]>
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 | ||
31f18b77 FG |
21 | #include "include/cpp-btree/btree_set.h" |
22 | ||
7c673cae FG |
23 | #include "BlueStore.h" |
24 | #include "os/kv.h" | |
25 | #include "include/compat.h" | |
26 | #include "include/intarith.h" | |
27 | #include "include/stringify.h" | |
11fdf7f2 TL |
28 | #include "include/str_map.h" |
29 | #include "include/util.h" | |
7c673cae FG |
30 | #include "common/errno.h" |
31 | #include "common/safe_io.h" | |
91327a77 | 32 | #include "common/PriorityCache.h" |
7c673cae FG |
33 | #include "Allocator.h" |
34 | #include "FreelistManager.h" | |
35 | #include "BlueFS.h" | |
36 | #include "BlueRocksEnv.h" | |
37 | #include "auth/Crypto.h" | |
38 | #include "common/EventTrace.h" | |
91327a77 | 39 | #include "perfglue/heap_profiler.h" |
11fdf7f2 TL |
40 | #include "common/blkdev.h" |
41 | #include "common/numa.h" | |
7c673cae FG |
42 | |
43 | #define dout_context cct | |
44 | #define dout_subsys ceph_subsys_bluestore | |
45 | ||
31f18b77 FG |
46 | using bid_t = decltype(BlueStore::Blob::id); |
47 | ||
48 | // bluestore_cache_onode | |
7c673cae | 49 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Onode, bluestore_onode, |
31f18b77 | 50 | bluestore_cache_onode); |
7c673cae | 51 | |
31f18b77 | 52 | // bluestore_cache_other |
7c673cae | 53 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Buffer, bluestore_buffer, |
31f18b77 | 54 | bluestore_cache_other); |
7c673cae | 55 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Extent, bluestore_extent, |
31f18b77 | 56 | bluestore_cache_other); |
7c673cae | 57 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Blob, bluestore_blob, |
31f18b77 | 58 | bluestore_cache_other); |
7c673cae | 59 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::SharedBlob, bluestore_shared_blob, |
31f18b77 FG |
60 | bluestore_cache_other); |
61 | ||
62 | // bluestore_txc | |
63 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::TransContext, bluestore_transcontext, | |
64 | bluestore_txc); | |
65 | ||
7c673cae FG |
66 | |
67 | // kv store prefixes | |
11fdf7f2 TL |
68 | const string PREFIX_SUPER = "S"; // field -> value |
69 | const string PREFIX_STAT = "T"; // field -> value(int64 array) | |
70 | const string PREFIX_COLL = "C"; // collection name -> cnode_t | |
71 | const string PREFIX_OBJ = "O"; // object name -> onode_t | |
72 | const string PREFIX_OMAP = "M"; // u64 + keyname -> value | |
73 | const string PREFIX_PGMETA_OMAP = "P"; // u64 + keyname -> value(for meta coll) | |
74 | const string PREFIX_DEFERRED = "L"; // id -> deferred_transaction_t | |
75 | const string PREFIX_ALLOC = "B"; // u64 offset -> u64 length (freelist) | |
76 | const string PREFIX_ALLOC_BITMAP = "b";// (see BitmapFreelistManager) | |
7c673cae FG |
77 | const string PREFIX_SHARED_BLOB = "X"; // u64 offset -> shared_blob_t |
78 | ||
11fdf7f2 TL |
79 | const string BLUESTORE_GLOBAL_STATFS_KEY = "bluestore_statfs"; |
80 | ||
7c673cae FG |
81 | // write a label in the first block. always use this size. note that |
82 | // bluefs makes a matching assumption about the location of its | |
83 | // superblock (always the second block of the device). | |
84 | #define BDEV_LABEL_BLOCK_SIZE 4096 | |
85 | ||
86 | // reserve: label (4k) + bluefs super (4k), which means we start at 8k. | |
87 | #define SUPER_RESERVED 8192 | |
88 | ||
89 | #define OBJECT_MAX_SIZE 0xffffffff // 32 bits | |
90 | ||
91 | ||
92 | /* | |
93 | * extent map blob encoding | |
94 | * | |
95 | * we use the low bits of the blobid field to indicate some common scenarios | |
96 | * and spanning vs local ids. See ExtentMap::{encode,decode}_some(). | |
97 | */ | |
98 | #define BLOBID_FLAG_CONTIGUOUS 0x1 // this extent starts at end of previous | |
99 | #define BLOBID_FLAG_ZEROOFFSET 0x2 // blob_offset is 0 | |
100 | #define BLOBID_FLAG_SAMELENGTH 0x4 // length matches previous extent | |
101 | #define BLOBID_FLAG_SPANNING 0x8 // has spanning blob id | |
102 | #define BLOBID_SHIFT_BITS 4 | |
103 | ||
104 | /* | |
105 | * object name key structure | |
106 | * | |
107 | * encoded u8: shard + 2^7 (so that it sorts properly) | |
108 | * encoded u64: poolid + 2^63 (so that it sorts properly) | |
109 | * encoded u32: hash (bit reversed) | |
110 | * | |
111 | * escaped string: namespace | |
112 | * | |
113 | * escaped string: key or object name | |
114 | * 1 char: '<', '=', or '>'. if =, then object key == object name, and | |
115 | * we are done. otherwise, we are followed by the object name. | |
116 | * escaped string: object name (unless '=' above) | |
117 | * | |
118 | * encoded u64: snap | |
119 | * encoded u64: generation | |
120 | * 'o' | |
121 | */ | |
122 | #define ONODE_KEY_SUFFIX 'o' | |
123 | ||
124 | /* | |
125 | * extent shard key | |
126 | * | |
127 | * object prefix key | |
128 | * u32 | |
129 | * 'x' | |
130 | */ | |
131 | #define EXTENT_SHARD_KEY_SUFFIX 'x' | |
132 | ||
11fdf7f2 TL |
133 | // FIXME minor: make a BlueStore method once we have effecient way to |
134 | // map idx to counter nickname | |
135 | #define LOG_LATENCY_I(logger, cct, idx, v, info) { \ | |
136 | ceph::timespan lat = v; \ | |
137 | logger->tinc(idx, lat); \ | |
138 | if (cct->_conf->bluestore_log_op_age > 0.0 && \ | |
139 | lat >= make_timespan(cct->_conf->bluestore_log_op_age)) { \ | |
140 | dout(0) << __func__ << " slow operation observed for " #idx \ | |
141 | << ", latency = " << lat \ | |
142 | << info \ | |
143 | << dendl; \ | |
144 | } \ | |
145 | } | |
146 | ||
147 | #define LOG_LATENCY_FN(logger, cct, idx, v, fn) { \ | |
148 | ceph::timespan lat = v; \ | |
149 | logger->tinc(idx, lat); \ | |
150 | if (cct->_conf->bluestore_log_op_age > 0.0 && \ | |
151 | lat >= make_timespan(cct->_conf->bluestore_log_op_age)) { \ | |
152 | dout(0) << __func__ << " slow operation observed for " #idx \ | |
153 | << ", latency = " << lat \ | |
154 | << fn(lat) \ | |
155 | << dendl; \ | |
156 | } \ | |
157 | } | |
158 | ||
159 | #define LOG_LATENCY(logger, cct, idx, v) LOG_LATENCY_I(logger, cct, idx, v, "") | |
160 | ||
7c673cae FG |
161 | /* |
162 | * string encoding in the key | |
163 | * | |
164 | * The key string needs to lexicographically sort the same way that | |
165 | * ghobject_t does. We do this by escaping anything <= to '#' with # | |
166 | * plus a 2 digit hex string, and anything >= '~' with ~ plus the two | |
167 | * hex digits. | |
168 | * | |
169 | * We use ! as a terminator for strings; this works because it is < # | |
170 | * and will get escaped if it is present in the string. | |
171 | * | |
172 | */ | |
173 | template<typename S> | |
174 | static void append_escaped(const string &in, S *out) | |
175 | { | |
224ce89b WB |
176 | char hexbyte[in.length() * 3 + 1]; |
177 | char* ptr = &hexbyte[0]; | |
7c673cae FG |
178 | for (string::const_iterator i = in.begin(); i != in.end(); ++i) { |
179 | if (*i <= '#') { | |
224ce89b WB |
180 | *ptr++ = '#'; |
181 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
182 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 183 | } else if (*i >= '~') { |
224ce89b WB |
184 | *ptr++ = '~'; |
185 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
186 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 187 | } else { |
224ce89b | 188 | *ptr++ = *i; |
7c673cae FG |
189 | } |
190 | } | |
224ce89b WB |
191 | *ptr++ = '!'; |
192 | out->append(hexbyte, ptr - &hexbyte[0]); | |
193 | } | |
194 | ||
195 | inline unsigned h2i(char c) | |
196 | { | |
197 | if ((c >= '0') && (c <= '9')) { | |
198 | return c - 0x30; | |
199 | } else if ((c >= 'a') && (c <= 'f')) { | |
200 | return c - 'a' + 10; | |
201 | } else if ((c >= 'A') && (c <= 'F')) { | |
202 | return c - 'A' + 10; | |
203 | } else { | |
204 | return 256; // make it always larger than 255 | |
205 | } | |
7c673cae FG |
206 | } |
207 | ||
208 | static int decode_escaped(const char *p, string *out) | |
209 | { | |
224ce89b WB |
210 | char buff[256]; |
211 | char* ptr = &buff[0]; | |
212 | char* max = &buff[252]; | |
7c673cae FG |
213 | const char *orig_p = p; |
214 | while (*p && *p != '!') { | |
215 | if (*p == '#' || *p == '~') { | |
224ce89b WB |
216 | unsigned hex = 0; |
217 | p++; | |
218 | hex = h2i(*p++) << 4; | |
219 | if (hex > 255) { | |
220 | return -EINVAL; | |
221 | } | |
222 | hex |= h2i(*p++); | |
223 | if (hex > 255) { | |
224 | return -EINVAL; | |
225 | } | |
226 | *ptr++ = hex; | |
7c673cae | 227 | } else { |
224ce89b WB |
228 | *ptr++ = *p++; |
229 | } | |
230 | if (ptr > max) { | |
231 | out->append(buff, ptr-buff); | |
232 | ptr = &buff[0]; | |
7c673cae FG |
233 | } |
234 | } | |
224ce89b WB |
235 | if (ptr != buff) { |
236 | out->append(buff, ptr-buff); | |
237 | } | |
7c673cae FG |
238 | return p - orig_p; |
239 | } | |
240 | ||
241 | // some things we encode in binary (as le32 or le64); print the | |
242 | // resulting key strings nicely | |
243 | template<typename S> | |
244 | static string pretty_binary_string(const S& in) | |
245 | { | |
246 | char buf[10]; | |
247 | string out; | |
248 | out.reserve(in.length() * 3); | |
249 | enum { NONE, HEX, STRING } mode = NONE; | |
250 | unsigned from = 0, i; | |
251 | for (i=0; i < in.length(); ++i) { | |
252 | if ((in[i] < 32 || (unsigned char)in[i] > 126) || | |
253 | (mode == HEX && in.length() - i >= 4 && | |
254 | ((in[i] < 32 || (unsigned char)in[i] > 126) || | |
255 | (in[i+1] < 32 || (unsigned char)in[i+1] > 126) || | |
256 | (in[i+2] < 32 || (unsigned char)in[i+2] > 126) || | |
257 | (in[i+3] < 32 || (unsigned char)in[i+3] > 126)))) { | |
258 | if (mode == STRING) { | |
259 | out.append(in.c_str() + from, i - from); | |
260 | out.push_back('\''); | |
261 | } | |
262 | if (mode != HEX) { | |
263 | out.append("0x"); | |
264 | mode = HEX; | |
265 | } | |
266 | if (in.length() - i >= 4) { | |
267 | // print a whole u32 at once | |
268 | snprintf(buf, sizeof(buf), "%08x", | |
269 | (uint32_t)(((unsigned char)in[i] << 24) | | |
270 | ((unsigned char)in[i+1] << 16) | | |
271 | ((unsigned char)in[i+2] << 8) | | |
272 | ((unsigned char)in[i+3] << 0))); | |
273 | i += 3; | |
274 | } else { | |
275 | snprintf(buf, sizeof(buf), "%02x", (int)(unsigned char)in[i]); | |
276 | } | |
277 | out.append(buf); | |
278 | } else { | |
279 | if (mode != STRING) { | |
280 | out.push_back('\''); | |
281 | mode = STRING; | |
282 | from = i; | |
283 | } | |
284 | } | |
285 | } | |
286 | if (mode == STRING) { | |
287 | out.append(in.c_str() + from, i - from); | |
288 | out.push_back('\''); | |
289 | } | |
290 | return out; | |
291 | } | |
292 | ||
293 | template<typename T> | |
294 | static void _key_encode_shard(shard_id_t shard, T *key) | |
295 | { | |
296 | key->push_back((char)((uint8_t)shard.id + (uint8_t)0x80)); | |
297 | } | |
298 | ||
299 | static const char *_key_decode_shard(const char *key, shard_id_t *pshard) | |
300 | { | |
301 | pshard->id = (uint8_t)*key - (uint8_t)0x80; | |
302 | return key + 1; | |
303 | } | |
304 | ||
305 | static void get_coll_key_range(const coll_t& cid, int bits, | |
306 | string *temp_start, string *temp_end, | |
307 | string *start, string *end) | |
308 | { | |
309 | temp_start->clear(); | |
310 | temp_end->clear(); | |
311 | start->clear(); | |
312 | end->clear(); | |
313 | ||
314 | spg_t pgid; | |
315 | if (cid.is_pg(&pgid)) { | |
316 | _key_encode_shard(pgid.shard, start); | |
317 | *temp_start = *start; | |
318 | ||
319 | _key_encode_u64(pgid.pool() + 0x8000000000000000ull, start); | |
320 | _key_encode_u64((-2ll - pgid.pool()) + 0x8000000000000000ull, temp_start); | |
321 | ||
322 | *end = *start; | |
323 | *temp_end = *temp_start; | |
324 | ||
325 | uint32_t reverse_hash = hobject_t::_reverse_bits(pgid.ps()); | |
326 | _key_encode_u32(reverse_hash, start); | |
327 | _key_encode_u32(reverse_hash, temp_start); | |
328 | ||
329 | uint64_t end_hash = reverse_hash + (1ull << (32 - bits)); | |
330 | if (end_hash > 0xffffffffull) | |
331 | end_hash = 0xffffffffull; | |
332 | ||
333 | _key_encode_u32(end_hash, end); | |
334 | _key_encode_u32(end_hash, temp_end); | |
335 | } else { | |
336 | _key_encode_shard(shard_id_t::NO_SHARD, start); | |
337 | _key_encode_u64(-1ull + 0x8000000000000000ull, start); | |
338 | *end = *start; | |
339 | _key_encode_u32(0, start); | |
340 | _key_encode_u32(0xffffffff, end); | |
341 | ||
342 | // no separate temp section | |
343 | *temp_start = *end; | |
344 | *temp_end = *end; | |
345 | } | |
346 | } | |
347 | ||
348 | static void get_shared_blob_key(uint64_t sbid, string *key) | |
349 | { | |
350 | key->clear(); | |
351 | _key_encode_u64(sbid, key); | |
352 | } | |
353 | ||
354 | static int get_key_shared_blob(const string& key, uint64_t *sbid) | |
355 | { | |
356 | const char *p = key.c_str(); | |
357 | if (key.length() < sizeof(uint64_t)) | |
358 | return -1; | |
224ce89b | 359 | _key_decode_u64(p, sbid); |
7c673cae FG |
360 | return 0; |
361 | } | |
362 | ||
363 | template<typename S> | |
364 | static int get_key_object(const S& key, ghobject_t *oid) | |
365 | { | |
366 | int r; | |
367 | const char *p = key.c_str(); | |
368 | ||
369 | if (key.length() < 1 + 8 + 4) | |
370 | return -1; | |
371 | p = _key_decode_shard(p, &oid->shard_id); | |
372 | ||
373 | uint64_t pool; | |
374 | p = _key_decode_u64(p, &pool); | |
375 | oid->hobj.pool = pool - 0x8000000000000000ull; | |
376 | ||
377 | unsigned hash; | |
378 | p = _key_decode_u32(p, &hash); | |
379 | ||
380 | oid->hobj.set_bitwise_key_u32(hash); | |
381 | ||
382 | r = decode_escaped(p, &oid->hobj.nspace); | |
383 | if (r < 0) | |
384 | return -2; | |
385 | p += r + 1; | |
386 | ||
387 | string k; | |
388 | r = decode_escaped(p, &k); | |
389 | if (r < 0) | |
390 | return -3; | |
391 | p += r + 1; | |
392 | if (*p == '=') { | |
393 | // no key | |
394 | ++p; | |
395 | oid->hobj.oid.name = k; | |
396 | } else if (*p == '<' || *p == '>') { | |
397 | // key + name | |
398 | ++p; | |
399 | r = decode_escaped(p, &oid->hobj.oid.name); | |
400 | if (r < 0) | |
401 | return -5; | |
402 | p += r + 1; | |
403 | oid->hobj.set_key(k); | |
404 | } else { | |
405 | // malformed | |
406 | return -6; | |
407 | } | |
408 | ||
409 | p = _key_decode_u64(p, &oid->hobj.snap.val); | |
410 | p = _key_decode_u64(p, &oid->generation); | |
411 | ||
412 | if (*p != ONODE_KEY_SUFFIX) { | |
413 | return -7; | |
414 | } | |
415 | p++; | |
416 | if (*p) { | |
417 | // if we get something other than a null terminator here, | |
418 | // something goes wrong. | |
419 | return -8; | |
420 | } | |
421 | ||
422 | return 0; | |
423 | } | |
424 | ||
425 | template<typename S> | |
426 | static void get_object_key(CephContext *cct, const ghobject_t& oid, S *key) | |
427 | { | |
428 | key->clear(); | |
429 | ||
430 | size_t max_len = 1 + 8 + 4 + | |
431 | (oid.hobj.nspace.length() * 3 + 1) + | |
432 | (oid.hobj.get_key().length() * 3 + 1) + | |
433 | 1 + // for '<', '=', or '>' | |
434 | (oid.hobj.oid.name.length() * 3 + 1) + | |
435 | 8 + 8 + 1; | |
436 | key->reserve(max_len); | |
437 | ||
438 | _key_encode_shard(oid.shard_id, key); | |
439 | _key_encode_u64(oid.hobj.pool + 0x8000000000000000ull, key); | |
440 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), key); | |
441 | ||
442 | append_escaped(oid.hobj.nspace, key); | |
443 | ||
444 | if (oid.hobj.get_key().length()) { | |
445 | // is a key... could be < = or >. | |
446 | append_escaped(oid.hobj.get_key(), key); | |
447 | // (ASCII chars < = and > sort in that order, yay) | |
448 | int r = oid.hobj.get_key().compare(oid.hobj.oid.name); | |
449 | if (r) { | |
450 | key->append(r > 0 ? ">" : "<"); | |
451 | append_escaped(oid.hobj.oid.name, key); | |
452 | } else { | |
453 | // same as no key | |
454 | key->append("="); | |
455 | } | |
456 | } else { | |
457 | // no key | |
458 | append_escaped(oid.hobj.oid.name, key); | |
459 | key->append("="); | |
460 | } | |
461 | ||
462 | _key_encode_u64(oid.hobj.snap, key); | |
463 | _key_encode_u64(oid.generation, key); | |
464 | ||
465 | key->push_back(ONODE_KEY_SUFFIX); | |
466 | ||
467 | // sanity check | |
468 | if (true) { | |
469 | ghobject_t t; | |
470 | int r = get_key_object(*key, &t); | |
471 | if (r || t != oid) { | |
472 | derr << " r " << r << dendl; | |
473 | derr << "key " << pretty_binary_string(*key) << dendl; | |
474 | derr << "oid " << oid << dendl; | |
475 | derr << " t " << t << dendl; | |
11fdf7f2 | 476 | ceph_assert(r == 0 && t == oid); |
7c673cae FG |
477 | } |
478 | } | |
479 | } | |
480 | ||
481 | ||
482 | // extent shard keys are the onode key, plus a u32, plus 'x'. the trailing | |
483 | // char lets us quickly test whether it is a shard key without decoding any | |
484 | // of the prefix bytes. | |
485 | template<typename S> | |
486 | static void get_extent_shard_key(const S& onode_key, uint32_t offset, | |
487 | string *key) | |
488 | { | |
489 | key->clear(); | |
490 | key->reserve(onode_key.length() + 4 + 1); | |
491 | key->append(onode_key.c_str(), onode_key.size()); | |
492 | _key_encode_u32(offset, key); | |
493 | key->push_back(EXTENT_SHARD_KEY_SUFFIX); | |
494 | } | |
495 | ||
496 | static void rewrite_extent_shard_key(uint32_t offset, string *key) | |
497 | { | |
11fdf7f2 TL |
498 | ceph_assert(key->size() > sizeof(uint32_t) + 1); |
499 | ceph_assert(*key->rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
500 | _key_encode_u32(offset, key->size() - sizeof(uint32_t) - 1, key); |
501 | } | |
502 | ||
503 | template<typename S> | |
504 | static void generate_extent_shard_key_and_apply( | |
505 | const S& onode_key, | |
506 | uint32_t offset, | |
507 | string *key, | |
508 | std::function<void(const string& final_key)> apply) | |
509 | { | |
510 | if (key->empty()) { // make full key | |
11fdf7f2 | 511 | ceph_assert(!onode_key.empty()); |
7c673cae FG |
512 | get_extent_shard_key(onode_key, offset, key); |
513 | } else { | |
514 | rewrite_extent_shard_key(offset, key); | |
515 | } | |
516 | apply(*key); | |
517 | } | |
518 | ||
519 | int get_key_extent_shard(const string& key, string *onode_key, uint32_t *offset) | |
520 | { | |
11fdf7f2 TL |
521 | ceph_assert(key.size() > sizeof(uint32_t) + 1); |
522 | ceph_assert(*key.rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
523 | int okey_len = key.size() - sizeof(uint32_t) - 1; |
524 | *onode_key = key.substr(0, okey_len); | |
525 | const char *p = key.data() + okey_len; | |
224ce89b | 526 | _key_decode_u32(p, offset); |
7c673cae FG |
527 | return 0; |
528 | } | |
529 | ||
530 | static bool is_extent_shard_key(const string& key) | |
531 | { | |
532 | return *key.rbegin() == EXTENT_SHARD_KEY_SUFFIX; | |
533 | } | |
534 | ||
535 | // '-' < '.' < '~' | |
536 | static void get_omap_header(uint64_t id, string *out) | |
537 | { | |
538 | _key_encode_u64(id, out); | |
539 | out->push_back('-'); | |
540 | } | |
541 | ||
542 | // hmm, I don't think there's any need to escape the user key since we | |
543 | // have a clean prefix. | |
544 | static void get_omap_key(uint64_t id, const string& key, string *out) | |
545 | { | |
546 | _key_encode_u64(id, out); | |
547 | out->push_back('.'); | |
548 | out->append(key); | |
549 | } | |
550 | ||
551 | static void rewrite_omap_key(uint64_t id, string old, string *out) | |
552 | { | |
553 | _key_encode_u64(id, out); | |
554 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
555 | } | |
556 | ||
557 | static void decode_omap_key(const string& key, string *user_key) | |
558 | { | |
559 | *user_key = key.substr(sizeof(uint64_t) + 1); | |
560 | } | |
561 | ||
562 | static void get_omap_tail(uint64_t id, string *out) | |
563 | { | |
564 | _key_encode_u64(id, out); | |
565 | out->push_back('~'); | |
566 | } | |
567 | ||
568 | static void get_deferred_key(uint64_t seq, string *out) | |
569 | { | |
570 | _key_encode_u64(seq, out); | |
571 | } | |
572 | ||
11fdf7f2 TL |
573 | static void get_pool_stat_key(int64_t pool_id, string *key) |
574 | { | |
575 | key->clear(); | |
576 | _key_encode_u64(pool_id, key); | |
577 | } | |
578 | ||
579 | static int get_key_pool_stat(const string& key, uint64_t* pool_id) | |
580 | { | |
581 | const char *p = key.c_str(); | |
582 | if (key.length() < sizeof(uint64_t)) | |
583 | return -1; | |
584 | _key_decode_u64(p, pool_id); | |
585 | return 0; | |
586 | } | |
7c673cae | 587 | |
81eedcae TL |
588 | template <int LogLevelV> |
589 | void _dump_extent_map(CephContext *cct, const BlueStore::ExtentMap &em) | |
590 | { | |
591 | uint64_t pos = 0; | |
592 | for (auto& s : em.shards) { | |
593 | dout(LogLevelV) << __func__ << " shard " << *s.shard_info | |
594 | << (s.loaded ? " (loaded)" : "") | |
595 | << (s.dirty ? " (dirty)" : "") | |
596 | << dendl; | |
597 | } | |
598 | for (auto& e : em.extent_map) { | |
599 | dout(LogLevelV) << __func__ << " " << e << dendl; | |
600 | ceph_assert(e.logical_offset >= pos); | |
601 | pos = e.logical_offset + e.length; | |
602 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
603 | if (blob.has_csum()) { | |
604 | vector<uint64_t> v; | |
605 | unsigned n = blob.get_csum_count(); | |
606 | for (unsigned i = 0; i < n; ++i) | |
607 | v.push_back(blob.get_csum_item(i)); | |
608 | dout(LogLevelV) << __func__ << " csum: " << std::hex << v << std::dec | |
609 | << dendl; | |
610 | } | |
611 | std::lock_guard l(e.blob->shared_blob->get_cache()->lock); | |
612 | for (auto& i : e.blob->shared_blob->bc.buffer_map) { | |
613 | dout(LogLevelV) << __func__ << " 0x" << std::hex << i.first | |
614 | << "~" << i.second->length << std::dec | |
615 | << " " << *i.second << dendl; | |
616 | } | |
617 | } | |
618 | } | |
619 | ||
620 | template <int LogLevelV> | |
621 | void _dump_onode(CephContext *cct, const BlueStore::Onode& o) | |
622 | { | |
623 | if (!cct->_conf->subsys.should_gather<ceph_subsys_bluestore, LogLevelV>()) | |
624 | return; | |
625 | dout(LogLevelV) << __func__ << " " << &o << " " << o.oid | |
626 | << " nid " << o.onode.nid | |
627 | << " size 0x" << std::hex << o.onode.size | |
628 | << " (" << std::dec << o.onode.size << ")" | |
629 | << " expected_object_size " << o.onode.expected_object_size | |
630 | << " expected_write_size " << o.onode.expected_write_size | |
631 | << " in " << o.onode.extent_map_shards.size() << " shards" | |
632 | << ", " << o.extent_map.spanning_blob_map.size() | |
633 | << " spanning blobs" | |
634 | << dendl; | |
635 | for (auto p = o.onode.attrs.begin(); | |
636 | p != o.onode.attrs.end(); | |
637 | ++p) { | |
638 | dout(LogLevelV) << __func__ << " attr " << p->first | |
639 | << " len " << p->second.length() << dendl; | |
640 | } | |
641 | _dump_extent_map<LogLevelV>(cct, o.extent_map); | |
642 | } | |
643 | ||
644 | template <int LogLevelV> | |
645 | void _dump_transaction(CephContext *cct, ObjectStore::Transaction *t) | |
646 | { | |
647 | dout(LogLevelV) << __func__ << " transaction dump:\n"; | |
648 | JSONFormatter f(true); | |
649 | f.open_object_section("transaction"); | |
650 | t->dump(&f); | |
651 | f.close_section(); | |
652 | f.flush(*_dout); | |
653 | *_dout << dendl; | |
654 | } | |
655 | ||
7c673cae FG |
656 | // merge operators |
657 | ||
658 | struct Int64ArrayMergeOperator : public KeyValueDB::MergeOperator { | |
659 | void merge_nonexistent( | |
660 | const char *rdata, size_t rlen, std::string *new_value) override { | |
661 | *new_value = std::string(rdata, rlen); | |
662 | } | |
663 | void merge( | |
664 | const char *ldata, size_t llen, | |
665 | const char *rdata, size_t rlen, | |
666 | std::string *new_value) override { | |
11fdf7f2 TL |
667 | ceph_assert(llen == rlen); |
668 | ceph_assert((rlen % 8) == 0); | |
7c673cae FG |
669 | new_value->resize(rlen); |
670 | const __le64* lv = (const __le64*)ldata; | |
671 | const __le64* rv = (const __le64*)rdata; | |
672 | __le64* nv = &(__le64&)new_value->at(0); | |
673 | for (size_t i = 0; i < rlen >> 3; ++i) { | |
674 | nv[i] = lv[i] + rv[i]; | |
675 | } | |
676 | } | |
677 | // We use each operator name and each prefix to construct the | |
678 | // overall RocksDB operator name for consistency check at open time. | |
91327a77 | 679 | const char *name() const override { |
7c673cae FG |
680 | return "int64_array"; |
681 | } | |
682 | }; | |
683 | ||
684 | ||
685 | // Buffer | |
686 | ||
687 | ostream& operator<<(ostream& out, const BlueStore::Buffer& b) | |
688 | { | |
689 | out << "buffer(" << &b << " space " << b.space << " 0x" << std::hex | |
690 | << b.offset << "~" << b.length << std::dec | |
691 | << " " << BlueStore::Buffer::get_state_name(b.state); | |
692 | if (b.flags) | |
693 | out << " " << BlueStore::Buffer::get_flag_name(b.flags); | |
694 | return out << ")"; | |
695 | } | |
696 | ||
697 | // Garbage Collector | |
698 | ||
699 | void BlueStore::GarbageCollector::process_protrusive_extents( | |
700 | const BlueStore::ExtentMap& extent_map, | |
701 | uint64_t start_offset, | |
702 | uint64_t end_offset, | |
703 | uint64_t start_touch_offset, | |
704 | uint64_t end_touch_offset, | |
705 | uint64_t min_alloc_size) | |
706 | { | |
11fdf7f2 | 707 | ceph_assert(start_offset <= start_touch_offset && end_offset>= end_touch_offset); |
7c673cae | 708 | |
11fdf7f2 TL |
709 | uint64_t lookup_start_offset = p2align(start_offset, min_alloc_size); |
710 | uint64_t lookup_end_offset = round_up_to(end_offset, min_alloc_size); | |
7c673cae FG |
711 | |
712 | dout(30) << __func__ << " (hex): [" << std::hex | |
713 | << lookup_start_offset << ", " << lookup_end_offset | |
714 | << ")" << std::dec << dendl; | |
715 | ||
716 | for (auto it = extent_map.seek_lextent(lookup_start_offset); | |
717 | it != extent_map.extent_map.end() && | |
718 | it->logical_offset < lookup_end_offset; | |
719 | ++it) { | |
720 | uint64_t alloc_unit_start = it->logical_offset / min_alloc_size; | |
721 | uint64_t alloc_unit_end = (it->logical_end() - 1) / min_alloc_size; | |
722 | ||
723 | dout(30) << __func__ << " " << *it | |
724 | << "alloc_units: " << alloc_unit_start << ".." << alloc_unit_end | |
725 | << dendl; | |
726 | ||
727 | Blob* b = it->blob.get(); | |
728 | ||
729 | if (it->logical_offset >=start_touch_offset && | |
730 | it->logical_end() <= end_touch_offset) { | |
731 | // Process extents within the range affected by | |
732 | // the current write request. | |
733 | // Need to take into account if existing extents | |
734 | // can be merged with them (uncompressed case) | |
735 | if (!b->get_blob().is_compressed()) { | |
736 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
737 | --blob_info_counted->expected_allocations; // don't need to allocate | |
738 | // new AU for compressed | |
739 | // data since another | |
740 | // collocated uncompressed | |
741 | // blob already exists | |
742 | dout(30) << __func__ << " --expected:" | |
743 | << alloc_unit_start << dendl; | |
744 | } | |
745 | used_alloc_unit = alloc_unit_end; | |
746 | blob_info_counted = nullptr; | |
747 | } | |
748 | } else if (b->get_blob().is_compressed()) { | |
749 | ||
750 | // additionally we take compressed blobs that were not impacted | |
751 | // by the write into account too | |
752 | BlobInfo& bi = | |
753 | affected_blobs.emplace( | |
754 | b, BlobInfo(b->get_referenced_bytes())).first->second; | |
755 | ||
756 | int adjust = | |
757 | (used_alloc_unit && used_alloc_unit == alloc_unit_start) ? 0 : 1; | |
758 | bi.expected_allocations += alloc_unit_end - alloc_unit_start + adjust; | |
759 | dout(30) << __func__ << " expected_allocations=" | |
760 | << bi.expected_allocations << " end_au:" | |
761 | << alloc_unit_end << dendl; | |
762 | ||
763 | blob_info_counted = &bi; | |
764 | used_alloc_unit = alloc_unit_end; | |
765 | ||
11fdf7f2 | 766 | ceph_assert(it->length <= bi.referenced_bytes); |
7c673cae FG |
767 | bi.referenced_bytes -= it->length; |
768 | dout(30) << __func__ << " affected_blob:" << *b | |
769 | << " unref 0x" << std::hex << it->length | |
770 | << " referenced = 0x" << bi.referenced_bytes | |
771 | << std::dec << dendl; | |
772 | // NOTE: we can't move specific blob to resulting GC list here | |
773 | // when reference counter == 0 since subsequent extents might | |
774 | // decrement its expected_allocation. | |
775 | // Hence need to enumerate all the extents first. | |
776 | if (!bi.collect_candidate) { | |
777 | bi.first_lextent = it; | |
778 | bi.collect_candidate = true; | |
779 | } | |
780 | bi.last_lextent = it; | |
781 | } else { | |
782 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
783 | // don't need to allocate new AU for compressed data since another | |
784 | // collocated uncompressed blob already exists | |
785 | --blob_info_counted->expected_allocations; | |
786 | dout(30) << __func__ << " --expected_allocations:" | |
787 | << alloc_unit_start << dendl; | |
788 | } | |
789 | used_alloc_unit = alloc_unit_end; | |
790 | blob_info_counted = nullptr; | |
791 | } | |
792 | } | |
793 | ||
794 | for (auto b_it = affected_blobs.begin(); | |
795 | b_it != affected_blobs.end(); | |
796 | ++b_it) { | |
797 | Blob* b = b_it->first; | |
798 | BlobInfo& bi = b_it->second; | |
799 | if (bi.referenced_bytes == 0) { | |
800 | uint64_t len_on_disk = b_it->first->get_blob().get_ondisk_length(); | |
801 | int64_t blob_expected_for_release = | |
11fdf7f2 | 802 | round_up_to(len_on_disk, min_alloc_size) / min_alloc_size; |
7c673cae FG |
803 | |
804 | dout(30) << __func__ << " " << *(b_it->first) | |
805 | << " expected4release=" << blob_expected_for_release | |
806 | << " expected_allocations=" << bi.expected_allocations | |
807 | << dendl; | |
808 | int64_t benefit = blob_expected_for_release - bi.expected_allocations; | |
11fdf7f2 | 809 | if (benefit >= g_conf()->bluestore_gc_enable_blob_threshold) { |
7c673cae FG |
810 | if (bi.collect_candidate) { |
811 | auto it = bi.first_lextent; | |
812 | bool bExit = false; | |
813 | do { | |
814 | if (it->blob.get() == b) { | |
815 | extents_to_collect.emplace_back(it->logical_offset, it->length); | |
816 | } | |
817 | bExit = it == bi.last_lextent; | |
818 | ++it; | |
31f18b77 | 819 | } while (!bExit); |
7c673cae FG |
820 | } |
821 | expected_for_release += blob_expected_for_release; | |
822 | expected_allocations += bi.expected_allocations; | |
823 | } | |
824 | } | |
825 | } | |
826 | } | |
827 | ||
828 | int64_t BlueStore::GarbageCollector::estimate( | |
829 | uint64_t start_offset, | |
830 | uint64_t length, | |
831 | const BlueStore::ExtentMap& extent_map, | |
832 | const BlueStore::old_extent_map_t& old_extents, | |
833 | uint64_t min_alloc_size) | |
834 | { | |
835 | ||
836 | affected_blobs.clear(); | |
837 | extents_to_collect.clear(); | |
838 | used_alloc_unit = boost::optional<uint64_t >(); | |
839 | blob_info_counted = nullptr; | |
840 | ||
841 | gc_start_offset = start_offset; | |
842 | gc_end_offset = start_offset + length; | |
843 | ||
844 | uint64_t end_offset = start_offset + length; | |
845 | ||
846 | for (auto it = old_extents.begin(); it != old_extents.end(); ++it) { | |
847 | Blob* b = it->e.blob.get(); | |
848 | if (b->get_blob().is_compressed()) { | |
849 | ||
850 | // update gc_start_offset/gc_end_offset if needed | |
851 | gc_start_offset = min(gc_start_offset, (uint64_t)it->e.blob_start()); | |
11fdf7f2 | 852 | gc_end_offset = std::max(gc_end_offset, (uint64_t)it->e.blob_end()); |
7c673cae FG |
853 | |
854 | auto o = it->e.logical_offset; | |
855 | auto l = it->e.length; | |
856 | ||
857 | uint64_t ref_bytes = b->get_referenced_bytes(); | |
858 | // micro optimization to bypass blobs that have no more references | |
859 | if (ref_bytes != 0) { | |
860 | dout(30) << __func__ << " affected_blob:" << *b | |
861 | << " unref 0x" << std::hex << o << "~" << l | |
862 | << std::dec << dendl; | |
863 | affected_blobs.emplace(b, BlobInfo(ref_bytes)); | |
864 | } | |
865 | } | |
866 | } | |
867 | dout(30) << __func__ << " gc range(hex): [" << std::hex | |
868 | << gc_start_offset << ", " << gc_end_offset | |
869 | << ")" << std::dec << dendl; | |
870 | ||
871 | // enumerate preceeding extents to check if they reference affected blobs | |
872 | if (gc_start_offset < start_offset || gc_end_offset > end_offset) { | |
873 | process_protrusive_extents(extent_map, | |
874 | gc_start_offset, | |
875 | gc_end_offset, | |
876 | start_offset, | |
877 | end_offset, | |
878 | min_alloc_size); | |
879 | } | |
880 | return expected_for_release - expected_allocations; | |
881 | } | |
882 | ||
883 | // Cache | |
884 | ||
885 | BlueStore::Cache *BlueStore::Cache::create(CephContext* cct, string type, | |
886 | PerfCounters *logger) | |
887 | { | |
888 | Cache *c = nullptr; | |
889 | ||
890 | if (type == "lru") | |
891 | c = new LRUCache(cct); | |
892 | else if (type == "2q") | |
893 | c = new TwoQCache(cct); | |
894 | else | |
11fdf7f2 | 895 | ceph_abort_msg("unrecognized cache type"); |
7c673cae FG |
896 | |
897 | c->logger = logger; | |
898 | return c; | |
899 | } | |
900 | ||
91327a77 | 901 | void BlueStore::Cache::trim(uint64_t onode_max, uint64_t buffer_max) |
7c673cae | 902 | { |
11fdf7f2 | 903 | std::lock_guard l(lock); |
91327a77 | 904 | _trim(onode_max, buffer_max); |
7c673cae FG |
905 | } |
906 | ||
91327a77 | 907 | void BlueStore::Cache::trim_all() |
7c673cae | 908 | { |
11fdf7f2 | 909 | std::lock_guard l(lock); |
91327a77 | 910 | _trim(0, 0); |
7c673cae FG |
911 | } |
912 | ||
7c673cae FG |
913 | // LRUCache |
914 | #undef dout_prefix | |
915 | #define dout_prefix *_dout << "bluestore.LRUCache(" << this << ") " | |
916 | ||
917 | void BlueStore::LRUCache::_touch_onode(OnodeRef& o) | |
918 | { | |
919 | auto p = onode_lru.iterator_to(*o); | |
920 | onode_lru.erase(p); | |
921 | onode_lru.push_front(*o); | |
922 | } | |
923 | ||
924 | void BlueStore::LRUCache::_trim(uint64_t onode_max, uint64_t buffer_max) | |
925 | { | |
926 | dout(20) << __func__ << " onodes " << onode_lru.size() << " / " << onode_max | |
927 | << " buffers " << buffer_size << " / " << buffer_max | |
928 | << dendl; | |
929 | ||
930 | _audit("trim start"); | |
931 | ||
932 | // buffers | |
933 | while (buffer_size > buffer_max) { | |
934 | auto i = buffer_lru.rbegin(); | |
935 | if (i == buffer_lru.rend()) { | |
936 | // stop if buffer_lru is now empty | |
937 | break; | |
938 | } | |
939 | ||
940 | Buffer *b = &*i; | |
11fdf7f2 | 941 | ceph_assert(b->is_clean()); |
7c673cae FG |
942 | dout(20) << __func__ << " rm " << *b << dendl; |
943 | b->space->_rm_buffer(this, b); | |
944 | } | |
945 | ||
946 | // onodes | |
91327a77 | 947 | if (onode_max >= onode_lru.size()) { |
7c673cae | 948 | return; // don't even try |
91327a77 AA |
949 | } |
950 | uint64_t num = onode_lru.size() - onode_max; | |
7c673cae FG |
951 | |
952 | auto p = onode_lru.end(); | |
11fdf7f2 | 953 | ceph_assert(p != onode_lru.begin()); |
7c673cae FG |
954 | --p; |
955 | int skipped = 0; | |
11fdf7f2 | 956 | int max_skipped = g_conf()->bluestore_cache_trim_max_skip_pinned; |
7c673cae FG |
957 | while (num > 0) { |
958 | Onode *o = &*p; | |
959 | int refs = o->nref.load(); | |
960 | if (refs > 1) { | |
961 | dout(20) << __func__ << " " << o->oid << " has " << refs | |
962 | << " refs, skipping" << dendl; | |
963 | if (++skipped >= max_skipped) { | |
964 | dout(20) << __func__ << " maximum skip pinned reached; stopping with " | |
965 | << num << " left to trim" << dendl; | |
966 | break; | |
967 | } | |
968 | ||
969 | if (p == onode_lru.begin()) { | |
970 | break; | |
971 | } else { | |
972 | p--; | |
973 | num--; | |
974 | continue; | |
975 | } | |
976 | } | |
977 | dout(30) << __func__ << " rm " << o->oid << dendl; | |
978 | if (p != onode_lru.begin()) { | |
979 | onode_lru.erase(p--); | |
980 | } else { | |
981 | onode_lru.erase(p); | |
11fdf7f2 | 982 | ceph_assert(num == 1); |
7c673cae FG |
983 | } |
984 | o->get(); // paranoia | |
985 | o->c->onode_map.remove(o->oid); | |
986 | o->put(); | |
987 | --num; | |
988 | } | |
989 | } | |
990 | ||
991 | #ifdef DEBUG_CACHE | |
992 | void BlueStore::LRUCache::_audit(const char *when) | |
993 | { | |
994 | dout(10) << __func__ << " " << when << " start" << dendl; | |
995 | uint64_t s = 0; | |
996 | for (auto i = buffer_lru.begin(); i != buffer_lru.end(); ++i) { | |
997 | s += i->length; | |
998 | } | |
999 | if (s != buffer_size) { | |
1000 | derr << __func__ << " buffer_size " << buffer_size << " actual " << s | |
1001 | << dendl; | |
1002 | for (auto i = buffer_lru.begin(); i != buffer_lru.end(); ++i) { | |
1003 | derr << __func__ << " " << *i << dendl; | |
1004 | } | |
11fdf7f2 | 1005 | ceph_assert(s == buffer_size); |
7c673cae FG |
1006 | } |
1007 | dout(20) << __func__ << " " << when << " buffer_size " << buffer_size | |
1008 | << " ok" << dendl; | |
1009 | } | |
1010 | #endif | |
1011 | ||
1012 | // TwoQCache | |
1013 | #undef dout_prefix | |
1014 | #define dout_prefix *_dout << "bluestore.2QCache(" << this << ") " | |
1015 | ||
1016 | ||
1017 | void BlueStore::TwoQCache::_touch_onode(OnodeRef& o) | |
1018 | { | |
1019 | auto p = onode_lru.iterator_to(*o); | |
1020 | onode_lru.erase(p); | |
1021 | onode_lru.push_front(*o); | |
1022 | } | |
1023 | ||
1024 | void BlueStore::TwoQCache::_add_buffer(Buffer *b, int level, Buffer *near) | |
1025 | { | |
1026 | dout(20) << __func__ << " level " << level << " near " << near | |
1027 | << " on " << *b | |
1028 | << " which has cache_private " << b->cache_private << dendl; | |
1029 | if (near) { | |
1030 | b->cache_private = near->cache_private; | |
1031 | switch (b->cache_private) { | |
1032 | case BUFFER_WARM_IN: | |
1033 | buffer_warm_in.insert(buffer_warm_in.iterator_to(*near), *b); | |
1034 | break; | |
1035 | case BUFFER_WARM_OUT: | |
11fdf7f2 | 1036 | ceph_assert(b->is_empty()); |
7c673cae FG |
1037 | buffer_warm_out.insert(buffer_warm_out.iterator_to(*near), *b); |
1038 | break; | |
1039 | case BUFFER_HOT: | |
1040 | buffer_hot.insert(buffer_hot.iterator_to(*near), *b); | |
1041 | break; | |
1042 | default: | |
11fdf7f2 | 1043 | ceph_abort_msg("bad cache_private"); |
7c673cae FG |
1044 | } |
1045 | } else if (b->cache_private == BUFFER_NEW) { | |
1046 | b->cache_private = BUFFER_WARM_IN; | |
1047 | if (level > 0) { | |
1048 | buffer_warm_in.push_front(*b); | |
1049 | } else { | |
1050 | // take caller hint to start at the back of the warm queue | |
1051 | buffer_warm_in.push_back(*b); | |
1052 | } | |
1053 | } else { | |
1054 | // we got a hint from discard | |
1055 | switch (b->cache_private) { | |
1056 | case BUFFER_WARM_IN: | |
1057 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1058 | // do this. | |
1059 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1060 | buffer_warm_in.push_front(*b); | |
1061 | break; | |
1062 | case BUFFER_WARM_OUT: | |
1063 | b->cache_private = BUFFER_HOT; | |
1064 | // move to hot. fall-thru | |
1065 | case BUFFER_HOT: | |
1066 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1067 | buffer_hot.push_front(*b); | |
1068 | break; | |
1069 | default: | |
11fdf7f2 | 1070 | ceph_abort_msg("bad cache_private"); |
7c673cae FG |
1071 | } |
1072 | } | |
1073 | if (!b->is_empty()) { | |
1074 | buffer_bytes += b->length; | |
1075 | buffer_list_bytes[b->cache_private] += b->length; | |
1076 | } | |
1077 | } | |
1078 | ||
1079 | void BlueStore::TwoQCache::_rm_buffer(Buffer *b) | |
1080 | { | |
1081 | dout(20) << __func__ << " " << *b << dendl; | |
1082 | if (!b->is_empty()) { | |
11fdf7f2 | 1083 | ceph_assert(buffer_bytes >= b->length); |
7c673cae | 1084 | buffer_bytes -= b->length; |
11fdf7f2 | 1085 | ceph_assert(buffer_list_bytes[b->cache_private] >= b->length); |
7c673cae FG |
1086 | buffer_list_bytes[b->cache_private] -= b->length; |
1087 | } | |
1088 | switch (b->cache_private) { | |
1089 | case BUFFER_WARM_IN: | |
1090 | buffer_warm_in.erase(buffer_warm_in.iterator_to(*b)); | |
1091 | break; | |
1092 | case BUFFER_WARM_OUT: | |
1093 | buffer_warm_out.erase(buffer_warm_out.iterator_to(*b)); | |
1094 | break; | |
1095 | case BUFFER_HOT: | |
1096 | buffer_hot.erase(buffer_hot.iterator_to(*b)); | |
1097 | break; | |
1098 | default: | |
11fdf7f2 | 1099 | ceph_abort_msg("bad cache_private"); |
7c673cae FG |
1100 | } |
1101 | } | |
1102 | ||
1103 | void BlueStore::TwoQCache::_move_buffer(Cache *srcc, Buffer *b) | |
1104 | { | |
1105 | TwoQCache *src = static_cast<TwoQCache*>(srcc); | |
1106 | src->_rm_buffer(b); | |
1107 | ||
1108 | // preserve which list we're on (even if we can't preserve the order!) | |
1109 | switch (b->cache_private) { | |
1110 | case BUFFER_WARM_IN: | |
11fdf7f2 | 1111 | ceph_assert(!b->is_empty()); |
7c673cae FG |
1112 | buffer_warm_in.push_back(*b); |
1113 | break; | |
1114 | case BUFFER_WARM_OUT: | |
11fdf7f2 | 1115 | ceph_assert(b->is_empty()); |
7c673cae FG |
1116 | buffer_warm_out.push_back(*b); |
1117 | break; | |
1118 | case BUFFER_HOT: | |
11fdf7f2 | 1119 | ceph_assert(!b->is_empty()); |
7c673cae FG |
1120 | buffer_hot.push_back(*b); |
1121 | break; | |
1122 | default: | |
11fdf7f2 | 1123 | ceph_abort_msg("bad cache_private"); |
7c673cae FG |
1124 | } |
1125 | if (!b->is_empty()) { | |
1126 | buffer_bytes += b->length; | |
1127 | buffer_list_bytes[b->cache_private] += b->length; | |
1128 | } | |
1129 | } | |
1130 | ||
1131 | void BlueStore::TwoQCache::_adjust_buffer_size(Buffer *b, int64_t delta) | |
1132 | { | |
1133 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1134 | if (!b->is_empty()) { | |
11fdf7f2 | 1135 | ceph_assert((int64_t)buffer_bytes + delta >= 0); |
7c673cae | 1136 | buffer_bytes += delta; |
11fdf7f2 | 1137 | ceph_assert((int64_t)buffer_list_bytes[b->cache_private] + delta >= 0); |
7c673cae FG |
1138 | buffer_list_bytes[b->cache_private] += delta; |
1139 | } | |
1140 | } | |
1141 | ||
1142 | void BlueStore::TwoQCache::_trim(uint64_t onode_max, uint64_t buffer_max) | |
1143 | { | |
1144 | dout(20) << __func__ << " onodes " << onode_lru.size() << " / " << onode_max | |
1145 | << " buffers " << buffer_bytes << " / " << buffer_max | |
1146 | << dendl; | |
1147 | ||
1148 | _audit("trim start"); | |
1149 | ||
1150 | // buffers | |
1151 | if (buffer_bytes > buffer_max) { | |
1152 | uint64_t kin = buffer_max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1153 | uint64_t khot = buffer_max - kin; | |
1154 | ||
1155 | // pre-calculate kout based on average buffer size too, | |
1156 | // which is typical(the warm_in and hot lists may change later) | |
1157 | uint64_t kout = 0; | |
1158 | uint64_t buffer_num = buffer_hot.size() + buffer_warm_in.size(); | |
1159 | if (buffer_num) { | |
1160 | uint64_t buffer_avg_size = buffer_bytes / buffer_num; | |
11fdf7f2 | 1161 | ceph_assert(buffer_avg_size); |
7c673cae FG |
1162 | uint64_t calculated_buffer_num = buffer_max / buffer_avg_size; |
1163 | kout = calculated_buffer_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1164 | } | |
1165 | ||
1166 | if (buffer_list_bytes[BUFFER_HOT] < khot) { | |
1167 | // hot is small, give slack to warm_in | |
1168 | kin += khot - buffer_list_bytes[BUFFER_HOT]; | |
1169 | } else if (buffer_list_bytes[BUFFER_WARM_IN] < kin) { | |
1170 | // warm_in is small, give slack to hot | |
1171 | khot += kin - buffer_list_bytes[BUFFER_WARM_IN]; | |
1172 | } | |
1173 | ||
1174 | // adjust warm_in list | |
1175 | int64_t to_evict_bytes = buffer_list_bytes[BUFFER_WARM_IN] - kin; | |
1176 | uint64_t evicted = 0; | |
1177 | ||
1178 | while (to_evict_bytes > 0) { | |
1179 | auto p = buffer_warm_in.rbegin(); | |
1180 | if (p == buffer_warm_in.rend()) { | |
1181 | // stop if warm_in list is now empty | |
1182 | break; | |
1183 | } | |
1184 | ||
1185 | Buffer *b = &*p; | |
11fdf7f2 | 1186 | ceph_assert(b->is_clean()); |
7c673cae | 1187 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; |
11fdf7f2 | 1188 | ceph_assert(buffer_bytes >= b->length); |
7c673cae | 1189 | buffer_bytes -= b->length; |
11fdf7f2 | 1190 | ceph_assert(buffer_list_bytes[BUFFER_WARM_IN] >= b->length); |
7c673cae FG |
1191 | buffer_list_bytes[BUFFER_WARM_IN] -= b->length; |
1192 | to_evict_bytes -= b->length; | |
1193 | evicted += b->length; | |
1194 | b->state = Buffer::STATE_EMPTY; | |
1195 | b->data.clear(); | |
1196 | buffer_warm_in.erase(buffer_warm_in.iterator_to(*b)); | |
1197 | buffer_warm_out.push_front(*b); | |
1198 | b->cache_private = BUFFER_WARM_OUT; | |
1199 | } | |
1200 | ||
1201 | if (evicted > 0) { | |
1adf2230 | 1202 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) |
7c673cae FG |
1203 | << " from warm_in list, done evicting warm_in buffers" |
1204 | << dendl; | |
1205 | } | |
1206 | ||
1207 | // adjust hot list | |
1208 | to_evict_bytes = buffer_list_bytes[BUFFER_HOT] - khot; | |
1209 | evicted = 0; | |
1210 | ||
1211 | while (to_evict_bytes > 0) { | |
1212 | auto p = buffer_hot.rbegin(); | |
1213 | if (p == buffer_hot.rend()) { | |
1214 | // stop if hot list is now empty | |
1215 | break; | |
1216 | } | |
1217 | ||
1218 | Buffer *b = &*p; | |
1219 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
11fdf7f2 | 1220 | ceph_assert(b->is_clean()); |
7c673cae FG |
1221 | // adjust evict size before buffer goes invalid |
1222 | to_evict_bytes -= b->length; | |
1223 | evicted += b->length; | |
1224 | b->space->_rm_buffer(this, b); | |
1225 | } | |
1226 | ||
1227 | if (evicted > 0) { | |
1adf2230 | 1228 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) |
7c673cae FG |
1229 | << " from hot list, done evicting hot buffers" |
1230 | << dendl; | |
1231 | } | |
1232 | ||
1233 | // adjust warm out list too, if necessary | |
1234 | int64_t num = buffer_warm_out.size() - kout; | |
1235 | while (num-- > 0) { | |
1236 | Buffer *b = &*buffer_warm_out.rbegin(); | |
11fdf7f2 | 1237 | ceph_assert(b->is_empty()); |
7c673cae FG |
1238 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; |
1239 | b->space->_rm_buffer(this, b); | |
1240 | } | |
1241 | } | |
1242 | ||
1243 | // onodes | |
91327a77 | 1244 | if (onode_max >= onode_lru.size()) { |
7c673cae | 1245 | return; // don't even try |
91327a77 AA |
1246 | } |
1247 | uint64_t num = onode_lru.size() - onode_max; | |
7c673cae FG |
1248 | |
1249 | auto p = onode_lru.end(); | |
11fdf7f2 | 1250 | ceph_assert(p != onode_lru.begin()); |
7c673cae FG |
1251 | --p; |
1252 | int skipped = 0; | |
11fdf7f2 | 1253 | int max_skipped = g_conf()->bluestore_cache_trim_max_skip_pinned; |
7c673cae FG |
1254 | while (num > 0) { |
1255 | Onode *o = &*p; | |
1256 | dout(20) << __func__ << " considering " << o << dendl; | |
1257 | int refs = o->nref.load(); | |
1258 | if (refs > 1) { | |
1259 | dout(20) << __func__ << " " << o->oid << " has " << refs | |
1260 | << " refs; skipping" << dendl; | |
1261 | if (++skipped >= max_skipped) { | |
1262 | dout(20) << __func__ << " maximum skip pinned reached; stopping with " | |
1263 | << num << " left to trim" << dendl; | |
1264 | break; | |
1265 | } | |
1266 | ||
1267 | if (p == onode_lru.begin()) { | |
1268 | break; | |
1269 | } else { | |
1270 | p--; | |
1271 | num--; | |
1272 | continue; | |
1273 | } | |
1274 | } | |
1275 | dout(30) << __func__ << " " << o->oid << " num=" << num <<" lru size="<<onode_lru.size()<< dendl; | |
1276 | if (p != onode_lru.begin()) { | |
1277 | onode_lru.erase(p--); | |
1278 | } else { | |
1279 | onode_lru.erase(p); | |
11fdf7f2 | 1280 | ceph_assert(num == 1); |
7c673cae FG |
1281 | } |
1282 | o->get(); // paranoia | |
1283 | o->c->onode_map.remove(o->oid); | |
1284 | o->put(); | |
1285 | --num; | |
1286 | } | |
1287 | } | |
1288 | ||
1289 | #ifdef DEBUG_CACHE | |
1290 | void BlueStore::TwoQCache::_audit(const char *when) | |
1291 | { | |
1292 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1293 | uint64_t s = 0; | |
1294 | for (auto i = buffer_hot.begin(); i != buffer_hot.end(); ++i) { | |
1295 | s += i->length; | |
1296 | } | |
1297 | ||
1298 | uint64_t hot_bytes = s; | |
1299 | if (hot_bytes != buffer_list_bytes[BUFFER_HOT]) { | |
1300 | derr << __func__ << " hot_list_bytes " | |
1301 | << buffer_list_bytes[BUFFER_HOT] | |
1302 | << " != actual " << hot_bytes | |
1303 | << dendl; | |
11fdf7f2 | 1304 | ceph_assert(hot_bytes == buffer_list_bytes[BUFFER_HOT]); |
7c673cae FG |
1305 | } |
1306 | ||
1307 | for (auto i = buffer_warm_in.begin(); i != buffer_warm_in.end(); ++i) { | |
1308 | s += i->length; | |
1309 | } | |
1310 | ||
1311 | uint64_t warm_in_bytes = s - hot_bytes; | |
1312 | if (warm_in_bytes != buffer_list_bytes[BUFFER_WARM_IN]) { | |
1313 | derr << __func__ << " warm_in_list_bytes " | |
1314 | << buffer_list_bytes[BUFFER_WARM_IN] | |
1315 | << " != actual " << warm_in_bytes | |
1316 | << dendl; | |
11fdf7f2 | 1317 | ceph_assert(warm_in_bytes == buffer_list_bytes[BUFFER_WARM_IN]); |
7c673cae FG |
1318 | } |
1319 | ||
1320 | if (s != buffer_bytes) { | |
1321 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1322 | << dendl; | |
11fdf7f2 | 1323 | ceph_assert(s == buffer_bytes); |
7c673cae FG |
1324 | } |
1325 | ||
1326 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes | |
1327 | << " ok" << dendl; | |
1328 | } | |
1329 | #endif | |
1330 | ||
1331 | ||
1332 | // BufferSpace | |
1333 | ||
1334 | #undef dout_prefix | |
1335 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1336 | ||
1337 | void BlueStore::BufferSpace::_clear(Cache* cache) | |
1338 | { | |
1339 | // note: we already hold cache->lock | |
1340 | ldout(cache->cct, 20) << __func__ << dendl; | |
1341 | while (!buffer_map.empty()) { | |
1342 | _rm_buffer(cache, buffer_map.begin()); | |
1343 | } | |
1344 | } | |
1345 | ||
1346 | int BlueStore::BufferSpace::_discard(Cache* cache, uint32_t offset, uint32_t length) | |
1347 | { | |
1348 | // note: we already hold cache->lock | |
1349 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1350 | << std::dec << dendl; | |
1351 | int cache_private = 0; | |
1352 | cache->_audit("discard start"); | |
1353 | auto i = _data_lower_bound(offset); | |
1354 | uint32_t end = offset + length; | |
1355 | while (i != buffer_map.end()) { | |
1356 | Buffer *b = i->second.get(); | |
1357 | if (b->offset >= end) { | |
1358 | break; | |
1359 | } | |
1360 | if (b->cache_private > cache_private) { | |
1361 | cache_private = b->cache_private; | |
1362 | } | |
1363 | if (b->offset < offset) { | |
1364 | int64_t front = offset - b->offset; | |
1365 | if (b->end() > end) { | |
1366 | // drop middle (split) | |
1367 | uint32_t tail = b->end() - end; | |
1368 | if (b->data.length()) { | |
1369 | bufferlist bl; | |
1370 | bl.substr_of(b->data, b->length - tail, tail); | |
31f18b77 FG |
1371 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1372 | nb->maybe_rebuild(); | |
1373 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1374 | } else { |
31f18b77 FG |
1375 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail), |
1376 | 0, b); | |
7c673cae FG |
1377 | } |
1378 | if (!b->is_writing()) { | |
1379 | cache->_adjust_buffer_size(b, front - (int64_t)b->length); | |
1380 | } | |
1381 | b->truncate(front); | |
31f18b77 | 1382 | b->maybe_rebuild(); |
7c673cae FG |
1383 | cache->_audit("discard end 1"); |
1384 | break; | |
1385 | } else { | |
1386 | // drop tail | |
1387 | if (!b->is_writing()) { | |
1388 | cache->_adjust_buffer_size(b, front - (int64_t)b->length); | |
1389 | } | |
1390 | b->truncate(front); | |
31f18b77 | 1391 | b->maybe_rebuild(); |
7c673cae FG |
1392 | ++i; |
1393 | continue; | |
1394 | } | |
1395 | } | |
1396 | if (b->end() <= end) { | |
1397 | // drop entire buffer | |
1398 | _rm_buffer(cache, i++); | |
1399 | continue; | |
1400 | } | |
1401 | // drop front | |
1402 | uint32_t keep = b->end() - end; | |
1403 | if (b->data.length()) { | |
1404 | bufferlist bl; | |
1405 | bl.substr_of(b->data, b->length - keep, keep); | |
31f18b77 FG |
1406 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1407 | nb->maybe_rebuild(); | |
1408 | _add_buffer(cache, nb, 0, b); | |
7c673cae FG |
1409 | } else { |
1410 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep), 0, b); | |
1411 | } | |
1412 | _rm_buffer(cache, i); | |
1413 | cache->_audit("discard end 2"); | |
1414 | break; | |
1415 | } | |
1416 | return cache_private; | |
1417 | } | |
1418 | ||
1419 | void BlueStore::BufferSpace::read( | |
1420 | Cache* cache, | |
224ce89b WB |
1421 | uint32_t offset, |
1422 | uint32_t length, | |
7c673cae | 1423 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1424 | interval_set<uint32_t>& res_intervals, |
1425 | int flags) | |
7c673cae | 1426 | { |
7c673cae FG |
1427 | res.clear(); |
1428 | res_intervals.clear(); | |
1429 | uint32_t want_bytes = length; | |
1430 | uint32_t end = offset + length; | |
224ce89b WB |
1431 | |
1432 | { | |
11fdf7f2 | 1433 | std::lock_guard l(cache->lock); |
224ce89b WB |
1434 | for (auto i = _data_lower_bound(offset); |
1435 | i != buffer_map.end() && offset < end && i->first < end; | |
1436 | ++i) { | |
1437 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1438 | ceph_assert(b->end() > offset); |
91327a77 AA |
1439 | |
1440 | bool val = false; | |
1441 | if (flags & BYPASS_CLEAN_CACHE) | |
1442 | val = b->is_writing(); | |
1443 | else | |
1444 | val = b->is_writing() || b->is_clean(); | |
1445 | if (val) { | |
224ce89b WB |
1446 | if (b->offset < offset) { |
1447 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1448 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1449 | res[offset].substr_of(b->data, skip, l); |
1450 | res_intervals.insert(offset, l); | |
1451 | offset += l; | |
1452 | length -= l; | |
1453 | if (!b->is_writing()) { | |
1454 | cache->_touch_buffer(b); | |
1455 | } | |
1456 | continue; | |
1457 | } | |
1458 | if (b->offset > offset) { | |
1459 | uint32_t gap = b->offset - offset; | |
1460 | if (length <= gap) { | |
1461 | break; | |
1462 | } | |
1463 | offset += gap; | |
1464 | length -= gap; | |
1465 | } | |
1466 | if (!b->is_writing()) { | |
7c673cae | 1467 | cache->_touch_buffer(b); |
224ce89b WB |
1468 | } |
1469 | if (b->length > length) { | |
1470 | res[offset].substr_of(b->data, 0, length); | |
1471 | res_intervals.insert(offset, length); | |
7c673cae | 1472 | break; |
224ce89b WB |
1473 | } else { |
1474 | res[offset].append(b->data); | |
1475 | res_intervals.insert(offset, b->length); | |
1476 | if (b->length == length) | |
1477 | break; | |
1478 | offset += b->length; | |
1479 | length -= b->length; | |
1480 | } | |
7c673cae FG |
1481 | } |
1482 | } | |
1483 | } | |
1484 | ||
1485 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1486 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1487 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1488 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1489 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1490 | } | |
1491 | ||
f64942e4 | 1492 | void BlueStore::BufferSpace::_finish_write(Cache* cache, uint64_t seq) |
7c673cae | 1493 | { |
7c673cae FG |
1494 | auto i = writing.begin(); |
1495 | while (i != writing.end()) { | |
1496 | if (i->seq > seq) { | |
1497 | break; | |
1498 | } | |
1499 | if (i->seq < seq) { | |
1500 | ++i; | |
1501 | continue; | |
1502 | } | |
1503 | ||
1504 | Buffer *b = &*i; | |
11fdf7f2 | 1505 | ceph_assert(b->is_writing()); |
7c673cae FG |
1506 | |
1507 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1508 | writing.erase(i++); | |
1509 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1510 | buffer_map.erase(b->offset); | |
1511 | } else { | |
1512 | b->state = Buffer::STATE_CLEAN; | |
1513 | writing.erase(i++); | |
31f18b77 FG |
1514 | b->maybe_rebuild(); |
1515 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
7c673cae FG |
1516 | cache->_add_buffer(b, 1, nullptr); |
1517 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; | |
1518 | } | |
1519 | } | |
1520 | ||
1521 | cache->_audit("finish_write end"); | |
1522 | } | |
1523 | ||
1524 | void BlueStore::BufferSpace::split(Cache* cache, size_t pos, BlueStore::BufferSpace &r) | |
1525 | { | |
11fdf7f2 | 1526 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1527 | if (buffer_map.empty()) |
1528 | return; | |
1529 | ||
1530 | auto p = --buffer_map.end(); | |
1531 | while (true) { | |
1532 | if (p->second->end() <= pos) | |
1533 | break; | |
1534 | ||
1535 | if (p->second->offset < pos) { | |
1536 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1537 | size_t left = pos - p->second->offset; | |
1538 | size_t right = p->second->length - left; | |
1539 | if (p->second->data.length()) { | |
1540 | bufferlist bl; | |
1541 | bl.substr_of(p->second->data, left, right); | |
1542 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, bl), | |
1543 | 0, p->second.get()); | |
1544 | } else { | |
1545 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, right), | |
1546 | 0, p->second.get()); | |
1547 | } | |
1548 | cache->_adjust_buffer_size(p->second.get(), -right); | |
1549 | p->second->truncate(left); | |
1550 | break; | |
1551 | } | |
1552 | ||
11fdf7f2 | 1553 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1554 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1555 | if (p->second->data.length()) { | |
1556 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1557 | p->second->offset - pos, p->second->data), | |
1558 | 0, p->second.get()); | |
1559 | } else { | |
1560 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1561 | p->second->offset - pos, p->second->length), | |
1562 | 0, p->second.get()); | |
1563 | } | |
1564 | if (p == buffer_map.begin()) { | |
1565 | _rm_buffer(cache, p); | |
1566 | break; | |
1567 | } else { | |
1568 | _rm_buffer(cache, p--); | |
1569 | } | |
1570 | } | |
11fdf7f2 | 1571 | ceph_assert(writing.empty()); |
7c673cae FG |
1572 | } |
1573 | ||
1574 | // OnodeSpace | |
1575 | ||
1576 | #undef dout_prefix | |
1577 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1578 | ||
1579 | BlueStore::OnodeRef BlueStore::OnodeSpace::add(const ghobject_t& oid, OnodeRef o) | |
1580 | { | |
11fdf7f2 | 1581 | std::lock_guard l(cache->lock); |
7c673cae FG |
1582 | auto p = onode_map.find(oid); |
1583 | if (p != onode_map.end()) { | |
1584 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o | |
1585 | << " raced, returning existing " << p->second | |
1586 | << dendl; | |
1587 | return p->second; | |
1588 | } | |
1589 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o << dendl; | |
1590 | onode_map[oid] = o; | |
1591 | cache->_add_onode(o, 1); | |
1592 | return o; | |
1593 | } | |
1594 | ||
1595 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) | |
1596 | { | |
7c673cae | 1597 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b WB |
1598 | OnodeRef o; |
1599 | bool hit = false; | |
1600 | ||
1601 | { | |
11fdf7f2 | 1602 | std::lock_guard l(cache->lock); |
224ce89b WB |
1603 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1604 | if (p == onode_map.end()) { | |
1605 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
1606 | } else { | |
1607 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
1608 | << dendl; | |
1609 | cache->_touch_onode(p->second); | |
1610 | hit = true; | |
1611 | o = p->second; | |
1612 | } | |
1613 | } | |
1614 | ||
1615 | if (hit) { | |
1616 | cache->logger->inc(l_bluestore_onode_hits); | |
1617 | } else { | |
7c673cae | 1618 | cache->logger->inc(l_bluestore_onode_misses); |
7c673cae | 1619 | } |
224ce89b | 1620 | return o; |
7c673cae FG |
1621 | } |
1622 | ||
1623 | void BlueStore::OnodeSpace::clear() | |
1624 | { | |
11fdf7f2 | 1625 | std::lock_guard l(cache->lock); |
7c673cae FG |
1626 | ldout(cache->cct, 10) << __func__ << dendl; |
1627 | for (auto &p : onode_map) { | |
1628 | cache->_rm_onode(p.second); | |
1629 | } | |
1630 | onode_map.clear(); | |
1631 | } | |
1632 | ||
1633 | bool BlueStore::OnodeSpace::empty() | |
1634 | { | |
11fdf7f2 | 1635 | std::lock_guard l(cache->lock); |
7c673cae FG |
1636 | return onode_map.empty(); |
1637 | } | |
1638 | ||
1639 | void BlueStore::OnodeSpace::rename( | |
1640 | OnodeRef& oldo, | |
1641 | const ghobject_t& old_oid, | |
1642 | const ghobject_t& new_oid, | |
31f18b77 | 1643 | const mempool::bluestore_cache_other::string& new_okey) |
7c673cae | 1644 | { |
11fdf7f2 | 1645 | std::lock_guard l(cache->lock); |
7c673cae FG |
1646 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
1647 | << dendl; | |
1648 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
1649 | po = onode_map.find(old_oid); | |
1650 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 1651 | ceph_assert(po != pn); |
7c673cae | 1652 | |
11fdf7f2 | 1653 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
1654 | if (pn != onode_map.end()) { |
1655 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
1656 | << dendl; | |
1657 | cache->_rm_onode(pn->second); | |
1658 | onode_map.erase(pn); | |
1659 | } | |
1660 | OnodeRef o = po->second; | |
1661 | ||
1662 | // install a non-existent onode at old location | |
1663 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
1664 | po->second = oldo; | |
1665 | cache->_add_onode(po->second, 1); | |
1666 | ||
1667 | // add at new position and fix oid, key | |
1668 | onode_map.insert(make_pair(new_oid, o)); | |
1669 | cache->_touch_onode(o); | |
1670 | o->oid = new_oid; | |
1671 | o->key = new_okey; | |
1672 | } | |
1673 | ||
1674 | bool BlueStore::OnodeSpace::map_any(std::function<bool(OnodeRef)> f) | |
1675 | { | |
11fdf7f2 | 1676 | std::lock_guard l(cache->lock); |
7c673cae FG |
1677 | ldout(cache->cct, 20) << __func__ << dendl; |
1678 | for (auto& i : onode_map) { | |
1679 | if (f(i.second)) { | |
1680 | return true; | |
1681 | } | |
1682 | } | |
1683 | return false; | |
1684 | } | |
1685 | ||
11fdf7f2 TL |
1686 | template <int LogLevelV = 30> |
1687 | void BlueStore::OnodeSpace::dump(CephContext *cct) | |
3efd9988 FG |
1688 | { |
1689 | for (auto& i : onode_map) { | |
11fdf7f2 | 1690 | ldout(cct, LogLevelV) << i.first << " : " << i.second << dendl; |
3efd9988 FG |
1691 | } |
1692 | } | |
7c673cae FG |
1693 | |
1694 | // SharedBlob | |
1695 | ||
1696 | #undef dout_prefix | |
1697 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
1698 | ||
1699 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
1700 | { | |
1701 | out << "SharedBlob(" << &sb; | |
1702 | ||
1703 | if (sb.loaded) { | |
1704 | out << " loaded " << *sb.persistent; | |
1705 | } else { | |
1706 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
1707 | } | |
1708 | return out << ")"; | |
1709 | } | |
1710 | ||
1711 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
1712 | : coll(_coll), sbid_unloaded(i) | |
1713 | { | |
11fdf7f2 | 1714 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
1715 | if (get_cache()) { |
1716 | get_cache()->add_blob(); | |
1717 | } | |
1718 | } | |
1719 | ||
1720 | BlueStore::SharedBlob::~SharedBlob() | |
1721 | { | |
7c673cae FG |
1722 | if (loaded && persistent) { |
1723 | delete persistent; | |
1724 | } | |
1725 | } | |
1726 | ||
1727 | void BlueStore::SharedBlob::put() | |
1728 | { | |
1729 | if (--nref == 0) { | |
1730 | ldout(coll->store->cct, 20) << __func__ << " " << this | |
1731 | << " removing self from set " << get_parent() | |
1732 | << dendl; | |
1adf2230 AA |
1733 | again: |
1734 | auto coll_snap = coll; | |
1735 | if (coll_snap) { | |
11fdf7f2 | 1736 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
1737 | if (coll_snap != coll) { |
1738 | goto again; | |
1739 | } | |
91327a77 AA |
1740 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
1741 | // race with lookup | |
1742 | return; | |
1743 | } | |
1adf2230 AA |
1744 | bc._clear(coll_snap->cache); |
1745 | coll_snap->cache->rm_blob(); | |
7c673cae | 1746 | } |
28e407b8 | 1747 | delete this; |
7c673cae FG |
1748 | } |
1749 | } | |
1750 | ||
1751 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
1752 | { | |
11fdf7f2 | 1753 | ceph_assert(persistent); |
7c673cae FG |
1754 | persistent->ref_map.get(offset, length); |
1755 | } | |
1756 | ||
1757 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 1758 | PExtentVector *r, |
11fdf7f2 | 1759 | bool *unshare) |
7c673cae | 1760 | { |
11fdf7f2 TL |
1761 | ceph_assert(persistent); |
1762 | persistent->ref_map.put(offset, length, r, | |
1763 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
1764 | } |
1765 | ||
f64942e4 AA |
1766 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
1767 | { | |
1768 | while (true) { | |
1769 | Cache *cache = coll->cache; | |
11fdf7f2 | 1770 | std::lock_guard l(cache->lock); |
f64942e4 AA |
1771 | if (coll->cache != cache) { |
1772 | ldout(coll->store->cct, 20) << __func__ | |
1773 | << " raced with sb cache update, was " << cache | |
1774 | << ", now " << coll->cache << ", retrying" | |
1775 | << dendl; | |
1776 | continue; | |
1777 | } | |
1778 | bc._finish_write(cache, seq); | |
1779 | break; | |
1780 | } | |
1781 | } | |
1782 | ||
3efd9988 FG |
1783 | // SharedBlobSet |
1784 | ||
1785 | #undef dout_prefix | |
1786 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
1787 | ||
11fdf7f2 TL |
1788 | template <int LogLevelV = 30> |
1789 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 1790 | { |
11fdf7f2 | 1791 | std::lock_guard l(lock); |
3efd9988 | 1792 | for (auto& i : sb_map) { |
11fdf7f2 | 1793 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
1794 | } |
1795 | } | |
1796 | ||
7c673cae FG |
1797 | // Blob |
1798 | ||
1799 | #undef dout_prefix | |
1800 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
1801 | ||
1802 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) | |
1803 | { | |
1804 | out << "Blob(" << &b; | |
1805 | if (b.is_spanning()) { | |
1806 | out << " spanning " << b.id; | |
1807 | } | |
35e4c445 FG |
1808 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
1809 | if (b.shared_blob) { | |
1810 | out << " " << *b.shared_blob; | |
1811 | } else { | |
1812 | out << " (shared_blob=NULL)"; | |
1813 | } | |
1814 | out << ")"; | |
7c673cae FG |
1815 | return out; |
1816 | } | |
1817 | ||
1818 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
1819 | { | |
224ce89b | 1820 | if (get_blob().is_shared()) { |
7c673cae FG |
1821 | return; |
1822 | } | |
224ce89b | 1823 | if (get_blob().is_compressed()) { |
7c673cae FG |
1824 | bool discard = false; |
1825 | bool all_invalid = true; | |
224ce89b | 1826 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
1827 | if (!e.is_valid()) { |
1828 | discard = true; | |
1829 | } else { | |
1830 | all_invalid = false; | |
1831 | } | |
1832 | } | |
11fdf7f2 | 1833 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
1834 | // or none pextents are invalid. |
1835 | if (discard) { | |
224ce89b WB |
1836 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
1837 | get_blob().get_logical_length()); | |
7c673cae FG |
1838 | } |
1839 | } else { | |
1840 | size_t pos = 0; | |
224ce89b | 1841 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
1842 | if (!e.is_valid()) { |
1843 | ldout(coll->store->cct, 20) << __func__ << " 0x" << std::hex << pos | |
1844 | << "~" << e.length | |
1845 | << std::dec << dendl; | |
1846 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); | |
1847 | } | |
1848 | pos += e.length; | |
1849 | } | |
224ce89b WB |
1850 | if (get_blob().can_prune_tail()) { |
1851 | dirty_blob().prune_tail(); | |
1852 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
7c673cae | 1853 | auto cct = coll->store->cct; //used by dout |
224ce89b | 1854 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
1855 | } |
1856 | } | |
1857 | } | |
1858 | ||
1859 | void BlueStore::Blob::get_ref( | |
1860 | Collection *coll, | |
1861 | uint32_t offset, | |
1862 | uint32_t length) | |
1863 | { | |
1864 | // Caller has to initialize Blob's logical length prior to increment | |
1865 | // references. Otherwise one is neither unable to determine required | |
1866 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
1867 | // for single counter mode. | |
11fdf7f2 | 1868 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
1869 | auto cct = coll->store->cct; |
1870 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
1871 | << std::dec << " " << *this << dendl; | |
1872 | ||
1873 | if (used_in_blob.is_empty()) { | |
1874 | uint32_t min_release_size = | |
224ce89b WB |
1875 | get_blob().get_release_size(coll->store->min_alloc_size); |
1876 | uint64_t l = get_blob().get_logical_length(); | |
1877 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
1878 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
1879 | used_in_blob.init(l, min_release_size); |
1880 | } | |
1881 | used_in_blob.get( | |
1882 | offset, | |
1883 | length); | |
1884 | } | |
1885 | ||
1886 | bool BlueStore::Blob::put_ref( | |
1887 | Collection *coll, | |
1888 | uint32_t offset, | |
1889 | uint32_t length, | |
1890 | PExtentVector *r) | |
1891 | { | |
1892 | PExtentVector logical; | |
1893 | ||
1894 | auto cct = coll->store->cct; | |
1895 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
1896 | << std::dec << " " << *this << dendl; | |
1897 | ||
1898 | bool empty = used_in_blob.put( | |
1899 | offset, | |
1900 | length, | |
1901 | &logical); | |
1902 | r->clear(); | |
1903 | // nothing to release | |
1904 | if (!empty && logical.empty()) { | |
1905 | return false; | |
1906 | } | |
1907 | ||
1908 | bluestore_blob_t& b = dirty_blob(); | |
1909 | return b.release_extents(empty, logical, r); | |
1910 | } | |
1911 | ||
224ce89b | 1912 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
1913 | uint32_t target_blob_size, |
1914 | uint32_t b_offset, | |
1915 | uint32_t *length0) { | |
11fdf7f2 TL |
1916 | ceph_assert(min_alloc_size); |
1917 | ceph_assert(target_blob_size); | |
7c673cae FG |
1918 | if (!get_blob().is_mutable()) { |
1919 | return false; | |
1920 | } | |
1921 | ||
1922 | uint32_t length = *length0; | |
1923 | uint32_t end = b_offset + length; | |
1924 | ||
1925 | // Currently for the sake of simplicity we omit blob reuse if data is | |
1926 | // unaligned with csum chunk. Later we can perform padding if needed. | |
1927 | if (get_blob().has_csum() && | |
1928 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
1929 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
1930 | return false; | |
1931 | } | |
1932 | ||
1933 | auto blen = get_blob().get_logical_length(); | |
1934 | uint32_t new_blen = blen; | |
1935 | ||
1936 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 1937 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
1938 | |
1939 | if (b_offset >= blen) { | |
224ce89b WB |
1940 | // new data totally stands out of the existing blob |
1941 | new_blen = end; | |
7c673cae | 1942 | } else { |
224ce89b | 1943 | // new data overlaps with the existing blob |
11fdf7f2 | 1944 | new_blen = std::max(blen, end); |
224ce89b WB |
1945 | |
1946 | uint32_t overlap = 0; | |
1947 | if (new_blen > blen) { | |
1948 | overlap = blen - b_offset; | |
1949 | } else { | |
1950 | overlap = length; | |
1951 | } | |
1952 | ||
1953 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
1954 | // abort if any piece of the overlap has already been allocated | |
1955 | return false; | |
7c673cae FG |
1956 | } |
1957 | } | |
224ce89b | 1958 | |
7c673cae FG |
1959 | if (new_blen > blen) { |
1960 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
1961 | // Unable to decrease the provided length to fit into max_blob_size | |
1962 | if (overflow >= length) { | |
1963 | return false; | |
1964 | } | |
1965 | ||
1966 | // FIXME: in some cases we could reduce unused resolution | |
1967 | if (get_blob().has_unused()) { | |
1968 | return false; | |
1969 | } | |
1970 | ||
1971 | if (overflow > 0) { | |
1972 | new_blen -= overflow; | |
1973 | length -= overflow; | |
1974 | *length0 = length; | |
1975 | } | |
224ce89b | 1976 | |
7c673cae FG |
1977 | if (new_blen > blen) { |
1978 | dirty_blob().add_tail(new_blen); | |
1979 | used_in_blob.add_tail(new_blen, | |
224ce89b | 1980 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
1981 | } |
1982 | } | |
1983 | return true; | |
1984 | } | |
1985 | ||
1986 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
1987 | { | |
1988 | auto cct = coll->store->cct; //used by dout | |
1989 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
1990 | << " start " << *this << dendl; | |
11fdf7f2 TL |
1991 | ceph_assert(blob.can_split()); |
1992 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
1993 | bluestore_blob_t &lb = dirty_blob(); |
1994 | bluestore_blob_t &rb = r->dirty_blob(); | |
1995 | ||
1996 | used_in_blob.split( | |
1997 | blob_offset, | |
1998 | &(r->used_in_blob)); | |
1999 | ||
2000 | lb.split(blob_offset, rb); | |
2001 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2002 | ||
2003 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2004 | << " finish " << *this << dendl; | |
2005 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2006 | << " and " << *r << dendl; | |
2007 | } | |
2008 | ||
2009 | #ifndef CACHE_BLOB_BL | |
2010 | void BlueStore::Blob::decode( | |
2011 | Collection *coll, | |
11fdf7f2 | 2012 | bufferptr::const_iterator& p, |
7c673cae FG |
2013 | uint64_t struct_v, |
2014 | uint64_t* sbid, | |
2015 | bool include_ref_map) | |
2016 | { | |
2017 | denc(blob, p, struct_v); | |
2018 | if (blob.is_shared()) { | |
2019 | denc(*sbid, p); | |
2020 | } | |
2021 | if (include_ref_map) { | |
2022 | if (struct_v > 1) { | |
2023 | used_in_blob.decode(p); | |
2024 | } else { | |
2025 | used_in_blob.clear(); | |
2026 | bluestore_extent_ref_map_t legacy_ref_map; | |
2027 | legacy_ref_map.decode(p); | |
2028 | for (auto r : legacy_ref_map.ref_map) { | |
2029 | get_ref( | |
2030 | coll, | |
2031 | r.first, | |
2032 | r.second.refs * r.second.length); | |
2033 | } | |
2034 | } | |
2035 | } | |
2036 | } | |
2037 | #endif | |
2038 | ||
2039 | // Extent | |
2040 | ||
2041 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) | |
2042 | { | |
2043 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2044 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2045 | << " " << *e.blob; | |
2046 | } | |
2047 | ||
2048 | // OldExtent | |
2049 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2050 | uint32_t lo, | |
2051 | uint32_t o, | |
2052 | uint32_t l, | |
2053 | BlobRef& b) { | |
2054 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2055 | b->put_ref(c.get(), o, l, &(oe->r)); | |
2056 | oe->blob_empty = b->get_referenced_bytes() == 0; | |
2057 | return oe; | |
2058 | } | |
2059 | ||
2060 | // ExtentMap | |
2061 | ||
2062 | #undef dout_prefix | |
2063 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
2064 | ||
2065 | BlueStore::ExtentMap::ExtentMap(Onode *o) | |
2066 | : onode(o), | |
2067 | inline_bl( | |
2068 | o->c->store->cct->_conf->bluestore_extent_map_inline_shard_prealloc_size) { | |
2069 | } | |
2070 | ||
11fdf7f2 TL |
2071 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2072 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2073 | uint64_t& length, uint64_t& dstoff) { | |
2074 | ||
2075 | auto cct = onode->c->store->cct; | |
2076 | bool inject_21040 = | |
2077 | cct->_conf->bluestore_debug_inject_bug21040; | |
2078 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2079 | for (auto& e : oldo->extent_map.extent_map) { | |
2080 | e.blob->last_encoded_id = -1; | |
2081 | } | |
2082 | ||
2083 | int n = 0; | |
2084 | uint64_t end = srcoff + length; | |
2085 | uint32_t dirty_range_begin = 0; | |
2086 | uint32_t dirty_range_end = 0; | |
2087 | bool src_dirty = false; | |
2088 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2089 | ep != oldo->extent_map.extent_map.end(); | |
2090 | ++ep) { | |
2091 | auto& e = *ep; | |
2092 | if (e.logical_offset >= end) { | |
2093 | break; | |
2094 | } | |
2095 | dout(20) << __func__ << " src " << e << dendl; | |
2096 | BlobRef cb; | |
2097 | bool blob_duped = true; | |
2098 | if (e.blob->last_encoded_id >= 0) { | |
2099 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2100 | blob_duped = false; | |
2101 | } else { | |
2102 | // dup the blob | |
2103 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2104 | // make sure it is shared | |
2105 | if (!blob.is_shared()) { | |
2106 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2107 | if (!inject_21040 && !src_dirty) { | |
2108 | src_dirty = true; | |
2109 | dirty_range_begin = e.logical_offset; | |
2110 | } else if (inject_21040 && | |
2111 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2112 | dirty_range_begin = e.logical_offset; | |
2113 | } | |
2114 | ceph_assert(e.logical_end() > 0); | |
2115 | // -1 to exclude next potential shard | |
2116 | dirty_range_end = e.logical_end() - 1; | |
2117 | } else { | |
2118 | c->load_shared_blob(e.blob->shared_blob); | |
2119 | } | |
2120 | cb = new Blob(); | |
2121 | e.blob->last_encoded_id = n; | |
2122 | id_to_blob[n] = cb; | |
2123 | e.blob->dup(*cb); | |
2124 | // bump the extent refs on the copied blob's extents | |
2125 | for (auto p : blob.get_extents()) { | |
2126 | if (p.is_valid()) { | |
2127 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2128 | } | |
2129 | } | |
2130 | txc->write_shared_blob(e.blob->shared_blob); | |
2131 | dout(20) << __func__ << " new " << *cb << dendl; | |
2132 | } | |
2133 | ||
2134 | int skip_front, skip_back; | |
2135 | if (e.logical_offset < srcoff) { | |
2136 | skip_front = srcoff - e.logical_offset; | |
2137 | } else { | |
2138 | skip_front = 0; | |
2139 | } | |
2140 | if (e.logical_end() > end) { | |
2141 | skip_back = e.logical_end() - end; | |
2142 | } else { | |
2143 | skip_back = 0; | |
2144 | } | |
2145 | ||
2146 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2147 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2148 | newo->extent_map.extent_map.insert(*ne); | |
2149 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2150 | // fixme: we may leave parts of new blob unreferenced that could | |
2151 | // be freed (relative to the shared_blob). | |
2152 | txc->statfs_delta.stored() += ne->length; | |
2153 | if (e.blob->get_blob().is_compressed()) { | |
2154 | txc->statfs_delta.compressed_original() += ne->length; | |
2155 | if (blob_duped) { | |
2156 | txc->statfs_delta.compressed() += | |
2157 | cb->get_blob().get_compressed_payload_length(); | |
2158 | } | |
2159 | } | |
2160 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2161 | ++n; | |
2162 | } | |
2163 | if ((!inject_21040 && src_dirty) || | |
2164 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2165 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2166 | dirty_range_end - dirty_range_begin); | |
2167 | txc->write_onode(oldo); | |
2168 | } | |
2169 | txc->write_onode(newo); | |
2170 | ||
2171 | if (dstoff + length > newo->onode.size) { | |
2172 | newo->onode.size = dstoff + length; | |
2173 | } | |
2174 | newo->extent_map.dirty_range(dstoff, length); | |
2175 | } | |
7c673cae FG |
2176 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2177 | bool force) | |
2178 | { | |
2179 | auto cct = onode->c->store->cct; //used by dout | |
2180 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2181 | if (onode->onode.extent_map_shards.empty()) { | |
2182 | if (inline_bl.length() == 0) { | |
2183 | unsigned n; | |
2184 | // we need to encode inline_bl to measure encoded length | |
2185 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
3efd9988 | 2186 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_cache_other); |
11fdf7f2 | 2187 | ceph_assert(!never_happen); |
7c673cae FG |
2188 | size_t len = inline_bl.length(); |
2189 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2190 | << " extents" << dendl; | |
2191 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2192 | request_reshard(0, OBJECT_MAX_SIZE); | |
2193 | return; | |
2194 | } | |
2195 | } | |
2196 | // will persist in the onode key. | |
2197 | } else { | |
2198 | // pending shard update | |
2199 | struct dirty_shard_t { | |
2200 | Shard *shard; | |
2201 | bufferlist bl; | |
2202 | dirty_shard_t(Shard *s) : shard(s) {} | |
2203 | }; | |
2204 | vector<dirty_shard_t> encoded_shards; | |
2205 | // allocate slots for all shards in a single call instead of | |
2206 | // doing multiple allocations - one per each dirty shard | |
2207 | encoded_shards.reserve(shards.size()); | |
2208 | ||
2209 | auto p = shards.begin(); | |
2210 | auto prev_p = p; | |
2211 | while (p != shards.end()) { | |
11fdf7f2 | 2212 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2213 | auto n = p; |
2214 | ++n; | |
2215 | if (p->dirty) { | |
2216 | uint32_t endoff; | |
2217 | if (n == shards.end()) { | |
2218 | endoff = OBJECT_MAX_SIZE; | |
2219 | } else { | |
2220 | endoff = n->shard_info->offset; | |
2221 | } | |
2222 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2223 | bufferlist& bl = encoded_shards.back().bl; | |
2224 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2225 | bl, &p->extents)) { | |
2226 | if (force) { | |
2227 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2228 | ceph_assert(!force); |
7c673cae FG |
2229 | } |
2230 | } | |
2231 | size_t len = bl.length(); | |
2232 | ||
2233 | dout(20) << __func__ << " shard 0x" << std::hex | |
2234 | << p->shard_info->offset << std::dec << " is " << len | |
2235 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2236 | << p->extents << " extents" << dendl; | |
2237 | ||
2238 | if (!force) { | |
2239 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2240 | // we are big; reshard ourselves | |
2241 | request_reshard(p->shard_info->offset, endoff); | |
2242 | } | |
2243 | // avoid resharding the trailing shard, even if it is small | |
2244 | else if (n != shards.end() && | |
11fdf7f2 TL |
2245 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2246 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2247 | if (p == shards.begin()) { |
2248 | // we are the first shard, combine with next shard | |
7c673cae | 2249 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2250 | } else { |
31f18b77 FG |
2251 | // combine either with the previous shard or the next, |
2252 | // whichever is smaller | |
7c673cae FG |
2253 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2254 | request_reshard(p->shard_info->offset, endoff + 1); | |
2255 | } else { | |
2256 | request_reshard(prev_p->shard_info->offset, endoff); | |
2257 | } | |
2258 | } | |
2259 | } | |
2260 | } | |
2261 | } | |
2262 | prev_p = p; | |
2263 | p = n; | |
2264 | } | |
2265 | if (needs_reshard()) { | |
2266 | return; | |
2267 | } | |
2268 | ||
2269 | // schedule DB update for dirty shards | |
2270 | string key; | |
2271 | for (auto& it : encoded_shards) { | |
2272 | it.shard->dirty = false; | |
2273 | it.shard->shard_info->bytes = it.bl.length(); | |
2274 | generate_extent_shard_key_and_apply( | |
2275 | onode->key, | |
2276 | it.shard->shard_info->offset, | |
2277 | &key, | |
2278 | [&](const string& final_key) { | |
2279 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2280 | } | |
2281 | ); | |
2282 | } | |
2283 | } | |
2284 | } | |
2285 | ||
31f18b77 FG |
2286 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2287 | { | |
2288 | if (spanning_blob_map.empty()) | |
2289 | return 0; | |
2290 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2291 | // bid is valid and available. | |
2292 | if (bid >= 0) | |
2293 | return bid; | |
2294 | // Find next unused bid; | |
2295 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2296 | const auto begin_bid = bid; | |
2297 | do { | |
2298 | if (!spanning_blob_map.count(bid)) | |
2299 | return bid; | |
2300 | else { | |
2301 | bid++; | |
2302 | if (bid < 0) bid = 0; | |
2303 | } | |
2304 | } while (bid != begin_bid); | |
81eedcae TL |
2305 | auto cct = onode->c->store->cct; // used by dout |
2306 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2307 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2308 | } |
2309 | ||
7c673cae FG |
2310 | void BlueStore::ExtentMap::reshard( |
2311 | KeyValueDB *db, | |
2312 | KeyValueDB::Transaction t) | |
2313 | { | |
2314 | auto cct = onode->c->store->cct; // used by dout | |
2315 | ||
2316 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2317 | << needs_reshard_end << ")" << std::dec | |
2318 | << " of " << onode->onode.extent_map_shards.size() | |
2319 | << " shards on " << onode->oid << dendl; | |
2320 | for (auto& p : spanning_blob_map) { | |
2321 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2322 | << dendl; | |
2323 | } | |
2324 | // determine shard index range | |
2325 | unsigned si_begin = 0, si_end = 0; | |
2326 | if (!shards.empty()) { | |
2327 | while (si_begin + 1 < shards.size() && | |
2328 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2329 | ++si_begin; | |
2330 | } | |
2331 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2332 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2333 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2334 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2335 | break; | |
2336 | } | |
2337 | } | |
2338 | if (si_end == shards.size()) { | |
2339 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2340 | } | |
2341 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2342 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2343 | << needs_reshard_end << ")" << std::dec << dendl; | |
2344 | } | |
2345 | ||
181888fb | 2346 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2347 | |
2348 | // we may need to fault in a larger interval later must have all | |
2349 | // referring extents for spanning blobs loaded in order to have | |
2350 | // accurate use_tracker values. | |
2351 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2352 | uint32_t spanning_scan_end = needs_reshard_end; | |
2353 | ||
2354 | // remove old keys | |
2355 | string key; | |
2356 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2357 | generate_extent_shard_key_and_apply( | |
2358 | onode->key, shards[i].shard_info->offset, &key, | |
2359 | [&](const string& final_key) { | |
2360 | t->rmkey(PREFIX_OBJ, final_key); | |
2361 | } | |
2362 | ); | |
2363 | } | |
2364 | ||
2365 | // calculate average extent size | |
2366 | unsigned bytes = 0; | |
2367 | unsigned extents = 0; | |
2368 | if (onode->onode.extent_map_shards.empty()) { | |
2369 | bytes = inline_bl.length(); | |
2370 | extents = extent_map.size(); | |
2371 | } else { | |
2372 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2373 | bytes += shards[i].shard_info->bytes; | |
2374 | extents += shards[i].extents; | |
2375 | } | |
2376 | } | |
2377 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2378 | unsigned slop = target * | |
2379 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2380 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2381 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2382 | << ", slop " << slop << dendl; | |
2383 | ||
2384 | // reshard | |
2385 | unsigned estimate = 0; | |
31f18b77 | 2386 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2387 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2388 | unsigned max_blob_end = 0; | |
2389 | Extent dummy(needs_reshard_begin); | |
2390 | for (auto e = extent_map.lower_bound(dummy); | |
2391 | e != extent_map.end(); | |
2392 | ++e) { | |
2393 | if (e->logical_offset >= needs_reshard_end) { | |
2394 | break; | |
2395 | } | |
2396 | dout(30) << " extent " << *e << dendl; | |
2397 | ||
2398 | // disfavor shard boundaries that span a blob | |
2399 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2400 | if (estimate && | |
2401 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2402 | // new shard | |
31f18b77 | 2403 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2404 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2405 | new_shard_info.back().offset = offset; | |
2406 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2407 | << std::dec << dendl; |
7c673cae FG |
2408 | } |
2409 | offset = e->logical_offset; | |
2410 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2411 | new_shard_info.back().offset = offset; | |
2412 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2413 | << std::dec << dendl; | |
2414 | estimate = 0; | |
2415 | } | |
2416 | estimate += extent_avg; | |
31f18b77 FG |
2417 | unsigned bs = e->blob_start(); |
2418 | if (bs < spanning_scan_begin) { | |
2419 | spanning_scan_begin = bs; | |
7c673cae FG |
2420 | } |
2421 | uint32_t be = e->blob_end(); | |
2422 | if (be > max_blob_end) { | |
2423 | max_blob_end = be; | |
2424 | } | |
2425 | if (be > spanning_scan_end) { | |
2426 | spanning_scan_end = be; | |
2427 | } | |
2428 | } | |
2429 | if (new_shard_info.empty() && (si_begin > 0 || | |
2430 | si_end < shards.size())) { | |
2431 | // we resharded a partial range; we must produce at least one output | |
2432 | // shard | |
2433 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2434 | new_shard_info.back().offset = needs_reshard_begin; | |
2435 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2436 | << std::dec << " (singleton degenerate case)" << dendl; | |
2437 | } | |
2438 | ||
2439 | auto& sv = onode->onode.extent_map_shards; | |
2440 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2441 | dout(20) << __func__ << " old " << sv << dendl; | |
2442 | if (sv.empty()) { | |
2443 | // no old shards to keep | |
2444 | sv.swap(new_shard_info); | |
2445 | init_shards(true, true); | |
2446 | } else { | |
2447 | // splice in new shards | |
2448 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2449 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2450 | sv.insert( | |
2451 | sv.begin() + si_begin, | |
2452 | new_shard_info.begin(), | |
2453 | new_shard_info.end()); | |
2454 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2455 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2456 | |
11fdf7f2 | 2457 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2458 | |
2459 | // note that we need to update every shard_info of shards here, | |
2460 | // as sv might have been totally re-allocated above | |
2461 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2462 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2463 | } |
2464 | ||
2465 | // mark newly added shards as dirty | |
2466 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2467 | shards[i].loaded = true; |
2468 | shards[i].dirty = true; | |
2469 | } | |
7c673cae FG |
2470 | } |
2471 | dout(20) << __func__ << " fin " << sv << dendl; | |
2472 | inline_bl.clear(); | |
2473 | ||
2474 | if (sv.empty()) { | |
2475 | // no more shards; unspan all previously spanning blobs | |
2476 | auto p = spanning_blob_map.begin(); | |
2477 | while (p != spanning_blob_map.end()) { | |
2478 | p->second->id = -1; | |
2479 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2480 | p = spanning_blob_map.erase(p); | |
2481 | } | |
2482 | } else { | |
2483 | // identify new spanning blobs | |
2484 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2485 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2486 | if (spanning_scan_begin < needs_reshard_begin) { | |
2487 | fault_range(db, spanning_scan_begin, | |
2488 | needs_reshard_begin - spanning_scan_begin); | |
2489 | } | |
2490 | if (spanning_scan_end > needs_reshard_end) { | |
2491 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2492 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2493 | } |
2494 | auto sp = sv.begin() + si_begin; | |
2495 | auto esp = sv.end(); | |
2496 | unsigned shard_start = sp->offset; | |
2497 | unsigned shard_end; | |
2498 | ++sp; | |
2499 | if (sp == esp) { | |
2500 | shard_end = OBJECT_MAX_SIZE; | |
2501 | } else { | |
2502 | shard_end = sp->offset; | |
2503 | } | |
7c673cae FG |
2504 | Extent dummy(needs_reshard_begin); |
2505 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { | |
2506 | if (e->logical_offset >= needs_reshard_end) { | |
2507 | break; | |
2508 | } | |
2509 | dout(30) << " extent " << *e << dendl; | |
2510 | while (e->logical_offset >= shard_end) { | |
2511 | shard_start = shard_end; | |
11fdf7f2 | 2512 | ceph_assert(sp != esp); |
7c673cae FG |
2513 | ++sp; |
2514 | if (sp == esp) { | |
2515 | shard_end = OBJECT_MAX_SIZE; | |
2516 | } else { | |
2517 | shard_end = sp->offset; | |
2518 | } | |
2519 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2520 | << " to 0x" << shard_end << std::dec << dendl; | |
2521 | } | |
2522 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { | |
2523 | if (!e->blob->is_spanning()) { | |
2524 | // We have two options: (1) split the blob into pieces at the | |
2525 | // shard boundaries (and adjust extents accordingly), or (2) | |
2526 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2527 | // we may have to split it multiple times--potentially at every | |
2528 | // shard boundary. | |
2529 | bool must_span = false; | |
2530 | BlobRef b = e->blob; | |
2531 | if (b->can_split()) { | |
2532 | uint32_t bstart = e->blob_start(); | |
2533 | uint32_t bend = e->blob_end(); | |
2534 | for (const auto& sh : shards) { | |
2535 | if (bstart < sh.shard_info->offset && | |
2536 | bend > sh.shard_info->offset) { | |
2537 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2538 | if (b->can_split_at(blob_offset)) { | |
2539 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2540 | << std::hex << bstart << " blob_offset 0x" | |
2541 | << blob_offset << std::dec << " " << *b << dendl; | |
2542 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2543 | // switch b to the new right-hand side, in case it | |
2544 | // *also* has to get split. | |
2545 | bstart += blob_offset; | |
2546 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2547 | } else { | |
2548 | must_span = true; | |
2549 | break; | |
2550 | } | |
2551 | } | |
2552 | } | |
2553 | } else { | |
2554 | must_span = true; | |
2555 | } | |
2556 | if (must_span) { | |
31f18b77 FG |
2557 | auto bid = allocate_spanning_blob_id(); |
2558 | b->id = bid; | |
7c673cae FG |
2559 | spanning_blob_map[b->id] = b; |
2560 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
2561 | } | |
2562 | } | |
2563 | } else { | |
2564 | if (e->blob->is_spanning()) { | |
2565 | spanning_blob_map.erase(e->blob->id); | |
2566 | e->blob->id = -1; | |
2567 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
2568 | } | |
2569 | } | |
2570 | } | |
2571 | } | |
2572 | ||
2573 | clear_needs_reshard(); | |
2574 | } | |
2575 | ||
2576 | bool BlueStore::ExtentMap::encode_some( | |
2577 | uint32_t offset, | |
2578 | uint32_t length, | |
2579 | bufferlist& bl, | |
2580 | unsigned *pn) | |
2581 | { | |
2582 | auto cct = onode->c->store->cct; //used by dout | |
2583 | Extent dummy(offset); | |
2584 | auto start = extent_map.lower_bound(dummy); | |
2585 | uint32_t end = offset + length; | |
2586 | ||
2587 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
2588 | // serialization only. Hence there is no specific | |
2589 | // handling at ExtentMap level. | |
2590 | ||
2591 | unsigned n = 0; | |
2592 | size_t bound = 0; | |
7c673cae FG |
2593 | bool must_reshard = false; |
2594 | for (auto p = start; | |
2595 | p != extent_map.end() && p->logical_offset < end; | |
2596 | ++p, ++n) { | |
11fdf7f2 | 2597 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
2598 | p->blob->last_encoded_id = -1; |
2599 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
2600 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2601 | << std::dec << " hit new spanning blob " << *p << dendl; | |
2602 | request_reshard(p->blob_start(), p->blob_end()); | |
2603 | must_reshard = true; | |
2604 | } | |
31f18b77 FG |
2605 | if (!must_reshard) { |
2606 | denc_varint(0, bound); // blobid | |
2607 | denc_varint(0, bound); // logical_offset | |
2608 | denc_varint(0, bound); // len | |
2609 | denc_varint(0, bound); // blob_offset | |
7c673cae | 2610 | |
31f18b77 FG |
2611 | p->blob->bound_encode( |
2612 | bound, | |
2613 | struct_v, | |
2614 | p->blob->shared_blob->get_sbid(), | |
2615 | false); | |
2616 | } | |
7c673cae FG |
2617 | } |
2618 | if (must_reshard) { | |
2619 | return true; | |
2620 | } | |
2621 | ||
31f18b77 FG |
2622 | denc(struct_v, bound); |
2623 | denc_varint(0, bound); // number of extents | |
2624 | ||
7c673cae FG |
2625 | { |
2626 | auto app = bl.get_contiguous_appender(bound); | |
2627 | denc(struct_v, app); | |
2628 | denc_varint(n, app); | |
2629 | if (pn) { | |
2630 | *pn = n; | |
2631 | } | |
2632 | ||
2633 | n = 0; | |
2634 | uint64_t pos = 0; | |
2635 | uint64_t prev_len = 0; | |
2636 | for (auto p = start; | |
2637 | p != extent_map.end() && p->logical_offset < end; | |
2638 | ++p, ++n) { | |
2639 | unsigned blobid; | |
2640 | bool include_blob = false; | |
2641 | if (p->blob->is_spanning()) { | |
2642 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
2643 | blobid |= BLOBID_FLAG_SPANNING; | |
2644 | } else if (p->blob->last_encoded_id < 0) { | |
2645 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
2646 | include_blob = true; | |
2647 | blobid = 0; // the decoder will infer the id from n | |
2648 | } else { | |
2649 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
2650 | } | |
2651 | if (p->logical_offset == pos) { | |
2652 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
2653 | } | |
2654 | if (p->blob_offset == 0) { | |
2655 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
2656 | } | |
2657 | if (p->length == prev_len) { | |
2658 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
2659 | } else { | |
2660 | prev_len = p->length; | |
2661 | } | |
2662 | denc_varint(blobid, app); | |
2663 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
2664 | denc_varint_lowz(p->logical_offset - pos, app); | |
2665 | } | |
2666 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
2667 | denc_varint_lowz(p->blob_offset, app); | |
2668 | } | |
2669 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
2670 | denc_varint_lowz(p->length, app); | |
2671 | } | |
2672 | pos = p->logical_end(); | |
2673 | if (include_blob) { | |
2674 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
2675 | } | |
2676 | } | |
2677 | } | |
2678 | /*derr << __func__ << bl << dendl; | |
2679 | derr << __func__ << ":"; | |
2680 | bl.hexdump(*_dout); | |
2681 | *_dout << dendl; | |
2682 | */ | |
2683 | return false; | |
2684 | } | |
2685 | ||
2686 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
2687 | { | |
2688 | auto cct = onode->c->store->cct; //used by dout | |
2689 | /* | |
2690 | derr << __func__ << ":"; | |
2691 | bl.hexdump(*_dout); | |
2692 | *_dout << dendl; | |
2693 | */ | |
2694 | ||
11fdf7f2 | 2695 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae FG |
2696 | auto p = bl.front().begin_deep(); |
2697 | __u8 struct_v; | |
2698 | denc(struct_v, p); | |
2699 | // Version 2 differs from v1 in blob's ref_map | |
2700 | // serialization only. Hence there is no specific | |
2701 | // handling at ExtentMap level below. | |
11fdf7f2 | 2702 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
2703 | |
2704 | uint32_t num; | |
2705 | denc_varint(num, p); | |
2706 | vector<BlobRef> blobs(num); | |
2707 | uint64_t pos = 0; | |
2708 | uint64_t prev_len = 0; | |
2709 | unsigned n = 0; | |
2710 | ||
2711 | while (!p.end()) { | |
2712 | Extent *le = new Extent(); | |
2713 | uint64_t blobid; | |
2714 | denc_varint(blobid, p); | |
2715 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
2716 | uint64_t gap; | |
2717 | denc_varint_lowz(gap, p); | |
2718 | pos += gap; | |
2719 | } | |
2720 | le->logical_offset = pos; | |
2721 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
2722 | denc_varint_lowz(le->blob_offset, p); | |
2723 | } else { | |
2724 | le->blob_offset = 0; | |
2725 | } | |
2726 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
2727 | denc_varint_lowz(prev_len, p); | |
2728 | } | |
2729 | le->length = prev_len; | |
2730 | ||
2731 | if (blobid & BLOBID_FLAG_SPANNING) { | |
2732 | dout(30) << __func__ << " getting spanning blob " | |
2733 | << (blobid >> BLOBID_SHIFT_BITS) << dendl; | |
2734 | le->assign_blob(get_spanning_blob(blobid >> BLOBID_SHIFT_BITS)); | |
2735 | } else { | |
2736 | blobid >>= BLOBID_SHIFT_BITS; | |
2737 | if (blobid) { | |
2738 | le->assign_blob(blobs[blobid - 1]); | |
11fdf7f2 | 2739 | ceph_assert(le->blob); |
7c673cae FG |
2740 | } else { |
2741 | Blob *b = new Blob(); | |
2742 | uint64_t sbid = 0; | |
2743 | b->decode(onode->c, p, struct_v, &sbid, false); | |
2744 | blobs[n] = b; | |
2745 | onode->c->open_shared_blob(sbid, b); | |
2746 | le->assign_blob(b); | |
2747 | } | |
2748 | // we build ref_map dynamically for non-spanning blobs | |
2749 | le->blob->get_ref( | |
2750 | onode->c, | |
2751 | le->blob_offset, | |
2752 | le->length); | |
2753 | } | |
2754 | pos += prev_len; | |
2755 | ++n; | |
2756 | extent_map.insert(*le); | |
2757 | } | |
2758 | ||
11fdf7f2 | 2759 | ceph_assert(n == num); |
7c673cae FG |
2760 | return num; |
2761 | } | |
2762 | ||
2763 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
2764 | { | |
2765 | // Version 2 differs from v1 in blob's ref_map | |
2766 | // serialization only. Hence there is no specific | |
2767 | // handling at ExtentMap level. | |
2768 | __u8 struct_v = 2; | |
2769 | ||
2770 | denc(struct_v, p); | |
2771 | denc_varint((uint32_t)0, p); | |
2772 | size_t key_size = 0; | |
2773 | denc_varint((uint32_t)0, key_size); | |
2774 | p += spanning_blob_map.size() * key_size; | |
2775 | for (const auto& i : spanning_blob_map) { | |
2776 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
2777 | } | |
2778 | } | |
2779 | ||
2780 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
2781 | bufferlist::contiguous_appender& p) | |
2782 | { | |
2783 | // Version 2 differs from v1 in blob's ref_map | |
2784 | // serialization only. Hence there is no specific | |
2785 | // handling at ExtentMap level. | |
2786 | __u8 struct_v = 2; | |
2787 | ||
2788 | denc(struct_v, p); | |
2789 | denc_varint(spanning_blob_map.size(), p); | |
2790 | for (auto& i : spanning_blob_map) { | |
2791 | denc_varint(i.second->id, p); | |
2792 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
2793 | } | |
2794 | } | |
2795 | ||
2796 | void BlueStore::ExtentMap::decode_spanning_blobs( | |
11fdf7f2 | 2797 | bufferptr::const_iterator& p) |
7c673cae FG |
2798 | { |
2799 | __u8 struct_v; | |
2800 | denc(struct_v, p); | |
2801 | // Version 2 differs from v1 in blob's ref_map | |
2802 | // serialization only. Hence there is no specific | |
2803 | // handling at ExtentMap level. | |
11fdf7f2 | 2804 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
2805 | |
2806 | unsigned n; | |
2807 | denc_varint(n, p); | |
2808 | while (n--) { | |
2809 | BlobRef b(new Blob()); | |
2810 | denc_varint(b->id, p); | |
2811 | spanning_blob_map[b->id] = b; | |
2812 | uint64_t sbid = 0; | |
2813 | b->decode(onode->c, p, struct_v, &sbid, true); | |
2814 | onode->c->open_shared_blob(sbid, b); | |
2815 | } | |
2816 | } | |
2817 | ||
2818 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) | |
2819 | { | |
2820 | shards.resize(onode->onode.extent_map_shards.size()); | |
2821 | unsigned i = 0; | |
2822 | for (auto &s : onode->onode.extent_map_shards) { | |
2823 | shards[i].shard_info = &s; | |
2824 | shards[i].loaded = loaded; | |
2825 | shards[i].dirty = dirty; | |
2826 | ++i; | |
2827 | } | |
2828 | } | |
2829 | ||
2830 | void BlueStore::ExtentMap::fault_range( | |
2831 | KeyValueDB *db, | |
2832 | uint32_t offset, | |
2833 | uint32_t length) | |
2834 | { | |
2835 | auto cct = onode->c->store->cct; //used by dout | |
2836 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2837 | << std::dec << dendl; | |
2838 | auto start = seek_shard(offset); | |
2839 | auto last = seek_shard(offset + length); | |
2840 | ||
2841 | if (start < 0) | |
2842 | return; | |
2843 | ||
11fdf7f2 | 2844 | ceph_assert(last >= start); |
7c673cae FG |
2845 | string key; |
2846 | while (start <= last) { | |
11fdf7f2 | 2847 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
2848 | auto p = &shards[start]; |
2849 | if (!p->loaded) { | |
2850 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
2851 | << p->shard_info->offset << std::dec << dendl; | |
2852 | bufferlist v; | |
2853 | generate_extent_shard_key_and_apply( | |
2854 | onode->key, p->shard_info->offset, &key, | |
2855 | [&](const string& final_key) { | |
2856 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
2857 | if (r < 0) { | |
2858 | derr << __func__ << " missing shard 0x" << std::hex | |
2859 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
2860 | << dendl; | |
11fdf7f2 | 2861 | ceph_assert(r >= 0); |
7c673cae FG |
2862 | } |
2863 | } | |
2864 | ); | |
2865 | p->extents = decode_some(v); | |
2866 | p->loaded = true; | |
2867 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
2868 | << p->shard_info->offset |
2869 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 2870 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
2871 | ceph_assert(p->dirty == false); |
2872 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
2873 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
2874 | } else { | |
2875 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
2876 | } | |
2877 | ++start; | |
2878 | } | |
2879 | } | |
2880 | ||
2881 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
2882 | uint32_t offset, |
2883 | uint32_t length) | |
2884 | { | |
2885 | auto cct = onode->c->store->cct; //used by dout | |
2886 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2887 | << std::dec << dendl; | |
2888 | if (shards.empty()) { | |
2889 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
2890 | inline_bl.clear(); | |
2891 | return; | |
2892 | } | |
2893 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
2894 | if (length == 0) { |
2895 | length = 1; | |
2896 | } | |
2897 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
2898 | if (start < 0) |
2899 | return; | |
2900 | ||
11fdf7f2 | 2901 | ceph_assert(last >= start); |
7c673cae | 2902 | while (start <= last) { |
11fdf7f2 | 2903 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
2904 | auto p = &shards[start]; |
2905 | if (!p->loaded) { | |
11fdf7f2 TL |
2906 | derr << __func__ << "on write 0x" << std::hex << offset |
2907 | << "~" << length << " shard 0x" << p->shard_info->offset | |
2908 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
2909 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
2910 | } |
2911 | if (!p->dirty) { | |
2912 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
2913 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
2914 | p->dirty = true; | |
2915 | } | |
2916 | ++start; | |
2917 | } | |
2918 | } | |
2919 | ||
2920 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
2921 | uint64_t offset) | |
2922 | { | |
2923 | Extent dummy(offset); | |
2924 | return extent_map.find(dummy); | |
2925 | } | |
2926 | ||
7c673cae FG |
2927 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
2928 | uint64_t offset) | |
2929 | { | |
2930 | Extent dummy(offset); | |
2931 | auto fp = extent_map.lower_bound(dummy); | |
2932 | if (fp != extent_map.begin()) { | |
2933 | --fp; | |
2934 | if (fp->logical_end() <= offset) { | |
2935 | ++fp; | |
2936 | } | |
2937 | } | |
2938 | return fp; | |
2939 | } | |
2940 | ||
2941 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
2942 | uint64_t offset) const | |
2943 | { | |
2944 | Extent dummy(offset); | |
2945 | auto fp = extent_map.lower_bound(dummy); | |
2946 | if (fp != extent_map.begin()) { | |
2947 | --fp; | |
2948 | if (fp->logical_end() <= offset) { | |
2949 | ++fp; | |
2950 | } | |
2951 | } | |
2952 | return fp; | |
2953 | } | |
2954 | ||
2955 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
2956 | { | |
2957 | auto fp = seek_lextent(offset); | |
2958 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
2959 | return false; | |
2960 | } | |
2961 | return true; | |
2962 | } | |
2963 | ||
2964 | int BlueStore::ExtentMap::compress_extent_map( | |
2965 | uint64_t offset, | |
2966 | uint64_t length) | |
2967 | { | |
2968 | auto cct = onode->c->store->cct; //used by dout | |
2969 | if (extent_map.empty()) | |
2970 | return 0; | |
2971 | int removed = 0; | |
2972 | auto p = seek_lextent(offset); | |
2973 | if (p != extent_map.begin()) { | |
2974 | --p; // start to the left of offset | |
2975 | } | |
2976 | // the caller should have just written to this region | |
11fdf7f2 | 2977 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
2978 | |
2979 | // identify the *next* shard | |
2980 | auto pshard = shards.begin(); | |
2981 | while (pshard != shards.end() && | |
2982 | p->logical_offset >= pshard->shard_info->offset) { | |
2983 | ++pshard; | |
2984 | } | |
2985 | uint64_t shard_end; | |
2986 | if (pshard != shards.end()) { | |
2987 | shard_end = pshard->shard_info->offset; | |
2988 | } else { | |
2989 | shard_end = OBJECT_MAX_SIZE; | |
2990 | } | |
2991 | ||
2992 | auto n = p; | |
2993 | for (++n; n != extent_map.end(); p = n++) { | |
2994 | if (n->logical_offset > offset + length) { | |
2995 | break; // stop after end | |
2996 | } | |
2997 | while (n != extent_map.end() && | |
2998 | p->logical_end() == n->logical_offset && | |
2999 | p->blob == n->blob && | |
3000 | p->blob_offset + p->length == n->blob_offset && | |
3001 | n->logical_offset < shard_end) { | |
3002 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3003 | << " next shard 0x" << shard_end << std::dec | |
3004 | << " merging " << *p << " and " << *n << dendl; | |
3005 | p->length += n->length; | |
3006 | rm(n++); | |
3007 | ++removed; | |
3008 | } | |
3009 | if (n == extent_map.end()) { | |
3010 | break; | |
3011 | } | |
3012 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3013 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3014 | ++pshard; |
3015 | if (pshard != shards.end()) { | |
3016 | shard_end = pshard->shard_info->offset; | |
3017 | } else { | |
3018 | shard_end = OBJECT_MAX_SIZE; | |
3019 | } | |
3020 | } | |
3021 | } | |
11fdf7f2 | 3022 | if (removed) { |
7c673cae FG |
3023 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3024 | } | |
3025 | return removed; | |
3026 | } | |
3027 | ||
3028 | void BlueStore::ExtentMap::punch_hole( | |
3029 | CollectionRef &c, | |
3030 | uint64_t offset, | |
3031 | uint64_t length, | |
3032 | old_extent_map_t *old_extents) | |
3033 | { | |
3034 | auto p = seek_lextent(offset); | |
3035 | uint64_t end = offset + length; | |
3036 | while (p != extent_map.end()) { | |
3037 | if (p->logical_offset >= end) { | |
3038 | break; | |
3039 | } | |
3040 | if (p->logical_offset < offset) { | |
3041 | if (p->logical_end() > end) { | |
3042 | // split and deref middle | |
3043 | uint64_t front = offset - p->logical_offset; | |
3044 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3045 | length, p->blob); | |
3046 | old_extents->push_back(*oe); | |
3047 | add(end, | |
3048 | p->blob_offset + front + length, | |
3049 | p->length - front - length, | |
3050 | p->blob); | |
3051 | p->length = front; | |
3052 | break; | |
3053 | } else { | |
3054 | // deref tail | |
11fdf7f2 | 3055 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3056 | uint64_t keep = offset - p->logical_offset; |
3057 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3058 | p->length - keep, p->blob); | |
3059 | old_extents->push_back(*oe); | |
3060 | p->length = keep; | |
3061 | ++p; | |
3062 | continue; | |
3063 | } | |
3064 | } | |
3065 | if (p->logical_offset + p->length <= end) { | |
3066 | // deref whole lextent | |
3067 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3068 | p->length, p->blob); | |
3069 | old_extents->push_back(*oe); | |
3070 | rm(p++); | |
3071 | continue; | |
3072 | } | |
3073 | // deref head | |
3074 | uint64_t keep = p->logical_end() - end; | |
3075 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3076 | p->length - keep, p->blob); | |
3077 | old_extents->push_back(*oe); | |
3078 | ||
3079 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3080 | rm(p); | |
3081 | break; | |
3082 | } | |
3083 | } | |
3084 | ||
3085 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3086 | CollectionRef &c, | |
3087 | uint64_t logical_offset, | |
3088 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3089 | old_extent_map_t *old_extents) | |
3090 | { | |
3091 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3092 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3093 | |
3094 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3095 | // old_extents list if we overwre the blob totally | |
3096 | // This might happen during WAL overwrite. | |
3097 | b->get_ref(onode->c, blob_offset, length); | |
3098 | ||
3099 | if (old_extents) { | |
3100 | punch_hole(c, logical_offset, length, old_extents); | |
3101 | } | |
3102 | ||
3103 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3104 | extent_map.insert(*le); | |
3105 | if (spans_shard(logical_offset, length)) { | |
3106 | request_reshard(logical_offset, logical_offset + length); | |
3107 | } | |
3108 | return le; | |
3109 | } | |
3110 | ||
3111 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3112 | BlobRef lb, | |
3113 | uint32_t blob_offset, | |
3114 | uint32_t pos) | |
3115 | { | |
3116 | auto cct = onode->c->store->cct; //used by dout | |
3117 | ||
3118 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; | |
3119 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3120 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3121 | << dendl; | |
3122 | BlobRef rb = onode->c->new_blob(); | |
3123 | lb->split(onode->c, blob_offset, rb.get()); | |
3124 | ||
3125 | for (auto ep = seek_lextent(pos); | |
3126 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3127 | ++ep) { | |
3128 | if (ep->blob != lb) { | |
3129 | continue; | |
3130 | } | |
3131 | if (ep->logical_offset < pos) { | |
3132 | // split extent | |
3133 | size_t left = pos - ep->logical_offset; | |
3134 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3135 | extent_map.insert(*ne); | |
3136 | ep->length = left; | |
3137 | dout(30) << __func__ << " split " << *ep << dendl; | |
3138 | dout(30) << __func__ << " to " << *ne << dendl; | |
3139 | } else { | |
3140 | // switch blob | |
11fdf7f2 | 3141 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3142 | |
3143 | ep->blob = rb; | |
3144 | ep->blob_offset -= blob_offset; | |
3145 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3146 | } | |
3147 | } | |
3148 | return rb; | |
3149 | } | |
3150 | ||
3151 | // Onode | |
3152 | ||
3153 | #undef dout_prefix | |
3154 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3155 | ||
3156 | void BlueStore::Onode::flush() | |
3157 | { | |
3158 | if (flushing_count.load()) { | |
3159 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
11fdf7f2 | 3160 | std::unique_lock l(flush_lock); |
7c673cae FG |
3161 | while (flushing_count.load()) { |
3162 | flush_cond.wait(l); | |
3163 | } | |
3164 | } | |
3165 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3166 | } | |
3167 | ||
3168 | // ======================================================= | |
3169 | // WriteContext | |
3170 | ||
3171 | /// Checks for writes to the same pextent within a blob | |
3172 | bool BlueStore::WriteContext::has_conflict( | |
3173 | BlobRef b, | |
3174 | uint64_t loffs, | |
3175 | uint64_t loffs_end, | |
3176 | uint64_t min_alloc_size) | |
3177 | { | |
11fdf7f2 TL |
3178 | ceph_assert((loffs % min_alloc_size) == 0); |
3179 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3180 | for (auto w : writes) { |
3181 | if (b == w.b) { | |
11fdf7f2 TL |
3182 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3183 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3184 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3185 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3186 | return true; |
3187 | } | |
3188 | } | |
3189 | } | |
3190 | return false; | |
3191 | } | |
3192 | ||
3193 | // ======================================================= | |
3194 | ||
3195 | // DeferredBatch | |
3196 | #undef dout_prefix | |
3197 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
3198 | ||
3199 | void BlueStore::DeferredBatch::prepare_write( | |
3200 | CephContext *cct, | |
3201 | uint64_t seq, uint64_t offset, uint64_t length, | |
3202 | bufferlist::const_iterator& blp) | |
3203 | { | |
3204 | _discard(cct, offset, length); | |
3205 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3206 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3207 | i.first->second.seq = seq; |
3208 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3209 | i.first->second.bl.reassign_to_mempool( |
3210 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3211 | dout(20) << __func__ << " seq " << seq |
3212 | << " 0x" << std::hex << offset << "~" << length | |
3213 | << " crc " << i.first->second.bl.crc32c(-1) | |
3214 | << std::dec << dendl; | |
3215 | seq_bytes[seq] += length; | |
3216 | #ifdef DEBUG_DEFERRED | |
3217 | _audit(cct); | |
3218 | #endif | |
3219 | } | |
3220 | ||
3221 | void BlueStore::DeferredBatch::_discard( | |
3222 | CephContext *cct, uint64_t offset, uint64_t length) | |
3223 | { | |
3224 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3225 | << std::dec << dendl; | |
3226 | auto p = iomap.lower_bound(offset); | |
3227 | if (p != iomap.begin()) { | |
3228 | --p; | |
3229 | auto end = p->first + p->second.bl.length(); | |
3230 | if (end > offset) { | |
3231 | bufferlist head; | |
3232 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3233 | dout(20) << __func__ << " keep head " << p->second.seq | |
3234 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3235 | << " -> 0x" << head.length() << std::dec << dendl; | |
3236 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3237 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3238 | if (end > offset + length) { |
3239 | bufferlist tail; | |
3240 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3241 | end - (offset + length)); | |
3242 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3243 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3244 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3245 | auto &n = iomap[offset + length]; | |
3246 | n.bl.swap(tail); | |
3247 | n.seq = p->second.seq; | |
3248 | i->second -= length; | |
3249 | } else { | |
3250 | i->second -= end - offset; | |
3251 | } | |
11fdf7f2 | 3252 | ceph_assert(i->second >= 0); |
7c673cae FG |
3253 | p->second.bl.swap(head); |
3254 | } | |
3255 | ++p; | |
3256 | } | |
3257 | while (p != iomap.end()) { | |
3258 | if (p->first >= offset + length) { | |
3259 | break; | |
3260 | } | |
3261 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3262 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3263 | auto end = p->first + p->second.bl.length(); |
3264 | if (end > offset + length) { | |
3265 | unsigned drop_front = offset + length - p->first; | |
3266 | unsigned keep_tail = end - (offset + length); | |
3267 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3268 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3269 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3270 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3271 | << std::dec << dendl; | |
3272 | auto &s = iomap[offset + length]; | |
3273 | s.seq = p->second.seq; | |
3274 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3275 | i->second -= drop_front; | |
3276 | } else { | |
3277 | dout(20) << __func__ << " drop " << p->second.seq | |
3278 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3279 | << std::dec << dendl; | |
3280 | i->second -= p->second.bl.length(); | |
3281 | } | |
11fdf7f2 | 3282 | ceph_assert(i->second >= 0); |
7c673cae FG |
3283 | p = iomap.erase(p); |
3284 | } | |
3285 | } | |
3286 | ||
3287 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3288 | { | |
3289 | map<uint64_t,int> sb; | |
3290 | for (auto p : seq_bytes) { | |
3291 | sb[p.first] = 0; // make sure we have the same set of keys | |
3292 | } | |
3293 | uint64_t pos = 0; | |
3294 | for (auto& p : iomap) { | |
11fdf7f2 | 3295 | ceph_assert(p.first >= pos); |
7c673cae FG |
3296 | sb[p.second.seq] += p.second.bl.length(); |
3297 | pos = p.first + p.second.bl.length(); | |
3298 | } | |
11fdf7f2 | 3299 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3300 | } |
3301 | ||
3302 | ||
3303 | // Collection | |
3304 | ||
3305 | #undef dout_prefix | |
3306 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3307 | ||
11fdf7f2 TL |
3308 | BlueStore::Collection::Collection(BlueStore *store_, Cache *c, coll_t cid) |
3309 | : CollectionImpl(cid), | |
3310 | store(store_), | |
7c673cae | 3311 | cache(c), |
7c673cae FG |
3312 | lock("BlueStore::Collection::lock", true, false), |
3313 | exists(true), | |
11fdf7f2 TL |
3314 | onode_map(c), |
3315 | commit_queue(nullptr) | |
3316 | { | |
3317 | } | |
3318 | ||
3319 | bool BlueStore::Collection::flush_commit(Context *c) | |
3320 | { | |
3321 | return osr->flush_commit(c); | |
3322 | } | |
3323 | ||
3324 | void BlueStore::Collection::flush() | |
3325 | { | |
3326 | osr->flush(); | |
3327 | } | |
3328 | ||
3329 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3330 | { |
11fdf7f2 | 3331 | osr->flush_all_but_last(); |
7c673cae FG |
3332 | } |
3333 | ||
3334 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3335 | { | |
11fdf7f2 | 3336 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3337 | const bluestore_blob_t& blob = b->get_blob(); |
3338 | if (!blob.is_shared()) { | |
3339 | b->shared_blob = new SharedBlob(this); | |
3340 | return; | |
3341 | } | |
3342 | ||
3343 | b->shared_blob = shared_blob_set.lookup(sbid); | |
3344 | if (b->shared_blob) { | |
3345 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3346 | << std::dec << " had " << *b->shared_blob << dendl; | |
3347 | } else { | |
3348 | b->shared_blob = new SharedBlob(sbid, this); | |
3349 | shared_blob_set.add(this, b->shared_blob.get()); | |
3350 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3351 | << std::dec << " opened " << *b->shared_blob | |
3352 | << dendl; | |
3353 | } | |
3354 | } | |
3355 | ||
3356 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
3357 | { | |
3358 | if (!sb->is_loaded()) { | |
3359 | ||
3360 | bufferlist v; | |
3361 | string key; | |
3362 | auto sbid = sb->get_sbid(); | |
3363 | get_shared_blob_key(sbid, &key); | |
3364 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
3365 | if (r < 0) { | |
3366 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
3367 | << std::dec << " not found at key " | |
3368 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 3369 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
3370 | } |
3371 | ||
3372 | sb->loaded = true; | |
3373 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
3374 | auto p = v.cbegin(); |
3375 | decode(*(sb->persistent), p); | |
7c673cae FG |
3376 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
3377 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
3378 | } | |
3379 | } | |
3380 | ||
3381 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
3382 | { | |
7c673cae | 3383 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 3384 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
3385 | |
3386 | // update blob | |
31f18b77 | 3387 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 3388 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
3389 | |
3390 | // update shared blob | |
3391 | b->shared_blob->loaded = true; | |
3392 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
3393 | shared_blob_set.add(this, b->shared_blob.get()); | |
3394 | for (auto p : blob.get_extents()) { | |
3395 | if (p.is_valid()) { | |
3396 | b->shared_blob->get_ref( | |
3397 | p.offset, | |
3398 | p.length); | |
3399 | } | |
3400 | } | |
3401 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
3402 | } | |
3403 | ||
31f18b77 FG |
3404 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
3405 | { | |
3406 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 3407 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
3408 | |
3409 | uint64_t sbid = sb->get_sbid(); | |
3410 | shared_blob_set.remove(sb); | |
3411 | sb->loaded = false; | |
3412 | delete sb->persistent; | |
3413 | sb->sbid_unloaded = 0; | |
3414 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
3415 | return sbid; | |
3416 | } | |
3417 | ||
7c673cae FG |
3418 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
3419 | const ghobject_t& oid, | |
3420 | bool create) | |
3421 | { | |
11fdf7f2 | 3422 | ceph_assert(create ? lock.is_wlocked() : lock.is_locked()); |
7c673cae FG |
3423 | |
3424 | spg_t pgid; | |
3425 | if (cid.is_pg(&pgid)) { | |
3426 | if (!oid.match(cnode.bits, pgid.ps())) { | |
3427 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
3428 | << pgid << " bits " << cnode.bits << dendl; | |
3429 | ceph_abort(); | |
3430 | } | |
3431 | } | |
3432 | ||
3433 | OnodeRef o = onode_map.lookup(oid); | |
3434 | if (o) | |
3435 | return o; | |
3436 | ||
31f18b77 | 3437 | mempool::bluestore_cache_other::string key; |
7c673cae FG |
3438 | get_object_key(store->cct, oid, &key); |
3439 | ||
3440 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
3441 | << pretty_binary_string(key) << dendl; | |
3442 | ||
3443 | bufferlist v; | |
3444 | int r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
3445 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
3446 | Onode *on; | |
3447 | if (v.length() == 0) { | |
11fdf7f2 | 3448 | ceph_assert(r == -ENOENT); |
7c673cae FG |
3449 | if (!store->cct->_conf->bluestore_debug_misc && |
3450 | !create) | |
3451 | return OnodeRef(); | |
3452 | ||
3453 | // new object, new onode | |
3454 | on = new Onode(this, oid, key); | |
3455 | } else { | |
3456 | // loaded | |
11fdf7f2 | 3457 | ceph_assert(r >= 0); |
7c673cae FG |
3458 | on = new Onode(this, oid, key); |
3459 | on->exists = true; | |
11fdf7f2 | 3460 | auto p = v.front().begin_deep(); |
7c673cae | 3461 | on->onode.decode(p); |
3efd9988 FG |
3462 | for (auto& i : on->onode.attrs) { |
3463 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
3464 | } | |
7c673cae FG |
3465 | |
3466 | // initialize extent_map | |
3467 | on->extent_map.decode_spanning_blobs(p); | |
3468 | if (on->onode.extent_map_shards.empty()) { | |
3469 | denc(on->extent_map.inline_bl, p); | |
3470 | on->extent_map.decode_some(on->extent_map.inline_bl); | |
3efd9988 FG |
3471 | on->extent_map.inline_bl.reassign_to_mempool( |
3472 | mempool::mempool_bluestore_cache_other); | |
7c673cae FG |
3473 | } else { |
3474 | on->extent_map.init_shards(false, false); | |
3475 | } | |
3476 | } | |
3477 | o.reset(on); | |
3478 | return onode_map.add(oid, o); | |
3479 | } | |
3480 | ||
3481 | void BlueStore::Collection::split_cache( | |
3482 | Collection *dest) | |
3483 | { | |
3484 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
3485 | ||
3486 | // lock (one or both) cache shards | |
3487 | std::lock(cache->lock, dest->cache->lock); | |
11fdf7f2 TL |
3488 | std::lock_guard l(cache->lock, std::adopt_lock); |
3489 | std::lock_guard l2(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
3490 | |
3491 | int destbits = dest->cnode.bits; | |
3492 | spg_t destpg; | |
3493 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 3494 | ceph_assert(is_pg); |
7c673cae FG |
3495 | |
3496 | auto p = onode_map.onode_map.begin(); | |
3497 | while (p != onode_map.onode_map.end()) { | |
11fdf7f2 | 3498 | OnodeRef o = p->second; |
7c673cae FG |
3499 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
3500 | // onode does not belong to this child | |
11fdf7f2 TL |
3501 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
3502 | << dendl; | |
7c673cae FG |
3503 | ++p; |
3504 | } else { | |
7c673cae FG |
3505 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
3506 | << dendl; | |
3507 | ||
3508 | cache->_rm_onode(p->second); | |
3509 | p = onode_map.onode_map.erase(p); | |
3510 | ||
3511 | o->c = dest; | |
3512 | dest->cache->_add_onode(o, 1); | |
3513 | dest->onode_map.onode_map[o->oid] = o; | |
3514 | dest->onode_map.cache = dest->cache; | |
3515 | ||
3516 | // move over shared blobs and buffers. cover shared blobs from | |
3517 | // both extent map and spanning blob map (the full extent map | |
3518 | // may not be faulted in) | |
3519 | vector<SharedBlob*> sbvec; | |
3520 | for (auto& e : o->extent_map.extent_map) { | |
3521 | sbvec.push_back(e.blob->shared_blob.get()); | |
3522 | } | |
3523 | for (auto& b : o->extent_map.spanning_blob_map) { | |
3524 | sbvec.push_back(b.second->shared_blob.get()); | |
3525 | } | |
3526 | for (auto sb : sbvec) { | |
3527 | if (sb->coll == dest) { | |
3528 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
3529 | << dendl; | |
3530 | continue; | |
3531 | } | |
3532 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
3533 | if (sb->get_sbid()) { |
3534 | ldout(store->cct, 20) << __func__ | |
3535 | << " moving registration " << *sb << dendl; | |
3536 | shared_blob_set.remove(sb); | |
3537 | dest->shared_blob_set.add(dest, sb); | |
3538 | } | |
3efd9988 | 3539 | sb->coll = dest; |
7c673cae | 3540 | if (dest->cache != cache) { |
7c673cae FG |
3541 | for (auto& i : sb->bc.buffer_map) { |
3542 | if (!i.second->is_writing()) { | |
3543 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
3544 | << dendl; | |
3545 | dest->cache->_move_buffer(cache, i.second.get()); | |
3546 | } | |
3547 | } | |
3548 | } | |
3549 | } | |
7c673cae FG |
3550 | } |
3551 | } | |
3552 | } | |
3553 | ||
7c673cae FG |
3554 | // ======================================================= |
3555 | ||
91327a77 AA |
3556 | // MempoolThread |
3557 | ||
3558 | #undef dout_prefix | |
3559 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
3560 | ||
7c673cae FG |
3561 | void *BlueStore::MempoolThread::entry() |
3562 | { | |
11fdf7f2 TL |
3563 | std::unique_lock l(lock); |
3564 | ||
3565 | std::list<std::shared_ptr<PriorityCache::PriCache>> caches; | |
3566 | binned_kv_cache = store->db->get_priority_cache(); | |
3567 | if (binned_kv_cache != nullptr) { | |
3568 | caches.push_back(binned_kv_cache); | |
3569 | } | |
3570 | caches.push_back(meta_cache); | |
3571 | caches.push_back(data_cache); | |
31f18b77 | 3572 | |
91327a77 AA |
3573 | autotune_cache_size = store->osd_memory_cache_min; |
3574 | ||
3575 | utime_t next_balance = ceph_clock_now(); | |
3576 | utime_t next_resize = ceph_clock_now(); | |
31f18b77 | 3577 | |
91327a77 AA |
3578 | bool interval_stats_trim = false; |
3579 | bool interval_stats_resize = false; | |
3580 | while (!stop) { | |
91327a77 AA |
3581 | // Before we trim, check and see if it's time to rebalance/resize. |
3582 | double autotune_interval = store->cache_autotune_interval; | |
3583 | double resize_interval = store->osd_memory_cache_resize_interval; | |
3584 | ||
3585 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { | |
11fdf7f2 TL |
3586 | _adjust_cache_settings(); |
3587 | ||
91327a77 AA |
3588 | // Log events at 5 instead of 20 when balance happens. |
3589 | interval_stats_resize = true; | |
3590 | interval_stats_trim = true; | |
3591 | if (store->cache_autotune) { | |
3592 | _balance_cache(caches); | |
3593 | } | |
31f18b77 | 3594 | |
91327a77 AA |
3595 | next_balance = ceph_clock_now(); |
3596 | next_balance += autotune_interval; | |
3597 | } | |
3598 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { | |
3599 | if (ceph_using_tcmalloc() && store->cache_autotune) { | |
3600 | _tune_cache_size(interval_stats_resize); | |
3601 | interval_stats_resize = false; | |
3602 | } | |
3603 | next_resize = ceph_clock_now(); | |
3604 | next_resize += resize_interval; | |
31f18b77 FG |
3605 | } |
3606 | ||
91327a77 AA |
3607 | // Now Trim |
3608 | _trim_shards(interval_stats_trim); | |
3609 | interval_stats_trim = false; | |
31f18b77 | 3610 | |
91327a77 | 3611 | store->_update_cache_logger(); |
11fdf7f2 TL |
3612 | auto wait = ceph::make_timespan( |
3613 | store->cct->_conf->bluestore_cache_trim_interval); | |
3614 | cond.wait_for(l, wait); | |
7c673cae FG |
3615 | } |
3616 | stop = false; | |
3617 | return NULL; | |
3618 | } | |
3619 | ||
91327a77 AA |
3620 | void BlueStore::MempoolThread::_adjust_cache_settings() |
3621 | { | |
11fdf7f2 TL |
3622 | if (binned_kv_cache != nullptr) { |
3623 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
3624 | } | |
3625 | meta_cache->set_cache_ratio(store->cache_meta_ratio); | |
3626 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
91327a77 AA |
3627 | } |
3628 | ||
3629 | void BlueStore::MempoolThread::_trim_shards(bool interval_stats) | |
3630 | { | |
3631 | auto cct = store->cct; | |
3632 | size_t num_shards = store->cache_shards.size(); | |
3633 | ||
3634 | int64_t kv_used = store->db->get_cache_usage(); | |
11fdf7f2 TL |
3635 | int64_t meta_used = meta_cache->_get_used_bytes(); |
3636 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
3637 | |
3638 | uint64_t cache_size = store->cache_size; | |
3639 | int64_t kv_alloc = | |
11fdf7f2 | 3640 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
91327a77 | 3641 | int64_t meta_alloc = |
11fdf7f2 | 3642 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 3643 | int64_t data_alloc = |
11fdf7f2 | 3644 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 3645 | |
11fdf7f2 | 3646 | if (binned_kv_cache != nullptr && store->cache_autotune) { |
91327a77 AA |
3647 | cache_size = autotune_cache_size; |
3648 | ||
11fdf7f2 TL |
3649 | kv_alloc = binned_kv_cache->get_committed_size(); |
3650 | meta_alloc = meta_cache->get_committed_size(); | |
3651 | data_alloc = data_cache->get_committed_size(); | |
91327a77 AA |
3652 | } |
3653 | ||
3654 | if (interval_stats) { | |
3655 | ldout(cct, 5) << __func__ << " cache_size: " << cache_size | |
3656 | << " kv_alloc: " << kv_alloc | |
3657 | << " kv_used: " << kv_used | |
3658 | << " meta_alloc: " << meta_alloc | |
3659 | << " meta_used: " << meta_used | |
3660 | << " data_alloc: " << data_alloc | |
3661 | << " data_used: " << data_used << dendl; | |
3662 | } else { | |
3663 | ldout(cct, 20) << __func__ << " cache_size: " << cache_size | |
3664 | << " kv_alloc: " << kv_alloc | |
3665 | << " kv_used: " << kv_used | |
3666 | << " meta_alloc: " << meta_alloc | |
3667 | << " meta_used: " << meta_used | |
3668 | << " data_alloc: " << data_alloc | |
3669 | << " data_used: " << data_used << dendl; | |
3670 | } | |
3671 | ||
3672 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
11fdf7f2 | 3673 | (meta_alloc / (double) num_shards) / meta_cache->get_bytes_per_onode()); |
91327a77 AA |
3674 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / num_shards); |
3675 | ||
3676 | ldout(cct, 30) << __func__ << " max_shard_onodes: " << max_shard_onodes | |
3677 | << " max_shard_buffer: " << max_shard_buffer << dendl; | |
3678 | ||
3679 | for (auto i : store->cache_shards) { | |
3680 | i->trim(max_shard_onodes, max_shard_buffer); | |
3681 | } | |
3682 | } | |
3683 | ||
3684 | void BlueStore::MempoolThread::_tune_cache_size(bool interval_stats) | |
3685 | { | |
3686 | auto cct = store->cct; | |
3687 | uint64_t target = store->osd_memory_target; | |
3688 | uint64_t base = store->osd_memory_base; | |
3689 | double fragmentation = store->osd_memory_expected_fragmentation; | |
91327a77 | 3690 | uint64_t cache_min = store->osd_memory_cache_min; |
f64942e4 AA |
3691 | uint64_t cache_max = cache_min; |
3692 | uint64_t limited_target = (1.0 - fragmentation) * target; | |
3693 | if (limited_target > base + cache_min) { | |
3694 | cache_max = limited_target - base; | |
3695 | } | |
91327a77 AA |
3696 | |
3697 | size_t heap_size = 0; | |
3698 | size_t unmapped = 0; | |
3699 | uint64_t mapped = 0; | |
3700 | ||
3701 | ceph_heap_release_free_memory(); | |
3702 | ceph_heap_get_numeric_property("generic.heap_size", &heap_size); | |
3703 | ceph_heap_get_numeric_property("tcmalloc.pageheap_unmapped_bytes", &unmapped); | |
3704 | mapped = heap_size - unmapped; | |
3705 | ||
3706 | uint64_t new_size = autotune_cache_size; | |
3707 | new_size = (new_size < cache_max) ? new_size : cache_max; | |
3708 | new_size = (new_size > cache_min) ? new_size : cache_min; | |
3709 | ||
3710 | // Approach the min/max slowly, but bounce away quickly. | |
3711 | if ((uint64_t) mapped < target) { | |
3712 | double ratio = 1 - ((double) mapped / target); | |
3713 | new_size += ratio * (cache_max - new_size); | |
3714 | } else { | |
3715 | double ratio = 1 - ((double) target / mapped); | |
3716 | new_size -= ratio * (new_size - cache_min); | |
3717 | } | |
3718 | ||
3719 | if (interval_stats) { | |
3720 | ldout(cct, 5) << __func__ | |
3721 | << " target: " << target | |
3722 | << " heap: " << heap_size | |
3723 | << " unmapped: " << unmapped | |
3724 | << " mapped: " << mapped | |
3725 | << " old cache_size: " << autotune_cache_size | |
3726 | << " new cache size: " << new_size << dendl; | |
3727 | } else { | |
3728 | ldout(cct, 20) << __func__ | |
3729 | << " target: " << target | |
3730 | << " heap: " << heap_size | |
3731 | << " unmapped: " << unmapped | |
3732 | << " mapped: " << mapped | |
3733 | << " old cache_size: " << autotune_cache_size | |
3734 | << " new cache size: " << new_size << dendl; | |
3735 | } | |
3736 | autotune_cache_size = new_size; | |
3737 | } | |
3738 | ||
3739 | void BlueStore::MempoolThread::_balance_cache( | |
11fdf7f2 | 3740 | const std::list<std::shared_ptr<PriorityCache::PriCache>>& caches) |
91327a77 AA |
3741 | { |
3742 | int64_t mem_avail = autotune_cache_size; | |
11fdf7f2 TL |
3743 | /* Each cache is going to get at least 1 chunk's worth of memory from get_chunk |
3744 | * so shrink the available memory here to compensate. Don't shrink the amount of | |
3745 | * memory below 0 however. | |
3746 | */ | |
3747 | mem_avail -= PriorityCache::get_chunk(1, autotune_cache_size) * caches.size(); | |
3748 | if (mem_avail < 0) { | |
3749 | mem_avail = 0; | |
3750 | } | |
91327a77 AA |
3751 | |
3752 | // Assign memory for each priority level | |
3753 | for (int i = 0; i < PriorityCache::Priority::LAST + 1; i++) { | |
3754 | ldout(store->cct, 10) << __func__ << " assigning cache bytes for PRI: " << i << dendl; | |
3755 | PriorityCache::Priority pri = static_cast<PriorityCache::Priority>(i); | |
3756 | _balance_cache_pri(&mem_avail, caches, pri); | |
3757 | } | |
3758 | // Assign any leftover memory based on the default ratios. | |
3759 | if (mem_avail > 0) { | |
3760 | for (auto it = caches.begin(); it != caches.end(); it++) { | |
3761 | int64_t fair_share = | |
3762 | static_cast<int64_t>((*it)->get_cache_ratio() * mem_avail); | |
3763 | if (fair_share > 0) { | |
3764 | (*it)->add_cache_bytes(PriorityCache::Priority::LAST, fair_share); | |
3765 | } | |
3766 | } | |
3767 | } | |
3768 | // assert if we assigned more memory than is available. | |
11fdf7f2 | 3769 | ceph_assert(mem_avail >= 0); |
91327a77 AA |
3770 | |
3771 | // Finally commit the new cache sizes | |
3772 | for (auto it = caches.begin(); it != caches.end(); it++) { | |
11fdf7f2 | 3773 | (*it)->commit_cache_size(autotune_cache_size); |
91327a77 AA |
3774 | } |
3775 | } | |
3776 | ||
3777 | void BlueStore::MempoolThread::_balance_cache_pri(int64_t *mem_avail, | |
11fdf7f2 TL |
3778 | const std::list<std::shared_ptr<PriorityCache::PriCache>>& caches, |
3779 | PriorityCache::Priority pri) | |
91327a77 | 3780 | { |
11fdf7f2 | 3781 | std::list<std::shared_ptr<PriorityCache::PriCache>> tmp_caches = caches; |
91327a77 AA |
3782 | double cur_ratios = 0; |
3783 | double new_ratios = 0; | |
3784 | ||
3785 | // Zero this priority's bytes, sum the initial ratios. | |
3786 | for (auto it = tmp_caches.begin(); it != tmp_caches.end(); it++) { | |
3787 | (*it)->set_cache_bytes(pri, 0); | |
3788 | cur_ratios += (*it)->get_cache_ratio(); | |
3789 | } | |
3790 | ||
3791 | // For this priority, loop until caches are satisified or we run out of memory. | |
3792 | // Since we can't allocate fractional bytes, stop if we have fewer bytes left | |
3793 | // than the number of participating caches. | |
3794 | while (!tmp_caches.empty() && *mem_avail > static_cast<int64_t>(tmp_caches.size())) { | |
3795 | uint64_t total_assigned = 0; | |
3796 | ||
3797 | for (auto it = tmp_caches.begin(); it != tmp_caches.end(); ) { | |
11fdf7f2 | 3798 | int64_t cache_wants = (*it)->request_cache_bytes(pri, autotune_cache_size); |
91327a77 AA |
3799 | |
3800 | // Usually the ratio should be set to the fraction of the current caches' | |
3801 | // assigned ratio compared to the total ratio of all caches that still | |
3802 | // want memory. There is a special case where the only caches left are | |
3803 | // all assigned 0% ratios but still want memory. In that case, give | |
3804 | // them an equal shot at the remaining memory for this priority. | |
3805 | double ratio = 1.0 / tmp_caches.size(); | |
3806 | if (cur_ratios > 0) { | |
3807 | ratio = (*it)->get_cache_ratio() / cur_ratios; | |
3808 | } | |
3809 | int64_t fair_share = static_cast<int64_t>(*mem_avail * ratio); | |
3810 | ||
3811 | if (cache_wants > fair_share) { | |
3812 | // If we want too much, take what we can get but stick around for more | |
3813 | (*it)->add_cache_bytes(pri, fair_share); | |
3814 | total_assigned += fair_share; | |
3815 | ||
3816 | new_ratios += (*it)->get_cache_ratio(); | |
3817 | ldout(store->cct, 20) << __func__ << " " << (*it)->get_cache_name() | |
3818 | << " wanted: " << cache_wants << " fair_share: " << fair_share | |
3819 | << " mem_avail: " << *mem_avail | |
3820 | << " staying in list. Size: " << tmp_caches.size() | |
3821 | << dendl; | |
3822 | ++it; | |
3823 | } else { | |
3824 | // Otherwise assign only what we want | |
3825 | if (cache_wants > 0) { | |
3826 | (*it)->add_cache_bytes(pri, cache_wants); | |
3827 | total_assigned += cache_wants; | |
3828 | ||
3829 | ldout(store->cct, 20) << __func__ << " " << (*it)->get_cache_name() | |
3830 | << " wanted: " << cache_wants << " fair_share: " << fair_share | |
3831 | << " mem_avail: " << *mem_avail | |
3832 | << " removing from list. New size: " << tmp_caches.size() - 1 | |
3833 | << dendl; | |
3834 | ||
3835 | } | |
3836 | // Either the cache didn't want anything or got what it wanted, so remove it from the tmp list. | |
3837 | it = tmp_caches.erase(it); | |
3838 | } | |
3839 | } | |
3840 | // Reset the ratios | |
3841 | *mem_avail -= total_assigned; | |
3842 | cur_ratios = new_ratios; | |
3843 | new_ratios = 0; | |
3844 | } | |
3845 | } | |
3846 | ||
7c673cae FG |
3847 | // ======================================================= |
3848 | ||
31f18b77 FG |
3849 | // OmapIteratorImpl |
3850 | ||
3851 | #undef dout_prefix | |
3852 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
3853 | ||
3854 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
3855 | CollectionRef c, OnodeRef o, KeyValueDB::Iterator it) | |
3856 | : c(c), o(o), it(it) | |
3857 | { | |
3858 | RWLock::RLocker l(c->lock); | |
3859 | if (o->onode.has_omap()) { | |
3860 | get_omap_key(o->onode.nid, string(), &head); | |
3861 | get_omap_tail(o->onode.nid, &tail); | |
3862 | it->lower_bound(head); | |
3863 | } | |
3864 | } | |
3865 | ||
11fdf7f2 TL |
3866 | string BlueStore::OmapIteratorImpl::_stringify() const |
3867 | { | |
3868 | stringstream s; | |
3869 | s << " omap_iterator(cid = " << c->cid | |
3870 | <<", oid = " << o->oid << ")"; | |
3871 | return s.str(); | |
3872 | } | |
3873 | ||
31f18b77 FG |
3874 | int BlueStore::OmapIteratorImpl::seek_to_first() |
3875 | { | |
3876 | RWLock::RLocker l(c->lock); | |
11fdf7f2 | 3877 | auto start1 = mono_clock::now(); |
31f18b77 FG |
3878 | if (o->onode.has_omap()) { |
3879 | it->lower_bound(head); | |
3880 | } else { | |
3881 | it = KeyValueDB::Iterator(); | |
3882 | } | |
11fdf7f2 TL |
3883 | c->store->log_latency_fn( |
3884 | l_bluestore_omap_seek_to_first_lat, | |
3885 | mono_clock::now() - start1, | |
3886 | [&] (const ceph::timespan& lat) { | |
3887 | return ", lat = " + timespan_str(lat) + _stringify(); | |
3888 | } | |
3889 | ); | |
3890 | ||
31f18b77 FG |
3891 | return 0; |
3892 | } | |
3893 | ||
3894 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
3895 | { | |
3896 | RWLock::RLocker l(c->lock); | |
11fdf7f2 | 3897 | auto start1 = mono_clock::now(); |
31f18b77 FG |
3898 | if (o->onode.has_omap()) { |
3899 | string key; | |
3900 | get_omap_key(o->onode.nid, after, &key); | |
3901 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " | |
3902 | << pretty_binary_string(key) << dendl; | |
3903 | it->upper_bound(key); | |
3904 | } else { | |
3905 | it = KeyValueDB::Iterator(); | |
3906 | } | |
11fdf7f2 TL |
3907 | c->store->log_latency_fn( |
3908 | l_bluestore_omap_upper_bound_lat, | |
3909 | mono_clock::now() - start1, | |
3910 | [&] (const ceph::timespan& lat) { | |
3911 | return ", after = " + after + ", lat = " + timespan_str(lat) + | |
3912 | _stringify(); | |
3913 | } | |
3914 | ); | |
31f18b77 FG |
3915 | return 0; |
3916 | } | |
3917 | ||
3918 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
3919 | { | |
3920 | RWLock::RLocker l(c->lock); | |
11fdf7f2 | 3921 | auto start1 = mono_clock::now(); |
31f18b77 FG |
3922 | if (o->onode.has_omap()) { |
3923 | string key; | |
3924 | get_omap_key(o->onode.nid, to, &key); | |
3925 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " | |
3926 | << pretty_binary_string(key) << dendl; | |
3927 | it->lower_bound(key); | |
3928 | } else { | |
3929 | it = KeyValueDB::Iterator(); | |
3930 | } | |
11fdf7f2 TL |
3931 | c->store->log_latency_fn( |
3932 | l_bluestore_omap_lower_bound_lat, | |
3933 | mono_clock::now() - start1, | |
3934 | [&] (const ceph::timespan& lat) { | |
3935 | return ", to = " + to + ", lat = " + timespan_str(lat) + | |
3936 | _stringify(); | |
3937 | } | |
3938 | ); | |
31f18b77 FG |
3939 | return 0; |
3940 | } | |
3941 | ||
3942 | bool BlueStore::OmapIteratorImpl::valid() | |
3943 | { | |
3944 | RWLock::RLocker l(c->lock); | |
3945 | bool r = o->onode.has_omap() && it && it->valid() && | |
3946 | it->raw_key().second <= tail; | |
3947 | if (it && it->valid()) { | |
3948 | ldout(c->store->cct,20) << __func__ << " is at " | |
3949 | << pretty_binary_string(it->raw_key().second) | |
3950 | << dendl; | |
3951 | } | |
3952 | return r; | |
3953 | } | |
3954 | ||
11fdf7f2 | 3955 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 3956 | { |
11fdf7f2 | 3957 | int r = -1; |
31f18b77 | 3958 | RWLock::RLocker l(c->lock); |
11fdf7f2 | 3959 | auto start1 = mono_clock::now(); |
31f18b77 FG |
3960 | if (o->onode.has_omap()) { |
3961 | it->next(); | |
11fdf7f2 | 3962 | r = 0; |
31f18b77 | 3963 | } |
11fdf7f2 TL |
3964 | c->store->log_latency_fn( |
3965 | l_bluestore_omap_next_lat, | |
3966 | mono_clock::now() - start1, | |
3967 | [&] (const ceph::timespan& lat) { | |
3968 | return ", lat = " + timespan_str(lat) + _stringify(); | |
3969 | } | |
3970 | ); | |
3971 | ||
3972 | return r; | |
31f18b77 FG |
3973 | } |
3974 | ||
3975 | string BlueStore::OmapIteratorImpl::key() | |
3976 | { | |
3977 | RWLock::RLocker l(c->lock); | |
11fdf7f2 | 3978 | ceph_assert(it->valid()); |
31f18b77 FG |
3979 | string db_key = it->raw_key().second; |
3980 | string user_key; | |
3981 | decode_omap_key(db_key, &user_key); | |
3982 | return user_key; | |
3983 | } | |
3984 | ||
3985 | bufferlist BlueStore::OmapIteratorImpl::value() | |
3986 | { | |
3987 | RWLock::RLocker l(c->lock); | |
11fdf7f2 | 3988 | ceph_assert(it->valid()); |
31f18b77 FG |
3989 | return it->value(); |
3990 | } | |
3991 | ||
3992 | ||
3993 | // ===================================== | |
3994 | ||
7c673cae FG |
3995 | #undef dout_prefix |
3996 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
3997 | ||
3998 | ||
3999 | static void aio_cb(void *priv, void *priv2) | |
4000 | { | |
4001 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4002 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4003 | c->aio_finish(store); | |
4004 | } | |
4005 | ||
11fdf7f2 TL |
4006 | static void discard_cb(void *priv, void *priv2) |
4007 | { | |
4008 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4009 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4010 | store->handle_discard(*tmp); | |
4011 | } | |
4012 | ||
4013 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4014 | { | |
4015 | dout(10) << __func__ << dendl; | |
4016 | ceph_assert(alloc); | |
4017 | alloc->release(to_release); | |
4018 | } | |
4019 | ||
7c673cae FG |
4020 | BlueStore::BlueStore(CephContext *cct, const string& path) |
4021 | : ObjectStore(cct, path), | |
4022 | throttle_bytes(cct, "bluestore_throttle_bytes", | |
4023 | cct->_conf->bluestore_throttle_bytes), | |
4024 | throttle_deferred_bytes(cct, "bluestore_throttle_deferred_bytes", | |
4025 | cct->_conf->bluestore_throttle_bytes + | |
4026 | cct->_conf->bluestore_throttle_deferred_bytes), | |
181888fb | 4027 | deferred_finisher(cct, "defered_finisher", "dfin"), |
11fdf7f2 | 4028 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4029 | kv_sync_thread(this), |
31f18b77 | 4030 | kv_finalize_thread(this), |
7c673cae FG |
4031 | mempool_thread(this) |
4032 | { | |
4033 | _init_logger(); | |
11fdf7f2 | 4034 | cct->_conf.add_observer(this); |
7c673cae | 4035 | set_cache_shards(1); |
7c673cae FG |
4036 | } |
4037 | ||
4038 | BlueStore::BlueStore(CephContext *cct, | |
4039 | const string& path, | |
4040 | uint64_t _min_alloc_size) | |
4041 | : ObjectStore(cct, path), | |
4042 | throttle_bytes(cct, "bluestore_throttle_bytes", | |
4043 | cct->_conf->bluestore_throttle_bytes), | |
4044 | throttle_deferred_bytes(cct, "bluestore_throttle_deferred_bytes", | |
4045 | cct->_conf->bluestore_throttle_bytes + | |
4046 | cct->_conf->bluestore_throttle_deferred_bytes), | |
181888fb | 4047 | deferred_finisher(cct, "defered_finisher", "dfin"), |
11fdf7f2 | 4048 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4049 | kv_sync_thread(this), |
31f18b77 | 4050 | kv_finalize_thread(this), |
7c673cae FG |
4051 | min_alloc_size(_min_alloc_size), |
4052 | min_alloc_size_order(ctz(_min_alloc_size)), | |
4053 | mempool_thread(this) | |
4054 | { | |
4055 | _init_logger(); | |
11fdf7f2 | 4056 | cct->_conf.add_observer(this); |
7c673cae | 4057 | set_cache_shards(1); |
7c673cae FG |
4058 | } |
4059 | ||
4060 | BlueStore::~BlueStore() | |
4061 | { | |
11fdf7f2 | 4062 | cct->_conf.remove_observer(this); |
7c673cae | 4063 | _shutdown_logger(); |
11fdf7f2 TL |
4064 | ceph_assert(!mounted); |
4065 | ceph_assert(db == NULL); | |
4066 | ceph_assert(bluefs == NULL); | |
4067 | ceph_assert(fsid_fd < 0); | |
4068 | ceph_assert(path_fd < 0); | |
7c673cae FG |
4069 | for (auto i : cache_shards) { |
4070 | delete i; | |
4071 | } | |
4072 | cache_shards.clear(); | |
4073 | } | |
4074 | ||
4075 | const char **BlueStore::get_tracked_conf_keys() const | |
4076 | { | |
4077 | static const char* KEYS[] = { | |
4078 | "bluestore_csum_type", | |
4079 | "bluestore_compression_mode", | |
4080 | "bluestore_compression_algorithm", | |
4081 | "bluestore_compression_min_blob_size", | |
4082 | "bluestore_compression_min_blob_size_ssd", | |
4083 | "bluestore_compression_min_blob_size_hdd", | |
4084 | "bluestore_compression_max_blob_size", | |
4085 | "bluestore_compression_max_blob_size_ssd", | |
4086 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4087 | "bluestore_compression_required_ratio", |
7c673cae FG |
4088 | "bluestore_max_alloc_size", |
4089 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4090 | "bluestore_prefer_deferred_size_hdd", |
4091 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4092 | "bluestore_deferred_batch_ops", |
4093 | "bluestore_deferred_batch_ops_hdd", | |
4094 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4095 | "bluestore_throttle_bytes", |
4096 | "bluestore_throttle_deferred_bytes", | |
4097 | "bluestore_throttle_cost_per_io_hdd", | |
4098 | "bluestore_throttle_cost_per_io_ssd", | |
4099 | "bluestore_throttle_cost_per_io", | |
4100 | "bluestore_max_blob_size", | |
4101 | "bluestore_max_blob_size_ssd", | |
4102 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4103 | "osd_memory_target", |
4104 | "osd_memory_target_cgroup_limit_ratio", | |
4105 | "osd_memory_base", | |
4106 | "osd_memory_cache_min", | |
4107 | "bluestore_cache_autotune", | |
4108 | "bluestore_cache_autotune_interval", | |
81eedcae TL |
4109 | "bluestore_no_per_pool_stats_tolerance", |
4110 | "bluestore_warn_on_legacy_statfs", | |
7c673cae FG |
4111 | NULL |
4112 | }; | |
4113 | return KEYS; | |
4114 | } | |
4115 | ||
11fdf7f2 | 4116 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4117 | const std::set<std::string> &changed) |
4118 | { | |
81eedcae TL |
4119 | if (changed.count("bluestore_no_per_pool_stats_tolerance") || |
4120 | changed.count("bluestore_warn_on_legacy_statfs")) { | |
4121 | _check_legacy_statfs_alert(); | |
4122 | } | |
4123 | ||
7c673cae FG |
4124 | if (changed.count("bluestore_csum_type")) { |
4125 | _set_csum(); | |
4126 | } | |
4127 | if (changed.count("bluestore_compression_mode") || | |
4128 | changed.count("bluestore_compression_algorithm") || | |
4129 | changed.count("bluestore_compression_min_blob_size") || | |
4130 | changed.count("bluestore_compression_max_blob_size")) { | |
4131 | if (bdev) { | |
4132 | _set_compression(); | |
4133 | } | |
4134 | } | |
4135 | if (changed.count("bluestore_max_blob_size") || | |
4136 | changed.count("bluestore_max_blob_size_ssd") || | |
4137 | changed.count("bluestore_max_blob_size_hdd")) { | |
4138 | if (bdev) { | |
4139 | // only after startup | |
4140 | _set_blob_size(); | |
4141 | } | |
4142 | } | |
4143 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4144 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4145 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4146 | changed.count("bluestore_max_alloc_size") || |
4147 | changed.count("bluestore_deferred_batch_ops") || | |
4148 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4149 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4150 | if (bdev) { | |
4151 | // only after startup | |
4152 | _set_alloc_sizes(); | |
4153 | } | |
4154 | } | |
4155 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4156 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4157 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4158 | if (bdev) { | |
4159 | _set_throttle_params(); | |
4160 | } | |
4161 | } | |
4162 | if (changed.count("bluestore_throttle_bytes")) { | |
4163 | throttle_bytes.reset_max(conf->bluestore_throttle_bytes); | |
4164 | throttle_deferred_bytes.reset_max( | |
4165 | conf->bluestore_throttle_bytes + conf->bluestore_throttle_deferred_bytes); | |
4166 | } | |
4167 | if (changed.count("bluestore_throttle_deferred_bytes")) { | |
4168 | throttle_deferred_bytes.reset_max( | |
4169 | conf->bluestore_throttle_bytes + conf->bluestore_throttle_deferred_bytes); | |
4170 | } | |
4171 | } | |
4172 | ||
4173 | void BlueStore::_set_compression() | |
4174 | { | |
224ce89b WB |
4175 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4176 | if (m) { | |
11fdf7f2 | 4177 | _clear_compression_alert(); |
224ce89b WB |
4178 | comp_mode = *m; |
4179 | } else { | |
4180 | derr << __func__ << " unrecognized value '" | |
4181 | << cct->_conf->bluestore_compression_mode | |
4182 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4183 | << dendl; | |
4184 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4185 | string s("unknown mode: "); |
4186 | s += cct->_conf->bluestore_compression_mode; | |
4187 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4188 | } |
4189 | ||
4190 | compressor = nullptr; | |
4191 | ||
4192 | if (comp_mode == Compressor::COMP_NONE) { | |
4193 | dout(10) << __func__ << " compression mode set to 'none', " | |
11fdf7f2 | 4194 | << "ignore other compression settings" << dendl; |
224ce89b WB |
4195 | return; |
4196 | } | |
4197 | ||
3efd9988 FG |
4198 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4199 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4200 | } else { |
11fdf7f2 | 4201 | ceph_assert(bdev); |
7c673cae FG |
4202 | if (bdev->is_rotational()) { |
4203 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; | |
4204 | } else { | |
4205 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4206 | } | |
4207 | } | |
4208 | ||
4209 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4210 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4211 | } else { | |
11fdf7f2 | 4212 | ceph_assert(bdev); |
7c673cae FG |
4213 | if (bdev->is_rotational()) { |
4214 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; | |
4215 | } else { | |
4216 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4217 | } | |
4218 | } | |
4219 | ||
7c673cae FG |
4220 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4221 | if (!alg_name.empty()) { | |
4222 | compressor = Compressor::create(cct, alg_name); | |
4223 | if (!compressor) { | |
4224 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4225 | << dendl; | |
11fdf7f2 | 4226 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4227 | } |
4228 | } | |
4229 | ||
4230 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4231 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4232 | << " min_blob " << comp_min_blob_size |
4233 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4234 | << dendl; |
4235 | } | |
4236 | ||
4237 | void BlueStore::_set_csum() | |
4238 | { | |
4239 | csum_type = Checksummer::CSUM_NONE; | |
4240 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4241 | if (t > Checksummer::CSUM_NONE) | |
4242 | csum_type = t; | |
4243 | ||
4244 | dout(10) << __func__ << " csum_type " | |
4245 | << Checksummer::get_csum_type_string(csum_type) | |
4246 | << dendl; | |
4247 | } | |
4248 | ||
4249 | void BlueStore::_set_throttle_params() | |
4250 | { | |
4251 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4252 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4253 | } else { | |
11fdf7f2 | 4254 | ceph_assert(bdev); |
7c673cae FG |
4255 | if (bdev->is_rotational()) { |
4256 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; | |
4257 | } else { | |
4258 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4259 | } | |
4260 | } | |
4261 | ||
4262 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4263 | << dendl; | |
4264 | } | |
4265 | void BlueStore::_set_blob_size() | |
4266 | { | |
4267 | if (cct->_conf->bluestore_max_blob_size) { | |
4268 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4269 | } else { | |
11fdf7f2 | 4270 | ceph_assert(bdev); |
7c673cae FG |
4271 | if (bdev->is_rotational()) { |
4272 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; | |
4273 | } else { | |
4274 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4275 | } | |
4276 | } | |
4277 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4278 | << std::dec << dendl; | |
4279 | } | |
4280 | ||
11fdf7f2 | 4281 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4282 | { |
11fdf7f2 TL |
4283 | // set osd_memory_target *default* based on cgroup limit? |
4284 | // (do this before we fetch the osd_memory_target value!) | |
4285 | double cgroup_ratio = cct->_conf.get_val<double>( | |
4286 | "osd_memory_target_cgroup_limit_ratio"); | |
4287 | if (cgroup_ratio > 0.0) { | |
4288 | uint64_t cgroup_limit = 0; | |
4289 | if (get_cgroup_memory_limit(&cgroup_limit) == 0 && | |
4290 | cgroup_limit) { | |
4291 | uint64_t def = cgroup_limit * cgroup_ratio; | |
4292 | dout(10) << __func__ << " osd_memory_target_cgroup_limit_ratio " | |
4293 | << cgroup_ratio << ", cgroup_limit " << cgroup_limit | |
4294 | << ", defaulting osd_memory_target to " << def | |
4295 | << dendl; | |
4296 | cct->_conf.set_val_default("osd_memory_target", stringify(def)); | |
1adf2230 AA |
4297 | } |
4298 | } | |
1adf2230 | 4299 | |
11fdf7f2 TL |
4300 | ceph_assert(bdev); |
4301 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4302 | cache_autotune_interval = |
11fdf7f2 TL |
4303 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
4304 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4305 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4306 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4307 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4308 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4309 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4310 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4311 | |
224ce89b WB |
4312 | if (cct->_conf->bluestore_cache_size) { |
4313 | cache_size = cct->_conf->bluestore_cache_size; | |
4314 | } else { | |
4315 | // choose global cache size based on backend type | |
4316 | if (bdev->is_rotational()) { | |
4317 | cache_size = cct->_conf->bluestore_cache_size_hdd; | |
4318 | } else { | |
4319 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4320 | } | |
4321 | } | |
31f18b77 | 4322 | |
91327a77 | 4323 | cache_meta_ratio = cct->_conf->bluestore_cache_meta_ratio; |
224ce89b | 4324 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4325 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4326 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4327 | return -EINVAL; |
4328 | } | |
91327a77 AA |
4329 | |
4330 | cache_kv_ratio = cct->_conf->bluestore_cache_kv_ratio; | |
224ce89b | 4331 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4332 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4333 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4334 | return -EINVAL; |
4335 | } | |
91327a77 | 4336 | |
31f18b77 | 4337 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4338 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4339 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4340 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4341 | << dendl; | |
31f18b77 FG |
4342 | return -EINVAL; |
4343 | } | |
91327a77 AA |
4344 | |
4345 | cache_data_ratio = | |
4346 | (double)1.0 - (double)cache_meta_ratio - (double)cache_kv_ratio; | |
31f18b77 FG |
4347 | if (cache_data_ratio < 0) { |
4348 | // deal with floating point imprecision | |
4349 | cache_data_ratio = 0; | |
4350 | } | |
91327a77 | 4351 | |
224ce89b WB |
4352 | dout(1) << __func__ << " cache_size " << cache_size |
4353 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4354 | << " kv " << cache_kv_ratio |
4355 | << " data " << cache_data_ratio | |
4356 | << dendl; | |
4357 | return 0; | |
4358 | } | |
4359 | ||
3efd9988 FG |
4360 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4361 | { | |
4362 | bluestore_bdev_label_t label; | |
4363 | string p = path + "/block"; | |
4364 | int r = _read_bdev_label(cct, p, &label); | |
4365 | if (r < 0) { | |
4366 | return ObjectStore::write_meta(key, value); | |
4367 | } | |
4368 | label.meta[key] = value; | |
4369 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4370 | ceph_assert(r == 0); |
3efd9988 FG |
4371 | return ObjectStore::write_meta(key, value); |
4372 | } | |
4373 | ||
4374 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4375 | { | |
4376 | bluestore_bdev_label_t label; | |
4377 | string p = path + "/block"; | |
4378 | int r = _read_bdev_label(cct, p, &label); | |
4379 | if (r < 0) { | |
4380 | return ObjectStore::read_meta(key, value); | |
4381 | } | |
4382 | auto i = label.meta.find(key); | |
4383 | if (i == label.meta.end()) { | |
4384 | return ObjectStore::read_meta(key, value); | |
4385 | } | |
4386 | *value = i->second; | |
4387 | return 0; | |
4388 | } | |
4389 | ||
7c673cae FG |
4390 | void BlueStore::_init_logger() |
4391 | { | |
4392 | PerfCountersBuilder b(cct, "bluestore", | |
4393 | l_bluestore_first, l_bluestore_last); | |
4394 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
4395 | "Average kv_thread flush latency", | |
4396 | "fl_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4397 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
4398 | "Average kv_thread commit latency"); | |
11fdf7f2 TL |
4399 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", |
4400 | "Average kv_sync thread latency", | |
4401 | "ks_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4402 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
4403 | "Average kv_finalize thread latency", | |
4404 | "kf_l", PerfCountersBuilder::PRIO_INTERESTING); | |
7c673cae FG |
4405 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
4406 | "Average prepare state latency"); | |
4407 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", | |
4408 | "Average aio_wait state latency", | |
4409 | "io_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4410 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", | |
4411 | "Average io_done state latency"); | |
4412 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", | |
4413 | "Average kv_queued state latency"); | |
4414 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", | |
4415 | "Average kv_commiting state latency"); | |
4416 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", | |
4417 | "Average kv_done state latency"); | |
4418 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", | |
4419 | "Average deferred_queued state latency"); | |
4420 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", | |
4421 | "Average aio_wait state latency"); | |
4422 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", | |
4423 | "Average cleanup state latency"); | |
4424 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
4425 | "Average finishing state latency"); | |
4426 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
4427 | "Average done state latency"); | |
4428 | b.add_time_avg(l_bluestore_throttle_lat, "throttle_lat", | |
4429 | "Average submit throttle latency", | |
4430 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4431 | b.add_time_avg(l_bluestore_submit_lat, "submit_lat", | |
4432 | "Average submit latency", | |
4433 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4434 | b.add_time_avg(l_bluestore_commit_lat, "commit_lat", | |
4435 | "Average commit latency", | |
4436 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4437 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
4438 | "Average read latency", | |
4439 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4440 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", | |
4441 | "Average read onode metadata latency"); | |
4442 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", | |
4443 | "Average read latency"); | |
4444 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
4445 | "Average compress latency"); | |
4446 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
4447 | "Average decompress latency"); | |
4448 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", | |
4449 | "Average checksum latency"); | |
4450 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
4451 | "Sum for beneficial compress ops"); | |
4452 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
4453 | "Sum for compress ops rejected due to low net gain of space"); | |
4454 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", | |
11fdf7f2 | 4455 | "Sum for write-op padded bytes", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4456 | b.add_u64_counter(l_bluestore_deferred_write_ops, "deferred_write_ops", |
4457 | "Sum for deferred write op"); | |
4458 | b.add_u64_counter(l_bluestore_deferred_write_bytes, "deferred_write_bytes", | |
11fdf7f2 | 4459 | "Sum for deferred write bytes", "def", 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4460 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", |
4461 | "Sum for write penalty read ops"); | |
4462 | b.add_u64(l_bluestore_allocated, "bluestore_allocated", | |
4463 | "Sum for allocated bytes"); | |
4464 | b.add_u64(l_bluestore_stored, "bluestore_stored", | |
4465 | "Sum for stored bytes"); | |
4466 | b.add_u64(l_bluestore_compressed, "bluestore_compressed", | |
4467 | "Sum for stored compressed bytes"); | |
4468 | b.add_u64(l_bluestore_compressed_allocated, "bluestore_compressed_allocated", | |
4469 | "Sum for bytes allocated for compressed data"); | |
4470 | b.add_u64(l_bluestore_compressed_original, "bluestore_compressed_original", | |
4471 | "Sum for original bytes that were compressed"); | |
4472 | ||
4473 | b.add_u64(l_bluestore_onodes, "bluestore_onodes", | |
4474 | "Number of onodes in cache"); | |
4475 | b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits", | |
4476 | "Sum for onode-lookups hit in the cache"); | |
4477 | b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses", | |
4478 | "Sum for onode-lookups missed in the cache"); | |
4479 | b.add_u64_counter(l_bluestore_onode_shard_hits, "bluestore_onode_shard_hits", | |
4480 | "Sum for onode-shard lookups hit in the cache"); | |
4481 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
4482 | "bluestore_onode_shard_misses", | |
4483 | "Sum for onode-shard lookups missed in the cache"); | |
4484 | b.add_u64(l_bluestore_extents, "bluestore_extents", | |
4485 | "Number of extents in cache"); | |
4486 | b.add_u64(l_bluestore_blobs, "bluestore_blobs", | |
4487 | "Number of blobs in cache"); | |
4488 | b.add_u64(l_bluestore_buffers, "bluestore_buffers", | |
4489 | "Number of buffers in cache"); | |
4490 | b.add_u64(l_bluestore_buffer_bytes, "bluestore_buffer_bytes", | |
11fdf7f2 | 4491 | "Number of buffer bytes in cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4492 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "bluestore_buffer_hit_bytes", |
11fdf7f2 | 4493 | "Sum for bytes of read hit in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4494 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "bluestore_buffer_miss_bytes", |
11fdf7f2 | 4495 | "Sum for bytes of read missed in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4496 | |
4497 | b.add_u64_counter(l_bluestore_write_big, "bluestore_write_big", | |
4498 | "Large aligned writes into fresh blobs"); | |
4499 | b.add_u64_counter(l_bluestore_write_big_bytes, "bluestore_write_big_bytes", | |
11fdf7f2 | 4500 | "Large aligned writes into fresh blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4501 | b.add_u64_counter(l_bluestore_write_big_blobs, "bluestore_write_big_blobs", |
4502 | "Large aligned writes into fresh blobs (blobs)"); | |
4503 | b.add_u64_counter(l_bluestore_write_small, "bluestore_write_small", | |
4504 | "Small writes into existing or sparse small blobs"); | |
4505 | b.add_u64_counter(l_bluestore_write_small_bytes, "bluestore_write_small_bytes", | |
11fdf7f2 | 4506 | "Small writes into existing or sparse small blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4507 | b.add_u64_counter(l_bluestore_write_small_unused, |
4508 | "bluestore_write_small_unused", | |
4509 | "Small writes into unused portion of existing blob"); | |
4510 | b.add_u64_counter(l_bluestore_write_small_deferred, | |
4511 | "bluestore_write_small_deferred", | |
4512 | "Small overwrites using deferred"); | |
4513 | b.add_u64_counter(l_bluestore_write_small_pre_read, | |
4514 | "bluestore_write_small_pre_read", | |
4515 | "Small writes that required we read some data (possibly " | |
4516 | "cached) to fill out the block"); | |
4517 | b.add_u64_counter(l_bluestore_write_small_new, "bluestore_write_small_new", | |
4518 | "Small write into new (sparse) blob"); | |
4519 | ||
4520 | b.add_u64_counter(l_bluestore_txc, "bluestore_txc", "Transactions committed"); | |
4521 | b.add_u64_counter(l_bluestore_onode_reshard, "bluestore_onode_reshard", | |
4522 | "Onode extent map reshard events"); | |
4523 | b.add_u64_counter(l_bluestore_blob_split, "bluestore_blob_split", | |
4524 | "Sum for blob splitting due to resharding"); | |
4525 | b.add_u64_counter(l_bluestore_extent_compress, "bluestore_extent_compress", | |
4526 | "Sum for extents that have been removed due to compression"); | |
4527 | b.add_u64_counter(l_bluestore_gc_merged, "bluestore_gc_merged", | |
4528 | "Sum for extents that have been merged due to garbage " | |
4529 | "collection"); | |
b32b8144 FG |
4530 | b.add_u64_counter(l_bluestore_read_eio, "bluestore_read_eio", |
4531 | "Read EIO errors propagated to high level callers"); | |
f64942e4 AA |
4532 | b.add_u64_counter(l_bluestore_reads_with_retries, "bluestore_reads_with_retries", |
4533 | "Read operations that required at least one retry due to failed checksum validation"); | |
a8e16298 TL |
4534 | b.add_u64(l_bluestore_fragmentation, "bluestore_fragmentation_micros", |
4535 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
11fdf7f2 TL |
4536 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
4537 | "Average omap iterator seek_to_first call latency"); | |
4538 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", | |
4539 | "Average omap iterator upper_bound call latency"); | |
4540 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", | |
4541 | "Average omap iterator lower_bound call latency"); | |
4542 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", | |
4543 | "Average omap iterator next call latency"); | |
7c673cae FG |
4544 | logger = b.create_perf_counters(); |
4545 | cct->get_perfcounters_collection()->add(logger); | |
4546 | } | |
4547 | ||
4548 | int BlueStore::_reload_logger() | |
4549 | { | |
4550 | struct store_statfs_t store_statfs; | |
7c673cae | 4551 | int r = statfs(&store_statfs); |
11fdf7f2 | 4552 | if (r >= 0) { |
7c673cae | 4553 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
4554 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
4555 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
4556 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
4557 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
4558 | } |
4559 | return r; | |
4560 | } | |
4561 | ||
4562 | void BlueStore::_shutdown_logger() | |
4563 | { | |
4564 | cct->get_perfcounters_collection()->remove(logger); | |
4565 | delete logger; | |
4566 | } | |
4567 | ||
4568 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
4569 | uuid_d *fsid) | |
4570 | { | |
4571 | bluestore_bdev_label_t label; | |
4572 | int r = _read_bdev_label(cct, path, &label); | |
4573 | if (r < 0) | |
4574 | return r; | |
4575 | *fsid = label.osd_uuid; | |
4576 | return 0; | |
4577 | } | |
4578 | ||
4579 | int BlueStore::_open_path() | |
4580 | { | |
b32b8144 | 4581 | // sanity check(s) |
11fdf7f2 TL |
4582 | auto osd_max_object_size = |
4583 | cct->_conf.get_val<Option::size_t>("osd_max_object_size"); | |
4584 | if (osd_max_object_size >= (size_t)OBJECT_MAX_SIZE) { | |
4585 | derr << __func__ << " osd_max_object_size >= 0x" << std::hex << OBJECT_MAX_SIZE | |
4586 | << "; BlueStore has hard limit of 0x" << OBJECT_MAX_SIZE << "." << std::dec << dendl; | |
b32b8144 FG |
4587 | return -EINVAL; |
4588 | } | |
11fdf7f2 | 4589 | ceph_assert(path_fd < 0); |
91327a77 | 4590 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
4591 | if (path_fd < 0) { |
4592 | int r = -errno; | |
4593 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
4594 | << dendl; | |
4595 | return r; | |
4596 | } | |
4597 | return 0; | |
4598 | } | |
4599 | ||
4600 | void BlueStore::_close_path() | |
4601 | { | |
4602 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
4603 | path_fd = -1; | |
4604 | } | |
4605 | ||
3efd9988 FG |
4606 | int BlueStore::_write_bdev_label(CephContext *cct, |
4607 | string path, bluestore_bdev_label_t label) | |
7c673cae FG |
4608 | { |
4609 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
4610 | bufferlist bl; | |
11fdf7f2 | 4611 | encode(label, bl); |
7c673cae | 4612 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
4613 | encode(crc, bl); |
4614 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
4615 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
4616 | z.zero(); | |
4617 | bl.append(std::move(z)); | |
4618 | ||
91327a77 | 4619 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC)); |
7c673cae FG |
4620 | if (fd < 0) { |
4621 | fd = -errno; | |
4622 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
4623 | << dendl; | |
4624 | return fd; | |
4625 | } | |
4626 | int r = bl.write_fd(fd); | |
4627 | if (r < 0) { | |
4628 | derr << __func__ << " failed to write to " << path | |
4629 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 4630 | goto out; |
7c673cae | 4631 | } |
3efd9988 FG |
4632 | r = ::fsync(fd); |
4633 | if (r < 0) { | |
4634 | derr << __func__ << " failed to fsync " << path | |
4635 | << ": " << cpp_strerror(r) << dendl; | |
4636 | } | |
11fdf7f2 | 4637 | out: |
7c673cae FG |
4638 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
4639 | return r; | |
4640 | } | |
4641 | ||
4642 | int BlueStore::_read_bdev_label(CephContext* cct, string path, | |
4643 | bluestore_bdev_label_t *label) | |
4644 | { | |
4645 | dout(10) << __func__ << dendl; | |
91327a77 | 4646 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
4647 | if (fd < 0) { |
4648 | fd = -errno; | |
4649 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
4650 | << dendl; | |
4651 | return fd; | |
4652 | } | |
4653 | bufferlist bl; | |
4654 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
4655 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
4656 | if (r < 0) { | |
4657 | derr << __func__ << " failed to read from " << path | |
4658 | << ": " << cpp_strerror(r) << dendl; | |
4659 | return r; | |
4660 | } | |
4661 | ||
4662 | uint32_t crc, expected_crc; | |
11fdf7f2 | 4663 | auto p = bl.cbegin(); |
7c673cae | 4664 | try { |
11fdf7f2 | 4665 | decode(*label, p); |
7c673cae FG |
4666 | bufferlist t; |
4667 | t.substr_of(bl, 0, p.get_off()); | |
4668 | crc = t.crc32c(-1); | |
11fdf7f2 | 4669 | decode(expected_crc, p); |
7c673cae FG |
4670 | } |
4671 | catch (buffer::error& e) { | |
b32b8144 | 4672 | dout(2) << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
4673 | << ": " << e.what() |
4674 | << dendl; | |
b32b8144 | 4675 | return -ENOENT; |
7c673cae FG |
4676 | } |
4677 | if (crc != expected_crc) { | |
4678 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
4679 | << " != actual " << crc << dendl; | |
4680 | return -EIO; | |
4681 | } | |
4682 | dout(10) << __func__ << " got " << *label << dendl; | |
4683 | return 0; | |
4684 | } | |
4685 | ||
4686 | int BlueStore::_check_or_set_bdev_label( | |
4687 | string path, uint64_t size, string desc, bool create) | |
4688 | { | |
4689 | bluestore_bdev_label_t label; | |
4690 | if (create) { | |
4691 | label.osd_uuid = fsid; | |
4692 | label.size = size; | |
4693 | label.btime = ceph_clock_now(); | |
4694 | label.description = desc; | |
3efd9988 | 4695 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
4696 | if (r < 0) |
4697 | return r; | |
4698 | } else { | |
4699 | int r = _read_bdev_label(cct, path, &label); | |
4700 | if (r < 0) | |
4701 | return r; | |
31f18b77 FG |
4702 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
4703 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
4704 | << " and fsid " << fsid << " check bypassed" << dendl; | |
4705 | } | |
4706 | else if (label.osd_uuid != fsid) { | |
7c673cae FG |
4707 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
4708 | << " does not match our fsid " << fsid << dendl; | |
4709 | return -EIO; | |
4710 | } | |
4711 | } | |
4712 | return 0; | |
4713 | } | |
4714 | ||
4715 | void BlueStore::_set_alloc_sizes(void) | |
4716 | { | |
7c673cae FG |
4717 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
4718 | ||
4719 | if (cct->_conf->bluestore_prefer_deferred_size) { | |
4720 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
4721 | } else { | |
11fdf7f2 | 4722 | ceph_assert(bdev); |
7c673cae FG |
4723 | if (bdev->is_rotational()) { |
4724 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; | |
4725 | } else { | |
4726 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
4727 | } | |
4728 | } | |
4729 | ||
4730 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
4731 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
4732 | } else { | |
11fdf7f2 | 4733 | ceph_assert(bdev); |
7c673cae FG |
4734 | if (bdev->is_rotational()) { |
4735 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; | |
4736 | } else { | |
4737 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
4738 | } | |
4739 | } | |
4740 | ||
4741 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 4742 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
4743 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
4744 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
4745 | << std::dec | |
4746 | << " deferred_batch_ops " << deferred_batch_ops | |
4747 | << dendl; | |
4748 | } | |
4749 | ||
4750 | int BlueStore::_open_bdev(bool create) | |
4751 | { | |
11fdf7f2 | 4752 | ceph_assert(bdev == NULL); |
7c673cae | 4753 | string p = path + "/block"; |
11fdf7f2 | 4754 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
4755 | int r = bdev->open(p); |
4756 | if (r < 0) | |
4757 | goto fail; | |
4758 | ||
11fdf7f2 TL |
4759 | if (create && cct->_conf->bdev_enable_discard) { |
4760 | bdev->discard(0, bdev->get_size()); | |
4761 | } | |
4762 | ||
7c673cae FG |
4763 | if (bdev->supported_bdev_label()) { |
4764 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
4765 | if (r < 0) | |
4766 | goto fail_close; | |
4767 | } | |
4768 | ||
4769 | // initialize global block parameters | |
4770 | block_size = bdev->get_block_size(); | |
4771 | block_mask = ~(block_size - 1); | |
4772 | block_size_order = ctz(block_size); | |
11fdf7f2 | 4773 | ceph_assert(block_size == 1u << block_size_order); |
224ce89b WB |
4774 | // and set cache_size based on device type |
4775 | r = _set_cache_sizes(); | |
4776 | if (r < 0) { | |
4777 | goto fail_close; | |
4778 | } | |
7c673cae FG |
4779 | return 0; |
4780 | ||
4781 | fail_close: | |
4782 | bdev->close(); | |
4783 | fail: | |
4784 | delete bdev; | |
4785 | bdev = NULL; | |
4786 | return r; | |
4787 | } | |
4788 | ||
11fdf7f2 TL |
4789 | void BlueStore::_validate_bdev() |
4790 | { | |
4791 | ceph_assert(bdev); | |
4792 | ceph_assert(min_alloc_size); // _get_odisk_reserved depends on that | |
4793 | uint64_t dev_size = bdev->get_size(); | |
4794 | if (dev_size < | |
4795 | _get_ondisk_reserved() + cct->_conf->bluestore_bluefs_min) { | |
4796 | dout(1) << __func__ << " main device size " << byte_u_t(dev_size) | |
4797 | << " is too small, disable bluestore_bluefs_min for now" | |
4798 | << dendl; | |
4799 | ceph_assert(dev_size >= _get_ondisk_reserved()); | |
4800 | ||
4801 | int r = cct->_conf.set_val("bluestore_bluefs_min", "0"); | |
4802 | ceph_assert(r == 0); | |
4803 | } | |
4804 | } | |
4805 | ||
7c673cae FG |
4806 | void BlueStore::_close_bdev() |
4807 | { | |
11fdf7f2 | 4808 | ceph_assert(bdev); |
7c673cae FG |
4809 | bdev->close(); |
4810 | delete bdev; | |
4811 | bdev = NULL; | |
4812 | } | |
4813 | ||
11fdf7f2 | 4814 | int BlueStore::_open_fm(KeyValueDB::Transaction t) |
7c673cae | 4815 | { |
11fdf7f2 TL |
4816 | ceph_assert(fm == NULL); |
4817 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); | |
4818 | ceph_assert(fm); | |
4819 | if (t) { | |
4820 | // create mode. initialize freespace | |
7c673cae | 4821 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
4822 | { |
4823 | bufferlist bl; | |
4824 | bl.append(freelist_type); | |
4825 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
4826 | } | |
b32b8144 FG |
4827 | // being able to allocate in units less than bdev block size |
4828 | // seems to be a bad idea. | |
11fdf7f2 | 4829 | ceph_assert( cct->_conf->bdev_block_size <= (int64_t)min_alloc_size); |
b32b8144 | 4830 | fm->create(bdev->get_size(), (int64_t)min_alloc_size, t); |
7c673cae FG |
4831 | |
4832 | // allocate superblock reserved space. note that we do not mark | |
4833 | // bluefs space as allocated in the freelist; we instead rely on | |
4834 | // bluefs_extents. | |
11fdf7f2 | 4835 | auto reserved = _get_ondisk_reserved(); |
3efd9988 | 4836 | fm->allocate(0, reserved, t); |
7c673cae | 4837 | |
7c673cae | 4838 | if (cct->_conf->bluestore_bluefs) { |
11fdf7f2 | 4839 | ceph_assert(bluefs_extents.num_intervals() == 1); |
7c673cae | 4840 | interval_set<uint64_t>::iterator p = bluefs_extents.begin(); |
11fdf7f2 | 4841 | reserved = round_up_to(p.get_start() + p.get_len(), min_alloc_size); |
7c673cae FG |
4842 | dout(20) << __func__ << " reserved 0x" << std::hex << reserved << std::dec |
4843 | << " for bluefs" << dendl; | |
7c673cae FG |
4844 | } |
4845 | ||
4846 | if (cct->_conf->bluestore_debug_prefill > 0) { | |
4847 | uint64_t end = bdev->get_size() - reserved; | |
4848 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
4849 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
4850 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 4851 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
4852 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
4853 | float r = cct->_conf->bluestore_debug_prefill; | |
4854 | r /= 1.0 - r; | |
4855 | bool stop = false; | |
4856 | ||
4857 | while (!stop && start < end) { | |
4858 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
4859 | if (start + l > end) { | |
4860 | l = end - start; | |
11fdf7f2 | 4861 | l = p2align(l, min_alloc_size); |
7c673cae | 4862 | } |
11fdf7f2 | 4863 | ceph_assert(start + l <= end); |
7c673cae FG |
4864 | |
4865 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 4866 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
4867 | if (start + l + u > end) { |
4868 | u = end - (start + l); | |
4869 | // trim to align so we don't overflow again | |
11fdf7f2 | 4870 | u = p2align(u, min_alloc_size); |
7c673cae FG |
4871 | stop = true; |
4872 | } | |
11fdf7f2 | 4873 | ceph_assert(start + l + u <= end); |
7c673cae | 4874 | |
11fdf7f2 | 4875 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
4876 | << " use 0x" << u << std::dec << dendl; |
4877 | ||
4878 | if (u == 0) { | |
4879 | // break if u has been trimmed to nothing | |
4880 | break; | |
4881 | } | |
4882 | ||
4883 | fm->allocate(start + l, u, t); | |
4884 | start += l + u; | |
4885 | } | |
4886 | } | |
7c673cae FG |
4887 | } |
4888 | ||
11fdf7f2 | 4889 | int r = fm->init(db); |
7c673cae FG |
4890 | if (r < 0) { |
4891 | derr << __func__ << " freelist init failed: " << cpp_strerror(r) << dendl; | |
4892 | delete fm; | |
4893 | fm = NULL; | |
4894 | return r; | |
4895 | } | |
81eedcae TL |
4896 | // if space size tracked by free list manager is that higher than actual |
4897 | // dev size one can hit out-of-space allocation which will result | |
4898 | // in data loss and/or assertions | |
4899 | // Probably user altered the device size somehow. | |
4900 | // The only fix for now is to redeploy OSD. | |
4901 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
4902 | ostringstream ss; | |
4903 | ss << "slow device size mismatch detected, " | |
4904 | << " fm size(" << fm->get_size() | |
4905 | << ") > slow device size(" << bdev->get_size() | |
4906 | << "), Please stop using this OSD as it might cause data loss."; | |
4907 | _set_disk_size_mismatch_alert(ss.str()); | |
4908 | } | |
7c673cae FG |
4909 | return 0; |
4910 | } | |
4911 | ||
4912 | void BlueStore::_close_fm() | |
4913 | { | |
4914 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 4915 | ceph_assert(fm); |
7c673cae FG |
4916 | fm->shutdown(); |
4917 | delete fm; | |
4918 | fm = NULL; | |
4919 | } | |
4920 | ||
4921 | int BlueStore::_open_alloc() | |
4922 | { | |
11fdf7f2 TL |
4923 | ceph_assert(alloc == NULL); |
4924 | ceph_assert(bdev->get_size()); | |
4925 | ||
4926 | if (bluefs) { | |
4927 | bluefs_extents.clear(); | |
4928 | auto r = bluefs->get_block_extents(bluefs_shared_bdev, &bluefs_extents); | |
4929 | if (r < 0) { | |
4930 | lderr(cct) << __func__ << " failed to retrieve bluefs_extents: " | |
4931 | << cpp_strerror(r) << dendl; | |
4932 | ||
4933 | return r; | |
4934 | } | |
4935 | dout(10) << __func__ << " bluefs extents 0x" | |
4936 | << std::hex << bluefs_extents << std::dec | |
4937 | << dendl; | |
4938 | } | |
4939 | ||
7c673cae FG |
4940 | alloc = Allocator::create(cct, cct->_conf->bluestore_allocator, |
4941 | bdev->get_size(), | |
4942 | min_alloc_size); | |
4943 | if (!alloc) { | |
4944 | lderr(cct) << __func__ << " Allocator::unknown alloc type " | |
4945 | << cct->_conf->bluestore_allocator | |
4946 | << dendl; | |
4947 | return -EINVAL; | |
4948 | } | |
4949 | ||
4950 | uint64_t num = 0, bytes = 0; | |
4951 | ||
4952 | dout(1) << __func__ << " opening allocation metadata" << dendl; | |
4953 | // initialize from freelist | |
4954 | fm->enumerate_reset(); | |
4955 | uint64_t offset, length; | |
11fdf7f2 | 4956 | while (fm->enumerate_next(db, &offset, &length)) { |
7c673cae FG |
4957 | alloc->init_add_free(offset, length); |
4958 | ++num; | |
4959 | bytes += length; | |
4960 | } | |
224ce89b | 4961 | fm->enumerate_reset(); |
1adf2230 | 4962 | dout(1) << __func__ << " loaded " << byte_u_t(bytes) |
7c673cae FG |
4963 | << " in " << num << " extents" |
4964 | << dendl; | |
4965 | ||
4966 | // also mark bluefs space as allocated | |
4967 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
4968 | alloc->init_rm_free(e.get_start(), e.get_len()); | |
4969 | } | |
7c673cae FG |
4970 | |
4971 | return 0; | |
4972 | } | |
4973 | ||
4974 | void BlueStore::_close_alloc() | |
4975 | { | |
11fdf7f2 TL |
4976 | ceph_assert(bdev); |
4977 | bdev->discard_drain(); | |
4978 | ||
4979 | ceph_assert(alloc); | |
7c673cae FG |
4980 | alloc->shutdown(); |
4981 | delete alloc; | |
4982 | alloc = NULL; | |
11fdf7f2 | 4983 | bluefs_extents.clear(); |
7c673cae FG |
4984 | } |
4985 | ||
4986 | int BlueStore::_open_fsid(bool create) | |
4987 | { | |
11fdf7f2 | 4988 | ceph_assert(fsid_fd < 0); |
91327a77 | 4989 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
4990 | if (create) |
4991 | flags |= O_CREAT; | |
4992 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
4993 | if (fsid_fd < 0) { | |
4994 | int err = -errno; | |
4995 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
4996 | return err; | |
4997 | } | |
4998 | return 0; | |
4999 | } | |
5000 | ||
5001 | int BlueStore::_read_fsid(uuid_d *uuid) | |
5002 | { | |
5003 | char fsid_str[40]; | |
5004 | memset(fsid_str, 0, sizeof(fsid_str)); | |
5005 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
5006 | if (ret < 0) { | |
5007 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
5008 | return ret; | |
5009 | } | |
5010 | if (ret > 36) | |
5011 | fsid_str[36] = 0; | |
5012 | else | |
5013 | fsid_str[ret] = 0; | |
5014 | if (!uuid->parse(fsid_str)) { | |
5015 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
5016 | return -EINVAL; | |
5017 | } | |
5018 | return 0; | |
5019 | } | |
5020 | ||
5021 | int BlueStore::_write_fsid() | |
5022 | { | |
5023 | int r = ::ftruncate(fsid_fd, 0); | |
5024 | if (r < 0) { | |
5025 | r = -errno; | |
5026 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
5027 | return r; | |
5028 | } | |
5029 | string str = stringify(fsid) + "\n"; | |
5030 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
5031 | if (r < 0) { | |
5032 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
5033 | return r; | |
5034 | } | |
5035 | r = ::fsync(fsid_fd); | |
5036 | if (r < 0) { | |
5037 | r = -errno; | |
5038 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
5039 | return r; | |
5040 | } | |
5041 | return 0; | |
5042 | } | |
5043 | ||
5044 | void BlueStore::_close_fsid() | |
5045 | { | |
5046 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
5047 | fsid_fd = -1; | |
5048 | } | |
5049 | ||
5050 | int BlueStore::_lock_fsid() | |
5051 | { | |
5052 | struct flock l; | |
5053 | memset(&l, 0, sizeof(l)); | |
5054 | l.l_type = F_WRLCK; | |
5055 | l.l_whence = SEEK_SET; | |
5056 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
5057 | if (r < 0) { | |
5058 | int err = errno; | |
5059 | derr << __func__ << " failed to lock " << path << "/fsid" | |
5060 | << " (is another ceph-osd still running?)" | |
5061 | << cpp_strerror(err) << dendl; | |
5062 | return -err; | |
5063 | } | |
5064 | return 0; | |
5065 | } | |
5066 | ||
31f18b77 FG |
5067 | bool BlueStore::is_rotational() |
5068 | { | |
5069 | if (bdev) { | |
5070 | return bdev->is_rotational(); | |
5071 | } | |
5072 | ||
5073 | bool rotational = true; | |
5074 | int r = _open_path(); | |
5075 | if (r < 0) | |
5076 | goto out; | |
5077 | r = _open_fsid(false); | |
5078 | if (r < 0) | |
5079 | goto out_path; | |
5080 | r = _read_fsid(&fsid); | |
5081 | if (r < 0) | |
5082 | goto out_fsid; | |
5083 | r = _lock_fsid(); | |
5084 | if (r < 0) | |
5085 | goto out_fsid; | |
5086 | r = _open_bdev(false); | |
5087 | if (r < 0) | |
5088 | goto out_fsid; | |
5089 | rotational = bdev->is_rotational(); | |
5090 | _close_bdev(); | |
5091 | out_fsid: | |
5092 | _close_fsid(); | |
5093 | out_path: | |
5094 | _close_path(); | |
5095 | out: | |
5096 | return rotational; | |
5097 | } | |
5098 | ||
d2e6a577 FG |
5099 | bool BlueStore::is_journal_rotational() |
5100 | { | |
5101 | if (!bluefs) { | |
5102 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
5103 | << dendl; | |
5104 | return is_rotational(); | |
5105 | } | |
5106 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
5107 | return bluefs->wal_is_rotational(); | |
5108 | } | |
5109 | ||
7c673cae FG |
5110 | bool BlueStore::test_mount_in_use() |
5111 | { | |
5112 | // most error conditions mean the mount is not in use (e.g., because | |
5113 | // it doesn't exist). only if we fail to lock do we conclude it is | |
5114 | // in use. | |
5115 | bool ret = false; | |
5116 | int r = _open_path(); | |
5117 | if (r < 0) | |
5118 | return false; | |
5119 | r = _open_fsid(false); | |
5120 | if (r < 0) | |
5121 | goto out_path; | |
5122 | r = _lock_fsid(); | |
5123 | if (r < 0) | |
5124 | ret = true; // if we can't lock, it is in use | |
5125 | _close_fsid(); | |
5126 | out_path: | |
5127 | _close_path(); | |
5128 | return ret; | |
5129 | } | |
5130 | ||
11fdf7f2 | 5131 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
5132 | { |
5133 | int r; | |
11fdf7f2 | 5134 | bluefs = new BlueFS(cct); |
7c673cae | 5135 | |
11fdf7f2 TL |
5136 | string bfn; |
5137 | struct stat st; | |
5138 | ||
5139 | bfn = path + "/block.db"; | |
5140 | if (::stat(bfn.c_str(), &st) == 0) { | |
5141 | r = bluefs->add_block_device(BlueFS::BDEV_DB, bfn, | |
5142 | create && cct->_conf->bdev_enable_discard); | |
7c673cae | 5143 | if (r < 0) { |
11fdf7f2 TL |
5144 | derr << __func__ << " add block device(" << bfn << ") returned: " |
5145 | << cpp_strerror(r) << dendl; | |
5146 | goto free_bluefs; | |
7c673cae | 5147 | } |
7c673cae | 5148 | |
11fdf7f2 TL |
5149 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
5150 | r = _check_or_set_bdev_label( | |
5151 | bfn, | |
5152 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
5153 | "bluefs db", create); | |
5154 | if (r < 0) { | |
5155 | derr << __func__ | |
5156 | << " check block device(" << bfn << ") label returned: " | |
5157 | << cpp_strerror(r) << dendl; | |
5158 | goto free_bluefs; | |
5159 | } | |
7c673cae | 5160 | } |
11fdf7f2 TL |
5161 | if (create) { |
5162 | bluefs->add_block_extent( | |
5163 | BlueFS::BDEV_DB, | |
5164 | SUPER_RESERVED, | |
5165 | bluefs->get_block_device_size(BlueFS::BDEV_DB) - SUPER_RESERVED); | |
5166 | } | |
5167 | bluefs_shared_bdev = BlueFS::BDEV_SLOW; | |
5168 | bluefs_single_shared_device = false; | |
5169 | } else { | |
5170 | r = -errno; | |
5171 | if (::lstat(bfn.c_str(), &st) == -1) { | |
5172 | r = 0; | |
5173 | bluefs_shared_bdev = BlueFS::BDEV_DB; | |
7c673cae | 5174 | } else { |
11fdf7f2 TL |
5175 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
5176 | << cpp_strerror(r) << dendl; | |
5177 | goto free_bluefs; | |
7c673cae FG |
5178 | } |
5179 | } | |
7c673cae | 5180 | |
11fdf7f2 TL |
5181 | // shared device |
5182 | bfn = path + "/block"; | |
5183 | // never trim here | |
5184 | r = bluefs->add_block_device(bluefs_shared_bdev, bfn, false, | |
5185 | true /* shared with bluestore */); | |
5186 | if (r < 0) { | |
5187 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5188 | << cpp_strerror(r) << dendl; | |
5189 | goto free_bluefs; | |
5190 | } | |
5191 | if (create) { | |
5192 | // note: we always leave the first SUPER_RESERVED (8k) of the device unused | |
5193 | uint64_t initial = | |
5194 | bdev->get_size() * (cct->_conf->bluestore_bluefs_min_ratio + | |
5195 | cct->_conf->bluestore_bluefs_gift_ratio); | |
5196 | initial = std::max(initial, cct->_conf->bluestore_bluefs_min); | |
5197 | if (cct->_conf->bluefs_alloc_size % min_alloc_size) { | |
5198 | derr << __func__ << " bluefs_alloc_size 0x" << std::hex | |
5199 | << cct->_conf->bluefs_alloc_size << " is not a multiple of " | |
5200 | << "min_alloc_size 0x" << min_alloc_size << std::dec << dendl; | |
5201 | r = -EINVAL; | |
5202 | goto free_bluefs; | |
7c673cae | 5203 | } |
11fdf7f2 TL |
5204 | // align to bluefs's alloc_size |
5205 | initial = p2roundup(initial, cct->_conf->bluefs_alloc_size); | |
5206 | // put bluefs in the middle of the device in case it is an HDD | |
5207 | uint64_t start = p2align((bdev->get_size() - initial) / 2, | |
5208 | cct->_conf->bluefs_alloc_size); | |
5209 | //avoiding superblock overwrite | |
5210 | ceph_assert(cct->_conf->bluefs_alloc_size > _get_ondisk_reserved()); | |
5211 | start = std::max(cct->_conf->bluefs_alloc_size, start); | |
7c673cae | 5212 | |
11fdf7f2 TL |
5213 | bluefs->add_block_extent(bluefs_shared_bdev, start, initial); |
5214 | bluefs_extents.insert(start, initial); | |
5215 | ++out_of_sync_fm; | |
5216 | } | |
5217 | ||
5218 | bfn = path + "/block.wal"; | |
5219 | if (::stat(bfn.c_str(), &st) == 0) { | |
5220 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
5221 | create && cct->_conf->bdev_enable_discard); | |
5222 | if (r < 0) { | |
5223 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5224 | << cpp_strerror(r) << dendl; | |
5225 | goto free_bluefs; | |
5226 | } | |
7c673cae | 5227 | |
11fdf7f2 TL |
5228 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
5229 | r = _check_or_set_bdev_label( | |
5230 | bfn, | |
5231 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
5232 | "bluefs wal", create); | |
7c673cae | 5233 | if (r < 0) { |
11fdf7f2 TL |
5234 | derr << __func__ << " check block device(" << bfn |
5235 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
5236 | goto free_bluefs; |
5237 | } | |
7c673cae FG |
5238 | } |
5239 | ||
11fdf7f2 TL |
5240 | if (create) { |
5241 | bluefs->add_block_extent( | |
5242 | BlueFS::BDEV_WAL, BDEV_LABEL_BLOCK_SIZE, | |
5243 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) - | |
5244 | BDEV_LABEL_BLOCK_SIZE); | |
5245 | } | |
5246 | bluefs_single_shared_device = false; | |
5247 | } else { | |
5248 | r = 0; | |
5249 | if (::lstat(bfn.c_str(), &st) != -1) { | |
5250 | r = -errno; | |
5251 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
5252 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
5253 | goto free_bluefs; |
5254 | } | |
11fdf7f2 TL |
5255 | } |
5256 | return 0; | |
7c673cae | 5257 | |
11fdf7f2 TL |
5258 | free_bluefs: |
5259 | ceph_assert(bluefs); | |
5260 | delete bluefs; | |
5261 | bluefs = NULL; | |
5262 | return r; | |
5263 | } | |
7c673cae | 5264 | |
11fdf7f2 TL |
5265 | int BlueStore::_open_bluefs(bool create) |
5266 | { | |
5267 | int r = _minimal_open_bluefs(create); | |
5268 | if (r < 0) { | |
5269 | return r; | |
5270 | } | |
5271 | if (create) { | |
5272 | bluefs->mkfs(fsid); | |
5273 | } | |
5274 | r = bluefs->mount(); | |
5275 | if (r < 0) { | |
5276 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
5277 | } | |
5278 | return r; | |
5279 | } | |
5280 | ||
5281 | void BlueStore::_close_bluefs() | |
5282 | { | |
5283 | bluefs->umount(); | |
5284 | _minimal_close_bluefs(); | |
5285 | } | |
5286 | ||
5287 | void BlueStore::_minimal_close_bluefs() | |
5288 | { | |
5289 | delete bluefs; | |
5290 | bluefs = NULL; | |
5291 | } | |
5292 | ||
5293 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
5294 | { | |
5295 | if (create) { | |
5296 | *ret = cct->_conf->bluestore_bluefs; | |
5297 | } else { | |
5298 | string s; | |
5299 | int r = read_meta("bluefs", &s); | |
5300 | if (r < 0) { | |
5301 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
5302 | return -EIO; | |
5303 | } | |
5304 | if (s == "1") { | |
5305 | *ret = true; | |
5306 | } else if (s == "0") { | |
5307 | *ret = false; | |
31f18b77 | 5308 | } else { |
11fdf7f2 TL |
5309 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
5310 | << dendl; | |
5311 | return -EIO; | |
5312 | } | |
5313 | } | |
5314 | return 0; | |
5315 | } | |
5316 | ||
5317 | /* | |
5318 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
5319 | * in the proper order | |
5320 | */ | |
5321 | int BlueStore::_open_db_and_around(bool read_only) | |
5322 | { | |
5323 | int r; | |
5324 | bool do_bluefs = false; | |
5325 | _is_bluefs(false, &do_bluefs); // ignore err code | |
5326 | if (do_bluefs) { | |
5327 | // open in read-only first to read FM list and init allocator | |
5328 | // as they might be needed for some BlueFS procedures | |
5329 | r = _open_db(false, false, true); | |
5330 | if (r < 0) | |
5331 | return r; | |
5332 | ||
5333 | r = _open_super_meta(); | |
5334 | if (r < 0) { | |
5335 | goto out_db; | |
5336 | } | |
5337 | ||
5338 | r = _open_fm(nullptr); | |
5339 | if (r < 0) | |
5340 | goto out_db; | |
5341 | ||
5342 | r = _open_alloc(); | |
5343 | if (r < 0) | |
5344 | goto out_fm; | |
5345 | ||
5346 | // now open in R/W mode | |
5347 | if (!read_only) { | |
5348 | _close_db(); | |
5349 | ||
5350 | r = _open_db(false, false, false); | |
5351 | if (r < 0) { | |
5352 | _close_alloc(); | |
5353 | _close_fm(); | |
5354 | return r; | |
28e407b8 | 5355 | } |
7c673cae | 5356 | } |
11fdf7f2 TL |
5357 | } else { |
5358 | r = _open_db(false, false); | |
5359 | if (r < 0) { | |
5360 | return r; | |
5361 | } | |
5362 | r = _open_super_meta(); | |
5363 | if (r < 0) { | |
5364 | goto out_db; | |
5365 | } | |
7c673cae | 5366 | |
11fdf7f2 TL |
5367 | r = _open_fm(nullptr); |
5368 | if (r < 0) | |
5369 | goto out_db; | |
5370 | ||
5371 | r = _open_alloc(); | |
5372 | if (r < 0) | |
5373 | goto out_fm; | |
5374 | } | |
5375 | return 0; | |
5376 | ||
5377 | out_fm: | |
5378 | _close_fm(); | |
5379 | out_db: | |
5380 | _close_db(); | |
5381 | return r; | |
5382 | } | |
5383 | ||
5384 | void BlueStore::_close_db_and_around() | |
5385 | { | |
5386 | if (bluefs) { | |
5387 | if (out_of_sync_fm.fetch_and(0)) { | |
5388 | _sync_bluefs_and_fm(); | |
5389 | } | |
5390 | _close_db(); | |
5391 | while(out_of_sync_fm.fetch_and(0)) { | |
5392 | // if seen some allocations during close - repeat open_db, sync fm, close | |
5393 | dout(0) << __func__ << " syncing FreelistManager" << dendl; | |
5394 | int r = _open_db(false, false, false); | |
5395 | if (r < 0) { | |
5396 | derr << __func__ | |
5397 | << " unable to open db, FreelistManager is probably out of sync" | |
5398 | << dendl; | |
5399 | break; | |
5400 | } | |
5401 | _sync_bluefs_and_fm(); | |
5402 | _close_db(); | |
7c673cae | 5403 | } |
11fdf7f2 TL |
5404 | if (!_kv_only) { |
5405 | _close_alloc(); | |
5406 | _close_fm(); | |
5407 | } | |
5408 | } else { | |
5409 | _close_alloc(); | |
5410 | _close_fm(); | |
5411 | _close_db(); | |
5412 | } | |
5413 | } | |
5414 | ||
5415 | // updates legacy bluefs related recs in DB to a state valid for | |
5416 | // downgrades from nautilus. | |
5417 | void BlueStore::_sync_bluefs_and_fm() | |
5418 | { | |
5419 | if (cct->_conf->bluestore_bluefs_db_compatibility) { | |
5420 | bufferlist bl; | |
5421 | encode(bluefs_extents, bl); | |
5422 | dout(20) << __func__ << " bluefs_extents at KV is now 0x" | |
5423 | << std::hex << bluefs_extents << std::dec | |
5424 | << dendl; | |
5425 | KeyValueDB::Transaction synct = db->get_transaction(); | |
5426 | synct->set(PREFIX_SUPER, "bluefs_extents", bl); | |
5427 | synct->set(PREFIX_SUPER, "bluefs_extents_back", bl); | |
5428 | ||
5429 | // Nice thing is that we don't need to update FreelistManager here. | |
5430 | // It always has corresponding bits set to 'Free' for both Nautilus+ and | |
5431 | // pre-Nautilis releases. | |
5432 | // So once we get an extent to bluefs_extents this means it's | |
5433 | // been free in allocator and hence it's free in FM too. | |
5434 | ||
5435 | db->submit_transaction_sync(synct); | |
5436 | } | |
5437 | } | |
5438 | ||
5439 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) | |
5440 | { | |
5441 | int r; | |
5442 | ceph_assert(!db); | |
5443 | ceph_assert(!(create && read_only)); | |
5444 | string fn = path + "/db"; | |
5445 | string options; | |
5446 | stringstream err; | |
5447 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); | |
5448 | ||
5449 | string kv_backend; | |
5450 | std::vector<KeyValueDB::ColumnFamily> cfs; | |
5451 | ||
5452 | if (create) { | |
5453 | kv_backend = cct->_conf->bluestore_kvbackend; | |
5454 | } else { | |
5455 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 5456 | if (r < 0) { |
11fdf7f2 TL |
5457 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
5458 | return -EIO; | |
5459 | } | |
5460 | } | |
5461 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
5462 | ||
5463 | bool do_bluefs; | |
5464 | r = _is_bluefs(create, &do_bluefs); | |
5465 | if (r < 0) { | |
5466 | return r; | |
5467 | } | |
5468 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
5469 | ||
5470 | map<string,string> kv_options; | |
5471 | // force separate wal dir for all new deployments. | |
5472 | kv_options["separate_wal_dir"] = 1; | |
5473 | rocksdb::Env *env = NULL; | |
5474 | if (do_bluefs) { | |
5475 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
5476 | if (kv_backend != "rocksdb") { | |
5477 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
5478 | return -EINVAL; | |
7c673cae | 5479 | } |
11fdf7f2 TL |
5480 | |
5481 | r = _open_bluefs(create); | |
5482 | if (r < 0) { | |
5483 | return r; | |
5484 | } | |
5485 | bluefs->set_slow_device_expander(this); | |
5486 | ||
7c673cae FG |
5487 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
5488 | rocksdb::Env *a = new BlueRocksEnv(bluefs); | |
5489 | rocksdb::Env *b = rocksdb::Env::Default(); | |
5490 | if (create) { | |
5491 | string cmd = "rm -rf " + path + "/db " + | |
5492 | path + "/db.slow " + | |
5493 | path + "/db.wal"; | |
5494 | int r = system(cmd.c_str()); | |
5495 | (void)r; | |
5496 | } | |
5497 | env = new rocksdb::EnvMirror(b, a, false, true); | |
5498 | } else { | |
5499 | env = new BlueRocksEnv(bluefs); | |
5500 | ||
5501 | // simplify the dir names, too, as "seen" by rocksdb | |
5502 | fn = "db"; | |
5503 | } | |
5504 | ||
5505 | if (bluefs_shared_bdev == BlueFS::BDEV_SLOW) { | |
5506 | // we have both block.db and block; tell rocksdb! | |
5507 | // note: the second (last) size value doesn't really matter | |
5508 | ostringstream db_paths; | |
5509 | uint64_t db_size = bluefs->get_block_device_size(BlueFS::BDEV_DB); | |
5510 | uint64_t slow_size = bluefs->get_block_device_size(BlueFS::BDEV_SLOW); | |
5511 | db_paths << fn << "," | |
5512 | << (uint64_t)(db_size * 95 / 100) << " " | |
5513 | << fn + ".slow" << "," | |
5514 | << (uint64_t)(slow_size * 95 / 100); | |
11fdf7f2 TL |
5515 | kv_options["db_paths"] = db_paths.str(); |
5516 | dout(10) << __func__ << " set db_paths to " << db_paths.str() << dendl; | |
7c673cae FG |
5517 | } |
5518 | ||
5519 | if (create) { | |
5520 | env->CreateDir(fn); | |
11fdf7f2 TL |
5521 | env->CreateDir(fn + ".wal"); |
5522 | env->CreateDir(fn + ".slow"); | |
5523 | } else { | |
5524 | std::vector<std::string> res; | |
5525 | // check for dir presence | |
5526 | auto r = env->GetChildren(fn+".wal", &res); | |
5527 | if (r.IsNotFound()) { | |
5528 | kv_options.erase("separate_wal_dir"); | |
5529 | } | |
7c673cae | 5530 | } |
11fdf7f2 TL |
5531 | } else { |
5532 | string walfn = path + "/db.wal"; | |
7c673cae | 5533 | |
11fdf7f2 TL |
5534 | if (create) { |
5535 | int r = ::mkdir(fn.c_str(), 0755); | |
5536 | if (r < 0) | |
5537 | r = -errno; | |
5538 | if (r < 0 && r != -EEXIST) { | |
5539 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
5540 | << dendl; | |
5541 | return r; | |
5542 | } | |
5543 | ||
5544 | // wal_dir, too! | |
7c673cae FG |
5545 | r = ::mkdir(walfn.c_str(), 0755); |
5546 | if (r < 0) | |
5547 | r = -errno; | |
5548 | if (r < 0 && r != -EEXIST) { | |
5549 | derr << __func__ << " failed to create " << walfn | |
5550 | << ": " << cpp_strerror(r) | |
5551 | << dendl; | |
5552 | return r; | |
5553 | } | |
11fdf7f2 TL |
5554 | } else { |
5555 | struct stat st; | |
5556 | r = ::stat(walfn.c_str(), &st); | |
5557 | if (r < 0 && errno == ENOENT) { | |
5558 | kv_options.erase("separate_wal_dir"); | |
5559 | } | |
7c673cae FG |
5560 | } |
5561 | } | |
5562 | ||
91327a77 | 5563 | |
7c673cae FG |
5564 | db = KeyValueDB::create(cct, |
5565 | kv_backend, | |
5566 | fn, | |
11fdf7f2 | 5567 | kv_options, |
7c673cae FG |
5568 | static_cast<void*>(env)); |
5569 | if (!db) { | |
5570 | derr << __func__ << " error creating db" << dendl; | |
5571 | if (bluefs) { | |
11fdf7f2 | 5572 | _close_bluefs(); |
7c673cae FG |
5573 | } |
5574 | // delete env manually here since we can't depend on db to do this | |
5575 | // under this case | |
5576 | delete env; | |
5577 | env = NULL; | |
5578 | return -EIO; | |
5579 | } | |
5580 | ||
5581 | FreelistManager::setup_merge_operators(db); | |
5582 | db->set_merge_operator(PREFIX_STAT, merge_op); | |
91327a77 | 5583 | db->set_cache_size(cache_kv_ratio * cache_size); |
31f18b77 | 5584 | |
11fdf7f2 | 5585 | if (kv_backend == "rocksdb") { |
7c673cae | 5586 | options = cct->_conf->bluestore_rocksdb_options; |
11fdf7f2 TL |
5587 | |
5588 | map<string,string> cf_map; | |
5589 | cct->_conf.with_val<string>("bluestore_rocksdb_cfs", | |
5590 | get_str_map, | |
5591 | &cf_map, | |
5592 | " \t"); | |
5593 | for (auto& i : cf_map) { | |
5594 | dout(10) << "column family " << i.first << ": " << i.second << dendl; | |
5595 | cfs.push_back(KeyValueDB::ColumnFamily(i.first, i.second)); | |
5596 | } | |
5597 | } | |
5598 | ||
7c673cae | 5599 | db->init(options); |
11fdf7f2 TL |
5600 | if (to_repair_db) |
5601 | return 0; | |
5602 | if (create) { | |
5603 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { | |
5604 | r = db->create_and_open(err, cfs); | |
5605 | } else { | |
5606 | r = db->create_and_open(err); | |
5607 | } | |
5608 | } else { | |
5609 | // we pass in cf list here, but it is only used if the db already has | |
5610 | // column families created. | |
5611 | r = read_only ? | |
5612 | db->open_read_only(err, cfs) : | |
5613 | db->open(err, cfs); | |
5614 | } | |
7c673cae FG |
5615 | if (r) { |
5616 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
11fdf7f2 | 5617 | _close_db(); |
7c673cae FG |
5618 | return -EIO; |
5619 | } | |
5620 | dout(1) << __func__ << " opened " << kv_backend | |
5621 | << " path " << fn << " options " << options << dendl; | |
5622 | return 0; | |
7c673cae FG |
5623 | } |
5624 | ||
5625 | void BlueStore::_close_db() | |
5626 | { | |
11fdf7f2 | 5627 | ceph_assert(db); |
7c673cae FG |
5628 | delete db; |
5629 | db = NULL; | |
5630 | if (bluefs) { | |
11fdf7f2 | 5631 | _close_bluefs(); |
7c673cae FG |
5632 | } |
5633 | } | |
5634 | ||
11fdf7f2 | 5635 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 5636 | { |
11fdf7f2 TL |
5637 | auto dump_interval = |
5638 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
5639 | if (dump_interval > 0 && | |
5640 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
5641 | alloc->dump(); | |
5642 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); | |
5643 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 5644 | } |
11fdf7f2 | 5645 | } |
7c673cae | 5646 | |
7c673cae | 5647 | |
11fdf7f2 TL |
5648 | int BlueStore::allocate_bluefs_freespace( |
5649 | uint64_t min_size, | |
5650 | uint64_t size, | |
5651 | PExtentVector* extents_out) | |
5652 | { | |
5653 | ceph_assert(min_size <= size); | |
5654 | if (size) { | |
5655 | // round up to alloc size | |
5656 | min_size = p2roundup(min_size, cct->_conf->bluefs_alloc_size); | |
5657 | size = p2roundup(size, cct->_conf->bluefs_alloc_size); | |
5658 | ||
5659 | PExtentVector extents_local; | |
5660 | PExtentVector* extents = extents_out ? extents_out : &extents_local; | |
5661 | ||
5662 | ||
5663 | uint64_t gift; | |
5664 | uint64_t allocated = 0; | |
5665 | int64_t alloc_len; | |
5666 | do { | |
5667 | // hard cap to fit into 32 bits | |
5668 | gift = std::min<uint64_t>(size, 1ull << 31); | |
5669 | dout(10) << __func__ << " gifting " << gift | |
5670 | << " (" << byte_u_t(gift) << ")" << dendl; | |
5671 | ||
5672 | alloc_len = alloc->allocate(gift, cct->_conf->bluefs_alloc_size, | |
5673 | 0, 0, extents); | |
5674 | if (alloc_len) { | |
5675 | allocated += alloc_len; | |
5676 | size -= alloc_len; | |
5677 | } | |
5678 | ||
5679 | if (alloc_len < (int64_t)gift && (min_size > allocated)) { | |
5680 | derr << __func__ | |
5681 | << " failed to allocate on 0x" << std::hex << gift | |
5682 | << " min_size 0x" << min_size | |
5683 | << " > allocated total 0x" << allocated | |
5684 | << " bluefs_alloc_size 0x" << cct->_conf->bluefs_alloc_size | |
5685 | << " allocated 0x" << alloc_len | |
5686 | << " available 0x " << alloc->get_free() | |
5687 | << std::dec << dendl; | |
7c673cae | 5688 | |
11fdf7f2 TL |
5689 | alloc->dump(); |
5690 | alloc->release(*extents); | |
5691 | extents->clear(); | |
5692 | return -ENOSPC; | |
5693 | } | |
5694 | } while (size && alloc_len > 0); | |
5695 | for (auto& e : *extents) { | |
5696 | dout(5) << __func__ << " gifting " << e << " to bluefs" << dendl; | |
5697 | bluefs_extents.insert(e.offset, e.length); | |
5698 | ++out_of_sync_fm; | |
5699 | // apply to bluefs if not requested from outside | |
5700 | if (!extents_out) { | |
5701 | bluefs->add_block_extent(bluefs_shared_bdev, e.offset, e.length); | |
5702 | } | |
7c673cae FG |
5703 | } |
5704 | } | |
7c673cae FG |
5705 | return 0; |
5706 | } | |
5707 | ||
11fdf7f2 | 5708 | int64_t BlueStore::_get_bluefs_size_delta(uint64_t bluefs_free, uint64_t bluefs_total) |
f64942e4 | 5709 | { |
7c673cae FG |
5710 | float bluefs_free_ratio = (float)bluefs_free / (float)bluefs_total; |
5711 | ||
5712 | uint64_t my_free = alloc->get_free(); | |
5713 | uint64_t total = bdev->get_size(); | |
5714 | float my_free_ratio = (float)my_free / (float)total; | |
5715 | ||
5716 | uint64_t total_free = bluefs_free + my_free; | |
5717 | ||
5718 | float bluefs_ratio = (float)bluefs_free / (float)total_free; | |
5719 | ||
5720 | dout(10) << __func__ | |
1adf2230 | 5721 | << " bluefs " << byte_u_t(bluefs_free) |
7c673cae | 5722 | << " free (" << bluefs_free_ratio |
1adf2230 | 5723 | << ") bluestore " << byte_u_t(my_free) |
7c673cae FG |
5724 | << " free (" << my_free_ratio |
5725 | << "), bluefs_ratio " << bluefs_ratio | |
5726 | << dendl; | |
5727 | ||
5728 | uint64_t gift = 0; | |
5729 | uint64_t reclaim = 0; | |
5730 | if (bluefs_ratio < cct->_conf->bluestore_bluefs_min_ratio) { | |
5731 | gift = cct->_conf->bluestore_bluefs_gift_ratio * total_free; | |
5732 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio | |
5733 | << " < min_ratio " << cct->_conf->bluestore_bluefs_min_ratio | |
1adf2230 | 5734 | << ", should gift " << byte_u_t(gift) << dendl; |
7c673cae FG |
5735 | } else if (bluefs_ratio > cct->_conf->bluestore_bluefs_max_ratio) { |
5736 | reclaim = cct->_conf->bluestore_bluefs_reclaim_ratio * total_free; | |
5737 | if (bluefs_total - reclaim < cct->_conf->bluestore_bluefs_min) | |
5738 | reclaim = bluefs_total - cct->_conf->bluestore_bluefs_min; | |
5739 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio | |
5740 | << " > max_ratio " << cct->_conf->bluestore_bluefs_max_ratio | |
1adf2230 | 5741 | << ", should reclaim " << byte_u_t(reclaim) << dendl; |
7c673cae | 5742 | } |
3efd9988 FG |
5743 | |
5744 | // don't take over too much of the freespace | |
5745 | uint64_t free_cap = cct->_conf->bluestore_bluefs_max_ratio * total_free; | |
7c673cae | 5746 | if (bluefs_total < cct->_conf->bluestore_bluefs_min && |
3efd9988 | 5747 | cct->_conf->bluestore_bluefs_min < free_cap) { |
7c673cae FG |
5748 | uint64_t g = cct->_conf->bluestore_bluefs_min - bluefs_total; |
5749 | dout(10) << __func__ << " bluefs_total " << bluefs_total | |
5750 | << " < min " << cct->_conf->bluestore_bluefs_min | |
1adf2230 | 5751 | << ", should gift " << byte_u_t(g) << dendl; |
7c673cae FG |
5752 | if (g > gift) |
5753 | gift = g; | |
5754 | reclaim = 0; | |
5755 | } | |
11fdf7f2 | 5756 | uint64_t min_free = cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); |
3efd9988 FG |
5757 | if (bluefs_free < min_free && |
5758 | min_free < free_cap) { | |
5759 | uint64_t g = min_free - bluefs_free; | |
11fdf7f2 | 5760 | dout(10) << __func__ << " bluefs_free " << bluefs_free |
3efd9988 | 5761 | << " < min " << min_free |
1adf2230 | 5762 | << ", should gift " << byte_u_t(g) << dendl; |
3efd9988 FG |
5763 | if (g > gift) |
5764 | gift = g; | |
5765 | reclaim = 0; | |
5766 | } | |
11fdf7f2 TL |
5767 | ceph_assert((int64_t)gift >= 0); |
5768 | ceph_assert((int64_t)reclaim >= 0); | |
5769 | return gift > 0 ? (int64_t)gift : -(int64_t)reclaim; | |
5770 | } | |
7c673cae | 5771 | |
11fdf7f2 TL |
5772 | int BlueStore::_balance_bluefs_freespace() |
5773 | { | |
5774 | int ret = 0; | |
5775 | ceph_assert(bluefs); | |
7c673cae | 5776 | |
11fdf7f2 TL |
5777 | vector<pair<uint64_t,uint64_t>> bluefs_usage; // <free, total> ... |
5778 | bluefs->get_usage(&bluefs_usage); | |
5779 | ceph_assert(bluefs_usage.size() > bluefs_shared_bdev); | |
7c673cae | 5780 | |
11fdf7f2 TL |
5781 | bool clear_alert = true; |
5782 | if (bluefs_shared_bdev == BlueFS::BDEV_SLOW) { | |
5783 | auto& p = bluefs_usage[bluefs_shared_bdev]; | |
5784 | if (p.first != p.second) { | |
5785 | auto& db = bluefs_usage[BlueFS::BDEV_DB]; | |
5786 | ostringstream ss; | |
5787 | ss << "spilled over " << byte_u_t(p.second - p.first) | |
5788 | << " metadata from 'db' device (" << byte_u_t(db.second - db.first) | |
5789 | << " used of " << byte_u_t(db.second) << ") to slow device"; | |
5790 | _set_spillover_alert(ss.str()); | |
5791 | clear_alert = false; | |
5792 | } | |
5793 | } | |
5794 | if (clear_alert) { | |
5795 | _clear_spillover_alert(); | |
7c673cae FG |
5796 | } |
5797 | ||
11fdf7f2 TL |
5798 | // fixme: look at primary bdev only for now |
5799 | int64_t delta = _get_bluefs_size_delta( | |
5800 | bluefs_usage[bluefs_shared_bdev].first, | |
5801 | bluefs_usage[bluefs_shared_bdev].second); | |
5802 | ||
7c673cae | 5803 | // reclaim from bluefs? |
11fdf7f2 | 5804 | if (delta < 0) { |
7c673cae | 5805 | // round up to alloc size |
11fdf7f2 | 5806 | auto reclaim = p2roundup(uint64_t(-delta), cct->_conf->bluefs_alloc_size); |
7c673cae FG |
5807 | |
5808 | // hard cap to fit into 32 bits | |
11fdf7f2 | 5809 | reclaim = std::min<uint64_t>(reclaim, 1ull << 31); |
7c673cae | 5810 | dout(10) << __func__ << " reclaiming " << reclaim |
1adf2230 | 5811 | << " (" << byte_u_t(reclaim) << ")" << dendl; |
7c673cae FG |
5812 | |
5813 | while (reclaim > 0) { | |
5814 | // NOTE: this will block and do IO. | |
a8e16298 | 5815 | PExtentVector extents; |
7c673cae FG |
5816 | int r = bluefs->reclaim_blocks(bluefs_shared_bdev, reclaim, |
5817 | &extents); | |
5818 | if (r < 0) { | |
5819 | derr << __func__ << " failed to reclaim space from bluefs" | |
5820 | << dendl; | |
5821 | break; | |
5822 | } | |
5823 | for (auto e : extents) { | |
11fdf7f2 | 5824 | ++out_of_sync_fm; |
7c673cae FG |
5825 | bluefs_extents.erase(e.offset, e.length); |
5826 | bluefs_extents_reclaiming.insert(e.offset, e.length); | |
5827 | reclaim -= e.length; | |
5828 | } | |
5829 | } | |
5830 | ||
5831 | ret = 1; | |
5832 | } | |
5833 | ||
5834 | return ret; | |
5835 | } | |
5836 | ||
7c673cae FG |
5837 | int BlueStore::_open_collections(int *errors) |
5838 | { | |
28e407b8 | 5839 | dout(10) << __func__ << dendl; |
11fdf7f2 | 5840 | ceph_assert(coll_map.empty()); |
7c673cae FG |
5841 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
5842 | for (it->upper_bound(string()); | |
5843 | it->valid(); | |
5844 | it->next()) { | |
5845 | coll_t cid; | |
5846 | if (cid.parse(it->key())) { | |
5847 | CollectionRef c( | |
5848 | new Collection( | |
5849 | this, | |
5850 | cache_shards[cid.hash_to_shard(cache_shards.size())], | |
5851 | cid)); | |
5852 | bufferlist bl = it->value(); | |
11fdf7f2 | 5853 | auto p = bl.cbegin(); |
7c673cae | 5854 | try { |
11fdf7f2 | 5855 | decode(c->cnode, p); |
7c673cae FG |
5856 | } catch (buffer::error& e) { |
5857 | derr << __func__ << " failed to decode cnode, key:" | |
5858 | << pretty_binary_string(it->key()) << dendl; | |
5859 | return -EIO; | |
5860 | } | |
28e407b8 AA |
5861 | dout(20) << __func__ << " opened " << cid << " " << c |
5862 | << " " << c->cnode << dendl; | |
11fdf7f2 | 5863 | _osr_attach(c.get()); |
7c673cae | 5864 | coll_map[cid] = c; |
11fdf7f2 | 5865 | |
7c673cae FG |
5866 | } else { |
5867 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
5868 | if (errors) | |
5869 | (*errors)++; | |
5870 | } | |
5871 | } | |
5872 | return 0; | |
5873 | } | |
5874 | ||
224ce89b | 5875 | void BlueStore::_open_statfs() |
31f18b77 | 5876 | { |
11fdf7f2 TL |
5877 | osd_pools.clear(); |
5878 | vstatfs.reset(); | |
5879 | ||
31f18b77 | 5880 | bufferlist bl; |
11fdf7f2 | 5881 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 5882 | if (r >= 0) { |
11fdf7f2 | 5883 | per_pool_stat_collection = false; |
31f18b77 | 5884 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 5885 | auto it = bl.cbegin(); |
31f18b77 | 5886 | vstatfs.decode(it); |
11fdf7f2 | 5887 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 5888 | } else { |
31f18b77 FG |
5889 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
5890 | } | |
81eedcae | 5891 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
5892 | } else if (cct->_conf->bluestore_no_per_pool_stats_tolerance == "enforce") { |
5893 | per_pool_stat_collection = false; | |
5894 | dout(10) << __func__ << " store_statfs is requested but missing, using empty" << dendl; | |
5895 | } else { | |
5896 | per_pool_stat_collection = true; | |
5897 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
5898 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT); | |
5899 | for (it->upper_bound(string()); | |
5900 | it->valid(); | |
5901 | it->next()) { | |
5902 | ||
5903 | uint64_t pool_id; | |
5904 | int r = get_key_pool_stat(it->key(), &pool_id); | |
5905 | ceph_assert(r == 0); | |
5906 | ||
5907 | bufferlist bl; | |
5908 | bl = it->value(); | |
5909 | auto p = bl.cbegin(); | |
5910 | auto& st = osd_pools[pool_id]; | |
5911 | try { | |
5912 | st.decode(p); | |
5913 | vstatfs += st; | |
5914 | ||
5915 | dout(30) << __func__ << " pool " << pool_id | |
5916 | << " statfs " << st << dendl; | |
5917 | } catch (buffer::error& e) { | |
5918 | derr << __func__ << " failed to decode pool stats, key:" | |
5919 | << pretty_binary_string(it->key()) << dendl; | |
5920 | } | |
5921 | } | |
31f18b77 | 5922 | } |
11fdf7f2 TL |
5923 | dout(30) << __func__ << " statfs " << vstatfs << dendl; |
5924 | ||
31f18b77 FG |
5925 | } |
5926 | ||
7c673cae FG |
5927 | int BlueStore::_setup_block_symlink_or_file( |
5928 | string name, | |
5929 | string epath, | |
5930 | uint64_t size, | |
5931 | bool create) | |
5932 | { | |
5933 | dout(20) << __func__ << " name " << name << " path " << epath | |
5934 | << " size " << size << " create=" << (int)create << dendl; | |
5935 | int r = 0; | |
91327a77 | 5936 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
5937 | if (create) |
5938 | flags |= O_CREAT; | |
5939 | if (epath.length()) { | |
5940 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
5941 | if (r < 0) { | |
5942 | r = -errno; | |
5943 | derr << __func__ << " failed to create " << name << " symlink to " | |
5944 | << epath << ": " << cpp_strerror(r) << dendl; | |
5945 | return r; | |
5946 | } | |
5947 | ||
5948 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
5949 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
5950 | if (fd < 0) { | |
5951 | r = -errno; | |
5952 | derr << __func__ << " failed to open " << epath << " file: " | |
5953 | << cpp_strerror(r) << dendl; | |
5954 | return r; | |
5955 | } | |
11fdf7f2 TL |
5956 | // write the Transport ID of the NVMe device |
5957 | // a transport id looks like: "trtype:PCIe traddr:0000:02:00.0" | |
5958 | // where "0000:02:00.0" is the selector of a PCI device, see | |
5959 | // the first column of "lspci -mm -n -D" | |
5960 | string trid{"trtype:PCIe "}; | |
5961 | trid += "traddr:"; | |
5962 | trid += epath.substr(strlen(SPDK_PREFIX)); | |
5963 | r = ::write(fd, trid.c_str(), trid.size()); | |
5964 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
5965 | dout(1) << __func__ << " created " << name << " symlink to " |
5966 | << epath << dendl; | |
5967 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5968 | } | |
5969 | } | |
5970 | if (size) { | |
5971 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
5972 | if (fd >= 0) { | |
5973 | // block file is present | |
5974 | struct stat st; | |
5975 | int r = ::fstat(fd, &st); | |
5976 | if (r == 0 && | |
5977 | S_ISREG(st.st_mode) && // if it is a regular file | |
5978 | st.st_size == 0) { // and is 0 bytes | |
5979 | r = ::ftruncate(fd, size); | |
5980 | if (r < 0) { | |
5981 | r = -errno; | |
5982 | derr << __func__ << " failed to resize " << name << " file to " | |
5983 | << size << ": " << cpp_strerror(r) << dendl; | |
5984 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5985 | return r; | |
5986 | } | |
5987 | ||
5988 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
5989 | r = ::ceph_posix_fallocate(fd, 0, size); |
5990 | if (r > 0) { | |
7c673cae FG |
5991 | derr << __func__ << " failed to prefallocate " << name << " file to " |
5992 | << size << ": " << cpp_strerror(r) << dendl; | |
5993 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
5994 | return -r; | |
5995 | } | |
7c673cae FG |
5996 | } |
5997 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 5998 | << byte_u_t(size) << dendl; |
7c673cae FG |
5999 | } |
6000 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6001 | } else { | |
6002 | int r = -errno; | |
6003 | if (r != -ENOENT) { | |
6004 | derr << __func__ << " failed to open " << name << " file: " | |
6005 | << cpp_strerror(r) << dendl; | |
6006 | return r; | |
6007 | } | |
6008 | } | |
6009 | } | |
6010 | return 0; | |
6011 | } | |
6012 | ||
6013 | int BlueStore::mkfs() | |
6014 | { | |
6015 | dout(1) << __func__ << " path " << path << dendl; | |
6016 | int r; | |
6017 | uuid_d old_fsid; | |
6018 | ||
6019 | { | |
6020 | string done; | |
6021 | r = read_meta("mkfs_done", &done); | |
6022 | if (r == 0) { | |
6023 | dout(1) << __func__ << " already created" << dendl; | |
6024 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
6025 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6026 | if (r < 0) { | |
6027 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
6028 | << dendl; | |
6029 | return r; | |
6030 | } | |
6031 | if (r > 0) { | |
6032 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
6033 | r = -EIO; | |
6034 | } | |
6035 | } | |
6036 | return r; // idempotent | |
6037 | } | |
6038 | } | |
6039 | ||
6040 | { | |
6041 | string type; | |
6042 | r = read_meta("type", &type); | |
6043 | if (r == 0) { | |
6044 | if (type != "bluestore") { | |
6045 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6046 | return -EIO; | |
6047 | } | |
6048 | } else { | |
6049 | r = write_meta("type", "bluestore"); | |
6050 | if (r < 0) | |
6051 | return r; | |
6052 | } | |
6053 | } | |
6054 | ||
6055 | freelist_type = "bitmap"; | |
6056 | ||
6057 | r = _open_path(); | |
6058 | if (r < 0) | |
6059 | return r; | |
6060 | ||
6061 | r = _open_fsid(true); | |
6062 | if (r < 0) | |
6063 | goto out_path_fd; | |
6064 | ||
6065 | r = _lock_fsid(); | |
6066 | if (r < 0) | |
6067 | goto out_close_fsid; | |
6068 | ||
6069 | r = _read_fsid(&old_fsid); | |
6070 | if (r < 0 || old_fsid.is_zero()) { | |
6071 | if (fsid.is_zero()) { | |
6072 | fsid.generate_random(); | |
6073 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
6074 | } else { | |
6075 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
6076 | } | |
6077 | // we'll write it later. | |
6078 | } else { | |
6079 | if (!fsid.is_zero() && fsid != old_fsid) { | |
6080 | derr << __func__ << " on-disk fsid " << old_fsid | |
6081 | << " != provided " << fsid << dendl; | |
6082 | r = -EINVAL; | |
6083 | goto out_close_fsid; | |
6084 | } | |
6085 | fsid = old_fsid; | |
6086 | } | |
6087 | ||
6088 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
6089 | cct->_conf->bluestore_block_size, | |
6090 | cct->_conf->bluestore_block_create); | |
6091 | if (r < 0) | |
6092 | goto out_close_fsid; | |
6093 | if (cct->_conf->bluestore_bluefs) { | |
6094 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
6095 | cct->_conf->bluestore_block_wal_size, | |
6096 | cct->_conf->bluestore_block_wal_create); | |
6097 | if (r < 0) | |
6098 | goto out_close_fsid; | |
6099 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
6100 | cct->_conf->bluestore_block_db_size, | |
6101 | cct->_conf->bluestore_block_db_create); | |
6102 | if (r < 0) | |
6103 | goto out_close_fsid; | |
6104 | } | |
6105 | ||
6106 | r = _open_bdev(true); | |
6107 | if (r < 0) | |
6108 | goto out_close_fsid; | |
6109 | ||
3efd9988 FG |
6110 | // choose min_alloc_size |
6111 | if (cct->_conf->bluestore_min_alloc_size) { | |
6112 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; | |
6113 | } else { | |
11fdf7f2 | 6114 | ceph_assert(bdev); |
3efd9988 FG |
6115 | if (bdev->is_rotational()) { |
6116 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; | |
6117 | } else { | |
6118 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
6119 | } | |
6120 | } | |
11fdf7f2 | 6121 | _validate_bdev(); |
3efd9988 FG |
6122 | |
6123 | // make sure min_alloc_size is power of 2 aligned. | |
11fdf7f2 | 6124 | if (!isp2(min_alloc_size)) { |
3efd9988 FG |
6125 | derr << __func__ << " min_alloc_size 0x" |
6126 | << std::hex << min_alloc_size << std::dec | |
6127 | << " is not power of 2 aligned!" | |
6128 | << dendl; | |
6129 | r = -EINVAL; | |
6130 | goto out_close_bdev; | |
6131 | } | |
6132 | ||
7c673cae FG |
6133 | r = _open_db(true); |
6134 | if (r < 0) | |
6135 | goto out_close_bdev; | |
6136 | ||
7c673cae FG |
6137 | { |
6138 | KeyValueDB::Transaction t = db->get_transaction(); | |
11fdf7f2 TL |
6139 | r = _open_fm(t); |
6140 | if (r < 0) | |
6141 | goto out_close_db; | |
7c673cae FG |
6142 | { |
6143 | bufferlist bl; | |
11fdf7f2 | 6144 | encode((uint64_t)0, bl); |
7c673cae FG |
6145 | t->set(PREFIX_SUPER, "nid_max", bl); |
6146 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
6147 | } | |
6148 | ||
7c673cae FG |
6149 | { |
6150 | bufferlist bl; | |
11fdf7f2 | 6151 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
6152 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
6153 | } | |
6154 | ||
6155 | ondisk_format = latest_ondisk_format; | |
6156 | _prepare_ondisk_format_super(t); | |
6157 | db->submit_transaction_sync(t); | |
6158 | } | |
6159 | ||
7c673cae FG |
6160 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
6161 | if (r < 0) | |
224ce89b WB |
6162 | goto out_close_fm; |
6163 | ||
3efd9988 | 6164 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 6165 | if (r < 0) |
224ce89b | 6166 | goto out_close_fm; |
7c673cae FG |
6167 | |
6168 | if (fsid != old_fsid) { | |
6169 | r = _write_fsid(); | |
6170 | if (r < 0) { | |
6171 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 6172 | goto out_close_fm; |
7c673cae FG |
6173 | } |
6174 | } | |
6175 | ||
11fdf7f2 TL |
6176 | if (out_of_sync_fm.fetch_and(0)) { |
6177 | _sync_bluefs_and_fm(); | |
6178 | } | |
6179 | ||
7c673cae FG |
6180 | out_close_fm: |
6181 | _close_fm(); | |
6182 | out_close_db: | |
6183 | _close_db(); | |
6184 | out_close_bdev: | |
6185 | _close_bdev(); | |
6186 | out_close_fsid: | |
6187 | _close_fsid(); | |
6188 | out_path_fd: | |
6189 | _close_path(); | |
6190 | ||
6191 | if (r == 0 && | |
6192 | cct->_conf->bluestore_fsck_on_mkfs) { | |
6193 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6194 | if (rc < 0) | |
6195 | return rc; | |
6196 | if (rc > 0) { | |
6197 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6198 | r = -EIO; | |
6199 | } | |
11fdf7f2 TL |
6200 | } |
6201 | ||
6202 | if (r == 0) { | |
6203 | // indicate success by writing the 'mkfs_done' file | |
6204 | r = write_meta("mkfs_done", "yes"); | |
6205 | } | |
6206 | ||
6207 | if (r < 0) { | |
6208 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6209 | } else { | |
6210 | dout(0) << __func__ << " success" << dendl; | |
6211 | } | |
6212 | return r; | |
6213 | } | |
6214 | ||
6215 | int BlueStore::_mount_for_bluefs() | |
6216 | { | |
6217 | int r = _open_path(); | |
6218 | ceph_assert(r == 0); | |
6219 | r = _open_fsid(false); | |
6220 | ceph_assert(r == 0); | |
6221 | r = _read_fsid(&fsid); | |
6222 | ceph_assert(r == 0); | |
6223 | r = _lock_fsid(); | |
6224 | ceph_assert(r == 0); | |
6225 | r = _open_bluefs(false); | |
6226 | ceph_assert(r == 0); | |
6227 | return r; | |
6228 | } | |
6229 | ||
6230 | void BlueStore::_umount_for_bluefs() | |
6231 | { | |
6232 | _close_bluefs(); | |
6233 | _close_fsid(); | |
6234 | _close_path(); | |
6235 | } | |
6236 | ||
6237 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) | |
6238 | { | |
6239 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6240 | int r; | |
6241 | ceph_assert(path_fd < 0); | |
6242 | ||
6243 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6244 | ||
6245 | if (!cct->_conf->bluestore_bluefs) { | |
6246 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6247 | return -EIO; | |
6248 | } | |
6249 | ||
6250 | r = _mount_for_bluefs(); | |
6251 | ||
6252 | int reserved = 0; | |
6253 | if (id == BlueFS::BDEV_NEWWAL) { | |
6254 | string p = path + "/block.wal"; | |
6255 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
6256 | cct->_conf->bluestore_block_wal_size, | |
6257 | true); | |
6258 | ceph_assert(r == 0); | |
6259 | ||
6260 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
6261 | cct->_conf->bdev_enable_discard); | |
6262 | ceph_assert(r == 0); | |
6263 | ||
6264 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6265 | r = _check_or_set_bdev_label( | |
6266 | p, | |
6267 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6268 | "bluefs wal", | |
6269 | true); | |
6270 | ceph_assert(r == 0); | |
6271 | } | |
6272 | ||
6273 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
6274 | } else if (id == BlueFS::BDEV_NEWDB) { | |
6275 | string p = path + "/block.db"; | |
6276 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
6277 | cct->_conf->bluestore_block_db_size, | |
6278 | true); | |
6279 | ceph_assert(r == 0); | |
6280 | ||
6281 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
6282 | cct->_conf->bdev_enable_discard); | |
6283 | ceph_assert(r == 0); | |
6284 | ||
6285 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6286 | r = _check_or_set_bdev_label( | |
6287 | p, | |
6288 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6289 | "bluefs db", | |
6290 | true); | |
6291 | ceph_assert(r == 0); | |
6292 | } | |
6293 | reserved = SUPER_RESERVED; | |
6294 | } | |
6295 | ||
6296 | bluefs->umount(); | |
6297 | bluefs->mount(); | |
6298 | ||
6299 | bluefs->add_block_extent( | |
6300 | id, | |
6301 | reserved, | |
6302 | bluefs->get_block_device_size(id) - reserved); | |
6303 | ||
6304 | r = bluefs->prepare_new_device(id); | |
6305 | ceph_assert(r == 0); | |
6306 | ||
6307 | if (r < 0) { | |
6308 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6309 | } else { | |
6310 | dout(0) << __func__ << " success" << dendl; | |
6311 | } | |
6312 | ||
6313 | _umount_for_bluefs(); | |
6314 | return r; | |
6315 | } | |
6316 | ||
6317 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
6318 | int id) | |
6319 | { | |
6320 | dout(10) << __func__ << " id:" << id << dendl; | |
6321 | ceph_assert(path_fd < 0); | |
6322 | ||
6323 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
6324 | ||
6325 | if (!cct->_conf->bluestore_bluefs) { | |
6326 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6327 | return -EIO; | |
6328 | } | |
6329 | ||
6330 | int r = _mount_for_bluefs(); | |
6331 | ||
6332 | // require bluestore_bluefs_min_free to be free at target device! | |
6333 | uint64_t used_space = cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
6334 | for(auto src_id : devs_source) { | |
6335 | used_space += bluefs->get_total(src_id) - bluefs->get_free(src_id); | |
6336 | } | |
6337 | uint64_t target_free = bluefs->get_free(id); | |
6338 | if (id == BlueFS::BDEV_SLOW && target_free < used_space) { | |
6339 | // will need to remount full BlueStore instance to allocate more space | |
6340 | _umount_for_bluefs(); | |
6341 | ||
6342 | r = mount(); | |
6343 | ceph_assert(r == 0); | |
6344 | dout(1) << __func__ | |
6345 | << " Allocating more space at slow device for BlueFS: +" | |
6346 | << used_space - target_free << " bytes" << dendl; | |
6347 | r = allocate_bluefs_freespace( | |
6348 | used_space - target_free, | |
6349 | used_space - target_free, | |
6350 | nullptr); | |
6351 | ||
6352 | umount(); | |
6353 | if (r != 0) { | |
6354 | derr << __func__ | |
6355 | << " can't migrate, unable to allocate extra space: " | |
6356 | << used_space - target_free << " at target:" << id | |
6357 | << dendl; | |
6358 | return -ENOSPC; | |
6359 | } | |
6360 | ||
6361 | r = _mount_for_bluefs(); | |
6362 | ceph_assert(r == 0); | |
6363 | } else if (target_free < used_space) { | |
6364 | derr << __func__ | |
6365 | << " can't migrate, free space at target: " << target_free | |
6366 | << " is less than required space: " << used_space | |
6367 | << dendl; | |
6368 | return -ENOSPC; | |
6369 | } | |
6370 | r = bluefs->device_migrate_to_existing(cct, devs_source, id); | |
6371 | if (r < 0) { | |
6372 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
6373 | goto shutdown; | |
6374 | } | |
6375 | ||
6376 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
6377 | r = unlink(string(path + "/block.db").c_str()); | |
6378 | ceph_assert(r == 0); | |
6379 | } | |
6380 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6381 | r = unlink(string(path + "/block.wal").c_str()); | |
6382 | ceph_assert(r == 0); | |
6383 | } | |
6384 | ||
6385 | shutdown: | |
6386 | _umount_for_bluefs(); | |
6387 | return r; | |
6388 | } | |
6389 | ||
6390 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
6391 | int id, | |
6392 | const string& dev_path) | |
6393 | { | |
6394 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6395 | int r; | |
6396 | ceph_assert(path_fd < 0); | |
6397 | ||
6398 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6399 | ||
6400 | if (!cct->_conf->bluestore_bluefs) { | |
6401 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6402 | return -EIO; | |
6403 | } | |
6404 | ||
6405 | r = _mount_for_bluefs(); | |
6406 | ||
6407 | int reserved = 0; | |
6408 | string link_db; | |
6409 | string link_wal; | |
6410 | if (devs_source.count(BlueFS::BDEV_DB) && | |
6411 | bluefs_shared_bdev != BlueFS::BDEV_DB) { | |
6412 | link_db = path + "/block.db"; | |
6413 | } | |
6414 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6415 | link_wal = path + "/block.wal"; | |
6416 | } | |
6417 | ||
6418 | size_t target_size; | |
6419 | string target_name; | |
6420 | if (id == BlueFS::BDEV_NEWWAL) { | |
6421 | target_name = "block.wal"; | |
6422 | target_size = cct->_conf->bluestore_block_wal_size; | |
6423 | ||
6424 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
6425 | cct->_conf->bdev_enable_discard); | |
6426 | ceph_assert(r == 0); | |
6427 | ||
6428 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6429 | r = _check_or_set_bdev_label( | |
6430 | dev_path, | |
6431 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6432 | "bluefs wal", | |
6433 | true); | |
6434 | ceph_assert(r == 0); | |
6435 | } | |
6436 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
6437 | } else if (id == BlueFS::BDEV_NEWDB) { | |
6438 | target_name = "block.db"; | |
6439 | target_size = cct->_conf->bluestore_block_db_size; | |
31f18b77 | 6440 | |
11fdf7f2 TL |
6441 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
6442 | cct->_conf->bdev_enable_discard); | |
6443 | ceph_assert(r == 0); | |
6444 | ||
6445 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6446 | r = _check_or_set_bdev_label( | |
6447 | dev_path, | |
6448 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6449 | "bluefs db", | |
6450 | true); | |
6451 | ceph_assert(r == 0); | |
6452 | } | |
6453 | reserved = SUPER_RESERVED; | |
31f18b77 FG |
6454 | } |
6455 | ||
11fdf7f2 TL |
6456 | bluefs->umount(); |
6457 | bluefs->mount(); | |
6458 | ||
6459 | bluefs->add_block_extent( | |
6460 | id, reserved, bluefs->get_block_device_size(id) - reserved); | |
6461 | ||
6462 | r = bluefs->device_migrate_to_new(cct, devs_source, id); | |
6463 | ||
7c673cae | 6464 | if (r < 0) { |
11fdf7f2 TL |
6465 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
6466 | goto shutdown; | |
6467 | } | |
6468 | ||
6469 | if (!link_db.empty()) { | |
6470 | r = unlink(link_db.c_str()); | |
6471 | ceph_assert(r == 0); | |
6472 | } | |
6473 | if (!link_wal.empty()) { | |
6474 | r = unlink(link_wal.c_str()); | |
6475 | ceph_assert(r == 0); | |
6476 | } | |
6477 | r = _setup_block_symlink_or_file( | |
6478 | target_name, | |
6479 | dev_path, | |
6480 | target_size, | |
6481 | true); | |
6482 | ceph_assert(r == 0); | |
6483 | dout(0) << __func__ << " success" << dendl; | |
6484 | ||
6485 | shutdown: | |
6486 | _umount_for_bluefs(); | |
6487 | return r; | |
6488 | } | |
6489 | ||
6490 | string BlueStore::get_device_path(unsigned id) | |
6491 | { | |
6492 | string res; | |
6493 | if (id < BlueFS::MAX_BDEV) { | |
6494 | switch (id) { | |
6495 | case BlueFS::BDEV_WAL: | |
6496 | res = path + "/block.wal"; | |
6497 | break; | |
6498 | case BlueFS::BDEV_DB: | |
6499 | if (id == bluefs_shared_bdev) { | |
6500 | res = path + "/block"; | |
6501 | } else { | |
6502 | res = path + "/block.db"; | |
6503 | } | |
6504 | break; | |
6505 | case BlueFS::BDEV_SLOW: | |
6506 | res = path + "/block"; | |
6507 | break; | |
6508 | } | |
6509 | } | |
6510 | return res; | |
6511 | } | |
6512 | ||
6513 | int BlueStore::expand_devices(ostream& out) | |
6514 | { | |
6515 | int r = _mount(false); | |
6516 | ceph_assert(r == 0); | |
6517 | bluefs->dump_block_extents(out); | |
6518 | out << "Expanding..." << std::endl; | |
6519 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { | |
6520 | if (devid == bluefs_shared_bdev ) { | |
6521 | continue; | |
6522 | } | |
6523 | uint64_t size = bluefs->get_block_device_size(devid); | |
6524 | if (size == 0) { | |
6525 | // no bdev | |
6526 | continue; | |
6527 | } | |
6528 | ||
6529 | interval_set<uint64_t> before; | |
6530 | bluefs->get_block_extents(devid, &before); | |
6531 | ceph_assert(!before.empty()); | |
6532 | uint64_t end = before.range_end(); | |
6533 | if (end < size) { | |
6534 | out << devid | |
6535 | <<" : expanding " << " from 0x" << std::hex | |
6536 | << end << " to 0x" << size << std::dec << std::endl; | |
6537 | bluefs->add_block_extent(devid, end, size-end); | |
6538 | string p = get_device_path(devid); | |
6539 | const char* path = p.c_str(); | |
6540 | if (path == nullptr) { | |
6541 | derr << devid | |
6542 | <<": can't find device path " << dendl; | |
6543 | continue; | |
6544 | } | |
6545 | bluestore_bdev_label_t label; | |
6546 | int r = _read_bdev_label(cct, path, &label); | |
6547 | if (r < 0) { | |
6548 | derr << "unable to read label for " << path << ": " | |
6549 | << cpp_strerror(r) << dendl; | |
6550 | continue; | |
6551 | } | |
6552 | label.size = size; | |
6553 | r = _write_bdev_label(cct, path, label); | |
6554 | if (r < 0) { | |
6555 | derr << "unable to write label for " << path << ": " | |
6556 | << cpp_strerror(r) << dendl; | |
6557 | continue; | |
6558 | } | |
6559 | out << devid | |
6560 | <<" : size label updated to " << size | |
6561 | << std::endl; | |
6562 | } | |
6563 | } | |
6564 | uint64_t size0 = fm->get_size(); | |
6565 | uint64_t size = bdev->get_size(); | |
6566 | if (size0 < size) { | |
6567 | out << bluefs_shared_bdev | |
6568 | <<" : expanding " << " from 0x" << std::hex | |
6569 | << size0 << " to 0x" << size << std::dec << std::endl; | |
6570 | KeyValueDB::Transaction txn; | |
6571 | txn = db->get_transaction(); | |
6572 | int r = fm->expand(size, txn); | |
6573 | ceph_assert(r == 0); | |
6574 | db->submit_transaction_sync(txn); | |
6575 | ||
6576 | // always reference to slow device here | |
6577 | string p = get_device_path(BlueFS::BDEV_SLOW); | |
6578 | ceph_assert(!p.empty()); | |
6579 | const char* path = p.c_str(); | |
6580 | bluestore_bdev_label_t label; | |
6581 | r = _read_bdev_label(cct, path, &label); | |
6582 | if (r < 0) { | |
6583 | derr << "unable to read label for " << path << ": " | |
6584 | << cpp_strerror(r) << dendl; | |
6585 | } else { | |
6586 | label.size = size; | |
6587 | r = _write_bdev_label(cct, path, label); | |
6588 | if (r < 0) { | |
6589 | derr << "unable to write label for " << path << ": " | |
6590 | << cpp_strerror(r) << dendl; | |
6591 | } else { | |
6592 | out << bluefs_shared_bdev | |
6593 | <<" : size label updated to " << size | |
6594 | << std::endl; | |
6595 | } | |
6596 | } | |
7c673cae | 6597 | } |
11fdf7f2 | 6598 | umount(); |
7c673cae FG |
6599 | return r; |
6600 | } | |
6601 | ||
6602 | void BlueStore::set_cache_shards(unsigned num) | |
6603 | { | |
6604 | dout(10) << __func__ << " " << num << dendl; | |
6605 | size_t old = cache_shards.size(); | |
11fdf7f2 | 6606 | ceph_assert(num >= old); |
7c673cae FG |
6607 | cache_shards.resize(num); |
6608 | for (unsigned i = old; i < num; ++i) { | |
6609 | cache_shards[i] = Cache::create(cct, cct->_conf->bluestore_cache_type, | |
6610 | logger); | |
6611 | } | |
6612 | } | |
6613 | ||
11fdf7f2 | 6614 | int BlueStore::_mount(bool kv_only, bool open_db) |
7c673cae FG |
6615 | { |
6616 | dout(1) << __func__ << " path " << path << dendl; | |
6617 | ||
3efd9988 FG |
6618 | _kv_only = kv_only; |
6619 | ||
7c673cae FG |
6620 | { |
6621 | string type; | |
6622 | int r = read_meta("type", &type); | |
6623 | if (r < 0) { | |
6624 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
6625 | << dendl; | |
6626 | return r; | |
6627 | } | |
6628 | ||
6629 | if (type != "bluestore") { | |
6630 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6631 | return -EIO; | |
6632 | } | |
6633 | } | |
6634 | ||
6635 | if (cct->_conf->bluestore_fsck_on_mount) { | |
6636 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); | |
6637 | if (rc < 0) | |
6638 | return rc; | |
6639 | if (rc > 0) { | |
6640 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6641 | return -EIO; | |
6642 | } | |
6643 | } | |
6644 | ||
6645 | int r = _open_path(); | |
6646 | if (r < 0) | |
6647 | return r; | |
6648 | r = _open_fsid(false); | |
6649 | if (r < 0) | |
6650 | goto out_path; | |
6651 | ||
6652 | r = _read_fsid(&fsid); | |
6653 | if (r < 0) | |
6654 | goto out_fsid; | |
6655 | ||
6656 | r = _lock_fsid(); | |
6657 | if (r < 0) | |
6658 | goto out_fsid; | |
6659 | ||
6660 | r = _open_bdev(false); | |
6661 | if (r < 0) | |
6662 | goto out_fsid; | |
6663 | ||
11fdf7f2 TL |
6664 | if (open_db) { |
6665 | r = _open_db_and_around(false); | |
6666 | } else { | |
6667 | // we can bypass db open exclusively in case of kv_only mode | |
6668 | ceph_assert(kv_only); | |
6669 | r = _open_db(false, true); | |
6670 | if (r < 0) | |
6671 | goto out_bdev; | |
6672 | } | |
7c673cae FG |
6673 | |
6674 | if (kv_only) | |
6675 | return 0; | |
6676 | ||
11fdf7f2 TL |
6677 | r = _upgrade_super(); |
6678 | if (r < 0) { | |
7c673cae | 6679 | goto out_db; |
11fdf7f2 | 6680 | } |
7c673cae FG |
6681 | |
6682 | r = _open_collections(); | |
6683 | if (r < 0) | |
11fdf7f2 | 6684 | goto out_db; |
7c673cae FG |
6685 | |
6686 | r = _reload_logger(); | |
6687 | if (r < 0) | |
6688 | goto out_coll; | |
6689 | ||
31f18b77 | 6690 | _kv_start(); |
7c673cae FG |
6691 | |
6692 | r = _deferred_replay(); | |
6693 | if (r < 0) | |
6694 | goto out_stop; | |
6695 | ||
6696 | mempool_thread.init(); | |
6697 | ||
7c673cae FG |
6698 | mounted = true; |
6699 | return 0; | |
6700 | ||
6701 | out_stop: | |
6702 | _kv_stop(); | |
7c673cae | 6703 | out_coll: |
31f18b77 | 6704 | _flush_cache(); |
7c673cae | 6705 | out_db: |
11fdf7f2 | 6706 | _close_db_and_around(); |
7c673cae FG |
6707 | out_bdev: |
6708 | _close_bdev(); | |
6709 | out_fsid: | |
6710 | _close_fsid(); | |
6711 | out_path: | |
6712 | _close_path(); | |
6713 | return r; | |
6714 | } | |
6715 | ||
6716 | int BlueStore::umount() | |
6717 | { | |
11fdf7f2 | 6718 | ceph_assert(_kv_only || mounted); |
7c673cae FG |
6719 | dout(1) << __func__ << dendl; |
6720 | ||
6721 | _osr_drain_all(); | |
7c673cae | 6722 | |
7c673cae | 6723 | mounted = false; |
3efd9988 FG |
6724 | if (!_kv_only) { |
6725 | mempool_thread.shutdown(); | |
6726 | dout(20) << __func__ << " stopping kv thread" << dendl; | |
6727 | _kv_stop(); | |
3efd9988 FG |
6728 | _flush_cache(); |
6729 | dout(20) << __func__ << " closing" << dendl; | |
6730 | ||
3efd9988 | 6731 | } |
11fdf7f2 | 6732 | _close_db_and_around(); |
7c673cae FG |
6733 | _close_bdev(); |
6734 | _close_fsid(); | |
6735 | _close_path(); | |
6736 | ||
6737 | if (cct->_conf->bluestore_fsck_on_umount) { | |
6738 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); | |
6739 | if (rc < 0) | |
6740 | return rc; | |
6741 | if (rc > 0) { | |
6742 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6743 | return -EIO; | |
6744 | } | |
6745 | } | |
6746 | return 0; | |
6747 | } | |
6748 | ||
6749 | static void apply(uint64_t off, | |
6750 | uint64_t len, | |
6751 | uint64_t granularity, | |
6752 | BlueStore::mempool_dynamic_bitset &bitset, | |
7c673cae FG |
6753 | std::function<void(uint64_t, |
6754 | BlueStore::mempool_dynamic_bitset &)> f) { | |
11fdf7f2 | 6755 | auto end = round_up_to(off + len, granularity); |
7c673cae FG |
6756 | while (off < end) { |
6757 | uint64_t pos = off / granularity; | |
6758 | f(pos, bitset); | |
6759 | off += granularity; | |
6760 | } | |
6761 | } | |
6762 | ||
6763 | int BlueStore::_fsck_check_extents( | |
11fdf7f2 | 6764 | const coll_t& cid, |
7c673cae FG |
6765 | const ghobject_t& oid, |
6766 | const PExtentVector& extents, | |
6767 | bool compressed, | |
6768 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 6769 | uint64_t granularity, |
11fdf7f2 | 6770 | BlueStoreRepairer* repairer, |
7c673cae FG |
6771 | store_statfs_t& expected_statfs) |
6772 | { | |
6773 | dout(30) << __func__ << " oid " << oid << " extents " << extents << dendl; | |
6774 | int errors = 0; | |
6775 | for (auto e : extents) { | |
6776 | if (!e.is_valid()) | |
6777 | continue; | |
6778 | expected_statfs.allocated += e.length; | |
6779 | if (compressed) { | |
11fdf7f2 | 6780 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae FG |
6781 | } |
6782 | bool already = false; | |
6783 | apply( | |
b32b8144 | 6784 | e.offset, e.length, granularity, used_blocks, |
7c673cae | 6785 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
11fdf7f2 TL |
6786 | ceph_assert(pos < bs.size()); |
6787 | if (bs.test(pos)) { | |
6788 | if (repairer) { | |
6789 | repairer->note_misreference( | |
6790 | pos * min_alloc_size, min_alloc_size, !already); | |
6791 | } | |
6792 | if (!already) { | |
6793 | derr << "fsck error: " << oid << " extent " << e | |
6794 | << " or a subset is already allocated (misreferenced)" << dendl; | |
6795 | ++errors; | |
6796 | already = true; | |
6797 | } | |
6798 | } | |
7c673cae FG |
6799 | else |
6800 | bs.set(pos); | |
6801 | }); | |
11fdf7f2 TL |
6802 | if (repairer) { |
6803 | repairer->get_space_usage_tracker().set_used( e.offset, e.length, cid, oid); | |
6804 | } | |
6805 | ||
7c673cae | 6806 | if (e.end() > bdev->get_size()) { |
11fdf7f2 | 6807 | derr << "fsck error: " << oid << " extent " << e |
7c673cae FG |
6808 | << " past end of block device" << dendl; |
6809 | ++errors; | |
6810 | } | |
6811 | } | |
6812 | return errors; | |
6813 | } | |
6814 | ||
11fdf7f2 TL |
6815 | void BlueStore::_fsck_check_pool_statfs( |
6816 | BlueStore::per_pool_statfs& expected_pool_statfs, | |
6817 | bool need_per_pool_stats, | |
6818 | int& errors, | |
6819 | BlueStoreRepairer* repairer) | |
6820 | { | |
6821 | auto it = db->get_iterator(PREFIX_STAT); | |
6822 | if (it) { | |
6823 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
6824 | string key = it->key(); | |
6825 | if (key == BLUESTORE_GLOBAL_STATFS_KEY) { | |
6826 | if (repairer) { | |
6827 | if (need_per_pool_stats) { | |
6828 | ++errors; | |
6829 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
6830 | derr << "fsck error: " << "legacy statfs record found, removing" << dendl; | |
6831 | } else { | |
6832 | derr << "fsck warning: " << "legacy statfs record found, bypassing" << dendl; | |
6833 | } | |
6834 | } else { | |
6835 | const char* s = "fsck warning: "; | |
6836 | if (need_per_pool_stats) { | |
6837 | ++errors; | |
6838 | s = "fsck error: "; | |
6839 | } | |
6840 | derr << s << "legacy statfs record found, suggest to " | |
6841 | "run store repair to get consistent statistic reports" | |
6842 | << dendl; | |
6843 | } | |
6844 | continue; | |
6845 | } | |
6846 | if (!need_per_pool_stats) { | |
6847 | continue; | |
6848 | } | |
6849 | uint64_t pool_id; | |
6850 | if (get_key_pool_stat(key, &pool_id) < 0) { | |
6851 | derr << "fsck error: bad key " << key | |
6852 | << "in statfs namespece" << dendl; | |
6853 | if (repairer) { | |
6854 | repairer->remove_key(db, PREFIX_STAT, key); | |
6855 | } | |
6856 | ++errors; | |
6857 | continue; | |
6858 | } | |
6859 | ||
6860 | volatile_statfs vstatfs; | |
6861 | bufferlist bl = it->value(); | |
6862 | auto blp = bl.cbegin(); | |
6863 | try { | |
6864 | vstatfs.decode(blp); | |
6865 | } catch (buffer::error& e) { | |
6866 | derr << "fsck error: failed to decode Pool StatFS record" | |
6867 | << pretty_binary_string(key) << dendl; | |
6868 | if (repairer) { | |
6869 | dout(20) << __func__ << " undecodable Pool StatFS record, key:'" | |
6870 | << pretty_binary_string(key) | |
6871 | << "', removing" << dendl; | |
6872 | repairer->remove_key(db, PREFIX_STAT, key); | |
6873 | } | |
6874 | ++errors; | |
6875 | vstatfs.reset(); | |
6876 | } | |
6877 | auto stat_it = expected_pool_statfs.find(pool_id); | |
6878 | if (stat_it == expected_pool_statfs.end()) { | |
6879 | if (vstatfs.is_empty()) { | |
6880 | // we don't consider that as an error since empty pool statfs | |
6881 | // are left in DB for now | |
6882 | dout(20) << "fsck inf: found empty stray Pool StatFS record for pool id 0x" | |
6883 | << std::hex << pool_id << std::dec << dendl; | |
6884 | if (repairer) { | |
6885 | // but we need to increment error count in case of repair | |
6886 | // to have proper counters at the end | |
6887 | // (as repairer increments recovery counter anyway). | |
6888 | ++errors; | |
6889 | } | |
6890 | } else { | |
6891 | derr << "fsck error: found stray Pool StatFS record for pool id 0x" | |
6892 | << std::hex << pool_id << std::dec << dendl; | |
6893 | ++errors; | |
6894 | } | |
6895 | if (repairer) { | |
6896 | repairer->remove_key(db, PREFIX_SHARED_BLOB, key); | |
6897 | } | |
6898 | continue; | |
6899 | } | |
6900 | store_statfs_t statfs; | |
6901 | vstatfs.publish(&statfs); | |
6902 | if (!(stat_it->second == statfs)) { | |
6903 | derr << "fsck error: actual " << statfs | |
6904 | << " != expected " << stat_it->second | |
6905 | << " for pool " | |
6906 | << std::hex << pool_id << std::dec << dendl; | |
6907 | if (repairer) { | |
6908 | repairer->fix_statfs(db, key, stat_it->second); | |
6909 | } | |
6910 | ++errors; | |
6911 | } | |
6912 | expected_pool_statfs.erase(stat_it); | |
6913 | } | |
6914 | } // if (it) | |
6915 | for( auto s = expected_pool_statfs.begin(); s != expected_pool_statfs.end(); | |
6916 | ++s) { | |
6917 | if (s->second.is_zero()) { | |
6918 | // we might lack empty statfs recs in DB | |
6919 | continue; | |
6920 | } | |
6921 | derr << "fsck error: missing Pool StatFS record for pool " | |
6922 | << std::hex << s->first << std::dec << dendl; | |
6923 | if (repairer) { | |
6924 | string key; | |
6925 | get_pool_stat_key(s->first, &key); | |
6926 | repairer->fix_statfs(db, key, s->second); | |
6927 | } | |
6928 | ++errors; | |
6929 | } | |
6930 | } | |
6931 | ||
6932 | /** | |
6933 | An overview for currently implemented repair logics | |
6934 | performed in fsck in two stages: detection(+preparation) and commit. | |
6935 | Detection stage (in processing order): | |
6936 | (Issue -> Repair action to schedule) | |
6937 | - Detect undecodable keys for Shared Blobs -> Remove | |
6938 | - Detect undecodable records for Shared Blobs -> Remove | |
6939 | (might trigger missed Shared Blob detection below) | |
6940 | - Detect stray records for Shared Blobs -> Remove | |
6941 | - Detect misreferenced pextents -> Fix | |
6942 | Prepare Bloom-like filter to track cid/oid -> pextent | |
6943 | Prepare list of extents that are improperly referenced | |
6944 | Enumerate Onode records that might use 'misreferenced' pextents | |
6945 | (Bloom-like filter applied to reduce computation) | |
6946 | Per each questinable Onode enumerate all blobs and identify broken ones | |
6947 | (i.e. blobs having 'misreferences') | |
6948 | Rewrite each broken blob data by allocating another extents and | |
6949 | copying data there | |
6950 | If blob is shared - unshare it and mark corresponding Shared Blob | |
6951 | for removal | |
6952 | Release previously allocated space | |
6953 | Update Extent Map | |
6954 | - Detect missed Shared Blobs -> Recreate | |
6955 | - Detect undecodable deferred transaction -> Remove | |
6956 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
6957 | - Detect Freelist Manager's leaked entries -> Mark as free | |
6958 | - Detect statfs inconsistency - Update | |
6959 | Commit stage (separate DB commit per each step): | |
6960 | - Apply leaked FM entries fix | |
6961 | - Apply 'false free' FM entries fix | |
6962 | - Apply 'Remove' actions | |
6963 | - Apply fix for misreference pextents | |
6964 | - Apply Shared Blob recreate | |
6965 | (can be merged with the step above if misreferences were dectected) | |
6966 | - Apply StatFS update | |
6967 | */ | |
3efd9988 | 6968 | int BlueStore::_fsck(bool deep, bool repair) |
7c673cae | 6969 | { |
3efd9988 | 6970 | dout(1) << __func__ |
11fdf7f2 TL |
6971 | << " <<<START>>>" |
6972 | << (repair ? " repair" : " check") | |
3efd9988 | 6973 | << (deep ? " (deep)" : " (shallow)") << " start" << dendl; |
7c673cae | 6974 | int errors = 0; |
11fdf7f2 | 6975 | unsigned repaired = 0; |
31f18b77 FG |
6976 | |
6977 | typedef btree::btree_set< | |
6978 | uint64_t,std::less<uint64_t>, | |
6979 | mempool::bluestore_fsck::pool_allocator<uint64_t>> uint64_t_btree_t; | |
6980 | uint64_t_btree_t used_nids; | |
6981 | uint64_t_btree_t used_omap_head; | |
11fdf7f2 | 6982 | uint64_t_btree_t used_pgmeta_omap_head; |
31f18b77 FG |
6983 | uint64_t_btree_t used_sbids; |
6984 | ||
7c673cae | 6985 | mempool_dynamic_bitset used_blocks; |
7c673cae | 6986 | KeyValueDB::Iterator it; |
11fdf7f2 TL |
6987 | store_statfs_t expected_store_statfs, actual_statfs; |
6988 | per_pool_statfs expected_pool_statfs; | |
6989 | ||
7c673cae | 6990 | struct sb_info_t { |
11fdf7f2 TL |
6991 | coll_t cid; |
6992 | int64_t pool_id = INT64_MIN; | |
7c673cae FG |
6993 | list<ghobject_t> oids; |
6994 | SharedBlobRef sb; | |
6995 | bluestore_extent_ref_map_t ref_map; | |
11fdf7f2 TL |
6996 | bool compressed = false; |
6997 | bool passed = false; | |
6998 | bool updated = false; | |
7c673cae FG |
6999 | }; |
7000 | mempool::bluestore_fsck::map<uint64_t,sb_info_t> sb_info; | |
7001 | ||
7002 | uint64_t num_objects = 0; | |
7003 | uint64_t num_extents = 0; | |
7004 | uint64_t num_blobs = 0; | |
7005 | uint64_t num_spanning_blobs = 0; | |
7006 | uint64_t num_shared_blobs = 0; | |
7007 | uint64_t num_sharded_objects = 0; | |
7008 | uint64_t num_object_shards = 0; | |
11fdf7f2 TL |
7009 | BlueStoreRepairer repairer; |
7010 | store_statfs_t* expected_statfs = nullptr; | |
7011 | // in deep mode we need R/W write access to be able to replay deferred ops | |
7012 | bool read_only = !(repair || deep); | |
7c673cae FG |
7013 | |
7014 | utime_t start = ceph_clock_now(); | |
11fdf7f2 TL |
7015 | const auto& no_pps_mode = cct->_conf->bluestore_no_per_pool_stats_tolerance; |
7016 | bool need_per_pool_stats = no_pps_mode == "until_fsck" || | |
7017 | (no_pps_mode == "until_repair" && repair); | |
7018 | bool enforce_no_per_pool_stats = no_pps_mode == "enforce"; | |
7c673cae FG |
7019 | |
7020 | int r = _open_path(); | |
7021 | if (r < 0) | |
7022 | return r; | |
7023 | r = _open_fsid(false); | |
7024 | if (r < 0) | |
7025 | goto out_path; | |
7026 | ||
7027 | r = _read_fsid(&fsid); | |
7028 | if (r < 0) | |
7029 | goto out_fsid; | |
7030 | ||
7031 | r = _lock_fsid(); | |
7032 | if (r < 0) | |
7033 | goto out_fsid; | |
7034 | ||
7035 | r = _open_bdev(false); | |
7036 | if (r < 0) | |
7037 | goto out_fsid; | |
7038 | ||
11fdf7f2 | 7039 | r = _open_db_and_around(read_only); |
7c673cae FG |
7040 | if (r < 0) |
7041 | goto out_bdev; | |
7042 | ||
11fdf7f2 TL |
7043 | if (!read_only) { |
7044 | r = _upgrade_super(); | |
7045 | if (r < 0) { | |
7046 | goto out_db; | |
7047 | } | |
7048 | } | |
7c673cae FG |
7049 | |
7050 | r = _open_collections(&errors); | |
7051 | if (r < 0) | |
11fdf7f2 | 7052 | goto out_db; |
7c673cae FG |
7053 | |
7054 | mempool_thread.init(); | |
7055 | ||
11fdf7f2 TL |
7056 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
7057 | // enable in repair or deep mode modes only | |
7058 | if (!read_only) { | |
7059 | _kv_start(); | |
7060 | r = _deferred_replay(); | |
7061 | _kv_stop(); | |
7062 | } | |
7c673cae FG |
7063 | if (r < 0) |
7064 | goto out_scan; | |
7065 | ||
b32b8144 | 7066 | used_blocks.resize(fm->get_alloc_units()); |
7c673cae | 7067 | apply( |
11fdf7f2 | 7068 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), fm->get_alloc_size(), used_blocks, |
7c673cae | 7069 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
11fdf7f2 | 7070 | ceph_assert(pos < bs.size()); |
7c673cae FG |
7071 | bs.set(pos); |
7072 | } | |
7073 | ); | |
11fdf7f2 TL |
7074 | if (repair) { |
7075 | repairer.get_space_usage_tracker().init( | |
7076 | bdev->get_size(), | |
7077 | min_alloc_size); | |
7078 | } | |
7c673cae FG |
7079 | |
7080 | if (bluefs) { | |
11fdf7f2 TL |
7081 | if( cct->_conf->bluestore_bluefs_db_compatibility) { |
7082 | interval_set<uint64_t> bluefs_extents_db; | |
7083 | bufferlist bl; | |
7084 | db->get(PREFIX_SUPER, "bluefs_extents", &bl); | |
7085 | auto p = bl.cbegin(); | |
7086 | auto prev_errors = errors; | |
7087 | try { | |
7088 | decode(bluefs_extents_db, p); | |
7089 | bluefs_extents_db.union_of(bluefs_extents); | |
7090 | bluefs_extents_db.subtract(bluefs_extents); | |
7091 | if (!bluefs_extents_db.empty()) { | |
7092 | derr << "fsck error: bluefs_extents inconsistency, " | |
7093 | << "downgrade to previous releases might be broken." | |
7094 | << dendl; | |
7095 | ++errors; | |
7096 | } | |
7097 | } | |
7098 | catch (buffer::error& e) { | |
7099 | derr << "fsck error: failed to retrieve bluefs_extents from kv" << dendl; | |
7100 | ++errors; | |
7101 | } | |
7102 | if (errors != prev_errors && repair) { | |
7103 | repairer.fix_bluefs_extents(out_of_sync_fm); | |
7104 | } | |
7105 | } | |
7106 | ||
7c673cae FG |
7107 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { |
7108 | apply( | |
b32b8144 | 7109 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae | 7110 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
11fdf7f2 | 7111 | ceph_assert(pos < bs.size()); |
7c673cae FG |
7112 | bs.set(pos); |
7113 | } | |
7114 | ); | |
7115 | } | |
7116 | r = bluefs->fsck(); | |
7117 | if (r < 0) { | |
7118 | goto out_scan; | |
7119 | } | |
7120 | if (r > 0) | |
7121 | errors += r; | |
7122 | } | |
7123 | ||
11fdf7f2 | 7124 | // get expected statfs; reset unaffected fields to be able to compare |
7c673cae FG |
7125 | // structs |
7126 | statfs(&actual_statfs); | |
11fdf7f2 TL |
7127 | actual_statfs.total = 0; |
7128 | actual_statfs.internally_reserved = 0; | |
7129 | actual_statfs.available = 0; | |
7130 | actual_statfs.internal_metadata = 0; | |
7131 | actual_statfs.omap_allocated = 0; | |
7132 | ||
7133 | need_per_pool_stats = per_pool_stat_collection || need_per_pool_stats; | |
7c673cae FG |
7134 | |
7135 | // walk PREFIX_OBJ | |
7136 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
7137 | it = db->get_iterator(PREFIX_OBJ); | |
7138 | if (it) { | |
11fdf7f2 TL |
7139 | //fill global if not overriden below |
7140 | expected_statfs = &expected_store_statfs; | |
7141 | ||
7c673cae FG |
7142 | CollectionRef c; |
7143 | spg_t pgid; | |
7144 | mempool::bluestore_fsck::list<string> expecting_shards; | |
7145 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
11fdf7f2 | 7146 | if (g_conf()->bluestore_debug_fsck_abort) { |
31f18b77 FG |
7147 | goto out_scan; |
7148 | } | |
11fdf7f2 TL |
7149 | dout(30) << __func__ << " key " |
7150 | << pretty_binary_string(it->key()) << dendl; | |
7c673cae FG |
7151 | if (is_extent_shard_key(it->key())) { |
7152 | while (!expecting_shards.empty() && | |
7153 | expecting_shards.front() < it->key()) { | |
3efd9988 | 7154 | derr << "fsck error: missing shard key " |
7c673cae FG |
7155 | << pretty_binary_string(expecting_shards.front()) |
7156 | << dendl; | |
7157 | ++errors; | |
7158 | expecting_shards.pop_front(); | |
7159 | } | |
7160 | if (!expecting_shards.empty() && | |
7161 | expecting_shards.front() == it->key()) { | |
7162 | // all good | |
7163 | expecting_shards.pop_front(); | |
7164 | continue; | |
7165 | } | |
7166 | ||
7167 | uint32_t offset; | |
7168 | string okey; | |
7169 | get_key_extent_shard(it->key(), &okey, &offset); | |
3efd9988 | 7170 | derr << "fsck error: stray shard 0x" << std::hex << offset |
7c673cae FG |
7171 | << std::dec << dendl; |
7172 | if (expecting_shards.empty()) { | |
3efd9988 | 7173 | derr << "fsck error: " << pretty_binary_string(it->key()) |
7c673cae FG |
7174 | << " is unexpected" << dendl; |
7175 | ++errors; | |
7176 | continue; | |
7177 | } | |
7178 | while (expecting_shards.front() > it->key()) { | |
3efd9988 | 7179 | derr << "fsck error: saw " << pretty_binary_string(it->key()) |
7c673cae | 7180 | << dendl; |
3efd9988 | 7181 | derr << "fsck error: exp " |
7c673cae FG |
7182 | << pretty_binary_string(expecting_shards.front()) << dendl; |
7183 | ++errors; | |
7184 | expecting_shards.pop_front(); | |
7185 | if (expecting_shards.empty()) { | |
7186 | break; | |
7187 | } | |
7188 | } | |
7189 | continue; | |
7190 | } | |
7191 | ||
7192 | ghobject_t oid; | |
7193 | int r = get_key_object(it->key(), &oid); | |
7194 | if (r < 0) { | |
3efd9988 | 7195 | derr << "fsck error: bad object key " |
7c673cae FG |
7196 | << pretty_binary_string(it->key()) << dendl; |
7197 | ++errors; | |
7198 | continue; | |
7199 | } | |
7200 | if (!c || | |
7201 | oid.shard_id != pgid.shard || | |
11fdf7f2 | 7202 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || |
7c673cae FG |
7203 | !c->contains(oid)) { |
7204 | c = nullptr; | |
11fdf7f2 TL |
7205 | for (auto& p : coll_map) { |
7206 | if (p.second->contains(oid)) { | |
7207 | c = p.second; | |
7c673cae FG |
7208 | break; |
7209 | } | |
7210 | } | |
7211 | if (!c) { | |
3efd9988 | 7212 | derr << "fsck error: stray object " << oid |
7c673cae FG |
7213 | << " not owned by any collection" << dendl; |
7214 | ++errors; | |
7215 | continue; | |
7216 | } | |
11fdf7f2 TL |
7217 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; |
7218 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
7219 | << dendl; | |
7220 | if (need_per_pool_stats) { | |
7221 | expected_statfs = &expected_pool_statfs[pool_id]; | |
7222 | } | |
7223 | ||
28e407b8 AA |
7224 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode |
7225 | << dendl; | |
7c673cae FG |
7226 | } |
7227 | ||
7228 | if (!expecting_shards.empty()) { | |
7229 | for (auto &k : expecting_shards) { | |
3efd9988 | 7230 | derr << "fsck error: missing shard key " |
7c673cae FG |
7231 | << pretty_binary_string(k) << dendl; |
7232 | } | |
7233 | ++errors; | |
7234 | expecting_shards.clear(); | |
7235 | } | |
7236 | ||
7237 | dout(10) << __func__ << " " << oid << dendl; | |
11fdf7f2 | 7238 | store_statfs_t onode_statfs; |
7c673cae FG |
7239 | RWLock::RLocker l(c->lock); |
7240 | OnodeRef o = c->get_onode(oid, false); | |
7241 | if (o->onode.nid) { | |
7242 | if (o->onode.nid > nid_max) { | |
3efd9988 | 7243 | derr << "fsck error: " << oid << " nid " << o->onode.nid |
7c673cae FG |
7244 | << " > nid_max " << nid_max << dendl; |
7245 | ++errors; | |
7246 | } | |
7247 | if (used_nids.count(o->onode.nid)) { | |
3efd9988 | 7248 | derr << "fsck error: " << oid << " nid " << o->onode.nid |
7c673cae FG |
7249 | << " already in use" << dendl; |
7250 | ++errors; | |
7251 | continue; // go for next object | |
7252 | } | |
7253 | used_nids.insert(o->onode.nid); | |
7254 | } | |
7255 | ++num_objects; | |
7256 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); | |
7257 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
81eedcae | 7258 | _dump_onode<30>(cct, *o); |
7c673cae FG |
7259 | // shards |
7260 | if (!o->extent_map.shards.empty()) { | |
7261 | ++num_sharded_objects; | |
7262 | num_object_shards += o->extent_map.shards.size(); | |
7263 | } | |
7264 | for (auto& s : o->extent_map.shards) { | |
7265 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
7266 | expecting_shards.push_back(string()); | |
7267 | get_extent_shard_key(o->key, s.shard_info->offset, | |
7268 | &expecting_shards.back()); | |
7269 | if (s.shard_info->offset >= o->onode.size) { | |
3efd9988 | 7270 | derr << "fsck error: " << oid << " shard 0x" << std::hex |
7c673cae FG |
7271 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size |
7272 | << std::dec << dendl; | |
7273 | ++errors; | |
7274 | } | |
7275 | } | |
7276 | // lextents | |
7277 | map<BlobRef,bluestore_blob_t::unused_t> referenced; | |
7278 | uint64_t pos = 0; | |
7279 | mempool::bluestore_fsck::map<BlobRef, | |
7280 | bluestore_blob_use_tracker_t> ref_map; | |
7281 | for (auto& l : o->extent_map.extent_map) { | |
7282 | dout(20) << __func__ << " " << l << dendl; | |
7283 | if (l.logical_offset < pos) { | |
3efd9988 | 7284 | derr << "fsck error: " << oid << " lextent at 0x" |
7c673cae FG |
7285 | << std::hex << l.logical_offset |
7286 | << " overlaps with the previous, which ends at 0x" << pos | |
7287 | << std::dec << dendl; | |
7288 | ++errors; | |
7289 | } | |
7290 | if (o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
3efd9988 | 7291 | derr << "fsck error: " << oid << " lextent at 0x" |
7c673cae FG |
7292 | << std::hex << l.logical_offset << "~" << l.length |
7293 | << " spans a shard boundary" | |
7294 | << std::dec << dendl; | |
7295 | ++errors; | |
7296 | } | |
7297 | pos = l.logical_offset + l.length; | |
11fdf7f2 TL |
7298 | onode_statfs.data_stored += l.length; |
7299 | ceph_assert(l.blob); | |
7c673cae FG |
7300 | const bluestore_blob_t& blob = l.blob->get_blob(); |
7301 | ||
7302 | auto& ref = ref_map[l.blob]; | |
7303 | if (ref.is_empty()) { | |
7304 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
7305 | uint32_t l = blob.get_logical_length(); | |
7306 | ref.init(l, min_release_size); | |
7307 | } | |
7308 | ref.get( | |
7309 | l.blob_offset, | |
7310 | l.length); | |
7311 | ++num_extents; | |
7312 | if (blob.has_unused()) { | |
7313 | auto p = referenced.find(l.blob); | |
7314 | bluestore_blob_t::unused_t *pu; | |
7315 | if (p == referenced.end()) { | |
7316 | pu = &referenced[l.blob]; | |
7317 | } else { | |
7318 | pu = &p->second; | |
7319 | } | |
7320 | uint64_t blob_len = blob.get_logical_length(); | |
11fdf7f2 TL |
7321 | ceph_assert((blob_len % (sizeof(*pu)*8)) == 0); |
7322 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
7c673cae FG |
7323 | uint64_t chunk_size = blob_len / (sizeof(*pu)*8); |
7324 | uint64_t start = l.blob_offset / chunk_size; | |
7325 | uint64_t end = | |
11fdf7f2 | 7326 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; |
7c673cae FG |
7327 | for (auto i = start; i < end; ++i) { |
7328 | (*pu) |= (1u << i); | |
7329 | } | |
7330 | } | |
7331 | } | |
7332 | for (auto &i : referenced) { | |
7333 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
7334 | << std::dec << " for " << *i.first << dendl; | |
7335 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7336 | if (i.second & blob.unused) { | |
3efd9988 | 7337 | derr << "fsck error: " << oid << " blob claims unused 0x" |
7c673cae | 7338 | << std::hex << blob.unused |
11fdf7f2 | 7339 | << " but extents reference 0x" << i.second << std::dec |
7c673cae FG |
7340 | << " on blob " << *i.first << dendl; |
7341 | ++errors; | |
7342 | } | |
7343 | if (blob.has_csum()) { | |
7344 | uint64_t blob_len = blob.get_logical_length(); | |
7345 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused)*8); | |
7346 | unsigned csum_count = blob.get_csum_count(); | |
7347 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
7348 | for (unsigned p = 0; p < csum_count; ++p) { | |
7349 | unsigned pos = p * csum_chunk_size; | |
7350 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
7351 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
7352 | unsigned mask = 1u << firstbit; | |
7353 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
7354 | mask |= 1u << b; | |
7355 | } | |
7356 | if ((blob.unused & mask) == mask) { | |
7357 | // this csum chunk region is marked unused | |
7358 | if (blob.get_csum_item(p) != 0) { | |
3efd9988 | 7359 | derr << "fsck error: " << oid |
7c673cae FG |
7360 | << " blob claims csum chunk 0x" << std::hex << pos |
7361 | << "~" << csum_chunk_size | |
7362 | << " is unused (mask 0x" << mask << " of unused 0x" | |
7363 | << blob.unused << ") but csum is non-zero 0x" | |
7364 | << blob.get_csum_item(p) << std::dec << " on blob " | |
7365 | << *i.first << dendl; | |
7366 | ++errors; | |
7367 | } | |
7368 | } | |
7369 | } | |
7370 | } | |
7371 | } | |
7372 | for (auto &i : ref_map) { | |
7373 | ++num_blobs; | |
7374 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7375 | bool equal = i.first->get_blob_use_tracker().equal(i.second); | |
7376 | if (!equal) { | |
3efd9988 | 7377 | derr << "fsck error: " << oid << " blob " << *i.first |
7c673cae FG |
7378 | << " doesn't match expected ref_map " << i.second << dendl; |
7379 | ++errors; | |
7380 | } | |
7381 | if (blob.is_compressed()) { | |
11fdf7f2 TL |
7382 | onode_statfs.data_compressed += blob.get_compressed_payload_length(); |
7383 | onode_statfs.data_compressed_original += | |
7c673cae FG |
7384 | i.first->get_referenced_bytes(); |
7385 | } | |
7386 | if (blob.is_shared()) { | |
7387 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
3efd9988 | 7388 | derr << "fsck error: " << oid << " blob " << blob |
7c673cae FG |
7389 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " |
7390 | << blobid_max << dendl; | |
7391 | ++errors; | |
7392 | } else if (i.first->shared_blob->get_sbid() == 0) { | |
3efd9988 | 7393 | derr << "fsck error: " << oid << " blob " << blob |
7c673cae FG |
7394 | << " marked as shared but has uninitialized sbid" |
7395 | << dendl; | |
7396 | ++errors; | |
7397 | } | |
7398 | sb_info_t& sbi = sb_info[i.first->shared_blob->get_sbid()]; | |
11fdf7f2 TL |
7399 | ceph_assert(sbi.cid == coll_t() || sbi.cid == c->cid); |
7400 | ceph_assert(sbi.pool_id == INT64_MIN || | |
7401 | sbi.pool_id == oid.hobj.get_logical_pool()); | |
7402 | sbi.cid = c->cid; | |
7403 | sbi.pool_id = oid.hobj.get_logical_pool(); | |
7c673cae FG |
7404 | sbi.sb = i.first->shared_blob; |
7405 | sbi.oids.push_back(oid); | |
7406 | sbi.compressed = blob.is_compressed(); | |
7407 | for (auto e : blob.get_extents()) { | |
7408 | if (e.is_valid()) { | |
7409 | sbi.ref_map.get(e.offset, e.length); | |
7410 | } | |
7411 | } | |
7412 | } else { | |
11fdf7f2 | 7413 | errors += _fsck_check_extents(c->cid, oid, blob.get_extents(), |
7c673cae FG |
7414 | blob.is_compressed(), |
7415 | used_blocks, | |
b32b8144 | 7416 | fm->get_alloc_size(), |
11fdf7f2 TL |
7417 | repair ? &repairer : nullptr, |
7418 | onode_statfs); | |
7c673cae FG |
7419 | } |
7420 | } | |
7421 | if (deep) { | |
7422 | bufferlist bl; | |
a8e16298 TL |
7423 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; |
7424 | uint64_t offset = 0; | |
7425 | do { | |
7426 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
7427 | int r = _do_read(c.get(), o, offset, l, bl, | |
7428 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
7429 | if (r < 0) { | |
7430 | ++errors; | |
7431 | derr << "fsck error: " << oid << std::hex | |
7432 | << " error during read: " | |
7433 | << " " << offset << "~" << l | |
7434 | << " " << cpp_strerror(r) << std::dec | |
7435 | << dendl; | |
7436 | break; | |
7437 | } | |
7438 | offset += l; | |
7439 | } while (offset < o->onode.size); | |
7c673cae FG |
7440 | } |
7441 | // omap | |
7442 | if (o->onode.has_omap()) { | |
11fdf7f2 TL |
7443 | auto& m = |
7444 | o->onode.is_pgmeta_omap() ? used_pgmeta_omap_head : used_omap_head; | |
7445 | if (m.count(o->onode.nid)) { | |
3efd9988 | 7446 | derr << "fsck error: " << oid << " omap_head " << o->onode.nid |
7c673cae FG |
7447 | << " already in use" << dendl; |
7448 | ++errors; | |
7449 | } else { | |
11fdf7f2 | 7450 | m.insert(o->onode.nid); |
7c673cae FG |
7451 | } |
7452 | } | |
11fdf7f2 TL |
7453 | expected_statfs->add(onode_statfs); |
7454 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
7455 | } // if (it) | |
7456 | ||
7c673cae FG |
7457 | dout(1) << __func__ << " checking shared_blobs" << dendl; |
7458 | it = db->get_iterator(PREFIX_SHARED_BLOB); | |
7459 | if (it) { | |
11fdf7f2 TL |
7460 | //fill global if not overriden below |
7461 | expected_statfs = &expected_store_statfs; | |
7462 | ||
7c673cae FG |
7463 | for (it->lower_bound(string()); it->valid(); it->next()) { |
7464 | string key = it->key(); | |
7465 | uint64_t sbid; | |
7466 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 7467 | derr << "fsck error: bad key '" << key |
7c673cae | 7468 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
7469 | if (repair) { |
7470 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
7471 | } | |
7c673cae FG |
7472 | ++errors; |
7473 | continue; | |
7474 | } | |
7475 | auto p = sb_info.find(sbid); | |
7476 | if (p == sb_info.end()) { | |
3efd9988 | 7477 | derr << "fsck error: found stray shared blob data for sbid 0x" |
7c673cae | 7478 | << std::hex << sbid << std::dec << dendl; |
11fdf7f2 TL |
7479 | if (repair) { |
7480 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
7481 | } | |
7c673cae FG |
7482 | ++errors; |
7483 | } else { | |
7484 | ++num_shared_blobs; | |
7485 | sb_info_t& sbi = p->second; | |
7486 | bluestore_shared_blob_t shared_blob(sbid); | |
7487 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
7488 | auto blp = bl.cbegin(); |
7489 | try { | |
7490 | decode(shared_blob, blp); | |
7491 | } catch (buffer::error& e) { | |
7492 | ++errors; | |
7493 | // Force update and don't report as missing | |
7494 | sbi.updated = sbi.passed = true; | |
7495 | ||
7496 | derr << "fsck error: failed to decode Shared Blob" | |
7497 | << pretty_binary_string(it->key()) << dendl; | |
7498 | if (repair) { | |
7499 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
7500 | << pretty_binary_string(it->key()) | |
7501 | << "', removing" << dendl; | |
7502 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
7503 | } | |
7504 | continue; | |
7505 | } | |
7c673cae FG |
7506 | dout(20) << __func__ << " " << *sbi.sb << " " << shared_blob << dendl; |
7507 | if (shared_blob.ref_map != sbi.ref_map) { | |
3efd9988 | 7508 | derr << "fsck error: shared blob 0x" << std::hex << sbid |
11fdf7f2 TL |
7509 | << std::dec << " ref_map " << shared_blob.ref_map |
7510 | << " != expected " << sbi.ref_map << dendl; | |
7511 | sbi.updated = true; // will update later in repair mode only! | |
7c673cae FG |
7512 | ++errors; |
7513 | } | |
7514 | PExtentVector extents; | |
7515 | for (auto &r : shared_blob.ref_map.ref_map) { | |
7516 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
7517 | } | |
11fdf7f2 TL |
7518 | if (need_per_pool_stats) { |
7519 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; | |
7520 | } | |
7521 | errors += _fsck_check_extents(sbi.cid, | |
7522 | p->second.oids.front(), | |
7c673cae FG |
7523 | extents, |
7524 | p->second.compressed, | |
b32b8144 FG |
7525 | used_blocks, |
7526 | fm->get_alloc_size(), | |
11fdf7f2 TL |
7527 | repair ? &repairer : nullptr, |
7528 | *expected_statfs); | |
7529 | sbi.passed = true; | |
7530 | } | |
7531 | } | |
7532 | } // if (it) | |
7533 | ||
7534 | if (repair && repairer.preprocess_misreference(db)) { | |
7535 | ||
7536 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
7537 | auto& space_tracker = repairer.get_space_usage_tracker(); | |
7538 | auto& misref_extents = repairer.get_misreferences(); | |
7539 | interval_set<uint64_t> to_release; | |
7540 | it = db->get_iterator(PREFIX_OBJ); | |
7541 | if (it) { | |
7542 | //fill global if not overriden below | |
7543 | expected_statfs = &expected_store_statfs; | |
7544 | ||
7545 | CollectionRef c; | |
7546 | spg_t pgid; | |
7547 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
7548 | bool bypass_rest = false; | |
7549 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
7550 | it->next()) { | |
7551 | dout(30) << __func__ << " key " | |
7552 | << pretty_binary_string(it->key()) << dendl; | |
7553 | if (is_extent_shard_key(it->key())) { | |
7554 | continue; | |
7555 | } | |
7556 | ||
7557 | ghobject_t oid; | |
7558 | int r = get_key_object(it->key(), &oid); | |
7559 | if (r < 0 || !space_tracker.is_used(oid)) { | |
7560 | continue; | |
7561 | } | |
7562 | ||
7563 | if (!c || | |
7564 | oid.shard_id != pgid.shard || | |
7565 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
7566 | !c->contains(oid)) { | |
7567 | c = nullptr; | |
7568 | for (auto& p : coll_map) { | |
7569 | if (p.second->contains(oid)) { | |
7570 | c = p.second; | |
7571 | break; | |
7572 | } | |
7573 | } | |
7574 | if (!c) { | |
7575 | continue; | |
7576 | } | |
7577 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
7578 | if (need_per_pool_stats) { | |
7579 | expected_statfs = &expected_pool_statfs[pool_id]; | |
7580 | } | |
7581 | } | |
7582 | if (!space_tracker.is_used(c->cid)) { | |
7583 | continue; | |
7584 | } | |
7585 | ||
7586 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
7587 | << " obj:" << oid << dendl; | |
7588 | ||
7589 | RWLock::RLocker l(c->lock); | |
7590 | OnodeRef o = c->get_onode(oid, false); | |
7591 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
7592 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
7593 | ||
7594 | for (auto& e : o->extent_map.extent_map) { | |
7595 | blobs.insert(e.blob); | |
7596 | } | |
7597 | bool need_onode_update = false; | |
7598 | bool first_dump = true; | |
7599 | for(auto b : blobs) { | |
7600 | bool broken_blob = false; | |
7601 | auto& pextents = b->dirty_blob().dirty_extents(); | |
7602 | for (auto& e : pextents) { | |
7603 | if (!e.is_valid()) { | |
7604 | continue; | |
7605 | } | |
7606 | // for the sake of simplicity and proper shared blob handling | |
7607 | // always rewrite the whole blob even when it's partially | |
7608 | // misreferenced. | |
7609 | if (misref_extents.intersects(e.offset, e.length)) { | |
7610 | if (first_dump) { | |
7611 | first_dump = false; | |
81eedcae | 7612 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
7613 | } |
7614 | broken_blob = true; | |
7615 | break; | |
7616 | } | |
7617 | } | |
7618 | if (!broken_blob) | |
7619 | continue; | |
7620 | bool compressed = b->get_blob().is_compressed(); | |
7621 | need_onode_update = true; | |
7622 | dout(10) << __func__ | |
7623 | << " fix misreferences in oid:" << oid | |
7624 | << " " << *b << dendl; | |
7625 | uint64_t b_off = 0; | |
7626 | PExtentVector pext_to_release; | |
7627 | pext_to_release.reserve(pextents.size()); | |
7628 | // rewriting all valid pextents | |
7629 | for (auto e = pextents.begin(); e != pextents.end(); | |
7630 | b_off += e->length, e++) { | |
7631 | if (!e->is_valid()) { | |
7632 | continue; | |
7633 | } | |
7634 | PExtentVector exts; | |
7635 | int64_t alloc_len = alloc->allocate(e->length, min_alloc_size, | |
7636 | 0, 0, &exts); | |
7637 | if (alloc_len < (int64_t)e->length) { | |
7638 | derr << __func__ | |
7639 | << " failed to allocate 0x" << std::hex << e->length | |
7640 | << " allocated 0x " << alloc_len | |
7641 | << " min_alloc_size 0x" << min_alloc_size | |
7642 | << " available 0x " << alloc->get_free() | |
7643 | << std::dec << dendl; | |
7644 | if (alloc_len > 0) { | |
7645 | alloc->release(exts); | |
7646 | } | |
7647 | bypass_rest = true; | |
7648 | break; | |
7649 | } | |
7650 | expected_statfs->allocated += e->length; | |
7651 | if (compressed) { | |
7652 | expected_statfs->data_compressed_allocated += e->length; | |
7653 | } | |
7654 | ||
7655 | bufferlist bl; | |
7656 | IOContext ioc(cct, NULL, true); // allow EIO | |
7657 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); | |
7658 | if (r < 0) { | |
7659 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
7660 | <<"~" << e->length << std::dec << dendl; | |
7661 | ceph_abort_msg("read failed, wtf"); | |
7662 | } | |
7663 | pext_to_release.push_back(*e); | |
7664 | e = pextents.erase(e); | |
7665 | e = pextents.insert(e, exts.begin(), exts.end()); | |
7666 | b->get_blob().map_bl( | |
7667 | b_off, bl, | |
7668 | [&](uint64_t offset, bufferlist& t) { | |
7669 | int r = bdev->write(offset, t, false); | |
7670 | ceph_assert(r == 0); | |
7671 | }); | |
7672 | e += exts.size() - 1; | |
7673 | for (auto& p : exts) { | |
7674 | fm->allocate(p.offset, p.length, txn); | |
7675 | } | |
7676 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
7677 | ||
7678 | if (b->get_blob().is_shared()) { | |
7679 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
7680 | ||
7681 | auto sb_it = sb_info.find(b->shared_blob->get_sbid()); | |
7682 | ceph_assert(sb_it != sb_info.end()); | |
7683 | sb_info_t& sbi = sb_it->second; | |
7684 | ||
7685 | for (auto& r : sbi.ref_map.ref_map) { | |
7686 | expected_statfs->allocated -= r.second.length; | |
7687 | if (sbi.compressed) { | |
7688 | // NB: it's crucial to use compressed flag from sb_info_t | |
7689 | // as we originally used that value while accumulating | |
7690 | // expected_statfs | |
7691 | expected_statfs->data_compressed_allocated -= r.second.length; | |
7692 | } | |
7693 | } | |
7694 | sbi.updated = sbi.passed = true; | |
7695 | sbi.ref_map.clear(); | |
7696 | ||
7697 | // relying on blob's pextents to decide what to release. | |
7698 | for (auto& p : pext_to_release) { | |
7699 | to_release.union_insert(p.offset, p.length); | |
7700 | } | |
7701 | } else { | |
7702 | for (auto& p : pext_to_release) { | |
7703 | expected_statfs->allocated -= p.length; | |
7704 | if (compressed) { | |
7705 | expected_statfs->data_compressed_allocated -= p.length; | |
7706 | } | |
7707 | to_release.union_insert(p.offset, p.length); | |
7708 | } | |
7709 | } | |
7710 | if (bypass_rest) { | |
7711 | break; | |
7712 | } | |
7713 | } // for(auto b : blobs) | |
7714 | if (need_onode_update) { | |
7715 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
7716 | _record_onode(o, txn); | |
7717 | } | |
7718 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
7719 | ||
7720 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
7721 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
7722 | << "~" << it.get_len() << std::dec << dendl; | |
7723 | fm->release(it.get_start(), it.get_len(), txn); | |
7724 | } | |
7725 | alloc->release(to_release); | |
7726 | to_release.clear(); | |
7727 | } // if (it) { | |
7728 | } //if (repair && repairer.preprocess_misreference()) { | |
7729 | ||
7730 | for (auto &p : sb_info) { | |
7731 | sb_info_t& sbi = p.second; | |
7732 | if (!sbi.passed) { | |
7733 | derr << "fsck error: missing " << *sbi.sb << dendl; | |
7734 | ++errors; | |
7735 | } | |
7736 | if (repair && (!sbi.passed || sbi.updated)) { | |
7737 | auto sbid = p.first; | |
7738 | if (sbi.ref_map.empty()) { | |
7739 | ceph_assert(sbi.passed); | |
7740 | dout(20) << __func__ << " " << *sbi.sb | |
7741 | << " is empty, removing" << dendl; | |
7742 | repairer.fix_shared_blob(db, sbid, nullptr); | |
7743 | } else { | |
7744 | bufferlist bl; | |
7745 | bluestore_shared_blob_t persistent(sbid, std::move(sbi.ref_map)); | |
7746 | encode(persistent, bl); | |
7747 | dout(20) << __func__ << " " << *sbi.sb | |
7748 | << " is " << bl.length() << " bytes, updating" << dendl; | |
7749 | ||
7750 | repairer.fix_shared_blob(db, sbid, &bl); | |
7c673cae FG |
7751 | } |
7752 | } | |
7753 | } | |
11fdf7f2 TL |
7754 | sb_info.clear(); |
7755 | ||
7756 | // check global stats if no-pps is enforced only | |
7757 | if (!need_per_pool_stats) { | |
7758 | if (!(actual_statfs == expected_store_statfs)) { | |
7759 | derr << "fsck error: actual " << actual_statfs | |
7760 | << " != expected " << expected_store_statfs << dendl; | |
7761 | if (repair) { | |
7762 | repairer.fix_statfs(db, BLUESTORE_GLOBAL_STATFS_KEY, | |
7763 | expected_store_statfs); | |
7764 | } | |
7765 | ++errors; | |
7766 | } | |
7c673cae | 7767 | } |
11fdf7f2 TL |
7768 | if (!enforce_no_per_pool_stats) { |
7769 | dout(1) << __func__ << " checking pool_statfs" << dendl; | |
7770 | _fsck_check_pool_statfs(expected_pool_statfs, need_per_pool_stats, | |
7771 | errors, repair ? &repairer : nullptr); | |
7c673cae FG |
7772 | } |
7773 | ||
7774 | dout(1) << __func__ << " checking for stray omap data" << dendl; | |
7775 | it = db->get_iterator(PREFIX_OMAP); | |
7776 | if (it) { | |
7777 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7778 | uint64_t omap_head; | |
7779 | _key_decode_u64(it->key().c_str(), &omap_head); | |
7780 | if (used_omap_head.count(omap_head) == 0) { | |
3efd9988 | 7781 | derr << "fsck error: found stray omap data on omap_head " |
7c673cae FG |
7782 | << omap_head << dendl; |
7783 | ++errors; | |
7784 | } | |
7785 | } | |
7786 | } | |
11fdf7f2 TL |
7787 | it = db->get_iterator(PREFIX_PGMETA_OMAP); |
7788 | if (it) { | |
7789 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7790 | uint64_t omap_head; | |
7791 | _key_decode_u64(it->key().c_str(), &omap_head); | |
7792 | if (used_pgmeta_omap_head.count(omap_head) == 0) { | |
7793 | derr << "fsck error: found stray omap data on omap_head " | |
7794 | << omap_head << dendl; | |
7795 | ++errors; | |
7796 | } | |
7797 | } | |
7798 | } | |
7c673cae FG |
7799 | |
7800 | dout(1) << __func__ << " checking deferred events" << dendl; | |
7801 | it = db->get_iterator(PREFIX_DEFERRED); | |
7802 | if (it) { | |
7803 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7804 | bufferlist bl = it->value(); | |
11fdf7f2 | 7805 | auto p = bl.cbegin(); |
7c673cae FG |
7806 | bluestore_deferred_transaction_t wt; |
7807 | try { | |
11fdf7f2 | 7808 | decode(wt, p); |
7c673cae | 7809 | } catch (buffer::error& e) { |
3efd9988 | 7810 | derr << "fsck error: failed to decode deferred txn " |
7c673cae | 7811 | << pretty_binary_string(it->key()) << dendl; |
11fdf7f2 TL |
7812 | if (repair) { |
7813 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
7814 | << pretty_binary_string(it->key()) | |
7815 | << "', removing" << dendl; | |
7816 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
7817 | } | |
7818 | continue; | |
7c673cae FG |
7819 | } |
7820 | dout(20) << __func__ << " deferred " << wt.seq | |
7821 | << " ops " << wt.ops.size() | |
7822 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
7823 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
7824 | apply( | |
b32b8144 | 7825 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae | 7826 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
11fdf7f2 | 7827 | ceph_assert(pos < bs.size()); |
7c673cae FG |
7828 | bs.set(pos); |
7829 | } | |
7830 | ); | |
7831 | } | |
7832 | } | |
7833 | } | |
7834 | ||
7835 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
7836 | { | |
7837 | // remove bluefs_extents from used set since the freelist doesn't | |
7838 | // know they are allocated. | |
7839 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
7840 | apply( | |
b32b8144 | 7841 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae | 7842 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
11fdf7f2 | 7843 | ceph_assert(pos < bs.size()); |
7c673cae FG |
7844 | bs.reset(pos); |
7845 | } | |
7846 | ); | |
7847 | } | |
7848 | fm->enumerate_reset(); | |
7849 | uint64_t offset, length; | |
11fdf7f2 | 7850 | while (fm->enumerate_next(db, &offset, &length)) { |
7c673cae FG |
7851 | bool intersects = false; |
7852 | apply( | |
b32b8144 | 7853 | offset, length, fm->get_alloc_size(), used_blocks, |
7c673cae | 7854 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
11fdf7f2 | 7855 | ceph_assert(pos < bs.size()); |
7c673cae | 7856 | if (bs.test(pos)) { |
11fdf7f2 TL |
7857 | if (offset == SUPER_RESERVED && |
7858 | length == min_alloc_size - SUPER_RESERVED) { | |
7859 | // this is due to the change just after luminous to min_alloc_size | |
7860 | // granularity allocations, and our baked in assumption at the top | |
7861 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
7862 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
7863 | // since we will never allocate this region below min_alloc_size. | |
7864 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
7865 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
7866 | << length << std::dec << dendl; | |
7867 | } else { | |
7868 | intersects = true; | |
7869 | if (repair) { | |
7870 | repairer.fix_false_free(db, fm, | |
7871 | pos * min_alloc_size, | |
7872 | min_alloc_size); | |
7873 | } | |
7874 | } | |
7c673cae FG |
7875 | } else { |
7876 | bs.set(pos); | |
7877 | } | |
7878 | } | |
7879 | ); | |
7880 | if (intersects) { | |
11fdf7f2 TL |
7881 | derr << "fsck error: free extent 0x" << std::hex << offset |
7882 | << "~" << length << std::dec | |
7883 | << " intersects allocated blocks" << dendl; | |
7884 | ++errors; | |
b5b8bbf5 FG |
7885 | } |
7886 | } | |
3efd9988 FG |
7887 | fm->enumerate_reset(); |
7888 | size_t count = used_blocks.count(); | |
7c673cae | 7889 | if (used_blocks.size() != count) { |
11fdf7f2 | 7890 | ceph_assert(used_blocks.size() > count); |
b5b8bbf5 FG |
7891 | used_blocks.flip(); |
7892 | size_t start = used_blocks.find_first(); | |
7893 | while (start != decltype(used_blocks)::npos) { | |
7894 | size_t cur = start; | |
7895 | while (true) { | |
7896 | size_t next = used_blocks.find_next(cur); | |
7897 | if (next != cur + 1) { | |
11fdf7f2 | 7898 | ++errors; |
3efd9988 | 7899 | derr << "fsck error: leaked extent 0x" << std::hex |
b32b8144 FG |
7900 | << ((uint64_t)start * fm->get_alloc_size()) << "~" |
7901 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
b5b8bbf5 | 7902 | << dendl; |
11fdf7f2 TL |
7903 | if (repair) { |
7904 | repairer.fix_leaked(db, | |
7905 | fm, | |
7906 | start * min_alloc_size, | |
7907 | (cur + 1 - start) * min_alloc_size); | |
7908 | } | |
b5b8bbf5 FG |
7909 | start = next; |
7910 | break; | |
7911 | } | |
7912 | cur = next; | |
7913 | } | |
7914 | } | |
7915 | used_blocks.flip(); | |
7c673cae FG |
7916 | } |
7917 | } | |
11fdf7f2 TL |
7918 | if (repair) { |
7919 | dout(5) << __func__ << " applying repair results" << dendl; | |
7920 | repaired = repairer.apply(db); | |
7921 | dout(5) << __func__ << " repair applied" << dendl; | |
7922 | } | |
7c673cae FG |
7923 | out_scan: |
7924 | mempool_thread.shutdown(); | |
31f18b77 | 7925 | _flush_cache(); |
7c673cae FG |
7926 | out_db: |
7927 | it.reset(); // before db is closed | |
11fdf7f2 | 7928 | _close_db_and_around(); |
7c673cae FG |
7929 | out_bdev: |
7930 | _close_bdev(); | |
7931 | out_fsid: | |
7932 | _close_fsid(); | |
7933 | out_path: | |
7934 | _close_path(); | |
7935 | ||
7936 | // fatal errors take precedence | |
7937 | if (r < 0) | |
7938 | return r; | |
7939 | ||
7940 | dout(2) << __func__ << " " << num_objects << " objects, " | |
7941 | << num_sharded_objects << " of them sharded. " | |
7942 | << dendl; | |
7943 | dout(2) << __func__ << " " << num_extents << " extents to " | |
7944 | << num_blobs << " blobs, " | |
7945 | << num_spanning_blobs << " spanning, " | |
7946 | << num_shared_blobs << " shared." | |
7947 | << dendl; | |
7948 | ||
7949 | utime_t duration = ceph_clock_now() - start; | |
11fdf7f2 TL |
7950 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " << repaired |
7951 | << " repaired, " << (errors - (int)repaired) << " remaining in " | |
7c673cae | 7952 | << duration << " seconds" << dendl; |
11fdf7f2 TL |
7953 | return errors - (int)repaired; |
7954 | } | |
7955 | ||
7956 | /// methods to inject various errors fsck can repair | |
7957 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
7958 | const bufferlist& bl) | |
7959 | { | |
7960 | KeyValueDB::Transaction txn; | |
7961 | txn = db->get_transaction(); | |
7962 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
7963 | db->submit_transaction_sync(txn); | |
7964 | }; | |
7965 | ||
7966 | void BlueStore::inject_leaked(uint64_t len) | |
7967 | { | |
7968 | KeyValueDB::Transaction txn; | |
7969 | txn = db->get_transaction(); | |
7970 | ||
7971 | PExtentVector exts; | |
7972 | int64_t alloc_len = alloc->allocate(len, min_alloc_size, | |
7973 | min_alloc_size * 256, 0, &exts); | |
7974 | ceph_assert(alloc_len >= (int64_t)len); | |
7975 | for (auto& p : exts) { | |
7976 | fm->allocate(p.offset, p.length, txn); | |
7977 | } | |
7978 | db->submit_transaction_sync(txn); | |
7979 | } | |
7980 | ||
7981 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
7982 | { | |
7983 | KeyValueDB::Transaction txn; | |
7984 | OnodeRef o; | |
7985 | CollectionRef c = _get_collection(cid); | |
7986 | ceph_assert(c); | |
7987 | { | |
7988 | RWLock::WLocker l(c->lock); // just to avoid internal asserts | |
7989 | o = c->get_onode(oid, false); | |
7990 | ceph_assert(o); | |
7991 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
7992 | } | |
7993 | ||
7994 | bool injected = false; | |
7995 | txn = db->get_transaction(); | |
7996 | auto& em = o->extent_map.extent_map; | |
7997 | std::vector<const PExtentVector*> v; | |
7998 | if (em.size()) { | |
7999 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
8000 | } | |
8001 | if (em.size() > 1) { | |
8002 | auto it = em.end(); | |
8003 | --it; | |
8004 | v.push_back(&(it->blob->get_blob().get_extents())); | |
8005 | } | |
8006 | for (auto pext : v) { | |
8007 | if (pext->size()) { | |
8008 | auto p = pext->begin(); | |
8009 | while (p != pext->end()) { | |
8010 | if (p->is_valid()) { | |
8011 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
8012 | << "~" << p->length << std::dec << dendl; | |
8013 | fm->release(p->offset, p->length, txn); | |
8014 | injected = true; | |
8015 | break; | |
8016 | } | |
8017 | ++p; | |
8018 | } | |
8019 | } | |
8020 | } | |
8021 | ceph_assert(injected); | |
8022 | db->submit_transaction_sync(txn); | |
8023 | } | |
8024 | ||
8025 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) | |
8026 | { | |
8027 | BlueStoreRepairer repairer; | |
8028 | repairer.fix_statfs(db, key, new_statfs); | |
8029 | repairer.apply(db); | |
8030 | } | |
8031 | ||
8032 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, | |
8033 | coll_t cid2, ghobject_t oid2, | |
8034 | uint64_t offset) | |
8035 | { | |
8036 | OnodeRef o1; | |
8037 | CollectionRef c1 = _get_collection(cid1); | |
8038 | ceph_assert(c1); | |
8039 | { | |
8040 | RWLock::WLocker l(c1->lock); // just to avoid internal asserts | |
8041 | o1 = c1->get_onode(oid1, false); | |
8042 | ceph_assert(o1); | |
8043 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
8044 | } | |
8045 | OnodeRef o2; | |
8046 | CollectionRef c2 = _get_collection(cid2); | |
8047 | ceph_assert(c2); | |
8048 | { | |
8049 | RWLock::WLocker l(c2->lock); // just to avoid internal asserts | |
8050 | o2 = c2->get_onode(oid2, false); | |
8051 | ceph_assert(o2); | |
8052 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
8053 | } | |
8054 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
8055 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
8056 | ||
8057 | // require onode/extent layout to be the same (and simple) | |
8058 | // to make things easier | |
8059 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
8060 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
8061 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
8062 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
8063 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
8064 | ceph_assert(e1.length == e2.length); | |
8065 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
8066 | ||
8067 | KeyValueDB::Transaction txn; | |
8068 | txn = db->get_transaction(); | |
8069 | ||
8070 | // along with misreference error this will create space leaks errors | |
8071 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
8072 | o2->extent_map.dirty_range(offset, e2.length); | |
8073 | o2->extent_map.update(txn, false); | |
8074 | ||
8075 | _record_onode(o2, txn); | |
8076 | db->submit_transaction_sync(txn); | |
7c673cae FG |
8077 | } |
8078 | ||
8079 | void BlueStore::collect_metadata(map<string,string> *pm) | |
8080 | { | |
8081 | dout(10) << __func__ << dendl; | |
8082 | bdev->collect_metadata("bluestore_bdev_", pm); | |
8083 | if (bluefs) { | |
8084 | (*pm)["bluefs"] = "1"; | |
8085 | (*pm)["bluefs_single_shared_device"] = stringify((int)bluefs_single_shared_device); | |
11fdf7f2 | 8086 | bluefs->collect_metadata(pm, bluefs_shared_bdev); |
7c673cae FG |
8087 | } else { |
8088 | (*pm)["bluefs"] = "0"; | |
8089 | } | |
11fdf7f2 TL |
8090 | |
8091 | // report numa mapping for underlying devices | |
8092 | int node = -1; | |
8093 | set<int> nodes; | |
8094 | set<string> failed; | |
8095 | int r = get_numa_node(&node, &nodes, &failed); | |
8096 | if (r >= 0) { | |
8097 | if (!failed.empty()) { | |
8098 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
8099 | } | |
8100 | if (!nodes.empty()) { | |
8101 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
8102 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
8103 | } | |
8104 | if (node >= 0) { | |
8105 | (*pm)["objectstore_numa_node"] = stringify(node); | |
8106 | } | |
8107 | } | |
8108 | } | |
8109 | ||
8110 | int BlueStore::get_numa_node( | |
8111 | int *final_node, | |
8112 | set<int> *out_nodes, | |
8113 | set<string> *out_failed) | |
8114 | { | |
8115 | int node = -1; | |
8116 | set<string> devices; | |
8117 | get_devices(&devices); | |
8118 | set<int> nodes; | |
8119 | set<string> failed; | |
8120 | for (auto& devname : devices) { | |
8121 | int n; | |
8122 | BlkDev bdev(devname); | |
8123 | int r = bdev.get_numa_node(&n); | |
8124 | if (r < 0) { | |
8125 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
8126 | << dendl; | |
8127 | failed.insert(devname); | |
8128 | continue; | |
8129 | } | |
8130 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
8131 | << dendl; | |
8132 | nodes.insert(n); | |
8133 | if (node < 0) { | |
8134 | node = n; | |
8135 | } | |
8136 | } | |
8137 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
8138 | *final_node = node; | |
8139 | } | |
8140 | if (out_nodes) { | |
8141 | *out_nodes = nodes; | |
8142 | } | |
8143 | if (out_failed) { | |
8144 | *out_failed = failed; | |
8145 | } | |
8146 | return 0; | |
8147 | } | |
8148 | ||
8149 | int BlueStore::get_devices(set<string> *ls) | |
8150 | { | |
8151 | if (bdev) { | |
8152 | bdev->get_devices(ls); | |
8153 | if (bluefs) { | |
8154 | bluefs->get_devices(ls); | |
8155 | } | |
8156 | return 0; | |
8157 | } | |
8158 | ||
8159 | // grumble, we haven't started up yet. | |
8160 | int r = _open_path(); | |
8161 | if (r < 0) | |
8162 | goto out; | |
8163 | r = _open_fsid(false); | |
8164 | if (r < 0) | |
8165 | goto out_path; | |
8166 | r = _read_fsid(&fsid); | |
8167 | if (r < 0) | |
8168 | goto out_fsid; | |
8169 | r = _lock_fsid(); | |
8170 | if (r < 0) | |
8171 | goto out_fsid; | |
8172 | r = _open_bdev(false); | |
8173 | if (r < 0) | |
8174 | goto out_fsid; | |
8175 | r = _minimal_open_bluefs(false); | |
8176 | if (r < 0) | |
8177 | goto out_bdev; | |
8178 | bdev->get_devices(ls); | |
8179 | if (bluefs) { | |
8180 | bluefs->get_devices(ls); | |
8181 | } | |
8182 | r = 0; | |
8183 | _minimal_close_bluefs(); | |
8184 | out_bdev: | |
8185 | _close_bdev(); | |
8186 | out_fsid: | |
8187 | _close_fsid(); | |
8188 | out_path: | |
8189 | _close_path(); | |
8190 | out: | |
8191 | return r; | |
7c673cae FG |
8192 | } |
8193 | ||
11fdf7f2 | 8194 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
8195 | { |
8196 | buf->reset(); | |
11fdf7f2 TL |
8197 | |
8198 | buf->omap_allocated = db->estimate_prefix_size(PREFIX_OMAP); | |
8199 | ||
8200 | uint64_t bfree = alloc->get_free(); | |
7c673cae FG |
8201 | |
8202 | if (bluefs) { | |
11fdf7f2 TL |
8203 | int64_t bluefs_total = bluefs->get_total(bluefs_shared_bdev); |
8204 | int64_t bluefs_free = bluefs->get_free(bluefs_shared_bdev); | |
94b18763 FG |
8205 | // part of our shared device is "free" according to BlueFS, but we |
8206 | // can't touch bluestore_bluefs_min of it. | |
8207 | int64_t shared_available = std::min( | |
11fdf7f2 TL |
8208 | bluefs_free, |
8209 | int64_t(bluefs_total - cct->_conf->bluestore_bluefs_min)); | |
8210 | buf->internally_reserved = bluefs_total - shared_available; | |
94b18763 | 8211 | if (shared_available > 0) { |
11fdf7f2 TL |
8212 | bfree += shared_available; |
8213 | } | |
8214 | // include dedicated db, too, if that isn't the shared device. | |
8215 | if (bluefs_shared_bdev != BlueFS::BDEV_DB) { | |
8216 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); | |
7c673cae | 8217 | } |
11fdf7f2 TL |
8218 | // call any non-omap bluefs space "internal metadata" |
8219 | buf->internal_metadata = | |
8220 | std::max(bluefs->get_used(), (uint64_t)cct->_conf->bluestore_bluefs_min) | |
8221 | - buf->omap_allocated; | |
7c673cae FG |
8222 | } |
8223 | ||
11fdf7f2 TL |
8224 | uint64_t thin_total, thin_avail; |
8225 | if (bdev->get_thin_utilization(&thin_total, &thin_avail)) { | |
8226 | buf->total += thin_total; | |
8227 | ||
8228 | // we are limited by both the size of the virtual device and the | |
8229 | // underlying physical device. | |
8230 | bfree = std::min(bfree, thin_avail); | |
8231 | ||
8232 | buf->allocated = thin_total - thin_avail; | |
8233 | } else { | |
8234 | buf->total += bdev->get_size(); | |
8235 | } | |
8236 | buf->available = bfree; | |
8237 | } | |
8238 | ||
8239 | int BlueStore::statfs(struct store_statfs_t *buf, | |
8240 | osd_alert_list_t* alerts) | |
8241 | { | |
8242 | if (alerts) { | |
8243 | alerts->clear(); | |
8244 | _log_alerts(*alerts); | |
8245 | } | |
8246 | _get_statfs_overall(buf); | |
31f18b77 | 8247 | { |
11fdf7f2 | 8248 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 8249 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
8250 | buf->data_stored = vstatfs.stored(); |
8251 | buf->data_compressed = vstatfs.compressed(); | |
8252 | buf->data_compressed_original = vstatfs.compressed_original(); | |
8253 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
8254 | } | |
8255 | ||
8256 | dout(20) << __func__ << " " << *buf << dendl; | |
8257 | return 0; | |
8258 | } | |
8259 | ||
8260 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf) | |
8261 | { | |
8262 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 8263 | |
11fdf7f2 TL |
8264 | if (!per_pool_stat_collection) { |
8265 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
8266 | return -ENOTSUP; | |
7c673cae | 8267 | } |
11fdf7f2 | 8268 | buf->reset(); |
7c673cae | 8269 | |
11fdf7f2 TL |
8270 | { |
8271 | std::lock_guard l(vstatfs_lock); | |
8272 | osd_pools[pool_id].publish(buf); | |
8273 | } | |
8274 | dout(10) << __func__ << *buf << dendl; | |
7c673cae FG |
8275 | return 0; |
8276 | } | |
8277 | ||
81eedcae TL |
8278 | void BlueStore::_check_legacy_statfs_alert() |
8279 | { | |
8280 | string s; | |
8281 | if (!per_pool_stat_collection && | |
8282 | cct->_conf->bluestore_no_per_pool_stats_tolerance != "enforce" && | |
8283 | cct->_conf->bluestore_warn_on_legacy_statfs) { | |
8284 | s = "legacy statfs reporting detected, " | |
8285 | "suggest to run store repair to get consistent statistic reports"; | |
8286 | } | |
8287 | std::lock_guard l(qlock); | |
8288 | legacy_statfs_alert = s; | |
8289 | } | |
8290 | ||
7c673cae FG |
8291 | // --------------- |
8292 | // cache | |
8293 | ||
8294 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
8295 | { | |
8296 | RWLock::RLocker l(coll_lock); | |
8297 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); | |
8298 | if (cp == coll_map.end()) | |
8299 | return CollectionRef(); | |
8300 | return cp->second; | |
8301 | } | |
8302 | ||
8303 | void BlueStore::_queue_reap_collection(CollectionRef& c) | |
8304 | { | |
8305 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
8306 | // _reap_collections and this in the same thread, |
8307 | // so no need a lock. | |
7c673cae FG |
8308 | removed_collections.push_back(c); |
8309 | } | |
8310 | ||
8311 | void BlueStore::_reap_collections() | |
8312 | { | |
94b18763 | 8313 | |
7c673cae FG |
8314 | list<CollectionRef> removed_colls; |
8315 | { | |
94b18763 FG |
8316 | // _queue_reap_collection and this in the same thread. |
8317 | // So no need a lock. | |
8318 | if (!removed_collections.empty()) | |
8319 | removed_colls.swap(removed_collections); | |
8320 | else | |
8321 | return; | |
7c673cae FG |
8322 | } |
8323 | ||
94b18763 FG |
8324 | list<CollectionRef>::iterator p = removed_colls.begin(); |
8325 | while (p != removed_colls.end()) { | |
7c673cae FG |
8326 | CollectionRef c = *p; |
8327 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
8328 | if (c->onode_map.map_any([&](OnodeRef o) { | |
11fdf7f2 | 8329 | ceph_assert(!o->exists); |
7c673cae FG |
8330 | if (o->flushing_count.load()) { |
8331 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
8332 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 8333 | return true; |
7c673cae | 8334 | } |
94b18763 | 8335 | return false; |
7c673cae | 8336 | })) { |
94b18763 | 8337 | ++p; |
7c673cae FG |
8338 | continue; |
8339 | } | |
8340 | c->onode_map.clear(); | |
94b18763 | 8341 | p = removed_colls.erase(p); |
7c673cae FG |
8342 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
8343 | } | |
94b18763 | 8344 | if (removed_colls.empty()) { |
7c673cae | 8345 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
8346 | } else { |
8347 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
8348 | } |
8349 | } | |
8350 | ||
8351 | void BlueStore::_update_cache_logger() | |
8352 | { | |
8353 | uint64_t num_onodes = 0; | |
8354 | uint64_t num_extents = 0; | |
8355 | uint64_t num_blobs = 0; | |
8356 | uint64_t num_buffers = 0; | |
8357 | uint64_t num_buffer_bytes = 0; | |
8358 | for (auto c : cache_shards) { | |
8359 | c->add_stats(&num_onodes, &num_extents, &num_blobs, | |
8360 | &num_buffers, &num_buffer_bytes); | |
8361 | } | |
8362 | logger->set(l_bluestore_onodes, num_onodes); | |
8363 | logger->set(l_bluestore_extents, num_extents); | |
8364 | logger->set(l_bluestore_blobs, num_blobs); | |
8365 | logger->set(l_bluestore_buffers, num_buffers); | |
8366 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
8367 | } | |
8368 | ||
8369 | // --------------- | |
8370 | // read operations | |
8371 | ||
8372 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
8373 | { | |
8374 | return _get_collection(cid); | |
8375 | } | |
8376 | ||
11fdf7f2 TL |
8377 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
8378 | const coll_t& cid) | |
7c673cae | 8379 | { |
11fdf7f2 TL |
8380 | RWLock::WLocker l(coll_lock); |
8381 | Collection *c = new Collection( | |
8382 | this, | |
8383 | cache_shards[cid.hash_to_shard(cache_shards.size())], | |
8384 | cid); | |
8385 | new_coll_map[cid] = c; | |
8386 | _osr_attach(c); | |
8387 | return c; | |
8388 | } | |
8389 | ||
8390 | void BlueStore::set_collection_commit_queue( | |
8391 | const coll_t& cid, | |
8392 | ContextQueue *commit_queue) | |
8393 | { | |
8394 | if (commit_queue) { | |
8395 | RWLock::RLocker l(coll_lock); | |
8396 | if (coll_map.count(cid)) { | |
8397 | coll_map[cid]->commit_queue = commit_queue; | |
8398 | } else if (new_coll_map.count(cid)) { | |
8399 | new_coll_map[cid]->commit_queue = commit_queue; | |
8400 | } | |
8401 | } | |
7c673cae FG |
8402 | } |
8403 | ||
11fdf7f2 | 8404 | |
7c673cae FG |
8405 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
8406 | { | |
8407 | Collection *c = static_cast<Collection *>(c_.get()); | |
8408 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
8409 | if (!c->exists) | |
8410 | return false; | |
8411 | ||
8412 | bool r = true; | |
8413 | ||
8414 | { | |
8415 | RWLock::RLocker l(c->lock); | |
8416 | OnodeRef o = c->get_onode(oid, false); | |
8417 | if (!o || !o->exists) | |
8418 | r = false; | |
8419 | } | |
8420 | ||
7c673cae FG |
8421 | return r; |
8422 | } | |
8423 | ||
7c673cae FG |
8424 | int BlueStore::stat( |
8425 | CollectionHandle &c_, | |
8426 | const ghobject_t& oid, | |
8427 | struct stat *st, | |
8428 | bool allow_eio) | |
8429 | { | |
8430 | Collection *c = static_cast<Collection *>(c_.get()); | |
8431 | if (!c->exists) | |
8432 | return -ENOENT; | |
8433 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
8434 | ||
8435 | { | |
8436 | RWLock::RLocker l(c->lock); | |
8437 | OnodeRef o = c->get_onode(oid, false); | |
8438 | if (!o || !o->exists) | |
8439 | return -ENOENT; | |
8440 | st->st_size = o->onode.size; | |
8441 | st->st_blksize = 4096; | |
8442 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
8443 | st->st_nlink = 1; | |
8444 | } | |
8445 | ||
7c673cae FG |
8446 | int r = 0; |
8447 | if (_debug_mdata_eio(oid)) { | |
8448 | r = -EIO; | |
8449 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
8450 | } | |
8451 | return r; | |
8452 | } | |
8453 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 8454 | CollectionHandle& ch, |
7c673cae FG |
8455 | const pool_opts_t& opts) |
8456 | { | |
7c673cae | 8457 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 8458 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
8459 | if (!c->exists) |
8460 | return -ENOENT; | |
8461 | RWLock::WLocker l(c->lock); | |
8462 | c->pool_opts = opts; | |
8463 | return 0; | |
8464 | } | |
8465 | ||
7c673cae FG |
8466 | int BlueStore::read( |
8467 | CollectionHandle &c_, | |
8468 | const ghobject_t& oid, | |
8469 | uint64_t offset, | |
8470 | size_t length, | |
8471 | bufferlist& bl, | |
224ce89b | 8472 | uint32_t op_flags) |
7c673cae | 8473 | { |
11fdf7f2 | 8474 | auto start = mono_clock::now(); |
7c673cae FG |
8475 | Collection *c = static_cast<Collection *>(c_.get()); |
8476 | const coll_t &cid = c->get_cid(); | |
8477 | dout(15) << __func__ << " " << cid << " " << oid | |
8478 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
8479 | << dendl; | |
8480 | if (!c->exists) | |
8481 | return -ENOENT; | |
8482 | ||
8483 | bl.clear(); | |
8484 | int r; | |
8485 | { | |
8486 | RWLock::RLocker l(c->lock); | |
11fdf7f2 | 8487 | auto start1 = mono_clock::now(); |
7c673cae | 8488 | OnodeRef o = c->get_onode(oid, false); |
11fdf7f2 | 8489 | LOG_LATENCY(logger, cct, l_bluestore_read_onode_meta_lat, mono_clock::now() - start1); |
7c673cae FG |
8490 | if (!o || !o->exists) { |
8491 | r = -ENOENT; | |
8492 | goto out; | |
8493 | } | |
8494 | ||
8495 | if (offset == length && offset == 0) | |
8496 | length = o->onode.size; | |
8497 | ||
8498 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
8499 | if (r == -EIO) { |
8500 | logger->inc(l_bluestore_read_eio); | |
8501 | } | |
7c673cae FG |
8502 | } |
8503 | ||
8504 | out: | |
28e407b8 | 8505 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
8506 | r = -EIO; |
8507 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
8508 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
8509 | cct->_conf->bluestore_debug_random_read_err && | |
8510 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
8511 | 100.0)) == 0) { | |
224ce89b WB |
8512 | dout(0) << __func__ << ": inject random EIO" << dendl; |
8513 | r = -EIO; | |
7c673cae FG |
8514 | } |
8515 | dout(10) << __func__ << " " << cid << " " << oid | |
8516 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
8517 | << " = " << r << dendl; | |
11fdf7f2 | 8518 | LOG_LATENCY(logger, cct, l_bluestore_read_lat, mono_clock::now() - start); |
7c673cae FG |
8519 | return r; |
8520 | } | |
8521 | ||
8522 | // -------------------------------------------------------- | |
8523 | // intermediate data structures used while reading | |
8524 | struct region_t { | |
8525 | uint64_t logical_offset; | |
8526 | uint64_t blob_xoffset; //region offset within the blob | |
8527 | uint64_t length; | |
7c673cae FG |
8528 | |
8529 | // used later in read process | |
8530 | uint64_t front = 0; | |
7c673cae | 8531 | |
11fdf7f2 | 8532 | region_t(uint64_t offset, uint64_t b_offs, uint64_t len, uint64_t front = 0) |
7c673cae FG |
8533 | : logical_offset(offset), |
8534 | blob_xoffset(b_offs), | |
11fdf7f2 TL |
8535 | length(len), |
8536 | front(front){} | |
7c673cae FG |
8537 | region_t(const region_t& from) |
8538 | : logical_offset(from.logical_offset), | |
8539 | blob_xoffset(from.blob_xoffset), | |
11fdf7f2 TL |
8540 | length(from.length), |
8541 | front(from.front){} | |
7c673cae FG |
8542 | |
8543 | friend ostream& operator<<(ostream& out, const region_t& r) { | |
8544 | return out << "0x" << std::hex << r.logical_offset << ":" | |
8545 | << r.blob_xoffset << "~" << r.length << std::dec; | |
8546 | } | |
8547 | }; | |
8548 | ||
11fdf7f2 TL |
8549 | // merged blob read request |
8550 | struct read_req_t { | |
8551 | uint64_t r_off = 0; | |
8552 | uint64_t r_len = 0; | |
8553 | bufferlist bl; | |
8554 | std::list<region_t> regs; // original read regions | |
8555 | ||
8556 | read_req_t(uint64_t off, uint64_t len) : r_off(off), r_len(len) {} | |
8557 | ||
8558 | friend ostream& operator<<(ostream& out, const read_req_t& r) { | |
8559 | out << "{<0x" << std::hex << r.r_off << ", 0x" << r.r_len << "> : ["; | |
8560 | for (const auto& reg : r.regs) | |
8561 | out << reg; | |
8562 | return out << "]}" << std::dec; | |
8563 | } | |
8564 | }; | |
8565 | ||
8566 | typedef list<read_req_t> regions2read_t; | |
7c673cae FG |
8567 | typedef map<BlueStore::BlobRef, regions2read_t> blobs2read_t; |
8568 | ||
8569 | int BlueStore::_do_read( | |
8570 | Collection *c, | |
8571 | OnodeRef o, | |
8572 | uint64_t offset, | |
8573 | size_t length, | |
8574 | bufferlist& bl, | |
f64942e4 AA |
8575 | uint32_t op_flags, |
8576 | uint64_t retry_count) | |
7c673cae | 8577 | { |
11fdf7f2 | 8578 | FUNCTRACE(cct); |
7c673cae | 8579 | int r = 0; |
91327a77 | 8580 | int read_cache_policy = 0; // do not bypass clean or dirty cache |
7c673cae FG |
8581 | |
8582 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
8583 | << " size 0x" << o->onode.size << " (" << std::dec | |
8584 | << o->onode.size << ")" << dendl; | |
8585 | bl.clear(); | |
8586 | ||
8587 | if (offset >= o->onode.size) { | |
8588 | return r; | |
8589 | } | |
8590 | ||
8591 | // generally, don't buffer anything, unless the client explicitly requests | |
8592 | // it. | |
8593 | bool buffered = false; | |
8594 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
8595 | dout(20) << __func__ << " will do buffered read" << dendl; | |
8596 | buffered = true; | |
8597 | } else if (cct->_conf->bluestore_default_buffered_read && | |
8598 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
8599 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
8600 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
8601 | buffered = true; | |
8602 | } | |
8603 | ||
8604 | if (offset + length > o->onode.size) { | |
8605 | length = o->onode.size - offset; | |
8606 | } | |
8607 | ||
11fdf7f2 | 8608 | auto start = mono_clock::now(); |
7c673cae | 8609 | o->extent_map.fault_range(db, offset, length); |
11fdf7f2 | 8610 | LOG_LATENCY(logger, cct, l_bluestore_read_onode_meta_lat, mono_clock::now() - start); |
81eedcae | 8611 | _dump_onode<30>(cct, *o); |
7c673cae FG |
8612 | |
8613 | ready_regions_t ready_regions; | |
8614 | ||
91327a77 AA |
8615 | // for deep-scrub, we only read dirty cache and bypass clean cache in |
8616 | // order to read underlying block device in case there are silent disk errors. | |
8617 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
8618 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
8619 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
8620 | } | |
8621 | ||
7c673cae FG |
8622 | // build blob-wise list to of stuff read (that isn't cached) |
8623 | blobs2read_t blobs2read; | |
8624 | unsigned left = length; | |
8625 | uint64_t pos = offset; | |
8626 | unsigned num_regions = 0; | |
8627 | auto lp = o->extent_map.seek_lextent(offset); | |
8628 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
8629 | if (pos < lp->logical_offset) { | |
8630 | unsigned hole = lp->logical_offset - pos; | |
8631 | if (hole >= left) { | |
8632 | break; | |
8633 | } | |
8634 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
8635 | << std::dec << dendl; | |
8636 | pos += hole; | |
8637 | left -= hole; | |
8638 | } | |
94b18763 | 8639 | BlobRef& bptr = lp->blob; |
7c673cae FG |
8640 | unsigned l_off = pos - lp->logical_offset; |
8641 | unsigned b_off = l_off + lp->blob_offset; | |
8642 | unsigned b_len = std::min(left, lp->length - l_off); | |
8643 | ||
8644 | ready_regions_t cache_res; | |
8645 | interval_set<uint32_t> cache_interval; | |
8646 | bptr->shared_blob->bc.read( | |
91327a77 AA |
8647 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
8648 | read_cache_policy); | |
7c673cae FG |
8649 | dout(20) << __func__ << " blob " << *bptr << std::hex |
8650 | << " need 0x" << b_off << "~" << b_len | |
8651 | << " cache has 0x" << cache_interval | |
8652 | << std::dec << dendl; | |
8653 | ||
8654 | auto pc = cache_res.begin(); | |
11fdf7f2 | 8655 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
8656 | while (b_len > 0) { |
8657 | unsigned l; | |
8658 | if (pc != cache_res.end() && | |
8659 | pc->first == b_off) { | |
8660 | l = pc->second.length(); | |
8661 | ready_regions[pos].claim(pc->second); | |
8662 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" | |
8663 | << b_off << "~" << l << std::dec << dendl; | |
8664 | ++pc; | |
8665 | } else { | |
8666 | l = b_len; | |
8667 | if (pc != cache_res.end()) { | |
11fdf7f2 | 8668 | ceph_assert(pc->first > b_off); |
7c673cae FG |
8669 | l = pc->first - b_off; |
8670 | } | |
8671 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
8672 | << b_off << "~" << l << std::dec << dendl; | |
11fdf7f2 TL |
8673 | // merge regions |
8674 | { | |
8675 | uint64_t r_off = b_off; | |
8676 | uint64_t r_len = l; | |
8677 | uint64_t front = r_off % chunk_size; | |
8678 | if (front) { | |
8679 | r_off -= front; | |
8680 | r_len += front; | |
8681 | } | |
8682 | unsigned tail = r_len % chunk_size; | |
8683 | if (tail) { | |
8684 | r_len += chunk_size - tail; | |
8685 | } | |
8686 | bool merged = false; | |
8687 | regions2read_t& r2r = blobs2read[bptr]; | |
8688 | if (r2r.size()) { | |
8689 | read_req_t& pre = r2r.back(); | |
8690 | if (r_off <= (pre.r_off + pre.r_len)) { | |
8691 | front += (r_off - pre.r_off); | |
8692 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
8693 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
8694 | merged = true; | |
8695 | } | |
8696 | } | |
8697 | if (!merged) { | |
8698 | read_req_t req(r_off, r_len); | |
8699 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
8700 | r2r.emplace_back(std::move(req)); | |
8701 | } | |
8702 | } | |
7c673cae FG |
8703 | ++num_regions; |
8704 | } | |
8705 | pos += l; | |
8706 | b_off += l; | |
8707 | left -= l; | |
8708 | b_len -= l; | |
8709 | } | |
8710 | ++lp; | |
8711 | } | |
8712 | ||
8713 | // read raw blob data. use aio if we have >1 blobs to read. | |
11fdf7f2 TL |
8714 | start = mono_clock::now(); // for the sake of simplicity |
8715 | // measure the whole block below. | |
8716 | // The error isn't that much... | |
7c673cae | 8717 | vector<bufferlist> compressed_blob_bls; |
b32b8144 | 8718 | IOContext ioc(cct, NULL, true); // allow EIO |
7c673cae | 8719 | for (auto& p : blobs2read) { |
94b18763 | 8720 | const BlobRef& bptr = p.first; |
11fdf7f2 | 8721 | regions2read_t& r2r = p.second; |
7c673cae | 8722 | dout(20) << __func__ << " blob " << *bptr << std::hex |
11fdf7f2 | 8723 | << " need " << r2r << std::dec << dendl; |
7c673cae FG |
8724 | if (bptr->get_blob().is_compressed()) { |
8725 | // read the whole thing | |
8726 | if (compressed_blob_bls.empty()) { | |
8727 | // ensure we avoid any reallocation on subsequent blobs | |
8728 | compressed_blob_bls.reserve(blobs2read.size()); | |
8729 | } | |
8730 | compressed_blob_bls.push_back(bufferlist()); | |
8731 | bufferlist& bl = compressed_blob_bls.back(); | |
8732 | r = bptr->get_blob().map( | |
8733 | 0, bptr->get_blob().get_ondisk_length(), | |
8734 | [&](uint64_t offset, uint64_t length) { | |
8735 | int r; | |
8736 | // use aio if there are more regions to read than those in this blob | |
11fdf7f2 | 8737 | if (num_regions > r2r.size()) { |
7c673cae FG |
8738 | r = bdev->aio_read(offset, length, &bl, &ioc); |
8739 | } else { | |
8740 | r = bdev->read(offset, length, &bl, &ioc, false); | |
8741 | } | |
8742 | if (r < 0) | |
8743 | return r; | |
8744 | return 0; | |
8745 | }); | |
b32b8144 FG |
8746 | if (r < 0) { |
8747 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
8748 | if (r == -EIO) { | |
8749 | // propagate EIO to caller | |
8750 | return r; | |
8751 | } | |
11fdf7f2 | 8752 | ceph_assert(r == 0); |
b32b8144 | 8753 | } |
7c673cae FG |
8754 | } else { |
8755 | // read the pieces | |
11fdf7f2 | 8756 | for (auto& req : r2r) { |
7c673cae | 8757 | dout(20) << __func__ << " region 0x" << std::hex |
11fdf7f2 TL |
8758 | << req.regs.front().logical_offset |
8759 | << ": 0x" << req.regs.front().blob_xoffset | |
8760 | << " reading 0x" << req.r_off | |
8761 | << "~" << req.r_len << std::dec | |
7c673cae FG |
8762 | << dendl; |
8763 | ||
8764 | // read it | |
8765 | r = bptr->get_blob().map( | |
11fdf7f2 | 8766 | req.r_off, req.r_len, |
7c673cae FG |
8767 | [&](uint64_t offset, uint64_t length) { |
8768 | int r; | |
8769 | // use aio if there is more than one region to read | |
8770 | if (num_regions > 1) { | |
11fdf7f2 | 8771 | r = bdev->aio_read(offset, length, &req.bl, &ioc); |
7c673cae | 8772 | } else { |
11fdf7f2 | 8773 | r = bdev->read(offset, length, &req.bl, &ioc, false); |
7c673cae FG |
8774 | } |
8775 | if (r < 0) | |
8776 | return r; | |
8777 | return 0; | |
8778 | }); | |
b32b8144 FG |
8779 | if (r < 0) { |
8780 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
8781 | << dendl; | |
8782 | if (r == -EIO) { | |
8783 | // propagate EIO to caller | |
8784 | return r; | |
8785 | } | |
11fdf7f2 | 8786 | ceph_assert(r == 0); |
b32b8144 | 8787 | } |
11fdf7f2 | 8788 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
8789 | } |
8790 | } | |
8791 | } | |
11fdf7f2 TL |
8792 | |
8793 | int64_t num_ios = length; | |
7c673cae | 8794 | if (ioc.has_pending_aios()) { |
11fdf7f2 | 8795 | num_ios = -ioc.get_num_ios(); |
7c673cae FG |
8796 | bdev->aio_submit(&ioc); |
8797 | dout(20) << __func__ << " waiting for aio" << dendl; | |
8798 | ioc.aio_wait(); | |
b32b8144 FG |
8799 | r = ioc.get_return_value(); |
8800 | if (r < 0) { | |
11fdf7f2 | 8801 | ceph_assert(r == -EIO); // no other errors allowed |
b32b8144 FG |
8802 | return -EIO; |
8803 | } | |
7c673cae | 8804 | } |
11fdf7f2 TL |
8805 | LOG_LATENCY_FN(logger, cct, l_bluestore_read_wait_aio_lat, |
8806 | mono_clock::now() - start, | |
8807 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
8808 | ); | |
7c673cae FG |
8809 | |
8810 | // enumerate and decompress desired blobs | |
8811 | auto p = compressed_blob_bls.begin(); | |
8812 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
8813 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 8814 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 8815 | regions2read_t& r2r = b2r_it->second; |
7c673cae | 8816 | dout(20) << __func__ << " blob " << *bptr << std::hex |
11fdf7f2 | 8817 | << " need 0x" << r2r << std::dec << dendl; |
7c673cae | 8818 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 8819 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
8820 | bufferlist& compressed_bl = *p++; |
8821 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
11fdf7f2 | 8822 | r2r.front().regs.front().logical_offset) < 0) { |
f64942e4 AA |
8823 | // Handles spurious read errors caused by a kernel bug. |
8824 | // We sometimes get all-zero pages as a result of the read under | |
11fdf7f2 TL |
8825 | // high memory pressure. Retrying the failing read succeeds in most |
8826 | // cases. | |
f64942e4 AA |
8827 | // See also: http://tracker.ceph.com/issues/22464 |
8828 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
8829 | return -EIO; | |
8830 | } | |
8831 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
7c673cae FG |
8832 | } |
8833 | bufferlist raw_bl; | |
8834 | r = _decompress(compressed_bl, &raw_bl); | |
8835 | if (r < 0) | |
8836 | return r; | |
8837 | if (buffered) { | |
8838 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, | |
8839 | raw_bl); | |
8840 | } | |
11fdf7f2 TL |
8841 | for (auto& req : r2r) { |
8842 | for (auto& r : req.regs) { | |
8843 | ready_regions[r.logical_offset].substr_of( | |
8844 | raw_bl, r.blob_xoffset, r.length); | |
8845 | } | |
7c673cae FG |
8846 | } |
8847 | } else { | |
11fdf7f2 TL |
8848 | for (auto& req : r2r) { |
8849 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, | |
8850 | req.regs.front().logical_offset) < 0) { | |
f64942e4 AA |
8851 | // Handles spurious read errors caused by a kernel bug. |
8852 | // We sometimes get all-zero pages as a result of the read under | |
11fdf7f2 TL |
8853 | // high memory pressure. Retrying the failing read succeeds in most |
8854 | // cases. | |
f64942e4 AA |
8855 | // See also: http://tracker.ceph.com/issues/22464 |
8856 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
8857 | return -EIO; | |
8858 | } | |
8859 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
7c673cae FG |
8860 | } |
8861 | if (buffered) { | |
8862 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
11fdf7f2 | 8863 | req.r_off, req.bl); |
7c673cae FG |
8864 | } |
8865 | ||
8866 | // prune and keep result | |
11fdf7f2 TL |
8867 | for (const auto& r : req.regs) { |
8868 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
8869 | } | |
7c673cae FG |
8870 | } |
8871 | } | |
8872 | ++b2r_it; | |
8873 | } | |
8874 | ||
8875 | // generate a resulting buffer | |
8876 | auto pr = ready_regions.begin(); | |
8877 | auto pr_end = ready_regions.end(); | |
8878 | pos = 0; | |
8879 | while (pos < length) { | |
8880 | if (pr != pr_end && pr->first == pos + offset) { | |
8881 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
8882 | << ": data from 0x" << pr->first << "~" << pr->second.length() | |
8883 | << std::dec << dendl; | |
8884 | pos += pr->second.length(); | |
8885 | bl.claim_append(pr->second); | |
8886 | ++pr; | |
8887 | } else { | |
8888 | uint64_t l = length - pos; | |
8889 | if (pr != pr_end) { | |
11fdf7f2 | 8890 | ceph_assert(pr->first > pos + offset); |
7c673cae FG |
8891 | l = pr->first - (pos + offset); |
8892 | } | |
8893 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
8894 | << ": zeros for 0x" << (pos + offset) << "~" << l | |
8895 | << std::dec << dendl; | |
8896 | bl.append_zero(l); | |
8897 | pos += l; | |
8898 | } | |
8899 | } | |
11fdf7f2 TL |
8900 | ceph_assert(bl.length() == length); |
8901 | ceph_assert(pos == length); | |
8902 | ceph_assert(pr == pr_end); | |
7c673cae | 8903 | r = bl.length(); |
f64942e4 AA |
8904 | if (retry_count) { |
8905 | logger->inc(l_bluestore_reads_with_retries); | |
8906 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
8907 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
8908 | } | |
7c673cae FG |
8909 | return r; |
8910 | } | |
8911 | ||
8912 | int BlueStore::_verify_csum(OnodeRef& o, | |
8913 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
8914 | const bufferlist& bl, | |
8915 | uint64_t logical_offset) const | |
8916 | { | |
8917 | int bad; | |
8918 | uint64_t bad_csum; | |
11fdf7f2 | 8919 | auto start = mono_clock::now(); |
7c673cae | 8920 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
8921 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
8922 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
8923 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
8924 | bad = blob_xoffset; | |
8925 | r = -1; | |
8926 | bad_csum = 0xDEADBEEF; | |
8927 | } | |
7c673cae FG |
8928 | if (r < 0) { |
8929 | if (r == -1) { | |
8930 | PExtentVector pex; | |
8931 | blob->map( | |
8932 | bad, | |
8933 | blob->get_csum_chunk_size(), | |
8934 | [&](uint64_t offset, uint64_t length) { | |
8935 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
8936 | return 0; | |
8937 | }); | |
8938 | derr << __func__ << " bad " | |
8939 | << Checksummer::get_csum_type_string(blob->csum_type) | |
8940 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
8941 | << " checksum at blob offset 0x" << bad | |
8942 | << ", got 0x" << bad_csum << ", expected 0x" | |
8943 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
8944 | << ", device location " << pex | |
8945 | << ", logical extent 0x" << std::hex | |
8946 | << (logical_offset + bad - blob_xoffset) << "~" | |
8947 | << blob->get_csum_chunk_size() << std::dec | |
8948 | << ", object " << o->oid | |
8949 | << dendl; | |
8950 | } else { | |
8951 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
8952 | } | |
8953 | } | |
11fdf7f2 TL |
8954 | LOG_LATENCY(logger, cct, l_bluestore_csum_lat, mono_clock::now() - start); |
8955 | if (cct->_conf->bluestore_ignore_data_csum) { | |
8956 | return 0; | |
8957 | } | |
7c673cae FG |
8958 | return r; |
8959 | } | |
8960 | ||
8961 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
8962 | { | |
8963 | int r = 0; | |
11fdf7f2 TL |
8964 | auto start = mono_clock::now(); |
8965 | auto i = source.cbegin(); | |
7c673cae | 8966 | bluestore_compression_header_t chdr; |
11fdf7f2 | 8967 | decode(chdr, i); |
7c673cae FG |
8968 | int alg = int(chdr.type); |
8969 | CompressorRef cp = compressor; | |
8970 | if (!cp || (int)cp->get_type() != alg) { | |
8971 | cp = Compressor::create(cct, alg); | |
8972 | } | |
8973 | ||
8974 | if (!cp.get()) { | |
8975 | // if compressor isn't available - error, because cannot return | |
8976 | // decompressed data? | |
11fdf7f2 TL |
8977 | |
8978 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
8979 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
8980 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
8981 | r = -EIO; |
8982 | } else { | |
8983 | r = cp->decompress(i, chdr.length, *result); | |
8984 | if (r < 0) { | |
8985 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
8986 | r = -EIO; | |
8987 | } | |
8988 | } | |
11fdf7f2 | 8989 | LOG_LATENCY(logger, cct, l_bluestore_decompress_lat, mono_clock::now() - start); |
7c673cae FG |
8990 | return r; |
8991 | } | |
8992 | ||
8993 | // this stores fiemap into interval_set, other variations | |
8994 | // use it internally | |
8995 | int BlueStore::_fiemap( | |
8996 | CollectionHandle &c_, | |
8997 | const ghobject_t& oid, | |
8998 | uint64_t offset, | |
8999 | size_t length, | |
9000 | interval_set<uint64_t>& destset) | |
9001 | { | |
9002 | Collection *c = static_cast<Collection *>(c_.get()); | |
9003 | if (!c->exists) | |
9004 | return -ENOENT; | |
9005 | { | |
9006 | RWLock::RLocker l(c->lock); | |
9007 | ||
9008 | OnodeRef o = c->get_onode(oid, false); | |
9009 | if (!o || !o->exists) { | |
9010 | return -ENOENT; | |
9011 | } | |
81eedcae | 9012 | _dump_onode<30>(cct, *o); |
7c673cae FG |
9013 | |
9014 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
9015 | << " size 0x" << o->onode.size << std::dec << dendl; | |
9016 | ||
9017 | boost::intrusive::set<Extent>::iterator ep, eend; | |
9018 | if (offset >= o->onode.size) | |
9019 | goto out; | |
9020 | ||
9021 | if (offset + length > o->onode.size) { | |
9022 | length = o->onode.size - offset; | |
9023 | } | |
9024 | ||
9025 | o->extent_map.fault_range(db, offset, length); | |
9026 | eend = o->extent_map.extent_map.end(); | |
9027 | ep = o->extent_map.seek_lextent(offset); | |
9028 | while (length > 0) { | |
9029 | dout(20) << __func__ << " offset " << offset << dendl; | |
9030 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
9031 | ++ep; | |
9032 | continue; | |
9033 | } | |
9034 | ||
9035 | uint64_t x_len = length; | |
9036 | if (ep != eend && ep->logical_offset <= offset) { | |
9037 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 9038 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
9039 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
9040 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
9041 | destset.insert(offset, x_len); | |
9042 | length -= x_len; | |
9043 | offset += x_len; | |
9044 | if (x_off + x_len == ep->length) | |
9045 | ++ep; | |
9046 | continue; | |
9047 | } | |
9048 | if (ep != eend && | |
9049 | ep->logical_offset > offset && | |
9050 | ep->logical_offset - offset < x_len) { | |
9051 | x_len = ep->logical_offset - offset; | |
9052 | } | |
9053 | offset += x_len; | |
9054 | length -= x_len; | |
9055 | } | |
9056 | } | |
9057 | ||
9058 | out: | |
7c673cae FG |
9059 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
9060 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
9061 | return 0; | |
9062 | } | |
9063 | ||
7c673cae FG |
9064 | int BlueStore::fiemap( |
9065 | CollectionHandle &c_, | |
9066 | const ghobject_t& oid, | |
9067 | uint64_t offset, | |
9068 | size_t length, | |
9069 | bufferlist& bl) | |
9070 | { | |
9071 | interval_set<uint64_t> m; | |
9072 | int r = _fiemap(c_, oid, offset, length, m); | |
9073 | if (r >= 0) { | |
11fdf7f2 | 9074 | encode(m, bl); |
7c673cae FG |
9075 | } |
9076 | return r; | |
9077 | } | |
9078 | ||
7c673cae FG |
9079 | int BlueStore::fiemap( |
9080 | CollectionHandle &c_, | |
9081 | const ghobject_t& oid, | |
9082 | uint64_t offset, | |
9083 | size_t length, | |
9084 | map<uint64_t, uint64_t>& destmap) | |
9085 | { | |
9086 | interval_set<uint64_t> m; | |
9087 | int r = _fiemap(c_, oid, offset, length, m); | |
9088 | if (r >= 0) { | |
9089 | m.move_into(destmap); | |
9090 | } | |
9091 | return r; | |
9092 | } | |
9093 | ||
7c673cae FG |
9094 | int BlueStore::getattr( |
9095 | CollectionHandle &c_, | |
9096 | const ghobject_t& oid, | |
9097 | const char *name, | |
9098 | bufferptr& value) | |
9099 | { | |
9100 | Collection *c = static_cast<Collection *>(c_.get()); | |
9101 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
9102 | if (!c->exists) | |
9103 | return -ENOENT; | |
9104 | ||
9105 | int r; | |
9106 | { | |
9107 | RWLock::RLocker l(c->lock); | |
31f18b77 | 9108 | mempool::bluestore_cache_other::string k(name); |
7c673cae FG |
9109 | |
9110 | OnodeRef o = c->get_onode(oid, false); | |
9111 | if (!o || !o->exists) { | |
9112 | r = -ENOENT; | |
9113 | goto out; | |
9114 | } | |
9115 | ||
9116 | if (!o->onode.attrs.count(k)) { | |
9117 | r = -ENODATA; | |
9118 | goto out; | |
9119 | } | |
9120 | value = o->onode.attrs[k]; | |
9121 | r = 0; | |
9122 | } | |
9123 | out: | |
7c673cae FG |
9124 | if (r == 0 && _debug_mdata_eio(oid)) { |
9125 | r = -EIO; | |
9126 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9127 | } | |
9128 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
9129 | << " = " << r << dendl; | |
9130 | return r; | |
9131 | } | |
9132 | ||
7c673cae FG |
9133 | int BlueStore::getattrs( |
9134 | CollectionHandle &c_, | |
9135 | const ghobject_t& oid, | |
9136 | map<string,bufferptr>& aset) | |
9137 | { | |
9138 | Collection *c = static_cast<Collection *>(c_.get()); | |
9139 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
9140 | if (!c->exists) | |
9141 | return -ENOENT; | |
9142 | ||
9143 | int r; | |
9144 | { | |
9145 | RWLock::RLocker l(c->lock); | |
9146 | ||
9147 | OnodeRef o = c->get_onode(oid, false); | |
9148 | if (!o || !o->exists) { | |
9149 | r = -ENOENT; | |
9150 | goto out; | |
9151 | } | |
9152 | for (auto& i : o->onode.attrs) { | |
9153 | aset.emplace(i.first.c_str(), i.second); | |
9154 | } | |
9155 | r = 0; | |
9156 | } | |
9157 | ||
9158 | out: | |
7c673cae FG |
9159 | if (r == 0 && _debug_mdata_eio(oid)) { |
9160 | r = -EIO; | |
9161 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9162 | } | |
9163 | dout(10) << __func__ << " " << c->cid << " " << oid | |
9164 | << " = " << r << dendl; | |
9165 | return r; | |
9166 | } | |
9167 | ||
9168 | int BlueStore::list_collections(vector<coll_t>& ls) | |
9169 | { | |
9170 | RWLock::RLocker l(coll_lock); | |
11fdf7f2 | 9171 | ls.reserve(coll_map.size()); |
7c673cae FG |
9172 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
9173 | p != coll_map.end(); | |
9174 | ++p) | |
9175 | ls.push_back(p->first); | |
9176 | return 0; | |
9177 | } | |
9178 | ||
9179 | bool BlueStore::collection_exists(const coll_t& c) | |
9180 | { | |
9181 | RWLock::RLocker l(coll_lock); | |
9182 | return coll_map.count(c); | |
9183 | } | |
9184 | ||
11fdf7f2 | 9185 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 9186 | { |
11fdf7f2 | 9187 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
9188 | vector<ghobject_t> ls; |
9189 | ghobject_t next; | |
11fdf7f2 | 9190 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
9191 | &ls, &next); |
9192 | if (r < 0) { | |
9193 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
9194 | << dendl; | |
9195 | return r; | |
9196 | } | |
9197 | *empty = ls.empty(); | |
11fdf7f2 | 9198 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
9199 | return 0; |
9200 | } | |
9201 | ||
11fdf7f2 | 9202 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 9203 | { |
11fdf7f2 TL |
9204 | dout(15) << __func__ << " " << ch->cid << dendl; |
9205 | Collection *c = static_cast<Collection*>(ch.get()); | |
7c673cae | 9206 | RWLock::RLocker l(c->lock); |
11fdf7f2 | 9207 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
9208 | return c->cnode.bits; |
9209 | } | |
9210 | ||
7c673cae FG |
9211 | int BlueStore::collection_list( |
9212 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
9213 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
9214 | { | |
9215 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 9216 | c->flush(); |
7c673cae FG |
9217 | dout(15) << __func__ << " " << c->cid |
9218 | << " start " << start << " end " << end << " max " << max << dendl; | |
9219 | int r; | |
9220 | { | |
9221 | RWLock::RLocker l(c->lock); | |
9222 | r = _collection_list(c, start, end, max, ls, pnext); | |
9223 | } | |
9224 | ||
7c673cae FG |
9225 | dout(10) << __func__ << " " << c->cid |
9226 | << " start " << start << " end " << end << " max " << max | |
9227 | << " = " << r << ", ls.size() = " << ls->size() | |
9228 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
9229 | return r; | |
9230 | } | |
9231 | ||
9232 | int BlueStore::_collection_list( | |
9233 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
9234 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
9235 | { | |
9236 | ||
9237 | if (!c->exists) | |
9238 | return -ENOENT; | |
9239 | ||
9240 | int r = 0; | |
9241 | ghobject_t static_next; | |
9242 | KeyValueDB::Iterator it; | |
9243 | string temp_start_key, temp_end_key; | |
9244 | string start_key, end_key; | |
9245 | bool set_next = false; | |
9246 | string pend; | |
9247 | bool temp; | |
9248 | ||
9249 | if (!pnext) | |
9250 | pnext = &static_next; | |
9251 | ||
11fdf7f2 | 9252 | if (start.is_max() || start.hobj.is_max()) { |
7c673cae FG |
9253 | goto out; |
9254 | } | |
9255 | get_coll_key_range(c->cid, c->cnode.bits, &temp_start_key, &temp_end_key, | |
9256 | &start_key, &end_key); | |
9257 | dout(20) << __func__ | |
9258 | << " range " << pretty_binary_string(temp_start_key) | |
9259 | << " to " << pretty_binary_string(temp_end_key) | |
9260 | << " and " << pretty_binary_string(start_key) | |
9261 | << " to " << pretty_binary_string(end_key) | |
9262 | << " start " << start << dendl; | |
9263 | it = db->get_iterator(PREFIX_OBJ); | |
9264 | if (start == ghobject_t() || | |
9265 | start.hobj == hobject_t() || | |
9266 | start == c->cid.get_min_hobj()) { | |
9267 | it->upper_bound(temp_start_key); | |
9268 | temp = true; | |
9269 | } else { | |
9270 | string k; | |
9271 | get_object_key(cct, start, &k); | |
9272 | if (start.hobj.is_temp()) { | |
9273 | temp = true; | |
11fdf7f2 | 9274 | ceph_assert(k >= temp_start_key && k < temp_end_key); |
7c673cae FG |
9275 | } else { |
9276 | temp = false; | |
11fdf7f2 | 9277 | ceph_assert(k >= start_key && k < end_key); |
7c673cae | 9278 | } |
11fdf7f2 | 9279 | dout(20) << __func__ << " start from " << pretty_binary_string(k) |
7c673cae FG |
9280 | << " temp=" << (int)temp << dendl; |
9281 | it->lower_bound(k); | |
9282 | } | |
9283 | if (end.hobj.is_max()) { | |
9284 | pend = temp ? temp_end_key : end_key; | |
9285 | } else { | |
9286 | get_object_key(cct, end, &end_key); | |
9287 | if (end.hobj.is_temp()) { | |
9288 | if (temp) | |
9289 | pend = end_key; | |
9290 | else | |
9291 | goto out; | |
9292 | } else { | |
9293 | pend = temp ? temp_end_key : end_key; | |
9294 | } | |
9295 | } | |
9296 | dout(20) << __func__ << " pend " << pretty_binary_string(pend) << dendl; | |
9297 | while (true) { | |
9298 | if (!it->valid() || it->key() >= pend) { | |
9299 | if (!it->valid()) | |
9300 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
9301 | else | |
9302 | dout(20) << __func__ << " key " << pretty_binary_string(it->key()) | |
9303 | << " >= " << end << dendl; | |
9304 | if (temp) { | |
9305 | if (end.hobj.is_temp()) { | |
9306 | break; | |
9307 | } | |
9308 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
9309 | temp = false; | |
9310 | it->upper_bound(start_key); | |
9311 | pend = end_key; | |
9312 | dout(30) << __func__ << " pend " << pretty_binary_string(pend) << dendl; | |
9313 | continue; | |
9314 | } | |
9315 | break; | |
9316 | } | |
9317 | dout(30) << __func__ << " key " << pretty_binary_string(it->key()) << dendl; | |
9318 | if (is_extent_shard_key(it->key())) { | |
9319 | it->next(); | |
9320 | continue; | |
9321 | } | |
9322 | ghobject_t oid; | |
9323 | int r = get_key_object(it->key(), &oid); | |
11fdf7f2 | 9324 | ceph_assert(r == 0); |
7c673cae FG |
9325 | dout(20) << __func__ << " oid " << oid << " end " << end << dendl; |
9326 | if (ls->size() >= (unsigned)max) { | |
9327 | dout(20) << __func__ << " reached max " << max << dendl; | |
9328 | *pnext = oid; | |
9329 | set_next = true; | |
9330 | break; | |
9331 | } | |
9332 | ls->push_back(oid); | |
9333 | it->next(); | |
9334 | } | |
9335 | out: | |
9336 | if (!set_next) { | |
9337 | *pnext = ghobject_t::get_max(); | |
9338 | } | |
9339 | ||
9340 | return r; | |
9341 | } | |
9342 | ||
7c673cae FG |
9343 | int BlueStore::omap_get( |
9344 | CollectionHandle &c_, ///< [in] Collection containing oid | |
9345 | const ghobject_t &oid, ///< [in] Object containing omap | |
9346 | bufferlist *header, ///< [out] omap header | |
9347 | map<string, bufferlist> *out /// < [out] Key to value map | |
9348 | ) | |
9349 | { | |
9350 | Collection *c = static_cast<Collection *>(c_.get()); | |
9351 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
9352 | if (!c->exists) | |
9353 | return -ENOENT; | |
9354 | RWLock::RLocker l(c->lock); | |
9355 | int r = 0; | |
9356 | OnodeRef o = c->get_onode(oid, false); | |
9357 | if (!o || !o->exists) { | |
9358 | r = -ENOENT; | |
9359 | goto out; | |
9360 | } | |
9361 | if (!o->onode.has_omap()) | |
9362 | goto out; | |
9363 | o->flush(); | |
9364 | { | |
11fdf7f2 TL |
9365 | const string& prefix = |
9366 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
9367 | KeyValueDB::Iterator it = db->get_iterator(prefix); | |
7c673cae FG |
9368 | string head, tail; |
9369 | get_omap_header(o->onode.nid, &head); | |
9370 | get_omap_tail(o->onode.nid, &tail); | |
9371 | it->lower_bound(head); | |
9372 | while (it->valid()) { | |
9373 | if (it->key() == head) { | |
9374 | dout(30) << __func__ << " got header" << dendl; | |
9375 | *header = it->value(); | |
9376 | } else if (it->key() >= tail) { | |
9377 | dout(30) << __func__ << " reached tail" << dendl; | |
9378 | break; | |
9379 | } else { | |
9380 | string user_key; | |
9381 | decode_omap_key(it->key(), &user_key); | |
11fdf7f2 | 9382 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
9383 | << " -> " << user_key << dendl; |
9384 | (*out)[user_key] = it->value(); | |
9385 | } | |
9386 | it->next(); | |
9387 | } | |
9388 | } | |
9389 | out: | |
9390 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
9391 | << dendl; | |
9392 | return r; | |
9393 | } | |
9394 | ||
7c673cae FG |
9395 | int BlueStore::omap_get_header( |
9396 | CollectionHandle &c_, ///< [in] Collection containing oid | |
9397 | const ghobject_t &oid, ///< [in] Object containing omap | |
9398 | bufferlist *header, ///< [out] omap header | |
9399 | bool allow_eio ///< [in] don't assert on eio | |
9400 | ) | |
9401 | { | |
9402 | Collection *c = static_cast<Collection *>(c_.get()); | |
9403 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
9404 | if (!c->exists) | |
9405 | return -ENOENT; | |
9406 | RWLock::RLocker l(c->lock); | |
9407 | int r = 0; | |
9408 | OnodeRef o = c->get_onode(oid, false); | |
9409 | if (!o || !o->exists) { | |
9410 | r = -ENOENT; | |
9411 | goto out; | |
9412 | } | |
9413 | if (!o->onode.has_omap()) | |
9414 | goto out; | |
9415 | o->flush(); | |
9416 | { | |
9417 | string head; | |
9418 | get_omap_header(o->onode.nid, &head); | |
11fdf7f2 TL |
9419 | if (db->get(o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP, |
9420 | head, header) >= 0) { | |
7c673cae FG |
9421 | dout(30) << __func__ << " got header" << dendl; |
9422 | } else { | |
9423 | dout(30) << __func__ << " no header" << dendl; | |
9424 | } | |
9425 | } | |
9426 | out: | |
9427 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
9428 | << dendl; | |
9429 | return r; | |
9430 | } | |
9431 | ||
7c673cae FG |
9432 | int BlueStore::omap_get_keys( |
9433 | CollectionHandle &c_, ///< [in] Collection containing oid | |
9434 | const ghobject_t &oid, ///< [in] Object containing omap | |
9435 | set<string> *keys ///< [out] Keys defined on oid | |
9436 | ) | |
9437 | { | |
9438 | Collection *c = static_cast<Collection *>(c_.get()); | |
9439 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
9440 | if (!c->exists) | |
9441 | return -ENOENT; | |
9442 | RWLock::RLocker l(c->lock); | |
9443 | int r = 0; | |
9444 | OnodeRef o = c->get_onode(oid, false); | |
9445 | if (!o || !o->exists) { | |
9446 | r = -ENOENT; | |
9447 | goto out; | |
9448 | } | |
9449 | if (!o->onode.has_omap()) | |
9450 | goto out; | |
9451 | o->flush(); | |
9452 | { | |
11fdf7f2 TL |
9453 | const string& prefix = |
9454 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
9455 | KeyValueDB::Iterator it = db->get_iterator(prefix); | |
7c673cae FG |
9456 | string head, tail; |
9457 | get_omap_key(o->onode.nid, string(), &head); | |
9458 | get_omap_tail(o->onode.nid, &tail); | |
9459 | it->lower_bound(head); | |
9460 | while (it->valid()) { | |
9461 | if (it->key() >= tail) { | |
9462 | dout(30) << __func__ << " reached tail" << dendl; | |
9463 | break; | |
9464 | } | |
9465 | string user_key; | |
9466 | decode_omap_key(it->key(), &user_key); | |
11fdf7f2 | 9467 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
9468 | << " -> " << user_key << dendl; |
9469 | keys->insert(user_key); | |
9470 | it->next(); | |
11fdf7f2 TL |
9471 | } |
9472 | } | |
9473 | out: | |
9474 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
9475 | << dendl; | |
9476 | return r; | |
7c673cae FG |
9477 | } |
9478 | ||
9479 | int BlueStore::omap_get_values( | |
9480 | CollectionHandle &c_, ///< [in] Collection containing oid | |
9481 | const ghobject_t &oid, ///< [in] Object containing omap | |
9482 | const set<string> &keys, ///< [in] Keys to get | |
9483 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
9484 | ) | |
9485 | { | |
9486 | Collection *c = static_cast<Collection *>(c_.get()); | |
9487 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
9488 | if (!c->exists) | |
9489 | return -ENOENT; | |
9490 | RWLock::RLocker l(c->lock); | |
9491 | int r = 0; | |
9492 | string final_key; | |
9493 | OnodeRef o = c->get_onode(oid, false); | |
9494 | if (!o || !o->exists) { | |
9495 | r = -ENOENT; | |
9496 | goto out; | |
9497 | } | |
9498 | if (!o->onode.has_omap()) | |
9499 | goto out; | |
11fdf7f2 TL |
9500 | { |
9501 | const string& prefix = | |
9502 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
9503 | o->flush(); | |
9504 | _key_encode_u64(o->onode.nid, &final_key); | |
9505 | final_key.push_back('.'); | |
9506 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { | |
9507 | final_key.resize(9); // keep prefix | |
9508 | final_key += *p; | |
9509 | bufferlist val; | |
9510 | if (db->get(prefix, final_key, &val) >= 0) { | |
9511 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
9512 | << " -> " << *p << dendl; | |
9513 | out->insert(make_pair(*p, val)); | |
9514 | } | |
7c673cae FG |
9515 | } |
9516 | } | |
9517 | out: | |
9518 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
9519 | << dendl; | |
9520 | return r; | |
9521 | } | |
9522 | ||
7c673cae FG |
9523 | int BlueStore::omap_check_keys( |
9524 | CollectionHandle &c_, ///< [in] Collection containing oid | |
9525 | const ghobject_t &oid, ///< [in] Object containing omap | |
9526 | const set<string> &keys, ///< [in] Keys to check | |
9527 | set<string> *out ///< [out] Subset of keys defined on oid | |
9528 | ) | |
9529 | { | |
9530 | Collection *c = static_cast<Collection *>(c_.get()); | |
9531 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
9532 | if (!c->exists) | |
9533 | return -ENOENT; | |
9534 | RWLock::RLocker l(c->lock); | |
9535 | int r = 0; | |
9536 | string final_key; | |
9537 | OnodeRef o = c->get_onode(oid, false); | |
9538 | if (!o || !o->exists) { | |
9539 | r = -ENOENT; | |
9540 | goto out; | |
9541 | } | |
9542 | if (!o->onode.has_omap()) | |
9543 | goto out; | |
11fdf7f2 TL |
9544 | { |
9545 | const string& prefix = | |
9546 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
9547 | o->flush(); | |
9548 | _key_encode_u64(o->onode.nid, &final_key); | |
9549 | final_key.push_back('.'); | |
9550 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { | |
9551 | final_key.resize(9); // keep prefix | |
9552 | final_key += *p; | |
9553 | bufferlist val; | |
9554 | if (db->get(prefix, final_key, &val) >= 0) { | |
9555 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
9556 | << " -> " << *p << dendl; | |
9557 | out->insert(*p); | |
9558 | } else { | |
9559 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
9560 | << " -> " << *p << dendl; | |
9561 | } | |
7c673cae FG |
9562 | } |
9563 | } | |
9564 | out: | |
9565 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
9566 | << dendl; | |
9567 | return r; | |
9568 | } | |
9569 | ||
7c673cae FG |
9570 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
9571 | CollectionHandle &c_, ///< [in] collection | |
9572 | const ghobject_t &oid ///< [in] object | |
9573 | ) | |
9574 | { | |
9575 | Collection *c = static_cast<Collection *>(c_.get()); | |
9576 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
9577 | if (!c->exists) { | |
9578 | return ObjectMap::ObjectMapIterator(); | |
9579 | } | |
9580 | RWLock::RLocker l(c->lock); | |
9581 | OnodeRef o = c->get_onode(oid, false); | |
9582 | if (!o || !o->exists) { | |
9583 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
9584 | return ObjectMap::ObjectMapIterator(); | |
9585 | } | |
9586 | o->flush(); | |
9587 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
11fdf7f2 TL |
9588 | KeyValueDB::Iterator it = db->get_iterator( |
9589 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP); | |
7c673cae FG |
9590 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(c, o, it)); |
9591 | } | |
9592 | ||
9593 | // ----------------- | |
9594 | // write helpers | |
9595 | ||
11fdf7f2 TL |
9596 | uint64_t BlueStore::_get_ondisk_reserved() const { |
9597 | return round_up_to( | |
9598 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
9599 | } | |
9600 | ||
7c673cae FG |
9601 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
9602 | { | |
9603 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
9604 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
9605 | << dendl; | |
11fdf7f2 | 9606 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
9607 | { |
9608 | bufferlist bl; | |
11fdf7f2 | 9609 | encode(ondisk_format, bl); |
7c673cae FG |
9610 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
9611 | } | |
9612 | { | |
9613 | bufferlist bl; | |
11fdf7f2 | 9614 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
9615 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
9616 | } | |
9617 | } | |
9618 | ||
9619 | int BlueStore::_open_super_meta() | |
9620 | { | |
9621 | // nid | |
9622 | { | |
9623 | nid_max = 0; | |
9624 | bufferlist bl; | |
9625 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 9626 | auto p = bl.cbegin(); |
7c673cae FG |
9627 | try { |
9628 | uint64_t v; | |
11fdf7f2 | 9629 | decode(v, p); |
7c673cae FG |
9630 | nid_max = v; |
9631 | } catch (buffer::error& e) { | |
9632 | derr << __func__ << " unable to read nid_max" << dendl; | |
9633 | return -EIO; | |
9634 | } | |
9635 | dout(10) << __func__ << " old nid_max " << nid_max << dendl; | |
9636 | nid_last = nid_max.load(); | |
9637 | } | |
9638 | ||
9639 | // blobid | |
9640 | { | |
9641 | blobid_max = 0; | |
9642 | bufferlist bl; | |
9643 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 9644 | auto p = bl.cbegin(); |
7c673cae FG |
9645 | try { |
9646 | uint64_t v; | |
11fdf7f2 | 9647 | decode(v, p); |
7c673cae FG |
9648 | blobid_max = v; |
9649 | } catch (buffer::error& e) { | |
9650 | derr << __func__ << " unable to read blobid_max" << dendl; | |
9651 | return -EIO; | |
9652 | } | |
9653 | dout(10) << __func__ << " old blobid_max " << blobid_max << dendl; | |
9654 | blobid_last = blobid_max.load(); | |
9655 | } | |
9656 | ||
9657 | // freelist | |
9658 | { | |
9659 | bufferlist bl; | |
9660 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
9661 | if (bl.length()) { | |
9662 | freelist_type = std::string(bl.c_str(), bl.length()); | |
9663 | dout(10) << __func__ << " freelist_type " << freelist_type << dendl; | |
9664 | } else { | |
11fdf7f2 | 9665 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 9666 | } |
7c673cae FG |
9667 | } |
9668 | ||
9669 | // ondisk format | |
9670 | int32_t compat_ondisk_format = 0; | |
9671 | { | |
9672 | bufferlist bl; | |
9673 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
9674 | if (r < 0) { | |
9675 | // base case: kraken bluestore is v1 and readable by v1 | |
9676 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
9677 | << dendl; | |
9678 | ondisk_format = 1; | |
9679 | compat_ondisk_format = 1; | |
9680 | } else { | |
11fdf7f2 | 9681 | auto p = bl.cbegin(); |
7c673cae | 9682 | try { |
11fdf7f2 | 9683 | decode(ondisk_format, p); |
7c673cae FG |
9684 | } catch (buffer::error& e) { |
9685 | derr << __func__ << " unable to read ondisk_format" << dendl; | |
9686 | return -EIO; | |
9687 | } | |
9688 | bl.clear(); | |
9689 | { | |
9690 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
9691 | ceph_assert(!r); |
9692 | auto p = bl.cbegin(); | |
7c673cae | 9693 | try { |
11fdf7f2 | 9694 | decode(compat_ondisk_format, p); |
7c673cae FG |
9695 | } catch (buffer::error& e) { |
9696 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; | |
9697 | return -EIO; | |
9698 | } | |
9699 | } | |
9700 | } | |
9701 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
9702 | << " compat_ondisk_format " << compat_ondisk_format | |
9703 | << dendl; | |
9704 | } | |
9705 | ||
9706 | if (latest_ondisk_format < compat_ondisk_format) { | |
9707 | derr << __func__ << " compat_ondisk_format is " | |
9708 | << compat_ondisk_format << " but we only understand version " | |
9709 | << latest_ondisk_format << dendl; | |
9710 | return -EPERM; | |
9711 | } | |
7c673cae FG |
9712 | |
9713 | { | |
9714 | bufferlist bl; | |
9715 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 9716 | auto p = bl.cbegin(); |
7c673cae FG |
9717 | try { |
9718 | uint64_t val; | |
11fdf7f2 | 9719 | decode(val, p); |
7c673cae | 9720 | min_alloc_size = val; |
224ce89b | 9721 | min_alloc_size_order = ctz(val); |
11fdf7f2 | 9722 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
7c673cae FG |
9723 | } catch (buffer::error& e) { |
9724 | derr << __func__ << " unable to read min_alloc_size" << dendl; | |
9725 | return -EIO; | |
9726 | } | |
9727 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
9728 | << std::dec << dendl; | |
9729 | } | |
224ce89b | 9730 | _open_statfs(); |
7c673cae FG |
9731 | _set_alloc_sizes(); |
9732 | _set_throttle_params(); | |
9733 | ||
9734 | _set_csum(); | |
9735 | _set_compression(); | |
9736 | _set_blob_size(); | |
9737 | ||
11fdf7f2 | 9738 | _validate_bdev(); |
7c673cae FG |
9739 | return 0; |
9740 | } | |
9741 | ||
9742 | int BlueStore::_upgrade_super() | |
9743 | { | |
9744 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
9745 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
9746 | if (ondisk_format < latest_ondisk_format) { |
9747 | ceph_assert(ondisk_format > 0); | |
9748 | ceph_assert(ondisk_format < latest_ondisk_format); | |
9749 | ||
9750 | if (ondisk_format == 1) { | |
9751 | // changes: | |
9752 | // - super: added ondisk_format | |
9753 | // - super: added min_readable_ondisk_format | |
9754 | // - super: added min_compat_ondisk_format | |
9755 | // - super: added min_alloc_size | |
9756 | // - super: removed min_min_alloc_size | |
9757 | KeyValueDB::Transaction t = db->get_transaction(); | |
9758 | { | |
9759 | bufferlist bl; | |
9760 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
9761 | auto p = bl.cbegin(); | |
9762 | try { | |
9763 | uint64_t val; | |
9764 | decode(val, p); | |
9765 | min_alloc_size = val; | |
9766 | } catch (buffer::error& e) { | |
9767 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; | |
9768 | return -EIO; | |
9769 | } | |
9770 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
9771 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 9772 | } |
11fdf7f2 TL |
9773 | ondisk_format = 2; |
9774 | _prepare_ondisk_format_super(t); | |
9775 | int r = db->submit_transaction_sync(t); | |
9776 | ceph_assert(r == 0); | |
7c673cae | 9777 | } |
7c673cae | 9778 | } |
7c673cae FG |
9779 | // done |
9780 | dout(1) << __func__ << " done" << dendl; | |
9781 | return 0; | |
9782 | } | |
9783 | ||
9784 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef o) | |
9785 | { | |
224ce89b | 9786 | if (o->onode.nid) { |
11fdf7f2 | 9787 | ceph_assert(o->exists); |
7c673cae | 9788 | return; |
224ce89b | 9789 | } |
7c673cae FG |
9790 | uint64_t nid = ++nid_last; |
9791 | dout(20) << __func__ << " " << nid << dendl; | |
9792 | o->onode.nid = nid; | |
9793 | txc->last_nid = nid; | |
224ce89b | 9794 | o->exists = true; |
7c673cae FG |
9795 | } |
9796 | ||
9797 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
9798 | { | |
9799 | uint64_t bid = ++blobid_last; | |
9800 | dout(20) << __func__ << " " << bid << dendl; | |
9801 | txc->last_blobid = bid; | |
9802 | return bid; | |
9803 | } | |
9804 | ||
9805 | void BlueStore::get_db_statistics(Formatter *f) | |
9806 | { | |
9807 | db->get_statistics(f); | |
9808 | } | |
9809 | ||
11fdf7f2 TL |
9810 | BlueStore::TransContext *BlueStore::_txc_create( |
9811 | Collection *c, OpSequencer *osr, | |
9812 | list<Context*> *on_commits) | |
7c673cae | 9813 | { |
11fdf7f2 | 9814 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae FG |
9815 | txc->t = db->get_transaction(); |
9816 | osr->queue_new(txc); | |
9817 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
9818 | << " seq " << txc->seq << dendl; | |
9819 | return txc; | |
9820 | } | |
9821 | ||
9822 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
9823 | { | |
11fdf7f2 TL |
9824 | // one "io" for the kv commit |
9825 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
9826 | auto cost = throttle_cost_per_io.load(); |
9827 | txc->cost = ios * cost + txc->bytes; | |
9828 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" | |
9829 | << ios << " ios * " << cost << " + " << txc->bytes | |
9830 | << " bytes)" << dendl; | |
9831 | } | |
9832 | ||
9833 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
9834 | { | |
9835 | if (txc->statfs_delta.is_empty()) | |
9836 | return; | |
9837 | ||
9838 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
9839 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
9840 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
9841 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
9842 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
9843 | ||
9844 | bufferlist bl; | |
9845 | txc->statfs_delta.encode(bl); | |
11fdf7f2 TL |
9846 | if (per_pool_stat_collection) { |
9847 | string key; | |
9848 | get_pool_stat_key(txc->osd_pool_id, &key); | |
9849 | txc->t->merge(PREFIX_STAT, key, bl); | |
9850 | ||
9851 | std::lock_guard l(vstatfs_lock); | |
9852 | auto& stats = osd_pools[txc->osd_pool_id]; | |
9853 | stats += txc->statfs_delta; | |
9854 | ||
9855 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
9856 | ||
9857 | } else { | |
9858 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
7c673cae | 9859 | |
11fdf7f2 TL |
9860 | std::lock_guard l(vstatfs_lock); |
9861 | vstatfs += txc->statfs_delta; | |
9862 | } | |
7c673cae FG |
9863 | txc->statfs_delta.reset(); |
9864 | } | |
9865 | ||
9866 | void BlueStore::_txc_state_proc(TransContext *txc) | |
9867 | { | |
9868 | while (true) { | |
9869 | dout(10) << __func__ << " txc " << txc | |
9870 | << " " << txc->get_state_name() << dendl; | |
9871 | switch (txc->state) { | |
9872 | case TransContext::STATE_PREPARE: | |
9873 | txc->log_state_latency(logger, l_bluestore_state_prepare_lat); | |
9874 | if (txc->ioc.has_pending_aios()) { | |
9875 | txc->state = TransContext::STATE_AIO_WAIT; | |
9876 | txc->had_ios = true; | |
9877 | _txc_aio_submit(txc); | |
9878 | return; | |
9879 | } | |
9880 | // ** fall-thru ** | |
9881 | ||
9882 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 TL |
9883 | { |
9884 | utime_t lat = txc->log_state_latency(logger, l_bluestore_state_aio_wait_lat); | |
9885 | if (lat >= cct->_conf->bluestore_log_op_age) { | |
9886 | dout(0) << __func__ << " slow aio_wait, txc = " << txc | |
9887 | << ", latency = " << lat | |
9888 | << dendl; | |
9889 | } | |
9890 | } | |
9891 | ||
7c673cae FG |
9892 | _txc_finish_io(txc); // may trigger blocked txc's too |
9893 | return; | |
9894 | ||
9895 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 9896 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
9897 | if (txc->had_ios) { |
9898 | ++txc->osr->txc_with_unstable_io; | |
9899 | } | |
9900 | txc->log_state_latency(logger, l_bluestore_state_io_done_lat); | |
9901 | txc->state = TransContext::STATE_KV_QUEUED; | |
9902 | if (cct->_conf->bluestore_sync_submit_transaction) { | |
9903 | if (txc->last_nid >= nid_max || | |
9904 | txc->last_blobid >= blobid_max) { | |
9905 | dout(20) << __func__ | |
9906 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
9907 | << dendl; | |
9908 | } else if (txc->osr->kv_committing_serially) { | |
9909 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
9910 | << dendl; | |
9911 | // note: this is starvation-prone. once we have a txc in a busy | |
9912 | // sequencer that is committing serially it is possible to keep | |
9913 | // submitting new transactions fast enough that we get stuck doing | |
9914 | // so. the alternative is to block here... fixme? | |
9915 | } else if (txc->osr->txc_with_unstable_io) { | |
9916 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
9917 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
9918 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
9919 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
9920 | == 0) { | |
9921 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
9922 | << dendl; | |
9923 | } else { | |
9924 | txc->state = TransContext::STATE_KV_SUBMITTED; | |
31f18b77 | 9925 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); |
11fdf7f2 | 9926 | ceph_assert(r == 0); |
7c673cae FG |
9927 | _txc_applied_kv(txc); |
9928 | } | |
9929 | } | |
9930 | { | |
11fdf7f2 | 9931 | std::lock_guard l(kv_lock); |
7c673cae FG |
9932 | kv_queue.push_back(txc); |
9933 | kv_cond.notify_one(); | |
9934 | if (txc->state != TransContext::STATE_KV_SUBMITTED) { | |
9935 | kv_queue_unsubmitted.push_back(txc); | |
9936 | ++txc->osr->kv_committing_serially; | |
9937 | } | |
31f18b77 FG |
9938 | if (txc->had_ios) |
9939 | kv_ios++; | |
9940 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
9941 | } |
9942 | return; | |
9943 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
9944 | _txc_committed_kv(txc); |
9945 | // ** fall-thru ** | |
9946 | ||
9947 | case TransContext::STATE_KV_DONE: | |
9948 | txc->log_state_latency(logger, l_bluestore_state_kv_done_lat); | |
9949 | if (txc->deferred_txn) { | |
9950 | txc->state = TransContext::STATE_DEFERRED_QUEUED; | |
9951 | _deferred_queue(txc); | |
9952 | return; | |
9953 | } | |
9954 | txc->state = TransContext::STATE_FINISHING; | |
9955 | break; | |
9956 | ||
9957 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9958 | txc->log_state_latency(logger, l_bluestore_state_deferred_cleanup_lat); | |
9959 | txc->state = TransContext::STATE_FINISHING; | |
9960 | // ** fall-thru ** | |
9961 | ||
9962 | case TransContext::STATE_FINISHING: | |
9963 | txc->log_state_latency(logger, l_bluestore_state_finishing_lat); | |
9964 | _txc_finish(txc); | |
9965 | return; | |
9966 | ||
9967 | default: | |
9968 | derr << __func__ << " unexpected txc " << txc | |
9969 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 9970 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
9971 | return; |
9972 | } | |
9973 | } | |
9974 | } | |
9975 | ||
9976 | void BlueStore::_txc_finish_io(TransContext *txc) | |
9977 | { | |
9978 | dout(20) << __func__ << " " << txc << dendl; | |
9979 | ||
9980 | /* | |
9981 | * we need to preserve the order of kv transactions, | |
9982 | * even though aio will complete in any order. | |
9983 | */ | |
9984 | ||
9985 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 9986 | std::lock_guard l(osr->qlock); |
7c673cae | 9987 | txc->state = TransContext::STATE_IO_DONE; |
11fdf7f2 | 9988 | txc->ioc.release_running_aios(); |
7c673cae FG |
9989 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
9990 | while (p != osr->q.begin()) { | |
9991 | --p; | |
9992 | if (p->state < TransContext::STATE_IO_DONE) { | |
9993 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " | |
9994 | << p->get_state_name() << dendl; | |
9995 | return; | |
9996 | } | |
9997 | if (p->state > TransContext::STATE_IO_DONE) { | |
9998 | ++p; | |
9999 | break; | |
10000 | } | |
10001 | } | |
10002 | do { | |
10003 | _txc_state_proc(&*p++); | |
10004 | } while (p != osr->q.end() && | |
10005 | p->state == TransContext::STATE_IO_DONE); | |
10006 | ||
11fdf7f2 | 10007 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
10008 | osr->qcond.notify_all(); |
10009 | } | |
10010 | } | |
10011 | ||
10012 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
10013 | { | |
10014 | dout(20) << __func__ << " txc " << txc | |
10015 | << " onodes " << txc->onodes | |
10016 | << " shared_blobs " << txc->shared_blobs | |
10017 | << dendl; | |
10018 | ||
10019 | // finalize onodes | |
10020 | for (auto o : txc->onodes) { | |
11fdf7f2 | 10021 | _record_onode(o, t); |
7c673cae FG |
10022 | o->flushing_count++; |
10023 | } | |
10024 | ||
10025 | // objects we modified but didn't affect the onode | |
10026 | auto p = txc->modified_objects.begin(); | |
10027 | while (p != txc->modified_objects.end()) { | |
10028 | if (txc->onodes.count(*p) == 0) { | |
10029 | (*p)->flushing_count++; | |
10030 | ++p; | |
10031 | } else { | |
10032 | // remove dups with onodes list to avoid problems in _txc_finish | |
10033 | p = txc->modified_objects.erase(p); | |
10034 | } | |
10035 | } | |
10036 | ||
10037 | // finalize shared_blobs | |
10038 | for (auto sb : txc->shared_blobs) { | |
10039 | string key; | |
10040 | auto sbid = sb->get_sbid(); | |
10041 | get_shared_blob_key(sbid, &key); | |
10042 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
10043 | dout(20) << __func__ << " shared_blob 0x" |
10044 | << std::hex << sbid << std::dec | |
7c673cae FG |
10045 | << " is empty" << dendl; |
10046 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
10047 | } else { | |
10048 | bufferlist bl; | |
11fdf7f2 TL |
10049 | encode(*(sb->persistent), bl); |
10050 | dout(20) << __func__ << " shared_blob 0x" | |
10051 | << std::hex << sbid << std::dec | |
31f18b77 | 10052 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
10053 | t->set(PREFIX_SHARED_BLOB, key, bl); |
10054 | } | |
10055 | } | |
10056 | } | |
10057 | ||
10058 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
10059 | PerfCounters &logger) | |
10060 | { | |
11fdf7f2 TL |
10061 | os_commit_latency_ns.consume_next( |
10062 | logger.get_tavg_ns( | |
7c673cae | 10063 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
10064 | os_apply_latency_ns.consume_next( |
10065 | logger.get_tavg_ns( | |
7c673cae FG |
10066 | l_bluestore_commit_lat)); |
10067 | } | |
10068 | ||
10069 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) | |
10070 | { | |
10071 | dout(20) << __func__ << " txc " << txc << std::hex | |
10072 | << " allocated 0x" << txc->allocated | |
10073 | << " released 0x" << txc->released | |
10074 | << std::dec << dendl; | |
10075 | ||
10076 | // We have to handle the case where we allocate *and* deallocate the | |
10077 | // same region in this transaction. The freelist doesn't like that. | |
10078 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
10079 | // debug check. But that's important.) | |
10080 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
10081 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
10082 | interval_set<uint64_t> *preleased = &txc->released; | |
10083 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
10084 | interval_set<uint64_t> overlap; | |
10085 | overlap.intersection_of(txc->allocated, txc->released); | |
10086 | if (!overlap.empty()) { | |
10087 | tmp_allocated = txc->allocated; | |
10088 | tmp_allocated.subtract(overlap); | |
10089 | tmp_released = txc->released; | |
10090 | tmp_released.subtract(overlap); | |
10091 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
10092 | << ", new allocated 0x" << tmp_allocated | |
10093 | << " released 0x" << tmp_released << std::dec | |
10094 | << dendl; | |
10095 | pallocated = &tmp_allocated; | |
10096 | preleased = &tmp_released; | |
10097 | } | |
10098 | } | |
10099 | ||
10100 | // update freelist with non-overlap sets | |
10101 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
10102 | p != pallocated->end(); | |
10103 | ++p) { | |
10104 | fm->allocate(p.get_start(), p.get_len(), t); | |
10105 | } | |
10106 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
10107 | p != preleased->end(); | |
10108 | ++p) { | |
10109 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
10110 | << "~" << p.get_len() << std::dec << dendl; | |
10111 | fm->release(p.get_start(), p.get_len(), t); | |
10112 | } | |
10113 | ||
10114 | _txc_update_store_statfs(txc); | |
10115 | } | |
10116 | ||
10117 | void BlueStore::_txc_applied_kv(TransContext *txc) | |
10118 | { | |
10119 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { | |
10120 | for (auto& o : *ls) { | |
10121 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
10122 | << dendl; | |
10123 | if (--o->flushing_count == 0) { | |
11fdf7f2 | 10124 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
10125 | o->flush_cond.notify_all(); |
10126 | } | |
10127 | } | |
10128 | } | |
10129 | } | |
10130 | ||
10131 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
10132 | { | |
10133 | dout(20) << __func__ << " txc " << txc << dendl; | |
1adf2230 | 10134 | { |
11fdf7f2 | 10135 | std::lock_guard l(txc->osr->qlock); |
1adf2230 | 10136 | txc->state = TransContext::STATE_KV_DONE; |
11fdf7f2 TL |
10137 | if (txc->ch->commit_queue) { |
10138 | txc->ch->commit_queue->queue(txc->oncommits); | |
10139 | } else { | |
10140 | finisher.queue(txc->oncommits); | |
1adf2230 | 10141 | } |
7c673cae | 10142 | } |
1adf2230 | 10143 | txc->log_state_latency(logger, l_bluestore_state_kv_committing_lat); |
11fdf7f2 TL |
10144 | LOG_LATENCY_FN(logger, cct, |
10145 | l_bluestore_commit_lat, | |
10146 | ceph::make_timespan(ceph_clock_now() - txc->start), | |
10147 | [&](auto lat) { | |
10148 | return ", txc = " + stringify(txc); | |
10149 | } | |
10150 | ); | |
7c673cae FG |
10151 | } |
10152 | ||
10153 | void BlueStore::_txc_finish(TransContext *txc) | |
10154 | { | |
10155 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
11fdf7f2 | 10156 | ceph_assert(txc->state == TransContext::STATE_FINISHING); |
7c673cae FG |
10157 | |
10158 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 10159 | sb->finish_write(txc->seq); |
7c673cae FG |
10160 | } |
10161 | txc->shared_blobs_written.clear(); | |
10162 | ||
10163 | while (!txc->removed_collections.empty()) { | |
10164 | _queue_reap_collection(txc->removed_collections.front()); | |
10165 | txc->removed_collections.pop_front(); | |
10166 | } | |
10167 | ||
10168 | OpSequencerRef osr = txc->osr; | |
7c673cae | 10169 | bool empty = false; |
31f18b77 | 10170 | bool submit_deferred = false; |
7c673cae FG |
10171 | OpSequencer::q_list_t releasing_txc; |
10172 | { | |
11fdf7f2 | 10173 | std::lock_guard l(osr->qlock); |
7c673cae FG |
10174 | txc->state = TransContext::STATE_DONE; |
10175 | bool notify = false; | |
10176 | while (!osr->q.empty()) { | |
10177 | TransContext *txc = &osr->q.front(); | |
10178 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
10179 | << dendl; | |
10180 | if (txc->state != TransContext::STATE_DONE) { | |
10181 | if (txc->state == TransContext::STATE_PREPARE && | |
10182 | deferred_aggressive) { | |
10183 | // for _osr_drain_preceding() | |
10184 | notify = true; | |
10185 | } | |
31f18b77 | 10186 | if (txc->state == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 10187 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
10188 | submit_deferred = true; |
10189 | } | |
7c673cae FG |
10190 | break; |
10191 | } | |
10192 | ||
7c673cae FG |
10193 | osr->q.pop_front(); |
10194 | releasing_txc.push_back(*txc); | |
10195 | notify = true; | |
10196 | } | |
10197 | if (notify) { | |
10198 | osr->qcond.notify_all(); | |
10199 | } | |
10200 | if (osr->q.empty()) { | |
10201 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
10202 | empty = true; | |
10203 | } | |
10204 | } | |
10205 | while (!releasing_txc.empty()) { | |
10206 | // release to allocator only after all preceding txc's have also | |
10207 | // finished any deferred writes that potentially land in these | |
10208 | // blocks | |
10209 | auto txc = &releasing_txc.front(); | |
10210 | _txc_release_alloc(txc); | |
10211 | releasing_txc.pop_front(); | |
10212 | txc->log_state_latency(logger, l_bluestore_state_done_lat); | |
10213 | delete txc; | |
10214 | } | |
10215 | ||
31f18b77 FG |
10216 | if (submit_deferred) { |
10217 | // we're pinning memory; flush! we could be more fine-grained here but | |
10218 | // i'm not sure it's worth the bother. | |
10219 | deferred_try_submit(); | |
7c673cae FG |
10220 | } |
10221 | ||
7c673cae | 10222 | if (empty && osr->zombie) { |
11fdf7f2 TL |
10223 | std::lock_guard l(zombie_osr_lock); |
10224 | if (zombie_osr_set.erase(osr->cid)) { | |
10225 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
10226 | } else { | |
10227 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
10228 | << dendl; | |
10229 | } | |
7c673cae | 10230 | } |
11fdf7f2 | 10231 | } |
7c673cae FG |
10232 | |
10233 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
10234 | { | |
a8e16298 | 10235 | // it's expected we're called with lazy_release_lock already taken! |
11fdf7f2 TL |
10236 | if (likely(!cct->_conf->bluestore_debug_no_reuse_blocks)) { |
10237 | int r = 0; | |
10238 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
10239 | r = bdev->queue_discard(txc->released); | |
10240 | if (r == 0) { | |
10241 | dout(10) << __func__ << "(queued) " << txc << " " << std::hex | |
10242 | << txc->released << std::dec << dendl; | |
10243 | goto out; | |
10244 | } | |
10245 | } else if (cct->_conf->bdev_enable_discard) { | |
10246 | for (auto p = txc->released.begin(); p != txc->released.end(); ++p) { | |
10247 | bdev->discard(p.get_start(), p.get_len()); | |
10248 | } | |
10249 | } | |
10250 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
94b18763 | 10251 | << txc->released << std::dec << dendl; |
11fdf7f2 | 10252 | alloc->release(txc->released); |
7c673cae FG |
10253 | } |
10254 | ||
11fdf7f2 | 10255 | out: |
7c673cae FG |
10256 | txc->allocated.clear(); |
10257 | txc->released.clear(); | |
10258 | } | |
10259 | ||
11fdf7f2 TL |
10260 | void BlueStore::_osr_attach(Collection *c) |
10261 | { | |
10262 | // note: caller has RWLock on coll_map | |
10263 | auto q = coll_map.find(c->cid); | |
10264 | if (q != coll_map.end()) { | |
10265 | c->osr = q->second->osr; | |
10266 | ldout(cct, 10) << __func__ << " " << c->cid | |
10267 | << " reusing osr " << c->osr << " from existing coll " | |
10268 | << q->second << dendl; | |
10269 | } else { | |
10270 | std::lock_guard l(zombie_osr_lock); | |
10271 | auto p = zombie_osr_set.find(c->cid); | |
10272 | if (p == zombie_osr_set.end()) { | |
10273 | c->osr = new OpSequencer(this, c->cid); | |
10274 | ldout(cct, 10) << __func__ << " " << c->cid | |
10275 | << " fresh osr " << c->osr << dendl; | |
10276 | } else { | |
10277 | c->osr = p->second; | |
10278 | zombie_osr_set.erase(p); | |
10279 | ldout(cct, 10) << __func__ << " " << c->cid | |
10280 | << " resurrecting zombie osr " << c->osr << dendl; | |
10281 | c->osr->zombie = false; | |
10282 | } | |
10283 | } | |
10284 | } | |
10285 | ||
10286 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
10287 | { | |
10288 | std::lock_guard l(zombie_osr_lock); | |
10289 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
10290 | osr->zombie = true; | |
10291 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
10292 | // this is either a new insertion or the same osr is already there | |
10293 | ceph_assert(i.second || i.first->second == osr); | |
10294 | } | |
10295 | ||
7c673cae FG |
10296 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
10297 | { | |
10298 | OpSequencer *osr = txc->osr.get(); | |
10299 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
10300 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
10301 | { | |
10302 | // submit anything pending | |
224ce89b | 10303 | deferred_lock.lock(); |
11fdf7f2 | 10304 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
10305 | _deferred_submit_unlock(osr); |
10306 | } else { | |
10307 | deferred_lock.unlock(); | |
7c673cae FG |
10308 | } |
10309 | } | |
10310 | { | |
10311 | // wake up any previously finished deferred events | |
11fdf7f2 | 10312 | std::lock_guard l(kv_lock); |
7c673cae FG |
10313 | kv_cond.notify_one(); |
10314 | } | |
10315 | osr->drain_preceding(txc); | |
10316 | --deferred_aggressive; | |
10317 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
10318 | } | |
10319 | ||
11fdf7f2 TL |
10320 | void BlueStore::_osr_drain(OpSequencer *osr) |
10321 | { | |
10322 | dout(10) << __func__ << " " << osr << dendl; | |
10323 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
10324 | { | |
10325 | // submit anything pending | |
10326 | deferred_lock.lock(); | |
10327 | if (osr->deferred_pending && !osr->deferred_running) { | |
10328 | _deferred_submit_unlock(osr); | |
10329 | } else { | |
10330 | deferred_lock.unlock(); | |
10331 | } | |
10332 | } | |
10333 | { | |
10334 | // wake up any previously finished deferred events | |
10335 | std::lock_guard l(kv_lock); | |
10336 | kv_cond.notify_one(); | |
10337 | } | |
10338 | osr->drain(); | |
10339 | --deferred_aggressive; | |
10340 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
10341 | } | |
10342 | ||
7c673cae FG |
10343 | void BlueStore::_osr_drain_all() |
10344 | { | |
10345 | dout(10) << __func__ << dendl; | |
10346 | ||
10347 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
10348 | vector<OpSequencerRef> zombies; |
10349 | { | |
10350 | RWLock::RLocker l(coll_lock); | |
10351 | for (auto& i : coll_map) { | |
10352 | s.insert(i.second->osr); | |
10353 | } | |
10354 | } | |
7c673cae | 10355 | { |
11fdf7f2 TL |
10356 | std::lock_guard l(zombie_osr_lock); |
10357 | for (auto& i : zombie_osr_set) { | |
10358 | s.insert(i.second); | |
10359 | zombies.push_back(i.second); | |
10360 | } | |
7c673cae FG |
10361 | } |
10362 | dout(20) << __func__ << " osr_set " << s << dendl; | |
10363 | ||
10364 | ++deferred_aggressive; | |
10365 | { | |
10366 | // submit anything pending | |
224ce89b | 10367 | deferred_try_submit(); |
7c673cae FG |
10368 | } |
10369 | { | |
10370 | // wake up any previously finished deferred events | |
11fdf7f2 | 10371 | std::lock_guard l(kv_lock); |
7c673cae FG |
10372 | kv_cond.notify_one(); |
10373 | } | |
31f18b77 | 10374 | { |
11fdf7f2 | 10375 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
10376 | kv_finalize_cond.notify_one(); |
10377 | } | |
7c673cae FG |
10378 | for (auto osr : s) { |
10379 | dout(20) << __func__ << " drain " << osr << dendl; | |
10380 | osr->drain(); | |
10381 | } | |
10382 | --deferred_aggressive; | |
10383 | ||
7c673cae | 10384 | { |
11fdf7f2 TL |
10385 | std::lock_guard l(zombie_osr_lock); |
10386 | for (auto& osr : zombies) { | |
10387 | if (zombie_osr_set.erase(osr->cid)) { | |
10388 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
10389 | ceph_assert(osr->q.empty()); | |
10390 | } else if (osr->zombie) { | |
10391 | dout(10) << __func__ << " empty zombie osr " << osr | |
10392 | << " already reaped" << dendl; | |
10393 | ceph_assert(osr->q.empty()); | |
10394 | } else { | |
10395 | dout(10) << __func__ << " empty zombie osr " << osr | |
10396 | << " resurrected" << dendl; | |
10397 | } | |
7c673cae FG |
10398 | } |
10399 | } | |
11fdf7f2 TL |
10400 | |
10401 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
10402 | } |
10403 | ||
11fdf7f2 | 10404 | |
31f18b77 FG |
10405 | void BlueStore::_kv_start() |
10406 | { | |
10407 | dout(10) << __func__ << dendl; | |
10408 | ||
181888fb | 10409 | deferred_finisher.start(); |
11fdf7f2 | 10410 | finisher.start(); |
31f18b77 FG |
10411 | kv_sync_thread.create("bstore_kv_sync"); |
10412 | kv_finalize_thread.create("bstore_kv_final"); | |
10413 | } | |
10414 | ||
10415 | void BlueStore::_kv_stop() | |
10416 | { | |
10417 | dout(10) << __func__ << dendl; | |
10418 | { | |
11fdf7f2 | 10419 | std::unique_lock l(kv_lock); |
31f18b77 FG |
10420 | while (!kv_sync_started) { |
10421 | kv_cond.wait(l); | |
10422 | } | |
10423 | kv_stop = true; | |
10424 | kv_cond.notify_all(); | |
10425 | } | |
10426 | { | |
11fdf7f2 | 10427 | std::unique_lock l(kv_finalize_lock); |
31f18b77 FG |
10428 | while (!kv_finalize_started) { |
10429 | kv_finalize_cond.wait(l); | |
10430 | } | |
10431 | kv_finalize_stop = true; | |
10432 | kv_finalize_cond.notify_all(); | |
10433 | } | |
10434 | kv_sync_thread.join(); | |
10435 | kv_finalize_thread.join(); | |
11fdf7f2 | 10436 | ceph_assert(removed_collections.empty()); |
31f18b77 | 10437 | { |
11fdf7f2 | 10438 | std::lock_guard l(kv_lock); |
31f18b77 FG |
10439 | kv_stop = false; |
10440 | } | |
10441 | { | |
11fdf7f2 | 10442 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
10443 | kv_finalize_stop = false; |
10444 | } | |
10445 | dout(10) << __func__ << " stopping finishers" << dendl; | |
181888fb FG |
10446 | deferred_finisher.wait_for_empty(); |
10447 | deferred_finisher.stop(); | |
11fdf7f2 TL |
10448 | finisher.wait_for_empty(); |
10449 | finisher.stop(); | |
31f18b77 FG |
10450 | dout(10) << __func__ << " stopped" << dendl; |
10451 | } | |
10452 | ||
7c673cae FG |
10453 | void BlueStore::_kv_sync_thread() |
10454 | { | |
10455 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
10456 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
10457 | std::unique_lock l(kv_lock); | |
10458 | ceph_assert(!kv_sync_started); | |
31f18b77 FG |
10459 | kv_sync_started = true; |
10460 | kv_cond.notify_all(); | |
7c673cae | 10461 | while (true) { |
11fdf7f2 | 10462 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
10463 | if (kv_queue.empty() && |
10464 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 10465 | !deferred_aggressive)) { |
7c673cae FG |
10466 | if (kv_stop) |
10467 | break; | |
10468 | dout(20) << __func__ << " sleep" << dendl; | |
11fdf7f2 | 10469 | kv_cond.wait(l); |
7c673cae FG |
10470 | dout(20) << __func__ << " wake" << dendl; |
10471 | } else { | |
10472 | deque<TransContext*> kv_submitting; | |
10473 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
10474 | uint64_t aios = 0, costs = 0; |
10475 | ||
7c673cae FG |
10476 | dout(20) << __func__ << " committing " << kv_queue.size() |
10477 | << " submitting " << kv_queue_unsubmitted.size() | |
10478 | << " deferred done " << deferred_done_queue.size() | |
10479 | << " stable " << deferred_stable_queue.size() | |
10480 | << dendl; | |
10481 | kv_committing.swap(kv_queue); | |
10482 | kv_submitting.swap(kv_queue_unsubmitted); | |
10483 | deferred_done.swap(deferred_done_queue); | |
10484 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
10485 | aios = kv_ios; |
10486 | costs = kv_throttle_costs; | |
10487 | kv_ios = 0; | |
10488 | kv_throttle_costs = 0; | |
7c673cae FG |
10489 | l.unlock(); |
10490 | ||
10491 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
10492 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
10493 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
10494 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
10495 | ||
11fdf7f2 TL |
10496 | auto start = mono_clock::now(); |
10497 | ||
7c673cae FG |
10498 | bool force_flush = false; |
10499 | // if bluefs is sharing the same device as data (only), then we | |
10500 | // can rely on the bluefs commit to flush the device and make | |
10501 | // deferred aios stable. that means that if we do have done deferred | |
10502 | // txcs AND we are not on a single device, we need to force a flush. | |
10503 | if (bluefs_single_shared_device && bluefs) { | |
31f18b77 | 10504 | if (aios) { |
7c673cae | 10505 | force_flush = true; |
11fdf7f2 | 10506 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
10507 | force_flush = true; // there's nothing else to commit! |
10508 | } else if (deferred_aggressive) { | |
10509 | force_flush = true; | |
10510 | } | |
11fdf7f2 TL |
10511 | } else { |
10512 | if (aios || !deferred_done.empty()) { | |
10513 | force_flush = true; | |
10514 | } else { | |
10515 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
10516 | } | |
10517 | } | |
7c673cae FG |
10518 | |
10519 | if (force_flush) { | |
31f18b77 | 10520 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
10521 | << " force_flush=" << (int)force_flush |
10522 | << ", flushing, deferred done->stable" << dendl; | |
10523 | // flush/barrier on block device | |
10524 | bdev->flush(); | |
10525 | ||
10526 | // if we flush then deferred done are now deferred stable | |
10527 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
10528 | deferred_done.end()); | |
10529 | deferred_done.clear(); | |
10530 | } | |
11fdf7f2 | 10531 | auto after_flush = mono_clock::now(); |
7c673cae FG |
10532 | |
10533 | // we will use one final transaction to force a sync | |
10534 | KeyValueDB::Transaction synct = db->get_transaction(); | |
10535 | ||
10536 | // increase {nid,blobid}_max? note that this covers both the | |
10537 | // case where we are approaching the max and the case we passed | |
10538 | // it. in either case, we increase the max in the earlier txn | |
10539 | // we submit. | |
10540 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
10541 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
10542 | KeyValueDB::Transaction t = | |
10543 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
10544 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
10545 | bufferlist bl; | |
11fdf7f2 | 10546 | encode(new_nid_max, bl); |
7c673cae FG |
10547 | t->set(PREFIX_SUPER, "nid_max", bl); |
10548 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
10549 | } | |
10550 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
10551 | KeyValueDB::Transaction t = | |
10552 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
10553 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
10554 | bufferlist bl; | |
11fdf7f2 | 10555 | encode(new_blobid_max, bl); |
7c673cae FG |
10556 | t->set(PREFIX_SUPER, "blobid_max", bl); |
10557 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
10558 | } | |
c07f9fc5 FG |
10559 | |
10560 | for (auto txc : kv_committing) { | |
10561 | if (txc->state == TransContext::STATE_KV_QUEUED) { | |
10562 | txc->log_state_latency(logger, l_bluestore_state_kv_queued_lat); | |
10563 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); | |
11fdf7f2 | 10564 | ceph_assert(r == 0); |
c07f9fc5 FG |
10565 | _txc_applied_kv(txc); |
10566 | --txc->osr->kv_committing_serially; | |
10567 | txc->state = TransContext::STATE_KV_SUBMITTED; | |
10568 | if (txc->osr->kv_submitted_waiters) { | |
11fdf7f2 TL |
10569 | std::lock_guard l(txc->osr->qlock); |
10570 | txc->osr->qcond.notify_all(); | |
7c673cae | 10571 | } |
c07f9fc5 FG |
10572 | |
10573 | } else { | |
11fdf7f2 | 10574 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
c07f9fc5 | 10575 | txc->log_state_latency(logger, l_bluestore_state_kv_queued_lat); |
7c673cae | 10576 | } |
7c673cae FG |
10577 | if (txc->had_ios) { |
10578 | --txc->osr->txc_with_unstable_io; | |
10579 | } | |
7c673cae FG |
10580 | } |
10581 | ||
31f18b77 FG |
10582 | // release throttle *before* we commit. this allows new ops |
10583 | // to be prepared and enter pipeline while we are waiting on | |
10584 | // the kv commit sync/flush. then hopefully on the next | |
10585 | // iteration there will already be ops awake. otherwise, we | |
10586 | // end up going to sleep, and then wake up when the very first | |
10587 | // transaction is ready for commit. | |
10588 | throttle_bytes.put(costs); | |
10589 | ||
7c673cae FG |
10590 | if (bluefs && |
10591 | after_flush - bluefs_last_balance > | |
11fdf7f2 | 10592 | ceph::make_timespan(cct->_conf->bluestore_bluefs_balance_interval)) { |
7c673cae | 10593 | bluefs_last_balance = after_flush; |
11fdf7f2 TL |
10594 | int r = _balance_bluefs_freespace(); |
10595 | ceph_assert(r >= 0); | |
7c673cae FG |
10596 | } |
10597 | ||
10598 | // cleanup sync deferred keys | |
10599 | for (auto b : deferred_stable) { | |
10600 | for (auto& txc : b->txcs) { | |
10601 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 10602 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
10603 | string key; |
10604 | get_deferred_key(wt.seq, &key); | |
10605 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
10606 | } | |
10607 | } | |
10608 | ||
10609 | // submit synct synchronously (block and wait for it to commit) | |
31f18b77 | 10610 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
10611 | ceph_assert(r == 0); |
10612 | ||
10613 | { | |
10614 | std::unique_lock m(kv_finalize_lock); | |
10615 | if (kv_committing_to_finalize.empty()) { | |
10616 | kv_committing_to_finalize.swap(kv_committing); | |
10617 | } else { | |
10618 | kv_committing_to_finalize.insert( | |
10619 | kv_committing_to_finalize.end(), | |
10620 | kv_committing.begin(), | |
10621 | kv_committing.end()); | |
10622 | kv_committing.clear(); | |
10623 | } | |
10624 | if (deferred_stable_to_finalize.empty()) { | |
10625 | deferred_stable_to_finalize.swap(deferred_stable); | |
10626 | } else { | |
10627 | deferred_stable_to_finalize.insert( | |
10628 | deferred_stable_to_finalize.end(), | |
10629 | deferred_stable.begin(), | |
10630 | deferred_stable.end()); | |
10631 | deferred_stable.clear(); | |
10632 | } | |
10633 | kv_finalize_cond.notify_one(); | |
10634 | } | |
7c673cae FG |
10635 | |
10636 | if (new_nid_max) { | |
10637 | nid_max = new_nid_max; | |
10638 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
10639 | } | |
10640 | if (new_blobid_max) { | |
10641 | blobid_max = new_blobid_max; | |
10642 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
10643 | } | |
10644 | ||
224ce89b | 10645 | { |
11fdf7f2 TL |
10646 | auto finish = mono_clock::now(); |
10647 | ceph::timespan dur_flush = after_flush - start; | |
10648 | ceph::timespan dur_kv = finish - after_flush; | |
10649 | ceph::timespan dur = finish - start; | |
224ce89b WB |
10650 | dout(20) << __func__ << " committed " << kv_committing.size() |
10651 | << " cleaned " << deferred_stable.size() | |
10652 | << " in " << dur | |
10653 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
10654 | << dendl; | |
11fdf7f2 TL |
10655 | LOG_LATENCY(logger, cct, l_bluestore_kv_flush_lat, dur_flush); |
10656 | LOG_LATENCY(logger, cct, l_bluestore_kv_commit_lat, dur_kv); | |
10657 | LOG_LATENCY(logger, cct, l_bluestore_kv_sync_lat, dur); | |
7c673cae | 10658 | } |
31f18b77 FG |
10659 | |
10660 | if (bluefs) { | |
11fdf7f2 TL |
10661 | if (!bluefs_extents_reclaiming.empty()) { |
10662 | dout(0) << __func__ << " releasing old bluefs 0x" << std::hex | |
10663 | << bluefs_extents_reclaiming << std::dec << dendl; | |
81eedcae TL |
10664 | int r = 0; |
10665 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
10666 | r = bdev->queue_discard(bluefs_extents_reclaiming); | |
10667 | if (r == 0) { | |
10668 | goto clear; | |
10669 | } | |
10670 | } else if (cct->_conf->bdev_enable_discard) { | |
10671 | for (auto p = bluefs_extents_reclaiming.begin(); p != bluefs_extents_reclaiming.end(); ++p) { | |
10672 | bdev->discard(p.get_start(), p.get_len()); | |
10673 | } | |
10674 | } | |
10675 | ||
11fdf7f2 | 10676 | alloc->release(bluefs_extents_reclaiming); |
81eedcae | 10677 | clear: |
11fdf7f2 | 10678 | bluefs_extents_reclaiming.clear(); |
31f18b77 | 10679 | } |
31f18b77 FG |
10680 | } |
10681 | ||
10682 | l.lock(); | |
10683 | // previously deferred "done" are now "stable" by virtue of this | |
10684 | // commit cycle. | |
10685 | deferred_stable_queue.swap(deferred_done); | |
10686 | } | |
10687 | } | |
10688 | dout(10) << __func__ << " finish" << dendl; | |
10689 | kv_sync_started = false; | |
10690 | } | |
10691 | ||
10692 | void BlueStore::_kv_finalize_thread() | |
10693 | { | |
10694 | deque<TransContext*> kv_committed; | |
10695 | deque<DeferredBatch*> deferred_stable; | |
10696 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
10697 | std::unique_lock l(kv_finalize_lock); |
10698 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
10699 | kv_finalize_started = true; |
10700 | kv_finalize_cond.notify_all(); | |
10701 | while (true) { | |
11fdf7f2 TL |
10702 | ceph_assert(kv_committed.empty()); |
10703 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
10704 | if (kv_committing_to_finalize.empty() && |
10705 | deferred_stable_to_finalize.empty()) { | |
10706 | if (kv_finalize_stop) | |
10707 | break; | |
10708 | dout(20) << __func__ << " sleep" << dendl; | |
10709 | kv_finalize_cond.wait(l); | |
10710 | dout(20) << __func__ << " wake" << dendl; | |
10711 | } else { | |
10712 | kv_committed.swap(kv_committing_to_finalize); | |
10713 | deferred_stable.swap(deferred_stable_to_finalize); | |
10714 | l.unlock(); | |
10715 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
10716 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
10717 | ||
11fdf7f2 TL |
10718 | auto start = mono_clock::now(); |
10719 | ||
31f18b77 FG |
10720 | while (!kv_committed.empty()) { |
10721 | TransContext *txc = kv_committed.front(); | |
11fdf7f2 | 10722 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 10723 | _txc_state_proc(txc); |
31f18b77 | 10724 | kv_committed.pop_front(); |
7c673cae | 10725 | } |
31f18b77 | 10726 | |
7c673cae FG |
10727 | for (auto b : deferred_stable) { |
10728 | auto p = b->txcs.begin(); | |
10729 | while (p != b->txcs.end()) { | |
10730 | TransContext *txc = &*p; | |
10731 | p = b->txcs.erase(p); // unlink here because | |
10732 | _txc_state_proc(txc); // this may destroy txc | |
10733 | } | |
10734 | delete b; | |
10735 | } | |
31f18b77 | 10736 | deferred_stable.clear(); |
7c673cae FG |
10737 | |
10738 | if (!deferred_aggressive) { | |
31f18b77 | 10739 | if (deferred_queue_size >= deferred_batch_ops.load() || |
7c673cae | 10740 | throttle_deferred_bytes.past_midpoint()) { |
224ce89b | 10741 | deferred_try_submit(); |
7c673cae FG |
10742 | } |
10743 | } | |
10744 | ||
10745 | // this is as good a place as any ... | |
10746 | _reap_collections(); | |
10747 | ||
11fdf7f2 TL |
10748 | logger->set(l_bluestore_fragmentation, |
10749 | (uint64_t)(alloc->get_fragmentation(min_alloc_size) * 1000)); | |
10750 | ||
10751 | LOG_LATENCY(logger, cct, l_bluestore_kv_final_lat, mono_clock::now() - start); | |
10752 | ||
7c673cae | 10753 | l.lock(); |
7c673cae FG |
10754 | } |
10755 | } | |
10756 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 10757 | kv_finalize_started = false; |
7c673cae FG |
10758 | } |
10759 | ||
10760 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( | |
10761 | TransContext *txc, OnodeRef o) | |
10762 | { | |
10763 | if (!txc->deferred_txn) { | |
10764 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
10765 | } | |
10766 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
10767 | return &txc->deferred_txn->ops.back(); | |
10768 | } | |
10769 | ||
10770 | void BlueStore::_deferred_queue(TransContext *txc) | |
10771 | { | |
10772 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
224ce89b | 10773 | deferred_lock.lock(); |
7c673cae FG |
10774 | if (!txc->osr->deferred_pending && |
10775 | !txc->osr->deferred_running) { | |
10776 | deferred_queue.push_back(*txc->osr); | |
10777 | } | |
10778 | if (!txc->osr->deferred_pending) { | |
10779 | txc->osr->deferred_pending = new DeferredBatch(cct, txc->osr.get()); | |
10780 | } | |
10781 | ++deferred_queue_size; | |
10782 | txc->osr->deferred_pending->txcs.push_back(*txc); | |
10783 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; | |
10784 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
10785 | const auto& op = *opi; | |
11fdf7f2 | 10786 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
10787 | bufferlist::const_iterator p = op.data.begin(); |
10788 | for (auto e : op.extents) { | |
10789 | txc->osr->deferred_pending->prepare_write( | |
10790 | cct, wt.seq, e.offset, e.length, p); | |
10791 | } | |
10792 | } | |
10793 | if (deferred_aggressive && | |
10794 | !txc->osr->deferred_running) { | |
224ce89b WB |
10795 | _deferred_submit_unlock(txc->osr.get()); |
10796 | } else { | |
10797 | deferred_lock.unlock(); | |
7c673cae FG |
10798 | } |
10799 | } | |
10800 | ||
224ce89b | 10801 | void BlueStore::deferred_try_submit() |
7c673cae FG |
10802 | { |
10803 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
10804 | << deferred_queue_size << " txcs" << dendl; | |
11fdf7f2 | 10805 | std::lock_guard l(deferred_lock); |
224ce89b WB |
10806 | vector<OpSequencerRef> osrs; |
10807 | osrs.reserve(deferred_queue.size()); | |
7c673cae | 10808 | for (auto& osr : deferred_queue) { |
224ce89b WB |
10809 | osrs.push_back(&osr); |
10810 | } | |
10811 | for (auto& osr : osrs) { | |
181888fb FG |
10812 | if (osr->deferred_pending) { |
10813 | if (!osr->deferred_running) { | |
10814 | _deferred_submit_unlock(osr.get()); | |
10815 | deferred_lock.lock(); | |
10816 | } else { | |
10817 | dout(20) << __func__ << " osr " << osr << " already has running" | |
10818 | << dendl; | |
10819 | } | |
10820 | } else { | |
10821 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; | |
7c673cae FG |
10822 | } |
10823 | } | |
10824 | } | |
10825 | ||
224ce89b | 10826 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
10827 | { |
10828 | dout(10) << __func__ << " osr " << osr | |
10829 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
10830 | << dendl; | |
11fdf7f2 TL |
10831 | ceph_assert(osr->deferred_pending); |
10832 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
10833 | |
10834 | auto b = osr->deferred_pending; | |
10835 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 10836 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
10837 | |
10838 | osr->deferred_running = osr->deferred_pending; | |
10839 | osr->deferred_pending = nullptr; | |
10840 | ||
11fdf7f2 TL |
10841 | deferred_lock.unlock(); |
10842 | ||
10843 | for (auto& txc : b->txcs) { | |
10844 | txc.log_state_latency(logger, l_bluestore_state_deferred_queued_lat); | |
10845 | } | |
7c673cae FG |
10846 | uint64_t start = 0, pos = 0; |
10847 | bufferlist bl; | |
10848 | auto i = b->iomap.begin(); | |
10849 | while (true) { | |
10850 | if (i == b->iomap.end() || i->first != pos) { | |
10851 | if (bl.length()) { | |
10852 | dout(20) << __func__ << " write 0x" << std::hex | |
10853 | << start << "~" << bl.length() | |
10854 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 10855 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
10856 | logger->inc(l_bluestore_deferred_write_ops); |
10857 | logger->inc(l_bluestore_deferred_write_bytes, bl.length()); | |
10858 | int r = bdev->aio_write(start, bl, &b->ioc, false); | |
11fdf7f2 | 10859 | ceph_assert(r == 0); |
7c673cae FG |
10860 | } |
10861 | } | |
10862 | if (i == b->iomap.end()) { | |
10863 | break; | |
10864 | } | |
10865 | start = 0; | |
10866 | pos = i->first; | |
10867 | bl.clear(); | |
10868 | } | |
10869 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
10870 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
10871 | << dendl; | |
10872 | if (!bl.length()) { | |
10873 | start = pos; | |
10874 | } | |
10875 | pos += i->second.bl.length(); | |
10876 | bl.claim_append(i->second.bl); | |
10877 | ++i; | |
10878 | } | |
224ce89b | 10879 | |
7c673cae FG |
10880 | bdev->aio_submit(&b->ioc); |
10881 | } | |
10882 | ||
3efd9988 FG |
10883 | struct C_DeferredTrySubmit : public Context { |
10884 | BlueStore *store; | |
10885 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
10886 | void finish(int r) { | |
10887 | store->deferred_try_submit(); | |
10888 | } | |
10889 | }; | |
10890 | ||
7c673cae FG |
10891 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
10892 | { | |
10893 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 10894 | ceph_assert(osr->deferred_running); |
7c673cae FG |
10895 | DeferredBatch *b = osr->deferred_running; |
10896 | ||
10897 | { | |
11fdf7f2 TL |
10898 | std::lock_guard l(deferred_lock); |
10899 | ceph_assert(osr->deferred_running == b); | |
7c673cae FG |
10900 | osr->deferred_running = nullptr; |
10901 | if (!osr->deferred_pending) { | |
181888fb | 10902 | dout(20) << __func__ << " dequeueing" << dendl; |
7c673cae FG |
10903 | auto q = deferred_queue.iterator_to(*osr); |
10904 | deferred_queue.erase(q); | |
10905 | } else if (deferred_aggressive) { | |
224ce89b | 10906 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; |
3efd9988 | 10907 | deferred_finisher.queue(new C_DeferredTrySubmit(this)); |
181888fb FG |
10908 | } else { |
10909 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
7c673cae FG |
10910 | } |
10911 | } | |
10912 | ||
10913 | { | |
31f18b77 | 10914 | uint64_t costs = 0; |
11fdf7f2 TL |
10915 | { |
10916 | std::lock_guard l2(osr->qlock); | |
10917 | for (auto& i : b->txcs) { | |
10918 | TransContext *txc = &i; | |
10919 | txc->log_state_latency(logger, l_bluestore_state_deferred_aio_wait_lat); | |
10920 | txc->state = TransContext::STATE_DEFERRED_CLEANUP; | |
10921 | costs += txc->cost; | |
10922 | } | |
7c673cae | 10923 | } |
31f18b77 | 10924 | throttle_deferred_bytes.put(costs); |
11fdf7f2 | 10925 | std::lock_guard l(kv_lock); |
7c673cae FG |
10926 | deferred_done_queue.emplace_back(b); |
10927 | } | |
10928 | ||
10929 | // in the normal case, do not bother waking up the kv thread; it will | |
10930 | // catch us on the next commit anyway. | |
10931 | if (deferred_aggressive) { | |
11fdf7f2 | 10932 | std::lock_guard l(kv_lock); |
7c673cae FG |
10933 | kv_cond.notify_one(); |
10934 | } | |
10935 | } | |
10936 | ||
10937 | int BlueStore::_deferred_replay() | |
10938 | { | |
10939 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
10940 | int count = 0; |
10941 | int r = 0; | |
11fdf7f2 TL |
10942 | CollectionRef ch = _get_collection(coll_t::meta()); |
10943 | bool fake_ch = false; | |
10944 | if (!ch) { | |
10945 | // hmm, replaying initial mkfs? | |
10946 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
10947 | fake_ch = true; | |
10948 | } | |
10949 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
10950 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
10951 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
10952 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
10953 | << dendl; | |
10954 | bluestore_deferred_transaction_t *deferred_txn = | |
10955 | new bluestore_deferred_transaction_t; | |
10956 | bufferlist bl = it->value(); | |
11fdf7f2 | 10957 | auto p = bl.cbegin(); |
7c673cae | 10958 | try { |
11fdf7f2 | 10959 | decode(*deferred_txn, p); |
7c673cae FG |
10960 | } catch (buffer::error& e) { |
10961 | derr << __func__ << " failed to decode deferred txn " | |
10962 | << pretty_binary_string(it->key()) << dendl; | |
10963 | delete deferred_txn; | |
10964 | r = -EIO; | |
10965 | goto out; | |
10966 | } | |
11fdf7f2 | 10967 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); |
7c673cae FG |
10968 | txc->deferred_txn = deferred_txn; |
10969 | txc->state = TransContext::STATE_KV_DONE; | |
10970 | _txc_state_proc(txc); | |
10971 | } | |
10972 | out: | |
10973 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 10974 | _osr_register_zombie(osr); |
7c673cae | 10975 | _osr_drain_all(); |
11fdf7f2 TL |
10976 | if (fake_ch) { |
10977 | new_coll_map.clear(); | |
10978 | } | |
7c673cae FG |
10979 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
10980 | return r; | |
10981 | } | |
10982 | ||
10983 | // --------------------------- | |
10984 | // transactions | |
10985 | ||
10986 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
10987 | CollectionHandle& ch, |
10988 | vector<Transaction>& tls, | |
10989 | TrackedOpRef op, | |
10990 | ThreadPool::TPHandle *handle) | |
10991 | { | |
10992 | FUNCTRACE(cct); | |
10993 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 10994 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 10995 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae FG |
10996 | |
10997 | if (cct->_conf->objectstore_blackhole) { | |
10998 | dout(0) << __func__ << " objectstore_blackhole = TRUE, dropping transaction" | |
10999 | << dendl; | |
11fdf7f2 TL |
11000 | for (auto& l : { on_applied, on_commit, on_applied_sync }) { |
11001 | for (auto c : l) { | |
11002 | delete c; | |
11003 | } | |
11004 | } | |
7c673cae FG |
11005 | return 0; |
11006 | } | |
11fdf7f2 TL |
11007 | auto start = mono_clock::now(); |
11008 | ||
11009 | Collection *c = static_cast<Collection*>(ch.get()); | |
11010 | OpSequencer *osr = c->osr.get(); | |
11011 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae FG |
11012 | |
11013 | // prepare | |
11fdf7f2 TL |
11014 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, |
11015 | &on_commit); | |
7c673cae FG |
11016 | |
11017 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { | |
7c673cae FG |
11018 | txc->bytes += (*p).get_num_bytes(); |
11019 | _txc_add_transaction(txc, &(*p)); | |
11020 | } | |
11021 | _txc_calc_cost(txc); | |
11022 | ||
11023 | _txc_write_nodes(txc, txc->t); | |
11024 | ||
11025 | // journal deferred items | |
11026 | if (txc->deferred_txn) { | |
11027 | txc->deferred_txn->seq = ++deferred_seq; | |
11028 | bufferlist bl; | |
11fdf7f2 | 11029 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
11030 | string key; |
11031 | get_deferred_key(txc->deferred_txn->seq, &key); | |
11032 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
11033 | } | |
11034 | ||
11035 | _txc_finalize_kv(txc, txc->t); | |
11036 | if (handle) | |
11037 | handle->suspend_tp_timeout(); | |
11038 | ||
11fdf7f2 | 11039 | auto tstart = mono_clock::now(); |
7c673cae FG |
11040 | throttle_bytes.get(txc->cost); |
11041 | if (txc->deferred_txn) { | |
11042 | // ensure we do not block here because of deferred writes | |
11043 | if (!throttle_deferred_bytes.get_or_fail(txc->cost)) { | |
d2e6a577 FG |
11044 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
11045 | << dendl; | |
11046 | ++deferred_aggressive; | |
7c673cae | 11047 | deferred_try_submit(); |
3efd9988 FG |
11048 | { |
11049 | // wake up any previously finished deferred events | |
11fdf7f2 | 11050 | std::lock_guard l(kv_lock); |
3efd9988 FG |
11051 | kv_cond.notify_one(); |
11052 | } | |
7c673cae | 11053 | throttle_deferred_bytes.get(txc->cost); |
d2e6a577 FG |
11054 | --deferred_aggressive; |
11055 | } | |
7c673cae | 11056 | } |
11fdf7f2 | 11057 | auto tend = mono_clock::now(); |
7c673cae FG |
11058 | |
11059 | if (handle) | |
11060 | handle->reset_tp_timeout(); | |
11061 | ||
11062 | logger->inc(l_bluestore_txc); | |
11063 | ||
11064 | // execute (start) | |
11065 | _txc_state_proc(txc); | |
11066 | ||
11fdf7f2 TL |
11067 | // we're immediately readable (unlike FileStore) |
11068 | for (auto c : on_applied_sync) { | |
11069 | c->complete(0); | |
11070 | } | |
11071 | if (!on_applied.empty()) { | |
11072 | if (c->commit_queue) { | |
11073 | c->commit_queue->queue(on_applied); | |
11074 | } else { | |
11075 | finisher.queue(on_applied); | |
11076 | } | |
11077 | } | |
11078 | ||
11079 | LOG_LATENCY(logger, cct, l_bluestore_submit_lat, mono_clock::now() - start); | |
11080 | LOG_LATENCY(logger, cct, l_bluestore_throttle_lat, tend - tstart); | |
7c673cae FG |
11081 | return 0; |
11082 | } | |
11083 | ||
11084 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
11085 | { | |
11086 | dout(10) << __func__ << " txc " << txc << dendl; | |
11087 | bdev->aio_submit(&txc->ioc); | |
11088 | } | |
11089 | ||
11090 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
11091 | { | |
11092 | Transaction::iterator i = t->begin(); | |
11093 | ||
81eedcae | 11094 | _dump_transaction<30>(cct, t); |
7c673cae FG |
11095 | |
11096 | vector<CollectionRef> cvec(i.colls.size()); | |
11097 | unsigned j = 0; | |
11098 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
11099 | ++p, ++j) { | |
11100 | cvec[j] = _get_collection(*p); | |
7c673cae | 11101 | } |
11fdf7f2 | 11102 | |
7c673cae FG |
11103 | vector<OnodeRef> ovec(i.objects.size()); |
11104 | ||
11105 | for (int pos = 0; i.have_op(); ++pos) { | |
11106 | Transaction::Op *op = i.decode_op(); | |
11107 | int r = 0; | |
11108 | ||
11109 | // no coll or obj | |
11110 | if (op->op == Transaction::OP_NOP) | |
11111 | continue; | |
11112 | ||
11fdf7f2 | 11113 | |
7c673cae FG |
11114 | // collection operations |
11115 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
11116 | |
11117 | // initialize osd_pool_id and do a smoke test that all collections belong | |
11118 | // to the same pool | |
11119 | spg_t pgid; | |
11120 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
11121 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
11122 | txc->osd_pool_id == pgid.pool()); | |
11123 | txc->osd_pool_id = pgid.pool(); | |
11124 | } | |
11125 | ||
7c673cae FG |
11126 | switch (op->op) { |
11127 | case Transaction::OP_RMCOLL: | |
11128 | { | |
11129 | const coll_t &cid = i.get_cid(op->cid); | |
11130 | r = _remove_collection(txc, cid, &c); | |
11131 | if (!r) | |
11132 | continue; | |
11133 | } | |
11134 | break; | |
11135 | ||
11136 | case Transaction::OP_MKCOLL: | |
11137 | { | |
11fdf7f2 | 11138 | ceph_assert(!c); |
7c673cae FG |
11139 | const coll_t &cid = i.get_cid(op->cid); |
11140 | r = _create_collection(txc, cid, op->split_bits, &c); | |
11141 | if (!r) | |
11142 | continue; | |
11143 | } | |
11144 | break; | |
11145 | ||
11146 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 11147 | ceph_abort_msg("deprecated"); |
7c673cae FG |
11148 | break; |
11149 | ||
11150 | case Transaction::OP_SPLIT_COLLECTION2: | |
11151 | { | |
11152 | uint32_t bits = op->split_bits; | |
11153 | uint32_t rem = op->split_rem; | |
11154 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
11155 | if (!r) | |
11156 | continue; | |
11157 | } | |
11158 | break; | |
11159 | ||
11fdf7f2 TL |
11160 | case Transaction::OP_MERGE_COLLECTION: |
11161 | { | |
11162 | uint32_t bits = op->split_bits; | |
11163 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
11164 | if (!r) | |
11165 | continue; | |
11166 | } | |
11167 | break; | |
11168 | ||
7c673cae FG |
11169 | case Transaction::OP_COLL_HINT: |
11170 | { | |
11171 | uint32_t type = op->hint_type; | |
11172 | bufferlist hint; | |
11173 | i.decode_bl(hint); | |
11fdf7f2 | 11174 | auto hiter = hint.cbegin(); |
7c673cae FG |
11175 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
11176 | uint32_t pg_num; | |
11177 | uint64_t num_objs; | |
11fdf7f2 TL |
11178 | decode(pg_num, hiter); |
11179 | decode(num_objs, hiter); | |
7c673cae FG |
11180 | dout(10) << __func__ << " collection hint objects is a no-op, " |
11181 | << " pg_num " << pg_num << " num_objects " << num_objs | |
11182 | << dendl; | |
11183 | } else { | |
11184 | // Ignore the hint | |
11185 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
11186 | } | |
11187 | continue; | |
11188 | } | |
11189 | break; | |
11190 | ||
11191 | case Transaction::OP_COLL_SETATTR: | |
11192 | r = -EOPNOTSUPP; | |
11193 | break; | |
11194 | ||
11195 | case Transaction::OP_COLL_RMATTR: | |
11196 | r = -EOPNOTSUPP; | |
11197 | break; | |
11198 | ||
11199 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 11200 | ceph_abort_msg("not implemented"); |
7c673cae FG |
11201 | break; |
11202 | } | |
11203 | if (r < 0) { | |
11204 | derr << __func__ << " error " << cpp_strerror(r) | |
11205 | << " not handled on operation " << op->op | |
11206 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 11207 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 11208 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
11209 | } |
11210 | ||
11211 | // these operations implicity create the object | |
11212 | bool create = false; | |
11213 | if (op->op == Transaction::OP_TOUCH || | |
11214 | op->op == Transaction::OP_WRITE || | |
11215 | op->op == Transaction::OP_ZERO) { | |
11216 | create = true; | |
11217 | } | |
11218 | ||
11219 | // object operations | |
11220 | RWLock::WLocker l(c->lock); | |
11221 | OnodeRef &o = ovec[op->oid]; | |
11222 | if (!o) { | |
11223 | ghobject_t oid = i.get_oid(op->oid); | |
11224 | o = c->get_onode(oid, create); | |
11225 | } | |
11226 | if (!create && (!o || !o->exists)) { | |
11227 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
11228 | << i.get_oid(op->oid) << dendl; | |
11229 | r = -ENOENT; | |
11230 | goto endop; | |
11231 | } | |
11232 | ||
11233 | switch (op->op) { | |
11234 | case Transaction::OP_TOUCH: | |
11235 | r = _touch(txc, c, o); | |
11236 | break; | |
11237 | ||
11238 | case Transaction::OP_WRITE: | |
11239 | { | |
11240 | uint64_t off = op->off; | |
11241 | uint64_t len = op->len; | |
11242 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
11243 | bufferlist bl; | |
11244 | i.decode_bl(bl); | |
11245 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
11246 | } | |
11247 | break; | |
11248 | ||
11249 | case Transaction::OP_ZERO: | |
11250 | { | |
11251 | uint64_t off = op->off; | |
11252 | uint64_t len = op->len; | |
11253 | r = _zero(txc, c, o, off, len); | |
11254 | } | |
11255 | break; | |
11256 | ||
11257 | case Transaction::OP_TRIMCACHE: | |
11258 | { | |
11259 | // deprecated, no-op | |
11260 | } | |
11261 | break; | |
11262 | ||
11263 | case Transaction::OP_TRUNCATE: | |
11264 | { | |
11265 | uint64_t off = op->off; | |
35e4c445 | 11266 | r = _truncate(txc, c, o, off); |
7c673cae FG |
11267 | } |
11268 | break; | |
11269 | ||
11270 | case Transaction::OP_REMOVE: | |
11271 | { | |
11272 | r = _remove(txc, c, o); | |
11273 | } | |
11274 | break; | |
11275 | ||
11276 | case Transaction::OP_SETATTR: | |
11277 | { | |
11278 | string name = i.decode_string(); | |
11279 | bufferptr bp; | |
11280 | i.decode_bp(bp); | |
11281 | r = _setattr(txc, c, o, name, bp); | |
11282 | } | |
11283 | break; | |
11284 | ||
11285 | case Transaction::OP_SETATTRS: | |
11286 | { | |
11287 | map<string, bufferptr> aset; | |
11288 | i.decode_attrset(aset); | |
11289 | r = _setattrs(txc, c, o, aset); | |
11290 | } | |
11291 | break; | |
11292 | ||
11293 | case Transaction::OP_RMATTR: | |
11294 | { | |
11295 | string name = i.decode_string(); | |
11296 | r = _rmattr(txc, c, o, name); | |
11297 | } | |
11298 | break; | |
11299 | ||
11300 | case Transaction::OP_RMATTRS: | |
11301 | { | |
11302 | r = _rmattrs(txc, c, o); | |
11303 | } | |
11304 | break; | |
11305 | ||
11306 | case Transaction::OP_CLONE: | |
11307 | { | |
11308 | OnodeRef& no = ovec[op->dest_oid]; | |
11309 | if (!no) { | |
11310 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
11311 | no = c->get_onode(noid, true); | |
11312 | } | |
11313 | r = _clone(txc, c, o, no); | |
11314 | } | |
11315 | break; | |
11316 | ||
11317 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 11318 | ceph_abort_msg("deprecated"); |
7c673cae FG |
11319 | break; |
11320 | ||
11321 | case Transaction::OP_CLONERANGE2: | |
11322 | { | |
11323 | OnodeRef& no = ovec[op->dest_oid]; | |
11324 | if (!no) { | |
11325 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
11326 | no = c->get_onode(noid, true); | |
11327 | } | |
11328 | uint64_t srcoff = op->off; | |
11329 | uint64_t len = op->len; | |
11330 | uint64_t dstoff = op->dest_off; | |
11331 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
11332 | } | |
11333 | break; | |
11334 | ||
11335 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 11336 | ceph_abort_msg("not implemented"); |
7c673cae FG |
11337 | break; |
11338 | ||
11339 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 11340 | ceph_abort_msg("not implemented"); |
7c673cae FG |
11341 | break; |
11342 | ||
11343 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 11344 | ceph_abort_msg("deprecated"); |
7c673cae FG |
11345 | break; |
11346 | ||
11347 | case Transaction::OP_COLL_MOVE_RENAME: | |
11348 | case Transaction::OP_TRY_RENAME: | |
11349 | { | |
11fdf7f2 | 11350 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
11351 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
11352 | OnodeRef& no = ovec[op->dest_oid]; | |
11353 | if (!no) { | |
11354 | no = c->get_onode(noid, false); | |
11355 | } | |
11356 | r = _rename(txc, c, o, no, noid); | |
11357 | } | |
11358 | break; | |
11359 | ||
11360 | case Transaction::OP_OMAP_CLEAR: | |
11361 | { | |
11362 | r = _omap_clear(txc, c, o); | |
11363 | } | |
11364 | break; | |
11365 | case Transaction::OP_OMAP_SETKEYS: | |
11366 | { | |
11367 | bufferlist aset_bl; | |
11368 | i.decode_attrset_bl(&aset_bl); | |
11369 | r = _omap_setkeys(txc, c, o, aset_bl); | |
11370 | } | |
11371 | break; | |
11372 | case Transaction::OP_OMAP_RMKEYS: | |
11373 | { | |
11374 | bufferlist keys_bl; | |
11375 | i.decode_keyset_bl(&keys_bl); | |
11376 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
11377 | } | |
11378 | break; | |
11379 | case Transaction::OP_OMAP_RMKEYRANGE: | |
11380 | { | |
11381 | string first, last; | |
11382 | first = i.decode_string(); | |
11383 | last = i.decode_string(); | |
11384 | r = _omap_rmkey_range(txc, c, o, first, last); | |
11385 | } | |
11386 | break; | |
11387 | case Transaction::OP_OMAP_SETHEADER: | |
11388 | { | |
11389 | bufferlist bl; | |
11390 | i.decode_bl(bl); | |
11391 | r = _omap_setheader(txc, c, o, bl); | |
11392 | } | |
11393 | break; | |
11394 | ||
11395 | case Transaction::OP_SETALLOCHINT: | |
11396 | { | |
11397 | r = _set_alloc_hint(txc, c, o, | |
11398 | op->expected_object_size, | |
11399 | op->expected_write_size, | |
11400 | op->alloc_hint_flags); | |
11401 | } | |
11402 | break; | |
11403 | ||
11404 | default: | |
11fdf7f2 | 11405 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
11406 | ceph_abort(); |
11407 | } | |
11408 | ||
11409 | endop: | |
11410 | if (r < 0) { | |
11411 | bool ok = false; | |
11412 | ||
11413 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
11414 | op->op == Transaction::OP_CLONE || | |
11415 | op->op == Transaction::OP_CLONERANGE2 || | |
11416 | op->op == Transaction::OP_COLL_ADD || | |
11417 | op->op == Transaction::OP_SETATTR || | |
11418 | op->op == Transaction::OP_SETATTRS || | |
11419 | op->op == Transaction::OP_RMATTR || | |
11420 | op->op == Transaction::OP_OMAP_SETKEYS || | |
11421 | op->op == Transaction::OP_OMAP_RMKEYS || | |
11422 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
11423 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
11424 | // -ENOENT is usually okay | |
11425 | ok = true; | |
11426 | if (r == -ENODATA) | |
11427 | ok = true; | |
11428 | ||
11429 | if (!ok) { | |
11430 | const char *msg = "unexpected error code"; | |
11431 | ||
11432 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
11433 | op->op == Transaction::OP_CLONE || | |
11434 | op->op == Transaction::OP_CLONERANGE2)) | |
11435 | msg = "ENOENT on clone suggests osd bug"; | |
11436 | ||
11437 | if (r == -ENOSPC) | |
11438 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
11439 | // by partially applying transactions. | |
11440 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
11441 | ||
11442 | if (r == -ENOTEMPTY) { | |
11443 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
11444 | } | |
11445 | ||
11446 | derr << __func__ << " error " << cpp_strerror(r) | |
11447 | << " not handled on operation " << op->op | |
11448 | << " (op " << pos << ", counting from 0)" | |
11449 | << dendl; | |
11450 | derr << msg << dendl; | |
81eedcae | 11451 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 11452 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
11453 | } |
11454 | } | |
11455 | } | |
11456 | } | |
11457 | ||
11458 | ||
11459 | ||
11460 | // ----------------- | |
11461 | // write operations | |
11462 | ||
11463 | int BlueStore::_touch(TransContext *txc, | |
11464 | CollectionRef& c, | |
11465 | OnodeRef &o) | |
11466 | { | |
11467 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
11468 | int r = 0; | |
7c673cae FG |
11469 | _assign_nid(txc, o); |
11470 | txc->write_onode(o); | |
11471 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
11472 | return r; | |
11473 | } | |
11474 | ||
7c673cae FG |
11475 | void BlueStore::_pad_zeros( |
11476 | bufferlist *bl, uint64_t *offset, | |
11477 | uint64_t chunk_size) | |
11478 | { | |
11479 | auto length = bl->length(); | |
11480 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
11481 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
11482 | dout(40) << "before:\n"; | |
11483 | bl->hexdump(*_dout); | |
11484 | *_dout << dendl; | |
11485 | // front | |
11486 | size_t front_pad = *offset % chunk_size; | |
11487 | size_t back_pad = 0; | |
11488 | size_t pad_count = 0; | |
11489 | if (front_pad) { | |
11fdf7f2 TL |
11490 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
11491 | bufferptr z = buffer::create_small_page_aligned(chunk_size); | |
224ce89b | 11492 | z.zero(0, front_pad, false); |
7c673cae | 11493 | pad_count += front_pad; |
224ce89b | 11494 | bl->copy(0, front_copy, z.c_str() + front_pad); |
7c673cae FG |
11495 | if (front_copy + front_pad < chunk_size) { |
11496 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 11497 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
11498 | pad_count += back_pad; |
11499 | } | |
11500 | bufferlist old, t; | |
11501 | old.swap(*bl); | |
11502 | t.substr_of(old, front_copy, length - front_copy); | |
11503 | bl->append(z); | |
11504 | bl->claim_append(t); | |
11505 | *offset -= front_pad; | |
224ce89b | 11506 | length += pad_count; |
7c673cae FG |
11507 | } |
11508 | ||
11509 | // back | |
11510 | uint64_t end = *offset + length; | |
11511 | unsigned back_copy = end % chunk_size; | |
11512 | if (back_copy) { | |
11fdf7f2 | 11513 | ceph_assert(back_pad == 0); |
7c673cae | 11514 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 11515 | ceph_assert(back_copy <= length); |
7c673cae | 11516 | bufferptr tail(chunk_size); |
224ce89b WB |
11517 | bl->copy(length - back_copy, back_copy, tail.c_str()); |
11518 | tail.zero(back_copy, back_pad, false); | |
7c673cae FG |
11519 | bufferlist old; |
11520 | old.swap(*bl); | |
11521 | bl->substr_of(old, 0, length - back_copy); | |
11522 | bl->append(tail); | |
11523 | length += back_pad; | |
11524 | pad_count += back_pad; | |
11525 | } | |
11526 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
11527 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
11528 | << length << std::dec << dendl; | |
11529 | dout(40) << "after:\n"; | |
11530 | bl->hexdump(*_dout); | |
11531 | *_dout << dendl; | |
11532 | if (pad_count) | |
11533 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 11534 | ceph_assert(bl->length() == length); |
7c673cae FG |
11535 | } |
11536 | ||
11537 | void BlueStore::_do_write_small( | |
11538 | TransContext *txc, | |
11539 | CollectionRef &c, | |
11540 | OnodeRef o, | |
11541 | uint64_t offset, uint64_t length, | |
11542 | bufferlist::iterator& blp, | |
11543 | WriteContext *wctx) | |
11544 | { | |
11545 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
11546 | << std::dec << dendl; | |
11fdf7f2 | 11547 | ceph_assert(length < min_alloc_size); |
7c673cae FG |
11548 | uint64_t end_offs = offset + length; |
11549 | ||
11550 | logger->inc(l_bluestore_write_small); | |
11551 | logger->inc(l_bluestore_write_small_bytes, length); | |
11552 | ||
11553 | bufferlist bl; | |
11554 | blp.copy(length, bl); | |
11555 | ||
81eedcae TL |
11556 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
11557 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
11558 | uint32_t alloc_len = min_alloc_size; | |
11559 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
11560 | ||
11561 | bool any_change; | |
11562 | ||
11563 | // search suitable extent in both forward and reverse direction in | |
11564 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
11565 | // then check if blob can be reused via can_reuse_blob func or apply | |
11566 | // direct/deferred write (the latter for extents including or higher | |
11567 | // than 'offset' only). | |
11568 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
11569 | ||
7c673cae FG |
11570 | // Look for an existing mutable blob we can use. |
11571 | auto begin = o->extent_map.extent_map.begin(); | |
11572 | auto end = o->extent_map.extent_map.end(); | |
11573 | auto ep = o->extent_map.seek_lextent(offset); | |
11574 | if (ep != begin) { | |
11575 | --ep; | |
11576 | if (ep->blob_end() <= offset) { | |
11577 | ++ep; | |
11578 | } | |
11579 | } | |
11580 | auto prev_ep = ep; | |
11581 | if (prev_ep != begin) { | |
11582 | --prev_ep; | |
11583 | } else { | |
11584 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
11585 | } | |
11586 | ||
7c673cae FG |
11587 | do { |
11588 | any_change = false; | |
11589 | ||
11590 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
11591 | BlobRef b = ep->blob; | |
11592 | auto bstart = ep->blob_start(); | |
11593 | dout(20) << __func__ << " considering " << *b | |
11594 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
11595 | if (bstart >= end_offs) { | |
11596 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
11597 | } else if (!b->get_blob().is_mutable()) { | |
11598 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
11599 | } else if (ep->logical_offset % min_alloc_size != | |
11600 | ep->blob_offset % min_alloc_size) { | |
11601 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
11602 | } else { | |
11603 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
11604 | // can we pad our head/tail out with zeros? | |
11605 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
11606 | head_pad = p2phase(offset, chunk_size); |
11607 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
11608 | if (head_pad || tail_pad) { |
11609 | o->extent_map.fault_range(db, offset - head_pad, | |
11610 | end_offs - offset + head_pad + tail_pad); | |
11611 | } | |
11612 | if (head_pad && | |
11613 | o->extent_map.has_any_lextents(offset - head_pad, chunk_size)) { | |
11614 | head_pad = 0; | |
11615 | } | |
11616 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
11617 | tail_pad = 0; | |
11618 | } | |
11619 | ||
11620 | uint64_t b_off = offset - head_pad - bstart; | |
11621 | uint64_t b_len = length + head_pad + tail_pad; | |
11622 | ||
11623 | // direct write into unused blocks of an existing mutable blob? | |
11624 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
11625 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
11626 | b->get_blob().is_unused(b_off, b_len) && | |
11627 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 11628 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
11629 | |
11630 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
11631 | << b_off << "~" << b_len | |
11632 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
11633 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 11634 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
11635 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
11636 | ||
11fdf7f2 | 11637 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
11638 | if (b_len <= prefer_deferred_size) { |
11639 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
11640 | << b_len << std::dec << " unused write via deferred" << dendl; | |
11641 | bluestore_deferred_op_t *op = _get_deferred_op(txc, o); | |
11642 | op->op = bluestore_deferred_op_t::OP_WRITE; | |
11643 | b->get_blob().map( | |
11644 | b_off, b_len, | |
11645 | [&](uint64_t offset, uint64_t length) { | |
11646 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
11647 | return 0; | |
11648 | }); | |
224ce89b | 11649 | op->data = bl; |
7c673cae FG |
11650 | } else { |
11651 | b->get_blob().map_bl( | |
224ce89b | 11652 | b_off, bl, |
7c673cae FG |
11653 | [&](uint64_t offset, bufferlist& t) { |
11654 | bdev->aio_write(offset, t, | |
11655 | &txc->ioc, wctx->buffered); | |
11656 | }); | |
11657 | } | |
11658 | } | |
224ce89b | 11659 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
11660 | dout(20) << __func__ << " lex old " << *ep << dendl; |
11661 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
11662 | b, | |
11663 | &wctx->old_extents); | |
11664 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
11665 | txc->statfs_delta.stored() += le->length; | |
11666 | dout(20) << __func__ << " lex " << *le << dendl; | |
11667 | logger->inc(l_bluestore_write_small_unused); | |
11668 | return; | |
11669 | } | |
11670 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
11671 | uint64_t head_read = p2phase(b_off, chunk_size); |
11672 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
11673 | if ((head_read || tail_read) && |
11674 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
11675 | head_read + tail_read < min_alloc_size) { | |
11676 | b_off -= head_read; | |
11677 | b_len += head_read + tail_read; | |
11678 | ||
11679 | } else { | |
11680 | head_read = tail_read = 0; | |
11681 | } | |
11682 | ||
11683 | // chunk-aligned deferred overwrite? | |
11684 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
11685 | b_off % chunk_size == 0 && | |
11686 | b_len % chunk_size == 0 && | |
11687 | b->get_blob().is_allocated(b_off, b_len)) { | |
11688 | ||
224ce89b | 11689 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
11690 | |
11691 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
11692 | << " and tail 0x" << tail_read << std::dec << dendl; | |
11693 | if (head_read) { | |
11694 | bufferlist head_bl; | |
11695 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
11696 | head_bl, 0); | |
11fdf7f2 | 11697 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
11698 | size_t zlen = head_read - r; |
11699 | if (zlen) { | |
11700 | head_bl.append_zero(zlen); | |
11701 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
11702 | } | |
11fdf7f2 TL |
11703 | head_bl.claim_append(bl); |
11704 | bl.swap(head_bl); | |
7c673cae FG |
11705 | logger->inc(l_bluestore_write_penalty_read_ops); |
11706 | } | |
11707 | if (tail_read) { | |
11708 | bufferlist tail_bl; | |
11709 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
11710 | tail_bl, 0); | |
11fdf7f2 | 11711 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
11712 | size_t zlen = tail_read - r; |
11713 | if (zlen) { | |
11714 | tail_bl.append_zero(zlen); | |
11715 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
11716 | } | |
224ce89b | 11717 | bl.claim_append(tail_bl); |
7c673cae FG |
11718 | logger->inc(l_bluestore_write_penalty_read_ops); |
11719 | } | |
11720 | logger->inc(l_bluestore_write_small_pre_read); | |
11721 | ||
224ce89b | 11722 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
11723 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
11724 | ||
7c673cae | 11725 | if (b->get_blob().csum_type) { |
224ce89b | 11726 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae | 11727 | } |
11fdf7f2 TL |
11728 | |
11729 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
11730 | bluestore_deferred_op_t *op = _get_deferred_op(txc, o); | |
11731 | op->op = bluestore_deferred_op_t::OP_WRITE; | |
11732 | int r = b->get_blob().map( | |
11733 | b_off, b_len, | |
11734 | [&](uint64_t offset, uint64_t length) { | |
11735 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
11736 | return 0; | |
11737 | }); | |
11738 | ceph_assert(r == 0); | |
11739 | op->data.claim(bl); | |
11740 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" | |
11741 | << b_len << std::dec << " of mutable " << *b | |
11742 | << " at " << op->extents << dendl; | |
11743 | } | |
11744 | ||
7c673cae FG |
11745 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
11746 | b, &wctx->old_extents); | |
11747 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
11748 | txc->statfs_delta.stored() += le->length; | |
11749 | dout(20) << __func__ << " lex " << *le << dendl; | |
11750 | logger->inc(l_bluestore_write_small_deferred); | |
11751 | return; | |
11752 | } | |
224ce89b WB |
11753 | // try to reuse blob if we can |
11754 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
11755 | max_bsize, |
11756 | offset0 - bstart, | |
11757 | &alloc_len)) { | |
11fdf7f2 | 11758 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
11759 | // fit into reused blob |
11760 | // Need to check for pending writes desiring to | |
11761 | // reuse the same pextent. The rationale is that during GC two chunks | |
11762 | // from garbage blobs(compressed?) can share logical space within the same | |
11763 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
11764 | // Hence the second write will fail in an attempt to reuse blob at | |
11765 | // do_alloc_write(). | |
11766 | if (!wctx->has_conflict(b, | |
11767 | offset0, | |
11768 | offset0 + alloc_len, | |
11769 | min_alloc_size)) { | |
11770 | ||
11771 | // we can't reuse pad_head/pad_tail since they might be truncated | |
11772 | // due to existent extents | |
11773 | uint64_t b_off = offset - bstart; | |
11774 | uint64_t b_off0 = b_off; | |
11775 | _pad_zeros(&bl, &b_off0, chunk_size); | |
11776 | ||
11777 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
11778 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
11779 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
11780 | << std::dec << dendl; |
11781 | ||
11782 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11783 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
11784 | false, false); | |
11785 | logger->inc(l_bluestore_write_small_unused); | |
11786 | return; | |
11787 | } | |
11788 | } | |
11789 | } | |
11790 | ++ep; | |
11791 | any_change = true; | |
11792 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
11793 | ||
11794 | // check extent for reuse in reverse order | |
11795 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
11796 | BlobRef b = prev_ep->blob; | |
11797 | auto bstart = prev_ep->blob_start(); | |
11798 | dout(20) << __func__ << " considering " << *b | |
11799 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 11800 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
11801 | max_bsize, |
11802 | offset0 - bstart, | |
11803 | &alloc_len)) { | |
11fdf7f2 | 11804 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
11805 | // fit into reused blob |
11806 | // Need to check for pending writes desiring to | |
11807 | // reuse the same pextent. The rationale is that during GC two chunks | |
11808 | // from garbage blobs(compressed?) can share logical space within the same | |
11809 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
11810 | // Hence the second write will fail in an attempt to reuse blob at | |
11811 | // do_alloc_write(). | |
11812 | if (!wctx->has_conflict(b, | |
11813 | offset0, | |
11814 | offset0 + alloc_len, | |
11815 | min_alloc_size)) { | |
11816 | ||
11817 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
11818 | uint64_t b_off = offset - bstart; | |
11819 | uint64_t b_off0 = b_off; | |
11820 | _pad_zeros(&bl, &b_off0, chunk_size); | |
11821 | ||
11822 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
11823 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
11824 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
11825 | << std::dec << dendl; |
11826 | ||
11827 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11828 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
11829 | false, false); | |
11830 | logger->inc(l_bluestore_write_small_unused); | |
11831 | return; | |
11832 | } | |
11833 | } | |
11834 | if (prev_ep != begin) { | |
11835 | --prev_ep; | |
11836 | any_change = true; | |
11837 | } else { | |
11838 | prev_ep = end; // to avoid useless first extent re-check | |
11839 | } | |
11840 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
11841 | } while (any_change); | |
11842 | ||
11843 | // new blob. | |
7c673cae | 11844 | BlobRef b = c->new_blob(); |
11fdf7f2 | 11845 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae FG |
11846 | uint64_t b_off0 = b_off; |
11847 | _pad_zeros(&bl, &b_off0, block_size); | |
11848 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11849 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, true, true); | |
7c673cae FG |
11850 | |
11851 | return; | |
11852 | } | |
11853 | ||
11854 | void BlueStore::_do_write_big( | |
11855 | TransContext *txc, | |
11856 | CollectionRef &c, | |
11857 | OnodeRef o, | |
11858 | uint64_t offset, uint64_t length, | |
11859 | bufferlist::iterator& blp, | |
11860 | WriteContext *wctx) | |
11861 | { | |
11862 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
11863 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
11864 | << " compress " << (int)wctx->compress | |
11865 | << dendl; | |
11866 | logger->inc(l_bluestore_write_big); | |
11867 | logger->inc(l_bluestore_write_big_bytes, length); | |
11868 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11fdf7f2 | 11869 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae FG |
11870 | while (length > 0) { |
11871 | bool new_blob = false; | |
11fdf7f2 | 11872 | uint32_t l = std::min(max_bsize, length); |
7c673cae FG |
11873 | BlobRef b; |
11874 | uint32_t b_off = 0; | |
11875 | ||
11876 | //attempting to reuse existing blob | |
11877 | if (!wctx->compress) { | |
11878 | // look for an existing mutable blob we can reuse | |
11879 | auto begin = o->extent_map.extent_map.begin(); | |
11880 | auto end = o->extent_map.extent_map.end(); | |
11881 | auto ep = o->extent_map.seek_lextent(offset); | |
11882 | auto prev_ep = ep; | |
11883 | if (prev_ep != begin) { | |
11884 | --prev_ep; | |
11885 | } else { | |
11886 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
11887 | } | |
11888 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
11889 | // search suitable extent in both forward and reverse direction in | |
11890 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 11891 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
11892 | bool any_change; |
11893 | do { | |
11894 | any_change = false; | |
11895 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
11896 | if (offset >= ep->blob_start() && | |
224ce89b | 11897 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
11898 | offset - ep->blob_start(), |
11899 | &l)) { | |
11900 | b = ep->blob; | |
11901 | b_off = offset - ep->blob_start(); | |
11902 | prev_ep = end; // to avoid check below | |
11903 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 11904 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
11905 | } else { |
11906 | ++ep; | |
11907 | any_change = true; | |
11908 | } | |
11909 | } | |
11910 | ||
11911 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
224ce89b | 11912 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
11913 | offset - prev_ep->blob_start(), |
11914 | &l)) { | |
11915 | b = prev_ep->blob; | |
11916 | b_off = offset - prev_ep->blob_start(); | |
11917 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 11918 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
11919 | } else if (prev_ep != begin) { |
11920 | --prev_ep; | |
11921 | any_change = true; | |
11922 | } else { | |
11923 | prev_ep = end; // to avoid useless first extent re-check | |
11924 | } | |
11925 | } | |
11926 | } while (b == nullptr && any_change); | |
11927 | } | |
11928 | if (b == nullptr) { | |
11929 | b = c->new_blob(); | |
11930 | b_off = 0; | |
11931 | new_blob = true; | |
11932 | } | |
11933 | ||
11934 | bufferlist t; | |
11935 | blp.copy(l, t); | |
11936 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); | |
11937 | offset += l; | |
11938 | length -= l; | |
11939 | logger->inc(l_bluestore_write_big_blobs); | |
11940 | } | |
11941 | } | |
11942 | ||
11943 | int BlueStore::_do_alloc_write( | |
11944 | TransContext *txc, | |
11945 | CollectionRef coll, | |
11946 | OnodeRef o, | |
11947 | WriteContext *wctx) | |
11948 | { | |
11949 | dout(20) << __func__ << " txc " << txc | |
11950 | << " " << wctx->writes.size() << " blobs" | |
11951 | << dendl; | |
3efd9988 FG |
11952 | if (wctx->writes.empty()) { |
11953 | return 0; | |
7c673cae FG |
11954 | } |
11955 | ||
7c673cae FG |
11956 | CompressorRef c; |
11957 | double crr = 0; | |
11958 | if (wctx->compress) { | |
11959 | c = select_option( | |
11960 | "compression_algorithm", | |
11961 | compressor, | |
11962 | [&]() { | |
11963 | string val; | |
11964 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
11965 | CompressorRef cp = compressor; | |
11966 | if (!cp || cp->get_type_name() != val) { | |
11967 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
11968 | if (!cp) { |
11969 | if (_set_compression_alert(false, val.c_str())) { | |
11970 | derr << __func__ << " unable to initialize " << val.c_str() | |
11971 | << " compressor" << dendl; | |
11972 | } | |
11973 | } | |
7c673cae FG |
11974 | } |
11975 | return boost::optional<CompressorRef>(cp); | |
11976 | } | |
11977 | return boost::optional<CompressorRef>(); | |
11978 | } | |
11979 | ); | |
11980 | ||
11981 | crr = select_option( | |
11982 | "compression_required_ratio", | |
11983 | cct->_conf->bluestore_compression_required_ratio, | |
11984 | [&]() { | |
11985 | double val; | |
3efd9988 | 11986 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
7c673cae FG |
11987 | return boost::optional<double>(val); |
11988 | } | |
11989 | return boost::optional<double>(); | |
11990 | } | |
11991 | ); | |
11992 | } | |
11993 | ||
11994 | // checksum | |
11fdf7f2 | 11995 | int64_t csum = csum_type.load(); |
7c673cae FG |
11996 | csum = select_option( |
11997 | "csum_type", | |
11998 | csum, | |
11999 | [&]() { | |
11fdf7f2 | 12000 | int64_t val; |
3efd9988 | 12001 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
11fdf7f2 | 12002 | return boost::optional<int64_t>(val); |
7c673cae | 12003 | } |
11fdf7f2 | 12004 | return boost::optional<int64_t>(); |
7c673cae FG |
12005 | } |
12006 | ); | |
12007 | ||
3efd9988 FG |
12008 | // compress (as needed) and calc needed space |
12009 | uint64_t need = 0; | |
11fdf7f2 | 12010 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 12011 | for (auto& wi : wctx->writes) { |
3efd9988 | 12012 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 12013 | auto start = mono_clock::now(); |
7c673cae FG |
12014 | |
12015 | // compress | |
11fdf7f2 TL |
12016 | ceph_assert(wi.b_off == 0); |
12017 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 12018 | |
7c673cae FG |
12019 | // FIXME: memory alignment here is bad |
12020 | bufferlist t; | |
3efd9988 | 12021 | int r = c->compress(wi.bl, t); |
3efd9988 | 12022 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 12023 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
12024 | bool rejected = false; |
12025 | uint64_t compressed_len = t.length(); | |
12026 | // do an approximate (fast) estimation for resulting blob size | |
12027 | // that doesn't take header overhead into account | |
11fdf7f2 | 12028 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
12029 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
12030 | bluestore_compression_header_t chdr; | |
12031 | chdr.type = c->get_type(); | |
12032 | chdr.length = t.length(); | |
12033 | encode(chdr, wi.compressed_bl); | |
12034 | wi.compressed_bl.claim_append(t); | |
12035 | ||
12036 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 12037 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
12038 | if (result_len <= want_len && result_len < wi.blob_length) { |
12039 | // Cool. We compressed at least as much as we were hoping to. | |
12040 | // pad out to min_alloc_size | |
12041 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
12042 | wi.compressed_len = compressed_len; | |
12043 | wi.compressed = true; | |
12044 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
12045 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
12046 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
12047 | << " with " << c->get_type() | |
12048 | << std::dec << dendl; | |
12049 | txc->statfs_delta.compressed() += compressed_len; | |
12050 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
12051 | txc->statfs_delta.compressed_allocated() += result_len; | |
12052 | logger->inc(l_bluestore_compress_success_count); | |
12053 | need += result_len; | |
12054 | } else { | |
12055 | rejected = true; | |
12056 | } | |
12057 | } else if (r != 0) { | |
12058 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
12059 | << " bytes compressed using " << c->get_type_name() | |
12060 | << std::dec | |
12061 | << " failed with errcode = " << r | |
12062 | << ", leaving uncompressed" | |
12063 | << dendl; | |
12064 | logger->inc(l_bluestore_compress_rejected_count); | |
12065 | need += wi.blob_length; | |
7c673cae | 12066 | } else { |
a8e16298 TL |
12067 | rejected = true; |
12068 | } | |
12069 | ||
12070 | if (rejected) { | |
3efd9988 | 12071 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 12072 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
12073 | << " with " << c->get_type() |
12074 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 12075 | << " -> 0x" << want_len |
3efd9988 FG |
12076 | << ", leaving uncompressed" |
12077 | << std::dec << dendl; | |
12078 | logger->inc(l_bluestore_compress_rejected_count); | |
12079 | need += wi.blob_length; | |
7c673cae | 12080 | } |
11fdf7f2 TL |
12081 | LOG_LATENCY(logger, cct, l_bluestore_compress_lat, |
12082 | mono_clock::now() - start); | |
3efd9988 FG |
12083 | } else { |
12084 | need += wi.blob_length; | |
7c673cae | 12085 | } |
3efd9988 | 12086 | } |
a8e16298 | 12087 | PExtentVector prealloc; |
3efd9988 | 12088 | prealloc.reserve(2 * wctx->writes.size());; |
11fdf7f2 | 12089 | int64_t prealloc_left = 0; |
3efd9988 FG |
12090 | prealloc_left = alloc->allocate( |
12091 | need, min_alloc_size, need, | |
12092 | 0, &prealloc); | |
11fdf7f2 TL |
12093 | if (prealloc_left < (int64_t)need) { |
12094 | derr << __func__ << " failed to allocate 0x" << std::hex << need | |
12095 | << " allocated 0x " << prealloc_left | |
12096 | << " min_alloc_size 0x" << min_alloc_size | |
12097 | << " available 0x " << alloc->get_free() | |
12098 | << std::dec << dendl; | |
12099 | if (prealloc.size()) { | |
12100 | alloc->release(prealloc); | |
12101 | } | |
a8e16298 TL |
12102 | return -ENOSPC; |
12103 | } | |
a8e16298 | 12104 | |
3efd9988 FG |
12105 | dout(20) << __func__ << " prealloc " << prealloc << dendl; |
12106 | auto prealloc_pos = prealloc.begin(); | |
12107 | ||
12108 | for (auto& wi : wctx->writes) { | |
12109 | BlobRef b = wi.b; | |
12110 | bluestore_blob_t& dblob = b->dirty_blob(); | |
12111 | uint64_t b_off = wi.b_off; | |
12112 | bufferlist *l = &wi.bl; | |
12113 | uint64_t final_length = wi.blob_length; | |
12114 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
12115 | if (wi.compressed) { |
12116 | final_length = wi.compressed_bl.length(); | |
12117 | csum_length = final_length; | |
3efd9988 FG |
12118 | l = &wi.compressed_bl; |
12119 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
12120 | } else if (wi.new_blob) { | |
7c673cae | 12121 | // initialize newly created blob only |
11fdf7f2 TL |
12122 | ceph_assert(dblob.is_mutable()); |
12123 | unsigned csum_order; | |
7c673cae FG |
12124 | if (l->length() != wi.blob_length) { |
12125 | // hrm, maybe we could do better here, but let's not bother. | |
12126 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
12127 | << block_size_order << dendl; | |
31f18b77 | 12128 | csum_order = block_size_order; |
7c673cae FG |
12129 | } else { |
12130 | csum_order = std::min(wctx->csum_order, ctz(l->length())); | |
12131 | } | |
12132 | // try to align blob with max_blob_size to improve | |
12133 | // its reuse ratio, e.g. in case of reverse write | |
12134 | uint32_t suggested_boff = | |
12135 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
12136 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
12137 | suggested_boff + final_length <= max_bsize && | |
12138 | suggested_boff > b_off) { | |
181888fb | 12139 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 12140 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 12141 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
12142 | csum_length += suggested_boff - b_off; |
12143 | b_off = suggested_boff; | |
12144 | } | |
181888fb FG |
12145 | if (csum != Checksummer::CSUM_NONE) { |
12146 | dout(20) << __func__ << " initialize csum setting for new blob " << *b | |
12147 | << " csum_type " << Checksummer::get_csum_type_string(csum) | |
12148 | << " csum_order " << csum_order | |
12149 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
12150 | << dendl; | |
12151 | dblob.init_csum(csum, csum_order, csum_length); | |
12152 | } | |
7c673cae FG |
12153 | } |
12154 | ||
a8e16298 | 12155 | PExtentVector extents; |
3efd9988 FG |
12156 | int64_t left = final_length; |
12157 | while (left > 0) { | |
11fdf7f2 | 12158 | ceph_assert(prealloc_left > 0); |
3efd9988 FG |
12159 | if (prealloc_pos->length <= left) { |
12160 | prealloc_left -= prealloc_pos->length; | |
12161 | left -= prealloc_pos->length; | |
12162 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
12163 | extents.push_back(*prealloc_pos); | |
12164 | ++prealloc_pos; | |
12165 | } else { | |
12166 | extents.emplace_back(prealloc_pos->offset, left); | |
12167 | prealloc_pos->offset += left; | |
12168 | prealloc_pos->length -= left; | |
12169 | prealloc_left -= left; | |
12170 | txc->statfs_delta.allocated() += left; | |
12171 | left = 0; | |
12172 | break; | |
12173 | } | |
12174 | } | |
7c673cae | 12175 | for (auto& p : extents) { |
3efd9988 | 12176 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 12177 | } |
11fdf7f2 | 12178 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 12179 | |
181888fb FG |
12180 | dout(20) << __func__ << " blob " << *b << dendl; |
12181 | if (dblob.has_csum()) { | |
7c673cae FG |
12182 | dblob.calc_csum(b_off, *l); |
12183 | } | |
181888fb | 12184 | |
7c673cae FG |
12185 | if (wi.mark_unused) { |
12186 | auto b_end = b_off + wi.bl.length(); | |
12187 | if (b_off) { | |
12188 | dblob.add_unused(0, b_off); | |
12189 | } | |
12190 | if (b_end < wi.blob_length) { | |
12191 | dblob.add_unused(b_end, wi.blob_length - b_end); | |
12192 | } | |
12193 | } | |
12194 | ||
12195 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
12196 | b_off + (wi.b_off0 - wi.b_off), | |
12197 | wi.length0, | |
12198 | wi.b, | |
12199 | nullptr); | |
12200 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
12201 | txc->statfs_delta.stored() += le->length; | |
12202 | dout(20) << __func__ << " lex " << *le << dendl; | |
12203 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
12204 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
12205 | ||
12206 | // queue io | |
11fdf7f2 | 12207 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
12208 | if (l->length() <= prefer_deferred_size.load()) { |
12209 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
12210 | << l->length() << std::dec << " write via deferred" << dendl; | |
12211 | bluestore_deferred_op_t *op = _get_deferred_op(txc, o); | |
12212 | op->op = bluestore_deferred_op_t::OP_WRITE; | |
12213 | int r = b->get_blob().map( | |
12214 | b_off, l->length(), | |
12215 | [&](uint64_t offset, uint64_t length) { | |
12216 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
12217 | return 0; | |
12218 | }); | |
11fdf7f2 | 12219 | ceph_assert(r == 0); |
7c673cae | 12220 | op->data = *l; |
81eedcae | 12221 | logger->inc(l_bluestore_write_small_deferred); |
7c673cae FG |
12222 | } else { |
12223 | b->get_blob().map_bl( | |
12224 | b_off, *l, | |
12225 | [&](uint64_t offset, bufferlist& t) { | |
12226 | bdev->aio_write(offset, t, &txc->ioc, false); | |
12227 | }); | |
81eedcae | 12228 | logger->inc(l_bluestore_write_small_new); |
7c673cae FG |
12229 | } |
12230 | } | |
12231 | } | |
11fdf7f2 TL |
12232 | ceph_assert(prealloc_pos == prealloc.end()); |
12233 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
12234 | return 0; |
12235 | } | |
12236 | ||
12237 | void BlueStore::_wctx_finish( | |
12238 | TransContext *txc, | |
12239 | CollectionRef& c, | |
12240 | OnodeRef o, | |
31f18b77 FG |
12241 | WriteContext *wctx, |
12242 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
12243 | { |
12244 | auto oep = wctx->old_extents.begin(); | |
12245 | while (oep != wctx->old_extents.end()) { | |
12246 | auto &lo = *oep; | |
12247 | oep = wctx->old_extents.erase(oep); | |
12248 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
12249 | BlobRef b = lo.e.blob; | |
12250 | const bluestore_blob_t& blob = b->get_blob(); | |
12251 | if (blob.is_compressed()) { | |
12252 | if (lo.blob_empty) { | |
12253 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
12254 | } | |
12255 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
12256 | } | |
12257 | auto& r = lo.r; | |
12258 | txc->statfs_delta.stored() -= lo.e.length; | |
12259 | if (!r.empty()) { | |
12260 | dout(20) << __func__ << " blob release " << r << dendl; | |
12261 | if (blob.is_shared()) { | |
12262 | PExtentVector final; | |
12263 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
12264 | bool unshare = false; |
12265 | bool* unshare_ptr = | |
12266 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 12267 | for (auto e : r) { |
31f18b77 FG |
12268 | b->shared_blob->put_ref( |
12269 | e.offset, e.length, &final, | |
11fdf7f2 TL |
12270 | unshare_ptr); |
12271 | } | |
12272 | if (unshare) { | |
12273 | ceph_assert(maybe_unshared_blobs); | |
12274 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
12275 | } |
12276 | dout(20) << __func__ << " shared_blob release " << final | |
12277 | << " from " << *b->shared_blob << dendl; | |
12278 | txc->write_shared_blob(b->shared_blob); | |
12279 | r.clear(); | |
12280 | r.swap(final); | |
12281 | } | |
12282 | } | |
12283 | // we can't invalidate our logical extents as we drop them because | |
12284 | // other lextents (either in our onode or others) may still | |
12285 | // reference them. but we can throw out anything that is no | |
12286 | // longer allocated. Note that this will leave behind edge bits | |
12287 | // that are no longer referenced but not deallocated (until they | |
12288 | // age out of the cache naturally). | |
12289 | b->discard_unallocated(c.get()); | |
12290 | for (auto e : r) { | |
12291 | dout(20) << __func__ << " release " << e << dendl; | |
12292 | txc->released.insert(e.offset, e.length); | |
12293 | txc->statfs_delta.allocated() -= e.length; | |
12294 | if (blob.is_compressed()) { | |
12295 | txc->statfs_delta.compressed_allocated() -= e.length; | |
12296 | } | |
12297 | } | |
12298 | delete &lo; | |
12299 | if (b->is_spanning() && !b->is_referenced()) { | |
12300 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b | |
12301 | << dendl; | |
12302 | o->extent_map.spanning_blob_map.erase(b->id); | |
12303 | } | |
12304 | } | |
12305 | } | |
12306 | ||
12307 | void BlueStore::_do_write_data( | |
12308 | TransContext *txc, | |
12309 | CollectionRef& c, | |
12310 | OnodeRef o, | |
12311 | uint64_t offset, | |
12312 | uint64_t length, | |
12313 | bufferlist& bl, | |
12314 | WriteContext *wctx) | |
12315 | { | |
12316 | uint64_t end = offset + length; | |
12317 | bufferlist::iterator p = bl.begin(); | |
12318 | ||
12319 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
12320 | (length != min_alloc_size)) { | |
12321 | // we fall within the same block | |
12322 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
12323 | } else { | |
12324 | uint64_t head_offset, head_length; | |
12325 | uint64_t middle_offset, middle_length; | |
12326 | uint64_t tail_offset, tail_length; | |
12327 | ||
12328 | head_offset = offset; | |
11fdf7f2 | 12329 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 12330 | |
11fdf7f2 TL |
12331 | tail_offset = p2align(end, min_alloc_size); |
12332 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
12333 | |
12334 | middle_offset = head_offset + head_length; | |
12335 | middle_length = length - head_length - tail_length; | |
12336 | ||
12337 | if (head_length) { | |
12338 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
12339 | } | |
12340 | ||
12341 | if (middle_length) { | |
12342 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); | |
12343 | } | |
12344 | ||
12345 | if (tail_length) { | |
12346 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
12347 | } | |
12348 | } | |
12349 | } | |
12350 | ||
31f18b77 FG |
12351 | void BlueStore::_choose_write_options( |
12352 | CollectionRef& c, | |
12353 | OnodeRef o, | |
12354 | uint32_t fadvise_flags, | |
12355 | WriteContext *wctx) | |
7c673cae | 12356 | { |
7c673cae FG |
12357 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
12358 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 12359 | wctx->buffered = true; |
7c673cae FG |
12360 | } else if (cct->_conf->bluestore_default_buffered_write && |
12361 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
12362 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
12363 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 12364 | wctx->buffered = true; |
7c673cae FG |
12365 | } |
12366 | ||
31f18b77 FG |
12367 | // apply basic csum block size |
12368 | wctx->csum_order = block_size_order; | |
7c673cae FG |
12369 | |
12370 | // compression parameters | |
12371 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
12372 | auto cm = select_option( | |
12373 | "compression_mode", | |
31f18b77 | 12374 | comp_mode.load(), |
7c673cae FG |
12375 | [&]() { |
12376 | string val; | |
11fdf7f2 | 12377 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
31f18b77 FG |
12378 | return boost::optional<Compressor::CompressionMode>( |
12379 | Compressor::get_comp_mode_type(val)); | |
7c673cae FG |
12380 | } |
12381 | return boost::optional<Compressor::CompressionMode>(); | |
12382 | } | |
12383 | ); | |
31f18b77 FG |
12384 | |
12385 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
12386 | ((cm == Compressor::COMP_FORCE) || |
12387 | (cm == Compressor::COMP_AGGRESSIVE && | |
12388 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
12389 | (cm == Compressor::COMP_PASSIVE && | |
12390 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
12391 | ||
12392 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
12393 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
12394 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
12395 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 12396 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 12397 | |
7c673cae | 12398 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 12399 | |
7c673cae | 12400 | if (o->onode.expected_write_size) { |
224ce89b | 12401 | wctx->csum_order = std::max(min_alloc_size_order, |
31f18b77 | 12402 | (uint8_t)ctz(o->onode.expected_write_size)); |
7c673cae | 12403 | } else { |
224ce89b | 12404 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
12405 | } |
12406 | ||
31f18b77 FG |
12407 | if (wctx->compress) { |
12408 | wctx->target_blob_size = select_option( | |
7c673cae | 12409 | "compression_max_blob_size", |
31f18b77 | 12410 | comp_max_blob_size.load(), |
7c673cae | 12411 | [&]() { |
11fdf7f2 TL |
12412 | int64_t val; |
12413 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
7c673cae FG |
12414 | return boost::optional<uint64_t>((uint64_t)val); |
12415 | } | |
12416 | return boost::optional<uint64_t>(); | |
12417 | } | |
12418 | ); | |
12419 | } | |
12420 | } else { | |
31f18b77 FG |
12421 | if (wctx->compress) { |
12422 | wctx->target_blob_size = select_option( | |
7c673cae | 12423 | "compression_min_blob_size", |
31f18b77 | 12424 | comp_min_blob_size.load(), |
7c673cae | 12425 | [&]() { |
11fdf7f2 TL |
12426 | int64_t val; |
12427 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
7c673cae FG |
12428 | return boost::optional<uint64_t>((uint64_t)val); |
12429 | } | |
12430 | return boost::optional<uint64_t>(); | |
12431 | } | |
12432 | ); | |
12433 | } | |
12434 | } | |
31f18b77 | 12435 | |
7c673cae | 12436 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
12437 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
12438 | wctx->target_blob_size = max_bsize; | |
7c673cae | 12439 | } |
31f18b77 | 12440 | |
7c673cae FG |
12441 | // set the min blob size floor at 2x the min_alloc_size, or else we |
12442 | // won't be able to allocate a smaller extent for the compressed | |
12443 | // data. | |
31f18b77 FG |
12444 | if (wctx->compress && |
12445 | wctx->target_blob_size < min_alloc_size * 2) { | |
12446 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 12447 | } |
31f18b77 FG |
12448 | |
12449 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
12450 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
12451 | << " compress=" << (int)wctx->compress |
12452 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
12453 | << std::dec << dendl; |
12454 | } | |
12455 | ||
12456 | int BlueStore::_do_gc( | |
12457 | TransContext *txc, | |
12458 | CollectionRef& c, | |
12459 | OnodeRef o, | |
12460 | const GarbageCollector& gc, | |
12461 | const WriteContext& wctx, | |
12462 | uint64_t *dirty_start, | |
12463 | uint64_t *dirty_end) | |
12464 | { | |
12465 | auto& extents_to_collect = gc.get_extents_to_collect(); | |
12466 | ||
1adf2230 | 12467 | bool dirty_range_updated = false; |
31f18b77 | 12468 | WriteContext wctx_gc; |
7c673cae | 12469 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 12470 | |
31f18b77 FG |
12471 | for (auto it = extents_to_collect.begin(); |
12472 | it != extents_to_collect.end(); | |
12473 | ++it) { | |
12474 | bufferlist bl; | |
12475 | int r = _do_read(c.get(), o, it->offset, it->length, bl, 0); | |
11fdf7f2 | 12476 | ceph_assert(r == (int)it->length); |
31f18b77 | 12477 | |
31f18b77 FG |
12478 | _do_write_data(txc, c, o, it->offset, it->length, bl, &wctx_gc); |
12479 | logger->inc(l_bluestore_gc_merged, it->length); | |
12480 | ||
12481 | if (*dirty_start > it->offset) { | |
12482 | *dirty_start = it->offset; | |
1adf2230 | 12483 | dirty_range_updated = true; |
31f18b77 FG |
12484 | } |
12485 | ||
12486 | if (*dirty_end < it->offset + it->length) { | |
12487 | *dirty_end = it->offset + it->length; | |
1adf2230 | 12488 | dirty_range_updated = true; |
31f18b77 FG |
12489 | } |
12490 | } | |
1adf2230 AA |
12491 | if (dirty_range_updated) { |
12492 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
12493 | } | |
31f18b77 FG |
12494 | |
12495 | dout(30) << __func__ << " alloc write" << dendl; | |
12496 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
12497 | if (r < 0) { | |
12498 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
12499 | << dendl; | |
12500 | return r; | |
12501 | } | |
12502 | ||
12503 | _wctx_finish(txc, c, o, &wctx_gc); | |
12504 | return 0; | |
12505 | } | |
12506 | ||
12507 | int BlueStore::_do_write( | |
12508 | TransContext *txc, | |
12509 | CollectionRef& c, | |
12510 | OnodeRef o, | |
12511 | uint64_t offset, | |
12512 | uint64_t length, | |
12513 | bufferlist& bl, | |
12514 | uint32_t fadvise_flags) | |
12515 | { | |
12516 | int r = 0; | |
12517 | ||
12518 | dout(20) << __func__ | |
12519 | << " " << o->oid | |
12520 | << " 0x" << std::hex << offset << "~" << length | |
12521 | << " - have 0x" << o->onode.size | |
12522 | << " (" << std::dec << o->onode.size << ")" | |
12523 | << " bytes" | |
12524 | << " fadvise_flags 0x" << std::hex << fadvise_flags << std::dec | |
12525 | << dendl; | |
81eedcae | 12526 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
12527 | |
12528 | if (length == 0) { | |
12529 | return 0; | |
12530 | } | |
12531 | ||
12532 | uint64_t end = offset + length; | |
12533 | ||
12534 | GarbageCollector gc(c->store->cct); | |
12535 | int64_t benefit; | |
12536 | auto dirty_start = offset; | |
12537 | auto dirty_end = end; | |
12538 | ||
12539 | WriteContext wctx; | |
12540 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
12541 | o->extent_map.fault_range(db, offset, length); |
12542 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
12543 | r = _do_alloc_write(txc, c, o, &wctx); |
12544 | if (r < 0) { | |
12545 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
12546 | << dendl; | |
12547 | goto out; | |
12548 | } | |
12549 | ||
31f18b77 FG |
12550 | // NB: _wctx_finish() will empty old_extents |
12551 | // so we must do gc estimation before that | |
7c673cae | 12552 | benefit = gc.estimate(offset, |
31f18b77 FG |
12553 | length, |
12554 | o->extent_map, | |
12555 | wctx.old_extents, | |
12556 | min_alloc_size); | |
7c673cae FG |
12557 | |
12558 | _wctx_finish(txc, c, o, &wctx); | |
12559 | if (end > o->onode.size) { | |
12560 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 12561 | << std::dec << dendl; |
7c673cae FG |
12562 | o->onode.size = end; |
12563 | } | |
12564 | ||
11fdf7f2 | 12565 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
31f18b77 FG |
12566 | if (!gc.get_extents_to_collect().empty()) { |
12567 | dout(20) << __func__ << " perform garbage collection, " | |
12568 | << "expected benefit = " << benefit << " AUs" << dendl; | |
12569 | r = _do_gc(txc, c, o, gc, wctx, &dirty_start, &dirty_end); | |
12570 | if (r < 0) { | |
12571 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
12572 | << dendl; | |
12573 | goto out; | |
7c673cae | 12574 | } |
1adf2230 AA |
12575 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
12576 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae FG |
12577 | } |
12578 | } | |
7c673cae | 12579 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
12580 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
12581 | ||
7c673cae FG |
12582 | r = 0; |
12583 | ||
12584 | out: | |
12585 | return r; | |
12586 | } | |
12587 | ||
12588 | int BlueStore::_write(TransContext *txc, | |
12589 | CollectionRef& c, | |
12590 | OnodeRef& o, | |
31f18b77 FG |
12591 | uint64_t offset, size_t length, |
12592 | bufferlist& bl, | |
12593 | uint32_t fadvise_flags) | |
7c673cae FG |
12594 | { |
12595 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12596 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
12597 | << dendl; | |
35e4c445 FG |
12598 | int r = 0; |
12599 | if (offset + length >= OBJECT_MAX_SIZE) { | |
12600 | r = -E2BIG; | |
12601 | } else { | |
12602 | _assign_nid(txc, o); | |
12603 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
12604 | txc->write_onode(o); | |
12605 | } | |
7c673cae FG |
12606 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
12607 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
12608 | << " = " << r << dendl; | |
12609 | return r; | |
12610 | } | |
12611 | ||
12612 | int BlueStore::_zero(TransContext *txc, | |
12613 | CollectionRef& c, | |
12614 | OnodeRef& o, | |
12615 | uint64_t offset, size_t length) | |
12616 | { | |
12617 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12618 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
12619 | << dendl; | |
35e4c445 FG |
12620 | int r = 0; |
12621 | if (offset + length >= OBJECT_MAX_SIZE) { | |
12622 | r = -E2BIG; | |
12623 | } else { | |
12624 | _assign_nid(txc, o); | |
12625 | r = _do_zero(txc, c, o, offset, length); | |
12626 | } | |
7c673cae FG |
12627 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
12628 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
12629 | << " = " << r << dendl; | |
12630 | return r; | |
12631 | } | |
12632 | ||
12633 | int BlueStore::_do_zero(TransContext *txc, | |
12634 | CollectionRef& c, | |
12635 | OnodeRef& o, | |
12636 | uint64_t offset, size_t length) | |
12637 | { | |
12638 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12639 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
12640 | << dendl; | |
12641 | int r = 0; | |
12642 | ||
81eedcae | 12643 | _dump_onode<30>(cct, *o); |
7c673cae FG |
12644 | |
12645 | WriteContext wctx; | |
12646 | o->extent_map.fault_range(db, offset, length); | |
12647 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 12648 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
12649 | _wctx_finish(txc, c, o, &wctx); |
12650 | ||
b32b8144 | 12651 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
12652 | o->onode.size = offset + length; |
12653 | dout(20) << __func__ << " extending size to " << offset + length | |
12654 | << dendl; | |
12655 | } | |
12656 | txc->write_onode(o); | |
12657 | ||
12658 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
12659 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
12660 | << " = " << r << dendl; | |
12661 | return r; | |
12662 | } | |
12663 | ||
12664 | void BlueStore::_do_truncate( | |
31f18b77 FG |
12665 | TransContext *txc, CollectionRef& c, OnodeRef o, uint64_t offset, |
12666 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
12667 | { |
12668 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12669 | << " 0x" << std::hex << offset << std::dec << dendl; | |
12670 | ||
81eedcae | 12671 | _dump_onode<30>(cct, *o); |
7c673cae FG |
12672 | |
12673 | if (offset == o->onode.size) | |
31f18b77 | 12674 | return; |
7c673cae FG |
12675 | |
12676 | if (offset < o->onode.size) { | |
12677 | WriteContext wctx; | |
12678 | uint64_t length = o->onode.size - offset; | |
12679 | o->extent_map.fault_range(db, offset, length); | |
12680 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 FG |
12681 | o->extent_map.dirty_range(offset, length); |
12682 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); | |
7c673cae FG |
12683 | |
12684 | // if we have shards past EOF, ask for a reshard | |
12685 | if (!o->onode.extent_map_shards.empty() && | |
12686 | o->onode.extent_map_shards.back().offset >= offset) { | |
12687 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
12688 | if (offset) { | |
12689 | o->extent_map.request_reshard(offset - 1, offset + length); | |
12690 | } else { | |
12691 | o->extent_map.request_reshard(0, length); | |
12692 | } | |
12693 | } | |
12694 | } | |
12695 | ||
12696 | o->onode.size = offset; | |
12697 | ||
12698 | txc->write_onode(o); | |
12699 | } | |
12700 | ||
35e4c445 | 12701 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
12702 | CollectionRef& c, |
12703 | OnodeRef& o, | |
12704 | uint64_t offset) | |
12705 | { | |
12706 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12707 | << " 0x" << std::hex << offset << std::dec | |
12708 | << dendl; | |
35e4c445 FG |
12709 | int r = 0; |
12710 | if (offset >= OBJECT_MAX_SIZE) { | |
12711 | r = -E2BIG; | |
12712 | } else { | |
12713 | _do_truncate(txc, c, o, offset); | |
12714 | } | |
12715 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
12716 | << " 0x" << std::hex << offset << std::dec | |
12717 | << " = " << r << dendl; | |
12718 | return r; | |
7c673cae FG |
12719 | } |
12720 | ||
12721 | int BlueStore::_do_remove( | |
12722 | TransContext *txc, | |
12723 | CollectionRef& c, | |
12724 | OnodeRef o) | |
12725 | { | |
31f18b77 | 12726 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
12727 | bool is_gen = !o->oid.is_no_gen(); |
12728 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
12729 | if (o->onode.has_omap()) { |
12730 | o->flush(); | |
11fdf7f2 TL |
12731 | _do_omap_clear(txc, |
12732 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP, | |
12733 | o->onode.nid); | |
7c673cae FG |
12734 | } |
12735 | o->exists = false; | |
12736 | string key; | |
12737 | for (auto &s : o->extent_map.shards) { | |
12738 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
12739 | << s.shard_info->offset << std::dec << dendl; | |
12740 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
12741 | [&](const string& final_key) { | |
12742 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
12743 | } | |
12744 | ); | |
12745 | } | |
12746 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 12747 | txc->note_removed_object(o); |
7c673cae FG |
12748 | o->extent_map.clear(); |
12749 | o->onode = bluestore_onode_t(); | |
12750 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 12751 | |
224ce89b WB |
12752 | if (!is_gen || maybe_unshared_blobs.empty()) { |
12753 | return 0; | |
12754 | } | |
31f18b77 | 12755 | |
224ce89b WB |
12756 | // see if we can unshare blobs still referenced by the head |
12757 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
12758 | << maybe_unshared_blobs << dendl; | |
12759 | ghobject_t nogen = o->oid; | |
12760 | nogen.generation = ghobject_t::NO_GEN; | |
12761 | OnodeRef h = c->onode_map.lookup(nogen); | |
12762 | ||
12763 | if (!h || !h->exists) { | |
12764 | return 0; | |
12765 | } | |
12766 | ||
12767 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
12768 | << " " << h->oid << dendl; | |
12769 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
12770 | for (auto& e : h->extent_map.extent_map) { | |
12771 | const bluestore_blob_t& b = e.blob->get_blob(); | |
12772 | SharedBlob *sb = e.blob->shared_blob.get(); | |
12773 | if (b.is_shared() && | |
12774 | sb->loaded && | |
12775 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
12776 | if (b.is_compressed()) { |
12777 | expect[sb].get(0, b.get_ondisk_length()); | |
12778 | } else { | |
12779 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
12780 | expect[sb].get(off, len); | |
12781 | return 0; | |
12782 | }); | |
12783 | } | |
224ce89b WB |
12784 | } |
12785 | } | |
31f18b77 | 12786 | |
224ce89b WB |
12787 | vector<SharedBlob*> unshared_blobs; |
12788 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
12789 | for (auto& p : expect) { | |
12790 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
12791 | if (p.first->persistent->ref_map == p.second) { | |
12792 | SharedBlob *sb = p.first; | |
12793 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
12794 | unshared_blobs.push_back(sb); | |
12795 | txc->unshare_blob(sb); | |
12796 | uint64_t sbid = c->make_blob_unshared(sb); | |
12797 | string key; | |
12798 | get_shared_blob_key(sbid, &key); | |
12799 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
12800 | } | |
12801 | } | |
12802 | ||
12803 | if (unshared_blobs.empty()) { | |
12804 | return 0; | |
12805 | } | |
12806 | ||
224ce89b WB |
12807 | for (auto& e : h->extent_map.extent_map) { |
12808 | const bluestore_blob_t& b = e.blob->get_blob(); | |
12809 | SharedBlob *sb = e.blob->shared_blob.get(); | |
12810 | if (b.is_shared() && | |
12811 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
12812 | sb) != unshared_blobs.end()) { | |
12813 | dout(20) << __func__ << " unsharing " << e << dendl; | |
12814 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
12815 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 12816 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
12817 | } |
12818 | } | |
224ce89b WB |
12819 | txc->write_onode(h); |
12820 | ||
7c673cae FG |
12821 | return 0; |
12822 | } | |
12823 | ||
12824 | int BlueStore::_remove(TransContext *txc, | |
12825 | CollectionRef& c, | |
12826 | OnodeRef &o) | |
12827 | { | |
11fdf7f2 TL |
12828 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
12829 | << " onode " << o.get() | |
12830 | << " txc "<< txc << dendl; | |
7c673cae FG |
12831 | int r = _do_remove(txc, c, o); |
12832 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
12833 | return r; | |
12834 | } | |
12835 | ||
12836 | int BlueStore::_setattr(TransContext *txc, | |
12837 | CollectionRef& c, | |
12838 | OnodeRef& o, | |
12839 | const string& name, | |
12840 | bufferptr& val) | |
12841 | { | |
12842 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12843 | << " " << name << " (" << val.length() << " bytes)" | |
12844 | << dendl; | |
12845 | int r = 0; | |
3efd9988 FG |
12846 | if (val.is_partial()) { |
12847 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
12848 | val.length()); | |
12849 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
12850 | } else { | |
12851 | auto& b = o->onode.attrs[name.c_str()] = val; | |
12852 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
12853 | } | |
7c673cae FG |
12854 | txc->write_onode(o); |
12855 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
12856 | << " " << name << " (" << val.length() << " bytes)" | |
12857 | << " = " << r << dendl; | |
12858 | return r; | |
12859 | } | |
12860 | ||
12861 | int BlueStore::_setattrs(TransContext *txc, | |
12862 | CollectionRef& c, | |
12863 | OnodeRef& o, | |
12864 | const map<string,bufferptr>& aset) | |
12865 | { | |
12866 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12867 | << " " << aset.size() << " keys" | |
12868 | << dendl; | |
12869 | int r = 0; | |
12870 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
12871 | p != aset.end(); ++p) { | |
3efd9988 FG |
12872 | if (p->second.is_partial()) { |
12873 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 12874 | bufferptr(p->second.c_str(), p->second.length()); |
3efd9988 FG |
12875 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); |
12876 | } else { | |
12877 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
12878 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
12879 | } | |
7c673cae FG |
12880 | } |
12881 | txc->write_onode(o); | |
12882 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
12883 | << " " << aset.size() << " keys" | |
12884 | << " = " << r << dendl; | |
12885 | return r; | |
12886 | } | |
12887 | ||
12888 | ||
12889 | int BlueStore::_rmattr(TransContext *txc, | |
12890 | CollectionRef& c, | |
12891 | OnodeRef& o, | |
12892 | const string& name) | |
12893 | { | |
12894 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
12895 | << " " << name << dendl; | |
12896 | int r = 0; | |
12897 | auto it = o->onode.attrs.find(name.c_str()); | |
12898 | if (it == o->onode.attrs.end()) | |
12899 | goto out; | |
12900 | ||
12901 | o->onode.attrs.erase(it); | |
12902 | txc->write_onode(o); | |
12903 | ||
12904 | out: | |
12905 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
12906 | << " " << name << " = " << r << dendl; | |
12907 | return r; | |
12908 | } | |
12909 | ||
12910 | int BlueStore::_rmattrs(TransContext *txc, | |
12911 | CollectionRef& c, | |
12912 | OnodeRef& o) | |
12913 | { | |
12914 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
12915 | int r = 0; | |
12916 | ||
12917 | if (o->onode.attrs.empty()) | |
12918 | goto out; | |
12919 | ||
12920 | o->onode.attrs.clear(); | |
12921 | txc->write_onode(o); | |
12922 | ||
12923 | out: | |
12924 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
12925 | return r; | |
12926 | } | |
12927 | ||
11fdf7f2 TL |
12928 | void BlueStore::_do_omap_clear(TransContext *txc, const string& omap_prefix, |
12929 | uint64_t id) | |
7c673cae | 12930 | { |
7c673cae FG |
12931 | string prefix, tail; |
12932 | get_omap_header(id, &prefix); | |
12933 | get_omap_tail(id, &tail); | |
11fdf7f2 TL |
12934 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
12935 | dout(20) << __func__ << " remove range start: " | |
12936 | << pretty_binary_string(prefix) << " end: " | |
12937 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
12938 | } |
12939 | ||
12940 | int BlueStore::_omap_clear(TransContext *txc, | |
12941 | CollectionRef& c, | |
12942 | OnodeRef& o) | |
12943 | { | |
12944 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
12945 | int r = 0; | |
12946 | if (o->onode.has_omap()) { | |
12947 | o->flush(); | |
11fdf7f2 TL |
12948 | _do_omap_clear(txc, |
12949 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP, | |
12950 | o->onode.nid); | |
7c673cae FG |
12951 | o->onode.clear_omap_flag(); |
12952 | txc->write_onode(o); | |
12953 | } | |
12954 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
12955 | return r; | |
12956 | } | |
12957 | ||
12958 | int BlueStore::_omap_setkeys(TransContext *txc, | |
12959 | CollectionRef& c, | |
12960 | OnodeRef& o, | |
12961 | bufferlist &bl) | |
12962 | { | |
12963 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
12964 | int r; | |
11fdf7f2 | 12965 | auto p = bl.cbegin(); |
7c673cae FG |
12966 | __u32 num; |
12967 | if (!o->onode.has_omap()) { | |
12968 | o->onode.set_omap_flag(); | |
11fdf7f2 TL |
12969 | if (o->oid.is_pgmeta()) { |
12970 | o->onode.flags |= bluestore_onode_t::FLAG_PGMETA_OMAP; | |
12971 | } | |
7c673cae FG |
12972 | txc->write_onode(o); |
12973 | } else { | |
12974 | txc->note_modified_object(o); | |
12975 | } | |
11fdf7f2 TL |
12976 | const string& prefix = |
12977 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
7c673cae FG |
12978 | string final_key; |
12979 | _key_encode_u64(o->onode.nid, &final_key); | |
12980 | final_key.push_back('.'); | |
11fdf7f2 | 12981 | decode(num, p); |
7c673cae FG |
12982 | while (num--) { |
12983 | string key; | |
12984 | bufferlist value; | |
11fdf7f2 TL |
12985 | decode(key, p); |
12986 | decode(value, p); | |
7c673cae FG |
12987 | final_key.resize(9); // keep prefix |
12988 | final_key += key; | |
11fdf7f2 | 12989 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 12990 | << " <- " << key << dendl; |
11fdf7f2 | 12991 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
12992 | } |
12993 | r = 0; | |
12994 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
12995 | return r; | |
12996 | } | |
12997 | ||
12998 | int BlueStore::_omap_setheader(TransContext *txc, | |
12999 | CollectionRef& c, | |
13000 | OnodeRef &o, | |
13001 | bufferlist& bl) | |
13002 | { | |
13003 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
13004 | int r; | |
13005 | string key; | |
13006 | if (!o->onode.has_omap()) { | |
13007 | o->onode.set_omap_flag(); | |
11fdf7f2 TL |
13008 | if (o->oid.is_pgmeta()) { |
13009 | o->onode.flags |= bluestore_onode_t::FLAG_PGMETA_OMAP; | |
13010 | } | |
7c673cae FG |
13011 | txc->write_onode(o); |
13012 | } else { | |
13013 | txc->note_modified_object(o); | |
13014 | } | |
11fdf7f2 TL |
13015 | const string& prefix = |
13016 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
7c673cae | 13017 | get_omap_header(o->onode.nid, &key); |
11fdf7f2 | 13018 | txc->t->set(prefix, key, bl); |
7c673cae FG |
13019 | r = 0; |
13020 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
13021 | return r; | |
13022 | } | |
13023 | ||
13024 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
13025 | CollectionRef& c, | |
13026 | OnodeRef& o, | |
13027 | bufferlist& bl) | |
13028 | { | |
13029 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
13030 | int r = 0; | |
11fdf7f2 | 13031 | auto p = bl.cbegin(); |
7c673cae FG |
13032 | __u32 num; |
13033 | string final_key; | |
13034 | ||
13035 | if (!o->onode.has_omap()) { | |
13036 | goto out; | |
13037 | } | |
11fdf7f2 TL |
13038 | { |
13039 | const string& prefix = | |
13040 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
13041 | _key_encode_u64(o->onode.nid, &final_key); | |
13042 | final_key.push_back('.'); | |
13043 | decode(num, p); | |
13044 | while (num--) { | |
13045 | string key; | |
13046 | decode(key, p); | |
13047 | final_key.resize(9); // keep prefix | |
13048 | final_key += key; | |
13049 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
13050 | << " <- " << key << dendl; | |
13051 | txc->t->rmkey(prefix, final_key); | |
13052 | } | |
7c673cae FG |
13053 | } |
13054 | txc->note_modified_object(o); | |
13055 | ||
13056 | out: | |
13057 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
13058 | return r; | |
13059 | } | |
13060 | ||
13061 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
13062 | CollectionRef& c, | |
13063 | OnodeRef& o, | |
13064 | const string& first, const string& last) | |
13065 | { | |
13066 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
13067 | string key_first, key_last; |
13068 | int r = 0; | |
13069 | if (!o->onode.has_omap()) { | |
13070 | goto out; | |
13071 | } | |
11fdf7f2 TL |
13072 | { |
13073 | const string& prefix = | |
13074 | o->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
13075 | o->flush(); | |
13076 | get_omap_key(o->onode.nid, first, &key_first); | |
13077 | get_omap_key(o->onode.nid, last, &key_last); | |
13078 | txc->t->rm_range_keys(prefix, key_first, key_last); | |
13079 | dout(20) << __func__ << " remove range start: " | |
13080 | << pretty_binary_string(key_first) << " end: " | |
13081 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
13082 | } |
13083 | txc->note_modified_object(o); | |
13084 | ||
13085 | out: | |
13086 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
13087 | return r; | |
13088 | } | |
13089 | ||
13090 | int BlueStore::_set_alloc_hint( | |
13091 | TransContext *txc, | |
13092 | CollectionRef& c, | |
13093 | OnodeRef& o, | |
13094 | uint64_t expected_object_size, | |
13095 | uint64_t expected_write_size, | |
13096 | uint32_t flags) | |
13097 | { | |
13098 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
13099 | << " object_size " << expected_object_size | |
13100 | << " write_size " << expected_write_size | |
13101 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
13102 | << dendl; | |
13103 | int r = 0; | |
13104 | o->onode.expected_object_size = expected_object_size; | |
13105 | o->onode.expected_write_size = expected_write_size; | |
13106 | o->onode.alloc_hint_flags = flags; | |
13107 | txc->write_onode(o); | |
13108 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
13109 | << " object_size " << expected_object_size | |
13110 | << " write_size " << expected_write_size | |
13111 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
13112 | << " = " << r << dendl; | |
13113 | return r; | |
13114 | } | |
13115 | ||
13116 | int BlueStore::_clone(TransContext *txc, | |
13117 | CollectionRef& c, | |
13118 | OnodeRef& oldo, | |
13119 | OnodeRef& newo) | |
13120 | { | |
13121 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
13122 | << newo->oid << dendl; | |
13123 | int r = 0; | |
13124 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
13125 | derr << __func__ << " mismatched hash on " << oldo->oid | |
13126 | << " and " << newo->oid << dendl; | |
13127 | return -EINVAL; | |
13128 | } | |
13129 | ||
7c673cae FG |
13130 | _assign_nid(txc, newo); |
13131 | ||
13132 | // clone data | |
13133 | oldo->flush(); | |
13134 | _do_truncate(txc, c, newo, 0); | |
13135 | if (cct->_conf->bluestore_clone_cow) { | |
13136 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
13137 | } else { | |
13138 | bufferlist bl; | |
13139 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
13140 | if (r < 0) | |
13141 | goto out; | |
13142 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
13143 | if (r < 0) | |
13144 | goto out; | |
13145 | } | |
13146 | ||
13147 | // clone attrs | |
13148 | newo->onode.attrs = oldo->onode.attrs; | |
13149 | ||
13150 | // clone omap | |
13151 | if (newo->onode.has_omap()) { | |
13152 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
13153 | newo->flush(); | |
11fdf7f2 TL |
13154 | _do_omap_clear(txc, |
13155 | newo->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP | |
13156 | : PREFIX_OMAP, | |
13157 | newo->onode.nid); | |
7c673cae FG |
13158 | } |
13159 | if (oldo->onode.has_omap()) { | |
13160 | dout(20) << __func__ << " copying omap data" << dendl; | |
13161 | if (!newo->onode.has_omap()) { | |
13162 | newo->onode.set_omap_flag(); | |
11fdf7f2 TL |
13163 | if (newo->oid.is_pgmeta()) { |
13164 | newo->onode.flags |= bluestore_onode_t::FLAG_PGMETA_OMAP; | |
13165 | } | |
7c673cae | 13166 | } |
11fdf7f2 TL |
13167 | const string& prefix = |
13168 | newo->onode.is_pgmeta_omap() ? PREFIX_PGMETA_OMAP : PREFIX_OMAP; | |
13169 | KeyValueDB::Iterator it = db->get_iterator(prefix); | |
7c673cae FG |
13170 | string head, tail; |
13171 | get_omap_header(oldo->onode.nid, &head); | |
13172 | get_omap_tail(oldo->onode.nid, &tail); | |
13173 | it->lower_bound(head); | |
13174 | while (it->valid()) { | |
13175 | if (it->key() >= tail) { | |
13176 | dout(30) << __func__ << " reached tail" << dendl; | |
13177 | break; | |
13178 | } else { | |
13179 | dout(30) << __func__ << " got header/data " | |
13180 | << pretty_binary_string(it->key()) << dendl; | |
13181 | string key; | |
13182 | rewrite_omap_key(newo->onode.nid, it->key(), &key); | |
11fdf7f2 | 13183 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
13184 | } |
13185 | it->next(); | |
13186 | } | |
13187 | } else { | |
13188 | newo->onode.clear_omap_flag(); | |
13189 | } | |
13190 | ||
13191 | txc->write_onode(newo); | |
13192 | r = 0; | |
13193 | ||
13194 | out: | |
13195 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
13196 | << newo->oid << " = " << r << dendl; | |
13197 | return r; | |
13198 | } | |
13199 | ||
13200 | int BlueStore::_do_clone_range( | |
13201 | TransContext *txc, | |
13202 | CollectionRef& c, | |
13203 | OnodeRef& oldo, | |
13204 | OnodeRef& newo, | |
224ce89b WB |
13205 | uint64_t srcoff, |
13206 | uint64_t length, | |
13207 | uint64_t dstoff) | |
7c673cae FG |
13208 | { |
13209 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
13210 | << newo->oid | |
13211 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
13212 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
13213 | oldo->extent_map.fault_range(db, srcoff, length); | |
13214 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
13215 | _dump_onode<30>(cct, *oldo); |
13216 | _dump_onode<30>(cct, *newo); | |
7c673cae | 13217 | |
11fdf7f2 | 13218 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
81eedcae TL |
13219 | _dump_onode<30>(cct, *oldo); |
13220 | _dump_onode<30>(cct, *newo); | |
7c673cae FG |
13221 | return 0; |
13222 | } | |
13223 | ||
13224 | int BlueStore::_clone_range(TransContext *txc, | |
13225 | CollectionRef& c, | |
13226 | OnodeRef& oldo, | |
13227 | OnodeRef& newo, | |
13228 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
13229 | { | |
13230 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
13231 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
13232 | << " to offset 0x" << dstoff << std::dec << dendl; | |
13233 | int r = 0; | |
13234 | ||
35e4c445 FG |
13235 | if (srcoff + length >= OBJECT_MAX_SIZE || |
13236 | dstoff + length >= OBJECT_MAX_SIZE) { | |
13237 | r = -E2BIG; | |
13238 | goto out; | |
13239 | } | |
7c673cae FG |
13240 | if (srcoff + length > oldo->onode.size) { |
13241 | r = -EINVAL; | |
13242 | goto out; | |
13243 | } | |
13244 | ||
7c673cae FG |
13245 | _assign_nid(txc, newo); |
13246 | ||
13247 | if (length > 0) { | |
13248 | if (cct->_conf->bluestore_clone_cow) { | |
13249 | _do_zero(txc, c, newo, dstoff, length); | |
13250 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
13251 | } else { | |
13252 | bufferlist bl; | |
13253 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
13254 | if (r < 0) | |
13255 | goto out; | |
13256 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
13257 | if (r < 0) | |
13258 | goto out; | |
13259 | } | |
13260 | } | |
13261 | ||
13262 | txc->write_onode(newo); | |
13263 | r = 0; | |
13264 | ||
13265 | out: | |
13266 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
13267 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
13268 | << " to offset 0x" << dstoff << std::dec | |
13269 | << " = " << r << dendl; | |
13270 | return r; | |
13271 | } | |
13272 | ||
13273 | int BlueStore::_rename(TransContext *txc, | |
13274 | CollectionRef& c, | |
13275 | OnodeRef& oldo, | |
13276 | OnodeRef& newo, | |
13277 | const ghobject_t& new_oid) | |
13278 | { | |
13279 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
13280 | << new_oid << dendl; | |
13281 | int r; | |
13282 | ghobject_t old_oid = oldo->oid; | |
31f18b77 | 13283 | mempool::bluestore_cache_other::string new_okey; |
7c673cae FG |
13284 | |
13285 | if (newo) { | |
13286 | if (newo->exists) { | |
13287 | r = -EEXIST; | |
13288 | goto out; | |
13289 | } | |
11fdf7f2 | 13290 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
13291 | } |
13292 | ||
13293 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
13294 | ||
13295 | // rewrite shards | |
13296 | { | |
13297 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
13298 | get_object_key(cct, new_oid, &new_okey); | |
13299 | string key; | |
13300 | for (auto &s : oldo->extent_map.shards) { | |
13301 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
13302 | [&](const string& final_key) { | |
13303 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
13304 | } | |
13305 | ); | |
13306 | s.dirty = true; | |
13307 | } | |
13308 | } | |
13309 | ||
13310 | newo = oldo; | |
13311 | txc->write_onode(newo); | |
13312 | ||
13313 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
13314 | // Onode in the old slot | |
13315 | c->onode_map.rename(oldo, old_oid, new_oid, new_okey); | |
13316 | r = 0; | |
13317 | ||
f64942e4 AA |
13318 | // hold a ref to new Onode in old name position, to ensure we don't drop |
13319 | // it from the cache before this txc commits (or else someone may come along | |
13320 | // and read newo's metadata via the old name). | |
13321 | txc->note_modified_object(oldo); | |
13322 | ||
7c673cae FG |
13323 | out: |
13324 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
13325 | << new_oid << " = " << r << dendl; | |
13326 | return r; | |
13327 | } | |
13328 | ||
13329 | // collections | |
13330 | ||
13331 | int BlueStore::_create_collection( | |
13332 | TransContext *txc, | |
13333 | const coll_t &cid, | |
13334 | unsigned bits, | |
13335 | CollectionRef *c) | |
13336 | { | |
13337 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
13338 | int r; | |
13339 | bufferlist bl; | |
13340 | ||
13341 | { | |
13342 | RWLock::WLocker l(coll_lock); | |
13343 | if (*c) { | |
13344 | r = -EEXIST; | |
13345 | goto out; | |
13346 | } | |
11fdf7f2 TL |
13347 | auto p = new_coll_map.find(cid); |
13348 | ceph_assert(p != new_coll_map.end()); | |
13349 | *c = p->second; | |
7c673cae FG |
13350 | (*c)->cnode.bits = bits; |
13351 | coll_map[cid] = *c; | |
11fdf7f2 | 13352 | new_coll_map.erase(p); |
7c673cae | 13353 | } |
11fdf7f2 | 13354 | encode((*c)->cnode, bl); |
7c673cae FG |
13355 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
13356 | r = 0; | |
13357 | ||
13358 | out: | |
13359 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
13360 | return r; | |
13361 | } | |
13362 | ||
13363 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
13364 | CollectionRef *c) | |
13365 | { | |
13366 | dout(15) << __func__ << " " << cid << dendl; | |
13367 | int r; | |
13368 | ||
11fdf7f2 | 13369 | (*c)->flush_all_but_last(); |
7c673cae FG |
13370 | { |
13371 | RWLock::WLocker l(coll_lock); | |
13372 | if (!*c) { | |
13373 | r = -ENOENT; | |
13374 | goto out; | |
13375 | } | |
13376 | size_t nonexistent_count = 0; | |
11fdf7f2 | 13377 | ceph_assert((*c)->exists); |
7c673cae FG |
13378 | if ((*c)->onode_map.map_any([&](OnodeRef o) { |
13379 | if (o->exists) { | |
13380 | dout(10) << __func__ << " " << o->oid << " " << o | |
13381 | << " exists in onode_map" << dendl; | |
13382 | return true; | |
13383 | } | |
13384 | ++nonexistent_count; | |
13385 | return false; | |
13386 | })) { | |
13387 | r = -ENOTEMPTY; | |
13388 | goto out; | |
13389 | } | |
13390 | ||
13391 | vector<ghobject_t> ls; | |
13392 | ghobject_t next; | |
13393 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
13394 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 13395 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae FG |
13396 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
13397 | nonexistent_count + 1, &ls, &next); | |
13398 | if (r >= 0) { | |
11fdf7f2 TL |
13399 | // If true mean collecton has more objects than nonexistent_count, |
13400 | // so bypass check. | |
13401 | bool exists = (!next.is_max()); | |
7c673cae FG |
13402 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
13403 | dout(10) << __func__ << " oid " << *it << dendl; | |
13404 | auto onode = (*c)->onode_map.lookup(*it); | |
13405 | exists = !onode || onode->exists; | |
13406 | if (exists) { | |
13407 | dout(10) << __func__ << " " << *it | |
11fdf7f2 TL |
13408 | << " exists in db, " |
13409 | << (!onode ? "not present in ram" : "present in ram") | |
13410 | << dendl; | |
7c673cae FG |
13411 | } |
13412 | } | |
13413 | if (!exists) { | |
11fdf7f2 | 13414 | _do_remove_collection(txc, c); |
7c673cae FG |
13415 | r = 0; |
13416 | } else { | |
13417 | dout(10) << __func__ << " " << cid | |
13418 | << " is non-empty" << dendl; | |
13419 | r = -ENOTEMPTY; | |
13420 | } | |
13421 | } | |
13422 | } | |
13423 | ||
13424 | out: | |
13425 | dout(10) << __func__ << " " << cid << " = " << r << dendl; | |
13426 | return r; | |
13427 | } | |
13428 | ||
11fdf7f2 TL |
13429 | void BlueStore::_do_remove_collection(TransContext *txc, |
13430 | CollectionRef *c) | |
13431 | { | |
13432 | coll_map.erase((*c)->cid); | |
13433 | txc->removed_collections.push_back(*c); | |
13434 | (*c)->exists = false; | |
13435 | _osr_register_zombie((*c)->osr.get()); | |
13436 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
13437 | c->reset(); | |
13438 | } | |
13439 | ||
7c673cae FG |
13440 | int BlueStore::_split_collection(TransContext *txc, |
13441 | CollectionRef& c, | |
13442 | CollectionRef& d, | |
13443 | unsigned bits, int rem) | |
13444 | { | |
13445 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
13446 | << " bits " << bits << dendl; | |
13447 | RWLock::WLocker l(c->lock); | |
13448 | RWLock::WLocker l2(d->lock); | |
13449 | int r; | |
13450 | ||
13451 | // flush all previous deferred writes on this sequencer. this is a bit | |
13452 | // heavyweight, but we need to make sure all deferred writes complete | |
13453 | // before we split as the new collection's sequencer may need to order | |
13454 | // this after those writes, and we don't bother with the complexity of | |
13455 | // moving those TransContexts over to the new osr. | |
13456 | _osr_drain_preceding(txc); | |
13457 | ||
13458 | // move any cached items (onodes and referenced shared blobs) that will | |
13459 | // belong to the child collection post-split. leave everything else behind. | |
13460 | // this may include things that don't strictly belong to the now-smaller | |
13461 | // parent split, but the OSD will always send us a split for every new | |
13462 | // child. | |
13463 | ||
13464 | spg_t pgid, dest_pgid; | |
13465 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 13466 | ceph_assert(is_pg); |
7c673cae | 13467 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 13468 | ceph_assert(is_pg); |
7c673cae FG |
13469 | |
13470 | // the destination should initially be empty. | |
11fdf7f2 TL |
13471 | ceph_assert(d->onode_map.empty()); |
13472 | ceph_assert(d->shared_blob_set.empty()); | |
13473 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
13474 | |
13475 | c->split_cache(d.get()); | |
13476 | ||
13477 | // adjust bits. note that this will be redundant for all but the first | |
13478 | // split call for this parent (first child). | |
13479 | c->cnode.bits = bits; | |
11fdf7f2 | 13480 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
13481 | r = 0; |
13482 | ||
13483 | bufferlist bl; | |
11fdf7f2 | 13484 | encode(c->cnode, bl); |
7c673cae FG |
13485 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
13486 | ||
13487 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
13488 | << " bits " << bits << " = " << r << dendl; | |
13489 | return r; | |
13490 | } | |
13491 | ||
11fdf7f2 TL |
13492 | int BlueStore::_merge_collection( |
13493 | TransContext *txc, | |
13494 | CollectionRef *c, | |
13495 | CollectionRef& d, | |
13496 | unsigned bits) | |
13497 | { | |
13498 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
13499 | << " bits " << bits << dendl; | |
13500 | RWLock::WLocker l((*c)->lock); | |
13501 | RWLock::WLocker l2(d->lock); | |
13502 | int r; | |
13503 | ||
13504 | coll_t cid = (*c)->cid; | |
13505 | ||
13506 | // flush all previous deferred writes on the source collection to ensure | |
13507 | // that all deferred writes complete before we merge as the target collection's | |
13508 | // sequencer may need to order new ops after those writes. | |
13509 | ||
13510 | _osr_drain((*c)->osr.get()); | |
13511 | ||
13512 | // move any cached items (onodes and referenced shared blobs) that will | |
13513 | // belong to the child collection post-split. leave everything else behind. | |
13514 | // this may include things that don't strictly belong to the now-smaller | |
13515 | // parent split, but the OSD will always send us a split for every new | |
13516 | // child. | |
13517 | ||
13518 | spg_t pgid, dest_pgid; | |
13519 | bool is_pg = cid.is_pg(&pgid); | |
13520 | ceph_assert(is_pg); | |
13521 | is_pg = d->cid.is_pg(&dest_pgid); | |
13522 | ceph_assert(is_pg); | |
13523 | ||
13524 | // adjust bits. note that this will be redundant for all but the first | |
13525 | // merge call for the parent/target. | |
13526 | d->cnode.bits = bits; | |
13527 | ||
13528 | // behavior depends on target (d) bits, so this after that is updated. | |
13529 | (*c)->split_cache(d.get()); | |
13530 | ||
13531 | // remove source collection | |
13532 | { | |
13533 | RWLock::WLocker l3(coll_lock); | |
13534 | _do_remove_collection(txc, c); | |
13535 | } | |
13536 | ||
13537 | r = 0; | |
13538 | ||
13539 | bufferlist bl; | |
13540 | encode(d->cnode, bl); | |
13541 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
13542 | ||
13543 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
13544 | << " bits " << bits << " = " << r << dendl; | |
13545 | return r; | |
13546 | } | |
13547 | ||
13548 | void BlueStore::log_latency_fn( | |
13549 | int idx, | |
13550 | const ceph::timespan& l, | |
13551 | std::function<string (const ceph::timespan& lat)> fn) | |
13552 | { | |
13553 | LOG_LATENCY_FN(logger, cct, idx, l, fn); | |
13554 | } | |
13555 | ||
13556 | ||
7c673cae FG |
13557 | // DB key value Histogram |
13558 | #define KEY_SLAB 32 | |
13559 | #define VALUE_SLAB 64 | |
13560 | ||
13561 | const string prefix_onode = "o"; | |
13562 | const string prefix_onode_shard = "x"; | |
13563 | const string prefix_other = "Z"; | |
13564 | ||
13565 | int BlueStore::DBHistogram::get_key_slab(size_t sz) | |
13566 | { | |
13567 | return (sz/KEY_SLAB); | |
13568 | } | |
13569 | ||
13570 | string BlueStore::DBHistogram::get_key_slab_to_range(int slab) | |
13571 | { | |
13572 | int lower_bound = slab * KEY_SLAB; | |
13573 | int upper_bound = (slab + 1) * KEY_SLAB; | |
13574 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
13575 | return ret; | |
13576 | } | |
13577 | ||
13578 | int BlueStore::DBHistogram::get_value_slab(size_t sz) | |
13579 | { | |
13580 | return (sz/VALUE_SLAB); | |
13581 | } | |
13582 | ||
13583 | string BlueStore::DBHistogram::get_value_slab_to_range(int slab) | |
13584 | { | |
13585 | int lower_bound = slab * VALUE_SLAB; | |
13586 | int upper_bound = (slab + 1) * VALUE_SLAB; | |
13587 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
13588 | return ret; | |
13589 | } | |
13590 | ||
13591 | void BlueStore::DBHistogram::update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
13592 | const string &prefix, size_t key_size, size_t value_size) | |
13593 | { | |
13594 | uint32_t key_slab = get_key_slab(key_size); | |
13595 | uint32_t value_slab = get_value_slab(value_size); | |
13596 | key_hist[prefix][key_slab].count++; | |
11fdf7f2 TL |
13597 | key_hist[prefix][key_slab].max_len = |
13598 | std::max<size_t>(key_size, key_hist[prefix][key_slab].max_len); | |
7c673cae FG |
13599 | key_hist[prefix][key_slab].val_map[value_slab].count++; |
13600 | key_hist[prefix][key_slab].val_map[value_slab].max_len = | |
11fdf7f2 TL |
13601 | std::max<size_t>(value_size, |
13602 | key_hist[prefix][key_slab].val_map[value_slab].max_len); | |
7c673cae FG |
13603 | } |
13604 | ||
13605 | void BlueStore::DBHistogram::dump(Formatter *f) | |
13606 | { | |
13607 | f->open_object_section("rocksdb_value_distribution"); | |
13608 | for (auto i : value_hist) { | |
13609 | f->dump_unsigned(get_value_slab_to_range(i.first).data(), i.second); | |
13610 | } | |
13611 | f->close_section(); | |
13612 | ||
13613 | f->open_object_section("rocksdb_key_value_histogram"); | |
13614 | for (auto i : key_hist) { | |
13615 | f->dump_string("prefix", i.first); | |
13616 | f->open_object_section("key_hist"); | |
13617 | for ( auto k : i.second) { | |
13618 | f->dump_unsigned(get_key_slab_to_range(k.first).data(), k.second.count); | |
13619 | f->dump_unsigned("max_len", k.second.max_len); | |
13620 | f->open_object_section("value_hist"); | |
13621 | for ( auto j : k.second.val_map) { | |
13622 | f->dump_unsigned(get_value_slab_to_range(j.first).data(), j.second.count); | |
13623 | f->dump_unsigned("max_len", j.second.max_len); | |
13624 | } | |
13625 | f->close_section(); | |
13626 | } | |
13627 | f->close_section(); | |
13628 | } | |
13629 | f->close_section(); | |
13630 | } | |
13631 | ||
13632 | //Itrerates through the db and collects the stats | |
13633 | void BlueStore::generate_db_histogram(Formatter *f) | |
13634 | { | |
13635 | //globals | |
13636 | uint64_t num_onodes = 0; | |
13637 | uint64_t num_shards = 0; | |
13638 | uint64_t num_super = 0; | |
13639 | uint64_t num_coll = 0; | |
13640 | uint64_t num_omap = 0; | |
11fdf7f2 | 13641 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
13642 | uint64_t num_deferred = 0; |
13643 | uint64_t num_alloc = 0; | |
13644 | uint64_t num_stat = 0; | |
13645 | uint64_t num_others = 0; | |
13646 | uint64_t num_shared_shards = 0; | |
13647 | size_t max_key_size =0, max_value_size = 0; | |
13648 | uint64_t total_key_size = 0, total_value_size = 0; | |
13649 | size_t key_size = 0, value_size = 0; | |
13650 | DBHistogram hist; | |
13651 | ||
11fdf7f2 | 13652 | auto start = coarse_mono_clock::now(); |
7c673cae | 13653 | |
11fdf7f2 | 13654 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
13655 | iter->seek_to_first(); |
13656 | while (iter->valid()) { | |
13657 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
13658 | key_size = iter->key_size(); | |
13659 | value_size = iter->value_size(); | |
13660 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
13661 | max_key_size = std::max(max_key_size, key_size); |
13662 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
13663 | total_key_size += key_size; |
13664 | total_value_size += value_size; | |
13665 | ||
13666 | pair<string,string> key(iter->raw_key()); | |
13667 | ||
13668 | if (key.first == PREFIX_SUPER) { | |
13669 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
13670 | num_super++; | |
13671 | } else if (key.first == PREFIX_STAT) { | |
13672 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
13673 | num_stat++; | |
13674 | } else if (key.first == PREFIX_COLL) { | |
13675 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
13676 | num_coll++; | |
13677 | } else if (key.first == PREFIX_OBJ) { | |
13678 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
13679 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
13680 | num_onodes++; | |
13681 | } else { | |
13682 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
13683 | num_shards++; | |
13684 | } | |
13685 | } else if (key.first == PREFIX_OMAP) { | |
13686 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
13687 | num_omap++; | |
11fdf7f2 TL |
13688 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
13689 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
13690 | num_pgmeta_omap++; | |
7c673cae FG |
13691 | } else if (key.first == PREFIX_DEFERRED) { |
13692 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
13693 | num_deferred++; | |
11fdf7f2 | 13694 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
13695 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
13696 | num_alloc++; | |
13697 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
13698 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
13699 | num_shared_shards++; | |
13700 | } else { | |
13701 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
13702 | num_others++; | |
13703 | } | |
13704 | iter->next(); | |
13705 | } | |
13706 | ||
11fdf7f2 | 13707 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
13708 | f->open_object_section("rocksdb_key_value_stats"); |
13709 | f->dump_unsigned("num_onodes", num_onodes); | |
13710 | f->dump_unsigned("num_shards", num_shards); | |
13711 | f->dump_unsigned("num_super", num_super); | |
13712 | f->dump_unsigned("num_coll", num_coll); | |
13713 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 13714 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
13715 | f->dump_unsigned("num_deferred", num_deferred); |
13716 | f->dump_unsigned("num_alloc", num_alloc); | |
13717 | f->dump_unsigned("num_stat", num_stat); | |
13718 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
13719 | f->dump_unsigned("num_others", num_others); | |
13720 | f->dump_unsigned("max_key_size", max_key_size); | |
13721 | f->dump_unsigned("max_value_size", max_value_size); | |
13722 | f->dump_unsigned("total_key_size", total_key_size); | |
13723 | f->dump_unsigned("total_value_size", total_value_size); | |
13724 | f->close_section(); | |
13725 | ||
13726 | hist.dump(f); | |
13727 | ||
13728 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
13729 | ||
13730 | } | |
13731 | ||
31f18b77 | 13732 | void BlueStore::_flush_cache() |
7c673cae FG |
13733 | { |
13734 | dout(10) << __func__ << dendl; | |
13735 | for (auto i : cache_shards) { | |
13736 | i->trim_all(); | |
11fdf7f2 | 13737 | ceph_assert(i->empty()); |
7c673cae FG |
13738 | } |
13739 | for (auto& p : coll_map) { | |
3efd9988 | 13740 | if (!p.second->onode_map.empty()) { |
11fdf7f2 TL |
13741 | derr << __func__ << " stray onodes on " << p.first << dendl; |
13742 | p.second->onode_map.dump<0>(cct); | |
3efd9988 FG |
13743 | } |
13744 | if (!p.second->shared_blob_set.empty()) { | |
13745 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 13746 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 13747 | } |
11fdf7f2 TL |
13748 | ceph_assert(p.second->onode_map.empty()); |
13749 | ceph_assert(p.second->shared_blob_set.empty()); | |
7c673cae FG |
13750 | } |
13751 | coll_map.clear(); | |
13752 | } | |
13753 | ||
31f18b77 FG |
13754 | // For external caller. |
13755 | // We use a best-effort policy instead, e.g., | |
13756 | // we don't care if there are still some pinned onodes/data in the cache | |
13757 | // after this command is completed. | |
11fdf7f2 | 13758 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
13759 | { |
13760 | dout(10) << __func__ << dendl; | |
13761 | for (auto i : cache_shards) { | |
13762 | i->trim_all(); | |
13763 | } | |
11fdf7f2 TL |
13764 | |
13765 | return 0; | |
31f18b77 FG |
13766 | } |
13767 | ||
7c673cae FG |
13768 | void BlueStore::_apply_padding(uint64_t head_pad, |
13769 | uint64_t tail_pad, | |
7c673cae FG |
13770 | bufferlist& padded) |
13771 | { | |
7c673cae | 13772 | if (head_pad) { |
224ce89b | 13773 | padded.prepend_zero(head_pad); |
7c673cae FG |
13774 | } |
13775 | if (tail_pad) { | |
13776 | padded.append_zero(tail_pad); | |
13777 | } | |
13778 | if (head_pad || tail_pad) { | |
13779 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
13780 | << " tail 0x" << tail_pad << std::dec << dendl; | |
13781 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
13782 | } | |
13783 | } | |
13784 | ||
11fdf7f2 TL |
13785 | void BlueStore::_record_onode(OnodeRef &o, KeyValueDB::Transaction &txn) |
13786 | { | |
13787 | // finalize extent_map shards | |
13788 | o->extent_map.update(txn, false); | |
13789 | if (o->extent_map.needs_reshard()) { | |
13790 | o->extent_map.reshard(db, txn); | |
13791 | o->extent_map.update(txn, true); | |
13792 | if (o->extent_map.needs_reshard()) { | |
13793 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
13794 | << dendl; | |
13795 | o->extent_map.clear_needs_reshard(); | |
13796 | } | |
13797 | logger->inc(l_bluestore_onode_reshard); | |
13798 | } | |
13799 | ||
13800 | // bound encode | |
13801 | size_t bound = 0; | |
13802 | denc(o->onode, bound); | |
13803 | o->extent_map.bound_encode_spanning_blobs(bound); | |
13804 | if (o->onode.extent_map_shards.empty()) { | |
13805 | denc(o->extent_map.inline_bl, bound); | |
13806 | } | |
13807 | ||
13808 | // encode | |
13809 | bufferlist bl; | |
13810 | unsigned onode_part, blob_part, extent_part; | |
13811 | { | |
13812 | auto p = bl.get_contiguous_appender(bound, true); | |
13813 | denc(o->onode, p); | |
13814 | onode_part = p.get_logical_offset(); | |
13815 | o->extent_map.encode_spanning_blobs(p); | |
13816 | blob_part = p.get_logical_offset() - onode_part; | |
13817 | if (o->onode.extent_map_shards.empty()) { | |
13818 | denc(o->extent_map.inline_bl, p); | |
13819 | } | |
13820 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
13821 | } | |
13822 | ||
13823 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
13824 | << " (" << onode_part << " bytes onode + " | |
13825 | << blob_part << " bytes spanning blobs + " | |
13826 | << extent_part << " bytes inline extents)" | |
13827 | << dendl; | |
13828 | ||
13829 | ||
13830 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
13831 | } | |
13832 | ||
13833 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
13834 | { | |
13835 | std::lock_guard l(qlock); | |
13836 | ||
81eedcae TL |
13837 | if (!disk_size_mismatch_alert.empty()) { |
13838 | alerts.emplace( | |
13839 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
13840 | disk_size_mismatch_alert); | |
13841 | } | |
13842 | if (!legacy_statfs_alert.empty()) { | |
13843 | alerts.emplace( | |
13844 | "BLUESTORE_LEGACY_STATFS", | |
13845 | legacy_statfs_alert); | |
13846 | } | |
11fdf7f2 TL |
13847 | if (!spillover_alert.empty() && |
13848 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
13849 | alerts.emplace( | |
13850 | "BLUEFS_SPILLOVER", | |
13851 | spillover_alert); | |
13852 | } | |
13853 | string s0(failed_cmode); | |
13854 | ||
13855 | if (!failed_compressors.empty()) { | |
13856 | if (!s0.empty()) { | |
13857 | s0 += ", "; | |
13858 | } | |
13859 | s0 += "unable to load:"; | |
13860 | bool first = true; | |
13861 | for (auto& s : failed_compressors) { | |
13862 | if (first) { | |
13863 | first = false; | |
13864 | } else { | |
13865 | s0 += ", "; | |
13866 | } | |
13867 | s0 += s; | |
13868 | } | |
13869 | alerts.emplace( | |
13870 | "BLUESTORE_NO_COMPRESSION", | |
13871 | s0); | |
13872 | } | |
13873 | } | |
13874 | ||
7c673cae | 13875 | // =========================================== |
11fdf7f2 TL |
13876 | // BlueStoreRepairer |
13877 | ||
13878 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
13879 | const interval_set<uint64_t>& extents) | |
13880 | { | |
13881 | ceph_assert(granularity); // initialized | |
13882 | // can't call for the second time | |
13883 | ceph_assert(!was_filtered_out); | |
13884 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
13885 | ||
13886 | uint64_t prev_pos = 0; | |
13887 | uint64_t npos = collections_bfs.size(); | |
13888 | ||
13889 | bloom_vector collections_reduced; | |
13890 | bloom_vector objects_reduced; | |
13891 | ||
13892 | for (auto e : extents) { | |
13893 | if (e.second == 0) { | |
13894 | continue; | |
13895 | } | |
13896 | uint64_t pos = max(e.first / granularity, prev_pos); | |
13897 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
13898 | while (pos != npos && pos < end_pos) { | |
13899 | ceph_assert( collections_bfs[pos].element_count() == | |
13900 | objects_bfs[pos].element_count()); | |
13901 | if (collections_bfs[pos].element_count()) { | |
13902 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
13903 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
13904 | } | |
13905 | ++pos; | |
13906 | } | |
13907 | prev_pos = end_pos; | |
13908 | } | |
13909 | collections_reduced.swap(collections_bfs); | |
13910 | objects_reduced.swap(objects_bfs); | |
13911 | was_filtered_out = true; | |
13912 | return collections_bfs.size(); | |
13913 | } | |
13914 | ||
13915 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
13916 | const string& prefix, | |
13917 | const string& key) | |
13918 | { | |
13919 | if (!remove_key_txn) { | |
13920 | remove_key_txn = db->get_transaction(); | |
13921 | } | |
13922 | ++to_repair_cnt; | |
13923 | remove_key_txn->rmkey(prefix, key); | |
13924 | ||
13925 | return true; | |
13926 | } | |
13927 | ||
13928 | bool BlueStoreRepairer::fix_shared_blob( | |
13929 | KeyValueDB *db, | |
13930 | uint64_t sbid, | |
13931 | const bufferlist* bl) | |
13932 | { | |
13933 | KeyValueDB::Transaction txn; | |
13934 | if (fix_misreferences_txn) { // reuse this txn | |
13935 | txn = fix_misreferences_txn; | |
13936 | } else { | |
13937 | if (!fix_shared_blob_txn) { | |
13938 | fix_shared_blob_txn = db->get_transaction(); | |
13939 | } | |
13940 | txn = fix_shared_blob_txn; | |
13941 | } | |
13942 | string key; | |
13943 | get_shared_blob_key(sbid, &key); | |
13944 | ||
13945 | ++to_repair_cnt; | |
13946 | if (bl) { | |
13947 | txn->set(PREFIX_SHARED_BLOB, key, *bl); | |
13948 | } else { | |
13949 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
13950 | } | |
13951 | return true; | |
13952 | } | |
13953 | ||
13954 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
13955 | const string& key, | |
13956 | const store_statfs_t& new_statfs) | |
13957 | { | |
13958 | if (!fix_statfs_txn) { | |
13959 | fix_statfs_txn = db->get_transaction(); | |
13960 | } | |
13961 | BlueStore::volatile_statfs vstatfs; | |
13962 | vstatfs = new_statfs; | |
13963 | bufferlist bl; | |
13964 | vstatfs.encode(bl); | |
13965 | ++to_repair_cnt; | |
13966 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
13967 | return true; | |
13968 | } | |
13969 | ||
13970 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
13971 | FreelistManager* fm, | |
13972 | uint64_t offset, uint64_t len) | |
13973 | { | |
13974 | if (!fix_fm_leaked_txn) { | |
13975 | fix_fm_leaked_txn = db->get_transaction(); | |
13976 | } | |
13977 | ++to_repair_cnt; | |
13978 | fm->release(offset, len, fix_fm_leaked_txn); | |
13979 | return true; | |
13980 | } | |
13981 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
13982 | FreelistManager* fm, | |
13983 | uint64_t offset, uint64_t len) | |
13984 | { | |
13985 | if (!fix_fm_false_free_txn) { | |
13986 | fix_fm_false_free_txn = db->get_transaction(); | |
13987 | } | |
13988 | ++to_repair_cnt; | |
13989 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
13990 | return true; | |
13991 | } | |
13992 | ||
13993 | bool BlueStoreRepairer::fix_bluefs_extents(std::atomic<uint64_t>& out_of_sync_flag) | |
13994 | { | |
13995 | // this is just a stub to count num of repairs properly, | |
13996 | // actual repair happens in BlueStore::_close_db_and_around() | |
13997 | // while doing _sync_bluefs_and_fm | |
13998 | ++out_of_sync_flag; | |
13999 | ++to_repair_cnt; | |
14000 | return true; | |
14001 | } | |
14002 | ||
14003 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) | |
14004 | { | |
14005 | if (misreferenced_extents.size()) { | |
14006 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
14007 | ceph_assert(n > 0); | |
14008 | if (!fix_misreferences_txn) { | |
14009 | fix_misreferences_txn = db->get_transaction(); | |
14010 | } | |
14011 | return true; | |
14012 | } | |
14013 | return false; | |
14014 | } | |
14015 | ||
14016 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
14017 | { | |
14018 | if (fix_fm_leaked_txn) { | |
14019 | db->submit_transaction_sync(fix_fm_leaked_txn); | |
14020 | fix_fm_leaked_txn = nullptr; | |
14021 | } | |
14022 | if (fix_fm_false_free_txn) { | |
14023 | db->submit_transaction_sync(fix_fm_false_free_txn); | |
14024 | fix_fm_false_free_txn = nullptr; | |
14025 | } | |
14026 | if (remove_key_txn) { | |
14027 | db->submit_transaction_sync(remove_key_txn); | |
14028 | remove_key_txn = nullptr; | |
14029 | } | |
14030 | if (fix_misreferences_txn) { | |
14031 | db->submit_transaction_sync(fix_misreferences_txn); | |
14032 | fix_misreferences_txn = nullptr; | |
14033 | } | |
14034 | if (fix_shared_blob_txn) { | |
14035 | db->submit_transaction_sync(fix_shared_blob_txn); | |
14036 | fix_shared_blob_txn = nullptr; | |
14037 | } | |
14038 | ||
14039 | if (fix_statfs_txn) { | |
14040 | db->submit_transaction_sync(fix_statfs_txn); | |
14041 | fix_statfs_txn = nullptr; | |
14042 | } | |
14043 | unsigned repaired = to_repair_cnt; | |
14044 | to_repair_cnt = 0; | |
14045 | return repaired; | |
14046 | } | |
14047 | ||
14048 | // ======================================================= | |
14049 |