]>
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 | ||
9f95a23c | 26 | #include "bluestore_common.h" |
7c673cae FG |
27 | #include "BlueStore.h" |
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" | |
7c673cae | 47 | |
9f95a23c TL |
48 | #if defined(WITH_LTTNG) |
49 | #define TRACEPOINT_DEFINE | |
50 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
51 | #include "tracing/bluestore.h" | |
52 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
53 | #undef TRACEPOINT_DEFINE | |
54 | #else | |
55 | #define tracepoint(...) | |
56 | #endif | |
57 | ||
7c673cae FG |
58 | #define dout_context cct |
59 | #define dout_subsys ceph_subsys_bluestore | |
60 | ||
31f18b77 FG |
61 | using bid_t = decltype(BlueStore::Blob::id); |
62 | ||
63 | // bluestore_cache_onode | |
7c673cae | 64 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Onode, bluestore_onode, |
31f18b77 | 65 | bluestore_cache_onode); |
7c673cae | 66 | |
31f18b77 | 67 | // bluestore_cache_other |
7c673cae | 68 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Buffer, bluestore_buffer, |
f91f0fd5 | 69 | bluestore_Buffer); |
7c673cae | 70 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Extent, bluestore_extent, |
f91f0fd5 | 71 | bluestore_Extent); |
7c673cae | 72 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Blob, bluestore_blob, |
f91f0fd5 | 73 | bluestore_Blob); |
7c673cae | 74 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::SharedBlob, bluestore_shared_blob, |
f91f0fd5 | 75 | bluestore_SharedBlob); |
31f18b77 FG |
76 | |
77 | // bluestore_txc | |
78 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::TransContext, bluestore_transcontext, | |
79 | bluestore_txc); | |
80 | ||
7c673cae FG |
81 | |
82 | // kv store prefixes | |
11fdf7f2 TL |
83 | const string PREFIX_SUPER = "S"; // field -> value |
84 | const string PREFIX_STAT = "T"; // field -> value(int64 array) | |
85 | const string PREFIX_COLL = "C"; // collection name -> cnode_t | |
86 | const string PREFIX_OBJ = "O"; // object name -> onode_t | |
87 | const string PREFIX_OMAP = "M"; // u64 + keyname -> value | |
88 | const string PREFIX_PGMETA_OMAP = "P"; // u64 + keyname -> value(for meta coll) | |
9f95a23c | 89 | const string PREFIX_PERPOOL_OMAP = "m"; // s64 + u64 + keyname -> value |
11fdf7f2 TL |
90 | const string PREFIX_DEFERRED = "L"; // id -> deferred_transaction_t |
91 | const string PREFIX_ALLOC = "B"; // u64 offset -> u64 length (freelist) | |
92 | const string PREFIX_ALLOC_BITMAP = "b";// (see BitmapFreelistManager) | |
7c673cae FG |
93 | const string PREFIX_SHARED_BLOB = "X"; // u64 offset -> shared_blob_t |
94 | ||
11fdf7f2 TL |
95 | const string BLUESTORE_GLOBAL_STATFS_KEY = "bluestore_statfs"; |
96 | ||
7c673cae FG |
97 | // write a label in the first block. always use this size. note that |
98 | // bluefs makes a matching assumption about the location of its | |
99 | // superblock (always the second block of the device). | |
100 | #define BDEV_LABEL_BLOCK_SIZE 4096 | |
101 | ||
102 | // reserve: label (4k) + bluefs super (4k), which means we start at 8k. | |
103 | #define SUPER_RESERVED 8192 | |
104 | ||
105 | #define OBJECT_MAX_SIZE 0xffffffff // 32 bits | |
106 | ||
107 | ||
108 | /* | |
109 | * extent map blob encoding | |
110 | * | |
111 | * we use the low bits of the blobid field to indicate some common scenarios | |
112 | * and spanning vs local ids. See ExtentMap::{encode,decode}_some(). | |
113 | */ | |
114 | #define BLOBID_FLAG_CONTIGUOUS 0x1 // this extent starts at end of previous | |
115 | #define BLOBID_FLAG_ZEROOFFSET 0x2 // blob_offset is 0 | |
116 | #define BLOBID_FLAG_SAMELENGTH 0x4 // length matches previous extent | |
117 | #define BLOBID_FLAG_SPANNING 0x8 // has spanning blob id | |
118 | #define BLOBID_SHIFT_BITS 4 | |
119 | ||
120 | /* | |
121 | * object name key structure | |
122 | * | |
123 | * encoded u8: shard + 2^7 (so that it sorts properly) | |
124 | * encoded u64: poolid + 2^63 (so that it sorts properly) | |
125 | * encoded u32: hash (bit reversed) | |
126 | * | |
127 | * escaped string: namespace | |
128 | * | |
129 | * escaped string: key or object name | |
130 | * 1 char: '<', '=', or '>'. if =, then object key == object name, and | |
131 | * we are done. otherwise, we are followed by the object name. | |
132 | * escaped string: object name (unless '=' above) | |
133 | * | |
134 | * encoded u64: snap | |
135 | * encoded u64: generation | |
136 | * 'o' | |
137 | */ | |
138 | #define ONODE_KEY_SUFFIX 'o' | |
139 | ||
140 | /* | |
141 | * extent shard key | |
142 | * | |
143 | * object prefix key | |
144 | * u32 | |
145 | * 'x' | |
146 | */ | |
147 | #define EXTENT_SHARD_KEY_SUFFIX 'x' | |
148 | ||
149 | /* | |
150 | * string encoding in the key | |
151 | * | |
152 | * The key string needs to lexicographically sort the same way that | |
153 | * ghobject_t does. We do this by escaping anything <= to '#' with # | |
154 | * plus a 2 digit hex string, and anything >= '~' with ~ plus the two | |
155 | * hex digits. | |
156 | * | |
157 | * We use ! as a terminator for strings; this works because it is < # | |
158 | * and will get escaped if it is present in the string. | |
159 | * | |
f91f0fd5 TL |
160 | * NOTE: There is a bug in this implementation: due to implicit |
161 | * character type conversion in comparison it may produce unexpected | |
162 | * ordering. Unfortunately fixing the bug would mean invalidating the | |
163 | * keys in existing deployments. Instead we do additional sorting | |
164 | * where it is needed. | |
7c673cae FG |
165 | */ |
166 | template<typename S> | |
167 | static void append_escaped(const string &in, S *out) | |
168 | { | |
224ce89b WB |
169 | char hexbyte[in.length() * 3 + 1]; |
170 | char* ptr = &hexbyte[0]; | |
7c673cae | 171 | for (string::const_iterator i = in.begin(); i != in.end(); ++i) { |
f91f0fd5 | 172 | if (*i <= '#') { // bug: unexpected result for *i > 0x7f |
224ce89b WB |
173 | *ptr++ = '#'; |
174 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
175 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
f91f0fd5 | 176 | } else if (*i >= '~') { // bug: unexpected result for *i > 0x7f |
224ce89b WB |
177 | *ptr++ = '~'; |
178 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
179 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 180 | } else { |
224ce89b | 181 | *ptr++ = *i; |
7c673cae FG |
182 | } |
183 | } | |
224ce89b WB |
184 | *ptr++ = '!'; |
185 | out->append(hexbyte, ptr - &hexbyte[0]); | |
186 | } | |
187 | ||
188 | inline unsigned h2i(char c) | |
189 | { | |
190 | if ((c >= '0') && (c <= '9')) { | |
191 | return c - 0x30; | |
192 | } else if ((c >= 'a') && (c <= 'f')) { | |
193 | return c - 'a' + 10; | |
194 | } else if ((c >= 'A') && (c <= 'F')) { | |
195 | return c - 'A' + 10; | |
196 | } else { | |
197 | return 256; // make it always larger than 255 | |
198 | } | |
7c673cae FG |
199 | } |
200 | ||
201 | static int decode_escaped(const char *p, string *out) | |
202 | { | |
224ce89b WB |
203 | char buff[256]; |
204 | char* ptr = &buff[0]; | |
205 | char* max = &buff[252]; | |
7c673cae FG |
206 | const char *orig_p = p; |
207 | while (*p && *p != '!') { | |
208 | if (*p == '#' || *p == '~') { | |
224ce89b WB |
209 | unsigned hex = 0; |
210 | p++; | |
211 | hex = h2i(*p++) << 4; | |
212 | if (hex > 255) { | |
213 | return -EINVAL; | |
214 | } | |
215 | hex |= h2i(*p++); | |
216 | if (hex > 255) { | |
217 | return -EINVAL; | |
218 | } | |
219 | *ptr++ = hex; | |
7c673cae | 220 | } else { |
224ce89b WB |
221 | *ptr++ = *p++; |
222 | } | |
223 | if (ptr > max) { | |
224 | out->append(buff, ptr-buff); | |
225 | ptr = &buff[0]; | |
7c673cae FG |
226 | } |
227 | } | |
224ce89b WB |
228 | if (ptr != buff) { |
229 | out->append(buff, ptr-buff); | |
230 | } | |
7c673cae FG |
231 | return p - orig_p; |
232 | } | |
233 | ||
234 | // some things we encode in binary (as le32 or le64); print the | |
235 | // resulting key strings nicely | |
236 | template<typename S> | |
237 | static string pretty_binary_string(const S& in) | |
238 | { | |
239 | char buf[10]; | |
240 | string out; | |
241 | out.reserve(in.length() * 3); | |
242 | enum { NONE, HEX, STRING } mode = NONE; | |
243 | unsigned from = 0, i; | |
244 | for (i=0; i < in.length(); ++i) { | |
245 | if ((in[i] < 32 || (unsigned char)in[i] > 126) || | |
246 | (mode == HEX && in.length() - i >= 4 && | |
247 | ((in[i] < 32 || (unsigned char)in[i] > 126) || | |
248 | (in[i+1] < 32 || (unsigned char)in[i+1] > 126) || | |
249 | (in[i+2] < 32 || (unsigned char)in[i+2] > 126) || | |
250 | (in[i+3] < 32 || (unsigned char)in[i+3] > 126)))) { | |
251 | if (mode == STRING) { | |
252 | out.append(in.c_str() + from, i - from); | |
253 | out.push_back('\''); | |
254 | } | |
255 | if (mode != HEX) { | |
256 | out.append("0x"); | |
257 | mode = HEX; | |
258 | } | |
259 | if (in.length() - i >= 4) { | |
260 | // print a whole u32 at once | |
261 | snprintf(buf, sizeof(buf), "%08x", | |
262 | (uint32_t)(((unsigned char)in[i] << 24) | | |
263 | ((unsigned char)in[i+1] << 16) | | |
264 | ((unsigned char)in[i+2] << 8) | | |
265 | ((unsigned char)in[i+3] << 0))); | |
266 | i += 3; | |
267 | } else { | |
268 | snprintf(buf, sizeof(buf), "%02x", (int)(unsigned char)in[i]); | |
269 | } | |
270 | out.append(buf); | |
271 | } else { | |
272 | if (mode != STRING) { | |
273 | out.push_back('\''); | |
274 | mode = STRING; | |
275 | from = i; | |
276 | } | |
277 | } | |
278 | } | |
279 | if (mode == STRING) { | |
280 | out.append(in.c_str() + from, i - from); | |
281 | out.push_back('\''); | |
282 | } | |
283 | return out; | |
284 | } | |
285 | ||
286 | template<typename T> | |
287 | static void _key_encode_shard(shard_id_t shard, T *key) | |
288 | { | |
289 | key->push_back((char)((uint8_t)shard.id + (uint8_t)0x80)); | |
290 | } | |
291 | ||
292 | static const char *_key_decode_shard(const char *key, shard_id_t *pshard) | |
293 | { | |
294 | pshard->id = (uint8_t)*key - (uint8_t)0x80; | |
295 | return key + 1; | |
296 | } | |
297 | ||
f91f0fd5 TL |
298 | static void get_coll_range(const coll_t& cid, int bits, |
299 | ghobject_t *temp_start, ghobject_t *temp_end, | |
300 | ghobject_t *start, ghobject_t *end) | |
7c673cae | 301 | { |
7c673cae FG |
302 | spg_t pgid; |
303 | if (cid.is_pg(&pgid)) { | |
f91f0fd5 | 304 | start->shard_id = pgid.shard; |
7c673cae FG |
305 | *temp_start = *start; |
306 | ||
f91f0fd5 TL |
307 | start->hobj.pool = pgid.pool(); |
308 | temp_start->hobj.pool = -2ll - pgid.pool(); | |
7c673cae FG |
309 | |
310 | *end = *start; | |
311 | *temp_end = *temp_start; | |
312 | ||
313 | uint32_t reverse_hash = hobject_t::_reverse_bits(pgid.ps()); | |
f91f0fd5 TL |
314 | start->hobj.set_bitwise_key_u32(reverse_hash); |
315 | temp_start->hobj.set_bitwise_key_u32(reverse_hash); | |
7c673cae FG |
316 | |
317 | uint64_t end_hash = reverse_hash + (1ull << (32 - bits)); | |
318 | if (end_hash > 0xffffffffull) | |
319 | end_hash = 0xffffffffull; | |
320 | ||
f91f0fd5 TL |
321 | end->hobj.set_bitwise_key_u32(end_hash); |
322 | temp_end->hobj.set_bitwise_key_u32(end_hash); | |
7c673cae | 323 | } else { |
f91f0fd5 TL |
324 | start->shard_id = shard_id_t::NO_SHARD; |
325 | start->hobj.pool = -1ull; | |
326 | ||
7c673cae | 327 | *end = *start; |
f91f0fd5 TL |
328 | start->hobj.set_bitwise_key_u32(0); |
329 | end->hobj.set_bitwise_key_u32(0xffffffff); | |
7c673cae FG |
330 | |
331 | // no separate temp section | |
332 | *temp_start = *end; | |
333 | *temp_end = *end; | |
334 | } | |
f91f0fd5 TL |
335 | |
336 | start->generation = 0; | |
337 | end->generation = 0; | |
338 | temp_start->generation = 0; | |
339 | temp_end->generation = 0; | |
7c673cae FG |
340 | } |
341 | ||
342 | static void get_shared_blob_key(uint64_t sbid, string *key) | |
343 | { | |
344 | key->clear(); | |
345 | _key_encode_u64(sbid, key); | |
346 | } | |
347 | ||
348 | static int get_key_shared_blob(const string& key, uint64_t *sbid) | |
349 | { | |
350 | const char *p = key.c_str(); | |
351 | if (key.length() < sizeof(uint64_t)) | |
352 | return -1; | |
224ce89b | 353 | _key_decode_u64(p, sbid); |
7c673cae FG |
354 | return 0; |
355 | } | |
356 | ||
357 | template<typename S> | |
f91f0fd5 | 358 | static void _key_encode_prefix(const ghobject_t& oid, S *key) |
7c673cae | 359 | { |
f91f0fd5 TL |
360 | _key_encode_shard(oid.shard_id, key); |
361 | _key_encode_u64(oid.hobj.pool + 0x8000000000000000ull, key); | |
362 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), key); | |
363 | } | |
7c673cae | 364 | |
f91f0fd5 TL |
365 | static const char *_key_decode_prefix(const char *p, ghobject_t *oid) |
366 | { | |
7c673cae FG |
367 | p = _key_decode_shard(p, &oid->shard_id); |
368 | ||
369 | uint64_t pool; | |
370 | p = _key_decode_u64(p, &pool); | |
371 | oid->hobj.pool = pool - 0x8000000000000000ull; | |
372 | ||
373 | unsigned hash; | |
374 | p = _key_decode_u32(p, &hash); | |
375 | ||
376 | oid->hobj.set_bitwise_key_u32(hash); | |
377 | ||
f91f0fd5 TL |
378 | return p; |
379 | } | |
380 | ||
381 | #define ENCODED_KEY_PREFIX_LEN (1 + 8 + 4) | |
382 | ||
383 | template<typename S> | |
384 | static int get_key_object(const S& key, ghobject_t *oid) | |
385 | { | |
386 | int r; | |
387 | const char *p = key.c_str(); | |
388 | ||
389 | if (key.length() < ENCODED_KEY_PREFIX_LEN) | |
390 | return -1; | |
391 | ||
392 | p = _key_decode_prefix(p, oid); | |
393 | ||
394 | if (key.length() == ENCODED_KEY_PREFIX_LEN) | |
395 | return -2; | |
396 | ||
7c673cae FG |
397 | r = decode_escaped(p, &oid->hobj.nspace); |
398 | if (r < 0) | |
399 | return -2; | |
400 | p += r + 1; | |
401 | ||
402 | string k; | |
403 | r = decode_escaped(p, &k); | |
404 | if (r < 0) | |
405 | return -3; | |
406 | p += r + 1; | |
407 | if (*p == '=') { | |
408 | // no key | |
409 | ++p; | |
410 | oid->hobj.oid.name = k; | |
411 | } else if (*p == '<' || *p == '>') { | |
412 | // key + name | |
413 | ++p; | |
414 | r = decode_escaped(p, &oid->hobj.oid.name); | |
415 | if (r < 0) | |
416 | return -5; | |
417 | p += r + 1; | |
418 | oid->hobj.set_key(k); | |
419 | } else { | |
420 | // malformed | |
421 | return -6; | |
422 | } | |
423 | ||
424 | p = _key_decode_u64(p, &oid->hobj.snap.val); | |
425 | p = _key_decode_u64(p, &oid->generation); | |
426 | ||
427 | if (*p != ONODE_KEY_SUFFIX) { | |
428 | return -7; | |
429 | } | |
430 | p++; | |
431 | if (*p) { | |
432 | // if we get something other than a null terminator here, | |
433 | // something goes wrong. | |
434 | return -8; | |
435 | } | |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
440 | template<typename S> | |
441 | static void get_object_key(CephContext *cct, const ghobject_t& oid, S *key) | |
442 | { | |
443 | key->clear(); | |
444 | ||
f91f0fd5 | 445 | size_t max_len = ENCODED_KEY_PREFIX_LEN + |
7c673cae FG |
446 | (oid.hobj.nspace.length() * 3 + 1) + |
447 | (oid.hobj.get_key().length() * 3 + 1) + | |
448 | 1 + // for '<', '=', or '>' | |
449 | (oid.hobj.oid.name.length() * 3 + 1) + | |
450 | 8 + 8 + 1; | |
451 | key->reserve(max_len); | |
452 | ||
f91f0fd5 | 453 | _key_encode_prefix(oid, key); |
7c673cae FG |
454 | |
455 | append_escaped(oid.hobj.nspace, key); | |
456 | ||
457 | if (oid.hobj.get_key().length()) { | |
458 | // is a key... could be < = or >. | |
459 | append_escaped(oid.hobj.get_key(), key); | |
460 | // (ASCII chars < = and > sort in that order, yay) | |
461 | int r = oid.hobj.get_key().compare(oid.hobj.oid.name); | |
462 | if (r) { | |
463 | key->append(r > 0 ? ">" : "<"); | |
464 | append_escaped(oid.hobj.oid.name, key); | |
465 | } else { | |
466 | // same as no key | |
467 | key->append("="); | |
468 | } | |
469 | } else { | |
470 | // no key | |
471 | append_escaped(oid.hobj.oid.name, key); | |
472 | key->append("="); | |
473 | } | |
474 | ||
475 | _key_encode_u64(oid.hobj.snap, key); | |
476 | _key_encode_u64(oid.generation, key); | |
477 | ||
478 | key->push_back(ONODE_KEY_SUFFIX); | |
479 | ||
480 | // sanity check | |
481 | if (true) { | |
482 | ghobject_t t; | |
483 | int r = get_key_object(*key, &t); | |
484 | if (r || t != oid) { | |
485 | derr << " r " << r << dendl; | |
486 | derr << "key " << pretty_binary_string(*key) << dendl; | |
487 | derr << "oid " << oid << dendl; | |
488 | derr << " t " << t << dendl; | |
11fdf7f2 | 489 | ceph_assert(r == 0 && t == oid); |
7c673cae FG |
490 | } |
491 | } | |
492 | } | |
493 | ||
494 | ||
495 | // extent shard keys are the onode key, plus a u32, plus 'x'. the trailing | |
496 | // char lets us quickly test whether it is a shard key without decoding any | |
497 | // of the prefix bytes. | |
498 | template<typename S> | |
499 | static void get_extent_shard_key(const S& onode_key, uint32_t offset, | |
500 | string *key) | |
501 | { | |
502 | key->clear(); | |
503 | key->reserve(onode_key.length() + 4 + 1); | |
504 | key->append(onode_key.c_str(), onode_key.size()); | |
505 | _key_encode_u32(offset, key); | |
506 | key->push_back(EXTENT_SHARD_KEY_SUFFIX); | |
507 | } | |
508 | ||
509 | static void rewrite_extent_shard_key(uint32_t offset, string *key) | |
510 | { | |
11fdf7f2 TL |
511 | ceph_assert(key->size() > sizeof(uint32_t) + 1); |
512 | ceph_assert(*key->rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
513 | _key_encode_u32(offset, key->size() - sizeof(uint32_t) - 1, key); |
514 | } | |
515 | ||
516 | template<typename S> | |
517 | static void generate_extent_shard_key_and_apply( | |
518 | const S& onode_key, | |
519 | uint32_t offset, | |
520 | string *key, | |
521 | std::function<void(const string& final_key)> apply) | |
522 | { | |
523 | if (key->empty()) { // make full key | |
11fdf7f2 | 524 | ceph_assert(!onode_key.empty()); |
7c673cae FG |
525 | get_extent_shard_key(onode_key, offset, key); |
526 | } else { | |
527 | rewrite_extent_shard_key(offset, key); | |
528 | } | |
529 | apply(*key); | |
530 | } | |
531 | ||
532 | int get_key_extent_shard(const string& key, string *onode_key, uint32_t *offset) | |
533 | { | |
11fdf7f2 TL |
534 | ceph_assert(key.size() > sizeof(uint32_t) + 1); |
535 | ceph_assert(*key.rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
536 | int okey_len = key.size() - sizeof(uint32_t) - 1; |
537 | *onode_key = key.substr(0, okey_len); | |
538 | const char *p = key.data() + okey_len; | |
224ce89b | 539 | _key_decode_u32(p, offset); |
7c673cae FG |
540 | return 0; |
541 | } | |
542 | ||
543 | static bool is_extent_shard_key(const string& key) | |
544 | { | |
545 | return *key.rbegin() == EXTENT_SHARD_KEY_SUFFIX; | |
546 | } | |
547 | ||
7c673cae FG |
548 | static void get_deferred_key(uint64_t seq, string *out) |
549 | { | |
550 | _key_encode_u64(seq, out); | |
551 | } | |
552 | ||
11fdf7f2 TL |
553 | static void get_pool_stat_key(int64_t pool_id, string *key) |
554 | { | |
555 | key->clear(); | |
556 | _key_encode_u64(pool_id, key); | |
557 | } | |
558 | ||
559 | static int get_key_pool_stat(const string& key, uint64_t* pool_id) | |
560 | { | |
561 | const char *p = key.c_str(); | |
562 | if (key.length() < sizeof(uint64_t)) | |
563 | return -1; | |
564 | _key_decode_u64(p, pool_id); | |
565 | return 0; | |
566 | } | |
7c673cae | 567 | |
81eedcae TL |
568 | template <int LogLevelV> |
569 | void _dump_extent_map(CephContext *cct, const BlueStore::ExtentMap &em) | |
570 | { | |
571 | uint64_t pos = 0; | |
572 | for (auto& s : em.shards) { | |
573 | dout(LogLevelV) << __func__ << " shard " << *s.shard_info | |
574 | << (s.loaded ? " (loaded)" : "") | |
575 | << (s.dirty ? " (dirty)" : "") | |
576 | << dendl; | |
577 | } | |
578 | for (auto& e : em.extent_map) { | |
579 | dout(LogLevelV) << __func__ << " " << e << dendl; | |
580 | ceph_assert(e.logical_offset >= pos); | |
581 | pos = e.logical_offset + e.length; | |
582 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
583 | if (blob.has_csum()) { | |
584 | vector<uint64_t> v; | |
585 | unsigned n = blob.get_csum_count(); | |
586 | for (unsigned i = 0; i < n; ++i) | |
587 | v.push_back(blob.get_csum_item(i)); | |
588 | dout(LogLevelV) << __func__ << " csum: " << std::hex << v << std::dec | |
589 | << dendl; | |
590 | } | |
591 | std::lock_guard l(e.blob->shared_blob->get_cache()->lock); | |
592 | for (auto& i : e.blob->shared_blob->bc.buffer_map) { | |
593 | dout(LogLevelV) << __func__ << " 0x" << std::hex << i.first | |
594 | << "~" << i.second->length << std::dec | |
595 | << " " << *i.second << dendl; | |
596 | } | |
597 | } | |
598 | } | |
599 | ||
600 | template <int LogLevelV> | |
601 | void _dump_onode(CephContext *cct, const BlueStore::Onode& o) | |
602 | { | |
603 | if (!cct->_conf->subsys.should_gather<ceph_subsys_bluestore, LogLevelV>()) | |
604 | return; | |
605 | dout(LogLevelV) << __func__ << " " << &o << " " << o.oid | |
606 | << " nid " << o.onode.nid | |
607 | << " size 0x" << std::hex << o.onode.size | |
608 | << " (" << std::dec << o.onode.size << ")" | |
609 | << " expected_object_size " << o.onode.expected_object_size | |
610 | << " expected_write_size " << o.onode.expected_write_size | |
611 | << " in " << o.onode.extent_map_shards.size() << " shards" | |
612 | << ", " << o.extent_map.spanning_blob_map.size() | |
613 | << " spanning blobs" | |
614 | << dendl; | |
615 | for (auto p = o.onode.attrs.begin(); | |
616 | p != o.onode.attrs.end(); | |
617 | ++p) { | |
618 | dout(LogLevelV) << __func__ << " attr " << p->first | |
619 | << " len " << p->second.length() << dendl; | |
620 | } | |
621 | _dump_extent_map<LogLevelV>(cct, o.extent_map); | |
622 | } | |
623 | ||
624 | template <int LogLevelV> | |
625 | void _dump_transaction(CephContext *cct, ObjectStore::Transaction *t) | |
626 | { | |
627 | dout(LogLevelV) << __func__ << " transaction dump:\n"; | |
628 | JSONFormatter f(true); | |
629 | f.open_object_section("transaction"); | |
630 | t->dump(&f); | |
631 | f.close_section(); | |
632 | f.flush(*_dout); | |
633 | *_dout << dendl; | |
634 | } | |
635 | ||
7c673cae FG |
636 | // merge operators |
637 | ||
638 | struct Int64ArrayMergeOperator : public KeyValueDB::MergeOperator { | |
639 | void merge_nonexistent( | |
640 | const char *rdata, size_t rlen, std::string *new_value) override { | |
641 | *new_value = std::string(rdata, rlen); | |
642 | } | |
643 | void merge( | |
644 | const char *ldata, size_t llen, | |
645 | const char *rdata, size_t rlen, | |
646 | std::string *new_value) override { | |
11fdf7f2 TL |
647 | ceph_assert(llen == rlen); |
648 | ceph_assert((rlen % 8) == 0); | |
7c673cae | 649 | new_value->resize(rlen); |
eafe8130 TL |
650 | const ceph_le64* lv = (const ceph_le64*)ldata; |
651 | const ceph_le64* rv = (const ceph_le64*)rdata; | |
652 | ceph_le64* nv = &(ceph_le64&)new_value->at(0); | |
7c673cae FG |
653 | for (size_t i = 0; i < rlen >> 3; ++i) { |
654 | nv[i] = lv[i] + rv[i]; | |
655 | } | |
656 | } | |
657 | // We use each operator name and each prefix to construct the | |
658 | // overall RocksDB operator name for consistency check at open time. | |
91327a77 | 659 | const char *name() const override { |
7c673cae FG |
660 | return "int64_array"; |
661 | } | |
662 | }; | |
663 | ||
664 | ||
665 | // Buffer | |
666 | ||
667 | ostream& operator<<(ostream& out, const BlueStore::Buffer& b) | |
668 | { | |
669 | out << "buffer(" << &b << " space " << b.space << " 0x" << std::hex | |
670 | << b.offset << "~" << b.length << std::dec | |
671 | << " " << BlueStore::Buffer::get_state_name(b.state); | |
672 | if (b.flags) | |
673 | out << " " << BlueStore::Buffer::get_flag_name(b.flags); | |
674 | return out << ")"; | |
675 | } | |
676 | ||
f91f0fd5 TL |
677 | namespace { |
678 | ||
679 | /* | |
680 | * Due to a bug in key string encoding (see a comment for append_escaped) | |
681 | * the KeyValueDB iterator does not lexicographically sort the same | |
682 | * way that ghobject_t does: objects with the same hash may have wrong order. | |
683 | * | |
684 | * This is the iterator wrapper that fixes the keys order. | |
685 | */ | |
686 | ||
687 | class CollectionListIterator { | |
688 | public: | |
689 | CollectionListIterator(const KeyValueDB::Iterator &it) | |
690 | : m_it(it) { | |
691 | } | |
692 | virtual ~CollectionListIterator() { | |
693 | } | |
694 | ||
695 | virtual bool valid() const = 0; | |
696 | virtual const ghobject_t &oid() const = 0; | |
697 | virtual void lower_bound(const ghobject_t &oid) = 0; | |
698 | virtual void upper_bound(const ghobject_t &oid) = 0; | |
699 | virtual void next() = 0; | |
700 | ||
701 | protected: | |
702 | KeyValueDB::Iterator m_it; | |
703 | }; | |
704 | ||
705 | class SimpleCollectionListIterator : public CollectionListIterator { | |
706 | public: | |
707 | SimpleCollectionListIterator(CephContext *cct, const KeyValueDB::Iterator &it) | |
708 | : CollectionListIterator(it), m_cct(cct) { | |
709 | } | |
710 | ||
711 | bool valid() const override { | |
712 | return m_it->valid(); | |
713 | } | |
714 | ||
715 | const ghobject_t &oid() const override { | |
716 | ceph_assert(valid()); | |
717 | ||
718 | return m_oid; | |
719 | } | |
720 | ||
721 | void lower_bound(const ghobject_t &oid) override { | |
722 | string key; | |
723 | get_object_key(m_cct, oid, &key); | |
724 | ||
725 | m_it->lower_bound(key); | |
726 | get_oid(); | |
727 | } | |
728 | ||
729 | void upper_bound(const ghobject_t &oid) override { | |
730 | string key; | |
731 | get_object_key(m_cct, oid, &key); | |
732 | ||
733 | m_it->upper_bound(key); | |
734 | get_oid(); | |
735 | } | |
736 | ||
737 | void next() override { | |
738 | ceph_assert(valid()); | |
739 | ||
740 | m_it->next(); | |
741 | get_oid(); | |
742 | } | |
743 | ||
744 | private: | |
745 | CephContext *m_cct; | |
746 | ghobject_t m_oid; | |
747 | ||
748 | void get_oid() { | |
749 | if (!valid()) { | |
750 | return; | |
751 | } | |
752 | ||
753 | if (is_extent_shard_key(m_it->key())) { | |
754 | next(); | |
755 | return; | |
756 | } | |
757 | ||
758 | m_oid = ghobject_t(); | |
759 | int r = get_key_object(m_it->key(), &m_oid); | |
760 | ceph_assert(r == 0); | |
761 | } | |
762 | }; | |
763 | ||
764 | class SortedCollectionListIterator : public CollectionListIterator { | |
765 | public: | |
766 | SortedCollectionListIterator(const KeyValueDB::Iterator &it) | |
767 | : CollectionListIterator(it), m_chunk_iter(m_chunk.end()) { | |
768 | } | |
769 | ||
770 | bool valid() const override { | |
771 | return m_chunk_iter != m_chunk.end(); | |
772 | } | |
773 | ||
774 | const ghobject_t &oid() const override { | |
775 | ceph_assert(valid()); | |
776 | ||
777 | return m_chunk_iter->first; | |
778 | } | |
779 | ||
780 | void lower_bound(const ghobject_t &oid) override { | |
781 | std::string key; | |
782 | _key_encode_prefix(oid, &key); | |
783 | ||
784 | m_it->lower_bound(key); | |
785 | m_chunk_iter = m_chunk.end(); | |
786 | if (!get_next_chunk()) { | |
787 | return; | |
788 | } | |
789 | ||
790 | if (this->oid().shard_id != oid.shard_id || | |
791 | this->oid().hobj.pool != oid.hobj.pool || | |
792 | this->oid().hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
793 | return; | |
794 | } | |
795 | ||
796 | m_chunk_iter = m_chunk.lower_bound(oid); | |
797 | if (m_chunk_iter == m_chunk.end()) { | |
798 | get_next_chunk(); | |
799 | } | |
800 | } | |
801 | ||
802 | void upper_bound(const ghobject_t &oid) override { | |
803 | lower_bound(oid); | |
804 | ||
805 | if (valid() && this->oid() == oid) { | |
806 | next(); | |
807 | } | |
808 | } | |
809 | ||
810 | void next() override { | |
811 | ceph_assert(valid()); | |
812 | ||
813 | m_chunk_iter++; | |
814 | if (m_chunk_iter == m_chunk.end()) { | |
815 | get_next_chunk(); | |
816 | } | |
817 | } | |
818 | ||
819 | private: | |
820 | std::map<ghobject_t, std::string> m_chunk; | |
821 | std::map<ghobject_t, std::string>::iterator m_chunk_iter; | |
822 | ||
823 | bool get_next_chunk() { | |
824 | while (m_it->valid() && is_extent_shard_key(m_it->key())) { | |
825 | m_it->next(); | |
826 | } | |
827 | ||
828 | if (!m_it->valid()) { | |
829 | return false; | |
830 | } | |
831 | ||
832 | ghobject_t oid; | |
833 | int r = get_key_object(m_it->key(), &oid); | |
834 | ceph_assert(r == 0); | |
835 | ||
836 | m_chunk.clear(); | |
837 | while (true) { | |
838 | m_chunk.insert({oid, m_it->key()}); | |
839 | ||
840 | do { | |
841 | m_it->next(); | |
842 | } while (m_it->valid() && is_extent_shard_key(m_it->key())); | |
843 | ||
844 | if (!m_it->valid()) { | |
845 | break; | |
846 | } | |
847 | ||
848 | ghobject_t next; | |
849 | r = get_key_object(m_it->key(), &next); | |
850 | ceph_assert(r == 0); | |
851 | if (next.shard_id != oid.shard_id || | |
852 | next.hobj.pool != oid.hobj.pool || | |
853 | next.hobj.get_bitwise_key_u32() != oid.hobj.get_bitwise_key_u32()) { | |
854 | break; | |
855 | } | |
856 | oid = next; | |
857 | } | |
858 | ||
859 | m_chunk_iter = m_chunk.begin(); | |
860 | return true; | |
861 | } | |
862 | }; | |
863 | ||
864 | } // anonymous namespace | |
865 | ||
7c673cae FG |
866 | // Garbage Collector |
867 | ||
868 | void BlueStore::GarbageCollector::process_protrusive_extents( | |
869 | const BlueStore::ExtentMap& extent_map, | |
870 | uint64_t start_offset, | |
871 | uint64_t end_offset, | |
872 | uint64_t start_touch_offset, | |
873 | uint64_t end_touch_offset, | |
874 | uint64_t min_alloc_size) | |
875 | { | |
11fdf7f2 | 876 | ceph_assert(start_offset <= start_touch_offset && end_offset>= end_touch_offset); |
7c673cae | 877 | |
11fdf7f2 TL |
878 | uint64_t lookup_start_offset = p2align(start_offset, min_alloc_size); |
879 | uint64_t lookup_end_offset = round_up_to(end_offset, min_alloc_size); | |
7c673cae FG |
880 | |
881 | dout(30) << __func__ << " (hex): [" << std::hex | |
882 | << lookup_start_offset << ", " << lookup_end_offset | |
883 | << ")" << std::dec << dendl; | |
884 | ||
885 | for (auto it = extent_map.seek_lextent(lookup_start_offset); | |
886 | it != extent_map.extent_map.end() && | |
887 | it->logical_offset < lookup_end_offset; | |
888 | ++it) { | |
889 | uint64_t alloc_unit_start = it->logical_offset / min_alloc_size; | |
890 | uint64_t alloc_unit_end = (it->logical_end() - 1) / min_alloc_size; | |
891 | ||
892 | dout(30) << __func__ << " " << *it | |
893 | << "alloc_units: " << alloc_unit_start << ".." << alloc_unit_end | |
894 | << dendl; | |
895 | ||
896 | Blob* b = it->blob.get(); | |
897 | ||
898 | if (it->logical_offset >=start_touch_offset && | |
899 | it->logical_end() <= end_touch_offset) { | |
900 | // Process extents within the range affected by | |
901 | // the current write request. | |
902 | // Need to take into account if existing extents | |
903 | // can be merged with them (uncompressed case) | |
904 | if (!b->get_blob().is_compressed()) { | |
905 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
906 | --blob_info_counted->expected_allocations; // don't need to allocate | |
907 | // new AU for compressed | |
908 | // data since another | |
909 | // collocated uncompressed | |
910 | // blob already exists | |
911 | dout(30) << __func__ << " --expected:" | |
912 | << alloc_unit_start << dendl; | |
913 | } | |
914 | used_alloc_unit = alloc_unit_end; | |
915 | blob_info_counted = nullptr; | |
916 | } | |
917 | } else if (b->get_blob().is_compressed()) { | |
918 | ||
919 | // additionally we take compressed blobs that were not impacted | |
920 | // by the write into account too | |
921 | BlobInfo& bi = | |
922 | affected_blobs.emplace( | |
923 | b, BlobInfo(b->get_referenced_bytes())).first->second; | |
924 | ||
925 | int adjust = | |
926 | (used_alloc_unit && used_alloc_unit == alloc_unit_start) ? 0 : 1; | |
927 | bi.expected_allocations += alloc_unit_end - alloc_unit_start + adjust; | |
928 | dout(30) << __func__ << " expected_allocations=" | |
929 | << bi.expected_allocations << " end_au:" | |
930 | << alloc_unit_end << dendl; | |
931 | ||
932 | blob_info_counted = &bi; | |
933 | used_alloc_unit = alloc_unit_end; | |
934 | ||
11fdf7f2 | 935 | ceph_assert(it->length <= bi.referenced_bytes); |
7c673cae FG |
936 | bi.referenced_bytes -= it->length; |
937 | dout(30) << __func__ << " affected_blob:" << *b | |
938 | << " unref 0x" << std::hex << it->length | |
939 | << " referenced = 0x" << bi.referenced_bytes | |
940 | << std::dec << dendl; | |
941 | // NOTE: we can't move specific blob to resulting GC list here | |
942 | // when reference counter == 0 since subsequent extents might | |
943 | // decrement its expected_allocation. | |
944 | // Hence need to enumerate all the extents first. | |
945 | if (!bi.collect_candidate) { | |
946 | bi.first_lextent = it; | |
947 | bi.collect_candidate = true; | |
948 | } | |
949 | bi.last_lextent = it; | |
950 | } else { | |
951 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
952 | // don't need to allocate new AU for compressed data since another | |
953 | // collocated uncompressed blob already exists | |
954 | --blob_info_counted->expected_allocations; | |
955 | dout(30) << __func__ << " --expected_allocations:" | |
956 | << alloc_unit_start << dendl; | |
957 | } | |
958 | used_alloc_unit = alloc_unit_end; | |
959 | blob_info_counted = nullptr; | |
960 | } | |
961 | } | |
962 | ||
963 | for (auto b_it = affected_blobs.begin(); | |
964 | b_it != affected_blobs.end(); | |
965 | ++b_it) { | |
966 | Blob* b = b_it->first; | |
967 | BlobInfo& bi = b_it->second; | |
968 | if (bi.referenced_bytes == 0) { | |
969 | uint64_t len_on_disk = b_it->first->get_blob().get_ondisk_length(); | |
970 | int64_t blob_expected_for_release = | |
11fdf7f2 | 971 | round_up_to(len_on_disk, min_alloc_size) / min_alloc_size; |
7c673cae FG |
972 | |
973 | dout(30) << __func__ << " " << *(b_it->first) | |
974 | << " expected4release=" << blob_expected_for_release | |
975 | << " expected_allocations=" << bi.expected_allocations | |
976 | << dendl; | |
977 | int64_t benefit = blob_expected_for_release - bi.expected_allocations; | |
11fdf7f2 | 978 | if (benefit >= g_conf()->bluestore_gc_enable_blob_threshold) { |
7c673cae FG |
979 | if (bi.collect_candidate) { |
980 | auto it = bi.first_lextent; | |
981 | bool bExit = false; | |
982 | do { | |
983 | if (it->blob.get() == b) { | |
eafe8130 | 984 | extents_to_collect.insert(it->logical_offset, it->length); |
7c673cae FG |
985 | } |
986 | bExit = it == bi.last_lextent; | |
987 | ++it; | |
31f18b77 | 988 | } while (!bExit); |
7c673cae FG |
989 | } |
990 | expected_for_release += blob_expected_for_release; | |
991 | expected_allocations += bi.expected_allocations; | |
992 | } | |
993 | } | |
994 | } | |
995 | } | |
996 | ||
997 | int64_t BlueStore::GarbageCollector::estimate( | |
998 | uint64_t start_offset, | |
999 | uint64_t length, | |
1000 | const BlueStore::ExtentMap& extent_map, | |
1001 | const BlueStore::old_extent_map_t& old_extents, | |
1002 | uint64_t min_alloc_size) | |
1003 | { | |
1004 | ||
1005 | affected_blobs.clear(); | |
1006 | extents_to_collect.clear(); | |
1007 | used_alloc_unit = boost::optional<uint64_t >(); | |
1008 | blob_info_counted = nullptr; | |
1009 | ||
eafe8130 TL |
1010 | uint64_t gc_start_offset = start_offset; |
1011 | uint64_t gc_end_offset = start_offset + length; | |
7c673cae FG |
1012 | |
1013 | uint64_t end_offset = start_offset + length; | |
1014 | ||
1015 | for (auto it = old_extents.begin(); it != old_extents.end(); ++it) { | |
1016 | Blob* b = it->e.blob.get(); | |
1017 | if (b->get_blob().is_compressed()) { | |
1018 | ||
1019 | // update gc_start_offset/gc_end_offset if needed | |
1020 | gc_start_offset = min(gc_start_offset, (uint64_t)it->e.blob_start()); | |
11fdf7f2 | 1021 | gc_end_offset = std::max(gc_end_offset, (uint64_t)it->e.blob_end()); |
7c673cae FG |
1022 | |
1023 | auto o = it->e.logical_offset; | |
1024 | auto l = it->e.length; | |
1025 | ||
1026 | uint64_t ref_bytes = b->get_referenced_bytes(); | |
1027 | // micro optimization to bypass blobs that have no more references | |
1028 | if (ref_bytes != 0) { | |
1029 | dout(30) << __func__ << " affected_blob:" << *b | |
1030 | << " unref 0x" << std::hex << o << "~" << l | |
1031 | << std::dec << dendl; | |
1032 | affected_blobs.emplace(b, BlobInfo(ref_bytes)); | |
1033 | } | |
1034 | } | |
1035 | } | |
1036 | dout(30) << __func__ << " gc range(hex): [" << std::hex | |
1037 | << gc_start_offset << ", " << gc_end_offset | |
1038 | << ")" << std::dec << dendl; | |
1039 | ||
1040 | // enumerate preceeding extents to check if they reference affected blobs | |
1041 | if (gc_start_offset < start_offset || gc_end_offset > end_offset) { | |
1042 | process_protrusive_extents(extent_map, | |
1043 | gc_start_offset, | |
1044 | gc_end_offset, | |
1045 | start_offset, | |
1046 | end_offset, | |
1047 | min_alloc_size); | |
1048 | } | |
1049 | return expected_for_release - expected_allocations; | |
1050 | } | |
1051 | ||
9f95a23c TL |
1052 | // LruOnodeCacheShard |
1053 | struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard { | |
1054 | typedef boost::intrusive::list< | |
1055 | BlueStore::Onode, | |
1056 | boost::intrusive::member_hook< | |
1057 | BlueStore::Onode, | |
1058 | boost::intrusive::list_member_hook<>, | |
1059 | &BlueStore::Onode::lru_item> > list_t; | |
7c673cae | 1060 | |
9f95a23c | 1061 | list_t lru; |
7c673cae | 1062 | |
9f95a23c | 1063 | explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {} |
7c673cae | 1064 | |
f6b5b4d7 | 1065 | void _add(BlueStore::Onode* o, int level) override |
9f95a23c | 1066 | { |
f6b5b4d7 | 1067 | if (o->put_cache()) { |
9f95a23c | 1068 | (level > 0) ? lru.push_front(*o) : lru.push_back(*o); |
f6b5b4d7 TL |
1069 | } else { |
1070 | ++num_pinned; | |
9f95a23c | 1071 | } |
f6b5b4d7 TL |
1072 | ++num; // we count both pinned and unpinned entries |
1073 | dout(20) << __func__ << " " << this << " " << o->oid << " added, num=" << num << dendl; | |
eafe8130 | 1074 | } |
f6b5b4d7 | 1075 | void _rm(BlueStore::Onode* o) override |
9f95a23c | 1076 | { |
f6b5b4d7 | 1077 | if (o->pop_cache()) { |
9f95a23c | 1078 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
1079 | } else { |
1080 | ceph_assert(num_pinned); | |
1081 | --num_pinned; | |
9f95a23c | 1082 | } |
f6b5b4d7 TL |
1083 | ceph_assert(num); |
1084 | --num; | |
1085 | dout(20) << __func__ << " " << this << " " << " " << o->oid << " removed, num=" << num << dendl; | |
9f95a23c | 1086 | } |
f6b5b4d7 | 1087 | void _pin(BlueStore::Onode* o) override |
9f95a23c | 1088 | { |
9f95a23c | 1089 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
1090 | ++num_pinned; |
1091 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " pinned" << dendl; | |
9f95a23c | 1092 | } |
f6b5b4d7 | 1093 | void _unpin(BlueStore::Onode* o) override |
9f95a23c | 1094 | { |
f6b5b4d7 TL |
1095 | lru.push_front(*o); |
1096 | ceph_assert(num_pinned); | |
1097 | --num_pinned; | |
1098 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " unpinned" << dendl; | |
9f95a23c | 1099 | } |
f6b5b4d7 | 1100 | |
9f95a23c TL |
1101 | void _trim_to(uint64_t new_size) override |
1102 | { | |
1103 | if (new_size >= lru.size()) { | |
1104 | return; // don't even try | |
1105 | } | |
1106 | uint64_t n = lru.size() - new_size; | |
1107 | auto p = lru.end(); | |
1108 | ceph_assert(p != lru.begin()); | |
1109 | --p; | |
f6b5b4d7 TL |
1110 | ceph_assert(num >= n); |
1111 | num -= n; | |
1112 | while (n-- > 0) { | |
9f95a23c | 1113 | BlueStore::Onode *o = &*p; |
f6b5b4d7 TL |
1114 | dout(20) << __func__ << " rm " << o->oid << " " |
1115 | << o->nref << " " << o->cached << " " << o->pinned << dendl; | |
9f95a23c TL |
1116 | if (p != lru.begin()) { |
1117 | lru.erase(p--); | |
1118 | } else { | |
f6b5b4d7 | 1119 | ceph_assert(n == 0); |
9f95a23c | 1120 | lru.erase(p); |
9f95a23c | 1121 | } |
f6b5b4d7 TL |
1122 | auto pinned = !o->pop_cache(); |
1123 | ceph_assert(!pinned); | |
1124 | o->c->onode_map._remove(o->oid); | |
9f95a23c | 1125 | } |
f6b5b4d7 TL |
1126 | } |
1127 | void move_pinned(OnodeCacheShard *to, BlueStore::Onode *o) override | |
1128 | { | |
1129 | if (to == this) { | |
1130 | return; | |
1131 | } | |
1132 | ceph_assert(o->cached); | |
1133 | ceph_assert(o->pinned); | |
1134 | ceph_assert(num); | |
1135 | ceph_assert(num_pinned); | |
1136 | --num_pinned; | |
1137 | --num; | |
1138 | ++to->num_pinned; | |
1139 | ++to->num; | |
9f95a23c TL |
1140 | } |
1141 | void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override | |
1142 | { | |
f6b5b4d7 | 1143 | *onodes += num; |
9f95a23c TL |
1144 | *pinned_onodes += num_pinned; |
1145 | } | |
1146 | }; | |
7c673cae | 1147 | |
9f95a23c TL |
1148 | // OnodeCacheShard |
1149 | BlueStore::OnodeCacheShard *BlueStore::OnodeCacheShard::create( | |
1150 | CephContext* cct, | |
1151 | string type, | |
1152 | PerfCounters *logger) | |
7c673cae | 1153 | { |
9f95a23c TL |
1154 | BlueStore::OnodeCacheShard *c = nullptr; |
1155 | // Currently we only implement an LRU cache for onodes | |
1156 | c = new LruOnodeCacheShard(cct); | |
1157 | c->logger = logger; | |
1158 | return c; | |
7c673cae FG |
1159 | } |
1160 | ||
9f95a23c TL |
1161 | // LruBufferCacheShard |
1162 | struct LruBufferCacheShard : public BlueStore::BufferCacheShard { | |
1163 | typedef boost::intrusive::list< | |
1164 | BlueStore::Buffer, | |
1165 | boost::intrusive::member_hook< | |
1166 | BlueStore::Buffer, | |
1167 | boost::intrusive::list_member_hook<>, | |
1168 | &BlueStore::Buffer::lru_item> > list_t; | |
1169 | list_t lru; | |
1170 | ||
1171 | explicit LruBufferCacheShard(CephContext *cct) : BlueStore::BufferCacheShard(cct) {} | |
1172 | ||
1173 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override { | |
1174 | if (near) { | |
1175 | auto q = lru.iterator_to(*near); | |
1176 | lru.insert(q, *b); | |
1177 | } else if (level > 0) { | |
1178 | lru.push_front(*b); | |
1179 | } else { | |
1180 | lru.push_back(*b); | |
7c673cae | 1181 | } |
9f95a23c TL |
1182 | buffer_bytes += b->length; |
1183 | num = lru.size(); | |
1184 | } | |
1185 | void _rm(BlueStore::Buffer *b) override { | |
1186 | ceph_assert(buffer_bytes >= b->length); | |
1187 | buffer_bytes -= b->length; | |
1188 | auto q = lru.iterator_to(*b); | |
1189 | lru.erase(q); | |
1190 | num = lru.size(); | |
1191 | } | |
1192 | void _move(BlueStore::BufferCacheShard *src, BlueStore::Buffer *b) override { | |
1193 | src->_rm(b); | |
1194 | _add(b, 0, nullptr); | |
1195 | } | |
1196 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override { | |
1197 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1198 | buffer_bytes += delta; | |
1199 | } | |
1200 | void _touch(BlueStore::Buffer *b) override { | |
1201 | auto p = lru.iterator_to(*b); | |
1202 | lru.erase(p); | |
1203 | lru.push_front(*b); | |
1204 | num = lru.size(); | |
1205 | _audit("_touch_buffer end"); | |
1206 | } | |
7c673cae | 1207 | |
9f95a23c TL |
1208 | void _trim_to(uint64_t max) override |
1209 | { | |
1210 | while (buffer_bytes > max) { | |
1211 | auto i = lru.rbegin(); | |
1212 | if (i == lru.rend()) { | |
1213 | // stop if lru is now empty | |
7c673cae FG |
1214 | break; |
1215 | } | |
1216 | ||
9f95a23c TL |
1217 | BlueStore::Buffer *b = &*i; |
1218 | ceph_assert(b->is_clean()); | |
1219 | dout(20) << __func__ << " rm " << *b << dendl; | |
1220 | b->space->_rm_buffer(this, b); | |
7c673cae | 1221 | } |
9f95a23c | 1222 | num = lru.size(); |
7c673cae | 1223 | } |
7c673cae | 1224 | |
9f95a23c TL |
1225 | void add_stats(uint64_t *extents, |
1226 | uint64_t *blobs, | |
1227 | uint64_t *buffers, | |
1228 | uint64_t *bytes) override { | |
1229 | *extents += num_extents; | |
1230 | *blobs += num_blobs; | |
1231 | *buffers += num; | |
1232 | *bytes += buffer_bytes; | |
7c673cae | 1233 | } |
9f95a23c TL |
1234 | #ifdef DEBUG_CACHE |
1235 | void _audit(const char *s) override | |
1236 | { | |
1237 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1238 | uint64_t s = 0; | |
1239 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1240 | s += i->length; | |
1241 | } | |
1242 | if (s != buffer_bytes) { | |
1243 | derr << __func__ << " buffer_size " << buffer_bytes << " actual " << s | |
1244 | << dendl; | |
1245 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1246 | derr << __func__ << " " << *i << dendl; | |
1247 | } | |
1248 | ceph_assert(s == buffer_bytes); | |
7c673cae | 1249 | } |
9f95a23c TL |
1250 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1251 | << " ok" << dendl; | |
7c673cae | 1252 | } |
7c673cae | 1253 | #endif |
9f95a23c | 1254 | }; |
7c673cae | 1255 | |
9f95a23c TL |
1256 | // TwoQBufferCacheShard |
1257 | ||
1258 | struct TwoQBufferCacheShard : public BlueStore::BufferCacheShard { | |
1259 | typedef boost::intrusive::list< | |
1260 | BlueStore::Buffer, | |
1261 | boost::intrusive::member_hook< | |
1262 | BlueStore::Buffer, | |
1263 | boost::intrusive::list_member_hook<>, | |
1264 | &BlueStore::Buffer::lru_item> > list_t; | |
1265 | list_t hot; ///< "Am" hot buffers | |
1266 | list_t warm_in; ///< "A1in" newly warm buffers | |
1267 | list_t warm_out; ///< "A1out" empty buffers we've evicted | |
1268 | uint64_t buffer_bytes = 0; ///< bytes | |
1269 | ||
1270 | enum { | |
1271 | BUFFER_NEW = 0, | |
1272 | BUFFER_WARM_IN, ///< in warm_in | |
1273 | BUFFER_WARM_OUT, ///< in warm_out | |
1274 | BUFFER_HOT, ///< in hot | |
1275 | BUFFER_TYPE_MAX | |
1276 | }; | |
7c673cae | 1277 | |
9f95a23c | 1278 | uint64_t list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type |
7c673cae | 1279 | |
9f95a23c TL |
1280 | public: |
1281 | explicit TwoQBufferCacheShard(CephContext *cct) : BufferCacheShard(cct) {} | |
7c673cae | 1282 | |
9f95a23c TL |
1283 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override |
1284 | { | |
1285 | dout(20) << __func__ << " level " << level << " near " << near | |
1286 | << " on " << *b | |
1287 | << " which has cache_private " << b->cache_private << dendl; | |
1288 | if (near) { | |
1289 | b->cache_private = near->cache_private; | |
1290 | switch (b->cache_private) { | |
1291 | case BUFFER_WARM_IN: | |
1292 | warm_in.insert(warm_in.iterator_to(*near), *b); | |
1293 | break; | |
1294 | case BUFFER_WARM_OUT: | |
1295 | ceph_assert(b->is_empty()); | |
1296 | warm_out.insert(warm_out.iterator_to(*near), *b); | |
1297 | break; | |
1298 | case BUFFER_HOT: | |
1299 | hot.insert(hot.iterator_to(*near), *b); | |
1300 | break; | |
1301 | default: | |
1302 | ceph_abort_msg("bad cache_private"); | |
1303 | } | |
1304 | } else if (b->cache_private == BUFFER_NEW) { | |
1305 | b->cache_private = BUFFER_WARM_IN; | |
1306 | if (level > 0) { | |
1307 | warm_in.push_front(*b); | |
1308 | } else { | |
1309 | // take caller hint to start at the back of the warm queue | |
1310 | warm_in.push_back(*b); | |
1311 | } | |
1312 | } else { | |
1313 | // we got a hint from discard | |
1314 | switch (b->cache_private) { | |
1315 | case BUFFER_WARM_IN: | |
1316 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1317 | // do this. | |
1318 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1319 | warm_in.push_front(*b); | |
1320 | break; | |
1321 | case BUFFER_WARM_OUT: | |
1322 | b->cache_private = BUFFER_HOT; | |
1323 | // move to hot. fall-thru | |
1324 | case BUFFER_HOT: | |
1325 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1326 | hot.push_front(*b); | |
1327 | break; | |
1328 | default: | |
1329 | ceph_abort_msg("bad cache_private"); | |
1330 | } | |
1331 | } | |
1332 | if (!b->is_empty()) { | |
1333 | buffer_bytes += b->length; | |
1334 | list_bytes[b->cache_private] += b->length; | |
1335 | } | |
1336 | num = hot.size() + warm_in.size(); | |
1337 | } | |
1338 | ||
1339 | void _rm(BlueStore::Buffer *b) override | |
1340 | { | |
1341 | dout(20) << __func__ << " " << *b << dendl; | |
1342 | if (!b->is_empty()) { | |
1343 | ceph_assert(buffer_bytes >= b->length); | |
1344 | buffer_bytes -= b->length; | |
1345 | ceph_assert(list_bytes[b->cache_private] >= b->length); | |
1346 | list_bytes[b->cache_private] -= b->length; | |
1347 | } | |
7c673cae FG |
1348 | switch (b->cache_private) { |
1349 | case BUFFER_WARM_IN: | |
9f95a23c | 1350 | warm_in.erase(warm_in.iterator_to(*b)); |
7c673cae FG |
1351 | break; |
1352 | case BUFFER_WARM_OUT: | |
9f95a23c | 1353 | warm_out.erase(warm_out.iterator_to(*b)); |
7c673cae FG |
1354 | break; |
1355 | case BUFFER_HOT: | |
9f95a23c | 1356 | hot.erase(hot.iterator_to(*b)); |
7c673cae FG |
1357 | break; |
1358 | default: | |
11fdf7f2 | 1359 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1360 | } |
9f95a23c TL |
1361 | num = hot.size() + warm_in.size(); |
1362 | } | |
1363 | ||
1364 | void _move(BlueStore::BufferCacheShard *srcc, BlueStore::Buffer *b) override | |
1365 | { | |
1366 | TwoQBufferCacheShard *src = static_cast<TwoQBufferCacheShard*>(srcc); | |
1367 | src->_rm(b); | |
1368 | ||
1369 | // preserve which list we're on (even if we can't preserve the order!) | |
7c673cae FG |
1370 | switch (b->cache_private) { |
1371 | case BUFFER_WARM_IN: | |
9f95a23c TL |
1372 | ceph_assert(!b->is_empty()); |
1373 | warm_in.push_back(*b); | |
7c673cae FG |
1374 | break; |
1375 | case BUFFER_WARM_OUT: | |
9f95a23c TL |
1376 | ceph_assert(b->is_empty()); |
1377 | warm_out.push_back(*b); | |
1378 | break; | |
7c673cae | 1379 | case BUFFER_HOT: |
9f95a23c TL |
1380 | ceph_assert(!b->is_empty()); |
1381 | hot.push_back(*b); | |
7c673cae FG |
1382 | break; |
1383 | default: | |
11fdf7f2 | 1384 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1385 | } |
9f95a23c TL |
1386 | if (!b->is_empty()) { |
1387 | buffer_bytes += b->length; | |
1388 | list_bytes[b->cache_private] += b->length; | |
1389 | } | |
1390 | num = hot.size() + warm_in.size(); | |
7c673cae | 1391 | } |
7c673cae | 1392 | |
9f95a23c TL |
1393 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override |
1394 | { | |
1395 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1396 | if (!b->is_empty()) { | |
1397 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1398 | buffer_bytes += delta; | |
1399 | ceph_assert((int64_t)list_bytes[b->cache_private] + delta >= 0); | |
1400 | list_bytes[b->cache_private] += delta; | |
1401 | } | |
7c673cae | 1402 | } |
7c673cae | 1403 | |
9f95a23c TL |
1404 | void _touch(BlueStore::Buffer *b) override { |
1405 | switch (b->cache_private) { | |
1406 | case BUFFER_WARM_IN: | |
1407 | // do nothing (somewhat counter-intuitively!) | |
1408 | break; | |
1409 | case BUFFER_WARM_OUT: | |
1410 | // move from warm_out to hot LRU | |
1411 | ceph_abort_msg("this happens via discard hint"); | |
1412 | break; | |
1413 | case BUFFER_HOT: | |
1414 | // move to front of hot LRU | |
1415 | hot.erase(hot.iterator_to(*b)); | |
1416 | hot.push_front(*b); | |
1417 | break; | |
1418 | } | |
1419 | num = hot.size() + warm_in.size(); | |
1420 | _audit("_touch_buffer end"); | |
7c673cae | 1421 | } |
7c673cae | 1422 | |
9f95a23c TL |
1423 | void _trim_to(uint64_t max) override |
1424 | { | |
1425 | if (buffer_bytes > max) { | |
1426 | uint64_t kin = max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1427 | uint64_t khot = max - kin; | |
1428 | ||
1429 | // pre-calculate kout based on average buffer size too, | |
1430 | // which is typical(the warm_in and hot lists may change later) | |
1431 | uint64_t kout = 0; | |
1432 | uint64_t buffer_num = hot.size() + warm_in.size(); | |
1433 | if (buffer_num) { | |
1434 | uint64_t avg_size = buffer_bytes / buffer_num; | |
1435 | ceph_assert(avg_size); | |
1436 | uint64_t calculated_num = max / avg_size; | |
1437 | kout = calculated_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1438 | } | |
1439 | ||
1440 | if (list_bytes[BUFFER_HOT] < khot) { | |
1441 | // hot is small, give slack to warm_in | |
1442 | kin += khot - list_bytes[BUFFER_HOT]; | |
1443 | } else if (list_bytes[BUFFER_WARM_IN] < kin) { | |
1444 | // warm_in is small, give slack to hot | |
1445 | khot += kin - list_bytes[BUFFER_WARM_IN]; | |
1446 | } | |
1447 | ||
1448 | // adjust warm_in list | |
1449 | int64_t to_evict_bytes = list_bytes[BUFFER_WARM_IN] - kin; | |
1450 | uint64_t evicted = 0; | |
1451 | ||
1452 | while (to_evict_bytes > 0) { | |
1453 | auto p = warm_in.rbegin(); | |
1454 | if (p == warm_in.rend()) { | |
1455 | // stop if warm_in list is now empty | |
1456 | break; | |
1457 | } | |
7c673cae | 1458 | |
9f95a23c TL |
1459 | BlueStore::Buffer *b = &*p; |
1460 | ceph_assert(b->is_clean()); | |
1461 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; | |
1462 | ceph_assert(buffer_bytes >= b->length); | |
1463 | buffer_bytes -= b->length; | |
1464 | ceph_assert(list_bytes[BUFFER_WARM_IN] >= b->length); | |
1465 | list_bytes[BUFFER_WARM_IN] -= b->length; | |
1466 | to_evict_bytes -= b->length; | |
1467 | evicted += b->length; | |
1468 | b->state = BlueStore::Buffer::STATE_EMPTY; | |
1469 | b->data.clear(); | |
1470 | warm_in.erase(warm_in.iterator_to(*b)); | |
1471 | warm_out.push_front(*b); | |
1472 | b->cache_private = BUFFER_WARM_OUT; | |
1473 | } | |
1474 | ||
1475 | if (evicted > 0) { | |
1476 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1477 | << " from warm_in list, done evicting warm_in buffers" | |
1478 | << dendl; | |
1479 | } | |
7c673cae | 1480 | |
9f95a23c TL |
1481 | // adjust hot list |
1482 | to_evict_bytes = list_bytes[BUFFER_HOT] - khot; | |
1483 | evicted = 0; | |
7c673cae | 1484 | |
9f95a23c TL |
1485 | while (to_evict_bytes > 0) { |
1486 | auto p = hot.rbegin(); | |
1487 | if (p == hot.rend()) { | |
1488 | // stop if hot list is now empty | |
1489 | break; | |
1490 | } | |
7c673cae | 1491 | |
9f95a23c TL |
1492 | BlueStore::Buffer *b = &*p; |
1493 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
1494 | ceph_assert(b->is_clean()); | |
1495 | // adjust evict size before buffer goes invalid | |
1496 | to_evict_bytes -= b->length; | |
1497 | evicted += b->length; | |
1498 | b->space->_rm_buffer(this, b); | |
1499 | } | |
7c673cae | 1500 | |
9f95a23c TL |
1501 | if (evicted > 0) { |
1502 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1503 | << " from hot list, done evicting hot buffers" | |
1504 | << dendl; | |
7c673cae FG |
1505 | } |
1506 | ||
9f95a23c TL |
1507 | // adjust warm out list too, if necessary |
1508 | int64_t n = warm_out.size() - kout; | |
1509 | while (n-- > 0) { | |
1510 | BlueStore::Buffer *b = &*warm_out.rbegin(); | |
1511 | ceph_assert(b->is_empty()); | |
1512 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; | |
1513 | b->space->_rm_buffer(this, b); | |
1514 | } | |
7c673cae | 1515 | } |
9f95a23c TL |
1516 | num = hot.size() + warm_in.size(); |
1517 | } | |
7c673cae | 1518 | |
9f95a23c TL |
1519 | void add_stats(uint64_t *extents, |
1520 | uint64_t *blobs, | |
1521 | uint64_t *buffers, | |
1522 | uint64_t *bytes) override { | |
1523 | *extents += num_extents; | |
1524 | *blobs += num_blobs; | |
1525 | *buffers += num; | |
1526 | *bytes += buffer_bytes; | |
1527 | } | |
7c673cae | 1528 | |
9f95a23c TL |
1529 | #ifdef DEBUG_CACHE |
1530 | void _audit(const char *s) override | |
1531 | { | |
1532 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1533 | uint64_t s = 0; | |
1534 | for (auto i = hot.begin(); i != hot.end(); ++i) { | |
1535 | s += i->length; | |
7c673cae FG |
1536 | } |
1537 | ||
9f95a23c TL |
1538 | uint64_t hot_bytes = s; |
1539 | if (hot_bytes != list_bytes[BUFFER_HOT]) { | |
1540 | derr << __func__ << " hot_list_bytes " | |
1541 | << list_bytes[BUFFER_HOT] | |
1542 | << " != actual " << hot_bytes | |
1543 | << dendl; | |
1544 | ceph_assert(hot_bytes == list_bytes[BUFFER_HOT]); | |
7c673cae FG |
1545 | } |
1546 | ||
9f95a23c TL |
1547 | for (auto i = warm_in.begin(); i != warm_in.end(); ++i) { |
1548 | s += i->length; | |
7c673cae | 1549 | } |
7c673cae | 1550 | |
9f95a23c TL |
1551 | uint64_t warm_in_bytes = s - hot_bytes; |
1552 | if (warm_in_bytes != list_bytes[BUFFER_WARM_IN]) { | |
1553 | derr << __func__ << " warm_in_list_bytes " | |
1554 | << list_bytes[BUFFER_WARM_IN] | |
1555 | << " != actual " << warm_in_bytes | |
1556 | << dendl; | |
1557 | ceph_assert(warm_in_bytes == list_bytes[BUFFER_WARM_IN]); | |
7c673cae | 1558 | } |
7c673cae | 1559 | |
9f95a23c TL |
1560 | if (s != buffer_bytes) { |
1561 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1562 | << dendl; | |
1563 | ceph_assert(s == buffer_bytes); | |
1564 | } | |
7c673cae | 1565 | |
9f95a23c TL |
1566 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1567 | << " ok" << dendl; | |
7c673cae | 1568 | } |
9f95a23c TL |
1569 | #endif |
1570 | }; | |
7c673cae | 1571 | |
9f95a23c | 1572 | // BuferCacheShard |
7c673cae | 1573 | |
9f95a23c TL |
1574 | BlueStore::BufferCacheShard *BlueStore::BufferCacheShard::create( |
1575 | CephContext* cct, | |
1576 | string type, | |
1577 | PerfCounters *logger) | |
1578 | { | |
1579 | BufferCacheShard *c = nullptr; | |
1580 | if (type == "lru") | |
1581 | c = new LruBufferCacheShard(cct); | |
1582 | else if (type == "2q") | |
1583 | c = new TwoQBufferCacheShard(cct); | |
1584 | else | |
1585 | ceph_abort_msg("unrecognized cache type"); | |
1586 | c->logger = logger; | |
1587 | return c; | |
7c673cae | 1588 | } |
7c673cae FG |
1589 | |
1590 | // BufferSpace | |
1591 | ||
1592 | #undef dout_prefix | |
1593 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1594 | ||
9f95a23c | 1595 | void BlueStore::BufferSpace::_clear(BufferCacheShard* cache) |
7c673cae FG |
1596 | { |
1597 | // note: we already hold cache->lock | |
1598 | ldout(cache->cct, 20) << __func__ << dendl; | |
1599 | while (!buffer_map.empty()) { | |
1600 | _rm_buffer(cache, buffer_map.begin()); | |
1601 | } | |
1602 | } | |
1603 | ||
9f95a23c | 1604 | int BlueStore::BufferSpace::_discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) |
7c673cae FG |
1605 | { |
1606 | // note: we already hold cache->lock | |
1607 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1608 | << std::dec << dendl; | |
1609 | int cache_private = 0; | |
1610 | cache->_audit("discard start"); | |
1611 | auto i = _data_lower_bound(offset); | |
1612 | uint32_t end = offset + length; | |
1613 | while (i != buffer_map.end()) { | |
1614 | Buffer *b = i->second.get(); | |
1615 | if (b->offset >= end) { | |
1616 | break; | |
1617 | } | |
1618 | if (b->cache_private > cache_private) { | |
1619 | cache_private = b->cache_private; | |
1620 | } | |
1621 | if (b->offset < offset) { | |
1622 | int64_t front = offset - b->offset; | |
1623 | if (b->end() > end) { | |
1624 | // drop middle (split) | |
1625 | uint32_t tail = b->end() - end; | |
1626 | if (b->data.length()) { | |
1627 | bufferlist bl; | |
1628 | bl.substr_of(b->data, b->length - tail, tail); | |
31f18b77 FG |
1629 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1630 | nb->maybe_rebuild(); | |
1631 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1632 | } else { |
31f18b77 FG |
1633 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail), |
1634 | 0, b); | |
7c673cae FG |
1635 | } |
1636 | if (!b->is_writing()) { | |
9f95a23c | 1637 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1638 | } |
1639 | b->truncate(front); | |
31f18b77 | 1640 | b->maybe_rebuild(); |
7c673cae FG |
1641 | cache->_audit("discard end 1"); |
1642 | break; | |
1643 | } else { | |
1644 | // drop tail | |
1645 | if (!b->is_writing()) { | |
9f95a23c | 1646 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1647 | } |
1648 | b->truncate(front); | |
31f18b77 | 1649 | b->maybe_rebuild(); |
7c673cae FG |
1650 | ++i; |
1651 | continue; | |
1652 | } | |
1653 | } | |
1654 | if (b->end() <= end) { | |
1655 | // drop entire buffer | |
1656 | _rm_buffer(cache, i++); | |
1657 | continue; | |
1658 | } | |
1659 | // drop front | |
1660 | uint32_t keep = b->end() - end; | |
1661 | if (b->data.length()) { | |
1662 | bufferlist bl; | |
1663 | bl.substr_of(b->data, b->length - keep, keep); | |
31f18b77 FG |
1664 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1665 | nb->maybe_rebuild(); | |
1666 | _add_buffer(cache, nb, 0, b); | |
7c673cae FG |
1667 | } else { |
1668 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep), 0, b); | |
1669 | } | |
1670 | _rm_buffer(cache, i); | |
1671 | cache->_audit("discard end 2"); | |
1672 | break; | |
1673 | } | |
1674 | return cache_private; | |
1675 | } | |
1676 | ||
1677 | void BlueStore::BufferSpace::read( | |
9f95a23c | 1678 | BufferCacheShard* cache, |
224ce89b WB |
1679 | uint32_t offset, |
1680 | uint32_t length, | |
7c673cae | 1681 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1682 | interval_set<uint32_t>& res_intervals, |
1683 | int flags) | |
7c673cae | 1684 | { |
7c673cae FG |
1685 | res.clear(); |
1686 | res_intervals.clear(); | |
1687 | uint32_t want_bytes = length; | |
1688 | uint32_t end = offset + length; | |
224ce89b WB |
1689 | |
1690 | { | |
11fdf7f2 | 1691 | std::lock_guard l(cache->lock); |
224ce89b WB |
1692 | for (auto i = _data_lower_bound(offset); |
1693 | i != buffer_map.end() && offset < end && i->first < end; | |
1694 | ++i) { | |
1695 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1696 | ceph_assert(b->end() > offset); |
91327a77 AA |
1697 | |
1698 | bool val = false; | |
1699 | if (flags & BYPASS_CLEAN_CACHE) | |
1700 | val = b->is_writing(); | |
1701 | else | |
1702 | val = b->is_writing() || b->is_clean(); | |
1703 | if (val) { | |
224ce89b WB |
1704 | if (b->offset < offset) { |
1705 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1706 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1707 | res[offset].substr_of(b->data, skip, l); |
1708 | res_intervals.insert(offset, l); | |
1709 | offset += l; | |
1710 | length -= l; | |
1711 | if (!b->is_writing()) { | |
9f95a23c | 1712 | cache->_touch(b); |
224ce89b WB |
1713 | } |
1714 | continue; | |
1715 | } | |
1716 | if (b->offset > offset) { | |
1717 | uint32_t gap = b->offset - offset; | |
1718 | if (length <= gap) { | |
1719 | break; | |
1720 | } | |
1721 | offset += gap; | |
1722 | length -= gap; | |
1723 | } | |
1724 | if (!b->is_writing()) { | |
9f95a23c | 1725 | cache->_touch(b); |
224ce89b WB |
1726 | } |
1727 | if (b->length > length) { | |
1728 | res[offset].substr_of(b->data, 0, length); | |
1729 | res_intervals.insert(offset, length); | |
7c673cae | 1730 | break; |
224ce89b WB |
1731 | } else { |
1732 | res[offset].append(b->data); | |
1733 | res_intervals.insert(offset, b->length); | |
1734 | if (b->length == length) | |
1735 | break; | |
1736 | offset += b->length; | |
1737 | length -= b->length; | |
1738 | } | |
7c673cae FG |
1739 | } |
1740 | } | |
1741 | } | |
1742 | ||
1743 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1744 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1745 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1746 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1747 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1748 | } | |
1749 | ||
9f95a23c | 1750 | void BlueStore::BufferSpace::_finish_write(BufferCacheShard* cache, uint64_t seq) |
7c673cae | 1751 | { |
7c673cae FG |
1752 | auto i = writing.begin(); |
1753 | while (i != writing.end()) { | |
1754 | if (i->seq > seq) { | |
1755 | break; | |
1756 | } | |
1757 | if (i->seq < seq) { | |
1758 | ++i; | |
1759 | continue; | |
1760 | } | |
1761 | ||
1762 | Buffer *b = &*i; | |
11fdf7f2 | 1763 | ceph_assert(b->is_writing()); |
7c673cae FG |
1764 | |
1765 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1766 | writing.erase(i++); | |
1767 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1768 | buffer_map.erase(b->offset); | |
1769 | } else { | |
1770 | b->state = Buffer::STATE_CLEAN; | |
1771 | writing.erase(i++); | |
31f18b77 FG |
1772 | b->maybe_rebuild(); |
1773 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
9f95a23c | 1774 | cache->_add(b, 1, nullptr); |
7c673cae FG |
1775 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; |
1776 | } | |
1777 | } | |
9f95a23c | 1778 | cache->_trim(); |
7c673cae FG |
1779 | cache->_audit("finish_write end"); |
1780 | } | |
1781 | ||
9f95a23c | 1782 | void BlueStore::BufferSpace::split(BufferCacheShard* cache, size_t pos, BlueStore::BufferSpace &r) |
7c673cae | 1783 | { |
11fdf7f2 | 1784 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1785 | if (buffer_map.empty()) |
1786 | return; | |
1787 | ||
1788 | auto p = --buffer_map.end(); | |
1789 | while (true) { | |
1790 | if (p->second->end() <= pos) | |
1791 | break; | |
1792 | ||
1793 | if (p->second->offset < pos) { | |
1794 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1795 | size_t left = pos - p->second->offset; | |
1796 | size_t right = p->second->length - left; | |
1797 | if (p->second->data.length()) { | |
1798 | bufferlist bl; | |
1799 | bl.substr_of(p->second->data, left, right); | |
1800 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, bl), | |
1801 | 0, p->second.get()); | |
1802 | } else { | |
1803 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, right), | |
1804 | 0, p->second.get()); | |
1805 | } | |
9f95a23c | 1806 | cache->_adjust_size(p->second.get(), -right); |
7c673cae FG |
1807 | p->second->truncate(left); |
1808 | break; | |
1809 | } | |
1810 | ||
11fdf7f2 | 1811 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1812 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1813 | if (p->second->data.length()) { | |
1814 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1815 | p->second->offset - pos, p->second->data), | |
1816 | 0, p->second.get()); | |
1817 | } else { | |
1818 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1819 | p->second->offset - pos, p->second->length), | |
1820 | 0, p->second.get()); | |
1821 | } | |
1822 | if (p == buffer_map.begin()) { | |
1823 | _rm_buffer(cache, p); | |
1824 | break; | |
1825 | } else { | |
1826 | _rm_buffer(cache, p--); | |
1827 | } | |
1828 | } | |
11fdf7f2 | 1829 | ceph_assert(writing.empty()); |
9f95a23c | 1830 | cache->_trim(); |
7c673cae FG |
1831 | } |
1832 | ||
1833 | // OnodeSpace | |
1834 | ||
1835 | #undef dout_prefix | |
1836 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1837 | ||
f6b5b4d7 TL |
1838 | BlueStore::OnodeRef BlueStore::OnodeSpace::add(const ghobject_t& oid, |
1839 | OnodeRef& o) | |
7c673cae | 1840 | { |
11fdf7f2 | 1841 | std::lock_guard l(cache->lock); |
7c673cae FG |
1842 | auto p = onode_map.find(oid); |
1843 | if (p != onode_map.end()) { | |
1844 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o | |
1845 | << " raced, returning existing " << p->second | |
1846 | << dendl; | |
1847 | return p->second; | |
1848 | } | |
f6b5b4d7 | 1849 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << o << dendl; |
7c673cae | 1850 | onode_map[oid] = o; |
f6b5b4d7 | 1851 | cache->_add(o.get(), 1); |
9f95a23c | 1852 | cache->_trim(); |
7c673cae FG |
1853 | return o; |
1854 | } | |
1855 | ||
f6b5b4d7 TL |
1856 | void BlueStore::OnodeSpace::_remove(const ghobject_t& oid) |
1857 | { | |
1858 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << dendl; | |
1859 | onode_map.erase(oid); | |
1860 | } | |
1861 | ||
7c673cae FG |
1862 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) |
1863 | { | |
7c673cae | 1864 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b WB |
1865 | OnodeRef o; |
1866 | bool hit = false; | |
1867 | ||
1868 | { | |
11fdf7f2 | 1869 | std::lock_guard l(cache->lock); |
224ce89b WB |
1870 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1871 | if (p == onode_map.end()) { | |
1872 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
1873 | } else { | |
1874 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
f6b5b4d7 TL |
1875 | << " " << p->second->nref |
1876 | << " " << p->second->cached | |
1877 | << " " << p->second->pinned | |
224ce89b | 1878 | << dendl; |
f6b5b4d7 TL |
1879 | // This will pin onode and implicitly touch the cache when Onode |
1880 | // eventually will become unpinned | |
224ce89b | 1881 | o = p->second; |
f6b5b4d7 TL |
1882 | ceph_assert(!o->cached || o->pinned); |
1883 | ||
1884 | hit = true; | |
224ce89b WB |
1885 | } |
1886 | } | |
1887 | ||
1888 | if (hit) { | |
1889 | cache->logger->inc(l_bluestore_onode_hits); | |
1890 | } else { | |
7c673cae | 1891 | cache->logger->inc(l_bluestore_onode_misses); |
7c673cae | 1892 | } |
224ce89b | 1893 | return o; |
7c673cae FG |
1894 | } |
1895 | ||
1896 | void BlueStore::OnodeSpace::clear() | |
1897 | { | |
11fdf7f2 | 1898 | std::lock_guard l(cache->lock); |
f6b5b4d7 | 1899 | ldout(cache->cct, 10) << __func__ << " " << onode_map.size()<< dendl; |
7c673cae | 1900 | for (auto &p : onode_map) { |
f6b5b4d7 | 1901 | cache->_rm(p.second.get()); |
7c673cae FG |
1902 | } |
1903 | onode_map.clear(); | |
1904 | } | |
1905 | ||
1906 | bool BlueStore::OnodeSpace::empty() | |
1907 | { | |
11fdf7f2 | 1908 | std::lock_guard l(cache->lock); |
7c673cae FG |
1909 | return onode_map.empty(); |
1910 | } | |
1911 | ||
1912 | void BlueStore::OnodeSpace::rename( | |
1913 | OnodeRef& oldo, | |
1914 | const ghobject_t& old_oid, | |
1915 | const ghobject_t& new_oid, | |
f91f0fd5 | 1916 | const mempool::bluestore_cache_meta::string& new_okey) |
7c673cae | 1917 | { |
11fdf7f2 | 1918 | std::lock_guard l(cache->lock); |
7c673cae FG |
1919 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
1920 | << dendl; | |
1921 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
1922 | po = onode_map.find(old_oid); | |
1923 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 1924 | ceph_assert(po != pn); |
7c673cae | 1925 | |
11fdf7f2 | 1926 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
1927 | if (pn != onode_map.end()) { |
1928 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
1929 | << dendl; | |
f6b5b4d7 | 1930 | cache->_rm(pn->second.get()); |
7c673cae FG |
1931 | onode_map.erase(pn); |
1932 | } | |
1933 | OnodeRef o = po->second; | |
1934 | ||
1935 | // install a non-existent onode at old location | |
1936 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
1937 | po->second = oldo; | |
f6b5b4d7 TL |
1938 | cache->_add(oldo.get(), 1); |
1939 | // add at new position and fix oid, key. | |
1940 | // This will pin 'o' and implicitly touch cache | |
1941 | // when it will eventually become unpinned | |
7c673cae | 1942 | onode_map.insert(make_pair(new_oid, o)); |
f6b5b4d7 TL |
1943 | ceph_assert(o->pinned); |
1944 | ||
7c673cae FG |
1945 | o->oid = new_oid; |
1946 | o->key = new_okey; | |
9f95a23c | 1947 | cache->_trim(); |
7c673cae FG |
1948 | } |
1949 | ||
1950 | bool BlueStore::OnodeSpace::map_any(std::function<bool(OnodeRef)> f) | |
1951 | { | |
11fdf7f2 | 1952 | std::lock_guard l(cache->lock); |
7c673cae FG |
1953 | ldout(cache->cct, 20) << __func__ << dendl; |
1954 | for (auto& i : onode_map) { | |
1955 | if (f(i.second)) { | |
1956 | return true; | |
1957 | } | |
1958 | } | |
1959 | return false; | |
1960 | } | |
1961 | ||
11fdf7f2 TL |
1962 | template <int LogLevelV = 30> |
1963 | void BlueStore::OnodeSpace::dump(CephContext *cct) | |
3efd9988 FG |
1964 | { |
1965 | for (auto& i : onode_map) { | |
f6b5b4d7 TL |
1966 | ldout(cct, LogLevelV) << i.first << " : " << i.second |
1967 | << " " << i.second->nref | |
1968 | << " " << i.second->cached | |
1969 | << " " << i.second->pinned | |
1970 | << dendl; | |
3efd9988 FG |
1971 | } |
1972 | } | |
7c673cae FG |
1973 | |
1974 | // SharedBlob | |
1975 | ||
1976 | #undef dout_prefix | |
1977 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
9f95a23c TL |
1978 | #undef dout_context |
1979 | #define dout_context coll->store->cct | |
7c673cae | 1980 | |
9f95a23c | 1981 | void BlueStore::SharedBlob::dump(Formatter* f) const |
7c673cae | 1982 | { |
9f95a23c TL |
1983 | f->dump_bool("loaded", loaded); |
1984 | if (loaded) { | |
1985 | persistent->dump(f); | |
1986 | } else { | |
1987 | f->dump_unsigned("sbid_unloaded", sbid_unloaded); | |
1988 | } | |
1989 | } | |
1990 | ||
1991 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
1992 | { | |
1993 | out << "SharedBlob(" << &sb; | |
1994 | ||
7c673cae FG |
1995 | if (sb.loaded) { |
1996 | out << " loaded " << *sb.persistent; | |
1997 | } else { | |
1998 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
1999 | } | |
2000 | return out << ")"; | |
2001 | } | |
2002 | ||
2003 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
2004 | : coll(_coll), sbid_unloaded(i) | |
2005 | { | |
11fdf7f2 | 2006 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
2007 | if (get_cache()) { |
2008 | get_cache()->add_blob(); | |
2009 | } | |
2010 | } | |
2011 | ||
2012 | BlueStore::SharedBlob::~SharedBlob() | |
2013 | { | |
7c673cae FG |
2014 | if (loaded && persistent) { |
2015 | delete persistent; | |
2016 | } | |
2017 | } | |
2018 | ||
2019 | void BlueStore::SharedBlob::put() | |
2020 | { | |
2021 | if (--nref == 0) { | |
9f95a23c TL |
2022 | dout(20) << __func__ << " " << this |
2023 | << " removing self from set " << get_parent() | |
2024 | << dendl; | |
1adf2230 AA |
2025 | again: |
2026 | auto coll_snap = coll; | |
2027 | if (coll_snap) { | |
11fdf7f2 | 2028 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
2029 | if (coll_snap != coll) { |
2030 | goto again; | |
2031 | } | |
91327a77 AA |
2032 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
2033 | // race with lookup | |
2034 | return; | |
2035 | } | |
1adf2230 AA |
2036 | bc._clear(coll_snap->cache); |
2037 | coll_snap->cache->rm_blob(); | |
7c673cae | 2038 | } |
28e407b8 | 2039 | delete this; |
7c673cae FG |
2040 | } |
2041 | } | |
2042 | ||
2043 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
2044 | { | |
11fdf7f2 | 2045 | ceph_assert(persistent); |
7c673cae FG |
2046 | persistent->ref_map.get(offset, length); |
2047 | } | |
2048 | ||
2049 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 2050 | PExtentVector *r, |
11fdf7f2 | 2051 | bool *unshare) |
7c673cae | 2052 | { |
11fdf7f2 TL |
2053 | ceph_assert(persistent); |
2054 | persistent->ref_map.put(offset, length, r, | |
2055 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
2056 | } |
2057 | ||
f64942e4 AA |
2058 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
2059 | { | |
2060 | while (true) { | |
9f95a23c | 2061 | BufferCacheShard *cache = coll->cache; |
11fdf7f2 | 2062 | std::lock_guard l(cache->lock); |
f64942e4 | 2063 | if (coll->cache != cache) { |
9f95a23c TL |
2064 | dout(20) << __func__ |
2065 | << " raced with sb cache update, was " << cache | |
2066 | << ", now " << coll->cache << ", retrying" | |
2067 | << dendl; | |
f64942e4 AA |
2068 | continue; |
2069 | } | |
2070 | bc._finish_write(cache, seq); | |
2071 | break; | |
2072 | } | |
2073 | } | |
2074 | ||
3efd9988 FG |
2075 | // SharedBlobSet |
2076 | ||
2077 | #undef dout_prefix | |
2078 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
2079 | ||
11fdf7f2 TL |
2080 | template <int LogLevelV = 30> |
2081 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 2082 | { |
11fdf7f2 | 2083 | std::lock_guard l(lock); |
3efd9988 | 2084 | for (auto& i : sb_map) { |
11fdf7f2 | 2085 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
2086 | } |
2087 | } | |
2088 | ||
7c673cae FG |
2089 | // Blob |
2090 | ||
2091 | #undef dout_prefix | |
2092 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
2093 | ||
9f95a23c TL |
2094 | void BlueStore::Blob::dump(Formatter* f) const |
2095 | { | |
2096 | if (is_spanning()) { | |
2097 | f->dump_unsigned("spanning_id ", id); | |
2098 | } | |
2099 | blob.dump(f); | |
2100 | if (shared_blob) { | |
2101 | f->dump_object("shared", *shared_blob); | |
2102 | } | |
2103 | } | |
2104 | ||
7c673cae FG |
2105 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) |
2106 | { | |
2107 | out << "Blob(" << &b; | |
2108 | if (b.is_spanning()) { | |
2109 | out << " spanning " << b.id; | |
2110 | } | |
35e4c445 FG |
2111 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
2112 | if (b.shared_blob) { | |
2113 | out << " " << *b.shared_blob; | |
2114 | } else { | |
2115 | out << " (shared_blob=NULL)"; | |
2116 | } | |
2117 | out << ")"; | |
7c673cae FG |
2118 | return out; |
2119 | } | |
2120 | ||
2121 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
2122 | { | |
224ce89b | 2123 | if (get_blob().is_shared()) { |
7c673cae FG |
2124 | return; |
2125 | } | |
224ce89b | 2126 | if (get_blob().is_compressed()) { |
7c673cae FG |
2127 | bool discard = false; |
2128 | bool all_invalid = true; | |
224ce89b | 2129 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
2130 | if (!e.is_valid()) { |
2131 | discard = true; | |
2132 | } else { | |
2133 | all_invalid = false; | |
2134 | } | |
2135 | } | |
11fdf7f2 | 2136 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
2137 | // or none pextents are invalid. |
2138 | if (discard) { | |
224ce89b WB |
2139 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
2140 | get_blob().get_logical_length()); | |
7c673cae FG |
2141 | } |
2142 | } else { | |
2143 | size_t pos = 0; | |
224ce89b | 2144 | for (auto e : get_blob().get_extents()) { |
7c673cae | 2145 | if (!e.is_valid()) { |
9f95a23c TL |
2146 | dout(20) << __func__ << " 0x" << std::hex << pos |
2147 | << "~" << e.length | |
2148 | << std::dec << dendl; | |
7c673cae FG |
2149 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); |
2150 | } | |
2151 | pos += e.length; | |
2152 | } | |
224ce89b WB |
2153 | if (get_blob().can_prune_tail()) { |
2154 | dirty_blob().prune_tail(); | |
2155 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
224ce89b | 2156 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
2157 | } |
2158 | } | |
2159 | } | |
2160 | ||
2161 | void BlueStore::Blob::get_ref( | |
2162 | Collection *coll, | |
2163 | uint32_t offset, | |
2164 | uint32_t length) | |
2165 | { | |
2166 | // Caller has to initialize Blob's logical length prior to increment | |
2167 | // references. Otherwise one is neither unable to determine required | |
2168 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
2169 | // for single counter mode. | |
11fdf7f2 | 2170 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
2171 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2172 | << std::dec << " " << *this << dendl; | |
2173 | ||
2174 | if (used_in_blob.is_empty()) { | |
2175 | uint32_t min_release_size = | |
224ce89b WB |
2176 | get_blob().get_release_size(coll->store->min_alloc_size); |
2177 | uint64_t l = get_blob().get_logical_length(); | |
2178 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
2179 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
2180 | used_in_blob.init(l, min_release_size); |
2181 | } | |
2182 | used_in_blob.get( | |
2183 | offset, | |
2184 | length); | |
2185 | } | |
2186 | ||
2187 | bool BlueStore::Blob::put_ref( | |
2188 | Collection *coll, | |
2189 | uint32_t offset, | |
2190 | uint32_t length, | |
2191 | PExtentVector *r) | |
2192 | { | |
2193 | PExtentVector logical; | |
2194 | ||
7c673cae FG |
2195 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
2196 | << std::dec << " " << *this << dendl; | |
2197 | ||
2198 | bool empty = used_in_blob.put( | |
2199 | offset, | |
2200 | length, | |
2201 | &logical); | |
2202 | r->clear(); | |
2203 | // nothing to release | |
2204 | if (!empty && logical.empty()) { | |
2205 | return false; | |
2206 | } | |
2207 | ||
2208 | bluestore_blob_t& b = dirty_blob(); | |
2209 | return b.release_extents(empty, logical, r); | |
2210 | } | |
2211 | ||
224ce89b | 2212 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
2213 | uint32_t target_blob_size, |
2214 | uint32_t b_offset, | |
2215 | uint32_t *length0) { | |
11fdf7f2 TL |
2216 | ceph_assert(min_alloc_size); |
2217 | ceph_assert(target_blob_size); | |
7c673cae FG |
2218 | if (!get_blob().is_mutable()) { |
2219 | return false; | |
2220 | } | |
2221 | ||
2222 | uint32_t length = *length0; | |
2223 | uint32_t end = b_offset + length; | |
2224 | ||
2225 | // Currently for the sake of simplicity we omit blob reuse if data is | |
2226 | // unaligned with csum chunk. Later we can perform padding if needed. | |
2227 | if (get_blob().has_csum() && | |
2228 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
2229 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
2230 | return false; | |
2231 | } | |
2232 | ||
2233 | auto blen = get_blob().get_logical_length(); | |
2234 | uint32_t new_blen = blen; | |
2235 | ||
2236 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 2237 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
2238 | |
2239 | if (b_offset >= blen) { | |
224ce89b WB |
2240 | // new data totally stands out of the existing blob |
2241 | new_blen = end; | |
7c673cae | 2242 | } else { |
224ce89b | 2243 | // new data overlaps with the existing blob |
11fdf7f2 | 2244 | new_blen = std::max(blen, end); |
224ce89b WB |
2245 | |
2246 | uint32_t overlap = 0; | |
2247 | if (new_blen > blen) { | |
2248 | overlap = blen - b_offset; | |
2249 | } else { | |
2250 | overlap = length; | |
2251 | } | |
2252 | ||
2253 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
2254 | // abort if any piece of the overlap has already been allocated | |
2255 | return false; | |
7c673cae FG |
2256 | } |
2257 | } | |
224ce89b | 2258 | |
7c673cae FG |
2259 | if (new_blen > blen) { |
2260 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
2261 | // Unable to decrease the provided length to fit into max_blob_size | |
2262 | if (overflow >= length) { | |
2263 | return false; | |
2264 | } | |
2265 | ||
2266 | // FIXME: in some cases we could reduce unused resolution | |
2267 | if (get_blob().has_unused()) { | |
2268 | return false; | |
2269 | } | |
2270 | ||
2271 | if (overflow > 0) { | |
2272 | new_blen -= overflow; | |
2273 | length -= overflow; | |
2274 | *length0 = length; | |
2275 | } | |
224ce89b | 2276 | |
7c673cae FG |
2277 | if (new_blen > blen) { |
2278 | dirty_blob().add_tail(new_blen); | |
2279 | used_in_blob.add_tail(new_blen, | |
224ce89b | 2280 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
2281 | } |
2282 | } | |
2283 | return true; | |
2284 | } | |
2285 | ||
2286 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
2287 | { | |
7c673cae FG |
2288 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec |
2289 | << " start " << *this << dendl; | |
11fdf7f2 TL |
2290 | ceph_assert(blob.can_split()); |
2291 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
2292 | bluestore_blob_t &lb = dirty_blob(); |
2293 | bluestore_blob_t &rb = r->dirty_blob(); | |
2294 | ||
2295 | used_in_blob.split( | |
2296 | blob_offset, | |
2297 | &(r->used_in_blob)); | |
2298 | ||
2299 | lb.split(blob_offset, rb); | |
2300 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2301 | ||
2302 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2303 | << " finish " << *this << dendl; | |
2304 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2305 | << " and " << *r << dendl; | |
2306 | } | |
2307 | ||
2308 | #ifndef CACHE_BLOB_BL | |
2309 | void BlueStore::Blob::decode( | |
2310 | Collection *coll, | |
11fdf7f2 | 2311 | bufferptr::const_iterator& p, |
7c673cae FG |
2312 | uint64_t struct_v, |
2313 | uint64_t* sbid, | |
2314 | bool include_ref_map) | |
2315 | { | |
2316 | denc(blob, p, struct_v); | |
2317 | if (blob.is_shared()) { | |
2318 | denc(*sbid, p); | |
2319 | } | |
2320 | if (include_ref_map) { | |
2321 | if (struct_v > 1) { | |
2322 | used_in_blob.decode(p); | |
2323 | } else { | |
2324 | used_in_blob.clear(); | |
2325 | bluestore_extent_ref_map_t legacy_ref_map; | |
2326 | legacy_ref_map.decode(p); | |
2327 | for (auto r : legacy_ref_map.ref_map) { | |
2328 | get_ref( | |
2329 | coll, | |
2330 | r.first, | |
2331 | r.second.refs * r.second.length); | |
2332 | } | |
2333 | } | |
2334 | } | |
2335 | } | |
2336 | #endif | |
2337 | ||
2338 | // Extent | |
2339 | ||
9f95a23c TL |
2340 | void BlueStore::Extent::dump(Formatter* f) const |
2341 | { | |
2342 | f->dump_unsigned("logical_offset", logical_offset); | |
2343 | f->dump_unsigned("length", length); | |
2344 | f->dump_unsigned("blob_offset", blob_offset); | |
2345 | f->dump_object("blob", *blob); | |
2346 | } | |
2347 | ||
7c673cae FG |
2348 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) |
2349 | { | |
2350 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2351 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2352 | << " " << *e.blob; | |
2353 | } | |
2354 | ||
2355 | // OldExtent | |
2356 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2357 | uint32_t lo, | |
2358 | uint32_t o, | |
2359 | uint32_t l, | |
2360 | BlobRef& b) { | |
2361 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2362 | b->put_ref(c.get(), o, l, &(oe->r)); | |
2363 | oe->blob_empty = b->get_referenced_bytes() == 0; | |
2364 | return oe; | |
2365 | } | |
2366 | ||
2367 | // ExtentMap | |
2368 | ||
2369 | #undef dout_prefix | |
2370 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
9f95a23c TL |
2371 | #undef dout_context |
2372 | #define dout_context onode->c->store->cct | |
7c673cae FG |
2373 | |
2374 | BlueStore::ExtentMap::ExtentMap(Onode *o) | |
2375 | : onode(o), | |
2376 | inline_bl( | |
2377 | o->c->store->cct->_conf->bluestore_extent_map_inline_shard_prealloc_size) { | |
2378 | } | |
2379 | ||
9f95a23c TL |
2380 | void BlueStore::ExtentMap::dump(Formatter* f) const |
2381 | { | |
2382 | f->open_array_section("extents"); | |
2383 | ||
2384 | for (auto& e : extent_map) { | |
2385 | f->dump_object("extent", e); | |
2386 | } | |
2387 | f->close_section(); | |
2388 | } | |
2389 | ||
11fdf7f2 TL |
2390 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2391 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2392 | uint64_t& length, uint64_t& dstoff) { | |
2393 | ||
2394 | auto cct = onode->c->store->cct; | |
2395 | bool inject_21040 = | |
2396 | cct->_conf->bluestore_debug_inject_bug21040; | |
2397 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2398 | for (auto& e : oldo->extent_map.extent_map) { | |
2399 | e.blob->last_encoded_id = -1; | |
2400 | } | |
2401 | ||
2402 | int n = 0; | |
2403 | uint64_t end = srcoff + length; | |
2404 | uint32_t dirty_range_begin = 0; | |
2405 | uint32_t dirty_range_end = 0; | |
2406 | bool src_dirty = false; | |
2407 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2408 | ep != oldo->extent_map.extent_map.end(); | |
2409 | ++ep) { | |
2410 | auto& e = *ep; | |
2411 | if (e.logical_offset >= end) { | |
2412 | break; | |
2413 | } | |
2414 | dout(20) << __func__ << " src " << e << dendl; | |
2415 | BlobRef cb; | |
2416 | bool blob_duped = true; | |
2417 | if (e.blob->last_encoded_id >= 0) { | |
2418 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2419 | blob_duped = false; | |
2420 | } else { | |
2421 | // dup the blob | |
2422 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2423 | // make sure it is shared | |
2424 | if (!blob.is_shared()) { | |
2425 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2426 | if (!inject_21040 && !src_dirty) { | |
2427 | src_dirty = true; | |
2428 | dirty_range_begin = e.logical_offset; | |
2429 | } else if (inject_21040 && | |
2430 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2431 | dirty_range_begin = e.logical_offset; | |
2432 | } | |
2433 | ceph_assert(e.logical_end() > 0); | |
2434 | // -1 to exclude next potential shard | |
2435 | dirty_range_end = e.logical_end() - 1; | |
2436 | } else { | |
2437 | c->load_shared_blob(e.blob->shared_blob); | |
2438 | } | |
2439 | cb = new Blob(); | |
2440 | e.blob->last_encoded_id = n; | |
2441 | id_to_blob[n] = cb; | |
2442 | e.blob->dup(*cb); | |
2443 | // bump the extent refs on the copied blob's extents | |
2444 | for (auto p : blob.get_extents()) { | |
2445 | if (p.is_valid()) { | |
2446 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2447 | } | |
2448 | } | |
2449 | txc->write_shared_blob(e.blob->shared_blob); | |
2450 | dout(20) << __func__ << " new " << *cb << dendl; | |
2451 | } | |
2452 | ||
2453 | int skip_front, skip_back; | |
2454 | if (e.logical_offset < srcoff) { | |
2455 | skip_front = srcoff - e.logical_offset; | |
2456 | } else { | |
2457 | skip_front = 0; | |
2458 | } | |
2459 | if (e.logical_end() > end) { | |
2460 | skip_back = e.logical_end() - end; | |
2461 | } else { | |
2462 | skip_back = 0; | |
2463 | } | |
2464 | ||
2465 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2466 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2467 | newo->extent_map.extent_map.insert(*ne); | |
2468 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2469 | // fixme: we may leave parts of new blob unreferenced that could | |
2470 | // be freed (relative to the shared_blob). | |
2471 | txc->statfs_delta.stored() += ne->length; | |
2472 | if (e.blob->get_blob().is_compressed()) { | |
2473 | txc->statfs_delta.compressed_original() += ne->length; | |
2474 | if (blob_duped) { | |
2475 | txc->statfs_delta.compressed() += | |
2476 | cb->get_blob().get_compressed_payload_length(); | |
2477 | } | |
2478 | } | |
2479 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2480 | ++n; | |
2481 | } | |
2482 | if ((!inject_21040 && src_dirty) || | |
2483 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2484 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2485 | dirty_range_end - dirty_range_begin); | |
2486 | txc->write_onode(oldo); | |
2487 | } | |
2488 | txc->write_onode(newo); | |
2489 | ||
2490 | if (dstoff + length > newo->onode.size) { | |
2491 | newo->onode.size = dstoff + length; | |
2492 | } | |
2493 | newo->extent_map.dirty_range(dstoff, length); | |
2494 | } | |
7c673cae FG |
2495 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2496 | bool force) | |
2497 | { | |
2498 | auto cct = onode->c->store->cct; //used by dout | |
2499 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2500 | if (onode->onode.extent_map_shards.empty()) { | |
2501 | if (inline_bl.length() == 0) { | |
2502 | unsigned n; | |
2503 | // we need to encode inline_bl to measure encoded length | |
2504 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
f91f0fd5 | 2505 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_inline_bl); |
11fdf7f2 | 2506 | ceph_assert(!never_happen); |
7c673cae FG |
2507 | size_t len = inline_bl.length(); |
2508 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2509 | << " extents" << dendl; | |
2510 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2511 | request_reshard(0, OBJECT_MAX_SIZE); | |
2512 | return; | |
2513 | } | |
2514 | } | |
2515 | // will persist in the onode key. | |
2516 | } else { | |
2517 | // pending shard update | |
2518 | struct dirty_shard_t { | |
2519 | Shard *shard; | |
2520 | bufferlist bl; | |
2521 | dirty_shard_t(Shard *s) : shard(s) {} | |
2522 | }; | |
2523 | vector<dirty_shard_t> encoded_shards; | |
2524 | // allocate slots for all shards in a single call instead of | |
2525 | // doing multiple allocations - one per each dirty shard | |
2526 | encoded_shards.reserve(shards.size()); | |
2527 | ||
2528 | auto p = shards.begin(); | |
2529 | auto prev_p = p; | |
2530 | while (p != shards.end()) { | |
11fdf7f2 | 2531 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2532 | auto n = p; |
2533 | ++n; | |
2534 | if (p->dirty) { | |
2535 | uint32_t endoff; | |
2536 | if (n == shards.end()) { | |
2537 | endoff = OBJECT_MAX_SIZE; | |
2538 | } else { | |
2539 | endoff = n->shard_info->offset; | |
2540 | } | |
2541 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2542 | bufferlist& bl = encoded_shards.back().bl; | |
2543 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2544 | bl, &p->extents)) { | |
2545 | if (force) { | |
2546 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2547 | ceph_assert(!force); |
7c673cae FG |
2548 | } |
2549 | } | |
2550 | size_t len = bl.length(); | |
2551 | ||
2552 | dout(20) << __func__ << " shard 0x" << std::hex | |
2553 | << p->shard_info->offset << std::dec << " is " << len | |
2554 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2555 | << p->extents << " extents" << dendl; | |
2556 | ||
2557 | if (!force) { | |
2558 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2559 | // we are big; reshard ourselves | |
2560 | request_reshard(p->shard_info->offset, endoff); | |
2561 | } | |
2562 | // avoid resharding the trailing shard, even if it is small | |
2563 | else if (n != shards.end() && | |
11fdf7f2 TL |
2564 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2565 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2566 | if (p == shards.begin()) { |
2567 | // we are the first shard, combine with next shard | |
7c673cae | 2568 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2569 | } else { |
31f18b77 FG |
2570 | // combine either with the previous shard or the next, |
2571 | // whichever is smaller | |
7c673cae FG |
2572 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2573 | request_reshard(p->shard_info->offset, endoff + 1); | |
2574 | } else { | |
2575 | request_reshard(prev_p->shard_info->offset, endoff); | |
2576 | } | |
2577 | } | |
2578 | } | |
2579 | } | |
2580 | } | |
2581 | prev_p = p; | |
2582 | p = n; | |
2583 | } | |
2584 | if (needs_reshard()) { | |
2585 | return; | |
2586 | } | |
2587 | ||
2588 | // schedule DB update for dirty shards | |
2589 | string key; | |
2590 | for (auto& it : encoded_shards) { | |
2591 | it.shard->dirty = false; | |
2592 | it.shard->shard_info->bytes = it.bl.length(); | |
2593 | generate_extent_shard_key_and_apply( | |
2594 | onode->key, | |
2595 | it.shard->shard_info->offset, | |
2596 | &key, | |
2597 | [&](const string& final_key) { | |
2598 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2599 | } | |
2600 | ); | |
2601 | } | |
2602 | } | |
2603 | } | |
2604 | ||
31f18b77 FG |
2605 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2606 | { | |
2607 | if (spanning_blob_map.empty()) | |
2608 | return 0; | |
2609 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2610 | // bid is valid and available. | |
2611 | if (bid >= 0) | |
2612 | return bid; | |
2613 | // Find next unused bid; | |
2614 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2615 | const auto begin_bid = bid; | |
2616 | do { | |
2617 | if (!spanning_blob_map.count(bid)) | |
2618 | return bid; | |
2619 | else { | |
2620 | bid++; | |
2621 | if (bid < 0) bid = 0; | |
2622 | } | |
2623 | } while (bid != begin_bid); | |
81eedcae TL |
2624 | auto cct = onode->c->store->cct; // used by dout |
2625 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2626 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2627 | } |
2628 | ||
7c673cae FG |
2629 | void BlueStore::ExtentMap::reshard( |
2630 | KeyValueDB *db, | |
2631 | KeyValueDB::Transaction t) | |
2632 | { | |
2633 | auto cct = onode->c->store->cct; // used by dout | |
2634 | ||
2635 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2636 | << needs_reshard_end << ")" << std::dec | |
2637 | << " of " << onode->onode.extent_map_shards.size() | |
2638 | << " shards on " << onode->oid << dendl; | |
2639 | for (auto& p : spanning_blob_map) { | |
2640 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2641 | << dendl; | |
2642 | } | |
2643 | // determine shard index range | |
2644 | unsigned si_begin = 0, si_end = 0; | |
2645 | if (!shards.empty()) { | |
2646 | while (si_begin + 1 < shards.size() && | |
2647 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2648 | ++si_begin; | |
2649 | } | |
2650 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2651 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2652 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2653 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2654 | break; | |
2655 | } | |
2656 | } | |
2657 | if (si_end == shards.size()) { | |
2658 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2659 | } | |
2660 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2661 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2662 | << needs_reshard_end << ")" << std::dec << dendl; | |
2663 | } | |
2664 | ||
181888fb | 2665 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2666 | |
2667 | // we may need to fault in a larger interval later must have all | |
2668 | // referring extents for spanning blobs loaded in order to have | |
2669 | // accurate use_tracker values. | |
2670 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2671 | uint32_t spanning_scan_end = needs_reshard_end; | |
2672 | ||
2673 | // remove old keys | |
2674 | string key; | |
2675 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2676 | generate_extent_shard_key_and_apply( | |
2677 | onode->key, shards[i].shard_info->offset, &key, | |
2678 | [&](const string& final_key) { | |
2679 | t->rmkey(PREFIX_OBJ, final_key); | |
2680 | } | |
2681 | ); | |
2682 | } | |
2683 | ||
2684 | // calculate average extent size | |
2685 | unsigned bytes = 0; | |
2686 | unsigned extents = 0; | |
2687 | if (onode->onode.extent_map_shards.empty()) { | |
2688 | bytes = inline_bl.length(); | |
2689 | extents = extent_map.size(); | |
2690 | } else { | |
2691 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2692 | bytes += shards[i].shard_info->bytes; | |
2693 | extents += shards[i].extents; | |
2694 | } | |
2695 | } | |
2696 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2697 | unsigned slop = target * | |
2698 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2699 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2700 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2701 | << ", slop " << slop << dendl; | |
2702 | ||
2703 | // reshard | |
2704 | unsigned estimate = 0; | |
31f18b77 | 2705 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2706 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2707 | unsigned max_blob_end = 0; | |
2708 | Extent dummy(needs_reshard_begin); | |
2709 | for (auto e = extent_map.lower_bound(dummy); | |
2710 | e != extent_map.end(); | |
2711 | ++e) { | |
2712 | if (e->logical_offset >= needs_reshard_end) { | |
2713 | break; | |
2714 | } | |
2715 | dout(30) << " extent " << *e << dendl; | |
2716 | ||
2717 | // disfavor shard boundaries that span a blob | |
2718 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2719 | if (estimate && | |
2720 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2721 | // new shard | |
31f18b77 | 2722 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2723 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2724 | new_shard_info.back().offset = offset; | |
2725 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2726 | << std::dec << dendl; |
7c673cae FG |
2727 | } |
2728 | offset = e->logical_offset; | |
2729 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2730 | new_shard_info.back().offset = offset; | |
2731 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2732 | << std::dec << dendl; | |
2733 | estimate = 0; | |
2734 | } | |
2735 | estimate += extent_avg; | |
31f18b77 FG |
2736 | unsigned bs = e->blob_start(); |
2737 | if (bs < spanning_scan_begin) { | |
2738 | spanning_scan_begin = bs; | |
7c673cae FG |
2739 | } |
2740 | uint32_t be = e->blob_end(); | |
2741 | if (be > max_blob_end) { | |
2742 | max_blob_end = be; | |
2743 | } | |
2744 | if (be > spanning_scan_end) { | |
2745 | spanning_scan_end = be; | |
2746 | } | |
2747 | } | |
2748 | if (new_shard_info.empty() && (si_begin > 0 || | |
2749 | si_end < shards.size())) { | |
2750 | // we resharded a partial range; we must produce at least one output | |
2751 | // shard | |
2752 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2753 | new_shard_info.back().offset = needs_reshard_begin; | |
2754 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2755 | << std::dec << " (singleton degenerate case)" << dendl; | |
2756 | } | |
2757 | ||
2758 | auto& sv = onode->onode.extent_map_shards; | |
2759 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2760 | dout(20) << __func__ << " old " << sv << dendl; | |
2761 | if (sv.empty()) { | |
2762 | // no old shards to keep | |
2763 | sv.swap(new_shard_info); | |
2764 | init_shards(true, true); | |
2765 | } else { | |
2766 | // splice in new shards | |
2767 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2768 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2769 | sv.insert( | |
2770 | sv.begin() + si_begin, | |
2771 | new_shard_info.begin(), | |
2772 | new_shard_info.end()); | |
2773 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2774 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2775 | |
11fdf7f2 | 2776 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2777 | |
2778 | // note that we need to update every shard_info of shards here, | |
2779 | // as sv might have been totally re-allocated above | |
2780 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2781 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2782 | } |
2783 | ||
2784 | // mark newly added shards as dirty | |
2785 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2786 | shards[i].loaded = true; |
2787 | shards[i].dirty = true; | |
2788 | } | |
7c673cae FG |
2789 | } |
2790 | dout(20) << __func__ << " fin " << sv << dendl; | |
2791 | inline_bl.clear(); | |
2792 | ||
2793 | if (sv.empty()) { | |
2794 | // no more shards; unspan all previously spanning blobs | |
2795 | auto p = spanning_blob_map.begin(); | |
2796 | while (p != spanning_blob_map.end()) { | |
2797 | p->second->id = -1; | |
2798 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2799 | p = spanning_blob_map.erase(p); | |
2800 | } | |
2801 | } else { | |
2802 | // identify new spanning blobs | |
2803 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2804 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2805 | if (spanning_scan_begin < needs_reshard_begin) { | |
2806 | fault_range(db, spanning_scan_begin, | |
2807 | needs_reshard_begin - spanning_scan_begin); | |
2808 | } | |
2809 | if (spanning_scan_end > needs_reshard_end) { | |
2810 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2811 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2812 | } |
2813 | auto sp = sv.begin() + si_begin; | |
2814 | auto esp = sv.end(); | |
2815 | unsigned shard_start = sp->offset; | |
2816 | unsigned shard_end; | |
2817 | ++sp; | |
2818 | if (sp == esp) { | |
2819 | shard_end = OBJECT_MAX_SIZE; | |
2820 | } else { | |
2821 | shard_end = sp->offset; | |
2822 | } | |
7c673cae | 2823 | Extent dummy(needs_reshard_begin); |
9f95a23c TL |
2824 | |
2825 | bool was_too_many_blobs_check = false; | |
2826 | auto too_many_blobs_threshold = | |
2827 | g_conf()->bluestore_debug_too_many_blobs_threshold; | |
2828 | auto& dumped_onodes = onode->c->onode_map.cache->dumped_onodes; | |
2829 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oid_slot = nullptr; | |
2830 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oldest_slot = nullptr; | |
2831 | ||
7c673cae FG |
2832 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { |
2833 | if (e->logical_offset >= needs_reshard_end) { | |
2834 | break; | |
2835 | } | |
2836 | dout(30) << " extent " << *e << dendl; | |
2837 | while (e->logical_offset >= shard_end) { | |
2838 | shard_start = shard_end; | |
11fdf7f2 | 2839 | ceph_assert(sp != esp); |
7c673cae FG |
2840 | ++sp; |
2841 | if (sp == esp) { | |
2842 | shard_end = OBJECT_MAX_SIZE; | |
2843 | } else { | |
2844 | shard_end = sp->offset; | |
2845 | } | |
2846 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2847 | << " to 0x" << shard_end << std::dec << dendl; | |
2848 | } | |
9f95a23c | 2849 | |
7c673cae FG |
2850 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { |
2851 | if (!e->blob->is_spanning()) { | |
2852 | // We have two options: (1) split the blob into pieces at the | |
2853 | // shard boundaries (and adjust extents accordingly), or (2) | |
2854 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2855 | // we may have to split it multiple times--potentially at every | |
2856 | // shard boundary. | |
2857 | bool must_span = false; | |
2858 | BlobRef b = e->blob; | |
2859 | if (b->can_split()) { | |
2860 | uint32_t bstart = e->blob_start(); | |
2861 | uint32_t bend = e->blob_end(); | |
2862 | for (const auto& sh : shards) { | |
2863 | if (bstart < sh.shard_info->offset && | |
2864 | bend > sh.shard_info->offset) { | |
2865 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2866 | if (b->can_split_at(blob_offset)) { | |
2867 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2868 | << std::hex << bstart << " blob_offset 0x" | |
2869 | << blob_offset << std::dec << " " << *b << dendl; | |
2870 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2871 | // switch b to the new right-hand side, in case it | |
2872 | // *also* has to get split. | |
2873 | bstart += blob_offset; | |
2874 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2875 | } else { | |
2876 | must_span = true; | |
2877 | break; | |
2878 | } | |
2879 | } | |
2880 | } | |
2881 | } else { | |
2882 | must_span = true; | |
2883 | } | |
2884 | if (must_span) { | |
31f18b77 FG |
2885 | auto bid = allocate_spanning_blob_id(); |
2886 | b->id = bid; | |
7c673cae FG |
2887 | spanning_blob_map[b->id] = b; |
2888 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
9f95a23c TL |
2889 | if (!was_too_many_blobs_check && |
2890 | too_many_blobs_threshold && | |
2891 | spanning_blob_map.size() >= size_t(too_many_blobs_threshold)) { | |
2892 | ||
2893 | was_too_many_blobs_check = true; | |
2894 | for (size_t i = 0; i < dumped_onodes.size(); ++i) { | |
2895 | if (dumped_onodes[i].first == onode->oid) { | |
2896 | oid_slot = &dumped_onodes[i]; | |
2897 | break; | |
2898 | } | |
2899 | if (!oldest_slot || (oldest_slot && | |
2900 | dumped_onodes[i].second < oldest_slot->second)) { | |
2901 | oldest_slot = &dumped_onodes[i]; | |
2902 | } | |
2903 | } | |
2904 | } | |
7c673cae FG |
2905 | } |
2906 | } | |
2907 | } else { | |
2908 | if (e->blob->is_spanning()) { | |
2909 | spanning_blob_map.erase(e->blob->id); | |
2910 | e->blob->id = -1; | |
2911 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
2912 | } | |
2913 | } | |
2914 | } | |
9f95a23c TL |
2915 | bool do_dump = (!oid_slot && was_too_many_blobs_check) || |
2916 | (oid_slot && | |
2917 | (mono_clock::now() - oid_slot->second >= make_timespan(5 * 60))); | |
2918 | if (do_dump) { | |
2919 | dout(0) << __func__ | |
2920 | << " spanning blob count exceeds threshold, " | |
2921 | << spanning_blob_map.size() << " spanning blobs" | |
2922 | << dendl; | |
2923 | _dump_onode<0>(cct, *onode); | |
2924 | if (oid_slot) { | |
2925 | oid_slot->second = mono_clock::now(); | |
2926 | } else { | |
2927 | ceph_assert(oldest_slot); | |
2928 | oldest_slot->first = onode->oid; | |
2929 | oldest_slot->second = mono_clock::now(); | |
2930 | } | |
2931 | } | |
7c673cae FG |
2932 | } |
2933 | ||
2934 | clear_needs_reshard(); | |
2935 | } | |
2936 | ||
2937 | bool BlueStore::ExtentMap::encode_some( | |
2938 | uint32_t offset, | |
2939 | uint32_t length, | |
2940 | bufferlist& bl, | |
2941 | unsigned *pn) | |
2942 | { | |
7c673cae FG |
2943 | Extent dummy(offset); |
2944 | auto start = extent_map.lower_bound(dummy); | |
2945 | uint32_t end = offset + length; | |
2946 | ||
2947 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
2948 | // serialization only. Hence there is no specific | |
2949 | // handling at ExtentMap level. | |
2950 | ||
2951 | unsigned n = 0; | |
2952 | size_t bound = 0; | |
7c673cae FG |
2953 | bool must_reshard = false; |
2954 | for (auto p = start; | |
2955 | p != extent_map.end() && p->logical_offset < end; | |
2956 | ++p, ++n) { | |
11fdf7f2 | 2957 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
2958 | p->blob->last_encoded_id = -1; |
2959 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
2960 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2961 | << std::dec << " hit new spanning blob " << *p << dendl; | |
2962 | request_reshard(p->blob_start(), p->blob_end()); | |
2963 | must_reshard = true; | |
2964 | } | |
31f18b77 FG |
2965 | if (!must_reshard) { |
2966 | denc_varint(0, bound); // blobid | |
2967 | denc_varint(0, bound); // logical_offset | |
2968 | denc_varint(0, bound); // len | |
2969 | denc_varint(0, bound); // blob_offset | |
7c673cae | 2970 | |
31f18b77 FG |
2971 | p->blob->bound_encode( |
2972 | bound, | |
2973 | struct_v, | |
2974 | p->blob->shared_blob->get_sbid(), | |
2975 | false); | |
2976 | } | |
7c673cae FG |
2977 | } |
2978 | if (must_reshard) { | |
2979 | return true; | |
2980 | } | |
2981 | ||
31f18b77 FG |
2982 | denc(struct_v, bound); |
2983 | denc_varint(0, bound); // number of extents | |
2984 | ||
7c673cae FG |
2985 | { |
2986 | auto app = bl.get_contiguous_appender(bound); | |
2987 | denc(struct_v, app); | |
2988 | denc_varint(n, app); | |
2989 | if (pn) { | |
2990 | *pn = n; | |
2991 | } | |
2992 | ||
2993 | n = 0; | |
2994 | uint64_t pos = 0; | |
2995 | uint64_t prev_len = 0; | |
2996 | for (auto p = start; | |
2997 | p != extent_map.end() && p->logical_offset < end; | |
2998 | ++p, ++n) { | |
2999 | unsigned blobid; | |
3000 | bool include_blob = false; | |
3001 | if (p->blob->is_spanning()) { | |
3002 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
3003 | blobid |= BLOBID_FLAG_SPANNING; | |
3004 | } else if (p->blob->last_encoded_id < 0) { | |
3005 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
3006 | include_blob = true; | |
3007 | blobid = 0; // the decoder will infer the id from n | |
3008 | } else { | |
3009 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
3010 | } | |
3011 | if (p->logical_offset == pos) { | |
3012 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
3013 | } | |
3014 | if (p->blob_offset == 0) { | |
3015 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
3016 | } | |
3017 | if (p->length == prev_len) { | |
3018 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
3019 | } else { | |
3020 | prev_len = p->length; | |
3021 | } | |
3022 | denc_varint(blobid, app); | |
3023 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3024 | denc_varint_lowz(p->logical_offset - pos, app); | |
3025 | } | |
3026 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3027 | denc_varint_lowz(p->blob_offset, app); | |
3028 | } | |
3029 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3030 | denc_varint_lowz(p->length, app); | |
3031 | } | |
3032 | pos = p->logical_end(); | |
3033 | if (include_blob) { | |
3034 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
3035 | } | |
3036 | } | |
3037 | } | |
3038 | /*derr << __func__ << bl << dendl; | |
3039 | derr << __func__ << ":"; | |
3040 | bl.hexdump(*_dout); | |
3041 | *_dout << dendl; | |
3042 | */ | |
3043 | return false; | |
3044 | } | |
3045 | ||
3046 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
3047 | { | |
7c673cae FG |
3048 | /* |
3049 | derr << __func__ << ":"; | |
3050 | bl.hexdump(*_dout); | |
3051 | *_dout << dendl; | |
3052 | */ | |
3053 | ||
11fdf7f2 | 3054 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae FG |
3055 | auto p = bl.front().begin_deep(); |
3056 | __u8 struct_v; | |
3057 | denc(struct_v, p); | |
3058 | // Version 2 differs from v1 in blob's ref_map | |
3059 | // serialization only. Hence there is no specific | |
3060 | // handling at ExtentMap level below. | |
11fdf7f2 | 3061 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
3062 | |
3063 | uint32_t num; | |
3064 | denc_varint(num, p); | |
3065 | vector<BlobRef> blobs(num); | |
3066 | uint64_t pos = 0; | |
3067 | uint64_t prev_len = 0; | |
3068 | unsigned n = 0; | |
3069 | ||
3070 | while (!p.end()) { | |
3071 | Extent *le = new Extent(); | |
3072 | uint64_t blobid; | |
3073 | denc_varint(blobid, p); | |
3074 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
3075 | uint64_t gap; | |
3076 | denc_varint_lowz(gap, p); | |
3077 | pos += gap; | |
3078 | } | |
3079 | le->logical_offset = pos; | |
3080 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
3081 | denc_varint_lowz(le->blob_offset, p); | |
3082 | } else { | |
3083 | le->blob_offset = 0; | |
3084 | } | |
3085 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
3086 | denc_varint_lowz(prev_len, p); | |
3087 | } | |
3088 | le->length = prev_len; | |
3089 | ||
3090 | if (blobid & BLOBID_FLAG_SPANNING) { | |
3091 | dout(30) << __func__ << " getting spanning blob " | |
3092 | << (blobid >> BLOBID_SHIFT_BITS) << dendl; | |
3093 | le->assign_blob(get_spanning_blob(blobid >> BLOBID_SHIFT_BITS)); | |
3094 | } else { | |
3095 | blobid >>= BLOBID_SHIFT_BITS; | |
3096 | if (blobid) { | |
3097 | le->assign_blob(blobs[blobid - 1]); | |
11fdf7f2 | 3098 | ceph_assert(le->blob); |
7c673cae FG |
3099 | } else { |
3100 | Blob *b = new Blob(); | |
3101 | uint64_t sbid = 0; | |
3102 | b->decode(onode->c, p, struct_v, &sbid, false); | |
3103 | blobs[n] = b; | |
3104 | onode->c->open_shared_blob(sbid, b); | |
3105 | le->assign_blob(b); | |
3106 | } | |
3107 | // we build ref_map dynamically for non-spanning blobs | |
3108 | le->blob->get_ref( | |
3109 | onode->c, | |
3110 | le->blob_offset, | |
3111 | le->length); | |
3112 | } | |
3113 | pos += prev_len; | |
3114 | ++n; | |
3115 | extent_map.insert(*le); | |
3116 | } | |
3117 | ||
11fdf7f2 | 3118 | ceph_assert(n == num); |
7c673cae FG |
3119 | return num; |
3120 | } | |
3121 | ||
3122 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
3123 | { | |
3124 | // Version 2 differs from v1 in blob's ref_map | |
3125 | // serialization only. Hence there is no specific | |
3126 | // handling at ExtentMap level. | |
3127 | __u8 struct_v = 2; | |
3128 | ||
3129 | denc(struct_v, p); | |
3130 | denc_varint((uint32_t)0, p); | |
3131 | size_t key_size = 0; | |
3132 | denc_varint((uint32_t)0, key_size); | |
3133 | p += spanning_blob_map.size() * key_size; | |
3134 | for (const auto& i : spanning_blob_map) { | |
3135 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3136 | } | |
3137 | } | |
3138 | ||
3139 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
3140 | bufferlist::contiguous_appender& p) | |
3141 | { | |
3142 | // Version 2 differs from v1 in blob's ref_map | |
3143 | // serialization only. Hence there is no specific | |
3144 | // handling at ExtentMap level. | |
3145 | __u8 struct_v = 2; | |
3146 | ||
3147 | denc(struct_v, p); | |
3148 | denc_varint(spanning_blob_map.size(), p); | |
3149 | for (auto& i : spanning_blob_map) { | |
3150 | denc_varint(i.second->id, p); | |
3151 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
3152 | } | |
3153 | } | |
3154 | ||
3155 | void BlueStore::ExtentMap::decode_spanning_blobs( | |
11fdf7f2 | 3156 | bufferptr::const_iterator& p) |
7c673cae FG |
3157 | { |
3158 | __u8 struct_v; | |
3159 | denc(struct_v, p); | |
3160 | // Version 2 differs from v1 in blob's ref_map | |
3161 | // serialization only. Hence there is no specific | |
3162 | // handling at ExtentMap level. | |
11fdf7f2 | 3163 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
3164 | |
3165 | unsigned n; | |
3166 | denc_varint(n, p); | |
3167 | while (n--) { | |
3168 | BlobRef b(new Blob()); | |
3169 | denc_varint(b->id, p); | |
3170 | spanning_blob_map[b->id] = b; | |
3171 | uint64_t sbid = 0; | |
3172 | b->decode(onode->c, p, struct_v, &sbid, true); | |
3173 | onode->c->open_shared_blob(sbid, b); | |
3174 | } | |
3175 | } | |
3176 | ||
3177 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) | |
3178 | { | |
3179 | shards.resize(onode->onode.extent_map_shards.size()); | |
3180 | unsigned i = 0; | |
3181 | for (auto &s : onode->onode.extent_map_shards) { | |
3182 | shards[i].shard_info = &s; | |
3183 | shards[i].loaded = loaded; | |
3184 | shards[i].dirty = dirty; | |
3185 | ++i; | |
3186 | } | |
3187 | } | |
3188 | ||
3189 | void BlueStore::ExtentMap::fault_range( | |
3190 | KeyValueDB *db, | |
3191 | uint32_t offset, | |
3192 | uint32_t length) | |
3193 | { | |
7c673cae FG |
3194 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3195 | << std::dec << dendl; | |
3196 | auto start = seek_shard(offset); | |
3197 | auto last = seek_shard(offset + length); | |
3198 | ||
3199 | if (start < 0) | |
3200 | return; | |
3201 | ||
11fdf7f2 | 3202 | ceph_assert(last >= start); |
7c673cae FG |
3203 | string key; |
3204 | while (start <= last) { | |
11fdf7f2 | 3205 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3206 | auto p = &shards[start]; |
3207 | if (!p->loaded) { | |
3208 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
3209 | << p->shard_info->offset << std::dec << dendl; | |
3210 | bufferlist v; | |
3211 | generate_extent_shard_key_and_apply( | |
3212 | onode->key, p->shard_info->offset, &key, | |
3213 | [&](const string& final_key) { | |
3214 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
3215 | if (r < 0) { | |
3216 | derr << __func__ << " missing shard 0x" << std::hex | |
3217 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
3218 | << dendl; | |
11fdf7f2 | 3219 | ceph_assert(r >= 0); |
7c673cae FG |
3220 | } |
3221 | } | |
3222 | ); | |
3223 | p->extents = decode_some(v); | |
3224 | p->loaded = true; | |
3225 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
3226 | << p->shard_info->offset |
3227 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 3228 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
3229 | ceph_assert(p->dirty == false); |
3230 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
3231 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
3232 | } else { | |
3233 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
3234 | } | |
3235 | ++start; | |
3236 | } | |
3237 | } | |
3238 | ||
3239 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
3240 | uint32_t offset, |
3241 | uint32_t length) | |
3242 | { | |
7c673cae FG |
3243 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3244 | << std::dec << dendl; | |
3245 | if (shards.empty()) { | |
3246 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
3247 | inline_bl.clear(); | |
3248 | return; | |
3249 | } | |
3250 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
3251 | if (length == 0) { |
3252 | length = 1; | |
3253 | } | |
3254 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
3255 | if (start < 0) |
3256 | return; | |
3257 | ||
11fdf7f2 | 3258 | ceph_assert(last >= start); |
7c673cae | 3259 | while (start <= last) { |
11fdf7f2 | 3260 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3261 | auto p = &shards[start]; |
3262 | if (!p->loaded) { | |
11fdf7f2 TL |
3263 | derr << __func__ << "on write 0x" << std::hex << offset |
3264 | << "~" << length << " shard 0x" << p->shard_info->offset | |
3265 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
3266 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
3267 | } |
3268 | if (!p->dirty) { | |
3269 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
3270 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
3271 | p->dirty = true; | |
3272 | } | |
3273 | ++start; | |
3274 | } | |
3275 | } | |
3276 | ||
3277 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
3278 | uint64_t offset) | |
3279 | { | |
3280 | Extent dummy(offset); | |
3281 | return extent_map.find(dummy); | |
3282 | } | |
3283 | ||
7c673cae FG |
3284 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
3285 | uint64_t offset) | |
3286 | { | |
3287 | Extent dummy(offset); | |
3288 | auto fp = extent_map.lower_bound(dummy); | |
3289 | if (fp != extent_map.begin()) { | |
3290 | --fp; | |
3291 | if (fp->logical_end() <= offset) { | |
3292 | ++fp; | |
3293 | } | |
3294 | } | |
3295 | return fp; | |
3296 | } | |
3297 | ||
3298 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
3299 | uint64_t offset) const | |
3300 | { | |
3301 | Extent dummy(offset); | |
3302 | auto fp = extent_map.lower_bound(dummy); | |
3303 | if (fp != extent_map.begin()) { | |
3304 | --fp; | |
3305 | if (fp->logical_end() <= offset) { | |
3306 | ++fp; | |
3307 | } | |
3308 | } | |
3309 | return fp; | |
3310 | } | |
3311 | ||
3312 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
3313 | { | |
3314 | auto fp = seek_lextent(offset); | |
3315 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
3316 | return false; | |
3317 | } | |
3318 | return true; | |
3319 | } | |
3320 | ||
3321 | int BlueStore::ExtentMap::compress_extent_map( | |
3322 | uint64_t offset, | |
3323 | uint64_t length) | |
3324 | { | |
7c673cae FG |
3325 | if (extent_map.empty()) |
3326 | return 0; | |
3327 | int removed = 0; | |
3328 | auto p = seek_lextent(offset); | |
3329 | if (p != extent_map.begin()) { | |
3330 | --p; // start to the left of offset | |
3331 | } | |
3332 | // the caller should have just written to this region | |
11fdf7f2 | 3333 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
3334 | |
3335 | // identify the *next* shard | |
3336 | auto pshard = shards.begin(); | |
3337 | while (pshard != shards.end() && | |
3338 | p->logical_offset >= pshard->shard_info->offset) { | |
3339 | ++pshard; | |
3340 | } | |
3341 | uint64_t shard_end; | |
3342 | if (pshard != shards.end()) { | |
3343 | shard_end = pshard->shard_info->offset; | |
3344 | } else { | |
3345 | shard_end = OBJECT_MAX_SIZE; | |
3346 | } | |
3347 | ||
3348 | auto n = p; | |
3349 | for (++n; n != extent_map.end(); p = n++) { | |
3350 | if (n->logical_offset > offset + length) { | |
3351 | break; // stop after end | |
3352 | } | |
3353 | while (n != extent_map.end() && | |
3354 | p->logical_end() == n->logical_offset && | |
3355 | p->blob == n->blob && | |
3356 | p->blob_offset + p->length == n->blob_offset && | |
3357 | n->logical_offset < shard_end) { | |
3358 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3359 | << " next shard 0x" << shard_end << std::dec | |
3360 | << " merging " << *p << " and " << *n << dendl; | |
3361 | p->length += n->length; | |
3362 | rm(n++); | |
3363 | ++removed; | |
3364 | } | |
3365 | if (n == extent_map.end()) { | |
3366 | break; | |
3367 | } | |
3368 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3369 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3370 | ++pshard; |
3371 | if (pshard != shards.end()) { | |
3372 | shard_end = pshard->shard_info->offset; | |
3373 | } else { | |
3374 | shard_end = OBJECT_MAX_SIZE; | |
3375 | } | |
3376 | } | |
3377 | } | |
11fdf7f2 | 3378 | if (removed) { |
7c673cae FG |
3379 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3380 | } | |
3381 | return removed; | |
3382 | } | |
3383 | ||
3384 | void BlueStore::ExtentMap::punch_hole( | |
3385 | CollectionRef &c, | |
3386 | uint64_t offset, | |
3387 | uint64_t length, | |
3388 | old_extent_map_t *old_extents) | |
3389 | { | |
3390 | auto p = seek_lextent(offset); | |
3391 | uint64_t end = offset + length; | |
3392 | while (p != extent_map.end()) { | |
3393 | if (p->logical_offset >= end) { | |
3394 | break; | |
3395 | } | |
3396 | if (p->logical_offset < offset) { | |
3397 | if (p->logical_end() > end) { | |
3398 | // split and deref middle | |
3399 | uint64_t front = offset - p->logical_offset; | |
3400 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3401 | length, p->blob); | |
3402 | old_extents->push_back(*oe); | |
3403 | add(end, | |
3404 | p->blob_offset + front + length, | |
3405 | p->length - front - length, | |
3406 | p->blob); | |
3407 | p->length = front; | |
3408 | break; | |
3409 | } else { | |
3410 | // deref tail | |
11fdf7f2 | 3411 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3412 | uint64_t keep = offset - p->logical_offset; |
3413 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3414 | p->length - keep, p->blob); | |
3415 | old_extents->push_back(*oe); | |
3416 | p->length = keep; | |
3417 | ++p; | |
3418 | continue; | |
3419 | } | |
3420 | } | |
3421 | if (p->logical_offset + p->length <= end) { | |
3422 | // deref whole lextent | |
3423 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3424 | p->length, p->blob); | |
3425 | old_extents->push_back(*oe); | |
3426 | rm(p++); | |
3427 | continue; | |
3428 | } | |
3429 | // deref head | |
3430 | uint64_t keep = p->logical_end() - end; | |
3431 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3432 | p->length - keep, p->blob); | |
3433 | old_extents->push_back(*oe); | |
3434 | ||
3435 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3436 | rm(p); | |
3437 | break; | |
3438 | } | |
3439 | } | |
3440 | ||
3441 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3442 | CollectionRef &c, | |
3443 | uint64_t logical_offset, | |
3444 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3445 | old_extent_map_t *old_extents) | |
3446 | { | |
3447 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3448 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3449 | |
3450 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3451 | // old_extents list if we overwre the blob totally | |
3452 | // This might happen during WAL overwrite. | |
3453 | b->get_ref(onode->c, blob_offset, length); | |
3454 | ||
3455 | if (old_extents) { | |
3456 | punch_hole(c, logical_offset, length, old_extents); | |
3457 | } | |
3458 | ||
3459 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3460 | extent_map.insert(*le); | |
3461 | if (spans_shard(logical_offset, length)) { | |
3462 | request_reshard(logical_offset, logical_offset + length); | |
3463 | } | |
3464 | return le; | |
3465 | } | |
3466 | ||
3467 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3468 | BlobRef lb, | |
3469 | uint32_t blob_offset, | |
3470 | uint32_t pos) | |
3471 | { | |
7c673cae FG |
3472 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; |
3473 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3474 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3475 | << dendl; | |
3476 | BlobRef rb = onode->c->new_blob(); | |
3477 | lb->split(onode->c, blob_offset, rb.get()); | |
3478 | ||
3479 | for (auto ep = seek_lextent(pos); | |
3480 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3481 | ++ep) { | |
3482 | if (ep->blob != lb) { | |
3483 | continue; | |
3484 | } | |
3485 | if (ep->logical_offset < pos) { | |
3486 | // split extent | |
3487 | size_t left = pos - ep->logical_offset; | |
3488 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3489 | extent_map.insert(*ne); | |
3490 | ep->length = left; | |
3491 | dout(30) << __func__ << " split " << *ep << dendl; | |
3492 | dout(30) << __func__ << " to " << *ne << dendl; | |
3493 | } else { | |
3494 | // switch blob | |
11fdf7f2 | 3495 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3496 | |
3497 | ep->blob = rb; | |
3498 | ep->blob_offset -= blob_offset; | |
3499 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3500 | } | |
3501 | } | |
3502 | return rb; | |
3503 | } | |
3504 | ||
3505 | // Onode | |
3506 | ||
3507 | #undef dout_prefix | |
3508 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3509 | ||
f6b5b4d7 TL |
3510 | // |
3511 | // A tricky thing about Onode's ref counter is that we do an additional | |
3512 | // increment when newly pinned instance is detected. And -1 on unpin. | |
3513 | // This prevents from a conflict with a delete call (when nref == 0). | |
3514 | // The latter might happen while the thread is in unpin() function | |
3515 | // (and e.g. waiting for lock acquisition) since nref is already | |
3516 | // decremented. And another 'putting' thread on the instance will release it. | |
3517 | // | |
3518 | void BlueStore::Onode::get() { | |
3519 | if (++nref == 2) { | |
3520 | c->get_onode_cache()->pin(this, [&]() { | |
3521 | bool was_pinned = pinned; | |
3522 | pinned = nref >= 2; | |
3523 | // additional increment for newly pinned instance | |
3524 | bool r = !was_pinned && pinned; | |
3525 | if (r) { | |
3526 | ++nref; | |
3527 | } | |
3528 | return cached && r; | |
3529 | }); | |
3530 | } | |
3531 | } | |
3532 | void BlueStore::Onode::put() { | |
3533 | if (--nref == 2) { | |
3534 | c->get_onode_cache()->unpin(this, [&]() { | |
3535 | bool was_pinned = pinned; | |
3536 | pinned = pinned && nref > 2; // intentionally use > not >= as we have | |
3537 | // +1 due to pinned state | |
3538 | bool r = was_pinned && !pinned; | |
3539 | // additional decrement for newly unpinned instance | |
3540 | if (r) { | |
3541 | --nref; | |
3542 | } | |
3543 | return cached && r; | |
3544 | }); | |
3545 | } | |
3546 | if (nref == 0) { | |
3547 | delete this; | |
3548 | } | |
3549 | } | |
3550 | ||
eafe8130 TL |
3551 | BlueStore::Onode* BlueStore::Onode::decode( |
3552 | CollectionRef c, | |
3553 | const ghobject_t& oid, | |
3554 | const string& key, | |
3555 | const bufferlist& v) | |
3556 | { | |
3557 | Onode* on = new Onode(c.get(), oid, key); | |
3558 | on->exists = true; | |
3559 | auto p = v.front().begin_deep(); | |
3560 | on->onode.decode(p); | |
3561 | for (auto& i : on->onode.attrs) { | |
f91f0fd5 | 3562 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
eafe8130 TL |
3563 | } |
3564 | ||
3565 | // initialize extent_map | |
3566 | on->extent_map.decode_spanning_blobs(p); | |
3567 | if (on->onode.extent_map_shards.empty()) { | |
3568 | denc(on->extent_map.inline_bl, p); | |
3569 | on->extent_map.decode_some(on->extent_map.inline_bl); | |
3570 | on->extent_map.inline_bl.reassign_to_mempool( | |
f91f0fd5 | 3571 | mempool::mempool_bluestore_cache_data); |
eafe8130 TL |
3572 | } |
3573 | else { | |
3574 | on->extent_map.init_shards(false, false); | |
3575 | } | |
3576 | return on; | |
3577 | } | |
3578 | ||
7c673cae FG |
3579 | void BlueStore::Onode::flush() |
3580 | { | |
3581 | if (flushing_count.load()) { | |
3582 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
9f95a23c | 3583 | waiting_count++; |
11fdf7f2 | 3584 | std::unique_lock l(flush_lock); |
7c673cae FG |
3585 | while (flushing_count.load()) { |
3586 | flush_cond.wait(l); | |
3587 | } | |
9f95a23c | 3588 | waiting_count--; |
7c673cae FG |
3589 | } |
3590 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3591 | } | |
3592 | ||
9f95a23c TL |
3593 | void BlueStore::Onode::dump(Formatter* f) const |
3594 | { | |
3595 | onode.dump(f); | |
3596 | extent_map.dump(f); | |
3597 | } | |
3598 | ||
3599 | ||
3600 | const string& BlueStore::Onode::get_omap_prefix() | |
3601 | { | |
3602 | if (onode.is_pgmeta_omap()) { | |
3603 | return PREFIX_PGMETA_OMAP; | |
3604 | } | |
3605 | if (onode.is_perpool_omap()) { | |
3606 | return PREFIX_PERPOOL_OMAP; | |
3607 | } | |
3608 | return PREFIX_OMAP; | |
3609 | } | |
3610 | ||
3611 | // '-' < '.' < '~' | |
3612 | ||
3613 | void BlueStore::Onode::get_omap_header(string *out) | |
3614 | { | |
3615 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3616 | _key_encode_u64(c->pool(), out); | |
3617 | } | |
3618 | _key_encode_u64(onode.nid, out); | |
3619 | out->push_back('-'); | |
3620 | } | |
3621 | ||
3622 | void BlueStore::Onode::get_omap_key(const string& key, string *out) | |
3623 | { | |
3624 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3625 | _key_encode_u64(c->pool(), out); | |
3626 | } | |
3627 | _key_encode_u64(onode.nid, out); | |
3628 | out->push_back('.'); | |
3629 | out->append(key); | |
3630 | } | |
3631 | ||
3632 | void BlueStore::Onode::rewrite_omap_key(const string& old, string *out) | |
3633 | { | |
3634 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3635 | _key_encode_u64(c->pool(), out); | |
3636 | } | |
3637 | _key_encode_u64(onode.nid, out); | |
3638 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
3639 | } | |
3640 | ||
3641 | void BlueStore::Onode::get_omap_tail(string *out) | |
3642 | { | |
3643 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3644 | _key_encode_u64(c->pool(), out); | |
3645 | } | |
3646 | _key_encode_u64(onode.nid, out); | |
3647 | out->push_back('~'); | |
3648 | } | |
3649 | ||
3650 | void BlueStore::Onode::decode_omap_key(const string& key, string *user_key) | |
3651 | { | |
3652 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3653 | *user_key = key.substr(sizeof(uint64_t)*2 + 1); | |
3654 | } else { | |
3655 | *user_key = key.substr(sizeof(uint64_t) + 1); | |
3656 | } | |
3657 | } | |
3658 | ||
3659 | ||
7c673cae FG |
3660 | // ======================================================= |
3661 | // WriteContext | |
3662 | ||
3663 | /// Checks for writes to the same pextent within a blob | |
3664 | bool BlueStore::WriteContext::has_conflict( | |
3665 | BlobRef b, | |
3666 | uint64_t loffs, | |
3667 | uint64_t loffs_end, | |
3668 | uint64_t min_alloc_size) | |
3669 | { | |
11fdf7f2 TL |
3670 | ceph_assert((loffs % min_alloc_size) == 0); |
3671 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3672 | for (auto w : writes) { |
3673 | if (b == w.b) { | |
11fdf7f2 TL |
3674 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3675 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3676 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3677 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3678 | return true; |
3679 | } | |
3680 | } | |
3681 | } | |
3682 | return false; | |
3683 | } | |
3684 | ||
3685 | // ======================================================= | |
3686 | ||
3687 | // DeferredBatch | |
3688 | #undef dout_prefix | |
3689 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
9f95a23c TL |
3690 | #undef dout_context |
3691 | #define dout_context cct | |
7c673cae FG |
3692 | |
3693 | void BlueStore::DeferredBatch::prepare_write( | |
3694 | CephContext *cct, | |
3695 | uint64_t seq, uint64_t offset, uint64_t length, | |
3696 | bufferlist::const_iterator& blp) | |
3697 | { | |
3698 | _discard(cct, offset, length); | |
3699 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3700 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3701 | i.first->second.seq = seq; |
3702 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3703 | i.first->second.bl.reassign_to_mempool( |
3704 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3705 | dout(20) << __func__ << " seq " << seq |
3706 | << " 0x" << std::hex << offset << "~" << length | |
3707 | << " crc " << i.first->second.bl.crc32c(-1) | |
3708 | << std::dec << dendl; | |
3709 | seq_bytes[seq] += length; | |
3710 | #ifdef DEBUG_DEFERRED | |
3711 | _audit(cct); | |
3712 | #endif | |
3713 | } | |
3714 | ||
3715 | void BlueStore::DeferredBatch::_discard( | |
3716 | CephContext *cct, uint64_t offset, uint64_t length) | |
3717 | { | |
3718 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3719 | << std::dec << dendl; | |
3720 | auto p = iomap.lower_bound(offset); | |
3721 | if (p != iomap.begin()) { | |
3722 | --p; | |
3723 | auto end = p->first + p->second.bl.length(); | |
3724 | if (end > offset) { | |
3725 | bufferlist head; | |
3726 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3727 | dout(20) << __func__ << " keep head " << p->second.seq | |
3728 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3729 | << " -> 0x" << head.length() << std::dec << dendl; | |
3730 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3731 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3732 | if (end > offset + length) { |
3733 | bufferlist tail; | |
3734 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3735 | end - (offset + length)); | |
3736 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3737 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3738 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3739 | auto &n = iomap[offset + length]; | |
3740 | n.bl.swap(tail); | |
3741 | n.seq = p->second.seq; | |
3742 | i->second -= length; | |
3743 | } else { | |
3744 | i->second -= end - offset; | |
3745 | } | |
11fdf7f2 | 3746 | ceph_assert(i->second >= 0); |
7c673cae FG |
3747 | p->second.bl.swap(head); |
3748 | } | |
3749 | ++p; | |
3750 | } | |
3751 | while (p != iomap.end()) { | |
3752 | if (p->first >= offset + length) { | |
3753 | break; | |
3754 | } | |
3755 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3756 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3757 | auto end = p->first + p->second.bl.length(); |
3758 | if (end > offset + length) { | |
3759 | unsigned drop_front = offset + length - p->first; | |
3760 | unsigned keep_tail = end - (offset + length); | |
3761 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3762 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3763 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3764 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3765 | << std::dec << dendl; | |
3766 | auto &s = iomap[offset + length]; | |
3767 | s.seq = p->second.seq; | |
3768 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3769 | i->second -= drop_front; | |
3770 | } else { | |
3771 | dout(20) << __func__ << " drop " << p->second.seq | |
3772 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3773 | << std::dec << dendl; | |
3774 | i->second -= p->second.bl.length(); | |
3775 | } | |
11fdf7f2 | 3776 | ceph_assert(i->second >= 0); |
7c673cae FG |
3777 | p = iomap.erase(p); |
3778 | } | |
3779 | } | |
3780 | ||
3781 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3782 | { | |
3783 | map<uint64_t,int> sb; | |
3784 | for (auto p : seq_bytes) { | |
3785 | sb[p.first] = 0; // make sure we have the same set of keys | |
3786 | } | |
3787 | uint64_t pos = 0; | |
3788 | for (auto& p : iomap) { | |
11fdf7f2 | 3789 | ceph_assert(p.first >= pos); |
7c673cae FG |
3790 | sb[p.second.seq] += p.second.bl.length(); |
3791 | pos = p.first + p.second.bl.length(); | |
3792 | } | |
11fdf7f2 | 3793 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3794 | } |
3795 | ||
3796 | ||
3797 | // Collection | |
3798 | ||
3799 | #undef dout_prefix | |
3800 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3801 | ||
9f95a23c TL |
3802 | BlueStore::Collection::Collection(BlueStore *store_, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t cid) |
3803 | : CollectionImpl(store_->cct, cid), | |
11fdf7f2 | 3804 | store(store_), |
9f95a23c | 3805 | cache(bc), |
7c673cae | 3806 | exists(true), |
9f95a23c | 3807 | onode_map(oc), |
11fdf7f2 TL |
3808 | commit_queue(nullptr) |
3809 | { | |
3810 | } | |
3811 | ||
3812 | bool BlueStore::Collection::flush_commit(Context *c) | |
3813 | { | |
3814 | return osr->flush_commit(c); | |
3815 | } | |
3816 | ||
3817 | void BlueStore::Collection::flush() | |
3818 | { | |
3819 | osr->flush(); | |
3820 | } | |
3821 | ||
3822 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3823 | { |
11fdf7f2 | 3824 | osr->flush_all_but_last(); |
7c673cae FG |
3825 | } |
3826 | ||
3827 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3828 | { | |
11fdf7f2 | 3829 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3830 | const bluestore_blob_t& blob = b->get_blob(); |
3831 | if (!blob.is_shared()) { | |
3832 | b->shared_blob = new SharedBlob(this); | |
3833 | return; | |
3834 | } | |
3835 | ||
3836 | b->shared_blob = shared_blob_set.lookup(sbid); | |
3837 | if (b->shared_blob) { | |
3838 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3839 | << std::dec << " had " << *b->shared_blob << dendl; | |
3840 | } else { | |
3841 | b->shared_blob = new SharedBlob(sbid, this); | |
3842 | shared_blob_set.add(this, b->shared_blob.get()); | |
3843 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3844 | << std::dec << " opened " << *b->shared_blob | |
3845 | << dendl; | |
3846 | } | |
3847 | } | |
3848 | ||
3849 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
3850 | { | |
3851 | if (!sb->is_loaded()) { | |
3852 | ||
3853 | bufferlist v; | |
3854 | string key; | |
3855 | auto sbid = sb->get_sbid(); | |
3856 | get_shared_blob_key(sbid, &key); | |
3857 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
3858 | if (r < 0) { | |
3859 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
3860 | << std::dec << " not found at key " | |
3861 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 3862 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
3863 | } |
3864 | ||
3865 | sb->loaded = true; | |
3866 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
3867 | auto p = v.cbegin(); |
3868 | decode(*(sb->persistent), p); | |
7c673cae FG |
3869 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
3870 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
3871 | } | |
3872 | } | |
3873 | ||
3874 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
3875 | { | |
7c673cae | 3876 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 3877 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
3878 | |
3879 | // update blob | |
31f18b77 | 3880 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 3881 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
3882 | |
3883 | // update shared blob | |
3884 | b->shared_blob->loaded = true; | |
3885 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
3886 | shared_blob_set.add(this, b->shared_blob.get()); | |
3887 | for (auto p : blob.get_extents()) { | |
3888 | if (p.is_valid()) { | |
3889 | b->shared_blob->get_ref( | |
3890 | p.offset, | |
3891 | p.length); | |
3892 | } | |
3893 | } | |
3894 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
3895 | } | |
3896 | ||
31f18b77 FG |
3897 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
3898 | { | |
3899 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 3900 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
3901 | |
3902 | uint64_t sbid = sb->get_sbid(); | |
3903 | shared_blob_set.remove(sb); | |
3904 | sb->loaded = false; | |
3905 | delete sb->persistent; | |
3906 | sb->sbid_unloaded = 0; | |
3907 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
3908 | return sbid; | |
3909 | } | |
3910 | ||
7c673cae FG |
3911 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
3912 | const ghobject_t& oid, | |
9f95a23c TL |
3913 | bool create, |
3914 | bool is_createop) | |
7c673cae | 3915 | { |
9f95a23c | 3916 | ceph_assert(create ? ceph_mutex_is_wlocked(lock) : ceph_mutex_is_locked(lock)); |
7c673cae FG |
3917 | |
3918 | spg_t pgid; | |
3919 | if (cid.is_pg(&pgid)) { | |
3920 | if (!oid.match(cnode.bits, pgid.ps())) { | |
3921 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
3922 | << pgid << " bits " << cnode.bits << dendl; | |
3923 | ceph_abort(); | |
3924 | } | |
3925 | } | |
3926 | ||
3927 | OnodeRef o = onode_map.lookup(oid); | |
3928 | if (o) | |
3929 | return o; | |
3930 | ||
eafe8130 | 3931 | string key; |
7c673cae FG |
3932 | get_object_key(store->cct, oid, &key); |
3933 | ||
3934 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
3935 | << pretty_binary_string(key) << dendl; | |
3936 | ||
3937 | bufferlist v; | |
9f95a23c | 3938 | int r = -ENOENT; |
7c673cae | 3939 | Onode *on; |
9f95a23c TL |
3940 | if (!is_createop) { |
3941 | r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
3942 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
3943 | } | |
7c673cae | 3944 | if (v.length() == 0) { |
11fdf7f2 | 3945 | ceph_assert(r == -ENOENT); |
7c673cae FG |
3946 | if (!store->cct->_conf->bluestore_debug_misc && |
3947 | !create) | |
3948 | return OnodeRef(); | |
3949 | ||
3950 | // new object, new onode | |
3951 | on = new Onode(this, oid, key); | |
3952 | } else { | |
3953 | // loaded | |
11fdf7f2 | 3954 | ceph_assert(r >= 0); |
eafe8130 | 3955 | on = Onode::decode(this, oid, key, v); |
7c673cae FG |
3956 | } |
3957 | o.reset(on); | |
3958 | return onode_map.add(oid, o); | |
3959 | } | |
3960 | ||
3961 | void BlueStore::Collection::split_cache( | |
3962 | Collection *dest) | |
3963 | { | |
3964 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
3965 | ||
3966 | // lock (one or both) cache shards | |
3967 | std::lock(cache->lock, dest->cache->lock); | |
11fdf7f2 TL |
3968 | std::lock_guard l(cache->lock, std::adopt_lock); |
3969 | std::lock_guard l2(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
3970 | |
3971 | int destbits = dest->cnode.bits; | |
3972 | spg_t destpg; | |
3973 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 3974 | ceph_assert(is_pg); |
7c673cae FG |
3975 | |
3976 | auto p = onode_map.onode_map.begin(); | |
3977 | while (p != onode_map.onode_map.end()) { | |
11fdf7f2 | 3978 | OnodeRef o = p->second; |
7c673cae FG |
3979 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
3980 | // onode does not belong to this child | |
11fdf7f2 TL |
3981 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
3982 | << dendl; | |
7c673cae FG |
3983 | ++p; |
3984 | } else { | |
7c673cae FG |
3985 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
3986 | << dendl; | |
3987 | ||
f6b5b4d7 TL |
3988 | // ensuring that nref is always >= 2 and hence onode is pinned and |
3989 | // physically out of cache during the transition | |
3990 | OnodeRef o_pin = o; | |
3991 | ceph_assert(o->pinned); | |
3992 | ||
7c673cae | 3993 | p = onode_map.onode_map.erase(p); |
7c673cae | 3994 | dest->onode_map.onode_map[o->oid] = o; |
f6b5b4d7 TL |
3995 | if (get_onode_cache() != dest->get_onode_cache()) { |
3996 | get_onode_cache()->move_pinned(dest->get_onode_cache(), o.get()); | |
9f95a23c | 3997 | } |
f6b5b4d7 | 3998 | o->c = dest; |
7c673cae FG |
3999 | |
4000 | // move over shared blobs and buffers. cover shared blobs from | |
4001 | // both extent map and spanning blob map (the full extent map | |
4002 | // may not be faulted in) | |
4003 | vector<SharedBlob*> sbvec; | |
4004 | for (auto& e : o->extent_map.extent_map) { | |
4005 | sbvec.push_back(e.blob->shared_blob.get()); | |
4006 | } | |
4007 | for (auto& b : o->extent_map.spanning_blob_map) { | |
4008 | sbvec.push_back(b.second->shared_blob.get()); | |
4009 | } | |
4010 | for (auto sb : sbvec) { | |
4011 | if (sb->coll == dest) { | |
4012 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
4013 | << dendl; | |
4014 | continue; | |
4015 | } | |
4016 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
4017 | if (sb->get_sbid()) { |
4018 | ldout(store->cct, 20) << __func__ | |
4019 | << " moving registration " << *sb << dendl; | |
4020 | shared_blob_set.remove(sb); | |
4021 | dest->shared_blob_set.add(dest, sb); | |
4022 | } | |
3efd9988 | 4023 | sb->coll = dest; |
7c673cae | 4024 | if (dest->cache != cache) { |
7c673cae FG |
4025 | for (auto& i : sb->bc.buffer_map) { |
4026 | if (!i.second->is_writing()) { | |
4027 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
4028 | << dendl; | |
9f95a23c | 4029 | dest->cache->_move(cache, i.second.get()); |
7c673cae FG |
4030 | } |
4031 | } | |
4032 | } | |
4033 | } | |
7c673cae FG |
4034 | } |
4035 | } | |
9f95a23c | 4036 | dest->cache->_trim(); |
7c673cae FG |
4037 | } |
4038 | ||
7c673cae FG |
4039 | // ======================================================= |
4040 | ||
91327a77 AA |
4041 | // MempoolThread |
4042 | ||
4043 | #undef dout_prefix | |
4044 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
9f95a23c TL |
4045 | #undef dout_context |
4046 | #define dout_context store->cct | |
91327a77 | 4047 | |
7c673cae FG |
4048 | void *BlueStore::MempoolThread::entry() |
4049 | { | |
9f95a23c | 4050 | std::unique_lock l{lock}; |
11fdf7f2 | 4051 | |
92f5a8d4 | 4052 | uint32_t prev_config_change = store->config_changed.load(); |
eafe8130 TL |
4053 | uint64_t base = store->osd_memory_base; |
4054 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4055 | uint64_t target = store->osd_memory_target; | |
4056 | uint64_t min = store->osd_memory_cache_min; | |
4057 | uint64_t max = min; | |
4058 | ||
4059 | // When setting the maximum amount of memory to use for cache, first | |
4060 | // assume some base amount of memory for the OSD and then fudge in | |
4061 | // some overhead for fragmentation that scales with cache usage. | |
4062 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4063 | if (ltarget > base + min) { | |
4064 | max = ltarget - base; | |
11fdf7f2 | 4065 | } |
31f18b77 | 4066 | |
eafe8130 TL |
4067 | binned_kv_cache = store->db->get_priority_cache(); |
4068 | if (store->cache_autotune && binned_kv_cache != nullptr) { | |
4069 | pcm = std::make_shared<PriorityCache::Manager>( | |
4070 | store->cct, min, max, target, true); | |
4071 | pcm->insert("kv", binned_kv_cache, true); | |
4072 | pcm->insert("meta", meta_cache, true); | |
4073 | pcm->insert("data", data_cache, true); | |
4074 | } | |
91327a77 AA |
4075 | |
4076 | utime_t next_balance = ceph_clock_now(); | |
4077 | utime_t next_resize = ceph_clock_now(); | |
9f95a23c TL |
4078 | utime_t next_deferred_force_submit = ceph_clock_now(); |
4079 | utime_t alloc_stats_dump_clock = ceph_clock_now(); | |
31f18b77 | 4080 | |
91327a77 | 4081 | bool interval_stats_trim = false; |
91327a77 | 4082 | while (!stop) { |
92f5a8d4 TL |
4083 | // Update pcm cache settings if related configuration was changed |
4084 | uint32_t cur_config_change = store->config_changed.load(); | |
4085 | if (cur_config_change != prev_config_change) { | |
4086 | _update_cache_settings(); | |
4087 | prev_config_change = cur_config_change; | |
4088 | } | |
4089 | ||
91327a77 AA |
4090 | // Before we trim, check and see if it's time to rebalance/resize. |
4091 | double autotune_interval = store->cache_autotune_interval; | |
4092 | double resize_interval = store->osd_memory_cache_resize_interval; | |
9f95a23c TL |
4093 | double max_defer_interval = store->max_defer_interval; |
4094 | ||
4095 | double alloc_stats_dump_interval = | |
4096 | store->cct->_conf->bluestore_alloc_stats_dump_interval; | |
91327a77 | 4097 | |
9f95a23c TL |
4098 | if (alloc_stats_dump_interval > 0 && |
4099 | alloc_stats_dump_clock + alloc_stats_dump_interval < ceph_clock_now()) { | |
4100 | store->_record_allocation_stats(); | |
4101 | alloc_stats_dump_clock = ceph_clock_now(); | |
4102 | } | |
91327a77 | 4103 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { |
11fdf7f2 TL |
4104 | _adjust_cache_settings(); |
4105 | ||
91327a77 | 4106 | // Log events at 5 instead of 20 when balance happens. |
91327a77 | 4107 | interval_stats_trim = true; |
eafe8130 TL |
4108 | |
4109 | if (pcm != nullptr) { | |
4110 | pcm->balance(); | |
91327a77 | 4111 | } |
31f18b77 | 4112 | |
91327a77 AA |
4113 | next_balance = ceph_clock_now(); |
4114 | next_balance += autotune_interval; | |
4115 | } | |
4116 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { | |
eafe8130 TL |
4117 | if (ceph_using_tcmalloc() && pcm != nullptr) { |
4118 | pcm->tune_memory(); | |
91327a77 AA |
4119 | } |
4120 | next_resize = ceph_clock_now(); | |
4121 | next_resize += resize_interval; | |
31f18b77 FG |
4122 | } |
4123 | ||
9f95a23c TL |
4124 | if (max_defer_interval > 0 && |
4125 | next_deferred_force_submit < ceph_clock_now()) { | |
4126 | if (store->get_deferred_last_submitted() + max_defer_interval < | |
4127 | ceph_clock_now()) { | |
4128 | store->deferred_try_submit(); | |
4129 | } | |
4130 | next_deferred_force_submit = ceph_clock_now(); | |
4131 | next_deferred_force_submit += max_defer_interval/3; | |
4132 | } | |
4133 | ||
4134 | // Now Resize the shards | |
4135 | _resize_shards(interval_stats_trim); | |
91327a77 | 4136 | interval_stats_trim = false; |
31f18b77 | 4137 | |
91327a77 | 4138 | store->_update_cache_logger(); |
11fdf7f2 TL |
4139 | auto wait = ceph::make_timespan( |
4140 | store->cct->_conf->bluestore_cache_trim_interval); | |
4141 | cond.wait_for(l, wait); | |
7c673cae | 4142 | } |
9f95a23c TL |
4143 | // do final dump |
4144 | store->_record_allocation_stats(); | |
7c673cae FG |
4145 | stop = false; |
4146 | return NULL; | |
4147 | } | |
4148 | ||
91327a77 AA |
4149 | void BlueStore::MempoolThread::_adjust_cache_settings() |
4150 | { | |
11fdf7f2 TL |
4151 | if (binned_kv_cache != nullptr) { |
4152 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
4153 | } | |
4154 | meta_cache->set_cache_ratio(store->cache_meta_ratio); | |
4155 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
91327a77 AA |
4156 | } |
4157 | ||
9f95a23c | 4158 | void BlueStore::MempoolThread::_resize_shards(bool interval_stats) |
91327a77 | 4159 | { |
9f95a23c TL |
4160 | size_t onode_shards = store->onode_cache_shards.size(); |
4161 | size_t buffer_shards = store->buffer_cache_shards.size(); | |
91327a77 | 4162 | int64_t kv_used = store->db->get_cache_usage(); |
11fdf7f2 TL |
4163 | int64_t meta_used = meta_cache->_get_used_bytes(); |
4164 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
4165 | |
4166 | uint64_t cache_size = store->cache_size; | |
4167 | int64_t kv_alloc = | |
11fdf7f2 | 4168 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
91327a77 | 4169 | int64_t meta_alloc = |
11fdf7f2 | 4170 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 4171 | int64_t data_alloc = |
11fdf7f2 | 4172 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 4173 | |
eafe8130 TL |
4174 | if (pcm != nullptr && binned_kv_cache != nullptr) { |
4175 | cache_size = pcm->get_tuned_mem(); | |
11fdf7f2 TL |
4176 | kv_alloc = binned_kv_cache->get_committed_size(); |
4177 | meta_alloc = meta_cache->get_committed_size(); | |
4178 | data_alloc = data_cache->get_committed_size(); | |
91327a77 AA |
4179 | } |
4180 | ||
4181 | if (interval_stats) { | |
9f95a23c | 4182 | dout(5) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4183 | << " kv_alloc: " << kv_alloc |
4184 | << " kv_used: " << kv_used | |
4185 | << " meta_alloc: " << meta_alloc | |
4186 | << " meta_used: " << meta_used | |
4187 | << " data_alloc: " << data_alloc | |
4188 | << " data_used: " << data_used << dendl; | |
4189 | } else { | |
9f95a23c | 4190 | dout(20) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
4191 | << " kv_alloc: " << kv_alloc |
4192 | << " kv_used: " << kv_used | |
4193 | << " meta_alloc: " << meta_alloc | |
4194 | << " meta_used: " << meta_used | |
4195 | << " data_alloc: " << data_alloc | |
4196 | << " data_used: " << data_used << dendl; | |
4197 | } | |
4198 | ||
4199 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
9f95a23c TL |
4200 | (meta_alloc / (double) onode_shards) / meta_cache->get_bytes_per_onode()); |
4201 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / buffer_shards); | |
91327a77 | 4202 | |
9f95a23c | 4203 | dout(30) << __func__ << " max_shard_onodes: " << max_shard_onodes |
91327a77 AA |
4204 | << " max_shard_buffer: " << max_shard_buffer << dendl; |
4205 | ||
9f95a23c TL |
4206 | for (auto i : store->onode_cache_shards) { |
4207 | i->set_max(max_shard_onodes); | |
4208 | } | |
4209 | for (auto i : store->buffer_cache_shards) { | |
4210 | i->set_max(max_shard_buffer); | |
91327a77 AA |
4211 | } |
4212 | } | |
4213 | ||
92f5a8d4 TL |
4214 | void BlueStore::MempoolThread::_update_cache_settings() |
4215 | { | |
4216 | // Nothing to do if pcm is not used. | |
4217 | if (pcm == nullptr) { | |
4218 | return; | |
4219 | } | |
4220 | ||
92f5a8d4 TL |
4221 | uint64_t target = store->osd_memory_target; |
4222 | uint64_t base = store->osd_memory_base; | |
4223 | uint64_t min = store->osd_memory_cache_min; | |
4224 | uint64_t max = min; | |
4225 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4226 | ||
4227 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4228 | if (ltarget > base + min) { | |
4229 | max = ltarget - base; | |
4230 | } | |
4231 | ||
4232 | // set pcm cache levels | |
4233 | pcm->set_target_memory(target); | |
4234 | pcm->set_min_memory(min); | |
4235 | pcm->set_max_memory(max); | |
4236 | ||
9f95a23c | 4237 | dout(5) << __func__ << " updated pcm target: " << target |
92f5a8d4 TL |
4238 | << " pcm min: " << min |
4239 | << " pcm max: " << max | |
4240 | << dendl; | |
4241 | } | |
4242 | ||
7c673cae FG |
4243 | // ======================================================= |
4244 | ||
31f18b77 FG |
4245 | // OmapIteratorImpl |
4246 | ||
4247 | #undef dout_prefix | |
4248 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
4249 | ||
4250 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
4251 | CollectionRef c, OnodeRef o, KeyValueDB::Iterator it) | |
4252 | : c(c), o(o), it(it) | |
4253 | { | |
9f95a23c | 4254 | std::shared_lock l(c->lock); |
31f18b77 | 4255 | if (o->onode.has_omap()) { |
9f95a23c TL |
4256 | o->get_omap_key(string(), &head); |
4257 | o->get_omap_tail(&tail); | |
31f18b77 FG |
4258 | it->lower_bound(head); |
4259 | } | |
4260 | } | |
4261 | ||
11fdf7f2 TL |
4262 | string BlueStore::OmapIteratorImpl::_stringify() const |
4263 | { | |
4264 | stringstream s; | |
4265 | s << " omap_iterator(cid = " << c->cid | |
4266 | <<", oid = " << o->oid << ")"; | |
4267 | return s.str(); | |
4268 | } | |
4269 | ||
31f18b77 FG |
4270 | int BlueStore::OmapIteratorImpl::seek_to_first() |
4271 | { | |
9f95a23c | 4272 | std::shared_lock l(c->lock); |
11fdf7f2 | 4273 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4274 | if (o->onode.has_omap()) { |
4275 | it->lower_bound(head); | |
4276 | } else { | |
4277 | it = KeyValueDB::Iterator(); | |
4278 | } | |
494da23a TL |
4279 | c->store->log_latency( |
4280 | __func__, | |
11fdf7f2 TL |
4281 | l_bluestore_omap_seek_to_first_lat, |
4282 | mono_clock::now() - start1, | |
494da23a | 4283 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 | 4284 | |
31f18b77 FG |
4285 | return 0; |
4286 | } | |
4287 | ||
4288 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
4289 | { | |
9f95a23c | 4290 | std::shared_lock l(c->lock); |
11fdf7f2 | 4291 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4292 | if (o->onode.has_omap()) { |
4293 | string key; | |
9f95a23c | 4294 | o->get_omap_key(after, &key); |
31f18b77 FG |
4295 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " |
4296 | << pretty_binary_string(key) << dendl; | |
4297 | it->upper_bound(key); | |
4298 | } else { | |
4299 | it = KeyValueDB::Iterator(); | |
4300 | } | |
11fdf7f2 | 4301 | c->store->log_latency_fn( |
494da23a | 4302 | __func__, |
11fdf7f2 TL |
4303 | l_bluestore_omap_upper_bound_lat, |
4304 | mono_clock::now() - start1, | |
494da23a | 4305 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4306 | [&] (const ceph::timespan& lat) { |
494da23a | 4307 | return ", after = " + after + |
11fdf7f2 TL |
4308 | _stringify(); |
4309 | } | |
4310 | ); | |
31f18b77 FG |
4311 | return 0; |
4312 | } | |
4313 | ||
4314 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
4315 | { | |
9f95a23c | 4316 | std::shared_lock l(c->lock); |
11fdf7f2 | 4317 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4318 | if (o->onode.has_omap()) { |
4319 | string key; | |
9f95a23c | 4320 | o->get_omap_key(to, &key); |
31f18b77 FG |
4321 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " |
4322 | << pretty_binary_string(key) << dendl; | |
4323 | it->lower_bound(key); | |
4324 | } else { | |
4325 | it = KeyValueDB::Iterator(); | |
4326 | } | |
11fdf7f2 | 4327 | c->store->log_latency_fn( |
494da23a | 4328 | __func__, |
11fdf7f2 TL |
4329 | l_bluestore_omap_lower_bound_lat, |
4330 | mono_clock::now() - start1, | |
494da23a | 4331 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4332 | [&] (const ceph::timespan& lat) { |
494da23a | 4333 | return ", to = " + to + |
11fdf7f2 TL |
4334 | _stringify(); |
4335 | } | |
4336 | ); | |
31f18b77 FG |
4337 | return 0; |
4338 | } | |
4339 | ||
4340 | bool BlueStore::OmapIteratorImpl::valid() | |
4341 | { | |
9f95a23c | 4342 | std::shared_lock l(c->lock); |
31f18b77 | 4343 | bool r = o->onode.has_omap() && it && it->valid() && |
494da23a | 4344 | it->raw_key().second < tail; |
31f18b77 FG |
4345 | if (it && it->valid()) { |
4346 | ldout(c->store->cct,20) << __func__ << " is at " | |
4347 | << pretty_binary_string(it->raw_key().second) | |
4348 | << dendl; | |
4349 | } | |
4350 | return r; | |
4351 | } | |
4352 | ||
11fdf7f2 | 4353 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 4354 | { |
11fdf7f2 | 4355 | int r = -1; |
9f95a23c | 4356 | std::shared_lock l(c->lock); |
11fdf7f2 | 4357 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4358 | if (o->onode.has_omap()) { |
4359 | it->next(); | |
11fdf7f2 | 4360 | r = 0; |
31f18b77 | 4361 | } |
494da23a TL |
4362 | c->store->log_latency( |
4363 | __func__, | |
11fdf7f2 TL |
4364 | l_bluestore_omap_next_lat, |
4365 | mono_clock::now() - start1, | |
494da23a | 4366 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 TL |
4367 | |
4368 | return r; | |
31f18b77 FG |
4369 | } |
4370 | ||
4371 | string BlueStore::OmapIteratorImpl::key() | |
4372 | { | |
9f95a23c | 4373 | std::shared_lock l(c->lock); |
11fdf7f2 | 4374 | ceph_assert(it->valid()); |
31f18b77 FG |
4375 | string db_key = it->raw_key().second; |
4376 | string user_key; | |
9f95a23c | 4377 | o->decode_omap_key(db_key, &user_key); |
494da23a | 4378 | |
31f18b77 FG |
4379 | return user_key; |
4380 | } | |
4381 | ||
4382 | bufferlist BlueStore::OmapIteratorImpl::value() | |
4383 | { | |
9f95a23c | 4384 | std::shared_lock l(c->lock); |
11fdf7f2 | 4385 | ceph_assert(it->valid()); |
31f18b77 FG |
4386 | return it->value(); |
4387 | } | |
4388 | ||
4389 | ||
4390 | // ===================================== | |
4391 | ||
7c673cae FG |
4392 | #undef dout_prefix |
4393 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
9f95a23c TL |
4394 | #undef dout_context |
4395 | #define dout_context cct | |
7c673cae FG |
4396 | |
4397 | ||
4398 | static void aio_cb(void *priv, void *priv2) | |
4399 | { | |
4400 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4401 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4402 | c->aio_finish(store); | |
4403 | } | |
4404 | ||
11fdf7f2 TL |
4405 | static void discard_cb(void *priv, void *priv2) |
4406 | { | |
4407 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4408 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4409 | store->handle_discard(*tmp); | |
4410 | } | |
4411 | ||
4412 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4413 | { | |
4414 | dout(10) << __func__ << dendl; | |
4415 | ceph_assert(alloc); | |
4416 | alloc->release(to_release); | |
4417 | } | |
4418 | ||
7c673cae | 4419 | BlueStore::BlueStore(CephContext *cct, const string& path) |
9f95a23c | 4420 | : BlueStore(cct, path, 0) {} |
7c673cae FG |
4421 | |
4422 | BlueStore::BlueStore(CephContext *cct, | |
4423 | const string& path, | |
4424 | uint64_t _min_alloc_size) | |
4425 | : ObjectStore(cct, path), | |
9f95a23c | 4426 | throttle(cct), |
11fdf7f2 | 4427 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4428 | kv_sync_thread(this), |
31f18b77 | 4429 | kv_finalize_thread(this), |
7c673cae FG |
4430 | min_alloc_size(_min_alloc_size), |
4431 | min_alloc_size_order(ctz(_min_alloc_size)), | |
4432 | mempool_thread(this) | |
4433 | { | |
4434 | _init_logger(); | |
11fdf7f2 | 4435 | cct->_conf.add_observer(this); |
7c673cae | 4436 | set_cache_shards(1); |
7c673cae FG |
4437 | } |
4438 | ||
4439 | BlueStore::~BlueStore() | |
4440 | { | |
11fdf7f2 | 4441 | cct->_conf.remove_observer(this); |
7c673cae | 4442 | _shutdown_logger(); |
11fdf7f2 TL |
4443 | ceph_assert(!mounted); |
4444 | ceph_assert(db == NULL); | |
4445 | ceph_assert(bluefs == NULL); | |
4446 | ceph_assert(fsid_fd < 0); | |
4447 | ceph_assert(path_fd < 0); | |
9f95a23c TL |
4448 | for (auto i : onode_cache_shards) { |
4449 | delete i; | |
4450 | } | |
4451 | for (auto i : buffer_cache_shards) { | |
7c673cae FG |
4452 | delete i; |
4453 | } | |
9f95a23c TL |
4454 | onode_cache_shards.clear(); |
4455 | buffer_cache_shards.clear(); | |
7c673cae FG |
4456 | } |
4457 | ||
4458 | const char **BlueStore::get_tracked_conf_keys() const | |
4459 | { | |
4460 | static const char* KEYS[] = { | |
4461 | "bluestore_csum_type", | |
4462 | "bluestore_compression_mode", | |
4463 | "bluestore_compression_algorithm", | |
4464 | "bluestore_compression_min_blob_size", | |
4465 | "bluestore_compression_min_blob_size_ssd", | |
4466 | "bluestore_compression_min_blob_size_hdd", | |
4467 | "bluestore_compression_max_blob_size", | |
4468 | "bluestore_compression_max_blob_size_ssd", | |
4469 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4470 | "bluestore_compression_required_ratio", |
7c673cae FG |
4471 | "bluestore_max_alloc_size", |
4472 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4473 | "bluestore_prefer_deferred_size_hdd", |
4474 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4475 | "bluestore_deferred_batch_ops", |
4476 | "bluestore_deferred_batch_ops_hdd", | |
4477 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4478 | "bluestore_throttle_bytes", |
4479 | "bluestore_throttle_deferred_bytes", | |
4480 | "bluestore_throttle_cost_per_io_hdd", | |
4481 | "bluestore_throttle_cost_per_io_ssd", | |
4482 | "bluestore_throttle_cost_per_io", | |
4483 | "bluestore_max_blob_size", | |
4484 | "bluestore_max_blob_size_ssd", | |
4485 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4486 | "osd_memory_target", |
4487 | "osd_memory_target_cgroup_limit_ratio", | |
4488 | "osd_memory_base", | |
4489 | "osd_memory_cache_min", | |
92f5a8d4 | 4490 | "osd_memory_expected_fragmentation", |
11fdf7f2 TL |
4491 | "bluestore_cache_autotune", |
4492 | "bluestore_cache_autotune_interval", | |
81eedcae | 4493 | "bluestore_warn_on_legacy_statfs", |
9f95a23c TL |
4494 | "bluestore_warn_on_no_per_pool_omap", |
4495 | "bluestore_max_defer_interval", | |
7c673cae FG |
4496 | NULL |
4497 | }; | |
4498 | return KEYS; | |
4499 | } | |
4500 | ||
11fdf7f2 | 4501 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4502 | const std::set<std::string> &changed) |
4503 | { | |
eafe8130 | 4504 | if (changed.count("bluestore_warn_on_legacy_statfs")) { |
81eedcae TL |
4505 | _check_legacy_statfs_alert(); |
4506 | } | |
9f95a23c TL |
4507 | if (changed.count("bluestore_warn_on_no_per_pool_omap")) { |
4508 | _check_no_per_pool_omap_alert(); | |
4509 | } | |
81eedcae | 4510 | |
7c673cae FG |
4511 | if (changed.count("bluestore_csum_type")) { |
4512 | _set_csum(); | |
4513 | } | |
4514 | if (changed.count("bluestore_compression_mode") || | |
4515 | changed.count("bluestore_compression_algorithm") || | |
4516 | changed.count("bluestore_compression_min_blob_size") || | |
4517 | changed.count("bluestore_compression_max_blob_size")) { | |
4518 | if (bdev) { | |
4519 | _set_compression(); | |
4520 | } | |
4521 | } | |
4522 | if (changed.count("bluestore_max_blob_size") || | |
4523 | changed.count("bluestore_max_blob_size_ssd") || | |
4524 | changed.count("bluestore_max_blob_size_hdd")) { | |
4525 | if (bdev) { | |
4526 | // only after startup | |
4527 | _set_blob_size(); | |
4528 | } | |
4529 | } | |
4530 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4531 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4532 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4533 | changed.count("bluestore_max_alloc_size") || |
4534 | changed.count("bluestore_deferred_batch_ops") || | |
4535 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4536 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4537 | if (bdev) { | |
4538 | // only after startup | |
4539 | _set_alloc_sizes(); | |
4540 | } | |
4541 | } | |
4542 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4543 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4544 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4545 | if (bdev) { | |
4546 | _set_throttle_params(); | |
4547 | } | |
4548 | } | |
9f95a23c TL |
4549 | if (changed.count("bluestore_throttle_bytes") || |
4550 | changed.count("bluestore_throttle_deferred_bytes") || | |
4551 | changed.count("bluestore_throttle_trace_rate")) { | |
4552 | throttle.reset_throttle(conf); | |
7c673cae | 4553 | } |
9f95a23c TL |
4554 | if (changed.count("bluestore_max_defer_interval")) { |
4555 | if (bdev) { | |
4556 | _set_max_defer_interval(); | |
4557 | } | |
7c673cae | 4558 | } |
92f5a8d4 TL |
4559 | if (changed.count("osd_memory_target") || |
4560 | changed.count("osd_memory_base") || | |
4561 | changed.count("osd_memory_cache_min") || | |
4562 | changed.count("osd_memory_expected_fragmentation")) { | |
4563 | _update_osd_memory_options(); | |
4564 | } | |
7c673cae FG |
4565 | } |
4566 | ||
4567 | void BlueStore::_set_compression() | |
4568 | { | |
224ce89b WB |
4569 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4570 | if (m) { | |
11fdf7f2 | 4571 | _clear_compression_alert(); |
224ce89b WB |
4572 | comp_mode = *m; |
4573 | } else { | |
4574 | derr << __func__ << " unrecognized value '" | |
4575 | << cct->_conf->bluestore_compression_mode | |
4576 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4577 | << dendl; | |
4578 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4579 | string s("unknown mode: "); |
4580 | s += cct->_conf->bluestore_compression_mode; | |
4581 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4582 | } |
4583 | ||
4584 | compressor = nullptr; | |
4585 | ||
3efd9988 FG |
4586 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4587 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4588 | } else { |
11fdf7f2 | 4589 | ceph_assert(bdev); |
9f95a23c | 4590 | if (_use_rotational_settings()) { |
7c673cae FG |
4591 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; |
4592 | } else { | |
4593 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4594 | } | |
4595 | } | |
4596 | ||
4597 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4598 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4599 | } else { | |
11fdf7f2 | 4600 | ceph_assert(bdev); |
9f95a23c | 4601 | if (_use_rotational_settings()) { |
7c673cae FG |
4602 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; |
4603 | } else { | |
4604 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4605 | } | |
4606 | } | |
4607 | ||
7c673cae FG |
4608 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4609 | if (!alg_name.empty()) { | |
4610 | compressor = Compressor::create(cct, alg_name); | |
4611 | if (!compressor) { | |
4612 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4613 | << dendl; | |
11fdf7f2 | 4614 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4615 | } |
4616 | } | |
4617 | ||
4618 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4619 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4620 | << " min_blob " << comp_min_blob_size |
4621 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4622 | << dendl; |
4623 | } | |
4624 | ||
4625 | void BlueStore::_set_csum() | |
4626 | { | |
4627 | csum_type = Checksummer::CSUM_NONE; | |
4628 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4629 | if (t > Checksummer::CSUM_NONE) | |
4630 | csum_type = t; | |
4631 | ||
4632 | dout(10) << __func__ << " csum_type " | |
4633 | << Checksummer::get_csum_type_string(csum_type) | |
4634 | << dendl; | |
4635 | } | |
4636 | ||
4637 | void BlueStore::_set_throttle_params() | |
4638 | { | |
4639 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4640 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4641 | } else { | |
11fdf7f2 | 4642 | ceph_assert(bdev); |
9f95a23c | 4643 | if (_use_rotational_settings()) { |
7c673cae FG |
4644 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; |
4645 | } else { | |
4646 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4647 | } | |
4648 | } | |
4649 | ||
4650 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4651 | << dendl; | |
4652 | } | |
4653 | void BlueStore::_set_blob_size() | |
4654 | { | |
4655 | if (cct->_conf->bluestore_max_blob_size) { | |
4656 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4657 | } else { | |
11fdf7f2 | 4658 | ceph_assert(bdev); |
9f95a23c | 4659 | if (_use_rotational_settings()) { |
7c673cae FG |
4660 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; |
4661 | } else { | |
4662 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4663 | } | |
4664 | } | |
4665 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4666 | << std::dec << dendl; | |
4667 | } | |
4668 | ||
92f5a8d4 TL |
4669 | void BlueStore::_update_osd_memory_options() |
4670 | { | |
4671 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4672 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
4673 | osd_memory_expected_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
4674 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
4675 | config_changed++; | |
4676 | dout(10) << __func__ | |
4677 | << " osd_memory_target " << osd_memory_target | |
4678 | << " osd_memory_base " << osd_memory_base | |
4679 | << " osd_memory_expected_fragmentation " << osd_memory_expected_fragmentation | |
4680 | << " osd_memory_cache_min " << osd_memory_cache_min | |
4681 | << dendl; | |
4682 | } | |
4683 | ||
11fdf7f2 | 4684 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4685 | { |
11fdf7f2 TL |
4686 | ceph_assert(bdev); |
4687 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4688 | cache_autotune_interval = |
11fdf7f2 TL |
4689 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
4690 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4691 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4692 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4693 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4694 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4695 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4696 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4697 | |
224ce89b WB |
4698 | if (cct->_conf->bluestore_cache_size) { |
4699 | cache_size = cct->_conf->bluestore_cache_size; | |
4700 | } else { | |
4701 | // choose global cache size based on backend type | |
9f95a23c | 4702 | if (_use_rotational_settings()) { |
224ce89b WB |
4703 | cache_size = cct->_conf->bluestore_cache_size_hdd; |
4704 | } else { | |
4705 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4706 | } | |
4707 | } | |
31f18b77 | 4708 | |
91327a77 | 4709 | cache_meta_ratio = cct->_conf->bluestore_cache_meta_ratio; |
224ce89b | 4710 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4711 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4712 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4713 | return -EINVAL; |
4714 | } | |
91327a77 AA |
4715 | |
4716 | cache_kv_ratio = cct->_conf->bluestore_cache_kv_ratio; | |
224ce89b | 4717 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4718 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4719 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4720 | return -EINVAL; |
4721 | } | |
91327a77 | 4722 | |
31f18b77 | 4723 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4724 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4725 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4726 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4727 | << dendl; | |
31f18b77 FG |
4728 | return -EINVAL; |
4729 | } | |
91327a77 AA |
4730 | |
4731 | cache_data_ratio = | |
4732 | (double)1.0 - (double)cache_meta_ratio - (double)cache_kv_ratio; | |
31f18b77 FG |
4733 | if (cache_data_ratio < 0) { |
4734 | // deal with floating point imprecision | |
4735 | cache_data_ratio = 0; | |
4736 | } | |
91327a77 | 4737 | |
224ce89b WB |
4738 | dout(1) << __func__ << " cache_size " << cache_size |
4739 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4740 | << " kv " << cache_kv_ratio |
4741 | << " data " << cache_data_ratio | |
4742 | << dendl; | |
4743 | return 0; | |
4744 | } | |
4745 | ||
3efd9988 FG |
4746 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4747 | { | |
4748 | bluestore_bdev_label_t label; | |
4749 | string p = path + "/block"; | |
4750 | int r = _read_bdev_label(cct, p, &label); | |
4751 | if (r < 0) { | |
4752 | return ObjectStore::write_meta(key, value); | |
4753 | } | |
4754 | label.meta[key] = value; | |
4755 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4756 | ceph_assert(r == 0); |
3efd9988 FG |
4757 | return ObjectStore::write_meta(key, value); |
4758 | } | |
4759 | ||
4760 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4761 | { | |
4762 | bluestore_bdev_label_t label; | |
4763 | string p = path + "/block"; | |
4764 | int r = _read_bdev_label(cct, p, &label); | |
4765 | if (r < 0) { | |
4766 | return ObjectStore::read_meta(key, value); | |
4767 | } | |
4768 | auto i = label.meta.find(key); | |
4769 | if (i == label.meta.end()) { | |
4770 | return ObjectStore::read_meta(key, value); | |
4771 | } | |
4772 | *value = i->second; | |
4773 | return 0; | |
4774 | } | |
4775 | ||
7c673cae FG |
4776 | void BlueStore::_init_logger() |
4777 | { | |
4778 | PerfCountersBuilder b(cct, "bluestore", | |
4779 | l_bluestore_first, l_bluestore_last); | |
4780 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
4781 | "Average kv_thread flush latency", | |
4782 | "fl_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4783 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
4784 | "Average kv_thread commit latency"); | |
11fdf7f2 TL |
4785 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", |
4786 | "Average kv_sync thread latency", | |
4787 | "ks_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4788 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
4789 | "Average kv_finalize thread latency", | |
4790 | "kf_l", PerfCountersBuilder::PRIO_INTERESTING); | |
7c673cae FG |
4791 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
4792 | "Average prepare state latency"); | |
4793 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", | |
4794 | "Average aio_wait state latency", | |
4795 | "io_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4796 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", | |
4797 | "Average io_done state latency"); | |
4798 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", | |
4799 | "Average kv_queued state latency"); | |
4800 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", | |
4801 | "Average kv_commiting state latency"); | |
4802 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", | |
4803 | "Average kv_done state latency"); | |
4804 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", | |
4805 | "Average deferred_queued state latency"); | |
4806 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", | |
4807 | "Average aio_wait state latency"); | |
4808 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", | |
4809 | "Average cleanup state latency"); | |
4810 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
4811 | "Average finishing state latency"); | |
4812 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
4813 | "Average done state latency"); | |
4814 | b.add_time_avg(l_bluestore_throttle_lat, "throttle_lat", | |
4815 | "Average submit throttle latency", | |
4816 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4817 | b.add_time_avg(l_bluestore_submit_lat, "submit_lat", | |
4818 | "Average submit latency", | |
4819 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4820 | b.add_time_avg(l_bluestore_commit_lat, "commit_lat", | |
4821 | "Average commit latency", | |
4822 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4823 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
4824 | "Average read latency", | |
4825 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4826 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", | |
4827 | "Average read onode metadata latency"); | |
4828 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", | |
4829 | "Average read latency"); | |
4830 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
4831 | "Average compress latency"); | |
4832 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
4833 | "Average decompress latency"); | |
4834 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", | |
4835 | "Average checksum latency"); | |
4836 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
4837 | "Sum for beneficial compress ops"); | |
4838 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
4839 | "Sum for compress ops rejected due to low net gain of space"); | |
4840 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", | |
11fdf7f2 | 4841 | "Sum for write-op padded bytes", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4842 | b.add_u64_counter(l_bluestore_deferred_write_ops, "deferred_write_ops", |
4843 | "Sum for deferred write op"); | |
4844 | b.add_u64_counter(l_bluestore_deferred_write_bytes, "deferred_write_bytes", | |
11fdf7f2 | 4845 | "Sum for deferred write bytes", "def", 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4846 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", |
4847 | "Sum for write penalty read ops"); | |
4848 | b.add_u64(l_bluestore_allocated, "bluestore_allocated", | |
4849 | "Sum for allocated bytes"); | |
4850 | b.add_u64(l_bluestore_stored, "bluestore_stored", | |
4851 | "Sum for stored bytes"); | |
4852 | b.add_u64(l_bluestore_compressed, "bluestore_compressed", | |
92f5a8d4 TL |
4853 | "Sum for stored compressed bytes", |
4854 | "c", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4855 | b.add_u64(l_bluestore_compressed_allocated, "bluestore_compressed_allocated", |
92f5a8d4 TL |
4856 | "Sum for bytes allocated for compressed data", |
4857 | "c_a", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4858 | b.add_u64(l_bluestore_compressed_original, "bluestore_compressed_original", |
92f5a8d4 TL |
4859 | "Sum for original bytes that were compressed", |
4860 | "c_o", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae FG |
4861 | b.add_u64(l_bluestore_onodes, "bluestore_onodes", |
4862 | "Number of onodes in cache"); | |
9f95a23c TL |
4863 | b.add_u64(l_bluestore_pinned_onodes, "bluestore_pinned_onodes", |
4864 | "Number of pinned onodes in cache"); | |
7c673cae FG |
4865 | b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits", |
4866 | "Sum for onode-lookups hit in the cache"); | |
4867 | b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses", | |
4868 | "Sum for onode-lookups missed in the cache"); | |
4869 | b.add_u64_counter(l_bluestore_onode_shard_hits, "bluestore_onode_shard_hits", | |
4870 | "Sum for onode-shard lookups hit in the cache"); | |
4871 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
4872 | "bluestore_onode_shard_misses", | |
4873 | "Sum for onode-shard lookups missed in the cache"); | |
4874 | b.add_u64(l_bluestore_extents, "bluestore_extents", | |
4875 | "Number of extents in cache"); | |
4876 | b.add_u64(l_bluestore_blobs, "bluestore_blobs", | |
4877 | "Number of blobs in cache"); | |
4878 | b.add_u64(l_bluestore_buffers, "bluestore_buffers", | |
4879 | "Number of buffers in cache"); | |
4880 | b.add_u64(l_bluestore_buffer_bytes, "bluestore_buffer_bytes", | |
11fdf7f2 | 4881 | "Number of buffer bytes in cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4882 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "bluestore_buffer_hit_bytes", |
11fdf7f2 | 4883 | "Sum for bytes of read hit in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4884 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "bluestore_buffer_miss_bytes", |
11fdf7f2 | 4885 | "Sum for bytes of read missed in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4886 | |
4887 | b.add_u64_counter(l_bluestore_write_big, "bluestore_write_big", | |
4888 | "Large aligned writes into fresh blobs"); | |
4889 | b.add_u64_counter(l_bluestore_write_big_bytes, "bluestore_write_big_bytes", | |
11fdf7f2 | 4890 | "Large aligned writes into fresh blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4891 | b.add_u64_counter(l_bluestore_write_big_blobs, "bluestore_write_big_blobs", |
4892 | "Large aligned writes into fresh blobs (blobs)"); | |
4893 | b.add_u64_counter(l_bluestore_write_small, "bluestore_write_small", | |
4894 | "Small writes into existing or sparse small blobs"); | |
4895 | b.add_u64_counter(l_bluestore_write_small_bytes, "bluestore_write_small_bytes", | |
11fdf7f2 | 4896 | "Small writes into existing or sparse small blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4897 | b.add_u64_counter(l_bluestore_write_small_unused, |
4898 | "bluestore_write_small_unused", | |
4899 | "Small writes into unused portion of existing blob"); | |
4900 | b.add_u64_counter(l_bluestore_write_small_deferred, | |
4901 | "bluestore_write_small_deferred", | |
4902 | "Small overwrites using deferred"); | |
4903 | b.add_u64_counter(l_bluestore_write_small_pre_read, | |
4904 | "bluestore_write_small_pre_read", | |
4905 | "Small writes that required we read some data (possibly " | |
4906 | "cached) to fill out the block"); | |
4907 | b.add_u64_counter(l_bluestore_write_small_new, "bluestore_write_small_new", | |
4908 | "Small write into new (sparse) blob"); | |
4909 | ||
4910 | b.add_u64_counter(l_bluestore_txc, "bluestore_txc", "Transactions committed"); | |
4911 | b.add_u64_counter(l_bluestore_onode_reshard, "bluestore_onode_reshard", | |
4912 | "Onode extent map reshard events"); | |
4913 | b.add_u64_counter(l_bluestore_blob_split, "bluestore_blob_split", | |
4914 | "Sum for blob splitting due to resharding"); | |
4915 | b.add_u64_counter(l_bluestore_extent_compress, "bluestore_extent_compress", | |
4916 | "Sum for extents that have been removed due to compression"); | |
4917 | b.add_u64_counter(l_bluestore_gc_merged, "bluestore_gc_merged", | |
4918 | "Sum for extents that have been merged due to garbage " | |
4919 | "collection"); | |
b32b8144 FG |
4920 | b.add_u64_counter(l_bluestore_read_eio, "bluestore_read_eio", |
4921 | "Read EIO errors propagated to high level callers"); | |
f64942e4 AA |
4922 | b.add_u64_counter(l_bluestore_reads_with_retries, "bluestore_reads_with_retries", |
4923 | "Read operations that required at least one retry due to failed checksum validation"); | |
a8e16298 TL |
4924 | b.add_u64(l_bluestore_fragmentation, "bluestore_fragmentation_micros", |
4925 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
11fdf7f2 TL |
4926 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
4927 | "Average omap iterator seek_to_first call latency"); | |
4928 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", | |
4929 | "Average omap iterator upper_bound call latency"); | |
4930 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", | |
4931 | "Average omap iterator lower_bound call latency"); | |
4932 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", | |
4933 | "Average omap iterator next call latency"); | |
494da23a TL |
4934 | b.add_time_avg(l_bluestore_clist_lat, "clist_lat", |
4935 | "Average collection listing latency"); | |
7c673cae FG |
4936 | logger = b.create_perf_counters(); |
4937 | cct->get_perfcounters_collection()->add(logger); | |
4938 | } | |
4939 | ||
4940 | int BlueStore::_reload_logger() | |
4941 | { | |
4942 | struct store_statfs_t store_statfs; | |
7c673cae | 4943 | int r = statfs(&store_statfs); |
11fdf7f2 | 4944 | if (r >= 0) { |
7c673cae | 4945 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
4946 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
4947 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
4948 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
4949 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
4950 | } |
4951 | return r; | |
4952 | } | |
4953 | ||
4954 | void BlueStore::_shutdown_logger() | |
4955 | { | |
4956 | cct->get_perfcounters_collection()->remove(logger); | |
4957 | delete logger; | |
4958 | } | |
4959 | ||
4960 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
4961 | uuid_d *fsid) | |
4962 | { | |
4963 | bluestore_bdev_label_t label; | |
4964 | int r = _read_bdev_label(cct, path, &label); | |
4965 | if (r < 0) | |
4966 | return r; | |
4967 | *fsid = label.osd_uuid; | |
4968 | return 0; | |
4969 | } | |
4970 | ||
4971 | int BlueStore::_open_path() | |
4972 | { | |
b32b8144 | 4973 | // sanity check(s) |
11fdf7f2 | 4974 | ceph_assert(path_fd < 0); |
91327a77 | 4975 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
4976 | if (path_fd < 0) { |
4977 | int r = -errno; | |
4978 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
4979 | << dendl; | |
4980 | return r; | |
4981 | } | |
4982 | return 0; | |
4983 | } | |
4984 | ||
4985 | void BlueStore::_close_path() | |
4986 | { | |
4987 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
4988 | path_fd = -1; | |
4989 | } | |
4990 | ||
3efd9988 FG |
4991 | int BlueStore::_write_bdev_label(CephContext *cct, |
4992 | string path, bluestore_bdev_label_t label) | |
7c673cae FG |
4993 | { |
4994 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
4995 | bufferlist bl; | |
11fdf7f2 | 4996 | encode(label, bl); |
7c673cae | 4997 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
4998 | encode(crc, bl); |
4999 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
5000 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
5001 | z.zero(); | |
5002 | bl.append(std::move(z)); | |
5003 | ||
91327a77 | 5004 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC)); |
7c673cae FG |
5005 | if (fd < 0) { |
5006 | fd = -errno; | |
5007 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5008 | << dendl; | |
5009 | return fd; | |
5010 | } | |
5011 | int r = bl.write_fd(fd); | |
5012 | if (r < 0) { | |
5013 | derr << __func__ << " failed to write to " << path | |
5014 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 5015 | goto out; |
7c673cae | 5016 | } |
3efd9988 FG |
5017 | r = ::fsync(fd); |
5018 | if (r < 0) { | |
5019 | derr << __func__ << " failed to fsync " << path | |
5020 | << ": " << cpp_strerror(r) << dendl; | |
5021 | } | |
11fdf7f2 | 5022 | out: |
7c673cae FG |
5023 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
5024 | return r; | |
5025 | } | |
5026 | ||
5027 | int BlueStore::_read_bdev_label(CephContext* cct, string path, | |
5028 | bluestore_bdev_label_t *label) | |
5029 | { | |
5030 | dout(10) << __func__ << dendl; | |
91327a77 | 5031 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
5032 | if (fd < 0) { |
5033 | fd = -errno; | |
5034 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
5035 | << dendl; | |
5036 | return fd; | |
5037 | } | |
5038 | bufferlist bl; | |
5039 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
5040 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5041 | if (r < 0) { | |
5042 | derr << __func__ << " failed to read from " << path | |
5043 | << ": " << cpp_strerror(r) << dendl; | |
5044 | return r; | |
5045 | } | |
5046 | ||
5047 | uint32_t crc, expected_crc; | |
11fdf7f2 | 5048 | auto p = bl.cbegin(); |
7c673cae | 5049 | try { |
11fdf7f2 | 5050 | decode(*label, p); |
7c673cae FG |
5051 | bufferlist t; |
5052 | t.substr_of(bl, 0, p.get_off()); | |
5053 | crc = t.crc32c(-1); | |
11fdf7f2 | 5054 | decode(expected_crc, p); |
7c673cae FG |
5055 | } |
5056 | catch (buffer::error& e) { | |
b32b8144 | 5057 | dout(2) << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
5058 | << ": " << e.what() |
5059 | << dendl; | |
b32b8144 | 5060 | return -ENOENT; |
7c673cae FG |
5061 | } |
5062 | if (crc != expected_crc) { | |
5063 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
5064 | << " != actual " << crc << dendl; | |
5065 | return -EIO; | |
5066 | } | |
5067 | dout(10) << __func__ << " got " << *label << dendl; | |
5068 | return 0; | |
5069 | } | |
5070 | ||
5071 | int BlueStore::_check_or_set_bdev_label( | |
5072 | string path, uint64_t size, string desc, bool create) | |
5073 | { | |
5074 | bluestore_bdev_label_t label; | |
5075 | if (create) { | |
5076 | label.osd_uuid = fsid; | |
5077 | label.size = size; | |
5078 | label.btime = ceph_clock_now(); | |
5079 | label.description = desc; | |
3efd9988 | 5080 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
5081 | if (r < 0) |
5082 | return r; | |
5083 | } else { | |
5084 | int r = _read_bdev_label(cct, path, &label); | |
5085 | if (r < 0) | |
5086 | return r; | |
31f18b77 FG |
5087 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
5088 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
5089 | << " and fsid " << fsid << " check bypassed" << dendl; | |
1911f103 | 5090 | } else if (label.osd_uuid != fsid) { |
7c673cae FG |
5091 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
5092 | << " does not match our fsid " << fsid << dendl; | |
5093 | return -EIO; | |
5094 | } | |
5095 | } | |
5096 | return 0; | |
5097 | } | |
5098 | ||
5099 | void BlueStore::_set_alloc_sizes(void) | |
5100 | { | |
7c673cae FG |
5101 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
5102 | ||
5103 | if (cct->_conf->bluestore_prefer_deferred_size) { | |
5104 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
5105 | } else { | |
11fdf7f2 | 5106 | ceph_assert(bdev); |
9f95a23c | 5107 | if (_use_rotational_settings()) { |
7c673cae FG |
5108 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; |
5109 | } else { | |
5110 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
5111 | } | |
5112 | } | |
5113 | ||
5114 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
5115 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
5116 | } else { | |
11fdf7f2 | 5117 | ceph_assert(bdev); |
9f95a23c | 5118 | if (_use_rotational_settings()) { |
7c673cae FG |
5119 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; |
5120 | } else { | |
5121 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
5122 | } | |
5123 | } | |
5124 | ||
5125 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 5126 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
5127 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
5128 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
5129 | << std::dec | |
5130 | << " deferred_batch_ops " << deferred_batch_ops | |
5131 | << dendl; | |
5132 | } | |
5133 | ||
5134 | int BlueStore::_open_bdev(bool create) | |
5135 | { | |
11fdf7f2 | 5136 | ceph_assert(bdev == NULL); |
7c673cae | 5137 | string p = path + "/block"; |
11fdf7f2 | 5138 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
5139 | int r = bdev->open(p); |
5140 | if (r < 0) | |
5141 | goto fail; | |
5142 | ||
11fdf7f2 TL |
5143 | if (create && cct->_conf->bdev_enable_discard) { |
5144 | bdev->discard(0, bdev->get_size()); | |
5145 | } | |
5146 | ||
7c673cae FG |
5147 | if (bdev->supported_bdev_label()) { |
5148 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
5149 | if (r < 0) | |
5150 | goto fail_close; | |
5151 | } | |
5152 | ||
5153 | // initialize global block parameters | |
5154 | block_size = bdev->get_block_size(); | |
5155 | block_mask = ~(block_size - 1); | |
5156 | block_size_order = ctz(block_size); | |
11fdf7f2 | 5157 | ceph_assert(block_size == 1u << block_size_order); |
9f95a23c | 5158 | _set_max_defer_interval(); |
224ce89b WB |
5159 | // and set cache_size based on device type |
5160 | r = _set_cache_sizes(); | |
5161 | if (r < 0) { | |
5162 | goto fail_close; | |
5163 | } | |
7c673cae FG |
5164 | return 0; |
5165 | ||
5166 | fail_close: | |
5167 | bdev->close(); | |
5168 | fail: | |
5169 | delete bdev; | |
5170 | bdev = NULL; | |
5171 | return r; | |
5172 | } | |
5173 | ||
11fdf7f2 TL |
5174 | void BlueStore::_validate_bdev() |
5175 | { | |
5176 | ceph_assert(bdev); | |
5177 | ceph_assert(min_alloc_size); // _get_odisk_reserved depends on that | |
5178 | uint64_t dev_size = bdev->get_size(); | |
5179 | if (dev_size < | |
5180 | _get_ondisk_reserved() + cct->_conf->bluestore_bluefs_min) { | |
5181 | dout(1) << __func__ << " main device size " << byte_u_t(dev_size) | |
5182 | << " is too small, disable bluestore_bluefs_min for now" | |
5183 | << dendl; | |
5184 | ceph_assert(dev_size >= _get_ondisk_reserved()); | |
5185 | ||
5186 | int r = cct->_conf.set_val("bluestore_bluefs_min", "0"); | |
5187 | ceph_assert(r == 0); | |
5188 | } | |
5189 | } | |
5190 | ||
7c673cae FG |
5191 | void BlueStore::_close_bdev() |
5192 | { | |
11fdf7f2 | 5193 | ceph_assert(bdev); |
7c673cae FG |
5194 | bdev->close(); |
5195 | delete bdev; | |
5196 | bdev = NULL; | |
5197 | } | |
5198 | ||
1911f103 | 5199 | int BlueStore::_open_fm(KeyValueDB::Transaction t, bool read_only) |
7c673cae | 5200 | { |
1911f103 TL |
5201 | int r; |
5202 | bluestore_bdev_label_t label; | |
5203 | ||
11fdf7f2 TL |
5204 | ceph_assert(fm == NULL); |
5205 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); | |
5206 | ceph_assert(fm); | |
5207 | if (t) { | |
5208 | // create mode. initialize freespace | |
7c673cae | 5209 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
5210 | { |
5211 | bufferlist bl; | |
5212 | bl.append(freelist_type); | |
5213 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
5214 | } | |
b32b8144 FG |
5215 | // being able to allocate in units less than bdev block size |
5216 | // seems to be a bad idea. | |
11fdf7f2 | 5217 | ceph_assert( cct->_conf->bdev_block_size <= (int64_t)min_alloc_size); |
b32b8144 | 5218 | fm->create(bdev->get_size(), (int64_t)min_alloc_size, t); |
7c673cae FG |
5219 | |
5220 | // allocate superblock reserved space. note that we do not mark | |
5221 | // bluefs space as allocated in the freelist; we instead rely on | |
5222 | // bluefs_extents. | |
11fdf7f2 | 5223 | auto reserved = _get_ondisk_reserved(); |
3efd9988 | 5224 | fm->allocate(0, reserved, t); |
7c673cae | 5225 | |
7c673cae | 5226 | if (cct->_conf->bluestore_bluefs) { |
11fdf7f2 | 5227 | ceph_assert(bluefs_extents.num_intervals() == 1); |
7c673cae | 5228 | interval_set<uint64_t>::iterator p = bluefs_extents.begin(); |
11fdf7f2 | 5229 | reserved = round_up_to(p.get_start() + p.get_len(), min_alloc_size); |
7c673cae FG |
5230 | dout(20) << __func__ << " reserved 0x" << std::hex << reserved << std::dec |
5231 | << " for bluefs" << dendl; | |
7c673cae FG |
5232 | } |
5233 | ||
5234 | if (cct->_conf->bluestore_debug_prefill > 0) { | |
5235 | uint64_t end = bdev->get_size() - reserved; | |
5236 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
5237 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
5238 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 5239 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
5240 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
5241 | float r = cct->_conf->bluestore_debug_prefill; | |
5242 | r /= 1.0 - r; | |
5243 | bool stop = false; | |
5244 | ||
5245 | while (!stop && start < end) { | |
5246 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
5247 | if (start + l > end) { | |
5248 | l = end - start; | |
11fdf7f2 | 5249 | l = p2align(l, min_alloc_size); |
7c673cae | 5250 | } |
11fdf7f2 | 5251 | ceph_assert(start + l <= end); |
7c673cae FG |
5252 | |
5253 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 5254 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
5255 | if (start + l + u > end) { |
5256 | u = end - (start + l); | |
5257 | // trim to align so we don't overflow again | |
11fdf7f2 | 5258 | u = p2align(u, min_alloc_size); |
7c673cae FG |
5259 | stop = true; |
5260 | } | |
11fdf7f2 | 5261 | ceph_assert(start + l + u <= end); |
7c673cae | 5262 | |
11fdf7f2 | 5263 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
5264 | << " use 0x" << u << std::dec << dendl; |
5265 | ||
5266 | if (u == 0) { | |
5267 | // break if u has been trimmed to nothing | |
5268 | break; | |
5269 | } | |
5270 | ||
5271 | fm->allocate(start + l, u, t); | |
5272 | start += l + u; | |
5273 | } | |
5274 | } | |
1911f103 TL |
5275 | r = _write_out_fm_meta(0, false, &label); |
5276 | ceph_assert(r == 0); | |
5277 | } else { | |
5278 | string p = path + "/block"; | |
5279 | r = _read_bdev_label(cct, p, &label); | |
5280 | if (r < 0) { | |
5281 | derr << __func__ << " freelist init failed, error reading bdev label: " << cpp_strerror(r) << dendl; | |
5282 | delete fm; | |
5283 | fm = NULL; | |
5284 | return r; | |
5285 | } | |
7c673cae | 5286 | } |
1911f103 | 5287 | r = fm->init(label, db, read_only); |
7c673cae FG |
5288 | if (r < 0) { |
5289 | derr << __func__ << " freelist init failed: " << cpp_strerror(r) << dendl; | |
5290 | delete fm; | |
5291 | fm = NULL; | |
5292 | return r; | |
5293 | } | |
81eedcae TL |
5294 | // if space size tracked by free list manager is that higher than actual |
5295 | // dev size one can hit out-of-space allocation which will result | |
5296 | // in data loss and/or assertions | |
5297 | // Probably user altered the device size somehow. | |
5298 | // The only fix for now is to redeploy OSD. | |
5299 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
5300 | ostringstream ss; | |
5301 | ss << "slow device size mismatch detected, " | |
5302 | << " fm size(" << fm->get_size() | |
5303 | << ") > slow device size(" << bdev->get_size() | |
5304 | << "), Please stop using this OSD as it might cause data loss."; | |
5305 | _set_disk_size_mismatch_alert(ss.str()); | |
5306 | } | |
7c673cae FG |
5307 | return 0; |
5308 | } | |
5309 | ||
5310 | void BlueStore::_close_fm() | |
5311 | { | |
5312 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 5313 | ceph_assert(fm); |
7c673cae FG |
5314 | fm->shutdown(); |
5315 | delete fm; | |
5316 | fm = NULL; | |
5317 | } | |
5318 | ||
1911f103 TL |
5319 | int BlueStore::_write_out_fm_meta(uint64_t target_size, |
5320 | bool update_root_size, | |
5321 | bluestore_bdev_label_t* res_label) | |
5322 | { | |
5323 | string p = path + "/block"; | |
5324 | ||
5325 | std::vector<std::pair<string, string>> fm_meta; | |
5326 | fm->get_meta(target_size, &fm_meta); | |
5327 | ||
5328 | bluestore_bdev_label_t label; | |
5329 | int r = _read_bdev_label(cct, p, &label); | |
5330 | if (r < 0) | |
5331 | return r; | |
5332 | ||
5333 | for (auto& m : fm_meta) { | |
5334 | label.meta[m.first] = m.second; | |
5335 | } | |
5336 | if (update_root_size) { | |
5337 | label.size = target_size; | |
5338 | } | |
5339 | r = _write_bdev_label(cct, p, label); | |
5340 | if (res_label) { | |
5341 | *res_label = label; | |
5342 | } | |
5343 | ||
5344 | return r; | |
5345 | } | |
5346 | ||
7c673cae FG |
5347 | int BlueStore::_open_alloc() |
5348 | { | |
11fdf7f2 TL |
5349 | ceph_assert(alloc == NULL); |
5350 | ceph_assert(bdev->get_size()); | |
5351 | ||
5352 | if (bluefs) { | |
5353 | bluefs_extents.clear(); | |
9f95a23c TL |
5354 | auto r = bluefs->get_block_extents(bluefs_layout.shared_bdev, |
5355 | &bluefs_extents); | |
11fdf7f2 TL |
5356 | if (r < 0) { |
5357 | lderr(cct) << __func__ << " failed to retrieve bluefs_extents: " | |
5358 | << cpp_strerror(r) << dendl; | |
5359 | ||
5360 | return r; | |
5361 | } | |
5362 | dout(10) << __func__ << " bluefs extents 0x" | |
5363 | << std::hex << bluefs_extents << std::dec | |
5364 | << dendl; | |
5365 | } | |
5366 | ||
7c673cae FG |
5367 | alloc = Allocator::create(cct, cct->_conf->bluestore_allocator, |
5368 | bdev->get_size(), | |
eafe8130 | 5369 | min_alloc_size, "block"); |
7c673cae FG |
5370 | if (!alloc) { |
5371 | lderr(cct) << __func__ << " Allocator::unknown alloc type " | |
5372 | << cct->_conf->bluestore_allocator | |
5373 | << dendl; | |
5374 | return -EINVAL; | |
5375 | } | |
5376 | ||
5377 | uint64_t num = 0, bytes = 0; | |
5378 | ||
5379 | dout(1) << __func__ << " opening allocation metadata" << dendl; | |
5380 | // initialize from freelist | |
5381 | fm->enumerate_reset(); | |
5382 | uint64_t offset, length; | |
11fdf7f2 | 5383 | while (fm->enumerate_next(db, &offset, &length)) { |
7c673cae FG |
5384 | alloc->init_add_free(offset, length); |
5385 | ++num; | |
5386 | bytes += length; | |
5387 | } | |
224ce89b | 5388 | fm->enumerate_reset(); |
7c673cae FG |
5389 | |
5390 | // also mark bluefs space as allocated | |
5391 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
5392 | alloc->init_rm_free(e.get_start(), e.get_len()); | |
5393 | } | |
7c673cae | 5394 | |
1911f103 TL |
5395 | dout(1) << __func__ << " loaded " << byte_u_t(bytes) |
5396 | << " in " << num << " extents" | |
5397 | << " available " << byte_u_t(alloc->get_free()) | |
5398 | << dendl; | |
5399 | ||
7c673cae FG |
5400 | return 0; |
5401 | } | |
5402 | ||
5403 | void BlueStore::_close_alloc() | |
5404 | { | |
11fdf7f2 TL |
5405 | ceph_assert(bdev); |
5406 | bdev->discard_drain(); | |
5407 | ||
5408 | ceph_assert(alloc); | |
7c673cae FG |
5409 | alloc->shutdown(); |
5410 | delete alloc; | |
5411 | alloc = NULL; | |
11fdf7f2 | 5412 | bluefs_extents.clear(); |
7c673cae FG |
5413 | } |
5414 | ||
5415 | int BlueStore::_open_fsid(bool create) | |
5416 | { | |
11fdf7f2 | 5417 | ceph_assert(fsid_fd < 0); |
91327a77 | 5418 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
5419 | if (create) |
5420 | flags |= O_CREAT; | |
5421 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
5422 | if (fsid_fd < 0) { | |
5423 | int err = -errno; | |
5424 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
5425 | return err; | |
5426 | } | |
5427 | return 0; | |
5428 | } | |
5429 | ||
5430 | int BlueStore::_read_fsid(uuid_d *uuid) | |
5431 | { | |
5432 | char fsid_str[40]; | |
5433 | memset(fsid_str, 0, sizeof(fsid_str)); | |
5434 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
5435 | if (ret < 0) { | |
5436 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
5437 | return ret; | |
5438 | } | |
5439 | if (ret > 36) | |
5440 | fsid_str[36] = 0; | |
5441 | else | |
5442 | fsid_str[ret] = 0; | |
5443 | if (!uuid->parse(fsid_str)) { | |
5444 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
5445 | return -EINVAL; | |
5446 | } | |
5447 | return 0; | |
5448 | } | |
5449 | ||
5450 | int BlueStore::_write_fsid() | |
5451 | { | |
5452 | int r = ::ftruncate(fsid_fd, 0); | |
5453 | if (r < 0) { | |
5454 | r = -errno; | |
5455 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
5456 | return r; | |
5457 | } | |
5458 | string str = stringify(fsid) + "\n"; | |
5459 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
5460 | if (r < 0) { | |
5461 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
5462 | return r; | |
5463 | } | |
5464 | r = ::fsync(fsid_fd); | |
5465 | if (r < 0) { | |
5466 | r = -errno; | |
5467 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
5468 | return r; | |
5469 | } | |
5470 | return 0; | |
5471 | } | |
5472 | ||
5473 | void BlueStore::_close_fsid() | |
5474 | { | |
5475 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
5476 | fsid_fd = -1; | |
5477 | } | |
5478 | ||
5479 | int BlueStore::_lock_fsid() | |
5480 | { | |
5481 | struct flock l; | |
5482 | memset(&l, 0, sizeof(l)); | |
5483 | l.l_type = F_WRLCK; | |
5484 | l.l_whence = SEEK_SET; | |
5485 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
5486 | if (r < 0) { | |
5487 | int err = errno; | |
5488 | derr << __func__ << " failed to lock " << path << "/fsid" | |
5489 | << " (is another ceph-osd still running?)" | |
5490 | << cpp_strerror(err) << dendl; | |
5491 | return -err; | |
5492 | } | |
5493 | return 0; | |
5494 | } | |
5495 | ||
31f18b77 FG |
5496 | bool BlueStore::is_rotational() |
5497 | { | |
5498 | if (bdev) { | |
5499 | return bdev->is_rotational(); | |
5500 | } | |
5501 | ||
5502 | bool rotational = true; | |
5503 | int r = _open_path(); | |
5504 | if (r < 0) | |
5505 | goto out; | |
5506 | r = _open_fsid(false); | |
5507 | if (r < 0) | |
5508 | goto out_path; | |
5509 | r = _read_fsid(&fsid); | |
5510 | if (r < 0) | |
5511 | goto out_fsid; | |
5512 | r = _lock_fsid(); | |
5513 | if (r < 0) | |
5514 | goto out_fsid; | |
5515 | r = _open_bdev(false); | |
5516 | if (r < 0) | |
5517 | goto out_fsid; | |
5518 | rotational = bdev->is_rotational(); | |
5519 | _close_bdev(); | |
5520 | out_fsid: | |
5521 | _close_fsid(); | |
5522 | out_path: | |
5523 | _close_path(); | |
5524 | out: | |
5525 | return rotational; | |
5526 | } | |
5527 | ||
d2e6a577 FG |
5528 | bool BlueStore::is_journal_rotational() |
5529 | { | |
5530 | if (!bluefs) { | |
5531 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
5532 | << dendl; | |
5533 | return is_rotational(); | |
5534 | } | |
5535 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
5536 | return bluefs->wal_is_rotational(); | |
5537 | } | |
5538 | ||
9f95a23c TL |
5539 | bool BlueStore::_use_rotational_settings() |
5540 | { | |
5541 | if (cct->_conf->bluestore_debug_enforce_settings == "hdd") { | |
5542 | return true; | |
5543 | } | |
5544 | if (cct->_conf->bluestore_debug_enforce_settings == "ssd") { | |
5545 | return false; | |
5546 | } | |
5547 | return bdev->is_rotational(); | |
5548 | } | |
5549 | ||
7c673cae FG |
5550 | bool BlueStore::test_mount_in_use() |
5551 | { | |
5552 | // most error conditions mean the mount is not in use (e.g., because | |
5553 | // it doesn't exist). only if we fail to lock do we conclude it is | |
5554 | // in use. | |
5555 | bool ret = false; | |
5556 | int r = _open_path(); | |
5557 | if (r < 0) | |
5558 | return false; | |
5559 | r = _open_fsid(false); | |
5560 | if (r < 0) | |
5561 | goto out_path; | |
5562 | r = _lock_fsid(); | |
5563 | if (r < 0) | |
5564 | ret = true; // if we can't lock, it is in use | |
5565 | _close_fsid(); | |
5566 | out_path: | |
5567 | _close_path(); | |
5568 | return ret; | |
5569 | } | |
5570 | ||
11fdf7f2 | 5571 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
5572 | { |
5573 | int r; | |
11fdf7f2 | 5574 | bluefs = new BlueFS(cct); |
7c673cae | 5575 | |
11fdf7f2 TL |
5576 | string bfn; |
5577 | struct stat st; | |
5578 | ||
5579 | bfn = path + "/block.db"; | |
5580 | if (::stat(bfn.c_str(), &st) == 0) { | |
eafe8130 TL |
5581 | r = bluefs->add_block_device( |
5582 | BlueFS::BDEV_DB, bfn, | |
5583 | create && cct->_conf->bdev_enable_discard); | |
7c673cae | 5584 | if (r < 0) { |
11fdf7f2 TL |
5585 | derr << __func__ << " add block device(" << bfn << ") returned: " |
5586 | << cpp_strerror(r) << dendl; | |
5587 | goto free_bluefs; | |
7c673cae | 5588 | } |
7c673cae | 5589 | |
11fdf7f2 TL |
5590 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
5591 | r = _check_or_set_bdev_label( | |
5592 | bfn, | |
5593 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
5594 | "bluefs db", create); | |
5595 | if (r < 0) { | |
5596 | derr << __func__ | |
5597 | << " check block device(" << bfn << ") label returned: " | |
5598 | << cpp_strerror(r) << dendl; | |
5599 | goto free_bluefs; | |
5600 | } | |
7c673cae | 5601 | } |
11fdf7f2 TL |
5602 | if (create) { |
5603 | bluefs->add_block_extent( | |
5604 | BlueFS::BDEV_DB, | |
5605 | SUPER_RESERVED, | |
5606 | bluefs->get_block_device_size(BlueFS::BDEV_DB) - SUPER_RESERVED); | |
5607 | } | |
9f95a23c TL |
5608 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
5609 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
5610 | } else { |
5611 | r = -errno; | |
5612 | if (::lstat(bfn.c_str(), &st) == -1) { | |
5613 | r = 0; | |
9f95a23c | 5614 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7c673cae | 5615 | } else { |
11fdf7f2 TL |
5616 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
5617 | << cpp_strerror(r) << dendl; | |
5618 | goto free_bluefs; | |
7c673cae FG |
5619 | } |
5620 | } | |
7c673cae | 5621 | |
11fdf7f2 TL |
5622 | // shared device |
5623 | bfn = path + "/block"; | |
5624 | // never trim here | |
9f95a23c | 5625 | r = bluefs->add_block_device(bluefs_layout.shared_bdev, bfn, false, |
11fdf7f2 TL |
5626 | true /* shared with bluestore */); |
5627 | if (r < 0) { | |
5628 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5629 | << cpp_strerror(r) << dendl; | |
5630 | goto free_bluefs; | |
5631 | } | |
5632 | if (create) { | |
5633 | // note: we always leave the first SUPER_RESERVED (8k) of the device unused | |
5634 | uint64_t initial = | |
5635 | bdev->get_size() * (cct->_conf->bluestore_bluefs_min_ratio + | |
5636 | cct->_conf->bluestore_bluefs_gift_ratio); | |
5637 | initial = std::max(initial, cct->_conf->bluestore_bluefs_min); | |
eafe8130 TL |
5638 | uint64_t alloc_size = cct->_conf->bluefs_shared_alloc_size; |
5639 | if (alloc_size % min_alloc_size) { | |
5640 | derr << __func__ << " bluefs_shared_alloc_size 0x" << std::hex | |
5641 | << alloc_size << " is not a multiple of " | |
11fdf7f2 TL |
5642 | << "min_alloc_size 0x" << min_alloc_size << std::dec << dendl; |
5643 | r = -EINVAL; | |
5644 | goto free_bluefs; | |
7c673cae | 5645 | } |
11fdf7f2 | 5646 | // align to bluefs's alloc_size |
eafe8130 | 5647 | initial = p2roundup(initial, alloc_size); |
11fdf7f2 | 5648 | // put bluefs in the middle of the device in case it is an HDD |
eafe8130 | 5649 | uint64_t start = p2align((bdev->get_size() - initial) / 2, alloc_size); |
11fdf7f2 | 5650 | //avoiding superblock overwrite |
eafe8130 TL |
5651 | start = std::max(alloc_size, start); |
5652 | ceph_assert(start >=_get_ondisk_reserved()); | |
7c673cae | 5653 | |
9f95a23c | 5654 | bluefs->add_block_extent(bluefs_layout.shared_bdev, start, initial); |
11fdf7f2 TL |
5655 | bluefs_extents.insert(start, initial); |
5656 | ++out_of_sync_fm; | |
5657 | } | |
5658 | ||
5659 | bfn = path + "/block.wal"; | |
5660 | if (::stat(bfn.c_str(), &st) == 0) { | |
5661 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
eafe8130 | 5662 | create && cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
5663 | if (r < 0) { |
5664 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5665 | << cpp_strerror(r) << dendl; | |
5666 | goto free_bluefs; | |
5667 | } | |
7c673cae | 5668 | |
11fdf7f2 TL |
5669 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
5670 | r = _check_or_set_bdev_label( | |
5671 | bfn, | |
5672 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
5673 | "bluefs wal", create); | |
7c673cae | 5674 | if (r < 0) { |
11fdf7f2 TL |
5675 | derr << __func__ << " check block device(" << bfn |
5676 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
5677 | goto free_bluefs; |
5678 | } | |
7c673cae FG |
5679 | } |
5680 | ||
11fdf7f2 TL |
5681 | if (create) { |
5682 | bluefs->add_block_extent( | |
5683 | BlueFS::BDEV_WAL, BDEV_LABEL_BLOCK_SIZE, | |
5684 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) - | |
5685 | BDEV_LABEL_BLOCK_SIZE); | |
5686 | } | |
9f95a23c | 5687 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
5688 | } else { |
5689 | r = 0; | |
5690 | if (::lstat(bfn.c_str(), &st) != -1) { | |
5691 | r = -errno; | |
5692 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
5693 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
5694 | goto free_bluefs; |
5695 | } | |
11fdf7f2 TL |
5696 | } |
5697 | return 0; | |
7c673cae | 5698 | |
11fdf7f2 TL |
5699 | free_bluefs: |
5700 | ceph_assert(bluefs); | |
5701 | delete bluefs; | |
5702 | bluefs = NULL; | |
5703 | return r; | |
5704 | } | |
7c673cae | 5705 | |
11fdf7f2 TL |
5706 | int BlueStore::_open_bluefs(bool create) |
5707 | { | |
5708 | int r = _minimal_open_bluefs(create); | |
5709 | if (r < 0) { | |
5710 | return r; | |
5711 | } | |
9f95a23c TL |
5712 | RocksDBBlueFSVolumeSelector* vselector = nullptr; |
5713 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { | |
5714 | ||
5715 | string options = cct->_conf->bluestore_rocksdb_options; | |
5716 | ||
5717 | rocksdb::Options rocks_opts; | |
5718 | int r = RocksDBStore::ParseOptionsFromStringStatic( | |
5719 | cct, | |
5720 | options, | |
5721 | rocks_opts, | |
5722 | nullptr); | |
5723 | if (r < 0) { | |
5724 | return r; | |
5725 | } | |
5726 | ||
5727 | double reserved_factor = cct->_conf->bluestore_volume_selection_reserved_factor; | |
5728 | vselector = | |
5729 | new RocksDBBlueFSVolumeSelector( | |
5730 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
5731 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
5732 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100, | |
5733 | 1024 * 1024 * 1024, //FIXME: set expected l0 size here | |
5734 | rocks_opts.max_bytes_for_level_base, | |
5735 | rocks_opts.max_bytes_for_level_multiplier, | |
5736 | reserved_factor, | |
5737 | cct->_conf->bluestore_volume_selection_reserved, | |
5738 | cct->_conf->bluestore_volume_selection_policy != "rocksdb_original"); | |
5739 | } | |
11fdf7f2 | 5740 | if (create) { |
9f95a23c | 5741 | bluefs->mkfs(fsid, bluefs_layout); |
11fdf7f2 | 5742 | } |
9f95a23c | 5743 | bluefs->set_volume_selector(vselector); |
11fdf7f2 TL |
5744 | r = bluefs->mount(); |
5745 | if (r < 0) { | |
5746 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
5747 | } | |
9f95a23c | 5748 | ceph_assert_always(bluefs->maybe_verify_layout(bluefs_layout) == 0); |
11fdf7f2 TL |
5749 | return r; |
5750 | } | |
5751 | ||
1911f103 | 5752 | void BlueStore::_close_bluefs(bool cold_close) |
11fdf7f2 | 5753 | { |
1911f103 | 5754 | bluefs->umount(cold_close); |
11fdf7f2 TL |
5755 | _minimal_close_bluefs(); |
5756 | } | |
5757 | ||
5758 | void BlueStore::_minimal_close_bluefs() | |
5759 | { | |
5760 | delete bluefs; | |
5761 | bluefs = NULL; | |
5762 | } | |
5763 | ||
5764 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
5765 | { | |
5766 | if (create) { | |
5767 | *ret = cct->_conf->bluestore_bluefs; | |
5768 | } else { | |
5769 | string s; | |
5770 | int r = read_meta("bluefs", &s); | |
5771 | if (r < 0) { | |
5772 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
5773 | return -EIO; | |
5774 | } | |
5775 | if (s == "1") { | |
5776 | *ret = true; | |
5777 | } else if (s == "0") { | |
5778 | *ret = false; | |
31f18b77 | 5779 | } else { |
11fdf7f2 TL |
5780 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
5781 | << dendl; | |
5782 | return -EIO; | |
5783 | } | |
5784 | } | |
5785 | return 0; | |
5786 | } | |
5787 | ||
5788 | /* | |
5789 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
5790 | * in the proper order | |
5791 | */ | |
5792 | int BlueStore::_open_db_and_around(bool read_only) | |
5793 | { | |
5794 | int r; | |
5795 | bool do_bluefs = false; | |
5796 | _is_bluefs(false, &do_bluefs); // ignore err code | |
5797 | if (do_bluefs) { | |
5798 | // open in read-only first to read FM list and init allocator | |
5799 | // as they might be needed for some BlueFS procedures | |
5800 | r = _open_db(false, false, true); | |
5801 | if (r < 0) | |
5802 | return r; | |
5803 | ||
5804 | r = _open_super_meta(); | |
5805 | if (r < 0) { | |
5806 | goto out_db; | |
5807 | } | |
5808 | ||
1911f103 | 5809 | r = _open_fm(nullptr, true); |
11fdf7f2 TL |
5810 | if (r < 0) |
5811 | goto out_db; | |
5812 | ||
5813 | r = _open_alloc(); | |
5814 | if (r < 0) | |
5815 | goto out_fm; | |
5816 | ||
5817 | // now open in R/W mode | |
5818 | if (!read_only) { | |
1911f103 | 5819 | _close_db(true); |
11fdf7f2 TL |
5820 | |
5821 | r = _open_db(false, false, false); | |
5822 | if (r < 0) { | |
5823 | _close_alloc(); | |
5824 | _close_fm(); | |
5825 | return r; | |
28e407b8 | 5826 | } |
1911f103 | 5827 | fm->sync(db); |
7c673cae | 5828 | } |
11fdf7f2 TL |
5829 | } else { |
5830 | r = _open_db(false, false); | |
5831 | if (r < 0) { | |
5832 | return r; | |
5833 | } | |
5834 | r = _open_super_meta(); | |
5835 | if (r < 0) { | |
5836 | goto out_db; | |
5837 | } | |
7c673cae | 5838 | |
1911f103 | 5839 | r = _open_fm(nullptr, false); |
11fdf7f2 TL |
5840 | if (r < 0) |
5841 | goto out_db; | |
5842 | ||
5843 | r = _open_alloc(); | |
5844 | if (r < 0) | |
5845 | goto out_fm; | |
5846 | } | |
5847 | return 0; | |
5848 | ||
5849 | out_fm: | |
5850 | _close_fm(); | |
5851 | out_db: | |
1911f103 | 5852 | _close_db(read_only); |
11fdf7f2 TL |
5853 | return r; |
5854 | } | |
5855 | ||
1911f103 | 5856 | void BlueStore::_close_db_and_around(bool read_only) |
11fdf7f2 TL |
5857 | { |
5858 | if (bluefs) { | |
1911f103 | 5859 | if (!read_only && out_of_sync_fm.fetch_and(0)) { |
11fdf7f2 TL |
5860 | _sync_bluefs_and_fm(); |
5861 | } | |
1911f103 TL |
5862 | _close_db(read_only); |
5863 | while(!read_only && out_of_sync_fm.fetch_and(0)) { | |
11fdf7f2 TL |
5864 | // if seen some allocations during close - repeat open_db, sync fm, close |
5865 | dout(0) << __func__ << " syncing FreelistManager" << dendl; | |
5866 | int r = _open_db(false, false, false); | |
5867 | if (r < 0) { | |
5868 | derr << __func__ | |
5869 | << " unable to open db, FreelistManager is probably out of sync" | |
5870 | << dendl; | |
5871 | break; | |
5872 | } | |
5873 | _sync_bluefs_and_fm(); | |
1911f103 | 5874 | _close_db(false); |
7c673cae | 5875 | } |
11fdf7f2 TL |
5876 | if (!_kv_only) { |
5877 | _close_alloc(); | |
5878 | _close_fm(); | |
5879 | } | |
5880 | } else { | |
5881 | _close_alloc(); | |
5882 | _close_fm(); | |
1911f103 | 5883 | _close_db(read_only); |
11fdf7f2 TL |
5884 | } |
5885 | } | |
5886 | ||
5887 | // updates legacy bluefs related recs in DB to a state valid for | |
5888 | // downgrades from nautilus. | |
5889 | void BlueStore::_sync_bluefs_and_fm() | |
5890 | { | |
5891 | if (cct->_conf->bluestore_bluefs_db_compatibility) { | |
5892 | bufferlist bl; | |
5893 | encode(bluefs_extents, bl); | |
5894 | dout(20) << __func__ << " bluefs_extents at KV is now 0x" | |
5895 | << std::hex << bluefs_extents << std::dec | |
5896 | << dendl; | |
5897 | KeyValueDB::Transaction synct = db->get_transaction(); | |
5898 | synct->set(PREFIX_SUPER, "bluefs_extents", bl); | |
5899 | synct->set(PREFIX_SUPER, "bluefs_extents_back", bl); | |
5900 | ||
5901 | // Nice thing is that we don't need to update FreelistManager here. | |
5902 | // It always has corresponding bits set to 'Free' for both Nautilus+ and | |
5903 | // pre-Nautilis releases. | |
5904 | // So once we get an extent to bluefs_extents this means it's | |
5905 | // been free in allocator and hence it's free in FM too. | |
5906 | ||
5907 | db->submit_transaction_sync(synct); | |
5908 | } | |
5909 | } | |
5910 | ||
5911 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) | |
5912 | { | |
5913 | int r; | |
5914 | ceph_assert(!db); | |
5915 | ceph_assert(!(create && read_only)); | |
5916 | string fn = path + "/db"; | |
5917 | string options; | |
5918 | stringstream err; | |
5919 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); | |
5920 | ||
5921 | string kv_backend; | |
5922 | std::vector<KeyValueDB::ColumnFamily> cfs; | |
5923 | ||
5924 | if (create) { | |
5925 | kv_backend = cct->_conf->bluestore_kvbackend; | |
5926 | } else { | |
5927 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 5928 | if (r < 0) { |
11fdf7f2 TL |
5929 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
5930 | return -EIO; | |
5931 | } | |
5932 | } | |
5933 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
5934 | ||
5935 | bool do_bluefs; | |
5936 | r = _is_bluefs(create, &do_bluefs); | |
5937 | if (r < 0) { | |
5938 | return r; | |
5939 | } | |
5940 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
5941 | ||
5942 | map<string,string> kv_options; | |
5943 | // force separate wal dir for all new deployments. | |
5944 | kv_options["separate_wal_dir"] = 1; | |
5945 | rocksdb::Env *env = NULL; | |
5946 | if (do_bluefs) { | |
5947 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
5948 | if (kv_backend != "rocksdb") { | |
5949 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
5950 | return -EINVAL; | |
7c673cae | 5951 | } |
11fdf7f2 TL |
5952 | |
5953 | r = _open_bluefs(create); | |
5954 | if (r < 0) { | |
5955 | return r; | |
5956 | } | |
11fdf7f2 | 5957 | |
7c673cae | 5958 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
9f95a23c TL |
5959 | rocksdb::Env* a = new BlueRocksEnv(bluefs); |
5960 | rocksdb::Env* b = rocksdb::Env::Default(); | |
7c673cae | 5961 | if (create) { |
9f95a23c TL |
5962 | string cmd = "rm -rf " + path + "/db " + |
5963 | path + "/db.slow " + | |
5964 | path + "/db.wal"; | |
5965 | int r = system(cmd.c_str()); | |
5966 | (void)r; | |
7c673cae FG |
5967 | } |
5968 | env = new rocksdb::EnvMirror(b, a, false, true); | |
1911f103 | 5969 | } else { |
7c673cae FG |
5970 | env = new BlueRocksEnv(bluefs); |
5971 | ||
5972 | // simplify the dir names, too, as "seen" by rocksdb | |
5973 | fn = "db"; | |
5974 | } | |
9f95a23c TL |
5975 | bluefs->set_slow_device_expander(this); |
5976 | BlueFSVolumeSelector::paths paths; | |
5977 | bluefs->get_vselector_paths(fn, paths); | |
7c673cae | 5978 | |
9f95a23c | 5979 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
7c673cae FG |
5980 | // we have both block.db and block; tell rocksdb! |
5981 | // note: the second (last) size value doesn't really matter | |
5982 | ostringstream db_paths; | |
9f95a23c TL |
5983 | bool first = true; |
5984 | for (auto& p : paths) { | |
5985 | if (!first) { | |
5986 | db_paths << " "; | |
5987 | } | |
5988 | first = false; | |
5989 | db_paths << p.first << "," << p.second; | |
5990 | ||
5991 | } | |
11fdf7f2 | 5992 | kv_options["db_paths"] = db_paths.str(); |
9f95a23c | 5993 | dout(1) << __func__ << " set db_paths to " << db_paths.str() << dendl; |
7c673cae FG |
5994 | } |
5995 | ||
5996 | if (create) { | |
9f95a23c TL |
5997 | for (auto& p : paths) { |
5998 | env->CreateDir(p.first); | |
5999 | } | |
6000 | // Selectors don't provide wal path so far hence create explicitly | |
11fdf7f2 | 6001 | env->CreateDir(fn + ".wal"); |
11fdf7f2 TL |
6002 | } else { |
6003 | std::vector<std::string> res; | |
6004 | // check for dir presence | |
6005 | auto r = env->GetChildren(fn+".wal", &res); | |
6006 | if (r.IsNotFound()) { | |
6007 | kv_options.erase("separate_wal_dir"); | |
6008 | } | |
7c673cae | 6009 | } |
11fdf7f2 TL |
6010 | } else { |
6011 | string walfn = path + "/db.wal"; | |
7c673cae | 6012 | |
11fdf7f2 TL |
6013 | if (create) { |
6014 | int r = ::mkdir(fn.c_str(), 0755); | |
6015 | if (r < 0) | |
6016 | r = -errno; | |
6017 | if (r < 0 && r != -EEXIST) { | |
6018 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
6019 | << dendl; | |
6020 | return r; | |
6021 | } | |
6022 | ||
6023 | // wal_dir, too! | |
7c673cae FG |
6024 | r = ::mkdir(walfn.c_str(), 0755); |
6025 | if (r < 0) | |
6026 | r = -errno; | |
6027 | if (r < 0 && r != -EEXIST) { | |
6028 | derr << __func__ << " failed to create " << walfn | |
6029 | << ": " << cpp_strerror(r) | |
6030 | << dendl; | |
6031 | return r; | |
6032 | } | |
11fdf7f2 TL |
6033 | } else { |
6034 | struct stat st; | |
6035 | r = ::stat(walfn.c_str(), &st); | |
6036 | if (r < 0 && errno == ENOENT) { | |
6037 | kv_options.erase("separate_wal_dir"); | |
6038 | } | |
7c673cae FG |
6039 | } |
6040 | } | |
6041 | ||
91327a77 | 6042 | |
7c673cae FG |
6043 | db = KeyValueDB::create(cct, |
6044 | kv_backend, | |
6045 | fn, | |
11fdf7f2 | 6046 | kv_options, |
7c673cae FG |
6047 | static_cast<void*>(env)); |
6048 | if (!db) { | |
6049 | derr << __func__ << " error creating db" << dendl; | |
6050 | if (bluefs) { | |
1911f103 | 6051 | _close_bluefs(read_only); |
7c673cae FG |
6052 | } |
6053 | // delete env manually here since we can't depend on db to do this | |
6054 | // under this case | |
6055 | delete env; | |
6056 | env = NULL; | |
6057 | return -EIO; | |
6058 | } | |
6059 | ||
6060 | FreelistManager::setup_merge_operators(db); | |
6061 | db->set_merge_operator(PREFIX_STAT, merge_op); | |
91327a77 | 6062 | db->set_cache_size(cache_kv_ratio * cache_size); |
31f18b77 | 6063 | |
11fdf7f2 | 6064 | if (kv_backend == "rocksdb") { |
7c673cae | 6065 | options = cct->_conf->bluestore_rocksdb_options; |
11fdf7f2 TL |
6066 | |
6067 | map<string,string> cf_map; | |
6068 | cct->_conf.with_val<string>("bluestore_rocksdb_cfs", | |
6069 | get_str_map, | |
6070 | &cf_map, | |
6071 | " \t"); | |
6072 | for (auto& i : cf_map) { | |
6073 | dout(10) << "column family " << i.first << ": " << i.second << dendl; | |
6074 | cfs.push_back(KeyValueDB::ColumnFamily(i.first, i.second)); | |
6075 | } | |
6076 | } | |
6077 | ||
7c673cae | 6078 | db->init(options); |
11fdf7f2 TL |
6079 | if (to_repair_db) |
6080 | return 0; | |
6081 | if (create) { | |
6082 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { | |
6083 | r = db->create_and_open(err, cfs); | |
6084 | } else { | |
6085 | r = db->create_and_open(err); | |
6086 | } | |
6087 | } else { | |
6088 | // we pass in cf list here, but it is only used if the db already has | |
6089 | // column families created. | |
6090 | r = read_only ? | |
6091 | db->open_read_only(err, cfs) : | |
6092 | db->open(err, cfs); | |
6093 | } | |
7c673cae FG |
6094 | if (r) { |
6095 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
1911f103 | 6096 | _close_db(read_only); |
7c673cae FG |
6097 | return -EIO; |
6098 | } | |
6099 | dout(1) << __func__ << " opened " << kv_backend | |
6100 | << " path " << fn << " options " << options << dendl; | |
6101 | return 0; | |
7c673cae FG |
6102 | } |
6103 | ||
1911f103 | 6104 | void BlueStore::_close_db(bool cold_close) |
7c673cae | 6105 | { |
11fdf7f2 | 6106 | ceph_assert(db); |
7c673cae FG |
6107 | delete db; |
6108 | db = NULL; | |
6109 | if (bluefs) { | |
1911f103 | 6110 | _close_bluefs(cold_close); |
7c673cae FG |
6111 | } |
6112 | } | |
6113 | ||
11fdf7f2 | 6114 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 6115 | { |
11fdf7f2 TL |
6116 | auto dump_interval = |
6117 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
6118 | if (dump_interval > 0 && | |
6119 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
6120 | alloc->dump(); | |
6121 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); | |
6122 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 6123 | } |
11fdf7f2 | 6124 | } |
7c673cae | 6125 | |
7c673cae | 6126 | |
11fdf7f2 TL |
6127 | int BlueStore::allocate_bluefs_freespace( |
6128 | uint64_t min_size, | |
6129 | uint64_t size, | |
6130 | PExtentVector* extents_out) | |
6131 | { | |
6132 | ceph_assert(min_size <= size); | |
6133 | if (size) { | |
6134 | // round up to alloc size | |
9f95a23c | 6135 | uint64_t alloc_size = bluefs->get_alloc_size(bluefs_layout.shared_bdev); |
eafe8130 TL |
6136 | min_size = p2roundup(min_size, alloc_size); |
6137 | size = p2roundup(size, alloc_size); | |
11fdf7f2 TL |
6138 | |
6139 | PExtentVector extents_local; | |
6140 | PExtentVector* extents = extents_out ? extents_out : &extents_local; | |
6141 | ||
6142 | ||
6143 | uint64_t gift; | |
6144 | uint64_t allocated = 0; | |
6145 | int64_t alloc_len; | |
9f95a23c TL |
6146 | auto need = size; |
6147 | auto extent_count0 = extents->size(); | |
11fdf7f2 TL |
6148 | do { |
6149 | // hard cap to fit into 32 bits | |
9f95a23c | 6150 | gift = std::min<uint64_t>(size, 1ull << 30); |
11fdf7f2 TL |
6151 | dout(10) << __func__ << " gifting " << gift |
6152 | << " (" << byte_u_t(gift) << ")" << dendl; | |
6153 | ||
eafe8130 TL |
6154 | alloc_len = alloc->allocate(gift, alloc_size, 0, 0, extents); |
6155 | if (alloc_len > 0) { | |
11fdf7f2 TL |
6156 | allocated += alloc_len; |
6157 | size -= alloc_len; | |
6158 | } | |
6159 | ||
eafe8130 TL |
6160 | if (alloc_len < 0 || |
6161 | (alloc_len < (int64_t)gift && (min_size > allocated))) { | |
11fdf7f2 TL |
6162 | derr << __func__ |
6163 | << " failed to allocate on 0x" << std::hex << gift | |
6164 | << " min_size 0x" << min_size | |
6165 | << " > allocated total 0x" << allocated | |
eafe8130 TL |
6166 | << " bluefs_shared_alloc_size 0x" << alloc_size |
6167 | << " allocated 0x" << (alloc_len < 0 ? 0 : alloc_len) | |
11fdf7f2 TL |
6168 | << " available 0x " << alloc->get_free() |
6169 | << std::dec << dendl; | |
7c673cae | 6170 | |
494da23a | 6171 | _dump_alloc_on_failure(); |
11fdf7f2 TL |
6172 | alloc->release(*extents); |
6173 | extents->clear(); | |
6174 | return -ENOSPC; | |
6175 | } | |
6176 | } while (size && alloc_len > 0); | |
9f95a23c TL |
6177 | _collect_allocation_stats(need, alloc_size, extents->size() - extent_count0); |
6178 | ||
11fdf7f2 TL |
6179 | for (auto& e : *extents) { |
6180 | dout(5) << __func__ << " gifting " << e << " to bluefs" << dendl; | |
6181 | bluefs_extents.insert(e.offset, e.length); | |
6182 | ++out_of_sync_fm; | |
6183 | // apply to bluefs if not requested from outside | |
6184 | if (!extents_out) { | |
9f95a23c | 6185 | bluefs->add_block_extent(bluefs_layout.shared_bdev, e.offset, e.length); |
11fdf7f2 | 6186 | } |
7c673cae FG |
6187 | } |
6188 | } | |
7c673cae FG |
6189 | return 0; |
6190 | } | |
6191 | ||
9f95a23c TL |
6192 | uint64_t BlueStore::available_freespace(uint64_t alloc_size) { |
6193 | uint64_t total = 0; | |
6194 | auto iterated_allocation = [&](uint64_t off, uint64_t len) { | |
eafe8130 | 6195 | //only count in size that is alloc_size aligned |
9f95a23c TL |
6196 | uint64_t dist_to_alignment; |
6197 | uint64_t offset_in_block = off & (alloc_size - 1); | |
eafe8130 TL |
6198 | if (offset_in_block == 0) |
6199 | dist_to_alignment = 0; | |
6200 | else | |
6201 | dist_to_alignment = alloc_size - offset_in_block; | |
6202 | if (dist_to_alignment >= len) | |
6203 | return; | |
6204 | len -= dist_to_alignment; | |
6205 | total += p2align(len, alloc_size); | |
6206 | }; | |
6207 | alloc->dump(iterated_allocation); | |
6208 | return total; | |
6209 | } | |
6210 | ||
11fdf7f2 | 6211 | int64_t BlueStore::_get_bluefs_size_delta(uint64_t bluefs_free, uint64_t bluefs_total) |
f64942e4 | 6212 | { |
7c673cae FG |
6213 | float bluefs_free_ratio = (float)bluefs_free / (float)bluefs_total; |
6214 | ||
6215 | uint64_t my_free = alloc->get_free(); | |
6216 | uint64_t total = bdev->get_size(); | |
6217 | float my_free_ratio = (float)my_free / (float)total; | |
6218 | ||
6219 | uint64_t total_free = bluefs_free + my_free; | |
6220 | ||
6221 | float bluefs_ratio = (float)bluefs_free / (float)total_free; | |
6222 | ||
6223 | dout(10) << __func__ | |
1adf2230 | 6224 | << " bluefs " << byte_u_t(bluefs_free) |
7c673cae | 6225 | << " free (" << bluefs_free_ratio |
1adf2230 | 6226 | << ") bluestore " << byte_u_t(my_free) |
7c673cae FG |
6227 | << " free (" << my_free_ratio |
6228 | << "), bluefs_ratio " << bluefs_ratio | |
6229 | << dendl; | |
6230 | ||
6231 | uint64_t gift = 0; | |
6232 | uint64_t reclaim = 0; | |
6233 | if (bluefs_ratio < cct->_conf->bluestore_bluefs_min_ratio) { | |
6234 | gift = cct->_conf->bluestore_bluefs_gift_ratio * total_free; | |
9f95a23c TL |
6235 | if (gift >= my_free) |
6236 | gift = my_free / 2; | |
7c673cae FG |
6237 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio |
6238 | << " < min_ratio " << cct->_conf->bluestore_bluefs_min_ratio | |
1adf2230 | 6239 | << ", should gift " << byte_u_t(gift) << dendl; |
7c673cae FG |
6240 | } else if (bluefs_ratio > cct->_conf->bluestore_bluefs_max_ratio) { |
6241 | reclaim = cct->_conf->bluestore_bluefs_reclaim_ratio * total_free; | |
6242 | if (bluefs_total - reclaim < cct->_conf->bluestore_bluefs_min) | |
6243 | reclaim = bluefs_total - cct->_conf->bluestore_bluefs_min; | |
9f95a23c TL |
6244 | if (reclaim >= bluefs_free) |
6245 | reclaim = bluefs_free / 2; | |
7c673cae FG |
6246 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio |
6247 | << " > max_ratio " << cct->_conf->bluestore_bluefs_max_ratio | |
1adf2230 | 6248 | << ", should reclaim " << byte_u_t(reclaim) << dendl; |
7c673cae | 6249 | } |
3efd9988 FG |
6250 | |
6251 | // don't take over too much of the freespace | |
6252 | uint64_t free_cap = cct->_conf->bluestore_bluefs_max_ratio * total_free; | |
7c673cae | 6253 | if (bluefs_total < cct->_conf->bluestore_bluefs_min && |
3efd9988 | 6254 | cct->_conf->bluestore_bluefs_min < free_cap) { |
7c673cae FG |
6255 | uint64_t g = cct->_conf->bluestore_bluefs_min - bluefs_total; |
6256 | dout(10) << __func__ << " bluefs_total " << bluefs_total | |
6257 | << " < min " << cct->_conf->bluestore_bluefs_min | |
1adf2230 | 6258 | << ", should gift " << byte_u_t(g) << dendl; |
7c673cae FG |
6259 | if (g > gift) |
6260 | gift = g; | |
6261 | reclaim = 0; | |
6262 | } | |
9f95a23c TL |
6263 | uint64_t min_free = |
6264 | cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
3efd9988 FG |
6265 | if (bluefs_free < min_free && |
6266 | min_free < free_cap) { | |
6267 | uint64_t g = min_free - bluefs_free; | |
11fdf7f2 | 6268 | dout(10) << __func__ << " bluefs_free " << bluefs_free |
3efd9988 | 6269 | << " < min " << min_free |
1adf2230 | 6270 | << ", should gift " << byte_u_t(g) << dendl; |
3efd9988 FG |
6271 | if (g > gift) |
6272 | gift = g; | |
6273 | reclaim = 0; | |
6274 | } | |
9f95a23c TL |
6275 | uint64_t max_free = |
6276 | cct->_conf.get_val<Option::size_t>("bluestore_bluefs_max_free"); | |
6277 | if (bluefs_free > max_free) { | |
6278 | dout(10) << __func__ << " bluefs_free " << bluefs_free | |
6279 | << " > max " << max_free | |
6280 | << ", stop gifting for now" << dendl; | |
6281 | gift = 0; | |
6282 | } | |
11fdf7f2 TL |
6283 | ceph_assert((int64_t)gift >= 0); |
6284 | ceph_assert((int64_t)reclaim >= 0); | |
6285 | return gift > 0 ? (int64_t)gift : -(int64_t)reclaim; | |
6286 | } | |
7c673cae | 6287 | |
11fdf7f2 TL |
6288 | int BlueStore::_balance_bluefs_freespace() |
6289 | { | |
6290 | int ret = 0; | |
6291 | ceph_assert(bluefs); | |
7c673cae | 6292 | |
11fdf7f2 TL |
6293 | vector<pair<uint64_t,uint64_t>> bluefs_usage; // <free, total> ... |
6294 | bluefs->get_usage(&bluefs_usage); | |
9f95a23c | 6295 | ceph_assert(bluefs_usage.size() > bluefs_layout.shared_bdev); |
7c673cae | 6296 | |
11fdf7f2 | 6297 | bool clear_alert = true; |
9f95a23c TL |
6298 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
6299 | auto& p = bluefs_usage[bluefs_layout.shared_bdev]; | |
11fdf7f2 TL |
6300 | if (p.first != p.second) { |
6301 | auto& db = bluefs_usage[BlueFS::BDEV_DB]; | |
6302 | ostringstream ss; | |
6303 | ss << "spilled over " << byte_u_t(p.second - p.first) | |
6304 | << " metadata from 'db' device (" << byte_u_t(db.second - db.first) | |
6305 | << " used of " << byte_u_t(db.second) << ") to slow device"; | |
6306 | _set_spillover_alert(ss.str()); | |
6307 | clear_alert = false; | |
6308 | } | |
6309 | } | |
6310 | if (clear_alert) { | |
6311 | _clear_spillover_alert(); | |
7c673cae FG |
6312 | } |
6313 | ||
11fdf7f2 TL |
6314 | // fixme: look at primary bdev only for now |
6315 | int64_t delta = _get_bluefs_size_delta( | |
9f95a23c TL |
6316 | bluefs_usage[bluefs_layout.shared_bdev].first, |
6317 | bluefs_usage[bluefs_layout.shared_bdev].second); | |
11fdf7f2 | 6318 | |
7c673cae | 6319 | // reclaim from bluefs? |
11fdf7f2 | 6320 | if (delta < 0) { |
7c673cae | 6321 | // round up to alloc size |
9f95a23c | 6322 | uint64_t alloc_size = bluefs->get_alloc_size(bluefs_layout.shared_bdev); |
eafe8130 | 6323 | auto reclaim = p2roundup(uint64_t(-delta), alloc_size); |
7c673cae FG |
6324 | |
6325 | // hard cap to fit into 32 bits | |
9f95a23c | 6326 | reclaim = std::min<uint64_t>(reclaim, 1ull << 30); |
7c673cae | 6327 | dout(10) << __func__ << " reclaiming " << reclaim |
1adf2230 | 6328 | << " (" << byte_u_t(reclaim) << ")" << dendl; |
7c673cae FG |
6329 | |
6330 | while (reclaim > 0) { | |
6331 | // NOTE: this will block and do IO. | |
a8e16298 | 6332 | PExtentVector extents; |
9f95a23c | 6333 | int r = bluefs->reclaim_blocks(bluefs_layout.shared_bdev, reclaim, |
7c673cae FG |
6334 | &extents); |
6335 | if (r < 0) { | |
6336 | derr << __func__ << " failed to reclaim space from bluefs" | |
6337 | << dendl; | |
6338 | break; | |
6339 | } | |
6340 | for (auto e : extents) { | |
11fdf7f2 | 6341 | ++out_of_sync_fm; |
7c673cae FG |
6342 | bluefs_extents.erase(e.offset, e.length); |
6343 | bluefs_extents_reclaiming.insert(e.offset, e.length); | |
6344 | reclaim -= e.length; | |
6345 | } | |
6346 | } | |
6347 | ||
6348 | ret = 1; | |
6349 | } | |
6350 | ||
6351 | return ret; | |
6352 | } | |
6353 | ||
eafe8130 | 6354 | int BlueStore::_open_collections() |
7c673cae | 6355 | { |
28e407b8 | 6356 | dout(10) << __func__ << dendl; |
eafe8130 | 6357 | collections_had_errors = false; |
11fdf7f2 | 6358 | ceph_assert(coll_map.empty()); |
7c673cae FG |
6359 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
6360 | for (it->upper_bound(string()); | |
6361 | it->valid(); | |
6362 | it->next()) { | |
6363 | coll_t cid; | |
6364 | if (cid.parse(it->key())) { | |
9f95a23c | 6365 | auto c = ceph::make_ref<Collection>( |
7c673cae | 6366 | this, |
9f95a23c TL |
6367 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
6368 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
6369 | cid); | |
7c673cae | 6370 | bufferlist bl = it->value(); |
11fdf7f2 | 6371 | auto p = bl.cbegin(); |
7c673cae | 6372 | try { |
11fdf7f2 | 6373 | decode(c->cnode, p); |
7c673cae FG |
6374 | } catch (buffer::error& e) { |
6375 | derr << __func__ << " failed to decode cnode, key:" | |
6376 | << pretty_binary_string(it->key()) << dendl; | |
6377 | return -EIO; | |
6378 | } | |
28e407b8 AA |
6379 | dout(20) << __func__ << " opened " << cid << " " << c |
6380 | << " " << c->cnode << dendl; | |
11fdf7f2 | 6381 | _osr_attach(c.get()); |
7c673cae | 6382 | coll_map[cid] = c; |
11fdf7f2 | 6383 | |
7c673cae FG |
6384 | } else { |
6385 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
eafe8130 | 6386 | collections_had_errors = true; |
7c673cae FG |
6387 | } |
6388 | } | |
6389 | return 0; | |
6390 | } | |
6391 | ||
eafe8130 TL |
6392 | void BlueStore::_fsck_collections(int64_t* errors) |
6393 | { | |
6394 | if (collections_had_errors) { | |
6395 | dout(10) << __func__ << dendl; | |
6396 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); | |
6397 | for (it->upper_bound(string()); | |
6398 | it->valid(); | |
6399 | it->next()) { | |
6400 | coll_t cid; | |
6401 | if (!cid.parse(it->key())) { | |
6402 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
6403 | if (errors) { | |
6404 | (*errors)++; | |
6405 | } | |
6406 | } | |
6407 | } | |
6408 | } | |
6409 | } | |
6410 | ||
9f95a23c TL |
6411 | void BlueStore::_set_per_pool_omap() |
6412 | { | |
6413 | per_pool_omap = false; | |
6414 | bufferlist bl; | |
6415 | db->get(PREFIX_SUPER, "per_pool_omap", &bl); | |
6416 | if (bl.length()) { | |
6417 | per_pool_omap = true; | |
6418 | dout(10) << __func__ << " per_pool_omap=1" << dendl; | |
6419 | } else { | |
6420 | dout(10) << __func__ << " per_pool_omap not present" << dendl; | |
6421 | } | |
6422 | _check_no_per_pool_omap_alert(); | |
6423 | } | |
6424 | ||
224ce89b | 6425 | void BlueStore::_open_statfs() |
31f18b77 | 6426 | { |
11fdf7f2 TL |
6427 | osd_pools.clear(); |
6428 | vstatfs.reset(); | |
6429 | ||
31f18b77 | 6430 | bufferlist bl; |
11fdf7f2 | 6431 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 6432 | if (r >= 0) { |
11fdf7f2 | 6433 | per_pool_stat_collection = false; |
31f18b77 | 6434 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 6435 | auto it = bl.cbegin(); |
31f18b77 | 6436 | vstatfs.decode(it); |
11fdf7f2 | 6437 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 6438 | } else { |
31f18b77 FG |
6439 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
6440 | } | |
81eedcae | 6441 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
6442 | } else { |
6443 | per_pool_stat_collection = true; | |
6444 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
6445 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT); | |
6446 | for (it->upper_bound(string()); | |
6447 | it->valid(); | |
6448 | it->next()) { | |
6449 | ||
6450 | uint64_t pool_id; | |
6451 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6452 | ceph_assert(r == 0); | |
6453 | ||
6454 | bufferlist bl; | |
6455 | bl = it->value(); | |
6456 | auto p = bl.cbegin(); | |
6457 | auto& st = osd_pools[pool_id]; | |
6458 | try { | |
6459 | st.decode(p); | |
6460 | vstatfs += st; | |
6461 | ||
6462 | dout(30) << __func__ << " pool " << pool_id | |
6463 | << " statfs " << st << dendl; | |
6464 | } catch (buffer::error& e) { | |
6465 | derr << __func__ << " failed to decode pool stats, key:" | |
6466 | << pretty_binary_string(it->key()) << dendl; | |
6467 | } | |
6468 | } | |
31f18b77 | 6469 | } |
11fdf7f2 TL |
6470 | dout(30) << __func__ << " statfs " << vstatfs << dendl; |
6471 | ||
31f18b77 FG |
6472 | } |
6473 | ||
7c673cae FG |
6474 | int BlueStore::_setup_block_symlink_or_file( |
6475 | string name, | |
6476 | string epath, | |
6477 | uint64_t size, | |
6478 | bool create) | |
6479 | { | |
6480 | dout(20) << __func__ << " name " << name << " path " << epath | |
6481 | << " size " << size << " create=" << (int)create << dendl; | |
6482 | int r = 0; | |
91327a77 | 6483 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6484 | if (create) |
6485 | flags |= O_CREAT; | |
6486 | if (epath.length()) { | |
6487 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
6488 | if (r < 0) { | |
6489 | r = -errno; | |
6490 | derr << __func__ << " failed to create " << name << " symlink to " | |
6491 | << epath << ": " << cpp_strerror(r) << dendl; | |
6492 | return r; | |
6493 | } | |
6494 | ||
6495 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
6496 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
6497 | if (fd < 0) { | |
6498 | r = -errno; | |
6499 | derr << __func__ << " failed to open " << epath << " file: " | |
6500 | << cpp_strerror(r) << dendl; | |
6501 | return r; | |
6502 | } | |
11fdf7f2 TL |
6503 | // write the Transport ID of the NVMe device |
6504 | // a transport id looks like: "trtype:PCIe traddr:0000:02:00.0" | |
6505 | // where "0000:02:00.0" is the selector of a PCI device, see | |
6506 | // the first column of "lspci -mm -n -D" | |
6507 | string trid{"trtype:PCIe "}; | |
6508 | trid += "traddr:"; | |
6509 | trid += epath.substr(strlen(SPDK_PREFIX)); | |
6510 | r = ::write(fd, trid.c_str(), trid.size()); | |
6511 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
6512 | dout(1) << __func__ << " created " << name << " symlink to " |
6513 | << epath << dendl; | |
6514 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6515 | } | |
6516 | } | |
6517 | if (size) { | |
6518 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
6519 | if (fd >= 0) { | |
6520 | // block file is present | |
6521 | struct stat st; | |
6522 | int r = ::fstat(fd, &st); | |
6523 | if (r == 0 && | |
6524 | S_ISREG(st.st_mode) && // if it is a regular file | |
6525 | st.st_size == 0) { // and is 0 bytes | |
6526 | r = ::ftruncate(fd, size); | |
6527 | if (r < 0) { | |
6528 | r = -errno; | |
6529 | derr << __func__ << " failed to resize " << name << " file to " | |
6530 | << size << ": " << cpp_strerror(r) << dendl; | |
6531 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6532 | return r; | |
6533 | } | |
6534 | ||
6535 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
6536 | r = ::ceph_posix_fallocate(fd, 0, size); |
6537 | if (r > 0) { | |
7c673cae FG |
6538 | derr << __func__ << " failed to prefallocate " << name << " file to " |
6539 | << size << ": " << cpp_strerror(r) << dendl; | |
6540 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6541 | return -r; | |
6542 | } | |
7c673cae FG |
6543 | } |
6544 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 6545 | << byte_u_t(size) << dendl; |
7c673cae FG |
6546 | } |
6547 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6548 | } else { | |
6549 | int r = -errno; | |
6550 | if (r != -ENOENT) { | |
6551 | derr << __func__ << " failed to open " << name << " file: " | |
6552 | << cpp_strerror(r) << dendl; | |
6553 | return r; | |
6554 | } | |
6555 | } | |
6556 | } | |
6557 | return 0; | |
6558 | } | |
6559 | ||
6560 | int BlueStore::mkfs() | |
6561 | { | |
6562 | dout(1) << __func__ << " path " << path << dendl; | |
6563 | int r; | |
6564 | uuid_d old_fsid; | |
6565 | ||
eafe8130 TL |
6566 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6567 | derr << __func__ << " osd_max_object_size " | |
6568 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6569 | << OBJECT_MAX_SIZE << dendl; | |
6570 | return -EINVAL; | |
6571 | } | |
6572 | ||
7c673cae FG |
6573 | { |
6574 | string done; | |
6575 | r = read_meta("mkfs_done", &done); | |
6576 | if (r == 0) { | |
6577 | dout(1) << __func__ << " already created" << dendl; | |
6578 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
6579 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6580 | if (r < 0) { | |
6581 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
6582 | << dendl; | |
6583 | return r; | |
6584 | } | |
6585 | if (r > 0) { | |
6586 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
6587 | r = -EIO; | |
6588 | } | |
6589 | } | |
6590 | return r; // idempotent | |
6591 | } | |
6592 | } | |
6593 | ||
6594 | { | |
6595 | string type; | |
6596 | r = read_meta("type", &type); | |
6597 | if (r == 0) { | |
6598 | if (type != "bluestore") { | |
6599 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6600 | return -EIO; | |
6601 | } | |
6602 | } else { | |
6603 | r = write_meta("type", "bluestore"); | |
6604 | if (r < 0) | |
6605 | return r; | |
6606 | } | |
6607 | } | |
6608 | ||
6609 | freelist_type = "bitmap"; | |
6610 | ||
6611 | r = _open_path(); | |
6612 | if (r < 0) | |
6613 | return r; | |
6614 | ||
6615 | r = _open_fsid(true); | |
6616 | if (r < 0) | |
6617 | goto out_path_fd; | |
6618 | ||
6619 | r = _lock_fsid(); | |
6620 | if (r < 0) | |
6621 | goto out_close_fsid; | |
6622 | ||
6623 | r = _read_fsid(&old_fsid); | |
6624 | if (r < 0 || old_fsid.is_zero()) { | |
6625 | if (fsid.is_zero()) { | |
6626 | fsid.generate_random(); | |
6627 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
6628 | } else { | |
6629 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
6630 | } | |
6631 | // we'll write it later. | |
6632 | } else { | |
6633 | if (!fsid.is_zero() && fsid != old_fsid) { | |
6634 | derr << __func__ << " on-disk fsid " << old_fsid | |
6635 | << " != provided " << fsid << dendl; | |
6636 | r = -EINVAL; | |
6637 | goto out_close_fsid; | |
6638 | } | |
6639 | fsid = old_fsid; | |
6640 | } | |
6641 | ||
6642 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
6643 | cct->_conf->bluestore_block_size, | |
6644 | cct->_conf->bluestore_block_create); | |
6645 | if (r < 0) | |
6646 | goto out_close_fsid; | |
6647 | if (cct->_conf->bluestore_bluefs) { | |
6648 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
6649 | cct->_conf->bluestore_block_wal_size, | |
6650 | cct->_conf->bluestore_block_wal_create); | |
6651 | if (r < 0) | |
6652 | goto out_close_fsid; | |
6653 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
6654 | cct->_conf->bluestore_block_db_size, | |
6655 | cct->_conf->bluestore_block_db_create); | |
6656 | if (r < 0) | |
6657 | goto out_close_fsid; | |
6658 | } | |
6659 | ||
6660 | r = _open_bdev(true); | |
6661 | if (r < 0) | |
6662 | goto out_close_fsid; | |
6663 | ||
3efd9988 FG |
6664 | // choose min_alloc_size |
6665 | if (cct->_conf->bluestore_min_alloc_size) { | |
6666 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; | |
6667 | } else { | |
11fdf7f2 | 6668 | ceph_assert(bdev); |
3efd9988 FG |
6669 | if (bdev->is_rotational()) { |
6670 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; | |
6671 | } else { | |
6672 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
6673 | } | |
6674 | } | |
11fdf7f2 | 6675 | _validate_bdev(); |
3efd9988 FG |
6676 | |
6677 | // make sure min_alloc_size is power of 2 aligned. | |
11fdf7f2 | 6678 | if (!isp2(min_alloc_size)) { |
3efd9988 FG |
6679 | derr << __func__ << " min_alloc_size 0x" |
6680 | << std::hex << min_alloc_size << std::dec | |
6681 | << " is not power of 2 aligned!" | |
6682 | << dendl; | |
6683 | r = -EINVAL; | |
6684 | goto out_close_bdev; | |
6685 | } | |
6686 | ||
7c673cae FG |
6687 | r = _open_db(true); |
6688 | if (r < 0) | |
6689 | goto out_close_bdev; | |
6690 | ||
7c673cae FG |
6691 | { |
6692 | KeyValueDB::Transaction t = db->get_transaction(); | |
1911f103 | 6693 | r = _open_fm(t, true); |
11fdf7f2 TL |
6694 | if (r < 0) |
6695 | goto out_close_db; | |
7c673cae FG |
6696 | { |
6697 | bufferlist bl; | |
11fdf7f2 | 6698 | encode((uint64_t)0, bl); |
7c673cae FG |
6699 | t->set(PREFIX_SUPER, "nid_max", bl); |
6700 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
6701 | } | |
6702 | ||
7c673cae FG |
6703 | { |
6704 | bufferlist bl; | |
11fdf7f2 | 6705 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
6706 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
6707 | } | |
9f95a23c TL |
6708 | { |
6709 | bufferlist bl; | |
6710 | bl.append("1"); | |
6711 | t->set(PREFIX_SUPER, "per_pool_omap", bl); | |
6712 | } | |
7c673cae FG |
6713 | ondisk_format = latest_ondisk_format; |
6714 | _prepare_ondisk_format_super(t); | |
6715 | db->submit_transaction_sync(t); | |
6716 | } | |
6717 | ||
7c673cae FG |
6718 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
6719 | if (r < 0) | |
224ce89b WB |
6720 | goto out_close_fm; |
6721 | ||
3efd9988 | 6722 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 6723 | if (r < 0) |
224ce89b | 6724 | goto out_close_fm; |
7c673cae FG |
6725 | |
6726 | if (fsid != old_fsid) { | |
6727 | r = _write_fsid(); | |
6728 | if (r < 0) { | |
6729 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 6730 | goto out_close_fm; |
7c673cae FG |
6731 | } |
6732 | } | |
6733 | ||
11fdf7f2 TL |
6734 | if (out_of_sync_fm.fetch_and(0)) { |
6735 | _sync_bluefs_and_fm(); | |
6736 | } | |
6737 | ||
7c673cae FG |
6738 | out_close_fm: |
6739 | _close_fm(); | |
6740 | out_close_db: | |
1911f103 | 6741 | _close_db(false); |
7c673cae FG |
6742 | out_close_bdev: |
6743 | _close_bdev(); | |
6744 | out_close_fsid: | |
6745 | _close_fsid(); | |
6746 | out_path_fd: | |
6747 | _close_path(); | |
6748 | ||
6749 | if (r == 0 && | |
6750 | cct->_conf->bluestore_fsck_on_mkfs) { | |
6751 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6752 | if (rc < 0) | |
6753 | return rc; | |
6754 | if (rc > 0) { | |
6755 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6756 | r = -EIO; | |
6757 | } | |
11fdf7f2 TL |
6758 | } |
6759 | ||
6760 | if (r == 0) { | |
6761 | // indicate success by writing the 'mkfs_done' file | |
6762 | r = write_meta("mkfs_done", "yes"); | |
6763 | } | |
6764 | ||
6765 | if (r < 0) { | |
6766 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6767 | } else { | |
6768 | dout(0) << __func__ << " success" << dendl; | |
6769 | } | |
6770 | return r; | |
6771 | } | |
6772 | ||
6773 | int BlueStore::_mount_for_bluefs() | |
6774 | { | |
6775 | int r = _open_path(); | |
6776 | ceph_assert(r == 0); | |
6777 | r = _open_fsid(false); | |
6778 | ceph_assert(r == 0); | |
6779 | r = _read_fsid(&fsid); | |
6780 | ceph_assert(r == 0); | |
6781 | r = _lock_fsid(); | |
6782 | ceph_assert(r == 0); | |
6783 | r = _open_bluefs(false); | |
6784 | ceph_assert(r == 0); | |
6785 | return r; | |
6786 | } | |
6787 | ||
6788 | void BlueStore::_umount_for_bluefs() | |
6789 | { | |
1911f103 | 6790 | _close_bluefs(false); |
11fdf7f2 TL |
6791 | _close_fsid(); |
6792 | _close_path(); | |
6793 | } | |
6794 | ||
6795 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) | |
6796 | { | |
6797 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6798 | int r; | |
6799 | ceph_assert(path_fd < 0); | |
6800 | ||
6801 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6802 | ||
6803 | if (!cct->_conf->bluestore_bluefs) { | |
6804 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6805 | return -EIO; | |
6806 | } | |
6807 | ||
6808 | r = _mount_for_bluefs(); | |
6809 | ||
6810 | int reserved = 0; | |
6811 | if (id == BlueFS::BDEV_NEWWAL) { | |
6812 | string p = path + "/block.wal"; | |
6813 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
6814 | cct->_conf->bluestore_block_wal_size, | |
6815 | true); | |
6816 | ceph_assert(r == 0); | |
6817 | ||
6818 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
eafe8130 | 6819 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6820 | ceph_assert(r == 0); |
6821 | ||
6822 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6823 | r = _check_or_set_bdev_label( | |
6824 | p, | |
6825 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6826 | "bluefs wal", | |
6827 | true); | |
6828 | ceph_assert(r == 0); | |
6829 | } | |
6830 | ||
6831 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
9f95a23c | 6832 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6833 | } else if (id == BlueFS::BDEV_NEWDB) { |
6834 | string p = path + "/block.db"; | |
6835 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
6836 | cct->_conf->bluestore_block_db_size, | |
6837 | true); | |
6838 | ceph_assert(r == 0); | |
6839 | ||
6840 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
eafe8130 | 6841 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6842 | ceph_assert(r == 0); |
6843 | ||
6844 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6845 | r = _check_or_set_bdev_label( | |
6846 | p, | |
6847 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6848 | "bluefs db", | |
6849 | true); | |
6850 | ceph_assert(r == 0); | |
6851 | } | |
6852 | reserved = SUPER_RESERVED; | |
9f95a23c TL |
6853 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6854 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
6855 | } |
6856 | ||
6857 | bluefs->umount(); | |
6858 | bluefs->mount(); | |
6859 | ||
6860 | bluefs->add_block_extent( | |
6861 | id, | |
6862 | reserved, | |
1911f103 | 6863 | bluefs->get_block_device_size(id) - reserved, true); |
11fdf7f2 | 6864 | |
9f95a23c | 6865 | r = bluefs->prepare_new_device(id, bluefs_layout); |
11fdf7f2 TL |
6866 | ceph_assert(r == 0); |
6867 | ||
6868 | if (r < 0) { | |
6869 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6870 | } else { | |
6871 | dout(0) << __func__ << " success" << dendl; | |
6872 | } | |
6873 | ||
6874 | _umount_for_bluefs(); | |
6875 | return r; | |
6876 | } | |
6877 | ||
6878 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
6879 | int id) | |
6880 | { | |
6881 | dout(10) << __func__ << " id:" << id << dendl; | |
6882 | ceph_assert(path_fd < 0); | |
6883 | ||
6884 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
6885 | ||
6886 | if (!cct->_conf->bluestore_bluefs) { | |
6887 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6888 | return -EIO; | |
6889 | } | |
6890 | ||
6891 | int r = _mount_for_bluefs(); | |
6892 | ||
6893 | // require bluestore_bluefs_min_free to be free at target device! | |
6894 | uint64_t used_space = cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
6895 | for(auto src_id : devs_source) { | |
6896 | used_space += bluefs->get_total(src_id) - bluefs->get_free(src_id); | |
6897 | } | |
6898 | uint64_t target_free = bluefs->get_free(id); | |
6899 | if (id == BlueFS::BDEV_SLOW && target_free < used_space) { | |
6900 | // will need to remount full BlueStore instance to allocate more space | |
6901 | _umount_for_bluefs(); | |
6902 | ||
6903 | r = mount(); | |
6904 | ceph_assert(r == 0); | |
6905 | dout(1) << __func__ | |
6906 | << " Allocating more space at slow device for BlueFS: +" | |
6907 | << used_space - target_free << " bytes" << dendl; | |
6908 | r = allocate_bluefs_freespace( | |
6909 | used_space - target_free, | |
6910 | used_space - target_free, | |
6911 | nullptr); | |
6912 | ||
6913 | umount(); | |
6914 | if (r != 0) { | |
6915 | derr << __func__ | |
6916 | << " can't migrate, unable to allocate extra space: " | |
6917 | << used_space - target_free << " at target:" << id | |
6918 | << dendl; | |
6919 | return -ENOSPC; | |
6920 | } | |
6921 | ||
6922 | r = _mount_for_bluefs(); | |
6923 | ceph_assert(r == 0); | |
6924 | } else if (target_free < used_space) { | |
6925 | derr << __func__ | |
6926 | << " can't migrate, free space at target: " << target_free | |
6927 | << " is less than required space: " << used_space | |
6928 | << dendl; | |
6929 | return -ENOSPC; | |
6930 | } | |
9f95a23c TL |
6931 | if (devs_source.count(BlueFS::BDEV_DB)) { |
6932 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; | |
6933 | bluefs_layout.dedicated_db = false; | |
6934 | } | |
6935 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6936 | bluefs_layout.dedicated_wal = false; | |
6937 | } | |
6938 | r = bluefs->device_migrate_to_existing(cct, devs_source, id, bluefs_layout); | |
11fdf7f2 TL |
6939 | if (r < 0) { |
6940 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
6941 | goto shutdown; | |
6942 | } | |
6943 | ||
6944 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
6945 | r = unlink(string(path + "/block.db").c_str()); | |
6946 | ceph_assert(r == 0); | |
6947 | } | |
6948 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6949 | r = unlink(string(path + "/block.wal").c_str()); | |
6950 | ceph_assert(r == 0); | |
6951 | } | |
6952 | ||
6953 | shutdown: | |
6954 | _umount_for_bluefs(); | |
6955 | return r; | |
6956 | } | |
6957 | ||
6958 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
6959 | int id, | |
6960 | const string& dev_path) | |
6961 | { | |
6962 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6963 | int r; | |
6964 | ceph_assert(path_fd < 0); | |
6965 | ||
6966 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6967 | ||
6968 | if (!cct->_conf->bluestore_bluefs) { | |
6969 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6970 | return -EIO; | |
6971 | } | |
6972 | ||
6973 | r = _mount_for_bluefs(); | |
6974 | ||
6975 | int reserved = 0; | |
6976 | string link_db; | |
6977 | string link_wal; | |
6978 | if (devs_source.count(BlueFS::BDEV_DB) && | |
9f95a23c | 6979 | bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 6980 | link_db = path + "/block.db"; |
9f95a23c TL |
6981 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
6982 | bluefs_layout.dedicated_db = false; | |
11fdf7f2 TL |
6983 | } |
6984 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6985 | link_wal = path + "/block.wal"; | |
9f95a23c | 6986 | bluefs_layout.dedicated_wal = false; |
11fdf7f2 TL |
6987 | } |
6988 | ||
6989 | size_t target_size; | |
6990 | string target_name; | |
6991 | if (id == BlueFS::BDEV_NEWWAL) { | |
6992 | target_name = "block.wal"; | |
6993 | target_size = cct->_conf->bluestore_block_wal_size; | |
9f95a23c | 6994 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6995 | |
6996 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
eafe8130 | 6997 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6998 | ceph_assert(r == 0); |
6999 | ||
7000 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
7001 | r = _check_or_set_bdev_label( | |
7002 | dev_path, | |
7003 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
7004 | "bluefs wal", | |
7005 | true); | |
7006 | ceph_assert(r == 0); | |
7007 | } | |
7008 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
7009 | } else if (id == BlueFS::BDEV_NEWDB) { | |
7010 | target_name = "block.db"; | |
7011 | target_size = cct->_conf->bluestore_block_db_size; | |
9f95a23c TL |
7012 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
7013 | bluefs_layout.dedicated_db = true; | |
31f18b77 | 7014 | |
11fdf7f2 | 7015 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
eafe8130 | 7016 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
7017 | ceph_assert(r == 0); |
7018 | ||
7019 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
7020 | r = _check_or_set_bdev_label( | |
7021 | dev_path, | |
7022 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
7023 | "bluefs db", | |
7024 | true); | |
7025 | ceph_assert(r == 0); | |
7026 | } | |
7027 | reserved = SUPER_RESERVED; | |
31f18b77 FG |
7028 | } |
7029 | ||
11fdf7f2 TL |
7030 | bluefs->umount(); |
7031 | bluefs->mount(); | |
7032 | ||
7033 | bluefs->add_block_extent( | |
7034 | id, reserved, bluefs->get_block_device_size(id) - reserved); | |
7035 | ||
9f95a23c | 7036 | r = bluefs->device_migrate_to_new(cct, devs_source, id, bluefs_layout); |
11fdf7f2 | 7037 | |
7c673cae | 7038 | if (r < 0) { |
11fdf7f2 TL |
7039 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
7040 | goto shutdown; | |
7041 | } | |
7042 | ||
7043 | if (!link_db.empty()) { | |
7044 | r = unlink(link_db.c_str()); | |
7045 | ceph_assert(r == 0); | |
7046 | } | |
7047 | if (!link_wal.empty()) { | |
7048 | r = unlink(link_wal.c_str()); | |
7049 | ceph_assert(r == 0); | |
7050 | } | |
7051 | r = _setup_block_symlink_or_file( | |
7052 | target_name, | |
7053 | dev_path, | |
7054 | target_size, | |
7055 | true); | |
7056 | ceph_assert(r == 0); | |
7057 | dout(0) << __func__ << " success" << dendl; | |
7058 | ||
7059 | shutdown: | |
7060 | _umount_for_bluefs(); | |
7061 | return r; | |
7062 | } | |
7063 | ||
7064 | string BlueStore::get_device_path(unsigned id) | |
7065 | { | |
7066 | string res; | |
7067 | if (id < BlueFS::MAX_BDEV) { | |
7068 | switch (id) { | |
7069 | case BlueFS::BDEV_WAL: | |
7070 | res = path + "/block.wal"; | |
7071 | break; | |
7072 | case BlueFS::BDEV_DB: | |
9f95a23c | 7073 | if (id == bluefs_layout.shared_bdev) { |
11fdf7f2 TL |
7074 | res = path + "/block"; |
7075 | } else { | |
7076 | res = path + "/block.db"; | |
7077 | } | |
7078 | break; | |
7079 | case BlueFS::BDEV_SLOW: | |
7080 | res = path + "/block"; | |
7081 | break; | |
7082 | } | |
7083 | } | |
7084 | return res; | |
7085 | } | |
7086 | ||
7087 | int BlueStore::expand_devices(ostream& out) | |
7088 | { | |
1911f103 | 7089 | int r = cold_open(); |
11fdf7f2 TL |
7090 | ceph_assert(r == 0); |
7091 | bluefs->dump_block_extents(out); | |
1911f103 | 7092 | out << "Expanding DB/WAL..." << std::endl; |
11fdf7f2 | 7093 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { |
9f95a23c | 7094 | if (devid == bluefs_layout.shared_bdev ) { |
11fdf7f2 TL |
7095 | continue; |
7096 | } | |
7097 | uint64_t size = bluefs->get_block_device_size(devid); | |
7098 | if (size == 0) { | |
7099 | // no bdev | |
7100 | continue; | |
7101 | } | |
7102 | ||
7103 | interval_set<uint64_t> before; | |
7104 | bluefs->get_block_extents(devid, &before); | |
7105 | ceph_assert(!before.empty()); | |
7106 | uint64_t end = before.range_end(); | |
7107 | if (end < size) { | |
7108 | out << devid | |
7109 | <<" : expanding " << " from 0x" << std::hex | |
7110 | << end << " to 0x" << size << std::dec << std::endl; | |
7111 | bluefs->add_block_extent(devid, end, size-end); | |
7112 | string p = get_device_path(devid); | |
7113 | const char* path = p.c_str(); | |
7114 | if (path == nullptr) { | |
7115 | derr << devid | |
7116 | <<": can't find device path " << dendl; | |
7117 | continue; | |
7118 | } | |
7119 | bluestore_bdev_label_t label; | |
7120 | int r = _read_bdev_label(cct, path, &label); | |
7121 | if (r < 0) { | |
7122 | derr << "unable to read label for " << path << ": " | |
7123 | << cpp_strerror(r) << dendl; | |
7124 | continue; | |
7125 | } | |
7126 | label.size = size; | |
7127 | r = _write_bdev_label(cct, path, label); | |
7128 | if (r < 0) { | |
7129 | derr << "unable to write label for " << path << ": " | |
7130 | << cpp_strerror(r) << dendl; | |
7131 | continue; | |
7132 | } | |
7133 | out << devid | |
7134 | <<" : size label updated to " << size | |
7135 | << std::endl; | |
7136 | } | |
7137 | } | |
7138 | uint64_t size0 = fm->get_size(); | |
7139 | uint64_t size = bdev->get_size(); | |
7140 | if (size0 < size) { | |
9f95a23c | 7141 | out << bluefs_layout.shared_bdev |
1911f103 TL |
7142 | << " : expanding " << " from 0x" << std::hex |
7143 | << size0 << " to 0x" << size << std::dec << std::endl; | |
7144 | _write_out_fm_meta(size, true); | |
7145 | cold_close(); | |
7146 | ||
7147 | // mount in read/write to sync expansion changes | |
7148 | r = _mount(false); | |
11fdf7f2 | 7149 | ceph_assert(r == 0); |
1911f103 TL |
7150 | umount(); |
7151 | } else { | |
7152 | cold_close(); | |
7c673cae | 7153 | } |
1911f103 TL |
7154 | return r; |
7155 | } | |
7156 | ||
7157 | int BlueStore::dump_bluefs_sizes(ostream& out) | |
7158 | { | |
7159 | int r = cold_open(); | |
7160 | ceph_assert(r == 0); | |
7161 | bluefs->dump_block_extents(out); | |
7162 | cold_close(); | |
7c673cae FG |
7163 | return r; |
7164 | } | |
7165 | ||
7166 | void BlueStore::set_cache_shards(unsigned num) | |
7167 | { | |
7168 | dout(10) << __func__ << " " << num << dendl; | |
9f95a23c TL |
7169 | size_t oold = onode_cache_shards.size(); |
7170 | size_t bold = buffer_cache_shards.size(); | |
7171 | ceph_assert(num >= oold && num >= bold); | |
7172 | onode_cache_shards.resize(num); | |
7173 | buffer_cache_shards.resize(num); | |
7174 | for (unsigned i = oold; i < num; ++i) { | |
7175 | onode_cache_shards[i] = | |
7176 | OnodeCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
7177 | logger); | |
7178 | } | |
7179 | for (unsigned i = bold; i < num; ++i) { | |
7180 | buffer_cache_shards[i] = | |
7181 | BufferCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
7182 | logger); | |
7c673cae FG |
7183 | } |
7184 | } | |
7185 | ||
11fdf7f2 | 7186 | int BlueStore::_mount(bool kv_only, bool open_db) |
7c673cae FG |
7187 | { |
7188 | dout(1) << __func__ << " path " << path << dendl; | |
7189 | ||
3efd9988 FG |
7190 | _kv_only = kv_only; |
7191 | ||
7c673cae FG |
7192 | { |
7193 | string type; | |
7194 | int r = read_meta("type", &type); | |
7195 | if (r < 0) { | |
7196 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
7197 | << dendl; | |
7198 | return r; | |
7199 | } | |
7200 | ||
7201 | if (type != "bluestore") { | |
7202 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
7203 | return -EIO; | |
7204 | } | |
7205 | } | |
7206 | ||
7207 | if (cct->_conf->bluestore_fsck_on_mount) { | |
7208 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); | |
7209 | if (rc < 0) | |
7210 | return rc; | |
7211 | if (rc > 0) { | |
7212 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7213 | return -EIO; | |
7214 | } | |
7215 | } | |
7216 | ||
eafe8130 TL |
7217 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
7218 | derr << __func__ << " osd_max_object_size " | |
7219 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
7220 | << OBJECT_MAX_SIZE << dendl; | |
7221 | return -EINVAL; | |
7222 | } | |
7223 | ||
7c673cae FG |
7224 | int r = _open_path(); |
7225 | if (r < 0) | |
7226 | return r; | |
7227 | r = _open_fsid(false); | |
7228 | if (r < 0) | |
7229 | goto out_path; | |
7230 | ||
7231 | r = _read_fsid(&fsid); | |
7232 | if (r < 0) | |
7233 | goto out_fsid; | |
7234 | ||
7235 | r = _lock_fsid(); | |
7236 | if (r < 0) | |
7237 | goto out_fsid; | |
7238 | ||
7239 | r = _open_bdev(false); | |
7240 | if (r < 0) | |
7241 | goto out_fsid; | |
7242 | ||
11fdf7f2 TL |
7243 | if (open_db) { |
7244 | r = _open_db_and_around(false); | |
7245 | } else { | |
7246 | // we can bypass db open exclusively in case of kv_only mode | |
7247 | ceph_assert(kv_only); | |
7248 | r = _open_db(false, true); | |
9f95a23c TL |
7249 | } |
7250 | if (r < 0) { | |
7251 | goto out_bdev; | |
11fdf7f2 | 7252 | } |
7c673cae FG |
7253 | |
7254 | if (kv_only) | |
7255 | return 0; | |
7256 | ||
11fdf7f2 TL |
7257 | r = _upgrade_super(); |
7258 | if (r < 0) { | |
7c673cae | 7259 | goto out_db; |
11fdf7f2 | 7260 | } |
7c673cae FG |
7261 | |
7262 | r = _open_collections(); | |
7263 | if (r < 0) | |
11fdf7f2 | 7264 | goto out_db; |
7c673cae FG |
7265 | |
7266 | r = _reload_logger(); | |
7267 | if (r < 0) | |
7268 | goto out_coll; | |
7269 | ||
31f18b77 | 7270 | _kv_start(); |
7c673cae FG |
7271 | |
7272 | r = _deferred_replay(); | |
7273 | if (r < 0) | |
7274 | goto out_stop; | |
7275 | ||
7276 | mempool_thread.init(); | |
7277 | ||
9f95a23c | 7278 | if ((!per_pool_stat_collection || !per_pool_omap) && |
eafe8130 | 7279 | cct->_conf->bluestore_fsck_quick_fix_on_mount == true) { |
9f95a23c TL |
7280 | |
7281 | bool was_per_pool_omap = per_pool_omap; | |
7282 | ||
eafe8130 TL |
7283 | dout(1) << __func__ << " quick-fix on mount" << dendl; |
7284 | _fsck_on_open(FSCK_SHALLOW, true); | |
7285 | ||
7286 | //reread statfs | |
7287 | //FIXME minor: replace with actual open/close? | |
7288 | _open_statfs(); | |
eafe8130 | 7289 | _check_legacy_statfs_alert(); |
9f95a23c TL |
7290 | |
7291 | //set again as hopefully it has been fixed | |
7292 | if (!was_per_pool_omap) { | |
7293 | _set_per_pool_omap(); | |
7294 | } | |
eafe8130 TL |
7295 | } |
7296 | ||
7c673cae FG |
7297 | mounted = true; |
7298 | return 0; | |
7299 | ||
7300 | out_stop: | |
7301 | _kv_stop(); | |
7c673cae | 7302 | out_coll: |
f6b5b4d7 | 7303 | _shutdown_cache(); |
7c673cae | 7304 | out_db: |
1911f103 | 7305 | _close_db_and_around(false); |
7c673cae FG |
7306 | out_bdev: |
7307 | _close_bdev(); | |
7308 | out_fsid: | |
7309 | _close_fsid(); | |
7310 | out_path: | |
7311 | _close_path(); | |
7312 | return r; | |
7313 | } | |
7314 | ||
7315 | int BlueStore::umount() | |
7316 | { | |
11fdf7f2 | 7317 | ceph_assert(_kv_only || mounted); |
7c673cae FG |
7318 | dout(1) << __func__ << dendl; |
7319 | ||
7320 | _osr_drain_all(); | |
7c673cae | 7321 | |
7c673cae | 7322 | mounted = false; |
3efd9988 FG |
7323 | if (!_kv_only) { |
7324 | mempool_thread.shutdown(); | |
7325 | dout(20) << __func__ << " stopping kv thread" << dendl; | |
7326 | _kv_stop(); | |
f6b5b4d7 | 7327 | _shutdown_cache(); |
3efd9988 FG |
7328 | dout(20) << __func__ << " closing" << dendl; |
7329 | ||
3efd9988 | 7330 | } |
1911f103 | 7331 | _close_db_and_around(false); |
7c673cae FG |
7332 | _close_bdev(); |
7333 | _close_fsid(); | |
7334 | _close_path(); | |
7335 | ||
7336 | if (cct->_conf->bluestore_fsck_on_umount) { | |
7337 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); | |
7338 | if (rc < 0) | |
7339 | return rc; | |
7340 | if (rc > 0) { | |
7341 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7342 | return -EIO; | |
7343 | } | |
7344 | } | |
7345 | return 0; | |
7346 | } | |
7347 | ||
eafe8130 TL |
7348 | int BlueStore::cold_open() |
7349 | { | |
7350 | int r = _open_path(); | |
7351 | if (r < 0) | |
7352 | return r; | |
7353 | r = _open_fsid(false); | |
7354 | if (r < 0) | |
7355 | goto out_path; | |
7356 | ||
7357 | r = _read_fsid(&fsid); | |
7358 | if (r < 0) | |
7359 | goto out_fsid; | |
7360 | ||
7361 | r = _lock_fsid(); | |
7362 | if (r < 0) | |
7363 | goto out_fsid; | |
7364 | ||
7365 | r = _open_bdev(false); | |
7366 | if (r < 0) | |
7367 | goto out_fsid; | |
7368 | r = _open_db_and_around(true); | |
7369 | if (r < 0) { | |
7370 | goto out_bdev; | |
7371 | } | |
7372 | return 0; | |
7373 | out_bdev: | |
7374 | _close_bdev(); | |
7375 | out_fsid: | |
7376 | _close_fsid(); | |
7377 | out_path: | |
7378 | _close_path(); | |
7379 | return r; | |
7380 | } | |
7381 | int BlueStore::cold_close() | |
7382 | { | |
1911f103 | 7383 | _close_db_and_around(true); |
eafe8130 TL |
7384 | _close_bdev(); |
7385 | _close_fsid(); | |
7386 | _close_path(); | |
7387 | return 0; | |
7388 | } | |
7389 | ||
9f95a23c TL |
7390 | // derr wrapper to limit enormous output and avoid log flooding. |
7391 | // Of limited use where such output is expected for now | |
7392 | #define fsck_derr(err_cnt, threshold) \ | |
7393 | if (err_cnt <= threshold) { \ | |
7394 | bool need_skip_print = err_cnt == threshold; \ | |
7395 | derr | |
7396 | ||
7397 | #define fsck_dendl \ | |
7398 | dendl; \ | |
7399 | if (need_skip_print) \ | |
7400 | derr << "more error lines skipped..." << dendl; \ | |
7c673cae | 7401 | } |
7c673cae | 7402 | |
eafe8130 TL |
7403 | int _fsck_sum_extents( |
7404 | const PExtentVector& extents, | |
7405 | bool compressed, | |
7406 | store_statfs_t& expected_statfs) | |
7407 | { | |
7408 | for (auto e : extents) { | |
7409 | if (!e.is_valid()) | |
7410 | continue; | |
7411 | expected_statfs.allocated += e.length; | |
7412 | if (compressed) { | |
7413 | expected_statfs.data_compressed_allocated += e.length; | |
7414 | } | |
7415 | } | |
7416 | return 0; | |
7417 | } | |
7418 | ||
7c673cae | 7419 | int BlueStore::_fsck_check_extents( |
11fdf7f2 | 7420 | const coll_t& cid, |
7c673cae FG |
7421 | const ghobject_t& oid, |
7422 | const PExtentVector& extents, | |
7423 | bool compressed, | |
7424 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 7425 | uint64_t granularity, |
11fdf7f2 | 7426 | BlueStoreRepairer* repairer, |
eafe8130 TL |
7427 | store_statfs_t& expected_statfs, |
7428 | FSCKDepth depth) | |
7c673cae FG |
7429 | { |
7430 | dout(30) << __func__ << " oid " << oid << " extents " << extents << dendl; | |
7431 | int errors = 0; | |
7432 | for (auto e : extents) { | |
7433 | if (!e.is_valid()) | |
7434 | continue; | |
7435 | expected_statfs.allocated += e.length; | |
7436 | if (compressed) { | |
11fdf7f2 | 7437 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae | 7438 | } |
eafe8130 TL |
7439 | if (depth != FSCK_SHALLOW) { |
7440 | bool already = false; | |
9f95a23c | 7441 | apply_for_bitset_range( |
eafe8130 TL |
7442 | e.offset, e.length, granularity, used_blocks, |
7443 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
7444 | if (bs.test(pos)) { |
7445 | if (repairer) { | |
7446 | repairer->note_misreference( | |
7447 | pos * min_alloc_size, min_alloc_size, !already); | |
7448 | } | |
7449 | if (!already) { | |
7450 | derr << "fsck error: " << oid << " extent " << e | |
7451 | << " or a subset is already allocated (misreferenced)" << dendl; | |
7452 | ++errors; | |
7453 | already = true; | |
7454 | } | |
11fdf7f2 | 7455 | } |
eafe8130 TL |
7456 | else |
7457 | bs.set(pos); | |
7458 | }); | |
7459 | if (repairer) { | |
7460 | repairer->get_space_usage_tracker().set_used( e.offset, e.length, cid, oid); | |
7461 | } | |
11fdf7f2 | 7462 | |
eafe8130 TL |
7463 | if (e.end() > bdev->get_size()) { |
7464 | derr << "fsck error: " << oid << " extent " << e | |
7465 | << " past end of block device" << dendl; | |
7466 | ++errors; | |
7467 | } | |
7c673cae FG |
7468 | } |
7469 | } | |
7470 | return errors; | |
7471 | } | |
7472 | ||
11fdf7f2 TL |
7473 | void BlueStore::_fsck_check_pool_statfs( |
7474 | BlueStore::per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
7475 | int64_t& errors, |
7476 | int64_t& warnings, | |
11fdf7f2 TL |
7477 | BlueStoreRepairer* repairer) |
7478 | { | |
7479 | auto it = db->get_iterator(PREFIX_STAT); | |
7480 | if (it) { | |
7481 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7482 | string key = it->key(); | |
7483 | if (key == BLUESTORE_GLOBAL_STATFS_KEY) { | |
7484 | if (repairer) { | |
eafe8130 TL |
7485 | ++errors; |
7486 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
7487 | derr << "fsck error: " << "legacy statfs record found, removing" | |
11fdf7f2 TL |
7488 | << dendl; |
7489 | } | |
7490 | continue; | |
7491 | } | |
11fdf7f2 TL |
7492 | uint64_t pool_id; |
7493 | if (get_key_pool_stat(key, &pool_id) < 0) { | |
7494 | derr << "fsck error: bad key " << key | |
7495 | << "in statfs namespece" << dendl; | |
7496 | if (repairer) { | |
7497 | repairer->remove_key(db, PREFIX_STAT, key); | |
7498 | } | |
7499 | ++errors; | |
7500 | continue; | |
7501 | } | |
7502 | ||
7503 | volatile_statfs vstatfs; | |
7504 | bufferlist bl = it->value(); | |
7505 | auto blp = bl.cbegin(); | |
7506 | try { | |
7507 | vstatfs.decode(blp); | |
7508 | } catch (buffer::error& e) { | |
7509 | derr << "fsck error: failed to decode Pool StatFS record" | |
7510 | << pretty_binary_string(key) << dendl; | |
7511 | if (repairer) { | |
7512 | dout(20) << __func__ << " undecodable Pool StatFS record, key:'" | |
7513 | << pretty_binary_string(key) | |
7514 | << "', removing" << dendl; | |
7515 | repairer->remove_key(db, PREFIX_STAT, key); | |
7516 | } | |
7517 | ++errors; | |
7518 | vstatfs.reset(); | |
7519 | } | |
7520 | auto stat_it = expected_pool_statfs.find(pool_id); | |
7521 | if (stat_it == expected_pool_statfs.end()) { | |
7522 | if (vstatfs.is_empty()) { | |
7523 | // we don't consider that as an error since empty pool statfs | |
7524 | // are left in DB for now | |
7525 | dout(20) << "fsck inf: found empty stray Pool StatFS record for pool id 0x" | |
7526 | << std::hex << pool_id << std::dec << dendl; | |
7527 | if (repairer) { | |
7528 | // but we need to increment error count in case of repair | |
7529 | // to have proper counters at the end | |
7530 | // (as repairer increments recovery counter anyway). | |
7531 | ++errors; | |
7532 | } | |
7533 | } else { | |
7534 | derr << "fsck error: found stray Pool StatFS record for pool id 0x" | |
7535 | << std::hex << pool_id << std::dec << dendl; | |
7536 | ++errors; | |
7537 | } | |
7538 | if (repairer) { | |
7539 | repairer->remove_key(db, PREFIX_SHARED_BLOB, key); | |
7540 | } | |
7541 | continue; | |
7542 | } | |
7543 | store_statfs_t statfs; | |
7544 | vstatfs.publish(&statfs); | |
7545 | if (!(stat_it->second == statfs)) { | |
7546 | derr << "fsck error: actual " << statfs | |
7547 | << " != expected " << stat_it->second | |
7548 | << " for pool " | |
7549 | << std::hex << pool_id << std::dec << dendl; | |
7550 | if (repairer) { | |
7551 | repairer->fix_statfs(db, key, stat_it->second); | |
7552 | } | |
7553 | ++errors; | |
7554 | } | |
7555 | expected_pool_statfs.erase(stat_it); | |
7556 | } | |
7557 | } // if (it) | |
eafe8130 TL |
7558 | for (auto& s : expected_pool_statfs) { |
7559 | if (s.second.is_zero()) { | |
11fdf7f2 TL |
7560 | // we might lack empty statfs recs in DB |
7561 | continue; | |
7562 | } | |
7563 | derr << "fsck error: missing Pool StatFS record for pool " | |
eafe8130 | 7564 | << std::hex << s.first << std::dec << dendl; |
11fdf7f2 TL |
7565 | if (repairer) { |
7566 | string key; | |
eafe8130 TL |
7567 | get_pool_stat_key(s.first, &key); |
7568 | repairer->fix_statfs(db, key, s.second); | |
11fdf7f2 TL |
7569 | } |
7570 | ++errors; | |
7571 | } | |
eafe8130 | 7572 | if (!per_pool_stat_collection && |
eafe8130 TL |
7573 | repairer) { |
7574 | // by virtue of running this method, we correct the top-level | |
7575 | // error of having global stats | |
7576 | repairer->inc_repaired(); | |
7577 | } | |
11fdf7f2 TL |
7578 | } |
7579 | ||
eafe8130 TL |
7580 | BlueStore::OnodeRef BlueStore::fsck_check_objects_shallow( |
7581 | BlueStore::FSCKDepth depth, | |
7582 | int64_t pool_id, | |
7583 | BlueStore::CollectionRef c, | |
7584 | const ghobject_t& oid, | |
7585 | const string& key, | |
7586 | const bufferlist& value, | |
9f95a23c | 7587 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
7588 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
7589 | const BlueStore::FSCK_ObjectCtx& ctx) | |
7590 | { | |
7591 | auto& errors = ctx.errors; | |
7592 | auto& num_objects = ctx.num_objects; | |
7593 | auto& num_extents = ctx.num_extents; | |
7594 | auto& num_blobs = ctx.num_blobs; | |
7595 | auto& num_sharded_objects = ctx.num_sharded_objects; | |
7596 | auto& num_spanning_blobs = ctx.num_spanning_blobs; | |
7597 | auto used_blocks = ctx.used_blocks; | |
7598 | auto sb_info_lock = ctx.sb_info_lock; | |
7599 | auto& sb_info = ctx.sb_info; | |
7600 | auto repairer = ctx.repairer; | |
7601 | ||
7602 | store_statfs_t* res_statfs = (per_pool_stat_collection || repairer) ? | |
7603 | &ctx.expected_pool_statfs[pool_id] : | |
7604 | &ctx.expected_store_statfs; | |
7605 | ||
7606 | dout(10) << __func__ << " " << oid << dendl; | |
7607 | OnodeRef o; | |
7608 | o.reset(Onode::decode(c, oid, key, value)); | |
7609 | ++num_objects; | |
7c673cae | 7610 | |
eafe8130 | 7611 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); |
7c673cae | 7612 | |
eafe8130 TL |
7613 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
7614 | _dump_onode<30>(cct, *o); | |
7615 | // shards | |
7616 | if (!o->extent_map.shards.empty()) { | |
7617 | ++num_sharded_objects; | |
7618 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 7619 | ceph_assert(expecting_shards); |
eafe8130 TL |
7620 | for (auto& s : o->extent_map.shards) { |
7621 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
9f95a23c | 7622 | expecting_shards->push_back(string()); |
eafe8130 | 7623 | get_extent_shard_key(o->key, s.shard_info->offset, |
9f95a23c | 7624 | &expecting_shards->back()); |
eafe8130 TL |
7625 | if (s.shard_info->offset >= o->onode.size) { |
7626 | derr << "fsck error: " << oid << " shard 0x" << std::hex | |
7627 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size | |
7628 | << std::dec << dendl; | |
7629 | ++errors; | |
7630 | } | |
7631 | } | |
7632 | } | |
7633 | } | |
7c673cae | 7634 | |
eafe8130 TL |
7635 | // lextents |
7636 | uint64_t pos = 0; | |
7637 | mempool::bluestore_fsck::map<BlobRef, | |
7638 | bluestore_blob_use_tracker_t> ref_map; | |
7639 | for (auto& l : o->extent_map.extent_map) { | |
7640 | dout(20) << __func__ << " " << l << dendl; | |
7641 | if (l.logical_offset < pos) { | |
7642 | derr << "fsck error: " << oid << " lextent at 0x" | |
7643 | << std::hex << l.logical_offset | |
7644 | << " overlaps with the previous, which ends at 0x" << pos | |
7645 | << std::dec << dendl; | |
7646 | ++errors; | |
7647 | } | |
7648 | if (depth != FSCK_SHALLOW && | |
7649 | o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
7650 | derr << "fsck error: " << oid << " lextent at 0x" | |
7651 | << std::hex << l.logical_offset << "~" << l.length | |
7652 | << " spans a shard boundary" | |
7653 | << std::dec << dendl; | |
7654 | ++errors; | |
7655 | } | |
7656 | pos = l.logical_offset + l.length; | |
7657 | res_statfs->data_stored += l.length; | |
7658 | ceph_assert(l.blob); | |
7659 | const bluestore_blob_t& blob = l.blob->get_blob(); | |
7660 | ||
7661 | auto& ref = ref_map[l.blob]; | |
7662 | if (ref.is_empty()) { | |
7663 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
7664 | uint32_t l = blob.get_logical_length(); | |
7665 | ref.init(l, min_release_size); | |
7666 | } | |
7667 | ref.get( | |
7668 | l.blob_offset, | |
7669 | l.length); | |
7670 | ++num_extents; | |
7671 | if (depth != FSCK_SHALLOW && | |
7672 | blob.has_unused()) { | |
7673 | ceph_assert(referenced); | |
7674 | auto p = referenced->find(l.blob); | |
7675 | bluestore_blob_t::unused_t* pu; | |
7676 | if (p == referenced->end()) { | |
7677 | pu = &(*referenced)[l.blob]; | |
7678 | } | |
7679 | else { | |
7680 | pu = &p->second; | |
7681 | } | |
7682 | uint64_t blob_len = blob.get_logical_length(); | |
7683 | ceph_assert((blob_len % (sizeof(*pu) * 8)) == 0); | |
7684 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
7685 | uint64_t chunk_size = blob_len / (sizeof(*pu) * 8); | |
7686 | uint64_t start = l.blob_offset / chunk_size; | |
7687 | uint64_t end = | |
7688 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; | |
7689 | for (auto i = start; i < end; ++i) { | |
7690 | (*pu) |= (1u << i); | |
7691 | } | |
7692 | } | |
7693 | } //for (auto& l : o->extent_map.extent_map) | |
7694 | ||
7695 | for (auto& i : ref_map) { | |
7696 | ++num_blobs; | |
7697 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7698 | bool equal = | |
7699 | depth == FSCK_SHALLOW ? true : | |
7700 | i.first->get_blob_use_tracker().equal(i.second); | |
7701 | if (!equal) { | |
7702 | derr << "fsck error: " << oid << " blob " << *i.first | |
7703 | << " doesn't match expected ref_map " << i.second << dendl; | |
7704 | ++errors; | |
7705 | } | |
7706 | if (blob.is_compressed()) { | |
7707 | res_statfs->data_compressed += blob.get_compressed_payload_length(); | |
7708 | res_statfs->data_compressed_original += | |
7709 | i.first->get_referenced_bytes(); | |
7710 | } | |
7711 | if (blob.is_shared()) { | |
7712 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
7713 | derr << "fsck error: " << oid << " blob " << blob | |
7714 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " | |
7715 | << blobid_max << dendl; | |
7716 | ++errors; | |
7717 | } | |
7718 | else if (i.first->shared_blob->get_sbid() == 0) { | |
7719 | derr << "fsck error: " << oid << " blob " << blob | |
7720 | << " marked as shared but has uninitialized sbid" | |
7721 | << dendl; | |
7722 | ++errors; | |
7723 | } | |
7724 | // the below lock is optional and provided in multithreading mode only | |
7725 | if (sb_info_lock) { | |
7726 | sb_info_lock->lock(); | |
7727 | } | |
7728 | sb_info_t& sbi = sb_info[i.first->shared_blob->get_sbid()]; | |
7729 | ceph_assert(sbi.cid == coll_t() || sbi.cid == c->cid); | |
7730 | ceph_assert(sbi.pool_id == INT64_MIN || | |
7731 | sbi.pool_id == oid.hobj.get_logical_pool()); | |
7732 | sbi.cid = c->cid; | |
7733 | sbi.pool_id = oid.hobj.get_logical_pool(); | |
7734 | sbi.sb = i.first->shared_blob; | |
7735 | sbi.oids.push_back(oid); | |
7736 | sbi.compressed = blob.is_compressed(); | |
7737 | for (auto e : blob.get_extents()) { | |
7738 | if (e.is_valid()) { | |
7739 | sbi.ref_map.get(e.offset, e.length); | |
7740 | } | |
7741 | } | |
7742 | if (sb_info_lock) { | |
7743 | sb_info_lock->unlock(); | |
7744 | } | |
7745 | } else if (depth != FSCK_SHALLOW) { | |
7746 | ceph_assert(used_blocks); | |
7747 | errors += _fsck_check_extents(c->cid, oid, blob.get_extents(), | |
7748 | blob.is_compressed(), | |
7749 | *used_blocks, | |
7750 | fm->get_alloc_size(), | |
7751 | repairer, | |
7752 | *res_statfs, | |
7753 | depth); | |
7754 | } else { | |
7755 | errors += _fsck_sum_extents( | |
7756 | blob.get_extents(), | |
7757 | blob.is_compressed(), | |
7758 | *res_statfs); | |
7759 | } | |
7760 | } // for (auto& i : ref_map) | |
9f95a23c TL |
7761 | |
7762 | if (o->onode.has_omap()) { | |
7763 | _fsck_check_object_omap(depth, o, ctx); | |
7764 | } | |
7765 | ||
eafe8130 TL |
7766 | return o; |
7767 | } | |
7768 | ||
7769 | #include "common/WorkQueue.h" | |
7770 | ||
7771 | class ShallowFSCKThreadPool : public ThreadPool | |
7772 | { | |
7773 | public: | |
7774 | ShallowFSCKThreadPool(CephContext* cct_, std::string nm, std::string tn, int n) : | |
7775 | ThreadPool(cct_, nm, tn, n) { | |
7776 | } | |
7777 | void worker(ThreadPool::WorkThread* wt) override { | |
7778 | int next_wq = 0; | |
7779 | while (!_stop) { | |
7780 | next_wq %= work_queues.size(); | |
7781 | WorkQueue_ *wq = work_queues[next_wq++]; | |
7782 | ||
7783 | void* item = wq->_void_dequeue(); | |
7784 | if (item) { | |
7785 | processing++; | |
7786 | TPHandle tp_handle(cct, nullptr, wq->timeout_interval, wq->suicide_interval); | |
7787 | wq->_void_process(item, tp_handle); | |
7788 | processing--; | |
7789 | } | |
7790 | } | |
7791 | } | |
7792 | template <size_t BatchLen> | |
7793 | struct FSCKWorkQueue : public ThreadPool::WorkQueue_ | |
7794 | { | |
7795 | struct Entry { | |
7796 | int64_t pool_id; | |
7797 | BlueStore::CollectionRef c; | |
7798 | ghobject_t oid; | |
7799 | string key; | |
7800 | bufferlist value; | |
7801 | }; | |
7802 | struct Batch { | |
7803 | std::atomic<size_t> running = { 0 }; | |
7804 | size_t entry_count = 0; | |
7805 | std::array<Entry, BatchLen> entries; | |
7806 | ||
7807 | int64_t errors = 0; | |
7808 | int64_t warnings = 0; | |
7809 | uint64_t num_objects = 0; | |
7810 | uint64_t num_extents = 0; | |
7811 | uint64_t num_blobs = 0; | |
7812 | uint64_t num_sharded_objects = 0; | |
7813 | uint64_t num_spanning_blobs = 0; | |
7814 | store_statfs_t expected_store_statfs; | |
7815 | BlueStore::per_pool_statfs expected_pool_statfs; | |
7816 | }; | |
7817 | ||
7818 | size_t batchCount; | |
7819 | BlueStore* store = nullptr; | |
7820 | ||
eafe8130 TL |
7821 | ceph::mutex* sb_info_lock = nullptr; |
7822 | BlueStore::sb_info_map_t* sb_info = nullptr; | |
7823 | BlueStoreRepairer* repairer = nullptr; | |
7824 | ||
7825 | Batch* batches = nullptr; | |
7826 | size_t last_batch_pos = 0; | |
7827 | bool batch_acquired = false; | |
7828 | ||
7829 | FSCKWorkQueue(std::string n, | |
7830 | size_t _batchCount, | |
7831 | BlueStore* _store, | |
eafe8130 TL |
7832 | ceph::mutex* _sb_info_lock, |
7833 | BlueStore::sb_info_map_t& _sb_info, | |
7834 | BlueStoreRepairer* _repairer) : | |
7835 | WorkQueue_(n, time_t(), time_t()), | |
7836 | batchCount(_batchCount), | |
7837 | store(_store), | |
eafe8130 TL |
7838 | sb_info_lock(_sb_info_lock), |
7839 | sb_info(&_sb_info), | |
7840 | repairer(_repairer) | |
7841 | { | |
7842 | batches = new Batch[batchCount]; | |
7843 | } | |
7844 | ~FSCKWorkQueue() { | |
7845 | delete[] batches; | |
7846 | } | |
7847 | ||
7848 | /// Remove all work items from the queue. | |
7849 | void _clear() override { | |
7850 | //do nothing | |
7851 | } | |
7852 | /// Check whether there is anything to do. | |
7853 | bool _empty() override { | |
7854 | ceph_assert(false); | |
7855 | } | |
7856 | ||
7857 | /// Get the next work item to process. | |
7858 | void* _void_dequeue() override { | |
7859 | size_t pos = rand() % batchCount; | |
7860 | size_t pos0 = pos; | |
7861 | do { | |
7862 | auto& batch = batches[pos]; | |
7863 | if (batch.running.fetch_add(1) == 0) { | |
7864 | if (batch.entry_count) { | |
7865 | return &batch; | |
7866 | } | |
7867 | } | |
7868 | batch.running--; | |
7869 | pos++; | |
7870 | pos %= batchCount; | |
7871 | } while (pos != pos0); | |
7872 | return nullptr; | |
7873 | } | |
7874 | /** @brief Process the work item. | |
7875 | * This function will be called several times in parallel | |
7876 | * and must therefore be thread-safe. */ | |
7877 | void _void_process(void* item, TPHandle& handle) override { | |
7878 | Batch* batch = (Batch*)item; | |
7879 | ||
7880 | BlueStore::FSCK_ObjectCtx ctx( | |
7881 | batch->errors, | |
7882 | batch->warnings, | |
7883 | batch->num_objects, | |
7884 | batch->num_extents, | |
7885 | batch->num_blobs, | |
7886 | batch->num_sharded_objects, | |
7887 | batch->num_spanning_blobs, | |
7888 | nullptr, // used_blocks | |
9f95a23c | 7889 | nullptr, //used_omap_head |
eafe8130 TL |
7890 | sb_info_lock, |
7891 | *sb_info, | |
7892 | batch->expected_store_statfs, | |
7893 | batch->expected_pool_statfs, | |
7894 | repairer); | |
7895 | ||
7896 | for (size_t i = 0; i < batch->entry_count; i++) { | |
7897 | auto& entry = batch->entries[i]; | |
7898 | ||
7899 | store->fsck_check_objects_shallow( | |
7900 | BlueStore::FSCK_SHALLOW, | |
7901 | entry.pool_id, | |
7902 | entry.c, | |
7903 | entry.oid, | |
7904 | entry.key, | |
7905 | entry.value, | |
9f95a23c | 7906 | nullptr, // expecting_shards - this will need a protection if passed |
eafe8130 TL |
7907 | nullptr, // referenced |
7908 | ctx); | |
7909 | } | |
7910 | //std::cout << "processed " << batch << std::endl; | |
7911 | batch->entry_count = 0; | |
7912 | batch->running--; | |
7913 | } | |
7914 | /** @brief Synchronously finish processing a work item. | |
7915 | * This function is called after _void_process with the global thread pool lock held, | |
7916 | * so at most one copy will execute simultaneously for a given thread pool. | |
7917 | * It can be used for non-thread-safe finalization. */ | |
7918 | void _void_process_finish(void*) override { | |
7919 | ceph_assert(false); | |
7920 | } | |
7921 | ||
7922 | bool queue( | |
7923 | int64_t pool_id, | |
7924 | BlueStore::CollectionRef c, | |
7925 | const ghobject_t& oid, | |
7926 | const string& key, | |
7927 | const bufferlist& value) { | |
7928 | bool res = false; | |
7929 | size_t pos0 = last_batch_pos; | |
7930 | if (!batch_acquired) { | |
7931 | do { | |
7932 | auto& batch = batches[last_batch_pos]; | |
7933 | if (batch.running.fetch_add(1) == 0) { | |
7934 | if (batch.entry_count < BatchLen) { | |
7935 | batch_acquired = true; | |
7936 | break; | |
7937 | } | |
7938 | } | |
7939 | batch.running.fetch_sub(1); | |
7940 | last_batch_pos++; | |
7941 | last_batch_pos %= batchCount; | |
7942 | } while (last_batch_pos != pos0); | |
7943 | } | |
7944 | if (batch_acquired) { | |
7945 | auto& batch = batches[last_batch_pos]; | |
7946 | ceph_assert(batch.running); | |
7947 | ceph_assert(batch.entry_count < BatchLen); | |
7948 | ||
7949 | auto& entry = batch.entries[batch.entry_count]; | |
7950 | entry.pool_id = pool_id; | |
7951 | entry.c = c; | |
7952 | entry.oid = oid; | |
7953 | entry.key = key; | |
7954 | entry.value = value; | |
7955 | ||
7956 | ++batch.entry_count; | |
7957 | if (batch.entry_count == BatchLen) { | |
7958 | batch_acquired = false; | |
7959 | batch.running.fetch_sub(1); | |
7960 | last_batch_pos++; | |
7961 | last_batch_pos %= batchCount; | |
7962 | } | |
7963 | res = true; | |
7964 | } | |
7965 | return res; | |
7966 | } | |
7967 | ||
7968 | void finalize(ThreadPool& tp, | |
7969 | BlueStore::FSCK_ObjectCtx& ctx) { | |
7970 | if (batch_acquired) { | |
7971 | auto& batch = batches[last_batch_pos]; | |
7972 | ceph_assert(batch.running); | |
7973 | batch.running.fetch_sub(1); | |
7974 | } | |
7975 | tp.stop(); | |
7976 | ||
7977 | for (size_t i = 0; i < batchCount; i++) { | |
7978 | auto& batch = batches[i]; | |
7979 | ||
7980 | //process leftovers if any | |
7981 | if (batch.entry_count) { | |
7982 | TPHandle tp_handle(store->cct, | |
7983 | nullptr, | |
7984 | timeout_interval, | |
7985 | suicide_interval); | |
7986 | ceph_assert(batch.running == 0); | |
7987 | ||
7988 | batch.running++; // just to be on-par with the regular call | |
7989 | _void_process(&batch, tp_handle); | |
7990 | } | |
7991 | ceph_assert(batch.entry_count == 0); | |
7992 | ||
7993 | ctx.errors += batch.errors; | |
7994 | ctx.warnings += batch.warnings; | |
7995 | ctx.num_objects += batch.num_objects; | |
7996 | ctx.num_extents += batch.num_extents; | |
7997 | ctx.num_blobs += batch.num_blobs; | |
7998 | ctx.num_sharded_objects += batch.num_sharded_objects; | |
7999 | ctx.num_spanning_blobs += batch.num_spanning_blobs; | |
9f95a23c | 8000 | |
eafe8130 TL |
8001 | ctx.expected_store_statfs.add(batch.expected_store_statfs); |
8002 | ||
8003 | for (auto it = batch.expected_pool_statfs.begin(); | |
8004 | it != batch.expected_pool_statfs.end(); | |
8005 | it++) { | |
8006 | ctx.expected_pool_statfs[it->first].add(it->second); | |
8007 | } | |
8008 | } | |
8009 | } | |
8010 | }; | |
8011 | }; | |
8012 | ||
9f95a23c TL |
8013 | void BlueStore::_fsck_check_object_omap(FSCKDepth depth, |
8014 | OnodeRef& o, | |
8015 | const BlueStore::FSCK_ObjectCtx& ctx) | |
eafe8130 | 8016 | { |
9f95a23c TL |
8017 | auto& errors = ctx.errors; |
8018 | auto& warnings = ctx.warnings; | |
8019 | auto repairer = ctx.repairer; | |
8020 | ||
8021 | ceph_assert(o->onode.has_omap()); | |
8022 | if (!o->onode.is_perpool_omap() && !o->onode.is_pgmeta_omap()) { | |
8023 | if (per_pool_omap) { | |
8024 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8025 | << "fsck error: " << o->oid | |
8026 | << " has omap that is not per-pool or pgmeta" | |
8027 | << fsck_dendl; | |
8028 | ++errors; | |
8029 | } else { | |
8030 | const char* w; | |
8031 | int64_t num; | |
8032 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8033 | ++errors; | |
8034 | num = errors; | |
8035 | w = "error"; | |
8036 | } else { | |
8037 | ++warnings; | |
8038 | num = warnings; | |
8039 | w = "warning"; | |
8040 | } | |
8041 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
8042 | << "fsck " << w << ": " << o->oid | |
8043 | << " has omap that is not per-pool or pgmeta" | |
8044 | << fsck_dendl; | |
8045 | } | |
8046 | } | |
8047 | if (repairer && | |
8048 | !o->onode.is_perpool_omap() && | |
8049 | !o->onode.is_pgmeta_omap()) { | |
8050 | dout(10) << "fsck converting " << o->oid << " omap to per-pool" << dendl; | |
8051 | bufferlist h; | |
8052 | map<string, bufferlist> kv; | |
8053 | int r = _onode_omap_get(o, &h, &kv); | |
8054 | if (r < 0) { | |
8055 | derr << " got " << r << " " << cpp_strerror(r) << dendl; | |
8056 | } else { | |
8057 | KeyValueDB::Transaction txn = db->get_transaction(); | |
8058 | // remove old keys | |
8059 | const string& old_omap_prefix = o->get_omap_prefix(); | |
8060 | string old_head, old_tail; | |
8061 | o->get_omap_header(&old_head); | |
8062 | o->get_omap_tail(&old_tail); | |
8063 | txn->rm_range_keys(old_omap_prefix, old_head, old_tail); | |
8064 | txn->rmkey(old_omap_prefix, old_tail); | |
8065 | // set flag | |
8066 | o->onode.set_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP); | |
8067 | _record_onode(o, txn); | |
8068 | const string& new_omap_prefix = o->get_omap_prefix(); | |
8069 | // head | |
8070 | if (h.length()) { | |
8071 | string new_head; | |
8072 | o->get_omap_header(&new_head); | |
8073 | txn->set(new_omap_prefix, new_head, h); | |
8074 | } | |
8075 | // tail | |
8076 | string new_tail; | |
8077 | o->get_omap_tail(&new_tail); | |
8078 | bufferlist empty; | |
8079 | txn->set(new_omap_prefix, new_tail, empty); | |
8080 | // values | |
8081 | string final_key; | |
8082 | o->get_omap_key(string(), &final_key); | |
8083 | size_t base_key_len = final_key.size(); | |
8084 | for (auto& i : kv) { | |
8085 | final_key.resize(base_key_len); | |
8086 | final_key += i.first; | |
8087 | txn->set(new_omap_prefix, final_key, i.second); | |
8088 | } | |
8089 | db->submit_transaction_sync(txn); | |
8090 | repairer->inc_repaired(); | |
8091 | } | |
eafe8130 | 8092 | } |
9f95a23c | 8093 | } |
eafe8130 | 8094 | |
9f95a23c TL |
8095 | void BlueStore::_fsck_check_objects(FSCKDepth depth, |
8096 | BlueStore::FSCK_ObjectCtx& ctx) | |
8097 | { | |
eafe8130 | 8098 | auto& errors = ctx.errors; |
eafe8130 TL |
8099 | auto sb_info_lock = ctx.sb_info_lock; |
8100 | auto& sb_info = ctx.sb_info; | |
8101 | auto repairer = ctx.repairer; | |
8102 | ||
8103 | uint64_t_btree_t used_nids; | |
8104 | ||
8105 | size_t processed_myself = 0; | |
8106 | ||
8107 | auto it = db->get_iterator(PREFIX_OBJ); | |
8108 | mempool::bluestore_fsck::list<string> expecting_shards; | |
8109 | if (it) { | |
8110 | const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads; | |
8111 | typedef ShallowFSCKThreadPool::FSCKWorkQueue<256> WQ; | |
8112 | std::unique_ptr<WQ> wq( | |
8113 | new WQ( | |
8114 | "FSCKWorkQueue", | |
8115 | (thread_count ? : 1) * 32, | |
8116 | this, | |
eafe8130 TL |
8117 | sb_info_lock, |
8118 | sb_info, | |
8119 | repairer)); | |
8120 | ||
8121 | ShallowFSCKThreadPool thread_pool(cct, "ShallowFSCKThreadPool", "ShallowFSCK", thread_count); | |
8122 | ||
8123 | thread_pool.add_work_queue(wq.get()); | |
8124 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8125 | //not the best place but let's check anyway | |
8126 | ceph_assert(sb_info_lock); | |
8127 | thread_pool.start(); | |
8128 | } | |
8129 | ||
8130 | //fill global if not overriden below | |
8131 | CollectionRef c; | |
8132 | int64_t pool_id = -1; | |
8133 | spg_t pgid; | |
8134 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8135 | dout(30) << __func__ << " key " | |
8136 | << pretty_binary_string(it->key()) << dendl; | |
8137 | if (is_extent_shard_key(it->key())) { | |
8138 | if (depth == FSCK_SHALLOW) { | |
8139 | continue; | |
8140 | } | |
8141 | while (!expecting_shards.empty() && | |
8142 | expecting_shards.front() < it->key()) { | |
8143 | derr << "fsck error: missing shard key " | |
8144 | << pretty_binary_string(expecting_shards.front()) | |
8145 | << dendl; | |
8146 | ++errors; | |
8147 | expecting_shards.pop_front(); | |
8148 | } | |
8149 | if (!expecting_shards.empty() && | |
8150 | expecting_shards.front() == it->key()) { | |
8151 | // all good | |
8152 | expecting_shards.pop_front(); | |
8153 | continue; | |
8154 | } | |
8155 | ||
8156 | uint32_t offset; | |
8157 | string okey; | |
8158 | get_key_extent_shard(it->key(), &okey, &offset); | |
8159 | derr << "fsck error: stray shard 0x" << std::hex << offset | |
8160 | << std::dec << dendl; | |
8161 | if (expecting_shards.empty()) { | |
8162 | derr << "fsck error: " << pretty_binary_string(it->key()) | |
8163 | << " is unexpected" << dendl; | |
8164 | ++errors; | |
8165 | continue; | |
8166 | } | |
8167 | while (expecting_shards.front() > it->key()) { | |
8168 | derr << "fsck error: saw " << pretty_binary_string(it->key()) | |
8169 | << dendl; | |
8170 | derr << "fsck error: exp " | |
8171 | << pretty_binary_string(expecting_shards.front()) << dendl; | |
8172 | ++errors; | |
8173 | expecting_shards.pop_front(); | |
8174 | if (expecting_shards.empty()) { | |
8175 | break; | |
8176 | } | |
8177 | } | |
8178 | continue; | |
8179 | } | |
8180 | ||
8181 | ghobject_t oid; | |
8182 | int r = get_key_object(it->key(), &oid); | |
8183 | if (r < 0) { | |
8184 | derr << "fsck error: bad object key " | |
8185 | << pretty_binary_string(it->key()) << dendl; | |
8186 | ++errors; | |
8187 | continue; | |
8188 | } | |
8189 | if (!c || | |
8190 | oid.shard_id != pgid.shard || | |
8191 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8192 | !c->contains(oid)) { | |
8193 | c = nullptr; | |
8194 | for (auto& p : coll_map) { | |
8195 | if (p.second->contains(oid)) { | |
8196 | c = p.second; | |
8197 | break; | |
8198 | } | |
8199 | } | |
8200 | if (!c) { | |
8201 | derr << "fsck error: stray object " << oid | |
8202 | << " not owned by any collection" << dendl; | |
8203 | ++errors; | |
8204 | continue; | |
8205 | } | |
8206 | pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
8207 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
8208 | << dendl; | |
8209 | } | |
8210 | ||
8211 | if (depth != FSCK_SHALLOW && | |
8212 | !expecting_shards.empty()) { | |
8213 | for (auto& k : expecting_shards) { | |
8214 | derr << "fsck error: missing shard key " | |
8215 | << pretty_binary_string(k) << dendl; | |
8216 | } | |
8217 | ++errors; | |
8218 | expecting_shards.clear(); | |
8219 | } | |
8220 | ||
8221 | bool queued = false; | |
8222 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8223 | queued = wq->queue( | |
8224 | pool_id, | |
8225 | c, | |
8226 | oid, | |
8227 | it->key(), | |
8228 | it->value()); | |
8229 | } | |
8230 | OnodeRef o; | |
8231 | map<BlobRef, bluestore_blob_t::unused_t> referenced; | |
8232 | ||
8233 | if (!queued) { | |
8234 | ++processed_myself; | |
8235 | ||
8236 | o = fsck_check_objects_shallow( | |
8237 | depth, | |
8238 | pool_id, | |
8239 | c, | |
8240 | oid, | |
8241 | it->key(), | |
8242 | it->value(), | |
9f95a23c | 8243 | &expecting_shards, |
eafe8130 TL |
8244 | &referenced, |
8245 | ctx); | |
8246 | } | |
8247 | ||
8248 | if (depth != FSCK_SHALLOW) { | |
8249 | ceph_assert(o != nullptr); | |
8250 | if (o->onode.nid) { | |
8251 | if (o->onode.nid > nid_max) { | |
8252 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8253 | << " > nid_max " << nid_max << dendl; | |
8254 | ++errors; | |
8255 | } | |
8256 | if (used_nids.count(o->onode.nid)) { | |
8257 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8258 | << " already in use" << dendl; | |
8259 | ++errors; | |
8260 | continue; // go for next object | |
8261 | } | |
8262 | used_nids.insert(o->onode.nid); | |
8263 | } | |
8264 | for (auto& i : referenced) { | |
8265 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
8266 | << std::dec << " for " << *i.first << dendl; | |
8267 | const bluestore_blob_t& blob = i.first->get_blob(); | |
8268 | if (i.second & blob.unused) { | |
8269 | derr << "fsck error: " << oid << " blob claims unused 0x" | |
8270 | << std::hex << blob.unused | |
8271 | << " but extents reference 0x" << i.second << std::dec | |
8272 | << " on blob " << *i.first << dendl; | |
8273 | ++errors; | |
8274 | } | |
8275 | if (blob.has_csum()) { | |
8276 | uint64_t blob_len = blob.get_logical_length(); | |
8277 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused) * 8); | |
8278 | unsigned csum_count = blob.get_csum_count(); | |
8279 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
8280 | for (unsigned p = 0; p < csum_count; ++p) { | |
8281 | unsigned pos = p * csum_chunk_size; | |
8282 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
8283 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
8284 | unsigned mask = 1u << firstbit; | |
8285 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
8286 | mask |= 1u << b; | |
8287 | } | |
8288 | if ((blob.unused & mask) == mask) { | |
8289 | // this csum chunk region is marked unused | |
8290 | if (blob.get_csum_item(p) != 0) { | |
8291 | derr << "fsck error: " << oid | |
8292 | << " blob claims csum chunk 0x" << std::hex << pos | |
8293 | << "~" << csum_chunk_size | |
8294 | << " is unused (mask 0x" << mask << " of unused 0x" | |
8295 | << blob.unused << ") but csum is non-zero 0x" | |
8296 | << blob.get_csum_item(p) << std::dec << " on blob " | |
8297 | << *i.first << dendl; | |
8298 | ++errors; | |
8299 | } | |
8300 | } | |
8301 | } | |
8302 | } | |
8303 | } | |
8304 | // omap | |
8305 | if (o->onode.has_omap()) { | |
9f95a23c TL |
8306 | ceph_assert(ctx.used_omap_head); |
8307 | if (ctx.used_omap_head->count(o->onode.nid)) { | |
8308 | derr << "fsck error: " << o->oid << " omap_head " << o->onode.nid | |
8309 | << " already in use" << dendl; | |
eafe8130 TL |
8310 | ++errors; |
8311 | } else { | |
9f95a23c | 8312 | ctx.used_omap_head->insert(o->onode.nid); |
eafe8130 | 8313 | } |
9f95a23c | 8314 | } // if (o->onode.has_omap()) |
eafe8130 TL |
8315 | if (depth == FSCK_DEEP) { |
8316 | bufferlist bl; | |
8317 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; | |
8318 | uint64_t offset = 0; | |
8319 | do { | |
8320 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
8321 | int r = _do_read(c.get(), o, offset, l, bl, | |
8322 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
8323 | if (r < 0) { | |
8324 | ++errors; | |
8325 | derr << "fsck error: " << oid << std::hex | |
8326 | << " error during read: " | |
8327 | << " " << offset << "~" << l | |
8328 | << " " << cpp_strerror(r) << std::dec | |
8329 | << dendl; | |
8330 | break; | |
8331 | } | |
8332 | offset += l; | |
8333 | } while (offset < o->onode.size); | |
8334 | } // deep | |
8335 | } //if (depth != FSCK_SHALLOW) | |
8336 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8337 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8338 | wq->finalize(thread_pool, ctx); | |
8339 | if (processed_myself) { | |
8340 | // may be needs more threads? | |
8341 | dout(0) << __func__ << " partial offload" | |
8342 | << ", done myself " << processed_myself | |
8343 | << " of " << ctx.num_objects | |
8344 | << "objects, threads " << thread_count | |
8345 | << dendl; | |
8346 | } | |
8347 | } | |
8348 | } // if (it) | |
8349 | } | |
8350 | /** | |
8351 | An overview for currently implemented repair logics | |
8352 | performed in fsck in two stages: detection(+preparation) and commit. | |
8353 | Detection stage (in processing order): | |
8354 | (Issue -> Repair action to schedule) | |
8355 | - Detect undecodable keys for Shared Blobs -> Remove | |
8356 | - Detect undecodable records for Shared Blobs -> Remove | |
8357 | (might trigger missed Shared Blob detection below) | |
8358 | - Detect stray records for Shared Blobs -> Remove | |
8359 | - Detect misreferenced pextents -> Fix | |
8360 | Prepare Bloom-like filter to track cid/oid -> pextent | |
8361 | Prepare list of extents that are improperly referenced | |
8362 | Enumerate Onode records that might use 'misreferenced' pextents | |
8363 | (Bloom-like filter applied to reduce computation) | |
8364 | Per each questinable Onode enumerate all blobs and identify broken ones | |
8365 | (i.e. blobs having 'misreferences') | |
8366 | Rewrite each broken blob data by allocating another extents and | |
8367 | copying data there | |
8368 | If blob is shared - unshare it and mark corresponding Shared Blob | |
8369 | for removal | |
8370 | Release previously allocated space | |
8371 | Update Extent Map | |
8372 | - Detect missed Shared Blobs -> Recreate | |
8373 | - Detect undecodable deferred transaction -> Remove | |
8374 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
8375 | - Detect Freelist Manager's leaked entries -> Mark as free | |
8376 | - Detect statfs inconsistency - Update | |
8377 | Commit stage (separate DB commit per each step): | |
8378 | - Apply leaked FM entries fix | |
8379 | - Apply 'false free' FM entries fix | |
8380 | - Apply 'Remove' actions | |
8381 | - Apply fix for misreference pextents | |
8382 | - Apply Shared Blob recreate | |
8383 | (can be merged with the step above if misreferences were dectected) | |
8384 | - Apply StatFS update | |
8385 | */ | |
8386 | int BlueStore::_fsck(BlueStore::FSCKDepth depth, bool repair) | |
8387 | { | |
8388 | dout(1) << __func__ | |
8389 | << (repair ? " repair" : " check") | |
8390 | << (depth == FSCK_DEEP ? " (deep)" : | |
8391 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8392 | << dendl; | |
8393 | ||
8394 | // in deep mode we need R/W write access to be able to replay deferred ops | |
8395 | bool read_only = !(repair || depth == FSCK_DEEP); | |
8396 | ||
8397 | int r = _open_path(); | |
8398 | if (r < 0) | |
8399 | return r; | |
7c673cae FG |
8400 | r = _open_fsid(false); |
8401 | if (r < 0) | |
8402 | goto out_path; | |
8403 | ||
8404 | r = _read_fsid(&fsid); | |
8405 | if (r < 0) | |
8406 | goto out_fsid; | |
8407 | ||
8408 | r = _lock_fsid(); | |
8409 | if (r < 0) | |
8410 | goto out_fsid; | |
8411 | ||
8412 | r = _open_bdev(false); | |
8413 | if (r < 0) | |
8414 | goto out_fsid; | |
8415 | ||
11fdf7f2 | 8416 | r = _open_db_and_around(read_only); |
7c673cae FG |
8417 | if (r < 0) |
8418 | goto out_bdev; | |
8419 | ||
11fdf7f2 TL |
8420 | if (!read_only) { |
8421 | r = _upgrade_super(); | |
8422 | if (r < 0) { | |
8423 | goto out_db; | |
8424 | } | |
8425 | } | |
7c673cae | 8426 | |
eafe8130 | 8427 | r = _open_collections(); |
7c673cae | 8428 | if (r < 0) |
11fdf7f2 | 8429 | goto out_db; |
7c673cae FG |
8430 | |
8431 | mempool_thread.init(); | |
8432 | ||
11fdf7f2 TL |
8433 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
8434 | // enable in repair or deep mode modes only | |
8435 | if (!read_only) { | |
8436 | _kv_start(); | |
8437 | r = _deferred_replay(); | |
8438 | _kv_stop(); | |
8439 | } | |
7c673cae FG |
8440 | if (r < 0) |
8441 | goto out_scan; | |
8442 | ||
eafe8130 TL |
8443 | r = _fsck_on_open(depth, repair); |
8444 | ||
8445 | out_scan: | |
8446 | mempool_thread.shutdown(); | |
f6b5b4d7 | 8447 | _shutdown_cache(); |
eafe8130 | 8448 | out_db: |
1911f103 | 8449 | _close_db_and_around(false); |
eafe8130 TL |
8450 | out_bdev: |
8451 | _close_bdev(); | |
8452 | out_fsid: | |
8453 | _close_fsid(); | |
8454 | out_path: | |
8455 | _close_path(); | |
8456 | ||
8457 | return r; | |
8458 | } | |
8459 | ||
8460 | int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair) | |
8461 | { | |
8462 | dout(1) << __func__ | |
8463 | << " <<<START>>>" | |
8464 | << (repair ? " repair" : " check") | |
8465 | << (depth == FSCK_DEEP ? " (deep)" : | |
8466 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8467 | << " start" << dendl; | |
8468 | int64_t errors = 0; | |
8469 | int64_t warnings = 0; | |
8470 | unsigned repaired = 0; | |
8471 | ||
8472 | uint64_t_btree_t used_omap_head; | |
eafe8130 TL |
8473 | uint64_t_btree_t used_sbids; |
8474 | ||
8475 | mempool_dynamic_bitset used_blocks; | |
8476 | KeyValueDB::Iterator it; | |
8477 | store_statfs_t expected_store_statfs, actual_statfs; | |
8478 | per_pool_statfs expected_pool_statfs; | |
8479 | ||
8480 | sb_info_map_t sb_info; | |
8481 | ||
8482 | uint64_t num_objects = 0; | |
8483 | uint64_t num_extents = 0; | |
8484 | uint64_t num_blobs = 0; | |
8485 | uint64_t num_spanning_blobs = 0; | |
8486 | uint64_t num_shared_blobs = 0; | |
8487 | uint64_t num_sharded_objects = 0; | |
8488 | BlueStoreRepairer repairer; | |
8489 | ||
8490 | utime_t start = ceph_clock_now(); | |
8491 | ||
8492 | _fsck_collections(&errors); | |
b32b8144 | 8493 | used_blocks.resize(fm->get_alloc_units()); |
9f95a23c | 8494 | apply_for_bitset_range( |
11fdf7f2 | 8495 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), fm->get_alloc_size(), used_blocks, |
7c673cae FG |
8496 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
8497 | bs.set(pos); | |
8498 | } | |
8499 | ); | |
11fdf7f2 TL |
8500 | if (repair) { |
8501 | repairer.get_space_usage_tracker().init( | |
8502 | bdev->get_size(), | |
8503 | min_alloc_size); | |
8504 | } | |
7c673cae FG |
8505 | |
8506 | if (bluefs) { | |
11fdf7f2 TL |
8507 | if( cct->_conf->bluestore_bluefs_db_compatibility) { |
8508 | interval_set<uint64_t> bluefs_extents_db; | |
8509 | bufferlist bl; | |
8510 | db->get(PREFIX_SUPER, "bluefs_extents", &bl); | |
8511 | auto p = bl.cbegin(); | |
8512 | auto prev_errors = errors; | |
8513 | try { | |
8514 | decode(bluefs_extents_db, p); | |
8515 | bluefs_extents_db.union_of(bluefs_extents); | |
8516 | bluefs_extents_db.subtract(bluefs_extents); | |
8517 | if (!bluefs_extents_db.empty()) { | |
8518 | derr << "fsck error: bluefs_extents inconsistency, " | |
8519 | << "downgrade to previous releases might be broken." | |
8520 | << dendl; | |
8521 | ++errors; | |
8522 | } | |
8523 | } | |
8524 | catch (buffer::error& e) { | |
8525 | derr << "fsck error: failed to retrieve bluefs_extents from kv" << dendl; | |
8526 | ++errors; | |
8527 | } | |
8528 | if (errors != prev_errors && repair) { | |
8529 | repairer.fix_bluefs_extents(out_of_sync_fm); | |
8530 | } | |
8531 | } | |
8532 | ||
7c673cae | 8533 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { |
9f95a23c | 8534 | apply_for_bitset_range( |
b32b8144 | 8535 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae FG |
8536 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
8537 | bs.set(pos); | |
1911f103 | 8538 | }); |
7c673cae | 8539 | } |
eafe8130 | 8540 | int r = bluefs->fsck(); |
7c673cae | 8541 | if (r < 0) { |
eafe8130 | 8542 | return r; |
7c673cae FG |
8543 | } |
8544 | if (r > 0) | |
8545 | errors += r; | |
8546 | } | |
8547 | ||
eafe8130 TL |
8548 | if (!per_pool_stat_collection) { |
8549 | const char *w; | |
8550 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_stats) { | |
8551 | w = "error"; | |
8552 | ++errors; | |
8553 | } else { | |
8554 | w = "warning"; | |
8555 | ++warnings; | |
8556 | } | |
8557 | derr << "fsck " << w << ": store not yet converted to per-pool stats" | |
8558 | << dendl; | |
8559 | } | |
9f95a23c TL |
8560 | if (!per_pool_omap) { |
8561 | const char *w; | |
8562 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8563 | w = "error"; | |
8564 | ++errors; | |
8565 | } else { | |
8566 | w = "warning"; | |
8567 | ++warnings; | |
8568 | } | |
8569 | derr << "fsck " << w << ": store not yet converted to per-pool omap" | |
8570 | << dendl; | |
8571 | } | |
8572 | ||
11fdf7f2 | 8573 | // get expected statfs; reset unaffected fields to be able to compare |
7c673cae FG |
8574 | // structs |
8575 | statfs(&actual_statfs); | |
11fdf7f2 TL |
8576 | actual_statfs.total = 0; |
8577 | actual_statfs.internally_reserved = 0; | |
8578 | actual_statfs.available = 0; | |
8579 | actual_statfs.internal_metadata = 0; | |
8580 | actual_statfs.omap_allocated = 0; | |
8581 | ||
eafe8130 TL |
8582 | if (g_conf()->bluestore_debug_fsck_abort) { |
8583 | dout(1) << __func__ << " debug abort" << dendl; | |
8584 | goto out_scan; | |
8585 | } | |
7c673cae | 8586 | // walk PREFIX_OBJ |
eafe8130 TL |
8587 | { |
8588 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
8589 | ceph::mutex sb_info_lock = ceph::make_mutex("BlueStore::fsck::sbinfo_lock"); | |
8590 | BlueStore::FSCK_ObjectCtx ctx( | |
8591 | errors, | |
8592 | warnings, | |
8593 | num_objects, | |
8594 | num_extents, | |
8595 | num_blobs, | |
8596 | num_sharded_objects, | |
8597 | num_spanning_blobs, | |
8598 | &used_blocks, | |
8599 | &used_omap_head, | |
9f95a23c TL |
8600 | //no need for the below lock when in non-shallow mode as |
8601 | // there is no multithreading in this case | |
8602 | depth == FSCK_SHALLOW ? &sb_info_lock : nullptr, | |
eafe8130 TL |
8603 | sb_info, |
8604 | expected_store_statfs, | |
8605 | expected_pool_statfs, | |
8606 | repair ? &repairer : nullptr); | |
9f95a23c TL |
8607 | |
8608 | _fsck_check_objects(depth, ctx); | |
eafe8130 | 8609 | } |
11fdf7f2 | 8610 | |
7c673cae FG |
8611 | dout(1) << __func__ << " checking shared_blobs" << dendl; |
8612 | it = db->get_iterator(PREFIX_SHARED_BLOB); | |
8613 | if (it) { | |
eafe8130 TL |
8614 | // FIXME minor: perhaps simplify for shallow mode? |
8615 | // fill global if not overriden below | |
8616 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 | 8617 | |
7c673cae FG |
8618 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8619 | string key = it->key(); | |
8620 | uint64_t sbid; | |
8621 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 8622 | derr << "fsck error: bad key '" << key |
7c673cae | 8623 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
8624 | if (repair) { |
8625 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8626 | } | |
7c673cae FG |
8627 | ++errors; |
8628 | continue; | |
8629 | } | |
8630 | auto p = sb_info.find(sbid); | |
8631 | if (p == sb_info.end()) { | |
3efd9988 | 8632 | derr << "fsck error: found stray shared blob data for sbid 0x" |
7c673cae | 8633 | << std::hex << sbid << std::dec << dendl; |
11fdf7f2 TL |
8634 | if (repair) { |
8635 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8636 | } | |
7c673cae FG |
8637 | ++errors; |
8638 | } else { | |
8639 | ++num_shared_blobs; | |
8640 | sb_info_t& sbi = p->second; | |
8641 | bluestore_shared_blob_t shared_blob(sbid); | |
8642 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
8643 | auto blp = bl.cbegin(); |
8644 | try { | |
8645 | decode(shared_blob, blp); | |
8646 | } catch (buffer::error& e) { | |
8647 | ++errors; | |
8648 | // Force update and don't report as missing | |
8649 | sbi.updated = sbi.passed = true; | |
8650 | ||
8651 | derr << "fsck error: failed to decode Shared Blob" | |
8652 | << pretty_binary_string(it->key()) << dendl; | |
8653 | if (repair) { | |
8654 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
8655 | << pretty_binary_string(it->key()) | |
8656 | << "', removing" << dendl; | |
8657 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8658 | } | |
8659 | continue; | |
8660 | } | |
7c673cae FG |
8661 | dout(20) << __func__ << " " << *sbi.sb << " " << shared_blob << dendl; |
8662 | if (shared_blob.ref_map != sbi.ref_map) { | |
3efd9988 | 8663 | derr << "fsck error: shared blob 0x" << std::hex << sbid |
11fdf7f2 TL |
8664 | << std::dec << " ref_map " << shared_blob.ref_map |
8665 | << " != expected " << sbi.ref_map << dendl; | |
8666 | sbi.updated = true; // will update later in repair mode only! | |
7c673cae FG |
8667 | ++errors; |
8668 | } | |
8669 | PExtentVector extents; | |
8670 | for (auto &r : shared_blob.ref_map.ref_map) { | |
8671 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
8672 | } | |
eafe8130 | 8673 | if (per_pool_stat_collection || repair) { |
11fdf7f2 TL |
8674 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; |
8675 | } | |
8676 | errors += _fsck_check_extents(sbi.cid, | |
8677 | p->second.oids.front(), | |
7c673cae FG |
8678 | extents, |
8679 | p->second.compressed, | |
b32b8144 FG |
8680 | used_blocks, |
8681 | fm->get_alloc_size(), | |
11fdf7f2 | 8682 | repair ? &repairer : nullptr, |
eafe8130 TL |
8683 | *expected_statfs, |
8684 | depth); | |
11fdf7f2 TL |
8685 | sbi.passed = true; |
8686 | } | |
8687 | } | |
8688 | } // if (it) | |
8689 | ||
8690 | if (repair && repairer.preprocess_misreference(db)) { | |
8691 | ||
8692 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
8693 | auto& space_tracker = repairer.get_space_usage_tracker(); | |
8694 | auto& misref_extents = repairer.get_misreferences(); | |
8695 | interval_set<uint64_t> to_release; | |
8696 | it = db->get_iterator(PREFIX_OBJ); | |
8697 | if (it) { | |
eafe8130 TL |
8698 | // fill global if not overriden below |
8699 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 TL |
8700 | |
8701 | CollectionRef c; | |
8702 | spg_t pgid; | |
8703 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
8704 | bool bypass_rest = false; | |
8705 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
8706 | it->next()) { | |
8707 | dout(30) << __func__ << " key " | |
8708 | << pretty_binary_string(it->key()) << dendl; | |
8709 | if (is_extent_shard_key(it->key())) { | |
8710 | continue; | |
8711 | } | |
8712 | ||
8713 | ghobject_t oid; | |
8714 | int r = get_key_object(it->key(), &oid); | |
8715 | if (r < 0 || !space_tracker.is_used(oid)) { | |
8716 | continue; | |
8717 | } | |
8718 | ||
8719 | if (!c || | |
8720 | oid.shard_id != pgid.shard || | |
8721 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8722 | !c->contains(oid)) { | |
8723 | c = nullptr; | |
8724 | for (auto& p : coll_map) { | |
8725 | if (p.second->contains(oid)) { | |
8726 | c = p.second; | |
8727 | break; | |
8728 | } | |
8729 | } | |
8730 | if (!c) { | |
8731 | continue; | |
8732 | } | |
eafe8130 TL |
8733 | if (per_pool_stat_collection || repair) { |
8734 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
11fdf7f2 TL |
8735 | expected_statfs = &expected_pool_statfs[pool_id]; |
8736 | } | |
8737 | } | |
8738 | if (!space_tracker.is_used(c->cid)) { | |
8739 | continue; | |
8740 | } | |
8741 | ||
8742 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
8743 | << " obj:" << oid << dendl; | |
8744 | ||
eafe8130 TL |
8745 | OnodeRef o; |
8746 | o.reset(Onode::decode(c, oid, it->key(), it->value())); | |
11fdf7f2 TL |
8747 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8748 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
8749 | ||
8750 | for (auto& e : o->extent_map.extent_map) { | |
8751 | blobs.insert(e.blob); | |
8752 | } | |
8753 | bool need_onode_update = false; | |
8754 | bool first_dump = true; | |
8755 | for(auto b : blobs) { | |
8756 | bool broken_blob = false; | |
8757 | auto& pextents = b->dirty_blob().dirty_extents(); | |
8758 | for (auto& e : pextents) { | |
8759 | if (!e.is_valid()) { | |
8760 | continue; | |
8761 | } | |
8762 | // for the sake of simplicity and proper shared blob handling | |
8763 | // always rewrite the whole blob even when it's partially | |
8764 | // misreferenced. | |
8765 | if (misref_extents.intersects(e.offset, e.length)) { | |
8766 | if (first_dump) { | |
8767 | first_dump = false; | |
81eedcae | 8768 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
8769 | } |
8770 | broken_blob = true; | |
8771 | break; | |
8772 | } | |
8773 | } | |
8774 | if (!broken_blob) | |
8775 | continue; | |
8776 | bool compressed = b->get_blob().is_compressed(); | |
8777 | need_onode_update = true; | |
8778 | dout(10) << __func__ | |
8779 | << " fix misreferences in oid:" << oid | |
8780 | << " " << *b << dendl; | |
8781 | uint64_t b_off = 0; | |
8782 | PExtentVector pext_to_release; | |
8783 | pext_to_release.reserve(pextents.size()); | |
8784 | // rewriting all valid pextents | |
8785 | for (auto e = pextents.begin(); e != pextents.end(); | |
8786 | b_off += e->length, e++) { | |
8787 | if (!e->is_valid()) { | |
8788 | continue; | |
8789 | } | |
8790 | PExtentVector exts; | |
8791 | int64_t alloc_len = alloc->allocate(e->length, min_alloc_size, | |
8792 | 0, 0, &exts); | |
eafe8130 | 8793 | if (alloc_len < 0 || alloc_len < (int64_t)e->length) { |
11fdf7f2 TL |
8794 | derr << __func__ |
8795 | << " failed to allocate 0x" << std::hex << e->length | |
eafe8130 | 8796 | << " allocated 0x " << (alloc_len < 0 ? 0 : alloc_len) |
11fdf7f2 TL |
8797 | << " min_alloc_size 0x" << min_alloc_size |
8798 | << " available 0x " << alloc->get_free() | |
8799 | << std::dec << dendl; | |
8800 | if (alloc_len > 0) { | |
8801 | alloc->release(exts); | |
8802 | } | |
8803 | bypass_rest = true; | |
8804 | break; | |
8805 | } | |
8806 | expected_statfs->allocated += e->length; | |
8807 | if (compressed) { | |
8808 | expected_statfs->data_compressed_allocated += e->length; | |
8809 | } | |
8810 | ||
8811 | bufferlist bl; | |
8812 | IOContext ioc(cct, NULL, true); // allow EIO | |
8813 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); | |
8814 | if (r < 0) { | |
8815 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
8816 | <<"~" << e->length << std::dec << dendl; | |
8817 | ceph_abort_msg("read failed, wtf"); | |
8818 | } | |
8819 | pext_to_release.push_back(*e); | |
8820 | e = pextents.erase(e); | |
8821 | e = pextents.insert(e, exts.begin(), exts.end()); | |
8822 | b->get_blob().map_bl( | |
8823 | b_off, bl, | |
8824 | [&](uint64_t offset, bufferlist& t) { | |
8825 | int r = bdev->write(offset, t, false); | |
8826 | ceph_assert(r == 0); | |
8827 | }); | |
8828 | e += exts.size() - 1; | |
8829 | for (auto& p : exts) { | |
8830 | fm->allocate(p.offset, p.length, txn); | |
8831 | } | |
8832 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
8833 | ||
8834 | if (b->get_blob().is_shared()) { | |
8835 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
8836 | ||
8837 | auto sb_it = sb_info.find(b->shared_blob->get_sbid()); | |
8838 | ceph_assert(sb_it != sb_info.end()); | |
8839 | sb_info_t& sbi = sb_it->second; | |
8840 | ||
8841 | for (auto& r : sbi.ref_map.ref_map) { | |
8842 | expected_statfs->allocated -= r.second.length; | |
8843 | if (sbi.compressed) { | |
8844 | // NB: it's crucial to use compressed flag from sb_info_t | |
8845 | // as we originally used that value while accumulating | |
8846 | // expected_statfs | |
8847 | expected_statfs->data_compressed_allocated -= r.second.length; | |
8848 | } | |
8849 | } | |
8850 | sbi.updated = sbi.passed = true; | |
8851 | sbi.ref_map.clear(); | |
8852 | ||
8853 | // relying on blob's pextents to decide what to release. | |
8854 | for (auto& p : pext_to_release) { | |
8855 | to_release.union_insert(p.offset, p.length); | |
8856 | } | |
8857 | } else { | |
8858 | for (auto& p : pext_to_release) { | |
8859 | expected_statfs->allocated -= p.length; | |
8860 | if (compressed) { | |
8861 | expected_statfs->data_compressed_allocated -= p.length; | |
8862 | } | |
8863 | to_release.union_insert(p.offset, p.length); | |
8864 | } | |
8865 | } | |
8866 | if (bypass_rest) { | |
8867 | break; | |
8868 | } | |
8869 | } // for(auto b : blobs) | |
8870 | if (need_onode_update) { | |
8871 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
8872 | _record_onode(o, txn); | |
8873 | } | |
8874 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8875 | ||
8876 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
8877 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
8878 | << "~" << it.get_len() << std::dec << dendl; | |
8879 | fm->release(it.get_start(), it.get_len(), txn); | |
8880 | } | |
8881 | alloc->release(to_release); | |
8882 | to_release.clear(); | |
8883 | } // if (it) { | |
8884 | } //if (repair && repairer.preprocess_misreference()) { | |
8885 | ||
eafe8130 TL |
8886 | if (depth != FSCK_SHALLOW) { |
8887 | for (auto &p : sb_info) { | |
8888 | sb_info_t& sbi = p.second; | |
8889 | if (!sbi.passed) { | |
8890 | derr << "fsck error: missing " << *sbi.sb << dendl; | |
8891 | ++errors; | |
8892 | } | |
8893 | if (repair && (!sbi.passed || sbi.updated)) { | |
8894 | auto sbid = p.first; | |
8895 | if (sbi.ref_map.empty()) { | |
8896 | ceph_assert(sbi.passed); | |
8897 | dout(20) << __func__ << " " << *sbi.sb | |
8898 | << " is empty, removing" << dendl; | |
8899 | repairer.fix_shared_blob(db, sbid, nullptr); | |
8900 | } else { | |
8901 | bufferlist bl; | |
8902 | bluestore_shared_blob_t persistent(sbid, std::move(sbi.ref_map)); | |
8903 | encode(persistent, bl); | |
8904 | dout(20) << __func__ << " " << *sbi.sb | |
8905 | << " is " << bl.length() << " bytes, updating" << dendl; | |
11fdf7f2 | 8906 | |
eafe8130 TL |
8907 | repairer.fix_shared_blob(db, sbid, &bl); |
8908 | } | |
7c673cae FG |
8909 | } |
8910 | } | |
8911 | } | |
11fdf7f2 TL |
8912 | sb_info.clear(); |
8913 | ||
eafe8130 TL |
8914 | // check global stats only if fscking (not repairing) w/o per-pool stats |
8915 | if (!per_pool_stat_collection && | |
8916 | !repair && | |
8917 | !(actual_statfs == expected_store_statfs)) { | |
8918 | derr << "fsck error: actual " << actual_statfs | |
8919 | << " != expected " << expected_store_statfs << dendl; | |
8920 | if (repair) { | |
8921 | repairer.fix_statfs(db, BLUESTORE_GLOBAL_STATFS_KEY, | |
8922 | expected_store_statfs); | |
11fdf7f2 | 8923 | } |
eafe8130 | 8924 | ++errors; |
7c673cae FG |
8925 | } |
8926 | ||
eafe8130 TL |
8927 | dout(1) << __func__ << " checking pool_statfs" << dendl; |
8928 | _fsck_check_pool_statfs(expected_pool_statfs, | |
8929 | errors, warnings, repair ? &repairer : nullptr); | |
8930 | ||
8931 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 8932 | dout(1) << __func__ << " checking for stray omap data " << dendl; |
eafe8130 TL |
8933 | it = db->get_iterator(PREFIX_OMAP); |
8934 | if (it) { | |
9f95a23c | 8935 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8936 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8937 | uint64_t omap_head; | |
8938 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8939 | if (used_omap_head.count(omap_head) == 0 && |
8940 | omap_head != last_omap_head) { | |
8941 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8942 | << "fsck error: found stray omap data on omap_head " | |
8943 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head)<< fsck_dendl; | |
eafe8130 | 8944 | ++errors; |
9f95a23c | 8945 | last_omap_head = omap_head; |
eafe8130 | 8946 | } |
7c673cae FG |
8947 | } |
8948 | } | |
eafe8130 TL |
8949 | it = db->get_iterator(PREFIX_PGMETA_OMAP); |
8950 | if (it) { | |
9f95a23c | 8951 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8952 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8953 | uint64_t omap_head; | |
8954 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8955 | if (used_omap_head.count(omap_head) == 0 && |
8956 | omap_head != last_omap_head) { | |
8957 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8958 | << "fsck error: found stray (pgmeta) omap data on omap_head " | |
8959 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8960 | last_omap_head = omap_head; | |
eafe8130 TL |
8961 | ++errors; |
8962 | } | |
11fdf7f2 TL |
8963 | } |
8964 | } | |
9f95a23c TL |
8965 | it = db->get_iterator(PREFIX_PERPOOL_OMAP); |
8966 | if (it) { | |
8967 | uint64_t last_omap_head = 0; | |
8968 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8969 | uint64_t pool; | |
8970 | uint64_t omap_head; | |
8971 | string k = it->key(); | |
8972 | const char *c = k.c_str(); | |
8973 | c = _key_decode_u64(c, &pool); | |
8974 | c = _key_decode_u64(c, &omap_head); | |
8975 | if (used_omap_head.count(omap_head) == 0 && | |
8976 | omap_head != last_omap_head) { | |
8977 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8978 | << "fsck error: found stray (per-pool) omap data on omap_head " | |
8979 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8980 | ++errors; | |
8981 | last_omap_head = omap_head; | |
8982 | } | |
8983 | } | |
8984 | } | |
eafe8130 TL |
8985 | dout(1) << __func__ << " checking deferred events" << dendl; |
8986 | it = db->get_iterator(PREFIX_DEFERRED); | |
8987 | if (it) { | |
8988 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8989 | bufferlist bl = it->value(); | |
8990 | auto p = bl.cbegin(); | |
8991 | bluestore_deferred_transaction_t wt; | |
8992 | try { | |
8993 | decode(wt, p); | |
8994 | } catch (buffer::error& e) { | |
8995 | derr << "fsck error: failed to decode deferred txn " | |
8996 | << pretty_binary_string(it->key()) << dendl; | |
8997 | if (repair) { | |
8998 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
8999 | << pretty_binary_string(it->key()) | |
9000 | << "', removing" << dendl; | |
9001 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
9002 | } | |
9003 | continue; | |
9004 | } | |
9005 | dout(20) << __func__ << " deferred " << wt.seq | |
9006 | << " ops " << wt.ops.size() | |
9007 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
9008 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
9f95a23c | 9009 | apply_for_bitset_range( |
eafe8130 TL |
9010 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
9011 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
9012 | bs.set(pos); |
9013 | } | |
9014 | ); | |
9015 | } | |
7c673cae | 9016 | } |
eafe8130 TL |
9017 | } |
9018 | ||
9019 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
9020 | { | |
9021 | // remove bluefs_extents from used set since the freelist doesn't | |
9022 | // know they are allocated. | |
9023 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
9f95a23c | 9024 | apply_for_bitset_range( |
b32b8144 | 9025 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae | 9026 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
eafe8130 | 9027 | bs.reset(pos); |
7c673cae FG |
9028 | } |
9029 | ); | |
9030 | } | |
eafe8130 TL |
9031 | fm->enumerate_reset(); |
9032 | uint64_t offset, length; | |
9033 | while (fm->enumerate_next(db, &offset, &length)) { | |
9034 | bool intersects = false; | |
9f95a23c | 9035 | apply_for_bitset_range( |
eafe8130 TL |
9036 | offset, length, fm->get_alloc_size(), used_blocks, |
9037 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
9038 | if (bs.test(pos)) { |
9039 | if (offset == SUPER_RESERVED && | |
9040 | length == min_alloc_size - SUPER_RESERVED) { | |
9041 | // this is due to the change just after luminous to min_alloc_size | |
9042 | // granularity allocations, and our baked in assumption at the top | |
9043 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
9044 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
9045 | // since we will never allocate this region below min_alloc_size. | |
9046 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
9047 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
9048 | << length << std::dec << dendl; | |
9049 | } else { | |
9050 | intersects = true; | |
9051 | if (repair) { | |
9052 | repairer.fix_false_free(db, fm, | |
9053 | pos * min_alloc_size, | |
9054 | min_alloc_size); | |
9055 | } | |
11fdf7f2 | 9056 | } |
eafe8130 TL |
9057 | } else { |
9058 | bs.set(pos); | |
9059 | } | |
7c673cae | 9060 | } |
eafe8130 TL |
9061 | ); |
9062 | if (intersects) { | |
9063 | derr << "fsck error: free extent 0x" << std::hex << offset | |
9064 | << "~" << length << std::dec | |
9065 | << " intersects allocated blocks" << dendl; | |
9066 | ++errors; | |
7c673cae | 9067 | } |
b5b8bbf5 | 9068 | } |
eafe8130 TL |
9069 | fm->enumerate_reset(); |
9070 | size_t count = used_blocks.count(); | |
9071 | if (used_blocks.size() != count) { | |
9072 | ceph_assert(used_blocks.size() > count); | |
9073 | used_blocks.flip(); | |
9074 | size_t start = used_blocks.find_first(); | |
9075 | while (start != decltype(used_blocks)::npos) { | |
9076 | size_t cur = start; | |
9077 | while (true) { | |
9078 | size_t next = used_blocks.find_next(cur); | |
9079 | if (next != cur + 1) { | |
9080 | ++errors; | |
9081 | derr << "fsck error: leaked extent 0x" << std::hex | |
9082 | << ((uint64_t)start * fm->get_alloc_size()) << "~" | |
9083 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
9084 | << dendl; | |
9085 | if (repair) { | |
9086 | repairer.fix_leaked(db, | |
9087 | fm, | |
9088 | start * min_alloc_size, | |
9089 | (cur + 1 - start) * min_alloc_size); | |
9090 | } | |
9091 | start = next; | |
9092 | break; | |
11fdf7f2 | 9093 | } |
eafe8130 | 9094 | cur = next; |
b5b8bbf5 | 9095 | } |
eafe8130 TL |
9096 | } |
9097 | used_blocks.flip(); | |
b5b8bbf5 | 9098 | } |
7c673cae FG |
9099 | } |
9100 | } | |
11fdf7f2 | 9101 | if (repair) { |
9f95a23c TL |
9102 | if (!per_pool_omap) { |
9103 | dout(5) << __func__ << " marking per_pool_omap=1" << dendl; | |
9104 | repairer.fix_per_pool_omap(db); | |
9105 | } | |
9106 | ||
11fdf7f2 TL |
9107 | dout(5) << __func__ << " applying repair results" << dendl; |
9108 | repaired = repairer.apply(db); | |
9109 | dout(5) << __func__ << " repair applied" << dendl; | |
9110 | } | |
7c673cae | 9111 | |
eafe8130 | 9112 | out_scan: |
7c673cae FG |
9113 | dout(2) << __func__ << " " << num_objects << " objects, " |
9114 | << num_sharded_objects << " of them sharded. " | |
9115 | << dendl; | |
9116 | dout(2) << __func__ << " " << num_extents << " extents to " | |
9117 | << num_blobs << " blobs, " | |
9118 | << num_spanning_blobs << " spanning, " | |
9119 | << num_shared_blobs << " shared." | |
9120 | << dendl; | |
9121 | ||
9122 | utime_t duration = ceph_clock_now() - start; | |
9f95a23c TL |
9123 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " |
9124 | << warnings << " warnings, " | |
9125 | << repaired << " repaired, " | |
9126 | << (errors + warnings - (int)repaired) << " remaining in " | |
7c673cae | 9127 | << duration << " seconds" << dendl; |
9f95a23c TL |
9128 | |
9129 | // In non-repair mode we should return error count only as | |
9130 | // it indicates if store status is OK. | |
9131 | // In repair mode both errors and warnings are taken into account | |
9132 | // since repaired counter relates to them both. | |
9133 | return repair ? errors + warnings - (int)repaired : errors; | |
11fdf7f2 TL |
9134 | } |
9135 | ||
9136 | /// methods to inject various errors fsck can repair | |
9137 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
9138 | const bufferlist& bl) | |
9139 | { | |
9140 | KeyValueDB::Transaction txn; | |
9141 | txn = db->get_transaction(); | |
9142 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
9143 | db->submit_transaction_sync(txn); | |
9144 | }; | |
9145 | ||
9146 | void BlueStore::inject_leaked(uint64_t len) | |
9147 | { | |
9148 | KeyValueDB::Transaction txn; | |
9149 | txn = db->get_transaction(); | |
9150 | ||
9151 | PExtentVector exts; | |
9152 | int64_t alloc_len = alloc->allocate(len, min_alloc_size, | |
9153 | min_alloc_size * 256, 0, &exts); | |
9154 | ceph_assert(alloc_len >= (int64_t)len); | |
9155 | for (auto& p : exts) { | |
9156 | fm->allocate(p.offset, p.length, txn); | |
9157 | } | |
9158 | db->submit_transaction_sync(txn); | |
9159 | } | |
9160 | ||
9161 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
9162 | { | |
9163 | KeyValueDB::Transaction txn; | |
9164 | OnodeRef o; | |
9165 | CollectionRef c = _get_collection(cid); | |
9166 | ceph_assert(c); | |
9167 | { | |
9f95a23c | 9168 | std::unique_lock l{c->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9169 | o = c->get_onode(oid, false); |
9170 | ceph_assert(o); | |
9171 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
9172 | } | |
9173 | ||
9174 | bool injected = false; | |
9175 | txn = db->get_transaction(); | |
9176 | auto& em = o->extent_map.extent_map; | |
9177 | std::vector<const PExtentVector*> v; | |
9178 | if (em.size()) { | |
9179 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
9180 | } | |
9181 | if (em.size() > 1) { | |
9182 | auto it = em.end(); | |
9183 | --it; | |
9184 | v.push_back(&(it->blob->get_blob().get_extents())); | |
9185 | } | |
9186 | for (auto pext : v) { | |
9187 | if (pext->size()) { | |
9188 | auto p = pext->begin(); | |
9189 | while (p != pext->end()) { | |
9190 | if (p->is_valid()) { | |
9191 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
9192 | << "~" << p->length << std::dec << dendl; | |
9193 | fm->release(p->offset, p->length, txn); | |
9194 | injected = true; | |
9195 | break; | |
9196 | } | |
9197 | ++p; | |
9198 | } | |
9199 | } | |
9200 | } | |
9201 | ceph_assert(injected); | |
9202 | db->submit_transaction_sync(txn); | |
9203 | } | |
9204 | ||
9f95a23c TL |
9205 | void BlueStore::inject_legacy_omap() |
9206 | { | |
9207 | dout(1) << __func__ << dendl; | |
9208 | per_pool_omap = false; | |
9209 | KeyValueDB::Transaction txn; | |
9210 | txn = db->get_transaction(); | |
9211 | txn->rmkey(PREFIX_SUPER, "per_pool_omap"); | |
9212 | db->submit_transaction_sync(txn); | |
9213 | } | |
9214 | ||
9215 | void BlueStore::inject_legacy_omap(coll_t cid, ghobject_t oid) | |
9216 | { | |
9217 | dout(1) << __func__ << " " | |
9218 | << cid << " " << oid | |
9219 | <<dendl; | |
9220 | KeyValueDB::Transaction txn; | |
9221 | OnodeRef o; | |
9222 | CollectionRef c = _get_collection(cid); | |
9223 | ceph_assert(c); | |
9224 | { | |
9225 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
9226 | o = c->get_onode(oid, false); | |
9227 | ceph_assert(o); | |
9228 | } | |
9229 | o->onode.clear_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP | bluestore_onode_t::FLAG_PGMETA_OMAP); | |
9230 | txn = db->get_transaction(); | |
9231 | _record_onode(o, txn); | |
9232 | db->submit_transaction_sync(txn); | |
9233 | } | |
9234 | ||
9235 | ||
11fdf7f2 TL |
9236 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) |
9237 | { | |
9238 | BlueStoreRepairer repairer; | |
9239 | repairer.fix_statfs(db, key, new_statfs); | |
9240 | repairer.apply(db); | |
9241 | } | |
9242 | ||
eafe8130 TL |
9243 | void BlueStore::inject_global_statfs(const store_statfs_t& new_statfs) |
9244 | { | |
9245 | KeyValueDB::Transaction t = db->get_transaction(); | |
9246 | volatile_statfs v; | |
9247 | v = new_statfs; | |
9248 | bufferlist bl; | |
9249 | v.encode(bl); | |
9250 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
9251 | db->submit_transaction_sync(t); | |
9252 | } | |
9253 | ||
11fdf7f2 TL |
9254 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, |
9255 | coll_t cid2, ghobject_t oid2, | |
9256 | uint64_t offset) | |
9257 | { | |
9258 | OnodeRef o1; | |
9259 | CollectionRef c1 = _get_collection(cid1); | |
9260 | ceph_assert(c1); | |
9261 | { | |
9f95a23c | 9262 | std::unique_lock l{c1->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9263 | o1 = c1->get_onode(oid1, false); |
9264 | ceph_assert(o1); | |
9265 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9266 | } | |
9267 | OnodeRef o2; | |
9268 | CollectionRef c2 = _get_collection(cid2); | |
9269 | ceph_assert(c2); | |
9270 | { | |
9f95a23c | 9271 | std::unique_lock l{c2->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9272 | o2 = c2->get_onode(oid2, false); |
9273 | ceph_assert(o2); | |
9274 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9275 | } | |
9276 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
9277 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
9278 | ||
9279 | // require onode/extent layout to be the same (and simple) | |
9280 | // to make things easier | |
9281 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
9282 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
9283 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
9284 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
9285 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
9286 | ceph_assert(e1.length == e2.length); | |
9287 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
9288 | ||
9289 | KeyValueDB::Transaction txn; | |
9290 | txn = db->get_transaction(); | |
9291 | ||
9292 | // along with misreference error this will create space leaks errors | |
9293 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
9294 | o2->extent_map.dirty_range(offset, e2.length); | |
9295 | o2->extent_map.update(txn, false); | |
9296 | ||
9297 | _record_onode(o2, txn); | |
9298 | db->submit_transaction_sync(txn); | |
7c673cae FG |
9299 | } |
9300 | ||
9301 | void BlueStore::collect_metadata(map<string,string> *pm) | |
9302 | { | |
9303 | dout(10) << __func__ << dendl; | |
9304 | bdev->collect_metadata("bluestore_bdev_", pm); | |
9305 | if (bluefs) { | |
9306 | (*pm)["bluefs"] = "1"; | |
9f95a23c TL |
9307 | // this value is for backward compatibility only |
9308 | (*pm)["bluefs_single_shared_device"] = \ | |
9309 | stringify((int)bluefs_layout.single_shared_device()); | |
9310 | (*pm)["bluefs_dedicated_db"] = \ | |
9311 | stringify((int)bluefs_layout.dedicated_db); | |
9312 | (*pm)["bluefs_dedicated_wal"] = \ | |
9313 | stringify((int)bluefs_layout.dedicated_wal); | |
9314 | bluefs->collect_metadata(pm, bluefs_layout.shared_bdev); | |
7c673cae FG |
9315 | } else { |
9316 | (*pm)["bluefs"] = "0"; | |
9317 | } | |
11fdf7f2 TL |
9318 | |
9319 | // report numa mapping for underlying devices | |
9320 | int node = -1; | |
9321 | set<int> nodes; | |
9322 | set<string> failed; | |
9323 | int r = get_numa_node(&node, &nodes, &failed); | |
9324 | if (r >= 0) { | |
9325 | if (!failed.empty()) { | |
9326 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
9327 | } | |
9328 | if (!nodes.empty()) { | |
9329 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
9330 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
9331 | } | |
9332 | if (node >= 0) { | |
9333 | (*pm)["objectstore_numa_node"] = stringify(node); | |
9334 | } | |
9335 | } | |
9336 | } | |
9337 | ||
9338 | int BlueStore::get_numa_node( | |
9339 | int *final_node, | |
9340 | set<int> *out_nodes, | |
9341 | set<string> *out_failed) | |
9342 | { | |
9343 | int node = -1; | |
9344 | set<string> devices; | |
9345 | get_devices(&devices); | |
9346 | set<int> nodes; | |
9347 | set<string> failed; | |
9348 | for (auto& devname : devices) { | |
9349 | int n; | |
9350 | BlkDev bdev(devname); | |
9351 | int r = bdev.get_numa_node(&n); | |
9352 | if (r < 0) { | |
9353 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
9354 | << dendl; | |
9355 | failed.insert(devname); | |
9356 | continue; | |
9357 | } | |
9358 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
9359 | << dendl; | |
9360 | nodes.insert(n); | |
9361 | if (node < 0) { | |
9362 | node = n; | |
9363 | } | |
9364 | } | |
9365 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
9366 | *final_node = node; | |
9367 | } | |
9368 | if (out_nodes) { | |
9369 | *out_nodes = nodes; | |
9370 | } | |
9371 | if (out_failed) { | |
9372 | *out_failed = failed; | |
9373 | } | |
9374 | return 0; | |
9375 | } | |
9376 | ||
9377 | int BlueStore::get_devices(set<string> *ls) | |
9378 | { | |
9379 | if (bdev) { | |
9380 | bdev->get_devices(ls); | |
9381 | if (bluefs) { | |
9382 | bluefs->get_devices(ls); | |
9383 | } | |
9384 | return 0; | |
9385 | } | |
9386 | ||
9387 | // grumble, we haven't started up yet. | |
9388 | int r = _open_path(); | |
9389 | if (r < 0) | |
9390 | goto out; | |
9391 | r = _open_fsid(false); | |
9392 | if (r < 0) | |
9393 | goto out_path; | |
9394 | r = _read_fsid(&fsid); | |
9395 | if (r < 0) | |
9396 | goto out_fsid; | |
9397 | r = _lock_fsid(); | |
9398 | if (r < 0) | |
9399 | goto out_fsid; | |
9400 | r = _open_bdev(false); | |
9401 | if (r < 0) | |
9402 | goto out_fsid; | |
9403 | r = _minimal_open_bluefs(false); | |
9404 | if (r < 0) | |
9405 | goto out_bdev; | |
9406 | bdev->get_devices(ls); | |
9407 | if (bluefs) { | |
9408 | bluefs->get_devices(ls); | |
9409 | } | |
9410 | r = 0; | |
9411 | _minimal_close_bluefs(); | |
9412 | out_bdev: | |
9413 | _close_bdev(); | |
9414 | out_fsid: | |
9415 | _close_fsid(); | |
9416 | out_path: | |
9417 | _close_path(); | |
9418 | out: | |
9419 | return r; | |
7c673cae FG |
9420 | } |
9421 | ||
11fdf7f2 | 9422 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
9423 | { |
9424 | buf->reset(); | |
11fdf7f2 | 9425 | |
9f95a23c TL |
9426 | buf->omap_allocated = |
9427 | db->estimate_prefix_size(PREFIX_OMAP, string()) + | |
9428 | db->estimate_prefix_size(PREFIX_PERPOOL_OMAP, string()); | |
11fdf7f2 TL |
9429 | |
9430 | uint64_t bfree = alloc->get_free(); | |
7c673cae FG |
9431 | |
9432 | if (bluefs) { | |
9f95a23c TL |
9433 | int64_t bluefs_total = bluefs->get_total(bluefs_layout.shared_bdev); |
9434 | int64_t bluefs_free = bluefs->get_free(bluefs_layout.shared_bdev); | |
94b18763 FG |
9435 | // part of our shared device is "free" according to BlueFS, but we |
9436 | // can't touch bluestore_bluefs_min of it. | |
9437 | int64_t shared_available = std::min( | |
11fdf7f2 TL |
9438 | bluefs_free, |
9439 | int64_t(bluefs_total - cct->_conf->bluestore_bluefs_min)); | |
9440 | buf->internally_reserved = bluefs_total - shared_available; | |
94b18763 | 9441 | if (shared_available > 0) { |
11fdf7f2 TL |
9442 | bfree += shared_available; |
9443 | } | |
9444 | // include dedicated db, too, if that isn't the shared device. | |
9f95a23c | 9445 | if (bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 9446 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); |
7c673cae | 9447 | } |
11fdf7f2 TL |
9448 | // call any non-omap bluefs space "internal metadata" |
9449 | buf->internal_metadata = | |
9450 | std::max(bluefs->get_used(), (uint64_t)cct->_conf->bluestore_bluefs_min) | |
9451 | - buf->omap_allocated; | |
7c673cae FG |
9452 | } |
9453 | ||
11fdf7f2 TL |
9454 | uint64_t thin_total, thin_avail; |
9455 | if (bdev->get_thin_utilization(&thin_total, &thin_avail)) { | |
9456 | buf->total += thin_total; | |
9457 | ||
9458 | // we are limited by both the size of the virtual device and the | |
9459 | // underlying physical device. | |
9460 | bfree = std::min(bfree, thin_avail); | |
9461 | ||
9462 | buf->allocated = thin_total - thin_avail; | |
9463 | } else { | |
9464 | buf->total += bdev->get_size(); | |
9465 | } | |
9466 | buf->available = bfree; | |
9467 | } | |
9468 | ||
9469 | int BlueStore::statfs(struct store_statfs_t *buf, | |
9470 | osd_alert_list_t* alerts) | |
9471 | { | |
9472 | if (alerts) { | |
9473 | alerts->clear(); | |
9474 | _log_alerts(*alerts); | |
9475 | } | |
9476 | _get_statfs_overall(buf); | |
31f18b77 | 9477 | { |
11fdf7f2 | 9478 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 9479 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
9480 | buf->data_stored = vstatfs.stored(); |
9481 | buf->data_compressed = vstatfs.compressed(); | |
9482 | buf->data_compressed_original = vstatfs.compressed_original(); | |
9483 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
9484 | } | |
9485 | ||
9486 | dout(20) << __func__ << " " << *buf << dendl; | |
9487 | return 0; | |
9488 | } | |
9489 | ||
9f95a23c TL |
9490 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
9491 | bool *out_per_pool_omap) | |
11fdf7f2 TL |
9492 | { |
9493 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 9494 | |
11fdf7f2 TL |
9495 | if (!per_pool_stat_collection) { |
9496 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
9497 | return -ENOTSUP; | |
7c673cae | 9498 | } |
11fdf7f2 | 9499 | buf->reset(); |
7c673cae | 9500 | |
11fdf7f2 TL |
9501 | { |
9502 | std::lock_guard l(vstatfs_lock); | |
9503 | osd_pools[pool_id].publish(buf); | |
9504 | } | |
9f95a23c TL |
9505 | |
9506 | string key_prefix; | |
9507 | _key_encode_u64(pool_id, &key_prefix); | |
9508 | buf->omap_allocated = db->estimate_prefix_size(PREFIX_PERPOOL_OMAP, | |
9509 | key_prefix); | |
9510 | *out_per_pool_omap = per_pool_omap; | |
9511 | ||
11fdf7f2 | 9512 | dout(10) << __func__ << *buf << dendl; |
7c673cae FG |
9513 | return 0; |
9514 | } | |
9515 | ||
81eedcae TL |
9516 | void BlueStore::_check_legacy_statfs_alert() |
9517 | { | |
9518 | string s; | |
9519 | if (!per_pool_stat_collection && | |
eafe8130 | 9520 | cct->_conf->bluestore_warn_on_legacy_statfs) { |
81eedcae TL |
9521 | s = "legacy statfs reporting detected, " |
9522 | "suggest to run store repair to get consistent statistic reports"; | |
9523 | } | |
9524 | std::lock_guard l(qlock); | |
9525 | legacy_statfs_alert = s; | |
9526 | } | |
9527 | ||
9f95a23c TL |
9528 | void BlueStore::_check_no_per_pool_omap_alert() |
9529 | { | |
9530 | string s; | |
9531 | if (!per_pool_omap && | |
9532 | cct->_conf->bluestore_warn_on_no_per_pool_omap) { | |
9533 | s = "legacy (not per-pool) omap detected, " | |
9534 | "suggest to run store repair to measure per-pool omap usage"; | |
9535 | } | |
9536 | std::lock_guard l(qlock); | |
9537 | no_per_pool_omap_alert = s; | |
9538 | } | |
9539 | ||
7c673cae FG |
9540 | // --------------- |
9541 | // cache | |
9542 | ||
9543 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
9544 | { | |
9f95a23c | 9545 | std::shared_lock l(coll_lock); |
7c673cae FG |
9546 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); |
9547 | if (cp == coll_map.end()) | |
9548 | return CollectionRef(); | |
9549 | return cp->second; | |
9550 | } | |
9551 | ||
9552 | void BlueStore::_queue_reap_collection(CollectionRef& c) | |
9553 | { | |
9554 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
9555 | // _reap_collections and this in the same thread, |
9556 | // so no need a lock. | |
7c673cae FG |
9557 | removed_collections.push_back(c); |
9558 | } | |
9559 | ||
9560 | void BlueStore::_reap_collections() | |
9561 | { | |
94b18763 | 9562 | |
7c673cae FG |
9563 | list<CollectionRef> removed_colls; |
9564 | { | |
94b18763 FG |
9565 | // _queue_reap_collection and this in the same thread. |
9566 | // So no need a lock. | |
9567 | if (!removed_collections.empty()) | |
9568 | removed_colls.swap(removed_collections); | |
9569 | else | |
9570 | return; | |
7c673cae FG |
9571 | } |
9572 | ||
94b18763 FG |
9573 | list<CollectionRef>::iterator p = removed_colls.begin(); |
9574 | while (p != removed_colls.end()) { | |
7c673cae FG |
9575 | CollectionRef c = *p; |
9576 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
9577 | if (c->onode_map.map_any([&](OnodeRef o) { | |
11fdf7f2 | 9578 | ceph_assert(!o->exists); |
7c673cae FG |
9579 | if (o->flushing_count.load()) { |
9580 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
9581 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 9582 | return true; |
7c673cae | 9583 | } |
94b18763 | 9584 | return false; |
7c673cae | 9585 | })) { |
94b18763 | 9586 | ++p; |
7c673cae FG |
9587 | continue; |
9588 | } | |
9589 | c->onode_map.clear(); | |
94b18763 | 9590 | p = removed_colls.erase(p); |
7c673cae FG |
9591 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
9592 | } | |
94b18763 | 9593 | if (removed_colls.empty()) { |
7c673cae | 9594 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
9595 | } else { |
9596 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
9597 | } |
9598 | } | |
9599 | ||
9600 | void BlueStore::_update_cache_logger() | |
9601 | { | |
9602 | uint64_t num_onodes = 0; | |
9f95a23c | 9603 | uint64_t num_pinned_onodes = 0; |
7c673cae FG |
9604 | uint64_t num_extents = 0; |
9605 | uint64_t num_blobs = 0; | |
9606 | uint64_t num_buffers = 0; | |
9607 | uint64_t num_buffer_bytes = 0; | |
9f95a23c TL |
9608 | for (auto c : onode_cache_shards) { |
9609 | c->add_stats(&num_onodes, &num_pinned_onodes); | |
9610 | } | |
9611 | for (auto c : buffer_cache_shards) { | |
9612 | c->add_stats(&num_extents, &num_blobs, | |
9613 | &num_buffers, &num_buffer_bytes); | |
7c673cae FG |
9614 | } |
9615 | logger->set(l_bluestore_onodes, num_onodes); | |
9f95a23c | 9616 | logger->set(l_bluestore_pinned_onodes, num_pinned_onodes); |
7c673cae FG |
9617 | logger->set(l_bluestore_extents, num_extents); |
9618 | logger->set(l_bluestore_blobs, num_blobs); | |
9619 | logger->set(l_bluestore_buffers, num_buffers); | |
9620 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
9621 | } | |
9622 | ||
9623 | // --------------- | |
9624 | // read operations | |
9625 | ||
9626 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
9627 | { | |
9628 | return _get_collection(cid); | |
9629 | } | |
9630 | ||
11fdf7f2 TL |
9631 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
9632 | const coll_t& cid) | |
7c673cae | 9633 | { |
9f95a23c TL |
9634 | std::unique_lock l{coll_lock}; |
9635 | auto c = ceph::make_ref<Collection>( | |
11fdf7f2 | 9636 | this, |
9f95a23c TL |
9637 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
9638 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
11fdf7f2 TL |
9639 | cid); |
9640 | new_coll_map[cid] = c; | |
9f95a23c | 9641 | _osr_attach(c.get()); |
11fdf7f2 TL |
9642 | return c; |
9643 | } | |
9644 | ||
9645 | void BlueStore::set_collection_commit_queue( | |
9646 | const coll_t& cid, | |
9647 | ContextQueue *commit_queue) | |
9648 | { | |
9649 | if (commit_queue) { | |
9f95a23c | 9650 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
9651 | if (coll_map.count(cid)) { |
9652 | coll_map[cid]->commit_queue = commit_queue; | |
9653 | } else if (new_coll_map.count(cid)) { | |
9654 | new_coll_map[cid]->commit_queue = commit_queue; | |
9655 | } | |
9656 | } | |
7c673cae FG |
9657 | } |
9658 | ||
11fdf7f2 | 9659 | |
7c673cae FG |
9660 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
9661 | { | |
9662 | Collection *c = static_cast<Collection *>(c_.get()); | |
9663 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
9664 | if (!c->exists) | |
9665 | return false; | |
9666 | ||
9667 | bool r = true; | |
9668 | ||
9669 | { | |
9f95a23c | 9670 | std::shared_lock l(c->lock); |
7c673cae FG |
9671 | OnodeRef o = c->get_onode(oid, false); |
9672 | if (!o || !o->exists) | |
9673 | r = false; | |
9674 | } | |
9675 | ||
7c673cae FG |
9676 | return r; |
9677 | } | |
9678 | ||
7c673cae FG |
9679 | int BlueStore::stat( |
9680 | CollectionHandle &c_, | |
9681 | const ghobject_t& oid, | |
9682 | struct stat *st, | |
9683 | bool allow_eio) | |
9684 | { | |
9685 | Collection *c = static_cast<Collection *>(c_.get()); | |
9686 | if (!c->exists) | |
9687 | return -ENOENT; | |
9688 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
9689 | ||
9690 | { | |
9f95a23c | 9691 | std::shared_lock l(c->lock); |
7c673cae FG |
9692 | OnodeRef o = c->get_onode(oid, false); |
9693 | if (!o || !o->exists) | |
9694 | return -ENOENT; | |
9695 | st->st_size = o->onode.size; | |
9696 | st->st_blksize = 4096; | |
9697 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
9698 | st->st_nlink = 1; | |
9699 | } | |
9700 | ||
7c673cae FG |
9701 | int r = 0; |
9702 | if (_debug_mdata_eio(oid)) { | |
9703 | r = -EIO; | |
9704 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9705 | } | |
9706 | return r; | |
9707 | } | |
9708 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 9709 | CollectionHandle& ch, |
7c673cae FG |
9710 | const pool_opts_t& opts) |
9711 | { | |
7c673cae | 9712 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 9713 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
9714 | if (!c->exists) |
9715 | return -ENOENT; | |
9f95a23c | 9716 | std::unique_lock l{c->lock}; |
7c673cae FG |
9717 | c->pool_opts = opts; |
9718 | return 0; | |
9719 | } | |
9720 | ||
7c673cae FG |
9721 | int BlueStore::read( |
9722 | CollectionHandle &c_, | |
9723 | const ghobject_t& oid, | |
9724 | uint64_t offset, | |
9725 | size_t length, | |
9726 | bufferlist& bl, | |
224ce89b | 9727 | uint32_t op_flags) |
7c673cae | 9728 | { |
11fdf7f2 | 9729 | auto start = mono_clock::now(); |
7c673cae FG |
9730 | Collection *c = static_cast<Collection *>(c_.get()); |
9731 | const coll_t &cid = c->get_cid(); | |
9732 | dout(15) << __func__ << " " << cid << " " << oid | |
9733 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9734 | << dendl; | |
9735 | if (!c->exists) | |
9736 | return -ENOENT; | |
9737 | ||
9738 | bl.clear(); | |
9739 | int r; | |
9740 | { | |
9f95a23c | 9741 | std::shared_lock l(c->lock); |
11fdf7f2 | 9742 | auto start1 = mono_clock::now(); |
7c673cae | 9743 | OnodeRef o = c->get_onode(oid, false); |
494da23a TL |
9744 | log_latency("get_onode@read", |
9745 | l_bluestore_read_onode_meta_lat, | |
9746 | mono_clock::now() - start1, | |
9747 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9748 | if (!o || !o->exists) { |
9749 | r = -ENOENT; | |
9750 | goto out; | |
9751 | } | |
9752 | ||
9753 | if (offset == length && offset == 0) | |
9754 | length = o->onode.size; | |
9755 | ||
9756 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
9757 | if (r == -EIO) { |
9758 | logger->inc(l_bluestore_read_eio); | |
9759 | } | |
7c673cae FG |
9760 | } |
9761 | ||
9762 | out: | |
28e407b8 | 9763 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
9764 | r = -EIO; |
9765 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
9766 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
9767 | cct->_conf->bluestore_debug_random_read_err && | |
9768 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
9769 | 100.0)) == 0) { | |
224ce89b WB |
9770 | dout(0) << __func__ << ": inject random EIO" << dendl; |
9771 | r = -EIO; | |
7c673cae FG |
9772 | } |
9773 | dout(10) << __func__ << " " << cid << " " << oid | |
9774 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9775 | << " = " << r << dendl; | |
494da23a TL |
9776 | log_latency(__func__, |
9777 | l_bluestore_read_lat, | |
9778 | mono_clock::now() - start, | |
9779 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9780 | return r; |
9781 | } | |
9782 | ||
9f95a23c | 9783 | void BlueStore::_read_cache( |
7c673cae FG |
9784 | OnodeRef o, |
9785 | uint64_t offset, | |
9786 | size_t length, | |
9f95a23c TL |
9787 | int read_cache_policy, |
9788 | ready_regions_t& ready_regions, | |
9789 | blobs2read_t& blobs2read) | |
7c673cae | 9790 | { |
7c673cae | 9791 | // build blob-wise list to of stuff read (that isn't cached) |
7c673cae FG |
9792 | unsigned left = length; |
9793 | uint64_t pos = offset; | |
7c673cae FG |
9794 | auto lp = o->extent_map.seek_lextent(offset); |
9795 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
9796 | if (pos < lp->logical_offset) { | |
9797 | unsigned hole = lp->logical_offset - pos; | |
9798 | if (hole >= left) { | |
9f95a23c | 9799 | break; |
7c673cae FG |
9800 | } |
9801 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
9f95a23c | 9802 | << std::dec << dendl; |
7c673cae FG |
9803 | pos += hole; |
9804 | left -= hole; | |
9805 | } | |
94b18763 | 9806 | BlobRef& bptr = lp->blob; |
7c673cae FG |
9807 | unsigned l_off = pos - lp->logical_offset; |
9808 | unsigned b_off = l_off + lp->blob_offset; | |
9809 | unsigned b_len = std::min(left, lp->length - l_off); | |
9810 | ||
9811 | ready_regions_t cache_res; | |
9812 | interval_set<uint32_t> cache_interval; | |
9813 | bptr->shared_blob->bc.read( | |
91327a77 AA |
9814 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
9815 | read_cache_policy); | |
7c673cae | 9816 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c TL |
9817 | << " need 0x" << b_off << "~" << b_len |
9818 | << " cache has 0x" << cache_interval | |
9819 | << std::dec << dendl; | |
7c673cae FG |
9820 | |
9821 | auto pc = cache_res.begin(); | |
11fdf7f2 | 9822 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
9823 | while (b_len > 0) { |
9824 | unsigned l; | |
9825 | if (pc != cache_res.end() && | |
9f95a23c TL |
9826 | pc->first == b_off) { |
9827 | l = pc->second.length(); | |
9828 | ready_regions[pos].claim(pc->second); | |
9829 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" | |
9830 | << b_off << "~" << l << std::dec << dendl; | |
9831 | ++pc; | |
7c673cae | 9832 | } else { |
9f95a23c TL |
9833 | l = b_len; |
9834 | if (pc != cache_res.end()) { | |
9835 | ceph_assert(pc->first > b_off); | |
9836 | l = pc->first - b_off; | |
9837 | } | |
9838 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
9839 | << b_off << "~" << l << std::dec << dendl; | |
9840 | // merge regions | |
9841 | { | |
9842 | uint64_t r_off = b_off; | |
9843 | uint64_t r_len = l; | |
9844 | uint64_t front = r_off % chunk_size; | |
9845 | if (front) { | |
9846 | r_off -= front; | |
9847 | r_len += front; | |
9848 | } | |
9849 | unsigned tail = r_len % chunk_size; | |
9850 | if (tail) { | |
9851 | r_len += chunk_size - tail; | |
9852 | } | |
9853 | bool merged = false; | |
9854 | regions2read_t& r2r = blobs2read[bptr]; | |
9855 | if (r2r.size()) { | |
9856 | read_req_t& pre = r2r.back(); | |
9857 | if (r_off <= (pre.r_off + pre.r_len)) { | |
9858 | front += (r_off - pre.r_off); | |
9859 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
9860 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9861 | merged = true; | |
9862 | } | |
9863 | } | |
9864 | if (!merged) { | |
9865 | read_req_t req(r_off, r_len); | |
9866 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9867 | r2r.emplace_back(std::move(req)); | |
9868 | } | |
9869 | } | |
7c673cae FG |
9870 | } |
9871 | pos += l; | |
9872 | b_off += l; | |
9873 | left -= l; | |
9874 | b_len -= l; | |
9875 | } | |
9876 | ++lp; | |
9877 | } | |
9f95a23c | 9878 | } |
7c673cae | 9879 | |
9f95a23c TL |
9880 | int BlueStore::_prepare_read_ioc( |
9881 | blobs2read_t& blobs2read, | |
9882 | vector<bufferlist>* compressed_blob_bls, | |
9883 | IOContext* ioc) | |
9884 | { | |
7c673cae | 9885 | for (auto& p : blobs2read) { |
94b18763 | 9886 | const BlobRef& bptr = p.first; |
11fdf7f2 | 9887 | regions2read_t& r2r = p.second; |
7c673cae | 9888 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9889 | << " need " << r2r << std::dec << dendl; |
7c673cae FG |
9890 | if (bptr->get_blob().is_compressed()) { |
9891 | // read the whole thing | |
9f95a23c TL |
9892 | if (compressed_blob_bls->empty()) { |
9893 | // ensure we avoid any reallocation on subsequent blobs | |
9894 | compressed_blob_bls->reserve(blobs2read.size()); | |
9895 | } | |
9896 | compressed_blob_bls->push_back(bufferlist()); | |
9897 | bufferlist& bl = compressed_blob_bls->back(); | |
9898 | auto r = bptr->get_blob().map( | |
9899 | 0, bptr->get_blob().get_ondisk_length(), | |
9900 | [&](uint64_t offset, uint64_t length) { | |
9901 | int r = bdev->aio_read(offset, length, &bl, ioc); | |
9902 | if (r < 0) | |
7c673cae FG |
9903 | return r; |
9904 | return 0; | |
9f95a23c | 9905 | }); |
b32b8144 FG |
9906 | if (r < 0) { |
9907 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
9908 | if (r == -EIO) { | |
9909 | // propagate EIO to caller | |
9910 | return r; | |
9911 | } | |
11fdf7f2 | 9912 | ceph_assert(r == 0); |
b32b8144 | 9913 | } |
7c673cae FG |
9914 | } else { |
9915 | // read the pieces | |
11fdf7f2 | 9916 | for (auto& req : r2r) { |
9f95a23c TL |
9917 | dout(20) << __func__ << " region 0x" << std::hex |
9918 | << req.regs.front().logical_offset | |
9919 | << ": 0x" << req.regs.front().blob_xoffset | |
9920 | << " reading 0x" << req.r_off | |
9921 | << "~" << req.r_len << std::dec | |
9922 | << dendl; | |
7c673cae | 9923 | |
9f95a23c TL |
9924 | // read it |
9925 | auto r = bptr->get_blob().map( | |
9926 | req.r_off, req.r_len, | |
9927 | [&](uint64_t offset, uint64_t length) { | |
9928 | int r = bdev->aio_read(offset, length, &req.bl, ioc); | |
9929 | if (r < 0) | |
7c673cae FG |
9930 | return r; |
9931 | return 0; | |
9f95a23c | 9932 | }); |
b32b8144 FG |
9933 | if (r < 0) { |
9934 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
9935 | << dendl; | |
9936 | if (r == -EIO) { | |
9937 | // propagate EIO to caller | |
9938 | return r; | |
9939 | } | |
11fdf7f2 | 9940 | ceph_assert(r == 0); |
b32b8144 | 9941 | } |
9f95a23c | 9942 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
9943 | } |
9944 | } | |
9945 | } | |
9f95a23c TL |
9946 | return 0; |
9947 | } | |
11fdf7f2 | 9948 | |
9f95a23c TL |
9949 | int BlueStore::_generate_read_result_bl( |
9950 | OnodeRef o, | |
9951 | uint64_t offset, | |
9952 | size_t length, | |
9953 | ready_regions_t& ready_regions, | |
9954 | vector<bufferlist>& compressed_blob_bls, | |
9955 | blobs2read_t& blobs2read, | |
9956 | bool buffered, | |
9957 | bool* csum_error, | |
9958 | bufferlist& bl) | |
9959 | { | |
9960 | // enumerate and decompress desired blobs | |
7c673cae FG |
9961 | auto p = compressed_blob_bls.begin(); |
9962 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
9963 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 9964 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 9965 | regions2read_t& r2r = b2r_it->second; |
7c673cae | 9966 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9967 | << " need 0x" << r2r << std::dec << dendl; |
7c673cae | 9968 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 9969 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
9970 | bufferlist& compressed_bl = *p++; |
9971 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
9f95a23c TL |
9972 | r2r.front().regs.front().logical_offset) < 0) { |
9973 | *csum_error = true; | |
9974 | return -EIO; | |
7c673cae FG |
9975 | } |
9976 | bufferlist raw_bl; | |
9f95a23c | 9977 | auto r = _decompress(compressed_bl, &raw_bl); |
7c673cae | 9978 | if (r < 0) |
9f95a23c | 9979 | return r; |
7c673cae | 9980 | if (buffered) { |
9f95a23c TL |
9981 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, |
9982 | raw_bl); | |
7c673cae | 9983 | } |
11fdf7f2 TL |
9984 | for (auto& req : r2r) { |
9985 | for (auto& r : req.regs) { | |
9986 | ready_regions[r.logical_offset].substr_of( | |
9987 | raw_bl, r.blob_xoffset, r.length); | |
9988 | } | |
7c673cae FG |
9989 | } |
9990 | } else { | |
11fdf7f2 | 9991 | for (auto& req : r2r) { |
9f95a23c TL |
9992 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, |
9993 | req.regs.front().logical_offset) < 0) { | |
9994 | *csum_error = true; | |
9995 | return -EIO; | |
9996 | } | |
9997 | if (buffered) { | |
9998 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
9999 | req.r_off, req.bl); | |
10000 | } | |
7c673cae | 10001 | |
9f95a23c TL |
10002 | // prune and keep result |
10003 | for (const auto& r : req.regs) { | |
10004 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
11fdf7f2 | 10005 | } |
7c673cae FG |
10006 | } |
10007 | } | |
10008 | ++b2r_it; | |
10009 | } | |
10010 | ||
10011 | // generate a resulting buffer | |
10012 | auto pr = ready_regions.begin(); | |
10013 | auto pr_end = ready_regions.end(); | |
9f95a23c | 10014 | uint64_t pos = 0; |
7c673cae FG |
10015 | while (pos < length) { |
10016 | if (pr != pr_end && pr->first == pos + offset) { | |
10017 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
10018 | << ": data from 0x" << pr->first << "~" << pr->second.length() |
10019 | << std::dec << dendl; | |
7c673cae FG |
10020 | pos += pr->second.length(); |
10021 | bl.claim_append(pr->second); | |
10022 | ++pr; | |
10023 | } else { | |
10024 | uint64_t l = length - pos; | |
10025 | if (pr != pr_end) { | |
11fdf7f2 | 10026 | ceph_assert(pr->first > pos + offset); |
9f95a23c | 10027 | l = pr->first - (pos + offset); |
7c673cae FG |
10028 | } |
10029 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
10030 | << ": zeros for 0x" << (pos + offset) << "~" << l |
10031 | << std::dec << dendl; | |
7c673cae FG |
10032 | bl.append_zero(l); |
10033 | pos += l; | |
10034 | } | |
10035 | } | |
11fdf7f2 TL |
10036 | ceph_assert(bl.length() == length); |
10037 | ceph_assert(pos == length); | |
10038 | ceph_assert(pr == pr_end); | |
9f95a23c TL |
10039 | return 0; |
10040 | } | |
10041 | ||
10042 | int BlueStore::_do_read( | |
10043 | Collection *c, | |
10044 | OnodeRef o, | |
10045 | uint64_t offset, | |
10046 | size_t length, | |
10047 | bufferlist& bl, | |
10048 | uint32_t op_flags, | |
10049 | uint64_t retry_count) | |
10050 | { | |
10051 | FUNCTRACE(cct); | |
10052 | int r = 0; | |
10053 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
10054 | ||
10055 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10056 | << " size 0x" << o->onode.size << " (" << std::dec | |
10057 | << o->onode.size << ")" << dendl; | |
10058 | bl.clear(); | |
10059 | ||
10060 | if (offset >= o->onode.size) { | |
10061 | return r; | |
10062 | } | |
10063 | ||
10064 | // generally, don't buffer anything, unless the client explicitly requests | |
10065 | // it. | |
10066 | bool buffered = false; | |
10067 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
10068 | dout(20) << __func__ << " will do buffered read" << dendl; | |
10069 | buffered = true; | |
10070 | } else if (cct->_conf->bluestore_default_buffered_read && | |
10071 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
10072 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
10073 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
10074 | buffered = true; | |
10075 | } | |
10076 | ||
10077 | if (offset + length > o->onode.size) { | |
10078 | length = o->onode.size - offset; | |
10079 | } | |
10080 | ||
10081 | auto start = mono_clock::now(); | |
10082 | o->extent_map.fault_range(db, offset, length); | |
10083 | log_latency(__func__, | |
10084 | l_bluestore_read_onode_meta_lat, | |
10085 | mono_clock::now() - start, | |
10086 | cct->_conf->bluestore_log_op_age); | |
10087 | _dump_onode<30>(cct, *o); | |
10088 | ||
10089 | // for deep-scrub, we only read dirty cache and bypass clean cache in | |
10090 | // order to read underlying block device in case there are silent disk errors. | |
10091 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
10092 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
10093 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
10094 | } | |
10095 | ||
10096 | // build blob-wise list to of stuff read (that isn't cached) | |
10097 | ready_regions_t ready_regions; | |
10098 | blobs2read_t blobs2read; | |
10099 | _read_cache(o, offset, length, read_cache_policy, ready_regions, blobs2read); | |
10100 | ||
10101 | ||
10102 | // read raw blob data. | |
10103 | start = mono_clock::now(); // for the sake of simplicity | |
10104 | // measure the whole block below. | |
10105 | // The error isn't that much... | |
10106 | vector<bufferlist> compressed_blob_bls; | |
10107 | IOContext ioc(cct, NULL, true); // allow EIO | |
10108 | r = _prepare_read_ioc(blobs2read, &compressed_blob_bls, &ioc); | |
10109 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10110 | if (r < 0) | |
10111 | return r; | |
10112 | ||
10113 | int64_t num_ios = length; | |
10114 | if (ioc.has_pending_aios()) { | |
10115 | num_ios = -ioc.get_num_ios(); | |
10116 | bdev->aio_submit(&ioc); | |
10117 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10118 | ioc.aio_wait(); | |
10119 | r = ioc.get_return_value(); | |
10120 | if (r < 0) { | |
10121 | ceph_assert(r == -EIO); // no other errors allowed | |
10122 | return -EIO; | |
10123 | } | |
10124 | } | |
10125 | log_latency_fn(__func__, | |
10126 | l_bluestore_read_wait_aio_lat, | |
10127 | mono_clock::now() - start, | |
10128 | cct->_conf->bluestore_log_op_age, | |
10129 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10130 | ); | |
10131 | ||
10132 | bool csum_error = false; | |
10133 | r = _generate_read_result_bl(o, offset, length, ready_regions, | |
10134 | compressed_blob_bls, blobs2read, | |
10135 | buffered, &csum_error, bl); | |
10136 | if (csum_error) { | |
10137 | // Handles spurious read errors caused by a kernel bug. | |
10138 | // We sometimes get all-zero pages as a result of the read under | |
10139 | // high memory pressure. Retrying the failing read succeeds in most | |
10140 | // cases. | |
10141 | // See also: http://tracker.ceph.com/issues/22464 | |
10142 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10143 | return -EIO; | |
10144 | } | |
10145 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
10146 | } | |
7c673cae | 10147 | r = bl.length(); |
f64942e4 AA |
10148 | if (retry_count) { |
10149 | logger->inc(l_bluestore_reads_with_retries); | |
10150 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
10151 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
10152 | } | |
7c673cae FG |
10153 | return r; |
10154 | } | |
10155 | ||
10156 | int BlueStore::_verify_csum(OnodeRef& o, | |
10157 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
10158 | const bufferlist& bl, | |
10159 | uint64_t logical_offset) const | |
10160 | { | |
10161 | int bad; | |
10162 | uint64_t bad_csum; | |
11fdf7f2 | 10163 | auto start = mono_clock::now(); |
7c673cae | 10164 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
10165 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
10166 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
10167 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
10168 | bad = blob_xoffset; | |
10169 | r = -1; | |
10170 | bad_csum = 0xDEADBEEF; | |
10171 | } | |
7c673cae FG |
10172 | if (r < 0) { |
10173 | if (r == -1) { | |
10174 | PExtentVector pex; | |
10175 | blob->map( | |
10176 | bad, | |
10177 | blob->get_csum_chunk_size(), | |
10178 | [&](uint64_t offset, uint64_t length) { | |
10179 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
10180 | return 0; | |
10181 | }); | |
10182 | derr << __func__ << " bad " | |
10183 | << Checksummer::get_csum_type_string(blob->csum_type) | |
10184 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
10185 | << " checksum at blob offset 0x" << bad | |
10186 | << ", got 0x" << bad_csum << ", expected 0x" | |
10187 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
10188 | << ", device location " << pex | |
10189 | << ", logical extent 0x" << std::hex | |
10190 | << (logical_offset + bad - blob_xoffset) << "~" | |
10191 | << blob->get_csum_chunk_size() << std::dec | |
10192 | << ", object " << o->oid | |
10193 | << dendl; | |
10194 | } else { | |
10195 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
10196 | } | |
10197 | } | |
494da23a TL |
10198 | log_latency(__func__, |
10199 | l_bluestore_csum_lat, | |
10200 | mono_clock::now() - start, | |
10201 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 TL |
10202 | if (cct->_conf->bluestore_ignore_data_csum) { |
10203 | return 0; | |
10204 | } | |
7c673cae FG |
10205 | return r; |
10206 | } | |
10207 | ||
10208 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
10209 | { | |
10210 | int r = 0; | |
11fdf7f2 TL |
10211 | auto start = mono_clock::now(); |
10212 | auto i = source.cbegin(); | |
7c673cae | 10213 | bluestore_compression_header_t chdr; |
11fdf7f2 | 10214 | decode(chdr, i); |
7c673cae FG |
10215 | int alg = int(chdr.type); |
10216 | CompressorRef cp = compressor; | |
10217 | if (!cp || (int)cp->get_type() != alg) { | |
10218 | cp = Compressor::create(cct, alg); | |
10219 | } | |
10220 | ||
10221 | if (!cp.get()) { | |
10222 | // if compressor isn't available - error, because cannot return | |
10223 | // decompressed data? | |
11fdf7f2 TL |
10224 | |
10225 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
10226 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
10227 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
10228 | r = -EIO; |
10229 | } else { | |
10230 | r = cp->decompress(i, chdr.length, *result); | |
10231 | if (r < 0) { | |
10232 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
10233 | r = -EIO; | |
10234 | } | |
10235 | } | |
494da23a TL |
10236 | log_latency(__func__, |
10237 | l_bluestore_decompress_lat, | |
10238 | mono_clock::now() - start, | |
10239 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
10240 | return r; |
10241 | } | |
10242 | ||
10243 | // this stores fiemap into interval_set, other variations | |
10244 | // use it internally | |
10245 | int BlueStore::_fiemap( | |
10246 | CollectionHandle &c_, | |
10247 | const ghobject_t& oid, | |
10248 | uint64_t offset, | |
10249 | size_t length, | |
10250 | interval_set<uint64_t>& destset) | |
10251 | { | |
10252 | Collection *c = static_cast<Collection *>(c_.get()); | |
10253 | if (!c->exists) | |
10254 | return -ENOENT; | |
10255 | { | |
9f95a23c | 10256 | std::shared_lock l(c->lock); |
7c673cae FG |
10257 | |
10258 | OnodeRef o = c->get_onode(oid, false); | |
10259 | if (!o || !o->exists) { | |
10260 | return -ENOENT; | |
10261 | } | |
81eedcae | 10262 | _dump_onode<30>(cct, *o); |
7c673cae FG |
10263 | |
10264 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10265 | << " size 0x" << o->onode.size << std::dec << dendl; | |
10266 | ||
10267 | boost::intrusive::set<Extent>::iterator ep, eend; | |
10268 | if (offset >= o->onode.size) | |
10269 | goto out; | |
10270 | ||
10271 | if (offset + length > o->onode.size) { | |
10272 | length = o->onode.size - offset; | |
10273 | } | |
10274 | ||
10275 | o->extent_map.fault_range(db, offset, length); | |
10276 | eend = o->extent_map.extent_map.end(); | |
10277 | ep = o->extent_map.seek_lextent(offset); | |
10278 | while (length > 0) { | |
10279 | dout(20) << __func__ << " offset " << offset << dendl; | |
10280 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
10281 | ++ep; | |
10282 | continue; | |
10283 | } | |
10284 | ||
10285 | uint64_t x_len = length; | |
10286 | if (ep != eend && ep->logical_offset <= offset) { | |
10287 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 10288 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
10289 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
10290 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
10291 | destset.insert(offset, x_len); | |
10292 | length -= x_len; | |
10293 | offset += x_len; | |
10294 | if (x_off + x_len == ep->length) | |
10295 | ++ep; | |
10296 | continue; | |
10297 | } | |
10298 | if (ep != eend && | |
10299 | ep->logical_offset > offset && | |
10300 | ep->logical_offset - offset < x_len) { | |
10301 | x_len = ep->logical_offset - offset; | |
10302 | } | |
10303 | offset += x_len; | |
10304 | length -= x_len; | |
10305 | } | |
10306 | } | |
9f95a23c TL |
10307 | |
10308 | out: | |
10309 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10310 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
10311 | return 0; | |
10312 | } | |
10313 | ||
10314 | int BlueStore::fiemap( | |
10315 | CollectionHandle &c_, | |
10316 | const ghobject_t& oid, | |
10317 | uint64_t offset, | |
10318 | size_t length, | |
10319 | bufferlist& bl) | |
10320 | { | |
10321 | interval_set<uint64_t> m; | |
10322 | int r = _fiemap(c_, oid, offset, length, m); | |
10323 | if (r >= 0) { | |
10324 | encode(m, bl); | |
10325 | } | |
10326 | return r; | |
10327 | } | |
10328 | ||
10329 | int BlueStore::fiemap( | |
10330 | CollectionHandle &c_, | |
10331 | const ghobject_t& oid, | |
10332 | uint64_t offset, | |
10333 | size_t length, | |
10334 | map<uint64_t, uint64_t>& destmap) | |
10335 | { | |
10336 | interval_set<uint64_t> m; | |
10337 | int r = _fiemap(c_, oid, offset, length, m); | |
10338 | if (r >= 0) { | |
10339 | destmap = std::move(m).detach(); | |
10340 | } | |
10341 | return r; | |
10342 | } | |
10343 | ||
10344 | int BlueStore::readv( | |
10345 | CollectionHandle &c_, | |
10346 | const ghobject_t& oid, | |
10347 | interval_set<uint64_t>& m, | |
10348 | bufferlist& bl, | |
10349 | uint32_t op_flags) | |
10350 | { | |
10351 | auto start = mono_clock::now(); | |
10352 | Collection *c = static_cast<Collection *>(c_.get()); | |
10353 | const coll_t &cid = c->get_cid(); | |
10354 | dout(15) << __func__ << " " << cid << " " << oid | |
10355 | << " fiemap " << m | |
10356 | << dendl; | |
10357 | if (!c->exists) | |
10358 | return -ENOENT; | |
10359 | ||
10360 | bl.clear(); | |
10361 | int r; | |
10362 | { | |
10363 | std::shared_lock l(c->lock); | |
10364 | auto start1 = mono_clock::now(); | |
10365 | OnodeRef o = c->get_onode(oid, false); | |
10366 | log_latency("get_onode@read", | |
10367 | l_bluestore_read_onode_meta_lat, | |
10368 | mono_clock::now() - start1, | |
10369 | cct->_conf->bluestore_log_op_age); | |
10370 | if (!o || !o->exists) { | |
10371 | r = -ENOENT; | |
10372 | goto out; | |
10373 | } | |
10374 | ||
10375 | if (m.empty()) { | |
10376 | r = 0; | |
10377 | goto out; | |
10378 | } | |
10379 | ||
10380 | r = _do_readv(c, o, m, bl, op_flags); | |
10381 | if (r == -EIO) { | |
10382 | logger->inc(l_bluestore_read_eio); | |
10383 | } | |
10384 | } | |
10385 | ||
10386 | out: | |
10387 | if (r >= 0 && _debug_data_eio(oid)) { | |
10388 | r = -EIO; | |
10389 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10390 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ | |
10391 | cct->_conf->bluestore_debug_random_read_err && | |
10392 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
10393 | 100.0)) == 0) { | |
10394 | dout(0) << __func__ << ": inject random EIO" << dendl; | |
10395 | r = -EIO; | |
10396 | } | |
10397 | dout(10) << __func__ << " " << cid << " " << oid | |
10398 | << " fiemap " << m << std::dec | |
10399 | << " = " << r << dendl; | |
10400 | log_latency(__func__, | |
10401 | l_bluestore_read_lat, | |
10402 | mono_clock::now() - start, | |
10403 | cct->_conf->bluestore_log_op_age); | |
10404 | return r; | |
10405 | } | |
10406 | ||
10407 | int BlueStore::_do_readv( | |
10408 | Collection *c, | |
10409 | OnodeRef o, | |
10410 | const interval_set<uint64_t>& m, | |
10411 | bufferlist& bl, | |
10412 | uint32_t op_flags, | |
10413 | uint64_t retry_count) | |
10414 | { | |
10415 | FUNCTRACE(cct); | |
10416 | int r = 0; | |
10417 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
10418 | ||
10419 | dout(20) << __func__ << " fiemap " << m << std::hex | |
10420 | << " size 0x" << o->onode.size << " (" << std::dec | |
10421 | << o->onode.size << ")" << dendl; | |
10422 | ||
10423 | // generally, don't buffer anything, unless the client explicitly requests | |
10424 | // it. | |
10425 | bool buffered = false; | |
10426 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
10427 | dout(20) << __func__ << " will do buffered read" << dendl; | |
10428 | buffered = true; | |
10429 | } else if (cct->_conf->bluestore_default_buffered_read && | |
10430 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
10431 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
10432 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
10433 | buffered = true; | |
10434 | } | |
10435 | // this method must be idempotent since we may call it several times | |
10436 | // before we finally read the expected result. | |
10437 | bl.clear(); | |
10438 | ||
10439 | // call fiemap first! | |
10440 | ceph_assert(m.range_start() <= o->onode.size); | |
10441 | ceph_assert(m.range_end() <= o->onode.size); | |
10442 | auto start = mono_clock::now(); | |
10443 | o->extent_map.fault_range(db, m.range_start(), m.range_end() - m.range_start()); | |
10444 | log_latency(__func__, | |
10445 | l_bluestore_read_onode_meta_lat, | |
10446 | mono_clock::now() - start, | |
10447 | cct->_conf->bluestore_log_op_age); | |
10448 | _dump_onode<30>(cct, *o); | |
10449 | ||
10450 | IOContext ioc(cct, NULL, true); // allow EIO | |
10451 | vector<std::tuple<ready_regions_t, vector<bufferlist>, blobs2read_t>> raw_results; | |
10452 | raw_results.reserve(m.num_intervals()); | |
10453 | int i = 0; | |
10454 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10455 | raw_results.push_back({}); | |
10456 | _read_cache(o, p.get_start(), p.get_len(), read_cache_policy, | |
10457 | std::get<0>(raw_results[i]), std::get<2>(raw_results[i])); | |
10458 | r = _prepare_read_ioc(std::get<2>(raw_results[i]), &std::get<1>(raw_results[i]), &ioc); | |
10459 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10460 | if (r < 0) | |
10461 | return r; | |
10462 | } | |
10463 | ||
10464 | auto num_ios = m.size(); | |
10465 | if (ioc.has_pending_aios()) { | |
10466 | num_ios = ioc.get_num_ios(); | |
10467 | bdev->aio_submit(&ioc); | |
10468 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10469 | ioc.aio_wait(); | |
10470 | r = ioc.get_return_value(); | |
10471 | if (r < 0) { | |
10472 | ceph_assert(r == -EIO); // no other errors allowed | |
10473 | return -EIO; | |
10474 | } | |
10475 | } | |
10476 | log_latency_fn(__func__, | |
10477 | l_bluestore_read_wait_aio_lat, | |
10478 | mono_clock::now() - start, | |
10479 | cct->_conf->bluestore_log_op_age, | |
10480 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10481 | ); | |
10482 | ||
10483 | ceph_assert(raw_results.size() == (size_t)m.num_intervals()); | |
10484 | i = 0; | |
10485 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10486 | bool csum_error = false; | |
10487 | bufferlist t; | |
10488 | r = _generate_read_result_bl(o, p.get_start(), p.get_len(), | |
10489 | std::get<0>(raw_results[i]), | |
10490 | std::get<1>(raw_results[i]), | |
10491 | std::get<2>(raw_results[i]), | |
10492 | buffered, &csum_error, t); | |
10493 | if (csum_error) { | |
10494 | // Handles spurious read errors caused by a kernel bug. | |
10495 | // We sometimes get all-zero pages as a result of the read under | |
10496 | // high memory pressure. Retrying the failing read succeeds in most | |
10497 | // cases. | |
10498 | // See also: http://tracker.ceph.com/issues/22464 | |
10499 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10500 | return -EIO; | |
10501 | } | |
10502 | return _do_readv(c, o, m, bl, op_flags, retry_count + 1); | |
10503 | } | |
10504 | bl.claim_append(t); | |
10505 | } | |
10506 | if (retry_count) { | |
10507 | logger->inc(l_bluestore_reads_with_retries); | |
10508 | dout(5) << __func__ << " read fiemap " << m | |
10509 | << " failed " << retry_count << " times before succeeding" | |
10510 | << dendl; | |
10511 | } | |
10512 | return bl.length(); | |
7c673cae FG |
10513 | } |
10514 | ||
9f95a23c | 10515 | int BlueStore::dump_onode(CollectionHandle &c_, |
7c673cae | 10516 | const ghobject_t& oid, |
9f95a23c TL |
10517 | const string& section_name, |
10518 | Formatter *f) | |
7c673cae | 10519 | { |
9f95a23c TL |
10520 | Collection *c = static_cast<Collection *>(c_.get()); |
10521 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10522 | if (!c->exists) | |
10523 | return -ENOENT; | |
7c673cae | 10524 | |
9f95a23c TL |
10525 | int r; |
10526 | { | |
10527 | std::shared_lock l(c->lock); | |
10528 | ||
10529 | OnodeRef o = c->get_onode(oid, false); | |
10530 | if (!o || !o->exists) { | |
10531 | r = -ENOENT; | |
10532 | goto out; | |
10533 | } | |
10534 | // FIXME minor: actually the next line isn't enough to | |
10535 | // load shared blobs. Leaving as is for now.. | |
10536 | // | |
10537 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10538 | ||
10539 | _dump_onode<0>(cct, *o); | |
10540 | f->open_object_section(section_name.c_str()); | |
10541 | o->dump(f); | |
10542 | f->close_section(); | |
10543 | r = 0; | |
7c673cae | 10544 | } |
9f95a23c TL |
10545 | out: |
10546 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10547 | << " = " << r << dendl; | |
7c673cae FG |
10548 | return r; |
10549 | } | |
10550 | ||
7c673cae FG |
10551 | int BlueStore::getattr( |
10552 | CollectionHandle &c_, | |
10553 | const ghobject_t& oid, | |
10554 | const char *name, | |
10555 | bufferptr& value) | |
10556 | { | |
10557 | Collection *c = static_cast<Collection *>(c_.get()); | |
10558 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
10559 | if (!c->exists) | |
10560 | return -ENOENT; | |
10561 | ||
10562 | int r; | |
10563 | { | |
9f95a23c | 10564 | std::shared_lock l(c->lock); |
f91f0fd5 | 10565 | mempool::bluestore_cache_meta::string k(name); |
7c673cae FG |
10566 | |
10567 | OnodeRef o = c->get_onode(oid, false); | |
10568 | if (!o || !o->exists) { | |
10569 | r = -ENOENT; | |
10570 | goto out; | |
10571 | } | |
10572 | ||
10573 | if (!o->onode.attrs.count(k)) { | |
10574 | r = -ENODATA; | |
10575 | goto out; | |
10576 | } | |
10577 | value = o->onode.attrs[k]; | |
10578 | r = 0; | |
10579 | } | |
10580 | out: | |
7c673cae FG |
10581 | if (r == 0 && _debug_mdata_eio(oid)) { |
10582 | r = -EIO; | |
10583 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10584 | } | |
10585 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
10586 | << " = " << r << dendl; | |
10587 | return r; | |
10588 | } | |
10589 | ||
7c673cae FG |
10590 | int BlueStore::getattrs( |
10591 | CollectionHandle &c_, | |
10592 | const ghobject_t& oid, | |
10593 | map<string,bufferptr>& aset) | |
10594 | { | |
10595 | Collection *c = static_cast<Collection *>(c_.get()); | |
10596 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10597 | if (!c->exists) | |
10598 | return -ENOENT; | |
10599 | ||
10600 | int r; | |
10601 | { | |
9f95a23c | 10602 | std::shared_lock l(c->lock); |
7c673cae FG |
10603 | |
10604 | OnodeRef o = c->get_onode(oid, false); | |
10605 | if (!o || !o->exists) { | |
10606 | r = -ENOENT; | |
10607 | goto out; | |
10608 | } | |
10609 | for (auto& i : o->onode.attrs) { | |
10610 | aset.emplace(i.first.c_str(), i.second); | |
10611 | } | |
10612 | r = 0; | |
10613 | } | |
10614 | ||
10615 | out: | |
7c673cae FG |
10616 | if (r == 0 && _debug_mdata_eio(oid)) { |
10617 | r = -EIO; | |
10618 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10619 | } | |
10620 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10621 | << " = " << r << dendl; | |
10622 | return r; | |
10623 | } | |
10624 | ||
10625 | int BlueStore::list_collections(vector<coll_t>& ls) | |
10626 | { | |
9f95a23c | 10627 | std::shared_lock l(coll_lock); |
11fdf7f2 | 10628 | ls.reserve(coll_map.size()); |
7c673cae FG |
10629 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
10630 | p != coll_map.end(); | |
10631 | ++p) | |
10632 | ls.push_back(p->first); | |
10633 | return 0; | |
10634 | } | |
10635 | ||
10636 | bool BlueStore::collection_exists(const coll_t& c) | |
10637 | { | |
9f95a23c | 10638 | std::shared_lock l(coll_lock); |
7c673cae FG |
10639 | return coll_map.count(c); |
10640 | } | |
10641 | ||
11fdf7f2 | 10642 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 10643 | { |
11fdf7f2 | 10644 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
10645 | vector<ghobject_t> ls; |
10646 | ghobject_t next; | |
11fdf7f2 | 10647 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
10648 | &ls, &next); |
10649 | if (r < 0) { | |
10650 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
10651 | << dendl; | |
10652 | return r; | |
10653 | } | |
10654 | *empty = ls.empty(); | |
11fdf7f2 | 10655 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
10656 | return 0; |
10657 | } | |
10658 | ||
11fdf7f2 | 10659 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 10660 | { |
11fdf7f2 TL |
10661 | dout(15) << __func__ << " " << ch->cid << dendl; |
10662 | Collection *c = static_cast<Collection*>(ch.get()); | |
9f95a23c | 10663 | std::shared_lock l(c->lock); |
11fdf7f2 | 10664 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
10665 | return c->cnode.bits; |
10666 | } | |
10667 | ||
7c673cae FG |
10668 | int BlueStore::collection_list( |
10669 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10670 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10671 | { | |
10672 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 10673 | c->flush(); |
7c673cae FG |
10674 | dout(15) << __func__ << " " << c->cid |
10675 | << " start " << start << " end " << end << " max " << max << dendl; | |
10676 | int r; | |
10677 | { | |
9f95a23c | 10678 | std::shared_lock l(c->lock); |
f91f0fd5 TL |
10679 | r = _collection_list(c, start, end, max, false, ls, pnext); |
10680 | } | |
10681 | ||
10682 | dout(10) << __func__ << " " << c->cid | |
10683 | << " start " << start << " end " << end << " max " << max | |
10684 | << " = " << r << ", ls.size() = " << ls->size() | |
10685 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10686 | return r; | |
10687 | } | |
10688 | ||
10689 | int BlueStore::collection_list_legacy( | |
10690 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10691 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10692 | { | |
10693 | Collection *c = static_cast<Collection *>(c_.get()); | |
10694 | c->flush(); | |
10695 | dout(15) << __func__ << " " << c->cid | |
10696 | << " start " << start << " end " << end << " max " << max << dendl; | |
10697 | int r; | |
10698 | { | |
10699 | std::shared_lock l(c->lock); | |
10700 | r = _collection_list(c, start, end, max, true, ls, pnext); | |
7c673cae FG |
10701 | } |
10702 | ||
7c673cae FG |
10703 | dout(10) << __func__ << " " << c->cid |
10704 | << " start " << start << " end " << end << " max " << max | |
10705 | << " = " << r << ", ls.size() = " << ls->size() | |
10706 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10707 | return r; | |
10708 | } | |
10709 | ||
10710 | int BlueStore::_collection_list( | |
10711 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
f91f0fd5 | 10712 | bool legacy, vector<ghobject_t> *ls, ghobject_t *pnext) |
7c673cae FG |
10713 | { |
10714 | ||
10715 | if (!c->exists) | |
10716 | return -ENOENT; | |
10717 | ||
494da23a | 10718 | auto start_time = mono_clock::now(); |
7c673cae FG |
10719 | int r = 0; |
10720 | ghobject_t static_next; | |
f91f0fd5 TL |
10721 | std::unique_ptr<CollectionListIterator> it; |
10722 | ghobject_t coll_range_temp_start, coll_range_temp_end; | |
10723 | ghobject_t coll_range_start, coll_range_end; | |
7c673cae | 10724 | bool set_next = false; |
f91f0fd5 | 10725 | ghobject_t pend; |
7c673cae FG |
10726 | bool temp; |
10727 | ||
10728 | if (!pnext) | |
10729 | pnext = &static_next; | |
10730 | ||
11fdf7f2 | 10731 | if (start.is_max() || start.hobj.is_max()) { |
7c673cae FG |
10732 | goto out; |
10733 | } | |
f91f0fd5 TL |
10734 | get_coll_range(c->cid, c->cnode.bits, &coll_range_temp_start, |
10735 | &coll_range_temp_end, &coll_range_start, &coll_range_end); | |
7c673cae | 10736 | dout(20) << __func__ |
f91f0fd5 TL |
10737 | << " range " << coll_range_temp_start |
10738 | << " to " << coll_range_temp_end | |
10739 | << " and " << coll_range_start | |
10740 | << " to " << coll_range_end | |
7c673cae | 10741 | << " start " << start << dendl; |
f91f0fd5 TL |
10742 | if (legacy) { |
10743 | it = std::make_unique<SimpleCollectionListIterator>( | |
10744 | cct, db->get_iterator(PREFIX_OBJ)); | |
10745 | } else { | |
10746 | it = std::make_unique<SortedCollectionListIterator>( | |
10747 | db->get_iterator(PREFIX_OBJ)); | |
10748 | } | |
7c673cae FG |
10749 | if (start == ghobject_t() || |
10750 | start.hobj == hobject_t() || | |
10751 | start == c->cid.get_min_hobj()) { | |
f91f0fd5 | 10752 | it->upper_bound(coll_range_temp_start); |
7c673cae FG |
10753 | temp = true; |
10754 | } else { | |
7c673cae FG |
10755 | if (start.hobj.is_temp()) { |
10756 | temp = true; | |
f91f0fd5 | 10757 | ceph_assert(start >= coll_range_temp_start && start < coll_range_temp_end); |
7c673cae FG |
10758 | } else { |
10759 | temp = false; | |
f91f0fd5 | 10760 | ceph_assert(start >= coll_range_start && start < coll_range_end); |
7c673cae | 10761 | } |
f91f0fd5 TL |
10762 | dout(20) << __func__ << " temp=" << (int)temp << dendl; |
10763 | it->lower_bound(start); | |
7c673cae FG |
10764 | } |
10765 | if (end.hobj.is_max()) { | |
f91f0fd5 | 10766 | pend = temp ? coll_range_temp_end : coll_range_end; |
7c673cae | 10767 | } else { |
7c673cae FG |
10768 | if (end.hobj.is_temp()) { |
10769 | if (temp) | |
f91f0fd5 | 10770 | pend = end; |
7c673cae | 10771 | else |
f91f0fd5 | 10772 | goto out; |
7c673cae | 10773 | } else { |
f91f0fd5 | 10774 | pend = temp ? coll_range_temp_end : end; |
7c673cae FG |
10775 | } |
10776 | } | |
f91f0fd5 | 10777 | dout(20) << __func__ << " pend " << pend << dendl; |
7c673cae | 10778 | while (true) { |
f91f0fd5 | 10779 | if (!it->valid() || it->oid() >= pend) { |
7c673cae FG |
10780 | if (!it->valid()) |
10781 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
10782 | else | |
f91f0fd5 | 10783 | dout(20) << __func__ << " oid " << it->oid() << " >= " << pend << dendl; |
7c673cae FG |
10784 | if (temp) { |
10785 | if (end.hobj.is_temp()) { | |
f91f0fd5 TL |
10786 | if (it->valid() && it->oid() < coll_range_temp_end) { |
10787 | *pnext = it->oid(); | |
10788 | set_next = true; | |
10789 | } | |
7c673cae FG |
10790 | break; |
10791 | } | |
10792 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
10793 | temp = false; | |
f91f0fd5 TL |
10794 | it->upper_bound(coll_range_start); |
10795 | if (end.hobj.is_max()) | |
10796 | pend = coll_range_end; | |
10797 | else | |
10798 | pend = end; | |
10799 | dout(30) << __func__ << " pend " << pend << dendl; | |
7c673cae FG |
10800 | continue; |
10801 | } | |
f91f0fd5 TL |
10802 | if (it->valid() && it->oid() < coll_range_end) { |
10803 | *pnext = it->oid(); | |
10804 | set_next = true; | |
10805 | } | |
7c673cae FG |
10806 | break; |
10807 | } | |
f91f0fd5 | 10808 | dout(20) << __func__ << " oid " << it->oid() << " end " << end << dendl; |
7c673cae FG |
10809 | if (ls->size() >= (unsigned)max) { |
10810 | dout(20) << __func__ << " reached max " << max << dendl; | |
f91f0fd5 | 10811 | *pnext = it->oid(); |
7c673cae FG |
10812 | set_next = true; |
10813 | break; | |
10814 | } | |
f91f0fd5 | 10815 | ls->push_back(it->oid()); |
7c673cae FG |
10816 | it->next(); |
10817 | } | |
10818 | out: | |
10819 | if (!set_next) { | |
10820 | *pnext = ghobject_t::get_max(); | |
10821 | } | |
494da23a TL |
10822 | log_latency_fn( |
10823 | __func__, | |
10824 | l_bluestore_clist_lat, | |
10825 | mono_clock::now() - start_time, | |
10826 | cct->_conf->bluestore_log_collection_list_age, | |
10827 | [&] (const ceph::timespan& lat) { | |
10828 | ostringstream ostr; | |
10829 | ostr << ", lat = " << timespan_str(lat) | |
10830 | << " cid =" << c->cid | |
10831 | << " start " << start << " end " << end | |
10832 | << " max " << max; | |
10833 | return ostr.str(); | |
10834 | } | |
10835 | ); | |
7c673cae FG |
10836 | return r; |
10837 | } | |
10838 | ||
7c673cae FG |
10839 | int BlueStore::omap_get( |
10840 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10841 | const ghobject_t &oid, ///< [in] Object containing omap | |
10842 | bufferlist *header, ///< [out] omap header | |
10843 | map<string, bufferlist> *out /// < [out] Key to value map | |
10844 | ) | |
10845 | { | |
10846 | Collection *c = static_cast<Collection *>(c_.get()); | |
9f95a23c TL |
10847 | return _omap_get(c, oid, header, out); |
10848 | } | |
10849 | ||
10850 | int BlueStore::_omap_get( | |
10851 | Collection *c, ///< [in] Collection containing oid | |
10852 | const ghobject_t &oid, ///< [in] Object containing omap | |
10853 | bufferlist *header, ///< [out] omap header | |
10854 | map<string, bufferlist> *out /// < [out] Key to value map | |
10855 | ) | |
10856 | { | |
7c673cae FG |
10857 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; |
10858 | if (!c->exists) | |
10859 | return -ENOENT; | |
9f95a23c | 10860 | std::shared_lock l(c->lock); |
7c673cae FG |
10861 | int r = 0; |
10862 | OnodeRef o = c->get_onode(oid, false); | |
10863 | if (!o || !o->exists) { | |
10864 | r = -ENOENT; | |
10865 | goto out; | |
10866 | } | |
9f95a23c TL |
10867 | r = _onode_omap_get(o, header, out); |
10868 | out: | |
10869 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10870 | << dendl; | |
10871 | return r; | |
10872 | } | |
10873 | ||
10874 | int BlueStore::_onode_omap_get( | |
10875 | const OnodeRef &o, ///< [in] Object containing omap | |
10876 | bufferlist *header, ///< [out] omap header | |
10877 | map<string, bufferlist> *out /// < [out] Key to value map | |
10878 | ) | |
10879 | { | |
10880 | int r = 0; | |
10881 | if (!o || !o->exists) { | |
10882 | r = -ENOENT; | |
10883 | goto out; | |
10884 | } | |
7c673cae FG |
10885 | if (!o->onode.has_omap()) |
10886 | goto out; | |
10887 | o->flush(); | |
10888 | { | |
9f95a23c | 10889 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10890 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10891 | string head, tail; |
9f95a23c TL |
10892 | o->get_omap_header(&head); |
10893 | o->get_omap_tail(&tail); | |
7c673cae FG |
10894 | it->lower_bound(head); |
10895 | while (it->valid()) { | |
10896 | if (it->key() == head) { | |
9f95a23c TL |
10897 | dout(30) << __func__ << " got header" << dendl; |
10898 | *header = it->value(); | |
7c673cae | 10899 | } else if (it->key() >= tail) { |
9f95a23c TL |
10900 | dout(30) << __func__ << " reached tail" << dendl; |
10901 | break; | |
7c673cae | 10902 | } else { |
9f95a23c TL |
10903 | string user_key; |
10904 | o->decode_omap_key(it->key(), &user_key); | |
10905 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
10906 | << " -> " << user_key << dendl; | |
10907 | (*out)[user_key] = it->value(); | |
7c673cae FG |
10908 | } |
10909 | it->next(); | |
10910 | } | |
10911 | } | |
9f95a23c | 10912 | out: |
7c673cae FG |
10913 | return r; |
10914 | } | |
10915 | ||
7c673cae FG |
10916 | int BlueStore::omap_get_header( |
10917 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10918 | const ghobject_t &oid, ///< [in] Object containing omap | |
10919 | bufferlist *header, ///< [out] omap header | |
10920 | bool allow_eio ///< [in] don't assert on eio | |
10921 | ) | |
10922 | { | |
10923 | Collection *c = static_cast<Collection *>(c_.get()); | |
10924 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10925 | if (!c->exists) | |
10926 | return -ENOENT; | |
9f95a23c | 10927 | std::shared_lock l(c->lock); |
7c673cae FG |
10928 | int r = 0; |
10929 | OnodeRef o = c->get_onode(oid, false); | |
10930 | if (!o || !o->exists) { | |
10931 | r = -ENOENT; | |
10932 | goto out; | |
10933 | } | |
10934 | if (!o->onode.has_omap()) | |
10935 | goto out; | |
10936 | o->flush(); | |
10937 | { | |
10938 | string head; | |
9f95a23c TL |
10939 | o->get_omap_header(&head); |
10940 | if (db->get(o->get_omap_prefix(), head, header) >= 0) { | |
7c673cae FG |
10941 | dout(30) << __func__ << " got header" << dendl; |
10942 | } else { | |
10943 | dout(30) << __func__ << " no header" << dendl; | |
10944 | } | |
10945 | } | |
10946 | out: | |
10947 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10948 | << dendl; | |
10949 | return r; | |
10950 | } | |
10951 | ||
7c673cae FG |
10952 | int BlueStore::omap_get_keys( |
10953 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10954 | const ghobject_t &oid, ///< [in] Object containing omap | |
10955 | set<string> *keys ///< [out] Keys defined on oid | |
10956 | ) | |
10957 | { | |
10958 | Collection *c = static_cast<Collection *>(c_.get()); | |
10959 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10960 | if (!c->exists) | |
10961 | return -ENOENT; | |
9f95a23c | 10962 | std::shared_lock l(c->lock); |
7c673cae FG |
10963 | int r = 0; |
10964 | OnodeRef o = c->get_onode(oid, false); | |
10965 | if (!o || !o->exists) { | |
10966 | r = -ENOENT; | |
10967 | goto out; | |
10968 | } | |
10969 | if (!o->onode.has_omap()) | |
10970 | goto out; | |
10971 | o->flush(); | |
10972 | { | |
9f95a23c | 10973 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10974 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10975 | string head, tail; |
9f95a23c TL |
10976 | o->get_omap_key(string(), &head); |
10977 | o->get_omap_tail(&tail); | |
7c673cae FG |
10978 | it->lower_bound(head); |
10979 | while (it->valid()) { | |
10980 | if (it->key() >= tail) { | |
10981 | dout(30) << __func__ << " reached tail" << dendl; | |
10982 | break; | |
10983 | } | |
10984 | string user_key; | |
9f95a23c | 10985 | o->decode_omap_key(it->key(), &user_key); |
11fdf7f2 | 10986 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
10987 | << " -> " << user_key << dendl; |
10988 | keys->insert(user_key); | |
10989 | it->next(); | |
11fdf7f2 TL |
10990 | } |
10991 | } | |
10992 | out: | |
10993 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10994 | << dendl; | |
10995 | return r; | |
7c673cae FG |
10996 | } |
10997 | ||
10998 | int BlueStore::omap_get_values( | |
10999 | CollectionHandle &c_, ///< [in] Collection containing oid | |
11000 | const ghobject_t &oid, ///< [in] Object containing omap | |
11001 | const set<string> &keys, ///< [in] Keys to get | |
11002 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
11003 | ) | |
11004 | { | |
11005 | Collection *c = static_cast<Collection *>(c_.get()); | |
11006 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
11007 | if (!c->exists) | |
11008 | return -ENOENT; | |
9f95a23c | 11009 | std::shared_lock l(c->lock); |
7c673cae FG |
11010 | int r = 0; |
11011 | string final_key; | |
11012 | OnodeRef o = c->get_onode(oid, false); | |
11013 | if (!o || !o->exists) { | |
11014 | r = -ENOENT; | |
11015 | goto out; | |
11016 | } | |
9f95a23c | 11017 | if (!o->onode.has_omap()) { |
7c673cae | 11018 | goto out; |
9f95a23c TL |
11019 | } |
11020 | o->flush(); | |
11fdf7f2 | 11021 | { |
9f95a23c TL |
11022 | const string& prefix = o->get_omap_prefix(); |
11023 | o->get_omap_key(string(), &final_key); | |
11024 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 11025 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 11026 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
11027 | final_key += *p; |
11028 | bufferlist val; | |
11029 | if (db->get(prefix, final_key, &val) >= 0) { | |
11030 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
11031 | << " -> " << *p << dendl; | |
11032 | out->insert(make_pair(*p, val)); | |
11033 | } | |
7c673cae FG |
11034 | } |
11035 | } | |
11036 | out: | |
11037 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
11038 | << dendl; | |
11039 | return r; | |
11040 | } | |
11041 | ||
9f95a23c TL |
11042 | #ifdef WITH_SEASTAR |
11043 | int BlueStore::omap_get_values( | |
11044 | CollectionHandle &c_, ///< [in] Collection containing oid | |
11045 | const ghobject_t &oid, ///< [in] Object containing omap | |
11046 | const std::optional<string> &start_after, ///< [in] Keys to get | |
11047 | map<string, bufferlist> *output ///< [out] Returned keys and values | |
11048 | ) | |
11049 | { | |
11050 | Collection *c = static_cast<Collection *>(c_.get()); | |
11051 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
11052 | if (!c->exists) | |
11053 | return -ENOENT; | |
11054 | std::shared_lock l(c->lock); | |
11055 | int r = 0; | |
11056 | OnodeRef o = c->get_onode(oid, false); | |
11057 | if (!o || !o->exists) { | |
11058 | r = -ENOENT; | |
11059 | goto out; | |
11060 | } | |
11061 | if (!o->onode.has_omap()) { | |
11062 | goto out; | |
11063 | } | |
11064 | o->flush(); | |
11065 | { | |
11066 | ObjectMap::ObjectMapIterator iter = get_omap_iterator(c_, oid); | |
11067 | if (!iter) { | |
11068 | r = -ENOENT; | |
11069 | goto out; | |
11070 | } | |
11071 | iter->upper_bound(*start_after); | |
11072 | for (; iter->valid(); iter->next()) { | |
11073 | output->insert(make_pair(iter->key(), iter->value())); | |
11074 | } | |
11075 | } | |
11076 | ||
11077 | out: | |
11078 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
11079 | << dendl; | |
11080 | return r; | |
11081 | } | |
11082 | #endif | |
11083 | ||
7c673cae FG |
11084 | int BlueStore::omap_check_keys( |
11085 | CollectionHandle &c_, ///< [in] Collection containing oid | |
11086 | const ghobject_t &oid, ///< [in] Object containing omap | |
11087 | const set<string> &keys, ///< [in] Keys to check | |
11088 | set<string> *out ///< [out] Subset of keys defined on oid | |
11089 | ) | |
11090 | { | |
11091 | Collection *c = static_cast<Collection *>(c_.get()); | |
11092 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
11093 | if (!c->exists) | |
11094 | return -ENOENT; | |
9f95a23c | 11095 | std::shared_lock l(c->lock); |
7c673cae FG |
11096 | int r = 0; |
11097 | string final_key; | |
11098 | OnodeRef o = c->get_onode(oid, false); | |
11099 | if (!o || !o->exists) { | |
11100 | r = -ENOENT; | |
11101 | goto out; | |
11102 | } | |
9f95a23c | 11103 | if (!o->onode.has_omap()) { |
7c673cae | 11104 | goto out; |
9f95a23c TL |
11105 | } |
11106 | o->flush(); | |
11fdf7f2 | 11107 | { |
9f95a23c TL |
11108 | const string& prefix = o->get_omap_prefix(); |
11109 | o->get_omap_key(string(), &final_key); | |
11110 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 11111 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 11112 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
11113 | final_key += *p; |
11114 | bufferlist val; | |
11115 | if (db->get(prefix, final_key, &val) >= 0) { | |
11116 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
11117 | << " -> " << *p << dendl; | |
11118 | out->insert(*p); | |
11119 | } else { | |
11120 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
11121 | << " -> " << *p << dendl; | |
11122 | } | |
7c673cae FG |
11123 | } |
11124 | } | |
11125 | out: | |
11126 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
11127 | << dendl; | |
11128 | return r; | |
11129 | } | |
11130 | ||
7c673cae FG |
11131 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
11132 | CollectionHandle &c_, ///< [in] collection | |
11133 | const ghobject_t &oid ///< [in] object | |
11134 | ) | |
11135 | { | |
11136 | Collection *c = static_cast<Collection *>(c_.get()); | |
11137 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
11138 | if (!c->exists) { | |
11139 | return ObjectMap::ObjectMapIterator(); | |
11140 | } | |
9f95a23c | 11141 | std::shared_lock l(c->lock); |
7c673cae FG |
11142 | OnodeRef o = c->get_onode(oid, false); |
11143 | if (!o || !o->exists) { | |
11144 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
11145 | return ObjectMap::ObjectMapIterator(); | |
11146 | } | |
11147 | o->flush(); | |
11148 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
9f95a23c | 11149 | KeyValueDB::Iterator it = db->get_iterator(o->get_omap_prefix()); |
7c673cae FG |
11150 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(c, o, it)); |
11151 | } | |
11152 | ||
11153 | // ----------------- | |
11154 | // write helpers | |
11155 | ||
11fdf7f2 TL |
11156 | uint64_t BlueStore::_get_ondisk_reserved() const { |
11157 | return round_up_to( | |
11158 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
11159 | } | |
11160 | ||
7c673cae FG |
11161 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
11162 | { | |
11163 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
11164 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
11165 | << dendl; | |
11fdf7f2 | 11166 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
11167 | { |
11168 | bufferlist bl; | |
11fdf7f2 | 11169 | encode(ondisk_format, bl); |
7c673cae FG |
11170 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
11171 | } | |
11172 | { | |
11173 | bufferlist bl; | |
11fdf7f2 | 11174 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
11175 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
11176 | } | |
11177 | } | |
11178 | ||
11179 | int BlueStore::_open_super_meta() | |
11180 | { | |
11181 | // nid | |
11182 | { | |
11183 | nid_max = 0; | |
11184 | bufferlist bl; | |
11185 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 11186 | auto p = bl.cbegin(); |
7c673cae FG |
11187 | try { |
11188 | uint64_t v; | |
11fdf7f2 | 11189 | decode(v, p); |
7c673cae FG |
11190 | nid_max = v; |
11191 | } catch (buffer::error& e) { | |
11192 | derr << __func__ << " unable to read nid_max" << dendl; | |
11193 | return -EIO; | |
11194 | } | |
11195 | dout(10) << __func__ << " old nid_max " << nid_max << dendl; | |
11196 | nid_last = nid_max.load(); | |
11197 | } | |
11198 | ||
11199 | // blobid | |
11200 | { | |
11201 | blobid_max = 0; | |
11202 | bufferlist bl; | |
11203 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 11204 | auto p = bl.cbegin(); |
7c673cae FG |
11205 | try { |
11206 | uint64_t v; | |
11fdf7f2 | 11207 | decode(v, p); |
7c673cae FG |
11208 | blobid_max = v; |
11209 | } catch (buffer::error& e) { | |
11210 | derr << __func__ << " unable to read blobid_max" << dendl; | |
11211 | return -EIO; | |
11212 | } | |
11213 | dout(10) << __func__ << " old blobid_max " << blobid_max << dendl; | |
11214 | blobid_last = blobid_max.load(); | |
11215 | } | |
11216 | ||
11217 | // freelist | |
11218 | { | |
11219 | bufferlist bl; | |
11220 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
11221 | if (bl.length()) { | |
11222 | freelist_type = std::string(bl.c_str(), bl.length()); | |
11223 | dout(10) << __func__ << " freelist_type " << freelist_type << dendl; | |
11224 | } else { | |
11fdf7f2 | 11225 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 11226 | } |
7c673cae FG |
11227 | } |
11228 | ||
11229 | // ondisk format | |
11230 | int32_t compat_ondisk_format = 0; | |
11231 | { | |
11232 | bufferlist bl; | |
11233 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
11234 | if (r < 0) { | |
11235 | // base case: kraken bluestore is v1 and readable by v1 | |
11236 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
11237 | << dendl; | |
11238 | ondisk_format = 1; | |
11239 | compat_ondisk_format = 1; | |
11240 | } else { | |
11fdf7f2 | 11241 | auto p = bl.cbegin(); |
7c673cae | 11242 | try { |
11fdf7f2 | 11243 | decode(ondisk_format, p); |
7c673cae FG |
11244 | } catch (buffer::error& e) { |
11245 | derr << __func__ << " unable to read ondisk_format" << dendl; | |
11246 | return -EIO; | |
11247 | } | |
11248 | bl.clear(); | |
11249 | { | |
11250 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
11251 | ceph_assert(!r); |
11252 | auto p = bl.cbegin(); | |
7c673cae | 11253 | try { |
11fdf7f2 | 11254 | decode(compat_ondisk_format, p); |
7c673cae FG |
11255 | } catch (buffer::error& e) { |
11256 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; | |
11257 | return -EIO; | |
11258 | } | |
11259 | } | |
11260 | } | |
11261 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
11262 | << " compat_ondisk_format " << compat_ondisk_format | |
11263 | << dendl; | |
11264 | } | |
11265 | ||
11266 | if (latest_ondisk_format < compat_ondisk_format) { | |
11267 | derr << __func__ << " compat_ondisk_format is " | |
11268 | << compat_ondisk_format << " but we only understand version " | |
11269 | << latest_ondisk_format << dendl; | |
11270 | return -EPERM; | |
11271 | } | |
7c673cae FG |
11272 | |
11273 | { | |
11274 | bufferlist bl; | |
11275 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 11276 | auto p = bl.cbegin(); |
7c673cae FG |
11277 | try { |
11278 | uint64_t val; | |
11fdf7f2 | 11279 | decode(val, p); |
7c673cae | 11280 | min_alloc_size = val; |
224ce89b | 11281 | min_alloc_size_order = ctz(val); |
11fdf7f2 | 11282 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
7c673cae FG |
11283 | } catch (buffer::error& e) { |
11284 | derr << __func__ << " unable to read min_alloc_size" << dendl; | |
11285 | return -EIO; | |
11286 | } | |
11287 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11288 | << std::dec << dendl; | |
11289 | } | |
9f95a23c TL |
11290 | |
11291 | _set_per_pool_omap(); | |
11292 | ||
224ce89b | 11293 | _open_statfs(); |
7c673cae FG |
11294 | _set_alloc_sizes(); |
11295 | _set_throttle_params(); | |
11296 | ||
11297 | _set_csum(); | |
11298 | _set_compression(); | |
11299 | _set_blob_size(); | |
11300 | ||
11fdf7f2 | 11301 | _validate_bdev(); |
7c673cae FG |
11302 | return 0; |
11303 | } | |
11304 | ||
11305 | int BlueStore::_upgrade_super() | |
11306 | { | |
11307 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
11308 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
11309 | if (ondisk_format < latest_ondisk_format) { |
11310 | ceph_assert(ondisk_format > 0); | |
11311 | ceph_assert(ondisk_format < latest_ondisk_format); | |
11312 | ||
1911f103 | 11313 | KeyValueDB::Transaction t = db->get_transaction(); |
11fdf7f2 TL |
11314 | if (ondisk_format == 1) { |
11315 | // changes: | |
11316 | // - super: added ondisk_format | |
11317 | // - super: added min_readable_ondisk_format | |
11318 | // - super: added min_compat_ondisk_format | |
11319 | // - super: added min_alloc_size | |
11320 | // - super: removed min_min_alloc_size | |
11fdf7f2 TL |
11321 | { |
11322 | bufferlist bl; | |
11323 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
11324 | auto p = bl.cbegin(); | |
11325 | try { | |
11326 | uint64_t val; | |
11327 | decode(val, p); | |
11328 | min_alloc_size = val; | |
11329 | } catch (buffer::error& e) { | |
11330 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; | |
11331 | return -EIO; | |
11332 | } | |
11333 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
11334 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 11335 | } |
11fdf7f2 | 11336 | ondisk_format = 2; |
7c673cae | 11337 | } |
9f95a23c TL |
11338 | if (ondisk_format == 2) { |
11339 | // changes: | |
11340 | // - onode has FLAG_PER_POOL_OMAP. Note that we do not know that *all* | |
11341 | // ondes are using the per-pool prefix until a repair is run; at that | |
11342 | // point the per_pool_omap=1 key will be set. | |
11343 | // - super: added per_pool_omap key, which indicates that *all* objects | |
11344 | // are using the new prefix and key format | |
11345 | ondisk_format = 3; | |
1911f103 TL |
11346 | } |
11347 | if (ondisk_format == 3) { | |
11348 | // changes: | |
11349 | // - FreelistManager keeps meta within bdev label | |
11350 | int r = _write_out_fm_meta(0); | |
9f95a23c | 11351 | ceph_assert(r == 0); |
1911f103 | 11352 | ondisk_format = 4; |
9f95a23c | 11353 | } |
1911f103 TL |
11354 | // This to be the last operation |
11355 | _prepare_ondisk_format_super(t); | |
11356 | int r = db->submit_transaction_sync(t); | |
11357 | ceph_assert(r == 0); | |
7c673cae | 11358 | } |
7c673cae FG |
11359 | // done |
11360 | dout(1) << __func__ << " done" << dendl; | |
11361 | return 0; | |
11362 | } | |
11363 | ||
11364 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef o) | |
11365 | { | |
224ce89b | 11366 | if (o->onode.nid) { |
11fdf7f2 | 11367 | ceph_assert(o->exists); |
7c673cae | 11368 | return; |
224ce89b | 11369 | } |
7c673cae FG |
11370 | uint64_t nid = ++nid_last; |
11371 | dout(20) << __func__ << " " << nid << dendl; | |
11372 | o->onode.nid = nid; | |
11373 | txc->last_nid = nid; | |
224ce89b | 11374 | o->exists = true; |
7c673cae FG |
11375 | } |
11376 | ||
11377 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
11378 | { | |
11379 | uint64_t bid = ++blobid_last; | |
11380 | dout(20) << __func__ << " " << bid << dendl; | |
11381 | txc->last_blobid = bid; | |
11382 | return bid; | |
11383 | } | |
11384 | ||
11385 | void BlueStore::get_db_statistics(Formatter *f) | |
11386 | { | |
11387 | db->get_statistics(f); | |
11388 | } | |
11389 | ||
11fdf7f2 TL |
11390 | BlueStore::TransContext *BlueStore::_txc_create( |
11391 | Collection *c, OpSequencer *osr, | |
11392 | list<Context*> *on_commits) | |
7c673cae | 11393 | { |
11fdf7f2 | 11394 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae FG |
11395 | txc->t = db->get_transaction(); |
11396 | osr->queue_new(txc); | |
11397 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
11398 | << " seq " << txc->seq << dendl; | |
11399 | return txc; | |
11400 | } | |
11401 | ||
11402 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
11403 | { | |
11fdf7f2 TL |
11404 | // one "io" for the kv commit |
11405 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
11406 | auto cost = throttle_cost_per_io.load(); |
11407 | txc->cost = ios * cost + txc->bytes; | |
9f95a23c | 11408 | txc->ios = ios; |
7c673cae FG |
11409 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" |
11410 | << ios << " ios * " << cost << " + " << txc->bytes | |
11411 | << " bytes)" << dendl; | |
11412 | } | |
11413 | ||
11414 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
11415 | { | |
11416 | if (txc->statfs_delta.is_empty()) | |
11417 | return; | |
11418 | ||
11419 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
11420 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
11421 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
11422 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
11423 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
11424 | ||
11425 | bufferlist bl; | |
11426 | txc->statfs_delta.encode(bl); | |
11fdf7f2 TL |
11427 | if (per_pool_stat_collection) { |
11428 | string key; | |
11429 | get_pool_stat_key(txc->osd_pool_id, &key); | |
11430 | txc->t->merge(PREFIX_STAT, key, bl); | |
11431 | ||
11432 | std::lock_guard l(vstatfs_lock); | |
11433 | auto& stats = osd_pools[txc->osd_pool_id]; | |
11434 | stats += txc->statfs_delta; | |
11435 | ||
11436 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
11437 | ||
11438 | } else { | |
11439 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
7c673cae | 11440 | |
11fdf7f2 TL |
11441 | std::lock_guard l(vstatfs_lock); |
11442 | vstatfs += txc->statfs_delta; | |
11443 | } | |
7c673cae FG |
11444 | txc->statfs_delta.reset(); |
11445 | } | |
11446 | ||
11447 | void BlueStore::_txc_state_proc(TransContext *txc) | |
11448 | { | |
11449 | while (true) { | |
11450 | dout(10) << __func__ << " txc " << txc | |
11451 | << " " << txc->get_state_name() << dendl; | |
11452 | switch (txc->state) { | |
11453 | case TransContext::STATE_PREPARE: | |
9f95a23c | 11454 | throttle.log_state_latency(*txc, logger, l_bluestore_state_prepare_lat); |
7c673cae FG |
11455 | if (txc->ioc.has_pending_aios()) { |
11456 | txc->state = TransContext::STATE_AIO_WAIT; | |
11457 | txc->had_ios = true; | |
11458 | _txc_aio_submit(txc); | |
11459 | return; | |
11460 | } | |
11461 | // ** fall-thru ** | |
11462 | ||
11463 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 | 11464 | { |
9f95a23c TL |
11465 | mono_clock::duration lat = throttle.log_state_latency( |
11466 | *txc, logger, l_bluestore_state_aio_wait_lat); | |
11467 | if (ceph::to_seconds<double>(lat) >= cct->_conf->bluestore_log_op_age) { | |
11fdf7f2 TL |
11468 | dout(0) << __func__ << " slow aio_wait, txc = " << txc |
11469 | << ", latency = " << lat | |
11470 | << dendl; | |
11471 | } | |
11472 | } | |
11473 | ||
7c673cae FG |
11474 | _txc_finish_io(txc); // may trigger blocked txc's too |
11475 | return; | |
11476 | ||
11477 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 11478 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
11479 | if (txc->had_ios) { |
11480 | ++txc->osr->txc_with_unstable_io; | |
11481 | } | |
9f95a23c | 11482 | throttle.log_state_latency(*txc, logger, l_bluestore_state_io_done_lat); |
7c673cae FG |
11483 | txc->state = TransContext::STATE_KV_QUEUED; |
11484 | if (cct->_conf->bluestore_sync_submit_transaction) { | |
11485 | if (txc->last_nid >= nid_max || | |
11486 | txc->last_blobid >= blobid_max) { | |
11487 | dout(20) << __func__ | |
11488 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
11489 | << dendl; | |
11490 | } else if (txc->osr->kv_committing_serially) { | |
11491 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
11492 | << dendl; | |
11493 | // note: this is starvation-prone. once we have a txc in a busy | |
11494 | // sequencer that is committing serially it is possible to keep | |
11495 | // submitting new transactions fast enough that we get stuck doing | |
11496 | // so. the alternative is to block here... fixme? | |
11497 | } else if (txc->osr->txc_with_unstable_io) { | |
11498 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
11499 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
11500 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
11501 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
11502 | == 0) { | |
11503 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
11504 | << dendl; | |
11505 | } else { | |
9f95a23c | 11506 | _txc_apply_kv(txc, true); |
7c673cae FG |
11507 | } |
11508 | } | |
11509 | { | |
11fdf7f2 | 11510 | std::lock_guard l(kv_lock); |
7c673cae | 11511 | kv_queue.push_back(txc); |
9f95a23c TL |
11512 | if (!kv_sync_in_progress) { |
11513 | kv_sync_in_progress = true; | |
11514 | kv_cond.notify_one(); | |
11515 | } | |
7c673cae FG |
11516 | if (txc->state != TransContext::STATE_KV_SUBMITTED) { |
11517 | kv_queue_unsubmitted.push_back(txc); | |
11518 | ++txc->osr->kv_committing_serially; | |
11519 | } | |
31f18b77 FG |
11520 | if (txc->had_ios) |
11521 | kv_ios++; | |
11522 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
11523 | } |
11524 | return; | |
11525 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
11526 | _txc_committed_kv(txc); |
11527 | // ** fall-thru ** | |
11528 | ||
11529 | case TransContext::STATE_KV_DONE: | |
9f95a23c | 11530 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_done_lat); |
7c673cae FG |
11531 | if (txc->deferred_txn) { |
11532 | txc->state = TransContext::STATE_DEFERRED_QUEUED; | |
11533 | _deferred_queue(txc); | |
11534 | return; | |
11535 | } | |
11536 | txc->state = TransContext::STATE_FINISHING; | |
11537 | break; | |
11538 | ||
11539 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9f95a23c | 11540 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_cleanup_lat); |
7c673cae FG |
11541 | txc->state = TransContext::STATE_FINISHING; |
11542 | // ** fall-thru ** | |
11543 | ||
11544 | case TransContext::STATE_FINISHING: | |
9f95a23c | 11545 | throttle.log_state_latency(*txc, logger, l_bluestore_state_finishing_lat); |
7c673cae FG |
11546 | _txc_finish(txc); |
11547 | return; | |
11548 | ||
11549 | default: | |
11550 | derr << __func__ << " unexpected txc " << txc | |
11551 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 11552 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
11553 | return; |
11554 | } | |
11555 | } | |
11556 | } | |
11557 | ||
11558 | void BlueStore::_txc_finish_io(TransContext *txc) | |
11559 | { | |
11560 | dout(20) << __func__ << " " << txc << dendl; | |
11561 | ||
11562 | /* | |
11563 | * we need to preserve the order of kv transactions, | |
11564 | * even though aio will complete in any order. | |
11565 | */ | |
11566 | ||
11567 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 11568 | std::lock_guard l(osr->qlock); |
7c673cae | 11569 | txc->state = TransContext::STATE_IO_DONE; |
11fdf7f2 | 11570 | txc->ioc.release_running_aios(); |
7c673cae FG |
11571 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
11572 | while (p != osr->q.begin()) { | |
11573 | --p; | |
11574 | if (p->state < TransContext::STATE_IO_DONE) { | |
11575 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " | |
11576 | << p->get_state_name() << dendl; | |
11577 | return; | |
11578 | } | |
11579 | if (p->state > TransContext::STATE_IO_DONE) { | |
11580 | ++p; | |
11581 | break; | |
11582 | } | |
11583 | } | |
11584 | do { | |
11585 | _txc_state_proc(&*p++); | |
11586 | } while (p != osr->q.end() && | |
11587 | p->state == TransContext::STATE_IO_DONE); | |
11588 | ||
11fdf7f2 | 11589 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
11590 | osr->qcond.notify_all(); |
11591 | } | |
11592 | } | |
11593 | ||
11594 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
11595 | { | |
11596 | dout(20) << __func__ << " txc " << txc | |
11597 | << " onodes " << txc->onodes | |
11598 | << " shared_blobs " << txc->shared_blobs | |
11599 | << dendl; | |
11600 | ||
11601 | // finalize onodes | |
11602 | for (auto o : txc->onodes) { | |
11fdf7f2 | 11603 | _record_onode(o, t); |
7c673cae FG |
11604 | o->flushing_count++; |
11605 | } | |
11606 | ||
11607 | // objects we modified but didn't affect the onode | |
11608 | auto p = txc->modified_objects.begin(); | |
11609 | while (p != txc->modified_objects.end()) { | |
11610 | if (txc->onodes.count(*p) == 0) { | |
11611 | (*p)->flushing_count++; | |
11612 | ++p; | |
11613 | } else { | |
11614 | // remove dups with onodes list to avoid problems in _txc_finish | |
11615 | p = txc->modified_objects.erase(p); | |
11616 | } | |
11617 | } | |
11618 | ||
11619 | // finalize shared_blobs | |
11620 | for (auto sb : txc->shared_blobs) { | |
11621 | string key; | |
11622 | auto sbid = sb->get_sbid(); | |
11623 | get_shared_blob_key(sbid, &key); | |
11624 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
11625 | dout(20) << __func__ << " shared_blob 0x" |
11626 | << std::hex << sbid << std::dec | |
7c673cae FG |
11627 | << " is empty" << dendl; |
11628 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
11629 | } else { | |
11630 | bufferlist bl; | |
11fdf7f2 TL |
11631 | encode(*(sb->persistent), bl); |
11632 | dout(20) << __func__ << " shared_blob 0x" | |
11633 | << std::hex << sbid << std::dec | |
31f18b77 | 11634 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
11635 | t->set(PREFIX_SHARED_BLOB, key, bl); |
11636 | } | |
11637 | } | |
11638 | } | |
11639 | ||
11640 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
11641 | PerfCounters &logger) | |
11642 | { | |
11fdf7f2 TL |
11643 | os_commit_latency_ns.consume_next( |
11644 | logger.get_tavg_ns( | |
7c673cae | 11645 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
11646 | os_apply_latency_ns.consume_next( |
11647 | logger.get_tavg_ns( | |
7c673cae FG |
11648 | l_bluestore_commit_lat)); |
11649 | } | |
11650 | ||
11651 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) | |
11652 | { | |
11653 | dout(20) << __func__ << " txc " << txc << std::hex | |
11654 | << " allocated 0x" << txc->allocated | |
11655 | << " released 0x" << txc->released | |
11656 | << std::dec << dendl; | |
11657 | ||
11658 | // We have to handle the case where we allocate *and* deallocate the | |
11659 | // same region in this transaction. The freelist doesn't like that. | |
11660 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
11661 | // debug check. But that's important.) | |
11662 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
11663 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
11664 | interval_set<uint64_t> *preleased = &txc->released; | |
11665 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
11666 | interval_set<uint64_t> overlap; | |
11667 | overlap.intersection_of(txc->allocated, txc->released); | |
11668 | if (!overlap.empty()) { | |
11669 | tmp_allocated = txc->allocated; | |
11670 | tmp_allocated.subtract(overlap); | |
11671 | tmp_released = txc->released; | |
11672 | tmp_released.subtract(overlap); | |
11673 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
11674 | << ", new allocated 0x" << tmp_allocated | |
11675 | << " released 0x" << tmp_released << std::dec | |
11676 | << dendl; | |
11677 | pallocated = &tmp_allocated; | |
11678 | preleased = &tmp_released; | |
11679 | } | |
11680 | } | |
11681 | ||
11682 | // update freelist with non-overlap sets | |
11683 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
11684 | p != pallocated->end(); | |
11685 | ++p) { | |
11686 | fm->allocate(p.get_start(), p.get_len(), t); | |
11687 | } | |
11688 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
11689 | p != preleased->end(); | |
11690 | ++p) { | |
11691 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
11692 | << "~" << p.get_len() << std::dec << dendl; | |
11693 | fm->release(p.get_start(), p.get_len(), t); | |
11694 | } | |
11695 | ||
11696 | _txc_update_store_statfs(txc); | |
11697 | } | |
11698 | ||
9f95a23c | 11699 | void BlueStore::_txc_apply_kv(TransContext *txc, bool sync_submit_transaction) |
7c673cae | 11700 | { |
9f95a23c TL |
11701 | ceph_assert(txc->state == TransContext::STATE_KV_QUEUED); |
11702 | { | |
11703 | #if defined(WITH_LTTNG) | |
11704 | auto start = mono_clock::now(); | |
11705 | #endif | |
11706 | ||
11707 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); | |
11708 | ceph_assert(r == 0); | |
11709 | txc->state = TransContext::STATE_KV_SUBMITTED; | |
11710 | if (txc->osr->kv_submitted_waiters) { | |
11711 | std::lock_guard l(txc->osr->qlock); | |
11712 | txc->osr->qcond.notify_all(); | |
11713 | } | |
11714 | ||
11715 | #if defined(WITH_LTTNG) | |
11716 | if (txc->tracing) { | |
11717 | tracepoint( | |
11718 | bluestore, | |
11719 | transaction_kv_submit_latency, | |
11720 | txc->osr->get_sequencer_id(), | |
11721 | txc->seq, | |
11722 | sync_submit_transaction, | |
11723 | ceph::to_seconds<double>(mono_clock::now() - start)); | |
11724 | } | |
11725 | #endif | |
11726 | } | |
11727 | ||
7c673cae FG |
11728 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { |
11729 | for (auto& o : *ls) { | |
11730 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
11731 | << dendl; | |
9f95a23c | 11732 | if (--o->flushing_count == 0 && o->waiting_count.load()) { |
11fdf7f2 | 11733 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
11734 | o->flush_cond.notify_all(); |
11735 | } | |
11736 | } | |
11737 | } | |
11738 | } | |
11739 | ||
11740 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
11741 | { | |
11742 | dout(20) << __func__ << " txc " << txc << dendl; | |
9f95a23c | 11743 | throttle.complete_kv(*txc); |
1adf2230 | 11744 | { |
11fdf7f2 | 11745 | std::lock_guard l(txc->osr->qlock); |
1adf2230 | 11746 | txc->state = TransContext::STATE_KV_DONE; |
11fdf7f2 TL |
11747 | if (txc->ch->commit_queue) { |
11748 | txc->ch->commit_queue->queue(txc->oncommits); | |
11749 | } else { | |
11750 | finisher.queue(txc->oncommits); | |
1adf2230 | 11751 | } |
7c673cae | 11752 | } |
9f95a23c | 11753 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_committing_lat); |
494da23a TL |
11754 | log_latency_fn( |
11755 | __func__, | |
11756 | l_bluestore_commit_lat, | |
9f95a23c | 11757 | mono_clock::now() - txc->start, |
494da23a TL |
11758 | cct->_conf->bluestore_log_op_age, |
11759 | [&](auto lat) { | |
11760 | return ", txc = " + stringify(txc); | |
11761 | } | |
11fdf7f2 | 11762 | ); |
7c673cae FG |
11763 | } |
11764 | ||
11765 | void BlueStore::_txc_finish(TransContext *txc) | |
11766 | { | |
11767 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
11fdf7f2 | 11768 | ceph_assert(txc->state == TransContext::STATE_FINISHING); |
7c673cae FG |
11769 | |
11770 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 11771 | sb->finish_write(txc->seq); |
7c673cae FG |
11772 | } |
11773 | txc->shared_blobs_written.clear(); | |
11774 | ||
11775 | while (!txc->removed_collections.empty()) { | |
11776 | _queue_reap_collection(txc->removed_collections.front()); | |
11777 | txc->removed_collections.pop_front(); | |
11778 | } | |
11779 | ||
11780 | OpSequencerRef osr = txc->osr; | |
7c673cae | 11781 | bool empty = false; |
31f18b77 | 11782 | bool submit_deferred = false; |
7c673cae FG |
11783 | OpSequencer::q_list_t releasing_txc; |
11784 | { | |
11fdf7f2 | 11785 | std::lock_guard l(osr->qlock); |
7c673cae FG |
11786 | txc->state = TransContext::STATE_DONE; |
11787 | bool notify = false; | |
11788 | while (!osr->q.empty()) { | |
11789 | TransContext *txc = &osr->q.front(); | |
11790 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
11791 | << dendl; | |
11792 | if (txc->state != TransContext::STATE_DONE) { | |
11793 | if (txc->state == TransContext::STATE_PREPARE && | |
11794 | deferred_aggressive) { | |
11795 | // for _osr_drain_preceding() | |
11796 | notify = true; | |
11797 | } | |
31f18b77 | 11798 | if (txc->state == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 11799 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
11800 | submit_deferred = true; |
11801 | } | |
7c673cae FG |
11802 | break; |
11803 | } | |
11804 | ||
7c673cae FG |
11805 | osr->q.pop_front(); |
11806 | releasing_txc.push_back(*txc); | |
7c673cae | 11807 | } |
9f95a23c | 11808 | |
7c673cae FG |
11809 | if (osr->q.empty()) { |
11810 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
11811 | empty = true; | |
11812 | } | |
9f95a23c TL |
11813 | |
11814 | // only drain()/drain_preceding() need wakeup, | |
11815 | // other cases use kv_submitted_waiters | |
11816 | if (notify || empty) { | |
11817 | osr->qcond.notify_all(); | |
11818 | } | |
7c673cae | 11819 | } |
9f95a23c | 11820 | |
7c673cae FG |
11821 | while (!releasing_txc.empty()) { |
11822 | // release to allocator only after all preceding txc's have also | |
11823 | // finished any deferred writes that potentially land in these | |
11824 | // blocks | |
11825 | auto txc = &releasing_txc.front(); | |
11826 | _txc_release_alloc(txc); | |
11827 | releasing_txc.pop_front(); | |
9f95a23c TL |
11828 | throttle.log_state_latency(*txc, logger, l_bluestore_state_done_lat); |
11829 | throttle.complete(*txc); | |
7c673cae FG |
11830 | delete txc; |
11831 | } | |
11832 | ||
31f18b77 FG |
11833 | if (submit_deferred) { |
11834 | // we're pinning memory; flush! we could be more fine-grained here but | |
11835 | // i'm not sure it's worth the bother. | |
11836 | deferred_try_submit(); | |
7c673cae FG |
11837 | } |
11838 | ||
7c673cae | 11839 | if (empty && osr->zombie) { |
11fdf7f2 TL |
11840 | std::lock_guard l(zombie_osr_lock); |
11841 | if (zombie_osr_set.erase(osr->cid)) { | |
11842 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11843 | } else { | |
11844 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
11845 | << dendl; | |
11846 | } | |
7c673cae | 11847 | } |
9f95a23c | 11848 | } |
7c673cae FG |
11849 | |
11850 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
11851 | { | |
a8e16298 | 11852 | // it's expected we're called with lazy_release_lock already taken! |
11fdf7f2 TL |
11853 | if (likely(!cct->_conf->bluestore_debug_no_reuse_blocks)) { |
11854 | int r = 0; | |
11855 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
11856 | r = bdev->queue_discard(txc->released); | |
11857 | if (r == 0) { | |
11858 | dout(10) << __func__ << "(queued) " << txc << " " << std::hex | |
11859 | << txc->released << std::dec << dendl; | |
11860 | goto out; | |
11861 | } | |
11862 | } else if (cct->_conf->bdev_enable_discard) { | |
11863 | for (auto p = txc->released.begin(); p != txc->released.end(); ++p) { | |
11864 | bdev->discard(p.get_start(), p.get_len()); | |
11865 | } | |
11866 | } | |
11867 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
94b18763 | 11868 | << txc->released << std::dec << dendl; |
11fdf7f2 | 11869 | alloc->release(txc->released); |
7c673cae FG |
11870 | } |
11871 | ||
11fdf7f2 | 11872 | out: |
7c673cae FG |
11873 | txc->allocated.clear(); |
11874 | txc->released.clear(); | |
11875 | } | |
11876 | ||
11fdf7f2 TL |
11877 | void BlueStore::_osr_attach(Collection *c) |
11878 | { | |
11879 | // note: caller has RWLock on coll_map | |
11880 | auto q = coll_map.find(c->cid); | |
11881 | if (q != coll_map.end()) { | |
11882 | c->osr = q->second->osr; | |
11883 | ldout(cct, 10) << __func__ << " " << c->cid | |
11884 | << " reusing osr " << c->osr << " from existing coll " | |
11885 | << q->second << dendl; | |
11886 | } else { | |
11887 | std::lock_guard l(zombie_osr_lock); | |
11888 | auto p = zombie_osr_set.find(c->cid); | |
11889 | if (p == zombie_osr_set.end()) { | |
9f95a23c | 11890 | c->osr = ceph::make_ref<OpSequencer>(this, next_sequencer_id++, c->cid); |
11fdf7f2 TL |
11891 | ldout(cct, 10) << __func__ << " " << c->cid |
11892 | << " fresh osr " << c->osr << dendl; | |
11893 | } else { | |
11894 | c->osr = p->second; | |
11895 | zombie_osr_set.erase(p); | |
11896 | ldout(cct, 10) << __func__ << " " << c->cid | |
11897 | << " resurrecting zombie osr " << c->osr << dendl; | |
11898 | c->osr->zombie = false; | |
11899 | } | |
11900 | } | |
11901 | } | |
11902 | ||
11903 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
11904 | { | |
11905 | std::lock_guard l(zombie_osr_lock); | |
11906 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
11907 | osr->zombie = true; | |
11908 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
11909 | // this is either a new insertion or the same osr is already there | |
11910 | ceph_assert(i.second || i.first->second == osr); | |
11911 | } | |
11912 | ||
7c673cae FG |
11913 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
11914 | { | |
11915 | OpSequencer *osr = txc->osr.get(); | |
11916 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
11917 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11918 | { | |
11919 | // submit anything pending | |
224ce89b | 11920 | deferred_lock.lock(); |
11fdf7f2 | 11921 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
11922 | _deferred_submit_unlock(osr); |
11923 | } else { | |
11924 | deferred_lock.unlock(); | |
7c673cae FG |
11925 | } |
11926 | } | |
11927 | { | |
11928 | // wake up any previously finished deferred events | |
11fdf7f2 | 11929 | std::lock_guard l(kv_lock); |
9f95a23c TL |
11930 | if (!kv_sync_in_progress) { |
11931 | kv_sync_in_progress = true; | |
11932 | kv_cond.notify_one(); | |
11933 | } | |
7c673cae FG |
11934 | } |
11935 | osr->drain_preceding(txc); | |
11936 | --deferred_aggressive; | |
11937 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11938 | } | |
11939 | ||
11fdf7f2 TL |
11940 | void BlueStore::_osr_drain(OpSequencer *osr) |
11941 | { | |
11942 | dout(10) << __func__ << " " << osr << dendl; | |
11943 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11944 | { | |
11945 | // submit anything pending | |
11946 | deferred_lock.lock(); | |
11947 | if (osr->deferred_pending && !osr->deferred_running) { | |
11948 | _deferred_submit_unlock(osr); | |
11949 | } else { | |
11950 | deferred_lock.unlock(); | |
11951 | } | |
11952 | } | |
11953 | { | |
11954 | // wake up any previously finished deferred events | |
11955 | std::lock_guard l(kv_lock); | |
9f95a23c TL |
11956 | if (!kv_sync_in_progress) { |
11957 | kv_sync_in_progress = true; | |
11958 | kv_cond.notify_one(); | |
11959 | } | |
11fdf7f2 TL |
11960 | } |
11961 | osr->drain(); | |
11962 | --deferred_aggressive; | |
11963 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11964 | } | |
11965 | ||
7c673cae FG |
11966 | void BlueStore::_osr_drain_all() |
11967 | { | |
11968 | dout(10) << __func__ << dendl; | |
11969 | ||
11970 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
11971 | vector<OpSequencerRef> zombies; |
11972 | { | |
9f95a23c | 11973 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
11974 | for (auto& i : coll_map) { |
11975 | s.insert(i.second->osr); | |
11976 | } | |
11977 | } | |
7c673cae | 11978 | { |
11fdf7f2 TL |
11979 | std::lock_guard l(zombie_osr_lock); |
11980 | for (auto& i : zombie_osr_set) { | |
11981 | s.insert(i.second); | |
11982 | zombies.push_back(i.second); | |
11983 | } | |
7c673cae FG |
11984 | } |
11985 | dout(20) << __func__ << " osr_set " << s << dendl; | |
11986 | ||
11987 | ++deferred_aggressive; | |
11988 | { | |
11989 | // submit anything pending | |
224ce89b | 11990 | deferred_try_submit(); |
7c673cae FG |
11991 | } |
11992 | { | |
11993 | // wake up any previously finished deferred events | |
11fdf7f2 | 11994 | std::lock_guard l(kv_lock); |
7c673cae FG |
11995 | kv_cond.notify_one(); |
11996 | } | |
31f18b77 | 11997 | { |
11fdf7f2 | 11998 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11999 | kv_finalize_cond.notify_one(); |
12000 | } | |
7c673cae FG |
12001 | for (auto osr : s) { |
12002 | dout(20) << __func__ << " drain " << osr << dendl; | |
12003 | osr->drain(); | |
12004 | } | |
12005 | --deferred_aggressive; | |
12006 | ||
7c673cae | 12007 | { |
11fdf7f2 TL |
12008 | std::lock_guard l(zombie_osr_lock); |
12009 | for (auto& osr : zombies) { | |
12010 | if (zombie_osr_set.erase(osr->cid)) { | |
12011 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
12012 | ceph_assert(osr->q.empty()); | |
12013 | } else if (osr->zombie) { | |
12014 | dout(10) << __func__ << " empty zombie osr " << osr | |
12015 | << " already reaped" << dendl; | |
12016 | ceph_assert(osr->q.empty()); | |
12017 | } else { | |
12018 | dout(10) << __func__ << " empty zombie osr " << osr | |
12019 | << " resurrected" << dendl; | |
12020 | } | |
7c673cae FG |
12021 | } |
12022 | } | |
11fdf7f2 TL |
12023 | |
12024 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
12025 | } |
12026 | ||
11fdf7f2 | 12027 | |
31f18b77 FG |
12028 | void BlueStore::_kv_start() |
12029 | { | |
12030 | dout(10) << __func__ << dendl; | |
12031 | ||
11fdf7f2 | 12032 | finisher.start(); |
31f18b77 FG |
12033 | kv_sync_thread.create("bstore_kv_sync"); |
12034 | kv_finalize_thread.create("bstore_kv_final"); | |
12035 | } | |
12036 | ||
12037 | void BlueStore::_kv_stop() | |
12038 | { | |
12039 | dout(10) << __func__ << dendl; | |
12040 | { | |
9f95a23c | 12041 | std::unique_lock l{kv_lock}; |
31f18b77 FG |
12042 | while (!kv_sync_started) { |
12043 | kv_cond.wait(l); | |
12044 | } | |
12045 | kv_stop = true; | |
12046 | kv_cond.notify_all(); | |
12047 | } | |
12048 | { | |
9f95a23c | 12049 | std::unique_lock l{kv_finalize_lock}; |
31f18b77 FG |
12050 | while (!kv_finalize_started) { |
12051 | kv_finalize_cond.wait(l); | |
12052 | } | |
12053 | kv_finalize_stop = true; | |
12054 | kv_finalize_cond.notify_all(); | |
12055 | } | |
12056 | kv_sync_thread.join(); | |
12057 | kv_finalize_thread.join(); | |
11fdf7f2 | 12058 | ceph_assert(removed_collections.empty()); |
31f18b77 | 12059 | { |
11fdf7f2 | 12060 | std::lock_guard l(kv_lock); |
31f18b77 FG |
12061 | kv_stop = false; |
12062 | } | |
12063 | { | |
11fdf7f2 | 12064 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
12065 | kv_finalize_stop = false; |
12066 | } | |
12067 | dout(10) << __func__ << " stopping finishers" << dendl; | |
11fdf7f2 TL |
12068 | finisher.wait_for_empty(); |
12069 | finisher.stop(); | |
31f18b77 FG |
12070 | dout(10) << __func__ << " stopped" << dendl; |
12071 | } | |
12072 | ||
7c673cae FG |
12073 | void BlueStore::_kv_sync_thread() |
12074 | { | |
12075 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 | 12076 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
9f95a23c | 12077 | std::unique_lock l{kv_lock}; |
11fdf7f2 | 12078 | ceph_assert(!kv_sync_started); |
31f18b77 FG |
12079 | kv_sync_started = true; |
12080 | kv_cond.notify_all(); | |
7c673cae | 12081 | while (true) { |
11fdf7f2 | 12082 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
12083 | if (kv_queue.empty() && |
12084 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 12085 | !deferred_aggressive)) { |
7c673cae FG |
12086 | if (kv_stop) |
12087 | break; | |
12088 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 12089 | kv_sync_in_progress = false; |
11fdf7f2 | 12090 | kv_cond.wait(l); |
7c673cae FG |
12091 | dout(20) << __func__ << " wake" << dendl; |
12092 | } else { | |
12093 | deque<TransContext*> kv_submitting; | |
12094 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
12095 | uint64_t aios = 0, costs = 0; |
12096 | ||
7c673cae FG |
12097 | dout(20) << __func__ << " committing " << kv_queue.size() |
12098 | << " submitting " << kv_queue_unsubmitted.size() | |
12099 | << " deferred done " << deferred_done_queue.size() | |
12100 | << " stable " << deferred_stable_queue.size() | |
12101 | << dendl; | |
12102 | kv_committing.swap(kv_queue); | |
12103 | kv_submitting.swap(kv_queue_unsubmitted); | |
12104 | deferred_done.swap(deferred_done_queue); | |
12105 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
12106 | aios = kv_ios; |
12107 | costs = kv_throttle_costs; | |
12108 | kv_ios = 0; | |
12109 | kv_throttle_costs = 0; | |
7c673cae FG |
12110 | l.unlock(); |
12111 | ||
12112 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
12113 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
12114 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
12115 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12116 | ||
11fdf7f2 TL |
12117 | auto start = mono_clock::now(); |
12118 | ||
7c673cae FG |
12119 | bool force_flush = false; |
12120 | // if bluefs is sharing the same device as data (only), then we | |
12121 | // can rely on the bluefs commit to flush the device and make | |
12122 | // deferred aios stable. that means that if we do have done deferred | |
12123 | // txcs AND we are not on a single device, we need to force a flush. | |
9f95a23c | 12124 | if (bluefs && bluefs_layout.single_shared_device()) { |
31f18b77 | 12125 | if (aios) { |
7c673cae | 12126 | force_flush = true; |
11fdf7f2 | 12127 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
12128 | force_flush = true; // there's nothing else to commit! |
12129 | } else if (deferred_aggressive) { | |
12130 | force_flush = true; | |
12131 | } | |
11fdf7f2 TL |
12132 | } else { |
12133 | if (aios || !deferred_done.empty()) { | |
12134 | force_flush = true; | |
12135 | } else { | |
12136 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
12137 | } | |
12138 | } | |
7c673cae FG |
12139 | |
12140 | if (force_flush) { | |
31f18b77 | 12141 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
12142 | << " force_flush=" << (int)force_flush |
12143 | << ", flushing, deferred done->stable" << dendl; | |
12144 | // flush/barrier on block device | |
12145 | bdev->flush(); | |
12146 | ||
12147 | // if we flush then deferred done are now deferred stable | |
12148 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
12149 | deferred_done.end()); | |
12150 | deferred_done.clear(); | |
12151 | } | |
11fdf7f2 | 12152 | auto after_flush = mono_clock::now(); |
7c673cae FG |
12153 | |
12154 | // we will use one final transaction to force a sync | |
12155 | KeyValueDB::Transaction synct = db->get_transaction(); | |
12156 | ||
12157 | // increase {nid,blobid}_max? note that this covers both the | |
12158 | // case where we are approaching the max and the case we passed | |
12159 | // it. in either case, we increase the max in the earlier txn | |
12160 | // we submit. | |
12161 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
12162 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
12163 | KeyValueDB::Transaction t = | |
12164 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
12165 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
12166 | bufferlist bl; | |
11fdf7f2 | 12167 | encode(new_nid_max, bl); |
7c673cae FG |
12168 | t->set(PREFIX_SUPER, "nid_max", bl); |
12169 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
12170 | } | |
12171 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
12172 | KeyValueDB::Transaction t = | |
12173 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
12174 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
12175 | bufferlist bl; | |
11fdf7f2 | 12176 | encode(new_blobid_max, bl); |
7c673cae FG |
12177 | t->set(PREFIX_SUPER, "blobid_max", bl); |
12178 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
12179 | } | |
c07f9fc5 FG |
12180 | |
12181 | for (auto txc : kv_committing) { | |
9f95a23c | 12182 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_queued_lat); |
c07f9fc5 | 12183 | if (txc->state == TransContext::STATE_KV_QUEUED) { |
9f95a23c | 12184 | _txc_apply_kv(txc, false); |
c07f9fc5 | 12185 | --txc->osr->kv_committing_serially; |
c07f9fc5 | 12186 | } else { |
11fdf7f2 | 12187 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12188 | } |
7c673cae FG |
12189 | if (txc->had_ios) { |
12190 | --txc->osr->txc_with_unstable_io; | |
12191 | } | |
7c673cae FG |
12192 | } |
12193 | ||
31f18b77 FG |
12194 | // release throttle *before* we commit. this allows new ops |
12195 | // to be prepared and enter pipeline while we are waiting on | |
12196 | // the kv commit sync/flush. then hopefully on the next | |
12197 | // iteration there will already be ops awake. otherwise, we | |
12198 | // end up going to sleep, and then wake up when the very first | |
12199 | // transaction is ready for commit. | |
9f95a23c | 12200 | throttle.release_kv_throttle(costs); |
31f18b77 | 12201 | |
7c673cae FG |
12202 | if (bluefs && |
12203 | after_flush - bluefs_last_balance > | |
11fdf7f2 | 12204 | ceph::make_timespan(cct->_conf->bluestore_bluefs_balance_interval)) { |
7c673cae | 12205 | bluefs_last_balance = after_flush; |
11fdf7f2 TL |
12206 | int r = _balance_bluefs_freespace(); |
12207 | ceph_assert(r >= 0); | |
7c673cae FG |
12208 | } |
12209 | ||
12210 | // cleanup sync deferred keys | |
12211 | for (auto b : deferred_stable) { | |
12212 | for (auto& txc : b->txcs) { | |
12213 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 12214 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
12215 | string key; |
12216 | get_deferred_key(wt.seq, &key); | |
12217 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
12218 | } | |
12219 | } | |
12220 | ||
9f95a23c TL |
12221 | #if defined(WITH_LTTNG) |
12222 | auto sync_start = mono_clock::now(); | |
12223 | #endif | |
7c673cae | 12224 | // submit synct synchronously (block and wait for it to commit) |
31f18b77 | 12225 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
12226 | ceph_assert(r == 0); |
12227 | ||
9f95a23c TL |
12228 | int committing_size = kv_committing.size(); |
12229 | int deferred_size = deferred_stable.size(); | |
12230 | ||
12231 | #if defined(WITH_LTTNG) | |
12232 | double sync_latency = ceph::to_seconds<double>(mono_clock::now() - sync_start); | |
12233 | for (auto txc: kv_committing) { | |
12234 | if (txc->tracing) { | |
12235 | tracepoint( | |
12236 | bluestore, | |
12237 | transaction_kv_sync_latency, | |
12238 | txc->osr->get_sequencer_id(), | |
12239 | txc->seq, | |
12240 | kv_committing.size(), | |
12241 | deferred_done.size(), | |
12242 | deferred_stable.size(), | |
12243 | sync_latency); | |
12244 | } | |
12245 | } | |
12246 | #endif | |
12247 | ||
11fdf7f2 | 12248 | { |
9f95a23c | 12249 | std::unique_lock m{kv_finalize_lock}; |
11fdf7f2 TL |
12250 | if (kv_committing_to_finalize.empty()) { |
12251 | kv_committing_to_finalize.swap(kv_committing); | |
12252 | } else { | |
12253 | kv_committing_to_finalize.insert( | |
12254 | kv_committing_to_finalize.end(), | |
12255 | kv_committing.begin(), | |
12256 | kv_committing.end()); | |
12257 | kv_committing.clear(); | |
12258 | } | |
12259 | if (deferred_stable_to_finalize.empty()) { | |
12260 | deferred_stable_to_finalize.swap(deferred_stable); | |
12261 | } else { | |
12262 | deferred_stable_to_finalize.insert( | |
12263 | deferred_stable_to_finalize.end(), | |
12264 | deferred_stable.begin(), | |
12265 | deferred_stable.end()); | |
12266 | deferred_stable.clear(); | |
12267 | } | |
9f95a23c TL |
12268 | if (!kv_finalize_in_progress) { |
12269 | kv_finalize_in_progress = true; | |
12270 | kv_finalize_cond.notify_one(); | |
12271 | } | |
11fdf7f2 | 12272 | } |
7c673cae FG |
12273 | |
12274 | if (new_nid_max) { | |
12275 | nid_max = new_nid_max; | |
12276 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
12277 | } | |
12278 | if (new_blobid_max) { | |
12279 | blobid_max = new_blobid_max; | |
12280 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
12281 | } | |
12282 | ||
224ce89b | 12283 | { |
11fdf7f2 TL |
12284 | auto finish = mono_clock::now(); |
12285 | ceph::timespan dur_flush = after_flush - start; | |
12286 | ceph::timespan dur_kv = finish - after_flush; | |
12287 | ceph::timespan dur = finish - start; | |
9f95a23c TL |
12288 | dout(20) << __func__ << " committed " << committing_size |
12289 | << " cleaned " << deferred_size | |
224ce89b WB |
12290 | << " in " << dur |
12291 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
12292 | << dendl; | |
494da23a TL |
12293 | log_latency("kv_flush", |
12294 | l_bluestore_kv_flush_lat, | |
12295 | dur_flush, | |
12296 | cct->_conf->bluestore_log_op_age); | |
12297 | log_latency("kv_commit", | |
12298 | l_bluestore_kv_commit_lat, | |
12299 | dur_kv, | |
12300 | cct->_conf->bluestore_log_op_age); | |
12301 | log_latency("kv_sync", | |
12302 | l_bluestore_kv_sync_lat, | |
12303 | dur, | |
12304 | cct->_conf->bluestore_log_op_age); | |
7c673cae | 12305 | } |
31f18b77 FG |
12306 | |
12307 | if (bluefs) { | |
11fdf7f2 TL |
12308 | if (!bluefs_extents_reclaiming.empty()) { |
12309 | dout(0) << __func__ << " releasing old bluefs 0x" << std::hex | |
12310 | << bluefs_extents_reclaiming << std::dec << dendl; | |
81eedcae TL |
12311 | int r = 0; |
12312 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
12313 | r = bdev->queue_discard(bluefs_extents_reclaiming); | |
12314 | if (r == 0) { | |
12315 | goto clear; | |
12316 | } | |
12317 | } else if (cct->_conf->bdev_enable_discard) { | |
12318 | for (auto p = bluefs_extents_reclaiming.begin(); p != bluefs_extents_reclaiming.end(); ++p) { | |
12319 | bdev->discard(p.get_start(), p.get_len()); | |
12320 | } | |
12321 | } | |
12322 | ||
11fdf7f2 | 12323 | alloc->release(bluefs_extents_reclaiming); |
81eedcae | 12324 | clear: |
11fdf7f2 | 12325 | bluefs_extents_reclaiming.clear(); |
31f18b77 | 12326 | } |
31f18b77 FG |
12327 | } |
12328 | ||
12329 | l.lock(); | |
12330 | // previously deferred "done" are now "stable" by virtue of this | |
12331 | // commit cycle. | |
12332 | deferred_stable_queue.swap(deferred_done); | |
12333 | } | |
12334 | } | |
12335 | dout(10) << __func__ << " finish" << dendl; | |
12336 | kv_sync_started = false; | |
12337 | } | |
12338 | ||
12339 | void BlueStore::_kv_finalize_thread() | |
12340 | { | |
12341 | deque<TransContext*> kv_committed; | |
12342 | deque<DeferredBatch*> deferred_stable; | |
12343 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
12344 | std::unique_lock l(kv_finalize_lock); |
12345 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
12346 | kv_finalize_started = true; |
12347 | kv_finalize_cond.notify_all(); | |
12348 | while (true) { | |
11fdf7f2 TL |
12349 | ceph_assert(kv_committed.empty()); |
12350 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
12351 | if (kv_committing_to_finalize.empty() && |
12352 | deferred_stable_to_finalize.empty()) { | |
12353 | if (kv_finalize_stop) | |
12354 | break; | |
12355 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 12356 | kv_finalize_in_progress = false; |
31f18b77 FG |
12357 | kv_finalize_cond.wait(l); |
12358 | dout(20) << __func__ << " wake" << dendl; | |
12359 | } else { | |
12360 | kv_committed.swap(kv_committing_to_finalize); | |
12361 | deferred_stable.swap(deferred_stable_to_finalize); | |
12362 | l.unlock(); | |
12363 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
12364 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12365 | ||
11fdf7f2 TL |
12366 | auto start = mono_clock::now(); |
12367 | ||
31f18b77 FG |
12368 | while (!kv_committed.empty()) { |
12369 | TransContext *txc = kv_committed.front(); | |
11fdf7f2 | 12370 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12371 | _txc_state_proc(txc); |
31f18b77 | 12372 | kv_committed.pop_front(); |
7c673cae | 12373 | } |
31f18b77 | 12374 | |
7c673cae FG |
12375 | for (auto b : deferred_stable) { |
12376 | auto p = b->txcs.begin(); | |
12377 | while (p != b->txcs.end()) { | |
12378 | TransContext *txc = &*p; | |
12379 | p = b->txcs.erase(p); // unlink here because | |
12380 | _txc_state_proc(txc); // this may destroy txc | |
12381 | } | |
12382 | delete b; | |
12383 | } | |
31f18b77 | 12384 | deferred_stable.clear(); |
7c673cae FG |
12385 | |
12386 | if (!deferred_aggressive) { | |
31f18b77 | 12387 | if (deferred_queue_size >= deferred_batch_ops.load() || |
9f95a23c | 12388 | throttle.should_submit_deferred()) { |
224ce89b | 12389 | deferred_try_submit(); |
7c673cae FG |
12390 | } |
12391 | } | |
12392 | ||
12393 | // this is as good a place as any ... | |
12394 | _reap_collections(); | |
12395 | ||
11fdf7f2 | 12396 | logger->set(l_bluestore_fragmentation, |
9f95a23c | 12397 | (uint64_t)(alloc->get_fragmentation() * 1000)); |
11fdf7f2 | 12398 | |
494da23a TL |
12399 | log_latency("kv_final", |
12400 | l_bluestore_kv_final_lat, | |
12401 | mono_clock::now() - start, | |
12402 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 | 12403 | |
7c673cae | 12404 | l.lock(); |
7c673cae FG |
12405 | } |
12406 | } | |
12407 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 12408 | kv_finalize_started = false; |
7c673cae FG |
12409 | } |
12410 | ||
12411 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( | |
9f95a23c | 12412 | TransContext *txc) |
7c673cae FG |
12413 | { |
12414 | if (!txc->deferred_txn) { | |
12415 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
12416 | } | |
12417 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
12418 | return &txc->deferred_txn->ops.back(); | |
12419 | } | |
12420 | ||
12421 | void BlueStore::_deferred_queue(TransContext *txc) | |
12422 | { | |
12423 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
224ce89b | 12424 | deferred_lock.lock(); |
7c673cae FG |
12425 | if (!txc->osr->deferred_pending && |
12426 | !txc->osr->deferred_running) { | |
12427 | deferred_queue.push_back(*txc->osr); | |
12428 | } | |
12429 | if (!txc->osr->deferred_pending) { | |
12430 | txc->osr->deferred_pending = new DeferredBatch(cct, txc->osr.get()); | |
12431 | } | |
12432 | ++deferred_queue_size; | |
12433 | txc->osr->deferred_pending->txcs.push_back(*txc); | |
12434 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; | |
12435 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
12436 | const auto& op = *opi; | |
11fdf7f2 | 12437 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
12438 | bufferlist::const_iterator p = op.data.begin(); |
12439 | for (auto e : op.extents) { | |
12440 | txc->osr->deferred_pending->prepare_write( | |
12441 | cct, wt.seq, e.offset, e.length, p); | |
12442 | } | |
12443 | } | |
12444 | if (deferred_aggressive && | |
12445 | !txc->osr->deferred_running) { | |
224ce89b WB |
12446 | _deferred_submit_unlock(txc->osr.get()); |
12447 | } else { | |
12448 | deferred_lock.unlock(); | |
7c673cae FG |
12449 | } |
12450 | } | |
12451 | ||
224ce89b | 12452 | void BlueStore::deferred_try_submit() |
7c673cae FG |
12453 | { |
12454 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
12455 | << deferred_queue_size << " txcs" << dendl; | |
11fdf7f2 | 12456 | std::lock_guard l(deferred_lock); |
224ce89b WB |
12457 | vector<OpSequencerRef> osrs; |
12458 | osrs.reserve(deferred_queue.size()); | |
7c673cae | 12459 | for (auto& osr : deferred_queue) { |
224ce89b WB |
12460 | osrs.push_back(&osr); |
12461 | } | |
12462 | for (auto& osr : osrs) { | |
181888fb FG |
12463 | if (osr->deferred_pending) { |
12464 | if (!osr->deferred_running) { | |
12465 | _deferred_submit_unlock(osr.get()); | |
12466 | deferred_lock.lock(); | |
12467 | } else { | |
12468 | dout(20) << __func__ << " osr " << osr << " already has running" | |
12469 | << dendl; | |
12470 | } | |
12471 | } else { | |
12472 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; | |
7c673cae FG |
12473 | } |
12474 | } | |
9f95a23c TL |
12475 | |
12476 | deferred_last_submitted = ceph_clock_now(); | |
7c673cae FG |
12477 | } |
12478 | ||
224ce89b | 12479 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
12480 | { |
12481 | dout(10) << __func__ << " osr " << osr | |
12482 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
12483 | << dendl; | |
11fdf7f2 TL |
12484 | ceph_assert(osr->deferred_pending); |
12485 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
12486 | |
12487 | auto b = osr->deferred_pending; | |
12488 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 12489 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
12490 | |
12491 | osr->deferred_running = osr->deferred_pending; | |
12492 | osr->deferred_pending = nullptr; | |
12493 | ||
11fdf7f2 TL |
12494 | deferred_lock.unlock(); |
12495 | ||
12496 | for (auto& txc : b->txcs) { | |
9f95a23c | 12497 | throttle.log_state_latency(txc, logger, l_bluestore_state_deferred_queued_lat); |
11fdf7f2 | 12498 | } |
7c673cae FG |
12499 | uint64_t start = 0, pos = 0; |
12500 | bufferlist bl; | |
12501 | auto i = b->iomap.begin(); | |
12502 | while (true) { | |
12503 | if (i == b->iomap.end() || i->first != pos) { | |
12504 | if (bl.length()) { | |
12505 | dout(20) << __func__ << " write 0x" << std::hex | |
12506 | << start << "~" << bl.length() | |
12507 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 12508 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
12509 | logger->inc(l_bluestore_deferred_write_ops); |
12510 | logger->inc(l_bluestore_deferred_write_bytes, bl.length()); | |
12511 | int r = bdev->aio_write(start, bl, &b->ioc, false); | |
11fdf7f2 | 12512 | ceph_assert(r == 0); |
7c673cae FG |
12513 | } |
12514 | } | |
12515 | if (i == b->iomap.end()) { | |
12516 | break; | |
12517 | } | |
12518 | start = 0; | |
12519 | pos = i->first; | |
12520 | bl.clear(); | |
12521 | } | |
12522 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
12523 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
12524 | << dendl; | |
12525 | if (!bl.length()) { | |
12526 | start = pos; | |
12527 | } | |
12528 | pos += i->second.bl.length(); | |
12529 | bl.claim_append(i->second.bl); | |
12530 | ++i; | |
12531 | } | |
224ce89b | 12532 | |
7c673cae FG |
12533 | bdev->aio_submit(&b->ioc); |
12534 | } | |
12535 | ||
3efd9988 FG |
12536 | struct C_DeferredTrySubmit : public Context { |
12537 | BlueStore *store; | |
12538 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
12539 | void finish(int r) { | |
12540 | store->deferred_try_submit(); | |
12541 | } | |
12542 | }; | |
12543 | ||
7c673cae FG |
12544 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
12545 | { | |
12546 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 12547 | ceph_assert(osr->deferred_running); |
7c673cae FG |
12548 | DeferredBatch *b = osr->deferred_running; |
12549 | ||
12550 | { | |
9f95a23c | 12551 | deferred_lock.lock(); |
11fdf7f2 | 12552 | ceph_assert(osr->deferred_running == b); |
7c673cae FG |
12553 | osr->deferred_running = nullptr; |
12554 | if (!osr->deferred_pending) { | |
181888fb | 12555 | dout(20) << __func__ << " dequeueing" << dendl; |
7c673cae FG |
12556 | auto q = deferred_queue.iterator_to(*osr); |
12557 | deferred_queue.erase(q); | |
9f95a23c | 12558 | deferred_lock.unlock(); |
181888fb | 12559 | } else { |
9f95a23c TL |
12560 | deferred_lock.unlock(); |
12561 | if (deferred_aggressive) { | |
12562 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; | |
12563 | finisher.queue(new C_DeferredTrySubmit(this)); | |
12564 | } else { | |
12565 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
12566 | } | |
7c673cae FG |
12567 | } |
12568 | } | |
12569 | ||
12570 | { | |
31f18b77 | 12571 | uint64_t costs = 0; |
11fdf7f2 | 12572 | { |
11fdf7f2 TL |
12573 | for (auto& i : b->txcs) { |
12574 | TransContext *txc = &i; | |
9f95a23c | 12575 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_aio_wait_lat); |
11fdf7f2 TL |
12576 | txc->state = TransContext::STATE_DEFERRED_CLEANUP; |
12577 | costs += txc->cost; | |
12578 | } | |
7c673cae | 12579 | } |
9f95a23c | 12580 | throttle.release_deferred_throttle(costs); |
7c673cae FG |
12581 | } |
12582 | ||
9f95a23c | 12583 | { |
11fdf7f2 | 12584 | std::lock_guard l(kv_lock); |
9f95a23c TL |
12585 | deferred_done_queue.emplace_back(b); |
12586 | ||
12587 | // in the normal case, do not bother waking up the kv thread; it will | |
12588 | // catch us on the next commit anyway. | |
12589 | if (deferred_aggressive && !kv_sync_in_progress) { | |
12590 | kv_sync_in_progress = true; | |
12591 | kv_cond.notify_one(); | |
12592 | } | |
7c673cae FG |
12593 | } |
12594 | } | |
12595 | ||
12596 | int BlueStore::_deferred_replay() | |
12597 | { | |
12598 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
12599 | int count = 0; |
12600 | int r = 0; | |
11fdf7f2 TL |
12601 | CollectionRef ch = _get_collection(coll_t::meta()); |
12602 | bool fake_ch = false; | |
12603 | if (!ch) { | |
12604 | // hmm, replaying initial mkfs? | |
12605 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
12606 | fake_ch = true; | |
12607 | } | |
12608 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
12609 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
12610 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
12611 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
12612 | << dendl; | |
12613 | bluestore_deferred_transaction_t *deferred_txn = | |
12614 | new bluestore_deferred_transaction_t; | |
12615 | bufferlist bl = it->value(); | |
11fdf7f2 | 12616 | auto p = bl.cbegin(); |
7c673cae | 12617 | try { |
11fdf7f2 | 12618 | decode(*deferred_txn, p); |
7c673cae FG |
12619 | } catch (buffer::error& e) { |
12620 | derr << __func__ << " failed to decode deferred txn " | |
12621 | << pretty_binary_string(it->key()) << dendl; | |
12622 | delete deferred_txn; | |
12623 | r = -EIO; | |
12624 | goto out; | |
12625 | } | |
11fdf7f2 | 12626 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); |
7c673cae FG |
12627 | txc->deferred_txn = deferred_txn; |
12628 | txc->state = TransContext::STATE_KV_DONE; | |
12629 | _txc_state_proc(txc); | |
12630 | } | |
12631 | out: | |
12632 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 12633 | _osr_register_zombie(osr); |
7c673cae | 12634 | _osr_drain_all(); |
11fdf7f2 TL |
12635 | if (fake_ch) { |
12636 | new_coll_map.clear(); | |
12637 | } | |
7c673cae FG |
12638 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
12639 | return r; | |
12640 | } | |
12641 | ||
12642 | // --------------------------- | |
12643 | // transactions | |
12644 | ||
12645 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
12646 | CollectionHandle& ch, |
12647 | vector<Transaction>& tls, | |
12648 | TrackedOpRef op, | |
12649 | ThreadPool::TPHandle *handle) | |
12650 | { | |
12651 | FUNCTRACE(cct); | |
12652 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 12653 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 12654 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae | 12655 | |
11fdf7f2 TL |
12656 | auto start = mono_clock::now(); |
12657 | ||
12658 | Collection *c = static_cast<Collection*>(ch.get()); | |
12659 | OpSequencer *osr = c->osr.get(); | |
12660 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae FG |
12661 | |
12662 | // prepare | |
11fdf7f2 TL |
12663 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, |
12664 | &on_commit); | |
7c673cae FG |
12665 | |
12666 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { | |
7c673cae FG |
12667 | txc->bytes += (*p).get_num_bytes(); |
12668 | _txc_add_transaction(txc, &(*p)); | |
12669 | } | |
12670 | _txc_calc_cost(txc); | |
12671 | ||
12672 | _txc_write_nodes(txc, txc->t); | |
12673 | ||
12674 | // journal deferred items | |
12675 | if (txc->deferred_txn) { | |
12676 | txc->deferred_txn->seq = ++deferred_seq; | |
12677 | bufferlist bl; | |
11fdf7f2 | 12678 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
12679 | string key; |
12680 | get_deferred_key(txc->deferred_txn->seq, &key); | |
12681 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
12682 | } | |
12683 | ||
12684 | _txc_finalize_kv(txc, txc->t); | |
12685 | if (handle) | |
12686 | handle->suspend_tp_timeout(); | |
12687 | ||
11fdf7f2 | 12688 | auto tstart = mono_clock::now(); |
9f95a23c TL |
12689 | |
12690 | if (!throttle.try_start_transaction( | |
12691 | *db, | |
12692 | *txc, | |
12693 | tstart)) { | |
7c673cae | 12694 | // ensure we do not block here because of deferred writes |
9f95a23c TL |
12695 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
12696 | << dendl; | |
12697 | ++deferred_aggressive; | |
12698 | deferred_try_submit(); | |
12699 | { | |
12700 | // wake up any previously finished deferred events | |
12701 | std::lock_guard l(kv_lock); | |
12702 | if (!kv_sync_in_progress) { | |
12703 | kv_sync_in_progress = true; | |
3efd9988 FG |
12704 | kv_cond.notify_one(); |
12705 | } | |
9f95a23c TL |
12706 | } |
12707 | throttle.finish_start_transaction(*db, *txc, tstart); | |
12708 | --deferred_aggressive; | |
7c673cae | 12709 | } |
11fdf7f2 | 12710 | auto tend = mono_clock::now(); |
7c673cae FG |
12711 | |
12712 | if (handle) | |
12713 | handle->reset_tp_timeout(); | |
12714 | ||
12715 | logger->inc(l_bluestore_txc); | |
12716 | ||
12717 | // execute (start) | |
12718 | _txc_state_proc(txc); | |
12719 | ||
11fdf7f2 TL |
12720 | // we're immediately readable (unlike FileStore) |
12721 | for (auto c : on_applied_sync) { | |
12722 | c->complete(0); | |
12723 | } | |
12724 | if (!on_applied.empty()) { | |
12725 | if (c->commit_queue) { | |
12726 | c->commit_queue->queue(on_applied); | |
12727 | } else { | |
12728 | finisher.queue(on_applied); | |
12729 | } | |
12730 | } | |
12731 | ||
494da23a TL |
12732 | log_latency("submit_transact", |
12733 | l_bluestore_submit_lat, | |
12734 | mono_clock::now() - start, | |
12735 | cct->_conf->bluestore_log_op_age); | |
12736 | log_latency("throttle_transact", | |
12737 | l_bluestore_throttle_lat, | |
12738 | tend - tstart, | |
12739 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
12740 | return 0; |
12741 | } | |
12742 | ||
12743 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
12744 | { | |
12745 | dout(10) << __func__ << " txc " << txc << dendl; | |
12746 | bdev->aio_submit(&txc->ioc); | |
12747 | } | |
12748 | ||
12749 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
12750 | { | |
12751 | Transaction::iterator i = t->begin(); | |
12752 | ||
81eedcae | 12753 | _dump_transaction<30>(cct, t); |
7c673cae FG |
12754 | |
12755 | vector<CollectionRef> cvec(i.colls.size()); | |
12756 | unsigned j = 0; | |
12757 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
12758 | ++p, ++j) { | |
12759 | cvec[j] = _get_collection(*p); | |
7c673cae | 12760 | } |
11fdf7f2 | 12761 | |
7c673cae FG |
12762 | vector<OnodeRef> ovec(i.objects.size()); |
12763 | ||
12764 | for (int pos = 0; i.have_op(); ++pos) { | |
12765 | Transaction::Op *op = i.decode_op(); | |
12766 | int r = 0; | |
12767 | ||
12768 | // no coll or obj | |
12769 | if (op->op == Transaction::OP_NOP) | |
12770 | continue; | |
12771 | ||
11fdf7f2 | 12772 | |
7c673cae FG |
12773 | // collection operations |
12774 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
12775 | |
12776 | // initialize osd_pool_id and do a smoke test that all collections belong | |
12777 | // to the same pool | |
12778 | spg_t pgid; | |
12779 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
12780 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
12781 | txc->osd_pool_id == pgid.pool()); | |
12782 | txc->osd_pool_id = pgid.pool(); | |
12783 | } | |
12784 | ||
7c673cae FG |
12785 | switch (op->op) { |
12786 | case Transaction::OP_RMCOLL: | |
12787 | { | |
12788 | const coll_t &cid = i.get_cid(op->cid); | |
12789 | r = _remove_collection(txc, cid, &c); | |
12790 | if (!r) | |
12791 | continue; | |
12792 | } | |
12793 | break; | |
12794 | ||
12795 | case Transaction::OP_MKCOLL: | |
12796 | { | |
11fdf7f2 | 12797 | ceph_assert(!c); |
7c673cae FG |
12798 | const coll_t &cid = i.get_cid(op->cid); |
12799 | r = _create_collection(txc, cid, op->split_bits, &c); | |
12800 | if (!r) | |
12801 | continue; | |
12802 | } | |
12803 | break; | |
12804 | ||
12805 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 12806 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12807 | break; |
12808 | ||
12809 | case Transaction::OP_SPLIT_COLLECTION2: | |
12810 | { | |
12811 | uint32_t bits = op->split_bits; | |
12812 | uint32_t rem = op->split_rem; | |
12813 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
12814 | if (!r) | |
12815 | continue; | |
12816 | } | |
12817 | break; | |
12818 | ||
11fdf7f2 TL |
12819 | case Transaction::OP_MERGE_COLLECTION: |
12820 | { | |
12821 | uint32_t bits = op->split_bits; | |
12822 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
12823 | if (!r) | |
12824 | continue; | |
12825 | } | |
12826 | break; | |
12827 | ||
7c673cae FG |
12828 | case Transaction::OP_COLL_HINT: |
12829 | { | |
12830 | uint32_t type = op->hint_type; | |
12831 | bufferlist hint; | |
12832 | i.decode_bl(hint); | |
11fdf7f2 | 12833 | auto hiter = hint.cbegin(); |
7c673cae FG |
12834 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
12835 | uint32_t pg_num; | |
12836 | uint64_t num_objs; | |
11fdf7f2 TL |
12837 | decode(pg_num, hiter); |
12838 | decode(num_objs, hiter); | |
7c673cae FG |
12839 | dout(10) << __func__ << " collection hint objects is a no-op, " |
12840 | << " pg_num " << pg_num << " num_objects " << num_objs | |
12841 | << dendl; | |
12842 | } else { | |
12843 | // Ignore the hint | |
12844 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
12845 | } | |
12846 | continue; | |
12847 | } | |
12848 | break; | |
12849 | ||
12850 | case Transaction::OP_COLL_SETATTR: | |
12851 | r = -EOPNOTSUPP; | |
12852 | break; | |
12853 | ||
12854 | case Transaction::OP_COLL_RMATTR: | |
12855 | r = -EOPNOTSUPP; | |
12856 | break; | |
12857 | ||
12858 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 12859 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12860 | break; |
12861 | } | |
12862 | if (r < 0) { | |
12863 | derr << __func__ << " error " << cpp_strerror(r) | |
12864 | << " not handled on operation " << op->op | |
12865 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 12866 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12867 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12868 | } |
12869 | ||
12870 | // these operations implicity create the object | |
12871 | bool create = false; | |
12872 | if (op->op == Transaction::OP_TOUCH || | |
9f95a23c | 12873 | op->op == Transaction::OP_CREATE || |
7c673cae FG |
12874 | op->op == Transaction::OP_WRITE || |
12875 | op->op == Transaction::OP_ZERO) { | |
12876 | create = true; | |
12877 | } | |
12878 | ||
12879 | // object operations | |
9f95a23c | 12880 | std::unique_lock l(c->lock); |
7c673cae FG |
12881 | OnodeRef &o = ovec[op->oid]; |
12882 | if (!o) { | |
12883 | ghobject_t oid = i.get_oid(op->oid); | |
9f95a23c | 12884 | o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE); |
7c673cae FG |
12885 | } |
12886 | if (!create && (!o || !o->exists)) { | |
12887 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
12888 | << i.get_oid(op->oid) << dendl; | |
12889 | r = -ENOENT; | |
12890 | goto endop; | |
12891 | } | |
12892 | ||
12893 | switch (op->op) { | |
9f95a23c | 12894 | case Transaction::OP_CREATE: |
7c673cae FG |
12895 | case Transaction::OP_TOUCH: |
12896 | r = _touch(txc, c, o); | |
12897 | break; | |
12898 | ||
12899 | case Transaction::OP_WRITE: | |
12900 | { | |
12901 | uint64_t off = op->off; | |
12902 | uint64_t len = op->len; | |
12903 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
12904 | bufferlist bl; | |
12905 | i.decode_bl(bl); | |
12906 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
12907 | } | |
12908 | break; | |
12909 | ||
12910 | case Transaction::OP_ZERO: | |
12911 | { | |
12912 | uint64_t off = op->off; | |
12913 | uint64_t len = op->len; | |
12914 | r = _zero(txc, c, o, off, len); | |
12915 | } | |
12916 | break; | |
12917 | ||
12918 | case Transaction::OP_TRIMCACHE: | |
12919 | { | |
12920 | // deprecated, no-op | |
12921 | } | |
12922 | break; | |
12923 | ||
12924 | case Transaction::OP_TRUNCATE: | |
12925 | { | |
12926 | uint64_t off = op->off; | |
35e4c445 | 12927 | r = _truncate(txc, c, o, off); |
7c673cae FG |
12928 | } |
12929 | break; | |
12930 | ||
12931 | case Transaction::OP_REMOVE: | |
12932 | { | |
12933 | r = _remove(txc, c, o); | |
12934 | } | |
12935 | break; | |
12936 | ||
12937 | case Transaction::OP_SETATTR: | |
12938 | { | |
12939 | string name = i.decode_string(); | |
12940 | bufferptr bp; | |
12941 | i.decode_bp(bp); | |
12942 | r = _setattr(txc, c, o, name, bp); | |
12943 | } | |
12944 | break; | |
12945 | ||
12946 | case Transaction::OP_SETATTRS: | |
12947 | { | |
12948 | map<string, bufferptr> aset; | |
12949 | i.decode_attrset(aset); | |
12950 | r = _setattrs(txc, c, o, aset); | |
12951 | } | |
12952 | break; | |
12953 | ||
12954 | case Transaction::OP_RMATTR: | |
12955 | { | |
12956 | string name = i.decode_string(); | |
12957 | r = _rmattr(txc, c, o, name); | |
12958 | } | |
12959 | break; | |
12960 | ||
12961 | case Transaction::OP_RMATTRS: | |
12962 | { | |
12963 | r = _rmattrs(txc, c, o); | |
12964 | } | |
12965 | break; | |
12966 | ||
12967 | case Transaction::OP_CLONE: | |
12968 | { | |
12969 | OnodeRef& no = ovec[op->dest_oid]; | |
12970 | if (!no) { | |
12971 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12972 | no = c->get_onode(noid, true); | |
12973 | } | |
12974 | r = _clone(txc, c, o, no); | |
12975 | } | |
12976 | break; | |
12977 | ||
12978 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 12979 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12980 | break; |
12981 | ||
12982 | case Transaction::OP_CLONERANGE2: | |
12983 | { | |
12984 | OnodeRef& no = ovec[op->dest_oid]; | |
12985 | if (!no) { | |
12986 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12987 | no = c->get_onode(noid, true); | |
12988 | } | |
12989 | uint64_t srcoff = op->off; | |
12990 | uint64_t len = op->len; | |
12991 | uint64_t dstoff = op->dest_off; | |
12992 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
12993 | } | |
12994 | break; | |
12995 | ||
12996 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 12997 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12998 | break; |
12999 | ||
13000 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 13001 | ceph_abort_msg("not implemented"); |
7c673cae FG |
13002 | break; |
13003 | ||
13004 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 13005 | ceph_abort_msg("deprecated"); |
7c673cae FG |
13006 | break; |
13007 | ||
13008 | case Transaction::OP_COLL_MOVE_RENAME: | |
13009 | case Transaction::OP_TRY_RENAME: | |
13010 | { | |
11fdf7f2 | 13011 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
13012 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
13013 | OnodeRef& no = ovec[op->dest_oid]; | |
13014 | if (!no) { | |
13015 | no = c->get_onode(noid, false); | |
13016 | } | |
13017 | r = _rename(txc, c, o, no, noid); | |
13018 | } | |
13019 | break; | |
13020 | ||
13021 | case Transaction::OP_OMAP_CLEAR: | |
13022 | { | |
13023 | r = _omap_clear(txc, c, o); | |
13024 | } | |
13025 | break; | |
13026 | case Transaction::OP_OMAP_SETKEYS: | |
13027 | { | |
13028 | bufferlist aset_bl; | |
13029 | i.decode_attrset_bl(&aset_bl); | |
13030 | r = _omap_setkeys(txc, c, o, aset_bl); | |
13031 | } | |
13032 | break; | |
13033 | case Transaction::OP_OMAP_RMKEYS: | |
13034 | { | |
13035 | bufferlist keys_bl; | |
13036 | i.decode_keyset_bl(&keys_bl); | |
13037 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
13038 | } | |
13039 | break; | |
13040 | case Transaction::OP_OMAP_RMKEYRANGE: | |
13041 | { | |
13042 | string first, last; | |
13043 | first = i.decode_string(); | |
13044 | last = i.decode_string(); | |
13045 | r = _omap_rmkey_range(txc, c, o, first, last); | |
13046 | } | |
13047 | break; | |
13048 | case Transaction::OP_OMAP_SETHEADER: | |
13049 | { | |
13050 | bufferlist bl; | |
13051 | i.decode_bl(bl); | |
13052 | r = _omap_setheader(txc, c, o, bl); | |
13053 | } | |
13054 | break; | |
13055 | ||
13056 | case Transaction::OP_SETALLOCHINT: | |
13057 | { | |
13058 | r = _set_alloc_hint(txc, c, o, | |
13059 | op->expected_object_size, | |
13060 | op->expected_write_size, | |
13061 | op->alloc_hint_flags); | |
13062 | } | |
13063 | break; | |
13064 | ||
13065 | default: | |
11fdf7f2 | 13066 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
13067 | ceph_abort(); |
13068 | } | |
13069 | ||
13070 | endop: | |
13071 | if (r < 0) { | |
13072 | bool ok = false; | |
13073 | ||
13074 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
13075 | op->op == Transaction::OP_CLONE || | |
13076 | op->op == Transaction::OP_CLONERANGE2 || | |
13077 | op->op == Transaction::OP_COLL_ADD || | |
13078 | op->op == Transaction::OP_SETATTR || | |
13079 | op->op == Transaction::OP_SETATTRS || | |
13080 | op->op == Transaction::OP_RMATTR || | |
13081 | op->op == Transaction::OP_OMAP_SETKEYS || | |
13082 | op->op == Transaction::OP_OMAP_RMKEYS || | |
13083 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
13084 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
13085 | // -ENOENT is usually okay | |
13086 | ok = true; | |
13087 | if (r == -ENODATA) | |
13088 | ok = true; | |
13089 | ||
13090 | if (!ok) { | |
13091 | const char *msg = "unexpected error code"; | |
13092 | ||
13093 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
13094 | op->op == Transaction::OP_CLONE || | |
13095 | op->op == Transaction::OP_CLONERANGE2)) | |
13096 | msg = "ENOENT on clone suggests osd bug"; | |
13097 | ||
13098 | if (r == -ENOSPC) | |
13099 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
13100 | // by partially applying transactions. | |
13101 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
13102 | ||
13103 | if (r == -ENOTEMPTY) { | |
13104 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
13105 | } | |
13106 | ||
13107 | derr << __func__ << " error " << cpp_strerror(r) | |
13108 | << " not handled on operation " << op->op | |
13109 | << " (op " << pos << ", counting from 0)" | |
13110 | << dendl; | |
13111 | derr << msg << dendl; | |
81eedcae | 13112 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 13113 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
13114 | } |
13115 | } | |
13116 | } | |
13117 | } | |
13118 | ||
13119 | ||
13120 | ||
13121 | // ----------------- | |
13122 | // write operations | |
13123 | ||
13124 | int BlueStore::_touch(TransContext *txc, | |
13125 | CollectionRef& c, | |
13126 | OnodeRef &o) | |
13127 | { | |
13128 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
13129 | int r = 0; | |
7c673cae FG |
13130 | _assign_nid(txc, o); |
13131 | txc->write_onode(o); | |
13132 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
13133 | return r; | |
13134 | } | |
13135 | ||
7c673cae FG |
13136 | void BlueStore::_pad_zeros( |
13137 | bufferlist *bl, uint64_t *offset, | |
13138 | uint64_t chunk_size) | |
13139 | { | |
13140 | auto length = bl->length(); | |
13141 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
13142 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
13143 | dout(40) << "before:\n"; | |
13144 | bl->hexdump(*_dout); | |
13145 | *_dout << dendl; | |
13146 | // front | |
13147 | size_t front_pad = *offset % chunk_size; | |
13148 | size_t back_pad = 0; | |
13149 | size_t pad_count = 0; | |
13150 | if (front_pad) { | |
11fdf7f2 TL |
13151 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
13152 | bufferptr z = buffer::create_small_page_aligned(chunk_size); | |
224ce89b | 13153 | z.zero(0, front_pad, false); |
7c673cae | 13154 | pad_count += front_pad; |
9f95a23c | 13155 | bl->begin().copy(front_copy, z.c_str() + front_pad); |
7c673cae FG |
13156 | if (front_copy + front_pad < chunk_size) { |
13157 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 13158 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
13159 | pad_count += back_pad; |
13160 | } | |
13161 | bufferlist old, t; | |
13162 | old.swap(*bl); | |
13163 | t.substr_of(old, front_copy, length - front_copy); | |
13164 | bl->append(z); | |
13165 | bl->claim_append(t); | |
13166 | *offset -= front_pad; | |
224ce89b | 13167 | length += pad_count; |
7c673cae FG |
13168 | } |
13169 | ||
13170 | // back | |
13171 | uint64_t end = *offset + length; | |
13172 | unsigned back_copy = end % chunk_size; | |
13173 | if (back_copy) { | |
11fdf7f2 | 13174 | ceph_assert(back_pad == 0); |
7c673cae | 13175 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 13176 | ceph_assert(back_copy <= length); |
7c673cae | 13177 | bufferptr tail(chunk_size); |
9f95a23c | 13178 | bl->begin(length - back_copy).copy(back_copy, tail.c_str()); |
224ce89b | 13179 | tail.zero(back_copy, back_pad, false); |
7c673cae FG |
13180 | bufferlist old; |
13181 | old.swap(*bl); | |
13182 | bl->substr_of(old, 0, length - back_copy); | |
13183 | bl->append(tail); | |
13184 | length += back_pad; | |
13185 | pad_count += back_pad; | |
13186 | } | |
13187 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
13188 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
13189 | << length << std::dec << dendl; | |
13190 | dout(40) << "after:\n"; | |
13191 | bl->hexdump(*_dout); | |
13192 | *_dout << dendl; | |
13193 | if (pad_count) | |
13194 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 13195 | ceph_assert(bl->length() == length); |
7c673cae FG |
13196 | } |
13197 | ||
13198 | void BlueStore::_do_write_small( | |
13199 | TransContext *txc, | |
13200 | CollectionRef &c, | |
13201 | OnodeRef o, | |
13202 | uint64_t offset, uint64_t length, | |
13203 | bufferlist::iterator& blp, | |
13204 | WriteContext *wctx) | |
13205 | { | |
13206 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13207 | << std::dec << dendl; | |
11fdf7f2 | 13208 | ceph_assert(length < min_alloc_size); |
7c673cae FG |
13209 | uint64_t end_offs = offset + length; |
13210 | ||
13211 | logger->inc(l_bluestore_write_small); | |
13212 | logger->inc(l_bluestore_write_small_bytes, length); | |
13213 | ||
13214 | bufferlist bl; | |
13215 | blp.copy(length, bl); | |
13216 | ||
81eedcae TL |
13217 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
13218 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
13219 | uint32_t alloc_len = min_alloc_size; | |
13220 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
13221 | ||
13222 | bool any_change; | |
13223 | ||
13224 | // search suitable extent in both forward and reverse direction in | |
13225 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
13226 | // then check if blob can be reused via can_reuse_blob func or apply | |
13227 | // direct/deferred write (the latter for extents including or higher | |
13228 | // than 'offset' only). | |
13229 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
13230 | ||
7c673cae FG |
13231 | // Look for an existing mutable blob we can use. |
13232 | auto begin = o->extent_map.extent_map.begin(); | |
13233 | auto end = o->extent_map.extent_map.end(); | |
13234 | auto ep = o->extent_map.seek_lextent(offset); | |
13235 | if (ep != begin) { | |
13236 | --ep; | |
13237 | if (ep->blob_end() <= offset) { | |
13238 | ++ep; | |
13239 | } | |
13240 | } | |
13241 | auto prev_ep = ep; | |
13242 | if (prev_ep != begin) { | |
13243 | --prev_ep; | |
13244 | } else { | |
13245 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
13246 | } | |
13247 | ||
eafe8130 TL |
13248 | boost::container::flat_set<const bluestore_blob_t*> inspected_blobs; |
13249 | // We don't want to have more blobs than min alloc units fit | |
13250 | // into 2 max blobs | |
13251 | size_t blob_threshold = max_blob_size / min_alloc_size * 2 + 1; | |
13252 | bool above_blob_threshold = false; | |
13253 | ||
13254 | inspected_blobs.reserve(blob_threshold); | |
13255 | ||
13256 | uint64_t max_off = 0; | |
13257 | auto start_ep = ep; | |
13258 | auto end_ep = ep; // exclusively | |
7c673cae FG |
13259 | do { |
13260 | any_change = false; | |
13261 | ||
13262 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13263 | BlobRef b = ep->blob; | |
eafe8130 TL |
13264 | if (!above_blob_threshold) { |
13265 | inspected_blobs.insert(&b->get_blob()); | |
13266 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13267 | } | |
13268 | max_off = ep->logical_end(); | |
7c673cae | 13269 | auto bstart = ep->blob_start(); |
eafe8130 | 13270 | |
7c673cae FG |
13271 | dout(20) << __func__ << " considering " << *b |
13272 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
13273 | if (bstart >= end_offs) { | |
13274 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
13275 | } else if (!b->get_blob().is_mutable()) { | |
13276 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
13277 | } else if (ep->logical_offset % min_alloc_size != | |
13278 | ep->blob_offset % min_alloc_size) { | |
13279 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
13280 | } else { | |
13281 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13282 | // can we pad our head/tail out with zeros? | |
13283 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
13284 | head_pad = p2phase(offset, chunk_size); |
13285 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
13286 | if (head_pad || tail_pad) { |
13287 | o->extent_map.fault_range(db, offset - head_pad, | |
13288 | end_offs - offset + head_pad + tail_pad); | |
13289 | } | |
13290 | if (head_pad && | |
13291 | o->extent_map.has_any_lextents(offset - head_pad, chunk_size)) { | |
13292 | head_pad = 0; | |
13293 | } | |
13294 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
13295 | tail_pad = 0; | |
13296 | } | |
13297 | ||
13298 | uint64_t b_off = offset - head_pad - bstart; | |
13299 | uint64_t b_len = length + head_pad + tail_pad; | |
13300 | ||
13301 | // direct write into unused blocks of an existing mutable blob? | |
13302 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
13303 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13304 | b->get_blob().is_unused(b_off, b_len) && | |
13305 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 13306 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13307 | |
13308 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
13309 | << b_off << "~" << b_len | |
13310 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
13311 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 13312 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13313 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13314 | ||
11fdf7f2 | 13315 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13316 | if (b_len <= prefer_deferred_size) { |
13317 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13318 | << b_len << std::dec << " unused write via deferred" << dendl; | |
9f95a23c | 13319 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13320 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13321 | b->get_blob().map( | |
13322 | b_off, b_len, | |
13323 | [&](uint64_t offset, uint64_t length) { | |
13324 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13325 | return 0; | |
13326 | }); | |
224ce89b | 13327 | op->data = bl; |
7c673cae FG |
13328 | } else { |
13329 | b->get_blob().map_bl( | |
224ce89b | 13330 | b_off, bl, |
7c673cae FG |
13331 | [&](uint64_t offset, bufferlist& t) { |
13332 | bdev->aio_write(offset, t, | |
13333 | &txc->ioc, wctx->buffered); | |
13334 | }); | |
13335 | } | |
13336 | } | |
224ce89b | 13337 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
13338 | dout(20) << __func__ << " lex old " << *ep << dendl; |
13339 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
13340 | b, | |
13341 | &wctx->old_extents); | |
13342 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13343 | txc->statfs_delta.stored() += le->length; | |
13344 | dout(20) << __func__ << " lex " << *le << dendl; | |
13345 | logger->inc(l_bluestore_write_small_unused); | |
13346 | return; | |
13347 | } | |
13348 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
13349 | uint64_t head_read = p2phase(b_off, chunk_size); |
13350 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
13351 | if ((head_read || tail_read) && |
13352 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
13353 | head_read + tail_read < min_alloc_size) { | |
13354 | b_off -= head_read; | |
13355 | b_len += head_read + tail_read; | |
13356 | ||
13357 | } else { | |
13358 | head_read = tail_read = 0; | |
13359 | } | |
13360 | ||
13361 | // chunk-aligned deferred overwrite? | |
13362 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13363 | b_off % chunk_size == 0 && | |
13364 | b_len % chunk_size == 0 && | |
13365 | b->get_blob().is_allocated(b_off, b_len)) { | |
13366 | ||
224ce89b | 13367 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13368 | |
13369 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
13370 | << " and tail 0x" << tail_read << std::dec << dendl; | |
13371 | if (head_read) { | |
13372 | bufferlist head_bl; | |
13373 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
13374 | head_bl, 0); | |
11fdf7f2 | 13375 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
13376 | size_t zlen = head_read - r; |
13377 | if (zlen) { | |
13378 | head_bl.append_zero(zlen); | |
13379 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13380 | } | |
11fdf7f2 TL |
13381 | head_bl.claim_append(bl); |
13382 | bl.swap(head_bl); | |
7c673cae FG |
13383 | logger->inc(l_bluestore_write_penalty_read_ops); |
13384 | } | |
13385 | if (tail_read) { | |
13386 | bufferlist tail_bl; | |
13387 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
13388 | tail_bl, 0); | |
11fdf7f2 | 13389 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
13390 | size_t zlen = tail_read - r; |
13391 | if (zlen) { | |
13392 | tail_bl.append_zero(zlen); | |
13393 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13394 | } | |
224ce89b | 13395 | bl.claim_append(tail_bl); |
7c673cae FG |
13396 | logger->inc(l_bluestore_write_penalty_read_ops); |
13397 | } | |
13398 | logger->inc(l_bluestore_write_small_pre_read); | |
13399 | ||
224ce89b | 13400 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13401 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13402 | ||
7c673cae | 13403 | if (b->get_blob().csum_type) { |
224ce89b | 13404 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae | 13405 | } |
11fdf7f2 TL |
13406 | |
13407 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
9f95a23c | 13408 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
11fdf7f2 TL |
13409 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13410 | int r = b->get_blob().map( | |
13411 | b_off, b_len, | |
13412 | [&](uint64_t offset, uint64_t length) { | |
13413 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13414 | return 0; | |
13415 | }); | |
13416 | ceph_assert(r == 0); | |
13417 | op->data.claim(bl); | |
13418 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" | |
13419 | << b_len << std::dec << " of mutable " << *b | |
13420 | << " at " << op->extents << dendl; | |
13421 | } | |
13422 | ||
7c673cae FG |
13423 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
13424 | b, &wctx->old_extents); | |
13425 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13426 | txc->statfs_delta.stored() += le->length; | |
13427 | dout(20) << __func__ << " lex " << *le << dendl; | |
13428 | logger->inc(l_bluestore_write_small_deferred); | |
13429 | return; | |
13430 | } | |
224ce89b WB |
13431 | // try to reuse blob if we can |
13432 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
13433 | max_bsize, |
13434 | offset0 - bstart, | |
13435 | &alloc_len)) { | |
11fdf7f2 | 13436 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13437 | // fit into reused blob |
13438 | // Need to check for pending writes desiring to | |
13439 | // reuse the same pextent. The rationale is that during GC two chunks | |
13440 | // from garbage blobs(compressed?) can share logical space within the same | |
13441 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13442 | // Hence the second write will fail in an attempt to reuse blob at | |
13443 | // do_alloc_write(). | |
13444 | if (!wctx->has_conflict(b, | |
13445 | offset0, | |
13446 | offset0 + alloc_len, | |
13447 | min_alloc_size)) { | |
13448 | ||
13449 | // we can't reuse pad_head/pad_tail since they might be truncated | |
13450 | // due to existent extents | |
13451 | uint64_t b_off = offset - bstart; | |
13452 | uint64_t b_off0 = b_off; | |
13453 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13454 | ||
13455 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13456 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13457 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13458 | << std::dec << dendl; |
13459 | ||
13460 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13461 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13462 | false, false); | |
13463 | logger->inc(l_bluestore_write_small_unused); | |
13464 | return; | |
13465 | } | |
13466 | } | |
13467 | } | |
13468 | ++ep; | |
eafe8130 | 13469 | end_ep = ep; |
7c673cae FG |
13470 | any_change = true; |
13471 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
13472 | ||
13473 | // check extent for reuse in reverse order | |
13474 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
13475 | BlobRef b = prev_ep->blob; | |
eafe8130 TL |
13476 | if (!above_blob_threshold) { |
13477 | inspected_blobs.insert(&b->get_blob()); | |
13478 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13479 | } | |
13480 | start_ep = prev_ep; | |
7c673cae FG |
13481 | auto bstart = prev_ep->blob_start(); |
13482 | dout(20) << __func__ << " considering " << *b | |
13483 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 13484 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
13485 | max_bsize, |
13486 | offset0 - bstart, | |
13487 | &alloc_len)) { | |
11fdf7f2 | 13488 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13489 | // fit into reused blob |
13490 | // Need to check for pending writes desiring to | |
13491 | // reuse the same pextent. The rationale is that during GC two chunks | |
13492 | // from garbage blobs(compressed?) can share logical space within the same | |
13493 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13494 | // Hence the second write will fail in an attempt to reuse blob at | |
13495 | // do_alloc_write(). | |
13496 | if (!wctx->has_conflict(b, | |
13497 | offset0, | |
13498 | offset0 + alloc_len, | |
13499 | min_alloc_size)) { | |
13500 | ||
13501 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13502 | uint64_t b_off = offset - bstart; | |
13503 | uint64_t b_off0 = b_off; | |
13504 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13505 | ||
13506 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13507 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13508 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13509 | << std::dec << dendl; |
13510 | ||
13511 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13512 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13513 | false, false); | |
13514 | logger->inc(l_bluestore_write_small_unused); | |
13515 | return; | |
13516 | } | |
13517 | } | |
13518 | if (prev_ep != begin) { | |
13519 | --prev_ep; | |
13520 | any_change = true; | |
13521 | } else { | |
13522 | prev_ep = end; // to avoid useless first extent re-check | |
13523 | } | |
13524 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
13525 | } while (any_change); | |
13526 | ||
eafe8130 TL |
13527 | if (above_blob_threshold) { |
13528 | dout(10) << __func__ << " request GC, blobs >= " << inspected_blobs.size() | |
13529 | << " " << std::hex << min_off << "~" << max_off << std::dec | |
13530 | << dendl; | |
13531 | ceph_assert(start_ep != end_ep); | |
13532 | for (auto ep = start_ep; ep != end_ep; ++ep) { | |
13533 | dout(20) << __func__ << " inserting for GC " | |
13534 | << std::hex << ep->logical_offset << "~" << ep->length | |
13535 | << std::dec << dendl; | |
13536 | ||
13537 | wctx->extents_to_gc.union_insert(ep->logical_offset, ep->length); | |
13538 | } | |
13539 | // insert newly written extent to GC | |
13540 | wctx->extents_to_gc.union_insert(offset, length); | |
13541 | dout(20) << __func__ << " inserting (last) for GC " | |
13542 | << std::hex << offset << "~" << length | |
13543 | << std::dec << dendl; | |
13544 | } | |
7c673cae | 13545 | // new blob. |
7c673cae | 13546 | BlobRef b = c->new_blob(); |
11fdf7f2 | 13547 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae FG |
13548 | uint64_t b_off0 = b_off; |
13549 | _pad_zeros(&bl, &b_off0, block_size); | |
13550 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
1911f103 TL |
13551 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, |
13552 | min_alloc_size != block_size, // use 'unused' bitmap when alloc granularity | |
13553 | // doesn't match disk one only | |
13554 | true); | |
7c673cae FG |
13555 | |
13556 | return; | |
13557 | } | |
13558 | ||
13559 | void BlueStore::_do_write_big( | |
13560 | TransContext *txc, | |
13561 | CollectionRef &c, | |
13562 | OnodeRef o, | |
13563 | uint64_t offset, uint64_t length, | |
13564 | bufferlist::iterator& blp, | |
13565 | WriteContext *wctx) | |
13566 | { | |
13567 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13568 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
13569 | << " compress " << (int)wctx->compress | |
13570 | << dendl; | |
13571 | logger->inc(l_bluestore_write_big); | |
13572 | logger->inc(l_bluestore_write_big_bytes, length); | |
13573 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11fdf7f2 | 13574 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae FG |
13575 | while (length > 0) { |
13576 | bool new_blob = false; | |
11fdf7f2 | 13577 | uint32_t l = std::min(max_bsize, length); |
7c673cae FG |
13578 | BlobRef b; |
13579 | uint32_t b_off = 0; | |
13580 | ||
13581 | //attempting to reuse existing blob | |
13582 | if (!wctx->compress) { | |
13583 | // look for an existing mutable blob we can reuse | |
13584 | auto begin = o->extent_map.extent_map.begin(); | |
13585 | auto end = o->extent_map.extent_map.end(); | |
13586 | auto ep = o->extent_map.seek_lextent(offset); | |
13587 | auto prev_ep = ep; | |
13588 | if (prev_ep != begin) { | |
13589 | --prev_ep; | |
13590 | } else { | |
13591 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
13592 | } | |
13593 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
13594 | // search suitable extent in both forward and reverse direction in | |
13595 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 13596 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
13597 | bool any_change; |
13598 | do { | |
13599 | any_change = false; | |
13600 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13601 | if (offset >= ep->blob_start() && | |
224ce89b | 13602 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13603 | offset - ep->blob_start(), |
13604 | &l)) { | |
13605 | b = ep->blob; | |
13606 | b_off = offset - ep->blob_start(); | |
13607 | prev_ep = end; // to avoid check below | |
13608 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13609 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13610 | } else { |
13611 | ++ep; | |
13612 | any_change = true; | |
13613 | } | |
13614 | } | |
13615 | ||
13616 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
224ce89b | 13617 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13618 | offset - prev_ep->blob_start(), |
13619 | &l)) { | |
13620 | b = prev_ep->blob; | |
13621 | b_off = offset - prev_ep->blob_start(); | |
13622 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13623 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13624 | } else if (prev_ep != begin) { |
13625 | --prev_ep; | |
13626 | any_change = true; | |
13627 | } else { | |
13628 | prev_ep = end; // to avoid useless first extent re-check | |
13629 | } | |
13630 | } | |
13631 | } while (b == nullptr && any_change); | |
13632 | } | |
13633 | if (b == nullptr) { | |
13634 | b = c->new_blob(); | |
13635 | b_off = 0; | |
13636 | new_blob = true; | |
13637 | } | |
13638 | ||
13639 | bufferlist t; | |
13640 | blp.copy(l, t); | |
13641 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); | |
13642 | offset += l; | |
13643 | length -= l; | |
13644 | logger->inc(l_bluestore_write_big_blobs); | |
13645 | } | |
13646 | } | |
13647 | ||
13648 | int BlueStore::_do_alloc_write( | |
13649 | TransContext *txc, | |
13650 | CollectionRef coll, | |
13651 | OnodeRef o, | |
13652 | WriteContext *wctx) | |
13653 | { | |
13654 | dout(20) << __func__ << " txc " << txc | |
13655 | << " " << wctx->writes.size() << " blobs" | |
13656 | << dendl; | |
3efd9988 FG |
13657 | if (wctx->writes.empty()) { |
13658 | return 0; | |
7c673cae FG |
13659 | } |
13660 | ||
7c673cae FG |
13661 | CompressorRef c; |
13662 | double crr = 0; | |
13663 | if (wctx->compress) { | |
13664 | c = select_option( | |
13665 | "compression_algorithm", | |
13666 | compressor, | |
13667 | [&]() { | |
13668 | string val; | |
13669 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
13670 | CompressorRef cp = compressor; | |
13671 | if (!cp || cp->get_type_name() != val) { | |
13672 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
13673 | if (!cp) { |
13674 | if (_set_compression_alert(false, val.c_str())) { | |
13675 | derr << __func__ << " unable to initialize " << val.c_str() | |
13676 | << " compressor" << dendl; | |
13677 | } | |
13678 | } | |
7c673cae FG |
13679 | } |
13680 | return boost::optional<CompressorRef>(cp); | |
13681 | } | |
13682 | return boost::optional<CompressorRef>(); | |
13683 | } | |
13684 | ); | |
13685 | ||
13686 | crr = select_option( | |
13687 | "compression_required_ratio", | |
13688 | cct->_conf->bluestore_compression_required_ratio, | |
13689 | [&]() { | |
13690 | double val; | |
3efd9988 | 13691 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
7c673cae FG |
13692 | return boost::optional<double>(val); |
13693 | } | |
13694 | return boost::optional<double>(); | |
13695 | } | |
13696 | ); | |
13697 | } | |
13698 | ||
13699 | // checksum | |
11fdf7f2 | 13700 | int64_t csum = csum_type.load(); |
7c673cae FG |
13701 | csum = select_option( |
13702 | "csum_type", | |
13703 | csum, | |
13704 | [&]() { | |
11fdf7f2 | 13705 | int64_t val; |
3efd9988 | 13706 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
11fdf7f2 | 13707 | return boost::optional<int64_t>(val); |
7c673cae | 13708 | } |
11fdf7f2 | 13709 | return boost::optional<int64_t>(); |
7c673cae FG |
13710 | } |
13711 | ); | |
13712 | ||
3efd9988 FG |
13713 | // compress (as needed) and calc needed space |
13714 | uint64_t need = 0; | |
11fdf7f2 | 13715 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 13716 | for (auto& wi : wctx->writes) { |
3efd9988 | 13717 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 13718 | auto start = mono_clock::now(); |
7c673cae FG |
13719 | |
13720 | // compress | |
11fdf7f2 TL |
13721 | ceph_assert(wi.b_off == 0); |
13722 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 13723 | |
7c673cae FG |
13724 | // FIXME: memory alignment here is bad |
13725 | bufferlist t; | |
3efd9988 | 13726 | int r = c->compress(wi.bl, t); |
3efd9988 | 13727 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 13728 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
13729 | bool rejected = false; |
13730 | uint64_t compressed_len = t.length(); | |
13731 | // do an approximate (fast) estimation for resulting blob size | |
13732 | // that doesn't take header overhead into account | |
11fdf7f2 | 13733 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13734 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
13735 | bluestore_compression_header_t chdr; | |
13736 | chdr.type = c->get_type(); | |
13737 | chdr.length = t.length(); | |
13738 | encode(chdr, wi.compressed_bl); | |
13739 | wi.compressed_bl.claim_append(t); | |
13740 | ||
13741 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 13742 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13743 | if (result_len <= want_len && result_len < wi.blob_length) { |
13744 | // Cool. We compressed at least as much as we were hoping to. | |
13745 | // pad out to min_alloc_size | |
13746 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
13747 | wi.compressed_len = compressed_len; | |
13748 | wi.compressed = true; | |
13749 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
13750 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
13751 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
13752 | << " with " << c->get_type() | |
13753 | << std::dec << dendl; | |
13754 | txc->statfs_delta.compressed() += compressed_len; | |
13755 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
13756 | txc->statfs_delta.compressed_allocated() += result_len; | |
13757 | logger->inc(l_bluestore_compress_success_count); | |
13758 | need += result_len; | |
13759 | } else { | |
13760 | rejected = true; | |
13761 | } | |
13762 | } else if (r != 0) { | |
13763 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
13764 | << " bytes compressed using " << c->get_type_name() | |
13765 | << std::dec | |
13766 | << " failed with errcode = " << r | |
13767 | << ", leaving uncompressed" | |
13768 | << dendl; | |
13769 | logger->inc(l_bluestore_compress_rejected_count); | |
13770 | need += wi.blob_length; | |
7c673cae | 13771 | } else { |
a8e16298 TL |
13772 | rejected = true; |
13773 | } | |
13774 | ||
13775 | if (rejected) { | |
3efd9988 | 13776 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 13777 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
13778 | << " with " << c->get_type() |
13779 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 13780 | << " -> 0x" << want_len |
3efd9988 FG |
13781 | << ", leaving uncompressed" |
13782 | << std::dec << dendl; | |
13783 | logger->inc(l_bluestore_compress_rejected_count); | |
13784 | need += wi.blob_length; | |
7c673cae | 13785 | } |
494da23a TL |
13786 | log_latency("compress@_do_alloc_write", |
13787 | l_bluestore_compress_lat, | |
13788 | mono_clock::now() - start, | |
13789 | cct->_conf->bluestore_log_op_age ); | |
3efd9988 FG |
13790 | } else { |
13791 | need += wi.blob_length; | |
7c673cae | 13792 | } |
3efd9988 | 13793 | } |
a8e16298 | 13794 | PExtentVector prealloc; |
3efd9988 | 13795 | prealloc.reserve(2 * wctx->writes.size());; |
11fdf7f2 | 13796 | int64_t prealloc_left = 0; |
3efd9988 FG |
13797 | prealloc_left = alloc->allocate( |
13798 | need, min_alloc_size, need, | |
13799 | 0, &prealloc); | |
eafe8130 | 13800 | if (prealloc_left < 0 || prealloc_left < (int64_t)need) { |
11fdf7f2 | 13801 | derr << __func__ << " failed to allocate 0x" << std::hex << need |
eafe8130 | 13802 | << " allocated 0x " << (prealloc_left < 0 ? 0 : prealloc_left) |
11fdf7f2 TL |
13803 | << " min_alloc_size 0x" << min_alloc_size |
13804 | << " available 0x " << alloc->get_free() | |
13805 | << std::dec << dendl; | |
13806 | if (prealloc.size()) { | |
13807 | alloc->release(prealloc); | |
13808 | } | |
a8e16298 TL |
13809 | return -ENOSPC; |
13810 | } | |
9f95a23c | 13811 | _collect_allocation_stats(need, min_alloc_size, prealloc.size()); |
a8e16298 | 13812 | |
3efd9988 FG |
13813 | dout(20) << __func__ << " prealloc " << prealloc << dendl; |
13814 | auto prealloc_pos = prealloc.begin(); | |
13815 | ||
13816 | for (auto& wi : wctx->writes) { | |
13817 | BlobRef b = wi.b; | |
13818 | bluestore_blob_t& dblob = b->dirty_blob(); | |
13819 | uint64_t b_off = wi.b_off; | |
13820 | bufferlist *l = &wi.bl; | |
13821 | uint64_t final_length = wi.blob_length; | |
13822 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
13823 | if (wi.compressed) { |
13824 | final_length = wi.compressed_bl.length(); | |
13825 | csum_length = final_length; | |
3efd9988 FG |
13826 | l = &wi.compressed_bl; |
13827 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
13828 | } else if (wi.new_blob) { | |
7c673cae | 13829 | // initialize newly created blob only |
11fdf7f2 TL |
13830 | ceph_assert(dblob.is_mutable()); |
13831 | unsigned csum_order; | |
7c673cae FG |
13832 | if (l->length() != wi.blob_length) { |
13833 | // hrm, maybe we could do better here, but let's not bother. | |
13834 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
13835 | << block_size_order << dendl; | |
31f18b77 | 13836 | csum_order = block_size_order; |
7c673cae FG |
13837 | } else { |
13838 | csum_order = std::min(wctx->csum_order, ctz(l->length())); | |
13839 | } | |
13840 | // try to align blob with max_blob_size to improve | |
13841 | // its reuse ratio, e.g. in case of reverse write | |
13842 | uint32_t suggested_boff = | |
13843 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
13844 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
13845 | suggested_boff + final_length <= max_bsize && | |
13846 | suggested_boff > b_off) { | |
181888fb | 13847 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 13848 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 13849 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
13850 | csum_length += suggested_boff - b_off; |
13851 | b_off = suggested_boff; | |
13852 | } | |
181888fb FG |
13853 | if (csum != Checksummer::CSUM_NONE) { |
13854 | dout(20) << __func__ << " initialize csum setting for new blob " << *b | |
13855 | << " csum_type " << Checksummer::get_csum_type_string(csum) | |
13856 | << " csum_order " << csum_order | |
13857 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
13858 | << dendl; | |
13859 | dblob.init_csum(csum, csum_order, csum_length); | |
13860 | } | |
7c673cae FG |
13861 | } |
13862 | ||
a8e16298 | 13863 | PExtentVector extents; |
3efd9988 FG |
13864 | int64_t left = final_length; |
13865 | while (left > 0) { | |
11fdf7f2 | 13866 | ceph_assert(prealloc_left > 0); |
3efd9988 FG |
13867 | if (prealloc_pos->length <= left) { |
13868 | prealloc_left -= prealloc_pos->length; | |
13869 | left -= prealloc_pos->length; | |
13870 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
13871 | extents.push_back(*prealloc_pos); | |
13872 | ++prealloc_pos; | |
13873 | } else { | |
13874 | extents.emplace_back(prealloc_pos->offset, left); | |
13875 | prealloc_pos->offset += left; | |
13876 | prealloc_pos->length -= left; | |
13877 | prealloc_left -= left; | |
13878 | txc->statfs_delta.allocated() += left; | |
13879 | left = 0; | |
13880 | break; | |
13881 | } | |
13882 | } | |
7c673cae | 13883 | for (auto& p : extents) { |
3efd9988 | 13884 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 13885 | } |
11fdf7f2 | 13886 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 13887 | |
181888fb FG |
13888 | dout(20) << __func__ << " blob " << *b << dendl; |
13889 | if (dblob.has_csum()) { | |
7c673cae FG |
13890 | dblob.calc_csum(b_off, *l); |
13891 | } | |
181888fb | 13892 | |
7c673cae | 13893 | if (wi.mark_unused) { |
1911f103 | 13894 | ceph_assert(!dblob.is_compressed()); |
7c673cae FG |
13895 | auto b_end = b_off + wi.bl.length(); |
13896 | if (b_off) { | |
13897 | dblob.add_unused(0, b_off); | |
13898 | } | |
1911f103 TL |
13899 | uint64_t llen = dblob.get_logical_length(); |
13900 | if (b_end < llen) { | |
13901 | dblob.add_unused(b_end, llen - b_end); | |
7c673cae FG |
13902 | } |
13903 | } | |
13904 | ||
13905 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
13906 | b_off + (wi.b_off0 - wi.b_off), | |
13907 | wi.length0, | |
13908 | wi.b, | |
13909 | nullptr); | |
13910 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13911 | txc->statfs_delta.stored() += le->length; | |
13912 | dout(20) << __func__ << " lex " << *le << dendl; | |
13913 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
13914 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
13915 | ||
13916 | // queue io | |
11fdf7f2 | 13917 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13918 | if (l->length() <= prefer_deferred_size.load()) { |
13919 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13920 | << l->length() << std::dec << " write via deferred" << dendl; | |
9f95a23c | 13921 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13922 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13923 | int r = b->get_blob().map( | |
13924 | b_off, l->length(), | |
13925 | [&](uint64_t offset, uint64_t length) { | |
13926 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13927 | return 0; | |
13928 | }); | |
11fdf7f2 | 13929 | ceph_assert(r == 0); |
7c673cae | 13930 | op->data = *l; |
81eedcae | 13931 | logger->inc(l_bluestore_write_small_deferred); |
7c673cae FG |
13932 | } else { |
13933 | b->get_blob().map_bl( | |
13934 | b_off, *l, | |
13935 | [&](uint64_t offset, bufferlist& t) { | |
13936 | bdev->aio_write(offset, t, &txc->ioc, false); | |
13937 | }); | |
81eedcae | 13938 | logger->inc(l_bluestore_write_small_new); |
7c673cae FG |
13939 | } |
13940 | } | |
13941 | } | |
11fdf7f2 TL |
13942 | ceph_assert(prealloc_pos == prealloc.end()); |
13943 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
13944 | return 0; |
13945 | } | |
13946 | ||
13947 | void BlueStore::_wctx_finish( | |
13948 | TransContext *txc, | |
13949 | CollectionRef& c, | |
13950 | OnodeRef o, | |
31f18b77 FG |
13951 | WriteContext *wctx, |
13952 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
13953 | { |
13954 | auto oep = wctx->old_extents.begin(); | |
13955 | while (oep != wctx->old_extents.end()) { | |
13956 | auto &lo = *oep; | |
13957 | oep = wctx->old_extents.erase(oep); | |
13958 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
13959 | BlobRef b = lo.e.blob; | |
13960 | const bluestore_blob_t& blob = b->get_blob(); | |
13961 | if (blob.is_compressed()) { | |
13962 | if (lo.blob_empty) { | |
13963 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
13964 | } | |
13965 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
13966 | } | |
13967 | auto& r = lo.r; | |
13968 | txc->statfs_delta.stored() -= lo.e.length; | |
13969 | if (!r.empty()) { | |
13970 | dout(20) << __func__ << " blob release " << r << dendl; | |
13971 | if (blob.is_shared()) { | |
13972 | PExtentVector final; | |
13973 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
13974 | bool unshare = false; |
13975 | bool* unshare_ptr = | |
13976 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 13977 | for (auto e : r) { |
31f18b77 FG |
13978 | b->shared_blob->put_ref( |
13979 | e.offset, e.length, &final, | |
11fdf7f2 TL |
13980 | unshare_ptr); |
13981 | } | |
13982 | if (unshare) { | |
13983 | ceph_assert(maybe_unshared_blobs); | |
13984 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
13985 | } |
13986 | dout(20) << __func__ << " shared_blob release " << final | |
13987 | << " from " << *b->shared_blob << dendl; | |
13988 | txc->write_shared_blob(b->shared_blob); | |
13989 | r.clear(); | |
13990 | r.swap(final); | |
13991 | } | |
13992 | } | |
13993 | // we can't invalidate our logical extents as we drop them because | |
13994 | // other lextents (either in our onode or others) may still | |
13995 | // reference them. but we can throw out anything that is no | |
13996 | // longer allocated. Note that this will leave behind edge bits | |
13997 | // that are no longer referenced but not deallocated (until they | |
13998 | // age out of the cache naturally). | |
13999 | b->discard_unallocated(c.get()); | |
14000 | for (auto e : r) { | |
14001 | dout(20) << __func__ << " release " << e << dendl; | |
14002 | txc->released.insert(e.offset, e.length); | |
14003 | txc->statfs_delta.allocated() -= e.length; | |
14004 | if (blob.is_compressed()) { | |
14005 | txc->statfs_delta.compressed_allocated() -= e.length; | |
14006 | } | |
14007 | } | |
9f95a23c TL |
14008 | |
14009 | if (b->is_spanning() && !b->is_referenced() && lo.blob_empty) { | |
7c673cae FG |
14010 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b |
14011 | << dendl; | |
14012 | o->extent_map.spanning_blob_map.erase(b->id); | |
14013 | } | |
9f95a23c | 14014 | delete &lo; |
7c673cae FG |
14015 | } |
14016 | } | |
14017 | ||
14018 | void BlueStore::_do_write_data( | |
14019 | TransContext *txc, | |
14020 | CollectionRef& c, | |
14021 | OnodeRef o, | |
14022 | uint64_t offset, | |
14023 | uint64_t length, | |
14024 | bufferlist& bl, | |
14025 | WriteContext *wctx) | |
14026 | { | |
14027 | uint64_t end = offset + length; | |
14028 | bufferlist::iterator p = bl.begin(); | |
14029 | ||
14030 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
14031 | (length != min_alloc_size)) { | |
14032 | // we fall within the same block | |
14033 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
14034 | } else { | |
14035 | uint64_t head_offset, head_length; | |
14036 | uint64_t middle_offset, middle_length; | |
14037 | uint64_t tail_offset, tail_length; | |
14038 | ||
14039 | head_offset = offset; | |
11fdf7f2 | 14040 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 14041 | |
11fdf7f2 TL |
14042 | tail_offset = p2align(end, min_alloc_size); |
14043 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
14044 | |
14045 | middle_offset = head_offset + head_length; | |
14046 | middle_length = length - head_length - tail_length; | |
14047 | ||
14048 | if (head_length) { | |
14049 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
14050 | } | |
14051 | ||
14052 | if (middle_length) { | |
14053 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); | |
14054 | } | |
14055 | ||
14056 | if (tail_length) { | |
14057 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
14058 | } | |
14059 | } | |
14060 | } | |
14061 | ||
31f18b77 FG |
14062 | void BlueStore::_choose_write_options( |
14063 | CollectionRef& c, | |
14064 | OnodeRef o, | |
14065 | uint32_t fadvise_flags, | |
14066 | WriteContext *wctx) | |
7c673cae | 14067 | { |
7c673cae FG |
14068 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
14069 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 14070 | wctx->buffered = true; |
7c673cae FG |
14071 | } else if (cct->_conf->bluestore_default_buffered_write && |
14072 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
14073 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
14074 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 14075 | wctx->buffered = true; |
7c673cae FG |
14076 | } |
14077 | ||
31f18b77 FG |
14078 | // apply basic csum block size |
14079 | wctx->csum_order = block_size_order; | |
7c673cae FG |
14080 | |
14081 | // compression parameters | |
14082 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
14083 | auto cm = select_option( | |
14084 | "compression_mode", | |
31f18b77 | 14085 | comp_mode.load(), |
7c673cae FG |
14086 | [&]() { |
14087 | string val; | |
11fdf7f2 | 14088 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
31f18b77 FG |
14089 | return boost::optional<Compressor::CompressionMode>( |
14090 | Compressor::get_comp_mode_type(val)); | |
7c673cae FG |
14091 | } |
14092 | return boost::optional<Compressor::CompressionMode>(); | |
14093 | } | |
14094 | ); | |
31f18b77 FG |
14095 | |
14096 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
14097 | ((cm == Compressor::COMP_FORCE) || |
14098 | (cm == Compressor::COMP_AGGRESSIVE && | |
14099 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
14100 | (cm == Compressor::COMP_PASSIVE && | |
14101 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
14102 | ||
14103 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
14104 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
14105 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
14106 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 14107 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 14108 | |
7c673cae | 14109 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 14110 | |
7c673cae | 14111 | if (o->onode.expected_write_size) { |
224ce89b | 14112 | wctx->csum_order = std::max(min_alloc_size_order, |
31f18b77 | 14113 | (uint8_t)ctz(o->onode.expected_write_size)); |
7c673cae | 14114 | } else { |
224ce89b | 14115 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
14116 | } |
14117 | ||
31f18b77 FG |
14118 | if (wctx->compress) { |
14119 | wctx->target_blob_size = select_option( | |
7c673cae | 14120 | "compression_max_blob_size", |
31f18b77 | 14121 | comp_max_blob_size.load(), |
7c673cae | 14122 | [&]() { |
11fdf7f2 TL |
14123 | int64_t val; |
14124 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
7c673cae FG |
14125 | return boost::optional<uint64_t>((uint64_t)val); |
14126 | } | |
14127 | return boost::optional<uint64_t>(); | |
14128 | } | |
14129 | ); | |
14130 | } | |
14131 | } else { | |
31f18b77 FG |
14132 | if (wctx->compress) { |
14133 | wctx->target_blob_size = select_option( | |
7c673cae | 14134 | "compression_min_blob_size", |
31f18b77 | 14135 | comp_min_blob_size.load(), |
7c673cae | 14136 | [&]() { |
11fdf7f2 TL |
14137 | int64_t val; |
14138 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
7c673cae FG |
14139 | return boost::optional<uint64_t>((uint64_t)val); |
14140 | } | |
14141 | return boost::optional<uint64_t>(); | |
14142 | } | |
14143 | ); | |
14144 | } | |
14145 | } | |
31f18b77 | 14146 | |
7c673cae | 14147 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
14148 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
14149 | wctx->target_blob_size = max_bsize; | |
7c673cae | 14150 | } |
31f18b77 | 14151 | |
7c673cae FG |
14152 | // set the min blob size floor at 2x the min_alloc_size, or else we |
14153 | // won't be able to allocate a smaller extent for the compressed | |
14154 | // data. | |
31f18b77 FG |
14155 | if (wctx->compress && |
14156 | wctx->target_blob_size < min_alloc_size * 2) { | |
14157 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 14158 | } |
31f18b77 FG |
14159 | |
14160 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
14161 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
14162 | << " compress=" << (int)wctx->compress |
14163 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
14164 | << std::dec << dendl; |
14165 | } | |
14166 | ||
14167 | int BlueStore::_do_gc( | |
14168 | TransContext *txc, | |
14169 | CollectionRef& c, | |
14170 | OnodeRef o, | |
31f18b77 FG |
14171 | const WriteContext& wctx, |
14172 | uint64_t *dirty_start, | |
14173 | uint64_t *dirty_end) | |
14174 | { | |
31f18b77 | 14175 | |
1adf2230 | 14176 | bool dirty_range_updated = false; |
31f18b77 | 14177 | WriteContext wctx_gc; |
7c673cae | 14178 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 14179 | |
eafe8130 | 14180 | auto & extents_to_collect = wctx.extents_to_gc; |
31f18b77 FG |
14181 | for (auto it = extents_to_collect.begin(); |
14182 | it != extents_to_collect.end(); | |
14183 | ++it) { | |
14184 | bufferlist bl; | |
eafe8130 TL |
14185 | auto offset = (*it).first; |
14186 | auto length = (*it).second; | |
14187 | dout(20) << __func__ << " processing " << std::hex | |
14188 | << offset << "~" << length << std::dec | |
14189 | << dendl; | |
14190 | int r = _do_read(c.get(), o, offset, length, bl, 0); | |
14191 | ceph_assert(r == (int)length); | |
31f18b77 | 14192 | |
eafe8130 TL |
14193 | _do_write_data(txc, c, o, offset, length, bl, &wctx_gc); |
14194 | logger->inc(l_bluestore_gc_merged, length); | |
31f18b77 | 14195 | |
eafe8130 TL |
14196 | if (*dirty_start > offset) { |
14197 | *dirty_start = offset; | |
1adf2230 | 14198 | dirty_range_updated = true; |
31f18b77 FG |
14199 | } |
14200 | ||
eafe8130 TL |
14201 | if (*dirty_end < offset + length) { |
14202 | *dirty_end = offset + length; | |
1adf2230 | 14203 | dirty_range_updated = true; |
31f18b77 FG |
14204 | } |
14205 | } | |
1adf2230 AA |
14206 | if (dirty_range_updated) { |
14207 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
14208 | } | |
31f18b77 FG |
14209 | |
14210 | dout(30) << __func__ << " alloc write" << dendl; | |
14211 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
14212 | if (r < 0) { | |
14213 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14214 | << dendl; | |
14215 | return r; | |
14216 | } | |
14217 | ||
14218 | _wctx_finish(txc, c, o, &wctx_gc); | |
14219 | return 0; | |
14220 | } | |
14221 | ||
14222 | int BlueStore::_do_write( | |
14223 | TransContext *txc, | |
14224 | CollectionRef& c, | |
14225 | OnodeRef o, | |
14226 | uint64_t offset, | |
14227 | uint64_t length, | |
14228 | bufferlist& bl, | |
14229 | uint32_t fadvise_flags) | |
14230 | { | |
14231 | int r = 0; | |
14232 | ||
14233 | dout(20) << __func__ | |
14234 | << " " << o->oid | |
14235 | << " 0x" << std::hex << offset << "~" << length | |
14236 | << " - have 0x" << o->onode.size | |
14237 | << " (" << std::dec << o->onode.size << ")" | |
14238 | << " bytes" | |
14239 | << " fadvise_flags 0x" << std::hex << fadvise_flags << std::dec | |
14240 | << dendl; | |
81eedcae | 14241 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
14242 | |
14243 | if (length == 0) { | |
14244 | return 0; | |
14245 | } | |
14246 | ||
14247 | uint64_t end = offset + length; | |
14248 | ||
14249 | GarbageCollector gc(c->store->cct); | |
eafe8130 | 14250 | int64_t benefit = 0; |
31f18b77 FG |
14251 | auto dirty_start = offset; |
14252 | auto dirty_end = end; | |
14253 | ||
14254 | WriteContext wctx; | |
14255 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
14256 | o->extent_map.fault_range(db, offset, length); |
14257 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
14258 | r = _do_alloc_write(txc, c, o, &wctx); |
14259 | if (r < 0) { | |
14260 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14261 | << dendl; | |
14262 | goto out; | |
14263 | } | |
14264 | ||
eafe8130 TL |
14265 | if (wctx.extents_to_gc.empty() || |
14266 | wctx.extents_to_gc.range_start() > offset || | |
14267 | wctx.extents_to_gc.range_end() < offset + length) { | |
14268 | benefit = gc.estimate(offset, | |
14269 | length, | |
14270 | o->extent_map, | |
14271 | wctx.old_extents, | |
14272 | min_alloc_size); | |
14273 | } | |
14274 | ||
31f18b77 FG |
14275 | // NB: _wctx_finish() will empty old_extents |
14276 | // so we must do gc estimation before that | |
7c673cae FG |
14277 | _wctx_finish(txc, c, o, &wctx); |
14278 | if (end > o->onode.size) { | |
14279 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 14280 | << std::dec << dendl; |
7c673cae FG |
14281 | o->onode.size = end; |
14282 | } | |
14283 | ||
11fdf7f2 | 14284 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
eafe8130 TL |
14285 | wctx.extents_to_gc.union_of(gc.get_extents_to_collect()); |
14286 | dout(20) << __func__ | |
14287 | << " perform garbage collection for compressed extents, " | |
14288 | << "expected benefit = " << benefit << " AUs" << dendl; | |
14289 | } | |
14290 | if (!wctx.extents_to_gc.empty()) { | |
14291 | dout(20) << __func__ << " perform garbage collection" << dendl; | |
14292 | ||
14293 | r = _do_gc(txc, c, o, | |
14294 | wctx, | |
14295 | &dirty_start, &dirty_end); | |
14296 | if (r < 0) { | |
14297 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
14298 | << dendl; | |
14299 | goto out; | |
7c673cae | 14300 | } |
eafe8130 TL |
14301 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
14302 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae | 14303 | } |
7c673cae | 14304 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
14305 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
14306 | ||
7c673cae FG |
14307 | r = 0; |
14308 | ||
14309 | out: | |
14310 | return r; | |
14311 | } | |
14312 | ||
14313 | int BlueStore::_write(TransContext *txc, | |
14314 | CollectionRef& c, | |
14315 | OnodeRef& o, | |
31f18b77 FG |
14316 | uint64_t offset, size_t length, |
14317 | bufferlist& bl, | |
14318 | uint32_t fadvise_flags) | |
7c673cae FG |
14319 | { |
14320 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14321 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14322 | << dendl; | |
35e4c445 FG |
14323 | int r = 0; |
14324 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14325 | r = -E2BIG; | |
14326 | } else { | |
14327 | _assign_nid(txc, o); | |
14328 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
14329 | txc->write_onode(o); | |
14330 | } | |
7c673cae FG |
14331 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14332 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14333 | << " = " << r << dendl; | |
14334 | return r; | |
14335 | } | |
14336 | ||
14337 | int BlueStore::_zero(TransContext *txc, | |
14338 | CollectionRef& c, | |
14339 | OnodeRef& o, | |
14340 | uint64_t offset, size_t length) | |
14341 | { | |
14342 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14343 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14344 | << dendl; | |
35e4c445 FG |
14345 | int r = 0; |
14346 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14347 | r = -E2BIG; | |
14348 | } else { | |
14349 | _assign_nid(txc, o); | |
14350 | r = _do_zero(txc, c, o, offset, length); | |
14351 | } | |
7c673cae FG |
14352 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14353 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14354 | << " = " << r << dendl; | |
14355 | return r; | |
14356 | } | |
14357 | ||
14358 | int BlueStore::_do_zero(TransContext *txc, | |
14359 | CollectionRef& c, | |
14360 | OnodeRef& o, | |
14361 | uint64_t offset, size_t length) | |
14362 | { | |
14363 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14364 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14365 | << dendl; | |
14366 | int r = 0; | |
14367 | ||
81eedcae | 14368 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14369 | |
14370 | WriteContext wctx; | |
14371 | o->extent_map.fault_range(db, offset, length); | |
14372 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 14373 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
14374 | _wctx_finish(txc, c, o, &wctx); |
14375 | ||
b32b8144 | 14376 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
14377 | o->onode.size = offset + length; |
14378 | dout(20) << __func__ << " extending size to " << offset + length | |
14379 | << dendl; | |
14380 | } | |
14381 | txc->write_onode(o); | |
14382 | ||
14383 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14384 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14385 | << " = " << r << dendl; | |
14386 | return r; | |
14387 | } | |
14388 | ||
14389 | void BlueStore::_do_truncate( | |
31f18b77 FG |
14390 | TransContext *txc, CollectionRef& c, OnodeRef o, uint64_t offset, |
14391 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14392 | { |
14393 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14394 | << " 0x" << std::hex << offset << std::dec << dendl; | |
14395 | ||
81eedcae | 14396 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14397 | |
14398 | if (offset == o->onode.size) | |
31f18b77 | 14399 | return; |
7c673cae FG |
14400 | |
14401 | if (offset < o->onode.size) { | |
14402 | WriteContext wctx; | |
14403 | uint64_t length = o->onode.size - offset; | |
14404 | o->extent_map.fault_range(db, offset, length); | |
14405 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 FG |
14406 | o->extent_map.dirty_range(offset, length); |
14407 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); | |
7c673cae FG |
14408 | |
14409 | // if we have shards past EOF, ask for a reshard | |
14410 | if (!o->onode.extent_map_shards.empty() && | |
14411 | o->onode.extent_map_shards.back().offset >= offset) { | |
14412 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
14413 | if (offset) { | |
14414 | o->extent_map.request_reshard(offset - 1, offset + length); | |
14415 | } else { | |
14416 | o->extent_map.request_reshard(0, length); | |
14417 | } | |
14418 | } | |
14419 | } | |
14420 | ||
14421 | o->onode.size = offset; | |
14422 | ||
14423 | txc->write_onode(o); | |
14424 | } | |
14425 | ||
35e4c445 | 14426 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
14427 | CollectionRef& c, |
14428 | OnodeRef& o, | |
14429 | uint64_t offset) | |
14430 | { | |
14431 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14432 | << " 0x" << std::hex << offset << std::dec | |
14433 | << dendl; | |
35e4c445 FG |
14434 | int r = 0; |
14435 | if (offset >= OBJECT_MAX_SIZE) { | |
14436 | r = -E2BIG; | |
14437 | } else { | |
14438 | _do_truncate(txc, c, o, offset); | |
14439 | } | |
14440 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14441 | << " 0x" << std::hex << offset << std::dec | |
14442 | << " = " << r << dendl; | |
14443 | return r; | |
7c673cae FG |
14444 | } |
14445 | ||
14446 | int BlueStore::_do_remove( | |
14447 | TransContext *txc, | |
14448 | CollectionRef& c, | |
14449 | OnodeRef o) | |
14450 | { | |
31f18b77 | 14451 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
14452 | bool is_gen = !o->oid.is_no_gen(); |
14453 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
14454 | if (o->onode.has_omap()) { |
14455 | o->flush(); | |
9f95a23c | 14456 | _do_omap_clear(txc, o); |
7c673cae FG |
14457 | } |
14458 | o->exists = false; | |
14459 | string key; | |
14460 | for (auto &s : o->extent_map.shards) { | |
14461 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
14462 | << s.shard_info->offset << std::dec << dendl; | |
14463 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
14464 | [&](const string& final_key) { | |
14465 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14466 | } | |
14467 | ); | |
14468 | } | |
14469 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 14470 | txc->note_removed_object(o); |
7c673cae FG |
14471 | o->extent_map.clear(); |
14472 | o->onode = bluestore_onode_t(); | |
14473 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 14474 | |
224ce89b WB |
14475 | if (!is_gen || maybe_unshared_blobs.empty()) { |
14476 | return 0; | |
14477 | } | |
31f18b77 | 14478 | |
224ce89b WB |
14479 | // see if we can unshare blobs still referenced by the head |
14480 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
14481 | << maybe_unshared_blobs << dendl; | |
14482 | ghobject_t nogen = o->oid; | |
14483 | nogen.generation = ghobject_t::NO_GEN; | |
14484 | OnodeRef h = c->onode_map.lookup(nogen); | |
14485 | ||
14486 | if (!h || !h->exists) { | |
14487 | return 0; | |
14488 | } | |
14489 | ||
14490 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
14491 | << " " << h->oid << dendl; | |
14492 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
14493 | for (auto& e : h->extent_map.extent_map) { | |
14494 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14495 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14496 | if (b.is_shared() && | |
14497 | sb->loaded && | |
14498 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
14499 | if (b.is_compressed()) { |
14500 | expect[sb].get(0, b.get_ondisk_length()); | |
14501 | } else { | |
14502 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
14503 | expect[sb].get(off, len); | |
14504 | return 0; | |
14505 | }); | |
14506 | } | |
224ce89b WB |
14507 | } |
14508 | } | |
31f18b77 | 14509 | |
224ce89b WB |
14510 | vector<SharedBlob*> unshared_blobs; |
14511 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
14512 | for (auto& p : expect) { | |
14513 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
14514 | if (p.first->persistent->ref_map == p.second) { | |
14515 | SharedBlob *sb = p.first; | |
14516 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
14517 | unshared_blobs.push_back(sb); | |
14518 | txc->unshare_blob(sb); | |
14519 | uint64_t sbid = c->make_blob_unshared(sb); | |
14520 | string key; | |
14521 | get_shared_blob_key(sbid, &key); | |
14522 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
14523 | } | |
14524 | } | |
14525 | ||
14526 | if (unshared_blobs.empty()) { | |
14527 | return 0; | |
14528 | } | |
14529 | ||
224ce89b WB |
14530 | for (auto& e : h->extent_map.extent_map) { |
14531 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14532 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14533 | if (b.is_shared() && | |
14534 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
14535 | sb) != unshared_blobs.end()) { | |
14536 | dout(20) << __func__ << " unsharing " << e << dendl; | |
14537 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
14538 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 14539 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
14540 | } |
14541 | } | |
224ce89b WB |
14542 | txc->write_onode(h); |
14543 | ||
7c673cae FG |
14544 | return 0; |
14545 | } | |
14546 | ||
14547 | int BlueStore::_remove(TransContext *txc, | |
14548 | CollectionRef& c, | |
14549 | OnodeRef &o) | |
14550 | { | |
11fdf7f2 TL |
14551 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
14552 | << " onode " << o.get() | |
14553 | << " txc "<< txc << dendl; | |
7c673cae FG |
14554 | int r = _do_remove(txc, c, o); |
14555 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14556 | return r; | |
14557 | } | |
14558 | ||
14559 | int BlueStore::_setattr(TransContext *txc, | |
14560 | CollectionRef& c, | |
14561 | OnodeRef& o, | |
14562 | const string& name, | |
14563 | bufferptr& val) | |
14564 | { | |
14565 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14566 | << " " << name << " (" << val.length() << " bytes)" | |
14567 | << dendl; | |
14568 | int r = 0; | |
3efd9988 FG |
14569 | if (val.is_partial()) { |
14570 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
14571 | val.length()); | |
f91f0fd5 | 14572 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
14573 | } else { |
14574 | auto& b = o->onode.attrs[name.c_str()] = val; | |
f91f0fd5 | 14575 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 14576 | } |
7c673cae FG |
14577 | txc->write_onode(o); |
14578 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14579 | << " " << name << " (" << val.length() << " bytes)" | |
14580 | << " = " << r << dendl; | |
14581 | return r; | |
14582 | } | |
14583 | ||
14584 | int BlueStore::_setattrs(TransContext *txc, | |
14585 | CollectionRef& c, | |
14586 | OnodeRef& o, | |
14587 | const map<string,bufferptr>& aset) | |
14588 | { | |
14589 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14590 | << " " << aset.size() << " keys" | |
14591 | << dendl; | |
14592 | int r = 0; | |
14593 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
14594 | p != aset.end(); ++p) { | |
3efd9988 FG |
14595 | if (p->second.is_partial()) { |
14596 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 14597 | bufferptr(p->second.c_str(), p->second.length()); |
f91f0fd5 | 14598 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 FG |
14599 | } else { |
14600 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
f91f0fd5 | 14601 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_meta); |
3efd9988 | 14602 | } |
7c673cae FG |
14603 | } |
14604 | txc->write_onode(o); | |
14605 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14606 | << " " << aset.size() << " keys" | |
14607 | << " = " << r << dendl; | |
14608 | return r; | |
14609 | } | |
14610 | ||
14611 | ||
14612 | int BlueStore::_rmattr(TransContext *txc, | |
14613 | CollectionRef& c, | |
14614 | OnodeRef& o, | |
14615 | const string& name) | |
14616 | { | |
14617 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14618 | << " " << name << dendl; | |
14619 | int r = 0; | |
14620 | auto it = o->onode.attrs.find(name.c_str()); | |
14621 | if (it == o->onode.attrs.end()) | |
14622 | goto out; | |
14623 | ||
14624 | o->onode.attrs.erase(it); | |
14625 | txc->write_onode(o); | |
14626 | ||
14627 | out: | |
14628 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14629 | << " " << name << " = " << r << dendl; | |
14630 | return r; | |
14631 | } | |
14632 | ||
14633 | int BlueStore::_rmattrs(TransContext *txc, | |
14634 | CollectionRef& c, | |
14635 | OnodeRef& o) | |
14636 | { | |
14637 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14638 | int r = 0; | |
14639 | ||
14640 | if (o->onode.attrs.empty()) | |
14641 | goto out; | |
14642 | ||
14643 | o->onode.attrs.clear(); | |
14644 | txc->write_onode(o); | |
14645 | ||
14646 | out: | |
14647 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14648 | return r; | |
14649 | } | |
14650 | ||
9f95a23c | 14651 | void BlueStore::_do_omap_clear(TransContext *txc, OnodeRef& o) |
7c673cae | 14652 | { |
9f95a23c | 14653 | const string& omap_prefix = o->get_omap_prefix(); |
7c673cae | 14654 | string prefix, tail; |
9f95a23c TL |
14655 | o->get_omap_header(&prefix); |
14656 | o->get_omap_tail(&tail); | |
11fdf7f2 | 14657 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
494da23a | 14658 | txc->t->rmkey(omap_prefix, tail); |
11fdf7f2 TL |
14659 | dout(20) << __func__ << " remove range start: " |
14660 | << pretty_binary_string(prefix) << " end: " | |
14661 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
14662 | } |
14663 | ||
14664 | int BlueStore::_omap_clear(TransContext *txc, | |
14665 | CollectionRef& c, | |
14666 | OnodeRef& o) | |
14667 | { | |
14668 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14669 | int r = 0; | |
14670 | if (o->onode.has_omap()) { | |
14671 | o->flush(); | |
9f95a23c | 14672 | _do_omap_clear(txc, o); |
7c673cae FG |
14673 | o->onode.clear_omap_flag(); |
14674 | txc->write_onode(o); | |
14675 | } | |
14676 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14677 | return r; | |
14678 | } | |
14679 | ||
14680 | int BlueStore::_omap_setkeys(TransContext *txc, | |
14681 | CollectionRef& c, | |
14682 | OnodeRef& o, | |
14683 | bufferlist &bl) | |
14684 | { | |
14685 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14686 | int r; | |
11fdf7f2 | 14687 | auto p = bl.cbegin(); |
7c673cae FG |
14688 | __u32 num; |
14689 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14690 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14691 | o->onode.set_omap_flags_pgmeta(); |
14692 | } else { | |
14693 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14694 | } |
7c673cae | 14695 | txc->write_onode(o); |
494da23a | 14696 | |
9f95a23c | 14697 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14698 | string key_tail; |
14699 | bufferlist tail; | |
9f95a23c | 14700 | o->get_omap_tail(&key_tail); |
494da23a | 14701 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14702 | } else { |
14703 | txc->note_modified_object(o); | |
14704 | } | |
9f95a23c | 14705 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 14706 | string final_key; |
9f95a23c TL |
14707 | o->get_omap_key(string(), &final_key); |
14708 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 14709 | decode(num, p); |
7c673cae FG |
14710 | while (num--) { |
14711 | string key; | |
14712 | bufferlist value; | |
11fdf7f2 TL |
14713 | decode(key, p); |
14714 | decode(value, p); | |
9f95a23c | 14715 | final_key.resize(base_key_len); // keep prefix |
7c673cae | 14716 | final_key += key; |
11fdf7f2 | 14717 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 14718 | << " <- " << key << dendl; |
11fdf7f2 | 14719 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
14720 | } |
14721 | r = 0; | |
14722 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14723 | return r; | |
14724 | } | |
14725 | ||
14726 | int BlueStore::_omap_setheader(TransContext *txc, | |
14727 | CollectionRef& c, | |
14728 | OnodeRef &o, | |
14729 | bufferlist& bl) | |
14730 | { | |
14731 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14732 | int r; | |
14733 | string key; | |
14734 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14735 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14736 | o->onode.set_omap_flags_pgmeta(); |
14737 | } else { | |
14738 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14739 | } |
7c673cae | 14740 | txc->write_onode(o); |
494da23a | 14741 | |
9f95a23c | 14742 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14743 | string key_tail; |
14744 | bufferlist tail; | |
9f95a23c | 14745 | o->get_omap_tail(&key_tail); |
494da23a | 14746 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14747 | } else { |
14748 | txc->note_modified_object(o); | |
14749 | } | |
9f95a23c TL |
14750 | const string& prefix = o->get_omap_prefix(); |
14751 | o->get_omap_header(&key); | |
11fdf7f2 | 14752 | txc->t->set(prefix, key, bl); |
7c673cae FG |
14753 | r = 0; |
14754 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14755 | return r; | |
14756 | } | |
14757 | ||
14758 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
14759 | CollectionRef& c, | |
14760 | OnodeRef& o, | |
14761 | bufferlist& bl) | |
14762 | { | |
14763 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14764 | int r = 0; | |
11fdf7f2 | 14765 | auto p = bl.cbegin(); |
7c673cae FG |
14766 | __u32 num; |
14767 | string final_key; | |
14768 | ||
14769 | if (!o->onode.has_omap()) { | |
14770 | goto out; | |
14771 | } | |
11fdf7f2 | 14772 | { |
9f95a23c TL |
14773 | const string& prefix = o->get_omap_prefix(); |
14774 | o->get_omap_key(string(), &final_key); | |
14775 | size_t base_key_len = final_key.size(); | |
11fdf7f2 TL |
14776 | decode(num, p); |
14777 | while (num--) { | |
14778 | string key; | |
14779 | decode(key, p); | |
9f95a23c | 14780 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
14781 | final_key += key; |
14782 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
14783 | << " <- " << key << dendl; | |
14784 | txc->t->rmkey(prefix, final_key); | |
14785 | } | |
7c673cae FG |
14786 | } |
14787 | txc->note_modified_object(o); | |
14788 | ||
14789 | out: | |
14790 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14791 | return r; | |
14792 | } | |
14793 | ||
14794 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
14795 | CollectionRef& c, | |
14796 | OnodeRef& o, | |
14797 | const string& first, const string& last) | |
14798 | { | |
14799 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
14800 | string key_first, key_last; |
14801 | int r = 0; | |
14802 | if (!o->onode.has_omap()) { | |
14803 | goto out; | |
14804 | } | |
11fdf7f2 | 14805 | { |
9f95a23c | 14806 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 14807 | o->flush(); |
9f95a23c TL |
14808 | o->get_omap_key(first, &key_first); |
14809 | o->get_omap_key(last, &key_last); | |
11fdf7f2 TL |
14810 | txc->t->rm_range_keys(prefix, key_first, key_last); |
14811 | dout(20) << __func__ << " remove range start: " | |
14812 | << pretty_binary_string(key_first) << " end: " | |
14813 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
14814 | } |
14815 | txc->note_modified_object(o); | |
14816 | ||
14817 | out: | |
14818 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14819 | return r; | |
14820 | } | |
14821 | ||
14822 | int BlueStore::_set_alloc_hint( | |
14823 | TransContext *txc, | |
14824 | CollectionRef& c, | |
14825 | OnodeRef& o, | |
14826 | uint64_t expected_object_size, | |
14827 | uint64_t expected_write_size, | |
14828 | uint32_t flags) | |
14829 | { | |
14830 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14831 | << " object_size " << expected_object_size | |
14832 | << " write_size " << expected_write_size | |
14833 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
14834 | << dendl; | |
14835 | int r = 0; | |
14836 | o->onode.expected_object_size = expected_object_size; | |
14837 | o->onode.expected_write_size = expected_write_size; | |
14838 | o->onode.alloc_hint_flags = flags; | |
14839 | txc->write_onode(o); | |
14840 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14841 | << " object_size " << expected_object_size | |
14842 | << " write_size " << expected_write_size | |
14843 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
14844 | << " = " << r << dendl; | |
14845 | return r; | |
14846 | } | |
14847 | ||
14848 | int BlueStore::_clone(TransContext *txc, | |
14849 | CollectionRef& c, | |
14850 | OnodeRef& oldo, | |
14851 | OnodeRef& newo) | |
14852 | { | |
14853 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14854 | << newo->oid << dendl; | |
14855 | int r = 0; | |
14856 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
14857 | derr << __func__ << " mismatched hash on " << oldo->oid | |
14858 | << " and " << newo->oid << dendl; | |
14859 | return -EINVAL; | |
14860 | } | |
14861 | ||
7c673cae FG |
14862 | _assign_nid(txc, newo); |
14863 | ||
14864 | // clone data | |
14865 | oldo->flush(); | |
14866 | _do_truncate(txc, c, newo, 0); | |
14867 | if (cct->_conf->bluestore_clone_cow) { | |
14868 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
14869 | } else { | |
14870 | bufferlist bl; | |
14871 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
14872 | if (r < 0) | |
14873 | goto out; | |
14874 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
14875 | if (r < 0) | |
14876 | goto out; | |
14877 | } | |
14878 | ||
14879 | // clone attrs | |
14880 | newo->onode.attrs = oldo->onode.attrs; | |
14881 | ||
14882 | // clone omap | |
14883 | if (newo->onode.has_omap()) { | |
14884 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
14885 | newo->flush(); | |
9f95a23c | 14886 | _do_omap_clear(txc, newo); |
494da23a | 14887 | newo->onode.clear_omap_flag(); |
7c673cae FG |
14888 | } |
14889 | if (oldo->onode.has_omap()) { | |
14890 | dout(20) << __func__ << " copying omap data" << dendl; | |
494da23a | 14891 | if (newo->oid.is_pgmeta()) { |
9f95a23c TL |
14892 | newo->onode.set_omap_flags_pgmeta(); |
14893 | } else { | |
14894 | newo->onode.set_omap_flags(); | |
7c673cae | 14895 | } |
9f95a23c | 14896 | const string& prefix = newo->get_omap_prefix(); |
11fdf7f2 | 14897 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 14898 | string head, tail; |
9f95a23c TL |
14899 | oldo->get_omap_header(&head); |
14900 | oldo->get_omap_tail(&tail); | |
7c673cae FG |
14901 | it->lower_bound(head); |
14902 | while (it->valid()) { | |
14903 | if (it->key() >= tail) { | |
14904 | dout(30) << __func__ << " reached tail" << dendl; | |
14905 | break; | |
14906 | } else { | |
14907 | dout(30) << __func__ << " got header/data " | |
14908 | << pretty_binary_string(it->key()) << dendl; | |
14909 | string key; | |
9f95a23c | 14910 | newo->rewrite_omap_key(it->key(), &key); |
11fdf7f2 | 14911 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
14912 | } |
14913 | it->next(); | |
14914 | } | |
494da23a TL |
14915 | string new_tail; |
14916 | bufferlist new_tail_value; | |
9f95a23c | 14917 | newo->get_omap_tail(&new_tail); |
494da23a | 14918 | txc->t->set(prefix, new_tail, new_tail_value); |
7c673cae FG |
14919 | } |
14920 | ||
14921 | txc->write_onode(newo); | |
14922 | r = 0; | |
14923 | ||
14924 | out: | |
14925 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14926 | << newo->oid << " = " << r << dendl; | |
14927 | return r; | |
14928 | } | |
14929 | ||
14930 | int BlueStore::_do_clone_range( | |
14931 | TransContext *txc, | |
14932 | CollectionRef& c, | |
14933 | OnodeRef& oldo, | |
14934 | OnodeRef& newo, | |
224ce89b WB |
14935 | uint64_t srcoff, |
14936 | uint64_t length, | |
14937 | uint64_t dstoff) | |
7c673cae FG |
14938 | { |
14939 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14940 | << newo->oid | |
14941 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
14942 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
14943 | oldo->extent_map.fault_range(db, srcoff, length); | |
14944 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
14945 | _dump_onode<30>(cct, *oldo); |
14946 | _dump_onode<30>(cct, *newo); | |
7c673cae | 14947 | |
11fdf7f2 | 14948 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
81eedcae TL |
14949 | _dump_onode<30>(cct, *oldo); |
14950 | _dump_onode<30>(cct, *newo); | |
7c673cae FG |
14951 | return 0; |
14952 | } | |
14953 | ||
14954 | int BlueStore::_clone_range(TransContext *txc, | |
14955 | CollectionRef& c, | |
14956 | OnodeRef& oldo, | |
14957 | OnodeRef& newo, | |
14958 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
14959 | { | |
14960 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14961 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
14962 | << " to offset 0x" << dstoff << std::dec << dendl; | |
14963 | int r = 0; | |
14964 | ||
35e4c445 FG |
14965 | if (srcoff + length >= OBJECT_MAX_SIZE || |
14966 | dstoff + length >= OBJECT_MAX_SIZE) { | |
14967 | r = -E2BIG; | |
14968 | goto out; | |
14969 | } | |
7c673cae FG |
14970 | if (srcoff + length > oldo->onode.size) { |
14971 | r = -EINVAL; | |
14972 | goto out; | |
14973 | } | |
14974 | ||
7c673cae FG |
14975 | _assign_nid(txc, newo); |
14976 | ||
14977 | if (length > 0) { | |
14978 | if (cct->_conf->bluestore_clone_cow) { | |
14979 | _do_zero(txc, c, newo, dstoff, length); | |
14980 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
14981 | } else { | |
14982 | bufferlist bl; | |
14983 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
14984 | if (r < 0) | |
14985 | goto out; | |
14986 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
14987 | if (r < 0) | |
14988 | goto out; | |
14989 | } | |
14990 | } | |
14991 | ||
14992 | txc->write_onode(newo); | |
14993 | r = 0; | |
14994 | ||
14995 | out: | |
14996 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14997 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
14998 | << " to offset 0x" << dstoff << std::dec | |
14999 | << " = " << r << dendl; | |
15000 | return r; | |
15001 | } | |
15002 | ||
15003 | int BlueStore::_rename(TransContext *txc, | |
15004 | CollectionRef& c, | |
15005 | OnodeRef& oldo, | |
15006 | OnodeRef& newo, | |
15007 | const ghobject_t& new_oid) | |
15008 | { | |
15009 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
15010 | << new_oid << dendl; | |
15011 | int r; | |
15012 | ghobject_t old_oid = oldo->oid; | |
f91f0fd5 | 15013 | mempool::bluestore_cache_meta::string new_okey; |
7c673cae FG |
15014 | |
15015 | if (newo) { | |
15016 | if (newo->exists) { | |
15017 | r = -EEXIST; | |
15018 | goto out; | |
15019 | } | |
11fdf7f2 | 15020 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
15021 | } |
15022 | ||
15023 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
15024 | ||
15025 | // rewrite shards | |
15026 | { | |
15027 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
15028 | get_object_key(cct, new_oid, &new_okey); | |
15029 | string key; | |
15030 | for (auto &s : oldo->extent_map.shards) { | |
15031 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
15032 | [&](const string& final_key) { | |
15033 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
15034 | } | |
15035 | ); | |
15036 | s.dirty = true; | |
15037 | } | |
15038 | } | |
15039 | ||
15040 | newo = oldo; | |
15041 | txc->write_onode(newo); | |
15042 | ||
15043 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
15044 | // Onode in the old slot | |
15045 | c->onode_map.rename(oldo, old_oid, new_oid, new_okey); | |
15046 | r = 0; | |
15047 | ||
f64942e4 AA |
15048 | // hold a ref to new Onode in old name position, to ensure we don't drop |
15049 | // it from the cache before this txc commits (or else someone may come along | |
15050 | // and read newo's metadata via the old name). | |
15051 | txc->note_modified_object(oldo); | |
15052 | ||
7c673cae FG |
15053 | out: |
15054 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
15055 | << new_oid << " = " << r << dendl; | |
15056 | return r; | |
15057 | } | |
15058 | ||
15059 | // collections | |
15060 | ||
15061 | int BlueStore::_create_collection( | |
15062 | TransContext *txc, | |
15063 | const coll_t &cid, | |
15064 | unsigned bits, | |
15065 | CollectionRef *c) | |
15066 | { | |
15067 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
15068 | int r; | |
15069 | bufferlist bl; | |
15070 | ||
15071 | { | |
9f95a23c | 15072 | std::unique_lock l(coll_lock); |
7c673cae FG |
15073 | if (*c) { |
15074 | r = -EEXIST; | |
15075 | goto out; | |
15076 | } | |
11fdf7f2 TL |
15077 | auto p = new_coll_map.find(cid); |
15078 | ceph_assert(p != new_coll_map.end()); | |
15079 | *c = p->second; | |
7c673cae FG |
15080 | (*c)->cnode.bits = bits; |
15081 | coll_map[cid] = *c; | |
11fdf7f2 | 15082 | new_coll_map.erase(p); |
7c673cae | 15083 | } |
11fdf7f2 | 15084 | encode((*c)->cnode, bl); |
7c673cae FG |
15085 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
15086 | r = 0; | |
15087 | ||
15088 | out: | |
15089 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
15090 | return r; | |
15091 | } | |
15092 | ||
15093 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
15094 | CollectionRef *c) | |
15095 | { | |
15096 | dout(15) << __func__ << " " << cid << dendl; | |
15097 | int r; | |
15098 | ||
11fdf7f2 | 15099 | (*c)->flush_all_but_last(); |
7c673cae | 15100 | { |
9f95a23c | 15101 | std::unique_lock l(coll_lock); |
7c673cae FG |
15102 | if (!*c) { |
15103 | r = -ENOENT; | |
15104 | goto out; | |
15105 | } | |
15106 | size_t nonexistent_count = 0; | |
11fdf7f2 | 15107 | ceph_assert((*c)->exists); |
7c673cae FG |
15108 | if ((*c)->onode_map.map_any([&](OnodeRef o) { |
15109 | if (o->exists) { | |
494da23a TL |
15110 | dout(1) << __func__ << " " << o->oid << " " << o |
15111 | << " exists in onode_map" << dendl; | |
7c673cae FG |
15112 | return true; |
15113 | } | |
15114 | ++nonexistent_count; | |
15115 | return false; | |
15116 | })) { | |
15117 | r = -ENOTEMPTY; | |
15118 | goto out; | |
15119 | } | |
15120 | ||
15121 | vector<ghobject_t> ls; | |
15122 | ghobject_t next; | |
15123 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
15124 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 15125 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae | 15126 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
f91f0fd5 | 15127 | nonexistent_count + 1, false, &ls, &next); |
7c673cae | 15128 | if (r >= 0) { |
11fdf7f2 TL |
15129 | // If true mean collecton has more objects than nonexistent_count, |
15130 | // so bypass check. | |
15131 | bool exists = (!next.is_max()); | |
7c673cae FG |
15132 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
15133 | dout(10) << __func__ << " oid " << *it << dendl; | |
15134 | auto onode = (*c)->onode_map.lookup(*it); | |
15135 | exists = !onode || onode->exists; | |
15136 | if (exists) { | |
494da23a TL |
15137 | dout(1) << __func__ << " " << *it |
15138 | << " exists in db, " | |
15139 | << (!onode ? "not present in ram" : "present in ram") | |
15140 | << dendl; | |
7c673cae FG |
15141 | } |
15142 | } | |
15143 | if (!exists) { | |
11fdf7f2 | 15144 | _do_remove_collection(txc, c); |
7c673cae FG |
15145 | r = 0; |
15146 | } else { | |
15147 | dout(10) << __func__ << " " << cid | |
15148 | << " is non-empty" << dendl; | |
15149 | r = -ENOTEMPTY; | |
15150 | } | |
15151 | } | |
15152 | } | |
15153 | ||
15154 | out: | |
15155 | dout(10) << __func__ << " " << cid << " = " << r << dendl; | |
15156 | return r; | |
15157 | } | |
15158 | ||
11fdf7f2 TL |
15159 | void BlueStore::_do_remove_collection(TransContext *txc, |
15160 | CollectionRef *c) | |
15161 | { | |
15162 | coll_map.erase((*c)->cid); | |
15163 | txc->removed_collections.push_back(*c); | |
15164 | (*c)->exists = false; | |
15165 | _osr_register_zombie((*c)->osr.get()); | |
15166 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
15167 | c->reset(); | |
15168 | } | |
15169 | ||
7c673cae FG |
15170 | int BlueStore::_split_collection(TransContext *txc, |
15171 | CollectionRef& c, | |
15172 | CollectionRef& d, | |
15173 | unsigned bits, int rem) | |
15174 | { | |
15175 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
15176 | << " bits " << bits << dendl; | |
9f95a23c TL |
15177 | std::unique_lock l(c->lock); |
15178 | std::unique_lock l2(d->lock); | |
7c673cae FG |
15179 | int r; |
15180 | ||
15181 | // flush all previous deferred writes on this sequencer. this is a bit | |
15182 | // heavyweight, but we need to make sure all deferred writes complete | |
15183 | // before we split as the new collection's sequencer may need to order | |
15184 | // this after those writes, and we don't bother with the complexity of | |
15185 | // moving those TransContexts over to the new osr. | |
15186 | _osr_drain_preceding(txc); | |
15187 | ||
15188 | // move any cached items (onodes and referenced shared blobs) that will | |
15189 | // belong to the child collection post-split. leave everything else behind. | |
15190 | // this may include things that don't strictly belong to the now-smaller | |
15191 | // parent split, but the OSD will always send us a split for every new | |
15192 | // child. | |
15193 | ||
15194 | spg_t pgid, dest_pgid; | |
15195 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 15196 | ceph_assert(is_pg); |
7c673cae | 15197 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 15198 | ceph_assert(is_pg); |
7c673cae FG |
15199 | |
15200 | // the destination should initially be empty. | |
11fdf7f2 TL |
15201 | ceph_assert(d->onode_map.empty()); |
15202 | ceph_assert(d->shared_blob_set.empty()); | |
15203 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
15204 | |
15205 | c->split_cache(d.get()); | |
15206 | ||
15207 | // adjust bits. note that this will be redundant for all but the first | |
15208 | // split call for this parent (first child). | |
15209 | c->cnode.bits = bits; | |
11fdf7f2 | 15210 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
15211 | r = 0; |
15212 | ||
15213 | bufferlist bl; | |
11fdf7f2 | 15214 | encode(c->cnode, bl); |
7c673cae FG |
15215 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
15216 | ||
15217 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
15218 | << " bits " << bits << " = " << r << dendl; | |
15219 | return r; | |
15220 | } | |
15221 | ||
11fdf7f2 TL |
15222 | int BlueStore::_merge_collection( |
15223 | TransContext *txc, | |
15224 | CollectionRef *c, | |
15225 | CollectionRef& d, | |
15226 | unsigned bits) | |
15227 | { | |
15228 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
15229 | << " bits " << bits << dendl; | |
9f95a23c TL |
15230 | std::unique_lock l((*c)->lock); |
15231 | std::unique_lock l2(d->lock); | |
11fdf7f2 TL |
15232 | int r; |
15233 | ||
15234 | coll_t cid = (*c)->cid; | |
15235 | ||
15236 | // flush all previous deferred writes on the source collection to ensure | |
15237 | // that all deferred writes complete before we merge as the target collection's | |
15238 | // sequencer may need to order new ops after those writes. | |
15239 | ||
15240 | _osr_drain((*c)->osr.get()); | |
15241 | ||
15242 | // move any cached items (onodes and referenced shared blobs) that will | |
15243 | // belong to the child collection post-split. leave everything else behind. | |
15244 | // this may include things that don't strictly belong to the now-smaller | |
15245 | // parent split, but the OSD will always send us a split for every new | |
15246 | // child. | |
15247 | ||
15248 | spg_t pgid, dest_pgid; | |
15249 | bool is_pg = cid.is_pg(&pgid); | |
15250 | ceph_assert(is_pg); | |
15251 | is_pg = d->cid.is_pg(&dest_pgid); | |
15252 | ceph_assert(is_pg); | |
15253 | ||
15254 | // adjust bits. note that this will be redundant for all but the first | |
15255 | // merge call for the parent/target. | |
15256 | d->cnode.bits = bits; | |
15257 | ||
15258 | // behavior depends on target (d) bits, so this after that is updated. | |
15259 | (*c)->split_cache(d.get()); | |
15260 | ||
15261 | // remove source collection | |
15262 | { | |
9f95a23c | 15263 | std::unique_lock l3(coll_lock); |
11fdf7f2 TL |
15264 | _do_remove_collection(txc, c); |
15265 | } | |
15266 | ||
15267 | r = 0; | |
15268 | ||
15269 | bufferlist bl; | |
15270 | encode(d->cnode, bl); | |
15271 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
15272 | ||
15273 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
15274 | << " bits " << bits << " = " << r << dendl; | |
15275 | return r; | |
15276 | } | |
15277 | ||
494da23a TL |
15278 | void BlueStore::log_latency( |
15279 | const char* name, | |
15280 | int idx, | |
15281 | const ceph::timespan& l, | |
15282 | double lat_threshold, | |
15283 | const char* info) const | |
15284 | { | |
15285 | logger->tinc(idx, l); | |
15286 | if (lat_threshold > 0.0 && | |
15287 | l >= make_timespan(lat_threshold)) { | |
15288 | dout(0) << __func__ << " slow operation observed for " << name | |
15289 | << ", latency = " << l | |
15290 | << info | |
15291 | << dendl; | |
15292 | } | |
15293 | } | |
15294 | ||
11fdf7f2 | 15295 | void BlueStore::log_latency_fn( |
494da23a | 15296 | const char* name, |
11fdf7f2 TL |
15297 | int idx, |
15298 | const ceph::timespan& l, | |
494da23a TL |
15299 | double lat_threshold, |
15300 | std::function<string (const ceph::timespan& lat)> fn) const | |
11fdf7f2 | 15301 | { |
494da23a TL |
15302 | logger->tinc(idx, l); |
15303 | if (lat_threshold > 0.0 && | |
15304 | l >= make_timespan(lat_threshold)) { | |
15305 | dout(0) << __func__ << " slow operation observed for " << name | |
15306 | << ", latency = " << l | |
15307 | << fn(l) | |
15308 | << dendl; | |
15309 | } | |
11fdf7f2 TL |
15310 | } |
15311 | ||
9f95a23c TL |
15312 | #if defined(WITH_LTTNG) |
15313 | void BlueStore::BlueStoreThrottle::emit_initial_tracepoint( | |
15314 | KeyValueDB &db, | |
15315 | TransContext &txc, | |
15316 | mono_clock::time_point start_throttle_acquire) | |
15317 | { | |
15318 | pending_kv_ios += txc.ios; | |
15319 | if (txc.deferred_txn) { | |
15320 | pending_deferred_ios += txc.ios; | |
15321 | } | |
15322 | ||
15323 | uint64_t started = 0; | |
15324 | uint64_t completed = 0; | |
15325 | if (should_trace(&started, &completed)) { | |
15326 | txc.tracing = true; | |
15327 | uint64_t rocksdb_base_level, | |
15328 | rocksdb_estimate_pending_compaction_bytes, | |
15329 | rocksdb_cur_size_all_mem_tables, | |
15330 | rocksdb_compaction_pending, | |
15331 | rocksdb_mem_table_flush_pending, | |
15332 | rocksdb_num_running_compactions, | |
15333 | rocksdb_num_running_flushes, | |
15334 | rocksdb_actual_delayed_write_rate; | |
15335 | db.get_property( | |
15336 | "rocksdb.base-level", | |
15337 | &rocksdb_base_level); | |
15338 | db.get_property( | |
15339 | "rocksdb.estimate-pending-compaction-bytes", | |
15340 | &rocksdb_estimate_pending_compaction_bytes); | |
15341 | db.get_property( | |
15342 | "rocksdb.cur-size-all-mem-tables", | |
15343 | &rocksdb_cur_size_all_mem_tables); | |
15344 | db.get_property( | |
15345 | "rocksdb.compaction-pending", | |
15346 | &rocksdb_compaction_pending); | |
15347 | db.get_property( | |
15348 | "rocksdb.mem-table-flush-pending", | |
15349 | &rocksdb_mem_table_flush_pending); | |
15350 | db.get_property( | |
15351 | "rocksdb.num-running-compactions", | |
15352 | &rocksdb_num_running_compactions); | |
15353 | db.get_property( | |
15354 | "rocksdb.num-running-flushes", | |
15355 | &rocksdb_num_running_flushes); | |
15356 | db.get_property( | |
15357 | "rocksdb.actual-delayed-write-rate", | |
15358 | &rocksdb_actual_delayed_write_rate); | |
15359 | ||
15360 | ||
15361 | tracepoint( | |
15362 | bluestore, | |
15363 | transaction_initial_state, | |
15364 | txc.osr->get_sequencer_id(), | |
15365 | txc.seq, | |
15366 | throttle_bytes.get_current(), | |
15367 | throttle_deferred_bytes.get_current(), | |
15368 | pending_kv_ios, | |
15369 | pending_deferred_ios, | |
15370 | started, | |
15371 | completed, | |
15372 | ceph::to_seconds<double>(mono_clock::now() - start_throttle_acquire)); | |
15373 | ||
15374 | tracepoint( | |
15375 | bluestore, | |
15376 | transaction_initial_state_rocksdb, | |
15377 | txc.osr->get_sequencer_id(), | |
15378 | txc.seq, | |
15379 | rocksdb_base_level, | |
15380 | rocksdb_estimate_pending_compaction_bytes, | |
15381 | rocksdb_cur_size_all_mem_tables, | |
15382 | rocksdb_compaction_pending, | |
15383 | rocksdb_mem_table_flush_pending, | |
15384 | rocksdb_num_running_compactions, | |
15385 | rocksdb_num_running_flushes, | |
15386 | rocksdb_actual_delayed_write_rate); | |
15387 | } | |
15388 | } | |
15389 | #endif | |
15390 | ||
15391 | mono_clock::duration BlueStore::BlueStoreThrottle::log_state_latency( | |
15392 | TransContext &txc, PerfCounters *logger, int state) | |
15393 | { | |
15394 | mono_clock::time_point now = mono_clock::now(); | |
15395 | mono_clock::duration lat = now - txc.last_stamp; | |
15396 | logger->tinc(state, lat); | |
15397 | #if defined(WITH_LTTNG) | |
15398 | if (txc.tracing && | |
15399 | state >= l_bluestore_state_prepare_lat && | |
15400 | state <= l_bluestore_state_done_lat) { | |
15401 | OID_ELAPSED("", lat.to_nsec() / 1000.0, txc.get_state_latency_name(state)); | |
15402 | tracepoint( | |
15403 | bluestore, | |
15404 | transaction_state_duration, | |
15405 | txc.osr->get_sequencer_id(), | |
15406 | txc.seq, | |
15407 | state, | |
15408 | ceph::to_seconds<double>(lat)); | |
15409 | } | |
15410 | #endif | |
15411 | txc.last_stamp = now; | |
15412 | return lat; | |
15413 | } | |
15414 | ||
15415 | bool BlueStore::BlueStoreThrottle::try_start_transaction( | |
15416 | KeyValueDB &db, | |
15417 | TransContext &txc, | |
15418 | mono_clock::time_point start_throttle_acquire) | |
15419 | { | |
15420 | throttle_bytes.get(txc.cost); | |
15421 | ||
15422 | if (!txc.deferred_txn || throttle_deferred_bytes.get_or_fail(txc.cost)) { | |
15423 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15424 | return true; | |
15425 | } else { | |
15426 | return false; | |
15427 | } | |
15428 | } | |
15429 | ||
15430 | void BlueStore::BlueStoreThrottle::finish_start_transaction( | |
15431 | KeyValueDB &db, | |
15432 | TransContext &txc, | |
15433 | mono_clock::time_point start_throttle_acquire) | |
15434 | { | |
15435 | ceph_assert(txc.deferred_txn); | |
15436 | throttle_deferred_bytes.get(txc.cost); | |
15437 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15438 | } | |
15439 | ||
15440 | #if defined(WITH_LTTNG) | |
15441 | void BlueStore::BlueStoreThrottle::complete_kv(TransContext &txc) | |
15442 | { | |
15443 | pending_kv_ios -= 1; | |
15444 | ios_completed_since_last_traced++; | |
15445 | if (txc.tracing) { | |
15446 | tracepoint( | |
15447 | bluestore, | |
15448 | transaction_commit_latency, | |
15449 | txc.osr->get_sequencer_id(), | |
15450 | txc.seq, | |
15451 | ceph::to_seconds<double>(mono_clock::now() - txc.start)); | |
15452 | } | |
15453 | } | |
15454 | #endif | |
15455 | ||
15456 | #if defined(WITH_LTTNG) | |
15457 | void BlueStore::BlueStoreThrottle::complete(TransContext &txc) | |
15458 | { | |
15459 | if (txc.deferred_txn) { | |
15460 | pending_deferred_ios -= 1; | |
15461 | } | |
15462 | if (txc.tracing) { | |
15463 | mono_clock::time_point now = mono_clock::now(); | |
15464 | mono_clock::duration lat = now - txc.start; | |
15465 | tracepoint( | |
15466 | bluestore, | |
15467 | transaction_total_duration, | |
15468 | txc.osr->get_sequencer_id(), | |
15469 | txc.seq, | |
15470 | ceph::to_seconds<double>(lat)); | |
15471 | } | |
15472 | } | |
15473 | #endif | |
11fdf7f2 | 15474 | |
7c673cae FG |
15475 | // DB key value Histogram |
15476 | #define KEY_SLAB 32 | |
15477 | #define VALUE_SLAB 64 | |
15478 | ||
15479 | const string prefix_onode = "o"; | |
15480 | const string prefix_onode_shard = "x"; | |
15481 | const string prefix_other = "Z"; | |
15482 | ||
15483 | int BlueStore::DBHistogram::get_key_slab(size_t sz) | |
15484 | { | |
15485 | return (sz/KEY_SLAB); | |
15486 | } | |
15487 | ||
15488 | string BlueStore::DBHistogram::get_key_slab_to_range(int slab) | |
15489 | { | |
15490 | int lower_bound = slab * KEY_SLAB; | |
15491 | int upper_bound = (slab + 1) * KEY_SLAB; | |
15492 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15493 | return ret; | |
15494 | } | |
15495 | ||
15496 | int BlueStore::DBHistogram::get_value_slab(size_t sz) | |
15497 | { | |
15498 | return (sz/VALUE_SLAB); | |
15499 | } | |
15500 | ||
15501 | string BlueStore::DBHistogram::get_value_slab_to_range(int slab) | |
15502 | { | |
15503 | int lower_bound = slab * VALUE_SLAB; | |
15504 | int upper_bound = (slab + 1) * VALUE_SLAB; | |
15505 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15506 | return ret; | |
15507 | } | |
15508 | ||
15509 | void BlueStore::DBHistogram::update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
15510 | const string &prefix, size_t key_size, size_t value_size) | |
15511 | { | |
15512 | uint32_t key_slab = get_key_slab(key_size); | |
15513 | uint32_t value_slab = get_value_slab(value_size); | |
15514 | key_hist[prefix][key_slab].count++; | |
11fdf7f2 TL |
15515 | key_hist[prefix][key_slab].max_len = |
15516 | std::max<size_t>(key_size, key_hist[prefix][key_slab].max_len); | |
7c673cae FG |
15517 | key_hist[prefix][key_slab].val_map[value_slab].count++; |
15518 | key_hist[prefix][key_slab].val_map[value_slab].max_len = | |
11fdf7f2 TL |
15519 | std::max<size_t>(value_size, |
15520 | key_hist[prefix][key_slab].val_map[value_slab].max_len); | |
7c673cae FG |
15521 | } |
15522 | ||
15523 | void BlueStore::DBHistogram::dump(Formatter *f) | |
15524 | { | |
15525 | f->open_object_section("rocksdb_value_distribution"); | |
15526 | for (auto i : value_hist) { | |
15527 | f->dump_unsigned(get_value_slab_to_range(i.first).data(), i.second); | |
15528 | } | |
15529 | f->close_section(); | |
15530 | ||
15531 | f->open_object_section("rocksdb_key_value_histogram"); | |
15532 | for (auto i : key_hist) { | |
15533 | f->dump_string("prefix", i.first); | |
15534 | f->open_object_section("key_hist"); | |
15535 | for ( auto k : i.second) { | |
15536 | f->dump_unsigned(get_key_slab_to_range(k.first).data(), k.second.count); | |
15537 | f->dump_unsigned("max_len", k.second.max_len); | |
15538 | f->open_object_section("value_hist"); | |
15539 | for ( auto j : k.second.val_map) { | |
15540 | f->dump_unsigned(get_value_slab_to_range(j.first).data(), j.second.count); | |
15541 | f->dump_unsigned("max_len", j.second.max_len); | |
15542 | } | |
15543 | f->close_section(); | |
15544 | } | |
15545 | f->close_section(); | |
15546 | } | |
15547 | f->close_section(); | |
15548 | } | |
15549 | ||
15550 | //Itrerates through the db and collects the stats | |
15551 | void BlueStore::generate_db_histogram(Formatter *f) | |
15552 | { | |
15553 | //globals | |
15554 | uint64_t num_onodes = 0; | |
15555 | uint64_t num_shards = 0; | |
15556 | uint64_t num_super = 0; | |
15557 | uint64_t num_coll = 0; | |
15558 | uint64_t num_omap = 0; | |
11fdf7f2 | 15559 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
15560 | uint64_t num_deferred = 0; |
15561 | uint64_t num_alloc = 0; | |
15562 | uint64_t num_stat = 0; | |
15563 | uint64_t num_others = 0; | |
15564 | uint64_t num_shared_shards = 0; | |
15565 | size_t max_key_size =0, max_value_size = 0; | |
15566 | uint64_t total_key_size = 0, total_value_size = 0; | |
15567 | size_t key_size = 0, value_size = 0; | |
15568 | DBHistogram hist; | |
15569 | ||
11fdf7f2 | 15570 | auto start = coarse_mono_clock::now(); |
7c673cae | 15571 | |
11fdf7f2 | 15572 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
15573 | iter->seek_to_first(); |
15574 | while (iter->valid()) { | |
15575 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
15576 | key_size = iter->key_size(); | |
15577 | value_size = iter->value_size(); | |
15578 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
15579 | max_key_size = std::max(max_key_size, key_size); |
15580 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
15581 | total_key_size += key_size; |
15582 | total_value_size += value_size; | |
15583 | ||
15584 | pair<string,string> key(iter->raw_key()); | |
15585 | ||
15586 | if (key.first == PREFIX_SUPER) { | |
15587 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
15588 | num_super++; | |
15589 | } else if (key.first == PREFIX_STAT) { | |
15590 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
15591 | num_stat++; | |
15592 | } else if (key.first == PREFIX_COLL) { | |
15593 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
15594 | num_coll++; | |
15595 | } else if (key.first == PREFIX_OBJ) { | |
15596 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
15597 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
15598 | num_onodes++; | |
15599 | } else { | |
15600 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
15601 | num_shards++; | |
15602 | } | |
15603 | } else if (key.first == PREFIX_OMAP) { | |
15604 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
15605 | num_omap++; | |
11fdf7f2 TL |
15606 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
15607 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
15608 | num_pgmeta_omap++; | |
7c673cae FG |
15609 | } else if (key.first == PREFIX_DEFERRED) { |
15610 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
15611 | num_deferred++; | |
11fdf7f2 | 15612 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
15613 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
15614 | num_alloc++; | |
15615 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
15616 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
15617 | num_shared_shards++; | |
15618 | } else { | |
15619 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
15620 | num_others++; | |
15621 | } | |
15622 | iter->next(); | |
15623 | } | |
15624 | ||
11fdf7f2 | 15625 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
15626 | f->open_object_section("rocksdb_key_value_stats"); |
15627 | f->dump_unsigned("num_onodes", num_onodes); | |
15628 | f->dump_unsigned("num_shards", num_shards); | |
15629 | f->dump_unsigned("num_super", num_super); | |
15630 | f->dump_unsigned("num_coll", num_coll); | |
15631 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 15632 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
15633 | f->dump_unsigned("num_deferred", num_deferred); |
15634 | f->dump_unsigned("num_alloc", num_alloc); | |
15635 | f->dump_unsigned("num_stat", num_stat); | |
15636 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
15637 | f->dump_unsigned("num_others", num_others); | |
15638 | f->dump_unsigned("max_key_size", max_key_size); | |
15639 | f->dump_unsigned("max_value_size", max_value_size); | |
15640 | f->dump_unsigned("total_key_size", total_key_size); | |
15641 | f->dump_unsigned("total_value_size", total_value_size); | |
15642 | f->close_section(); | |
15643 | ||
15644 | hist.dump(f); | |
15645 | ||
15646 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
15647 | ||
15648 | } | |
15649 | ||
f6b5b4d7 | 15650 | void BlueStore::_shutdown_cache() |
7c673cae FG |
15651 | { |
15652 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15653 | for (auto i : buffer_cache_shards) { |
15654 | i->flush(); | |
11fdf7f2 | 15655 | ceph_assert(i->empty()); |
7c673cae FG |
15656 | } |
15657 | for (auto& p : coll_map) { | |
f6b5b4d7 | 15658 | p.second->onode_map.clear(); |
3efd9988 FG |
15659 | if (!p.second->shared_blob_set.empty()) { |
15660 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 15661 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 15662 | } |
11fdf7f2 TL |
15663 | ceph_assert(p.second->onode_map.empty()); |
15664 | ceph_assert(p.second->shared_blob_set.empty()); | |
7c673cae FG |
15665 | } |
15666 | coll_map.clear(); | |
f6b5b4d7 TL |
15667 | for (auto i : onode_cache_shards) { |
15668 | ceph_assert(i->empty()); | |
15669 | } | |
7c673cae FG |
15670 | } |
15671 | ||
31f18b77 FG |
15672 | // For external caller. |
15673 | // We use a best-effort policy instead, e.g., | |
15674 | // we don't care if there are still some pinned onodes/data in the cache | |
15675 | // after this command is completed. | |
11fdf7f2 | 15676 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
15677 | { |
15678 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15679 | for (auto i : onode_cache_shards) { |
15680 | i->flush(); | |
15681 | } | |
15682 | for (auto i : buffer_cache_shards) { | |
15683 | i->flush(); | |
31f18b77 | 15684 | } |
11fdf7f2 TL |
15685 | |
15686 | return 0; | |
31f18b77 FG |
15687 | } |
15688 | ||
7c673cae FG |
15689 | void BlueStore::_apply_padding(uint64_t head_pad, |
15690 | uint64_t tail_pad, | |
7c673cae FG |
15691 | bufferlist& padded) |
15692 | { | |
7c673cae | 15693 | if (head_pad) { |
224ce89b | 15694 | padded.prepend_zero(head_pad); |
7c673cae FG |
15695 | } |
15696 | if (tail_pad) { | |
15697 | padded.append_zero(tail_pad); | |
15698 | } | |
15699 | if (head_pad || tail_pad) { | |
15700 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
15701 | << " tail 0x" << tail_pad << std::dec << dendl; | |
15702 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
15703 | } | |
15704 | } | |
15705 | ||
11fdf7f2 TL |
15706 | void BlueStore::_record_onode(OnodeRef &o, KeyValueDB::Transaction &txn) |
15707 | { | |
15708 | // finalize extent_map shards | |
15709 | o->extent_map.update(txn, false); | |
15710 | if (o->extent_map.needs_reshard()) { | |
15711 | o->extent_map.reshard(db, txn); | |
15712 | o->extent_map.update(txn, true); | |
15713 | if (o->extent_map.needs_reshard()) { | |
15714 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
15715 | << dendl; | |
15716 | o->extent_map.clear_needs_reshard(); | |
15717 | } | |
15718 | logger->inc(l_bluestore_onode_reshard); | |
15719 | } | |
15720 | ||
15721 | // bound encode | |
15722 | size_t bound = 0; | |
15723 | denc(o->onode, bound); | |
15724 | o->extent_map.bound_encode_spanning_blobs(bound); | |
15725 | if (o->onode.extent_map_shards.empty()) { | |
15726 | denc(o->extent_map.inline_bl, bound); | |
15727 | } | |
15728 | ||
15729 | // encode | |
15730 | bufferlist bl; | |
15731 | unsigned onode_part, blob_part, extent_part; | |
15732 | { | |
15733 | auto p = bl.get_contiguous_appender(bound, true); | |
15734 | denc(o->onode, p); | |
15735 | onode_part = p.get_logical_offset(); | |
15736 | o->extent_map.encode_spanning_blobs(p); | |
15737 | blob_part = p.get_logical_offset() - onode_part; | |
15738 | if (o->onode.extent_map_shards.empty()) { | |
15739 | denc(o->extent_map.inline_bl, p); | |
15740 | } | |
15741 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
15742 | } | |
15743 | ||
15744 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
15745 | << " (" << onode_part << " bytes onode + " | |
15746 | << blob_part << " bytes spanning blobs + " | |
15747 | << extent_part << " bytes inline extents)" | |
15748 | << dendl; | |
15749 | ||
15750 | ||
15751 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
15752 | } | |
15753 | ||
15754 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
15755 | { | |
15756 | std::lock_guard l(qlock); | |
15757 | ||
81eedcae TL |
15758 | if (!disk_size_mismatch_alert.empty()) { |
15759 | alerts.emplace( | |
15760 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
15761 | disk_size_mismatch_alert); | |
15762 | } | |
15763 | if (!legacy_statfs_alert.empty()) { | |
15764 | alerts.emplace( | |
15765 | "BLUESTORE_LEGACY_STATFS", | |
15766 | legacy_statfs_alert); | |
15767 | } | |
11fdf7f2 TL |
15768 | if (!spillover_alert.empty() && |
15769 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
15770 | alerts.emplace( | |
15771 | "BLUEFS_SPILLOVER", | |
15772 | spillover_alert); | |
15773 | } | |
9f95a23c TL |
15774 | if (!no_per_pool_omap_alert.empty()) { |
15775 | alerts.emplace( | |
15776 | "BLUESTORE_NO_PER_POOL_OMAP", | |
15777 | no_per_pool_omap_alert); | |
15778 | } | |
11fdf7f2 TL |
15779 | string s0(failed_cmode); |
15780 | ||
15781 | if (!failed_compressors.empty()) { | |
15782 | if (!s0.empty()) { | |
15783 | s0 += ", "; | |
15784 | } | |
15785 | s0 += "unable to load:"; | |
15786 | bool first = true; | |
15787 | for (auto& s : failed_compressors) { | |
15788 | if (first) { | |
15789 | first = false; | |
15790 | } else { | |
15791 | s0 += ", "; | |
15792 | } | |
15793 | s0 += s; | |
15794 | } | |
15795 | alerts.emplace( | |
15796 | "BLUESTORE_NO_COMPRESSION", | |
15797 | s0); | |
15798 | } | |
15799 | } | |
15800 | ||
9f95a23c TL |
15801 | void BlueStore::_collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
15802 | size_t extents) | |
15803 | { | |
15804 | alloc_stats_count++; | |
15805 | alloc_stats_fragments += extents; | |
15806 | alloc_stats_size += need; | |
15807 | } | |
15808 | ||
15809 | void BlueStore::_record_allocation_stats() | |
15810 | { | |
15811 | // don't care about data consistency, | |
15812 | // fields can be partially modified while making the tuple | |
15813 | auto t0 = std::make_tuple( | |
15814 | alloc_stats_count.exchange(0), | |
15815 | alloc_stats_fragments.exchange(0), | |
15816 | alloc_stats_size.exchange(0)); | |
15817 | ||
15818 | dout(0) << " allocation stats probe " | |
15819 | << probe_count << ":" | |
15820 | << " cnt: " << std::get<0>(t0) | |
15821 | << " frags: " << std::get<1>(t0) | |
15822 | << " size: " << std::get<2>(t0) | |
15823 | << dendl; | |
15824 | ||
15825 | ||
15826 | // | |
15827 | // Keep the history for probes from the power-of-two sequence: | |
15828 | // -1, -2, -4, -8, -16 | |
15829 | // | |
15830 | size_t base = 1; | |
15831 | for (auto& t : alloc_stats_history) { | |
15832 | dout(0) << " probe -" | |
15833 | << base + (probe_count % base) << ": " | |
15834 | << std::get<0>(t) | |
15835 | << ", " << std::get<1>(t) | |
15836 | << ", " << std::get<2>(t) | |
15837 | << dendl; | |
15838 | base <<= 1; | |
15839 | } | |
15840 | dout(0) << "------------" << dendl; | |
15841 | ||
15842 | auto prev = probe_count++; | |
15843 | auto mask = (1 << alloc_stats_history.size()) - 1; | |
15844 | probe_count &= mask; | |
15845 | ||
15846 | for (size_t i = cbits(prev ^ probe_count) - 1; i > 0 ; --i) { | |
15847 | alloc_stats_history[i] = alloc_stats_history[i - 1]; | |
15848 | } | |
15849 | alloc_stats_history[0].swap(t0); | |
15850 | } | |
15851 | ||
7c673cae | 15852 | // =========================================== |
11fdf7f2 TL |
15853 | // BlueStoreRepairer |
15854 | ||
15855 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
15856 | const interval_set<uint64_t>& extents) | |
15857 | { | |
15858 | ceph_assert(granularity); // initialized | |
15859 | // can't call for the second time | |
15860 | ceph_assert(!was_filtered_out); | |
15861 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
15862 | ||
15863 | uint64_t prev_pos = 0; | |
15864 | uint64_t npos = collections_bfs.size(); | |
15865 | ||
15866 | bloom_vector collections_reduced; | |
15867 | bloom_vector objects_reduced; | |
15868 | ||
15869 | for (auto e : extents) { | |
15870 | if (e.second == 0) { | |
15871 | continue; | |
15872 | } | |
15873 | uint64_t pos = max(e.first / granularity, prev_pos); | |
15874 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
15875 | while (pos != npos && pos < end_pos) { | |
15876 | ceph_assert( collections_bfs[pos].element_count() == | |
15877 | objects_bfs[pos].element_count()); | |
15878 | if (collections_bfs[pos].element_count()) { | |
15879 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
15880 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
15881 | } | |
15882 | ++pos; | |
15883 | } | |
15884 | prev_pos = end_pos; | |
15885 | } | |
15886 | collections_reduced.swap(collections_bfs); | |
15887 | objects_reduced.swap(objects_bfs); | |
15888 | was_filtered_out = true; | |
15889 | return collections_bfs.size(); | |
15890 | } | |
15891 | ||
15892 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
15893 | const string& prefix, | |
15894 | const string& key) | |
15895 | { | |
15896 | if (!remove_key_txn) { | |
15897 | remove_key_txn = db->get_transaction(); | |
15898 | } | |
15899 | ++to_repair_cnt; | |
15900 | remove_key_txn->rmkey(prefix, key); | |
15901 | ||
15902 | return true; | |
15903 | } | |
15904 | ||
9f95a23c TL |
15905 | void BlueStoreRepairer::fix_per_pool_omap(KeyValueDB *db) |
15906 | { | |
15907 | fix_per_pool_omap_txn = db->get_transaction(); | |
15908 | ++to_repair_cnt; | |
15909 | bufferlist bl; | |
15910 | bl.append("1"); | |
15911 | fix_per_pool_omap_txn->set(PREFIX_SUPER, "per_pool_omap", bl); | |
15912 | } | |
15913 | ||
11fdf7f2 TL |
15914 | bool BlueStoreRepairer::fix_shared_blob( |
15915 | KeyValueDB *db, | |
15916 | uint64_t sbid, | |
15917 | const bufferlist* bl) | |
15918 | { | |
15919 | KeyValueDB::Transaction txn; | |
15920 | if (fix_misreferences_txn) { // reuse this txn | |
15921 | txn = fix_misreferences_txn; | |
15922 | } else { | |
15923 | if (!fix_shared_blob_txn) { | |
15924 | fix_shared_blob_txn = db->get_transaction(); | |
15925 | } | |
15926 | txn = fix_shared_blob_txn; | |
15927 | } | |
15928 | string key; | |
15929 | get_shared_blob_key(sbid, &key); | |
15930 | ||
15931 | ++to_repair_cnt; | |
15932 | if (bl) { | |
15933 | txn->set(PREFIX_SHARED_BLOB, key, *bl); | |
15934 | } else { | |
15935 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
15936 | } | |
15937 | return true; | |
15938 | } | |
15939 | ||
15940 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
15941 | const string& key, | |
15942 | const store_statfs_t& new_statfs) | |
15943 | { | |
15944 | if (!fix_statfs_txn) { | |
15945 | fix_statfs_txn = db->get_transaction(); | |
15946 | } | |
15947 | BlueStore::volatile_statfs vstatfs; | |
15948 | vstatfs = new_statfs; | |
15949 | bufferlist bl; | |
15950 | vstatfs.encode(bl); | |
15951 | ++to_repair_cnt; | |
15952 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
15953 | return true; | |
15954 | } | |
15955 | ||
15956 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
15957 | FreelistManager* fm, | |
15958 | uint64_t offset, uint64_t len) | |
15959 | { | |
15960 | if (!fix_fm_leaked_txn) { | |
15961 | fix_fm_leaked_txn = db->get_transaction(); | |
15962 | } | |
15963 | ++to_repair_cnt; | |
15964 | fm->release(offset, len, fix_fm_leaked_txn); | |
15965 | return true; | |
15966 | } | |
15967 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
15968 | FreelistManager* fm, | |
15969 | uint64_t offset, uint64_t len) | |
15970 | { | |
15971 | if (!fix_fm_false_free_txn) { | |
15972 | fix_fm_false_free_txn = db->get_transaction(); | |
15973 | } | |
15974 | ++to_repair_cnt; | |
15975 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
15976 | return true; | |
15977 | } | |
15978 | ||
15979 | bool BlueStoreRepairer::fix_bluefs_extents(std::atomic<uint64_t>& out_of_sync_flag) | |
15980 | { | |
15981 | // this is just a stub to count num of repairs properly, | |
15982 | // actual repair happens in BlueStore::_close_db_and_around() | |
15983 | // while doing _sync_bluefs_and_fm | |
15984 | ++out_of_sync_flag; | |
15985 | ++to_repair_cnt; | |
15986 | return true; | |
15987 | } | |
15988 | ||
15989 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) | |
15990 | { | |
15991 | if (misreferenced_extents.size()) { | |
15992 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
15993 | ceph_assert(n > 0); | |
15994 | if (!fix_misreferences_txn) { | |
15995 | fix_misreferences_txn = db->get_transaction(); | |
15996 | } | |
15997 | return true; | |
15998 | } | |
15999 | return false; | |
16000 | } | |
16001 | ||
16002 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
16003 | { | |
9f95a23c TL |
16004 | if (fix_per_pool_omap_txn) { |
16005 | db->submit_transaction_sync(fix_per_pool_omap_txn); | |
16006 | fix_per_pool_omap_txn = nullptr; | |
16007 | } | |
11fdf7f2 TL |
16008 | if (fix_fm_leaked_txn) { |
16009 | db->submit_transaction_sync(fix_fm_leaked_txn); | |
16010 | fix_fm_leaked_txn = nullptr; | |
16011 | } | |
16012 | if (fix_fm_false_free_txn) { | |
16013 | db->submit_transaction_sync(fix_fm_false_free_txn); | |
16014 | fix_fm_false_free_txn = nullptr; | |
16015 | } | |
16016 | if (remove_key_txn) { | |
16017 | db->submit_transaction_sync(remove_key_txn); | |
16018 | remove_key_txn = nullptr; | |
16019 | } | |
16020 | if (fix_misreferences_txn) { | |
16021 | db->submit_transaction_sync(fix_misreferences_txn); | |
16022 | fix_misreferences_txn = nullptr; | |
16023 | } | |
16024 | if (fix_shared_blob_txn) { | |
16025 | db->submit_transaction_sync(fix_shared_blob_txn); | |
16026 | fix_shared_blob_txn = nullptr; | |
16027 | } | |
16028 | ||
16029 | if (fix_statfs_txn) { | |
16030 | db->submit_transaction_sync(fix_statfs_txn); | |
16031 | fix_statfs_txn = nullptr; | |
16032 | } | |
16033 | unsigned repaired = to_repair_cnt; | |
16034 | to_repair_cnt = 0; | |
16035 | return repaired; | |
16036 | } | |
16037 | ||
16038 | // ======================================================= | |
9f95a23c TL |
16039 | // RocksDBBlueFSVolumeSelector |
16040 | ||
16041 | uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) { | |
16042 | ceph_assert(h != nullptr); | |
16043 | uint64_t hint = reinterpret_cast<uint64_t>(h); | |
16044 | uint8_t res; | |
16045 | switch (hint) { | |
16046 | case LEVEL_SLOW: | |
16047 | res = BlueFS::BDEV_SLOW; | |
16048 | if (db_avail4slow > 0) { | |
16049 | // considering statically available db space vs. | |
16050 | // - observed maximums on DB dev for DB/WAL/UNSORTED data | |
16051 | // - observed maximum spillovers | |
16052 | uint64_t max_db_use = 0; // max db usage we potentially observed | |
f6b5b4d7 | 16053 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_LOG - LEVEL_FIRST); |
9f95a23c TL |
16054 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_WAL - LEVEL_FIRST); |
16055 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_DB - LEVEL_FIRST); | |
16056 | // this could go to db hence using it in the estimation | |
16057 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_SLOW, LEVEL_DB - LEVEL_FIRST); | |
16058 | ||
16059 | auto db_total = l_totals[LEVEL_DB - LEVEL_FIRST]; | |
16060 | uint64_t avail = min( | |
16061 | db_avail4slow, | |
16062 | max_db_use < db_total ? db_total - max_db_use : 0); | |
16063 | ||
16064 | // considering current DB dev usage for SLOW data | |
16065 | if (avail > per_level_per_dev_usage.at(BlueFS::BDEV_DB, LEVEL_SLOW - LEVEL_FIRST)) { | |
16066 | res = BlueFS::BDEV_DB; | |
16067 | } | |
16068 | } | |
16069 | break; | |
f6b5b4d7 | 16070 | case LEVEL_LOG: |
9f95a23c TL |
16071 | case LEVEL_WAL: |
16072 | res = BlueFS::BDEV_WAL; | |
16073 | break; | |
16074 | case LEVEL_DB: | |
16075 | default: | |
16076 | res = BlueFS::BDEV_DB; | |
16077 | break; | |
16078 | } | |
16079 | return res; | |
16080 | } | |
16081 | ||
16082 | void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const | |
16083 | { | |
16084 | res.emplace_back(base, l_totals[LEVEL_DB - LEVEL_FIRST]); | |
16085 | res.emplace_back(base + ".slow", l_totals[LEVEL_SLOW - LEVEL_FIRST]); | |
16086 | } | |
16087 | ||
16088 | void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(const string& dirname) const { | |
16089 | uint8_t res = LEVEL_DB; | |
16090 | if (dirname.length() > 5) { | |
16091 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
16092 | // match up with bluestore. the slow device is always the second | |
16093 | // one (when a dedicated block.db device is present and used at | |
16094 | // bdev 0). the wal device is always last. | |
16095 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
16096 | res = LEVEL_SLOW; | |
16097 | } | |
16098 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
16099 | res = LEVEL_WAL; | |
16100 | } | |
16101 | } | |
16102 | return reinterpret_cast<void*>(res); | |
16103 | } | |
16104 | ||
16105 | void RocksDBBlueFSVolumeSelector::dump(ostream& sout) { | |
16106 | auto max_x = per_level_per_dev_usage.get_max_x(); | |
16107 | auto max_y = per_level_per_dev_usage.get_max_y(); | |
16108 | sout << "RocksDBBlueFSVolumeSelector: wal_total:" << l_totals[LEVEL_WAL - LEVEL_FIRST] | |
16109 | << ", db_total:" << l_totals[LEVEL_DB - LEVEL_FIRST] | |
16110 | << ", slow_total:" << l_totals[LEVEL_SLOW - LEVEL_FIRST] | |
16111 | << ", db_avail:" << db_avail4slow << std::endl | |
16112 | << "Usage matrix:" << std::endl; | |
f6b5b4d7 | 16113 | constexpr std::array<const char*, 8> names{ { |
9f95a23c TL |
16114 | "DEV/LEV", |
16115 | "WAL", | |
16116 | "DB", | |
16117 | "SLOW", | |
16118 | "*", | |
16119 | "*", | |
f6b5b4d7 TL |
16120 | "REAL", |
16121 | "FILES", | |
9f95a23c TL |
16122 | } }; |
16123 | const size_t width = 12; | |
16124 | for (size_t i = 0; i < names.size(); ++i) { | |
16125 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16126 | sout.width(width); | |
16127 | sout << names[i]; | |
16128 | } | |
16129 | sout << std::endl; | |
16130 | for (size_t l = 0; l < max_y; l++) { | |
16131 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16132 | sout.width(width); | |
16133 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
16134 | case LEVEL_LOG: |
16135 | sout << "LOG"; break; | |
9f95a23c TL |
16136 | case LEVEL_WAL: |
16137 | sout << "WAL"; break; | |
16138 | case LEVEL_DB: | |
16139 | sout << "DB"; break; | |
16140 | case LEVEL_SLOW: | |
16141 | sout << "SLOW"; break; | |
16142 | case LEVEL_MAX: | |
16143 | sout << "TOTALS"; break; | |
16144 | } | |
f6b5b4d7 | 16145 | for (size_t d = 0; d < max_x; d++) { |
9f95a23c TL |
16146 | sout.setf(std::ios::left, std::ios::adjustfield); |
16147 | sout.width(width); | |
16148 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(d, l))); | |
16149 | } | |
16150 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16151 | sout.width(width); | |
f6b5b4d7 | 16152 | sout << stringify(per_level_files[l]) << std::endl; |
9f95a23c TL |
16153 | } |
16154 | ceph_assert(max_x == per_level_per_dev_max.get_max_x()); | |
16155 | ceph_assert(max_y == per_level_per_dev_max.get_max_y()); | |
16156 | sout << "MAXIMUMS:" << std::endl; | |
16157 | for (size_t l = 0; l < max_y; l++) { | |
16158 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16159 | sout.width(width); | |
16160 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
16161 | case LEVEL_LOG: |
16162 | sout << "LOG"; break; | |
9f95a23c TL |
16163 | case LEVEL_WAL: |
16164 | sout << "WAL"; break; | |
16165 | case LEVEL_DB: | |
16166 | sout << "DB"; break; | |
16167 | case LEVEL_SLOW: | |
16168 | sout << "SLOW"; break; | |
16169 | case LEVEL_MAX: | |
16170 | sout << "TOTALS"; break; | |
16171 | } | |
16172 | for (size_t d = 0; d < max_x - 1; d++) { | |
16173 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16174 | sout.width(width); | |
16175 | sout << stringify(byte_u_t(per_level_per_dev_max.at(d, l))); | |
16176 | } | |
16177 | sout.setf(std::ios::left, std::ios::adjustfield); | |
16178 | sout.width(width); | |
16179 | sout << stringify(byte_u_t(per_level_per_dev_max.at(max_x - 1, l))); | |
16180 | if (l < max_y - 1) { | |
16181 | sout << std::endl; | |
16182 | } | |
16183 | } | |
16184 | } | |
11fdf7f2 | 16185 | |
9f95a23c | 16186 | // ======================================================= |