CollectionRef& c, OnodeRef o, uint64_t offset, |
14658 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14659 | { |
14660 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14661 | << " 0x" << std::hex << offset << std::dec << dendl; | |
14662 | ||
81eedcae | 14663 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14664 | |
14665 | if (offset == o->onode.size) | |
31f18b77 | 14666 | return; |
7c673cae | 14667 | |
f67539c2 | 14668 | WriteContext wctx; |
7c673cae | 14669 | if (offset < o->onode.size) { |
7c673cae FG |
14670 | uint64_t length = o->onode.size - offset; |
14671 | o->extent_map.fault_range(db, offset, length); | |
14672 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 FG |
14673 | o->extent_map.dirty_range(offset, length); |
14674 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); | |
7c673cae FG |
14675 | |
14676 | // if we have shards past EOF, ask for a reshard | |
14677 | if (!o->onode.extent_map_shards.empty() && | |
14678 | o->onode.extent_map_shards.back().offset >= offset) { | |
14679 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
14680 | if (offset) { | |
14681 | o->extent_map.request_reshard(offset - 1, offset + length); | |
14682 | } else { | |
14683 | o->extent_map.request_reshard(0, length); | |
14684 | } | |
14685 | } | |
14686 | } | |
14687 | ||
14688 | o->onode.size = offset; | |
14689 | ||
f67539c2 TL |
14690 | if (bdev->is_smr()) { |
14691 | // On zoned devices, we currently support only removing an object or | |
14692 | // truncating it to zero size, both of which fall through this code path. | |
14693 | ceph_assert(offset == 0 && !wctx.old_extents.empty()); | |
14694 | int64_t ondisk_offset = wctx.old_extents.begin()->r.begin()->offset; | |
14695 | txc->zoned_note_truncated_object(o, ondisk_offset); | |
14696 | } | |
14697 | ||
7c673cae FG |
14698 | txc->write_onode(o); |
14699 | } | |
14700 | ||
35e4c445 | 14701 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
14702 | CollectionRef& c, |
14703 | OnodeRef& o, | |
14704 | uint64_t offset) | |
14705 | { | |
14706 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14707 | << " 0x" << std::hex << offset << std::dec | |
14708 | << dendl; | |
35e4c445 FG |
14709 | int r = 0; |
14710 | if (offset >= OBJECT_MAX_SIZE) { | |
14711 | r = -E2BIG; | |
14712 | } else { | |
14713 | _do_truncate(txc, c, o, offset); | |
14714 | } | |
14715 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14716 | << " 0x" << std::hex << offset << std::dec | |
14717 | << " = " << r << dendl; | |
14718 | return r; | |
7c673cae FG |
14719 | } |
14720 | ||
14721 | int BlueStore::_do_remove( | |
14722 | TransContext *txc, | |
14723 | CollectionRef& c, | |
14724 | OnodeRef o) | |
14725 | { | |
31f18b77 | 14726 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
14727 | bool is_gen = !o->oid.is_no_gen(); |
14728 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
14729 | if (o->onode.has_omap()) { |
14730 | o->flush(); | |
9f95a23c | 14731 | _do_omap_clear(txc, o); |
7c673cae FG |
14732 | } |
14733 | o->exists = false; | |
14734 | string key; | |
14735 | for (auto &s : o->extent_map.shards) { | |
14736 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
14737 | << s.shard_info->offset << std::dec << dendl; | |
14738 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
14739 | [&](const string& final_key) { | |
14740 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14741 | } | |
14742 | ); | |
14743 | } | |
14744 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 14745 | txc->note_removed_object(o); |
7c673cae FG |
14746 | o->extent_map.clear(); |
14747 | o->onode = bluestore_onode_t(); | |
14748 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 14749 | |
224ce89b WB |
14750 | if (!is_gen || maybe_unshared_blobs.empty()) { |
14751 | return 0; | |
14752 | } | |
31f18b77 | 14753 | |
224ce89b WB |
14754 | // see if we can unshare blobs still referenced by the head |
14755 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
14756 | << maybe_unshared_blobs << dendl; | |
14757 | ghobject_t nogen = o->oid; | |
14758 | nogen.generation = ghobject_t::NO_GEN; | |
f67539c2 | 14759 | OnodeRef h = c->get_onode(nogen, false); |
224ce89b WB |
14760 | |
14761 | if (!h || !h->exists) { | |
14762 | return 0; | |
14763 | } | |
14764 | ||
14765 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
14766 | << " " << h->oid << dendl; | |
14767 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
14768 | for (auto& e : h->extent_map.extent_map) { | |
14769 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14770 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14771 | if (b.is_shared() && | |
14772 | sb->loaded && | |
14773 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
14774 | if (b.is_compressed()) { |
14775 | expect[sb].get(0, b.get_ondisk_length()); | |
14776 | } else { | |
14777 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
14778 | expect[sb].get(off, len); | |
14779 | return 0; | |
14780 | }); | |
14781 | } | |
224ce89b WB |
14782 | } |
14783 | } | |
31f18b77 | 14784 | |
224ce89b WB |
14785 | vector<SharedBlob*> unshared_blobs; |
14786 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
14787 | for (auto& p : expect) { | |
14788 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
14789 | if (p.first->persistent->ref_map == p.second) { | |
14790 | SharedBlob *sb = p.first; | |
14791 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
14792 | unshared_blobs.push_back(sb); | |
14793 | txc->unshare_blob(sb); | |
14794 | uint64_t sbid = c->make_blob_unshared(sb); | |
14795 | string key; | |
14796 | get_shared_blob_key(sbid, &key); | |
14797 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
14798 | } | |
14799 | } | |
14800 | ||
14801 | if (unshared_blobs.empty()) { | |
14802 | return 0; | |
14803 | } | |
14804 | ||
224ce89b WB |
14805 | for (auto& e : h->extent_map.extent_map) { |
14806 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14807 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14808 | if (b.is_shared() && | |
14809 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
14810 | sb) != unshared_blobs.end()) { | |
14811 | dout(20) << __func__ << " unsharing " << e << dendl; | |
14812 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
14813 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 14814 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
14815 | } |
14816 | } | |
224ce89b WB |
14817 | txc->write_onode(h); |
14818 | ||
7c673cae FG |
14819 | return 0; |
14820 | } | |
14821 | ||
14822 | int BlueStore::_remove(TransContext *txc, | |
14823 | CollectionRef& c, | |
14824 | OnodeRef &o) | |
14825 | { | |
11fdf7f2 TL |
14826 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
14827 | << " onode " << o.get() | |
14828 | << " txc "<< txc << dendl; | |
adb31ebb TL |
14829 | |
14830 | auto start_time = mono_clock::now(); | |
7c673cae | 14831 | int r = _do_remove(txc, c, o); |
adb31ebb TL |
14832 | log_latency_fn( |
14833 | __func__, | |
14834 | l_bluestore_remove_lat, | |
14835 | mono_clock::now() - start_time, | |
14836 | cct->_conf->bluestore_log_op_age, | |
14837 | [&](const ceph::timespan& lat) { | |
14838 | ostringstream ostr; | |
14839 | ostr << ", lat = " << timespan_str(lat) | |
14840 | << " cid =" << c->cid | |
14841 | << " oid =" << o->oid; | |
14842 | return ostr.str(); | |
14843 | } | |
14844 | ); | |
14845 | ||
7c673cae FG |
14846 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; |
14847 | return r; | |
14848 | } | |
14849 | ||
14850 | int BlueStore::_setattr(TransContext *txc, | |
14851 | CollectionRef& c, | |
14852 | OnodeRef& o, | |
14853 | const string& name, | |
14854 | bufferptr& val) | |
14855 | { | |
14856 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14857 | << " " << name << " (" << val.length() << " bytes)" | |
14858 | << dendl; | |
14859 | int r = 0; | |
3efd9988 FG |
14860 | if (val.is_partial()) { |
14861 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
14862 | val.length()); | |
f91f0fd5 | 14863 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
14864 | } else { |
14865 | auto& b = o->onode.attrs[name.c_str()] = val; | |
f91f0fd5 | 14866 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 14867 | } |
7c673cae FG |
14868 | txc->write_onode(o); |
14869 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14870 | << " " << name << " (" << val.length() << " bytes)" | |
14871 | << " = " << r << dendl; | |
14872 | return r; | |
14873 | } | |
14874 | ||
14875 | int BlueStore::_setattrs(TransContext *txc, | |
14876 | CollectionRef& c, | |
14877 | OnodeRef& o, | |
14878 | const map<string,bufferptr>& aset) | |
14879 | { | |
14880 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14881 | << " " << aset.size() << " keys" | |
14882 | << dendl; | |
14883 | int r = 0; | |
14884 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
14885 | p != aset.end(); ++p) { | |
3efd9988 FG |
14886 | if (p->second.is_partial()) { |
14887 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 14888 | bufferptr(p->second.c_str(), p->second.length()); |
f91f0fd5 | 14889 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
14890 | } else { |
14891 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
f91f0fd5 | 14892 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 14893 | } |
7c673cae FG |
14894 | } |
14895 | txc->write_onode(o); | |
14896 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14897 | << " " << aset.size() << " keys" | |
14898 | << " = " << r << dendl; | |
14899 | return r; | |
14900 | } | |
14901 | ||
14902 | ||
14903 | int BlueStore::_rmattr(TransContext *txc, | |
14904 | CollectionRef& c, | |
14905 | OnodeRef& o, | |
14906 | const string& name) | |
14907 | { | |
14908 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14909 | << " " << name << dendl; | |
14910 | int r = 0; | |
14911 | auto it = o->onode.attrs.find(name.c_str()); | |
14912 | if (it == o->onode.attrs.end()) | |
14913 | goto out; | |
14914 | ||
14915 | o->onode.attrs.erase(it); | |
14916 | txc->write_onode(o); | |
14917 | ||
14918 | out: | |
14919 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14920 | << " " << name << " = " << r << dendl; | |
14921 | return r; | |
14922 | } | |
14923 | ||
14924 | int BlueStore::_rmattrs(TransContext *txc, | |
14925 | CollectionRef& c, | |
14926 | OnodeRef& o) | |
14927 | { | |
14928 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14929 | int r = 0; | |
14930 | ||
14931 | if (o->onode.attrs.empty()) | |
14932 | goto out; | |
14933 | ||
14934 | o->onode.attrs.clear(); | |
14935 | txc->write_onode(o); | |
14936 | ||
14937 | out: | |
14938 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14939 | return r; | |
14940 | } | |
14941 | ||
9f95a23c | 14942 | void BlueStore::_do_omap_clear(TransContext *txc, OnodeRef& o) |
7c673cae | 14943 | { |
9f95a23c | 14944 | const string& omap_prefix = o->get_omap_prefix(); |
7c673cae | 14945 | string prefix, tail; |
9f95a23c TL |
14946 | o->get_omap_header(&prefix); |
14947 | o->get_omap_tail(&tail); | |
11fdf7f2 | 14948 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
494da23a | 14949 | txc->t->rmkey(omap_prefix, tail); |
11fdf7f2 TL |
14950 | dout(20) << __func__ << " remove range start: " |
14951 | << pretty_binary_string(prefix) << " end: " | |
14952 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
14953 | } |
14954 | ||
14955 | int BlueStore::_omap_clear(TransContext *txc, | |
14956 | CollectionRef& c, | |
14957 | OnodeRef& o) | |
14958 | { | |
14959 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14960 | int r = 0; | |
14961 | if (o->onode.has_omap()) { | |
14962 | o->flush(); | |
9f95a23c | 14963 | _do_omap_clear(txc, o); |
7c673cae FG |
14964 | o->onode.clear_omap_flag(); |
14965 | txc->write_onode(o); | |
14966 | } | |
14967 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14968 | return r; | |
14969 | } | |
14970 | ||
14971 | int BlueStore::_omap_setkeys(TransContext *txc, | |
14972 | CollectionRef& c, | |
14973 | OnodeRef& o, | |
14974 | bufferlist &bl) | |
14975 | { | |
14976 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14977 | int r; | |
11fdf7f2 | 14978 | auto p = bl.cbegin(); |
7c673cae FG |
14979 | __u32 num; |
14980 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14981 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14982 | o->onode.set_omap_flags_pgmeta(); |
14983 | } else { | |
14984 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14985 | } |
7c673cae | 14986 | txc->write_onode(o); |
494da23a | 14987 | |
9f95a23c | 14988 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14989 | string key_tail; |
14990 | bufferlist tail; | |
9f95a23c | 14991 | o->get_omap_tail(&key_tail); |
494da23a | 14992 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14993 | } else { |
14994 | txc->note_modified_object(o); | |
14995 | } | |
9f95a23c | 14996 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 14997 | string final_key; |
9f95a23c TL |
14998 | o->get_omap_key(string(), &final_key); |
14999 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 15000 | decode(num, p); |
7c673cae FG |
15001 | while (num--) { |
15002 | string key; | |
15003 | bufferlist value; | |
11fdf7f2 TL |
15004 | decode(key, p); |
15005 | decode(value, p); | |
9f95a23c | 15006 | final_key.resize(base_key_len); // keep prefix |
7c673cae | 15007 | final_key += key; |
11fdf7f2 | 15008 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 15009 | << " <- " << key << dendl; |
11fdf7f2 | 15010 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
15011 | } |
15012 | r = 0; | |
15013 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15014 | return r; | |
15015 | } | |
15016 | ||
15017 | int BlueStore::_omap_setheader(TransContext *txc, | |
15018 | CollectionRef& c, | |
15019 | OnodeRef &o, | |
15020 | bufferlist& bl) | |
15021 | { | |
15022 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15023 | int r; | |
15024 | string key; | |
15025 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 15026 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
15027 | o->onode.set_omap_flags_pgmeta(); |
15028 | } else { | |
15029 | o->onode.set_omap_flags(); | |
11fdf7f2 | 15030 | } |
7c673cae | 15031 | txc->write_onode(o); |
494da23a | 15032 | |
9f95a23c | 15033 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
15034 | string key_tail; |
15035 | bufferlist tail; | |
9f95a23c | 15036 | o->get_omap_tail(&key_tail); |
494da23a | 15037 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
15038 | } else { |
15039 | txc->note_modified_object(o); | |
15040 | } | |
9f95a23c TL |
15041 | const string& prefix = o->get_omap_prefix(); |
15042 | o->get_omap_header(&key); | |
11fdf7f2 | 15043 | txc->t->set(prefix, key, bl); |
7c673cae FG |
15044 | r = 0; |
15045 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15046 | return r; | |
15047 | } | |
15048 | ||
15049 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
15050 | CollectionRef& c, | |
15051 | OnodeRef& o, | |
15052 | bufferlist& bl) | |
15053 | { | |
15054 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15055 | int r = 0; | |
11fdf7f2 | 15056 | auto p = bl.cbegin(); |
7c673cae FG |
15057 | __u32 num; |
15058 | string final_key; | |
15059 | ||
15060 | if (!o->onode.has_omap()) { | |
15061 | goto out; | |
15062 | } | |
11fdf7f2 | 15063 | { |
9f95a23c TL |
15064 | const string& prefix = o->get_omap_prefix(); |
15065 | o->get_omap_key(string(), &final_key); | |
15066 | size_t base_key_len = final_key.size(); | |
11fdf7f2 TL |
15067 | decode(num, p); |
15068 | while (num--) { | |
15069 | string key; | |
15070 | decode(key, p); | |
9f95a23c | 15071 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
15072 | final_key += key; |
15073 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
15074 | << " <- " << key << dendl; | |
15075 | txc->t->rmkey(prefix, final_key); | |
15076 | } | |
7c673cae FG |
15077 | } |
15078 | txc->note_modified_object(o); | |
15079 | ||
15080 | out: | |
15081 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15082 | return r; | |
15083 | } | |
15084 | ||
15085 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
15086 | CollectionRef& c, | |
15087 | OnodeRef& o, | |
15088 | const string& first, const string& last) | |
15089 | { | |
15090 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
15091 | string key_first, key_last; |
15092 | int r = 0; | |
15093 | if (!o->onode.has_omap()) { | |
15094 | goto out; | |
15095 | } | |
11fdf7f2 | 15096 | { |
9f95a23c | 15097 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 15098 | o->flush(); |
9f95a23c TL |
15099 | o->get_omap_key(first, &key_first); |
15100 | o->get_omap_key(last, &key_last); | |
11fdf7f2 TL |
15101 | txc->t->rm_range_keys(prefix, key_first, key_last); |
15102 | dout(20) << __func__ << " remove range start: " | |
15103 | << pretty_binary_string(key_first) << " end: " | |
15104 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
15105 | } |
15106 | txc->note_modified_object(o); | |
15107 | ||
15108 | out: | |
15109 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15110 | return r; | |
15111 | } | |
15112 | ||
15113 | int BlueStore::_set_alloc_hint( | |
15114 | TransContext *txc, | |
15115 | CollectionRef& c, | |
15116 | OnodeRef& o, | |
15117 | uint64_t expected_object_size, | |
15118 | uint64_t expected_write_size, | |
15119 | uint32_t flags) | |
15120 | { | |
15121 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
15122 | << " object_size " << expected_object_size | |
15123 | << " write_size " << expected_write_size | |
15124 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
15125 | << dendl; | |
15126 | int r = 0; | |
15127 | o->onode.expected_object_size = expected_object_size; | |
15128 | o->onode.expected_write_size = expected_write_size; | |
15129 | o->onode.alloc_hint_flags = flags; | |
15130 | txc->write_onode(o); | |
15131 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
15132 | << " object_size " << expected_object_size | |
15133 | << " write_size " << expected_write_size | |
15134 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
15135 | << " = " << r << dendl; | |
15136 | return r; | |
15137 | } | |
15138 | ||
15139 | int BlueStore::_clone(TransContext *txc, | |
15140 | CollectionRef& c, | |
15141 | OnodeRef& oldo, | |
15142 | OnodeRef& newo) | |
15143 | { | |
15144 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15145 | << newo->oid << dendl; | |
15146 | int r = 0; | |
15147 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
15148 | derr << __func__ << " mismatched hash on " << oldo->oid | |
15149 | << " and " << newo->oid << dendl; | |
15150 | return -EINVAL; | |
15151 | } | |
15152 | ||
7c673cae FG |
15153 | _assign_nid(txc, newo); |
15154 | ||
15155 | // clone data | |
15156 | oldo->flush(); | |
15157 | _do_truncate(txc, c, newo, 0); | |
15158 | if (cct->_conf->bluestore_clone_cow) { | |
15159 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
15160 | } else { | |
15161 | bufferlist bl; | |
15162 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
15163 | if (r < 0) | |
15164 | goto out; | |
15165 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
15166 | if (r < 0) | |
15167 | goto out; | |
15168 | } | |
15169 | ||
15170 | // clone attrs | |
15171 | newo->onode.attrs = oldo->onode.attrs; | |
15172 | ||
15173 | // clone omap | |
15174 | if (newo->onode.has_omap()) { | |
15175 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
15176 | newo->flush(); | |
9f95a23c | 15177 | _do_omap_clear(txc, newo); |
494da23a | 15178 | newo->onode.clear_omap_flag(); |
7c673cae FG |
15179 | } |
15180 | if (oldo->onode.has_omap()) { | |
15181 | dout(20) << __func__ << " copying omap data" << dendl; | |
494da23a | 15182 | if (newo->oid.is_pgmeta()) { |
9f95a23c TL |
15183 | newo->onode.set_omap_flags_pgmeta(); |
15184 | } else { | |
15185 | newo->onode.set_omap_flags(); | |
7c673cae | 15186 | } |
9f95a23c | 15187 | const string& prefix = newo->get_omap_prefix(); |
11fdf7f2 | 15188 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 15189 | string head, tail; |
9f95a23c TL |
15190 | oldo->get_omap_header(&head); |
15191 | oldo->get_omap_tail(&tail); | |
7c673cae FG |
15192 | it->lower_bound(head); |
15193 | while (it->valid()) { | |
15194 | if (it->key() >= tail) { | |
15195 | dout(30) << __func__ << " reached tail" << dendl; | |
15196 | break; | |
15197 | } else { | |
15198 | dout(30) << __func__ << " got header/data " | |
15199 | << pretty_binary_string(it->key()) << dendl; | |
15200 | string key; | |
9f95a23c | 15201 | newo->rewrite_omap_key(it->key(), &key); |
11fdf7f2 | 15202 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
15203 | } |
15204 | it->next(); | |
15205 | } | |
494da23a TL |
15206 | string new_tail; |
15207 | bufferlist new_tail_value; | |
9f95a23c | 15208 | newo->get_omap_tail(&new_tail); |
494da23a | 15209 | txc->t->set(prefix, new_tail, new_tail_value); |
7c673cae FG |
15210 | } |
15211 | ||
15212 | txc->write_onode(newo); | |
15213 | r = 0; | |
15214 | ||
15215 | out: | |
15216 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15217 | << newo->oid << " = " << r << dendl; | |
15218 | return r; | |
15219 | } | |
15220 | ||
15221 | int BlueStore::_do_clone_range( | |
15222 | TransContext *txc, | |
15223 | CollectionRef& c, | |
15224 | OnodeRef& oldo, | |
15225 | OnodeRef& newo, | |
224ce89b WB |
15226 | uint64_t srcoff, |
15227 | uint64_t length, | |
15228 | uint64_t dstoff) | |
7c673cae FG |
15229 | { |
15230 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15231 | << newo->oid | |
15232 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
15233 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
15234 | oldo->extent_map.fault_range(db, srcoff, length); | |
15235 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
15236 | _dump_onode<30>(cct, *oldo); |
15237 | _dump_onode<30>(cct, *newo); | |
7c673cae | 15238 | |
11fdf7f2 | 15239 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
81eedcae TL |
15240 | _dump_onode<30>(cct, *oldo); |
15241 | _dump_onode<30>(cct, *newo); | |
7c673cae FG |
15242 | return 0; |
15243 | } | |
15244 | ||
15245 | int BlueStore::_clone_range(TransContext *txc, | |
15246 | CollectionRef& c, | |
15247 | OnodeRef& oldo, | |
15248 | OnodeRef& newo, | |
15249 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
15250 | { | |
15251 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15252 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
15253 | << " to offset 0x" << dstoff << std::dec << dendl; | |
15254 | int r = 0; | |
15255 | ||
35e4c445 FG |
15256 | if (srcoff + length >= OBJECT_MAX_SIZE || |
15257 | dstoff + length >= OBJECT_MAX_SIZE) { | |
15258 | r = -E2BIG; | |
15259 | goto out; | |
15260 | } | |
7c673cae FG |
15261 | if (srcoff + length > oldo->onode.size) { |
15262 | r = -EINVAL; | |
15263 | goto out; | |
15264 | } | |
15265 | ||
7c673cae FG |
15266 | _assign_nid(txc, newo); |
15267 | ||
15268 | if (length > 0) { | |
15269 | if (cct->_conf->bluestore_clone_cow) { | |
15270 | _do_zero(txc, c, newo, dstoff, length); | |
15271 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
15272 | } else { | |
15273 | bufferlist bl; | |
15274 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
15275 | if (r < 0) | |
15276 | goto out; | |
15277 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
15278 | if (r < 0) | |
15279 | goto out; | |
15280 | } | |
15281 | } | |
15282 | ||
15283 | txc->write_onode(newo); | |
15284 | r = 0; | |
15285 | ||
15286 | out: | |
15287 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15288 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
15289 | << " to offset 0x" << dstoff << std::dec | |
15290 | << " = " << r << dendl; | |
15291 | return r; | |
15292 | } | |
15293 | ||
15294 | int BlueStore::_rename(TransContext *txc, | |
15295 | CollectionRef& c, | |
15296 | OnodeRef& oldo, | |
15297 | OnodeRef& newo, | |
15298 | const ghobject_t& new_oid) | |
15299 | { | |
15300 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15301 | << new_oid << dendl; | |
15302 | int r; | |
15303 | ghobject_t old_oid = oldo->oid; | |
f91f0fd5 | 15304 | mempool::bluestore_cache_meta::string new_okey; |
7c673cae FG |
15305 | |
15306 | if (newo) { | |
15307 | if (newo->exists) { | |
15308 | r = -EEXIST; | |
15309 | goto out; | |
15310 | } | |
11fdf7f2 | 15311 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
15312 | } |
15313 | ||
15314 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
15315 | ||
15316 | // rewrite shards | |
15317 | { | |
15318 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
15319 | get_object_key(cct, new_oid, &new_okey); | |
15320 | string key; | |
15321 | for (auto &s : oldo->extent_map.shards) { | |
15322 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
15323 | [&](const string& final_key) { | |
15324 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
15325 | } | |
15326 | ); | |
15327 | s.dirty = true; | |
15328 | } | |
15329 | } | |
15330 | ||
15331 | newo = oldo; | |
15332 | txc->write_onode(newo); | |
15333 | ||
15334 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
15335 | // Onode in the old slot | |
15336 | c->onode_map.rename(oldo, old_oid, new_oid, new_okey); | |
15337 | r = 0; | |
15338 | ||
f64942e4 AA |
15339 | // hold a ref to new Onode in old name position, to ensure we don't drop |
15340 | // it from the cache before this txc commits (or else someone may come along | |
15341 | // and read newo's metadata via the old name). | |
15342 | txc->note_modified_object(oldo); | |
15343 | ||
7c673cae FG |
15344 | out: |
15345 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
15346 | << new_oid << " = " << r << dendl; | |
15347 | return r; | |
15348 | } | |
15349 | ||
15350 | // collections | |
15351 | ||
15352 | int BlueStore::_create_collection( | |
15353 | TransContext *txc, | |
15354 | const coll_t &cid, | |
15355 | unsigned bits, | |
15356 | CollectionRef *c) | |
15357 | { | |
15358 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
15359 | int r; | |
15360 | bufferlist bl; | |
15361 | ||
15362 | { | |
9f95a23c | 15363 | std::unique_lock l(coll_lock); |
7c673cae FG |
15364 | if (*c) { |
15365 | r = -EEXIST; | |
15366 | goto out; | |
15367 | } | |
11fdf7f2 TL |
15368 | auto p = new_coll_map.find(cid); |
15369 | ceph_assert(p != new_coll_map.end()); | |
15370 | *c = p->second; | |
7c673cae FG |
15371 | (*c)->cnode.bits = bits; |
15372 | coll_map[cid] = *c; | |
11fdf7f2 | 15373 | new_coll_map.erase(p); |
7c673cae | 15374 | } |
11fdf7f2 | 15375 | encode((*c)->cnode, bl); |
7c673cae FG |
15376 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
15377 | r = 0; | |
15378 | ||
15379 | out: | |
15380 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
15381 | return r; | |
15382 | } | |
15383 | ||
15384 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
15385 | CollectionRef *c) | |
15386 | { | |
15387 | dout(15) << __func__ << " " << cid << dendl; | |
15388 | int r; | |
15389 | ||
11fdf7f2 | 15390 | (*c)->flush_all_but_last(); |
7c673cae | 15391 | { |
9f95a23c | 15392 | std::unique_lock l(coll_lock); |
7c673cae FG |
15393 | if (!*c) { |
15394 | r = -ENOENT; | |
15395 | goto out; | |
15396 | } | |
15397 | size_t nonexistent_count = 0; | |
11fdf7f2 | 15398 | ceph_assert((*c)->exists); |
adb31ebb | 15399 | if ((*c)->onode_map.map_any([&](Onode* o) { |
f67539c2 TL |
15400 | if (o->exists) { |
15401 | dout(1) << __func__ << " " << o->oid << " " << o | |
15402 | << " exists in onode_map" << dendl; | |
7c673cae | 15403 | return true; |
f67539c2 TL |
15404 | } |
15405 | ++nonexistent_count; | |
15406 | return false; | |
15407 | })) { | |
7c673cae FG |
15408 | r = -ENOTEMPTY; |
15409 | goto out; | |
15410 | } | |
7c673cae FG |
15411 | vector<ghobject_t> ls; |
15412 | ghobject_t next; | |
15413 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
15414 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 15415 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae | 15416 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
f91f0fd5 | 15417 | nonexistent_count + 1, false, &ls, &next); |
7c673cae | 15418 | if (r >= 0) { |
11fdf7f2 TL |
15419 | // If true mean collecton has more objects than nonexistent_count, |
15420 | // so bypass check. | |
15421 | bool exists = (!next.is_max()); | |
7c673cae FG |
15422 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
15423 | dout(10) << __func__ << " oid " << *it << dendl; | |
15424 | auto onode = (*c)->onode_map.lookup(*it); | |
15425 | exists = !onode || onode->exists; | |
15426 | if (exists) { | |
494da23a | 15427 | dout(1) << __func__ << " " << *it |
f67539c2 TL |
15428 | << " exists in db, " |
15429 | << (!onode ? "not present in ram" : "present in ram") | |
15430 | << dendl; | |
7c673cae FG |
15431 | } |
15432 | } | |
15433 | if (!exists) { | |
f67539c2 | 15434 | _do_remove_collection(txc, c); |
7c673cae FG |
15435 | r = 0; |
15436 | } else { | |
15437 | dout(10) << __func__ << " " << cid | |
15438 | << " is non-empty" << dendl; | |
f67539c2 | 15439 | r = -ENOTEMPTY; |
7c673cae FG |
15440 | } |
15441 | } | |
15442 | } | |
f67539c2 | 15443 | out: |
7c673cae FG |
15444 | dout(10) << __func__ << " " << cid << " = " << r << dendl; |
15445 | return r; | |
15446 | } | |
15447 | ||
11fdf7f2 TL |
15448 | void BlueStore::_do_remove_collection(TransContext *txc, |
15449 | CollectionRef *c) | |
15450 | { | |
15451 | coll_map.erase((*c)->cid); | |
15452 | txc->removed_collections.push_back(*c); | |
15453 | (*c)->exists = false; | |
15454 | _osr_register_zombie((*c)->osr.get()); | |
15455 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
15456 | c->reset(); | |
15457 | } | |
15458 | ||
7c673cae FG |
15459 | int BlueStore::_split_collection(TransContext *txc, |
15460 | CollectionRef& c, | |
15461 | CollectionRef& d, | |
15462 | unsigned bits, int rem) | |
15463 | { | |
15464 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
15465 | << " bits " << bits << dendl; | |
9f95a23c TL |
15466 | std::unique_lock l(c->lock); |
15467 | std::unique_lock l2(d->lock); | |
7c673cae FG |
15468 | int r; |
15469 | ||
15470 | // flush all previous deferred writes on this sequencer. this is a bit | |
15471 | // heavyweight, but we need to make sure all deferred writes complete | |
15472 | // before we split as the new collection's sequencer may need to order | |
15473 | // this after those writes, and we don't bother with the complexity of | |
15474 | // moving those TransContexts over to the new osr. | |
15475 | _osr_drain_preceding(txc); | |
15476 | ||
15477 | // move any cached items (onodes and referenced shared blobs) that will | |
15478 | // belong to the child collection post-split. leave everything else behind. | |
15479 | // this may include things that don't strictly belong to the now-smaller | |
15480 | // parent split, but the OSD will always send us a split for every new | |
15481 | // child. | |
15482 | ||
15483 | spg_t pgid, dest_pgid; | |
15484 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 15485 | ceph_assert(is_pg); |
7c673cae | 15486 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 15487 | ceph_assert(is_pg); |
7c673cae FG |
15488 | |
15489 | // the destination should initially be empty. | |
11fdf7f2 TL |
15490 | ceph_assert(d->onode_map.empty()); |
15491 | ceph_assert(d->shared_blob_set.empty()); | |
15492 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
15493 | |
15494 | c->split_cache(d.get()); | |
15495 | ||
15496 | // adjust bits. note that this will be redundant for all but the first | |
15497 | // split call for this parent (first child). | |
15498 | c->cnode.bits = bits; | |
11fdf7f2 | 15499 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
15500 | r = 0; |
15501 | ||
15502 | bufferlist bl; | |
11fdf7f2 | 15503 | encode(c->cnode, bl); |
7c673cae FG |
15504 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
15505 | ||
15506 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
15507 | << " bits " << bits << " = " << r << dendl; | |
15508 | return r; | |
15509 | } | |
15510 | ||
11fdf7f2 TL |
15511 | int BlueStore::_merge_collection( |
15512 | TransContext *txc, | |
15513 | CollectionRef *c, | |
15514 | CollectionRef& d, | |
15515 | unsigned bits) | |
15516 | { | |
15517 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
15518 | << " bits " << bits << dendl; | |
9f95a23c TL |
15519 | std::unique_lock l((*c)->lock); |
15520 | std::unique_lock l2(d->lock); | |
11fdf7f2 TL |
15521 | int r; |
15522 | ||
15523 | coll_t cid = (*c)->cid; | |
15524 | ||
15525 | // flush all previous deferred writes on the source collection to ensure | |
15526 | // that all deferred writes complete before we merge as the target collection's | |
15527 | // sequencer may need to order new ops after those writes. | |
15528 | ||
15529 | _osr_drain((*c)->osr.get()); | |
15530 | ||
15531 | // move any cached items (onodes and referenced shared blobs) that will | |
15532 | // belong to the child collection post-split. leave everything else behind. | |
15533 | // this may include things that don't strictly belong to the now-smaller | |
15534 | // parent split, but the OSD will always send us a split for every new | |
15535 | // child. | |
15536 | ||
15537 | spg_t pgid, dest_pgid; | |
15538 | bool is_pg = cid.is_pg(&pgid); | |
15539 | ceph_assert(is_pg); | |
15540 | is_pg = d->cid.is_pg(&dest_pgid); | |
15541 | ceph_assert(is_pg); | |
15542 | ||
15543 | // adjust bits. note that this will be redundant for all but the first | |
15544 | // merge call for the parent/target. | |
15545 | d->cnode.bits = bits; | |
15546 | ||
15547 | // behavior depends on target (d) bits, so this after that is updated. | |
15548 | (*c)->split_cache(d.get()); | |
15549 | ||
15550 | // remove source collection | |
15551 | { | |
9f95a23c | 15552 | std::unique_lock l3(coll_lock); |
11fdf7f2 TL |
15553 | _do_remove_collection(txc, c); |
15554 | } | |
15555 | ||
15556 | r = 0; | |
15557 | ||
15558 | bufferlist bl; | |
15559 | encode(d->cnode, bl); | |
15560 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
15561 | ||
15562 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
15563 | << " bits " << bits << " = " << r << dendl; | |
15564 | return r; | |
15565 | } | |
15566 | ||
494da23a TL |
15567 | void BlueStore::log_latency( |
15568 | const char* name, | |
15569 | int idx, | |
15570 | const ceph::timespan& l, | |
15571 | double lat_threshold, | |
15572 | const char* info) const | |
15573 | { | |
15574 | logger->tinc(idx, l); | |
15575 | if (lat_threshold > 0.0 && | |
15576 | l >= make_timespan(lat_threshold)) { | |
15577 | dout(0) << __func__ << " slow operation observed for " << name | |
15578 | << ", latency = " << l | |
15579 | << info | |
15580 | << dendl; | |
15581 | } | |
15582 | } | |
15583 | ||
11fdf7f2 | 15584 | void BlueStore::log_latency_fn( |
494da23a | 15585 | const char* name, |
11fdf7f2 TL |
15586 | int idx, |
15587 | const ceph::timespan& l, | |
494da23a TL |
15588 | double lat_threshold, |
15589 | std::function<string (const ceph::timespan& lat)> fn) const | |
11fdf7f2 | 15590 | { |
494da23a TL |
15591 | logger->tinc(idx, l); |
15592 | if (lat_threshold > 0.0 && | |
15593 | l >= make_timespan(lat_threshold)) { | |
15594 | dout(0) << __func__ << " slow operation observed for " << name | |
15595 | << ", latency = " << l | |
15596 | << fn(l) | |
15597 | << dendl; | |
15598 | } | |
11fdf7f2 TL |
15599 | } |
15600 | ||
9f95a23c TL |
15601 | #if defined(WITH_LTTNG) |
15602 | void BlueStore::BlueStoreThrottle::emit_initial_tracepoint( | |
15603 | KeyValueDB &db, | |
15604 | TransContext &txc, | |
15605 | mono_clock::time_point start_throttle_acquire) | |
15606 | { | |
15607 | pending_kv_ios += txc.ios; | |
15608 | if (txc.deferred_txn) { | |
15609 | pending_deferred_ios += txc.ios; | |
15610 | } | |
15611 | ||
15612 | uint64_t started = 0; | |
15613 | uint64_t completed = 0; | |
15614 | if (should_trace(&started, &completed)) { | |
15615 | txc.tracing = true; | |
15616 | uint64_t rocksdb_base_level, | |
15617 | rocksdb_estimate_pending_compaction_bytes, | |
15618 | rocksdb_cur_size_all_mem_tables, | |
15619 | rocksdb_compaction_pending, | |
15620 | rocksdb_mem_table_flush_pending, | |
15621 | rocksdb_num_running_compactions, | |
15622 | rocksdb_num_running_flushes, | |
15623 | rocksdb_actual_delayed_write_rate; | |
15624 | db.get_property( | |
15625 | "rocksdb.base-level", | |
15626 | &rocksdb_base_level); | |
15627 | db.get_property( | |
15628 | "rocksdb.estimate-pending-compaction-bytes", | |
15629 | &rocksdb_estimate_pending_compaction_bytes); | |
15630 | db.get_property( | |
15631 | "rocksdb.cur-size-all-mem-tables", | |
15632 | &rocksdb_cur_size_all_mem_tables); | |
15633 | db.get_property( | |
15634 | "rocksdb.compaction-pending", | |
15635 | &rocksdb_compaction_pending); | |
15636 | db.get_property( | |
15637 | "rocksdb.mem-table-flush-pending", | |
15638 | &rocksdb_mem_table_flush_pending); | |
15639 | db.get_property( | |
15640 | "rocksdb.num-running-compactions", | |
15641 | &rocksdb_num_running_compactions); | |
15642 | db.get_property( | |
15643 | "rocksdb.num-running-flushes", | |
15644 | &rocksdb_num_running_flushes); | |
15645 | db.get_property( | |
15646 | "rocksdb.actual-delayed-write-rate", | |
15647 | &rocksdb_actual_delayed_write_rate); | |
15648 | ||
15649 | ||
15650 | tracepoint( | |
15651 | bluestore, | |
15652 | transaction_initial_state, | |
15653 | txc.osr->get_sequencer_id(), | |
15654 | txc.seq, | |
15655 | throttle_bytes.get_current(), | |
15656 | throttle_deferred_bytes.get_current(), | |
15657 | pending_kv_ios, | |
15658 | pending_deferred_ios, | |
15659 | started, | |
15660 | completed, | |
15661 | ceph::to_seconds<double>(mono_clock::now() - start_throttle_acquire)); | |
15662 | ||
15663 | tracepoint( | |
15664 | bluestore, | |
15665 | transaction_initial_state_rocksdb, | |
15666 | txc.osr->get_sequencer_id(), | |
15667 | txc.seq, | |
15668 | rocksdb_base_level, | |
15669 | rocksdb_estimate_pending_compaction_bytes, | |
15670 | rocksdb_cur_size_all_mem_tables, | |
15671 | rocksdb_compaction_pending, | |
15672 | rocksdb_mem_table_flush_pending, | |
15673 | rocksdb_num_running_compactions, | |
15674 | rocksdb_num_running_flushes, | |
15675 | rocksdb_actual_delayed_write_rate); | |
15676 | } | |
15677 | } | |
15678 | #endif | |
15679 | ||
15680 | mono_clock::duration BlueStore::BlueStoreThrottle::log_state_latency( | |
15681 | TransContext &txc, PerfCounters *logger, int state) | |
15682 | { | |
15683 | mono_clock::time_point now = mono_clock::now(); | |
15684 | mono_clock::duration lat = now - txc.last_stamp; | |
15685 | logger->tinc(state, lat); | |
15686 | #if defined(WITH_LTTNG) | |
15687 | if (txc.tracing && | |
15688 | state >= l_bluestore_state_prepare_lat && | |
15689 | state <= l_bluestore_state_done_lat) { | |
15690 | OID_ELAPSED("", lat.to_nsec() / 1000.0, txc.get_state_latency_name(state)); | |
15691 | tracepoint( | |
15692 | bluestore, | |
15693 | transaction_state_duration, | |
15694 | txc.osr->get_sequencer_id(), | |
15695 | txc.seq, | |
15696 | state, | |
15697 | ceph::to_seconds<double>(lat)); | |
15698 | } | |
15699 | #endif | |
15700 | txc.last_stamp = now; | |
15701 | return lat; | |
15702 | } | |
15703 | ||
15704 | bool BlueStore::BlueStoreThrottle::try_start_transaction( | |
15705 | KeyValueDB &db, | |
15706 | TransContext &txc, | |
15707 | mono_clock::time_point start_throttle_acquire) | |
15708 | { | |
15709 | throttle_bytes.get(txc.cost); | |
15710 | ||
15711 | if (!txc.deferred_txn || throttle_deferred_bytes.get_or_fail(txc.cost)) { | |
15712 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15713 | return true; | |
15714 | } else { | |
15715 | return false; | |
15716 | } | |
15717 | } | |
15718 | ||
15719 | void BlueStore::BlueStoreThrottle::finish_start_transaction( | |
15720 | KeyValueDB &db, | |
15721 | TransContext &txc, | |
15722 | mono_clock::time_point start_throttle_acquire) | |
15723 | { | |
15724 | ceph_assert(txc.deferred_txn); | |
15725 | throttle_deferred_bytes.get(txc.cost); | |
15726 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15727 | } | |
15728 | ||
15729 | #if defined(WITH_LTTNG) | |
15730 | void BlueStore::BlueStoreThrottle::complete_kv(TransContext &txc) | |
15731 | { | |
15732 | pending_kv_ios -= 1; | |
15733 | ios_completed_since_last_traced++; | |
15734 | if (txc.tracing) { | |
15735 | tracepoint( | |
15736 | bluestore, | |
15737 | transaction_commit_latency, | |
15738 | txc.osr->get_sequencer_id(), | |
15739 | txc.seq, | |
15740 | ceph::to_seconds<double>(mono_clock::now() - txc.start)); | |
15741 | } | |
15742 | } | |
15743 | #endif | |
15744 | ||
15745 | #if defined(WITH_LTTNG) | |
15746 | void BlueStore::BlueStoreThrottle::complete(TransContext &txc) | |
15747 | { | |
15748 | if (txc.deferred_txn) { | |
15749 | pending_deferred_ios -= 1; | |
15750 | } | |
15751 | if (txc.tracing) { | |
15752 | mono_clock::time_point now = mono_clock::now(); | |
15753 | mono_clock::duration lat = now - txc.start; | |
15754 | tracepoint( | |
15755 | bluestore, | |
15756 | transaction_total_duration, | |
15757 | txc.osr->get_sequencer_id(), | |
15758 | txc.seq, | |
15759 | ceph::to_seconds<double>(lat)); | |
15760 | } | |
15761 | } | |
15762 | #endif | |
11fdf7f2 | 15763 | |
7c673cae FG |
15764 | // DB key value Histogram |
15765 | #define KEY_SLAB 32 | |
15766 | #define VALUE_SLAB 64 | |
15767 | ||
15768 | const string prefix_onode = "o"; | |
15769 | const string prefix_onode_shard = "x"; | |
15770 | const string prefix_other = "Z"; | |
15771 | ||
15772 | int BlueStore::DBHistogram::get_key_slab(size_t sz) | |
15773 | { | |
15774 | return (sz/KEY_SLAB); | |
15775 | } | |
15776 | ||
15777 | string BlueStore::DBHistogram::get_key_slab_to_range(int slab) | |
15778 | { | |
15779 | int lower_bound = slab * KEY_SLAB; | |
15780 | int upper_bound = (slab + 1) * KEY_SLAB; | |
15781 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15782 | return ret; | |
15783 | } | |
15784 | ||
15785 | int BlueStore::DBHistogram::get_value_slab(size_t sz) | |
15786 | { | |
15787 | return (sz/VALUE_SLAB); | |
15788 | } | |
15789 | ||
15790 | string BlueStore::DBHistogram::get_value_slab_to_range(int slab) | |
15791 | { | |
15792 | int lower_bound = slab * VALUE_SLAB; | |
15793 | int upper_bound = (slab + 1) * VALUE_SLAB; | |
15794 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15795 | return ret; | |
15796 | } | |
15797 | ||
15798 | void BlueStore::DBHistogram::update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
15799 | const string &prefix, size_t key_size, size_t value_size) | |
15800 | { | |
15801 | uint32_t key_slab = get_key_slab(key_size); | |
15802 | uint32_t value_slab = get_value_slab(value_size); | |
15803 | key_hist[prefix][key_slab].count++; | |
11fdf7f2 TL |
15804 | key_hist[prefix][key_slab].max_len = |
15805 | std::max<size_t>(key_size, key_hist[prefix][key_slab].max_len); | |
7c673cae FG |
15806 | key_hist[prefix][key_slab].val_map[value_slab].count++; |
15807 | key_hist[prefix][key_slab].val_map[value_slab].max_len = | |
11fdf7f2 TL |
15808 | std::max<size_t>(value_size, |
15809 | key_hist[prefix][key_slab].val_map[value_slab].max_len); | |
7c673cae FG |
15810 | } |
15811 | ||
15812 | void BlueStore::DBHistogram::dump(Formatter *f) | |
15813 | { | |
15814 | f->open_object_section("rocksdb_value_distribution"); | |
15815 | for (auto i : value_hist) { | |
15816 | f->dump_unsigned(get_value_slab_to_range(i.first).data(), i.second); | |
15817 | } | |
15818 | f->close_section(); | |
15819 | ||
15820 | f->open_object_section("rocksdb_key_value_histogram"); | |
15821 | for (auto i : key_hist) { | |
15822 | f->dump_string("prefix", i.first); | |
15823 | f->open_object_section("key_hist"); | |
15824 | for ( auto k : i.second) { | |
15825 | f->dump_unsigned(get_key_slab_to_range(k.first).data(), k.second.count); | |
15826 | f->dump_unsigned("max_len", k.second.max_len); | |
15827 | f->open_object_section("value_hist"); | |
15828 | for ( auto j : k.second.val_map) { | |
15829 | f->dump_unsigned(get_value_slab_to_range(j.first).data(), j.second.count); | |
15830 | f->dump_unsigned("max_len", j.second.max_len); | |
15831 | } | |
15832 | f->close_section(); | |
15833 | } | |
15834 | f->close_section(); | |
15835 | } | |
15836 | f->close_section(); | |
15837 | } | |
15838 | ||
15839 | //Itrerates through the db and collects the stats | |
15840 | void BlueStore::generate_db_histogram(Formatter *f) | |
15841 | { | |
15842 | //globals | |
15843 | uint64_t num_onodes = 0; | |
15844 | uint64_t num_shards = 0; | |
15845 | uint64_t num_super = 0; | |
15846 | uint64_t num_coll = 0; | |
15847 | uint64_t num_omap = 0; | |
11fdf7f2 | 15848 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
15849 | uint64_t num_deferred = 0; |
15850 | uint64_t num_alloc = 0; | |
15851 | uint64_t num_stat = 0; | |
15852 | uint64_t num_others = 0; | |
15853 | uint64_t num_shared_shards = 0; | |
15854 | size_t max_key_size =0, max_value_size = 0; | |
15855 | uint64_t total_key_size = 0, total_value_size = 0; | |
15856 | size_t key_size = 0, value_size = 0; | |
15857 | DBHistogram hist; | |
15858 | ||
11fdf7f2 | 15859 | auto start = coarse_mono_clock::now(); |
7c673cae | 15860 | |
11fdf7f2 | 15861 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
15862 | iter->seek_to_first(); |
15863 | while (iter->valid()) { | |
15864 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
15865 | key_size = iter->key_size(); | |
15866 | value_size = iter->value_size(); | |
15867 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
15868 | max_key_size = std::max(max_key_size, key_size); |
15869 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
15870 | total_key_size += key_size; |
15871 | total_value_size += value_size; | |
15872 | ||
15873 | pair<string,string> key(iter->raw_key()); | |
15874 | ||
15875 | if (key.first == PREFIX_SUPER) { | |
15876 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
15877 | num_super++; | |
15878 | } else if (key.first == PREFIX_STAT) { | |
15879 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
15880 | num_stat++; | |
15881 | } else if (key.first == PREFIX_COLL) { | |
15882 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
15883 | num_coll++; | |
15884 | } else if (key.first == PREFIX_OBJ) { | |
15885 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
15886 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
15887 | num_onodes++; | |
15888 | } else { | |
15889 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
15890 | num_shards++; | |
15891 | } | |
15892 | } else if (key.first == PREFIX_OMAP) { | |
15893 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
15894 | num_omap++; | |
f67539c2 TL |
15895 | } else if (key.first == PREFIX_PERPOOL_OMAP) { |
15896 | hist.update_hist_entry(hist.key_hist, PREFIX_PERPOOL_OMAP, key_size, value_size); | |
15897 | num_omap++; | |
15898 | } else if (key.first == PREFIX_PERPG_OMAP) { | |
15899 | hist.update_hist_entry(hist.key_hist, PREFIX_PERPG_OMAP, key_size, value_size); | |
15900 | num_omap++; | |
11fdf7f2 TL |
15901 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
15902 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
15903 | num_pgmeta_omap++; | |
7c673cae FG |
15904 | } else if (key.first == PREFIX_DEFERRED) { |
15905 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
15906 | num_deferred++; | |
11fdf7f2 | 15907 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
15908 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
15909 | num_alloc++; | |
15910 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
15911 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
15912 | num_shared_shards++; | |
15913 | } else { | |
15914 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
15915 | num_others++; | |
15916 | } | |
15917 | iter->next(); | |
15918 | } | |
15919 | ||
11fdf7f2 | 15920 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
15921 | f->open_object_section("rocksdb_key_value_stats"); |
15922 | f->dump_unsigned("num_onodes", num_onodes); | |
15923 | f->dump_unsigned("num_shards", num_shards); | |
15924 | f->dump_unsigned("num_super", num_super); | |
15925 | f->dump_unsigned("num_coll", num_coll); | |
15926 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 15927 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
15928 | f->dump_unsigned("num_deferred", num_deferred); |
15929 | f->dump_unsigned("num_alloc", num_alloc); | |
15930 | f->dump_unsigned("num_stat", num_stat); | |
15931 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
15932 | f->dump_unsigned("num_others", num_others); | |
15933 | f->dump_unsigned("max_key_size", max_key_size); | |
15934 | f->dump_unsigned("max_value_size", max_value_size); | |
15935 | f->dump_unsigned("total_key_size", total_key_size); | |
15936 | f->dump_unsigned("total_value_size", total_value_size); | |
15937 | f->close_section(); | |
15938 | ||
15939 | hist.dump(f); | |
15940 | ||
15941 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
15942 | ||
15943 | } | |
15944 | ||
f6b5b4d7 | 15945 | void BlueStore::_shutdown_cache() |
7c673cae FG |
15946 | { |
15947 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15948 | for (auto i : buffer_cache_shards) { |
15949 | i->flush(); | |
11fdf7f2 | 15950 | ceph_assert(i->empty()); |
7c673cae FG |
15951 | } |
15952 | for (auto& p : coll_map) { | |
f6b5b4d7 | 15953 | p.second->onode_map.clear(); |
3efd9988 FG |
15954 | if (!p.second->shared_blob_set.empty()) { |
15955 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 15956 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 15957 | } |
11fdf7f2 TL |
15958 | ceph_assert(p.second->onode_map.empty()); |
15959 | ceph_assert(p.second->shared_blob_set.empty()); | |
7c673cae FG |
15960 | } |
15961 | coll_map.clear(); | |
f6b5b4d7 TL |
15962 | for (auto i : onode_cache_shards) { |
15963 | ceph_assert(i->empty()); | |
15964 | } | |
7c673cae FG |
15965 | } |
15966 | ||
31f18b77 FG |
15967 | // For external caller. |
15968 | // We use a best-effort policy instead, e.g., | |
15969 | // we don't care if there are still some pinned onodes/data in the cache | |
15970 | // after this command is completed. | |
11fdf7f2 | 15971 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
15972 | { |
15973 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15974 | for (auto i : onode_cache_shards) { |
15975 | i->flush(); | |
15976 | } | |
15977 | for (auto i : buffer_cache_shards) { | |
15978 | i->flush(); | |
31f18b77 | 15979 | } |
11fdf7f2 TL |
15980 | |
15981 | return 0; | |
31f18b77 FG |
15982 | } |
15983 | ||
7c673cae FG |
15984 | void BlueStore::_apply_padding(uint64_t head_pad, |
15985 | uint64_t tail_pad, | |
7c673cae FG |
15986 | bufferlist& padded) |
15987 | { | |
7c673cae | 15988 | if (head_pad) { |
224ce89b | 15989 | padded.prepend_zero(head_pad); |
7c673cae FG |
15990 | } |
15991 | if (tail_pad) { | |
15992 | padded.append_zero(tail_pad); | |
15993 | } | |
15994 | if (head_pad || tail_pad) { | |
15995 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
15996 | << " tail 0x" << tail_pad << std::dec << dendl; | |
15997 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
15998 | } | |
15999 | } | |
16000 | ||
11fdf7f2 TL |
16001 | void BlueStore::_record_onode(OnodeRef &o, KeyValueDB::Transaction &txn) |
16002 | { | |
16003 | // finalize extent_map shards | |
16004 | o->extent_map.update(txn, false); | |
16005 | if (o->extent_map.needs_reshard()) { | |
16006 | o->extent_map.reshard(db, txn); | |
16007 | o->extent_map.update(txn, true); | |
16008 | if (o->extent_map.needs_reshard()) { | |
16009 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
16010 | << dendl; | |
16011 | o->extent_map.clear_needs_reshard(); | |
16012 | } | |
16013 | logger->inc(l_bluestore_onode_reshard); | |
16014 | } | |
16015 | ||
16016 | // bound encode | |
16017 | size_t bound = 0; | |
16018 | denc(o->onode, bound); | |
16019 | o->extent_map.bound_encode_spanning_blobs(bound); | |
16020 | if (o->onode.extent_map_shards.empty()) { | |
16021 | denc(o->extent_map.inline_bl, bound); | |
16022 | } | |
16023 | ||
16024 | // encode | |
16025 | bufferlist bl; | |
16026 | unsigned onode_part, blob_part, extent_part; | |
16027 | { | |
16028 | auto p = bl.get_contiguous_appender(bound, true); | |
16029 | denc(o->onode, p); | |
16030 | onode_part = p.get_logical_offset(); | |
16031 | o->extent_map.encode_spanning_blobs(p); | |
16032 | blob_part = p.get_logical_offset() - onode_part; | |
16033 | if (o->onode.extent_map_shards.empty()) { | |
16034 | denc(o->extent_map.inline_bl, p); | |
16035 | } | |
16036 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
16037 | } | |
16038 | ||
16039 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
16040 | << " (" << onode_part << " bytes onode + " | |
16041 | << blob_part << " bytes spanning blobs + " | |
16042 | << extent_part << " bytes inline extents)" | |
16043 | << dendl; | |
16044 | ||
16045 | ||
16046 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
16047 | } | |
16048 | ||
16049 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
16050 | { | |
16051 | std::lock_guard l(qlock); | |
16052 | ||
f67539c2 TL |
16053 | if (!spurious_read_errors_alert.empty()) { |
16054 | alerts.emplace( | |
16055 | "BLUESTORE_SPURIOUS_READ_ERRORS", | |
16056 | spurious_read_errors_alert); | |
16057 | } | |
81eedcae TL |
16058 | if (!disk_size_mismatch_alert.empty()) { |
16059 | alerts.emplace( | |
16060 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
16061 | disk_size_mismatch_alert); | |
16062 | } | |
16063 | if (!legacy_statfs_alert.empty()) { | |
16064 | alerts.emplace( | |
16065 | "BLUESTORE_LEGACY_STATFS", | |
16066 | legacy_statfs_alert); | |
16067 | } | |
11fdf7f2 TL |
16068 | if (!spillover_alert.empty() && |
16069 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
16070 | alerts.emplace( | |
16071 | "BLUEFS_SPILLOVER", | |
16072 | spillover_alert); | |
16073 | } | |
f67539c2 TL |
16074 | if (!no_per_pg_omap_alert.empty()) { |
16075 | alerts.emplace( | |
16076 | "BLUESTORE_NO_PER_PG_OMAP", | |
16077 | no_per_pg_omap_alert); | |
16078 | } | |
9f95a23c TL |
16079 | if (!no_per_pool_omap_alert.empty()) { |
16080 | alerts.emplace( | |
16081 | "BLUESTORE_NO_PER_POOL_OMAP", | |
16082 | no_per_pool_omap_alert); | |
16083 | } | |
11fdf7f2 TL |
16084 | string s0(failed_cmode); |
16085 | ||
16086 | if (!failed_compressors.empty()) { | |
16087 | if (!s0.empty()) { | |
16088 | s0 += ", "; | |
16089 | } | |
16090 | s0 += "unable to load:"; | |
16091 | bool first = true; | |
16092 | for (auto& s : failed_compressors) { | |
16093 | if (first) { | |
16094 | first = false; | |
16095 | } else { | |
16096 | s0 += ", "; | |
16097 | } | |
16098 | s0 += s; | |
16099 | } | |
16100 | alerts.emplace( | |
16101 | "BLUESTORE_NO_COMPRESSION", | |
16102 | s0); | |
16103 | } | |
16104 | } | |
16105 | ||
9f95a23c TL |
16106 | void BlueStore::_collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
16107 | size_t extents) | |
16108 | { | |
16109 | alloc_stats_count++; | |
16110 | alloc_stats_fragments += extents; | |
16111 | alloc_stats_size += need; | |
16112 | } | |
16113 | ||
16114 | void BlueStore::_record_allocation_stats() | |
16115 | { | |
16116 | // don't care about data consistency, | |
16117 | // fields can be partially modified while making the tuple | |
16118 | auto t0 = std::make_tuple( | |
16119 | alloc_stats_count.exchange(0), | |
16120 | alloc_stats_fragments.exchange(0), | |
16121 | alloc_stats_size.exchange(0)); | |
16122 | ||
16123 | dout(0) << " allocation stats probe " | |
16124 | << probe_count << ":" | |
16125 | << " cnt: " << std::get<0>(t0) | |
16126 | << " frags: " << std::get<1>(t0) | |
16127 | << " size: " << std::get<2>(t0) | |
16128 | << dendl; | |
16129 | ||
16130 | ||
16131 | // | |
16132 | // Keep the history for probes from the power-of-two sequence: | |
16133 | // -1, -2, -4, -8, -16 | |
16134 | // | |
16135 | size_t base = 1; | |
16136 | for (auto& t : alloc_stats_history) { | |
16137 | dout(0) << " probe -" | |
16138 | << base + (probe_count % base) << ": " | |
16139 | << std::get<0>(t) | |
16140 | << ", " << std::get<1>(t) | |
16141 | << ", " << std::get<2>(t) | |
16142 | << dendl; | |
16143 | base <<= 1; | |
16144 | } | |
16145 | dout(0) << "------------" << dendl; | |
16146 | ||
f67539c2 | 16147 | ++ probe_count; |
9f95a23c | 16148 | |
f67539c2 TL |
16149 | for (ssize_t i = alloc_stats_history.size() - 1 ; i > 0 ; --i) { |
16150 | if ((probe_count % (1 << i)) == 0) { | |
16151 | alloc_stats_history[i] = alloc_stats_history[i - 1]; | |
16152 | } | |
9f95a23c TL |
16153 | } |
16154 | alloc_stats_history[0].swap(t0); | |
16155 | } | |
16156 | ||
7c673cae | 16157 | // =========================================== |
11fdf7f2 TL |
16158 | // BlueStoreRepairer |
16159 | ||
16160 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
16161 | const interval_set<uint64_t>& extents) | |
16162 | { | |
16163 | ceph_assert(granularity); // initialized | |
16164 | // can't call for the second time | |
16165 | ceph_assert(!was_filtered_out); | |
16166 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
16167 | ||
16168 | uint64_t prev_pos = 0; | |
16169 | uint64_t npos = collections_bfs.size(); | |
16170 | ||
16171 | bloom_vector collections_reduced; | |
16172 | bloom_vector objects_reduced; | |
16173 | ||
16174 | for (auto e : extents) { | |
16175 | if (e.second == 0) { | |
16176 | continue; | |
16177 | } | |
16178 | uint64_t pos = max(e.first / granularity, prev_pos); | |
16179 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
16180 | while (pos != npos && pos < end_pos) { | |
16181 | ceph_assert( collections_bfs[pos].element_count() == | |
16182 | objects_bfs[pos].element_count()); | |
16183 | if (collections_bfs[pos].element_count()) { | |
16184 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
16185 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
16186 | } | |
16187 | ++pos; | |
16188 | } | |
16189 | prev_pos = end_pos; | |
16190 | } | |
16191 | collections_reduced.swap(collections_bfs); | |
16192 | objects_reduced.swap(objects_bfs); | |
16193 | was_filtered_out = true; | |
16194 | return collections_bfs.size(); | |
16195 | } | |
16196 | ||
16197 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
16198 | const string& prefix, | |
16199 | const string& key) | |
16200 | { | |
b3b6e05e | 16201 | std::lock_guard l(lock); |
11fdf7f2 TL |
16202 | if (!remove_key_txn) { |
16203 | remove_key_txn = db->get_transaction(); | |
16204 | } | |
16205 | ++to_repair_cnt; | |
16206 | remove_key_txn->rmkey(prefix, key); | |
16207 | ||
16208 | return true; | |
16209 | } | |
16210 | ||
f67539c2 | 16211 | void BlueStoreRepairer::fix_per_pool_omap(KeyValueDB *db, int val) |
9f95a23c | 16212 | { |
b3b6e05e TL |
16213 | std::lock_guard l(lock); // possibly redundant |
16214 | ceph_assert(fix_per_pool_omap_txn == nullptr); | |
9f95a23c TL |
16215 | fix_per_pool_omap_txn = db->get_transaction(); |
16216 | ++to_repair_cnt; | |
16217 | bufferlist bl; | |
f67539c2 | 16218 | bl.append(stringify(val)); |
9f95a23c TL |
16219 | fix_per_pool_omap_txn->set(PREFIX_SUPER, "per_pool_omap", bl); |
16220 | } | |
16221 | ||
11fdf7f2 TL |
16222 | bool BlueStoreRepairer::fix_shared_blob( |
16223 | KeyValueDB *db, | |
16224 | uint64_t sbid, | |
16225 | const bufferlist* bl) | |
16226 | { | |
b3b6e05e | 16227 | std::lock_guard l(lock); // possibly redundant |
11fdf7f2 TL |
16228 | KeyValueDB::Transaction txn; |
16229 | if (fix_misreferences_txn) { // reuse this txn | |
16230 | txn = fix_misreferences_txn; | |
16231 | } else { | |
16232 | if (!fix_shared_blob_txn) { | |
16233 | fix_shared_blob_txn = db->get_transaction(); | |
16234 | } | |
16235 | txn = fix_shared_blob_txn; | |
16236 | } | |
16237 | string key; | |
16238 | get_shared_blob_key(sbid, &key); | |
16239 | ||
16240 | ++to_repair_cnt; | |
16241 | if (bl) { | |
16242 | txn->set(PREFIX_SHARED_BLOB, key, *bl); | |
16243 | } else { | |
16244 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
16245 | } | |
16246 | return true; | |
16247 | } | |
16248 | ||
16249 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
16250 | const string& key, | |
16251 | const store_statfs_t& new_statfs) | |
16252 | { | |
b3b6e05e | 16253 | std::lock_guard l(lock); |
11fdf7f2 TL |
16254 | if (!fix_statfs_txn) { |
16255 | fix_statfs_txn = db->get_transaction(); | |
16256 | } | |
16257 | BlueStore::volatile_statfs vstatfs; | |
16258 | vstatfs = new_statfs; | |
16259 | bufferlist bl; | |
16260 | vstatfs.encode(bl); | |
16261 | ++to_repair_cnt; | |
16262 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
16263 | return true; | |
16264 | } | |
16265 | ||
16266 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
16267 | FreelistManager* fm, | |
16268 | uint64_t offset, uint64_t len) | |
16269 | { | |
b3b6e05e | 16270 | std::lock_guard l(lock); |
11fdf7f2 TL |
16271 | if (!fix_fm_leaked_txn) { |
16272 | fix_fm_leaked_txn = db->get_transaction(); | |
16273 | } | |
16274 | ++to_repair_cnt; | |
16275 | fm->release(offset, len, fix_fm_leaked_txn); | |
16276 | return true; | |
16277 | } | |
16278 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
16279 | FreelistManager* fm, | |
16280 | uint64_t offset, uint64_t len) | |
16281 | { | |
b3b6e05e | 16282 | std::lock_guard l(lock); |
11fdf7f2 TL |
16283 | if (!fix_fm_false_free_txn) { |
16284 | fix_fm_false_free_txn = db->get_transaction(); | |
16285 | } | |
16286 | ++to_repair_cnt; | |
16287 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
16288 | return true; | |
16289 | } | |
16290 | ||
b3b6e05e TL |
16291 | bool BlueStoreRepairer::fix_spanning_blobs( |
16292 | KeyValueDB* db, | |
16293 | std::function<void(KeyValueDB::Transaction)> f) | |
adb31ebb | 16294 | { |
b3b6e05e | 16295 | std::lock_guard l(lock); |
adb31ebb TL |
16296 | if (!fix_onode_txn) { |
16297 | fix_onode_txn = db->get_transaction(); | |
16298 | } | |
b3b6e05e | 16299 | f(fix_onode_txn); |
adb31ebb | 16300 | ++to_repair_cnt; |
b3b6e05e | 16301 | return true; |
adb31ebb TL |
16302 | } |
16303 | ||
11fdf7f2 TL |
16304 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) |
16305 | { | |
b3b6e05e | 16306 | //NB: not for use in multithreading mode!!! |
11fdf7f2 TL |
16307 | if (misreferenced_extents.size()) { |
16308 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
16309 | ceph_assert(n > 0); | |
16310 | if (!fix_misreferences_txn) { | |
16311 | fix_misreferences_txn = db->get_transaction(); | |
16312 | } | |
16313 | return true; | |
16314 | } | |
16315 | return false; | |
16316 | } | |
16317 | ||
16318 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
16319 | { | |
b3b6e05e | 16320 | //NB: not for use in multithreading mode!!! |
9f95a23c TL |
16321 | if (fix_per_pool_omap_txn) { |
16322 | db->submit_transaction_sync(fix_per_pool_omap_txn); | |
16323 | fix_per_pool_omap_txn = nullptr; | |
16324 | } | |
11fdf7f2 TL |
16325 | if (fix_fm_leaked_txn) { |
16326 | db->submit_transaction_sync(fix_fm_leaked_txn); | |
16327 | fix_fm_leaked_txn = nullptr; | |
16328 | } | |
16329 | if (fix_fm_false_free_txn) { | |
16330 | db->submit_transaction_sync(fix_fm_false_free_txn); | |
16331 | fix_fm_false_free_txn = nullptr; | |
16332 | } | |
16333 | if (remove_key_txn) { | |
16334 | db->submit_transaction_sync(remove_key_txn); | |
16335 | remove_key_txn = nullptr; | |
16336 | } | |
16337 | if (fix_misreferences_txn) { | |
16338 | db->submit_transaction_sync(fix_misreferences_txn); | |
16339 | fix_misreferences_txn = nullptr; | |
16340 | } | |
adb31ebb TL |
16341 | if (fix_onode_txn) { |
16342 | db->submit_transaction_sync(fix_onode_txn); | |
16343 | fix_onode_txn = nullptr; | |
16344 | } | |
11fdf7f2 TL |
16345 | if (fix_shared_blob_txn) { |
16346 | db->submit_transaction_sync(fix_shared_blob_txn); | |
16347 | fix_shared_blob_txn = nullptr; | |
16348 | } | |
16349 | ||
16350 | if (fix_statfs_txn) { | |
16351 | db->submit_transaction_sync(fix_statfs_txn); | |
16352 | fix_statfs_txn = nullptr; | |
16353 | } | |
16354 | unsigned repaired = to_repair_cnt; | |
16355 | to_repair_cnt = 0; | |
16356 | return repaired; | |
16357 | } | |
16358 | ||
16359 | // ======================================================= | |
9f95a23c TL |
16360 | // RocksDBBlueFSVolumeSelector |
16361 | ||
16362 | uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) { | |
16363 | ceph_assert(h != nullptr); | |
16364 | uint64_t hint = reinterpret_cast<uint64_t>(h); | |
16365 | uint8_t res; | |
16366 | switch (hint) { | |
16367 | case LEVEL_SLOW: | |
16368 | res = BlueFS::BDEV_SLOW; | |
16369 | if (db_avail4slow > 0) { | |
16370 | // considering statically available db space vs. | |
16371 | // - observed maximums on DB dev for DB/WAL/UNSORTED data | |
16372 | // - observed maximum spillovers | |
16373 | uint64_t max_db_use = 0; // max db usage we potentially observed | |
f6b5b4d7 | 16374 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_LOG - LEVEL_FIRST); |
9f95a23c TL |
16375 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_WAL - LEVEL_FIRST); |
16376 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_DB - LEVEL_FIRST); | |
16377 | // this could go to db hence using it in the estimation | |
16378 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_SLOW, LEVEL_DB - LEVEL_FIRST); | |
16379 | ||
16380 | auto db_total = l_totals[LEVEL_DB - LEVEL_FIRST]; | |
16381 | uint64_t avail = min( | |
16382 | db_avail4slow, | |
16383 | max_db_use < db_total ? db_total - max_db_use : 0); | |
16384 | ||
16385 | // considering current DB dev usage for SLOW data | |
16386 | if (avail > per_level_per_dev_usage.at(BlueFS::BDEV_DB, LEVEL_SLOW - LEVEL_FIRST)) { | |
16387 | res = BlueFS::BDEV_DB; | |
16388 | } | |
16389 | } | |
16390 | break; | |
f6b5b4d7 | 16391 | case LEVEL_LOG: |
9f95a23c TL |
16392 | case LEVEL_WAL: |
16393 | res = BlueFS::BDEV_WAL; | |
16394 | break; | |
16395 | case LEVEL_DB: | |
16396 | default: | |
16397 | res = BlueFS::BDEV_DB; | |
16398 | break; | |
16399 | } | |
16400 | return res; | |
16401 | } | |
16402 | ||
16403 | void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const | |
16404 | { | |
16405 | res.emplace_back(base, l_totals[LEVEL_DB - LEVEL_FIRST]); | |
16406 | res.emplace_back(base + ".slow", l_totals[LEVEL_SLOW - LEVEL_FIRST]); | |
16407 | } | |
16408 | ||
b3b6e05e | 16409 | void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(std::string_view dirname) const { |
9f95a23c TL |
16410 | uint8_t res = LEVEL_DB; |
16411 | if (dirname.length() > 5) { | |
16412 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
16413 | // match up with bluestore. the slow device is always the second | |
16414 | // one (when a dedicated block.db device is present and used at | |
16415 | // bdev 0). the wal device is always last. | |
16416 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
16417 | res = LEVEL_SLOW; | |
16418 | } | |
16419 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
16420 | res = LEVEL_WAL; | |
16421 | } | |
16422 | } | |
16423 | return reinterpret_cast<void*>(res); | |
16424 | } | |
16425 | ||
16426 | void RocksDBBlueFSVolumeSelector::dump(ostream& sout) { | |
16427 | auto max_x = per_level_per_dev_usage.get_max_x(); | |
16428 | auto max_y = per_level_per_dev_usage.get_max_y(); | |
16429 | sout << "RocksDBBlueFSVolumeSelector: wal_total:" << l_totals[LEVEL_WAL - LEVEL_FIRST] | |
16430 | << ", db_total:" << l_totals[LEVEL_DB - LEVEL_FIRST] | |
16431 | << ", slow_total:" << l_totals[LEVEL_SLOW - LEVEL_FIRST] | |
16432 | << ", db_avail:" << db_avail4slow << std::endl | |
16433 | << "Usage matrix:" << std::endl; | |
f6b5b4d7 | 16434 | constexpr std::array<const char*, 8> names{ { |
9f95a23c TL |
16435 | "DEV/LEV", |
16436 | "WAL", | |
16437 | "DB", | |
16438 | "SLOW", | |
16439 | "*", | |
16440 | "*", | |
f6b5b4d7 TL |
16441 | "REAL", |
16442 | "FILES", | |
9f95a23c TL |
16443 | } }; |
16444 | const size_t width = 12; | |
16445 | for (size_t i = 0; i < names.size(); ++i) { | |
16446 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16447 | sout.width(width); | |
16448 | sout << names[i]; | |
16449 | } | |
16450 | sout << std::endl; | |
16451 | for (size_t l = 0; l < max_y; l++) { | |
16452 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16453 | sout.width(width); | |
16454 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
16455 | case LEVEL_LOG: |
16456 | sout << "LOG"; break; | |
9f95a23c TL |
16457 | case LEVEL_WAL: |
16458 | sout << "WAL"; break; | |
16459 | case LEVEL_DB: | |
16460 | sout << "DB"; break; | |
16461 | case LEVEL_SLOW: | |
16462 | sout << "SLOW"; break; | |
16463 | case LEVEL_MAX: | |
16464 | sout << "TOTALS"; break; | |
16465 | } | |
f6b5b4d7 | 16466 | for (size_t d = 0; d < max_x; d++) { |
9f95a23c TL |
16467 | sout.setf(std::ios::left, std::ios::adjustfield); |
16468 | sout.width(width); | |
16469 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(d, l))); | |
16470 | } | |
16471 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16472 | sout.width(width); | |
f6b5b4d7 | 16473 | sout << stringify(per_level_files[l]) << std::endl; |
9f95a23c TL |
16474 | } |
16475 | ceph_assert(max_x == per_level_per_dev_max.get_max_x()); | |
16476 | ceph_assert(max_y == per_level_per_dev_max.get_max_y()); | |
16477 | sout << "MAXIMUMS:" << std::endl; | |
16478 | for (size_t l = 0; l < max_y; l++) { | |
16479 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16480 | sout.width(width); | |
16481 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
16482 | case LEVEL_LOG: |
16483 | sout << "LOG"; break; | |
9f95a23c TL |
16484 | case LEVEL_WAL: |
16485 | sout << "WAL"; break; | |
16486 | case LEVEL_DB: | |
16487 | sout << "DB"; break; | |
16488 | case LEVEL_SLOW: | |
16489 | sout << "SLOW"; break; | |
16490 | case LEVEL_MAX: | |
16491 | sout << "TOTALS"; break; | |
16492 | } | |
16493 | for (size_t d = 0; d < max_x - 1; d++) { | |
16494 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16495 | sout.width(width); | |
16496 | sout << stringify(byte_u_t(per_level_per_dev_max.at(d, l))); | |
16497 | } | |
16498 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16499 | sout.width(width); | |
16500 | sout << stringify(byte_u_t(per_level_per_dev_max.at(max_x - 1, l))); | |
16501 | if (l < max_y - 1) { | |
16502 | sout << std::endl; | |
16503 | } | |
16504 | } | |
16505 | } | |
11fdf7f2 | 16506 | |
9f95a23c | 16507 | // ======================================================= |