]>
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 | 277 | ghobject_t *temp_start, ghobject_t *temp_end, |
a4b75251 | 278 | ghobject_t *start, ghobject_t *end, bool legacy) |
7c673cae | 279 | { |
7c673cae | 280 | spg_t pgid; |
a4b75251 TL |
281 | constexpr uint32_t MAX_HASH = std::numeric_limits<uint32_t>::max(); |
282 | // use different nspaces due to we use different schemes when encoding | |
283 | // keys for listing objects | |
284 | const std::string_view MAX_NSPACE = legacy ? "\x7f" : "\xff"; | |
7c673cae | 285 | if (cid.is_pg(&pgid)) { |
f91f0fd5 | 286 | start->shard_id = pgid.shard; |
7c673cae FG |
287 | *temp_start = *start; |
288 | ||
f91f0fd5 TL |
289 | start->hobj.pool = pgid.pool(); |
290 | temp_start->hobj.pool = -2ll - pgid.pool(); | |
7c673cae FG |
291 | |
292 | *end = *start; | |
293 | *temp_end = *temp_start; | |
294 | ||
295 | uint32_t reverse_hash = hobject_t::_reverse_bits(pgid.ps()); | |
f91f0fd5 TL |
296 | start->hobj.set_bitwise_key_u32(reverse_hash); |
297 | temp_start->hobj.set_bitwise_key_u32(reverse_hash); | |
7c673cae FG |
298 | |
299 | uint64_t end_hash = reverse_hash + (1ull << (32 - bits)); | |
a4b75251 TL |
300 | if (end_hash > MAX_HASH) { |
301 | // make sure end hobj is even greater than the maximum possible hobj | |
302 | end->hobj.set_bitwise_key_u32(MAX_HASH); | |
303 | temp_end->hobj.set_bitwise_key_u32(MAX_HASH); | |
304 | end->hobj.nspace = MAX_NSPACE; | |
305 | } else { | |
306 | end->hobj.set_bitwise_key_u32(end_hash); | |
307 | temp_end->hobj.set_bitwise_key_u32(end_hash); | |
308 | } | |
7c673cae | 309 | } else { |
f91f0fd5 TL |
310 | start->shard_id = shard_id_t::NO_SHARD; |
311 | start->hobj.pool = -1ull; | |
312 | ||
7c673cae | 313 | *end = *start; |
f91f0fd5 | 314 | start->hobj.set_bitwise_key_u32(0); |
a4b75251 TL |
315 | end->hobj.set_bitwise_key_u32(MAX_HASH); |
316 | end->hobj.nspace = MAX_NSPACE; | |
7c673cae FG |
317 | // no separate temp section |
318 | *temp_start = *end; | |
319 | *temp_end = *end; | |
320 | } | |
f91f0fd5 TL |
321 | |
322 | start->generation = 0; | |
323 | end->generation = 0; | |
324 | temp_start->generation = 0; | |
325 | temp_end->generation = 0; | |
7c673cae FG |
326 | } |
327 | ||
328 | static void get_shared_blob_key(uint64_t sbid, string *key) | |
329 | { | |
330 | key->clear(); | |
331 | _key_encode_u64(sbid, key); | |
332 | } | |
333 | ||
334 | static int get_key_shared_blob(const string& key, uint64_t *sbid) | |
335 | { | |
336 | const char *p = key.c_str(); | |
337 | if (key.length() < sizeof(uint64_t)) | |
338 | return -1; | |
224ce89b | 339 | _key_decode_u64(p, sbid); |
7c673cae FG |
340 | return 0; |
341 | } | |
342 | ||
343 | template<typename S> | |
f91f0fd5 | 344 | static void _key_encode_prefix(const ghobject_t& oid, S *key) |
7c673cae | 345 | { |
f91f0fd5 TL |
346 | _key_encode_shard(oid.shard_id, key); |
347 | _key_encode_u64(oid.hobj.pool + 0x8000000000000000ull, key); | |
348 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), key); | |
349 | } | |
7c673cae | 350 | |
f91f0fd5 TL |
351 | static const char *_key_decode_prefix(const char *p, ghobject_t *oid) |
352 | { | |
7c673cae FG |
353 | p = _key_decode_shard(p, &oid->shard_id); |
354 | ||
355 | uint64_t pool; | |
356 | p = _key_decode_u64(p, &pool); | |
357 | oid->hobj.pool = pool - 0x8000000000000000ull; | |
358 | ||
359 | unsigned hash; | |
360 | p = _key_decode_u32(p, &hash); | |
361 | ||
362 | oid->hobj.set_bitwise_key_u32(hash); | |
363 | ||
f91f0fd5 TL |
364 | return p; |
365 | } | |
366 | ||
367 | #define ENCODED_KEY_PREFIX_LEN (1 + 8 + 4) | |
368 | ||
369 | template<typename S> | |
370 | static int get_key_object(const S& key, ghobject_t *oid) | |
371 | { | |
372 | int r; | |
373 | const char *p = key.c_str(); | |
374 | ||
375 | if (key.length() < ENCODED_KEY_PREFIX_LEN) | |
376 | return -1; | |
377 | ||
378 | p = _key_decode_prefix(p, oid); | |
379 | ||
380 | if (key.length() == ENCODED_KEY_PREFIX_LEN) | |
381 | return -2; | |
382 | ||
7c673cae FG |
383 | r = decode_escaped(p, &oid->hobj.nspace); |
384 | if (r < 0) | |
385 | return -2; | |
386 | p += r + 1; | |
387 | ||
388 | string k; | |
389 | r = decode_escaped(p, &k); | |
390 | if (r < 0) | |
391 | return -3; | |
392 | p += r + 1; | |
393 | if (*p == '=') { | |
394 | // no key | |
395 | ++p; | |
396 | oid->hobj.oid.name = k; | |
397 | } else if (*p == '<' || *p == '>') { | |
398 | // key + name | |
399 | ++p; | |
400 | r = decode_escaped(p, &oid->hobj.oid.name); | |
401 | if (r < 0) | |
402 | return -5; | |
403 | p += r + 1; | |
404 | oid->hobj.set_key(k); | |
405 | } else { | |
406 | // malformed | |
407 | return -6; | |
408 | } | |
409 | ||
410 | p = _key_decode_u64(p, &oid->hobj.snap.val); | |
411 | p = _key_decode_u64(p, &oid->generation); | |
412 | ||
413 | if (*p != ONODE_KEY_SUFFIX) { | |
414 | return -7; | |
415 | } | |
416 | p++; | |
417 | if (*p) { | |
418 | // if we get something other than a null terminator here, | |
419 | // something goes wrong. | |
420 | return -8; | |
421 | } | |
422 | ||
423 | return 0; | |
424 | } | |
425 | ||
426 | template<typename S> | |
427 | static void get_object_key(CephContext *cct, const ghobject_t& oid, S *key) | |
428 | { | |
429 | key->clear(); | |
430 | ||
f91f0fd5 | 431 | size_t max_len = ENCODED_KEY_PREFIX_LEN + |
7c673cae FG |
432 | (oid.hobj.nspace.length() * 3 + 1) + |
433 | (oid.hobj.get_key().length() * 3 + 1) + | |
434 | 1 + // for '<', '=', or '>' | |
435 | (oid.hobj.oid.name.length() * 3 + 1) + | |
436 | 8 + 8 + 1; | |
437 | key->reserve(max_len); | |
438 | ||
f91f0fd5 | 439 | _key_encode_prefix(oid, key); |
7c673cae FG |
440 | |
441 | append_escaped(oid.hobj.nspace, key); | |
442 | ||
443 | if (oid.hobj.get_key().length()) { | |
444 | // is a key... could be < = or >. | |
445 | append_escaped(oid.hobj.get_key(), key); | |
446 | // (ASCII chars < = and > sort in that order, yay) | |
447 | int r = oid.hobj.get_key().compare(oid.hobj.oid.name); | |
448 | if (r) { | |
449 | key->append(r > 0 ? ">" : "<"); | |
450 | append_escaped(oid.hobj.oid.name, key); | |
451 | } else { | |
452 | // same as no key | |
453 | key->append("="); | |
454 | } | |
455 | } else { | |
456 | // no key | |
457 | append_escaped(oid.hobj.oid.name, key); | |
458 | key->append("="); | |
459 | } | |
460 | ||
461 | _key_encode_u64(oid.hobj.snap, key); | |
462 | _key_encode_u64(oid.generation, key); | |
463 | ||
464 | key->push_back(ONODE_KEY_SUFFIX); | |
465 | ||
466 | // sanity check | |
467 | if (true) { | |
468 | ghobject_t t; | |
469 | int r = get_key_object(*key, &t); | |
470 | if (r || t != oid) { | |
471 | derr << " r " << r << dendl; | |
472 | derr << "key " << pretty_binary_string(*key) << dendl; | |
473 | derr << "oid " << oid << dendl; | |
474 | derr << " t " << t << dendl; | |
11fdf7f2 | 475 | ceph_assert(r == 0 && t == oid); |
7c673cae FG |
476 | } |
477 | } | |
478 | } | |
479 | ||
7c673cae FG |
480 | // extent shard keys are the onode key, plus a u32, plus 'x'. the trailing |
481 | // char lets us quickly test whether it is a shard key without decoding any | |
482 | // of the prefix bytes. | |
483 | template<typename S> | |
484 | static void get_extent_shard_key(const S& onode_key, uint32_t offset, | |
485 | string *key) | |
486 | { | |
487 | key->clear(); | |
488 | key->reserve(onode_key.length() + 4 + 1); | |
489 | key->append(onode_key.c_str(), onode_key.size()); | |
490 | _key_encode_u32(offset, key); | |
491 | key->push_back(EXTENT_SHARD_KEY_SUFFIX); | |
492 | } | |
493 | ||
494 | static void rewrite_extent_shard_key(uint32_t offset, string *key) | |
495 | { | |
11fdf7f2 TL |
496 | ceph_assert(key->size() > sizeof(uint32_t) + 1); |
497 | ceph_assert(*key->rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
498 | _key_encode_u32(offset, key->size() - sizeof(uint32_t) - 1, key); |
499 | } | |
500 | ||
501 | template<typename S> | |
502 | static void generate_extent_shard_key_and_apply( | |
503 | const S& onode_key, | |
504 | uint32_t offset, | |
505 | string *key, | |
506 | std::function<void(const string& final_key)> apply) | |
507 | { | |
508 | if (key->empty()) { // make full key | |
11fdf7f2 | 509 | ceph_assert(!onode_key.empty()); |
7c673cae FG |
510 | get_extent_shard_key(onode_key, offset, key); |
511 | } else { | |
512 | rewrite_extent_shard_key(offset, key); | |
513 | } | |
514 | apply(*key); | |
515 | } | |
516 | ||
517 | int get_key_extent_shard(const string& key, string *onode_key, uint32_t *offset) | |
518 | { | |
11fdf7f2 TL |
519 | ceph_assert(key.size() > sizeof(uint32_t) + 1); |
520 | ceph_assert(*key.rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
521 | int okey_len = key.size() - sizeof(uint32_t) - 1; |
522 | *onode_key = key.substr(0, okey_len); | |
523 | const char *p = key.data() + okey_len; | |
224ce89b | 524 | _key_decode_u32(p, offset); |
7c673cae FG |
525 | return 0; |
526 | } | |
527 | ||
528 | static bool is_extent_shard_key(const string& key) | |
529 | { | |
530 | return *key.rbegin() == EXTENT_SHARD_KEY_SUFFIX; | |
531 | } | |
532 | ||
7c673cae FG |
533 | static void get_deferred_key(uint64_t seq, string *out) |
534 | { | |
535 | _key_encode_u64(seq, out); | |
536 | } | |
537 | ||
11fdf7f2 TL |
538 | static void get_pool_stat_key(int64_t pool_id, string *key) |
539 | { | |
540 | key->clear(); | |
541 | _key_encode_u64(pool_id, key); | |
542 | } | |
543 | ||
544 | static int get_key_pool_stat(const string& key, uint64_t* pool_id) | |
545 | { | |
546 | const char *p = key.c_str(); | |
547 | if (key.length() < sizeof(uint64_t)) | |
548 | return -1; | |
549 | _key_decode_u64(p, pool_id); | |
550 | return 0; | |
551 | } | |
7c673cae | 552 | |
522d829b | 553 | |
81eedcae TL |
554 | template <int LogLevelV> |
555 | void _dump_extent_map(CephContext *cct, const BlueStore::ExtentMap &em) | |
556 | { | |
557 | uint64_t pos = 0; | |
558 | for (auto& s : em.shards) { | |
559 | dout(LogLevelV) << __func__ << " shard " << *s.shard_info | |
560 | << (s.loaded ? " (loaded)" : "") | |
561 | << (s.dirty ? " (dirty)" : "") | |
562 | << dendl; | |
563 | } | |
564 | for (auto& e : em.extent_map) { | |
565 | dout(LogLevelV) << __func__ << " " << e << dendl; | |
566 | ceph_assert(e.logical_offset >= pos); | |
567 | pos = e.logical_offset + e.length; | |
568 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
569 | if (blob.has_csum()) { | |
570 | vector<uint64_t> v; | |
571 | unsigned n = blob.get_csum_count(); | |
572 | for (unsigned i = 0; i < n; ++i) | |
573 | v.push_back(blob.get_csum_item(i)); | |
574 | dout(LogLevelV) << __func__ << " csum: " << std::hex << v << std::dec | |
575 | << dendl; | |
576 | } | |
577 | std::lock_guard l(e.blob->shared_blob->get_cache()->lock); | |
578 | for (auto& i : e.blob->shared_blob->bc.buffer_map) { | |
579 | dout(LogLevelV) << __func__ << " 0x" << std::hex << i.first | |
580 | << "~" << i.second->length << std::dec | |
581 | << " " << *i.second << dendl; | |
582 | } | |
583 | } | |
584 | } | |
585 | ||
586 | template <int LogLevelV> | |
587 | void _dump_onode(CephContext *cct, const BlueStore::Onode& o) | |
588 | { | |
589 | if (!cct->_conf->subsys.should_gather<ceph_subsys_bluestore, LogLevelV>()) | |
590 | return; | |
591 | dout(LogLevelV) << __func__ << " " << &o << " " << o.oid | |
592 | << " nid " << o.onode.nid | |
593 | << " size 0x" << std::hex << o.onode.size | |
594 | << " (" << std::dec << o.onode.size << ")" | |
595 | << " expected_object_size " << o.onode.expected_object_size | |
596 | << " expected_write_size " << o.onode.expected_write_size | |
597 | << " in " << o.onode.extent_map_shards.size() << " shards" | |
598 | << ", " << o.extent_map.spanning_blob_map.size() | |
599 | << " spanning blobs" | |
600 | << dendl; | |
601 | for (auto p = o.onode.attrs.begin(); | |
602 | p != o.onode.attrs.end(); | |
603 | ++p) { | |
604 | dout(LogLevelV) << __func__ << " attr " << p->first | |
605 | << " len " << p->second.length() << dendl; | |
606 | } | |
607 | _dump_extent_map<LogLevelV>(cct, o.extent_map); | |
608 | } | |
609 | ||
610 | template <int LogLevelV> | |
611 | void _dump_transaction(CephContext *cct, ObjectStore::Transaction *t) | |
612 | { | |
613 | dout(LogLevelV) << __func__ << " transaction dump:\n"; | |
614 | JSONFormatter f(true); | |
615 | f.open_object_section("transaction"); | |
616 | t->dump(&f); | |
617 | f.close_section(); | |
618 | f.flush(*_dout); | |
619 | *_dout << dendl; | |
620 | } | |
621 | ||
7c673cae FG |
622 | // Buffer |
623 | ||
624 | ostream& operator<<(ostream& out, const BlueStore::Buffer& b) | |
625 | { | |
626 | out << "buffer(" << &b << " space " << b.space << " 0x" << std::hex | |
627 | << b.offset << "~" << b.length << std::dec | |
628 | << " " << BlueStore::Buffer::get_state_name(b.state); | |
629 | if (b.flags) | |
630 | out << " " << BlueStore::Buffer::get_flag_name(b.flags); | |
631 | return out << ")"; | |
632 | } | |
633 | ||
f91f0fd5 TL |
634 | namespace { |
635 | ||
636 | /* | |
637 | * Due to a bug in key string encoding (see a comment for append_escaped) | |
638 | * the KeyValueDB iterator does not lexicographically sort the same | |
639 | * way that ghobject_t does: objects with the same hash may have wrong order. | |
640 | * | |
641 | * This is the iterator wrapper that fixes the keys order. | |
642 | */ | |
643 | ||
644 | class CollectionListIterator { | |
645 | public: | |
646 | CollectionListIterator(const KeyValueDB::Iterator &it) | |
647 | : m_it(it) { | |
648 | } | |
649 | virtual ~CollectionListIterator() { | |
650 | } | |
651 | ||
652 | virtual bool valid() const = 0; | |
653 | virtual const ghobject_t &oid() const = 0; | |
654 | virtual void lower_bound(const ghobject_t &oid) = 0; | |
655 | virtual void upper_bound(const ghobject_t &oid) = 0; | |
656 | virtual void next() = 0; | |
657 | ||
adb31ebb TL |
658 | virtual int cmp(const ghobject_t &oid) const = 0; |
659 | ||
660 | bool is_ge(const ghobject_t &oid) const { | |
661 | return cmp(oid) >= 0; | |
662 | } | |
663 | ||
664 | bool is_lt(const ghobject_t &oid) const { | |
665 | return cmp(oid) < 0; | |
666 | } | |
667 | ||
f91f0fd5 TL |
668 | protected: |
669 | KeyValueDB::Iterator m_it; | |
670 | }; | |
671 | ||
672 | class SimpleCollectionListIterator : public CollectionListIterator { | |
673 | public: | |
674 | SimpleCollectionListIterator(CephContext *cct, const KeyValueDB::Iterator &it) | |
675 | : CollectionListIterator(it), m_cct(cct) { | |
676 | } | |
677 | ||
678 | bool valid() const override { | |
679 | return m_it->valid(); | |
680 | } | |
681 | ||
682 | const ghobject_t &oid() const override { | |
683 | ceph_assert(valid()); | |
684 | ||
685 | return m_oid; | |
686 | } | |
687 | ||
688 | void lower_bound(const ghobject_t &oid) override { | |
689 | string key; | |
690 | get_object_key(m_cct, oid, &key); | |
691 | ||
692 | m_it->lower_bound(key); | |
693 | get_oid(); | |
694 | } | |
695 | ||
696 | void upper_bound(const ghobject_t &oid) override { | |
697 | string key; | |
698 | get_object_key(m_cct, oid, &key); | |
699 | ||
700 | m_it->upper_bound(key); | |
701 | get_oid(); | |
702 | } | |
703 | ||
704 | void next() override { | |
705 | ceph_assert(valid()); | |
706 | ||
707 | m_it->next(); | |
708 | get_oid(); | |
709 | } | |
710 | ||
adb31ebb TL |
711 | int cmp(const ghobject_t &oid) const override { |
712 | ceph_assert(valid()); | |
713 | ||
714 | string key; | |
715 | get_object_key(m_cct, oid, &key); | |
716 | ||
717 | return m_it->key().compare(key); | |
718 | } | |
719 | ||
f91f0fd5 TL |
720 | private: |
721 | CephContext *m_cct; | |
722 | ghobject_t m_oid; | |
723 | ||
724 | void get_oid() { | |
f67539c2 TL |
725 | m_oid = ghobject_t(); |
726 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
727 | m_it->next(); | |
f91f0fd5 | 728 | } |
f67539c2 | 729 | if (!valid()) { |
f91f0fd5 TL |
730 | return; |
731 | } | |
732 | ||
f91f0fd5 TL |
733 | int r = get_key_object(m_it->key(), &m_oid); |
734 | ceph_assert(r == 0); | |
735 | } | |
736 | }; | |
737 | ||
738 | class SortedCollectionListIterator : public CollectionListIterator { | |
739 | public: | |
740 | SortedCollectionListIterator(const KeyValueDB::Iterator &it) | |
741 | : CollectionListIterator(it), m_chunk_iter(m_chunk.end()) { | |
742 | } | |
743 | ||
744 | bool valid() const override { | |
745 | return m_chunk_iter != m_chunk.end(); | |
746 | } | |
747 | ||
748 | const ghobject_t &oid() const override { | |
749 | ceph_assert(valid()); | |
750 | ||
751 | return m_chunk_iter->first; | |
752 | } | |
753 | ||
754 | void lower_bound(const ghobject_t &oid) override { | |
755 | std::string key; | |
756 | _key_encode_prefix(oid, &key); | |
757 | ||
758 | m_it->lower_bound(key); | |
759 | m_chunk_iter = m_chunk.end(); | |
760 | if (!get_next_chunk()) { | |
761 | return; | |
762 | } | |
763 | ||
764 | if (this->oid().shard_id != oid.shard_id || | |
765 | this->oid().hobj.pool != oid.hobj.pool || | |
766 | this->oid().hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
767 | return; | |
768 | } | |
769 | ||
770 | m_chunk_iter = m_chunk.lower_bound(oid); | |
771 | if (m_chunk_iter == m_chunk.end()) { | |
772 | get_next_chunk(); | |
773 | } | |
774 | } | |
775 | ||
776 | void upper_bound(const ghobject_t &oid) override { | |
777 | lower_bound(oid); | |
778 | ||
779 | if (valid() && this->oid() == oid) { | |
780 | next(); | |
781 | } | |
782 | } | |
783 | ||
784 | void next() override { | |
785 | ceph_assert(valid()); | |
786 | ||
787 | m_chunk_iter++; | |
788 | if (m_chunk_iter == m_chunk.end()) { | |
789 | get_next_chunk(); | |
790 | } | |
791 | } | |
792 | ||
adb31ebb TL |
793 | int cmp(const ghobject_t &oid) const override { |
794 | ceph_assert(valid()); | |
795 | ||
796 | if (this->oid() < oid) { | |
797 | return -1; | |
798 | } | |
799 | if (this->oid() > oid) { | |
800 | return 1; | |
801 | } | |
802 | return 0; | |
803 | } | |
804 | ||
f91f0fd5 TL |
805 | private: |
806 | std::map<ghobject_t, std::string> m_chunk; | |
807 | std::map<ghobject_t, std::string>::iterator m_chunk_iter; | |
808 | ||
809 | bool get_next_chunk() { | |
810 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
811 | m_it->next(); | |
812 | } | |
813 | ||
814 | if (!m_it->valid()) { | |
815 | return false; | |
816 | } | |
817 | ||
818 | ghobject_t oid; | |
819 | int r = get_key_object(m_it->key(), &oid); | |
820 | ceph_assert(r == 0); | |
821 | ||
822 | m_chunk.clear(); | |
823 | while (true) { | |
824 | m_chunk.insert({oid, m_it->key()}); | |
825 | ||
826 | do { | |
827 | m_it->next(); | |
828 | } while (m_it->valid() && is_extent_shard_key(m_it->key())); | |
829 | ||
830 | if (!m_it->valid()) { | |
831 | break; | |
832 | } | |
833 | ||
834 | ghobject_t next; | |
835 | r = get_key_object(m_it->key(), &next); | |
836 | ceph_assert(r == 0); | |
837 | if (next.shard_id != oid.shard_id || | |
838 | next.hobj.pool != oid.hobj.pool || | |
839 | next.hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
840 | break; | |
841 | } | |
842 | oid = next; | |
843 | } | |
844 | ||
845 | m_chunk_iter = m_chunk.begin(); | |
846 | return true; | |
847 | } | |
848 | }; | |
849 | ||
850 | } // anonymous namespace | |
851 | ||
7c673cae FG |
852 | // Garbage Collector |
853 | ||
854 | void BlueStore::GarbageCollector::process_protrusive_extents( | |
855 | const BlueStore::ExtentMap& extent_map, | |
856 | uint64_t start_offset, | |
857 | uint64_t end_offset, | |
858 | uint64_t start_touch_offset, | |
859 | uint64_t end_touch_offset, | |
860 | uint64_t min_alloc_size) | |
861 | { | |
11fdf7f2 | 862 | ceph_assert(start_offset <= start_touch_offset && end_offset>= end_touch_offset); |
7c673cae | 863 | |
11fdf7f2 TL |
864 | uint64_t lookup_start_offset = p2align(start_offset, min_alloc_size); |
865 | uint64_t lookup_end_offset = round_up_to(end_offset, min_alloc_size); | |
7c673cae FG |
866 | |
867 | dout(30) << __func__ << " (hex): [" << std::hex | |
868 | << lookup_start_offset << ", " << lookup_end_offset | |
869 | << ")" << std::dec << dendl; | |
870 | ||
871 | for (auto it = extent_map.seek_lextent(lookup_start_offset); | |
872 | it != extent_map.extent_map.end() && | |
873 | it->logical_offset < lookup_end_offset; | |
874 | ++it) { | |
875 | uint64_t alloc_unit_start = it->logical_offset / min_alloc_size; | |
876 | uint64_t alloc_unit_end = (it->logical_end() - 1) / min_alloc_size; | |
877 | ||
878 | dout(30) << __func__ << " " << *it | |
879 | << "alloc_units: " << alloc_unit_start << ".." << alloc_unit_end | |
880 | << dendl; | |
881 | ||
882 | Blob* b = it->blob.get(); | |
883 | ||
884 | if (it->logical_offset >=start_touch_offset && | |
885 | it->logical_end() <= end_touch_offset) { | |
886 | // Process extents within the range affected by | |
887 | // the current write request. | |
888 | // Need to take into account if existing extents | |
889 | // can be merged with them (uncompressed case) | |
890 | if (!b->get_blob().is_compressed()) { | |
891 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
892 | --blob_info_counted->expected_allocations; // don't need to allocate | |
893 | // new AU for compressed | |
894 | // data since another | |
895 | // collocated uncompressed | |
896 | // blob already exists | |
897 | dout(30) << __func__ << " --expected:" | |
898 | << alloc_unit_start << dendl; | |
899 | } | |
900 | used_alloc_unit = alloc_unit_end; | |
901 | blob_info_counted = nullptr; | |
902 | } | |
903 | } else if (b->get_blob().is_compressed()) { | |
904 | ||
905 | // additionally we take compressed blobs that were not impacted | |
906 | // by the write into account too | |
907 | BlobInfo& bi = | |
908 | affected_blobs.emplace( | |
909 | b, BlobInfo(b->get_referenced_bytes())).first->second; | |
910 | ||
911 | int adjust = | |
912 | (used_alloc_unit && used_alloc_unit == alloc_unit_start) ? 0 : 1; | |
913 | bi.expected_allocations += alloc_unit_end - alloc_unit_start + adjust; | |
914 | dout(30) << __func__ << " expected_allocations=" | |
915 | << bi.expected_allocations << " end_au:" | |
916 | << alloc_unit_end << dendl; | |
917 | ||
918 | blob_info_counted = &bi; | |
919 | used_alloc_unit = alloc_unit_end; | |
920 | ||
11fdf7f2 | 921 | ceph_assert(it->length <= bi.referenced_bytes); |
7c673cae FG |
922 | bi.referenced_bytes -= it->length; |
923 | dout(30) << __func__ << " affected_blob:" << *b | |
924 | << " unref 0x" << std::hex << it->length | |
925 | << " referenced = 0x" << bi.referenced_bytes | |
926 | << std::dec << dendl; | |
927 | // NOTE: we can't move specific blob to resulting GC list here | |
928 | // when reference counter == 0 since subsequent extents might | |
929 | // decrement its expected_allocation. | |
930 | // Hence need to enumerate all the extents first. | |
931 | if (!bi.collect_candidate) { | |
932 | bi.first_lextent = it; | |
933 | bi.collect_candidate = true; | |
934 | } | |
935 | bi.last_lextent = it; | |
936 | } else { | |
937 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
938 | // don't need to allocate new AU for compressed data since another | |
939 | // collocated uncompressed blob already exists | |
940 | --blob_info_counted->expected_allocations; | |
941 | dout(30) << __func__ << " --expected_allocations:" | |
942 | << alloc_unit_start << dendl; | |
943 | } | |
944 | used_alloc_unit = alloc_unit_end; | |
945 | blob_info_counted = nullptr; | |
946 | } | |
947 | } | |
948 | ||
949 | for (auto b_it = affected_blobs.begin(); | |
950 | b_it != affected_blobs.end(); | |
951 | ++b_it) { | |
952 | Blob* b = b_it->first; | |
953 | BlobInfo& bi = b_it->second; | |
954 | if (bi.referenced_bytes == 0) { | |
955 | uint64_t len_on_disk = b_it->first->get_blob().get_ondisk_length(); | |
956 | int64_t blob_expected_for_release = | |
11fdf7f2 | 957 | round_up_to(len_on_disk, min_alloc_size) / min_alloc_size; |
7c673cae FG |
958 | |
959 | dout(30) << __func__ << " " << *(b_it->first) | |
960 | << " expected4release=" << blob_expected_for_release | |
961 | << " expected_allocations=" << bi.expected_allocations | |
962 | << dendl; | |
963 | int64_t benefit = blob_expected_for_release - bi.expected_allocations; | |
11fdf7f2 | 964 | if (benefit >= g_conf()->bluestore_gc_enable_blob_threshold) { |
7c673cae FG |
965 | if (bi.collect_candidate) { |
966 | auto it = bi.first_lextent; | |
967 | bool bExit = false; | |
968 | do { | |
969 | if (it->blob.get() == b) { | |
eafe8130 | 970 | extents_to_collect.insert(it->logical_offset, it->length); |
7c673cae FG |
971 | } |
972 | bExit = it == bi.last_lextent; | |
973 | ++it; | |
31f18b77 | 974 | } while (!bExit); |
7c673cae FG |
975 | } |
976 | expected_for_release += blob_expected_for_release; | |
977 | expected_allocations += bi.expected_allocations; | |
978 | } | |
979 | } | |
980 | } | |
981 | } | |
982 | ||
983 | int64_t BlueStore::GarbageCollector::estimate( | |
984 | uint64_t start_offset, | |
985 | uint64_t length, | |
986 | const BlueStore::ExtentMap& extent_map, | |
987 | const BlueStore::old_extent_map_t& old_extents, | |
988 | uint64_t min_alloc_size) | |
989 | { | |
990 | ||
991 | affected_blobs.clear(); | |
992 | extents_to_collect.clear(); | |
993 | used_alloc_unit = boost::optional<uint64_t >(); | |
994 | blob_info_counted = nullptr; | |
995 | ||
eafe8130 TL |
996 | uint64_t gc_start_offset = start_offset; |
997 | uint64_t gc_end_offset = start_offset + length; | |
7c673cae FG |
998 | |
999 | uint64_t end_offset = start_offset + length; | |
1000 | ||
1001 | for (auto it = old_extents.begin(); it != old_extents.end(); ++it) { | |
1002 | Blob* b = it->e.blob.get(); | |
1003 | if (b->get_blob().is_compressed()) { | |
1004 | ||
1005 | // update gc_start_offset/gc_end_offset if needed | |
1006 | gc_start_offset = min(gc_start_offset, (uint64_t)it->e.blob_start()); | |
11fdf7f2 | 1007 | gc_end_offset = std::max(gc_end_offset, (uint64_t)it->e.blob_end()); |
7c673cae FG |
1008 | |
1009 | auto o = it->e.logical_offset; | |
1010 | auto l = it->e.length; | |
1011 | ||
1012 | uint64_t ref_bytes = b->get_referenced_bytes(); | |
1013 | // micro optimization to bypass blobs that have no more references | |
1014 | if (ref_bytes != 0) { | |
1015 | dout(30) << __func__ << " affected_blob:" << *b | |
1016 | << " unref 0x" << std::hex << o << "~" << l | |
1017 | << std::dec << dendl; | |
1018 | affected_blobs.emplace(b, BlobInfo(ref_bytes)); | |
1019 | } | |
1020 | } | |
1021 | } | |
1022 | dout(30) << __func__ << " gc range(hex): [" << std::hex | |
1023 | << gc_start_offset << ", " << gc_end_offset | |
1024 | << ")" << std::dec << dendl; | |
1025 | ||
1026 | // enumerate preceeding extents to check if they reference affected blobs | |
1027 | if (gc_start_offset < start_offset || gc_end_offset > end_offset) { | |
1028 | process_protrusive_extents(extent_map, | |
1029 | gc_start_offset, | |
1030 | gc_end_offset, | |
1031 | start_offset, | |
1032 | end_offset, | |
1033 | min_alloc_size); | |
1034 | } | |
1035 | return expected_for_release - expected_allocations; | |
1036 | } | |
1037 | ||
9f95a23c TL |
1038 | // LruOnodeCacheShard |
1039 | struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard { | |
1040 | typedef boost::intrusive::list< | |
1041 | BlueStore::Onode, | |
1042 | boost::intrusive::member_hook< | |
1043 | BlueStore::Onode, | |
1044 | boost::intrusive::list_member_hook<>, | |
1045 | &BlueStore::Onode::lru_item> > list_t; | |
7c673cae | 1046 | |
9f95a23c | 1047 | list_t lru; |
7c673cae | 1048 | |
9f95a23c | 1049 | explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {} |
7c673cae | 1050 | |
f6b5b4d7 | 1051 | void _add(BlueStore::Onode* o, int level) override |
9f95a23c | 1052 | { |
f6b5b4d7 | 1053 | if (o->put_cache()) { |
9f95a23c | 1054 | (level > 0) ? lru.push_front(*o) : lru.push_back(*o); |
f6b5b4d7 TL |
1055 | } else { |
1056 | ++num_pinned; | |
9f95a23c | 1057 | } |
f6b5b4d7 TL |
1058 | ++num; // we count both pinned and unpinned entries |
1059 | dout(20) << __func__ << " " << this << " " << o->oid << " added, num=" << num << dendl; | |
eafe8130 | 1060 | } |
f6b5b4d7 | 1061 | void _rm(BlueStore::Onode* o) override |
9f95a23c | 1062 | { |
f6b5b4d7 | 1063 | if (o->pop_cache()) { |
9f95a23c | 1064 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
1065 | } else { |
1066 | ceph_assert(num_pinned); | |
1067 | --num_pinned; | |
9f95a23c | 1068 | } |
f6b5b4d7 TL |
1069 | ceph_assert(num); |
1070 | --num; | |
1071 | dout(20) << __func__ << " " << this << " " << " " << o->oid << " removed, num=" << num << dendl; | |
9f95a23c | 1072 | } |
f6b5b4d7 | 1073 | void _pin(BlueStore::Onode* o) override |
9f95a23c | 1074 | { |
9f95a23c | 1075 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
1076 | ++num_pinned; |
1077 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " pinned" << dendl; | |
9f95a23c | 1078 | } |
f6b5b4d7 | 1079 | void _unpin(BlueStore::Onode* o) override |
9f95a23c | 1080 | { |
f6b5b4d7 TL |
1081 | lru.push_front(*o); |
1082 | ceph_assert(num_pinned); | |
1083 | --num_pinned; | |
1084 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " unpinned" << dendl; | |
9f95a23c | 1085 | } |
adb31ebb TL |
1086 | void _unpin_and_rm(BlueStore::Onode* o) override |
1087 | { | |
1088 | o->pop_cache(); | |
1089 | ceph_assert(num_pinned); | |
1090 | --num_pinned; | |
1091 | ceph_assert(num); | |
1092 | --num; | |
1093 | } | |
9f95a23c TL |
1094 | void _trim_to(uint64_t new_size) override |
1095 | { | |
1096 | if (new_size >= lru.size()) { | |
1097 | return; // don't even try | |
1098 | } | |
1099 | uint64_t n = lru.size() - new_size; | |
1100 | auto p = lru.end(); | |
1101 | ceph_assert(p != lru.begin()); | |
1102 | --p; | |
f6b5b4d7 TL |
1103 | ceph_assert(num >= n); |
1104 | num -= n; | |
1105 | while (n-- > 0) { | |
9f95a23c | 1106 | BlueStore::Onode *o = &*p; |
f6b5b4d7 TL |
1107 | dout(20) << __func__ << " rm " << o->oid << " " |
1108 | << o->nref << " " << o->cached << " " << o->pinned << dendl; | |
9f95a23c TL |
1109 | if (p != lru.begin()) { |
1110 | lru.erase(p--); | |
1111 | } else { | |
f6b5b4d7 | 1112 | ceph_assert(n == 0); |
9f95a23c | 1113 | lru.erase(p); |
9f95a23c | 1114 | } |
f6b5b4d7 TL |
1115 | auto pinned = !o->pop_cache(); |
1116 | ceph_assert(!pinned); | |
1117 | o->c->onode_map._remove(o->oid); | |
9f95a23c | 1118 | } |
f6b5b4d7 TL |
1119 | } |
1120 | void move_pinned(OnodeCacheShard *to, BlueStore::Onode *o) override | |
1121 | { | |
1122 | if (to == this) { | |
1123 | return; | |
1124 | } | |
1125 | ceph_assert(o->cached); | |
1126 | ceph_assert(o->pinned); | |
1127 | ceph_assert(num); | |
1128 | ceph_assert(num_pinned); | |
1129 | --num_pinned; | |
1130 | --num; | |
1131 | ++to->num_pinned; | |
1132 | ++to->num; | |
9f95a23c TL |
1133 | } |
1134 | void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override | |
1135 | { | |
f6b5b4d7 | 1136 | *onodes += num; |
9f95a23c TL |
1137 | *pinned_onodes += num_pinned; |
1138 | } | |
1139 | }; | |
7c673cae | 1140 | |
9f95a23c TL |
1141 | // OnodeCacheShard |
1142 | BlueStore::OnodeCacheShard *BlueStore::OnodeCacheShard::create( | |
1143 | CephContext* cct, | |
1144 | string type, | |
1145 | PerfCounters *logger) | |
7c673cae | 1146 | { |
9f95a23c TL |
1147 | BlueStore::OnodeCacheShard *c = nullptr; |
1148 | // Currently we only implement an LRU cache for onodes | |
1149 | c = new LruOnodeCacheShard(cct); | |
1150 | c->logger = logger; | |
1151 | return c; | |
7c673cae FG |
1152 | } |
1153 | ||
9f95a23c TL |
1154 | // LruBufferCacheShard |
1155 | struct LruBufferCacheShard : public BlueStore::BufferCacheShard { | |
1156 | typedef boost::intrusive::list< | |
1157 | BlueStore::Buffer, | |
1158 | boost::intrusive::member_hook< | |
1159 | BlueStore::Buffer, | |
1160 | boost::intrusive::list_member_hook<>, | |
1161 | &BlueStore::Buffer::lru_item> > list_t; | |
1162 | list_t lru; | |
1163 | ||
1164 | explicit LruBufferCacheShard(CephContext *cct) : BlueStore::BufferCacheShard(cct) {} | |
1165 | ||
1166 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override { | |
1167 | if (near) { | |
1168 | auto q = lru.iterator_to(*near); | |
1169 | lru.insert(q, *b); | |
1170 | } else if (level > 0) { | |
1171 | lru.push_front(*b); | |
1172 | } else { | |
1173 | lru.push_back(*b); | |
7c673cae | 1174 | } |
9f95a23c TL |
1175 | buffer_bytes += b->length; |
1176 | num = lru.size(); | |
1177 | } | |
1178 | void _rm(BlueStore::Buffer *b) override { | |
1179 | ceph_assert(buffer_bytes >= b->length); | |
1180 | buffer_bytes -= b->length; | |
1181 | auto q = lru.iterator_to(*b); | |
1182 | lru.erase(q); | |
1183 | num = lru.size(); | |
1184 | } | |
1185 | void _move(BlueStore::BufferCacheShard *src, BlueStore::Buffer *b) override { | |
1186 | src->_rm(b); | |
1187 | _add(b, 0, nullptr); | |
1188 | } | |
1189 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override { | |
1190 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1191 | buffer_bytes += delta; | |
1192 | } | |
1193 | void _touch(BlueStore::Buffer *b) override { | |
1194 | auto p = lru.iterator_to(*b); | |
1195 | lru.erase(p); | |
1196 | lru.push_front(*b); | |
1197 | num = lru.size(); | |
1198 | _audit("_touch_buffer end"); | |
1199 | } | |
7c673cae | 1200 | |
9f95a23c TL |
1201 | void _trim_to(uint64_t max) override |
1202 | { | |
1203 | while (buffer_bytes > max) { | |
1204 | auto i = lru.rbegin(); | |
1205 | if (i == lru.rend()) { | |
1206 | // stop if lru is now empty | |
7c673cae FG |
1207 | break; |
1208 | } | |
1209 | ||
9f95a23c TL |
1210 | BlueStore::Buffer *b = &*i; |
1211 | ceph_assert(b->is_clean()); | |
1212 | dout(20) << __func__ << " rm " << *b << dendl; | |
1213 | b->space->_rm_buffer(this, b); | |
7c673cae | 1214 | } |
9f95a23c | 1215 | num = lru.size(); |
7c673cae | 1216 | } |
7c673cae | 1217 | |
9f95a23c TL |
1218 | void add_stats(uint64_t *extents, |
1219 | uint64_t *blobs, | |
1220 | uint64_t *buffers, | |
1221 | uint64_t *bytes) override { | |
1222 | *extents += num_extents; | |
1223 | *blobs += num_blobs; | |
1224 | *buffers += num; | |
1225 | *bytes += buffer_bytes; | |
7c673cae | 1226 | } |
9f95a23c TL |
1227 | #ifdef DEBUG_CACHE |
1228 | void _audit(const char *s) override | |
1229 | { | |
1230 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1231 | uint64_t s = 0; | |
1232 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1233 | s += i->length; | |
1234 | } | |
1235 | if (s != buffer_bytes) { | |
1236 | derr << __func__ << " buffer_size " << buffer_bytes << " actual " << s | |
1237 | << dendl; | |
1238 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1239 | derr << __func__ << " " << *i << dendl; | |
1240 | } | |
1241 | ceph_assert(s == buffer_bytes); | |
7c673cae | 1242 | } |
9f95a23c TL |
1243 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1244 | << " ok" << dendl; | |
7c673cae | 1245 | } |
7c673cae | 1246 | #endif |
9f95a23c | 1247 | }; |
7c673cae | 1248 | |
9f95a23c TL |
1249 | // TwoQBufferCacheShard |
1250 | ||
1251 | struct TwoQBufferCacheShard : public BlueStore::BufferCacheShard { | |
1252 | typedef boost::intrusive::list< | |
1253 | BlueStore::Buffer, | |
1254 | boost::intrusive::member_hook< | |
1255 | BlueStore::Buffer, | |
1256 | boost::intrusive::list_member_hook<>, | |
1257 | &BlueStore::Buffer::lru_item> > list_t; | |
1258 | list_t hot; ///< "Am" hot buffers | |
1259 | list_t warm_in; ///< "A1in" newly warm buffers | |
1260 | list_t warm_out; ///< "A1out" empty buffers we've evicted | |
9f95a23c TL |
1261 | |
1262 | enum { | |
1263 | BUFFER_NEW = 0, | |
1264 | BUFFER_WARM_IN, ///< in warm_in | |
1265 | BUFFER_WARM_OUT, ///< in warm_out | |
1266 | BUFFER_HOT, ///< in hot | |
1267 | BUFFER_TYPE_MAX | |
1268 | }; | |
7c673cae | 1269 | |
9f95a23c | 1270 | uint64_t list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type |
7c673cae | 1271 | |
9f95a23c TL |
1272 | public: |
1273 | explicit TwoQBufferCacheShard(CephContext *cct) : BufferCacheShard(cct) {} | |
7c673cae | 1274 | |
9f95a23c TL |
1275 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override |
1276 | { | |
1277 | dout(20) << __func__ << " level " << level << " near " << near | |
1278 | << " on " << *b | |
1279 | << " which has cache_private " << b->cache_private << dendl; | |
1280 | if (near) { | |
1281 | b->cache_private = near->cache_private; | |
1282 | switch (b->cache_private) { | |
1283 | case BUFFER_WARM_IN: | |
1284 | warm_in.insert(warm_in.iterator_to(*near), *b); | |
1285 | break; | |
1286 | case BUFFER_WARM_OUT: | |
1287 | ceph_assert(b->is_empty()); | |
1288 | warm_out.insert(warm_out.iterator_to(*near), *b); | |
1289 | break; | |
1290 | case BUFFER_HOT: | |
1291 | hot.insert(hot.iterator_to(*near), *b); | |
1292 | break; | |
1293 | default: | |
1294 | ceph_abort_msg("bad cache_private"); | |
1295 | } | |
1296 | } else if (b->cache_private == BUFFER_NEW) { | |
1297 | b->cache_private = BUFFER_WARM_IN; | |
1298 | if (level > 0) { | |
1299 | warm_in.push_front(*b); | |
1300 | } else { | |
1301 | // take caller hint to start at the back of the warm queue | |
1302 | warm_in.push_back(*b); | |
1303 | } | |
1304 | } else { | |
1305 | // we got a hint from discard | |
1306 | switch (b->cache_private) { | |
1307 | case BUFFER_WARM_IN: | |
1308 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1309 | // do this. | |
1310 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1311 | warm_in.push_front(*b); | |
1312 | break; | |
1313 | case BUFFER_WARM_OUT: | |
1314 | b->cache_private = BUFFER_HOT; | |
1315 | // move to hot. fall-thru | |
1316 | case BUFFER_HOT: | |
1317 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1318 | hot.push_front(*b); | |
1319 | break; | |
1320 | default: | |
1321 | ceph_abort_msg("bad cache_private"); | |
1322 | } | |
1323 | } | |
1324 | if (!b->is_empty()) { | |
1325 | buffer_bytes += b->length; | |
1326 | list_bytes[b->cache_private] += b->length; | |
1327 | } | |
1328 | num = hot.size() + warm_in.size(); | |
1329 | } | |
1330 | ||
1331 | void _rm(BlueStore::Buffer *b) override | |
1332 | { | |
1333 | dout(20) << __func__ << " " << *b << dendl; | |
1334 | if (!b->is_empty()) { | |
1335 | ceph_assert(buffer_bytes >= b->length); | |
1336 | buffer_bytes -= b->length; | |
1337 | ceph_assert(list_bytes[b->cache_private] >= b->length); | |
1338 | list_bytes[b->cache_private] -= b->length; | |
1339 | } | |
7c673cae FG |
1340 | switch (b->cache_private) { |
1341 | case BUFFER_WARM_IN: | |
9f95a23c | 1342 | warm_in.erase(warm_in.iterator_to(*b)); |
7c673cae FG |
1343 | break; |
1344 | case BUFFER_WARM_OUT: | |
9f95a23c | 1345 | warm_out.erase(warm_out.iterator_to(*b)); |
7c673cae FG |
1346 | break; |
1347 | case BUFFER_HOT: | |
9f95a23c | 1348 | hot.erase(hot.iterator_to(*b)); |
7c673cae FG |
1349 | break; |
1350 | default: | |
11fdf7f2 | 1351 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1352 | } |
9f95a23c TL |
1353 | num = hot.size() + warm_in.size(); |
1354 | } | |
1355 | ||
1356 | void _move(BlueStore::BufferCacheShard *srcc, BlueStore::Buffer *b) override | |
1357 | { | |
1358 | TwoQBufferCacheShard *src = static_cast<TwoQBufferCacheShard*>(srcc); | |
1359 | src->_rm(b); | |
1360 | ||
1361 | // preserve which list we're on (even if we can't preserve the order!) | |
7c673cae FG |
1362 | switch (b->cache_private) { |
1363 | case BUFFER_WARM_IN: | |
9f95a23c TL |
1364 | ceph_assert(!b->is_empty()); |
1365 | warm_in.push_back(*b); | |
7c673cae FG |
1366 | break; |
1367 | case BUFFER_WARM_OUT: | |
9f95a23c TL |
1368 | ceph_assert(b->is_empty()); |
1369 | warm_out.push_back(*b); | |
1370 | break; | |
7c673cae | 1371 | case BUFFER_HOT: |
9f95a23c TL |
1372 | ceph_assert(!b->is_empty()); |
1373 | hot.push_back(*b); | |
7c673cae FG |
1374 | break; |
1375 | default: | |
11fdf7f2 | 1376 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1377 | } |
9f95a23c TL |
1378 | if (!b->is_empty()) { |
1379 | buffer_bytes += b->length; | |
1380 | list_bytes[b->cache_private] += b->length; | |
1381 | } | |
1382 | num = hot.size() + warm_in.size(); | |
7c673cae | 1383 | } |
7c673cae | 1384 | |
9f95a23c TL |
1385 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override |
1386 | { | |
1387 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1388 | if (!b->is_empty()) { | |
1389 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1390 | buffer_bytes += delta; | |
1391 | ceph_assert((int64_t)list_bytes[b->cache_private] + delta >= 0); | |
1392 | list_bytes[b->cache_private] += delta; | |
1393 | } | |
7c673cae | 1394 | } |
7c673cae | 1395 | |
9f95a23c TL |
1396 | void _touch(BlueStore::Buffer *b) override { |
1397 | switch (b->cache_private) { | |
1398 | case BUFFER_WARM_IN: | |
1399 | // do nothing (somewhat counter-intuitively!) | |
1400 | break; | |
1401 | case BUFFER_WARM_OUT: | |
1402 | // move from warm_out to hot LRU | |
1403 | ceph_abort_msg("this happens via discard hint"); | |
1404 | break; | |
1405 | case BUFFER_HOT: | |
1406 | // move to front of hot LRU | |
1407 | hot.erase(hot.iterator_to(*b)); | |
1408 | hot.push_front(*b); | |
1409 | break; | |
1410 | } | |
1411 | num = hot.size() + warm_in.size(); | |
1412 | _audit("_touch_buffer end"); | |
7c673cae | 1413 | } |
7c673cae | 1414 | |
9f95a23c TL |
1415 | void _trim_to(uint64_t max) override |
1416 | { | |
1417 | if (buffer_bytes > max) { | |
1418 | uint64_t kin = max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1419 | uint64_t khot = max - kin; | |
1420 | ||
1421 | // pre-calculate kout based on average buffer size too, | |
1422 | // which is typical(the warm_in and hot lists may change later) | |
1423 | uint64_t kout = 0; | |
1424 | uint64_t buffer_num = hot.size() + warm_in.size(); | |
1425 | if (buffer_num) { | |
1426 | uint64_t avg_size = buffer_bytes / buffer_num; | |
1427 | ceph_assert(avg_size); | |
1428 | uint64_t calculated_num = max / avg_size; | |
1429 | kout = calculated_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1430 | } | |
1431 | ||
1432 | if (list_bytes[BUFFER_HOT] < khot) { | |
1433 | // hot is small, give slack to warm_in | |
1434 | kin += khot - list_bytes[BUFFER_HOT]; | |
1435 | } else if (list_bytes[BUFFER_WARM_IN] < kin) { | |
1436 | // warm_in is small, give slack to hot | |
1437 | khot += kin - list_bytes[BUFFER_WARM_IN]; | |
1438 | } | |
1439 | ||
1440 | // adjust warm_in list | |
1441 | int64_t to_evict_bytes = list_bytes[BUFFER_WARM_IN] - kin; | |
1442 | uint64_t evicted = 0; | |
1443 | ||
1444 | while (to_evict_bytes > 0) { | |
1445 | auto p = warm_in.rbegin(); | |
1446 | if (p == warm_in.rend()) { | |
1447 | // stop if warm_in list is now empty | |
1448 | break; | |
1449 | } | |
7c673cae | 1450 | |
9f95a23c TL |
1451 | BlueStore::Buffer *b = &*p; |
1452 | ceph_assert(b->is_clean()); | |
1453 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; | |
1454 | ceph_assert(buffer_bytes >= b->length); | |
1455 | buffer_bytes -= b->length; | |
1456 | ceph_assert(list_bytes[BUFFER_WARM_IN] >= b->length); | |
1457 | list_bytes[BUFFER_WARM_IN] -= b->length; | |
1458 | to_evict_bytes -= b->length; | |
1459 | evicted += b->length; | |
1460 | b->state = BlueStore::Buffer::STATE_EMPTY; | |
1461 | b->data.clear(); | |
1462 | warm_in.erase(warm_in.iterator_to(*b)); | |
1463 | warm_out.push_front(*b); | |
1464 | b->cache_private = BUFFER_WARM_OUT; | |
1465 | } | |
1466 | ||
1467 | if (evicted > 0) { | |
1468 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1469 | << " from warm_in list, done evicting warm_in buffers" | |
1470 | << dendl; | |
1471 | } | |
7c673cae | 1472 | |
9f95a23c TL |
1473 | // adjust hot list |
1474 | to_evict_bytes = list_bytes[BUFFER_HOT] - khot; | |
1475 | evicted = 0; | |
7c673cae | 1476 | |
9f95a23c TL |
1477 | while (to_evict_bytes > 0) { |
1478 | auto p = hot.rbegin(); | |
1479 | if (p == hot.rend()) { | |
1480 | // stop if hot list is now empty | |
1481 | break; | |
1482 | } | |
7c673cae | 1483 | |
9f95a23c TL |
1484 | BlueStore::Buffer *b = &*p; |
1485 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
1486 | ceph_assert(b->is_clean()); | |
1487 | // adjust evict size before buffer goes invalid | |
1488 | to_evict_bytes -= b->length; | |
1489 | evicted += b->length; | |
1490 | b->space->_rm_buffer(this, b); | |
1491 | } | |
7c673cae | 1492 | |
9f95a23c TL |
1493 | if (evicted > 0) { |
1494 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1495 | << " from hot list, done evicting hot buffers" | |
1496 | << dendl; | |
7c673cae FG |
1497 | } |
1498 | ||
9f95a23c TL |
1499 | // adjust warm out list too, if necessary |
1500 | int64_t n = warm_out.size() - kout; | |
1501 | while (n-- > 0) { | |
1502 | BlueStore::Buffer *b = &*warm_out.rbegin(); | |
1503 | ceph_assert(b->is_empty()); | |
1504 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; | |
1505 | b->space->_rm_buffer(this, b); | |
1506 | } | |
7c673cae | 1507 | } |
9f95a23c TL |
1508 | num = hot.size() + warm_in.size(); |
1509 | } | |
7c673cae | 1510 | |
9f95a23c TL |
1511 | void add_stats(uint64_t *extents, |
1512 | uint64_t *blobs, | |
1513 | uint64_t *buffers, | |
1514 | uint64_t *bytes) override { | |
1515 | *extents += num_extents; | |
1516 | *blobs += num_blobs; | |
1517 | *buffers += num; | |
1518 | *bytes += buffer_bytes; | |
1519 | } | |
7c673cae | 1520 | |
9f95a23c TL |
1521 | #ifdef DEBUG_CACHE |
1522 | void _audit(const char *s) override | |
1523 | { | |
1524 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1525 | uint64_t s = 0; | |
1526 | for (auto i = hot.begin(); i != hot.end(); ++i) { | |
1527 | s += i->length; | |
7c673cae FG |
1528 | } |
1529 | ||
9f95a23c TL |
1530 | uint64_t hot_bytes = s; |
1531 | if (hot_bytes != list_bytes[BUFFER_HOT]) { | |
1532 | derr << __func__ << " hot_list_bytes " | |
1533 | << list_bytes[BUFFER_HOT] | |
1534 | << " != actual " << hot_bytes | |
1535 | << dendl; | |
1536 | ceph_assert(hot_bytes == list_bytes[BUFFER_HOT]); | |
7c673cae FG |
1537 | } |
1538 | ||
9f95a23c TL |
1539 | for (auto i = warm_in.begin(); i != warm_in.end(); ++i) { |
1540 | s += i->length; | |
7c673cae | 1541 | } |
7c673cae | 1542 | |
9f95a23c TL |
1543 | uint64_t warm_in_bytes = s - hot_bytes; |
1544 | if (warm_in_bytes != list_bytes[BUFFER_WARM_IN]) { | |
1545 | derr << __func__ << " warm_in_list_bytes " | |
1546 | << list_bytes[BUFFER_WARM_IN] | |
1547 | << " != actual " << warm_in_bytes | |
1548 | << dendl; | |
1549 | ceph_assert(warm_in_bytes == list_bytes[BUFFER_WARM_IN]); | |
7c673cae | 1550 | } |
7c673cae | 1551 | |
9f95a23c TL |
1552 | if (s != buffer_bytes) { |
1553 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1554 | << dendl; | |
1555 | ceph_assert(s == buffer_bytes); | |
1556 | } | |
7c673cae | 1557 | |
9f95a23c TL |
1558 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1559 | << " ok" << dendl; | |
7c673cae | 1560 | } |
9f95a23c TL |
1561 | #endif |
1562 | }; | |
7c673cae | 1563 | |
9f95a23c | 1564 | // BuferCacheShard |
7c673cae | 1565 | |
9f95a23c TL |
1566 | BlueStore::BufferCacheShard *BlueStore::BufferCacheShard::create( |
1567 | CephContext* cct, | |
1568 | string type, | |
1569 | PerfCounters *logger) | |
1570 | { | |
1571 | BufferCacheShard *c = nullptr; | |
1572 | if (type == "lru") | |
1573 | c = new LruBufferCacheShard(cct); | |
1574 | else if (type == "2q") | |
1575 | c = new TwoQBufferCacheShard(cct); | |
1576 | else | |
1577 | ceph_abort_msg("unrecognized cache type"); | |
1578 | c->logger = logger; | |
1579 | return c; | |
7c673cae | 1580 | } |
7c673cae FG |
1581 | |
1582 | // BufferSpace | |
1583 | ||
1584 | #undef dout_prefix | |
1585 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1586 | ||
9f95a23c | 1587 | void BlueStore::BufferSpace::_clear(BufferCacheShard* cache) |
7c673cae FG |
1588 | { |
1589 | // note: we already hold cache->lock | |
1590 | ldout(cache->cct, 20) << __func__ << dendl; | |
1591 | while (!buffer_map.empty()) { | |
1592 | _rm_buffer(cache, buffer_map.begin()); | |
1593 | } | |
1594 | } | |
1595 | ||
9f95a23c | 1596 | int BlueStore::BufferSpace::_discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) |
7c673cae FG |
1597 | { |
1598 | // note: we already hold cache->lock | |
1599 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1600 | << std::dec << dendl; | |
1601 | int cache_private = 0; | |
1602 | cache->_audit("discard start"); | |
1603 | auto i = _data_lower_bound(offset); | |
1604 | uint32_t end = offset + length; | |
1605 | while (i != buffer_map.end()) { | |
1606 | Buffer *b = i->second.get(); | |
1607 | if (b->offset >= end) { | |
1608 | break; | |
1609 | } | |
1610 | if (b->cache_private > cache_private) { | |
1611 | cache_private = b->cache_private; | |
1612 | } | |
1613 | if (b->offset < offset) { | |
1614 | int64_t front = offset - b->offset; | |
1615 | if (b->end() > end) { | |
1616 | // drop middle (split) | |
1617 | uint32_t tail = b->end() - end; | |
1618 | if (b->data.length()) { | |
1619 | bufferlist bl; | |
1620 | bl.substr_of(b->data, b->length - tail, tail); | |
f67539c2 | 1621 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl, b->flags); |
31f18b77 FG |
1622 | nb->maybe_rebuild(); |
1623 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1624 | } else { |
f67539c2 TL |
1625 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail, |
1626 | b->flags), | |
1627 | 0, b); | |
7c673cae FG |
1628 | } |
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 | cache->_audit("discard end 1"); |
1635 | break; | |
1636 | } else { | |
1637 | // drop tail | |
1638 | if (!b->is_writing()) { | |
9f95a23c | 1639 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1640 | } |
1641 | b->truncate(front); | |
31f18b77 | 1642 | b->maybe_rebuild(); |
7c673cae FG |
1643 | ++i; |
1644 | continue; | |
1645 | } | |
1646 | } | |
1647 | if (b->end() <= end) { | |
1648 | // drop entire buffer | |
1649 | _rm_buffer(cache, i++); | |
1650 | continue; | |
1651 | } | |
1652 | // drop front | |
1653 | uint32_t keep = b->end() - end; | |
1654 | if (b->data.length()) { | |
1655 | bufferlist bl; | |
1656 | bl.substr_of(b->data, b->length - keep, keep); | |
f67539c2 | 1657 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl, b->flags); |
31f18b77 FG |
1658 | nb->maybe_rebuild(); |
1659 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1660 | } else { |
f67539c2 TL |
1661 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep, |
1662 | b->flags), | |
1663 | 0, b); | |
7c673cae FG |
1664 | } |
1665 | _rm_buffer(cache, i); | |
1666 | cache->_audit("discard end 2"); | |
1667 | break; | |
1668 | } | |
1669 | return cache_private; | |
1670 | } | |
1671 | ||
1672 | void BlueStore::BufferSpace::read( | |
9f95a23c | 1673 | BufferCacheShard* cache, |
224ce89b WB |
1674 | uint32_t offset, |
1675 | uint32_t length, | |
7c673cae | 1676 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1677 | interval_set<uint32_t>& res_intervals, |
1678 | int flags) | |
7c673cae | 1679 | { |
7c673cae FG |
1680 | res.clear(); |
1681 | res_intervals.clear(); | |
1682 | uint32_t want_bytes = length; | |
1683 | uint32_t end = offset + length; | |
224ce89b WB |
1684 | |
1685 | { | |
11fdf7f2 | 1686 | std::lock_guard l(cache->lock); |
224ce89b WB |
1687 | for (auto i = _data_lower_bound(offset); |
1688 | i != buffer_map.end() && offset < end && i->first < end; | |
1689 | ++i) { | |
1690 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1691 | ceph_assert(b->end() > offset); |
91327a77 AA |
1692 | |
1693 | bool val = false; | |
1694 | if (flags & BYPASS_CLEAN_CACHE) | |
1695 | val = b->is_writing(); | |
1696 | else | |
1697 | val = b->is_writing() || b->is_clean(); | |
1698 | if (val) { | |
224ce89b WB |
1699 | if (b->offset < offset) { |
1700 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1701 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1702 | res[offset].substr_of(b->data, skip, l); |
1703 | res_intervals.insert(offset, l); | |
1704 | offset += l; | |
1705 | length -= l; | |
1706 | if (!b->is_writing()) { | |
9f95a23c | 1707 | cache->_touch(b); |
f67539c2 | 1708 | } |
224ce89b WB |
1709 | continue; |
1710 | } | |
1711 | if (b->offset > offset) { | |
1712 | uint32_t gap = b->offset - offset; | |
1713 | if (length <= gap) { | |
1714 | break; | |
1715 | } | |
1716 | offset += gap; | |
1717 | length -= gap; | |
1718 | } | |
1719 | if (!b->is_writing()) { | |
9f95a23c | 1720 | cache->_touch(b); |
224ce89b WB |
1721 | } |
1722 | if (b->length > length) { | |
1723 | res[offset].substr_of(b->data, 0, length); | |
1724 | res_intervals.insert(offset, length); | |
7c673cae | 1725 | break; |
224ce89b WB |
1726 | } else { |
1727 | res[offset].append(b->data); | |
1728 | res_intervals.insert(offset, b->length); | |
1729 | if (b->length == length) | |
1730 | break; | |
1731 | offset += b->length; | |
1732 | length -= b->length; | |
1733 | } | |
7c673cae FG |
1734 | } |
1735 | } | |
1736 | } | |
1737 | ||
1738 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1739 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1740 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1741 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1742 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1743 | } | |
1744 | ||
9f95a23c | 1745 | void BlueStore::BufferSpace::_finish_write(BufferCacheShard* cache, uint64_t seq) |
7c673cae | 1746 | { |
7c673cae FG |
1747 | auto i = writing.begin(); |
1748 | while (i != writing.end()) { | |
1749 | if (i->seq > seq) { | |
1750 | break; | |
1751 | } | |
1752 | if (i->seq < seq) { | |
1753 | ++i; | |
1754 | continue; | |
1755 | } | |
1756 | ||
1757 | Buffer *b = &*i; | |
11fdf7f2 | 1758 | ceph_assert(b->is_writing()); |
7c673cae FG |
1759 | |
1760 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1761 | writing.erase(i++); | |
1762 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1763 | buffer_map.erase(b->offset); | |
1764 | } else { | |
1765 | b->state = Buffer::STATE_CLEAN; | |
1766 | writing.erase(i++); | |
31f18b77 FG |
1767 | b->maybe_rebuild(); |
1768 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
9f95a23c | 1769 | cache->_add(b, 1, nullptr); |
7c673cae FG |
1770 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; |
1771 | } | |
1772 | } | |
9f95a23c | 1773 | cache->_trim(); |
7c673cae FG |
1774 | cache->_audit("finish_write end"); |
1775 | } | |
1776 | ||
9f95a23c | 1777 | void BlueStore::BufferSpace::split(BufferCacheShard* cache, size_t pos, BlueStore::BufferSpace &r) |
7c673cae | 1778 | { |
11fdf7f2 | 1779 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1780 | if (buffer_map.empty()) |
1781 | return; | |
1782 | ||
1783 | auto p = --buffer_map.end(); | |
1784 | while (true) { | |
1785 | if (p->second->end() <= pos) | |
1786 | break; | |
1787 | ||
1788 | if (p->second->offset < pos) { | |
1789 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1790 | size_t left = pos - p->second->offset; | |
1791 | size_t right = p->second->length - left; | |
1792 | if (p->second->data.length()) { | |
1793 | bufferlist bl; | |
1794 | bl.substr_of(p->second->data, left, right); | |
f67539c2 TL |
1795 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, |
1796 | 0, bl, p->second->flags), | |
7c673cae FG |
1797 | 0, p->second.get()); |
1798 | } else { | |
f67539c2 TL |
1799 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, |
1800 | 0, right, p->second->flags), | |
7c673cae FG |
1801 | 0, p->second.get()); |
1802 | } | |
9f95a23c | 1803 | cache->_adjust_size(p->second.get(), -right); |
7c673cae FG |
1804 | p->second->truncate(left); |
1805 | break; | |
1806 | } | |
1807 | ||
11fdf7f2 | 1808 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1809 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1810 | if (p->second->data.length()) { | |
1811 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
f67539c2 | 1812 | p->second->offset - pos, p->second->data, p->second->flags), |
7c673cae FG |
1813 | 0, p->second.get()); |
1814 | } else { | |
1815 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
f67539c2 | 1816 | p->second->offset - pos, p->second->length, p->second->flags), |
7c673cae FG |
1817 | 0, p->second.get()); |
1818 | } | |
1819 | if (p == buffer_map.begin()) { | |
1820 | _rm_buffer(cache, p); | |
1821 | break; | |
1822 | } else { | |
1823 | _rm_buffer(cache, p--); | |
1824 | } | |
1825 | } | |
11fdf7f2 | 1826 | ceph_assert(writing.empty()); |
9f95a23c | 1827 | cache->_trim(); |
7c673cae FG |
1828 | } |
1829 | ||
1830 | // OnodeSpace | |
1831 | ||
1832 | #undef dout_prefix | |
1833 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1834 | ||
f6b5b4d7 TL |
1835 | BlueStore::OnodeRef BlueStore::OnodeSpace::add(const ghobject_t& oid, |
1836 | OnodeRef& o) | |
7c673cae | 1837 | { |
11fdf7f2 | 1838 | std::lock_guard l(cache->lock); |
7c673cae FG |
1839 | auto p = onode_map.find(oid); |
1840 | if (p != onode_map.end()) { | |
1841 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o | |
1842 | << " raced, returning existing " << p->second | |
1843 | << dendl; | |
1844 | return p->second; | |
1845 | } | |
f6b5b4d7 | 1846 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << o << dendl; |
7c673cae | 1847 | onode_map[oid] = o; |
f6b5b4d7 | 1848 | cache->_add(o.get(), 1); |
9f95a23c | 1849 | cache->_trim(); |
7c673cae FG |
1850 | return o; |
1851 | } | |
1852 | ||
f6b5b4d7 TL |
1853 | void BlueStore::OnodeSpace::_remove(const ghobject_t& oid) |
1854 | { | |
1855 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << dendl; | |
1856 | onode_map.erase(oid); | |
1857 | } | |
1858 | ||
7c673cae FG |
1859 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) |
1860 | { | |
7c673cae | 1861 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b WB |
1862 | OnodeRef o; |
1863 | bool hit = false; | |
1864 | ||
1865 | { | |
11fdf7f2 | 1866 | std::lock_guard l(cache->lock); |
224ce89b WB |
1867 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1868 | if (p == onode_map.end()) { | |
1869 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
1870 | } else { | |
1871 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
f6b5b4d7 TL |
1872 | << " " << p->second->nref |
1873 | << " " << p->second->cached | |
1874 | << " " << p->second->pinned | |
224ce89b | 1875 | << dendl; |
f6b5b4d7 TL |
1876 | // This will pin onode and implicitly touch the cache when Onode |
1877 | // eventually will become unpinned | |
224ce89b | 1878 | o = p->second; |
f6b5b4d7 TL |
1879 | ceph_assert(!o->cached || o->pinned); |
1880 | ||
1881 | hit = true; | |
224ce89b WB |
1882 | } |
1883 | } | |
1884 | ||
1885 | if (hit) { | |
1886 | cache->logger->inc(l_bluestore_onode_hits); | |
1887 | } else { | |
7c673cae | 1888 | cache->logger->inc(l_bluestore_onode_misses); |
7c673cae | 1889 | } |
224ce89b | 1890 | return o; |
7c673cae FG |
1891 | } |
1892 | ||
1893 | void BlueStore::OnodeSpace::clear() | |
1894 | { | |
11fdf7f2 | 1895 | std::lock_guard l(cache->lock); |
f6b5b4d7 | 1896 | ldout(cache->cct, 10) << __func__ << " " << onode_map.size()<< dendl; |
7c673cae | 1897 | for (auto &p : onode_map) { |
f6b5b4d7 | 1898 | cache->_rm(p.second.get()); |
7c673cae FG |
1899 | } |
1900 | onode_map.clear(); | |
1901 | } | |
1902 | ||
1903 | bool BlueStore::OnodeSpace::empty() | |
1904 | { | |
11fdf7f2 | 1905 | std::lock_guard l(cache->lock); |
7c673cae FG |
1906 | return onode_map.empty(); |
1907 | } | |
1908 | ||
1909 | void BlueStore::OnodeSpace::rename( | |
1910 | OnodeRef& oldo, | |
1911 | const ghobject_t& old_oid, | |
1912 | const ghobject_t& new_oid, | |
f91f0fd5 | 1913 | const mempool::bluestore_cache_meta::string& new_okey) |
7c673cae | 1914 | { |
11fdf7f2 | 1915 | std::lock_guard l(cache->lock); |
7c673cae FG |
1916 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
1917 | << dendl; | |
1918 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
1919 | po = onode_map.find(old_oid); | |
1920 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 1921 | ceph_assert(po != pn); |
7c673cae | 1922 | |
11fdf7f2 | 1923 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
1924 | if (pn != onode_map.end()) { |
1925 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
1926 | << dendl; | |
f6b5b4d7 | 1927 | cache->_rm(pn->second.get()); |
7c673cae FG |
1928 | onode_map.erase(pn); |
1929 | } | |
1930 | OnodeRef o = po->second; | |
1931 | ||
1932 | // install a non-existent onode at old location | |
1933 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
1934 | po->second = oldo; | |
f6b5b4d7 TL |
1935 | cache->_add(oldo.get(), 1); |
1936 | // add at new position and fix oid, key. | |
1937 | // This will pin 'o' and implicitly touch cache | |
1938 | // when it will eventually become unpinned | |
7c673cae | 1939 | onode_map.insert(make_pair(new_oid, o)); |
f6b5b4d7 TL |
1940 | ceph_assert(o->pinned); |
1941 | ||
7c673cae FG |
1942 | o->oid = new_oid; |
1943 | o->key = new_okey; | |
9f95a23c | 1944 | cache->_trim(); |
7c673cae FG |
1945 | } |
1946 | ||
adb31ebb | 1947 | bool BlueStore::OnodeSpace::map_any(std::function<bool(Onode*)> f) |
7c673cae | 1948 | { |
11fdf7f2 | 1949 | std::lock_guard l(cache->lock); |
7c673cae FG |
1950 | ldout(cache->cct, 20) << __func__ << dendl; |
1951 | for (auto& i : onode_map) { | |
adb31ebb | 1952 | if (f(i.second.get())) { |
7c673cae FG |
1953 | return true; |
1954 | } | |
1955 | } | |
1956 | return false; | |
1957 | } | |
1958 | ||
11fdf7f2 TL |
1959 | template <int LogLevelV = 30> |
1960 | void BlueStore::OnodeSpace::dump(CephContext *cct) | |
3efd9988 FG |
1961 | { |
1962 | for (auto& i : onode_map) { | |
f6b5b4d7 TL |
1963 | ldout(cct, LogLevelV) << i.first << " : " << i.second |
1964 | << " " << i.second->nref | |
1965 | << " " << i.second->cached | |
1966 | << " " << i.second->pinned | |
1967 | << dendl; | |
3efd9988 FG |
1968 | } |
1969 | } | |
7c673cae FG |
1970 | |
1971 | // SharedBlob | |
1972 | ||
1973 | #undef dout_prefix | |
1974 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
9f95a23c TL |
1975 | #undef dout_context |
1976 | #define dout_context coll->store->cct | |
7c673cae | 1977 | |
9f95a23c | 1978 | void BlueStore::SharedBlob::dump(Formatter* f) const |
7c673cae | 1979 | { |
9f95a23c TL |
1980 | f->dump_bool("loaded", loaded); |
1981 | if (loaded) { | |
1982 | persistent->dump(f); | |
1983 | } else { | |
1984 | f->dump_unsigned("sbid_unloaded", sbid_unloaded); | |
1985 | } | |
1986 | } | |
1987 | ||
1988 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
1989 | { | |
1990 | out << "SharedBlob(" << &sb; | |
1991 | ||
7c673cae FG |
1992 | if (sb.loaded) { |
1993 | out << " loaded " << *sb.persistent; | |
1994 | } else { | |
1995 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
1996 | } | |
1997 | return out << ")"; | |
1998 | } | |
1999 | ||
2000 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
2001 | : coll(_coll), sbid_unloaded(i) | |
2002 | { | |
11fdf7f2 | 2003 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
2004 | if (get_cache()) { |
2005 | get_cache()->add_blob(); | |
2006 | } | |
2007 | } | |
2008 | ||
2009 | BlueStore::SharedBlob::~SharedBlob() | |
2010 | { | |
7c673cae FG |
2011 | if (loaded && persistent) { |
2012 | delete persistent; | |
2013 | } | |
2014 | } | |
2015 | ||
2016 | void BlueStore::SharedBlob::put() | |
2017 | { | |
2018 | if (--nref == 0) { | |
9f95a23c TL |
2019 | dout(20) << __func__ << " " << this |
2020 | << " removing self from set " << get_parent() | |
2021 | << dendl; | |
1adf2230 AA |
2022 | again: |
2023 | auto coll_snap = coll; | |
2024 | if (coll_snap) { | |
11fdf7f2 | 2025 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
2026 | if (coll_snap != coll) { |
2027 | goto again; | |
2028 | } | |
91327a77 AA |
2029 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
2030 | // race with lookup | |
2031 | return; | |
2032 | } | |
1adf2230 AA |
2033 | bc._clear(coll_snap->cache); |
2034 | coll_snap->cache->rm_blob(); | |
7c673cae | 2035 | } |
28e407b8 | 2036 | delete this; |
7c673cae FG |
2037 | } |
2038 | } | |
2039 | ||
2040 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
2041 | { | |
11fdf7f2 | 2042 | ceph_assert(persistent); |
7c673cae FG |
2043 | persistent->ref_map.get(offset, length); |
2044 | } | |
2045 | ||
2046 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 2047 | PExtentVector *r, |
11fdf7f2 | 2048 | bool *unshare) |
7c673cae | 2049 | { |
11fdf7f2 TL |
2050 | ceph_assert(persistent); |
2051 | persistent->ref_map.put(offset, length, r, | |
2052 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
2053 | } |
2054 | ||
f64942e4 AA |
2055 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
2056 | { | |
2057 | while (true) { | |
9f95a23c | 2058 | BufferCacheShard *cache = coll->cache; |
11fdf7f2 | 2059 | std::lock_guard l(cache->lock); |
f64942e4 | 2060 | if (coll->cache != cache) { |
9f95a23c TL |
2061 | dout(20) << __func__ |
2062 | << " raced with sb cache update, was " << cache | |
2063 | << ", now " << coll->cache << ", retrying" | |
2064 | << dendl; | |
f64942e4 AA |
2065 | continue; |
2066 | } | |
2067 | bc._finish_write(cache, seq); | |
2068 | break; | |
2069 | } | |
2070 | } | |
2071 | ||
3efd9988 FG |
2072 | // SharedBlobSet |
2073 | ||
2074 | #undef dout_prefix | |
2075 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
2076 | ||
11fdf7f2 TL |
2077 | template <int LogLevelV = 30> |
2078 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 2079 | { |
11fdf7f2 | 2080 | std::lock_guard l(lock); |
3efd9988 | 2081 | for (auto& i : sb_map) { |
11fdf7f2 | 2082 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
2083 | } |
2084 | } | |
2085 | ||
7c673cae FG |
2086 | // Blob |
2087 | ||
2088 | #undef dout_prefix | |
2089 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
2090 | ||
9f95a23c TL |
2091 | void BlueStore::Blob::dump(Formatter* f) const |
2092 | { | |
2093 | if (is_spanning()) { | |
2094 | f->dump_unsigned("spanning_id ", id); | |
2095 | } | |
2096 | blob.dump(f); | |
2097 | if (shared_blob) { | |
2098 | f->dump_object("shared", *shared_blob); | |
2099 | } | |
2100 | } | |
2101 | ||
7c673cae FG |
2102 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) |
2103 | { | |
2104 | out << "Blob(" << &b; | |
2105 | if (b.is_spanning()) { | |
2106 | out << " spanning " << b.id; | |
2107 | } | |
35e4c445 FG |
2108 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
2109 | if (b.shared_blob) { | |
2110 | out << " " << *b.shared_blob; | |
2111 | } else { | |
2112 | out << " (shared_blob=NULL)"; | |
2113 | } | |
2114 | out << ")"; | |
7c673cae FG |
2115 | return out; |
2116 | } | |
2117 | ||
2118 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
2119 | { | |
224ce89b | 2120 | if (get_blob().is_shared()) { |
7c673cae FG |
2121 | return; |
2122 | } | |
224ce89b | 2123 | if (get_blob().is_compressed()) { |
7c673cae FG |
2124 | bool discard = false; |
2125 | bool all_invalid = true; | |
224ce89b | 2126 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
2127 | if (!e.is_valid()) { |
2128 | discard = true; | |
2129 | } else { | |
2130 | all_invalid = false; | |
2131 | } | |
2132 | } | |
11fdf7f2 | 2133 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
2134 | // or none pextents are invalid. |
2135 | if (discard) { | |
224ce89b WB |
2136 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
2137 | get_blob().get_logical_length()); | |
7c673cae FG |
2138 | } |
2139 | } else { | |
2140 | size_t pos = 0; | |
224ce89b | 2141 | for (auto e : get_blob().get_extents()) { |
7c673cae | 2142 | if (!e.is_valid()) { |
9f95a23c TL |
2143 | dout(20) << __func__ << " 0x" << std::hex << pos |
2144 | << "~" << e.length | |
2145 | << std::dec << dendl; | |
7c673cae FG |
2146 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); |
2147 | } | |
2148 | pos += e.length; | |
2149 | } | |
224ce89b WB |
2150 | if (get_blob().can_prune_tail()) { |
2151 | dirty_blob().prune_tail(); | |
2152 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
224ce89b | 2153 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
2154 | } |
2155 | } | |
2156 | } | |
2157 | ||
2158 | void BlueStore::Blob::get_ref( | |
2159 | Collection *coll, | |
2160 | uint32_t offset, | |
2161 | uint32_t length) | |
2162 | { | |
2163 | // Caller has to initialize Blob's logical length prior to increment | |
2164 | // references. Otherwise one is neither unable to determine required | |
2165 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
2166 | // for single counter mode. | |
11fdf7f2 | 2167 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
2168 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2169 | << std::dec << " " << *this << dendl; | |
2170 | ||
2171 | if (used_in_blob.is_empty()) { | |
2172 | uint32_t min_release_size = | |
224ce89b WB |
2173 | get_blob().get_release_size(coll->store->min_alloc_size); |
2174 | uint64_t l = get_blob().get_logical_length(); | |
2175 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
2176 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
2177 | used_in_blob.init(l, min_release_size); |
2178 | } | |
2179 | used_in_blob.get( | |
2180 | offset, | |
2181 | length); | |
2182 | } | |
2183 | ||
2184 | bool BlueStore::Blob::put_ref( | |
2185 | Collection *coll, | |
2186 | uint32_t offset, | |
2187 | uint32_t length, | |
2188 | PExtentVector *r) | |
2189 | { | |
2190 | PExtentVector logical; | |
2191 | ||
7c673cae FG |
2192 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2193 | << std::dec << " " << *this << dendl; | |
2194 | ||
2195 | bool empty = used_in_blob.put( | |
2196 | offset, | |
2197 | length, | |
2198 | &logical); | |
2199 | r->clear(); | |
2200 | // nothing to release | |
2201 | if (!empty && logical.empty()) { | |
2202 | return false; | |
2203 | } | |
2204 | ||
2205 | bluestore_blob_t& b = dirty_blob(); | |
2206 | return b.release_extents(empty, logical, r); | |
2207 | } | |
2208 | ||
224ce89b | 2209 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
2210 | uint32_t target_blob_size, |
2211 | uint32_t b_offset, | |
2212 | uint32_t *length0) { | |
11fdf7f2 TL |
2213 | ceph_assert(min_alloc_size); |
2214 | ceph_assert(target_blob_size); | |
7c673cae FG |
2215 | if (!get_blob().is_mutable()) { |
2216 | return false; | |
2217 | } | |
2218 | ||
2219 | uint32_t length = *length0; | |
2220 | uint32_t end = b_offset + length; | |
2221 | ||
2222 | // Currently for the sake of simplicity we omit blob reuse if data is | |
2223 | // unaligned with csum chunk. Later we can perform padding if needed. | |
2224 | if (get_blob().has_csum() && | |
2225 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
2226 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
2227 | return false; | |
2228 | } | |
2229 | ||
2230 | auto blen = get_blob().get_logical_length(); | |
2231 | uint32_t new_blen = blen; | |
2232 | ||
2233 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 2234 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
2235 | |
2236 | if (b_offset >= blen) { | |
224ce89b WB |
2237 | // new data totally stands out of the existing blob |
2238 | new_blen = end; | |
7c673cae | 2239 | } else { |
224ce89b | 2240 | // new data overlaps with the existing blob |
11fdf7f2 | 2241 | new_blen = std::max(blen, end); |
224ce89b WB |
2242 | |
2243 | uint32_t overlap = 0; | |
2244 | if (new_blen > blen) { | |
2245 | overlap = blen - b_offset; | |
2246 | } else { | |
2247 | overlap = length; | |
2248 | } | |
2249 | ||
2250 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
2251 | // abort if any piece of the overlap has already been allocated | |
2252 | return false; | |
7c673cae FG |
2253 | } |
2254 | } | |
224ce89b | 2255 | |
7c673cae FG |
2256 | if (new_blen > blen) { |
2257 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
2258 | // Unable to decrease the provided length to fit into max_blob_size | |
2259 | if (overflow >= length) { | |
2260 | return false; | |
2261 | } | |
2262 | ||
2263 | // FIXME: in some cases we could reduce unused resolution | |
2264 | if (get_blob().has_unused()) { | |
2265 | return false; | |
2266 | } | |
2267 | ||
2268 | if (overflow > 0) { | |
2269 | new_blen -= overflow; | |
2270 | length -= overflow; | |
2271 | *length0 = length; | |
2272 | } | |
224ce89b | 2273 | |
7c673cae FG |
2274 | if (new_blen > blen) { |
2275 | dirty_blob().add_tail(new_blen); | |
2276 | used_in_blob.add_tail(new_blen, | |
224ce89b | 2277 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
2278 | } |
2279 | } | |
2280 | return true; | |
2281 | } | |
2282 | ||
2283 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
2284 | { | |
7c673cae FG |
2285 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec |
2286 | << " start " << *this << dendl; | |
11fdf7f2 TL |
2287 | ceph_assert(blob.can_split()); |
2288 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
2289 | bluestore_blob_t &lb = dirty_blob(); |
2290 | bluestore_blob_t &rb = r->dirty_blob(); | |
2291 | ||
2292 | used_in_blob.split( | |
2293 | blob_offset, | |
2294 | &(r->used_in_blob)); | |
2295 | ||
2296 | lb.split(blob_offset, rb); | |
2297 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2298 | ||
2299 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2300 | << " finish " << *this << dendl; | |
2301 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2302 | << " and " << *r << dendl; | |
2303 | } | |
2304 | ||
2305 | #ifndef CACHE_BLOB_BL | |
2306 | void BlueStore::Blob::decode( | |
2307 | Collection *coll, | |
11fdf7f2 | 2308 | bufferptr::const_iterator& p, |
7c673cae FG |
2309 | uint64_t struct_v, |
2310 | uint64_t* sbid, | |
2311 | bool include_ref_map) | |
2312 | { | |
2313 | denc(blob, p, struct_v); | |
2314 | if (blob.is_shared()) { | |
2315 | denc(*sbid, p); | |
2316 | } | |
2317 | if (include_ref_map) { | |
2318 | if (struct_v > 1) { | |
2319 | used_in_blob.decode(p); | |
2320 | } else { | |
2321 | used_in_blob.clear(); | |
2322 | bluestore_extent_ref_map_t legacy_ref_map; | |
2323 | legacy_ref_map.decode(p); | |
2324 | for (auto r : legacy_ref_map.ref_map) { | |
2325 | get_ref( | |
2326 | coll, | |
2327 | r.first, | |
2328 | r.second.refs * r.second.length); | |
2329 | } | |
2330 | } | |
2331 | } | |
2332 | } | |
2333 | #endif | |
2334 | ||
2335 | // Extent | |
2336 | ||
9f95a23c TL |
2337 | void BlueStore::Extent::dump(Formatter* f) const |
2338 | { | |
2339 | f->dump_unsigned("logical_offset", logical_offset); | |
2340 | f->dump_unsigned("length", length); | |
2341 | f->dump_unsigned("blob_offset", blob_offset); | |
2342 | f->dump_object("blob", *blob); | |
2343 | } | |
2344 | ||
7c673cae FG |
2345 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) |
2346 | { | |
2347 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2348 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2349 | << " " << *e.blob; | |
2350 | } | |
2351 | ||
2352 | // OldExtent | |
2353 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2354 | uint32_t lo, | |
2355 | uint32_t o, | |
2356 | uint32_t l, | |
2357 | BlobRef& b) { | |
2358 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2359 | b->put_ref(c.get(), o, l, &(oe->r)); | |
adb31ebb | 2360 | oe->blob_empty = !b->is_referenced(); |
7c673cae FG |
2361 | return oe; |
2362 | } | |
2363 | ||
2364 | // ExtentMap | |
2365 | ||
2366 | #undef dout_prefix | |
2367 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
9f95a23c TL |
2368 | #undef dout_context |
2369 | #define dout_context onode->c->store->cct | |
7c673cae FG |
2370 | |
2371 | BlueStore::ExtentMap::ExtentMap(Onode *o) | |
2372 | : onode(o), | |
2373 | inline_bl( | |
2374 | o->c->store->cct->_conf->bluestore_extent_map_inline_shard_prealloc_size) { | |
2375 | } | |
2376 | ||
9f95a23c TL |
2377 | void BlueStore::ExtentMap::dump(Formatter* f) const |
2378 | { | |
2379 | f->open_array_section("extents"); | |
2380 | ||
2381 | for (auto& e : extent_map) { | |
2382 | f->dump_object("extent", e); | |
2383 | } | |
2384 | f->close_section(); | |
2385 | } | |
2386 | ||
11fdf7f2 TL |
2387 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2388 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2389 | uint64_t& length, uint64_t& dstoff) { | |
2390 | ||
2391 | auto cct = onode->c->store->cct; | |
2392 | bool inject_21040 = | |
2393 | cct->_conf->bluestore_debug_inject_bug21040; | |
2394 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2395 | for (auto& e : oldo->extent_map.extent_map) { | |
2396 | e.blob->last_encoded_id = -1; | |
2397 | } | |
2398 | ||
2399 | int n = 0; | |
2400 | uint64_t end = srcoff + length; | |
2401 | uint32_t dirty_range_begin = 0; | |
2402 | uint32_t dirty_range_end = 0; | |
2403 | bool src_dirty = false; | |
2404 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2405 | ep != oldo->extent_map.extent_map.end(); | |
2406 | ++ep) { | |
2407 | auto& e = *ep; | |
2408 | if (e.logical_offset >= end) { | |
2409 | break; | |
2410 | } | |
2411 | dout(20) << __func__ << " src " << e << dendl; | |
2412 | BlobRef cb; | |
2413 | bool blob_duped = true; | |
2414 | if (e.blob->last_encoded_id >= 0) { | |
2415 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2416 | blob_duped = false; | |
2417 | } else { | |
2418 | // dup the blob | |
2419 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2420 | // make sure it is shared | |
2421 | if (!blob.is_shared()) { | |
2422 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2423 | if (!inject_21040 && !src_dirty) { | |
2424 | src_dirty = true; | |
2425 | dirty_range_begin = e.logical_offset; | |
2426 | } else if (inject_21040 && | |
2427 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2428 | dirty_range_begin = e.logical_offset; | |
2429 | } | |
2430 | ceph_assert(e.logical_end() > 0); | |
2431 | // -1 to exclude next potential shard | |
2432 | dirty_range_end = e.logical_end() - 1; | |
2433 | } else { | |
2434 | c->load_shared_blob(e.blob->shared_blob); | |
2435 | } | |
2436 | cb = new Blob(); | |
2437 | e.blob->last_encoded_id = n; | |
2438 | id_to_blob[n] = cb; | |
2439 | e.blob->dup(*cb); | |
2440 | // bump the extent refs on the copied blob's extents | |
2441 | for (auto p : blob.get_extents()) { | |
2442 | if (p.is_valid()) { | |
2443 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2444 | } | |
2445 | } | |
2446 | txc->write_shared_blob(e.blob->shared_blob); | |
2447 | dout(20) << __func__ << " new " << *cb << dendl; | |
2448 | } | |
2449 | ||
2450 | int skip_front, skip_back; | |
2451 | if (e.logical_offset < srcoff) { | |
2452 | skip_front = srcoff - e.logical_offset; | |
2453 | } else { | |
2454 | skip_front = 0; | |
2455 | } | |
2456 | if (e.logical_end() > end) { | |
2457 | skip_back = e.logical_end() - end; | |
2458 | } else { | |
2459 | skip_back = 0; | |
2460 | } | |
2461 | ||
2462 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2463 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2464 | newo->extent_map.extent_map.insert(*ne); | |
2465 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2466 | // fixme: we may leave parts of new blob unreferenced that could | |
2467 | // be freed (relative to the shared_blob). | |
2468 | txc->statfs_delta.stored() += ne->length; | |
2469 | if (e.blob->get_blob().is_compressed()) { | |
2470 | txc->statfs_delta.compressed_original() += ne->length; | |
2471 | if (blob_duped) { | |
2472 | txc->statfs_delta.compressed() += | |
2473 | cb->get_blob().get_compressed_payload_length(); | |
2474 | } | |
2475 | } | |
2476 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2477 | ++n; | |
2478 | } | |
2479 | if ((!inject_21040 && src_dirty) || | |
2480 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2481 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2482 | dirty_range_end - dirty_range_begin); | |
2483 | txc->write_onode(oldo); | |
2484 | } | |
2485 | txc->write_onode(newo); | |
2486 | ||
2487 | if (dstoff + length > newo->onode.size) { | |
2488 | newo->onode.size = dstoff + length; | |
2489 | } | |
2490 | newo->extent_map.dirty_range(dstoff, length); | |
2491 | } | |
7c673cae FG |
2492 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2493 | bool force) | |
2494 | { | |
2495 | auto cct = onode->c->store->cct; //used by dout | |
2496 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2497 | if (onode->onode.extent_map_shards.empty()) { | |
2498 | if (inline_bl.length() == 0) { | |
2499 | unsigned n; | |
2500 | // we need to encode inline_bl to measure encoded length | |
2501 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
f91f0fd5 | 2502 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_inline_bl); |
11fdf7f2 | 2503 | ceph_assert(!never_happen); |
7c673cae FG |
2504 | size_t len = inline_bl.length(); |
2505 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2506 | << " extents" << dendl; | |
2507 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2508 | request_reshard(0, OBJECT_MAX_SIZE); | |
2509 | return; | |
2510 | } | |
2511 | } | |
2512 | // will persist in the onode key. | |
2513 | } else { | |
2514 | // pending shard update | |
2515 | struct dirty_shard_t { | |
2516 | Shard *shard; | |
2517 | bufferlist bl; | |
2518 | dirty_shard_t(Shard *s) : shard(s) {} | |
2519 | }; | |
2520 | vector<dirty_shard_t> encoded_shards; | |
2521 | // allocate slots for all shards in a single call instead of | |
2522 | // doing multiple allocations - one per each dirty shard | |
2523 | encoded_shards.reserve(shards.size()); | |
2524 | ||
2525 | auto p = shards.begin(); | |
2526 | auto prev_p = p; | |
2527 | while (p != shards.end()) { | |
11fdf7f2 | 2528 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2529 | auto n = p; |
2530 | ++n; | |
2531 | if (p->dirty) { | |
2532 | uint32_t endoff; | |
2533 | if (n == shards.end()) { | |
2534 | endoff = OBJECT_MAX_SIZE; | |
2535 | } else { | |
2536 | endoff = n->shard_info->offset; | |
2537 | } | |
2538 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2539 | bufferlist& bl = encoded_shards.back().bl; | |
2540 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2541 | bl, &p->extents)) { | |
2542 | if (force) { | |
2543 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2544 | ceph_assert(!force); |
7c673cae FG |
2545 | } |
2546 | } | |
2547 | size_t len = bl.length(); | |
2548 | ||
2549 | dout(20) << __func__ << " shard 0x" << std::hex | |
2550 | << p->shard_info->offset << std::dec << " is " << len | |
2551 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2552 | << p->extents << " extents" << dendl; | |
2553 | ||
2554 | if (!force) { | |
2555 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2556 | // we are big; reshard ourselves | |
2557 | request_reshard(p->shard_info->offset, endoff); | |
2558 | } | |
2559 | // avoid resharding the trailing shard, even if it is small | |
2560 | else if (n != shards.end() && | |
11fdf7f2 TL |
2561 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2562 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2563 | if (p == shards.begin()) { |
2564 | // we are the first shard, combine with next shard | |
7c673cae | 2565 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2566 | } else { |
31f18b77 FG |
2567 | // combine either with the previous shard or the next, |
2568 | // whichever is smaller | |
7c673cae FG |
2569 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2570 | request_reshard(p->shard_info->offset, endoff + 1); | |
2571 | } else { | |
2572 | request_reshard(prev_p->shard_info->offset, endoff); | |
2573 | } | |
2574 | } | |
2575 | } | |
2576 | } | |
2577 | } | |
2578 | prev_p = p; | |
2579 | p = n; | |
2580 | } | |
2581 | if (needs_reshard()) { | |
2582 | return; | |
2583 | } | |
2584 | ||
2585 | // schedule DB update for dirty shards | |
2586 | string key; | |
2587 | for (auto& it : encoded_shards) { | |
2588 | it.shard->dirty = false; | |
2589 | it.shard->shard_info->bytes = it.bl.length(); | |
2590 | generate_extent_shard_key_and_apply( | |
2591 | onode->key, | |
2592 | it.shard->shard_info->offset, | |
2593 | &key, | |
2594 | [&](const string& final_key) { | |
2595 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2596 | } | |
2597 | ); | |
2598 | } | |
2599 | } | |
2600 | } | |
2601 | ||
31f18b77 FG |
2602 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2603 | { | |
2604 | if (spanning_blob_map.empty()) | |
2605 | return 0; | |
2606 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2607 | // bid is valid and available. | |
2608 | if (bid >= 0) | |
2609 | return bid; | |
2610 | // Find next unused bid; | |
2611 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2612 | const auto begin_bid = bid; | |
2613 | do { | |
2614 | if (!spanning_blob_map.count(bid)) | |
2615 | return bid; | |
2616 | else { | |
2617 | bid++; | |
2618 | if (bid < 0) bid = 0; | |
2619 | } | |
2620 | } while (bid != begin_bid); | |
81eedcae TL |
2621 | auto cct = onode->c->store->cct; // used by dout |
2622 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2623 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2624 | } |
2625 | ||
7c673cae FG |
2626 | void BlueStore::ExtentMap::reshard( |
2627 | KeyValueDB *db, | |
2628 | KeyValueDB::Transaction t) | |
2629 | { | |
2630 | auto cct = onode->c->store->cct; // used by dout | |
2631 | ||
2632 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2633 | << needs_reshard_end << ")" << std::dec | |
2634 | << " of " << onode->onode.extent_map_shards.size() | |
2635 | << " shards on " << onode->oid << dendl; | |
2636 | for (auto& p : spanning_blob_map) { | |
2637 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2638 | << dendl; | |
2639 | } | |
2640 | // determine shard index range | |
2641 | unsigned si_begin = 0, si_end = 0; | |
2642 | if (!shards.empty()) { | |
2643 | while (si_begin + 1 < shards.size() && | |
2644 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2645 | ++si_begin; | |
2646 | } | |
2647 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2648 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2649 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2650 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2651 | break; | |
2652 | } | |
2653 | } | |
2654 | if (si_end == shards.size()) { | |
2655 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2656 | } | |
2657 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2658 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2659 | << needs_reshard_end << ")" << std::dec << dendl; | |
2660 | } | |
2661 | ||
181888fb | 2662 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2663 | |
2664 | // we may need to fault in a larger interval later must have all | |
2665 | // referring extents for spanning blobs loaded in order to have | |
2666 | // accurate use_tracker values. | |
2667 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2668 | uint32_t spanning_scan_end = needs_reshard_end; | |
2669 | ||
2670 | // remove old keys | |
2671 | string key; | |
2672 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2673 | generate_extent_shard_key_and_apply( | |
2674 | onode->key, shards[i].shard_info->offset, &key, | |
2675 | [&](const string& final_key) { | |
2676 | t->rmkey(PREFIX_OBJ, final_key); | |
2677 | } | |
2678 | ); | |
2679 | } | |
2680 | ||
2681 | // calculate average extent size | |
2682 | unsigned bytes = 0; | |
2683 | unsigned extents = 0; | |
2684 | if (onode->onode.extent_map_shards.empty()) { | |
2685 | bytes = inline_bl.length(); | |
2686 | extents = extent_map.size(); | |
2687 | } else { | |
2688 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2689 | bytes += shards[i].shard_info->bytes; | |
2690 | extents += shards[i].extents; | |
2691 | } | |
2692 | } | |
2693 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2694 | unsigned slop = target * | |
2695 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2696 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2697 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2698 | << ", slop " << slop << dendl; | |
2699 | ||
2700 | // reshard | |
2701 | unsigned estimate = 0; | |
31f18b77 | 2702 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2703 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2704 | unsigned max_blob_end = 0; | |
2705 | Extent dummy(needs_reshard_begin); | |
2706 | for (auto e = extent_map.lower_bound(dummy); | |
2707 | e != extent_map.end(); | |
2708 | ++e) { | |
2709 | if (e->logical_offset >= needs_reshard_end) { | |
2710 | break; | |
2711 | } | |
2712 | dout(30) << " extent " << *e << dendl; | |
2713 | ||
2714 | // disfavor shard boundaries that span a blob | |
2715 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2716 | if (estimate && | |
2717 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2718 | // new shard | |
31f18b77 | 2719 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2720 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2721 | new_shard_info.back().offset = offset; | |
2722 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2723 | << std::dec << dendl; |
7c673cae FG |
2724 | } |
2725 | offset = e->logical_offset; | |
2726 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2727 | new_shard_info.back().offset = offset; | |
2728 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2729 | << std::dec << dendl; | |
2730 | estimate = 0; | |
2731 | } | |
2732 | estimate += extent_avg; | |
31f18b77 FG |
2733 | unsigned bs = e->blob_start(); |
2734 | if (bs < spanning_scan_begin) { | |
2735 | spanning_scan_begin = bs; | |
7c673cae FG |
2736 | } |
2737 | uint32_t be = e->blob_end(); | |
2738 | if (be > max_blob_end) { | |
2739 | max_blob_end = be; | |
2740 | } | |
2741 | if (be > spanning_scan_end) { | |
2742 | spanning_scan_end = be; | |
2743 | } | |
2744 | } | |
2745 | if (new_shard_info.empty() && (si_begin > 0 || | |
2746 | si_end < shards.size())) { | |
2747 | // we resharded a partial range; we must produce at least one output | |
2748 | // shard | |
2749 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2750 | new_shard_info.back().offset = needs_reshard_begin; | |
2751 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2752 | << std::dec << " (singleton degenerate case)" << dendl; | |
2753 | } | |
2754 | ||
2755 | auto& sv = onode->onode.extent_map_shards; | |
2756 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2757 | dout(20) << __func__ << " old " << sv << dendl; | |
2758 | if (sv.empty()) { | |
2759 | // no old shards to keep | |
2760 | sv.swap(new_shard_info); | |
2761 | init_shards(true, true); | |
2762 | } else { | |
2763 | // splice in new shards | |
2764 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2765 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2766 | sv.insert( | |
2767 | sv.begin() + si_begin, | |
2768 | new_shard_info.begin(), | |
2769 | new_shard_info.end()); | |
2770 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2771 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2772 | |
11fdf7f2 | 2773 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2774 | |
2775 | // note that we need to update every shard_info of shards here, | |
2776 | // as sv might have been totally re-allocated above | |
2777 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2778 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2779 | } |
2780 | ||
2781 | // mark newly added shards as dirty | |
2782 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2783 | shards[i].loaded = true; |
2784 | shards[i].dirty = true; | |
2785 | } | |
7c673cae FG |
2786 | } |
2787 | dout(20) << __func__ << " fin " << sv << dendl; | |
2788 | inline_bl.clear(); | |
2789 | ||
2790 | if (sv.empty()) { | |
2791 | // no more shards; unspan all previously spanning blobs | |
2792 | auto p = spanning_blob_map.begin(); | |
2793 | while (p != spanning_blob_map.end()) { | |
2794 | p->second->id = -1; | |
2795 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2796 | p = spanning_blob_map.erase(p); | |
2797 | } | |
2798 | } else { | |
2799 | // identify new spanning blobs | |
2800 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2801 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2802 | if (spanning_scan_begin < needs_reshard_begin) { | |
2803 | fault_range(db, spanning_scan_begin, | |
2804 | needs_reshard_begin - spanning_scan_begin); | |
2805 | } | |
2806 | if (spanning_scan_end > needs_reshard_end) { | |
2807 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2808 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2809 | } |
2810 | auto sp = sv.begin() + si_begin; | |
2811 | auto esp = sv.end(); | |
2812 | unsigned shard_start = sp->offset; | |
2813 | unsigned shard_end; | |
2814 | ++sp; | |
2815 | if (sp == esp) { | |
2816 | shard_end = OBJECT_MAX_SIZE; | |
2817 | } else { | |
2818 | shard_end = sp->offset; | |
2819 | } | |
7c673cae | 2820 | Extent dummy(needs_reshard_begin); |
9f95a23c TL |
2821 | |
2822 | bool was_too_many_blobs_check = false; | |
2823 | auto too_many_blobs_threshold = | |
2824 | g_conf()->bluestore_debug_too_many_blobs_threshold; | |
2825 | auto& dumped_onodes = onode->c->onode_map.cache->dumped_onodes; | |
2826 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oid_slot = nullptr; | |
2827 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oldest_slot = nullptr; | |
2828 | ||
7c673cae FG |
2829 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { |
2830 | if (e->logical_offset >= needs_reshard_end) { | |
2831 | break; | |
2832 | } | |
2833 | dout(30) << " extent " << *e << dendl; | |
2834 | while (e->logical_offset >= shard_end) { | |
2835 | shard_start = shard_end; | |
11fdf7f2 | 2836 | ceph_assert(sp != esp); |
7c673cae FG |
2837 | ++sp; |
2838 | if (sp == esp) { | |
2839 | shard_end = OBJECT_MAX_SIZE; | |
2840 | } else { | |
2841 | shard_end = sp->offset; | |
2842 | } | |
2843 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2844 | << " to 0x" << shard_end << std::dec << dendl; | |
2845 | } | |
9f95a23c | 2846 | |
7c673cae FG |
2847 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { |
2848 | if (!e->blob->is_spanning()) { | |
2849 | // We have two options: (1) split the blob into pieces at the | |
2850 | // shard boundaries (and adjust extents accordingly), or (2) | |
2851 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2852 | // we may have to split it multiple times--potentially at every | |
2853 | // shard boundary. | |
2854 | bool must_span = false; | |
2855 | BlobRef b = e->blob; | |
2856 | if (b->can_split()) { | |
2857 | uint32_t bstart = e->blob_start(); | |
2858 | uint32_t bend = e->blob_end(); | |
2859 | for (const auto& sh : shards) { | |
2860 | if (bstart < sh.shard_info->offset && | |
2861 | bend > sh.shard_info->offset) { | |
2862 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2863 | if (b->can_split_at(blob_offset)) { | |
2864 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2865 | << std::hex << bstart << " blob_offset 0x" | |
2866 | << blob_offset << std::dec << " " << *b << dendl; | |
2867 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2868 | // switch b to the new right-hand side, in case it | |
2869 | // *also* has to get split. | |
2870 | bstart += blob_offset; | |
2871 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2872 | } else { | |
2873 | must_span = true; | |
2874 | break; | |
2875 | } | |
2876 | } | |
2877 | } | |
2878 | } else { | |
2879 | must_span = true; | |
2880 | } | |
2881 | if (must_span) { | |
31f18b77 FG |
2882 | auto bid = allocate_spanning_blob_id(); |
2883 | b->id = bid; | |
7c673cae FG |
2884 | spanning_blob_map[b->id] = b; |
2885 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
9f95a23c TL |
2886 | if (!was_too_many_blobs_check && |
2887 | too_many_blobs_threshold && | |
2888 | spanning_blob_map.size() >= size_t(too_many_blobs_threshold)) { | |
2889 | ||
2890 | was_too_many_blobs_check = true; | |
2891 | for (size_t i = 0; i < dumped_onodes.size(); ++i) { | |
2892 | if (dumped_onodes[i].first == onode->oid) { | |
2893 | oid_slot = &dumped_onodes[i]; | |
2894 | break; | |
2895 | } | |
2896 | if (!oldest_slot || (oldest_slot && | |
2897 | dumped_onodes[i].second < oldest_slot->second)) { | |
2898 | oldest_slot = &dumped_onodes[i]; | |
2899 | } | |
2900 | } | |
2901 | } | |
7c673cae FG |
2902 | } |
2903 | } | |
2904 | } else { | |
2905 | if (e->blob->is_spanning()) { | |
2906 | spanning_blob_map.erase(e->blob->id); | |
2907 | e->blob->id = -1; | |
2908 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
2909 | } | |
2910 | } | |
2911 | } | |
9f95a23c TL |
2912 | bool do_dump = (!oid_slot && was_too_many_blobs_check) || |
2913 | (oid_slot && | |
2914 | (mono_clock::now() - oid_slot->second >= make_timespan(5 * 60))); | |
2915 | if (do_dump) { | |
2916 | dout(0) << __func__ | |
2917 | << " spanning blob count exceeds threshold, " | |
2918 | << spanning_blob_map.size() << " spanning blobs" | |
2919 | << dendl; | |
2920 | _dump_onode<0>(cct, *onode); | |
2921 | if (oid_slot) { | |
2922 | oid_slot->second = mono_clock::now(); | |
2923 | } else { | |
2924 | ceph_assert(oldest_slot); | |
2925 | oldest_slot->first = onode->oid; | |
2926 | oldest_slot->second = mono_clock::now(); | |
2927 | } | |
2928 | } | |
7c673cae FG |
2929 | } |
2930 | ||
2931 | clear_needs_reshard(); | |
2932 | } | |
2933 | ||
2934 | bool BlueStore::ExtentMap::encode_some( | |
2935 | uint32_t offset, | |
2936 | uint32_t length, | |
2937 | bufferlist& bl, | |
2938 | unsigned *pn) | |
2939 | { | |
7c673cae FG |
2940 | Extent dummy(offset); |
2941 | auto start = extent_map.lower_bound(dummy); | |
2942 | uint32_t end = offset + length; | |
2943 | ||
2944 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
2945 | // serialization only. Hence there is no specific | |
2946 | // handling at ExtentMap level. | |
2947 | ||
2948 | unsigned n = 0; | |
2949 | size_t bound = 0; | |
7c673cae FG |
2950 | bool must_reshard = false; |
2951 | for (auto p = start; | |
2952 | p != extent_map.end() && p->logical_offset < end; | |
2953 | ++p, ++n) { | |
11fdf7f2 | 2954 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
2955 | p->blob->last_encoded_id = -1; |
2956 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
2957 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2958 | << std::dec << " hit new spanning blob " << *p << dendl; | |
2959 | request_reshard(p->blob_start(), p->blob_end()); | |
2960 | must_reshard = true; | |
2961 | } | |
31f18b77 FG |
2962 | if (!must_reshard) { |
2963 | denc_varint(0, bound); // blobid | |
2964 | denc_varint(0, bound); // logical_offset | |
2965 | denc_varint(0, bound); // len | |
2966 | denc_varint(0, bound); // blob_offset | |
7c673cae | 2967 | |
31f18b77 FG |
2968 | p->blob->bound_encode( |
2969 | bound, | |
2970 | struct_v, | |
2971 | p->blob->shared_blob->get_sbid(), | |
2972 | false); | |
2973 | } | |
7c673cae FG |
2974 | } |
2975 | if (must_reshard) { | |
2976 | return true; | |
2977 | } | |
2978 | ||
31f18b77 FG |
2979 | denc(struct_v, bound); |
2980 | denc_varint(0, bound); // number of extents | |
2981 | ||
7c673cae FG |
2982 | { |
2983 | auto app = bl.get_contiguous_appender(bound); | |
2984 | denc(struct_v, app); | |
2985 | denc_varint(n, app); | |
2986 | if (pn) { | |
2987 | *pn = n; | |
2988 | } | |
2989 | ||
2990 | n = 0; | |
2991 | uint64_t pos = 0; | |
2992 | uint64_t prev_len = 0; | |
2993 | for (auto p = start; | |
2994 | p != extent_map.end() && p->logical_offset < end; | |
2995 | ++p, ++n) { | |
2996 | unsigned blobid; | |
2997 | bool include_blob = false; | |
2998 | if (p->blob->is_spanning()) { | |
2999 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
3000 | blobid |= BLOBID_FLAG_SPANNING; | |
3001 | } else if (p->blob->last_encoded_id < 0) { | |
3002 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
3003 | include_blob = true; | |
3004 | blobid = 0; // the decoder will infer the id from n | |
3005 | } else { | |
3006 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
3007 | } | |
3008 | if (p->logical_offset == pos) { | |
3009 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
3010 | } | |
3011 | if (p->blob_offset == 0) { | |
3012 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
3013 | } | |
3014 | if (p->length == prev_len) { | |
3015 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
3016 | } else { | |
3017 | prev_len = p->length; | |
3018 | } | |
3019 | denc_varint(blobid, app); | |
3020 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3021 | denc_varint_lowz(p->logical_offset - pos, app); | |
3022 | } | |
3023 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3024 | denc_varint_lowz(p->blob_offset, app); | |
3025 | } | |
3026 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3027 | denc_varint_lowz(p->length, app); | |
3028 | } | |
3029 | pos = p->logical_end(); | |
3030 | if (include_blob) { | |
3031 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
3032 | } | |
3033 | } | |
3034 | } | |
3035 | /*derr << __func__ << bl << dendl; | |
3036 | derr << __func__ << ":"; | |
3037 | bl.hexdump(*_dout); | |
3038 | *_dout << dendl; | |
3039 | */ | |
3040 | return false; | |
3041 | } | |
3042 | ||
3043 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
3044 | { | |
7c673cae FG |
3045 | /* |
3046 | derr << __func__ << ":"; | |
3047 | bl.hexdump(*_dout); | |
3048 | *_dout << dendl; | |
3049 | */ | |
3050 | ||
11fdf7f2 | 3051 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae FG |
3052 | auto p = bl.front().begin_deep(); |
3053 | __u8 struct_v; | |
3054 | denc(struct_v, p); | |
3055 | // Version 2 differs from v1 in blob's ref_map | |
3056 | // serialization only. Hence there is no specific | |
3057 | // handling at ExtentMap level below. | |
11fdf7f2 | 3058 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
3059 | |
3060 | uint32_t num; | |
3061 | denc_varint(num, p); | |
3062 | vector<BlobRef> blobs(num); | |
3063 | uint64_t pos = 0; | |
3064 | uint64_t prev_len = 0; | |
3065 | unsigned n = 0; | |
3066 | ||
3067 | while (!p.end()) { | |
3068 | Extent *le = new Extent(); | |
3069 | uint64_t blobid; | |
3070 | denc_varint(blobid, p); | |
3071 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3072 | uint64_t gap; | |
3073 | denc_varint_lowz(gap, p); | |
3074 | pos += gap; | |
3075 | } | |
3076 | le->logical_offset = pos; | |
3077 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3078 | denc_varint_lowz(le->blob_offset, p); | |
3079 | } else { | |
3080 | le->blob_offset = 0; | |
3081 | } | |
3082 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3083 | denc_varint_lowz(prev_len, p); | |
3084 | } | |
3085 | le->length = prev_len; | |
3086 | ||
3087 | if (blobid & BLOBID_FLAG_SPANNING) { | |
3088 | dout(30) << __func__ << " getting spanning blob " | |
3089 | << (blobid >> BLOBID_SHIFT_BITS) << dendl; | |
3090 | le->assign_blob(get_spanning_blob(blobid >> BLOBID_SHIFT_BITS)); | |
3091 | } else { | |
3092 | blobid >>= BLOBID_SHIFT_BITS; | |
3093 | if (blobid) { | |
3094 | le->assign_blob(blobs[blobid - 1]); | |
11fdf7f2 | 3095 | ceph_assert(le->blob); |
7c673cae FG |
3096 | } else { |
3097 | Blob *b = new Blob(); | |
3098 | uint64_t sbid = 0; | |
3099 | b->decode(onode->c, p, struct_v, &sbid, false); | |
3100 | blobs[n] = b; | |
3101 | onode->c->open_shared_blob(sbid, b); | |
3102 | le->assign_blob(b); | |
3103 | } | |
3104 | // we build ref_map dynamically for non-spanning blobs | |
3105 | le->blob->get_ref( | |
3106 | onode->c, | |
3107 | le->blob_offset, | |
3108 | le->length); | |
3109 | } | |
3110 | pos += prev_len; | |
3111 | ++n; | |
3112 | extent_map.insert(*le); | |
3113 | } | |
3114 | ||
11fdf7f2 | 3115 | ceph_assert(n == num); |
7c673cae FG |
3116 | return num; |
3117 | } | |
3118 | ||
3119 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
3120 | { | |
3121 | // Version 2 differs from v1 in blob's ref_map | |
3122 | // serialization only. Hence there is no specific | |
3123 | // handling at ExtentMap level. | |
3124 | __u8 struct_v = 2; | |
3125 | ||
3126 | denc(struct_v, p); | |
3127 | denc_varint((uint32_t)0, p); | |
3128 | size_t key_size = 0; | |
3129 | denc_varint((uint32_t)0, key_size); | |
3130 | p += spanning_blob_map.size() * key_size; | |
3131 | for (const auto& i : spanning_blob_map) { | |
3132 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3133 | } | |
3134 | } | |
3135 | ||
3136 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
3137 | bufferlist::contiguous_appender& p) | |
3138 | { | |
3139 | // Version 2 differs from v1 in blob's ref_map | |
3140 | // serialization only. Hence there is no specific | |
3141 | // handling at ExtentMap level. | |
3142 | __u8 struct_v = 2; | |
3143 | ||
3144 | denc(struct_v, p); | |
3145 | denc_varint(spanning_blob_map.size(), p); | |
3146 | for (auto& i : spanning_blob_map) { | |
3147 | denc_varint(i.second->id, p); | |
3148 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3149 | } | |
3150 | } | |
3151 | ||
3152 | void BlueStore::ExtentMap::decode_spanning_blobs( | |
11fdf7f2 | 3153 | bufferptr::const_iterator& p) |
7c673cae FG |
3154 | { |
3155 | __u8 struct_v; | |
3156 | denc(struct_v, p); | |
3157 | // Version 2 differs from v1 in blob's ref_map | |
3158 | // serialization only. Hence there is no specific | |
3159 | // handling at ExtentMap level. | |
11fdf7f2 | 3160 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
3161 | |
3162 | unsigned n; | |
3163 | denc_varint(n, p); | |
3164 | while (n--) { | |
3165 | BlobRef b(new Blob()); | |
3166 | denc_varint(b->id, p); | |
3167 | spanning_blob_map[b->id] = b; | |
3168 | uint64_t sbid = 0; | |
3169 | b->decode(onode->c, p, struct_v, &sbid, true); | |
3170 | onode->c->open_shared_blob(sbid, b); | |
3171 | } | |
3172 | } | |
3173 | ||
3174 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) | |
3175 | { | |
3176 | shards.resize(onode->onode.extent_map_shards.size()); | |
3177 | unsigned i = 0; | |
3178 | for (auto &s : onode->onode.extent_map_shards) { | |
3179 | shards[i].shard_info = &s; | |
3180 | shards[i].loaded = loaded; | |
3181 | shards[i].dirty = dirty; | |
3182 | ++i; | |
3183 | } | |
3184 | } | |
3185 | ||
3186 | void BlueStore::ExtentMap::fault_range( | |
3187 | KeyValueDB *db, | |
3188 | uint32_t offset, | |
3189 | uint32_t length) | |
3190 | { | |
7c673cae FG |
3191 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3192 | << std::dec << dendl; | |
3193 | auto start = seek_shard(offset); | |
3194 | auto last = seek_shard(offset + length); | |
3195 | ||
3196 | if (start < 0) | |
3197 | return; | |
3198 | ||
11fdf7f2 | 3199 | ceph_assert(last >= start); |
7c673cae FG |
3200 | string key; |
3201 | while (start <= last) { | |
11fdf7f2 | 3202 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3203 | auto p = &shards[start]; |
3204 | if (!p->loaded) { | |
3205 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
3206 | << p->shard_info->offset << std::dec << dendl; | |
3207 | bufferlist v; | |
3208 | generate_extent_shard_key_and_apply( | |
3209 | onode->key, p->shard_info->offset, &key, | |
3210 | [&](const string& final_key) { | |
3211 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
3212 | if (r < 0) { | |
3213 | derr << __func__ << " missing shard 0x" << std::hex | |
3214 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
3215 | << dendl; | |
11fdf7f2 | 3216 | ceph_assert(r >= 0); |
7c673cae FG |
3217 | } |
3218 | } | |
3219 | ); | |
3220 | p->extents = decode_some(v); | |
3221 | p->loaded = true; | |
3222 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
3223 | << p->shard_info->offset |
3224 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 3225 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
3226 | ceph_assert(p->dirty == false); |
3227 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
3228 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
3229 | } else { | |
3230 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
3231 | } | |
3232 | ++start; | |
3233 | } | |
3234 | } | |
3235 | ||
3236 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
3237 | uint32_t offset, |
3238 | uint32_t length) | |
3239 | { | |
7c673cae FG |
3240 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3241 | << std::dec << dendl; | |
3242 | if (shards.empty()) { | |
3243 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
3244 | inline_bl.clear(); | |
3245 | return; | |
3246 | } | |
3247 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
3248 | if (length == 0) { |
3249 | length = 1; | |
3250 | } | |
3251 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
3252 | if (start < 0) |
3253 | return; | |
3254 | ||
11fdf7f2 | 3255 | ceph_assert(last >= start); |
7c673cae | 3256 | while (start <= last) { |
11fdf7f2 | 3257 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3258 | auto p = &shards[start]; |
3259 | if (!p->loaded) { | |
11fdf7f2 TL |
3260 | derr << __func__ << "on write 0x" << std::hex << offset |
3261 | << "~" << length << " shard 0x" << p->shard_info->offset | |
3262 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
3263 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
3264 | } |
3265 | if (!p->dirty) { | |
3266 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
3267 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
3268 | p->dirty = true; | |
3269 | } | |
3270 | ++start; | |
3271 | } | |
3272 | } | |
3273 | ||
3274 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
3275 | uint64_t offset) | |
3276 | { | |
3277 | Extent dummy(offset); | |
3278 | return extent_map.find(dummy); | |
3279 | } | |
3280 | ||
7c673cae FG |
3281 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
3282 | uint64_t offset) | |
3283 | { | |
3284 | Extent dummy(offset); | |
3285 | auto fp = extent_map.lower_bound(dummy); | |
3286 | if (fp != extent_map.begin()) { | |
3287 | --fp; | |
3288 | if (fp->logical_end() <= offset) { | |
3289 | ++fp; | |
3290 | } | |
3291 | } | |
3292 | return fp; | |
3293 | } | |
3294 | ||
3295 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
3296 | uint64_t offset) const | |
3297 | { | |
3298 | Extent dummy(offset); | |
3299 | auto fp = extent_map.lower_bound(dummy); | |
3300 | if (fp != extent_map.begin()) { | |
3301 | --fp; | |
3302 | if (fp->logical_end() <= offset) { | |
3303 | ++fp; | |
3304 | } | |
3305 | } | |
3306 | return fp; | |
3307 | } | |
3308 | ||
3309 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
3310 | { | |
3311 | auto fp = seek_lextent(offset); | |
3312 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
3313 | return false; | |
3314 | } | |
3315 | return true; | |
3316 | } | |
3317 | ||
3318 | int BlueStore::ExtentMap::compress_extent_map( | |
3319 | uint64_t offset, | |
3320 | uint64_t length) | |
3321 | { | |
7c673cae FG |
3322 | if (extent_map.empty()) |
3323 | return 0; | |
3324 | int removed = 0; | |
3325 | auto p = seek_lextent(offset); | |
3326 | if (p != extent_map.begin()) { | |
3327 | --p; // start to the left of offset | |
3328 | } | |
3329 | // the caller should have just written to this region | |
11fdf7f2 | 3330 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
3331 | |
3332 | // identify the *next* shard | |
3333 | auto pshard = shards.begin(); | |
3334 | while (pshard != shards.end() && | |
3335 | p->logical_offset >= pshard->shard_info->offset) { | |
3336 | ++pshard; | |
3337 | } | |
3338 | uint64_t shard_end; | |
3339 | if (pshard != shards.end()) { | |
3340 | shard_end = pshard->shard_info->offset; | |
3341 | } else { | |
3342 | shard_end = OBJECT_MAX_SIZE; | |
3343 | } | |
3344 | ||
3345 | auto n = p; | |
3346 | for (++n; n != extent_map.end(); p = n++) { | |
3347 | if (n->logical_offset > offset + length) { | |
3348 | break; // stop after end | |
3349 | } | |
3350 | while (n != extent_map.end() && | |
3351 | p->logical_end() == n->logical_offset && | |
3352 | p->blob == n->blob && | |
3353 | p->blob_offset + p->length == n->blob_offset && | |
3354 | n->logical_offset < shard_end) { | |
3355 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3356 | << " next shard 0x" << shard_end << std::dec | |
3357 | << " merging " << *p << " and " << *n << dendl; | |
3358 | p->length += n->length; | |
3359 | rm(n++); | |
3360 | ++removed; | |
3361 | } | |
3362 | if (n == extent_map.end()) { | |
3363 | break; | |
3364 | } | |
3365 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3366 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3367 | ++pshard; |
3368 | if (pshard != shards.end()) { | |
3369 | shard_end = pshard->shard_info->offset; | |
3370 | } else { | |
3371 | shard_end = OBJECT_MAX_SIZE; | |
3372 | } | |
3373 | } | |
3374 | } | |
11fdf7f2 | 3375 | if (removed) { |
7c673cae FG |
3376 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3377 | } | |
3378 | return removed; | |
3379 | } | |
3380 | ||
3381 | void BlueStore::ExtentMap::punch_hole( | |
3382 | CollectionRef &c, | |
3383 | uint64_t offset, | |
3384 | uint64_t length, | |
3385 | old_extent_map_t *old_extents) | |
3386 | { | |
3387 | auto p = seek_lextent(offset); | |
3388 | uint64_t end = offset + length; | |
3389 | while (p != extent_map.end()) { | |
3390 | if (p->logical_offset >= end) { | |
3391 | break; | |
3392 | } | |
3393 | if (p->logical_offset < offset) { | |
3394 | if (p->logical_end() > end) { | |
3395 | // split and deref middle | |
3396 | uint64_t front = offset - p->logical_offset; | |
3397 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3398 | length, p->blob); | |
3399 | old_extents->push_back(*oe); | |
3400 | add(end, | |
3401 | p->blob_offset + front + length, | |
3402 | p->length - front - length, | |
3403 | p->blob); | |
3404 | p->length = front; | |
3405 | break; | |
3406 | } else { | |
3407 | // deref tail | |
11fdf7f2 | 3408 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3409 | uint64_t keep = offset - p->logical_offset; |
3410 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3411 | p->length - keep, p->blob); | |
3412 | old_extents->push_back(*oe); | |
3413 | p->length = keep; | |
3414 | ++p; | |
3415 | continue; | |
3416 | } | |
3417 | } | |
3418 | if (p->logical_offset + p->length <= end) { | |
3419 | // deref whole lextent | |
3420 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3421 | p->length, p->blob); | |
3422 | old_extents->push_back(*oe); | |
3423 | rm(p++); | |
3424 | continue; | |
3425 | } | |
3426 | // deref head | |
3427 | uint64_t keep = p->logical_end() - end; | |
3428 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3429 | p->length - keep, p->blob); | |
3430 | old_extents->push_back(*oe); | |
3431 | ||
3432 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3433 | rm(p); | |
3434 | break; | |
3435 | } | |
3436 | } | |
3437 | ||
3438 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3439 | CollectionRef &c, | |
3440 | uint64_t logical_offset, | |
3441 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3442 | old_extent_map_t *old_extents) | |
3443 | { | |
3444 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3445 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3446 | |
3447 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3448 | // old_extents list if we overwre the blob totally | |
3449 | // This might happen during WAL overwrite. | |
3450 | b->get_ref(onode->c, blob_offset, length); | |
3451 | ||
3452 | if (old_extents) { | |
3453 | punch_hole(c, logical_offset, length, old_extents); | |
3454 | } | |
3455 | ||
3456 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3457 | extent_map.insert(*le); | |
3458 | if (spans_shard(logical_offset, length)) { | |
3459 | request_reshard(logical_offset, logical_offset + length); | |
3460 | } | |
3461 | return le; | |
3462 | } | |
3463 | ||
3464 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3465 | BlobRef lb, | |
3466 | uint32_t blob_offset, | |
3467 | uint32_t pos) | |
3468 | { | |
7c673cae FG |
3469 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; |
3470 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3471 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3472 | << dendl; | |
3473 | BlobRef rb = onode->c->new_blob(); | |
3474 | lb->split(onode->c, blob_offset, rb.get()); | |
3475 | ||
3476 | for (auto ep = seek_lextent(pos); | |
3477 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3478 | ++ep) { | |
3479 | if (ep->blob != lb) { | |
3480 | continue; | |
3481 | } | |
3482 | if (ep->logical_offset < pos) { | |
3483 | // split extent | |
3484 | size_t left = pos - ep->logical_offset; | |
3485 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3486 | extent_map.insert(*ne); | |
3487 | ep->length = left; | |
3488 | dout(30) << __func__ << " split " << *ep << dendl; | |
3489 | dout(30) << __func__ << " to " << *ne << dendl; | |
3490 | } else { | |
3491 | // switch blob | |
11fdf7f2 | 3492 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3493 | |
3494 | ep->blob = rb; | |
3495 | ep->blob_offset -= blob_offset; | |
3496 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3497 | } | |
3498 | } | |
3499 | return rb; | |
3500 | } | |
3501 | ||
3502 | // Onode | |
3503 | ||
3504 | #undef dout_prefix | |
3505 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3506 | ||
f6b5b4d7 TL |
3507 | // |
3508 | // A tricky thing about Onode's ref counter is that we do an additional | |
3509 | // increment when newly pinned instance is detected. And -1 on unpin. | |
3510 | // This prevents from a conflict with a delete call (when nref == 0). | |
3511 | // The latter might happen while the thread is in unpin() function | |
3512 | // (and e.g. waiting for lock acquisition) since nref is already | |
3513 | // decremented. And another 'putting' thread on the instance will release it. | |
3514 | // | |
3515 | void BlueStore::Onode::get() { | |
adb31ebb TL |
3516 | if (++nref >= 2 && !pinned) { |
3517 | OnodeCacheShard* ocs = c->get_onode_cache(); | |
f67539c2 TL |
3518 | ocs->lock.lock(); |
3519 | // It is possible that during waiting split_cache moved us to different OnodeCacheShard. | |
3520 | while (ocs != c->get_onode_cache()) { | |
3521 | ocs->lock.unlock(); | |
3522 | ocs = c->get_onode_cache(); | |
3523 | ocs->lock.lock(); | |
3524 | } | |
adb31ebb TL |
3525 | bool was_pinned = pinned; |
3526 | pinned = nref >= 2; | |
3527 | // additional increment for newly pinned instance | |
3528 | bool r = !was_pinned && pinned; | |
3529 | if (r) { | |
3530 | ++nref; | |
3531 | } | |
3532 | if (cached && r) { | |
3533 | ocs->_pin(this); | |
3534 | } | |
f67539c2 | 3535 | ocs->lock.unlock(); |
f6b5b4d7 TL |
3536 | } |
3537 | } | |
3538 | void BlueStore::Onode::put() { | |
adb31ebb TL |
3539 | int n = --nref; |
3540 | if (n == 2) { | |
3541 | OnodeCacheShard* ocs = c->get_onode_cache(); | |
f67539c2 TL |
3542 | ocs->lock.lock(); |
3543 | // It is possible that during waiting split_cache moved us to different OnodeCacheShard. | |
3544 | while (ocs != c->get_onode_cache()) { | |
3545 | ocs->lock.unlock(); | |
3546 | ocs = c->get_onode_cache(); | |
3547 | ocs->lock.lock(); | |
3548 | } | |
adb31ebb TL |
3549 | bool need_unpin = pinned; |
3550 | pinned = pinned && nref > 2; // intentionally use > not >= as we have | |
3551 | // +1 due to pinned state | |
3552 | need_unpin = need_unpin && !pinned; | |
3553 | if (cached && need_unpin) { | |
3554 | if (exists) { | |
3555 | ocs->_unpin(this); | |
3556 | } else { | |
3557 | ocs->_unpin_and_rm(this); | |
3558 | // remove will also decrement nref and delete Onode | |
3559 | c->onode_map._remove(oid); | |
3560 | } | |
3561 | } | |
3562 | // additional decrement for newly unpinned instance | |
3563 | // should be the last action since Onode can be released | |
3564 | // at any point after this decrement | |
3565 | if (need_unpin) { | |
3566 | n = --nref; | |
3567 | } | |
f67539c2 | 3568 | ocs->lock.unlock(); |
f6b5b4d7 | 3569 | } |
adb31ebb | 3570 | if (n == 0) { |
f6b5b4d7 TL |
3571 | delete this; |
3572 | } | |
3573 | } | |
3574 | ||
eafe8130 TL |
3575 | BlueStore::Onode* BlueStore::Onode::decode( |
3576 | CollectionRef c, | |
3577 | const ghobject_t& oid, | |
3578 | const string& key, | |
3579 | const bufferlist& v) | |
3580 | { | |
3581 | Onode* on = new Onode(c.get(), oid, key); | |
3582 | on->exists = true; | |
3583 | auto p = v.front().begin_deep(); | |
3584 | on->onode.decode(p); | |
3585 | for (auto& i : on->onode.attrs) { | |
f91f0fd5 | 3586 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
eafe8130 TL |
3587 | } |
3588 | ||
3589 | // initialize extent_map | |
3590 | on->extent_map.decode_spanning_blobs(p); | |
3591 | if (on->onode.extent_map_shards.empty()) { | |
3592 | denc(on->extent_map.inline_bl, p); | |
3593 | on->extent_map.decode_some(on->extent_map.inline_bl); | |
3594 | on->extent_map.inline_bl.reassign_to_mempool( | |
f91f0fd5 | 3595 | mempool::mempool_bluestore_cache_data); |
eafe8130 TL |
3596 | } |
3597 | else { | |
3598 | on->extent_map.init_shards(false, false); | |
3599 | } | |
3600 | return on; | |
3601 | } | |
3602 | ||
7c673cae FG |
3603 | void BlueStore::Onode::flush() |
3604 | { | |
3605 | if (flushing_count.load()) { | |
3606 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
9f95a23c | 3607 | waiting_count++; |
11fdf7f2 | 3608 | std::unique_lock l(flush_lock); |
7c673cae FG |
3609 | while (flushing_count.load()) { |
3610 | flush_cond.wait(l); | |
3611 | } | |
9f95a23c | 3612 | waiting_count--; |
7c673cae FG |
3613 | } |
3614 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3615 | } | |
3616 | ||
9f95a23c TL |
3617 | void BlueStore::Onode::dump(Formatter* f) const |
3618 | { | |
3619 | onode.dump(f); | |
3620 | extent_map.dump(f); | |
3621 | } | |
3622 | ||
522d829b | 3623 | const std::string& BlueStore::Onode::calc_omap_prefix(uint8_t flags) |
9f95a23c | 3624 | { |
522d829b | 3625 | if (bluestore_onode_t::is_pgmeta_omap(flags)) { |
9f95a23c TL |
3626 | return PREFIX_PGMETA_OMAP; |
3627 | } | |
522d829b | 3628 | if (bluestore_onode_t::is_perpg_omap(flags)) { |
f67539c2 TL |
3629 | return PREFIX_PERPG_OMAP; |
3630 | } | |
522d829b | 3631 | if (bluestore_onode_t::is_perpool_omap(flags)) { |
9f95a23c TL |
3632 | return PREFIX_PERPOOL_OMAP; |
3633 | } | |
3634 | return PREFIX_OMAP; | |
3635 | } | |
3636 | ||
3637 | // '-' < '.' < '~' | |
522d829b TL |
3638 | void BlueStore::Onode::calc_omap_header( |
3639 | uint8_t flags, | |
3640 | const Onode* o, | |
3641 | std::string* out) | |
9f95a23c | 3642 | { |
522d829b TL |
3643 | if (!bluestore_onode_t::is_pgmeta_omap(flags)) { |
3644 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3645 | _key_encode_u64(o->c->pool(), out); | |
3646 | _key_encode_u32(o->oid.hobj.get_bitwise_key_u32(), out); | |
3647 | } else if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3648 | _key_encode_u64(o->c->pool(), out); | |
f67539c2 | 3649 | } |
9f95a23c | 3650 | } |
522d829b | 3651 | _key_encode_u64(o->onode.nid, out); |
9f95a23c TL |
3652 | out->push_back('-'); |
3653 | } | |
3654 | ||
522d829b TL |
3655 | void BlueStore::Onode::calc_omap_key(uint8_t flags, |
3656 | const Onode* o, | |
3657 | const std::string& key, | |
3658 | std::string* out) | |
9f95a23c | 3659 | { |
522d829b TL |
3660 | if (!bluestore_onode_t::is_pgmeta_omap(flags)) { |
3661 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3662 | _key_encode_u64(o->c->pool(), out); | |
3663 | _key_encode_u32(o->oid.hobj.get_bitwise_key_u32(), out); | |
3664 | } else if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3665 | _key_encode_u64(o->c->pool(), out); | |
f67539c2 | 3666 | } |
9f95a23c | 3667 | } |
522d829b | 3668 | _key_encode_u64(o->onode.nid, out); |
9f95a23c TL |
3669 | out->push_back('.'); |
3670 | out->append(key); | |
3671 | } | |
3672 | ||
3673 | void BlueStore::Onode::rewrite_omap_key(const string& old, string *out) | |
3674 | { | |
f67539c2 TL |
3675 | if (!onode.is_pgmeta_omap()) { |
3676 | if (onode.is_perpg_omap()) { | |
3677 | _key_encode_u64(c->pool(), out); | |
3678 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), out); | |
3679 | } else if (onode.is_perpool_omap()) { | |
3680 | _key_encode_u64(c->pool(), out); | |
3681 | } | |
9f95a23c TL |
3682 | } |
3683 | _key_encode_u64(onode.nid, out); | |
3684 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
3685 | } | |
3686 | ||
522d829b TL |
3687 | void BlueStore::Onode::calc_omap_tail( |
3688 | uint8_t flags, | |
3689 | const Onode* o, | |
3690 | std::string* out) | |
9f95a23c | 3691 | { |
522d829b TL |
3692 | if (!bluestore_onode_t::is_pgmeta_omap(flags)) { |
3693 | if (bluestore_onode_t::is_perpg_omap(flags)) { | |
3694 | _key_encode_u64(o->c->pool(), out); | |
3695 | _key_encode_u32(o->oid.hobj.get_bitwise_key_u32(), out); | |
3696 | } else if (bluestore_onode_t::is_perpool_omap(flags)) { | |
3697 | _key_encode_u64(o->c->pool(), out); | |
f67539c2 | 3698 | } |
9f95a23c | 3699 | } |
522d829b | 3700 | _key_encode_u64(o->onode.nid, out); |
9f95a23c TL |
3701 | out->push_back('~'); |
3702 | } | |
3703 | ||
3704 | void BlueStore::Onode::decode_omap_key(const string& key, string *user_key) | |
3705 | { | |
f67539c2 TL |
3706 | size_t pos = sizeof(uint64_t) + 1; |
3707 | if (!onode.is_pgmeta_omap()) { | |
3708 | if (onode.is_perpg_omap()) { | |
3709 | pos += sizeof(uint64_t) + sizeof(uint32_t); | |
3710 | } else if (onode.is_perpool_omap()) { | |
3711 | pos += sizeof(uint64_t); | |
3712 | } | |
9f95a23c | 3713 | } |
f67539c2 | 3714 | *user_key = key.substr(pos); |
9f95a23c TL |
3715 | } |
3716 | ||
7c673cae FG |
3717 | // ======================================================= |
3718 | // WriteContext | |
3719 | ||
3720 | /// Checks for writes to the same pextent within a blob | |
3721 | bool BlueStore::WriteContext::has_conflict( | |
3722 | BlobRef b, | |
3723 | uint64_t loffs, | |
3724 | uint64_t loffs_end, | |
3725 | uint64_t min_alloc_size) | |
3726 | { | |
11fdf7f2 TL |
3727 | ceph_assert((loffs % min_alloc_size) == 0); |
3728 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3729 | for (auto w : writes) { |
3730 | if (b == w.b) { | |
11fdf7f2 TL |
3731 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3732 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3733 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3734 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3735 | return true; |
3736 | } | |
3737 | } | |
3738 | } | |
3739 | return false; | |
3740 | } | |
3741 | ||
3742 | // ======================================================= | |
3743 | ||
3744 | // DeferredBatch | |
3745 | #undef dout_prefix | |
3746 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
9f95a23c TL |
3747 | #undef dout_context |
3748 | #define dout_context cct | |
7c673cae FG |
3749 | |
3750 | void BlueStore::DeferredBatch::prepare_write( | |
3751 | CephContext *cct, | |
3752 | uint64_t seq, uint64_t offset, uint64_t length, | |
3753 | bufferlist::const_iterator& blp) | |
3754 | { | |
3755 | _discard(cct, offset, length); | |
3756 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3757 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3758 | i.first->second.seq = seq; |
3759 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3760 | i.first->second.bl.reassign_to_mempool( |
3761 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3762 | dout(20) << __func__ << " seq " << seq |
3763 | << " 0x" << std::hex << offset << "~" << length | |
3764 | << " crc " << i.first->second.bl.crc32c(-1) | |
3765 | << std::dec << dendl; | |
3766 | seq_bytes[seq] += length; | |
3767 | #ifdef DEBUG_DEFERRED | |
3768 | _audit(cct); | |
3769 | #endif | |
3770 | } | |
3771 | ||
3772 | void BlueStore::DeferredBatch::_discard( | |
3773 | CephContext *cct, uint64_t offset, uint64_t length) | |
3774 | { | |
3775 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3776 | << std::dec << dendl; | |
3777 | auto p = iomap.lower_bound(offset); | |
3778 | if (p != iomap.begin()) { | |
3779 | --p; | |
3780 | auto end = p->first + p->second.bl.length(); | |
3781 | if (end > offset) { | |
3782 | bufferlist head; | |
3783 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3784 | dout(20) << __func__ << " keep head " << p->second.seq | |
3785 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3786 | << " -> 0x" << head.length() << std::dec << dendl; | |
3787 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3788 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3789 | if (end > offset + length) { |
3790 | bufferlist tail; | |
3791 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3792 | end - (offset + length)); | |
3793 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3794 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3795 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3796 | auto &n = iomap[offset + length]; | |
3797 | n.bl.swap(tail); | |
3798 | n.seq = p->second.seq; | |
3799 | i->second -= length; | |
3800 | } else { | |
3801 | i->second -= end - offset; | |
3802 | } | |
11fdf7f2 | 3803 | ceph_assert(i->second >= 0); |
7c673cae FG |
3804 | p->second.bl.swap(head); |
3805 | } | |
3806 | ++p; | |
3807 | } | |
3808 | while (p != iomap.end()) { | |
3809 | if (p->first >= offset + length) { | |
3810 | break; | |
3811 | } | |
3812 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3813 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3814 | auto end = p->first + p->second.bl.length(); |
3815 | if (end > offset + length) { | |
3816 | unsigned drop_front = offset + length - p->first; | |
3817 | unsigned keep_tail = end - (offset + length); | |
3818 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3819 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3820 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3821 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3822 | << std::dec << dendl; | |
3823 | auto &s = iomap[offset + length]; | |
3824 | s.seq = p->second.seq; | |
3825 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3826 | i->second -= drop_front; | |
3827 | } else { | |
3828 | dout(20) << __func__ << " drop " << p->second.seq | |
3829 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3830 | << std::dec << dendl; | |
3831 | i->second -= p->second.bl.length(); | |
3832 | } | |
11fdf7f2 | 3833 | ceph_assert(i->second >= 0); |
7c673cae FG |
3834 | p = iomap.erase(p); |
3835 | } | |
3836 | } | |
3837 | ||
3838 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3839 | { | |
3840 | map<uint64_t,int> sb; | |
3841 | for (auto p : seq_bytes) { | |
3842 | sb[p.first] = 0; // make sure we have the same set of keys | |
3843 | } | |
3844 | uint64_t pos = 0; | |
3845 | for (auto& p : iomap) { | |
11fdf7f2 | 3846 | ceph_assert(p.first >= pos); |
7c673cae FG |
3847 | sb[p.second.seq] += p.second.bl.length(); |
3848 | pos = p.first + p.second.bl.length(); | |
3849 | } | |
11fdf7f2 | 3850 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3851 | } |
3852 | ||
3853 | ||
3854 | // Collection | |
3855 | ||
3856 | #undef dout_prefix | |
3857 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3858 | ||
9f95a23c TL |
3859 | BlueStore::Collection::Collection(BlueStore *store_, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t cid) |
3860 | : CollectionImpl(store_->cct, cid), | |
11fdf7f2 | 3861 | store(store_), |
9f95a23c | 3862 | cache(bc), |
7c673cae | 3863 | exists(true), |
9f95a23c | 3864 | onode_map(oc), |
11fdf7f2 TL |
3865 | commit_queue(nullptr) |
3866 | { | |
3867 | } | |
3868 | ||
3869 | bool BlueStore::Collection::flush_commit(Context *c) | |
3870 | { | |
3871 | return osr->flush_commit(c); | |
3872 | } | |
3873 | ||
3874 | void BlueStore::Collection::flush() | |
3875 | { | |
3876 | osr->flush(); | |
3877 | } | |
3878 | ||
3879 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3880 | { |
11fdf7f2 | 3881 | osr->flush_all_but_last(); |
7c673cae FG |
3882 | } |
3883 | ||
3884 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3885 | { | |
11fdf7f2 | 3886 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3887 | const bluestore_blob_t& blob = b->get_blob(); |
3888 | if (!blob.is_shared()) { | |
3889 | b->shared_blob = new SharedBlob(this); | |
3890 | return; | |
3891 | } | |
3892 | ||
3893 | b->shared_blob = shared_blob_set.lookup(sbid); | |
3894 | if (b->shared_blob) { | |
3895 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3896 | << std::dec << " had " << *b->shared_blob << dendl; | |
3897 | } else { | |
3898 | b->shared_blob = new SharedBlob(sbid, this); | |
3899 | shared_blob_set.add(this, b->shared_blob.get()); | |
3900 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3901 | << std::dec << " opened " << *b->shared_blob | |
3902 | << dendl; | |
3903 | } | |
3904 | } | |
3905 | ||
3906 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
3907 | { | |
3908 | if (!sb->is_loaded()) { | |
3909 | ||
3910 | bufferlist v; | |
3911 | string key; | |
3912 | auto sbid = sb->get_sbid(); | |
3913 | get_shared_blob_key(sbid, &key); | |
3914 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
3915 | if (r < 0) { | |
3916 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
3917 | << std::dec << " not found at key " | |
3918 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 3919 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
3920 | } |
3921 | ||
3922 | sb->loaded = true; | |
3923 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
3924 | auto p = v.cbegin(); |
3925 | decode(*(sb->persistent), p); | |
7c673cae FG |
3926 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
3927 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
3928 | } | |
3929 | } | |
3930 | ||
3931 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
3932 | { | |
7c673cae | 3933 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 3934 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
3935 | |
3936 | // update blob | |
31f18b77 | 3937 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 3938 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
3939 | |
3940 | // update shared blob | |
3941 | b->shared_blob->loaded = true; | |
3942 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
3943 | shared_blob_set.add(this, b->shared_blob.get()); | |
3944 | for (auto p : blob.get_extents()) { | |
3945 | if (p.is_valid()) { | |
3946 | b->shared_blob->get_ref( | |
3947 | p.offset, | |
3948 | p.length); | |
3949 | } | |
3950 | } | |
3951 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
3952 | } | |
3953 | ||
31f18b77 FG |
3954 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
3955 | { | |
3956 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 3957 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
3958 | |
3959 | uint64_t sbid = sb->get_sbid(); | |
3960 | shared_blob_set.remove(sb); | |
3961 | sb->loaded = false; | |
3962 | delete sb->persistent; | |
3963 | sb->sbid_unloaded = 0; | |
3964 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
3965 | return sbid; | |
3966 | } | |
3967 | ||
7c673cae FG |
3968 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
3969 | const ghobject_t& oid, | |
9f95a23c TL |
3970 | bool create, |
3971 | bool is_createop) | |
7c673cae | 3972 | { |
9f95a23c | 3973 | ceph_assert(create ? ceph_mutex_is_wlocked(lock) : ceph_mutex_is_locked(lock)); |
7c673cae FG |
3974 | |
3975 | spg_t pgid; | |
3976 | if (cid.is_pg(&pgid)) { | |
3977 | if (!oid.match(cnode.bits, pgid.ps())) { | |
3978 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
3979 | << pgid << " bits " << cnode.bits << dendl; | |
3980 | ceph_abort(); | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | OnodeRef o = onode_map.lookup(oid); | |
3985 | if (o) | |
3986 | return o; | |
3987 | ||
eafe8130 | 3988 | string key; |
7c673cae FG |
3989 | get_object_key(store->cct, oid, &key); |
3990 | ||
3991 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
3992 | << pretty_binary_string(key) << dendl; | |
3993 | ||
3994 | bufferlist v; | |
9f95a23c | 3995 | int r = -ENOENT; |
7c673cae | 3996 | Onode *on; |
9f95a23c TL |
3997 | if (!is_createop) { |
3998 | r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
3999 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
4000 | } | |
7c673cae | 4001 | if (v.length() == 0) { |
11fdf7f2 | 4002 | ceph_assert(r == -ENOENT); |
f67539c2 | 4003 | if (!create) |
7c673cae FG |
4004 | return OnodeRef(); |
4005 | ||
4006 | // new object, new onode | |
4007 | on = new Onode(this, oid, key); | |
4008 | } else { | |
4009 | // loaded | |
11fdf7f2 | 4010 | ceph_assert(r >= 0); |
eafe8130 | 4011 | on = Onode::decode(this, oid, key, v); |
7c673cae FG |
4012 | } |
4013 | o.reset(on); | |
4014 | return onode_map.add(oid, o); | |
4015 | } | |
4016 | ||
4017 | void BlueStore::Collection::split_cache( | |
4018 | Collection *dest) | |
4019 | { | |
4020 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
4021 | ||
f67539c2 TL |
4022 | auto *ocache = get_onode_cache(); |
4023 | auto *ocache_dest = dest->get_onode_cache(); | |
4024 | ||
4025 | // lock cache shards | |
4026 | std::lock(ocache->lock, ocache_dest->lock, cache->lock, dest->cache->lock); | |
4027 | std::lock_guard l(ocache->lock, std::adopt_lock); | |
4028 | std::lock_guard l2(ocache_dest->lock, std::adopt_lock); | |
4029 | std::lock_guard l3(cache->lock, std::adopt_lock); | |
4030 | std::lock_guard l4(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
4031 | |
4032 | int destbits = dest->cnode.bits; | |
4033 | spg_t destpg; | |
4034 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 4035 | ceph_assert(is_pg); |
7c673cae FG |
4036 | |
4037 | auto p = onode_map.onode_map.begin(); | |
4038 | while (p != onode_map.onode_map.end()) { | |
11fdf7f2 | 4039 | OnodeRef o = p->second; |
7c673cae FG |
4040 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
4041 | // onode does not belong to this child | |
11fdf7f2 TL |
4042 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
4043 | << dendl; | |
7c673cae FG |
4044 | ++p; |
4045 | } else { | |
7c673cae FG |
4046 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
4047 | << dendl; | |
4048 | ||
f6b5b4d7 TL |
4049 | // ensuring that nref is always >= 2 and hence onode is pinned and |
4050 | // physically out of cache during the transition | |
4051 | OnodeRef o_pin = o; | |
4052 | ceph_assert(o->pinned); | |
4053 | ||
7c673cae | 4054 | p = onode_map.onode_map.erase(p); |
7c673cae | 4055 | dest->onode_map.onode_map[o->oid] = o; |
adb31ebb | 4056 | if (o->cached) { |
f6b5b4d7 | 4057 | get_onode_cache()->move_pinned(dest->get_onode_cache(), o.get()); |
9f95a23c | 4058 | } |
f6b5b4d7 | 4059 | o->c = dest; |
7c673cae FG |
4060 | |
4061 | // move over shared blobs and buffers. cover shared blobs from | |
4062 | // both extent map and spanning blob map (the full extent map | |
4063 | // may not be faulted in) | |
4064 | vector<SharedBlob*> sbvec; | |
4065 | for (auto& e : o->extent_map.extent_map) { | |
4066 | sbvec.push_back(e.blob->shared_blob.get()); | |
4067 | } | |
4068 | for (auto& b : o->extent_map.spanning_blob_map) { | |
4069 | sbvec.push_back(b.second->shared_blob.get()); | |
4070 | } | |
4071 | for (auto sb : sbvec) { | |
4072 | if (sb->coll == dest) { | |
4073 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
4074 | << dendl; | |
4075 | continue; | |
4076 | } | |
4077 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
4078 | if (sb->get_sbid()) { |
4079 | ldout(store->cct, 20) << __func__ | |
4080 | << " moving registration " << *sb << dendl; | |
4081 | shared_blob_set.remove(sb); | |
4082 | dest->shared_blob_set.add(dest, sb); | |
4083 | } | |
3efd9988 | 4084 | sb->coll = dest; |
7c673cae | 4085 | if (dest->cache != cache) { |
7c673cae FG |
4086 | for (auto& i : sb->bc.buffer_map) { |
4087 | if (!i.second->is_writing()) { | |
4088 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
4089 | << dendl; | |
9f95a23c | 4090 | dest->cache->_move(cache, i.second.get()); |
7c673cae FG |
4091 | } |
4092 | } | |
4093 | } | |
4094 | } | |
7c673cae FG |
4095 | } |
4096 | } | |
9f95a23c | 4097 | dest->cache->_trim(); |
7c673cae FG |
4098 | } |
4099 | ||
7c673cae FG |
4100 | // ======================================================= |
4101 | ||
91327a77 AA |
4102 | // MempoolThread |
4103 | ||
4104 | #undef dout_prefix | |
4105 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
9f95a23c TL |
4106 | #undef dout_context |
4107 | #define dout_context store->cct | |
91327a77 | 4108 | |
7c673cae FG |
4109 | void *BlueStore::MempoolThread::entry() |
4110 | { | |
9f95a23c | 4111 | std::unique_lock l{lock}; |
11fdf7f2 | 4112 | |
92f5a8d4 | 4113 | uint32_t prev_config_change = store->config_changed.load(); |
eafe8130 TL |
4114 | uint64_t base = store->osd_memory_base; |
4115 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4116 | uint64_t target = store->osd_memory_target; | |
4117 | uint64_t min = store->osd_memory_cache_min; | |
4118 | uint64_t max = min; | |
4119 | ||
4120 | // When setting the maximum amount of memory to use for cache, first | |
4121 | // assume some base amount of memory for the OSD and then fudge in | |
4122 | // some overhead for fragmentation that scales with cache usage. | |
4123 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4124 | if (ltarget > base + min) { | |
4125 | max = ltarget - base; | |
11fdf7f2 | 4126 | } |
31f18b77 | 4127 | |
eafe8130 | 4128 | binned_kv_cache = store->db->get_priority_cache(); |
f67539c2 | 4129 | binned_kv_onode_cache = store->db->get_priority_cache(PREFIX_OBJ); |
eafe8130 TL |
4130 | if (store->cache_autotune && binned_kv_cache != nullptr) { |
4131 | pcm = std::make_shared<PriorityCache::Manager>( | |
f67539c2 | 4132 | store->cct, min, max, target, true, "bluestore-pricache"); |
eafe8130 TL |
4133 | pcm->insert("kv", binned_kv_cache, true); |
4134 | pcm->insert("meta", meta_cache, true); | |
4135 | pcm->insert("data", data_cache, true); | |
f67539c2 TL |
4136 | if (binned_kv_onode_cache != nullptr) { |
4137 | pcm->insert("kv_onode", binned_kv_onode_cache, true); | |
4138 | } | |
eafe8130 | 4139 | } |
91327a77 AA |
4140 | |
4141 | utime_t next_balance = ceph_clock_now(); | |
4142 | utime_t next_resize = ceph_clock_now(); | |
9f95a23c TL |
4143 | utime_t next_deferred_force_submit = ceph_clock_now(); |
4144 | utime_t alloc_stats_dump_clock = ceph_clock_now(); | |
31f18b77 | 4145 | |
91327a77 | 4146 | bool interval_stats_trim = false; |
91327a77 | 4147 | while (!stop) { |
92f5a8d4 TL |
4148 | // Update pcm cache settings if related configuration was changed |
4149 | uint32_t cur_config_change = store->config_changed.load(); | |
4150 | if (cur_config_change != prev_config_change) { | |
4151 | _update_cache_settings(); | |
4152 | prev_config_change = cur_config_change; | |
4153 | } | |
4154 | ||
91327a77 AA |
4155 | // Before we trim, check and see if it's time to rebalance/resize. |
4156 | double autotune_interval = store->cache_autotune_interval; | |
4157 | double resize_interval = store->osd_memory_cache_resize_interval; | |
9f95a23c TL |
4158 | double max_defer_interval = store->max_defer_interval; |
4159 | ||
4160 | double alloc_stats_dump_interval = | |
4161 | store->cct->_conf->bluestore_alloc_stats_dump_interval; | |
91327a77 | 4162 | |
9f95a23c TL |
4163 | if (alloc_stats_dump_interval > 0 && |
4164 | alloc_stats_dump_clock + alloc_stats_dump_interval < ceph_clock_now()) { | |
4165 | store->_record_allocation_stats(); | |
4166 | alloc_stats_dump_clock = ceph_clock_now(); | |
4167 | } | |
91327a77 | 4168 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { |
11fdf7f2 TL |
4169 | _adjust_cache_settings(); |
4170 | ||
91327a77 | 4171 | // Log events at 5 instead of 20 when balance happens. |
91327a77 | 4172 | interval_stats_trim = true; |
eafe8130 TL |
4173 | |
4174 | if (pcm != nullptr) { | |
4175 | pcm->balance(); | |
91327a77 | 4176 | } |
31f18b77 | 4177 | |
91327a77 AA |
4178 | next_balance = ceph_clock_now(); |
4179 | next_balance += autotune_interval; | |
4180 | } | |
4181 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { | |
eafe8130 TL |
4182 | if (ceph_using_tcmalloc() && pcm != nullptr) { |
4183 | pcm->tune_memory(); | |
91327a77 AA |
4184 | } |
4185 | next_resize = ceph_clock_now(); | |
4186 | next_resize += resize_interval; | |
31f18b77 FG |
4187 | } |
4188 | ||
9f95a23c TL |
4189 | if (max_defer_interval > 0 && |
4190 | next_deferred_force_submit < ceph_clock_now()) { | |
4191 | if (store->get_deferred_last_submitted() + max_defer_interval < | |
4192 | ceph_clock_now()) { | |
4193 | store->deferred_try_submit(); | |
4194 | } | |
4195 | next_deferred_force_submit = ceph_clock_now(); | |
4196 | next_deferred_force_submit += max_defer_interval/3; | |
4197 | } | |
4198 | ||
4199 | // Now Resize the shards | |
4200 | _resize_shards(interval_stats_trim); | |
91327a77 | 4201 | interval_stats_trim = false; |
31f18b77 | 4202 | |
91327a77 | 4203 | store->_update_cache_logger(); |
11fdf7f2 TL |
4204 | auto wait = ceph::make_timespan( |
4205 | store->cct->_conf->bluestore_cache_trim_interval); | |
4206 | cond.wait_for(l, wait); | |
7c673cae | 4207 | } |
9f95a23c TL |
4208 | // do final dump |
4209 | store->_record_allocation_stats(); | |
7c673cae | 4210 | stop = false; |
f67539c2 | 4211 | pcm = nullptr; |
7c673cae FG |
4212 | return NULL; |
4213 | } | |
4214 | ||
91327a77 AA |
4215 | void BlueStore::MempoolThread::_adjust_cache_settings() |
4216 | { | |
11fdf7f2 TL |
4217 | if (binned_kv_cache != nullptr) { |
4218 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
4219 | } | |
f67539c2 TL |
4220 | if (binned_kv_onode_cache != nullptr) { |
4221 | binned_kv_onode_cache->set_cache_ratio(store->cache_kv_onode_ratio); | |
4222 | } | |
11fdf7f2 TL |
4223 | meta_cache->set_cache_ratio(store->cache_meta_ratio); |
4224 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
91327a77 AA |
4225 | } |
4226 | ||
9f95a23c | 4227 | void BlueStore::MempoolThread::_resize_shards(bool interval_stats) |
91327a77 | 4228 | { |
9f95a23c TL |
4229 | size_t onode_shards = store->onode_cache_shards.size(); |
4230 | size_t buffer_shards = store->buffer_cache_shards.size(); | |
91327a77 | 4231 | int64_t kv_used = store->db->get_cache_usage(); |
f67539c2 | 4232 | int64_t kv_onode_used = store->db->get_cache_usage(PREFIX_OBJ); |
11fdf7f2 TL |
4233 | int64_t meta_used = meta_cache->_get_used_bytes(); |
4234 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
4235 | |
4236 | uint64_t cache_size = store->cache_size; | |
4237 | int64_t kv_alloc = | |
11fdf7f2 | 4238 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
f67539c2 TL |
4239 | int64_t kv_onode_alloc = |
4240 | static_cast<int64_t>(store->cache_kv_onode_ratio * cache_size); | |
91327a77 | 4241 | int64_t meta_alloc = |
11fdf7f2 | 4242 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 4243 | int64_t data_alloc = |
11fdf7f2 | 4244 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 4245 | |
eafe8130 TL |
4246 | if (pcm != nullptr && binned_kv_cache != nullptr) { |
4247 | cache_size = pcm->get_tuned_mem(); | |
11fdf7f2 TL |
4248 | kv_alloc = binned_kv_cache->get_committed_size(); |
4249 | meta_alloc = meta_cache->get_committed_size(); | |
4250 | data_alloc = data_cache->get_committed_size(); | |
f67539c2 TL |
4251 | if (binned_kv_onode_cache != nullptr) { |
4252 | kv_onode_alloc = binned_kv_onode_cache->get_committed_size(); | |
4253 | } | |
91327a77 AA |
4254 | } |
4255 | ||
4256 | if (interval_stats) { | |
9f95a23c | 4257 | dout(5) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4258 | << " kv_alloc: " << kv_alloc |
4259 | << " kv_used: " << kv_used | |
f67539c2 TL |
4260 | << " kv_onode_alloc: " << kv_onode_alloc |
4261 | << " kv_onode_used: " << kv_onode_used | |
91327a77 AA |
4262 | << " meta_alloc: " << meta_alloc |
4263 | << " meta_used: " << meta_used | |
4264 | << " data_alloc: " << data_alloc | |
4265 | << " data_used: " << data_used << dendl; | |
4266 | } else { | |
9f95a23c | 4267 | dout(20) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4268 | << " kv_alloc: " << kv_alloc |
4269 | << " kv_used: " << kv_used | |
f67539c2 TL |
4270 | << " kv_onode_alloc: " << kv_onode_alloc |
4271 | << " kv_onode_used: " << kv_onode_used | |
91327a77 AA |
4272 | << " meta_alloc: " << meta_alloc |
4273 | << " meta_used: " << meta_used | |
4274 | << " data_alloc: " << data_alloc | |
4275 | << " data_used: " << data_used << dendl; | |
4276 | } | |
4277 | ||
4278 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
9f95a23c TL |
4279 | (meta_alloc / (double) onode_shards) / meta_cache->get_bytes_per_onode()); |
4280 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / buffer_shards); | |
91327a77 | 4281 | |
9f95a23c | 4282 | dout(30) << __func__ << " max_shard_onodes: " << max_shard_onodes |
91327a77 AA |
4283 | << " max_shard_buffer: " << max_shard_buffer << dendl; |
4284 | ||
9f95a23c TL |
4285 | for (auto i : store->onode_cache_shards) { |
4286 | i->set_max(max_shard_onodes); | |
4287 | } | |
4288 | for (auto i : store->buffer_cache_shards) { | |
4289 | i->set_max(max_shard_buffer); | |
91327a77 AA |
4290 | } |
4291 | } | |
4292 | ||
92f5a8d4 TL |
4293 | void BlueStore::MempoolThread::_update_cache_settings() |
4294 | { | |
4295 | // Nothing to do if pcm is not used. | |
4296 | if (pcm == nullptr) { | |
4297 | return; | |
4298 | } | |
4299 | ||
92f5a8d4 TL |
4300 | uint64_t target = store->osd_memory_target; |
4301 | uint64_t base = store->osd_memory_base; | |
4302 | uint64_t min = store->osd_memory_cache_min; | |
4303 | uint64_t max = min; | |
4304 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4305 | ||
4306 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4307 | if (ltarget > base + min) { | |
4308 | max = ltarget - base; | |
4309 | } | |
4310 | ||
4311 | // set pcm cache levels | |
4312 | pcm->set_target_memory(target); | |
4313 | pcm->set_min_memory(min); | |
4314 | pcm->set_max_memory(max); | |
4315 | ||
9f95a23c | 4316 | dout(5) << __func__ << " updated pcm target: " << target |
92f5a8d4 TL |
4317 | << " pcm min: " << min |
4318 | << " pcm max: " << max | |
4319 | << dendl; | |
4320 | } | |
4321 | ||
7c673cae FG |
4322 | // ======================================================= |
4323 | ||
31f18b77 FG |
4324 | // OmapIteratorImpl |
4325 | ||
4326 | #undef dout_prefix | |
4327 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
4328 | ||
4329 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
4330 | CollectionRef c, OnodeRef o, KeyValueDB::Iterator it) | |
4331 | : c(c), o(o), it(it) | |
4332 | { | |
9f95a23c | 4333 | std::shared_lock l(c->lock); |
31f18b77 | 4334 | if (o->onode.has_omap()) { |
9f95a23c TL |
4335 | o->get_omap_key(string(), &head); |
4336 | o->get_omap_tail(&tail); | |
31f18b77 FG |
4337 | it->lower_bound(head); |
4338 | } | |
4339 | } | |
4340 | ||
11fdf7f2 TL |
4341 | string BlueStore::OmapIteratorImpl::_stringify() const |
4342 | { | |
4343 | stringstream s; | |
4344 | s << " omap_iterator(cid = " << c->cid | |
4345 | <<", oid = " << o->oid << ")"; | |
4346 | return s.str(); | |
4347 | } | |
4348 | ||
31f18b77 FG |
4349 | int BlueStore::OmapIteratorImpl::seek_to_first() |
4350 | { | |
9f95a23c | 4351 | std::shared_lock l(c->lock); |
11fdf7f2 | 4352 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4353 | if (o->onode.has_omap()) { |
4354 | it->lower_bound(head); | |
4355 | } else { | |
4356 | it = KeyValueDB::Iterator(); | |
4357 | } | |
494da23a TL |
4358 | c->store->log_latency( |
4359 | __func__, | |
11fdf7f2 TL |
4360 | l_bluestore_omap_seek_to_first_lat, |
4361 | mono_clock::now() - start1, | |
494da23a | 4362 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 | 4363 | |
31f18b77 FG |
4364 | return 0; |
4365 | } | |
4366 | ||
4367 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
4368 | { | |
9f95a23c | 4369 | std::shared_lock l(c->lock); |
11fdf7f2 | 4370 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4371 | if (o->onode.has_omap()) { |
4372 | string key; | |
9f95a23c | 4373 | o->get_omap_key(after, &key); |
31f18b77 FG |
4374 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " |
4375 | << pretty_binary_string(key) << dendl; | |
4376 | it->upper_bound(key); | |
4377 | } else { | |
4378 | it = KeyValueDB::Iterator(); | |
4379 | } | |
11fdf7f2 | 4380 | c->store->log_latency_fn( |
494da23a | 4381 | __func__, |
11fdf7f2 TL |
4382 | l_bluestore_omap_upper_bound_lat, |
4383 | mono_clock::now() - start1, | |
494da23a | 4384 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4385 | [&] (const ceph::timespan& lat) { |
494da23a | 4386 | return ", after = " + after + |
11fdf7f2 TL |
4387 | _stringify(); |
4388 | } | |
4389 | ); | |
31f18b77 FG |
4390 | return 0; |
4391 | } | |
4392 | ||
4393 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
4394 | { | |
9f95a23c | 4395 | std::shared_lock l(c->lock); |
11fdf7f2 | 4396 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4397 | if (o->onode.has_omap()) { |
4398 | string key; | |
9f95a23c | 4399 | o->get_omap_key(to, &key); |
31f18b77 FG |
4400 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " |
4401 | << pretty_binary_string(key) << dendl; | |
4402 | it->lower_bound(key); | |
4403 | } else { | |
4404 | it = KeyValueDB::Iterator(); | |
4405 | } | |
11fdf7f2 | 4406 | c->store->log_latency_fn( |
494da23a | 4407 | __func__, |
11fdf7f2 TL |
4408 | l_bluestore_omap_lower_bound_lat, |
4409 | mono_clock::now() - start1, | |
494da23a | 4410 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4411 | [&] (const ceph::timespan& lat) { |
494da23a | 4412 | return ", to = " + to + |
11fdf7f2 TL |
4413 | _stringify(); |
4414 | } | |
4415 | ); | |
31f18b77 FG |
4416 | return 0; |
4417 | } | |
4418 | ||
4419 | bool BlueStore::OmapIteratorImpl::valid() | |
4420 | { | |
9f95a23c | 4421 | std::shared_lock l(c->lock); |
31f18b77 | 4422 | bool r = o->onode.has_omap() && it && it->valid() && |
494da23a | 4423 | it->raw_key().second < tail; |
31f18b77 FG |
4424 | if (it && it->valid()) { |
4425 | ldout(c->store->cct,20) << __func__ << " is at " | |
4426 | << pretty_binary_string(it->raw_key().second) | |
4427 | << dendl; | |
4428 | } | |
4429 | return r; | |
4430 | } | |
4431 | ||
11fdf7f2 | 4432 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 4433 | { |
11fdf7f2 | 4434 | int r = -1; |
9f95a23c | 4435 | std::shared_lock l(c->lock); |
11fdf7f2 | 4436 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4437 | if (o->onode.has_omap()) { |
4438 | it->next(); | |
11fdf7f2 | 4439 | r = 0; |
31f18b77 | 4440 | } |
494da23a TL |
4441 | c->store->log_latency( |
4442 | __func__, | |
11fdf7f2 TL |
4443 | l_bluestore_omap_next_lat, |
4444 | mono_clock::now() - start1, | |
494da23a | 4445 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 TL |
4446 | |
4447 | return r; | |
31f18b77 FG |
4448 | } |
4449 | ||
4450 | string BlueStore::OmapIteratorImpl::key() | |
4451 | { | |
9f95a23c | 4452 | std::shared_lock l(c->lock); |
11fdf7f2 | 4453 | ceph_assert(it->valid()); |
31f18b77 FG |
4454 | string db_key = it->raw_key().second; |
4455 | string user_key; | |
9f95a23c | 4456 | o->decode_omap_key(db_key, &user_key); |
494da23a | 4457 | |
31f18b77 FG |
4458 | return user_key; |
4459 | } | |
4460 | ||
4461 | bufferlist BlueStore::OmapIteratorImpl::value() | |
4462 | { | |
9f95a23c | 4463 | std::shared_lock l(c->lock); |
11fdf7f2 | 4464 | ceph_assert(it->valid()); |
31f18b77 FG |
4465 | return it->value(); |
4466 | } | |
4467 | ||
4468 | ||
4469 | // ===================================== | |
4470 | ||
7c673cae FG |
4471 | #undef dout_prefix |
4472 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
9f95a23c TL |
4473 | #undef dout_context |
4474 | #define dout_context cct | |
7c673cae FG |
4475 | |
4476 | ||
4477 | static void aio_cb(void *priv, void *priv2) | |
4478 | { | |
4479 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4480 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4481 | c->aio_finish(store); | |
4482 | } | |
4483 | ||
11fdf7f2 TL |
4484 | static void discard_cb(void *priv, void *priv2) |
4485 | { | |
4486 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4487 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4488 | store->handle_discard(*tmp); | |
4489 | } | |
4490 | ||
4491 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4492 | { | |
4493 | dout(10) << __func__ << dendl; | |
f67539c2 TL |
4494 | ceph_assert(shared_alloc.a); |
4495 | shared_alloc.a->release(to_release); | |
11fdf7f2 TL |
4496 | } |
4497 | ||
7c673cae | 4498 | BlueStore::BlueStore(CephContext *cct, const string& path) |
9f95a23c | 4499 | : BlueStore(cct, path, 0) {} |
7c673cae FG |
4500 | |
4501 | BlueStore::BlueStore(CephContext *cct, | |
4502 | const string& path, | |
4503 | uint64_t _min_alloc_size) | |
4504 | : ObjectStore(cct, path), | |
9f95a23c | 4505 | throttle(cct), |
11fdf7f2 | 4506 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4507 | kv_sync_thread(this), |
31f18b77 | 4508 | kv_finalize_thread(this), |
f67539c2 | 4509 | zoned_cleaner_thread(this), |
7c673cae FG |
4510 | min_alloc_size(_min_alloc_size), |
4511 | min_alloc_size_order(ctz(_min_alloc_size)), | |
4512 | mempool_thread(this) | |
4513 | { | |
4514 | _init_logger(); | |
11fdf7f2 | 4515 | cct->_conf.add_observer(this); |
7c673cae | 4516 | set_cache_shards(1); |
7c673cae FG |
4517 | } |
4518 | ||
4519 | BlueStore::~BlueStore() | |
4520 | { | |
11fdf7f2 | 4521 | cct->_conf.remove_observer(this); |
7c673cae | 4522 | _shutdown_logger(); |
11fdf7f2 TL |
4523 | ceph_assert(!mounted); |
4524 | ceph_assert(db == NULL); | |
4525 | ceph_assert(bluefs == NULL); | |
4526 | ceph_assert(fsid_fd < 0); | |
4527 | ceph_assert(path_fd < 0); | |
9f95a23c TL |
4528 | for (auto i : onode_cache_shards) { |
4529 | delete i; | |
4530 | } | |
4531 | for (auto i : buffer_cache_shards) { | |
7c673cae FG |
4532 | delete i; |
4533 | } | |
9f95a23c TL |
4534 | onode_cache_shards.clear(); |
4535 | buffer_cache_shards.clear(); | |
7c673cae FG |
4536 | } |
4537 | ||
4538 | const char **BlueStore::get_tracked_conf_keys() const | |
4539 | { | |
4540 | static const char* KEYS[] = { | |
4541 | "bluestore_csum_type", | |
4542 | "bluestore_compression_mode", | |
4543 | "bluestore_compression_algorithm", | |
4544 | "bluestore_compression_min_blob_size", | |
4545 | "bluestore_compression_min_blob_size_ssd", | |
4546 | "bluestore_compression_min_blob_size_hdd", | |
4547 | "bluestore_compression_max_blob_size", | |
4548 | "bluestore_compression_max_blob_size_ssd", | |
4549 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4550 | "bluestore_compression_required_ratio", |
7c673cae FG |
4551 | "bluestore_max_alloc_size", |
4552 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4553 | "bluestore_prefer_deferred_size_hdd", |
4554 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4555 | "bluestore_deferred_batch_ops", |
4556 | "bluestore_deferred_batch_ops_hdd", | |
4557 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4558 | "bluestore_throttle_bytes", |
4559 | "bluestore_throttle_deferred_bytes", | |
4560 | "bluestore_throttle_cost_per_io_hdd", | |
4561 | "bluestore_throttle_cost_per_io_ssd", | |
4562 | "bluestore_throttle_cost_per_io", | |
4563 | "bluestore_max_blob_size", | |
4564 | "bluestore_max_blob_size_ssd", | |
4565 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4566 | "osd_memory_target", |
4567 | "osd_memory_target_cgroup_limit_ratio", | |
4568 | "osd_memory_base", | |
4569 | "osd_memory_cache_min", | |
92f5a8d4 | 4570 | "osd_memory_expected_fragmentation", |
11fdf7f2 TL |
4571 | "bluestore_cache_autotune", |
4572 | "bluestore_cache_autotune_interval", | |
81eedcae | 4573 | "bluestore_warn_on_legacy_statfs", |
9f95a23c TL |
4574 | "bluestore_warn_on_no_per_pool_omap", |
4575 | "bluestore_max_defer_interval", | |
7c673cae FG |
4576 | NULL |
4577 | }; | |
4578 | return KEYS; | |
4579 | } | |
4580 | ||
11fdf7f2 | 4581 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4582 | const std::set<std::string> &changed) |
4583 | { | |
eafe8130 | 4584 | if (changed.count("bluestore_warn_on_legacy_statfs")) { |
81eedcae TL |
4585 | _check_legacy_statfs_alert(); |
4586 | } | |
f67539c2 TL |
4587 | if (changed.count("bluestore_warn_on_no_per_pool_omap") || |
4588 | changed.count("bluestore_warn_on_no_per_pg_omap")) { | |
4589 | _check_no_per_pg_or_pool_omap_alert(); | |
9f95a23c | 4590 | } |
81eedcae | 4591 | |
7c673cae FG |
4592 | if (changed.count("bluestore_csum_type")) { |
4593 | _set_csum(); | |
4594 | } | |
4595 | if (changed.count("bluestore_compression_mode") || | |
4596 | changed.count("bluestore_compression_algorithm") || | |
4597 | changed.count("bluestore_compression_min_blob_size") || | |
4598 | changed.count("bluestore_compression_max_blob_size")) { | |
4599 | if (bdev) { | |
4600 | _set_compression(); | |
4601 | } | |
4602 | } | |
4603 | if (changed.count("bluestore_max_blob_size") || | |
4604 | changed.count("bluestore_max_blob_size_ssd") || | |
4605 | changed.count("bluestore_max_blob_size_hdd")) { | |
4606 | if (bdev) { | |
4607 | // only after startup | |
4608 | _set_blob_size(); | |
4609 | } | |
4610 | } | |
4611 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4612 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4613 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4614 | changed.count("bluestore_max_alloc_size") || |
4615 | changed.count("bluestore_deferred_batch_ops") || | |
4616 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4617 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4618 | if (bdev) { | |
4619 | // only after startup | |
4620 | _set_alloc_sizes(); | |
4621 | } | |
4622 | } | |
4623 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4624 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4625 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4626 | if (bdev) { | |
4627 | _set_throttle_params(); | |
4628 | } | |
4629 | } | |
9f95a23c TL |
4630 | if (changed.count("bluestore_throttle_bytes") || |
4631 | changed.count("bluestore_throttle_deferred_bytes") || | |
4632 | changed.count("bluestore_throttle_trace_rate")) { | |
4633 | throttle.reset_throttle(conf); | |
7c673cae | 4634 | } |
9f95a23c TL |
4635 | if (changed.count("bluestore_max_defer_interval")) { |
4636 | if (bdev) { | |
4637 | _set_max_defer_interval(); | |
4638 | } | |
7c673cae | 4639 | } |
92f5a8d4 TL |
4640 | if (changed.count("osd_memory_target") || |
4641 | changed.count("osd_memory_base") || | |
4642 | changed.count("osd_memory_cache_min") || | |
4643 | changed.count("osd_memory_expected_fragmentation")) { | |
4644 | _update_osd_memory_options(); | |
4645 | } | |
7c673cae FG |
4646 | } |
4647 | ||
4648 | void BlueStore::_set_compression() | |
4649 | { | |
224ce89b WB |
4650 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4651 | if (m) { | |
11fdf7f2 | 4652 | _clear_compression_alert(); |
224ce89b WB |
4653 | comp_mode = *m; |
4654 | } else { | |
4655 | derr << __func__ << " unrecognized value '" | |
4656 | << cct->_conf->bluestore_compression_mode | |
4657 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4658 | << dendl; | |
4659 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4660 | string s("unknown mode: "); |
4661 | s += cct->_conf->bluestore_compression_mode; | |
4662 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4663 | } |
4664 | ||
4665 | compressor = nullptr; | |
4666 | ||
3efd9988 FG |
4667 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4668 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4669 | } else { |
11fdf7f2 | 4670 | ceph_assert(bdev); |
9f95a23c | 4671 | if (_use_rotational_settings()) { |
7c673cae FG |
4672 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; |
4673 | } else { | |
4674 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4675 | } | |
4676 | } | |
4677 | ||
4678 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4679 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4680 | } else { | |
11fdf7f2 | 4681 | ceph_assert(bdev); |
9f95a23c | 4682 | if (_use_rotational_settings()) { |
7c673cae FG |
4683 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; |
4684 | } else { | |
4685 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4686 | } | |
4687 | } | |
4688 | ||
7c673cae FG |
4689 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4690 | if (!alg_name.empty()) { | |
4691 | compressor = Compressor::create(cct, alg_name); | |
4692 | if (!compressor) { | |
4693 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4694 | << dendl; | |
11fdf7f2 | 4695 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4696 | } |
4697 | } | |
4698 | ||
4699 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4700 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4701 | << " min_blob " << comp_min_blob_size |
4702 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4703 | << dendl; |
4704 | } | |
4705 | ||
4706 | void BlueStore::_set_csum() | |
4707 | { | |
4708 | csum_type = Checksummer::CSUM_NONE; | |
4709 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4710 | if (t > Checksummer::CSUM_NONE) | |
4711 | csum_type = t; | |
4712 | ||
4713 | dout(10) << __func__ << " csum_type " | |
4714 | << Checksummer::get_csum_type_string(csum_type) | |
4715 | << dendl; | |
4716 | } | |
4717 | ||
4718 | void BlueStore::_set_throttle_params() | |
4719 | { | |
4720 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4721 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4722 | } else { | |
11fdf7f2 | 4723 | ceph_assert(bdev); |
9f95a23c | 4724 | if (_use_rotational_settings()) { |
7c673cae FG |
4725 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; |
4726 | } else { | |
4727 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4728 | } | |
4729 | } | |
4730 | ||
4731 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4732 | << dendl; | |
4733 | } | |
4734 | void BlueStore::_set_blob_size() | |
4735 | { | |
4736 | if (cct->_conf->bluestore_max_blob_size) { | |
4737 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4738 | } else { | |
11fdf7f2 | 4739 | ceph_assert(bdev); |
9f95a23c | 4740 | if (_use_rotational_settings()) { |
7c673cae FG |
4741 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; |
4742 | } else { | |
4743 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4744 | } | |
4745 | } | |
4746 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4747 | << std::dec << dendl; | |
4748 | } | |
4749 | ||
92f5a8d4 TL |
4750 | void BlueStore::_update_osd_memory_options() |
4751 | { | |
4752 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4753 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
4754 | osd_memory_expected_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
4755 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
4756 | config_changed++; | |
4757 | dout(10) << __func__ | |
4758 | << " osd_memory_target " << osd_memory_target | |
4759 | << " osd_memory_base " << osd_memory_base | |
4760 | << " osd_memory_expected_fragmentation " << osd_memory_expected_fragmentation | |
4761 | << " osd_memory_cache_min " << osd_memory_cache_min | |
4762 | << dendl; | |
4763 | } | |
4764 | ||
11fdf7f2 | 4765 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4766 | { |
11fdf7f2 TL |
4767 | ceph_assert(bdev); |
4768 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4769 | cache_autotune_interval = |
11fdf7f2 TL |
4770 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
4771 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4772 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4773 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4774 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4775 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4776 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4777 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4778 | |
224ce89b WB |
4779 | if (cct->_conf->bluestore_cache_size) { |
4780 | cache_size = cct->_conf->bluestore_cache_size; | |
4781 | } else { | |
4782 | // choose global cache size based on backend type | |
9f95a23c | 4783 | if (_use_rotational_settings()) { |
224ce89b WB |
4784 | cache_size = cct->_conf->bluestore_cache_size_hdd; |
4785 | } else { | |
4786 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4787 | } | |
4788 | } | |
31f18b77 | 4789 | |
f67539c2 | 4790 | cache_meta_ratio = cct->_conf.get_val<double>("bluestore_cache_meta_ratio"); |
224ce89b | 4791 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4792 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4793 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4794 | return -EINVAL; |
4795 | } | |
91327a77 | 4796 | |
f67539c2 | 4797 | cache_kv_ratio = cct->_conf.get_val<double>("bluestore_cache_kv_ratio"); |
224ce89b | 4798 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4799 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4800 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4801 | return -EINVAL; |
4802 | } | |
91327a77 | 4803 | |
f67539c2 TL |
4804 | cache_kv_onode_ratio = cct->_conf.get_val<double>("bluestore_cache_kv_onode_ratio"); |
4805 | if (cache_kv_onode_ratio < 0 || cache_kv_onode_ratio > 1.0) { | |
4806 | derr << __func__ << " bluestore_cache_kv_onode_ratio (" << cache_kv_onode_ratio | |
4807 | << ") must be in range [0,1.0]" << dendl; | |
4808 | return -EINVAL; | |
4809 | } | |
4810 | ||
31f18b77 | 4811 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4812 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4813 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4814 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4815 | << dendl; | |
31f18b77 FG |
4816 | return -EINVAL; |
4817 | } | |
91327a77 | 4818 | |
f67539c2 TL |
4819 | cache_data_ratio = (double)1.0 - |
4820 | (double)cache_meta_ratio - | |
4821 | (double)cache_kv_ratio - | |
4822 | (double)cache_kv_onode_ratio; | |
31f18b77 FG |
4823 | if (cache_data_ratio < 0) { |
4824 | // deal with floating point imprecision | |
4825 | cache_data_ratio = 0; | |
4826 | } | |
91327a77 | 4827 | |
224ce89b WB |
4828 | dout(1) << __func__ << " cache_size " << cache_size |
4829 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4830 | << " kv " << cache_kv_ratio |
4831 | << " data " << cache_data_ratio | |
4832 | << dendl; | |
4833 | return 0; | |
4834 | } | |
4835 | ||
3efd9988 FG |
4836 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4837 | { | |
4838 | bluestore_bdev_label_t label; | |
4839 | string p = path + "/block"; | |
4840 | int r = _read_bdev_label(cct, p, &label); | |
4841 | if (r < 0) { | |
4842 | return ObjectStore::write_meta(key, value); | |
4843 | } | |
4844 | label.meta[key] = value; | |
4845 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4846 | ceph_assert(r == 0); |
3efd9988 FG |
4847 | return ObjectStore::write_meta(key, value); |
4848 | } | |
4849 | ||
4850 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4851 | { | |
4852 | bluestore_bdev_label_t label; | |
4853 | string p = path + "/block"; | |
4854 | int r = _read_bdev_label(cct, p, &label); | |
4855 | if (r < 0) { | |
4856 | return ObjectStore::read_meta(key, value); | |
4857 | } | |
4858 | auto i = label.meta.find(key); | |
4859 | if (i == label.meta.end()) { | |
4860 | return ObjectStore::read_meta(key, value); | |
4861 | } | |
4862 | *value = i->second; | |
4863 | return 0; | |
4864 | } | |
4865 | ||
7c673cae FG |
4866 | void BlueStore::_init_logger() |
4867 | { | |
4868 | PerfCountersBuilder b(cct, "bluestore", | |
4869 | l_bluestore_first, l_bluestore_last); | |
4870 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
4871 | "Average kv_thread flush latency", | |
4872 | "fl_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4873 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
4874 | "Average kv_thread commit latency"); | |
11fdf7f2 TL |
4875 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", |
4876 | "Average kv_sync thread latency", | |
4877 | "ks_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4878 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
4879 | "Average kv_finalize thread latency", | |
4880 | "kf_l", PerfCountersBuilder::PRIO_INTERESTING); | |
7c673cae FG |
4881 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
4882 | "Average prepare state latency"); | |
4883 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", | |
4884 | "Average aio_wait state latency", | |
4885 | "io_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4886 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", | |
4887 | "Average io_done state latency"); | |
4888 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", | |
4889 | "Average kv_queued state latency"); | |
4890 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", | |
4891 | "Average kv_commiting state latency"); | |
4892 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", | |
4893 | "Average kv_done state latency"); | |
4894 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", | |
4895 | "Average deferred_queued state latency"); | |
4896 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", | |
4897 | "Average aio_wait state latency"); | |
4898 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", | |
4899 | "Average cleanup state latency"); | |
4900 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
4901 | "Average finishing state latency"); | |
4902 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
4903 | "Average done state latency"); | |
4904 | b.add_time_avg(l_bluestore_throttle_lat, "throttle_lat", | |
4905 | "Average submit throttle latency", | |
4906 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4907 | b.add_time_avg(l_bluestore_submit_lat, "submit_lat", | |
4908 | "Average submit latency", | |
4909 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4910 | b.add_time_avg(l_bluestore_commit_lat, "commit_lat", | |
4911 | "Average commit latency", | |
4912 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4913 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
4914 | "Average read latency", | |
4915 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4916 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", | |
4917 | "Average read onode metadata latency"); | |
4918 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", | |
4919 | "Average read latency"); | |
4920 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
4921 | "Average compress latency"); | |
4922 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
4923 | "Average decompress latency"); | |
4924 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", | |
4925 | "Average checksum latency"); | |
4926 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
4927 | "Sum for beneficial compress ops"); | |
4928 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
4929 | "Sum for compress ops rejected due to low net gain of space"); | |
4930 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", | |
11fdf7f2 | 4931 | "Sum for write-op padded bytes", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4932 | b.add_u64_counter(l_bluestore_deferred_write_ops, "deferred_write_ops", |
4933 | "Sum for deferred write op"); | |
4934 | b.add_u64_counter(l_bluestore_deferred_write_bytes, "deferred_write_bytes", | |
11fdf7f2 | 4935 | "Sum for deferred write bytes", "def", 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4936 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", |
4937 | "Sum for write penalty read ops"); | |
4938 | b.add_u64(l_bluestore_allocated, "bluestore_allocated", | |
4939 | "Sum for allocated bytes"); | |
4940 | b.add_u64(l_bluestore_stored, "bluestore_stored", | |
4941 | "Sum for stored bytes"); | |
4942 | b.add_u64(l_bluestore_compressed, "bluestore_compressed", | |
92f5a8d4 TL |
4943 | "Sum for stored compressed bytes", |
4944 | "c", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4945 | b.add_u64(l_bluestore_compressed_allocated, "bluestore_compressed_allocated", |
92f5a8d4 TL |
4946 | "Sum for bytes allocated for compressed data", |
4947 | "c_a", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4948 | b.add_u64(l_bluestore_compressed_original, "bluestore_compressed_original", |
92f5a8d4 TL |
4949 | "Sum for original bytes that were compressed", |
4950 | "c_o", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae FG |
4951 | b.add_u64(l_bluestore_onodes, "bluestore_onodes", |
4952 | "Number of onodes in cache"); | |
9f95a23c TL |
4953 | b.add_u64(l_bluestore_pinned_onodes, "bluestore_pinned_onodes", |
4954 | "Number of pinned onodes in cache"); | |
7c673cae FG |
4955 | b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits", |
4956 | "Sum for onode-lookups hit in the cache"); | |
4957 | b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses", | |
4958 | "Sum for onode-lookups missed in the cache"); | |
4959 | b.add_u64_counter(l_bluestore_onode_shard_hits, "bluestore_onode_shard_hits", | |
4960 | "Sum for onode-shard lookups hit in the cache"); | |
4961 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
4962 | "bluestore_onode_shard_misses", | |
4963 | "Sum for onode-shard lookups missed in the cache"); | |
4964 | b.add_u64(l_bluestore_extents, "bluestore_extents", | |
4965 | "Number of extents in cache"); | |
4966 | b.add_u64(l_bluestore_blobs, "bluestore_blobs", | |
4967 | "Number of blobs in cache"); | |
4968 | b.add_u64(l_bluestore_buffers, "bluestore_buffers", | |
4969 | "Number of buffers in cache"); | |
4970 | b.add_u64(l_bluestore_buffer_bytes, "bluestore_buffer_bytes", | |
11fdf7f2 | 4971 | "Number of buffer bytes in cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4972 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "bluestore_buffer_hit_bytes", |
11fdf7f2 | 4973 | "Sum for bytes of read hit in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4974 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "bluestore_buffer_miss_bytes", |
11fdf7f2 | 4975 | "Sum for bytes of read missed in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4976 | |
4977 | b.add_u64_counter(l_bluestore_write_big, "bluestore_write_big", | |
4978 | "Large aligned writes into fresh blobs"); | |
4979 | b.add_u64_counter(l_bluestore_write_big_bytes, "bluestore_write_big_bytes", | |
11fdf7f2 | 4980 | "Large aligned writes into fresh blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4981 | b.add_u64_counter(l_bluestore_write_big_blobs, "bluestore_write_big_blobs", |
4982 | "Large aligned writes into fresh blobs (blobs)"); | |
f67539c2 TL |
4983 | b.add_u64_counter(l_bluestore_write_big_deferred, |
4984 | "bluestore_write_big_deferred", | |
4985 | "Big overwrites using deferred"); | |
7c673cae FG |
4986 | b.add_u64_counter(l_bluestore_write_small, "bluestore_write_small", |
4987 | "Small writes into existing or sparse small blobs"); | |
4988 | b.add_u64_counter(l_bluestore_write_small_bytes, "bluestore_write_small_bytes", | |
11fdf7f2 | 4989 | "Small writes into existing or sparse small blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4990 | b.add_u64_counter(l_bluestore_write_small_unused, |
4991 | "bluestore_write_small_unused", | |
4992 | "Small writes into unused portion of existing blob"); | |
f67539c2 TL |
4993 | b.add_u64_counter(l_bluestore_write_deferred, |
4994 | "bluestore_write_deferred", | |
522d829b TL |
4995 | "Total deferred writes submitted"); |
4996 | b.add_u64_counter(l_bluestore_write_deferred_bytes, | |
4997 | "bluestore_write_deferred_bytes", | |
4998 | "Total bytes submitted as deferred writes"); | |
7c673cae FG |
4999 | b.add_u64_counter(l_bluestore_write_small_pre_read, |
5000 | "bluestore_write_small_pre_read", | |
5001 | "Small writes that required we read some data (possibly " | |
5002 | "cached) to fill out the block"); | |
f67539c2 TL |
5003 | b.add_u64_counter(l_bluestore_write_new, "bluestore_write_new", |
5004 | "Write into new blob"); | |
7c673cae FG |
5005 | |
5006 | b.add_u64_counter(l_bluestore_txc, "bluestore_txc", "Transactions committed"); | |
5007 | b.add_u64_counter(l_bluestore_onode_reshard, "bluestore_onode_reshard", | |
5008 | "Onode extent map reshard events"); | |
5009 | b.add_u64_counter(l_bluestore_blob_split, "bluestore_blob_split", | |
5010 | "Sum for blob splitting due to resharding"); | |
5011 | b.add_u64_counter(l_bluestore_extent_compress, "bluestore_extent_compress", | |
5012 | "Sum for extents that have been removed due to compression"); | |
5013 | b.add_u64_counter(l_bluestore_gc_merged, "bluestore_gc_merged", | |
5014 | "Sum for extents that have been merged due to garbage " | |
5015 | "collection"); | |
b32b8144 FG |
5016 | b.add_u64_counter(l_bluestore_read_eio, "bluestore_read_eio", |
5017 | "Read EIO errors propagated to high level callers"); | |
f64942e4 AA |
5018 | b.add_u64_counter(l_bluestore_reads_with_retries, "bluestore_reads_with_retries", |
5019 | "Read operations that required at least one retry due to failed checksum validation"); | |
a8e16298 TL |
5020 | b.add_u64(l_bluestore_fragmentation, "bluestore_fragmentation_micros", |
5021 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
11fdf7f2 TL |
5022 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
5023 | "Average omap iterator seek_to_first call latency"); | |
5024 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", | |
5025 | "Average omap iterator upper_bound call latency"); | |
5026 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", | |
5027 | "Average omap iterator lower_bound call latency"); | |
5028 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", | |
5029 | "Average omap iterator next call latency"); | |
adb31ebb TL |
5030 | b.add_time_avg(l_bluestore_omap_get_keys_lat, "omap_get_keys_lat", |
5031 | "Average omap get_keys call latency"); | |
5032 | b.add_time_avg(l_bluestore_omap_get_values_lat, "omap_get_values_lat", | |
5033 | "Average omap get_values call latency"); | |
494da23a TL |
5034 | b.add_time_avg(l_bluestore_clist_lat, "clist_lat", |
5035 | "Average collection listing latency"); | |
adb31ebb TL |
5036 | b.add_time_avg(l_bluestore_remove_lat, "remove_lat", |
5037 | "Average removal latency"); | |
5038 | ||
7c673cae FG |
5039 | logger = b.create_perf_counters(); |
5040 | cct->get_perfcounters_collection()->add(logger); | |
5041 | } | |
5042 | ||
5043 | int BlueStore::_reload_logger() | |
5044 | { | |
5045 | struct store_statfs_t store_statfs; | |
7c673cae | 5046 | int r = statfs(&store_statfs); |
11fdf7f2 | 5047 | if (r >= 0) { |
7c673cae | 5048 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
5049 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
5050 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
5051 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
5052 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
5053 | } |
5054 | return r; | |
5055 | } | |
5056 | ||
5057 | void BlueStore::_shutdown_logger() | |
5058 | { | |
5059 | cct->get_perfcounters_collection()->remove(logger); | |
5060 | delete logger; | |
5061 | } | |
5062 | ||
5063 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
5064 | uuid_d *fsid) | |
5065 | { | |
5066 | bluestore_bdev_label_t label; | |
5067 | int r = _read_bdev_label(cct, path, &label); | |
5068 | if (r < 0) | |
5069 | return r; | |
5070 | *fsid = label.osd_uuid; | |
5071 | return 0; | |
5072 | } | |
5073 | ||
5074 | int BlueStore::_open_path() | |
5075 | { | |
b32b8144 | 5076 | // sanity check(s) |
11fdf7f2 | 5077 | ceph_assert(path_fd < 0); |
91327a77 | 5078 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
5079 | if (path_fd < 0) { |
5080 | int r = -errno; | |
5081 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
5082 | << dendl; | |
5083 | return r; | |
5084 | } | |
5085 | return 0; | |
5086 | } | |
5087 | ||
5088 | void BlueStore::_close_path() | |
5089 | { | |
5090 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
5091 | path_fd = -1; | |
5092 | } | |
5093 | ||
3efd9988 FG |
5094 | int BlueStore::_write_bdev_label(CephContext *cct, |
5095 | string path, bluestore_bdev_label_t label) | |
7c673cae FG |
5096 | { |
5097 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
5098 | bufferlist bl; | |
11fdf7f2 | 5099 | encode(label, bl); |
7c673cae | 5100 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
5101 | encode(crc, bl); |
5102 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
5103 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
5104 | z.zero(); | |
5105 | bl.append(std::move(z)); | |
5106 | ||
91327a77 | 5107 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC)); |
7c673cae FG |
5108 | if (fd < 0) { |
5109 | fd = -errno; | |
5110 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5111 | << dendl; | |
5112 | return fd; | |
5113 | } | |
5114 | int r = bl.write_fd(fd); | |
5115 | if (r < 0) { | |
5116 | derr << __func__ << " failed to write to " << path | |
5117 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 5118 | goto out; |
7c673cae | 5119 | } |
3efd9988 FG |
5120 | r = ::fsync(fd); |
5121 | if (r < 0) { | |
5122 | derr << __func__ << " failed to fsync " << path | |
5123 | << ": " << cpp_strerror(r) << dendl; | |
5124 | } | |
11fdf7f2 | 5125 | out: |
7c673cae FG |
5126 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
5127 | return r; | |
5128 | } | |
5129 | ||
5130 | int BlueStore::_read_bdev_label(CephContext* cct, string path, | |
5131 | bluestore_bdev_label_t *label) | |
5132 | { | |
5133 | dout(10) << __func__ << dendl; | |
91327a77 | 5134 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
5135 | if (fd < 0) { |
5136 | fd = -errno; | |
5137 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5138 | << dendl; | |
5139 | return fd; | |
5140 | } | |
5141 | bufferlist bl; | |
5142 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
5143 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5144 | if (r < 0) { | |
5145 | derr << __func__ << " failed to read from " << path | |
5146 | << ": " << cpp_strerror(r) << dendl; | |
5147 | return r; | |
5148 | } | |
5149 | ||
5150 | uint32_t crc, expected_crc; | |
11fdf7f2 | 5151 | auto p = bl.cbegin(); |
7c673cae | 5152 | try { |
11fdf7f2 | 5153 | decode(*label, p); |
7c673cae FG |
5154 | bufferlist t; |
5155 | t.substr_of(bl, 0, p.get_off()); | |
5156 | crc = t.crc32c(-1); | |
11fdf7f2 | 5157 | decode(expected_crc, p); |
7c673cae | 5158 | } |
f67539c2 | 5159 | catch (ceph::buffer::error& e) { |
b32b8144 | 5160 | dout(2) << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
5161 | << ": " << e.what() |
5162 | << dendl; | |
b32b8144 | 5163 | return -ENOENT; |
7c673cae FG |
5164 | } |
5165 | if (crc != expected_crc) { | |
5166 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
5167 | << " != actual " << crc << dendl; | |
5168 | return -EIO; | |
5169 | } | |
5170 | dout(10) << __func__ << " got " << *label << dendl; | |
5171 | return 0; | |
5172 | } | |
5173 | ||
5174 | int BlueStore::_check_or_set_bdev_label( | |
5175 | string path, uint64_t size, string desc, bool create) | |
5176 | { | |
5177 | bluestore_bdev_label_t label; | |
5178 | if (create) { | |
5179 | label.osd_uuid = fsid; | |
5180 | label.size = size; | |
5181 | label.btime = ceph_clock_now(); | |
5182 | label.description = desc; | |
3efd9988 | 5183 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
5184 | if (r < 0) |
5185 | return r; | |
5186 | } else { | |
5187 | int r = _read_bdev_label(cct, path, &label); | |
5188 | if (r < 0) | |
5189 | return r; | |
31f18b77 FG |
5190 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
5191 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
5192 | << " and fsid " << fsid << " check bypassed" << dendl; | |
1911f103 | 5193 | } else if (label.osd_uuid != fsid) { |
7c673cae FG |
5194 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
5195 | << " does not match our fsid " << fsid << dendl; | |
5196 | return -EIO; | |
5197 | } | |
5198 | } | |
5199 | return 0; | |
5200 | } | |
5201 | ||
5202 | void BlueStore::_set_alloc_sizes(void) | |
5203 | { | |
7c673cae FG |
5204 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
5205 | ||
5206 | if (cct->_conf->bluestore_prefer_deferred_size) { | |
5207 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
5208 | } else { | |
11fdf7f2 | 5209 | ceph_assert(bdev); |
9f95a23c | 5210 | if (_use_rotational_settings()) { |
7c673cae FG |
5211 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; |
5212 | } else { | |
5213 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
5214 | } | |
5215 | } | |
5216 | ||
5217 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
5218 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
5219 | } else { | |
11fdf7f2 | 5220 | ceph_assert(bdev); |
9f95a23c | 5221 | if (_use_rotational_settings()) { |
7c673cae FG |
5222 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; |
5223 | } else { | |
5224 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
5225 | } | |
5226 | } | |
5227 | ||
5228 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 5229 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
5230 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
5231 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
5232 | << std::dec | |
5233 | << " deferred_batch_ops " << deferred_batch_ops | |
5234 | << dendl; | |
5235 | } | |
5236 | ||
5237 | int BlueStore::_open_bdev(bool create) | |
5238 | { | |
11fdf7f2 | 5239 | ceph_assert(bdev == NULL); |
7c673cae | 5240 | string p = path + "/block"; |
11fdf7f2 | 5241 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
5242 | int r = bdev->open(p); |
5243 | if (r < 0) | |
5244 | goto fail; | |
5245 | ||
11fdf7f2 TL |
5246 | if (create && cct->_conf->bdev_enable_discard) { |
5247 | bdev->discard(0, bdev->get_size()); | |
5248 | } | |
5249 | ||
7c673cae FG |
5250 | if (bdev->supported_bdev_label()) { |
5251 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
5252 | if (r < 0) | |
5253 | goto fail_close; | |
5254 | } | |
5255 | ||
5256 | // initialize global block parameters | |
5257 | block_size = bdev->get_block_size(); | |
5258 | block_mask = ~(block_size - 1); | |
5259 | block_size_order = ctz(block_size); | |
11fdf7f2 | 5260 | ceph_assert(block_size == 1u << block_size_order); |
9f95a23c | 5261 | _set_max_defer_interval(); |
224ce89b WB |
5262 | // and set cache_size based on device type |
5263 | r = _set_cache_sizes(); | |
5264 | if (r < 0) { | |
5265 | goto fail_close; | |
5266 | } | |
f67539c2 TL |
5267 | |
5268 | if (bdev->is_smr()) { | |
5269 | freelist_type = "zoned"; | |
5270 | } | |
7c673cae FG |
5271 | return 0; |
5272 | ||
5273 | fail_close: | |
5274 | bdev->close(); | |
5275 | fail: | |
5276 | delete bdev; | |
5277 | bdev = NULL; | |
5278 | return r; | |
5279 | } | |
5280 | ||
11fdf7f2 TL |
5281 | void BlueStore::_validate_bdev() |
5282 | { | |
5283 | ceph_assert(bdev); | |
11fdf7f2 | 5284 | uint64_t dev_size = bdev->get_size(); |
f67539c2 | 5285 | ceph_assert(dev_size > _get_ondisk_reserved()); |
11fdf7f2 TL |
5286 | } |
5287 | ||
7c673cae FG |
5288 | void BlueStore::_close_bdev() |
5289 | { | |
11fdf7f2 | 5290 | ceph_assert(bdev); |
7c673cae FG |
5291 | bdev->close(); |
5292 | delete bdev; | |
5293 | bdev = NULL; | |
5294 | } | |
5295 | ||
1911f103 | 5296 | int BlueStore::_open_fm(KeyValueDB::Transaction t, bool read_only) |
7c673cae | 5297 | { |
1911f103 | 5298 | int r; |
1911f103 | 5299 | |
11fdf7f2 TL |
5300 | ceph_assert(fm == NULL); |
5301 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); | |
5302 | ceph_assert(fm); | |
5303 | if (t) { | |
5304 | // create mode. initialize freespace | |
7c673cae | 5305 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
5306 | { |
5307 | bufferlist bl; | |
5308 | bl.append(freelist_type); | |
5309 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
5310 | } | |
b32b8144 FG |
5311 | // being able to allocate in units less than bdev block size |
5312 | // seems to be a bad idea. | |
11fdf7f2 | 5313 | ceph_assert( cct->_conf->bdev_block_size <= (int64_t)min_alloc_size); |
f67539c2 TL |
5314 | |
5315 | uint64_t alloc_size = min_alloc_size; | |
5316 | if (bdev->is_smr()) { | |
5317 | alloc_size = _zoned_piggyback_device_parameters_onto(alloc_size); | |
5318 | } | |
5319 | ||
5320 | fm->create(bdev->get_size(), alloc_size, t); | |
7c673cae FG |
5321 | |
5322 | // allocate superblock reserved space. note that we do not mark | |
5323 | // bluefs space as allocated in the freelist; we instead rely on | |
f67539c2 | 5324 | // bluefs doing that itself. |
11fdf7f2 | 5325 | auto reserved = _get_ondisk_reserved(); |
3efd9988 | 5326 | fm->allocate(0, reserved, t); |
7c673cae | 5327 | |
7c673cae FG |
5328 | if (cct->_conf->bluestore_debug_prefill > 0) { |
5329 | uint64_t end = bdev->get_size() - reserved; | |
5330 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
5331 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
5332 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 5333 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
5334 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
5335 | float r = cct->_conf->bluestore_debug_prefill; | |
5336 | r /= 1.0 - r; | |
5337 | bool stop = false; | |
5338 | ||
5339 | while (!stop && start < end) { | |
5340 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
5341 | if (start + l > end) { | |
5342 | l = end - start; | |
11fdf7f2 | 5343 | l = p2align(l, min_alloc_size); |
7c673cae | 5344 | } |
11fdf7f2 | 5345 | ceph_assert(start + l <= end); |
7c673cae FG |
5346 | |
5347 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 5348 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
5349 | if (start + l + u > end) { |
5350 | u = end - (start + l); | |
5351 | // trim to align so we don't overflow again | |
11fdf7f2 | 5352 | u = p2align(u, min_alloc_size); |
7c673cae FG |
5353 | stop = true; |
5354 | } | |
11fdf7f2 | 5355 | ceph_assert(start + l + u <= end); |
7c673cae | 5356 | |
11fdf7f2 | 5357 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
5358 | << " use 0x" << u << std::dec << dendl; |
5359 | ||
5360 | if (u == 0) { | |
5361 | // break if u has been trimmed to nothing | |
5362 | break; | |
5363 | } | |
5364 | ||
5365 | fm->allocate(start + l, u, t); | |
5366 | start += l + u; | |
5367 | } | |
5368 | } | |
f67539c2 | 5369 | r = _write_out_fm_meta(0); |
1911f103 TL |
5370 | ceph_assert(r == 0); |
5371 | } else { | |
f67539c2 TL |
5372 | r = fm->init(db, read_only, |
5373 | [&](const std::string& key, std::string* result) { | |
5374 | return read_meta(key, result); | |
5375 | }); | |
1911f103 | 5376 | if (r < 0) { |
f67539c2 | 5377 | derr << __func__ << " freelist init failed: " << cpp_strerror(r) << dendl; |
1911f103 TL |
5378 | delete fm; |
5379 | fm = NULL; | |
5380 | return r; | |
5381 | } | |
7c673cae | 5382 | } |
81eedcae TL |
5383 | // if space size tracked by free list manager is that higher than actual |
5384 | // dev size one can hit out-of-space allocation which will result | |
5385 | // in data loss and/or assertions | |
5386 | // Probably user altered the device size somehow. | |
5387 | // The only fix for now is to redeploy OSD. | |
5388 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
5389 | ostringstream ss; | |
5390 | ss << "slow device size mismatch detected, " | |
5391 | << " fm size(" << fm->get_size() | |
5392 | << ") > slow device size(" << bdev->get_size() | |
5393 | << "), Please stop using this OSD as it might cause data loss."; | |
5394 | _set_disk_size_mismatch_alert(ss.str()); | |
5395 | } | |
7c673cae FG |
5396 | return 0; |
5397 | } | |
5398 | ||
5399 | void BlueStore::_close_fm() | |
5400 | { | |
5401 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 5402 | ceph_assert(fm); |
7c673cae FG |
5403 | fm->shutdown(); |
5404 | delete fm; | |
5405 | fm = NULL; | |
5406 | } | |
5407 | ||
f67539c2 | 5408 | int BlueStore::_write_out_fm_meta(uint64_t target_size) |
1911f103 | 5409 | { |
f67539c2 | 5410 | int r = 0; |
1911f103 TL |
5411 | string p = path + "/block"; |
5412 | ||
5413 | std::vector<std::pair<string, string>> fm_meta; | |
5414 | fm->get_meta(target_size, &fm_meta); | |
5415 | ||
1911f103 | 5416 | for (auto& m : fm_meta) { |
f67539c2 TL |
5417 | r = write_meta(m.first, m.second); |
5418 | ceph_assert(r == 0); | |
1911f103 | 5419 | } |
1911f103 TL |
5420 | return r; |
5421 | } | |
5422 | ||
f67539c2 | 5423 | int BlueStore::_create_alloc() |
7c673cae | 5424 | { |
f67539c2 | 5425 | ceph_assert(shared_alloc.a == NULL); |
11fdf7f2 TL |
5426 | ceph_assert(bdev->get_size()); |
5427 | ||
f67539c2 TL |
5428 | uint64_t alloc_size = min_alloc_size; |
5429 | if (bdev->is_smr()) { | |
5430 | int r = _zoned_check_config_settings(); | |
5431 | if (r < 0) | |
11fdf7f2 | 5432 | return r; |
f67539c2 | 5433 | alloc_size = _zoned_piggyback_device_parameters_onto(alloc_size); |
11fdf7f2 TL |
5434 | } |
5435 | ||
f67539c2 TL |
5436 | shared_alloc.set(Allocator::create(cct, cct->_conf->bluestore_allocator, |
5437 | bdev->get_size(), | |
5438 | alloc_size, "block")); | |
5439 | ||
5440 | if (!shared_alloc.a) { | |
5441 | lderr(cct) << __func__ << "Failed to create allocator:: " | |
5442 | << cct->_conf->bluestore_allocator | |
5443 | << dendl; | |
7c673cae FG |
5444 | return -EINVAL; |
5445 | } | |
f67539c2 TL |
5446 | return 0; |
5447 | } | |
5448 | ||
5449 | int BlueStore::_init_alloc() | |
5450 | { | |
5451 | int r = _create_alloc(); | |
5452 | if (r < 0) { | |
5453 | return r; | |
5454 | } | |
5455 | ceph_assert(shared_alloc.a != NULL); | |
5456 | ||
5457 | if (bdev->is_smr()) { | |
5458 | shared_alloc.a->zoned_set_zone_states(fm->get_zone_states(db)); | |
5459 | } | |
7c673cae FG |
5460 | |
5461 | uint64_t num = 0, bytes = 0; | |
5462 | ||
5463 | dout(1) << __func__ << " opening allocation metadata" << dendl; | |
5464 | // initialize from freelist | |
5465 | fm->enumerate_reset(); | |
5466 | uint64_t offset, length; | |
11fdf7f2 | 5467 | while (fm->enumerate_next(db, &offset, &length)) { |
f67539c2 | 5468 | shared_alloc.a->init_add_free(offset, length); |
7c673cae FG |
5469 | ++num; |
5470 | bytes += length; | |
5471 | } | |
224ce89b | 5472 | fm->enumerate_reset(); |
7c673cae | 5473 | |
f67539c2 TL |
5474 | dout(1) << __func__ |
5475 | << " loaded " << byte_u_t(bytes) << " in " << num << " extents" | |
5476 | << std::hex | |
5477 | << ", allocator type " << shared_alloc.a->get_type() | |
5478 | << ", capacity 0x" << shared_alloc.a->get_capacity() | |
5479 | << ", block size 0x" << shared_alloc.a->get_block_size() | |
5480 | << ", free 0x" << shared_alloc.a->get_free() | |
5481 | << ", fragmentation " << shared_alloc.a->get_fragmentation() | |
5482 | << std::dec << dendl; | |
1911f103 | 5483 | |
7c673cae FG |
5484 | return 0; |
5485 | } | |
5486 | ||
5487 | void BlueStore::_close_alloc() | |
5488 | { | |
11fdf7f2 TL |
5489 | ceph_assert(bdev); |
5490 | bdev->discard_drain(); | |
5491 | ||
f67539c2 TL |
5492 | ceph_assert(shared_alloc.a); |
5493 | shared_alloc.a->shutdown(); | |
5494 | delete shared_alloc.a; | |
5495 | shared_alloc.reset(); | |
7c673cae FG |
5496 | } |
5497 | ||
5498 | int BlueStore::_open_fsid(bool create) | |
5499 | { | |
11fdf7f2 | 5500 | ceph_assert(fsid_fd < 0); |
91327a77 | 5501 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
5502 | if (create) |
5503 | flags |= O_CREAT; | |
5504 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
5505 | if (fsid_fd < 0) { | |
5506 | int err = -errno; | |
5507 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
5508 | return err; | |
5509 | } | |
5510 | return 0; | |
5511 | } | |
5512 | ||
5513 | int BlueStore::_read_fsid(uuid_d *uuid) | |
5514 | { | |
5515 | char fsid_str[40]; | |
5516 | memset(fsid_str, 0, sizeof(fsid_str)); | |
5517 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
5518 | if (ret < 0) { | |
5519 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
5520 | return ret; | |
5521 | } | |
5522 | if (ret > 36) | |
5523 | fsid_str[36] = 0; | |
5524 | else | |
5525 | fsid_str[ret] = 0; | |
5526 | if (!uuid->parse(fsid_str)) { | |
5527 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
5528 | return -EINVAL; | |
5529 | } | |
5530 | return 0; | |
5531 | } | |
5532 | ||
5533 | int BlueStore::_write_fsid() | |
5534 | { | |
5535 | int r = ::ftruncate(fsid_fd, 0); | |
5536 | if (r < 0) { | |
5537 | r = -errno; | |
5538 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
5539 | return r; | |
5540 | } | |
5541 | string str = stringify(fsid) + "\n"; | |
5542 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
5543 | if (r < 0) { | |
5544 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
5545 | return r; | |
5546 | } | |
5547 | r = ::fsync(fsid_fd); | |
5548 | if (r < 0) { | |
5549 | r = -errno; | |
5550 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
5551 | return r; | |
5552 | } | |
5553 | return 0; | |
5554 | } | |
5555 | ||
5556 | void BlueStore::_close_fsid() | |
5557 | { | |
5558 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
5559 | fsid_fd = -1; | |
5560 | } | |
5561 | ||
5562 | int BlueStore::_lock_fsid() | |
5563 | { | |
5564 | struct flock l; | |
5565 | memset(&l, 0, sizeof(l)); | |
5566 | l.l_type = F_WRLCK; | |
5567 | l.l_whence = SEEK_SET; | |
5568 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
5569 | if (r < 0) { | |
5570 | int err = errno; | |
5571 | derr << __func__ << " failed to lock " << path << "/fsid" | |
5572 | << " (is another ceph-osd still running?)" | |
5573 | << cpp_strerror(err) << dendl; | |
5574 | return -err; | |
5575 | } | |
5576 | return 0; | |
5577 | } | |
5578 | ||
31f18b77 FG |
5579 | bool BlueStore::is_rotational() |
5580 | { | |
5581 | if (bdev) { | |
5582 | return bdev->is_rotational(); | |
5583 | } | |
5584 | ||
5585 | bool rotational = true; | |
5586 | int r = _open_path(); | |
5587 | if (r < 0) | |
5588 | goto out; | |
5589 | r = _open_fsid(false); | |
5590 | if (r < 0) | |
5591 | goto out_path; | |
5592 | r = _read_fsid(&fsid); | |
5593 | if (r < 0) | |
5594 | goto out_fsid; | |
5595 | r = _lock_fsid(); | |
5596 | if (r < 0) | |
5597 | goto out_fsid; | |
5598 | r = _open_bdev(false); | |
5599 | if (r < 0) | |
5600 | goto out_fsid; | |
5601 | rotational = bdev->is_rotational(); | |
5602 | _close_bdev(); | |
5603 | out_fsid: | |
5604 | _close_fsid(); | |
5605 | out_path: | |
5606 | _close_path(); | |
5607 | out: | |
5608 | return rotational; | |
5609 | } | |
5610 | ||
d2e6a577 FG |
5611 | bool BlueStore::is_journal_rotational() |
5612 | { | |
5613 | if (!bluefs) { | |
5614 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
5615 | << dendl; | |
5616 | return is_rotational(); | |
5617 | } | |
5618 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
5619 | return bluefs->wal_is_rotational(); | |
5620 | } | |
5621 | ||
9f95a23c TL |
5622 | bool BlueStore::_use_rotational_settings() |
5623 | { | |
5624 | if (cct->_conf->bluestore_debug_enforce_settings == "hdd") { | |
5625 | return true; | |
5626 | } | |
5627 | if (cct->_conf->bluestore_debug_enforce_settings == "ssd") { | |
5628 | return false; | |
5629 | } | |
5630 | return bdev->is_rotational(); | |
5631 | } | |
5632 | ||
7c673cae FG |
5633 | bool BlueStore::test_mount_in_use() |
5634 | { | |
5635 | // most error conditions mean the mount is not in use (e.g., because | |
5636 | // it doesn't exist). only if we fail to lock do we conclude it is | |
5637 | // in use. | |
5638 | bool ret = false; | |
5639 | int r = _open_path(); | |
5640 | if (r < 0) | |
5641 | return false; | |
5642 | r = _open_fsid(false); | |
5643 | if (r < 0) | |
5644 | goto out_path; | |
5645 | r = _lock_fsid(); | |
5646 | if (r < 0) | |
5647 | ret = true; // if we can't lock, it is in use | |
5648 | _close_fsid(); | |
5649 | out_path: | |
5650 | _close_path(); | |
5651 | return ret; | |
5652 | } | |
5653 | ||
11fdf7f2 | 5654 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
5655 | { |
5656 | int r; | |
11fdf7f2 | 5657 | bluefs = new BlueFS(cct); |
7c673cae | 5658 | |
11fdf7f2 TL |
5659 | string bfn; |
5660 | struct stat st; | |
5661 | ||
5662 | bfn = path + "/block.db"; | |
5663 | if (::stat(bfn.c_str(), &st) == 0) { | |
eafe8130 TL |
5664 | r = bluefs->add_block_device( |
5665 | BlueFS::BDEV_DB, bfn, | |
f67539c2 TL |
5666 | create && cct->_conf->bdev_enable_discard, |
5667 | SUPER_RESERVED); | |
7c673cae | 5668 | if (r < 0) { |
11fdf7f2 TL |
5669 | derr << __func__ << " add block device(" << bfn << ") returned: " |
5670 | << cpp_strerror(r) << dendl; | |
5671 | goto free_bluefs; | |
7c673cae | 5672 | } |
7c673cae | 5673 | |
11fdf7f2 TL |
5674 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
5675 | r = _check_or_set_bdev_label( | |
5676 | bfn, | |
5677 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
5678 | "bluefs db", create); | |
5679 | if (r < 0) { | |
5680 | derr << __func__ | |
5681 | << " check block device(" << bfn << ") label returned: " | |
5682 | << cpp_strerror(r) << dendl; | |
5683 | goto free_bluefs; | |
5684 | } | |
7c673cae | 5685 | } |
9f95a23c TL |
5686 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
5687 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
5688 | } else { |
5689 | r = -errno; | |
5690 | if (::lstat(bfn.c_str(), &st) == -1) { | |
5691 | r = 0; | |
9f95a23c | 5692 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7c673cae | 5693 | } else { |
11fdf7f2 TL |
5694 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
5695 | << cpp_strerror(r) << dendl; | |
5696 | goto free_bluefs; | |
7c673cae FG |
5697 | } |
5698 | } | |
7c673cae | 5699 | |
11fdf7f2 TL |
5700 | // shared device |
5701 | bfn = path + "/block"; | |
5702 | // never trim here | |
9f95a23c | 5703 | r = bluefs->add_block_device(bluefs_layout.shared_bdev, bfn, false, |
f67539c2 TL |
5704 | 0, // no need to provide valid 'reserved' for shared dev |
5705 | &shared_alloc); | |
11fdf7f2 TL |
5706 | if (r < 0) { |
5707 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5708 | << cpp_strerror(r) << dendl; | |
5709 | goto free_bluefs; | |
5710 | } | |
11fdf7f2 TL |
5711 | |
5712 | bfn = path + "/block.wal"; | |
5713 | if (::stat(bfn.c_str(), &st) == 0) { | |
5714 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
f67539c2 TL |
5715 | create && cct->_conf->bdev_enable_discard, |
5716 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
5717 | if (r < 0) { |
5718 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5719 | << cpp_strerror(r) << dendl; | |
5720 | goto free_bluefs; | |
5721 | } | |
7c673cae | 5722 | |
11fdf7f2 TL |
5723 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
5724 | r = _check_or_set_bdev_label( | |
5725 | bfn, | |
5726 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
5727 | "bluefs wal", create); | |
7c673cae | 5728 | if (r < 0) { |
11fdf7f2 TL |
5729 | derr << __func__ << " check block device(" << bfn |
5730 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
5731 | goto free_bluefs; |
5732 | } | |
7c673cae FG |
5733 | } |
5734 | ||
9f95a23c | 5735 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
5736 | } else { |
5737 | r = 0; | |
5738 | if (::lstat(bfn.c_str(), &st) != -1) { | |
5739 | r = -errno; | |
5740 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
5741 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
5742 | goto free_bluefs; |
5743 | } | |
11fdf7f2 TL |
5744 | } |
5745 | return 0; | |
7c673cae | 5746 | |
11fdf7f2 TL |
5747 | free_bluefs: |
5748 | ceph_assert(bluefs); | |
5749 | delete bluefs; | |
5750 | bluefs = NULL; | |
5751 | return r; | |
5752 | } | |
7c673cae | 5753 | |
f67539c2 | 5754 | int BlueStore::_open_bluefs(bool create, bool read_only) |
11fdf7f2 TL |
5755 | { |
5756 | int r = _minimal_open_bluefs(create); | |
5757 | if (r < 0) { | |
5758 | return r; | |
5759 | } | |
f67539c2 | 5760 | BlueFSVolumeSelector* vselector = nullptr; |
9f95a23c TL |
5761 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
5762 | ||
5763 | string options = cct->_conf->bluestore_rocksdb_options; | |
cd265ab1 TL |
5764 | string options_annex = cct->_conf->bluestore_rocksdb_options_annex; |
5765 | if (!options_annex.empty()) { | |
5766 | if (!options.empty() && | |
5767 | *options.rbegin() != ',') { | |
5768 | options += ','; | |
5769 | } | |
5770 | options += options_annex; | |
5771 | } | |
9f95a23c TL |
5772 | |
5773 | rocksdb::Options rocks_opts; | |
f67539c2 | 5774 | r = RocksDBStore::ParseOptionsFromStringStatic( |
9f95a23c TL |
5775 | cct, |
5776 | options, | |
5777 | rocks_opts, | |
5778 | nullptr); | |
5779 | if (r < 0) { | |
5780 | return r; | |
5781 | } | |
f67539c2 TL |
5782 | if (cct->_conf->bluestore_volume_selection_policy == "fit_to_fast") { |
5783 | vselector = new FitToFastVolumeSelector( | |
9f95a23c TL |
5784 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, |
5785 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
f67539c2 TL |
5786 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100); |
5787 | } else { | |
5788 | double reserved_factor = cct->_conf->bluestore_volume_selection_reserved_factor; | |
5789 | vselector = | |
5790 | new RocksDBBlueFSVolumeSelector( | |
5791 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
5792 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
5793 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100, | |
5794 | 1024 * 1024 * 1024, //FIXME: set expected l0 size here | |
5795 | rocks_opts.max_bytes_for_level_base, | |
5796 | rocks_opts.max_bytes_for_level_multiplier, | |
5797 | reserved_factor, | |
5798 | cct->_conf->bluestore_volume_selection_reserved, | |
5799 | cct->_conf->bluestore_volume_selection_policy == "use_some_extra"); | |
5800 | } | |
9f95a23c | 5801 | } |
11fdf7f2 | 5802 | if (create) { |
9f95a23c | 5803 | bluefs->mkfs(fsid, bluefs_layout); |
11fdf7f2 | 5804 | } |
9f95a23c | 5805 | bluefs->set_volume_selector(vselector); |
11fdf7f2 TL |
5806 | r = bluefs->mount(); |
5807 | if (r < 0) { | |
5808 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
5809 | } | |
9f95a23c | 5810 | ceph_assert_always(bluefs->maybe_verify_layout(bluefs_layout) == 0); |
11fdf7f2 TL |
5811 | return r; |
5812 | } | |
5813 | ||
1911f103 | 5814 | void BlueStore::_close_bluefs(bool cold_close) |
11fdf7f2 | 5815 | { |
1911f103 | 5816 | bluefs->umount(cold_close); |
11fdf7f2 TL |
5817 | _minimal_close_bluefs(); |
5818 | } | |
5819 | ||
5820 | void BlueStore::_minimal_close_bluefs() | |
5821 | { | |
5822 | delete bluefs; | |
5823 | bluefs = NULL; | |
5824 | } | |
5825 | ||
5826 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
5827 | { | |
5828 | if (create) { | |
5829 | *ret = cct->_conf->bluestore_bluefs; | |
5830 | } else { | |
5831 | string s; | |
5832 | int r = read_meta("bluefs", &s); | |
5833 | if (r < 0) { | |
5834 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
5835 | return -EIO; | |
5836 | } | |
5837 | if (s == "1") { | |
5838 | *ret = true; | |
5839 | } else if (s == "0") { | |
5840 | *ret = false; | |
31f18b77 | 5841 | } else { |
11fdf7f2 TL |
5842 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
5843 | << dendl; | |
5844 | return -EIO; | |
5845 | } | |
5846 | } | |
5847 | return 0; | |
5848 | } | |
5849 | ||
5850 | /* | |
5851 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
5852 | * in the proper order | |
5853 | */ | |
f67539c2 | 5854 | int BlueStore::_open_db_and_around(bool read_only, bool to_repair) |
11fdf7f2 | 5855 | { |
f67539c2 TL |
5856 | dout(0) << __func__ << " read-only:" << read_only |
5857 | << " repair:" << to_repair << dendl; | |
5858 | { | |
5859 | string type; | |
5860 | int r = read_meta("type", &type); | |
5861 | if (r < 0) { | |
5862 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
5863 | << dendl; | |
11fdf7f2 | 5864 | return r; |
f67539c2 | 5865 | } |
11fdf7f2 | 5866 | |
f67539c2 TL |
5867 | if (type != "bluestore") { |
5868 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
5869 | return -EIO; | |
11fdf7f2 | 5870 | } |
f67539c2 | 5871 | } |
11fdf7f2 | 5872 | |
f67539c2 TL |
5873 | int r = _open_path(); |
5874 | if (r < 0) | |
5875 | return r; | |
5876 | r = _open_fsid(false); | |
5877 | if (r < 0) | |
5878 | goto out_path; | |
11fdf7f2 | 5879 | |
f67539c2 TL |
5880 | r = _read_fsid(&fsid); |
5881 | if (r < 0) | |
5882 | goto out_fsid; | |
11fdf7f2 | 5883 | |
f67539c2 TL |
5884 | r = _lock_fsid(); |
5885 | if (r < 0) | |
5886 | goto out_fsid; | |
11fdf7f2 | 5887 | |
f67539c2 TL |
5888 | r = _open_bdev(false); |
5889 | if (r < 0) | |
5890 | goto out_fsid; | |
7c673cae | 5891 | |
f67539c2 TL |
5892 | // open in read-only first to read FM list and init allocator |
5893 | // as they might be needed for some BlueFS procedures | |
5894 | r = _open_db(false, false, true); | |
5895 | if (r < 0) | |
5896 | goto out_bdev; | |
11fdf7f2 | 5897 | |
f67539c2 TL |
5898 | r = _open_super_meta(); |
5899 | if (r < 0) { | |
5900 | goto out_db; | |
5901 | } | |
5902 | ||
5903 | r = _open_fm(nullptr, true); | |
5904 | if (r < 0) | |
5905 | goto out_db; | |
5906 | ||
5907 | r = _init_alloc(); | |
5908 | if (r < 0) | |
5909 | goto out_fm; | |
5910 | ||
5911 | // Re-open in the proper mode(s). | |
5912 | ||
5913 | // Can't simply bypass second open for read-only mode as we need to | |
5914 | // load allocated extents from bluefs into allocator. | |
5915 | // And now it's time to do that | |
5916 | // | |
5917 | _close_db(true); | |
5918 | ||
5919 | r = _open_db(false, to_repair, read_only); | |
5920 | if (r < 0) { | |
5921 | goto out_alloc; | |
11fdf7f2 TL |
5922 | } |
5923 | return 0; | |
5924 | ||
f67539c2 TL |
5925 | out_alloc: |
5926 | _close_alloc(); | |
5927 | out_fm: | |
11fdf7f2 TL |
5928 | _close_fm(); |
5929 | out_db: | |
1911f103 | 5930 | _close_db(read_only); |
f67539c2 TL |
5931 | out_bdev: |
5932 | _close_bdev(); | |
5933 | out_fsid: | |
5934 | _close_fsid(); | |
5935 | out_path: | |
5936 | _close_path(); | |
11fdf7f2 TL |
5937 | return r; |
5938 | } | |
5939 | ||
1911f103 | 5940 | void BlueStore::_close_db_and_around(bool read_only) |
11fdf7f2 | 5941 | { |
f67539c2 TL |
5942 | _close_db(read_only); |
5943 | _close_fm(); | |
5944 | _close_alloc(); | |
5945 | _close_bdev(); | |
5946 | _close_fsid(); | |
5947 | _close_path(); | |
5948 | } | |
5949 | ||
5950 | int BlueStore::open_db_environment(KeyValueDB **pdb, bool to_repair) | |
5951 | { | |
5952 | _kv_only = true; | |
5953 | int r = _open_db_and_around(false, to_repair); | |
5954 | if (r == 0) { | |
5955 | *pdb = db; | |
11fdf7f2 | 5956 | } else { |
f67539c2 | 5957 | *pdb = nullptr; |
11fdf7f2 | 5958 | } |
f67539c2 | 5959 | return r; |
11fdf7f2 TL |
5960 | } |
5961 | ||
f67539c2 | 5962 | int BlueStore::close_db_environment() |
11fdf7f2 | 5963 | { |
f67539c2 TL |
5964 | _close_db_and_around(false); |
5965 | return 0; | |
11fdf7f2 TL |
5966 | } |
5967 | ||
f67539c2 TL |
5968 | int BlueStore::_prepare_db_environment(bool create, bool read_only, |
5969 | std::string* _fn, std::string* _kv_backend) | |
11fdf7f2 TL |
5970 | { |
5971 | int r; | |
5972 | ceph_assert(!db); | |
f67539c2 TL |
5973 | std::string& fn=*_fn; |
5974 | std::string& kv_backend=*_kv_backend; | |
5975 | fn = path + "/db"; | |
11fdf7f2 TL |
5976 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); |
5977 | ||
11fdf7f2 TL |
5978 | if (create) { |
5979 | kv_backend = cct->_conf->bluestore_kvbackend; | |
5980 | } else { | |
5981 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 5982 | if (r < 0) { |
11fdf7f2 TL |
5983 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
5984 | return -EIO; | |
5985 | } | |
5986 | } | |
5987 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
5988 | ||
5989 | bool do_bluefs; | |
5990 | r = _is_bluefs(create, &do_bluefs); | |
5991 | if (r < 0) { | |
5992 | return r; | |
5993 | } | |
5994 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
5995 | ||
5996 | map<string,string> kv_options; | |
5997 | // force separate wal dir for all new deployments. | |
5998 | kv_options["separate_wal_dir"] = 1; | |
5999 | rocksdb::Env *env = NULL; | |
6000 | if (do_bluefs) { | |
6001 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
6002 | if (kv_backend != "rocksdb") { | |
6003 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
6004 | return -EINVAL; | |
7c673cae | 6005 | } |
11fdf7f2 | 6006 | |
f67539c2 | 6007 | r = _open_bluefs(create, read_only); |
11fdf7f2 TL |
6008 | if (r < 0) { |
6009 | return r; | |
6010 | } | |
11fdf7f2 | 6011 | |
7c673cae | 6012 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
9f95a23c TL |
6013 | rocksdb::Env* a = new BlueRocksEnv(bluefs); |
6014 | rocksdb::Env* b = rocksdb::Env::Default(); | |
7c673cae | 6015 | if (create) { |
9f95a23c TL |
6016 | string cmd = "rm -rf " + path + "/db " + |
6017 | path + "/db.slow " + | |
6018 | path + "/db.wal"; | |
6019 | int r = system(cmd.c_str()); | |
6020 | (void)r; | |
7c673cae FG |
6021 | } |
6022 | env = new rocksdb::EnvMirror(b, a, false, true); | |
1911f103 | 6023 | } else { |
7c673cae FG |
6024 | env = new BlueRocksEnv(bluefs); |
6025 | ||
6026 | // simplify the dir names, too, as "seen" by rocksdb | |
6027 | fn = "db"; | |
6028 | } | |
9f95a23c TL |
6029 | BlueFSVolumeSelector::paths paths; |
6030 | bluefs->get_vselector_paths(fn, paths); | |
7c673cae | 6031 | |
522d829b | 6032 | { |
7c673cae | 6033 | ostringstream db_paths; |
9f95a23c TL |
6034 | bool first = true; |
6035 | for (auto& p : paths) { | |
6036 | if (!first) { | |
6037 | db_paths << " "; | |
6038 | } | |
6039 | first = false; | |
6040 | db_paths << p.first << "," << p.second; | |
6041 | ||
6042 | } | |
11fdf7f2 | 6043 | kv_options["db_paths"] = db_paths.str(); |
9f95a23c | 6044 | dout(1) << __func__ << " set db_paths to " << db_paths.str() << dendl; |
7c673cae FG |
6045 | } |
6046 | ||
6047 | if (create) { | |
9f95a23c TL |
6048 | for (auto& p : paths) { |
6049 | env->CreateDir(p.first); | |
6050 | } | |
6051 | // Selectors don't provide wal path so far hence create explicitly | |
11fdf7f2 | 6052 | env->CreateDir(fn + ".wal"); |
11fdf7f2 TL |
6053 | } else { |
6054 | std::vector<std::string> res; | |
6055 | // check for dir presence | |
6056 | auto r = env->GetChildren(fn+".wal", &res); | |
6057 | if (r.IsNotFound()) { | |
6058 | kv_options.erase("separate_wal_dir"); | |
6059 | } | |
7c673cae | 6060 | } |
11fdf7f2 TL |
6061 | } else { |
6062 | string walfn = path + "/db.wal"; | |
7c673cae | 6063 | |
11fdf7f2 TL |
6064 | if (create) { |
6065 | int r = ::mkdir(fn.c_str(), 0755); | |
6066 | if (r < 0) | |
6067 | r = -errno; | |
6068 | if (r < 0 && r != -EEXIST) { | |
6069 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
6070 | << dendl; | |
6071 | return r; | |
6072 | } | |
6073 | ||
6074 | // wal_dir, too! | |
7c673cae FG |
6075 | r = ::mkdir(walfn.c_str(), 0755); |
6076 | if (r < 0) | |
6077 | r = -errno; | |
6078 | if (r < 0 && r != -EEXIST) { | |
6079 | derr << __func__ << " failed to create " << walfn | |
6080 | << ": " << cpp_strerror(r) | |
6081 | << dendl; | |
6082 | return r; | |
6083 | } | |
11fdf7f2 TL |
6084 | } else { |
6085 | struct stat st; | |
6086 | r = ::stat(walfn.c_str(), &st); | |
6087 | if (r < 0 && errno == ENOENT) { | |
6088 | kv_options.erase("separate_wal_dir"); | |
6089 | } | |
7c673cae FG |
6090 | } |
6091 | } | |
6092 | ||
91327a77 | 6093 | |
7c673cae FG |
6094 | db = KeyValueDB::create(cct, |
6095 | kv_backend, | |
6096 | fn, | |
11fdf7f2 | 6097 | kv_options, |
7c673cae FG |
6098 | static_cast<void*>(env)); |
6099 | if (!db) { | |
6100 | derr << __func__ << " error creating db" << dendl; | |
6101 | if (bluefs) { | |
1911f103 | 6102 | _close_bluefs(read_only); |
7c673cae FG |
6103 | } |
6104 | // delete env manually here since we can't depend on db to do this | |
6105 | // under this case | |
6106 | delete env; | |
6107 | env = NULL; | |
6108 | return -EIO; | |
6109 | } | |
6110 | ||
f67539c2 | 6111 | FreelistManager::setup_merge_operators(db, freelist_type); |
7c673cae | 6112 | db->set_merge_operator(PREFIX_STAT, merge_op); |
91327a77 | 6113 | db->set_cache_size(cache_kv_ratio * cache_size); |
f67539c2 TL |
6114 | return 0; |
6115 | } | |
31f18b77 | 6116 | |
f67539c2 TL |
6117 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) |
6118 | { | |
6119 | int r; | |
6120 | ceph_assert(!(create && read_only)); | |
6121 | string options; | |
6122 | string options_annex; | |
6123 | stringstream err; | |
6124 | string kv_dir_fn; | |
6125 | string kv_backend; | |
6126 | std::string sharding_def; | |
6127 | r = _prepare_db_environment(create, read_only, &kv_dir_fn, &kv_backend); | |
6128 | if (r < 0) { | |
6129 | derr << __func__ << " failed to prepare db environment: " << err.str() << dendl; | |
6130 | return -EIO; | |
6131 | } | |
11fdf7f2 | 6132 | if (kv_backend == "rocksdb") { |
7c673cae | 6133 | options = cct->_conf->bluestore_rocksdb_options; |
cd265ab1 TL |
6134 | options_annex = cct->_conf->bluestore_rocksdb_options_annex; |
6135 | if (!options_annex.empty()) { | |
6136 | if (!options.empty() && | |
6137 | *options.rbegin() != ',') { | |
6138 | options += ','; | |
6139 | } | |
6140 | options += options_annex; | |
6141 | } | |
11fdf7f2 | 6142 | |
f67539c2 TL |
6143 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { |
6144 | sharding_def = cct->_conf.get_val<std::string>("bluestore_rocksdb_cfs"); | |
11fdf7f2 TL |
6145 | } |
6146 | } | |
6147 | ||
7c673cae | 6148 | db->init(options); |
11fdf7f2 TL |
6149 | if (to_repair_db) |
6150 | return 0; | |
6151 | if (create) { | |
f67539c2 | 6152 | r = db->create_and_open(err, sharding_def); |
11fdf7f2 TL |
6153 | } else { |
6154 | // we pass in cf list here, but it is only used if the db already has | |
6155 | // column families created. | |
6156 | r = read_only ? | |
f67539c2 TL |
6157 | db->open_read_only(err, sharding_def) : |
6158 | db->open(err, sharding_def); | |
11fdf7f2 | 6159 | } |
7c673cae FG |
6160 | if (r) { |
6161 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
1911f103 | 6162 | _close_db(read_only); |
7c673cae FG |
6163 | return -EIO; |
6164 | } | |
6165 | dout(1) << __func__ << " opened " << kv_backend | |
f67539c2 | 6166 | << " path " << kv_dir_fn << " options " << options << dendl; |
7c673cae | 6167 | return 0; |
7c673cae FG |
6168 | } |
6169 | ||
1911f103 | 6170 | void BlueStore::_close_db(bool cold_close) |
7c673cae | 6171 | { |
11fdf7f2 | 6172 | ceph_assert(db); |
7c673cae FG |
6173 | delete db; |
6174 | db = NULL; | |
6175 | if (bluefs) { | |
1911f103 | 6176 | _close_bluefs(cold_close); |
7c673cae FG |
6177 | } |
6178 | } | |
6179 | ||
11fdf7f2 | 6180 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 6181 | { |
11fdf7f2 TL |
6182 | auto dump_interval = |
6183 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
6184 | if (dump_interval > 0 && | |
6185 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
f67539c2 | 6186 | shared_alloc.a->dump(); |
11fdf7f2 TL |
6187 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); |
6188 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 6189 | } |
11fdf7f2 | 6190 | } |
7c673cae | 6191 | |
eafe8130 | 6192 | int BlueStore::_open_collections() |
7c673cae | 6193 | { |
28e407b8 | 6194 | dout(10) << __func__ << dendl; |
eafe8130 | 6195 | collections_had_errors = false; |
11fdf7f2 | 6196 | ceph_assert(coll_map.empty()); |
7c673cae FG |
6197 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
6198 | for (it->upper_bound(string()); | |
6199 | it->valid(); | |
6200 | it->next()) { | |
6201 | coll_t cid; | |
6202 | if (cid.parse(it->key())) { | |
9f95a23c | 6203 | auto c = ceph::make_ref<Collection>( |
7c673cae | 6204 | this, |
9f95a23c TL |
6205 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
6206 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
6207 | cid); | |
7c673cae | 6208 | bufferlist bl = it->value(); |
11fdf7f2 | 6209 | auto p = bl.cbegin(); |
7c673cae | 6210 | try { |
11fdf7f2 | 6211 | decode(c->cnode, p); |
f67539c2 | 6212 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
6213 | derr << __func__ << " failed to decode cnode, key:" |
6214 | << pretty_binary_string(it->key()) << dendl; | |
6215 | return -EIO; | |
6216 | } | |
28e407b8 AA |
6217 | dout(20) << __func__ << " opened " << cid << " " << c |
6218 | << " " << c->cnode << dendl; | |
11fdf7f2 | 6219 | _osr_attach(c.get()); |
7c673cae | 6220 | coll_map[cid] = c; |
11fdf7f2 | 6221 | |
7c673cae FG |
6222 | } else { |
6223 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
eafe8130 | 6224 | collections_had_errors = true; |
7c673cae FG |
6225 | } |
6226 | } | |
6227 | return 0; | |
6228 | } | |
6229 | ||
eafe8130 TL |
6230 | void BlueStore::_fsck_collections(int64_t* errors) |
6231 | { | |
6232 | if (collections_had_errors) { | |
6233 | dout(10) << __func__ << dendl; | |
f67539c2 | 6234 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
6235 | for (it->upper_bound(string()); |
6236 | it->valid(); | |
6237 | it->next()) { | |
6238 | coll_t cid; | |
6239 | if (!cid.parse(it->key())) { | |
6240 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
6241 | if (errors) { | |
6242 | (*errors)++; | |
6243 | } | |
6244 | } | |
6245 | } | |
6246 | } | |
6247 | } | |
6248 | ||
9f95a23c TL |
6249 | void BlueStore::_set_per_pool_omap() |
6250 | { | |
f67539c2 | 6251 | per_pool_omap = OMAP_BULK; |
9f95a23c TL |
6252 | bufferlist bl; |
6253 | db->get(PREFIX_SUPER, "per_pool_omap", &bl); | |
6254 | if (bl.length()) { | |
f67539c2 TL |
6255 | auto s = bl.to_str(); |
6256 | if (s == stringify(OMAP_PER_POOL)) { | |
6257 | per_pool_omap = OMAP_PER_POOL; | |
a4b75251 | 6258 | } else if (s == stringify(OMAP_PER_PG)) { |
f67539c2 | 6259 | per_pool_omap = OMAP_PER_PG; |
a4b75251 TL |
6260 | } else { |
6261 | ceph_assert(s == stringify(OMAP_BULK)); | |
f67539c2 TL |
6262 | } |
6263 | dout(10) << __func__ << " per_pool_omap = " << per_pool_omap << dendl; | |
9f95a23c TL |
6264 | } else { |
6265 | dout(10) << __func__ << " per_pool_omap not present" << dendl; | |
6266 | } | |
f67539c2 | 6267 | _check_no_per_pg_or_pool_omap_alert(); |
9f95a23c TL |
6268 | } |
6269 | ||
224ce89b | 6270 | void BlueStore::_open_statfs() |
31f18b77 | 6271 | { |
11fdf7f2 TL |
6272 | osd_pools.clear(); |
6273 | vstatfs.reset(); | |
6274 | ||
31f18b77 | 6275 | bufferlist bl; |
11fdf7f2 | 6276 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 6277 | if (r >= 0) { |
11fdf7f2 | 6278 | per_pool_stat_collection = false; |
31f18b77 | 6279 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 6280 | auto it = bl.cbegin(); |
31f18b77 | 6281 | vstatfs.decode(it); |
11fdf7f2 | 6282 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 6283 | } else { |
31f18b77 FG |
6284 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
6285 | } | |
81eedcae | 6286 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
6287 | } else { |
6288 | per_pool_stat_collection = true; | |
6289 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
f67539c2 | 6290 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 TL |
6291 | for (it->upper_bound(string()); |
6292 | it->valid(); | |
6293 | it->next()) { | |
6294 | ||
6295 | uint64_t pool_id; | |
6296 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6297 | ceph_assert(r == 0); | |
6298 | ||
6299 | bufferlist bl; | |
6300 | bl = it->value(); | |
6301 | auto p = bl.cbegin(); | |
6302 | auto& st = osd_pools[pool_id]; | |
6303 | try { | |
6304 | st.decode(p); | |
6305 | vstatfs += st; | |
6306 | ||
6307 | dout(30) << __func__ << " pool " << pool_id | |
6308 | << " statfs " << st << dendl; | |
f67539c2 | 6309 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
6310 | derr << __func__ << " failed to decode pool stats, key:" |
6311 | << pretty_binary_string(it->key()) << dendl; | |
6312 | } | |
6313 | } | |
31f18b77 | 6314 | } |
11fdf7f2 TL |
6315 | dout(30) << __func__ << " statfs " << vstatfs << dendl; |
6316 | ||
31f18b77 FG |
6317 | } |
6318 | ||
7c673cae FG |
6319 | int BlueStore::_setup_block_symlink_or_file( |
6320 | string name, | |
6321 | string epath, | |
6322 | uint64_t size, | |
6323 | bool create) | |
6324 | { | |
6325 | dout(20) << __func__ << " name " << name << " path " << epath | |
6326 | << " size " << size << " create=" << (int)create << dendl; | |
6327 | int r = 0; | |
91327a77 | 6328 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6329 | if (create) |
6330 | flags |= O_CREAT; | |
6331 | if (epath.length()) { | |
6332 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
6333 | if (r < 0) { | |
6334 | r = -errno; | |
6335 | derr << __func__ << " failed to create " << name << " symlink to " | |
6336 | << epath << ": " << cpp_strerror(r) << dendl; | |
6337 | return r; | |
6338 | } | |
6339 | ||
6340 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
6341 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
6342 | if (fd < 0) { | |
6343 | r = -errno; | |
6344 | derr << __func__ << " failed to open " << epath << " file: " | |
6345 | << cpp_strerror(r) << dendl; | |
6346 | return r; | |
6347 | } | |
11fdf7f2 TL |
6348 | // write the Transport ID of the NVMe device |
6349 | // a transport id looks like: "trtype:PCIe traddr:0000:02:00.0" | |
6350 | // where "0000:02:00.0" is the selector of a PCI device, see | |
6351 | // the first column of "lspci -mm -n -D" | |
6352 | string trid{"trtype:PCIe "}; | |
6353 | trid += "traddr:"; | |
6354 | trid += epath.substr(strlen(SPDK_PREFIX)); | |
6355 | r = ::write(fd, trid.c_str(), trid.size()); | |
6356 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
6357 | dout(1) << __func__ << " created " << name << " symlink to " |
6358 | << epath << dendl; | |
6359 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6360 | } | |
6361 | } | |
6362 | if (size) { | |
6363 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
6364 | if (fd >= 0) { | |
6365 | // block file is present | |
6366 | struct stat st; | |
6367 | int r = ::fstat(fd, &st); | |
6368 | if (r == 0 && | |
6369 | S_ISREG(st.st_mode) && // if it is a regular file | |
6370 | st.st_size == 0) { // and is 0 bytes | |
6371 | r = ::ftruncate(fd, size); | |
6372 | if (r < 0) { | |
6373 | r = -errno; | |
6374 | derr << __func__ << " failed to resize " << name << " file to " | |
6375 | << size << ": " << cpp_strerror(r) << dendl; | |
6376 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6377 | return r; | |
6378 | } | |
6379 | ||
6380 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
6381 | r = ::ceph_posix_fallocate(fd, 0, size); |
6382 | if (r > 0) { | |
7c673cae FG |
6383 | derr << __func__ << " failed to prefallocate " << name << " file to " |
6384 | << size << ": " << cpp_strerror(r) << dendl; | |
6385 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6386 | return -r; | |
6387 | } | |
7c673cae FG |
6388 | } |
6389 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 6390 | << byte_u_t(size) << dendl; |
7c673cae FG |
6391 | } |
6392 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6393 | } else { | |
6394 | int r = -errno; | |
6395 | if (r != -ENOENT) { | |
6396 | derr << __func__ << " failed to open " << name << " file: " | |
6397 | << cpp_strerror(r) << dendl; | |
6398 | return r; | |
6399 | } | |
6400 | } | |
6401 | } | |
6402 | return 0; | |
6403 | } | |
6404 | ||
6405 | int BlueStore::mkfs() | |
6406 | { | |
6407 | dout(1) << __func__ << " path " << path << dendl; | |
6408 | int r; | |
6409 | uuid_d old_fsid; | |
f67539c2 | 6410 | uint64_t reserved; |
eafe8130 TL |
6411 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6412 | derr << __func__ << " osd_max_object_size " | |
6413 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6414 | << OBJECT_MAX_SIZE << dendl; | |
6415 | return -EINVAL; | |
6416 | } | |
6417 | ||
7c673cae FG |
6418 | { |
6419 | string done; | |
6420 | r = read_meta("mkfs_done", &done); | |
6421 | if (r == 0) { | |
6422 | dout(1) << __func__ << " already created" << dendl; | |
6423 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
6424 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6425 | if (r < 0) { | |
6426 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
6427 | << dendl; | |
6428 | return r; | |
6429 | } | |
6430 | if (r > 0) { | |
6431 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
6432 | r = -EIO; | |
6433 | } | |
6434 | } | |
6435 | return r; // idempotent | |
6436 | } | |
6437 | } | |
6438 | ||
6439 | { | |
6440 | string type; | |
6441 | r = read_meta("type", &type); | |
6442 | if (r == 0) { | |
6443 | if (type != "bluestore") { | |
6444 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6445 | return -EIO; | |
6446 | } | |
6447 | } else { | |
6448 | r = write_meta("type", "bluestore"); | |
6449 | if (r < 0) | |
6450 | return r; | |
6451 | } | |
6452 | } | |
6453 | ||
6454 | freelist_type = "bitmap"; | |
6455 | ||
6456 | r = _open_path(); | |
6457 | if (r < 0) | |
6458 | return r; | |
6459 | ||
6460 | r = _open_fsid(true); | |
6461 | if (r < 0) | |
6462 | goto out_path_fd; | |
6463 | ||
6464 | r = _lock_fsid(); | |
6465 | if (r < 0) | |
6466 | goto out_close_fsid; | |
6467 | ||
6468 | r = _read_fsid(&old_fsid); | |
6469 | if (r < 0 || old_fsid.is_zero()) { | |
6470 | if (fsid.is_zero()) { | |
6471 | fsid.generate_random(); | |
6472 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
6473 | } else { | |
6474 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
6475 | } | |
6476 | // we'll write it later. | |
6477 | } else { | |
6478 | if (!fsid.is_zero() && fsid != old_fsid) { | |
6479 | derr << __func__ << " on-disk fsid " << old_fsid | |
6480 | << " != provided " << fsid << dendl; | |
6481 | r = -EINVAL; | |
6482 | goto out_close_fsid; | |
6483 | } | |
6484 | fsid = old_fsid; | |
6485 | } | |
6486 | ||
6487 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
6488 | cct->_conf->bluestore_block_size, | |
6489 | cct->_conf->bluestore_block_create); | |
6490 | if (r < 0) | |
6491 | goto out_close_fsid; | |
6492 | if (cct->_conf->bluestore_bluefs) { | |
6493 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
6494 | cct->_conf->bluestore_block_wal_size, | |
6495 | cct->_conf->bluestore_block_wal_create); | |
6496 | if (r < 0) | |
6497 | goto out_close_fsid; | |
6498 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
6499 | cct->_conf->bluestore_block_db_size, | |
6500 | cct->_conf->bluestore_block_db_create); | |
6501 | if (r < 0) | |
6502 | goto out_close_fsid; | |
6503 | } | |
6504 | ||
6505 | r = _open_bdev(true); | |
6506 | if (r < 0) | |
6507 | goto out_close_fsid; | |
6508 | ||
3efd9988 FG |
6509 | // choose min_alloc_size |
6510 | if (cct->_conf->bluestore_min_alloc_size) { | |
6511 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; | |
6512 | } else { | |
11fdf7f2 | 6513 | ceph_assert(bdev); |
f67539c2 | 6514 | if (_use_rotational_settings()) { |
3efd9988 FG |
6515 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; |
6516 | } else { | |
6517 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
6518 | } | |
6519 | } | |
11fdf7f2 | 6520 | _validate_bdev(); |
3efd9988 FG |
6521 | |
6522 | // make sure min_alloc_size is power of 2 aligned. | |
11fdf7f2 | 6523 | if (!isp2(min_alloc_size)) { |
3efd9988 FG |
6524 | derr << __func__ << " min_alloc_size 0x" |
6525 | << std::hex << min_alloc_size << std::dec | |
6526 | << " is not power of 2 aligned!" | |
6527 | << dendl; | |
6528 | r = -EINVAL; | |
6529 | goto out_close_bdev; | |
6530 | } | |
6531 | ||
f67539c2 TL |
6532 | r = _create_alloc(); |
6533 | if (r < 0) { | |
6534 | goto out_close_bdev; | |
6535 | } | |
6536 | ||
6537 | reserved = _get_ondisk_reserved(); | |
6538 | shared_alloc.a->init_add_free(reserved, | |
6539 | p2align(bdev->get_size(), min_alloc_size) - reserved); | |
6540 | ||
7c673cae FG |
6541 | r = _open_db(true); |
6542 | if (r < 0) | |
f67539c2 | 6543 | goto out_close_alloc; |
7c673cae | 6544 | |
7c673cae FG |
6545 | { |
6546 | KeyValueDB::Transaction t = db->get_transaction(); | |
1911f103 | 6547 | r = _open_fm(t, true); |
11fdf7f2 TL |
6548 | if (r < 0) |
6549 | goto out_close_db; | |
7c673cae FG |
6550 | { |
6551 | bufferlist bl; | |
11fdf7f2 | 6552 | encode((uint64_t)0, bl); |
7c673cae FG |
6553 | t->set(PREFIX_SUPER, "nid_max", bl); |
6554 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
6555 | } | |
6556 | ||
7c673cae FG |
6557 | { |
6558 | bufferlist bl; | |
11fdf7f2 | 6559 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
6560 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
6561 | } | |
9f95a23c TL |
6562 | { |
6563 | bufferlist bl; | |
a4b75251 TL |
6564 | if (cct->_conf.get_val<bool>("bluestore_debug_legacy_omap")) { |
6565 | bl.append(stringify(OMAP_BULK)); | |
6566 | } else { | |
6567 | bl.append(stringify(OMAP_PER_PG)); | |
6568 | } | |
9f95a23c TL |
6569 | t->set(PREFIX_SUPER, "per_pool_omap", bl); |
6570 | } | |
7c673cae FG |
6571 | ondisk_format = latest_ondisk_format; |
6572 | _prepare_ondisk_format_super(t); | |
6573 | db->submit_transaction_sync(t); | |
6574 | } | |
6575 | ||
7c673cae FG |
6576 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
6577 | if (r < 0) | |
224ce89b WB |
6578 | goto out_close_fm; |
6579 | ||
3efd9988 | 6580 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 6581 | if (r < 0) |
224ce89b | 6582 | goto out_close_fm; |
7c673cae FG |
6583 | |
6584 | if (fsid != old_fsid) { | |
6585 | r = _write_fsid(); | |
6586 | if (r < 0) { | |
6587 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 6588 | goto out_close_fm; |
7c673cae FG |
6589 | } |
6590 | } | |
6591 | ||
7c673cae FG |
6592 | out_close_fm: |
6593 | _close_fm(); | |
6594 | out_close_db: | |
1911f103 | 6595 | _close_db(false); |
f67539c2 TL |
6596 | out_close_alloc: |
6597 | _close_alloc(); | |
7c673cae FG |
6598 | out_close_bdev: |
6599 | _close_bdev(); | |
6600 | out_close_fsid: | |
6601 | _close_fsid(); | |
6602 | out_path_fd: | |
6603 | _close_path(); | |
6604 | ||
6605 | if (r == 0 && | |
6606 | cct->_conf->bluestore_fsck_on_mkfs) { | |
6607 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6608 | if (rc < 0) | |
6609 | return rc; | |
6610 | if (rc > 0) { | |
6611 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6612 | r = -EIO; | |
6613 | } | |
11fdf7f2 TL |
6614 | } |
6615 | ||
6616 | if (r == 0) { | |
6617 | // indicate success by writing the 'mkfs_done' file | |
6618 | r = write_meta("mkfs_done", "yes"); | |
6619 | } | |
6620 | ||
6621 | if (r < 0) { | |
6622 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6623 | } else { | |
6624 | dout(0) << __func__ << " success" << dendl; | |
6625 | } | |
6626 | return r; | |
6627 | } | |
6628 | ||
11fdf7f2 TL |
6629 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) |
6630 | { | |
6631 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6632 | int r; | |
6633 | ceph_assert(path_fd < 0); | |
6634 | ||
6635 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6636 | ||
6637 | if (!cct->_conf->bluestore_bluefs) { | |
6638 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6639 | return -EIO; | |
6640 | } | |
6641 | ||
f67539c2 | 6642 | r = _open_db_and_around(true); |
11fdf7f2 | 6643 | |
11fdf7f2 TL |
6644 | if (id == BlueFS::BDEV_NEWWAL) { |
6645 | string p = path + "/block.wal"; | |
6646 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
6647 | cct->_conf->bluestore_block_wal_size, | |
6648 | true); | |
6649 | ceph_assert(r == 0); | |
6650 | ||
6651 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
f67539c2 TL |
6652 | cct->_conf->bdev_enable_discard, |
6653 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
6654 | ceph_assert(r == 0); |
6655 | ||
6656 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6657 | r = _check_or_set_bdev_label( | |
6658 | p, | |
6659 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6660 | "bluefs wal", | |
6661 | true); | |
6662 | ceph_assert(r == 0); | |
6663 | } | |
6664 | ||
9f95a23c | 6665 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6666 | } else if (id == BlueFS::BDEV_NEWDB) { |
6667 | string p = path + "/block.db"; | |
6668 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
6669 | cct->_conf->bluestore_block_db_size, | |
6670 | true); | |
6671 | ceph_assert(r == 0); | |
6672 | ||
6673 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
f67539c2 TL |
6674 | cct->_conf->bdev_enable_discard, |
6675 | SUPER_RESERVED); | |
11fdf7f2 TL |
6676 | ceph_assert(r == 0); |
6677 | ||
6678 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6679 | r = _check_or_set_bdev_label( | |
6680 | p, | |
6681 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6682 | "bluefs db", | |
6683 | true); | |
6684 | ceph_assert(r == 0); | |
6685 | } | |
9f95a23c TL |
6686 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6687 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
6688 | } |
6689 | ||
6690 | bluefs->umount(); | |
6691 | bluefs->mount(); | |
6692 | ||
9f95a23c | 6693 | r = bluefs->prepare_new_device(id, bluefs_layout); |
11fdf7f2 TL |
6694 | ceph_assert(r == 0); |
6695 | ||
6696 | if (r < 0) { | |
6697 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6698 | } else { | |
6699 | dout(0) << __func__ << " success" << dendl; | |
6700 | } | |
6701 | ||
f67539c2 | 6702 | _close_db_and_around(true); |
11fdf7f2 TL |
6703 | return r; |
6704 | } | |
6705 | ||
6706 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
6707 | int id) | |
6708 | { | |
6709 | dout(10) << __func__ << " id:" << id << dendl; | |
6710 | ceph_assert(path_fd < 0); | |
6711 | ||
6712 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
6713 | ||
6714 | if (!cct->_conf->bluestore_bluefs) { | |
6715 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6716 | return -EIO; | |
6717 | } | |
6718 | ||
f67539c2 | 6719 | int r = _open_db_and_around(true); |
11fdf7f2 | 6720 | |
f67539c2 | 6721 | uint64_t used_space = 0; |
11fdf7f2 | 6722 | for(auto src_id : devs_source) { |
f67539c2 | 6723 | used_space += bluefs->get_used(src_id); |
11fdf7f2 TL |
6724 | } |
6725 | uint64_t target_free = bluefs->get_free(id); | |
f67539c2 | 6726 | if (target_free < used_space) { |
11fdf7f2 TL |
6727 | derr << __func__ |
6728 | << " can't migrate, free space at target: " << target_free | |
6729 | << " is less than required space: " << used_space | |
6730 | << dendl; | |
f67539c2 TL |
6731 | r = -ENOSPC; |
6732 | goto shutdown; | |
11fdf7f2 | 6733 | } |
9f95a23c TL |
6734 | if (devs_source.count(BlueFS::BDEV_DB)) { |
6735 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; | |
6736 | bluefs_layout.dedicated_db = false; | |
6737 | } | |
6738 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6739 | bluefs_layout.dedicated_wal = false; | |
6740 | } | |
6741 | r = bluefs->device_migrate_to_existing(cct, devs_source, id, bluefs_layout); | |
11fdf7f2 TL |
6742 | if (r < 0) { |
6743 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
6744 | goto shutdown; | |
6745 | } | |
6746 | ||
6747 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
6748 | r = unlink(string(path + "/block.db").c_str()); | |
6749 | ceph_assert(r == 0); | |
6750 | } | |
6751 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6752 | r = unlink(string(path + "/block.wal").c_str()); | |
6753 | ceph_assert(r == 0); | |
6754 | } | |
6755 | ||
6756 | shutdown: | |
f67539c2 | 6757 | _close_db_and_around(true); |
11fdf7f2 TL |
6758 | return r; |
6759 | } | |
6760 | ||
6761 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
6762 | int id, | |
6763 | const string& dev_path) | |
6764 | { | |
6765 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6766 | int r; | |
6767 | ceph_assert(path_fd < 0); | |
6768 | ||
6769 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6770 | ||
6771 | if (!cct->_conf->bluestore_bluefs) { | |
6772 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6773 | return -EIO; | |
6774 | } | |
6775 | ||
f67539c2 | 6776 | r = _open_db_and_around(true); |
11fdf7f2 | 6777 | |
11fdf7f2 TL |
6778 | string link_db; |
6779 | string link_wal; | |
6780 | if (devs_source.count(BlueFS::BDEV_DB) && | |
9f95a23c | 6781 | bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 6782 | link_db = path + "/block.db"; |
9f95a23c TL |
6783 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
6784 | bluefs_layout.dedicated_db = false; | |
11fdf7f2 TL |
6785 | } |
6786 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6787 | link_wal = path + "/block.wal"; | |
9f95a23c | 6788 | bluefs_layout.dedicated_wal = false; |
11fdf7f2 TL |
6789 | } |
6790 | ||
6791 | size_t target_size; | |
6792 | string target_name; | |
6793 | if (id == BlueFS::BDEV_NEWWAL) { | |
6794 | target_name = "block.wal"; | |
6795 | target_size = cct->_conf->bluestore_block_wal_size; | |
9f95a23c | 6796 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6797 | |
6798 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
f67539c2 TL |
6799 | cct->_conf->bdev_enable_discard, |
6800 | BDEV_LABEL_BLOCK_SIZE); | |
11fdf7f2 TL |
6801 | ceph_assert(r == 0); |
6802 | ||
6803 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6804 | r = _check_or_set_bdev_label( | |
6805 | dev_path, | |
6806 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6807 | "bluefs wal", | |
6808 | true); | |
6809 | ceph_assert(r == 0); | |
6810 | } | |
11fdf7f2 TL |
6811 | } else if (id == BlueFS::BDEV_NEWDB) { |
6812 | target_name = "block.db"; | |
6813 | target_size = cct->_conf->bluestore_block_db_size; | |
9f95a23c TL |
6814 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6815 | bluefs_layout.dedicated_db = true; | |
31f18b77 | 6816 | |
11fdf7f2 | 6817 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
f67539c2 TL |
6818 | cct->_conf->bdev_enable_discard, |
6819 | SUPER_RESERVED); | |
11fdf7f2 TL |
6820 | ceph_assert(r == 0); |
6821 | ||
6822 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6823 | r = _check_or_set_bdev_label( | |
6824 | dev_path, | |
6825 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6826 | "bluefs db", | |
6827 | true); | |
6828 | ceph_assert(r == 0); | |
6829 | } | |
31f18b77 FG |
6830 | } |
6831 | ||
11fdf7f2 TL |
6832 | bluefs->umount(); |
6833 | bluefs->mount(); | |
6834 | ||
9f95a23c | 6835 | r = bluefs->device_migrate_to_new(cct, devs_source, id, bluefs_layout); |
11fdf7f2 | 6836 | |
7c673cae | 6837 | if (r < 0) { |
11fdf7f2 TL |
6838 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
6839 | goto shutdown; | |
6840 | } | |
6841 | ||
6842 | if (!link_db.empty()) { | |
6843 | r = unlink(link_db.c_str()); | |
6844 | ceph_assert(r == 0); | |
6845 | } | |
6846 | if (!link_wal.empty()) { | |
6847 | r = unlink(link_wal.c_str()); | |
6848 | ceph_assert(r == 0); | |
6849 | } | |
6850 | r = _setup_block_symlink_or_file( | |
6851 | target_name, | |
6852 | dev_path, | |
6853 | target_size, | |
6854 | true); | |
6855 | ceph_assert(r == 0); | |
6856 | dout(0) << __func__ << " success" << dendl; | |
6857 | ||
6858 | shutdown: | |
f67539c2 TL |
6859 | _close_db_and_around(true); |
6860 | ||
11fdf7f2 TL |
6861 | return r; |
6862 | } | |
6863 | ||
6864 | string BlueStore::get_device_path(unsigned id) | |
6865 | { | |
6866 | string res; | |
6867 | if (id < BlueFS::MAX_BDEV) { | |
6868 | switch (id) { | |
6869 | case BlueFS::BDEV_WAL: | |
6870 | res = path + "/block.wal"; | |
6871 | break; | |
6872 | case BlueFS::BDEV_DB: | |
9f95a23c | 6873 | if (id == bluefs_layout.shared_bdev) { |
11fdf7f2 TL |
6874 | res = path + "/block"; |
6875 | } else { | |
6876 | res = path + "/block.db"; | |
6877 | } | |
6878 | break; | |
6879 | case BlueFS::BDEV_SLOW: | |
6880 | res = path + "/block"; | |
6881 | break; | |
6882 | } | |
6883 | } | |
6884 | return res; | |
6885 | } | |
6886 | ||
f67539c2 TL |
6887 | int BlueStore::_set_bdev_label_size(const string& path, uint64_t size) |
6888 | { | |
6889 | bluestore_bdev_label_t label; | |
6890 | int r = _read_bdev_label(cct, path, &label); | |
6891 | if (r < 0) { | |
6892 | derr << "unable to read label for " << path << ": " | |
6893 | << cpp_strerror(r) << dendl; | |
6894 | } else { | |
6895 | label.size = size; | |
6896 | r = _write_bdev_label(cct, path, label); | |
6897 | if (r < 0) { | |
6898 | derr << "unable to write label for " << path << ": " | |
6899 | << cpp_strerror(r) << dendl; | |
6900 | } | |
6901 | } | |
6902 | return r; | |
6903 | } | |
6904 | ||
11fdf7f2 TL |
6905 | int BlueStore::expand_devices(ostream& out) |
6906 | { | |
f67539c2 | 6907 | int r = _open_db_and_around(true); |
11fdf7f2 TL |
6908 | ceph_assert(r == 0); |
6909 | bluefs->dump_block_extents(out); | |
1911f103 | 6910 | out << "Expanding DB/WAL..." << std::endl; |
11fdf7f2 | 6911 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { |
9f95a23c | 6912 | if (devid == bluefs_layout.shared_bdev ) { |
11fdf7f2 TL |
6913 | continue; |
6914 | } | |
6915 | uint64_t size = bluefs->get_block_device_size(devid); | |
6916 | if (size == 0) { | |
6917 | // no bdev | |
6918 | continue; | |
6919 | } | |
6920 | ||
f67539c2 TL |
6921 | out << devid |
6922 | <<" : expanding " << " to 0x" << size << std::dec << std::endl; | |
6923 | string p = get_device_path(devid); | |
6924 | const char* path = p.c_str(); | |
6925 | if (path == nullptr) { | |
6926 | derr << devid | |
6927 | <<": can't find device path " << dendl; | |
6928 | continue; | |
6929 | } | |
6930 | if (bluefs->bdev_support_label(devid)) { | |
6931 | if (_set_bdev_label_size(p, size) >= 0) { | |
6932 | out << devid | |
6933 | << " : size label updated to " << size | |
6934 | << std::endl; | |
11fdf7f2 | 6935 | } |
11fdf7f2 TL |
6936 | } |
6937 | } | |
6938 | uint64_t size0 = fm->get_size(); | |
6939 | uint64_t size = bdev->get_size(); | |
6940 | if (size0 < size) { | |
9f95a23c | 6941 | out << bluefs_layout.shared_bdev |
1911f103 TL |
6942 | << " : expanding " << " from 0x" << std::hex |
6943 | << size0 << " to 0x" << size << std::dec << std::endl; | |
f67539c2 TL |
6944 | _write_out_fm_meta(size); |
6945 | if (bdev->supported_bdev_label()) { | |
6946 | if (_set_bdev_label_size(path, size) >= 0) { | |
6947 | out << bluefs_layout.shared_bdev | |
6948 | << " : size label updated to " << size | |
6949 | << std::endl; | |
6950 | } | |
6951 | } | |
6952 | _close_db_and_around(true); | |
1911f103 TL |
6953 | |
6954 | // mount in read/write to sync expansion changes | |
f67539c2 | 6955 | r = _mount(); |
11fdf7f2 | 6956 | ceph_assert(r == 0); |
1911f103 TL |
6957 | umount(); |
6958 | } else { | |
f67539c2 | 6959 | _close_db_and_around(true); |
7c673cae | 6960 | } |
1911f103 TL |
6961 | return r; |
6962 | } | |
6963 | ||
6964 | int BlueStore::dump_bluefs_sizes(ostream& out) | |
6965 | { | |
f67539c2 | 6966 | int r = _open_db_and_around(true); |
1911f103 TL |
6967 | ceph_assert(r == 0); |
6968 | bluefs->dump_block_extents(out); | |
f67539c2 | 6969 | _close_db_and_around(true); |
7c673cae FG |
6970 | return r; |
6971 | } | |
6972 | ||
6973 | void BlueStore::set_cache_shards(unsigned num) | |
6974 | { | |
6975 | dout(10) << __func__ << " " << num << dendl; | |
9f95a23c TL |
6976 | size_t oold = onode_cache_shards.size(); |
6977 | size_t bold = buffer_cache_shards.size(); | |
6978 | ceph_assert(num >= oold && num >= bold); | |
6979 | onode_cache_shards.resize(num); | |
6980 | buffer_cache_shards.resize(num); | |
6981 | for (unsigned i = oold; i < num; ++i) { | |
6982 | onode_cache_shards[i] = | |
6983 | OnodeCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6984 | logger); | |
6985 | } | |
6986 | for (unsigned i = bold; i < num; ++i) { | |
6987 | buffer_cache_shards[i] = | |
6988 | BufferCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6989 | logger); | |
7c673cae FG |
6990 | } |
6991 | } | |
6992 | ||
f67539c2 | 6993 | int BlueStore::_mount() |
7c673cae FG |
6994 | { |
6995 | dout(1) << __func__ << " path " << path << dendl; | |
6996 | ||
f67539c2 | 6997 | _kv_only = false; |
7c673cae FG |
6998 | if (cct->_conf->bluestore_fsck_on_mount) { |
6999 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); | |
7000 | if (rc < 0) | |
7001 | return rc; | |
7002 | if (rc > 0) { | |
7003 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7004 | return -EIO; | |
7005 | } | |
7006 | } | |
7007 | ||
eafe8130 TL |
7008 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
7009 | derr << __func__ << " osd_max_object_size " | |
7010 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
7011 | << OBJECT_MAX_SIZE << dendl; | |
7012 | return -EINVAL; | |
7013 | } | |
7014 | ||
f67539c2 | 7015 | int r = _open_db_and_around(false); |
9f95a23c | 7016 | if (r < 0) { |
f67539c2 | 7017 | return r; |
11fdf7f2 | 7018 | } |
7c673cae | 7019 | |
11fdf7f2 TL |
7020 | r = _upgrade_super(); |
7021 | if (r < 0) { | |
7c673cae | 7022 | goto out_db; |
11fdf7f2 | 7023 | } |
7c673cae FG |
7024 | |
7025 | r = _open_collections(); | |
7026 | if (r < 0) | |
11fdf7f2 | 7027 | goto out_db; |
7c673cae FG |
7028 | |
7029 | r = _reload_logger(); | |
7030 | if (r < 0) | |
7031 | goto out_coll; | |
7032 | ||
31f18b77 | 7033 | _kv_start(); |
7c673cae | 7034 | |
f67539c2 TL |
7035 | if (bdev->is_smr()) { |
7036 | _zoned_cleaner_start(); | |
7037 | } | |
7038 | ||
7c673cae FG |
7039 | r = _deferred_replay(); |
7040 | if (r < 0) | |
7041 | goto out_stop; | |
7042 | ||
7043 | mempool_thread.init(); | |
7044 | ||
f67539c2 | 7045 | if ((!per_pool_stat_collection || per_pool_omap != OMAP_PER_PG) && |
eafe8130 | 7046 | cct->_conf->bluestore_fsck_quick_fix_on_mount == true) { |
9f95a23c | 7047 | |
f67539c2 | 7048 | auto was_per_pool_omap = per_pool_omap; |
9f95a23c | 7049 | |
eafe8130 TL |
7050 | dout(1) << __func__ << " quick-fix on mount" << dendl; |
7051 | _fsck_on_open(FSCK_SHALLOW, true); | |
7052 | ||
7053 | //reread statfs | |
7054 | //FIXME minor: replace with actual open/close? | |
7055 | _open_statfs(); | |
eafe8130 | 7056 | _check_legacy_statfs_alert(); |
9f95a23c TL |
7057 | |
7058 | //set again as hopefully it has been fixed | |
f67539c2 | 7059 | if (was_per_pool_omap != OMAP_PER_PG) { |
9f95a23c TL |
7060 | _set_per_pool_omap(); |
7061 | } | |
eafe8130 TL |
7062 | } |
7063 | ||
7c673cae FG |
7064 | mounted = true; |
7065 | return 0; | |
7066 | ||
7067 | out_stop: | |
f67539c2 TL |
7068 | if (bdev->is_smr()) { |
7069 | _zoned_cleaner_stop(); | |
7070 | } | |
7c673cae | 7071 | _kv_stop(); |
7c673cae | 7072 | out_coll: |
f6b5b4d7 | 7073 | _shutdown_cache(); |
7c673cae | 7074 | out_db: |
1911f103 | 7075 | _close_db_and_around(false); |
7c673cae FG |
7076 | return r; |
7077 | } | |
7078 | ||
7079 | int BlueStore::umount() | |
7080 | { | |
11fdf7f2 | 7081 | ceph_assert(_kv_only || mounted); |
7c673cae FG |
7082 | dout(1) << __func__ << dendl; |
7083 | ||
7084 | _osr_drain_all(); | |
7c673cae | 7085 | |
7c673cae | 7086 | mounted = false; |
3efd9988 FG |
7087 | if (!_kv_only) { |
7088 | mempool_thread.shutdown(); | |
f67539c2 TL |
7089 | if (bdev->is_smr()) { |
7090 | dout(20) << __func__ << " stopping zone cleaner thread" << dendl; | |
7091 | _zoned_cleaner_stop(); | |
7092 | } | |
3efd9988 FG |
7093 | dout(20) << __func__ << " stopping kv thread" << dendl; |
7094 | _kv_stop(); | |
f6b5b4d7 | 7095 | _shutdown_cache(); |
3efd9988 FG |
7096 | dout(20) << __func__ << " closing" << dendl; |
7097 | ||
3efd9988 | 7098 | } |
1911f103 | 7099 | _close_db_and_around(false); |
7c673cae FG |
7100 | |
7101 | if (cct->_conf->bluestore_fsck_on_umount) { | |
7102 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); | |
7103 | if (rc < 0) | |
7104 | return rc; | |
7105 | if (rc > 0) { | |
7106 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7107 | return -EIO; | |
7108 | } | |
7109 | } | |
7110 | return 0; | |
7111 | } | |
7112 | ||
eafe8130 TL |
7113 | int BlueStore::cold_open() |
7114 | { | |
f67539c2 | 7115 | return _open_db_and_around(true); |
eafe8130 | 7116 | } |
f67539c2 | 7117 | |
eafe8130 TL |
7118 | int BlueStore::cold_close() |
7119 | { | |
1911f103 | 7120 | _close_db_and_around(true); |
eafe8130 TL |
7121 | return 0; |
7122 | } | |
7123 | ||
9f95a23c TL |
7124 | // derr wrapper to limit enormous output and avoid log flooding. |
7125 | // Of limited use where such output is expected for now | |
7126 | #define fsck_derr(err_cnt, threshold) \ | |
7127 | if (err_cnt <= threshold) { \ | |
7128 | bool need_skip_print = err_cnt == threshold; \ | |
7129 | derr | |
7130 | ||
7131 | #define fsck_dendl \ | |
7132 | dendl; \ | |
7133 | if (need_skip_print) \ | |
7134 | derr << "more error lines skipped..." << dendl; \ | |
7c673cae | 7135 | } |
7c673cae | 7136 | |
eafe8130 TL |
7137 | int _fsck_sum_extents( |
7138 | const PExtentVector& extents, | |
7139 | bool compressed, | |
7140 | store_statfs_t& expected_statfs) | |
7141 | { | |
7142 | for (auto e : extents) { | |
7143 | if (!e.is_valid()) | |
7144 | continue; | |
7145 | expected_statfs.allocated += e.length; | |
7146 | if (compressed) { | |
7147 | expected_statfs.data_compressed_allocated += e.length; | |
7148 | } | |
7149 | } | |
7150 | return 0; | |
7151 | } | |
7152 | ||
7c673cae | 7153 | int BlueStore::_fsck_check_extents( |
11fdf7f2 | 7154 | const coll_t& cid, |
7c673cae FG |
7155 | const ghobject_t& oid, |
7156 | const PExtentVector& extents, | |
7157 | bool compressed, | |
7158 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 7159 | uint64_t granularity, |
11fdf7f2 | 7160 | BlueStoreRepairer* repairer, |
eafe8130 TL |
7161 | store_statfs_t& expected_statfs, |
7162 | FSCKDepth depth) | |
7c673cae FG |
7163 | { |
7164 | dout(30) << __func__ << " oid " << oid << " extents " << extents << dendl; | |
7165 | int errors = 0; | |
7166 | for (auto e : extents) { | |
7167 | if (!e.is_valid()) | |
7168 | continue; | |
7169 | expected_statfs.allocated += e.length; | |
7170 | if (compressed) { | |
11fdf7f2 | 7171 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae | 7172 | } |
eafe8130 TL |
7173 | if (depth != FSCK_SHALLOW) { |
7174 | bool already = false; | |
9f95a23c | 7175 | apply_for_bitset_range( |
eafe8130 TL |
7176 | e.offset, e.length, granularity, used_blocks, |
7177 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
7178 | if (bs.test(pos)) { |
7179 | if (repairer) { | |
7180 | repairer->note_misreference( | |
7181 | pos * min_alloc_size, min_alloc_size, !already); | |
7182 | } | |
7183 | if (!already) { | |
7184 | derr << "fsck error: " << oid << " extent " << e | |
7185 | << " or a subset is already allocated (misreferenced)" << dendl; | |
7186 | ++errors; | |
7187 | already = true; | |
7188 | } | |
11fdf7f2 | 7189 | } |
eafe8130 TL |
7190 | else |
7191 | bs.set(pos); | |
7192 | }); | |
7193 | if (repairer) { | |
b3b6e05e | 7194 | repairer->set_space_used(e.offset, e.length, cid, oid); |
eafe8130 | 7195 | } |
11fdf7f2 | 7196 | |
eafe8130 TL |
7197 | if (e.end() > bdev->get_size()) { |
7198 | derr << "fsck error: " << oid << " extent " << e | |
7199 | << " past end of block device" << dendl; | |
7200 | ++errors; | |
7201 | } | |
7c673cae FG |
7202 | } |
7203 | } | |
7204 | return errors; | |
7205 | } | |
7206 | ||
11fdf7f2 TL |
7207 | void BlueStore::_fsck_check_pool_statfs( |
7208 | BlueStore::per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
7209 | int64_t& errors, |
7210 | int64_t& warnings, | |
11fdf7f2 TL |
7211 | BlueStoreRepairer* repairer) |
7212 | { | |
f67539c2 | 7213 | auto it = db->get_iterator(PREFIX_STAT, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 TL |
7214 | if (it) { |
7215 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7216 | string key = it->key(); | |
7217 | if (key == BLUESTORE_GLOBAL_STATFS_KEY) { | |
7218 | if (repairer) { | |
eafe8130 TL |
7219 | ++errors; |
7220 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
7221 | derr << "fsck error: " << "legacy statfs record found, removing" | |
11fdf7f2 TL |
7222 | << dendl; |
7223 | } | |
7224 | continue; | |
7225 | } | |
11fdf7f2 TL |
7226 | uint64_t pool_id; |
7227 | if (get_key_pool_stat(key, &pool_id) < 0) { | |
7228 | derr << "fsck error: bad key " << key | |
7229 | << "in statfs namespece" << dendl; | |
7230 | if (repairer) { | |
7231 | repairer->remove_key(db, PREFIX_STAT, key); | |
7232 | } | |
7233 | ++errors; | |
7234 | continue; | |
7235 | } | |
7236 | ||
7237 | volatile_statfs vstatfs; | |
7238 | bufferlist bl = it->value(); | |
7239 | auto blp = bl.cbegin(); | |
7240 | try { | |
7241 | vstatfs.decode(blp); | |
f67539c2 | 7242 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
7243 | derr << "fsck error: failed to decode Pool StatFS record" |
7244 | << pretty_binary_string(key) << dendl; | |
7245 | if (repairer) { | |
7246 | dout(20) << __func__ << " undecodable Pool StatFS record, key:'" | |
7247 | << pretty_binary_string(key) | |
7248 | << "', removing" << dendl; | |
7249 | repairer->remove_key(db, PREFIX_STAT, key); | |
7250 | } | |
7251 | ++errors; | |
7252 | vstatfs.reset(); | |
7253 | } | |
7254 | auto stat_it = expected_pool_statfs.find(pool_id); | |
7255 | if (stat_it == expected_pool_statfs.end()) { | |
7256 | if (vstatfs.is_empty()) { | |
7257 | // we don't consider that as an error since empty pool statfs | |
7258 | // are left in DB for now | |
7259 | dout(20) << "fsck inf: found empty stray Pool StatFS record for pool id 0x" | |
7260 | << std::hex << pool_id << std::dec << dendl; | |
7261 | if (repairer) { | |
7262 | // but we need to increment error count in case of repair | |
7263 | // to have proper counters at the end | |
7264 | // (as repairer increments recovery counter anyway). | |
7265 | ++errors; | |
7266 | } | |
7267 | } else { | |
7268 | derr << "fsck error: found stray Pool StatFS record for pool id 0x" | |
7269 | << std::hex << pool_id << std::dec << dendl; | |
7270 | ++errors; | |
7271 | } | |
7272 | if (repairer) { | |
522d829b | 7273 | repairer->remove_key(db, PREFIX_STAT, key); |
11fdf7f2 TL |
7274 | } |
7275 | continue; | |
7276 | } | |
7277 | store_statfs_t statfs; | |
7278 | vstatfs.publish(&statfs); | |
7279 | if (!(stat_it->second == statfs)) { | |
7280 | derr << "fsck error: actual " << statfs | |
7281 | << " != expected " << stat_it->second | |
7282 | << " for pool " | |
7283 | << std::hex << pool_id << std::dec << dendl; | |
7284 | if (repairer) { | |
7285 | repairer->fix_statfs(db, key, stat_it->second); | |
7286 | } | |
7287 | ++errors; | |
7288 | } | |
7289 | expected_pool_statfs.erase(stat_it); | |
7290 | } | |
7291 | } // if (it) | |
eafe8130 TL |
7292 | for (auto& s : expected_pool_statfs) { |
7293 | if (s.second.is_zero()) { | |
11fdf7f2 TL |
7294 | // we might lack empty statfs recs in DB |
7295 | continue; | |
7296 | } | |
7297 | derr << "fsck error: missing Pool StatFS record for pool " | |
eafe8130 | 7298 | << std::hex << s.first << std::dec << dendl; |
11fdf7f2 TL |
7299 | if (repairer) { |
7300 | string key; | |
eafe8130 TL |
7301 | get_pool_stat_key(s.first, &key); |
7302 | repairer->fix_statfs(db, key, s.second); | |
11fdf7f2 TL |
7303 | } |
7304 | ++errors; | |
7305 | } | |
eafe8130 | 7306 | if (!per_pool_stat_collection && |
eafe8130 TL |
7307 | repairer) { |
7308 | // by virtue of running this method, we correct the top-level | |
7309 | // error of having global stats | |
7310 | repairer->inc_repaired(); | |
7311 | } | |
11fdf7f2 TL |
7312 | } |
7313 | ||
eafe8130 TL |
7314 | BlueStore::OnodeRef BlueStore::fsck_check_objects_shallow( |
7315 | BlueStore::FSCKDepth depth, | |
7316 | int64_t pool_id, | |
7317 | BlueStore::CollectionRef c, | |
7318 | const ghobject_t& oid, | |
7319 | const string& key, | |
7320 | const bufferlist& value, | |
9f95a23c | 7321 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
7322 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
7323 | const BlueStore::FSCK_ObjectCtx& ctx) | |
7324 | { | |
7325 | auto& errors = ctx.errors; | |
7326 | auto& num_objects = ctx.num_objects; | |
7327 | auto& num_extents = ctx.num_extents; | |
7328 | auto& num_blobs = ctx.num_blobs; | |
7329 | auto& num_sharded_objects = ctx.num_sharded_objects; | |
7330 | auto& num_spanning_blobs = ctx.num_spanning_blobs; | |
7331 | auto used_blocks = ctx.used_blocks; | |
7332 | auto sb_info_lock = ctx.sb_info_lock; | |
7333 | auto& sb_info = ctx.sb_info; | |
7334 | auto repairer = ctx.repairer; | |
7335 | ||
7336 | store_statfs_t* res_statfs = (per_pool_stat_collection || repairer) ? | |
7337 | &ctx.expected_pool_statfs[pool_id] : | |
7338 | &ctx.expected_store_statfs; | |
7339 | ||
7340 | dout(10) << __func__ << " " << oid << dendl; | |
7341 | OnodeRef o; | |
7342 | o.reset(Onode::decode(c, oid, key, value)); | |
7343 | ++num_objects; | |
7c673cae | 7344 | |
eafe8130 | 7345 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); |
7c673cae | 7346 | |
eafe8130 TL |
7347 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
7348 | _dump_onode<30>(cct, *o); | |
7349 | // shards | |
7350 | if (!o->extent_map.shards.empty()) { | |
7351 | ++num_sharded_objects; | |
7352 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 7353 | ceph_assert(expecting_shards); |
eafe8130 TL |
7354 | for (auto& s : o->extent_map.shards) { |
7355 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
9f95a23c | 7356 | expecting_shards->push_back(string()); |
eafe8130 | 7357 | get_extent_shard_key(o->key, s.shard_info->offset, |
9f95a23c | 7358 | &expecting_shards->back()); |
eafe8130 TL |
7359 | if (s.shard_info->offset >= o->onode.size) { |
7360 | derr << "fsck error: " << oid << " shard 0x" << std::hex | |
7361 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size | |
7362 | << std::dec << dendl; | |
7363 | ++errors; | |
7364 | } | |
7365 | } | |
7366 | } | |
7367 | } | |
7c673cae | 7368 | |
eafe8130 TL |
7369 | // lextents |
7370 | uint64_t pos = 0; | |
7371 | mempool::bluestore_fsck::map<BlobRef, | |
7372 | bluestore_blob_use_tracker_t> ref_map; | |
7373 | for (auto& l : o->extent_map.extent_map) { | |
7374 | dout(20) << __func__ << " " << l << dendl; | |
7375 | if (l.logical_offset < pos) { | |
7376 | derr << "fsck error: " << oid << " lextent at 0x" | |
7377 | << std::hex << l.logical_offset | |
7378 | << " overlaps with the previous, which ends at 0x" << pos | |
7379 | << std::dec << dendl; | |
7380 | ++errors; | |
7381 | } | |
7382 | if (depth != FSCK_SHALLOW && | |
7383 | o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
7384 | derr << "fsck error: " << oid << " lextent at 0x" | |
7385 | << std::hex << l.logical_offset << "~" << l.length | |
7386 | << " spans a shard boundary" | |
7387 | << std::dec << dendl; | |
7388 | ++errors; | |
7389 | } | |
7390 | pos = l.logical_offset + l.length; | |
7391 | res_statfs->data_stored += l.length; | |
7392 | ceph_assert(l.blob); | |
7393 | const bluestore_blob_t& blob = l.blob->get_blob(); | |
7394 | ||
7395 | auto& ref = ref_map[l.blob]; | |
7396 | if (ref.is_empty()) { | |
7397 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
7398 | uint32_t l = blob.get_logical_length(); | |
7399 | ref.init(l, min_release_size); | |
7400 | } | |
7401 | ref.get( | |
7402 | l.blob_offset, | |
7403 | l.length); | |
7404 | ++num_extents; | |
7405 | if (depth != FSCK_SHALLOW && | |
7406 | blob.has_unused()) { | |
7407 | ceph_assert(referenced); | |
7408 | auto p = referenced->find(l.blob); | |
7409 | bluestore_blob_t::unused_t* pu; | |
7410 | if (p == referenced->end()) { | |
7411 | pu = &(*referenced)[l.blob]; | |
7412 | } | |
7413 | else { | |
7414 | pu = &p->second; | |
7415 | } | |
7416 | uint64_t blob_len = blob.get_logical_length(); | |
7417 | ceph_assert((blob_len % (sizeof(*pu) * 8)) == 0); | |
7418 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
7419 | uint64_t chunk_size = blob_len / (sizeof(*pu) * 8); | |
7420 | uint64_t start = l.blob_offset / chunk_size; | |
7421 | uint64_t end = | |
7422 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; | |
7423 | for (auto i = start; i < end; ++i) { | |
7424 | (*pu) |= (1u << i); | |
7425 | } | |
7426 | } | |
7427 | } //for (auto& l : o->extent_map.extent_map) | |
7428 | ||
7429 | for (auto& i : ref_map) { | |
7430 | ++num_blobs; | |
7431 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7432 | bool equal = | |
7433 | depth == FSCK_SHALLOW ? true : | |
7434 | i.first->get_blob_use_tracker().equal(i.second); | |
7435 | if (!equal) { | |
7436 | derr << "fsck error: " << oid << " blob " << *i.first | |
7437 | << " doesn't match expected ref_map " << i.second << dendl; | |
7438 | ++errors; | |
7439 | } | |
7440 | if (blob.is_compressed()) { | |
7441 | res_statfs->data_compressed += blob.get_compressed_payload_length(); | |
7442 | res_statfs->data_compressed_original += | |
7443 | i.first->get_referenced_bytes(); | |
7444 | } | |
7445 | if (blob.is_shared()) { | |
7446 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
7447 | derr << "fsck error: " << oid << " blob " << blob | |
7448 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " | |
7449 | << blobid_max << dendl; | |
7450 | ++errors; | |
7451 | } | |
7452 | else if (i.first->shared_blob->get_sbid() == 0) { | |
7453 | derr << "fsck error: " << oid << " blob " << blob | |
7454 | << " marked as shared but has uninitialized sbid" | |
7455 | << dendl; | |
7456 | ++errors; | |
7457 | } | |
7458 | // the below lock is optional and provided in multithreading mode only | |
7459 | if (sb_info_lock) { | |
7460 | sb_info_lock->lock(); | |
7461 | } | |
7462 | sb_info_t& sbi = sb_info[i.first->shared_blob->get_sbid()]; | |
7463 | ceph_assert(sbi.cid == coll_t() || sbi.cid == c->cid); | |
7464 | ceph_assert(sbi.pool_id == INT64_MIN || | |
7465 | sbi.pool_id == oid.hobj.get_logical_pool()); | |
7466 | sbi.cid = c->cid; | |
7467 | sbi.pool_id = oid.hobj.get_logical_pool(); | |
7468 | sbi.sb = i.first->shared_blob; | |
7469 | sbi.oids.push_back(oid); | |
7470 | sbi.compressed = blob.is_compressed(); | |
7471 | for (auto e : blob.get_extents()) { | |
7472 | if (e.is_valid()) { | |
7473 | sbi.ref_map.get(e.offset, e.length); | |
7474 | } | |
7475 | } | |
7476 | if (sb_info_lock) { | |
7477 | sb_info_lock->unlock(); | |
7478 | } | |
7479 | } else if (depth != FSCK_SHALLOW) { | |
7480 | ceph_assert(used_blocks); | |
7481 | errors += _fsck_check_extents(c->cid, oid, blob.get_extents(), | |
7482 | blob.is_compressed(), | |
7483 | *used_blocks, | |
7484 | fm->get_alloc_size(), | |
7485 | repairer, | |
7486 | *res_statfs, | |
7487 | depth); | |
7488 | } else { | |
7489 | errors += _fsck_sum_extents( | |
7490 | blob.get_extents(), | |
7491 | blob.is_compressed(), | |
7492 | *res_statfs); | |
7493 | } | |
7494 | } // for (auto& i : ref_map) | |
9f95a23c | 7495 | |
adb31ebb TL |
7496 | { |
7497 | auto &sbm = o->extent_map.spanning_blob_map; | |
7498 | size_t broken = 0; | |
7499 | BlobRef first_broken; | |
7500 | for (auto it = sbm.begin(); it != sbm.end();) { | |
7501 | auto it1 = it++; | |
7502 | if (ref_map.count(it1->second) == 0) { | |
7503 | if (!broken) { | |
7504 | first_broken = it1->second; | |
7505 | ++errors; | |
7506 | } | |
7507 | broken++; | |
7508 | if (repairer) { | |
7509 | sbm.erase(it1); | |
7510 | } | |
7511 | } | |
7512 | } | |
7513 | if (broken) { | |
7514 | derr << "fsck error: " << oid << " - " << broken | |
7515 | << " zombie spanning blob(s) found, the first one: " | |
7516 | << *first_broken << dendl; | |
7517 | if(repairer) { | |
b3b6e05e TL |
7518 | repairer->fix_spanning_blobs( |
7519 | db, | |
7520 | [&](KeyValueDB::Transaction txn) { | |
7521 | _record_onode(o, txn); | |
7522 | }); | |
adb31ebb TL |
7523 | } |
7524 | } | |
7525 | } | |
7526 | ||
9f95a23c TL |
7527 | if (o->onode.has_omap()) { |
7528 | _fsck_check_object_omap(depth, o, ctx); | |
7529 | } | |
7530 | ||
eafe8130 TL |
7531 | return o; |
7532 | } | |
7533 | ||
7534 | #include "common/WorkQueue.h" | |
7535 | ||
7536 | class ShallowFSCKThreadPool : public ThreadPool | |
7537 | { | |
7538 | public: | |
7539 | ShallowFSCKThreadPool(CephContext* cct_, std::string nm, std::string tn, int n) : | |
7540 | ThreadPool(cct_, nm, tn, n) { | |
7541 | } | |
7542 | void worker(ThreadPool::WorkThread* wt) override { | |
7543 | int next_wq = 0; | |
7544 | while (!_stop) { | |
7545 | next_wq %= work_queues.size(); | |
7546 | WorkQueue_ *wq = work_queues[next_wq++]; | |
7547 | ||
7548 | void* item = wq->_void_dequeue(); | |
7549 | if (item) { | |
7550 | processing++; | |
7551 | TPHandle tp_handle(cct, nullptr, wq->timeout_interval, wq->suicide_interval); | |
7552 | wq->_void_process(item, tp_handle); | |
7553 | processing--; | |
7554 | } | |
7555 | } | |
7556 | } | |
7557 | template <size_t BatchLen> | |
7558 | struct FSCKWorkQueue : public ThreadPool::WorkQueue_ | |
7559 | { | |
7560 | struct Entry { | |
7561 | int64_t pool_id; | |
7562 | BlueStore::CollectionRef c; | |
7563 | ghobject_t oid; | |
7564 | string key; | |
7565 | bufferlist value; | |
7566 | }; | |
7567 | struct Batch { | |
7568 | std::atomic<size_t> running = { 0 }; | |
7569 | size_t entry_count = 0; | |
7570 | std::array<Entry, BatchLen> entries; | |
7571 | ||
7572 | int64_t errors = 0; | |
7573 | int64_t warnings = 0; | |
7574 | uint64_t num_objects = 0; | |
7575 | uint64_t num_extents = 0; | |
7576 | uint64_t num_blobs = 0; | |
7577 | uint64_t num_sharded_objects = 0; | |
7578 | uint64_t num_spanning_blobs = 0; | |
7579 | store_statfs_t expected_store_statfs; | |
7580 | BlueStore::per_pool_statfs expected_pool_statfs; | |
7581 | }; | |
7582 | ||
7583 | size_t batchCount; | |
7584 | BlueStore* store = nullptr; | |
7585 | ||
eafe8130 TL |
7586 | ceph::mutex* sb_info_lock = nullptr; |
7587 | BlueStore::sb_info_map_t* sb_info = nullptr; | |
7588 | BlueStoreRepairer* repairer = nullptr; | |
7589 | ||
7590 | Batch* batches = nullptr; | |
7591 | size_t last_batch_pos = 0; | |
7592 | bool batch_acquired = false; | |
7593 | ||
7594 | FSCKWorkQueue(std::string n, | |
7595 | size_t _batchCount, | |
7596 | BlueStore* _store, | |
eafe8130 TL |
7597 | ceph::mutex* _sb_info_lock, |
7598 | BlueStore::sb_info_map_t& _sb_info, | |
7599 | BlueStoreRepairer* _repairer) : | |
f67539c2 | 7600 | WorkQueue_(n, ceph::timespan::zero(), ceph::timespan::zero()), |
eafe8130 TL |
7601 | batchCount(_batchCount), |
7602 | store(_store), | |
eafe8130 TL |
7603 | sb_info_lock(_sb_info_lock), |
7604 | sb_info(&_sb_info), | |
7605 | repairer(_repairer) | |
7606 | { | |
7607 | batches = new Batch[batchCount]; | |
7608 | } | |
7609 | ~FSCKWorkQueue() { | |
7610 | delete[] batches; | |
7611 | } | |
7612 | ||
7613 | /// Remove all work items from the queue. | |
7614 | void _clear() override { | |
7615 | //do nothing | |
7616 | } | |
7617 | /// Check whether there is anything to do. | |
7618 | bool _empty() override { | |
7619 | ceph_assert(false); | |
7620 | } | |
7621 | ||
7622 | /// Get the next work item to process. | |
7623 | void* _void_dequeue() override { | |
7624 | size_t pos = rand() % batchCount; | |
7625 | size_t pos0 = pos; | |
7626 | do { | |
7627 | auto& batch = batches[pos]; | |
7628 | if (batch.running.fetch_add(1) == 0) { | |
7629 | if (batch.entry_count) { | |
7630 | return &batch; | |
7631 | } | |
7632 | } | |
7633 | batch.running--; | |
7634 | pos++; | |
7635 | pos %= batchCount; | |
7636 | } while (pos != pos0); | |
7637 | return nullptr; | |
7638 | } | |
7639 | /** @brief Process the work item. | |
7640 | * This function will be called several times in parallel | |
7641 | * and must therefore be thread-safe. */ | |
7642 | void _void_process(void* item, TPHandle& handle) override { | |
7643 | Batch* batch = (Batch*)item; | |
7644 | ||
7645 | BlueStore::FSCK_ObjectCtx ctx( | |
7646 | batch->errors, | |
7647 | batch->warnings, | |
7648 | batch->num_objects, | |
7649 | batch->num_extents, | |
7650 | batch->num_blobs, | |
7651 | batch->num_sharded_objects, | |
7652 | batch->num_spanning_blobs, | |
7653 | nullptr, // used_blocks | |
9f95a23c | 7654 | nullptr, //used_omap_head |
eafe8130 TL |
7655 | sb_info_lock, |
7656 | *sb_info, | |
7657 | batch->expected_store_statfs, | |
7658 | batch->expected_pool_statfs, | |
7659 | repairer); | |
7660 | ||
7661 | for (size_t i = 0; i < batch->entry_count; i++) { | |
7662 | auto& entry = batch->entries[i]; | |
7663 | ||
7664 | store->fsck_check_objects_shallow( | |
7665 | BlueStore::FSCK_SHALLOW, | |
7666 | entry.pool_id, | |
7667 | entry.c, | |
7668 | entry.oid, | |
7669 | entry.key, | |
7670 | entry.value, | |
9f95a23c | 7671 | nullptr, // expecting_shards - this will need a protection if passed |
eafe8130 TL |
7672 | nullptr, // referenced |
7673 | ctx); | |
7674 | } | |
7675 | //std::cout << "processed " << batch << std::endl; | |
7676 | batch->entry_count = 0; | |
7677 | batch->running--; | |
7678 | } | |
7679 | /** @brief Synchronously finish processing a work item. | |
7680 | * This function is called after _void_process with the global thread pool lock held, | |
7681 | * so at most one copy will execute simultaneously for a given thread pool. | |
7682 | * It can be used for non-thread-safe finalization. */ | |
7683 | void _void_process_finish(void*) override { | |
7684 | ceph_assert(false); | |
7685 | } | |
7686 | ||
7687 | bool queue( | |
7688 | int64_t pool_id, | |
7689 | BlueStore::CollectionRef c, | |
7690 | const ghobject_t& oid, | |
7691 | const string& key, | |
7692 | const bufferlist& value) { | |
7693 | bool res = false; | |
7694 | size_t pos0 = last_batch_pos; | |
7695 | if (!batch_acquired) { | |
7696 | do { | |
7697 | auto& batch = batches[last_batch_pos]; | |
7698 | if (batch.running.fetch_add(1) == 0) { | |
7699 | if (batch.entry_count < BatchLen) { | |
7700 | batch_acquired = true; | |
7701 | break; | |
7702 | } | |
7703 | } | |
7704 | batch.running.fetch_sub(1); | |
7705 | last_batch_pos++; | |
7706 | last_batch_pos %= batchCount; | |
7707 | } while (last_batch_pos != pos0); | |
7708 | } | |
7709 | if (batch_acquired) { | |
7710 | auto& batch = batches[last_batch_pos]; | |
7711 | ceph_assert(batch.running); | |
7712 | ceph_assert(batch.entry_count < BatchLen); | |
7713 | ||
7714 | auto& entry = batch.entries[batch.entry_count]; | |
7715 | entry.pool_id = pool_id; | |
7716 | entry.c = c; | |
7717 | entry.oid = oid; | |
7718 | entry.key = key; | |
7719 | entry.value = value; | |
7720 | ||
7721 | ++batch.entry_count; | |
7722 | if (batch.entry_count == BatchLen) { | |
7723 | batch_acquired = false; | |
7724 | batch.running.fetch_sub(1); | |
7725 | last_batch_pos++; | |
7726 | last_batch_pos %= batchCount; | |
7727 | } | |
7728 | res = true; | |
7729 | } | |
7730 | return res; | |
7731 | } | |
7732 | ||
7733 | void finalize(ThreadPool& tp, | |
7734 | BlueStore::FSCK_ObjectCtx& ctx) { | |
7735 | if (batch_acquired) { | |
7736 | auto& batch = batches[last_batch_pos]; | |
7737 | ceph_assert(batch.running); | |
7738 | batch.running.fetch_sub(1); | |
7739 | } | |
7740 | tp.stop(); | |
7741 | ||
7742 | for (size_t i = 0; i < batchCount; i++) { | |
7743 | auto& batch = batches[i]; | |
7744 | ||
7745 | //process leftovers if any | |
7746 | if (batch.entry_count) { | |
7747 | TPHandle tp_handle(store->cct, | |
7748 | nullptr, | |
7749 | timeout_interval, | |
7750 | suicide_interval); | |
7751 | ceph_assert(batch.running == 0); | |
7752 | ||
7753 | batch.running++; // just to be on-par with the regular call | |
7754 | _void_process(&batch, tp_handle); | |
7755 | } | |
7756 | ceph_assert(batch.entry_count == 0); | |
7757 | ||
7758 | ctx.errors += batch.errors; | |
7759 | ctx.warnings += batch.warnings; | |
7760 | ctx.num_objects += batch.num_objects; | |
7761 | ctx.num_extents += batch.num_extents; | |
7762 | ctx.num_blobs += batch.num_blobs; | |
7763 | ctx.num_sharded_objects += batch.num_sharded_objects; | |
7764 | ctx.num_spanning_blobs += batch.num_spanning_blobs; | |
9f95a23c | 7765 | |
eafe8130 TL |
7766 | ctx.expected_store_statfs.add(batch.expected_store_statfs); |
7767 | ||
7768 | for (auto it = batch.expected_pool_statfs.begin(); | |
7769 | it != batch.expected_pool_statfs.end(); | |
7770 | it++) { | |
7771 | ctx.expected_pool_statfs[it->first].add(it->second); | |
7772 | } | |
7773 | } | |
7774 | } | |
7775 | }; | |
7776 | }; | |
7777 | ||
9f95a23c TL |
7778 | void BlueStore::_fsck_check_object_omap(FSCKDepth depth, |
7779 | OnodeRef& o, | |
7780 | const BlueStore::FSCK_ObjectCtx& ctx) | |
eafe8130 | 7781 | { |
9f95a23c TL |
7782 | auto& errors = ctx.errors; |
7783 | auto& warnings = ctx.warnings; | |
7784 | auto repairer = ctx.repairer; | |
7785 | ||
7786 | ceph_assert(o->onode.has_omap()); | |
7787 | if (!o->onode.is_perpool_omap() && !o->onode.is_pgmeta_omap()) { | |
f67539c2 | 7788 | if (per_pool_omap == OMAP_PER_POOL) { |
9f95a23c TL |
7789 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
7790 | << "fsck error: " << o->oid | |
7791 | << " has omap that is not per-pool or pgmeta" | |
7792 | << fsck_dendl; | |
7793 | ++errors; | |
7794 | } else { | |
7795 | const char* w; | |
7796 | int64_t num; | |
7797 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
7798 | ++errors; | |
7799 | num = errors; | |
7800 | w = "error"; | |
7801 | } else { | |
7802 | ++warnings; | |
7803 | num = warnings; | |
7804 | w = "warning"; | |
7805 | } | |
7806 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
7807 | << "fsck " << w << ": " << o->oid | |
7808 | << " has omap that is not per-pool or pgmeta" | |
7809 | << fsck_dendl; | |
7810 | } | |
f67539c2 TL |
7811 | } else if (!o->onode.is_perpg_omap() && !o->onode.is_pgmeta_omap()) { |
7812 | if (per_pool_omap == OMAP_PER_PG) { | |
7813 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
7814 | << "fsck error: " << o->oid | |
7815 | << " has omap that is not per-pg or pgmeta" | |
7816 | << fsck_dendl; | |
7817 | ++errors; | |
7818 | } else { | |
7819 | const char* w; | |
7820 | int64_t num; | |
7821 | if (cct->_conf->bluestore_fsck_error_on_no_per_pg_omap) { | |
7822 | ++errors; | |
7823 | num = errors; | |
7824 | w = "error"; | |
7825 | } else { | |
7826 | ++warnings; | |
7827 | num = warnings; | |
7828 | w = "warning"; | |
7829 | } | |
7830 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
7831 | << "fsck " << w << ": " << o->oid | |
7832 | << " has omap that is not per-pg or pgmeta" | |
7833 | << fsck_dendl; | |
7834 | } | |
9f95a23c TL |
7835 | } |
7836 | if (repairer && | |
f67539c2 | 7837 | !o->onode.is_perpg_omap() && |
9f95a23c | 7838 | !o->onode.is_pgmeta_omap()) { |
f67539c2 | 7839 | dout(10) << "fsck converting " << o->oid << " omap to per-pg" << dendl; |
522d829b | 7840 | bufferlist header; |
9f95a23c | 7841 | map<string, bufferlist> kv; |
522d829b TL |
7842 | { |
7843 | KeyValueDB::Transaction txn = db->get_transaction(); | |
7844 | uint64_t txn_cost = 0; | |
7845 | const string& prefix = Onode::calc_omap_prefix(o->onode.flags); | |
7846 | uint8_t new_flags = o->onode.flags | | |
7847 | bluestore_onode_t::FLAG_PERPOOL_OMAP | | |
7848 | bluestore_onode_t::FLAG_PERPG_OMAP; | |
7849 | const string& new_omap_prefix = Onode::calc_omap_prefix(new_flags); | |
7850 | ||
7851 | KeyValueDB::Iterator it = db->get_iterator(prefix); | |
7852 | string head, tail; | |
7853 | o->get_omap_header(&head); | |
7854 | o->get_omap_tail(&tail); | |
7855 | it->lower_bound(head); | |
7856 | // head | |
7857 | if (it->valid() && it->key() == head) { | |
7858 | dout(30) << __func__ << " got header" << dendl; | |
7859 | header = it->value(); | |
7860 | if (header.length()) { | |
7861 | string new_head; | |
7862 | Onode::calc_omap_header(new_flags, o.get(), &new_head); | |
7863 | txn->set(new_omap_prefix, new_head, header); | |
7864 | txn_cost += new_head.length() + header.length(); | |
7865 | } | |
a4b75251 | 7866 | it->next(); |
522d829b TL |
7867 | } |
7868 | // tail | |
7869 | { | |
7870 | string new_tail; | |
7871 | Onode::calc_omap_tail(new_flags, o.get(), &new_tail); | |
7872 | bufferlist empty; | |
7873 | txn->set(new_omap_prefix, new_tail, empty); | |
7874 | txn_cost += new_tail.length() + new_tail.length(); | |
7875 | } | |
7876 | // values | |
7877 | string final_key; | |
7878 | Onode::calc_omap_key(new_flags, o.get(), string(), &final_key); | |
7879 | size_t base_key_len = final_key.size(); | |
7880 | while (it->valid() && it->key() < tail) { | |
7881 | string user_key; | |
7882 | o->decode_omap_key(it->key(), &user_key); | |
7883 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
7884 | << " -> " << user_key << dendl; | |
7885 | ||
7886 | final_key.resize(base_key_len); | |
a4b75251 | 7887 | final_key += user_key; |
522d829b TL |
7888 | auto v = it->value(); |
7889 | txn->set(new_omap_prefix, final_key, v); | |
7890 | txn_cost += final_key.length() + v.length(); | |
7891 | ||
7892 | // submit a portion if cost exceeds 16MB | |
7893 | if (txn_cost >= 16 * (1 << 20) ) { | |
7894 | db->submit_transaction_sync(txn); | |
7895 | txn = db->get_transaction(); | |
7896 | txn_cost = 0; | |
7897 | } | |
7898 | it->next(); | |
7899 | } | |
7900 | if (txn_cost > 0) { | |
7901 | db->submit_transaction_sync(txn); | |
7902 | } | |
7903 | } | |
7904 | // finalize: remove legacy data | |
7905 | { | |
9f95a23c TL |
7906 | KeyValueDB::Transaction txn = db->get_transaction(); |
7907 | // remove old keys | |
7908 | const string& old_omap_prefix = o->get_omap_prefix(); | |
7909 | string old_head, old_tail; | |
7910 | o->get_omap_header(&old_head); | |
7911 | o->get_omap_tail(&old_tail); | |
7912 | txn->rm_range_keys(old_omap_prefix, old_head, old_tail); | |
7913 | txn->rmkey(old_omap_prefix, old_tail); | |
7914 | // set flag | |
f67539c2 | 7915 | o->onode.set_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP | bluestore_onode_t::FLAG_PERPG_OMAP); |
9f95a23c | 7916 | _record_onode(o, txn); |
9f95a23c TL |
7917 | db->submit_transaction_sync(txn); |
7918 | repairer->inc_repaired(); | |
522d829b | 7919 | repairer->request_compaction(); |
9f95a23c | 7920 | } |
eafe8130 | 7921 | } |
9f95a23c | 7922 | } |
eafe8130 | 7923 | |
9f95a23c TL |
7924 | void BlueStore::_fsck_check_objects(FSCKDepth depth, |
7925 | BlueStore::FSCK_ObjectCtx& ctx) | |
7926 | { | |
eafe8130 | 7927 | auto& errors = ctx.errors; |
eafe8130 TL |
7928 | auto sb_info_lock = ctx.sb_info_lock; |
7929 | auto& sb_info = ctx.sb_info; | |
7930 | auto repairer = ctx.repairer; | |
7931 | ||
7932 | uint64_t_btree_t used_nids; | |
7933 | ||
7934 | size_t processed_myself = 0; | |
7935 | ||
f67539c2 | 7936 | auto it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
7937 | mempool::bluestore_fsck::list<string> expecting_shards; |
7938 | if (it) { | |
7939 | const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads; | |
7940 | typedef ShallowFSCKThreadPool::FSCKWorkQueue<256> WQ; | |
7941 | std::unique_ptr<WQ> wq( | |
7942 | new WQ( | |
7943 | "FSCKWorkQueue", | |
7944 | (thread_count ? : 1) * 32, | |
7945 | this, | |
eafe8130 TL |
7946 | sb_info_lock, |
7947 | sb_info, | |
7948 | repairer)); | |
7949 | ||
7950 | ShallowFSCKThreadPool thread_pool(cct, "ShallowFSCKThreadPool", "ShallowFSCK", thread_count); | |
7951 | ||
7952 | thread_pool.add_work_queue(wq.get()); | |
7953 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
7954 | //not the best place but let's check anyway | |
7955 | ceph_assert(sb_info_lock); | |
7956 | thread_pool.start(); | |
7957 | } | |
7958 | ||
7959 | //fill global if not overriden below | |
7960 | CollectionRef c; | |
7961 | int64_t pool_id = -1; | |
7962 | spg_t pgid; | |
7963 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7964 | dout(30) << __func__ << " key " | |
7965 | << pretty_binary_string(it->key()) << dendl; | |
7966 | if (is_extent_shard_key(it->key())) { | |
7967 | if (depth == FSCK_SHALLOW) { | |
7968 | continue; | |
7969 | } | |
7970 | while (!expecting_shards.empty() && | |
7971 | expecting_shards.front() < it->key()) { | |
7972 | derr << "fsck error: missing shard key " | |
7973 | << pretty_binary_string(expecting_shards.front()) | |
7974 | << dendl; | |
7975 | ++errors; | |
7976 | expecting_shards.pop_front(); | |
7977 | } | |
7978 | if (!expecting_shards.empty() && | |
7979 | expecting_shards.front() == it->key()) { | |
7980 | // all good | |
7981 | expecting_shards.pop_front(); | |
7982 | continue; | |
7983 | } | |
7984 | ||
7985 | uint32_t offset; | |
7986 | string okey; | |
7987 | get_key_extent_shard(it->key(), &okey, &offset); | |
7988 | derr << "fsck error: stray shard 0x" << std::hex << offset | |
7989 | << std::dec << dendl; | |
7990 | if (expecting_shards.empty()) { | |
7991 | derr << "fsck error: " << pretty_binary_string(it->key()) | |
7992 | << " is unexpected" << dendl; | |
7993 | ++errors; | |
7994 | continue; | |
7995 | } | |
7996 | while (expecting_shards.front() > it->key()) { | |
7997 | derr << "fsck error: saw " << pretty_binary_string(it->key()) | |
7998 | << dendl; | |
7999 | derr << "fsck error: exp " | |
8000 | << pretty_binary_string(expecting_shards.front()) << dendl; | |
8001 | ++errors; | |
8002 | expecting_shards.pop_front(); | |
8003 | if (expecting_shards.empty()) { | |
8004 | break; | |
8005 | } | |
8006 | } | |
8007 | continue; | |
8008 | } | |
8009 | ||
8010 | ghobject_t oid; | |
8011 | int r = get_key_object(it->key(), &oid); | |
8012 | if (r < 0) { | |
8013 | derr << "fsck error: bad object key " | |
8014 | << pretty_binary_string(it->key()) << dendl; | |
8015 | ++errors; | |
8016 | continue; | |
8017 | } | |
8018 | if (!c || | |
8019 | oid.shard_id != pgid.shard || | |
8020 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8021 | !c->contains(oid)) { | |
8022 | c = nullptr; | |
8023 | for (auto& p : coll_map) { | |
8024 | if (p.second->contains(oid)) { | |
8025 | c = p.second; | |
8026 | break; | |
8027 | } | |
8028 | } | |
8029 | if (!c) { | |
8030 | derr << "fsck error: stray object " << oid | |
8031 | << " not owned by any collection" << dendl; | |
8032 | ++errors; | |
8033 | continue; | |
8034 | } | |
8035 | pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
8036 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
8037 | << dendl; | |
8038 | } | |
8039 | ||
8040 | if (depth != FSCK_SHALLOW && | |
8041 | !expecting_shards.empty()) { | |
8042 | for (auto& k : expecting_shards) { | |
8043 | derr << "fsck error: missing shard key " | |
8044 | << pretty_binary_string(k) << dendl; | |
8045 | } | |
8046 | ++errors; | |
8047 | expecting_shards.clear(); | |
8048 | } | |
8049 | ||
8050 | bool queued = false; | |
8051 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8052 | queued = wq->queue( | |
8053 | pool_id, | |
8054 | c, | |
8055 | oid, | |
8056 | it->key(), | |
8057 | it->value()); | |
8058 | } | |
8059 | OnodeRef o; | |
8060 | map<BlobRef, bluestore_blob_t::unused_t> referenced; | |
8061 | ||
8062 | if (!queued) { | |
8063 | ++processed_myself; | |
8064 | ||
8065 | o = fsck_check_objects_shallow( | |
8066 | depth, | |
8067 | pool_id, | |
8068 | c, | |
8069 | oid, | |
8070 | it->key(), | |
8071 | it->value(), | |
9f95a23c | 8072 | &expecting_shards, |
eafe8130 TL |
8073 | &referenced, |
8074 | ctx); | |
8075 | } | |
8076 | ||
8077 | if (depth != FSCK_SHALLOW) { | |
8078 | ceph_assert(o != nullptr); | |
8079 | if (o->onode.nid) { | |
8080 | if (o->onode.nid > nid_max) { | |
8081 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8082 | << " > nid_max " << nid_max << dendl; | |
8083 | ++errors; | |
8084 | } | |
8085 | if (used_nids.count(o->onode.nid)) { | |
8086 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8087 | << " already in use" << dendl; | |
8088 | ++errors; | |
8089 | continue; // go for next object | |
8090 | } | |
8091 | used_nids.insert(o->onode.nid); | |
8092 | } | |
8093 | for (auto& i : referenced) { | |
8094 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
8095 | << std::dec << " for " << *i.first << dendl; | |
8096 | const bluestore_blob_t& blob = i.first->get_blob(); | |
8097 | if (i.second & blob.unused) { | |
8098 | derr << "fsck error: " << oid << " blob claims unused 0x" | |
8099 | << std::hex << blob.unused | |
8100 | << " but extents reference 0x" << i.second << std::dec | |
8101 | << " on blob " << *i.first << dendl; | |
8102 | ++errors; | |
8103 | } | |
8104 | if (blob.has_csum()) { | |
8105 | uint64_t blob_len = blob.get_logical_length(); | |
8106 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused) * 8); | |
8107 | unsigned csum_count = blob.get_csum_count(); | |
8108 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
8109 | for (unsigned p = 0; p < csum_count; ++p) { | |
8110 | unsigned pos = p * csum_chunk_size; | |
8111 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
8112 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
8113 | unsigned mask = 1u << firstbit; | |
8114 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
8115 | mask |= 1u << b; | |
8116 | } | |
8117 | if ((blob.unused & mask) == mask) { | |
8118 | // this csum chunk region is marked unused | |
8119 | if (blob.get_csum_item(p) != 0) { | |
8120 | derr << "fsck error: " << oid | |
8121 | << " blob claims csum chunk 0x" << std::hex << pos | |
8122 | << "~" << csum_chunk_size | |
8123 | << " is unused (mask 0x" << mask << " of unused 0x" | |
8124 | << blob.unused << ") but csum is non-zero 0x" | |
8125 | << blob.get_csum_item(p) << std::dec << " on blob " | |
8126 | << *i.first << dendl; | |
8127 | ++errors; | |
8128 | } | |
8129 | } | |
8130 | } | |
8131 | } | |
8132 | } | |
8133 | // omap | |
8134 | if (o->onode.has_omap()) { | |
9f95a23c TL |
8135 | ceph_assert(ctx.used_omap_head); |
8136 | if (ctx.used_omap_head->count(o->onode.nid)) { | |
8137 | derr << "fsck error: " << o->oid << " omap_head " << o->onode.nid | |
8138 | << " already in use" << dendl; | |
eafe8130 TL |
8139 | ++errors; |
8140 | } else { | |
9f95a23c | 8141 | ctx.used_omap_head->insert(o->onode.nid); |
eafe8130 | 8142 | } |
9f95a23c | 8143 | } // if (o->onode.has_omap()) |
eafe8130 TL |
8144 | if (depth == FSCK_DEEP) { |
8145 | bufferlist bl; | |
8146 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; | |
8147 | uint64_t offset = 0; | |
8148 | do { | |
8149 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
8150 | int r = _do_read(c.get(), o, offset, l, bl, | |
8151 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
8152 | if (r < 0) { | |
8153 | ++errors; | |
8154 | derr << "fsck error: " << oid << std::hex | |
8155 | << " error during read: " | |
8156 | << " " << offset << "~" << l | |
8157 | << " " << cpp_strerror(r) << std::dec | |
8158 | << dendl; | |
8159 | break; | |
8160 | } | |
8161 | offset += l; | |
8162 | } while (offset < o->onode.size); | |
8163 | } // deep | |
8164 | } //if (depth != FSCK_SHALLOW) | |
8165 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8166 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8167 | wq->finalize(thread_pool, ctx); | |
8168 | if (processed_myself) { | |
8169 | // may be needs more threads? | |
8170 | dout(0) << __func__ << " partial offload" | |
8171 | << ", done myself " << processed_myself | |
8172 | << " of " << ctx.num_objects | |
8173 | << "objects, threads " << thread_count | |
8174 | << dendl; | |
8175 | } | |
8176 | } | |
8177 | } // if (it) | |
8178 | } | |
8179 | /** | |
8180 | An overview for currently implemented repair logics | |
8181 | performed in fsck in two stages: detection(+preparation) and commit. | |
8182 | Detection stage (in processing order): | |
8183 | (Issue -> Repair action to schedule) | |
8184 | - Detect undecodable keys for Shared Blobs -> Remove | |
8185 | - Detect undecodable records for Shared Blobs -> Remove | |
8186 | (might trigger missed Shared Blob detection below) | |
8187 | - Detect stray records for Shared Blobs -> Remove | |
8188 | - Detect misreferenced pextents -> Fix | |
8189 | Prepare Bloom-like filter to track cid/oid -> pextent | |
8190 | Prepare list of extents that are improperly referenced | |
8191 | Enumerate Onode records that might use 'misreferenced' pextents | |
8192 | (Bloom-like filter applied to reduce computation) | |
8193 | Per each questinable Onode enumerate all blobs and identify broken ones | |
8194 | (i.e. blobs having 'misreferences') | |
8195 | Rewrite each broken blob data by allocating another extents and | |
8196 | copying data there | |
8197 | If blob is shared - unshare it and mark corresponding Shared Blob | |
8198 | for removal | |
8199 | Release previously allocated space | |
8200 | Update Extent Map | |
8201 | - Detect missed Shared Blobs -> Recreate | |
8202 | - Detect undecodable deferred transaction -> Remove | |
8203 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
8204 | - Detect Freelist Manager's leaked entries -> Mark as free | |
8205 | - Detect statfs inconsistency - Update | |
8206 | Commit stage (separate DB commit per each step): | |
8207 | - Apply leaked FM entries fix | |
8208 | - Apply 'false free' FM entries fix | |
8209 | - Apply 'Remove' actions | |
8210 | - Apply fix for misreference pextents | |
8211 | - Apply Shared Blob recreate | |
8212 | (can be merged with the step above if misreferences were dectected) | |
8213 | - Apply StatFS update | |
8214 | */ | |
8215 | int BlueStore::_fsck(BlueStore::FSCKDepth depth, bool repair) | |
8216 | { | |
8217 | dout(1) << __func__ | |
8218 | << (repair ? " repair" : " check") | |
8219 | << (depth == FSCK_DEEP ? " (deep)" : | |
8220 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8221 | << dendl; | |
8222 | ||
8223 | // in deep mode we need R/W write access to be able to replay deferred ops | |
8224 | bool read_only = !(repair || depth == FSCK_DEEP); | |
8225 | ||
f67539c2 | 8226 | int r = _open_db_and_around(read_only); |
eafe8130 TL |
8227 | if (r < 0) |
8228 | return r; | |
7c673cae | 8229 | |
11fdf7f2 TL |
8230 | if (!read_only) { |
8231 | r = _upgrade_super(); | |
8232 | if (r < 0) { | |
8233 | goto out_db; | |
8234 | } | |
8235 | } | |
7c673cae | 8236 | |
eafe8130 | 8237 | r = _open_collections(); |
7c673cae | 8238 | if (r < 0) |
11fdf7f2 | 8239 | goto out_db; |
7c673cae FG |
8240 | |
8241 | mempool_thread.init(); | |
8242 | ||
11fdf7f2 TL |
8243 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
8244 | // enable in repair or deep mode modes only | |
8245 | if (!read_only) { | |
8246 | _kv_start(); | |
8247 | r = _deferred_replay(); | |
8248 | _kv_stop(); | |
8249 | } | |
7c673cae FG |
8250 | if (r < 0) |
8251 | goto out_scan; | |
8252 | ||
eafe8130 TL |
8253 | r = _fsck_on_open(depth, repair); |
8254 | ||
8255 | out_scan: | |
8256 | mempool_thread.shutdown(); | |
f6b5b4d7 | 8257 | _shutdown_cache(); |
eafe8130 | 8258 | out_db: |
1911f103 | 8259 | _close_db_and_around(false); |
eafe8130 TL |
8260 | |
8261 | return r; | |
8262 | } | |
8263 | ||
8264 | int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair) | |
8265 | { | |
8266 | dout(1) << __func__ | |
8267 | << " <<<START>>>" | |
8268 | << (repair ? " repair" : " check") | |
8269 | << (depth == FSCK_DEEP ? " (deep)" : | |
8270 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8271 | << " start" << dendl; | |
8272 | int64_t errors = 0; | |
8273 | int64_t warnings = 0; | |
8274 | unsigned repaired = 0; | |
8275 | ||
8276 | uint64_t_btree_t used_omap_head; | |
eafe8130 TL |
8277 | uint64_t_btree_t used_sbids; |
8278 | ||
f67539c2 | 8279 | mempool_dynamic_bitset used_blocks, bluefs_used_blocks; |
eafe8130 TL |
8280 | KeyValueDB::Iterator it; |
8281 | store_statfs_t expected_store_statfs, actual_statfs; | |
8282 | per_pool_statfs expected_pool_statfs; | |
8283 | ||
8284 | sb_info_map_t sb_info; | |
8285 | ||
8286 | uint64_t num_objects = 0; | |
8287 | uint64_t num_extents = 0; | |
8288 | uint64_t num_blobs = 0; | |
8289 | uint64_t num_spanning_blobs = 0; | |
8290 | uint64_t num_shared_blobs = 0; | |
8291 | uint64_t num_sharded_objects = 0; | |
8292 | BlueStoreRepairer repairer; | |
8293 | ||
f67539c2 TL |
8294 | auto alloc_size = fm->get_alloc_size(); |
8295 | ||
eafe8130 TL |
8296 | utime_t start = ceph_clock_now(); |
8297 | ||
8298 | _fsck_collections(&errors); | |
b32b8144 | 8299 | used_blocks.resize(fm->get_alloc_units()); |
7c673cae FG |
8300 | |
8301 | if (bluefs) { | |
f67539c2 | 8302 | interval_set<uint64_t> bluefs_extents; |
11fdf7f2 | 8303 | |
f67539c2 TL |
8304 | int r = bluefs->get_block_extents(bluefs_layout.shared_bdev, &bluefs_extents); |
8305 | ceph_assert(r == 0); | |
8306 | for (auto [start, len] : bluefs_extents) { | |
8307 | apply_for_bitset_range(start, len, alloc_size, used_blocks, | |
8308 | [&](uint64_t pos, mempool_dynamic_bitset& bs) { | |
8309 | ceph_assert(pos < bs.size()); | |
7c673cae | 8310 | bs.set(pos); |
f67539c2 TL |
8311 | } |
8312 | ); | |
8313 | } | |
8314 | } | |
8315 | ||
8316 | bluefs_used_blocks = used_blocks; | |
8317 | ||
8318 | apply_for_bitset_range( | |
8319 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), alloc_size, used_blocks, | |
8320 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
8321 | bs.set(pos); | |
7c673cae | 8322 | } |
f67539c2 TL |
8323 | ); |
8324 | ||
8325 | ||
8326 | if (repair) { | |
b3b6e05e | 8327 | repairer.init_space_usage_tracker( |
f67539c2 TL |
8328 | bdev->get_size(), |
8329 | min_alloc_size); | |
8330 | } | |
8331 | ||
8332 | if (bluefs) { | |
eafe8130 | 8333 | int r = bluefs->fsck(); |
7c673cae | 8334 | if (r < 0) { |
eafe8130 | 8335 | return r; |
7c673cae FG |
8336 | } |
8337 | if (r > 0) | |
8338 | errors += r; | |
8339 | } | |
8340 | ||
eafe8130 TL |
8341 | if (!per_pool_stat_collection) { |
8342 | const char *w; | |
8343 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_stats) { | |
8344 | w = "error"; | |
8345 | ++errors; | |
8346 | } else { | |
8347 | w = "warning"; | |
8348 | ++warnings; | |
8349 | } | |
8350 | derr << "fsck " << w << ": store not yet converted to per-pool stats" | |
8351 | << dendl; | |
8352 | } | |
f67539c2 | 8353 | if (per_pool_omap != OMAP_PER_PG) { |
9f95a23c TL |
8354 | const char *w; |
8355 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8356 | w = "error"; | |
8357 | ++errors; | |
8358 | } else { | |
8359 | w = "warning"; | |
8360 | ++warnings; | |
8361 | } | |
f67539c2 | 8362 | derr << "fsck " << w << ": store not yet converted to per-pg omap" |
9f95a23c TL |
8363 | << dendl; |
8364 | } | |
8365 | ||
11fdf7f2 | 8366 | // get expected statfs; reset unaffected fields to be able to compare |
7c673cae FG |
8367 | // structs |
8368 | statfs(&actual_statfs); | |
11fdf7f2 TL |
8369 | actual_statfs.total = 0; |
8370 | actual_statfs.internally_reserved = 0; | |
8371 | actual_statfs.available = 0; | |
8372 | actual_statfs.internal_metadata = 0; | |
8373 | actual_statfs.omap_allocated = 0; | |
8374 | ||
eafe8130 TL |
8375 | if (g_conf()->bluestore_debug_fsck_abort) { |
8376 | dout(1) << __func__ << " debug abort" << dendl; | |
8377 | goto out_scan; | |
8378 | } | |
7c673cae | 8379 | // walk PREFIX_OBJ |
eafe8130 TL |
8380 | { |
8381 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
8382 | ceph::mutex sb_info_lock = ceph::make_mutex("BlueStore::fsck::sbinfo_lock"); | |
8383 | BlueStore::FSCK_ObjectCtx ctx( | |
8384 | errors, | |
8385 | warnings, | |
8386 | num_objects, | |
8387 | num_extents, | |
8388 | num_blobs, | |
8389 | num_sharded_objects, | |
8390 | num_spanning_blobs, | |
8391 | &used_blocks, | |
8392 | &used_omap_head, | |
9f95a23c TL |
8393 | //no need for the below lock when in non-shallow mode as |
8394 | // there is no multithreading in this case | |
8395 | depth == FSCK_SHALLOW ? &sb_info_lock : nullptr, | |
eafe8130 TL |
8396 | sb_info, |
8397 | expected_store_statfs, | |
8398 | expected_pool_statfs, | |
8399 | repair ? &repairer : nullptr); | |
9f95a23c TL |
8400 | |
8401 | _fsck_check_objects(depth, ctx); | |
eafe8130 | 8402 | } |
11fdf7f2 | 8403 | |
7c673cae | 8404 | dout(1) << __func__ << " checking shared_blobs" << dendl; |
f67539c2 | 8405 | it = db->get_iterator(PREFIX_SHARED_BLOB, KeyValueDB::ITERATOR_NOCACHE); |
7c673cae | 8406 | if (it) { |
eafe8130 TL |
8407 | // FIXME minor: perhaps simplify for shallow mode? |
8408 | // fill global if not overriden below | |
8409 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 | 8410 | |
7c673cae FG |
8411 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8412 | string key = it->key(); | |
8413 | uint64_t sbid; | |
8414 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 8415 | derr << "fsck error: bad key '" << key |
7c673cae | 8416 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
8417 | if (repair) { |
8418 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8419 | } | |
7c673cae FG |
8420 | ++errors; |
8421 | continue; | |
8422 | } | |
8423 | auto p = sb_info.find(sbid); | |
8424 | if (p == sb_info.end()) { | |
3efd9988 | 8425 | derr << "fsck error: found stray shared blob data for sbid 0x" |
7c673cae | 8426 | << std::hex << sbid << std::dec << dendl; |
11fdf7f2 TL |
8427 | if (repair) { |
8428 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8429 | } | |
7c673cae FG |
8430 | ++errors; |
8431 | } else { | |
8432 | ++num_shared_blobs; | |
8433 | sb_info_t& sbi = p->second; | |
8434 | bluestore_shared_blob_t shared_blob(sbid); | |
8435 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
8436 | auto blp = bl.cbegin(); |
8437 | try { | |
8438 | decode(shared_blob, blp); | |
f67539c2 | 8439 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
8440 | ++errors; |
8441 | // Force update and don't report as missing | |
8442 | sbi.updated = sbi.passed = true; | |
8443 | ||
8444 | derr << "fsck error: failed to decode Shared Blob" | |
8445 | << pretty_binary_string(it->key()) << dendl; | |
8446 | if (repair) { | |
8447 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
8448 | << pretty_binary_string(it->key()) | |
8449 | << "', removing" << dendl; | |
8450 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8451 | } | |
8452 | continue; | |
8453 | } | |
7c673cae FG |
8454 | dout(20) << __func__ << " " << *sbi.sb << " " << shared_blob << dendl; |
8455 | if (shared_blob.ref_map != sbi.ref_map) { | |
3efd9988 | 8456 | derr << "fsck error: shared blob 0x" << std::hex << sbid |
11fdf7f2 TL |
8457 | << std::dec << " ref_map " << shared_blob.ref_map |
8458 | << " != expected " << sbi.ref_map << dendl; | |
8459 | sbi.updated = true; // will update later in repair mode only! | |
7c673cae FG |
8460 | ++errors; |
8461 | } | |
8462 | PExtentVector extents; | |
8463 | for (auto &r : shared_blob.ref_map.ref_map) { | |
8464 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
8465 | } | |
eafe8130 | 8466 | if (per_pool_stat_collection || repair) { |
11fdf7f2 TL |
8467 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; |
8468 | } | |
8469 | errors += _fsck_check_extents(sbi.cid, | |
a4b75251 | 8470 | sbi.oids.front(), |
7c673cae | 8471 | extents, |
a4b75251 | 8472 | sbi.compressed, |
b32b8144 FG |
8473 | used_blocks, |
8474 | fm->get_alloc_size(), | |
11fdf7f2 | 8475 | repair ? &repairer : nullptr, |
eafe8130 TL |
8476 | *expected_statfs, |
8477 | depth); | |
11fdf7f2 TL |
8478 | sbi.passed = true; |
8479 | } | |
8480 | } | |
8481 | } // if (it) | |
8482 | ||
8483 | if (repair && repairer.preprocess_misreference(db)) { | |
8484 | ||
8485 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
11fdf7f2 TL |
8486 | auto& misref_extents = repairer.get_misreferences(); |
8487 | interval_set<uint64_t> to_release; | |
f67539c2 | 8488 | it = db->get_iterator(PREFIX_OBJ, KeyValueDB::ITERATOR_NOCACHE); |
11fdf7f2 | 8489 | if (it) { |
eafe8130 TL |
8490 | // fill global if not overriden below |
8491 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 TL |
8492 | |
8493 | CollectionRef c; | |
8494 | spg_t pgid; | |
8495 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
8496 | bool bypass_rest = false; | |
8497 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
8498 | it->next()) { | |
8499 | dout(30) << __func__ << " key " | |
8500 | << pretty_binary_string(it->key()) << dendl; | |
8501 | if (is_extent_shard_key(it->key())) { | |
8502 | continue; | |
8503 | } | |
8504 | ||
8505 | ghobject_t oid; | |
8506 | int r = get_key_object(it->key(), &oid); | |
b3b6e05e | 8507 | if (r < 0 || !repairer.is_used(oid)) { |
11fdf7f2 TL |
8508 | continue; |
8509 | } | |
8510 | ||
8511 | if (!c || | |
8512 | oid.shard_id != pgid.shard || | |
8513 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8514 | !c->contains(oid)) { | |
8515 | c = nullptr; | |
8516 | for (auto& p : coll_map) { | |
8517 | if (p.second->contains(oid)) { | |
8518 | c = p.second; | |
8519 | break; | |
8520 | } | |
8521 | } | |
8522 | if (!c) { | |
8523 | continue; | |
8524 | } | |
eafe8130 TL |
8525 | if (per_pool_stat_collection || repair) { |
8526 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
11fdf7f2 TL |
8527 | expected_statfs = &expected_pool_statfs[pool_id]; |
8528 | } | |
8529 | } | |
b3b6e05e | 8530 | if (!repairer.is_used(c->cid)) { |
11fdf7f2 TL |
8531 | continue; |
8532 | } | |
8533 | ||
8534 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
8535 | << " obj:" << oid << dendl; | |
8536 | ||
eafe8130 TL |
8537 | OnodeRef o; |
8538 | o.reset(Onode::decode(c, oid, it->key(), it->value())); | |
11fdf7f2 TL |
8539 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8540 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
8541 | ||
8542 | for (auto& e : o->extent_map.extent_map) { | |
8543 | blobs.insert(e.blob); | |
8544 | } | |
8545 | bool need_onode_update = false; | |
8546 | bool first_dump = true; | |
8547 | for(auto b : blobs) { | |
8548 | bool broken_blob = false; | |
8549 | auto& pextents = b->dirty_blob().dirty_extents(); | |
8550 | for (auto& e : pextents) { | |
8551 | if (!e.is_valid()) { | |
8552 | continue; | |
8553 | } | |
8554 | // for the sake of simplicity and proper shared blob handling | |
8555 | // always rewrite the whole blob even when it's partially | |
8556 | // misreferenced. | |
8557 | if (misref_extents.intersects(e.offset, e.length)) { | |
8558 | if (first_dump) { | |
8559 | first_dump = false; | |
81eedcae | 8560 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
8561 | } |
8562 | broken_blob = true; | |
8563 | break; | |
8564 | } | |
8565 | } | |
8566 | if (!broken_blob) | |
8567 | continue; | |
8568 | bool compressed = b->get_blob().is_compressed(); | |
8569 | need_onode_update = true; | |
8570 | dout(10) << __func__ | |
8571 | << " fix misreferences in oid:" << oid | |
8572 | << " " << *b << dendl; | |
8573 | uint64_t b_off = 0; | |
8574 | PExtentVector pext_to_release; | |
8575 | pext_to_release.reserve(pextents.size()); | |
8576 | // rewriting all valid pextents | |
8577 | for (auto e = pextents.begin(); e != pextents.end(); | |
a4b75251 TL |
8578 | e++) { |
8579 | auto b_off_cur = b_off; | |
8580 | b_off += e->length; | |
11fdf7f2 TL |
8581 | if (!e->is_valid()) { |
8582 | continue; | |
8583 | } | |
8584 | PExtentVector exts; | |
f67539c2 TL |
8585 | int64_t alloc_len = |
8586 | shared_alloc.a->allocate(e->length, min_alloc_size, | |
8587 | 0, 0, &exts); | |
eafe8130 | 8588 | if (alloc_len < 0 || alloc_len < (int64_t)e->length) { |
11fdf7f2 TL |
8589 | derr << __func__ |
8590 | << " failed to allocate 0x" << std::hex << e->length | |
eafe8130 | 8591 | << " allocated 0x " << (alloc_len < 0 ? 0 : alloc_len) |
11fdf7f2 | 8592 | << " min_alloc_size 0x" << min_alloc_size |
f67539c2 | 8593 | << " available 0x " << shared_alloc.a->get_free() |
11fdf7f2 TL |
8594 | << std::dec << dendl; |
8595 | if (alloc_len > 0) { | |
f67539c2 | 8596 | shared_alloc.a->release(exts); |
11fdf7f2 TL |
8597 | } |
8598 | bypass_rest = true; | |
8599 | break; | |
8600 | } | |
8601 | expected_statfs->allocated += e->length; | |
8602 | if (compressed) { | |
8603 | expected_statfs->data_compressed_allocated += e->length; | |
8604 | } | |
8605 | ||
8606 | bufferlist bl; | |
8607 | IOContext ioc(cct, NULL, true); // allow EIO | |
8608 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); | |
8609 | if (r < 0) { | |
8610 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
8611 | <<"~" << e->length << std::dec << dendl; | |
8612 | ceph_abort_msg("read failed, wtf"); | |
8613 | } | |
8614 | pext_to_release.push_back(*e); | |
8615 | e = pextents.erase(e); | |
8616 | e = pextents.insert(e, exts.begin(), exts.end()); | |
8617 | b->get_blob().map_bl( | |
a4b75251 TL |
8618 | b_off_cur, |
8619 | bl, | |
11fdf7f2 TL |
8620 | [&](uint64_t offset, bufferlist& t) { |
8621 | int r = bdev->write(offset, t, false); | |
8622 | ceph_assert(r == 0); | |
8623 | }); | |
8624 | e += exts.size() - 1; | |
8625 | for (auto& p : exts) { | |
8626 | fm->allocate(p.offset, p.length, txn); | |
8627 | } | |
8628 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
8629 | ||
8630 | if (b->get_blob().is_shared()) { | |
8631 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
8632 | ||
8633 | auto sb_it = sb_info.find(b->shared_blob->get_sbid()); | |
8634 | ceph_assert(sb_it != sb_info.end()); | |
8635 | sb_info_t& sbi = sb_it->second; | |
8636 | ||
8637 | for (auto& r : sbi.ref_map.ref_map) { | |
8638 | expected_statfs->allocated -= r.second.length; | |
8639 | if (sbi.compressed) { | |
8640 | // NB: it's crucial to use compressed flag from sb_info_t | |
8641 | // as we originally used that value while accumulating | |
8642 | // expected_statfs | |
8643 | expected_statfs->data_compressed_allocated -= r.second.length; | |
8644 | } | |
8645 | } | |
8646 | sbi.updated = sbi.passed = true; | |
8647 | sbi.ref_map.clear(); | |
8648 | ||
8649 | // relying on blob's pextents to decide what to release. | |
8650 | for (auto& p : pext_to_release) { | |
8651 | to_release.union_insert(p.offset, p.length); | |
8652 | } | |
8653 | } else { | |
8654 | for (auto& p : pext_to_release) { | |
8655 | expected_statfs->allocated -= p.length; | |
8656 | if (compressed) { | |
8657 | expected_statfs->data_compressed_allocated -= p.length; | |
8658 | } | |
8659 | to_release.union_insert(p.offset, p.length); | |
8660 | } | |
8661 | } | |
8662 | if (bypass_rest) { | |
8663 | break; | |
8664 | } | |
8665 | } // for(auto b : blobs) | |
8666 | if (need_onode_update) { | |
8667 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
8668 | _record_onode(o, txn); | |
8669 | } | |
8670 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8671 | ||
8672 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
8673 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
8674 | << "~" << it.get_len() << std::dec << dendl; | |
8675 | fm->release(it.get_start(), it.get_len(), txn); | |
8676 | } | |
f67539c2 | 8677 | shared_alloc.a->release(to_release); |
11fdf7f2 TL |
8678 | to_release.clear(); |
8679 | } // if (it) { | |
8680 | } //if (repair && repairer.preprocess_misreference()) { | |
8681 | ||
eafe8130 TL |
8682 | if (depth != FSCK_SHALLOW) { |
8683 | for (auto &p : sb_info) { | |
8684 | sb_info_t& sbi = p.second; | |
8685 | if (!sbi.passed) { | |
8686 | derr << "fsck error: missing " << *sbi.sb << dendl; | |
8687 | ++errors; | |
8688 | } | |
8689 | if (repair && (!sbi.passed || sbi.updated)) { | |
8690 | auto sbid = p.first; | |
8691 | if (sbi.ref_map.empty()) { | |
8692 | ceph_assert(sbi.passed); | |
8693 | dout(20) << __func__ << " " << *sbi.sb | |
8694 | << " is empty, removing" << dendl; | |
8695 | repairer.fix_shared_blob(db, sbid, nullptr); | |
8696 | } else { | |
8697 | bufferlist bl; | |
8698 | bluestore_shared_blob_t persistent(sbid, std::move(sbi.ref_map)); | |
8699 | encode(persistent, bl); | |
8700 | dout(20) << __func__ << " " << *sbi.sb | |
a4b75251 TL |
8701 | << " is " << bl.length() << " bytes, updating" |
8702 | << dendl; | |
11fdf7f2 | 8703 | |
eafe8130 | 8704 | repairer.fix_shared_blob(db, sbid, &bl); |
a4b75251 TL |
8705 | // we need to account for shared blob pextents at both |
8706 | // stats and used blocks to avoid related errors. | |
8707 | PExtentVector extents; | |
8708 | for (auto& r : persistent.ref_map.ref_map) { | |
8709 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
8710 | } | |
8711 | auto* expected_statfs = &expected_pool_statfs[sbi.pool_id]; | |
8712 | int errors = _fsck_check_extents(sbi.cid, | |
8713 | ghobject_t(), // doesn't matter | |
8714 | extents, | |
8715 | sbi.compressed, | |
8716 | used_blocks, | |
8717 | fm->get_alloc_size(), | |
8718 | nullptr, | |
8719 | *expected_statfs, | |
8720 | depth); | |
8721 | if (errors) { | |
8722 | derr << __func__ << " " << errors | |
8723 | << " unexpected error(s) after missed shared blob repair," | |
8724 | << " perhaps worth one more repair attempt" | |
8725 | << dendl; | |
8726 | } | |
eafe8130 | 8727 | } |
7c673cae FG |
8728 | } |
8729 | } | |
8730 | } | |
11fdf7f2 TL |
8731 | sb_info.clear(); |
8732 | ||
eafe8130 TL |
8733 | // check global stats only if fscking (not repairing) w/o per-pool stats |
8734 | if (!per_pool_stat_collection && | |
8735 | !repair && | |
8736 | !(actual_statfs == expected_store_statfs)) { | |
8737 | derr << "fsck error: actual " << actual_statfs | |
8738 | << " != expected " << expected_store_statfs << dendl; | |
8739 | if (repair) { | |
8740 | repairer.fix_statfs(db, BLUESTORE_GLOBAL_STATFS_KEY, | |
8741 | expected_store_statfs); | |
11fdf7f2 | 8742 | } |
eafe8130 | 8743 | ++errors; |
7c673cae FG |
8744 | } |
8745 | ||
eafe8130 TL |
8746 | dout(1) << __func__ << " checking pool_statfs" << dendl; |
8747 | _fsck_check_pool_statfs(expected_pool_statfs, | |
8748 | errors, warnings, repair ? &repairer : nullptr); | |
8749 | ||
8750 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 8751 | dout(1) << __func__ << " checking for stray omap data " << dendl; |
f67539c2 | 8752 | it = db->get_iterator(PREFIX_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 | 8753 | if (it) { |
9f95a23c | 8754 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8755 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8756 | uint64_t omap_head; | |
f67539c2 | 8757 | |
eafe8130 | 8758 | _key_decode_u64(it->key().c_str(), &omap_head); |
f67539c2 | 8759 | |
9f95a23c | 8760 | if (used_omap_head.count(omap_head) == 0 && |
f67539c2 | 8761 | omap_head != last_omap_head) { |
9f95a23c TL |
8762 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
8763 | << "fsck error: found stray omap data on omap_head " | |
f67539c2 TL |
8764 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; |
8765 | ++errors; | |
8766 | last_omap_head = omap_head; | |
eafe8130 | 8767 | } |
7c673cae FG |
8768 | } |
8769 | } | |
f67539c2 | 8770 | it = db->get_iterator(PREFIX_PGMETA_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 | 8771 | if (it) { |
9f95a23c | 8772 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8773 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8774 | uint64_t omap_head; | |
8775 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8776 | if (used_omap_head.count(omap_head) == 0 && |
8777 | omap_head != last_omap_head) { | |
8778 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8779 | << "fsck error: found stray (pgmeta) omap data on omap_head " | |
8780 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8781 | last_omap_head = omap_head; | |
eafe8130 TL |
8782 | ++errors; |
8783 | } | |
11fdf7f2 TL |
8784 | } |
8785 | } | |
f67539c2 | 8786 | it = db->get_iterator(PREFIX_PERPOOL_OMAP, KeyValueDB::ITERATOR_NOCACHE); |
9f95a23c TL |
8787 | if (it) { |
8788 | uint64_t last_omap_head = 0; | |
8789 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8790 | uint64_t pool; | |
8791 | uint64_t omap_head; | |
8792 | string k = it->key(); | |
8793 | const char *c = k.c_str(); | |
8794 | c = _key_decode_u64(c, &pool); | |
8795 | c = _key_decode_u64(c, &omap_head); | |
8796 | if (used_omap_head.count(omap_head) == 0 && | |
f67539c2 | 8797 | omap_head != last_omap_head) { |
9f95a23c TL |
8798 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) |
8799 | << "fsck error: found stray (per-pool) omap data on omap_head " | |
8800 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8801 | ++errors; | |
f67539c2 TL |
8802 | last_omap_head = omap_head; |
8803 | } | |
8804 | } | |
8805 | } | |
8806 | it = db->get_iterator(PREFIX_PERPG_OMAP, KeyValueDB::ITERATOR_NOCACHE); | |
8807 | if (it) { | |
8808 | uint64_t last_omap_head = 0; | |
8809 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8810 | uint64_t pool; | |
8811 | uint32_t hash; | |
8812 | uint64_t omap_head; | |
8813 | string k = it->key(); | |
8814 | const char* c = k.c_str(); | |
8815 | c = _key_decode_u64(c, &pool); | |
8816 | c = _key_decode_u32(c, &hash); | |
8817 | c = _key_decode_u64(c, &omap_head); | |
8818 | if (used_omap_head.count(omap_head) == 0 && | |
8819 | omap_head != last_omap_head) { | |
8820 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8821 | << "fsck error: found stray (per-pg) omap data on omap_head " | |
8822 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8823 | ++errors; | |
8824 | last_omap_head = omap_head; | |
9f95a23c TL |
8825 | } |
8826 | } | |
8827 | } | |
eafe8130 | 8828 | dout(1) << __func__ << " checking deferred events" << dendl; |
f67539c2 | 8829 | it = db->get_iterator(PREFIX_DEFERRED, KeyValueDB::ITERATOR_NOCACHE); |
eafe8130 TL |
8830 | if (it) { |
8831 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8832 | bufferlist bl = it->value(); | |
8833 | auto p = bl.cbegin(); | |
8834 | bluestore_deferred_transaction_t wt; | |
8835 | try { | |
8836 | decode(wt, p); | |
f67539c2 | 8837 | } catch (ceph::buffer::error& e) { |
eafe8130 TL |
8838 | derr << "fsck error: failed to decode deferred txn " |
8839 | << pretty_binary_string(it->key()) << dendl; | |
8840 | if (repair) { | |
8841 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
8842 | << pretty_binary_string(it->key()) | |
8843 | << "', removing" << dendl; | |
8844 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8845 | } | |
8846 | continue; | |
8847 | } | |
8848 | dout(20) << __func__ << " deferred " << wt.seq | |
8849 | << " ops " << wt.ops.size() | |
8850 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
8851 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
9f95a23c | 8852 | apply_for_bitset_range( |
f67539c2 | 8853 | e.get_start(), e.get_len(), alloc_size, used_blocks, |
eafe8130 | 8854 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
eafe8130 TL |
8855 | bs.set(pos); |
8856 | } | |
8857 | ); | |
8858 | } | |
7c673cae | 8859 | } |
eafe8130 TL |
8860 | } |
8861 | ||
8862 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
8863 | { | |
eafe8130 TL |
8864 | fm->enumerate_reset(); |
8865 | uint64_t offset, length; | |
8866 | while (fm->enumerate_next(db, &offset, &length)) { | |
8867 | bool intersects = false; | |
9f95a23c | 8868 | apply_for_bitset_range( |
f67539c2 | 8869 | offset, length, alloc_size, used_blocks, |
eafe8130 | 8870 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
f67539c2 TL |
8871 | ceph_assert(pos < bs.size()); |
8872 | if (bs.test(pos) && !bluefs_used_blocks.test(pos)) { | |
eafe8130 TL |
8873 | if (offset == SUPER_RESERVED && |
8874 | length == min_alloc_size - SUPER_RESERVED) { | |
8875 | // this is due to the change just after luminous to min_alloc_size | |
8876 | // granularity allocations, and our baked in assumption at the top | |
8877 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
8878 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
8879 | // since we will never allocate this region below min_alloc_size. | |
8880 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
8881 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
8882 | << length << std::dec << dendl; | |
8883 | } else { | |
8884 | intersects = true; | |
8885 | if (repair) { | |
8886 | repairer.fix_false_free(db, fm, | |
8887 | pos * min_alloc_size, | |
8888 | min_alloc_size); | |
8889 | } | |
11fdf7f2 | 8890 | } |
eafe8130 TL |
8891 | } else { |
8892 | bs.set(pos); | |
8893 | } | |
7c673cae | 8894 | } |
eafe8130 TL |
8895 | ); |
8896 | if (intersects) { | |
8897 | derr << "fsck error: free extent 0x" << std::hex << offset | |
8898 | << "~" << length << std::dec | |
8899 | << " intersects allocated blocks" << dendl; | |
8900 | ++errors; | |
7c673cae | 8901 | } |
b5b8bbf5 | 8902 | } |
eafe8130 TL |
8903 | fm->enumerate_reset(); |
8904 | size_t count = used_blocks.count(); | |
8905 | if (used_blocks.size() != count) { | |
8906 | ceph_assert(used_blocks.size() > count); | |
8907 | used_blocks.flip(); | |
8908 | size_t start = used_blocks.find_first(); | |
8909 | while (start != decltype(used_blocks)::npos) { | |
8910 | size_t cur = start; | |
8911 | while (true) { | |
8912 | size_t next = used_blocks.find_next(cur); | |
8913 | if (next != cur + 1) { | |
8914 | ++errors; | |
8915 | derr << "fsck error: leaked extent 0x" << std::hex | |
8916 | << ((uint64_t)start * fm->get_alloc_size()) << "~" | |
8917 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
8918 | << dendl; | |
8919 | if (repair) { | |
8920 | repairer.fix_leaked(db, | |
8921 | fm, | |
8922 | start * min_alloc_size, | |
8923 | (cur + 1 - start) * min_alloc_size); | |
8924 | } | |
8925 | start = next; | |
8926 | break; | |
11fdf7f2 | 8927 | } |
eafe8130 | 8928 | cur = next; |
b5b8bbf5 | 8929 | } |
eafe8130 TL |
8930 | } |
8931 | used_blocks.flip(); | |
b5b8bbf5 | 8932 | } |
7c673cae FG |
8933 | } |
8934 | } | |
11fdf7f2 | 8935 | if (repair) { |
f67539c2 TL |
8936 | if (per_pool_omap != OMAP_PER_PG) { |
8937 | dout(5) << __func__ << " fixing per_pg_omap" << dendl; | |
8938 | repairer.fix_per_pool_omap(db, OMAP_PER_PG); | |
9f95a23c TL |
8939 | } |
8940 | ||
11fdf7f2 TL |
8941 | dout(5) << __func__ << " applying repair results" << dendl; |
8942 | repaired = repairer.apply(db); | |
8943 | dout(5) << __func__ << " repair applied" << dendl; | |
8944 | } | |
7c673cae | 8945 | |
eafe8130 | 8946 | out_scan: |
7c673cae FG |
8947 | dout(2) << __func__ << " " << num_objects << " objects, " |
8948 | << num_sharded_objects << " of them sharded. " | |
8949 | << dendl; | |
8950 | dout(2) << __func__ << " " << num_extents << " extents to " | |
8951 | << num_blobs << " blobs, " | |
8952 | << num_spanning_blobs << " spanning, " | |
8953 | << num_shared_blobs << " shared." | |
8954 | << dendl; | |
8955 | ||
8956 | utime_t duration = ceph_clock_now() - start; | |
9f95a23c TL |
8957 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " |
8958 | << warnings << " warnings, " | |
8959 | << repaired << " repaired, " | |
8960 | << (errors + warnings - (int)repaired) << " remaining in " | |
7c673cae | 8961 | << duration << " seconds" << dendl; |
9f95a23c TL |
8962 | |
8963 | // In non-repair mode we should return error count only as | |
8964 | // it indicates if store status is OK. | |
8965 | // In repair mode both errors and warnings are taken into account | |
8966 | // since repaired counter relates to them both. | |
8967 | return repair ? errors + warnings - (int)repaired : errors; | |
11fdf7f2 TL |
8968 | } |
8969 | ||
8970 | /// methods to inject various errors fsck can repair | |
8971 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
8972 | const bufferlist& bl) | |
8973 | { | |
8974 | KeyValueDB::Transaction txn; | |
8975 | txn = db->get_transaction(); | |
8976 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
8977 | db->submit_transaction_sync(txn); | |
8978 | }; | |
8979 | ||
a4b75251 TL |
8980 | void BlueStore::inject_no_shared_blob_key() |
8981 | { | |
8982 | KeyValueDB::Transaction txn; | |
8983 | txn = db->get_transaction(); | |
8984 | ceph_assert(blobid_last > 0); | |
8985 | // kill the last used sbid, this can be broken due to blobid preallocation | |
8986 | // in rare cases, leaving as-is for the sake of simplicity | |
8987 | uint64_t sbid = blobid_last; | |
8988 | ||
8989 | string key; | |
8990 | dout(5) << __func__<< " " << sbid << dendl; | |
8991 | get_shared_blob_key(sbid, &key); | |
8992 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
8993 | db->submit_transaction_sync(txn); | |
8994 | }; | |
8995 | ||
8996 | ||
11fdf7f2 TL |
8997 | void BlueStore::inject_leaked(uint64_t len) |
8998 | { | |
8999 | KeyValueDB::Transaction txn; | |
9000 | txn = db->get_transaction(); | |
9001 | ||
9002 | PExtentVector exts; | |
f67539c2 | 9003 | int64_t alloc_len = shared_alloc.a->allocate(len, min_alloc_size, |
11fdf7f2 TL |
9004 | min_alloc_size * 256, 0, &exts); |
9005 | ceph_assert(alloc_len >= (int64_t)len); | |
9006 | for (auto& p : exts) { | |
9007 | fm->allocate(p.offset, p.length, txn); | |
9008 | } | |
9009 | db->submit_transaction_sync(txn); | |
9010 | } | |
9011 | ||
9012 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
9013 | { | |
9014 | KeyValueDB::Transaction txn; | |
9015 | OnodeRef o; | |
9016 | CollectionRef c = _get_collection(cid); | |
9017 | ceph_assert(c); | |
9018 | { | |
9f95a23c | 9019 | std::unique_lock l{c->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9020 | o = c->get_onode(oid, false); |
9021 | ceph_assert(o); | |
9022 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
9023 | } | |
9024 | ||
9025 | bool injected = false; | |
9026 | txn = db->get_transaction(); | |
9027 | auto& em = o->extent_map.extent_map; | |
9028 | std::vector<const PExtentVector*> v; | |
9029 | if (em.size()) { | |
9030 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
9031 | } | |
9032 | if (em.size() > 1) { | |
9033 | auto it = em.end(); | |
9034 | --it; | |
9035 | v.push_back(&(it->blob->get_blob().get_extents())); | |
9036 | } | |
9037 | for (auto pext : v) { | |
9038 | if (pext->size()) { | |
9039 | auto p = pext->begin(); | |
9040 | while (p != pext->end()) { | |
9041 | if (p->is_valid()) { | |
9042 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
9043 | << "~" << p->length << std::dec << dendl; | |
9044 | fm->release(p->offset, p->length, txn); | |
9045 | injected = true; | |
9046 | break; | |
9047 | } | |
9048 | ++p; | |
9049 | } | |
9050 | } | |
9051 | } | |
9052 | ceph_assert(injected); | |
9053 | db->submit_transaction_sync(txn); | |
9054 | } | |
9055 | ||
9f95a23c TL |
9056 | void BlueStore::inject_legacy_omap() |
9057 | { | |
9058 | dout(1) << __func__ << dendl; | |
f67539c2 | 9059 | per_pool_omap = OMAP_BULK; |
9f95a23c TL |
9060 | KeyValueDB::Transaction txn; |
9061 | txn = db->get_transaction(); | |
9062 | txn->rmkey(PREFIX_SUPER, "per_pool_omap"); | |
9063 | db->submit_transaction_sync(txn); | |
9064 | } | |
9065 | ||
9066 | void BlueStore::inject_legacy_omap(coll_t cid, ghobject_t oid) | |
9067 | { | |
9068 | dout(1) << __func__ << " " | |
9069 | << cid << " " << oid | |
9070 | <<dendl; | |
9071 | KeyValueDB::Transaction txn; | |
9072 | OnodeRef o; | |
9073 | CollectionRef c = _get_collection(cid); | |
9074 | ceph_assert(c); | |
9075 | { | |
9076 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
9077 | o = c->get_onode(oid, false); | |
9078 | ceph_assert(o); | |
9079 | } | |
f67539c2 TL |
9080 | o->onode.clear_flag( |
9081 | bluestore_onode_t::FLAG_PERPG_OMAP | | |
9082 | bluestore_onode_t::FLAG_PERPOOL_OMAP | | |
9083 | bluestore_onode_t::FLAG_PGMETA_OMAP); | |
9f95a23c TL |
9084 | txn = db->get_transaction(); |
9085 | _record_onode(o, txn); | |
9086 | db->submit_transaction_sync(txn); | |
9087 | } | |
9088 | ||
9089 | ||
11fdf7f2 TL |
9090 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) |
9091 | { | |
9092 | BlueStoreRepairer repairer; | |
9093 | repairer.fix_statfs(db, key, new_statfs); | |
9094 | repairer.apply(db); | |
9095 | } | |
9096 | ||
eafe8130 TL |
9097 | void BlueStore::inject_global_statfs(const store_statfs_t& new_statfs) |
9098 | { | |
9099 | KeyValueDB::Transaction t = db->get_transaction(); | |
9100 | volatile_statfs v; | |
9101 | v = new_statfs; | |
9102 | bufferlist bl; | |
9103 | v.encode(bl); | |
9104 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
9105 | db->submit_transaction_sync(t); | |
9106 | } | |
9107 | ||
11fdf7f2 TL |
9108 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, |
9109 | coll_t cid2, ghobject_t oid2, | |
9110 | uint64_t offset) | |
9111 | { | |
9112 | OnodeRef o1; | |
9113 | CollectionRef c1 = _get_collection(cid1); | |
9114 | ceph_assert(c1); | |
9115 | { | |
9f95a23c | 9116 | std::unique_lock l{c1->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9117 | o1 = c1->get_onode(oid1, false); |
9118 | ceph_assert(o1); | |
9119 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9120 | } | |
9121 | OnodeRef o2; | |
9122 | CollectionRef c2 = _get_collection(cid2); | |
9123 | ceph_assert(c2); | |
9124 | { | |
9f95a23c | 9125 | std::unique_lock l{c2->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9126 | o2 = c2->get_onode(oid2, false); |
9127 | ceph_assert(o2); | |
9128 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9129 | } | |
9130 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
9131 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
9132 | ||
9133 | // require onode/extent layout to be the same (and simple) | |
9134 | // to make things easier | |
9135 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
9136 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
9137 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
9138 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
9139 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
9140 | ceph_assert(e1.length == e2.length); | |
9141 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
9142 | ||
9143 | KeyValueDB::Transaction txn; | |
9144 | txn = db->get_transaction(); | |
9145 | ||
9146 | // along with misreference error this will create space leaks errors | |
9147 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
9148 | o2->extent_map.dirty_range(offset, e2.length); | |
9149 | o2->extent_map.update(txn, false); | |
9150 | ||
9151 | _record_onode(o2, txn); | |
9152 | db->submit_transaction_sync(txn); | |
7c673cae FG |
9153 | } |
9154 | ||
adb31ebb TL |
9155 | void BlueStore::inject_zombie_spanning_blob(coll_t cid, ghobject_t oid, |
9156 | int16_t blob_id) | |
9157 | { | |
9158 | OnodeRef o; | |
9159 | CollectionRef c = _get_collection(cid); | |
9160 | ceph_assert(c); | |
9161 | { | |
9162 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
9163 | o = c->get_onode(oid, false); | |
9164 | ceph_assert(o); | |
9165 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
9166 | } | |
9167 | ||
9168 | BlobRef b = c->new_blob(); | |
9169 | b->id = blob_id; | |
9170 | o->extent_map.spanning_blob_map[blob_id] = b; | |
9171 | ||
9172 | KeyValueDB::Transaction txn; | |
9173 | txn = db->get_transaction(); | |
9174 | ||
9175 | _record_onode(o, txn); | |
9176 | db->submit_transaction_sync(txn); | |
9177 | } | |
9178 | ||
a4b75251 TL |
9179 | void BlueStore::inject_bluefs_file(std::string_view dir, std::string_view name, size_t new_size) |
9180 | { | |
9181 | ceph_assert(bluefs); | |
9182 | ||
9183 | BlueFS::FileWriter* p_handle = nullptr; | |
9184 | auto ret = bluefs->open_for_write(dir, name, &p_handle, false); | |
9185 | ceph_assert(ret == 0); | |
9186 | ||
9187 | std::string s('0', new_size); | |
9188 | bufferlist bl; | |
9189 | bl.append(s); | |
9190 | p_handle->append(bl); | |
9191 | ||
9192 | bluefs->fsync(p_handle); | |
9193 | bluefs->close_writer(p_handle); | |
9194 | } | |
9195 | ||
7c673cae FG |
9196 | void BlueStore::collect_metadata(map<string,string> *pm) |
9197 | { | |
9198 | dout(10) << __func__ << dendl; | |
9199 | bdev->collect_metadata("bluestore_bdev_", pm); | |
9200 | if (bluefs) { | |
9201 | (*pm)["bluefs"] = "1"; | |
9f95a23c TL |
9202 | // this value is for backward compatibility only |
9203 | (*pm)["bluefs_single_shared_device"] = \ | |
9204 | stringify((int)bluefs_layout.single_shared_device()); | |
9205 | (*pm)["bluefs_dedicated_db"] = \ | |
9206 | stringify((int)bluefs_layout.dedicated_db); | |
9207 | (*pm)["bluefs_dedicated_wal"] = \ | |
9208 | stringify((int)bluefs_layout.dedicated_wal); | |
9209 | bluefs->collect_metadata(pm, bluefs_layout.shared_bdev); | |
7c673cae FG |
9210 | } else { |
9211 | (*pm)["bluefs"] = "0"; | |
9212 | } | |
11fdf7f2 TL |
9213 | |
9214 | // report numa mapping for underlying devices | |
9215 | int node = -1; | |
9216 | set<int> nodes; | |
9217 | set<string> failed; | |
9218 | int r = get_numa_node(&node, &nodes, &failed); | |
9219 | if (r >= 0) { | |
9220 | if (!failed.empty()) { | |
9221 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
9222 | } | |
9223 | if (!nodes.empty()) { | |
9224 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
9225 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
9226 | } | |
9227 | if (node >= 0) { | |
9228 | (*pm)["objectstore_numa_node"] = stringify(node); | |
9229 | } | |
9230 | } | |
9231 | } | |
9232 | ||
9233 | int BlueStore::get_numa_node( | |
9234 | int *final_node, | |
9235 | set<int> *out_nodes, | |
9236 | set<string> *out_failed) | |
9237 | { | |
9238 | int node = -1; | |
9239 | set<string> devices; | |
9240 | get_devices(&devices); | |
9241 | set<int> nodes; | |
9242 | set<string> failed; | |
9243 | for (auto& devname : devices) { | |
9244 | int n; | |
9245 | BlkDev bdev(devname); | |
9246 | int r = bdev.get_numa_node(&n); | |
9247 | if (r < 0) { | |
9248 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
9249 | << dendl; | |
9250 | failed.insert(devname); | |
9251 | continue; | |
9252 | } | |
9253 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
9254 | << dendl; | |
9255 | nodes.insert(n); | |
9256 | if (node < 0) { | |
9257 | node = n; | |
9258 | } | |
9259 | } | |
9260 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
9261 | *final_node = node; | |
9262 | } | |
9263 | if (out_nodes) { | |
9264 | *out_nodes = nodes; | |
9265 | } | |
9266 | if (out_failed) { | |
9267 | *out_failed = failed; | |
9268 | } | |
9269 | return 0; | |
9270 | } | |
9271 | ||
9272 | int BlueStore::get_devices(set<string> *ls) | |
9273 | { | |
9274 | if (bdev) { | |
9275 | bdev->get_devices(ls); | |
9276 | if (bluefs) { | |
9277 | bluefs->get_devices(ls); | |
9278 | } | |
9279 | return 0; | |
9280 | } | |
9281 | ||
9282 | // grumble, we haven't started up yet. | |
9283 | int r = _open_path(); | |
9284 | if (r < 0) | |
9285 | goto out; | |
9286 | r = _open_fsid(false); | |
9287 | if (r < 0) | |
9288 | goto out_path; | |
9289 | r = _read_fsid(&fsid); | |
9290 | if (r < 0) | |
9291 | goto out_fsid; | |
9292 | r = _lock_fsid(); | |
9293 | if (r < 0) | |
9294 | goto out_fsid; | |
9295 | r = _open_bdev(false); | |
9296 | if (r < 0) | |
9297 | goto out_fsid; | |
9298 | r = _minimal_open_bluefs(false); | |
9299 | if (r < 0) | |
9300 | goto out_bdev; | |
9301 | bdev->get_devices(ls); | |
9302 | if (bluefs) { | |
9303 | bluefs->get_devices(ls); | |
9304 | } | |
9305 | r = 0; | |
9306 | _minimal_close_bluefs(); | |
9307 | out_bdev: | |
9308 | _close_bdev(); | |
9309 | out_fsid: | |
9310 | _close_fsid(); | |
9311 | out_path: | |
9312 | _close_path(); | |
9313 | out: | |
9314 | return r; | |
7c673cae FG |
9315 | } |
9316 | ||
11fdf7f2 | 9317 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
9318 | { |
9319 | buf->reset(); | |
11fdf7f2 | 9320 | |
f67539c2 TL |
9321 | auto prefix = per_pool_omap == OMAP_BULK ? |
9322 | PREFIX_OMAP : | |
9323 | per_pool_omap == OMAP_PER_POOL ? | |
9324 | PREFIX_PERPOOL_OMAP : | |
9325 | PREFIX_PERPG_OMAP; | |
9f95a23c | 9326 | buf->omap_allocated = |
f67539c2 | 9327 | db->estimate_prefix_size(prefix, string()); |
11fdf7f2 | 9328 | |
f67539c2 | 9329 | uint64_t bfree = shared_alloc.a->get_free(); |
7c673cae FG |
9330 | |
9331 | if (bluefs) { | |
f67539c2 | 9332 | buf->internally_reserved = 0; |
11fdf7f2 | 9333 | // include dedicated db, too, if that isn't the shared device. |
9f95a23c | 9334 | if (bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 9335 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); |
7c673cae | 9336 | } |
11fdf7f2 TL |
9337 | // call any non-omap bluefs space "internal metadata" |
9338 | buf->internal_metadata = | |
f67539c2 | 9339 | bluefs->get_used() |
11fdf7f2 | 9340 | - buf->omap_allocated; |
7c673cae FG |
9341 | } |
9342 | ||
11fdf7f2 TL |
9343 | uint64_t thin_total, thin_avail; |
9344 | if (bdev->get_thin_utilization(&thin_total, &thin_avail)) { | |
9345 | buf->total += thin_total; | |
9346 | ||
9347 | // we are limited by both the size of the virtual device and the | |
9348 | // underlying physical device. | |
9349 | bfree = std::min(bfree, thin_avail); | |
9350 | ||
9351 | buf->allocated = thin_total - thin_avail; | |
9352 | } else { | |
9353 | buf->total += bdev->get_size(); | |
9354 | } | |
9355 | buf->available = bfree; | |
9356 | } | |
9357 | ||
9358 | int BlueStore::statfs(struct store_statfs_t *buf, | |
9359 | osd_alert_list_t* alerts) | |
9360 | { | |
9361 | if (alerts) { | |
9362 | alerts->clear(); | |
9363 | _log_alerts(*alerts); | |
9364 | } | |
9365 | _get_statfs_overall(buf); | |
31f18b77 | 9366 | { |
11fdf7f2 | 9367 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 9368 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
9369 | buf->data_stored = vstatfs.stored(); |
9370 | buf->data_compressed = vstatfs.compressed(); | |
9371 | buf->data_compressed_original = vstatfs.compressed_original(); | |
9372 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
9373 | } | |
9374 | ||
9375 | dout(20) << __func__ << " " << *buf << dendl; | |
9376 | return 0; | |
9377 | } | |
9378 | ||
9f95a23c TL |
9379 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
9380 | bool *out_per_pool_omap) | |
11fdf7f2 TL |
9381 | { |
9382 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 9383 | |
11fdf7f2 TL |
9384 | if (!per_pool_stat_collection) { |
9385 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
9386 | return -ENOTSUP; | |
7c673cae | 9387 | } |
11fdf7f2 | 9388 | buf->reset(); |
7c673cae | 9389 | |
11fdf7f2 TL |
9390 | { |
9391 | std::lock_guard l(vstatfs_lock); | |
9392 | osd_pools[pool_id].publish(buf); | |
9393 | } | |
9f95a23c TL |
9394 | |
9395 | string key_prefix; | |
9396 | _key_encode_u64(pool_id, &key_prefix); | |
f67539c2 TL |
9397 | *out_per_pool_omap = per_pool_omap != OMAP_BULK; |
9398 | if (*out_per_pool_omap) { | |
9399 | auto prefix = per_pool_omap == OMAP_PER_POOL ? | |
9400 | PREFIX_PERPOOL_OMAP : | |
9401 | PREFIX_PERPG_OMAP; | |
9402 | buf->omap_allocated = db->estimate_prefix_size(prefix, key_prefix); | |
9403 | } | |
9f95a23c | 9404 | |
11fdf7f2 | 9405 | dout(10) << __func__ << *buf << dendl; |
7c673cae FG |
9406 | return 0; |
9407 | } | |
9408 | ||
81eedcae TL |
9409 | void BlueStore::_check_legacy_statfs_alert() |
9410 | { | |
9411 | string s; | |
9412 | if (!per_pool_stat_collection && | |
eafe8130 | 9413 | cct->_conf->bluestore_warn_on_legacy_statfs) { |
81eedcae TL |
9414 | s = "legacy statfs reporting detected, " |
9415 | "suggest to run store repair to get consistent statistic reports"; | |
9416 | } | |
9417 | std::lock_guard l(qlock); | |
9418 | legacy_statfs_alert = s; | |
9419 | } | |
9420 | ||
f67539c2 | 9421 | void BlueStore::_check_no_per_pg_or_pool_omap_alert() |
9f95a23c | 9422 | { |
f67539c2 TL |
9423 | string per_pg, per_pool; |
9424 | if (per_pool_omap != OMAP_PER_PG) { | |
9425 | if (cct->_conf->bluestore_warn_on_no_per_pg_omap) { | |
9426 | per_pg = "legacy (not per-pg) omap detected, " | |
9427 | "suggest to run store repair to benefit from faster PG removal"; | |
9428 | } | |
9429 | if (per_pool_omap != OMAP_PER_POOL) { | |
9430 | if (cct->_conf->bluestore_warn_on_no_per_pool_omap) { | |
9431 | per_pool = "legacy (not per-pool) omap detected, " | |
9432 | "suggest to run store repair to benefit from per-pool omap usage statistics"; | |
9433 | } | |
9434 | } | |
9f95a23c TL |
9435 | } |
9436 | std::lock_guard l(qlock); | |
f67539c2 TL |
9437 | no_per_pg_omap_alert = per_pg; |
9438 | no_per_pool_omap_alert = per_pool; | |
9f95a23c TL |
9439 | } |
9440 | ||
7c673cae FG |
9441 | // --------------- |
9442 | // cache | |
9443 | ||
9444 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
9445 | { | |
9f95a23c | 9446 | std::shared_lock l(coll_lock); |
7c673cae FG |
9447 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); |
9448 | if (cp == coll_map.end()) | |
9449 | return CollectionRef(); | |
9450 | return cp->second; | |
9451 | } | |
9452 | ||
9453 | void BlueStore::_queue_reap_collection(CollectionRef& c) | |
9454 | { | |
9455 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
9456 | // _reap_collections and this in the same thread, |
9457 | // so no need a lock. | |
7c673cae FG |
9458 | removed_collections.push_back(c); |
9459 | } | |
9460 | ||
9461 | void BlueStore::_reap_collections() | |
9462 | { | |
94b18763 | 9463 | |
7c673cae FG |
9464 | list<CollectionRef> removed_colls; |
9465 | { | |
94b18763 FG |
9466 | // _queue_reap_collection and this in the same thread. |
9467 | // So no need a lock. | |
9468 | if (!removed_collections.empty()) | |
9469 | removed_colls.swap(removed_collections); | |
9470 | else | |
9471 | return; | |
7c673cae FG |
9472 | } |
9473 | ||
94b18763 FG |
9474 | list<CollectionRef>::iterator p = removed_colls.begin(); |
9475 | while (p != removed_colls.end()) { | |
7c673cae FG |
9476 | CollectionRef c = *p; |
9477 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
adb31ebb | 9478 | if (c->onode_map.map_any([&](Onode* o) { |
11fdf7f2 | 9479 | ceph_assert(!o->exists); |
7c673cae FG |
9480 | if (o->flushing_count.load()) { |
9481 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
9482 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 9483 | return true; |
7c673cae | 9484 | } |
94b18763 | 9485 | return false; |
7c673cae | 9486 | })) { |
94b18763 | 9487 | ++p; |
7c673cae FG |
9488 | continue; |
9489 | } | |
9490 | c->onode_map.clear(); | |
94b18763 | 9491 | p = removed_colls.erase(p); |
7c673cae FG |
9492 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
9493 | } | |
94b18763 | 9494 | if (removed_colls.empty()) { |
7c673cae | 9495 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
9496 | } else { |
9497 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
9498 | } |
9499 | } | |
9500 | ||
9501 | void BlueStore::_update_cache_logger() | |
9502 | { | |
9503 | uint64_t num_onodes = 0; | |
9f95a23c | 9504 | uint64_t num_pinned_onodes = 0; |
7c673cae FG |
9505 | uint64_t num_extents = 0; |
9506 | uint64_t num_blobs = 0; | |
9507 | uint64_t num_buffers = 0; | |
9508 | uint64_t num_buffer_bytes = 0; | |
9f95a23c TL |
9509 | for (auto c : onode_cache_shards) { |
9510 | c->add_stats(&num_onodes, &num_pinned_onodes); | |
9511 | } | |
9512 | for (auto c : buffer_cache_shards) { | |
9513 | c->add_stats(&num_extents, &num_blobs, | |
9514 | &num_buffers, &num_buffer_bytes); | |
7c673cae FG |
9515 | } |
9516 | logger->set(l_bluestore_onodes, num_onodes); | |
9f95a23c | 9517 | logger->set(l_bluestore_pinned_onodes, num_pinned_onodes); |
7c673cae FG |
9518 | logger->set(l_bluestore_extents, num_extents); |
9519 | logger->set(l_bluestore_blobs, num_blobs); | |
9520 | logger->set(l_bluestore_buffers, num_buffers); | |
9521 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
9522 | } | |
9523 | ||
9524 | // --------------- | |
9525 | // read operations | |
9526 | ||
9527 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
9528 | { | |
9529 | return _get_collection(cid); | |
9530 | } | |
9531 | ||
11fdf7f2 TL |
9532 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
9533 | const coll_t& cid) | |
7c673cae | 9534 | { |
9f95a23c TL |
9535 | std::unique_lock l{coll_lock}; |
9536 | auto c = ceph::make_ref<Collection>( | |
11fdf7f2 | 9537 | this, |
9f95a23c TL |
9538 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
9539 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
11fdf7f2 TL |
9540 | cid); |
9541 | new_coll_map[cid] = c; | |
9f95a23c | 9542 | _osr_attach(c.get()); |
11fdf7f2 TL |
9543 | return c; |
9544 | } | |
9545 | ||
9546 | void BlueStore::set_collection_commit_queue( | |
9547 | const coll_t& cid, | |
9548 | ContextQueue *commit_queue) | |
9549 | { | |
9550 | if (commit_queue) { | |
9f95a23c | 9551 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
9552 | if (coll_map.count(cid)) { |
9553 | coll_map[cid]->commit_queue = commit_queue; | |
9554 | } else if (new_coll_map.count(cid)) { | |
9555 | new_coll_map[cid]->commit_queue = commit_queue; | |
9556 | } | |
9557 | } | |
7c673cae FG |
9558 | } |
9559 | ||
11fdf7f2 | 9560 | |
7c673cae FG |
9561 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
9562 | { | |
9563 | Collection *c = static_cast<Collection *>(c_.get()); | |
9564 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
9565 | if (!c->exists) | |
9566 | return false; | |
9567 | ||
9568 | bool r = true; | |
9569 | ||
9570 | { | |
9f95a23c | 9571 | std::shared_lock l(c->lock); |
7c673cae FG |
9572 | OnodeRef o = c->get_onode(oid, false); |
9573 | if (!o || !o->exists) | |
9574 | r = false; | |
9575 | } | |
9576 | ||
7c673cae FG |
9577 | return r; |
9578 | } | |
9579 | ||
7c673cae FG |
9580 | int BlueStore::stat( |
9581 | CollectionHandle &c_, | |
9582 | const ghobject_t& oid, | |
9583 | struct stat *st, | |
9584 | bool allow_eio) | |
9585 | { | |
9586 | Collection *c = static_cast<Collection *>(c_.get()); | |
9587 | if (!c->exists) | |
9588 | return -ENOENT; | |
9589 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
9590 | ||
9591 | { | |
9f95a23c | 9592 | std::shared_lock l(c->lock); |
7c673cae FG |
9593 | OnodeRef o = c->get_onode(oid, false); |
9594 | if (!o || !o->exists) | |
9595 | return -ENOENT; | |
9596 | st->st_size = o->onode.size; | |
9597 | st->st_blksize = 4096; | |
9598 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
9599 | st->st_nlink = 1; | |
9600 | } | |
9601 | ||
7c673cae FG |
9602 | int r = 0; |
9603 | if (_debug_mdata_eio(oid)) { | |
9604 | r = -EIO; | |
9605 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9606 | } | |
9607 | return r; | |
9608 | } | |
9609 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 9610 | CollectionHandle& ch, |
7c673cae FG |
9611 | const pool_opts_t& opts) |
9612 | { | |
7c673cae | 9613 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 9614 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
9615 | if (!c->exists) |
9616 | return -ENOENT; | |
9f95a23c | 9617 | std::unique_lock l{c->lock}; |
7c673cae FG |
9618 | c->pool_opts = opts; |
9619 | return 0; | |
9620 | } | |
9621 | ||
7c673cae FG |
9622 | int BlueStore::read( |
9623 | CollectionHandle &c_, | |
9624 | const ghobject_t& oid, | |
9625 | uint64_t offset, | |
9626 | size_t length, | |
9627 | bufferlist& bl, | |
224ce89b | 9628 | uint32_t op_flags) |
7c673cae | 9629 | { |
11fdf7f2 | 9630 | auto start = mono_clock::now(); |
7c673cae FG |
9631 | Collection *c = static_cast<Collection *>(c_.get()); |
9632 | const coll_t &cid = c->get_cid(); | |
9633 | dout(15) << __func__ << " " << cid << " " << oid | |
9634 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9635 | << dendl; | |
9636 | if (!c->exists) | |
9637 | return -ENOENT; | |
9638 | ||
9639 | bl.clear(); | |
9640 | int r; | |
9641 | { | |
9f95a23c | 9642 | std::shared_lock l(c->lock); |
11fdf7f2 | 9643 | auto start1 = mono_clock::now(); |
7c673cae | 9644 | OnodeRef o = c->get_onode(oid, false); |
494da23a TL |
9645 | log_latency("get_onode@read", |
9646 | l_bluestore_read_onode_meta_lat, | |
9647 | mono_clock::now() - start1, | |
9648 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9649 | if (!o || !o->exists) { |
9650 | r = -ENOENT; | |
9651 | goto out; | |
9652 | } | |
9653 | ||
9654 | if (offset == length && offset == 0) | |
9655 | length = o->onode.size; | |
9656 | ||
9657 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
9658 | if (r == -EIO) { |
9659 | logger->inc(l_bluestore_read_eio); | |
9660 | } | |
7c673cae FG |
9661 | } |
9662 | ||
9663 | out: | |
28e407b8 | 9664 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
9665 | r = -EIO; |
9666 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
9667 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
9668 | cct->_conf->bluestore_debug_random_read_err && | |
9669 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
9670 | 100.0)) == 0) { | |
224ce89b WB |
9671 | dout(0) << __func__ << ": inject random EIO" << dendl; |
9672 | r = -EIO; | |
7c673cae FG |
9673 | } |
9674 | dout(10) << __func__ << " " << cid << " " << oid | |
9675 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9676 | << " = " << r << dendl; | |
494da23a TL |
9677 | log_latency(__func__, |
9678 | l_bluestore_read_lat, | |
9679 | mono_clock::now() - start, | |
9680 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9681 | return r; |
9682 | } | |
9683 | ||
9f95a23c | 9684 | void BlueStore::_read_cache( |
7c673cae FG |
9685 | OnodeRef o, |
9686 | uint64_t offset, | |
9687 | size_t length, | |
9f95a23c TL |
9688 | int read_cache_policy, |
9689 | ready_regions_t& ready_regions, | |
9690 | blobs2read_t& blobs2read) | |
7c673cae | 9691 | { |
7c673cae | 9692 | // build blob-wise list to of stuff read (that isn't cached) |
7c673cae FG |
9693 | unsigned left = length; |
9694 | uint64_t pos = offset; | |
7c673cae FG |
9695 | auto lp = o->extent_map.seek_lextent(offset); |
9696 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
9697 | if (pos < lp->logical_offset) { | |
9698 | unsigned hole = lp->logical_offset - pos; | |
9699 | if (hole >= left) { | |
9f95a23c | 9700 | break; |
7c673cae FG |
9701 | } |
9702 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
9f95a23c | 9703 | << std::dec << dendl; |
7c673cae FG |
9704 | pos += hole; |
9705 | left -= hole; | |
9706 | } | |
94b18763 | 9707 | BlobRef& bptr = lp->blob; |
7c673cae FG |
9708 | unsigned l_off = pos - lp->logical_offset; |
9709 | unsigned b_off = l_off + lp->blob_offset; | |
9710 | unsigned b_len = std::min(left, lp->length - l_off); | |
9711 | ||
9712 | ready_regions_t cache_res; | |
9713 | interval_set<uint32_t> cache_interval; | |
9714 | bptr->shared_blob->bc.read( | |
91327a77 AA |
9715 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
9716 | read_cache_policy); | |
7c673cae | 9717 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c TL |
9718 | << " need 0x" << b_off << "~" << b_len |
9719 | << " cache has 0x" << cache_interval | |
9720 | << std::dec << dendl; | |
7c673cae FG |
9721 | |
9722 | auto pc = cache_res.begin(); | |
11fdf7f2 | 9723 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
9724 | while (b_len > 0) { |
9725 | unsigned l; | |
9726 | if (pc != cache_res.end() && | |
9f95a23c TL |
9727 | pc->first == b_off) { |
9728 | l = pc->second.length(); | |
f67539c2 | 9729 | ready_regions[pos] = std::move(pc->second); |
9f95a23c TL |
9730 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" |
9731 | << b_off << "~" << l << std::dec << dendl; | |
9732 | ++pc; | |
7c673cae | 9733 | } else { |
9f95a23c TL |
9734 | l = b_len; |
9735 | if (pc != cache_res.end()) { | |
9736 | ceph_assert(pc->first > b_off); | |
9737 | l = pc->first - b_off; | |
9738 | } | |
9739 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
9740 | << b_off << "~" << l << std::dec << dendl; | |
9741 | // merge regions | |
9742 | { | |
9743 | uint64_t r_off = b_off; | |
9744 | uint64_t r_len = l; | |
9745 | uint64_t front = r_off % chunk_size; | |
9746 | if (front) { | |
9747 | r_off -= front; | |
9748 | r_len += front; | |
9749 | } | |
9750 | unsigned tail = r_len % chunk_size; | |
9751 | if (tail) { | |
9752 | r_len += chunk_size - tail; | |
9753 | } | |
9754 | bool merged = false; | |
9755 | regions2read_t& r2r = blobs2read[bptr]; | |
9756 | if (r2r.size()) { | |
9757 | read_req_t& pre = r2r.back(); | |
9758 | if (r_off <= (pre.r_off + pre.r_len)) { | |
9759 | front += (r_off - pre.r_off); | |
9760 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
9761 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9762 | merged = true; | |
9763 | } | |
9764 | } | |
9765 | if (!merged) { | |
9766 | read_req_t req(r_off, r_len); | |
9767 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9768 | r2r.emplace_back(std::move(req)); | |
9769 | } | |
9770 | } | |
7c673cae FG |
9771 | } |
9772 | pos += l; | |
9773 | b_off += l; | |
9774 | left -= l; | |
9775 | b_len -= l; | |
9776 | } | |
9777 | ++lp; | |
9778 | } | |
9f95a23c | 9779 | } |
7c673cae | 9780 | |
9f95a23c TL |
9781 | int BlueStore::_prepare_read_ioc( |
9782 | blobs2read_t& blobs2read, | |
9783 | vector<bufferlist>* compressed_blob_bls, | |
9784 | IOContext* ioc) | |
9785 | { | |
7c673cae | 9786 | for (auto& p : blobs2read) { |
94b18763 | 9787 | const BlobRef& bptr = p.first; |
11fdf7f2 | 9788 | regions2read_t& r2r = p.second; |
7c673cae | 9789 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9790 | << " need " << r2r << std::dec << dendl; |
7c673cae FG |
9791 | if (bptr->get_blob().is_compressed()) { |
9792 | // read the whole thing | |
9f95a23c TL |
9793 | if (compressed_blob_bls->empty()) { |
9794 | // ensure we avoid any reallocation on subsequent blobs | |
9795 | compressed_blob_bls->reserve(blobs2read.size()); | |
9796 | } | |
9797 | compressed_blob_bls->push_back(bufferlist()); | |
9798 | bufferlist& bl = compressed_blob_bls->back(); | |
9799 | auto r = bptr->get_blob().map( | |
9800 | 0, bptr->get_blob().get_ondisk_length(), | |
9801 | [&](uint64_t offset, uint64_t length) { | |
9802 | int r = bdev->aio_read(offset, length, &bl, ioc); | |
9803 | if (r < 0) | |
7c673cae FG |
9804 | return r; |
9805 | return 0; | |
9f95a23c | 9806 | }); |
b32b8144 FG |
9807 | if (r < 0) { |
9808 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
9809 | if (r == -EIO) { | |
9810 | // propagate EIO to caller | |
9811 | return r; | |
9812 | } | |
11fdf7f2 | 9813 | ceph_assert(r == 0); |
b32b8144 | 9814 | } |
7c673cae FG |
9815 | } else { |
9816 | // read the pieces | |
11fdf7f2 | 9817 | for (auto& req : r2r) { |
9f95a23c TL |
9818 | dout(20) << __func__ << " region 0x" << std::hex |
9819 | << req.regs.front().logical_offset | |
9820 | << ": 0x" << req.regs.front().blob_xoffset | |
9821 | << " reading 0x" << req.r_off | |
9822 | << "~" << req.r_len << std::dec | |
9823 | << dendl; | |
7c673cae | 9824 | |
9f95a23c TL |
9825 | // read it |
9826 | auto r = bptr->get_blob().map( | |
9827 | req.r_off, req.r_len, | |
9828 | [&](uint64_t offset, uint64_t length) { | |
9829 | int r = bdev->aio_read(offset, length, &req.bl, ioc); | |
9830 | if (r < 0) | |
7c673cae FG |
9831 | return r; |
9832 | return 0; | |
9f95a23c | 9833 | }); |
b32b8144 FG |
9834 | if (r < 0) { |
9835 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
9836 | << dendl; | |
9837 | if (r == -EIO) { | |
9838 | // propagate EIO to caller | |
9839 | return r; | |
9840 | } | |
11fdf7f2 | 9841 | ceph_assert(r == 0); |
b32b8144 | 9842 | } |
9f95a23c | 9843 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
9844 | } |
9845 | } | |
9846 | } | |
9f95a23c TL |
9847 | return 0; |
9848 | } | |
11fdf7f2 | 9849 | |
9f95a23c TL |
9850 | int BlueStore::_generate_read_result_bl( |
9851 | OnodeRef o, | |
9852 | uint64_t offset, | |
9853 | size_t length, | |
9854 | ready_regions_t& ready_regions, | |
9855 | vector<bufferlist>& compressed_blob_bls, | |
9856 | blobs2read_t& blobs2read, | |
9857 | bool buffered, | |
9858 | bool* csum_error, | |
9859 | bufferlist& bl) | |
9860 | { | |
9861 | // enumerate and decompress desired blobs | |
7c673cae FG |
9862 | auto p = compressed_blob_bls.begin(); |
9863 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
9864 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 9865 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 9866 | regions2read_t& r2r = b2r_it->second; |
7c673cae | 9867 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9868 | << " need 0x" << r2r << std::dec << dendl; |
7c673cae | 9869 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 9870 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
9871 | bufferlist& compressed_bl = *p++; |
9872 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
9f95a23c TL |
9873 | r2r.front().regs.front().logical_offset) < 0) { |
9874 | *csum_error = true; | |
9875 | return -EIO; | |
7c673cae FG |
9876 | } |
9877 | bufferlist raw_bl; | |
9f95a23c | 9878 | auto r = _decompress(compressed_bl, &raw_bl); |
7c673cae | 9879 | if (r < 0) |
9f95a23c | 9880 | return r; |
7c673cae | 9881 | if (buffered) { |
9f95a23c TL |
9882 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, |
9883 | raw_bl); | |
7c673cae | 9884 | } |
11fdf7f2 TL |
9885 | for (auto& req : r2r) { |
9886 | for (auto& r : req.regs) { | |
9887 | ready_regions[r.logical_offset].substr_of( | |
9888 | raw_bl, r.blob_xoffset, r.length); | |
9889 | } | |
7c673cae FG |
9890 | } |
9891 | } else { | |
11fdf7f2 | 9892 | for (auto& req : r2r) { |
9f95a23c TL |
9893 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, |
9894 | req.regs.front().logical_offset) < 0) { | |
9895 | *csum_error = true; | |
9896 | return -EIO; | |
9897 | } | |
9898 | if (buffered) { | |
9899 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
9900 | req.r_off, req.bl); | |
9901 | } | |
7c673cae | 9902 | |
9f95a23c TL |
9903 | // prune and keep result |
9904 | for (const auto& r : req.regs) { | |
9905 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
11fdf7f2 | 9906 | } |
7c673cae FG |
9907 | } |
9908 | } | |
9909 | ++b2r_it; | |
9910 | } | |
9911 | ||
9912 | // generate a resulting buffer | |
9913 | auto pr = ready_regions.begin(); | |
9914 | auto pr_end = ready_regions.end(); | |
9f95a23c | 9915 | uint64_t pos = 0; |
7c673cae FG |
9916 | while (pos < length) { |
9917 | if (pr != pr_end && pr->first == pos + offset) { | |
9918 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9919 | << ": data from 0x" << pr->first << "~" << pr->second.length() |
9920 | << std::dec << dendl; | |
7c673cae FG |
9921 | pos += pr->second.length(); |
9922 | bl.claim_append(pr->second); | |
9923 | ++pr; | |
9924 | } else { | |
9925 | uint64_t l = length - pos; | |
9926 | if (pr != pr_end) { | |
11fdf7f2 | 9927 | ceph_assert(pr->first > pos + offset); |
9f95a23c | 9928 | l = pr->first - (pos + offset); |
7c673cae FG |
9929 | } |
9930 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9931 | << ": zeros for 0x" << (pos + offset) << "~" << l |
9932 | << std::dec << dendl; | |
7c673cae FG |
9933 | bl.append_zero(l); |
9934 | pos += l; | |
9935 | } | |
9936 | } | |
11fdf7f2 TL |
9937 | ceph_assert(bl.length() == length); |
9938 | ceph_assert(pos == length); | |
9939 | ceph_assert(pr == pr_end); | |
9f95a23c TL |
9940 | return 0; |
9941 | } | |
9942 | ||
9943 | int BlueStore::_do_read( | |
9944 | Collection *c, | |
9945 | OnodeRef o, | |
9946 | uint64_t offset, | |
9947 | size_t length, | |
9948 | bufferlist& bl, | |
9949 | uint32_t op_flags, | |
9950 | uint64_t retry_count) | |
9951 | { | |
9952 | FUNCTRACE(cct); | |
9953 | int r = 0; | |
9954 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
9955 | ||
9956 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
9957 | << " size 0x" << o->onode.size << " (" << std::dec | |
9958 | << o->onode.size << ")" << dendl; | |
9959 | bl.clear(); | |
9960 | ||
9961 | if (offset >= o->onode.size) { | |
9962 | return r; | |
9963 | } | |
9964 | ||
9965 | // generally, don't buffer anything, unless the client explicitly requests | |
9966 | // it. | |
9967 | bool buffered = false; | |
9968 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
9969 | dout(20) << __func__ << " will do buffered read" << dendl; | |
9970 | buffered = true; | |
9971 | } else if (cct->_conf->bluestore_default_buffered_read && | |
9972 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
9973 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
9974 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
9975 | buffered = true; | |
9976 | } | |
9977 | ||
9978 | if (offset + length > o->onode.size) { | |
9979 | length = o->onode.size - offset; | |
9980 | } | |
9981 | ||
9982 | auto start = mono_clock::now(); | |
9983 | o->extent_map.fault_range(db, offset, length); | |
9984 | log_latency(__func__, | |
9985 | l_bluestore_read_onode_meta_lat, | |
9986 | mono_clock::now() - start, | |
9987 | cct->_conf->bluestore_log_op_age); | |
9988 | _dump_onode<30>(cct, *o); | |
9989 | ||
9990 | // for deep-scrub, we only read dirty cache and bypass clean cache in | |
9991 | // order to read underlying block device in case there are silent disk errors. | |
9992 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
9993 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
9994 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
9995 | } | |
9996 | ||
9997 | // build blob-wise list to of stuff read (that isn't cached) | |
9998 | ready_regions_t ready_regions; | |
9999 | blobs2read_t blobs2read; | |
10000 | _read_cache(o, offset, length, read_cache_policy, ready_regions, blobs2read); | |
10001 | ||
10002 | ||
10003 | // read raw blob data. | |
10004 | start = mono_clock::now(); // for the sake of simplicity | |
10005 | // measure the whole block below. | |
10006 | // The error isn't that much... | |
10007 | vector<bufferlist> compressed_blob_bls; | |
10008 | IOContext ioc(cct, NULL, true); // allow EIO | |
10009 | r = _prepare_read_ioc(blobs2read, &compressed_blob_bls, &ioc); | |
10010 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10011 | if (r < 0) | |
10012 | return r; | |
10013 | ||
f67539c2 | 10014 | int64_t num_ios = blobs2read.size(); |
9f95a23c | 10015 | if (ioc.has_pending_aios()) { |
f67539c2 | 10016 | num_ios = ioc.get_num_ios(); |
9f95a23c TL |
10017 | bdev->aio_submit(&ioc); |
10018 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10019 | ioc.aio_wait(); | |
10020 | r = ioc.get_return_value(); | |
10021 | if (r < 0) { | |
10022 | ceph_assert(r == -EIO); // no other errors allowed | |
10023 | return -EIO; | |
10024 | } | |
10025 | } | |
10026 | log_latency_fn(__func__, | |
10027 | l_bluestore_read_wait_aio_lat, | |
10028 | mono_clock::now() - start, | |
10029 | cct->_conf->bluestore_log_op_age, | |
10030 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10031 | ); | |
10032 | ||
10033 | bool csum_error = false; | |
10034 | r = _generate_read_result_bl(o, offset, length, ready_regions, | |
10035 | compressed_blob_bls, blobs2read, | |
10036 | buffered, &csum_error, bl); | |
10037 | if (csum_error) { | |
10038 | // Handles spurious read errors caused by a kernel bug. | |
10039 | // We sometimes get all-zero pages as a result of the read under | |
10040 | // high memory pressure. Retrying the failing read succeeds in most | |
10041 | // cases. | |
10042 | // See also: http://tracker.ceph.com/issues/22464 | |
10043 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10044 | return -EIO; | |
10045 | } | |
10046 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
10047 | } | |
7c673cae | 10048 | r = bl.length(); |
f64942e4 AA |
10049 | if (retry_count) { |
10050 | logger->inc(l_bluestore_reads_with_retries); | |
10051 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
10052 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
f67539c2 TL |
10053 | stringstream s; |
10054 | s << " reads with retries: " << logger->get(l_bluestore_reads_with_retries); | |
10055 | _set_spurious_read_errors_alert(s.str()); | |
f64942e4 | 10056 | } |
7c673cae FG |
10057 | return r; |
10058 | } | |
10059 | ||
10060 | int BlueStore::_verify_csum(OnodeRef& o, | |
10061 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
10062 | const bufferlist& bl, | |
10063 | uint64_t logical_offset) const | |
10064 | { | |
10065 | int bad; | |
10066 | uint64_t bad_csum; | |
11fdf7f2 | 10067 | auto start = mono_clock::now(); |
7c673cae | 10068 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
10069 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
10070 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
10071 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
10072 | bad = blob_xoffset; | |
10073 | r = -1; | |
10074 | bad_csum = 0xDEADBEEF; | |
10075 | } | |
7c673cae FG |
10076 | if (r < 0) { |
10077 | if (r == -1) { | |
10078 | PExtentVector pex; | |
10079 | blob->map( | |
10080 | bad, | |
10081 | blob->get_csum_chunk_size(), | |
10082 | [&](uint64_t offset, uint64_t length) { | |
10083 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
10084 | return 0; | |
10085 | }); | |
10086 | derr << __func__ << " bad " | |
10087 | << Checksummer::get_csum_type_string(blob->csum_type) | |
10088 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
10089 | << " checksum at blob offset 0x" << bad | |
10090 | << ", got 0x" << bad_csum << ", expected 0x" | |
10091 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
10092 | << ", device location " << pex | |
10093 | << ", logical extent 0x" << std::hex | |
10094 | << (logical_offset + bad - blob_xoffset) << "~" | |
10095 | << blob->get_csum_chunk_size() << std::dec | |
10096 | << ", object " << o->oid | |
10097 | << dendl; | |
10098 | } else { | |
10099 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
10100 | } | |
10101 | } | |
494da23a TL |
10102 | log_latency(__func__, |
10103 | l_bluestore_csum_lat, | |
10104 | mono_clock::now() - start, | |
10105 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 TL |
10106 | if (cct->_conf->bluestore_ignore_data_csum) { |
10107 | return 0; | |
10108 | } | |
7c673cae FG |
10109 | return r; |
10110 | } | |
10111 | ||
10112 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
10113 | { | |
10114 | int r = 0; | |
11fdf7f2 TL |
10115 | auto start = mono_clock::now(); |
10116 | auto i = source.cbegin(); | |
7c673cae | 10117 | bluestore_compression_header_t chdr; |
11fdf7f2 | 10118 | decode(chdr, i); |
7c673cae FG |
10119 | int alg = int(chdr.type); |
10120 | CompressorRef cp = compressor; | |
10121 | if (!cp || (int)cp->get_type() != alg) { | |
10122 | cp = Compressor::create(cct, alg); | |
10123 | } | |
10124 | ||
10125 | if (!cp.get()) { | |
10126 | // if compressor isn't available - error, because cannot return | |
10127 | // decompressed data? | |
11fdf7f2 TL |
10128 | |
10129 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
10130 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
10131 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
10132 | r = -EIO; |
10133 | } else { | |
f67539c2 | 10134 | r = cp->decompress(i, chdr.length, *result, chdr.compressor_message); |
7c673cae FG |
10135 | if (r < 0) { |
10136 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
10137 | r = -EIO; | |
10138 | } | |
10139 | } | |
494da23a TL |
10140 | log_latency(__func__, |
10141 | l_bluestore_decompress_lat, | |
10142 | mono_clock::now() - start, | |
10143 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
10144 | return r; |
10145 | } | |
10146 | ||
10147 | // this stores fiemap into interval_set, other variations | |
10148 | // use it internally | |
10149 | int BlueStore::_fiemap( | |
10150 | CollectionHandle &c_, | |
10151 | const ghobject_t& oid, | |
10152 | uint64_t offset, | |
10153 | size_t length, | |
10154 | interval_set<uint64_t>& destset) | |
10155 | { | |
10156 | Collection *c = static_cast<Collection *>(c_.get()); | |
10157 | if (!c->exists) | |
10158 | return -ENOENT; | |
10159 | { | |
9f95a23c | 10160 | std::shared_lock l(c->lock); |
7c673cae FG |
10161 | |
10162 | OnodeRef o = c->get_onode(oid, false); | |
10163 | if (!o || !o->exists) { | |
10164 | return -ENOENT; | |
10165 | } | |
81eedcae | 10166 | _dump_onode<30>(cct, *o); |
7c673cae FG |
10167 | |
10168 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10169 | << " size 0x" << o->onode.size << std::dec << dendl; | |
10170 | ||
10171 | boost::intrusive::set<Extent>::iterator ep, eend; | |
10172 | if (offset >= o->onode.size) | |
10173 | goto out; | |
10174 | ||
10175 | if (offset + length > o->onode.size) { | |
10176 | length = o->onode.size - offset; | |
10177 | } | |
10178 | ||
10179 | o->extent_map.fault_range(db, offset, length); | |
10180 | eend = o->extent_map.extent_map.end(); | |
10181 | ep = o->extent_map.seek_lextent(offset); | |
10182 | while (length > 0) { | |
10183 | dout(20) << __func__ << " offset " << offset << dendl; | |
10184 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
10185 | ++ep; | |
10186 | continue; | |
10187 | } | |
10188 | ||
10189 | uint64_t x_len = length; | |
10190 | if (ep != eend && ep->logical_offset <= offset) { | |
10191 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 10192 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
10193 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
10194 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
10195 | destset.insert(offset, x_len); | |
10196 | length -= x_len; | |
10197 | offset += x_len; | |
10198 | if (x_off + x_len == ep->length) | |
10199 | ++ep; | |
10200 | continue; | |
10201 | } | |
10202 | if (ep != eend && | |
10203 | ep->logical_offset > offset && | |
10204 | ep->logical_offset - offset < x_len) { | |
10205 | x_len = ep->logical_offset - offset; | |
10206 | } | |
10207 | offset += x_len; | |
10208 | length -= x_len; | |
10209 | } | |
10210 | } | |
9f95a23c TL |
10211 | |
10212 | out: | |
10213 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10214 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
10215 | return 0; | |
10216 | } | |
10217 | ||
10218 | int BlueStore::fiemap( | |
10219 | CollectionHandle &c_, | |
10220 | const ghobject_t& oid, | |
10221 | uint64_t offset, | |
10222 | size_t length, | |
10223 | bufferlist& bl) | |
10224 | { | |
10225 | interval_set<uint64_t> m; | |
10226 | int r = _fiemap(c_, oid, offset, length, m); | |
10227 | if (r >= 0) { | |
10228 | encode(m, bl); | |
10229 | } | |
10230 | return r; | |
10231 | } | |
10232 | ||
10233 | int BlueStore::fiemap( | |
10234 | CollectionHandle &c_, | |
10235 | const ghobject_t& oid, | |
10236 | uint64_t offset, | |
10237 | size_t length, | |
10238 | map<uint64_t, uint64_t>& destmap) | |
10239 | { | |
10240 | interval_set<uint64_t> m; | |
10241 | int r = _fiemap(c_, oid, offset, length, m); | |
10242 | if (r >= 0) { | |
10243 | destmap = std::move(m).detach(); | |
10244 | } | |
10245 | return r; | |
10246 | } | |
10247 | ||
10248 | int BlueStore::readv( | |
10249 | CollectionHandle &c_, | |
10250 | const ghobject_t& oid, | |
10251 | interval_set<uint64_t>& m, | |
10252 | bufferlist& bl, | |
10253 | uint32_t op_flags) | |
10254 | { | |
10255 | auto start = mono_clock::now(); | |
10256 | Collection *c = static_cast<Collection *>(c_.get()); | |
10257 | const coll_t &cid = c->get_cid(); | |
10258 | dout(15) << __func__ << " " << cid << " " << oid | |
10259 | << " fiemap " << m | |
10260 | << dendl; | |
10261 | if (!c->exists) | |
10262 | return -ENOENT; | |
10263 | ||
10264 | bl.clear(); | |
10265 | int r; | |
10266 | { | |
10267 | std::shared_lock l(c->lock); | |
10268 | auto start1 = mono_clock::now(); | |
10269 | OnodeRef o = c->get_onode(oid, false); | |
10270 | log_latency("get_onode@read", | |
10271 | l_bluestore_read_onode_meta_lat, | |
10272 | mono_clock::now() - start1, | |
10273 | cct->_conf->bluestore_log_op_age); | |
10274 | if (!o || !o->exists) { | |
10275 | r = -ENOENT; | |
10276 | goto out; | |
10277 | } | |
10278 | ||
10279 | if (m.empty()) { | |
10280 | r = 0; | |
10281 | goto out; | |
10282 | } | |
10283 | ||
10284 | r = _do_readv(c, o, m, bl, op_flags); | |
10285 | if (r == -EIO) { | |
10286 | logger->inc(l_bluestore_read_eio); | |
10287 | } | |
10288 | } | |
10289 | ||
10290 | out: | |
10291 | if (r >= 0 && _debug_data_eio(oid)) { | |
10292 | r = -EIO; | |
10293 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10294 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ | |
10295 | cct->_conf->bluestore_debug_random_read_err && | |
10296 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
10297 | 100.0)) == 0) { | |
10298 | dout(0) << __func__ << ": inject random EIO" << dendl; | |
10299 | r = -EIO; | |
10300 | } | |
10301 | dout(10) << __func__ << " " << cid << " " << oid | |
10302 | << " fiemap " << m << std::dec | |
10303 | << " = " << r << dendl; | |
10304 | log_latency(__func__, | |
10305 | l_bluestore_read_lat, | |
10306 | mono_clock::now() - start, | |
10307 | cct->_conf->bluestore_log_op_age); | |
10308 | return r; | |
10309 | } | |
10310 | ||
10311 | int BlueStore::_do_readv( | |
10312 | Collection *c, | |
10313 | OnodeRef o, | |
10314 | const interval_set<uint64_t>& m, | |
10315 | bufferlist& bl, | |
10316 | uint32_t op_flags, | |
10317 | uint64_t retry_count) | |
10318 | { | |
10319 | FUNCTRACE(cct); | |
10320 | int r = 0; | |
10321 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
10322 | ||
10323 | dout(20) << __func__ << " fiemap " << m << std::hex | |
10324 | << " size 0x" << o->onode.size << " (" << std::dec | |
10325 | << o->onode.size << ")" << dendl; | |
10326 | ||
10327 | // generally, don't buffer anything, unless the client explicitly requests | |
10328 | // it. | |
10329 | bool buffered = false; | |
10330 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
10331 | dout(20) << __func__ << " will do buffered read" << dendl; | |
10332 | buffered = true; | |
10333 | } else if (cct->_conf->bluestore_default_buffered_read && | |
10334 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
10335 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
10336 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
10337 | buffered = true; | |
10338 | } | |
10339 | // this method must be idempotent since we may call it several times | |
10340 | // before we finally read the expected result. | |
10341 | bl.clear(); | |
10342 | ||
10343 | // call fiemap first! | |
10344 | ceph_assert(m.range_start() <= o->onode.size); | |
10345 | ceph_assert(m.range_end() <= o->onode.size); | |
10346 | auto start = mono_clock::now(); | |
10347 | o->extent_map.fault_range(db, m.range_start(), m.range_end() - m.range_start()); | |
10348 | log_latency(__func__, | |
10349 | l_bluestore_read_onode_meta_lat, | |
10350 | mono_clock::now() - start, | |
10351 | cct->_conf->bluestore_log_op_age); | |
10352 | _dump_onode<30>(cct, *o); | |
10353 | ||
10354 | IOContext ioc(cct, NULL, true); // allow EIO | |
10355 | vector<std::tuple<ready_regions_t, vector<bufferlist>, blobs2read_t>> raw_results; | |
10356 | raw_results.reserve(m.num_intervals()); | |
10357 | int i = 0; | |
10358 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10359 | raw_results.push_back({}); | |
10360 | _read_cache(o, p.get_start(), p.get_len(), read_cache_policy, | |
10361 | std::get<0>(raw_results[i]), std::get<2>(raw_results[i])); | |
10362 | r = _prepare_read_ioc(std::get<2>(raw_results[i]), &std::get<1>(raw_results[i]), &ioc); | |
10363 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10364 | if (r < 0) | |
10365 | return r; | |
10366 | } | |
10367 | ||
10368 | auto num_ios = m.size(); | |
10369 | if (ioc.has_pending_aios()) { | |
10370 | num_ios = ioc.get_num_ios(); | |
10371 | bdev->aio_submit(&ioc); | |
10372 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10373 | ioc.aio_wait(); | |
10374 | r = ioc.get_return_value(); | |
10375 | if (r < 0) { | |
10376 | ceph_assert(r == -EIO); // no other errors allowed | |
10377 | return -EIO; | |
10378 | } | |
10379 | } | |
10380 | log_latency_fn(__func__, | |
10381 | l_bluestore_read_wait_aio_lat, | |
10382 | mono_clock::now() - start, | |
10383 | cct->_conf->bluestore_log_op_age, | |
10384 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10385 | ); | |
10386 | ||
10387 | ceph_assert(raw_results.size() == (size_t)m.num_intervals()); | |
10388 | i = 0; | |
10389 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10390 | bool csum_error = false; | |
10391 | bufferlist t; | |
10392 | r = _generate_read_result_bl(o, p.get_start(), p.get_len(), | |
10393 | std::get<0>(raw_results[i]), | |
10394 | std::get<1>(raw_results[i]), | |
10395 | std::get<2>(raw_results[i]), | |
10396 | buffered, &csum_error, t); | |
10397 | if (csum_error) { | |
10398 | // Handles spurious read errors caused by a kernel bug. | |
10399 | // We sometimes get all-zero pages as a result of the read under | |
10400 | // high memory pressure. Retrying the failing read succeeds in most | |
10401 | // cases. | |
10402 | // See also: http://tracker.ceph.com/issues/22464 | |
10403 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10404 | return -EIO; | |
10405 | } | |
10406 | return _do_readv(c, o, m, bl, op_flags, retry_count + 1); | |
10407 | } | |
10408 | bl.claim_append(t); | |
10409 | } | |
10410 | if (retry_count) { | |
10411 | logger->inc(l_bluestore_reads_with_retries); | |
10412 | dout(5) << __func__ << " read fiemap " << m | |
10413 | << " failed " << retry_count << " times before succeeding" | |
10414 | << dendl; | |
10415 | } | |
10416 | return bl.length(); | |
7c673cae FG |
10417 | } |
10418 | ||
9f95a23c | 10419 | int BlueStore::dump_onode(CollectionHandle &c_, |
7c673cae | 10420 | const ghobject_t& oid, |
9f95a23c TL |
10421 | const string& section_name, |
10422 | Formatter *f) | |
7c673cae | 10423 | { |
9f95a23c TL |
10424 | Collection *c = static_cast<Collection *>(c_.get()); |
10425 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10426 | if (!c->exists) | |
10427 | return -ENOENT; | |
7c673cae | 10428 | |
9f95a23c TL |
10429 | int r; |
10430 | { | |
10431 | std::shared_lock l(c->lock); | |
10432 | ||
10433 | OnodeRef o = c->get_onode(oid, false); | |
10434 | if (!o || !o->exists) { | |
10435 | r = -ENOENT; | |
10436 | goto out; | |
10437 | } | |
10438 | // FIXME minor: actually the next line isn't enough to | |
10439 | // load shared blobs. Leaving as is for now.. | |
10440 | // | |
10441 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10442 | ||
10443 | _dump_onode<0>(cct, *o); | |
10444 | f->open_object_section(section_name.c_str()); | |
10445 | o->dump(f); | |
10446 | f->close_section(); | |
10447 | r = 0; | |
7c673cae | 10448 | } |
9f95a23c TL |
10449 | out: |
10450 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10451 | << " = " << r << dendl; | |
7c673cae FG |
10452 | return r; |
10453 | } | |
10454 | ||
7c673cae FG |
10455 | int BlueStore::getattr( |
10456 | CollectionHandle &c_, | |
10457 | const ghobject_t& oid, | |
10458 | const char *name, | |
10459 | bufferptr& value) | |
10460 | { | |
10461 | Collection *c = static_cast<Collection *>(c_.get()); | |
10462 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
10463 | if (!c->exists) | |
10464 | return -ENOENT; | |
10465 | ||
10466 | int r; | |
10467 | { | |
9f95a23c | 10468 | std::shared_lock l(c->lock); |
f91f0fd5 | 10469 | mempool::bluestore_cache_meta::string k(name); |
7c673cae FG |
10470 | |
10471 | OnodeRef o = c->get_onode(oid, false); | |
10472 | if (!o || !o->exists) { | |
10473 | r = -ENOENT; | |
10474 | goto out; | |
10475 | } | |
10476 | ||
10477 | if (!o->onode.attrs.count(k)) { | |
10478 | r = -ENODATA; | |
10479 | goto out; | |
10480 | } | |
10481 | value = o->onode.attrs[k]; | |
10482 | r = 0; | |
10483 | } | |
10484 | out: | |
7c673cae FG |
10485 | if (r == 0 && _debug_mdata_eio(oid)) { |
10486 | r = -EIO; | |
10487 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10488 | } | |
10489 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
10490 | << " = " << r << dendl; | |
10491 | return r; | |
10492 | } | |
10493 | ||
7c673cae FG |
10494 | int BlueStore::getattrs( |
10495 | CollectionHandle &c_, | |
10496 | const ghobject_t& oid, | |
10497 | map<string,bufferptr>& aset) | |
10498 | { | |
10499 | Collection *c = static_cast<Collection *>(c_.get()); | |
10500 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10501 | if (!c->exists) | |
10502 | return -ENOENT; | |
10503 | ||
10504 | int r; | |
10505 | { | |
9f95a23c | 10506 | std::shared_lock l(c->lock); |
7c673cae FG |
10507 | |
10508 | OnodeRef o = c->get_onode(oid, false); | |
10509 | if (!o || !o->exists) { | |
10510 | r = -ENOENT; | |
10511 | goto out; | |
10512 | } | |
10513 | for (auto& i : o->onode.attrs) { | |
10514 | aset.emplace(i.first.c_str(), i.second); | |
10515 | } | |
10516 | r = 0; | |
10517 | } | |
10518 | ||
10519 | out: | |
7c673cae FG |
10520 | if (r == 0 && _debug_mdata_eio(oid)) { |
10521 | r = -EIO; | |
10522 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10523 | } | |
10524 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10525 | << " = " << r << dendl; | |
10526 | return r; | |
10527 | } | |
10528 | ||
10529 | int BlueStore::list_collections(vector<coll_t>& ls) | |
10530 | { | |
9f95a23c | 10531 | std::shared_lock l(coll_lock); |
11fdf7f2 | 10532 | ls.reserve(coll_map.size()); |
7c673cae FG |
10533 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
10534 | p != coll_map.end(); | |
10535 | ++p) | |
10536 | ls.push_back(p->first); | |
10537 | return 0; | |
10538 | } | |
10539 | ||
10540 | bool BlueStore::collection_exists(const coll_t& c) | |
10541 | { | |
9f95a23c | 10542 | std::shared_lock l(coll_lock); |
7c673cae FG |
10543 | return coll_map.count(c); |
10544 | } | |
10545 | ||
11fdf7f2 | 10546 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 10547 | { |
11fdf7f2 | 10548 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
10549 | vector<ghobject_t> ls; |
10550 | ghobject_t next; | |
11fdf7f2 | 10551 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
10552 | &ls, &next); |
10553 | if (r < 0) { | |
10554 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
10555 | << dendl; | |
10556 | return r; | |
10557 | } | |
10558 | *empty = ls.empty(); | |
11fdf7f2 | 10559 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
10560 | return 0; |
10561 | } | |
10562 | ||
11fdf7f2 | 10563 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 10564 | { |
11fdf7f2 TL |
10565 | dout(15) << __func__ << " " << ch->cid << dendl; |
10566 | Collection *c = static_cast<Collection*>(ch.get()); | |
9f95a23c | 10567 | std::shared_lock l(c->lock); |
11fdf7f2 | 10568 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
10569 | return c->cnode.bits; |
10570 | } | |
10571 | ||
7c673cae FG |
10572 | int BlueStore::collection_list( |
10573 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10574 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10575 | { | |
10576 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 10577 | c->flush(); |
7c673cae FG |
10578 | dout(15) << __func__ << " " << c->cid |
10579 | << " start " << start << " end " << end << " max " << max << dendl; | |
10580 | int r; | |
10581 | { | |
9f95a23c | 10582 | std::shared_lock l(c->lock); |
f91f0fd5 TL |
10583 | r = _collection_list(c, start, end, max, false, ls, pnext); |
10584 | } | |
10585 | ||
10586 | dout(10) << __func__ << " " << c->cid | |
10587 | << " start " << start << " end " << end << " max " << max | |
10588 | << " = " << r << ", ls.size() = " << ls->size() | |
10589 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10590 | return r; | |
10591 | } | |
10592 | ||
10593 | int BlueStore::collection_list_legacy( | |
10594 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10595 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10596 | { | |
10597 | Collection *c = static_cast<Collection *>(c_.get()); | |
10598 | c->flush(); | |
10599 | dout(15) << __func__ << " " << c->cid | |
10600 | << " start " << start << " end " << end << " max " << max << dendl; | |
10601 | int r; | |
10602 | { | |
10603 | std::shared_lock l(c->lock); | |
10604 | r = _collection_list(c, start, end, max, true, ls, pnext); | |
7c673cae FG |
10605 | } |
10606 | ||
7c673cae FG |
10607 | dout(10) << __func__ << " " << c->cid |
10608 | << " start " << start << " end " << end << " max " << max | |
10609 | << " = " << r << ", ls.size() = " << ls->size() | |
10610 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10611 | return r; | |
10612 | } | |
10613 | ||
10614 | int BlueStore::_collection_list( | |
10615 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
f91f0fd5 | 10616 | bool legacy, vector<ghobject_t> *ls, ghobject_t *pnext) |
7c673cae FG |
10617 | { |
10618 | ||
10619 | if (!c->exists) | |
10620 | return -ENOENT; | |
10621 | ||
7c673cae | 10622 | ghobject_t static_next; |
f91f0fd5 TL |
10623 | std::unique_ptr<CollectionListIterator> it; |
10624 | ghobject_t coll_range_temp_start, coll_range_temp_end; | |
10625 | ghobject_t coll_range_start, coll_range_end; | |
f91f0fd5 | 10626 | ghobject_t pend; |
7c673cae FG |
10627 | bool temp; |
10628 | ||
10629 | if (!pnext) | |
10630 | pnext = &static_next; | |
10631 | ||
a4b75251 TL |
10632 | auto log_latency = make_scope_guard( |
10633 | [&, start_time = mono_clock::now(), func_name = __func__] { | |
10634 | log_latency_fn( | |
10635 | func_name, | |
10636 | l_bluestore_remove_lat, | |
10637 | mono_clock::now() - start_time, | |
10638 | cct->_conf->bluestore_log_collection_list_age, | |
10639 | [&](const ceph::timespan& lat) { | |
10640 | ostringstream ostr; | |
10641 | ostr << ", lat = " << timespan_str(lat) | |
10642 | << " cid =" << c->cid | |
10643 | << " start " << start << " end " << end | |
10644 | << " max " << max; | |
10645 | return ostr.str(); | |
10646 | }); | |
10647 | }); | |
10648 | ||
11fdf7f2 | 10649 | if (start.is_max() || start.hobj.is_max()) { |
a4b75251 TL |
10650 | *pnext = ghobject_t::get_max(); |
10651 | return 0; | |
7c673cae | 10652 | } |
f91f0fd5 | 10653 | get_coll_range(c->cid, c->cnode.bits, &coll_range_temp_start, |
a4b75251 | 10654 | &coll_range_temp_end, &coll_range_start, &coll_range_end, legacy); |
7c673cae | 10655 | dout(20) << __func__ |
f91f0fd5 TL |
10656 | << " range " << coll_range_temp_start |
10657 | << " to " << coll_range_temp_end | |
10658 | << " and " << coll_range_start | |
10659 | << " to " << coll_range_end | |
7c673cae | 10660 | << " start " << start << dendl; |
f91f0fd5 TL |
10661 | if (legacy) { |
10662 | it = std::make_unique<SimpleCollectionListIterator>( | |
10663 | cct, db->get_iterator(PREFIX_OBJ)); | |
10664 | } else { | |
10665 | it = std::make_unique<SortedCollectionListIterator>( | |
10666 | db->get_iterator(PREFIX_OBJ)); | |
10667 | } | |
7c673cae FG |
10668 | if (start == ghobject_t() || |
10669 | start.hobj == hobject_t() || | |
10670 | start == c->cid.get_min_hobj()) { | |
f91f0fd5 | 10671 | it->upper_bound(coll_range_temp_start); |
7c673cae FG |
10672 | temp = true; |
10673 | } else { | |
7c673cae FG |
10674 | if (start.hobj.is_temp()) { |
10675 | temp = true; | |
f91f0fd5 | 10676 | ceph_assert(start >= coll_range_temp_start && start < coll_range_temp_end); |
7c673cae FG |
10677 | } else { |
10678 | temp = false; | |
f91f0fd5 | 10679 | ceph_assert(start >= coll_range_start && start < coll_range_end); |
7c673cae | 10680 | } |
f91f0fd5 TL |
10681 | dout(20) << __func__ << " temp=" << (int)temp << dendl; |
10682 | it->lower_bound(start); | |
7c673cae FG |
10683 | } |
10684 | if (end.hobj.is_max()) { | |
f91f0fd5 | 10685 | pend = temp ? coll_range_temp_end : coll_range_end; |
7c673cae | 10686 | } else { |
7c673cae | 10687 | if (end.hobj.is_temp()) { |
a4b75251 | 10688 | if (temp) { |
f91f0fd5 | 10689 | pend = end; |
a4b75251 TL |
10690 | } else { |
10691 | *pnext = ghobject_t::get_max(); | |
10692 | return 0; | |
10693 | } | |
7c673cae | 10694 | } else { |
f91f0fd5 | 10695 | pend = temp ? coll_range_temp_end : end; |
7c673cae FG |
10696 | } |
10697 | } | |
f91f0fd5 | 10698 | dout(20) << __func__ << " pend " << pend << dendl; |
7c673cae | 10699 | while (true) { |
adb31ebb | 10700 | if (!it->valid() || it->is_ge(pend)) { |
7c673cae FG |
10701 | if (!it->valid()) |
10702 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
10703 | else | |
f91f0fd5 | 10704 | dout(20) << __func__ << " oid " << it->oid() << " >= " << pend << dendl; |
7c673cae FG |
10705 | if (temp) { |
10706 | if (end.hobj.is_temp()) { | |
adb31ebb | 10707 | if (it->valid() && it->is_lt(coll_range_temp_end)) { |
f91f0fd5 | 10708 | *pnext = it->oid(); |
a4b75251 | 10709 | return 0; |
f91f0fd5 | 10710 | } |
7c673cae FG |
10711 | break; |
10712 | } | |
10713 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
10714 | temp = false; | |
f91f0fd5 TL |
10715 | it->upper_bound(coll_range_start); |
10716 | if (end.hobj.is_max()) | |
10717 | pend = coll_range_end; | |
10718 | else | |
10719 | pend = end; | |
10720 | dout(30) << __func__ << " pend " << pend << dendl; | |
7c673cae FG |
10721 | continue; |
10722 | } | |
adb31ebb | 10723 | if (it->valid() && it->is_lt(coll_range_end)) { |
f91f0fd5 | 10724 | *pnext = it->oid(); |
a4b75251 | 10725 | return 0; |
f91f0fd5 | 10726 | } |
7c673cae FG |
10727 | break; |
10728 | } | |
f91f0fd5 | 10729 | dout(20) << __func__ << " oid " << it->oid() << " end " << end << dendl; |
7c673cae FG |
10730 | if (ls->size() >= (unsigned)max) { |
10731 | dout(20) << __func__ << " reached max " << max << dendl; | |
f91f0fd5 | 10732 | *pnext = it->oid(); |
a4b75251 | 10733 | return 0; |
7c673cae | 10734 | } |
f91f0fd5 | 10735 | ls->push_back(it->oid()); |
7c673cae FG |
10736 | it->next(); |
10737 | } | |
a4b75251 TL |
10738 | *pnext = ghobject_t::get_max(); |
10739 | return 0; | |
7c673cae FG |
10740 | } |
10741 | ||
7c673cae FG |
10742 | int BlueStore::omap_get( |
10743 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10744 | const ghobject_t &oid, ///< [in] Object containing omap | |
10745 | bufferlist *header, ///< [out] omap header | |
10746 | map<string, bufferlist> *out /// < [out] Key to value map | |
10747 | ) | |
10748 | { | |
10749 | Collection *c = static_cast<Collection *>(c_.get()); | |
9f95a23c TL |
10750 | return _omap_get(c, oid, header, out); |
10751 | } | |
10752 | ||
10753 | int BlueStore::_omap_get( | |
10754 | Collection *c, ///< [in] Collection containing oid | |
10755 | const ghobject_t &oid, ///< [in] Object containing omap | |
10756 | bufferlist *header, ///< [out] omap header | |
10757 | map<string, bufferlist> *out /// < [out] Key to value map | |
10758 | ) | |
10759 | { | |
7c673cae FG |
10760 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; |
10761 | if (!c->exists) | |
10762 | return -ENOENT; | |
9f95a23c | 10763 | std::shared_lock l(c->lock); |
7c673cae FG |
10764 | int r = 0; |
10765 | OnodeRef o = c->get_onode(oid, false); | |
10766 | if (!o || !o->exists) { | |
10767 | r = -ENOENT; | |
10768 | goto out; | |
10769 | } | |
9f95a23c TL |
10770 | r = _onode_omap_get(o, header, out); |
10771 | out: | |
10772 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10773 | << dendl; | |
10774 | return r; | |
10775 | } | |
10776 | ||
10777 | int BlueStore::_onode_omap_get( | |
10778 | const OnodeRef &o, ///< [in] Object containing omap | |
10779 | bufferlist *header, ///< [out] omap header | |
10780 | map<string, bufferlist> *out /// < [out] Key to value map | |
10781 | ) | |
10782 | { | |
10783 | int r = 0; | |
10784 | if (!o || !o->exists) { | |
10785 | r = -ENOENT; | |
10786 | goto out; | |
10787 | } | |
7c673cae FG |
10788 | if (!o->onode.has_omap()) |
10789 | goto out; | |
10790 | o->flush(); | |
10791 | { | |
9f95a23c | 10792 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10793 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10794 | string head, tail; |
9f95a23c TL |
10795 | o->get_omap_header(&head); |
10796 | o->get_omap_tail(&tail); | |
7c673cae FG |
10797 | it->lower_bound(head); |
10798 | while (it->valid()) { | |
10799 | if (it->key() == head) { | |
9f95a23c TL |
10800 | dout(30) << __func__ << " got header" << dendl; |
10801 | *header = it->value(); | |
7c673cae | 10802 | } else if (it->key() >= tail) { |
9f95a23c TL |
10803 | dout(30) << __func__ << " reached tail" << dendl; |
10804 | break; | |
7c673cae | 10805 | } else { |
9f95a23c TL |
10806 | string user_key; |
10807 | o->decode_omap_key(it->key(), &user_key); | |
10808 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
10809 | << " -> " << user_key << dendl; | |
10810 | (*out)[user_key] = it->value(); | |
7c673cae FG |
10811 | } |
10812 | it->next(); | |
10813 | } | |
10814 | } | |
9f95a23c | 10815 | out: |
7c673cae FG |
10816 | return r; |
10817 | } | |
10818 | ||
7c673cae FG |
10819 | int BlueStore::omap_get_header( |
10820 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10821 | const ghobject_t &oid, ///< [in] Object containing omap | |
10822 | bufferlist *header, ///< [out] omap header | |
10823 | bool allow_eio ///< [in] don't assert on eio | |
10824 | ) | |
10825 | { | |
10826 | Collection *c = static_cast<Collection *>(c_.get()); | |
10827 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10828 | if (!c->exists) | |
10829 | return -ENOENT; | |
9f95a23c | 10830 | std::shared_lock l(c->lock); |
7c673cae FG |
10831 | int r = 0; |
10832 | OnodeRef o = c->get_onode(oid, false); | |
10833 | if (!o || !o->exists) { | |
10834 | r = -ENOENT; | |
10835 | goto out; | |
10836 | } | |
10837 | if (!o->onode.has_omap()) | |
10838 | goto out; | |
10839 | o->flush(); | |
10840 | { | |
10841 | string head; | |
9f95a23c TL |
10842 | o->get_omap_header(&head); |
10843 | if (db->get(o->get_omap_prefix(), head, header) >= 0) { | |
7c673cae FG |
10844 | dout(30) << __func__ << " got header" << dendl; |
10845 | } else { | |
10846 | dout(30) << __func__ << " no header" << dendl; | |
10847 | } | |
10848 | } | |
10849 | out: | |
10850 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10851 | << dendl; | |
10852 | return r; | |
10853 | } | |
10854 | ||
7c673cae FG |
10855 | int BlueStore::omap_get_keys( |
10856 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10857 | const ghobject_t &oid, ///< [in] Object containing omap | |
10858 | set<string> *keys ///< [out] Keys defined on oid | |
10859 | ) | |
10860 | { | |
10861 | Collection *c = static_cast<Collection *>(c_.get()); | |
10862 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10863 | if (!c->exists) | |
10864 | return -ENOENT; | |
adb31ebb | 10865 | auto start1 = mono_clock::now(); |
9f95a23c | 10866 | std::shared_lock l(c->lock); |
7c673cae FG |
10867 | int r = 0; |
10868 | OnodeRef o = c->get_onode(oid, false); | |
10869 | if (!o || !o->exists) { | |
10870 | r = -ENOENT; | |
10871 | goto out; | |
10872 | } | |
10873 | if (!o->onode.has_omap()) | |
10874 | goto out; | |
10875 | o->flush(); | |
10876 | { | |
9f95a23c | 10877 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10878 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10879 | string head, tail; |
9f95a23c TL |
10880 | o->get_omap_key(string(), &head); |
10881 | o->get_omap_tail(&tail); | |
7c673cae FG |
10882 | it->lower_bound(head); |
10883 | while (it->valid()) { | |
10884 | if (it->key() >= tail) { | |
10885 | dout(30) << __func__ << " reached tail" << dendl; | |
10886 | break; | |
10887 | } | |
10888 | string user_key; | |
9f95a23c | 10889 | o->decode_omap_key(it->key(), &user_key); |
11fdf7f2 | 10890 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
10891 | << " -> " << user_key << dendl; |
10892 | keys->insert(user_key); | |
10893 | it->next(); | |
11fdf7f2 TL |
10894 | } |
10895 | } | |
10896 | out: | |
adb31ebb TL |
10897 | c->store->log_latency( |
10898 | __func__, | |
10899 | l_bluestore_omap_get_keys_lat, | |
10900 | mono_clock::now() - start1, | |
10901 | c->store->cct->_conf->bluestore_log_omap_iterator_age); | |
10902 | ||
11fdf7f2 TL |
10903 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r |
10904 | << dendl; | |
10905 | return r; | |
7c673cae FG |
10906 | } |
10907 | ||
10908 | int BlueStore::omap_get_values( | |
10909 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10910 | const ghobject_t &oid, ///< [in] Object containing omap | |
10911 | const set<string> &keys, ///< [in] Keys to get | |
10912 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
10913 | ) | |
10914 | { | |
10915 | Collection *c = static_cast<Collection *>(c_.get()); | |
10916 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10917 | if (!c->exists) | |
10918 | return -ENOENT; | |
9f95a23c | 10919 | std::shared_lock l(c->lock); |
adb31ebb | 10920 | auto start1 = mono_clock::now(); |
7c673cae FG |
10921 | int r = 0; |
10922 | string final_key; | |
10923 | OnodeRef o = c->get_onode(oid, false); | |
10924 | if (!o || !o->exists) { | |
10925 | r = -ENOENT; | |
10926 | goto out; | |
10927 | } | |
9f95a23c | 10928 | if (!o->onode.has_omap()) { |
7c673cae | 10929 | goto out; |
9f95a23c TL |
10930 | } |
10931 | o->flush(); | |
11fdf7f2 | 10932 | { |
9f95a23c TL |
10933 | const string& prefix = o->get_omap_prefix(); |
10934 | o->get_omap_key(string(), &final_key); | |
10935 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10936 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10937 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10938 | final_key += *p; |
10939 | bufferlist val; | |
10940 | if (db->get(prefix, final_key, &val) >= 0) { | |
10941 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
10942 | << " -> " << *p << dendl; | |
10943 | out->insert(make_pair(*p, val)); | |
10944 | } | |
7c673cae FG |
10945 | } |
10946 | } | |
10947 | out: | |
adb31ebb TL |
10948 | c->store->log_latency( |
10949 | __func__, | |
10950 | l_bluestore_omap_get_values_lat, | |
10951 | mono_clock::now() - start1, | |
10952 | c->store->cct->_conf->bluestore_log_omap_iterator_age); | |
10953 | ||
7c673cae FG |
10954 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r |
10955 | << dendl; | |
10956 | return r; | |
10957 | } | |
10958 | ||
9f95a23c TL |
10959 | #ifdef WITH_SEASTAR |
10960 | int BlueStore::omap_get_values( | |
10961 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10962 | const ghobject_t &oid, ///< [in] Object containing omap | |
10963 | const std::optional<string> &start_after, ///< [in] Keys to get | |
10964 | map<string, bufferlist> *output ///< [out] Returned keys and values | |
10965 | ) | |
10966 | { | |
10967 | Collection *c = static_cast<Collection *>(c_.get()); | |
10968 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10969 | if (!c->exists) | |
10970 | return -ENOENT; | |
10971 | std::shared_lock l(c->lock); | |
10972 | int r = 0; | |
10973 | OnodeRef o = c->get_onode(oid, false); | |
10974 | if (!o || !o->exists) { | |
10975 | r = -ENOENT; | |
10976 | goto out; | |
10977 | } | |
10978 | if (!o->onode.has_omap()) { | |
10979 | goto out; | |
10980 | } | |
10981 | o->flush(); | |
10982 | { | |
10983 | ObjectMap::ObjectMapIterator iter = get_omap_iterator(c_, oid); | |
10984 | if (!iter) { | |
10985 | r = -ENOENT; | |
10986 | goto out; | |
10987 | } | |
10988 | iter->upper_bound(*start_after); | |
10989 | for (; iter->valid(); iter->next()) { | |
10990 | output->insert(make_pair(iter->key(), iter->value())); | |
10991 | } | |
10992 | } | |
10993 | ||
10994 | out: | |
10995 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10996 | << dendl; | |
10997 | return r; | |
10998 | } | |
10999 | #endif | |
11000 | ||
7c673cae FG |
11001 | int BlueStore::omap_check_keys( |
11002 | CollectionHandle &c_, ///< [in] Collection containing oid | |
11003 | const ghobject_t &oid, ///< [in] Object containing omap | |
11004 | const set<string> &keys, ///< [in] Keys to check | |
11005 | set<string> *out ///< [out] Subset of keys defined on oid | |
11006 | ) | |
11007 | { | |
11008 | Collection *c = static_cast<Collection *>(c_.get()); | |
11009 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
11010 | if (!c->exists) | |
11011 | return -ENOENT; | |
9f95a23c | 11012 | std::shared_lock l(c->lock); |
7c673cae FG |
11013 | int r = 0; |
11014 | string final_key; | |
11015 | OnodeRef o = c->get_onode(oid, false); | |
11016 | if (!o || !o->exists) { | |
11017 | r = -ENOENT; | |
11018 | goto out; | |
11019 | } | |
9f95a23c | 11020 | if (!o->onode.has_omap()) { |
7c673cae | 11021 | goto out; |
9f95a23c TL |
11022 | } |
11023 | o->flush(); | |
11fdf7f2 | 11024 | { |
9f95a23c TL |
11025 | const string& prefix = o->get_omap_prefix(); |
11026 | o->get_omap_key(string(), &final_key); | |
11027 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 11028 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 11029 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
11030 | final_key += *p; |
11031 | bufferlist val; | |
11032 | if (db->get(prefix, final_key, &val) >= 0) { | |
11033 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
11034 | << " -> " << *p << dendl; | |
11035 | out->insert(*p); | |
11036 | } else { | |
11037 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
11038 | << " -> " << *p << dendl; | |
11039 | } | |
7c673cae FG |
11040 | } |
11041 | } | |
11042 | out: | |
11043 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
11044 | << dendl; | |
11045 | return r; | |
11046 | } | |
11047 | ||
7c673cae FG |
11048 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
11049 | CollectionHandle &c_, ///< [in] collection | |
11050 | const ghobject_t &oid ///< [in] object | |
11051 | ) | |
11052 | { | |
11053 | Collection *c = static_cast<Collection *>(c_.get()); | |
11054 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
11055 | if (!c->exists) { | |
11056 | return ObjectMap::ObjectMapIterator(); | |
11057 | } | |
9f95a23c | 11058 | std::shared_lock l(c->lock); |
7c673cae FG |
11059 | OnodeRef o = c->get_onode(oid, false); |
11060 | if (!o || !o->exists) { | |
11061 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
11062 | return ObjectMap::ObjectMapIterator(); | |
11063 | } | |
11064 | o->flush(); | |
11065 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
9f95a23c | 11066 | KeyValueDB::Iterator it = db->get_iterator(o->get_omap_prefix()); |
7c673cae FG |
11067 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(c, o, it)); |
11068 | } | |
11069 | ||
11070 | // ----------------- | |
11071 | // write helpers | |
11072 | ||
11fdf7f2 | 11073 | uint64_t BlueStore::_get_ondisk_reserved() const { |
f67539c2 | 11074 | ceph_assert(min_alloc_size); |
11fdf7f2 TL |
11075 | return round_up_to( |
11076 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
11077 | } | |
11078 | ||
7c673cae FG |
11079 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
11080 | { | |
11081 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
11082 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
11083 | << dendl; | |
11fdf7f2 | 11084 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
11085 | { |
11086 | bufferlist bl; | |
11fdf7f2 | 11087 | encode(ondisk_format, bl); |
7c673cae FG |
11088 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
11089 | } | |
11090 | { | |
11091 | bufferlist bl; | |
11fdf7f2 | 11092 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
11093 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
11094 | } | |
11095 | } | |
11096 | ||
11097 | int BlueStore::_open_super_meta() | |
11098 | { | |
11099 | // nid | |
11100 | { | |
11101 | nid_max = 0; | |
11102 | bufferlist bl; | |
11103 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 11104 | auto p = bl.cbegin(); |
7c673cae FG |
11105 | try { |
11106 | uint64_t v; | |
11fdf7f2 | 11107 | decode(v, p); |
7c673cae | 11108 | nid_max = v; |
f67539c2 | 11109 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11110 | derr << __func__ << " unable to read nid_max" << dendl; |
11111 | return -EIO; | |
11112 | } | |
f67539c2 | 11113 | dout(1) << __func__ << " old nid_max " << nid_max << dendl; |
7c673cae FG |
11114 | nid_last = nid_max.load(); |
11115 | } | |
11116 | ||
11117 | // blobid | |
11118 | { | |
11119 | blobid_max = 0; | |
11120 | bufferlist bl; | |
11121 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 11122 | auto p = bl.cbegin(); |
7c673cae FG |
11123 | try { |
11124 | uint64_t v; | |
11fdf7f2 | 11125 | decode(v, p); |
7c673cae | 11126 | blobid_max = v; |
f67539c2 | 11127 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11128 | derr << __func__ << " unable to read blobid_max" << dendl; |
11129 | return -EIO; | |
11130 | } | |
f67539c2 | 11131 | dout(1) << __func__ << " old blobid_max " << blobid_max << dendl; |
7c673cae FG |
11132 | blobid_last = blobid_max.load(); |
11133 | } | |
11134 | ||
11135 | // freelist | |
11136 | { | |
11137 | bufferlist bl; | |
11138 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
11139 | if (bl.length()) { | |
11140 | freelist_type = std::string(bl.c_str(), bl.length()); | |
f67539c2 | 11141 | dout(1) << __func__ << " freelist_type " << freelist_type << dendl; |
7c673cae | 11142 | } else { |
11fdf7f2 | 11143 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 11144 | } |
7c673cae FG |
11145 | } |
11146 | ||
11147 | // ondisk format | |
11148 | int32_t compat_ondisk_format = 0; | |
11149 | { | |
11150 | bufferlist bl; | |
11151 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
11152 | if (r < 0) { | |
11153 | // base case: kraken bluestore is v1 and readable by v1 | |
11154 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
11155 | << dendl; | |
11156 | ondisk_format = 1; | |
11157 | compat_ondisk_format = 1; | |
11158 | } else { | |
11fdf7f2 | 11159 | auto p = bl.cbegin(); |
7c673cae | 11160 | try { |
11fdf7f2 | 11161 | decode(ondisk_format, p); |
f67539c2 | 11162 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11163 | derr << __func__ << " unable to read ondisk_format" << dendl; |
11164 | return -EIO; | |
11165 | } | |
11166 | bl.clear(); | |
11167 | { | |
11168 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
11169 | ceph_assert(!r); |
11170 | auto p = bl.cbegin(); | |
7c673cae | 11171 | try { |
11fdf7f2 | 11172 | decode(compat_ondisk_format, p); |
f67539c2 | 11173 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11174 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; |
11175 | return -EIO; | |
11176 | } | |
11177 | } | |
11178 | } | |
f67539c2 | 11179 | dout(1) << __func__ << " ondisk_format " << ondisk_format |
7c673cae FG |
11180 | << " compat_ondisk_format " << compat_ondisk_format |
11181 | << dendl; | |
11182 | } | |
11183 | ||
11184 | if (latest_ondisk_format < compat_ondisk_format) { | |
11185 | derr << __func__ << " compat_ondisk_format is " | |
11186 | << compat_ondisk_format << " but we only understand version " | |
11187 | << latest_ondisk_format << dendl; | |
11188 | return -EPERM; | |
11189 | } | |
7c673cae FG |
11190 | |
11191 | { | |
11192 | bufferlist bl; | |
11193 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 11194 | auto p = bl.cbegin(); |
7c673cae FG |
11195 | try { |
11196 | uint64_t val; | |
11fdf7f2 | 11197 | decode(val, p); |
7c673cae | 11198 | min_alloc_size = val; |
224ce89b | 11199 | min_alloc_size_order = ctz(val); |
11fdf7f2 | 11200 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
f67539c2 | 11201 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
11202 | derr << __func__ << " unable to read min_alloc_size" << dendl; |
11203 | return -EIO; | |
11204 | } | |
f67539c2 | 11205 | dout(1) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size |
7c673cae FG |
11206 | << std::dec << dendl; |
11207 | } | |
9f95a23c TL |
11208 | |
11209 | _set_per_pool_omap(); | |
11210 | ||
224ce89b | 11211 | _open_statfs(); |
7c673cae FG |
11212 | _set_alloc_sizes(); |
11213 | _set_throttle_params(); | |
11214 | ||
11215 | _set_csum(); | |
11216 | _set_compression(); | |
11217 | _set_blob_size(); | |
11218 | ||
11fdf7f2 | 11219 | _validate_bdev(); |
7c673cae FG |
11220 | return 0; |
11221 | } | |
11222 | ||
11223 | int BlueStore::_upgrade_super() | |
11224 | { | |
11225 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
11226 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
11227 | if (ondisk_format < latest_ondisk_format) { |
11228 | ceph_assert(ondisk_format > 0); | |
11229 | ceph_assert(ondisk_format < latest_ondisk_format); | |
11230 | ||
1911f103 | 11231 | KeyValueDB::Transaction t = db->get_transaction(); |
11fdf7f2 TL |
11232 | if (ondisk_format == 1) { |
11233 | // changes: | |
11234 | // - super: added ondisk_format | |
11235 | // - super: added min_readable_ondisk_format | |
11236 | // - super: added min_compat_ondisk_format | |
11237 | // - super: added min_alloc_size | |
11238 | // - super: removed min_min_alloc_size | |
11fdf7f2 TL |
11239 | { |
11240 | bufferlist bl; | |
11241 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
11242 | auto p = bl.cbegin(); | |
11243 | try { | |
11244 | uint64_t val; | |
11245 | decode(val, p); | |
11246 | min_alloc_size = val; | |
f67539c2 | 11247 | } catch (ceph::buffer::error& e) { |
11fdf7f2 TL |
11248 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; |
11249 | return -EIO; | |
11250 | } | |
11251 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
11252 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 11253 | } |
11fdf7f2 | 11254 | ondisk_format = 2; |
7c673cae | 11255 | } |
9f95a23c TL |
11256 | if (ondisk_format == 2) { |
11257 | // changes: | |
f67539c2 TL |
11258 | // - onode has FLAG_PERPOOL_OMAP. Note that we do not know that *all* |
11259 | // oondes are using the per-pool prefix until a repair is run; at that | |
9f95a23c TL |
11260 | // point the per_pool_omap=1 key will be set. |
11261 | // - super: added per_pool_omap key, which indicates that *all* objects | |
11262 | // are using the new prefix and key format | |
11263 | ondisk_format = 3; | |
1911f103 TL |
11264 | } |
11265 | if (ondisk_format == 3) { | |
11266 | // changes: | |
11267 | // - FreelistManager keeps meta within bdev label | |
11268 | int r = _write_out_fm_meta(0); | |
9f95a23c | 11269 | ceph_assert(r == 0); |
1911f103 | 11270 | ondisk_format = 4; |
9f95a23c | 11271 | } |
1911f103 TL |
11272 | // This to be the last operation |
11273 | _prepare_ondisk_format_super(t); | |
11274 | int r = db->submit_transaction_sync(t); | |
11275 | ceph_assert(r == 0); | |
7c673cae | 11276 | } |
7c673cae FG |
11277 | // done |
11278 | dout(1) << __func__ << " done" << dendl; | |
11279 | return 0; | |
11280 | } | |
11281 | ||
11282 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef o) | |
11283 | { | |
224ce89b | 11284 | if (o->onode.nid) { |
11fdf7f2 | 11285 | ceph_assert(o->exists); |
7c673cae | 11286 | return; |
224ce89b | 11287 | } |
7c673cae FG |
11288 | uint64_t nid = ++nid_last; |
11289 | dout(20) << __func__ << " " << nid << dendl; | |
11290 | o->onode.nid = nid; | |
11291 | txc->last_nid = nid; | |
224ce89b | 11292 | o->exists = true; |
7c673cae FG |
11293 | } |
11294 | ||
11295 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
11296 | { | |
11297 | uint64_t bid = ++blobid_last; | |
11298 | dout(20) << __func__ << " " << bid << dendl; | |
11299 | txc->last_blobid = bid; | |
11300 | return bid; | |
11301 | } | |
11302 | ||
11303 | void BlueStore::get_db_statistics(Formatter *f) | |
11304 | { | |
11305 | db->get_statistics(f); | |
11306 | } | |
11307 | ||
11fdf7f2 TL |
11308 | BlueStore::TransContext *BlueStore::_txc_create( |
11309 | Collection *c, OpSequencer *osr, | |
f67539c2 TL |
11310 | list<Context*> *on_commits, |
11311 | TrackedOpRef osd_op) | |
7c673cae | 11312 | { |
11fdf7f2 | 11313 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae | 11314 | txc->t = db->get_transaction(); |
f67539c2 TL |
11315 | |
11316 | #ifdef WITH_BLKIN | |
11317 | if (osd_op && osd_op->pg_trace) { | |
11318 | txc->trace.init("TransContext", &trace_endpoint, | |
11319 | &osd_op->pg_trace); | |
11320 | txc->trace.event("txc create"); | |
11321 | txc->trace.keyval("txc seq", txc->seq); | |
11322 | } | |
11323 | #endif | |
11324 | ||
7c673cae FG |
11325 | osr->queue_new(txc); |
11326 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
11327 | << " seq " << txc->seq << dendl; | |
11328 | return txc; | |
11329 | } | |
11330 | ||
11331 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
11332 | { | |
11fdf7f2 TL |
11333 | // one "io" for the kv commit |
11334 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
11335 | auto cost = throttle_cost_per_io.load(); |
11336 | txc->cost = ios * cost + txc->bytes; | |
9f95a23c | 11337 | txc->ios = ios; |
7c673cae FG |
11338 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" |
11339 | << ios << " ios * " << cost << " + " << txc->bytes | |
11340 | << " bytes)" << dendl; | |
11341 | } | |
11342 | ||
11343 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
11344 | { | |
11345 | if (txc->statfs_delta.is_empty()) | |
11346 | return; | |
11347 | ||
11348 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
11349 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
11350 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
11351 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
11352 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
11353 | ||
11354 | bufferlist bl; | |
11355 | txc->statfs_delta.encode(bl); | |
11fdf7f2 TL |
11356 | if (per_pool_stat_collection) { |
11357 | string key; | |
11358 | get_pool_stat_key(txc->osd_pool_id, &key); | |
11359 | txc->t->merge(PREFIX_STAT, key, bl); | |
11360 | ||
11361 | std::lock_guard l(vstatfs_lock); | |
11362 | auto& stats = osd_pools[txc->osd_pool_id]; | |
11363 | stats += txc->statfs_delta; | |
11364 | ||
11365 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
11366 | ||
11367 | } else { | |
11368 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
7c673cae | 11369 | |
11fdf7f2 TL |
11370 | std::lock_guard l(vstatfs_lock); |
11371 | vstatfs += txc->statfs_delta; | |
11372 | } | |
7c673cae FG |
11373 | txc->statfs_delta.reset(); |
11374 | } | |
11375 | ||
11376 | void BlueStore::_txc_state_proc(TransContext *txc) | |
11377 | { | |
11378 | while (true) { | |
11379 | dout(10) << __func__ << " txc " << txc | |
11380 | << " " << txc->get_state_name() << dendl; | |
f67539c2 | 11381 | switch (txc->get_state()) { |
7c673cae | 11382 | case TransContext::STATE_PREPARE: |
9f95a23c | 11383 | throttle.log_state_latency(*txc, logger, l_bluestore_state_prepare_lat); |
7c673cae | 11384 | if (txc->ioc.has_pending_aios()) { |
f67539c2 TL |
11385 | txc->set_state(TransContext::STATE_AIO_WAIT); |
11386 | #ifdef WITH_BLKIN | |
11387 | if (txc->trace) { | |
11388 | txc->trace.keyval("pending aios", txc->ioc.num_pending.load()); | |
11389 | } | |
11390 | #endif | |
7c673cae FG |
11391 | txc->had_ios = true; |
11392 | _txc_aio_submit(txc); | |
11393 | return; | |
11394 | } | |
11395 | // ** fall-thru ** | |
11396 | ||
11397 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 | 11398 | { |
9f95a23c TL |
11399 | mono_clock::duration lat = throttle.log_state_latency( |
11400 | *txc, logger, l_bluestore_state_aio_wait_lat); | |
11401 | if (ceph::to_seconds<double>(lat) >= cct->_conf->bluestore_log_op_age) { | |
11fdf7f2 TL |
11402 | dout(0) << __func__ << " slow aio_wait, txc = " << txc |
11403 | << ", latency = " << lat | |
11404 | << dendl; | |
11405 | } | |
11406 | } | |
11407 | ||
7c673cae FG |
11408 | _txc_finish_io(txc); // may trigger blocked txc's too |
11409 | return; | |
11410 | ||
11411 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 11412 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
11413 | if (txc->had_ios) { |
11414 | ++txc->osr->txc_with_unstable_io; | |
11415 | } | |
9f95a23c | 11416 | throttle.log_state_latency(*txc, logger, l_bluestore_state_io_done_lat); |
f67539c2 | 11417 | txc->set_state(TransContext::STATE_KV_QUEUED); |
7c673cae FG |
11418 | if (cct->_conf->bluestore_sync_submit_transaction) { |
11419 | if (txc->last_nid >= nid_max || | |
11420 | txc->last_blobid >= blobid_max) { | |
11421 | dout(20) << __func__ | |
11422 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
11423 | << dendl; | |
11424 | } else if (txc->osr->kv_committing_serially) { | |
11425 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
11426 | << dendl; | |
11427 | // note: this is starvation-prone. once we have a txc in a busy | |
11428 | // sequencer that is committing serially it is possible to keep | |
11429 | // submitting new transactions fast enough that we get stuck doing | |
11430 | // so. the alternative is to block here... fixme? | |
11431 | } else if (txc->osr->txc_with_unstable_io) { | |
11432 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
11433 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
11434 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
11435 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
11436 | == 0) { | |
11437 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
11438 | << dendl; | |
11439 | } else { | |
9f95a23c | 11440 | _txc_apply_kv(txc, true); |
7c673cae FG |
11441 | } |
11442 | } | |
11443 | { | |
11fdf7f2 | 11444 | std::lock_guard l(kv_lock); |
7c673cae | 11445 | kv_queue.push_back(txc); |
9f95a23c TL |
11446 | if (!kv_sync_in_progress) { |
11447 | kv_sync_in_progress = true; | |
11448 | kv_cond.notify_one(); | |
11449 | } | |
f67539c2 | 11450 | if (txc->get_state() != TransContext::STATE_KV_SUBMITTED) { |
7c673cae FG |
11451 | kv_queue_unsubmitted.push_back(txc); |
11452 | ++txc->osr->kv_committing_serially; | |
11453 | } | |
31f18b77 FG |
11454 | if (txc->had_ios) |
11455 | kv_ios++; | |
11456 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
11457 | } |
11458 | return; | |
11459 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
11460 | _txc_committed_kv(txc); |
11461 | // ** fall-thru ** | |
11462 | ||
11463 | case TransContext::STATE_KV_DONE: | |
9f95a23c | 11464 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_done_lat); |
7c673cae | 11465 | if (txc->deferred_txn) { |
f67539c2 | 11466 | txc->set_state(TransContext::STATE_DEFERRED_QUEUED); |
7c673cae FG |
11467 | _deferred_queue(txc); |
11468 | return; | |
11469 | } | |
f67539c2 | 11470 | txc->set_state(TransContext::STATE_FINISHING); |
7c673cae FG |
11471 | break; |
11472 | ||
11473 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9f95a23c | 11474 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_cleanup_lat); |
f67539c2 | 11475 | txc->set_state(TransContext::STATE_FINISHING); |
7c673cae FG |
11476 | // ** fall-thru ** |
11477 | ||
11478 | case TransContext::STATE_FINISHING: | |
9f95a23c | 11479 | throttle.log_state_latency(*txc, logger, l_bluestore_state_finishing_lat); |
7c673cae FG |
11480 | _txc_finish(txc); |
11481 | return; | |
11482 | ||
11483 | default: | |
11484 | derr << __func__ << " unexpected txc " << txc | |
11485 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 11486 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
11487 | return; |
11488 | } | |
11489 | } | |
11490 | } | |
11491 | ||
11492 | void BlueStore::_txc_finish_io(TransContext *txc) | |
11493 | { | |
11494 | dout(20) << __func__ << " " << txc << dendl; | |
11495 | ||
11496 | /* | |
11497 | * we need to preserve the order of kv transactions, | |
11498 | * even though aio will complete in any order. | |
11499 | */ | |
11500 | ||
11501 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 11502 | std::lock_guard l(osr->qlock); |
f67539c2 | 11503 | txc->set_state(TransContext::STATE_IO_DONE); |
11fdf7f2 | 11504 | txc->ioc.release_running_aios(); |
7c673cae FG |
11505 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
11506 | while (p != osr->q.begin()) { | |
11507 | --p; | |
f67539c2 | 11508 | if (p->get_state() < TransContext::STATE_IO_DONE) { |
7c673cae FG |
11509 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " |
11510 | << p->get_state_name() << dendl; | |
11511 | return; | |
11512 | } | |
f67539c2 | 11513 | if (p->get_state() > TransContext::STATE_IO_DONE) { |
7c673cae FG |
11514 | ++p; |
11515 | break; | |
11516 | } | |
11517 | } | |
11518 | do { | |
11519 | _txc_state_proc(&*p++); | |
11520 | } while (p != osr->q.end() && | |
f67539c2 | 11521 | p->get_state() == TransContext::STATE_IO_DONE); |
7c673cae | 11522 | |
11fdf7f2 | 11523 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
11524 | osr->qcond.notify_all(); |
11525 | } | |
11526 | } | |
11527 | ||
11528 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
11529 | { | |
11530 | dout(20) << __func__ << " txc " << txc | |
11531 | << " onodes " << txc->onodes | |
11532 | << " shared_blobs " << txc->shared_blobs | |
11533 | << dendl; | |
11534 | ||
11535 | // finalize onodes | |
11536 | for (auto o : txc->onodes) { | |
11fdf7f2 | 11537 | _record_onode(o, t); |
7c673cae FG |
11538 | o->flushing_count++; |
11539 | } | |
11540 | ||
11541 | // objects we modified but didn't affect the onode | |
11542 | auto p = txc->modified_objects.begin(); | |
11543 | while (p != txc->modified_objects.end()) { | |
11544 | if (txc->onodes.count(*p) == 0) { | |
11545 | (*p)->flushing_count++; | |
11546 | ++p; | |
11547 | } else { | |
11548 | // remove dups with onodes list to avoid problems in _txc_finish | |
11549 | p = txc->modified_objects.erase(p); | |
11550 | } | |
11551 | } | |
11552 | ||
11553 | // finalize shared_blobs | |
11554 | for (auto sb : txc->shared_blobs) { | |
11555 | string key; | |
11556 | auto sbid = sb->get_sbid(); | |
11557 | get_shared_blob_key(sbid, &key); | |
11558 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
11559 | dout(20) << __func__ << " shared_blob 0x" |
11560 | << std::hex << sbid << std::dec | |
7c673cae FG |
11561 | << " is empty" << dendl; |
11562 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
11563 | } else { | |
11564 | bufferlist bl; | |
11fdf7f2 TL |
11565 | encode(*(sb->persistent), bl); |
11566 | dout(20) << __func__ << " shared_blob 0x" | |
11567 | << std::hex << sbid << std::dec | |
31f18b77 | 11568 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
11569 | t->set(PREFIX_SHARED_BLOB, key, bl); |
11570 | } | |
11571 | } | |
11572 | } | |
11573 | ||
11574 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
11575 | PerfCounters &logger) | |
11576 | { | |
11fdf7f2 TL |
11577 | os_commit_latency_ns.consume_next( |
11578 | logger.get_tavg_ns( | |
7c673cae | 11579 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
11580 | os_apply_latency_ns.consume_next( |
11581 | logger.get_tavg_ns( | |
7c673cae FG |
11582 | l_bluestore_commit_lat)); |
11583 | } | |
11584 | ||
f67539c2 TL |
11585 | // For every object we maintain <zone_num+oid, offset> tuple in the key-value |
11586 | // store. When a new object written to a zone, we insert the corresponding | |
11587 | // tuple to the database. When an object is truncated, we remove the | |
11588 | // corresponding tuple. When an object is overwritten, we remove the old tuple | |
11589 | // and insert a new tuple corresponding to the new location of the object. The | |
11590 | // cleaner can now identify live objects within the zone <zone_num> by | |
11591 | // enumerating all the keys starting with <zone_num> prefix. | |
11592 | void BlueStore::_zoned_update_cleaning_metadata(TransContext *txc) { | |
11593 | for (const auto &[o, offsets] : txc->zoned_onode_to_offset_map) { | |
11594 | std::string key; | |
11595 | get_object_key(cct, o->oid, &key); | |
11596 | for (auto offset : offsets) { | |
11597 | if (offset > 0) { | |
11598 | bufferlist offset_bl; | |
11599 | encode(offset, offset_bl); | |
11600 | txc->t->set(_zoned_get_prefix(offset), key, offset_bl); | |
11601 | } else { | |
11602 | txc->t->rmkey(_zoned_get_prefix(-offset), key); | |
11603 | } | |
11604 | } | |
11605 | } | |
11606 | } | |
11607 | ||
11608 | std::string BlueStore::_zoned_get_prefix(uint64_t offset) { | |
11609 | uint64_t zone_num = offset / bdev->get_zone_size(); | |
11610 | std::string zone_key; | |
11611 | _key_encode_u64(zone_num, &zone_key); | |
11612 | return PREFIX_ZONED_CL_INFO + zone_key; | |
11613 | } | |
11614 | ||
11615 | // For now, to avoid interface changes we piggyback zone_size (in MiB) and the | |
11616 | // first sequential zone number onto min_alloc_size and pass it to functions | |
11617 | // Allocator::create and FreelistManager::create. | |
11618 | uint64_t BlueStore::_zoned_piggyback_device_parameters_onto(uint64_t min_alloc_size) { | |
11619 | uint64_t zone_size = bdev->get_zone_size(); | |
11620 | uint64_t zone_size_mb = zone_size / (1024 * 1024); | |
11621 | uint64_t first_seq_zone = bdev->get_conventional_region_size() / zone_size; | |
11622 | min_alloc_size |= (zone_size_mb << 32); | |
11623 | min_alloc_size |= (first_seq_zone << 48); | |
11624 | return min_alloc_size; | |
11625 | } | |
11626 | ||
11627 | int BlueStore::_zoned_check_config_settings() { | |
11628 | if (cct->_conf->bluestore_allocator != "zoned") { | |
11629 | dout(1) << __func__ << " The drive is HM-SMR but " | |
11630 | << cct->_conf->bluestore_allocator << " allocator is specified. " | |
11631 | << "Only zoned allocator can be used with HM-SMR drive." << dendl; | |
11632 | return -EINVAL; | |
11633 | } | |
11634 | ||
11635 | // At least for now we want to use large min_alloc_size with HM-SMR drives. | |
11636 | // Populating used_blocks bitset on a debug build of ceph-osd takes about 5 | |
11637 | // minutes with a 14 TB HM-SMR drive and 4 KiB min_alloc_size. | |
11638 | if (min_alloc_size < 64 * 1024) { | |
11639 | dout(1) << __func__ << " The drive is HM-SMR but min_alloc_size is " | |
11640 | << min_alloc_size << ". " | |
11641 | << "Please set to at least 64 KiB." << dendl; | |
11642 | return -EINVAL; | |
11643 | } | |
11644 | ||
11645 | // We don't want to defer writes with HM-SMR because it violates sequential | |
11646 | // write requirement. | |
11647 | if (prefer_deferred_size) { | |
11648 | dout(1) << __func__ << " The drive is HM-SMR but prefer_deferred_size is " | |
11649 | << prefer_deferred_size << ". " | |
11650 | << "Please set to 0." << dendl; | |
11651 | return -EINVAL; | |
11652 | } | |
11653 | return 0; | |
11654 | } | |
11655 | ||
7c673cae FG |
11656 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) |
11657 | { | |
11658 | dout(20) << __func__ << " txc " << txc << std::hex | |
11659 | << " allocated 0x" << txc->allocated | |
11660 | << " released 0x" << txc->released | |
11661 | << std::dec << dendl; | |
11662 | ||
11663 | // We have to handle the case where we allocate *and* deallocate the | |
11664 | // same region in this transaction. The freelist doesn't like that. | |
11665 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
11666 | // debug check. But that's important.) | |
11667 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
11668 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
11669 | interval_set<uint64_t> *preleased = &txc->released; | |
11670 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
11671 | interval_set<uint64_t> overlap; | |
11672 | overlap.intersection_of(txc->allocated, txc->released); | |
11673 | if (!overlap.empty()) { | |
11674 | tmp_allocated = txc->allocated; | |
11675 | tmp_allocated.subtract(overlap); | |
11676 | tmp_released = txc->released; | |
11677 | tmp_released.subtract(overlap); | |
11678 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
11679 | << ", new allocated 0x" << tmp_allocated | |
11680 | << " released 0x" << tmp_released << std::dec | |
11681 | << dendl; | |
11682 | pallocated = &tmp_allocated; | |
11683 | preleased = &tmp_released; | |
11684 | } | |
11685 | } | |
11686 | ||
11687 | // update freelist with non-overlap sets | |
11688 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
11689 | p != pallocated->end(); | |
11690 | ++p) { | |
11691 | fm->allocate(p.get_start(), p.get_len(), t); | |
11692 | } | |
11693 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
11694 | p != preleased->end(); | |
11695 | ++p) { | |
11696 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
11697 | << "~" << p.get_len() << std::dec << dendl; | |
11698 | fm->release(p.get_start(), p.get_len(), t); | |
11699 | } | |
11700 | ||
f67539c2 TL |
11701 | if (bdev->is_smr()) { |
11702 | _zoned_update_cleaning_metadata(txc); | |
11703 | } | |
11704 | ||
7c673cae FG |
11705 | _txc_update_store_statfs(txc); |
11706 | } | |
11707 | ||
9f95a23c | 11708 | void BlueStore::_txc_apply_kv(TransContext *txc, bool sync_submit_transaction) |
7c673cae | 11709 | { |
f67539c2 | 11710 | ceph_assert(txc->get_state() == TransContext::STATE_KV_QUEUED); |
9f95a23c TL |
11711 | { |
11712 | #if defined(WITH_LTTNG) | |
11713 | auto start = mono_clock::now(); | |
11714 | #endif | |
11715 | ||
f67539c2 TL |
11716 | #ifdef WITH_BLKIN |
11717 | if (txc->trace) { | |
11718 | txc->trace.event("db async submit"); | |
11719 | } | |
11720 | #endif | |
11721 | ||
9f95a23c TL |
11722 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); |
11723 | ceph_assert(r == 0); | |
f67539c2 | 11724 | txc->set_state(TransContext::STATE_KV_SUBMITTED); |
9f95a23c TL |
11725 | if (txc->osr->kv_submitted_waiters) { |
11726 | std::lock_guard l(txc->osr->qlock); | |
11727 | txc->osr->qcond.notify_all(); | |
11728 | } | |
11729 | ||
11730 | #if defined(WITH_LTTNG) | |
11731 | if (txc->tracing) { | |
11732 | tracepoint( | |
11733 | bluestore, | |
11734 | transaction_kv_submit_latency, | |
11735 | txc->osr->get_sequencer_id(), | |
11736 | txc->seq, | |
11737 | sync_submit_transaction, | |
11738 | ceph::to_seconds<double>(mono_clock::now() - start)); | |
11739 | } | |
11740 | #endif | |
11741 | } | |
11742 | ||
7c673cae FG |
11743 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { |
11744 | for (auto& o : *ls) { | |
11745 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
11746 | << dendl; | |
9f95a23c | 11747 | if (--o->flushing_count == 0 && o->waiting_count.load()) { |
11fdf7f2 | 11748 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
11749 | o->flush_cond.notify_all(); |
11750 | } | |
11751 | } | |
11752 | } | |
11753 | } | |
11754 | ||
11755 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
11756 | { | |
11757 | dout(20) << __func__ << " txc " << txc << dendl; | |
9f95a23c | 11758 | throttle.complete_kv(*txc); |
1adf2230 | 11759 | { |
11fdf7f2 | 11760 | std::lock_guard l(txc->osr->qlock); |
f67539c2 | 11761 | txc->set_state(TransContext::STATE_KV_DONE); |
11fdf7f2 TL |
11762 | if (txc->ch->commit_queue) { |
11763 | txc->ch->commit_queue->queue(txc->oncommits); | |
11764 | } else { | |
11765 | finisher.queue(txc->oncommits); | |
1adf2230 | 11766 | } |
7c673cae | 11767 | } |
9f95a23c | 11768 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_committing_lat); |
494da23a TL |
11769 | log_latency_fn( |
11770 | __func__, | |
11771 | l_bluestore_commit_lat, | |
9f95a23c | 11772 | mono_clock::now() - txc->start, |
494da23a TL |
11773 | cct->_conf->bluestore_log_op_age, |
11774 | [&](auto lat) { | |
11775 | return ", txc = " + stringify(txc); | |
11776 | } | |
11fdf7f2 | 11777 | ); |
7c673cae FG |
11778 | } |
11779 | ||
11780 | void BlueStore::_txc_finish(TransContext *txc) | |
11781 | { | |
11782 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
f67539c2 | 11783 | ceph_assert(txc->get_state() == TransContext::STATE_FINISHING); |
7c673cae FG |
11784 | |
11785 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 11786 | sb->finish_write(txc->seq); |
7c673cae FG |
11787 | } |
11788 | txc->shared_blobs_written.clear(); | |
11789 | ||
11790 | while (!txc->removed_collections.empty()) { | |
11791 | _queue_reap_collection(txc->removed_collections.front()); | |
11792 | txc->removed_collections.pop_front(); | |
11793 | } | |
11794 | ||
11795 | OpSequencerRef osr = txc->osr; | |
7c673cae | 11796 | bool empty = false; |
31f18b77 | 11797 | bool submit_deferred = false; |
7c673cae FG |
11798 | OpSequencer::q_list_t releasing_txc; |
11799 | { | |
11fdf7f2 | 11800 | std::lock_guard l(osr->qlock); |
f67539c2 | 11801 | txc->set_state(TransContext::STATE_DONE); |
7c673cae FG |
11802 | bool notify = false; |
11803 | while (!osr->q.empty()) { | |
11804 | TransContext *txc = &osr->q.front(); | |
11805 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
11806 | << dendl; | |
f67539c2 TL |
11807 | if (txc->get_state() != TransContext::STATE_DONE) { |
11808 | if (txc->get_state() == TransContext::STATE_PREPARE && | |
7c673cae FG |
11809 | deferred_aggressive) { |
11810 | // for _osr_drain_preceding() | |
11811 | notify = true; | |
11812 | } | |
f67539c2 | 11813 | if (txc->get_state() == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 11814 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
11815 | submit_deferred = true; |
11816 | } | |
7c673cae FG |
11817 | break; |
11818 | } | |
11819 | ||
7c673cae FG |
11820 | osr->q.pop_front(); |
11821 | releasing_txc.push_back(*txc); | |
7c673cae | 11822 | } |
9f95a23c | 11823 | |
7c673cae FG |
11824 | if (osr->q.empty()) { |
11825 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
11826 | empty = true; | |
11827 | } | |
9f95a23c TL |
11828 | |
11829 | // only drain()/drain_preceding() need wakeup, | |
11830 | // other cases use kv_submitted_waiters | |
11831 | if (notify || empty) { | |
11832 | osr->qcond.notify_all(); | |
11833 | } | |
7c673cae | 11834 | } |
9f95a23c | 11835 | |
7c673cae FG |
11836 | while (!releasing_txc.empty()) { |
11837 | // release to allocator only after all preceding txc's have also | |
11838 | // finished any deferred writes that potentially land in these | |
11839 | // blocks | |
11840 | auto txc = &releasing_txc.front(); | |
11841 | _txc_release_alloc(txc); | |
11842 | releasing_txc.pop_front(); | |
9f95a23c TL |
11843 | throttle.log_state_latency(*txc, logger, l_bluestore_state_done_lat); |
11844 | throttle.complete(*txc); | |
7c673cae FG |
11845 | delete txc; |
11846 | } | |
11847 | ||
31f18b77 FG |
11848 | if (submit_deferred) { |
11849 | // we're pinning memory; flush! we could be more fine-grained here but | |
11850 | // i'm not sure it's worth the bother. | |
11851 | deferred_try_submit(); | |
7c673cae FG |
11852 | } |
11853 | ||
7c673cae | 11854 | if (empty && osr->zombie) { |
11fdf7f2 TL |
11855 | std::lock_guard l(zombie_osr_lock); |
11856 | if (zombie_osr_set.erase(osr->cid)) { | |
11857 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11858 | } else { | |
11859 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
11860 | << dendl; | |
11861 | } | |
7c673cae | 11862 | } |
9f95a23c | 11863 | } |
7c673cae FG |
11864 | |
11865 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
11866 | { | |
a8e16298 | 11867 | // it's expected we're called with lazy_release_lock already taken! |
11fdf7f2 TL |
11868 | if (likely(!cct->_conf->bluestore_debug_no_reuse_blocks)) { |
11869 | int r = 0; | |
11870 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
11871 | r = bdev->queue_discard(txc->released); | |
11872 | if (r == 0) { | |
11873 | dout(10) << __func__ << "(queued) " << txc << " " << std::hex | |
11874 | << txc->released << std::dec << dendl; | |
11875 | goto out; | |
11876 | } | |
11877 | } else if (cct->_conf->bdev_enable_discard) { | |
11878 | for (auto p = txc->released.begin(); p != txc->released.end(); ++p) { | |
11879 | bdev->discard(p.get_start(), p.get_len()); | |
11880 | } | |
11881 | } | |
11882 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
94b18763 | 11883 | << txc->released << std::dec << dendl; |
f67539c2 | 11884 | shared_alloc.a->release(txc->released); |
7c673cae FG |
11885 | } |
11886 | ||
11fdf7f2 | 11887 | out: |
7c673cae FG |
11888 | txc->allocated.clear(); |
11889 | txc->released.clear(); | |
11890 | } | |
11891 | ||
11fdf7f2 TL |
11892 | void BlueStore::_osr_attach(Collection *c) |
11893 | { | |
11894 | // note: caller has RWLock on coll_map | |
11895 | auto q = coll_map.find(c->cid); | |
11896 | if (q != coll_map.end()) { | |
11897 | c->osr = q->second->osr; | |
11898 | ldout(cct, 10) << __func__ << " " << c->cid | |
11899 | << " reusing osr " << c->osr << " from existing coll " | |
11900 | << q->second << dendl; | |
11901 | } else { | |
11902 | std::lock_guard l(zombie_osr_lock); | |
11903 | auto p = zombie_osr_set.find(c->cid); | |
11904 | if (p == zombie_osr_set.end()) { | |
9f95a23c | 11905 | c->osr = ceph::make_ref<OpSequencer>(this, next_sequencer_id++, c->cid); |
11fdf7f2 TL |
11906 | ldout(cct, 10) << __func__ << " " << c->cid |
11907 | << " fresh osr " << c->osr << dendl; | |
11908 | } else { | |
11909 | c->osr = p->second; | |
11910 | zombie_osr_set.erase(p); | |
11911 | ldout(cct, 10) << __func__ << " " << c->cid | |
11912 | << " resurrecting zombie osr " << c->osr << dendl; | |
11913 | c->osr->zombie = false; | |
11914 | } | |
11915 | } | |
11916 | } | |
11917 | ||
11918 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
11919 | { | |
11920 | std::lock_guard l(zombie_osr_lock); | |
11921 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
11922 | osr->zombie = true; | |
11923 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
11924 | // this is either a new insertion or the same osr is already there | |
11925 | ceph_assert(i.second || i.first->second == osr); | |
11926 | } | |
11927 | ||
7c673cae FG |
11928 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
11929 | { | |
11930 | OpSequencer *osr = txc->osr.get(); | |
11931 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
11932 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11933 | { | |
11934 | // submit anything pending | |
f67539c2 | 11935 | osr->deferred_lock.lock(); |
11fdf7f2 | 11936 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
11937 | _deferred_submit_unlock(osr); |
11938 | } else { | |
f67539c2 | 11939 | osr->deferred_lock.unlock(); |
7c673cae FG |
11940 | } |
11941 | } | |
11942 | { | |
11943 | // wake up any previously finished deferred events | |
11fdf7f2 | 11944 | std::lock_guard l(kv_lock); |
9f95a23c TL |
11945 | if (!kv_sync_in_progress) { |
11946 | kv_sync_in_progress = true; | |
11947 | kv_cond.notify_one(); | |
11948 | } | |
7c673cae FG |
11949 | } |
11950 | osr->drain_preceding(txc); | |
11951 | --deferred_aggressive; | |
11952 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11953 | } | |
11954 | ||
11fdf7f2 TL |
11955 | void BlueStore::_osr_drain(OpSequencer *osr) |
11956 | { | |
11957 | dout(10) << __func__ << " " << osr << dendl; | |
11958 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11959 | { | |
11960 | // submit anything pending | |
f67539c2 | 11961 | osr->deferred_lock.lock(); |
11fdf7f2 TL |
11962 | if (osr->deferred_pending && !osr->deferred_running) { |
11963 | _deferred_submit_unlock(osr); | |
11964 | } else { | |
f67539c2 | 11965 | osr->deferred_lock.unlock(); |
11fdf7f2 TL |
11966 | } |
11967 | } | |
11968 | { | |
11969 | // wake up any previously finished deferred events | |
11970 | std::lock_guard l(kv_lock); | |
9f95a23c TL |
11971 | if (!kv_sync_in_progress) { |
11972 | kv_sync_in_progress = true; | |
11973 | kv_cond.notify_one(); | |
11974 | } | |
11fdf7f2 TL |
11975 | } |
11976 | osr->drain(); | |
11977 | --deferred_aggressive; | |
11978 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11979 | } | |
11980 | ||
7c673cae FG |
11981 | void BlueStore::_osr_drain_all() |
11982 | { | |
11983 | dout(10) << __func__ << dendl; | |
11984 | ||
11985 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
11986 | vector<OpSequencerRef> zombies; |
11987 | { | |
9f95a23c | 11988 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
11989 | for (auto& i : coll_map) { |
11990 | s.insert(i.second->osr); | |
11991 | } | |
11992 | } | |
7c673cae | 11993 | { |
11fdf7f2 TL |
11994 | std::lock_guard l(zombie_osr_lock); |
11995 | for (auto& i : zombie_osr_set) { | |
11996 | s.insert(i.second); | |
11997 | zombies.push_back(i.second); | |
11998 | } | |
7c673cae FG |
11999 | } |
12000 | dout(20) << __func__ << " osr_set " << s << dendl; | |
12001 | ||
12002 | ++deferred_aggressive; | |
12003 | { | |
12004 | // submit anything pending | |
224ce89b | 12005 | deferred_try_submit(); |
7c673cae FG |
12006 | } |
12007 | { | |
12008 | // wake up any previously finished deferred events | |
11fdf7f2 | 12009 | std::lock_guard l(kv_lock); |
7c673cae FG |
12010 | kv_cond.notify_one(); |
12011 | } | |
31f18b77 | 12012 | { |
11fdf7f2 | 12013 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
12014 | kv_finalize_cond.notify_one(); |
12015 | } | |
7c673cae FG |
12016 | for (auto osr : s) { |
12017 | dout(20) << __func__ << " drain " << osr << dendl; | |
12018 | osr->drain(); | |
12019 | } | |
12020 | --deferred_aggressive; | |
12021 | ||
7c673cae | 12022 | { |
11fdf7f2 TL |
12023 | std::lock_guard l(zombie_osr_lock); |
12024 | for (auto& osr : zombies) { | |
12025 | if (zombie_osr_set.erase(osr->cid)) { | |
12026 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
12027 | ceph_assert(osr->q.empty()); | |
12028 | } else if (osr->zombie) { | |
12029 | dout(10) << __func__ << " empty zombie osr " << osr | |
12030 | << " already reaped" << dendl; | |
12031 | ceph_assert(osr->q.empty()); | |
12032 | } else { | |
12033 | dout(10) << __func__ << " empty zombie osr " << osr | |
12034 | << " resurrected" << dendl; | |
12035 | } | |
7c673cae FG |
12036 | } |
12037 | } | |
11fdf7f2 TL |
12038 | |
12039 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
12040 | } |
12041 | ||
11fdf7f2 | 12042 | |
31f18b77 FG |
12043 | void BlueStore::_kv_start() |
12044 | { | |
12045 | dout(10) << __func__ << dendl; | |
12046 | ||
11fdf7f2 | 12047 | finisher.start(); |
31f18b77 FG |
12048 | kv_sync_thread.create("bstore_kv_sync"); |
12049 | kv_finalize_thread.create("bstore_kv_final"); | |
12050 | } | |
12051 | ||
12052 | void BlueStore::_kv_stop() | |
12053 | { | |
12054 | dout(10) << __func__ << dendl; | |
12055 | { | |
9f95a23c | 12056 | std::unique_lock l{kv_lock}; |
31f18b77 FG |
12057 | while (!kv_sync_started) { |
12058 | kv_cond.wait(l); | |
12059 | } | |
12060 | kv_stop = true; | |
12061 | kv_cond.notify_all(); | |
12062 | } | |
12063 | { | |
9f95a23c | 12064 | std::unique_lock l{kv_finalize_lock}; |
31f18b77 FG |
12065 | while (!kv_finalize_started) { |
12066 | kv_finalize_cond.wait(l); | |
12067 | } | |
12068 | kv_finalize_stop = true; | |
12069 | kv_finalize_cond.notify_all(); | |
12070 | } | |
12071 | kv_sync_thread.join(); | |
12072 | kv_finalize_thread.join(); | |
11fdf7f2 | 12073 | ceph_assert(removed_collections.empty()); |
31f18b77 | 12074 | { |
11fdf7f2 | 12075 | std::lock_guard l(kv_lock); |
31f18b77 FG |
12076 | kv_stop = false; |
12077 | } | |
12078 | { | |
11fdf7f2 | 12079 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
12080 | kv_finalize_stop = false; |
12081 | } | |
12082 | dout(10) << __func__ << " stopping finishers" << dendl; | |
11fdf7f2 TL |
12083 | finisher.wait_for_empty(); |
12084 | finisher.stop(); | |
31f18b77 FG |
12085 | dout(10) << __func__ << " stopped" << dendl; |
12086 | } | |
12087 | ||
7c673cae FG |
12088 | void BlueStore::_kv_sync_thread() |
12089 | { | |
12090 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 | 12091 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
9f95a23c | 12092 | std::unique_lock l{kv_lock}; |
11fdf7f2 | 12093 | ceph_assert(!kv_sync_started); |
31f18b77 FG |
12094 | kv_sync_started = true; |
12095 | kv_cond.notify_all(); | |
adb31ebb TL |
12096 | |
12097 | auto t0 = mono_clock::now(); | |
12098 | timespan twait = ceph::make_timespan(0); | |
12099 | size_t kv_submitted = 0; | |
12100 | ||
7c673cae | 12101 | while (true) { |
adb31ebb TL |
12102 | auto period = cct->_conf->bluestore_kv_sync_util_logging_s; |
12103 | auto observation_period = | |
12104 | ceph::make_timespan(period); | |
12105 | auto elapsed = mono_clock::now() - t0; | |
12106 | if (period && elapsed >= observation_period) { | |
12107 | dout(5) << __func__ << " utilization: idle " | |
12108 | << twait << " of " << elapsed | |
12109 | << ", submitted: " << kv_submitted | |
12110 | <<dendl; | |
12111 | t0 = mono_clock::now(); | |
12112 | twait = ceph::make_timespan(0); | |
12113 | kv_submitted = 0; | |
12114 | } | |
11fdf7f2 | 12115 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
12116 | if (kv_queue.empty() && |
12117 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 12118 | !deferred_aggressive)) { |
7c673cae FG |
12119 | if (kv_stop) |
12120 | break; | |
12121 | dout(20) << __func__ << " sleep" << dendl; | |
adb31ebb | 12122 | auto t = mono_clock::now(); |
9f95a23c | 12123 | kv_sync_in_progress = false; |
11fdf7f2 | 12124 | kv_cond.wait(l); |
adb31ebb TL |
12125 | twait += mono_clock::now() - t; |
12126 | ||
7c673cae FG |
12127 | dout(20) << __func__ << " wake" << dendl; |
12128 | } else { | |
12129 | deque<TransContext*> kv_submitting; | |
12130 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
12131 | uint64_t aios = 0, costs = 0; |
12132 | ||
7c673cae FG |
12133 | dout(20) << __func__ << " committing " << kv_queue.size() |
12134 | << " submitting " << kv_queue_unsubmitted.size() | |
12135 | << " deferred done " << deferred_done_queue.size() | |
12136 | << " stable " << deferred_stable_queue.size() | |
12137 | << dendl; | |
12138 | kv_committing.swap(kv_queue); | |
12139 | kv_submitting.swap(kv_queue_unsubmitted); | |
12140 | deferred_done.swap(deferred_done_queue); | |
12141 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
12142 | aios = kv_ios; |
12143 | costs = kv_throttle_costs; | |
12144 | kv_ios = 0; | |
12145 | kv_throttle_costs = 0; | |
7c673cae FG |
12146 | l.unlock(); |
12147 | ||
12148 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
12149 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
12150 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
12151 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12152 | ||
11fdf7f2 TL |
12153 | auto start = mono_clock::now(); |
12154 | ||
7c673cae FG |
12155 | bool force_flush = false; |
12156 | // if bluefs is sharing the same device as data (only), then we | |
12157 | // can rely on the bluefs commit to flush the device and make | |
12158 | // deferred aios stable. that means that if we do have done deferred | |
12159 | // txcs AND we are not on a single device, we need to force a flush. | |
9f95a23c | 12160 | if (bluefs && bluefs_layout.single_shared_device()) { |
31f18b77 | 12161 | if (aios) { |
7c673cae | 12162 | force_flush = true; |
11fdf7f2 | 12163 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
12164 | force_flush = true; // there's nothing else to commit! |
12165 | } else if (deferred_aggressive) { | |
12166 | force_flush = true; | |
12167 | } | |
11fdf7f2 TL |
12168 | } else { |
12169 | if (aios || !deferred_done.empty()) { | |
12170 | force_flush = true; | |
12171 | } else { | |
12172 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
12173 | } | |
12174 | } | |
7c673cae FG |
12175 | |
12176 | if (force_flush) { | |
31f18b77 | 12177 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
12178 | << " force_flush=" << (int)force_flush |
12179 | << ", flushing, deferred done->stable" << dendl; | |
12180 | // flush/barrier on block device | |
12181 | bdev->flush(); | |
12182 | ||
12183 | // if we flush then deferred done are now deferred stable | |
12184 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
12185 | deferred_done.end()); | |
12186 | deferred_done.clear(); | |
12187 | } | |
11fdf7f2 | 12188 | auto after_flush = mono_clock::now(); |
7c673cae FG |
12189 | |
12190 | // we will use one final transaction to force a sync | |
12191 | KeyValueDB::Transaction synct = db->get_transaction(); | |
12192 | ||
12193 | // increase {nid,blobid}_max? note that this covers both the | |
12194 | // case where we are approaching the max and the case we passed | |
12195 | // it. in either case, we increase the max in the earlier txn | |
12196 | // we submit. | |
12197 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
12198 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
12199 | KeyValueDB::Transaction t = | |
12200 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
12201 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
12202 | bufferlist bl; | |
11fdf7f2 | 12203 | encode(new_nid_max, bl); |
7c673cae FG |
12204 | t->set(PREFIX_SUPER, "nid_max", bl); |
12205 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
12206 | } | |
12207 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
12208 | KeyValueDB::Transaction t = | |
12209 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
12210 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
12211 | bufferlist bl; | |
11fdf7f2 | 12212 | encode(new_blobid_max, bl); |
7c673cae FG |
12213 | t->set(PREFIX_SUPER, "blobid_max", bl); |
12214 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
12215 | } | |
c07f9fc5 FG |
12216 | |
12217 | for (auto txc : kv_committing) { | |
9f95a23c | 12218 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_queued_lat); |
f67539c2 | 12219 | if (txc->get_state() == TransContext::STATE_KV_QUEUED) { |
adb31ebb | 12220 | ++kv_submitted; |
9f95a23c | 12221 | _txc_apply_kv(txc, false); |
c07f9fc5 | 12222 | --txc->osr->kv_committing_serially; |
c07f9fc5 | 12223 | } else { |
f67539c2 | 12224 | ceph_assert(txc->get_state() == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12225 | } |
7c673cae FG |
12226 | if (txc->had_ios) { |
12227 | --txc->osr->txc_with_unstable_io; | |
12228 | } | |
7c673cae FG |
12229 | } |
12230 | ||
31f18b77 FG |
12231 | // release throttle *before* we commit. this allows new ops |
12232 | // to be prepared and enter pipeline while we are waiting on | |
12233 | // the kv commit sync/flush. then hopefully on the next | |
12234 | // iteration there will already be ops awake. otherwise, we | |
12235 | // end up going to sleep, and then wake up when the very first | |
12236 | // transaction is ready for commit. | |
9f95a23c | 12237 | throttle.release_kv_throttle(costs); |
31f18b77 | 12238 | |
7c673cae FG |
12239 | // cleanup sync deferred keys |
12240 | for (auto b : deferred_stable) { | |
12241 | for (auto& txc : b->txcs) { | |
12242 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 12243 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
12244 | string key; |
12245 | get_deferred_key(wt.seq, &key); | |
12246 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
12247 | } | |
12248 | } | |
12249 | ||
9f95a23c TL |
12250 | #if defined(WITH_LTTNG) |
12251 | auto sync_start = mono_clock::now(); | |
12252 | #endif | |
7c673cae | 12253 | // submit synct synchronously (block and wait for it to commit) |
31f18b77 | 12254 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
12255 | ceph_assert(r == 0); |
12256 | ||
f67539c2 TL |
12257 | #ifdef WITH_BLKIN |
12258 | for (auto txc : kv_committing) { | |
12259 | if (txc->trace) { | |
12260 | txc->trace.event("db sync submit"); | |
12261 | txc->trace.keyval("kv_committing size", kv_committing.size()); | |
12262 | } | |
12263 | } | |
12264 | #endif | |
12265 | ||
9f95a23c TL |
12266 | int committing_size = kv_committing.size(); |
12267 | int deferred_size = deferred_stable.size(); | |
12268 | ||
12269 | #if defined(WITH_LTTNG) | |
12270 | double sync_latency = ceph::to_seconds<double>(mono_clock::now() - sync_start); | |
12271 | for (auto txc: kv_committing) { | |
12272 | if (txc->tracing) { | |
12273 | tracepoint( | |
12274 | bluestore, | |
12275 | transaction_kv_sync_latency, | |
12276 | txc->osr->get_sequencer_id(), | |
12277 | txc->seq, | |
12278 | kv_committing.size(), | |
12279 | deferred_done.size(), | |
12280 | deferred_stable.size(), | |
12281 | sync_latency); | |
12282 | } | |
12283 | } | |
12284 | #endif | |
12285 | ||
11fdf7f2 | 12286 | { |
9f95a23c | 12287 | std::unique_lock m{kv_finalize_lock}; |
11fdf7f2 TL |
12288 | if (kv_committing_to_finalize.empty()) { |
12289 | kv_committing_to_finalize.swap(kv_committing); | |
12290 | } else { | |
12291 | kv_committing_to_finalize.insert( | |
12292 | kv_committing_to_finalize.end(), | |
12293 | kv_committing.begin(), | |
12294 | kv_committing.end()); | |
12295 | kv_committing.clear(); | |
12296 | } | |
12297 | if (deferred_stable_to_finalize.empty()) { | |
12298 | deferred_stable_to_finalize.swap(deferred_stable); | |
12299 | } else { | |
12300 | deferred_stable_to_finalize.insert( | |
12301 | deferred_stable_to_finalize.end(), | |
12302 | deferred_stable.begin(), | |
12303 | deferred_stable.end()); | |
12304 | deferred_stable.clear(); | |
12305 | } | |
9f95a23c TL |
12306 | if (!kv_finalize_in_progress) { |
12307 | kv_finalize_in_progress = true; | |
12308 | kv_finalize_cond.notify_one(); | |
12309 | } | |
11fdf7f2 | 12310 | } |
7c673cae FG |
12311 | |
12312 | if (new_nid_max) { | |
12313 | nid_max = new_nid_max; | |
12314 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
12315 | } | |
12316 | if (new_blobid_max) { | |
12317 | blobid_max = new_blobid_max; | |
12318 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
12319 | } | |
12320 | ||
224ce89b | 12321 | { |
11fdf7f2 TL |
12322 | auto finish = mono_clock::now(); |
12323 | ceph::timespan dur_flush = after_flush - start; | |
12324 | ceph::timespan dur_kv = finish - after_flush; | |
12325 | ceph::timespan dur = finish - start; | |
9f95a23c TL |
12326 | dout(20) << __func__ << " committed " << committing_size |
12327 | << " cleaned " << deferred_size | |
224ce89b WB |
12328 | << " in " << dur |
12329 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
12330 | << dendl; | |
494da23a TL |
12331 | log_latency("kv_flush", |
12332 | l_bluestore_kv_flush_lat, | |
12333 | dur_flush, | |
12334 | cct->_conf->bluestore_log_op_age); | |
12335 | log_latency("kv_commit", | |
12336 | l_bluestore_kv_commit_lat, | |
12337 | dur_kv, | |
12338 | cct->_conf->bluestore_log_op_age); | |
12339 | log_latency("kv_sync", | |
12340 | l_bluestore_kv_sync_lat, | |
12341 | dur, | |
12342 | cct->_conf->bluestore_log_op_age); | |
7c673cae | 12343 | } |
31f18b77 | 12344 | |
31f18b77 FG |
12345 | l.lock(); |
12346 | // previously deferred "done" are now "stable" by virtue of this | |
12347 | // commit cycle. | |
12348 | deferred_stable_queue.swap(deferred_done); | |
12349 | } | |
12350 | } | |
12351 | dout(10) << __func__ << " finish" << dendl; | |
12352 | kv_sync_started = false; | |
12353 | } | |
12354 | ||
12355 | void BlueStore::_kv_finalize_thread() | |
12356 | { | |
12357 | deque<TransContext*> kv_committed; | |
12358 | deque<DeferredBatch*> deferred_stable; | |
12359 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
12360 | std::unique_lock l(kv_finalize_lock); |
12361 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
12362 | kv_finalize_started = true; |
12363 | kv_finalize_cond.notify_all(); | |
12364 | while (true) { | |
11fdf7f2 TL |
12365 | ceph_assert(kv_committed.empty()); |
12366 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
12367 | if (kv_committing_to_finalize.empty() && |
12368 | deferred_stable_to_finalize.empty()) { | |
12369 | if (kv_finalize_stop) | |
12370 | break; | |
12371 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 12372 | kv_finalize_in_progress = false; |
31f18b77 FG |
12373 | kv_finalize_cond.wait(l); |
12374 | dout(20) << __func__ << " wake" << dendl; | |
12375 | } else { | |
12376 | kv_committed.swap(kv_committing_to_finalize); | |
12377 | deferred_stable.swap(deferred_stable_to_finalize); | |
12378 | l.unlock(); | |
12379 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
12380 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12381 | ||
11fdf7f2 TL |
12382 | auto start = mono_clock::now(); |
12383 | ||
31f18b77 FG |
12384 | while (!kv_committed.empty()) { |
12385 | TransContext *txc = kv_committed.front(); | |
f67539c2 | 12386 | ceph_assert(txc->get_state() == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12387 | _txc_state_proc(txc); |
31f18b77 | 12388 | kv_committed.pop_front(); |
7c673cae | 12389 | } |
31f18b77 | 12390 | |
7c673cae FG |
12391 | for (auto b : deferred_stable) { |
12392 | auto p = b->txcs.begin(); | |
12393 | while (p != b->txcs.end()) { | |
12394 | TransContext *txc = &*p; | |
12395 | p = b->txcs.erase(p); // unlink here because | |
12396 | _txc_state_proc(txc); // this may destroy txc | |
12397 | } | |
12398 | delete b; | |
12399 | } | |
31f18b77 | 12400 | deferred_stable.clear(); |
7c673cae FG |
12401 | |
12402 | if (!deferred_aggressive) { | |
31f18b77 | 12403 | if (deferred_queue_size >= deferred_batch_ops.load() || |
9f95a23c | 12404 | throttle.should_submit_deferred()) { |
224ce89b | 12405 | deferred_try_submit(); |
7c673cae FG |
12406 | } |
12407 | } | |
12408 | ||
12409 | // this is as good a place as any ... | |
12410 | _reap_collections(); | |
12411 | ||
11fdf7f2 | 12412 | logger->set(l_bluestore_fragmentation, |
f67539c2 | 12413 | (uint64_t)(shared_alloc.a->get_fragmentation() * 1000)); |
11fdf7f2 | 12414 | |
494da23a TL |
12415 | log_latency("kv_final", |
12416 | l_bluestore_kv_final_lat, | |
12417 | mono_clock::now() - start, | |
12418 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 | 12419 | |
7c673cae | 12420 | l.lock(); |
7c673cae FG |
12421 | } |
12422 | } | |
12423 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 12424 | kv_finalize_started = false; |
7c673cae FG |
12425 | } |
12426 | ||
f67539c2 TL |
12427 | void BlueStore::_zoned_cleaner_start() { |
12428 | dout(10) << __func__ << dendl; | |
12429 | ||
12430 | zoned_cleaner_thread.create("bstore_zcleaner"); | |
12431 | } | |
12432 | ||
12433 | void BlueStore::_zoned_cleaner_stop() { | |
12434 | dout(10) << __func__ << dendl; | |
12435 | { | |
12436 | std::unique_lock l{zoned_cleaner_lock}; | |
12437 | while (!zoned_cleaner_started) { | |
12438 | zoned_cleaner_cond.wait(l); | |
12439 | } | |
12440 | zoned_cleaner_stop = true; | |
12441 | zoned_cleaner_cond.notify_all(); | |
12442 | } | |
12443 | zoned_cleaner_thread.join(); | |
12444 | { | |
12445 | std::lock_guard l{zoned_cleaner_lock}; | |
12446 | zoned_cleaner_stop = false; | |
12447 | } | |
12448 | dout(10) << __func__ << " done" << dendl; | |
12449 | } | |
12450 | ||
12451 | void BlueStore::_zoned_cleaner_thread() { | |
12452 | dout(10) << __func__ << " start" << dendl; | |
12453 | std::unique_lock l{zoned_cleaner_lock}; | |
12454 | ceph_assert(!zoned_cleaner_started); | |
12455 | zoned_cleaner_started = true; | |
12456 | zoned_cleaner_cond.notify_all(); | |
12457 | std::deque<uint64_t> zones_to_clean; | |
12458 | while (true) { | |
12459 | if (zoned_cleaner_queue.empty()) { | |
12460 | if (zoned_cleaner_stop) { | |
12461 | break; | |
12462 | } | |
12463 | dout(20) << __func__ << " sleep" << dendl; | |
12464 | zoned_cleaner_cond.wait(l); | |
12465 | dout(20) << __func__ << " wake" << dendl; | |
12466 | } else { | |
12467 | zones_to_clean.swap(zoned_cleaner_queue); | |
12468 | l.unlock(); | |
12469 | while (!zones_to_clean.empty()) { | |
12470 | _zoned_clean_zone(zones_to_clean.front()); | |
12471 | zones_to_clean.pop_front(); | |
12472 | } | |
12473 | l.lock(); | |
12474 | } | |
12475 | } | |
12476 | dout(10) << __func__ << " finish" << dendl; | |
12477 | zoned_cleaner_started = false; | |
12478 | } | |
12479 | ||
12480 | void BlueStore::_zoned_clean_zone(uint64_t zone_num) { | |
12481 | dout(10) << __func__ << " cleaning zone " << zone_num << dendl; | |
12482 | } | |
12483 | ||
7c673cae | 12484 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( |
522d829b | 12485 | TransContext *txc, uint64_t len) |
7c673cae FG |
12486 | { |
12487 | if (!txc->deferred_txn) { | |
12488 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
12489 | } | |
12490 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
522d829b TL |
12491 | logger->inc(l_bluestore_write_deferred); |
12492 | logger->inc(l_bluestore_write_deferred_bytes, len); | |
7c673cae FG |
12493 | return &txc->deferred_txn->ops.back(); |
12494 | } | |
12495 | ||
12496 | void BlueStore::_deferred_queue(TransContext *txc) | |
12497 | { | |
12498 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
f67539c2 TL |
12499 | |
12500 | DeferredBatch *tmp; | |
12501 | txc->osr->deferred_lock.lock(); | |
12502 | { | |
12503 | if (!txc->osr->deferred_pending) { | |
12504 | tmp = new DeferredBatch(cct, txc->osr.get()); | |
12505 | } else { | |
12506 | tmp = txc->osr->deferred_pending; | |
12507 | } | |
7c673cae | 12508 | } |
f67539c2 TL |
12509 | |
12510 | tmp->txcs.push_back(*txc); | |
7c673cae FG |
12511 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; |
12512 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
12513 | const auto& op = *opi; | |
11fdf7f2 | 12514 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
12515 | bufferlist::const_iterator p = op.data.begin(); |
12516 | for (auto e : op.extents) { | |
f67539c2 | 12517 | tmp->prepare_write(cct, wt.seq, e.offset, e.length, p); |
7c673cae FG |
12518 | } |
12519 | } | |
f67539c2 TL |
12520 | |
12521 | { | |
12522 | ++deferred_queue_size; | |
12523 | txc->osr->deferred_pending = tmp; | |
12524 | // condition "tmp->txcs.size() == 1" mean deferred_pending was originally empty. | |
12525 | // So we should add osr into deferred_queue. | |
12526 | if (!txc->osr->deferred_running && (tmp->txcs.size() == 1)) { | |
12527 | deferred_lock.lock(); | |
12528 | deferred_queue.push_back(*txc->osr); | |
12529 | deferred_lock.unlock(); | |
12530 | } | |
12531 | ||
12532 | if (deferred_aggressive && | |
12533 | !txc->osr->deferred_running) { | |
12534 | _deferred_submit_unlock(txc->osr.get()); | |
12535 | } else { | |
12536 | txc->osr->deferred_lock.unlock(); | |
12537 | } | |
7c673cae | 12538 | } |
f67539c2 | 12539 | } |
7c673cae | 12540 | |
224ce89b | 12541 | void BlueStore::deferred_try_submit() |
7c673cae FG |
12542 | { |
12543 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
12544 | << deferred_queue_size << " txcs" << dendl; | |
224ce89b | 12545 | vector<OpSequencerRef> osrs; |
f67539c2 TL |
12546 | |
12547 | { | |
12548 | std::lock_guard l(deferred_lock); | |
12549 | osrs.reserve(deferred_queue.size()); | |
12550 | for (auto& osr : deferred_queue) { | |
12551 | osrs.push_back(&osr); | |
12552 | } | |
224ce89b | 12553 | } |
f67539c2 | 12554 | |
224ce89b | 12555 | for (auto& osr : osrs) { |
f67539c2 | 12556 | osr->deferred_lock.lock(); |
181888fb FG |
12557 | if (osr->deferred_pending) { |
12558 | if (!osr->deferred_running) { | |
12559 | _deferred_submit_unlock(osr.get()); | |
181888fb | 12560 | } else { |
f67539c2 | 12561 | osr->deferred_lock.unlock(); |
181888fb FG |
12562 | dout(20) << __func__ << " osr " << osr << " already has running" |
12563 | << dendl; | |
12564 | } | |
12565 | } else { | |
f67539c2 | 12566 | osr->deferred_lock.unlock(); |
181888fb | 12567 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; |
7c673cae FG |
12568 | } |
12569 | } | |
9f95a23c | 12570 | |
f67539c2 TL |
12571 | { |
12572 | std::lock_guard l(deferred_lock); | |
12573 | deferred_last_submitted = ceph_clock_now(); | |
12574 | } | |
7c673cae FG |
12575 | } |
12576 | ||
224ce89b | 12577 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
12578 | { |
12579 | dout(10) << __func__ << " osr " << osr | |
12580 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
12581 | << dendl; | |
11fdf7f2 TL |
12582 | ceph_assert(osr->deferred_pending); |
12583 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
12584 | |
12585 | auto b = osr->deferred_pending; | |
12586 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 12587 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
12588 | |
12589 | osr->deferred_running = osr->deferred_pending; | |
12590 | osr->deferred_pending = nullptr; | |
12591 | ||
f67539c2 | 12592 | osr->deferred_lock.unlock(); |
11fdf7f2 TL |
12593 | |
12594 | for (auto& txc : b->txcs) { | |
9f95a23c | 12595 | throttle.log_state_latency(txc, logger, l_bluestore_state_deferred_queued_lat); |
11fdf7f2 | 12596 | } |
7c673cae FG |
12597 | uint64_t start = 0, pos = 0; |
12598 | bufferlist bl; | |
12599 | auto i = b->iomap.begin(); | |
12600 | while (true) { | |
12601 | if (i == b->iomap.end() || i->first != pos) { | |
12602 | if (bl.length()) { | |
12603 | dout(20) << __func__ << " write 0x" << std::hex | |
12604 | << start << "~" << bl.length() | |
12605 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 12606 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
12607 | logger->inc(l_bluestore_deferred_write_ops); |
12608 | logger->inc(l_bluestore_deferred_write_bytes, bl.length()); | |
12609 | int r = bdev->aio_write(start, bl, &b->ioc, false); | |
11fdf7f2 | 12610 | ceph_assert(r == 0); |
7c673cae FG |
12611 | } |
12612 | } | |
12613 | if (i == b->iomap.end()) { | |
12614 | break; | |
12615 | } | |
12616 | start = 0; | |
12617 | pos = i->first; | |
12618 | bl.clear(); | |
12619 | } | |
12620 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
12621 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
12622 | << dendl; | |
12623 | if (!bl.length()) { | |
12624 | start = pos; | |
12625 | } | |
12626 | pos += i->second.bl.length(); | |
12627 | bl.claim_append(i->second.bl); | |
12628 | ++i; | |
12629 | } | |
224ce89b | 12630 | |
7c673cae FG |
12631 | bdev->aio_submit(&b->ioc); |
12632 | } | |
12633 | ||
3efd9988 FG |
12634 | struct C_DeferredTrySubmit : public Context { |
12635 | BlueStore *store; | |
12636 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
12637 | void finish(int r) { | |
12638 | store->deferred_try_submit(); | |
12639 | } | |
12640 | }; | |
12641 | ||
7c673cae FG |
12642 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
12643 | { | |
12644 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 12645 | ceph_assert(osr->deferred_running); |
7c673cae FG |
12646 | DeferredBatch *b = osr->deferred_running; |
12647 | ||
12648 | { | |
f67539c2 | 12649 | osr->deferred_lock.lock(); |
11fdf7f2 | 12650 | ceph_assert(osr->deferred_running == b); |
7c673cae FG |
12651 | osr->deferred_running = nullptr; |
12652 | if (!osr->deferred_pending) { | |
181888fb | 12653 | dout(20) << __func__ << " dequeueing" << dendl; |
f67539c2 TL |
12654 | { |
12655 | deferred_lock.lock(); | |
12656 | auto q = deferred_queue.iterator_to(*osr); | |
12657 | deferred_queue.erase(q); | |
12658 | deferred_lock.unlock(); | |
12659 | } | |
12660 | osr->deferred_lock.unlock(); | |
181888fb | 12661 | } else { |
f67539c2 | 12662 | osr->deferred_lock.unlock(); |
9f95a23c TL |
12663 | if (deferred_aggressive) { |
12664 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; | |
12665 | finisher.queue(new C_DeferredTrySubmit(this)); | |
12666 | } else { | |
12667 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
12668 | } | |
7c673cae FG |
12669 | } |
12670 | } | |
12671 | ||
12672 | { | |
31f18b77 | 12673 | uint64_t costs = 0; |
11fdf7f2 | 12674 | { |
11fdf7f2 TL |
12675 | for (auto& i : b->txcs) { |
12676 | TransContext *txc = &i; | |
9f95a23c | 12677 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_aio_wait_lat); |
f67539c2 | 12678 | txc->set_state(TransContext::STATE_DEFERRED_CLEANUP); |
11fdf7f2 TL |
12679 | costs += txc->cost; |
12680 | } | |
7c673cae | 12681 | } |
9f95a23c | 12682 | throttle.release_deferred_throttle(costs); |
7c673cae FG |
12683 | } |
12684 | ||
9f95a23c | 12685 | { |
11fdf7f2 | 12686 | std::lock_guard l(kv_lock); |
9f95a23c TL |
12687 | deferred_done_queue.emplace_back(b); |
12688 | ||
12689 | // in the normal case, do not bother waking up the kv thread; it will | |
12690 | // catch us on the next commit anyway. | |
12691 | if (deferred_aggressive && !kv_sync_in_progress) { | |
12692 | kv_sync_in_progress = true; | |
12693 | kv_cond.notify_one(); | |
12694 | } | |
7c673cae FG |
12695 | } |
12696 | } | |
12697 | ||
12698 | int BlueStore::_deferred_replay() | |
12699 | { | |
12700 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
12701 | int count = 0; |
12702 | int r = 0; | |
11fdf7f2 TL |
12703 | CollectionRef ch = _get_collection(coll_t::meta()); |
12704 | bool fake_ch = false; | |
12705 | if (!ch) { | |
12706 | // hmm, replaying initial mkfs? | |
12707 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
12708 | fake_ch = true; | |
12709 | } | |
12710 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
12711 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
12712 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
12713 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
12714 | << dendl; | |
12715 | bluestore_deferred_transaction_t *deferred_txn = | |
12716 | new bluestore_deferred_transaction_t; | |
12717 | bufferlist bl = it->value(); | |
11fdf7f2 | 12718 | auto p = bl.cbegin(); |
7c673cae | 12719 | try { |
11fdf7f2 | 12720 | decode(*deferred_txn, p); |
f67539c2 | 12721 | } catch (ceph::buffer::error& e) { |
7c673cae FG |
12722 | derr << __func__ << " failed to decode deferred txn " |
12723 | << pretty_binary_string(it->key()) << dendl; | |
12724 | delete deferred_txn; | |
12725 | r = -EIO; | |
12726 | goto out; | |
12727 | } | |
11fdf7f2 | 12728 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); |
7c673cae | 12729 | txc->deferred_txn = deferred_txn; |
f67539c2 | 12730 | txc->set_state(TransContext::STATE_KV_DONE); |
7c673cae FG |
12731 | _txc_state_proc(txc); |
12732 | } | |
12733 | out: | |
12734 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 12735 | _osr_register_zombie(osr); |
7c673cae | 12736 | _osr_drain_all(); |
11fdf7f2 TL |
12737 | if (fake_ch) { |
12738 | new_coll_map.clear(); | |
12739 | } | |
7c673cae FG |
12740 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
12741 | return r; | |
12742 | } | |
12743 | ||
12744 | // --------------------------- | |
12745 | // transactions | |
12746 | ||
12747 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
12748 | CollectionHandle& ch, |
12749 | vector<Transaction>& tls, | |
12750 | TrackedOpRef op, | |
12751 | ThreadPool::TPHandle *handle) | |
12752 | { | |
12753 | FUNCTRACE(cct); | |
12754 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 12755 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 12756 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae | 12757 | |
11fdf7f2 TL |
12758 | auto start = mono_clock::now(); |
12759 | ||
12760 | Collection *c = static_cast<Collection*>(ch.get()); | |
12761 | OpSequencer *osr = c->osr.get(); | |
12762 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae FG |
12763 | |
12764 | // prepare | |
11fdf7f2 | 12765 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, |
f67539c2 | 12766 | &on_commit, op); |
7c673cae | 12767 | |
f67539c2 TL |
12768 | // With HM-SMR drives (and ZNS SSDs) we want the I/O allocation and I/O |
12769 | // submission to happen atomically because if I/O submission happens in a | |
12770 | // different order than I/O allocation, we end up issuing non-sequential | |
12771 | // writes to the drive. This is a temporary solution until ZONE APPEND | |
12772 | // support matures in the kernel. For more information please see: | |
12773 | // https://www.usenix.org/conference/vault20/presentation/bjorling | |
12774 | if (bdev->is_smr()) { | |
12775 | atomic_alloc_and_submit_lock.lock(); | |
12776 | } | |
7c673cae | 12777 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { |
7c673cae FG |
12778 | txc->bytes += (*p).get_num_bytes(); |
12779 | _txc_add_transaction(txc, &(*p)); | |
12780 | } | |
12781 | _txc_calc_cost(txc); | |
12782 | ||
12783 | _txc_write_nodes(txc, txc->t); | |
12784 | ||
12785 | // journal deferred items | |
12786 | if (txc->deferred_txn) { | |
12787 | txc->deferred_txn->seq = ++deferred_seq; | |
12788 | bufferlist bl; | |
11fdf7f2 | 12789 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
12790 | string key; |
12791 | get_deferred_key(txc->deferred_txn->seq, &key); | |
12792 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
12793 | } | |
12794 | ||
12795 | _txc_finalize_kv(txc, txc->t); | |
f67539c2 TL |
12796 | |
12797 | #ifdef WITH_BLKIN | |
12798 | if (txc->trace) { | |
12799 | txc->trace.event("txc encode finished"); | |
12800 | } | |
12801 | #endif | |
12802 | ||
7c673cae FG |
12803 | if (handle) |
12804 | handle->suspend_tp_timeout(); | |
12805 | ||
11fdf7f2 | 12806 | auto tstart = mono_clock::now(); |
9f95a23c TL |
12807 | |
12808 | if (!throttle.try_start_transaction( | |
12809 | *db, | |
12810 | *txc, | |
12811 | tstart)) { | |
7c673cae | 12812 | // ensure we do not block here because of deferred writes |
9f95a23c TL |
12813 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
12814 | << dendl; | |
12815 | ++deferred_aggressive; | |
12816 | deferred_try_submit(); | |
12817 | { | |
12818 | // wake up any previously finished deferred events | |
12819 | std::lock_guard l(kv_lock); | |
12820 | if (!kv_sync_in_progress) { | |
12821 | kv_sync_in_progress = true; | |
3efd9988 FG |
12822 | kv_cond.notify_one(); |
12823 | } | |
9f95a23c TL |
12824 | } |
12825 | throttle.finish_start_transaction(*db, *txc, tstart); | |
12826 | --deferred_aggressive; | |
7c673cae | 12827 | } |
11fdf7f2 | 12828 | auto tend = mono_clock::now(); |
7c673cae FG |
12829 | |
12830 | if (handle) | |
12831 | handle->reset_tp_timeout(); | |
12832 | ||
12833 | logger->inc(l_bluestore_txc); | |
12834 | ||
12835 | // execute (start) | |
12836 | _txc_state_proc(txc); | |
12837 | ||
f67539c2 TL |
12838 | if (bdev->is_smr()) { |
12839 | atomic_alloc_and_submit_lock.unlock(); | |
12840 | } | |
12841 | ||
11fdf7f2 TL |
12842 | // we're immediately readable (unlike FileStore) |
12843 | for (auto c : on_applied_sync) { | |
12844 | c->complete(0); | |
12845 | } | |
12846 | if (!on_applied.empty()) { | |
12847 | if (c->commit_queue) { | |
12848 | c->commit_queue->queue(on_applied); | |
12849 | } else { | |
12850 | finisher.queue(on_applied); | |
12851 | } | |
12852 | } | |
12853 | ||
f67539c2 TL |
12854 | #ifdef WITH_BLKIN |
12855 | if (txc->trace) { | |
12856 | txc->trace.event("txc applied"); | |
12857 | } | |
12858 | #endif | |
12859 | ||
494da23a TL |
12860 | log_latency("submit_transact", |
12861 | l_bluestore_submit_lat, | |
12862 | mono_clock::now() - start, | |
12863 | cct->_conf->bluestore_log_op_age); | |
12864 | log_latency("throttle_transact", | |
12865 | l_bluestore_throttle_lat, | |
12866 | tend - tstart, | |
12867 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
12868 | return 0; |
12869 | } | |
12870 | ||
12871 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
12872 | { | |
12873 | dout(10) << __func__ << " txc " << txc << dendl; | |
12874 | bdev->aio_submit(&txc->ioc); | |
12875 | } | |
12876 | ||
12877 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
12878 | { | |
12879 | Transaction::iterator i = t->begin(); | |
12880 | ||
81eedcae | 12881 | _dump_transaction<30>(cct, t); |
7c673cae FG |
12882 | |
12883 | vector<CollectionRef> cvec(i.colls.size()); | |
12884 | unsigned j = 0; | |
12885 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
12886 | ++p, ++j) { | |
12887 | cvec[j] = _get_collection(*p); | |
7c673cae | 12888 | } |
11fdf7f2 | 12889 | |
7c673cae FG |
12890 | vector<OnodeRef> ovec(i.objects.size()); |
12891 | ||
12892 | for (int pos = 0; i.have_op(); ++pos) { | |
12893 | Transaction::Op *op = i.decode_op(); | |
12894 | int r = 0; | |
12895 | ||
12896 | // no coll or obj | |
12897 | if (op->op == Transaction::OP_NOP) | |
12898 | continue; | |
12899 | ||
11fdf7f2 | 12900 | |
7c673cae FG |
12901 | // collection operations |
12902 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
12903 | |
12904 | // initialize osd_pool_id and do a smoke test that all collections belong | |
12905 | // to the same pool | |
12906 | spg_t pgid; | |
12907 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
12908 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
12909 | txc->osd_pool_id == pgid.pool()); | |
12910 | txc->osd_pool_id = pgid.pool(); | |
12911 | } | |
12912 | ||
7c673cae FG |
12913 | switch (op->op) { |
12914 | case Transaction::OP_RMCOLL: | |
12915 | { | |
12916 | const coll_t &cid = i.get_cid(op->cid); | |
12917 | r = _remove_collection(txc, cid, &c); | |
12918 | if (!r) | |
12919 | continue; | |
12920 | } | |
12921 | break; | |
12922 | ||
12923 | case Transaction::OP_MKCOLL: | |
12924 | { | |
11fdf7f2 | 12925 | ceph_assert(!c); |
7c673cae FG |
12926 | const coll_t &cid = i.get_cid(op->cid); |
12927 | r = _create_collection(txc, cid, op->split_bits, &c); | |
12928 | if (!r) | |
12929 | continue; | |
12930 | } | |
12931 | break; | |
12932 | ||
12933 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 12934 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12935 | break; |
12936 | ||
12937 | case Transaction::OP_SPLIT_COLLECTION2: | |
12938 | { | |
12939 | uint32_t bits = op->split_bits; | |
12940 | uint32_t rem = op->split_rem; | |
12941 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
12942 | if (!r) | |
12943 | continue; | |
12944 | } | |
12945 | break; | |
12946 | ||
11fdf7f2 TL |
12947 | case Transaction::OP_MERGE_COLLECTION: |
12948 | { | |
12949 | uint32_t bits = op->split_bits; | |
12950 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
12951 | if (!r) | |
12952 | continue; | |
12953 | } | |
12954 | break; | |
12955 | ||
7c673cae FG |
12956 | case Transaction::OP_COLL_HINT: |
12957 | { | |
f67539c2 | 12958 | uint32_t type = op->hint; |
7c673cae FG |
12959 | bufferlist hint; |
12960 | i.decode_bl(hint); | |
11fdf7f2 | 12961 | auto hiter = hint.cbegin(); |
7c673cae FG |
12962 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
12963 | uint32_t pg_num; | |
12964 | uint64_t num_objs; | |
11fdf7f2 TL |
12965 | decode(pg_num, hiter); |
12966 | decode(num_objs, hiter); | |
7c673cae FG |
12967 | dout(10) << __func__ << " collection hint objects is a no-op, " |
12968 | << " pg_num " << pg_num << " num_objects " << num_objs | |
12969 | << dendl; | |
12970 | } else { | |
12971 | // Ignore the hint | |
12972 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
12973 | } | |
12974 | continue; | |
12975 | } | |
12976 | break; | |
12977 | ||
12978 | case Transaction::OP_COLL_SETATTR: | |
12979 | r = -EOPNOTSUPP; | |
12980 | break; | |
12981 | ||
12982 | case Transaction::OP_COLL_RMATTR: | |
12983 | r = -EOPNOTSUPP; | |
12984 | break; | |
12985 | ||
12986 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 12987 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12988 | break; |
12989 | } | |
12990 | if (r < 0) { | |
12991 | derr << __func__ << " error " << cpp_strerror(r) | |
12992 | << " not handled on operation " << op->op | |
12993 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 12994 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12995 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12996 | } |
12997 | ||
12998 | // these operations implicity create the object | |
12999 | bool create = false; | |
13000 | if (op->op == Transaction::OP_TOUCH || | |
9f95a23c | 13001 | op->op == Transaction::OP_CREATE || |
7c673cae FG |
13002 | op->op == Transaction::OP_WRITE || |
13003 | op->op == Transaction::OP_ZERO) { | |
13004 | create = true; | |
13005 | } | |
13006 | ||
13007 | // object operations | |
9f95a23c | 13008 | std::unique_lock l(c->lock); |
7c673cae FG |
13009 | OnodeRef &o = ovec[op->oid]; |
13010 | if (!o) { | |
13011 | ghobject_t oid = i.get_oid(op->oid); | |
9f95a23c | 13012 | o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE); |
7c673cae FG |
13013 | } |
13014 | if (!create && (!o || !o->exists)) { | |
13015 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
13016 | << i.get_oid(op->oid) << dendl; | |
13017 | r = -ENOENT; | |
13018 | goto endop; | |
13019 | } | |
13020 | ||
13021 | switch (op->op) { | |
9f95a23c | 13022 | case Transaction::OP_CREATE: |
7c673cae FG |
13023 | case Transaction::OP_TOUCH: |
13024 | r = _touch(txc, c, o); | |
13025 | break; | |
13026 | ||
13027 | case Transaction::OP_WRITE: | |
13028 | { | |
13029 | uint64_t off = op->off; | |
13030 | uint64_t len = op->len; | |
13031 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
13032 | bufferlist bl; | |
13033 | i.decode_bl(bl); | |
13034 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
13035 | } | |
13036 | break; | |
13037 | ||
13038 | case Transaction::OP_ZERO: | |
13039 | { | |
13040 | uint64_t off = op->off; | |
13041 | uint64_t len = op->len; | |
13042 | r = _zero(txc, c, o, off, len); | |
13043 | } | |
13044 | break; | |
13045 | ||
13046 | case Transaction::OP_TRIMCACHE: | |
13047 | { | |
13048 | // deprecated, no-op | |
13049 | } | |
13050 | break; | |
13051 | ||
13052 | case Transaction::OP_TRUNCATE: | |
13053 | { | |
13054 | uint64_t off = op->off; | |
35e4c445 | 13055 | r = _truncate(txc, c, o, off); |
7c673cae FG |
13056 | } |
13057 | break; | |
13058 | ||
13059 | case Transaction::OP_REMOVE: | |
13060 | { | |
13061 | r = _remove(txc, c, o); | |
13062 | } | |
13063 | break; | |
13064 | ||
13065 | case Transaction::OP_SETATTR: | |
13066 | { | |
13067 | string name = i.decode_string(); | |
13068 | bufferptr bp; | |
13069 | i.decode_bp(bp); | |
13070 | r = _setattr(txc, c, o, name, bp); | |
13071 | } | |
13072 | break; | |
13073 | ||
13074 | case Transaction::OP_SETATTRS: | |
13075 | { | |
13076 | map<string, bufferptr> aset; | |
13077 | i.decode_attrset(aset); | |
13078 | r = _setattrs(txc, c, o, aset); | |
13079 | } | |
13080 | break; | |
13081 | ||
13082 | case Transaction::OP_RMATTR: | |
13083 | { | |
13084 | string name = i.decode_string(); | |
13085 | r = _rmattr(txc, c, o, name); | |
13086 | } | |
13087 | break; | |
13088 | ||
13089 | case Transaction::OP_RMATTRS: | |
13090 | { | |
13091 | r = _rmattrs(txc, c, o); | |
13092 | } | |
13093 | break; | |
13094 | ||
13095 | case Transaction::OP_CLONE: | |
13096 | { | |
13097 | OnodeRef& no = ovec[op->dest_oid]; | |
13098 | if (!no) { | |
13099 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
13100 | no = c->get_onode(noid, true); | |
13101 | } | |
13102 | r = _clone(txc, c, o, no); | |
13103 | } | |
13104 | break; | |
13105 | ||
13106 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 13107 | ceph_abort_msg("deprecated"); |
7c673cae FG |
13108 | break; |
13109 | ||
13110 | case Transaction::OP_CLONERANGE2: | |
13111 | { | |
13112 | OnodeRef& no = ovec[op->dest_oid]; | |
13113 | if (!no) { | |
13114 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
13115 | no = c->get_onode(noid, true); | |
13116 | } | |
13117 | uint64_t srcoff = op->off; | |
13118 | uint64_t len = op->len; | |
13119 | uint64_t dstoff = op->dest_off; | |
13120 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
13121 | } | |
13122 | break; | |
13123 | ||
13124 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 13125 | ceph_abort_msg("not implemented"); |
7c673cae FG |
13126 | break; |
13127 | ||
13128 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 13129 | ceph_abort_msg("not implemented"); |
7c673cae FG |
13130 | break; |
13131 | ||
13132 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 13133 | ceph_abort_msg("deprecated"); |
7c673cae FG |
13134 | break; |
13135 | ||
13136 | case Transaction::OP_COLL_MOVE_RENAME: | |
13137 | case Transaction::OP_TRY_RENAME: | |
13138 | { | |
11fdf7f2 | 13139 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
13140 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
13141 | OnodeRef& no = ovec[op->dest_oid]; | |
13142 | if (!no) { | |
13143 | no = c->get_onode(noid, false); | |
13144 | } | |
13145 | r = _rename(txc, c, o, no, noid); | |
13146 | } | |
13147 | break; | |
13148 | ||
13149 | case Transaction::OP_OMAP_CLEAR: | |
13150 | { | |
13151 | r = _omap_clear(txc, c, o); | |
13152 | } | |
13153 | break; | |
13154 | case Transaction::OP_OMAP_SETKEYS: | |
13155 | { | |
13156 | bufferlist aset_bl; | |
13157 | i.decode_attrset_bl(&aset_bl); | |
13158 | r = _omap_setkeys(txc, c, o, aset_bl); | |
13159 | } | |
13160 | break; | |
13161 | case Transaction::OP_OMAP_RMKEYS: | |
13162 | { | |
13163 | bufferlist keys_bl; | |
13164 | i.decode_keyset_bl(&keys_bl); | |
13165 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
13166 | } | |
13167 | break; | |
13168 | case Transaction::OP_OMAP_RMKEYRANGE: | |
13169 | { | |
13170 | string first, last; | |
13171 | first = i.decode_string(); | |
13172 | last = i.decode_string(); | |
13173 | r = _omap_rmkey_range(txc, c, o, first, last); | |
13174 | } | |
13175 | break; | |
13176 | case Transaction::OP_OMAP_SETHEADER: | |
13177 | { | |
13178 | bufferlist bl; | |
13179 | i.decode_bl(bl); | |
13180 | r = _omap_setheader(txc, c, o, bl); | |
13181 | } | |
13182 | break; | |
13183 | ||
13184 | case Transaction::OP_SETALLOCHINT: | |
13185 | { | |
13186 | r = _set_alloc_hint(txc, c, o, | |
13187 | op->expected_object_size, | |
13188 | op->expected_write_size, | |
f67539c2 | 13189 | op->hint); |
7c673cae FG |
13190 | } |
13191 | break; | |
13192 | ||
13193 | default: | |
11fdf7f2 | 13194 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
13195 | ceph_abort(); |
13196 | } | |
13197 | ||
13198 | endop: | |
13199 | if (r < 0) { | |
13200 | bool ok = false; | |
13201 | ||
13202 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
13203 | op->op == Transaction::OP_CLONE || | |
13204 | op->op == Transaction::OP_CLONERANGE2 || | |
13205 | op->op == Transaction::OP_COLL_ADD || | |
13206 | op->op == Transaction::OP_SETATTR || | |
13207 | op->op == Transaction::OP_SETATTRS || | |
13208 | op->op == Transaction::OP_RMATTR || | |
13209 | op->op == Transaction::OP_OMAP_SETKEYS || | |
13210 | op->op == Transaction::OP_OMAP_RMKEYS || | |
13211 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
13212 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
13213 | // -ENOENT is usually okay | |
13214 | ok = true; | |
13215 | if (r == -ENODATA) | |
13216 | ok = true; | |
13217 | ||
13218 | if (!ok) { | |
13219 | const char *msg = "unexpected error code"; | |
13220 | ||
13221 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
13222 | op->op == Transaction::OP_CLONE || | |
13223 | op->op == Transaction::OP_CLONERANGE2)) | |
13224 | msg = "ENOENT on clone suggests osd bug"; | |
13225 | ||
13226 | if (r == -ENOSPC) | |
13227 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
13228 | // by partially applying transactions. | |
13229 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
13230 | ||
13231 | if (r == -ENOTEMPTY) { | |
13232 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
13233 | } | |
13234 | ||
13235 | derr << __func__ << " error " << cpp_strerror(r) | |
13236 | << " not handled on operation " << op->op | |
13237 | << " (op " << pos << ", counting from 0)" | |
13238 | << dendl; | |
13239 | derr << msg << dendl; | |
81eedcae | 13240 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 13241 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
13242 | } |
13243 | } | |
13244 | } | |
13245 | } | |
13246 | ||
13247 | ||
13248 | ||
13249 | // ----------------- | |
13250 | // write operations | |
13251 | ||
13252 | int BlueStore::_touch(TransContext *txc, | |
13253 | CollectionRef& c, | |
13254 | OnodeRef &o) | |
13255 | { | |
13256 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
13257 | int r = 0; | |
7c673cae FG |
13258 | _assign_nid(txc, o); |
13259 | txc->write_onode(o); | |
13260 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
13261 | return r; | |
13262 | } | |
13263 | ||
7c673cae FG |
13264 | void BlueStore::_pad_zeros( |
13265 | bufferlist *bl, uint64_t *offset, | |
13266 | uint64_t chunk_size) | |
13267 | { | |
13268 | auto length = bl->length(); | |
13269 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
13270 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
13271 | dout(40) << "before:\n"; | |
13272 | bl->hexdump(*_dout); | |
13273 | *_dout << dendl; | |
13274 | // front | |
13275 | size_t front_pad = *offset % chunk_size; | |
13276 | size_t back_pad = 0; | |
13277 | size_t pad_count = 0; | |
13278 | if (front_pad) { | |
11fdf7f2 | 13279 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
f67539c2 | 13280 | bufferptr z = ceph::buffer::create_small_page_aligned(chunk_size); |
224ce89b | 13281 | z.zero(0, front_pad, false); |
7c673cae | 13282 | pad_count += front_pad; |
9f95a23c | 13283 | bl->begin().copy(front_copy, z.c_str() + front_pad); |
7c673cae FG |
13284 | if (front_copy + front_pad < chunk_size) { |
13285 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 13286 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
13287 | pad_count += back_pad; |
13288 | } | |
13289 | bufferlist old, t; | |
13290 | old.swap(*bl); | |
13291 | t.substr_of(old, front_copy, length - front_copy); | |
13292 | bl->append(z); | |
13293 | bl->claim_append(t); | |
13294 | *offset -= front_pad; | |
224ce89b | 13295 | length += pad_count; |
7c673cae FG |
13296 | } |
13297 | ||
13298 | // back | |
13299 | uint64_t end = *offset + length; | |
13300 | unsigned back_copy = end % chunk_size; | |
13301 | if (back_copy) { | |
11fdf7f2 | 13302 | ceph_assert(back_pad == 0); |
7c673cae | 13303 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 13304 | ceph_assert(back_copy <= length); |
7c673cae | 13305 | bufferptr tail(chunk_size); |
9f95a23c | 13306 | bl->begin(length - back_copy).copy(back_copy, tail.c_str()); |
224ce89b | 13307 | tail.zero(back_copy, back_pad, false); |
7c673cae FG |
13308 | bufferlist old; |
13309 | old.swap(*bl); | |
13310 | bl->substr_of(old, 0, length - back_copy); | |
13311 | bl->append(tail); | |
13312 | length += back_pad; | |
13313 | pad_count += back_pad; | |
13314 | } | |
13315 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
13316 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
13317 | << length << std::dec << dendl; | |
13318 | dout(40) << "after:\n"; | |
13319 | bl->hexdump(*_dout); | |
13320 | *_dout << dendl; | |
13321 | if (pad_count) | |
13322 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 13323 | ceph_assert(bl->length() == length); |
7c673cae FG |
13324 | } |
13325 | ||
13326 | void BlueStore::_do_write_small( | |
13327 | TransContext *txc, | |
13328 | CollectionRef &c, | |
13329 | OnodeRef o, | |
13330 | uint64_t offset, uint64_t length, | |
13331 | bufferlist::iterator& blp, | |
13332 | WriteContext *wctx) | |
13333 | { | |
13334 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13335 | << std::dec << dendl; | |
11fdf7f2 | 13336 | ceph_assert(length < min_alloc_size); |
f67539c2 | 13337 | |
7c673cae FG |
13338 | uint64_t end_offs = offset + length; |
13339 | ||
13340 | logger->inc(l_bluestore_write_small); | |
13341 | logger->inc(l_bluestore_write_small_bytes, length); | |
13342 | ||
13343 | bufferlist bl; | |
13344 | blp.copy(length, bl); | |
13345 | ||
81eedcae TL |
13346 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
13347 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
13348 | uint32_t alloc_len = min_alloc_size; | |
13349 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
13350 | ||
13351 | bool any_change; | |
13352 | ||
13353 | // search suitable extent in both forward and reverse direction in | |
13354 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
13355 | // then check if blob can be reused via can_reuse_blob func or apply | |
13356 | // direct/deferred write (the latter for extents including or higher | |
13357 | // than 'offset' only). | |
13358 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
13359 | ||
f67539c2 TL |
13360 | // On zoned devices, the first goal is to support non-overwrite workloads, |
13361 | // such as RGW, with large, aligned objects. Therefore, for user writes | |
13362 | // _do_write_small should not trigger. OSDs, however, write and update a tiny | |
13363 | // amount of metadata, such as OSD maps, to disk. For those cases, we | |
13364 | // temporarily just pad them to min_alloc_size and write them to a new place | |
13365 | // on every update. | |
13366 | if (bdev->is_smr()) { | |
13367 | BlobRef b = c->new_blob(); | |
13368 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); | |
13369 | uint64_t b_off0 = b_off; | |
13370 | _pad_zeros(&bl, &b_off0, min_alloc_size); | |
13371 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13372 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, false, true); | |
13373 | return; | |
13374 | } | |
13375 | ||
7c673cae FG |
13376 | // Look for an existing mutable blob we can use. |
13377 | auto begin = o->extent_map.extent_map.begin(); | |
13378 | auto end = o->extent_map.extent_map.end(); | |
13379 | auto ep = o->extent_map.seek_lextent(offset); | |
13380 | if (ep != begin) { | |
13381 | --ep; | |
13382 | if (ep->blob_end() <= offset) { | |
13383 | ++ep; | |
13384 | } | |
13385 | } | |
f67539c2 TL |
13386 | auto prev_ep = end; |
13387 | if (ep != begin) { | |
13388 | prev_ep = ep; | |
7c673cae | 13389 | --prev_ep; |
7c673cae FG |
13390 | } |
13391 | ||
eafe8130 TL |
13392 | boost::container::flat_set<const bluestore_blob_t*> inspected_blobs; |
13393 | // We don't want to have more blobs than min alloc units fit | |
13394 | // into 2 max blobs | |
13395 | size_t blob_threshold = max_blob_size / min_alloc_size * 2 + 1; | |
13396 | bool above_blob_threshold = false; | |
13397 | ||
13398 | inspected_blobs.reserve(blob_threshold); | |
13399 | ||
13400 | uint64_t max_off = 0; | |
13401 | auto start_ep = ep; | |
13402 | auto end_ep = ep; // exclusively | |
7c673cae FG |
13403 | do { |
13404 | any_change = false; | |
13405 | ||
13406 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13407 | BlobRef b = ep->blob; | |
eafe8130 TL |
13408 | if (!above_blob_threshold) { |
13409 | inspected_blobs.insert(&b->get_blob()); | |
13410 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13411 | } | |
13412 | max_off = ep->logical_end(); | |
7c673cae | 13413 | auto bstart = ep->blob_start(); |
eafe8130 | 13414 | |
7c673cae FG |
13415 | dout(20) << __func__ << " considering " << *b |
13416 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
13417 | if (bstart >= end_offs) { | |
13418 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
13419 | } else if (!b->get_blob().is_mutable()) { | |
13420 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
13421 | } else if (ep->logical_offset % min_alloc_size != | |
13422 | ep->blob_offset % min_alloc_size) { | |
13423 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
13424 | } else { | |
13425 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13426 | // can we pad our head/tail out with zeros? | |
13427 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
13428 | head_pad = p2phase(offset, chunk_size); |
13429 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
13430 | if (head_pad || tail_pad) { |
13431 | o->extent_map.fault_range(db, offset - head_pad, | |
13432 | end_offs - offset + head_pad + tail_pad); | |
13433 | } | |
13434 | if (head_pad && | |
a4b75251 | 13435 | o->extent_map.has_any_lextents(offset - head_pad, head_pad)) { |
7c673cae FG |
13436 | head_pad = 0; |
13437 | } | |
13438 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
13439 | tail_pad = 0; | |
13440 | } | |
13441 | ||
13442 | uint64_t b_off = offset - head_pad - bstart; | |
13443 | uint64_t b_len = length + head_pad + tail_pad; | |
13444 | ||
13445 | // direct write into unused blocks of an existing mutable blob? | |
13446 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
13447 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13448 | b->get_blob().is_unused(b_off, b_len) && | |
13449 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 13450 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13451 | |
13452 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
13453 | << b_off << "~" << b_len | |
13454 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
13455 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 13456 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13457 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13458 | ||
11fdf7f2 | 13459 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
522d829b | 13460 | if (b_len < prefer_deferred_size) { |
7c673cae FG |
13461 | dout(20) << __func__ << " deferring small 0x" << std::hex |
13462 | << b_len << std::dec << " unused write via deferred" << dendl; | |
522d829b | 13463 | bluestore_deferred_op_t *op = _get_deferred_op(txc, bl.length()); |
7c673cae FG |
13464 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13465 | b->get_blob().map( | |
13466 | b_off, b_len, | |
13467 | [&](uint64_t offset, uint64_t length) { | |
13468 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13469 | return 0; | |
13470 | }); | |
224ce89b | 13471 | op->data = bl; |
7c673cae FG |
13472 | } else { |
13473 | b->get_blob().map_bl( | |
224ce89b | 13474 | b_off, bl, |
7c673cae FG |
13475 | [&](uint64_t offset, bufferlist& t) { |
13476 | bdev->aio_write(offset, t, | |
13477 | &txc->ioc, wctx->buffered); | |
13478 | }); | |
13479 | } | |
13480 | } | |
224ce89b | 13481 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
13482 | dout(20) << __func__ << " lex old " << *ep << dendl; |
13483 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
13484 | b, | |
13485 | &wctx->old_extents); | |
13486 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
f67539c2 | 13487 | |
7c673cae FG |
13488 | txc->statfs_delta.stored() += le->length; |
13489 | dout(20) << __func__ << " lex " << *le << dendl; | |
13490 | logger->inc(l_bluestore_write_small_unused); | |
13491 | return; | |
13492 | } | |
13493 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
13494 | uint64_t head_read = p2phase(b_off, chunk_size); |
13495 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
13496 | if ((head_read || tail_read) && |
13497 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
13498 | head_read + tail_read < min_alloc_size) { | |
13499 | b_off -= head_read; | |
13500 | b_len += head_read + tail_read; | |
13501 | ||
13502 | } else { | |
13503 | head_read = tail_read = 0; | |
13504 | } | |
13505 | ||
13506 | // chunk-aligned deferred overwrite? | |
13507 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13508 | b_off % chunk_size == 0 && | |
13509 | b_len % chunk_size == 0 && | |
13510 | b->get_blob().is_allocated(b_off, b_len)) { | |
13511 | ||
224ce89b | 13512 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13513 | |
13514 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
13515 | << " and tail 0x" << tail_read << std::dec << dendl; | |
13516 | if (head_read) { | |
13517 | bufferlist head_bl; | |
13518 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
13519 | head_bl, 0); | |
11fdf7f2 | 13520 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
13521 | size_t zlen = head_read - r; |
13522 | if (zlen) { | |
13523 | head_bl.append_zero(zlen); | |
13524 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13525 | } | |
11fdf7f2 TL |
13526 | head_bl.claim_append(bl); |
13527 | bl.swap(head_bl); | |
7c673cae FG |
13528 | logger->inc(l_bluestore_write_penalty_read_ops); |
13529 | } | |
13530 | if (tail_read) { | |
13531 | bufferlist tail_bl; | |
13532 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
13533 | tail_bl, 0); | |
11fdf7f2 | 13534 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
13535 | size_t zlen = tail_read - r; |
13536 | if (zlen) { | |
13537 | tail_bl.append_zero(zlen); | |
13538 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13539 | } | |
224ce89b | 13540 | bl.claim_append(tail_bl); |
7c673cae FG |
13541 | logger->inc(l_bluestore_write_penalty_read_ops); |
13542 | } | |
f67539c2 | 13543 | logger->inc(l_bluestore_write_small_pre_read); |
7c673cae | 13544 | |
224ce89b | 13545 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13546 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13547 | ||
f67539c2 | 13548 | b->dirty_blob().calc_csum(b_off, bl); |
11fdf7f2 TL |
13549 | |
13550 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
522d829b | 13551 | bluestore_deferred_op_t *op = _get_deferred_op(txc, bl.length()); |
11fdf7f2 TL |
13552 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13553 | int r = b->get_blob().map( | |
13554 | b_off, b_len, | |
13555 | [&](uint64_t offset, uint64_t length) { | |
13556 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13557 | return 0; | |
13558 | }); | |
13559 | ceph_assert(r == 0); | |
f67539c2 | 13560 | op->data = std::move(bl); |
11fdf7f2 TL |
13561 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" |
13562 | << b_len << std::dec << " of mutable " << *b | |
13563 | << " at " << op->extents << dendl; | |
13564 | } | |
13565 | ||
7c673cae FG |
13566 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
13567 | b, &wctx->old_extents); | |
13568 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13569 | txc->statfs_delta.stored() += le->length; | |
13570 | dout(20) << __func__ << " lex " << *le << dendl; | |
7c673cae FG |
13571 | return; |
13572 | } | |
224ce89b WB |
13573 | // try to reuse blob if we can |
13574 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
13575 | max_bsize, |
13576 | offset0 - bstart, | |
13577 | &alloc_len)) { | |
11fdf7f2 | 13578 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13579 | // fit into reused blob |
13580 | // Need to check for pending writes desiring to | |
13581 | // reuse the same pextent. The rationale is that during GC two chunks | |
13582 | // from garbage blobs(compressed?) can share logical space within the same | |
13583 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13584 | // Hence the second write will fail in an attempt to reuse blob at | |
13585 | // do_alloc_write(). | |
13586 | if (!wctx->has_conflict(b, | |
13587 | offset0, | |
13588 | offset0 + alloc_len, | |
13589 | min_alloc_size)) { | |
13590 | ||
13591 | // we can't reuse pad_head/pad_tail since they might be truncated | |
13592 | // due to existent extents | |
13593 | uint64_t b_off = offset - bstart; | |
13594 | uint64_t b_off0 = b_off; | |
13595 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13596 | ||
13597 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13598 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13599 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13600 | << std::dec << dendl; |
13601 | ||
13602 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13603 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13604 | false, false); | |
13605 | logger->inc(l_bluestore_write_small_unused); | |
13606 | return; | |
13607 | } | |
13608 | } | |
13609 | } | |
13610 | ++ep; | |
eafe8130 | 13611 | end_ep = ep; |
7c673cae FG |
13612 | any_change = true; |
13613 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
13614 | ||
13615 | // check extent for reuse in reverse order | |
13616 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
13617 | BlobRef b = prev_ep->blob; | |
eafe8130 TL |
13618 | if (!above_blob_threshold) { |
13619 | inspected_blobs.insert(&b->get_blob()); | |
13620 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13621 | } | |
13622 | start_ep = prev_ep; | |
7c673cae FG |
13623 | auto bstart = prev_ep->blob_start(); |
13624 | dout(20) << __func__ << " considering " << *b | |
13625 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 13626 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
13627 | max_bsize, |
13628 | offset0 - bstart, | |
13629 | &alloc_len)) { | |
11fdf7f2 | 13630 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13631 | // fit into reused blob |
13632 | // Need to check for pending writes desiring to | |
13633 | // reuse the same pextent. The rationale is that during GC two chunks | |
13634 | // from garbage blobs(compressed?) can share logical space within the same | |
13635 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13636 | // Hence the second write will fail in an attempt to reuse blob at | |
13637 | // do_alloc_write(). | |
13638 | if (!wctx->has_conflict(b, | |
13639 | offset0, | |
13640 | offset0 + alloc_len, | |
13641 | min_alloc_size)) { | |
13642 | ||
13643 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13644 | uint64_t b_off = offset - bstart; | |
13645 | uint64_t b_off0 = b_off; | |
13646 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13647 | ||
13648 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13649 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13650 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13651 | << std::dec << dendl; |
13652 | ||
13653 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13654 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13655 | false, false); | |
13656 | logger->inc(l_bluestore_write_small_unused); | |
13657 | return; | |
13658 | } | |
13659 | } | |
13660 | if (prev_ep != begin) { | |
13661 | --prev_ep; | |
13662 | any_change = true; | |
13663 | } else { | |
13664 | prev_ep = end; // to avoid useless first extent re-check | |
13665 | } | |
13666 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
13667 | } while (any_change); | |
13668 | ||
eafe8130 TL |
13669 | if (above_blob_threshold) { |
13670 | dout(10) << __func__ << " request GC, blobs >= " << inspected_blobs.size() | |
13671 | << " " << std::hex << min_off << "~" << max_off << std::dec | |
13672 | << dendl; | |
13673 | ceph_assert(start_ep != end_ep); | |
13674 | for (auto ep = start_ep; ep != end_ep; ++ep) { | |
13675 | dout(20) << __func__ << " inserting for GC " | |
13676 | << std::hex << ep->logical_offset << "~" << ep->length | |
13677 | << std::dec << dendl; | |
13678 | ||
13679 | wctx->extents_to_gc.union_insert(ep->logical_offset, ep->length); | |
13680 | } | |
13681 | // insert newly written extent to GC | |
13682 | wctx->extents_to_gc.union_insert(offset, length); | |
13683 | dout(20) << __func__ << " inserting (last) for GC " | |
13684 | << std::hex << offset << "~" << length | |
13685 | << std::dec << dendl; | |
13686 | } | |
7c673cae | 13687 | // new blob. |
7c673cae | 13688 | BlobRef b = c->new_blob(); |
11fdf7f2 | 13689 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae FG |
13690 | uint64_t b_off0 = b_off; |
13691 | _pad_zeros(&bl, &b_off0, block_size); | |
13692 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
1911f103 TL |
13693 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, |
13694 | min_alloc_size != block_size, // use 'unused' bitmap when alloc granularity | |
13695 | // doesn't match disk one only | |
13696 | true); | |
7c673cae FG |
13697 | |
13698 | return; | |
13699 | } | |
13700 | ||
f67539c2 TL |
13701 | bool BlueStore::BigDeferredWriteContext::can_defer( |
13702 | BlueStore::extent_map_t::iterator ep, | |
13703 | uint64_t prefer_deferred_size, | |
13704 | uint64_t block_size, | |
13705 | uint64_t offset, | |
13706 | uint64_t l) | |
13707 | { | |
13708 | bool res = false; | |
13709 | auto& blob = ep->blob->get_blob(); | |
13710 | if (offset >= ep->blob_start() && | |
13711 | blob.is_mutable()) { | |
13712 | off = offset; | |
13713 | b_off = offset - ep->blob_start(); | |
13714 | uint64_t chunk_size = blob.get_chunk_size(block_size); | |
13715 | uint64_t ondisk = blob.get_ondisk_length(); | |
13716 | used = std::min(l, ondisk - b_off); | |
13717 | ||
13718 | // will read some data to fill out the chunk? | |
13719 | head_read = p2phase<uint64_t>(b_off, chunk_size); | |
13720 | tail_read = p2nphase<uint64_t>(b_off + used, chunk_size); | |
13721 | b_off -= head_read; | |
13722 | ||
13723 | ceph_assert(b_off % chunk_size == 0); | |
13724 | ceph_assert(blob_aligned_len() % chunk_size == 0); | |
13725 | ||
522d829b | 13726 | res = blob_aligned_len() < prefer_deferred_size && |
f67539c2 TL |
13727 | blob_aligned_len() <= ondisk && |
13728 | blob.is_allocated(b_off, blob_aligned_len()); | |
13729 | if (res) { | |
13730 | blob_ref = ep->blob; | |
13731 | blob_start = ep->blob_start(); | |
13732 | } | |
13733 | } | |
13734 | return res; | |
13735 | } | |
13736 | ||
13737 | bool BlueStore::BigDeferredWriteContext::apply_defer() | |
13738 | { | |
13739 | int r = blob_ref->get_blob().map( | |
13740 | b_off, blob_aligned_len(), | |
13741 | [&](const bluestore_pextent_t& pext, | |
13742 | uint64_t offset, | |
13743 | uint64_t length) { | |
13744 | // apply deferred if overwrite breaks blob continuity only. | |
13745 | // if it totally overlaps some pextent - fallback to regular write | |
13746 | if (pext.offset < offset || | |
13747 | pext.end() > offset + length) { | |
13748 | res_extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13749 | return 0; | |
13750 | } | |
13751 | return -1; | |
13752 | }); | |
13753 | return r >= 0; | |
13754 | } | |
13755 | ||
13756 | void BlueStore::_do_write_big_apply_deferred( | |
13757 | TransContext* txc, | |
13758 | CollectionRef& c, | |
13759 | OnodeRef o, | |
13760 | BlueStore::BigDeferredWriteContext& dctx, | |
13761 | bufferlist::iterator& blp, | |
13762 | WriteContext* wctx) | |
13763 | { | |
13764 | bufferlist bl; | |
13765 | dout(20) << __func__ << " reading head 0x" << std::hex << dctx.head_read | |
13766 | << " and tail 0x" << dctx.tail_read << std::dec << dendl; | |
13767 | if (dctx.head_read) { | |
13768 | int r = _do_read(c.get(), o, | |
13769 | dctx.off - dctx.head_read, | |
13770 | dctx.head_read, | |
13771 | bl, | |
13772 | 0); | |
13773 | ceph_assert(r >= 0 && r <= (int)dctx.head_read); | |
13774 | size_t zlen = dctx.head_read - r; | |
13775 | if (zlen) { | |
13776 | bl.append_zero(zlen); | |
13777 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13778 | } | |
13779 | logger->inc(l_bluestore_write_penalty_read_ops); | |
13780 | } | |
13781 | blp.copy(dctx.used, bl); | |
13782 | ||
13783 | if (dctx.tail_read) { | |
13784 | bufferlist tail_bl; | |
13785 | int r = _do_read(c.get(), o, | |
13786 | dctx.off + dctx.used, dctx.tail_read, | |
13787 | tail_bl, 0); | |
13788 | ceph_assert(r >= 0 && r <= (int)dctx.tail_read); | |
13789 | size_t zlen = dctx.tail_read - r; | |
13790 | if (zlen) { | |
13791 | tail_bl.append_zero(zlen); | |
13792 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13793 | } | |
13794 | bl.claim_append(tail_bl); | |
13795 | logger->inc(l_bluestore_write_penalty_read_ops); | |
13796 | } | |
13797 | auto& b0 = dctx.blob_ref; | |
13798 | _buffer_cache_write(txc, b0, dctx.b_off, bl, | |
13799 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
13800 | ||
13801 | b0->dirty_blob().calc_csum(dctx.b_off, bl); | |
13802 | ||
13803 | Extent* le = o->extent_map.set_lextent(c, dctx.off, | |
13804 | dctx.off - dctx.blob_start, dctx.used, b0, &wctx->old_extents); | |
13805 | ||
13806 | // in fact this is a no-op for big writes but left here to maintain | |
13807 | // uniformity and avoid missing after some refactor. | |
13808 | b0->dirty_blob().mark_used(le->blob_offset, le->length); | |
13809 | txc->statfs_delta.stored() += le->length; | |
13810 | ||
13811 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
522d829b | 13812 | bluestore_deferred_op_t* op = _get_deferred_op(txc, bl.length()); |
f67539c2 TL |
13813 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13814 | op->extents.swap(dctx.res_extents); | |
13815 | op->data = std::move(bl); | |
13816 | } | |
13817 | } | |
13818 | ||
7c673cae FG |
13819 | void BlueStore::_do_write_big( |
13820 | TransContext *txc, | |
13821 | CollectionRef &c, | |
13822 | OnodeRef o, | |
13823 | uint64_t offset, uint64_t length, | |
13824 | bufferlist::iterator& blp, | |
13825 | WriteContext *wctx) | |
13826 | { | |
13827 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13828 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
13829 | << " compress " << (int)wctx->compress | |
13830 | << dendl; | |
13831 | logger->inc(l_bluestore_write_big); | |
13832 | logger->inc(l_bluestore_write_big_bytes, length); | |
11fdf7f2 | 13833 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
f67539c2 | 13834 | uint64_t prefer_deferred_size_snapshot = prefer_deferred_size.load(); |
7c673cae FG |
13835 | while (length > 0) { |
13836 | bool new_blob = false; | |
7c673cae FG |
13837 | BlobRef b; |
13838 | uint32_t b_off = 0; | |
522d829b | 13839 | uint32_t l = 0; |
7c673cae FG |
13840 | |
13841 | //attempting to reuse existing blob | |
13842 | if (!wctx->compress) { | |
522d829b TL |
13843 | // enforce target blob alignment with max_bsize |
13844 | l = max_bsize - p2phase(offset, max_bsize); | |
13845 | l = std::min(uint64_t(l), length); | |
13846 | ||
7c673cae | 13847 | auto end = o->extent_map.extent_map.end(); |
f67539c2 | 13848 | |
522d829b TL |
13849 | dout(20) << __func__ << " may be defer: 0x" << std::hex |
13850 | << offset << "~" << l | |
13851 | << std::dec << dendl; | |
13852 | ||
f67539c2 TL |
13853 | if (prefer_deferred_size_snapshot && |
13854 | l <= prefer_deferred_size_snapshot * 2) { | |
13855 | // Single write that spans two adjusted existing blobs can result | |
13856 | // in up to two deferred blocks of 'prefer_deferred_size' | |
13857 | // So we're trying to minimize the amount of resulting blobs | |
13858 | // and preserve 2 blobs rather than inserting one more in between | |
13859 | // E.g. write 0x10000~20000 over existing blobs | |
13860 | // (0x0~20000 and 0x20000~20000) is better (from subsequent reading | |
13861 | // performance point of view) to result in two deferred writes to | |
13862 | // existing blobs than having 3 blobs: 0x0~10000, 0x10000~20000, 0x30000~10000 | |
13863 | ||
13864 | // look for an existing mutable blob we can write into | |
13865 | auto ep = o->extent_map.seek_lextent(offset); | |
13866 | auto ep_next = end; | |
13867 | BigDeferredWriteContext head_info, tail_info; | |
13868 | ||
13869 | bool will_defer = ep != end ? | |
13870 | head_info.can_defer(ep, | |
13871 | prefer_deferred_size_snapshot, | |
13872 | block_size, | |
13873 | offset, | |
13874 | l) : | |
13875 | false; | |
13876 | auto offset_next = offset + head_info.used; | |
13877 | auto remaining = l - head_info.used; | |
13878 | if (will_defer && remaining) { | |
13879 | will_defer = false; | |
13880 | if (remaining <= prefer_deferred_size_snapshot) { | |
13881 | ep_next = o->extent_map.seek_lextent(offset_next); | |
13882 | // check if we can defer remaining totally | |
13883 | will_defer = ep_next == end ? | |
13884 | false : | |
13885 | tail_info.can_defer(ep_next, | |
13886 | prefer_deferred_size_snapshot, | |
13887 | block_size, | |
13888 | offset_next, | |
13889 | remaining); | |
13890 | will_defer = will_defer && remaining == tail_info.used; | |
13891 | } | |
13892 | } | |
13893 | if (will_defer) { | |
13894 | dout(20) << __func__ << " " << *(head_info.blob_ref) | |
13895 | << " deferring big " << std::hex | |
13896 | << " (0x" << head_info.b_off << "~" << head_info.blob_aligned_len() << ")" | |
13897 | << std::dec << " write via deferred" | |
13898 | << dendl; | |
13899 | if (remaining) { | |
13900 | dout(20) << __func__ << " " << *(tail_info.blob_ref) | |
13901 | << " deferring big " << std::hex | |
13902 | << " (0x" << tail_info.b_off << "~" << tail_info.blob_aligned_len() << ")" | |
13903 | << std::dec << " write via deferred" | |
13904 | << dendl; | |
13905 | } | |
13906 | ||
13907 | will_defer = head_info.apply_defer(); | |
13908 | if (!will_defer) { | |
13909 | dout(20) << __func__ | |
13910 | << " deferring big fell back, head isn't continuous" | |
13911 | << dendl; | |
13912 | } else if (remaining) { | |
13913 | will_defer = tail_info.apply_defer(); | |
13914 | if (!will_defer) { | |
13915 | dout(20) << __func__ | |
13916 | << " deferring big fell back, tail isn't continuous" | |
13917 | << dendl; | |
13918 | } | |
13919 | } | |
13920 | } | |
13921 | if (will_defer) { | |
13922 | _do_write_big_apply_deferred(txc, c, o, head_info, blp, wctx); | |
13923 | if (remaining) { | |
13924 | _do_write_big_apply_deferred(txc, c, o, tail_info, | |
13925 | blp, wctx); | |
13926 | } | |
522d829b TL |
13927 | dout(20) << __func__ << " defer big: 0x" << std::hex |
13928 | << offset << "~" << l | |
13929 | << std::dec << dendl; | |
f67539c2 TL |
13930 | offset += l; |
13931 | length -= l; | |
13932 | logger->inc(l_bluestore_write_big_blobs, remaining ? 2 : 1); | |
13933 | logger->inc(l_bluestore_write_big_deferred, remaining ? 2 : 1); | |
13934 | continue; | |
13935 | } | |
13936 | } | |
522d829b | 13937 | dout(20) << __func__ << " lookup for blocks to reuse..." << dendl; |
f67539c2 TL |
13938 | |
13939 | o->extent_map.punch_hole(c, offset, l, &wctx->old_extents); | |
13940 | ||
13941 | // seek again as punch_hole could invalidate ep | |
7c673cae | 13942 | auto ep = o->extent_map.seek_lextent(offset); |
f67539c2 TL |
13943 | auto begin = o->extent_map.extent_map.begin(); |
13944 | auto prev_ep = end; | |
13945 | if (ep != begin) { | |
13946 | prev_ep = ep; | |
7c673cae | 13947 | --prev_ep; |
7c673cae | 13948 | } |
f67539c2 | 13949 | |
7c673cae FG |
13950 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; |
13951 | // search suitable extent in both forward and reverse direction in | |
13952 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 13953 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
13954 | bool any_change; |
13955 | do { | |
13956 | any_change = false; | |
13957 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
522d829b TL |
13958 | dout(20) << __func__ << " considering " << *ep |
13959 | << " bstart 0x" << std::hex << ep->blob_start() << std::dec << dendl; | |
f67539c2 TL |
13960 | |
13961 | if (offset >= ep->blob_start() && | |
224ce89b | 13962 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13963 | offset - ep->blob_start(), |
13964 | &l)) { | |
13965 | b = ep->blob; | |
f67539c2 | 13966 | b_off = offset - ep->blob_start(); |
7c673cae FG |
13967 | prev_ep = end; // to avoid check below |
13968 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13969 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13970 | } else { |
13971 | ++ep; | |
13972 | any_change = true; | |
13973 | } | |
13974 | } | |
13975 | ||
13976 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
522d829b TL |
13977 | dout(20) << __func__ << " considering rev " << *prev_ep |
13978 | << " bstart 0x" << std::hex << prev_ep->blob_start() << std::dec << dendl; | |
f67539c2 | 13979 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13980 | offset - prev_ep->blob_start(), |
13981 | &l)) { | |
13982 | b = prev_ep->blob; | |
13983 | b_off = offset - prev_ep->blob_start(); | |
13984 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13985 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13986 | } else if (prev_ep != begin) { |
13987 | --prev_ep; | |
13988 | any_change = true; | |
13989 | } else { | |
13990 | prev_ep = end; // to avoid useless first extent re-check | |
13991 | } | |
13992 | } | |
13993 | } while (b == nullptr && any_change); | |
f67539c2 | 13994 | } else { |
522d829b TL |
13995 | // trying to utilize as longer chunk as permitted in case of compression. |
13996 | l = std::min(max_bsize, length); | |
f67539c2 TL |
13997 | o->extent_map.punch_hole(c, offset, l, &wctx->old_extents); |
13998 | } // if (!wctx->compress) | |
13999 | ||
7c673cae FG |
14000 | if (b == nullptr) { |
14001 | b = c->new_blob(); | |
14002 | b_off = 0; | |
14003 | new_blob = true; | |
14004 | } | |
7c673cae FG |
14005 | bufferlist t; |
14006 | blp.copy(l, t); | |
14007 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); | |
522d829b TL |
14008 | dout(20) << __func__ << " schedule write big: 0x" |
14009 | << std::hex << offset << "~" << l << std::dec | |
14010 | << (new_blob ? " new " : " reuse ") | |
14011 | << *b << dendl; | |
7c673cae FG |
14012 | offset += l; |
14013 | length -= l; | |
14014 | logger->inc(l_bluestore_write_big_blobs); | |
14015 | } | |
14016 | } | |
14017 | ||
14018 | int BlueStore::_do_alloc_write( | |
14019 | TransContext *txc, | |
14020 | CollectionRef coll, | |
14021 | OnodeRef o, | |
14022 | WriteContext *wctx) | |
14023 | { | |
14024 | dout(20) << __func__ << " txc " << txc | |
14025 | << " " << wctx->writes.size() << " blobs" | |
14026 | << dendl; | |
3efd9988 FG |
14027 | if (wctx->writes.empty()) { |
14028 | return 0; | |
7c673cae FG |
14029 | } |
14030 | ||
7c673cae FG |
14031 | CompressorRef c; |
14032 | double crr = 0; | |
14033 | if (wctx->compress) { | |
14034 | c = select_option( | |
14035 | "compression_algorithm", | |
14036 | compressor, | |
14037 | [&]() { | |
14038 | string val; | |
14039 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
14040 | CompressorRef cp = compressor; | |
14041 | if (!cp || cp->get_type_name() != val) { | |
14042 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
14043 | if (!cp) { |
14044 | if (_set_compression_alert(false, val.c_str())) { | |
14045 | derr << __func__ << " unable to initialize " << val.c_str() | |
14046 | << " compressor" << dendl; | |
14047 | } | |
14048 | } | |
7c673cae FG |
14049 | } |
14050 | return boost::optional<CompressorRef>(cp); | |
14051 | } | |
14052 | return boost::optional<CompressorRef>(); | |
14053 | } | |
14054 | ); | |
14055 | ||
14056 | crr = select_option( | |
14057 | "compression_required_ratio", | |
14058 | cct->_conf->bluestore_compression_required_ratio, | |
14059 | [&]() { | |
14060 | double val; | |
3efd9988 | 14061 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
7c673cae FG |
14062 | return boost::optional<double>(val); |
14063 | } | |
14064 | return boost::optional<double>(); | |
14065 | } | |
14066 | ); | |
14067 | } | |
14068 | ||
14069 | // checksum | |
11fdf7f2 | 14070 | int64_t csum = csum_type.load(); |
7c673cae FG |
14071 | csum = select_option( |
14072 | "csum_type", | |
14073 | csum, | |
14074 | [&]() { | |
11fdf7f2 | 14075 | int64_t val; |
3efd9988 | 14076 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
11fdf7f2 | 14077 | return boost::optional<int64_t>(val); |
7c673cae | 14078 | } |
11fdf7f2 | 14079 | return boost::optional<int64_t>(); |
7c673cae FG |
14080 | } |
14081 | ); | |
14082 | ||
3efd9988 FG |
14083 | // compress (as needed) and calc needed space |
14084 | uint64_t need = 0; | |
11fdf7f2 | 14085 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 14086 | for (auto& wi : wctx->writes) { |
3efd9988 | 14087 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 14088 | auto start = mono_clock::now(); |
7c673cae FG |
14089 | |
14090 | // compress | |
11fdf7f2 TL |
14091 | ceph_assert(wi.b_off == 0); |
14092 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 14093 | |
7c673cae FG |
14094 | // FIXME: memory alignment here is bad |
14095 | bufferlist t; | |
f67539c2 TL |
14096 | boost::optional<int32_t> compressor_message; |
14097 | int r = c->compress(wi.bl, t, compressor_message); | |
3efd9988 | 14098 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 14099 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
14100 | bool rejected = false; |
14101 | uint64_t compressed_len = t.length(); | |
14102 | // do an approximate (fast) estimation for resulting blob size | |
14103 | // that doesn't take header overhead into account | |
11fdf7f2 | 14104 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
14105 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
14106 | bluestore_compression_header_t chdr; | |
14107 | chdr.type = c->get_type(); | |
14108 | chdr.length = t.length(); | |
f67539c2 | 14109 | chdr.compressor_message = compressor_message; |
a8e16298 TL |
14110 | encode(chdr, wi.compressed_bl); |
14111 | wi.compressed_bl.claim_append(t); | |
14112 | ||
14113 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 14114 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
14115 | if (result_len <= want_len && result_len < wi.blob_length) { |
14116 | // Cool. We compressed at least as much as we were hoping to. | |
14117 | // pad out to min_alloc_size | |
14118 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
14119 | wi.compressed_len = compressed_len; | |
14120 | wi.compressed = true; | |
14121 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
14122 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
14123 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
14124 | << " with " << c->get_type() | |
14125 | << std::dec << dendl; | |
14126 | txc->statfs_delta.compressed() += compressed_len; | |
14127 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
14128 | txc->statfs_delta.compressed_allocated() += result_len; | |
14129 | logger->inc(l_bluestore_compress_success_count); | |
14130 | need += result_len; | |
14131 | } else { | |
14132 | rejected = true; | |
14133 | } | |
14134 | } else if (r != 0) { | |
14135 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
14136 | << " bytes compressed using " << c->get_type_name() | |
14137 | << std::dec | |
14138 | << " failed with errcode = " << r | |
14139 | << ", leaving uncompressed" | |
14140 | << dendl; | |
14141 | logger->inc(l_bluestore_compress_rejected_count); | |
14142 | need += wi.blob_length; | |
7c673cae | 14143 | } else { |
a8e16298 TL |
14144 | rejected = true; |
14145 | } | |
14146 | ||
14147 | if (rejected) { | |
3efd9988 | 14148 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 14149 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
14150 | << " with " << c->get_type() |
14151 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 14152 | << " -> 0x" << want_len |
3efd9988 FG |
14153 | << ", leaving uncompressed" |
14154 | << std::dec << dendl; | |
14155 | logger->inc(l_bluestore_compress_rejected_count); | |
14156 | need += wi.blob_length; | |
7c673cae | 14157 | } |
494da23a TL |
14158 | log_latency("compress@_do_alloc_write", |
14159 | l_bluestore_compress_lat, | |
14160 | mono_clock::now() - start, | |
14161 | cct->_conf->bluestore_log_op_age ); | |
3efd9988 FG |
14162 | } else { |
14163 | need += wi.blob_length; | |
7c673cae | 14164 | } |
3efd9988 | 14165 | } |
a8e16298 | 14166 | PExtentVector prealloc; |
3efd9988 | 14167 | prealloc.reserve(2 * wctx->writes.size());; |
11fdf7f2 | 14168 | int64_t prealloc_left = 0; |
f67539c2 | 14169 | prealloc_left = shared_alloc.a->allocate( |
3efd9988 FG |
14170 | need, min_alloc_size, need, |
14171 | 0, &prealloc); | |
eafe8130 | 14172 | if (prealloc_left < 0 || prealloc_left < (int64_t)need) { |
11fdf7f2 | 14173 | derr << __func__ << " failed to allocate 0x" << std::hex << need |
eafe8130 | 14174 | << " allocated 0x " << (prealloc_left < 0 ? 0 : prealloc_left) |
11fdf7f2 | 14175 | << " min_alloc_size 0x" << min_alloc_size |
f67539c2 | 14176 | << " available 0x " << shared_alloc.a->get_free() |
11fdf7f2 TL |
14177 | << std::dec << dendl; |
14178 | if (prealloc.size()) { | |
f67539c2 | 14179 | shared_alloc.a->release(prealloc); |
11fdf7f2 | 14180 | } |
a8e16298 TL |
14181 | return -ENOSPC; |
14182 | } | |
9f95a23c | 14183 | _collect_allocation_stats(need, min_alloc_size, prealloc.size()); |
a8e16298 | 14184 | |
f67539c2 TL |
14185 | if (bdev->is_smr()) { |
14186 | std::deque<uint64_t> zones_to_clean; | |
14187 | if (shared_alloc.a->zoned_get_zones_to_clean(&zones_to_clean)) { | |
14188 | std::lock_guard l{zoned_cleaner_lock}; | |
14189 | zoned_cleaner_queue.swap(zones_to_clean); | |
14190 | zoned_cleaner_cond.notify_one(); | |
14191 | } | |
14192 | } | |
14193 | ||
3efd9988 FG |
14194 | dout(20) << __func__ << " prealloc " << prealloc << dendl; |
14195 | auto prealloc_pos = prealloc.begin(); | |
522d829b TL |
14196 | ceph_assert(prealloc_pos != prealloc.end()); |
14197 | uint64_t prealloc_pos_length = prealloc_pos->length; | |
3efd9988 FG |
14198 | |
14199 | for (auto& wi : wctx->writes) { | |
522d829b | 14200 | bluestore_blob_t& dblob = wi.b->dirty_blob(); |
3efd9988 FG |
14201 | uint64_t b_off = wi.b_off; |
14202 | bufferlist *l = &wi.bl; | |
14203 | uint64_t final_length = wi.blob_length; | |
14204 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
14205 | if (wi.compressed) { |
14206 | final_length = wi.compressed_bl.length(); | |
14207 | csum_length = final_length; | |
adb31ebb | 14208 | unsigned csum_order = ctz(csum_length); |
3efd9988 FG |
14209 | l = &wi.compressed_bl; |
14210 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
adb31ebb | 14211 | if (csum != Checksummer::CSUM_NONE) { |
522d829b TL |
14212 | dout(20) << __func__ |
14213 | << " initialize csum setting for compressed blob " << *wi.b | |
adb31ebb TL |
14214 | << " csum_type " << Checksummer::get_csum_type_string(csum) |
14215 | << " csum_order " << csum_order | |
14216 | << " csum_length 0x" << std::hex << csum_length | |
14217 | << " blob_length 0x" << wi.blob_length | |
14218 | << " compressed_length 0x" << wi.compressed_len << std::dec | |
14219 | << dendl; | |
14220 | dblob.init_csum(csum, csum_order, csum_length); | |
14221 | } | |
3efd9988 | 14222 | } else if (wi.new_blob) { |
adb31ebb | 14223 | unsigned csum_order; |
7c673cae | 14224 | // initialize newly created blob only |
11fdf7f2 | 14225 | ceph_assert(dblob.is_mutable()); |
7c673cae FG |
14226 | if (l->length() != wi.blob_length) { |
14227 | // hrm, maybe we could do better here, but let's not bother. | |
14228 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
14229 | << block_size_order << dendl; | |
31f18b77 | 14230 | csum_order = block_size_order; |
7c673cae FG |
14231 | } else { |
14232 | csum_order = std::min(wctx->csum_order, ctz(l->length())); | |
14233 | } | |
14234 | // try to align blob with max_blob_size to improve | |
14235 | // its reuse ratio, e.g. in case of reverse write | |
14236 | uint32_t suggested_boff = | |
14237 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
14238 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
14239 | suggested_boff + final_length <= max_bsize && | |
14240 | suggested_boff > b_off) { | |
181888fb | 14241 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 14242 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 14243 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
14244 | csum_length += suggested_boff - b_off; |
14245 | b_off = suggested_boff; | |
14246 | } | |
181888fb | 14247 | if (csum != Checksummer::CSUM_NONE) { |
522d829b TL |
14248 | dout(20) << __func__ |
14249 | << " initialize csum setting for new blob " << *wi.b | |
181888fb FG |
14250 | << " csum_type " << Checksummer::get_csum_type_string(csum) |
14251 | << " csum_order " << csum_order | |
14252 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
14253 | << dendl; | |
14254 | dblob.init_csum(csum, csum_order, csum_length); | |
14255 | } | |
7c673cae FG |
14256 | } |
14257 | ||
a8e16298 | 14258 | PExtentVector extents; |
3efd9988 | 14259 | int64_t left = final_length; |
522d829b TL |
14260 | bool has_chunk2defer = false; |
14261 | auto prefer_deferred_size_snapshot = prefer_deferred_size.load(); | |
3efd9988 | 14262 | while (left > 0) { |
11fdf7f2 | 14263 | ceph_assert(prealloc_left > 0); |
522d829b | 14264 | has_chunk2defer |= (prealloc_pos_length < prefer_deferred_size_snapshot); |
3efd9988 FG |
14265 | if (prealloc_pos->length <= left) { |
14266 | prealloc_left -= prealloc_pos->length; | |
14267 | left -= prealloc_pos->length; | |
14268 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
14269 | extents.push_back(*prealloc_pos); | |
14270 | ++prealloc_pos; | |
522d829b TL |
14271 | if (prealloc_pos != prealloc.end()) { |
14272 | prealloc_pos_length = prealloc_pos->length; | |
14273 | } | |
3efd9988 FG |
14274 | } else { |
14275 | extents.emplace_back(prealloc_pos->offset, left); | |
14276 | prealloc_pos->offset += left; | |
14277 | prealloc_pos->length -= left; | |
14278 | prealloc_left -= left; | |
14279 | txc->statfs_delta.allocated() += left; | |
14280 | left = 0; | |
14281 | break; | |
14282 | } | |
14283 | } | |
7c673cae | 14284 | for (auto& p : extents) { |
3efd9988 | 14285 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 14286 | } |
11fdf7f2 | 14287 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 14288 | |
522d829b | 14289 | dout(20) << __func__ << " blob " << *wi.b << dendl; |
181888fb | 14290 | if (dblob.has_csum()) { |
7c673cae FG |
14291 | dblob.calc_csum(b_off, *l); |
14292 | } | |
181888fb | 14293 | |
7c673cae | 14294 | if (wi.mark_unused) { |
1911f103 | 14295 | ceph_assert(!dblob.is_compressed()); |
7c673cae FG |
14296 | auto b_end = b_off + wi.bl.length(); |
14297 | if (b_off) { | |
14298 | dblob.add_unused(0, b_off); | |
14299 | } | |
1911f103 TL |
14300 | uint64_t llen = dblob.get_logical_length(); |
14301 | if (b_end < llen) { | |
14302 | dblob.add_unused(b_end, llen - b_end); | |
7c673cae FG |
14303 | } |
14304 | } | |
14305 | ||
14306 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
14307 | b_off + (wi.b_off0 - wi.b_off), | |
14308 | wi.length0, | |
14309 | wi.b, | |
14310 | nullptr); | |
14311 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
14312 | txc->statfs_delta.stored() += le->length; | |
14313 | dout(20) << __func__ << " lex " << *le << dendl; | |
14314 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
14315 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
14316 | ||
14317 | // queue io | |
11fdf7f2 | 14318 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
522d829b | 14319 | if (has_chunk2defer && l->length() < prefer_deferred_size_snapshot) { |
f67539c2 | 14320 | dout(20) << __func__ << " deferring 0x" << std::hex |
7c673cae | 14321 | << l->length() << std::dec << " write via deferred" << dendl; |
522d829b | 14322 | bluestore_deferred_op_t *op = _get_deferred_op(txc, l->length()); |
7c673cae | 14323 | op->op = bluestore_deferred_op_t::OP_WRITE; |
522d829b | 14324 | int r = wi.b->get_blob().map( |
7c673cae FG |
14325 | b_off, l->length(), |
14326 | [&](uint64_t offset, uint64_t length) { | |
14327 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
14328 | return 0; | |
14329 | }); | |
11fdf7f2 | 14330 | ceph_assert(r == 0); |
7c673cae FG |
14331 | op->data = *l; |
14332 | } else { | |
522d829b | 14333 | wi.b->get_blob().map_bl( |
7c673cae FG |
14334 | b_off, *l, |
14335 | [&](uint64_t offset, bufferlist& t) { | |
14336 | bdev->aio_write(offset, t, &txc->ioc, false); | |
14337 | }); | |
f67539c2 | 14338 | logger->inc(l_bluestore_write_new); |
7c673cae FG |
14339 | } |
14340 | } | |
14341 | } | |
11fdf7f2 TL |
14342 | ceph_assert(prealloc_pos == prealloc.end()); |
14343 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
14344 | return 0; |
14345 | } | |
14346 | ||
14347 | void BlueStore::_wctx_finish( | |
14348 | TransContext *txc, | |
14349 | CollectionRef& c, | |
14350 | OnodeRef o, | |
31f18b77 FG |
14351 | WriteContext *wctx, |
14352 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14353 | { |
14354 | auto oep = wctx->old_extents.begin(); | |
14355 | while (oep != wctx->old_extents.end()) { | |
14356 | auto &lo = *oep; | |
14357 | oep = wctx->old_extents.erase(oep); | |
14358 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
14359 | BlobRef b = lo.e.blob; | |
14360 | const bluestore_blob_t& blob = b->get_blob(); | |
14361 | if (blob.is_compressed()) { | |
14362 | if (lo.blob_empty) { | |
14363 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
14364 | } | |
14365 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
14366 | } | |
14367 | auto& r = lo.r; | |
14368 | txc->statfs_delta.stored() -= lo.e.length; | |
14369 | if (!r.empty()) { | |
f67539c2 | 14370 | dout(20) << __func__ << " blob " << *b << " release " << r << dendl; |
7c673cae FG |
14371 | if (blob.is_shared()) { |
14372 | PExtentVector final; | |
14373 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
14374 | bool unshare = false; |
14375 | bool* unshare_ptr = | |
14376 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 14377 | for (auto e : r) { |
31f18b77 FG |
14378 | b->shared_blob->put_ref( |
14379 | e.offset, e.length, &final, | |
11fdf7f2 TL |
14380 | unshare_ptr); |
14381 | } | |
14382 | if (unshare) { | |
14383 | ceph_assert(maybe_unshared_blobs); | |
14384 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
14385 | } |
14386 | dout(20) << __func__ << " shared_blob release " << final | |
14387 | << " from " << *b->shared_blob << dendl; | |
14388 | txc->write_shared_blob(b->shared_blob); | |
14389 | r.clear(); | |
14390 | r.swap(final); | |
14391 | } | |
14392 | } | |
14393 | // we can't invalidate our logical extents as we drop them because | |
14394 | // other lextents (either in our onode or others) may still | |
14395 | // reference them. but we can throw out anything that is no | |
14396 | // longer allocated. Note that this will leave behind edge bits | |
14397 | // that are no longer referenced but not deallocated (until they | |
14398 | // age out of the cache naturally). | |
14399 | b->discard_unallocated(c.get()); | |
14400 | for (auto e : r) { | |
14401 | dout(20) << __func__ << " release " << e << dendl; | |
14402 | txc->released.insert(e.offset, e.length); | |
14403 | txc->statfs_delta.allocated() -= e.length; | |
14404 | if (blob.is_compressed()) { | |
14405 | txc->statfs_delta.compressed_allocated() -= e.length; | |
14406 | } | |
14407 | } | |
9f95a23c TL |
14408 | |
14409 | if (b->is_spanning() && !b->is_referenced() && lo.blob_empty) { | |
7c673cae FG |
14410 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b |
14411 | << dendl; | |
14412 | o->extent_map.spanning_blob_map.erase(b->id); | |
14413 | } | |
9f95a23c | 14414 | delete &lo; |
7c673cae FG |
14415 | } |
14416 | } | |
14417 | ||
14418 | void BlueStore::_do_write_data( | |
14419 | TransContext *txc, | |
14420 | CollectionRef& c, | |
14421 | OnodeRef o, | |
14422 | uint64_t offset, | |
14423 | uint64_t length, | |
14424 | bufferlist& bl, | |
14425 | WriteContext *wctx) | |
14426 | { | |
14427 | uint64_t end = offset + length; | |
14428 | bufferlist::iterator p = bl.begin(); | |
14429 | ||
14430 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
14431 | (length != min_alloc_size)) { | |
14432 | // we fall within the same block | |
14433 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
14434 | } else { | |
14435 | uint64_t head_offset, head_length; | |
14436 | uint64_t middle_offset, middle_length; | |
14437 | uint64_t tail_offset, tail_length; | |
14438 | ||
14439 | head_offset = offset; | |
11fdf7f2 | 14440 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 14441 | |
11fdf7f2 TL |
14442 | tail_offset = p2align(end, min_alloc_size); |
14443 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
14444 | |
14445 | middle_offset = head_offset + head_length; | |
14446 | middle_length = length - head_length - tail_length; | |
14447 | ||
14448 | if (head_length) { | |
14449 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
14450 | } | |
14451 | ||
f67539c2 | 14452 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); |
7c673cae FG |
14453 | |
14454 | if (tail_length) { | |
14455 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
14456 | } | |
14457 | } | |
14458 | } | |
14459 | ||
31f18b77 FG |
14460 | void BlueStore::_choose_write_options( |
14461 | CollectionRef& c, | |
14462 | OnodeRef o, | |
14463 | uint32_t fadvise_flags, | |
14464 | WriteContext *wctx) | |
7c673cae | 14465 | { |
7c673cae FG |
14466 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
14467 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 14468 | wctx->buffered = true; |
7c673cae FG |
14469 | } else if (cct->_conf->bluestore_default_buffered_write && |
14470 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
14471 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
14472 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 14473 | wctx->buffered = true; |
7c673cae FG |
14474 | } |
14475 | ||
31f18b77 FG |
14476 | // apply basic csum block size |
14477 | wctx->csum_order = block_size_order; | |
7c673cae FG |
14478 | |
14479 | // compression parameters | |
14480 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
14481 | auto cm = select_option( | |
14482 | "compression_mode", | |
31f18b77 | 14483 | comp_mode.load(), |
7c673cae FG |
14484 | [&]() { |
14485 | string val; | |
11fdf7f2 | 14486 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
31f18b77 FG |
14487 | return boost::optional<Compressor::CompressionMode>( |
14488 | Compressor::get_comp_mode_type(val)); | |
7c673cae FG |
14489 | } |
14490 | return boost::optional<Compressor::CompressionMode>(); | |
14491 | } | |
14492 | ); | |
31f18b77 FG |
14493 | |
14494 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
14495 | ((cm == Compressor::COMP_FORCE) || |
14496 | (cm == Compressor::COMP_AGGRESSIVE && | |
14497 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
14498 | (cm == Compressor::COMP_PASSIVE && | |
14499 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
14500 | ||
14501 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
14502 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
14503 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
14504 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 14505 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 14506 | |
7c673cae | 14507 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 14508 | |
7c673cae | 14509 | if (o->onode.expected_write_size) { |
224ce89b | 14510 | wctx->csum_order = std::max(min_alloc_size_order, |
31f18b77 | 14511 | (uint8_t)ctz(o->onode.expected_write_size)); |
7c673cae | 14512 | } else { |
224ce89b | 14513 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
14514 | } |
14515 | ||
31f18b77 FG |
14516 | if (wctx->compress) { |
14517 | wctx->target_blob_size = select_option( | |
7c673cae | 14518 | "compression_max_blob_size", |
31f18b77 | 14519 | comp_max_blob_size.load(), |
7c673cae | 14520 | [&]() { |
11fdf7f2 TL |
14521 | int64_t val; |
14522 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
7c673cae FG |
14523 | return boost::optional<uint64_t>((uint64_t)val); |
14524 | } | |
14525 | return boost::optional<uint64_t>(); | |
14526 | } | |
14527 | ); | |
14528 | } | |
14529 | } else { | |
31f18b77 FG |
14530 | if (wctx->compress) { |
14531 | wctx->target_blob_size = select_option( | |
7c673cae | 14532 | "compression_min_blob_size", |
31f18b77 | 14533 | comp_min_blob_size.load(), |
7c673cae | 14534 | [&]() { |
11fdf7f2 TL |
14535 | int64_t val; |
14536 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
7c673cae FG |
14537 | return boost::optional<uint64_t>((uint64_t)val); |
14538 | } | |
14539 | return boost::optional<uint64_t>(); | |
14540 | } | |
14541 | ); | |
14542 | } | |
14543 | } | |
31f18b77 | 14544 | |
7c673cae | 14545 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
14546 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
14547 | wctx->target_blob_size = max_bsize; | |
7c673cae | 14548 | } |
31f18b77 | 14549 | |
7c673cae FG |
14550 | // set the min blob size floor at 2x the min_alloc_size, or else we |
14551 | // won't be able to allocate a smaller extent for the compressed | |
14552 | // data. | |
31f18b77 FG |
14553 | if (wctx->compress && |
14554 | wctx->target_blob_size < min_alloc_size * 2) { | |
14555 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 14556 | } |
31f18b77 FG |
14557 | |
14558 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
14559 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
14560 | << " compress=" << (int)wctx->compress |
14561 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
14562 | << std::dec << dendl; |
14563 | } | |
14564 | ||
14565 | int BlueStore::_do_gc( | |
14566 | TransContext *txc, | |
14567 | CollectionRef& c, | |
14568 | OnodeRef o, | |
31f18b77 FG |
14569 | const WriteContext& wctx, |
14570 | uint64_t *dirty_start, | |
14571 | uint64_t *dirty_end) | |
14572 | { | |
31f18b77 | 14573 | |
1adf2230 | 14574 | bool dirty_range_updated = false; |
31f18b77 | 14575 | WriteContext wctx_gc; |
7c673cae | 14576 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 14577 | |
eafe8130 | 14578 | auto & extents_to_collect = wctx.extents_to_gc; |
31f18b77 FG |
14579 | for (auto it = extents_to_collect.begin(); |
14580 | it != extents_to_collect.end(); | |
14581 | ++it) { | |
14582 | bufferlist bl; | |
eafe8130 TL |
14583 | auto offset = (*it).first; |
14584 | auto length = (*it).second; | |
14585 | dout(20) << __func__ << " processing " << std::hex | |
14586 | << offset << "~" << length << std::dec | |
14587 | << dendl; | |
14588 | int r = _do_read(c.get(), o, offset, length, bl, 0); | |
14589 | ceph_assert(r == (int)length); | |
31f18b77 | 14590 | |
eafe8130 TL |
14591 | _do_write_data(txc, c, o, offset, length, bl, &wctx_gc); |
14592 | logger->inc(l_bluestore_gc_merged, length); | |
31f18b77 | 14593 | |
eafe8130 TL |
14594 | if (*dirty_start > offset) { |
14595 | *dirty_start = offset; | |
1adf2230 | 14596 | dirty_range_updated = true; |
31f18b77 FG |
14597 | } |
14598 | ||
eafe8130 TL |
14599 | if (*dirty_end < offset + length) { |
14600 | *dirty_end = offset + length; | |
1adf2230 | 14601 | dirty_range_updated = true; |
31f18b77 FG |
14602 | } |
14603 | } | |
1adf2230 AA |
14604 | if (dirty_range_updated) { |
14605 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
14606 | } | |
31f18b77 FG |
14607 | |
14608 | dout(30) << __func__ << " alloc write" << dendl; | |
14609 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
14610 | if (r < 0) { | |
14611 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14612 | << dendl; | |
14613 | return r; | |
14614 | } | |
14615 | ||
14616 | _wctx_finish(txc, c, o, &wctx_gc); | |
14617 | return 0; | |
14618 | } | |
14619 | ||
14620 | int BlueStore::_do_write( | |
14621 | TransContext *txc, | |
14622 | CollectionRef& c, | |
14623 | OnodeRef o, | |
14624 | uint64_t offset, | |
14625 | uint64_t length, | |
14626 | bufferlist& bl, | |
14627 | uint32_t fadvise_flags) | |
14628 | { | |
14629 | int r = 0; | |
14630 | ||
14631 | dout(20) << __func__ | |
14632 | << " " << o->oid | |
14633 | << " 0x" << std::hex << offset << "~" << length | |
14634 | << " - have 0x" << o->onode.size | |
14635 | << " (" << std::dec << o->onode.size << ")" | |
f67539c2 TL |
14636 | << " bytes" << std::hex |
14637 | << " fadvise_flags 0x" << fadvise_flags | |
14638 | << " alloc_hint 0x" << o->onode.alloc_hint_flags | |
14639 | << " expected_object_size " << o->onode.expected_object_size | |
14640 | << " expected_write_size " << o->onode.expected_write_size | |
14641 | << std::dec | |
31f18b77 | 14642 | << dendl; |
81eedcae | 14643 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
14644 | |
14645 | if (length == 0) { | |
14646 | return 0; | |
14647 | } | |
14648 | ||
14649 | uint64_t end = offset + length; | |
14650 | ||
14651 | GarbageCollector gc(c->store->cct); | |
eafe8130 | 14652 | int64_t benefit = 0; |
31f18b77 FG |
14653 | auto dirty_start = offset; |
14654 | auto dirty_end = end; | |
14655 | ||
14656 | WriteContext wctx; | |
14657 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
14658 | o->extent_map.fault_range(db, offset, length); |
14659 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
14660 | r = _do_alloc_write(txc, c, o, &wctx); |
14661 | if (r < 0) { | |
14662 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14663 | << dendl; | |
14664 | goto out; | |
14665 | } | |
14666 | ||
eafe8130 TL |
14667 | if (wctx.extents_to_gc.empty() || |
14668 | wctx.extents_to_gc.range_start() > offset || | |
14669 | wctx.extents_to_gc.range_end() < offset + length) { | |
14670 | benefit = gc.estimate(offset, | |
14671 | length, | |
14672 | o->extent_map, | |
14673 | wctx.old_extents, | |
14674 | min_alloc_size); | |
14675 | } | |
14676 | ||
f67539c2 TL |
14677 | if (bdev->is_smr()) { |
14678 | if (wctx.old_extents.empty()) { | |
14679 | txc->zoned_note_new_object(o); | |
14680 | } else { | |
14681 | int64_t old_ondisk_offset = wctx.old_extents.begin()->r.begin()->offset; | |
14682 | txc->zoned_note_updated_object(o, old_ondisk_offset); | |
14683 | } | |
14684 | } | |
14685 | ||
31f18b77 FG |
14686 | // NB: _wctx_finish() will empty old_extents |
14687 | // so we must do gc estimation before that | |
7c673cae FG |
14688 | _wctx_finish(txc, c, o, &wctx); |
14689 | if (end > o->onode.size) { | |
14690 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 14691 | << std::dec << dendl; |
7c673cae FG |
14692 | o->onode.size = end; |
14693 | } | |
14694 | ||
11fdf7f2 | 14695 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
eafe8130 TL |
14696 | wctx.extents_to_gc.union_of(gc.get_extents_to_collect()); |
14697 | dout(20) << __func__ | |
14698 | << " perform garbage collection for compressed extents, " | |
14699 | << "expected benefit = " << benefit << " AUs" << dendl; | |
14700 | } | |
14701 | if (!wctx.extents_to_gc.empty()) { | |
14702 | dout(20) << __func__ << " perform garbage collection" << dendl; | |
14703 | ||
14704 | r = _do_gc(txc, c, o, | |
14705 | wctx, | |
14706 | &dirty_start, &dirty_end); | |
14707 | if (r < 0) { | |
14708 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
14709 | << dendl; | |
14710 | goto out; | |
7c673cae | 14711 | } |
eafe8130 TL |
14712 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
14713 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae | 14714 | } |
7c673cae | 14715 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
14716 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
14717 | ||
7c673cae FG |
14718 | r = 0; |
14719 | ||
14720 | out: | |
14721 | return r; | |
14722 | } | |
14723 | ||
14724 | int BlueStore::_write(TransContext *txc, | |
14725 | CollectionRef& c, | |
14726 | OnodeRef& o, | |
31f18b77 FG |
14727 | uint64_t offset, size_t length, |
14728 | bufferlist& bl, | |
14729 | uint32_t fadvise_flags) | |
7c673cae FG |
14730 | { |
14731 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14732 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14733 | << dendl; | |
35e4c445 FG |
14734 | int r = 0; |
14735 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14736 | r = -E2BIG; | |
14737 | } else { | |
14738 | _assign_nid(txc, o); | |
14739 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
14740 | txc->write_onode(o); | |
14741 | } | |
7c673cae FG |
14742 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14743 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14744 | << " = " << r << dendl; | |
14745 | return r; | |
14746 | } | |
14747 | ||
14748 | int BlueStore::_zero(TransContext *txc, | |
14749 | CollectionRef& c, | |
14750 | OnodeRef& o, | |
14751 | uint64_t offset, size_t length) | |
14752 | { | |
14753 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14754 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14755 | << dendl; | |
35e4c445 FG |
14756 | int r = 0; |
14757 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14758 | r = -E2BIG; | |
14759 | } else { | |
14760 | _assign_nid(txc, o); | |
14761 | r = _do_zero(txc, c, o, offset, length); | |
14762 | } | |
7c673cae FG |
14763 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14764 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14765 | << " = " << r << dendl; | |
14766 | return r; | |
14767 | } | |
14768 | ||
14769 | int BlueStore::_do_zero(TransContext *txc, | |
14770 | CollectionRef& c, | |
14771 | OnodeRef& o, | |
14772 | uint64_t offset, size_t length) | |
14773 | { | |
14774 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14775 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14776 | << dendl; | |
14777 | int r = 0; | |
14778 | ||
81eedcae | 14779 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14780 | |
14781 | WriteContext wctx; | |
14782 | o->extent_map.fault_range(db, offset, length); | |
14783 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 14784 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
14785 | _wctx_finish(txc, c, o, &wctx); |
14786 | ||
b32b8144 | 14787 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
14788 | o->onode.size = offset + length; |
14789 | dout(20) << __func__ << " extending size to " << offset + length | |
14790 | << dendl; | |
14791 | } | |
14792 | txc->write_onode(o); | |
14793 | ||
14794 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14795 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14796 | << " = " << r << dendl; | |
14797 | return r; | |
14798 | } | |
14799 | ||
14800 | void BlueStore::_do_truncate( | |
31f18b77 FG |
14801 | TransContext *txc, CollectionRef& c, OnodeRef o, uint64_t offset, |
14802 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14803 | { |
14804 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14805 | << " 0x" << std::hex << offset << std::dec << dendl; | |
14806 | ||
81eedcae | 14807 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14808 | |
14809 | if (offset == o->onode.size) | |
31f18b77 | 14810 | return; |
7c673cae | 14811 | |
f67539c2 | 14812 | WriteContext wctx; |
7c673cae | 14813 | if (offset < o->onode.size) { |
7c673cae FG |
14814 | uint64_t length = o->onode.size - offset; |
14815 | o->extent_map.fault_range(db, offset, length); | |
14816 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 FG |
14817 | o->extent_map.dirty_range(offset, length); |
14818 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); | |
7c673cae FG |
14819 | |
14820 | // if we have shards past EOF, ask for a reshard | |
14821 | if (!o->onode.extent_map_shards.empty() && | |
14822 | o->onode.extent_map_shards.back().offset >= offset) { | |
14823 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
14824 | if (offset) { | |
14825 | o->extent_map.request_reshard(offset - 1, offset + length); | |
14826 | } else { | |
14827 | o->extent_map.request_reshard(0, length); | |
14828 | } | |
14829 | } | |
14830 | } | |
14831 | ||
14832 | o->onode.size = offset; | |
14833 | ||
f67539c2 TL |
14834 | if (bdev->is_smr()) { |
14835 | // On zoned devices, we currently support only removing an object or | |
14836 | // truncating it to zero size, both of which fall through this code path. | |
14837 | ceph_assert(offset == 0 && !wctx.old_extents.empty()); | |
14838 | int64_t ondisk_offset = wctx.old_extents.begin()->r.begin()->offset; | |
14839 | txc->zoned_note_truncated_object(o, ondisk_offset); | |
14840 | } | |
14841 | ||
7c673cae FG |
14842 | txc->write_onode(o); |
14843 | } | |
14844 | ||
35e4c445 | 14845 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
14846 | CollectionRef& c, |
14847 | OnodeRef& o, | |
14848 | uint64_t offset) | |
14849 | { | |
14850 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14851 | << " 0x" << std::hex << offset << std::dec | |
14852 | << dendl; | |
35e4c445 FG |
14853 | int r = 0; |
14854 | if (offset >= OBJECT_MAX_SIZE) { | |
14855 | r = -E2BIG; | |
14856 | } else { | |
14857 | _do_truncate(txc, c, o, offset); | |
14858 | } | |
14859 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14860 | << " 0x" << std::hex << offset << std::dec | |
14861 | << " = " << r << dendl; | |
14862 | return r; | |
7c673cae FG |
14863 | } |
14864 | ||
14865 | int BlueStore::_do_remove( | |
14866 | TransContext *txc, | |
14867 | CollectionRef& c, | |
14868 | OnodeRef o) | |
14869 | { | |
31f18b77 | 14870 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
14871 | bool is_gen = !o->oid.is_no_gen(); |
14872 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
14873 | if (o->onode.has_omap()) { |
14874 | o->flush(); | |
9f95a23c | 14875 | _do_omap_clear(txc, o); |
7c673cae FG |
14876 | } |
14877 | o->exists = false; | |
14878 | string key; | |
14879 | for (auto &s : o->extent_map.shards) { | |
14880 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
14881 | << s.shard_info->offset << std::dec << dendl; | |
14882 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
14883 | [&](const string& final_key) { | |
14884 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14885 | } | |
14886 | ); | |
14887 | } | |
14888 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 14889 | txc->note_removed_object(o); |
7c673cae FG |
14890 | o->extent_map.clear(); |
14891 | o->onode = bluestore_onode_t(); | |
14892 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 14893 | |
224ce89b WB |
14894 | if (!is_gen || maybe_unshared_blobs.empty()) { |
14895 | return 0; | |
14896 | } | |
31f18b77 | 14897 | |
224ce89b WB |
14898 | // see if we can unshare blobs still referenced by the head |
14899 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
14900 | << maybe_unshared_blobs << dendl; | |
14901 | ghobject_t nogen = o->oid; | |
14902 | nogen.generation = ghobject_t::NO_GEN; | |
f67539c2 | 14903 | OnodeRef h = c->get_onode(nogen, false); |
224ce89b WB |
14904 | |
14905 | if (!h || !h->exists) { | |
14906 | return 0; | |
14907 | } | |
14908 | ||
14909 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
14910 | << " " << h->oid << dendl; | |
14911 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
14912 | for (auto& e : h->extent_map.extent_map) { | |
14913 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14914 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14915 | if (b.is_shared() && | |
14916 | sb->loaded && | |
14917 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
14918 | if (b.is_compressed()) { |
14919 | expect[sb].get(0, b.get_ondisk_length()); | |
14920 | } else { | |
14921 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
14922 | expect[sb].get(off, len); | |
14923 | return 0; | |
14924 | }); | |
14925 | } | |
224ce89b WB |
14926 | } |
14927 | } | |
31f18b77 | 14928 | |
224ce89b WB |
14929 | vector<SharedBlob*> unshared_blobs; |
14930 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
14931 | for (auto& p : expect) { | |
14932 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
14933 | if (p.first->persistent->ref_map == p.second) { | |
14934 | SharedBlob *sb = p.first; | |
14935 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
14936 | unshared_blobs.push_back(sb); | |
14937 | txc->unshare_blob(sb); | |
14938 | uint64_t sbid = c->make_blob_unshared(sb); | |
14939 | string key; | |
14940 | get_shared_blob_key(sbid, &key); | |
14941 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
14942 | } | |
14943 | } | |
14944 | ||
14945 | if (unshared_blobs.empty()) { | |
14946 | return 0; | |
14947 | } | |
14948 | ||
224ce89b WB |
14949 | for (auto& e : h->extent_map.extent_map) { |
14950 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14951 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14952 | if (b.is_shared() && | |
14953 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
14954 | sb) != unshared_blobs.end()) { | |
14955 | dout(20) << __func__ << " unsharing " << e << dendl; | |
14956 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
14957 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 14958 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
14959 | } |
14960 | } | |
224ce89b WB |
14961 | txc->write_onode(h); |
14962 | ||
7c673cae FG |
14963 | return 0; |
14964 | } | |
14965 | ||
14966 | int BlueStore::_remove(TransContext *txc, | |
14967 | CollectionRef& c, | |
14968 | OnodeRef &o) | |
14969 | { | |
11fdf7f2 TL |
14970 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
14971 | << " onode " << o.get() | |
14972 | << " txc "<< txc << dendl; | |
adb31ebb TL |
14973 | |
14974 | auto start_time = mono_clock::now(); | |
7c673cae | 14975 | int r = _do_remove(txc, c, o); |
adb31ebb TL |
14976 | log_latency_fn( |
14977 | __func__, | |
14978 | l_bluestore_remove_lat, | |
14979 | mono_clock::now() - start_time, | |
14980 | cct->_conf->bluestore_log_op_age, | |
14981 | [&](const ceph::timespan& lat) { | |
14982 | ostringstream ostr; | |
14983 | ostr << ", lat = " << timespan_str(lat) | |
14984 | << " cid =" << c->cid | |
14985 | << " oid =" << o->oid; | |
14986 | return ostr.str(); | |
14987 | } | |
14988 | ); | |
14989 | ||
7c673cae FG |
14990 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; |
14991 | return r; | |
14992 | } | |
14993 | ||
14994 | int BlueStore::_setattr(TransContext *txc, | |
14995 | CollectionRef& c, | |
14996 | OnodeRef& o, | |
14997 | const string& name, | |
14998 | bufferptr& val) | |
14999 | { | |
15000 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
15001 | << " " << name << " (" << val.length() << " bytes)" | |
15002 | << dendl; | |
15003 | int r = 0; | |
3efd9988 FG |
15004 | if (val.is_partial()) { |
15005 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
15006 | val.length()); | |
f91f0fd5 | 15007 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
15008 | } else { |
15009 | auto& b = o->onode.attrs[name.c_str()] = val; | |
f91f0fd5 | 15010 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 15011 | } |
7c673cae FG |
15012 | txc->write_onode(o); |
15013 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
15014 | << " " << name << " (" << val.length() << " bytes)" | |
15015 | << " = " << r << dendl; | |
15016 | return r; | |
15017 | } | |
15018 | ||
15019 | int BlueStore::_setattrs(TransContext *txc, | |
15020 | CollectionRef& c, | |
15021 | OnodeRef& o, | |
15022 | const map<string,bufferptr>& aset) | |
15023 | { | |
15024 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
15025 | << " " << aset.size() << " keys" | |
15026 | << dendl; | |
15027 | int r = 0; | |
15028 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
15029 | p != aset.end(); ++p) { | |
3efd9988 FG |
15030 | if (p->second.is_partial()) { |
15031 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 15032 | bufferptr(p->second.c_str(), p->second.length()); |
f91f0fd5 | 15033 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
15034 | } else { |
15035 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
f91f0fd5 | 15036 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 15037 | } |
7c673cae FG |
15038 | } |
15039 | txc->write_onode(o); | |
15040 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
15041 | << " " << aset.size() << " keys" | |
15042 | << " = " << r << dendl; | |
15043 | return r; | |
15044 | } | |
15045 | ||
15046 | ||
15047 | int BlueStore::_rmattr(TransContext *txc, | |
15048 | CollectionRef& c, | |
15049 | OnodeRef& o, | |
15050 | const string& name) | |
15051 | { | |
15052 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
15053 | << " " << name << dendl; | |
15054 | int r = 0; | |
15055 | auto it = o->onode.attrs.find(name.c_str()); | |
15056 | if (it == o->onode.attrs.end()) | |
15057 | goto out; | |
15058 | ||
15059 | o->onode.attrs.erase(it); | |
15060 | txc->write_onode(o); | |
15061 | ||
15062 | out: | |
15063 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
15064 | << " " << name << " = " << r << dendl; | |
15065 | return r; | |
15066 | } | |
15067 | ||
15068 | int BlueStore::_rmattrs(TransContext *txc, | |
15069 | CollectionRef& c, | |
15070 | OnodeRef& o) | |
15071 | { | |
15072 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15073 | int r = 0; | |
15074 | ||
15075 | if (o->onode.attrs.empty()) | |
15076 | goto out; | |
15077 | ||
15078 | o->onode.attrs.clear(); | |
15079 | txc->write_onode(o); | |
15080 | ||
15081 | out: | |
15082 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15083 | return r; | |
15084 | } | |
15085 | ||
9f95a23c | 15086 | void BlueStore::_do_omap_clear(TransContext *txc, OnodeRef& o) |
7c673cae | 15087 | { |
9f95a23c | 15088 | const string& omap_prefix = o->get_omap_prefix(); |
7c673cae | 15089 | string prefix, tail; |
9f95a23c TL |
15090 | o->get_omap_header(&prefix); |
15091 | o->get_omap_tail(&tail); | |
11fdf7f2 | 15092 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
494da23a | 15093 | txc->t->rmkey(omap_prefix, tail); |
11fdf7f2 TL |
15094 | dout(20) << __func__ << " remove range start: " |
15095 | << pretty_binary_string(prefix) << " end: " | |
15096 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
15097 | } |
15098 | ||
15099 | int BlueStore::_omap_clear(TransContext *txc, | |
15100 | CollectionRef& c, | |
15101 | OnodeRef& o) | |
15102 | { | |
15103 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15104 | int r = 0; | |
15105 | if (o->onode.has_omap()) { | |
15106 | o->flush(); | |
9f95a23c | 15107 | _do_omap_clear(txc, o); |
7c673cae FG |
15108 | o->onode.clear_omap_flag(); |
15109 | txc->write_onode(o); | |
15110 | } | |
15111 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15112 | return r; | |
15113 | } | |
15114 | ||
15115 | int BlueStore::_omap_setkeys(TransContext *txc, | |
15116 | CollectionRef& c, | |
15117 | OnodeRef& o, | |
15118 | bufferlist &bl) | |
15119 | { | |
15120 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15121 | int r; | |
11fdf7f2 | 15122 | auto p = bl.cbegin(); |
7c673cae FG |
15123 | __u32 num; |
15124 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 15125 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
15126 | o->onode.set_omap_flags_pgmeta(); |
15127 | } else { | |
522d829b | 15128 | o->onode.set_omap_flags(per_pool_omap == OMAP_BULK); |
11fdf7f2 | 15129 | } |
7c673cae | 15130 | txc->write_onode(o); |
494da23a | 15131 | |
9f95a23c | 15132 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
15133 | string key_tail; |
15134 | bufferlist tail; | |
9f95a23c | 15135 | o->get_omap_tail(&key_tail); |
494da23a | 15136 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
15137 | } else { |
15138 | txc->note_modified_object(o); | |
15139 | } | |
9f95a23c | 15140 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 15141 | string final_key; |
9f95a23c TL |
15142 | o->get_omap_key(string(), &final_key); |
15143 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 15144 | decode(num, p); |
7c673cae FG |
15145 | while (num--) { |
15146 | string key; | |
15147 | bufferlist value; | |
11fdf7f2 TL |
15148 | decode(key, p); |
15149 | decode(value, p); | |
9f95a23c | 15150 | final_key.resize(base_key_len); // keep prefix |
7c673cae | 15151 | final_key += key; |
11fdf7f2 | 15152 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 15153 | << " <- " << key << dendl; |
11fdf7f2 | 15154 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
15155 | } |
15156 | r = 0; | |
15157 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15158 | return r; | |
15159 | } | |
15160 | ||
15161 | int BlueStore::_omap_setheader(TransContext *txc, | |
15162 | CollectionRef& c, | |
15163 | OnodeRef &o, | |
15164 | bufferlist& bl) | |
15165 | { | |
15166 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15167 | int r; | |
15168 | string key; | |
15169 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 15170 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
15171 | o->onode.set_omap_flags_pgmeta(); |
15172 | } else { | |
522d829b | 15173 | o->onode.set_omap_flags(per_pool_omap == OMAP_BULK); |
11fdf7f2 | 15174 | } |
7c673cae | 15175 | txc->write_onode(o); |
494da23a | 15176 | |
9f95a23c | 15177 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
15178 | string key_tail; |
15179 | bufferlist tail; | |
9f95a23c | 15180 | o->get_omap_tail(&key_tail); |
494da23a | 15181 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
15182 | } else { |
15183 | txc->note_modified_object(o); | |
15184 | } | |
9f95a23c TL |
15185 | const string& prefix = o->get_omap_prefix(); |
15186 | o->get_omap_header(&key); | |
11fdf7f2 | 15187 | txc->t->set(prefix, key, bl); |
7c673cae FG |
15188 | r = 0; |
15189 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15190 | return r; | |
15191 | } | |
15192 | ||
15193 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
15194 | CollectionRef& c, | |
15195 | OnodeRef& o, | |
15196 | bufferlist& bl) | |
15197 | { | |
15198 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
15199 | int r = 0; | |
11fdf7f2 | 15200 | auto p = bl.cbegin(); |
7c673cae FG |
15201 | __u32 num; |
15202 | string final_key; | |
15203 | ||
15204 | if (!o->onode.has_omap()) { | |
15205 | goto out; | |
15206 | } | |
11fdf7f2 | 15207 | { |
9f95a23c TL |
15208 | const string& prefix = o->get_omap_prefix(); |
15209 | o->get_omap_key(string(), &final_key); | |
15210 | size_t base_key_len = final_key.size(); | |
11fdf7f2 TL |
15211 | decode(num, p); |
15212 | while (num--) { | |
15213 | string key; | |
15214 | decode(key, p); | |
9f95a23c | 15215 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
15216 | final_key += key; |
15217 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
15218 | << " <- " << key << dendl; | |
15219 | txc->t->rmkey(prefix, final_key); | |
15220 | } | |
7c673cae FG |
15221 | } |
15222 | txc->note_modified_object(o); | |
15223 | ||
15224 | out: | |
15225 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15226 | return r; | |
15227 | } | |
15228 | ||
15229 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
15230 | CollectionRef& c, | |
15231 | OnodeRef& o, | |
15232 | const string& first, const string& last) | |
15233 | { | |
15234 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
15235 | string key_first, key_last; |
15236 | int r = 0; | |
15237 | if (!o->onode.has_omap()) { | |
15238 | goto out; | |
15239 | } | |
11fdf7f2 | 15240 | { |
9f95a23c | 15241 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 15242 | o->flush(); |
9f95a23c TL |
15243 | o->get_omap_key(first, &key_first); |
15244 | o->get_omap_key(last, &key_last); | |
11fdf7f2 TL |
15245 | txc->t->rm_range_keys(prefix, key_first, key_last); |
15246 | dout(20) << __func__ << " remove range start: " | |
15247 | << pretty_binary_string(key_first) << " end: " | |
15248 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
15249 | } |
15250 | txc->note_modified_object(o); | |
15251 | ||
15252 | out: | |
15253 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
15254 | return r; | |
15255 | } | |
15256 | ||
15257 | int BlueStore::_set_alloc_hint( | |
15258 | TransContext *txc, | |
15259 | CollectionRef& c, | |
15260 | OnodeRef& o, | |
15261 | uint64_t expected_object_size, | |
15262 | uint64_t expected_write_size, | |
15263 | uint32_t flags) | |
15264 | { | |
15265 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
15266 | << " object_size " << expected_object_size | |
15267 | << " write_size " << expected_write_size | |
15268 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
15269 | << dendl; | |
15270 | int r = 0; | |
15271 | o->onode.expected_object_size = expected_object_size; | |
15272 | o->onode.expected_write_size = expected_write_size; | |
15273 | o->onode.alloc_hint_flags = flags; | |
15274 | txc->write_onode(o); | |
15275 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
15276 | << " object_size " << expected_object_size | |
15277 | << " write_size " << expected_write_size | |
15278 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
15279 | << " = " << r << dendl; | |
15280 | return r; | |
15281 | } | |
15282 | ||
15283 | int BlueStore::_clone(TransContext *txc, | |
15284 | CollectionRef& c, | |
15285 | OnodeRef& oldo, | |
15286 | OnodeRef& newo) | |
15287 | { | |
15288 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15289 | << newo->oid << dendl; | |
15290 | int r = 0; | |
15291 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
15292 | derr << __func__ << " mismatched hash on " << oldo->oid | |
15293 | << " and " << newo->oid << dendl; | |
15294 | return -EINVAL; | |
15295 | } | |
15296 | ||
7c673cae FG |
15297 | _assign_nid(txc, newo); |
15298 | ||
15299 | // clone data | |
15300 | oldo->flush(); | |
15301 | _do_truncate(txc, c, newo, 0); | |
15302 | if (cct->_conf->bluestore_clone_cow) { | |
15303 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
15304 | } else { | |
15305 | bufferlist bl; | |
15306 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
15307 | if (r < 0) | |
15308 | goto out; | |
15309 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
15310 | if (r < 0) | |
15311 | goto out; | |
15312 | } | |
15313 | ||
15314 | // clone attrs | |
15315 | newo->onode.attrs = oldo->onode.attrs; | |
15316 | ||
15317 | // clone omap | |
15318 | if (newo->onode.has_omap()) { | |
15319 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
15320 | newo->flush(); | |
9f95a23c | 15321 | _do_omap_clear(txc, newo); |
494da23a | 15322 | newo->onode.clear_omap_flag(); |
7c673cae FG |
15323 | } |
15324 | if (oldo->onode.has_omap()) { | |
15325 | dout(20) << __func__ << " copying omap data" << dendl; | |
494da23a | 15326 | if (newo->oid.is_pgmeta()) { |
9f95a23c TL |
15327 | newo->onode.set_omap_flags_pgmeta(); |
15328 | } else { | |
522d829b | 15329 | newo->onode.set_omap_flags(per_pool_omap == OMAP_BULK); |
7c673cae | 15330 | } |
9f95a23c | 15331 | const string& prefix = newo->get_omap_prefix(); |
11fdf7f2 | 15332 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 15333 | string head, tail; |
9f95a23c TL |
15334 | oldo->get_omap_header(&head); |
15335 | oldo->get_omap_tail(&tail); | |
7c673cae FG |
15336 | it->lower_bound(head); |
15337 | while (it->valid()) { | |
15338 | if (it->key() >= tail) { | |
15339 | dout(30) << __func__ << " reached tail" << dendl; | |
15340 | break; | |
15341 | } else { | |
15342 | dout(30) << __func__ << " got header/data " | |
15343 | << pretty_binary_string(it->key()) << dendl; | |
15344 | string key; | |
9f95a23c | 15345 | newo->rewrite_omap_key(it->key(), &key); |
11fdf7f2 | 15346 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
15347 | } |
15348 | it->next(); | |
15349 | } | |
494da23a TL |
15350 | string new_tail; |
15351 | bufferlist new_tail_value; | |
9f95a23c | 15352 | newo->get_omap_tail(&new_tail); |
494da23a | 15353 | txc->t->set(prefix, new_tail, new_tail_value); |
7c673cae FG |
15354 | } |
15355 | ||
15356 | txc->write_onode(newo); | |
15357 | r = 0; | |
15358 | ||
15359 | out: | |
15360 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15361 | << newo->oid << " = " << r << dendl; | |
15362 | return r; | |
15363 | } | |
15364 | ||
15365 | int BlueStore::_do_clone_range( | |
15366 | TransContext *txc, | |
15367 | CollectionRef& c, | |
15368 | OnodeRef& oldo, | |
15369 | OnodeRef& newo, | |
224ce89b WB |
15370 | uint64_t srcoff, |
15371 | uint64_t length, | |
15372 | uint64_t dstoff) | |
7c673cae FG |
15373 | { |
15374 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15375 | << newo->oid | |
15376 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
15377 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
15378 | oldo->extent_map.fault_range(db, srcoff, length); | |
15379 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
15380 | _dump_onode<30>(cct, *oldo); |
15381 | _dump_onode<30>(cct, *newo); | |
7c673cae | 15382 | |
11fdf7f2 | 15383 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
81eedcae TL |
15384 | _dump_onode<30>(cct, *oldo); |
15385 | _dump_onode<30>(cct, *newo); | |
7c673cae FG |
15386 | return 0; |
15387 | } | |
15388 | ||
15389 | int BlueStore::_clone_range(TransContext *txc, | |
15390 | CollectionRef& c, | |
15391 | OnodeRef& oldo, | |
15392 | OnodeRef& newo, | |
15393 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
15394 | { | |
15395 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15396 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
15397 | << " to offset 0x" << dstoff << std::dec << dendl; | |
15398 | int r = 0; | |
15399 | ||
35e4c445 FG |
15400 | if (srcoff + length >= OBJECT_MAX_SIZE || |
15401 | dstoff + length >= OBJECT_MAX_SIZE) { | |
15402 | r = -E2BIG; | |
15403 | goto out; | |
15404 | } | |
7c673cae FG |
15405 | if (srcoff + length > oldo->onode.size) { |
15406 | r = -EINVAL; | |
15407 | goto out; | |
15408 | } | |
15409 | ||
7c673cae FG |
15410 | _assign_nid(txc, newo); |
15411 | ||
15412 | if (length > 0) { | |
15413 | if (cct->_conf->bluestore_clone_cow) { | |
15414 | _do_zero(txc, c, newo, dstoff, length); | |
15415 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
15416 | } else { | |
15417 | bufferlist bl; | |
15418 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
15419 | if (r < 0) | |
15420 | goto out; | |
15421 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
15422 | if (r < 0) | |
15423 | goto out; | |
15424 | } | |
15425 | } | |
15426 | ||
15427 | txc->write_onode(newo); | |
15428 | r = 0; | |
15429 | ||
15430 | out: | |
15431 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15432 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
15433 | << " to offset 0x" << dstoff << std::dec | |
15434 | << " = " << r << dendl; | |
15435 | return r; | |
15436 | } | |
15437 | ||
15438 | int BlueStore::_rename(TransContext *txc, | |
15439 | CollectionRef& c, | |
15440 | OnodeRef& oldo, | |
15441 | OnodeRef& newo, | |
15442 | const ghobject_t& new_oid) | |
15443 | { | |
15444 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15445 | << new_oid << dendl; | |
15446 | int r; | |
15447 | ghobject_t old_oid = oldo->oid; | |
f91f0fd5 | 15448 | mempool::bluestore_cache_meta::string new_okey; |
7c673cae FG |
15449 | |
15450 | if (newo) { | |
15451 | if (newo->exists) { | |
15452 | r = -EEXIST; | |
15453 | goto out; | |
15454 | } | |
11fdf7f2 | 15455 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
15456 | } |
15457 | ||
15458 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
15459 | ||
15460 | // rewrite shards | |
15461 | { | |
15462 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
15463 | get_object_key(cct, new_oid, &new_okey); | |
15464 | string key; | |
15465 | for (auto &s : oldo->extent_map.shards) { | |
15466 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
15467 | [&](const string& final_key) { | |
15468 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
15469 | } | |
15470 | ); | |
15471 | s.dirty = true; | |
15472 | } | |
15473 | } | |
15474 | ||
15475 | newo = oldo; | |
15476 | txc->write_onode(newo); | |
15477 | ||
15478 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
15479 | // Onode in the old slot | |
15480 | c->onode_map.rename(oldo, old_oid, new_oid, new_okey); | |
15481 | r = 0; | |
15482 | ||
f64942e4 AA |
15483 | // hold a ref to new Onode in old name position, to ensure we don't drop |
15484 | // it from the cache before this txc commits (or else someone may come along | |
15485 | // and read newo's metadata via the old name). | |
15486 | txc->note_modified_object(oldo); | |
15487 | ||
7c673cae FG |
15488 | out: |
15489 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
15490 | << new_oid << " = " << r << dendl; | |
15491 | return r; | |
15492 | } | |
15493 | ||
15494 | // collections | |
15495 | ||
15496 | int BlueStore::_create_collection( | |
15497 | TransContext *txc, | |
15498 | const coll_t &cid, | |
15499 | unsigned bits, | |
15500 | CollectionRef *c) | |
15501 | { | |
15502 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
15503 | int r; | |
15504 | bufferlist bl; | |
15505 | ||
15506 | { | |
9f95a23c | 15507 | std::unique_lock l(coll_lock); |
7c673cae FG |
15508 | if (*c) { |
15509 | r = -EEXIST; | |
15510 | goto out; | |
15511 | } | |
11fdf7f2 TL |
15512 | auto p = new_coll_map.find(cid); |
15513 | ceph_assert(p != new_coll_map.end()); | |
15514 | *c = p->second; | |
7c673cae FG |
15515 | (*c)->cnode.bits = bits; |
15516 | coll_map[cid] = *c; | |
11fdf7f2 | 15517 | new_coll_map.erase(p); |
7c673cae | 15518 | } |
11fdf7f2 | 15519 | encode((*c)->cnode, bl); |
7c673cae FG |
15520 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
15521 | r = 0; | |
15522 | ||
15523 | out: | |
15524 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
15525 | return r; | |
15526 | } | |
15527 | ||
15528 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
15529 | CollectionRef *c) | |
15530 | { | |
15531 | dout(15) << __func__ << " " << cid << dendl; | |
15532 | int r; | |
15533 | ||
11fdf7f2 | 15534 | (*c)->flush_all_but_last(); |
7c673cae | 15535 | { |
9f95a23c | 15536 | std::unique_lock l(coll_lock); |
7c673cae FG |
15537 | if (!*c) { |
15538 | r = -ENOENT; | |
15539 | goto out; | |
15540 | } | |
15541 | size_t nonexistent_count = 0; | |
11fdf7f2 | 15542 | ceph_assert((*c)->exists); |
adb31ebb | 15543 | if ((*c)->onode_map.map_any([&](Onode* o) { |
f67539c2 TL |
15544 | if (o->exists) { |
15545 | dout(1) << __func__ << " " << o->oid << " " << o | |
15546 | << " exists in onode_map" << dendl; | |
7c673cae | 15547 | return true; |
f67539c2 TL |
15548 | } |
15549 | ++nonexistent_count; | |
15550 | return false; | |
15551 | })) { | |
7c673cae FG |
15552 | r = -ENOTEMPTY; |
15553 | goto out; | |
15554 | } | |
7c673cae FG |
15555 | vector<ghobject_t> ls; |
15556 | ghobject_t next; | |
15557 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
15558 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 15559 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae | 15560 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
f91f0fd5 | 15561 | nonexistent_count + 1, false, &ls, &next); |
7c673cae | 15562 | if (r >= 0) { |
11fdf7f2 TL |
15563 | // If true mean collecton has more objects than nonexistent_count, |
15564 | // so bypass check. | |
15565 | bool exists = (!next.is_max()); | |
7c673cae FG |
15566 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
15567 | dout(10) << __func__ << " oid " << *it << dendl; | |
15568 | auto onode = (*c)->onode_map.lookup(*it); | |
15569 | exists = !onode || onode->exists; | |
15570 | if (exists) { | |
494da23a | 15571 | dout(1) << __func__ << " " << *it |
f67539c2 TL |
15572 | << " exists in db, " |
15573 | << (!onode ? "not present in ram" : "present in ram") | |
15574 | << dendl; | |
7c673cae FG |
15575 | } |
15576 | } | |
15577 | if (!exists) { | |
f67539c2 | 15578 | _do_remove_collection(txc, c); |
7c673cae FG |
15579 | r = 0; |
15580 | } else { | |
15581 | dout(10) << __func__ << " " << cid | |
15582 | << " is non-empty" << dendl; | |
f67539c2 | 15583 | r = -ENOTEMPTY; |
7c673cae FG |
15584 | } |
15585 | } | |
15586 | } | |
f67539c2 | 15587 | out: |
7c673cae FG |
15588 | dout(10) << __func__ << " " << cid << " = " << r << dendl; |
15589 | return r; | |
15590 | } | |
15591 | ||
11fdf7f2 TL |
15592 | void BlueStore::_do_remove_collection(TransContext *txc, |
15593 | CollectionRef *c) | |
15594 | { | |
15595 | coll_map.erase((*c)->cid); | |
15596 | txc->removed_collections.push_back(*c); | |
15597 | (*c)->exists = false; | |
15598 | _osr_register_zombie((*c)->osr.get()); | |
15599 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
15600 | c->reset(); | |
15601 | } | |
15602 | ||
7c673cae FG |
15603 | int BlueStore::_split_collection(TransContext *txc, |
15604 | CollectionRef& c, | |
15605 | CollectionRef& d, | |
15606 | unsigned bits, int rem) | |
15607 | { | |
15608 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
15609 | << " bits " << bits << dendl; | |
9f95a23c TL |
15610 | std::unique_lock l(c->lock); |
15611 | std::unique_lock l2(d->lock); | |
7c673cae FG |
15612 | int r; |
15613 | ||
15614 | // flush all previous deferred writes on this sequencer. this is a bit | |
15615 | // heavyweight, but we need to make sure all deferred writes complete | |
15616 | // before we split as the new collection's sequencer may need to order | |
15617 | // this after those writes, and we don't bother with the complexity of | |
15618 | // moving those TransContexts over to the new osr. | |
15619 | _osr_drain_preceding(txc); | |
15620 | ||
15621 | // move any cached items (onodes and referenced shared blobs) that will | |
15622 | // belong to the child collection post-split. leave everything else behind. | |
15623 | // this may include things that don't strictly belong to the now-smaller | |
15624 | // parent split, but the OSD will always send us a split for every new | |
15625 | // child. | |
15626 | ||
15627 | spg_t pgid, dest_pgid; | |
15628 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 15629 | ceph_assert(is_pg); |
7c673cae | 15630 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 15631 | ceph_assert(is_pg); |
7c673cae FG |
15632 | |
15633 | // the destination should initially be empty. | |
11fdf7f2 TL |
15634 | ceph_assert(d->onode_map.empty()); |
15635 | ceph_assert(d->shared_blob_set.empty()); | |
15636 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
15637 | |
15638 | c->split_cache(d.get()); | |
15639 | ||
15640 | // adjust bits. note that this will be redundant for all but the first | |
15641 | // split call for this parent (first child). | |
15642 | c->cnode.bits = bits; | |
11fdf7f2 | 15643 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
15644 | r = 0; |
15645 | ||
15646 | bufferlist bl; | |
11fdf7f2 | 15647 | encode(c->cnode, bl); |
7c673cae FG |
15648 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
15649 | ||
15650 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
15651 | << " bits " << bits << " = " << r << dendl; | |
15652 | return r; | |
15653 | } | |
15654 | ||
11fdf7f2 TL |
15655 | int BlueStore::_merge_collection( |
15656 | TransContext *txc, | |
15657 | CollectionRef *c, | |
15658 | CollectionRef& d, | |
15659 | unsigned bits) | |
15660 | { | |
15661 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
15662 | << " bits " << bits << dendl; | |
9f95a23c TL |
15663 | std::unique_lock l((*c)->lock); |
15664 | std::unique_lock l2(d->lock); | |
11fdf7f2 TL |
15665 | int r; |
15666 | ||
15667 | coll_t cid = (*c)->cid; | |
15668 | ||
15669 | // flush all previous deferred writes on the source collection to ensure | |
15670 | // that all deferred writes complete before we merge as the target collection's | |
15671 | // sequencer may need to order new ops after those writes. | |
15672 | ||
15673 | _osr_drain((*c)->osr.get()); | |
15674 | ||
15675 | // move any cached items (onodes and referenced shared blobs) that will | |
15676 | // belong to the child collection post-split. leave everything else behind. | |
15677 | // this may include things that don't strictly belong to the now-smaller | |
15678 | // parent split, but the OSD will always send us a split for every new | |
15679 | // child. | |
15680 | ||
15681 | spg_t pgid, dest_pgid; | |
15682 | bool is_pg = cid.is_pg(&pgid); | |
15683 | ceph_assert(is_pg); | |
15684 | is_pg = d->cid.is_pg(&dest_pgid); | |
15685 | ceph_assert(is_pg); | |
15686 | ||
15687 | // adjust bits. note that this will be redundant for all but the first | |
15688 | // merge call for the parent/target. | |
15689 | d->cnode.bits = bits; | |
15690 | ||
15691 | // behavior depends on target (d) bits, so this after that is updated. | |
15692 | (*c)->split_cache(d.get()); | |
15693 | ||
15694 | // remove source collection | |
15695 | { | |
9f95a23c | 15696 | std::unique_lock l3(coll_lock); |
11fdf7f2 TL |
15697 | _do_remove_collection(txc, c); |
15698 | } | |
15699 | ||
15700 | r = 0; | |
15701 | ||
15702 | bufferlist bl; | |
15703 | encode(d->cnode, bl); | |
15704 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
15705 | ||
15706 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
15707 | << " bits " << bits << " = " << r << dendl; | |
15708 | return r; | |
15709 | } | |
15710 | ||
494da23a TL |
15711 | void BlueStore::log_latency( |
15712 | const char* name, | |
15713 | int idx, | |
15714 | const ceph::timespan& l, | |
15715 | double lat_threshold, | |
15716 | const char* info) const | |
15717 | { | |
15718 | logger->tinc(idx, l); | |
15719 | if (lat_threshold > 0.0 && | |
15720 | l >= make_timespan(lat_threshold)) { | |
15721 | dout(0) << __func__ << " slow operation observed for " << name | |
15722 | << ", latency = " << l | |
15723 | << info | |
15724 | << dendl; | |
15725 | } | |
15726 | } | |
15727 | ||
11fdf7f2 | 15728 | void BlueStore::log_latency_fn( |
494da23a | 15729 | const char* name, |
11fdf7f2 TL |
15730 | int idx, |
15731 | const ceph::timespan& l, | |
494da23a TL |
15732 | double lat_threshold, |
15733 | std::function<string (const ceph::timespan& lat)> fn) const | |
11fdf7f2 | 15734 | { |
494da23a TL |
15735 | logger->tinc(idx, l); |
15736 | if (lat_threshold > 0.0 && | |
15737 | l >= make_timespan(lat_threshold)) { | |
15738 | dout(0) << __func__ << " slow operation observed for " << name | |
15739 | << ", latency = " << l | |
15740 | << fn(l) | |
15741 | << dendl; | |
15742 | } | |
11fdf7f2 TL |
15743 | } |
15744 | ||
9f95a23c TL |
15745 | #if defined(WITH_LTTNG) |
15746 | void BlueStore::BlueStoreThrottle::emit_initial_tracepoint( | |
15747 | KeyValueDB &db, | |
15748 | TransContext &txc, | |
15749 | mono_clock::time_point start_throttle_acquire) | |
15750 | { | |
15751 | pending_kv_ios += txc.ios; | |
15752 | if (txc.deferred_txn) { | |
15753 | pending_deferred_ios += txc.ios; | |
15754 | } | |
15755 | ||
15756 | uint64_t started = 0; | |
15757 | uint64_t completed = 0; | |
15758 | if (should_trace(&started, &completed)) { | |
15759 | txc.tracing = true; | |
15760 | uint64_t rocksdb_base_level, | |
15761 | rocksdb_estimate_pending_compaction_bytes, | |
15762 | rocksdb_cur_size_all_mem_tables, | |
15763 | rocksdb_compaction_pending, | |
15764 | rocksdb_mem_table_flush_pending, | |
15765 | rocksdb_num_running_compactions, | |
15766 | rocksdb_num_running_flushes, | |
15767 | rocksdb_actual_delayed_write_rate; | |
15768 | db.get_property( | |
15769 | "rocksdb.base-level", | |
15770 | &rocksdb_base_level); | |
15771 | db.get_property( | |
15772 | "rocksdb.estimate-pending-compaction-bytes", | |
15773 | &rocksdb_estimate_pending_compaction_bytes); | |
15774 | db.get_property( | |
15775 | "rocksdb.cur-size-all-mem-tables", | |
15776 | &rocksdb_cur_size_all_mem_tables); | |
15777 | db.get_property( | |
15778 | "rocksdb.compaction-pending", | |
15779 | &rocksdb_compaction_pending); | |
15780 | db.get_property( | |
15781 | "rocksdb.mem-table-flush-pending", | |
15782 | &rocksdb_mem_table_flush_pending); | |
15783 | db.get_property( | |
15784 | "rocksdb.num-running-compactions", | |
15785 | &rocksdb_num_running_compactions); | |
15786 | db.get_property( | |
15787 | "rocksdb.num-running-flushes", | |
15788 | &rocksdb_num_running_flushes); | |
15789 | db.get_property( | |
15790 | "rocksdb.actual-delayed-write-rate", | |
15791 | &rocksdb_actual_delayed_write_rate); | |
15792 | ||
15793 | ||
15794 | tracepoint( | |
15795 | bluestore, | |
15796 | transaction_initial_state, | |
15797 | txc.osr->get_sequencer_id(), | |
15798 | txc.seq, | |
15799 | throttle_bytes.get_current(), | |
15800 | throttle_deferred_bytes.get_current(), | |
15801 | pending_kv_ios, | |
15802 | pending_deferred_ios, | |
15803 | started, | |
15804 | completed, | |
15805 | ceph::to_seconds<double>(mono_clock::now() - start_throttle_acquire)); | |
15806 | ||
15807 | tracepoint( | |
15808 | bluestore, | |
15809 | transaction_initial_state_rocksdb, | |
15810 | txc.osr->get_sequencer_id(), | |
15811 | txc.seq, | |
15812 | rocksdb_base_level, | |
15813 | rocksdb_estimate_pending_compaction_bytes, | |
15814 | rocksdb_cur_size_all_mem_tables, | |
15815 | rocksdb_compaction_pending, | |
15816 | rocksdb_mem_table_flush_pending, | |
15817 | rocksdb_num_running_compactions, | |
15818 | rocksdb_num_running_flushes, | |
15819 | rocksdb_actual_delayed_write_rate); | |
15820 | } | |
15821 | } | |
15822 | #endif | |
15823 | ||
15824 | mono_clock::duration BlueStore::BlueStoreThrottle::log_state_latency( | |
15825 | TransContext &txc, PerfCounters *logger, int state) | |
15826 | { | |
15827 | mono_clock::time_point now = mono_clock::now(); | |
15828 | mono_clock::duration lat = now - txc.last_stamp; | |
15829 | logger->tinc(state, lat); | |
15830 | #if defined(WITH_LTTNG) | |
15831 | if (txc.tracing && | |
15832 | state >= l_bluestore_state_prepare_lat && | |
15833 | state <= l_bluestore_state_done_lat) { | |
15834 | OID_ELAPSED("", lat.to_nsec() / 1000.0, txc.get_state_latency_name(state)); | |
15835 | tracepoint( | |
15836 | bluestore, | |
15837 | transaction_state_duration, | |
15838 | txc.osr->get_sequencer_id(), | |
15839 | txc.seq, | |
15840 | state, | |
15841 | ceph::to_seconds<double>(lat)); | |
15842 | } | |
15843 | #endif | |
15844 | txc.last_stamp = now; | |
15845 | return lat; | |
15846 | } | |
15847 | ||
15848 | bool BlueStore::BlueStoreThrottle::try_start_transaction( | |
15849 | KeyValueDB &db, | |
15850 | TransContext &txc, | |
15851 | mono_clock::time_point start_throttle_acquire) | |
15852 | { | |
15853 | throttle_bytes.get(txc.cost); | |
15854 | ||
15855 | if (!txc.deferred_txn || throttle_deferred_bytes.get_or_fail(txc.cost)) { | |
15856 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15857 | return true; | |
15858 | } else { | |
15859 | return false; | |
15860 | } | |
15861 | } | |
15862 | ||
15863 | void BlueStore::BlueStoreThrottle::finish_start_transaction( | |
15864 | KeyValueDB &db, | |
15865 | TransContext &txc, | |
15866 | mono_clock::time_point start_throttle_acquire) | |
15867 | { | |
15868 | ceph_assert(txc.deferred_txn); | |
15869 | throttle_deferred_bytes.get(txc.cost); | |
15870 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15871 | } | |
15872 | ||
15873 | #if defined(WITH_LTTNG) | |
15874 | void BlueStore::BlueStoreThrottle::complete_kv(TransContext &txc) | |
15875 | { | |
15876 | pending_kv_ios -= 1; | |
15877 | ios_completed_since_last_traced++; | |
15878 | if (txc.tracing) { | |
15879 | tracepoint( | |
15880 | bluestore, | |
15881 | transaction_commit_latency, | |
15882 | txc.osr->get_sequencer_id(), | |
15883 | txc.seq, | |
15884 | ceph::to_seconds<double>(mono_clock::now() - txc.start)); | |
15885 | } | |
15886 | } | |
15887 | #endif | |
15888 | ||
15889 | #if defined(WITH_LTTNG) | |
15890 | void BlueStore::BlueStoreThrottle::complete(TransContext &txc) | |
15891 | { | |
15892 | if (txc.deferred_txn) { | |
15893 | pending_deferred_ios -= 1; | |
15894 | } | |
15895 | if (txc.tracing) { | |
15896 | mono_clock::time_point now = mono_clock::now(); | |
15897 | mono_clock::duration lat = now - txc.start; | |
15898 | tracepoint( | |
15899 | bluestore, | |
15900 | transaction_total_duration, | |
15901 | txc.osr->get_sequencer_id(), | |
15902 | txc.seq, | |
15903 | ceph::to_seconds<double>(lat)); | |
15904 | } | |
15905 | } | |
15906 | #endif | |
11fdf7f2 | 15907 | |
7c673cae FG |
15908 | // DB key value Histogram |
15909 | #define KEY_SLAB 32 | |
15910 | #define VALUE_SLAB 64 | |
15911 | ||
15912 | const string prefix_onode = "o"; | |
15913 | const string prefix_onode_shard = "x"; | |
15914 | const string prefix_other = "Z"; | |
15915 | ||
15916 | int BlueStore::DBHistogram::get_key_slab(size_t sz) | |
15917 | { | |
15918 | return (sz/KEY_SLAB); | |
15919 | } | |
15920 | ||
15921 | string BlueStore::DBHistogram::get_key_slab_to_range(int slab) | |
15922 | { | |
15923 | int lower_bound = slab * KEY_SLAB; | |
15924 | int upper_bound = (slab + 1) * KEY_SLAB; | |
15925 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15926 | return ret; | |
15927 | } | |
15928 | ||
15929 | int BlueStore::DBHistogram::get_value_slab(size_t sz) | |
15930 | { | |
15931 | return (sz/VALUE_SLAB); | |
15932 | } | |
15933 | ||
15934 | string BlueStore::DBHistogram::get_value_slab_to_range(int slab) | |
15935 | { | |
15936 | int lower_bound = slab * VALUE_SLAB; | |
15937 | int upper_bound = (slab + 1) * VALUE_SLAB; | |
15938 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15939 | return ret; | |
15940 | } | |
15941 | ||
15942 | void BlueStore::DBHistogram::update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
15943 | const string &prefix, size_t key_size, size_t value_size) | |
15944 | { | |
15945 | uint32_t key_slab = get_key_slab(key_size); | |
15946 | uint32_t value_slab = get_value_slab(value_size); | |
15947 | key_hist[prefix][key_slab].count++; | |
11fdf7f2 TL |
15948 | key_hist[prefix][key_slab].max_len = |
15949 | std::max<size_t>(key_size, key_hist[prefix][key_slab].max_len); | |
7c673cae FG |
15950 | key_hist[prefix][key_slab].val_map[value_slab].count++; |
15951 | key_hist[prefix][key_slab].val_map[value_slab].max_len = | |
11fdf7f2 TL |
15952 | std::max<size_t>(value_size, |
15953 | key_hist[prefix][key_slab].val_map[value_slab].max_len); | |
7c673cae FG |
15954 | } |
15955 | ||
15956 | void BlueStore::DBHistogram::dump(Formatter *f) | |
15957 | { | |
15958 | f->open_object_section("rocksdb_value_distribution"); | |
15959 | for (auto i : value_hist) { | |
15960 | f->dump_unsigned(get_value_slab_to_range(i.first).data(), i.second); | |
15961 | } | |
15962 | f->close_section(); | |
15963 | ||
15964 | f->open_object_section("rocksdb_key_value_histogram"); | |
15965 | for (auto i : key_hist) { | |
15966 | f->dump_string("prefix", i.first); | |
15967 | f->open_object_section("key_hist"); | |
15968 | for ( auto k : i.second) { | |
15969 | f->dump_unsigned(get_key_slab_to_range(k.first).data(), k.second.count); | |
15970 | f->dump_unsigned("max_len", k.second.max_len); | |
15971 | f->open_object_section("value_hist"); | |
15972 | for ( auto j : k.second.val_map) { | |
15973 | f->dump_unsigned(get_value_slab_to_range(j.first).data(), j.second.count); | |
15974 | f->dump_unsigned("max_len", j.second.max_len); | |
15975 | } | |
15976 | f->close_section(); | |
15977 | } | |
15978 | f->close_section(); | |
15979 | } | |
15980 | f->close_section(); | |
15981 | } | |
15982 | ||
15983 | //Itrerates through the db and collects the stats | |
15984 | void BlueStore::generate_db_histogram(Formatter *f) | |
15985 | { | |
15986 | //globals | |
15987 | uint64_t num_onodes = 0; | |
15988 | uint64_t num_shards = 0; | |
15989 | uint64_t num_super = 0; | |
15990 | uint64_t num_coll = 0; | |
15991 | uint64_t num_omap = 0; | |
11fdf7f2 | 15992 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
15993 | uint64_t num_deferred = 0; |
15994 | uint64_t num_alloc = 0; | |
15995 | uint64_t num_stat = 0; | |
15996 | uint64_t num_others = 0; | |
15997 | uint64_t num_shared_shards = 0; | |
15998 | size_t max_key_size =0, max_value_size = 0; | |
15999 | uint64_t total_key_size = 0, total_value_size = 0; | |
16000 | size_t key_size = 0, value_size = 0; | |
16001 | DBHistogram hist; | |
16002 | ||
11fdf7f2 | 16003 | auto start = coarse_mono_clock::now(); |
7c673cae | 16004 | |
11fdf7f2 | 16005 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
16006 | iter->seek_to_first(); |
16007 | while (iter->valid()) { | |
16008 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
16009 | key_size = iter->key_size(); | |
16010 | value_size = iter->value_size(); | |
16011 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
16012 | max_key_size = std::max(max_key_size, key_size); |
16013 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
16014 | total_key_size += key_size; |
16015 | total_value_size += value_size; | |
16016 | ||
16017 | pair<string,string> key(iter->raw_key()); | |
16018 | ||
16019 | if (key.first == PREFIX_SUPER) { | |
16020 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
16021 | num_super++; | |
16022 | } else if (key.first == PREFIX_STAT) { | |
16023 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
16024 | num_stat++; | |
16025 | } else if (key.first == PREFIX_COLL) { | |
16026 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
16027 | num_coll++; | |
16028 | } else if (key.first == PREFIX_OBJ) { | |
16029 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
16030 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
16031 | num_onodes++; | |
16032 | } else { | |
16033 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
16034 | num_shards++; | |
16035 | } | |
16036 | } else if (key.first == PREFIX_OMAP) { | |
16037 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
16038 | num_omap++; | |
f67539c2 TL |
16039 | } else if (key.first == PREFIX_PERPOOL_OMAP) { |
16040 | hist.update_hist_entry(hist.key_hist, PREFIX_PERPOOL_OMAP, key_size, value_size); | |
16041 | num_omap++; | |
16042 | } else if (key.first == PREFIX_PERPG_OMAP) { | |
16043 | hist.update_hist_entry(hist.key_hist, PREFIX_PERPG_OMAP, key_size, value_size); | |
16044 | num_omap++; | |
11fdf7f2 TL |
16045 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
16046 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
16047 | num_pgmeta_omap++; | |
7c673cae FG |
16048 | } else if (key.first == PREFIX_DEFERRED) { |
16049 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
16050 | num_deferred++; | |
11fdf7f2 | 16051 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
16052 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
16053 | num_alloc++; | |
16054 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
16055 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
16056 | num_shared_shards++; | |
16057 | } else { | |
16058 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
16059 | num_others++; | |
16060 | } | |
16061 | iter->next(); | |
16062 | } | |
16063 | ||
11fdf7f2 | 16064 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
16065 | f->open_object_section("rocksdb_key_value_stats"); |
16066 | f->dump_unsigned("num_onodes", num_onodes); | |
16067 | f->dump_unsigned("num_shards", num_shards); | |
16068 | f->dump_unsigned("num_super", num_super); | |
16069 | f->dump_unsigned("num_coll", num_coll); | |
16070 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 16071 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
16072 | f->dump_unsigned("num_deferred", num_deferred); |
16073 | f->dump_unsigned("num_alloc", num_alloc); | |
16074 | f->dump_unsigned("num_stat", num_stat); | |
16075 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
16076 | f->dump_unsigned("num_others", num_others); | |
16077 | f->dump_unsigned("max_key_size", max_key_size); | |
16078 | f->dump_unsigned("max_value_size", max_value_size); | |
16079 | f->dump_unsigned("total_key_size", total_key_size); | |
16080 | f->dump_unsigned("total_value_size", total_value_size); | |
16081 | f->close_section(); | |
16082 | ||
16083 | hist.dump(f); | |
16084 | ||
16085 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
16086 | ||
16087 | } | |
16088 | ||
f6b5b4d7 | 16089 | void BlueStore::_shutdown_cache() |
7c673cae FG |
16090 | { |
16091 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
16092 | for (auto i : buffer_cache_shards) { |
16093 | i->flush(); | |
11fdf7f2 | 16094 | ceph_assert(i->empty()); |
7c673cae FG |
16095 | } |
16096 | for (auto& p : coll_map) { | |
f6b5b4d7 | 16097 | p.second->onode_map.clear(); |
3efd9988 FG |
16098 | if (!p.second->shared_blob_set.empty()) { |
16099 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 16100 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 16101 | } |
11fdf7f2 TL |
16102 | ceph_assert(p.second->onode_map.empty()); |
16103 | ceph_assert(p.second->shared_blob_set.empty()); | |
7c673cae FG |
16104 | } |
16105 | coll_map.clear(); | |
f6b5b4d7 TL |
16106 | for (auto i : onode_cache_shards) { |
16107 | ceph_assert(i->empty()); | |
16108 | } | |
7c673cae FG |
16109 | } |
16110 | ||
31f18b77 FG |
16111 | // For external caller. |
16112 | // We use a best-effort policy instead, e.g., | |
16113 | // we don't care if there are still some pinned onodes/data in the cache | |
16114 | // after this command is completed. | |
11fdf7f2 | 16115 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
16116 | { |
16117 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
16118 | for (auto i : onode_cache_shards) { |
16119 | i->flush(); | |
16120 | } | |
16121 | for (auto i : buffer_cache_shards) { | |
16122 | i->flush(); | |
31f18b77 | 16123 | } |
11fdf7f2 TL |
16124 | |
16125 | return 0; | |
31f18b77 FG |
16126 | } |
16127 | ||
7c673cae FG |
16128 | void BlueStore::_apply_padding(uint64_t head_pad, |
16129 | uint64_t tail_pad, | |
7c673cae FG |
16130 | bufferlist& padded) |
16131 | { | |
7c673cae | 16132 | if (head_pad) { |
224ce89b | 16133 | padded.prepend_zero(head_pad); |
7c673cae FG |
16134 | } |
16135 | if (tail_pad) { | |
16136 | padded.append_zero(tail_pad); | |
16137 | } | |
16138 | if (head_pad || tail_pad) { | |
16139 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
16140 | << " tail 0x" << tail_pad << std::dec << dendl; | |
16141 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
16142 | } | |
16143 | } | |
16144 | ||
11fdf7f2 TL |
16145 | void BlueStore::_record_onode(OnodeRef &o, KeyValueDB::Transaction &txn) |
16146 | { | |
16147 | // finalize extent_map shards | |
16148 | o->extent_map.update(txn, false); | |
16149 | if (o->extent_map.needs_reshard()) { | |
16150 | o->extent_map.reshard(db, txn); | |
16151 | o->extent_map.update(txn, true); | |
16152 | if (o->extent_map.needs_reshard()) { | |
16153 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
16154 | << dendl; | |
16155 | o->extent_map.clear_needs_reshard(); | |
16156 | } | |
16157 | logger->inc(l_bluestore_onode_reshard); | |
16158 | } | |
16159 | ||
16160 | // bound encode | |
16161 | size_t bound = 0; | |
16162 | denc(o->onode, bound); | |
16163 | o->extent_map.bound_encode_spanning_blobs(bound); | |
16164 | if (o->onode.extent_map_shards.empty()) { | |
16165 | denc(o->extent_map.inline_bl, bound); | |
16166 | } | |
16167 | ||
16168 | // encode | |
16169 | bufferlist bl; | |
16170 | unsigned onode_part, blob_part, extent_part; | |
16171 | { | |
16172 | auto p = bl.get_contiguous_appender(bound, true); | |
16173 | denc(o->onode, p); | |
16174 | onode_part = p.get_logical_offset(); | |
16175 | o->extent_map.encode_spanning_blobs(p); | |
16176 | blob_part = p.get_logical_offset() - onode_part; | |
16177 | if (o->onode.extent_map_shards.empty()) { | |
16178 | denc(o->extent_map.inline_bl, p); | |
16179 | } | |
16180 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
16181 | } | |
16182 | ||
16183 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
16184 | << " (" << onode_part << " bytes onode + " | |
16185 | << blob_part << " bytes spanning blobs + " | |
16186 | << extent_part << " bytes inline extents)" | |
16187 | << dendl; | |
16188 | ||
16189 | ||
16190 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
16191 | } | |
16192 | ||
16193 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
16194 | { | |
16195 | std::lock_guard l(qlock); | |
16196 | ||
522d829b TL |
16197 | if (!spurious_read_errors_alert.empty() && |
16198 | cct->_conf->bluestore_warn_on_spurious_read_errors) { | |
f67539c2 TL |
16199 | alerts.emplace( |
16200 | "BLUESTORE_SPURIOUS_READ_ERRORS", | |
16201 | spurious_read_errors_alert); | |
16202 | } | |
81eedcae TL |
16203 | if (!disk_size_mismatch_alert.empty()) { |
16204 | alerts.emplace( | |
16205 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
16206 | disk_size_mismatch_alert); | |
16207 | } | |
16208 | if (!legacy_statfs_alert.empty()) { | |
16209 | alerts.emplace( | |
16210 | "BLUESTORE_LEGACY_STATFS", | |
16211 | legacy_statfs_alert); | |
16212 | } | |
11fdf7f2 TL |
16213 | if (!spillover_alert.empty() && |
16214 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
16215 | alerts.emplace( | |
16216 | "BLUEFS_SPILLOVER", | |
16217 | spillover_alert); | |
16218 | } | |
f67539c2 TL |
16219 | if (!no_per_pg_omap_alert.empty()) { |
16220 | alerts.emplace( | |
16221 | "BLUESTORE_NO_PER_PG_OMAP", | |
16222 | no_per_pg_omap_alert); | |
16223 | } | |
9f95a23c TL |
16224 | if (!no_per_pool_omap_alert.empty()) { |
16225 | alerts.emplace( | |
16226 | "BLUESTORE_NO_PER_POOL_OMAP", | |
16227 | no_per_pool_omap_alert); | |
16228 | } | |
11fdf7f2 TL |
16229 | string s0(failed_cmode); |
16230 | ||
16231 | if (!failed_compressors.empty()) { | |
16232 | if (!s0.empty()) { | |
16233 | s0 += ", "; | |
16234 | } | |
16235 | s0 += "unable to load:"; | |
16236 | bool first = true; | |
16237 | for (auto& s : failed_compressors) { | |
16238 | if (first) { | |
16239 | first = false; | |
16240 | } else { | |
16241 | s0 += ", "; | |
16242 | } | |
16243 | s0 += s; | |
16244 | } | |
16245 | alerts.emplace( | |
16246 | "BLUESTORE_NO_COMPRESSION", | |
16247 | s0); | |
16248 | } | |
16249 | } | |
16250 | ||
9f95a23c TL |
16251 | void BlueStore::_collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
16252 | size_t extents) | |
16253 | { | |
16254 | alloc_stats_count++; | |
16255 | alloc_stats_fragments += extents; | |
16256 | alloc_stats_size += need; | |
16257 | } | |
16258 | ||
16259 | void BlueStore::_record_allocation_stats() | |
16260 | { | |
16261 | // don't care about data consistency, | |
16262 | // fields can be partially modified while making the tuple | |
16263 | auto t0 = std::make_tuple( | |
16264 | alloc_stats_count.exchange(0), | |
16265 | alloc_stats_fragments.exchange(0), | |
16266 | alloc_stats_size.exchange(0)); | |
16267 | ||
16268 | dout(0) << " allocation stats probe " | |
16269 | << probe_count << ":" | |
16270 | << " cnt: " << std::get<0>(t0) | |
16271 | << " frags: " << std::get<1>(t0) | |
16272 | << " size: " << std::get<2>(t0) | |
16273 | << dendl; | |
16274 | ||
16275 | ||
16276 | // | |
16277 | // Keep the history for probes from the power-of-two sequence: | |
16278 | // -1, -2, -4, -8, -16 | |
16279 | // | |
16280 | size_t base = 1; | |
16281 | for (auto& t : alloc_stats_history) { | |
16282 | dout(0) << " probe -" | |
16283 | << base + (probe_count % base) << ": " | |
16284 | << std::get<0>(t) | |
16285 | << ", " << std::get<1>(t) | |
16286 | << ", " << std::get<2>(t) | |
16287 | << dendl; | |
16288 | base <<= 1; | |
16289 | } | |
16290 | dout(0) << "------------" << dendl; | |
16291 | ||
f67539c2 | 16292 | ++ probe_count; |
9f95a23c | 16293 | |
f67539c2 TL |
16294 | for (ssize_t i = alloc_stats_history.size() - 1 ; i > 0 ; --i) { |
16295 | if ((probe_count % (1 << i)) == 0) { | |
16296 | alloc_stats_history[i] = alloc_stats_history[i - 1]; | |
16297 | } | |
9f95a23c TL |
16298 | } |
16299 | alloc_stats_history[0].swap(t0); | |
16300 | } | |
16301 | ||
7c673cae | 16302 | // =========================================== |
11fdf7f2 TL |
16303 | // BlueStoreRepairer |
16304 | ||
16305 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
16306 | const interval_set<uint64_t>& extents) | |
16307 | { | |
16308 | ceph_assert(granularity); // initialized | |
16309 | // can't call for the second time | |
16310 | ceph_assert(!was_filtered_out); | |
16311 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
16312 | ||
16313 | uint64_t prev_pos = 0; | |
16314 | uint64_t npos = collections_bfs.size(); | |
16315 | ||
16316 | bloom_vector collections_reduced; | |
16317 | bloom_vector objects_reduced; | |
16318 | ||
16319 | for (auto e : extents) { | |
16320 | if (e.second == 0) { | |
16321 | continue; | |
16322 | } | |
16323 | uint64_t pos = max(e.first / granularity, prev_pos); | |
16324 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
16325 | while (pos != npos && pos < end_pos) { | |
16326 | ceph_assert( collections_bfs[pos].element_count() == | |
16327 | objects_bfs[pos].element_count()); | |
16328 | if (collections_bfs[pos].element_count()) { | |
16329 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
16330 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
16331 | } | |
16332 | ++pos; | |
16333 | } | |
16334 | prev_pos = end_pos; | |
16335 | } | |
16336 | collections_reduced.swap(collections_bfs); | |
16337 | objects_reduced.swap(objects_bfs); | |
16338 | was_filtered_out = true; | |
16339 | return collections_bfs.size(); | |
16340 | } | |
16341 | ||
16342 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
16343 | const string& prefix, | |
16344 | const string& key) | |
16345 | { | |
b3b6e05e | 16346 | std::lock_guard l(lock); |
11fdf7f2 TL |
16347 | if (!remove_key_txn) { |
16348 | remove_key_txn = db->get_transaction(); | |
16349 | } | |
16350 | ++to_repair_cnt; | |
16351 | remove_key_txn->rmkey(prefix, key); | |
16352 | ||
16353 | return true; | |
16354 | } | |
16355 | ||
f67539c2 | 16356 | void BlueStoreRepairer::fix_per_pool_omap(KeyValueDB *db, int val) |
9f95a23c | 16357 | { |
b3b6e05e TL |
16358 | std::lock_guard l(lock); // possibly redundant |
16359 | ceph_assert(fix_per_pool_omap_txn == nullptr); | |
9f95a23c TL |
16360 | fix_per_pool_omap_txn = db->get_transaction(); |
16361 | ++to_repair_cnt; | |
16362 | bufferlist bl; | |
f67539c2 | 16363 | bl.append(stringify(val)); |
9f95a23c TL |
16364 | fix_per_pool_omap_txn->set(PREFIX_SUPER, "per_pool_omap", bl); |
16365 | } | |
16366 | ||
11fdf7f2 TL |
16367 | bool BlueStoreRepairer::fix_shared_blob( |
16368 | KeyValueDB *db, | |
16369 | uint64_t sbid, | |
16370 | const bufferlist* bl) | |
16371 | { | |
b3b6e05e | 16372 | std::lock_guard l(lock); // possibly redundant |
11fdf7f2 TL |
16373 | KeyValueDB::Transaction txn; |
16374 | if (fix_misreferences_txn) { // reuse this txn | |
16375 | txn = fix_misreferences_txn; | |
16376 | } else { | |
16377 | if (!fix_shared_blob_txn) { | |
16378 | fix_shared_blob_txn = db->get_transaction(); | |
16379 | } | |
16380 | txn = fix_shared_blob_txn; | |
16381 | } | |
16382 | string key; | |
16383 | get_shared_blob_key(sbid, &key); | |
16384 | ||
16385 | ++to_repair_cnt; | |
16386 | if (bl) { | |
16387 | txn->set(PREFIX_SHARED_BLOB, key, *bl); | |
16388 | } else { | |
16389 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
16390 | } | |
16391 | return true; | |
16392 | } | |
16393 | ||
16394 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
16395 | const string& key, | |
16396 | const store_statfs_t& new_statfs) | |
16397 | { | |
b3b6e05e | 16398 | std::lock_guard l(lock); |
11fdf7f2 TL |
16399 | if (!fix_statfs_txn) { |
16400 | fix_statfs_txn = db->get_transaction(); | |
16401 | } | |
16402 | BlueStore::volatile_statfs vstatfs; | |
16403 | vstatfs = new_statfs; | |
16404 | bufferlist bl; | |
16405 | vstatfs.encode(bl); | |
16406 | ++to_repair_cnt; | |
16407 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
16408 | return true; | |
16409 | } | |
16410 | ||
16411 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
16412 | FreelistManager* fm, | |
16413 | uint64_t offset, uint64_t len) | |
16414 | { | |
b3b6e05e | 16415 | std::lock_guard l(lock); |
11fdf7f2 TL |
16416 | if (!fix_fm_leaked_txn) { |
16417 | fix_fm_leaked_txn = db->get_transaction(); | |
16418 | } | |
16419 | ++to_repair_cnt; | |
16420 | fm->release(offset, len, fix_fm_leaked_txn); | |
16421 | return true; | |
16422 | } | |
16423 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
16424 | FreelistManager* fm, | |
16425 | uint64_t offset, uint64_t len) | |
16426 | { | |
b3b6e05e | 16427 | std::lock_guard l(lock); |
11fdf7f2 TL |
16428 | if (!fix_fm_false_free_txn) { |
16429 | fix_fm_false_free_txn = db->get_transaction(); | |
16430 | } | |
16431 | ++to_repair_cnt; | |
16432 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
16433 | return true; | |
16434 | } | |
16435 | ||
b3b6e05e TL |
16436 | bool BlueStoreRepairer::fix_spanning_blobs( |
16437 | KeyValueDB* db, | |
16438 | std::function<void(KeyValueDB::Transaction)> f) | |
adb31ebb | 16439 | { |
b3b6e05e | 16440 | std::lock_guard l(lock); |
adb31ebb TL |
16441 | if (!fix_onode_txn) { |
16442 | fix_onode_txn = db->get_transaction(); | |
16443 | } | |
b3b6e05e | 16444 | f(fix_onode_txn); |
adb31ebb | 16445 | ++to_repair_cnt; |
b3b6e05e | 16446 | return true; |
adb31ebb TL |
16447 | } |
16448 | ||
11fdf7f2 TL |
16449 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) |
16450 | { | |
b3b6e05e | 16451 | //NB: not for use in multithreading mode!!! |
11fdf7f2 TL |
16452 | if (misreferenced_extents.size()) { |
16453 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
16454 | ceph_assert(n > 0); | |
16455 | if (!fix_misreferences_txn) { | |
16456 | fix_misreferences_txn = db->get_transaction(); | |
16457 | } | |
16458 | return true; | |
16459 | } | |
16460 | return false; | |
16461 | } | |
16462 | ||
16463 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
16464 | { | |
b3b6e05e | 16465 | //NB: not for use in multithreading mode!!! |
9f95a23c TL |
16466 | if (fix_per_pool_omap_txn) { |
16467 | db->submit_transaction_sync(fix_per_pool_omap_txn); | |
16468 | fix_per_pool_omap_txn = nullptr; | |
16469 | } | |
11fdf7f2 TL |
16470 | if (fix_fm_leaked_txn) { |
16471 | db->submit_transaction_sync(fix_fm_leaked_txn); | |
16472 | fix_fm_leaked_txn = nullptr; | |
16473 | } | |
16474 | if (fix_fm_false_free_txn) { | |
16475 | db->submit_transaction_sync(fix_fm_false_free_txn); | |
16476 | fix_fm_false_free_txn = nullptr; | |
16477 | } | |
16478 | if (remove_key_txn) { | |
16479 | db->submit_transaction_sync(remove_key_txn); | |
16480 | remove_key_txn = nullptr; | |
16481 | } | |
16482 | if (fix_misreferences_txn) { | |
16483 | db->submit_transaction_sync(fix_misreferences_txn); | |
16484 | fix_misreferences_txn = nullptr; | |
16485 | } | |
adb31ebb TL |
16486 | if (fix_onode_txn) { |
16487 | db->submit_transaction_sync(fix_onode_txn); | |
16488 | fix_onode_txn = nullptr; | |
16489 | } | |
11fdf7f2 TL |
16490 | if (fix_shared_blob_txn) { |
16491 | db->submit_transaction_sync(fix_shared_blob_txn); | |
16492 | fix_shared_blob_txn = nullptr; | |
16493 | } | |
16494 | ||
16495 | if (fix_statfs_txn) { | |
16496 | db->submit_transaction_sync(fix_statfs_txn); | |
16497 | fix_statfs_txn = nullptr; | |
16498 | } | |
522d829b TL |
16499 | if (need_compact) { |
16500 | db->compact(); | |
16501 | need_compact = false; | |
16502 | } | |
11fdf7f2 TL |
16503 | unsigned repaired = to_repair_cnt; |
16504 | to_repair_cnt = 0; | |
16505 | return repaired; | |
16506 | } | |
16507 | ||
16508 | // ======================================================= | |
9f95a23c TL |
16509 | // RocksDBBlueFSVolumeSelector |
16510 | ||
16511 | uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) { | |
16512 | ceph_assert(h != nullptr); | |
16513 | uint64_t hint = reinterpret_cast<uint64_t>(h); | |
16514 | uint8_t res; | |
16515 | switch (hint) { | |
16516 | case LEVEL_SLOW: | |
16517 | res = BlueFS::BDEV_SLOW; | |
16518 | if (db_avail4slow > 0) { | |
16519 | // considering statically available db space vs. | |
16520 | // - observed maximums on DB dev for DB/WAL/UNSORTED data | |
16521 | // - observed maximum spillovers | |
16522 | uint64_t max_db_use = 0; // max db usage we potentially observed | |
f6b5b4d7 | 16523 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_LOG - LEVEL_FIRST); |
9f95a23c TL |
16524 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_WAL - LEVEL_FIRST); |
16525 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_DB - LEVEL_FIRST); | |
16526 | // this could go to db hence using it in the estimation | |
16527 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_SLOW, LEVEL_DB - LEVEL_FIRST); | |
16528 | ||
16529 | auto db_total = l_totals[LEVEL_DB - LEVEL_FIRST]; | |
16530 | uint64_t avail = min( | |
16531 | db_avail4slow, | |
16532 | max_db_use < db_total ? db_total - max_db_use : 0); | |
16533 | ||
16534 | // considering current DB dev usage for SLOW data | |
16535 | if (avail > per_level_per_dev_usage.at(BlueFS::BDEV_DB, LEVEL_SLOW - LEVEL_FIRST)) { | |
16536 | res = BlueFS::BDEV_DB; | |
16537 | } | |
16538 | } | |
16539 | break; | |
f6b5b4d7 | 16540 | case LEVEL_LOG: |
9f95a23c TL |
16541 | case LEVEL_WAL: |
16542 | res = BlueFS::BDEV_WAL; | |
16543 | break; | |
16544 | case LEVEL_DB: | |
16545 | default: | |
16546 | res = BlueFS::BDEV_DB; | |
16547 | break; | |
16548 | } | |
16549 | return res; | |
16550 | } | |
16551 | ||
16552 | void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const | |
16553 | { | |
a4b75251 TL |
16554 | auto db_size = l_totals[LEVEL_DB - LEVEL_FIRST]; |
16555 | res.emplace_back(base, db_size); | |
16556 | auto slow_size = l_totals[LEVEL_SLOW - LEVEL_FIRST]; | |
16557 | if (slow_size == 0) { | |
16558 | slow_size = db_size; | |
16559 | } | |
16560 | res.emplace_back(base + ".slow", slow_size); | |
9f95a23c TL |
16561 | } |
16562 | ||
b3b6e05e | 16563 | void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(std::string_view dirname) const { |
9f95a23c TL |
16564 | uint8_t res = LEVEL_DB; |
16565 | if (dirname.length() > 5) { | |
16566 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
16567 | // match up with bluestore. the slow device is always the second | |
16568 | // one (when a dedicated block.db device is present and used at | |
16569 | // bdev 0). the wal device is always last. | |
16570 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
16571 | res = LEVEL_SLOW; | |
16572 | } | |
16573 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
16574 | res = LEVEL_WAL; | |
16575 | } | |
16576 | } | |
16577 | return reinterpret_cast<void*>(res); | |
16578 | } | |
16579 | ||
16580 | void RocksDBBlueFSVolumeSelector::dump(ostream& sout) { | |
16581 | auto max_x = per_level_per_dev_usage.get_max_x(); | |
16582 | auto max_y = per_level_per_dev_usage.get_max_y(); | |
16583 | sout << "RocksDBBlueFSVolumeSelector: wal_total:" << l_totals[LEVEL_WAL - LEVEL_FIRST] | |
16584 | << ", db_total:" << l_totals[LEVEL_DB - LEVEL_FIRST] | |
16585 | << ", slow_total:" << l_totals[LEVEL_SLOW - LEVEL_FIRST] | |
16586 | << ", db_avail:" << db_avail4slow << std::endl | |
16587 | << "Usage matrix:" << std::endl; | |
f6b5b4d7 | 16588 | constexpr std::array<const char*, 8> names{ { |
9f95a23c TL |
16589 | "DEV/LEV", |
16590 | "WAL", | |
16591 | "DB", | |
16592 | "SLOW", | |
16593 | "*", | |
16594 | "*", | |
f6b5b4d7 TL |
16595 | "REAL", |
16596 | "FILES", | |
9f95a23c TL |
16597 | } }; |
16598 | const size_t width = 12; | |
16599 | for (size_t i = 0; i < names.size(); ++i) { | |
16600 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16601 | sout.width(width); | |
16602 | sout << names[i]; | |
16603 | } | |
16604 | sout << std::endl; | |
16605 | for (size_t l = 0; l < max_y; l++) { | |
16606 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16607 | sout.width(width); | |
16608 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
16609 | case LEVEL_LOG: |
16610 | sout << "LOG"; break; | |
9f95a23c TL |
16611 | case LEVEL_WAL: |
16612 | sout << "WAL"; break; | |
16613 | case LEVEL_DB: | |
16614 | sout << "DB"; break; | |
16615 | case LEVEL_SLOW: | |
16616 | sout << "SLOW"; break; | |
16617 | case LEVEL_MAX: | |
16618 | sout << "TOTALS"; break; | |
16619 | } | |
f6b5b4d7 | 16620 | for (size_t d = 0; d < max_x; d++) { |
9f95a23c TL |
16621 | sout.setf(std::ios::left, std::ios::adjustfield); |
16622 | sout.width(width); | |
16623 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(d, l))); | |
16624 | } | |
16625 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16626 | sout.width(width); | |
f6b5b4d7 | 16627 | sout << stringify(per_level_files[l]) << std::endl; |
9f95a23c TL |
16628 | } |
16629 | ceph_assert(max_x == per_level_per_dev_max.get_max_x()); | |
16630 | ceph_assert(max_y == per_level_per_dev_max.get_max_y()); | |
16631 | sout << "MAXIMUMS:" << std::endl; | |
16632 | for (size_t l = 0; l < max_y; l++) { | |
16633 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16634 | sout.width(width); | |
16635 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
16636 | case LEVEL_LOG: |
16637 | sout << "LOG"; break; | |
9f95a23c TL |
16638 | case LEVEL_WAL: |
16639 | sout << "WAL"; break; | |
16640 | case LEVEL_DB: | |
16641 | sout << "DB"; break; | |
16642 | case LEVEL_SLOW: | |
16643 | sout << "SLOW"; break; | |
16644 | case LEVEL_MAX: | |
16645 | sout << "TOTALS"; break; | |
16646 | } | |
16647 | for (size_t d = 0; d < max_x - 1; d++) { | |
16648 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16649 | sout.width(width); | |
16650 | sout << stringify(byte_u_t(per_level_per_dev_max.at(d, l))); | |
16651 | } | |
16652 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16653 | sout.width(width); | |
16654 | sout << stringify(byte_u_t(per_level_per_dev_max.at(max_x - 1, l))); | |
16655 | if (l < max_y - 1) { | |
16656 | sout << std::endl; | |
16657 | } | |
16658 | } | |
16659 | } | |
11fdf7f2 | 16660 | |
9f95a23c | 16661 | // ======================================================= |