]>
Commit | Line | Data |
---|---|---|
11fdf7f2 | 1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
7c673cae FG |
2 | // vim: ts=8 sw=2 smarttab |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2014 Red Hat | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
15 | #include <unistd.h> | |
16 | #include <stdlib.h> | |
17 | #include <sys/types.h> | |
18 | #include <sys/stat.h> | |
19 | #include <fcntl.h> | |
20 | ||
eafe8130 | 21 | #include <boost/container/flat_set.hpp> |
9f95a23c | 22 | #include "boost/algorithm/string.hpp" |
eafe8130 | 23 | |
31f18b77 FG |
24 | #include "include/cpp-btree/btree_set.h" |
25 | ||
9f95a23c | 26 | #include "bluestore_common.h" |
7c673cae FG |
27 | #include "BlueStore.h" |
28 | #include "os/kv.h" | |
29 | #include "include/compat.h" | |
30 | #include "include/intarith.h" | |
31 | #include "include/stringify.h" | |
11fdf7f2 TL |
32 | #include "include/str_map.h" |
33 | #include "include/util.h" | |
7c673cae FG |
34 | #include "common/errno.h" |
35 | #include "common/safe_io.h" | |
91327a77 | 36 | #include "common/PriorityCache.h" |
9f95a23c | 37 | #include "common/RWLock.h" |
7c673cae FG |
38 | #include "Allocator.h" |
39 | #include "FreelistManager.h" | |
40 | #include "BlueFS.h" | |
41 | #include "BlueRocksEnv.h" | |
42 | #include "auth/Crypto.h" | |
43 | #include "common/EventTrace.h" | |
91327a77 | 44 | #include "perfglue/heap_profiler.h" |
11fdf7f2 TL |
45 | #include "common/blkdev.h" |
46 | #include "common/numa.h" | |
7c673cae | 47 | |
9f95a23c TL |
48 | #if defined(WITH_LTTNG) |
49 | #define TRACEPOINT_DEFINE | |
50 | #define TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
51 | #include "tracing/bluestore.h" | |
52 | #undef TRACEPOINT_PROBE_DYNAMIC_LINKAGE | |
53 | #undef TRACEPOINT_DEFINE | |
54 | #else | |
55 | #define tracepoint(...) | |
56 | #endif | |
57 | ||
7c673cae FG |
58 | #define dout_context cct |
59 | #define dout_subsys ceph_subsys_bluestore | |
60 | ||
31f18b77 FG |
61 | using bid_t = decltype(BlueStore::Blob::id); |
62 | ||
63 | // bluestore_cache_onode | |
7c673cae | 64 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Onode, bluestore_onode, |
31f18b77 | 65 | bluestore_cache_onode); |
7c673cae | 66 | |
31f18b77 | 67 | // bluestore_cache_other |
7c673cae | 68 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Buffer, bluestore_buffer, |
31f18b77 | 69 | bluestore_cache_other); |
7c673cae | 70 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Extent, bluestore_extent, |
31f18b77 | 71 | bluestore_cache_other); |
7c673cae | 72 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::Blob, bluestore_blob, |
31f18b77 | 73 | bluestore_cache_other); |
7c673cae | 74 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::SharedBlob, bluestore_shared_blob, |
31f18b77 FG |
75 | bluestore_cache_other); |
76 | ||
77 | // bluestore_txc | |
78 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueStore::TransContext, bluestore_transcontext, | |
79 | bluestore_txc); | |
80 | ||
7c673cae FG |
81 | |
82 | // kv store prefixes | |
11fdf7f2 TL |
83 | const string PREFIX_SUPER = "S"; // field -> value |
84 | const string PREFIX_STAT = "T"; // field -> value(int64 array) | |
85 | const string PREFIX_COLL = "C"; // collection name -> cnode_t | |
86 | const string PREFIX_OBJ = "O"; // object name -> onode_t | |
87 | const string PREFIX_OMAP = "M"; // u64 + keyname -> value | |
88 | const string PREFIX_PGMETA_OMAP = "P"; // u64 + keyname -> value(for meta coll) | |
9f95a23c | 89 | const string PREFIX_PERPOOL_OMAP = "m"; // s64 + u64 + keyname -> value |
11fdf7f2 TL |
90 | const string PREFIX_DEFERRED = "L"; // id -> deferred_transaction_t |
91 | const string PREFIX_ALLOC = "B"; // u64 offset -> u64 length (freelist) | |
92 | const string PREFIX_ALLOC_BITMAP = "b";// (see BitmapFreelistManager) | |
7c673cae FG |
93 | const string PREFIX_SHARED_BLOB = "X"; // u64 offset -> shared_blob_t |
94 | ||
11fdf7f2 TL |
95 | const string BLUESTORE_GLOBAL_STATFS_KEY = "bluestore_statfs"; |
96 | ||
7c673cae FG |
97 | // write a label in the first block. always use this size. note that |
98 | // bluefs makes a matching assumption about the location of its | |
99 | // superblock (always the second block of the device). | |
100 | #define BDEV_LABEL_BLOCK_SIZE 4096 | |
101 | ||
102 | // reserve: label (4k) + bluefs super (4k), which means we start at 8k. | |
103 | #define SUPER_RESERVED 8192 | |
104 | ||
105 | #define OBJECT_MAX_SIZE 0xffffffff // 32 bits | |
106 | ||
107 | ||
108 | /* | |
109 | * extent map blob encoding | |
110 | * | |
111 | * we use the low bits of the blobid field to indicate some common scenarios | |
112 | * and spanning vs local ids. See ExtentMap::{encode,decode}_some(). | |
113 | */ | |
114 | #define BLOBID_FLAG_CONTIGUOUS 0x1 // this extent starts at end of previous | |
115 | #define BLOBID_FLAG_ZEROOFFSET 0x2 // blob_offset is 0 | |
116 | #define BLOBID_FLAG_SAMELENGTH 0x4 // length matches previous extent | |
117 | #define BLOBID_FLAG_SPANNING 0x8 // has spanning blob id | |
118 | #define BLOBID_SHIFT_BITS 4 | |
119 | ||
120 | /* | |
121 | * object name key structure | |
122 | * | |
123 | * encoded u8: shard + 2^7 (so that it sorts properly) | |
124 | * encoded u64: poolid + 2^63 (so that it sorts properly) | |
125 | * encoded u32: hash (bit reversed) | |
126 | * | |
127 | * escaped string: namespace | |
128 | * | |
129 | * escaped string: key or object name | |
130 | * 1 char: '<', '=', or '>'. if =, then object key == object name, and | |
131 | * we are done. otherwise, we are followed by the object name. | |
132 | * escaped string: object name (unless '=' above) | |
133 | * | |
134 | * encoded u64: snap | |
135 | * encoded u64: generation | |
136 | * 'o' | |
137 | */ | |
138 | #define ONODE_KEY_SUFFIX 'o' | |
139 | ||
140 | /* | |
141 | * extent shard key | |
142 | * | |
143 | * object prefix key | |
144 | * u32 | |
145 | * 'x' | |
146 | */ | |
147 | #define EXTENT_SHARD_KEY_SUFFIX 'x' | |
148 | ||
149 | /* | |
150 | * string encoding in the key | |
151 | * | |
152 | * The key string needs to lexicographically sort the same way that | |
153 | * ghobject_t does. We do this by escaping anything <= to '#' with # | |
154 | * plus a 2 digit hex string, and anything >= '~' with ~ plus the two | |
155 | * hex digits. | |
156 | * | |
157 | * We use ! as a terminator for strings; this works because it is < # | |
158 | * and will get escaped if it is present in the string. | |
159 | * | |
160 | */ | |
161 | template<typename S> | |
162 | static void append_escaped(const string &in, S *out) | |
163 | { | |
224ce89b WB |
164 | char hexbyte[in.length() * 3 + 1]; |
165 | char* ptr = &hexbyte[0]; | |
7c673cae FG |
166 | for (string::const_iterator i = in.begin(); i != in.end(); ++i) { |
167 | if (*i <= '#') { | |
224ce89b WB |
168 | *ptr++ = '#'; |
169 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
170 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 171 | } else if (*i >= '~') { |
224ce89b WB |
172 | *ptr++ = '~'; |
173 | *ptr++ = "0123456789abcdef"[(*i >> 4) & 0x0f]; | |
174 | *ptr++ = "0123456789abcdef"[*i & 0x0f]; | |
7c673cae | 175 | } else { |
224ce89b | 176 | *ptr++ = *i; |
7c673cae FG |
177 | } |
178 | } | |
224ce89b WB |
179 | *ptr++ = '!'; |
180 | out->append(hexbyte, ptr - &hexbyte[0]); | |
181 | } | |
182 | ||
183 | inline unsigned h2i(char c) | |
184 | { | |
185 | if ((c >= '0') && (c <= '9')) { | |
186 | return c - 0x30; | |
187 | } else if ((c >= 'a') && (c <= 'f')) { | |
188 | return c - 'a' + 10; | |
189 | } else if ((c >= 'A') && (c <= 'F')) { | |
190 | return c - 'A' + 10; | |
191 | } else { | |
192 | return 256; // make it always larger than 255 | |
193 | } | |
7c673cae FG |
194 | } |
195 | ||
196 | static int decode_escaped(const char *p, string *out) | |
197 | { | |
224ce89b WB |
198 | char buff[256]; |
199 | char* ptr = &buff[0]; | |
200 | char* max = &buff[252]; | |
7c673cae FG |
201 | const char *orig_p = p; |
202 | while (*p && *p != '!') { | |
203 | if (*p == '#' || *p == '~') { | |
224ce89b WB |
204 | unsigned hex = 0; |
205 | p++; | |
206 | hex = h2i(*p++) << 4; | |
207 | if (hex > 255) { | |
208 | return -EINVAL; | |
209 | } | |
210 | hex |= h2i(*p++); | |
211 | if (hex > 255) { | |
212 | return -EINVAL; | |
213 | } | |
214 | *ptr++ = hex; | |
7c673cae | 215 | } else { |
224ce89b WB |
216 | *ptr++ = *p++; |
217 | } | |
218 | if (ptr > max) { | |
219 | out->append(buff, ptr-buff); | |
220 | ptr = &buff[0]; | |
7c673cae FG |
221 | } |
222 | } | |
224ce89b WB |
223 | if (ptr != buff) { |
224 | out->append(buff, ptr-buff); | |
225 | } | |
7c673cae FG |
226 | return p - orig_p; |
227 | } | |
228 | ||
229 | // some things we encode in binary (as le32 or le64); print the | |
230 | // resulting key strings nicely | |
231 | template<typename S> | |
232 | static string pretty_binary_string(const S& in) | |
233 | { | |
234 | char buf[10]; | |
235 | string out; | |
236 | out.reserve(in.length() * 3); | |
237 | enum { NONE, HEX, STRING } mode = NONE; | |
238 | unsigned from = 0, i; | |
239 | for (i=0; i < in.length(); ++i) { | |
240 | if ((in[i] < 32 || (unsigned char)in[i] > 126) || | |
241 | (mode == HEX && in.length() - i >= 4 && | |
242 | ((in[i] < 32 || (unsigned char)in[i] > 126) || | |
243 | (in[i+1] < 32 || (unsigned char)in[i+1] > 126) || | |
244 | (in[i+2] < 32 || (unsigned char)in[i+2] > 126) || | |
245 | (in[i+3] < 32 || (unsigned char)in[i+3] > 126)))) { | |
246 | if (mode == STRING) { | |
247 | out.append(in.c_str() + from, i - from); | |
248 | out.push_back('\''); | |
249 | } | |
250 | if (mode != HEX) { | |
251 | out.append("0x"); | |
252 | mode = HEX; | |
253 | } | |
254 | if (in.length() - i >= 4) { | |
255 | // print a whole u32 at once | |
256 | snprintf(buf, sizeof(buf), "%08x", | |
257 | (uint32_t)(((unsigned char)in[i] << 24) | | |
258 | ((unsigned char)in[i+1] << 16) | | |
259 | ((unsigned char)in[i+2] << 8) | | |
260 | ((unsigned char)in[i+3] << 0))); | |
261 | i += 3; | |
262 | } else { | |
263 | snprintf(buf, sizeof(buf), "%02x", (int)(unsigned char)in[i]); | |
264 | } | |
265 | out.append(buf); | |
266 | } else { | |
267 | if (mode != STRING) { | |
268 | out.push_back('\''); | |
269 | mode = STRING; | |
270 | from = i; | |
271 | } | |
272 | } | |
273 | } | |
274 | if (mode == STRING) { | |
275 | out.append(in.c_str() + from, i - from); | |
276 | out.push_back('\''); | |
277 | } | |
278 | return out; | |
279 | } | |
280 | ||
281 | template<typename T> | |
282 | static void _key_encode_shard(shard_id_t shard, T *key) | |
283 | { | |
284 | key->push_back((char)((uint8_t)shard.id + (uint8_t)0x80)); | |
285 | } | |
286 | ||
287 | static const char *_key_decode_shard(const char *key, shard_id_t *pshard) | |
288 | { | |
289 | pshard->id = (uint8_t)*key - (uint8_t)0x80; | |
290 | return key + 1; | |
291 | } | |
292 | ||
293 | static void get_coll_key_range(const coll_t& cid, int bits, | |
294 | string *temp_start, string *temp_end, | |
295 | string *start, string *end) | |
296 | { | |
297 | temp_start->clear(); | |
298 | temp_end->clear(); | |
299 | start->clear(); | |
300 | end->clear(); | |
301 | ||
302 | spg_t pgid; | |
303 | if (cid.is_pg(&pgid)) { | |
304 | _key_encode_shard(pgid.shard, start); | |
305 | *temp_start = *start; | |
306 | ||
307 | _key_encode_u64(pgid.pool() + 0x8000000000000000ull, start); | |
308 | _key_encode_u64((-2ll - pgid.pool()) + 0x8000000000000000ull, temp_start); | |
309 | ||
310 | *end = *start; | |
311 | *temp_end = *temp_start; | |
312 | ||
313 | uint32_t reverse_hash = hobject_t::_reverse_bits(pgid.ps()); | |
314 | _key_encode_u32(reverse_hash, start); | |
315 | _key_encode_u32(reverse_hash, temp_start); | |
316 | ||
317 | uint64_t end_hash = reverse_hash + (1ull << (32 - bits)); | |
318 | if (end_hash > 0xffffffffull) | |
319 | end_hash = 0xffffffffull; | |
320 | ||
321 | _key_encode_u32(end_hash, end); | |
322 | _key_encode_u32(end_hash, temp_end); | |
323 | } else { | |
324 | _key_encode_shard(shard_id_t::NO_SHARD, start); | |
325 | _key_encode_u64(-1ull + 0x8000000000000000ull, start); | |
326 | *end = *start; | |
327 | _key_encode_u32(0, start); | |
328 | _key_encode_u32(0xffffffff, end); | |
329 | ||
330 | // no separate temp section | |
331 | *temp_start = *end; | |
332 | *temp_end = *end; | |
333 | } | |
334 | } | |
335 | ||
336 | static void get_shared_blob_key(uint64_t sbid, string *key) | |
337 | { | |
338 | key->clear(); | |
339 | _key_encode_u64(sbid, key); | |
340 | } | |
341 | ||
342 | static int get_key_shared_blob(const string& key, uint64_t *sbid) | |
343 | { | |
344 | const char *p = key.c_str(); | |
345 | if (key.length() < sizeof(uint64_t)) | |
346 | return -1; | |
224ce89b | 347 | _key_decode_u64(p, sbid); |
7c673cae FG |
348 | return 0; |
349 | } | |
350 | ||
351 | template<typename S> | |
352 | static int get_key_object(const S& key, ghobject_t *oid) | |
353 | { | |
354 | int r; | |
355 | const char *p = key.c_str(); | |
356 | ||
357 | if (key.length() < 1 + 8 + 4) | |
358 | return -1; | |
359 | p = _key_decode_shard(p, &oid->shard_id); | |
360 | ||
361 | uint64_t pool; | |
362 | p = _key_decode_u64(p, &pool); | |
363 | oid->hobj.pool = pool - 0x8000000000000000ull; | |
364 | ||
365 | unsigned hash; | |
366 | p = _key_decode_u32(p, &hash); | |
367 | ||
368 | oid->hobj.set_bitwise_key_u32(hash); | |
369 | ||
370 | r = decode_escaped(p, &oid->hobj.nspace); | |
371 | if (r < 0) | |
372 | return -2; | |
373 | p += r + 1; | |
374 | ||
375 | string k; | |
376 | r = decode_escaped(p, &k); | |
377 | if (r < 0) | |
378 | return -3; | |
379 | p += r + 1; | |
380 | if (*p == '=') { | |
381 | // no key | |
382 | ++p; | |
383 | oid->hobj.oid.name = k; | |
384 | } else if (*p == '<' || *p == '>') { | |
385 | // key + name | |
386 | ++p; | |
387 | r = decode_escaped(p, &oid->hobj.oid.name); | |
388 | if (r < 0) | |
389 | return -5; | |
390 | p += r + 1; | |
391 | oid->hobj.set_key(k); | |
392 | } else { | |
393 | // malformed | |
394 | return -6; | |
395 | } | |
396 | ||
397 | p = _key_decode_u64(p, &oid->hobj.snap.val); | |
398 | p = _key_decode_u64(p, &oid->generation); | |
399 | ||
400 | if (*p != ONODE_KEY_SUFFIX) { | |
401 | return -7; | |
402 | } | |
403 | p++; | |
404 | if (*p) { | |
405 | // if we get something other than a null terminator here, | |
406 | // something goes wrong. | |
407 | return -8; | |
408 | } | |
409 | ||
410 | return 0; | |
411 | } | |
412 | ||
413 | template<typename S> | |
414 | static void get_object_key(CephContext *cct, const ghobject_t& oid, S *key) | |
415 | { | |
416 | key->clear(); | |
417 | ||
418 | size_t max_len = 1 + 8 + 4 + | |
419 | (oid.hobj.nspace.length() * 3 + 1) + | |
420 | (oid.hobj.get_key().length() * 3 + 1) + | |
421 | 1 + // for '<', '=', or '>' | |
422 | (oid.hobj.oid.name.length() * 3 + 1) + | |
423 | 8 + 8 + 1; | |
424 | key->reserve(max_len); | |
425 | ||
426 | _key_encode_shard(oid.shard_id, key); | |
427 | _key_encode_u64(oid.hobj.pool + 0x8000000000000000ull, key); | |
428 | _key_encode_u32(oid.hobj.get_bitwise_key_u32(), key); | |
429 | ||
430 | append_escaped(oid.hobj.nspace, key); | |
431 | ||
432 | if (oid.hobj.get_key().length()) { | |
433 | // is a key... could be < = or >. | |
434 | append_escaped(oid.hobj.get_key(), key); | |
435 | // (ASCII chars < = and > sort in that order, yay) | |
436 | int r = oid.hobj.get_key().compare(oid.hobj.oid.name); | |
437 | if (r) { | |
438 | key->append(r > 0 ? ">" : "<"); | |
439 | append_escaped(oid.hobj.oid.name, key); | |
440 | } else { | |
441 | // same as no key | |
442 | key->append("="); | |
443 | } | |
444 | } else { | |
445 | // no key | |
446 | append_escaped(oid.hobj.oid.name, key); | |
447 | key->append("="); | |
448 | } | |
449 | ||
450 | _key_encode_u64(oid.hobj.snap, key); | |
451 | _key_encode_u64(oid.generation, key); | |
452 | ||
453 | key->push_back(ONODE_KEY_SUFFIX); | |
454 | ||
455 | // sanity check | |
456 | if (true) { | |
457 | ghobject_t t; | |
458 | int r = get_key_object(*key, &t); | |
459 | if (r || t != oid) { | |
460 | derr << " r " << r << dendl; | |
461 | derr << "key " << pretty_binary_string(*key) << dendl; | |
462 | derr << "oid " << oid << dendl; | |
463 | derr << " t " << t << dendl; | |
11fdf7f2 | 464 | ceph_assert(r == 0 && t == oid); |
7c673cae FG |
465 | } |
466 | } | |
467 | } | |
468 | ||
469 | ||
470 | // extent shard keys are the onode key, plus a u32, plus 'x'. the trailing | |
471 | // char lets us quickly test whether it is a shard key without decoding any | |
472 | // of the prefix bytes. | |
473 | template<typename S> | |
474 | static void get_extent_shard_key(const S& onode_key, uint32_t offset, | |
475 | string *key) | |
476 | { | |
477 | key->clear(); | |
478 | key->reserve(onode_key.length() + 4 + 1); | |
479 | key->append(onode_key.c_str(), onode_key.size()); | |
480 | _key_encode_u32(offset, key); | |
481 | key->push_back(EXTENT_SHARD_KEY_SUFFIX); | |
482 | } | |
483 | ||
484 | static void rewrite_extent_shard_key(uint32_t offset, string *key) | |
485 | { | |
11fdf7f2 TL |
486 | ceph_assert(key->size() > sizeof(uint32_t) + 1); |
487 | ceph_assert(*key->rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
488 | _key_encode_u32(offset, key->size() - sizeof(uint32_t) - 1, key); |
489 | } | |
490 | ||
491 | template<typename S> | |
492 | static void generate_extent_shard_key_and_apply( | |
493 | const S& onode_key, | |
494 | uint32_t offset, | |
495 | string *key, | |
496 | std::function<void(const string& final_key)> apply) | |
497 | { | |
498 | if (key->empty()) { // make full key | |
11fdf7f2 | 499 | ceph_assert(!onode_key.empty()); |
7c673cae FG |
500 | get_extent_shard_key(onode_key, offset, key); |
501 | } else { | |
502 | rewrite_extent_shard_key(offset, key); | |
503 | } | |
504 | apply(*key); | |
505 | } | |
506 | ||
507 | int get_key_extent_shard(const string& key, string *onode_key, uint32_t *offset) | |
508 | { | |
11fdf7f2 TL |
509 | ceph_assert(key.size() > sizeof(uint32_t) + 1); |
510 | ceph_assert(*key.rbegin() == EXTENT_SHARD_KEY_SUFFIX); | |
7c673cae FG |
511 | int okey_len = key.size() - sizeof(uint32_t) - 1; |
512 | *onode_key = key.substr(0, okey_len); | |
513 | const char *p = key.data() + okey_len; | |
224ce89b | 514 | _key_decode_u32(p, offset); |
7c673cae FG |
515 | return 0; |
516 | } | |
517 | ||
518 | static bool is_extent_shard_key(const string& key) | |
519 | { | |
520 | return *key.rbegin() == EXTENT_SHARD_KEY_SUFFIX; | |
521 | } | |
522 | ||
7c673cae FG |
523 | static void get_deferred_key(uint64_t seq, string *out) |
524 | { | |
525 | _key_encode_u64(seq, out); | |
526 | } | |
527 | ||
11fdf7f2 TL |
528 | static void get_pool_stat_key(int64_t pool_id, string *key) |
529 | { | |
530 | key->clear(); | |
531 | _key_encode_u64(pool_id, key); | |
532 | } | |
533 | ||
534 | static int get_key_pool_stat(const string& key, uint64_t* pool_id) | |
535 | { | |
536 | const char *p = key.c_str(); | |
537 | if (key.length() < sizeof(uint64_t)) | |
538 | return -1; | |
539 | _key_decode_u64(p, pool_id); | |
540 | return 0; | |
541 | } | |
7c673cae | 542 | |
81eedcae TL |
543 | template <int LogLevelV> |
544 | void _dump_extent_map(CephContext *cct, const BlueStore::ExtentMap &em) | |
545 | { | |
546 | uint64_t pos = 0; | |
547 | for (auto& s : em.shards) { | |
548 | dout(LogLevelV) << __func__ << " shard " << *s.shard_info | |
549 | << (s.loaded ? " (loaded)" : "") | |
550 | << (s.dirty ? " (dirty)" : "") | |
551 | << dendl; | |
552 | } | |
553 | for (auto& e : em.extent_map) { | |
554 | dout(LogLevelV) << __func__ << " " << e << dendl; | |
555 | ceph_assert(e.logical_offset >= pos); | |
556 | pos = e.logical_offset + e.length; | |
557 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
558 | if (blob.has_csum()) { | |
559 | vector<uint64_t> v; | |
560 | unsigned n = blob.get_csum_count(); | |
561 | for (unsigned i = 0; i < n; ++i) | |
562 | v.push_back(blob.get_csum_item(i)); | |
563 | dout(LogLevelV) << __func__ << " csum: " << std::hex << v << std::dec | |
564 | << dendl; | |
565 | } | |
566 | std::lock_guard l(e.blob->shared_blob->get_cache()->lock); | |
567 | for (auto& i : e.blob->shared_blob->bc.buffer_map) { | |
568 | dout(LogLevelV) << __func__ << " 0x" << std::hex << i.first | |
569 | << "~" << i.second->length << std::dec | |
570 | << " " << *i.second << dendl; | |
571 | } | |
572 | } | |
573 | } | |
574 | ||
575 | template <int LogLevelV> | |
576 | void _dump_onode(CephContext *cct, const BlueStore::Onode& o) | |
577 | { | |
578 | if (!cct->_conf->subsys.should_gather<ceph_subsys_bluestore, LogLevelV>()) | |
579 | return; | |
580 | dout(LogLevelV) << __func__ << " " << &o << " " << o.oid | |
581 | << " nid " << o.onode.nid | |
582 | << " size 0x" << std::hex << o.onode.size | |
583 | << " (" << std::dec << o.onode.size << ")" | |
584 | << " expected_object_size " << o.onode.expected_object_size | |
585 | << " expected_write_size " << o.onode.expected_write_size | |
586 | << " in " << o.onode.extent_map_shards.size() << " shards" | |
587 | << ", " << o.extent_map.spanning_blob_map.size() | |
588 | << " spanning blobs" | |
589 | << dendl; | |
590 | for (auto p = o.onode.attrs.begin(); | |
591 | p != o.onode.attrs.end(); | |
592 | ++p) { | |
593 | dout(LogLevelV) << __func__ << " attr " << p->first | |
594 | << " len " << p->second.length() << dendl; | |
595 | } | |
596 | _dump_extent_map<LogLevelV>(cct, o.extent_map); | |
597 | } | |
598 | ||
599 | template <int LogLevelV> | |
600 | void _dump_transaction(CephContext *cct, ObjectStore::Transaction *t) | |
601 | { | |
602 | dout(LogLevelV) << __func__ << " transaction dump:\n"; | |
603 | JSONFormatter f(true); | |
604 | f.open_object_section("transaction"); | |
605 | t->dump(&f); | |
606 | f.close_section(); | |
607 | f.flush(*_dout); | |
608 | *_dout << dendl; | |
609 | } | |
610 | ||
7c673cae FG |
611 | // merge operators |
612 | ||
613 | struct Int64ArrayMergeOperator : public KeyValueDB::MergeOperator { | |
614 | void merge_nonexistent( | |
615 | const char *rdata, size_t rlen, std::string *new_value) override { | |
616 | *new_value = std::string(rdata, rlen); | |
617 | } | |
618 | void merge( | |
619 | const char *ldata, size_t llen, | |
620 | const char *rdata, size_t rlen, | |
621 | std::string *new_value) override { | |
11fdf7f2 TL |
622 | ceph_assert(llen == rlen); |
623 | ceph_assert((rlen % 8) == 0); | |
7c673cae | 624 | new_value->resize(rlen); |
eafe8130 TL |
625 | const ceph_le64* lv = (const ceph_le64*)ldata; |
626 | const ceph_le64* rv = (const ceph_le64*)rdata; | |
627 | ceph_le64* nv = &(ceph_le64&)new_value->at(0); | |
7c673cae FG |
628 | for (size_t i = 0; i < rlen >> 3; ++i) { |
629 | nv[i] = lv[i] + rv[i]; | |
630 | } | |
631 | } | |
632 | // We use each operator name and each prefix to construct the | |
633 | // overall RocksDB operator name for consistency check at open time. | |
91327a77 | 634 | const char *name() const override { |
7c673cae FG |
635 | return "int64_array"; |
636 | } | |
637 | }; | |
638 | ||
639 | ||
640 | // Buffer | |
641 | ||
642 | ostream& operator<<(ostream& out, const BlueStore::Buffer& b) | |
643 | { | |
644 | out << "buffer(" << &b << " space " << b.space << " 0x" << std::hex | |
645 | << b.offset << "~" << b.length << std::dec | |
646 | << " " << BlueStore::Buffer::get_state_name(b.state); | |
647 | if (b.flags) | |
648 | out << " " << BlueStore::Buffer::get_flag_name(b.flags); | |
649 | return out << ")"; | |
650 | } | |
651 | ||
652 | // Garbage Collector | |
653 | ||
654 | void BlueStore::GarbageCollector::process_protrusive_extents( | |
655 | const BlueStore::ExtentMap& extent_map, | |
656 | uint64_t start_offset, | |
657 | uint64_t end_offset, | |
658 | uint64_t start_touch_offset, | |
659 | uint64_t end_touch_offset, | |
660 | uint64_t min_alloc_size) | |
661 | { | |
11fdf7f2 | 662 | ceph_assert(start_offset <= start_touch_offset && end_offset>= end_touch_offset); |
7c673cae | 663 | |
11fdf7f2 TL |
664 | uint64_t lookup_start_offset = p2align(start_offset, min_alloc_size); |
665 | uint64_t lookup_end_offset = round_up_to(end_offset, min_alloc_size); | |
7c673cae FG |
666 | |
667 | dout(30) << __func__ << " (hex): [" << std::hex | |
668 | << lookup_start_offset << ", " << lookup_end_offset | |
669 | << ")" << std::dec << dendl; | |
670 | ||
671 | for (auto it = extent_map.seek_lextent(lookup_start_offset); | |
672 | it != extent_map.extent_map.end() && | |
673 | it->logical_offset < lookup_end_offset; | |
674 | ++it) { | |
675 | uint64_t alloc_unit_start = it->logical_offset / min_alloc_size; | |
676 | uint64_t alloc_unit_end = (it->logical_end() - 1) / min_alloc_size; | |
677 | ||
678 | dout(30) << __func__ << " " << *it | |
679 | << "alloc_units: " << alloc_unit_start << ".." << alloc_unit_end | |
680 | << dendl; | |
681 | ||
682 | Blob* b = it->blob.get(); | |
683 | ||
684 | if (it->logical_offset >=start_touch_offset && | |
685 | it->logical_end() <= end_touch_offset) { | |
686 | // Process extents within the range affected by | |
687 | // the current write request. | |
688 | // Need to take into account if existing extents | |
689 | // can be merged with them (uncompressed case) | |
690 | if (!b->get_blob().is_compressed()) { | |
691 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
692 | --blob_info_counted->expected_allocations; // don't need to allocate | |
693 | // new AU for compressed | |
694 | // data since another | |
695 | // collocated uncompressed | |
696 | // blob already exists | |
697 | dout(30) << __func__ << " --expected:" | |
698 | << alloc_unit_start << dendl; | |
699 | } | |
700 | used_alloc_unit = alloc_unit_end; | |
701 | blob_info_counted = nullptr; | |
702 | } | |
703 | } else if (b->get_blob().is_compressed()) { | |
704 | ||
705 | // additionally we take compressed blobs that were not impacted | |
706 | // by the write into account too | |
707 | BlobInfo& bi = | |
708 | affected_blobs.emplace( | |
709 | b, BlobInfo(b->get_referenced_bytes())).first->second; | |
710 | ||
711 | int adjust = | |
712 | (used_alloc_unit && used_alloc_unit == alloc_unit_start) ? 0 : 1; | |
713 | bi.expected_allocations += alloc_unit_end - alloc_unit_start + adjust; | |
714 | dout(30) << __func__ << " expected_allocations=" | |
715 | << bi.expected_allocations << " end_au:" | |
716 | << alloc_unit_end << dendl; | |
717 | ||
718 | blob_info_counted = &bi; | |
719 | used_alloc_unit = alloc_unit_end; | |
720 | ||
11fdf7f2 | 721 | ceph_assert(it->length <= bi.referenced_bytes); |
7c673cae FG |
722 | bi.referenced_bytes -= it->length; |
723 | dout(30) << __func__ << " affected_blob:" << *b | |
724 | << " unref 0x" << std::hex << it->length | |
725 | << " referenced = 0x" << bi.referenced_bytes | |
726 | << std::dec << dendl; | |
727 | // NOTE: we can't move specific blob to resulting GC list here | |
728 | // when reference counter == 0 since subsequent extents might | |
729 | // decrement its expected_allocation. | |
730 | // Hence need to enumerate all the extents first. | |
731 | if (!bi.collect_candidate) { | |
732 | bi.first_lextent = it; | |
733 | bi.collect_candidate = true; | |
734 | } | |
735 | bi.last_lextent = it; | |
736 | } else { | |
737 | if (blob_info_counted && used_alloc_unit == alloc_unit_start) { | |
738 | // don't need to allocate new AU for compressed data since another | |
739 | // collocated uncompressed blob already exists | |
740 | --blob_info_counted->expected_allocations; | |
741 | dout(30) << __func__ << " --expected_allocations:" | |
742 | << alloc_unit_start << dendl; | |
743 | } | |
744 | used_alloc_unit = alloc_unit_end; | |
745 | blob_info_counted = nullptr; | |
746 | } | |
747 | } | |
748 | ||
749 | for (auto b_it = affected_blobs.begin(); | |
750 | b_it != affected_blobs.end(); | |
751 | ++b_it) { | |
752 | Blob* b = b_it->first; | |
753 | BlobInfo& bi = b_it->second; | |
754 | if (bi.referenced_bytes == 0) { | |
755 | uint64_t len_on_disk = b_it->first->get_blob().get_ondisk_length(); | |
756 | int64_t blob_expected_for_release = | |
11fdf7f2 | 757 | round_up_to(len_on_disk, min_alloc_size) / min_alloc_size; |
7c673cae FG |
758 | |
759 | dout(30) << __func__ << " " << *(b_it->first) | |
760 | << " expected4release=" << blob_expected_for_release | |
761 | << " expected_allocations=" << bi.expected_allocations | |
762 | << dendl; | |
763 | int64_t benefit = blob_expected_for_release - bi.expected_allocations; | |
11fdf7f2 | 764 | if (benefit >= g_conf()->bluestore_gc_enable_blob_threshold) { |
7c673cae FG |
765 | if (bi.collect_candidate) { |
766 | auto it = bi.first_lextent; | |
767 | bool bExit = false; | |
768 | do { | |
769 | if (it->blob.get() == b) { | |
eafe8130 | 770 | extents_to_collect.insert(it->logical_offset, it->length); |
7c673cae FG |
771 | } |
772 | bExit = it == bi.last_lextent; | |
773 | ++it; | |
31f18b77 | 774 | } while (!bExit); |
7c673cae FG |
775 | } |
776 | expected_for_release += blob_expected_for_release; | |
777 | expected_allocations += bi.expected_allocations; | |
778 | } | |
779 | } | |
780 | } | |
781 | } | |
782 | ||
783 | int64_t BlueStore::GarbageCollector::estimate( | |
784 | uint64_t start_offset, | |
785 | uint64_t length, | |
786 | const BlueStore::ExtentMap& extent_map, | |
787 | const BlueStore::old_extent_map_t& old_extents, | |
788 | uint64_t min_alloc_size) | |
789 | { | |
790 | ||
791 | affected_blobs.clear(); | |
792 | extents_to_collect.clear(); | |
793 | used_alloc_unit = boost::optional<uint64_t >(); | |
794 | blob_info_counted = nullptr; | |
795 | ||
eafe8130 TL |
796 | uint64_t gc_start_offset = start_offset; |
797 | uint64_t gc_end_offset = start_offset + length; | |
7c673cae FG |
798 | |
799 | uint64_t end_offset = start_offset + length; | |
800 | ||
801 | for (auto it = old_extents.begin(); it != old_extents.end(); ++it) { | |
802 | Blob* b = it->e.blob.get(); | |
803 | if (b->get_blob().is_compressed()) { | |
804 | ||
805 | // update gc_start_offset/gc_end_offset if needed | |
806 | gc_start_offset = min(gc_start_offset, (uint64_t)it->e.blob_start()); | |
11fdf7f2 | 807 | gc_end_offset = std::max(gc_end_offset, (uint64_t)it->e.blob_end()); |
7c673cae FG |
808 | |
809 | auto o = it->e.logical_offset; | |
810 | auto l = it->e.length; | |
811 | ||
812 | uint64_t ref_bytes = b->get_referenced_bytes(); | |
813 | // micro optimization to bypass blobs that have no more references | |
814 | if (ref_bytes != 0) { | |
815 | dout(30) << __func__ << " affected_blob:" << *b | |
816 | << " unref 0x" << std::hex << o << "~" << l | |
817 | << std::dec << dendl; | |
818 | affected_blobs.emplace(b, BlobInfo(ref_bytes)); | |
819 | } | |
820 | } | |
821 | } | |
822 | dout(30) << __func__ << " gc range(hex): [" << std::hex | |
823 | << gc_start_offset << ", " << gc_end_offset | |
824 | << ")" << std::dec << dendl; | |
825 | ||
826 | // enumerate preceeding extents to check if they reference affected blobs | |
827 | if (gc_start_offset < start_offset || gc_end_offset > end_offset) { | |
828 | process_protrusive_extents(extent_map, | |
829 | gc_start_offset, | |
830 | gc_end_offset, | |
831 | start_offset, | |
832 | end_offset, | |
833 | min_alloc_size); | |
834 | } | |
835 | return expected_for_release - expected_allocations; | |
836 | } | |
837 | ||
9f95a23c TL |
838 | // LruOnodeCacheShard |
839 | struct LruOnodeCacheShard : public BlueStore::OnodeCacheShard { | |
840 | typedef boost::intrusive::list< | |
841 | BlueStore::Onode, | |
842 | boost::intrusive::member_hook< | |
843 | BlueStore::Onode, | |
844 | boost::intrusive::list_member_hook<>, | |
845 | &BlueStore::Onode::lru_item> > list_t; | |
846 | typedef boost::intrusive::list< | |
847 | BlueStore::Onode, | |
848 | boost::intrusive::member_hook< | |
849 | BlueStore::Onode, | |
850 | boost::intrusive::list_member_hook<>, | |
851 | &BlueStore::Onode::pin_item> > pin_list_t; | |
7c673cae | 852 | |
9f95a23c TL |
853 | list_t lru; |
854 | pin_list_t pin_list; | |
7c673cae | 855 | |
9f95a23c | 856 | explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {} |
7c673cae | 857 | |
9f95a23c TL |
858 | void _add(BlueStore::OnodeRef& o, int level) override |
859 | { | |
860 | ceph_assert(o->s == nullptr); | |
861 | o->s = this; | |
862 | if (o->nref > 1) { | |
863 | pin_list.push_front(*o); | |
864 | o->pinned = true; | |
865 | num_pinned = pin_list.size(); | |
866 | } else { | |
867 | (level > 0) ? lru.push_front(*o) : lru.push_back(*o); | |
868 | } | |
869 | num = lru.size(); | |
eafe8130 | 870 | } |
9f95a23c TL |
871 | void _rm(BlueStore::OnodeRef& o) override |
872 | { | |
873 | o->s = nullptr; | |
874 | if (o->pinned) { | |
875 | o->pinned = false; | |
876 | pin_list.erase(pin_list.iterator_to(*o)); | |
877 | } else { | |
878 | lru.erase(lru.iterator_to(*o)); | |
879 | } | |
880 | num = lru.size(); | |
881 | num_pinned = pin_list.size(); | |
882 | } | |
883 | void _touch(BlueStore::OnodeRef& o) override | |
884 | { | |
885 | if (o->pinned) { | |
886 | return; | |
887 | } | |
888 | lru.erase(lru.iterator_to(*o)); | |
889 | lru.push_front(*o); | |
890 | num = lru.size(); | |
891 | } | |
892 | void _pin(BlueStore::Onode& o) override | |
893 | { | |
894 | if (o.pinned == true) { | |
895 | return; | |
896 | } | |
897 | lru.erase(lru.iterator_to(o)); | |
898 | pin_list.push_front(o); | |
899 | o.pinned = true; | |
900 | num = lru.size(); | |
901 | num_pinned = pin_list.size(); | |
902 | dout(30) << __func__ << " " << o.oid << " pinned" << dendl; | |
7c673cae | 903 | |
9f95a23c TL |
904 | } |
905 | void _unpin(BlueStore::Onode& o) override | |
906 | { | |
907 | if (o.pinned == false) { | |
908 | return; | |
909 | } | |
910 | pin_list.erase(pin_list.iterator_to(o)); | |
911 | lru.push_front(o); | |
912 | o.pinned = false; | |
913 | num = lru.size(); | |
914 | num_pinned = pin_list.size(); | |
915 | dout(30) << __func__ << " " << o.oid << " unpinned" << dendl; | |
916 | } | |
917 | void _trim_to(uint64_t new_size) override | |
918 | { | |
919 | if (new_size >= lru.size()) { | |
920 | return; // don't even try | |
921 | } | |
922 | uint64_t n = lru.size() - new_size; | |
923 | auto p = lru.end(); | |
924 | ceph_assert(p != lru.begin()); | |
925 | --p; | |
926 | while (n > 0) { | |
927 | BlueStore::Onode *o = &*p; | |
928 | dout(30) << __func__ << " rm " << o->oid << dendl; | |
929 | if (p != lru.begin()) { | |
930 | lru.erase(p--); | |
931 | } else { | |
932 | lru.erase(p); | |
933 | ceph_assert(n == 1); | |
934 | } | |
935 | o->s = nullptr; | |
936 | o->get(); // paranoia | |
937 | o->c->onode_map.remove(o->oid); | |
938 | o->put(); | |
939 | --n; | |
940 | } | |
941 | num = lru.size(); | |
942 | } | |
943 | void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override | |
944 | { | |
945 | *onodes += num + num_pinned; | |
946 | *pinned_onodes += num_pinned; | |
947 | } | |
948 | }; | |
7c673cae | 949 | |
9f95a23c TL |
950 | // OnodeCacheShard |
951 | BlueStore::OnodeCacheShard *BlueStore::OnodeCacheShard::create( | |
952 | CephContext* cct, | |
953 | string type, | |
954 | PerfCounters *logger) | |
7c673cae | 955 | { |
9f95a23c TL |
956 | BlueStore::OnodeCacheShard *c = nullptr; |
957 | // Currently we only implement an LRU cache for onodes | |
958 | c = new LruOnodeCacheShard(cct); | |
959 | c->logger = logger; | |
960 | return c; | |
7c673cae FG |
961 | } |
962 | ||
9f95a23c TL |
963 | // LruBufferCacheShard |
964 | struct LruBufferCacheShard : public BlueStore::BufferCacheShard { | |
965 | typedef boost::intrusive::list< | |
966 | BlueStore::Buffer, | |
967 | boost::intrusive::member_hook< | |
968 | BlueStore::Buffer, | |
969 | boost::intrusive::list_member_hook<>, | |
970 | &BlueStore::Buffer::lru_item> > list_t; | |
971 | list_t lru; | |
972 | ||
973 | explicit LruBufferCacheShard(CephContext *cct) : BlueStore::BufferCacheShard(cct) {} | |
974 | ||
975 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override { | |
976 | if (near) { | |
977 | auto q = lru.iterator_to(*near); | |
978 | lru.insert(q, *b); | |
979 | } else if (level > 0) { | |
980 | lru.push_front(*b); | |
981 | } else { | |
982 | lru.push_back(*b); | |
7c673cae | 983 | } |
9f95a23c TL |
984 | buffer_bytes += b->length; |
985 | num = lru.size(); | |
986 | } | |
987 | void _rm(BlueStore::Buffer *b) override { | |
988 | ceph_assert(buffer_bytes >= b->length); | |
989 | buffer_bytes -= b->length; | |
990 | auto q = lru.iterator_to(*b); | |
991 | lru.erase(q); | |
992 | num = lru.size(); | |
993 | } | |
994 | void _move(BlueStore::BufferCacheShard *src, BlueStore::Buffer *b) override { | |
995 | src->_rm(b); | |
996 | _add(b, 0, nullptr); | |
997 | } | |
998 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override { | |
999 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1000 | buffer_bytes += delta; | |
1001 | } | |
1002 | void _touch(BlueStore::Buffer *b) override { | |
1003 | auto p = lru.iterator_to(*b); | |
1004 | lru.erase(p); | |
1005 | lru.push_front(*b); | |
1006 | num = lru.size(); | |
1007 | _audit("_touch_buffer end"); | |
1008 | } | |
7c673cae | 1009 | |
9f95a23c TL |
1010 | void _trim_to(uint64_t max) override |
1011 | { | |
1012 | while (buffer_bytes > max) { | |
1013 | auto i = lru.rbegin(); | |
1014 | if (i == lru.rend()) { | |
1015 | // stop if lru is now empty | |
7c673cae FG |
1016 | break; |
1017 | } | |
1018 | ||
9f95a23c TL |
1019 | BlueStore::Buffer *b = &*i; |
1020 | ceph_assert(b->is_clean()); | |
1021 | dout(20) << __func__ << " rm " << *b << dendl; | |
1022 | b->space->_rm_buffer(this, b); | |
7c673cae | 1023 | } |
9f95a23c | 1024 | num = lru.size(); |
7c673cae | 1025 | } |
7c673cae | 1026 | |
9f95a23c TL |
1027 | void add_stats(uint64_t *extents, |
1028 | uint64_t *blobs, | |
1029 | uint64_t *buffers, | |
1030 | uint64_t *bytes) override { | |
1031 | *extents += num_extents; | |
1032 | *blobs += num_blobs; | |
1033 | *buffers += num; | |
1034 | *bytes += buffer_bytes; | |
7c673cae | 1035 | } |
9f95a23c TL |
1036 | #ifdef DEBUG_CACHE |
1037 | void _audit(const char *s) override | |
1038 | { | |
1039 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1040 | uint64_t s = 0; | |
1041 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1042 | s += i->length; | |
1043 | } | |
1044 | if (s != buffer_bytes) { | |
1045 | derr << __func__ << " buffer_size " << buffer_bytes << " actual " << s | |
1046 | << dendl; | |
1047 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1048 | derr << __func__ << " " << *i << dendl; | |
1049 | } | |
1050 | ceph_assert(s == buffer_bytes); | |
7c673cae | 1051 | } |
9f95a23c TL |
1052 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1053 | << " ok" << dendl; | |
7c673cae | 1054 | } |
7c673cae | 1055 | #endif |
9f95a23c | 1056 | }; |
7c673cae | 1057 | |
9f95a23c TL |
1058 | // TwoQBufferCacheShard |
1059 | ||
1060 | struct TwoQBufferCacheShard : public BlueStore::BufferCacheShard { | |
1061 | typedef boost::intrusive::list< | |
1062 | BlueStore::Buffer, | |
1063 | boost::intrusive::member_hook< | |
1064 | BlueStore::Buffer, | |
1065 | boost::intrusive::list_member_hook<>, | |
1066 | &BlueStore::Buffer::lru_item> > list_t; | |
1067 | list_t hot; ///< "Am" hot buffers | |
1068 | list_t warm_in; ///< "A1in" newly warm buffers | |
1069 | list_t warm_out; ///< "A1out" empty buffers we've evicted | |
1070 | uint64_t buffer_bytes = 0; ///< bytes | |
1071 | ||
1072 | enum { | |
1073 | BUFFER_NEW = 0, | |
1074 | BUFFER_WARM_IN, ///< in warm_in | |
1075 | BUFFER_WARM_OUT, ///< in warm_out | |
1076 | BUFFER_HOT, ///< in hot | |
1077 | BUFFER_TYPE_MAX | |
1078 | }; | |
7c673cae | 1079 | |
9f95a23c | 1080 | uint64_t list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type |
7c673cae | 1081 | |
9f95a23c TL |
1082 | public: |
1083 | explicit TwoQBufferCacheShard(CephContext *cct) : BufferCacheShard(cct) {} | |
7c673cae | 1084 | |
9f95a23c TL |
1085 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override |
1086 | { | |
1087 | dout(20) << __func__ << " level " << level << " near " << near | |
1088 | << " on " << *b | |
1089 | << " which has cache_private " << b->cache_private << dendl; | |
1090 | if (near) { | |
1091 | b->cache_private = near->cache_private; | |
1092 | switch (b->cache_private) { | |
1093 | case BUFFER_WARM_IN: | |
1094 | warm_in.insert(warm_in.iterator_to(*near), *b); | |
1095 | break; | |
1096 | case BUFFER_WARM_OUT: | |
1097 | ceph_assert(b->is_empty()); | |
1098 | warm_out.insert(warm_out.iterator_to(*near), *b); | |
1099 | break; | |
1100 | case BUFFER_HOT: | |
1101 | hot.insert(hot.iterator_to(*near), *b); | |
1102 | break; | |
1103 | default: | |
1104 | ceph_abort_msg("bad cache_private"); | |
1105 | } | |
1106 | } else if (b->cache_private == BUFFER_NEW) { | |
1107 | b->cache_private = BUFFER_WARM_IN; | |
1108 | if (level > 0) { | |
1109 | warm_in.push_front(*b); | |
1110 | } else { | |
1111 | // take caller hint to start at the back of the warm queue | |
1112 | warm_in.push_back(*b); | |
1113 | } | |
1114 | } else { | |
1115 | // we got a hint from discard | |
1116 | switch (b->cache_private) { | |
1117 | case BUFFER_WARM_IN: | |
1118 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1119 | // do this. | |
1120 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1121 | warm_in.push_front(*b); | |
1122 | break; | |
1123 | case BUFFER_WARM_OUT: | |
1124 | b->cache_private = BUFFER_HOT; | |
1125 | // move to hot. fall-thru | |
1126 | case BUFFER_HOT: | |
1127 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1128 | hot.push_front(*b); | |
1129 | break; | |
1130 | default: | |
1131 | ceph_abort_msg("bad cache_private"); | |
1132 | } | |
1133 | } | |
1134 | if (!b->is_empty()) { | |
1135 | buffer_bytes += b->length; | |
1136 | list_bytes[b->cache_private] += b->length; | |
1137 | } | |
1138 | num = hot.size() + warm_in.size(); | |
1139 | } | |
1140 | ||
1141 | void _rm(BlueStore::Buffer *b) override | |
1142 | { | |
1143 | dout(20) << __func__ << " " << *b << dendl; | |
1144 | if (!b->is_empty()) { | |
1145 | ceph_assert(buffer_bytes >= b->length); | |
1146 | buffer_bytes -= b->length; | |
1147 | ceph_assert(list_bytes[b->cache_private] >= b->length); | |
1148 | list_bytes[b->cache_private] -= b->length; | |
1149 | } | |
7c673cae FG |
1150 | switch (b->cache_private) { |
1151 | case BUFFER_WARM_IN: | |
9f95a23c | 1152 | warm_in.erase(warm_in.iterator_to(*b)); |
7c673cae FG |
1153 | break; |
1154 | case BUFFER_WARM_OUT: | |
9f95a23c | 1155 | warm_out.erase(warm_out.iterator_to(*b)); |
7c673cae FG |
1156 | break; |
1157 | case BUFFER_HOT: | |
9f95a23c | 1158 | hot.erase(hot.iterator_to(*b)); |
7c673cae FG |
1159 | break; |
1160 | default: | |
11fdf7f2 | 1161 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1162 | } |
9f95a23c TL |
1163 | num = hot.size() + warm_in.size(); |
1164 | } | |
1165 | ||
1166 | void _move(BlueStore::BufferCacheShard *srcc, BlueStore::Buffer *b) override | |
1167 | { | |
1168 | TwoQBufferCacheShard *src = static_cast<TwoQBufferCacheShard*>(srcc); | |
1169 | src->_rm(b); | |
1170 | ||
1171 | // preserve which list we're on (even if we can't preserve the order!) | |
7c673cae FG |
1172 | switch (b->cache_private) { |
1173 | case BUFFER_WARM_IN: | |
9f95a23c TL |
1174 | ceph_assert(!b->is_empty()); |
1175 | warm_in.push_back(*b); | |
7c673cae FG |
1176 | break; |
1177 | case BUFFER_WARM_OUT: | |
9f95a23c TL |
1178 | ceph_assert(b->is_empty()); |
1179 | warm_out.push_back(*b); | |
1180 | break; | |
7c673cae | 1181 | case BUFFER_HOT: |
9f95a23c TL |
1182 | ceph_assert(!b->is_empty()); |
1183 | hot.push_back(*b); | |
7c673cae FG |
1184 | break; |
1185 | default: | |
11fdf7f2 | 1186 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1187 | } |
9f95a23c TL |
1188 | if (!b->is_empty()) { |
1189 | buffer_bytes += b->length; | |
1190 | list_bytes[b->cache_private] += b->length; | |
1191 | } | |
1192 | num = hot.size() + warm_in.size(); | |
7c673cae | 1193 | } |
7c673cae | 1194 | |
9f95a23c TL |
1195 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override |
1196 | { | |
1197 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1198 | if (!b->is_empty()) { | |
1199 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1200 | buffer_bytes += delta; | |
1201 | ceph_assert((int64_t)list_bytes[b->cache_private] + delta >= 0); | |
1202 | list_bytes[b->cache_private] += delta; | |
1203 | } | |
7c673cae | 1204 | } |
7c673cae | 1205 | |
9f95a23c TL |
1206 | void _touch(BlueStore::Buffer *b) override { |
1207 | switch (b->cache_private) { | |
1208 | case BUFFER_WARM_IN: | |
1209 | // do nothing (somewhat counter-intuitively!) | |
1210 | break; | |
1211 | case BUFFER_WARM_OUT: | |
1212 | // move from warm_out to hot LRU | |
1213 | ceph_abort_msg("this happens via discard hint"); | |
1214 | break; | |
1215 | case BUFFER_HOT: | |
1216 | // move to front of hot LRU | |
1217 | hot.erase(hot.iterator_to(*b)); | |
1218 | hot.push_front(*b); | |
1219 | break; | |
1220 | } | |
1221 | num = hot.size() + warm_in.size(); | |
1222 | _audit("_touch_buffer end"); | |
7c673cae | 1223 | } |
7c673cae | 1224 | |
9f95a23c TL |
1225 | void _trim_to(uint64_t max) override |
1226 | { | |
1227 | if (buffer_bytes > max) { | |
1228 | uint64_t kin = max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1229 | uint64_t khot = max - kin; | |
1230 | ||
1231 | // pre-calculate kout based on average buffer size too, | |
1232 | // which is typical(the warm_in and hot lists may change later) | |
1233 | uint64_t kout = 0; | |
1234 | uint64_t buffer_num = hot.size() + warm_in.size(); | |
1235 | if (buffer_num) { | |
1236 | uint64_t avg_size = buffer_bytes / buffer_num; | |
1237 | ceph_assert(avg_size); | |
1238 | uint64_t calculated_num = max / avg_size; | |
1239 | kout = calculated_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1240 | } | |
1241 | ||
1242 | if (list_bytes[BUFFER_HOT] < khot) { | |
1243 | // hot is small, give slack to warm_in | |
1244 | kin += khot - list_bytes[BUFFER_HOT]; | |
1245 | } else if (list_bytes[BUFFER_WARM_IN] < kin) { | |
1246 | // warm_in is small, give slack to hot | |
1247 | khot += kin - list_bytes[BUFFER_WARM_IN]; | |
1248 | } | |
1249 | ||
1250 | // adjust warm_in list | |
1251 | int64_t to_evict_bytes = list_bytes[BUFFER_WARM_IN] - kin; | |
1252 | uint64_t evicted = 0; | |
1253 | ||
1254 | while (to_evict_bytes > 0) { | |
1255 | auto p = warm_in.rbegin(); | |
1256 | if (p == warm_in.rend()) { | |
1257 | // stop if warm_in list is now empty | |
1258 | break; | |
1259 | } | |
7c673cae | 1260 | |
9f95a23c TL |
1261 | BlueStore::Buffer *b = &*p; |
1262 | ceph_assert(b->is_clean()); | |
1263 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; | |
1264 | ceph_assert(buffer_bytes >= b->length); | |
1265 | buffer_bytes -= b->length; | |
1266 | ceph_assert(list_bytes[BUFFER_WARM_IN] >= b->length); | |
1267 | list_bytes[BUFFER_WARM_IN] -= b->length; | |
1268 | to_evict_bytes -= b->length; | |
1269 | evicted += b->length; | |
1270 | b->state = BlueStore::Buffer::STATE_EMPTY; | |
1271 | b->data.clear(); | |
1272 | warm_in.erase(warm_in.iterator_to(*b)); | |
1273 | warm_out.push_front(*b); | |
1274 | b->cache_private = BUFFER_WARM_OUT; | |
1275 | } | |
1276 | ||
1277 | if (evicted > 0) { | |
1278 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1279 | << " from warm_in list, done evicting warm_in buffers" | |
1280 | << dendl; | |
1281 | } | |
7c673cae | 1282 | |
9f95a23c TL |
1283 | // adjust hot list |
1284 | to_evict_bytes = list_bytes[BUFFER_HOT] - khot; | |
1285 | evicted = 0; | |
7c673cae | 1286 | |
9f95a23c TL |
1287 | while (to_evict_bytes > 0) { |
1288 | auto p = hot.rbegin(); | |
1289 | if (p == hot.rend()) { | |
1290 | // stop if hot list is now empty | |
1291 | break; | |
1292 | } | |
7c673cae | 1293 | |
9f95a23c TL |
1294 | BlueStore::Buffer *b = &*p; |
1295 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
1296 | ceph_assert(b->is_clean()); | |
1297 | // adjust evict size before buffer goes invalid | |
1298 | to_evict_bytes -= b->length; | |
1299 | evicted += b->length; | |
1300 | b->space->_rm_buffer(this, b); | |
1301 | } | |
7c673cae | 1302 | |
9f95a23c TL |
1303 | if (evicted > 0) { |
1304 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1305 | << " from hot list, done evicting hot buffers" | |
1306 | << dendl; | |
7c673cae FG |
1307 | } |
1308 | ||
9f95a23c TL |
1309 | // adjust warm out list too, if necessary |
1310 | int64_t n = warm_out.size() - kout; | |
1311 | while (n-- > 0) { | |
1312 | BlueStore::Buffer *b = &*warm_out.rbegin(); | |
1313 | ceph_assert(b->is_empty()); | |
1314 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; | |
1315 | b->space->_rm_buffer(this, b); | |
1316 | } | |
7c673cae | 1317 | } |
9f95a23c TL |
1318 | num = hot.size() + warm_in.size(); |
1319 | } | |
7c673cae | 1320 | |
9f95a23c TL |
1321 | void add_stats(uint64_t *extents, |
1322 | uint64_t *blobs, | |
1323 | uint64_t *buffers, | |
1324 | uint64_t *bytes) override { | |
1325 | *extents += num_extents; | |
1326 | *blobs += num_blobs; | |
1327 | *buffers += num; | |
1328 | *bytes += buffer_bytes; | |
1329 | } | |
7c673cae | 1330 | |
9f95a23c TL |
1331 | #ifdef DEBUG_CACHE |
1332 | void _audit(const char *s) override | |
1333 | { | |
1334 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1335 | uint64_t s = 0; | |
1336 | for (auto i = hot.begin(); i != hot.end(); ++i) { | |
1337 | s += i->length; | |
7c673cae FG |
1338 | } |
1339 | ||
9f95a23c TL |
1340 | uint64_t hot_bytes = s; |
1341 | if (hot_bytes != list_bytes[BUFFER_HOT]) { | |
1342 | derr << __func__ << " hot_list_bytes " | |
1343 | << list_bytes[BUFFER_HOT] | |
1344 | << " != actual " << hot_bytes | |
1345 | << dendl; | |
1346 | ceph_assert(hot_bytes == list_bytes[BUFFER_HOT]); | |
7c673cae FG |
1347 | } |
1348 | ||
9f95a23c TL |
1349 | for (auto i = warm_in.begin(); i != warm_in.end(); ++i) { |
1350 | s += i->length; | |
7c673cae | 1351 | } |
7c673cae | 1352 | |
9f95a23c TL |
1353 | uint64_t warm_in_bytes = s - hot_bytes; |
1354 | if (warm_in_bytes != list_bytes[BUFFER_WARM_IN]) { | |
1355 | derr << __func__ << " warm_in_list_bytes " | |
1356 | << list_bytes[BUFFER_WARM_IN] | |
1357 | << " != actual " << warm_in_bytes | |
1358 | << dendl; | |
1359 | ceph_assert(warm_in_bytes == list_bytes[BUFFER_WARM_IN]); | |
7c673cae | 1360 | } |
7c673cae | 1361 | |
9f95a23c TL |
1362 | if (s != buffer_bytes) { |
1363 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1364 | << dendl; | |
1365 | ceph_assert(s == buffer_bytes); | |
1366 | } | |
7c673cae | 1367 | |
9f95a23c TL |
1368 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1369 | << " ok" << dendl; | |
7c673cae | 1370 | } |
9f95a23c TL |
1371 | #endif |
1372 | }; | |
7c673cae | 1373 | |
9f95a23c | 1374 | // BuferCacheShard |
7c673cae | 1375 | |
9f95a23c TL |
1376 | BlueStore::BufferCacheShard *BlueStore::BufferCacheShard::create( |
1377 | CephContext* cct, | |
1378 | string type, | |
1379 | PerfCounters *logger) | |
1380 | { | |
1381 | BufferCacheShard *c = nullptr; | |
1382 | if (type == "lru") | |
1383 | c = new LruBufferCacheShard(cct); | |
1384 | else if (type == "2q") | |
1385 | c = new TwoQBufferCacheShard(cct); | |
1386 | else | |
1387 | ceph_abort_msg("unrecognized cache type"); | |
1388 | c->logger = logger; | |
1389 | return c; | |
7c673cae | 1390 | } |
7c673cae FG |
1391 | |
1392 | // BufferSpace | |
1393 | ||
1394 | #undef dout_prefix | |
1395 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1396 | ||
9f95a23c | 1397 | void BlueStore::BufferSpace::_clear(BufferCacheShard* cache) |
7c673cae FG |
1398 | { |
1399 | // note: we already hold cache->lock | |
1400 | ldout(cache->cct, 20) << __func__ << dendl; | |
1401 | while (!buffer_map.empty()) { | |
1402 | _rm_buffer(cache, buffer_map.begin()); | |
1403 | } | |
1404 | } | |
1405 | ||
9f95a23c | 1406 | int BlueStore::BufferSpace::_discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) |
7c673cae FG |
1407 | { |
1408 | // note: we already hold cache->lock | |
1409 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1410 | << std::dec << dendl; | |
1411 | int cache_private = 0; | |
1412 | cache->_audit("discard start"); | |
1413 | auto i = _data_lower_bound(offset); | |
1414 | uint32_t end = offset + length; | |
1415 | while (i != buffer_map.end()) { | |
1416 | Buffer *b = i->second.get(); | |
1417 | if (b->offset >= end) { | |
1418 | break; | |
1419 | } | |
1420 | if (b->cache_private > cache_private) { | |
1421 | cache_private = b->cache_private; | |
1422 | } | |
1423 | if (b->offset < offset) { | |
1424 | int64_t front = offset - b->offset; | |
1425 | if (b->end() > end) { | |
1426 | // drop middle (split) | |
1427 | uint32_t tail = b->end() - end; | |
1428 | if (b->data.length()) { | |
1429 | bufferlist bl; | |
1430 | bl.substr_of(b->data, b->length - tail, tail); | |
31f18b77 FG |
1431 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1432 | nb->maybe_rebuild(); | |
1433 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1434 | } else { |
31f18b77 FG |
1435 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail), |
1436 | 0, b); | |
7c673cae FG |
1437 | } |
1438 | if (!b->is_writing()) { | |
9f95a23c | 1439 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1440 | } |
1441 | b->truncate(front); | |
31f18b77 | 1442 | b->maybe_rebuild(); |
7c673cae FG |
1443 | cache->_audit("discard end 1"); |
1444 | break; | |
1445 | } else { | |
1446 | // drop tail | |
1447 | if (!b->is_writing()) { | |
9f95a23c | 1448 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1449 | } |
1450 | b->truncate(front); | |
31f18b77 | 1451 | b->maybe_rebuild(); |
7c673cae FG |
1452 | ++i; |
1453 | continue; | |
1454 | } | |
1455 | } | |
1456 | if (b->end() <= end) { | |
1457 | // drop entire buffer | |
1458 | _rm_buffer(cache, i++); | |
1459 | continue; | |
1460 | } | |
1461 | // drop front | |
1462 | uint32_t keep = b->end() - end; | |
1463 | if (b->data.length()) { | |
1464 | bufferlist bl; | |
1465 | bl.substr_of(b->data, b->length - keep, keep); | |
31f18b77 FG |
1466 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1467 | nb->maybe_rebuild(); | |
1468 | _add_buffer(cache, nb, 0, b); | |
7c673cae FG |
1469 | } else { |
1470 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep), 0, b); | |
1471 | } | |
1472 | _rm_buffer(cache, i); | |
1473 | cache->_audit("discard end 2"); | |
1474 | break; | |
1475 | } | |
1476 | return cache_private; | |
1477 | } | |
1478 | ||
1479 | void BlueStore::BufferSpace::read( | |
9f95a23c | 1480 | BufferCacheShard* cache, |
224ce89b WB |
1481 | uint32_t offset, |
1482 | uint32_t length, | |
7c673cae | 1483 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1484 | interval_set<uint32_t>& res_intervals, |
1485 | int flags) | |
7c673cae | 1486 | { |
7c673cae FG |
1487 | res.clear(); |
1488 | res_intervals.clear(); | |
1489 | uint32_t want_bytes = length; | |
1490 | uint32_t end = offset + length; | |
224ce89b WB |
1491 | |
1492 | { | |
11fdf7f2 | 1493 | std::lock_guard l(cache->lock); |
224ce89b WB |
1494 | for (auto i = _data_lower_bound(offset); |
1495 | i != buffer_map.end() && offset < end && i->first < end; | |
1496 | ++i) { | |
1497 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1498 | ceph_assert(b->end() > offset); |
91327a77 AA |
1499 | |
1500 | bool val = false; | |
1501 | if (flags & BYPASS_CLEAN_CACHE) | |
1502 | val = b->is_writing(); | |
1503 | else | |
1504 | val = b->is_writing() || b->is_clean(); | |
1505 | if (val) { | |
224ce89b WB |
1506 | if (b->offset < offset) { |
1507 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1508 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1509 | res[offset].substr_of(b->data, skip, l); |
1510 | res_intervals.insert(offset, l); | |
1511 | offset += l; | |
1512 | length -= l; | |
1513 | if (!b->is_writing()) { | |
9f95a23c | 1514 | cache->_touch(b); |
224ce89b WB |
1515 | } |
1516 | continue; | |
1517 | } | |
1518 | if (b->offset > offset) { | |
1519 | uint32_t gap = b->offset - offset; | |
1520 | if (length <= gap) { | |
1521 | break; | |
1522 | } | |
1523 | offset += gap; | |
1524 | length -= gap; | |
1525 | } | |
1526 | if (!b->is_writing()) { | |
9f95a23c | 1527 | cache->_touch(b); |
224ce89b WB |
1528 | } |
1529 | if (b->length > length) { | |
1530 | res[offset].substr_of(b->data, 0, length); | |
1531 | res_intervals.insert(offset, length); | |
7c673cae | 1532 | break; |
224ce89b WB |
1533 | } else { |
1534 | res[offset].append(b->data); | |
1535 | res_intervals.insert(offset, b->length); | |
1536 | if (b->length == length) | |
1537 | break; | |
1538 | offset += b->length; | |
1539 | length -= b->length; | |
1540 | } | |
7c673cae FG |
1541 | } |
1542 | } | |
1543 | } | |
1544 | ||
1545 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1546 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1547 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1548 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1549 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1550 | } | |
1551 | ||
9f95a23c | 1552 | void BlueStore::BufferSpace::_finish_write(BufferCacheShard* cache, uint64_t seq) |
7c673cae | 1553 | { |
7c673cae FG |
1554 | auto i = writing.begin(); |
1555 | while (i != writing.end()) { | |
1556 | if (i->seq > seq) { | |
1557 | break; | |
1558 | } | |
1559 | if (i->seq < seq) { | |
1560 | ++i; | |
1561 | continue; | |
1562 | } | |
1563 | ||
1564 | Buffer *b = &*i; | |
11fdf7f2 | 1565 | ceph_assert(b->is_writing()); |
7c673cae FG |
1566 | |
1567 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1568 | writing.erase(i++); | |
1569 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1570 | buffer_map.erase(b->offset); | |
1571 | } else { | |
1572 | b->state = Buffer::STATE_CLEAN; | |
1573 | writing.erase(i++); | |
31f18b77 FG |
1574 | b->maybe_rebuild(); |
1575 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
9f95a23c | 1576 | cache->_add(b, 1, nullptr); |
7c673cae FG |
1577 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; |
1578 | } | |
1579 | } | |
9f95a23c | 1580 | cache->_trim(); |
7c673cae FG |
1581 | cache->_audit("finish_write end"); |
1582 | } | |
1583 | ||
9f95a23c | 1584 | void BlueStore::BufferSpace::split(BufferCacheShard* cache, size_t pos, BlueStore::BufferSpace &r) |
7c673cae | 1585 | { |
11fdf7f2 | 1586 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1587 | if (buffer_map.empty()) |
1588 | return; | |
1589 | ||
1590 | auto p = --buffer_map.end(); | |
1591 | while (true) { | |
1592 | if (p->second->end() <= pos) | |
1593 | break; | |
1594 | ||
1595 | if (p->second->offset < pos) { | |
1596 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1597 | size_t left = pos - p->second->offset; | |
1598 | size_t right = p->second->length - left; | |
1599 | if (p->second->data.length()) { | |
1600 | bufferlist bl; | |
1601 | bl.substr_of(p->second->data, left, right); | |
1602 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, bl), | |
1603 | 0, p->second.get()); | |
1604 | } else { | |
1605 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, right), | |
1606 | 0, p->second.get()); | |
1607 | } | |
9f95a23c | 1608 | cache->_adjust_size(p->second.get(), -right); |
7c673cae FG |
1609 | p->second->truncate(left); |
1610 | break; | |
1611 | } | |
1612 | ||
11fdf7f2 | 1613 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1614 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1615 | if (p->second->data.length()) { | |
1616 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1617 | p->second->offset - pos, p->second->data), | |
1618 | 0, p->second.get()); | |
1619 | } else { | |
1620 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1621 | p->second->offset - pos, p->second->length), | |
1622 | 0, p->second.get()); | |
1623 | } | |
1624 | if (p == buffer_map.begin()) { | |
1625 | _rm_buffer(cache, p); | |
1626 | break; | |
1627 | } else { | |
1628 | _rm_buffer(cache, p--); | |
1629 | } | |
1630 | } | |
11fdf7f2 | 1631 | ceph_assert(writing.empty()); |
9f95a23c | 1632 | cache->_trim(); |
7c673cae FG |
1633 | } |
1634 | ||
1635 | // OnodeSpace | |
1636 | ||
1637 | #undef dout_prefix | |
1638 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1639 | ||
1640 | BlueStore::OnodeRef BlueStore::OnodeSpace::add(const ghobject_t& oid, OnodeRef o) | |
1641 | { | |
11fdf7f2 | 1642 | std::lock_guard l(cache->lock); |
7c673cae FG |
1643 | auto p = onode_map.find(oid); |
1644 | if (p != onode_map.end()) { | |
1645 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o | |
1646 | << " raced, returning existing " << p->second | |
1647 | << dendl; | |
1648 | return p->second; | |
1649 | } | |
1650 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o << dendl; | |
1651 | onode_map[oid] = o; | |
9f95a23c TL |
1652 | cache->_add(o, 1); |
1653 | cache->_trim(); | |
7c673cae FG |
1654 | return o; |
1655 | } | |
1656 | ||
1657 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) | |
1658 | { | |
7c673cae | 1659 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b WB |
1660 | OnodeRef o; |
1661 | bool hit = false; | |
1662 | ||
1663 | { | |
11fdf7f2 | 1664 | std::lock_guard l(cache->lock); |
224ce89b WB |
1665 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1666 | if (p == onode_map.end()) { | |
1667 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
1668 | } else { | |
1669 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
1670 | << dendl; | |
9f95a23c | 1671 | cache->_touch(p->second); |
224ce89b WB |
1672 | hit = true; |
1673 | o = p->second; | |
1674 | } | |
1675 | } | |
1676 | ||
1677 | if (hit) { | |
1678 | cache->logger->inc(l_bluestore_onode_hits); | |
1679 | } else { | |
7c673cae | 1680 | cache->logger->inc(l_bluestore_onode_misses); |
7c673cae | 1681 | } |
224ce89b | 1682 | return o; |
7c673cae FG |
1683 | } |
1684 | ||
1685 | void BlueStore::OnodeSpace::clear() | |
1686 | { | |
11fdf7f2 | 1687 | std::lock_guard l(cache->lock); |
7c673cae FG |
1688 | ldout(cache->cct, 10) << __func__ << dendl; |
1689 | for (auto &p : onode_map) { | |
9f95a23c | 1690 | cache->_rm(p.second); |
7c673cae FG |
1691 | } |
1692 | onode_map.clear(); | |
1693 | } | |
1694 | ||
1695 | bool BlueStore::OnodeSpace::empty() | |
1696 | { | |
11fdf7f2 | 1697 | std::lock_guard l(cache->lock); |
7c673cae FG |
1698 | return onode_map.empty(); |
1699 | } | |
1700 | ||
1701 | void BlueStore::OnodeSpace::rename( | |
1702 | OnodeRef& oldo, | |
1703 | const ghobject_t& old_oid, | |
1704 | const ghobject_t& new_oid, | |
31f18b77 | 1705 | const mempool::bluestore_cache_other::string& new_okey) |
7c673cae | 1706 | { |
11fdf7f2 | 1707 | std::lock_guard l(cache->lock); |
7c673cae FG |
1708 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
1709 | << dendl; | |
1710 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
1711 | po = onode_map.find(old_oid); | |
1712 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 1713 | ceph_assert(po != pn); |
7c673cae | 1714 | |
11fdf7f2 | 1715 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
1716 | if (pn != onode_map.end()) { |
1717 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
1718 | << dendl; | |
9f95a23c | 1719 | cache->_rm(pn->second); |
7c673cae FG |
1720 | onode_map.erase(pn); |
1721 | } | |
1722 | OnodeRef o = po->second; | |
1723 | ||
1724 | // install a non-existent onode at old location | |
1725 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
1726 | po->second = oldo; | |
9f95a23c | 1727 | cache->_add(po->second, 1); |
7c673cae FG |
1728 | // add at new position and fix oid, key |
1729 | onode_map.insert(make_pair(new_oid, o)); | |
9f95a23c | 1730 | cache->_touch(o); |
7c673cae FG |
1731 | o->oid = new_oid; |
1732 | o->key = new_okey; | |
9f95a23c | 1733 | cache->_trim(); |
7c673cae FG |
1734 | } |
1735 | ||
1736 | bool BlueStore::OnodeSpace::map_any(std::function<bool(OnodeRef)> f) | |
1737 | { | |
11fdf7f2 | 1738 | std::lock_guard l(cache->lock); |
7c673cae FG |
1739 | ldout(cache->cct, 20) << __func__ << dendl; |
1740 | for (auto& i : onode_map) { | |
1741 | if (f(i.second)) { | |
1742 | return true; | |
1743 | } | |
1744 | } | |
1745 | return false; | |
1746 | } | |
1747 | ||
11fdf7f2 TL |
1748 | template <int LogLevelV = 30> |
1749 | void BlueStore::OnodeSpace::dump(CephContext *cct) | |
3efd9988 FG |
1750 | { |
1751 | for (auto& i : onode_map) { | |
11fdf7f2 | 1752 | ldout(cct, LogLevelV) << i.first << " : " << i.second << dendl; |
3efd9988 FG |
1753 | } |
1754 | } | |
7c673cae FG |
1755 | |
1756 | // SharedBlob | |
1757 | ||
1758 | #undef dout_prefix | |
1759 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
9f95a23c TL |
1760 | #undef dout_context |
1761 | #define dout_context coll->store->cct | |
7c673cae | 1762 | |
9f95a23c | 1763 | void BlueStore::SharedBlob::dump(Formatter* f) const |
7c673cae | 1764 | { |
9f95a23c TL |
1765 | f->dump_bool("loaded", loaded); |
1766 | if (loaded) { | |
1767 | persistent->dump(f); | |
1768 | } else { | |
1769 | f->dump_unsigned("sbid_unloaded", sbid_unloaded); | |
1770 | } | |
1771 | } | |
1772 | ||
1773 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
1774 | { | |
1775 | out << "SharedBlob(" << &sb; | |
1776 | ||
7c673cae FG |
1777 | if (sb.loaded) { |
1778 | out << " loaded " << *sb.persistent; | |
1779 | } else { | |
1780 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
1781 | } | |
1782 | return out << ")"; | |
1783 | } | |
1784 | ||
1785 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
1786 | : coll(_coll), sbid_unloaded(i) | |
1787 | { | |
11fdf7f2 | 1788 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
1789 | if (get_cache()) { |
1790 | get_cache()->add_blob(); | |
1791 | } | |
1792 | } | |
1793 | ||
1794 | BlueStore::SharedBlob::~SharedBlob() | |
1795 | { | |
7c673cae FG |
1796 | if (loaded && persistent) { |
1797 | delete persistent; | |
1798 | } | |
1799 | } | |
1800 | ||
1801 | void BlueStore::SharedBlob::put() | |
1802 | { | |
1803 | if (--nref == 0) { | |
9f95a23c TL |
1804 | dout(20) << __func__ << " " << this |
1805 | << " removing self from set " << get_parent() | |
1806 | << dendl; | |
1adf2230 AA |
1807 | again: |
1808 | auto coll_snap = coll; | |
1809 | if (coll_snap) { | |
11fdf7f2 | 1810 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
1811 | if (coll_snap != coll) { |
1812 | goto again; | |
1813 | } | |
91327a77 AA |
1814 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
1815 | // race with lookup | |
1816 | return; | |
1817 | } | |
1adf2230 AA |
1818 | bc._clear(coll_snap->cache); |
1819 | coll_snap->cache->rm_blob(); | |
7c673cae | 1820 | } |
28e407b8 | 1821 | delete this; |
7c673cae FG |
1822 | } |
1823 | } | |
1824 | ||
1825 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
1826 | { | |
11fdf7f2 | 1827 | ceph_assert(persistent); |
7c673cae FG |
1828 | persistent->ref_map.get(offset, length); |
1829 | } | |
1830 | ||
1831 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 1832 | PExtentVector *r, |
11fdf7f2 | 1833 | bool *unshare) |
7c673cae | 1834 | { |
11fdf7f2 TL |
1835 | ceph_assert(persistent); |
1836 | persistent->ref_map.put(offset, length, r, | |
1837 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
1838 | } |
1839 | ||
f64942e4 AA |
1840 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
1841 | { | |
1842 | while (true) { | |
9f95a23c | 1843 | BufferCacheShard *cache = coll->cache; |
11fdf7f2 | 1844 | std::lock_guard l(cache->lock); |
f64942e4 | 1845 | if (coll->cache != cache) { |
9f95a23c TL |
1846 | dout(20) << __func__ |
1847 | << " raced with sb cache update, was " << cache | |
1848 | << ", now " << coll->cache << ", retrying" | |
1849 | << dendl; | |
f64942e4 AA |
1850 | continue; |
1851 | } | |
1852 | bc._finish_write(cache, seq); | |
1853 | break; | |
1854 | } | |
1855 | } | |
1856 | ||
3efd9988 FG |
1857 | // SharedBlobSet |
1858 | ||
1859 | #undef dout_prefix | |
1860 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
1861 | ||
11fdf7f2 TL |
1862 | template <int LogLevelV = 30> |
1863 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 1864 | { |
11fdf7f2 | 1865 | std::lock_guard l(lock); |
3efd9988 | 1866 | for (auto& i : sb_map) { |
11fdf7f2 | 1867 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
1868 | } |
1869 | } | |
1870 | ||
7c673cae FG |
1871 | // Blob |
1872 | ||
1873 | #undef dout_prefix | |
1874 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
1875 | ||
9f95a23c TL |
1876 | void BlueStore::Blob::dump(Formatter* f) const |
1877 | { | |
1878 | if (is_spanning()) { | |
1879 | f->dump_unsigned("spanning_id ", id); | |
1880 | } | |
1881 | blob.dump(f); | |
1882 | if (shared_blob) { | |
1883 | f->dump_object("shared", *shared_blob); | |
1884 | } | |
1885 | } | |
1886 | ||
7c673cae FG |
1887 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) |
1888 | { | |
1889 | out << "Blob(" << &b; | |
1890 | if (b.is_spanning()) { | |
1891 | out << " spanning " << b.id; | |
1892 | } | |
35e4c445 FG |
1893 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
1894 | if (b.shared_blob) { | |
1895 | out << " " << *b.shared_blob; | |
1896 | } else { | |
1897 | out << " (shared_blob=NULL)"; | |
1898 | } | |
1899 | out << ")"; | |
7c673cae FG |
1900 | return out; |
1901 | } | |
1902 | ||
1903 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
1904 | { | |
224ce89b | 1905 | if (get_blob().is_shared()) { |
7c673cae FG |
1906 | return; |
1907 | } | |
224ce89b | 1908 | if (get_blob().is_compressed()) { |
7c673cae FG |
1909 | bool discard = false; |
1910 | bool all_invalid = true; | |
224ce89b | 1911 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
1912 | if (!e.is_valid()) { |
1913 | discard = true; | |
1914 | } else { | |
1915 | all_invalid = false; | |
1916 | } | |
1917 | } | |
11fdf7f2 | 1918 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
1919 | // or none pextents are invalid. |
1920 | if (discard) { | |
224ce89b WB |
1921 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
1922 | get_blob().get_logical_length()); | |
7c673cae FG |
1923 | } |
1924 | } else { | |
1925 | size_t pos = 0; | |
224ce89b | 1926 | for (auto e : get_blob().get_extents()) { |
7c673cae | 1927 | if (!e.is_valid()) { |
9f95a23c TL |
1928 | dout(20) << __func__ << " 0x" << std::hex << pos |
1929 | << "~" << e.length | |
1930 | << std::dec << dendl; | |
7c673cae FG |
1931 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); |
1932 | } | |
1933 | pos += e.length; | |
1934 | } | |
224ce89b WB |
1935 | if (get_blob().can_prune_tail()) { |
1936 | dirty_blob().prune_tail(); | |
1937 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
224ce89b | 1938 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
1939 | } |
1940 | } | |
1941 | } | |
1942 | ||
1943 | void BlueStore::Blob::get_ref( | |
1944 | Collection *coll, | |
1945 | uint32_t offset, | |
1946 | uint32_t length) | |
1947 | { | |
1948 | // Caller has to initialize Blob's logical length prior to increment | |
1949 | // references. Otherwise one is neither unable to determine required | |
1950 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
1951 | // for single counter mode. | |
11fdf7f2 | 1952 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
1953 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
1954 | << std::dec << " " << *this << dendl; | |
1955 | ||
1956 | if (used_in_blob.is_empty()) { | |
1957 | uint32_t min_release_size = | |
224ce89b WB |
1958 | get_blob().get_release_size(coll->store->min_alloc_size); |
1959 | uint64_t l = get_blob().get_logical_length(); | |
1960 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
1961 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
1962 | used_in_blob.init(l, min_release_size); |
1963 | } | |
1964 | used_in_blob.get( | |
1965 | offset, | |
1966 | length); | |
1967 | } | |
1968 | ||
1969 | bool BlueStore::Blob::put_ref( | |
1970 | Collection *coll, | |
1971 | uint32_t offset, | |
1972 | uint32_t length, | |
1973 | PExtentVector *r) | |
1974 | { | |
1975 | PExtentVector logical; | |
1976 | ||
7c673cae FG |
1977 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
1978 | << std::dec << " " << *this << dendl; | |
1979 | ||
1980 | bool empty = used_in_blob.put( | |
1981 | offset, | |
1982 | length, | |
1983 | &logical); | |
1984 | r->clear(); | |
1985 | // nothing to release | |
1986 | if (!empty && logical.empty()) { | |
1987 | return false; | |
1988 | } | |
1989 | ||
1990 | bluestore_blob_t& b = dirty_blob(); | |
1991 | return b.release_extents(empty, logical, r); | |
1992 | } | |
1993 | ||
224ce89b | 1994 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
1995 | uint32_t target_blob_size, |
1996 | uint32_t b_offset, | |
1997 | uint32_t *length0) { | |
11fdf7f2 TL |
1998 | ceph_assert(min_alloc_size); |
1999 | ceph_assert(target_blob_size); | |
7c673cae FG |
2000 | if (!get_blob().is_mutable()) { |
2001 | return false; | |
2002 | } | |
2003 | ||
2004 | uint32_t length = *length0; | |
2005 | uint32_t end = b_offset + length; | |
2006 | ||
2007 | // Currently for the sake of simplicity we omit blob reuse if data is | |
2008 | // unaligned with csum chunk. Later we can perform padding if needed. | |
2009 | if (get_blob().has_csum() && | |
2010 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
2011 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
2012 | return false; | |
2013 | } | |
2014 | ||
2015 | auto blen = get_blob().get_logical_length(); | |
2016 | uint32_t new_blen = blen; | |
2017 | ||
2018 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 2019 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
2020 | |
2021 | if (b_offset >= blen) { | |
224ce89b WB |
2022 | // new data totally stands out of the existing blob |
2023 | new_blen = end; | |
7c673cae | 2024 | } else { |
224ce89b | 2025 | // new data overlaps with the existing blob |
11fdf7f2 | 2026 | new_blen = std::max(blen, end); |
224ce89b WB |
2027 | |
2028 | uint32_t overlap = 0; | |
2029 | if (new_blen > blen) { | |
2030 | overlap = blen - b_offset; | |
2031 | } else { | |
2032 | overlap = length; | |
2033 | } | |
2034 | ||
2035 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
2036 | // abort if any piece of the overlap has already been allocated | |
2037 | return false; | |
7c673cae FG |
2038 | } |
2039 | } | |
224ce89b | 2040 | |
7c673cae FG |
2041 | if (new_blen > blen) { |
2042 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
2043 | // Unable to decrease the provided length to fit into max_blob_size | |
2044 | if (overflow >= length) { | |
2045 | return false; | |
2046 | } | |
2047 | ||
2048 | // FIXME: in some cases we could reduce unused resolution | |
2049 | if (get_blob().has_unused()) { | |
2050 | return false; | |
2051 | } | |
2052 | ||
2053 | if (overflow > 0) { | |
2054 | new_blen -= overflow; | |
2055 | length -= overflow; | |
2056 | *length0 = length; | |
2057 | } | |
224ce89b | 2058 | |
7c673cae FG |
2059 | if (new_blen > blen) { |
2060 | dirty_blob().add_tail(new_blen); | |
2061 | used_in_blob.add_tail(new_blen, | |
224ce89b | 2062 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
2063 | } |
2064 | } | |
2065 | return true; | |
2066 | } | |
2067 | ||
2068 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
2069 | { | |
7c673cae FG |
2070 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec |
2071 | << " start " << *this << dendl; | |
11fdf7f2 TL |
2072 | ceph_assert(blob.can_split()); |
2073 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
2074 | bluestore_blob_t &lb = dirty_blob(); |
2075 | bluestore_blob_t &rb = r->dirty_blob(); | |
2076 | ||
2077 | used_in_blob.split( | |
2078 | blob_offset, | |
2079 | &(r->used_in_blob)); | |
2080 | ||
2081 | lb.split(blob_offset, rb); | |
2082 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2083 | ||
2084 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2085 | << " finish " << *this << dendl; | |
2086 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2087 | << " and " << *r << dendl; | |
2088 | } | |
2089 | ||
2090 | #ifndef CACHE_BLOB_BL | |
2091 | void BlueStore::Blob::decode( | |
2092 | Collection *coll, | |
11fdf7f2 | 2093 | bufferptr::const_iterator& p, |
7c673cae FG |
2094 | uint64_t struct_v, |
2095 | uint64_t* sbid, | |
2096 | bool include_ref_map) | |
2097 | { | |
2098 | denc(blob, p, struct_v); | |
2099 | if (blob.is_shared()) { | |
2100 | denc(*sbid, p); | |
2101 | } | |
2102 | if (include_ref_map) { | |
2103 | if (struct_v > 1) { | |
2104 | used_in_blob.decode(p); | |
2105 | } else { | |
2106 | used_in_blob.clear(); | |
2107 | bluestore_extent_ref_map_t legacy_ref_map; | |
2108 | legacy_ref_map.decode(p); | |
2109 | for (auto r : legacy_ref_map.ref_map) { | |
2110 | get_ref( | |
2111 | coll, | |
2112 | r.first, | |
2113 | r.second.refs * r.second.length); | |
2114 | } | |
2115 | } | |
2116 | } | |
2117 | } | |
2118 | #endif | |
2119 | ||
2120 | // Extent | |
2121 | ||
9f95a23c TL |
2122 | void BlueStore::Extent::dump(Formatter* f) const |
2123 | { | |
2124 | f->dump_unsigned("logical_offset", logical_offset); | |
2125 | f->dump_unsigned("length", length); | |
2126 | f->dump_unsigned("blob_offset", blob_offset); | |
2127 | f->dump_object("blob", *blob); | |
2128 | } | |
2129 | ||
7c673cae FG |
2130 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) |
2131 | { | |
2132 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2133 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2134 | << " " << *e.blob; | |
2135 | } | |
2136 | ||
2137 | // OldExtent | |
2138 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2139 | uint32_t lo, | |
2140 | uint32_t o, | |
2141 | uint32_t l, | |
2142 | BlobRef& b) { | |
2143 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2144 | b->put_ref(c.get(), o, l, &(oe->r)); | |
2145 | oe->blob_empty = b->get_referenced_bytes() == 0; | |
2146 | return oe; | |
2147 | } | |
2148 | ||
2149 | // ExtentMap | |
2150 | ||
2151 | #undef dout_prefix | |
2152 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
9f95a23c TL |
2153 | #undef dout_context |
2154 | #define dout_context onode->c->store->cct | |
7c673cae FG |
2155 | |
2156 | BlueStore::ExtentMap::ExtentMap(Onode *o) | |
2157 | : onode(o), | |
2158 | inline_bl( | |
2159 | o->c->store->cct->_conf->bluestore_extent_map_inline_shard_prealloc_size) { | |
2160 | } | |
2161 | ||
9f95a23c TL |
2162 | void BlueStore::ExtentMap::dump(Formatter* f) const |
2163 | { | |
2164 | f->open_array_section("extents"); | |
2165 | ||
2166 | for (auto& e : extent_map) { | |
2167 | f->dump_object("extent", e); | |
2168 | } | |
2169 | f->close_section(); | |
2170 | } | |
2171 | ||
11fdf7f2 TL |
2172 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2173 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2174 | uint64_t& length, uint64_t& dstoff) { | |
2175 | ||
2176 | auto cct = onode->c->store->cct; | |
2177 | bool inject_21040 = | |
2178 | cct->_conf->bluestore_debug_inject_bug21040; | |
2179 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2180 | for (auto& e : oldo->extent_map.extent_map) { | |
2181 | e.blob->last_encoded_id = -1; | |
2182 | } | |
2183 | ||
2184 | int n = 0; | |
2185 | uint64_t end = srcoff + length; | |
2186 | uint32_t dirty_range_begin = 0; | |
2187 | uint32_t dirty_range_end = 0; | |
2188 | bool src_dirty = false; | |
2189 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2190 | ep != oldo->extent_map.extent_map.end(); | |
2191 | ++ep) { | |
2192 | auto& e = *ep; | |
2193 | if (e.logical_offset >= end) { | |
2194 | break; | |
2195 | } | |
2196 | dout(20) << __func__ << " src " << e << dendl; | |
2197 | BlobRef cb; | |
2198 | bool blob_duped = true; | |
2199 | if (e.blob->last_encoded_id >= 0) { | |
2200 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2201 | blob_duped = false; | |
2202 | } else { | |
2203 | // dup the blob | |
2204 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2205 | // make sure it is shared | |
2206 | if (!blob.is_shared()) { | |
2207 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2208 | if (!inject_21040 && !src_dirty) { | |
2209 | src_dirty = true; | |
2210 | dirty_range_begin = e.logical_offset; | |
2211 | } else if (inject_21040 && | |
2212 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2213 | dirty_range_begin = e.logical_offset; | |
2214 | } | |
2215 | ceph_assert(e.logical_end() > 0); | |
2216 | // -1 to exclude next potential shard | |
2217 | dirty_range_end = e.logical_end() - 1; | |
2218 | } else { | |
2219 | c->load_shared_blob(e.blob->shared_blob); | |
2220 | } | |
2221 | cb = new Blob(); | |
2222 | e.blob->last_encoded_id = n; | |
2223 | id_to_blob[n] = cb; | |
2224 | e.blob->dup(*cb); | |
2225 | // bump the extent refs on the copied blob's extents | |
2226 | for (auto p : blob.get_extents()) { | |
2227 | if (p.is_valid()) { | |
2228 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2229 | } | |
2230 | } | |
2231 | txc->write_shared_blob(e.blob->shared_blob); | |
2232 | dout(20) << __func__ << " new " << *cb << dendl; | |
2233 | } | |
2234 | ||
2235 | int skip_front, skip_back; | |
2236 | if (e.logical_offset < srcoff) { | |
2237 | skip_front = srcoff - e.logical_offset; | |
2238 | } else { | |
2239 | skip_front = 0; | |
2240 | } | |
2241 | if (e.logical_end() > end) { | |
2242 | skip_back = e.logical_end() - end; | |
2243 | } else { | |
2244 | skip_back = 0; | |
2245 | } | |
2246 | ||
2247 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2248 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2249 | newo->extent_map.extent_map.insert(*ne); | |
2250 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2251 | // fixme: we may leave parts of new blob unreferenced that could | |
2252 | // be freed (relative to the shared_blob). | |
2253 | txc->statfs_delta.stored() += ne->length; | |
2254 | if (e.blob->get_blob().is_compressed()) { | |
2255 | txc->statfs_delta.compressed_original() += ne->length; | |
2256 | if (blob_duped) { | |
2257 | txc->statfs_delta.compressed() += | |
2258 | cb->get_blob().get_compressed_payload_length(); | |
2259 | } | |
2260 | } | |
2261 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2262 | ++n; | |
2263 | } | |
2264 | if ((!inject_21040 && src_dirty) || | |
2265 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2266 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2267 | dirty_range_end - dirty_range_begin); | |
2268 | txc->write_onode(oldo); | |
2269 | } | |
2270 | txc->write_onode(newo); | |
2271 | ||
2272 | if (dstoff + length > newo->onode.size) { | |
2273 | newo->onode.size = dstoff + length; | |
2274 | } | |
2275 | newo->extent_map.dirty_range(dstoff, length); | |
2276 | } | |
7c673cae FG |
2277 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2278 | bool force) | |
2279 | { | |
2280 | auto cct = onode->c->store->cct; //used by dout | |
2281 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2282 | if (onode->onode.extent_map_shards.empty()) { | |
2283 | if (inline_bl.length() == 0) { | |
2284 | unsigned n; | |
2285 | // we need to encode inline_bl to measure encoded length | |
2286 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
3efd9988 | 2287 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_cache_other); |
11fdf7f2 | 2288 | ceph_assert(!never_happen); |
7c673cae FG |
2289 | size_t len = inline_bl.length(); |
2290 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2291 | << " extents" << dendl; | |
2292 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2293 | request_reshard(0, OBJECT_MAX_SIZE); | |
2294 | return; | |
2295 | } | |
2296 | } | |
2297 | // will persist in the onode key. | |
2298 | } else { | |
2299 | // pending shard update | |
2300 | struct dirty_shard_t { | |
2301 | Shard *shard; | |
2302 | bufferlist bl; | |
2303 | dirty_shard_t(Shard *s) : shard(s) {} | |
2304 | }; | |
2305 | vector<dirty_shard_t> encoded_shards; | |
2306 | // allocate slots for all shards in a single call instead of | |
2307 | // doing multiple allocations - one per each dirty shard | |
2308 | encoded_shards.reserve(shards.size()); | |
2309 | ||
2310 | auto p = shards.begin(); | |
2311 | auto prev_p = p; | |
2312 | while (p != shards.end()) { | |
11fdf7f2 | 2313 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2314 | auto n = p; |
2315 | ++n; | |
2316 | if (p->dirty) { | |
2317 | uint32_t endoff; | |
2318 | if (n == shards.end()) { | |
2319 | endoff = OBJECT_MAX_SIZE; | |
2320 | } else { | |
2321 | endoff = n->shard_info->offset; | |
2322 | } | |
2323 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2324 | bufferlist& bl = encoded_shards.back().bl; | |
2325 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2326 | bl, &p->extents)) { | |
2327 | if (force) { | |
2328 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2329 | ceph_assert(!force); |
7c673cae FG |
2330 | } |
2331 | } | |
2332 | size_t len = bl.length(); | |
2333 | ||
2334 | dout(20) << __func__ << " shard 0x" << std::hex | |
2335 | << p->shard_info->offset << std::dec << " is " << len | |
2336 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2337 | << p->extents << " extents" << dendl; | |
2338 | ||
2339 | if (!force) { | |
2340 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2341 | // we are big; reshard ourselves | |
2342 | request_reshard(p->shard_info->offset, endoff); | |
2343 | } | |
2344 | // avoid resharding the trailing shard, even if it is small | |
2345 | else if (n != shards.end() && | |
11fdf7f2 TL |
2346 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2347 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2348 | if (p == shards.begin()) { |
2349 | // we are the first shard, combine with next shard | |
7c673cae | 2350 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2351 | } else { |
31f18b77 FG |
2352 | // combine either with the previous shard or the next, |
2353 | // whichever is smaller | |
7c673cae FG |
2354 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2355 | request_reshard(p->shard_info->offset, endoff + 1); | |
2356 | } else { | |
2357 | request_reshard(prev_p->shard_info->offset, endoff); | |
2358 | } | |
2359 | } | |
2360 | } | |
2361 | } | |
2362 | } | |
2363 | prev_p = p; | |
2364 | p = n; | |
2365 | } | |
2366 | if (needs_reshard()) { | |
2367 | return; | |
2368 | } | |
2369 | ||
2370 | // schedule DB update for dirty shards | |
2371 | string key; | |
2372 | for (auto& it : encoded_shards) { | |
2373 | it.shard->dirty = false; | |
2374 | it.shard->shard_info->bytes = it.bl.length(); | |
2375 | generate_extent_shard_key_and_apply( | |
2376 | onode->key, | |
2377 | it.shard->shard_info->offset, | |
2378 | &key, | |
2379 | [&](const string& final_key) { | |
2380 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2381 | } | |
2382 | ); | |
2383 | } | |
2384 | } | |
2385 | } | |
2386 | ||
31f18b77 FG |
2387 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2388 | { | |
2389 | if (spanning_blob_map.empty()) | |
2390 | return 0; | |
2391 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2392 | // bid is valid and available. | |
2393 | if (bid >= 0) | |
2394 | return bid; | |
2395 | // Find next unused bid; | |
2396 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2397 | const auto begin_bid = bid; | |
2398 | do { | |
2399 | if (!spanning_blob_map.count(bid)) | |
2400 | return bid; | |
2401 | else { | |
2402 | bid++; | |
2403 | if (bid < 0) bid = 0; | |
2404 | } | |
2405 | } while (bid != begin_bid); | |
81eedcae TL |
2406 | auto cct = onode->c->store->cct; // used by dout |
2407 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2408 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2409 | } |
2410 | ||
7c673cae FG |
2411 | void BlueStore::ExtentMap::reshard( |
2412 | KeyValueDB *db, | |
2413 | KeyValueDB::Transaction t) | |
2414 | { | |
2415 | auto cct = onode->c->store->cct; // used by dout | |
2416 | ||
2417 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2418 | << needs_reshard_end << ")" << std::dec | |
2419 | << " of " << onode->onode.extent_map_shards.size() | |
2420 | << " shards on " << onode->oid << dendl; | |
2421 | for (auto& p : spanning_blob_map) { | |
2422 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2423 | << dendl; | |
2424 | } | |
2425 | // determine shard index range | |
2426 | unsigned si_begin = 0, si_end = 0; | |
2427 | if (!shards.empty()) { | |
2428 | while (si_begin + 1 < shards.size() && | |
2429 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2430 | ++si_begin; | |
2431 | } | |
2432 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2433 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2434 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2435 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2436 | break; | |
2437 | } | |
2438 | } | |
2439 | if (si_end == shards.size()) { | |
2440 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2441 | } | |
2442 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2443 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2444 | << needs_reshard_end << ")" << std::dec << dendl; | |
2445 | } | |
2446 | ||
181888fb | 2447 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2448 | |
2449 | // we may need to fault in a larger interval later must have all | |
2450 | // referring extents for spanning blobs loaded in order to have | |
2451 | // accurate use_tracker values. | |
2452 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2453 | uint32_t spanning_scan_end = needs_reshard_end; | |
2454 | ||
2455 | // remove old keys | |
2456 | string key; | |
2457 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2458 | generate_extent_shard_key_and_apply( | |
2459 | onode->key, shards[i].shard_info->offset, &key, | |
2460 | [&](const string& final_key) { | |
2461 | t->rmkey(PREFIX_OBJ, final_key); | |
2462 | } | |
2463 | ); | |
2464 | } | |
2465 | ||
2466 | // calculate average extent size | |
2467 | unsigned bytes = 0; | |
2468 | unsigned extents = 0; | |
2469 | if (onode->onode.extent_map_shards.empty()) { | |
2470 | bytes = inline_bl.length(); | |
2471 | extents = extent_map.size(); | |
2472 | } else { | |
2473 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2474 | bytes += shards[i].shard_info->bytes; | |
2475 | extents += shards[i].extents; | |
2476 | } | |
2477 | } | |
2478 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2479 | unsigned slop = target * | |
2480 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2481 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2482 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2483 | << ", slop " << slop << dendl; | |
2484 | ||
2485 | // reshard | |
2486 | unsigned estimate = 0; | |
31f18b77 | 2487 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2488 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2489 | unsigned max_blob_end = 0; | |
2490 | Extent dummy(needs_reshard_begin); | |
2491 | for (auto e = extent_map.lower_bound(dummy); | |
2492 | e != extent_map.end(); | |
2493 | ++e) { | |
2494 | if (e->logical_offset >= needs_reshard_end) { | |
2495 | break; | |
2496 | } | |
2497 | dout(30) << " extent " << *e << dendl; | |
2498 | ||
2499 | // disfavor shard boundaries that span a blob | |
2500 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2501 | if (estimate && | |
2502 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2503 | // new shard | |
31f18b77 | 2504 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2505 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2506 | new_shard_info.back().offset = offset; | |
2507 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2508 | << std::dec << dendl; |
7c673cae FG |
2509 | } |
2510 | offset = e->logical_offset; | |
2511 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2512 | new_shard_info.back().offset = offset; | |
2513 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2514 | << std::dec << dendl; | |
2515 | estimate = 0; | |
2516 | } | |
2517 | estimate += extent_avg; | |
31f18b77 FG |
2518 | unsigned bs = e->blob_start(); |
2519 | if (bs < spanning_scan_begin) { | |
2520 | spanning_scan_begin = bs; | |
7c673cae FG |
2521 | } |
2522 | uint32_t be = e->blob_end(); | |
2523 | if (be > max_blob_end) { | |
2524 | max_blob_end = be; | |
2525 | } | |
2526 | if (be > spanning_scan_end) { | |
2527 | spanning_scan_end = be; | |
2528 | } | |
2529 | } | |
2530 | if (new_shard_info.empty() && (si_begin > 0 || | |
2531 | si_end < shards.size())) { | |
2532 | // we resharded a partial range; we must produce at least one output | |
2533 | // shard | |
2534 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2535 | new_shard_info.back().offset = needs_reshard_begin; | |
2536 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2537 | << std::dec << " (singleton degenerate case)" << dendl; | |
2538 | } | |
2539 | ||
2540 | auto& sv = onode->onode.extent_map_shards; | |
2541 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2542 | dout(20) << __func__ << " old " << sv << dendl; | |
2543 | if (sv.empty()) { | |
2544 | // no old shards to keep | |
2545 | sv.swap(new_shard_info); | |
2546 | init_shards(true, true); | |
2547 | } else { | |
2548 | // splice in new shards | |
2549 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2550 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2551 | sv.insert( | |
2552 | sv.begin() + si_begin, | |
2553 | new_shard_info.begin(), | |
2554 | new_shard_info.end()); | |
2555 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2556 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2557 | |
11fdf7f2 | 2558 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2559 | |
2560 | // note that we need to update every shard_info of shards here, | |
2561 | // as sv might have been totally re-allocated above | |
2562 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2563 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2564 | } |
2565 | ||
2566 | // mark newly added shards as dirty | |
2567 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2568 | shards[i].loaded = true; |
2569 | shards[i].dirty = true; | |
2570 | } | |
7c673cae FG |
2571 | } |
2572 | dout(20) << __func__ << " fin " << sv << dendl; | |
2573 | inline_bl.clear(); | |
2574 | ||
2575 | if (sv.empty()) { | |
2576 | // no more shards; unspan all previously spanning blobs | |
2577 | auto p = spanning_blob_map.begin(); | |
2578 | while (p != spanning_blob_map.end()) { | |
2579 | p->second->id = -1; | |
2580 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2581 | p = spanning_blob_map.erase(p); | |
2582 | } | |
2583 | } else { | |
2584 | // identify new spanning blobs | |
2585 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2586 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2587 | if (spanning_scan_begin < needs_reshard_begin) { | |
2588 | fault_range(db, spanning_scan_begin, | |
2589 | needs_reshard_begin - spanning_scan_begin); | |
2590 | } | |
2591 | if (spanning_scan_end > needs_reshard_end) { | |
2592 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2593 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2594 | } |
2595 | auto sp = sv.begin() + si_begin; | |
2596 | auto esp = sv.end(); | |
2597 | unsigned shard_start = sp->offset; | |
2598 | unsigned shard_end; | |
2599 | ++sp; | |
2600 | if (sp == esp) { | |
2601 | shard_end = OBJECT_MAX_SIZE; | |
2602 | } else { | |
2603 | shard_end = sp->offset; | |
2604 | } | |
7c673cae | 2605 | Extent dummy(needs_reshard_begin); |
9f95a23c TL |
2606 | |
2607 | bool was_too_many_blobs_check = false; | |
2608 | auto too_many_blobs_threshold = | |
2609 | g_conf()->bluestore_debug_too_many_blobs_threshold; | |
2610 | auto& dumped_onodes = onode->c->onode_map.cache->dumped_onodes; | |
2611 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oid_slot = nullptr; | |
2612 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oldest_slot = nullptr; | |
2613 | ||
7c673cae FG |
2614 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { |
2615 | if (e->logical_offset >= needs_reshard_end) { | |
2616 | break; | |
2617 | } | |
2618 | dout(30) << " extent " << *e << dendl; | |
2619 | while (e->logical_offset >= shard_end) { | |
2620 | shard_start = shard_end; | |
11fdf7f2 | 2621 | ceph_assert(sp != esp); |
7c673cae FG |
2622 | ++sp; |
2623 | if (sp == esp) { | |
2624 | shard_end = OBJECT_MAX_SIZE; | |
2625 | } else { | |
2626 | shard_end = sp->offset; | |
2627 | } | |
2628 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2629 | << " to 0x" << shard_end << std::dec << dendl; | |
2630 | } | |
9f95a23c | 2631 | |
7c673cae FG |
2632 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { |
2633 | if (!e->blob->is_spanning()) { | |
2634 | // We have two options: (1) split the blob into pieces at the | |
2635 | // shard boundaries (and adjust extents accordingly), or (2) | |
2636 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2637 | // we may have to split it multiple times--potentially at every | |
2638 | // shard boundary. | |
2639 | bool must_span = false; | |
2640 | BlobRef b = e->blob; | |
2641 | if (b->can_split()) { | |
2642 | uint32_t bstart = e->blob_start(); | |
2643 | uint32_t bend = e->blob_end(); | |
2644 | for (const auto& sh : shards) { | |
2645 | if (bstart < sh.shard_info->offset && | |
2646 | bend > sh.shard_info->offset) { | |
2647 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2648 | if (b->can_split_at(blob_offset)) { | |
2649 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2650 | << std::hex << bstart << " blob_offset 0x" | |
2651 | << blob_offset << std::dec << " " << *b << dendl; | |
2652 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2653 | // switch b to the new right-hand side, in case it | |
2654 | // *also* has to get split. | |
2655 | bstart += blob_offset; | |
2656 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2657 | } else { | |
2658 | must_span = true; | |
2659 | break; | |
2660 | } | |
2661 | } | |
2662 | } | |
2663 | } else { | |
2664 | must_span = true; | |
2665 | } | |
2666 | if (must_span) { | |
31f18b77 FG |
2667 | auto bid = allocate_spanning_blob_id(); |
2668 | b->id = bid; | |
7c673cae FG |
2669 | spanning_blob_map[b->id] = b; |
2670 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
9f95a23c TL |
2671 | if (!was_too_many_blobs_check && |
2672 | too_many_blobs_threshold && | |
2673 | spanning_blob_map.size() >= size_t(too_many_blobs_threshold)) { | |
2674 | ||
2675 | was_too_many_blobs_check = true; | |
2676 | for (size_t i = 0; i < dumped_onodes.size(); ++i) { | |
2677 | if (dumped_onodes[i].first == onode->oid) { | |
2678 | oid_slot = &dumped_onodes[i]; | |
2679 | break; | |
2680 | } | |
2681 | if (!oldest_slot || (oldest_slot && | |
2682 | dumped_onodes[i].second < oldest_slot->second)) { | |
2683 | oldest_slot = &dumped_onodes[i]; | |
2684 | } | |
2685 | } | |
2686 | } | |
7c673cae FG |
2687 | } |
2688 | } | |
2689 | } else { | |
2690 | if (e->blob->is_spanning()) { | |
2691 | spanning_blob_map.erase(e->blob->id); | |
2692 | e->blob->id = -1; | |
2693 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
2694 | } | |
2695 | } | |
2696 | } | |
9f95a23c TL |
2697 | bool do_dump = (!oid_slot && was_too_many_blobs_check) || |
2698 | (oid_slot && | |
2699 | (mono_clock::now() - oid_slot->second >= make_timespan(5 * 60))); | |
2700 | if (do_dump) { | |
2701 | dout(0) << __func__ | |
2702 | << " spanning blob count exceeds threshold, " | |
2703 | << spanning_blob_map.size() << " spanning blobs" | |
2704 | << dendl; | |
2705 | _dump_onode<0>(cct, *onode); | |
2706 | if (oid_slot) { | |
2707 | oid_slot->second = mono_clock::now(); | |
2708 | } else { | |
2709 | ceph_assert(oldest_slot); | |
2710 | oldest_slot->first = onode->oid; | |
2711 | oldest_slot->second = mono_clock::now(); | |
2712 | } | |
2713 | } | |
7c673cae FG |
2714 | } |
2715 | ||
2716 | clear_needs_reshard(); | |
2717 | } | |
2718 | ||
2719 | bool BlueStore::ExtentMap::encode_some( | |
2720 | uint32_t offset, | |
2721 | uint32_t length, | |
2722 | bufferlist& bl, | |
2723 | unsigned *pn) | |
2724 | { | |
7c673cae FG |
2725 | Extent dummy(offset); |
2726 | auto start = extent_map.lower_bound(dummy); | |
2727 | uint32_t end = offset + length; | |
2728 | ||
2729 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
2730 | // serialization only. Hence there is no specific | |
2731 | // handling at ExtentMap level. | |
2732 | ||
2733 | unsigned n = 0; | |
2734 | size_t bound = 0; | |
7c673cae FG |
2735 | bool must_reshard = false; |
2736 | for (auto p = start; | |
2737 | p != extent_map.end() && p->logical_offset < end; | |
2738 | ++p, ++n) { | |
11fdf7f2 | 2739 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
2740 | p->blob->last_encoded_id = -1; |
2741 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
2742 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2743 | << std::dec << " hit new spanning blob " << *p << dendl; | |
2744 | request_reshard(p->blob_start(), p->blob_end()); | |
2745 | must_reshard = true; | |
2746 | } | |
31f18b77 FG |
2747 | if (!must_reshard) { |
2748 | denc_varint(0, bound); // blobid | |
2749 | denc_varint(0, bound); // logical_offset | |
2750 | denc_varint(0, bound); // len | |
2751 | denc_varint(0, bound); // blob_offset | |
7c673cae | 2752 | |
31f18b77 FG |
2753 | p->blob->bound_encode( |
2754 | bound, | |
2755 | struct_v, | |
2756 | p->blob->shared_blob->get_sbid(), | |
2757 | false); | |
2758 | } | |
7c673cae FG |
2759 | } |
2760 | if (must_reshard) { | |
2761 | return true; | |
2762 | } | |
2763 | ||
31f18b77 FG |
2764 | denc(struct_v, bound); |
2765 | denc_varint(0, bound); // number of extents | |
2766 | ||
7c673cae FG |
2767 | { |
2768 | auto app = bl.get_contiguous_appender(bound); | |
2769 | denc(struct_v, app); | |
2770 | denc_varint(n, app); | |
2771 | if (pn) { | |
2772 | *pn = n; | |
2773 | } | |
2774 | ||
2775 | n = 0; | |
2776 | uint64_t pos = 0; | |
2777 | uint64_t prev_len = 0; | |
2778 | for (auto p = start; | |
2779 | p != extent_map.end() && p->logical_offset < end; | |
2780 | ++p, ++n) { | |
2781 | unsigned blobid; | |
2782 | bool include_blob = false; | |
2783 | if (p->blob->is_spanning()) { | |
2784 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
2785 | blobid |= BLOBID_FLAG_SPANNING; | |
2786 | } else if (p->blob->last_encoded_id < 0) { | |
2787 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
2788 | include_blob = true; | |
2789 | blobid = 0; // the decoder will infer the id from n | |
2790 | } else { | |
2791 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
2792 | } | |
2793 | if (p->logical_offset == pos) { | |
2794 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
2795 | } | |
2796 | if (p->blob_offset == 0) { | |
2797 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
2798 | } | |
2799 | if (p->length == prev_len) { | |
2800 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
2801 | } else { | |
2802 | prev_len = p->length; | |
2803 | } | |
2804 | denc_varint(blobid, app); | |
2805 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
2806 | denc_varint_lowz(p->logical_offset - pos, app); | |
2807 | } | |
2808 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
2809 | denc_varint_lowz(p->blob_offset, app); | |
2810 | } | |
2811 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
2812 | denc_varint_lowz(p->length, app); | |
2813 | } | |
2814 | pos = p->logical_end(); | |
2815 | if (include_blob) { | |
2816 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
2817 | } | |
2818 | } | |
2819 | } | |
2820 | /*derr << __func__ << bl << dendl; | |
2821 | derr << __func__ << ":"; | |
2822 | bl.hexdump(*_dout); | |
2823 | *_dout << dendl; | |
2824 | */ | |
2825 | return false; | |
2826 | } | |
2827 | ||
2828 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
2829 | { | |
7c673cae FG |
2830 | /* |
2831 | derr << __func__ << ":"; | |
2832 | bl.hexdump(*_dout); | |
2833 | *_dout << dendl; | |
2834 | */ | |
2835 | ||
11fdf7f2 | 2836 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae FG |
2837 | auto p = bl.front().begin_deep(); |
2838 | __u8 struct_v; | |
2839 | denc(struct_v, p); | |
2840 | // Version 2 differs from v1 in blob's ref_map | |
2841 | // serialization only. Hence there is no specific | |
2842 | // handling at ExtentMap level below. | |
11fdf7f2 | 2843 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
2844 | |
2845 | uint32_t num; | |
2846 | denc_varint(num, p); | |
2847 | vector<BlobRef> blobs(num); | |
2848 | uint64_t pos = 0; | |
2849 | uint64_t prev_len = 0; | |
2850 | unsigned n = 0; | |
2851 | ||
2852 | while (!p.end()) { | |
2853 | Extent *le = new Extent(); | |
2854 | uint64_t blobid; | |
2855 | denc_varint(blobid, p); | |
2856 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
2857 | uint64_t gap; | |
2858 | denc_varint_lowz(gap, p); | |
2859 | pos += gap; | |
2860 | } | |
2861 | le->logical_offset = pos; | |
2862 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
2863 | denc_varint_lowz(le->blob_offset, p); | |
2864 | } else { | |
2865 | le->blob_offset = 0; | |
2866 | } | |
2867 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
2868 | denc_varint_lowz(prev_len, p); | |
2869 | } | |
2870 | le->length = prev_len; | |
2871 | ||
2872 | if (blobid & BLOBID_FLAG_SPANNING) { | |
2873 | dout(30) << __func__ << " getting spanning blob " | |
2874 | << (blobid >> BLOBID_SHIFT_BITS) << dendl; | |
2875 | le->assign_blob(get_spanning_blob(blobid >> BLOBID_SHIFT_BITS)); | |
2876 | } else { | |
2877 | blobid >>= BLOBID_SHIFT_BITS; | |
2878 | if (blobid) { | |
2879 | le->assign_blob(blobs[blobid - 1]); | |
11fdf7f2 | 2880 | ceph_assert(le->blob); |
7c673cae FG |
2881 | } else { |
2882 | Blob *b = new Blob(); | |
2883 | uint64_t sbid = 0; | |
2884 | b->decode(onode->c, p, struct_v, &sbid, false); | |
2885 | blobs[n] = b; | |
2886 | onode->c->open_shared_blob(sbid, b); | |
2887 | le->assign_blob(b); | |
2888 | } | |
2889 | // we build ref_map dynamically for non-spanning blobs | |
2890 | le->blob->get_ref( | |
2891 | onode->c, | |
2892 | le->blob_offset, | |
2893 | le->length); | |
2894 | } | |
2895 | pos += prev_len; | |
2896 | ++n; | |
2897 | extent_map.insert(*le); | |
2898 | } | |
2899 | ||
11fdf7f2 | 2900 | ceph_assert(n == num); |
7c673cae FG |
2901 | return num; |
2902 | } | |
2903 | ||
2904 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
2905 | { | |
2906 | // Version 2 differs from v1 in blob's ref_map | |
2907 | // serialization only. Hence there is no specific | |
2908 | // handling at ExtentMap level. | |
2909 | __u8 struct_v = 2; | |
2910 | ||
2911 | denc(struct_v, p); | |
2912 | denc_varint((uint32_t)0, p); | |
2913 | size_t key_size = 0; | |
2914 | denc_varint((uint32_t)0, key_size); | |
2915 | p += spanning_blob_map.size() * key_size; | |
2916 | for (const auto& i : spanning_blob_map) { | |
2917 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
2918 | } | |
2919 | } | |
2920 | ||
2921 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
2922 | bufferlist::contiguous_appender& p) | |
2923 | { | |
2924 | // Version 2 differs from v1 in blob's ref_map | |
2925 | // serialization only. Hence there is no specific | |
2926 | // handling at ExtentMap level. | |
2927 | __u8 struct_v = 2; | |
2928 | ||
2929 | denc(struct_v, p); | |
2930 | denc_varint(spanning_blob_map.size(), p); | |
2931 | for (auto& i : spanning_blob_map) { | |
2932 | denc_varint(i.second->id, p); | |
2933 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
2934 | } | |
2935 | } | |
2936 | ||
2937 | void BlueStore::ExtentMap::decode_spanning_blobs( | |
11fdf7f2 | 2938 | bufferptr::const_iterator& p) |
7c673cae FG |
2939 | { |
2940 | __u8 struct_v; | |
2941 | denc(struct_v, p); | |
2942 | // Version 2 differs from v1 in blob's ref_map | |
2943 | // serialization only. Hence there is no specific | |
2944 | // handling at ExtentMap level. | |
11fdf7f2 | 2945 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
2946 | |
2947 | unsigned n; | |
2948 | denc_varint(n, p); | |
2949 | while (n--) { | |
2950 | BlobRef b(new Blob()); | |
2951 | denc_varint(b->id, p); | |
2952 | spanning_blob_map[b->id] = b; | |
2953 | uint64_t sbid = 0; | |
2954 | b->decode(onode->c, p, struct_v, &sbid, true); | |
2955 | onode->c->open_shared_blob(sbid, b); | |
2956 | } | |
2957 | } | |
2958 | ||
2959 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) | |
2960 | { | |
2961 | shards.resize(onode->onode.extent_map_shards.size()); | |
2962 | unsigned i = 0; | |
2963 | for (auto &s : onode->onode.extent_map_shards) { | |
2964 | shards[i].shard_info = &s; | |
2965 | shards[i].loaded = loaded; | |
2966 | shards[i].dirty = dirty; | |
2967 | ++i; | |
2968 | } | |
2969 | } | |
2970 | ||
2971 | void BlueStore::ExtentMap::fault_range( | |
2972 | KeyValueDB *db, | |
2973 | uint32_t offset, | |
2974 | uint32_t length) | |
2975 | { | |
7c673cae FG |
2976 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
2977 | << std::dec << dendl; | |
2978 | auto start = seek_shard(offset); | |
2979 | auto last = seek_shard(offset + length); | |
2980 | ||
2981 | if (start < 0) | |
2982 | return; | |
2983 | ||
11fdf7f2 | 2984 | ceph_assert(last >= start); |
7c673cae FG |
2985 | string key; |
2986 | while (start <= last) { | |
11fdf7f2 | 2987 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
2988 | auto p = &shards[start]; |
2989 | if (!p->loaded) { | |
2990 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
2991 | << p->shard_info->offset << std::dec << dendl; | |
2992 | bufferlist v; | |
2993 | generate_extent_shard_key_and_apply( | |
2994 | onode->key, p->shard_info->offset, &key, | |
2995 | [&](const string& final_key) { | |
2996 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
2997 | if (r < 0) { | |
2998 | derr << __func__ << " missing shard 0x" << std::hex | |
2999 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
3000 | << dendl; | |
11fdf7f2 | 3001 | ceph_assert(r >= 0); |
7c673cae FG |
3002 | } |
3003 | } | |
3004 | ); | |
3005 | p->extents = decode_some(v); | |
3006 | p->loaded = true; | |
3007 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
3008 | << p->shard_info->offset |
3009 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 3010 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
3011 | ceph_assert(p->dirty == false); |
3012 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
3013 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
3014 | } else { | |
3015 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
3016 | } | |
3017 | ++start; | |
3018 | } | |
3019 | } | |
3020 | ||
3021 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
3022 | uint32_t offset, |
3023 | uint32_t length) | |
3024 | { | |
7c673cae FG |
3025 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3026 | << std::dec << dendl; | |
3027 | if (shards.empty()) { | |
3028 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
3029 | inline_bl.clear(); | |
3030 | return; | |
3031 | } | |
3032 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
3033 | if (length == 0) { |
3034 | length = 1; | |
3035 | } | |
3036 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
3037 | if (start < 0) |
3038 | return; | |
3039 | ||
11fdf7f2 | 3040 | ceph_assert(last >= start); |
7c673cae | 3041 | while (start <= last) { |
11fdf7f2 | 3042 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3043 | auto p = &shards[start]; |
3044 | if (!p->loaded) { | |
11fdf7f2 TL |
3045 | derr << __func__ << "on write 0x" << std::hex << offset |
3046 | << "~" << length << " shard 0x" << p->shard_info->offset | |
3047 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
3048 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
3049 | } |
3050 | if (!p->dirty) { | |
3051 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
3052 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
3053 | p->dirty = true; | |
3054 | } | |
3055 | ++start; | |
3056 | } | |
3057 | } | |
3058 | ||
3059 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
3060 | uint64_t offset) | |
3061 | { | |
3062 | Extent dummy(offset); | |
3063 | return extent_map.find(dummy); | |
3064 | } | |
3065 | ||
7c673cae FG |
3066 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
3067 | uint64_t offset) | |
3068 | { | |
3069 | Extent dummy(offset); | |
3070 | auto fp = extent_map.lower_bound(dummy); | |
3071 | if (fp != extent_map.begin()) { | |
3072 | --fp; | |
3073 | if (fp->logical_end() <= offset) { | |
3074 | ++fp; | |
3075 | } | |
3076 | } | |
3077 | return fp; | |
3078 | } | |
3079 | ||
3080 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
3081 | uint64_t offset) const | |
3082 | { | |
3083 | Extent dummy(offset); | |
3084 | auto fp = extent_map.lower_bound(dummy); | |
3085 | if (fp != extent_map.begin()) { | |
3086 | --fp; | |
3087 | if (fp->logical_end() <= offset) { | |
3088 | ++fp; | |
3089 | } | |
3090 | } | |
3091 | return fp; | |
3092 | } | |
3093 | ||
3094 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
3095 | { | |
3096 | auto fp = seek_lextent(offset); | |
3097 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
3098 | return false; | |
3099 | } | |
3100 | return true; | |
3101 | } | |
3102 | ||
3103 | int BlueStore::ExtentMap::compress_extent_map( | |
3104 | uint64_t offset, | |
3105 | uint64_t length) | |
3106 | { | |
7c673cae FG |
3107 | if (extent_map.empty()) |
3108 | return 0; | |
3109 | int removed = 0; | |
3110 | auto p = seek_lextent(offset); | |
3111 | if (p != extent_map.begin()) { | |
3112 | --p; // start to the left of offset | |
3113 | } | |
3114 | // the caller should have just written to this region | |
11fdf7f2 | 3115 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
3116 | |
3117 | // identify the *next* shard | |
3118 | auto pshard = shards.begin(); | |
3119 | while (pshard != shards.end() && | |
3120 | p->logical_offset >= pshard->shard_info->offset) { | |
3121 | ++pshard; | |
3122 | } | |
3123 | uint64_t shard_end; | |
3124 | if (pshard != shards.end()) { | |
3125 | shard_end = pshard->shard_info->offset; | |
3126 | } else { | |
3127 | shard_end = OBJECT_MAX_SIZE; | |
3128 | } | |
3129 | ||
3130 | auto n = p; | |
3131 | for (++n; n != extent_map.end(); p = n++) { | |
3132 | if (n->logical_offset > offset + length) { | |
3133 | break; // stop after end | |
3134 | } | |
3135 | while (n != extent_map.end() && | |
3136 | p->logical_end() == n->logical_offset && | |
3137 | p->blob == n->blob && | |
3138 | p->blob_offset + p->length == n->blob_offset && | |
3139 | n->logical_offset < shard_end) { | |
3140 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3141 | << " next shard 0x" << shard_end << std::dec | |
3142 | << " merging " << *p << " and " << *n << dendl; | |
3143 | p->length += n->length; | |
3144 | rm(n++); | |
3145 | ++removed; | |
3146 | } | |
3147 | if (n == extent_map.end()) { | |
3148 | break; | |
3149 | } | |
3150 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3151 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3152 | ++pshard; |
3153 | if (pshard != shards.end()) { | |
3154 | shard_end = pshard->shard_info->offset; | |
3155 | } else { | |
3156 | shard_end = OBJECT_MAX_SIZE; | |
3157 | } | |
3158 | } | |
3159 | } | |
11fdf7f2 | 3160 | if (removed) { |
7c673cae FG |
3161 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3162 | } | |
3163 | return removed; | |
3164 | } | |
3165 | ||
3166 | void BlueStore::ExtentMap::punch_hole( | |
3167 | CollectionRef &c, | |
3168 | uint64_t offset, | |
3169 | uint64_t length, | |
3170 | old_extent_map_t *old_extents) | |
3171 | { | |
3172 | auto p = seek_lextent(offset); | |
3173 | uint64_t end = offset + length; | |
3174 | while (p != extent_map.end()) { | |
3175 | if (p->logical_offset >= end) { | |
3176 | break; | |
3177 | } | |
3178 | if (p->logical_offset < offset) { | |
3179 | if (p->logical_end() > end) { | |
3180 | // split and deref middle | |
3181 | uint64_t front = offset - p->logical_offset; | |
3182 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3183 | length, p->blob); | |
3184 | old_extents->push_back(*oe); | |
3185 | add(end, | |
3186 | p->blob_offset + front + length, | |
3187 | p->length - front - length, | |
3188 | p->blob); | |
3189 | p->length = front; | |
3190 | break; | |
3191 | } else { | |
3192 | // deref tail | |
11fdf7f2 | 3193 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3194 | uint64_t keep = offset - p->logical_offset; |
3195 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3196 | p->length - keep, p->blob); | |
3197 | old_extents->push_back(*oe); | |
3198 | p->length = keep; | |
3199 | ++p; | |
3200 | continue; | |
3201 | } | |
3202 | } | |
3203 | if (p->logical_offset + p->length <= end) { | |
3204 | // deref whole lextent | |
3205 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3206 | p->length, p->blob); | |
3207 | old_extents->push_back(*oe); | |
3208 | rm(p++); | |
3209 | continue; | |
3210 | } | |
3211 | // deref head | |
3212 | uint64_t keep = p->logical_end() - end; | |
3213 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3214 | p->length - keep, p->blob); | |
3215 | old_extents->push_back(*oe); | |
3216 | ||
3217 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3218 | rm(p); | |
3219 | break; | |
3220 | } | |
3221 | } | |
3222 | ||
3223 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3224 | CollectionRef &c, | |
3225 | uint64_t logical_offset, | |
3226 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3227 | old_extent_map_t *old_extents) | |
3228 | { | |
3229 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3230 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3231 | |
3232 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3233 | // old_extents list if we overwre the blob totally | |
3234 | // This might happen during WAL overwrite. | |
3235 | b->get_ref(onode->c, blob_offset, length); | |
3236 | ||
3237 | if (old_extents) { | |
3238 | punch_hole(c, logical_offset, length, old_extents); | |
3239 | } | |
3240 | ||
3241 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3242 | extent_map.insert(*le); | |
3243 | if (spans_shard(logical_offset, length)) { | |
3244 | request_reshard(logical_offset, logical_offset + length); | |
3245 | } | |
3246 | return le; | |
3247 | } | |
3248 | ||
3249 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3250 | BlobRef lb, | |
3251 | uint32_t blob_offset, | |
3252 | uint32_t pos) | |
3253 | { | |
7c673cae FG |
3254 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; |
3255 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3256 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3257 | << dendl; | |
3258 | BlobRef rb = onode->c->new_blob(); | |
3259 | lb->split(onode->c, blob_offset, rb.get()); | |
3260 | ||
3261 | for (auto ep = seek_lextent(pos); | |
3262 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3263 | ++ep) { | |
3264 | if (ep->blob != lb) { | |
3265 | continue; | |
3266 | } | |
3267 | if (ep->logical_offset < pos) { | |
3268 | // split extent | |
3269 | size_t left = pos - ep->logical_offset; | |
3270 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3271 | extent_map.insert(*ne); | |
3272 | ep->length = left; | |
3273 | dout(30) << __func__ << " split " << *ep << dendl; | |
3274 | dout(30) << __func__ << " to " << *ne << dendl; | |
3275 | } else { | |
3276 | // switch blob | |
11fdf7f2 | 3277 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3278 | |
3279 | ep->blob = rb; | |
3280 | ep->blob_offset -= blob_offset; | |
3281 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3282 | } | |
3283 | } | |
3284 | return rb; | |
3285 | } | |
3286 | ||
3287 | // Onode | |
3288 | ||
3289 | #undef dout_prefix | |
3290 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3291 | ||
eafe8130 TL |
3292 | BlueStore::Onode* BlueStore::Onode::decode( |
3293 | CollectionRef c, | |
3294 | const ghobject_t& oid, | |
3295 | const string& key, | |
3296 | const bufferlist& v) | |
3297 | { | |
3298 | Onode* on = new Onode(c.get(), oid, key); | |
3299 | on->exists = true; | |
3300 | auto p = v.front().begin_deep(); | |
3301 | on->onode.decode(p); | |
3302 | for (auto& i : on->onode.attrs) { | |
3303 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
3304 | } | |
3305 | ||
3306 | // initialize extent_map | |
3307 | on->extent_map.decode_spanning_blobs(p); | |
3308 | if (on->onode.extent_map_shards.empty()) { | |
3309 | denc(on->extent_map.inline_bl, p); | |
3310 | on->extent_map.decode_some(on->extent_map.inline_bl); | |
3311 | on->extent_map.inline_bl.reassign_to_mempool( | |
3312 | mempool::mempool_bluestore_cache_other); | |
3313 | } | |
3314 | else { | |
3315 | on->extent_map.init_shards(false, false); | |
3316 | } | |
3317 | return on; | |
3318 | } | |
3319 | ||
7c673cae FG |
3320 | void BlueStore::Onode::flush() |
3321 | { | |
3322 | if (flushing_count.load()) { | |
3323 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
9f95a23c | 3324 | waiting_count++; |
11fdf7f2 | 3325 | std::unique_lock l(flush_lock); |
7c673cae FG |
3326 | while (flushing_count.load()) { |
3327 | flush_cond.wait(l); | |
3328 | } | |
9f95a23c | 3329 | waiting_count--; |
7c673cae FG |
3330 | } |
3331 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3332 | } | |
3333 | ||
9f95a23c TL |
3334 | void BlueStore::Onode::dump(Formatter* f) const |
3335 | { | |
3336 | onode.dump(f); | |
3337 | extent_map.dump(f); | |
3338 | } | |
3339 | ||
3340 | ||
3341 | const string& BlueStore::Onode::get_omap_prefix() | |
3342 | { | |
3343 | if (onode.is_pgmeta_omap()) { | |
3344 | return PREFIX_PGMETA_OMAP; | |
3345 | } | |
3346 | if (onode.is_perpool_omap()) { | |
3347 | return PREFIX_PERPOOL_OMAP; | |
3348 | } | |
3349 | return PREFIX_OMAP; | |
3350 | } | |
3351 | ||
3352 | // '-' < '.' < '~' | |
3353 | ||
3354 | void BlueStore::Onode::get_omap_header(string *out) | |
3355 | { | |
3356 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3357 | _key_encode_u64(c->pool(), out); | |
3358 | } | |
3359 | _key_encode_u64(onode.nid, out); | |
3360 | out->push_back('-'); | |
3361 | } | |
3362 | ||
3363 | void BlueStore::Onode::get_omap_key(const string& key, string *out) | |
3364 | { | |
3365 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3366 | _key_encode_u64(c->pool(), out); | |
3367 | } | |
3368 | _key_encode_u64(onode.nid, out); | |
3369 | out->push_back('.'); | |
3370 | out->append(key); | |
3371 | } | |
3372 | ||
3373 | void BlueStore::Onode::rewrite_omap_key(const string& old, string *out) | |
3374 | { | |
3375 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3376 | _key_encode_u64(c->pool(), out); | |
3377 | } | |
3378 | _key_encode_u64(onode.nid, out); | |
3379 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
3380 | } | |
3381 | ||
3382 | void BlueStore::Onode::get_omap_tail(string *out) | |
3383 | { | |
3384 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3385 | _key_encode_u64(c->pool(), out); | |
3386 | } | |
3387 | _key_encode_u64(onode.nid, out); | |
3388 | out->push_back('~'); | |
3389 | } | |
3390 | ||
3391 | void BlueStore::Onode::decode_omap_key(const string& key, string *user_key) | |
3392 | { | |
3393 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3394 | *user_key = key.substr(sizeof(uint64_t)*2 + 1); | |
3395 | } else { | |
3396 | *user_key = key.substr(sizeof(uint64_t) + 1); | |
3397 | } | |
3398 | } | |
3399 | ||
3400 | ||
7c673cae FG |
3401 | // ======================================================= |
3402 | // WriteContext | |
3403 | ||
3404 | /// Checks for writes to the same pextent within a blob | |
3405 | bool BlueStore::WriteContext::has_conflict( | |
3406 | BlobRef b, | |
3407 | uint64_t loffs, | |
3408 | uint64_t loffs_end, | |
3409 | uint64_t min_alloc_size) | |
3410 | { | |
11fdf7f2 TL |
3411 | ceph_assert((loffs % min_alloc_size) == 0); |
3412 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3413 | for (auto w : writes) { |
3414 | if (b == w.b) { | |
11fdf7f2 TL |
3415 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3416 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3417 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3418 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3419 | return true; |
3420 | } | |
3421 | } | |
3422 | } | |
3423 | return false; | |
3424 | } | |
3425 | ||
3426 | // ======================================================= | |
3427 | ||
3428 | // DeferredBatch | |
3429 | #undef dout_prefix | |
3430 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
9f95a23c TL |
3431 | #undef dout_context |
3432 | #define dout_context cct | |
7c673cae FG |
3433 | |
3434 | void BlueStore::DeferredBatch::prepare_write( | |
3435 | CephContext *cct, | |
3436 | uint64_t seq, uint64_t offset, uint64_t length, | |
3437 | bufferlist::const_iterator& blp) | |
3438 | { | |
3439 | _discard(cct, offset, length); | |
3440 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3441 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3442 | i.first->second.seq = seq; |
3443 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3444 | i.first->second.bl.reassign_to_mempool( |
3445 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3446 | dout(20) << __func__ << " seq " << seq |
3447 | << " 0x" << std::hex << offset << "~" << length | |
3448 | << " crc " << i.first->second.bl.crc32c(-1) | |
3449 | << std::dec << dendl; | |
3450 | seq_bytes[seq] += length; | |
3451 | #ifdef DEBUG_DEFERRED | |
3452 | _audit(cct); | |
3453 | #endif | |
3454 | } | |
3455 | ||
3456 | void BlueStore::DeferredBatch::_discard( | |
3457 | CephContext *cct, uint64_t offset, uint64_t length) | |
3458 | { | |
3459 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3460 | << std::dec << dendl; | |
3461 | auto p = iomap.lower_bound(offset); | |
3462 | if (p != iomap.begin()) { | |
3463 | --p; | |
3464 | auto end = p->first + p->second.bl.length(); | |
3465 | if (end > offset) { | |
3466 | bufferlist head; | |
3467 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3468 | dout(20) << __func__ << " keep head " << p->second.seq | |
3469 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3470 | << " -> 0x" << head.length() << std::dec << dendl; | |
3471 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3472 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3473 | if (end > offset + length) { |
3474 | bufferlist tail; | |
3475 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3476 | end - (offset + length)); | |
3477 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3478 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3479 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3480 | auto &n = iomap[offset + length]; | |
3481 | n.bl.swap(tail); | |
3482 | n.seq = p->second.seq; | |
3483 | i->second -= length; | |
3484 | } else { | |
3485 | i->second -= end - offset; | |
3486 | } | |
11fdf7f2 | 3487 | ceph_assert(i->second >= 0); |
7c673cae FG |
3488 | p->second.bl.swap(head); |
3489 | } | |
3490 | ++p; | |
3491 | } | |
3492 | while (p != iomap.end()) { | |
3493 | if (p->first >= offset + length) { | |
3494 | break; | |
3495 | } | |
3496 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3497 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3498 | auto end = p->first + p->second.bl.length(); |
3499 | if (end > offset + length) { | |
3500 | unsigned drop_front = offset + length - p->first; | |
3501 | unsigned keep_tail = end - (offset + length); | |
3502 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3503 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3504 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3505 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3506 | << std::dec << dendl; | |
3507 | auto &s = iomap[offset + length]; | |
3508 | s.seq = p->second.seq; | |
3509 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3510 | i->second -= drop_front; | |
3511 | } else { | |
3512 | dout(20) << __func__ << " drop " << p->second.seq | |
3513 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3514 | << std::dec << dendl; | |
3515 | i->second -= p->second.bl.length(); | |
3516 | } | |
11fdf7f2 | 3517 | ceph_assert(i->second >= 0); |
7c673cae FG |
3518 | p = iomap.erase(p); |
3519 | } | |
3520 | } | |
3521 | ||
3522 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3523 | { | |
3524 | map<uint64_t,int> sb; | |
3525 | for (auto p : seq_bytes) { | |
3526 | sb[p.first] = 0; // make sure we have the same set of keys | |
3527 | } | |
3528 | uint64_t pos = 0; | |
3529 | for (auto& p : iomap) { | |
11fdf7f2 | 3530 | ceph_assert(p.first >= pos); |
7c673cae FG |
3531 | sb[p.second.seq] += p.second.bl.length(); |
3532 | pos = p.first + p.second.bl.length(); | |
3533 | } | |
11fdf7f2 | 3534 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3535 | } |
3536 | ||
3537 | ||
3538 | // Collection | |
3539 | ||
3540 | #undef dout_prefix | |
3541 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3542 | ||
9f95a23c TL |
3543 | BlueStore::Collection::Collection(BlueStore *store_, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t cid) |
3544 | : CollectionImpl(store_->cct, cid), | |
11fdf7f2 | 3545 | store(store_), |
9f95a23c | 3546 | cache(bc), |
7c673cae | 3547 | exists(true), |
9f95a23c | 3548 | onode_map(oc), |
11fdf7f2 TL |
3549 | commit_queue(nullptr) |
3550 | { | |
3551 | } | |
3552 | ||
3553 | bool BlueStore::Collection::flush_commit(Context *c) | |
3554 | { | |
3555 | return osr->flush_commit(c); | |
3556 | } | |
3557 | ||
3558 | void BlueStore::Collection::flush() | |
3559 | { | |
3560 | osr->flush(); | |
3561 | } | |
3562 | ||
3563 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3564 | { |
11fdf7f2 | 3565 | osr->flush_all_but_last(); |
7c673cae FG |
3566 | } |
3567 | ||
3568 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3569 | { | |
11fdf7f2 | 3570 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3571 | const bluestore_blob_t& blob = b->get_blob(); |
3572 | if (!blob.is_shared()) { | |
3573 | b->shared_blob = new SharedBlob(this); | |
3574 | return; | |
3575 | } | |
3576 | ||
3577 | b->shared_blob = shared_blob_set.lookup(sbid); | |
3578 | if (b->shared_blob) { | |
3579 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3580 | << std::dec << " had " << *b->shared_blob << dendl; | |
3581 | } else { | |
3582 | b->shared_blob = new SharedBlob(sbid, this); | |
3583 | shared_blob_set.add(this, b->shared_blob.get()); | |
3584 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3585 | << std::dec << " opened " << *b->shared_blob | |
3586 | << dendl; | |
3587 | } | |
3588 | } | |
3589 | ||
3590 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
3591 | { | |
3592 | if (!sb->is_loaded()) { | |
3593 | ||
3594 | bufferlist v; | |
3595 | string key; | |
3596 | auto sbid = sb->get_sbid(); | |
3597 | get_shared_blob_key(sbid, &key); | |
3598 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
3599 | if (r < 0) { | |
3600 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
3601 | << std::dec << " not found at key " | |
3602 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 3603 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
3604 | } |
3605 | ||
3606 | sb->loaded = true; | |
3607 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
3608 | auto p = v.cbegin(); |
3609 | decode(*(sb->persistent), p); | |
7c673cae FG |
3610 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
3611 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
3612 | } | |
3613 | } | |
3614 | ||
3615 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
3616 | { | |
7c673cae | 3617 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 3618 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
3619 | |
3620 | // update blob | |
31f18b77 | 3621 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 3622 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
3623 | |
3624 | // update shared blob | |
3625 | b->shared_blob->loaded = true; | |
3626 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
3627 | shared_blob_set.add(this, b->shared_blob.get()); | |
3628 | for (auto p : blob.get_extents()) { | |
3629 | if (p.is_valid()) { | |
3630 | b->shared_blob->get_ref( | |
3631 | p.offset, | |
3632 | p.length); | |
3633 | } | |
3634 | } | |
3635 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
3636 | } | |
3637 | ||
31f18b77 FG |
3638 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
3639 | { | |
3640 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 3641 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
3642 | |
3643 | uint64_t sbid = sb->get_sbid(); | |
3644 | shared_blob_set.remove(sb); | |
3645 | sb->loaded = false; | |
3646 | delete sb->persistent; | |
3647 | sb->sbid_unloaded = 0; | |
3648 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
3649 | return sbid; | |
3650 | } | |
3651 | ||
7c673cae FG |
3652 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
3653 | const ghobject_t& oid, | |
9f95a23c TL |
3654 | bool create, |
3655 | bool is_createop) | |
7c673cae | 3656 | { |
9f95a23c | 3657 | ceph_assert(create ? ceph_mutex_is_wlocked(lock) : ceph_mutex_is_locked(lock)); |
7c673cae FG |
3658 | |
3659 | spg_t pgid; | |
3660 | if (cid.is_pg(&pgid)) { | |
3661 | if (!oid.match(cnode.bits, pgid.ps())) { | |
3662 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
3663 | << pgid << " bits " << cnode.bits << dendl; | |
3664 | ceph_abort(); | |
3665 | } | |
3666 | } | |
3667 | ||
3668 | OnodeRef o = onode_map.lookup(oid); | |
3669 | if (o) | |
3670 | return o; | |
3671 | ||
eafe8130 | 3672 | string key; |
7c673cae FG |
3673 | get_object_key(store->cct, oid, &key); |
3674 | ||
3675 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
3676 | << pretty_binary_string(key) << dendl; | |
3677 | ||
3678 | bufferlist v; | |
9f95a23c | 3679 | int r = -ENOENT; |
7c673cae | 3680 | Onode *on; |
9f95a23c TL |
3681 | if (!is_createop) { |
3682 | r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
3683 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
3684 | } | |
7c673cae | 3685 | if (v.length() == 0) { |
11fdf7f2 | 3686 | ceph_assert(r == -ENOENT); |
7c673cae FG |
3687 | if (!store->cct->_conf->bluestore_debug_misc && |
3688 | !create) | |
3689 | return OnodeRef(); | |
3690 | ||
3691 | // new object, new onode | |
3692 | on = new Onode(this, oid, key); | |
3693 | } else { | |
3694 | // loaded | |
11fdf7f2 | 3695 | ceph_assert(r >= 0); |
eafe8130 | 3696 | on = Onode::decode(this, oid, key, v); |
7c673cae FG |
3697 | } |
3698 | o.reset(on); | |
3699 | return onode_map.add(oid, o); | |
3700 | } | |
3701 | ||
3702 | void BlueStore::Collection::split_cache( | |
3703 | Collection *dest) | |
3704 | { | |
3705 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
3706 | ||
3707 | // lock (one or both) cache shards | |
3708 | std::lock(cache->lock, dest->cache->lock); | |
11fdf7f2 TL |
3709 | std::lock_guard l(cache->lock, std::adopt_lock); |
3710 | std::lock_guard l2(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
3711 | |
3712 | int destbits = dest->cnode.bits; | |
3713 | spg_t destpg; | |
3714 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 3715 | ceph_assert(is_pg); |
7c673cae FG |
3716 | |
3717 | auto p = onode_map.onode_map.begin(); | |
3718 | while (p != onode_map.onode_map.end()) { | |
11fdf7f2 | 3719 | OnodeRef o = p->second; |
7c673cae FG |
3720 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
3721 | // onode does not belong to this child | |
11fdf7f2 TL |
3722 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
3723 | << dendl; | |
7c673cae FG |
3724 | ++p; |
3725 | } else { | |
7c673cae FG |
3726 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
3727 | << dendl; | |
3728 | ||
9f95a23c TL |
3729 | // move the onode to the new map before futzing with the cache |
3730 | // shard, ensuring that nref is always >= 2, and no racing | |
3731 | // thread can trigger a pin or unpin (which does *not* behave | |
3732 | // well when we are clearing and resetting the 's' shard | |
3733 | // pointer!). | |
7c673cae | 3734 | p = onode_map.onode_map.erase(p); |
7c673cae | 3735 | dest->onode_map.onode_map[o->oid] = o; |
9f95a23c TL |
3736 | |
3737 | if (onode_map.cache != dest->onode_map.cache) { | |
3738 | // move onode to a different cache shard | |
3739 | onode_map.cache->_rm(o); | |
3740 | o->c = dest; | |
3741 | dest->onode_map.cache->_add(o, 1); | |
3742 | } else { | |
3743 | // the onode is in the same cache shard, making our move simpler. | |
3744 | o->c = dest; | |
3745 | } | |
7c673cae FG |
3746 | |
3747 | // move over shared blobs and buffers. cover shared blobs from | |
3748 | // both extent map and spanning blob map (the full extent map | |
3749 | // may not be faulted in) | |
3750 | vector<SharedBlob*> sbvec; | |
3751 | for (auto& e : o->extent_map.extent_map) { | |
3752 | sbvec.push_back(e.blob->shared_blob.get()); | |
3753 | } | |
3754 | for (auto& b : o->extent_map.spanning_blob_map) { | |
3755 | sbvec.push_back(b.second->shared_blob.get()); | |
3756 | } | |
3757 | for (auto sb : sbvec) { | |
3758 | if (sb->coll == dest) { | |
3759 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
3760 | << dendl; | |
3761 | continue; | |
3762 | } | |
3763 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
3764 | if (sb->get_sbid()) { |
3765 | ldout(store->cct, 20) << __func__ | |
3766 | << " moving registration " << *sb << dendl; | |
3767 | shared_blob_set.remove(sb); | |
3768 | dest->shared_blob_set.add(dest, sb); | |
3769 | } | |
3efd9988 | 3770 | sb->coll = dest; |
7c673cae | 3771 | if (dest->cache != cache) { |
7c673cae FG |
3772 | for (auto& i : sb->bc.buffer_map) { |
3773 | if (!i.second->is_writing()) { | |
3774 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
3775 | << dendl; | |
9f95a23c | 3776 | dest->cache->_move(cache, i.second.get()); |
7c673cae FG |
3777 | } |
3778 | } | |
3779 | } | |
3780 | } | |
7c673cae FG |
3781 | } |
3782 | } | |
9f95a23c | 3783 | dest->cache->_trim(); |
7c673cae FG |
3784 | } |
3785 | ||
7c673cae FG |
3786 | // ======================================================= |
3787 | ||
91327a77 AA |
3788 | // MempoolThread |
3789 | ||
3790 | #undef dout_prefix | |
3791 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
9f95a23c TL |
3792 | #undef dout_context |
3793 | #define dout_context store->cct | |
91327a77 | 3794 | |
7c673cae FG |
3795 | void *BlueStore::MempoolThread::entry() |
3796 | { | |
9f95a23c | 3797 | std::unique_lock l{lock}; |
11fdf7f2 | 3798 | |
92f5a8d4 | 3799 | uint32_t prev_config_change = store->config_changed.load(); |
eafe8130 TL |
3800 | uint64_t base = store->osd_memory_base; |
3801 | double fragmentation = store->osd_memory_expected_fragmentation; | |
3802 | uint64_t target = store->osd_memory_target; | |
3803 | uint64_t min = store->osd_memory_cache_min; | |
3804 | uint64_t max = min; | |
3805 | ||
3806 | // When setting the maximum amount of memory to use for cache, first | |
3807 | // assume some base amount of memory for the OSD and then fudge in | |
3808 | // some overhead for fragmentation that scales with cache usage. | |
3809 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
3810 | if (ltarget > base + min) { | |
3811 | max = ltarget - base; | |
11fdf7f2 | 3812 | } |
31f18b77 | 3813 | |
eafe8130 TL |
3814 | binned_kv_cache = store->db->get_priority_cache(); |
3815 | if (store->cache_autotune && binned_kv_cache != nullptr) { | |
3816 | pcm = std::make_shared<PriorityCache::Manager>( | |
3817 | store->cct, min, max, target, true); | |
3818 | pcm->insert("kv", binned_kv_cache, true); | |
3819 | pcm->insert("meta", meta_cache, true); | |
3820 | pcm->insert("data", data_cache, true); | |
3821 | } | |
91327a77 AA |
3822 | |
3823 | utime_t next_balance = ceph_clock_now(); | |
3824 | utime_t next_resize = ceph_clock_now(); | |
9f95a23c TL |
3825 | utime_t next_deferred_force_submit = ceph_clock_now(); |
3826 | utime_t alloc_stats_dump_clock = ceph_clock_now(); | |
31f18b77 | 3827 | |
91327a77 | 3828 | bool interval_stats_trim = false; |
91327a77 | 3829 | while (!stop) { |
92f5a8d4 TL |
3830 | // Update pcm cache settings if related configuration was changed |
3831 | uint32_t cur_config_change = store->config_changed.load(); | |
3832 | if (cur_config_change != prev_config_change) { | |
3833 | _update_cache_settings(); | |
3834 | prev_config_change = cur_config_change; | |
3835 | } | |
3836 | ||
91327a77 AA |
3837 | // Before we trim, check and see if it's time to rebalance/resize. |
3838 | double autotune_interval = store->cache_autotune_interval; | |
3839 | double resize_interval = store->osd_memory_cache_resize_interval; | |
9f95a23c TL |
3840 | double max_defer_interval = store->max_defer_interval; |
3841 | ||
3842 | double alloc_stats_dump_interval = | |
3843 | store->cct->_conf->bluestore_alloc_stats_dump_interval; | |
91327a77 | 3844 | |
9f95a23c TL |
3845 | if (alloc_stats_dump_interval > 0 && |
3846 | alloc_stats_dump_clock + alloc_stats_dump_interval < ceph_clock_now()) { | |
3847 | store->_record_allocation_stats(); | |
3848 | alloc_stats_dump_clock = ceph_clock_now(); | |
3849 | } | |
91327a77 | 3850 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { |
11fdf7f2 TL |
3851 | _adjust_cache_settings(); |
3852 | ||
91327a77 | 3853 | // Log events at 5 instead of 20 when balance happens. |
91327a77 | 3854 | interval_stats_trim = true; |
eafe8130 TL |
3855 | |
3856 | if (pcm != nullptr) { | |
3857 | pcm->balance(); | |
91327a77 | 3858 | } |
31f18b77 | 3859 | |
91327a77 AA |
3860 | next_balance = ceph_clock_now(); |
3861 | next_balance += autotune_interval; | |
3862 | } | |
3863 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { | |
eafe8130 TL |
3864 | if (ceph_using_tcmalloc() && pcm != nullptr) { |
3865 | pcm->tune_memory(); | |
91327a77 AA |
3866 | } |
3867 | next_resize = ceph_clock_now(); | |
3868 | next_resize += resize_interval; | |
31f18b77 FG |
3869 | } |
3870 | ||
9f95a23c TL |
3871 | if (max_defer_interval > 0 && |
3872 | next_deferred_force_submit < ceph_clock_now()) { | |
3873 | if (store->get_deferred_last_submitted() + max_defer_interval < | |
3874 | ceph_clock_now()) { | |
3875 | store->deferred_try_submit(); | |
3876 | } | |
3877 | next_deferred_force_submit = ceph_clock_now(); | |
3878 | next_deferred_force_submit += max_defer_interval/3; | |
3879 | } | |
3880 | ||
3881 | // Now Resize the shards | |
3882 | _resize_shards(interval_stats_trim); | |
91327a77 | 3883 | interval_stats_trim = false; |
31f18b77 | 3884 | |
91327a77 | 3885 | store->_update_cache_logger(); |
11fdf7f2 TL |
3886 | auto wait = ceph::make_timespan( |
3887 | store->cct->_conf->bluestore_cache_trim_interval); | |
3888 | cond.wait_for(l, wait); | |
7c673cae | 3889 | } |
9f95a23c TL |
3890 | // do final dump |
3891 | store->_record_allocation_stats(); | |
7c673cae FG |
3892 | stop = false; |
3893 | return NULL; | |
3894 | } | |
3895 | ||
91327a77 AA |
3896 | void BlueStore::MempoolThread::_adjust_cache_settings() |
3897 | { | |
11fdf7f2 TL |
3898 | if (binned_kv_cache != nullptr) { |
3899 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
3900 | } | |
3901 | meta_cache->set_cache_ratio(store->cache_meta_ratio); | |
3902 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
91327a77 AA |
3903 | } |
3904 | ||
9f95a23c | 3905 | void BlueStore::MempoolThread::_resize_shards(bool interval_stats) |
91327a77 | 3906 | { |
9f95a23c TL |
3907 | size_t onode_shards = store->onode_cache_shards.size(); |
3908 | size_t buffer_shards = store->buffer_cache_shards.size(); | |
91327a77 | 3909 | int64_t kv_used = store->db->get_cache_usage(); |
11fdf7f2 TL |
3910 | int64_t meta_used = meta_cache->_get_used_bytes(); |
3911 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
3912 | |
3913 | uint64_t cache_size = store->cache_size; | |
3914 | int64_t kv_alloc = | |
11fdf7f2 | 3915 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
91327a77 | 3916 | int64_t meta_alloc = |
11fdf7f2 | 3917 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 3918 | int64_t data_alloc = |
11fdf7f2 | 3919 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 3920 | |
eafe8130 TL |
3921 | if (pcm != nullptr && binned_kv_cache != nullptr) { |
3922 | cache_size = pcm->get_tuned_mem(); | |
11fdf7f2 TL |
3923 | kv_alloc = binned_kv_cache->get_committed_size(); |
3924 | meta_alloc = meta_cache->get_committed_size(); | |
3925 | data_alloc = data_cache->get_committed_size(); | |
91327a77 AA |
3926 | } |
3927 | ||
3928 | if (interval_stats) { | |
9f95a23c | 3929 | dout(5) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
3930 | << " kv_alloc: " << kv_alloc |
3931 | << " kv_used: " << kv_used | |
3932 | << " meta_alloc: " << meta_alloc | |
3933 | << " meta_used: " << meta_used | |
3934 | << " data_alloc: " << data_alloc | |
3935 | << " data_used: " << data_used << dendl; | |
3936 | } else { | |
9f95a23c | 3937 | dout(20) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
3938 | << " kv_alloc: " << kv_alloc |
3939 | << " kv_used: " << kv_used | |
3940 | << " meta_alloc: " << meta_alloc | |
3941 | << " meta_used: " << meta_used | |
3942 | << " data_alloc: " << data_alloc | |
3943 | << " data_used: " << data_used << dendl; | |
3944 | } | |
3945 | ||
3946 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
9f95a23c TL |
3947 | (meta_alloc / (double) onode_shards) / meta_cache->get_bytes_per_onode()); |
3948 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / buffer_shards); | |
91327a77 | 3949 | |
9f95a23c | 3950 | dout(30) << __func__ << " max_shard_onodes: " << max_shard_onodes |
91327a77 AA |
3951 | << " max_shard_buffer: " << max_shard_buffer << dendl; |
3952 | ||
9f95a23c TL |
3953 | for (auto i : store->onode_cache_shards) { |
3954 | i->set_max(max_shard_onodes); | |
3955 | } | |
3956 | for (auto i : store->buffer_cache_shards) { | |
3957 | i->set_max(max_shard_buffer); | |
91327a77 AA |
3958 | } |
3959 | } | |
3960 | ||
92f5a8d4 TL |
3961 | void BlueStore::MempoolThread::_update_cache_settings() |
3962 | { | |
3963 | // Nothing to do if pcm is not used. | |
3964 | if (pcm == nullptr) { | |
3965 | return; | |
3966 | } | |
3967 | ||
92f5a8d4 TL |
3968 | uint64_t target = store->osd_memory_target; |
3969 | uint64_t base = store->osd_memory_base; | |
3970 | uint64_t min = store->osd_memory_cache_min; | |
3971 | uint64_t max = min; | |
3972 | double fragmentation = store->osd_memory_expected_fragmentation; | |
3973 | ||
3974 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
3975 | if (ltarget > base + min) { | |
3976 | max = ltarget - base; | |
3977 | } | |
3978 | ||
3979 | // set pcm cache levels | |
3980 | pcm->set_target_memory(target); | |
3981 | pcm->set_min_memory(min); | |
3982 | pcm->set_max_memory(max); | |
3983 | ||
9f95a23c | 3984 | dout(5) << __func__ << " updated pcm target: " << target |
92f5a8d4 TL |
3985 | << " pcm min: " << min |
3986 | << " pcm max: " << max | |
3987 | << dendl; | |
3988 | } | |
3989 | ||
7c673cae FG |
3990 | // ======================================================= |
3991 | ||
31f18b77 FG |
3992 | // OmapIteratorImpl |
3993 | ||
3994 | #undef dout_prefix | |
3995 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
3996 | ||
3997 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
3998 | CollectionRef c, OnodeRef o, KeyValueDB::Iterator it) | |
3999 | : c(c), o(o), it(it) | |
4000 | { | |
9f95a23c | 4001 | std::shared_lock l(c->lock); |
31f18b77 | 4002 | if (o->onode.has_omap()) { |
9f95a23c TL |
4003 | o->get_omap_key(string(), &head); |
4004 | o->get_omap_tail(&tail); | |
31f18b77 FG |
4005 | it->lower_bound(head); |
4006 | } | |
4007 | } | |
4008 | ||
11fdf7f2 TL |
4009 | string BlueStore::OmapIteratorImpl::_stringify() const |
4010 | { | |
4011 | stringstream s; | |
4012 | s << " omap_iterator(cid = " << c->cid | |
4013 | <<", oid = " << o->oid << ")"; | |
4014 | return s.str(); | |
4015 | } | |
4016 | ||
31f18b77 FG |
4017 | int BlueStore::OmapIteratorImpl::seek_to_first() |
4018 | { | |
9f95a23c | 4019 | std::shared_lock l(c->lock); |
11fdf7f2 | 4020 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4021 | if (o->onode.has_omap()) { |
4022 | it->lower_bound(head); | |
4023 | } else { | |
4024 | it = KeyValueDB::Iterator(); | |
4025 | } | |
494da23a TL |
4026 | c->store->log_latency( |
4027 | __func__, | |
11fdf7f2 TL |
4028 | l_bluestore_omap_seek_to_first_lat, |
4029 | mono_clock::now() - start1, | |
494da23a | 4030 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 | 4031 | |
31f18b77 FG |
4032 | return 0; |
4033 | } | |
4034 | ||
4035 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
4036 | { | |
9f95a23c | 4037 | std::shared_lock l(c->lock); |
11fdf7f2 | 4038 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4039 | if (o->onode.has_omap()) { |
4040 | string key; | |
9f95a23c | 4041 | o->get_omap_key(after, &key); |
31f18b77 FG |
4042 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " |
4043 | << pretty_binary_string(key) << dendl; | |
4044 | it->upper_bound(key); | |
4045 | } else { | |
4046 | it = KeyValueDB::Iterator(); | |
4047 | } | |
11fdf7f2 | 4048 | c->store->log_latency_fn( |
494da23a | 4049 | __func__, |
11fdf7f2 TL |
4050 | l_bluestore_omap_upper_bound_lat, |
4051 | mono_clock::now() - start1, | |
494da23a | 4052 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4053 | [&] (const ceph::timespan& lat) { |
494da23a | 4054 | return ", after = " + after + |
11fdf7f2 TL |
4055 | _stringify(); |
4056 | } | |
4057 | ); | |
31f18b77 FG |
4058 | return 0; |
4059 | } | |
4060 | ||
4061 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
4062 | { | |
9f95a23c | 4063 | std::shared_lock l(c->lock); |
11fdf7f2 | 4064 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4065 | if (o->onode.has_omap()) { |
4066 | string key; | |
9f95a23c | 4067 | o->get_omap_key(to, &key); |
31f18b77 FG |
4068 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " |
4069 | << pretty_binary_string(key) << dendl; | |
4070 | it->lower_bound(key); | |
4071 | } else { | |
4072 | it = KeyValueDB::Iterator(); | |
4073 | } | |
11fdf7f2 | 4074 | c->store->log_latency_fn( |
494da23a | 4075 | __func__, |
11fdf7f2 TL |
4076 | l_bluestore_omap_lower_bound_lat, |
4077 | mono_clock::now() - start1, | |
494da23a | 4078 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4079 | [&] (const ceph::timespan& lat) { |
494da23a | 4080 | return ", to = " + to + |
11fdf7f2 TL |
4081 | _stringify(); |
4082 | } | |
4083 | ); | |
31f18b77 FG |
4084 | return 0; |
4085 | } | |
4086 | ||
4087 | bool BlueStore::OmapIteratorImpl::valid() | |
4088 | { | |
9f95a23c | 4089 | std::shared_lock l(c->lock); |
31f18b77 | 4090 | bool r = o->onode.has_omap() && it && it->valid() && |
494da23a | 4091 | it->raw_key().second < tail; |
31f18b77 FG |
4092 | if (it && it->valid()) { |
4093 | ldout(c->store->cct,20) << __func__ << " is at " | |
4094 | << pretty_binary_string(it->raw_key().second) | |
4095 | << dendl; | |
4096 | } | |
4097 | return r; | |
4098 | } | |
4099 | ||
11fdf7f2 | 4100 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 4101 | { |
11fdf7f2 | 4102 | int r = -1; |
9f95a23c | 4103 | std::shared_lock l(c->lock); |
11fdf7f2 | 4104 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4105 | if (o->onode.has_omap()) { |
4106 | it->next(); | |
11fdf7f2 | 4107 | r = 0; |
31f18b77 | 4108 | } |
494da23a TL |
4109 | c->store->log_latency( |
4110 | __func__, | |
11fdf7f2 TL |
4111 | l_bluestore_omap_next_lat, |
4112 | mono_clock::now() - start1, | |
494da23a | 4113 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 TL |
4114 | |
4115 | return r; | |
31f18b77 FG |
4116 | } |
4117 | ||
4118 | string BlueStore::OmapIteratorImpl::key() | |
4119 | { | |
9f95a23c | 4120 | std::shared_lock l(c->lock); |
11fdf7f2 | 4121 | ceph_assert(it->valid()); |
31f18b77 FG |
4122 | string db_key = it->raw_key().second; |
4123 | string user_key; | |
9f95a23c | 4124 | o->decode_omap_key(db_key, &user_key); |
494da23a | 4125 | |
31f18b77 FG |
4126 | return user_key; |
4127 | } | |
4128 | ||
4129 | bufferlist BlueStore::OmapIteratorImpl::value() | |
4130 | { | |
9f95a23c | 4131 | std::shared_lock l(c->lock); |
11fdf7f2 | 4132 | ceph_assert(it->valid()); |
31f18b77 FG |
4133 | return it->value(); |
4134 | } | |
4135 | ||
4136 | ||
4137 | // ===================================== | |
4138 | ||
7c673cae FG |
4139 | #undef dout_prefix |
4140 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
9f95a23c TL |
4141 | #undef dout_context |
4142 | #define dout_context cct | |
7c673cae FG |
4143 | |
4144 | ||
4145 | static void aio_cb(void *priv, void *priv2) | |
4146 | { | |
4147 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4148 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4149 | c->aio_finish(store); | |
4150 | } | |
4151 | ||
11fdf7f2 TL |
4152 | static void discard_cb(void *priv, void *priv2) |
4153 | { | |
4154 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4155 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4156 | store->handle_discard(*tmp); | |
4157 | } | |
4158 | ||
4159 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4160 | { | |
4161 | dout(10) << __func__ << dendl; | |
4162 | ceph_assert(alloc); | |
4163 | alloc->release(to_release); | |
4164 | } | |
4165 | ||
7c673cae | 4166 | BlueStore::BlueStore(CephContext *cct, const string& path) |
9f95a23c | 4167 | : BlueStore(cct, path, 0) {} |
7c673cae FG |
4168 | |
4169 | BlueStore::BlueStore(CephContext *cct, | |
4170 | const string& path, | |
4171 | uint64_t _min_alloc_size) | |
4172 | : ObjectStore(cct, path), | |
9f95a23c | 4173 | throttle(cct), |
11fdf7f2 | 4174 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4175 | kv_sync_thread(this), |
31f18b77 | 4176 | kv_finalize_thread(this), |
7c673cae FG |
4177 | min_alloc_size(_min_alloc_size), |
4178 | min_alloc_size_order(ctz(_min_alloc_size)), | |
4179 | mempool_thread(this) | |
4180 | { | |
4181 | _init_logger(); | |
11fdf7f2 | 4182 | cct->_conf.add_observer(this); |
7c673cae | 4183 | set_cache_shards(1); |
7c673cae FG |
4184 | } |
4185 | ||
4186 | BlueStore::~BlueStore() | |
4187 | { | |
11fdf7f2 | 4188 | cct->_conf.remove_observer(this); |
7c673cae | 4189 | _shutdown_logger(); |
11fdf7f2 TL |
4190 | ceph_assert(!mounted); |
4191 | ceph_assert(db == NULL); | |
4192 | ceph_assert(bluefs == NULL); | |
4193 | ceph_assert(fsid_fd < 0); | |
4194 | ceph_assert(path_fd < 0); | |
9f95a23c TL |
4195 | for (auto i : onode_cache_shards) { |
4196 | delete i; | |
4197 | } | |
4198 | for (auto i : buffer_cache_shards) { | |
7c673cae FG |
4199 | delete i; |
4200 | } | |
9f95a23c TL |
4201 | onode_cache_shards.clear(); |
4202 | buffer_cache_shards.clear(); | |
7c673cae FG |
4203 | } |
4204 | ||
4205 | const char **BlueStore::get_tracked_conf_keys() const | |
4206 | { | |
4207 | static const char* KEYS[] = { | |
4208 | "bluestore_csum_type", | |
4209 | "bluestore_compression_mode", | |
4210 | "bluestore_compression_algorithm", | |
4211 | "bluestore_compression_min_blob_size", | |
4212 | "bluestore_compression_min_blob_size_ssd", | |
4213 | "bluestore_compression_min_blob_size_hdd", | |
4214 | "bluestore_compression_max_blob_size", | |
4215 | "bluestore_compression_max_blob_size_ssd", | |
4216 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4217 | "bluestore_compression_required_ratio", |
7c673cae FG |
4218 | "bluestore_max_alloc_size", |
4219 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4220 | "bluestore_prefer_deferred_size_hdd", |
4221 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4222 | "bluestore_deferred_batch_ops", |
4223 | "bluestore_deferred_batch_ops_hdd", | |
4224 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4225 | "bluestore_throttle_bytes", |
4226 | "bluestore_throttle_deferred_bytes", | |
4227 | "bluestore_throttle_cost_per_io_hdd", | |
4228 | "bluestore_throttle_cost_per_io_ssd", | |
4229 | "bluestore_throttle_cost_per_io", | |
4230 | "bluestore_max_blob_size", | |
4231 | "bluestore_max_blob_size_ssd", | |
4232 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4233 | "osd_memory_target", |
4234 | "osd_memory_target_cgroup_limit_ratio", | |
4235 | "osd_memory_base", | |
4236 | "osd_memory_cache_min", | |
92f5a8d4 | 4237 | "osd_memory_expected_fragmentation", |
11fdf7f2 TL |
4238 | "bluestore_cache_autotune", |
4239 | "bluestore_cache_autotune_interval", | |
81eedcae | 4240 | "bluestore_warn_on_legacy_statfs", |
9f95a23c TL |
4241 | "bluestore_warn_on_no_per_pool_omap", |
4242 | "bluestore_max_defer_interval", | |
7c673cae FG |
4243 | NULL |
4244 | }; | |
4245 | return KEYS; | |
4246 | } | |
4247 | ||
11fdf7f2 | 4248 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4249 | const std::set<std::string> &changed) |
4250 | { | |
eafe8130 | 4251 | if (changed.count("bluestore_warn_on_legacy_statfs")) { |
81eedcae TL |
4252 | _check_legacy_statfs_alert(); |
4253 | } | |
9f95a23c TL |
4254 | if (changed.count("bluestore_warn_on_no_per_pool_omap")) { |
4255 | _check_no_per_pool_omap_alert(); | |
4256 | } | |
81eedcae | 4257 | |
7c673cae FG |
4258 | if (changed.count("bluestore_csum_type")) { |
4259 | _set_csum(); | |
4260 | } | |
4261 | if (changed.count("bluestore_compression_mode") || | |
4262 | changed.count("bluestore_compression_algorithm") || | |
4263 | changed.count("bluestore_compression_min_blob_size") || | |
4264 | changed.count("bluestore_compression_max_blob_size")) { | |
4265 | if (bdev) { | |
4266 | _set_compression(); | |
4267 | } | |
4268 | } | |
4269 | if (changed.count("bluestore_max_blob_size") || | |
4270 | changed.count("bluestore_max_blob_size_ssd") || | |
4271 | changed.count("bluestore_max_blob_size_hdd")) { | |
4272 | if (bdev) { | |
4273 | // only after startup | |
4274 | _set_blob_size(); | |
4275 | } | |
4276 | } | |
4277 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4278 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4279 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4280 | changed.count("bluestore_max_alloc_size") || |
4281 | changed.count("bluestore_deferred_batch_ops") || | |
4282 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4283 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4284 | if (bdev) { | |
4285 | // only after startup | |
4286 | _set_alloc_sizes(); | |
4287 | } | |
4288 | } | |
4289 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4290 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4291 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4292 | if (bdev) { | |
4293 | _set_throttle_params(); | |
4294 | } | |
4295 | } | |
9f95a23c TL |
4296 | if (changed.count("bluestore_throttle_bytes") || |
4297 | changed.count("bluestore_throttle_deferred_bytes") || | |
4298 | changed.count("bluestore_throttle_trace_rate")) { | |
4299 | throttle.reset_throttle(conf); | |
7c673cae | 4300 | } |
9f95a23c TL |
4301 | if (changed.count("bluestore_max_defer_interval")) { |
4302 | if (bdev) { | |
4303 | _set_max_defer_interval(); | |
4304 | } | |
7c673cae | 4305 | } |
92f5a8d4 TL |
4306 | if (changed.count("osd_memory_target") || |
4307 | changed.count("osd_memory_base") || | |
4308 | changed.count("osd_memory_cache_min") || | |
4309 | changed.count("osd_memory_expected_fragmentation")) { | |
4310 | _update_osd_memory_options(); | |
4311 | } | |
7c673cae FG |
4312 | } |
4313 | ||
4314 | void BlueStore::_set_compression() | |
4315 | { | |
224ce89b WB |
4316 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4317 | if (m) { | |
11fdf7f2 | 4318 | _clear_compression_alert(); |
224ce89b WB |
4319 | comp_mode = *m; |
4320 | } else { | |
4321 | derr << __func__ << " unrecognized value '" | |
4322 | << cct->_conf->bluestore_compression_mode | |
4323 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4324 | << dendl; | |
4325 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4326 | string s("unknown mode: "); |
4327 | s += cct->_conf->bluestore_compression_mode; | |
4328 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4329 | } |
4330 | ||
4331 | compressor = nullptr; | |
4332 | ||
3efd9988 FG |
4333 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4334 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4335 | } else { |
11fdf7f2 | 4336 | ceph_assert(bdev); |
9f95a23c | 4337 | if (_use_rotational_settings()) { |
7c673cae FG |
4338 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; |
4339 | } else { | |
4340 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4341 | } | |
4342 | } | |
4343 | ||
4344 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4345 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4346 | } else { | |
11fdf7f2 | 4347 | ceph_assert(bdev); |
9f95a23c | 4348 | if (_use_rotational_settings()) { |
7c673cae FG |
4349 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; |
4350 | } else { | |
4351 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4352 | } | |
4353 | } | |
4354 | ||
7c673cae FG |
4355 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4356 | if (!alg_name.empty()) { | |
4357 | compressor = Compressor::create(cct, alg_name); | |
4358 | if (!compressor) { | |
4359 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4360 | << dendl; | |
11fdf7f2 | 4361 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4362 | } |
4363 | } | |
4364 | ||
4365 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4366 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4367 | << " min_blob " << comp_min_blob_size |
4368 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4369 | << dendl; |
4370 | } | |
4371 | ||
4372 | void BlueStore::_set_csum() | |
4373 | { | |
4374 | csum_type = Checksummer::CSUM_NONE; | |
4375 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4376 | if (t > Checksummer::CSUM_NONE) | |
4377 | csum_type = t; | |
4378 | ||
4379 | dout(10) << __func__ << " csum_type " | |
4380 | << Checksummer::get_csum_type_string(csum_type) | |
4381 | << dendl; | |
4382 | } | |
4383 | ||
4384 | void BlueStore::_set_throttle_params() | |
4385 | { | |
4386 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4387 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4388 | } else { | |
11fdf7f2 | 4389 | ceph_assert(bdev); |
9f95a23c | 4390 | if (_use_rotational_settings()) { |
7c673cae FG |
4391 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; |
4392 | } else { | |
4393 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4394 | } | |
4395 | } | |
4396 | ||
4397 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4398 | << dendl; | |
4399 | } | |
4400 | void BlueStore::_set_blob_size() | |
4401 | { | |
4402 | if (cct->_conf->bluestore_max_blob_size) { | |
4403 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4404 | } else { | |
11fdf7f2 | 4405 | ceph_assert(bdev); |
9f95a23c | 4406 | if (_use_rotational_settings()) { |
7c673cae FG |
4407 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; |
4408 | } else { | |
4409 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4410 | } | |
4411 | } | |
4412 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4413 | << std::dec << dendl; | |
4414 | } | |
4415 | ||
92f5a8d4 TL |
4416 | void BlueStore::_update_osd_memory_options() |
4417 | { | |
4418 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4419 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
4420 | osd_memory_expected_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
4421 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
4422 | config_changed++; | |
4423 | dout(10) << __func__ | |
4424 | << " osd_memory_target " << osd_memory_target | |
4425 | << " osd_memory_base " << osd_memory_base | |
4426 | << " osd_memory_expected_fragmentation " << osd_memory_expected_fragmentation | |
4427 | << " osd_memory_cache_min " << osd_memory_cache_min | |
4428 | << dendl; | |
4429 | } | |
4430 | ||
11fdf7f2 | 4431 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4432 | { |
11fdf7f2 TL |
4433 | ceph_assert(bdev); |
4434 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4435 | cache_autotune_interval = |
11fdf7f2 TL |
4436 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
4437 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4438 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4439 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4440 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4441 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4442 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4443 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4444 | |
224ce89b WB |
4445 | if (cct->_conf->bluestore_cache_size) { |
4446 | cache_size = cct->_conf->bluestore_cache_size; | |
4447 | } else { | |
4448 | // choose global cache size based on backend type | |
9f95a23c | 4449 | if (_use_rotational_settings()) { |
224ce89b WB |
4450 | cache_size = cct->_conf->bluestore_cache_size_hdd; |
4451 | } else { | |
4452 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4453 | } | |
4454 | } | |
31f18b77 | 4455 | |
91327a77 | 4456 | cache_meta_ratio = cct->_conf->bluestore_cache_meta_ratio; |
224ce89b | 4457 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4458 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4459 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4460 | return -EINVAL; |
4461 | } | |
91327a77 AA |
4462 | |
4463 | cache_kv_ratio = cct->_conf->bluestore_cache_kv_ratio; | |
224ce89b | 4464 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4465 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4466 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4467 | return -EINVAL; |
4468 | } | |
91327a77 | 4469 | |
31f18b77 | 4470 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4471 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4472 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4473 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4474 | << dendl; | |
31f18b77 FG |
4475 | return -EINVAL; |
4476 | } | |
91327a77 AA |
4477 | |
4478 | cache_data_ratio = | |
4479 | (double)1.0 - (double)cache_meta_ratio - (double)cache_kv_ratio; | |
31f18b77 FG |
4480 | if (cache_data_ratio < 0) { |
4481 | // deal with floating point imprecision | |
4482 | cache_data_ratio = 0; | |
4483 | } | |
91327a77 | 4484 | |
224ce89b WB |
4485 | dout(1) << __func__ << " cache_size " << cache_size |
4486 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4487 | << " kv " << cache_kv_ratio |
4488 | << " data " << cache_data_ratio | |
4489 | << dendl; | |
4490 | return 0; | |
4491 | } | |
4492 | ||
3efd9988 FG |
4493 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4494 | { | |
4495 | bluestore_bdev_label_t label; | |
4496 | string p = path + "/block"; | |
4497 | int r = _read_bdev_label(cct, p, &label); | |
4498 | if (r < 0) { | |
4499 | return ObjectStore::write_meta(key, value); | |
4500 | } | |
4501 | label.meta[key] = value; | |
4502 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4503 | ceph_assert(r == 0); |
3efd9988 FG |
4504 | return ObjectStore::write_meta(key, value); |
4505 | } | |
4506 | ||
4507 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4508 | { | |
4509 | bluestore_bdev_label_t label; | |
4510 | string p = path + "/block"; | |
4511 | int r = _read_bdev_label(cct, p, &label); | |
4512 | if (r < 0) { | |
4513 | return ObjectStore::read_meta(key, value); | |
4514 | } | |
4515 | auto i = label.meta.find(key); | |
4516 | if (i == label.meta.end()) { | |
4517 | return ObjectStore::read_meta(key, value); | |
4518 | } | |
4519 | *value = i->second; | |
4520 | return 0; | |
4521 | } | |
4522 | ||
7c673cae FG |
4523 | void BlueStore::_init_logger() |
4524 | { | |
4525 | PerfCountersBuilder b(cct, "bluestore", | |
4526 | l_bluestore_first, l_bluestore_last); | |
4527 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
4528 | "Average kv_thread flush latency", | |
4529 | "fl_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4530 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
4531 | "Average kv_thread commit latency"); | |
11fdf7f2 TL |
4532 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", |
4533 | "Average kv_sync thread latency", | |
4534 | "ks_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4535 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
4536 | "Average kv_finalize thread latency", | |
4537 | "kf_l", PerfCountersBuilder::PRIO_INTERESTING); | |
7c673cae FG |
4538 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
4539 | "Average prepare state latency"); | |
4540 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", | |
4541 | "Average aio_wait state latency", | |
4542 | "io_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4543 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", | |
4544 | "Average io_done state latency"); | |
4545 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", | |
4546 | "Average kv_queued state latency"); | |
4547 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", | |
4548 | "Average kv_commiting state latency"); | |
4549 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", | |
4550 | "Average kv_done state latency"); | |
4551 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", | |
4552 | "Average deferred_queued state latency"); | |
4553 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", | |
4554 | "Average aio_wait state latency"); | |
4555 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", | |
4556 | "Average cleanup state latency"); | |
4557 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
4558 | "Average finishing state latency"); | |
4559 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
4560 | "Average done state latency"); | |
4561 | b.add_time_avg(l_bluestore_throttle_lat, "throttle_lat", | |
4562 | "Average submit throttle latency", | |
4563 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4564 | b.add_time_avg(l_bluestore_submit_lat, "submit_lat", | |
4565 | "Average submit latency", | |
4566 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4567 | b.add_time_avg(l_bluestore_commit_lat, "commit_lat", | |
4568 | "Average commit latency", | |
4569 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4570 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
4571 | "Average read latency", | |
4572 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4573 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", | |
4574 | "Average read onode metadata latency"); | |
4575 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", | |
4576 | "Average read latency"); | |
4577 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
4578 | "Average compress latency"); | |
4579 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
4580 | "Average decompress latency"); | |
4581 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", | |
4582 | "Average checksum latency"); | |
4583 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
4584 | "Sum for beneficial compress ops"); | |
4585 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
4586 | "Sum for compress ops rejected due to low net gain of space"); | |
4587 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", | |
11fdf7f2 | 4588 | "Sum for write-op padded bytes", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4589 | b.add_u64_counter(l_bluestore_deferred_write_ops, "deferred_write_ops", |
4590 | "Sum for deferred write op"); | |
4591 | b.add_u64_counter(l_bluestore_deferred_write_bytes, "deferred_write_bytes", | |
11fdf7f2 | 4592 | "Sum for deferred write bytes", "def", 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4593 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", |
4594 | "Sum for write penalty read ops"); | |
4595 | b.add_u64(l_bluestore_allocated, "bluestore_allocated", | |
4596 | "Sum for allocated bytes"); | |
4597 | b.add_u64(l_bluestore_stored, "bluestore_stored", | |
4598 | "Sum for stored bytes"); | |
4599 | b.add_u64(l_bluestore_compressed, "bluestore_compressed", | |
92f5a8d4 TL |
4600 | "Sum for stored compressed bytes", |
4601 | "c", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4602 | b.add_u64(l_bluestore_compressed_allocated, "bluestore_compressed_allocated", |
92f5a8d4 TL |
4603 | "Sum for bytes allocated for compressed data", |
4604 | "c_a", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4605 | b.add_u64(l_bluestore_compressed_original, "bluestore_compressed_original", |
92f5a8d4 TL |
4606 | "Sum for original bytes that were compressed", |
4607 | "c_o", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae FG |
4608 | b.add_u64(l_bluestore_onodes, "bluestore_onodes", |
4609 | "Number of onodes in cache"); | |
9f95a23c TL |
4610 | b.add_u64(l_bluestore_pinned_onodes, "bluestore_pinned_onodes", |
4611 | "Number of pinned onodes in cache"); | |
7c673cae FG |
4612 | b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits", |
4613 | "Sum for onode-lookups hit in the cache"); | |
4614 | b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses", | |
4615 | "Sum for onode-lookups missed in the cache"); | |
4616 | b.add_u64_counter(l_bluestore_onode_shard_hits, "bluestore_onode_shard_hits", | |
4617 | "Sum for onode-shard lookups hit in the cache"); | |
4618 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
4619 | "bluestore_onode_shard_misses", | |
4620 | "Sum for onode-shard lookups missed in the cache"); | |
4621 | b.add_u64(l_bluestore_extents, "bluestore_extents", | |
4622 | "Number of extents in cache"); | |
4623 | b.add_u64(l_bluestore_blobs, "bluestore_blobs", | |
4624 | "Number of blobs in cache"); | |
4625 | b.add_u64(l_bluestore_buffers, "bluestore_buffers", | |
4626 | "Number of buffers in cache"); | |
4627 | b.add_u64(l_bluestore_buffer_bytes, "bluestore_buffer_bytes", | |
11fdf7f2 | 4628 | "Number of buffer bytes in cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4629 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "bluestore_buffer_hit_bytes", |
11fdf7f2 | 4630 | "Sum for bytes of read hit in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4631 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "bluestore_buffer_miss_bytes", |
11fdf7f2 | 4632 | "Sum for bytes of read missed in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4633 | |
4634 | b.add_u64_counter(l_bluestore_write_big, "bluestore_write_big", | |
4635 | "Large aligned writes into fresh blobs"); | |
4636 | b.add_u64_counter(l_bluestore_write_big_bytes, "bluestore_write_big_bytes", | |
11fdf7f2 | 4637 | "Large aligned writes into fresh blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4638 | b.add_u64_counter(l_bluestore_write_big_blobs, "bluestore_write_big_blobs", |
4639 | "Large aligned writes into fresh blobs (blobs)"); | |
4640 | b.add_u64_counter(l_bluestore_write_small, "bluestore_write_small", | |
4641 | "Small writes into existing or sparse small blobs"); | |
4642 | b.add_u64_counter(l_bluestore_write_small_bytes, "bluestore_write_small_bytes", | |
11fdf7f2 | 4643 | "Small writes into existing or sparse small blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4644 | b.add_u64_counter(l_bluestore_write_small_unused, |
4645 | "bluestore_write_small_unused", | |
4646 | "Small writes into unused portion of existing blob"); | |
4647 | b.add_u64_counter(l_bluestore_write_small_deferred, | |
4648 | "bluestore_write_small_deferred", | |
4649 | "Small overwrites using deferred"); | |
4650 | b.add_u64_counter(l_bluestore_write_small_pre_read, | |
4651 | "bluestore_write_small_pre_read", | |
4652 | "Small writes that required we read some data (possibly " | |
4653 | "cached) to fill out the block"); | |
4654 | b.add_u64_counter(l_bluestore_write_small_new, "bluestore_write_small_new", | |
4655 | "Small write into new (sparse) blob"); | |
4656 | ||
4657 | b.add_u64_counter(l_bluestore_txc, "bluestore_txc", "Transactions committed"); | |
4658 | b.add_u64_counter(l_bluestore_onode_reshard, "bluestore_onode_reshard", | |
4659 | "Onode extent map reshard events"); | |
4660 | b.add_u64_counter(l_bluestore_blob_split, "bluestore_blob_split", | |
4661 | "Sum for blob splitting due to resharding"); | |
4662 | b.add_u64_counter(l_bluestore_extent_compress, "bluestore_extent_compress", | |
4663 | "Sum for extents that have been removed due to compression"); | |
4664 | b.add_u64_counter(l_bluestore_gc_merged, "bluestore_gc_merged", | |
4665 | "Sum for extents that have been merged due to garbage " | |
4666 | "collection"); | |
b32b8144 FG |
4667 | b.add_u64_counter(l_bluestore_read_eio, "bluestore_read_eio", |
4668 | "Read EIO errors propagated to high level callers"); | |
f64942e4 AA |
4669 | b.add_u64_counter(l_bluestore_reads_with_retries, "bluestore_reads_with_retries", |
4670 | "Read operations that required at least one retry due to failed checksum validation"); | |
a8e16298 TL |
4671 | b.add_u64(l_bluestore_fragmentation, "bluestore_fragmentation_micros", |
4672 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
11fdf7f2 TL |
4673 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
4674 | "Average omap iterator seek_to_first call latency"); | |
4675 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", | |
4676 | "Average omap iterator upper_bound call latency"); | |
4677 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", | |
4678 | "Average omap iterator lower_bound call latency"); | |
4679 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", | |
4680 | "Average omap iterator next call latency"); | |
494da23a TL |
4681 | b.add_time_avg(l_bluestore_clist_lat, "clist_lat", |
4682 | "Average collection listing latency"); | |
7c673cae FG |
4683 | logger = b.create_perf_counters(); |
4684 | cct->get_perfcounters_collection()->add(logger); | |
4685 | } | |
4686 | ||
4687 | int BlueStore::_reload_logger() | |
4688 | { | |
4689 | struct store_statfs_t store_statfs; | |
7c673cae | 4690 | int r = statfs(&store_statfs); |
11fdf7f2 | 4691 | if (r >= 0) { |
7c673cae | 4692 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
4693 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
4694 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
4695 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
4696 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
4697 | } |
4698 | return r; | |
4699 | } | |
4700 | ||
4701 | void BlueStore::_shutdown_logger() | |
4702 | { | |
4703 | cct->get_perfcounters_collection()->remove(logger); | |
4704 | delete logger; | |
4705 | } | |
4706 | ||
4707 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
4708 | uuid_d *fsid) | |
4709 | { | |
4710 | bluestore_bdev_label_t label; | |
4711 | int r = _read_bdev_label(cct, path, &label); | |
4712 | if (r < 0) | |
4713 | return r; | |
4714 | *fsid = label.osd_uuid; | |
4715 | return 0; | |
4716 | } | |
4717 | ||
4718 | int BlueStore::_open_path() | |
4719 | { | |
b32b8144 | 4720 | // sanity check(s) |
11fdf7f2 | 4721 | ceph_assert(path_fd < 0); |
91327a77 | 4722 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
4723 | if (path_fd < 0) { |
4724 | int r = -errno; | |
4725 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
4726 | << dendl; | |
4727 | return r; | |
4728 | } | |
4729 | return 0; | |
4730 | } | |
4731 | ||
4732 | void BlueStore::_close_path() | |
4733 | { | |
4734 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
4735 | path_fd = -1; | |
4736 | } | |
4737 | ||
3efd9988 FG |
4738 | int BlueStore::_write_bdev_label(CephContext *cct, |
4739 | string path, bluestore_bdev_label_t label) | |
7c673cae FG |
4740 | { |
4741 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
4742 | bufferlist bl; | |
11fdf7f2 | 4743 | encode(label, bl); |
7c673cae | 4744 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
4745 | encode(crc, bl); |
4746 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
4747 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
4748 | z.zero(); | |
4749 | bl.append(std::move(z)); | |
4750 | ||
91327a77 | 4751 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC)); |
7c673cae FG |
4752 | if (fd < 0) { |
4753 | fd = -errno; | |
4754 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
4755 | << dendl; | |
4756 | return fd; | |
4757 | } | |
4758 | int r = bl.write_fd(fd); | |
4759 | if (r < 0) { | |
4760 | derr << __func__ << " failed to write to " << path | |
4761 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 4762 | goto out; |
7c673cae | 4763 | } |
3efd9988 FG |
4764 | r = ::fsync(fd); |
4765 | if (r < 0) { | |
4766 | derr << __func__ << " failed to fsync " << path | |
4767 | << ": " << cpp_strerror(r) << dendl; | |
4768 | } | |
11fdf7f2 | 4769 | out: |
7c673cae FG |
4770 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
4771 | return r; | |
4772 | } | |
4773 | ||
4774 | int BlueStore::_read_bdev_label(CephContext* cct, string path, | |
4775 | bluestore_bdev_label_t *label) | |
4776 | { | |
4777 | dout(10) << __func__ << dendl; | |
91327a77 | 4778 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
4779 | if (fd < 0) { |
4780 | fd = -errno; | |
4781 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
4782 | << dendl; | |
4783 | return fd; | |
4784 | } | |
4785 | bufferlist bl; | |
4786 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
4787 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
4788 | if (r < 0) { | |
4789 | derr << __func__ << " failed to read from " << path | |
4790 | << ": " << cpp_strerror(r) << dendl; | |
4791 | return r; | |
4792 | } | |
4793 | ||
4794 | uint32_t crc, expected_crc; | |
11fdf7f2 | 4795 | auto p = bl.cbegin(); |
7c673cae | 4796 | try { |
11fdf7f2 | 4797 | decode(*label, p); |
7c673cae FG |
4798 | bufferlist t; |
4799 | t.substr_of(bl, 0, p.get_off()); | |
4800 | crc = t.crc32c(-1); | |
11fdf7f2 | 4801 | decode(expected_crc, p); |
7c673cae FG |
4802 | } |
4803 | catch (buffer::error& e) { | |
b32b8144 | 4804 | dout(2) << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
4805 | << ": " << e.what() |
4806 | << dendl; | |
b32b8144 | 4807 | return -ENOENT; |
7c673cae FG |
4808 | } |
4809 | if (crc != expected_crc) { | |
4810 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
4811 | << " != actual " << crc << dendl; | |
4812 | return -EIO; | |
4813 | } | |
4814 | dout(10) << __func__ << " got " << *label << dendl; | |
4815 | return 0; | |
4816 | } | |
4817 | ||
4818 | int BlueStore::_check_or_set_bdev_label( | |
4819 | string path, uint64_t size, string desc, bool create) | |
4820 | { | |
4821 | bluestore_bdev_label_t label; | |
4822 | if (create) { | |
4823 | label.osd_uuid = fsid; | |
4824 | label.size = size; | |
4825 | label.btime = ceph_clock_now(); | |
4826 | label.description = desc; | |
3efd9988 | 4827 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
4828 | if (r < 0) |
4829 | return r; | |
4830 | } else { | |
4831 | int r = _read_bdev_label(cct, path, &label); | |
4832 | if (r < 0) | |
4833 | return r; | |
31f18b77 FG |
4834 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
4835 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
4836 | << " and fsid " << fsid << " check bypassed" << dendl; | |
4837 | } | |
4838 | else if (label.osd_uuid != fsid) { | |
7c673cae FG |
4839 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
4840 | << " does not match our fsid " << fsid << dendl; | |
4841 | return -EIO; | |
4842 | } | |
4843 | } | |
4844 | return 0; | |
4845 | } | |
4846 | ||
4847 | void BlueStore::_set_alloc_sizes(void) | |
4848 | { | |
7c673cae FG |
4849 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
4850 | ||
4851 | if (cct->_conf->bluestore_prefer_deferred_size) { | |
4852 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
4853 | } else { | |
11fdf7f2 | 4854 | ceph_assert(bdev); |
9f95a23c | 4855 | if (_use_rotational_settings()) { |
7c673cae FG |
4856 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; |
4857 | } else { | |
4858 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
4859 | } | |
4860 | } | |
4861 | ||
4862 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
4863 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
4864 | } else { | |
11fdf7f2 | 4865 | ceph_assert(bdev); |
9f95a23c | 4866 | if (_use_rotational_settings()) { |
7c673cae FG |
4867 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; |
4868 | } else { | |
4869 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
4870 | } | |
4871 | } | |
4872 | ||
4873 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 4874 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
4875 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
4876 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
4877 | << std::dec | |
4878 | << " deferred_batch_ops " << deferred_batch_ops | |
4879 | << dendl; | |
4880 | } | |
4881 | ||
4882 | int BlueStore::_open_bdev(bool create) | |
4883 | { | |
11fdf7f2 | 4884 | ceph_assert(bdev == NULL); |
7c673cae | 4885 | string p = path + "/block"; |
11fdf7f2 | 4886 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
4887 | int r = bdev->open(p); |
4888 | if (r < 0) | |
4889 | goto fail; | |
4890 | ||
11fdf7f2 TL |
4891 | if (create && cct->_conf->bdev_enable_discard) { |
4892 | bdev->discard(0, bdev->get_size()); | |
4893 | } | |
4894 | ||
7c673cae FG |
4895 | if (bdev->supported_bdev_label()) { |
4896 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
4897 | if (r < 0) | |
4898 | goto fail_close; | |
4899 | } | |
4900 | ||
4901 | // initialize global block parameters | |
4902 | block_size = bdev->get_block_size(); | |
4903 | block_mask = ~(block_size - 1); | |
4904 | block_size_order = ctz(block_size); | |
11fdf7f2 | 4905 | ceph_assert(block_size == 1u << block_size_order); |
9f95a23c | 4906 | _set_max_defer_interval(); |
224ce89b WB |
4907 | // and set cache_size based on device type |
4908 | r = _set_cache_sizes(); | |
4909 | if (r < 0) { | |
4910 | goto fail_close; | |
4911 | } | |
7c673cae FG |
4912 | return 0; |
4913 | ||
4914 | fail_close: | |
4915 | bdev->close(); | |
4916 | fail: | |
4917 | delete bdev; | |
4918 | bdev = NULL; | |
4919 | return r; | |
4920 | } | |
4921 | ||
11fdf7f2 TL |
4922 | void BlueStore::_validate_bdev() |
4923 | { | |
4924 | ceph_assert(bdev); | |
4925 | ceph_assert(min_alloc_size); // _get_odisk_reserved depends on that | |
4926 | uint64_t dev_size = bdev->get_size(); | |
4927 | if (dev_size < | |
4928 | _get_ondisk_reserved() + cct->_conf->bluestore_bluefs_min) { | |
4929 | dout(1) << __func__ << " main device size " << byte_u_t(dev_size) | |
4930 | << " is too small, disable bluestore_bluefs_min for now" | |
4931 | << dendl; | |
4932 | ceph_assert(dev_size >= _get_ondisk_reserved()); | |
4933 | ||
4934 | int r = cct->_conf.set_val("bluestore_bluefs_min", "0"); | |
4935 | ceph_assert(r == 0); | |
4936 | } | |
4937 | } | |
4938 | ||
7c673cae FG |
4939 | void BlueStore::_close_bdev() |
4940 | { | |
11fdf7f2 | 4941 | ceph_assert(bdev); |
7c673cae FG |
4942 | bdev->close(); |
4943 | delete bdev; | |
4944 | bdev = NULL; | |
4945 | } | |
4946 | ||
11fdf7f2 | 4947 | int BlueStore::_open_fm(KeyValueDB::Transaction t) |
7c673cae | 4948 | { |
11fdf7f2 TL |
4949 | ceph_assert(fm == NULL); |
4950 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); | |
4951 | ceph_assert(fm); | |
4952 | if (t) { | |
4953 | // create mode. initialize freespace | |
7c673cae | 4954 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
4955 | { |
4956 | bufferlist bl; | |
4957 | bl.append(freelist_type); | |
4958 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
4959 | } | |
b32b8144 FG |
4960 | // being able to allocate in units less than bdev block size |
4961 | // seems to be a bad idea. | |
11fdf7f2 | 4962 | ceph_assert( cct->_conf->bdev_block_size <= (int64_t)min_alloc_size); |
b32b8144 | 4963 | fm->create(bdev->get_size(), (int64_t)min_alloc_size, t); |
7c673cae FG |
4964 | |
4965 | // allocate superblock reserved space. note that we do not mark | |
4966 | // bluefs space as allocated in the freelist; we instead rely on | |
4967 | // bluefs_extents. | |
11fdf7f2 | 4968 | auto reserved = _get_ondisk_reserved(); |
3efd9988 | 4969 | fm->allocate(0, reserved, t); |
7c673cae | 4970 | |
7c673cae | 4971 | if (cct->_conf->bluestore_bluefs) { |
11fdf7f2 | 4972 | ceph_assert(bluefs_extents.num_intervals() == 1); |
7c673cae | 4973 | interval_set<uint64_t>::iterator p = bluefs_extents.begin(); |
11fdf7f2 | 4974 | reserved = round_up_to(p.get_start() + p.get_len(), min_alloc_size); |
7c673cae FG |
4975 | dout(20) << __func__ << " reserved 0x" << std::hex << reserved << std::dec |
4976 | << " for bluefs" << dendl; | |
7c673cae FG |
4977 | } |
4978 | ||
4979 | if (cct->_conf->bluestore_debug_prefill > 0) { | |
4980 | uint64_t end = bdev->get_size() - reserved; | |
4981 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
4982 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
4983 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 4984 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
4985 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
4986 | float r = cct->_conf->bluestore_debug_prefill; | |
4987 | r /= 1.0 - r; | |
4988 | bool stop = false; | |
4989 | ||
4990 | while (!stop && start < end) { | |
4991 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
4992 | if (start + l > end) { | |
4993 | l = end - start; | |
11fdf7f2 | 4994 | l = p2align(l, min_alloc_size); |
7c673cae | 4995 | } |
11fdf7f2 | 4996 | ceph_assert(start + l <= end); |
7c673cae FG |
4997 | |
4998 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 4999 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
5000 | if (start + l + u > end) { |
5001 | u = end - (start + l); | |
5002 | // trim to align so we don't overflow again | |
11fdf7f2 | 5003 | u = p2align(u, min_alloc_size); |
7c673cae FG |
5004 | stop = true; |
5005 | } | |
11fdf7f2 | 5006 | ceph_assert(start + l + u <= end); |
7c673cae | 5007 | |
11fdf7f2 | 5008 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
5009 | << " use 0x" << u << std::dec << dendl; |
5010 | ||
5011 | if (u == 0) { | |
5012 | // break if u has been trimmed to nothing | |
5013 | break; | |
5014 | } | |
5015 | ||
5016 | fm->allocate(start + l, u, t); | |
5017 | start += l + u; | |
5018 | } | |
5019 | } | |
7c673cae FG |
5020 | } |
5021 | ||
11fdf7f2 | 5022 | int r = fm->init(db); |
7c673cae FG |
5023 | if (r < 0) { |
5024 | derr << __func__ << " freelist init failed: " << cpp_strerror(r) << dendl; | |
5025 | delete fm; | |
5026 | fm = NULL; | |
5027 | return r; | |
5028 | } | |
81eedcae TL |
5029 | // if space size tracked by free list manager is that higher than actual |
5030 | // dev size one can hit out-of-space allocation which will result | |
5031 | // in data loss and/or assertions | |
5032 | // Probably user altered the device size somehow. | |
5033 | // The only fix for now is to redeploy OSD. | |
5034 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
5035 | ostringstream ss; | |
5036 | ss << "slow device size mismatch detected, " | |
5037 | << " fm size(" << fm->get_size() | |
5038 | << ") > slow device size(" << bdev->get_size() | |
5039 | << "), Please stop using this OSD as it might cause data loss."; | |
5040 | _set_disk_size_mismatch_alert(ss.str()); | |
5041 | } | |
7c673cae FG |
5042 | return 0; |
5043 | } | |
5044 | ||
5045 | void BlueStore::_close_fm() | |
5046 | { | |
5047 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 5048 | ceph_assert(fm); |
7c673cae FG |
5049 | fm->shutdown(); |
5050 | delete fm; | |
5051 | fm = NULL; | |
5052 | } | |
5053 | ||
5054 | int BlueStore::_open_alloc() | |
5055 | { | |
11fdf7f2 TL |
5056 | ceph_assert(alloc == NULL); |
5057 | ceph_assert(bdev->get_size()); | |
5058 | ||
5059 | if (bluefs) { | |
5060 | bluefs_extents.clear(); | |
9f95a23c TL |
5061 | auto r = bluefs->get_block_extents(bluefs_layout.shared_bdev, |
5062 | &bluefs_extents); | |
11fdf7f2 TL |
5063 | if (r < 0) { |
5064 | lderr(cct) << __func__ << " failed to retrieve bluefs_extents: " | |
5065 | << cpp_strerror(r) << dendl; | |
5066 | ||
5067 | return r; | |
5068 | } | |
5069 | dout(10) << __func__ << " bluefs extents 0x" | |
5070 | << std::hex << bluefs_extents << std::dec | |
5071 | << dendl; | |
5072 | } | |
5073 | ||
7c673cae FG |
5074 | alloc = Allocator::create(cct, cct->_conf->bluestore_allocator, |
5075 | bdev->get_size(), | |
eafe8130 | 5076 | min_alloc_size, "block"); |
7c673cae FG |
5077 | if (!alloc) { |
5078 | lderr(cct) << __func__ << " Allocator::unknown alloc type " | |
5079 | << cct->_conf->bluestore_allocator | |
5080 | << dendl; | |
5081 | return -EINVAL; | |
5082 | } | |
5083 | ||
5084 | uint64_t num = 0, bytes = 0; | |
5085 | ||
5086 | dout(1) << __func__ << " opening allocation metadata" << dendl; | |
5087 | // initialize from freelist | |
5088 | fm->enumerate_reset(); | |
5089 | uint64_t offset, length; | |
11fdf7f2 | 5090 | while (fm->enumerate_next(db, &offset, &length)) { |
7c673cae FG |
5091 | alloc->init_add_free(offset, length); |
5092 | ++num; | |
5093 | bytes += length; | |
5094 | } | |
224ce89b | 5095 | fm->enumerate_reset(); |
1adf2230 | 5096 | dout(1) << __func__ << " loaded " << byte_u_t(bytes) |
7c673cae FG |
5097 | << " in " << num << " extents" |
5098 | << dendl; | |
5099 | ||
5100 | // also mark bluefs space as allocated | |
5101 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
5102 | alloc->init_rm_free(e.get_start(), e.get_len()); | |
5103 | } | |
7c673cae FG |
5104 | |
5105 | return 0; | |
5106 | } | |
5107 | ||
5108 | void BlueStore::_close_alloc() | |
5109 | { | |
11fdf7f2 TL |
5110 | ceph_assert(bdev); |
5111 | bdev->discard_drain(); | |
5112 | ||
5113 | ceph_assert(alloc); | |
7c673cae FG |
5114 | alloc->shutdown(); |
5115 | delete alloc; | |
5116 | alloc = NULL; | |
11fdf7f2 | 5117 | bluefs_extents.clear(); |
7c673cae FG |
5118 | } |
5119 | ||
5120 | int BlueStore::_open_fsid(bool create) | |
5121 | { | |
11fdf7f2 | 5122 | ceph_assert(fsid_fd < 0); |
91327a77 | 5123 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
5124 | if (create) |
5125 | flags |= O_CREAT; | |
5126 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
5127 | if (fsid_fd < 0) { | |
5128 | int err = -errno; | |
5129 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
5130 | return err; | |
5131 | } | |
5132 | return 0; | |
5133 | } | |
5134 | ||
5135 | int BlueStore::_read_fsid(uuid_d *uuid) | |
5136 | { | |
5137 | char fsid_str[40]; | |
5138 | memset(fsid_str, 0, sizeof(fsid_str)); | |
5139 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
5140 | if (ret < 0) { | |
5141 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
5142 | return ret; | |
5143 | } | |
5144 | if (ret > 36) | |
5145 | fsid_str[36] = 0; | |
5146 | else | |
5147 | fsid_str[ret] = 0; | |
5148 | if (!uuid->parse(fsid_str)) { | |
5149 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
5150 | return -EINVAL; | |
5151 | } | |
5152 | return 0; | |
5153 | } | |
5154 | ||
5155 | int BlueStore::_write_fsid() | |
5156 | { | |
5157 | int r = ::ftruncate(fsid_fd, 0); | |
5158 | if (r < 0) { | |
5159 | r = -errno; | |
5160 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
5161 | return r; | |
5162 | } | |
5163 | string str = stringify(fsid) + "\n"; | |
5164 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
5165 | if (r < 0) { | |
5166 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
5167 | return r; | |
5168 | } | |
5169 | r = ::fsync(fsid_fd); | |
5170 | if (r < 0) { | |
5171 | r = -errno; | |
5172 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
5173 | return r; | |
5174 | } | |
5175 | return 0; | |
5176 | } | |
5177 | ||
5178 | void BlueStore::_close_fsid() | |
5179 | { | |
5180 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
5181 | fsid_fd = -1; | |
5182 | } | |
5183 | ||
5184 | int BlueStore::_lock_fsid() | |
5185 | { | |
5186 | struct flock l; | |
5187 | memset(&l, 0, sizeof(l)); | |
5188 | l.l_type = F_WRLCK; | |
5189 | l.l_whence = SEEK_SET; | |
5190 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
5191 | if (r < 0) { | |
5192 | int err = errno; | |
5193 | derr << __func__ << " failed to lock " << path << "/fsid" | |
5194 | << " (is another ceph-osd still running?)" | |
5195 | << cpp_strerror(err) << dendl; | |
5196 | return -err; | |
5197 | } | |
5198 | return 0; | |
5199 | } | |
5200 | ||
31f18b77 FG |
5201 | bool BlueStore::is_rotational() |
5202 | { | |
5203 | if (bdev) { | |
5204 | return bdev->is_rotational(); | |
5205 | } | |
5206 | ||
5207 | bool rotational = true; | |
5208 | int r = _open_path(); | |
5209 | if (r < 0) | |
5210 | goto out; | |
5211 | r = _open_fsid(false); | |
5212 | if (r < 0) | |
5213 | goto out_path; | |
5214 | r = _read_fsid(&fsid); | |
5215 | if (r < 0) | |
5216 | goto out_fsid; | |
5217 | r = _lock_fsid(); | |
5218 | if (r < 0) | |
5219 | goto out_fsid; | |
5220 | r = _open_bdev(false); | |
5221 | if (r < 0) | |
5222 | goto out_fsid; | |
5223 | rotational = bdev->is_rotational(); | |
5224 | _close_bdev(); | |
5225 | out_fsid: | |
5226 | _close_fsid(); | |
5227 | out_path: | |
5228 | _close_path(); | |
5229 | out: | |
5230 | return rotational; | |
5231 | } | |
5232 | ||
d2e6a577 FG |
5233 | bool BlueStore::is_journal_rotational() |
5234 | { | |
5235 | if (!bluefs) { | |
5236 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
5237 | << dendl; | |
5238 | return is_rotational(); | |
5239 | } | |
5240 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
5241 | return bluefs->wal_is_rotational(); | |
5242 | } | |
5243 | ||
9f95a23c TL |
5244 | bool BlueStore::_use_rotational_settings() |
5245 | { | |
5246 | if (cct->_conf->bluestore_debug_enforce_settings == "hdd") { | |
5247 | return true; | |
5248 | } | |
5249 | if (cct->_conf->bluestore_debug_enforce_settings == "ssd") { | |
5250 | return false; | |
5251 | } | |
5252 | return bdev->is_rotational(); | |
5253 | } | |
5254 | ||
7c673cae FG |
5255 | bool BlueStore::test_mount_in_use() |
5256 | { | |
5257 | // most error conditions mean the mount is not in use (e.g., because | |
5258 | // it doesn't exist). only if we fail to lock do we conclude it is | |
5259 | // in use. | |
5260 | bool ret = false; | |
5261 | int r = _open_path(); | |
5262 | if (r < 0) | |
5263 | return false; | |
5264 | r = _open_fsid(false); | |
5265 | if (r < 0) | |
5266 | goto out_path; | |
5267 | r = _lock_fsid(); | |
5268 | if (r < 0) | |
5269 | ret = true; // if we can't lock, it is in use | |
5270 | _close_fsid(); | |
5271 | out_path: | |
5272 | _close_path(); | |
5273 | return ret; | |
5274 | } | |
5275 | ||
11fdf7f2 | 5276 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
5277 | { |
5278 | int r; | |
11fdf7f2 | 5279 | bluefs = new BlueFS(cct); |
7c673cae | 5280 | |
11fdf7f2 TL |
5281 | string bfn; |
5282 | struct stat st; | |
5283 | ||
5284 | bfn = path + "/block.db"; | |
5285 | if (::stat(bfn.c_str(), &st) == 0) { | |
eafe8130 TL |
5286 | r = bluefs->add_block_device( |
5287 | BlueFS::BDEV_DB, bfn, | |
5288 | create && cct->_conf->bdev_enable_discard); | |
7c673cae | 5289 | if (r < 0) { |
11fdf7f2 TL |
5290 | derr << __func__ << " add block device(" << bfn << ") returned: " |
5291 | << cpp_strerror(r) << dendl; | |
5292 | goto free_bluefs; | |
7c673cae | 5293 | } |
7c673cae | 5294 | |
11fdf7f2 TL |
5295 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
5296 | r = _check_or_set_bdev_label( | |
5297 | bfn, | |
5298 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
5299 | "bluefs db", create); | |
5300 | if (r < 0) { | |
5301 | derr << __func__ | |
5302 | << " check block device(" << bfn << ") label returned: " | |
5303 | << cpp_strerror(r) << dendl; | |
5304 | goto free_bluefs; | |
5305 | } | |
7c673cae | 5306 | } |
11fdf7f2 TL |
5307 | if (create) { |
5308 | bluefs->add_block_extent( | |
5309 | BlueFS::BDEV_DB, | |
5310 | SUPER_RESERVED, | |
5311 | bluefs->get_block_device_size(BlueFS::BDEV_DB) - SUPER_RESERVED); | |
5312 | } | |
9f95a23c TL |
5313 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
5314 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
5315 | } else { |
5316 | r = -errno; | |
5317 | if (::lstat(bfn.c_str(), &st) == -1) { | |
5318 | r = 0; | |
9f95a23c | 5319 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7c673cae | 5320 | } else { |
11fdf7f2 TL |
5321 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
5322 | << cpp_strerror(r) << dendl; | |
5323 | goto free_bluefs; | |
7c673cae FG |
5324 | } |
5325 | } | |
7c673cae | 5326 | |
11fdf7f2 TL |
5327 | // shared device |
5328 | bfn = path + "/block"; | |
5329 | // never trim here | |
9f95a23c | 5330 | r = bluefs->add_block_device(bluefs_layout.shared_bdev, bfn, false, |
11fdf7f2 TL |
5331 | true /* shared with bluestore */); |
5332 | if (r < 0) { | |
5333 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5334 | << cpp_strerror(r) << dendl; | |
5335 | goto free_bluefs; | |
5336 | } | |
5337 | if (create) { | |
5338 | // note: we always leave the first SUPER_RESERVED (8k) of the device unused | |
5339 | uint64_t initial = | |
5340 | bdev->get_size() * (cct->_conf->bluestore_bluefs_min_ratio + | |
5341 | cct->_conf->bluestore_bluefs_gift_ratio); | |
5342 | initial = std::max(initial, cct->_conf->bluestore_bluefs_min); | |
eafe8130 TL |
5343 | uint64_t alloc_size = cct->_conf->bluefs_shared_alloc_size; |
5344 | if (alloc_size % min_alloc_size) { | |
5345 | derr << __func__ << " bluefs_shared_alloc_size 0x" << std::hex | |
5346 | << alloc_size << " is not a multiple of " | |
11fdf7f2 TL |
5347 | << "min_alloc_size 0x" << min_alloc_size << std::dec << dendl; |
5348 | r = -EINVAL; | |
5349 | goto free_bluefs; | |
7c673cae | 5350 | } |
11fdf7f2 | 5351 | // align to bluefs's alloc_size |
eafe8130 | 5352 | initial = p2roundup(initial, alloc_size); |
11fdf7f2 | 5353 | // put bluefs in the middle of the device in case it is an HDD |
eafe8130 | 5354 | uint64_t start = p2align((bdev->get_size() - initial) / 2, alloc_size); |
11fdf7f2 | 5355 | //avoiding superblock overwrite |
eafe8130 TL |
5356 | start = std::max(alloc_size, start); |
5357 | ceph_assert(start >=_get_ondisk_reserved()); | |
7c673cae | 5358 | |
9f95a23c | 5359 | bluefs->add_block_extent(bluefs_layout.shared_bdev, start, initial); |
11fdf7f2 TL |
5360 | bluefs_extents.insert(start, initial); |
5361 | ++out_of_sync_fm; | |
5362 | } | |
5363 | ||
5364 | bfn = path + "/block.wal"; | |
5365 | if (::stat(bfn.c_str(), &st) == 0) { | |
5366 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
eafe8130 | 5367 | create && cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
5368 | if (r < 0) { |
5369 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5370 | << cpp_strerror(r) << dendl; | |
5371 | goto free_bluefs; | |
5372 | } | |
7c673cae | 5373 | |
11fdf7f2 TL |
5374 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
5375 | r = _check_or_set_bdev_label( | |
5376 | bfn, | |
5377 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
5378 | "bluefs wal", create); | |
7c673cae | 5379 | if (r < 0) { |
11fdf7f2 TL |
5380 | derr << __func__ << " check block device(" << bfn |
5381 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
5382 | goto free_bluefs; |
5383 | } | |
7c673cae FG |
5384 | } |
5385 | ||
11fdf7f2 TL |
5386 | if (create) { |
5387 | bluefs->add_block_extent( | |
5388 | BlueFS::BDEV_WAL, BDEV_LABEL_BLOCK_SIZE, | |
5389 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) - | |
5390 | BDEV_LABEL_BLOCK_SIZE); | |
5391 | } | |
9f95a23c | 5392 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
5393 | } else { |
5394 | r = 0; | |
5395 | if (::lstat(bfn.c_str(), &st) != -1) { | |
5396 | r = -errno; | |
5397 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
5398 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
5399 | goto free_bluefs; |
5400 | } | |
11fdf7f2 TL |
5401 | } |
5402 | return 0; | |
7c673cae | 5403 | |
11fdf7f2 TL |
5404 | free_bluefs: |
5405 | ceph_assert(bluefs); | |
5406 | delete bluefs; | |
5407 | bluefs = NULL; | |
5408 | return r; | |
5409 | } | |
7c673cae | 5410 | |
11fdf7f2 TL |
5411 | int BlueStore::_open_bluefs(bool create) |
5412 | { | |
5413 | int r = _minimal_open_bluefs(create); | |
5414 | if (r < 0) { | |
5415 | return r; | |
5416 | } | |
9f95a23c TL |
5417 | RocksDBBlueFSVolumeSelector* vselector = nullptr; |
5418 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { | |
5419 | ||
5420 | string options = cct->_conf->bluestore_rocksdb_options; | |
5421 | ||
5422 | rocksdb::Options rocks_opts; | |
5423 | int r = RocksDBStore::ParseOptionsFromStringStatic( | |
5424 | cct, | |
5425 | options, | |
5426 | rocks_opts, | |
5427 | nullptr); | |
5428 | if (r < 0) { | |
5429 | return r; | |
5430 | } | |
5431 | ||
5432 | double reserved_factor = cct->_conf->bluestore_volume_selection_reserved_factor; | |
5433 | vselector = | |
5434 | new RocksDBBlueFSVolumeSelector( | |
5435 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
5436 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
5437 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100, | |
5438 | 1024 * 1024 * 1024, //FIXME: set expected l0 size here | |
5439 | rocks_opts.max_bytes_for_level_base, | |
5440 | rocks_opts.max_bytes_for_level_multiplier, | |
5441 | reserved_factor, | |
5442 | cct->_conf->bluestore_volume_selection_reserved, | |
5443 | cct->_conf->bluestore_volume_selection_policy != "rocksdb_original"); | |
5444 | } | |
11fdf7f2 | 5445 | if (create) { |
9f95a23c | 5446 | bluefs->mkfs(fsid, bluefs_layout); |
11fdf7f2 | 5447 | } |
9f95a23c | 5448 | bluefs->set_volume_selector(vselector); |
11fdf7f2 TL |
5449 | r = bluefs->mount(); |
5450 | if (r < 0) { | |
5451 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
5452 | } | |
9f95a23c | 5453 | ceph_assert_always(bluefs->maybe_verify_layout(bluefs_layout) == 0); |
11fdf7f2 TL |
5454 | return r; |
5455 | } | |
5456 | ||
5457 | void BlueStore::_close_bluefs() | |
5458 | { | |
5459 | bluefs->umount(); | |
5460 | _minimal_close_bluefs(); | |
5461 | } | |
5462 | ||
5463 | void BlueStore::_minimal_close_bluefs() | |
5464 | { | |
5465 | delete bluefs; | |
5466 | bluefs = NULL; | |
5467 | } | |
5468 | ||
5469 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
5470 | { | |
5471 | if (create) { | |
5472 | *ret = cct->_conf->bluestore_bluefs; | |
5473 | } else { | |
5474 | string s; | |
5475 | int r = read_meta("bluefs", &s); | |
5476 | if (r < 0) { | |
5477 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
5478 | return -EIO; | |
5479 | } | |
5480 | if (s == "1") { | |
5481 | *ret = true; | |
5482 | } else if (s == "0") { | |
5483 | *ret = false; | |
31f18b77 | 5484 | } else { |
11fdf7f2 TL |
5485 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
5486 | << dendl; | |
5487 | return -EIO; | |
5488 | } | |
5489 | } | |
5490 | return 0; | |
5491 | } | |
5492 | ||
5493 | /* | |
5494 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
5495 | * in the proper order | |
5496 | */ | |
5497 | int BlueStore::_open_db_and_around(bool read_only) | |
5498 | { | |
5499 | int r; | |
5500 | bool do_bluefs = false; | |
5501 | _is_bluefs(false, &do_bluefs); // ignore err code | |
5502 | if (do_bluefs) { | |
5503 | // open in read-only first to read FM list and init allocator | |
5504 | // as they might be needed for some BlueFS procedures | |
5505 | r = _open_db(false, false, true); | |
5506 | if (r < 0) | |
5507 | return r; | |
5508 | ||
5509 | r = _open_super_meta(); | |
5510 | if (r < 0) { | |
5511 | goto out_db; | |
5512 | } | |
5513 | ||
5514 | r = _open_fm(nullptr); | |
5515 | if (r < 0) | |
5516 | goto out_db; | |
5517 | ||
5518 | r = _open_alloc(); | |
5519 | if (r < 0) | |
5520 | goto out_fm; | |
5521 | ||
5522 | // now open in R/W mode | |
5523 | if (!read_only) { | |
5524 | _close_db(); | |
5525 | ||
5526 | r = _open_db(false, false, false); | |
5527 | if (r < 0) { | |
5528 | _close_alloc(); | |
5529 | _close_fm(); | |
5530 | return r; | |
28e407b8 | 5531 | } |
7c673cae | 5532 | } |
11fdf7f2 TL |
5533 | } else { |
5534 | r = _open_db(false, false); | |
5535 | if (r < 0) { | |
5536 | return r; | |
5537 | } | |
5538 | r = _open_super_meta(); | |
5539 | if (r < 0) { | |
5540 | goto out_db; | |
5541 | } | |
7c673cae | 5542 | |
11fdf7f2 TL |
5543 | r = _open_fm(nullptr); |
5544 | if (r < 0) | |
5545 | goto out_db; | |
5546 | ||
5547 | r = _open_alloc(); | |
5548 | if (r < 0) | |
5549 | goto out_fm; | |
5550 | } | |
5551 | return 0; | |
5552 | ||
5553 | out_fm: | |
5554 | _close_fm(); | |
5555 | out_db: | |
5556 | _close_db(); | |
5557 | return r; | |
5558 | } | |
5559 | ||
5560 | void BlueStore::_close_db_and_around() | |
5561 | { | |
5562 | if (bluefs) { | |
5563 | if (out_of_sync_fm.fetch_and(0)) { | |
5564 | _sync_bluefs_and_fm(); | |
5565 | } | |
5566 | _close_db(); | |
5567 | while(out_of_sync_fm.fetch_and(0)) { | |
5568 | // if seen some allocations during close - repeat open_db, sync fm, close | |
5569 | dout(0) << __func__ << " syncing FreelistManager" << dendl; | |
5570 | int r = _open_db(false, false, false); | |
5571 | if (r < 0) { | |
5572 | derr << __func__ | |
5573 | << " unable to open db, FreelistManager is probably out of sync" | |
5574 | << dendl; | |
5575 | break; | |
5576 | } | |
5577 | _sync_bluefs_and_fm(); | |
5578 | _close_db(); | |
7c673cae | 5579 | } |
11fdf7f2 TL |
5580 | if (!_kv_only) { |
5581 | _close_alloc(); | |
5582 | _close_fm(); | |
5583 | } | |
5584 | } else { | |
5585 | _close_alloc(); | |
5586 | _close_fm(); | |
5587 | _close_db(); | |
5588 | } | |
5589 | } | |
5590 | ||
5591 | // updates legacy bluefs related recs in DB to a state valid for | |
5592 | // downgrades from nautilus. | |
5593 | void BlueStore::_sync_bluefs_and_fm() | |
5594 | { | |
5595 | if (cct->_conf->bluestore_bluefs_db_compatibility) { | |
5596 | bufferlist bl; | |
5597 | encode(bluefs_extents, bl); | |
5598 | dout(20) << __func__ << " bluefs_extents at KV is now 0x" | |
5599 | << std::hex << bluefs_extents << std::dec | |
5600 | << dendl; | |
5601 | KeyValueDB::Transaction synct = db->get_transaction(); | |
5602 | synct->set(PREFIX_SUPER, "bluefs_extents", bl); | |
5603 | synct->set(PREFIX_SUPER, "bluefs_extents_back", bl); | |
5604 | ||
5605 | // Nice thing is that we don't need to update FreelistManager here. | |
5606 | // It always has corresponding bits set to 'Free' for both Nautilus+ and | |
5607 | // pre-Nautilis releases. | |
5608 | // So once we get an extent to bluefs_extents this means it's | |
5609 | // been free in allocator and hence it's free in FM too. | |
5610 | ||
5611 | db->submit_transaction_sync(synct); | |
5612 | } | |
5613 | } | |
5614 | ||
5615 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) | |
5616 | { | |
5617 | int r; | |
5618 | ceph_assert(!db); | |
5619 | ceph_assert(!(create && read_only)); | |
5620 | string fn = path + "/db"; | |
5621 | string options; | |
5622 | stringstream err; | |
5623 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); | |
5624 | ||
5625 | string kv_backend; | |
5626 | std::vector<KeyValueDB::ColumnFamily> cfs; | |
5627 | ||
5628 | if (create) { | |
5629 | kv_backend = cct->_conf->bluestore_kvbackend; | |
5630 | } else { | |
5631 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 5632 | if (r < 0) { |
11fdf7f2 TL |
5633 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
5634 | return -EIO; | |
5635 | } | |
5636 | } | |
5637 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
5638 | ||
5639 | bool do_bluefs; | |
5640 | r = _is_bluefs(create, &do_bluefs); | |
5641 | if (r < 0) { | |
5642 | return r; | |
5643 | } | |
5644 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
5645 | ||
5646 | map<string,string> kv_options; | |
5647 | // force separate wal dir for all new deployments. | |
5648 | kv_options["separate_wal_dir"] = 1; | |
5649 | rocksdb::Env *env = NULL; | |
5650 | if (do_bluefs) { | |
5651 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
5652 | if (kv_backend != "rocksdb") { | |
5653 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
5654 | return -EINVAL; | |
7c673cae | 5655 | } |
11fdf7f2 TL |
5656 | |
5657 | r = _open_bluefs(create); | |
5658 | if (r < 0) { | |
5659 | return r; | |
5660 | } | |
11fdf7f2 | 5661 | |
7c673cae | 5662 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
9f95a23c TL |
5663 | rocksdb::Env* a = new BlueRocksEnv(bluefs); |
5664 | rocksdb::Env* b = rocksdb::Env::Default(); | |
7c673cae | 5665 | if (create) { |
9f95a23c TL |
5666 | string cmd = "rm -rf " + path + "/db " + |
5667 | path + "/db.slow " + | |
5668 | path + "/db.wal"; | |
5669 | int r = system(cmd.c_str()); | |
5670 | (void)r; | |
7c673cae FG |
5671 | } |
5672 | env = new rocksdb::EnvMirror(b, a, false, true); | |
9f95a23c TL |
5673 | } |
5674 | else { | |
7c673cae FG |
5675 | env = new BlueRocksEnv(bluefs); |
5676 | ||
5677 | // simplify the dir names, too, as "seen" by rocksdb | |
5678 | fn = "db"; | |
5679 | } | |
9f95a23c TL |
5680 | bluefs->set_slow_device_expander(this); |
5681 | BlueFSVolumeSelector::paths paths; | |
5682 | bluefs->get_vselector_paths(fn, paths); | |
7c673cae | 5683 | |
9f95a23c | 5684 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
7c673cae FG |
5685 | // we have both block.db and block; tell rocksdb! |
5686 | // note: the second (last) size value doesn't really matter | |
5687 | ostringstream db_paths; | |
9f95a23c TL |
5688 | bool first = true; |
5689 | for (auto& p : paths) { | |
5690 | if (!first) { | |
5691 | db_paths << " "; | |
5692 | } | |
5693 | first = false; | |
5694 | db_paths << p.first << "," << p.second; | |
5695 | ||
5696 | } | |
11fdf7f2 | 5697 | kv_options["db_paths"] = db_paths.str(); |
9f95a23c | 5698 | dout(1) << __func__ << " set db_paths to " << db_paths.str() << dendl; |
7c673cae FG |
5699 | } |
5700 | ||
5701 | if (create) { | |
9f95a23c TL |
5702 | for (auto& p : paths) { |
5703 | env->CreateDir(p.first); | |
5704 | } | |
5705 | // Selectors don't provide wal path so far hence create explicitly | |
11fdf7f2 | 5706 | env->CreateDir(fn + ".wal"); |
11fdf7f2 TL |
5707 | } else { |
5708 | std::vector<std::string> res; | |
5709 | // check for dir presence | |
5710 | auto r = env->GetChildren(fn+".wal", &res); | |
5711 | if (r.IsNotFound()) { | |
5712 | kv_options.erase("separate_wal_dir"); | |
5713 | } | |
7c673cae | 5714 | } |
11fdf7f2 TL |
5715 | } else { |
5716 | string walfn = path + "/db.wal"; | |
7c673cae | 5717 | |
11fdf7f2 TL |
5718 | if (create) { |
5719 | int r = ::mkdir(fn.c_str(), 0755); | |
5720 | if (r < 0) | |
5721 | r = -errno; | |
5722 | if (r < 0 && r != -EEXIST) { | |
5723 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
5724 | << dendl; | |
5725 | return r; | |
5726 | } | |
5727 | ||
5728 | // wal_dir, too! | |
7c673cae FG |
5729 | r = ::mkdir(walfn.c_str(), 0755); |
5730 | if (r < 0) | |
5731 | r = -errno; | |
5732 | if (r < 0 && r != -EEXIST) { | |
5733 | derr << __func__ << " failed to create " << walfn | |
5734 | << ": " << cpp_strerror(r) | |
5735 | << dendl; | |
5736 | return r; | |
5737 | } | |
11fdf7f2 TL |
5738 | } else { |
5739 | struct stat st; | |
5740 | r = ::stat(walfn.c_str(), &st); | |
5741 | if (r < 0 && errno == ENOENT) { | |
5742 | kv_options.erase("separate_wal_dir"); | |
5743 | } | |
7c673cae FG |
5744 | } |
5745 | } | |
5746 | ||
91327a77 | 5747 | |
7c673cae FG |
5748 | db = KeyValueDB::create(cct, |
5749 | kv_backend, | |
5750 | fn, | |
11fdf7f2 | 5751 | kv_options, |
7c673cae FG |
5752 | static_cast<void*>(env)); |
5753 | if (!db) { | |
5754 | derr << __func__ << " error creating db" << dendl; | |
5755 | if (bluefs) { | |
11fdf7f2 | 5756 | _close_bluefs(); |
7c673cae FG |
5757 | } |
5758 | // delete env manually here since we can't depend on db to do this | |
5759 | // under this case | |
5760 | delete env; | |
5761 | env = NULL; | |
5762 | return -EIO; | |
5763 | } | |
5764 | ||
5765 | FreelistManager::setup_merge_operators(db); | |
5766 | db->set_merge_operator(PREFIX_STAT, merge_op); | |
91327a77 | 5767 | db->set_cache_size(cache_kv_ratio * cache_size); |
31f18b77 | 5768 | |
11fdf7f2 | 5769 | if (kv_backend == "rocksdb") { |
7c673cae | 5770 | options = cct->_conf->bluestore_rocksdb_options; |
11fdf7f2 TL |
5771 | |
5772 | map<string,string> cf_map; | |
5773 | cct->_conf.with_val<string>("bluestore_rocksdb_cfs", | |
5774 | get_str_map, | |
5775 | &cf_map, | |
5776 | " \t"); | |
5777 | for (auto& i : cf_map) { | |
5778 | dout(10) << "column family " << i.first << ": " << i.second << dendl; | |
5779 | cfs.push_back(KeyValueDB::ColumnFamily(i.first, i.second)); | |
5780 | } | |
5781 | } | |
5782 | ||
7c673cae | 5783 | db->init(options); |
11fdf7f2 TL |
5784 | if (to_repair_db) |
5785 | return 0; | |
5786 | if (create) { | |
5787 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { | |
5788 | r = db->create_and_open(err, cfs); | |
5789 | } else { | |
5790 | r = db->create_and_open(err); | |
5791 | } | |
5792 | } else { | |
5793 | // we pass in cf list here, but it is only used if the db already has | |
5794 | // column families created. | |
5795 | r = read_only ? | |
5796 | db->open_read_only(err, cfs) : | |
5797 | db->open(err, cfs); | |
5798 | } | |
7c673cae FG |
5799 | if (r) { |
5800 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
11fdf7f2 | 5801 | _close_db(); |
7c673cae FG |
5802 | return -EIO; |
5803 | } | |
5804 | dout(1) << __func__ << " opened " << kv_backend | |
5805 | << " path " << fn << " options " << options << dendl; | |
5806 | return 0; | |
7c673cae FG |
5807 | } |
5808 | ||
5809 | void BlueStore::_close_db() | |
5810 | { | |
11fdf7f2 | 5811 | ceph_assert(db); |
7c673cae FG |
5812 | delete db; |
5813 | db = NULL; | |
5814 | if (bluefs) { | |
11fdf7f2 | 5815 | _close_bluefs(); |
7c673cae FG |
5816 | } |
5817 | } | |
5818 | ||
11fdf7f2 | 5819 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 5820 | { |
11fdf7f2 TL |
5821 | auto dump_interval = |
5822 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
5823 | if (dump_interval > 0 && | |
5824 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
5825 | alloc->dump(); | |
5826 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); | |
5827 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 5828 | } |
11fdf7f2 | 5829 | } |
7c673cae | 5830 | |
7c673cae | 5831 | |
11fdf7f2 TL |
5832 | int BlueStore::allocate_bluefs_freespace( |
5833 | uint64_t min_size, | |
5834 | uint64_t size, | |
5835 | PExtentVector* extents_out) | |
5836 | { | |
5837 | ceph_assert(min_size <= size); | |
5838 | if (size) { | |
5839 | // round up to alloc size | |
9f95a23c | 5840 | uint64_t alloc_size = bluefs->get_alloc_size(bluefs_layout.shared_bdev); |
eafe8130 TL |
5841 | min_size = p2roundup(min_size, alloc_size); |
5842 | size = p2roundup(size, alloc_size); | |
11fdf7f2 TL |
5843 | |
5844 | PExtentVector extents_local; | |
5845 | PExtentVector* extents = extents_out ? extents_out : &extents_local; | |
5846 | ||
5847 | ||
5848 | uint64_t gift; | |
5849 | uint64_t allocated = 0; | |
5850 | int64_t alloc_len; | |
9f95a23c TL |
5851 | auto need = size; |
5852 | auto extent_count0 = extents->size(); | |
11fdf7f2 TL |
5853 | do { |
5854 | // hard cap to fit into 32 bits | |
9f95a23c | 5855 | gift = std::min<uint64_t>(size, 1ull << 30); |
11fdf7f2 TL |
5856 | dout(10) << __func__ << " gifting " << gift |
5857 | << " (" << byte_u_t(gift) << ")" << dendl; | |
5858 | ||
eafe8130 TL |
5859 | alloc_len = alloc->allocate(gift, alloc_size, 0, 0, extents); |
5860 | if (alloc_len > 0) { | |
11fdf7f2 TL |
5861 | allocated += alloc_len; |
5862 | size -= alloc_len; | |
5863 | } | |
5864 | ||
eafe8130 TL |
5865 | if (alloc_len < 0 || |
5866 | (alloc_len < (int64_t)gift && (min_size > allocated))) { | |
11fdf7f2 TL |
5867 | derr << __func__ |
5868 | << " failed to allocate on 0x" << std::hex << gift | |
5869 | << " min_size 0x" << min_size | |
5870 | << " > allocated total 0x" << allocated | |
eafe8130 TL |
5871 | << " bluefs_shared_alloc_size 0x" << alloc_size |
5872 | << " allocated 0x" << (alloc_len < 0 ? 0 : alloc_len) | |
11fdf7f2 TL |
5873 | << " available 0x " << alloc->get_free() |
5874 | << std::dec << dendl; | |
7c673cae | 5875 | |
494da23a | 5876 | _dump_alloc_on_failure(); |
11fdf7f2 TL |
5877 | alloc->release(*extents); |
5878 | extents->clear(); | |
5879 | return -ENOSPC; | |
5880 | } | |
5881 | } while (size && alloc_len > 0); | |
9f95a23c TL |
5882 | _collect_allocation_stats(need, alloc_size, extents->size() - extent_count0); |
5883 | ||
11fdf7f2 TL |
5884 | for (auto& e : *extents) { |
5885 | dout(5) << __func__ << " gifting " << e << " to bluefs" << dendl; | |
5886 | bluefs_extents.insert(e.offset, e.length); | |
5887 | ++out_of_sync_fm; | |
5888 | // apply to bluefs if not requested from outside | |
5889 | if (!extents_out) { | |
9f95a23c | 5890 | bluefs->add_block_extent(bluefs_layout.shared_bdev, e.offset, e.length); |
11fdf7f2 | 5891 | } |
7c673cae FG |
5892 | } |
5893 | } | |
7c673cae FG |
5894 | return 0; |
5895 | } | |
5896 | ||
9f95a23c TL |
5897 | uint64_t BlueStore::available_freespace(uint64_t alloc_size) { |
5898 | uint64_t total = 0; | |
5899 | auto iterated_allocation = [&](uint64_t off, uint64_t len) { | |
eafe8130 | 5900 | //only count in size that is alloc_size aligned |
9f95a23c TL |
5901 | uint64_t dist_to_alignment; |
5902 | uint64_t offset_in_block = off & (alloc_size - 1); | |
eafe8130 TL |
5903 | if (offset_in_block == 0) |
5904 | dist_to_alignment = 0; | |
5905 | else | |
5906 | dist_to_alignment = alloc_size - offset_in_block; | |
5907 | if (dist_to_alignment >= len) | |
5908 | return; | |
5909 | len -= dist_to_alignment; | |
5910 | total += p2align(len, alloc_size); | |
5911 | }; | |
5912 | alloc->dump(iterated_allocation); | |
5913 | return total; | |
5914 | } | |
5915 | ||
11fdf7f2 | 5916 | int64_t BlueStore::_get_bluefs_size_delta(uint64_t bluefs_free, uint64_t bluefs_total) |
f64942e4 | 5917 | { |
7c673cae FG |
5918 | float bluefs_free_ratio = (float)bluefs_free / (float)bluefs_total; |
5919 | ||
5920 | uint64_t my_free = alloc->get_free(); | |
5921 | uint64_t total = bdev->get_size(); | |
5922 | float my_free_ratio = (float)my_free / (float)total; | |
5923 | ||
5924 | uint64_t total_free = bluefs_free + my_free; | |
5925 | ||
5926 | float bluefs_ratio = (float)bluefs_free / (float)total_free; | |
5927 | ||
5928 | dout(10) << __func__ | |
1adf2230 | 5929 | << " bluefs " << byte_u_t(bluefs_free) |
7c673cae | 5930 | << " free (" << bluefs_free_ratio |
1adf2230 | 5931 | << ") bluestore " << byte_u_t(my_free) |
7c673cae FG |
5932 | << " free (" << my_free_ratio |
5933 | << "), bluefs_ratio " << bluefs_ratio | |
5934 | << dendl; | |
5935 | ||
5936 | uint64_t gift = 0; | |
5937 | uint64_t reclaim = 0; | |
5938 | if (bluefs_ratio < cct->_conf->bluestore_bluefs_min_ratio) { | |
5939 | gift = cct->_conf->bluestore_bluefs_gift_ratio * total_free; | |
9f95a23c TL |
5940 | if (gift >= my_free) |
5941 | gift = my_free / 2; | |
7c673cae FG |
5942 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio |
5943 | << " < min_ratio " << cct->_conf->bluestore_bluefs_min_ratio | |
1adf2230 | 5944 | << ", should gift " << byte_u_t(gift) << dendl; |
7c673cae FG |
5945 | } else if (bluefs_ratio > cct->_conf->bluestore_bluefs_max_ratio) { |
5946 | reclaim = cct->_conf->bluestore_bluefs_reclaim_ratio * total_free; | |
5947 | if (bluefs_total - reclaim < cct->_conf->bluestore_bluefs_min) | |
5948 | reclaim = bluefs_total - cct->_conf->bluestore_bluefs_min; | |
9f95a23c TL |
5949 | if (reclaim >= bluefs_free) |
5950 | reclaim = bluefs_free / 2; | |
7c673cae FG |
5951 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio |
5952 | << " > max_ratio " << cct->_conf->bluestore_bluefs_max_ratio | |
1adf2230 | 5953 | << ", should reclaim " << byte_u_t(reclaim) << dendl; |
7c673cae | 5954 | } |
3efd9988 FG |
5955 | |
5956 | // don't take over too much of the freespace | |
5957 | uint64_t free_cap = cct->_conf->bluestore_bluefs_max_ratio * total_free; | |
7c673cae | 5958 | if (bluefs_total < cct->_conf->bluestore_bluefs_min && |
3efd9988 | 5959 | cct->_conf->bluestore_bluefs_min < free_cap) { |
7c673cae FG |
5960 | uint64_t g = cct->_conf->bluestore_bluefs_min - bluefs_total; |
5961 | dout(10) << __func__ << " bluefs_total " << bluefs_total | |
5962 | << " < min " << cct->_conf->bluestore_bluefs_min | |
1adf2230 | 5963 | << ", should gift " << byte_u_t(g) << dendl; |
7c673cae FG |
5964 | if (g > gift) |
5965 | gift = g; | |
5966 | reclaim = 0; | |
5967 | } | |
9f95a23c TL |
5968 | uint64_t min_free = |
5969 | cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
3efd9988 FG |
5970 | if (bluefs_free < min_free && |
5971 | min_free < free_cap) { | |
5972 | uint64_t g = min_free - bluefs_free; | |
11fdf7f2 | 5973 | dout(10) << __func__ << " bluefs_free " << bluefs_free |
3efd9988 | 5974 | << " < min " << min_free |
1adf2230 | 5975 | << ", should gift " << byte_u_t(g) << dendl; |
3efd9988 FG |
5976 | if (g > gift) |
5977 | gift = g; | |
5978 | reclaim = 0; | |
5979 | } | |
9f95a23c TL |
5980 | uint64_t max_free = |
5981 | cct->_conf.get_val<Option::size_t>("bluestore_bluefs_max_free"); | |
5982 | if (bluefs_free > max_free) { | |
5983 | dout(10) << __func__ << " bluefs_free " << bluefs_free | |
5984 | << " > max " << max_free | |
5985 | << ", stop gifting for now" << dendl; | |
5986 | gift = 0; | |
5987 | } | |
11fdf7f2 TL |
5988 | ceph_assert((int64_t)gift >= 0); |
5989 | ceph_assert((int64_t)reclaim >= 0); | |
5990 | return gift > 0 ? (int64_t)gift : -(int64_t)reclaim; | |
5991 | } | |
7c673cae | 5992 | |
11fdf7f2 TL |
5993 | int BlueStore::_balance_bluefs_freespace() |
5994 | { | |
5995 | int ret = 0; | |
5996 | ceph_assert(bluefs); | |
7c673cae | 5997 | |
11fdf7f2 TL |
5998 | vector<pair<uint64_t,uint64_t>> bluefs_usage; // <free, total> ... |
5999 | bluefs->get_usage(&bluefs_usage); | |
9f95a23c | 6000 | ceph_assert(bluefs_usage.size() > bluefs_layout.shared_bdev); |
7c673cae | 6001 | |
11fdf7f2 | 6002 | bool clear_alert = true; |
9f95a23c TL |
6003 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
6004 | auto& p = bluefs_usage[bluefs_layout.shared_bdev]; | |
11fdf7f2 TL |
6005 | if (p.first != p.second) { |
6006 | auto& db = bluefs_usage[BlueFS::BDEV_DB]; | |
6007 | ostringstream ss; | |
6008 | ss << "spilled over " << byte_u_t(p.second - p.first) | |
6009 | << " metadata from 'db' device (" << byte_u_t(db.second - db.first) | |
6010 | << " used of " << byte_u_t(db.second) << ") to slow device"; | |
6011 | _set_spillover_alert(ss.str()); | |
6012 | clear_alert = false; | |
6013 | } | |
6014 | } | |
6015 | if (clear_alert) { | |
6016 | _clear_spillover_alert(); | |
7c673cae FG |
6017 | } |
6018 | ||
11fdf7f2 TL |
6019 | // fixme: look at primary bdev only for now |
6020 | int64_t delta = _get_bluefs_size_delta( | |
9f95a23c TL |
6021 | bluefs_usage[bluefs_layout.shared_bdev].first, |
6022 | bluefs_usage[bluefs_layout.shared_bdev].second); | |
11fdf7f2 | 6023 | |
7c673cae | 6024 | // reclaim from bluefs? |
11fdf7f2 | 6025 | if (delta < 0) { |
7c673cae | 6026 | // round up to alloc size |
9f95a23c | 6027 | uint64_t alloc_size = bluefs->get_alloc_size(bluefs_layout.shared_bdev); |
eafe8130 | 6028 | auto reclaim = p2roundup(uint64_t(-delta), alloc_size); |
7c673cae FG |
6029 | |
6030 | // hard cap to fit into 32 bits | |
9f95a23c | 6031 | reclaim = std::min<uint64_t>(reclaim, 1ull << 30); |
7c673cae | 6032 | dout(10) << __func__ << " reclaiming " << reclaim |
1adf2230 | 6033 | << " (" << byte_u_t(reclaim) << ")" << dendl; |
7c673cae FG |
6034 | |
6035 | while (reclaim > 0) { | |
6036 | // NOTE: this will block and do IO. | |
a8e16298 | 6037 | PExtentVector extents; |
9f95a23c | 6038 | int r = bluefs->reclaim_blocks(bluefs_layout.shared_bdev, reclaim, |
7c673cae FG |
6039 | &extents); |
6040 | if (r < 0) { | |
6041 | derr << __func__ << " failed to reclaim space from bluefs" | |
6042 | << dendl; | |
6043 | break; | |
6044 | } | |
6045 | for (auto e : extents) { | |
11fdf7f2 | 6046 | ++out_of_sync_fm; |
7c673cae FG |
6047 | bluefs_extents.erase(e.offset, e.length); |
6048 | bluefs_extents_reclaiming.insert(e.offset, e.length); | |
6049 | reclaim -= e.length; | |
6050 | } | |
6051 | } | |
6052 | ||
6053 | ret = 1; | |
6054 | } | |
6055 | ||
6056 | return ret; | |
6057 | } | |
6058 | ||
eafe8130 | 6059 | int BlueStore::_open_collections() |
7c673cae | 6060 | { |
28e407b8 | 6061 | dout(10) << __func__ << dendl; |
eafe8130 | 6062 | collections_had_errors = false; |
11fdf7f2 | 6063 | ceph_assert(coll_map.empty()); |
7c673cae FG |
6064 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
6065 | for (it->upper_bound(string()); | |
6066 | it->valid(); | |
6067 | it->next()) { | |
6068 | coll_t cid; | |
6069 | if (cid.parse(it->key())) { | |
9f95a23c | 6070 | auto c = ceph::make_ref<Collection>( |
7c673cae | 6071 | this, |
9f95a23c TL |
6072 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
6073 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
6074 | cid); | |
7c673cae | 6075 | bufferlist bl = it->value(); |
11fdf7f2 | 6076 | auto p = bl.cbegin(); |
7c673cae | 6077 | try { |
11fdf7f2 | 6078 | decode(c->cnode, p); |
7c673cae FG |
6079 | } catch (buffer::error& e) { |
6080 | derr << __func__ << " failed to decode cnode, key:" | |
6081 | << pretty_binary_string(it->key()) << dendl; | |
6082 | return -EIO; | |
6083 | } | |
28e407b8 AA |
6084 | dout(20) << __func__ << " opened " << cid << " " << c |
6085 | << " " << c->cnode << dendl; | |
11fdf7f2 | 6086 | _osr_attach(c.get()); |
7c673cae | 6087 | coll_map[cid] = c; |
11fdf7f2 | 6088 | |
7c673cae FG |
6089 | } else { |
6090 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
eafe8130 | 6091 | collections_had_errors = true; |
7c673cae FG |
6092 | } |
6093 | } | |
6094 | return 0; | |
6095 | } | |
6096 | ||
eafe8130 TL |
6097 | void BlueStore::_fsck_collections(int64_t* errors) |
6098 | { | |
6099 | if (collections_had_errors) { | |
6100 | dout(10) << __func__ << dendl; | |
6101 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); | |
6102 | for (it->upper_bound(string()); | |
6103 | it->valid(); | |
6104 | it->next()) { | |
6105 | coll_t cid; | |
6106 | if (!cid.parse(it->key())) { | |
6107 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
6108 | if (errors) { | |
6109 | (*errors)++; | |
6110 | } | |
6111 | } | |
6112 | } | |
6113 | } | |
6114 | } | |
6115 | ||
9f95a23c TL |
6116 | void BlueStore::_set_per_pool_omap() |
6117 | { | |
6118 | per_pool_omap = false; | |
6119 | bufferlist bl; | |
6120 | db->get(PREFIX_SUPER, "per_pool_omap", &bl); | |
6121 | if (bl.length()) { | |
6122 | per_pool_omap = true; | |
6123 | dout(10) << __func__ << " per_pool_omap=1" << dendl; | |
6124 | } else { | |
6125 | dout(10) << __func__ << " per_pool_omap not present" << dendl; | |
6126 | } | |
6127 | _check_no_per_pool_omap_alert(); | |
6128 | } | |
6129 | ||
224ce89b | 6130 | void BlueStore::_open_statfs() |
31f18b77 | 6131 | { |
11fdf7f2 TL |
6132 | osd_pools.clear(); |
6133 | vstatfs.reset(); | |
6134 | ||
31f18b77 | 6135 | bufferlist bl; |
11fdf7f2 | 6136 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 6137 | if (r >= 0) { |
11fdf7f2 | 6138 | per_pool_stat_collection = false; |
31f18b77 | 6139 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 6140 | auto it = bl.cbegin(); |
31f18b77 | 6141 | vstatfs.decode(it); |
11fdf7f2 | 6142 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 6143 | } else { |
31f18b77 FG |
6144 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
6145 | } | |
81eedcae | 6146 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
6147 | } else { |
6148 | per_pool_stat_collection = true; | |
6149 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
6150 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT); | |
6151 | for (it->upper_bound(string()); | |
6152 | it->valid(); | |
6153 | it->next()) { | |
6154 | ||
6155 | uint64_t pool_id; | |
6156 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6157 | ceph_assert(r == 0); | |
6158 | ||
6159 | bufferlist bl; | |
6160 | bl = it->value(); | |
6161 | auto p = bl.cbegin(); | |
6162 | auto& st = osd_pools[pool_id]; | |
6163 | try { | |
6164 | st.decode(p); | |
6165 | vstatfs += st; | |
6166 | ||
6167 | dout(30) << __func__ << " pool " << pool_id | |
6168 | << " statfs " << st << dendl; | |
6169 | } catch (buffer::error& e) { | |
6170 | derr << __func__ << " failed to decode pool stats, key:" | |
6171 | << pretty_binary_string(it->key()) << dendl; | |
6172 | } | |
6173 | } | |
31f18b77 | 6174 | } |
11fdf7f2 TL |
6175 | dout(30) << __func__ << " statfs " << vstatfs << dendl; |
6176 | ||
31f18b77 FG |
6177 | } |
6178 | ||
7c673cae FG |
6179 | int BlueStore::_setup_block_symlink_or_file( |
6180 | string name, | |
6181 | string epath, | |
6182 | uint64_t size, | |
6183 | bool create) | |
6184 | { | |
6185 | dout(20) << __func__ << " name " << name << " path " << epath | |
6186 | << " size " << size << " create=" << (int)create << dendl; | |
6187 | int r = 0; | |
91327a77 | 6188 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6189 | if (create) |
6190 | flags |= O_CREAT; | |
6191 | if (epath.length()) { | |
6192 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
6193 | if (r < 0) { | |
6194 | r = -errno; | |
6195 | derr << __func__ << " failed to create " << name << " symlink to " | |
6196 | << epath << ": " << cpp_strerror(r) << dendl; | |
6197 | return r; | |
6198 | } | |
6199 | ||
6200 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
6201 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
6202 | if (fd < 0) { | |
6203 | r = -errno; | |
6204 | derr << __func__ << " failed to open " << epath << " file: " | |
6205 | << cpp_strerror(r) << dendl; | |
6206 | return r; | |
6207 | } | |
11fdf7f2 TL |
6208 | // write the Transport ID of the NVMe device |
6209 | // a transport id looks like: "trtype:PCIe traddr:0000:02:00.0" | |
6210 | // where "0000:02:00.0" is the selector of a PCI device, see | |
6211 | // the first column of "lspci -mm -n -D" | |
6212 | string trid{"trtype:PCIe "}; | |
6213 | trid += "traddr:"; | |
6214 | trid += epath.substr(strlen(SPDK_PREFIX)); | |
6215 | r = ::write(fd, trid.c_str(), trid.size()); | |
6216 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
6217 | dout(1) << __func__ << " created " << name << " symlink to " |
6218 | << epath << dendl; | |
6219 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6220 | } | |
6221 | } | |
6222 | if (size) { | |
6223 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
6224 | if (fd >= 0) { | |
6225 | // block file is present | |
6226 | struct stat st; | |
6227 | int r = ::fstat(fd, &st); | |
6228 | if (r == 0 && | |
6229 | S_ISREG(st.st_mode) && // if it is a regular file | |
6230 | st.st_size == 0) { // and is 0 bytes | |
6231 | r = ::ftruncate(fd, size); | |
6232 | if (r < 0) { | |
6233 | r = -errno; | |
6234 | derr << __func__ << " failed to resize " << name << " file to " | |
6235 | << size << ": " << cpp_strerror(r) << dendl; | |
6236 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6237 | return r; | |
6238 | } | |
6239 | ||
6240 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
6241 | r = ::ceph_posix_fallocate(fd, 0, size); |
6242 | if (r > 0) { | |
7c673cae FG |
6243 | derr << __func__ << " failed to prefallocate " << name << " file to " |
6244 | << size << ": " << cpp_strerror(r) << dendl; | |
6245 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6246 | return -r; | |
6247 | } | |
7c673cae FG |
6248 | } |
6249 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 6250 | << byte_u_t(size) << dendl; |
7c673cae FG |
6251 | } |
6252 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6253 | } else { | |
6254 | int r = -errno; | |
6255 | if (r != -ENOENT) { | |
6256 | derr << __func__ << " failed to open " << name << " file: " | |
6257 | << cpp_strerror(r) << dendl; | |
6258 | return r; | |
6259 | } | |
6260 | } | |
6261 | } | |
6262 | return 0; | |
6263 | } | |
6264 | ||
6265 | int BlueStore::mkfs() | |
6266 | { | |
6267 | dout(1) << __func__ << " path " << path << dendl; | |
6268 | int r; | |
6269 | uuid_d old_fsid; | |
6270 | ||
eafe8130 TL |
6271 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6272 | derr << __func__ << " osd_max_object_size " | |
6273 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6274 | << OBJECT_MAX_SIZE << dendl; | |
6275 | return -EINVAL; | |
6276 | } | |
6277 | ||
7c673cae FG |
6278 | { |
6279 | string done; | |
6280 | r = read_meta("mkfs_done", &done); | |
6281 | if (r == 0) { | |
6282 | dout(1) << __func__ << " already created" << dendl; | |
6283 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
6284 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6285 | if (r < 0) { | |
6286 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
6287 | << dendl; | |
6288 | return r; | |
6289 | } | |
6290 | if (r > 0) { | |
6291 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
6292 | r = -EIO; | |
6293 | } | |
6294 | } | |
6295 | return r; // idempotent | |
6296 | } | |
6297 | } | |
6298 | ||
6299 | { | |
6300 | string type; | |
6301 | r = read_meta("type", &type); | |
6302 | if (r == 0) { | |
6303 | if (type != "bluestore") { | |
6304 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6305 | return -EIO; | |
6306 | } | |
6307 | } else { | |
6308 | r = write_meta("type", "bluestore"); | |
6309 | if (r < 0) | |
6310 | return r; | |
6311 | } | |
6312 | } | |
6313 | ||
6314 | freelist_type = "bitmap"; | |
6315 | ||
6316 | r = _open_path(); | |
6317 | if (r < 0) | |
6318 | return r; | |
6319 | ||
6320 | r = _open_fsid(true); | |
6321 | if (r < 0) | |
6322 | goto out_path_fd; | |
6323 | ||
6324 | r = _lock_fsid(); | |
6325 | if (r < 0) | |
6326 | goto out_close_fsid; | |
6327 | ||
6328 | r = _read_fsid(&old_fsid); | |
6329 | if (r < 0 || old_fsid.is_zero()) { | |
6330 | if (fsid.is_zero()) { | |
6331 | fsid.generate_random(); | |
6332 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
6333 | } else { | |
6334 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
6335 | } | |
6336 | // we'll write it later. | |
6337 | } else { | |
6338 | if (!fsid.is_zero() && fsid != old_fsid) { | |
6339 | derr << __func__ << " on-disk fsid " << old_fsid | |
6340 | << " != provided " << fsid << dendl; | |
6341 | r = -EINVAL; | |
6342 | goto out_close_fsid; | |
6343 | } | |
6344 | fsid = old_fsid; | |
6345 | } | |
6346 | ||
6347 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
6348 | cct->_conf->bluestore_block_size, | |
6349 | cct->_conf->bluestore_block_create); | |
6350 | if (r < 0) | |
6351 | goto out_close_fsid; | |
6352 | if (cct->_conf->bluestore_bluefs) { | |
6353 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
6354 | cct->_conf->bluestore_block_wal_size, | |
6355 | cct->_conf->bluestore_block_wal_create); | |
6356 | if (r < 0) | |
6357 | goto out_close_fsid; | |
6358 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
6359 | cct->_conf->bluestore_block_db_size, | |
6360 | cct->_conf->bluestore_block_db_create); | |
6361 | if (r < 0) | |
6362 | goto out_close_fsid; | |
6363 | } | |
6364 | ||
6365 | r = _open_bdev(true); | |
6366 | if (r < 0) | |
6367 | goto out_close_fsid; | |
6368 | ||
3efd9988 FG |
6369 | // choose min_alloc_size |
6370 | if (cct->_conf->bluestore_min_alloc_size) { | |
6371 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; | |
6372 | } else { | |
11fdf7f2 | 6373 | ceph_assert(bdev); |
3efd9988 FG |
6374 | if (bdev->is_rotational()) { |
6375 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; | |
6376 | } else { | |
6377 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
6378 | } | |
6379 | } | |
11fdf7f2 | 6380 | _validate_bdev(); |
3efd9988 FG |
6381 | |
6382 | // make sure min_alloc_size is power of 2 aligned. | |
11fdf7f2 | 6383 | if (!isp2(min_alloc_size)) { |
3efd9988 FG |
6384 | derr << __func__ << " min_alloc_size 0x" |
6385 | << std::hex << min_alloc_size << std::dec | |
6386 | << " is not power of 2 aligned!" | |
6387 | << dendl; | |
6388 | r = -EINVAL; | |
6389 | goto out_close_bdev; | |
6390 | } | |
6391 | ||
7c673cae FG |
6392 | r = _open_db(true); |
6393 | if (r < 0) | |
6394 | goto out_close_bdev; | |
6395 | ||
7c673cae FG |
6396 | { |
6397 | KeyValueDB::Transaction t = db->get_transaction(); | |
11fdf7f2 TL |
6398 | r = _open_fm(t); |
6399 | if (r < 0) | |
6400 | goto out_close_db; | |
7c673cae FG |
6401 | { |
6402 | bufferlist bl; | |
11fdf7f2 | 6403 | encode((uint64_t)0, bl); |
7c673cae FG |
6404 | t->set(PREFIX_SUPER, "nid_max", bl); |
6405 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
6406 | } | |
6407 | ||
7c673cae FG |
6408 | { |
6409 | bufferlist bl; | |
11fdf7f2 | 6410 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
6411 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
6412 | } | |
9f95a23c TL |
6413 | { |
6414 | bufferlist bl; | |
6415 | bl.append("1"); | |
6416 | t->set(PREFIX_SUPER, "per_pool_omap", bl); | |
6417 | } | |
7c673cae FG |
6418 | ondisk_format = latest_ondisk_format; |
6419 | _prepare_ondisk_format_super(t); | |
6420 | db->submit_transaction_sync(t); | |
6421 | } | |
6422 | ||
7c673cae FG |
6423 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
6424 | if (r < 0) | |
224ce89b WB |
6425 | goto out_close_fm; |
6426 | ||
3efd9988 | 6427 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 6428 | if (r < 0) |
224ce89b | 6429 | goto out_close_fm; |
7c673cae FG |
6430 | |
6431 | if (fsid != old_fsid) { | |
6432 | r = _write_fsid(); | |
6433 | if (r < 0) { | |
6434 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 6435 | goto out_close_fm; |
7c673cae FG |
6436 | } |
6437 | } | |
6438 | ||
11fdf7f2 TL |
6439 | if (out_of_sync_fm.fetch_and(0)) { |
6440 | _sync_bluefs_and_fm(); | |
6441 | } | |
6442 | ||
7c673cae FG |
6443 | out_close_fm: |
6444 | _close_fm(); | |
6445 | out_close_db: | |
6446 | _close_db(); | |
6447 | out_close_bdev: | |
6448 | _close_bdev(); | |
6449 | out_close_fsid: | |
6450 | _close_fsid(); | |
6451 | out_path_fd: | |
6452 | _close_path(); | |
6453 | ||
6454 | if (r == 0 && | |
6455 | cct->_conf->bluestore_fsck_on_mkfs) { | |
6456 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6457 | if (rc < 0) | |
6458 | return rc; | |
6459 | if (rc > 0) { | |
6460 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6461 | r = -EIO; | |
6462 | } | |
11fdf7f2 TL |
6463 | } |
6464 | ||
6465 | if (r == 0) { | |
6466 | // indicate success by writing the 'mkfs_done' file | |
6467 | r = write_meta("mkfs_done", "yes"); | |
6468 | } | |
6469 | ||
6470 | if (r < 0) { | |
6471 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6472 | } else { | |
6473 | dout(0) << __func__ << " success" << dendl; | |
6474 | } | |
6475 | return r; | |
6476 | } | |
6477 | ||
6478 | int BlueStore::_mount_for_bluefs() | |
6479 | { | |
6480 | int r = _open_path(); | |
6481 | ceph_assert(r == 0); | |
6482 | r = _open_fsid(false); | |
6483 | ceph_assert(r == 0); | |
6484 | r = _read_fsid(&fsid); | |
6485 | ceph_assert(r == 0); | |
6486 | r = _lock_fsid(); | |
6487 | ceph_assert(r == 0); | |
6488 | r = _open_bluefs(false); | |
6489 | ceph_assert(r == 0); | |
6490 | return r; | |
6491 | } | |
6492 | ||
6493 | void BlueStore::_umount_for_bluefs() | |
6494 | { | |
6495 | _close_bluefs(); | |
6496 | _close_fsid(); | |
6497 | _close_path(); | |
6498 | } | |
6499 | ||
6500 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) | |
6501 | { | |
6502 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6503 | int r; | |
6504 | ceph_assert(path_fd < 0); | |
6505 | ||
6506 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6507 | ||
6508 | if (!cct->_conf->bluestore_bluefs) { | |
6509 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6510 | return -EIO; | |
6511 | } | |
6512 | ||
6513 | r = _mount_for_bluefs(); | |
6514 | ||
6515 | int reserved = 0; | |
6516 | if (id == BlueFS::BDEV_NEWWAL) { | |
6517 | string p = path + "/block.wal"; | |
6518 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
6519 | cct->_conf->bluestore_block_wal_size, | |
6520 | true); | |
6521 | ceph_assert(r == 0); | |
6522 | ||
6523 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
eafe8130 | 6524 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6525 | ceph_assert(r == 0); |
6526 | ||
6527 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6528 | r = _check_or_set_bdev_label( | |
6529 | p, | |
6530 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6531 | "bluefs wal", | |
6532 | true); | |
6533 | ceph_assert(r == 0); | |
6534 | } | |
6535 | ||
6536 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
9f95a23c | 6537 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6538 | } else if (id == BlueFS::BDEV_NEWDB) { |
6539 | string p = path + "/block.db"; | |
6540 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
6541 | cct->_conf->bluestore_block_db_size, | |
6542 | true); | |
6543 | ceph_assert(r == 0); | |
6544 | ||
6545 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
eafe8130 | 6546 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6547 | ceph_assert(r == 0); |
6548 | ||
6549 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6550 | r = _check_or_set_bdev_label( | |
6551 | p, | |
6552 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6553 | "bluefs db", | |
6554 | true); | |
6555 | ceph_assert(r == 0); | |
6556 | } | |
6557 | reserved = SUPER_RESERVED; | |
9f95a23c TL |
6558 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6559 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
6560 | } |
6561 | ||
6562 | bluefs->umount(); | |
6563 | bluefs->mount(); | |
6564 | ||
6565 | bluefs->add_block_extent( | |
6566 | id, | |
6567 | reserved, | |
6568 | bluefs->get_block_device_size(id) - reserved); | |
6569 | ||
9f95a23c | 6570 | r = bluefs->prepare_new_device(id, bluefs_layout); |
11fdf7f2 TL |
6571 | ceph_assert(r == 0); |
6572 | ||
6573 | if (r < 0) { | |
6574 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6575 | } else { | |
6576 | dout(0) << __func__ << " success" << dendl; | |
6577 | } | |
6578 | ||
6579 | _umount_for_bluefs(); | |
6580 | return r; | |
6581 | } | |
6582 | ||
6583 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
6584 | int id) | |
6585 | { | |
6586 | dout(10) << __func__ << " id:" << id << dendl; | |
6587 | ceph_assert(path_fd < 0); | |
6588 | ||
6589 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
6590 | ||
6591 | if (!cct->_conf->bluestore_bluefs) { | |
6592 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6593 | return -EIO; | |
6594 | } | |
6595 | ||
6596 | int r = _mount_for_bluefs(); | |
6597 | ||
6598 | // require bluestore_bluefs_min_free to be free at target device! | |
6599 | uint64_t used_space = cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
6600 | for(auto src_id : devs_source) { | |
6601 | used_space += bluefs->get_total(src_id) - bluefs->get_free(src_id); | |
6602 | } | |
6603 | uint64_t target_free = bluefs->get_free(id); | |
6604 | if (id == BlueFS::BDEV_SLOW && target_free < used_space) { | |
6605 | // will need to remount full BlueStore instance to allocate more space | |
6606 | _umount_for_bluefs(); | |
6607 | ||
6608 | r = mount(); | |
6609 | ceph_assert(r == 0); | |
6610 | dout(1) << __func__ | |
6611 | << " Allocating more space at slow device for BlueFS: +" | |
6612 | << used_space - target_free << " bytes" << dendl; | |
6613 | r = allocate_bluefs_freespace( | |
6614 | used_space - target_free, | |
6615 | used_space - target_free, | |
6616 | nullptr); | |
6617 | ||
6618 | umount(); | |
6619 | if (r != 0) { | |
6620 | derr << __func__ | |
6621 | << " can't migrate, unable to allocate extra space: " | |
6622 | << used_space - target_free << " at target:" << id | |
6623 | << dendl; | |
6624 | return -ENOSPC; | |
6625 | } | |
6626 | ||
6627 | r = _mount_for_bluefs(); | |
6628 | ceph_assert(r == 0); | |
6629 | } else if (target_free < used_space) { | |
6630 | derr << __func__ | |
6631 | << " can't migrate, free space at target: " << target_free | |
6632 | << " is less than required space: " << used_space | |
6633 | << dendl; | |
6634 | return -ENOSPC; | |
6635 | } | |
9f95a23c TL |
6636 | if (devs_source.count(BlueFS::BDEV_DB)) { |
6637 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; | |
6638 | bluefs_layout.dedicated_db = false; | |
6639 | } | |
6640 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6641 | bluefs_layout.dedicated_wal = false; | |
6642 | } | |
6643 | r = bluefs->device_migrate_to_existing(cct, devs_source, id, bluefs_layout); | |
11fdf7f2 TL |
6644 | if (r < 0) { |
6645 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
6646 | goto shutdown; | |
6647 | } | |
6648 | ||
6649 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
6650 | r = unlink(string(path + "/block.db").c_str()); | |
6651 | ceph_assert(r == 0); | |
6652 | } | |
6653 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6654 | r = unlink(string(path + "/block.wal").c_str()); | |
6655 | ceph_assert(r == 0); | |
6656 | } | |
6657 | ||
6658 | shutdown: | |
6659 | _umount_for_bluefs(); | |
6660 | return r; | |
6661 | } | |
6662 | ||
6663 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
6664 | int id, | |
6665 | const string& dev_path) | |
6666 | { | |
6667 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6668 | int r; | |
6669 | ceph_assert(path_fd < 0); | |
6670 | ||
6671 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6672 | ||
6673 | if (!cct->_conf->bluestore_bluefs) { | |
6674 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6675 | return -EIO; | |
6676 | } | |
6677 | ||
6678 | r = _mount_for_bluefs(); | |
6679 | ||
6680 | int reserved = 0; | |
6681 | string link_db; | |
6682 | string link_wal; | |
6683 | if (devs_source.count(BlueFS::BDEV_DB) && | |
9f95a23c | 6684 | bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 6685 | link_db = path + "/block.db"; |
9f95a23c TL |
6686 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
6687 | bluefs_layout.dedicated_db = false; | |
11fdf7f2 TL |
6688 | } |
6689 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6690 | link_wal = path + "/block.wal"; | |
9f95a23c | 6691 | bluefs_layout.dedicated_wal = false; |
11fdf7f2 TL |
6692 | } |
6693 | ||
6694 | size_t target_size; | |
6695 | string target_name; | |
6696 | if (id == BlueFS::BDEV_NEWWAL) { | |
6697 | target_name = "block.wal"; | |
6698 | target_size = cct->_conf->bluestore_block_wal_size; | |
9f95a23c | 6699 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6700 | |
6701 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
eafe8130 | 6702 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6703 | ceph_assert(r == 0); |
6704 | ||
6705 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6706 | r = _check_or_set_bdev_label( | |
6707 | dev_path, | |
6708 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6709 | "bluefs wal", | |
6710 | true); | |
6711 | ceph_assert(r == 0); | |
6712 | } | |
6713 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
6714 | } else if (id == BlueFS::BDEV_NEWDB) { | |
6715 | target_name = "block.db"; | |
6716 | target_size = cct->_conf->bluestore_block_db_size; | |
9f95a23c TL |
6717 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6718 | bluefs_layout.dedicated_db = true; | |
31f18b77 | 6719 | |
11fdf7f2 | 6720 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
eafe8130 | 6721 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6722 | ceph_assert(r == 0); |
6723 | ||
6724 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6725 | r = _check_or_set_bdev_label( | |
6726 | dev_path, | |
6727 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6728 | "bluefs db", | |
6729 | true); | |
6730 | ceph_assert(r == 0); | |
6731 | } | |
6732 | reserved = SUPER_RESERVED; | |
31f18b77 FG |
6733 | } |
6734 | ||
11fdf7f2 TL |
6735 | bluefs->umount(); |
6736 | bluefs->mount(); | |
6737 | ||
6738 | bluefs->add_block_extent( | |
6739 | id, reserved, bluefs->get_block_device_size(id) - reserved); | |
6740 | ||
9f95a23c | 6741 | r = bluefs->device_migrate_to_new(cct, devs_source, id, bluefs_layout); |
11fdf7f2 | 6742 | |
7c673cae | 6743 | if (r < 0) { |
11fdf7f2 TL |
6744 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
6745 | goto shutdown; | |
6746 | } | |
6747 | ||
6748 | if (!link_db.empty()) { | |
6749 | r = unlink(link_db.c_str()); | |
6750 | ceph_assert(r == 0); | |
6751 | } | |
6752 | if (!link_wal.empty()) { | |
6753 | r = unlink(link_wal.c_str()); | |
6754 | ceph_assert(r == 0); | |
6755 | } | |
6756 | r = _setup_block_symlink_or_file( | |
6757 | target_name, | |
6758 | dev_path, | |
6759 | target_size, | |
6760 | true); | |
6761 | ceph_assert(r == 0); | |
6762 | dout(0) << __func__ << " success" << dendl; | |
6763 | ||
6764 | shutdown: | |
6765 | _umount_for_bluefs(); | |
6766 | return r; | |
6767 | } | |
6768 | ||
6769 | string BlueStore::get_device_path(unsigned id) | |
6770 | { | |
6771 | string res; | |
6772 | if (id < BlueFS::MAX_BDEV) { | |
6773 | switch (id) { | |
6774 | case BlueFS::BDEV_WAL: | |
6775 | res = path + "/block.wal"; | |
6776 | break; | |
6777 | case BlueFS::BDEV_DB: | |
9f95a23c | 6778 | if (id == bluefs_layout.shared_bdev) { |
11fdf7f2 TL |
6779 | res = path + "/block"; |
6780 | } else { | |
6781 | res = path + "/block.db"; | |
6782 | } | |
6783 | break; | |
6784 | case BlueFS::BDEV_SLOW: | |
6785 | res = path + "/block"; | |
6786 | break; | |
6787 | } | |
6788 | } | |
6789 | return res; | |
6790 | } | |
6791 | ||
6792 | int BlueStore::expand_devices(ostream& out) | |
6793 | { | |
6794 | int r = _mount(false); | |
6795 | ceph_assert(r == 0); | |
6796 | bluefs->dump_block_extents(out); | |
6797 | out << "Expanding..." << std::endl; | |
6798 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { | |
9f95a23c | 6799 | if (devid == bluefs_layout.shared_bdev ) { |
11fdf7f2 TL |
6800 | continue; |
6801 | } | |
6802 | uint64_t size = bluefs->get_block_device_size(devid); | |
6803 | if (size == 0) { | |
6804 | // no bdev | |
6805 | continue; | |
6806 | } | |
6807 | ||
6808 | interval_set<uint64_t> before; | |
6809 | bluefs->get_block_extents(devid, &before); | |
6810 | ceph_assert(!before.empty()); | |
6811 | uint64_t end = before.range_end(); | |
6812 | if (end < size) { | |
6813 | out << devid | |
6814 | <<" : expanding " << " from 0x" << std::hex | |
6815 | << end << " to 0x" << size << std::dec << std::endl; | |
6816 | bluefs->add_block_extent(devid, end, size-end); | |
6817 | string p = get_device_path(devid); | |
6818 | const char* path = p.c_str(); | |
6819 | if (path == nullptr) { | |
6820 | derr << devid | |
6821 | <<": can't find device path " << dendl; | |
6822 | continue; | |
6823 | } | |
6824 | bluestore_bdev_label_t label; | |
6825 | int r = _read_bdev_label(cct, path, &label); | |
6826 | if (r < 0) { | |
6827 | derr << "unable to read label for " << path << ": " | |
6828 | << cpp_strerror(r) << dendl; | |
6829 | continue; | |
6830 | } | |
6831 | label.size = size; | |
6832 | r = _write_bdev_label(cct, path, label); | |
6833 | if (r < 0) { | |
6834 | derr << "unable to write label for " << path << ": " | |
6835 | << cpp_strerror(r) << dendl; | |
6836 | continue; | |
6837 | } | |
6838 | out << devid | |
6839 | <<" : size label updated to " << size | |
6840 | << std::endl; | |
6841 | } | |
6842 | } | |
6843 | uint64_t size0 = fm->get_size(); | |
6844 | uint64_t size = bdev->get_size(); | |
6845 | if (size0 < size) { | |
9f95a23c | 6846 | out << bluefs_layout.shared_bdev |
11fdf7f2 TL |
6847 | <<" : expanding " << " from 0x" << std::hex |
6848 | << size0 << " to 0x" << size << std::dec << std::endl; | |
6849 | KeyValueDB::Transaction txn; | |
6850 | txn = db->get_transaction(); | |
6851 | int r = fm->expand(size, txn); | |
6852 | ceph_assert(r == 0); | |
6853 | db->submit_transaction_sync(txn); | |
6854 | ||
6855 | // always reference to slow device here | |
6856 | string p = get_device_path(BlueFS::BDEV_SLOW); | |
6857 | ceph_assert(!p.empty()); | |
6858 | const char* path = p.c_str(); | |
6859 | bluestore_bdev_label_t label; | |
6860 | r = _read_bdev_label(cct, path, &label); | |
6861 | if (r < 0) { | |
6862 | derr << "unable to read label for " << path << ": " | |
6863 | << cpp_strerror(r) << dendl; | |
6864 | } else { | |
6865 | label.size = size; | |
6866 | r = _write_bdev_label(cct, path, label); | |
6867 | if (r < 0) { | |
6868 | derr << "unable to write label for " << path << ": " | |
6869 | << cpp_strerror(r) << dendl; | |
6870 | } else { | |
9f95a23c | 6871 | out << bluefs_layout.shared_bdev |
11fdf7f2 TL |
6872 | <<" : size label updated to " << size |
6873 | << std::endl; | |
6874 | } | |
6875 | } | |
7c673cae | 6876 | } |
11fdf7f2 | 6877 | umount(); |
7c673cae FG |
6878 | return r; |
6879 | } | |
6880 | ||
6881 | void BlueStore::set_cache_shards(unsigned num) | |
6882 | { | |
6883 | dout(10) << __func__ << " " << num << dendl; | |
9f95a23c TL |
6884 | size_t oold = onode_cache_shards.size(); |
6885 | size_t bold = buffer_cache_shards.size(); | |
6886 | ceph_assert(num >= oold && num >= bold); | |
6887 | onode_cache_shards.resize(num); | |
6888 | buffer_cache_shards.resize(num); | |
6889 | for (unsigned i = oold; i < num; ++i) { | |
6890 | onode_cache_shards[i] = | |
6891 | OnodeCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6892 | logger); | |
6893 | } | |
6894 | for (unsigned i = bold; i < num; ++i) { | |
6895 | buffer_cache_shards[i] = | |
6896 | BufferCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6897 | logger); | |
7c673cae FG |
6898 | } |
6899 | } | |
6900 | ||
11fdf7f2 | 6901 | int BlueStore::_mount(bool kv_only, bool open_db) |
7c673cae FG |
6902 | { |
6903 | dout(1) << __func__ << " path " << path << dendl; | |
6904 | ||
3efd9988 FG |
6905 | _kv_only = kv_only; |
6906 | ||
7c673cae FG |
6907 | { |
6908 | string type; | |
6909 | int r = read_meta("type", &type); | |
6910 | if (r < 0) { | |
6911 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
6912 | << dendl; | |
6913 | return r; | |
6914 | } | |
6915 | ||
6916 | if (type != "bluestore") { | |
6917 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6918 | return -EIO; | |
6919 | } | |
6920 | } | |
6921 | ||
6922 | if (cct->_conf->bluestore_fsck_on_mount) { | |
6923 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); | |
6924 | if (rc < 0) | |
6925 | return rc; | |
6926 | if (rc > 0) { | |
6927 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6928 | return -EIO; | |
6929 | } | |
6930 | } | |
6931 | ||
eafe8130 TL |
6932 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6933 | derr << __func__ << " osd_max_object_size " | |
6934 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6935 | << OBJECT_MAX_SIZE << dendl; | |
6936 | return -EINVAL; | |
6937 | } | |
6938 | ||
7c673cae FG |
6939 | int r = _open_path(); |
6940 | if (r < 0) | |
6941 | return r; | |
6942 | r = _open_fsid(false); | |
6943 | if (r < 0) | |
6944 | goto out_path; | |
6945 | ||
6946 | r = _read_fsid(&fsid); | |
6947 | if (r < 0) | |
6948 | goto out_fsid; | |
6949 | ||
6950 | r = _lock_fsid(); | |
6951 | if (r < 0) | |
6952 | goto out_fsid; | |
6953 | ||
6954 | r = _open_bdev(false); | |
6955 | if (r < 0) | |
6956 | goto out_fsid; | |
6957 | ||
11fdf7f2 TL |
6958 | if (open_db) { |
6959 | r = _open_db_and_around(false); | |
6960 | } else { | |
6961 | // we can bypass db open exclusively in case of kv_only mode | |
6962 | ceph_assert(kv_only); | |
6963 | r = _open_db(false, true); | |
9f95a23c TL |
6964 | } |
6965 | if (r < 0) { | |
6966 | goto out_bdev; | |
11fdf7f2 | 6967 | } |
7c673cae FG |
6968 | |
6969 | if (kv_only) | |
6970 | return 0; | |
6971 | ||
11fdf7f2 TL |
6972 | r = _upgrade_super(); |
6973 | if (r < 0) { | |
7c673cae | 6974 | goto out_db; |
11fdf7f2 | 6975 | } |
7c673cae FG |
6976 | |
6977 | r = _open_collections(); | |
6978 | if (r < 0) | |
11fdf7f2 | 6979 | goto out_db; |
7c673cae FG |
6980 | |
6981 | r = _reload_logger(); | |
6982 | if (r < 0) | |
6983 | goto out_coll; | |
6984 | ||
31f18b77 | 6985 | _kv_start(); |
7c673cae FG |
6986 | |
6987 | r = _deferred_replay(); | |
6988 | if (r < 0) | |
6989 | goto out_stop; | |
6990 | ||
6991 | mempool_thread.init(); | |
6992 | ||
9f95a23c | 6993 | if ((!per_pool_stat_collection || !per_pool_omap) && |
eafe8130 | 6994 | cct->_conf->bluestore_fsck_quick_fix_on_mount == true) { |
9f95a23c TL |
6995 | |
6996 | bool was_per_pool_omap = per_pool_omap; | |
6997 | ||
eafe8130 TL |
6998 | dout(1) << __func__ << " quick-fix on mount" << dendl; |
6999 | _fsck_on_open(FSCK_SHALLOW, true); | |
7000 | ||
7001 | //reread statfs | |
7002 | //FIXME minor: replace with actual open/close? | |
7003 | _open_statfs(); | |
eafe8130 | 7004 | _check_legacy_statfs_alert(); |
9f95a23c TL |
7005 | |
7006 | //set again as hopefully it has been fixed | |
7007 | if (!was_per_pool_omap) { | |
7008 | _set_per_pool_omap(); | |
7009 | } | |
eafe8130 TL |
7010 | } |
7011 | ||
7c673cae FG |
7012 | mounted = true; |
7013 | return 0; | |
7014 | ||
7015 | out_stop: | |
7016 | _kv_stop(); | |
7c673cae | 7017 | out_coll: |
31f18b77 | 7018 | _flush_cache(); |
7c673cae | 7019 | out_db: |
11fdf7f2 | 7020 | _close_db_and_around(); |
7c673cae FG |
7021 | out_bdev: |
7022 | _close_bdev(); | |
7023 | out_fsid: | |
7024 | _close_fsid(); | |
7025 | out_path: | |
7026 | _close_path(); | |
7027 | return r; | |
7028 | } | |
7029 | ||
7030 | int BlueStore::umount() | |
7031 | { | |
11fdf7f2 | 7032 | ceph_assert(_kv_only || mounted); |
7c673cae FG |
7033 | dout(1) << __func__ << dendl; |
7034 | ||
7035 | _osr_drain_all(); | |
7c673cae | 7036 | |
7c673cae | 7037 | mounted = false; |
3efd9988 FG |
7038 | if (!_kv_only) { |
7039 | mempool_thread.shutdown(); | |
7040 | dout(20) << __func__ << " stopping kv thread" << dendl; | |
7041 | _kv_stop(); | |
3efd9988 FG |
7042 | _flush_cache(); |
7043 | dout(20) << __func__ << " closing" << dendl; | |
7044 | ||
3efd9988 | 7045 | } |
11fdf7f2 | 7046 | _close_db_and_around(); |
7c673cae FG |
7047 | _close_bdev(); |
7048 | _close_fsid(); | |
7049 | _close_path(); | |
7050 | ||
7051 | if (cct->_conf->bluestore_fsck_on_umount) { | |
7052 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); | |
7053 | if (rc < 0) | |
7054 | return rc; | |
7055 | if (rc > 0) { | |
7056 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7057 | return -EIO; | |
7058 | } | |
7059 | } | |
7060 | return 0; | |
7061 | } | |
7062 | ||
eafe8130 TL |
7063 | int BlueStore::cold_open() |
7064 | { | |
7065 | int r = _open_path(); | |
7066 | if (r < 0) | |
7067 | return r; | |
7068 | r = _open_fsid(false); | |
7069 | if (r < 0) | |
7070 | goto out_path; | |
7071 | ||
7072 | r = _read_fsid(&fsid); | |
7073 | if (r < 0) | |
7074 | goto out_fsid; | |
7075 | ||
7076 | r = _lock_fsid(); | |
7077 | if (r < 0) | |
7078 | goto out_fsid; | |
7079 | ||
7080 | r = _open_bdev(false); | |
7081 | if (r < 0) | |
7082 | goto out_fsid; | |
7083 | r = _open_db_and_around(true); | |
7084 | if (r < 0) { | |
7085 | goto out_bdev; | |
7086 | } | |
7087 | return 0; | |
7088 | out_bdev: | |
7089 | _close_bdev(); | |
7090 | out_fsid: | |
7091 | _close_fsid(); | |
7092 | out_path: | |
7093 | _close_path(); | |
7094 | return r; | |
7095 | } | |
7096 | int BlueStore::cold_close() | |
7097 | { | |
7098 | _close_db_and_around(); | |
7099 | _close_bdev(); | |
7100 | _close_fsid(); | |
7101 | _close_path(); | |
7102 | return 0; | |
7103 | } | |
7104 | ||
9f95a23c TL |
7105 | // derr wrapper to limit enormous output and avoid log flooding. |
7106 | // Of limited use where such output is expected for now | |
7107 | #define fsck_derr(err_cnt, threshold) \ | |
7108 | if (err_cnt <= threshold) { \ | |
7109 | bool need_skip_print = err_cnt == threshold; \ | |
7110 | derr | |
7111 | ||
7112 | #define fsck_dendl \ | |
7113 | dendl; \ | |
7114 | if (need_skip_print) \ | |
7115 | derr << "more error lines skipped..." << dendl; \ | |
7c673cae | 7116 | } |
7c673cae | 7117 | |
eafe8130 TL |
7118 | int _fsck_sum_extents( |
7119 | const PExtentVector& extents, | |
7120 | bool compressed, | |
7121 | store_statfs_t& expected_statfs) | |
7122 | { | |
7123 | for (auto e : extents) { | |
7124 | if (!e.is_valid()) | |
7125 | continue; | |
7126 | expected_statfs.allocated += e.length; | |
7127 | if (compressed) { | |
7128 | expected_statfs.data_compressed_allocated += e.length; | |
7129 | } | |
7130 | } | |
7131 | return 0; | |
7132 | } | |
7133 | ||
7c673cae | 7134 | int BlueStore::_fsck_check_extents( |
11fdf7f2 | 7135 | const coll_t& cid, |
7c673cae FG |
7136 | const ghobject_t& oid, |
7137 | const PExtentVector& extents, | |
7138 | bool compressed, | |
7139 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 7140 | uint64_t granularity, |
11fdf7f2 | 7141 | BlueStoreRepairer* repairer, |
eafe8130 TL |
7142 | store_statfs_t& expected_statfs, |
7143 | FSCKDepth depth) | |
7c673cae FG |
7144 | { |
7145 | dout(30) << __func__ << " oid " << oid << " extents " << extents << dendl; | |
7146 | int errors = 0; | |
7147 | for (auto e : extents) { | |
7148 | if (!e.is_valid()) | |
7149 | continue; | |
7150 | expected_statfs.allocated += e.length; | |
7151 | if (compressed) { | |
11fdf7f2 | 7152 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae | 7153 | } |
eafe8130 TL |
7154 | if (depth != FSCK_SHALLOW) { |
7155 | bool already = false; | |
9f95a23c | 7156 | apply_for_bitset_range( |
eafe8130 TL |
7157 | e.offset, e.length, granularity, used_blocks, |
7158 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
7159 | if (bs.test(pos)) { |
7160 | if (repairer) { | |
7161 | repairer->note_misreference( | |
7162 | pos * min_alloc_size, min_alloc_size, !already); | |
7163 | } | |
7164 | if (!already) { | |
7165 | derr << "fsck error: " << oid << " extent " << e | |
7166 | << " or a subset is already allocated (misreferenced)" << dendl; | |
7167 | ++errors; | |
7168 | already = true; | |
7169 | } | |
11fdf7f2 | 7170 | } |
eafe8130 TL |
7171 | else |
7172 | bs.set(pos); | |
7173 | }); | |
7174 | if (repairer) { | |
7175 | repairer->get_space_usage_tracker().set_used( e.offset, e.length, cid, oid); | |
7176 | } | |
11fdf7f2 | 7177 | |
eafe8130 TL |
7178 | if (e.end() > bdev->get_size()) { |
7179 | derr << "fsck error: " << oid << " extent " << e | |
7180 | << " past end of block device" << dendl; | |
7181 | ++errors; | |
7182 | } | |
7c673cae FG |
7183 | } |
7184 | } | |
7185 | return errors; | |
7186 | } | |
7187 | ||
11fdf7f2 TL |
7188 | void BlueStore::_fsck_check_pool_statfs( |
7189 | BlueStore::per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
7190 | int64_t& errors, |
7191 | int64_t& warnings, | |
11fdf7f2 TL |
7192 | BlueStoreRepairer* repairer) |
7193 | { | |
7194 | auto it = db->get_iterator(PREFIX_STAT); | |
7195 | if (it) { | |
7196 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7197 | string key = it->key(); | |
7198 | if (key == BLUESTORE_GLOBAL_STATFS_KEY) { | |
7199 | if (repairer) { | |
eafe8130 TL |
7200 | ++errors; |
7201 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
7202 | derr << "fsck error: " << "legacy statfs record found, removing" | |
11fdf7f2 TL |
7203 | << dendl; |
7204 | } | |
7205 | continue; | |
7206 | } | |
11fdf7f2 TL |
7207 | uint64_t pool_id; |
7208 | if (get_key_pool_stat(key, &pool_id) < 0) { | |
7209 | derr << "fsck error: bad key " << key | |
7210 | << "in statfs namespece" << dendl; | |
7211 | if (repairer) { | |
7212 | repairer->remove_key(db, PREFIX_STAT, key); | |
7213 | } | |
7214 | ++errors; | |
7215 | continue; | |
7216 | } | |
7217 | ||
7218 | volatile_statfs vstatfs; | |
7219 | bufferlist bl = it->value(); | |
7220 | auto blp = bl.cbegin(); | |
7221 | try { | |
7222 | vstatfs.decode(blp); | |
7223 | } catch (buffer::error& e) { | |
7224 | derr << "fsck error: failed to decode Pool StatFS record" | |
7225 | << pretty_binary_string(key) << dendl; | |
7226 | if (repairer) { | |
7227 | dout(20) << __func__ << " undecodable Pool StatFS record, key:'" | |
7228 | << pretty_binary_string(key) | |
7229 | << "', removing" << dendl; | |
7230 | repairer->remove_key(db, PREFIX_STAT, key); | |
7231 | } | |
7232 | ++errors; | |
7233 | vstatfs.reset(); | |
7234 | } | |
7235 | auto stat_it = expected_pool_statfs.find(pool_id); | |
7236 | if (stat_it == expected_pool_statfs.end()) { | |
7237 | if (vstatfs.is_empty()) { | |
7238 | // we don't consider that as an error since empty pool statfs | |
7239 | // are left in DB for now | |
7240 | dout(20) << "fsck inf: found empty stray Pool StatFS record for pool id 0x" | |
7241 | << std::hex << pool_id << std::dec << dendl; | |
7242 | if (repairer) { | |
7243 | // but we need to increment error count in case of repair | |
7244 | // to have proper counters at the end | |
7245 | // (as repairer increments recovery counter anyway). | |
7246 | ++errors; | |
7247 | } | |
7248 | } else { | |
7249 | derr << "fsck error: found stray Pool StatFS record for pool id 0x" | |
7250 | << std::hex << pool_id << std::dec << dendl; | |
7251 | ++errors; | |
7252 | } | |
7253 | if (repairer) { | |
7254 | repairer->remove_key(db, PREFIX_SHARED_BLOB, key); | |
7255 | } | |
7256 | continue; | |
7257 | } | |
7258 | store_statfs_t statfs; | |
7259 | vstatfs.publish(&statfs); | |
7260 | if (!(stat_it->second == statfs)) { | |
7261 | derr << "fsck error: actual " << statfs | |
7262 | << " != expected " << stat_it->second | |
7263 | << " for pool " | |
7264 | << std::hex << pool_id << std::dec << dendl; | |
7265 | if (repairer) { | |
7266 | repairer->fix_statfs(db, key, stat_it->second); | |
7267 | } | |
7268 | ++errors; | |
7269 | } | |
7270 | expected_pool_statfs.erase(stat_it); | |
7271 | } | |
7272 | } // if (it) | |
eafe8130 TL |
7273 | for (auto& s : expected_pool_statfs) { |
7274 | if (s.second.is_zero()) { | |
11fdf7f2 TL |
7275 | // we might lack empty statfs recs in DB |
7276 | continue; | |
7277 | } | |
7278 | derr << "fsck error: missing Pool StatFS record for pool " | |
eafe8130 | 7279 | << std::hex << s.first << std::dec << dendl; |
11fdf7f2 TL |
7280 | if (repairer) { |
7281 | string key; | |
eafe8130 TL |
7282 | get_pool_stat_key(s.first, &key); |
7283 | repairer->fix_statfs(db, key, s.second); | |
11fdf7f2 TL |
7284 | } |
7285 | ++errors; | |
7286 | } | |
eafe8130 | 7287 | if (!per_pool_stat_collection && |
eafe8130 TL |
7288 | repairer) { |
7289 | // by virtue of running this method, we correct the top-level | |
7290 | // error of having global stats | |
7291 | repairer->inc_repaired(); | |
7292 | } | |
11fdf7f2 TL |
7293 | } |
7294 | ||
eafe8130 TL |
7295 | BlueStore::OnodeRef BlueStore::fsck_check_objects_shallow( |
7296 | BlueStore::FSCKDepth depth, | |
7297 | int64_t pool_id, | |
7298 | BlueStore::CollectionRef c, | |
7299 | const ghobject_t& oid, | |
7300 | const string& key, | |
7301 | const bufferlist& value, | |
9f95a23c | 7302 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
7303 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
7304 | const BlueStore::FSCK_ObjectCtx& ctx) | |
7305 | { | |
7306 | auto& errors = ctx.errors; | |
7307 | auto& num_objects = ctx.num_objects; | |
7308 | auto& num_extents = ctx.num_extents; | |
7309 | auto& num_blobs = ctx.num_blobs; | |
7310 | auto& num_sharded_objects = ctx.num_sharded_objects; | |
7311 | auto& num_spanning_blobs = ctx.num_spanning_blobs; | |
7312 | auto used_blocks = ctx.used_blocks; | |
7313 | auto sb_info_lock = ctx.sb_info_lock; | |
7314 | auto& sb_info = ctx.sb_info; | |
7315 | auto repairer = ctx.repairer; | |
7316 | ||
7317 | store_statfs_t* res_statfs = (per_pool_stat_collection || repairer) ? | |
7318 | &ctx.expected_pool_statfs[pool_id] : | |
7319 | &ctx.expected_store_statfs; | |
7320 | ||
7321 | dout(10) << __func__ << " " << oid << dendl; | |
7322 | OnodeRef o; | |
7323 | o.reset(Onode::decode(c, oid, key, value)); | |
7324 | ++num_objects; | |
7c673cae | 7325 | |
eafe8130 | 7326 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); |
7c673cae | 7327 | |
eafe8130 TL |
7328 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
7329 | _dump_onode<30>(cct, *o); | |
7330 | // shards | |
7331 | if (!o->extent_map.shards.empty()) { | |
7332 | ++num_sharded_objects; | |
7333 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 7334 | ceph_assert(expecting_shards); |
eafe8130 TL |
7335 | for (auto& s : o->extent_map.shards) { |
7336 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
9f95a23c | 7337 | expecting_shards->push_back(string()); |
eafe8130 | 7338 | get_extent_shard_key(o->key, s.shard_info->offset, |
9f95a23c | 7339 | &expecting_shards->back()); |
eafe8130 TL |
7340 | if (s.shard_info->offset >= o->onode.size) { |
7341 | derr << "fsck error: " << oid << " shard 0x" << std::hex | |
7342 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size | |
7343 | << std::dec << dendl; | |
7344 | ++errors; | |
7345 | } | |
7346 | } | |
7347 | } | |
7348 | } | |
7c673cae | 7349 | |
eafe8130 TL |
7350 | // lextents |
7351 | uint64_t pos = 0; | |
7352 | mempool::bluestore_fsck::map<BlobRef, | |
7353 | bluestore_blob_use_tracker_t> ref_map; | |
7354 | for (auto& l : o->extent_map.extent_map) { | |
7355 | dout(20) << __func__ << " " << l << dendl; | |
7356 | if (l.logical_offset < pos) { | |
7357 | derr << "fsck error: " << oid << " lextent at 0x" | |
7358 | << std::hex << l.logical_offset | |
7359 | << " overlaps with the previous, which ends at 0x" << pos | |
7360 | << std::dec << dendl; | |
7361 | ++errors; | |
7362 | } | |
7363 | if (depth != FSCK_SHALLOW && | |
7364 | o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
7365 | derr << "fsck error: " << oid << " lextent at 0x" | |
7366 | << std::hex << l.logical_offset << "~" << l.length | |
7367 | << " spans a shard boundary" | |
7368 | << std::dec << dendl; | |
7369 | ++errors; | |
7370 | } | |
7371 | pos = l.logical_offset + l.length; | |
7372 | res_statfs->data_stored += l.length; | |
7373 | ceph_assert(l.blob); | |
7374 | const bluestore_blob_t& blob = l.blob->get_blob(); | |
7375 | ||
7376 | auto& ref = ref_map[l.blob]; | |
7377 | if (ref.is_empty()) { | |
7378 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
7379 | uint32_t l = blob.get_logical_length(); | |
7380 | ref.init(l, min_release_size); | |
7381 | } | |
7382 | ref.get( | |
7383 | l.blob_offset, | |
7384 | l.length); | |
7385 | ++num_extents; | |
7386 | if (depth != FSCK_SHALLOW && | |
7387 | blob.has_unused()) { | |
7388 | ceph_assert(referenced); | |
7389 | auto p = referenced->find(l.blob); | |
7390 | bluestore_blob_t::unused_t* pu; | |
7391 | if (p == referenced->end()) { | |
7392 | pu = &(*referenced)[l.blob]; | |
7393 | } | |
7394 | else { | |
7395 | pu = &p->second; | |
7396 | } | |
7397 | uint64_t blob_len = blob.get_logical_length(); | |
7398 | ceph_assert((blob_len % (sizeof(*pu) * 8)) == 0); | |
7399 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
7400 | uint64_t chunk_size = blob_len / (sizeof(*pu) * 8); | |
7401 | uint64_t start = l.blob_offset / chunk_size; | |
7402 | uint64_t end = | |
7403 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; | |
7404 | for (auto i = start; i < end; ++i) { | |
7405 | (*pu) |= (1u << i); | |
7406 | } | |
7407 | } | |
7408 | } //for (auto& l : o->extent_map.extent_map) | |
7409 | ||
7410 | for (auto& i : ref_map) { | |
7411 | ++num_blobs; | |
7412 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7413 | bool equal = | |
7414 | depth == FSCK_SHALLOW ? true : | |
7415 | i.first->get_blob_use_tracker().equal(i.second); | |
7416 | if (!equal) { | |
7417 | derr << "fsck error: " << oid << " blob " << *i.first | |
7418 | << " doesn't match expected ref_map " << i.second << dendl; | |
7419 | ++errors; | |
7420 | } | |
7421 | if (blob.is_compressed()) { | |
7422 | res_statfs->data_compressed += blob.get_compressed_payload_length(); | |
7423 | res_statfs->data_compressed_original += | |
7424 | i.first->get_referenced_bytes(); | |
7425 | } | |
7426 | if (blob.is_shared()) { | |
7427 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
7428 | derr << "fsck error: " << oid << " blob " << blob | |
7429 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " | |
7430 | << blobid_max << dendl; | |
7431 | ++errors; | |
7432 | } | |
7433 | else if (i.first->shared_blob->get_sbid() == 0) { | |
7434 | derr << "fsck error: " << oid << " blob " << blob | |
7435 | << " marked as shared but has uninitialized sbid" | |
7436 | << dendl; | |
7437 | ++errors; | |
7438 | } | |
7439 | // the below lock is optional and provided in multithreading mode only | |
7440 | if (sb_info_lock) { | |
7441 | sb_info_lock->lock(); | |
7442 | } | |
7443 | sb_info_t& sbi = sb_info[i.first->shared_blob->get_sbid()]; | |
7444 | ceph_assert(sbi.cid == coll_t() || sbi.cid == c->cid); | |
7445 | ceph_assert(sbi.pool_id == INT64_MIN || | |
7446 | sbi.pool_id == oid.hobj.get_logical_pool()); | |
7447 | sbi.cid = c->cid; | |
7448 | sbi.pool_id = oid.hobj.get_logical_pool(); | |
7449 | sbi.sb = i.first->shared_blob; | |
7450 | sbi.oids.push_back(oid); | |
7451 | sbi.compressed = blob.is_compressed(); | |
7452 | for (auto e : blob.get_extents()) { | |
7453 | if (e.is_valid()) { | |
7454 | sbi.ref_map.get(e.offset, e.length); | |
7455 | } | |
7456 | } | |
7457 | if (sb_info_lock) { | |
7458 | sb_info_lock->unlock(); | |
7459 | } | |
7460 | } else if (depth != FSCK_SHALLOW) { | |
7461 | ceph_assert(used_blocks); | |
7462 | errors += _fsck_check_extents(c->cid, oid, blob.get_extents(), | |
7463 | blob.is_compressed(), | |
7464 | *used_blocks, | |
7465 | fm->get_alloc_size(), | |
7466 | repairer, | |
7467 | *res_statfs, | |
7468 | depth); | |
7469 | } else { | |
7470 | errors += _fsck_sum_extents( | |
7471 | blob.get_extents(), | |
7472 | blob.is_compressed(), | |
7473 | *res_statfs); | |
7474 | } | |
7475 | } // for (auto& i : ref_map) | |
9f95a23c TL |
7476 | |
7477 | if (o->onode.has_omap()) { | |
7478 | _fsck_check_object_omap(depth, o, ctx); | |
7479 | } | |
7480 | ||
eafe8130 TL |
7481 | return o; |
7482 | } | |
7483 | ||
7484 | #include "common/WorkQueue.h" | |
7485 | ||
7486 | class ShallowFSCKThreadPool : public ThreadPool | |
7487 | { | |
7488 | public: | |
7489 | ShallowFSCKThreadPool(CephContext* cct_, std::string nm, std::string tn, int n) : | |
7490 | ThreadPool(cct_, nm, tn, n) { | |
7491 | } | |
7492 | void worker(ThreadPool::WorkThread* wt) override { | |
7493 | int next_wq = 0; | |
7494 | while (!_stop) { | |
7495 | next_wq %= work_queues.size(); | |
7496 | WorkQueue_ *wq = work_queues[next_wq++]; | |
7497 | ||
7498 | void* item = wq->_void_dequeue(); | |
7499 | if (item) { | |
7500 | processing++; | |
7501 | TPHandle tp_handle(cct, nullptr, wq->timeout_interval, wq->suicide_interval); | |
7502 | wq->_void_process(item, tp_handle); | |
7503 | processing--; | |
7504 | } | |
7505 | } | |
7506 | } | |
7507 | template <size_t BatchLen> | |
7508 | struct FSCKWorkQueue : public ThreadPool::WorkQueue_ | |
7509 | { | |
7510 | struct Entry { | |
7511 | int64_t pool_id; | |
7512 | BlueStore::CollectionRef c; | |
7513 | ghobject_t oid; | |
7514 | string key; | |
7515 | bufferlist value; | |
7516 | }; | |
7517 | struct Batch { | |
7518 | std::atomic<size_t> running = { 0 }; | |
7519 | size_t entry_count = 0; | |
7520 | std::array<Entry, BatchLen> entries; | |
7521 | ||
7522 | int64_t errors = 0; | |
7523 | int64_t warnings = 0; | |
7524 | uint64_t num_objects = 0; | |
7525 | uint64_t num_extents = 0; | |
7526 | uint64_t num_blobs = 0; | |
7527 | uint64_t num_sharded_objects = 0; | |
7528 | uint64_t num_spanning_blobs = 0; | |
7529 | store_statfs_t expected_store_statfs; | |
7530 | BlueStore::per_pool_statfs expected_pool_statfs; | |
7531 | }; | |
7532 | ||
7533 | size_t batchCount; | |
7534 | BlueStore* store = nullptr; | |
7535 | ||
eafe8130 TL |
7536 | ceph::mutex* sb_info_lock = nullptr; |
7537 | BlueStore::sb_info_map_t* sb_info = nullptr; | |
7538 | BlueStoreRepairer* repairer = nullptr; | |
7539 | ||
7540 | Batch* batches = nullptr; | |
7541 | size_t last_batch_pos = 0; | |
7542 | bool batch_acquired = false; | |
7543 | ||
7544 | FSCKWorkQueue(std::string n, | |
7545 | size_t _batchCount, | |
7546 | BlueStore* _store, | |
eafe8130 TL |
7547 | ceph::mutex* _sb_info_lock, |
7548 | BlueStore::sb_info_map_t& _sb_info, | |
7549 | BlueStoreRepairer* _repairer) : | |
7550 | WorkQueue_(n, time_t(), time_t()), | |
7551 | batchCount(_batchCount), | |
7552 | store(_store), | |
eafe8130 TL |
7553 | sb_info_lock(_sb_info_lock), |
7554 | sb_info(&_sb_info), | |
7555 | repairer(_repairer) | |
7556 | { | |
7557 | batches = new Batch[batchCount]; | |
7558 | } | |
7559 | ~FSCKWorkQueue() { | |
7560 | delete[] batches; | |
7561 | } | |
7562 | ||
7563 | /// Remove all work items from the queue. | |
7564 | void _clear() override { | |
7565 | //do nothing | |
7566 | } | |
7567 | /// Check whether there is anything to do. | |
7568 | bool _empty() override { | |
7569 | ceph_assert(false); | |
7570 | } | |
7571 | ||
7572 | /// Get the next work item to process. | |
7573 | void* _void_dequeue() override { | |
7574 | size_t pos = rand() % batchCount; | |
7575 | size_t pos0 = pos; | |
7576 | do { | |
7577 | auto& batch = batches[pos]; | |
7578 | if (batch.running.fetch_add(1) == 0) { | |
7579 | if (batch.entry_count) { | |
7580 | return &batch; | |
7581 | } | |
7582 | } | |
7583 | batch.running--; | |
7584 | pos++; | |
7585 | pos %= batchCount; | |
7586 | } while (pos != pos0); | |
7587 | return nullptr; | |
7588 | } | |
7589 | /** @brief Process the work item. | |
7590 | * This function will be called several times in parallel | |
7591 | * and must therefore be thread-safe. */ | |
7592 | void _void_process(void* item, TPHandle& handle) override { | |
7593 | Batch* batch = (Batch*)item; | |
7594 | ||
7595 | BlueStore::FSCK_ObjectCtx ctx( | |
7596 | batch->errors, | |
7597 | batch->warnings, | |
7598 | batch->num_objects, | |
7599 | batch->num_extents, | |
7600 | batch->num_blobs, | |
7601 | batch->num_sharded_objects, | |
7602 | batch->num_spanning_blobs, | |
7603 | nullptr, // used_blocks | |
9f95a23c | 7604 | nullptr, //used_omap_head |
eafe8130 TL |
7605 | sb_info_lock, |
7606 | *sb_info, | |
7607 | batch->expected_store_statfs, | |
7608 | batch->expected_pool_statfs, | |
7609 | repairer); | |
7610 | ||
7611 | for (size_t i = 0; i < batch->entry_count; i++) { | |
7612 | auto& entry = batch->entries[i]; | |
7613 | ||
7614 | store->fsck_check_objects_shallow( | |
7615 | BlueStore::FSCK_SHALLOW, | |
7616 | entry.pool_id, | |
7617 | entry.c, | |
7618 | entry.oid, | |
7619 | entry.key, | |
7620 | entry.value, | |
9f95a23c | 7621 | nullptr, // expecting_shards - this will need a protection if passed |
eafe8130 TL |
7622 | nullptr, // referenced |
7623 | ctx); | |
7624 | } | |
7625 | //std::cout << "processed " << batch << std::endl; | |
7626 | batch->entry_count = 0; | |
7627 | batch->running--; | |
7628 | } | |
7629 | /** @brief Synchronously finish processing a work item. | |
7630 | * This function is called after _void_process with the global thread pool lock held, | |
7631 | * so at most one copy will execute simultaneously for a given thread pool. | |
7632 | * It can be used for non-thread-safe finalization. */ | |
7633 | void _void_process_finish(void*) override { | |
7634 | ceph_assert(false); | |
7635 | } | |
7636 | ||
7637 | bool queue( | |
7638 | int64_t pool_id, | |
7639 | BlueStore::CollectionRef c, | |
7640 | const ghobject_t& oid, | |
7641 | const string& key, | |
7642 | const bufferlist& value) { | |
7643 | bool res = false; | |
7644 | size_t pos0 = last_batch_pos; | |
7645 | if (!batch_acquired) { | |
7646 | do { | |
7647 | auto& batch = batches[last_batch_pos]; | |
7648 | if (batch.running.fetch_add(1) == 0) { | |
7649 | if (batch.entry_count < BatchLen) { | |
7650 | batch_acquired = true; | |
7651 | break; | |
7652 | } | |
7653 | } | |
7654 | batch.running.fetch_sub(1); | |
7655 | last_batch_pos++; | |
7656 | last_batch_pos %= batchCount; | |
7657 | } while (last_batch_pos != pos0); | |
7658 | } | |
7659 | if (batch_acquired) { | |
7660 | auto& batch = batches[last_batch_pos]; | |
7661 | ceph_assert(batch.running); | |
7662 | ceph_assert(batch.entry_count < BatchLen); | |
7663 | ||
7664 | auto& entry = batch.entries[batch.entry_count]; | |
7665 | entry.pool_id = pool_id; | |
7666 | entry.c = c; | |
7667 | entry.oid = oid; | |
7668 | entry.key = key; | |
7669 | entry.value = value; | |
7670 | ||
7671 | ++batch.entry_count; | |
7672 | if (batch.entry_count == BatchLen) { | |
7673 | batch_acquired = false; | |
7674 | batch.running.fetch_sub(1); | |
7675 | last_batch_pos++; | |
7676 | last_batch_pos %= batchCount; | |
7677 | } | |
7678 | res = true; | |
7679 | } | |
7680 | return res; | |
7681 | } | |
7682 | ||
7683 | void finalize(ThreadPool& tp, | |
7684 | BlueStore::FSCK_ObjectCtx& ctx) { | |
7685 | if (batch_acquired) { | |
7686 | auto& batch = batches[last_batch_pos]; | |
7687 | ceph_assert(batch.running); | |
7688 | batch.running.fetch_sub(1); | |
7689 | } | |
7690 | tp.stop(); | |
7691 | ||
7692 | for (size_t i = 0; i < batchCount; i++) { | |
7693 | auto& batch = batches[i]; | |
7694 | ||
7695 | //process leftovers if any | |
7696 | if (batch.entry_count) { | |
7697 | TPHandle tp_handle(store->cct, | |
7698 | nullptr, | |
7699 | timeout_interval, | |
7700 | suicide_interval); | |
7701 | ceph_assert(batch.running == 0); | |
7702 | ||
7703 | batch.running++; // just to be on-par with the regular call | |
7704 | _void_process(&batch, tp_handle); | |
7705 | } | |
7706 | ceph_assert(batch.entry_count == 0); | |
7707 | ||
7708 | ctx.errors += batch.errors; | |
7709 | ctx.warnings += batch.warnings; | |
7710 | ctx.num_objects += batch.num_objects; | |
7711 | ctx.num_extents += batch.num_extents; | |
7712 | ctx.num_blobs += batch.num_blobs; | |
7713 | ctx.num_sharded_objects += batch.num_sharded_objects; | |
7714 | ctx.num_spanning_blobs += batch.num_spanning_blobs; | |
9f95a23c | 7715 | |
eafe8130 TL |
7716 | ctx.expected_store_statfs.add(batch.expected_store_statfs); |
7717 | ||
7718 | for (auto it = batch.expected_pool_statfs.begin(); | |
7719 | it != batch.expected_pool_statfs.end(); | |
7720 | it++) { | |
7721 | ctx.expected_pool_statfs[it->first].add(it->second); | |
7722 | } | |
7723 | } | |
7724 | } | |
7725 | }; | |
7726 | }; | |
7727 | ||
9f95a23c TL |
7728 | void BlueStore::_fsck_check_object_omap(FSCKDepth depth, |
7729 | OnodeRef& o, | |
7730 | const BlueStore::FSCK_ObjectCtx& ctx) | |
eafe8130 | 7731 | { |
9f95a23c TL |
7732 | auto& errors = ctx.errors; |
7733 | auto& warnings = ctx.warnings; | |
7734 | auto repairer = ctx.repairer; | |
7735 | ||
7736 | ceph_assert(o->onode.has_omap()); | |
7737 | if (!o->onode.is_perpool_omap() && !o->onode.is_pgmeta_omap()) { | |
7738 | if (per_pool_omap) { | |
7739 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
7740 | << "fsck error: " << o->oid | |
7741 | << " has omap that is not per-pool or pgmeta" | |
7742 | << fsck_dendl; | |
7743 | ++errors; | |
7744 | } else { | |
7745 | const char* w; | |
7746 | int64_t num; | |
7747 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
7748 | ++errors; | |
7749 | num = errors; | |
7750 | w = "error"; | |
7751 | } else { | |
7752 | ++warnings; | |
7753 | num = warnings; | |
7754 | w = "warning"; | |
7755 | } | |
7756 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
7757 | << "fsck " << w << ": " << o->oid | |
7758 | << " has omap that is not per-pool or pgmeta" | |
7759 | << fsck_dendl; | |
7760 | } | |
7761 | } | |
7762 | if (repairer && | |
7763 | !o->onode.is_perpool_omap() && | |
7764 | !o->onode.is_pgmeta_omap()) { | |
7765 | dout(10) << "fsck converting " << o->oid << " omap to per-pool" << dendl; | |
7766 | bufferlist h; | |
7767 | map<string, bufferlist> kv; | |
7768 | int r = _onode_omap_get(o, &h, &kv); | |
7769 | if (r < 0) { | |
7770 | derr << " got " << r << " " << cpp_strerror(r) << dendl; | |
7771 | } else { | |
7772 | KeyValueDB::Transaction txn = db->get_transaction(); | |
7773 | // remove old keys | |
7774 | const string& old_omap_prefix = o->get_omap_prefix(); | |
7775 | string old_head, old_tail; | |
7776 | o->get_omap_header(&old_head); | |
7777 | o->get_omap_tail(&old_tail); | |
7778 | txn->rm_range_keys(old_omap_prefix, old_head, old_tail); | |
7779 | txn->rmkey(old_omap_prefix, old_tail); | |
7780 | // set flag | |
7781 | o->onode.set_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP); | |
7782 | _record_onode(o, txn); | |
7783 | const string& new_omap_prefix = o->get_omap_prefix(); | |
7784 | // head | |
7785 | if (h.length()) { | |
7786 | string new_head; | |
7787 | o->get_omap_header(&new_head); | |
7788 | txn->set(new_omap_prefix, new_head, h); | |
7789 | } | |
7790 | // tail | |
7791 | string new_tail; | |
7792 | o->get_omap_tail(&new_tail); | |
7793 | bufferlist empty; | |
7794 | txn->set(new_omap_prefix, new_tail, empty); | |
7795 | // values | |
7796 | string final_key; | |
7797 | o->get_omap_key(string(), &final_key); | |
7798 | size_t base_key_len = final_key.size(); | |
7799 | for (auto& i : kv) { | |
7800 | final_key.resize(base_key_len); | |
7801 | final_key += i.first; | |
7802 | txn->set(new_omap_prefix, final_key, i.second); | |
7803 | } | |
7804 | db->submit_transaction_sync(txn); | |
7805 | repairer->inc_repaired(); | |
7806 | } | |
eafe8130 | 7807 | } |
9f95a23c | 7808 | } |
eafe8130 | 7809 | |
9f95a23c TL |
7810 | void BlueStore::_fsck_check_objects(FSCKDepth depth, |
7811 | BlueStore::FSCK_ObjectCtx& ctx) | |
7812 | { | |
eafe8130 | 7813 | auto& errors = ctx.errors; |
eafe8130 TL |
7814 | auto sb_info_lock = ctx.sb_info_lock; |
7815 | auto& sb_info = ctx.sb_info; | |
7816 | auto repairer = ctx.repairer; | |
7817 | ||
7818 | uint64_t_btree_t used_nids; | |
7819 | ||
7820 | size_t processed_myself = 0; | |
7821 | ||
7822 | auto it = db->get_iterator(PREFIX_OBJ); | |
7823 | mempool::bluestore_fsck::list<string> expecting_shards; | |
7824 | if (it) { | |
7825 | const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads; | |
7826 | typedef ShallowFSCKThreadPool::FSCKWorkQueue<256> WQ; | |
7827 | std::unique_ptr<WQ> wq( | |
7828 | new WQ( | |
7829 | "FSCKWorkQueue", | |
7830 | (thread_count ? : 1) * 32, | |
7831 | this, | |
eafe8130 TL |
7832 | sb_info_lock, |
7833 | sb_info, | |
7834 | repairer)); | |
7835 | ||
7836 | ShallowFSCKThreadPool thread_pool(cct, "ShallowFSCKThreadPool", "ShallowFSCK", thread_count); | |
7837 | ||
7838 | thread_pool.add_work_queue(wq.get()); | |
7839 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
7840 | //not the best place but let's check anyway | |
7841 | ceph_assert(sb_info_lock); | |
7842 | thread_pool.start(); | |
7843 | } | |
7844 | ||
7845 | //fill global if not overriden below | |
7846 | CollectionRef c; | |
7847 | int64_t pool_id = -1; | |
7848 | spg_t pgid; | |
7849 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7850 | dout(30) << __func__ << " key " | |
7851 | << pretty_binary_string(it->key()) << dendl; | |
7852 | if (is_extent_shard_key(it->key())) { | |
7853 | if (depth == FSCK_SHALLOW) { | |
7854 | continue; | |
7855 | } | |
7856 | while (!expecting_shards.empty() && | |
7857 | expecting_shards.front() < it->key()) { | |
7858 | derr << "fsck error: missing shard key " | |
7859 | << pretty_binary_string(expecting_shards.front()) | |
7860 | << dendl; | |
7861 | ++errors; | |
7862 | expecting_shards.pop_front(); | |
7863 | } | |
7864 | if (!expecting_shards.empty() && | |
7865 | expecting_shards.front() == it->key()) { | |
7866 | // all good | |
7867 | expecting_shards.pop_front(); | |
7868 | continue; | |
7869 | } | |
7870 | ||
7871 | uint32_t offset; | |
7872 | string okey; | |
7873 | get_key_extent_shard(it->key(), &okey, &offset); | |
7874 | derr << "fsck error: stray shard 0x" << std::hex << offset | |
7875 | << std::dec << dendl; | |
7876 | if (expecting_shards.empty()) { | |
7877 | derr << "fsck error: " << pretty_binary_string(it->key()) | |
7878 | << " is unexpected" << dendl; | |
7879 | ++errors; | |
7880 | continue; | |
7881 | } | |
7882 | while (expecting_shards.front() > it->key()) { | |
7883 | derr << "fsck error: saw " << pretty_binary_string(it->key()) | |
7884 | << dendl; | |
7885 | derr << "fsck error: exp " | |
7886 | << pretty_binary_string(expecting_shards.front()) << dendl; | |
7887 | ++errors; | |
7888 | expecting_shards.pop_front(); | |
7889 | if (expecting_shards.empty()) { | |
7890 | break; | |
7891 | } | |
7892 | } | |
7893 | continue; | |
7894 | } | |
7895 | ||
7896 | ghobject_t oid; | |
7897 | int r = get_key_object(it->key(), &oid); | |
7898 | if (r < 0) { | |
7899 | derr << "fsck error: bad object key " | |
7900 | << pretty_binary_string(it->key()) << dendl; | |
7901 | ++errors; | |
7902 | continue; | |
7903 | } | |
7904 | if (!c || | |
7905 | oid.shard_id != pgid.shard || | |
7906 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
7907 | !c->contains(oid)) { | |
7908 | c = nullptr; | |
7909 | for (auto& p : coll_map) { | |
7910 | if (p.second->contains(oid)) { | |
7911 | c = p.second; | |
7912 | break; | |
7913 | } | |
7914 | } | |
7915 | if (!c) { | |
7916 | derr << "fsck error: stray object " << oid | |
7917 | << " not owned by any collection" << dendl; | |
7918 | ++errors; | |
7919 | continue; | |
7920 | } | |
7921 | pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
7922 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
7923 | << dendl; | |
7924 | } | |
7925 | ||
7926 | if (depth != FSCK_SHALLOW && | |
7927 | !expecting_shards.empty()) { | |
7928 | for (auto& k : expecting_shards) { | |
7929 | derr << "fsck error: missing shard key " | |
7930 | << pretty_binary_string(k) << dendl; | |
7931 | } | |
7932 | ++errors; | |
7933 | expecting_shards.clear(); | |
7934 | } | |
7935 | ||
7936 | bool queued = false; | |
7937 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
7938 | queued = wq->queue( | |
7939 | pool_id, | |
7940 | c, | |
7941 | oid, | |
7942 | it->key(), | |
7943 | it->value()); | |
7944 | } | |
7945 | OnodeRef o; | |
7946 | map<BlobRef, bluestore_blob_t::unused_t> referenced; | |
7947 | ||
7948 | if (!queued) { | |
7949 | ++processed_myself; | |
7950 | ||
7951 | o = fsck_check_objects_shallow( | |
7952 | depth, | |
7953 | pool_id, | |
7954 | c, | |
7955 | oid, | |
7956 | it->key(), | |
7957 | it->value(), | |
9f95a23c | 7958 | &expecting_shards, |
eafe8130 TL |
7959 | &referenced, |
7960 | ctx); | |
7961 | } | |
7962 | ||
7963 | if (depth != FSCK_SHALLOW) { | |
7964 | ceph_assert(o != nullptr); | |
7965 | if (o->onode.nid) { | |
7966 | if (o->onode.nid > nid_max) { | |
7967 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
7968 | << " > nid_max " << nid_max << dendl; | |
7969 | ++errors; | |
7970 | } | |
7971 | if (used_nids.count(o->onode.nid)) { | |
7972 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
7973 | << " already in use" << dendl; | |
7974 | ++errors; | |
7975 | continue; // go for next object | |
7976 | } | |
7977 | used_nids.insert(o->onode.nid); | |
7978 | } | |
7979 | for (auto& i : referenced) { | |
7980 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
7981 | << std::dec << " for " << *i.first << dendl; | |
7982 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7983 | if (i.second & blob.unused) { | |
7984 | derr << "fsck error: " << oid << " blob claims unused 0x" | |
7985 | << std::hex << blob.unused | |
7986 | << " but extents reference 0x" << i.second << std::dec | |
7987 | << " on blob " << *i.first << dendl; | |
7988 | ++errors; | |
7989 | } | |
7990 | if (blob.has_csum()) { | |
7991 | uint64_t blob_len = blob.get_logical_length(); | |
7992 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused) * 8); | |
7993 | unsigned csum_count = blob.get_csum_count(); | |
7994 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
7995 | for (unsigned p = 0; p < csum_count; ++p) { | |
7996 | unsigned pos = p * csum_chunk_size; | |
7997 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
7998 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
7999 | unsigned mask = 1u << firstbit; | |
8000 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
8001 | mask |= 1u << b; | |
8002 | } | |
8003 | if ((blob.unused & mask) == mask) { | |
8004 | // this csum chunk region is marked unused | |
8005 | if (blob.get_csum_item(p) != 0) { | |
8006 | derr << "fsck error: " << oid | |
8007 | << " blob claims csum chunk 0x" << std::hex << pos | |
8008 | << "~" << csum_chunk_size | |
8009 | << " is unused (mask 0x" << mask << " of unused 0x" | |
8010 | << blob.unused << ") but csum is non-zero 0x" | |
8011 | << blob.get_csum_item(p) << std::dec << " on blob " | |
8012 | << *i.first << dendl; | |
8013 | ++errors; | |
8014 | } | |
8015 | } | |
8016 | } | |
8017 | } | |
8018 | } | |
8019 | // omap | |
8020 | if (o->onode.has_omap()) { | |
9f95a23c TL |
8021 | ceph_assert(ctx.used_omap_head); |
8022 | if (ctx.used_omap_head->count(o->onode.nid)) { | |
8023 | derr << "fsck error: " << o->oid << " omap_head " << o->onode.nid | |
8024 | << " already in use" << dendl; | |
eafe8130 TL |
8025 | ++errors; |
8026 | } else { | |
9f95a23c | 8027 | ctx.used_omap_head->insert(o->onode.nid); |
eafe8130 | 8028 | } |
9f95a23c | 8029 | } // if (o->onode.has_omap()) |
eafe8130 TL |
8030 | if (depth == FSCK_DEEP) { |
8031 | bufferlist bl; | |
8032 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; | |
8033 | uint64_t offset = 0; | |
8034 | do { | |
8035 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
8036 | int r = _do_read(c.get(), o, offset, l, bl, | |
8037 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
8038 | if (r < 0) { | |
8039 | ++errors; | |
8040 | derr << "fsck error: " << oid << std::hex | |
8041 | << " error during read: " | |
8042 | << " " << offset << "~" << l | |
8043 | << " " << cpp_strerror(r) << std::dec | |
8044 | << dendl; | |
8045 | break; | |
8046 | } | |
8047 | offset += l; | |
8048 | } while (offset < o->onode.size); | |
8049 | } // deep | |
8050 | } //if (depth != FSCK_SHALLOW) | |
8051 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8052 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8053 | wq->finalize(thread_pool, ctx); | |
8054 | if (processed_myself) { | |
8055 | // may be needs more threads? | |
8056 | dout(0) << __func__ << " partial offload" | |
8057 | << ", done myself " << processed_myself | |
8058 | << " of " << ctx.num_objects | |
8059 | << "objects, threads " << thread_count | |
8060 | << dendl; | |
8061 | } | |
8062 | } | |
8063 | } // if (it) | |
8064 | } | |
8065 | /** | |
8066 | An overview for currently implemented repair logics | |
8067 | performed in fsck in two stages: detection(+preparation) and commit. | |
8068 | Detection stage (in processing order): | |
8069 | (Issue -> Repair action to schedule) | |
8070 | - Detect undecodable keys for Shared Blobs -> Remove | |
8071 | - Detect undecodable records for Shared Blobs -> Remove | |
8072 | (might trigger missed Shared Blob detection below) | |
8073 | - Detect stray records for Shared Blobs -> Remove | |
8074 | - Detect misreferenced pextents -> Fix | |
8075 | Prepare Bloom-like filter to track cid/oid -> pextent | |
8076 | Prepare list of extents that are improperly referenced | |
8077 | Enumerate Onode records that might use 'misreferenced' pextents | |
8078 | (Bloom-like filter applied to reduce computation) | |
8079 | Per each questinable Onode enumerate all blobs and identify broken ones | |
8080 | (i.e. blobs having 'misreferences') | |
8081 | Rewrite each broken blob data by allocating another extents and | |
8082 | copying data there | |
8083 | If blob is shared - unshare it and mark corresponding Shared Blob | |
8084 | for removal | |
8085 | Release previously allocated space | |
8086 | Update Extent Map | |
8087 | - Detect missed Shared Blobs -> Recreate | |
8088 | - Detect undecodable deferred transaction -> Remove | |
8089 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
8090 | - Detect Freelist Manager's leaked entries -> Mark as free | |
8091 | - Detect statfs inconsistency - Update | |
8092 | Commit stage (separate DB commit per each step): | |
8093 | - Apply leaked FM entries fix | |
8094 | - Apply 'false free' FM entries fix | |
8095 | - Apply 'Remove' actions | |
8096 | - Apply fix for misreference pextents | |
8097 | - Apply Shared Blob recreate | |
8098 | (can be merged with the step above if misreferences were dectected) | |
8099 | - Apply StatFS update | |
8100 | */ | |
8101 | int BlueStore::_fsck(BlueStore::FSCKDepth depth, bool repair) | |
8102 | { | |
8103 | dout(1) << __func__ | |
8104 | << (repair ? " repair" : " check") | |
8105 | << (depth == FSCK_DEEP ? " (deep)" : | |
8106 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8107 | << dendl; | |
8108 | ||
8109 | // in deep mode we need R/W write access to be able to replay deferred ops | |
8110 | bool read_only = !(repair || depth == FSCK_DEEP); | |
8111 | ||
8112 | int r = _open_path(); | |
8113 | if (r < 0) | |
8114 | return r; | |
7c673cae FG |
8115 | r = _open_fsid(false); |
8116 | if (r < 0) | |
8117 | goto out_path; | |
8118 | ||
8119 | r = _read_fsid(&fsid); | |
8120 | if (r < 0) | |
8121 | goto out_fsid; | |
8122 | ||
8123 | r = _lock_fsid(); | |
8124 | if (r < 0) | |
8125 | goto out_fsid; | |
8126 | ||
8127 | r = _open_bdev(false); | |
8128 | if (r < 0) | |
8129 | goto out_fsid; | |
8130 | ||
11fdf7f2 | 8131 | r = _open_db_and_around(read_only); |
7c673cae FG |
8132 | if (r < 0) |
8133 | goto out_bdev; | |
8134 | ||
11fdf7f2 TL |
8135 | if (!read_only) { |
8136 | r = _upgrade_super(); | |
8137 | if (r < 0) { | |
8138 | goto out_db; | |
8139 | } | |
8140 | } | |
7c673cae | 8141 | |
eafe8130 | 8142 | r = _open_collections(); |
7c673cae | 8143 | if (r < 0) |
11fdf7f2 | 8144 | goto out_db; |
7c673cae FG |
8145 | |
8146 | mempool_thread.init(); | |
8147 | ||
11fdf7f2 TL |
8148 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
8149 | // enable in repair or deep mode modes only | |
8150 | if (!read_only) { | |
8151 | _kv_start(); | |
8152 | r = _deferred_replay(); | |
8153 | _kv_stop(); | |
8154 | } | |
7c673cae FG |
8155 | if (r < 0) |
8156 | goto out_scan; | |
8157 | ||
eafe8130 TL |
8158 | r = _fsck_on_open(depth, repair); |
8159 | ||
8160 | out_scan: | |
8161 | mempool_thread.shutdown(); | |
8162 | _flush_cache(); | |
8163 | out_db: | |
8164 | _close_db_and_around(); | |
8165 | out_bdev: | |
8166 | _close_bdev(); | |
8167 | out_fsid: | |
8168 | _close_fsid(); | |
8169 | out_path: | |
8170 | _close_path(); | |
8171 | ||
8172 | return r; | |
8173 | } | |
8174 | ||
8175 | int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair) | |
8176 | { | |
8177 | dout(1) << __func__ | |
8178 | << " <<<START>>>" | |
8179 | << (repair ? " repair" : " check") | |
8180 | << (depth == FSCK_DEEP ? " (deep)" : | |
8181 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8182 | << " start" << dendl; | |
8183 | int64_t errors = 0; | |
8184 | int64_t warnings = 0; | |
8185 | unsigned repaired = 0; | |
8186 | ||
8187 | uint64_t_btree_t used_omap_head; | |
eafe8130 TL |
8188 | uint64_t_btree_t used_sbids; |
8189 | ||
8190 | mempool_dynamic_bitset used_blocks; | |
8191 | KeyValueDB::Iterator it; | |
8192 | store_statfs_t expected_store_statfs, actual_statfs; | |
8193 | per_pool_statfs expected_pool_statfs; | |
8194 | ||
8195 | sb_info_map_t sb_info; | |
8196 | ||
8197 | uint64_t num_objects = 0; | |
8198 | uint64_t num_extents = 0; | |
8199 | uint64_t num_blobs = 0; | |
8200 | uint64_t num_spanning_blobs = 0; | |
8201 | uint64_t num_shared_blobs = 0; | |
8202 | uint64_t num_sharded_objects = 0; | |
8203 | BlueStoreRepairer repairer; | |
8204 | ||
8205 | utime_t start = ceph_clock_now(); | |
8206 | ||
8207 | _fsck_collections(&errors); | |
b32b8144 | 8208 | used_blocks.resize(fm->get_alloc_units()); |
9f95a23c | 8209 | apply_for_bitset_range( |
11fdf7f2 | 8210 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), fm->get_alloc_size(), used_blocks, |
7c673cae FG |
8211 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
8212 | bs.set(pos); | |
8213 | } | |
8214 | ); | |
11fdf7f2 TL |
8215 | if (repair) { |
8216 | repairer.get_space_usage_tracker().init( | |
8217 | bdev->get_size(), | |
8218 | min_alloc_size); | |
8219 | } | |
7c673cae FG |
8220 | |
8221 | if (bluefs) { | |
11fdf7f2 TL |
8222 | if( cct->_conf->bluestore_bluefs_db_compatibility) { |
8223 | interval_set<uint64_t> bluefs_extents_db; | |
8224 | bufferlist bl; | |
8225 | db->get(PREFIX_SUPER, "bluefs_extents", &bl); | |
8226 | auto p = bl.cbegin(); | |
8227 | auto prev_errors = errors; | |
8228 | try { | |
8229 | decode(bluefs_extents_db, p); | |
8230 | bluefs_extents_db.union_of(bluefs_extents); | |
8231 | bluefs_extents_db.subtract(bluefs_extents); | |
8232 | if (!bluefs_extents_db.empty()) { | |
8233 | derr << "fsck error: bluefs_extents inconsistency, " | |
8234 | << "downgrade to previous releases might be broken." | |
8235 | << dendl; | |
8236 | ++errors; | |
8237 | } | |
8238 | } | |
8239 | catch (buffer::error& e) { | |
8240 | derr << "fsck error: failed to retrieve bluefs_extents from kv" << dendl; | |
8241 | ++errors; | |
8242 | } | |
8243 | if (errors != prev_errors && repair) { | |
8244 | repairer.fix_bluefs_extents(out_of_sync_fm); | |
8245 | } | |
8246 | } | |
8247 | ||
7c673cae | 8248 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { |
9f95a23c | 8249 | apply_for_bitset_range( |
b32b8144 | 8250 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae FG |
8251 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
8252 | bs.set(pos); | |
8253 | } | |
8254 | ); | |
8255 | } | |
eafe8130 | 8256 | int r = bluefs->fsck(); |
7c673cae | 8257 | if (r < 0) { |
eafe8130 | 8258 | return r; |
7c673cae FG |
8259 | } |
8260 | if (r > 0) | |
8261 | errors += r; | |
8262 | } | |
8263 | ||
eafe8130 TL |
8264 | if (!per_pool_stat_collection) { |
8265 | const char *w; | |
8266 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_stats) { | |
8267 | w = "error"; | |
8268 | ++errors; | |
8269 | } else { | |
8270 | w = "warning"; | |
8271 | ++warnings; | |
8272 | } | |
8273 | derr << "fsck " << w << ": store not yet converted to per-pool stats" | |
8274 | << dendl; | |
8275 | } | |
9f95a23c TL |
8276 | if (!per_pool_omap) { |
8277 | const char *w; | |
8278 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8279 | w = "error"; | |
8280 | ++errors; | |
8281 | } else { | |
8282 | w = "warning"; | |
8283 | ++warnings; | |
8284 | } | |
8285 | derr << "fsck " << w << ": store not yet converted to per-pool omap" | |
8286 | << dendl; | |
8287 | } | |
8288 | ||
11fdf7f2 | 8289 | // get expected statfs; reset unaffected fields to be able to compare |
7c673cae FG |
8290 | // structs |
8291 | statfs(&actual_statfs); | |
11fdf7f2 TL |
8292 | actual_statfs.total = 0; |
8293 | actual_statfs.internally_reserved = 0; | |
8294 | actual_statfs.available = 0; | |
8295 | actual_statfs.internal_metadata = 0; | |
8296 | actual_statfs.omap_allocated = 0; | |
8297 | ||
eafe8130 TL |
8298 | if (g_conf()->bluestore_debug_fsck_abort) { |
8299 | dout(1) << __func__ << " debug abort" << dendl; | |
8300 | goto out_scan; | |
8301 | } | |
7c673cae | 8302 | // walk PREFIX_OBJ |
eafe8130 TL |
8303 | { |
8304 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
8305 | ceph::mutex sb_info_lock = ceph::make_mutex("BlueStore::fsck::sbinfo_lock"); | |
8306 | BlueStore::FSCK_ObjectCtx ctx( | |
8307 | errors, | |
8308 | warnings, | |
8309 | num_objects, | |
8310 | num_extents, | |
8311 | num_blobs, | |
8312 | num_sharded_objects, | |
8313 | num_spanning_blobs, | |
8314 | &used_blocks, | |
8315 | &used_omap_head, | |
9f95a23c TL |
8316 | //no need for the below lock when in non-shallow mode as |
8317 | // there is no multithreading in this case | |
8318 | depth == FSCK_SHALLOW ? &sb_info_lock : nullptr, | |
eafe8130 TL |
8319 | sb_info, |
8320 | expected_store_statfs, | |
8321 | expected_pool_statfs, | |
8322 | repair ? &repairer : nullptr); | |
9f95a23c TL |
8323 | |
8324 | _fsck_check_objects(depth, ctx); | |
eafe8130 | 8325 | } |
11fdf7f2 | 8326 | |
7c673cae FG |
8327 | dout(1) << __func__ << " checking shared_blobs" << dendl; |
8328 | it = db->get_iterator(PREFIX_SHARED_BLOB); | |
8329 | if (it) { | |
eafe8130 TL |
8330 | // FIXME minor: perhaps simplify for shallow mode? |
8331 | // fill global if not overriden below | |
8332 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 | 8333 | |
7c673cae FG |
8334 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8335 | string key = it->key(); | |
8336 | uint64_t sbid; | |
8337 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 8338 | derr << "fsck error: bad key '" << key |
7c673cae | 8339 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
8340 | if (repair) { |
8341 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8342 | } | |
7c673cae FG |
8343 | ++errors; |
8344 | continue; | |
8345 | } | |
8346 | auto p = sb_info.find(sbid); | |
8347 | if (p == sb_info.end()) { | |
3efd9988 | 8348 | derr << "fsck error: found stray shared blob data for sbid 0x" |
7c673cae | 8349 | << std::hex << sbid << std::dec << dendl; |
11fdf7f2 TL |
8350 | if (repair) { |
8351 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8352 | } | |
7c673cae FG |
8353 | ++errors; |
8354 | } else { | |
8355 | ++num_shared_blobs; | |
8356 | sb_info_t& sbi = p->second; | |
8357 | bluestore_shared_blob_t shared_blob(sbid); | |
8358 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
8359 | auto blp = bl.cbegin(); |
8360 | try { | |
8361 | decode(shared_blob, blp); | |
8362 | } catch (buffer::error& e) { | |
8363 | ++errors; | |
8364 | // Force update and don't report as missing | |
8365 | sbi.updated = sbi.passed = true; | |
8366 | ||
8367 | derr << "fsck error: failed to decode Shared Blob" | |
8368 | << pretty_binary_string(it->key()) << dendl; | |
8369 | if (repair) { | |
8370 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
8371 | << pretty_binary_string(it->key()) | |
8372 | << "', removing" << dendl; | |
8373 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8374 | } | |
8375 | continue; | |
8376 | } | |
7c673cae FG |
8377 | dout(20) << __func__ << " " << *sbi.sb << " " << shared_blob << dendl; |
8378 | if (shared_blob.ref_map != sbi.ref_map) { | |
3efd9988 | 8379 | derr << "fsck error: shared blob 0x" << std::hex << sbid |
11fdf7f2 TL |
8380 | << std::dec << " ref_map " << shared_blob.ref_map |
8381 | << " != expected " << sbi.ref_map << dendl; | |
8382 | sbi.updated = true; // will update later in repair mode only! | |
7c673cae FG |
8383 | ++errors; |
8384 | } | |
8385 | PExtentVector extents; | |
8386 | for (auto &r : shared_blob.ref_map.ref_map) { | |
8387 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
8388 | } | |
eafe8130 | 8389 | if (per_pool_stat_collection || repair) { |
11fdf7f2 TL |
8390 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; |
8391 | } | |
8392 | errors += _fsck_check_extents(sbi.cid, | |
8393 | p->second.oids.front(), | |
7c673cae FG |
8394 | extents, |
8395 | p->second.compressed, | |
b32b8144 FG |
8396 | used_blocks, |
8397 | fm->get_alloc_size(), | |
11fdf7f2 | 8398 | repair ? &repairer : nullptr, |
eafe8130 TL |
8399 | *expected_statfs, |
8400 | depth); | |
11fdf7f2 TL |
8401 | sbi.passed = true; |
8402 | } | |
8403 | } | |
8404 | } // if (it) | |
8405 | ||
8406 | if (repair && repairer.preprocess_misreference(db)) { | |
8407 | ||
8408 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
8409 | auto& space_tracker = repairer.get_space_usage_tracker(); | |
8410 | auto& misref_extents = repairer.get_misreferences(); | |
8411 | interval_set<uint64_t> to_release; | |
8412 | it = db->get_iterator(PREFIX_OBJ); | |
8413 | if (it) { | |
eafe8130 TL |
8414 | // fill global if not overriden below |
8415 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 TL |
8416 | |
8417 | CollectionRef c; | |
8418 | spg_t pgid; | |
8419 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
8420 | bool bypass_rest = false; | |
8421 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
8422 | it->next()) { | |
8423 | dout(30) << __func__ << " key " | |
8424 | << pretty_binary_string(it->key()) << dendl; | |
8425 | if (is_extent_shard_key(it->key())) { | |
8426 | continue; | |
8427 | } | |
8428 | ||
8429 | ghobject_t oid; | |
8430 | int r = get_key_object(it->key(), &oid); | |
8431 | if (r < 0 || !space_tracker.is_used(oid)) { | |
8432 | continue; | |
8433 | } | |
8434 | ||
8435 | if (!c || | |
8436 | oid.shard_id != pgid.shard || | |
8437 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8438 | !c->contains(oid)) { | |
8439 | c = nullptr; | |
8440 | for (auto& p : coll_map) { | |
8441 | if (p.second->contains(oid)) { | |
8442 | c = p.second; | |
8443 | break; | |
8444 | } | |
8445 | } | |
8446 | if (!c) { | |
8447 | continue; | |
8448 | } | |
eafe8130 TL |
8449 | if (per_pool_stat_collection || repair) { |
8450 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
11fdf7f2 TL |
8451 | expected_statfs = &expected_pool_statfs[pool_id]; |
8452 | } | |
8453 | } | |
8454 | if (!space_tracker.is_used(c->cid)) { | |
8455 | continue; | |
8456 | } | |
8457 | ||
8458 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
8459 | << " obj:" << oid << dendl; | |
8460 | ||
eafe8130 TL |
8461 | OnodeRef o; |
8462 | o.reset(Onode::decode(c, oid, it->key(), it->value())); | |
11fdf7f2 TL |
8463 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8464 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
8465 | ||
8466 | for (auto& e : o->extent_map.extent_map) { | |
8467 | blobs.insert(e.blob); | |
8468 | } | |
8469 | bool need_onode_update = false; | |
8470 | bool first_dump = true; | |
8471 | for(auto b : blobs) { | |
8472 | bool broken_blob = false; | |
8473 | auto& pextents = b->dirty_blob().dirty_extents(); | |
8474 | for (auto& e : pextents) { | |
8475 | if (!e.is_valid()) { | |
8476 | continue; | |
8477 | } | |
8478 | // for the sake of simplicity and proper shared blob handling | |
8479 | // always rewrite the whole blob even when it's partially | |
8480 | // misreferenced. | |
8481 | if (misref_extents.intersects(e.offset, e.length)) { | |
8482 | if (first_dump) { | |
8483 | first_dump = false; | |
81eedcae | 8484 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
8485 | } |
8486 | broken_blob = true; | |
8487 | break; | |
8488 | } | |
8489 | } | |
8490 | if (!broken_blob) | |
8491 | continue; | |
8492 | bool compressed = b->get_blob().is_compressed(); | |
8493 | need_onode_update = true; | |
8494 | dout(10) << __func__ | |
8495 | << " fix misreferences in oid:" << oid | |
8496 | << " " << *b << dendl; | |
8497 | uint64_t b_off = 0; | |
8498 | PExtentVector pext_to_release; | |
8499 | pext_to_release.reserve(pextents.size()); | |
8500 | // rewriting all valid pextents | |
8501 | for (auto e = pextents.begin(); e != pextents.end(); | |
8502 | b_off += e->length, e++) { | |
8503 | if (!e->is_valid()) { | |
8504 | continue; | |
8505 | } | |
8506 | PExtentVector exts; | |
8507 | int64_t alloc_len = alloc->allocate(e->length, min_alloc_size, | |
8508 | 0, 0, &exts); | |
eafe8130 | 8509 | if (alloc_len < 0 || alloc_len < (int64_t)e->length) { |
11fdf7f2 TL |
8510 | derr << __func__ |
8511 | << " failed to allocate 0x" << std::hex << e->length | |
eafe8130 | 8512 | << " allocated 0x " << (alloc_len < 0 ? 0 : alloc_len) |
11fdf7f2 TL |
8513 | << " min_alloc_size 0x" << min_alloc_size |
8514 | << " available 0x " << alloc->get_free() | |
8515 | << std::dec << dendl; | |
8516 | if (alloc_len > 0) { | |
8517 | alloc->release(exts); | |
8518 | } | |
8519 | bypass_rest = true; | |
8520 | break; | |
8521 | } | |
8522 | expected_statfs->allocated += e->length; | |
8523 | if (compressed) { | |
8524 | expected_statfs->data_compressed_allocated += e->length; | |
8525 | } | |
8526 | ||
8527 | bufferlist bl; | |
8528 | IOContext ioc(cct, NULL, true); // allow EIO | |
8529 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); | |
8530 | if (r < 0) { | |
8531 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
8532 | <<"~" << e->length << std::dec << dendl; | |
8533 | ceph_abort_msg("read failed, wtf"); | |
8534 | } | |
8535 | pext_to_release.push_back(*e); | |
8536 | e = pextents.erase(e); | |
8537 | e = pextents.insert(e, exts.begin(), exts.end()); | |
8538 | b->get_blob().map_bl( | |
8539 | b_off, bl, | |
8540 | [&](uint64_t offset, bufferlist& t) { | |
8541 | int r = bdev->write(offset, t, false); | |
8542 | ceph_assert(r == 0); | |
8543 | }); | |
8544 | e += exts.size() - 1; | |
8545 | for (auto& p : exts) { | |
8546 | fm->allocate(p.offset, p.length, txn); | |
8547 | } | |
8548 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
8549 | ||
8550 | if (b->get_blob().is_shared()) { | |
8551 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
8552 | ||
8553 | auto sb_it = sb_info.find(b->shared_blob->get_sbid()); | |
8554 | ceph_assert(sb_it != sb_info.end()); | |
8555 | sb_info_t& sbi = sb_it->second; | |
8556 | ||
8557 | for (auto& r : sbi.ref_map.ref_map) { | |
8558 | expected_statfs->allocated -= r.second.length; | |
8559 | if (sbi.compressed) { | |
8560 | // NB: it's crucial to use compressed flag from sb_info_t | |
8561 | // as we originally used that value while accumulating | |
8562 | // expected_statfs | |
8563 | expected_statfs->data_compressed_allocated -= r.second.length; | |
8564 | } | |
8565 | } | |
8566 | sbi.updated = sbi.passed = true; | |
8567 | sbi.ref_map.clear(); | |
8568 | ||
8569 | // relying on blob's pextents to decide what to release. | |
8570 | for (auto& p : pext_to_release) { | |
8571 | to_release.union_insert(p.offset, p.length); | |
8572 | } | |
8573 | } else { | |
8574 | for (auto& p : pext_to_release) { | |
8575 | expected_statfs->allocated -= p.length; | |
8576 | if (compressed) { | |
8577 | expected_statfs->data_compressed_allocated -= p.length; | |
8578 | } | |
8579 | to_release.union_insert(p.offset, p.length); | |
8580 | } | |
8581 | } | |
8582 | if (bypass_rest) { | |
8583 | break; | |
8584 | } | |
8585 | } // for(auto b : blobs) | |
8586 | if (need_onode_update) { | |
8587 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
8588 | _record_onode(o, txn); | |
8589 | } | |
8590 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8591 | ||
8592 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
8593 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
8594 | << "~" << it.get_len() << std::dec << dendl; | |
8595 | fm->release(it.get_start(), it.get_len(), txn); | |
8596 | } | |
8597 | alloc->release(to_release); | |
8598 | to_release.clear(); | |
8599 | } // if (it) { | |
8600 | } //if (repair && repairer.preprocess_misreference()) { | |
8601 | ||
eafe8130 TL |
8602 | if (depth != FSCK_SHALLOW) { |
8603 | for (auto &p : sb_info) { | |
8604 | sb_info_t& sbi = p.second; | |
8605 | if (!sbi.passed) { | |
8606 | derr << "fsck error: missing " << *sbi.sb << dendl; | |
8607 | ++errors; | |
8608 | } | |
8609 | if (repair && (!sbi.passed || sbi.updated)) { | |
8610 | auto sbid = p.first; | |
8611 | if (sbi.ref_map.empty()) { | |
8612 | ceph_assert(sbi.passed); | |
8613 | dout(20) << __func__ << " " << *sbi.sb | |
8614 | << " is empty, removing" << dendl; | |
8615 | repairer.fix_shared_blob(db, sbid, nullptr); | |
8616 | } else { | |
8617 | bufferlist bl; | |
8618 | bluestore_shared_blob_t persistent(sbid, std::move(sbi.ref_map)); | |
8619 | encode(persistent, bl); | |
8620 | dout(20) << __func__ << " " << *sbi.sb | |
8621 | << " is " << bl.length() << " bytes, updating" << dendl; | |
11fdf7f2 | 8622 | |
eafe8130 TL |
8623 | repairer.fix_shared_blob(db, sbid, &bl); |
8624 | } | |
7c673cae FG |
8625 | } |
8626 | } | |
8627 | } | |
11fdf7f2 TL |
8628 | sb_info.clear(); |
8629 | ||
eafe8130 TL |
8630 | // check global stats only if fscking (not repairing) w/o per-pool stats |
8631 | if (!per_pool_stat_collection && | |
8632 | !repair && | |
8633 | !(actual_statfs == expected_store_statfs)) { | |
8634 | derr << "fsck error: actual " << actual_statfs | |
8635 | << " != expected " << expected_store_statfs << dendl; | |
8636 | if (repair) { | |
8637 | repairer.fix_statfs(db, BLUESTORE_GLOBAL_STATFS_KEY, | |
8638 | expected_store_statfs); | |
11fdf7f2 | 8639 | } |
eafe8130 | 8640 | ++errors; |
7c673cae FG |
8641 | } |
8642 | ||
eafe8130 TL |
8643 | dout(1) << __func__ << " checking pool_statfs" << dendl; |
8644 | _fsck_check_pool_statfs(expected_pool_statfs, | |
8645 | errors, warnings, repair ? &repairer : nullptr); | |
8646 | ||
8647 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 8648 | dout(1) << __func__ << " checking for stray omap data " << dendl; |
eafe8130 TL |
8649 | it = db->get_iterator(PREFIX_OMAP); |
8650 | if (it) { | |
9f95a23c | 8651 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8652 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8653 | uint64_t omap_head; | |
8654 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8655 | if (used_omap_head.count(omap_head) == 0 && |
8656 | omap_head != last_omap_head) { | |
8657 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8658 | << "fsck error: found stray omap data on omap_head " | |
8659 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head)<< fsck_dendl; | |
eafe8130 | 8660 | ++errors; |
9f95a23c | 8661 | last_omap_head = omap_head; |
eafe8130 | 8662 | } |
7c673cae FG |
8663 | } |
8664 | } | |
eafe8130 TL |
8665 | it = db->get_iterator(PREFIX_PGMETA_OMAP); |
8666 | if (it) { | |
9f95a23c | 8667 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8668 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8669 | uint64_t omap_head; | |
8670 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8671 | if (used_omap_head.count(omap_head) == 0 && |
8672 | omap_head != last_omap_head) { | |
8673 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8674 | << "fsck error: found stray (pgmeta) omap data on omap_head " | |
8675 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8676 | last_omap_head = omap_head; | |
eafe8130 TL |
8677 | ++errors; |
8678 | } | |
11fdf7f2 TL |
8679 | } |
8680 | } | |
9f95a23c TL |
8681 | it = db->get_iterator(PREFIX_PERPOOL_OMAP); |
8682 | if (it) { | |
8683 | uint64_t last_omap_head = 0; | |
8684 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8685 | uint64_t pool; | |
8686 | uint64_t omap_head; | |
8687 | string k = it->key(); | |
8688 | const char *c = k.c_str(); | |
8689 | c = _key_decode_u64(c, &pool); | |
8690 | c = _key_decode_u64(c, &omap_head); | |
8691 | if (used_omap_head.count(omap_head) == 0 && | |
8692 | omap_head != last_omap_head) { | |
8693 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8694 | << "fsck error: found stray (per-pool) omap data on omap_head " | |
8695 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8696 | ++errors; | |
8697 | last_omap_head = omap_head; | |
8698 | } | |
8699 | } | |
8700 | } | |
eafe8130 TL |
8701 | dout(1) << __func__ << " checking deferred events" << dendl; |
8702 | it = db->get_iterator(PREFIX_DEFERRED); | |
8703 | if (it) { | |
8704 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8705 | bufferlist bl = it->value(); | |
8706 | auto p = bl.cbegin(); | |
8707 | bluestore_deferred_transaction_t wt; | |
8708 | try { | |
8709 | decode(wt, p); | |
8710 | } catch (buffer::error& e) { | |
8711 | derr << "fsck error: failed to decode deferred txn " | |
8712 | << pretty_binary_string(it->key()) << dendl; | |
8713 | if (repair) { | |
8714 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
8715 | << pretty_binary_string(it->key()) | |
8716 | << "', removing" << dendl; | |
8717 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8718 | } | |
8719 | continue; | |
8720 | } | |
8721 | dout(20) << __func__ << " deferred " << wt.seq | |
8722 | << " ops " << wt.ops.size() | |
8723 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
8724 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
9f95a23c | 8725 | apply_for_bitset_range( |
eafe8130 TL |
8726 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
8727 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
8728 | bs.set(pos); |
8729 | } | |
8730 | ); | |
8731 | } | |
7c673cae | 8732 | } |
eafe8130 TL |
8733 | } |
8734 | ||
8735 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
8736 | { | |
8737 | // remove bluefs_extents from used set since the freelist doesn't | |
8738 | // know they are allocated. | |
8739 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
9f95a23c | 8740 | apply_for_bitset_range( |
b32b8144 | 8741 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae | 8742 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
eafe8130 | 8743 | bs.reset(pos); |
7c673cae FG |
8744 | } |
8745 | ); | |
8746 | } | |
eafe8130 TL |
8747 | fm->enumerate_reset(); |
8748 | uint64_t offset, length; | |
8749 | while (fm->enumerate_next(db, &offset, &length)) { | |
8750 | bool intersects = false; | |
9f95a23c | 8751 | apply_for_bitset_range( |
eafe8130 TL |
8752 | offset, length, fm->get_alloc_size(), used_blocks, |
8753 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
8754 | if (bs.test(pos)) { |
8755 | if (offset == SUPER_RESERVED && | |
8756 | length == min_alloc_size - SUPER_RESERVED) { | |
8757 | // this is due to the change just after luminous to min_alloc_size | |
8758 | // granularity allocations, and our baked in assumption at the top | |
8759 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
8760 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
8761 | // since we will never allocate this region below min_alloc_size. | |
8762 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
8763 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
8764 | << length << std::dec << dendl; | |
8765 | } else { | |
8766 | intersects = true; | |
8767 | if (repair) { | |
8768 | repairer.fix_false_free(db, fm, | |
8769 | pos * min_alloc_size, | |
8770 | min_alloc_size); | |
8771 | } | |
11fdf7f2 | 8772 | } |
eafe8130 TL |
8773 | } else { |
8774 | bs.set(pos); | |
8775 | } | |
7c673cae | 8776 | } |
eafe8130 TL |
8777 | ); |
8778 | if (intersects) { | |
8779 | derr << "fsck error: free extent 0x" << std::hex << offset | |
8780 | << "~" << length << std::dec | |
8781 | << " intersects allocated blocks" << dendl; | |
8782 | ++errors; | |
7c673cae | 8783 | } |
b5b8bbf5 | 8784 | } |
eafe8130 TL |
8785 | fm->enumerate_reset(); |
8786 | size_t count = used_blocks.count(); | |
8787 | if (used_blocks.size() != count) { | |
8788 | ceph_assert(used_blocks.size() > count); | |
8789 | used_blocks.flip(); | |
8790 | size_t start = used_blocks.find_first(); | |
8791 | while (start != decltype(used_blocks)::npos) { | |
8792 | size_t cur = start; | |
8793 | while (true) { | |
8794 | size_t next = used_blocks.find_next(cur); | |
8795 | if (next != cur + 1) { | |
8796 | ++errors; | |
8797 | derr << "fsck error: leaked extent 0x" << std::hex | |
8798 | << ((uint64_t)start * fm->get_alloc_size()) << "~" | |
8799 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
8800 | << dendl; | |
8801 | if (repair) { | |
8802 | repairer.fix_leaked(db, | |
8803 | fm, | |
8804 | start * min_alloc_size, | |
8805 | (cur + 1 - start) * min_alloc_size); | |
8806 | } | |
8807 | start = next; | |
8808 | break; | |
11fdf7f2 | 8809 | } |
eafe8130 | 8810 | cur = next; |
b5b8bbf5 | 8811 | } |
eafe8130 TL |
8812 | } |
8813 | used_blocks.flip(); | |
b5b8bbf5 | 8814 | } |
7c673cae FG |
8815 | } |
8816 | } | |
11fdf7f2 | 8817 | if (repair) { |
9f95a23c TL |
8818 | if (!per_pool_omap) { |
8819 | dout(5) << __func__ << " marking per_pool_omap=1" << dendl; | |
8820 | repairer.fix_per_pool_omap(db); | |
8821 | } | |
8822 | ||
11fdf7f2 TL |
8823 | dout(5) << __func__ << " applying repair results" << dendl; |
8824 | repaired = repairer.apply(db); | |
8825 | dout(5) << __func__ << " repair applied" << dendl; | |
8826 | } | |
7c673cae | 8827 | |
eafe8130 | 8828 | out_scan: |
7c673cae FG |
8829 | dout(2) << __func__ << " " << num_objects << " objects, " |
8830 | << num_sharded_objects << " of them sharded. " | |
8831 | << dendl; | |
8832 | dout(2) << __func__ << " " << num_extents << " extents to " | |
8833 | << num_blobs << " blobs, " | |
8834 | << num_spanning_blobs << " spanning, " | |
8835 | << num_shared_blobs << " shared." | |
8836 | << dendl; | |
8837 | ||
8838 | utime_t duration = ceph_clock_now() - start; | |
9f95a23c TL |
8839 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " |
8840 | << warnings << " warnings, " | |
8841 | << repaired << " repaired, " | |
8842 | << (errors + warnings - (int)repaired) << " remaining in " | |
7c673cae | 8843 | << duration << " seconds" << dendl; |
9f95a23c TL |
8844 | |
8845 | // In non-repair mode we should return error count only as | |
8846 | // it indicates if store status is OK. | |
8847 | // In repair mode both errors and warnings are taken into account | |
8848 | // since repaired counter relates to them both. | |
8849 | return repair ? errors + warnings - (int)repaired : errors; | |
11fdf7f2 TL |
8850 | } |
8851 | ||
8852 | /// methods to inject various errors fsck can repair | |
8853 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
8854 | const bufferlist& bl) | |
8855 | { | |
8856 | KeyValueDB::Transaction txn; | |
8857 | txn = db->get_transaction(); | |
8858 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
8859 | db->submit_transaction_sync(txn); | |
8860 | }; | |
8861 | ||
8862 | void BlueStore::inject_leaked(uint64_t len) | |
8863 | { | |
8864 | KeyValueDB::Transaction txn; | |
8865 | txn = db->get_transaction(); | |
8866 | ||
8867 | PExtentVector exts; | |
8868 | int64_t alloc_len = alloc->allocate(len, min_alloc_size, | |
8869 | min_alloc_size * 256, 0, &exts); | |
8870 | ceph_assert(alloc_len >= (int64_t)len); | |
8871 | for (auto& p : exts) { | |
8872 | fm->allocate(p.offset, p.length, txn); | |
8873 | } | |
8874 | db->submit_transaction_sync(txn); | |
8875 | } | |
8876 | ||
8877 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
8878 | { | |
8879 | KeyValueDB::Transaction txn; | |
8880 | OnodeRef o; | |
8881 | CollectionRef c = _get_collection(cid); | |
8882 | ceph_assert(c); | |
8883 | { | |
9f95a23c | 8884 | std::unique_lock l{c->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
8885 | o = c->get_onode(oid, false); |
8886 | ceph_assert(o); | |
8887 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
8888 | } | |
8889 | ||
8890 | bool injected = false; | |
8891 | txn = db->get_transaction(); | |
8892 | auto& em = o->extent_map.extent_map; | |
8893 | std::vector<const PExtentVector*> v; | |
8894 | if (em.size()) { | |
8895 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
8896 | } | |
8897 | if (em.size() > 1) { | |
8898 | auto it = em.end(); | |
8899 | --it; | |
8900 | v.push_back(&(it->blob->get_blob().get_extents())); | |
8901 | } | |
8902 | for (auto pext : v) { | |
8903 | if (pext->size()) { | |
8904 | auto p = pext->begin(); | |
8905 | while (p != pext->end()) { | |
8906 | if (p->is_valid()) { | |
8907 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
8908 | << "~" << p->length << std::dec << dendl; | |
8909 | fm->release(p->offset, p->length, txn); | |
8910 | injected = true; | |
8911 | break; | |
8912 | } | |
8913 | ++p; | |
8914 | } | |
8915 | } | |
8916 | } | |
8917 | ceph_assert(injected); | |
8918 | db->submit_transaction_sync(txn); | |
8919 | } | |
8920 | ||
9f95a23c TL |
8921 | void BlueStore::inject_legacy_omap() |
8922 | { | |
8923 | dout(1) << __func__ << dendl; | |
8924 | per_pool_omap = false; | |
8925 | KeyValueDB::Transaction txn; | |
8926 | txn = db->get_transaction(); | |
8927 | txn->rmkey(PREFIX_SUPER, "per_pool_omap"); | |
8928 | db->submit_transaction_sync(txn); | |
8929 | } | |
8930 | ||
8931 | void BlueStore::inject_legacy_omap(coll_t cid, ghobject_t oid) | |
8932 | { | |
8933 | dout(1) << __func__ << " " | |
8934 | << cid << " " << oid | |
8935 | <<dendl; | |
8936 | KeyValueDB::Transaction txn; | |
8937 | OnodeRef o; | |
8938 | CollectionRef c = _get_collection(cid); | |
8939 | ceph_assert(c); | |
8940 | { | |
8941 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
8942 | o = c->get_onode(oid, false); | |
8943 | ceph_assert(o); | |
8944 | } | |
8945 | o->onode.clear_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP | bluestore_onode_t::FLAG_PGMETA_OMAP); | |
8946 | txn = db->get_transaction(); | |
8947 | _record_onode(o, txn); | |
8948 | db->submit_transaction_sync(txn); | |
8949 | } | |
8950 | ||
8951 | ||
11fdf7f2 TL |
8952 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) |
8953 | { | |
8954 | BlueStoreRepairer repairer; | |
8955 | repairer.fix_statfs(db, key, new_statfs); | |
8956 | repairer.apply(db); | |
8957 | } | |
8958 | ||
eafe8130 TL |
8959 | void BlueStore::inject_global_statfs(const store_statfs_t& new_statfs) |
8960 | { | |
8961 | KeyValueDB::Transaction t = db->get_transaction(); | |
8962 | volatile_statfs v; | |
8963 | v = new_statfs; | |
8964 | bufferlist bl; | |
8965 | v.encode(bl); | |
8966 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
8967 | db->submit_transaction_sync(t); | |
8968 | } | |
8969 | ||
11fdf7f2 TL |
8970 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, |
8971 | coll_t cid2, ghobject_t oid2, | |
8972 | uint64_t offset) | |
8973 | { | |
8974 | OnodeRef o1; | |
8975 | CollectionRef c1 = _get_collection(cid1); | |
8976 | ceph_assert(c1); | |
8977 | { | |
9f95a23c | 8978 | std::unique_lock l{c1->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
8979 | o1 = c1->get_onode(oid1, false); |
8980 | ceph_assert(o1); | |
8981 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
8982 | } | |
8983 | OnodeRef o2; | |
8984 | CollectionRef c2 = _get_collection(cid2); | |
8985 | ceph_assert(c2); | |
8986 | { | |
9f95a23c | 8987 | std::unique_lock l{c2->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
8988 | o2 = c2->get_onode(oid2, false); |
8989 | ceph_assert(o2); | |
8990 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
8991 | } | |
8992 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
8993 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
8994 | ||
8995 | // require onode/extent layout to be the same (and simple) | |
8996 | // to make things easier | |
8997 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
8998 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
8999 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
9000 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
9001 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
9002 | ceph_assert(e1.length == e2.length); | |
9003 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
9004 | ||
9005 | KeyValueDB::Transaction txn; | |
9006 | txn = db->get_transaction(); | |
9007 | ||
9008 | // along with misreference error this will create space leaks errors | |
9009 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
9010 | o2->extent_map.dirty_range(offset, e2.length); | |
9011 | o2->extent_map.update(txn, false); | |
9012 | ||
9013 | _record_onode(o2, txn); | |
9014 | db->submit_transaction_sync(txn); | |
7c673cae FG |
9015 | } |
9016 | ||
9017 | void BlueStore::collect_metadata(map<string,string> *pm) | |
9018 | { | |
9019 | dout(10) << __func__ << dendl; | |
9020 | bdev->collect_metadata("bluestore_bdev_", pm); | |
9021 | if (bluefs) { | |
9022 | (*pm)["bluefs"] = "1"; | |
9f95a23c TL |
9023 | // this value is for backward compatibility only |
9024 | (*pm)["bluefs_single_shared_device"] = \ | |
9025 | stringify((int)bluefs_layout.single_shared_device()); | |
9026 | (*pm)["bluefs_dedicated_db"] = \ | |
9027 | stringify((int)bluefs_layout.dedicated_db); | |
9028 | (*pm)["bluefs_dedicated_wal"] = \ | |
9029 | stringify((int)bluefs_layout.dedicated_wal); | |
9030 | bluefs->collect_metadata(pm, bluefs_layout.shared_bdev); | |
7c673cae FG |
9031 | } else { |
9032 | (*pm)["bluefs"] = "0"; | |
9033 | } | |
11fdf7f2 TL |
9034 | |
9035 | // report numa mapping for underlying devices | |
9036 | int node = -1; | |
9037 | set<int> nodes; | |
9038 | set<string> failed; | |
9039 | int r = get_numa_node(&node, &nodes, &failed); | |
9040 | if (r >= 0) { | |
9041 | if (!failed.empty()) { | |
9042 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
9043 | } | |
9044 | if (!nodes.empty()) { | |
9045 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
9046 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
9047 | } | |
9048 | if (node >= 0) { | |
9049 | (*pm)["objectstore_numa_node"] = stringify(node); | |
9050 | } | |
9051 | } | |
9052 | } | |
9053 | ||
9054 | int BlueStore::get_numa_node( | |
9055 | int *final_node, | |
9056 | set<int> *out_nodes, | |
9057 | set<string> *out_failed) | |
9058 | { | |
9059 | int node = -1; | |
9060 | set<string> devices; | |
9061 | get_devices(&devices); | |
9062 | set<int> nodes; | |
9063 | set<string> failed; | |
9064 | for (auto& devname : devices) { | |
9065 | int n; | |
9066 | BlkDev bdev(devname); | |
9067 | int r = bdev.get_numa_node(&n); | |
9068 | if (r < 0) { | |
9069 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
9070 | << dendl; | |
9071 | failed.insert(devname); | |
9072 | continue; | |
9073 | } | |
9074 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
9075 | << dendl; | |
9076 | nodes.insert(n); | |
9077 | if (node < 0) { | |
9078 | node = n; | |
9079 | } | |
9080 | } | |
9081 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
9082 | *final_node = node; | |
9083 | } | |
9084 | if (out_nodes) { | |
9085 | *out_nodes = nodes; | |
9086 | } | |
9087 | if (out_failed) { | |
9088 | *out_failed = failed; | |
9089 | } | |
9090 | return 0; | |
9091 | } | |
9092 | ||
9093 | int BlueStore::get_devices(set<string> *ls) | |
9094 | { | |
9095 | if (bdev) { | |
9096 | bdev->get_devices(ls); | |
9097 | if (bluefs) { | |
9098 | bluefs->get_devices(ls); | |
9099 | } | |
9100 | return 0; | |
9101 | } | |
9102 | ||
9103 | // grumble, we haven't started up yet. | |
9104 | int r = _open_path(); | |
9105 | if (r < 0) | |
9106 | goto out; | |
9107 | r = _open_fsid(false); | |
9108 | if (r < 0) | |
9109 | goto out_path; | |
9110 | r = _read_fsid(&fsid); | |
9111 | if (r < 0) | |
9112 | goto out_fsid; | |
9113 | r = _lock_fsid(); | |
9114 | if (r < 0) | |
9115 | goto out_fsid; | |
9116 | r = _open_bdev(false); | |
9117 | if (r < 0) | |
9118 | goto out_fsid; | |
9119 | r = _minimal_open_bluefs(false); | |
9120 | if (r < 0) | |
9121 | goto out_bdev; | |
9122 | bdev->get_devices(ls); | |
9123 | if (bluefs) { | |
9124 | bluefs->get_devices(ls); | |
9125 | } | |
9126 | r = 0; | |
9127 | _minimal_close_bluefs(); | |
9128 | out_bdev: | |
9129 | _close_bdev(); | |
9130 | out_fsid: | |
9131 | _close_fsid(); | |
9132 | out_path: | |
9133 | _close_path(); | |
9134 | out: | |
9135 | return r; | |
7c673cae FG |
9136 | } |
9137 | ||
11fdf7f2 | 9138 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
9139 | { |
9140 | buf->reset(); | |
11fdf7f2 | 9141 | |
9f95a23c TL |
9142 | buf->omap_allocated = |
9143 | db->estimate_prefix_size(PREFIX_OMAP, string()) + | |
9144 | db->estimate_prefix_size(PREFIX_PERPOOL_OMAP, string()); | |
11fdf7f2 TL |
9145 | |
9146 | uint64_t bfree = alloc->get_free(); | |
7c673cae FG |
9147 | |
9148 | if (bluefs) { | |
9f95a23c TL |
9149 | int64_t bluefs_total = bluefs->get_total(bluefs_layout.shared_bdev); |
9150 | int64_t bluefs_free = bluefs->get_free(bluefs_layout.shared_bdev); | |
94b18763 FG |
9151 | // part of our shared device is "free" according to BlueFS, but we |
9152 | // can't touch bluestore_bluefs_min of it. | |
9153 | int64_t shared_available = std::min( | |
11fdf7f2 TL |
9154 | bluefs_free, |
9155 | int64_t(bluefs_total - cct->_conf->bluestore_bluefs_min)); | |
9156 | buf->internally_reserved = bluefs_total - shared_available; | |
94b18763 | 9157 | if (shared_available > 0) { |
11fdf7f2 TL |
9158 | bfree += shared_available; |
9159 | } | |
9160 | // include dedicated db, too, if that isn't the shared device. | |
9f95a23c | 9161 | if (bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 9162 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); |
7c673cae | 9163 | } |
11fdf7f2 TL |
9164 | // call any non-omap bluefs space "internal metadata" |
9165 | buf->internal_metadata = | |
9166 | std::max(bluefs->get_used(), (uint64_t)cct->_conf->bluestore_bluefs_min) | |
9167 | - buf->omap_allocated; | |
7c673cae FG |
9168 | } |
9169 | ||
11fdf7f2 TL |
9170 | uint64_t thin_total, thin_avail; |
9171 | if (bdev->get_thin_utilization(&thin_total, &thin_avail)) { | |
9172 | buf->total += thin_total; | |
9173 | ||
9174 | // we are limited by both the size of the virtual device and the | |
9175 | // underlying physical device. | |
9176 | bfree = std::min(bfree, thin_avail); | |
9177 | ||
9178 | buf->allocated = thin_total - thin_avail; | |
9179 | } else { | |
9180 | buf->total += bdev->get_size(); | |
9181 | } | |
9182 | buf->available = bfree; | |
9183 | } | |
9184 | ||
9185 | int BlueStore::statfs(struct store_statfs_t *buf, | |
9186 | osd_alert_list_t* alerts) | |
9187 | { | |
9188 | if (alerts) { | |
9189 | alerts->clear(); | |
9190 | _log_alerts(*alerts); | |
9191 | } | |
9192 | _get_statfs_overall(buf); | |
31f18b77 | 9193 | { |
11fdf7f2 | 9194 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 9195 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
9196 | buf->data_stored = vstatfs.stored(); |
9197 | buf->data_compressed = vstatfs.compressed(); | |
9198 | buf->data_compressed_original = vstatfs.compressed_original(); | |
9199 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
9200 | } | |
9201 | ||
9202 | dout(20) << __func__ << " " << *buf << dendl; | |
9203 | return 0; | |
9204 | } | |
9205 | ||
9f95a23c TL |
9206 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
9207 | bool *out_per_pool_omap) | |
11fdf7f2 TL |
9208 | { |
9209 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 9210 | |
11fdf7f2 TL |
9211 | if (!per_pool_stat_collection) { |
9212 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
9213 | return -ENOTSUP; | |
7c673cae | 9214 | } |
11fdf7f2 | 9215 | buf->reset(); |
7c673cae | 9216 | |
11fdf7f2 TL |
9217 | { |
9218 | std::lock_guard l(vstatfs_lock); | |
9219 | osd_pools[pool_id].publish(buf); | |
9220 | } | |
9f95a23c TL |
9221 | |
9222 | string key_prefix; | |
9223 | _key_encode_u64(pool_id, &key_prefix); | |
9224 | buf->omap_allocated = db->estimate_prefix_size(PREFIX_PERPOOL_OMAP, | |
9225 | key_prefix); | |
9226 | *out_per_pool_omap = per_pool_omap; | |
9227 | ||
11fdf7f2 | 9228 | dout(10) << __func__ << *buf << dendl; |
7c673cae FG |
9229 | return 0; |
9230 | } | |
9231 | ||
81eedcae TL |
9232 | void BlueStore::_check_legacy_statfs_alert() |
9233 | { | |
9234 | string s; | |
9235 | if (!per_pool_stat_collection && | |
eafe8130 | 9236 | cct->_conf->bluestore_warn_on_legacy_statfs) { |
81eedcae TL |
9237 | s = "legacy statfs reporting detected, " |
9238 | "suggest to run store repair to get consistent statistic reports"; | |
9239 | } | |
9240 | std::lock_guard l(qlock); | |
9241 | legacy_statfs_alert = s; | |
9242 | } | |
9243 | ||
9f95a23c TL |
9244 | void BlueStore::_check_no_per_pool_omap_alert() |
9245 | { | |
9246 | string s; | |
9247 | if (!per_pool_omap && | |
9248 | cct->_conf->bluestore_warn_on_no_per_pool_omap) { | |
9249 | s = "legacy (not per-pool) omap detected, " | |
9250 | "suggest to run store repair to measure per-pool omap usage"; | |
9251 | } | |
9252 | std::lock_guard l(qlock); | |
9253 | no_per_pool_omap_alert = s; | |
9254 | } | |
9255 | ||
7c673cae FG |
9256 | // --------------- |
9257 | // cache | |
9258 | ||
9259 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
9260 | { | |
9f95a23c | 9261 | std::shared_lock l(coll_lock); |
7c673cae FG |
9262 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); |
9263 | if (cp == coll_map.end()) | |
9264 | return CollectionRef(); | |
9265 | return cp->second; | |
9266 | } | |
9267 | ||
9268 | void BlueStore::_queue_reap_collection(CollectionRef& c) | |
9269 | { | |
9270 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
9271 | // _reap_collections and this in the same thread, |
9272 | // so no need a lock. | |
7c673cae FG |
9273 | removed_collections.push_back(c); |
9274 | } | |
9275 | ||
9276 | void BlueStore::_reap_collections() | |
9277 | { | |
94b18763 | 9278 | |
7c673cae FG |
9279 | list<CollectionRef> removed_colls; |
9280 | { | |
94b18763 FG |
9281 | // _queue_reap_collection and this in the same thread. |
9282 | // So no need a lock. | |
9283 | if (!removed_collections.empty()) | |
9284 | removed_colls.swap(removed_collections); | |
9285 | else | |
9286 | return; | |
7c673cae FG |
9287 | } |
9288 | ||
94b18763 FG |
9289 | list<CollectionRef>::iterator p = removed_colls.begin(); |
9290 | while (p != removed_colls.end()) { | |
7c673cae FG |
9291 | CollectionRef c = *p; |
9292 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
9293 | if (c->onode_map.map_any([&](OnodeRef o) { | |
11fdf7f2 | 9294 | ceph_assert(!o->exists); |
7c673cae FG |
9295 | if (o->flushing_count.load()) { |
9296 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
9297 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 9298 | return true; |
7c673cae | 9299 | } |
94b18763 | 9300 | return false; |
7c673cae | 9301 | })) { |
94b18763 | 9302 | ++p; |
7c673cae FG |
9303 | continue; |
9304 | } | |
9305 | c->onode_map.clear(); | |
94b18763 | 9306 | p = removed_colls.erase(p); |
7c673cae FG |
9307 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
9308 | } | |
94b18763 | 9309 | if (removed_colls.empty()) { |
7c673cae | 9310 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
9311 | } else { |
9312 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
9313 | } |
9314 | } | |
9315 | ||
9316 | void BlueStore::_update_cache_logger() | |
9317 | { | |
9318 | uint64_t num_onodes = 0; | |
9f95a23c | 9319 | uint64_t num_pinned_onodes = 0; |
7c673cae FG |
9320 | uint64_t num_extents = 0; |
9321 | uint64_t num_blobs = 0; | |
9322 | uint64_t num_buffers = 0; | |
9323 | uint64_t num_buffer_bytes = 0; | |
9f95a23c TL |
9324 | for (auto c : onode_cache_shards) { |
9325 | c->add_stats(&num_onodes, &num_pinned_onodes); | |
9326 | } | |
9327 | for (auto c : buffer_cache_shards) { | |
9328 | c->add_stats(&num_extents, &num_blobs, | |
9329 | &num_buffers, &num_buffer_bytes); | |
7c673cae FG |
9330 | } |
9331 | logger->set(l_bluestore_onodes, num_onodes); | |
9f95a23c | 9332 | logger->set(l_bluestore_pinned_onodes, num_pinned_onodes); |
7c673cae FG |
9333 | logger->set(l_bluestore_extents, num_extents); |
9334 | logger->set(l_bluestore_blobs, num_blobs); | |
9335 | logger->set(l_bluestore_buffers, num_buffers); | |
9336 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
9337 | } | |
9338 | ||
9339 | // --------------- | |
9340 | // read operations | |
9341 | ||
9342 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
9343 | { | |
9344 | return _get_collection(cid); | |
9345 | } | |
9346 | ||
11fdf7f2 TL |
9347 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
9348 | const coll_t& cid) | |
7c673cae | 9349 | { |
9f95a23c TL |
9350 | std::unique_lock l{coll_lock}; |
9351 | auto c = ceph::make_ref<Collection>( | |
11fdf7f2 | 9352 | this, |
9f95a23c TL |
9353 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
9354 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
11fdf7f2 TL |
9355 | cid); |
9356 | new_coll_map[cid] = c; | |
9f95a23c | 9357 | _osr_attach(c.get()); |
11fdf7f2 TL |
9358 | return c; |
9359 | } | |
9360 | ||
9361 | void BlueStore::set_collection_commit_queue( | |
9362 | const coll_t& cid, | |
9363 | ContextQueue *commit_queue) | |
9364 | { | |
9365 | if (commit_queue) { | |
9f95a23c | 9366 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
9367 | if (coll_map.count(cid)) { |
9368 | coll_map[cid]->commit_queue = commit_queue; | |
9369 | } else if (new_coll_map.count(cid)) { | |
9370 | new_coll_map[cid]->commit_queue = commit_queue; | |
9371 | } | |
9372 | } | |
7c673cae FG |
9373 | } |
9374 | ||
11fdf7f2 | 9375 | |
7c673cae FG |
9376 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
9377 | { | |
9378 | Collection *c = static_cast<Collection *>(c_.get()); | |
9379 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
9380 | if (!c->exists) | |
9381 | return false; | |
9382 | ||
9383 | bool r = true; | |
9384 | ||
9385 | { | |
9f95a23c | 9386 | std::shared_lock l(c->lock); |
7c673cae FG |
9387 | OnodeRef o = c->get_onode(oid, false); |
9388 | if (!o || !o->exists) | |
9389 | r = false; | |
9390 | } | |
9391 | ||
7c673cae FG |
9392 | return r; |
9393 | } | |
9394 | ||
7c673cae FG |
9395 | int BlueStore::stat( |
9396 | CollectionHandle &c_, | |
9397 | const ghobject_t& oid, | |
9398 | struct stat *st, | |
9399 | bool allow_eio) | |
9400 | { | |
9401 | Collection *c = static_cast<Collection *>(c_.get()); | |
9402 | if (!c->exists) | |
9403 | return -ENOENT; | |
9404 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
9405 | ||
9406 | { | |
9f95a23c | 9407 | std::shared_lock l(c->lock); |
7c673cae FG |
9408 | OnodeRef o = c->get_onode(oid, false); |
9409 | if (!o || !o->exists) | |
9410 | return -ENOENT; | |
9411 | st->st_size = o->onode.size; | |
9412 | st->st_blksize = 4096; | |
9413 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
9414 | st->st_nlink = 1; | |
9415 | } | |
9416 | ||
7c673cae FG |
9417 | int r = 0; |
9418 | if (_debug_mdata_eio(oid)) { | |
9419 | r = -EIO; | |
9420 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9421 | } | |
9422 | return r; | |
9423 | } | |
9424 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 9425 | CollectionHandle& ch, |
7c673cae FG |
9426 | const pool_opts_t& opts) |
9427 | { | |
7c673cae | 9428 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 9429 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
9430 | if (!c->exists) |
9431 | return -ENOENT; | |
9f95a23c | 9432 | std::unique_lock l{c->lock}; |
7c673cae FG |
9433 | c->pool_opts = opts; |
9434 | return 0; | |
9435 | } | |
9436 | ||
7c673cae FG |
9437 | int BlueStore::read( |
9438 | CollectionHandle &c_, | |
9439 | const ghobject_t& oid, | |
9440 | uint64_t offset, | |
9441 | size_t length, | |
9442 | bufferlist& bl, | |
224ce89b | 9443 | uint32_t op_flags) |
7c673cae | 9444 | { |
11fdf7f2 | 9445 | auto start = mono_clock::now(); |
7c673cae FG |
9446 | Collection *c = static_cast<Collection *>(c_.get()); |
9447 | const coll_t &cid = c->get_cid(); | |
9448 | dout(15) << __func__ << " " << cid << " " << oid | |
9449 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9450 | << dendl; | |
9451 | if (!c->exists) | |
9452 | return -ENOENT; | |
9453 | ||
9454 | bl.clear(); | |
9455 | int r; | |
9456 | { | |
9f95a23c | 9457 | std::shared_lock l(c->lock); |
11fdf7f2 | 9458 | auto start1 = mono_clock::now(); |
7c673cae | 9459 | OnodeRef o = c->get_onode(oid, false); |
494da23a TL |
9460 | log_latency("get_onode@read", |
9461 | l_bluestore_read_onode_meta_lat, | |
9462 | mono_clock::now() - start1, | |
9463 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9464 | if (!o || !o->exists) { |
9465 | r = -ENOENT; | |
9466 | goto out; | |
9467 | } | |
9468 | ||
9469 | if (offset == length && offset == 0) | |
9470 | length = o->onode.size; | |
9471 | ||
9472 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
9473 | if (r == -EIO) { |
9474 | logger->inc(l_bluestore_read_eio); | |
9475 | } | |
7c673cae FG |
9476 | } |
9477 | ||
9478 | out: | |
28e407b8 | 9479 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
9480 | r = -EIO; |
9481 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
9482 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
9483 | cct->_conf->bluestore_debug_random_read_err && | |
9484 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
9485 | 100.0)) == 0) { | |
224ce89b WB |
9486 | dout(0) << __func__ << ": inject random EIO" << dendl; |
9487 | r = -EIO; | |
7c673cae FG |
9488 | } |
9489 | dout(10) << __func__ << " " << cid << " " << oid | |
9490 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9491 | << " = " << r << dendl; | |
494da23a TL |
9492 | log_latency(__func__, |
9493 | l_bluestore_read_lat, | |
9494 | mono_clock::now() - start, | |
9495 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9496 | return r; |
9497 | } | |
9498 | ||
9f95a23c | 9499 | void BlueStore::_read_cache( |
7c673cae FG |
9500 | OnodeRef o, |
9501 | uint64_t offset, | |
9502 | size_t length, | |
9f95a23c TL |
9503 | int read_cache_policy, |
9504 | ready_regions_t& ready_regions, | |
9505 | blobs2read_t& blobs2read) | |
7c673cae | 9506 | { |
7c673cae | 9507 | // build blob-wise list to of stuff read (that isn't cached) |
7c673cae FG |
9508 | unsigned left = length; |
9509 | uint64_t pos = offset; | |
7c673cae FG |
9510 | auto lp = o->extent_map.seek_lextent(offset); |
9511 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
9512 | if (pos < lp->logical_offset) { | |
9513 | unsigned hole = lp->logical_offset - pos; | |
9514 | if (hole >= left) { | |
9f95a23c | 9515 | break; |
7c673cae FG |
9516 | } |
9517 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
9f95a23c | 9518 | << std::dec << dendl; |
7c673cae FG |
9519 | pos += hole; |
9520 | left -= hole; | |
9521 | } | |
94b18763 | 9522 | BlobRef& bptr = lp->blob; |
7c673cae FG |
9523 | unsigned l_off = pos - lp->logical_offset; |
9524 | unsigned b_off = l_off + lp->blob_offset; | |
9525 | unsigned b_len = std::min(left, lp->length - l_off); | |
9526 | ||
9527 | ready_regions_t cache_res; | |
9528 | interval_set<uint32_t> cache_interval; | |
9529 | bptr->shared_blob->bc.read( | |
91327a77 AA |
9530 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
9531 | read_cache_policy); | |
7c673cae | 9532 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c TL |
9533 | << " need 0x" << b_off << "~" << b_len |
9534 | << " cache has 0x" << cache_interval | |
9535 | << std::dec << dendl; | |
7c673cae FG |
9536 | |
9537 | auto pc = cache_res.begin(); | |
11fdf7f2 | 9538 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
9539 | while (b_len > 0) { |
9540 | unsigned l; | |
9541 | if (pc != cache_res.end() && | |
9f95a23c TL |
9542 | pc->first == b_off) { |
9543 | l = pc->second.length(); | |
9544 | ready_regions[pos].claim(pc->second); | |
9545 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" | |
9546 | << b_off << "~" << l << std::dec << dendl; | |
9547 | ++pc; | |
7c673cae | 9548 | } else { |
9f95a23c TL |
9549 | l = b_len; |
9550 | if (pc != cache_res.end()) { | |
9551 | ceph_assert(pc->first > b_off); | |
9552 | l = pc->first - b_off; | |
9553 | } | |
9554 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
9555 | << b_off << "~" << l << std::dec << dendl; | |
9556 | // merge regions | |
9557 | { | |
9558 | uint64_t r_off = b_off; | |
9559 | uint64_t r_len = l; | |
9560 | uint64_t front = r_off % chunk_size; | |
9561 | if (front) { | |
9562 | r_off -= front; | |
9563 | r_len += front; | |
9564 | } | |
9565 | unsigned tail = r_len % chunk_size; | |
9566 | if (tail) { | |
9567 | r_len += chunk_size - tail; | |
9568 | } | |
9569 | bool merged = false; | |
9570 | regions2read_t& r2r = blobs2read[bptr]; | |
9571 | if (r2r.size()) { | |
9572 | read_req_t& pre = r2r.back(); | |
9573 | if (r_off <= (pre.r_off + pre.r_len)) { | |
9574 | front += (r_off - pre.r_off); | |
9575 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
9576 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9577 | merged = true; | |
9578 | } | |
9579 | } | |
9580 | if (!merged) { | |
9581 | read_req_t req(r_off, r_len); | |
9582 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9583 | r2r.emplace_back(std::move(req)); | |
9584 | } | |
9585 | } | |
7c673cae FG |
9586 | } |
9587 | pos += l; | |
9588 | b_off += l; | |
9589 | left -= l; | |
9590 | b_len -= l; | |
9591 | } | |
9592 | ++lp; | |
9593 | } | |
9f95a23c | 9594 | } |
7c673cae | 9595 | |
9f95a23c TL |
9596 | int BlueStore::_prepare_read_ioc( |
9597 | blobs2read_t& blobs2read, | |
9598 | vector<bufferlist>* compressed_blob_bls, | |
9599 | IOContext* ioc) | |
9600 | { | |
7c673cae | 9601 | for (auto& p : blobs2read) { |
94b18763 | 9602 | const BlobRef& bptr = p.first; |
11fdf7f2 | 9603 | regions2read_t& r2r = p.second; |
7c673cae | 9604 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9605 | << " need " << r2r << std::dec << dendl; |
7c673cae FG |
9606 | if (bptr->get_blob().is_compressed()) { |
9607 | // read the whole thing | |
9f95a23c TL |
9608 | if (compressed_blob_bls->empty()) { |
9609 | // ensure we avoid any reallocation on subsequent blobs | |
9610 | compressed_blob_bls->reserve(blobs2read.size()); | |
9611 | } | |
9612 | compressed_blob_bls->push_back(bufferlist()); | |
9613 | bufferlist& bl = compressed_blob_bls->back(); | |
9614 | auto r = bptr->get_blob().map( | |
9615 | 0, bptr->get_blob().get_ondisk_length(), | |
9616 | [&](uint64_t offset, uint64_t length) { | |
9617 | int r = bdev->aio_read(offset, length, &bl, ioc); | |
9618 | if (r < 0) | |
7c673cae FG |
9619 | return r; |
9620 | return 0; | |
9f95a23c | 9621 | }); |
b32b8144 FG |
9622 | if (r < 0) { |
9623 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
9624 | if (r == -EIO) { | |
9625 | // propagate EIO to caller | |
9626 | return r; | |
9627 | } | |
11fdf7f2 | 9628 | ceph_assert(r == 0); |
b32b8144 | 9629 | } |
7c673cae FG |
9630 | } else { |
9631 | // read the pieces | |
11fdf7f2 | 9632 | for (auto& req : r2r) { |
9f95a23c TL |
9633 | dout(20) << __func__ << " region 0x" << std::hex |
9634 | << req.regs.front().logical_offset | |
9635 | << ": 0x" << req.regs.front().blob_xoffset | |
9636 | << " reading 0x" << req.r_off | |
9637 | << "~" << req.r_len << std::dec | |
9638 | << dendl; | |
7c673cae | 9639 | |
9f95a23c TL |
9640 | // read it |
9641 | auto r = bptr->get_blob().map( | |
9642 | req.r_off, req.r_len, | |
9643 | [&](uint64_t offset, uint64_t length) { | |
9644 | int r = bdev->aio_read(offset, length, &req.bl, ioc); | |
9645 | if (r < 0) | |
7c673cae FG |
9646 | return r; |
9647 | return 0; | |
9f95a23c | 9648 | }); |
b32b8144 FG |
9649 | if (r < 0) { |
9650 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
9651 | << dendl; | |
9652 | if (r == -EIO) { | |
9653 | // propagate EIO to caller | |
9654 | return r; | |
9655 | } | |
11fdf7f2 | 9656 | ceph_assert(r == 0); |
b32b8144 | 9657 | } |
9f95a23c | 9658 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
9659 | } |
9660 | } | |
9661 | } | |
9f95a23c TL |
9662 | return 0; |
9663 | } | |
11fdf7f2 | 9664 | |
9f95a23c TL |
9665 | int BlueStore::_generate_read_result_bl( |
9666 | OnodeRef o, | |
9667 | uint64_t offset, | |
9668 | size_t length, | |
9669 | ready_regions_t& ready_regions, | |
9670 | vector<bufferlist>& compressed_blob_bls, | |
9671 | blobs2read_t& blobs2read, | |
9672 | bool buffered, | |
9673 | bool* csum_error, | |
9674 | bufferlist& bl) | |
9675 | { | |
9676 | // enumerate and decompress desired blobs | |
7c673cae FG |
9677 | auto p = compressed_blob_bls.begin(); |
9678 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
9679 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 9680 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 9681 | regions2read_t& r2r = b2r_it->second; |
7c673cae | 9682 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9683 | << " need 0x" << r2r << std::dec << dendl; |
7c673cae | 9684 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 9685 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
9686 | bufferlist& compressed_bl = *p++; |
9687 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
9f95a23c TL |
9688 | r2r.front().regs.front().logical_offset) < 0) { |
9689 | *csum_error = true; | |
9690 | return -EIO; | |
7c673cae FG |
9691 | } |
9692 | bufferlist raw_bl; | |
9f95a23c | 9693 | auto r = _decompress(compressed_bl, &raw_bl); |
7c673cae | 9694 | if (r < 0) |
9f95a23c | 9695 | return r; |
7c673cae | 9696 | if (buffered) { |
9f95a23c TL |
9697 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, |
9698 | raw_bl); | |
7c673cae | 9699 | } |
11fdf7f2 TL |
9700 | for (auto& req : r2r) { |
9701 | for (auto& r : req.regs) { | |
9702 | ready_regions[r.logical_offset].substr_of( | |
9703 | raw_bl, r.blob_xoffset, r.length); | |
9704 | } | |
7c673cae FG |
9705 | } |
9706 | } else { | |
11fdf7f2 | 9707 | for (auto& req : r2r) { |
9f95a23c TL |
9708 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, |
9709 | req.regs.front().logical_offset) < 0) { | |
9710 | *csum_error = true; | |
9711 | return -EIO; | |
9712 | } | |
9713 | if (buffered) { | |
9714 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
9715 | req.r_off, req.bl); | |
9716 | } | |
7c673cae | 9717 | |
9f95a23c TL |
9718 | // prune and keep result |
9719 | for (const auto& r : req.regs) { | |
9720 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
11fdf7f2 | 9721 | } |
7c673cae FG |
9722 | } |
9723 | } | |
9724 | ++b2r_it; | |
9725 | } | |
9726 | ||
9727 | // generate a resulting buffer | |
9728 | auto pr = ready_regions.begin(); | |
9729 | auto pr_end = ready_regions.end(); | |
9f95a23c | 9730 | uint64_t pos = 0; |
7c673cae FG |
9731 | while (pos < length) { |
9732 | if (pr != pr_end && pr->first == pos + offset) { | |
9733 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9734 | << ": data from 0x" << pr->first << "~" << pr->second.length() |
9735 | << std::dec << dendl; | |
7c673cae FG |
9736 | pos += pr->second.length(); |
9737 | bl.claim_append(pr->second); | |
9738 | ++pr; | |
9739 | } else { | |
9740 | uint64_t l = length - pos; | |
9741 | if (pr != pr_end) { | |
11fdf7f2 | 9742 | ceph_assert(pr->first > pos + offset); |
9f95a23c | 9743 | l = pr->first - (pos + offset); |
7c673cae FG |
9744 | } |
9745 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9746 | << ": zeros for 0x" << (pos + offset) << "~" << l |
9747 | << std::dec << dendl; | |
7c673cae FG |
9748 | bl.append_zero(l); |
9749 | pos += l; | |
9750 | } | |
9751 | } | |
11fdf7f2 TL |
9752 | ceph_assert(bl.length() == length); |
9753 | ceph_assert(pos == length); | |
9754 | ceph_assert(pr == pr_end); | |
9f95a23c TL |
9755 | return 0; |
9756 | } | |
9757 | ||
9758 | int BlueStore::_do_read( | |
9759 | Collection *c, | |
9760 | OnodeRef o, | |
9761 | uint64_t offset, | |
9762 | size_t length, | |
9763 | bufferlist& bl, | |
9764 | uint32_t op_flags, | |
9765 | uint64_t retry_count) | |
9766 | { | |
9767 | FUNCTRACE(cct); | |
9768 | int r = 0; | |
9769 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
9770 | ||
9771 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
9772 | << " size 0x" << o->onode.size << " (" << std::dec | |
9773 | << o->onode.size << ")" << dendl; | |
9774 | bl.clear(); | |
9775 | ||
9776 | if (offset >= o->onode.size) { | |
9777 | return r; | |
9778 | } | |
9779 | ||
9780 | // generally, don't buffer anything, unless the client explicitly requests | |
9781 | // it. | |
9782 | bool buffered = false; | |
9783 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
9784 | dout(20) << __func__ << " will do buffered read" << dendl; | |
9785 | buffered = true; | |
9786 | } else if (cct->_conf->bluestore_default_buffered_read && | |
9787 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
9788 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
9789 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
9790 | buffered = true; | |
9791 | } | |
9792 | ||
9793 | if (offset + length > o->onode.size) { | |
9794 | length = o->onode.size - offset; | |
9795 | } | |
9796 | ||
9797 | auto start = mono_clock::now(); | |
9798 | o->extent_map.fault_range(db, offset, length); | |
9799 | log_latency(__func__, | |
9800 | l_bluestore_read_onode_meta_lat, | |
9801 | mono_clock::now() - start, | |
9802 | cct->_conf->bluestore_log_op_age); | |
9803 | _dump_onode<30>(cct, *o); | |
9804 | ||
9805 | // for deep-scrub, we only read dirty cache and bypass clean cache in | |
9806 | // order to read underlying block device in case there are silent disk errors. | |
9807 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
9808 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
9809 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
9810 | } | |
9811 | ||
9812 | // build blob-wise list to of stuff read (that isn't cached) | |
9813 | ready_regions_t ready_regions; | |
9814 | blobs2read_t blobs2read; | |
9815 | _read_cache(o, offset, length, read_cache_policy, ready_regions, blobs2read); | |
9816 | ||
9817 | ||
9818 | // read raw blob data. | |
9819 | start = mono_clock::now(); // for the sake of simplicity | |
9820 | // measure the whole block below. | |
9821 | // The error isn't that much... | |
9822 | vector<bufferlist> compressed_blob_bls; | |
9823 | IOContext ioc(cct, NULL, true); // allow EIO | |
9824 | r = _prepare_read_ioc(blobs2read, &compressed_blob_bls, &ioc); | |
9825 | // we always issue aio for reading, so errors other than EIO are not allowed | |
9826 | if (r < 0) | |
9827 | return r; | |
9828 | ||
9829 | int64_t num_ios = length; | |
9830 | if (ioc.has_pending_aios()) { | |
9831 | num_ios = -ioc.get_num_ios(); | |
9832 | bdev->aio_submit(&ioc); | |
9833 | dout(20) << __func__ << " waiting for aio" << dendl; | |
9834 | ioc.aio_wait(); | |
9835 | r = ioc.get_return_value(); | |
9836 | if (r < 0) { | |
9837 | ceph_assert(r == -EIO); // no other errors allowed | |
9838 | return -EIO; | |
9839 | } | |
9840 | } | |
9841 | log_latency_fn(__func__, | |
9842 | l_bluestore_read_wait_aio_lat, | |
9843 | mono_clock::now() - start, | |
9844 | cct->_conf->bluestore_log_op_age, | |
9845 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
9846 | ); | |
9847 | ||
9848 | bool csum_error = false; | |
9849 | r = _generate_read_result_bl(o, offset, length, ready_regions, | |
9850 | compressed_blob_bls, blobs2read, | |
9851 | buffered, &csum_error, bl); | |
9852 | if (csum_error) { | |
9853 | // Handles spurious read errors caused by a kernel bug. | |
9854 | // We sometimes get all-zero pages as a result of the read under | |
9855 | // high memory pressure. Retrying the failing read succeeds in most | |
9856 | // cases. | |
9857 | // See also: http://tracker.ceph.com/issues/22464 | |
9858 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
9859 | return -EIO; | |
9860 | } | |
9861 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
9862 | } | |
7c673cae | 9863 | r = bl.length(); |
f64942e4 AA |
9864 | if (retry_count) { |
9865 | logger->inc(l_bluestore_reads_with_retries); | |
9866 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
9867 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
9868 | } | |
7c673cae FG |
9869 | return r; |
9870 | } | |
9871 | ||
9872 | int BlueStore::_verify_csum(OnodeRef& o, | |
9873 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
9874 | const bufferlist& bl, | |
9875 | uint64_t logical_offset) const | |
9876 | { | |
9877 | int bad; | |
9878 | uint64_t bad_csum; | |
11fdf7f2 | 9879 | auto start = mono_clock::now(); |
7c673cae | 9880 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
9881 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
9882 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
9883 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
9884 | bad = blob_xoffset; | |
9885 | r = -1; | |
9886 | bad_csum = 0xDEADBEEF; | |
9887 | } | |
7c673cae FG |
9888 | if (r < 0) { |
9889 | if (r == -1) { | |
9890 | PExtentVector pex; | |
9891 | blob->map( | |
9892 | bad, | |
9893 | blob->get_csum_chunk_size(), | |
9894 | [&](uint64_t offset, uint64_t length) { | |
9895 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
9896 | return 0; | |
9897 | }); | |
9898 | derr << __func__ << " bad " | |
9899 | << Checksummer::get_csum_type_string(blob->csum_type) | |
9900 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
9901 | << " checksum at blob offset 0x" << bad | |
9902 | << ", got 0x" << bad_csum << ", expected 0x" | |
9903 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
9904 | << ", device location " << pex | |
9905 | << ", logical extent 0x" << std::hex | |
9906 | << (logical_offset + bad - blob_xoffset) << "~" | |
9907 | << blob->get_csum_chunk_size() << std::dec | |
9908 | << ", object " << o->oid | |
9909 | << dendl; | |
9910 | } else { | |
9911 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
9912 | } | |
9913 | } | |
494da23a TL |
9914 | log_latency(__func__, |
9915 | l_bluestore_csum_lat, | |
9916 | mono_clock::now() - start, | |
9917 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 TL |
9918 | if (cct->_conf->bluestore_ignore_data_csum) { |
9919 | return 0; | |
9920 | } | |
7c673cae FG |
9921 | return r; |
9922 | } | |
9923 | ||
9924 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
9925 | { | |
9926 | int r = 0; | |
11fdf7f2 TL |
9927 | auto start = mono_clock::now(); |
9928 | auto i = source.cbegin(); | |
7c673cae | 9929 | bluestore_compression_header_t chdr; |
11fdf7f2 | 9930 | decode(chdr, i); |
7c673cae FG |
9931 | int alg = int(chdr.type); |
9932 | CompressorRef cp = compressor; | |
9933 | if (!cp || (int)cp->get_type() != alg) { | |
9934 | cp = Compressor::create(cct, alg); | |
9935 | } | |
9936 | ||
9937 | if (!cp.get()) { | |
9938 | // if compressor isn't available - error, because cannot return | |
9939 | // decompressed data? | |
11fdf7f2 TL |
9940 | |
9941 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
9942 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
9943 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
9944 | r = -EIO; |
9945 | } else { | |
9946 | r = cp->decompress(i, chdr.length, *result); | |
9947 | if (r < 0) { | |
9948 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
9949 | r = -EIO; | |
9950 | } | |
9951 | } | |
494da23a TL |
9952 | log_latency(__func__, |
9953 | l_bluestore_decompress_lat, | |
9954 | mono_clock::now() - start, | |
9955 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9956 | return r; |
9957 | } | |
9958 | ||
9959 | // this stores fiemap into interval_set, other variations | |
9960 | // use it internally | |
9961 | int BlueStore::_fiemap( | |
9962 | CollectionHandle &c_, | |
9963 | const ghobject_t& oid, | |
9964 | uint64_t offset, | |
9965 | size_t length, | |
9966 | interval_set<uint64_t>& destset) | |
9967 | { | |
9968 | Collection *c = static_cast<Collection *>(c_.get()); | |
9969 | if (!c->exists) | |
9970 | return -ENOENT; | |
9971 | { | |
9f95a23c | 9972 | std::shared_lock l(c->lock); |
7c673cae FG |
9973 | |
9974 | OnodeRef o = c->get_onode(oid, false); | |
9975 | if (!o || !o->exists) { | |
9976 | return -ENOENT; | |
9977 | } | |
81eedcae | 9978 | _dump_onode<30>(cct, *o); |
7c673cae FG |
9979 | |
9980 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
9981 | << " size 0x" << o->onode.size << std::dec << dendl; | |
9982 | ||
9983 | boost::intrusive::set<Extent>::iterator ep, eend; | |
9984 | if (offset >= o->onode.size) | |
9985 | goto out; | |
9986 | ||
9987 | if (offset + length > o->onode.size) { | |
9988 | length = o->onode.size - offset; | |
9989 | } | |
9990 | ||
9991 | o->extent_map.fault_range(db, offset, length); | |
9992 | eend = o->extent_map.extent_map.end(); | |
9993 | ep = o->extent_map.seek_lextent(offset); | |
9994 | while (length > 0) { | |
9995 | dout(20) << __func__ << " offset " << offset << dendl; | |
9996 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
9997 | ++ep; | |
9998 | continue; | |
9999 | } | |
10000 | ||
10001 | uint64_t x_len = length; | |
10002 | if (ep != eend && ep->logical_offset <= offset) { | |
10003 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 10004 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
10005 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
10006 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
10007 | destset.insert(offset, x_len); | |
10008 | length -= x_len; | |
10009 | offset += x_len; | |
10010 | if (x_off + x_len == ep->length) | |
10011 | ++ep; | |
10012 | continue; | |
10013 | } | |
10014 | if (ep != eend && | |
10015 | ep->logical_offset > offset && | |
10016 | ep->logical_offset - offset < x_len) { | |
10017 | x_len = ep->logical_offset - offset; | |
10018 | } | |
10019 | offset += x_len; | |
10020 | length -= x_len; | |
10021 | } | |
10022 | } | |
9f95a23c TL |
10023 | |
10024 | out: | |
10025 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10026 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
10027 | return 0; | |
10028 | } | |
10029 | ||
10030 | int BlueStore::fiemap( | |
10031 | CollectionHandle &c_, | |
10032 | const ghobject_t& oid, | |
10033 | uint64_t offset, | |
10034 | size_t length, | |
10035 | bufferlist& bl) | |
10036 | { | |
10037 | interval_set<uint64_t> m; | |
10038 | int r = _fiemap(c_, oid, offset, length, m); | |
10039 | if (r >= 0) { | |
10040 | encode(m, bl); | |
10041 | } | |
10042 | return r; | |
10043 | } | |
10044 | ||
10045 | int BlueStore::fiemap( | |
10046 | CollectionHandle &c_, | |
10047 | const ghobject_t& oid, | |
10048 | uint64_t offset, | |
10049 | size_t length, | |
10050 | map<uint64_t, uint64_t>& destmap) | |
10051 | { | |
10052 | interval_set<uint64_t> m; | |
10053 | int r = _fiemap(c_, oid, offset, length, m); | |
10054 | if (r >= 0) { | |
10055 | destmap = std::move(m).detach(); | |
10056 | } | |
10057 | return r; | |
10058 | } | |
10059 | ||
10060 | int BlueStore::readv( | |
10061 | CollectionHandle &c_, | |
10062 | const ghobject_t& oid, | |
10063 | interval_set<uint64_t>& m, | |
10064 | bufferlist& bl, | |
10065 | uint32_t op_flags) | |
10066 | { | |
10067 | auto start = mono_clock::now(); | |
10068 | Collection *c = static_cast<Collection *>(c_.get()); | |
10069 | const coll_t &cid = c->get_cid(); | |
10070 | dout(15) << __func__ << " " << cid << " " << oid | |
10071 | << " fiemap " << m | |
10072 | << dendl; | |
10073 | if (!c->exists) | |
10074 | return -ENOENT; | |
10075 | ||
10076 | bl.clear(); | |
10077 | int r; | |
10078 | { | |
10079 | std::shared_lock l(c->lock); | |
10080 | auto start1 = mono_clock::now(); | |
10081 | OnodeRef o = c->get_onode(oid, false); | |
10082 | log_latency("get_onode@read", | |
10083 | l_bluestore_read_onode_meta_lat, | |
10084 | mono_clock::now() - start1, | |
10085 | cct->_conf->bluestore_log_op_age); | |
10086 | if (!o || !o->exists) { | |
10087 | r = -ENOENT; | |
10088 | goto out; | |
10089 | } | |
10090 | ||
10091 | if (m.empty()) { | |
10092 | r = 0; | |
10093 | goto out; | |
10094 | } | |
10095 | ||
10096 | r = _do_readv(c, o, m, bl, op_flags); | |
10097 | if (r == -EIO) { | |
10098 | logger->inc(l_bluestore_read_eio); | |
10099 | } | |
10100 | } | |
10101 | ||
10102 | out: | |
10103 | if (r >= 0 && _debug_data_eio(oid)) { | |
10104 | r = -EIO; | |
10105 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10106 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ | |
10107 | cct->_conf->bluestore_debug_random_read_err && | |
10108 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
10109 | 100.0)) == 0) { | |
10110 | dout(0) << __func__ << ": inject random EIO" << dendl; | |
10111 | r = -EIO; | |
10112 | } | |
10113 | dout(10) << __func__ << " " << cid << " " << oid | |
10114 | << " fiemap " << m << std::dec | |
10115 | << " = " << r << dendl; | |
10116 | log_latency(__func__, | |
10117 | l_bluestore_read_lat, | |
10118 | mono_clock::now() - start, | |
10119 | cct->_conf->bluestore_log_op_age); | |
10120 | return r; | |
10121 | } | |
10122 | ||
10123 | int BlueStore::_do_readv( | |
10124 | Collection *c, | |
10125 | OnodeRef o, | |
10126 | const interval_set<uint64_t>& m, | |
10127 | bufferlist& bl, | |
10128 | uint32_t op_flags, | |
10129 | uint64_t retry_count) | |
10130 | { | |
10131 | FUNCTRACE(cct); | |
10132 | int r = 0; | |
10133 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
10134 | ||
10135 | dout(20) << __func__ << " fiemap " << m << std::hex | |
10136 | << " size 0x" << o->onode.size << " (" << std::dec | |
10137 | << o->onode.size << ")" << dendl; | |
10138 | ||
10139 | // generally, don't buffer anything, unless the client explicitly requests | |
10140 | // it. | |
10141 | bool buffered = false; | |
10142 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
10143 | dout(20) << __func__ << " will do buffered read" << dendl; | |
10144 | buffered = true; | |
10145 | } else if (cct->_conf->bluestore_default_buffered_read && | |
10146 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
10147 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
10148 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
10149 | buffered = true; | |
10150 | } | |
10151 | // this method must be idempotent since we may call it several times | |
10152 | // before we finally read the expected result. | |
10153 | bl.clear(); | |
10154 | ||
10155 | // call fiemap first! | |
10156 | ceph_assert(m.range_start() <= o->onode.size); | |
10157 | ceph_assert(m.range_end() <= o->onode.size); | |
10158 | auto start = mono_clock::now(); | |
10159 | o->extent_map.fault_range(db, m.range_start(), m.range_end() - m.range_start()); | |
10160 | log_latency(__func__, | |
10161 | l_bluestore_read_onode_meta_lat, | |
10162 | mono_clock::now() - start, | |
10163 | cct->_conf->bluestore_log_op_age); | |
10164 | _dump_onode<30>(cct, *o); | |
10165 | ||
10166 | IOContext ioc(cct, NULL, true); // allow EIO | |
10167 | vector<std::tuple<ready_regions_t, vector<bufferlist>, blobs2read_t>> raw_results; | |
10168 | raw_results.reserve(m.num_intervals()); | |
10169 | int i = 0; | |
10170 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10171 | raw_results.push_back({}); | |
10172 | _read_cache(o, p.get_start(), p.get_len(), read_cache_policy, | |
10173 | std::get<0>(raw_results[i]), std::get<2>(raw_results[i])); | |
10174 | r = _prepare_read_ioc(std::get<2>(raw_results[i]), &std::get<1>(raw_results[i]), &ioc); | |
10175 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10176 | if (r < 0) | |
10177 | return r; | |
10178 | } | |
10179 | ||
10180 | auto num_ios = m.size(); | |
10181 | if (ioc.has_pending_aios()) { | |
10182 | num_ios = ioc.get_num_ios(); | |
10183 | bdev->aio_submit(&ioc); | |
10184 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10185 | ioc.aio_wait(); | |
10186 | r = ioc.get_return_value(); | |
10187 | if (r < 0) { | |
10188 | ceph_assert(r == -EIO); // no other errors allowed | |
10189 | return -EIO; | |
10190 | } | |
10191 | } | |
10192 | log_latency_fn(__func__, | |
10193 | l_bluestore_read_wait_aio_lat, | |
10194 | mono_clock::now() - start, | |
10195 | cct->_conf->bluestore_log_op_age, | |
10196 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10197 | ); | |
10198 | ||
10199 | ceph_assert(raw_results.size() == (size_t)m.num_intervals()); | |
10200 | i = 0; | |
10201 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10202 | bool csum_error = false; | |
10203 | bufferlist t; | |
10204 | r = _generate_read_result_bl(o, p.get_start(), p.get_len(), | |
10205 | std::get<0>(raw_results[i]), | |
10206 | std::get<1>(raw_results[i]), | |
10207 | std::get<2>(raw_results[i]), | |
10208 | buffered, &csum_error, t); | |
10209 | if (csum_error) { | |
10210 | // Handles spurious read errors caused by a kernel bug. | |
10211 | // We sometimes get all-zero pages as a result of the read under | |
10212 | // high memory pressure. Retrying the failing read succeeds in most | |
10213 | // cases. | |
10214 | // See also: http://tracker.ceph.com/issues/22464 | |
10215 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10216 | return -EIO; | |
10217 | } | |
10218 | return _do_readv(c, o, m, bl, op_flags, retry_count + 1); | |
10219 | } | |
10220 | bl.claim_append(t); | |
10221 | } | |
10222 | if (retry_count) { | |
10223 | logger->inc(l_bluestore_reads_with_retries); | |
10224 | dout(5) << __func__ << " read fiemap " << m | |
10225 | << " failed " << retry_count << " times before succeeding" | |
10226 | << dendl; | |
10227 | } | |
10228 | return bl.length(); | |
7c673cae FG |
10229 | } |
10230 | ||
9f95a23c | 10231 | int BlueStore::dump_onode(CollectionHandle &c_, |
7c673cae | 10232 | const ghobject_t& oid, |
9f95a23c TL |
10233 | const string& section_name, |
10234 | Formatter *f) | |
7c673cae | 10235 | { |
9f95a23c TL |
10236 | Collection *c = static_cast<Collection *>(c_.get()); |
10237 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10238 | if (!c->exists) | |
10239 | return -ENOENT; | |
7c673cae | 10240 | |
9f95a23c TL |
10241 | int r; |
10242 | { | |
10243 | std::shared_lock l(c->lock); | |
10244 | ||
10245 | OnodeRef o = c->get_onode(oid, false); | |
10246 | if (!o || !o->exists) { | |
10247 | r = -ENOENT; | |
10248 | goto out; | |
10249 | } | |
10250 | // FIXME minor: actually the next line isn't enough to | |
10251 | // load shared blobs. Leaving as is for now.. | |
10252 | // | |
10253 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10254 | ||
10255 | _dump_onode<0>(cct, *o); | |
10256 | f->open_object_section(section_name.c_str()); | |
10257 | o->dump(f); | |
10258 | f->close_section(); | |
10259 | r = 0; | |
7c673cae | 10260 | } |
9f95a23c TL |
10261 | out: |
10262 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10263 | << " = " << r << dendl; | |
7c673cae FG |
10264 | return r; |
10265 | } | |
10266 | ||
7c673cae FG |
10267 | int BlueStore::getattr( |
10268 | CollectionHandle &c_, | |
10269 | const ghobject_t& oid, | |
10270 | const char *name, | |
10271 | bufferptr& value) | |
10272 | { | |
10273 | Collection *c = static_cast<Collection *>(c_.get()); | |
10274 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
10275 | if (!c->exists) | |
10276 | return -ENOENT; | |
10277 | ||
10278 | int r; | |
10279 | { | |
9f95a23c | 10280 | std::shared_lock l(c->lock); |
31f18b77 | 10281 | mempool::bluestore_cache_other::string k(name); |
7c673cae FG |
10282 | |
10283 | OnodeRef o = c->get_onode(oid, false); | |
10284 | if (!o || !o->exists) { | |
10285 | r = -ENOENT; | |
10286 | goto out; | |
10287 | } | |
10288 | ||
10289 | if (!o->onode.attrs.count(k)) { | |
10290 | r = -ENODATA; | |
10291 | goto out; | |
10292 | } | |
10293 | value = o->onode.attrs[k]; | |
10294 | r = 0; | |
10295 | } | |
10296 | out: | |
7c673cae FG |
10297 | if (r == 0 && _debug_mdata_eio(oid)) { |
10298 | r = -EIO; | |
10299 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10300 | } | |
10301 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
10302 | << " = " << r << dendl; | |
10303 | return r; | |
10304 | } | |
10305 | ||
7c673cae FG |
10306 | int BlueStore::getattrs( |
10307 | CollectionHandle &c_, | |
10308 | const ghobject_t& oid, | |
10309 | map<string,bufferptr>& aset) | |
10310 | { | |
10311 | Collection *c = static_cast<Collection *>(c_.get()); | |
10312 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10313 | if (!c->exists) | |
10314 | return -ENOENT; | |
10315 | ||
10316 | int r; | |
10317 | { | |
9f95a23c | 10318 | std::shared_lock l(c->lock); |
7c673cae FG |
10319 | |
10320 | OnodeRef o = c->get_onode(oid, false); | |
10321 | if (!o || !o->exists) { | |
10322 | r = -ENOENT; | |
10323 | goto out; | |
10324 | } | |
10325 | for (auto& i : o->onode.attrs) { | |
10326 | aset.emplace(i.first.c_str(), i.second); | |
10327 | } | |
10328 | r = 0; | |
10329 | } | |
10330 | ||
10331 | out: | |
7c673cae FG |
10332 | if (r == 0 && _debug_mdata_eio(oid)) { |
10333 | r = -EIO; | |
10334 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10335 | } | |
10336 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10337 | << " = " << r << dendl; | |
10338 | return r; | |
10339 | } | |
10340 | ||
10341 | int BlueStore::list_collections(vector<coll_t>& ls) | |
10342 | { | |
9f95a23c | 10343 | std::shared_lock l(coll_lock); |
11fdf7f2 | 10344 | ls.reserve(coll_map.size()); |
7c673cae FG |
10345 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
10346 | p != coll_map.end(); | |
10347 | ++p) | |
10348 | ls.push_back(p->first); | |
10349 | return 0; | |
10350 | } | |
10351 | ||
10352 | bool BlueStore::collection_exists(const coll_t& c) | |
10353 | { | |
9f95a23c | 10354 | std::shared_lock l(coll_lock); |
7c673cae FG |
10355 | return coll_map.count(c); |
10356 | } | |
10357 | ||
11fdf7f2 | 10358 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 10359 | { |
11fdf7f2 | 10360 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
10361 | vector<ghobject_t> ls; |
10362 | ghobject_t next; | |
11fdf7f2 | 10363 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
10364 | &ls, &next); |
10365 | if (r < 0) { | |
10366 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
10367 | << dendl; | |
10368 | return r; | |
10369 | } | |
10370 | *empty = ls.empty(); | |
11fdf7f2 | 10371 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
10372 | return 0; |
10373 | } | |
10374 | ||
11fdf7f2 | 10375 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 10376 | { |
11fdf7f2 TL |
10377 | dout(15) << __func__ << " " << ch->cid << dendl; |
10378 | Collection *c = static_cast<Collection*>(ch.get()); | |
9f95a23c | 10379 | std::shared_lock l(c->lock); |
11fdf7f2 | 10380 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
10381 | return c->cnode.bits; |
10382 | } | |
10383 | ||
7c673cae FG |
10384 | int BlueStore::collection_list( |
10385 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10386 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10387 | { | |
10388 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 10389 | c->flush(); |
7c673cae FG |
10390 | dout(15) << __func__ << " " << c->cid |
10391 | << " start " << start << " end " << end << " max " << max << dendl; | |
10392 | int r; | |
10393 | { | |
9f95a23c | 10394 | std::shared_lock l(c->lock); |
7c673cae FG |
10395 | r = _collection_list(c, start, end, max, ls, pnext); |
10396 | } | |
10397 | ||
7c673cae FG |
10398 | dout(10) << __func__ << " " << c->cid |
10399 | << " start " << start << " end " << end << " max " << max | |
10400 | << " = " << r << ", ls.size() = " << ls->size() | |
10401 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10402 | return r; | |
10403 | } | |
10404 | ||
10405 | int BlueStore::_collection_list( | |
10406 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
10407 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10408 | { | |
10409 | ||
10410 | if (!c->exists) | |
10411 | return -ENOENT; | |
10412 | ||
494da23a | 10413 | auto start_time = mono_clock::now(); |
7c673cae FG |
10414 | int r = 0; |
10415 | ghobject_t static_next; | |
10416 | KeyValueDB::Iterator it; | |
10417 | string temp_start_key, temp_end_key; | |
10418 | string start_key, end_key; | |
10419 | bool set_next = false; | |
10420 | string pend; | |
10421 | bool temp; | |
10422 | ||
10423 | if (!pnext) | |
10424 | pnext = &static_next; | |
10425 | ||
11fdf7f2 | 10426 | if (start.is_max() || start.hobj.is_max()) { |
7c673cae FG |
10427 | goto out; |
10428 | } | |
10429 | get_coll_key_range(c->cid, c->cnode.bits, &temp_start_key, &temp_end_key, | |
10430 | &start_key, &end_key); | |
10431 | dout(20) << __func__ | |
10432 | << " range " << pretty_binary_string(temp_start_key) | |
10433 | << " to " << pretty_binary_string(temp_end_key) | |
10434 | << " and " << pretty_binary_string(start_key) | |
10435 | << " to " << pretty_binary_string(end_key) | |
10436 | << " start " << start << dendl; | |
10437 | it = db->get_iterator(PREFIX_OBJ); | |
10438 | if (start == ghobject_t() || | |
10439 | start.hobj == hobject_t() || | |
10440 | start == c->cid.get_min_hobj()) { | |
10441 | it->upper_bound(temp_start_key); | |
10442 | temp = true; | |
10443 | } else { | |
10444 | string k; | |
10445 | get_object_key(cct, start, &k); | |
10446 | if (start.hobj.is_temp()) { | |
10447 | temp = true; | |
11fdf7f2 | 10448 | ceph_assert(k >= temp_start_key && k < temp_end_key); |
7c673cae FG |
10449 | } else { |
10450 | temp = false; | |
11fdf7f2 | 10451 | ceph_assert(k >= start_key && k < end_key); |
7c673cae | 10452 | } |
11fdf7f2 | 10453 | dout(20) << __func__ << " start from " << pretty_binary_string(k) |
7c673cae FG |
10454 | << " temp=" << (int)temp << dendl; |
10455 | it->lower_bound(k); | |
10456 | } | |
10457 | if (end.hobj.is_max()) { | |
10458 | pend = temp ? temp_end_key : end_key; | |
10459 | } else { | |
10460 | get_object_key(cct, end, &end_key); | |
10461 | if (end.hobj.is_temp()) { | |
10462 | if (temp) | |
10463 | pend = end_key; | |
10464 | else | |
10465 | goto out; | |
10466 | } else { | |
10467 | pend = temp ? temp_end_key : end_key; | |
10468 | } | |
10469 | } | |
10470 | dout(20) << __func__ << " pend " << pretty_binary_string(pend) << dendl; | |
10471 | while (true) { | |
10472 | if (!it->valid() || it->key() >= pend) { | |
10473 | if (!it->valid()) | |
10474 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
10475 | else | |
10476 | dout(20) << __func__ << " key " << pretty_binary_string(it->key()) | |
10477 | << " >= " << end << dendl; | |
10478 | if (temp) { | |
10479 | if (end.hobj.is_temp()) { | |
10480 | break; | |
10481 | } | |
10482 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
10483 | temp = false; | |
10484 | it->upper_bound(start_key); | |
10485 | pend = end_key; | |
10486 | dout(30) << __func__ << " pend " << pretty_binary_string(pend) << dendl; | |
10487 | continue; | |
10488 | } | |
10489 | break; | |
10490 | } | |
10491 | dout(30) << __func__ << " key " << pretty_binary_string(it->key()) << dendl; | |
10492 | if (is_extent_shard_key(it->key())) { | |
10493 | it->next(); | |
10494 | continue; | |
10495 | } | |
10496 | ghobject_t oid; | |
10497 | int r = get_key_object(it->key(), &oid); | |
11fdf7f2 | 10498 | ceph_assert(r == 0); |
7c673cae FG |
10499 | dout(20) << __func__ << " oid " << oid << " end " << end << dendl; |
10500 | if (ls->size() >= (unsigned)max) { | |
10501 | dout(20) << __func__ << " reached max " << max << dendl; | |
10502 | *pnext = oid; | |
10503 | set_next = true; | |
10504 | break; | |
10505 | } | |
10506 | ls->push_back(oid); | |
10507 | it->next(); | |
10508 | } | |
10509 | out: | |
10510 | if (!set_next) { | |
10511 | *pnext = ghobject_t::get_max(); | |
10512 | } | |
494da23a TL |
10513 | log_latency_fn( |
10514 | __func__, | |
10515 | l_bluestore_clist_lat, | |
10516 | mono_clock::now() - start_time, | |
10517 | cct->_conf->bluestore_log_collection_list_age, | |
10518 | [&] (const ceph::timespan& lat) { | |
10519 | ostringstream ostr; | |
10520 | ostr << ", lat = " << timespan_str(lat) | |
10521 | << " cid =" << c->cid | |
10522 | << " start " << start << " end " << end | |
10523 | << " max " << max; | |
10524 | return ostr.str(); | |
10525 | } | |
10526 | ); | |
7c673cae FG |
10527 | return r; |
10528 | } | |
10529 | ||
7c673cae FG |
10530 | int BlueStore::omap_get( |
10531 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10532 | const ghobject_t &oid, ///< [in] Object containing omap | |
10533 | bufferlist *header, ///< [out] omap header | |
10534 | map<string, bufferlist> *out /// < [out] Key to value map | |
10535 | ) | |
10536 | { | |
10537 | Collection *c = static_cast<Collection *>(c_.get()); | |
9f95a23c TL |
10538 | return _omap_get(c, oid, header, out); |
10539 | } | |
10540 | ||
10541 | int BlueStore::_omap_get( | |
10542 | Collection *c, ///< [in] Collection containing oid | |
10543 | const ghobject_t &oid, ///< [in] Object containing omap | |
10544 | bufferlist *header, ///< [out] omap header | |
10545 | map<string, bufferlist> *out /// < [out] Key to value map | |
10546 | ) | |
10547 | { | |
7c673cae FG |
10548 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; |
10549 | if (!c->exists) | |
10550 | return -ENOENT; | |
9f95a23c | 10551 | std::shared_lock l(c->lock); |
7c673cae FG |
10552 | int r = 0; |
10553 | OnodeRef o = c->get_onode(oid, false); | |
10554 | if (!o || !o->exists) { | |
10555 | r = -ENOENT; | |
10556 | goto out; | |
10557 | } | |
9f95a23c TL |
10558 | r = _onode_omap_get(o, header, out); |
10559 | out: | |
10560 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10561 | << dendl; | |
10562 | return r; | |
10563 | } | |
10564 | ||
10565 | int BlueStore::_onode_omap_get( | |
10566 | const OnodeRef &o, ///< [in] Object containing omap | |
10567 | bufferlist *header, ///< [out] omap header | |
10568 | map<string, bufferlist> *out /// < [out] Key to value map | |
10569 | ) | |
10570 | { | |
10571 | int r = 0; | |
10572 | if (!o || !o->exists) { | |
10573 | r = -ENOENT; | |
10574 | goto out; | |
10575 | } | |
7c673cae FG |
10576 | if (!o->onode.has_omap()) |
10577 | goto out; | |
10578 | o->flush(); | |
10579 | { | |
9f95a23c | 10580 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10581 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10582 | string head, tail; |
9f95a23c TL |
10583 | o->get_omap_header(&head); |
10584 | o->get_omap_tail(&tail); | |
7c673cae FG |
10585 | it->lower_bound(head); |
10586 | while (it->valid()) { | |
10587 | if (it->key() == head) { | |
9f95a23c TL |
10588 | dout(30) << __func__ << " got header" << dendl; |
10589 | *header = it->value(); | |
7c673cae | 10590 | } else if (it->key() >= tail) { |
9f95a23c TL |
10591 | dout(30) << __func__ << " reached tail" << dendl; |
10592 | break; | |
7c673cae | 10593 | } else { |
9f95a23c TL |
10594 | string user_key; |
10595 | o->decode_omap_key(it->key(), &user_key); | |
10596 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
10597 | << " -> " << user_key << dendl; | |
10598 | (*out)[user_key] = it->value(); | |
7c673cae FG |
10599 | } |
10600 | it->next(); | |
10601 | } | |
10602 | } | |
9f95a23c | 10603 | out: |
7c673cae FG |
10604 | return r; |
10605 | } | |
10606 | ||
7c673cae FG |
10607 | int BlueStore::omap_get_header( |
10608 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10609 | const ghobject_t &oid, ///< [in] Object containing omap | |
10610 | bufferlist *header, ///< [out] omap header | |
10611 | bool allow_eio ///< [in] don't assert on eio | |
10612 | ) | |
10613 | { | |
10614 | Collection *c = static_cast<Collection *>(c_.get()); | |
10615 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10616 | if (!c->exists) | |
10617 | return -ENOENT; | |
9f95a23c | 10618 | std::shared_lock l(c->lock); |
7c673cae FG |
10619 | int r = 0; |
10620 | OnodeRef o = c->get_onode(oid, false); | |
10621 | if (!o || !o->exists) { | |
10622 | r = -ENOENT; | |
10623 | goto out; | |
10624 | } | |
10625 | if (!o->onode.has_omap()) | |
10626 | goto out; | |
10627 | o->flush(); | |
10628 | { | |
10629 | string head; | |
9f95a23c TL |
10630 | o->get_omap_header(&head); |
10631 | if (db->get(o->get_omap_prefix(), head, header) >= 0) { | |
7c673cae FG |
10632 | dout(30) << __func__ << " got header" << dendl; |
10633 | } else { | |
10634 | dout(30) << __func__ << " no header" << dendl; | |
10635 | } | |
10636 | } | |
10637 | out: | |
10638 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10639 | << dendl; | |
10640 | return r; | |
10641 | } | |
10642 | ||
7c673cae FG |
10643 | int BlueStore::omap_get_keys( |
10644 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10645 | const ghobject_t &oid, ///< [in] Object containing omap | |
10646 | set<string> *keys ///< [out] Keys defined on oid | |
10647 | ) | |
10648 | { | |
10649 | Collection *c = static_cast<Collection *>(c_.get()); | |
10650 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10651 | if (!c->exists) | |
10652 | return -ENOENT; | |
9f95a23c | 10653 | std::shared_lock l(c->lock); |
7c673cae FG |
10654 | int r = 0; |
10655 | OnodeRef o = c->get_onode(oid, false); | |
10656 | if (!o || !o->exists) { | |
10657 | r = -ENOENT; | |
10658 | goto out; | |
10659 | } | |
10660 | if (!o->onode.has_omap()) | |
10661 | goto out; | |
10662 | o->flush(); | |
10663 | { | |
9f95a23c | 10664 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10665 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10666 | string head, tail; |
9f95a23c TL |
10667 | o->get_omap_key(string(), &head); |
10668 | o->get_omap_tail(&tail); | |
7c673cae FG |
10669 | it->lower_bound(head); |
10670 | while (it->valid()) { | |
10671 | if (it->key() >= tail) { | |
10672 | dout(30) << __func__ << " reached tail" << dendl; | |
10673 | break; | |
10674 | } | |
10675 | string user_key; | |
9f95a23c | 10676 | o->decode_omap_key(it->key(), &user_key); |
11fdf7f2 | 10677 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
10678 | << " -> " << user_key << dendl; |
10679 | keys->insert(user_key); | |
10680 | it->next(); | |
11fdf7f2 TL |
10681 | } |
10682 | } | |
10683 | out: | |
10684 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10685 | << dendl; | |
10686 | return r; | |
7c673cae FG |
10687 | } |
10688 | ||
10689 | int BlueStore::omap_get_values( | |
10690 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10691 | const ghobject_t &oid, ///< [in] Object containing omap | |
10692 | const set<string> &keys, ///< [in] Keys to get | |
10693 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
10694 | ) | |
10695 | { | |
10696 | Collection *c = static_cast<Collection *>(c_.get()); | |
10697 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10698 | if (!c->exists) | |
10699 | return -ENOENT; | |
9f95a23c | 10700 | std::shared_lock l(c->lock); |
7c673cae FG |
10701 | int r = 0; |
10702 | string final_key; | |
10703 | OnodeRef o = c->get_onode(oid, false); | |
10704 | if (!o || !o->exists) { | |
10705 | r = -ENOENT; | |
10706 | goto out; | |
10707 | } | |
9f95a23c | 10708 | if (!o->onode.has_omap()) { |
7c673cae | 10709 | goto out; |
9f95a23c TL |
10710 | } |
10711 | o->flush(); | |
11fdf7f2 | 10712 | { |
9f95a23c TL |
10713 | const string& prefix = o->get_omap_prefix(); |
10714 | o->get_omap_key(string(), &final_key); | |
10715 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10716 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10717 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10718 | final_key += *p; |
10719 | bufferlist val; | |
10720 | if (db->get(prefix, final_key, &val) >= 0) { | |
10721 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
10722 | << " -> " << *p << dendl; | |
10723 | out->insert(make_pair(*p, val)); | |
10724 | } | |
7c673cae FG |
10725 | } |
10726 | } | |
10727 | out: | |
10728 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10729 | << dendl; | |
10730 | return r; | |
10731 | } | |
10732 | ||
9f95a23c TL |
10733 | #ifdef WITH_SEASTAR |
10734 | int BlueStore::omap_get_values( | |
10735 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10736 | const ghobject_t &oid, ///< [in] Object containing omap | |
10737 | const std::optional<string> &start_after, ///< [in] Keys to get | |
10738 | map<string, bufferlist> *output ///< [out] Returned keys and values | |
10739 | ) | |
10740 | { | |
10741 | Collection *c = static_cast<Collection *>(c_.get()); | |
10742 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10743 | if (!c->exists) | |
10744 | return -ENOENT; | |
10745 | std::shared_lock l(c->lock); | |
10746 | int r = 0; | |
10747 | OnodeRef o = c->get_onode(oid, false); | |
10748 | if (!o || !o->exists) { | |
10749 | r = -ENOENT; | |
10750 | goto out; | |
10751 | } | |
10752 | if (!o->onode.has_omap()) { | |
10753 | goto out; | |
10754 | } | |
10755 | o->flush(); | |
10756 | { | |
10757 | ObjectMap::ObjectMapIterator iter = get_omap_iterator(c_, oid); | |
10758 | if (!iter) { | |
10759 | r = -ENOENT; | |
10760 | goto out; | |
10761 | } | |
10762 | iter->upper_bound(*start_after); | |
10763 | for (; iter->valid(); iter->next()) { | |
10764 | output->insert(make_pair(iter->key(), iter->value())); | |
10765 | } | |
10766 | } | |
10767 | ||
10768 | out: | |
10769 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10770 | << dendl; | |
10771 | return r; | |
10772 | } | |
10773 | #endif | |
10774 | ||
7c673cae FG |
10775 | int BlueStore::omap_check_keys( |
10776 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10777 | const ghobject_t &oid, ///< [in] Object containing omap | |
10778 | const set<string> &keys, ///< [in] Keys to check | |
10779 | set<string> *out ///< [out] Subset of keys defined on oid | |
10780 | ) | |
10781 | { | |
10782 | Collection *c = static_cast<Collection *>(c_.get()); | |
10783 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10784 | if (!c->exists) | |
10785 | return -ENOENT; | |
9f95a23c | 10786 | std::shared_lock l(c->lock); |
7c673cae FG |
10787 | int r = 0; |
10788 | string final_key; | |
10789 | OnodeRef o = c->get_onode(oid, false); | |
10790 | if (!o || !o->exists) { | |
10791 | r = -ENOENT; | |
10792 | goto out; | |
10793 | } | |
9f95a23c | 10794 | if (!o->onode.has_omap()) { |
7c673cae | 10795 | goto out; |
9f95a23c TL |
10796 | } |
10797 | o->flush(); | |
11fdf7f2 | 10798 | { |
9f95a23c TL |
10799 | const string& prefix = o->get_omap_prefix(); |
10800 | o->get_omap_key(string(), &final_key); | |
10801 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10802 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10803 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10804 | final_key += *p; |
10805 | bufferlist val; | |
10806 | if (db->get(prefix, final_key, &val) >= 0) { | |
10807 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
10808 | << " -> " << *p << dendl; | |
10809 | out->insert(*p); | |
10810 | } else { | |
10811 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
10812 | << " -> " << *p << dendl; | |
10813 | } | |
7c673cae FG |
10814 | } |
10815 | } | |
10816 | out: | |
10817 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10818 | << dendl; | |
10819 | return r; | |
10820 | } | |
10821 | ||
7c673cae FG |
10822 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
10823 | CollectionHandle &c_, ///< [in] collection | |
10824 | const ghobject_t &oid ///< [in] object | |
10825 | ) | |
10826 | { | |
10827 | Collection *c = static_cast<Collection *>(c_.get()); | |
10828 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
10829 | if (!c->exists) { | |
10830 | return ObjectMap::ObjectMapIterator(); | |
10831 | } | |
9f95a23c | 10832 | std::shared_lock l(c->lock); |
7c673cae FG |
10833 | OnodeRef o = c->get_onode(oid, false); |
10834 | if (!o || !o->exists) { | |
10835 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
10836 | return ObjectMap::ObjectMapIterator(); | |
10837 | } | |
10838 | o->flush(); | |
10839 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
9f95a23c | 10840 | KeyValueDB::Iterator it = db->get_iterator(o->get_omap_prefix()); |
7c673cae FG |
10841 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(c, o, it)); |
10842 | } | |
10843 | ||
10844 | // ----------------- | |
10845 | // write helpers | |
10846 | ||
11fdf7f2 TL |
10847 | uint64_t BlueStore::_get_ondisk_reserved() const { |
10848 | return round_up_to( | |
10849 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
10850 | } | |
10851 | ||
7c673cae FG |
10852 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
10853 | { | |
10854 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
10855 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
10856 | << dendl; | |
11fdf7f2 | 10857 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
10858 | { |
10859 | bufferlist bl; | |
11fdf7f2 | 10860 | encode(ondisk_format, bl); |
7c673cae FG |
10861 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
10862 | } | |
10863 | { | |
10864 | bufferlist bl; | |
11fdf7f2 | 10865 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
10866 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
10867 | } | |
10868 | } | |
10869 | ||
10870 | int BlueStore::_open_super_meta() | |
10871 | { | |
10872 | // nid | |
10873 | { | |
10874 | nid_max = 0; | |
10875 | bufferlist bl; | |
10876 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 10877 | auto p = bl.cbegin(); |
7c673cae FG |
10878 | try { |
10879 | uint64_t v; | |
11fdf7f2 | 10880 | decode(v, p); |
7c673cae FG |
10881 | nid_max = v; |
10882 | } catch (buffer::error& e) { | |
10883 | derr << __func__ << " unable to read nid_max" << dendl; | |
10884 | return -EIO; | |
10885 | } | |
10886 | dout(10) << __func__ << " old nid_max " << nid_max << dendl; | |
10887 | nid_last = nid_max.load(); | |
10888 | } | |
10889 | ||
10890 | // blobid | |
10891 | { | |
10892 | blobid_max = 0; | |
10893 | bufferlist bl; | |
10894 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 10895 | auto p = bl.cbegin(); |
7c673cae FG |
10896 | try { |
10897 | uint64_t v; | |
11fdf7f2 | 10898 | decode(v, p); |
7c673cae FG |
10899 | blobid_max = v; |
10900 | } catch (buffer::error& e) { | |
10901 | derr << __func__ << " unable to read blobid_max" << dendl; | |
10902 | return -EIO; | |
10903 | } | |
10904 | dout(10) << __func__ << " old blobid_max " << blobid_max << dendl; | |
10905 | blobid_last = blobid_max.load(); | |
10906 | } | |
10907 | ||
10908 | // freelist | |
10909 | { | |
10910 | bufferlist bl; | |
10911 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
10912 | if (bl.length()) { | |
10913 | freelist_type = std::string(bl.c_str(), bl.length()); | |
10914 | dout(10) << __func__ << " freelist_type " << freelist_type << dendl; | |
10915 | } else { | |
11fdf7f2 | 10916 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 10917 | } |
7c673cae FG |
10918 | } |
10919 | ||
10920 | // ondisk format | |
10921 | int32_t compat_ondisk_format = 0; | |
10922 | { | |
10923 | bufferlist bl; | |
10924 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
10925 | if (r < 0) { | |
10926 | // base case: kraken bluestore is v1 and readable by v1 | |
10927 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
10928 | << dendl; | |
10929 | ondisk_format = 1; | |
10930 | compat_ondisk_format = 1; | |
10931 | } else { | |
11fdf7f2 | 10932 | auto p = bl.cbegin(); |
7c673cae | 10933 | try { |
11fdf7f2 | 10934 | decode(ondisk_format, p); |
7c673cae FG |
10935 | } catch (buffer::error& e) { |
10936 | derr << __func__ << " unable to read ondisk_format" << dendl; | |
10937 | return -EIO; | |
10938 | } | |
10939 | bl.clear(); | |
10940 | { | |
10941 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
10942 | ceph_assert(!r); |
10943 | auto p = bl.cbegin(); | |
7c673cae | 10944 | try { |
11fdf7f2 | 10945 | decode(compat_ondisk_format, p); |
7c673cae FG |
10946 | } catch (buffer::error& e) { |
10947 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; | |
10948 | return -EIO; | |
10949 | } | |
10950 | } | |
10951 | } | |
10952 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
10953 | << " compat_ondisk_format " << compat_ondisk_format | |
10954 | << dendl; | |
10955 | } | |
10956 | ||
10957 | if (latest_ondisk_format < compat_ondisk_format) { | |
10958 | derr << __func__ << " compat_ondisk_format is " | |
10959 | << compat_ondisk_format << " but we only understand version " | |
10960 | << latest_ondisk_format << dendl; | |
10961 | return -EPERM; | |
10962 | } | |
7c673cae FG |
10963 | |
10964 | { | |
10965 | bufferlist bl; | |
10966 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 10967 | auto p = bl.cbegin(); |
7c673cae FG |
10968 | try { |
10969 | uint64_t val; | |
11fdf7f2 | 10970 | decode(val, p); |
7c673cae | 10971 | min_alloc_size = val; |
224ce89b | 10972 | min_alloc_size_order = ctz(val); |
11fdf7f2 | 10973 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
7c673cae FG |
10974 | } catch (buffer::error& e) { |
10975 | derr << __func__ << " unable to read min_alloc_size" << dendl; | |
10976 | return -EIO; | |
10977 | } | |
10978 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
10979 | << std::dec << dendl; | |
10980 | } | |
9f95a23c TL |
10981 | |
10982 | _set_per_pool_omap(); | |
10983 | ||
224ce89b | 10984 | _open_statfs(); |
7c673cae FG |
10985 | _set_alloc_sizes(); |
10986 | _set_throttle_params(); | |
10987 | ||
10988 | _set_csum(); | |
10989 | _set_compression(); | |
10990 | _set_blob_size(); | |
10991 | ||
11fdf7f2 | 10992 | _validate_bdev(); |
7c673cae FG |
10993 | return 0; |
10994 | } | |
10995 | ||
10996 | int BlueStore::_upgrade_super() | |
10997 | { | |
10998 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
10999 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
11000 | if (ondisk_format < latest_ondisk_format) { |
11001 | ceph_assert(ondisk_format > 0); | |
11002 | ceph_assert(ondisk_format < latest_ondisk_format); | |
11003 | ||
11004 | if (ondisk_format == 1) { | |
11005 | // changes: | |
11006 | // - super: added ondisk_format | |
11007 | // - super: added min_readable_ondisk_format | |
11008 | // - super: added min_compat_ondisk_format | |
11009 | // - super: added min_alloc_size | |
11010 | // - super: removed min_min_alloc_size | |
11011 | KeyValueDB::Transaction t = db->get_transaction(); | |
11012 | { | |
11013 | bufferlist bl; | |
11014 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
11015 | auto p = bl.cbegin(); | |
11016 | try { | |
11017 | uint64_t val; | |
11018 | decode(val, p); | |
11019 | min_alloc_size = val; | |
11020 | } catch (buffer::error& e) { | |
11021 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; | |
11022 | return -EIO; | |
11023 | } | |
11024 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
11025 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 11026 | } |
11fdf7f2 TL |
11027 | ondisk_format = 2; |
11028 | _prepare_ondisk_format_super(t); | |
11029 | int r = db->submit_transaction_sync(t); | |
11030 | ceph_assert(r == 0); | |
7c673cae | 11031 | } |
9f95a23c TL |
11032 | if (ondisk_format == 2) { |
11033 | // changes: | |
11034 | // - onode has FLAG_PER_POOL_OMAP. Note that we do not know that *all* | |
11035 | // ondes are using the per-pool prefix until a repair is run; at that | |
11036 | // point the per_pool_omap=1 key will be set. | |
11037 | // - super: added per_pool_omap key, which indicates that *all* objects | |
11038 | // are using the new prefix and key format | |
11039 | ondisk_format = 3; | |
11040 | KeyValueDB::Transaction t = db->get_transaction(); | |
11041 | _prepare_ondisk_format_super(t); | |
11042 | int r = db->submit_transaction_sync(t); | |
11043 | ceph_assert(r == 0); | |
11044 | } | |
7c673cae | 11045 | } |
7c673cae FG |
11046 | // done |
11047 | dout(1) << __func__ << " done" << dendl; | |
11048 | return 0; | |
11049 | } | |
11050 | ||
11051 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef o) | |
11052 | { | |
224ce89b | 11053 | if (o->onode.nid) { |
11fdf7f2 | 11054 | ceph_assert(o->exists); |
7c673cae | 11055 | return; |
224ce89b | 11056 | } |
7c673cae FG |
11057 | uint64_t nid = ++nid_last; |
11058 | dout(20) << __func__ << " " << nid << dendl; | |
11059 | o->onode.nid = nid; | |
11060 | txc->last_nid = nid; | |
224ce89b | 11061 | o->exists = true; |
7c673cae FG |
11062 | } |
11063 | ||
11064 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
11065 | { | |
11066 | uint64_t bid = ++blobid_last; | |
11067 | dout(20) << __func__ << " " << bid << dendl; | |
11068 | txc->last_blobid = bid; | |
11069 | return bid; | |
11070 | } | |
11071 | ||
11072 | void BlueStore::get_db_statistics(Formatter *f) | |
11073 | { | |
11074 | db->get_statistics(f); | |
11075 | } | |
11076 | ||
11fdf7f2 TL |
11077 | BlueStore::TransContext *BlueStore::_txc_create( |
11078 | Collection *c, OpSequencer *osr, | |
11079 | list<Context*> *on_commits) | |
7c673cae | 11080 | { |
11fdf7f2 | 11081 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae FG |
11082 | txc->t = db->get_transaction(); |
11083 | osr->queue_new(txc); | |
11084 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
11085 | << " seq " << txc->seq << dendl; | |
11086 | return txc; | |
11087 | } | |
11088 | ||
11089 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
11090 | { | |
11fdf7f2 TL |
11091 | // one "io" for the kv commit |
11092 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
11093 | auto cost = throttle_cost_per_io.load(); |
11094 | txc->cost = ios * cost + txc->bytes; | |
9f95a23c | 11095 | txc->ios = ios; |
7c673cae FG |
11096 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" |
11097 | << ios << " ios * " << cost << " + " << txc->bytes | |
11098 | << " bytes)" << dendl; | |
11099 | } | |
11100 | ||
11101 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
11102 | { | |
11103 | if (txc->statfs_delta.is_empty()) | |
11104 | return; | |
11105 | ||
11106 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
11107 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
11108 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
11109 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
11110 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
11111 | ||
11112 | bufferlist bl; | |
11113 | txc->statfs_delta.encode(bl); | |
11fdf7f2 TL |
11114 | if (per_pool_stat_collection) { |
11115 | string key; | |
11116 | get_pool_stat_key(txc->osd_pool_id, &key); | |
11117 | txc->t->merge(PREFIX_STAT, key, bl); | |
11118 | ||
11119 | std::lock_guard l(vstatfs_lock); | |
11120 | auto& stats = osd_pools[txc->osd_pool_id]; | |
11121 | stats += txc->statfs_delta; | |
11122 | ||
11123 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
11124 | ||
11125 | } else { | |
11126 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
7c673cae | 11127 | |
11fdf7f2 TL |
11128 | std::lock_guard l(vstatfs_lock); |
11129 | vstatfs += txc->statfs_delta; | |
11130 | } | |
7c673cae FG |
11131 | txc->statfs_delta.reset(); |
11132 | } | |
11133 | ||
11134 | void BlueStore::_txc_state_proc(TransContext *txc) | |
11135 | { | |
11136 | while (true) { | |
11137 | dout(10) << __func__ << " txc " << txc | |
11138 | << " " << txc->get_state_name() << dendl; | |
11139 | switch (txc->state) { | |
11140 | case TransContext::STATE_PREPARE: | |
9f95a23c | 11141 | throttle.log_state_latency(*txc, logger, l_bluestore_state_prepare_lat); |
7c673cae FG |
11142 | if (txc->ioc.has_pending_aios()) { |
11143 | txc->state = TransContext::STATE_AIO_WAIT; | |
11144 | txc->had_ios = true; | |
11145 | _txc_aio_submit(txc); | |
11146 | return; | |
11147 | } | |
11148 | // ** fall-thru ** | |
11149 | ||
11150 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 | 11151 | { |
9f95a23c TL |
11152 | mono_clock::duration lat = throttle.log_state_latency( |
11153 | *txc, logger, l_bluestore_state_aio_wait_lat); | |
11154 | if (ceph::to_seconds<double>(lat) >= cct->_conf->bluestore_log_op_age) { | |
11fdf7f2 TL |
11155 | dout(0) << __func__ << " slow aio_wait, txc = " << txc |
11156 | << ", latency = " << lat | |
11157 | << dendl; | |
11158 | } | |
11159 | } | |
11160 | ||
7c673cae FG |
11161 | _txc_finish_io(txc); // may trigger blocked txc's too |
11162 | return; | |
11163 | ||
11164 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 11165 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
11166 | if (txc->had_ios) { |
11167 | ++txc->osr->txc_with_unstable_io; | |
11168 | } | |
9f95a23c | 11169 | throttle.log_state_latency(*txc, logger, l_bluestore_state_io_done_lat); |
7c673cae FG |
11170 | txc->state = TransContext::STATE_KV_QUEUED; |
11171 | if (cct->_conf->bluestore_sync_submit_transaction) { | |
11172 | if (txc->last_nid >= nid_max || | |
11173 | txc->last_blobid >= blobid_max) { | |
11174 | dout(20) << __func__ | |
11175 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
11176 | << dendl; | |
11177 | } else if (txc->osr->kv_committing_serially) { | |
11178 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
11179 | << dendl; | |
11180 | // note: this is starvation-prone. once we have a txc in a busy | |
11181 | // sequencer that is committing serially it is possible to keep | |
11182 | // submitting new transactions fast enough that we get stuck doing | |
11183 | // so. the alternative is to block here... fixme? | |
11184 | } else if (txc->osr->txc_with_unstable_io) { | |
11185 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
11186 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
11187 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
11188 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
11189 | == 0) { | |
11190 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
11191 | << dendl; | |
11192 | } else { | |
9f95a23c | 11193 | _txc_apply_kv(txc, true); |
7c673cae FG |
11194 | } |
11195 | } | |
11196 | { | |
11fdf7f2 | 11197 | std::lock_guard l(kv_lock); |
7c673cae | 11198 | kv_queue.push_back(txc); |
9f95a23c TL |
11199 | if (!kv_sync_in_progress) { |
11200 | kv_sync_in_progress = true; | |
11201 | kv_cond.notify_one(); | |
11202 | } | |
7c673cae FG |
11203 | if (txc->state != TransContext::STATE_KV_SUBMITTED) { |
11204 | kv_queue_unsubmitted.push_back(txc); | |
11205 | ++txc->osr->kv_committing_serially; | |
11206 | } | |
31f18b77 FG |
11207 | if (txc->had_ios) |
11208 | kv_ios++; | |
11209 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
11210 | } |
11211 | return; | |
11212 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
11213 | _txc_committed_kv(txc); |
11214 | // ** fall-thru ** | |
11215 | ||
11216 | case TransContext::STATE_KV_DONE: | |
9f95a23c | 11217 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_done_lat); |
7c673cae FG |
11218 | if (txc->deferred_txn) { |
11219 | txc->state = TransContext::STATE_DEFERRED_QUEUED; | |
11220 | _deferred_queue(txc); | |
11221 | return; | |
11222 | } | |
11223 | txc->state = TransContext::STATE_FINISHING; | |
11224 | break; | |
11225 | ||
11226 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9f95a23c | 11227 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_cleanup_lat); |
7c673cae FG |
11228 | txc->state = TransContext::STATE_FINISHING; |
11229 | // ** fall-thru ** | |
11230 | ||
11231 | case TransContext::STATE_FINISHING: | |
9f95a23c | 11232 | throttle.log_state_latency(*txc, logger, l_bluestore_state_finishing_lat); |
7c673cae FG |
11233 | _txc_finish(txc); |
11234 | return; | |
11235 | ||
11236 | default: | |
11237 | derr << __func__ << " unexpected txc " << txc | |
11238 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 11239 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
11240 | return; |
11241 | } | |
11242 | } | |
11243 | } | |
11244 | ||
11245 | void BlueStore::_txc_finish_io(TransContext *txc) | |
11246 | { | |
11247 | dout(20) << __func__ << " " << txc << dendl; | |
11248 | ||
11249 | /* | |
11250 | * we need to preserve the order of kv transactions, | |
11251 | * even though aio will complete in any order. | |
11252 | */ | |
11253 | ||
11254 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 11255 | std::lock_guard l(osr->qlock); |
7c673cae | 11256 | txc->state = TransContext::STATE_IO_DONE; |
11fdf7f2 | 11257 | txc->ioc.release_running_aios(); |
7c673cae FG |
11258 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
11259 | while (p != osr->q.begin()) { | |
11260 | --p; | |
11261 | if (p->state < TransContext::STATE_IO_DONE) { | |
11262 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " | |
11263 | << p->get_state_name() << dendl; | |
11264 | return; | |
11265 | } | |
11266 | if (p->state > TransContext::STATE_IO_DONE) { | |
11267 | ++p; | |
11268 | break; | |
11269 | } | |
11270 | } | |
11271 | do { | |
11272 | _txc_state_proc(&*p++); | |
11273 | } while (p != osr->q.end() && | |
11274 | p->state == TransContext::STATE_IO_DONE); | |
11275 | ||
11fdf7f2 | 11276 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
11277 | osr->qcond.notify_all(); |
11278 | } | |
11279 | } | |
11280 | ||
11281 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
11282 | { | |
11283 | dout(20) << __func__ << " txc " << txc | |
11284 | << " onodes " << txc->onodes | |
11285 | << " shared_blobs " << txc->shared_blobs | |
11286 | << dendl; | |
11287 | ||
11288 | // finalize onodes | |
11289 | for (auto o : txc->onodes) { | |
11fdf7f2 | 11290 | _record_onode(o, t); |
7c673cae FG |
11291 | o->flushing_count++; |
11292 | } | |
11293 | ||
11294 | // objects we modified but didn't affect the onode | |
11295 | auto p = txc->modified_objects.begin(); | |
11296 | while (p != txc->modified_objects.end()) { | |
11297 | if (txc->onodes.count(*p) == 0) { | |
11298 | (*p)->flushing_count++; | |
11299 | ++p; | |
11300 | } else { | |
11301 | // remove dups with onodes list to avoid problems in _txc_finish | |
11302 | p = txc->modified_objects.erase(p); | |
11303 | } | |
11304 | } | |
11305 | ||
11306 | // finalize shared_blobs | |
11307 | for (auto sb : txc->shared_blobs) { | |
11308 | string key; | |
11309 | auto sbid = sb->get_sbid(); | |
11310 | get_shared_blob_key(sbid, &key); | |
11311 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
11312 | dout(20) << __func__ << " shared_blob 0x" |
11313 | << std::hex << sbid << std::dec | |
7c673cae FG |
11314 | << " is empty" << dendl; |
11315 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
11316 | } else { | |
11317 | bufferlist bl; | |
11fdf7f2 TL |
11318 | encode(*(sb->persistent), bl); |
11319 | dout(20) << __func__ << " shared_blob 0x" | |
11320 | << std::hex << sbid << std::dec | |
31f18b77 | 11321 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
11322 | t->set(PREFIX_SHARED_BLOB, key, bl); |
11323 | } | |
11324 | } | |
11325 | } | |
11326 | ||
11327 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
11328 | PerfCounters &logger) | |
11329 | { | |
11fdf7f2 TL |
11330 | os_commit_latency_ns.consume_next( |
11331 | logger.get_tavg_ns( | |
7c673cae | 11332 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
11333 | os_apply_latency_ns.consume_next( |
11334 | logger.get_tavg_ns( | |
7c673cae FG |
11335 | l_bluestore_commit_lat)); |
11336 | } | |
11337 | ||
11338 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) | |
11339 | { | |
11340 | dout(20) << __func__ << " txc " << txc << std::hex | |
11341 | << " allocated 0x" << txc->allocated | |
11342 | << " released 0x" << txc->released | |
11343 | << std::dec << dendl; | |
11344 | ||
11345 | // We have to handle the case where we allocate *and* deallocate the | |
11346 | // same region in this transaction. The freelist doesn't like that. | |
11347 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
11348 | // debug check. But that's important.) | |
11349 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
11350 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
11351 | interval_set<uint64_t> *preleased = &txc->released; | |
11352 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
11353 | interval_set<uint64_t> overlap; | |
11354 | overlap.intersection_of(txc->allocated, txc->released); | |
11355 | if (!overlap.empty()) { | |
11356 | tmp_allocated = txc->allocated; | |
11357 | tmp_allocated.subtract(overlap); | |
11358 | tmp_released = txc->released; | |
11359 | tmp_released.subtract(overlap); | |
11360 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
11361 | << ", new allocated 0x" << tmp_allocated | |
11362 | << " released 0x" << tmp_released << std::dec | |
11363 | << dendl; | |
11364 | pallocated = &tmp_allocated; | |
11365 | preleased = &tmp_released; | |
11366 | } | |
11367 | } | |
11368 | ||
11369 | // update freelist with non-overlap sets | |
11370 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
11371 | p != pallocated->end(); | |
11372 | ++p) { | |
11373 | fm->allocate(p.get_start(), p.get_len(), t); | |
11374 | } | |
11375 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
11376 | p != preleased->end(); | |
11377 | ++p) { | |
11378 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
11379 | << "~" << p.get_len() << std::dec << dendl; | |
11380 | fm->release(p.get_start(), p.get_len(), t); | |
11381 | } | |
11382 | ||
11383 | _txc_update_store_statfs(txc); | |
11384 | } | |
11385 | ||
9f95a23c | 11386 | void BlueStore::_txc_apply_kv(TransContext *txc, bool sync_submit_transaction) |
7c673cae | 11387 | { |
9f95a23c TL |
11388 | ceph_assert(txc->state == TransContext::STATE_KV_QUEUED); |
11389 | { | |
11390 | #if defined(WITH_LTTNG) | |
11391 | auto start = mono_clock::now(); | |
11392 | #endif | |
11393 | ||
11394 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); | |
11395 | ceph_assert(r == 0); | |
11396 | txc->state = TransContext::STATE_KV_SUBMITTED; | |
11397 | if (txc->osr->kv_submitted_waiters) { | |
11398 | std::lock_guard l(txc->osr->qlock); | |
11399 | txc->osr->qcond.notify_all(); | |
11400 | } | |
11401 | ||
11402 | #if defined(WITH_LTTNG) | |
11403 | if (txc->tracing) { | |
11404 | tracepoint( | |
11405 | bluestore, | |
11406 | transaction_kv_submit_latency, | |
11407 | txc->osr->get_sequencer_id(), | |
11408 | txc->seq, | |
11409 | sync_submit_transaction, | |
11410 | ceph::to_seconds<double>(mono_clock::now() - start)); | |
11411 | } | |
11412 | #endif | |
11413 | } | |
11414 | ||
7c673cae FG |
11415 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { |
11416 | for (auto& o : *ls) { | |
11417 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
11418 | << dendl; | |
9f95a23c | 11419 | if (--o->flushing_count == 0 && o->waiting_count.load()) { |
11fdf7f2 | 11420 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
11421 | o->flush_cond.notify_all(); |
11422 | } | |
11423 | } | |
11424 | } | |
11425 | } | |
11426 | ||
11427 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
11428 | { | |
11429 | dout(20) << __func__ << " txc " << txc << dendl; | |
9f95a23c | 11430 | throttle.complete_kv(*txc); |
1adf2230 | 11431 | { |
11fdf7f2 | 11432 | std::lock_guard l(txc->osr->qlock); |
1adf2230 | 11433 | txc->state = TransContext::STATE_KV_DONE; |
11fdf7f2 TL |
11434 | if (txc->ch->commit_queue) { |
11435 | txc->ch->commit_queue->queue(txc->oncommits); | |
11436 | } else { | |
11437 | finisher.queue(txc->oncommits); | |
1adf2230 | 11438 | } |
7c673cae | 11439 | } |
9f95a23c | 11440 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_committing_lat); |
494da23a TL |
11441 | log_latency_fn( |
11442 | __func__, | |
11443 | l_bluestore_commit_lat, | |
9f95a23c | 11444 | mono_clock::now() - txc->start, |
494da23a TL |
11445 | cct->_conf->bluestore_log_op_age, |
11446 | [&](auto lat) { | |
11447 | return ", txc = " + stringify(txc); | |
11448 | } | |
11fdf7f2 | 11449 | ); |
7c673cae FG |
11450 | } |
11451 | ||
11452 | void BlueStore::_txc_finish(TransContext *txc) | |
11453 | { | |
11454 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
11fdf7f2 | 11455 | ceph_assert(txc->state == TransContext::STATE_FINISHING); |
7c673cae FG |
11456 | |
11457 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 11458 | sb->finish_write(txc->seq); |
7c673cae FG |
11459 | } |
11460 | txc->shared_blobs_written.clear(); | |
11461 | ||
11462 | while (!txc->removed_collections.empty()) { | |
11463 | _queue_reap_collection(txc->removed_collections.front()); | |
11464 | txc->removed_collections.pop_front(); | |
11465 | } | |
11466 | ||
11467 | OpSequencerRef osr = txc->osr; | |
7c673cae | 11468 | bool empty = false; |
31f18b77 | 11469 | bool submit_deferred = false; |
7c673cae FG |
11470 | OpSequencer::q_list_t releasing_txc; |
11471 | { | |
11fdf7f2 | 11472 | std::lock_guard l(osr->qlock); |
7c673cae FG |
11473 | txc->state = TransContext::STATE_DONE; |
11474 | bool notify = false; | |
11475 | while (!osr->q.empty()) { | |
11476 | TransContext *txc = &osr->q.front(); | |
11477 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
11478 | << dendl; | |
11479 | if (txc->state != TransContext::STATE_DONE) { | |
11480 | if (txc->state == TransContext::STATE_PREPARE && | |
11481 | deferred_aggressive) { | |
11482 | // for _osr_drain_preceding() | |
11483 | notify = true; | |
11484 | } | |
31f18b77 | 11485 | if (txc->state == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 11486 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
11487 | submit_deferred = true; |
11488 | } | |
7c673cae FG |
11489 | break; |
11490 | } | |
11491 | ||
7c673cae FG |
11492 | osr->q.pop_front(); |
11493 | releasing_txc.push_back(*txc); | |
7c673cae | 11494 | } |
9f95a23c | 11495 | |
7c673cae FG |
11496 | if (osr->q.empty()) { |
11497 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
11498 | empty = true; | |
11499 | } | |
9f95a23c TL |
11500 | |
11501 | // only drain()/drain_preceding() need wakeup, | |
11502 | // other cases use kv_submitted_waiters | |
11503 | if (notify || empty) { | |
11504 | osr->qcond.notify_all(); | |
11505 | } | |
7c673cae | 11506 | } |
9f95a23c | 11507 | |
7c673cae FG |
11508 | while (!releasing_txc.empty()) { |
11509 | // release to allocator only after all preceding txc's have also | |
11510 | // finished any deferred writes that potentially land in these | |
11511 | // blocks | |
11512 | auto txc = &releasing_txc.front(); | |
11513 | _txc_release_alloc(txc); | |
11514 | releasing_txc.pop_front(); | |
9f95a23c TL |
11515 | throttle.log_state_latency(*txc, logger, l_bluestore_state_done_lat); |
11516 | throttle.complete(*txc); | |
7c673cae FG |
11517 | delete txc; |
11518 | } | |
11519 | ||
31f18b77 FG |
11520 | if (submit_deferred) { |
11521 | // we're pinning memory; flush! we could be more fine-grained here but | |
11522 | // i'm not sure it's worth the bother. | |
11523 | deferred_try_submit(); | |
7c673cae FG |
11524 | } |
11525 | ||
7c673cae | 11526 | if (empty && osr->zombie) { |
11fdf7f2 TL |
11527 | std::lock_guard l(zombie_osr_lock); |
11528 | if (zombie_osr_set.erase(osr->cid)) { | |
11529 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11530 | } else { | |
11531 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
11532 | << dendl; | |
11533 | } | |
7c673cae | 11534 | } |
9f95a23c | 11535 | } |
7c673cae FG |
11536 | |
11537 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
11538 | { | |
a8e16298 | 11539 | // it's expected we're called with lazy_release_lock already taken! |
11fdf7f2 TL |
11540 | if (likely(!cct->_conf->bluestore_debug_no_reuse_blocks)) { |
11541 | int r = 0; | |
11542 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
11543 | r = bdev->queue_discard(txc->released); | |
11544 | if (r == 0) { | |
11545 | dout(10) << __func__ << "(queued) " << txc << " " << std::hex | |
11546 | << txc->released << std::dec << dendl; | |
11547 | goto out; | |
11548 | } | |
11549 | } else if (cct->_conf->bdev_enable_discard) { | |
11550 | for (auto p = txc->released.begin(); p != txc->released.end(); ++p) { | |
11551 | bdev->discard(p.get_start(), p.get_len()); | |
11552 | } | |
11553 | } | |
11554 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
94b18763 | 11555 | << txc->released << std::dec << dendl; |
11fdf7f2 | 11556 | alloc->release(txc->released); |
7c673cae FG |
11557 | } |
11558 | ||
11fdf7f2 | 11559 | out: |
7c673cae FG |
11560 | txc->allocated.clear(); |
11561 | txc->released.clear(); | |
11562 | } | |
11563 | ||
11fdf7f2 TL |
11564 | void BlueStore::_osr_attach(Collection *c) |
11565 | { | |
11566 | // note: caller has RWLock on coll_map | |
11567 | auto q = coll_map.find(c->cid); | |
11568 | if (q != coll_map.end()) { | |
11569 | c->osr = q->second->osr; | |
11570 | ldout(cct, 10) << __func__ << " " << c->cid | |
11571 | << " reusing osr " << c->osr << " from existing coll " | |
11572 | << q->second << dendl; | |
11573 | } else { | |
11574 | std::lock_guard l(zombie_osr_lock); | |
11575 | auto p = zombie_osr_set.find(c->cid); | |
11576 | if (p == zombie_osr_set.end()) { | |
9f95a23c | 11577 | c->osr = ceph::make_ref<OpSequencer>(this, next_sequencer_id++, c->cid); |
11fdf7f2 TL |
11578 | ldout(cct, 10) << __func__ << " " << c->cid |
11579 | << " fresh osr " << c->osr << dendl; | |
11580 | } else { | |
11581 | c->osr = p->second; | |
11582 | zombie_osr_set.erase(p); | |
11583 | ldout(cct, 10) << __func__ << " " << c->cid | |
11584 | << " resurrecting zombie osr " << c->osr << dendl; | |
11585 | c->osr->zombie = false; | |
11586 | } | |
11587 | } | |
11588 | } | |
11589 | ||
11590 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
11591 | { | |
11592 | std::lock_guard l(zombie_osr_lock); | |
11593 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
11594 | osr->zombie = true; | |
11595 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
11596 | // this is either a new insertion or the same osr is already there | |
11597 | ceph_assert(i.second || i.first->second == osr); | |
11598 | } | |
11599 | ||
7c673cae FG |
11600 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
11601 | { | |
11602 | OpSequencer *osr = txc->osr.get(); | |
11603 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
11604 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11605 | { | |
11606 | // submit anything pending | |
224ce89b | 11607 | deferred_lock.lock(); |
11fdf7f2 | 11608 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
11609 | _deferred_submit_unlock(osr); |
11610 | } else { | |
11611 | deferred_lock.unlock(); | |
7c673cae FG |
11612 | } |
11613 | } | |
11614 | { | |
11615 | // wake up any previously finished deferred events | |
11fdf7f2 | 11616 | std::lock_guard l(kv_lock); |
9f95a23c TL |
11617 | if (!kv_sync_in_progress) { |
11618 | kv_sync_in_progress = true; | |
11619 | kv_cond.notify_one(); | |
11620 | } | |
7c673cae FG |
11621 | } |
11622 | osr->drain_preceding(txc); | |
11623 | --deferred_aggressive; | |
11624 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11625 | } | |
11626 | ||
11fdf7f2 TL |
11627 | void BlueStore::_osr_drain(OpSequencer *osr) |
11628 | { | |
11629 | dout(10) << __func__ << " " << osr << dendl; | |
11630 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11631 | { | |
11632 | // submit anything pending | |
11633 | deferred_lock.lock(); | |
11634 | if (osr->deferred_pending && !osr->deferred_running) { | |
11635 | _deferred_submit_unlock(osr); | |
11636 | } else { | |
11637 | deferred_lock.unlock(); | |
11638 | } | |
11639 | } | |
11640 | { | |
11641 | // wake up any previously finished deferred events | |
11642 | std::lock_guard l(kv_lock); | |
9f95a23c TL |
11643 | if (!kv_sync_in_progress) { |
11644 | kv_sync_in_progress = true; | |
11645 | kv_cond.notify_one(); | |
11646 | } | |
11fdf7f2 TL |
11647 | } |
11648 | osr->drain(); | |
11649 | --deferred_aggressive; | |
11650 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11651 | } | |
11652 | ||
7c673cae FG |
11653 | void BlueStore::_osr_drain_all() |
11654 | { | |
11655 | dout(10) << __func__ << dendl; | |
11656 | ||
11657 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
11658 | vector<OpSequencerRef> zombies; |
11659 | { | |
9f95a23c | 11660 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
11661 | for (auto& i : coll_map) { |
11662 | s.insert(i.second->osr); | |
11663 | } | |
11664 | } | |
7c673cae | 11665 | { |
11fdf7f2 TL |
11666 | std::lock_guard l(zombie_osr_lock); |
11667 | for (auto& i : zombie_osr_set) { | |
11668 | s.insert(i.second); | |
11669 | zombies.push_back(i.second); | |
11670 | } | |
7c673cae FG |
11671 | } |
11672 | dout(20) << __func__ << " osr_set " << s << dendl; | |
11673 | ||
11674 | ++deferred_aggressive; | |
11675 | { | |
11676 | // submit anything pending | |
224ce89b | 11677 | deferred_try_submit(); |
7c673cae FG |
11678 | } |
11679 | { | |
11680 | // wake up any previously finished deferred events | |
11fdf7f2 | 11681 | std::lock_guard l(kv_lock); |
7c673cae FG |
11682 | kv_cond.notify_one(); |
11683 | } | |
31f18b77 | 11684 | { |
11fdf7f2 | 11685 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11686 | kv_finalize_cond.notify_one(); |
11687 | } | |
7c673cae FG |
11688 | for (auto osr : s) { |
11689 | dout(20) << __func__ << " drain " << osr << dendl; | |
11690 | osr->drain(); | |
11691 | } | |
11692 | --deferred_aggressive; | |
11693 | ||
7c673cae | 11694 | { |
11fdf7f2 TL |
11695 | std::lock_guard l(zombie_osr_lock); |
11696 | for (auto& osr : zombies) { | |
11697 | if (zombie_osr_set.erase(osr->cid)) { | |
11698 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11699 | ceph_assert(osr->q.empty()); | |
11700 | } else if (osr->zombie) { | |
11701 | dout(10) << __func__ << " empty zombie osr " << osr | |
11702 | << " already reaped" << dendl; | |
11703 | ceph_assert(osr->q.empty()); | |
11704 | } else { | |
11705 | dout(10) << __func__ << " empty zombie osr " << osr | |
11706 | << " resurrected" << dendl; | |
11707 | } | |
7c673cae FG |
11708 | } |
11709 | } | |
11fdf7f2 TL |
11710 | |
11711 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
11712 | } |
11713 | ||
11fdf7f2 | 11714 | |
31f18b77 FG |
11715 | void BlueStore::_kv_start() |
11716 | { | |
11717 | dout(10) << __func__ << dendl; | |
11718 | ||
11fdf7f2 | 11719 | finisher.start(); |
31f18b77 FG |
11720 | kv_sync_thread.create("bstore_kv_sync"); |
11721 | kv_finalize_thread.create("bstore_kv_final"); | |
11722 | } | |
11723 | ||
11724 | void BlueStore::_kv_stop() | |
11725 | { | |
11726 | dout(10) << __func__ << dendl; | |
11727 | { | |
9f95a23c | 11728 | std::unique_lock l{kv_lock}; |
31f18b77 FG |
11729 | while (!kv_sync_started) { |
11730 | kv_cond.wait(l); | |
11731 | } | |
11732 | kv_stop = true; | |
11733 | kv_cond.notify_all(); | |
11734 | } | |
11735 | { | |
9f95a23c | 11736 | std::unique_lock l{kv_finalize_lock}; |
31f18b77 FG |
11737 | while (!kv_finalize_started) { |
11738 | kv_finalize_cond.wait(l); | |
11739 | } | |
11740 | kv_finalize_stop = true; | |
11741 | kv_finalize_cond.notify_all(); | |
11742 | } | |
11743 | kv_sync_thread.join(); | |
11744 | kv_finalize_thread.join(); | |
11fdf7f2 | 11745 | ceph_assert(removed_collections.empty()); |
31f18b77 | 11746 | { |
11fdf7f2 | 11747 | std::lock_guard l(kv_lock); |
31f18b77 FG |
11748 | kv_stop = false; |
11749 | } | |
11750 | { | |
11fdf7f2 | 11751 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11752 | kv_finalize_stop = false; |
11753 | } | |
11754 | dout(10) << __func__ << " stopping finishers" << dendl; | |
11fdf7f2 TL |
11755 | finisher.wait_for_empty(); |
11756 | finisher.stop(); | |
31f18b77 FG |
11757 | dout(10) << __func__ << " stopped" << dendl; |
11758 | } | |
11759 | ||
7c673cae FG |
11760 | void BlueStore::_kv_sync_thread() |
11761 | { | |
11762 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 | 11763 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
9f95a23c | 11764 | std::unique_lock l{kv_lock}; |
11fdf7f2 | 11765 | ceph_assert(!kv_sync_started); |
31f18b77 FG |
11766 | kv_sync_started = true; |
11767 | kv_cond.notify_all(); | |
7c673cae | 11768 | while (true) { |
11fdf7f2 | 11769 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
11770 | if (kv_queue.empty() && |
11771 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 11772 | !deferred_aggressive)) { |
7c673cae FG |
11773 | if (kv_stop) |
11774 | break; | |
11775 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 11776 | kv_sync_in_progress = false; |
11fdf7f2 | 11777 | kv_cond.wait(l); |
7c673cae FG |
11778 | dout(20) << __func__ << " wake" << dendl; |
11779 | } else { | |
11780 | deque<TransContext*> kv_submitting; | |
11781 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
11782 | uint64_t aios = 0, costs = 0; |
11783 | ||
7c673cae FG |
11784 | dout(20) << __func__ << " committing " << kv_queue.size() |
11785 | << " submitting " << kv_queue_unsubmitted.size() | |
11786 | << " deferred done " << deferred_done_queue.size() | |
11787 | << " stable " << deferred_stable_queue.size() | |
11788 | << dendl; | |
11789 | kv_committing.swap(kv_queue); | |
11790 | kv_submitting.swap(kv_queue_unsubmitted); | |
11791 | deferred_done.swap(deferred_done_queue); | |
11792 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
11793 | aios = kv_ios; |
11794 | costs = kv_throttle_costs; | |
11795 | kv_ios = 0; | |
11796 | kv_throttle_costs = 0; | |
7c673cae FG |
11797 | l.unlock(); |
11798 | ||
11799 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
11800 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
11801 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
11802 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
11803 | ||
11fdf7f2 TL |
11804 | auto start = mono_clock::now(); |
11805 | ||
7c673cae FG |
11806 | bool force_flush = false; |
11807 | // if bluefs is sharing the same device as data (only), then we | |
11808 | // can rely on the bluefs commit to flush the device and make | |
11809 | // deferred aios stable. that means that if we do have done deferred | |
11810 | // txcs AND we are not on a single device, we need to force a flush. | |
9f95a23c | 11811 | if (bluefs && bluefs_layout.single_shared_device()) { |
31f18b77 | 11812 | if (aios) { |
7c673cae | 11813 | force_flush = true; |
11fdf7f2 | 11814 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
11815 | force_flush = true; // there's nothing else to commit! |
11816 | } else if (deferred_aggressive) { | |
11817 | force_flush = true; | |
11818 | } | |
11fdf7f2 TL |
11819 | } else { |
11820 | if (aios || !deferred_done.empty()) { | |
11821 | force_flush = true; | |
11822 | } else { | |
11823 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
11824 | } | |
11825 | } | |
7c673cae FG |
11826 | |
11827 | if (force_flush) { | |
31f18b77 | 11828 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
11829 | << " force_flush=" << (int)force_flush |
11830 | << ", flushing, deferred done->stable" << dendl; | |
11831 | // flush/barrier on block device | |
11832 | bdev->flush(); | |
11833 | ||
11834 | // if we flush then deferred done are now deferred stable | |
11835 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
11836 | deferred_done.end()); | |
11837 | deferred_done.clear(); | |
11838 | } | |
11fdf7f2 | 11839 | auto after_flush = mono_clock::now(); |
7c673cae FG |
11840 | |
11841 | // we will use one final transaction to force a sync | |
11842 | KeyValueDB::Transaction synct = db->get_transaction(); | |
11843 | ||
11844 | // increase {nid,blobid}_max? note that this covers both the | |
11845 | // case where we are approaching the max and the case we passed | |
11846 | // it. in either case, we increase the max in the earlier txn | |
11847 | // we submit. | |
11848 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
11849 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
11850 | KeyValueDB::Transaction t = | |
11851 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
11852 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
11853 | bufferlist bl; | |
11fdf7f2 | 11854 | encode(new_nid_max, bl); |
7c673cae FG |
11855 | t->set(PREFIX_SUPER, "nid_max", bl); |
11856 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
11857 | } | |
11858 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
11859 | KeyValueDB::Transaction t = | |
11860 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
11861 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
11862 | bufferlist bl; | |
11fdf7f2 | 11863 | encode(new_blobid_max, bl); |
7c673cae FG |
11864 | t->set(PREFIX_SUPER, "blobid_max", bl); |
11865 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
11866 | } | |
c07f9fc5 FG |
11867 | |
11868 | for (auto txc : kv_committing) { | |
9f95a23c | 11869 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_queued_lat); |
c07f9fc5 | 11870 | if (txc->state == TransContext::STATE_KV_QUEUED) { |
9f95a23c | 11871 | _txc_apply_kv(txc, false); |
c07f9fc5 | 11872 | --txc->osr->kv_committing_serially; |
c07f9fc5 | 11873 | } else { |
11fdf7f2 | 11874 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 11875 | } |
7c673cae FG |
11876 | if (txc->had_ios) { |
11877 | --txc->osr->txc_with_unstable_io; | |
11878 | } | |
7c673cae FG |
11879 | } |
11880 | ||
31f18b77 FG |
11881 | // release throttle *before* we commit. this allows new ops |
11882 | // to be prepared and enter pipeline while we are waiting on | |
11883 | // the kv commit sync/flush. then hopefully on the next | |
11884 | // iteration there will already be ops awake. otherwise, we | |
11885 | // end up going to sleep, and then wake up when the very first | |
11886 | // transaction is ready for commit. | |
9f95a23c | 11887 | throttle.release_kv_throttle(costs); |
31f18b77 | 11888 | |
7c673cae FG |
11889 | if (bluefs && |
11890 | after_flush - bluefs_last_balance > | |
11fdf7f2 | 11891 | ceph::make_timespan(cct->_conf->bluestore_bluefs_balance_interval)) { |
7c673cae | 11892 | bluefs_last_balance = after_flush; |
11fdf7f2 TL |
11893 | int r = _balance_bluefs_freespace(); |
11894 | ceph_assert(r >= 0); | |
7c673cae FG |
11895 | } |
11896 | ||
11897 | // cleanup sync deferred keys | |
11898 | for (auto b : deferred_stable) { | |
11899 | for (auto& txc : b->txcs) { | |
11900 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 11901 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
11902 | string key; |
11903 | get_deferred_key(wt.seq, &key); | |
11904 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
11905 | } | |
11906 | } | |
11907 | ||
9f95a23c TL |
11908 | #if defined(WITH_LTTNG) |
11909 | auto sync_start = mono_clock::now(); | |
11910 | #endif | |
7c673cae | 11911 | // submit synct synchronously (block and wait for it to commit) |
31f18b77 | 11912 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
11913 | ceph_assert(r == 0); |
11914 | ||
9f95a23c TL |
11915 | int committing_size = kv_committing.size(); |
11916 | int deferred_size = deferred_stable.size(); | |
11917 | ||
11918 | #if defined(WITH_LTTNG) | |
11919 | double sync_latency = ceph::to_seconds<double>(mono_clock::now() - sync_start); | |
11920 | for (auto txc: kv_committing) { | |
11921 | if (txc->tracing) { | |
11922 | tracepoint( | |
11923 | bluestore, | |
11924 | transaction_kv_sync_latency, | |
11925 | txc->osr->get_sequencer_id(), | |
11926 | txc->seq, | |
11927 | kv_committing.size(), | |
11928 | deferred_done.size(), | |
11929 | deferred_stable.size(), | |
11930 | sync_latency); | |
11931 | } | |
11932 | } | |
11933 | #endif | |
11934 | ||
11fdf7f2 | 11935 | { |
9f95a23c | 11936 | std::unique_lock m{kv_finalize_lock}; |
11fdf7f2 TL |
11937 | if (kv_committing_to_finalize.empty()) { |
11938 | kv_committing_to_finalize.swap(kv_committing); | |
11939 | } else { | |
11940 | kv_committing_to_finalize.insert( | |
11941 | kv_committing_to_finalize.end(), | |
11942 | kv_committing.begin(), | |
11943 | kv_committing.end()); | |
11944 | kv_committing.clear(); | |
11945 | } | |
11946 | if (deferred_stable_to_finalize.empty()) { | |
11947 | deferred_stable_to_finalize.swap(deferred_stable); | |
11948 | } else { | |
11949 | deferred_stable_to_finalize.insert( | |
11950 | deferred_stable_to_finalize.end(), | |
11951 | deferred_stable.begin(), | |
11952 | deferred_stable.end()); | |
11953 | deferred_stable.clear(); | |
11954 | } | |
9f95a23c TL |
11955 | if (!kv_finalize_in_progress) { |
11956 | kv_finalize_in_progress = true; | |
11957 | kv_finalize_cond.notify_one(); | |
11958 | } | |
11fdf7f2 | 11959 | } |
7c673cae FG |
11960 | |
11961 | if (new_nid_max) { | |
11962 | nid_max = new_nid_max; | |
11963 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
11964 | } | |
11965 | if (new_blobid_max) { | |
11966 | blobid_max = new_blobid_max; | |
11967 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
11968 | } | |
11969 | ||
224ce89b | 11970 | { |
11fdf7f2 TL |
11971 | auto finish = mono_clock::now(); |
11972 | ceph::timespan dur_flush = after_flush - start; | |
11973 | ceph::timespan dur_kv = finish - after_flush; | |
11974 | ceph::timespan dur = finish - start; | |
9f95a23c TL |
11975 | dout(20) << __func__ << " committed " << committing_size |
11976 | << " cleaned " << deferred_size | |
224ce89b WB |
11977 | << " in " << dur |
11978 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
11979 | << dendl; | |
494da23a TL |
11980 | log_latency("kv_flush", |
11981 | l_bluestore_kv_flush_lat, | |
11982 | dur_flush, | |
11983 | cct->_conf->bluestore_log_op_age); | |
11984 | log_latency("kv_commit", | |
11985 | l_bluestore_kv_commit_lat, | |
11986 | dur_kv, | |
11987 | cct->_conf->bluestore_log_op_age); | |
11988 | log_latency("kv_sync", | |
11989 | l_bluestore_kv_sync_lat, | |
11990 | dur, | |
11991 | cct->_conf->bluestore_log_op_age); | |
7c673cae | 11992 | } |
31f18b77 FG |
11993 | |
11994 | if (bluefs) { | |
11fdf7f2 TL |
11995 | if (!bluefs_extents_reclaiming.empty()) { |
11996 | dout(0) << __func__ << " releasing old bluefs 0x" << std::hex | |
11997 | << bluefs_extents_reclaiming << std::dec << dendl; | |
81eedcae TL |
11998 | int r = 0; |
11999 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
12000 | r = bdev->queue_discard(bluefs_extents_reclaiming); | |
12001 | if (r == 0) { | |
12002 | goto clear; | |
12003 | } | |
12004 | } else if (cct->_conf->bdev_enable_discard) { | |
12005 | for (auto p = bluefs_extents_reclaiming.begin(); p != bluefs_extents_reclaiming.end(); ++p) { | |
12006 | bdev->discard(p.get_start(), p.get_len()); | |
12007 | } | |
12008 | } | |
12009 | ||
11fdf7f2 | 12010 | alloc->release(bluefs_extents_reclaiming); |
81eedcae | 12011 | clear: |
11fdf7f2 | 12012 | bluefs_extents_reclaiming.clear(); |
31f18b77 | 12013 | } |
31f18b77 FG |
12014 | } |
12015 | ||
12016 | l.lock(); | |
12017 | // previously deferred "done" are now "stable" by virtue of this | |
12018 | // commit cycle. | |
12019 | deferred_stable_queue.swap(deferred_done); | |
12020 | } | |
12021 | } | |
12022 | dout(10) << __func__ << " finish" << dendl; | |
12023 | kv_sync_started = false; | |
12024 | } | |
12025 | ||
12026 | void BlueStore::_kv_finalize_thread() | |
12027 | { | |
12028 | deque<TransContext*> kv_committed; | |
12029 | deque<DeferredBatch*> deferred_stable; | |
12030 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
12031 | std::unique_lock l(kv_finalize_lock); |
12032 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
12033 | kv_finalize_started = true; |
12034 | kv_finalize_cond.notify_all(); | |
12035 | while (true) { | |
11fdf7f2 TL |
12036 | ceph_assert(kv_committed.empty()); |
12037 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
12038 | if (kv_committing_to_finalize.empty() && |
12039 | deferred_stable_to_finalize.empty()) { | |
12040 | if (kv_finalize_stop) | |
12041 | break; | |
12042 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 12043 | kv_finalize_in_progress = false; |
31f18b77 FG |
12044 | kv_finalize_cond.wait(l); |
12045 | dout(20) << __func__ << " wake" << dendl; | |
12046 | } else { | |
12047 | kv_committed.swap(kv_committing_to_finalize); | |
12048 | deferred_stable.swap(deferred_stable_to_finalize); | |
12049 | l.unlock(); | |
12050 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
12051 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12052 | ||
11fdf7f2 TL |
12053 | auto start = mono_clock::now(); |
12054 | ||
31f18b77 FG |
12055 | while (!kv_committed.empty()) { |
12056 | TransContext *txc = kv_committed.front(); | |
11fdf7f2 | 12057 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12058 | _txc_state_proc(txc); |
31f18b77 | 12059 | kv_committed.pop_front(); |
7c673cae | 12060 | } |
31f18b77 | 12061 | |
7c673cae FG |
12062 | for (auto b : deferred_stable) { |
12063 | auto p = b->txcs.begin(); | |
12064 | while (p != b->txcs.end()) { | |
12065 | TransContext *txc = &*p; | |
12066 | p = b->txcs.erase(p); // unlink here because | |
12067 | _txc_state_proc(txc); // this may destroy txc | |
12068 | } | |
12069 | delete b; | |
12070 | } | |
31f18b77 | 12071 | deferred_stable.clear(); |
7c673cae FG |
12072 | |
12073 | if (!deferred_aggressive) { | |
31f18b77 | 12074 | if (deferred_queue_size >= deferred_batch_ops.load() || |
9f95a23c | 12075 | throttle.should_submit_deferred()) { |
224ce89b | 12076 | deferred_try_submit(); |
7c673cae FG |
12077 | } |
12078 | } | |
12079 | ||
12080 | // this is as good a place as any ... | |
12081 | _reap_collections(); | |
12082 | ||
11fdf7f2 | 12083 | logger->set(l_bluestore_fragmentation, |
9f95a23c | 12084 | (uint64_t)(alloc->get_fragmentation() * 1000)); |
11fdf7f2 | 12085 | |
494da23a TL |
12086 | log_latency("kv_final", |
12087 | l_bluestore_kv_final_lat, | |
12088 | mono_clock::now() - start, | |
12089 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 | 12090 | |
7c673cae | 12091 | l.lock(); |
7c673cae FG |
12092 | } |
12093 | } | |
12094 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 12095 | kv_finalize_started = false; |
7c673cae FG |
12096 | } |
12097 | ||
12098 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( | |
9f95a23c | 12099 | TransContext *txc) |
7c673cae FG |
12100 | { |
12101 | if (!txc->deferred_txn) { | |
12102 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
12103 | } | |
12104 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
12105 | return &txc->deferred_txn->ops.back(); | |
12106 | } | |
12107 | ||
12108 | void BlueStore::_deferred_queue(TransContext *txc) | |
12109 | { | |
12110 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
224ce89b | 12111 | deferred_lock.lock(); |
7c673cae FG |
12112 | if (!txc->osr->deferred_pending && |
12113 | !txc->osr->deferred_running) { | |
12114 | deferred_queue.push_back(*txc->osr); | |
12115 | } | |
12116 | if (!txc->osr->deferred_pending) { | |
12117 | txc->osr->deferred_pending = new DeferredBatch(cct, txc->osr.get()); | |
12118 | } | |
12119 | ++deferred_queue_size; | |
12120 | txc->osr->deferred_pending->txcs.push_back(*txc); | |
12121 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; | |
12122 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
12123 | const auto& op = *opi; | |
11fdf7f2 | 12124 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
12125 | bufferlist::const_iterator p = op.data.begin(); |
12126 | for (auto e : op.extents) { | |
12127 | txc->osr->deferred_pending->prepare_write( | |
12128 | cct, wt.seq, e.offset, e.length, p); | |
12129 | } | |
12130 | } | |
12131 | if (deferred_aggressive && | |
12132 | !txc->osr->deferred_running) { | |
224ce89b WB |
12133 | _deferred_submit_unlock(txc->osr.get()); |
12134 | } else { | |
12135 | deferred_lock.unlock(); | |
7c673cae FG |
12136 | } |
12137 | } | |
12138 | ||
224ce89b | 12139 | void BlueStore::deferred_try_submit() |
7c673cae FG |
12140 | { |
12141 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
12142 | << deferred_queue_size << " txcs" << dendl; | |
11fdf7f2 | 12143 | std::lock_guard l(deferred_lock); |
224ce89b WB |
12144 | vector<OpSequencerRef> osrs; |
12145 | osrs.reserve(deferred_queue.size()); | |
7c673cae | 12146 | for (auto& osr : deferred_queue) { |
224ce89b WB |
12147 | osrs.push_back(&osr); |
12148 | } | |
12149 | for (auto& osr : osrs) { | |
181888fb FG |
12150 | if (osr->deferred_pending) { |
12151 | if (!osr->deferred_running) { | |
12152 | _deferred_submit_unlock(osr.get()); | |
12153 | deferred_lock.lock(); | |
12154 | } else { | |
12155 | dout(20) << __func__ << " osr " << osr << " already has running" | |
12156 | << dendl; | |
12157 | } | |
12158 | } else { | |
12159 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; | |
7c673cae FG |
12160 | } |
12161 | } | |
9f95a23c TL |
12162 | |
12163 | deferred_last_submitted = ceph_clock_now(); | |
7c673cae FG |
12164 | } |
12165 | ||
224ce89b | 12166 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
12167 | { |
12168 | dout(10) << __func__ << " osr " << osr | |
12169 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
12170 | << dendl; | |
11fdf7f2 TL |
12171 | ceph_assert(osr->deferred_pending); |
12172 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
12173 | |
12174 | auto b = osr->deferred_pending; | |
12175 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 12176 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
12177 | |
12178 | osr->deferred_running = osr->deferred_pending; | |
12179 | osr->deferred_pending = nullptr; | |
12180 | ||
11fdf7f2 TL |
12181 | deferred_lock.unlock(); |
12182 | ||
12183 | for (auto& txc : b->txcs) { | |
9f95a23c | 12184 | throttle.log_state_latency(txc, logger, l_bluestore_state_deferred_queued_lat); |
11fdf7f2 | 12185 | } |
7c673cae FG |
12186 | uint64_t start = 0, pos = 0; |
12187 | bufferlist bl; | |
12188 | auto i = b->iomap.begin(); | |
12189 | while (true) { | |
12190 | if (i == b->iomap.end() || i->first != pos) { | |
12191 | if (bl.length()) { | |
12192 | dout(20) << __func__ << " write 0x" << std::hex | |
12193 | << start << "~" << bl.length() | |
12194 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 12195 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
12196 | logger->inc(l_bluestore_deferred_write_ops); |
12197 | logger->inc(l_bluestore_deferred_write_bytes, bl.length()); | |
12198 | int r = bdev->aio_write(start, bl, &b->ioc, false); | |
11fdf7f2 | 12199 | ceph_assert(r == 0); |
7c673cae FG |
12200 | } |
12201 | } | |
12202 | if (i == b->iomap.end()) { | |
12203 | break; | |
12204 | } | |
12205 | start = 0; | |
12206 | pos = i->first; | |
12207 | bl.clear(); | |
12208 | } | |
12209 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
12210 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
12211 | << dendl; | |
12212 | if (!bl.length()) { | |
12213 | start = pos; | |
12214 | } | |
12215 | pos += i->second.bl.length(); | |
12216 | bl.claim_append(i->second.bl); | |
12217 | ++i; | |
12218 | } | |
224ce89b | 12219 | |
7c673cae FG |
12220 | bdev->aio_submit(&b->ioc); |
12221 | } | |
12222 | ||
3efd9988 FG |
12223 | struct C_DeferredTrySubmit : public Context { |
12224 | BlueStore *store; | |
12225 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
12226 | void finish(int r) { | |
12227 | store->deferred_try_submit(); | |
12228 | } | |
12229 | }; | |
12230 | ||
7c673cae FG |
12231 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
12232 | { | |
12233 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 12234 | ceph_assert(osr->deferred_running); |
7c673cae FG |
12235 | DeferredBatch *b = osr->deferred_running; |
12236 | ||
12237 | { | |
9f95a23c | 12238 | deferred_lock.lock(); |
11fdf7f2 | 12239 | ceph_assert(osr->deferred_running == b); |
7c673cae FG |
12240 | osr->deferred_running = nullptr; |
12241 | if (!osr->deferred_pending) { | |
181888fb | 12242 | dout(20) << __func__ << " dequeueing" << dendl; |
7c673cae FG |
12243 | auto q = deferred_queue.iterator_to(*osr); |
12244 | deferred_queue.erase(q); | |
9f95a23c | 12245 | deferred_lock.unlock(); |
181888fb | 12246 | } else { |
9f95a23c TL |
12247 | deferred_lock.unlock(); |
12248 | if (deferred_aggressive) { | |
12249 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; | |
12250 | finisher.queue(new C_DeferredTrySubmit(this)); | |
12251 | } else { | |
12252 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
12253 | } | |
7c673cae FG |
12254 | } |
12255 | } | |
12256 | ||
12257 | { | |
31f18b77 | 12258 | uint64_t costs = 0; |
11fdf7f2 | 12259 | { |
11fdf7f2 TL |
12260 | for (auto& i : b->txcs) { |
12261 | TransContext *txc = &i; | |
9f95a23c | 12262 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_aio_wait_lat); |
11fdf7f2 TL |
12263 | txc->state = TransContext::STATE_DEFERRED_CLEANUP; |
12264 | costs += txc->cost; | |
12265 | } | |
7c673cae | 12266 | } |
9f95a23c | 12267 | throttle.release_deferred_throttle(costs); |
7c673cae FG |
12268 | } |
12269 | ||
9f95a23c | 12270 | { |
11fdf7f2 | 12271 | std::lock_guard l(kv_lock); |
9f95a23c TL |
12272 | deferred_done_queue.emplace_back(b); |
12273 | ||
12274 | // in the normal case, do not bother waking up the kv thread; it will | |
12275 | // catch us on the next commit anyway. | |
12276 | if (deferred_aggressive && !kv_sync_in_progress) { | |
12277 | kv_sync_in_progress = true; | |
12278 | kv_cond.notify_one(); | |
12279 | } | |
7c673cae FG |
12280 | } |
12281 | } | |
12282 | ||
12283 | int BlueStore::_deferred_replay() | |
12284 | { | |
12285 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
12286 | int count = 0; |
12287 | int r = 0; | |
11fdf7f2 TL |
12288 | CollectionRef ch = _get_collection(coll_t::meta()); |
12289 | bool fake_ch = false; | |
12290 | if (!ch) { | |
12291 | // hmm, replaying initial mkfs? | |
12292 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
12293 | fake_ch = true; | |
12294 | } | |
12295 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
12296 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
12297 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
12298 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
12299 | << dendl; | |
12300 | bluestore_deferred_transaction_t *deferred_txn = | |
12301 | new bluestore_deferred_transaction_t; | |
12302 | bufferlist bl = it->value(); | |
11fdf7f2 | 12303 | auto p = bl.cbegin(); |
7c673cae | 12304 | try { |
11fdf7f2 | 12305 | decode(*deferred_txn, p); |
7c673cae FG |
12306 | } catch (buffer::error& e) { |
12307 | derr << __func__ << " failed to decode deferred txn " | |
12308 | << pretty_binary_string(it->key()) << dendl; | |
12309 | delete deferred_txn; | |
12310 | r = -EIO; | |
12311 | goto out; | |
12312 | } | |
11fdf7f2 | 12313 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); |
7c673cae FG |
12314 | txc->deferred_txn = deferred_txn; |
12315 | txc->state = TransContext::STATE_KV_DONE; | |
12316 | _txc_state_proc(txc); | |
12317 | } | |
12318 | out: | |
12319 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 12320 | _osr_register_zombie(osr); |
7c673cae | 12321 | _osr_drain_all(); |
11fdf7f2 TL |
12322 | if (fake_ch) { |
12323 | new_coll_map.clear(); | |
12324 | } | |
7c673cae FG |
12325 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
12326 | return r; | |
12327 | } | |
12328 | ||
12329 | // --------------------------- | |
12330 | // transactions | |
12331 | ||
12332 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
12333 | CollectionHandle& ch, |
12334 | vector<Transaction>& tls, | |
12335 | TrackedOpRef op, | |
12336 | ThreadPool::TPHandle *handle) | |
12337 | { | |
12338 | FUNCTRACE(cct); | |
12339 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 12340 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 12341 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae | 12342 | |
11fdf7f2 TL |
12343 | auto start = mono_clock::now(); |
12344 | ||
12345 | Collection *c = static_cast<Collection*>(ch.get()); | |
12346 | OpSequencer *osr = c->osr.get(); | |
12347 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae FG |
12348 | |
12349 | // prepare | |
11fdf7f2 TL |
12350 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, |
12351 | &on_commit); | |
7c673cae FG |
12352 | |
12353 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { | |
7c673cae FG |
12354 | txc->bytes += (*p).get_num_bytes(); |
12355 | _txc_add_transaction(txc, &(*p)); | |
12356 | } | |
12357 | _txc_calc_cost(txc); | |
12358 | ||
12359 | _txc_write_nodes(txc, txc->t); | |
12360 | ||
12361 | // journal deferred items | |
12362 | if (txc->deferred_txn) { | |
12363 | txc->deferred_txn->seq = ++deferred_seq; | |
12364 | bufferlist bl; | |
11fdf7f2 | 12365 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
12366 | string key; |
12367 | get_deferred_key(txc->deferred_txn->seq, &key); | |
12368 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
12369 | } | |
12370 | ||
12371 | _txc_finalize_kv(txc, txc->t); | |
12372 | if (handle) | |
12373 | handle->suspend_tp_timeout(); | |
12374 | ||
11fdf7f2 | 12375 | auto tstart = mono_clock::now(); |
9f95a23c TL |
12376 | |
12377 | if (!throttle.try_start_transaction( | |
12378 | *db, | |
12379 | *txc, | |
12380 | tstart)) { | |
7c673cae | 12381 | // ensure we do not block here because of deferred writes |
9f95a23c TL |
12382 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
12383 | << dendl; | |
12384 | ++deferred_aggressive; | |
12385 | deferred_try_submit(); | |
12386 | { | |
12387 | // wake up any previously finished deferred events | |
12388 | std::lock_guard l(kv_lock); | |
12389 | if (!kv_sync_in_progress) { | |
12390 | kv_sync_in_progress = true; | |
3efd9988 FG |
12391 | kv_cond.notify_one(); |
12392 | } | |
9f95a23c TL |
12393 | } |
12394 | throttle.finish_start_transaction(*db, *txc, tstart); | |
12395 | --deferred_aggressive; | |
7c673cae | 12396 | } |
11fdf7f2 | 12397 | auto tend = mono_clock::now(); |
7c673cae FG |
12398 | |
12399 | if (handle) | |
12400 | handle->reset_tp_timeout(); | |
12401 | ||
12402 | logger->inc(l_bluestore_txc); | |
12403 | ||
12404 | // execute (start) | |
12405 | _txc_state_proc(txc); | |
12406 | ||
11fdf7f2 TL |
12407 | // we're immediately readable (unlike FileStore) |
12408 | for (auto c : on_applied_sync) { | |
12409 | c->complete(0); | |
12410 | } | |
12411 | if (!on_applied.empty()) { | |
12412 | if (c->commit_queue) { | |
12413 | c->commit_queue->queue(on_applied); | |
12414 | } else { | |
12415 | finisher.queue(on_applied); | |
12416 | } | |
12417 | } | |
12418 | ||
494da23a TL |
12419 | log_latency("submit_transact", |
12420 | l_bluestore_submit_lat, | |
12421 | mono_clock::now() - start, | |
12422 | cct->_conf->bluestore_log_op_age); | |
12423 | log_latency("throttle_transact", | |
12424 | l_bluestore_throttle_lat, | |
12425 | tend - tstart, | |
12426 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
12427 | return 0; |
12428 | } | |
12429 | ||
12430 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
12431 | { | |
12432 | dout(10) << __func__ << " txc " << txc << dendl; | |
12433 | bdev->aio_submit(&txc->ioc); | |
12434 | } | |
12435 | ||
12436 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
12437 | { | |
12438 | Transaction::iterator i = t->begin(); | |
12439 | ||
81eedcae | 12440 | _dump_transaction<30>(cct, t); |
7c673cae FG |
12441 | |
12442 | vector<CollectionRef> cvec(i.colls.size()); | |
12443 | unsigned j = 0; | |
12444 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
12445 | ++p, ++j) { | |
12446 | cvec[j] = _get_collection(*p); | |
7c673cae | 12447 | } |
11fdf7f2 | 12448 | |
7c673cae FG |
12449 | vector<OnodeRef> ovec(i.objects.size()); |
12450 | ||
12451 | for (int pos = 0; i.have_op(); ++pos) { | |
12452 | Transaction::Op *op = i.decode_op(); | |
12453 | int r = 0; | |
12454 | ||
12455 | // no coll or obj | |
12456 | if (op->op == Transaction::OP_NOP) | |
12457 | continue; | |
12458 | ||
11fdf7f2 | 12459 | |
7c673cae FG |
12460 | // collection operations |
12461 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
12462 | |
12463 | // initialize osd_pool_id and do a smoke test that all collections belong | |
12464 | // to the same pool | |
12465 | spg_t pgid; | |
12466 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
12467 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
12468 | txc->osd_pool_id == pgid.pool()); | |
12469 | txc->osd_pool_id = pgid.pool(); | |
12470 | } | |
12471 | ||
7c673cae FG |
12472 | switch (op->op) { |
12473 | case Transaction::OP_RMCOLL: | |
12474 | { | |
12475 | const coll_t &cid = i.get_cid(op->cid); | |
12476 | r = _remove_collection(txc, cid, &c); | |
12477 | if (!r) | |
12478 | continue; | |
12479 | } | |
12480 | break; | |
12481 | ||
12482 | case Transaction::OP_MKCOLL: | |
12483 | { | |
11fdf7f2 | 12484 | ceph_assert(!c); |
7c673cae FG |
12485 | const coll_t &cid = i.get_cid(op->cid); |
12486 | r = _create_collection(txc, cid, op->split_bits, &c); | |
12487 | if (!r) | |
12488 | continue; | |
12489 | } | |
12490 | break; | |
12491 | ||
12492 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 12493 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12494 | break; |
12495 | ||
12496 | case Transaction::OP_SPLIT_COLLECTION2: | |
12497 | { | |
12498 | uint32_t bits = op->split_bits; | |
12499 | uint32_t rem = op->split_rem; | |
12500 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
12501 | if (!r) | |
12502 | continue; | |
12503 | } | |
12504 | break; | |
12505 | ||
11fdf7f2 TL |
12506 | case Transaction::OP_MERGE_COLLECTION: |
12507 | { | |
12508 | uint32_t bits = op->split_bits; | |
12509 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
12510 | if (!r) | |
12511 | continue; | |
12512 | } | |
12513 | break; | |
12514 | ||
7c673cae FG |
12515 | case Transaction::OP_COLL_HINT: |
12516 | { | |
12517 | uint32_t type = op->hint_type; | |
12518 | bufferlist hint; | |
12519 | i.decode_bl(hint); | |
11fdf7f2 | 12520 | auto hiter = hint.cbegin(); |
7c673cae FG |
12521 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
12522 | uint32_t pg_num; | |
12523 | uint64_t num_objs; | |
11fdf7f2 TL |
12524 | decode(pg_num, hiter); |
12525 | decode(num_objs, hiter); | |
7c673cae FG |
12526 | dout(10) << __func__ << " collection hint objects is a no-op, " |
12527 | << " pg_num " << pg_num << " num_objects " << num_objs | |
12528 | << dendl; | |
12529 | } else { | |
12530 | // Ignore the hint | |
12531 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
12532 | } | |
12533 | continue; | |
12534 | } | |
12535 | break; | |
12536 | ||
12537 | case Transaction::OP_COLL_SETATTR: | |
12538 | r = -EOPNOTSUPP; | |
12539 | break; | |
12540 | ||
12541 | case Transaction::OP_COLL_RMATTR: | |
12542 | r = -EOPNOTSUPP; | |
12543 | break; | |
12544 | ||
12545 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 12546 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12547 | break; |
12548 | } | |
12549 | if (r < 0) { | |
12550 | derr << __func__ << " error " << cpp_strerror(r) | |
12551 | << " not handled on operation " << op->op | |
12552 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 12553 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12554 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12555 | } |
12556 | ||
12557 | // these operations implicity create the object | |
12558 | bool create = false; | |
12559 | if (op->op == Transaction::OP_TOUCH || | |
9f95a23c | 12560 | op->op == Transaction::OP_CREATE || |
7c673cae FG |
12561 | op->op == Transaction::OP_WRITE || |
12562 | op->op == Transaction::OP_ZERO) { | |
12563 | create = true; | |
12564 | } | |
12565 | ||
12566 | // object operations | |
9f95a23c | 12567 | std::unique_lock l(c->lock); |
7c673cae FG |
12568 | OnodeRef &o = ovec[op->oid]; |
12569 | if (!o) { | |
12570 | ghobject_t oid = i.get_oid(op->oid); | |
9f95a23c | 12571 | o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE); |
7c673cae FG |
12572 | } |
12573 | if (!create && (!o || !o->exists)) { | |
12574 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
12575 | << i.get_oid(op->oid) << dendl; | |
12576 | r = -ENOENT; | |
12577 | goto endop; | |
12578 | } | |
12579 | ||
12580 | switch (op->op) { | |
9f95a23c | 12581 | case Transaction::OP_CREATE: |
7c673cae FG |
12582 | case Transaction::OP_TOUCH: |
12583 | r = _touch(txc, c, o); | |
12584 | break; | |
12585 | ||
12586 | case Transaction::OP_WRITE: | |
12587 | { | |
12588 | uint64_t off = op->off; | |
12589 | uint64_t len = op->len; | |
12590 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
12591 | bufferlist bl; | |
12592 | i.decode_bl(bl); | |
12593 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
12594 | } | |
12595 | break; | |
12596 | ||
12597 | case Transaction::OP_ZERO: | |
12598 | { | |
12599 | uint64_t off = op->off; | |
12600 | uint64_t len = op->len; | |
12601 | r = _zero(txc, c, o, off, len); | |
12602 | } | |
12603 | break; | |
12604 | ||
12605 | case Transaction::OP_TRIMCACHE: | |
12606 | { | |
12607 | // deprecated, no-op | |
12608 | } | |
12609 | break; | |
12610 | ||
12611 | case Transaction::OP_TRUNCATE: | |
12612 | { | |
12613 | uint64_t off = op->off; | |
35e4c445 | 12614 | r = _truncate(txc, c, o, off); |
7c673cae FG |
12615 | } |
12616 | break; | |
12617 | ||
12618 | case Transaction::OP_REMOVE: | |
12619 | { | |
12620 | r = _remove(txc, c, o); | |
12621 | } | |
12622 | break; | |
12623 | ||
12624 | case Transaction::OP_SETATTR: | |
12625 | { | |
12626 | string name = i.decode_string(); | |
12627 | bufferptr bp; | |
12628 | i.decode_bp(bp); | |
12629 | r = _setattr(txc, c, o, name, bp); | |
12630 | } | |
12631 | break; | |
12632 | ||
12633 | case Transaction::OP_SETATTRS: | |
12634 | { | |
12635 | map<string, bufferptr> aset; | |
12636 | i.decode_attrset(aset); | |
12637 | r = _setattrs(txc, c, o, aset); | |
12638 | } | |
12639 | break; | |
12640 | ||
12641 | case Transaction::OP_RMATTR: | |
12642 | { | |
12643 | string name = i.decode_string(); | |
12644 | r = _rmattr(txc, c, o, name); | |
12645 | } | |
12646 | break; | |
12647 | ||
12648 | case Transaction::OP_RMATTRS: | |
12649 | { | |
12650 | r = _rmattrs(txc, c, o); | |
12651 | } | |
12652 | break; | |
12653 | ||
12654 | case Transaction::OP_CLONE: | |
12655 | { | |
12656 | OnodeRef& no = ovec[op->dest_oid]; | |
12657 | if (!no) { | |
12658 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12659 | no = c->get_onode(noid, true); | |
12660 | } | |
12661 | r = _clone(txc, c, o, no); | |
12662 | } | |
12663 | break; | |
12664 | ||
12665 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 12666 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12667 | break; |
12668 | ||
12669 | case Transaction::OP_CLONERANGE2: | |
12670 | { | |
12671 | OnodeRef& no = ovec[op->dest_oid]; | |
12672 | if (!no) { | |
12673 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12674 | no = c->get_onode(noid, true); | |
12675 | } | |
12676 | uint64_t srcoff = op->off; | |
12677 | uint64_t len = op->len; | |
12678 | uint64_t dstoff = op->dest_off; | |
12679 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
12680 | } | |
12681 | break; | |
12682 | ||
12683 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 12684 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12685 | break; |
12686 | ||
12687 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 12688 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12689 | break; |
12690 | ||
12691 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 12692 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12693 | break; |
12694 | ||
12695 | case Transaction::OP_COLL_MOVE_RENAME: | |
12696 | case Transaction::OP_TRY_RENAME: | |
12697 | { | |
11fdf7f2 | 12698 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
12699 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
12700 | OnodeRef& no = ovec[op->dest_oid]; | |
12701 | if (!no) { | |
12702 | no = c->get_onode(noid, false); | |
12703 | } | |
12704 | r = _rename(txc, c, o, no, noid); | |
12705 | } | |
12706 | break; | |
12707 | ||
12708 | case Transaction::OP_OMAP_CLEAR: | |
12709 | { | |
12710 | r = _omap_clear(txc, c, o); | |
12711 | } | |
12712 | break; | |
12713 | case Transaction::OP_OMAP_SETKEYS: | |
12714 | { | |
12715 | bufferlist aset_bl; | |
12716 | i.decode_attrset_bl(&aset_bl); | |
12717 | r = _omap_setkeys(txc, c, o, aset_bl); | |
12718 | } | |
12719 | break; | |
12720 | case Transaction::OP_OMAP_RMKEYS: | |
12721 | { | |
12722 | bufferlist keys_bl; | |
12723 | i.decode_keyset_bl(&keys_bl); | |
12724 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
12725 | } | |
12726 | break; | |
12727 | case Transaction::OP_OMAP_RMKEYRANGE: | |
12728 | { | |
12729 | string first, last; | |
12730 | first = i.decode_string(); | |
12731 | last = i.decode_string(); | |
12732 | r = _omap_rmkey_range(txc, c, o, first, last); | |
12733 | } | |
12734 | break; | |
12735 | case Transaction::OP_OMAP_SETHEADER: | |
12736 | { | |
12737 | bufferlist bl; | |
12738 | i.decode_bl(bl); | |
12739 | r = _omap_setheader(txc, c, o, bl); | |
12740 | } | |
12741 | break; | |
12742 | ||
12743 | case Transaction::OP_SETALLOCHINT: | |
12744 | { | |
12745 | r = _set_alloc_hint(txc, c, o, | |
12746 | op->expected_object_size, | |
12747 | op->expected_write_size, | |
12748 | op->alloc_hint_flags); | |
12749 | } | |
12750 | break; | |
12751 | ||
12752 | default: | |
11fdf7f2 | 12753 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
12754 | ceph_abort(); |
12755 | } | |
12756 | ||
12757 | endop: | |
12758 | if (r < 0) { | |
12759 | bool ok = false; | |
12760 | ||
12761 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
12762 | op->op == Transaction::OP_CLONE || | |
12763 | op->op == Transaction::OP_CLONERANGE2 || | |
12764 | op->op == Transaction::OP_COLL_ADD || | |
12765 | op->op == Transaction::OP_SETATTR || | |
12766 | op->op == Transaction::OP_SETATTRS || | |
12767 | op->op == Transaction::OP_RMATTR || | |
12768 | op->op == Transaction::OP_OMAP_SETKEYS || | |
12769 | op->op == Transaction::OP_OMAP_RMKEYS || | |
12770 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
12771 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
12772 | // -ENOENT is usually okay | |
12773 | ok = true; | |
12774 | if (r == -ENODATA) | |
12775 | ok = true; | |
12776 | ||
12777 | if (!ok) { | |
12778 | const char *msg = "unexpected error code"; | |
12779 | ||
12780 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
12781 | op->op == Transaction::OP_CLONE || | |
12782 | op->op == Transaction::OP_CLONERANGE2)) | |
12783 | msg = "ENOENT on clone suggests osd bug"; | |
12784 | ||
12785 | if (r == -ENOSPC) | |
12786 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
12787 | // by partially applying transactions. | |
12788 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
12789 | ||
12790 | if (r == -ENOTEMPTY) { | |
12791 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
12792 | } | |
12793 | ||
12794 | derr << __func__ << " error " << cpp_strerror(r) | |
12795 | << " not handled on operation " << op->op | |
12796 | << " (op " << pos << ", counting from 0)" | |
12797 | << dendl; | |
12798 | derr << msg << dendl; | |
81eedcae | 12799 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12800 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12801 | } |
12802 | } | |
12803 | } | |
12804 | } | |
12805 | ||
12806 | ||
12807 | ||
12808 | // ----------------- | |
12809 | // write operations | |
12810 | ||
12811 | int BlueStore::_touch(TransContext *txc, | |
12812 | CollectionRef& c, | |
12813 | OnodeRef &o) | |
12814 | { | |
12815 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
12816 | int r = 0; | |
7c673cae FG |
12817 | _assign_nid(txc, o); |
12818 | txc->write_onode(o); | |
12819 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
12820 | return r; | |
12821 | } | |
12822 | ||
7c673cae FG |
12823 | void BlueStore::_pad_zeros( |
12824 | bufferlist *bl, uint64_t *offset, | |
12825 | uint64_t chunk_size) | |
12826 | { | |
12827 | auto length = bl->length(); | |
12828 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
12829 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
12830 | dout(40) << "before:\n"; | |
12831 | bl->hexdump(*_dout); | |
12832 | *_dout << dendl; | |
12833 | // front | |
12834 | size_t front_pad = *offset % chunk_size; | |
12835 | size_t back_pad = 0; | |
12836 | size_t pad_count = 0; | |
12837 | if (front_pad) { | |
11fdf7f2 TL |
12838 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
12839 | bufferptr z = buffer::create_small_page_aligned(chunk_size); | |
224ce89b | 12840 | z.zero(0, front_pad, false); |
7c673cae | 12841 | pad_count += front_pad; |
9f95a23c | 12842 | bl->begin().copy(front_copy, z.c_str() + front_pad); |
7c673cae FG |
12843 | if (front_copy + front_pad < chunk_size) { |
12844 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 12845 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
12846 | pad_count += back_pad; |
12847 | } | |
12848 | bufferlist old, t; | |
12849 | old.swap(*bl); | |
12850 | t.substr_of(old, front_copy, length - front_copy); | |
12851 | bl->append(z); | |
12852 | bl->claim_append(t); | |
12853 | *offset -= front_pad; | |
224ce89b | 12854 | length += pad_count; |
7c673cae FG |
12855 | } |
12856 | ||
12857 | // back | |
12858 | uint64_t end = *offset + length; | |
12859 | unsigned back_copy = end % chunk_size; | |
12860 | if (back_copy) { | |
11fdf7f2 | 12861 | ceph_assert(back_pad == 0); |
7c673cae | 12862 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 12863 | ceph_assert(back_copy <= length); |
7c673cae | 12864 | bufferptr tail(chunk_size); |
9f95a23c | 12865 | bl->begin(length - back_copy).copy(back_copy, tail.c_str()); |
224ce89b | 12866 | tail.zero(back_copy, back_pad, false); |
7c673cae FG |
12867 | bufferlist old; |
12868 | old.swap(*bl); | |
12869 | bl->substr_of(old, 0, length - back_copy); | |
12870 | bl->append(tail); | |
12871 | length += back_pad; | |
12872 | pad_count += back_pad; | |
12873 | } | |
12874 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
12875 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
12876 | << length << std::dec << dendl; | |
12877 | dout(40) << "after:\n"; | |
12878 | bl->hexdump(*_dout); | |
12879 | *_dout << dendl; | |
12880 | if (pad_count) | |
12881 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 12882 | ceph_assert(bl->length() == length); |
7c673cae FG |
12883 | } |
12884 | ||
12885 | void BlueStore::_do_write_small( | |
12886 | TransContext *txc, | |
12887 | CollectionRef &c, | |
12888 | OnodeRef o, | |
12889 | uint64_t offset, uint64_t length, | |
12890 | bufferlist::iterator& blp, | |
12891 | WriteContext *wctx) | |
12892 | { | |
12893 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
12894 | << std::dec << dendl; | |
11fdf7f2 | 12895 | ceph_assert(length < min_alloc_size); |
7c673cae FG |
12896 | uint64_t end_offs = offset + length; |
12897 | ||
12898 | logger->inc(l_bluestore_write_small); | |
12899 | logger->inc(l_bluestore_write_small_bytes, length); | |
12900 | ||
12901 | bufferlist bl; | |
12902 | blp.copy(length, bl); | |
12903 | ||
81eedcae TL |
12904 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
12905 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
12906 | uint32_t alloc_len = min_alloc_size; | |
12907 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
12908 | ||
12909 | bool any_change; | |
12910 | ||
12911 | // search suitable extent in both forward and reverse direction in | |
12912 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
12913 | // then check if blob can be reused via can_reuse_blob func or apply | |
12914 | // direct/deferred write (the latter for extents including or higher | |
12915 | // than 'offset' only). | |
12916 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
12917 | ||
7c673cae FG |
12918 | // Look for an existing mutable blob we can use. |
12919 | auto begin = o->extent_map.extent_map.begin(); | |
12920 | auto end = o->extent_map.extent_map.end(); | |
12921 | auto ep = o->extent_map.seek_lextent(offset); | |
12922 | if (ep != begin) { | |
12923 | --ep; | |
12924 | if (ep->blob_end() <= offset) { | |
12925 | ++ep; | |
12926 | } | |
12927 | } | |
12928 | auto prev_ep = ep; | |
12929 | if (prev_ep != begin) { | |
12930 | --prev_ep; | |
12931 | } else { | |
12932 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
12933 | } | |
12934 | ||
eafe8130 TL |
12935 | boost::container::flat_set<const bluestore_blob_t*> inspected_blobs; |
12936 | // We don't want to have more blobs than min alloc units fit | |
12937 | // into 2 max blobs | |
12938 | size_t blob_threshold = max_blob_size / min_alloc_size * 2 + 1; | |
12939 | bool above_blob_threshold = false; | |
12940 | ||
12941 | inspected_blobs.reserve(blob_threshold); | |
12942 | ||
12943 | uint64_t max_off = 0; | |
12944 | auto start_ep = ep; | |
12945 | auto end_ep = ep; // exclusively | |
7c673cae FG |
12946 | do { |
12947 | any_change = false; | |
12948 | ||
12949 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
12950 | BlobRef b = ep->blob; | |
eafe8130 TL |
12951 | if (!above_blob_threshold) { |
12952 | inspected_blobs.insert(&b->get_blob()); | |
12953 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
12954 | } | |
12955 | max_off = ep->logical_end(); | |
7c673cae | 12956 | auto bstart = ep->blob_start(); |
eafe8130 | 12957 | |
7c673cae FG |
12958 | dout(20) << __func__ << " considering " << *b |
12959 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
12960 | if (bstart >= end_offs) { | |
12961 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
12962 | } else if (!b->get_blob().is_mutable()) { | |
12963 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
12964 | } else if (ep->logical_offset % min_alloc_size != | |
12965 | ep->blob_offset % min_alloc_size) { | |
12966 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
12967 | } else { | |
12968 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
12969 | // can we pad our head/tail out with zeros? | |
12970 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
12971 | head_pad = p2phase(offset, chunk_size); |
12972 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
12973 | if (head_pad || tail_pad) { |
12974 | o->extent_map.fault_range(db, offset - head_pad, | |
12975 | end_offs - offset + head_pad + tail_pad); | |
12976 | } | |
12977 | if (head_pad && | |
12978 | o->extent_map.has_any_lextents(offset - head_pad, chunk_size)) { | |
12979 | head_pad = 0; | |
12980 | } | |
12981 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
12982 | tail_pad = 0; | |
12983 | } | |
12984 | ||
12985 | uint64_t b_off = offset - head_pad - bstart; | |
12986 | uint64_t b_len = length + head_pad + tail_pad; | |
12987 | ||
12988 | // direct write into unused blocks of an existing mutable blob? | |
12989 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
12990 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
12991 | b->get_blob().is_unused(b_off, b_len) && | |
12992 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 12993 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
12994 | |
12995 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
12996 | << b_off << "~" << b_len | |
12997 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
12998 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 12999 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13000 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13001 | ||
11fdf7f2 | 13002 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13003 | if (b_len <= prefer_deferred_size) { |
13004 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13005 | << b_len << std::dec << " unused write via deferred" << dendl; | |
9f95a23c | 13006 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13007 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13008 | b->get_blob().map( | |
13009 | b_off, b_len, | |
13010 | [&](uint64_t offset, uint64_t length) { | |
13011 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13012 | return 0; | |
13013 | }); | |
224ce89b | 13014 | op->data = bl; |
7c673cae FG |
13015 | } else { |
13016 | b->get_blob().map_bl( | |
224ce89b | 13017 | b_off, bl, |
7c673cae FG |
13018 | [&](uint64_t offset, bufferlist& t) { |
13019 | bdev->aio_write(offset, t, | |
13020 | &txc->ioc, wctx->buffered); | |
13021 | }); | |
13022 | } | |
13023 | } | |
224ce89b | 13024 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
13025 | dout(20) << __func__ << " lex old " << *ep << dendl; |
13026 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
13027 | b, | |
13028 | &wctx->old_extents); | |
13029 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13030 | txc->statfs_delta.stored() += le->length; | |
13031 | dout(20) << __func__ << " lex " << *le << dendl; | |
13032 | logger->inc(l_bluestore_write_small_unused); | |
13033 | return; | |
13034 | } | |
13035 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
13036 | uint64_t head_read = p2phase(b_off, chunk_size); |
13037 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
13038 | if ((head_read || tail_read) && |
13039 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
13040 | head_read + tail_read < min_alloc_size) { | |
13041 | b_off -= head_read; | |
13042 | b_len += head_read + tail_read; | |
13043 | ||
13044 | } else { | |
13045 | head_read = tail_read = 0; | |
13046 | } | |
13047 | ||
13048 | // chunk-aligned deferred overwrite? | |
13049 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13050 | b_off % chunk_size == 0 && | |
13051 | b_len % chunk_size == 0 && | |
13052 | b->get_blob().is_allocated(b_off, b_len)) { | |
13053 | ||
224ce89b | 13054 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13055 | |
13056 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
13057 | << " and tail 0x" << tail_read << std::dec << dendl; | |
13058 | if (head_read) { | |
13059 | bufferlist head_bl; | |
13060 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
13061 | head_bl, 0); | |
11fdf7f2 | 13062 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
13063 | size_t zlen = head_read - r; |
13064 | if (zlen) { | |
13065 | head_bl.append_zero(zlen); | |
13066 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13067 | } | |
11fdf7f2 TL |
13068 | head_bl.claim_append(bl); |
13069 | bl.swap(head_bl); | |
7c673cae FG |
13070 | logger->inc(l_bluestore_write_penalty_read_ops); |
13071 | } | |
13072 | if (tail_read) { | |
13073 | bufferlist tail_bl; | |
13074 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
13075 | tail_bl, 0); | |
11fdf7f2 | 13076 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
13077 | size_t zlen = tail_read - r; |
13078 | if (zlen) { | |
13079 | tail_bl.append_zero(zlen); | |
13080 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13081 | } | |
224ce89b | 13082 | bl.claim_append(tail_bl); |
7c673cae FG |
13083 | logger->inc(l_bluestore_write_penalty_read_ops); |
13084 | } | |
13085 | logger->inc(l_bluestore_write_small_pre_read); | |
13086 | ||
224ce89b | 13087 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13088 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13089 | ||
7c673cae | 13090 | if (b->get_blob().csum_type) { |
224ce89b | 13091 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae | 13092 | } |
11fdf7f2 TL |
13093 | |
13094 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
9f95a23c | 13095 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
11fdf7f2 TL |
13096 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13097 | int r = b->get_blob().map( | |
13098 | b_off, b_len, | |
13099 | [&](uint64_t offset, uint64_t length) { | |
13100 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13101 | return 0; | |
13102 | }); | |
13103 | ceph_assert(r == 0); | |
13104 | op->data.claim(bl); | |
13105 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" | |
13106 | << b_len << std::dec << " of mutable " << *b | |
13107 | << " at " << op->extents << dendl; | |
13108 | } | |
13109 | ||
7c673cae FG |
13110 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
13111 | b, &wctx->old_extents); | |
13112 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13113 | txc->statfs_delta.stored() += le->length; | |
13114 | dout(20) << __func__ << " lex " << *le << dendl; | |
13115 | logger->inc(l_bluestore_write_small_deferred); | |
13116 | return; | |
13117 | } | |
224ce89b WB |
13118 | // try to reuse blob if we can |
13119 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
13120 | max_bsize, |
13121 | offset0 - bstart, | |
13122 | &alloc_len)) { | |
11fdf7f2 | 13123 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13124 | // fit into reused blob |
13125 | // Need to check for pending writes desiring to | |
13126 | // reuse the same pextent. The rationale is that during GC two chunks | |
13127 | // from garbage blobs(compressed?) can share logical space within the same | |
13128 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13129 | // Hence the second write will fail in an attempt to reuse blob at | |
13130 | // do_alloc_write(). | |
13131 | if (!wctx->has_conflict(b, | |
13132 | offset0, | |
13133 | offset0 + alloc_len, | |
13134 | min_alloc_size)) { | |
13135 | ||
13136 | // we can't reuse pad_head/pad_tail since they might be truncated | |
13137 | // due to existent extents | |
13138 | uint64_t b_off = offset - bstart; | |
13139 | uint64_t b_off0 = b_off; | |
13140 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13141 | ||
13142 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13143 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13144 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13145 | << std::dec << dendl; |
13146 | ||
13147 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13148 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13149 | false, false); | |
13150 | logger->inc(l_bluestore_write_small_unused); | |
13151 | return; | |
13152 | } | |
13153 | } | |
13154 | } | |
13155 | ++ep; | |
eafe8130 | 13156 | end_ep = ep; |
7c673cae FG |
13157 | any_change = true; |
13158 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
13159 | ||
13160 | // check extent for reuse in reverse order | |
13161 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
13162 | BlobRef b = prev_ep->blob; | |
eafe8130 TL |
13163 | if (!above_blob_threshold) { |
13164 | inspected_blobs.insert(&b->get_blob()); | |
13165 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13166 | } | |
13167 | start_ep = prev_ep; | |
7c673cae FG |
13168 | auto bstart = prev_ep->blob_start(); |
13169 | dout(20) << __func__ << " considering " << *b | |
13170 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 13171 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
13172 | max_bsize, |
13173 | offset0 - bstart, | |
13174 | &alloc_len)) { | |
11fdf7f2 | 13175 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13176 | // fit into reused blob |
13177 | // Need to check for pending writes desiring to | |
13178 | // reuse the same pextent. The rationale is that during GC two chunks | |
13179 | // from garbage blobs(compressed?) can share logical space within the same | |
13180 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13181 | // Hence the second write will fail in an attempt to reuse blob at | |
13182 | // do_alloc_write(). | |
13183 | if (!wctx->has_conflict(b, | |
13184 | offset0, | |
13185 | offset0 + alloc_len, | |
13186 | min_alloc_size)) { | |
13187 | ||
13188 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13189 | uint64_t b_off = offset - bstart; | |
13190 | uint64_t b_off0 = b_off; | |
13191 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13192 | ||
13193 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13194 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13195 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13196 | << std::dec << dendl; |
13197 | ||
13198 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13199 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13200 | false, false); | |
13201 | logger->inc(l_bluestore_write_small_unused); | |
13202 | return; | |
13203 | } | |
13204 | } | |
13205 | if (prev_ep != begin) { | |
13206 | --prev_ep; | |
13207 | any_change = true; | |
13208 | } else { | |
13209 | prev_ep = end; // to avoid useless first extent re-check | |
13210 | } | |
13211 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
13212 | } while (any_change); | |
13213 | ||
eafe8130 TL |
13214 | if (above_blob_threshold) { |
13215 | dout(10) << __func__ << " request GC, blobs >= " << inspected_blobs.size() | |
13216 | << " " << std::hex << min_off << "~" << max_off << std::dec | |
13217 | << dendl; | |
13218 | ceph_assert(start_ep != end_ep); | |
13219 | for (auto ep = start_ep; ep != end_ep; ++ep) { | |
13220 | dout(20) << __func__ << " inserting for GC " | |
13221 | << std::hex << ep->logical_offset << "~" << ep->length | |
13222 | << std::dec << dendl; | |
13223 | ||
13224 | wctx->extents_to_gc.union_insert(ep->logical_offset, ep->length); | |
13225 | } | |
13226 | // insert newly written extent to GC | |
13227 | wctx->extents_to_gc.union_insert(offset, length); | |
13228 | dout(20) << __func__ << " inserting (last) for GC " | |
13229 | << std::hex << offset << "~" << length | |
13230 | << std::dec << dendl; | |
13231 | } | |
7c673cae | 13232 | // new blob. |
7c673cae | 13233 | BlobRef b = c->new_blob(); |
11fdf7f2 | 13234 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae FG |
13235 | uint64_t b_off0 = b_off; |
13236 | _pad_zeros(&bl, &b_off0, block_size); | |
13237 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13238 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, true, true); | |
7c673cae FG |
13239 | |
13240 | return; | |
13241 | } | |
13242 | ||
13243 | void BlueStore::_do_write_big( | |
13244 | TransContext *txc, | |
13245 | CollectionRef &c, | |
13246 | OnodeRef o, | |
13247 | uint64_t offset, uint64_t length, | |
13248 | bufferlist::iterator& blp, | |
13249 | WriteContext *wctx) | |
13250 | { | |
13251 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13252 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
13253 | << " compress " << (int)wctx->compress | |
13254 | << dendl; | |
13255 | logger->inc(l_bluestore_write_big); | |
13256 | logger->inc(l_bluestore_write_big_bytes, length); | |
13257 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11fdf7f2 | 13258 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae FG |
13259 | while (length > 0) { |
13260 | bool new_blob = false; | |
11fdf7f2 | 13261 | uint32_t l = std::min(max_bsize, length); |
7c673cae FG |
13262 | BlobRef b; |
13263 | uint32_t b_off = 0; | |
13264 | ||
13265 | //attempting to reuse existing blob | |
13266 | if (!wctx->compress) { | |
13267 | // look for an existing mutable blob we can reuse | |
13268 | auto begin = o->extent_map.extent_map.begin(); | |
13269 | auto end = o->extent_map.extent_map.end(); | |
13270 | auto ep = o->extent_map.seek_lextent(offset); | |
13271 | auto prev_ep = ep; | |
13272 | if (prev_ep != begin) { | |
13273 | --prev_ep; | |
13274 | } else { | |
13275 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
13276 | } | |
13277 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
13278 | // search suitable extent in both forward and reverse direction in | |
13279 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 13280 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
13281 | bool any_change; |
13282 | do { | |
13283 | any_change = false; | |
13284 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13285 | if (offset >= ep->blob_start() && | |
224ce89b | 13286 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13287 | offset - ep->blob_start(), |
13288 | &l)) { | |
13289 | b = ep->blob; | |
13290 | b_off = offset - ep->blob_start(); | |
13291 | prev_ep = end; // to avoid check below | |
13292 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13293 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13294 | } else { |
13295 | ++ep; | |
13296 | any_change = true; | |
13297 | } | |
13298 | } | |
13299 | ||
13300 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
224ce89b | 13301 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13302 | offset - prev_ep->blob_start(), |
13303 | &l)) { | |
13304 | b = prev_ep->blob; | |
13305 | b_off = offset - prev_ep->blob_start(); | |
13306 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13307 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13308 | } else if (prev_ep != begin) { |
13309 | --prev_ep; | |
13310 | any_change = true; | |
13311 | } else { | |
13312 | prev_ep = end; // to avoid useless first extent re-check | |
13313 | } | |
13314 | } | |
13315 | } while (b == nullptr && any_change); | |
13316 | } | |
13317 | if (b == nullptr) { | |
13318 | b = c->new_blob(); | |
13319 | b_off = 0; | |
13320 | new_blob = true; | |
13321 | } | |
13322 | ||
13323 | bufferlist t; | |
13324 | blp.copy(l, t); | |
13325 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); | |
13326 | offset += l; | |
13327 | length -= l; | |
13328 | logger->inc(l_bluestore_write_big_blobs); | |
13329 | } | |
13330 | } | |
13331 | ||
13332 | int BlueStore::_do_alloc_write( | |
13333 | TransContext *txc, | |
13334 | CollectionRef coll, | |
13335 | OnodeRef o, | |
13336 | WriteContext *wctx) | |
13337 | { | |
13338 | dout(20) << __func__ << " txc " << txc | |
13339 | << " " << wctx->writes.size() << " blobs" | |
13340 | << dendl; | |
3efd9988 FG |
13341 | if (wctx->writes.empty()) { |
13342 | return 0; | |
7c673cae FG |
13343 | } |
13344 | ||
7c673cae FG |
13345 | CompressorRef c; |
13346 | double crr = 0; | |
13347 | if (wctx->compress) { | |
13348 | c = select_option( | |
13349 | "compression_algorithm", | |
13350 | compressor, | |
13351 | [&]() { | |
13352 | string val; | |
13353 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
13354 | CompressorRef cp = compressor; | |
13355 | if (!cp || cp->get_type_name() != val) { | |
13356 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
13357 | if (!cp) { |
13358 | if (_set_compression_alert(false, val.c_str())) { | |
13359 | derr << __func__ << " unable to initialize " << val.c_str() | |
13360 | << " compressor" << dendl; | |
13361 | } | |
13362 | } | |
7c673cae FG |
13363 | } |
13364 | return boost::optional<CompressorRef>(cp); | |
13365 | } | |
13366 | return boost::optional<CompressorRef>(); | |
13367 | } | |
13368 | ); | |
13369 | ||
13370 | crr = select_option( | |
13371 | "compression_required_ratio", | |
13372 | cct->_conf->bluestore_compression_required_ratio, | |
13373 | [&]() { | |
13374 | double val; | |
3efd9988 | 13375 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
7c673cae FG |
13376 | return boost::optional<double>(val); |
13377 | } | |
13378 | return boost::optional<double>(); | |
13379 | } | |
13380 | ); | |
13381 | } | |
13382 | ||
13383 | // checksum | |
11fdf7f2 | 13384 | int64_t csum = csum_type.load(); |
7c673cae FG |
13385 | csum = select_option( |
13386 | "csum_type", | |
13387 | csum, | |
13388 | [&]() { | |
11fdf7f2 | 13389 | int64_t val; |
3efd9988 | 13390 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
11fdf7f2 | 13391 | return boost::optional<int64_t>(val); |
7c673cae | 13392 | } |
11fdf7f2 | 13393 | return boost::optional<int64_t>(); |
7c673cae FG |
13394 | } |
13395 | ); | |
13396 | ||
3efd9988 FG |
13397 | // compress (as needed) and calc needed space |
13398 | uint64_t need = 0; | |
11fdf7f2 | 13399 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 13400 | for (auto& wi : wctx->writes) { |
3efd9988 | 13401 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 13402 | auto start = mono_clock::now(); |
7c673cae FG |
13403 | |
13404 | // compress | |
11fdf7f2 TL |
13405 | ceph_assert(wi.b_off == 0); |
13406 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 13407 | |
7c673cae FG |
13408 | // FIXME: memory alignment here is bad |
13409 | bufferlist t; | |
3efd9988 | 13410 | int r = c->compress(wi.bl, t); |
3efd9988 | 13411 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 13412 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
13413 | bool rejected = false; |
13414 | uint64_t compressed_len = t.length(); | |
13415 | // do an approximate (fast) estimation for resulting blob size | |
13416 | // that doesn't take header overhead into account | |
11fdf7f2 | 13417 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13418 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
13419 | bluestore_compression_header_t chdr; | |
13420 | chdr.type = c->get_type(); | |
13421 | chdr.length = t.length(); | |
13422 | encode(chdr, wi.compressed_bl); | |
13423 | wi.compressed_bl.claim_append(t); | |
13424 | ||
13425 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 13426 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13427 | if (result_len <= want_len && result_len < wi.blob_length) { |
13428 | // Cool. We compressed at least as much as we were hoping to. | |
13429 | // pad out to min_alloc_size | |
13430 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
13431 | wi.compressed_len = compressed_len; | |
13432 | wi.compressed = true; | |
13433 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
13434 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
13435 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
13436 | << " with " << c->get_type() | |
13437 | << std::dec << dendl; | |
13438 | txc->statfs_delta.compressed() += compressed_len; | |
13439 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
13440 | txc->statfs_delta.compressed_allocated() += result_len; | |
13441 | logger->inc(l_bluestore_compress_success_count); | |
13442 | need += result_len; | |
13443 | } else { | |
13444 | rejected = true; | |
13445 | } | |
13446 | } else if (r != 0) { | |
13447 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
13448 | << " bytes compressed using " << c->get_type_name() | |
13449 | << std::dec | |
13450 | << " failed with errcode = " << r | |
13451 | << ", leaving uncompressed" | |
13452 | << dendl; | |
13453 | logger->inc(l_bluestore_compress_rejected_count); | |
13454 | need += wi.blob_length; | |
7c673cae | 13455 | } else { |
a8e16298 TL |
13456 | rejected = true; |
13457 | } | |
13458 | ||
13459 | if (rejected) { | |
3efd9988 | 13460 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 13461 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
13462 | << " with " << c->get_type() |
13463 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 13464 | << " -> 0x" << want_len |
3efd9988 FG |
13465 | << ", leaving uncompressed" |
13466 | << std::dec << dendl; | |
13467 | logger->inc(l_bluestore_compress_rejected_count); | |
13468 | need += wi.blob_length; | |
7c673cae | 13469 | } |
494da23a TL |
13470 | log_latency("compress@_do_alloc_write", |
13471 | l_bluestore_compress_lat, | |
13472 | mono_clock::now() - start, | |
13473 | cct->_conf->bluestore_log_op_age ); | |
3efd9988 FG |
13474 | } else { |
13475 | need += wi.blob_length; | |
7c673cae | 13476 | } |
3efd9988 | 13477 | } |
a8e16298 | 13478 | PExtentVector prealloc; |
3efd9988 | 13479 | prealloc.reserve(2 * wctx->writes.size());; |
11fdf7f2 | 13480 | int64_t prealloc_left = 0; |
3efd9988 FG |
13481 | prealloc_left = alloc->allocate( |
13482 | need, min_alloc_size, need, | |
13483 | 0, &prealloc); | |
eafe8130 | 13484 | if (prealloc_left < 0 || prealloc_left < (int64_t)need) { |
11fdf7f2 | 13485 | derr << __func__ << " failed to allocate 0x" << std::hex << need |
eafe8130 | 13486 | << " allocated 0x " << (prealloc_left < 0 ? 0 : prealloc_left) |
11fdf7f2 TL |
13487 | << " min_alloc_size 0x" << min_alloc_size |
13488 | << " available 0x " << alloc->get_free() | |
13489 | << std::dec << dendl; | |
13490 | if (prealloc.size()) { | |
13491 | alloc->release(prealloc); | |
13492 | } | |
a8e16298 TL |
13493 | return -ENOSPC; |
13494 | } | |
9f95a23c | 13495 | _collect_allocation_stats(need, min_alloc_size, prealloc.size()); |
a8e16298 | 13496 | |
3efd9988 FG |
13497 | dout(20) << __func__ << " prealloc " << prealloc << dendl; |
13498 | auto prealloc_pos = prealloc.begin(); | |
13499 | ||
13500 | for (auto& wi : wctx->writes) { | |
13501 | BlobRef b = wi.b; | |
13502 | bluestore_blob_t& dblob = b->dirty_blob(); | |
13503 | uint64_t b_off = wi.b_off; | |
13504 | bufferlist *l = &wi.bl; | |
13505 | uint64_t final_length = wi.blob_length; | |
13506 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
13507 | if (wi.compressed) { |
13508 | final_length = wi.compressed_bl.length(); | |
13509 | csum_length = final_length; | |
3efd9988 FG |
13510 | l = &wi.compressed_bl; |
13511 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
13512 | } else if (wi.new_blob) { | |
7c673cae | 13513 | // initialize newly created blob only |
11fdf7f2 TL |
13514 | ceph_assert(dblob.is_mutable()); |
13515 | unsigned csum_order; | |
7c673cae FG |
13516 | if (l->length() != wi.blob_length) { |
13517 | // hrm, maybe we could do better here, but let's not bother. | |
13518 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
13519 | << block_size_order << dendl; | |
31f18b77 | 13520 | csum_order = block_size_order; |
7c673cae FG |
13521 | } else { |
13522 | csum_order = std::min(wctx->csum_order, ctz(l->length())); | |
13523 | } | |
13524 | // try to align blob with max_blob_size to improve | |
13525 | // its reuse ratio, e.g. in case of reverse write | |
13526 | uint32_t suggested_boff = | |
13527 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
13528 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
13529 | suggested_boff + final_length <= max_bsize && | |
13530 | suggested_boff > b_off) { | |
181888fb | 13531 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 13532 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 13533 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
13534 | csum_length += suggested_boff - b_off; |
13535 | b_off = suggested_boff; | |
13536 | } | |
181888fb FG |
13537 | if (csum != Checksummer::CSUM_NONE) { |
13538 | dout(20) << __func__ << " initialize csum setting for new blob " << *b | |
13539 | << " csum_type " << Checksummer::get_csum_type_string(csum) | |
13540 | << " csum_order " << csum_order | |
13541 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
13542 | << dendl; | |
13543 | dblob.init_csum(csum, csum_order, csum_length); | |
13544 | } | |
7c673cae FG |
13545 | } |
13546 | ||
a8e16298 | 13547 | PExtentVector extents; |
3efd9988 FG |
13548 | int64_t left = final_length; |
13549 | while (left > 0) { | |
11fdf7f2 | 13550 | ceph_assert(prealloc_left > 0); |
3efd9988 FG |
13551 | if (prealloc_pos->length <= left) { |
13552 | prealloc_left -= prealloc_pos->length; | |
13553 | left -= prealloc_pos->length; | |
13554 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
13555 | extents.push_back(*prealloc_pos); | |
13556 | ++prealloc_pos; | |
13557 | } else { | |
13558 | extents.emplace_back(prealloc_pos->offset, left); | |
13559 | prealloc_pos->offset += left; | |
13560 | prealloc_pos->length -= left; | |
13561 | prealloc_left -= left; | |
13562 | txc->statfs_delta.allocated() += left; | |
13563 | left = 0; | |
13564 | break; | |
13565 | } | |
13566 | } | |
7c673cae | 13567 | for (auto& p : extents) { |
3efd9988 | 13568 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 13569 | } |
11fdf7f2 | 13570 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 13571 | |
181888fb FG |
13572 | dout(20) << __func__ << " blob " << *b << dendl; |
13573 | if (dblob.has_csum()) { | |
7c673cae FG |
13574 | dblob.calc_csum(b_off, *l); |
13575 | } | |
181888fb | 13576 | |
7c673cae FG |
13577 | if (wi.mark_unused) { |
13578 | auto b_end = b_off + wi.bl.length(); | |
13579 | if (b_off) { | |
13580 | dblob.add_unused(0, b_off); | |
13581 | } | |
13582 | if (b_end < wi.blob_length) { | |
13583 | dblob.add_unused(b_end, wi.blob_length - b_end); | |
13584 | } | |
13585 | } | |
13586 | ||
13587 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
13588 | b_off + (wi.b_off0 - wi.b_off), | |
13589 | wi.length0, | |
13590 | wi.b, | |
13591 | nullptr); | |
13592 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13593 | txc->statfs_delta.stored() += le->length; | |
13594 | dout(20) << __func__ << " lex " << *le << dendl; | |
13595 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
13596 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
13597 | ||
13598 | // queue io | |
11fdf7f2 | 13599 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13600 | if (l->length() <= prefer_deferred_size.load()) { |
13601 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13602 | << l->length() << std::dec << " write via deferred" << dendl; | |
9f95a23c | 13603 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13604 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13605 | int r = b->get_blob().map( | |
13606 | b_off, l->length(), | |
13607 | [&](uint64_t offset, uint64_t length) { | |
13608 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13609 | return 0; | |
13610 | }); | |
11fdf7f2 | 13611 | ceph_assert(r == 0); |
7c673cae | 13612 | op->data = *l; |
81eedcae | 13613 | logger->inc(l_bluestore_write_small_deferred); |
7c673cae FG |
13614 | } else { |
13615 | b->get_blob().map_bl( | |
13616 | b_off, *l, | |
13617 | [&](uint64_t offset, bufferlist& t) { | |
13618 | bdev->aio_write(offset, t, &txc->ioc, false); | |
13619 | }); | |
81eedcae | 13620 | logger->inc(l_bluestore_write_small_new); |
7c673cae FG |
13621 | } |
13622 | } | |
13623 | } | |
11fdf7f2 TL |
13624 | ceph_assert(prealloc_pos == prealloc.end()); |
13625 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
13626 | return 0; |
13627 | } | |
13628 | ||
13629 | void BlueStore::_wctx_finish( | |
13630 | TransContext *txc, | |
13631 | CollectionRef& c, | |
13632 | OnodeRef o, | |
31f18b77 FG |
13633 | WriteContext *wctx, |
13634 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
13635 | { |
13636 | auto oep = wctx->old_extents.begin(); | |
13637 | while (oep != wctx->old_extents.end()) { | |
13638 | auto &lo = *oep; | |
13639 | oep = wctx->old_extents.erase(oep); | |
13640 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
13641 | BlobRef b = lo.e.blob; | |
13642 | const bluestore_blob_t& blob = b->get_blob(); | |
13643 | if (blob.is_compressed()) { | |
13644 | if (lo.blob_empty) { | |
13645 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
13646 | } | |
13647 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
13648 | } | |
13649 | auto& r = lo.r; | |
13650 | txc->statfs_delta.stored() -= lo.e.length; | |
13651 | if (!r.empty()) { | |
13652 | dout(20) << __func__ << " blob release " << r << dendl; | |
13653 | if (blob.is_shared()) { | |
13654 | PExtentVector final; | |
13655 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
13656 | bool unshare = false; |
13657 | bool* unshare_ptr = | |
13658 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 13659 | for (auto e : r) { |
31f18b77 FG |
13660 | b->shared_blob->put_ref( |
13661 | e.offset, e.length, &final, | |
11fdf7f2 TL |
13662 | unshare_ptr); |
13663 | } | |
13664 | if (unshare) { | |
13665 | ceph_assert(maybe_unshared_blobs); | |
13666 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
13667 | } |
13668 | dout(20) << __func__ << " shared_blob release " << final | |
13669 | << " from " << *b->shared_blob << dendl; | |
13670 | txc->write_shared_blob(b->shared_blob); | |
13671 | r.clear(); | |
13672 | r.swap(final); | |
13673 | } | |
13674 | } | |
13675 | // we can't invalidate our logical extents as we drop them because | |
13676 | // other lextents (either in our onode or others) may still | |
13677 | // reference them. but we can throw out anything that is no | |
13678 | // longer allocated. Note that this will leave behind edge bits | |
13679 | // that are no longer referenced but not deallocated (until they | |
13680 | // age out of the cache naturally). | |
13681 | b->discard_unallocated(c.get()); | |
13682 | for (auto e : r) { | |
13683 | dout(20) << __func__ << " release " << e << dendl; | |
13684 | txc->released.insert(e.offset, e.length); | |
13685 | txc->statfs_delta.allocated() -= e.length; | |
13686 | if (blob.is_compressed()) { | |
13687 | txc->statfs_delta.compressed_allocated() -= e.length; | |
13688 | } | |
13689 | } | |
9f95a23c TL |
13690 | |
13691 | if (b->is_spanning() && !b->is_referenced() && lo.blob_empty) { | |
7c673cae FG |
13692 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b |
13693 | << dendl; | |
13694 | o->extent_map.spanning_blob_map.erase(b->id); | |
13695 | } | |
9f95a23c | 13696 | delete &lo; |
7c673cae FG |
13697 | } |
13698 | } | |
13699 | ||
13700 | void BlueStore::_do_write_data( | |
13701 | TransContext *txc, | |
13702 | CollectionRef& c, | |
13703 | OnodeRef o, | |
13704 | uint64_t offset, | |
13705 | uint64_t length, | |
13706 | bufferlist& bl, | |
13707 | WriteContext *wctx) | |
13708 | { | |
13709 | uint64_t end = offset + length; | |
13710 | bufferlist::iterator p = bl.begin(); | |
13711 | ||
13712 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
13713 | (length != min_alloc_size)) { | |
13714 | // we fall within the same block | |
13715 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
13716 | } else { | |
13717 | uint64_t head_offset, head_length; | |
13718 | uint64_t middle_offset, middle_length; | |
13719 | uint64_t tail_offset, tail_length; | |
13720 | ||
13721 | head_offset = offset; | |
11fdf7f2 | 13722 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 13723 | |
11fdf7f2 TL |
13724 | tail_offset = p2align(end, min_alloc_size); |
13725 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
13726 | |
13727 | middle_offset = head_offset + head_length; | |
13728 | middle_length = length - head_length - tail_length; | |
13729 | ||
13730 | if (head_length) { | |
13731 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
13732 | } | |
13733 | ||
13734 | if (middle_length) { | |
13735 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); | |
13736 | } | |
13737 | ||
13738 | if (tail_length) { | |
13739 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
13740 | } | |
13741 | } | |
13742 | } | |
13743 | ||
31f18b77 FG |
13744 | void BlueStore::_choose_write_options( |
13745 | CollectionRef& c, | |
13746 | OnodeRef o, | |
13747 | uint32_t fadvise_flags, | |
13748 | WriteContext *wctx) | |
7c673cae | 13749 | { |
7c673cae FG |
13750 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
13751 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 13752 | wctx->buffered = true; |
7c673cae FG |
13753 | } else if (cct->_conf->bluestore_default_buffered_write && |
13754 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
13755 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
13756 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 13757 | wctx->buffered = true; |
7c673cae FG |
13758 | } |
13759 | ||
31f18b77 FG |
13760 | // apply basic csum block size |
13761 | wctx->csum_order = block_size_order; | |
7c673cae FG |
13762 | |
13763 | // compression parameters | |
13764 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
13765 | auto cm = select_option( | |
13766 | "compression_mode", | |
31f18b77 | 13767 | comp_mode.load(), |
7c673cae FG |
13768 | [&]() { |
13769 | string val; | |
11fdf7f2 | 13770 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
31f18b77 FG |
13771 | return boost::optional<Compressor::CompressionMode>( |
13772 | Compressor::get_comp_mode_type(val)); | |
7c673cae FG |
13773 | } |
13774 | return boost::optional<Compressor::CompressionMode>(); | |
13775 | } | |
13776 | ); | |
31f18b77 FG |
13777 | |
13778 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
13779 | ((cm == Compressor::COMP_FORCE) || |
13780 | (cm == Compressor::COMP_AGGRESSIVE && | |
13781 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
13782 | (cm == Compressor::COMP_PASSIVE && | |
13783 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
13784 | ||
13785 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
13786 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
13787 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
13788 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 13789 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 13790 | |
7c673cae | 13791 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 13792 | |
7c673cae | 13793 | if (o->onode.expected_write_size) { |
224ce89b | 13794 | wctx->csum_order = std::max(min_alloc_size_order, |
31f18b77 | 13795 | (uint8_t)ctz(o->onode.expected_write_size)); |
7c673cae | 13796 | } else { |
224ce89b | 13797 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
13798 | } |
13799 | ||
31f18b77 FG |
13800 | if (wctx->compress) { |
13801 | wctx->target_blob_size = select_option( | |
7c673cae | 13802 | "compression_max_blob_size", |
31f18b77 | 13803 | comp_max_blob_size.load(), |
7c673cae | 13804 | [&]() { |
11fdf7f2 TL |
13805 | int64_t val; |
13806 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
7c673cae FG |
13807 | return boost::optional<uint64_t>((uint64_t)val); |
13808 | } | |
13809 | return boost::optional<uint64_t>(); | |
13810 | } | |
13811 | ); | |
13812 | } | |
13813 | } else { | |
31f18b77 FG |
13814 | if (wctx->compress) { |
13815 | wctx->target_blob_size = select_option( | |
7c673cae | 13816 | "compression_min_blob_size", |
31f18b77 | 13817 | comp_min_blob_size.load(), |
7c673cae | 13818 | [&]() { |
11fdf7f2 TL |
13819 | int64_t val; |
13820 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
7c673cae FG |
13821 | return boost::optional<uint64_t>((uint64_t)val); |
13822 | } | |
13823 | return boost::optional<uint64_t>(); | |
13824 | } | |
13825 | ); | |
13826 | } | |
13827 | } | |
31f18b77 | 13828 | |
7c673cae | 13829 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
13830 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
13831 | wctx->target_blob_size = max_bsize; | |
7c673cae | 13832 | } |
31f18b77 | 13833 | |
7c673cae FG |
13834 | // set the min blob size floor at 2x the min_alloc_size, or else we |
13835 | // won't be able to allocate a smaller extent for the compressed | |
13836 | // data. | |
31f18b77 FG |
13837 | if (wctx->compress && |
13838 | wctx->target_blob_size < min_alloc_size * 2) { | |
13839 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 13840 | } |
31f18b77 FG |
13841 | |
13842 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
13843 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
13844 | << " compress=" << (int)wctx->compress |
13845 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
13846 | << std::dec << dendl; |
13847 | } | |
13848 | ||
13849 | int BlueStore::_do_gc( | |
13850 | TransContext *txc, | |
13851 | CollectionRef& c, | |
13852 | OnodeRef o, | |
31f18b77 FG |
13853 | const WriteContext& wctx, |
13854 | uint64_t *dirty_start, | |
13855 | uint64_t *dirty_end) | |
13856 | { | |
31f18b77 | 13857 | |
1adf2230 | 13858 | bool dirty_range_updated = false; |
31f18b77 | 13859 | WriteContext wctx_gc; |
7c673cae | 13860 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 13861 | |
eafe8130 | 13862 | auto & extents_to_collect = wctx.extents_to_gc; |
31f18b77 FG |
13863 | for (auto it = extents_to_collect.begin(); |
13864 | it != extents_to_collect.end(); | |
13865 | ++it) { | |
13866 | bufferlist bl; | |
eafe8130 TL |
13867 | auto offset = (*it).first; |
13868 | auto length = (*it).second; | |
13869 | dout(20) << __func__ << " processing " << std::hex | |
13870 | << offset << "~" << length << std::dec | |
13871 | << dendl; | |
13872 | int r = _do_read(c.get(), o, offset, length, bl, 0); | |
13873 | ceph_assert(r == (int)length); | |
31f18b77 | 13874 | |
eafe8130 TL |
13875 | _do_write_data(txc, c, o, offset, length, bl, &wctx_gc); |
13876 | logger->inc(l_bluestore_gc_merged, length); | |
31f18b77 | 13877 | |
eafe8130 TL |
13878 | if (*dirty_start > offset) { |
13879 | *dirty_start = offset; | |
1adf2230 | 13880 | dirty_range_updated = true; |
31f18b77 FG |
13881 | } |
13882 | ||
eafe8130 TL |
13883 | if (*dirty_end < offset + length) { |
13884 | *dirty_end = offset + length; | |
1adf2230 | 13885 | dirty_range_updated = true; |
31f18b77 FG |
13886 | } |
13887 | } | |
1adf2230 AA |
13888 | if (dirty_range_updated) { |
13889 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
13890 | } | |
31f18b77 FG |
13891 | |
13892 | dout(30) << __func__ << " alloc write" << dendl; | |
13893 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
13894 | if (r < 0) { | |
13895 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
13896 | << dendl; | |
13897 | return r; | |
13898 | } | |
13899 | ||
13900 | _wctx_finish(txc, c, o, &wctx_gc); | |
13901 | return 0; | |
13902 | } | |
13903 | ||
13904 | int BlueStore::_do_write( | |
13905 | TransContext *txc, | |
13906 | CollectionRef& c, | |
13907 | OnodeRef o, | |
13908 | uint64_t offset, | |
13909 | uint64_t length, | |
13910 | bufferlist& bl, | |
13911 | uint32_t fadvise_flags) | |
13912 | { | |
13913 | int r = 0; | |
13914 | ||
13915 | dout(20) << __func__ | |
13916 | << " " << o->oid | |
13917 | << " 0x" << std::hex << offset << "~" << length | |
13918 | << " - have 0x" << o->onode.size | |
13919 | << " (" << std::dec << o->onode.size << ")" | |
13920 | << " bytes" | |
13921 | << " fadvise_flags 0x" << std::hex << fadvise_flags << std::dec | |
13922 | << dendl; | |
81eedcae | 13923 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
13924 | |
13925 | if (length == 0) { | |
13926 | return 0; | |
13927 | } | |
13928 | ||
13929 | uint64_t end = offset + length; | |
13930 | ||
13931 | GarbageCollector gc(c->store->cct); | |
eafe8130 | 13932 | int64_t benefit = 0; |
31f18b77 FG |
13933 | auto dirty_start = offset; |
13934 | auto dirty_end = end; | |
13935 | ||
13936 | WriteContext wctx; | |
13937 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
13938 | o->extent_map.fault_range(db, offset, length); |
13939 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
13940 | r = _do_alloc_write(txc, c, o, &wctx); |
13941 | if (r < 0) { | |
13942 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
13943 | << dendl; | |
13944 | goto out; | |
13945 | } | |
13946 | ||
eafe8130 TL |
13947 | if (wctx.extents_to_gc.empty() || |
13948 | wctx.extents_to_gc.range_start() > offset || | |
13949 | wctx.extents_to_gc.range_end() < offset + length) { | |
13950 | benefit = gc.estimate(offset, | |
13951 | length, | |
13952 | o->extent_map, | |
13953 | wctx.old_extents, | |
13954 | min_alloc_size); | |
13955 | } | |
13956 | ||
31f18b77 FG |
13957 | // NB: _wctx_finish() will empty old_extents |
13958 | // so we must do gc estimation before that | |
7c673cae FG |
13959 | _wctx_finish(txc, c, o, &wctx); |
13960 | if (end > o->onode.size) { | |
13961 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 13962 | << std::dec << dendl; |
7c673cae FG |
13963 | o->onode.size = end; |
13964 | } | |
13965 | ||
11fdf7f2 | 13966 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
eafe8130 TL |
13967 | wctx.extents_to_gc.union_of(gc.get_extents_to_collect()); |
13968 | dout(20) << __func__ | |
13969 | << " perform garbage collection for compressed extents, " | |
13970 | << "expected benefit = " << benefit << " AUs" << dendl; | |
13971 | } | |
13972 | if (!wctx.extents_to_gc.empty()) { | |
13973 | dout(20) << __func__ << " perform garbage collection" << dendl; | |
13974 | ||
13975 | r = _do_gc(txc, c, o, | |
13976 | wctx, | |
13977 | &dirty_start, &dirty_end); | |
13978 | if (r < 0) { | |
13979 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
13980 | << dendl; | |
13981 | goto out; | |
7c673cae | 13982 | } |
eafe8130 TL |
13983 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
13984 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae | 13985 | } |
7c673cae | 13986 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
13987 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
13988 | ||
7c673cae FG |
13989 | r = 0; |
13990 | ||
13991 | out: | |
13992 | return r; | |
13993 | } | |
13994 | ||
13995 | int BlueStore::_write(TransContext *txc, | |
13996 | CollectionRef& c, | |
13997 | OnodeRef& o, | |
31f18b77 FG |
13998 | uint64_t offset, size_t length, |
13999 | bufferlist& bl, | |
14000 | uint32_t fadvise_flags) | |
7c673cae FG |
14001 | { |
14002 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14003 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14004 | << dendl; | |
35e4c445 FG |
14005 | int r = 0; |
14006 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14007 | r = -E2BIG; | |
14008 | } else { | |
14009 | _assign_nid(txc, o); | |
14010 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
14011 | txc->write_onode(o); | |
14012 | } | |
7c673cae FG |
14013 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14014 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14015 | << " = " << r << dendl; | |
14016 | return r; | |
14017 | } | |
14018 | ||
14019 | int BlueStore::_zero(TransContext *txc, | |
14020 | CollectionRef& c, | |
14021 | OnodeRef& o, | |
14022 | uint64_t offset, size_t length) | |
14023 | { | |
14024 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14025 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14026 | << dendl; | |
35e4c445 FG |
14027 | int r = 0; |
14028 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14029 | r = -E2BIG; | |
14030 | } else { | |
14031 | _assign_nid(txc, o); | |
14032 | r = _do_zero(txc, c, o, offset, length); | |
14033 | } | |
7c673cae FG |
14034 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14035 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14036 | << " = " << r << dendl; | |
14037 | return r; | |
14038 | } | |
14039 | ||
14040 | int BlueStore::_do_zero(TransContext *txc, | |
14041 | CollectionRef& c, | |
14042 | OnodeRef& o, | |
14043 | uint64_t offset, size_t length) | |
14044 | { | |
14045 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14046 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14047 | << dendl; | |
14048 | int r = 0; | |
14049 | ||
81eedcae | 14050 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14051 | |
14052 | WriteContext wctx; | |
14053 | o->extent_map.fault_range(db, offset, length); | |
14054 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 14055 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
14056 | _wctx_finish(txc, c, o, &wctx); |
14057 | ||
b32b8144 | 14058 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
14059 | o->onode.size = offset + length; |
14060 | dout(20) << __func__ << " extending size to " << offset + length | |
14061 | << dendl; | |
14062 | } | |
14063 | txc->write_onode(o); | |
14064 | ||
14065 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14066 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14067 | << " = " << r << dendl; | |
14068 | return r; | |
14069 | } | |
14070 | ||
14071 | void BlueStore::_do_truncate( | |
31f18b77 FG |
14072 | TransContext *txc, CollectionRef& c, OnodeRef o, uint64_t offset, |
14073 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14074 | { |
14075 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14076 | << " 0x" << std::hex << offset << std::dec << dendl; | |
14077 | ||
81eedcae | 14078 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14079 | |
14080 | if (offset == o->onode.size) | |
31f18b77 | 14081 | return; |
7c673cae FG |
14082 | |
14083 | if (offset < o->onode.size) { | |
14084 | WriteContext wctx; | |
14085 | uint64_t length = o->onode.size - offset; | |
14086 | o->extent_map.fault_range(db, offset, length); | |
14087 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 FG |
14088 | o->extent_map.dirty_range(offset, length); |
14089 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); | |
7c673cae FG |
14090 | |
14091 | // if we have shards past EOF, ask for a reshard | |
14092 | if (!o->onode.extent_map_shards.empty() && | |
14093 | o->onode.extent_map_shards.back().offset >= offset) { | |
14094 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
14095 | if (offset) { | |
14096 | o->extent_map.request_reshard(offset - 1, offset + length); | |
14097 | } else { | |
14098 | o->extent_map.request_reshard(0, length); | |
14099 | } | |
14100 | } | |
14101 | } | |
14102 | ||
14103 | o->onode.size = offset; | |
14104 | ||
14105 | txc->write_onode(o); | |
14106 | } | |
14107 | ||
35e4c445 | 14108 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
14109 | CollectionRef& c, |
14110 | OnodeRef& o, | |
14111 | uint64_t offset) | |
14112 | { | |
14113 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14114 | << " 0x" << std::hex << offset << std::dec | |
14115 | << dendl; | |
35e4c445 FG |
14116 | int r = 0; |
14117 | if (offset >= OBJECT_MAX_SIZE) { | |
14118 | r = -E2BIG; | |
14119 | } else { | |
14120 | _do_truncate(txc, c, o, offset); | |
14121 | } | |
14122 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14123 | << " 0x" << std::hex << offset << std::dec | |
14124 | << " = " << r << dendl; | |
14125 | return r; | |
7c673cae FG |
14126 | } |
14127 | ||
14128 | int BlueStore::_do_remove( | |
14129 | TransContext *txc, | |
14130 | CollectionRef& c, | |
14131 | OnodeRef o) | |
14132 | { | |
31f18b77 | 14133 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
14134 | bool is_gen = !o->oid.is_no_gen(); |
14135 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
14136 | if (o->onode.has_omap()) { |
14137 | o->flush(); | |
9f95a23c | 14138 | _do_omap_clear(txc, o); |
7c673cae FG |
14139 | } |
14140 | o->exists = false; | |
14141 | string key; | |
14142 | for (auto &s : o->extent_map.shards) { | |
14143 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
14144 | << s.shard_info->offset << std::dec << dendl; | |
14145 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
14146 | [&](const string& final_key) { | |
14147 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14148 | } | |
14149 | ); | |
14150 | } | |
14151 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 14152 | txc->note_removed_object(o); |
7c673cae FG |
14153 | o->extent_map.clear(); |
14154 | o->onode = bluestore_onode_t(); | |
14155 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 14156 | |
224ce89b WB |
14157 | if (!is_gen || maybe_unshared_blobs.empty()) { |
14158 | return 0; | |
14159 | } | |
31f18b77 | 14160 | |
224ce89b WB |
14161 | // see if we can unshare blobs still referenced by the head |
14162 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
14163 | << maybe_unshared_blobs << dendl; | |
14164 | ghobject_t nogen = o->oid; | |
14165 | nogen.generation = ghobject_t::NO_GEN; | |
14166 | OnodeRef h = c->onode_map.lookup(nogen); | |
14167 | ||
14168 | if (!h || !h->exists) { | |
14169 | return 0; | |
14170 | } | |
14171 | ||
14172 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
14173 | << " " << h->oid << dendl; | |
14174 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
14175 | for (auto& e : h->extent_map.extent_map) { | |
14176 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14177 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14178 | if (b.is_shared() && | |
14179 | sb->loaded && | |
14180 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
14181 | if (b.is_compressed()) { |
14182 | expect[sb].get(0, b.get_ondisk_length()); | |
14183 | } else { | |
14184 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
14185 | expect[sb].get(off, len); | |
14186 | return 0; | |
14187 | }); | |
14188 | } | |
224ce89b WB |
14189 | } |
14190 | } | |
31f18b77 | 14191 | |
224ce89b WB |
14192 | vector<SharedBlob*> unshared_blobs; |
14193 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
14194 | for (auto& p : expect) { | |
14195 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
14196 | if (p.first->persistent->ref_map == p.second) { | |
14197 | SharedBlob *sb = p.first; | |
14198 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
14199 | unshared_blobs.push_back(sb); | |
14200 | txc->unshare_blob(sb); | |
14201 | uint64_t sbid = c->make_blob_unshared(sb); | |
14202 | string key; | |
14203 | get_shared_blob_key(sbid, &key); | |
14204 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
14205 | } | |
14206 | } | |
14207 | ||
14208 | if (unshared_blobs.empty()) { | |
14209 | return 0; | |
14210 | } | |
14211 | ||
224ce89b WB |
14212 | for (auto& e : h->extent_map.extent_map) { |
14213 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14214 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14215 | if (b.is_shared() && | |
14216 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
14217 | sb) != unshared_blobs.end()) { | |
14218 | dout(20) << __func__ << " unsharing " << e << dendl; | |
14219 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
14220 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 14221 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
14222 | } |
14223 | } | |
224ce89b WB |
14224 | txc->write_onode(h); |
14225 | ||
7c673cae FG |
14226 | return 0; |
14227 | } | |
14228 | ||
14229 | int BlueStore::_remove(TransContext *txc, | |
14230 | CollectionRef& c, | |
14231 | OnodeRef &o) | |
14232 | { | |
11fdf7f2 TL |
14233 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
14234 | << " onode " << o.get() | |
14235 | << " txc "<< txc << dendl; | |
7c673cae FG |
14236 | int r = _do_remove(txc, c, o); |
14237 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14238 | return r; | |
14239 | } | |
14240 | ||
14241 | int BlueStore::_setattr(TransContext *txc, | |
14242 | CollectionRef& c, | |
14243 | OnodeRef& o, | |
14244 | const string& name, | |
14245 | bufferptr& val) | |
14246 | { | |
14247 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14248 | << " " << name << " (" << val.length() << " bytes)" | |
14249 | << dendl; | |
14250 | int r = 0; | |
3efd9988 FG |
14251 | if (val.is_partial()) { |
14252 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
14253 | val.length()); | |
14254 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
14255 | } else { | |
14256 | auto& b = o->onode.attrs[name.c_str()] = val; | |
14257 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
14258 | } | |
7c673cae FG |
14259 | txc->write_onode(o); |
14260 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14261 | << " " << name << " (" << val.length() << " bytes)" | |
14262 | << " = " << r << dendl; | |
14263 | return r; | |
14264 | } | |
14265 | ||
14266 | int BlueStore::_setattrs(TransContext *txc, | |
14267 | CollectionRef& c, | |
14268 | OnodeRef& o, | |
14269 | const map<string,bufferptr>& aset) | |
14270 | { | |
14271 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14272 | << " " << aset.size() << " keys" | |
14273 | << dendl; | |
14274 | int r = 0; | |
14275 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
14276 | p != aset.end(); ++p) { | |
3efd9988 FG |
14277 | if (p->second.is_partial()) { |
14278 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 14279 | bufferptr(p->second.c_str(), p->second.length()); |
3efd9988 FG |
14280 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); |
14281 | } else { | |
14282 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
14283 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
14284 | } | |
7c673cae FG |
14285 | } |
14286 | txc->write_onode(o); | |
14287 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14288 | << " " << aset.size() << " keys" | |
14289 | << " = " << r << dendl; | |
14290 | return r; | |
14291 | } | |
14292 | ||
14293 | ||
14294 | int BlueStore::_rmattr(TransContext *txc, | |
14295 | CollectionRef& c, | |
14296 | OnodeRef& o, | |
14297 | const string& name) | |
14298 | { | |
14299 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14300 | << " " << name << dendl; | |
14301 | int r = 0; | |
14302 | auto it = o->onode.attrs.find(name.c_str()); | |
14303 | if (it == o->onode.attrs.end()) | |
14304 | goto out; | |
14305 | ||
14306 | o->onode.attrs.erase(it); | |
14307 | txc->write_onode(o); | |
14308 | ||
14309 | out: | |
14310 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14311 | << " " << name << " = " << r << dendl; | |
14312 | return r; | |
14313 | } | |
14314 | ||
14315 | int BlueStore::_rmattrs(TransContext *txc, | |
14316 | CollectionRef& c, | |
14317 | OnodeRef& o) | |
14318 | { | |
14319 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14320 | int r = 0; | |
14321 | ||
14322 | if (o->onode.attrs.empty()) | |
14323 | goto out; | |
14324 | ||
14325 | o->onode.attrs.clear(); | |
14326 | txc->write_onode(o); | |
14327 | ||
14328 | out: | |
14329 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14330 | return r; | |
14331 | } | |
14332 | ||
9f95a23c | 14333 | void BlueStore::_do_omap_clear(TransContext *txc, OnodeRef& o) |
7c673cae | 14334 | { |
9f95a23c | 14335 | const string& omap_prefix = o->get_omap_prefix(); |
7c673cae | 14336 | string prefix, tail; |
9f95a23c TL |
14337 | o->get_omap_header(&prefix); |
14338 | o->get_omap_tail(&tail); | |
11fdf7f2 | 14339 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
494da23a | 14340 | txc->t->rmkey(omap_prefix, tail); |
11fdf7f2 TL |
14341 | dout(20) << __func__ << " remove range start: " |
14342 | << pretty_binary_string(prefix) << " end: " | |
14343 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
14344 | } |
14345 | ||
14346 | int BlueStore::_omap_clear(TransContext *txc, | |
14347 | CollectionRef& c, | |
14348 | OnodeRef& o) | |
14349 | { | |
14350 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14351 | int r = 0; | |
14352 | if (o->onode.has_omap()) { | |
14353 | o->flush(); | |
9f95a23c | 14354 | _do_omap_clear(txc, o); |
7c673cae FG |
14355 | o->onode.clear_omap_flag(); |
14356 | txc->write_onode(o); | |
14357 | } | |
14358 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14359 | return r; | |
14360 | } | |
14361 | ||
14362 | int BlueStore::_omap_setkeys(TransContext *txc, | |
14363 | CollectionRef& c, | |
14364 | OnodeRef& o, | |
14365 | bufferlist &bl) | |
14366 | { | |
14367 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14368 | int r; | |
11fdf7f2 | 14369 | auto p = bl.cbegin(); |
7c673cae FG |
14370 | __u32 num; |
14371 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14372 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14373 | o->onode.set_omap_flags_pgmeta(); |
14374 | } else { | |
14375 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14376 | } |
7c673cae | 14377 | txc->write_onode(o); |
494da23a | 14378 | |
9f95a23c | 14379 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14380 | string key_tail; |
14381 | bufferlist tail; | |
9f95a23c | 14382 | o->get_omap_tail(&key_tail); |
494da23a | 14383 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14384 | } else { |
14385 | txc->note_modified_object(o); | |
14386 | } | |
9f95a23c | 14387 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 14388 | string final_key; |
9f95a23c TL |
14389 | o->get_omap_key(string(), &final_key); |
14390 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 14391 | decode(num, p); |
7c673cae FG |
14392 | while (num--) { |
14393 | string key; | |
14394 | bufferlist value; | |
11fdf7f2 TL |
14395 | decode(key, p); |
14396 | decode(value, p); | |
9f95a23c | 14397 | final_key.resize(base_key_len); // keep prefix |
7c673cae | 14398 | final_key += key; |
11fdf7f2 | 14399 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 14400 | << " <- " << key << dendl; |
11fdf7f2 | 14401 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
14402 | } |
14403 | r = 0; | |
14404 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14405 | return r; | |
14406 | } | |
14407 | ||
14408 | int BlueStore::_omap_setheader(TransContext *txc, | |
14409 | CollectionRef& c, | |
14410 | OnodeRef &o, | |
14411 | bufferlist& bl) | |
14412 | { | |
14413 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14414 | int r; | |
14415 | string key; | |
14416 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14417 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14418 | o->onode.set_omap_flags_pgmeta(); |
14419 | } else { | |
14420 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14421 | } |
7c673cae | 14422 | txc->write_onode(o); |
494da23a | 14423 | |
9f95a23c | 14424 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14425 | string key_tail; |
14426 | bufferlist tail; | |
9f95a23c | 14427 | o->get_omap_tail(&key_tail); |
494da23a | 14428 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14429 | } else { |
14430 | txc->note_modified_object(o); | |
14431 | } | |
9f95a23c TL |
14432 | const string& prefix = o->get_omap_prefix(); |
14433 | o->get_omap_header(&key); | |
11fdf7f2 | 14434 | txc->t->set(prefix, key, bl); |
7c673cae FG |
14435 | r = 0; |
14436 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14437 | return r; | |
14438 | } | |
14439 | ||
14440 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
14441 | CollectionRef& c, | |
14442 | OnodeRef& o, | |
14443 | bufferlist& bl) | |
14444 | { | |
14445 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14446 | int r = 0; | |
11fdf7f2 | 14447 | auto p = bl.cbegin(); |
7c673cae FG |
14448 | __u32 num; |
14449 | string final_key; | |
14450 | ||
14451 | if (!o->onode.has_omap()) { | |
14452 | goto out; | |
14453 | } | |
11fdf7f2 | 14454 | { |
9f95a23c TL |
14455 | const string& prefix = o->get_omap_prefix(); |
14456 | o->get_omap_key(string(), &final_key); | |
14457 | size_t base_key_len = final_key.size(); | |
11fdf7f2 TL |
14458 | decode(num, p); |
14459 | while (num--) { | |
14460 | string key; | |
14461 | decode(key, p); | |
9f95a23c | 14462 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
14463 | final_key += key; |
14464 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
14465 | << " <- " << key << dendl; | |
14466 | txc->t->rmkey(prefix, final_key); | |
14467 | } | |
7c673cae FG |
14468 | } |
14469 | txc->note_modified_object(o); | |
14470 | ||
14471 | out: | |
14472 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14473 | return r; | |
14474 | } | |
14475 | ||
14476 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
14477 | CollectionRef& c, | |
14478 | OnodeRef& o, | |
14479 | const string& first, const string& last) | |
14480 | { | |
14481 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
14482 | string key_first, key_last; |
14483 | int r = 0; | |
14484 | if (!o->onode.has_omap()) { | |
14485 | goto out; | |
14486 | } | |
11fdf7f2 | 14487 | { |
9f95a23c | 14488 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 14489 | o->flush(); |
9f95a23c TL |
14490 | o->get_omap_key(first, &key_first); |
14491 | o->get_omap_key(last, &key_last); | |
11fdf7f2 TL |
14492 | txc->t->rm_range_keys(prefix, key_first, key_last); |
14493 | dout(20) << __func__ << " remove range start: " | |
14494 | << pretty_binary_string(key_first) << " end: " | |
14495 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
14496 | } |
14497 | txc->note_modified_object(o); | |
14498 | ||
14499 | out: | |
14500 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14501 | return r; | |
14502 | } | |
14503 | ||
14504 | int BlueStore::_set_alloc_hint( | |
14505 | TransContext *txc, | |
14506 | CollectionRef& c, | |
14507 | OnodeRef& o, | |
14508 | uint64_t expected_object_size, | |
14509 | uint64_t expected_write_size, | |
14510 | uint32_t flags) | |
14511 | { | |
14512 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14513 | << " object_size " << expected_object_size | |
14514 | << " write_size " << expected_write_size | |
14515 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
14516 | << dendl; | |
14517 | int r = 0; | |
14518 | o->onode.expected_object_size = expected_object_size; | |
14519 | o->onode.expected_write_size = expected_write_size; | |
14520 | o->onode.alloc_hint_flags = flags; | |
14521 | txc->write_onode(o); | |
14522 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14523 | << " object_size " << expected_object_size | |
14524 | << " write_size " << expected_write_size | |
14525 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
14526 | << " = " << r << dendl; | |
14527 | return r; | |
14528 | } | |
14529 | ||
14530 | int BlueStore::_clone(TransContext *txc, | |
14531 | CollectionRef& c, | |
14532 | OnodeRef& oldo, | |
14533 | OnodeRef& newo) | |
14534 | { | |
14535 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14536 | << newo->oid << dendl; | |
14537 | int r = 0; | |
14538 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
14539 | derr << __func__ << " mismatched hash on " << oldo->oid | |
14540 | << " and " << newo->oid << dendl; | |
14541 | return -EINVAL; | |
14542 | } | |
14543 | ||
7c673cae FG |
14544 | _assign_nid(txc, newo); |
14545 | ||
14546 | // clone data | |
14547 | oldo->flush(); | |
14548 | _do_truncate(txc, c, newo, 0); | |
14549 | if (cct->_conf->bluestore_clone_cow) { | |
14550 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
14551 | } else { | |
14552 | bufferlist bl; | |
14553 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
14554 | if (r < 0) | |
14555 | goto out; | |
14556 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
14557 | if (r < 0) | |
14558 | goto out; | |
14559 | } | |
14560 | ||
14561 | // clone attrs | |
14562 | newo->onode.attrs = oldo->onode.attrs; | |
14563 | ||
14564 | // clone omap | |
14565 | if (newo->onode.has_omap()) { | |
14566 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
14567 | newo->flush(); | |
9f95a23c | 14568 | _do_omap_clear(txc, newo); |
494da23a | 14569 | newo->onode.clear_omap_flag(); |
7c673cae FG |
14570 | } |
14571 | if (oldo->onode.has_omap()) { | |
14572 | dout(20) << __func__ << " copying omap data" << dendl; | |
494da23a | 14573 | if (newo->oid.is_pgmeta()) { |
9f95a23c TL |
14574 | newo->onode.set_omap_flags_pgmeta(); |
14575 | } else { | |
14576 | newo->onode.set_omap_flags(); | |
7c673cae | 14577 | } |
9f95a23c | 14578 | const string& prefix = newo->get_omap_prefix(); |
11fdf7f2 | 14579 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 14580 | string head, tail; |
9f95a23c TL |
14581 | oldo->get_omap_header(&head); |
14582 | oldo->get_omap_tail(&tail); | |
7c673cae FG |
14583 | it->lower_bound(head); |
14584 | while (it->valid()) { | |
14585 | if (it->key() >= tail) { | |
14586 | dout(30) << __func__ << " reached tail" << dendl; | |
14587 | break; | |
14588 | } else { | |
14589 | dout(30) << __func__ << " got header/data " | |
14590 | << pretty_binary_string(it->key()) << dendl; | |
14591 | string key; | |
9f95a23c | 14592 | newo->rewrite_omap_key(it->key(), &key); |
11fdf7f2 | 14593 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
14594 | } |
14595 | it->next(); | |
14596 | } | |
494da23a TL |
14597 | string new_tail; |
14598 | bufferlist new_tail_value; | |
9f95a23c | 14599 | newo->get_omap_tail(&new_tail); |
494da23a | 14600 | txc->t->set(prefix, new_tail, new_tail_value); |
7c673cae FG |
14601 | } |
14602 | ||
14603 | txc->write_onode(newo); | |
14604 | r = 0; | |
14605 | ||
14606 | out: | |
14607 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14608 | << newo->oid << " = " << r << dendl; | |
14609 | return r; | |
14610 | } | |
14611 | ||
14612 | int BlueStore::_do_clone_range( | |
14613 | TransContext *txc, | |
14614 | CollectionRef& c, | |
14615 | OnodeRef& oldo, | |
14616 | OnodeRef& newo, | |
224ce89b WB |
14617 | uint64_t srcoff, |
14618 | uint64_t length, | |
14619 | uint64_t dstoff) | |
7c673cae FG |
14620 | { |
14621 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14622 | << newo->oid | |
14623 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
14624 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
14625 | oldo->extent_map.fault_range(db, srcoff, length); | |
14626 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
14627 | _dump_onode<30>(cct, *oldo); |
14628 | _dump_onode<30>(cct, *newo); | |
7c673cae | 14629 | |
11fdf7f2 | 14630 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
81eedcae TL |
14631 | _dump_onode<30>(cct, *oldo); |
14632 | _dump_onode<30>(cct, *newo); | |
7c673cae FG |
14633 | return 0; |
14634 | } | |
14635 | ||
14636 | int BlueStore::_clone_range(TransContext *txc, | |
14637 | CollectionRef& c, | |
14638 | OnodeRef& oldo, | |
14639 | OnodeRef& newo, | |
14640 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
14641 | { | |
14642 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14643 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
14644 | << " to offset 0x" << dstoff << std::dec << dendl; | |
14645 | int r = 0; | |
14646 | ||
35e4c445 FG |
14647 | if (srcoff + length >= OBJECT_MAX_SIZE || |
14648 | dstoff + length >= OBJECT_MAX_SIZE) { | |
14649 | r = -E2BIG; | |
14650 | goto out; | |
14651 | } | |
7c673cae FG |
14652 | if (srcoff + length > oldo->onode.size) { |
14653 | r = -EINVAL; | |
14654 | goto out; | |
14655 | } | |
14656 | ||
7c673cae FG |
14657 | _assign_nid(txc, newo); |
14658 | ||
14659 | if (length > 0) { | |
14660 | if (cct->_conf->bluestore_clone_cow) { | |
14661 | _do_zero(txc, c, newo, dstoff, length); | |
14662 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
14663 | } else { | |
14664 | bufferlist bl; | |
14665 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
14666 | if (r < 0) | |
14667 | goto out; | |
14668 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
14669 | if (r < 0) | |
14670 | goto out; | |
14671 | } | |
14672 | } | |
14673 | ||
14674 | txc->write_onode(newo); | |
14675 | r = 0; | |
14676 | ||
14677 | out: | |
14678 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14679 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
14680 | << " to offset 0x" << dstoff << std::dec | |
14681 | << " = " << r << dendl; | |
14682 | return r; | |
14683 | } | |
14684 | ||
14685 | int BlueStore::_rename(TransContext *txc, | |
14686 | CollectionRef& c, | |
14687 | OnodeRef& oldo, | |
14688 | OnodeRef& newo, | |
14689 | const ghobject_t& new_oid) | |
14690 | { | |
14691 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14692 | << new_oid << dendl; | |
14693 | int r; | |
14694 | ghobject_t old_oid = oldo->oid; | |
31f18b77 | 14695 | mempool::bluestore_cache_other::string new_okey; |
7c673cae FG |
14696 | |
14697 | if (newo) { | |
14698 | if (newo->exists) { | |
14699 | r = -EEXIST; | |
14700 | goto out; | |
14701 | } | |
11fdf7f2 | 14702 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
14703 | } |
14704 | ||
14705 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
14706 | ||
14707 | // rewrite shards | |
14708 | { | |
14709 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
14710 | get_object_key(cct, new_oid, &new_okey); | |
14711 | string key; | |
14712 | for (auto &s : oldo->extent_map.shards) { | |
14713 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
14714 | [&](const string& final_key) { | |
14715 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14716 | } | |
14717 | ); | |
14718 | s.dirty = true; | |
14719 | } | |
14720 | } | |
14721 | ||
14722 | newo = oldo; | |
14723 | txc->write_onode(newo); | |
14724 | ||
14725 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
14726 | // Onode in the old slot | |
14727 | c->onode_map.rename(oldo, old_oid, new_oid, new_okey); | |
14728 | r = 0; | |
14729 | ||
f64942e4 AA |
14730 | // hold a ref to new Onode in old name position, to ensure we don't drop |
14731 | // it from the cache before this txc commits (or else someone may come along | |
14732 | // and read newo's metadata via the old name). | |
14733 | txc->note_modified_object(oldo); | |
14734 | ||
7c673cae FG |
14735 | out: |
14736 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
14737 | << new_oid << " = " << r << dendl; | |
14738 | return r; | |
14739 | } | |
14740 | ||
14741 | // collections | |
14742 | ||
14743 | int BlueStore::_create_collection( | |
14744 | TransContext *txc, | |
14745 | const coll_t &cid, | |
14746 | unsigned bits, | |
14747 | CollectionRef *c) | |
14748 | { | |
14749 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
14750 | int r; | |
14751 | bufferlist bl; | |
14752 | ||
14753 | { | |
9f95a23c | 14754 | std::unique_lock l(coll_lock); |
7c673cae FG |
14755 | if (*c) { |
14756 | r = -EEXIST; | |
14757 | goto out; | |
14758 | } | |
11fdf7f2 TL |
14759 | auto p = new_coll_map.find(cid); |
14760 | ceph_assert(p != new_coll_map.end()); | |
14761 | *c = p->second; | |
7c673cae FG |
14762 | (*c)->cnode.bits = bits; |
14763 | coll_map[cid] = *c; | |
11fdf7f2 | 14764 | new_coll_map.erase(p); |
7c673cae | 14765 | } |
11fdf7f2 | 14766 | encode((*c)->cnode, bl); |
7c673cae FG |
14767 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
14768 | r = 0; | |
14769 | ||
14770 | out: | |
14771 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
14772 | return r; | |
14773 | } | |
14774 | ||
14775 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
14776 | CollectionRef *c) | |
14777 | { | |
14778 | dout(15) << __func__ << " " << cid << dendl; | |
14779 | int r; | |
14780 | ||
11fdf7f2 | 14781 | (*c)->flush_all_but_last(); |
7c673cae | 14782 | { |
9f95a23c | 14783 | std::unique_lock l(coll_lock); |
7c673cae FG |
14784 | if (!*c) { |
14785 | r = -ENOENT; | |
14786 | goto out; | |
14787 | } | |
14788 | size_t nonexistent_count = 0; | |
11fdf7f2 | 14789 | ceph_assert((*c)->exists); |
7c673cae FG |
14790 | if ((*c)->onode_map.map_any([&](OnodeRef o) { |
14791 | if (o->exists) { | |
494da23a TL |
14792 | dout(1) << __func__ << " " << o->oid << " " << o |
14793 | << " exists in onode_map" << dendl; | |
7c673cae FG |
14794 | return true; |
14795 | } | |
14796 | ++nonexistent_count; | |
14797 | return false; | |
14798 | })) { | |
14799 | r = -ENOTEMPTY; | |
14800 | goto out; | |
14801 | } | |
14802 | ||
14803 | vector<ghobject_t> ls; | |
14804 | ghobject_t next; | |
14805 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
14806 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 14807 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae FG |
14808 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
14809 | nonexistent_count + 1, &ls, &next); | |
14810 | if (r >= 0) { | |
11fdf7f2 TL |
14811 | // If true mean collecton has more objects than nonexistent_count, |
14812 | // so bypass check. | |
14813 | bool exists = (!next.is_max()); | |
7c673cae FG |
14814 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
14815 | dout(10) << __func__ << " oid " << *it << dendl; | |
14816 | auto onode = (*c)->onode_map.lookup(*it); | |
14817 | exists = !onode || onode->exists; | |
14818 | if (exists) { | |
494da23a TL |
14819 | dout(1) << __func__ << " " << *it |
14820 | << " exists in db, " | |
14821 | << (!onode ? "not present in ram" : "present in ram") | |
14822 | << dendl; | |
7c673cae FG |
14823 | } |
14824 | } | |
14825 | if (!exists) { | |
11fdf7f2 | 14826 | _do_remove_collection(txc, c); |
7c673cae FG |
14827 | r = 0; |
14828 | } else { | |
14829 | dout(10) << __func__ << " " << cid | |
14830 | << " is non-empty" << dendl; | |
14831 | r = -ENOTEMPTY; | |
14832 | } | |
14833 | } | |
14834 | } | |
14835 | ||
14836 | out: | |
14837 | dout(10) << __func__ << " " << cid << " = " << r << dendl; | |
14838 | return r; | |
14839 | } | |
14840 | ||
11fdf7f2 TL |
14841 | void BlueStore::_do_remove_collection(TransContext *txc, |
14842 | CollectionRef *c) | |
14843 | { | |
14844 | coll_map.erase((*c)->cid); | |
14845 | txc->removed_collections.push_back(*c); | |
14846 | (*c)->exists = false; | |
14847 | _osr_register_zombie((*c)->osr.get()); | |
14848 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
14849 | c->reset(); | |
14850 | } | |
14851 | ||
7c673cae FG |
14852 | int BlueStore::_split_collection(TransContext *txc, |
14853 | CollectionRef& c, | |
14854 | CollectionRef& d, | |
14855 | unsigned bits, int rem) | |
14856 | { | |
14857 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
14858 | << " bits " << bits << dendl; | |
9f95a23c TL |
14859 | std::unique_lock l(c->lock); |
14860 | std::unique_lock l2(d->lock); | |
7c673cae FG |
14861 | int r; |
14862 | ||
14863 | // flush all previous deferred writes on this sequencer. this is a bit | |
14864 | // heavyweight, but we need to make sure all deferred writes complete | |
14865 | // before we split as the new collection's sequencer may need to order | |
14866 | // this after those writes, and we don't bother with the complexity of | |
14867 | // moving those TransContexts over to the new osr. | |
14868 | _osr_drain_preceding(txc); | |
14869 | ||
14870 | // move any cached items (onodes and referenced shared blobs) that will | |
14871 | // belong to the child collection post-split. leave everything else behind. | |
14872 | // this may include things that don't strictly belong to the now-smaller | |
14873 | // parent split, but the OSD will always send us a split for every new | |
14874 | // child. | |
14875 | ||
14876 | spg_t pgid, dest_pgid; | |
14877 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 14878 | ceph_assert(is_pg); |
7c673cae | 14879 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 14880 | ceph_assert(is_pg); |
7c673cae FG |
14881 | |
14882 | // the destination should initially be empty. | |
11fdf7f2 TL |
14883 | ceph_assert(d->onode_map.empty()); |
14884 | ceph_assert(d->shared_blob_set.empty()); | |
14885 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
14886 | |
14887 | c->split_cache(d.get()); | |
14888 | ||
14889 | // adjust bits. note that this will be redundant for all but the first | |
14890 | // split call for this parent (first child). | |
14891 | c->cnode.bits = bits; | |
11fdf7f2 | 14892 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
14893 | r = 0; |
14894 | ||
14895 | bufferlist bl; | |
11fdf7f2 | 14896 | encode(c->cnode, bl); |
7c673cae FG |
14897 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
14898 | ||
14899 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
14900 | << " bits " << bits << " = " << r << dendl; | |
14901 | return r; | |
14902 | } | |
14903 | ||
11fdf7f2 TL |
14904 | int BlueStore::_merge_collection( |
14905 | TransContext *txc, | |
14906 | CollectionRef *c, | |
14907 | CollectionRef& d, | |
14908 | unsigned bits) | |
14909 | { | |
14910 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
14911 | << " bits " << bits << dendl; | |
9f95a23c TL |
14912 | std::unique_lock l((*c)->lock); |
14913 | std::unique_lock l2(d->lock); | |
11fdf7f2 TL |
14914 | int r; |
14915 | ||
14916 | coll_t cid = (*c)->cid; | |
14917 | ||
14918 | // flush all previous deferred writes on the source collection to ensure | |
14919 | // that all deferred writes complete before we merge as the target collection's | |
14920 | // sequencer may need to order new ops after those writes. | |
14921 | ||
14922 | _osr_drain((*c)->osr.get()); | |
14923 | ||
14924 | // move any cached items (onodes and referenced shared blobs) that will | |
14925 | // belong to the child collection post-split. leave everything else behind. | |
14926 | // this may include things that don't strictly belong to the now-smaller | |
14927 | // parent split, but the OSD will always send us a split for every new | |
14928 | // child. | |
14929 | ||
14930 | spg_t pgid, dest_pgid; | |
14931 | bool is_pg = cid.is_pg(&pgid); | |
14932 | ceph_assert(is_pg); | |
14933 | is_pg = d->cid.is_pg(&dest_pgid); | |
14934 | ceph_assert(is_pg); | |
14935 | ||
14936 | // adjust bits. note that this will be redundant for all but the first | |
14937 | // merge call for the parent/target. | |
14938 | d->cnode.bits = bits; | |
14939 | ||
14940 | // behavior depends on target (d) bits, so this after that is updated. | |
14941 | (*c)->split_cache(d.get()); | |
14942 | ||
14943 | // remove source collection | |
14944 | { | |
9f95a23c | 14945 | std::unique_lock l3(coll_lock); |
11fdf7f2 TL |
14946 | _do_remove_collection(txc, c); |
14947 | } | |
14948 | ||
14949 | r = 0; | |
14950 | ||
14951 | bufferlist bl; | |
14952 | encode(d->cnode, bl); | |
14953 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
14954 | ||
14955 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
14956 | << " bits " << bits << " = " << r << dendl; | |
14957 | return r; | |
14958 | } | |
14959 | ||
494da23a TL |
14960 | void BlueStore::log_latency( |
14961 | const char* name, | |
14962 | int idx, | |
14963 | const ceph::timespan& l, | |
14964 | double lat_threshold, | |
14965 | const char* info) const | |
14966 | { | |
14967 | logger->tinc(idx, l); | |
14968 | if (lat_threshold > 0.0 && | |
14969 | l >= make_timespan(lat_threshold)) { | |
14970 | dout(0) << __func__ << " slow operation observed for " << name | |
14971 | << ", latency = " << l | |
14972 | << info | |
14973 | << dendl; | |
14974 | } | |
14975 | } | |
14976 | ||
11fdf7f2 | 14977 | void BlueStore::log_latency_fn( |
494da23a | 14978 | const char* name, |
11fdf7f2 TL |
14979 | int idx, |
14980 | const ceph::timespan& l, | |
494da23a TL |
14981 | double lat_threshold, |
14982 | std::function<string (const ceph::timespan& lat)> fn) const | |
11fdf7f2 | 14983 | { |
494da23a TL |
14984 | logger->tinc(idx, l); |
14985 | if (lat_threshold > 0.0 && | |
14986 | l >= make_timespan(lat_threshold)) { | |
14987 | dout(0) << __func__ << " slow operation observed for " << name | |
14988 | << ", latency = " << l | |
14989 | << fn(l) | |
14990 | << dendl; | |
14991 | } | |
11fdf7f2 TL |
14992 | } |
14993 | ||
9f95a23c TL |
14994 | #if defined(WITH_LTTNG) |
14995 | void BlueStore::BlueStoreThrottle::emit_initial_tracepoint( | |
14996 | KeyValueDB &db, | |
14997 | TransContext &txc, | |
14998 | mono_clock::time_point start_throttle_acquire) | |
14999 | { | |
15000 | pending_kv_ios += txc.ios; | |
15001 | if (txc.deferred_txn) { | |
15002 | pending_deferred_ios += txc.ios; | |
15003 | } | |
15004 | ||
15005 | uint64_t started = 0; | |
15006 | uint64_t completed = 0; | |
15007 | if (should_trace(&started, &completed)) { | |
15008 | txc.tracing = true; | |
15009 | uint64_t rocksdb_base_level, | |
15010 | rocksdb_estimate_pending_compaction_bytes, | |
15011 | rocksdb_cur_size_all_mem_tables, | |
15012 | rocksdb_compaction_pending, | |
15013 | rocksdb_mem_table_flush_pending, | |
15014 | rocksdb_num_running_compactions, | |
15015 | rocksdb_num_running_flushes, | |
15016 | rocksdb_actual_delayed_write_rate; | |
15017 | db.get_property( | |
15018 | "rocksdb.base-level", | |
15019 | &rocksdb_base_level); | |
15020 | db.get_property( | |
15021 | "rocksdb.estimate-pending-compaction-bytes", | |
15022 | &rocksdb_estimate_pending_compaction_bytes); | |
15023 | db.get_property( | |
15024 | "rocksdb.cur-size-all-mem-tables", | |
15025 | &rocksdb_cur_size_all_mem_tables); | |
15026 | db.get_property( | |
15027 | "rocksdb.compaction-pending", | |
15028 | &rocksdb_compaction_pending); | |
15029 | db.get_property( | |
15030 | "rocksdb.mem-table-flush-pending", | |
15031 | &rocksdb_mem_table_flush_pending); | |
15032 | db.get_property( | |
15033 | "rocksdb.num-running-compactions", | |
15034 | &rocksdb_num_running_compactions); | |
15035 | db.get_property( | |
15036 | "rocksdb.num-running-flushes", | |
15037 | &rocksdb_num_running_flushes); | |
15038 | db.get_property( | |
15039 | "rocksdb.actual-delayed-write-rate", | |
15040 | &rocksdb_actual_delayed_write_rate); | |
15041 | ||
15042 | ||
15043 | tracepoint( | |
15044 | bluestore, | |
15045 | transaction_initial_state, | |
15046 | txc.osr->get_sequencer_id(), | |
15047 | txc.seq, | |
15048 | throttle_bytes.get_current(), | |
15049 | throttle_deferred_bytes.get_current(), | |
15050 | pending_kv_ios, | |
15051 | pending_deferred_ios, | |
15052 | started, | |
15053 | completed, | |
15054 | ceph::to_seconds<double>(mono_clock::now() - start_throttle_acquire)); | |
15055 | ||
15056 | tracepoint( | |
15057 | bluestore, | |
15058 | transaction_initial_state_rocksdb, | |
15059 | txc.osr->get_sequencer_id(), | |
15060 | txc.seq, | |
15061 | rocksdb_base_level, | |
15062 | rocksdb_estimate_pending_compaction_bytes, | |
15063 | rocksdb_cur_size_all_mem_tables, | |
15064 | rocksdb_compaction_pending, | |
15065 | rocksdb_mem_table_flush_pending, | |
15066 | rocksdb_num_running_compactions, | |
15067 | rocksdb_num_running_flushes, | |
15068 | rocksdb_actual_delayed_write_rate); | |
15069 | } | |
15070 | } | |
15071 | #endif | |
15072 | ||
15073 | mono_clock::duration BlueStore::BlueStoreThrottle::log_state_latency( | |
15074 | TransContext &txc, PerfCounters *logger, int state) | |
15075 | { | |
15076 | mono_clock::time_point now = mono_clock::now(); | |
15077 | mono_clock::duration lat = now - txc.last_stamp; | |
15078 | logger->tinc(state, lat); | |
15079 | #if defined(WITH_LTTNG) | |
15080 | if (txc.tracing && | |
15081 | state >= l_bluestore_state_prepare_lat && | |
15082 | state <= l_bluestore_state_done_lat) { | |
15083 | OID_ELAPSED("", lat.to_nsec() / 1000.0, txc.get_state_latency_name(state)); | |
15084 | tracepoint( | |
15085 | bluestore, | |
15086 | transaction_state_duration, | |
15087 | txc.osr->get_sequencer_id(), | |
15088 | txc.seq, | |
15089 | state, | |
15090 | ceph::to_seconds<double>(lat)); | |
15091 | } | |
15092 | #endif | |
15093 | txc.last_stamp = now; | |
15094 | return lat; | |
15095 | } | |
15096 | ||
15097 | bool BlueStore::BlueStoreThrottle::try_start_transaction( | |
15098 | KeyValueDB &db, | |
15099 | TransContext &txc, | |
15100 | mono_clock::time_point start_throttle_acquire) | |
15101 | { | |
15102 | throttle_bytes.get(txc.cost); | |
15103 | ||
15104 | if (!txc.deferred_txn || throttle_deferred_bytes.get_or_fail(txc.cost)) { | |
15105 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15106 | return true; | |
15107 | } else { | |
15108 | return false; | |
15109 | } | |
15110 | } | |
15111 | ||
15112 | void BlueStore::BlueStoreThrottle::finish_start_transaction( | |
15113 | KeyValueDB &db, | |
15114 | TransContext &txc, | |
15115 | mono_clock::time_point start_throttle_acquire) | |
15116 | { | |
15117 | ceph_assert(txc.deferred_txn); | |
15118 | throttle_deferred_bytes.get(txc.cost); | |
15119 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15120 | } | |
15121 | ||
15122 | #if defined(WITH_LTTNG) | |
15123 | void BlueStore::BlueStoreThrottle::complete_kv(TransContext &txc) | |
15124 | { | |
15125 | pending_kv_ios -= 1; | |
15126 | ios_completed_since_last_traced++; | |
15127 | if (txc.tracing) { | |
15128 | tracepoint( | |
15129 | bluestore, | |
15130 | transaction_commit_latency, | |
15131 | txc.osr->get_sequencer_id(), | |
15132 | txc.seq, | |
15133 | ceph::to_seconds<double>(mono_clock::now() - txc.start)); | |
15134 | } | |
15135 | } | |
15136 | #endif | |
15137 | ||
15138 | #if defined(WITH_LTTNG) | |
15139 | void BlueStore::BlueStoreThrottle::complete(TransContext &txc) | |
15140 | { | |
15141 | if (txc.deferred_txn) { | |
15142 | pending_deferred_ios -= 1; | |
15143 | } | |
15144 | if (txc.tracing) { | |
15145 | mono_clock::time_point now = mono_clock::now(); | |
15146 | mono_clock::duration lat = now - txc.start; | |
15147 | tracepoint( | |
15148 | bluestore, | |
15149 | transaction_total_duration, | |
15150 | txc.osr->get_sequencer_id(), | |
15151 | txc.seq, | |
15152 | ceph::to_seconds<double>(lat)); | |
15153 | } | |
15154 | } | |
15155 | #endif | |
11fdf7f2 | 15156 | |
7c673cae FG |
15157 | // DB key value Histogram |
15158 | #define KEY_SLAB 32 | |
15159 | #define VALUE_SLAB 64 | |
15160 | ||
15161 | const string prefix_onode = "o"; | |
15162 | const string prefix_onode_shard = "x"; | |
15163 | const string prefix_other = "Z"; | |
15164 | ||
15165 | int BlueStore::DBHistogram::get_key_slab(size_t sz) | |
15166 | { | |
15167 | return (sz/KEY_SLAB); | |
15168 | } | |
15169 | ||
15170 | string BlueStore::DBHistogram::get_key_slab_to_range(int slab) | |
15171 | { | |
15172 | int lower_bound = slab * KEY_SLAB; | |
15173 | int upper_bound = (slab + 1) * KEY_SLAB; | |
15174 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15175 | return ret; | |
15176 | } | |
15177 | ||
15178 | int BlueStore::DBHistogram::get_value_slab(size_t sz) | |
15179 | { | |
15180 | return (sz/VALUE_SLAB); | |
15181 | } | |
15182 | ||
15183 | string BlueStore::DBHistogram::get_value_slab_to_range(int slab) | |
15184 | { | |
15185 | int lower_bound = slab * VALUE_SLAB; | |
15186 | int upper_bound = (slab + 1) * VALUE_SLAB; | |
15187 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15188 | return ret; | |
15189 | } | |
15190 | ||
15191 | void BlueStore::DBHistogram::update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
15192 | const string &prefix, size_t key_size, size_t value_size) | |
15193 | { | |
15194 | uint32_t key_slab = get_key_slab(key_size); | |
15195 | uint32_t value_slab = get_value_slab(value_size); | |
15196 | key_hist[prefix][key_slab].count++; | |
11fdf7f2 TL |
15197 | key_hist[prefix][key_slab].max_len = |
15198 | std::max<size_t>(key_size, key_hist[prefix][key_slab].max_len); | |
7c673cae FG |
15199 | key_hist[prefix][key_slab].val_map[value_slab].count++; |
15200 | key_hist[prefix][key_slab].val_map[value_slab].max_len = | |
11fdf7f2 TL |
15201 | std::max<size_t>(value_size, |
15202 | key_hist[prefix][key_slab].val_map[value_slab].max_len); | |
7c673cae FG |
15203 | } |
15204 | ||
15205 | void BlueStore::DBHistogram::dump(Formatter *f) | |
15206 | { | |
15207 | f->open_object_section("rocksdb_value_distribution"); | |
15208 | for (auto i : value_hist) { | |
15209 | f->dump_unsigned(get_value_slab_to_range(i.first).data(), i.second); | |
15210 | } | |
15211 | f->close_section(); | |
15212 | ||
15213 | f->open_object_section("rocksdb_key_value_histogram"); | |
15214 | for (auto i : key_hist) { | |
15215 | f->dump_string("prefix", i.first); | |
15216 | f->open_object_section("key_hist"); | |
15217 | for ( auto k : i.second) { | |
15218 | f->dump_unsigned(get_key_slab_to_range(k.first).data(), k.second.count); | |
15219 | f->dump_unsigned("max_len", k.second.max_len); | |
15220 | f->open_object_section("value_hist"); | |
15221 | for ( auto j : k.second.val_map) { | |
15222 | f->dump_unsigned(get_value_slab_to_range(j.first).data(), j.second.count); | |
15223 | f->dump_unsigned("max_len", j.second.max_len); | |
15224 | } | |
15225 | f->close_section(); | |
15226 | } | |
15227 | f->close_section(); | |
15228 | } | |
15229 | f->close_section(); | |
15230 | } | |
15231 | ||
15232 | //Itrerates through the db and collects the stats | |
15233 | void BlueStore::generate_db_histogram(Formatter *f) | |
15234 | { | |
15235 | //globals | |
15236 | uint64_t num_onodes = 0; | |
15237 | uint64_t num_shards = 0; | |
15238 | uint64_t num_super = 0; | |
15239 | uint64_t num_coll = 0; | |
15240 | uint64_t num_omap = 0; | |
11fdf7f2 | 15241 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
15242 | uint64_t num_deferred = 0; |
15243 | uint64_t num_alloc = 0; | |
15244 | uint64_t num_stat = 0; | |
15245 | uint64_t num_others = 0; | |
15246 | uint64_t num_shared_shards = 0; | |
15247 | size_t max_key_size =0, max_value_size = 0; | |
15248 | uint64_t total_key_size = 0, total_value_size = 0; | |
15249 | size_t key_size = 0, value_size = 0; | |
15250 | DBHistogram hist; | |
15251 | ||
11fdf7f2 | 15252 | auto start = coarse_mono_clock::now(); |
7c673cae | 15253 | |
11fdf7f2 | 15254 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
15255 | iter->seek_to_first(); |
15256 | while (iter->valid()) { | |
15257 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
15258 | key_size = iter->key_size(); | |
15259 | value_size = iter->value_size(); | |
15260 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
15261 | max_key_size = std::max(max_key_size, key_size); |
15262 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
15263 | total_key_size += key_size; |
15264 | total_value_size += value_size; | |
15265 | ||
15266 | pair<string,string> key(iter->raw_key()); | |
15267 | ||
15268 | if (key.first == PREFIX_SUPER) { | |
15269 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
15270 | num_super++; | |
15271 | } else if (key.first == PREFIX_STAT) { | |
15272 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
15273 | num_stat++; | |
15274 | } else if (key.first == PREFIX_COLL) { | |
15275 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
15276 | num_coll++; | |
15277 | } else if (key.first == PREFIX_OBJ) { | |
15278 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
15279 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
15280 | num_onodes++; | |
15281 | } else { | |
15282 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
15283 | num_shards++; | |
15284 | } | |
15285 | } else if (key.first == PREFIX_OMAP) { | |
15286 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
15287 | num_omap++; | |
11fdf7f2 TL |
15288 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
15289 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
15290 | num_pgmeta_omap++; | |
7c673cae FG |
15291 | } else if (key.first == PREFIX_DEFERRED) { |
15292 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
15293 | num_deferred++; | |
11fdf7f2 | 15294 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
15295 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
15296 | num_alloc++; | |
15297 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
15298 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
15299 | num_shared_shards++; | |
15300 | } else { | |
15301 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
15302 | num_others++; | |
15303 | } | |
15304 | iter->next(); | |
15305 | } | |
15306 | ||
11fdf7f2 | 15307 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
15308 | f->open_object_section("rocksdb_key_value_stats"); |
15309 | f->dump_unsigned("num_onodes", num_onodes); | |
15310 | f->dump_unsigned("num_shards", num_shards); | |
15311 | f->dump_unsigned("num_super", num_super); | |
15312 | f->dump_unsigned("num_coll", num_coll); | |
15313 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 15314 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
15315 | f->dump_unsigned("num_deferred", num_deferred); |
15316 | f->dump_unsigned("num_alloc", num_alloc); | |
15317 | f->dump_unsigned("num_stat", num_stat); | |
15318 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
15319 | f->dump_unsigned("num_others", num_others); | |
15320 | f->dump_unsigned("max_key_size", max_key_size); | |
15321 | f->dump_unsigned("max_value_size", max_value_size); | |
15322 | f->dump_unsigned("total_key_size", total_key_size); | |
15323 | f->dump_unsigned("total_value_size", total_value_size); | |
15324 | f->close_section(); | |
15325 | ||
15326 | hist.dump(f); | |
15327 | ||
15328 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
15329 | ||
15330 | } | |
15331 | ||
31f18b77 | 15332 | void BlueStore::_flush_cache() |
7c673cae FG |
15333 | { |
15334 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15335 | for (auto i : onode_cache_shards) { |
15336 | i->flush(); | |
15337 | ceph_assert(i->empty()); | |
15338 | } | |
15339 | for (auto i : buffer_cache_shards) { | |
15340 | i->flush(); | |
11fdf7f2 | 15341 | ceph_assert(i->empty()); |
7c673cae FG |
15342 | } |
15343 | for (auto& p : coll_map) { | |
3efd9988 | 15344 | if (!p.second->onode_map.empty()) { |
11fdf7f2 TL |
15345 | derr << __func__ << " stray onodes on " << p.first << dendl; |
15346 | p.second->onode_map.dump<0>(cct); | |
3efd9988 FG |
15347 | } |
15348 | if (!p.second->shared_blob_set.empty()) { | |
15349 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 15350 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 15351 | } |
11fdf7f2 TL |
15352 | ceph_assert(p.second->onode_map.empty()); |
15353 | ceph_assert(p.second->shared_blob_set.empty()); | |
7c673cae FG |
15354 | } |
15355 | coll_map.clear(); | |
15356 | } | |
15357 | ||
31f18b77 FG |
15358 | // For external caller. |
15359 | // We use a best-effort policy instead, e.g., | |
15360 | // we don't care if there are still some pinned onodes/data in the cache | |
15361 | // after this command is completed. | |
11fdf7f2 | 15362 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
15363 | { |
15364 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15365 | for (auto i : onode_cache_shards) { |
15366 | i->flush(); | |
15367 | } | |
15368 | for (auto i : buffer_cache_shards) { | |
15369 | i->flush(); | |
31f18b77 | 15370 | } |
11fdf7f2 TL |
15371 | |
15372 | return 0; | |
31f18b77 FG |
15373 | } |
15374 | ||
7c673cae FG |
15375 | void BlueStore::_apply_padding(uint64_t head_pad, |
15376 | uint64_t tail_pad, | |
7c673cae FG |
15377 | bufferlist& padded) |
15378 | { | |
7c673cae | 15379 | if (head_pad) { |
224ce89b | 15380 | padded.prepend_zero(head_pad); |
7c673cae FG |
15381 | } |
15382 | if (tail_pad) { | |
15383 | padded.append_zero(tail_pad); | |
15384 | } | |
15385 | if (head_pad || tail_pad) { | |
15386 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
15387 | << " tail 0x" << tail_pad << std::dec << dendl; | |
15388 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
15389 | } | |
15390 | } | |
15391 | ||
11fdf7f2 TL |
15392 | void BlueStore::_record_onode(OnodeRef &o, KeyValueDB::Transaction &txn) |
15393 | { | |
15394 | // finalize extent_map shards | |
15395 | o->extent_map.update(txn, false); | |
15396 | if (o->extent_map.needs_reshard()) { | |
15397 | o->extent_map.reshard(db, txn); | |
15398 | o->extent_map.update(txn, true); | |
15399 | if (o->extent_map.needs_reshard()) { | |
15400 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
15401 | << dendl; | |
15402 | o->extent_map.clear_needs_reshard(); | |
15403 | } | |
15404 | logger->inc(l_bluestore_onode_reshard); | |
15405 | } | |
15406 | ||
15407 | // bound encode | |
15408 | size_t bound = 0; | |
15409 | denc(o->onode, bound); | |
15410 | o->extent_map.bound_encode_spanning_blobs(bound); | |
15411 | if (o->onode.extent_map_shards.empty()) { | |
15412 | denc(o->extent_map.inline_bl, bound); | |
15413 | } | |
15414 | ||
15415 | // encode | |
15416 | bufferlist bl; | |
15417 | unsigned onode_part, blob_part, extent_part; | |
15418 | { | |
15419 | auto p = bl.get_contiguous_appender(bound, true); | |
15420 | denc(o->onode, p); | |
15421 | onode_part = p.get_logical_offset(); | |
15422 | o->extent_map.encode_spanning_blobs(p); | |
15423 | blob_part = p.get_logical_offset() - onode_part; | |
15424 | if (o->onode.extent_map_shards.empty()) { | |
15425 | denc(o->extent_map.inline_bl, p); | |
15426 | } | |
15427 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
15428 | } | |
15429 | ||
15430 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
15431 | << " (" << onode_part << " bytes onode + " | |
15432 | << blob_part << " bytes spanning blobs + " | |
15433 | << extent_part << " bytes inline extents)" | |
15434 | << dendl; | |
15435 | ||
15436 | ||
15437 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
15438 | } | |
15439 | ||
15440 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
15441 | { | |
15442 | std::lock_guard l(qlock); | |
15443 | ||
81eedcae TL |
15444 | if (!disk_size_mismatch_alert.empty()) { |
15445 | alerts.emplace( | |
15446 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
15447 | disk_size_mismatch_alert); | |
15448 | } | |
15449 | if (!legacy_statfs_alert.empty()) { | |
15450 | alerts.emplace( | |
15451 | "BLUESTORE_LEGACY_STATFS", | |
15452 | legacy_statfs_alert); | |
15453 | } | |
11fdf7f2 TL |
15454 | if (!spillover_alert.empty() && |
15455 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
15456 | alerts.emplace( | |
15457 | "BLUEFS_SPILLOVER", | |
15458 | spillover_alert); | |
15459 | } | |
9f95a23c TL |
15460 | if (!no_per_pool_omap_alert.empty()) { |
15461 | alerts.emplace( | |
15462 | "BLUESTORE_NO_PER_POOL_OMAP", | |
15463 | no_per_pool_omap_alert); | |
15464 | } | |
11fdf7f2 TL |
15465 | string s0(failed_cmode); |
15466 | ||
15467 | if (!failed_compressors.empty()) { | |
15468 | if (!s0.empty()) { | |
15469 | s0 += ", "; | |
15470 | } | |
15471 | s0 += "unable to load:"; | |
15472 | bool first = true; | |
15473 | for (auto& s : failed_compressors) { | |
15474 | if (first) { | |
15475 | first = false; | |
15476 | } else { | |
15477 | s0 += ", "; | |
15478 | } | |
15479 | s0 += s; | |
15480 | } | |
15481 | alerts.emplace( | |
15482 | "BLUESTORE_NO_COMPRESSION", | |
15483 | s0); | |
15484 | } | |
15485 | } | |
15486 | ||
9f95a23c TL |
15487 | void BlueStore::_collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
15488 | size_t extents) | |
15489 | { | |
15490 | alloc_stats_count++; | |
15491 | alloc_stats_fragments += extents; | |
15492 | alloc_stats_size += need; | |
15493 | } | |
15494 | ||
15495 | void BlueStore::_record_allocation_stats() | |
15496 | { | |
15497 | // don't care about data consistency, | |
15498 | // fields can be partially modified while making the tuple | |
15499 | auto t0 = std::make_tuple( | |
15500 | alloc_stats_count.exchange(0), | |
15501 | alloc_stats_fragments.exchange(0), | |
15502 | alloc_stats_size.exchange(0)); | |
15503 | ||
15504 | dout(0) << " allocation stats probe " | |
15505 | << probe_count << ":" | |
15506 | << " cnt: " << std::get<0>(t0) | |
15507 | << " frags: " << std::get<1>(t0) | |
15508 | << " size: " << std::get<2>(t0) | |
15509 | << dendl; | |
15510 | ||
15511 | ||
15512 | // | |
15513 | // Keep the history for probes from the power-of-two sequence: | |
15514 | // -1, -2, -4, -8, -16 | |
15515 | // | |
15516 | size_t base = 1; | |
15517 | for (auto& t : alloc_stats_history) { | |
15518 | dout(0) << " probe -" | |
15519 | << base + (probe_count % base) << ": " | |
15520 | << std::get<0>(t) | |
15521 | << ", " << std::get<1>(t) | |
15522 | << ", " << std::get<2>(t) | |
15523 | << dendl; | |
15524 | base <<= 1; | |
15525 | } | |
15526 | dout(0) << "------------" << dendl; | |
15527 | ||
15528 | auto prev = probe_count++; | |
15529 | auto mask = (1 << alloc_stats_history.size()) - 1; | |
15530 | probe_count &= mask; | |
15531 | ||
15532 | for (size_t i = cbits(prev ^ probe_count) - 1; i > 0 ; --i) { | |
15533 | alloc_stats_history[i] = alloc_stats_history[i - 1]; | |
15534 | } | |
15535 | alloc_stats_history[0].swap(t0); | |
15536 | } | |
15537 | ||
7c673cae | 15538 | // =========================================== |
11fdf7f2 TL |
15539 | // BlueStoreRepairer |
15540 | ||
15541 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
15542 | const interval_set<uint64_t>& extents) | |
15543 | { | |
15544 | ceph_assert(granularity); // initialized | |
15545 | // can't call for the second time | |
15546 | ceph_assert(!was_filtered_out); | |
15547 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
15548 | ||
15549 | uint64_t prev_pos = 0; | |
15550 | uint64_t npos = collections_bfs.size(); | |
15551 | ||
15552 | bloom_vector collections_reduced; | |
15553 | bloom_vector objects_reduced; | |
15554 | ||
15555 | for (auto e : extents) { | |
15556 | if (e.second == 0) { | |
15557 | continue; | |
15558 | } | |
15559 | uint64_t pos = max(e.first / granularity, prev_pos); | |
15560 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
15561 | while (pos != npos && pos < end_pos) { | |
15562 | ceph_assert( collections_bfs[pos].element_count() == | |
15563 | objects_bfs[pos].element_count()); | |
15564 | if (collections_bfs[pos].element_count()) { | |
15565 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
15566 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
15567 | } | |
15568 | ++pos; | |
15569 | } | |
15570 | prev_pos = end_pos; | |
15571 | } | |
15572 | collections_reduced.swap(collections_bfs); | |
15573 | objects_reduced.swap(objects_bfs); | |
15574 | was_filtered_out = true; | |
15575 | return collections_bfs.size(); | |
15576 | } | |
15577 | ||
15578 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
15579 | const string& prefix, | |
15580 | const string& key) | |
15581 | { | |
15582 | if (!remove_key_txn) { | |
15583 | remove_key_txn = db->get_transaction(); | |
15584 | } | |
15585 | ++to_repair_cnt; | |
15586 | remove_key_txn->rmkey(prefix, key); | |
15587 | ||
15588 | return true; | |
15589 | } | |
15590 | ||
9f95a23c TL |
15591 | void BlueStoreRepairer::fix_per_pool_omap(KeyValueDB *db) |
15592 | { | |
15593 | fix_per_pool_omap_txn = db->get_transaction(); | |
15594 | ++to_repair_cnt; | |
15595 | bufferlist bl; | |
15596 | bl.append("1"); | |
15597 | fix_per_pool_omap_txn->set(PREFIX_SUPER, "per_pool_omap", bl); | |
15598 | } | |
15599 | ||
11fdf7f2 TL |
15600 | bool BlueStoreRepairer::fix_shared_blob( |
15601 | KeyValueDB *db, | |
15602 | uint64_t sbid, | |
15603 | const bufferlist* bl) | |
15604 | { | |
15605 | KeyValueDB::Transaction txn; | |
15606 | if (fix_misreferences_txn) { // reuse this txn | |
15607 | txn = fix_misreferences_txn; | |
15608 | } else { | |
15609 | if (!fix_shared_blob_txn) { | |
15610 | fix_shared_blob_txn = db->get_transaction(); | |
15611 | } | |
15612 | txn = fix_shared_blob_txn; | |
15613 | } | |
15614 | string key; | |
15615 | get_shared_blob_key(sbid, &key); | |
15616 | ||
15617 | ++to_repair_cnt; | |
15618 | if (bl) { | |
15619 | txn->set(PREFIX_SHARED_BLOB, key, *bl); | |
15620 | } else { | |
15621 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
15622 | } | |
15623 | return true; | |
15624 | } | |
15625 | ||
15626 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
15627 | const string& key, | |
15628 | const store_statfs_t& new_statfs) | |
15629 | { | |
15630 | if (!fix_statfs_txn) { | |
15631 | fix_statfs_txn = db->get_transaction(); | |
15632 | } | |
15633 | BlueStore::volatile_statfs vstatfs; | |
15634 | vstatfs = new_statfs; | |
15635 | bufferlist bl; | |
15636 | vstatfs.encode(bl); | |
15637 | ++to_repair_cnt; | |
15638 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
15639 | return true; | |
15640 | } | |
15641 | ||
15642 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
15643 | FreelistManager* fm, | |
15644 | uint64_t offset, uint64_t len) | |
15645 | { | |
15646 | if (!fix_fm_leaked_txn) { | |
15647 | fix_fm_leaked_txn = db->get_transaction(); | |
15648 | } | |
15649 | ++to_repair_cnt; | |
15650 | fm->release(offset, len, fix_fm_leaked_txn); | |
15651 | return true; | |
15652 | } | |
15653 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
15654 | FreelistManager* fm, | |
15655 | uint64_t offset, uint64_t len) | |
15656 | { | |
15657 | if (!fix_fm_false_free_txn) { | |
15658 | fix_fm_false_free_txn = db->get_transaction(); | |
15659 | } | |
15660 | ++to_repair_cnt; | |
15661 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
15662 | return true; | |
15663 | } | |
15664 | ||
15665 | bool BlueStoreRepairer::fix_bluefs_extents(std::atomic<uint64_t>& out_of_sync_flag) | |
15666 | { | |
15667 | // this is just a stub to count num of repairs properly, | |
15668 | // actual repair happens in BlueStore::_close_db_and_around() | |
15669 | // while doing _sync_bluefs_and_fm | |
15670 | ++out_of_sync_flag; | |
15671 | ++to_repair_cnt; | |
15672 | return true; | |
15673 | } | |
15674 | ||
15675 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) | |
15676 | { | |
15677 | if (misreferenced_extents.size()) { | |
15678 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
15679 | ceph_assert(n > 0); | |
15680 | if (!fix_misreferences_txn) { | |
15681 | fix_misreferences_txn = db->get_transaction(); | |
15682 | } | |
15683 | return true; | |
15684 | } | |
15685 | return false; | |
15686 | } | |
15687 | ||
15688 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
15689 | { | |
9f95a23c TL |
15690 | if (fix_per_pool_omap_txn) { |
15691 | db->submit_transaction_sync(fix_per_pool_omap_txn); | |
15692 | fix_per_pool_omap_txn = nullptr; | |
15693 | } | |
11fdf7f2 TL |
15694 | if (fix_fm_leaked_txn) { |
15695 | db->submit_transaction_sync(fix_fm_leaked_txn); | |
15696 | fix_fm_leaked_txn = nullptr; | |
15697 | } | |
15698 | if (fix_fm_false_free_txn) { | |
15699 | db->submit_transaction_sync(fix_fm_false_free_txn); | |
15700 | fix_fm_false_free_txn = nullptr; | |
15701 | } | |
15702 | if (remove_key_txn) { | |
15703 | db->submit_transaction_sync(remove_key_txn); | |
15704 | remove_key_txn = nullptr; | |
15705 | } | |
15706 | if (fix_misreferences_txn) { | |
15707 | db->submit_transaction_sync(fix_misreferences_txn); | |
15708 | fix_misreferences_txn = nullptr; | |
15709 | } | |
15710 | if (fix_shared_blob_txn) { | |
15711 | db->submit_transaction_sync(fix_shared_blob_txn); | |
15712 | fix_shared_blob_txn = nullptr; | |
15713 | } | |
15714 | ||
15715 | if (fix_statfs_txn) { | |
15716 | db->submit_transaction_sync(fix_statfs_txn); | |
15717 | fix_statfs_txn = nullptr; | |
15718 | } | |
15719 | unsigned repaired = to_repair_cnt; | |
15720 | to_repair_cnt = 0; | |
15721 | return repaired; | |
15722 | } | |
15723 | ||
15724 | // ======================================================= | |
9f95a23c TL |
15725 | // RocksDBBlueFSVolumeSelector |
15726 | ||
15727 | uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) { | |
15728 | ceph_assert(h != nullptr); | |
15729 | uint64_t hint = reinterpret_cast<uint64_t>(h); | |
15730 | uint8_t res; | |
15731 | switch (hint) { | |
15732 | case LEVEL_SLOW: | |
15733 | res = BlueFS::BDEV_SLOW; | |
15734 | if (db_avail4slow > 0) { | |
15735 | // considering statically available db space vs. | |
15736 | // - observed maximums on DB dev for DB/WAL/UNSORTED data | |
15737 | // - observed maximum spillovers | |
15738 | uint64_t max_db_use = 0; // max db usage we potentially observed | |
15739 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_WAL - LEVEL_FIRST); | |
15740 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_DB - LEVEL_FIRST); | |
15741 | // this could go to db hence using it in the estimation | |
15742 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_SLOW, LEVEL_DB - LEVEL_FIRST); | |
15743 | ||
15744 | auto db_total = l_totals[LEVEL_DB - LEVEL_FIRST]; | |
15745 | uint64_t avail = min( | |
15746 | db_avail4slow, | |
15747 | max_db_use < db_total ? db_total - max_db_use : 0); | |
15748 | ||
15749 | // considering current DB dev usage for SLOW data | |
15750 | if (avail > per_level_per_dev_usage.at(BlueFS::BDEV_DB, LEVEL_SLOW - LEVEL_FIRST)) { | |
15751 | res = BlueFS::BDEV_DB; | |
15752 | } | |
15753 | } | |
15754 | break; | |
15755 | case LEVEL_WAL: | |
15756 | res = BlueFS::BDEV_WAL; | |
15757 | break; | |
15758 | case LEVEL_DB: | |
15759 | default: | |
15760 | res = BlueFS::BDEV_DB; | |
15761 | break; | |
15762 | } | |
15763 | return res; | |
15764 | } | |
15765 | ||
15766 | void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const | |
15767 | { | |
15768 | res.emplace_back(base, l_totals[LEVEL_DB - LEVEL_FIRST]); | |
15769 | res.emplace_back(base + ".slow", l_totals[LEVEL_SLOW - LEVEL_FIRST]); | |
15770 | } | |
15771 | ||
15772 | void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(const string& dirname) const { | |
15773 | uint8_t res = LEVEL_DB; | |
15774 | if (dirname.length() > 5) { | |
15775 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
15776 | // match up with bluestore. the slow device is always the second | |
15777 | // one (when a dedicated block.db device is present and used at | |
15778 | // bdev 0). the wal device is always last. | |
15779 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
15780 | res = LEVEL_SLOW; | |
15781 | } | |
15782 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
15783 | res = LEVEL_WAL; | |
15784 | } | |
15785 | } | |
15786 | return reinterpret_cast<void*>(res); | |
15787 | } | |
15788 | ||
15789 | void RocksDBBlueFSVolumeSelector::dump(ostream& sout) { | |
15790 | auto max_x = per_level_per_dev_usage.get_max_x(); | |
15791 | auto max_y = per_level_per_dev_usage.get_max_y(); | |
15792 | sout << "RocksDBBlueFSVolumeSelector: wal_total:" << l_totals[LEVEL_WAL - LEVEL_FIRST] | |
15793 | << ", db_total:" << l_totals[LEVEL_DB - LEVEL_FIRST] | |
15794 | << ", slow_total:" << l_totals[LEVEL_SLOW - LEVEL_FIRST] | |
15795 | << ", db_avail:" << db_avail4slow << std::endl | |
15796 | << "Usage matrix:" << std::endl; | |
15797 | constexpr std::array<const char*, 7> names{ { | |
15798 | "DEV/LEV", | |
15799 | "WAL", | |
15800 | "DB", | |
15801 | "SLOW", | |
15802 | "*", | |
15803 | "*", | |
15804 | "REAL" | |
15805 | } }; | |
15806 | const size_t width = 12; | |
15807 | for (size_t i = 0; i < names.size(); ++i) { | |
15808 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15809 | sout.width(width); | |
15810 | sout << names[i]; | |
15811 | } | |
15812 | sout << std::endl; | |
15813 | for (size_t l = 0; l < max_y; l++) { | |
15814 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15815 | sout.width(width); | |
15816 | switch (l + LEVEL_FIRST) { | |
15817 | case LEVEL_WAL: | |
15818 | sout << "WAL"; break; | |
15819 | case LEVEL_DB: | |
15820 | sout << "DB"; break; | |
15821 | case LEVEL_SLOW: | |
15822 | sout << "SLOW"; break; | |
15823 | case LEVEL_MAX: | |
15824 | sout << "TOTALS"; break; | |
15825 | } | |
15826 | for (size_t d = 0; d < max_x - 1; d++) { | |
15827 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15828 | sout.width(width); | |
15829 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(d, l))); | |
15830 | } | |
15831 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15832 | sout.width(width); | |
15833 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(max_x - 1, l))) | |
15834 | << std::endl; | |
15835 | } | |
15836 | ceph_assert(max_x == per_level_per_dev_max.get_max_x()); | |
15837 | ceph_assert(max_y == per_level_per_dev_max.get_max_y()); | |
15838 | sout << "MAXIMUMS:" << std::endl; | |
15839 | for (size_t l = 0; l < max_y; l++) { | |
15840 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15841 | sout.width(width); | |
15842 | switch (l + LEVEL_FIRST) { | |
15843 | case LEVEL_WAL: | |
15844 | sout << "WAL"; break; | |
15845 | case LEVEL_DB: | |
15846 | sout << "DB"; break; | |
15847 | case LEVEL_SLOW: | |
15848 | sout << "SLOW"; break; | |
15849 | case LEVEL_MAX: | |
15850 | sout << "TOTALS"; break; | |
15851 | } | |
15852 | for (size_t d = 0; d < max_x - 1; d++) { | |
15853 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15854 | sout.width(width); | |
15855 | sout << stringify(byte_u_t(per_level_per_dev_max.at(d, l))); | |
15856 | } | |
15857 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15858 | sout.width(width); | |
15859 | sout << stringify(byte_u_t(per_level_per_dev_max.at(max_x - 1, l))); | |
15860 | if (l < max_y - 1) { | |
15861 | sout << std::endl; | |
15862 | } | |
15863 | } | |
15864 | } | |
11fdf7f2 | 15865 | |
9f95a23c | 15866 | // ======================================================= |