]>
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; | |
7c673cae | 846 | |
9f95a23c | 847 | list_t lru; |
7c673cae | 848 | |
9f95a23c | 849 | explicit LruOnodeCacheShard(CephContext *cct) : BlueStore::OnodeCacheShard(cct) {} |
7c673cae | 850 | |
f6b5b4d7 | 851 | void _add(BlueStore::Onode* o, int level) override |
9f95a23c | 852 | { |
f6b5b4d7 | 853 | if (o->put_cache()) { |
9f95a23c | 854 | (level > 0) ? lru.push_front(*o) : lru.push_back(*o); |
f6b5b4d7 TL |
855 | } else { |
856 | ++num_pinned; | |
9f95a23c | 857 | } |
f6b5b4d7 TL |
858 | ++num; // we count both pinned and unpinned entries |
859 | dout(20) << __func__ << " " << this << " " << o->oid << " added, num=" << num << dendl; | |
eafe8130 | 860 | } |
f6b5b4d7 | 861 | void _rm(BlueStore::Onode* o) override |
9f95a23c | 862 | { |
f6b5b4d7 | 863 | if (o->pop_cache()) { |
9f95a23c | 864 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
865 | } else { |
866 | ceph_assert(num_pinned); | |
867 | --num_pinned; | |
9f95a23c | 868 | } |
f6b5b4d7 TL |
869 | ceph_assert(num); |
870 | --num; | |
871 | dout(20) << __func__ << " " << this << " " << " " << o->oid << " removed, num=" << num << dendl; | |
9f95a23c | 872 | } |
f6b5b4d7 | 873 | void _pin(BlueStore::Onode* o) override |
9f95a23c | 874 | { |
9f95a23c | 875 | lru.erase(lru.iterator_to(*o)); |
f6b5b4d7 TL |
876 | ++num_pinned; |
877 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " pinned" << dendl; | |
9f95a23c | 878 | } |
f6b5b4d7 | 879 | void _unpin(BlueStore::Onode* o) override |
9f95a23c | 880 | { |
f6b5b4d7 TL |
881 | lru.push_front(*o); |
882 | ceph_assert(num_pinned); | |
883 | --num_pinned; | |
884 | dout(20) << __func__ << this << " " << " " << " " << o->oid << " unpinned" << dendl; | |
9f95a23c | 885 | } |
f6b5b4d7 | 886 | |
9f95a23c TL |
887 | void _trim_to(uint64_t new_size) override |
888 | { | |
889 | if (new_size >= lru.size()) { | |
890 | return; // don't even try | |
891 | } | |
892 | uint64_t n = lru.size() - new_size; | |
893 | auto p = lru.end(); | |
894 | ceph_assert(p != lru.begin()); | |
895 | --p; | |
f6b5b4d7 TL |
896 | ceph_assert(num >= n); |
897 | num -= n; | |
898 | while (n-- > 0) { | |
9f95a23c | 899 | BlueStore::Onode *o = &*p; |
f6b5b4d7 TL |
900 | dout(20) << __func__ << " rm " << o->oid << " " |
901 | << o->nref << " " << o->cached << " " << o->pinned << dendl; | |
9f95a23c TL |
902 | if (p != lru.begin()) { |
903 | lru.erase(p--); | |
904 | } else { | |
f6b5b4d7 | 905 | ceph_assert(n == 0); |
9f95a23c | 906 | lru.erase(p); |
9f95a23c | 907 | } |
f6b5b4d7 TL |
908 | auto pinned = !o->pop_cache(); |
909 | ceph_assert(!pinned); | |
910 | o->c->onode_map._remove(o->oid); | |
9f95a23c | 911 | } |
f6b5b4d7 TL |
912 | } |
913 | void move_pinned(OnodeCacheShard *to, BlueStore::Onode *o) override | |
914 | { | |
915 | if (to == this) { | |
916 | return; | |
917 | } | |
918 | ceph_assert(o->cached); | |
919 | ceph_assert(o->pinned); | |
920 | ceph_assert(num); | |
921 | ceph_assert(num_pinned); | |
922 | --num_pinned; | |
923 | --num; | |
924 | ++to->num_pinned; | |
925 | ++to->num; | |
9f95a23c TL |
926 | } |
927 | void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) override | |
928 | { | |
f6b5b4d7 | 929 | *onodes += num; |
9f95a23c TL |
930 | *pinned_onodes += num_pinned; |
931 | } | |
932 | }; | |
7c673cae | 933 | |
9f95a23c TL |
934 | // OnodeCacheShard |
935 | BlueStore::OnodeCacheShard *BlueStore::OnodeCacheShard::create( | |
936 | CephContext* cct, | |
937 | string type, | |
938 | PerfCounters *logger) | |
7c673cae | 939 | { |
9f95a23c TL |
940 | BlueStore::OnodeCacheShard *c = nullptr; |
941 | // Currently we only implement an LRU cache for onodes | |
942 | c = new LruOnodeCacheShard(cct); | |
943 | c->logger = logger; | |
944 | return c; | |
7c673cae FG |
945 | } |
946 | ||
9f95a23c TL |
947 | // LruBufferCacheShard |
948 | struct LruBufferCacheShard : public BlueStore::BufferCacheShard { | |
949 | typedef boost::intrusive::list< | |
950 | BlueStore::Buffer, | |
951 | boost::intrusive::member_hook< | |
952 | BlueStore::Buffer, | |
953 | boost::intrusive::list_member_hook<>, | |
954 | &BlueStore::Buffer::lru_item> > list_t; | |
955 | list_t lru; | |
956 | ||
957 | explicit LruBufferCacheShard(CephContext *cct) : BlueStore::BufferCacheShard(cct) {} | |
958 | ||
959 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override { | |
960 | if (near) { | |
961 | auto q = lru.iterator_to(*near); | |
962 | lru.insert(q, *b); | |
963 | } else if (level > 0) { | |
964 | lru.push_front(*b); | |
965 | } else { | |
966 | lru.push_back(*b); | |
7c673cae | 967 | } |
9f95a23c TL |
968 | buffer_bytes += b->length; |
969 | num = lru.size(); | |
970 | } | |
971 | void _rm(BlueStore::Buffer *b) override { | |
972 | ceph_assert(buffer_bytes >= b->length); | |
973 | buffer_bytes -= b->length; | |
974 | auto q = lru.iterator_to(*b); | |
975 | lru.erase(q); | |
976 | num = lru.size(); | |
977 | } | |
978 | void _move(BlueStore::BufferCacheShard *src, BlueStore::Buffer *b) override { | |
979 | src->_rm(b); | |
980 | _add(b, 0, nullptr); | |
981 | } | |
982 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override { | |
983 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
984 | buffer_bytes += delta; | |
985 | } | |
986 | void _touch(BlueStore::Buffer *b) override { | |
987 | auto p = lru.iterator_to(*b); | |
988 | lru.erase(p); | |
989 | lru.push_front(*b); | |
990 | num = lru.size(); | |
991 | _audit("_touch_buffer end"); | |
992 | } | |
7c673cae | 993 | |
9f95a23c TL |
994 | void _trim_to(uint64_t max) override |
995 | { | |
996 | while (buffer_bytes > max) { | |
997 | auto i = lru.rbegin(); | |
998 | if (i == lru.rend()) { | |
999 | // stop if lru is now empty | |
7c673cae FG |
1000 | break; |
1001 | } | |
1002 | ||
9f95a23c TL |
1003 | BlueStore::Buffer *b = &*i; |
1004 | ceph_assert(b->is_clean()); | |
1005 | dout(20) << __func__ << " rm " << *b << dendl; | |
1006 | b->space->_rm_buffer(this, b); | |
7c673cae | 1007 | } |
9f95a23c | 1008 | num = lru.size(); |
7c673cae | 1009 | } |
7c673cae | 1010 | |
9f95a23c TL |
1011 | void add_stats(uint64_t *extents, |
1012 | uint64_t *blobs, | |
1013 | uint64_t *buffers, | |
1014 | uint64_t *bytes) override { | |
1015 | *extents += num_extents; | |
1016 | *blobs += num_blobs; | |
1017 | *buffers += num; | |
1018 | *bytes += buffer_bytes; | |
7c673cae | 1019 | } |
9f95a23c TL |
1020 | #ifdef DEBUG_CACHE |
1021 | void _audit(const char *s) override | |
1022 | { | |
1023 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1024 | uint64_t s = 0; | |
1025 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1026 | s += i->length; | |
1027 | } | |
1028 | if (s != buffer_bytes) { | |
1029 | derr << __func__ << " buffer_size " << buffer_bytes << " actual " << s | |
1030 | << dendl; | |
1031 | for (auto i = lru.begin(); i != lru.end(); ++i) { | |
1032 | derr << __func__ << " " << *i << dendl; | |
1033 | } | |
1034 | ceph_assert(s == buffer_bytes); | |
7c673cae | 1035 | } |
9f95a23c TL |
1036 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1037 | << " ok" << dendl; | |
7c673cae | 1038 | } |
7c673cae | 1039 | #endif |
9f95a23c | 1040 | }; |
7c673cae | 1041 | |
9f95a23c TL |
1042 | // TwoQBufferCacheShard |
1043 | ||
1044 | struct TwoQBufferCacheShard : public BlueStore::BufferCacheShard { | |
1045 | typedef boost::intrusive::list< | |
1046 | BlueStore::Buffer, | |
1047 | boost::intrusive::member_hook< | |
1048 | BlueStore::Buffer, | |
1049 | boost::intrusive::list_member_hook<>, | |
1050 | &BlueStore::Buffer::lru_item> > list_t; | |
1051 | list_t hot; ///< "Am" hot buffers | |
1052 | list_t warm_in; ///< "A1in" newly warm buffers | |
1053 | list_t warm_out; ///< "A1out" empty buffers we've evicted | |
1054 | uint64_t buffer_bytes = 0; ///< bytes | |
1055 | ||
1056 | enum { | |
1057 | BUFFER_NEW = 0, | |
1058 | BUFFER_WARM_IN, ///< in warm_in | |
1059 | BUFFER_WARM_OUT, ///< in warm_out | |
1060 | BUFFER_HOT, ///< in hot | |
1061 | BUFFER_TYPE_MAX | |
1062 | }; | |
7c673cae | 1063 | |
9f95a23c | 1064 | uint64_t list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type |
7c673cae | 1065 | |
9f95a23c TL |
1066 | public: |
1067 | explicit TwoQBufferCacheShard(CephContext *cct) : BufferCacheShard(cct) {} | |
7c673cae | 1068 | |
9f95a23c TL |
1069 | void _add(BlueStore::Buffer *b, int level, BlueStore::Buffer *near) override |
1070 | { | |
1071 | dout(20) << __func__ << " level " << level << " near " << near | |
1072 | << " on " << *b | |
1073 | << " which has cache_private " << b->cache_private << dendl; | |
1074 | if (near) { | |
1075 | b->cache_private = near->cache_private; | |
1076 | switch (b->cache_private) { | |
1077 | case BUFFER_WARM_IN: | |
1078 | warm_in.insert(warm_in.iterator_to(*near), *b); | |
1079 | break; | |
1080 | case BUFFER_WARM_OUT: | |
1081 | ceph_assert(b->is_empty()); | |
1082 | warm_out.insert(warm_out.iterator_to(*near), *b); | |
1083 | break; | |
1084 | case BUFFER_HOT: | |
1085 | hot.insert(hot.iterator_to(*near), *b); | |
1086 | break; | |
1087 | default: | |
1088 | ceph_abort_msg("bad cache_private"); | |
1089 | } | |
1090 | } else if (b->cache_private == BUFFER_NEW) { | |
1091 | b->cache_private = BUFFER_WARM_IN; | |
1092 | if (level > 0) { | |
1093 | warm_in.push_front(*b); | |
1094 | } else { | |
1095 | // take caller hint to start at the back of the warm queue | |
1096 | warm_in.push_back(*b); | |
1097 | } | |
1098 | } else { | |
1099 | // we got a hint from discard | |
1100 | switch (b->cache_private) { | |
1101 | case BUFFER_WARM_IN: | |
1102 | // stay in warm_in. move to front, even though 2Q doesn't actually | |
1103 | // do this. | |
1104 | dout(20) << __func__ << " move to front of warm " << *b << dendl; | |
1105 | warm_in.push_front(*b); | |
1106 | break; | |
1107 | case BUFFER_WARM_OUT: | |
1108 | b->cache_private = BUFFER_HOT; | |
1109 | // move to hot. fall-thru | |
1110 | case BUFFER_HOT: | |
1111 | dout(20) << __func__ << " move to front of hot " << *b << dendl; | |
1112 | hot.push_front(*b); | |
1113 | break; | |
1114 | default: | |
1115 | ceph_abort_msg("bad cache_private"); | |
1116 | } | |
1117 | } | |
1118 | if (!b->is_empty()) { | |
1119 | buffer_bytes += b->length; | |
1120 | list_bytes[b->cache_private] += b->length; | |
1121 | } | |
1122 | num = hot.size() + warm_in.size(); | |
1123 | } | |
1124 | ||
1125 | void _rm(BlueStore::Buffer *b) override | |
1126 | { | |
1127 | dout(20) << __func__ << " " << *b << dendl; | |
1128 | if (!b->is_empty()) { | |
1129 | ceph_assert(buffer_bytes >= b->length); | |
1130 | buffer_bytes -= b->length; | |
1131 | ceph_assert(list_bytes[b->cache_private] >= b->length); | |
1132 | list_bytes[b->cache_private] -= b->length; | |
1133 | } | |
7c673cae FG |
1134 | switch (b->cache_private) { |
1135 | case BUFFER_WARM_IN: | |
9f95a23c | 1136 | warm_in.erase(warm_in.iterator_to(*b)); |
7c673cae FG |
1137 | break; |
1138 | case BUFFER_WARM_OUT: | |
9f95a23c | 1139 | warm_out.erase(warm_out.iterator_to(*b)); |
7c673cae FG |
1140 | break; |
1141 | case BUFFER_HOT: | |
9f95a23c | 1142 | hot.erase(hot.iterator_to(*b)); |
7c673cae FG |
1143 | break; |
1144 | default: | |
11fdf7f2 | 1145 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1146 | } |
9f95a23c TL |
1147 | num = hot.size() + warm_in.size(); |
1148 | } | |
1149 | ||
1150 | void _move(BlueStore::BufferCacheShard *srcc, BlueStore::Buffer *b) override | |
1151 | { | |
1152 | TwoQBufferCacheShard *src = static_cast<TwoQBufferCacheShard*>(srcc); | |
1153 | src->_rm(b); | |
1154 | ||
1155 | // preserve which list we're on (even if we can't preserve the order!) | |
7c673cae FG |
1156 | switch (b->cache_private) { |
1157 | case BUFFER_WARM_IN: | |
9f95a23c TL |
1158 | ceph_assert(!b->is_empty()); |
1159 | warm_in.push_back(*b); | |
7c673cae FG |
1160 | break; |
1161 | case BUFFER_WARM_OUT: | |
9f95a23c TL |
1162 | ceph_assert(b->is_empty()); |
1163 | warm_out.push_back(*b); | |
1164 | break; | |
7c673cae | 1165 | case BUFFER_HOT: |
9f95a23c TL |
1166 | ceph_assert(!b->is_empty()); |
1167 | hot.push_back(*b); | |
7c673cae FG |
1168 | break; |
1169 | default: | |
11fdf7f2 | 1170 | ceph_abort_msg("bad cache_private"); |
7c673cae | 1171 | } |
9f95a23c TL |
1172 | if (!b->is_empty()) { |
1173 | buffer_bytes += b->length; | |
1174 | list_bytes[b->cache_private] += b->length; | |
1175 | } | |
1176 | num = hot.size() + warm_in.size(); | |
7c673cae | 1177 | } |
7c673cae | 1178 | |
9f95a23c TL |
1179 | void _adjust_size(BlueStore::Buffer *b, int64_t delta) override |
1180 | { | |
1181 | dout(20) << __func__ << " delta " << delta << " on " << *b << dendl; | |
1182 | if (!b->is_empty()) { | |
1183 | ceph_assert((int64_t)buffer_bytes + delta >= 0); | |
1184 | buffer_bytes += delta; | |
1185 | ceph_assert((int64_t)list_bytes[b->cache_private] + delta >= 0); | |
1186 | list_bytes[b->cache_private] += delta; | |
1187 | } | |
7c673cae | 1188 | } |
7c673cae | 1189 | |
9f95a23c TL |
1190 | void _touch(BlueStore::Buffer *b) override { |
1191 | switch (b->cache_private) { | |
1192 | case BUFFER_WARM_IN: | |
1193 | // do nothing (somewhat counter-intuitively!) | |
1194 | break; | |
1195 | case BUFFER_WARM_OUT: | |
1196 | // move from warm_out to hot LRU | |
1197 | ceph_abort_msg("this happens via discard hint"); | |
1198 | break; | |
1199 | case BUFFER_HOT: | |
1200 | // move to front of hot LRU | |
1201 | hot.erase(hot.iterator_to(*b)); | |
1202 | hot.push_front(*b); | |
1203 | break; | |
1204 | } | |
1205 | num = hot.size() + warm_in.size(); | |
1206 | _audit("_touch_buffer end"); | |
7c673cae | 1207 | } |
7c673cae | 1208 | |
9f95a23c TL |
1209 | void _trim_to(uint64_t max) override |
1210 | { | |
1211 | if (buffer_bytes > max) { | |
1212 | uint64_t kin = max * cct->_conf->bluestore_2q_cache_kin_ratio; | |
1213 | uint64_t khot = max - kin; | |
1214 | ||
1215 | // pre-calculate kout based on average buffer size too, | |
1216 | // which is typical(the warm_in and hot lists may change later) | |
1217 | uint64_t kout = 0; | |
1218 | uint64_t buffer_num = hot.size() + warm_in.size(); | |
1219 | if (buffer_num) { | |
1220 | uint64_t avg_size = buffer_bytes / buffer_num; | |
1221 | ceph_assert(avg_size); | |
1222 | uint64_t calculated_num = max / avg_size; | |
1223 | kout = calculated_num * cct->_conf->bluestore_2q_cache_kout_ratio; | |
1224 | } | |
1225 | ||
1226 | if (list_bytes[BUFFER_HOT] < khot) { | |
1227 | // hot is small, give slack to warm_in | |
1228 | kin += khot - list_bytes[BUFFER_HOT]; | |
1229 | } else if (list_bytes[BUFFER_WARM_IN] < kin) { | |
1230 | // warm_in is small, give slack to hot | |
1231 | khot += kin - list_bytes[BUFFER_WARM_IN]; | |
1232 | } | |
1233 | ||
1234 | // adjust warm_in list | |
1235 | int64_t to_evict_bytes = list_bytes[BUFFER_WARM_IN] - kin; | |
1236 | uint64_t evicted = 0; | |
1237 | ||
1238 | while (to_evict_bytes > 0) { | |
1239 | auto p = warm_in.rbegin(); | |
1240 | if (p == warm_in.rend()) { | |
1241 | // stop if warm_in list is now empty | |
1242 | break; | |
1243 | } | |
7c673cae | 1244 | |
9f95a23c TL |
1245 | BlueStore::Buffer *b = &*p; |
1246 | ceph_assert(b->is_clean()); | |
1247 | dout(20) << __func__ << " buffer_warm_in -> out " << *b << dendl; | |
1248 | ceph_assert(buffer_bytes >= b->length); | |
1249 | buffer_bytes -= b->length; | |
1250 | ceph_assert(list_bytes[BUFFER_WARM_IN] >= b->length); | |
1251 | list_bytes[BUFFER_WARM_IN] -= b->length; | |
1252 | to_evict_bytes -= b->length; | |
1253 | evicted += b->length; | |
1254 | b->state = BlueStore::Buffer::STATE_EMPTY; | |
1255 | b->data.clear(); | |
1256 | warm_in.erase(warm_in.iterator_to(*b)); | |
1257 | warm_out.push_front(*b); | |
1258 | b->cache_private = BUFFER_WARM_OUT; | |
1259 | } | |
1260 | ||
1261 | if (evicted > 0) { | |
1262 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1263 | << " from warm_in list, done evicting warm_in buffers" | |
1264 | << dendl; | |
1265 | } | |
7c673cae | 1266 | |
9f95a23c TL |
1267 | // adjust hot list |
1268 | to_evict_bytes = list_bytes[BUFFER_HOT] - khot; | |
1269 | evicted = 0; | |
7c673cae | 1270 | |
9f95a23c TL |
1271 | while (to_evict_bytes > 0) { |
1272 | auto p = hot.rbegin(); | |
1273 | if (p == hot.rend()) { | |
1274 | // stop if hot list is now empty | |
1275 | break; | |
1276 | } | |
7c673cae | 1277 | |
9f95a23c TL |
1278 | BlueStore::Buffer *b = &*p; |
1279 | dout(20) << __func__ << " buffer_hot rm " << *b << dendl; | |
1280 | ceph_assert(b->is_clean()); | |
1281 | // adjust evict size before buffer goes invalid | |
1282 | to_evict_bytes -= b->length; | |
1283 | evicted += b->length; | |
1284 | b->space->_rm_buffer(this, b); | |
1285 | } | |
7c673cae | 1286 | |
9f95a23c TL |
1287 | if (evicted > 0) { |
1288 | dout(20) << __func__ << " evicted " << byte_u_t(evicted) | |
1289 | << " from hot list, done evicting hot buffers" | |
1290 | << dendl; | |
7c673cae FG |
1291 | } |
1292 | ||
9f95a23c TL |
1293 | // adjust warm out list too, if necessary |
1294 | int64_t n = warm_out.size() - kout; | |
1295 | while (n-- > 0) { | |
1296 | BlueStore::Buffer *b = &*warm_out.rbegin(); | |
1297 | ceph_assert(b->is_empty()); | |
1298 | dout(20) << __func__ << " buffer_warm_out rm " << *b << dendl; | |
1299 | b->space->_rm_buffer(this, b); | |
1300 | } | |
7c673cae | 1301 | } |
9f95a23c TL |
1302 | num = hot.size() + warm_in.size(); |
1303 | } | |
7c673cae | 1304 | |
9f95a23c TL |
1305 | void add_stats(uint64_t *extents, |
1306 | uint64_t *blobs, | |
1307 | uint64_t *buffers, | |
1308 | uint64_t *bytes) override { | |
1309 | *extents += num_extents; | |
1310 | *blobs += num_blobs; | |
1311 | *buffers += num; | |
1312 | *bytes += buffer_bytes; | |
1313 | } | |
7c673cae | 1314 | |
9f95a23c TL |
1315 | #ifdef DEBUG_CACHE |
1316 | void _audit(const char *s) override | |
1317 | { | |
1318 | dout(10) << __func__ << " " << when << " start" << dendl; | |
1319 | uint64_t s = 0; | |
1320 | for (auto i = hot.begin(); i != hot.end(); ++i) { | |
1321 | s += i->length; | |
7c673cae FG |
1322 | } |
1323 | ||
9f95a23c TL |
1324 | uint64_t hot_bytes = s; |
1325 | if (hot_bytes != list_bytes[BUFFER_HOT]) { | |
1326 | derr << __func__ << " hot_list_bytes " | |
1327 | << list_bytes[BUFFER_HOT] | |
1328 | << " != actual " << hot_bytes | |
1329 | << dendl; | |
1330 | ceph_assert(hot_bytes == list_bytes[BUFFER_HOT]); | |
7c673cae FG |
1331 | } |
1332 | ||
9f95a23c TL |
1333 | for (auto i = warm_in.begin(); i != warm_in.end(); ++i) { |
1334 | s += i->length; | |
7c673cae | 1335 | } |
7c673cae | 1336 | |
9f95a23c TL |
1337 | uint64_t warm_in_bytes = s - hot_bytes; |
1338 | if (warm_in_bytes != list_bytes[BUFFER_WARM_IN]) { | |
1339 | derr << __func__ << " warm_in_list_bytes " | |
1340 | << list_bytes[BUFFER_WARM_IN] | |
1341 | << " != actual " << warm_in_bytes | |
1342 | << dendl; | |
1343 | ceph_assert(warm_in_bytes == list_bytes[BUFFER_WARM_IN]); | |
7c673cae | 1344 | } |
7c673cae | 1345 | |
9f95a23c TL |
1346 | if (s != buffer_bytes) { |
1347 | derr << __func__ << " buffer_bytes " << buffer_bytes << " actual " << s | |
1348 | << dendl; | |
1349 | ceph_assert(s == buffer_bytes); | |
1350 | } | |
7c673cae | 1351 | |
9f95a23c TL |
1352 | dout(20) << __func__ << " " << when << " buffer_bytes " << buffer_bytes |
1353 | << " ok" << dendl; | |
7c673cae | 1354 | } |
9f95a23c TL |
1355 | #endif |
1356 | }; | |
7c673cae | 1357 | |
9f95a23c | 1358 | // BuferCacheShard |
7c673cae | 1359 | |
9f95a23c TL |
1360 | BlueStore::BufferCacheShard *BlueStore::BufferCacheShard::create( |
1361 | CephContext* cct, | |
1362 | string type, | |
1363 | PerfCounters *logger) | |
1364 | { | |
1365 | BufferCacheShard *c = nullptr; | |
1366 | if (type == "lru") | |
1367 | c = new LruBufferCacheShard(cct); | |
1368 | else if (type == "2q") | |
1369 | c = new TwoQBufferCacheShard(cct); | |
1370 | else | |
1371 | ceph_abort_msg("unrecognized cache type"); | |
1372 | c->logger = logger; | |
1373 | return c; | |
7c673cae | 1374 | } |
7c673cae FG |
1375 | |
1376 | // BufferSpace | |
1377 | ||
1378 | #undef dout_prefix | |
1379 | #define dout_prefix *_dout << "bluestore.BufferSpace(" << this << " in " << cache << ") " | |
1380 | ||
9f95a23c | 1381 | void BlueStore::BufferSpace::_clear(BufferCacheShard* cache) |
7c673cae FG |
1382 | { |
1383 | // note: we already hold cache->lock | |
1384 | ldout(cache->cct, 20) << __func__ << dendl; | |
1385 | while (!buffer_map.empty()) { | |
1386 | _rm_buffer(cache, buffer_map.begin()); | |
1387 | } | |
1388 | } | |
1389 | ||
9f95a23c | 1390 | int BlueStore::BufferSpace::_discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) |
7c673cae FG |
1391 | { |
1392 | // note: we already hold cache->lock | |
1393 | ldout(cache->cct, 20) << __func__ << std::hex << " 0x" << offset << "~" << length | |
1394 | << std::dec << dendl; | |
1395 | int cache_private = 0; | |
1396 | cache->_audit("discard start"); | |
1397 | auto i = _data_lower_bound(offset); | |
1398 | uint32_t end = offset + length; | |
1399 | while (i != buffer_map.end()) { | |
1400 | Buffer *b = i->second.get(); | |
1401 | if (b->offset >= end) { | |
1402 | break; | |
1403 | } | |
1404 | if (b->cache_private > cache_private) { | |
1405 | cache_private = b->cache_private; | |
1406 | } | |
1407 | if (b->offset < offset) { | |
1408 | int64_t front = offset - b->offset; | |
1409 | if (b->end() > end) { | |
1410 | // drop middle (split) | |
1411 | uint32_t tail = b->end() - end; | |
1412 | if (b->data.length()) { | |
1413 | bufferlist bl; | |
1414 | bl.substr_of(b->data, b->length - tail, tail); | |
31f18b77 FG |
1415 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1416 | nb->maybe_rebuild(); | |
1417 | _add_buffer(cache, nb, 0, b); | |
7c673cae | 1418 | } else { |
31f18b77 FG |
1419 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, tail), |
1420 | 0, b); | |
7c673cae FG |
1421 | } |
1422 | if (!b->is_writing()) { | |
9f95a23c | 1423 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1424 | } |
1425 | b->truncate(front); | |
31f18b77 | 1426 | b->maybe_rebuild(); |
7c673cae FG |
1427 | cache->_audit("discard end 1"); |
1428 | break; | |
1429 | } else { | |
1430 | // drop tail | |
1431 | if (!b->is_writing()) { | |
9f95a23c | 1432 | cache->_adjust_size(b, front - (int64_t)b->length); |
7c673cae FG |
1433 | } |
1434 | b->truncate(front); | |
31f18b77 | 1435 | b->maybe_rebuild(); |
7c673cae FG |
1436 | ++i; |
1437 | continue; | |
1438 | } | |
1439 | } | |
1440 | if (b->end() <= end) { | |
1441 | // drop entire buffer | |
1442 | _rm_buffer(cache, i++); | |
1443 | continue; | |
1444 | } | |
1445 | // drop front | |
1446 | uint32_t keep = b->end() - end; | |
1447 | if (b->data.length()) { | |
1448 | bufferlist bl; | |
1449 | bl.substr_of(b->data, b->length - keep, keep); | |
31f18b77 FG |
1450 | Buffer *nb = new Buffer(this, b->state, b->seq, end, bl); |
1451 | nb->maybe_rebuild(); | |
1452 | _add_buffer(cache, nb, 0, b); | |
7c673cae FG |
1453 | } else { |
1454 | _add_buffer(cache, new Buffer(this, b->state, b->seq, end, keep), 0, b); | |
1455 | } | |
1456 | _rm_buffer(cache, i); | |
1457 | cache->_audit("discard end 2"); | |
1458 | break; | |
1459 | } | |
1460 | return cache_private; | |
1461 | } | |
1462 | ||
1463 | void BlueStore::BufferSpace::read( | |
9f95a23c | 1464 | BufferCacheShard* cache, |
224ce89b WB |
1465 | uint32_t offset, |
1466 | uint32_t length, | |
7c673cae | 1467 | BlueStore::ready_regions_t& res, |
91327a77 AA |
1468 | interval_set<uint32_t>& res_intervals, |
1469 | int flags) | |
7c673cae | 1470 | { |
7c673cae FG |
1471 | res.clear(); |
1472 | res_intervals.clear(); | |
1473 | uint32_t want_bytes = length; | |
1474 | uint32_t end = offset + length; | |
224ce89b WB |
1475 | |
1476 | { | |
11fdf7f2 | 1477 | std::lock_guard l(cache->lock); |
224ce89b WB |
1478 | for (auto i = _data_lower_bound(offset); |
1479 | i != buffer_map.end() && offset < end && i->first < end; | |
1480 | ++i) { | |
1481 | Buffer *b = i->second.get(); | |
11fdf7f2 | 1482 | ceph_assert(b->end() > offset); |
91327a77 AA |
1483 | |
1484 | bool val = false; | |
1485 | if (flags & BYPASS_CLEAN_CACHE) | |
1486 | val = b->is_writing(); | |
1487 | else | |
1488 | val = b->is_writing() || b->is_clean(); | |
1489 | if (val) { | |
224ce89b WB |
1490 | if (b->offset < offset) { |
1491 | uint32_t skip = offset - b->offset; | |
11fdf7f2 | 1492 | uint32_t l = min(length, b->length - skip); |
224ce89b WB |
1493 | res[offset].substr_of(b->data, skip, l); |
1494 | res_intervals.insert(offset, l); | |
1495 | offset += l; | |
1496 | length -= l; | |
1497 | if (!b->is_writing()) { | |
9f95a23c | 1498 | cache->_touch(b); |
224ce89b WB |
1499 | } |
1500 | continue; | |
1501 | } | |
1502 | if (b->offset > offset) { | |
1503 | uint32_t gap = b->offset - offset; | |
1504 | if (length <= gap) { | |
1505 | break; | |
1506 | } | |
1507 | offset += gap; | |
1508 | length -= gap; | |
1509 | } | |
1510 | if (!b->is_writing()) { | |
9f95a23c | 1511 | cache->_touch(b); |
224ce89b WB |
1512 | } |
1513 | if (b->length > length) { | |
1514 | res[offset].substr_of(b->data, 0, length); | |
1515 | res_intervals.insert(offset, length); | |
7c673cae | 1516 | break; |
224ce89b WB |
1517 | } else { |
1518 | res[offset].append(b->data); | |
1519 | res_intervals.insert(offset, b->length); | |
1520 | if (b->length == length) | |
1521 | break; | |
1522 | offset += b->length; | |
1523 | length -= b->length; | |
1524 | } | |
7c673cae FG |
1525 | } |
1526 | } | |
1527 | } | |
1528 | ||
1529 | uint64_t hit_bytes = res_intervals.size(); | |
11fdf7f2 | 1530 | ceph_assert(hit_bytes <= want_bytes); |
7c673cae FG |
1531 | uint64_t miss_bytes = want_bytes - hit_bytes; |
1532 | cache->logger->inc(l_bluestore_buffer_hit_bytes, hit_bytes); | |
1533 | cache->logger->inc(l_bluestore_buffer_miss_bytes, miss_bytes); | |
1534 | } | |
1535 | ||
9f95a23c | 1536 | void BlueStore::BufferSpace::_finish_write(BufferCacheShard* cache, uint64_t seq) |
7c673cae | 1537 | { |
7c673cae FG |
1538 | auto i = writing.begin(); |
1539 | while (i != writing.end()) { | |
1540 | if (i->seq > seq) { | |
1541 | break; | |
1542 | } | |
1543 | if (i->seq < seq) { | |
1544 | ++i; | |
1545 | continue; | |
1546 | } | |
1547 | ||
1548 | Buffer *b = &*i; | |
11fdf7f2 | 1549 | ceph_assert(b->is_writing()); |
7c673cae FG |
1550 | |
1551 | if (b->flags & Buffer::FLAG_NOCACHE) { | |
1552 | writing.erase(i++); | |
1553 | ldout(cache->cct, 20) << __func__ << " discard " << *b << dendl; | |
1554 | buffer_map.erase(b->offset); | |
1555 | } else { | |
1556 | b->state = Buffer::STATE_CLEAN; | |
1557 | writing.erase(i++); | |
31f18b77 FG |
1558 | b->maybe_rebuild(); |
1559 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
9f95a23c | 1560 | cache->_add(b, 1, nullptr); |
7c673cae FG |
1561 | ldout(cache->cct, 20) << __func__ << " added " << *b << dendl; |
1562 | } | |
1563 | } | |
9f95a23c | 1564 | cache->_trim(); |
7c673cae FG |
1565 | cache->_audit("finish_write end"); |
1566 | } | |
1567 | ||
9f95a23c | 1568 | void BlueStore::BufferSpace::split(BufferCacheShard* cache, size_t pos, BlueStore::BufferSpace &r) |
7c673cae | 1569 | { |
11fdf7f2 | 1570 | std::lock_guard lk(cache->lock); |
7c673cae FG |
1571 | if (buffer_map.empty()) |
1572 | return; | |
1573 | ||
1574 | auto p = --buffer_map.end(); | |
1575 | while (true) { | |
1576 | if (p->second->end() <= pos) | |
1577 | break; | |
1578 | ||
1579 | if (p->second->offset < pos) { | |
1580 | ldout(cache->cct, 30) << __func__ << " cut " << *p->second << dendl; | |
1581 | size_t left = pos - p->second->offset; | |
1582 | size_t right = p->second->length - left; | |
1583 | if (p->second->data.length()) { | |
1584 | bufferlist bl; | |
1585 | bl.substr_of(p->second->data, left, right); | |
1586 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, bl), | |
1587 | 0, p->second.get()); | |
1588 | } else { | |
1589 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, 0, right), | |
1590 | 0, p->second.get()); | |
1591 | } | |
9f95a23c | 1592 | cache->_adjust_size(p->second.get(), -right); |
7c673cae FG |
1593 | p->second->truncate(left); |
1594 | break; | |
1595 | } | |
1596 | ||
11fdf7f2 | 1597 | ceph_assert(p->second->end() > pos); |
7c673cae FG |
1598 | ldout(cache->cct, 30) << __func__ << " move " << *p->second << dendl; |
1599 | if (p->second->data.length()) { | |
1600 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1601 | p->second->offset - pos, p->second->data), | |
1602 | 0, p->second.get()); | |
1603 | } else { | |
1604 | r._add_buffer(cache, new Buffer(&r, p->second->state, p->second->seq, | |
1605 | p->second->offset - pos, p->second->length), | |
1606 | 0, p->second.get()); | |
1607 | } | |
1608 | if (p == buffer_map.begin()) { | |
1609 | _rm_buffer(cache, p); | |
1610 | break; | |
1611 | } else { | |
1612 | _rm_buffer(cache, p--); | |
1613 | } | |
1614 | } | |
11fdf7f2 | 1615 | ceph_assert(writing.empty()); |
9f95a23c | 1616 | cache->_trim(); |
7c673cae FG |
1617 | } |
1618 | ||
1619 | // OnodeSpace | |
1620 | ||
1621 | #undef dout_prefix | |
1622 | #define dout_prefix *_dout << "bluestore.OnodeSpace(" << this << " in " << cache << ") " | |
1623 | ||
f6b5b4d7 TL |
1624 | BlueStore::OnodeRef BlueStore::OnodeSpace::add(const ghobject_t& oid, |
1625 | OnodeRef& o) | |
7c673cae | 1626 | { |
11fdf7f2 | 1627 | std::lock_guard l(cache->lock); |
7c673cae FG |
1628 | auto p = onode_map.find(oid); |
1629 | if (p != onode_map.end()) { | |
1630 | ldout(cache->cct, 30) << __func__ << " " << oid << " " << o | |
1631 | << " raced, returning existing " << p->second | |
1632 | << dendl; | |
1633 | return p->second; | |
1634 | } | |
f6b5b4d7 | 1635 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << o << dendl; |
7c673cae | 1636 | onode_map[oid] = o; |
f6b5b4d7 | 1637 | cache->_add(o.get(), 1); |
9f95a23c | 1638 | cache->_trim(); |
7c673cae FG |
1639 | return o; |
1640 | } | |
1641 | ||
f6b5b4d7 TL |
1642 | void BlueStore::OnodeSpace::_remove(const ghobject_t& oid) |
1643 | { | |
1644 | ldout(cache->cct, 20) << __func__ << " " << oid << " " << dendl; | |
1645 | onode_map.erase(oid); | |
1646 | } | |
1647 | ||
7c673cae FG |
1648 | BlueStore::OnodeRef BlueStore::OnodeSpace::lookup(const ghobject_t& oid) |
1649 | { | |
7c673cae | 1650 | ldout(cache->cct, 30) << __func__ << dendl; |
224ce89b WB |
1651 | OnodeRef o; |
1652 | bool hit = false; | |
1653 | ||
1654 | { | |
11fdf7f2 | 1655 | std::lock_guard l(cache->lock); |
224ce89b WB |
1656 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator p = onode_map.find(oid); |
1657 | if (p == onode_map.end()) { | |
1658 | ldout(cache->cct, 30) << __func__ << " " << oid << " miss" << dendl; | |
1659 | } else { | |
1660 | ldout(cache->cct, 30) << __func__ << " " << oid << " hit " << p->second | |
f6b5b4d7 TL |
1661 | << " " << p->second->nref |
1662 | << " " << p->second->cached | |
1663 | << " " << p->second->pinned | |
224ce89b | 1664 | << dendl; |
f6b5b4d7 TL |
1665 | // This will pin onode and implicitly touch the cache when Onode |
1666 | // eventually will become unpinned | |
224ce89b | 1667 | o = p->second; |
f6b5b4d7 TL |
1668 | ceph_assert(!o->cached || o->pinned); |
1669 | ||
1670 | hit = true; | |
224ce89b WB |
1671 | } |
1672 | } | |
1673 | ||
1674 | if (hit) { | |
1675 | cache->logger->inc(l_bluestore_onode_hits); | |
1676 | } else { | |
7c673cae | 1677 | cache->logger->inc(l_bluestore_onode_misses); |
7c673cae | 1678 | } |
224ce89b | 1679 | return o; |
7c673cae FG |
1680 | } |
1681 | ||
1682 | void BlueStore::OnodeSpace::clear() | |
1683 | { | |
11fdf7f2 | 1684 | std::lock_guard l(cache->lock); |
f6b5b4d7 | 1685 | ldout(cache->cct, 10) << __func__ << " " << onode_map.size()<< dendl; |
7c673cae | 1686 | for (auto &p : onode_map) { |
f6b5b4d7 | 1687 | cache->_rm(p.second.get()); |
7c673cae FG |
1688 | } |
1689 | onode_map.clear(); | |
1690 | } | |
1691 | ||
1692 | bool BlueStore::OnodeSpace::empty() | |
1693 | { | |
11fdf7f2 | 1694 | std::lock_guard l(cache->lock); |
7c673cae FG |
1695 | return onode_map.empty(); |
1696 | } | |
1697 | ||
1698 | void BlueStore::OnodeSpace::rename( | |
1699 | OnodeRef& oldo, | |
1700 | const ghobject_t& old_oid, | |
1701 | const ghobject_t& new_oid, | |
31f18b77 | 1702 | const mempool::bluestore_cache_other::string& new_okey) |
7c673cae | 1703 | { |
11fdf7f2 | 1704 | std::lock_guard l(cache->lock); |
7c673cae FG |
1705 | ldout(cache->cct, 30) << __func__ << " " << old_oid << " -> " << new_oid |
1706 | << dendl; | |
1707 | ceph::unordered_map<ghobject_t,OnodeRef>::iterator po, pn; | |
1708 | po = onode_map.find(old_oid); | |
1709 | pn = onode_map.find(new_oid); | |
11fdf7f2 | 1710 | ceph_assert(po != pn); |
7c673cae | 1711 | |
11fdf7f2 | 1712 | ceph_assert(po != onode_map.end()); |
7c673cae FG |
1713 | if (pn != onode_map.end()) { |
1714 | ldout(cache->cct, 30) << __func__ << " removing target " << pn->second | |
1715 | << dendl; | |
f6b5b4d7 | 1716 | cache->_rm(pn->second.get()); |
7c673cae FG |
1717 | onode_map.erase(pn); |
1718 | } | |
1719 | OnodeRef o = po->second; | |
1720 | ||
1721 | // install a non-existent onode at old location | |
1722 | oldo.reset(new Onode(o->c, old_oid, o->key)); | |
1723 | po->second = oldo; | |
f6b5b4d7 TL |
1724 | cache->_add(oldo.get(), 1); |
1725 | // add at new position and fix oid, key. | |
1726 | // This will pin 'o' and implicitly touch cache | |
1727 | // when it will eventually become unpinned | |
7c673cae | 1728 | onode_map.insert(make_pair(new_oid, o)); |
f6b5b4d7 TL |
1729 | ceph_assert(o->pinned); |
1730 | ||
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) { | |
f6b5b4d7 TL |
1752 | ldout(cct, LogLevelV) << i.first << " : " << i.second |
1753 | << " " << i.second->nref | |
1754 | << " " << i.second->cached | |
1755 | << " " << i.second->pinned | |
1756 | << dendl; | |
3efd9988 FG |
1757 | } |
1758 | } | |
7c673cae FG |
1759 | |
1760 | // SharedBlob | |
1761 | ||
1762 | #undef dout_prefix | |
1763 | #define dout_prefix *_dout << "bluestore.sharedblob(" << this << ") " | |
9f95a23c TL |
1764 | #undef dout_context |
1765 | #define dout_context coll->store->cct | |
7c673cae | 1766 | |
9f95a23c | 1767 | void BlueStore::SharedBlob::dump(Formatter* f) const |
7c673cae | 1768 | { |
9f95a23c TL |
1769 | f->dump_bool("loaded", loaded); |
1770 | if (loaded) { | |
1771 | persistent->dump(f); | |
1772 | } else { | |
1773 | f->dump_unsigned("sbid_unloaded", sbid_unloaded); | |
1774 | } | |
1775 | } | |
1776 | ||
1777 | ostream& operator<<(ostream& out, const BlueStore::SharedBlob& sb) | |
1778 | { | |
1779 | out << "SharedBlob(" << &sb; | |
1780 | ||
7c673cae FG |
1781 | if (sb.loaded) { |
1782 | out << " loaded " << *sb.persistent; | |
1783 | } else { | |
1784 | out << " sbid 0x" << std::hex << sb.sbid_unloaded << std::dec; | |
1785 | } | |
1786 | return out << ")"; | |
1787 | } | |
1788 | ||
1789 | BlueStore::SharedBlob::SharedBlob(uint64_t i, Collection *_coll) | |
1790 | : coll(_coll), sbid_unloaded(i) | |
1791 | { | |
11fdf7f2 | 1792 | ceph_assert(sbid_unloaded > 0); |
7c673cae FG |
1793 | if (get_cache()) { |
1794 | get_cache()->add_blob(); | |
1795 | } | |
1796 | } | |
1797 | ||
1798 | BlueStore::SharedBlob::~SharedBlob() | |
1799 | { | |
7c673cae FG |
1800 | if (loaded && persistent) { |
1801 | delete persistent; | |
1802 | } | |
1803 | } | |
1804 | ||
1805 | void BlueStore::SharedBlob::put() | |
1806 | { | |
1807 | if (--nref == 0) { | |
9f95a23c TL |
1808 | dout(20) << __func__ << " " << this |
1809 | << " removing self from set " << get_parent() | |
1810 | << dendl; | |
1adf2230 AA |
1811 | again: |
1812 | auto coll_snap = coll; | |
1813 | if (coll_snap) { | |
11fdf7f2 | 1814 | std::lock_guard l(coll_snap->cache->lock); |
1adf2230 AA |
1815 | if (coll_snap != coll) { |
1816 | goto again; | |
1817 | } | |
91327a77 AA |
1818 | if (!coll_snap->shared_blob_set.remove(this, true)) { |
1819 | // race with lookup | |
1820 | return; | |
1821 | } | |
1adf2230 AA |
1822 | bc._clear(coll_snap->cache); |
1823 | coll_snap->cache->rm_blob(); | |
7c673cae | 1824 | } |
28e407b8 | 1825 | delete this; |
7c673cae FG |
1826 | } |
1827 | } | |
1828 | ||
1829 | void BlueStore::SharedBlob::get_ref(uint64_t offset, uint32_t length) | |
1830 | { | |
11fdf7f2 | 1831 | ceph_assert(persistent); |
7c673cae FG |
1832 | persistent->ref_map.get(offset, length); |
1833 | } | |
1834 | ||
1835 | void BlueStore::SharedBlob::put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 1836 | PExtentVector *r, |
11fdf7f2 | 1837 | bool *unshare) |
7c673cae | 1838 | { |
11fdf7f2 TL |
1839 | ceph_assert(persistent); |
1840 | persistent->ref_map.put(offset, length, r, | |
1841 | unshare && !*unshare ? unshare : nullptr); | |
7c673cae FG |
1842 | } |
1843 | ||
f64942e4 AA |
1844 | void BlueStore::SharedBlob::finish_write(uint64_t seq) |
1845 | { | |
1846 | while (true) { | |
9f95a23c | 1847 | BufferCacheShard *cache = coll->cache; |
11fdf7f2 | 1848 | std::lock_guard l(cache->lock); |
f64942e4 | 1849 | if (coll->cache != cache) { |
9f95a23c TL |
1850 | dout(20) << __func__ |
1851 | << " raced with sb cache update, was " << cache | |
1852 | << ", now " << coll->cache << ", retrying" | |
1853 | << dendl; | |
f64942e4 AA |
1854 | continue; |
1855 | } | |
1856 | bc._finish_write(cache, seq); | |
1857 | break; | |
1858 | } | |
1859 | } | |
1860 | ||
3efd9988 FG |
1861 | // SharedBlobSet |
1862 | ||
1863 | #undef dout_prefix | |
1864 | #define dout_prefix *_dout << "bluestore.sharedblobset(" << this << ") " | |
1865 | ||
11fdf7f2 TL |
1866 | template <int LogLevelV = 30> |
1867 | void BlueStore::SharedBlobSet::dump(CephContext *cct) | |
3efd9988 | 1868 | { |
11fdf7f2 | 1869 | std::lock_guard l(lock); |
3efd9988 | 1870 | for (auto& i : sb_map) { |
11fdf7f2 | 1871 | ldout(cct, LogLevelV) << i.first << " : " << *i.second << dendl; |
3efd9988 FG |
1872 | } |
1873 | } | |
1874 | ||
7c673cae FG |
1875 | // Blob |
1876 | ||
1877 | #undef dout_prefix | |
1878 | #define dout_prefix *_dout << "bluestore.blob(" << this << ") " | |
1879 | ||
9f95a23c TL |
1880 | void BlueStore::Blob::dump(Formatter* f) const |
1881 | { | |
1882 | if (is_spanning()) { | |
1883 | f->dump_unsigned("spanning_id ", id); | |
1884 | } | |
1885 | blob.dump(f); | |
1886 | if (shared_blob) { | |
1887 | f->dump_object("shared", *shared_blob); | |
1888 | } | |
1889 | } | |
1890 | ||
7c673cae FG |
1891 | ostream& operator<<(ostream& out, const BlueStore::Blob& b) |
1892 | { | |
1893 | out << "Blob(" << &b; | |
1894 | if (b.is_spanning()) { | |
1895 | out << " spanning " << b.id; | |
1896 | } | |
35e4c445 FG |
1897 | out << " " << b.get_blob() << " " << b.get_blob_use_tracker(); |
1898 | if (b.shared_blob) { | |
1899 | out << " " << *b.shared_blob; | |
1900 | } else { | |
1901 | out << " (shared_blob=NULL)"; | |
1902 | } | |
1903 | out << ")"; | |
7c673cae FG |
1904 | return out; |
1905 | } | |
1906 | ||
1907 | void BlueStore::Blob::discard_unallocated(Collection *coll) | |
1908 | { | |
224ce89b | 1909 | if (get_blob().is_shared()) { |
7c673cae FG |
1910 | return; |
1911 | } | |
224ce89b | 1912 | if (get_blob().is_compressed()) { |
7c673cae FG |
1913 | bool discard = false; |
1914 | bool all_invalid = true; | |
224ce89b | 1915 | for (auto e : get_blob().get_extents()) { |
7c673cae FG |
1916 | if (!e.is_valid()) { |
1917 | discard = true; | |
1918 | } else { | |
1919 | all_invalid = false; | |
1920 | } | |
1921 | } | |
11fdf7f2 | 1922 | ceph_assert(discard == all_invalid); // in case of compressed blob all |
7c673cae FG |
1923 | // or none pextents are invalid. |
1924 | if (discard) { | |
224ce89b WB |
1925 | shared_blob->bc.discard(shared_blob->get_cache(), 0, |
1926 | get_blob().get_logical_length()); | |
7c673cae FG |
1927 | } |
1928 | } else { | |
1929 | size_t pos = 0; | |
224ce89b | 1930 | for (auto e : get_blob().get_extents()) { |
7c673cae | 1931 | if (!e.is_valid()) { |
9f95a23c TL |
1932 | dout(20) << __func__ << " 0x" << std::hex << pos |
1933 | << "~" << e.length | |
1934 | << std::dec << dendl; | |
7c673cae FG |
1935 | shared_blob->bc.discard(shared_blob->get_cache(), pos, e.length); |
1936 | } | |
1937 | pos += e.length; | |
1938 | } | |
224ce89b WB |
1939 | if (get_blob().can_prune_tail()) { |
1940 | dirty_blob().prune_tail(); | |
1941 | used_in_blob.prune_tail(get_blob().get_ondisk_length()); | |
224ce89b | 1942 | dout(20) << __func__ << " pruned tail, now " << get_blob() << dendl; |
7c673cae FG |
1943 | } |
1944 | } | |
1945 | } | |
1946 | ||
1947 | void BlueStore::Blob::get_ref( | |
1948 | Collection *coll, | |
1949 | uint32_t offset, | |
1950 | uint32_t length) | |
1951 | { | |
1952 | // Caller has to initialize Blob's logical length prior to increment | |
1953 | // references. Otherwise one is neither unable to determine required | |
1954 | // amount of counters in case of per-au tracking nor obtain min_release_size | |
1955 | // for single counter mode. | |
11fdf7f2 | 1956 | ceph_assert(get_blob().get_logical_length() != 0); |
7c673cae FG |
1957 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
1958 | << std::dec << " " << *this << dendl; | |
1959 | ||
1960 | if (used_in_blob.is_empty()) { | |
1961 | uint32_t min_release_size = | |
224ce89b WB |
1962 | get_blob().get_release_size(coll->store->min_alloc_size); |
1963 | uint64_t l = get_blob().get_logical_length(); | |
1964 | dout(20) << __func__ << " init 0x" << std::hex << l << ", " | |
1965 | << min_release_size << std::dec << dendl; | |
7c673cae FG |
1966 | used_in_blob.init(l, min_release_size); |
1967 | } | |
1968 | used_in_blob.get( | |
1969 | offset, | |
1970 | length); | |
1971 | } | |
1972 | ||
1973 | bool BlueStore::Blob::put_ref( | |
1974 | Collection *coll, | |
1975 | uint32_t offset, | |
1976 | uint32_t length, | |
1977 | PExtentVector *r) | |
1978 | { | |
1979 | PExtentVector logical; | |
1980 | ||
7c673cae FG |
1981 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length |
1982 | << std::dec << " " << *this << dendl; | |
1983 | ||
1984 | bool empty = used_in_blob.put( | |
1985 | offset, | |
1986 | length, | |
1987 | &logical); | |
1988 | r->clear(); | |
1989 | // nothing to release | |
1990 | if (!empty && logical.empty()) { | |
1991 | return false; | |
1992 | } | |
1993 | ||
1994 | bluestore_blob_t& b = dirty_blob(); | |
1995 | return b.release_extents(empty, logical, r); | |
1996 | } | |
1997 | ||
224ce89b | 1998 | bool BlueStore::Blob::can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
1999 | uint32_t target_blob_size, |
2000 | uint32_t b_offset, | |
2001 | uint32_t *length0) { | |
11fdf7f2 TL |
2002 | ceph_assert(min_alloc_size); |
2003 | ceph_assert(target_blob_size); | |
7c673cae FG |
2004 | if (!get_blob().is_mutable()) { |
2005 | return false; | |
2006 | } | |
2007 | ||
2008 | uint32_t length = *length0; | |
2009 | uint32_t end = b_offset + length; | |
2010 | ||
2011 | // Currently for the sake of simplicity we omit blob reuse if data is | |
2012 | // unaligned with csum chunk. Later we can perform padding if needed. | |
2013 | if (get_blob().has_csum() && | |
2014 | ((b_offset % get_blob().get_csum_chunk_size()) != 0 || | |
2015 | (end % get_blob().get_csum_chunk_size()) != 0)) { | |
2016 | return false; | |
2017 | } | |
2018 | ||
2019 | auto blen = get_blob().get_logical_length(); | |
2020 | uint32_t new_blen = blen; | |
2021 | ||
2022 | // make sure target_blob_size isn't less than current blob len | |
11fdf7f2 | 2023 | target_blob_size = std::max(blen, target_blob_size); |
7c673cae FG |
2024 | |
2025 | if (b_offset >= blen) { | |
224ce89b WB |
2026 | // new data totally stands out of the existing blob |
2027 | new_blen = end; | |
7c673cae | 2028 | } else { |
224ce89b | 2029 | // new data overlaps with the existing blob |
11fdf7f2 | 2030 | new_blen = std::max(blen, end); |
224ce89b WB |
2031 | |
2032 | uint32_t overlap = 0; | |
2033 | if (new_blen > blen) { | |
2034 | overlap = blen - b_offset; | |
2035 | } else { | |
2036 | overlap = length; | |
2037 | } | |
2038 | ||
2039 | if (!get_blob().is_unallocated(b_offset, overlap)) { | |
2040 | // abort if any piece of the overlap has already been allocated | |
2041 | return false; | |
7c673cae FG |
2042 | } |
2043 | } | |
224ce89b | 2044 | |
7c673cae FG |
2045 | if (new_blen > blen) { |
2046 | int64_t overflow = int64_t(new_blen) - target_blob_size; | |
2047 | // Unable to decrease the provided length to fit into max_blob_size | |
2048 | if (overflow >= length) { | |
2049 | return false; | |
2050 | } | |
2051 | ||
2052 | // FIXME: in some cases we could reduce unused resolution | |
2053 | if (get_blob().has_unused()) { | |
2054 | return false; | |
2055 | } | |
2056 | ||
2057 | if (overflow > 0) { | |
2058 | new_blen -= overflow; | |
2059 | length -= overflow; | |
2060 | *length0 = length; | |
2061 | } | |
224ce89b | 2062 | |
7c673cae FG |
2063 | if (new_blen > blen) { |
2064 | dirty_blob().add_tail(new_blen); | |
2065 | used_in_blob.add_tail(new_blen, | |
224ce89b | 2066 | get_blob().get_release_size(min_alloc_size)); |
7c673cae FG |
2067 | } |
2068 | } | |
2069 | return true; | |
2070 | } | |
2071 | ||
2072 | void BlueStore::Blob::split(Collection *coll, uint32_t blob_offset, Blob *r) | |
2073 | { | |
7c673cae FG |
2074 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec |
2075 | << " start " << *this << dendl; | |
11fdf7f2 TL |
2076 | ceph_assert(blob.can_split()); |
2077 | ceph_assert(used_in_blob.can_split()); | |
7c673cae FG |
2078 | bluestore_blob_t &lb = dirty_blob(); |
2079 | bluestore_blob_t &rb = r->dirty_blob(); | |
2080 | ||
2081 | used_in_blob.split( | |
2082 | blob_offset, | |
2083 | &(r->used_in_blob)); | |
2084 | ||
2085 | lb.split(blob_offset, rb); | |
2086 | shared_blob->bc.split(shared_blob->get_cache(), blob_offset, r->shared_blob->bc); | |
2087 | ||
2088 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2089 | << " finish " << *this << dendl; | |
2090 | dout(10) << __func__ << " 0x" << std::hex << blob_offset << std::dec | |
2091 | << " and " << *r << dendl; | |
2092 | } | |
2093 | ||
2094 | #ifndef CACHE_BLOB_BL | |
2095 | void BlueStore::Blob::decode( | |
2096 | Collection *coll, | |
11fdf7f2 | 2097 | bufferptr::const_iterator& p, |
7c673cae FG |
2098 | uint64_t struct_v, |
2099 | uint64_t* sbid, | |
2100 | bool include_ref_map) | |
2101 | { | |
2102 | denc(blob, p, struct_v); | |
2103 | if (blob.is_shared()) { | |
2104 | denc(*sbid, p); | |
2105 | } | |
2106 | if (include_ref_map) { | |
2107 | if (struct_v > 1) { | |
2108 | used_in_blob.decode(p); | |
2109 | } else { | |
2110 | used_in_blob.clear(); | |
2111 | bluestore_extent_ref_map_t legacy_ref_map; | |
2112 | legacy_ref_map.decode(p); | |
2113 | for (auto r : legacy_ref_map.ref_map) { | |
2114 | get_ref( | |
2115 | coll, | |
2116 | r.first, | |
2117 | r.second.refs * r.second.length); | |
2118 | } | |
2119 | } | |
2120 | } | |
2121 | } | |
2122 | #endif | |
2123 | ||
2124 | // Extent | |
2125 | ||
9f95a23c TL |
2126 | void BlueStore::Extent::dump(Formatter* f) const |
2127 | { | |
2128 | f->dump_unsigned("logical_offset", logical_offset); | |
2129 | f->dump_unsigned("length", length); | |
2130 | f->dump_unsigned("blob_offset", blob_offset); | |
2131 | f->dump_object("blob", *blob); | |
2132 | } | |
2133 | ||
7c673cae FG |
2134 | ostream& operator<<(ostream& out, const BlueStore::Extent& e) |
2135 | { | |
2136 | return out << std::hex << "0x" << e.logical_offset << "~" << e.length | |
2137 | << ": 0x" << e.blob_offset << "~" << e.length << std::dec | |
2138 | << " " << *e.blob; | |
2139 | } | |
2140 | ||
2141 | // OldExtent | |
2142 | BlueStore::OldExtent* BlueStore::OldExtent::create(CollectionRef c, | |
2143 | uint32_t lo, | |
2144 | uint32_t o, | |
2145 | uint32_t l, | |
2146 | BlobRef& b) { | |
2147 | OldExtent* oe = new OldExtent(lo, o, l, b); | |
2148 | b->put_ref(c.get(), o, l, &(oe->r)); | |
2149 | oe->blob_empty = b->get_referenced_bytes() == 0; | |
2150 | return oe; | |
2151 | } | |
2152 | ||
2153 | // ExtentMap | |
2154 | ||
2155 | #undef dout_prefix | |
2156 | #define dout_prefix *_dout << "bluestore.extentmap(" << this << ") " | |
9f95a23c TL |
2157 | #undef dout_context |
2158 | #define dout_context onode->c->store->cct | |
7c673cae FG |
2159 | |
2160 | BlueStore::ExtentMap::ExtentMap(Onode *o) | |
2161 | : onode(o), | |
2162 | inline_bl( | |
2163 | o->c->store->cct->_conf->bluestore_extent_map_inline_shard_prealloc_size) { | |
2164 | } | |
2165 | ||
9f95a23c TL |
2166 | void BlueStore::ExtentMap::dump(Formatter* f) const |
2167 | { | |
2168 | f->open_array_section("extents"); | |
2169 | ||
2170 | for (auto& e : extent_map) { | |
2171 | f->dump_object("extent", e); | |
2172 | } | |
2173 | f->close_section(); | |
2174 | } | |
2175 | ||
11fdf7f2 TL |
2176 | void BlueStore::ExtentMap::dup(BlueStore* b, TransContext* txc, |
2177 | CollectionRef& c, OnodeRef& oldo, OnodeRef& newo, uint64_t& srcoff, | |
2178 | uint64_t& length, uint64_t& dstoff) { | |
2179 | ||
2180 | auto cct = onode->c->store->cct; | |
2181 | bool inject_21040 = | |
2182 | cct->_conf->bluestore_debug_inject_bug21040; | |
2183 | vector<BlobRef> id_to_blob(oldo->extent_map.extent_map.size()); | |
2184 | for (auto& e : oldo->extent_map.extent_map) { | |
2185 | e.blob->last_encoded_id = -1; | |
2186 | } | |
2187 | ||
2188 | int n = 0; | |
2189 | uint64_t end = srcoff + length; | |
2190 | uint32_t dirty_range_begin = 0; | |
2191 | uint32_t dirty_range_end = 0; | |
2192 | bool src_dirty = false; | |
2193 | for (auto ep = oldo->extent_map.seek_lextent(srcoff); | |
2194 | ep != oldo->extent_map.extent_map.end(); | |
2195 | ++ep) { | |
2196 | auto& e = *ep; | |
2197 | if (e.logical_offset >= end) { | |
2198 | break; | |
2199 | } | |
2200 | dout(20) << __func__ << " src " << e << dendl; | |
2201 | BlobRef cb; | |
2202 | bool blob_duped = true; | |
2203 | if (e.blob->last_encoded_id >= 0) { | |
2204 | cb = id_to_blob[e.blob->last_encoded_id]; | |
2205 | blob_duped = false; | |
2206 | } else { | |
2207 | // dup the blob | |
2208 | const bluestore_blob_t& blob = e.blob->get_blob(); | |
2209 | // make sure it is shared | |
2210 | if (!blob.is_shared()) { | |
2211 | c->make_blob_shared(b->_assign_blobid(txc), e.blob); | |
2212 | if (!inject_21040 && !src_dirty) { | |
2213 | src_dirty = true; | |
2214 | dirty_range_begin = e.logical_offset; | |
2215 | } else if (inject_21040 && | |
2216 | dirty_range_begin == 0 && dirty_range_end == 0) { | |
2217 | dirty_range_begin = e.logical_offset; | |
2218 | } | |
2219 | ceph_assert(e.logical_end() > 0); | |
2220 | // -1 to exclude next potential shard | |
2221 | dirty_range_end = e.logical_end() - 1; | |
2222 | } else { | |
2223 | c->load_shared_blob(e.blob->shared_blob); | |
2224 | } | |
2225 | cb = new Blob(); | |
2226 | e.blob->last_encoded_id = n; | |
2227 | id_to_blob[n] = cb; | |
2228 | e.blob->dup(*cb); | |
2229 | // bump the extent refs on the copied blob's extents | |
2230 | for (auto p : blob.get_extents()) { | |
2231 | if (p.is_valid()) { | |
2232 | e.blob->shared_blob->get_ref(p.offset, p.length); | |
2233 | } | |
2234 | } | |
2235 | txc->write_shared_blob(e.blob->shared_blob); | |
2236 | dout(20) << __func__ << " new " << *cb << dendl; | |
2237 | } | |
2238 | ||
2239 | int skip_front, skip_back; | |
2240 | if (e.logical_offset < srcoff) { | |
2241 | skip_front = srcoff - e.logical_offset; | |
2242 | } else { | |
2243 | skip_front = 0; | |
2244 | } | |
2245 | if (e.logical_end() > end) { | |
2246 | skip_back = e.logical_end() - end; | |
2247 | } else { | |
2248 | skip_back = 0; | |
2249 | } | |
2250 | ||
2251 | Extent* ne = new Extent(e.logical_offset + skip_front + dstoff - srcoff, | |
2252 | e.blob_offset + skip_front, e.length - skip_front - skip_back, cb); | |
2253 | newo->extent_map.extent_map.insert(*ne); | |
2254 | ne->blob->get_ref(c.get(), ne->blob_offset, ne->length); | |
2255 | // fixme: we may leave parts of new blob unreferenced that could | |
2256 | // be freed (relative to the shared_blob). | |
2257 | txc->statfs_delta.stored() += ne->length; | |
2258 | if (e.blob->get_blob().is_compressed()) { | |
2259 | txc->statfs_delta.compressed_original() += ne->length; | |
2260 | if (blob_duped) { | |
2261 | txc->statfs_delta.compressed() += | |
2262 | cb->get_blob().get_compressed_payload_length(); | |
2263 | } | |
2264 | } | |
2265 | dout(20) << __func__ << " dst " << *ne << dendl; | |
2266 | ++n; | |
2267 | } | |
2268 | if ((!inject_21040 && src_dirty) || | |
2269 | (inject_21040 && dirty_range_end > dirty_range_begin)) { | |
2270 | oldo->extent_map.dirty_range(dirty_range_begin, | |
2271 | dirty_range_end - dirty_range_begin); | |
2272 | txc->write_onode(oldo); | |
2273 | } | |
2274 | txc->write_onode(newo); | |
2275 | ||
2276 | if (dstoff + length > newo->onode.size) { | |
2277 | newo->onode.size = dstoff + length; | |
2278 | } | |
2279 | newo->extent_map.dirty_range(dstoff, length); | |
2280 | } | |
7c673cae FG |
2281 | void BlueStore::ExtentMap::update(KeyValueDB::Transaction t, |
2282 | bool force) | |
2283 | { | |
2284 | auto cct = onode->c->store->cct; //used by dout | |
2285 | dout(20) << __func__ << " " << onode->oid << (force ? " force" : "") << dendl; | |
2286 | if (onode->onode.extent_map_shards.empty()) { | |
2287 | if (inline_bl.length() == 0) { | |
2288 | unsigned n; | |
2289 | // we need to encode inline_bl to measure encoded length | |
2290 | bool never_happen = encode_some(0, OBJECT_MAX_SIZE, inline_bl, &n); | |
3efd9988 | 2291 | inline_bl.reassign_to_mempool(mempool::mempool_bluestore_cache_other); |
11fdf7f2 | 2292 | ceph_assert(!never_happen); |
7c673cae FG |
2293 | size_t len = inline_bl.length(); |
2294 | dout(20) << __func__ << " inline shard " << len << " bytes from " << n | |
2295 | << " extents" << dendl; | |
2296 | if (!force && len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2297 | request_reshard(0, OBJECT_MAX_SIZE); | |
2298 | return; | |
2299 | } | |
2300 | } | |
2301 | // will persist in the onode key. | |
2302 | } else { | |
2303 | // pending shard update | |
2304 | struct dirty_shard_t { | |
2305 | Shard *shard; | |
2306 | bufferlist bl; | |
2307 | dirty_shard_t(Shard *s) : shard(s) {} | |
2308 | }; | |
2309 | vector<dirty_shard_t> encoded_shards; | |
2310 | // allocate slots for all shards in a single call instead of | |
2311 | // doing multiple allocations - one per each dirty shard | |
2312 | encoded_shards.reserve(shards.size()); | |
2313 | ||
2314 | auto p = shards.begin(); | |
2315 | auto prev_p = p; | |
2316 | while (p != shards.end()) { | |
11fdf7f2 | 2317 | ceph_assert(p->shard_info->offset >= prev_p->shard_info->offset); |
7c673cae FG |
2318 | auto n = p; |
2319 | ++n; | |
2320 | if (p->dirty) { | |
2321 | uint32_t endoff; | |
2322 | if (n == shards.end()) { | |
2323 | endoff = OBJECT_MAX_SIZE; | |
2324 | } else { | |
2325 | endoff = n->shard_info->offset; | |
2326 | } | |
2327 | encoded_shards.emplace_back(dirty_shard_t(&(*p))); | |
2328 | bufferlist& bl = encoded_shards.back().bl; | |
2329 | if (encode_some(p->shard_info->offset, endoff - p->shard_info->offset, | |
2330 | bl, &p->extents)) { | |
2331 | if (force) { | |
2332 | derr << __func__ << " encode_some needs reshard" << dendl; | |
11fdf7f2 | 2333 | ceph_assert(!force); |
7c673cae FG |
2334 | } |
2335 | } | |
2336 | size_t len = bl.length(); | |
2337 | ||
2338 | dout(20) << __func__ << " shard 0x" << std::hex | |
2339 | << p->shard_info->offset << std::dec << " is " << len | |
2340 | << " bytes (was " << p->shard_info->bytes << ") from " | |
2341 | << p->extents << " extents" << dendl; | |
2342 | ||
2343 | if (!force) { | |
2344 | if (len > cct->_conf->bluestore_extent_map_shard_max_size) { | |
2345 | // we are big; reshard ourselves | |
2346 | request_reshard(p->shard_info->offset, endoff); | |
2347 | } | |
2348 | // avoid resharding the trailing shard, even if it is small | |
2349 | else if (n != shards.end() && | |
11fdf7f2 TL |
2350 | len < g_conf()->bluestore_extent_map_shard_min_size) { |
2351 | ceph_assert(endoff != OBJECT_MAX_SIZE); | |
31f18b77 FG |
2352 | if (p == shards.begin()) { |
2353 | // we are the first shard, combine with next shard | |
7c673cae | 2354 | request_reshard(p->shard_info->offset, endoff + 1); |
7c673cae | 2355 | } else { |
31f18b77 FG |
2356 | // combine either with the previous shard or the next, |
2357 | // whichever is smaller | |
7c673cae FG |
2358 | if (prev_p->shard_info->bytes > n->shard_info->bytes) { |
2359 | request_reshard(p->shard_info->offset, endoff + 1); | |
2360 | } else { | |
2361 | request_reshard(prev_p->shard_info->offset, endoff); | |
2362 | } | |
2363 | } | |
2364 | } | |
2365 | } | |
2366 | } | |
2367 | prev_p = p; | |
2368 | p = n; | |
2369 | } | |
2370 | if (needs_reshard()) { | |
2371 | return; | |
2372 | } | |
2373 | ||
2374 | // schedule DB update for dirty shards | |
2375 | string key; | |
2376 | for (auto& it : encoded_shards) { | |
2377 | it.shard->dirty = false; | |
2378 | it.shard->shard_info->bytes = it.bl.length(); | |
2379 | generate_extent_shard_key_and_apply( | |
2380 | onode->key, | |
2381 | it.shard->shard_info->offset, | |
2382 | &key, | |
2383 | [&](const string& final_key) { | |
2384 | t->set(PREFIX_OBJ, final_key, it.bl); | |
2385 | } | |
2386 | ); | |
2387 | } | |
2388 | } | |
2389 | } | |
2390 | ||
31f18b77 FG |
2391 | bid_t BlueStore::ExtentMap::allocate_spanning_blob_id() |
2392 | { | |
2393 | if (spanning_blob_map.empty()) | |
2394 | return 0; | |
2395 | bid_t bid = spanning_blob_map.rbegin()->first + 1; | |
2396 | // bid is valid and available. | |
2397 | if (bid >= 0) | |
2398 | return bid; | |
2399 | // Find next unused bid; | |
2400 | bid = rand() % (numeric_limits<bid_t>::max() + 1); | |
2401 | const auto begin_bid = bid; | |
2402 | do { | |
2403 | if (!spanning_blob_map.count(bid)) | |
2404 | return bid; | |
2405 | else { | |
2406 | bid++; | |
2407 | if (bid < 0) bid = 0; | |
2408 | } | |
2409 | } while (bid != begin_bid); | |
81eedcae TL |
2410 | auto cct = onode->c->store->cct; // used by dout |
2411 | _dump_onode<0>(cct, *onode); | |
11fdf7f2 | 2412 | ceph_abort_msg("no available blob id"); |
31f18b77 FG |
2413 | } |
2414 | ||
7c673cae FG |
2415 | void BlueStore::ExtentMap::reshard( |
2416 | KeyValueDB *db, | |
2417 | KeyValueDB::Transaction t) | |
2418 | { | |
2419 | auto cct = onode->c->store->cct; // used by dout | |
2420 | ||
2421 | dout(10) << __func__ << " 0x[" << std::hex << needs_reshard_begin << "," | |
2422 | << needs_reshard_end << ")" << std::dec | |
2423 | << " of " << onode->onode.extent_map_shards.size() | |
2424 | << " shards on " << onode->oid << dendl; | |
2425 | for (auto& p : spanning_blob_map) { | |
2426 | dout(20) << __func__ << " spanning blob " << p.first << " " << *p.second | |
2427 | << dendl; | |
2428 | } | |
2429 | // determine shard index range | |
2430 | unsigned si_begin = 0, si_end = 0; | |
2431 | if (!shards.empty()) { | |
2432 | while (si_begin + 1 < shards.size() && | |
2433 | shards[si_begin + 1].shard_info->offset <= needs_reshard_begin) { | |
2434 | ++si_begin; | |
2435 | } | |
2436 | needs_reshard_begin = shards[si_begin].shard_info->offset; | |
2437 | for (si_end = si_begin; si_end < shards.size(); ++si_end) { | |
2438 | if (shards[si_end].shard_info->offset >= needs_reshard_end) { | |
2439 | needs_reshard_end = shards[si_end].shard_info->offset; | |
2440 | break; | |
2441 | } | |
2442 | } | |
2443 | if (si_end == shards.size()) { | |
2444 | needs_reshard_end = OBJECT_MAX_SIZE; | |
2445 | } | |
2446 | dout(20) << __func__ << " shards [" << si_begin << "," << si_end << ")" | |
2447 | << " over 0x[" << std::hex << needs_reshard_begin << "," | |
2448 | << needs_reshard_end << ")" << std::dec << dendl; | |
2449 | } | |
2450 | ||
181888fb | 2451 | fault_range(db, needs_reshard_begin, (needs_reshard_end - needs_reshard_begin)); |
7c673cae FG |
2452 | |
2453 | // we may need to fault in a larger interval later must have all | |
2454 | // referring extents for spanning blobs loaded in order to have | |
2455 | // accurate use_tracker values. | |
2456 | uint32_t spanning_scan_begin = needs_reshard_begin; | |
2457 | uint32_t spanning_scan_end = needs_reshard_end; | |
2458 | ||
2459 | // remove old keys | |
2460 | string key; | |
2461 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2462 | generate_extent_shard_key_and_apply( | |
2463 | onode->key, shards[i].shard_info->offset, &key, | |
2464 | [&](const string& final_key) { | |
2465 | t->rmkey(PREFIX_OBJ, final_key); | |
2466 | } | |
2467 | ); | |
2468 | } | |
2469 | ||
2470 | // calculate average extent size | |
2471 | unsigned bytes = 0; | |
2472 | unsigned extents = 0; | |
2473 | if (onode->onode.extent_map_shards.empty()) { | |
2474 | bytes = inline_bl.length(); | |
2475 | extents = extent_map.size(); | |
2476 | } else { | |
2477 | for (unsigned i = si_begin; i < si_end; ++i) { | |
2478 | bytes += shards[i].shard_info->bytes; | |
2479 | extents += shards[i].extents; | |
2480 | } | |
2481 | } | |
2482 | unsigned target = cct->_conf->bluestore_extent_map_shard_target_size; | |
2483 | unsigned slop = target * | |
2484 | cct->_conf->bluestore_extent_map_shard_target_size_slop; | |
11fdf7f2 | 2485 | unsigned extent_avg = bytes / std::max(1u, extents); |
7c673cae FG |
2486 | dout(20) << __func__ << " extent_avg " << extent_avg << ", target " << target |
2487 | << ", slop " << slop << dendl; | |
2488 | ||
2489 | // reshard | |
2490 | unsigned estimate = 0; | |
31f18b77 | 2491 | unsigned offset = needs_reshard_begin; |
7c673cae FG |
2492 | vector<bluestore_onode_t::shard_info> new_shard_info; |
2493 | unsigned max_blob_end = 0; | |
2494 | Extent dummy(needs_reshard_begin); | |
2495 | for (auto e = extent_map.lower_bound(dummy); | |
2496 | e != extent_map.end(); | |
2497 | ++e) { | |
2498 | if (e->logical_offset >= needs_reshard_end) { | |
2499 | break; | |
2500 | } | |
2501 | dout(30) << " extent " << *e << dendl; | |
2502 | ||
2503 | // disfavor shard boundaries that span a blob | |
2504 | bool would_span = (e->logical_offset < max_blob_end) || e->blob_offset; | |
2505 | if (estimate && | |
2506 | estimate + extent_avg > target + (would_span ? slop : 0)) { | |
2507 | // new shard | |
31f18b77 | 2508 | if (offset == needs_reshard_begin) { |
7c673cae FG |
2509 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); |
2510 | new_shard_info.back().offset = offset; | |
2511 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
31f18b77 | 2512 | << std::dec << dendl; |
7c673cae FG |
2513 | } |
2514 | offset = e->logical_offset; | |
2515 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2516 | new_shard_info.back().offset = offset; | |
2517 | dout(20) << __func__ << " new shard 0x" << std::hex << offset | |
2518 | << std::dec << dendl; | |
2519 | estimate = 0; | |
2520 | } | |
2521 | estimate += extent_avg; | |
31f18b77 FG |
2522 | unsigned bs = e->blob_start(); |
2523 | if (bs < spanning_scan_begin) { | |
2524 | spanning_scan_begin = bs; | |
7c673cae FG |
2525 | } |
2526 | uint32_t be = e->blob_end(); | |
2527 | if (be > max_blob_end) { | |
2528 | max_blob_end = be; | |
2529 | } | |
2530 | if (be > spanning_scan_end) { | |
2531 | spanning_scan_end = be; | |
2532 | } | |
2533 | } | |
2534 | if (new_shard_info.empty() && (si_begin > 0 || | |
2535 | si_end < shards.size())) { | |
2536 | // we resharded a partial range; we must produce at least one output | |
2537 | // shard | |
2538 | new_shard_info.emplace_back(bluestore_onode_t::shard_info()); | |
2539 | new_shard_info.back().offset = needs_reshard_begin; | |
2540 | dout(20) << __func__ << " new shard 0x" << std::hex << needs_reshard_begin | |
2541 | << std::dec << " (singleton degenerate case)" << dendl; | |
2542 | } | |
2543 | ||
2544 | auto& sv = onode->onode.extent_map_shards; | |
2545 | dout(20) << __func__ << " new " << new_shard_info << dendl; | |
2546 | dout(20) << __func__ << " old " << sv << dendl; | |
2547 | if (sv.empty()) { | |
2548 | // no old shards to keep | |
2549 | sv.swap(new_shard_info); | |
2550 | init_shards(true, true); | |
2551 | } else { | |
2552 | // splice in new shards | |
2553 | sv.erase(sv.begin() + si_begin, sv.begin() + si_end); | |
2554 | shards.erase(shards.begin() + si_begin, shards.begin() + si_end); | |
2555 | sv.insert( | |
2556 | sv.begin() + si_begin, | |
2557 | new_shard_info.begin(), | |
2558 | new_shard_info.end()); | |
2559 | shards.insert(shards.begin() + si_begin, new_shard_info.size(), Shard()); | |
7c673cae | 2560 | si_end = si_begin + new_shard_info.size(); |
31f18b77 | 2561 | |
11fdf7f2 | 2562 | ceph_assert(sv.size() == shards.size()); |
31f18b77 FG |
2563 | |
2564 | // note that we need to update every shard_info of shards here, | |
2565 | // as sv might have been totally re-allocated above | |
2566 | for (unsigned i = 0; i < shards.size(); i++) { | |
7c673cae | 2567 | shards[i].shard_info = &sv[i]; |
31f18b77 FG |
2568 | } |
2569 | ||
2570 | // mark newly added shards as dirty | |
2571 | for (unsigned i = si_begin; i < si_end; ++i) { | |
7c673cae FG |
2572 | shards[i].loaded = true; |
2573 | shards[i].dirty = true; | |
2574 | } | |
7c673cae FG |
2575 | } |
2576 | dout(20) << __func__ << " fin " << sv << dendl; | |
2577 | inline_bl.clear(); | |
2578 | ||
2579 | if (sv.empty()) { | |
2580 | // no more shards; unspan all previously spanning blobs | |
2581 | auto p = spanning_blob_map.begin(); | |
2582 | while (p != spanning_blob_map.end()) { | |
2583 | p->second->id = -1; | |
2584 | dout(30) << __func__ << " un-spanning " << *p->second << dendl; | |
2585 | p = spanning_blob_map.erase(p); | |
2586 | } | |
2587 | } else { | |
2588 | // identify new spanning blobs | |
2589 | dout(20) << __func__ << " checking spanning blobs 0x[" << std::hex | |
2590 | << spanning_scan_begin << "," << spanning_scan_end << ")" << dendl; | |
2591 | if (spanning_scan_begin < needs_reshard_begin) { | |
2592 | fault_range(db, spanning_scan_begin, | |
2593 | needs_reshard_begin - spanning_scan_begin); | |
2594 | } | |
2595 | if (spanning_scan_end > needs_reshard_end) { | |
2596 | fault_range(db, needs_reshard_end, | |
31f18b77 | 2597 | spanning_scan_end - needs_reshard_end); |
7c673cae FG |
2598 | } |
2599 | auto sp = sv.begin() + si_begin; | |
2600 | auto esp = sv.end(); | |
2601 | unsigned shard_start = sp->offset; | |
2602 | unsigned shard_end; | |
2603 | ++sp; | |
2604 | if (sp == esp) { | |
2605 | shard_end = OBJECT_MAX_SIZE; | |
2606 | } else { | |
2607 | shard_end = sp->offset; | |
2608 | } | |
7c673cae | 2609 | Extent dummy(needs_reshard_begin); |
9f95a23c TL |
2610 | |
2611 | bool was_too_many_blobs_check = false; | |
2612 | auto too_many_blobs_threshold = | |
2613 | g_conf()->bluestore_debug_too_many_blobs_threshold; | |
2614 | auto& dumped_onodes = onode->c->onode_map.cache->dumped_onodes; | |
2615 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oid_slot = nullptr; | |
2616 | decltype(onode->c->onode_map.cache->dumped_onodes)::value_type* oldest_slot = nullptr; | |
2617 | ||
7c673cae FG |
2618 | for (auto e = extent_map.lower_bound(dummy); e != extent_map.end(); ++e) { |
2619 | if (e->logical_offset >= needs_reshard_end) { | |
2620 | break; | |
2621 | } | |
2622 | dout(30) << " extent " << *e << dendl; | |
2623 | while (e->logical_offset >= shard_end) { | |
2624 | shard_start = shard_end; | |
11fdf7f2 | 2625 | ceph_assert(sp != esp); |
7c673cae FG |
2626 | ++sp; |
2627 | if (sp == esp) { | |
2628 | shard_end = OBJECT_MAX_SIZE; | |
2629 | } else { | |
2630 | shard_end = sp->offset; | |
2631 | } | |
2632 | dout(30) << __func__ << " shard 0x" << std::hex << shard_start | |
2633 | << " to 0x" << shard_end << std::dec << dendl; | |
2634 | } | |
9f95a23c | 2635 | |
7c673cae FG |
2636 | if (e->blob_escapes_range(shard_start, shard_end - shard_start)) { |
2637 | if (!e->blob->is_spanning()) { | |
2638 | // We have two options: (1) split the blob into pieces at the | |
2639 | // shard boundaries (and adjust extents accordingly), or (2) | |
2640 | // mark it spanning. We prefer to cut the blob if we can. Note that | |
2641 | // we may have to split it multiple times--potentially at every | |
2642 | // shard boundary. | |
2643 | bool must_span = false; | |
2644 | BlobRef b = e->blob; | |
2645 | if (b->can_split()) { | |
2646 | uint32_t bstart = e->blob_start(); | |
2647 | uint32_t bend = e->blob_end(); | |
2648 | for (const auto& sh : shards) { | |
2649 | if (bstart < sh.shard_info->offset && | |
2650 | bend > sh.shard_info->offset) { | |
2651 | uint32_t blob_offset = sh.shard_info->offset - bstart; | |
2652 | if (b->can_split_at(blob_offset)) { | |
2653 | dout(20) << __func__ << " splitting blob, bstart 0x" | |
2654 | << std::hex << bstart << " blob_offset 0x" | |
2655 | << blob_offset << std::dec << " " << *b << dendl; | |
2656 | b = split_blob(b, blob_offset, sh.shard_info->offset); | |
2657 | // switch b to the new right-hand side, in case it | |
2658 | // *also* has to get split. | |
2659 | bstart += blob_offset; | |
2660 | onode->c->store->logger->inc(l_bluestore_blob_split); | |
2661 | } else { | |
2662 | must_span = true; | |
2663 | break; | |
2664 | } | |
2665 | } | |
2666 | } | |
2667 | } else { | |
2668 | must_span = true; | |
2669 | } | |
2670 | if (must_span) { | |
31f18b77 FG |
2671 | auto bid = allocate_spanning_blob_id(); |
2672 | b->id = bid; | |
7c673cae FG |
2673 | spanning_blob_map[b->id] = b; |
2674 | dout(20) << __func__ << " adding spanning " << *b << dendl; | |
9f95a23c TL |
2675 | if (!was_too_many_blobs_check && |
2676 | too_many_blobs_threshold && | |
2677 | spanning_blob_map.size() >= size_t(too_many_blobs_threshold)) { | |
2678 | ||
2679 | was_too_many_blobs_check = true; | |
2680 | for (size_t i = 0; i < dumped_onodes.size(); ++i) { | |
2681 | if (dumped_onodes[i].first == onode->oid) { | |
2682 | oid_slot = &dumped_onodes[i]; | |
2683 | break; | |
2684 | } | |
2685 | if (!oldest_slot || (oldest_slot && | |
2686 | dumped_onodes[i].second < oldest_slot->second)) { | |
2687 | oldest_slot = &dumped_onodes[i]; | |
2688 | } | |
2689 | } | |
2690 | } | |
7c673cae FG |
2691 | } |
2692 | } | |
2693 | } else { | |
2694 | if (e->blob->is_spanning()) { | |
2695 | spanning_blob_map.erase(e->blob->id); | |
2696 | e->blob->id = -1; | |
2697 | dout(30) << __func__ << " un-spanning " << *e->blob << dendl; | |
2698 | } | |
2699 | } | |
2700 | } | |
9f95a23c TL |
2701 | bool do_dump = (!oid_slot && was_too_many_blobs_check) || |
2702 | (oid_slot && | |
2703 | (mono_clock::now() - oid_slot->second >= make_timespan(5 * 60))); | |
2704 | if (do_dump) { | |
2705 | dout(0) << __func__ | |
2706 | << " spanning blob count exceeds threshold, " | |
2707 | << spanning_blob_map.size() << " spanning blobs" | |
2708 | << dendl; | |
2709 | _dump_onode<0>(cct, *onode); | |
2710 | if (oid_slot) { | |
2711 | oid_slot->second = mono_clock::now(); | |
2712 | } else { | |
2713 | ceph_assert(oldest_slot); | |
2714 | oldest_slot->first = onode->oid; | |
2715 | oldest_slot->second = mono_clock::now(); | |
2716 | } | |
2717 | } | |
7c673cae FG |
2718 | } |
2719 | ||
2720 | clear_needs_reshard(); | |
2721 | } | |
2722 | ||
2723 | bool BlueStore::ExtentMap::encode_some( | |
2724 | uint32_t offset, | |
2725 | uint32_t length, | |
2726 | bufferlist& bl, | |
2727 | unsigned *pn) | |
2728 | { | |
7c673cae FG |
2729 | Extent dummy(offset); |
2730 | auto start = extent_map.lower_bound(dummy); | |
2731 | uint32_t end = offset + length; | |
2732 | ||
2733 | __u8 struct_v = 2; // Version 2 differs from v1 in blob's ref_map | |
2734 | // serialization only. Hence there is no specific | |
2735 | // handling at ExtentMap level. | |
2736 | ||
2737 | unsigned n = 0; | |
2738 | size_t bound = 0; | |
7c673cae FG |
2739 | bool must_reshard = false; |
2740 | for (auto p = start; | |
2741 | p != extent_map.end() && p->logical_offset < end; | |
2742 | ++p, ++n) { | |
11fdf7f2 | 2743 | ceph_assert(p->logical_offset >= offset); |
7c673cae FG |
2744 | p->blob->last_encoded_id = -1; |
2745 | if (!p->blob->is_spanning() && p->blob_escapes_range(offset, length)) { | |
2746 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length | |
2747 | << std::dec << " hit new spanning blob " << *p << dendl; | |
2748 | request_reshard(p->blob_start(), p->blob_end()); | |
2749 | must_reshard = true; | |
2750 | } | |
31f18b77 FG |
2751 | if (!must_reshard) { |
2752 | denc_varint(0, bound); // blobid | |
2753 | denc_varint(0, bound); // logical_offset | |
2754 | denc_varint(0, bound); // len | |
2755 | denc_varint(0, bound); // blob_offset | |
7c673cae | 2756 | |
31f18b77 FG |
2757 | p->blob->bound_encode( |
2758 | bound, | |
2759 | struct_v, | |
2760 | p->blob->shared_blob->get_sbid(), | |
2761 | false); | |
2762 | } | |
7c673cae FG |
2763 | } |
2764 | if (must_reshard) { | |
2765 | return true; | |
2766 | } | |
2767 | ||
31f18b77 FG |
2768 | denc(struct_v, bound); |
2769 | denc_varint(0, bound); // number of extents | |
2770 | ||
7c673cae FG |
2771 | { |
2772 | auto app = bl.get_contiguous_appender(bound); | |
2773 | denc(struct_v, app); | |
2774 | denc_varint(n, app); | |
2775 | if (pn) { | |
2776 | *pn = n; | |
2777 | } | |
2778 | ||
2779 | n = 0; | |
2780 | uint64_t pos = 0; | |
2781 | uint64_t prev_len = 0; | |
2782 | for (auto p = start; | |
2783 | p != extent_map.end() && p->logical_offset < end; | |
2784 | ++p, ++n) { | |
2785 | unsigned blobid; | |
2786 | bool include_blob = false; | |
2787 | if (p->blob->is_spanning()) { | |
2788 | blobid = p->blob->id << BLOBID_SHIFT_BITS; | |
2789 | blobid |= BLOBID_FLAG_SPANNING; | |
2790 | } else if (p->blob->last_encoded_id < 0) { | |
2791 | p->blob->last_encoded_id = n + 1; // so it is always non-zero | |
2792 | include_blob = true; | |
2793 | blobid = 0; // the decoder will infer the id from n | |
2794 | } else { | |
2795 | blobid = p->blob->last_encoded_id << BLOBID_SHIFT_BITS; | |
2796 | } | |
2797 | if (p->logical_offset == pos) { | |
2798 | blobid |= BLOBID_FLAG_CONTIGUOUS; | |
2799 | } | |
2800 | if (p->blob_offset == 0) { | |
2801 | blobid |= BLOBID_FLAG_ZEROOFFSET; | |
2802 | } | |
2803 | if (p->length == prev_len) { | |
2804 | blobid |= BLOBID_FLAG_SAMELENGTH; | |
2805 | } else { | |
2806 | prev_len = p->length; | |
2807 | } | |
2808 | denc_varint(blobid, app); | |
2809 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
2810 | denc_varint_lowz(p->logical_offset - pos, app); | |
2811 | } | |
2812 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
2813 | denc_varint_lowz(p->blob_offset, app); | |
2814 | } | |
2815 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
2816 | denc_varint_lowz(p->length, app); | |
2817 | } | |
2818 | pos = p->logical_end(); | |
2819 | if (include_blob) { | |
2820 | p->blob->encode(app, struct_v, p->blob->shared_blob->get_sbid(), false); | |
2821 | } | |
2822 | } | |
2823 | } | |
2824 | /*derr << __func__ << bl << dendl; | |
2825 | derr << __func__ << ":"; | |
2826 | bl.hexdump(*_dout); | |
2827 | *_dout << dendl; | |
2828 | */ | |
2829 | return false; | |
2830 | } | |
2831 | ||
2832 | unsigned BlueStore::ExtentMap::decode_some(bufferlist& bl) | |
2833 | { | |
7c673cae FG |
2834 | /* |
2835 | derr << __func__ << ":"; | |
2836 | bl.hexdump(*_dout); | |
2837 | *_dout << dendl; | |
2838 | */ | |
2839 | ||
11fdf7f2 | 2840 | ceph_assert(bl.get_num_buffers() <= 1); |
7c673cae FG |
2841 | auto p = bl.front().begin_deep(); |
2842 | __u8 struct_v; | |
2843 | denc(struct_v, p); | |
2844 | // Version 2 differs from v1 in blob's ref_map | |
2845 | // serialization only. Hence there is no specific | |
2846 | // handling at ExtentMap level below. | |
11fdf7f2 | 2847 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
2848 | |
2849 | uint32_t num; | |
2850 | denc_varint(num, p); | |
2851 | vector<BlobRef> blobs(num); | |
2852 | uint64_t pos = 0; | |
2853 | uint64_t prev_len = 0; | |
2854 | unsigned n = 0; | |
2855 | ||
2856 | while (!p.end()) { | |
2857 | Extent *le = new Extent(); | |
2858 | uint64_t blobid; | |
2859 | denc_varint(blobid, p); | |
2860 | if ((blobid & BLOBID_FLAG_CONTIGUOUS) == 0) { | |
2861 | uint64_t gap; | |
2862 | denc_varint_lowz(gap, p); | |
2863 | pos += gap; | |
2864 | } | |
2865 | le->logical_offset = pos; | |
2866 | if ((blobid & BLOBID_FLAG_ZEROOFFSET) == 0) { | |
2867 | denc_varint_lowz(le->blob_offset, p); | |
2868 | } else { | |
2869 | le->blob_offset = 0; | |
2870 | } | |
2871 | if ((blobid & BLOBID_FLAG_SAMELENGTH) == 0) { | |
2872 | denc_varint_lowz(prev_len, p); | |
2873 | } | |
2874 | le->length = prev_len; | |
2875 | ||
2876 | if (blobid & BLOBID_FLAG_SPANNING) { | |
2877 | dout(30) << __func__ << " getting spanning blob " | |
2878 | << (blobid >> BLOBID_SHIFT_BITS) << dendl; | |
2879 | le->assign_blob(get_spanning_blob(blobid >> BLOBID_SHIFT_BITS)); | |
2880 | } else { | |
2881 | blobid >>= BLOBID_SHIFT_BITS; | |
2882 | if (blobid) { | |
2883 | le->assign_blob(blobs[blobid - 1]); | |
11fdf7f2 | 2884 | ceph_assert(le->blob); |
7c673cae FG |
2885 | } else { |
2886 | Blob *b = new Blob(); | |
2887 | uint64_t sbid = 0; | |
2888 | b->decode(onode->c, p, struct_v, &sbid, false); | |
2889 | blobs[n] = b; | |
2890 | onode->c->open_shared_blob(sbid, b); | |
2891 | le->assign_blob(b); | |
2892 | } | |
2893 | // we build ref_map dynamically for non-spanning blobs | |
2894 | le->blob->get_ref( | |
2895 | onode->c, | |
2896 | le->blob_offset, | |
2897 | le->length); | |
2898 | } | |
2899 | pos += prev_len; | |
2900 | ++n; | |
2901 | extent_map.insert(*le); | |
2902 | } | |
2903 | ||
11fdf7f2 | 2904 | ceph_assert(n == num); |
7c673cae FG |
2905 | return num; |
2906 | } | |
2907 | ||
2908 | void BlueStore::ExtentMap::bound_encode_spanning_blobs(size_t& p) | |
2909 | { | |
2910 | // Version 2 differs from v1 in blob's ref_map | |
2911 | // serialization only. Hence there is no specific | |
2912 | // handling at ExtentMap level. | |
2913 | __u8 struct_v = 2; | |
2914 | ||
2915 | denc(struct_v, p); | |
2916 | denc_varint((uint32_t)0, p); | |
2917 | size_t key_size = 0; | |
2918 | denc_varint((uint32_t)0, key_size); | |
2919 | p += spanning_blob_map.size() * key_size; | |
2920 | for (const auto& i : spanning_blob_map) { | |
2921 | i.second->bound_encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
2922 | } | |
2923 | } | |
2924 | ||
2925 | void BlueStore::ExtentMap::encode_spanning_blobs( | |
2926 | bufferlist::contiguous_appender& p) | |
2927 | { | |
2928 | // Version 2 differs from v1 in blob's ref_map | |
2929 | // serialization only. Hence there is no specific | |
2930 | // handling at ExtentMap level. | |
2931 | __u8 struct_v = 2; | |
2932 | ||
2933 | denc(struct_v, p); | |
2934 | denc_varint(spanning_blob_map.size(), p); | |
2935 | for (auto& i : spanning_blob_map) { | |
2936 | denc_varint(i.second->id, p); | |
2937 | i.second->encode(p, struct_v, i.second->shared_blob->get_sbid(), true); | |
2938 | } | |
2939 | } | |
2940 | ||
2941 | void BlueStore::ExtentMap::decode_spanning_blobs( | |
11fdf7f2 | 2942 | bufferptr::const_iterator& p) |
7c673cae FG |
2943 | { |
2944 | __u8 struct_v; | |
2945 | denc(struct_v, p); | |
2946 | // Version 2 differs from v1 in blob's ref_map | |
2947 | // serialization only. Hence there is no specific | |
2948 | // handling at ExtentMap level. | |
11fdf7f2 | 2949 | ceph_assert(struct_v == 1 || struct_v == 2); |
7c673cae FG |
2950 | |
2951 | unsigned n; | |
2952 | denc_varint(n, p); | |
2953 | while (n--) { | |
2954 | BlobRef b(new Blob()); | |
2955 | denc_varint(b->id, p); | |
2956 | spanning_blob_map[b->id] = b; | |
2957 | uint64_t sbid = 0; | |
2958 | b->decode(onode->c, p, struct_v, &sbid, true); | |
2959 | onode->c->open_shared_blob(sbid, b); | |
2960 | } | |
2961 | } | |
2962 | ||
2963 | void BlueStore::ExtentMap::init_shards(bool loaded, bool dirty) | |
2964 | { | |
2965 | shards.resize(onode->onode.extent_map_shards.size()); | |
2966 | unsigned i = 0; | |
2967 | for (auto &s : onode->onode.extent_map_shards) { | |
2968 | shards[i].shard_info = &s; | |
2969 | shards[i].loaded = loaded; | |
2970 | shards[i].dirty = dirty; | |
2971 | ++i; | |
2972 | } | |
2973 | } | |
2974 | ||
2975 | void BlueStore::ExtentMap::fault_range( | |
2976 | KeyValueDB *db, | |
2977 | uint32_t offset, | |
2978 | uint32_t length) | |
2979 | { | |
7c673cae FG |
2980 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
2981 | << std::dec << dendl; | |
2982 | auto start = seek_shard(offset); | |
2983 | auto last = seek_shard(offset + length); | |
2984 | ||
2985 | if (start < 0) | |
2986 | return; | |
2987 | ||
11fdf7f2 | 2988 | ceph_assert(last >= start); |
7c673cae FG |
2989 | string key; |
2990 | while (start <= last) { | |
11fdf7f2 | 2991 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
2992 | auto p = &shards[start]; |
2993 | if (!p->loaded) { | |
2994 | dout(30) << __func__ << " opening shard 0x" << std::hex | |
2995 | << p->shard_info->offset << std::dec << dendl; | |
2996 | bufferlist v; | |
2997 | generate_extent_shard_key_and_apply( | |
2998 | onode->key, p->shard_info->offset, &key, | |
2999 | [&](const string& final_key) { | |
3000 | int r = db->get(PREFIX_OBJ, final_key, &v); | |
3001 | if (r < 0) { | |
3002 | derr << __func__ << " missing shard 0x" << std::hex | |
3003 | << p->shard_info->offset << std::dec << " for " << onode->oid | |
3004 | << dendl; | |
11fdf7f2 | 3005 | ceph_assert(r >= 0); |
7c673cae FG |
3006 | } |
3007 | } | |
3008 | ); | |
3009 | p->extents = decode_some(v); | |
3010 | p->loaded = true; | |
3011 | dout(20) << __func__ << " open shard 0x" << std::hex | |
81eedcae TL |
3012 | << p->shard_info->offset |
3013 | << " for range 0x" << offset << "~" << length << std::dec | |
7c673cae | 3014 | << " (" << v.length() << " bytes)" << dendl; |
11fdf7f2 TL |
3015 | ceph_assert(p->dirty == false); |
3016 | ceph_assert(v.length() == p->shard_info->bytes); | |
7c673cae FG |
3017 | onode->c->store->logger->inc(l_bluestore_onode_shard_misses); |
3018 | } else { | |
3019 | onode->c->store->logger->inc(l_bluestore_onode_shard_hits); | |
3020 | } | |
3021 | ++start; | |
3022 | } | |
3023 | } | |
3024 | ||
3025 | void BlueStore::ExtentMap::dirty_range( | |
7c673cae FG |
3026 | uint32_t offset, |
3027 | uint32_t length) | |
3028 | { | |
7c673cae FG |
3029 | dout(30) << __func__ << " 0x" << std::hex << offset << "~" << length |
3030 | << std::dec << dendl; | |
3031 | if (shards.empty()) { | |
3032 | dout(20) << __func__ << " mark inline shard dirty" << dendl; | |
3033 | inline_bl.clear(); | |
3034 | return; | |
3035 | } | |
3036 | auto start = seek_shard(offset); | |
11fdf7f2 TL |
3037 | if (length == 0) { |
3038 | length = 1; | |
3039 | } | |
3040 | auto last = seek_shard(offset + length - 1); | |
7c673cae FG |
3041 | if (start < 0) |
3042 | return; | |
3043 | ||
11fdf7f2 | 3044 | ceph_assert(last >= start); |
7c673cae | 3045 | while (start <= last) { |
11fdf7f2 | 3046 | ceph_assert((size_t)start < shards.size()); |
7c673cae FG |
3047 | auto p = &shards[start]; |
3048 | if (!p->loaded) { | |
11fdf7f2 TL |
3049 | derr << __func__ << "on write 0x" << std::hex << offset |
3050 | << "~" << length << " shard 0x" << p->shard_info->offset | |
3051 | << std::dec << " is not loaded, can't mark dirty" << dendl; | |
3052 | ceph_abort_msg("can't mark unloaded shard dirty"); | |
7c673cae FG |
3053 | } |
3054 | if (!p->dirty) { | |
3055 | dout(20) << __func__ << " mark shard 0x" << std::hex | |
3056 | << p->shard_info->offset << std::dec << " dirty" << dendl; | |
3057 | p->dirty = true; | |
3058 | } | |
3059 | ++start; | |
3060 | } | |
3061 | } | |
3062 | ||
3063 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::find( | |
3064 | uint64_t offset) | |
3065 | { | |
3066 | Extent dummy(offset); | |
3067 | return extent_map.find(dummy); | |
3068 | } | |
3069 | ||
7c673cae FG |
3070 | BlueStore::extent_map_t::iterator BlueStore::ExtentMap::seek_lextent( |
3071 | uint64_t offset) | |
3072 | { | |
3073 | Extent dummy(offset); | |
3074 | auto fp = extent_map.lower_bound(dummy); | |
3075 | if (fp != extent_map.begin()) { | |
3076 | --fp; | |
3077 | if (fp->logical_end() <= offset) { | |
3078 | ++fp; | |
3079 | } | |
3080 | } | |
3081 | return fp; | |
3082 | } | |
3083 | ||
3084 | BlueStore::extent_map_t::const_iterator BlueStore::ExtentMap::seek_lextent( | |
3085 | uint64_t offset) const | |
3086 | { | |
3087 | Extent dummy(offset); | |
3088 | auto fp = extent_map.lower_bound(dummy); | |
3089 | if (fp != extent_map.begin()) { | |
3090 | --fp; | |
3091 | if (fp->logical_end() <= offset) { | |
3092 | ++fp; | |
3093 | } | |
3094 | } | |
3095 | return fp; | |
3096 | } | |
3097 | ||
3098 | bool BlueStore::ExtentMap::has_any_lextents(uint64_t offset, uint64_t length) | |
3099 | { | |
3100 | auto fp = seek_lextent(offset); | |
3101 | if (fp == extent_map.end() || fp->logical_offset >= offset + length) { | |
3102 | return false; | |
3103 | } | |
3104 | return true; | |
3105 | } | |
3106 | ||
3107 | int BlueStore::ExtentMap::compress_extent_map( | |
3108 | uint64_t offset, | |
3109 | uint64_t length) | |
3110 | { | |
7c673cae FG |
3111 | if (extent_map.empty()) |
3112 | return 0; | |
3113 | int removed = 0; | |
3114 | auto p = seek_lextent(offset); | |
3115 | if (p != extent_map.begin()) { | |
3116 | --p; // start to the left of offset | |
3117 | } | |
3118 | // the caller should have just written to this region | |
11fdf7f2 | 3119 | ceph_assert(p != extent_map.end()); |
7c673cae FG |
3120 | |
3121 | // identify the *next* shard | |
3122 | auto pshard = shards.begin(); | |
3123 | while (pshard != shards.end() && | |
3124 | p->logical_offset >= pshard->shard_info->offset) { | |
3125 | ++pshard; | |
3126 | } | |
3127 | uint64_t shard_end; | |
3128 | if (pshard != shards.end()) { | |
3129 | shard_end = pshard->shard_info->offset; | |
3130 | } else { | |
3131 | shard_end = OBJECT_MAX_SIZE; | |
3132 | } | |
3133 | ||
3134 | auto n = p; | |
3135 | for (++n; n != extent_map.end(); p = n++) { | |
3136 | if (n->logical_offset > offset + length) { | |
3137 | break; // stop after end | |
3138 | } | |
3139 | while (n != extent_map.end() && | |
3140 | p->logical_end() == n->logical_offset && | |
3141 | p->blob == n->blob && | |
3142 | p->blob_offset + p->length == n->blob_offset && | |
3143 | n->logical_offset < shard_end) { | |
3144 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3145 | << " next shard 0x" << shard_end << std::dec | |
3146 | << " merging " << *p << " and " << *n << dendl; | |
3147 | p->length += n->length; | |
3148 | rm(n++); | |
3149 | ++removed; | |
3150 | } | |
3151 | if (n == extent_map.end()) { | |
3152 | break; | |
3153 | } | |
3154 | if (n->logical_offset >= shard_end) { | |
11fdf7f2 | 3155 | ceph_assert(pshard != shards.end()); |
7c673cae FG |
3156 | ++pshard; |
3157 | if (pshard != shards.end()) { | |
3158 | shard_end = pshard->shard_info->offset; | |
3159 | } else { | |
3160 | shard_end = OBJECT_MAX_SIZE; | |
3161 | } | |
3162 | } | |
3163 | } | |
11fdf7f2 | 3164 | if (removed) { |
7c673cae FG |
3165 | onode->c->store->logger->inc(l_bluestore_extent_compress, removed); |
3166 | } | |
3167 | return removed; | |
3168 | } | |
3169 | ||
3170 | void BlueStore::ExtentMap::punch_hole( | |
3171 | CollectionRef &c, | |
3172 | uint64_t offset, | |
3173 | uint64_t length, | |
3174 | old_extent_map_t *old_extents) | |
3175 | { | |
3176 | auto p = seek_lextent(offset); | |
3177 | uint64_t end = offset + length; | |
3178 | while (p != extent_map.end()) { | |
3179 | if (p->logical_offset >= end) { | |
3180 | break; | |
3181 | } | |
3182 | if (p->logical_offset < offset) { | |
3183 | if (p->logical_end() > end) { | |
3184 | // split and deref middle | |
3185 | uint64_t front = offset - p->logical_offset; | |
3186 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + front, | |
3187 | length, p->blob); | |
3188 | old_extents->push_back(*oe); | |
3189 | add(end, | |
3190 | p->blob_offset + front + length, | |
3191 | p->length - front - length, | |
3192 | p->blob); | |
3193 | p->length = front; | |
3194 | break; | |
3195 | } else { | |
3196 | // deref tail | |
11fdf7f2 | 3197 | ceph_assert(p->logical_end() > offset); // else seek_lextent bug |
7c673cae FG |
3198 | uint64_t keep = offset - p->logical_offset; |
3199 | OldExtent* oe = OldExtent::create(c, offset, p->blob_offset + keep, | |
3200 | p->length - keep, p->blob); | |
3201 | old_extents->push_back(*oe); | |
3202 | p->length = keep; | |
3203 | ++p; | |
3204 | continue; | |
3205 | } | |
3206 | } | |
3207 | if (p->logical_offset + p->length <= end) { | |
3208 | // deref whole lextent | |
3209 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3210 | p->length, p->blob); | |
3211 | old_extents->push_back(*oe); | |
3212 | rm(p++); | |
3213 | continue; | |
3214 | } | |
3215 | // deref head | |
3216 | uint64_t keep = p->logical_end() - end; | |
3217 | OldExtent* oe = OldExtent::create(c, p->logical_offset, p->blob_offset, | |
3218 | p->length - keep, p->blob); | |
3219 | old_extents->push_back(*oe); | |
3220 | ||
3221 | add(end, p->blob_offset + p->length - keep, keep, p->blob); | |
3222 | rm(p); | |
3223 | break; | |
3224 | } | |
3225 | } | |
3226 | ||
3227 | BlueStore::Extent *BlueStore::ExtentMap::set_lextent( | |
3228 | CollectionRef &c, | |
3229 | uint64_t logical_offset, | |
3230 | uint64_t blob_offset, uint64_t length, BlobRef b, | |
3231 | old_extent_map_t *old_extents) | |
3232 | { | |
3233 | // We need to have completely initialized Blob to increment its ref counters. | |
11fdf7f2 | 3234 | ceph_assert(b->get_blob().get_logical_length() != 0); |
7c673cae FG |
3235 | |
3236 | // Do get_ref prior to punch_hole to prevent from putting reused blob into | |
3237 | // old_extents list if we overwre the blob totally | |
3238 | // This might happen during WAL overwrite. | |
3239 | b->get_ref(onode->c, blob_offset, length); | |
3240 | ||
3241 | if (old_extents) { | |
3242 | punch_hole(c, logical_offset, length, old_extents); | |
3243 | } | |
3244 | ||
3245 | Extent *le = new Extent(logical_offset, blob_offset, length, b); | |
3246 | extent_map.insert(*le); | |
3247 | if (spans_shard(logical_offset, length)) { | |
3248 | request_reshard(logical_offset, logical_offset + length); | |
3249 | } | |
3250 | return le; | |
3251 | } | |
3252 | ||
3253 | BlueStore::BlobRef BlueStore::ExtentMap::split_blob( | |
3254 | BlobRef lb, | |
3255 | uint32_t blob_offset, | |
3256 | uint32_t pos) | |
3257 | { | |
7c673cae FG |
3258 | uint32_t end_pos = pos + lb->get_blob().get_logical_length() - blob_offset; |
3259 | dout(20) << __func__ << " 0x" << std::hex << pos << " end 0x" << end_pos | |
3260 | << " blob_offset 0x" << blob_offset << std::dec << " " << *lb | |
3261 | << dendl; | |
3262 | BlobRef rb = onode->c->new_blob(); | |
3263 | lb->split(onode->c, blob_offset, rb.get()); | |
3264 | ||
3265 | for (auto ep = seek_lextent(pos); | |
3266 | ep != extent_map.end() && ep->logical_offset < end_pos; | |
3267 | ++ep) { | |
3268 | if (ep->blob != lb) { | |
3269 | continue; | |
3270 | } | |
3271 | if (ep->logical_offset < pos) { | |
3272 | // split extent | |
3273 | size_t left = pos - ep->logical_offset; | |
3274 | Extent *ne = new Extent(pos, 0, ep->length - left, rb); | |
3275 | extent_map.insert(*ne); | |
3276 | ep->length = left; | |
3277 | dout(30) << __func__ << " split " << *ep << dendl; | |
3278 | dout(30) << __func__ << " to " << *ne << dendl; | |
3279 | } else { | |
3280 | // switch blob | |
11fdf7f2 | 3281 | ceph_assert(ep->blob_offset >= blob_offset); |
7c673cae FG |
3282 | |
3283 | ep->blob = rb; | |
3284 | ep->blob_offset -= blob_offset; | |
3285 | dout(30) << __func__ << " adjusted " << *ep << dendl; | |
3286 | } | |
3287 | } | |
3288 | return rb; | |
3289 | } | |
3290 | ||
3291 | // Onode | |
3292 | ||
3293 | #undef dout_prefix | |
3294 | #define dout_prefix *_dout << "bluestore.onode(" << this << ")." << __func__ << " " | |
3295 | ||
f6b5b4d7 TL |
3296 | // |
3297 | // A tricky thing about Onode's ref counter is that we do an additional | |
3298 | // increment when newly pinned instance is detected. And -1 on unpin. | |
3299 | // This prevents from a conflict with a delete call (when nref == 0). | |
3300 | // The latter might happen while the thread is in unpin() function | |
3301 | // (and e.g. waiting for lock acquisition) since nref is already | |
3302 | // decremented. And another 'putting' thread on the instance will release it. | |
3303 | // | |
3304 | void BlueStore::Onode::get() { | |
3305 | if (++nref == 2) { | |
3306 | c->get_onode_cache()->pin(this, [&]() { | |
3307 | bool was_pinned = pinned; | |
3308 | pinned = nref >= 2; | |
3309 | // additional increment for newly pinned instance | |
3310 | bool r = !was_pinned && pinned; | |
3311 | if (r) { | |
3312 | ++nref; | |
3313 | } | |
3314 | return cached && r; | |
3315 | }); | |
3316 | } | |
3317 | } | |
3318 | void BlueStore::Onode::put() { | |
3319 | if (--nref == 2) { | |
3320 | c->get_onode_cache()->unpin(this, [&]() { | |
3321 | bool was_pinned = pinned; | |
3322 | pinned = pinned && nref > 2; // intentionally use > not >= as we have | |
3323 | // +1 due to pinned state | |
3324 | bool r = was_pinned && !pinned; | |
3325 | // additional decrement for newly unpinned instance | |
3326 | if (r) { | |
3327 | --nref; | |
3328 | } | |
3329 | return cached && r; | |
3330 | }); | |
3331 | } | |
3332 | if (nref == 0) { | |
3333 | delete this; | |
3334 | } | |
3335 | } | |
3336 | ||
eafe8130 TL |
3337 | BlueStore::Onode* BlueStore::Onode::decode( |
3338 | CollectionRef c, | |
3339 | const ghobject_t& oid, | |
3340 | const string& key, | |
3341 | const bufferlist& v) | |
3342 | { | |
3343 | Onode* on = new Onode(c.get(), oid, key); | |
3344 | on->exists = true; | |
3345 | auto p = v.front().begin_deep(); | |
3346 | on->onode.decode(p); | |
3347 | for (auto& i : on->onode.attrs) { | |
3348 | i.second.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
3349 | } | |
3350 | ||
3351 | // initialize extent_map | |
3352 | on->extent_map.decode_spanning_blobs(p); | |
3353 | if (on->onode.extent_map_shards.empty()) { | |
3354 | denc(on->extent_map.inline_bl, p); | |
3355 | on->extent_map.decode_some(on->extent_map.inline_bl); | |
3356 | on->extent_map.inline_bl.reassign_to_mempool( | |
3357 | mempool::mempool_bluestore_cache_other); | |
3358 | } | |
3359 | else { | |
3360 | on->extent_map.init_shards(false, false); | |
3361 | } | |
3362 | return on; | |
3363 | } | |
3364 | ||
7c673cae FG |
3365 | void BlueStore::Onode::flush() |
3366 | { | |
3367 | if (flushing_count.load()) { | |
3368 | ldout(c->store->cct, 20) << __func__ << " cnt:" << flushing_count << dendl; | |
9f95a23c | 3369 | waiting_count++; |
11fdf7f2 | 3370 | std::unique_lock l(flush_lock); |
7c673cae FG |
3371 | while (flushing_count.load()) { |
3372 | flush_cond.wait(l); | |
3373 | } | |
9f95a23c | 3374 | waiting_count--; |
7c673cae FG |
3375 | } |
3376 | ldout(c->store->cct, 20) << __func__ << " done" << dendl; | |
3377 | } | |
3378 | ||
9f95a23c TL |
3379 | void BlueStore::Onode::dump(Formatter* f) const |
3380 | { | |
3381 | onode.dump(f); | |
3382 | extent_map.dump(f); | |
3383 | } | |
3384 | ||
3385 | ||
3386 | const string& BlueStore::Onode::get_omap_prefix() | |
3387 | { | |
3388 | if (onode.is_pgmeta_omap()) { | |
3389 | return PREFIX_PGMETA_OMAP; | |
3390 | } | |
3391 | if (onode.is_perpool_omap()) { | |
3392 | return PREFIX_PERPOOL_OMAP; | |
3393 | } | |
3394 | return PREFIX_OMAP; | |
3395 | } | |
3396 | ||
3397 | // '-' < '.' < '~' | |
3398 | ||
3399 | void BlueStore::Onode::get_omap_header(string *out) | |
3400 | { | |
3401 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3402 | _key_encode_u64(c->pool(), out); | |
3403 | } | |
3404 | _key_encode_u64(onode.nid, out); | |
3405 | out->push_back('-'); | |
3406 | } | |
3407 | ||
3408 | void BlueStore::Onode::get_omap_key(const string& key, string *out) | |
3409 | { | |
3410 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3411 | _key_encode_u64(c->pool(), out); | |
3412 | } | |
3413 | _key_encode_u64(onode.nid, out); | |
3414 | out->push_back('.'); | |
3415 | out->append(key); | |
3416 | } | |
3417 | ||
3418 | void BlueStore::Onode::rewrite_omap_key(const string& old, string *out) | |
3419 | { | |
3420 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3421 | _key_encode_u64(c->pool(), out); | |
3422 | } | |
3423 | _key_encode_u64(onode.nid, out); | |
3424 | out->append(old.c_str() + out->length(), old.size() - out->length()); | |
3425 | } | |
3426 | ||
3427 | void BlueStore::Onode::get_omap_tail(string *out) | |
3428 | { | |
3429 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3430 | _key_encode_u64(c->pool(), out); | |
3431 | } | |
3432 | _key_encode_u64(onode.nid, out); | |
3433 | out->push_back('~'); | |
3434 | } | |
3435 | ||
3436 | void BlueStore::Onode::decode_omap_key(const string& key, string *user_key) | |
3437 | { | |
3438 | if (onode.is_perpool_omap() && !onode.is_pgmeta_omap()) { | |
3439 | *user_key = key.substr(sizeof(uint64_t)*2 + 1); | |
3440 | } else { | |
3441 | *user_key = key.substr(sizeof(uint64_t) + 1); | |
3442 | } | |
3443 | } | |
3444 | ||
3445 | ||
7c673cae FG |
3446 | // ======================================================= |
3447 | // WriteContext | |
3448 | ||
3449 | /// Checks for writes to the same pextent within a blob | |
3450 | bool BlueStore::WriteContext::has_conflict( | |
3451 | BlobRef b, | |
3452 | uint64_t loffs, | |
3453 | uint64_t loffs_end, | |
3454 | uint64_t min_alloc_size) | |
3455 | { | |
11fdf7f2 TL |
3456 | ceph_assert((loffs % min_alloc_size) == 0); |
3457 | ceph_assert((loffs_end % min_alloc_size) == 0); | |
7c673cae FG |
3458 | for (auto w : writes) { |
3459 | if (b == w.b) { | |
11fdf7f2 TL |
3460 | auto loffs2 = p2align(w.logical_offset, min_alloc_size); |
3461 | auto loffs2_end = p2roundup(w.logical_offset + w.length0, min_alloc_size); | |
7c673cae | 3462 | if ((loffs <= loffs2 && loffs_end > loffs2) || |
224ce89b | 3463 | (loffs >= loffs2 && loffs < loffs2_end)) { |
7c673cae FG |
3464 | return true; |
3465 | } | |
3466 | } | |
3467 | } | |
3468 | return false; | |
3469 | } | |
3470 | ||
3471 | // ======================================================= | |
3472 | ||
3473 | // DeferredBatch | |
3474 | #undef dout_prefix | |
3475 | #define dout_prefix *_dout << "bluestore.DeferredBatch(" << this << ") " | |
9f95a23c TL |
3476 | #undef dout_context |
3477 | #define dout_context cct | |
7c673cae FG |
3478 | |
3479 | void BlueStore::DeferredBatch::prepare_write( | |
3480 | CephContext *cct, | |
3481 | uint64_t seq, uint64_t offset, uint64_t length, | |
3482 | bufferlist::const_iterator& blp) | |
3483 | { | |
3484 | _discard(cct, offset, length); | |
3485 | auto i = iomap.insert(make_pair(offset, deferred_io())); | |
11fdf7f2 | 3486 | ceph_assert(i.second); // this should be a new insertion |
7c673cae FG |
3487 | i.first->second.seq = seq; |
3488 | blp.copy(length, i.first->second.bl); | |
31f18b77 FG |
3489 | i.first->second.bl.reassign_to_mempool( |
3490 | mempool::mempool_bluestore_writing_deferred); | |
7c673cae FG |
3491 | dout(20) << __func__ << " seq " << seq |
3492 | << " 0x" << std::hex << offset << "~" << length | |
3493 | << " crc " << i.first->second.bl.crc32c(-1) | |
3494 | << std::dec << dendl; | |
3495 | seq_bytes[seq] += length; | |
3496 | #ifdef DEBUG_DEFERRED | |
3497 | _audit(cct); | |
3498 | #endif | |
3499 | } | |
3500 | ||
3501 | void BlueStore::DeferredBatch::_discard( | |
3502 | CephContext *cct, uint64_t offset, uint64_t length) | |
3503 | { | |
3504 | generic_dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
3505 | << std::dec << dendl; | |
3506 | auto p = iomap.lower_bound(offset); | |
3507 | if (p != iomap.begin()) { | |
3508 | --p; | |
3509 | auto end = p->first + p->second.bl.length(); | |
3510 | if (end > offset) { | |
3511 | bufferlist head; | |
3512 | head.substr_of(p->second.bl, 0, offset - p->first); | |
3513 | dout(20) << __func__ << " keep head " << p->second.seq | |
3514 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3515 | << " -> 0x" << head.length() << std::dec << dendl; | |
3516 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3517 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3518 | if (end > offset + length) { |
3519 | bufferlist tail; | |
3520 | tail.substr_of(p->second.bl, offset + length - p->first, | |
3521 | end - (offset + length)); | |
3522 | dout(20) << __func__ << " keep tail " << p->second.seq | |
3523 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3524 | << " -> 0x" << tail.length() << std::dec << dendl; | |
3525 | auto &n = iomap[offset + length]; | |
3526 | n.bl.swap(tail); | |
3527 | n.seq = p->second.seq; | |
3528 | i->second -= length; | |
3529 | } else { | |
3530 | i->second -= end - offset; | |
3531 | } | |
11fdf7f2 | 3532 | ceph_assert(i->second >= 0); |
7c673cae FG |
3533 | p->second.bl.swap(head); |
3534 | } | |
3535 | ++p; | |
3536 | } | |
3537 | while (p != iomap.end()) { | |
3538 | if (p->first >= offset + length) { | |
3539 | break; | |
3540 | } | |
3541 | auto i = seq_bytes.find(p->second.seq); | |
11fdf7f2 | 3542 | ceph_assert(i != seq_bytes.end()); |
7c673cae FG |
3543 | auto end = p->first + p->second.bl.length(); |
3544 | if (end > offset + length) { | |
3545 | unsigned drop_front = offset + length - p->first; | |
3546 | unsigned keep_tail = end - (offset + length); | |
3547 | dout(20) << __func__ << " truncate front " << p->second.seq | |
3548 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3549 | << " drop_front 0x" << drop_front << " keep_tail 0x" << keep_tail | |
3550 | << " to 0x" << (offset + length) << "~" << keep_tail | |
3551 | << std::dec << dendl; | |
3552 | auto &s = iomap[offset + length]; | |
3553 | s.seq = p->second.seq; | |
3554 | s.bl.substr_of(p->second.bl, drop_front, keep_tail); | |
3555 | i->second -= drop_front; | |
3556 | } else { | |
3557 | dout(20) << __func__ << " drop " << p->second.seq | |
3558 | << " 0x" << std::hex << p->first << "~" << p->second.bl.length() | |
3559 | << std::dec << dendl; | |
3560 | i->second -= p->second.bl.length(); | |
3561 | } | |
11fdf7f2 | 3562 | ceph_assert(i->second >= 0); |
7c673cae FG |
3563 | p = iomap.erase(p); |
3564 | } | |
3565 | } | |
3566 | ||
3567 | void BlueStore::DeferredBatch::_audit(CephContext *cct) | |
3568 | { | |
3569 | map<uint64_t,int> sb; | |
3570 | for (auto p : seq_bytes) { | |
3571 | sb[p.first] = 0; // make sure we have the same set of keys | |
3572 | } | |
3573 | uint64_t pos = 0; | |
3574 | for (auto& p : iomap) { | |
11fdf7f2 | 3575 | ceph_assert(p.first >= pos); |
7c673cae FG |
3576 | sb[p.second.seq] += p.second.bl.length(); |
3577 | pos = p.first + p.second.bl.length(); | |
3578 | } | |
11fdf7f2 | 3579 | ceph_assert(sb == seq_bytes); |
7c673cae FG |
3580 | } |
3581 | ||
3582 | ||
3583 | // Collection | |
3584 | ||
3585 | #undef dout_prefix | |
3586 | #define dout_prefix *_dout << "bluestore(" << store->path << ").collection(" << cid << " " << this << ") " | |
3587 | ||
9f95a23c TL |
3588 | BlueStore::Collection::Collection(BlueStore *store_, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t cid) |
3589 | : CollectionImpl(store_->cct, cid), | |
11fdf7f2 | 3590 | store(store_), |
9f95a23c | 3591 | cache(bc), |
7c673cae | 3592 | exists(true), |
9f95a23c | 3593 | onode_map(oc), |
11fdf7f2 TL |
3594 | commit_queue(nullptr) |
3595 | { | |
3596 | } | |
3597 | ||
3598 | bool BlueStore::Collection::flush_commit(Context *c) | |
3599 | { | |
3600 | return osr->flush_commit(c); | |
3601 | } | |
3602 | ||
3603 | void BlueStore::Collection::flush() | |
3604 | { | |
3605 | osr->flush(); | |
3606 | } | |
3607 | ||
3608 | void BlueStore::Collection::flush_all_but_last() | |
7c673cae | 3609 | { |
11fdf7f2 | 3610 | osr->flush_all_but_last(); |
7c673cae FG |
3611 | } |
3612 | ||
3613 | void BlueStore::Collection::open_shared_blob(uint64_t sbid, BlobRef b) | |
3614 | { | |
11fdf7f2 | 3615 | ceph_assert(!b->shared_blob); |
7c673cae FG |
3616 | const bluestore_blob_t& blob = b->get_blob(); |
3617 | if (!blob.is_shared()) { | |
3618 | b->shared_blob = new SharedBlob(this); | |
3619 | return; | |
3620 | } | |
3621 | ||
3622 | b->shared_blob = shared_blob_set.lookup(sbid); | |
3623 | if (b->shared_blob) { | |
3624 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3625 | << std::dec << " had " << *b->shared_blob << dendl; | |
3626 | } else { | |
3627 | b->shared_blob = new SharedBlob(sbid, this); | |
3628 | shared_blob_set.add(this, b->shared_blob.get()); | |
3629 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid | |
3630 | << std::dec << " opened " << *b->shared_blob | |
3631 | << dendl; | |
3632 | } | |
3633 | } | |
3634 | ||
3635 | void BlueStore::Collection::load_shared_blob(SharedBlobRef sb) | |
3636 | { | |
3637 | if (!sb->is_loaded()) { | |
3638 | ||
3639 | bufferlist v; | |
3640 | string key; | |
3641 | auto sbid = sb->get_sbid(); | |
3642 | get_shared_blob_key(sbid, &key); | |
3643 | int r = store->db->get(PREFIX_SHARED_BLOB, key, &v); | |
3644 | if (r < 0) { | |
3645 | lderr(store->cct) << __func__ << " sbid 0x" << std::hex << sbid | |
3646 | << std::dec << " not found at key " | |
3647 | << pretty_binary_string(key) << dendl; | |
11fdf7f2 | 3648 | ceph_abort_msg("uh oh, missing shared_blob"); |
7c673cae FG |
3649 | } |
3650 | ||
3651 | sb->loaded = true; | |
3652 | sb->persistent = new bluestore_shared_blob_t(sbid); | |
11fdf7f2 TL |
3653 | auto p = v.cbegin(); |
3654 | decode(*(sb->persistent), p); | |
7c673cae FG |
3655 | ldout(store->cct, 10) << __func__ << " sbid 0x" << std::hex << sbid |
3656 | << std::dec << " loaded shared_blob " << *sb << dendl; | |
3657 | } | |
3658 | } | |
3659 | ||
3660 | void BlueStore::Collection::make_blob_shared(uint64_t sbid, BlobRef b) | |
3661 | { | |
7c673cae | 3662 | ldout(store->cct, 10) << __func__ << " " << *b << dendl; |
11fdf7f2 | 3663 | ceph_assert(!b->shared_blob->is_loaded()); |
7c673cae FG |
3664 | |
3665 | // update blob | |
31f18b77 | 3666 | bluestore_blob_t& blob = b->dirty_blob(); |
7c673cae | 3667 | blob.set_flag(bluestore_blob_t::FLAG_SHARED); |
7c673cae FG |
3668 | |
3669 | // update shared blob | |
3670 | b->shared_blob->loaded = true; | |
3671 | b->shared_blob->persistent = new bluestore_shared_blob_t(sbid); | |
3672 | shared_blob_set.add(this, b->shared_blob.get()); | |
3673 | for (auto p : blob.get_extents()) { | |
3674 | if (p.is_valid()) { | |
3675 | b->shared_blob->get_ref( | |
3676 | p.offset, | |
3677 | p.length); | |
3678 | } | |
3679 | } | |
3680 | ldout(store->cct, 20) << __func__ << " now " << *b << dendl; | |
3681 | } | |
3682 | ||
31f18b77 FG |
3683 | uint64_t BlueStore::Collection::make_blob_unshared(SharedBlob *sb) |
3684 | { | |
3685 | ldout(store->cct, 10) << __func__ << " " << *sb << dendl; | |
11fdf7f2 | 3686 | ceph_assert(sb->is_loaded()); |
31f18b77 FG |
3687 | |
3688 | uint64_t sbid = sb->get_sbid(); | |
3689 | shared_blob_set.remove(sb); | |
3690 | sb->loaded = false; | |
3691 | delete sb->persistent; | |
3692 | sb->sbid_unloaded = 0; | |
3693 | ldout(store->cct, 20) << __func__ << " now " << *sb << dendl; | |
3694 | return sbid; | |
3695 | } | |
3696 | ||
7c673cae FG |
3697 | BlueStore::OnodeRef BlueStore::Collection::get_onode( |
3698 | const ghobject_t& oid, | |
9f95a23c TL |
3699 | bool create, |
3700 | bool is_createop) | |
7c673cae | 3701 | { |
9f95a23c | 3702 | ceph_assert(create ? ceph_mutex_is_wlocked(lock) : ceph_mutex_is_locked(lock)); |
7c673cae FG |
3703 | |
3704 | spg_t pgid; | |
3705 | if (cid.is_pg(&pgid)) { | |
3706 | if (!oid.match(cnode.bits, pgid.ps())) { | |
3707 | lderr(store->cct) << __func__ << " oid " << oid << " not part of " | |
3708 | << pgid << " bits " << cnode.bits << dendl; | |
3709 | ceph_abort(); | |
3710 | } | |
3711 | } | |
3712 | ||
3713 | OnodeRef o = onode_map.lookup(oid); | |
3714 | if (o) | |
3715 | return o; | |
3716 | ||
eafe8130 | 3717 | string key; |
7c673cae FG |
3718 | get_object_key(store->cct, oid, &key); |
3719 | ||
3720 | ldout(store->cct, 20) << __func__ << " oid " << oid << " key " | |
3721 | << pretty_binary_string(key) << dendl; | |
3722 | ||
3723 | bufferlist v; | |
9f95a23c | 3724 | int r = -ENOENT; |
7c673cae | 3725 | Onode *on; |
9f95a23c TL |
3726 | if (!is_createop) { |
3727 | r = store->db->get(PREFIX_OBJ, key.c_str(), key.size(), &v); | |
3728 | ldout(store->cct, 20) << " r " << r << " v.len " << v.length() << dendl; | |
3729 | } | |
7c673cae | 3730 | if (v.length() == 0) { |
11fdf7f2 | 3731 | ceph_assert(r == -ENOENT); |
7c673cae FG |
3732 | if (!store->cct->_conf->bluestore_debug_misc && |
3733 | !create) | |
3734 | return OnodeRef(); | |
3735 | ||
3736 | // new object, new onode | |
3737 | on = new Onode(this, oid, key); | |
3738 | } else { | |
3739 | // loaded | |
11fdf7f2 | 3740 | ceph_assert(r >= 0); |
eafe8130 | 3741 | on = Onode::decode(this, oid, key, v); |
7c673cae FG |
3742 | } |
3743 | o.reset(on); | |
3744 | return onode_map.add(oid, o); | |
3745 | } | |
3746 | ||
3747 | void BlueStore::Collection::split_cache( | |
3748 | Collection *dest) | |
3749 | { | |
3750 | ldout(store->cct, 10) << __func__ << " to " << dest << dendl; | |
3751 | ||
3752 | // lock (one or both) cache shards | |
3753 | std::lock(cache->lock, dest->cache->lock); | |
11fdf7f2 TL |
3754 | std::lock_guard l(cache->lock, std::adopt_lock); |
3755 | std::lock_guard l2(dest->cache->lock, std::adopt_lock); | |
7c673cae FG |
3756 | |
3757 | int destbits = dest->cnode.bits; | |
3758 | spg_t destpg; | |
3759 | bool is_pg = dest->cid.is_pg(&destpg); | |
11fdf7f2 | 3760 | ceph_assert(is_pg); |
7c673cae FG |
3761 | |
3762 | auto p = onode_map.onode_map.begin(); | |
3763 | while (p != onode_map.onode_map.end()) { | |
11fdf7f2 | 3764 | OnodeRef o = p->second; |
7c673cae FG |
3765 | if (!p->second->oid.match(destbits, destpg.pgid.ps())) { |
3766 | // onode does not belong to this child | |
11fdf7f2 TL |
3767 | ldout(store->cct, 20) << __func__ << " not moving " << o << " " << o->oid |
3768 | << dendl; | |
7c673cae FG |
3769 | ++p; |
3770 | } else { | |
7c673cae FG |
3771 | ldout(store->cct, 20) << __func__ << " moving " << o << " " << o->oid |
3772 | << dendl; | |
3773 | ||
f6b5b4d7 TL |
3774 | // ensuring that nref is always >= 2 and hence onode is pinned and |
3775 | // physically out of cache during the transition | |
3776 | OnodeRef o_pin = o; | |
3777 | ceph_assert(o->pinned); | |
3778 | ||
7c673cae | 3779 | p = onode_map.onode_map.erase(p); |
7c673cae | 3780 | dest->onode_map.onode_map[o->oid] = o; |
f6b5b4d7 TL |
3781 | if (get_onode_cache() != dest->get_onode_cache()) { |
3782 | get_onode_cache()->move_pinned(dest->get_onode_cache(), o.get()); | |
9f95a23c | 3783 | } |
f6b5b4d7 | 3784 | o->c = dest; |
7c673cae FG |
3785 | |
3786 | // move over shared blobs and buffers. cover shared blobs from | |
3787 | // both extent map and spanning blob map (the full extent map | |
3788 | // may not be faulted in) | |
3789 | vector<SharedBlob*> sbvec; | |
3790 | for (auto& e : o->extent_map.extent_map) { | |
3791 | sbvec.push_back(e.blob->shared_blob.get()); | |
3792 | } | |
3793 | for (auto& b : o->extent_map.spanning_blob_map) { | |
3794 | sbvec.push_back(b.second->shared_blob.get()); | |
3795 | } | |
3796 | for (auto sb : sbvec) { | |
3797 | if (sb->coll == dest) { | |
3798 | ldout(store->cct, 20) << __func__ << " already moved " << *sb | |
3799 | << dendl; | |
3800 | continue; | |
3801 | } | |
3802 | ldout(store->cct, 20) << __func__ << " moving " << *sb << dendl; | |
31f18b77 FG |
3803 | if (sb->get_sbid()) { |
3804 | ldout(store->cct, 20) << __func__ | |
3805 | << " moving registration " << *sb << dendl; | |
3806 | shared_blob_set.remove(sb); | |
3807 | dest->shared_blob_set.add(dest, sb); | |
3808 | } | |
3efd9988 | 3809 | sb->coll = dest; |
7c673cae | 3810 | if (dest->cache != cache) { |
7c673cae FG |
3811 | for (auto& i : sb->bc.buffer_map) { |
3812 | if (!i.second->is_writing()) { | |
3813 | ldout(store->cct, 20) << __func__ << " moving " << *i.second | |
3814 | << dendl; | |
9f95a23c | 3815 | dest->cache->_move(cache, i.second.get()); |
7c673cae FG |
3816 | } |
3817 | } | |
3818 | } | |
3819 | } | |
7c673cae FG |
3820 | } |
3821 | } | |
9f95a23c | 3822 | dest->cache->_trim(); |
7c673cae FG |
3823 | } |
3824 | ||
7c673cae FG |
3825 | // ======================================================= |
3826 | ||
91327a77 AA |
3827 | // MempoolThread |
3828 | ||
3829 | #undef dout_prefix | |
3830 | #define dout_prefix *_dout << "bluestore.MempoolThread(" << this << ") " | |
9f95a23c TL |
3831 | #undef dout_context |
3832 | #define dout_context store->cct | |
91327a77 | 3833 | |
7c673cae FG |
3834 | void *BlueStore::MempoolThread::entry() |
3835 | { | |
9f95a23c | 3836 | std::unique_lock l{lock}; |
11fdf7f2 | 3837 | |
92f5a8d4 | 3838 | uint32_t prev_config_change = store->config_changed.load(); |
eafe8130 TL |
3839 | uint64_t base = store->osd_memory_base; |
3840 | double fragmentation = store->osd_memory_expected_fragmentation; | |
3841 | uint64_t target = store->osd_memory_target; | |
3842 | uint64_t min = store->osd_memory_cache_min; | |
3843 | uint64_t max = min; | |
3844 | ||
3845 | // When setting the maximum amount of memory to use for cache, first | |
3846 | // assume some base amount of memory for the OSD and then fudge in | |
3847 | // some overhead for fragmentation that scales with cache usage. | |
3848 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
3849 | if (ltarget > base + min) { | |
3850 | max = ltarget - base; | |
11fdf7f2 | 3851 | } |
31f18b77 | 3852 | |
eafe8130 TL |
3853 | binned_kv_cache = store->db->get_priority_cache(); |
3854 | if (store->cache_autotune && binned_kv_cache != nullptr) { | |
3855 | pcm = std::make_shared<PriorityCache::Manager>( | |
3856 | store->cct, min, max, target, true); | |
3857 | pcm->insert("kv", binned_kv_cache, true); | |
3858 | pcm->insert("meta", meta_cache, true); | |
3859 | pcm->insert("data", data_cache, true); | |
3860 | } | |
91327a77 AA |
3861 | |
3862 | utime_t next_balance = ceph_clock_now(); | |
3863 | utime_t next_resize = ceph_clock_now(); | |
9f95a23c TL |
3864 | utime_t next_deferred_force_submit = ceph_clock_now(); |
3865 | utime_t alloc_stats_dump_clock = ceph_clock_now(); | |
31f18b77 | 3866 | |
91327a77 | 3867 | bool interval_stats_trim = false; |
91327a77 | 3868 | while (!stop) { |
92f5a8d4 TL |
3869 | // Update pcm cache settings if related configuration was changed |
3870 | uint32_t cur_config_change = store->config_changed.load(); | |
3871 | if (cur_config_change != prev_config_change) { | |
3872 | _update_cache_settings(); | |
3873 | prev_config_change = cur_config_change; | |
3874 | } | |
3875 | ||
91327a77 AA |
3876 | // Before we trim, check and see if it's time to rebalance/resize. |
3877 | double autotune_interval = store->cache_autotune_interval; | |
3878 | double resize_interval = store->osd_memory_cache_resize_interval; | |
9f95a23c TL |
3879 | double max_defer_interval = store->max_defer_interval; |
3880 | ||
3881 | double alloc_stats_dump_interval = | |
3882 | store->cct->_conf->bluestore_alloc_stats_dump_interval; | |
91327a77 | 3883 | |
9f95a23c TL |
3884 | if (alloc_stats_dump_interval > 0 && |
3885 | alloc_stats_dump_clock + alloc_stats_dump_interval < ceph_clock_now()) { | |
3886 | store->_record_allocation_stats(); | |
3887 | alloc_stats_dump_clock = ceph_clock_now(); | |
3888 | } | |
91327a77 | 3889 | if (autotune_interval > 0 && next_balance < ceph_clock_now()) { |
11fdf7f2 TL |
3890 | _adjust_cache_settings(); |
3891 | ||
91327a77 | 3892 | // Log events at 5 instead of 20 when balance happens. |
91327a77 | 3893 | interval_stats_trim = true; |
eafe8130 TL |
3894 | |
3895 | if (pcm != nullptr) { | |
3896 | pcm->balance(); | |
91327a77 | 3897 | } |
31f18b77 | 3898 | |
91327a77 AA |
3899 | next_balance = ceph_clock_now(); |
3900 | next_balance += autotune_interval; | |
3901 | } | |
3902 | if (resize_interval > 0 && next_resize < ceph_clock_now()) { | |
eafe8130 TL |
3903 | if (ceph_using_tcmalloc() && pcm != nullptr) { |
3904 | pcm->tune_memory(); | |
91327a77 AA |
3905 | } |
3906 | next_resize = ceph_clock_now(); | |
3907 | next_resize += resize_interval; | |
31f18b77 FG |
3908 | } |
3909 | ||
9f95a23c TL |
3910 | if (max_defer_interval > 0 && |
3911 | next_deferred_force_submit < ceph_clock_now()) { | |
3912 | if (store->get_deferred_last_submitted() + max_defer_interval < | |
3913 | ceph_clock_now()) { | |
3914 | store->deferred_try_submit(); | |
3915 | } | |
3916 | next_deferred_force_submit = ceph_clock_now(); | |
3917 | next_deferred_force_submit += max_defer_interval/3; | |
3918 | } | |
3919 | ||
3920 | // Now Resize the shards | |
3921 | _resize_shards(interval_stats_trim); | |
91327a77 | 3922 | interval_stats_trim = false; |
31f18b77 | 3923 | |
91327a77 | 3924 | store->_update_cache_logger(); |
11fdf7f2 TL |
3925 | auto wait = ceph::make_timespan( |
3926 | store->cct->_conf->bluestore_cache_trim_interval); | |
3927 | cond.wait_for(l, wait); | |
7c673cae | 3928 | } |
9f95a23c TL |
3929 | // do final dump |
3930 | store->_record_allocation_stats(); | |
7c673cae FG |
3931 | stop = false; |
3932 | return NULL; | |
3933 | } | |
3934 | ||
91327a77 AA |
3935 | void BlueStore::MempoolThread::_adjust_cache_settings() |
3936 | { | |
11fdf7f2 TL |
3937 | if (binned_kv_cache != nullptr) { |
3938 | binned_kv_cache->set_cache_ratio(store->cache_kv_ratio); | |
3939 | } | |
3940 | meta_cache->set_cache_ratio(store->cache_meta_ratio); | |
3941 | data_cache->set_cache_ratio(store->cache_data_ratio); | |
91327a77 AA |
3942 | } |
3943 | ||
9f95a23c | 3944 | void BlueStore::MempoolThread::_resize_shards(bool interval_stats) |
91327a77 | 3945 | { |
9f95a23c TL |
3946 | size_t onode_shards = store->onode_cache_shards.size(); |
3947 | size_t buffer_shards = store->buffer_cache_shards.size(); | |
91327a77 | 3948 | int64_t kv_used = store->db->get_cache_usage(); |
11fdf7f2 TL |
3949 | int64_t meta_used = meta_cache->_get_used_bytes(); |
3950 | int64_t data_used = data_cache->_get_used_bytes(); | |
91327a77 AA |
3951 | |
3952 | uint64_t cache_size = store->cache_size; | |
3953 | int64_t kv_alloc = | |
11fdf7f2 | 3954 | static_cast<int64_t>(store->cache_kv_ratio * cache_size); |
91327a77 | 3955 | int64_t meta_alloc = |
11fdf7f2 | 3956 | static_cast<int64_t>(store->cache_meta_ratio * cache_size); |
91327a77 | 3957 | int64_t data_alloc = |
11fdf7f2 | 3958 | static_cast<int64_t>(store->cache_data_ratio * cache_size); |
91327a77 | 3959 | |
eafe8130 TL |
3960 | if (pcm != nullptr && binned_kv_cache != nullptr) { |
3961 | cache_size = pcm->get_tuned_mem(); | |
11fdf7f2 TL |
3962 | kv_alloc = binned_kv_cache->get_committed_size(); |
3963 | meta_alloc = meta_cache->get_committed_size(); | |
3964 | data_alloc = data_cache->get_committed_size(); | |
91327a77 AA |
3965 | } |
3966 | ||
3967 | if (interval_stats) { | |
9f95a23c | 3968 | dout(5) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
3969 | << " kv_alloc: " << kv_alloc |
3970 | << " kv_used: " << kv_used | |
3971 | << " meta_alloc: " << meta_alloc | |
3972 | << " meta_used: " << meta_used | |
3973 | << " data_alloc: " << data_alloc | |
3974 | << " data_used: " << data_used << dendl; | |
3975 | } else { | |
9f95a23c | 3976 | dout(20) << __func__ << " cache_size: " << cache_size |
91327a77 AA |
3977 | << " kv_alloc: " << kv_alloc |
3978 | << " kv_used: " << kv_used | |
3979 | << " meta_alloc: " << meta_alloc | |
3980 | << " meta_used: " << meta_used | |
3981 | << " data_alloc: " << data_alloc | |
3982 | << " data_used: " << data_used << dendl; | |
3983 | } | |
3984 | ||
3985 | uint64_t max_shard_onodes = static_cast<uint64_t>( | |
9f95a23c TL |
3986 | (meta_alloc / (double) onode_shards) / meta_cache->get_bytes_per_onode()); |
3987 | uint64_t max_shard_buffer = static_cast<uint64_t>(data_alloc / buffer_shards); | |
91327a77 | 3988 | |
9f95a23c | 3989 | dout(30) << __func__ << " max_shard_onodes: " << max_shard_onodes |
91327a77 AA |
3990 | << " max_shard_buffer: " << max_shard_buffer << dendl; |
3991 | ||
9f95a23c TL |
3992 | for (auto i : store->onode_cache_shards) { |
3993 | i->set_max(max_shard_onodes); | |
3994 | } | |
3995 | for (auto i : store->buffer_cache_shards) { | |
3996 | i->set_max(max_shard_buffer); | |
91327a77 AA |
3997 | } |
3998 | } | |
3999 | ||
92f5a8d4 TL |
4000 | void BlueStore::MempoolThread::_update_cache_settings() |
4001 | { | |
4002 | // Nothing to do if pcm is not used. | |
4003 | if (pcm == nullptr) { | |
4004 | return; | |
4005 | } | |
4006 | ||
92f5a8d4 TL |
4007 | uint64_t target = store->osd_memory_target; |
4008 | uint64_t base = store->osd_memory_base; | |
4009 | uint64_t min = store->osd_memory_cache_min; | |
4010 | uint64_t max = min; | |
4011 | double fragmentation = store->osd_memory_expected_fragmentation; | |
4012 | ||
4013 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
4014 | if (ltarget > base + min) { | |
4015 | max = ltarget - base; | |
4016 | } | |
4017 | ||
4018 | // set pcm cache levels | |
4019 | pcm->set_target_memory(target); | |
4020 | pcm->set_min_memory(min); | |
4021 | pcm->set_max_memory(max); | |
4022 | ||
9f95a23c | 4023 | dout(5) << __func__ << " updated pcm target: " << target |
92f5a8d4 TL |
4024 | << " pcm min: " << min |
4025 | << " pcm max: " << max | |
4026 | << dendl; | |
4027 | } | |
4028 | ||
7c673cae FG |
4029 | // ======================================================= |
4030 | ||
31f18b77 FG |
4031 | // OmapIteratorImpl |
4032 | ||
4033 | #undef dout_prefix | |
4034 | #define dout_prefix *_dout << "bluestore.OmapIteratorImpl(" << this << ") " | |
4035 | ||
4036 | BlueStore::OmapIteratorImpl::OmapIteratorImpl( | |
4037 | CollectionRef c, OnodeRef o, KeyValueDB::Iterator it) | |
4038 | : c(c), o(o), it(it) | |
4039 | { | |
9f95a23c | 4040 | std::shared_lock l(c->lock); |
31f18b77 | 4041 | if (o->onode.has_omap()) { |
9f95a23c TL |
4042 | o->get_omap_key(string(), &head); |
4043 | o->get_omap_tail(&tail); | |
31f18b77 FG |
4044 | it->lower_bound(head); |
4045 | } | |
4046 | } | |
4047 | ||
11fdf7f2 TL |
4048 | string BlueStore::OmapIteratorImpl::_stringify() const |
4049 | { | |
4050 | stringstream s; | |
4051 | s << " omap_iterator(cid = " << c->cid | |
4052 | <<", oid = " << o->oid << ")"; | |
4053 | return s.str(); | |
4054 | } | |
4055 | ||
31f18b77 FG |
4056 | int BlueStore::OmapIteratorImpl::seek_to_first() |
4057 | { | |
9f95a23c | 4058 | std::shared_lock l(c->lock); |
11fdf7f2 | 4059 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4060 | if (o->onode.has_omap()) { |
4061 | it->lower_bound(head); | |
4062 | } else { | |
4063 | it = KeyValueDB::Iterator(); | |
4064 | } | |
494da23a TL |
4065 | c->store->log_latency( |
4066 | __func__, | |
11fdf7f2 TL |
4067 | l_bluestore_omap_seek_to_first_lat, |
4068 | mono_clock::now() - start1, | |
494da23a | 4069 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 | 4070 | |
31f18b77 FG |
4071 | return 0; |
4072 | } | |
4073 | ||
4074 | int BlueStore::OmapIteratorImpl::upper_bound(const string& after) | |
4075 | { | |
9f95a23c | 4076 | std::shared_lock l(c->lock); |
11fdf7f2 | 4077 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4078 | if (o->onode.has_omap()) { |
4079 | string key; | |
9f95a23c | 4080 | o->get_omap_key(after, &key); |
31f18b77 FG |
4081 | ldout(c->store->cct,20) << __func__ << " after " << after << " key " |
4082 | << pretty_binary_string(key) << dendl; | |
4083 | it->upper_bound(key); | |
4084 | } else { | |
4085 | it = KeyValueDB::Iterator(); | |
4086 | } | |
11fdf7f2 | 4087 | c->store->log_latency_fn( |
494da23a | 4088 | __func__, |
11fdf7f2 TL |
4089 | l_bluestore_omap_upper_bound_lat, |
4090 | mono_clock::now() - start1, | |
494da23a | 4091 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4092 | [&] (const ceph::timespan& lat) { |
494da23a | 4093 | return ", after = " + after + |
11fdf7f2 TL |
4094 | _stringify(); |
4095 | } | |
4096 | ); | |
31f18b77 FG |
4097 | return 0; |
4098 | } | |
4099 | ||
4100 | int BlueStore::OmapIteratorImpl::lower_bound(const string& to) | |
4101 | { | |
9f95a23c | 4102 | std::shared_lock l(c->lock); |
11fdf7f2 | 4103 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4104 | if (o->onode.has_omap()) { |
4105 | string key; | |
9f95a23c | 4106 | o->get_omap_key(to, &key); |
31f18b77 FG |
4107 | ldout(c->store->cct,20) << __func__ << " to " << to << " key " |
4108 | << pretty_binary_string(key) << dendl; | |
4109 | it->lower_bound(key); | |
4110 | } else { | |
4111 | it = KeyValueDB::Iterator(); | |
4112 | } | |
11fdf7f2 | 4113 | c->store->log_latency_fn( |
494da23a | 4114 | __func__, |
11fdf7f2 TL |
4115 | l_bluestore_omap_lower_bound_lat, |
4116 | mono_clock::now() - start1, | |
494da23a | 4117 | c->store->cct->_conf->bluestore_log_omap_iterator_age, |
11fdf7f2 | 4118 | [&] (const ceph::timespan& lat) { |
494da23a | 4119 | return ", to = " + to + |
11fdf7f2 TL |
4120 | _stringify(); |
4121 | } | |
4122 | ); | |
31f18b77 FG |
4123 | return 0; |
4124 | } | |
4125 | ||
4126 | bool BlueStore::OmapIteratorImpl::valid() | |
4127 | { | |
9f95a23c | 4128 | std::shared_lock l(c->lock); |
31f18b77 | 4129 | bool r = o->onode.has_omap() && it && it->valid() && |
494da23a | 4130 | it->raw_key().second < tail; |
31f18b77 FG |
4131 | if (it && it->valid()) { |
4132 | ldout(c->store->cct,20) << __func__ << " is at " | |
4133 | << pretty_binary_string(it->raw_key().second) | |
4134 | << dendl; | |
4135 | } | |
4136 | return r; | |
4137 | } | |
4138 | ||
11fdf7f2 | 4139 | int BlueStore::OmapIteratorImpl::next() |
31f18b77 | 4140 | { |
11fdf7f2 | 4141 | int r = -1; |
9f95a23c | 4142 | std::shared_lock l(c->lock); |
11fdf7f2 | 4143 | auto start1 = mono_clock::now(); |
31f18b77 FG |
4144 | if (o->onode.has_omap()) { |
4145 | it->next(); | |
11fdf7f2 | 4146 | r = 0; |
31f18b77 | 4147 | } |
494da23a TL |
4148 | c->store->log_latency( |
4149 | __func__, | |
11fdf7f2 TL |
4150 | l_bluestore_omap_next_lat, |
4151 | mono_clock::now() - start1, | |
494da23a | 4152 | c->store->cct->_conf->bluestore_log_omap_iterator_age); |
11fdf7f2 TL |
4153 | |
4154 | return r; | |
31f18b77 FG |
4155 | } |
4156 | ||
4157 | string BlueStore::OmapIteratorImpl::key() | |
4158 | { | |
9f95a23c | 4159 | std::shared_lock l(c->lock); |
11fdf7f2 | 4160 | ceph_assert(it->valid()); |
31f18b77 FG |
4161 | string db_key = it->raw_key().second; |
4162 | string user_key; | |
9f95a23c | 4163 | o->decode_omap_key(db_key, &user_key); |
494da23a | 4164 | |
31f18b77 FG |
4165 | return user_key; |
4166 | } | |
4167 | ||
4168 | bufferlist BlueStore::OmapIteratorImpl::value() | |
4169 | { | |
9f95a23c | 4170 | std::shared_lock l(c->lock); |
11fdf7f2 | 4171 | ceph_assert(it->valid()); |
31f18b77 FG |
4172 | return it->value(); |
4173 | } | |
4174 | ||
4175 | ||
4176 | // ===================================== | |
4177 | ||
7c673cae FG |
4178 | #undef dout_prefix |
4179 | #define dout_prefix *_dout << "bluestore(" << path << ") " | |
9f95a23c TL |
4180 | #undef dout_context |
4181 | #define dout_context cct | |
7c673cae FG |
4182 | |
4183 | ||
4184 | static void aio_cb(void *priv, void *priv2) | |
4185 | { | |
4186 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4187 | BlueStore::AioContext *c = static_cast<BlueStore::AioContext*>(priv2); | |
4188 | c->aio_finish(store); | |
4189 | } | |
4190 | ||
11fdf7f2 TL |
4191 | static void discard_cb(void *priv, void *priv2) |
4192 | { | |
4193 | BlueStore *store = static_cast<BlueStore*>(priv); | |
4194 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
4195 | store->handle_discard(*tmp); | |
4196 | } | |
4197 | ||
4198 | void BlueStore::handle_discard(interval_set<uint64_t>& to_release) | |
4199 | { | |
4200 | dout(10) << __func__ << dendl; | |
4201 | ceph_assert(alloc); | |
4202 | alloc->release(to_release); | |
4203 | } | |
4204 | ||
7c673cae | 4205 | BlueStore::BlueStore(CephContext *cct, const string& path) |
9f95a23c | 4206 | : BlueStore(cct, path, 0) {} |
7c673cae FG |
4207 | |
4208 | BlueStore::BlueStore(CephContext *cct, | |
4209 | const string& path, | |
4210 | uint64_t _min_alloc_size) | |
4211 | : ObjectStore(cct, path), | |
9f95a23c | 4212 | throttle(cct), |
11fdf7f2 | 4213 | finisher(cct, "commit_finisher", "cfin"), |
7c673cae | 4214 | kv_sync_thread(this), |
31f18b77 | 4215 | kv_finalize_thread(this), |
7c673cae FG |
4216 | min_alloc_size(_min_alloc_size), |
4217 | min_alloc_size_order(ctz(_min_alloc_size)), | |
4218 | mempool_thread(this) | |
4219 | { | |
4220 | _init_logger(); | |
11fdf7f2 | 4221 | cct->_conf.add_observer(this); |
7c673cae | 4222 | set_cache_shards(1); |
7c673cae FG |
4223 | } |
4224 | ||
4225 | BlueStore::~BlueStore() | |
4226 | { | |
11fdf7f2 | 4227 | cct->_conf.remove_observer(this); |
7c673cae | 4228 | _shutdown_logger(); |
11fdf7f2 TL |
4229 | ceph_assert(!mounted); |
4230 | ceph_assert(db == NULL); | |
4231 | ceph_assert(bluefs == NULL); | |
4232 | ceph_assert(fsid_fd < 0); | |
4233 | ceph_assert(path_fd < 0); | |
9f95a23c TL |
4234 | for (auto i : onode_cache_shards) { |
4235 | delete i; | |
4236 | } | |
4237 | for (auto i : buffer_cache_shards) { | |
7c673cae FG |
4238 | delete i; |
4239 | } | |
9f95a23c TL |
4240 | onode_cache_shards.clear(); |
4241 | buffer_cache_shards.clear(); | |
7c673cae FG |
4242 | } |
4243 | ||
4244 | const char **BlueStore::get_tracked_conf_keys() const | |
4245 | { | |
4246 | static const char* KEYS[] = { | |
4247 | "bluestore_csum_type", | |
4248 | "bluestore_compression_mode", | |
4249 | "bluestore_compression_algorithm", | |
4250 | "bluestore_compression_min_blob_size", | |
4251 | "bluestore_compression_min_blob_size_ssd", | |
4252 | "bluestore_compression_min_blob_size_hdd", | |
4253 | "bluestore_compression_max_blob_size", | |
4254 | "bluestore_compression_max_blob_size_ssd", | |
4255 | "bluestore_compression_max_blob_size_hdd", | |
c07f9fc5 | 4256 | "bluestore_compression_required_ratio", |
7c673cae FG |
4257 | "bluestore_max_alloc_size", |
4258 | "bluestore_prefer_deferred_size", | |
181888fb FG |
4259 | "bluestore_prefer_deferred_size_hdd", |
4260 | "bluestore_prefer_deferred_size_ssd", | |
31f18b77 FG |
4261 | "bluestore_deferred_batch_ops", |
4262 | "bluestore_deferred_batch_ops_hdd", | |
4263 | "bluestore_deferred_batch_ops_ssd", | |
7c673cae FG |
4264 | "bluestore_throttle_bytes", |
4265 | "bluestore_throttle_deferred_bytes", | |
4266 | "bluestore_throttle_cost_per_io_hdd", | |
4267 | "bluestore_throttle_cost_per_io_ssd", | |
4268 | "bluestore_throttle_cost_per_io", | |
4269 | "bluestore_max_blob_size", | |
4270 | "bluestore_max_blob_size_ssd", | |
4271 | "bluestore_max_blob_size_hdd", | |
11fdf7f2 TL |
4272 | "osd_memory_target", |
4273 | "osd_memory_target_cgroup_limit_ratio", | |
4274 | "osd_memory_base", | |
4275 | "osd_memory_cache_min", | |
92f5a8d4 | 4276 | "osd_memory_expected_fragmentation", |
11fdf7f2 TL |
4277 | "bluestore_cache_autotune", |
4278 | "bluestore_cache_autotune_interval", | |
81eedcae | 4279 | "bluestore_warn_on_legacy_statfs", |
9f95a23c TL |
4280 | "bluestore_warn_on_no_per_pool_omap", |
4281 | "bluestore_max_defer_interval", | |
7c673cae FG |
4282 | NULL |
4283 | }; | |
4284 | return KEYS; | |
4285 | } | |
4286 | ||
11fdf7f2 | 4287 | void BlueStore::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
4288 | const std::set<std::string> &changed) |
4289 | { | |
eafe8130 | 4290 | if (changed.count("bluestore_warn_on_legacy_statfs")) { |
81eedcae TL |
4291 | _check_legacy_statfs_alert(); |
4292 | } | |
9f95a23c TL |
4293 | if (changed.count("bluestore_warn_on_no_per_pool_omap")) { |
4294 | _check_no_per_pool_omap_alert(); | |
4295 | } | |
81eedcae | 4296 | |
7c673cae FG |
4297 | if (changed.count("bluestore_csum_type")) { |
4298 | _set_csum(); | |
4299 | } | |
4300 | if (changed.count("bluestore_compression_mode") || | |
4301 | changed.count("bluestore_compression_algorithm") || | |
4302 | changed.count("bluestore_compression_min_blob_size") || | |
4303 | changed.count("bluestore_compression_max_blob_size")) { | |
4304 | if (bdev) { | |
4305 | _set_compression(); | |
4306 | } | |
4307 | } | |
4308 | if (changed.count("bluestore_max_blob_size") || | |
4309 | changed.count("bluestore_max_blob_size_ssd") || | |
4310 | changed.count("bluestore_max_blob_size_hdd")) { | |
4311 | if (bdev) { | |
4312 | // only after startup | |
4313 | _set_blob_size(); | |
4314 | } | |
4315 | } | |
4316 | if (changed.count("bluestore_prefer_deferred_size") || | |
181888fb FG |
4317 | changed.count("bluestore_prefer_deferred_size_hdd") || |
4318 | changed.count("bluestore_prefer_deferred_size_ssd") || | |
7c673cae FG |
4319 | changed.count("bluestore_max_alloc_size") || |
4320 | changed.count("bluestore_deferred_batch_ops") || | |
4321 | changed.count("bluestore_deferred_batch_ops_hdd") || | |
4322 | changed.count("bluestore_deferred_batch_ops_ssd")) { | |
4323 | if (bdev) { | |
4324 | // only after startup | |
4325 | _set_alloc_sizes(); | |
4326 | } | |
4327 | } | |
4328 | if (changed.count("bluestore_throttle_cost_per_io") || | |
4329 | changed.count("bluestore_throttle_cost_per_io_hdd") || | |
4330 | changed.count("bluestore_throttle_cost_per_io_ssd")) { | |
4331 | if (bdev) { | |
4332 | _set_throttle_params(); | |
4333 | } | |
4334 | } | |
9f95a23c TL |
4335 | if (changed.count("bluestore_throttle_bytes") || |
4336 | changed.count("bluestore_throttle_deferred_bytes") || | |
4337 | changed.count("bluestore_throttle_trace_rate")) { | |
4338 | throttle.reset_throttle(conf); | |
7c673cae | 4339 | } |
9f95a23c TL |
4340 | if (changed.count("bluestore_max_defer_interval")) { |
4341 | if (bdev) { | |
4342 | _set_max_defer_interval(); | |
4343 | } | |
7c673cae | 4344 | } |
92f5a8d4 TL |
4345 | if (changed.count("osd_memory_target") || |
4346 | changed.count("osd_memory_base") || | |
4347 | changed.count("osd_memory_cache_min") || | |
4348 | changed.count("osd_memory_expected_fragmentation")) { | |
4349 | _update_osd_memory_options(); | |
4350 | } | |
7c673cae FG |
4351 | } |
4352 | ||
4353 | void BlueStore::_set_compression() | |
4354 | { | |
224ce89b WB |
4355 | auto m = Compressor::get_comp_mode_type(cct->_conf->bluestore_compression_mode); |
4356 | if (m) { | |
11fdf7f2 | 4357 | _clear_compression_alert(); |
224ce89b WB |
4358 | comp_mode = *m; |
4359 | } else { | |
4360 | derr << __func__ << " unrecognized value '" | |
4361 | << cct->_conf->bluestore_compression_mode | |
4362 | << "' for bluestore_compression_mode, reverting to 'none'" | |
4363 | << dendl; | |
4364 | comp_mode = Compressor::COMP_NONE; | |
11fdf7f2 TL |
4365 | string s("unknown mode: "); |
4366 | s += cct->_conf->bluestore_compression_mode; | |
4367 | _set_compression_alert(true, s.c_str()); | |
224ce89b WB |
4368 | } |
4369 | ||
4370 | compressor = nullptr; | |
4371 | ||
3efd9988 FG |
4372 | if (cct->_conf->bluestore_compression_min_blob_size) { |
4373 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size; | |
7c673cae | 4374 | } else { |
11fdf7f2 | 4375 | ceph_assert(bdev); |
9f95a23c | 4376 | if (_use_rotational_settings()) { |
7c673cae FG |
4377 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_hdd; |
4378 | } else { | |
4379 | comp_min_blob_size = cct->_conf->bluestore_compression_min_blob_size_ssd; | |
4380 | } | |
4381 | } | |
4382 | ||
4383 | if (cct->_conf->bluestore_compression_max_blob_size) { | |
4384 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size; | |
4385 | } else { | |
11fdf7f2 | 4386 | ceph_assert(bdev); |
9f95a23c | 4387 | if (_use_rotational_settings()) { |
7c673cae FG |
4388 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_hdd; |
4389 | } else { | |
4390 | comp_max_blob_size = cct->_conf->bluestore_compression_max_blob_size_ssd; | |
4391 | } | |
4392 | } | |
4393 | ||
7c673cae FG |
4394 | auto& alg_name = cct->_conf->bluestore_compression_algorithm; |
4395 | if (!alg_name.empty()) { | |
4396 | compressor = Compressor::create(cct, alg_name); | |
4397 | if (!compressor) { | |
4398 | derr << __func__ << " unable to initialize " << alg_name.c_str() << " compressor" | |
4399 | << dendl; | |
11fdf7f2 | 4400 | _set_compression_alert(false, alg_name.c_str()); |
7c673cae FG |
4401 | } |
4402 | } | |
4403 | ||
4404 | dout(10) << __func__ << " mode " << Compressor::get_comp_mode_name(comp_mode) | |
4405 | << " alg " << (compressor ? compressor->get_type_name() : "(none)") | |
11fdf7f2 TL |
4406 | << " min_blob " << comp_min_blob_size |
4407 | << " max_blob " << comp_max_blob_size | |
7c673cae FG |
4408 | << dendl; |
4409 | } | |
4410 | ||
4411 | void BlueStore::_set_csum() | |
4412 | { | |
4413 | csum_type = Checksummer::CSUM_NONE; | |
4414 | int t = Checksummer::get_csum_string_type(cct->_conf->bluestore_csum_type); | |
4415 | if (t > Checksummer::CSUM_NONE) | |
4416 | csum_type = t; | |
4417 | ||
4418 | dout(10) << __func__ << " csum_type " | |
4419 | << Checksummer::get_csum_type_string(csum_type) | |
4420 | << dendl; | |
4421 | } | |
4422 | ||
4423 | void BlueStore::_set_throttle_params() | |
4424 | { | |
4425 | if (cct->_conf->bluestore_throttle_cost_per_io) { | |
4426 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io; | |
4427 | } else { | |
11fdf7f2 | 4428 | ceph_assert(bdev); |
9f95a23c | 4429 | if (_use_rotational_settings()) { |
7c673cae FG |
4430 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_hdd; |
4431 | } else { | |
4432 | throttle_cost_per_io = cct->_conf->bluestore_throttle_cost_per_io_ssd; | |
4433 | } | |
4434 | } | |
4435 | ||
4436 | dout(10) << __func__ << " throttle_cost_per_io " << throttle_cost_per_io | |
4437 | << dendl; | |
4438 | } | |
4439 | void BlueStore::_set_blob_size() | |
4440 | { | |
4441 | if (cct->_conf->bluestore_max_blob_size) { | |
4442 | max_blob_size = cct->_conf->bluestore_max_blob_size; | |
4443 | } else { | |
11fdf7f2 | 4444 | ceph_assert(bdev); |
9f95a23c | 4445 | if (_use_rotational_settings()) { |
7c673cae FG |
4446 | max_blob_size = cct->_conf->bluestore_max_blob_size_hdd; |
4447 | } else { | |
4448 | max_blob_size = cct->_conf->bluestore_max_blob_size_ssd; | |
4449 | } | |
4450 | } | |
4451 | dout(10) << __func__ << " max_blob_size 0x" << std::hex << max_blob_size | |
4452 | << std::dec << dendl; | |
4453 | } | |
4454 | ||
92f5a8d4 TL |
4455 | void BlueStore::_update_osd_memory_options() |
4456 | { | |
4457 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4458 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
4459 | osd_memory_expected_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
4460 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
4461 | config_changed++; | |
4462 | dout(10) << __func__ | |
4463 | << " osd_memory_target " << osd_memory_target | |
4464 | << " osd_memory_base " << osd_memory_base | |
4465 | << " osd_memory_expected_fragmentation " << osd_memory_expected_fragmentation | |
4466 | << " osd_memory_cache_min " << osd_memory_cache_min | |
4467 | << dendl; | |
4468 | } | |
4469 | ||
11fdf7f2 | 4470 | int BlueStore::_set_cache_sizes() |
1adf2230 | 4471 | { |
11fdf7f2 TL |
4472 | ceph_assert(bdev); |
4473 | cache_autotune = cct->_conf.get_val<bool>("bluestore_cache_autotune"); | |
91327a77 | 4474 | cache_autotune_interval = |
11fdf7f2 TL |
4475 | cct->_conf.get_val<double>("bluestore_cache_autotune_interval"); |
4476 | osd_memory_target = cct->_conf.get_val<Option::size_t>("osd_memory_target"); | |
4477 | osd_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
91327a77 | 4478 | osd_memory_expected_fragmentation = |
11fdf7f2 TL |
4479 | cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); |
4480 | osd_memory_cache_min = cct->_conf.get_val<Option::size_t>("osd_memory_cache_min"); | |
91327a77 | 4481 | osd_memory_cache_resize_interval = |
11fdf7f2 | 4482 | cct->_conf.get_val<double>("osd_memory_cache_resize_interval"); |
91327a77 | 4483 | |
224ce89b WB |
4484 | if (cct->_conf->bluestore_cache_size) { |
4485 | cache_size = cct->_conf->bluestore_cache_size; | |
4486 | } else { | |
4487 | // choose global cache size based on backend type | |
9f95a23c | 4488 | if (_use_rotational_settings()) { |
224ce89b WB |
4489 | cache_size = cct->_conf->bluestore_cache_size_hdd; |
4490 | } else { | |
4491 | cache_size = cct->_conf->bluestore_cache_size_ssd; | |
4492 | } | |
4493 | } | |
31f18b77 | 4494 | |
91327a77 | 4495 | cache_meta_ratio = cct->_conf->bluestore_cache_meta_ratio; |
224ce89b | 4496 | if (cache_meta_ratio < 0 || cache_meta_ratio > 1.0) { |
d2e6a577 | 4497 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 | 4498 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4499 | return -EINVAL; |
4500 | } | |
91327a77 AA |
4501 | |
4502 | cache_kv_ratio = cct->_conf->bluestore_cache_kv_ratio; | |
224ce89b | 4503 | if (cache_kv_ratio < 0 || cache_kv_ratio > 1.0) { |
d2e6a577 | 4504 | derr << __func__ << " bluestore_cache_kv_ratio (" << cache_kv_ratio |
91327a77 | 4505 | << ") must be in range [0,1.0]" << dendl; |
31f18b77 FG |
4506 | return -EINVAL; |
4507 | } | |
91327a77 | 4508 | |
31f18b77 | 4509 | if (cache_meta_ratio + cache_kv_ratio > 1.0) { |
d2e6a577 | 4510 | derr << __func__ << " bluestore_cache_meta_ratio (" << cache_meta_ratio |
91327a77 AA |
4511 | << ") + bluestore_cache_kv_ratio (" << cache_kv_ratio |
4512 | << ") = " << cache_meta_ratio + cache_kv_ratio << "; must be <= 1.0" | |
4513 | << dendl; | |
31f18b77 FG |
4514 | return -EINVAL; |
4515 | } | |
91327a77 AA |
4516 | |
4517 | cache_data_ratio = | |
4518 | (double)1.0 - (double)cache_meta_ratio - (double)cache_kv_ratio; | |
31f18b77 FG |
4519 | if (cache_data_ratio < 0) { |
4520 | // deal with floating point imprecision | |
4521 | cache_data_ratio = 0; | |
4522 | } | |
91327a77 | 4523 | |
224ce89b WB |
4524 | dout(1) << __func__ << " cache_size " << cache_size |
4525 | << " meta " << cache_meta_ratio | |
31f18b77 FG |
4526 | << " kv " << cache_kv_ratio |
4527 | << " data " << cache_data_ratio | |
4528 | << dendl; | |
4529 | return 0; | |
4530 | } | |
4531 | ||
3efd9988 FG |
4532 | int BlueStore::write_meta(const std::string& key, const std::string& value) |
4533 | { | |
4534 | bluestore_bdev_label_t label; | |
4535 | string p = path + "/block"; | |
4536 | int r = _read_bdev_label(cct, p, &label); | |
4537 | if (r < 0) { | |
4538 | return ObjectStore::write_meta(key, value); | |
4539 | } | |
4540 | label.meta[key] = value; | |
4541 | r = _write_bdev_label(cct, p, label); | |
11fdf7f2 | 4542 | ceph_assert(r == 0); |
3efd9988 FG |
4543 | return ObjectStore::write_meta(key, value); |
4544 | } | |
4545 | ||
4546 | int BlueStore::read_meta(const std::string& key, std::string *value) | |
4547 | { | |
4548 | bluestore_bdev_label_t label; | |
4549 | string p = path + "/block"; | |
4550 | int r = _read_bdev_label(cct, p, &label); | |
4551 | if (r < 0) { | |
4552 | return ObjectStore::read_meta(key, value); | |
4553 | } | |
4554 | auto i = label.meta.find(key); | |
4555 | if (i == label.meta.end()) { | |
4556 | return ObjectStore::read_meta(key, value); | |
4557 | } | |
4558 | *value = i->second; | |
4559 | return 0; | |
4560 | } | |
4561 | ||
7c673cae FG |
4562 | void BlueStore::_init_logger() |
4563 | { | |
4564 | PerfCountersBuilder b(cct, "bluestore", | |
4565 | l_bluestore_first, l_bluestore_last); | |
4566 | b.add_time_avg(l_bluestore_kv_flush_lat, "kv_flush_lat", | |
4567 | "Average kv_thread flush latency", | |
4568 | "fl_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4569 | b.add_time_avg(l_bluestore_kv_commit_lat, "kv_commit_lat", | |
4570 | "Average kv_thread commit latency"); | |
11fdf7f2 TL |
4571 | b.add_time_avg(l_bluestore_kv_sync_lat, "kv_sync_lat", |
4572 | "Average kv_sync thread latency", | |
4573 | "ks_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4574 | b.add_time_avg(l_bluestore_kv_final_lat, "kv_final_lat", | |
4575 | "Average kv_finalize thread latency", | |
4576 | "kf_l", PerfCountersBuilder::PRIO_INTERESTING); | |
7c673cae FG |
4577 | b.add_time_avg(l_bluestore_state_prepare_lat, "state_prepare_lat", |
4578 | "Average prepare state latency"); | |
4579 | b.add_time_avg(l_bluestore_state_aio_wait_lat, "state_aio_wait_lat", | |
4580 | "Average aio_wait state latency", | |
4581 | "io_l", PerfCountersBuilder::PRIO_INTERESTING); | |
4582 | b.add_time_avg(l_bluestore_state_io_done_lat, "state_io_done_lat", | |
4583 | "Average io_done state latency"); | |
4584 | b.add_time_avg(l_bluestore_state_kv_queued_lat, "state_kv_queued_lat", | |
4585 | "Average kv_queued state latency"); | |
4586 | b.add_time_avg(l_bluestore_state_kv_committing_lat, "state_kv_commiting_lat", | |
4587 | "Average kv_commiting state latency"); | |
4588 | b.add_time_avg(l_bluestore_state_kv_done_lat, "state_kv_done_lat", | |
4589 | "Average kv_done state latency"); | |
4590 | b.add_time_avg(l_bluestore_state_deferred_queued_lat, "state_deferred_queued_lat", | |
4591 | "Average deferred_queued state latency"); | |
4592 | b.add_time_avg(l_bluestore_state_deferred_aio_wait_lat, "state_deferred_aio_wait_lat", | |
4593 | "Average aio_wait state latency"); | |
4594 | b.add_time_avg(l_bluestore_state_deferred_cleanup_lat, "state_deferred_cleanup_lat", | |
4595 | "Average cleanup state latency"); | |
4596 | b.add_time_avg(l_bluestore_state_finishing_lat, "state_finishing_lat", | |
4597 | "Average finishing state latency"); | |
4598 | b.add_time_avg(l_bluestore_state_done_lat, "state_done_lat", | |
4599 | "Average done state latency"); | |
4600 | b.add_time_avg(l_bluestore_throttle_lat, "throttle_lat", | |
4601 | "Average submit throttle latency", | |
4602 | "th_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4603 | b.add_time_avg(l_bluestore_submit_lat, "submit_lat", | |
4604 | "Average submit latency", | |
4605 | "s_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4606 | b.add_time_avg(l_bluestore_commit_lat, "commit_lat", | |
4607 | "Average commit latency", | |
4608 | "c_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4609 | b.add_time_avg(l_bluestore_read_lat, "read_lat", | |
4610 | "Average read latency", | |
4611 | "r_l", PerfCountersBuilder::PRIO_CRITICAL); | |
4612 | b.add_time_avg(l_bluestore_read_onode_meta_lat, "read_onode_meta_lat", | |
4613 | "Average read onode metadata latency"); | |
4614 | b.add_time_avg(l_bluestore_read_wait_aio_lat, "read_wait_aio_lat", | |
4615 | "Average read latency"); | |
4616 | b.add_time_avg(l_bluestore_compress_lat, "compress_lat", | |
4617 | "Average compress latency"); | |
4618 | b.add_time_avg(l_bluestore_decompress_lat, "decompress_lat", | |
4619 | "Average decompress latency"); | |
4620 | b.add_time_avg(l_bluestore_csum_lat, "csum_lat", | |
4621 | "Average checksum latency"); | |
4622 | b.add_u64_counter(l_bluestore_compress_success_count, "compress_success_count", | |
4623 | "Sum for beneficial compress ops"); | |
4624 | b.add_u64_counter(l_bluestore_compress_rejected_count, "compress_rejected_count", | |
4625 | "Sum for compress ops rejected due to low net gain of space"); | |
4626 | b.add_u64_counter(l_bluestore_write_pad_bytes, "write_pad_bytes", | |
11fdf7f2 | 4627 | "Sum for write-op padded bytes", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4628 | b.add_u64_counter(l_bluestore_deferred_write_ops, "deferred_write_ops", |
4629 | "Sum for deferred write op"); | |
4630 | b.add_u64_counter(l_bluestore_deferred_write_bytes, "deferred_write_bytes", | |
11fdf7f2 | 4631 | "Sum for deferred write bytes", "def", 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4632 | b.add_u64_counter(l_bluestore_write_penalty_read_ops, "write_penalty_read_ops", |
4633 | "Sum for write penalty read ops"); | |
4634 | b.add_u64(l_bluestore_allocated, "bluestore_allocated", | |
4635 | "Sum for allocated bytes"); | |
4636 | b.add_u64(l_bluestore_stored, "bluestore_stored", | |
4637 | "Sum for stored bytes"); | |
4638 | b.add_u64(l_bluestore_compressed, "bluestore_compressed", | |
92f5a8d4 TL |
4639 | "Sum for stored compressed bytes", |
4640 | "c", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4641 | b.add_u64(l_bluestore_compressed_allocated, "bluestore_compressed_allocated", |
92f5a8d4 TL |
4642 | "Sum for bytes allocated for compressed data", |
4643 | "c_a", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae | 4644 | b.add_u64(l_bluestore_compressed_original, "bluestore_compressed_original", |
92f5a8d4 TL |
4645 | "Sum for original bytes that were compressed", |
4646 | "c_o", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
7c673cae FG |
4647 | b.add_u64(l_bluestore_onodes, "bluestore_onodes", |
4648 | "Number of onodes in cache"); | |
9f95a23c TL |
4649 | b.add_u64(l_bluestore_pinned_onodes, "bluestore_pinned_onodes", |
4650 | "Number of pinned onodes in cache"); | |
7c673cae FG |
4651 | b.add_u64_counter(l_bluestore_onode_hits, "bluestore_onode_hits", |
4652 | "Sum for onode-lookups hit in the cache"); | |
4653 | b.add_u64_counter(l_bluestore_onode_misses, "bluestore_onode_misses", | |
4654 | "Sum for onode-lookups missed in the cache"); | |
4655 | b.add_u64_counter(l_bluestore_onode_shard_hits, "bluestore_onode_shard_hits", | |
4656 | "Sum for onode-shard lookups hit in the cache"); | |
4657 | b.add_u64_counter(l_bluestore_onode_shard_misses, | |
4658 | "bluestore_onode_shard_misses", | |
4659 | "Sum for onode-shard lookups missed in the cache"); | |
4660 | b.add_u64(l_bluestore_extents, "bluestore_extents", | |
4661 | "Number of extents in cache"); | |
4662 | b.add_u64(l_bluestore_blobs, "bluestore_blobs", | |
4663 | "Number of blobs in cache"); | |
4664 | b.add_u64(l_bluestore_buffers, "bluestore_buffers", | |
4665 | "Number of buffers in cache"); | |
4666 | b.add_u64(l_bluestore_buffer_bytes, "bluestore_buffer_bytes", | |
11fdf7f2 | 4667 | "Number of buffer bytes in cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4668 | b.add_u64_counter(l_bluestore_buffer_hit_bytes, "bluestore_buffer_hit_bytes", |
11fdf7f2 | 4669 | "Sum for bytes of read hit in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
91327a77 | 4670 | b.add_u64_counter(l_bluestore_buffer_miss_bytes, "bluestore_buffer_miss_bytes", |
11fdf7f2 | 4671 | "Sum for bytes of read missed in the cache", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4672 | |
4673 | b.add_u64_counter(l_bluestore_write_big, "bluestore_write_big", | |
4674 | "Large aligned writes into fresh blobs"); | |
4675 | b.add_u64_counter(l_bluestore_write_big_bytes, "bluestore_write_big_bytes", | |
11fdf7f2 | 4676 | "Large aligned writes into fresh blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4677 | b.add_u64_counter(l_bluestore_write_big_blobs, "bluestore_write_big_blobs", |
4678 | "Large aligned writes into fresh blobs (blobs)"); | |
4679 | b.add_u64_counter(l_bluestore_write_small, "bluestore_write_small", | |
4680 | "Small writes into existing or sparse small blobs"); | |
4681 | b.add_u64_counter(l_bluestore_write_small_bytes, "bluestore_write_small_bytes", | |
11fdf7f2 | 4682 | "Small writes into existing or sparse small blobs (bytes)", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
4683 | b.add_u64_counter(l_bluestore_write_small_unused, |
4684 | "bluestore_write_small_unused", | |
4685 | "Small writes into unused portion of existing blob"); | |
4686 | b.add_u64_counter(l_bluestore_write_small_deferred, | |
4687 | "bluestore_write_small_deferred", | |
4688 | "Small overwrites using deferred"); | |
4689 | b.add_u64_counter(l_bluestore_write_small_pre_read, | |
4690 | "bluestore_write_small_pre_read", | |
4691 | "Small writes that required we read some data (possibly " | |
4692 | "cached) to fill out the block"); | |
4693 | b.add_u64_counter(l_bluestore_write_small_new, "bluestore_write_small_new", | |
4694 | "Small write into new (sparse) blob"); | |
4695 | ||
4696 | b.add_u64_counter(l_bluestore_txc, "bluestore_txc", "Transactions committed"); | |
4697 | b.add_u64_counter(l_bluestore_onode_reshard, "bluestore_onode_reshard", | |
4698 | "Onode extent map reshard events"); | |
4699 | b.add_u64_counter(l_bluestore_blob_split, "bluestore_blob_split", | |
4700 | "Sum for blob splitting due to resharding"); | |
4701 | b.add_u64_counter(l_bluestore_extent_compress, "bluestore_extent_compress", | |
4702 | "Sum for extents that have been removed due to compression"); | |
4703 | b.add_u64_counter(l_bluestore_gc_merged, "bluestore_gc_merged", | |
4704 | "Sum for extents that have been merged due to garbage " | |
4705 | "collection"); | |
b32b8144 FG |
4706 | b.add_u64_counter(l_bluestore_read_eio, "bluestore_read_eio", |
4707 | "Read EIO errors propagated to high level callers"); | |
f64942e4 AA |
4708 | b.add_u64_counter(l_bluestore_reads_with_retries, "bluestore_reads_with_retries", |
4709 | "Read operations that required at least one retry due to failed checksum validation"); | |
a8e16298 TL |
4710 | b.add_u64(l_bluestore_fragmentation, "bluestore_fragmentation_micros", |
4711 | "How fragmented bluestore free space is (free extents / max possible number of free extents) * 1000"); | |
11fdf7f2 TL |
4712 | b.add_time_avg(l_bluestore_omap_seek_to_first_lat, "omap_seek_to_first_lat", |
4713 | "Average omap iterator seek_to_first call latency"); | |
4714 | b.add_time_avg(l_bluestore_omap_upper_bound_lat, "omap_upper_bound_lat", | |
4715 | "Average omap iterator upper_bound call latency"); | |
4716 | b.add_time_avg(l_bluestore_omap_lower_bound_lat, "omap_lower_bound_lat", | |
4717 | "Average omap iterator lower_bound call latency"); | |
4718 | b.add_time_avg(l_bluestore_omap_next_lat, "omap_next_lat", | |
4719 | "Average omap iterator next call latency"); | |
494da23a TL |
4720 | b.add_time_avg(l_bluestore_clist_lat, "clist_lat", |
4721 | "Average collection listing latency"); | |
7c673cae FG |
4722 | logger = b.create_perf_counters(); |
4723 | cct->get_perfcounters_collection()->add(logger); | |
4724 | } | |
4725 | ||
4726 | int BlueStore::_reload_logger() | |
4727 | { | |
4728 | struct store_statfs_t store_statfs; | |
7c673cae | 4729 | int r = statfs(&store_statfs); |
11fdf7f2 | 4730 | if (r >= 0) { |
7c673cae | 4731 | logger->set(l_bluestore_allocated, store_statfs.allocated); |
11fdf7f2 TL |
4732 | logger->set(l_bluestore_stored, store_statfs.data_stored); |
4733 | logger->set(l_bluestore_compressed, store_statfs.data_compressed); | |
4734 | logger->set(l_bluestore_compressed_allocated, store_statfs.data_compressed_allocated); | |
4735 | logger->set(l_bluestore_compressed_original, store_statfs.data_compressed_original); | |
7c673cae FG |
4736 | } |
4737 | return r; | |
4738 | } | |
4739 | ||
4740 | void BlueStore::_shutdown_logger() | |
4741 | { | |
4742 | cct->get_perfcounters_collection()->remove(logger); | |
4743 | delete logger; | |
4744 | } | |
4745 | ||
4746 | int BlueStore::get_block_device_fsid(CephContext* cct, const string& path, | |
4747 | uuid_d *fsid) | |
4748 | { | |
4749 | bluestore_bdev_label_t label; | |
4750 | int r = _read_bdev_label(cct, path, &label); | |
4751 | if (r < 0) | |
4752 | return r; | |
4753 | *fsid = label.osd_uuid; | |
4754 | return 0; | |
4755 | } | |
4756 | ||
4757 | int BlueStore::_open_path() | |
4758 | { | |
b32b8144 | 4759 | // sanity check(s) |
11fdf7f2 | 4760 | ceph_assert(path_fd < 0); |
91327a77 | 4761 | path_fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_DIRECTORY|O_CLOEXEC)); |
7c673cae FG |
4762 | if (path_fd < 0) { |
4763 | int r = -errno; | |
4764 | derr << __func__ << " unable to open " << path << ": " << cpp_strerror(r) | |
4765 | << dendl; | |
4766 | return r; | |
4767 | } | |
4768 | return 0; | |
4769 | } | |
4770 | ||
4771 | void BlueStore::_close_path() | |
4772 | { | |
4773 | VOID_TEMP_FAILURE_RETRY(::close(path_fd)); | |
4774 | path_fd = -1; | |
4775 | } | |
4776 | ||
3efd9988 FG |
4777 | int BlueStore::_write_bdev_label(CephContext *cct, |
4778 | string path, bluestore_bdev_label_t label) | |
7c673cae FG |
4779 | { |
4780 | dout(10) << __func__ << " path " << path << " label " << label << dendl; | |
4781 | bufferlist bl; | |
11fdf7f2 | 4782 | encode(label, bl); |
7c673cae | 4783 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 TL |
4784 | encode(crc, bl); |
4785 | ceph_assert(bl.length() <= BDEV_LABEL_BLOCK_SIZE); | |
7c673cae FG |
4786 | bufferptr z(BDEV_LABEL_BLOCK_SIZE - bl.length()); |
4787 | z.zero(); | |
4788 | bl.append(std::move(z)); | |
4789 | ||
91327a77 | 4790 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_WRONLY|O_CLOEXEC)); |
7c673cae FG |
4791 | if (fd < 0) { |
4792 | fd = -errno; | |
4793 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
4794 | << dendl; | |
4795 | return fd; | |
4796 | } | |
4797 | int r = bl.write_fd(fd); | |
4798 | if (r < 0) { | |
4799 | derr << __func__ << " failed to write to " << path | |
4800 | << ": " << cpp_strerror(r) << dendl; | |
11fdf7f2 | 4801 | goto out; |
7c673cae | 4802 | } |
3efd9988 FG |
4803 | r = ::fsync(fd); |
4804 | if (r < 0) { | |
4805 | derr << __func__ << " failed to fsync " << path | |
4806 | << ": " << cpp_strerror(r) << dendl; | |
4807 | } | |
11fdf7f2 | 4808 | out: |
7c673cae FG |
4809 | VOID_TEMP_FAILURE_RETRY(::close(fd)); |
4810 | return r; | |
4811 | } | |
4812 | ||
4813 | int BlueStore::_read_bdev_label(CephContext* cct, string path, | |
4814 | bluestore_bdev_label_t *label) | |
4815 | { | |
4816 | dout(10) << __func__ << dendl; | |
91327a77 | 4817 | int fd = TEMP_FAILURE_RETRY(::open(path.c_str(), O_RDONLY|O_CLOEXEC)); |
7c673cae FG |
4818 | if (fd < 0) { |
4819 | fd = -errno; | |
4820 | derr << __func__ << " failed to open " << path << ": " << cpp_strerror(fd) | |
4821 | << dendl; | |
4822 | return fd; | |
4823 | } | |
4824 | bufferlist bl; | |
4825 | int r = bl.read_fd(fd, BDEV_LABEL_BLOCK_SIZE); | |
4826 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
4827 | if (r < 0) { | |
4828 | derr << __func__ << " failed to read from " << path | |
4829 | << ": " << cpp_strerror(r) << dendl; | |
4830 | return r; | |
4831 | } | |
4832 | ||
4833 | uint32_t crc, expected_crc; | |
11fdf7f2 | 4834 | auto p = bl.cbegin(); |
7c673cae | 4835 | try { |
11fdf7f2 | 4836 | decode(*label, p); |
7c673cae FG |
4837 | bufferlist t; |
4838 | t.substr_of(bl, 0, p.get_off()); | |
4839 | crc = t.crc32c(-1); | |
11fdf7f2 | 4840 | decode(expected_crc, p); |
7c673cae FG |
4841 | } |
4842 | catch (buffer::error& e) { | |
b32b8144 | 4843 | dout(2) << __func__ << " unable to decode label at offset " << p.get_off() |
7c673cae FG |
4844 | << ": " << e.what() |
4845 | << dendl; | |
b32b8144 | 4846 | return -ENOENT; |
7c673cae FG |
4847 | } |
4848 | if (crc != expected_crc) { | |
4849 | derr << __func__ << " bad crc on label, expected " << expected_crc | |
4850 | << " != actual " << crc << dendl; | |
4851 | return -EIO; | |
4852 | } | |
4853 | dout(10) << __func__ << " got " << *label << dendl; | |
4854 | return 0; | |
4855 | } | |
4856 | ||
4857 | int BlueStore::_check_or_set_bdev_label( | |
4858 | string path, uint64_t size, string desc, bool create) | |
4859 | { | |
4860 | bluestore_bdev_label_t label; | |
4861 | if (create) { | |
4862 | label.osd_uuid = fsid; | |
4863 | label.size = size; | |
4864 | label.btime = ceph_clock_now(); | |
4865 | label.description = desc; | |
3efd9988 | 4866 | int r = _write_bdev_label(cct, path, label); |
7c673cae FG |
4867 | if (r < 0) |
4868 | return r; | |
4869 | } else { | |
4870 | int r = _read_bdev_label(cct, path, &label); | |
4871 | if (r < 0) | |
4872 | return r; | |
31f18b77 FG |
4873 | if (cct->_conf->bluestore_debug_permit_any_bdev_label) { |
4874 | dout(20) << __func__ << " bdev " << path << " fsid " << label.osd_uuid | |
4875 | << " and fsid " << fsid << " check bypassed" << dendl; | |
1911f103 | 4876 | } else if (label.osd_uuid != fsid) { |
7c673cae FG |
4877 | derr << __func__ << " bdev " << path << " fsid " << label.osd_uuid |
4878 | << " does not match our fsid " << fsid << dendl; | |
4879 | return -EIO; | |
4880 | } | |
4881 | } | |
4882 | return 0; | |
4883 | } | |
4884 | ||
4885 | void BlueStore::_set_alloc_sizes(void) | |
4886 | { | |
7c673cae FG |
4887 | max_alloc_size = cct->_conf->bluestore_max_alloc_size; |
4888 | ||
4889 | if (cct->_conf->bluestore_prefer_deferred_size) { | |
4890 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size; | |
4891 | } else { | |
11fdf7f2 | 4892 | ceph_assert(bdev); |
9f95a23c | 4893 | if (_use_rotational_settings()) { |
7c673cae FG |
4894 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_hdd; |
4895 | } else { | |
4896 | prefer_deferred_size = cct->_conf->bluestore_prefer_deferred_size_ssd; | |
4897 | } | |
4898 | } | |
4899 | ||
4900 | if (cct->_conf->bluestore_deferred_batch_ops) { | |
4901 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops; | |
4902 | } else { | |
11fdf7f2 | 4903 | ceph_assert(bdev); |
9f95a23c | 4904 | if (_use_rotational_settings()) { |
7c673cae FG |
4905 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_hdd; |
4906 | } else { | |
4907 | deferred_batch_ops = cct->_conf->bluestore_deferred_batch_ops_ssd; | |
4908 | } | |
4909 | } | |
4910 | ||
4911 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11fdf7f2 | 4912 | << std::dec << " order " << (int)min_alloc_size_order |
7c673cae FG |
4913 | << " max_alloc_size 0x" << std::hex << max_alloc_size |
4914 | << " prefer_deferred_size 0x" << prefer_deferred_size | |
4915 | << std::dec | |
4916 | << " deferred_batch_ops " << deferred_batch_ops | |
4917 | << dendl; | |
4918 | } | |
4919 | ||
4920 | int BlueStore::_open_bdev(bool create) | |
4921 | { | |
11fdf7f2 | 4922 | ceph_assert(bdev == NULL); |
7c673cae | 4923 | string p = path + "/block"; |
11fdf7f2 | 4924 | bdev = BlockDevice::create(cct, p, aio_cb, static_cast<void*>(this), discard_cb, static_cast<void*>(this)); |
7c673cae FG |
4925 | int r = bdev->open(p); |
4926 | if (r < 0) | |
4927 | goto fail; | |
4928 | ||
11fdf7f2 TL |
4929 | if (create && cct->_conf->bdev_enable_discard) { |
4930 | bdev->discard(0, bdev->get_size()); | |
4931 | } | |
4932 | ||
7c673cae FG |
4933 | if (bdev->supported_bdev_label()) { |
4934 | r = _check_or_set_bdev_label(p, bdev->get_size(), "main", create); | |
4935 | if (r < 0) | |
4936 | goto fail_close; | |
4937 | } | |
4938 | ||
4939 | // initialize global block parameters | |
4940 | block_size = bdev->get_block_size(); | |
4941 | block_mask = ~(block_size - 1); | |
4942 | block_size_order = ctz(block_size); | |
11fdf7f2 | 4943 | ceph_assert(block_size == 1u << block_size_order); |
9f95a23c | 4944 | _set_max_defer_interval(); |
224ce89b WB |
4945 | // and set cache_size based on device type |
4946 | r = _set_cache_sizes(); | |
4947 | if (r < 0) { | |
4948 | goto fail_close; | |
4949 | } | |
7c673cae FG |
4950 | return 0; |
4951 | ||
4952 | fail_close: | |
4953 | bdev->close(); | |
4954 | fail: | |
4955 | delete bdev; | |
4956 | bdev = NULL; | |
4957 | return r; | |
4958 | } | |
4959 | ||
11fdf7f2 TL |
4960 | void BlueStore::_validate_bdev() |
4961 | { | |
4962 | ceph_assert(bdev); | |
4963 | ceph_assert(min_alloc_size); // _get_odisk_reserved depends on that | |
4964 | uint64_t dev_size = bdev->get_size(); | |
4965 | if (dev_size < | |
4966 | _get_ondisk_reserved() + cct->_conf->bluestore_bluefs_min) { | |
4967 | dout(1) << __func__ << " main device size " << byte_u_t(dev_size) | |
4968 | << " is too small, disable bluestore_bluefs_min for now" | |
4969 | << dendl; | |
4970 | ceph_assert(dev_size >= _get_ondisk_reserved()); | |
4971 | ||
4972 | int r = cct->_conf.set_val("bluestore_bluefs_min", "0"); | |
4973 | ceph_assert(r == 0); | |
4974 | } | |
4975 | } | |
4976 | ||
7c673cae FG |
4977 | void BlueStore::_close_bdev() |
4978 | { | |
11fdf7f2 | 4979 | ceph_assert(bdev); |
7c673cae FG |
4980 | bdev->close(); |
4981 | delete bdev; | |
4982 | bdev = NULL; | |
4983 | } | |
4984 | ||
1911f103 | 4985 | int BlueStore::_open_fm(KeyValueDB::Transaction t, bool read_only) |
7c673cae | 4986 | { |
1911f103 TL |
4987 | int r; |
4988 | bluestore_bdev_label_t label; | |
4989 | ||
11fdf7f2 TL |
4990 | ceph_assert(fm == NULL); |
4991 | fm = FreelistManager::create(cct, freelist_type, PREFIX_ALLOC); | |
4992 | ceph_assert(fm); | |
4993 | if (t) { | |
4994 | // create mode. initialize freespace | |
7c673cae | 4995 | dout(20) << __func__ << " initializing freespace" << dendl; |
7c673cae FG |
4996 | { |
4997 | bufferlist bl; | |
4998 | bl.append(freelist_type); | |
4999 | t->set(PREFIX_SUPER, "freelist_type", bl); | |
5000 | } | |
b32b8144 FG |
5001 | // being able to allocate in units less than bdev block size |
5002 | // seems to be a bad idea. | |
11fdf7f2 | 5003 | ceph_assert( cct->_conf->bdev_block_size <= (int64_t)min_alloc_size); |
b32b8144 | 5004 | fm->create(bdev->get_size(), (int64_t)min_alloc_size, t); |
7c673cae FG |
5005 | |
5006 | // allocate superblock reserved space. note that we do not mark | |
5007 | // bluefs space as allocated in the freelist; we instead rely on | |
5008 | // bluefs_extents. | |
11fdf7f2 | 5009 | auto reserved = _get_ondisk_reserved(); |
3efd9988 | 5010 | fm->allocate(0, reserved, t); |
7c673cae | 5011 | |
7c673cae | 5012 | if (cct->_conf->bluestore_bluefs) { |
11fdf7f2 | 5013 | ceph_assert(bluefs_extents.num_intervals() == 1); |
7c673cae | 5014 | interval_set<uint64_t>::iterator p = bluefs_extents.begin(); |
11fdf7f2 | 5015 | reserved = round_up_to(p.get_start() + p.get_len(), min_alloc_size); |
7c673cae FG |
5016 | dout(20) << __func__ << " reserved 0x" << std::hex << reserved << std::dec |
5017 | << " for bluefs" << dendl; | |
7c673cae FG |
5018 | } |
5019 | ||
5020 | if (cct->_conf->bluestore_debug_prefill > 0) { | |
5021 | uint64_t end = bdev->get_size() - reserved; | |
5022 | dout(1) << __func__ << " pre-fragmenting freespace, using " | |
5023 | << cct->_conf->bluestore_debug_prefill << " with max free extent " | |
5024 | << cct->_conf->bluestore_debug_prefragment_max << dendl; | |
11fdf7f2 | 5025 | uint64_t start = p2roundup(reserved, min_alloc_size); |
7c673cae FG |
5026 | uint64_t max_b = cct->_conf->bluestore_debug_prefragment_max / min_alloc_size; |
5027 | float r = cct->_conf->bluestore_debug_prefill; | |
5028 | r /= 1.0 - r; | |
5029 | bool stop = false; | |
5030 | ||
5031 | while (!stop && start < end) { | |
5032 | uint64_t l = (rand() % max_b + 1) * min_alloc_size; | |
5033 | if (start + l > end) { | |
5034 | l = end - start; | |
11fdf7f2 | 5035 | l = p2align(l, min_alloc_size); |
7c673cae | 5036 | } |
11fdf7f2 | 5037 | ceph_assert(start + l <= end); |
7c673cae FG |
5038 | |
5039 | uint64_t u = 1 + (uint64_t)(r * (double)l); | |
11fdf7f2 | 5040 | u = p2roundup(u, min_alloc_size); |
7c673cae FG |
5041 | if (start + l + u > end) { |
5042 | u = end - (start + l); | |
5043 | // trim to align so we don't overflow again | |
11fdf7f2 | 5044 | u = p2align(u, min_alloc_size); |
7c673cae FG |
5045 | stop = true; |
5046 | } | |
11fdf7f2 | 5047 | ceph_assert(start + l + u <= end); |
7c673cae | 5048 | |
11fdf7f2 | 5049 | dout(20) << __func__ << " free 0x" << std::hex << start << "~" << l |
7c673cae FG |
5050 | << " use 0x" << u << std::dec << dendl; |
5051 | ||
5052 | if (u == 0) { | |
5053 | // break if u has been trimmed to nothing | |
5054 | break; | |
5055 | } | |
5056 | ||
5057 | fm->allocate(start + l, u, t); | |
5058 | start += l + u; | |
5059 | } | |
5060 | } | |
1911f103 TL |
5061 | r = _write_out_fm_meta(0, false, &label); |
5062 | ceph_assert(r == 0); | |
5063 | } else { | |
5064 | string p = path + "/block"; | |
5065 | r = _read_bdev_label(cct, p, &label); | |
5066 | if (r < 0) { | |
5067 | derr << __func__ << " freelist init failed, error reading bdev label: " << cpp_strerror(r) << dendl; | |
5068 | delete fm; | |
5069 | fm = NULL; | |
5070 | return r; | |
5071 | } | |
7c673cae | 5072 | } |
1911f103 | 5073 | r = fm->init(label, db, read_only); |
7c673cae FG |
5074 | if (r < 0) { |
5075 | derr << __func__ << " freelist init failed: " << cpp_strerror(r) << dendl; | |
5076 | delete fm; | |
5077 | fm = NULL; | |
5078 | return r; | |
5079 | } | |
81eedcae TL |
5080 | // if space size tracked by free list manager is that higher than actual |
5081 | // dev size one can hit out-of-space allocation which will result | |
5082 | // in data loss and/or assertions | |
5083 | // Probably user altered the device size somehow. | |
5084 | // The only fix for now is to redeploy OSD. | |
5085 | if (fm->get_size() >= bdev->get_size() + min_alloc_size) { | |
5086 | ostringstream ss; | |
5087 | ss << "slow device size mismatch detected, " | |
5088 | << " fm size(" << fm->get_size() | |
5089 | << ") > slow device size(" << bdev->get_size() | |
5090 | << "), Please stop using this OSD as it might cause data loss."; | |
5091 | _set_disk_size_mismatch_alert(ss.str()); | |
5092 | } | |
7c673cae FG |
5093 | return 0; |
5094 | } | |
5095 | ||
5096 | void BlueStore::_close_fm() | |
5097 | { | |
5098 | dout(10) << __func__ << dendl; | |
11fdf7f2 | 5099 | ceph_assert(fm); |
7c673cae FG |
5100 | fm->shutdown(); |
5101 | delete fm; | |
5102 | fm = NULL; | |
5103 | } | |
5104 | ||
1911f103 TL |
5105 | int BlueStore::_write_out_fm_meta(uint64_t target_size, |
5106 | bool update_root_size, | |
5107 | bluestore_bdev_label_t* res_label) | |
5108 | { | |
5109 | string p = path + "/block"; | |
5110 | ||
5111 | std::vector<std::pair<string, string>> fm_meta; | |
5112 | fm->get_meta(target_size, &fm_meta); | |
5113 | ||
5114 | bluestore_bdev_label_t label; | |
5115 | int r = _read_bdev_label(cct, p, &label); | |
5116 | if (r < 0) | |
5117 | return r; | |
5118 | ||
5119 | for (auto& m : fm_meta) { | |
5120 | label.meta[m.first] = m.second; | |
5121 | } | |
5122 | if (update_root_size) { | |
5123 | label.size = target_size; | |
5124 | } | |
5125 | r = _write_bdev_label(cct, p, label); | |
5126 | if (res_label) { | |
5127 | *res_label = label; | |
5128 | } | |
5129 | ||
5130 | return r; | |
5131 | } | |
5132 | ||
7c673cae FG |
5133 | int BlueStore::_open_alloc() |
5134 | { | |
11fdf7f2 TL |
5135 | ceph_assert(alloc == NULL); |
5136 | ceph_assert(bdev->get_size()); | |
5137 | ||
5138 | if (bluefs) { | |
5139 | bluefs_extents.clear(); | |
9f95a23c TL |
5140 | auto r = bluefs->get_block_extents(bluefs_layout.shared_bdev, |
5141 | &bluefs_extents); | |
11fdf7f2 TL |
5142 | if (r < 0) { |
5143 | lderr(cct) << __func__ << " failed to retrieve bluefs_extents: " | |
5144 | << cpp_strerror(r) << dendl; | |
5145 | ||
5146 | return r; | |
5147 | } | |
5148 | dout(10) << __func__ << " bluefs extents 0x" | |
5149 | << std::hex << bluefs_extents << std::dec | |
5150 | << dendl; | |
5151 | } | |
5152 | ||
7c673cae FG |
5153 | alloc = Allocator::create(cct, cct->_conf->bluestore_allocator, |
5154 | bdev->get_size(), | |
eafe8130 | 5155 | min_alloc_size, "block"); |
7c673cae FG |
5156 | if (!alloc) { |
5157 | lderr(cct) << __func__ << " Allocator::unknown alloc type " | |
5158 | << cct->_conf->bluestore_allocator | |
5159 | << dendl; | |
5160 | return -EINVAL; | |
5161 | } | |
5162 | ||
5163 | uint64_t num = 0, bytes = 0; | |
5164 | ||
5165 | dout(1) << __func__ << " opening allocation metadata" << dendl; | |
5166 | // initialize from freelist | |
5167 | fm->enumerate_reset(); | |
5168 | uint64_t offset, length; | |
11fdf7f2 | 5169 | while (fm->enumerate_next(db, &offset, &length)) { |
7c673cae FG |
5170 | alloc->init_add_free(offset, length); |
5171 | ++num; | |
5172 | bytes += length; | |
5173 | } | |
224ce89b | 5174 | fm->enumerate_reset(); |
7c673cae FG |
5175 | |
5176 | // also mark bluefs space as allocated | |
5177 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
5178 | alloc->init_rm_free(e.get_start(), e.get_len()); | |
5179 | } | |
7c673cae | 5180 | |
1911f103 TL |
5181 | dout(1) << __func__ << " loaded " << byte_u_t(bytes) |
5182 | << " in " << num << " extents" | |
5183 | << " available " << byte_u_t(alloc->get_free()) | |
5184 | << dendl; | |
5185 | ||
7c673cae FG |
5186 | return 0; |
5187 | } | |
5188 | ||
5189 | void BlueStore::_close_alloc() | |
5190 | { | |
11fdf7f2 TL |
5191 | ceph_assert(bdev); |
5192 | bdev->discard_drain(); | |
5193 | ||
5194 | ceph_assert(alloc); | |
7c673cae FG |
5195 | alloc->shutdown(); |
5196 | delete alloc; | |
5197 | alloc = NULL; | |
11fdf7f2 | 5198 | bluefs_extents.clear(); |
7c673cae FG |
5199 | } |
5200 | ||
5201 | int BlueStore::_open_fsid(bool create) | |
5202 | { | |
11fdf7f2 | 5203 | ceph_assert(fsid_fd < 0); |
91327a77 | 5204 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
5205 | if (create) |
5206 | flags |= O_CREAT; | |
5207 | fsid_fd = ::openat(path_fd, "fsid", flags, 0644); | |
5208 | if (fsid_fd < 0) { | |
5209 | int err = -errno; | |
5210 | derr << __func__ << " " << cpp_strerror(err) << dendl; | |
5211 | return err; | |
5212 | } | |
5213 | return 0; | |
5214 | } | |
5215 | ||
5216 | int BlueStore::_read_fsid(uuid_d *uuid) | |
5217 | { | |
5218 | char fsid_str[40]; | |
5219 | memset(fsid_str, 0, sizeof(fsid_str)); | |
5220 | int ret = safe_read(fsid_fd, fsid_str, sizeof(fsid_str)); | |
5221 | if (ret < 0) { | |
5222 | derr << __func__ << " failed: " << cpp_strerror(ret) << dendl; | |
5223 | return ret; | |
5224 | } | |
5225 | if (ret > 36) | |
5226 | fsid_str[36] = 0; | |
5227 | else | |
5228 | fsid_str[ret] = 0; | |
5229 | if (!uuid->parse(fsid_str)) { | |
5230 | derr << __func__ << " unparsable uuid " << fsid_str << dendl; | |
5231 | return -EINVAL; | |
5232 | } | |
5233 | return 0; | |
5234 | } | |
5235 | ||
5236 | int BlueStore::_write_fsid() | |
5237 | { | |
5238 | int r = ::ftruncate(fsid_fd, 0); | |
5239 | if (r < 0) { | |
5240 | r = -errno; | |
5241 | derr << __func__ << " fsid truncate failed: " << cpp_strerror(r) << dendl; | |
5242 | return r; | |
5243 | } | |
5244 | string str = stringify(fsid) + "\n"; | |
5245 | r = safe_write(fsid_fd, str.c_str(), str.length()); | |
5246 | if (r < 0) { | |
5247 | derr << __func__ << " fsid write failed: " << cpp_strerror(r) << dendl; | |
5248 | return r; | |
5249 | } | |
5250 | r = ::fsync(fsid_fd); | |
5251 | if (r < 0) { | |
5252 | r = -errno; | |
5253 | derr << __func__ << " fsid fsync failed: " << cpp_strerror(r) << dendl; | |
5254 | return r; | |
5255 | } | |
5256 | return 0; | |
5257 | } | |
5258 | ||
5259 | void BlueStore::_close_fsid() | |
5260 | { | |
5261 | VOID_TEMP_FAILURE_RETRY(::close(fsid_fd)); | |
5262 | fsid_fd = -1; | |
5263 | } | |
5264 | ||
5265 | int BlueStore::_lock_fsid() | |
5266 | { | |
5267 | struct flock l; | |
5268 | memset(&l, 0, sizeof(l)); | |
5269 | l.l_type = F_WRLCK; | |
5270 | l.l_whence = SEEK_SET; | |
5271 | int r = ::fcntl(fsid_fd, F_SETLK, &l); | |
5272 | if (r < 0) { | |
5273 | int err = errno; | |
5274 | derr << __func__ << " failed to lock " << path << "/fsid" | |
5275 | << " (is another ceph-osd still running?)" | |
5276 | << cpp_strerror(err) << dendl; | |
5277 | return -err; | |
5278 | } | |
5279 | return 0; | |
5280 | } | |
5281 | ||
31f18b77 FG |
5282 | bool BlueStore::is_rotational() |
5283 | { | |
5284 | if (bdev) { | |
5285 | return bdev->is_rotational(); | |
5286 | } | |
5287 | ||
5288 | bool rotational = true; | |
5289 | int r = _open_path(); | |
5290 | if (r < 0) | |
5291 | goto out; | |
5292 | r = _open_fsid(false); | |
5293 | if (r < 0) | |
5294 | goto out_path; | |
5295 | r = _read_fsid(&fsid); | |
5296 | if (r < 0) | |
5297 | goto out_fsid; | |
5298 | r = _lock_fsid(); | |
5299 | if (r < 0) | |
5300 | goto out_fsid; | |
5301 | r = _open_bdev(false); | |
5302 | if (r < 0) | |
5303 | goto out_fsid; | |
5304 | rotational = bdev->is_rotational(); | |
5305 | _close_bdev(); | |
5306 | out_fsid: | |
5307 | _close_fsid(); | |
5308 | out_path: | |
5309 | _close_path(); | |
5310 | out: | |
5311 | return rotational; | |
5312 | } | |
5313 | ||
d2e6a577 FG |
5314 | bool BlueStore::is_journal_rotational() |
5315 | { | |
5316 | if (!bluefs) { | |
5317 | dout(5) << __func__ << " bluefs disabled, default to store media type" | |
5318 | << dendl; | |
5319 | return is_rotational(); | |
5320 | } | |
5321 | dout(10) << __func__ << " " << (int)bluefs->wal_is_rotational() << dendl; | |
5322 | return bluefs->wal_is_rotational(); | |
5323 | } | |
5324 | ||
9f95a23c TL |
5325 | bool BlueStore::_use_rotational_settings() |
5326 | { | |
5327 | if (cct->_conf->bluestore_debug_enforce_settings == "hdd") { | |
5328 | return true; | |
5329 | } | |
5330 | if (cct->_conf->bluestore_debug_enforce_settings == "ssd") { | |
5331 | return false; | |
5332 | } | |
5333 | return bdev->is_rotational(); | |
5334 | } | |
5335 | ||
7c673cae FG |
5336 | bool BlueStore::test_mount_in_use() |
5337 | { | |
5338 | // most error conditions mean the mount is not in use (e.g., because | |
5339 | // it doesn't exist). only if we fail to lock do we conclude it is | |
5340 | // in use. | |
5341 | bool ret = false; | |
5342 | int r = _open_path(); | |
5343 | if (r < 0) | |
5344 | return false; | |
5345 | r = _open_fsid(false); | |
5346 | if (r < 0) | |
5347 | goto out_path; | |
5348 | r = _lock_fsid(); | |
5349 | if (r < 0) | |
5350 | ret = true; // if we can't lock, it is in use | |
5351 | _close_fsid(); | |
5352 | out_path: | |
5353 | _close_path(); | |
5354 | return ret; | |
5355 | } | |
5356 | ||
11fdf7f2 | 5357 | int BlueStore::_minimal_open_bluefs(bool create) |
7c673cae FG |
5358 | { |
5359 | int r; | |
11fdf7f2 | 5360 | bluefs = new BlueFS(cct); |
7c673cae | 5361 | |
11fdf7f2 TL |
5362 | string bfn; |
5363 | struct stat st; | |
5364 | ||
5365 | bfn = path + "/block.db"; | |
5366 | if (::stat(bfn.c_str(), &st) == 0) { | |
eafe8130 TL |
5367 | r = bluefs->add_block_device( |
5368 | BlueFS::BDEV_DB, bfn, | |
5369 | create && cct->_conf->bdev_enable_discard); | |
7c673cae | 5370 | if (r < 0) { |
11fdf7f2 TL |
5371 | derr << __func__ << " add block device(" << bfn << ") returned: " |
5372 | << cpp_strerror(r) << dendl; | |
5373 | goto free_bluefs; | |
7c673cae | 5374 | } |
7c673cae | 5375 | |
11fdf7f2 TL |
5376 | if (bluefs->bdev_support_label(BlueFS::BDEV_DB)) { |
5377 | r = _check_or_set_bdev_label( | |
5378 | bfn, | |
5379 | bluefs->get_block_device_size(BlueFS::BDEV_DB), | |
5380 | "bluefs db", create); | |
5381 | if (r < 0) { | |
5382 | derr << __func__ | |
5383 | << " check block device(" << bfn << ") label returned: " | |
5384 | << cpp_strerror(r) << dendl; | |
5385 | goto free_bluefs; | |
5386 | } | |
7c673cae | 5387 | } |
11fdf7f2 TL |
5388 | if (create) { |
5389 | bluefs->add_block_extent( | |
5390 | BlueFS::BDEV_DB, | |
5391 | SUPER_RESERVED, | |
5392 | bluefs->get_block_device_size(BlueFS::BDEV_DB) - SUPER_RESERVED); | |
5393 | } | |
9f95a23c TL |
5394 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
5395 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
5396 | } else { |
5397 | r = -errno; | |
5398 | if (::lstat(bfn.c_str(), &st) == -1) { | |
5399 | r = 0; | |
9f95a23c | 5400 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
7c673cae | 5401 | } else { |
11fdf7f2 TL |
5402 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " |
5403 | << cpp_strerror(r) << dendl; | |
5404 | goto free_bluefs; | |
7c673cae FG |
5405 | } |
5406 | } | |
7c673cae | 5407 | |
11fdf7f2 TL |
5408 | // shared device |
5409 | bfn = path + "/block"; | |
5410 | // never trim here | |
9f95a23c | 5411 | r = bluefs->add_block_device(bluefs_layout.shared_bdev, bfn, false, |
11fdf7f2 TL |
5412 | true /* shared with bluestore */); |
5413 | if (r < 0) { | |
5414 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5415 | << cpp_strerror(r) << dendl; | |
5416 | goto free_bluefs; | |
5417 | } | |
5418 | if (create) { | |
5419 | // note: we always leave the first SUPER_RESERVED (8k) of the device unused | |
5420 | uint64_t initial = | |
5421 | bdev->get_size() * (cct->_conf->bluestore_bluefs_min_ratio + | |
5422 | cct->_conf->bluestore_bluefs_gift_ratio); | |
5423 | initial = std::max(initial, cct->_conf->bluestore_bluefs_min); | |
eafe8130 TL |
5424 | uint64_t alloc_size = cct->_conf->bluefs_shared_alloc_size; |
5425 | if (alloc_size % min_alloc_size) { | |
5426 | derr << __func__ << " bluefs_shared_alloc_size 0x" << std::hex | |
5427 | << alloc_size << " is not a multiple of " | |
11fdf7f2 TL |
5428 | << "min_alloc_size 0x" << min_alloc_size << std::dec << dendl; |
5429 | r = -EINVAL; | |
5430 | goto free_bluefs; | |
7c673cae | 5431 | } |
11fdf7f2 | 5432 | // align to bluefs's alloc_size |
eafe8130 | 5433 | initial = p2roundup(initial, alloc_size); |
11fdf7f2 | 5434 | // put bluefs in the middle of the device in case it is an HDD |
eafe8130 | 5435 | uint64_t start = p2align((bdev->get_size() - initial) / 2, alloc_size); |
11fdf7f2 | 5436 | //avoiding superblock overwrite |
eafe8130 TL |
5437 | start = std::max(alloc_size, start); |
5438 | ceph_assert(start >=_get_ondisk_reserved()); | |
7c673cae | 5439 | |
9f95a23c | 5440 | bluefs->add_block_extent(bluefs_layout.shared_bdev, start, initial); |
11fdf7f2 TL |
5441 | bluefs_extents.insert(start, initial); |
5442 | ++out_of_sync_fm; | |
5443 | } | |
5444 | ||
5445 | bfn = path + "/block.wal"; | |
5446 | if (::stat(bfn.c_str(), &st) == 0) { | |
5447 | r = bluefs->add_block_device(BlueFS::BDEV_WAL, bfn, | |
eafe8130 | 5448 | create && cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
5449 | if (r < 0) { |
5450 | derr << __func__ << " add block device(" << bfn << ") returned: " | |
5451 | << cpp_strerror(r) << dendl; | |
5452 | goto free_bluefs; | |
5453 | } | |
7c673cae | 5454 | |
11fdf7f2 TL |
5455 | if (bluefs->bdev_support_label(BlueFS::BDEV_WAL)) { |
5456 | r = _check_or_set_bdev_label( | |
5457 | bfn, | |
5458 | bluefs->get_block_device_size(BlueFS::BDEV_WAL), | |
5459 | "bluefs wal", create); | |
7c673cae | 5460 | if (r < 0) { |
11fdf7f2 TL |
5461 | derr << __func__ << " check block device(" << bfn |
5462 | << ") label returned: " << cpp_strerror(r) << dendl; | |
7c673cae FG |
5463 | goto free_bluefs; |
5464 | } | |
7c673cae FG |
5465 | } |
5466 | ||
11fdf7f2 TL |
5467 | if (create) { |
5468 | bluefs->add_block_extent( | |
5469 | BlueFS::BDEV_WAL, BDEV_LABEL_BLOCK_SIZE, | |
5470 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) - | |
5471 | BDEV_LABEL_BLOCK_SIZE); | |
5472 | } | |
9f95a23c | 5473 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
5474 | } else { |
5475 | r = 0; | |
5476 | if (::lstat(bfn.c_str(), &st) != -1) { | |
5477 | r = -errno; | |
5478 | derr << __func__ << " " << bfn << " symlink exists but target unusable: " | |
5479 | << cpp_strerror(r) << dendl; | |
7c673cae FG |
5480 | goto free_bluefs; |
5481 | } | |
11fdf7f2 TL |
5482 | } |
5483 | return 0; | |
7c673cae | 5484 | |
11fdf7f2 TL |
5485 | free_bluefs: |
5486 | ceph_assert(bluefs); | |
5487 | delete bluefs; | |
5488 | bluefs = NULL; | |
5489 | return r; | |
5490 | } | |
7c673cae | 5491 | |
11fdf7f2 TL |
5492 | int BlueStore::_open_bluefs(bool create) |
5493 | { | |
5494 | int r = _minimal_open_bluefs(create); | |
5495 | if (r < 0) { | |
5496 | return r; | |
5497 | } | |
9f95a23c TL |
5498 | RocksDBBlueFSVolumeSelector* vselector = nullptr; |
5499 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { | |
5500 | ||
5501 | string options = cct->_conf->bluestore_rocksdb_options; | |
5502 | ||
5503 | rocksdb::Options rocks_opts; | |
5504 | int r = RocksDBStore::ParseOptionsFromStringStatic( | |
5505 | cct, | |
5506 | options, | |
5507 | rocks_opts, | |
5508 | nullptr); | |
5509 | if (r < 0) { | |
5510 | return r; | |
5511 | } | |
5512 | ||
5513 | double reserved_factor = cct->_conf->bluestore_volume_selection_reserved_factor; | |
5514 | vselector = | |
5515 | new RocksDBBlueFSVolumeSelector( | |
5516 | bluefs->get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
5517 | bluefs->get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
5518 | bluefs->get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100, | |
5519 | 1024 * 1024 * 1024, //FIXME: set expected l0 size here | |
5520 | rocks_opts.max_bytes_for_level_base, | |
5521 | rocks_opts.max_bytes_for_level_multiplier, | |
5522 | reserved_factor, | |
5523 | cct->_conf->bluestore_volume_selection_reserved, | |
5524 | cct->_conf->bluestore_volume_selection_policy != "rocksdb_original"); | |
5525 | } | |
11fdf7f2 | 5526 | if (create) { |
9f95a23c | 5527 | bluefs->mkfs(fsid, bluefs_layout); |
11fdf7f2 | 5528 | } |
9f95a23c | 5529 | bluefs->set_volume_selector(vselector); |
11fdf7f2 TL |
5530 | r = bluefs->mount(); |
5531 | if (r < 0) { | |
5532 | derr << __func__ << " failed bluefs mount: " << cpp_strerror(r) << dendl; | |
5533 | } | |
9f95a23c | 5534 | ceph_assert_always(bluefs->maybe_verify_layout(bluefs_layout) == 0); |
11fdf7f2 TL |
5535 | return r; |
5536 | } | |
5537 | ||
1911f103 | 5538 | void BlueStore::_close_bluefs(bool cold_close) |
11fdf7f2 | 5539 | { |
1911f103 | 5540 | bluefs->umount(cold_close); |
11fdf7f2 TL |
5541 | _minimal_close_bluefs(); |
5542 | } | |
5543 | ||
5544 | void BlueStore::_minimal_close_bluefs() | |
5545 | { | |
5546 | delete bluefs; | |
5547 | bluefs = NULL; | |
5548 | } | |
5549 | ||
5550 | int BlueStore::_is_bluefs(bool create, bool* ret) | |
5551 | { | |
5552 | if (create) { | |
5553 | *ret = cct->_conf->bluestore_bluefs; | |
5554 | } else { | |
5555 | string s; | |
5556 | int r = read_meta("bluefs", &s); | |
5557 | if (r < 0) { | |
5558 | derr << __func__ << " unable to read 'bluefs' meta" << dendl; | |
5559 | return -EIO; | |
5560 | } | |
5561 | if (s == "1") { | |
5562 | *ret = true; | |
5563 | } else if (s == "0") { | |
5564 | *ret = false; | |
31f18b77 | 5565 | } else { |
11fdf7f2 TL |
5566 | derr << __func__ << " bluefs = " << s << " : not 0 or 1, aborting" |
5567 | << dendl; | |
5568 | return -EIO; | |
5569 | } | |
5570 | } | |
5571 | return 0; | |
5572 | } | |
5573 | ||
5574 | /* | |
5575 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
5576 | * in the proper order | |
5577 | */ | |
5578 | int BlueStore::_open_db_and_around(bool read_only) | |
5579 | { | |
5580 | int r; | |
5581 | bool do_bluefs = false; | |
5582 | _is_bluefs(false, &do_bluefs); // ignore err code | |
5583 | if (do_bluefs) { | |
5584 | // open in read-only first to read FM list and init allocator | |
5585 | // as they might be needed for some BlueFS procedures | |
5586 | r = _open_db(false, false, true); | |
5587 | if (r < 0) | |
5588 | return r; | |
5589 | ||
5590 | r = _open_super_meta(); | |
5591 | if (r < 0) { | |
5592 | goto out_db; | |
5593 | } | |
5594 | ||
1911f103 | 5595 | r = _open_fm(nullptr, true); |
11fdf7f2 TL |
5596 | if (r < 0) |
5597 | goto out_db; | |
5598 | ||
5599 | r = _open_alloc(); | |
5600 | if (r < 0) | |
5601 | goto out_fm; | |
5602 | ||
5603 | // now open in R/W mode | |
5604 | if (!read_only) { | |
1911f103 | 5605 | _close_db(true); |
11fdf7f2 TL |
5606 | |
5607 | r = _open_db(false, false, false); | |
5608 | if (r < 0) { | |
5609 | _close_alloc(); | |
5610 | _close_fm(); | |
5611 | return r; | |
28e407b8 | 5612 | } |
1911f103 | 5613 | fm->sync(db); |
7c673cae | 5614 | } |
11fdf7f2 TL |
5615 | } else { |
5616 | r = _open_db(false, false); | |
5617 | if (r < 0) { | |
5618 | return r; | |
5619 | } | |
5620 | r = _open_super_meta(); | |
5621 | if (r < 0) { | |
5622 | goto out_db; | |
5623 | } | |
7c673cae | 5624 | |
1911f103 | 5625 | r = _open_fm(nullptr, false); |
11fdf7f2 TL |
5626 | if (r < 0) |
5627 | goto out_db; | |
5628 | ||
5629 | r = _open_alloc(); | |
5630 | if (r < 0) | |
5631 | goto out_fm; | |
5632 | } | |
5633 | return 0; | |
5634 | ||
5635 | out_fm: | |
5636 | _close_fm(); | |
5637 | out_db: | |
1911f103 | 5638 | _close_db(read_only); |
11fdf7f2 TL |
5639 | return r; |
5640 | } | |
5641 | ||
1911f103 | 5642 | void BlueStore::_close_db_and_around(bool read_only) |
11fdf7f2 TL |
5643 | { |
5644 | if (bluefs) { | |
1911f103 | 5645 | if (!read_only && out_of_sync_fm.fetch_and(0)) { |
11fdf7f2 TL |
5646 | _sync_bluefs_and_fm(); |
5647 | } | |
1911f103 TL |
5648 | _close_db(read_only); |
5649 | while(!read_only && out_of_sync_fm.fetch_and(0)) { | |
11fdf7f2 TL |
5650 | // if seen some allocations during close - repeat open_db, sync fm, close |
5651 | dout(0) << __func__ << " syncing FreelistManager" << dendl; | |
5652 | int r = _open_db(false, false, false); | |
5653 | if (r < 0) { | |
5654 | derr << __func__ | |
5655 | << " unable to open db, FreelistManager is probably out of sync" | |
5656 | << dendl; | |
5657 | break; | |
5658 | } | |
5659 | _sync_bluefs_and_fm(); | |
1911f103 | 5660 | _close_db(false); |
7c673cae | 5661 | } |
11fdf7f2 TL |
5662 | if (!_kv_only) { |
5663 | _close_alloc(); | |
5664 | _close_fm(); | |
5665 | } | |
5666 | } else { | |
5667 | _close_alloc(); | |
5668 | _close_fm(); | |
1911f103 | 5669 | _close_db(read_only); |
11fdf7f2 TL |
5670 | } |
5671 | } | |
5672 | ||
5673 | // updates legacy bluefs related recs in DB to a state valid for | |
5674 | // downgrades from nautilus. | |
5675 | void BlueStore::_sync_bluefs_and_fm() | |
5676 | { | |
5677 | if (cct->_conf->bluestore_bluefs_db_compatibility) { | |
5678 | bufferlist bl; | |
5679 | encode(bluefs_extents, bl); | |
5680 | dout(20) << __func__ << " bluefs_extents at KV is now 0x" | |
5681 | << std::hex << bluefs_extents << std::dec | |
5682 | << dendl; | |
5683 | KeyValueDB::Transaction synct = db->get_transaction(); | |
5684 | synct->set(PREFIX_SUPER, "bluefs_extents", bl); | |
5685 | synct->set(PREFIX_SUPER, "bluefs_extents_back", bl); | |
5686 | ||
5687 | // Nice thing is that we don't need to update FreelistManager here. | |
5688 | // It always has corresponding bits set to 'Free' for both Nautilus+ and | |
5689 | // pre-Nautilis releases. | |
5690 | // So once we get an extent to bluefs_extents this means it's | |
5691 | // been free in allocator and hence it's free in FM too. | |
5692 | ||
5693 | db->submit_transaction_sync(synct); | |
5694 | } | |
5695 | } | |
5696 | ||
5697 | int BlueStore::_open_db(bool create, bool to_repair_db, bool read_only) | |
5698 | { | |
5699 | int r; | |
5700 | ceph_assert(!db); | |
5701 | ceph_assert(!(create && read_only)); | |
5702 | string fn = path + "/db"; | |
5703 | string options; | |
5704 | stringstream err; | |
5705 | std::shared_ptr<Int64ArrayMergeOperator> merge_op(new Int64ArrayMergeOperator); | |
5706 | ||
5707 | string kv_backend; | |
5708 | std::vector<KeyValueDB::ColumnFamily> cfs; | |
5709 | ||
5710 | if (create) { | |
5711 | kv_backend = cct->_conf->bluestore_kvbackend; | |
5712 | } else { | |
5713 | r = read_meta("kv_backend", &kv_backend); | |
7c673cae | 5714 | if (r < 0) { |
11fdf7f2 TL |
5715 | derr << __func__ << " unable to read 'kv_backend' meta" << dendl; |
5716 | return -EIO; | |
5717 | } | |
5718 | } | |
5719 | dout(10) << __func__ << " kv_backend = " << kv_backend << dendl; | |
5720 | ||
5721 | bool do_bluefs; | |
5722 | r = _is_bluefs(create, &do_bluefs); | |
5723 | if (r < 0) { | |
5724 | return r; | |
5725 | } | |
5726 | dout(10) << __func__ << " do_bluefs = " << do_bluefs << dendl; | |
5727 | ||
5728 | map<string,string> kv_options; | |
5729 | // force separate wal dir for all new deployments. | |
5730 | kv_options["separate_wal_dir"] = 1; | |
5731 | rocksdb::Env *env = NULL; | |
5732 | if (do_bluefs) { | |
5733 | dout(10) << __func__ << " initializing bluefs" << dendl; | |
5734 | if (kv_backend != "rocksdb") { | |
5735 | derr << " backend must be rocksdb to use bluefs" << dendl; | |
5736 | return -EINVAL; | |
7c673cae | 5737 | } |
11fdf7f2 TL |
5738 | |
5739 | r = _open_bluefs(create); | |
5740 | if (r < 0) { | |
5741 | return r; | |
5742 | } | |
11fdf7f2 | 5743 | |
7c673cae | 5744 | if (cct->_conf->bluestore_bluefs_env_mirror) { |
9f95a23c TL |
5745 | rocksdb::Env* a = new BlueRocksEnv(bluefs); |
5746 | rocksdb::Env* b = rocksdb::Env::Default(); | |
7c673cae | 5747 | if (create) { |
9f95a23c TL |
5748 | string cmd = "rm -rf " + path + "/db " + |
5749 | path + "/db.slow " + | |
5750 | path + "/db.wal"; | |
5751 | int r = system(cmd.c_str()); | |
5752 | (void)r; | |
7c673cae FG |
5753 | } |
5754 | env = new rocksdb::EnvMirror(b, a, false, true); | |
1911f103 | 5755 | } else { |
7c673cae FG |
5756 | env = new BlueRocksEnv(bluefs); |
5757 | ||
5758 | // simplify the dir names, too, as "seen" by rocksdb | |
5759 | fn = "db"; | |
5760 | } | |
9f95a23c TL |
5761 | bluefs->set_slow_device_expander(this); |
5762 | BlueFSVolumeSelector::paths paths; | |
5763 | bluefs->get_vselector_paths(fn, paths); | |
7c673cae | 5764 | |
9f95a23c | 5765 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
7c673cae FG |
5766 | // we have both block.db and block; tell rocksdb! |
5767 | // note: the second (last) size value doesn't really matter | |
5768 | ostringstream db_paths; | |
9f95a23c TL |
5769 | bool first = true; |
5770 | for (auto& p : paths) { | |
5771 | if (!first) { | |
5772 | db_paths << " "; | |
5773 | } | |
5774 | first = false; | |
5775 | db_paths << p.first << "," << p.second; | |
5776 | ||
5777 | } | |
11fdf7f2 | 5778 | kv_options["db_paths"] = db_paths.str(); |
9f95a23c | 5779 | dout(1) << __func__ << " set db_paths to " << db_paths.str() << dendl; |
7c673cae FG |
5780 | } |
5781 | ||
5782 | if (create) { | |
9f95a23c TL |
5783 | for (auto& p : paths) { |
5784 | env->CreateDir(p.first); | |
5785 | } | |
5786 | // Selectors don't provide wal path so far hence create explicitly | |
11fdf7f2 | 5787 | env->CreateDir(fn + ".wal"); |
11fdf7f2 TL |
5788 | } else { |
5789 | std::vector<std::string> res; | |
5790 | // check for dir presence | |
5791 | auto r = env->GetChildren(fn+".wal", &res); | |
5792 | if (r.IsNotFound()) { | |
5793 | kv_options.erase("separate_wal_dir"); | |
5794 | } | |
7c673cae | 5795 | } |
11fdf7f2 TL |
5796 | } else { |
5797 | string walfn = path + "/db.wal"; | |
7c673cae | 5798 | |
11fdf7f2 TL |
5799 | if (create) { |
5800 | int r = ::mkdir(fn.c_str(), 0755); | |
5801 | if (r < 0) | |
5802 | r = -errno; | |
5803 | if (r < 0 && r != -EEXIST) { | |
5804 | derr << __func__ << " failed to create " << fn << ": " << cpp_strerror(r) | |
5805 | << dendl; | |
5806 | return r; | |
5807 | } | |
5808 | ||
5809 | // wal_dir, too! | |
7c673cae FG |
5810 | r = ::mkdir(walfn.c_str(), 0755); |
5811 | if (r < 0) | |
5812 | r = -errno; | |
5813 | if (r < 0 && r != -EEXIST) { | |
5814 | derr << __func__ << " failed to create " << walfn | |
5815 | << ": " << cpp_strerror(r) | |
5816 | << dendl; | |
5817 | return r; | |
5818 | } | |
11fdf7f2 TL |
5819 | } else { |
5820 | struct stat st; | |
5821 | r = ::stat(walfn.c_str(), &st); | |
5822 | if (r < 0 && errno == ENOENT) { | |
5823 | kv_options.erase("separate_wal_dir"); | |
5824 | } | |
7c673cae FG |
5825 | } |
5826 | } | |
5827 | ||
91327a77 | 5828 | |
7c673cae FG |
5829 | db = KeyValueDB::create(cct, |
5830 | kv_backend, | |
5831 | fn, | |
11fdf7f2 | 5832 | kv_options, |
7c673cae FG |
5833 | static_cast<void*>(env)); |
5834 | if (!db) { | |
5835 | derr << __func__ << " error creating db" << dendl; | |
5836 | if (bluefs) { | |
1911f103 | 5837 | _close_bluefs(read_only); |
7c673cae FG |
5838 | } |
5839 | // delete env manually here since we can't depend on db to do this | |
5840 | // under this case | |
5841 | delete env; | |
5842 | env = NULL; | |
5843 | return -EIO; | |
5844 | } | |
5845 | ||
5846 | FreelistManager::setup_merge_operators(db); | |
5847 | db->set_merge_operator(PREFIX_STAT, merge_op); | |
91327a77 | 5848 | db->set_cache_size(cache_kv_ratio * cache_size); |
31f18b77 | 5849 | |
11fdf7f2 | 5850 | if (kv_backend == "rocksdb") { |
7c673cae | 5851 | options = cct->_conf->bluestore_rocksdb_options; |
11fdf7f2 TL |
5852 | |
5853 | map<string,string> cf_map; | |
5854 | cct->_conf.with_val<string>("bluestore_rocksdb_cfs", | |
5855 | get_str_map, | |
5856 | &cf_map, | |
5857 | " \t"); | |
5858 | for (auto& i : cf_map) { | |
5859 | dout(10) << "column family " << i.first << ": " << i.second << dendl; | |
5860 | cfs.push_back(KeyValueDB::ColumnFamily(i.first, i.second)); | |
5861 | } | |
5862 | } | |
5863 | ||
7c673cae | 5864 | db->init(options); |
11fdf7f2 TL |
5865 | if (to_repair_db) |
5866 | return 0; | |
5867 | if (create) { | |
5868 | if (cct->_conf.get_val<bool>("bluestore_rocksdb_cf")) { | |
5869 | r = db->create_and_open(err, cfs); | |
5870 | } else { | |
5871 | r = db->create_and_open(err); | |
5872 | } | |
5873 | } else { | |
5874 | // we pass in cf list here, but it is only used if the db already has | |
5875 | // column families created. | |
5876 | r = read_only ? | |
5877 | db->open_read_only(err, cfs) : | |
5878 | db->open(err, cfs); | |
5879 | } | |
7c673cae FG |
5880 | if (r) { |
5881 | derr << __func__ << " erroring opening db: " << err.str() << dendl; | |
1911f103 | 5882 | _close_db(read_only); |
7c673cae FG |
5883 | return -EIO; |
5884 | } | |
5885 | dout(1) << __func__ << " opened " << kv_backend | |
5886 | << " path " << fn << " options " << options << dendl; | |
5887 | return 0; | |
7c673cae FG |
5888 | } |
5889 | ||
1911f103 | 5890 | void BlueStore::_close_db(bool cold_close) |
7c673cae | 5891 | { |
11fdf7f2 | 5892 | ceph_assert(db); |
7c673cae FG |
5893 | delete db; |
5894 | db = NULL; | |
5895 | if (bluefs) { | |
1911f103 | 5896 | _close_bluefs(cold_close); |
7c673cae FG |
5897 | } |
5898 | } | |
5899 | ||
11fdf7f2 | 5900 | void BlueStore::_dump_alloc_on_failure() |
7c673cae | 5901 | { |
11fdf7f2 TL |
5902 | auto dump_interval = |
5903 | cct->_conf->bluestore_bluefs_alloc_failure_dump_interval; | |
5904 | if (dump_interval > 0 && | |
5905 | next_dump_on_bluefs_alloc_failure <= ceph_clock_now()) { | |
5906 | alloc->dump(); | |
5907 | next_dump_on_bluefs_alloc_failure = ceph_clock_now(); | |
5908 | next_dump_on_bluefs_alloc_failure += dump_interval; | |
7c673cae | 5909 | } |
11fdf7f2 | 5910 | } |
7c673cae | 5911 | |
7c673cae | 5912 | |
11fdf7f2 TL |
5913 | int BlueStore::allocate_bluefs_freespace( |
5914 | uint64_t min_size, | |
5915 | uint64_t size, | |
5916 | PExtentVector* extents_out) | |
5917 | { | |
5918 | ceph_assert(min_size <= size); | |
5919 | if (size) { | |
5920 | // round up to alloc size | |
9f95a23c | 5921 | uint64_t alloc_size = bluefs->get_alloc_size(bluefs_layout.shared_bdev); |
eafe8130 TL |
5922 | min_size = p2roundup(min_size, alloc_size); |
5923 | size = p2roundup(size, alloc_size); | |
11fdf7f2 TL |
5924 | |
5925 | PExtentVector extents_local; | |
5926 | PExtentVector* extents = extents_out ? extents_out : &extents_local; | |
5927 | ||
5928 | ||
5929 | uint64_t gift; | |
5930 | uint64_t allocated = 0; | |
5931 | int64_t alloc_len; | |
9f95a23c TL |
5932 | auto need = size; |
5933 | auto extent_count0 = extents->size(); | |
11fdf7f2 TL |
5934 | do { |
5935 | // hard cap to fit into 32 bits | |
9f95a23c | 5936 | gift = std::min<uint64_t>(size, 1ull << 30); |
11fdf7f2 TL |
5937 | dout(10) << __func__ << " gifting " << gift |
5938 | << " (" << byte_u_t(gift) << ")" << dendl; | |
5939 | ||
eafe8130 TL |
5940 | alloc_len = alloc->allocate(gift, alloc_size, 0, 0, extents); |
5941 | if (alloc_len > 0) { | |
11fdf7f2 TL |
5942 | allocated += alloc_len; |
5943 | size -= alloc_len; | |
5944 | } | |
5945 | ||
eafe8130 TL |
5946 | if (alloc_len < 0 || |
5947 | (alloc_len < (int64_t)gift && (min_size > allocated))) { | |
11fdf7f2 TL |
5948 | derr << __func__ |
5949 | << " failed to allocate on 0x" << std::hex << gift | |
5950 | << " min_size 0x" << min_size | |
5951 | << " > allocated total 0x" << allocated | |
eafe8130 TL |
5952 | << " bluefs_shared_alloc_size 0x" << alloc_size |
5953 | << " allocated 0x" << (alloc_len < 0 ? 0 : alloc_len) | |
11fdf7f2 TL |
5954 | << " available 0x " << alloc->get_free() |
5955 | << std::dec << dendl; | |
7c673cae | 5956 | |
494da23a | 5957 | _dump_alloc_on_failure(); |
11fdf7f2 TL |
5958 | alloc->release(*extents); |
5959 | extents->clear(); | |
5960 | return -ENOSPC; | |
5961 | } | |
5962 | } while (size && alloc_len > 0); | |
9f95a23c TL |
5963 | _collect_allocation_stats(need, alloc_size, extents->size() - extent_count0); |
5964 | ||
11fdf7f2 TL |
5965 | for (auto& e : *extents) { |
5966 | dout(5) << __func__ << " gifting " << e << " to bluefs" << dendl; | |
5967 | bluefs_extents.insert(e.offset, e.length); | |
5968 | ++out_of_sync_fm; | |
5969 | // apply to bluefs if not requested from outside | |
5970 | if (!extents_out) { | |
9f95a23c | 5971 | bluefs->add_block_extent(bluefs_layout.shared_bdev, e.offset, e.length); |
11fdf7f2 | 5972 | } |
7c673cae FG |
5973 | } |
5974 | } | |
7c673cae FG |
5975 | return 0; |
5976 | } | |
5977 | ||
9f95a23c TL |
5978 | uint64_t BlueStore::available_freespace(uint64_t alloc_size) { |
5979 | uint64_t total = 0; | |
5980 | auto iterated_allocation = [&](uint64_t off, uint64_t len) { | |
eafe8130 | 5981 | //only count in size that is alloc_size aligned |
9f95a23c TL |
5982 | uint64_t dist_to_alignment; |
5983 | uint64_t offset_in_block = off & (alloc_size - 1); | |
eafe8130 TL |
5984 | if (offset_in_block == 0) |
5985 | dist_to_alignment = 0; | |
5986 | else | |
5987 | dist_to_alignment = alloc_size - offset_in_block; | |
5988 | if (dist_to_alignment >= len) | |
5989 | return; | |
5990 | len -= dist_to_alignment; | |
5991 | total += p2align(len, alloc_size); | |
5992 | }; | |
5993 | alloc->dump(iterated_allocation); | |
5994 | return total; | |
5995 | } | |
5996 | ||
11fdf7f2 | 5997 | int64_t BlueStore::_get_bluefs_size_delta(uint64_t bluefs_free, uint64_t bluefs_total) |
f64942e4 | 5998 | { |
7c673cae FG |
5999 | float bluefs_free_ratio = (float)bluefs_free / (float)bluefs_total; |
6000 | ||
6001 | uint64_t my_free = alloc->get_free(); | |
6002 | uint64_t total = bdev->get_size(); | |
6003 | float my_free_ratio = (float)my_free / (float)total; | |
6004 | ||
6005 | uint64_t total_free = bluefs_free + my_free; | |
6006 | ||
6007 | float bluefs_ratio = (float)bluefs_free / (float)total_free; | |
6008 | ||
6009 | dout(10) << __func__ | |
1adf2230 | 6010 | << " bluefs " << byte_u_t(bluefs_free) |
7c673cae | 6011 | << " free (" << bluefs_free_ratio |
1adf2230 | 6012 | << ") bluestore " << byte_u_t(my_free) |
7c673cae FG |
6013 | << " free (" << my_free_ratio |
6014 | << "), bluefs_ratio " << bluefs_ratio | |
6015 | << dendl; | |
6016 | ||
6017 | uint64_t gift = 0; | |
6018 | uint64_t reclaim = 0; | |
6019 | if (bluefs_ratio < cct->_conf->bluestore_bluefs_min_ratio) { | |
6020 | gift = cct->_conf->bluestore_bluefs_gift_ratio * total_free; | |
9f95a23c TL |
6021 | if (gift >= my_free) |
6022 | gift = my_free / 2; | |
7c673cae FG |
6023 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio |
6024 | << " < min_ratio " << cct->_conf->bluestore_bluefs_min_ratio | |
1adf2230 | 6025 | << ", should gift " << byte_u_t(gift) << dendl; |
7c673cae FG |
6026 | } else if (bluefs_ratio > cct->_conf->bluestore_bluefs_max_ratio) { |
6027 | reclaim = cct->_conf->bluestore_bluefs_reclaim_ratio * total_free; | |
6028 | if (bluefs_total - reclaim < cct->_conf->bluestore_bluefs_min) | |
6029 | reclaim = bluefs_total - cct->_conf->bluestore_bluefs_min; | |
9f95a23c TL |
6030 | if (reclaim >= bluefs_free) |
6031 | reclaim = bluefs_free / 2; | |
7c673cae FG |
6032 | dout(10) << __func__ << " bluefs_ratio " << bluefs_ratio |
6033 | << " > max_ratio " << cct->_conf->bluestore_bluefs_max_ratio | |
1adf2230 | 6034 | << ", should reclaim " << byte_u_t(reclaim) << dendl; |
7c673cae | 6035 | } |
3efd9988 FG |
6036 | |
6037 | // don't take over too much of the freespace | |
6038 | uint64_t free_cap = cct->_conf->bluestore_bluefs_max_ratio * total_free; | |
7c673cae | 6039 | if (bluefs_total < cct->_conf->bluestore_bluefs_min && |
3efd9988 | 6040 | cct->_conf->bluestore_bluefs_min < free_cap) { |
7c673cae FG |
6041 | uint64_t g = cct->_conf->bluestore_bluefs_min - bluefs_total; |
6042 | dout(10) << __func__ << " bluefs_total " << bluefs_total | |
6043 | << " < min " << cct->_conf->bluestore_bluefs_min | |
1adf2230 | 6044 | << ", should gift " << byte_u_t(g) << dendl; |
7c673cae FG |
6045 | if (g > gift) |
6046 | gift = g; | |
6047 | reclaim = 0; | |
6048 | } | |
9f95a23c TL |
6049 | uint64_t min_free = |
6050 | cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
3efd9988 FG |
6051 | if (bluefs_free < min_free && |
6052 | min_free < free_cap) { | |
6053 | uint64_t g = min_free - bluefs_free; | |
11fdf7f2 | 6054 | dout(10) << __func__ << " bluefs_free " << bluefs_free |
3efd9988 | 6055 | << " < min " << min_free |
1adf2230 | 6056 | << ", should gift " << byte_u_t(g) << dendl; |
3efd9988 FG |
6057 | if (g > gift) |
6058 | gift = g; | |
6059 | reclaim = 0; | |
6060 | } | |
9f95a23c TL |
6061 | uint64_t max_free = |
6062 | cct->_conf.get_val<Option::size_t>("bluestore_bluefs_max_free"); | |
6063 | if (bluefs_free > max_free) { | |
6064 | dout(10) << __func__ << " bluefs_free " << bluefs_free | |
6065 | << " > max " << max_free | |
6066 | << ", stop gifting for now" << dendl; | |
6067 | gift = 0; | |
6068 | } | |
11fdf7f2 TL |
6069 | ceph_assert((int64_t)gift >= 0); |
6070 | ceph_assert((int64_t)reclaim >= 0); | |
6071 | return gift > 0 ? (int64_t)gift : -(int64_t)reclaim; | |
6072 | } | |
7c673cae | 6073 | |
11fdf7f2 TL |
6074 | int BlueStore::_balance_bluefs_freespace() |
6075 | { | |
6076 | int ret = 0; | |
6077 | ceph_assert(bluefs); | |
7c673cae | 6078 | |
11fdf7f2 TL |
6079 | vector<pair<uint64_t,uint64_t>> bluefs_usage; // <free, total> ... |
6080 | bluefs->get_usage(&bluefs_usage); | |
9f95a23c | 6081 | ceph_assert(bluefs_usage.size() > bluefs_layout.shared_bdev); |
7c673cae | 6082 | |
11fdf7f2 | 6083 | bool clear_alert = true; |
9f95a23c TL |
6084 | if (bluefs_layout.shared_bdev == BlueFS::BDEV_SLOW) { |
6085 | auto& p = bluefs_usage[bluefs_layout.shared_bdev]; | |
11fdf7f2 TL |
6086 | if (p.first != p.second) { |
6087 | auto& db = bluefs_usage[BlueFS::BDEV_DB]; | |
6088 | ostringstream ss; | |
6089 | ss << "spilled over " << byte_u_t(p.second - p.first) | |
6090 | << " metadata from 'db' device (" << byte_u_t(db.second - db.first) | |
6091 | << " used of " << byte_u_t(db.second) << ") to slow device"; | |
6092 | _set_spillover_alert(ss.str()); | |
6093 | clear_alert = false; | |
6094 | } | |
6095 | } | |
6096 | if (clear_alert) { | |
6097 | _clear_spillover_alert(); | |
7c673cae FG |
6098 | } |
6099 | ||
11fdf7f2 TL |
6100 | // fixme: look at primary bdev only for now |
6101 | int64_t delta = _get_bluefs_size_delta( | |
9f95a23c TL |
6102 | bluefs_usage[bluefs_layout.shared_bdev].first, |
6103 | bluefs_usage[bluefs_layout.shared_bdev].second); | |
11fdf7f2 | 6104 | |
7c673cae | 6105 | // reclaim from bluefs? |
11fdf7f2 | 6106 | if (delta < 0) { |
7c673cae | 6107 | // round up to alloc size |
9f95a23c | 6108 | uint64_t alloc_size = bluefs->get_alloc_size(bluefs_layout.shared_bdev); |
eafe8130 | 6109 | auto reclaim = p2roundup(uint64_t(-delta), alloc_size); |
7c673cae FG |
6110 | |
6111 | // hard cap to fit into 32 bits | |
9f95a23c | 6112 | reclaim = std::min<uint64_t>(reclaim, 1ull << 30); |
7c673cae | 6113 | dout(10) << __func__ << " reclaiming " << reclaim |
1adf2230 | 6114 | << " (" << byte_u_t(reclaim) << ")" << dendl; |
7c673cae FG |
6115 | |
6116 | while (reclaim > 0) { | |
6117 | // NOTE: this will block and do IO. | |
a8e16298 | 6118 | PExtentVector extents; |
9f95a23c | 6119 | int r = bluefs->reclaim_blocks(bluefs_layout.shared_bdev, reclaim, |
7c673cae FG |
6120 | &extents); |
6121 | if (r < 0) { | |
6122 | derr << __func__ << " failed to reclaim space from bluefs" | |
6123 | << dendl; | |
6124 | break; | |
6125 | } | |
6126 | for (auto e : extents) { | |
11fdf7f2 | 6127 | ++out_of_sync_fm; |
7c673cae FG |
6128 | bluefs_extents.erase(e.offset, e.length); |
6129 | bluefs_extents_reclaiming.insert(e.offset, e.length); | |
6130 | reclaim -= e.length; | |
6131 | } | |
6132 | } | |
6133 | ||
6134 | ret = 1; | |
6135 | } | |
6136 | ||
6137 | return ret; | |
6138 | } | |
6139 | ||
eafe8130 | 6140 | int BlueStore::_open_collections() |
7c673cae | 6141 | { |
28e407b8 | 6142 | dout(10) << __func__ << dendl; |
eafe8130 | 6143 | collections_had_errors = false; |
11fdf7f2 | 6144 | ceph_assert(coll_map.empty()); |
7c673cae FG |
6145 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); |
6146 | for (it->upper_bound(string()); | |
6147 | it->valid(); | |
6148 | it->next()) { | |
6149 | coll_t cid; | |
6150 | if (cid.parse(it->key())) { | |
9f95a23c | 6151 | auto c = ceph::make_ref<Collection>( |
7c673cae | 6152 | this, |
9f95a23c TL |
6153 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
6154 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
6155 | cid); | |
7c673cae | 6156 | bufferlist bl = it->value(); |
11fdf7f2 | 6157 | auto p = bl.cbegin(); |
7c673cae | 6158 | try { |
11fdf7f2 | 6159 | decode(c->cnode, p); |
7c673cae FG |
6160 | } catch (buffer::error& e) { |
6161 | derr << __func__ << " failed to decode cnode, key:" | |
6162 | << pretty_binary_string(it->key()) << dendl; | |
6163 | return -EIO; | |
6164 | } | |
28e407b8 AA |
6165 | dout(20) << __func__ << " opened " << cid << " " << c |
6166 | << " " << c->cnode << dendl; | |
11fdf7f2 | 6167 | _osr_attach(c.get()); |
7c673cae | 6168 | coll_map[cid] = c; |
11fdf7f2 | 6169 | |
7c673cae FG |
6170 | } else { |
6171 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
eafe8130 | 6172 | collections_had_errors = true; |
7c673cae FG |
6173 | } |
6174 | } | |
6175 | return 0; | |
6176 | } | |
6177 | ||
eafe8130 TL |
6178 | void BlueStore::_fsck_collections(int64_t* errors) |
6179 | { | |
6180 | if (collections_had_errors) { | |
6181 | dout(10) << __func__ << dendl; | |
6182 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_COLL); | |
6183 | for (it->upper_bound(string()); | |
6184 | it->valid(); | |
6185 | it->next()) { | |
6186 | coll_t cid; | |
6187 | if (!cid.parse(it->key())) { | |
6188 | derr << __func__ << " unrecognized collection " << it->key() << dendl; | |
6189 | if (errors) { | |
6190 | (*errors)++; | |
6191 | } | |
6192 | } | |
6193 | } | |
6194 | } | |
6195 | } | |
6196 | ||
9f95a23c TL |
6197 | void BlueStore::_set_per_pool_omap() |
6198 | { | |
6199 | per_pool_omap = false; | |
6200 | bufferlist bl; | |
6201 | db->get(PREFIX_SUPER, "per_pool_omap", &bl); | |
6202 | if (bl.length()) { | |
6203 | per_pool_omap = true; | |
6204 | dout(10) << __func__ << " per_pool_omap=1" << dendl; | |
6205 | } else { | |
6206 | dout(10) << __func__ << " per_pool_omap not present" << dendl; | |
6207 | } | |
6208 | _check_no_per_pool_omap_alert(); | |
6209 | } | |
6210 | ||
224ce89b | 6211 | void BlueStore::_open_statfs() |
31f18b77 | 6212 | { |
11fdf7f2 TL |
6213 | osd_pools.clear(); |
6214 | vstatfs.reset(); | |
6215 | ||
31f18b77 | 6216 | bufferlist bl; |
11fdf7f2 | 6217 | int r = db->get(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, &bl); |
31f18b77 | 6218 | if (r >= 0) { |
11fdf7f2 | 6219 | per_pool_stat_collection = false; |
31f18b77 | 6220 | if (size_t(bl.length()) >= sizeof(vstatfs.values)) { |
11fdf7f2 | 6221 | auto it = bl.cbegin(); |
31f18b77 | 6222 | vstatfs.decode(it); |
11fdf7f2 | 6223 | dout(10) << __func__ << " store_statfs is found" << dendl; |
224ce89b | 6224 | } else { |
31f18b77 FG |
6225 | dout(10) << __func__ << " store_statfs is corrupt, using empty" << dendl; |
6226 | } | |
81eedcae | 6227 | _check_legacy_statfs_alert(); |
11fdf7f2 TL |
6228 | } else { |
6229 | per_pool_stat_collection = true; | |
6230 | dout(10) << __func__ << " per-pool statfs is enabled" << dendl; | |
6231 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_STAT); | |
6232 | for (it->upper_bound(string()); | |
6233 | it->valid(); | |
6234 | it->next()) { | |
6235 | ||
6236 | uint64_t pool_id; | |
6237 | int r = get_key_pool_stat(it->key(), &pool_id); | |
6238 | ceph_assert(r == 0); | |
6239 | ||
6240 | bufferlist bl; | |
6241 | bl = it->value(); | |
6242 | auto p = bl.cbegin(); | |
6243 | auto& st = osd_pools[pool_id]; | |
6244 | try { | |
6245 | st.decode(p); | |
6246 | vstatfs += st; | |
6247 | ||
6248 | dout(30) << __func__ << " pool " << pool_id | |
6249 | << " statfs " << st << dendl; | |
6250 | } catch (buffer::error& e) { | |
6251 | derr << __func__ << " failed to decode pool stats, key:" | |
6252 | << pretty_binary_string(it->key()) << dendl; | |
6253 | } | |
6254 | } | |
31f18b77 | 6255 | } |
11fdf7f2 TL |
6256 | dout(30) << __func__ << " statfs " << vstatfs << dendl; |
6257 | ||
31f18b77 FG |
6258 | } |
6259 | ||
7c673cae FG |
6260 | int BlueStore::_setup_block_symlink_or_file( |
6261 | string name, | |
6262 | string epath, | |
6263 | uint64_t size, | |
6264 | bool create) | |
6265 | { | |
6266 | dout(20) << __func__ << " name " << name << " path " << epath | |
6267 | << " size " << size << " create=" << (int)create << dendl; | |
6268 | int r = 0; | |
91327a77 | 6269 | int flags = O_RDWR|O_CLOEXEC; |
7c673cae FG |
6270 | if (create) |
6271 | flags |= O_CREAT; | |
6272 | if (epath.length()) { | |
6273 | r = ::symlinkat(epath.c_str(), path_fd, name.c_str()); | |
6274 | if (r < 0) { | |
6275 | r = -errno; | |
6276 | derr << __func__ << " failed to create " << name << " symlink to " | |
6277 | << epath << ": " << cpp_strerror(r) << dendl; | |
6278 | return r; | |
6279 | } | |
6280 | ||
6281 | if (!epath.compare(0, strlen(SPDK_PREFIX), SPDK_PREFIX)) { | |
6282 | int fd = ::openat(path_fd, epath.c_str(), flags, 0644); | |
6283 | if (fd < 0) { | |
6284 | r = -errno; | |
6285 | derr << __func__ << " failed to open " << epath << " file: " | |
6286 | << cpp_strerror(r) << dendl; | |
6287 | return r; | |
6288 | } | |
11fdf7f2 TL |
6289 | // write the Transport ID of the NVMe device |
6290 | // a transport id looks like: "trtype:PCIe traddr:0000:02:00.0" | |
6291 | // where "0000:02:00.0" is the selector of a PCI device, see | |
6292 | // the first column of "lspci -mm -n -D" | |
6293 | string trid{"trtype:PCIe "}; | |
6294 | trid += "traddr:"; | |
6295 | trid += epath.substr(strlen(SPDK_PREFIX)); | |
6296 | r = ::write(fd, trid.c_str(), trid.size()); | |
6297 | ceph_assert(r == static_cast<int>(trid.size())); | |
7c673cae FG |
6298 | dout(1) << __func__ << " created " << name << " symlink to " |
6299 | << epath << dendl; | |
6300 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6301 | } | |
6302 | } | |
6303 | if (size) { | |
6304 | int fd = ::openat(path_fd, name.c_str(), flags, 0644); | |
6305 | if (fd >= 0) { | |
6306 | // block file is present | |
6307 | struct stat st; | |
6308 | int r = ::fstat(fd, &st); | |
6309 | if (r == 0 && | |
6310 | S_ISREG(st.st_mode) && // if it is a regular file | |
6311 | st.st_size == 0) { // and is 0 bytes | |
6312 | r = ::ftruncate(fd, size); | |
6313 | if (r < 0) { | |
6314 | r = -errno; | |
6315 | derr << __func__ << " failed to resize " << name << " file to " | |
6316 | << size << ": " << cpp_strerror(r) << dendl; | |
6317 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6318 | return r; | |
6319 | } | |
6320 | ||
6321 | if (cct->_conf->bluestore_block_preallocate_file) { | |
28e407b8 AA |
6322 | r = ::ceph_posix_fallocate(fd, 0, size); |
6323 | if (r > 0) { | |
7c673cae FG |
6324 | derr << __func__ << " failed to prefallocate " << name << " file to " |
6325 | << size << ": " << cpp_strerror(r) << dendl; | |
6326 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6327 | return -r; | |
6328 | } | |
7c673cae FG |
6329 | } |
6330 | dout(1) << __func__ << " resized " << name << " file to " | |
1adf2230 | 6331 | << byte_u_t(size) << dendl; |
7c673cae FG |
6332 | } |
6333 | VOID_TEMP_FAILURE_RETRY(::close(fd)); | |
6334 | } else { | |
6335 | int r = -errno; | |
6336 | if (r != -ENOENT) { | |
6337 | derr << __func__ << " failed to open " << name << " file: " | |
6338 | << cpp_strerror(r) << dendl; | |
6339 | return r; | |
6340 | } | |
6341 | } | |
6342 | } | |
6343 | return 0; | |
6344 | } | |
6345 | ||
6346 | int BlueStore::mkfs() | |
6347 | { | |
6348 | dout(1) << __func__ << " path " << path << dendl; | |
6349 | int r; | |
6350 | uuid_d old_fsid; | |
6351 | ||
eafe8130 TL |
6352 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
6353 | derr << __func__ << " osd_max_object_size " | |
6354 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
6355 | << OBJECT_MAX_SIZE << dendl; | |
6356 | return -EINVAL; | |
6357 | } | |
6358 | ||
7c673cae FG |
6359 | { |
6360 | string done; | |
6361 | r = read_meta("mkfs_done", &done); | |
6362 | if (r == 0) { | |
6363 | dout(1) << __func__ << " already created" << dendl; | |
6364 | if (cct->_conf->bluestore_fsck_on_mkfs) { | |
6365 | r = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6366 | if (r < 0) { | |
6367 | derr << __func__ << " fsck found fatal error: " << cpp_strerror(r) | |
6368 | << dendl; | |
6369 | return r; | |
6370 | } | |
6371 | if (r > 0) { | |
6372 | derr << __func__ << " fsck found " << r << " errors" << dendl; | |
6373 | r = -EIO; | |
6374 | } | |
6375 | } | |
6376 | return r; // idempotent | |
6377 | } | |
6378 | } | |
6379 | ||
6380 | { | |
6381 | string type; | |
6382 | r = read_meta("type", &type); | |
6383 | if (r == 0) { | |
6384 | if (type != "bluestore") { | |
6385 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6386 | return -EIO; | |
6387 | } | |
6388 | } else { | |
6389 | r = write_meta("type", "bluestore"); | |
6390 | if (r < 0) | |
6391 | return r; | |
6392 | } | |
6393 | } | |
6394 | ||
6395 | freelist_type = "bitmap"; | |
6396 | ||
6397 | r = _open_path(); | |
6398 | if (r < 0) | |
6399 | return r; | |
6400 | ||
6401 | r = _open_fsid(true); | |
6402 | if (r < 0) | |
6403 | goto out_path_fd; | |
6404 | ||
6405 | r = _lock_fsid(); | |
6406 | if (r < 0) | |
6407 | goto out_close_fsid; | |
6408 | ||
6409 | r = _read_fsid(&old_fsid); | |
6410 | if (r < 0 || old_fsid.is_zero()) { | |
6411 | if (fsid.is_zero()) { | |
6412 | fsid.generate_random(); | |
6413 | dout(1) << __func__ << " generated fsid " << fsid << dendl; | |
6414 | } else { | |
6415 | dout(1) << __func__ << " using provided fsid " << fsid << dendl; | |
6416 | } | |
6417 | // we'll write it later. | |
6418 | } else { | |
6419 | if (!fsid.is_zero() && fsid != old_fsid) { | |
6420 | derr << __func__ << " on-disk fsid " << old_fsid | |
6421 | << " != provided " << fsid << dendl; | |
6422 | r = -EINVAL; | |
6423 | goto out_close_fsid; | |
6424 | } | |
6425 | fsid = old_fsid; | |
6426 | } | |
6427 | ||
6428 | r = _setup_block_symlink_or_file("block", cct->_conf->bluestore_block_path, | |
6429 | cct->_conf->bluestore_block_size, | |
6430 | cct->_conf->bluestore_block_create); | |
6431 | if (r < 0) | |
6432 | goto out_close_fsid; | |
6433 | if (cct->_conf->bluestore_bluefs) { | |
6434 | r = _setup_block_symlink_or_file("block.wal", cct->_conf->bluestore_block_wal_path, | |
6435 | cct->_conf->bluestore_block_wal_size, | |
6436 | cct->_conf->bluestore_block_wal_create); | |
6437 | if (r < 0) | |
6438 | goto out_close_fsid; | |
6439 | r = _setup_block_symlink_or_file("block.db", cct->_conf->bluestore_block_db_path, | |
6440 | cct->_conf->bluestore_block_db_size, | |
6441 | cct->_conf->bluestore_block_db_create); | |
6442 | if (r < 0) | |
6443 | goto out_close_fsid; | |
6444 | } | |
6445 | ||
6446 | r = _open_bdev(true); | |
6447 | if (r < 0) | |
6448 | goto out_close_fsid; | |
6449 | ||
3efd9988 FG |
6450 | // choose min_alloc_size |
6451 | if (cct->_conf->bluestore_min_alloc_size) { | |
6452 | min_alloc_size = cct->_conf->bluestore_min_alloc_size; | |
6453 | } else { | |
11fdf7f2 | 6454 | ceph_assert(bdev); |
3efd9988 FG |
6455 | if (bdev->is_rotational()) { |
6456 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_hdd; | |
6457 | } else { | |
6458 | min_alloc_size = cct->_conf->bluestore_min_alloc_size_ssd; | |
6459 | } | |
6460 | } | |
11fdf7f2 | 6461 | _validate_bdev(); |
3efd9988 FG |
6462 | |
6463 | // make sure min_alloc_size is power of 2 aligned. | |
11fdf7f2 | 6464 | if (!isp2(min_alloc_size)) { |
3efd9988 FG |
6465 | derr << __func__ << " min_alloc_size 0x" |
6466 | << std::hex << min_alloc_size << std::dec | |
6467 | << " is not power of 2 aligned!" | |
6468 | << dendl; | |
6469 | r = -EINVAL; | |
6470 | goto out_close_bdev; | |
6471 | } | |
6472 | ||
7c673cae FG |
6473 | r = _open_db(true); |
6474 | if (r < 0) | |
6475 | goto out_close_bdev; | |
6476 | ||
7c673cae FG |
6477 | { |
6478 | KeyValueDB::Transaction t = db->get_transaction(); | |
1911f103 | 6479 | r = _open_fm(t, true); |
11fdf7f2 TL |
6480 | if (r < 0) |
6481 | goto out_close_db; | |
7c673cae FG |
6482 | { |
6483 | bufferlist bl; | |
11fdf7f2 | 6484 | encode((uint64_t)0, bl); |
7c673cae FG |
6485 | t->set(PREFIX_SUPER, "nid_max", bl); |
6486 | t->set(PREFIX_SUPER, "blobid_max", bl); | |
6487 | } | |
6488 | ||
7c673cae FG |
6489 | { |
6490 | bufferlist bl; | |
11fdf7f2 | 6491 | encode((uint64_t)min_alloc_size, bl); |
7c673cae FG |
6492 | t->set(PREFIX_SUPER, "min_alloc_size", bl); |
6493 | } | |
9f95a23c TL |
6494 | { |
6495 | bufferlist bl; | |
6496 | bl.append("1"); | |
6497 | t->set(PREFIX_SUPER, "per_pool_omap", bl); | |
6498 | } | |
7c673cae FG |
6499 | ondisk_format = latest_ondisk_format; |
6500 | _prepare_ondisk_format_super(t); | |
6501 | db->submit_transaction_sync(t); | |
6502 | } | |
6503 | ||
7c673cae FG |
6504 | r = write_meta("kv_backend", cct->_conf->bluestore_kvbackend); |
6505 | if (r < 0) | |
224ce89b WB |
6506 | goto out_close_fm; |
6507 | ||
3efd9988 | 6508 | r = write_meta("bluefs", stringify(bluefs ? 1 : 0)); |
7c673cae | 6509 | if (r < 0) |
224ce89b | 6510 | goto out_close_fm; |
7c673cae FG |
6511 | |
6512 | if (fsid != old_fsid) { | |
6513 | r = _write_fsid(); | |
6514 | if (r < 0) { | |
6515 | derr << __func__ << " error writing fsid: " << cpp_strerror(r) << dendl; | |
224ce89b | 6516 | goto out_close_fm; |
7c673cae FG |
6517 | } |
6518 | } | |
6519 | ||
11fdf7f2 TL |
6520 | if (out_of_sync_fm.fetch_and(0)) { |
6521 | _sync_bluefs_and_fm(); | |
6522 | } | |
6523 | ||
7c673cae FG |
6524 | out_close_fm: |
6525 | _close_fm(); | |
6526 | out_close_db: | |
1911f103 | 6527 | _close_db(false); |
7c673cae FG |
6528 | out_close_bdev: |
6529 | _close_bdev(); | |
6530 | out_close_fsid: | |
6531 | _close_fsid(); | |
6532 | out_path_fd: | |
6533 | _close_path(); | |
6534 | ||
6535 | if (r == 0 && | |
6536 | cct->_conf->bluestore_fsck_on_mkfs) { | |
6537 | int rc = fsck(cct->_conf->bluestore_fsck_on_mkfs_deep); | |
6538 | if (rc < 0) | |
6539 | return rc; | |
6540 | if (rc > 0) { | |
6541 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6542 | r = -EIO; | |
6543 | } | |
11fdf7f2 TL |
6544 | } |
6545 | ||
6546 | if (r == 0) { | |
6547 | // indicate success by writing the 'mkfs_done' file | |
6548 | r = write_meta("mkfs_done", "yes"); | |
6549 | } | |
6550 | ||
6551 | if (r < 0) { | |
6552 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6553 | } else { | |
6554 | dout(0) << __func__ << " success" << dendl; | |
6555 | } | |
6556 | return r; | |
6557 | } | |
6558 | ||
6559 | int BlueStore::_mount_for_bluefs() | |
6560 | { | |
6561 | int r = _open_path(); | |
6562 | ceph_assert(r == 0); | |
6563 | r = _open_fsid(false); | |
6564 | ceph_assert(r == 0); | |
6565 | r = _read_fsid(&fsid); | |
6566 | ceph_assert(r == 0); | |
6567 | r = _lock_fsid(); | |
6568 | ceph_assert(r == 0); | |
6569 | r = _open_bluefs(false); | |
6570 | ceph_assert(r == 0); | |
6571 | return r; | |
6572 | } | |
6573 | ||
6574 | void BlueStore::_umount_for_bluefs() | |
6575 | { | |
1911f103 | 6576 | _close_bluefs(false); |
11fdf7f2 TL |
6577 | _close_fsid(); |
6578 | _close_path(); | |
6579 | } | |
6580 | ||
6581 | int BlueStore::add_new_bluefs_device(int id, const string& dev_path) | |
6582 | { | |
6583 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6584 | int r; | |
6585 | ceph_assert(path_fd < 0); | |
6586 | ||
6587 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6588 | ||
6589 | if (!cct->_conf->bluestore_bluefs) { | |
6590 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6591 | return -EIO; | |
6592 | } | |
6593 | ||
6594 | r = _mount_for_bluefs(); | |
6595 | ||
6596 | int reserved = 0; | |
6597 | if (id == BlueFS::BDEV_NEWWAL) { | |
6598 | string p = path + "/block.wal"; | |
6599 | r = _setup_block_symlink_or_file("block.wal", dev_path, | |
6600 | cct->_conf->bluestore_block_wal_size, | |
6601 | true); | |
6602 | ceph_assert(r == 0); | |
6603 | ||
6604 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, p, | |
eafe8130 | 6605 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6606 | ceph_assert(r == 0); |
6607 | ||
6608 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6609 | r = _check_or_set_bdev_label( | |
6610 | p, | |
6611 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6612 | "bluefs wal", | |
6613 | true); | |
6614 | ceph_assert(r == 0); | |
6615 | } | |
6616 | ||
6617 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
9f95a23c | 6618 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6619 | } else if (id == BlueFS::BDEV_NEWDB) { |
6620 | string p = path + "/block.db"; | |
6621 | r = _setup_block_symlink_or_file("block.db", dev_path, | |
6622 | cct->_conf->bluestore_block_db_size, | |
6623 | true); | |
6624 | ceph_assert(r == 0); | |
6625 | ||
6626 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, p, | |
eafe8130 | 6627 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6628 | ceph_assert(r == 0); |
6629 | ||
6630 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6631 | r = _check_or_set_bdev_label( | |
6632 | p, | |
6633 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6634 | "bluefs db", | |
6635 | true); | |
6636 | ceph_assert(r == 0); | |
6637 | } | |
6638 | reserved = SUPER_RESERVED; | |
9f95a23c TL |
6639 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6640 | bluefs_layout.dedicated_db = true; | |
11fdf7f2 TL |
6641 | } |
6642 | ||
6643 | bluefs->umount(); | |
6644 | bluefs->mount(); | |
6645 | ||
6646 | bluefs->add_block_extent( | |
6647 | id, | |
6648 | reserved, | |
1911f103 | 6649 | bluefs->get_block_device_size(id) - reserved, true); |
11fdf7f2 | 6650 | |
9f95a23c | 6651 | r = bluefs->prepare_new_device(id, bluefs_layout); |
11fdf7f2 TL |
6652 | ceph_assert(r == 0); |
6653 | ||
6654 | if (r < 0) { | |
6655 | derr << __func__ << " failed, " << cpp_strerror(r) << dendl; | |
6656 | } else { | |
6657 | dout(0) << __func__ << " success" << dendl; | |
6658 | } | |
6659 | ||
6660 | _umount_for_bluefs(); | |
6661 | return r; | |
6662 | } | |
6663 | ||
6664 | int BlueStore::migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
6665 | int id) | |
6666 | { | |
6667 | dout(10) << __func__ << " id:" << id << dendl; | |
6668 | ceph_assert(path_fd < 0); | |
6669 | ||
6670 | ceph_assert(id == BlueFS::BDEV_SLOW || id == BlueFS::BDEV_DB); | |
6671 | ||
6672 | if (!cct->_conf->bluestore_bluefs) { | |
6673 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6674 | return -EIO; | |
6675 | } | |
6676 | ||
6677 | int r = _mount_for_bluefs(); | |
6678 | ||
6679 | // require bluestore_bluefs_min_free to be free at target device! | |
6680 | uint64_t used_space = cct->_conf.get_val<Option::size_t>("bluestore_bluefs_min_free"); | |
6681 | for(auto src_id : devs_source) { | |
6682 | used_space += bluefs->get_total(src_id) - bluefs->get_free(src_id); | |
6683 | } | |
6684 | uint64_t target_free = bluefs->get_free(id); | |
6685 | if (id == BlueFS::BDEV_SLOW && target_free < used_space) { | |
6686 | // will need to remount full BlueStore instance to allocate more space | |
6687 | _umount_for_bluefs(); | |
6688 | ||
6689 | r = mount(); | |
6690 | ceph_assert(r == 0); | |
6691 | dout(1) << __func__ | |
6692 | << " Allocating more space at slow device for BlueFS: +" | |
6693 | << used_space - target_free << " bytes" << dendl; | |
6694 | r = allocate_bluefs_freespace( | |
6695 | used_space - target_free, | |
6696 | used_space - target_free, | |
6697 | nullptr); | |
6698 | ||
6699 | umount(); | |
6700 | if (r != 0) { | |
6701 | derr << __func__ | |
6702 | << " can't migrate, unable to allocate extra space: " | |
6703 | << used_space - target_free << " at target:" << id | |
6704 | << dendl; | |
6705 | return -ENOSPC; | |
6706 | } | |
6707 | ||
6708 | r = _mount_for_bluefs(); | |
6709 | ceph_assert(r == 0); | |
6710 | } else if (target_free < used_space) { | |
6711 | derr << __func__ | |
6712 | << " can't migrate, free space at target: " << target_free | |
6713 | << " is less than required space: " << used_space | |
6714 | << dendl; | |
6715 | return -ENOSPC; | |
6716 | } | |
9f95a23c TL |
6717 | if (devs_source.count(BlueFS::BDEV_DB)) { |
6718 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; | |
6719 | bluefs_layout.dedicated_db = false; | |
6720 | } | |
6721 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6722 | bluefs_layout.dedicated_wal = false; | |
6723 | } | |
6724 | r = bluefs->device_migrate_to_existing(cct, devs_source, id, bluefs_layout); | |
11fdf7f2 TL |
6725 | if (r < 0) { |
6726 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; | |
6727 | goto shutdown; | |
6728 | } | |
6729 | ||
6730 | if (devs_source.count(BlueFS::BDEV_DB)) { | |
6731 | r = unlink(string(path + "/block.db").c_str()); | |
6732 | ceph_assert(r == 0); | |
6733 | } | |
6734 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6735 | r = unlink(string(path + "/block.wal").c_str()); | |
6736 | ceph_assert(r == 0); | |
6737 | } | |
6738 | ||
6739 | shutdown: | |
6740 | _umount_for_bluefs(); | |
6741 | return r; | |
6742 | } | |
6743 | ||
6744 | int BlueStore::migrate_to_new_bluefs_device(const set<int>& devs_source, | |
6745 | int id, | |
6746 | const string& dev_path) | |
6747 | { | |
6748 | dout(10) << __func__ << " path " << dev_path << " id:" << id << dendl; | |
6749 | int r; | |
6750 | ceph_assert(path_fd < 0); | |
6751 | ||
6752 | ceph_assert(id == BlueFS::BDEV_NEWWAL || id == BlueFS::BDEV_NEWDB); | |
6753 | ||
6754 | if (!cct->_conf->bluestore_bluefs) { | |
6755 | derr << __func__ << " bluefs isn't configured, can't add new device " << dendl; | |
6756 | return -EIO; | |
6757 | } | |
6758 | ||
6759 | r = _mount_for_bluefs(); | |
6760 | ||
6761 | int reserved = 0; | |
6762 | string link_db; | |
6763 | string link_wal; | |
6764 | if (devs_source.count(BlueFS::BDEV_DB) && | |
9f95a23c | 6765 | bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 6766 | link_db = path + "/block.db"; |
9f95a23c TL |
6767 | bluefs_layout.shared_bdev = BlueFS::BDEV_DB; |
6768 | bluefs_layout.dedicated_db = false; | |
11fdf7f2 TL |
6769 | } |
6770 | if (devs_source.count(BlueFS::BDEV_WAL)) { | |
6771 | link_wal = path + "/block.wal"; | |
9f95a23c | 6772 | bluefs_layout.dedicated_wal = false; |
11fdf7f2 TL |
6773 | } |
6774 | ||
6775 | size_t target_size; | |
6776 | string target_name; | |
6777 | if (id == BlueFS::BDEV_NEWWAL) { | |
6778 | target_name = "block.wal"; | |
6779 | target_size = cct->_conf->bluestore_block_wal_size; | |
9f95a23c | 6780 | bluefs_layout.dedicated_wal = true; |
11fdf7f2 TL |
6781 | |
6782 | r = bluefs->add_block_device(BlueFS::BDEV_NEWWAL, dev_path, | |
eafe8130 | 6783 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6784 | ceph_assert(r == 0); |
6785 | ||
6786 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWWAL)) { | |
6787 | r = _check_or_set_bdev_label( | |
6788 | dev_path, | |
6789 | bluefs->get_block_device_size(BlueFS::BDEV_NEWWAL), | |
6790 | "bluefs wal", | |
6791 | true); | |
6792 | ceph_assert(r == 0); | |
6793 | } | |
6794 | reserved = BDEV_LABEL_BLOCK_SIZE; | |
6795 | } else if (id == BlueFS::BDEV_NEWDB) { | |
6796 | target_name = "block.db"; | |
6797 | target_size = cct->_conf->bluestore_block_db_size; | |
9f95a23c TL |
6798 | bluefs_layout.shared_bdev = BlueFS::BDEV_SLOW; |
6799 | bluefs_layout.dedicated_db = true; | |
31f18b77 | 6800 | |
11fdf7f2 | 6801 | r = bluefs->add_block_device(BlueFS::BDEV_NEWDB, dev_path, |
eafe8130 | 6802 | cct->_conf->bdev_enable_discard); |
11fdf7f2 TL |
6803 | ceph_assert(r == 0); |
6804 | ||
6805 | if (bluefs->bdev_support_label(BlueFS::BDEV_NEWDB)) { | |
6806 | r = _check_or_set_bdev_label( | |
6807 | dev_path, | |
6808 | bluefs->get_block_device_size(BlueFS::BDEV_NEWDB), | |
6809 | "bluefs db", | |
6810 | true); | |
6811 | ceph_assert(r == 0); | |
6812 | } | |
6813 | reserved = SUPER_RESERVED; | |
31f18b77 FG |
6814 | } |
6815 | ||
11fdf7f2 TL |
6816 | bluefs->umount(); |
6817 | bluefs->mount(); | |
6818 | ||
6819 | bluefs->add_block_extent( | |
6820 | id, reserved, bluefs->get_block_device_size(id) - reserved); | |
6821 | ||
9f95a23c | 6822 | r = bluefs->device_migrate_to_new(cct, devs_source, id, bluefs_layout); |
11fdf7f2 | 6823 | |
7c673cae | 6824 | if (r < 0) { |
11fdf7f2 TL |
6825 | derr << __func__ << " failed during BlueFS migration, " << cpp_strerror(r) << dendl; |
6826 | goto shutdown; | |
6827 | } | |
6828 | ||
6829 | if (!link_db.empty()) { | |
6830 | r = unlink(link_db.c_str()); | |
6831 | ceph_assert(r == 0); | |
6832 | } | |
6833 | if (!link_wal.empty()) { | |
6834 | r = unlink(link_wal.c_str()); | |
6835 | ceph_assert(r == 0); | |
6836 | } | |
6837 | r = _setup_block_symlink_or_file( | |
6838 | target_name, | |
6839 | dev_path, | |
6840 | target_size, | |
6841 | true); | |
6842 | ceph_assert(r == 0); | |
6843 | dout(0) << __func__ << " success" << dendl; | |
6844 | ||
6845 | shutdown: | |
6846 | _umount_for_bluefs(); | |
6847 | return r; | |
6848 | } | |
6849 | ||
6850 | string BlueStore::get_device_path(unsigned id) | |
6851 | { | |
6852 | string res; | |
6853 | if (id < BlueFS::MAX_BDEV) { | |
6854 | switch (id) { | |
6855 | case BlueFS::BDEV_WAL: | |
6856 | res = path + "/block.wal"; | |
6857 | break; | |
6858 | case BlueFS::BDEV_DB: | |
9f95a23c | 6859 | if (id == bluefs_layout.shared_bdev) { |
11fdf7f2 TL |
6860 | res = path + "/block"; |
6861 | } else { | |
6862 | res = path + "/block.db"; | |
6863 | } | |
6864 | break; | |
6865 | case BlueFS::BDEV_SLOW: | |
6866 | res = path + "/block"; | |
6867 | break; | |
6868 | } | |
6869 | } | |
6870 | return res; | |
6871 | } | |
6872 | ||
6873 | int BlueStore::expand_devices(ostream& out) | |
6874 | { | |
1911f103 | 6875 | int r = cold_open(); |
11fdf7f2 TL |
6876 | ceph_assert(r == 0); |
6877 | bluefs->dump_block_extents(out); | |
1911f103 | 6878 | out << "Expanding DB/WAL..." << std::endl; |
11fdf7f2 | 6879 | for (auto devid : { BlueFS::BDEV_WAL, BlueFS::BDEV_DB}) { |
9f95a23c | 6880 | if (devid == bluefs_layout.shared_bdev ) { |
11fdf7f2 TL |
6881 | continue; |
6882 | } | |
6883 | uint64_t size = bluefs->get_block_device_size(devid); | |
6884 | if (size == 0) { | |
6885 | // no bdev | |
6886 | continue; | |
6887 | } | |
6888 | ||
6889 | interval_set<uint64_t> before; | |
6890 | bluefs->get_block_extents(devid, &before); | |
6891 | ceph_assert(!before.empty()); | |
6892 | uint64_t end = before.range_end(); | |
6893 | if (end < size) { | |
6894 | out << devid | |
6895 | <<" : expanding " << " from 0x" << std::hex | |
6896 | << end << " to 0x" << size << std::dec << std::endl; | |
6897 | bluefs->add_block_extent(devid, end, size-end); | |
6898 | string p = get_device_path(devid); | |
6899 | const char* path = p.c_str(); | |
6900 | if (path == nullptr) { | |
6901 | derr << devid | |
6902 | <<": can't find device path " << dendl; | |
6903 | continue; | |
6904 | } | |
6905 | bluestore_bdev_label_t label; | |
6906 | int r = _read_bdev_label(cct, path, &label); | |
6907 | if (r < 0) { | |
6908 | derr << "unable to read label for " << path << ": " | |
6909 | << cpp_strerror(r) << dendl; | |
6910 | continue; | |
6911 | } | |
6912 | label.size = size; | |
6913 | r = _write_bdev_label(cct, path, label); | |
6914 | if (r < 0) { | |
6915 | derr << "unable to write label for " << path << ": " | |
6916 | << cpp_strerror(r) << dendl; | |
6917 | continue; | |
6918 | } | |
6919 | out << devid | |
6920 | <<" : size label updated to " << size | |
6921 | << std::endl; | |
6922 | } | |
6923 | } | |
6924 | uint64_t size0 = fm->get_size(); | |
6925 | uint64_t size = bdev->get_size(); | |
6926 | if (size0 < size) { | |
9f95a23c | 6927 | out << bluefs_layout.shared_bdev |
1911f103 TL |
6928 | << " : expanding " << " from 0x" << std::hex |
6929 | << size0 << " to 0x" << size << std::dec << std::endl; | |
6930 | _write_out_fm_meta(size, true); | |
6931 | cold_close(); | |
6932 | ||
6933 | // mount in read/write to sync expansion changes | |
6934 | r = _mount(false); | |
11fdf7f2 | 6935 | ceph_assert(r == 0); |
1911f103 TL |
6936 | umount(); |
6937 | } else { | |
6938 | cold_close(); | |
7c673cae | 6939 | } |
1911f103 TL |
6940 | return r; |
6941 | } | |
6942 | ||
6943 | int BlueStore::dump_bluefs_sizes(ostream& out) | |
6944 | { | |
6945 | int r = cold_open(); | |
6946 | ceph_assert(r == 0); | |
6947 | bluefs->dump_block_extents(out); | |
6948 | cold_close(); | |
7c673cae FG |
6949 | return r; |
6950 | } | |
6951 | ||
6952 | void BlueStore::set_cache_shards(unsigned num) | |
6953 | { | |
6954 | dout(10) << __func__ << " " << num << dendl; | |
9f95a23c TL |
6955 | size_t oold = onode_cache_shards.size(); |
6956 | size_t bold = buffer_cache_shards.size(); | |
6957 | ceph_assert(num >= oold && num >= bold); | |
6958 | onode_cache_shards.resize(num); | |
6959 | buffer_cache_shards.resize(num); | |
6960 | for (unsigned i = oold; i < num; ++i) { | |
6961 | onode_cache_shards[i] = | |
6962 | OnodeCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6963 | logger); | |
6964 | } | |
6965 | for (unsigned i = bold; i < num; ++i) { | |
6966 | buffer_cache_shards[i] = | |
6967 | BufferCacheShard::create(cct, cct->_conf->bluestore_cache_type, | |
6968 | logger); | |
7c673cae FG |
6969 | } |
6970 | } | |
6971 | ||
11fdf7f2 | 6972 | int BlueStore::_mount(bool kv_only, bool open_db) |
7c673cae FG |
6973 | { |
6974 | dout(1) << __func__ << " path " << path << dendl; | |
6975 | ||
3efd9988 FG |
6976 | _kv_only = kv_only; |
6977 | ||
7c673cae FG |
6978 | { |
6979 | string type; | |
6980 | int r = read_meta("type", &type); | |
6981 | if (r < 0) { | |
6982 | derr << __func__ << " failed to load os-type: " << cpp_strerror(r) | |
6983 | << dendl; | |
6984 | return r; | |
6985 | } | |
6986 | ||
6987 | if (type != "bluestore") { | |
6988 | derr << __func__ << " expected bluestore, but type is " << type << dendl; | |
6989 | return -EIO; | |
6990 | } | |
6991 | } | |
6992 | ||
6993 | if (cct->_conf->bluestore_fsck_on_mount) { | |
6994 | int rc = fsck(cct->_conf->bluestore_fsck_on_mount_deep); | |
6995 | if (rc < 0) | |
6996 | return rc; | |
6997 | if (rc > 0) { | |
6998 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
6999 | return -EIO; | |
7000 | } | |
7001 | } | |
7002 | ||
eafe8130 TL |
7003 | if (cct->_conf->osd_max_object_size > OBJECT_MAX_SIZE) { |
7004 | derr << __func__ << " osd_max_object_size " | |
7005 | << cct->_conf->osd_max_object_size << " > bluestore max " | |
7006 | << OBJECT_MAX_SIZE << dendl; | |
7007 | return -EINVAL; | |
7008 | } | |
7009 | ||
7c673cae FG |
7010 | int r = _open_path(); |
7011 | if (r < 0) | |
7012 | return r; | |
7013 | r = _open_fsid(false); | |
7014 | if (r < 0) | |
7015 | goto out_path; | |
7016 | ||
7017 | r = _read_fsid(&fsid); | |
7018 | if (r < 0) | |
7019 | goto out_fsid; | |
7020 | ||
7021 | r = _lock_fsid(); | |
7022 | if (r < 0) | |
7023 | goto out_fsid; | |
7024 | ||
7025 | r = _open_bdev(false); | |
7026 | if (r < 0) | |
7027 | goto out_fsid; | |
7028 | ||
11fdf7f2 TL |
7029 | if (open_db) { |
7030 | r = _open_db_and_around(false); | |
7031 | } else { | |
7032 | // we can bypass db open exclusively in case of kv_only mode | |
7033 | ceph_assert(kv_only); | |
7034 | r = _open_db(false, true); | |
9f95a23c TL |
7035 | } |
7036 | if (r < 0) { | |
7037 | goto out_bdev; | |
11fdf7f2 | 7038 | } |
7c673cae FG |
7039 | |
7040 | if (kv_only) | |
7041 | return 0; | |
7042 | ||
11fdf7f2 TL |
7043 | r = _upgrade_super(); |
7044 | if (r < 0) { | |
7c673cae | 7045 | goto out_db; |
11fdf7f2 | 7046 | } |
7c673cae FG |
7047 | |
7048 | r = _open_collections(); | |
7049 | if (r < 0) | |
11fdf7f2 | 7050 | goto out_db; |
7c673cae FG |
7051 | |
7052 | r = _reload_logger(); | |
7053 | if (r < 0) | |
7054 | goto out_coll; | |
7055 | ||
31f18b77 | 7056 | _kv_start(); |
7c673cae FG |
7057 | |
7058 | r = _deferred_replay(); | |
7059 | if (r < 0) | |
7060 | goto out_stop; | |
7061 | ||
7062 | mempool_thread.init(); | |
7063 | ||
9f95a23c | 7064 | if ((!per_pool_stat_collection || !per_pool_omap) && |
eafe8130 | 7065 | cct->_conf->bluestore_fsck_quick_fix_on_mount == true) { |
9f95a23c TL |
7066 | |
7067 | bool was_per_pool_omap = per_pool_omap; | |
7068 | ||
eafe8130 TL |
7069 | dout(1) << __func__ << " quick-fix on mount" << dendl; |
7070 | _fsck_on_open(FSCK_SHALLOW, true); | |
7071 | ||
7072 | //reread statfs | |
7073 | //FIXME minor: replace with actual open/close? | |
7074 | _open_statfs(); | |
eafe8130 | 7075 | _check_legacy_statfs_alert(); |
9f95a23c TL |
7076 | |
7077 | //set again as hopefully it has been fixed | |
7078 | if (!was_per_pool_omap) { | |
7079 | _set_per_pool_omap(); | |
7080 | } | |
eafe8130 TL |
7081 | } |
7082 | ||
7c673cae FG |
7083 | mounted = true; |
7084 | return 0; | |
7085 | ||
7086 | out_stop: | |
7087 | _kv_stop(); | |
7c673cae | 7088 | out_coll: |
f6b5b4d7 | 7089 | _shutdown_cache(); |
7c673cae | 7090 | out_db: |
1911f103 | 7091 | _close_db_and_around(false); |
7c673cae FG |
7092 | out_bdev: |
7093 | _close_bdev(); | |
7094 | out_fsid: | |
7095 | _close_fsid(); | |
7096 | out_path: | |
7097 | _close_path(); | |
7098 | return r; | |
7099 | } | |
7100 | ||
7101 | int BlueStore::umount() | |
7102 | { | |
11fdf7f2 | 7103 | ceph_assert(_kv_only || mounted); |
7c673cae FG |
7104 | dout(1) << __func__ << dendl; |
7105 | ||
7106 | _osr_drain_all(); | |
7c673cae | 7107 | |
7c673cae | 7108 | mounted = false; |
3efd9988 FG |
7109 | if (!_kv_only) { |
7110 | mempool_thread.shutdown(); | |
7111 | dout(20) << __func__ << " stopping kv thread" << dendl; | |
7112 | _kv_stop(); | |
f6b5b4d7 | 7113 | _shutdown_cache(); |
3efd9988 FG |
7114 | dout(20) << __func__ << " closing" << dendl; |
7115 | ||
3efd9988 | 7116 | } |
1911f103 | 7117 | _close_db_and_around(false); |
7c673cae FG |
7118 | _close_bdev(); |
7119 | _close_fsid(); | |
7120 | _close_path(); | |
7121 | ||
7122 | if (cct->_conf->bluestore_fsck_on_umount) { | |
7123 | int rc = fsck(cct->_conf->bluestore_fsck_on_umount_deep); | |
7124 | if (rc < 0) | |
7125 | return rc; | |
7126 | if (rc > 0) { | |
7127 | derr << __func__ << " fsck found " << rc << " errors" << dendl; | |
7128 | return -EIO; | |
7129 | } | |
7130 | } | |
7131 | return 0; | |
7132 | } | |
7133 | ||
eafe8130 TL |
7134 | int BlueStore::cold_open() |
7135 | { | |
7136 | int r = _open_path(); | |
7137 | if (r < 0) | |
7138 | return r; | |
7139 | r = _open_fsid(false); | |
7140 | if (r < 0) | |
7141 | goto out_path; | |
7142 | ||
7143 | r = _read_fsid(&fsid); | |
7144 | if (r < 0) | |
7145 | goto out_fsid; | |
7146 | ||
7147 | r = _lock_fsid(); | |
7148 | if (r < 0) | |
7149 | goto out_fsid; | |
7150 | ||
7151 | r = _open_bdev(false); | |
7152 | if (r < 0) | |
7153 | goto out_fsid; | |
7154 | r = _open_db_and_around(true); | |
7155 | if (r < 0) { | |
7156 | goto out_bdev; | |
7157 | } | |
7158 | return 0; | |
7159 | out_bdev: | |
7160 | _close_bdev(); | |
7161 | out_fsid: | |
7162 | _close_fsid(); | |
7163 | out_path: | |
7164 | _close_path(); | |
7165 | return r; | |
7166 | } | |
7167 | int BlueStore::cold_close() | |
7168 | { | |
1911f103 | 7169 | _close_db_and_around(true); |
eafe8130 TL |
7170 | _close_bdev(); |
7171 | _close_fsid(); | |
7172 | _close_path(); | |
7173 | return 0; | |
7174 | } | |
7175 | ||
9f95a23c TL |
7176 | // derr wrapper to limit enormous output and avoid log flooding. |
7177 | // Of limited use where such output is expected for now | |
7178 | #define fsck_derr(err_cnt, threshold) \ | |
7179 | if (err_cnt <= threshold) { \ | |
7180 | bool need_skip_print = err_cnt == threshold; \ | |
7181 | derr | |
7182 | ||
7183 | #define fsck_dendl \ | |
7184 | dendl; \ | |
7185 | if (need_skip_print) \ | |
7186 | derr << "more error lines skipped..." << dendl; \ | |
7c673cae | 7187 | } |
7c673cae | 7188 | |
eafe8130 TL |
7189 | int _fsck_sum_extents( |
7190 | const PExtentVector& extents, | |
7191 | bool compressed, | |
7192 | store_statfs_t& expected_statfs) | |
7193 | { | |
7194 | for (auto e : extents) { | |
7195 | if (!e.is_valid()) | |
7196 | continue; | |
7197 | expected_statfs.allocated += e.length; | |
7198 | if (compressed) { | |
7199 | expected_statfs.data_compressed_allocated += e.length; | |
7200 | } | |
7201 | } | |
7202 | return 0; | |
7203 | } | |
7204 | ||
7c673cae | 7205 | int BlueStore::_fsck_check_extents( |
11fdf7f2 | 7206 | const coll_t& cid, |
7c673cae FG |
7207 | const ghobject_t& oid, |
7208 | const PExtentVector& extents, | |
7209 | bool compressed, | |
7210 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 7211 | uint64_t granularity, |
11fdf7f2 | 7212 | BlueStoreRepairer* repairer, |
eafe8130 TL |
7213 | store_statfs_t& expected_statfs, |
7214 | FSCKDepth depth) | |
7c673cae FG |
7215 | { |
7216 | dout(30) << __func__ << " oid " << oid << " extents " << extents << dendl; | |
7217 | int errors = 0; | |
7218 | for (auto e : extents) { | |
7219 | if (!e.is_valid()) | |
7220 | continue; | |
7221 | expected_statfs.allocated += e.length; | |
7222 | if (compressed) { | |
11fdf7f2 | 7223 | expected_statfs.data_compressed_allocated += e.length; |
7c673cae | 7224 | } |
eafe8130 TL |
7225 | if (depth != FSCK_SHALLOW) { |
7226 | bool already = false; | |
9f95a23c | 7227 | apply_for_bitset_range( |
eafe8130 TL |
7228 | e.offset, e.length, granularity, used_blocks, |
7229 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
7230 | if (bs.test(pos)) { |
7231 | if (repairer) { | |
7232 | repairer->note_misreference( | |
7233 | pos * min_alloc_size, min_alloc_size, !already); | |
7234 | } | |
7235 | if (!already) { | |
7236 | derr << "fsck error: " << oid << " extent " << e | |
7237 | << " or a subset is already allocated (misreferenced)" << dendl; | |
7238 | ++errors; | |
7239 | already = true; | |
7240 | } | |
11fdf7f2 | 7241 | } |
eafe8130 TL |
7242 | else |
7243 | bs.set(pos); | |
7244 | }); | |
7245 | if (repairer) { | |
7246 | repairer->get_space_usage_tracker().set_used( e.offset, e.length, cid, oid); | |
7247 | } | |
11fdf7f2 | 7248 | |
eafe8130 TL |
7249 | if (e.end() > bdev->get_size()) { |
7250 | derr << "fsck error: " << oid << " extent " << e | |
7251 | << " past end of block device" << dendl; | |
7252 | ++errors; | |
7253 | } | |
7c673cae FG |
7254 | } |
7255 | } | |
7256 | return errors; | |
7257 | } | |
7258 | ||
11fdf7f2 TL |
7259 | void BlueStore::_fsck_check_pool_statfs( |
7260 | BlueStore::per_pool_statfs& expected_pool_statfs, | |
eafe8130 TL |
7261 | int64_t& errors, |
7262 | int64_t& warnings, | |
11fdf7f2 TL |
7263 | BlueStoreRepairer* repairer) |
7264 | { | |
7265 | auto it = db->get_iterator(PREFIX_STAT); | |
7266 | if (it) { | |
7267 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7268 | string key = it->key(); | |
7269 | if (key == BLUESTORE_GLOBAL_STATFS_KEY) { | |
7270 | if (repairer) { | |
eafe8130 TL |
7271 | ++errors; |
7272 | repairer->remove_key(db, PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY); | |
7273 | derr << "fsck error: " << "legacy statfs record found, removing" | |
11fdf7f2 TL |
7274 | << dendl; |
7275 | } | |
7276 | continue; | |
7277 | } | |
11fdf7f2 TL |
7278 | uint64_t pool_id; |
7279 | if (get_key_pool_stat(key, &pool_id) < 0) { | |
7280 | derr << "fsck error: bad key " << key | |
7281 | << "in statfs namespece" << dendl; | |
7282 | if (repairer) { | |
7283 | repairer->remove_key(db, PREFIX_STAT, key); | |
7284 | } | |
7285 | ++errors; | |
7286 | continue; | |
7287 | } | |
7288 | ||
7289 | volatile_statfs vstatfs; | |
7290 | bufferlist bl = it->value(); | |
7291 | auto blp = bl.cbegin(); | |
7292 | try { | |
7293 | vstatfs.decode(blp); | |
7294 | } catch (buffer::error& e) { | |
7295 | derr << "fsck error: failed to decode Pool StatFS record" | |
7296 | << pretty_binary_string(key) << dendl; | |
7297 | if (repairer) { | |
7298 | dout(20) << __func__ << " undecodable Pool StatFS record, key:'" | |
7299 | << pretty_binary_string(key) | |
7300 | << "', removing" << dendl; | |
7301 | repairer->remove_key(db, PREFIX_STAT, key); | |
7302 | } | |
7303 | ++errors; | |
7304 | vstatfs.reset(); | |
7305 | } | |
7306 | auto stat_it = expected_pool_statfs.find(pool_id); | |
7307 | if (stat_it == expected_pool_statfs.end()) { | |
7308 | if (vstatfs.is_empty()) { | |
7309 | // we don't consider that as an error since empty pool statfs | |
7310 | // are left in DB for now | |
7311 | dout(20) << "fsck inf: found empty stray Pool StatFS record for pool id 0x" | |
7312 | << std::hex << pool_id << std::dec << dendl; | |
7313 | if (repairer) { | |
7314 | // but we need to increment error count in case of repair | |
7315 | // to have proper counters at the end | |
7316 | // (as repairer increments recovery counter anyway). | |
7317 | ++errors; | |
7318 | } | |
7319 | } else { | |
7320 | derr << "fsck error: found stray Pool StatFS record for pool id 0x" | |
7321 | << std::hex << pool_id << std::dec << dendl; | |
7322 | ++errors; | |
7323 | } | |
7324 | if (repairer) { | |
7325 | repairer->remove_key(db, PREFIX_SHARED_BLOB, key); | |
7326 | } | |
7327 | continue; | |
7328 | } | |
7329 | store_statfs_t statfs; | |
7330 | vstatfs.publish(&statfs); | |
7331 | if (!(stat_it->second == statfs)) { | |
7332 | derr << "fsck error: actual " << statfs | |
7333 | << " != expected " << stat_it->second | |
7334 | << " for pool " | |
7335 | << std::hex << pool_id << std::dec << dendl; | |
7336 | if (repairer) { | |
7337 | repairer->fix_statfs(db, key, stat_it->second); | |
7338 | } | |
7339 | ++errors; | |
7340 | } | |
7341 | expected_pool_statfs.erase(stat_it); | |
7342 | } | |
7343 | } // if (it) | |
eafe8130 TL |
7344 | for (auto& s : expected_pool_statfs) { |
7345 | if (s.second.is_zero()) { | |
11fdf7f2 TL |
7346 | // we might lack empty statfs recs in DB |
7347 | continue; | |
7348 | } | |
7349 | derr << "fsck error: missing Pool StatFS record for pool " | |
eafe8130 | 7350 | << std::hex << s.first << std::dec << dendl; |
11fdf7f2 TL |
7351 | if (repairer) { |
7352 | string key; | |
eafe8130 TL |
7353 | get_pool_stat_key(s.first, &key); |
7354 | repairer->fix_statfs(db, key, s.second); | |
11fdf7f2 TL |
7355 | } |
7356 | ++errors; | |
7357 | } | |
eafe8130 | 7358 | if (!per_pool_stat_collection && |
eafe8130 TL |
7359 | repairer) { |
7360 | // by virtue of running this method, we correct the top-level | |
7361 | // error of having global stats | |
7362 | repairer->inc_repaired(); | |
7363 | } | |
11fdf7f2 TL |
7364 | } |
7365 | ||
eafe8130 TL |
7366 | BlueStore::OnodeRef BlueStore::fsck_check_objects_shallow( |
7367 | BlueStore::FSCKDepth depth, | |
7368 | int64_t pool_id, | |
7369 | BlueStore::CollectionRef c, | |
7370 | const ghobject_t& oid, | |
7371 | const string& key, | |
7372 | const bufferlist& value, | |
9f95a23c | 7373 | mempool::bluestore_fsck::list<string>* expecting_shards, |
eafe8130 TL |
7374 | map<BlobRef, bluestore_blob_t::unused_t>* referenced, |
7375 | const BlueStore::FSCK_ObjectCtx& ctx) | |
7376 | { | |
7377 | auto& errors = ctx.errors; | |
7378 | auto& num_objects = ctx.num_objects; | |
7379 | auto& num_extents = ctx.num_extents; | |
7380 | auto& num_blobs = ctx.num_blobs; | |
7381 | auto& num_sharded_objects = ctx.num_sharded_objects; | |
7382 | auto& num_spanning_blobs = ctx.num_spanning_blobs; | |
7383 | auto used_blocks = ctx.used_blocks; | |
7384 | auto sb_info_lock = ctx.sb_info_lock; | |
7385 | auto& sb_info = ctx.sb_info; | |
7386 | auto repairer = ctx.repairer; | |
7387 | ||
7388 | store_statfs_t* res_statfs = (per_pool_stat_collection || repairer) ? | |
7389 | &ctx.expected_pool_statfs[pool_id] : | |
7390 | &ctx.expected_store_statfs; | |
7391 | ||
7392 | dout(10) << __func__ << " " << oid << dendl; | |
7393 | OnodeRef o; | |
7394 | o.reset(Onode::decode(c, oid, key, value)); | |
7395 | ++num_objects; | |
7c673cae | 7396 | |
eafe8130 | 7397 | num_spanning_blobs += o->extent_map.spanning_blob_map.size(); |
7c673cae | 7398 | |
eafe8130 TL |
7399 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
7400 | _dump_onode<30>(cct, *o); | |
7401 | // shards | |
7402 | if (!o->extent_map.shards.empty()) { | |
7403 | ++num_sharded_objects; | |
7404 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 7405 | ceph_assert(expecting_shards); |
eafe8130 TL |
7406 | for (auto& s : o->extent_map.shards) { |
7407 | dout(20) << __func__ << " shard " << *s.shard_info << dendl; | |
9f95a23c | 7408 | expecting_shards->push_back(string()); |
eafe8130 | 7409 | get_extent_shard_key(o->key, s.shard_info->offset, |
9f95a23c | 7410 | &expecting_shards->back()); |
eafe8130 TL |
7411 | if (s.shard_info->offset >= o->onode.size) { |
7412 | derr << "fsck error: " << oid << " shard 0x" << std::hex | |
7413 | << s.shard_info->offset << " past EOF at 0x" << o->onode.size | |
7414 | << std::dec << dendl; | |
7415 | ++errors; | |
7416 | } | |
7417 | } | |
7418 | } | |
7419 | } | |
7c673cae | 7420 | |
eafe8130 TL |
7421 | // lextents |
7422 | uint64_t pos = 0; | |
7423 | mempool::bluestore_fsck::map<BlobRef, | |
7424 | bluestore_blob_use_tracker_t> ref_map; | |
7425 | for (auto& l : o->extent_map.extent_map) { | |
7426 | dout(20) << __func__ << " " << l << dendl; | |
7427 | if (l.logical_offset < pos) { | |
7428 | derr << "fsck error: " << oid << " lextent at 0x" | |
7429 | << std::hex << l.logical_offset | |
7430 | << " overlaps with the previous, which ends at 0x" << pos | |
7431 | << std::dec << dendl; | |
7432 | ++errors; | |
7433 | } | |
7434 | if (depth != FSCK_SHALLOW && | |
7435 | o->extent_map.spans_shard(l.logical_offset, l.length)) { | |
7436 | derr << "fsck error: " << oid << " lextent at 0x" | |
7437 | << std::hex << l.logical_offset << "~" << l.length | |
7438 | << " spans a shard boundary" | |
7439 | << std::dec << dendl; | |
7440 | ++errors; | |
7441 | } | |
7442 | pos = l.logical_offset + l.length; | |
7443 | res_statfs->data_stored += l.length; | |
7444 | ceph_assert(l.blob); | |
7445 | const bluestore_blob_t& blob = l.blob->get_blob(); | |
7446 | ||
7447 | auto& ref = ref_map[l.blob]; | |
7448 | if (ref.is_empty()) { | |
7449 | uint32_t min_release_size = blob.get_release_size(min_alloc_size); | |
7450 | uint32_t l = blob.get_logical_length(); | |
7451 | ref.init(l, min_release_size); | |
7452 | } | |
7453 | ref.get( | |
7454 | l.blob_offset, | |
7455 | l.length); | |
7456 | ++num_extents; | |
7457 | if (depth != FSCK_SHALLOW && | |
7458 | blob.has_unused()) { | |
7459 | ceph_assert(referenced); | |
7460 | auto p = referenced->find(l.blob); | |
7461 | bluestore_blob_t::unused_t* pu; | |
7462 | if (p == referenced->end()) { | |
7463 | pu = &(*referenced)[l.blob]; | |
7464 | } | |
7465 | else { | |
7466 | pu = &p->second; | |
7467 | } | |
7468 | uint64_t blob_len = blob.get_logical_length(); | |
7469 | ceph_assert((blob_len % (sizeof(*pu) * 8)) == 0); | |
7470 | ceph_assert(l.blob_offset + l.length <= blob_len); | |
7471 | uint64_t chunk_size = blob_len / (sizeof(*pu) * 8); | |
7472 | uint64_t start = l.blob_offset / chunk_size; | |
7473 | uint64_t end = | |
7474 | round_up_to(l.blob_offset + l.length, chunk_size) / chunk_size; | |
7475 | for (auto i = start; i < end; ++i) { | |
7476 | (*pu) |= (1u << i); | |
7477 | } | |
7478 | } | |
7479 | } //for (auto& l : o->extent_map.extent_map) | |
7480 | ||
7481 | for (auto& i : ref_map) { | |
7482 | ++num_blobs; | |
7483 | const bluestore_blob_t& blob = i.first->get_blob(); | |
7484 | bool equal = | |
7485 | depth == FSCK_SHALLOW ? true : | |
7486 | i.first->get_blob_use_tracker().equal(i.second); | |
7487 | if (!equal) { | |
7488 | derr << "fsck error: " << oid << " blob " << *i.first | |
7489 | << " doesn't match expected ref_map " << i.second << dendl; | |
7490 | ++errors; | |
7491 | } | |
7492 | if (blob.is_compressed()) { | |
7493 | res_statfs->data_compressed += blob.get_compressed_payload_length(); | |
7494 | res_statfs->data_compressed_original += | |
7495 | i.first->get_referenced_bytes(); | |
7496 | } | |
7497 | if (blob.is_shared()) { | |
7498 | if (i.first->shared_blob->get_sbid() > blobid_max) { | |
7499 | derr << "fsck error: " << oid << " blob " << blob | |
7500 | << " sbid " << i.first->shared_blob->get_sbid() << " > blobid_max " | |
7501 | << blobid_max << dendl; | |
7502 | ++errors; | |
7503 | } | |
7504 | else if (i.first->shared_blob->get_sbid() == 0) { | |
7505 | derr << "fsck error: " << oid << " blob " << blob | |
7506 | << " marked as shared but has uninitialized sbid" | |
7507 | << dendl; | |
7508 | ++errors; | |
7509 | } | |
7510 | // the below lock is optional and provided in multithreading mode only | |
7511 | if (sb_info_lock) { | |
7512 | sb_info_lock->lock(); | |
7513 | } | |
7514 | sb_info_t& sbi = sb_info[i.first->shared_blob->get_sbid()]; | |
7515 | ceph_assert(sbi.cid == coll_t() || sbi.cid == c->cid); | |
7516 | ceph_assert(sbi.pool_id == INT64_MIN || | |
7517 | sbi.pool_id == oid.hobj.get_logical_pool()); | |
7518 | sbi.cid = c->cid; | |
7519 | sbi.pool_id = oid.hobj.get_logical_pool(); | |
7520 | sbi.sb = i.first->shared_blob; | |
7521 | sbi.oids.push_back(oid); | |
7522 | sbi.compressed = blob.is_compressed(); | |
7523 | for (auto e : blob.get_extents()) { | |
7524 | if (e.is_valid()) { | |
7525 | sbi.ref_map.get(e.offset, e.length); | |
7526 | } | |
7527 | } | |
7528 | if (sb_info_lock) { | |
7529 | sb_info_lock->unlock(); | |
7530 | } | |
7531 | } else if (depth != FSCK_SHALLOW) { | |
7532 | ceph_assert(used_blocks); | |
7533 | errors += _fsck_check_extents(c->cid, oid, blob.get_extents(), | |
7534 | blob.is_compressed(), | |
7535 | *used_blocks, | |
7536 | fm->get_alloc_size(), | |
7537 | repairer, | |
7538 | *res_statfs, | |
7539 | depth); | |
7540 | } else { | |
7541 | errors += _fsck_sum_extents( | |
7542 | blob.get_extents(), | |
7543 | blob.is_compressed(), | |
7544 | *res_statfs); | |
7545 | } | |
7546 | } // for (auto& i : ref_map) | |
9f95a23c TL |
7547 | |
7548 | if (o->onode.has_omap()) { | |
7549 | _fsck_check_object_omap(depth, o, ctx); | |
7550 | } | |
7551 | ||
eafe8130 TL |
7552 | return o; |
7553 | } | |
7554 | ||
7555 | #include "common/WorkQueue.h" | |
7556 | ||
7557 | class ShallowFSCKThreadPool : public ThreadPool | |
7558 | { | |
7559 | public: | |
7560 | ShallowFSCKThreadPool(CephContext* cct_, std::string nm, std::string tn, int n) : | |
7561 | ThreadPool(cct_, nm, tn, n) { | |
7562 | } | |
7563 | void worker(ThreadPool::WorkThread* wt) override { | |
7564 | int next_wq = 0; | |
7565 | while (!_stop) { | |
7566 | next_wq %= work_queues.size(); | |
7567 | WorkQueue_ *wq = work_queues[next_wq++]; | |
7568 | ||
7569 | void* item = wq->_void_dequeue(); | |
7570 | if (item) { | |
7571 | processing++; | |
7572 | TPHandle tp_handle(cct, nullptr, wq->timeout_interval, wq->suicide_interval); | |
7573 | wq->_void_process(item, tp_handle); | |
7574 | processing--; | |
7575 | } | |
7576 | } | |
7577 | } | |
7578 | template <size_t BatchLen> | |
7579 | struct FSCKWorkQueue : public ThreadPool::WorkQueue_ | |
7580 | { | |
7581 | struct Entry { | |
7582 | int64_t pool_id; | |
7583 | BlueStore::CollectionRef c; | |
7584 | ghobject_t oid; | |
7585 | string key; | |
7586 | bufferlist value; | |
7587 | }; | |
7588 | struct Batch { | |
7589 | std::atomic<size_t> running = { 0 }; | |
7590 | size_t entry_count = 0; | |
7591 | std::array<Entry, BatchLen> entries; | |
7592 | ||
7593 | int64_t errors = 0; | |
7594 | int64_t warnings = 0; | |
7595 | uint64_t num_objects = 0; | |
7596 | uint64_t num_extents = 0; | |
7597 | uint64_t num_blobs = 0; | |
7598 | uint64_t num_sharded_objects = 0; | |
7599 | uint64_t num_spanning_blobs = 0; | |
7600 | store_statfs_t expected_store_statfs; | |
7601 | BlueStore::per_pool_statfs expected_pool_statfs; | |
7602 | }; | |
7603 | ||
7604 | size_t batchCount; | |
7605 | BlueStore* store = nullptr; | |
7606 | ||
eafe8130 TL |
7607 | ceph::mutex* sb_info_lock = nullptr; |
7608 | BlueStore::sb_info_map_t* sb_info = nullptr; | |
7609 | BlueStoreRepairer* repairer = nullptr; | |
7610 | ||
7611 | Batch* batches = nullptr; | |
7612 | size_t last_batch_pos = 0; | |
7613 | bool batch_acquired = false; | |
7614 | ||
7615 | FSCKWorkQueue(std::string n, | |
7616 | size_t _batchCount, | |
7617 | BlueStore* _store, | |
eafe8130 TL |
7618 | ceph::mutex* _sb_info_lock, |
7619 | BlueStore::sb_info_map_t& _sb_info, | |
7620 | BlueStoreRepairer* _repairer) : | |
7621 | WorkQueue_(n, time_t(), time_t()), | |
7622 | batchCount(_batchCount), | |
7623 | store(_store), | |
eafe8130 TL |
7624 | sb_info_lock(_sb_info_lock), |
7625 | sb_info(&_sb_info), | |
7626 | repairer(_repairer) | |
7627 | { | |
7628 | batches = new Batch[batchCount]; | |
7629 | } | |
7630 | ~FSCKWorkQueue() { | |
7631 | delete[] batches; | |
7632 | } | |
7633 | ||
7634 | /// Remove all work items from the queue. | |
7635 | void _clear() override { | |
7636 | //do nothing | |
7637 | } | |
7638 | /// Check whether there is anything to do. | |
7639 | bool _empty() override { | |
7640 | ceph_assert(false); | |
7641 | } | |
7642 | ||
7643 | /// Get the next work item to process. | |
7644 | void* _void_dequeue() override { | |
7645 | size_t pos = rand() % batchCount; | |
7646 | size_t pos0 = pos; | |
7647 | do { | |
7648 | auto& batch = batches[pos]; | |
7649 | if (batch.running.fetch_add(1) == 0) { | |
7650 | if (batch.entry_count) { | |
7651 | return &batch; | |
7652 | } | |
7653 | } | |
7654 | batch.running--; | |
7655 | pos++; | |
7656 | pos %= batchCount; | |
7657 | } while (pos != pos0); | |
7658 | return nullptr; | |
7659 | } | |
7660 | /** @brief Process the work item. | |
7661 | * This function will be called several times in parallel | |
7662 | * and must therefore be thread-safe. */ | |
7663 | void _void_process(void* item, TPHandle& handle) override { | |
7664 | Batch* batch = (Batch*)item; | |
7665 | ||
7666 | BlueStore::FSCK_ObjectCtx ctx( | |
7667 | batch->errors, | |
7668 | batch->warnings, | |
7669 | batch->num_objects, | |
7670 | batch->num_extents, | |
7671 | batch->num_blobs, | |
7672 | batch->num_sharded_objects, | |
7673 | batch->num_spanning_blobs, | |
7674 | nullptr, // used_blocks | |
9f95a23c | 7675 | nullptr, //used_omap_head |
eafe8130 TL |
7676 | sb_info_lock, |
7677 | *sb_info, | |
7678 | batch->expected_store_statfs, | |
7679 | batch->expected_pool_statfs, | |
7680 | repairer); | |
7681 | ||
7682 | for (size_t i = 0; i < batch->entry_count; i++) { | |
7683 | auto& entry = batch->entries[i]; | |
7684 | ||
7685 | store->fsck_check_objects_shallow( | |
7686 | BlueStore::FSCK_SHALLOW, | |
7687 | entry.pool_id, | |
7688 | entry.c, | |
7689 | entry.oid, | |
7690 | entry.key, | |
7691 | entry.value, | |
9f95a23c | 7692 | nullptr, // expecting_shards - this will need a protection if passed |
eafe8130 TL |
7693 | nullptr, // referenced |
7694 | ctx); | |
7695 | } | |
7696 | //std::cout << "processed " << batch << std::endl; | |
7697 | batch->entry_count = 0; | |
7698 | batch->running--; | |
7699 | } | |
7700 | /** @brief Synchronously finish processing a work item. | |
7701 | * This function is called after _void_process with the global thread pool lock held, | |
7702 | * so at most one copy will execute simultaneously for a given thread pool. | |
7703 | * It can be used for non-thread-safe finalization. */ | |
7704 | void _void_process_finish(void*) override { | |
7705 | ceph_assert(false); | |
7706 | } | |
7707 | ||
7708 | bool queue( | |
7709 | int64_t pool_id, | |
7710 | BlueStore::CollectionRef c, | |
7711 | const ghobject_t& oid, | |
7712 | const string& key, | |
7713 | const bufferlist& value) { | |
7714 | bool res = false; | |
7715 | size_t pos0 = last_batch_pos; | |
7716 | if (!batch_acquired) { | |
7717 | do { | |
7718 | auto& batch = batches[last_batch_pos]; | |
7719 | if (batch.running.fetch_add(1) == 0) { | |
7720 | if (batch.entry_count < BatchLen) { | |
7721 | batch_acquired = true; | |
7722 | break; | |
7723 | } | |
7724 | } | |
7725 | batch.running.fetch_sub(1); | |
7726 | last_batch_pos++; | |
7727 | last_batch_pos %= batchCount; | |
7728 | } while (last_batch_pos != pos0); | |
7729 | } | |
7730 | if (batch_acquired) { | |
7731 | auto& batch = batches[last_batch_pos]; | |
7732 | ceph_assert(batch.running); | |
7733 | ceph_assert(batch.entry_count < BatchLen); | |
7734 | ||
7735 | auto& entry = batch.entries[batch.entry_count]; | |
7736 | entry.pool_id = pool_id; | |
7737 | entry.c = c; | |
7738 | entry.oid = oid; | |
7739 | entry.key = key; | |
7740 | entry.value = value; | |
7741 | ||
7742 | ++batch.entry_count; | |
7743 | if (batch.entry_count == BatchLen) { | |
7744 | batch_acquired = false; | |
7745 | batch.running.fetch_sub(1); | |
7746 | last_batch_pos++; | |
7747 | last_batch_pos %= batchCount; | |
7748 | } | |
7749 | res = true; | |
7750 | } | |
7751 | return res; | |
7752 | } | |
7753 | ||
7754 | void finalize(ThreadPool& tp, | |
7755 | BlueStore::FSCK_ObjectCtx& ctx) { | |
7756 | if (batch_acquired) { | |
7757 | auto& batch = batches[last_batch_pos]; | |
7758 | ceph_assert(batch.running); | |
7759 | batch.running.fetch_sub(1); | |
7760 | } | |
7761 | tp.stop(); | |
7762 | ||
7763 | for (size_t i = 0; i < batchCount; i++) { | |
7764 | auto& batch = batches[i]; | |
7765 | ||
7766 | //process leftovers if any | |
7767 | if (batch.entry_count) { | |
7768 | TPHandle tp_handle(store->cct, | |
7769 | nullptr, | |
7770 | timeout_interval, | |
7771 | suicide_interval); | |
7772 | ceph_assert(batch.running == 0); | |
7773 | ||
7774 | batch.running++; // just to be on-par with the regular call | |
7775 | _void_process(&batch, tp_handle); | |
7776 | } | |
7777 | ceph_assert(batch.entry_count == 0); | |
7778 | ||
7779 | ctx.errors += batch.errors; | |
7780 | ctx.warnings += batch.warnings; | |
7781 | ctx.num_objects += batch.num_objects; | |
7782 | ctx.num_extents += batch.num_extents; | |
7783 | ctx.num_blobs += batch.num_blobs; | |
7784 | ctx.num_sharded_objects += batch.num_sharded_objects; | |
7785 | ctx.num_spanning_blobs += batch.num_spanning_blobs; | |
9f95a23c | 7786 | |
eafe8130 TL |
7787 | ctx.expected_store_statfs.add(batch.expected_store_statfs); |
7788 | ||
7789 | for (auto it = batch.expected_pool_statfs.begin(); | |
7790 | it != batch.expected_pool_statfs.end(); | |
7791 | it++) { | |
7792 | ctx.expected_pool_statfs[it->first].add(it->second); | |
7793 | } | |
7794 | } | |
7795 | } | |
7796 | }; | |
7797 | }; | |
7798 | ||
9f95a23c TL |
7799 | void BlueStore::_fsck_check_object_omap(FSCKDepth depth, |
7800 | OnodeRef& o, | |
7801 | const BlueStore::FSCK_ObjectCtx& ctx) | |
eafe8130 | 7802 | { |
9f95a23c TL |
7803 | auto& errors = ctx.errors; |
7804 | auto& warnings = ctx.warnings; | |
7805 | auto repairer = ctx.repairer; | |
7806 | ||
7807 | ceph_assert(o->onode.has_omap()); | |
7808 | if (!o->onode.is_perpool_omap() && !o->onode.is_pgmeta_omap()) { | |
7809 | if (per_pool_omap) { | |
7810 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
7811 | << "fsck error: " << o->oid | |
7812 | << " has omap that is not per-pool or pgmeta" | |
7813 | << fsck_dendl; | |
7814 | ++errors; | |
7815 | } else { | |
7816 | const char* w; | |
7817 | int64_t num; | |
7818 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
7819 | ++errors; | |
7820 | num = errors; | |
7821 | w = "error"; | |
7822 | } else { | |
7823 | ++warnings; | |
7824 | num = warnings; | |
7825 | w = "warning"; | |
7826 | } | |
7827 | fsck_derr(num, MAX_FSCK_ERROR_LINES) | |
7828 | << "fsck " << w << ": " << o->oid | |
7829 | << " has omap that is not per-pool or pgmeta" | |
7830 | << fsck_dendl; | |
7831 | } | |
7832 | } | |
7833 | if (repairer && | |
7834 | !o->onode.is_perpool_omap() && | |
7835 | !o->onode.is_pgmeta_omap()) { | |
7836 | dout(10) << "fsck converting " << o->oid << " omap to per-pool" << dendl; | |
7837 | bufferlist h; | |
7838 | map<string, bufferlist> kv; | |
7839 | int r = _onode_omap_get(o, &h, &kv); | |
7840 | if (r < 0) { | |
7841 | derr << " got " << r << " " << cpp_strerror(r) << dendl; | |
7842 | } else { | |
7843 | KeyValueDB::Transaction txn = db->get_transaction(); | |
7844 | // remove old keys | |
7845 | const string& old_omap_prefix = o->get_omap_prefix(); | |
7846 | string old_head, old_tail; | |
7847 | o->get_omap_header(&old_head); | |
7848 | o->get_omap_tail(&old_tail); | |
7849 | txn->rm_range_keys(old_omap_prefix, old_head, old_tail); | |
7850 | txn->rmkey(old_omap_prefix, old_tail); | |
7851 | // set flag | |
7852 | o->onode.set_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP); | |
7853 | _record_onode(o, txn); | |
7854 | const string& new_omap_prefix = o->get_omap_prefix(); | |
7855 | // head | |
7856 | if (h.length()) { | |
7857 | string new_head; | |
7858 | o->get_omap_header(&new_head); | |
7859 | txn->set(new_omap_prefix, new_head, h); | |
7860 | } | |
7861 | // tail | |
7862 | string new_tail; | |
7863 | o->get_omap_tail(&new_tail); | |
7864 | bufferlist empty; | |
7865 | txn->set(new_omap_prefix, new_tail, empty); | |
7866 | // values | |
7867 | string final_key; | |
7868 | o->get_omap_key(string(), &final_key); | |
7869 | size_t base_key_len = final_key.size(); | |
7870 | for (auto& i : kv) { | |
7871 | final_key.resize(base_key_len); | |
7872 | final_key += i.first; | |
7873 | txn->set(new_omap_prefix, final_key, i.second); | |
7874 | } | |
7875 | db->submit_transaction_sync(txn); | |
7876 | repairer->inc_repaired(); | |
7877 | } | |
eafe8130 | 7878 | } |
9f95a23c | 7879 | } |
eafe8130 | 7880 | |
9f95a23c TL |
7881 | void BlueStore::_fsck_check_objects(FSCKDepth depth, |
7882 | BlueStore::FSCK_ObjectCtx& ctx) | |
7883 | { | |
eafe8130 | 7884 | auto& errors = ctx.errors; |
eafe8130 TL |
7885 | auto sb_info_lock = ctx.sb_info_lock; |
7886 | auto& sb_info = ctx.sb_info; | |
7887 | auto repairer = ctx.repairer; | |
7888 | ||
7889 | uint64_t_btree_t used_nids; | |
7890 | ||
7891 | size_t processed_myself = 0; | |
7892 | ||
7893 | auto it = db->get_iterator(PREFIX_OBJ); | |
7894 | mempool::bluestore_fsck::list<string> expecting_shards; | |
7895 | if (it) { | |
7896 | const size_t thread_count = cct->_conf->bluestore_fsck_quick_fix_threads; | |
7897 | typedef ShallowFSCKThreadPool::FSCKWorkQueue<256> WQ; | |
7898 | std::unique_ptr<WQ> wq( | |
7899 | new WQ( | |
7900 | "FSCKWorkQueue", | |
7901 | (thread_count ? : 1) * 32, | |
7902 | this, | |
eafe8130 TL |
7903 | sb_info_lock, |
7904 | sb_info, | |
7905 | repairer)); | |
7906 | ||
7907 | ShallowFSCKThreadPool thread_pool(cct, "ShallowFSCKThreadPool", "ShallowFSCK", thread_count); | |
7908 | ||
7909 | thread_pool.add_work_queue(wq.get()); | |
7910 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
7911 | //not the best place but let's check anyway | |
7912 | ceph_assert(sb_info_lock); | |
7913 | thread_pool.start(); | |
7914 | } | |
7915 | ||
7916 | //fill global if not overriden below | |
7917 | CollectionRef c; | |
7918 | int64_t pool_id = -1; | |
7919 | spg_t pgid; | |
7920 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
7921 | dout(30) << __func__ << " key " | |
7922 | << pretty_binary_string(it->key()) << dendl; | |
7923 | if (is_extent_shard_key(it->key())) { | |
7924 | if (depth == FSCK_SHALLOW) { | |
7925 | continue; | |
7926 | } | |
7927 | while (!expecting_shards.empty() && | |
7928 | expecting_shards.front() < it->key()) { | |
7929 | derr << "fsck error: missing shard key " | |
7930 | << pretty_binary_string(expecting_shards.front()) | |
7931 | << dendl; | |
7932 | ++errors; | |
7933 | expecting_shards.pop_front(); | |
7934 | } | |
7935 | if (!expecting_shards.empty() && | |
7936 | expecting_shards.front() == it->key()) { | |
7937 | // all good | |
7938 | expecting_shards.pop_front(); | |
7939 | continue; | |
7940 | } | |
7941 | ||
7942 | uint32_t offset; | |
7943 | string okey; | |
7944 | get_key_extent_shard(it->key(), &okey, &offset); | |
7945 | derr << "fsck error: stray shard 0x" << std::hex << offset | |
7946 | << std::dec << dendl; | |
7947 | if (expecting_shards.empty()) { | |
7948 | derr << "fsck error: " << pretty_binary_string(it->key()) | |
7949 | << " is unexpected" << dendl; | |
7950 | ++errors; | |
7951 | continue; | |
7952 | } | |
7953 | while (expecting_shards.front() > it->key()) { | |
7954 | derr << "fsck error: saw " << pretty_binary_string(it->key()) | |
7955 | << dendl; | |
7956 | derr << "fsck error: exp " | |
7957 | << pretty_binary_string(expecting_shards.front()) << dendl; | |
7958 | ++errors; | |
7959 | expecting_shards.pop_front(); | |
7960 | if (expecting_shards.empty()) { | |
7961 | break; | |
7962 | } | |
7963 | } | |
7964 | continue; | |
7965 | } | |
7966 | ||
7967 | ghobject_t oid; | |
7968 | int r = get_key_object(it->key(), &oid); | |
7969 | if (r < 0) { | |
7970 | derr << "fsck error: bad object key " | |
7971 | << pretty_binary_string(it->key()) << dendl; | |
7972 | ++errors; | |
7973 | continue; | |
7974 | } | |
7975 | if (!c || | |
7976 | oid.shard_id != pgid.shard || | |
7977 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
7978 | !c->contains(oid)) { | |
7979 | c = nullptr; | |
7980 | for (auto& p : coll_map) { | |
7981 | if (p.second->contains(oid)) { | |
7982 | c = p.second; | |
7983 | break; | |
7984 | } | |
7985 | } | |
7986 | if (!c) { | |
7987 | derr << "fsck error: stray object " << oid | |
7988 | << " not owned by any collection" << dendl; | |
7989 | ++errors; | |
7990 | continue; | |
7991 | } | |
7992 | pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
7993 | dout(20) << __func__ << " collection " << c->cid << " " << c->cnode | |
7994 | << dendl; | |
7995 | } | |
7996 | ||
7997 | if (depth != FSCK_SHALLOW && | |
7998 | !expecting_shards.empty()) { | |
7999 | for (auto& k : expecting_shards) { | |
8000 | derr << "fsck error: missing shard key " | |
8001 | << pretty_binary_string(k) << dendl; | |
8002 | } | |
8003 | ++errors; | |
8004 | expecting_shards.clear(); | |
8005 | } | |
8006 | ||
8007 | bool queued = false; | |
8008 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8009 | queued = wq->queue( | |
8010 | pool_id, | |
8011 | c, | |
8012 | oid, | |
8013 | it->key(), | |
8014 | it->value()); | |
8015 | } | |
8016 | OnodeRef o; | |
8017 | map<BlobRef, bluestore_blob_t::unused_t> referenced; | |
8018 | ||
8019 | if (!queued) { | |
8020 | ++processed_myself; | |
8021 | ||
8022 | o = fsck_check_objects_shallow( | |
8023 | depth, | |
8024 | pool_id, | |
8025 | c, | |
8026 | oid, | |
8027 | it->key(), | |
8028 | it->value(), | |
9f95a23c | 8029 | &expecting_shards, |
eafe8130 TL |
8030 | &referenced, |
8031 | ctx); | |
8032 | } | |
8033 | ||
8034 | if (depth != FSCK_SHALLOW) { | |
8035 | ceph_assert(o != nullptr); | |
8036 | if (o->onode.nid) { | |
8037 | if (o->onode.nid > nid_max) { | |
8038 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8039 | << " > nid_max " << nid_max << dendl; | |
8040 | ++errors; | |
8041 | } | |
8042 | if (used_nids.count(o->onode.nid)) { | |
8043 | derr << "fsck error: " << oid << " nid " << o->onode.nid | |
8044 | << " already in use" << dendl; | |
8045 | ++errors; | |
8046 | continue; // go for next object | |
8047 | } | |
8048 | used_nids.insert(o->onode.nid); | |
8049 | } | |
8050 | for (auto& i : referenced) { | |
8051 | dout(20) << __func__ << " referenced 0x" << std::hex << i.second | |
8052 | << std::dec << " for " << *i.first << dendl; | |
8053 | const bluestore_blob_t& blob = i.first->get_blob(); | |
8054 | if (i.second & blob.unused) { | |
8055 | derr << "fsck error: " << oid << " blob claims unused 0x" | |
8056 | << std::hex << blob.unused | |
8057 | << " but extents reference 0x" << i.second << std::dec | |
8058 | << " on blob " << *i.first << dendl; | |
8059 | ++errors; | |
8060 | } | |
8061 | if (blob.has_csum()) { | |
8062 | uint64_t blob_len = blob.get_logical_length(); | |
8063 | uint64_t unused_chunk_size = blob_len / (sizeof(blob.unused) * 8); | |
8064 | unsigned csum_count = blob.get_csum_count(); | |
8065 | unsigned csum_chunk_size = blob.get_csum_chunk_size(); | |
8066 | for (unsigned p = 0; p < csum_count; ++p) { | |
8067 | unsigned pos = p * csum_chunk_size; | |
8068 | unsigned firstbit = pos / unused_chunk_size; // [firstbit,lastbit] | |
8069 | unsigned lastbit = (pos + csum_chunk_size - 1) / unused_chunk_size; | |
8070 | unsigned mask = 1u << firstbit; | |
8071 | for (unsigned b = firstbit + 1; b <= lastbit; ++b) { | |
8072 | mask |= 1u << b; | |
8073 | } | |
8074 | if ((blob.unused & mask) == mask) { | |
8075 | // this csum chunk region is marked unused | |
8076 | if (blob.get_csum_item(p) != 0) { | |
8077 | derr << "fsck error: " << oid | |
8078 | << " blob claims csum chunk 0x" << std::hex << pos | |
8079 | << "~" << csum_chunk_size | |
8080 | << " is unused (mask 0x" << mask << " of unused 0x" | |
8081 | << blob.unused << ") but csum is non-zero 0x" | |
8082 | << blob.get_csum_item(p) << std::dec << " on blob " | |
8083 | << *i.first << dendl; | |
8084 | ++errors; | |
8085 | } | |
8086 | } | |
8087 | } | |
8088 | } | |
8089 | } | |
8090 | // omap | |
8091 | if (o->onode.has_omap()) { | |
9f95a23c TL |
8092 | ceph_assert(ctx.used_omap_head); |
8093 | if (ctx.used_omap_head->count(o->onode.nid)) { | |
8094 | derr << "fsck error: " << o->oid << " omap_head " << o->onode.nid | |
8095 | << " already in use" << dendl; | |
eafe8130 TL |
8096 | ++errors; |
8097 | } else { | |
9f95a23c | 8098 | ctx.used_omap_head->insert(o->onode.nid); |
eafe8130 | 8099 | } |
9f95a23c | 8100 | } // if (o->onode.has_omap()) |
eafe8130 TL |
8101 | if (depth == FSCK_DEEP) { |
8102 | bufferlist bl; | |
8103 | uint64_t max_read_block = cct->_conf->bluestore_fsck_read_bytes_cap; | |
8104 | uint64_t offset = 0; | |
8105 | do { | |
8106 | uint64_t l = std::min(uint64_t(o->onode.size - offset), max_read_block); | |
8107 | int r = _do_read(c.get(), o, offset, l, bl, | |
8108 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE); | |
8109 | if (r < 0) { | |
8110 | ++errors; | |
8111 | derr << "fsck error: " << oid << std::hex | |
8112 | << " error during read: " | |
8113 | << " " << offset << "~" << l | |
8114 | << " " << cpp_strerror(r) << std::dec | |
8115 | << dendl; | |
8116 | break; | |
8117 | } | |
8118 | offset += l; | |
8119 | } while (offset < o->onode.size); | |
8120 | } // deep | |
8121 | } //if (depth != FSCK_SHALLOW) | |
8122 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8123 | if (depth == FSCK_SHALLOW && thread_count > 0) { | |
8124 | wq->finalize(thread_pool, ctx); | |
8125 | if (processed_myself) { | |
8126 | // may be needs more threads? | |
8127 | dout(0) << __func__ << " partial offload" | |
8128 | << ", done myself " << processed_myself | |
8129 | << " of " << ctx.num_objects | |
8130 | << "objects, threads " << thread_count | |
8131 | << dendl; | |
8132 | } | |
8133 | } | |
8134 | } // if (it) | |
8135 | } | |
8136 | /** | |
8137 | An overview for currently implemented repair logics | |
8138 | performed in fsck in two stages: detection(+preparation) and commit. | |
8139 | Detection stage (in processing order): | |
8140 | (Issue -> Repair action to schedule) | |
8141 | - Detect undecodable keys for Shared Blobs -> Remove | |
8142 | - Detect undecodable records for Shared Blobs -> Remove | |
8143 | (might trigger missed Shared Blob detection below) | |
8144 | - Detect stray records for Shared Blobs -> Remove | |
8145 | - Detect misreferenced pextents -> Fix | |
8146 | Prepare Bloom-like filter to track cid/oid -> pextent | |
8147 | Prepare list of extents that are improperly referenced | |
8148 | Enumerate Onode records that might use 'misreferenced' pextents | |
8149 | (Bloom-like filter applied to reduce computation) | |
8150 | Per each questinable Onode enumerate all blobs and identify broken ones | |
8151 | (i.e. blobs having 'misreferences') | |
8152 | Rewrite each broken blob data by allocating another extents and | |
8153 | copying data there | |
8154 | If blob is shared - unshare it and mark corresponding Shared Blob | |
8155 | for removal | |
8156 | Release previously allocated space | |
8157 | Update Extent Map | |
8158 | - Detect missed Shared Blobs -> Recreate | |
8159 | - Detect undecodable deferred transaction -> Remove | |
8160 | - Detect Freelist Manager's 'false free' entries -> Mark as used | |
8161 | - Detect Freelist Manager's leaked entries -> Mark as free | |
8162 | - Detect statfs inconsistency - Update | |
8163 | Commit stage (separate DB commit per each step): | |
8164 | - Apply leaked FM entries fix | |
8165 | - Apply 'false free' FM entries fix | |
8166 | - Apply 'Remove' actions | |
8167 | - Apply fix for misreference pextents | |
8168 | - Apply Shared Blob recreate | |
8169 | (can be merged with the step above if misreferences were dectected) | |
8170 | - Apply StatFS update | |
8171 | */ | |
8172 | int BlueStore::_fsck(BlueStore::FSCKDepth depth, bool repair) | |
8173 | { | |
8174 | dout(1) << __func__ | |
8175 | << (repair ? " repair" : " check") | |
8176 | << (depth == FSCK_DEEP ? " (deep)" : | |
8177 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8178 | << dendl; | |
8179 | ||
8180 | // in deep mode we need R/W write access to be able to replay deferred ops | |
8181 | bool read_only = !(repair || depth == FSCK_DEEP); | |
8182 | ||
8183 | int r = _open_path(); | |
8184 | if (r < 0) | |
8185 | return r; | |
7c673cae FG |
8186 | r = _open_fsid(false); |
8187 | if (r < 0) | |
8188 | goto out_path; | |
8189 | ||
8190 | r = _read_fsid(&fsid); | |
8191 | if (r < 0) | |
8192 | goto out_fsid; | |
8193 | ||
8194 | r = _lock_fsid(); | |
8195 | if (r < 0) | |
8196 | goto out_fsid; | |
8197 | ||
8198 | r = _open_bdev(false); | |
8199 | if (r < 0) | |
8200 | goto out_fsid; | |
8201 | ||
11fdf7f2 | 8202 | r = _open_db_and_around(read_only); |
7c673cae FG |
8203 | if (r < 0) |
8204 | goto out_bdev; | |
8205 | ||
11fdf7f2 TL |
8206 | if (!read_only) { |
8207 | r = _upgrade_super(); | |
8208 | if (r < 0) { | |
8209 | goto out_db; | |
8210 | } | |
8211 | } | |
7c673cae | 8212 | |
eafe8130 | 8213 | r = _open_collections(); |
7c673cae | 8214 | if (r < 0) |
11fdf7f2 | 8215 | goto out_db; |
7c673cae FG |
8216 | |
8217 | mempool_thread.init(); | |
8218 | ||
11fdf7f2 TL |
8219 | // we need finisher and kv_{sync,finalize}_thread *just* for replay |
8220 | // enable in repair or deep mode modes only | |
8221 | if (!read_only) { | |
8222 | _kv_start(); | |
8223 | r = _deferred_replay(); | |
8224 | _kv_stop(); | |
8225 | } | |
7c673cae FG |
8226 | if (r < 0) |
8227 | goto out_scan; | |
8228 | ||
eafe8130 TL |
8229 | r = _fsck_on_open(depth, repair); |
8230 | ||
8231 | out_scan: | |
8232 | mempool_thread.shutdown(); | |
f6b5b4d7 | 8233 | _shutdown_cache(); |
eafe8130 | 8234 | out_db: |
1911f103 | 8235 | _close_db_and_around(false); |
eafe8130 TL |
8236 | out_bdev: |
8237 | _close_bdev(); | |
8238 | out_fsid: | |
8239 | _close_fsid(); | |
8240 | out_path: | |
8241 | _close_path(); | |
8242 | ||
8243 | return r; | |
8244 | } | |
8245 | ||
8246 | int BlueStore::_fsck_on_open(BlueStore::FSCKDepth depth, bool repair) | |
8247 | { | |
8248 | dout(1) << __func__ | |
8249 | << " <<<START>>>" | |
8250 | << (repair ? " repair" : " check") | |
8251 | << (depth == FSCK_DEEP ? " (deep)" : | |
8252 | depth == FSCK_SHALLOW ? " (shallow)" : " (regular)") | |
8253 | << " start" << dendl; | |
8254 | int64_t errors = 0; | |
8255 | int64_t warnings = 0; | |
8256 | unsigned repaired = 0; | |
8257 | ||
8258 | uint64_t_btree_t used_omap_head; | |
eafe8130 TL |
8259 | uint64_t_btree_t used_sbids; |
8260 | ||
8261 | mempool_dynamic_bitset used_blocks; | |
8262 | KeyValueDB::Iterator it; | |
8263 | store_statfs_t expected_store_statfs, actual_statfs; | |
8264 | per_pool_statfs expected_pool_statfs; | |
8265 | ||
8266 | sb_info_map_t sb_info; | |
8267 | ||
8268 | uint64_t num_objects = 0; | |
8269 | uint64_t num_extents = 0; | |
8270 | uint64_t num_blobs = 0; | |
8271 | uint64_t num_spanning_blobs = 0; | |
8272 | uint64_t num_shared_blobs = 0; | |
8273 | uint64_t num_sharded_objects = 0; | |
8274 | BlueStoreRepairer repairer; | |
8275 | ||
8276 | utime_t start = ceph_clock_now(); | |
8277 | ||
8278 | _fsck_collections(&errors); | |
b32b8144 | 8279 | used_blocks.resize(fm->get_alloc_units()); |
9f95a23c | 8280 | apply_for_bitset_range( |
11fdf7f2 | 8281 | 0, std::max<uint64_t>(min_alloc_size, SUPER_RESERVED), fm->get_alloc_size(), used_blocks, |
7c673cae FG |
8282 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
8283 | bs.set(pos); | |
8284 | } | |
8285 | ); | |
11fdf7f2 TL |
8286 | if (repair) { |
8287 | repairer.get_space_usage_tracker().init( | |
8288 | bdev->get_size(), | |
8289 | min_alloc_size); | |
8290 | } | |
7c673cae FG |
8291 | |
8292 | if (bluefs) { | |
11fdf7f2 TL |
8293 | if( cct->_conf->bluestore_bluefs_db_compatibility) { |
8294 | interval_set<uint64_t> bluefs_extents_db; | |
8295 | bufferlist bl; | |
8296 | db->get(PREFIX_SUPER, "bluefs_extents", &bl); | |
8297 | auto p = bl.cbegin(); | |
8298 | auto prev_errors = errors; | |
8299 | try { | |
8300 | decode(bluefs_extents_db, p); | |
8301 | bluefs_extents_db.union_of(bluefs_extents); | |
8302 | bluefs_extents_db.subtract(bluefs_extents); | |
8303 | if (!bluefs_extents_db.empty()) { | |
8304 | derr << "fsck error: bluefs_extents inconsistency, " | |
8305 | << "downgrade to previous releases might be broken." | |
8306 | << dendl; | |
8307 | ++errors; | |
8308 | } | |
8309 | } | |
8310 | catch (buffer::error& e) { | |
8311 | derr << "fsck error: failed to retrieve bluefs_extents from kv" << dendl; | |
8312 | ++errors; | |
8313 | } | |
8314 | if (errors != prev_errors && repair) { | |
8315 | repairer.fix_bluefs_extents(out_of_sync_fm); | |
8316 | } | |
8317 | } | |
8318 | ||
7c673cae | 8319 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { |
9f95a23c | 8320 | apply_for_bitset_range( |
b32b8144 | 8321 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae FG |
8322 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
8323 | bs.set(pos); | |
1911f103 | 8324 | }); |
7c673cae | 8325 | } |
eafe8130 | 8326 | int r = bluefs->fsck(); |
7c673cae | 8327 | if (r < 0) { |
eafe8130 | 8328 | return r; |
7c673cae FG |
8329 | } |
8330 | if (r > 0) | |
8331 | errors += r; | |
8332 | } | |
8333 | ||
eafe8130 TL |
8334 | if (!per_pool_stat_collection) { |
8335 | const char *w; | |
8336 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_stats) { | |
8337 | w = "error"; | |
8338 | ++errors; | |
8339 | } else { | |
8340 | w = "warning"; | |
8341 | ++warnings; | |
8342 | } | |
8343 | derr << "fsck " << w << ": store not yet converted to per-pool stats" | |
8344 | << dendl; | |
8345 | } | |
9f95a23c TL |
8346 | if (!per_pool_omap) { |
8347 | const char *w; | |
8348 | if (cct->_conf->bluestore_fsck_error_on_no_per_pool_omap) { | |
8349 | w = "error"; | |
8350 | ++errors; | |
8351 | } else { | |
8352 | w = "warning"; | |
8353 | ++warnings; | |
8354 | } | |
8355 | derr << "fsck " << w << ": store not yet converted to per-pool omap" | |
8356 | << dendl; | |
8357 | } | |
8358 | ||
11fdf7f2 | 8359 | // get expected statfs; reset unaffected fields to be able to compare |
7c673cae FG |
8360 | // structs |
8361 | statfs(&actual_statfs); | |
11fdf7f2 TL |
8362 | actual_statfs.total = 0; |
8363 | actual_statfs.internally_reserved = 0; | |
8364 | actual_statfs.available = 0; | |
8365 | actual_statfs.internal_metadata = 0; | |
8366 | actual_statfs.omap_allocated = 0; | |
8367 | ||
eafe8130 TL |
8368 | if (g_conf()->bluestore_debug_fsck_abort) { |
8369 | dout(1) << __func__ << " debug abort" << dendl; | |
8370 | goto out_scan; | |
8371 | } | |
7c673cae | 8372 | // walk PREFIX_OBJ |
eafe8130 TL |
8373 | { |
8374 | dout(1) << __func__ << " walking object keyspace" << dendl; | |
8375 | ceph::mutex sb_info_lock = ceph::make_mutex("BlueStore::fsck::sbinfo_lock"); | |
8376 | BlueStore::FSCK_ObjectCtx ctx( | |
8377 | errors, | |
8378 | warnings, | |
8379 | num_objects, | |
8380 | num_extents, | |
8381 | num_blobs, | |
8382 | num_sharded_objects, | |
8383 | num_spanning_blobs, | |
8384 | &used_blocks, | |
8385 | &used_omap_head, | |
9f95a23c TL |
8386 | //no need for the below lock when in non-shallow mode as |
8387 | // there is no multithreading in this case | |
8388 | depth == FSCK_SHALLOW ? &sb_info_lock : nullptr, | |
eafe8130 TL |
8389 | sb_info, |
8390 | expected_store_statfs, | |
8391 | expected_pool_statfs, | |
8392 | repair ? &repairer : nullptr); | |
9f95a23c TL |
8393 | |
8394 | _fsck_check_objects(depth, ctx); | |
eafe8130 | 8395 | } |
11fdf7f2 | 8396 | |
7c673cae FG |
8397 | dout(1) << __func__ << " checking shared_blobs" << dendl; |
8398 | it = db->get_iterator(PREFIX_SHARED_BLOB); | |
8399 | if (it) { | |
eafe8130 TL |
8400 | // FIXME minor: perhaps simplify for shallow mode? |
8401 | // fill global if not overriden below | |
8402 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 | 8403 | |
7c673cae FG |
8404 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8405 | string key = it->key(); | |
8406 | uint64_t sbid; | |
8407 | if (get_key_shared_blob(key, &sbid)) { | |
3efd9988 | 8408 | derr << "fsck error: bad key '" << key |
7c673cae | 8409 | << "' in shared blob namespace" << dendl; |
11fdf7f2 TL |
8410 | if (repair) { |
8411 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8412 | } | |
7c673cae FG |
8413 | ++errors; |
8414 | continue; | |
8415 | } | |
8416 | auto p = sb_info.find(sbid); | |
8417 | if (p == sb_info.end()) { | |
3efd9988 | 8418 | derr << "fsck error: found stray shared blob data for sbid 0x" |
7c673cae | 8419 | << std::hex << sbid << std::dec << dendl; |
11fdf7f2 TL |
8420 | if (repair) { |
8421 | repairer.remove_key(db, PREFIX_SHARED_BLOB, key); | |
8422 | } | |
7c673cae FG |
8423 | ++errors; |
8424 | } else { | |
8425 | ++num_shared_blobs; | |
8426 | sb_info_t& sbi = p->second; | |
8427 | bluestore_shared_blob_t shared_blob(sbid); | |
8428 | bufferlist bl = it->value(); | |
11fdf7f2 TL |
8429 | auto blp = bl.cbegin(); |
8430 | try { | |
8431 | decode(shared_blob, blp); | |
8432 | } catch (buffer::error& e) { | |
8433 | ++errors; | |
8434 | // Force update and don't report as missing | |
8435 | sbi.updated = sbi.passed = true; | |
8436 | ||
8437 | derr << "fsck error: failed to decode Shared Blob" | |
8438 | << pretty_binary_string(it->key()) << dendl; | |
8439 | if (repair) { | |
8440 | dout(20) << __func__ << " undecodable Shared Blob, key:'" | |
8441 | << pretty_binary_string(it->key()) | |
8442 | << "', removing" << dendl; | |
8443 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8444 | } | |
8445 | continue; | |
8446 | } | |
7c673cae FG |
8447 | dout(20) << __func__ << " " << *sbi.sb << " " << shared_blob << dendl; |
8448 | if (shared_blob.ref_map != sbi.ref_map) { | |
3efd9988 | 8449 | derr << "fsck error: shared blob 0x" << std::hex << sbid |
11fdf7f2 TL |
8450 | << std::dec << " ref_map " << shared_blob.ref_map |
8451 | << " != expected " << sbi.ref_map << dendl; | |
8452 | sbi.updated = true; // will update later in repair mode only! | |
7c673cae FG |
8453 | ++errors; |
8454 | } | |
8455 | PExtentVector extents; | |
8456 | for (auto &r : shared_blob.ref_map.ref_map) { | |
8457 | extents.emplace_back(bluestore_pextent_t(r.first, r.second.length)); | |
8458 | } | |
eafe8130 | 8459 | if (per_pool_stat_collection || repair) { |
11fdf7f2 TL |
8460 | expected_statfs = &expected_pool_statfs[sbi.pool_id]; |
8461 | } | |
8462 | errors += _fsck_check_extents(sbi.cid, | |
8463 | p->second.oids.front(), | |
7c673cae FG |
8464 | extents, |
8465 | p->second.compressed, | |
b32b8144 FG |
8466 | used_blocks, |
8467 | fm->get_alloc_size(), | |
11fdf7f2 | 8468 | repair ? &repairer : nullptr, |
eafe8130 TL |
8469 | *expected_statfs, |
8470 | depth); | |
11fdf7f2 TL |
8471 | sbi.passed = true; |
8472 | } | |
8473 | } | |
8474 | } // if (it) | |
8475 | ||
8476 | if (repair && repairer.preprocess_misreference(db)) { | |
8477 | ||
8478 | dout(1) << __func__ << " sorting out misreferenced extents" << dendl; | |
8479 | auto& space_tracker = repairer.get_space_usage_tracker(); | |
8480 | auto& misref_extents = repairer.get_misreferences(); | |
8481 | interval_set<uint64_t> to_release; | |
8482 | it = db->get_iterator(PREFIX_OBJ); | |
8483 | if (it) { | |
eafe8130 TL |
8484 | // fill global if not overriden below |
8485 | auto expected_statfs = &expected_store_statfs; | |
11fdf7f2 TL |
8486 | |
8487 | CollectionRef c; | |
8488 | spg_t pgid; | |
8489 | KeyValueDB::Transaction txn = repairer.get_fix_misreferences_txn(); | |
8490 | bool bypass_rest = false; | |
8491 | for (it->lower_bound(string()); it->valid() && !bypass_rest; | |
8492 | it->next()) { | |
8493 | dout(30) << __func__ << " key " | |
8494 | << pretty_binary_string(it->key()) << dendl; | |
8495 | if (is_extent_shard_key(it->key())) { | |
8496 | continue; | |
8497 | } | |
8498 | ||
8499 | ghobject_t oid; | |
8500 | int r = get_key_object(it->key(), &oid); | |
8501 | if (r < 0 || !space_tracker.is_used(oid)) { | |
8502 | continue; | |
8503 | } | |
8504 | ||
8505 | if (!c || | |
8506 | oid.shard_id != pgid.shard || | |
8507 | oid.hobj.get_logical_pool() != (int64_t)pgid.pool() || | |
8508 | !c->contains(oid)) { | |
8509 | c = nullptr; | |
8510 | for (auto& p : coll_map) { | |
8511 | if (p.second->contains(oid)) { | |
8512 | c = p.second; | |
8513 | break; | |
8514 | } | |
8515 | } | |
8516 | if (!c) { | |
8517 | continue; | |
8518 | } | |
eafe8130 TL |
8519 | if (per_pool_stat_collection || repair) { |
8520 | auto pool_id = c->cid.is_pg(&pgid) ? pgid.pool() : META_POOL_ID; | |
11fdf7f2 TL |
8521 | expected_statfs = &expected_pool_statfs[pool_id]; |
8522 | } | |
8523 | } | |
8524 | if (!space_tracker.is_used(c->cid)) { | |
8525 | continue; | |
8526 | } | |
8527 | ||
8528 | dout(20) << __func__ << " check misreference for col:" << c->cid | |
8529 | << " obj:" << oid << dendl; | |
8530 | ||
eafe8130 TL |
8531 | OnodeRef o; |
8532 | o.reset(Onode::decode(c, oid, it->key(), it->value())); | |
11fdf7f2 TL |
8533 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); |
8534 | mempool::bluestore_fsck::set<BlobRef> blobs; | |
8535 | ||
8536 | for (auto& e : o->extent_map.extent_map) { | |
8537 | blobs.insert(e.blob); | |
8538 | } | |
8539 | bool need_onode_update = false; | |
8540 | bool first_dump = true; | |
8541 | for(auto b : blobs) { | |
8542 | bool broken_blob = false; | |
8543 | auto& pextents = b->dirty_blob().dirty_extents(); | |
8544 | for (auto& e : pextents) { | |
8545 | if (!e.is_valid()) { | |
8546 | continue; | |
8547 | } | |
8548 | // for the sake of simplicity and proper shared blob handling | |
8549 | // always rewrite the whole blob even when it's partially | |
8550 | // misreferenced. | |
8551 | if (misref_extents.intersects(e.offset, e.length)) { | |
8552 | if (first_dump) { | |
8553 | first_dump = false; | |
81eedcae | 8554 | _dump_onode<10>(cct, *o); |
11fdf7f2 TL |
8555 | } |
8556 | broken_blob = true; | |
8557 | break; | |
8558 | } | |
8559 | } | |
8560 | if (!broken_blob) | |
8561 | continue; | |
8562 | bool compressed = b->get_blob().is_compressed(); | |
8563 | need_onode_update = true; | |
8564 | dout(10) << __func__ | |
8565 | << " fix misreferences in oid:" << oid | |
8566 | << " " << *b << dendl; | |
8567 | uint64_t b_off = 0; | |
8568 | PExtentVector pext_to_release; | |
8569 | pext_to_release.reserve(pextents.size()); | |
8570 | // rewriting all valid pextents | |
8571 | for (auto e = pextents.begin(); e != pextents.end(); | |
8572 | b_off += e->length, e++) { | |
8573 | if (!e->is_valid()) { | |
8574 | continue; | |
8575 | } | |
8576 | PExtentVector exts; | |
8577 | int64_t alloc_len = alloc->allocate(e->length, min_alloc_size, | |
8578 | 0, 0, &exts); | |
eafe8130 | 8579 | if (alloc_len < 0 || alloc_len < (int64_t)e->length) { |
11fdf7f2 TL |
8580 | derr << __func__ |
8581 | << " failed to allocate 0x" << std::hex << e->length | |
eafe8130 | 8582 | << " allocated 0x " << (alloc_len < 0 ? 0 : alloc_len) |
11fdf7f2 TL |
8583 | << " min_alloc_size 0x" << min_alloc_size |
8584 | << " available 0x " << alloc->get_free() | |
8585 | << std::dec << dendl; | |
8586 | if (alloc_len > 0) { | |
8587 | alloc->release(exts); | |
8588 | } | |
8589 | bypass_rest = true; | |
8590 | break; | |
8591 | } | |
8592 | expected_statfs->allocated += e->length; | |
8593 | if (compressed) { | |
8594 | expected_statfs->data_compressed_allocated += e->length; | |
8595 | } | |
8596 | ||
8597 | bufferlist bl; | |
8598 | IOContext ioc(cct, NULL, true); // allow EIO | |
8599 | r = bdev->read(e->offset, e->length, &bl, &ioc, false); | |
8600 | if (r < 0) { | |
8601 | derr << __func__ << " failed to read from 0x" << std::hex << e->offset | |
8602 | <<"~" << e->length << std::dec << dendl; | |
8603 | ceph_abort_msg("read failed, wtf"); | |
8604 | } | |
8605 | pext_to_release.push_back(*e); | |
8606 | e = pextents.erase(e); | |
8607 | e = pextents.insert(e, exts.begin(), exts.end()); | |
8608 | b->get_blob().map_bl( | |
8609 | b_off, bl, | |
8610 | [&](uint64_t offset, bufferlist& t) { | |
8611 | int r = bdev->write(offset, t, false); | |
8612 | ceph_assert(r == 0); | |
8613 | }); | |
8614 | e += exts.size() - 1; | |
8615 | for (auto& p : exts) { | |
8616 | fm->allocate(p.offset, p.length, txn); | |
8617 | } | |
8618 | } // for (auto e = pextents.begin(); e != pextents.end(); e++) { | |
8619 | ||
8620 | if (b->get_blob().is_shared()) { | |
8621 | b->dirty_blob().clear_flag(bluestore_blob_t::FLAG_SHARED); | |
8622 | ||
8623 | auto sb_it = sb_info.find(b->shared_blob->get_sbid()); | |
8624 | ceph_assert(sb_it != sb_info.end()); | |
8625 | sb_info_t& sbi = sb_it->second; | |
8626 | ||
8627 | for (auto& r : sbi.ref_map.ref_map) { | |
8628 | expected_statfs->allocated -= r.second.length; | |
8629 | if (sbi.compressed) { | |
8630 | // NB: it's crucial to use compressed flag from sb_info_t | |
8631 | // as we originally used that value while accumulating | |
8632 | // expected_statfs | |
8633 | expected_statfs->data_compressed_allocated -= r.second.length; | |
8634 | } | |
8635 | } | |
8636 | sbi.updated = sbi.passed = true; | |
8637 | sbi.ref_map.clear(); | |
8638 | ||
8639 | // relying on blob's pextents to decide what to release. | |
8640 | for (auto& p : pext_to_release) { | |
8641 | to_release.union_insert(p.offset, p.length); | |
8642 | } | |
8643 | } else { | |
8644 | for (auto& p : pext_to_release) { | |
8645 | expected_statfs->allocated -= p.length; | |
8646 | if (compressed) { | |
8647 | expected_statfs->data_compressed_allocated -= p.length; | |
8648 | } | |
8649 | to_release.union_insert(p.offset, p.length); | |
8650 | } | |
8651 | } | |
8652 | if (bypass_rest) { | |
8653 | break; | |
8654 | } | |
8655 | } // for(auto b : blobs) | |
8656 | if (need_onode_update) { | |
8657 | o->extent_map.dirty_range(0, OBJECT_MAX_SIZE); | |
8658 | _record_onode(o, txn); | |
8659 | } | |
8660 | } // for (it->lower_bound(string()); it->valid(); it->next()) | |
8661 | ||
8662 | for (auto it = to_release.begin(); it != to_release.end(); ++it) { | |
8663 | dout(10) << __func__ << " release 0x" << std::hex << it.get_start() | |
8664 | << "~" << it.get_len() << std::dec << dendl; | |
8665 | fm->release(it.get_start(), it.get_len(), txn); | |
8666 | } | |
8667 | alloc->release(to_release); | |
8668 | to_release.clear(); | |
8669 | } // if (it) { | |
8670 | } //if (repair && repairer.preprocess_misreference()) { | |
8671 | ||
eafe8130 TL |
8672 | if (depth != FSCK_SHALLOW) { |
8673 | for (auto &p : sb_info) { | |
8674 | sb_info_t& sbi = p.second; | |
8675 | if (!sbi.passed) { | |
8676 | derr << "fsck error: missing " << *sbi.sb << dendl; | |
8677 | ++errors; | |
8678 | } | |
8679 | if (repair && (!sbi.passed || sbi.updated)) { | |
8680 | auto sbid = p.first; | |
8681 | if (sbi.ref_map.empty()) { | |
8682 | ceph_assert(sbi.passed); | |
8683 | dout(20) << __func__ << " " << *sbi.sb | |
8684 | << " is empty, removing" << dendl; | |
8685 | repairer.fix_shared_blob(db, sbid, nullptr); | |
8686 | } else { | |
8687 | bufferlist bl; | |
8688 | bluestore_shared_blob_t persistent(sbid, std::move(sbi.ref_map)); | |
8689 | encode(persistent, bl); | |
8690 | dout(20) << __func__ << " " << *sbi.sb | |
8691 | << " is " << bl.length() << " bytes, updating" << dendl; | |
11fdf7f2 | 8692 | |
eafe8130 TL |
8693 | repairer.fix_shared_blob(db, sbid, &bl); |
8694 | } | |
7c673cae FG |
8695 | } |
8696 | } | |
8697 | } | |
11fdf7f2 TL |
8698 | sb_info.clear(); |
8699 | ||
eafe8130 TL |
8700 | // check global stats only if fscking (not repairing) w/o per-pool stats |
8701 | if (!per_pool_stat_collection && | |
8702 | !repair && | |
8703 | !(actual_statfs == expected_store_statfs)) { | |
8704 | derr << "fsck error: actual " << actual_statfs | |
8705 | << " != expected " << expected_store_statfs << dendl; | |
8706 | if (repair) { | |
8707 | repairer.fix_statfs(db, BLUESTORE_GLOBAL_STATFS_KEY, | |
8708 | expected_store_statfs); | |
11fdf7f2 | 8709 | } |
eafe8130 | 8710 | ++errors; |
7c673cae FG |
8711 | } |
8712 | ||
eafe8130 TL |
8713 | dout(1) << __func__ << " checking pool_statfs" << dendl; |
8714 | _fsck_check_pool_statfs(expected_pool_statfs, | |
8715 | errors, warnings, repair ? &repairer : nullptr); | |
8716 | ||
8717 | if (depth != FSCK_SHALLOW) { | |
9f95a23c | 8718 | dout(1) << __func__ << " checking for stray omap data " << dendl; |
eafe8130 TL |
8719 | it = db->get_iterator(PREFIX_OMAP); |
8720 | if (it) { | |
9f95a23c | 8721 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8722 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8723 | uint64_t omap_head; | |
8724 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8725 | if (used_omap_head.count(omap_head) == 0 && |
8726 | omap_head != last_omap_head) { | |
8727 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8728 | << "fsck error: found stray omap data on omap_head " | |
8729 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head)<< fsck_dendl; | |
eafe8130 | 8730 | ++errors; |
9f95a23c | 8731 | last_omap_head = omap_head; |
eafe8130 | 8732 | } |
7c673cae FG |
8733 | } |
8734 | } | |
eafe8130 TL |
8735 | it = db->get_iterator(PREFIX_PGMETA_OMAP); |
8736 | if (it) { | |
9f95a23c | 8737 | uint64_t last_omap_head = 0; |
eafe8130 TL |
8738 | for (it->lower_bound(string()); it->valid(); it->next()) { |
8739 | uint64_t omap_head; | |
8740 | _key_decode_u64(it->key().c_str(), &omap_head); | |
9f95a23c TL |
8741 | if (used_omap_head.count(omap_head) == 0 && |
8742 | omap_head != last_omap_head) { | |
8743 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8744 | << "fsck error: found stray (pgmeta) omap data on omap_head " | |
8745 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8746 | last_omap_head = omap_head; | |
eafe8130 TL |
8747 | ++errors; |
8748 | } | |
11fdf7f2 TL |
8749 | } |
8750 | } | |
9f95a23c TL |
8751 | it = db->get_iterator(PREFIX_PERPOOL_OMAP); |
8752 | if (it) { | |
8753 | uint64_t last_omap_head = 0; | |
8754 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8755 | uint64_t pool; | |
8756 | uint64_t omap_head; | |
8757 | string k = it->key(); | |
8758 | const char *c = k.c_str(); | |
8759 | c = _key_decode_u64(c, &pool); | |
8760 | c = _key_decode_u64(c, &omap_head); | |
8761 | if (used_omap_head.count(omap_head) == 0 && | |
8762 | omap_head != last_omap_head) { | |
8763 | fsck_derr(errors, MAX_FSCK_ERROR_LINES) | |
8764 | << "fsck error: found stray (per-pool) omap data on omap_head " | |
8765 | << omap_head << " " << last_omap_head << " " << used_omap_head.count(omap_head) << fsck_dendl; | |
8766 | ++errors; | |
8767 | last_omap_head = omap_head; | |
8768 | } | |
8769 | } | |
8770 | } | |
eafe8130 TL |
8771 | dout(1) << __func__ << " checking deferred events" << dendl; |
8772 | it = db->get_iterator(PREFIX_DEFERRED); | |
8773 | if (it) { | |
8774 | for (it->lower_bound(string()); it->valid(); it->next()) { | |
8775 | bufferlist bl = it->value(); | |
8776 | auto p = bl.cbegin(); | |
8777 | bluestore_deferred_transaction_t wt; | |
8778 | try { | |
8779 | decode(wt, p); | |
8780 | } catch (buffer::error& e) { | |
8781 | derr << "fsck error: failed to decode deferred txn " | |
8782 | << pretty_binary_string(it->key()) << dendl; | |
8783 | if (repair) { | |
8784 | dout(20) << __func__ << " undecodable deferred TXN record, key: '" | |
8785 | << pretty_binary_string(it->key()) | |
8786 | << "', removing" << dendl; | |
8787 | repairer.remove_key(db, PREFIX_DEFERRED, it->key()); | |
8788 | } | |
8789 | continue; | |
8790 | } | |
8791 | dout(20) << __func__ << " deferred " << wt.seq | |
8792 | << " ops " << wt.ops.size() | |
8793 | << " released 0x" << std::hex << wt.released << std::dec << dendl; | |
8794 | for (auto e = wt.released.begin(); e != wt.released.end(); ++e) { | |
9f95a23c | 8795 | apply_for_bitset_range( |
eafe8130 TL |
8796 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
8797 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
8798 | bs.set(pos); |
8799 | } | |
8800 | ); | |
8801 | } | |
7c673cae | 8802 | } |
eafe8130 TL |
8803 | } |
8804 | ||
8805 | dout(1) << __func__ << " checking freelist vs allocated" << dendl; | |
8806 | { | |
8807 | // remove bluefs_extents from used set since the freelist doesn't | |
8808 | // know they are allocated. | |
8809 | for (auto e = bluefs_extents.begin(); e != bluefs_extents.end(); ++e) { | |
9f95a23c | 8810 | apply_for_bitset_range( |
b32b8144 | 8811 | e.get_start(), e.get_len(), fm->get_alloc_size(), used_blocks, |
7c673cae | 8812 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { |
eafe8130 | 8813 | bs.reset(pos); |
7c673cae FG |
8814 | } |
8815 | ); | |
8816 | } | |
eafe8130 TL |
8817 | fm->enumerate_reset(); |
8818 | uint64_t offset, length; | |
8819 | while (fm->enumerate_next(db, &offset, &length)) { | |
8820 | bool intersects = false; | |
9f95a23c | 8821 | apply_for_bitset_range( |
eafe8130 TL |
8822 | offset, length, fm->get_alloc_size(), used_blocks, |
8823 | [&](uint64_t pos, mempool_dynamic_bitset &bs) { | |
eafe8130 TL |
8824 | if (bs.test(pos)) { |
8825 | if (offset == SUPER_RESERVED && | |
8826 | length == min_alloc_size - SUPER_RESERVED) { | |
8827 | // this is due to the change just after luminous to min_alloc_size | |
8828 | // granularity allocations, and our baked in assumption at the top | |
8829 | // of _fsck that 0~round_up_to(SUPER_RESERVED,min_alloc_size) is used | |
8830 | // (vs luminous's round_up_to(SUPER_RESERVED,block_size)). harmless, | |
8831 | // since we will never allocate this region below min_alloc_size. | |
8832 | dout(10) << __func__ << " ignoring free extent between SUPER_RESERVED" | |
8833 | << " and min_alloc_size, 0x" << std::hex << offset << "~" | |
8834 | << length << std::dec << dendl; | |
8835 | } else { | |
8836 | intersects = true; | |
8837 | if (repair) { | |
8838 | repairer.fix_false_free(db, fm, | |
8839 | pos * min_alloc_size, | |
8840 | min_alloc_size); | |
8841 | } | |
11fdf7f2 | 8842 | } |
eafe8130 TL |
8843 | } else { |
8844 | bs.set(pos); | |
8845 | } | |
7c673cae | 8846 | } |
eafe8130 TL |
8847 | ); |
8848 | if (intersects) { | |
8849 | derr << "fsck error: free extent 0x" << std::hex << offset | |
8850 | << "~" << length << std::dec | |
8851 | << " intersects allocated blocks" << dendl; | |
8852 | ++errors; | |
7c673cae | 8853 | } |
b5b8bbf5 | 8854 | } |
eafe8130 TL |
8855 | fm->enumerate_reset(); |
8856 | size_t count = used_blocks.count(); | |
8857 | if (used_blocks.size() != count) { | |
8858 | ceph_assert(used_blocks.size() > count); | |
8859 | used_blocks.flip(); | |
8860 | size_t start = used_blocks.find_first(); | |
8861 | while (start != decltype(used_blocks)::npos) { | |
8862 | size_t cur = start; | |
8863 | while (true) { | |
8864 | size_t next = used_blocks.find_next(cur); | |
8865 | if (next != cur + 1) { | |
8866 | ++errors; | |
8867 | derr << "fsck error: leaked extent 0x" << std::hex | |
8868 | << ((uint64_t)start * fm->get_alloc_size()) << "~" | |
8869 | << ((cur + 1 - start) * fm->get_alloc_size()) << std::dec | |
8870 | << dendl; | |
8871 | if (repair) { | |
8872 | repairer.fix_leaked(db, | |
8873 | fm, | |
8874 | start * min_alloc_size, | |
8875 | (cur + 1 - start) * min_alloc_size); | |
8876 | } | |
8877 | start = next; | |
8878 | break; | |
11fdf7f2 | 8879 | } |
eafe8130 | 8880 | cur = next; |
b5b8bbf5 | 8881 | } |
eafe8130 TL |
8882 | } |
8883 | used_blocks.flip(); | |
b5b8bbf5 | 8884 | } |
7c673cae FG |
8885 | } |
8886 | } | |
11fdf7f2 | 8887 | if (repair) { |
9f95a23c TL |
8888 | if (!per_pool_omap) { |
8889 | dout(5) << __func__ << " marking per_pool_omap=1" << dendl; | |
8890 | repairer.fix_per_pool_omap(db); | |
8891 | } | |
8892 | ||
11fdf7f2 TL |
8893 | dout(5) << __func__ << " applying repair results" << dendl; |
8894 | repaired = repairer.apply(db); | |
8895 | dout(5) << __func__ << " repair applied" << dendl; | |
8896 | } | |
7c673cae | 8897 | |
eafe8130 | 8898 | out_scan: |
7c673cae FG |
8899 | dout(2) << __func__ << " " << num_objects << " objects, " |
8900 | << num_sharded_objects << " of them sharded. " | |
8901 | << dendl; | |
8902 | dout(2) << __func__ << " " << num_extents << " extents to " | |
8903 | << num_blobs << " blobs, " | |
8904 | << num_spanning_blobs << " spanning, " | |
8905 | << num_shared_blobs << " shared." | |
8906 | << dendl; | |
8907 | ||
8908 | utime_t duration = ceph_clock_now() - start; | |
9f95a23c TL |
8909 | dout(1) << __func__ << " <<<FINISH>>> with " << errors << " errors, " |
8910 | << warnings << " warnings, " | |
8911 | << repaired << " repaired, " | |
8912 | << (errors + warnings - (int)repaired) << " remaining in " | |
7c673cae | 8913 | << duration << " seconds" << dendl; |
9f95a23c TL |
8914 | |
8915 | // In non-repair mode we should return error count only as | |
8916 | // it indicates if store status is OK. | |
8917 | // In repair mode both errors and warnings are taken into account | |
8918 | // since repaired counter relates to them both. | |
8919 | return repair ? errors + warnings - (int)repaired : errors; | |
11fdf7f2 TL |
8920 | } |
8921 | ||
8922 | /// methods to inject various errors fsck can repair | |
8923 | void BlueStore::inject_broken_shared_blob_key(const string& key, | |
8924 | const bufferlist& bl) | |
8925 | { | |
8926 | KeyValueDB::Transaction txn; | |
8927 | txn = db->get_transaction(); | |
8928 | txn->set(PREFIX_SHARED_BLOB, key, bl); | |
8929 | db->submit_transaction_sync(txn); | |
8930 | }; | |
8931 | ||
8932 | void BlueStore::inject_leaked(uint64_t len) | |
8933 | { | |
8934 | KeyValueDB::Transaction txn; | |
8935 | txn = db->get_transaction(); | |
8936 | ||
8937 | PExtentVector exts; | |
8938 | int64_t alloc_len = alloc->allocate(len, min_alloc_size, | |
8939 | min_alloc_size * 256, 0, &exts); | |
8940 | ceph_assert(alloc_len >= (int64_t)len); | |
8941 | for (auto& p : exts) { | |
8942 | fm->allocate(p.offset, p.length, txn); | |
8943 | } | |
8944 | db->submit_transaction_sync(txn); | |
8945 | } | |
8946 | ||
8947 | void BlueStore::inject_false_free(coll_t cid, ghobject_t oid) | |
8948 | { | |
8949 | KeyValueDB::Transaction txn; | |
8950 | OnodeRef o; | |
8951 | CollectionRef c = _get_collection(cid); | |
8952 | ceph_assert(c); | |
8953 | { | |
9f95a23c | 8954 | std::unique_lock l{c->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
8955 | o = c->get_onode(oid, false); |
8956 | ceph_assert(o); | |
8957 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
8958 | } | |
8959 | ||
8960 | bool injected = false; | |
8961 | txn = db->get_transaction(); | |
8962 | auto& em = o->extent_map.extent_map; | |
8963 | std::vector<const PExtentVector*> v; | |
8964 | if (em.size()) { | |
8965 | v.push_back(&em.begin()->blob->get_blob().get_extents()); | |
8966 | } | |
8967 | if (em.size() > 1) { | |
8968 | auto it = em.end(); | |
8969 | --it; | |
8970 | v.push_back(&(it->blob->get_blob().get_extents())); | |
8971 | } | |
8972 | for (auto pext : v) { | |
8973 | if (pext->size()) { | |
8974 | auto p = pext->begin(); | |
8975 | while (p != pext->end()) { | |
8976 | if (p->is_valid()) { | |
8977 | dout(20) << __func__ << " release 0x" << std::hex << p->offset | |
8978 | << "~" << p->length << std::dec << dendl; | |
8979 | fm->release(p->offset, p->length, txn); | |
8980 | injected = true; | |
8981 | break; | |
8982 | } | |
8983 | ++p; | |
8984 | } | |
8985 | } | |
8986 | } | |
8987 | ceph_assert(injected); | |
8988 | db->submit_transaction_sync(txn); | |
8989 | } | |
8990 | ||
9f95a23c TL |
8991 | void BlueStore::inject_legacy_omap() |
8992 | { | |
8993 | dout(1) << __func__ << dendl; | |
8994 | per_pool_omap = false; | |
8995 | KeyValueDB::Transaction txn; | |
8996 | txn = db->get_transaction(); | |
8997 | txn->rmkey(PREFIX_SUPER, "per_pool_omap"); | |
8998 | db->submit_transaction_sync(txn); | |
8999 | } | |
9000 | ||
9001 | void BlueStore::inject_legacy_omap(coll_t cid, ghobject_t oid) | |
9002 | { | |
9003 | dout(1) << __func__ << " " | |
9004 | << cid << " " << oid | |
9005 | <<dendl; | |
9006 | KeyValueDB::Transaction txn; | |
9007 | OnodeRef o; | |
9008 | CollectionRef c = _get_collection(cid); | |
9009 | ceph_assert(c); | |
9010 | { | |
9011 | std::unique_lock l{ c->lock }; // just to avoid internal asserts | |
9012 | o = c->get_onode(oid, false); | |
9013 | ceph_assert(o); | |
9014 | } | |
9015 | o->onode.clear_flag(bluestore_onode_t::FLAG_PERPOOL_OMAP | bluestore_onode_t::FLAG_PGMETA_OMAP); | |
9016 | txn = db->get_transaction(); | |
9017 | _record_onode(o, txn); | |
9018 | db->submit_transaction_sync(txn); | |
9019 | } | |
9020 | ||
9021 | ||
11fdf7f2 TL |
9022 | void BlueStore::inject_statfs(const string& key, const store_statfs_t& new_statfs) |
9023 | { | |
9024 | BlueStoreRepairer repairer; | |
9025 | repairer.fix_statfs(db, key, new_statfs); | |
9026 | repairer.apply(db); | |
9027 | } | |
9028 | ||
eafe8130 TL |
9029 | void BlueStore::inject_global_statfs(const store_statfs_t& new_statfs) |
9030 | { | |
9031 | KeyValueDB::Transaction t = db->get_transaction(); | |
9032 | volatile_statfs v; | |
9033 | v = new_statfs; | |
9034 | bufferlist bl; | |
9035 | v.encode(bl); | |
9036 | t->set(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
9037 | db->submit_transaction_sync(t); | |
9038 | } | |
9039 | ||
11fdf7f2 TL |
9040 | void BlueStore::inject_misreference(coll_t cid1, ghobject_t oid1, |
9041 | coll_t cid2, ghobject_t oid2, | |
9042 | uint64_t offset) | |
9043 | { | |
9044 | OnodeRef o1; | |
9045 | CollectionRef c1 = _get_collection(cid1); | |
9046 | ceph_assert(c1); | |
9047 | { | |
9f95a23c | 9048 | std::unique_lock l{c1->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9049 | o1 = c1->get_onode(oid1, false); |
9050 | ceph_assert(o1); | |
9051 | o1->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9052 | } | |
9053 | OnodeRef o2; | |
9054 | CollectionRef c2 = _get_collection(cid2); | |
9055 | ceph_assert(c2); | |
9056 | { | |
9f95a23c | 9057 | std::unique_lock l{c2->lock}; // just to avoid internal asserts |
11fdf7f2 TL |
9058 | o2 = c2->get_onode(oid2, false); |
9059 | ceph_assert(o2); | |
9060 | o2->extent_map.fault_range(db, offset, OBJECT_MAX_SIZE); | |
9061 | } | |
9062 | Extent& e1 = *(o1->extent_map.seek_lextent(offset)); | |
9063 | Extent& e2 = *(o2->extent_map.seek_lextent(offset)); | |
9064 | ||
9065 | // require onode/extent layout to be the same (and simple) | |
9066 | // to make things easier | |
9067 | ceph_assert(o1->onode.extent_map_shards.empty()); | |
9068 | ceph_assert(o2->onode.extent_map_shards.empty()); | |
9069 | ceph_assert(o1->extent_map.spanning_blob_map.size() == 0); | |
9070 | ceph_assert(o2->extent_map.spanning_blob_map.size() == 0); | |
9071 | ceph_assert(e1.logical_offset == e2.logical_offset); | |
9072 | ceph_assert(e1.length == e2.length); | |
9073 | ceph_assert(e1.blob_offset == e2.blob_offset); | |
9074 | ||
9075 | KeyValueDB::Transaction txn; | |
9076 | txn = db->get_transaction(); | |
9077 | ||
9078 | // along with misreference error this will create space leaks errors | |
9079 | e2.blob->dirty_blob() = e1.blob->get_blob(); | |
9080 | o2->extent_map.dirty_range(offset, e2.length); | |
9081 | o2->extent_map.update(txn, false); | |
9082 | ||
9083 | _record_onode(o2, txn); | |
9084 | db->submit_transaction_sync(txn); | |
7c673cae FG |
9085 | } |
9086 | ||
9087 | void BlueStore::collect_metadata(map<string,string> *pm) | |
9088 | { | |
9089 | dout(10) << __func__ << dendl; | |
9090 | bdev->collect_metadata("bluestore_bdev_", pm); | |
9091 | if (bluefs) { | |
9092 | (*pm)["bluefs"] = "1"; | |
9f95a23c TL |
9093 | // this value is for backward compatibility only |
9094 | (*pm)["bluefs_single_shared_device"] = \ | |
9095 | stringify((int)bluefs_layout.single_shared_device()); | |
9096 | (*pm)["bluefs_dedicated_db"] = \ | |
9097 | stringify((int)bluefs_layout.dedicated_db); | |
9098 | (*pm)["bluefs_dedicated_wal"] = \ | |
9099 | stringify((int)bluefs_layout.dedicated_wal); | |
9100 | bluefs->collect_metadata(pm, bluefs_layout.shared_bdev); | |
7c673cae FG |
9101 | } else { |
9102 | (*pm)["bluefs"] = "0"; | |
9103 | } | |
11fdf7f2 TL |
9104 | |
9105 | // report numa mapping for underlying devices | |
9106 | int node = -1; | |
9107 | set<int> nodes; | |
9108 | set<string> failed; | |
9109 | int r = get_numa_node(&node, &nodes, &failed); | |
9110 | if (r >= 0) { | |
9111 | if (!failed.empty()) { | |
9112 | (*pm)["objectstore_numa_unknown_devices"] = stringify(failed); | |
9113 | } | |
9114 | if (!nodes.empty()) { | |
9115 | dout(1) << __func__ << " devices span numa nodes " << nodes << dendl; | |
9116 | (*pm)["objectstore_numa_nodes"] = stringify(nodes); | |
9117 | } | |
9118 | if (node >= 0) { | |
9119 | (*pm)["objectstore_numa_node"] = stringify(node); | |
9120 | } | |
9121 | } | |
9122 | } | |
9123 | ||
9124 | int BlueStore::get_numa_node( | |
9125 | int *final_node, | |
9126 | set<int> *out_nodes, | |
9127 | set<string> *out_failed) | |
9128 | { | |
9129 | int node = -1; | |
9130 | set<string> devices; | |
9131 | get_devices(&devices); | |
9132 | set<int> nodes; | |
9133 | set<string> failed; | |
9134 | for (auto& devname : devices) { | |
9135 | int n; | |
9136 | BlkDev bdev(devname); | |
9137 | int r = bdev.get_numa_node(&n); | |
9138 | if (r < 0) { | |
9139 | dout(10) << __func__ << " bdev " << devname << " can't detect numa_node" | |
9140 | << dendl; | |
9141 | failed.insert(devname); | |
9142 | continue; | |
9143 | } | |
9144 | dout(10) << __func__ << " bdev " << devname << " on numa_node " << n | |
9145 | << dendl; | |
9146 | nodes.insert(n); | |
9147 | if (node < 0) { | |
9148 | node = n; | |
9149 | } | |
9150 | } | |
9151 | if (node >= 0 && nodes.size() == 1 && failed.empty()) { | |
9152 | *final_node = node; | |
9153 | } | |
9154 | if (out_nodes) { | |
9155 | *out_nodes = nodes; | |
9156 | } | |
9157 | if (out_failed) { | |
9158 | *out_failed = failed; | |
9159 | } | |
9160 | return 0; | |
9161 | } | |
9162 | ||
9163 | int BlueStore::get_devices(set<string> *ls) | |
9164 | { | |
9165 | if (bdev) { | |
9166 | bdev->get_devices(ls); | |
9167 | if (bluefs) { | |
9168 | bluefs->get_devices(ls); | |
9169 | } | |
9170 | return 0; | |
9171 | } | |
9172 | ||
9173 | // grumble, we haven't started up yet. | |
9174 | int r = _open_path(); | |
9175 | if (r < 0) | |
9176 | goto out; | |
9177 | r = _open_fsid(false); | |
9178 | if (r < 0) | |
9179 | goto out_path; | |
9180 | r = _read_fsid(&fsid); | |
9181 | if (r < 0) | |
9182 | goto out_fsid; | |
9183 | r = _lock_fsid(); | |
9184 | if (r < 0) | |
9185 | goto out_fsid; | |
9186 | r = _open_bdev(false); | |
9187 | if (r < 0) | |
9188 | goto out_fsid; | |
9189 | r = _minimal_open_bluefs(false); | |
9190 | if (r < 0) | |
9191 | goto out_bdev; | |
9192 | bdev->get_devices(ls); | |
9193 | if (bluefs) { | |
9194 | bluefs->get_devices(ls); | |
9195 | } | |
9196 | r = 0; | |
9197 | _minimal_close_bluefs(); | |
9198 | out_bdev: | |
9199 | _close_bdev(); | |
9200 | out_fsid: | |
9201 | _close_fsid(); | |
9202 | out_path: | |
9203 | _close_path(); | |
9204 | out: | |
9205 | return r; | |
7c673cae FG |
9206 | } |
9207 | ||
11fdf7f2 | 9208 | void BlueStore::_get_statfs_overall(struct store_statfs_t *buf) |
7c673cae FG |
9209 | { |
9210 | buf->reset(); | |
11fdf7f2 | 9211 | |
9f95a23c TL |
9212 | buf->omap_allocated = |
9213 | db->estimate_prefix_size(PREFIX_OMAP, string()) + | |
9214 | db->estimate_prefix_size(PREFIX_PERPOOL_OMAP, string()); | |
11fdf7f2 TL |
9215 | |
9216 | uint64_t bfree = alloc->get_free(); | |
7c673cae FG |
9217 | |
9218 | if (bluefs) { | |
9f95a23c TL |
9219 | int64_t bluefs_total = bluefs->get_total(bluefs_layout.shared_bdev); |
9220 | int64_t bluefs_free = bluefs->get_free(bluefs_layout.shared_bdev); | |
94b18763 FG |
9221 | // part of our shared device is "free" according to BlueFS, but we |
9222 | // can't touch bluestore_bluefs_min of it. | |
9223 | int64_t shared_available = std::min( | |
11fdf7f2 TL |
9224 | bluefs_free, |
9225 | int64_t(bluefs_total - cct->_conf->bluestore_bluefs_min)); | |
9226 | buf->internally_reserved = bluefs_total - shared_available; | |
94b18763 | 9227 | if (shared_available > 0) { |
11fdf7f2 TL |
9228 | bfree += shared_available; |
9229 | } | |
9230 | // include dedicated db, too, if that isn't the shared device. | |
9f95a23c | 9231 | if (bluefs_layout.shared_bdev != BlueFS::BDEV_DB) { |
11fdf7f2 | 9232 | buf->total += bluefs->get_total(BlueFS::BDEV_DB); |
7c673cae | 9233 | } |
11fdf7f2 TL |
9234 | // call any non-omap bluefs space "internal metadata" |
9235 | buf->internal_metadata = | |
9236 | std::max(bluefs->get_used(), (uint64_t)cct->_conf->bluestore_bluefs_min) | |
9237 | - buf->omap_allocated; | |
7c673cae FG |
9238 | } |
9239 | ||
11fdf7f2 TL |
9240 | uint64_t thin_total, thin_avail; |
9241 | if (bdev->get_thin_utilization(&thin_total, &thin_avail)) { | |
9242 | buf->total += thin_total; | |
9243 | ||
9244 | // we are limited by both the size of the virtual device and the | |
9245 | // underlying physical device. | |
9246 | bfree = std::min(bfree, thin_avail); | |
9247 | ||
9248 | buf->allocated = thin_total - thin_avail; | |
9249 | } else { | |
9250 | buf->total += bdev->get_size(); | |
9251 | } | |
9252 | buf->available = bfree; | |
9253 | } | |
9254 | ||
9255 | int BlueStore::statfs(struct store_statfs_t *buf, | |
9256 | osd_alert_list_t* alerts) | |
9257 | { | |
9258 | if (alerts) { | |
9259 | alerts->clear(); | |
9260 | _log_alerts(*alerts); | |
9261 | } | |
9262 | _get_statfs_overall(buf); | |
31f18b77 | 9263 | { |
11fdf7f2 | 9264 | std::lock_guard l(vstatfs_lock); |
31f18b77 | 9265 | buf->allocated = vstatfs.allocated(); |
11fdf7f2 TL |
9266 | buf->data_stored = vstatfs.stored(); |
9267 | buf->data_compressed = vstatfs.compressed(); | |
9268 | buf->data_compressed_original = vstatfs.compressed_original(); | |
9269 | buf->data_compressed_allocated = vstatfs.compressed_allocated(); | |
9270 | } | |
9271 | ||
9272 | dout(20) << __func__ << " " << *buf << dendl; | |
9273 | return 0; | |
9274 | } | |
9275 | ||
9f95a23c TL |
9276 | int BlueStore::pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, |
9277 | bool *out_per_pool_omap) | |
11fdf7f2 TL |
9278 | { |
9279 | dout(20) << __func__ << " pool " << pool_id<< dendl; | |
81eedcae | 9280 | |
11fdf7f2 TL |
9281 | if (!per_pool_stat_collection) { |
9282 | dout(20) << __func__ << " not supported in legacy mode " << dendl; | |
9283 | return -ENOTSUP; | |
7c673cae | 9284 | } |
11fdf7f2 | 9285 | buf->reset(); |
7c673cae | 9286 | |
11fdf7f2 TL |
9287 | { |
9288 | std::lock_guard l(vstatfs_lock); | |
9289 | osd_pools[pool_id].publish(buf); | |
9290 | } | |
9f95a23c TL |
9291 | |
9292 | string key_prefix; | |
9293 | _key_encode_u64(pool_id, &key_prefix); | |
9294 | buf->omap_allocated = db->estimate_prefix_size(PREFIX_PERPOOL_OMAP, | |
9295 | key_prefix); | |
9296 | *out_per_pool_omap = per_pool_omap; | |
9297 | ||
11fdf7f2 | 9298 | dout(10) << __func__ << *buf << dendl; |
7c673cae FG |
9299 | return 0; |
9300 | } | |
9301 | ||
81eedcae TL |
9302 | void BlueStore::_check_legacy_statfs_alert() |
9303 | { | |
9304 | string s; | |
9305 | if (!per_pool_stat_collection && | |
eafe8130 | 9306 | cct->_conf->bluestore_warn_on_legacy_statfs) { |
81eedcae TL |
9307 | s = "legacy statfs reporting detected, " |
9308 | "suggest to run store repair to get consistent statistic reports"; | |
9309 | } | |
9310 | std::lock_guard l(qlock); | |
9311 | legacy_statfs_alert = s; | |
9312 | } | |
9313 | ||
9f95a23c TL |
9314 | void BlueStore::_check_no_per_pool_omap_alert() |
9315 | { | |
9316 | string s; | |
9317 | if (!per_pool_omap && | |
9318 | cct->_conf->bluestore_warn_on_no_per_pool_omap) { | |
9319 | s = "legacy (not per-pool) omap detected, " | |
9320 | "suggest to run store repair to measure per-pool omap usage"; | |
9321 | } | |
9322 | std::lock_guard l(qlock); | |
9323 | no_per_pool_omap_alert = s; | |
9324 | } | |
9325 | ||
7c673cae FG |
9326 | // --------------- |
9327 | // cache | |
9328 | ||
9329 | BlueStore::CollectionRef BlueStore::_get_collection(const coll_t& cid) | |
9330 | { | |
9f95a23c | 9331 | std::shared_lock l(coll_lock); |
7c673cae FG |
9332 | ceph::unordered_map<coll_t,CollectionRef>::iterator cp = coll_map.find(cid); |
9333 | if (cp == coll_map.end()) | |
9334 | return CollectionRef(); | |
9335 | return cp->second; | |
9336 | } | |
9337 | ||
9338 | void BlueStore::_queue_reap_collection(CollectionRef& c) | |
9339 | { | |
9340 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
94b18763 FG |
9341 | // _reap_collections and this in the same thread, |
9342 | // so no need a lock. | |
7c673cae FG |
9343 | removed_collections.push_back(c); |
9344 | } | |
9345 | ||
9346 | void BlueStore::_reap_collections() | |
9347 | { | |
94b18763 | 9348 | |
7c673cae FG |
9349 | list<CollectionRef> removed_colls; |
9350 | { | |
94b18763 FG |
9351 | // _queue_reap_collection and this in the same thread. |
9352 | // So no need a lock. | |
9353 | if (!removed_collections.empty()) | |
9354 | removed_colls.swap(removed_collections); | |
9355 | else | |
9356 | return; | |
7c673cae FG |
9357 | } |
9358 | ||
94b18763 FG |
9359 | list<CollectionRef>::iterator p = removed_colls.begin(); |
9360 | while (p != removed_colls.end()) { | |
7c673cae FG |
9361 | CollectionRef c = *p; |
9362 | dout(10) << __func__ << " " << c << " " << c->cid << dendl; | |
9363 | if (c->onode_map.map_any([&](OnodeRef o) { | |
11fdf7f2 | 9364 | ceph_assert(!o->exists); |
7c673cae FG |
9365 | if (o->flushing_count.load()) { |
9366 | dout(10) << __func__ << " " << c << " " << c->cid << " " << o->oid | |
9367 | << " flush_txns " << o->flushing_count << dendl; | |
94b18763 | 9368 | return true; |
7c673cae | 9369 | } |
94b18763 | 9370 | return false; |
7c673cae | 9371 | })) { |
94b18763 | 9372 | ++p; |
7c673cae FG |
9373 | continue; |
9374 | } | |
9375 | c->onode_map.clear(); | |
94b18763 | 9376 | p = removed_colls.erase(p); |
7c673cae FG |
9377 | dout(10) << __func__ << " " << c << " " << c->cid << " done" << dendl; |
9378 | } | |
94b18763 | 9379 | if (removed_colls.empty()) { |
7c673cae | 9380 | dout(10) << __func__ << " all reaped" << dendl; |
94b18763 FG |
9381 | } else { |
9382 | removed_collections.splice(removed_collections.begin(), removed_colls); | |
7c673cae FG |
9383 | } |
9384 | } | |
9385 | ||
9386 | void BlueStore::_update_cache_logger() | |
9387 | { | |
9388 | uint64_t num_onodes = 0; | |
9f95a23c | 9389 | uint64_t num_pinned_onodes = 0; |
7c673cae FG |
9390 | uint64_t num_extents = 0; |
9391 | uint64_t num_blobs = 0; | |
9392 | uint64_t num_buffers = 0; | |
9393 | uint64_t num_buffer_bytes = 0; | |
9f95a23c TL |
9394 | for (auto c : onode_cache_shards) { |
9395 | c->add_stats(&num_onodes, &num_pinned_onodes); | |
9396 | } | |
9397 | for (auto c : buffer_cache_shards) { | |
9398 | c->add_stats(&num_extents, &num_blobs, | |
9399 | &num_buffers, &num_buffer_bytes); | |
7c673cae FG |
9400 | } |
9401 | logger->set(l_bluestore_onodes, num_onodes); | |
9f95a23c | 9402 | logger->set(l_bluestore_pinned_onodes, num_pinned_onodes); |
7c673cae FG |
9403 | logger->set(l_bluestore_extents, num_extents); |
9404 | logger->set(l_bluestore_blobs, num_blobs); | |
9405 | logger->set(l_bluestore_buffers, num_buffers); | |
9406 | logger->set(l_bluestore_buffer_bytes, num_buffer_bytes); | |
9407 | } | |
9408 | ||
9409 | // --------------- | |
9410 | // read operations | |
9411 | ||
9412 | ObjectStore::CollectionHandle BlueStore::open_collection(const coll_t& cid) | |
9413 | { | |
9414 | return _get_collection(cid); | |
9415 | } | |
9416 | ||
11fdf7f2 TL |
9417 | ObjectStore::CollectionHandle BlueStore::create_new_collection( |
9418 | const coll_t& cid) | |
7c673cae | 9419 | { |
9f95a23c TL |
9420 | std::unique_lock l{coll_lock}; |
9421 | auto c = ceph::make_ref<Collection>( | |
11fdf7f2 | 9422 | this, |
9f95a23c TL |
9423 | onode_cache_shards[cid.hash_to_shard(onode_cache_shards.size())], |
9424 | buffer_cache_shards[cid.hash_to_shard(buffer_cache_shards.size())], | |
11fdf7f2 TL |
9425 | cid); |
9426 | new_coll_map[cid] = c; | |
9f95a23c | 9427 | _osr_attach(c.get()); |
11fdf7f2 TL |
9428 | return c; |
9429 | } | |
9430 | ||
9431 | void BlueStore::set_collection_commit_queue( | |
9432 | const coll_t& cid, | |
9433 | ContextQueue *commit_queue) | |
9434 | { | |
9435 | if (commit_queue) { | |
9f95a23c | 9436 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
9437 | if (coll_map.count(cid)) { |
9438 | coll_map[cid]->commit_queue = commit_queue; | |
9439 | } else if (new_coll_map.count(cid)) { | |
9440 | new_coll_map[cid]->commit_queue = commit_queue; | |
9441 | } | |
9442 | } | |
7c673cae FG |
9443 | } |
9444 | ||
11fdf7f2 | 9445 | |
7c673cae FG |
9446 | bool BlueStore::exists(CollectionHandle &c_, const ghobject_t& oid) |
9447 | { | |
9448 | Collection *c = static_cast<Collection *>(c_.get()); | |
9449 | dout(10) << __func__ << " " << c->cid << " " << oid << dendl; | |
9450 | if (!c->exists) | |
9451 | return false; | |
9452 | ||
9453 | bool r = true; | |
9454 | ||
9455 | { | |
9f95a23c | 9456 | std::shared_lock l(c->lock); |
7c673cae FG |
9457 | OnodeRef o = c->get_onode(oid, false); |
9458 | if (!o || !o->exists) | |
9459 | r = false; | |
9460 | } | |
9461 | ||
7c673cae FG |
9462 | return r; |
9463 | } | |
9464 | ||
7c673cae FG |
9465 | int BlueStore::stat( |
9466 | CollectionHandle &c_, | |
9467 | const ghobject_t& oid, | |
9468 | struct stat *st, | |
9469 | bool allow_eio) | |
9470 | { | |
9471 | Collection *c = static_cast<Collection *>(c_.get()); | |
9472 | if (!c->exists) | |
9473 | return -ENOENT; | |
9474 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
9475 | ||
9476 | { | |
9f95a23c | 9477 | std::shared_lock l(c->lock); |
7c673cae FG |
9478 | OnodeRef o = c->get_onode(oid, false); |
9479 | if (!o || !o->exists) | |
9480 | return -ENOENT; | |
9481 | st->st_size = o->onode.size; | |
9482 | st->st_blksize = 4096; | |
9483 | st->st_blocks = (st->st_size + st->st_blksize - 1) / st->st_blksize; | |
9484 | st->st_nlink = 1; | |
9485 | } | |
9486 | ||
7c673cae FG |
9487 | int r = 0; |
9488 | if (_debug_mdata_eio(oid)) { | |
9489 | r = -EIO; | |
9490 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
9491 | } | |
9492 | return r; | |
9493 | } | |
9494 | int BlueStore::set_collection_opts( | |
11fdf7f2 | 9495 | CollectionHandle& ch, |
7c673cae FG |
9496 | const pool_opts_t& opts) |
9497 | { | |
7c673cae | 9498 | Collection *c = static_cast<Collection *>(ch.get()); |
11fdf7f2 | 9499 | dout(15) << __func__ << " " << ch->cid << " options " << opts << dendl; |
7c673cae FG |
9500 | if (!c->exists) |
9501 | return -ENOENT; | |
9f95a23c | 9502 | std::unique_lock l{c->lock}; |
7c673cae FG |
9503 | c->pool_opts = opts; |
9504 | return 0; | |
9505 | } | |
9506 | ||
7c673cae FG |
9507 | int BlueStore::read( |
9508 | CollectionHandle &c_, | |
9509 | const ghobject_t& oid, | |
9510 | uint64_t offset, | |
9511 | size_t length, | |
9512 | bufferlist& bl, | |
224ce89b | 9513 | uint32_t op_flags) |
7c673cae | 9514 | { |
11fdf7f2 | 9515 | auto start = mono_clock::now(); |
7c673cae FG |
9516 | Collection *c = static_cast<Collection *>(c_.get()); |
9517 | const coll_t &cid = c->get_cid(); | |
9518 | dout(15) << __func__ << " " << cid << " " << oid | |
9519 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9520 | << dendl; | |
9521 | if (!c->exists) | |
9522 | return -ENOENT; | |
9523 | ||
9524 | bl.clear(); | |
9525 | int r; | |
9526 | { | |
9f95a23c | 9527 | std::shared_lock l(c->lock); |
11fdf7f2 | 9528 | auto start1 = mono_clock::now(); |
7c673cae | 9529 | OnodeRef o = c->get_onode(oid, false); |
494da23a TL |
9530 | log_latency("get_onode@read", |
9531 | l_bluestore_read_onode_meta_lat, | |
9532 | mono_clock::now() - start1, | |
9533 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9534 | if (!o || !o->exists) { |
9535 | r = -ENOENT; | |
9536 | goto out; | |
9537 | } | |
9538 | ||
9539 | if (offset == length && offset == 0) | |
9540 | length = o->onode.size; | |
9541 | ||
9542 | r = _do_read(c, o, offset, length, bl, op_flags); | |
b32b8144 FG |
9543 | if (r == -EIO) { |
9544 | logger->inc(l_bluestore_read_eio); | |
9545 | } | |
7c673cae FG |
9546 | } |
9547 | ||
9548 | out: | |
28e407b8 | 9549 | if (r >= 0 && _debug_data_eio(oid)) { |
7c673cae FG |
9550 | r = -EIO; |
9551 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
11fdf7f2 TL |
9552 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ |
9553 | cct->_conf->bluestore_debug_random_read_err && | |
9554 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
9555 | 100.0)) == 0) { | |
224ce89b WB |
9556 | dout(0) << __func__ << ": inject random EIO" << dendl; |
9557 | r = -EIO; | |
7c673cae FG |
9558 | } |
9559 | dout(10) << __func__ << " " << cid << " " << oid | |
9560 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
9561 | << " = " << r << dendl; | |
494da23a TL |
9562 | log_latency(__func__, |
9563 | l_bluestore_read_lat, | |
9564 | mono_clock::now() - start, | |
9565 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
9566 | return r; |
9567 | } | |
9568 | ||
9f95a23c | 9569 | void BlueStore::_read_cache( |
7c673cae FG |
9570 | OnodeRef o, |
9571 | uint64_t offset, | |
9572 | size_t length, | |
9f95a23c TL |
9573 | int read_cache_policy, |
9574 | ready_regions_t& ready_regions, | |
9575 | blobs2read_t& blobs2read) | |
7c673cae | 9576 | { |
7c673cae | 9577 | // build blob-wise list to of stuff read (that isn't cached) |
7c673cae FG |
9578 | unsigned left = length; |
9579 | uint64_t pos = offset; | |
7c673cae FG |
9580 | auto lp = o->extent_map.seek_lextent(offset); |
9581 | while (left > 0 && lp != o->extent_map.extent_map.end()) { | |
9582 | if (pos < lp->logical_offset) { | |
9583 | unsigned hole = lp->logical_offset - pos; | |
9584 | if (hole >= left) { | |
9f95a23c | 9585 | break; |
7c673cae FG |
9586 | } |
9587 | dout(30) << __func__ << " hole 0x" << std::hex << pos << "~" << hole | |
9f95a23c | 9588 | << std::dec << dendl; |
7c673cae FG |
9589 | pos += hole; |
9590 | left -= hole; | |
9591 | } | |
94b18763 | 9592 | BlobRef& bptr = lp->blob; |
7c673cae FG |
9593 | unsigned l_off = pos - lp->logical_offset; |
9594 | unsigned b_off = l_off + lp->blob_offset; | |
9595 | unsigned b_len = std::min(left, lp->length - l_off); | |
9596 | ||
9597 | ready_regions_t cache_res; | |
9598 | interval_set<uint32_t> cache_interval; | |
9599 | bptr->shared_blob->bc.read( | |
91327a77 AA |
9600 | bptr->shared_blob->get_cache(), b_off, b_len, cache_res, cache_interval, |
9601 | read_cache_policy); | |
7c673cae | 9602 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c TL |
9603 | << " need 0x" << b_off << "~" << b_len |
9604 | << " cache has 0x" << cache_interval | |
9605 | << std::dec << dendl; | |
7c673cae FG |
9606 | |
9607 | auto pc = cache_res.begin(); | |
11fdf7f2 | 9608 | uint64_t chunk_size = bptr->get_blob().get_chunk_size(block_size); |
7c673cae FG |
9609 | while (b_len > 0) { |
9610 | unsigned l; | |
9611 | if (pc != cache_res.end() && | |
9f95a23c TL |
9612 | pc->first == b_off) { |
9613 | l = pc->second.length(); | |
9614 | ready_regions[pos].claim(pc->second); | |
9615 | dout(30) << __func__ << " use cache 0x" << std::hex << pos << ": 0x" | |
9616 | << b_off << "~" << l << std::dec << dendl; | |
9617 | ++pc; | |
7c673cae | 9618 | } else { |
9f95a23c TL |
9619 | l = b_len; |
9620 | if (pc != cache_res.end()) { | |
9621 | ceph_assert(pc->first > b_off); | |
9622 | l = pc->first - b_off; | |
9623 | } | |
9624 | dout(30) << __func__ << " will read 0x" << std::hex << pos << ": 0x" | |
9625 | << b_off << "~" << l << std::dec << dendl; | |
9626 | // merge regions | |
9627 | { | |
9628 | uint64_t r_off = b_off; | |
9629 | uint64_t r_len = l; | |
9630 | uint64_t front = r_off % chunk_size; | |
9631 | if (front) { | |
9632 | r_off -= front; | |
9633 | r_len += front; | |
9634 | } | |
9635 | unsigned tail = r_len % chunk_size; | |
9636 | if (tail) { | |
9637 | r_len += chunk_size - tail; | |
9638 | } | |
9639 | bool merged = false; | |
9640 | regions2read_t& r2r = blobs2read[bptr]; | |
9641 | if (r2r.size()) { | |
9642 | read_req_t& pre = r2r.back(); | |
9643 | if (r_off <= (pre.r_off + pre.r_len)) { | |
9644 | front += (r_off - pre.r_off); | |
9645 | pre.r_len += (r_off + r_len - pre.r_off - pre.r_len); | |
9646 | pre.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9647 | merged = true; | |
9648 | } | |
9649 | } | |
9650 | if (!merged) { | |
9651 | read_req_t req(r_off, r_len); | |
9652 | req.regs.emplace_back(region_t(pos, b_off, l, front)); | |
9653 | r2r.emplace_back(std::move(req)); | |
9654 | } | |
9655 | } | |
7c673cae FG |
9656 | } |
9657 | pos += l; | |
9658 | b_off += l; | |
9659 | left -= l; | |
9660 | b_len -= l; | |
9661 | } | |
9662 | ++lp; | |
9663 | } | |
9f95a23c | 9664 | } |
7c673cae | 9665 | |
9f95a23c TL |
9666 | int BlueStore::_prepare_read_ioc( |
9667 | blobs2read_t& blobs2read, | |
9668 | vector<bufferlist>* compressed_blob_bls, | |
9669 | IOContext* ioc) | |
9670 | { | |
7c673cae | 9671 | for (auto& p : blobs2read) { |
94b18763 | 9672 | const BlobRef& bptr = p.first; |
11fdf7f2 | 9673 | regions2read_t& r2r = p.second; |
7c673cae | 9674 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9675 | << " need " << r2r << std::dec << dendl; |
7c673cae FG |
9676 | if (bptr->get_blob().is_compressed()) { |
9677 | // read the whole thing | |
9f95a23c TL |
9678 | if (compressed_blob_bls->empty()) { |
9679 | // ensure we avoid any reallocation on subsequent blobs | |
9680 | compressed_blob_bls->reserve(blobs2read.size()); | |
9681 | } | |
9682 | compressed_blob_bls->push_back(bufferlist()); | |
9683 | bufferlist& bl = compressed_blob_bls->back(); | |
9684 | auto r = bptr->get_blob().map( | |
9685 | 0, bptr->get_blob().get_ondisk_length(), | |
9686 | [&](uint64_t offset, uint64_t length) { | |
9687 | int r = bdev->aio_read(offset, length, &bl, ioc); | |
9688 | if (r < 0) | |
7c673cae FG |
9689 | return r; |
9690 | return 0; | |
9f95a23c | 9691 | }); |
b32b8144 FG |
9692 | if (r < 0) { |
9693 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) << dendl; | |
9694 | if (r == -EIO) { | |
9695 | // propagate EIO to caller | |
9696 | return r; | |
9697 | } | |
11fdf7f2 | 9698 | ceph_assert(r == 0); |
b32b8144 | 9699 | } |
7c673cae FG |
9700 | } else { |
9701 | // read the pieces | |
11fdf7f2 | 9702 | for (auto& req : r2r) { |
9f95a23c TL |
9703 | dout(20) << __func__ << " region 0x" << std::hex |
9704 | << req.regs.front().logical_offset | |
9705 | << ": 0x" << req.regs.front().blob_xoffset | |
9706 | << " reading 0x" << req.r_off | |
9707 | << "~" << req.r_len << std::dec | |
9708 | << dendl; | |
7c673cae | 9709 | |
9f95a23c TL |
9710 | // read it |
9711 | auto r = bptr->get_blob().map( | |
9712 | req.r_off, req.r_len, | |
9713 | [&](uint64_t offset, uint64_t length) { | |
9714 | int r = bdev->aio_read(offset, length, &req.bl, ioc); | |
9715 | if (r < 0) | |
7c673cae FG |
9716 | return r; |
9717 | return 0; | |
9f95a23c | 9718 | }); |
b32b8144 FG |
9719 | if (r < 0) { |
9720 | derr << __func__ << " bdev-read failed: " << cpp_strerror(r) | |
9721 | << dendl; | |
9722 | if (r == -EIO) { | |
9723 | // propagate EIO to caller | |
9724 | return r; | |
9725 | } | |
11fdf7f2 | 9726 | ceph_assert(r == 0); |
b32b8144 | 9727 | } |
9f95a23c | 9728 | ceph_assert(req.bl.length() == req.r_len); |
7c673cae FG |
9729 | } |
9730 | } | |
9731 | } | |
9f95a23c TL |
9732 | return 0; |
9733 | } | |
11fdf7f2 | 9734 | |
9f95a23c TL |
9735 | int BlueStore::_generate_read_result_bl( |
9736 | OnodeRef o, | |
9737 | uint64_t offset, | |
9738 | size_t length, | |
9739 | ready_regions_t& ready_regions, | |
9740 | vector<bufferlist>& compressed_blob_bls, | |
9741 | blobs2read_t& blobs2read, | |
9742 | bool buffered, | |
9743 | bool* csum_error, | |
9744 | bufferlist& bl) | |
9745 | { | |
9746 | // enumerate and decompress desired blobs | |
7c673cae FG |
9747 | auto p = compressed_blob_bls.begin(); |
9748 | blobs2read_t::iterator b2r_it = blobs2read.begin(); | |
9749 | while (b2r_it != blobs2read.end()) { | |
94b18763 | 9750 | const BlobRef& bptr = b2r_it->first; |
11fdf7f2 | 9751 | regions2read_t& r2r = b2r_it->second; |
7c673cae | 9752 | dout(20) << __func__ << " blob " << *bptr << std::hex |
9f95a23c | 9753 | << " need 0x" << r2r << std::dec << dendl; |
7c673cae | 9754 | if (bptr->get_blob().is_compressed()) { |
11fdf7f2 | 9755 | ceph_assert(p != compressed_blob_bls.end()); |
7c673cae FG |
9756 | bufferlist& compressed_bl = *p++; |
9757 | if (_verify_csum(o, &bptr->get_blob(), 0, compressed_bl, | |
9f95a23c TL |
9758 | r2r.front().regs.front().logical_offset) < 0) { |
9759 | *csum_error = true; | |
9760 | return -EIO; | |
7c673cae FG |
9761 | } |
9762 | bufferlist raw_bl; | |
9f95a23c | 9763 | auto r = _decompress(compressed_bl, &raw_bl); |
7c673cae | 9764 | if (r < 0) |
9f95a23c | 9765 | return r; |
7c673cae | 9766 | if (buffered) { |
9f95a23c TL |
9767 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), 0, |
9768 | raw_bl); | |
7c673cae | 9769 | } |
11fdf7f2 TL |
9770 | for (auto& req : r2r) { |
9771 | for (auto& r : req.regs) { | |
9772 | ready_regions[r.logical_offset].substr_of( | |
9773 | raw_bl, r.blob_xoffset, r.length); | |
9774 | } | |
7c673cae FG |
9775 | } |
9776 | } else { | |
11fdf7f2 | 9777 | for (auto& req : r2r) { |
9f95a23c TL |
9778 | if (_verify_csum(o, &bptr->get_blob(), req.r_off, req.bl, |
9779 | req.regs.front().logical_offset) < 0) { | |
9780 | *csum_error = true; | |
9781 | return -EIO; | |
9782 | } | |
9783 | if (buffered) { | |
9784 | bptr->shared_blob->bc.did_read(bptr->shared_blob->get_cache(), | |
9785 | req.r_off, req.bl); | |
9786 | } | |
7c673cae | 9787 | |
9f95a23c TL |
9788 | // prune and keep result |
9789 | for (const auto& r : req.regs) { | |
9790 | ready_regions[r.logical_offset].substr_of(req.bl, r.front, r.length); | |
11fdf7f2 | 9791 | } |
7c673cae FG |
9792 | } |
9793 | } | |
9794 | ++b2r_it; | |
9795 | } | |
9796 | ||
9797 | // generate a resulting buffer | |
9798 | auto pr = ready_regions.begin(); | |
9799 | auto pr_end = ready_regions.end(); | |
9f95a23c | 9800 | uint64_t pos = 0; |
7c673cae FG |
9801 | while (pos < length) { |
9802 | if (pr != pr_end && pr->first == pos + offset) { | |
9803 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9804 | << ": data from 0x" << pr->first << "~" << pr->second.length() |
9805 | << std::dec << dendl; | |
7c673cae FG |
9806 | pos += pr->second.length(); |
9807 | bl.claim_append(pr->second); | |
9808 | ++pr; | |
9809 | } else { | |
9810 | uint64_t l = length - pos; | |
9811 | if (pr != pr_end) { | |
11fdf7f2 | 9812 | ceph_assert(pr->first > pos + offset); |
9f95a23c | 9813 | l = pr->first - (pos + offset); |
7c673cae FG |
9814 | } |
9815 | dout(30) << __func__ << " assemble 0x" << std::hex << pos | |
9f95a23c TL |
9816 | << ": zeros for 0x" << (pos + offset) << "~" << l |
9817 | << std::dec << dendl; | |
7c673cae FG |
9818 | bl.append_zero(l); |
9819 | pos += l; | |
9820 | } | |
9821 | } | |
11fdf7f2 TL |
9822 | ceph_assert(bl.length() == length); |
9823 | ceph_assert(pos == length); | |
9824 | ceph_assert(pr == pr_end); | |
9f95a23c TL |
9825 | return 0; |
9826 | } | |
9827 | ||
9828 | int BlueStore::_do_read( | |
9829 | Collection *c, | |
9830 | OnodeRef o, | |
9831 | uint64_t offset, | |
9832 | size_t length, | |
9833 | bufferlist& bl, | |
9834 | uint32_t op_flags, | |
9835 | uint64_t retry_count) | |
9836 | { | |
9837 | FUNCTRACE(cct); | |
9838 | int r = 0; | |
9839 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
9840 | ||
9841 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
9842 | << " size 0x" << o->onode.size << " (" << std::dec | |
9843 | << o->onode.size << ")" << dendl; | |
9844 | bl.clear(); | |
9845 | ||
9846 | if (offset >= o->onode.size) { | |
9847 | return r; | |
9848 | } | |
9849 | ||
9850 | // generally, don't buffer anything, unless the client explicitly requests | |
9851 | // it. | |
9852 | bool buffered = false; | |
9853 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
9854 | dout(20) << __func__ << " will do buffered read" << dendl; | |
9855 | buffered = true; | |
9856 | } else if (cct->_conf->bluestore_default_buffered_read && | |
9857 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
9858 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
9859 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
9860 | buffered = true; | |
9861 | } | |
9862 | ||
9863 | if (offset + length > o->onode.size) { | |
9864 | length = o->onode.size - offset; | |
9865 | } | |
9866 | ||
9867 | auto start = mono_clock::now(); | |
9868 | o->extent_map.fault_range(db, offset, length); | |
9869 | log_latency(__func__, | |
9870 | l_bluestore_read_onode_meta_lat, | |
9871 | mono_clock::now() - start, | |
9872 | cct->_conf->bluestore_log_op_age); | |
9873 | _dump_onode<30>(cct, *o); | |
9874 | ||
9875 | // for deep-scrub, we only read dirty cache and bypass clean cache in | |
9876 | // order to read underlying block device in case there are silent disk errors. | |
9877 | if (op_flags & CEPH_OSD_OP_FLAG_BYPASS_CLEAN_CACHE) { | |
9878 | dout(20) << __func__ << " will bypass cache and do direct read" << dendl; | |
9879 | read_cache_policy = BufferSpace::BYPASS_CLEAN_CACHE; | |
9880 | } | |
9881 | ||
9882 | // build blob-wise list to of stuff read (that isn't cached) | |
9883 | ready_regions_t ready_regions; | |
9884 | blobs2read_t blobs2read; | |
9885 | _read_cache(o, offset, length, read_cache_policy, ready_regions, blobs2read); | |
9886 | ||
9887 | ||
9888 | // read raw blob data. | |
9889 | start = mono_clock::now(); // for the sake of simplicity | |
9890 | // measure the whole block below. | |
9891 | // The error isn't that much... | |
9892 | vector<bufferlist> compressed_blob_bls; | |
9893 | IOContext ioc(cct, NULL, true); // allow EIO | |
9894 | r = _prepare_read_ioc(blobs2read, &compressed_blob_bls, &ioc); | |
9895 | // we always issue aio for reading, so errors other than EIO are not allowed | |
9896 | if (r < 0) | |
9897 | return r; | |
9898 | ||
9899 | int64_t num_ios = length; | |
9900 | if (ioc.has_pending_aios()) { | |
9901 | num_ios = -ioc.get_num_ios(); | |
9902 | bdev->aio_submit(&ioc); | |
9903 | dout(20) << __func__ << " waiting for aio" << dendl; | |
9904 | ioc.aio_wait(); | |
9905 | r = ioc.get_return_value(); | |
9906 | if (r < 0) { | |
9907 | ceph_assert(r == -EIO); // no other errors allowed | |
9908 | return -EIO; | |
9909 | } | |
9910 | } | |
9911 | log_latency_fn(__func__, | |
9912 | l_bluestore_read_wait_aio_lat, | |
9913 | mono_clock::now() - start, | |
9914 | cct->_conf->bluestore_log_op_age, | |
9915 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
9916 | ); | |
9917 | ||
9918 | bool csum_error = false; | |
9919 | r = _generate_read_result_bl(o, offset, length, ready_regions, | |
9920 | compressed_blob_bls, blobs2read, | |
9921 | buffered, &csum_error, bl); | |
9922 | if (csum_error) { | |
9923 | // Handles spurious read errors caused by a kernel bug. | |
9924 | // We sometimes get all-zero pages as a result of the read under | |
9925 | // high memory pressure. Retrying the failing read succeeds in most | |
9926 | // cases. | |
9927 | // See also: http://tracker.ceph.com/issues/22464 | |
9928 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
9929 | return -EIO; | |
9930 | } | |
9931 | return _do_read(c, o, offset, length, bl, op_flags, retry_count + 1); | |
9932 | } | |
7c673cae | 9933 | r = bl.length(); |
f64942e4 AA |
9934 | if (retry_count) { |
9935 | logger->inc(l_bluestore_reads_with_retries); | |
9936 | dout(5) << __func__ << " read at 0x" << std::hex << offset << "~" << length | |
9937 | << " failed " << std::dec << retry_count << " times before succeeding" << dendl; | |
9938 | } | |
7c673cae FG |
9939 | return r; |
9940 | } | |
9941 | ||
9942 | int BlueStore::_verify_csum(OnodeRef& o, | |
9943 | const bluestore_blob_t* blob, uint64_t blob_xoffset, | |
9944 | const bufferlist& bl, | |
9945 | uint64_t logical_offset) const | |
9946 | { | |
9947 | int bad; | |
9948 | uint64_t bad_csum; | |
11fdf7f2 | 9949 | auto start = mono_clock::now(); |
7c673cae | 9950 | int r = blob->verify_csum(blob_xoffset, bl, &bad, &bad_csum); |
f64942e4 AA |
9951 | if (cct->_conf->bluestore_debug_inject_csum_err_probability > 0 && |
9952 | (rand() % 10000) < cct->_conf->bluestore_debug_inject_csum_err_probability * 10000.0) { | |
9953 | derr << __func__ << " injecting bluestore checksum verifcation error" << dendl; | |
9954 | bad = blob_xoffset; | |
9955 | r = -1; | |
9956 | bad_csum = 0xDEADBEEF; | |
9957 | } | |
7c673cae FG |
9958 | if (r < 0) { |
9959 | if (r == -1) { | |
9960 | PExtentVector pex; | |
9961 | blob->map( | |
9962 | bad, | |
9963 | blob->get_csum_chunk_size(), | |
9964 | [&](uint64_t offset, uint64_t length) { | |
9965 | pex.emplace_back(bluestore_pextent_t(offset, length)); | |
9966 | return 0; | |
9967 | }); | |
9968 | derr << __func__ << " bad " | |
9969 | << Checksummer::get_csum_type_string(blob->csum_type) | |
9970 | << "/0x" << std::hex << blob->get_csum_chunk_size() | |
9971 | << " checksum at blob offset 0x" << bad | |
9972 | << ", got 0x" << bad_csum << ", expected 0x" | |
9973 | << blob->get_csum_item(bad / blob->get_csum_chunk_size()) << std::dec | |
9974 | << ", device location " << pex | |
9975 | << ", logical extent 0x" << std::hex | |
9976 | << (logical_offset + bad - blob_xoffset) << "~" | |
9977 | << blob->get_csum_chunk_size() << std::dec | |
9978 | << ", object " << o->oid | |
9979 | << dendl; | |
9980 | } else { | |
9981 | derr << __func__ << " failed with exit code: " << cpp_strerror(r) << dendl; | |
9982 | } | |
9983 | } | |
494da23a TL |
9984 | log_latency(__func__, |
9985 | l_bluestore_csum_lat, | |
9986 | mono_clock::now() - start, | |
9987 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 TL |
9988 | if (cct->_conf->bluestore_ignore_data_csum) { |
9989 | return 0; | |
9990 | } | |
7c673cae FG |
9991 | return r; |
9992 | } | |
9993 | ||
9994 | int BlueStore::_decompress(bufferlist& source, bufferlist* result) | |
9995 | { | |
9996 | int r = 0; | |
11fdf7f2 TL |
9997 | auto start = mono_clock::now(); |
9998 | auto i = source.cbegin(); | |
7c673cae | 9999 | bluestore_compression_header_t chdr; |
11fdf7f2 | 10000 | decode(chdr, i); |
7c673cae FG |
10001 | int alg = int(chdr.type); |
10002 | CompressorRef cp = compressor; | |
10003 | if (!cp || (int)cp->get_type() != alg) { | |
10004 | cp = Compressor::create(cct, alg); | |
10005 | } | |
10006 | ||
10007 | if (!cp.get()) { | |
10008 | // if compressor isn't available - error, because cannot return | |
10009 | // decompressed data? | |
11fdf7f2 TL |
10010 | |
10011 | const char* alg_name = Compressor::get_comp_alg_name(alg); | |
10012 | derr << __func__ << " can't load decompressor " << alg_name << dendl; | |
10013 | _set_compression_alert(false, alg_name); | |
7c673cae FG |
10014 | r = -EIO; |
10015 | } else { | |
10016 | r = cp->decompress(i, chdr.length, *result); | |
10017 | if (r < 0) { | |
10018 | derr << __func__ << " decompression failed with exit code " << r << dendl; | |
10019 | r = -EIO; | |
10020 | } | |
10021 | } | |
494da23a TL |
10022 | log_latency(__func__, |
10023 | l_bluestore_decompress_lat, | |
10024 | mono_clock::now() - start, | |
10025 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
10026 | return r; |
10027 | } | |
10028 | ||
10029 | // this stores fiemap into interval_set, other variations | |
10030 | // use it internally | |
10031 | int BlueStore::_fiemap( | |
10032 | CollectionHandle &c_, | |
10033 | const ghobject_t& oid, | |
10034 | uint64_t offset, | |
10035 | size_t length, | |
10036 | interval_set<uint64_t>& destset) | |
10037 | { | |
10038 | Collection *c = static_cast<Collection *>(c_.get()); | |
10039 | if (!c->exists) | |
10040 | return -ENOENT; | |
10041 | { | |
9f95a23c | 10042 | std::shared_lock l(c->lock); |
7c673cae FG |
10043 | |
10044 | OnodeRef o = c->get_onode(oid, false); | |
10045 | if (!o || !o->exists) { | |
10046 | return -ENOENT; | |
10047 | } | |
81eedcae | 10048 | _dump_onode<30>(cct, *o); |
7c673cae FG |
10049 | |
10050 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10051 | << " size 0x" << o->onode.size << std::dec << dendl; | |
10052 | ||
10053 | boost::intrusive::set<Extent>::iterator ep, eend; | |
10054 | if (offset >= o->onode.size) | |
10055 | goto out; | |
10056 | ||
10057 | if (offset + length > o->onode.size) { | |
10058 | length = o->onode.size - offset; | |
10059 | } | |
10060 | ||
10061 | o->extent_map.fault_range(db, offset, length); | |
10062 | eend = o->extent_map.extent_map.end(); | |
10063 | ep = o->extent_map.seek_lextent(offset); | |
10064 | while (length > 0) { | |
10065 | dout(20) << __func__ << " offset " << offset << dendl; | |
10066 | if (ep != eend && ep->logical_offset + ep->length <= offset) { | |
10067 | ++ep; | |
10068 | continue; | |
10069 | } | |
10070 | ||
10071 | uint64_t x_len = length; | |
10072 | if (ep != eend && ep->logical_offset <= offset) { | |
10073 | uint64_t x_off = offset - ep->logical_offset; | |
11fdf7f2 | 10074 | x_len = std::min(x_len, ep->length - x_off); |
7c673cae FG |
10075 | dout(30) << __func__ << " lextent 0x" << std::hex << offset << "~" |
10076 | << x_len << std::dec << " blob " << ep->blob << dendl; | |
10077 | destset.insert(offset, x_len); | |
10078 | length -= x_len; | |
10079 | offset += x_len; | |
10080 | if (x_off + x_len == ep->length) | |
10081 | ++ep; | |
10082 | continue; | |
10083 | } | |
10084 | if (ep != eend && | |
10085 | ep->logical_offset > offset && | |
10086 | ep->logical_offset - offset < x_len) { | |
10087 | x_len = ep->logical_offset - offset; | |
10088 | } | |
10089 | offset += x_len; | |
10090 | length -= x_len; | |
10091 | } | |
10092 | } | |
9f95a23c TL |
10093 | |
10094 | out: | |
10095 | dout(20) << __func__ << " 0x" << std::hex << offset << "~" << length | |
10096 | << " size = 0x(" << destset << ")" << std::dec << dendl; | |
10097 | return 0; | |
10098 | } | |
10099 | ||
10100 | int BlueStore::fiemap( | |
10101 | CollectionHandle &c_, | |
10102 | const ghobject_t& oid, | |
10103 | uint64_t offset, | |
10104 | size_t length, | |
10105 | bufferlist& bl) | |
10106 | { | |
10107 | interval_set<uint64_t> m; | |
10108 | int r = _fiemap(c_, oid, offset, length, m); | |
10109 | if (r >= 0) { | |
10110 | encode(m, bl); | |
10111 | } | |
10112 | return r; | |
10113 | } | |
10114 | ||
10115 | int BlueStore::fiemap( | |
10116 | CollectionHandle &c_, | |
10117 | const ghobject_t& oid, | |
10118 | uint64_t offset, | |
10119 | size_t length, | |
10120 | map<uint64_t, uint64_t>& destmap) | |
10121 | { | |
10122 | interval_set<uint64_t> m; | |
10123 | int r = _fiemap(c_, oid, offset, length, m); | |
10124 | if (r >= 0) { | |
10125 | destmap = std::move(m).detach(); | |
10126 | } | |
10127 | return r; | |
10128 | } | |
10129 | ||
10130 | int BlueStore::readv( | |
10131 | CollectionHandle &c_, | |
10132 | const ghobject_t& oid, | |
10133 | interval_set<uint64_t>& m, | |
10134 | bufferlist& bl, | |
10135 | uint32_t op_flags) | |
10136 | { | |
10137 | auto start = mono_clock::now(); | |
10138 | Collection *c = static_cast<Collection *>(c_.get()); | |
10139 | const coll_t &cid = c->get_cid(); | |
10140 | dout(15) << __func__ << " " << cid << " " << oid | |
10141 | << " fiemap " << m | |
10142 | << dendl; | |
10143 | if (!c->exists) | |
10144 | return -ENOENT; | |
10145 | ||
10146 | bl.clear(); | |
10147 | int r; | |
10148 | { | |
10149 | std::shared_lock l(c->lock); | |
10150 | auto start1 = mono_clock::now(); | |
10151 | OnodeRef o = c->get_onode(oid, false); | |
10152 | log_latency("get_onode@read", | |
10153 | l_bluestore_read_onode_meta_lat, | |
10154 | mono_clock::now() - start1, | |
10155 | cct->_conf->bluestore_log_op_age); | |
10156 | if (!o || !o->exists) { | |
10157 | r = -ENOENT; | |
10158 | goto out; | |
10159 | } | |
10160 | ||
10161 | if (m.empty()) { | |
10162 | r = 0; | |
10163 | goto out; | |
10164 | } | |
10165 | ||
10166 | r = _do_readv(c, o, m, bl, op_flags); | |
10167 | if (r == -EIO) { | |
10168 | logger->inc(l_bluestore_read_eio); | |
10169 | } | |
10170 | } | |
10171 | ||
10172 | out: | |
10173 | if (r >= 0 && _debug_data_eio(oid)) { | |
10174 | r = -EIO; | |
10175 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10176 | } else if (oid.hobj.pool > 0 && /* FIXME, see #23029 */ | |
10177 | cct->_conf->bluestore_debug_random_read_err && | |
10178 | (rand() % (int)(cct->_conf->bluestore_debug_random_read_err * | |
10179 | 100.0)) == 0) { | |
10180 | dout(0) << __func__ << ": inject random EIO" << dendl; | |
10181 | r = -EIO; | |
10182 | } | |
10183 | dout(10) << __func__ << " " << cid << " " << oid | |
10184 | << " fiemap " << m << std::dec | |
10185 | << " = " << r << dendl; | |
10186 | log_latency(__func__, | |
10187 | l_bluestore_read_lat, | |
10188 | mono_clock::now() - start, | |
10189 | cct->_conf->bluestore_log_op_age); | |
10190 | return r; | |
10191 | } | |
10192 | ||
10193 | int BlueStore::_do_readv( | |
10194 | Collection *c, | |
10195 | OnodeRef o, | |
10196 | const interval_set<uint64_t>& m, | |
10197 | bufferlist& bl, | |
10198 | uint32_t op_flags, | |
10199 | uint64_t retry_count) | |
10200 | { | |
10201 | FUNCTRACE(cct); | |
10202 | int r = 0; | |
10203 | int read_cache_policy = 0; // do not bypass clean or dirty cache | |
10204 | ||
10205 | dout(20) << __func__ << " fiemap " << m << std::hex | |
10206 | << " size 0x" << o->onode.size << " (" << std::dec | |
10207 | << o->onode.size << ")" << dendl; | |
10208 | ||
10209 | // generally, don't buffer anything, unless the client explicitly requests | |
10210 | // it. | |
10211 | bool buffered = false; | |
10212 | if (op_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { | |
10213 | dout(20) << __func__ << " will do buffered read" << dendl; | |
10214 | buffered = true; | |
10215 | } else if (cct->_conf->bluestore_default_buffered_read && | |
10216 | (op_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
10217 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
10218 | dout(20) << __func__ << " defaulting to buffered read" << dendl; | |
10219 | buffered = true; | |
10220 | } | |
10221 | // this method must be idempotent since we may call it several times | |
10222 | // before we finally read the expected result. | |
10223 | bl.clear(); | |
10224 | ||
10225 | // call fiemap first! | |
10226 | ceph_assert(m.range_start() <= o->onode.size); | |
10227 | ceph_assert(m.range_end() <= o->onode.size); | |
10228 | auto start = mono_clock::now(); | |
10229 | o->extent_map.fault_range(db, m.range_start(), m.range_end() - m.range_start()); | |
10230 | log_latency(__func__, | |
10231 | l_bluestore_read_onode_meta_lat, | |
10232 | mono_clock::now() - start, | |
10233 | cct->_conf->bluestore_log_op_age); | |
10234 | _dump_onode<30>(cct, *o); | |
10235 | ||
10236 | IOContext ioc(cct, NULL, true); // allow EIO | |
10237 | vector<std::tuple<ready_regions_t, vector<bufferlist>, blobs2read_t>> raw_results; | |
10238 | raw_results.reserve(m.num_intervals()); | |
10239 | int i = 0; | |
10240 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10241 | raw_results.push_back({}); | |
10242 | _read_cache(o, p.get_start(), p.get_len(), read_cache_policy, | |
10243 | std::get<0>(raw_results[i]), std::get<2>(raw_results[i])); | |
10244 | r = _prepare_read_ioc(std::get<2>(raw_results[i]), &std::get<1>(raw_results[i]), &ioc); | |
10245 | // we always issue aio for reading, so errors other than EIO are not allowed | |
10246 | if (r < 0) | |
10247 | return r; | |
10248 | } | |
10249 | ||
10250 | auto num_ios = m.size(); | |
10251 | if (ioc.has_pending_aios()) { | |
10252 | num_ios = ioc.get_num_ios(); | |
10253 | bdev->aio_submit(&ioc); | |
10254 | dout(20) << __func__ << " waiting for aio" << dendl; | |
10255 | ioc.aio_wait(); | |
10256 | r = ioc.get_return_value(); | |
10257 | if (r < 0) { | |
10258 | ceph_assert(r == -EIO); // no other errors allowed | |
10259 | return -EIO; | |
10260 | } | |
10261 | } | |
10262 | log_latency_fn(__func__, | |
10263 | l_bluestore_read_wait_aio_lat, | |
10264 | mono_clock::now() - start, | |
10265 | cct->_conf->bluestore_log_op_age, | |
10266 | [&](auto lat) { return ", num_ios = " + stringify(num_ios); } | |
10267 | ); | |
10268 | ||
10269 | ceph_assert(raw_results.size() == (size_t)m.num_intervals()); | |
10270 | i = 0; | |
10271 | for (auto p = m.begin(); p != m.end(); p++, i++) { | |
10272 | bool csum_error = false; | |
10273 | bufferlist t; | |
10274 | r = _generate_read_result_bl(o, p.get_start(), p.get_len(), | |
10275 | std::get<0>(raw_results[i]), | |
10276 | std::get<1>(raw_results[i]), | |
10277 | std::get<2>(raw_results[i]), | |
10278 | buffered, &csum_error, t); | |
10279 | if (csum_error) { | |
10280 | // Handles spurious read errors caused by a kernel bug. | |
10281 | // We sometimes get all-zero pages as a result of the read under | |
10282 | // high memory pressure. Retrying the failing read succeeds in most | |
10283 | // cases. | |
10284 | // See also: http://tracker.ceph.com/issues/22464 | |
10285 | if (retry_count >= cct->_conf->bluestore_retry_disk_reads) { | |
10286 | return -EIO; | |
10287 | } | |
10288 | return _do_readv(c, o, m, bl, op_flags, retry_count + 1); | |
10289 | } | |
10290 | bl.claim_append(t); | |
10291 | } | |
10292 | if (retry_count) { | |
10293 | logger->inc(l_bluestore_reads_with_retries); | |
10294 | dout(5) << __func__ << " read fiemap " << m | |
10295 | << " failed " << retry_count << " times before succeeding" | |
10296 | << dendl; | |
10297 | } | |
10298 | return bl.length(); | |
7c673cae FG |
10299 | } |
10300 | ||
9f95a23c | 10301 | int BlueStore::dump_onode(CollectionHandle &c_, |
7c673cae | 10302 | const ghobject_t& oid, |
9f95a23c TL |
10303 | const string& section_name, |
10304 | Formatter *f) | |
7c673cae | 10305 | { |
9f95a23c TL |
10306 | Collection *c = static_cast<Collection *>(c_.get()); |
10307 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10308 | if (!c->exists) | |
10309 | return -ENOENT; | |
7c673cae | 10310 | |
9f95a23c TL |
10311 | int r; |
10312 | { | |
10313 | std::shared_lock l(c->lock); | |
10314 | ||
10315 | OnodeRef o = c->get_onode(oid, false); | |
10316 | if (!o || !o->exists) { | |
10317 | r = -ENOENT; | |
10318 | goto out; | |
10319 | } | |
10320 | // FIXME minor: actually the next line isn't enough to | |
10321 | // load shared blobs. Leaving as is for now.. | |
10322 | // | |
10323 | o->extent_map.fault_range(db, 0, OBJECT_MAX_SIZE); | |
10324 | ||
10325 | _dump_onode<0>(cct, *o); | |
10326 | f->open_object_section(section_name.c_str()); | |
10327 | o->dump(f); | |
10328 | f->close_section(); | |
10329 | r = 0; | |
7c673cae | 10330 | } |
9f95a23c TL |
10331 | out: |
10332 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10333 | << " = " << r << dendl; | |
7c673cae FG |
10334 | return r; |
10335 | } | |
10336 | ||
7c673cae FG |
10337 | int BlueStore::getattr( |
10338 | CollectionHandle &c_, | |
10339 | const ghobject_t& oid, | |
10340 | const char *name, | |
10341 | bufferptr& value) | |
10342 | { | |
10343 | Collection *c = static_cast<Collection *>(c_.get()); | |
10344 | dout(15) << __func__ << " " << c->cid << " " << oid << " " << name << dendl; | |
10345 | if (!c->exists) | |
10346 | return -ENOENT; | |
10347 | ||
10348 | int r; | |
10349 | { | |
9f95a23c | 10350 | std::shared_lock l(c->lock); |
31f18b77 | 10351 | mempool::bluestore_cache_other::string k(name); |
7c673cae FG |
10352 | |
10353 | OnodeRef o = c->get_onode(oid, false); | |
10354 | if (!o || !o->exists) { | |
10355 | r = -ENOENT; | |
10356 | goto out; | |
10357 | } | |
10358 | ||
10359 | if (!o->onode.attrs.count(k)) { | |
10360 | r = -ENODATA; | |
10361 | goto out; | |
10362 | } | |
10363 | value = o->onode.attrs[k]; | |
10364 | r = 0; | |
10365 | } | |
10366 | out: | |
7c673cae FG |
10367 | if (r == 0 && _debug_mdata_eio(oid)) { |
10368 | r = -EIO; | |
10369 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10370 | } | |
10371 | dout(10) << __func__ << " " << c->cid << " " << oid << " " << name | |
10372 | << " = " << r << dendl; | |
10373 | return r; | |
10374 | } | |
10375 | ||
7c673cae FG |
10376 | int BlueStore::getattrs( |
10377 | CollectionHandle &c_, | |
10378 | const ghobject_t& oid, | |
10379 | map<string,bufferptr>& aset) | |
10380 | { | |
10381 | Collection *c = static_cast<Collection *>(c_.get()); | |
10382 | dout(15) << __func__ << " " << c->cid << " " << oid << dendl; | |
10383 | if (!c->exists) | |
10384 | return -ENOENT; | |
10385 | ||
10386 | int r; | |
10387 | { | |
9f95a23c | 10388 | std::shared_lock l(c->lock); |
7c673cae FG |
10389 | |
10390 | OnodeRef o = c->get_onode(oid, false); | |
10391 | if (!o || !o->exists) { | |
10392 | r = -ENOENT; | |
10393 | goto out; | |
10394 | } | |
10395 | for (auto& i : o->onode.attrs) { | |
10396 | aset.emplace(i.first.c_str(), i.second); | |
10397 | } | |
10398 | r = 0; | |
10399 | } | |
10400 | ||
10401 | out: | |
7c673cae FG |
10402 | if (r == 0 && _debug_mdata_eio(oid)) { |
10403 | r = -EIO; | |
10404 | derr << __func__ << " " << c->cid << " " << oid << " INJECT EIO" << dendl; | |
10405 | } | |
10406 | dout(10) << __func__ << " " << c->cid << " " << oid | |
10407 | << " = " << r << dendl; | |
10408 | return r; | |
10409 | } | |
10410 | ||
10411 | int BlueStore::list_collections(vector<coll_t>& ls) | |
10412 | { | |
9f95a23c | 10413 | std::shared_lock l(coll_lock); |
11fdf7f2 | 10414 | ls.reserve(coll_map.size()); |
7c673cae FG |
10415 | for (ceph::unordered_map<coll_t, CollectionRef>::iterator p = coll_map.begin(); |
10416 | p != coll_map.end(); | |
10417 | ++p) | |
10418 | ls.push_back(p->first); | |
10419 | return 0; | |
10420 | } | |
10421 | ||
10422 | bool BlueStore::collection_exists(const coll_t& c) | |
10423 | { | |
9f95a23c | 10424 | std::shared_lock l(coll_lock); |
7c673cae FG |
10425 | return coll_map.count(c); |
10426 | } | |
10427 | ||
11fdf7f2 | 10428 | int BlueStore::collection_empty(CollectionHandle& ch, bool *empty) |
7c673cae | 10429 | { |
11fdf7f2 | 10430 | dout(15) << __func__ << " " << ch->cid << dendl; |
7c673cae FG |
10431 | vector<ghobject_t> ls; |
10432 | ghobject_t next; | |
11fdf7f2 | 10433 | int r = collection_list(ch, ghobject_t(), ghobject_t::get_max(), 1, |
7c673cae FG |
10434 | &ls, &next); |
10435 | if (r < 0) { | |
10436 | derr << __func__ << " collection_list returned: " << cpp_strerror(r) | |
10437 | << dendl; | |
10438 | return r; | |
10439 | } | |
10440 | *empty = ls.empty(); | |
11fdf7f2 | 10441 | dout(10) << __func__ << " " << ch->cid << " = " << (int)(*empty) << dendl; |
7c673cae FG |
10442 | return 0; |
10443 | } | |
10444 | ||
11fdf7f2 | 10445 | int BlueStore::collection_bits(CollectionHandle& ch) |
7c673cae | 10446 | { |
11fdf7f2 TL |
10447 | dout(15) << __func__ << " " << ch->cid << dendl; |
10448 | Collection *c = static_cast<Collection*>(ch.get()); | |
9f95a23c | 10449 | std::shared_lock l(c->lock); |
11fdf7f2 | 10450 | dout(10) << __func__ << " " << ch->cid << " = " << c->cnode.bits << dendl; |
7c673cae FG |
10451 | return c->cnode.bits; |
10452 | } | |
10453 | ||
7c673cae FG |
10454 | int BlueStore::collection_list( |
10455 | CollectionHandle &c_, const ghobject_t& start, const ghobject_t& end, int max, | |
10456 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10457 | { | |
10458 | Collection *c = static_cast<Collection *>(c_.get()); | |
11fdf7f2 | 10459 | c->flush(); |
7c673cae FG |
10460 | dout(15) << __func__ << " " << c->cid |
10461 | << " start " << start << " end " << end << " max " << max << dendl; | |
10462 | int r; | |
10463 | { | |
9f95a23c | 10464 | std::shared_lock l(c->lock); |
7c673cae FG |
10465 | r = _collection_list(c, start, end, max, ls, pnext); |
10466 | } | |
10467 | ||
7c673cae FG |
10468 | dout(10) << __func__ << " " << c->cid |
10469 | << " start " << start << " end " << end << " max " << max | |
10470 | << " = " << r << ", ls.size() = " << ls->size() | |
10471 | << ", next = " << (pnext ? *pnext : ghobject_t()) << dendl; | |
10472 | return r; | |
10473 | } | |
10474 | ||
10475 | int BlueStore::_collection_list( | |
10476 | Collection *c, const ghobject_t& start, const ghobject_t& end, int max, | |
10477 | vector<ghobject_t> *ls, ghobject_t *pnext) | |
10478 | { | |
10479 | ||
10480 | if (!c->exists) | |
10481 | return -ENOENT; | |
10482 | ||
494da23a | 10483 | auto start_time = mono_clock::now(); |
7c673cae FG |
10484 | int r = 0; |
10485 | ghobject_t static_next; | |
10486 | KeyValueDB::Iterator it; | |
10487 | string temp_start_key, temp_end_key; | |
10488 | string start_key, end_key; | |
10489 | bool set_next = false; | |
10490 | string pend; | |
10491 | bool temp; | |
10492 | ||
10493 | if (!pnext) | |
10494 | pnext = &static_next; | |
10495 | ||
11fdf7f2 | 10496 | if (start.is_max() || start.hobj.is_max()) { |
7c673cae FG |
10497 | goto out; |
10498 | } | |
10499 | get_coll_key_range(c->cid, c->cnode.bits, &temp_start_key, &temp_end_key, | |
10500 | &start_key, &end_key); | |
10501 | dout(20) << __func__ | |
10502 | << " range " << pretty_binary_string(temp_start_key) | |
10503 | << " to " << pretty_binary_string(temp_end_key) | |
10504 | << " and " << pretty_binary_string(start_key) | |
10505 | << " to " << pretty_binary_string(end_key) | |
10506 | << " start " << start << dendl; | |
10507 | it = db->get_iterator(PREFIX_OBJ); | |
10508 | if (start == ghobject_t() || | |
10509 | start.hobj == hobject_t() || | |
10510 | start == c->cid.get_min_hobj()) { | |
10511 | it->upper_bound(temp_start_key); | |
10512 | temp = true; | |
10513 | } else { | |
10514 | string k; | |
10515 | get_object_key(cct, start, &k); | |
10516 | if (start.hobj.is_temp()) { | |
10517 | temp = true; | |
11fdf7f2 | 10518 | ceph_assert(k >= temp_start_key && k < temp_end_key); |
7c673cae FG |
10519 | } else { |
10520 | temp = false; | |
11fdf7f2 | 10521 | ceph_assert(k >= start_key && k < end_key); |
7c673cae | 10522 | } |
11fdf7f2 | 10523 | dout(20) << __func__ << " start from " << pretty_binary_string(k) |
7c673cae FG |
10524 | << " temp=" << (int)temp << dendl; |
10525 | it->lower_bound(k); | |
10526 | } | |
10527 | if (end.hobj.is_max()) { | |
10528 | pend = temp ? temp_end_key : end_key; | |
10529 | } else { | |
10530 | get_object_key(cct, end, &end_key); | |
10531 | if (end.hobj.is_temp()) { | |
10532 | if (temp) | |
10533 | pend = end_key; | |
10534 | else | |
10535 | goto out; | |
10536 | } else { | |
10537 | pend = temp ? temp_end_key : end_key; | |
10538 | } | |
10539 | } | |
10540 | dout(20) << __func__ << " pend " << pretty_binary_string(pend) << dendl; | |
10541 | while (true) { | |
10542 | if (!it->valid() || it->key() >= pend) { | |
10543 | if (!it->valid()) | |
10544 | dout(20) << __func__ << " iterator not valid (end of db?)" << dendl; | |
10545 | else | |
10546 | dout(20) << __func__ << " key " << pretty_binary_string(it->key()) | |
10547 | << " >= " << end << dendl; | |
10548 | if (temp) { | |
10549 | if (end.hobj.is_temp()) { | |
10550 | break; | |
10551 | } | |
10552 | dout(30) << __func__ << " switch to non-temp namespace" << dendl; | |
10553 | temp = false; | |
10554 | it->upper_bound(start_key); | |
10555 | pend = end_key; | |
10556 | dout(30) << __func__ << " pend " << pretty_binary_string(pend) << dendl; | |
10557 | continue; | |
10558 | } | |
10559 | break; | |
10560 | } | |
10561 | dout(30) << __func__ << " key " << pretty_binary_string(it->key()) << dendl; | |
10562 | if (is_extent_shard_key(it->key())) { | |
10563 | it->next(); | |
10564 | continue; | |
10565 | } | |
10566 | ghobject_t oid; | |
10567 | int r = get_key_object(it->key(), &oid); | |
11fdf7f2 | 10568 | ceph_assert(r == 0); |
7c673cae FG |
10569 | dout(20) << __func__ << " oid " << oid << " end " << end << dendl; |
10570 | if (ls->size() >= (unsigned)max) { | |
10571 | dout(20) << __func__ << " reached max " << max << dendl; | |
10572 | *pnext = oid; | |
10573 | set_next = true; | |
10574 | break; | |
10575 | } | |
10576 | ls->push_back(oid); | |
10577 | it->next(); | |
10578 | } | |
10579 | out: | |
10580 | if (!set_next) { | |
10581 | *pnext = ghobject_t::get_max(); | |
10582 | } | |
494da23a TL |
10583 | log_latency_fn( |
10584 | __func__, | |
10585 | l_bluestore_clist_lat, | |
10586 | mono_clock::now() - start_time, | |
10587 | cct->_conf->bluestore_log_collection_list_age, | |
10588 | [&] (const ceph::timespan& lat) { | |
10589 | ostringstream ostr; | |
10590 | ostr << ", lat = " << timespan_str(lat) | |
10591 | << " cid =" << c->cid | |
10592 | << " start " << start << " end " << end | |
10593 | << " max " << max; | |
10594 | return ostr.str(); | |
10595 | } | |
10596 | ); | |
7c673cae FG |
10597 | return r; |
10598 | } | |
10599 | ||
7c673cae FG |
10600 | int BlueStore::omap_get( |
10601 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10602 | const ghobject_t &oid, ///< [in] Object containing omap | |
10603 | bufferlist *header, ///< [out] omap header | |
10604 | map<string, bufferlist> *out /// < [out] Key to value map | |
10605 | ) | |
10606 | { | |
10607 | Collection *c = static_cast<Collection *>(c_.get()); | |
9f95a23c TL |
10608 | return _omap_get(c, oid, header, out); |
10609 | } | |
10610 | ||
10611 | int BlueStore::_omap_get( | |
10612 | Collection *c, ///< [in] Collection containing oid | |
10613 | const ghobject_t &oid, ///< [in] Object containing omap | |
10614 | bufferlist *header, ///< [out] omap header | |
10615 | map<string, bufferlist> *out /// < [out] Key to value map | |
10616 | ) | |
10617 | { | |
7c673cae FG |
10618 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; |
10619 | if (!c->exists) | |
10620 | return -ENOENT; | |
9f95a23c | 10621 | std::shared_lock l(c->lock); |
7c673cae FG |
10622 | int r = 0; |
10623 | OnodeRef o = c->get_onode(oid, false); | |
10624 | if (!o || !o->exists) { | |
10625 | r = -ENOENT; | |
10626 | goto out; | |
10627 | } | |
9f95a23c TL |
10628 | r = _onode_omap_get(o, header, out); |
10629 | out: | |
10630 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10631 | << dendl; | |
10632 | return r; | |
10633 | } | |
10634 | ||
10635 | int BlueStore::_onode_omap_get( | |
10636 | const OnodeRef &o, ///< [in] Object containing omap | |
10637 | bufferlist *header, ///< [out] omap header | |
10638 | map<string, bufferlist> *out /// < [out] Key to value map | |
10639 | ) | |
10640 | { | |
10641 | int r = 0; | |
10642 | if (!o || !o->exists) { | |
10643 | r = -ENOENT; | |
10644 | goto out; | |
10645 | } | |
7c673cae FG |
10646 | if (!o->onode.has_omap()) |
10647 | goto out; | |
10648 | o->flush(); | |
10649 | { | |
9f95a23c | 10650 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10651 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10652 | string head, tail; |
9f95a23c TL |
10653 | o->get_omap_header(&head); |
10654 | o->get_omap_tail(&tail); | |
7c673cae FG |
10655 | it->lower_bound(head); |
10656 | while (it->valid()) { | |
10657 | if (it->key() == head) { | |
9f95a23c TL |
10658 | dout(30) << __func__ << " got header" << dendl; |
10659 | *header = it->value(); | |
7c673cae | 10660 | } else if (it->key() >= tail) { |
9f95a23c TL |
10661 | dout(30) << __func__ << " reached tail" << dendl; |
10662 | break; | |
7c673cae | 10663 | } else { |
9f95a23c TL |
10664 | string user_key; |
10665 | o->decode_omap_key(it->key(), &user_key); | |
10666 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) | |
10667 | << " -> " << user_key << dendl; | |
10668 | (*out)[user_key] = it->value(); | |
7c673cae FG |
10669 | } |
10670 | it->next(); | |
10671 | } | |
10672 | } | |
9f95a23c | 10673 | out: |
7c673cae FG |
10674 | return r; |
10675 | } | |
10676 | ||
7c673cae FG |
10677 | int BlueStore::omap_get_header( |
10678 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10679 | const ghobject_t &oid, ///< [in] Object containing omap | |
10680 | bufferlist *header, ///< [out] omap header | |
10681 | bool allow_eio ///< [in] don't assert on eio | |
10682 | ) | |
10683 | { | |
10684 | Collection *c = static_cast<Collection *>(c_.get()); | |
10685 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10686 | if (!c->exists) | |
10687 | return -ENOENT; | |
9f95a23c | 10688 | std::shared_lock l(c->lock); |
7c673cae FG |
10689 | int r = 0; |
10690 | OnodeRef o = c->get_onode(oid, false); | |
10691 | if (!o || !o->exists) { | |
10692 | r = -ENOENT; | |
10693 | goto out; | |
10694 | } | |
10695 | if (!o->onode.has_omap()) | |
10696 | goto out; | |
10697 | o->flush(); | |
10698 | { | |
10699 | string head; | |
9f95a23c TL |
10700 | o->get_omap_header(&head); |
10701 | if (db->get(o->get_omap_prefix(), head, header) >= 0) { | |
7c673cae FG |
10702 | dout(30) << __func__ << " got header" << dendl; |
10703 | } else { | |
10704 | dout(30) << __func__ << " no header" << dendl; | |
10705 | } | |
10706 | } | |
10707 | out: | |
10708 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10709 | << dendl; | |
10710 | return r; | |
10711 | } | |
10712 | ||
7c673cae FG |
10713 | int BlueStore::omap_get_keys( |
10714 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10715 | const ghobject_t &oid, ///< [in] Object containing omap | |
10716 | set<string> *keys ///< [out] Keys defined on oid | |
10717 | ) | |
10718 | { | |
10719 | Collection *c = static_cast<Collection *>(c_.get()); | |
10720 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10721 | if (!c->exists) | |
10722 | return -ENOENT; | |
9f95a23c | 10723 | std::shared_lock l(c->lock); |
7c673cae FG |
10724 | int r = 0; |
10725 | OnodeRef o = c->get_onode(oid, false); | |
10726 | if (!o || !o->exists) { | |
10727 | r = -ENOENT; | |
10728 | goto out; | |
10729 | } | |
10730 | if (!o->onode.has_omap()) | |
10731 | goto out; | |
10732 | o->flush(); | |
10733 | { | |
9f95a23c | 10734 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 10735 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 10736 | string head, tail; |
9f95a23c TL |
10737 | o->get_omap_key(string(), &head); |
10738 | o->get_omap_tail(&tail); | |
7c673cae FG |
10739 | it->lower_bound(head); |
10740 | while (it->valid()) { | |
10741 | if (it->key() >= tail) { | |
10742 | dout(30) << __func__ << " reached tail" << dendl; | |
10743 | break; | |
10744 | } | |
10745 | string user_key; | |
9f95a23c | 10746 | o->decode_omap_key(it->key(), &user_key); |
11fdf7f2 | 10747 | dout(20) << __func__ << " got " << pretty_binary_string(it->key()) |
7c673cae FG |
10748 | << " -> " << user_key << dendl; |
10749 | keys->insert(user_key); | |
10750 | it->next(); | |
11fdf7f2 TL |
10751 | } |
10752 | } | |
10753 | out: | |
10754 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10755 | << dendl; | |
10756 | return r; | |
7c673cae FG |
10757 | } |
10758 | ||
10759 | int BlueStore::omap_get_values( | |
10760 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10761 | const ghobject_t &oid, ///< [in] Object containing omap | |
10762 | const set<string> &keys, ///< [in] Keys to get | |
10763 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
10764 | ) | |
10765 | { | |
10766 | Collection *c = static_cast<Collection *>(c_.get()); | |
10767 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10768 | if (!c->exists) | |
10769 | return -ENOENT; | |
9f95a23c | 10770 | std::shared_lock l(c->lock); |
7c673cae FG |
10771 | int r = 0; |
10772 | string final_key; | |
10773 | OnodeRef o = c->get_onode(oid, false); | |
10774 | if (!o || !o->exists) { | |
10775 | r = -ENOENT; | |
10776 | goto out; | |
10777 | } | |
9f95a23c | 10778 | if (!o->onode.has_omap()) { |
7c673cae | 10779 | goto out; |
9f95a23c TL |
10780 | } |
10781 | o->flush(); | |
11fdf7f2 | 10782 | { |
9f95a23c TL |
10783 | const string& prefix = o->get_omap_prefix(); |
10784 | o->get_omap_key(string(), &final_key); | |
10785 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10786 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10787 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10788 | final_key += *p; |
10789 | bufferlist val; | |
10790 | if (db->get(prefix, final_key, &val) >= 0) { | |
10791 | dout(30) << __func__ << " got " << pretty_binary_string(final_key) | |
10792 | << " -> " << *p << dendl; | |
10793 | out->insert(make_pair(*p, val)); | |
10794 | } | |
7c673cae FG |
10795 | } |
10796 | } | |
10797 | out: | |
10798 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10799 | << dendl; | |
10800 | return r; | |
10801 | } | |
10802 | ||
9f95a23c TL |
10803 | #ifdef WITH_SEASTAR |
10804 | int BlueStore::omap_get_values( | |
10805 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10806 | const ghobject_t &oid, ///< [in] Object containing omap | |
10807 | const std::optional<string> &start_after, ///< [in] Keys to get | |
10808 | map<string, bufferlist> *output ///< [out] Returned keys and values | |
10809 | ) | |
10810 | { | |
10811 | Collection *c = static_cast<Collection *>(c_.get()); | |
10812 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10813 | if (!c->exists) | |
10814 | return -ENOENT; | |
10815 | std::shared_lock l(c->lock); | |
10816 | int r = 0; | |
10817 | OnodeRef o = c->get_onode(oid, false); | |
10818 | if (!o || !o->exists) { | |
10819 | r = -ENOENT; | |
10820 | goto out; | |
10821 | } | |
10822 | if (!o->onode.has_omap()) { | |
10823 | goto out; | |
10824 | } | |
10825 | o->flush(); | |
10826 | { | |
10827 | ObjectMap::ObjectMapIterator iter = get_omap_iterator(c_, oid); | |
10828 | if (!iter) { | |
10829 | r = -ENOENT; | |
10830 | goto out; | |
10831 | } | |
10832 | iter->upper_bound(*start_after); | |
10833 | for (; iter->valid(); iter->next()) { | |
10834 | output->insert(make_pair(iter->key(), iter->value())); | |
10835 | } | |
10836 | } | |
10837 | ||
10838 | out: | |
10839 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10840 | << dendl; | |
10841 | return r; | |
10842 | } | |
10843 | #endif | |
10844 | ||
7c673cae FG |
10845 | int BlueStore::omap_check_keys( |
10846 | CollectionHandle &c_, ///< [in] Collection containing oid | |
10847 | const ghobject_t &oid, ///< [in] Object containing omap | |
10848 | const set<string> &keys, ///< [in] Keys to check | |
10849 | set<string> *out ///< [out] Subset of keys defined on oid | |
10850 | ) | |
10851 | { | |
10852 | Collection *c = static_cast<Collection *>(c_.get()); | |
10853 | dout(15) << __func__ << " " << c->get_cid() << " oid " << oid << dendl; | |
10854 | if (!c->exists) | |
10855 | return -ENOENT; | |
9f95a23c | 10856 | std::shared_lock l(c->lock); |
7c673cae FG |
10857 | int r = 0; |
10858 | string final_key; | |
10859 | OnodeRef o = c->get_onode(oid, false); | |
10860 | if (!o || !o->exists) { | |
10861 | r = -ENOENT; | |
10862 | goto out; | |
10863 | } | |
9f95a23c | 10864 | if (!o->onode.has_omap()) { |
7c673cae | 10865 | goto out; |
9f95a23c TL |
10866 | } |
10867 | o->flush(); | |
11fdf7f2 | 10868 | { |
9f95a23c TL |
10869 | const string& prefix = o->get_omap_prefix(); |
10870 | o->get_omap_key(string(), &final_key); | |
10871 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 10872 | for (set<string>::const_iterator p = keys.begin(); p != keys.end(); ++p) { |
9f95a23c | 10873 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
10874 | final_key += *p; |
10875 | bufferlist val; | |
10876 | if (db->get(prefix, final_key, &val) >= 0) { | |
10877 | dout(30) << __func__ << " have " << pretty_binary_string(final_key) | |
10878 | << " -> " << *p << dendl; | |
10879 | out->insert(*p); | |
10880 | } else { | |
10881 | dout(30) << __func__ << " miss " << pretty_binary_string(final_key) | |
10882 | << " -> " << *p << dendl; | |
10883 | } | |
7c673cae FG |
10884 | } |
10885 | } | |
10886 | out: | |
10887 | dout(10) << __func__ << " " << c->get_cid() << " oid " << oid << " = " << r | |
10888 | << dendl; | |
10889 | return r; | |
10890 | } | |
10891 | ||
7c673cae FG |
10892 | ObjectMap::ObjectMapIterator BlueStore::get_omap_iterator( |
10893 | CollectionHandle &c_, ///< [in] collection | |
10894 | const ghobject_t &oid ///< [in] object | |
10895 | ) | |
10896 | { | |
10897 | Collection *c = static_cast<Collection *>(c_.get()); | |
10898 | dout(10) << __func__ << " " << c->get_cid() << " " << oid << dendl; | |
10899 | if (!c->exists) { | |
10900 | return ObjectMap::ObjectMapIterator(); | |
10901 | } | |
9f95a23c | 10902 | std::shared_lock l(c->lock); |
7c673cae FG |
10903 | OnodeRef o = c->get_onode(oid, false); |
10904 | if (!o || !o->exists) { | |
10905 | dout(10) << __func__ << " " << oid << "doesn't exist" <<dendl; | |
10906 | return ObjectMap::ObjectMapIterator(); | |
10907 | } | |
10908 | o->flush(); | |
10909 | dout(10) << __func__ << " has_omap = " << (int)o->onode.has_omap() <<dendl; | |
9f95a23c | 10910 | KeyValueDB::Iterator it = db->get_iterator(o->get_omap_prefix()); |
7c673cae FG |
10911 | return ObjectMap::ObjectMapIterator(new OmapIteratorImpl(c, o, it)); |
10912 | } | |
10913 | ||
10914 | // ----------------- | |
10915 | // write helpers | |
10916 | ||
11fdf7f2 TL |
10917 | uint64_t BlueStore::_get_ondisk_reserved() const { |
10918 | return round_up_to( | |
10919 | std::max<uint64_t>(SUPER_RESERVED, min_alloc_size), min_alloc_size); | |
10920 | } | |
10921 | ||
7c673cae FG |
10922 | void BlueStore::_prepare_ondisk_format_super(KeyValueDB::Transaction& t) |
10923 | { | |
10924 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
10925 | << " min_compat_ondisk_format " << min_compat_ondisk_format | |
10926 | << dendl; | |
11fdf7f2 | 10927 | ceph_assert(ondisk_format == latest_ondisk_format); |
7c673cae FG |
10928 | { |
10929 | bufferlist bl; | |
11fdf7f2 | 10930 | encode(ondisk_format, bl); |
7c673cae FG |
10931 | t->set(PREFIX_SUPER, "ondisk_format", bl); |
10932 | } | |
10933 | { | |
10934 | bufferlist bl; | |
11fdf7f2 | 10935 | encode(min_compat_ondisk_format, bl); |
7c673cae FG |
10936 | t->set(PREFIX_SUPER, "min_compat_ondisk_format", bl); |
10937 | } | |
10938 | } | |
10939 | ||
10940 | int BlueStore::_open_super_meta() | |
10941 | { | |
10942 | // nid | |
10943 | { | |
10944 | nid_max = 0; | |
10945 | bufferlist bl; | |
10946 | db->get(PREFIX_SUPER, "nid_max", &bl); | |
11fdf7f2 | 10947 | auto p = bl.cbegin(); |
7c673cae FG |
10948 | try { |
10949 | uint64_t v; | |
11fdf7f2 | 10950 | decode(v, p); |
7c673cae FG |
10951 | nid_max = v; |
10952 | } catch (buffer::error& e) { | |
10953 | derr << __func__ << " unable to read nid_max" << dendl; | |
10954 | return -EIO; | |
10955 | } | |
10956 | dout(10) << __func__ << " old nid_max " << nid_max << dendl; | |
10957 | nid_last = nid_max.load(); | |
10958 | } | |
10959 | ||
10960 | // blobid | |
10961 | { | |
10962 | blobid_max = 0; | |
10963 | bufferlist bl; | |
10964 | db->get(PREFIX_SUPER, "blobid_max", &bl); | |
11fdf7f2 | 10965 | auto p = bl.cbegin(); |
7c673cae FG |
10966 | try { |
10967 | uint64_t v; | |
11fdf7f2 | 10968 | decode(v, p); |
7c673cae FG |
10969 | blobid_max = v; |
10970 | } catch (buffer::error& e) { | |
10971 | derr << __func__ << " unable to read blobid_max" << dendl; | |
10972 | return -EIO; | |
10973 | } | |
10974 | dout(10) << __func__ << " old blobid_max " << blobid_max << dendl; | |
10975 | blobid_last = blobid_max.load(); | |
10976 | } | |
10977 | ||
10978 | // freelist | |
10979 | { | |
10980 | bufferlist bl; | |
10981 | db->get(PREFIX_SUPER, "freelist_type", &bl); | |
10982 | if (bl.length()) { | |
10983 | freelist_type = std::string(bl.c_str(), bl.length()); | |
10984 | dout(10) << __func__ << " freelist_type " << freelist_type << dendl; | |
10985 | } else { | |
11fdf7f2 | 10986 | ceph_abort_msg("Not Support extent freelist manager"); |
7c673cae | 10987 | } |
7c673cae FG |
10988 | } |
10989 | ||
10990 | // ondisk format | |
10991 | int32_t compat_ondisk_format = 0; | |
10992 | { | |
10993 | bufferlist bl; | |
10994 | int r = db->get(PREFIX_SUPER, "ondisk_format", &bl); | |
10995 | if (r < 0) { | |
10996 | // base case: kraken bluestore is v1 and readable by v1 | |
10997 | dout(20) << __func__ << " missing ondisk_format; assuming kraken" | |
10998 | << dendl; | |
10999 | ondisk_format = 1; | |
11000 | compat_ondisk_format = 1; | |
11001 | } else { | |
11fdf7f2 | 11002 | auto p = bl.cbegin(); |
7c673cae | 11003 | try { |
11fdf7f2 | 11004 | decode(ondisk_format, p); |
7c673cae FG |
11005 | } catch (buffer::error& e) { |
11006 | derr << __func__ << " unable to read ondisk_format" << dendl; | |
11007 | return -EIO; | |
11008 | } | |
11009 | bl.clear(); | |
11010 | { | |
11011 | r = db->get(PREFIX_SUPER, "min_compat_ondisk_format", &bl); | |
11fdf7f2 TL |
11012 | ceph_assert(!r); |
11013 | auto p = bl.cbegin(); | |
7c673cae | 11014 | try { |
11fdf7f2 | 11015 | decode(compat_ondisk_format, p); |
7c673cae FG |
11016 | } catch (buffer::error& e) { |
11017 | derr << __func__ << " unable to read compat_ondisk_format" << dendl; | |
11018 | return -EIO; | |
11019 | } | |
11020 | } | |
11021 | } | |
11022 | dout(10) << __func__ << " ondisk_format " << ondisk_format | |
11023 | << " compat_ondisk_format " << compat_ondisk_format | |
11024 | << dendl; | |
11025 | } | |
11026 | ||
11027 | if (latest_ondisk_format < compat_ondisk_format) { | |
11028 | derr << __func__ << " compat_ondisk_format is " | |
11029 | << compat_ondisk_format << " but we only understand version " | |
11030 | << latest_ondisk_format << dendl; | |
11031 | return -EPERM; | |
11032 | } | |
7c673cae FG |
11033 | |
11034 | { | |
11035 | bufferlist bl; | |
11036 | db->get(PREFIX_SUPER, "min_alloc_size", &bl); | |
11fdf7f2 | 11037 | auto p = bl.cbegin(); |
7c673cae FG |
11038 | try { |
11039 | uint64_t val; | |
11fdf7f2 | 11040 | decode(val, p); |
7c673cae | 11041 | min_alloc_size = val; |
224ce89b | 11042 | min_alloc_size_order = ctz(val); |
11fdf7f2 | 11043 | ceph_assert(min_alloc_size == 1u << min_alloc_size_order); |
7c673cae FG |
11044 | } catch (buffer::error& e) { |
11045 | derr << __func__ << " unable to read min_alloc_size" << dendl; | |
11046 | return -EIO; | |
11047 | } | |
11048 | dout(10) << __func__ << " min_alloc_size 0x" << std::hex << min_alloc_size | |
11049 | << std::dec << dendl; | |
11050 | } | |
9f95a23c TL |
11051 | |
11052 | _set_per_pool_omap(); | |
11053 | ||
224ce89b | 11054 | _open_statfs(); |
7c673cae FG |
11055 | _set_alloc_sizes(); |
11056 | _set_throttle_params(); | |
11057 | ||
11058 | _set_csum(); | |
11059 | _set_compression(); | |
11060 | _set_blob_size(); | |
11061 | ||
11fdf7f2 | 11062 | _validate_bdev(); |
7c673cae FG |
11063 | return 0; |
11064 | } | |
11065 | ||
11066 | int BlueStore::_upgrade_super() | |
11067 | { | |
11068 | dout(1) << __func__ << " from " << ondisk_format << ", latest " | |
11069 | << latest_ondisk_format << dendl; | |
11fdf7f2 TL |
11070 | if (ondisk_format < latest_ondisk_format) { |
11071 | ceph_assert(ondisk_format > 0); | |
11072 | ceph_assert(ondisk_format < latest_ondisk_format); | |
11073 | ||
1911f103 | 11074 | KeyValueDB::Transaction t = db->get_transaction(); |
11fdf7f2 TL |
11075 | if (ondisk_format == 1) { |
11076 | // changes: | |
11077 | // - super: added ondisk_format | |
11078 | // - super: added min_readable_ondisk_format | |
11079 | // - super: added min_compat_ondisk_format | |
11080 | // - super: added min_alloc_size | |
11081 | // - super: removed min_min_alloc_size | |
11fdf7f2 TL |
11082 | { |
11083 | bufferlist bl; | |
11084 | db->get(PREFIX_SUPER, "min_min_alloc_size", &bl); | |
11085 | auto p = bl.cbegin(); | |
11086 | try { | |
11087 | uint64_t val; | |
11088 | decode(val, p); | |
11089 | min_alloc_size = val; | |
11090 | } catch (buffer::error& e) { | |
11091 | derr << __func__ << " failed to read min_min_alloc_size" << dendl; | |
11092 | return -EIO; | |
11093 | } | |
11094 | t->set(PREFIX_SUPER, "min_alloc_size", bl); | |
11095 | t->rmkey(PREFIX_SUPER, "min_min_alloc_size"); | |
7c673cae | 11096 | } |
11fdf7f2 | 11097 | ondisk_format = 2; |
7c673cae | 11098 | } |
9f95a23c TL |
11099 | if (ondisk_format == 2) { |
11100 | // changes: | |
11101 | // - onode has FLAG_PER_POOL_OMAP. Note that we do not know that *all* | |
11102 | // ondes are using the per-pool prefix until a repair is run; at that | |
11103 | // point the per_pool_omap=1 key will be set. | |
11104 | // - super: added per_pool_omap key, which indicates that *all* objects | |
11105 | // are using the new prefix and key format | |
11106 | ondisk_format = 3; | |
1911f103 TL |
11107 | } |
11108 | if (ondisk_format == 3) { | |
11109 | // changes: | |
11110 | // - FreelistManager keeps meta within bdev label | |
11111 | int r = _write_out_fm_meta(0); | |
9f95a23c | 11112 | ceph_assert(r == 0); |
1911f103 | 11113 | ondisk_format = 4; |
9f95a23c | 11114 | } |
1911f103 TL |
11115 | // This to be the last operation |
11116 | _prepare_ondisk_format_super(t); | |
11117 | int r = db->submit_transaction_sync(t); | |
11118 | ceph_assert(r == 0); | |
7c673cae | 11119 | } |
7c673cae FG |
11120 | // done |
11121 | dout(1) << __func__ << " done" << dendl; | |
11122 | return 0; | |
11123 | } | |
11124 | ||
11125 | void BlueStore::_assign_nid(TransContext *txc, OnodeRef o) | |
11126 | { | |
224ce89b | 11127 | if (o->onode.nid) { |
11fdf7f2 | 11128 | ceph_assert(o->exists); |
7c673cae | 11129 | return; |
224ce89b | 11130 | } |
7c673cae FG |
11131 | uint64_t nid = ++nid_last; |
11132 | dout(20) << __func__ << " " << nid << dendl; | |
11133 | o->onode.nid = nid; | |
11134 | txc->last_nid = nid; | |
224ce89b | 11135 | o->exists = true; |
7c673cae FG |
11136 | } |
11137 | ||
11138 | uint64_t BlueStore::_assign_blobid(TransContext *txc) | |
11139 | { | |
11140 | uint64_t bid = ++blobid_last; | |
11141 | dout(20) << __func__ << " " << bid << dendl; | |
11142 | txc->last_blobid = bid; | |
11143 | return bid; | |
11144 | } | |
11145 | ||
11146 | void BlueStore::get_db_statistics(Formatter *f) | |
11147 | { | |
11148 | db->get_statistics(f); | |
11149 | } | |
11150 | ||
11fdf7f2 TL |
11151 | BlueStore::TransContext *BlueStore::_txc_create( |
11152 | Collection *c, OpSequencer *osr, | |
11153 | list<Context*> *on_commits) | |
7c673cae | 11154 | { |
11fdf7f2 | 11155 | TransContext *txc = new TransContext(cct, c, osr, on_commits); |
7c673cae FG |
11156 | txc->t = db->get_transaction(); |
11157 | osr->queue_new(txc); | |
11158 | dout(20) << __func__ << " osr " << osr << " = " << txc | |
11159 | << " seq " << txc->seq << dendl; | |
11160 | return txc; | |
11161 | } | |
11162 | ||
11163 | void BlueStore::_txc_calc_cost(TransContext *txc) | |
11164 | { | |
11fdf7f2 TL |
11165 | // one "io" for the kv commit |
11166 | auto ios = 1 + txc->ioc.get_num_ios(); | |
7c673cae FG |
11167 | auto cost = throttle_cost_per_io.load(); |
11168 | txc->cost = ios * cost + txc->bytes; | |
9f95a23c | 11169 | txc->ios = ios; |
7c673cae FG |
11170 | dout(10) << __func__ << " " << txc << " cost " << txc->cost << " (" |
11171 | << ios << " ios * " << cost << " + " << txc->bytes | |
11172 | << " bytes)" << dendl; | |
11173 | } | |
11174 | ||
11175 | void BlueStore::_txc_update_store_statfs(TransContext *txc) | |
11176 | { | |
11177 | if (txc->statfs_delta.is_empty()) | |
11178 | return; | |
11179 | ||
11180 | logger->inc(l_bluestore_allocated, txc->statfs_delta.allocated()); | |
11181 | logger->inc(l_bluestore_stored, txc->statfs_delta.stored()); | |
11182 | logger->inc(l_bluestore_compressed, txc->statfs_delta.compressed()); | |
11183 | logger->inc(l_bluestore_compressed_allocated, txc->statfs_delta.compressed_allocated()); | |
11184 | logger->inc(l_bluestore_compressed_original, txc->statfs_delta.compressed_original()); | |
11185 | ||
11186 | bufferlist bl; | |
11187 | txc->statfs_delta.encode(bl); | |
11fdf7f2 TL |
11188 | if (per_pool_stat_collection) { |
11189 | string key; | |
11190 | get_pool_stat_key(txc->osd_pool_id, &key); | |
11191 | txc->t->merge(PREFIX_STAT, key, bl); | |
11192 | ||
11193 | std::lock_guard l(vstatfs_lock); | |
11194 | auto& stats = osd_pools[txc->osd_pool_id]; | |
11195 | stats += txc->statfs_delta; | |
11196 | ||
11197 | vstatfs += txc->statfs_delta; //non-persistent in this mode | |
11198 | ||
11199 | } else { | |
11200 | txc->t->merge(PREFIX_STAT, BLUESTORE_GLOBAL_STATFS_KEY, bl); | |
7c673cae | 11201 | |
11fdf7f2 TL |
11202 | std::lock_guard l(vstatfs_lock); |
11203 | vstatfs += txc->statfs_delta; | |
11204 | } | |
7c673cae FG |
11205 | txc->statfs_delta.reset(); |
11206 | } | |
11207 | ||
11208 | void BlueStore::_txc_state_proc(TransContext *txc) | |
11209 | { | |
11210 | while (true) { | |
11211 | dout(10) << __func__ << " txc " << txc | |
11212 | << " " << txc->get_state_name() << dendl; | |
11213 | switch (txc->state) { | |
11214 | case TransContext::STATE_PREPARE: | |
9f95a23c | 11215 | throttle.log_state_latency(*txc, logger, l_bluestore_state_prepare_lat); |
7c673cae FG |
11216 | if (txc->ioc.has_pending_aios()) { |
11217 | txc->state = TransContext::STATE_AIO_WAIT; | |
11218 | txc->had_ios = true; | |
11219 | _txc_aio_submit(txc); | |
11220 | return; | |
11221 | } | |
11222 | // ** fall-thru ** | |
11223 | ||
11224 | case TransContext::STATE_AIO_WAIT: | |
11fdf7f2 | 11225 | { |
9f95a23c TL |
11226 | mono_clock::duration lat = throttle.log_state_latency( |
11227 | *txc, logger, l_bluestore_state_aio_wait_lat); | |
11228 | if (ceph::to_seconds<double>(lat) >= cct->_conf->bluestore_log_op_age) { | |
11fdf7f2 TL |
11229 | dout(0) << __func__ << " slow aio_wait, txc = " << txc |
11230 | << ", latency = " << lat | |
11231 | << dendl; | |
11232 | } | |
11233 | } | |
11234 | ||
7c673cae FG |
11235 | _txc_finish_io(txc); // may trigger blocked txc's too |
11236 | return; | |
11237 | ||
11238 | case TransContext::STATE_IO_DONE: | |
11fdf7f2 | 11239 | ceph_assert(ceph_mutex_is_locked(txc->osr->qlock)); // see _txc_finish_io |
7c673cae FG |
11240 | if (txc->had_ios) { |
11241 | ++txc->osr->txc_with_unstable_io; | |
11242 | } | |
9f95a23c | 11243 | throttle.log_state_latency(*txc, logger, l_bluestore_state_io_done_lat); |
7c673cae FG |
11244 | txc->state = TransContext::STATE_KV_QUEUED; |
11245 | if (cct->_conf->bluestore_sync_submit_transaction) { | |
11246 | if (txc->last_nid >= nid_max || | |
11247 | txc->last_blobid >= blobid_max) { | |
11248 | dout(20) << __func__ | |
11249 | << " last_{nid,blobid} exceeds max, submit via kv thread" | |
11250 | << dendl; | |
11251 | } else if (txc->osr->kv_committing_serially) { | |
11252 | dout(20) << __func__ << " prior txc submitted via kv thread, us too" | |
11253 | << dendl; | |
11254 | // note: this is starvation-prone. once we have a txc in a busy | |
11255 | // sequencer that is committing serially it is possible to keep | |
11256 | // submitting new transactions fast enough that we get stuck doing | |
11257 | // so. the alternative is to block here... fixme? | |
11258 | } else if (txc->osr->txc_with_unstable_io) { | |
11259 | dout(20) << __func__ << " prior txc(s) with unstable ios " | |
11260 | << txc->osr->txc_with_unstable_io.load() << dendl; | |
11261 | } else if (cct->_conf->bluestore_debug_randomize_serial_transaction && | |
11262 | rand() % cct->_conf->bluestore_debug_randomize_serial_transaction | |
11263 | == 0) { | |
11264 | dout(20) << __func__ << " DEBUG randomly forcing submit via kv thread" | |
11265 | << dendl; | |
11266 | } else { | |
9f95a23c | 11267 | _txc_apply_kv(txc, true); |
7c673cae FG |
11268 | } |
11269 | } | |
11270 | { | |
11fdf7f2 | 11271 | std::lock_guard l(kv_lock); |
7c673cae | 11272 | kv_queue.push_back(txc); |
9f95a23c TL |
11273 | if (!kv_sync_in_progress) { |
11274 | kv_sync_in_progress = true; | |
11275 | kv_cond.notify_one(); | |
11276 | } | |
7c673cae FG |
11277 | if (txc->state != TransContext::STATE_KV_SUBMITTED) { |
11278 | kv_queue_unsubmitted.push_back(txc); | |
11279 | ++txc->osr->kv_committing_serially; | |
11280 | } | |
31f18b77 FG |
11281 | if (txc->had_ios) |
11282 | kv_ios++; | |
11283 | kv_throttle_costs += txc->cost; | |
7c673cae FG |
11284 | } |
11285 | return; | |
11286 | case TransContext::STATE_KV_SUBMITTED: | |
7c673cae FG |
11287 | _txc_committed_kv(txc); |
11288 | // ** fall-thru ** | |
11289 | ||
11290 | case TransContext::STATE_KV_DONE: | |
9f95a23c | 11291 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_done_lat); |
7c673cae FG |
11292 | if (txc->deferred_txn) { |
11293 | txc->state = TransContext::STATE_DEFERRED_QUEUED; | |
11294 | _deferred_queue(txc); | |
11295 | return; | |
11296 | } | |
11297 | txc->state = TransContext::STATE_FINISHING; | |
11298 | break; | |
11299 | ||
11300 | case TransContext::STATE_DEFERRED_CLEANUP: | |
9f95a23c | 11301 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_cleanup_lat); |
7c673cae FG |
11302 | txc->state = TransContext::STATE_FINISHING; |
11303 | // ** fall-thru ** | |
11304 | ||
11305 | case TransContext::STATE_FINISHING: | |
9f95a23c | 11306 | throttle.log_state_latency(*txc, logger, l_bluestore_state_finishing_lat); |
7c673cae FG |
11307 | _txc_finish(txc); |
11308 | return; | |
11309 | ||
11310 | default: | |
11311 | derr << __func__ << " unexpected txc " << txc | |
11312 | << " state " << txc->get_state_name() << dendl; | |
11fdf7f2 | 11313 | ceph_abort_msg("unexpected txc state"); |
7c673cae FG |
11314 | return; |
11315 | } | |
11316 | } | |
11317 | } | |
11318 | ||
11319 | void BlueStore::_txc_finish_io(TransContext *txc) | |
11320 | { | |
11321 | dout(20) << __func__ << " " << txc << dendl; | |
11322 | ||
11323 | /* | |
11324 | * we need to preserve the order of kv transactions, | |
11325 | * even though aio will complete in any order. | |
11326 | */ | |
11327 | ||
11328 | OpSequencer *osr = txc->osr.get(); | |
11fdf7f2 | 11329 | std::lock_guard l(osr->qlock); |
7c673cae | 11330 | txc->state = TransContext::STATE_IO_DONE; |
11fdf7f2 | 11331 | txc->ioc.release_running_aios(); |
7c673cae FG |
11332 | OpSequencer::q_list_t::iterator p = osr->q.iterator_to(*txc); |
11333 | while (p != osr->q.begin()) { | |
11334 | --p; | |
11335 | if (p->state < TransContext::STATE_IO_DONE) { | |
11336 | dout(20) << __func__ << " " << txc << " blocked by " << &*p << " " | |
11337 | << p->get_state_name() << dendl; | |
11338 | return; | |
11339 | } | |
11340 | if (p->state > TransContext::STATE_IO_DONE) { | |
11341 | ++p; | |
11342 | break; | |
11343 | } | |
11344 | } | |
11345 | do { | |
11346 | _txc_state_proc(&*p++); | |
11347 | } while (p != osr->q.end() && | |
11348 | p->state == TransContext::STATE_IO_DONE); | |
11349 | ||
11fdf7f2 | 11350 | if (osr->kv_submitted_waiters) { |
7c673cae FG |
11351 | osr->qcond.notify_all(); |
11352 | } | |
11353 | } | |
11354 | ||
11355 | void BlueStore::_txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t) | |
11356 | { | |
11357 | dout(20) << __func__ << " txc " << txc | |
11358 | << " onodes " << txc->onodes | |
11359 | << " shared_blobs " << txc->shared_blobs | |
11360 | << dendl; | |
11361 | ||
11362 | // finalize onodes | |
11363 | for (auto o : txc->onodes) { | |
11fdf7f2 | 11364 | _record_onode(o, t); |
7c673cae FG |
11365 | o->flushing_count++; |
11366 | } | |
11367 | ||
11368 | // objects we modified but didn't affect the onode | |
11369 | auto p = txc->modified_objects.begin(); | |
11370 | while (p != txc->modified_objects.end()) { | |
11371 | if (txc->onodes.count(*p) == 0) { | |
11372 | (*p)->flushing_count++; | |
11373 | ++p; | |
11374 | } else { | |
11375 | // remove dups with onodes list to avoid problems in _txc_finish | |
11376 | p = txc->modified_objects.erase(p); | |
11377 | } | |
11378 | } | |
11379 | ||
11380 | // finalize shared_blobs | |
11381 | for (auto sb : txc->shared_blobs) { | |
11382 | string key; | |
11383 | auto sbid = sb->get_sbid(); | |
11384 | get_shared_blob_key(sbid, &key); | |
11385 | if (sb->persistent->empty()) { | |
11fdf7f2 TL |
11386 | dout(20) << __func__ << " shared_blob 0x" |
11387 | << std::hex << sbid << std::dec | |
7c673cae FG |
11388 | << " is empty" << dendl; |
11389 | t->rmkey(PREFIX_SHARED_BLOB, key); | |
11390 | } else { | |
11391 | bufferlist bl; | |
11fdf7f2 TL |
11392 | encode(*(sb->persistent), bl); |
11393 | dout(20) << __func__ << " shared_blob 0x" | |
11394 | << std::hex << sbid << std::dec | |
31f18b77 | 11395 | << " is " << bl.length() << " " << *sb << dendl; |
7c673cae FG |
11396 | t->set(PREFIX_SHARED_BLOB, key, bl); |
11397 | } | |
11398 | } | |
11399 | } | |
11400 | ||
11401 | void BlueStore::BSPerfTracker::update_from_perfcounters( | |
11402 | PerfCounters &logger) | |
11403 | { | |
11fdf7f2 TL |
11404 | os_commit_latency_ns.consume_next( |
11405 | logger.get_tavg_ns( | |
7c673cae | 11406 | l_bluestore_commit_lat)); |
11fdf7f2 TL |
11407 | os_apply_latency_ns.consume_next( |
11408 | logger.get_tavg_ns( | |
7c673cae FG |
11409 | l_bluestore_commit_lat)); |
11410 | } | |
11411 | ||
11412 | void BlueStore::_txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t) | |
11413 | { | |
11414 | dout(20) << __func__ << " txc " << txc << std::hex | |
11415 | << " allocated 0x" << txc->allocated | |
11416 | << " released 0x" << txc->released | |
11417 | << std::dec << dendl; | |
11418 | ||
11419 | // We have to handle the case where we allocate *and* deallocate the | |
11420 | // same region in this transaction. The freelist doesn't like that. | |
11421 | // (Actually, the only thing that cares is the BitmapFreelistManager | |
11422 | // debug check. But that's important.) | |
11423 | interval_set<uint64_t> tmp_allocated, tmp_released; | |
11424 | interval_set<uint64_t> *pallocated = &txc->allocated; | |
11425 | interval_set<uint64_t> *preleased = &txc->released; | |
11426 | if (!txc->allocated.empty() && !txc->released.empty()) { | |
11427 | interval_set<uint64_t> overlap; | |
11428 | overlap.intersection_of(txc->allocated, txc->released); | |
11429 | if (!overlap.empty()) { | |
11430 | tmp_allocated = txc->allocated; | |
11431 | tmp_allocated.subtract(overlap); | |
11432 | tmp_released = txc->released; | |
11433 | tmp_released.subtract(overlap); | |
11434 | dout(20) << __func__ << " overlap 0x" << std::hex << overlap | |
11435 | << ", new allocated 0x" << tmp_allocated | |
11436 | << " released 0x" << tmp_released << std::dec | |
11437 | << dendl; | |
11438 | pallocated = &tmp_allocated; | |
11439 | preleased = &tmp_released; | |
11440 | } | |
11441 | } | |
11442 | ||
11443 | // update freelist with non-overlap sets | |
11444 | for (interval_set<uint64_t>::iterator p = pallocated->begin(); | |
11445 | p != pallocated->end(); | |
11446 | ++p) { | |
11447 | fm->allocate(p.get_start(), p.get_len(), t); | |
11448 | } | |
11449 | for (interval_set<uint64_t>::iterator p = preleased->begin(); | |
11450 | p != preleased->end(); | |
11451 | ++p) { | |
11452 | dout(20) << __func__ << " release 0x" << std::hex << p.get_start() | |
11453 | << "~" << p.get_len() << std::dec << dendl; | |
11454 | fm->release(p.get_start(), p.get_len(), t); | |
11455 | } | |
11456 | ||
11457 | _txc_update_store_statfs(txc); | |
11458 | } | |
11459 | ||
9f95a23c | 11460 | void BlueStore::_txc_apply_kv(TransContext *txc, bool sync_submit_transaction) |
7c673cae | 11461 | { |
9f95a23c TL |
11462 | ceph_assert(txc->state == TransContext::STATE_KV_QUEUED); |
11463 | { | |
11464 | #if defined(WITH_LTTNG) | |
11465 | auto start = mono_clock::now(); | |
11466 | #endif | |
11467 | ||
11468 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction(txc->t); | |
11469 | ceph_assert(r == 0); | |
11470 | txc->state = TransContext::STATE_KV_SUBMITTED; | |
11471 | if (txc->osr->kv_submitted_waiters) { | |
11472 | std::lock_guard l(txc->osr->qlock); | |
11473 | txc->osr->qcond.notify_all(); | |
11474 | } | |
11475 | ||
11476 | #if defined(WITH_LTTNG) | |
11477 | if (txc->tracing) { | |
11478 | tracepoint( | |
11479 | bluestore, | |
11480 | transaction_kv_submit_latency, | |
11481 | txc->osr->get_sequencer_id(), | |
11482 | txc->seq, | |
11483 | sync_submit_transaction, | |
11484 | ceph::to_seconds<double>(mono_clock::now() - start)); | |
11485 | } | |
11486 | #endif | |
11487 | } | |
11488 | ||
7c673cae FG |
11489 | for (auto ls : { &txc->onodes, &txc->modified_objects }) { |
11490 | for (auto& o : *ls) { | |
11491 | dout(20) << __func__ << " onode " << o << " had " << o->flushing_count | |
11492 | << dendl; | |
9f95a23c | 11493 | if (--o->flushing_count == 0 && o->waiting_count.load()) { |
11fdf7f2 | 11494 | std::lock_guard l(o->flush_lock); |
7c673cae FG |
11495 | o->flush_cond.notify_all(); |
11496 | } | |
11497 | } | |
11498 | } | |
11499 | } | |
11500 | ||
11501 | void BlueStore::_txc_committed_kv(TransContext *txc) | |
11502 | { | |
11503 | dout(20) << __func__ << " txc " << txc << dendl; | |
9f95a23c | 11504 | throttle.complete_kv(*txc); |
1adf2230 | 11505 | { |
11fdf7f2 | 11506 | std::lock_guard l(txc->osr->qlock); |
1adf2230 | 11507 | txc->state = TransContext::STATE_KV_DONE; |
11fdf7f2 TL |
11508 | if (txc->ch->commit_queue) { |
11509 | txc->ch->commit_queue->queue(txc->oncommits); | |
11510 | } else { | |
11511 | finisher.queue(txc->oncommits); | |
1adf2230 | 11512 | } |
7c673cae | 11513 | } |
9f95a23c | 11514 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_committing_lat); |
494da23a TL |
11515 | log_latency_fn( |
11516 | __func__, | |
11517 | l_bluestore_commit_lat, | |
9f95a23c | 11518 | mono_clock::now() - txc->start, |
494da23a TL |
11519 | cct->_conf->bluestore_log_op_age, |
11520 | [&](auto lat) { | |
11521 | return ", txc = " + stringify(txc); | |
11522 | } | |
11fdf7f2 | 11523 | ); |
7c673cae FG |
11524 | } |
11525 | ||
11526 | void BlueStore::_txc_finish(TransContext *txc) | |
11527 | { | |
11528 | dout(20) << __func__ << " " << txc << " onodes " << txc->onodes << dendl; | |
11fdf7f2 | 11529 | ceph_assert(txc->state == TransContext::STATE_FINISHING); |
7c673cae FG |
11530 | |
11531 | for (auto& sb : txc->shared_blobs_written) { | |
f64942e4 | 11532 | sb->finish_write(txc->seq); |
7c673cae FG |
11533 | } |
11534 | txc->shared_blobs_written.clear(); | |
11535 | ||
11536 | while (!txc->removed_collections.empty()) { | |
11537 | _queue_reap_collection(txc->removed_collections.front()); | |
11538 | txc->removed_collections.pop_front(); | |
11539 | } | |
11540 | ||
11541 | OpSequencerRef osr = txc->osr; | |
7c673cae | 11542 | bool empty = false; |
31f18b77 | 11543 | bool submit_deferred = false; |
7c673cae FG |
11544 | OpSequencer::q_list_t releasing_txc; |
11545 | { | |
11fdf7f2 | 11546 | std::lock_guard l(osr->qlock); |
7c673cae FG |
11547 | txc->state = TransContext::STATE_DONE; |
11548 | bool notify = false; | |
11549 | while (!osr->q.empty()) { | |
11550 | TransContext *txc = &osr->q.front(); | |
11551 | dout(20) << __func__ << " txc " << txc << " " << txc->get_state_name() | |
11552 | << dendl; | |
11553 | if (txc->state != TransContext::STATE_DONE) { | |
11554 | if (txc->state == TransContext::STATE_PREPARE && | |
11555 | deferred_aggressive) { | |
11556 | // for _osr_drain_preceding() | |
11557 | notify = true; | |
11558 | } | |
31f18b77 | 11559 | if (txc->state == TransContext::STATE_DEFERRED_QUEUED && |
11fdf7f2 | 11560 | osr->q.size() > g_conf()->bluestore_max_deferred_txc) { |
31f18b77 FG |
11561 | submit_deferred = true; |
11562 | } | |
7c673cae FG |
11563 | break; |
11564 | } | |
11565 | ||
7c673cae FG |
11566 | osr->q.pop_front(); |
11567 | releasing_txc.push_back(*txc); | |
7c673cae | 11568 | } |
9f95a23c | 11569 | |
7c673cae FG |
11570 | if (osr->q.empty()) { |
11571 | dout(20) << __func__ << " osr " << osr << " q now empty" << dendl; | |
11572 | empty = true; | |
11573 | } | |
9f95a23c TL |
11574 | |
11575 | // only drain()/drain_preceding() need wakeup, | |
11576 | // other cases use kv_submitted_waiters | |
11577 | if (notify || empty) { | |
11578 | osr->qcond.notify_all(); | |
11579 | } | |
7c673cae | 11580 | } |
9f95a23c | 11581 | |
7c673cae FG |
11582 | while (!releasing_txc.empty()) { |
11583 | // release to allocator only after all preceding txc's have also | |
11584 | // finished any deferred writes that potentially land in these | |
11585 | // blocks | |
11586 | auto txc = &releasing_txc.front(); | |
11587 | _txc_release_alloc(txc); | |
11588 | releasing_txc.pop_front(); | |
9f95a23c TL |
11589 | throttle.log_state_latency(*txc, logger, l_bluestore_state_done_lat); |
11590 | throttle.complete(*txc); | |
7c673cae FG |
11591 | delete txc; |
11592 | } | |
11593 | ||
31f18b77 FG |
11594 | if (submit_deferred) { |
11595 | // we're pinning memory; flush! we could be more fine-grained here but | |
11596 | // i'm not sure it's worth the bother. | |
11597 | deferred_try_submit(); | |
7c673cae FG |
11598 | } |
11599 | ||
7c673cae | 11600 | if (empty && osr->zombie) { |
11fdf7f2 TL |
11601 | std::lock_guard l(zombie_osr_lock); |
11602 | if (zombie_osr_set.erase(osr->cid)) { | |
11603 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11604 | } else { | |
11605 | dout(10) << __func__ << " empty zombie osr " << osr << " already reaped" | |
11606 | << dendl; | |
11607 | } | |
7c673cae | 11608 | } |
9f95a23c | 11609 | } |
7c673cae FG |
11610 | |
11611 | void BlueStore::_txc_release_alloc(TransContext *txc) | |
11612 | { | |
a8e16298 | 11613 | // it's expected we're called with lazy_release_lock already taken! |
11fdf7f2 TL |
11614 | if (likely(!cct->_conf->bluestore_debug_no_reuse_blocks)) { |
11615 | int r = 0; | |
11616 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
11617 | r = bdev->queue_discard(txc->released); | |
11618 | if (r == 0) { | |
11619 | dout(10) << __func__ << "(queued) " << txc << " " << std::hex | |
11620 | << txc->released << std::dec << dendl; | |
11621 | goto out; | |
11622 | } | |
11623 | } else if (cct->_conf->bdev_enable_discard) { | |
11624 | for (auto p = txc->released.begin(); p != txc->released.end(); ++p) { | |
11625 | bdev->discard(p.get_start(), p.get_len()); | |
11626 | } | |
11627 | } | |
11628 | dout(10) << __func__ << "(sync) " << txc << " " << std::hex | |
94b18763 | 11629 | << txc->released << std::dec << dendl; |
11fdf7f2 | 11630 | alloc->release(txc->released); |
7c673cae FG |
11631 | } |
11632 | ||
11fdf7f2 | 11633 | out: |
7c673cae FG |
11634 | txc->allocated.clear(); |
11635 | txc->released.clear(); | |
11636 | } | |
11637 | ||
11fdf7f2 TL |
11638 | void BlueStore::_osr_attach(Collection *c) |
11639 | { | |
11640 | // note: caller has RWLock on coll_map | |
11641 | auto q = coll_map.find(c->cid); | |
11642 | if (q != coll_map.end()) { | |
11643 | c->osr = q->second->osr; | |
11644 | ldout(cct, 10) << __func__ << " " << c->cid | |
11645 | << " reusing osr " << c->osr << " from existing coll " | |
11646 | << q->second << dendl; | |
11647 | } else { | |
11648 | std::lock_guard l(zombie_osr_lock); | |
11649 | auto p = zombie_osr_set.find(c->cid); | |
11650 | if (p == zombie_osr_set.end()) { | |
9f95a23c | 11651 | c->osr = ceph::make_ref<OpSequencer>(this, next_sequencer_id++, c->cid); |
11fdf7f2 TL |
11652 | ldout(cct, 10) << __func__ << " " << c->cid |
11653 | << " fresh osr " << c->osr << dendl; | |
11654 | } else { | |
11655 | c->osr = p->second; | |
11656 | zombie_osr_set.erase(p); | |
11657 | ldout(cct, 10) << __func__ << " " << c->cid | |
11658 | << " resurrecting zombie osr " << c->osr << dendl; | |
11659 | c->osr->zombie = false; | |
11660 | } | |
11661 | } | |
11662 | } | |
11663 | ||
11664 | void BlueStore::_osr_register_zombie(OpSequencer *osr) | |
11665 | { | |
11666 | std::lock_guard l(zombie_osr_lock); | |
11667 | dout(10) << __func__ << " " << osr << " " << osr->cid << dendl; | |
11668 | osr->zombie = true; | |
11669 | auto i = zombie_osr_set.emplace(osr->cid, osr); | |
11670 | // this is either a new insertion or the same osr is already there | |
11671 | ceph_assert(i.second || i.first->second == osr); | |
11672 | } | |
11673 | ||
7c673cae FG |
11674 | void BlueStore::_osr_drain_preceding(TransContext *txc) |
11675 | { | |
11676 | OpSequencer *osr = txc->osr.get(); | |
11677 | dout(10) << __func__ << " " << txc << " osr " << osr << dendl; | |
11678 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11679 | { | |
11680 | // submit anything pending | |
224ce89b | 11681 | deferred_lock.lock(); |
11fdf7f2 | 11682 | if (osr->deferred_pending && !osr->deferred_running) { |
224ce89b WB |
11683 | _deferred_submit_unlock(osr); |
11684 | } else { | |
11685 | deferred_lock.unlock(); | |
7c673cae FG |
11686 | } |
11687 | } | |
11688 | { | |
11689 | // wake up any previously finished deferred events | |
11fdf7f2 | 11690 | std::lock_guard l(kv_lock); |
9f95a23c TL |
11691 | if (!kv_sync_in_progress) { |
11692 | kv_sync_in_progress = true; | |
11693 | kv_cond.notify_one(); | |
11694 | } | |
7c673cae FG |
11695 | } |
11696 | osr->drain_preceding(txc); | |
11697 | --deferred_aggressive; | |
11698 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11699 | } | |
11700 | ||
11fdf7f2 TL |
11701 | void BlueStore::_osr_drain(OpSequencer *osr) |
11702 | { | |
11703 | dout(10) << __func__ << " " << osr << dendl; | |
11704 | ++deferred_aggressive; // FIXME: maybe osr-local aggressive flag? | |
11705 | { | |
11706 | // submit anything pending | |
11707 | deferred_lock.lock(); | |
11708 | if (osr->deferred_pending && !osr->deferred_running) { | |
11709 | _deferred_submit_unlock(osr); | |
11710 | } else { | |
11711 | deferred_lock.unlock(); | |
11712 | } | |
11713 | } | |
11714 | { | |
11715 | // wake up any previously finished deferred events | |
11716 | std::lock_guard l(kv_lock); | |
9f95a23c TL |
11717 | if (!kv_sync_in_progress) { |
11718 | kv_sync_in_progress = true; | |
11719 | kv_cond.notify_one(); | |
11720 | } | |
11fdf7f2 TL |
11721 | } |
11722 | osr->drain(); | |
11723 | --deferred_aggressive; | |
11724 | dout(10) << __func__ << " " << osr << " done" << dendl; | |
11725 | } | |
11726 | ||
7c673cae FG |
11727 | void BlueStore::_osr_drain_all() |
11728 | { | |
11729 | dout(10) << __func__ << dendl; | |
11730 | ||
11731 | set<OpSequencerRef> s; | |
11fdf7f2 TL |
11732 | vector<OpSequencerRef> zombies; |
11733 | { | |
9f95a23c | 11734 | std::shared_lock l(coll_lock); |
11fdf7f2 TL |
11735 | for (auto& i : coll_map) { |
11736 | s.insert(i.second->osr); | |
11737 | } | |
11738 | } | |
7c673cae | 11739 | { |
11fdf7f2 TL |
11740 | std::lock_guard l(zombie_osr_lock); |
11741 | for (auto& i : zombie_osr_set) { | |
11742 | s.insert(i.second); | |
11743 | zombies.push_back(i.second); | |
11744 | } | |
7c673cae FG |
11745 | } |
11746 | dout(20) << __func__ << " osr_set " << s << dendl; | |
11747 | ||
11748 | ++deferred_aggressive; | |
11749 | { | |
11750 | // submit anything pending | |
224ce89b | 11751 | deferred_try_submit(); |
7c673cae FG |
11752 | } |
11753 | { | |
11754 | // wake up any previously finished deferred events | |
11fdf7f2 | 11755 | std::lock_guard l(kv_lock); |
7c673cae FG |
11756 | kv_cond.notify_one(); |
11757 | } | |
31f18b77 | 11758 | { |
11fdf7f2 | 11759 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11760 | kv_finalize_cond.notify_one(); |
11761 | } | |
7c673cae FG |
11762 | for (auto osr : s) { |
11763 | dout(20) << __func__ << " drain " << osr << dendl; | |
11764 | osr->drain(); | |
11765 | } | |
11766 | --deferred_aggressive; | |
11767 | ||
7c673cae | 11768 | { |
11fdf7f2 TL |
11769 | std::lock_guard l(zombie_osr_lock); |
11770 | for (auto& osr : zombies) { | |
11771 | if (zombie_osr_set.erase(osr->cid)) { | |
11772 | dout(10) << __func__ << " reaping empty zombie osr " << osr << dendl; | |
11773 | ceph_assert(osr->q.empty()); | |
11774 | } else if (osr->zombie) { | |
11775 | dout(10) << __func__ << " empty zombie osr " << osr | |
11776 | << " already reaped" << dendl; | |
11777 | ceph_assert(osr->q.empty()); | |
11778 | } else { | |
11779 | dout(10) << __func__ << " empty zombie osr " << osr | |
11780 | << " resurrected" << dendl; | |
11781 | } | |
7c673cae FG |
11782 | } |
11783 | } | |
11fdf7f2 TL |
11784 | |
11785 | dout(10) << __func__ << " done" << dendl; | |
7c673cae FG |
11786 | } |
11787 | ||
11fdf7f2 | 11788 | |
31f18b77 FG |
11789 | void BlueStore::_kv_start() |
11790 | { | |
11791 | dout(10) << __func__ << dendl; | |
11792 | ||
11fdf7f2 | 11793 | finisher.start(); |
31f18b77 FG |
11794 | kv_sync_thread.create("bstore_kv_sync"); |
11795 | kv_finalize_thread.create("bstore_kv_final"); | |
11796 | } | |
11797 | ||
11798 | void BlueStore::_kv_stop() | |
11799 | { | |
11800 | dout(10) << __func__ << dendl; | |
11801 | { | |
9f95a23c | 11802 | std::unique_lock l{kv_lock}; |
31f18b77 FG |
11803 | while (!kv_sync_started) { |
11804 | kv_cond.wait(l); | |
11805 | } | |
11806 | kv_stop = true; | |
11807 | kv_cond.notify_all(); | |
11808 | } | |
11809 | { | |
9f95a23c | 11810 | std::unique_lock l{kv_finalize_lock}; |
31f18b77 FG |
11811 | while (!kv_finalize_started) { |
11812 | kv_finalize_cond.wait(l); | |
11813 | } | |
11814 | kv_finalize_stop = true; | |
11815 | kv_finalize_cond.notify_all(); | |
11816 | } | |
11817 | kv_sync_thread.join(); | |
11818 | kv_finalize_thread.join(); | |
11fdf7f2 | 11819 | ceph_assert(removed_collections.empty()); |
31f18b77 | 11820 | { |
11fdf7f2 | 11821 | std::lock_guard l(kv_lock); |
31f18b77 FG |
11822 | kv_stop = false; |
11823 | } | |
11824 | { | |
11fdf7f2 | 11825 | std::lock_guard l(kv_finalize_lock); |
31f18b77 FG |
11826 | kv_finalize_stop = false; |
11827 | } | |
11828 | dout(10) << __func__ << " stopping finishers" << dendl; | |
11fdf7f2 TL |
11829 | finisher.wait_for_empty(); |
11830 | finisher.stop(); | |
31f18b77 FG |
11831 | dout(10) << __func__ << " stopped" << dendl; |
11832 | } | |
11833 | ||
7c673cae FG |
11834 | void BlueStore::_kv_sync_thread() |
11835 | { | |
11836 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 | 11837 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable |
9f95a23c | 11838 | std::unique_lock l{kv_lock}; |
11fdf7f2 | 11839 | ceph_assert(!kv_sync_started); |
31f18b77 FG |
11840 | kv_sync_started = true; |
11841 | kv_cond.notify_all(); | |
7c673cae | 11842 | while (true) { |
11fdf7f2 | 11843 | ceph_assert(kv_committing.empty()); |
7c673cae FG |
11844 | if (kv_queue.empty() && |
11845 | ((deferred_done_queue.empty() && deferred_stable_queue.empty()) || | |
11fdf7f2 | 11846 | !deferred_aggressive)) { |
7c673cae FG |
11847 | if (kv_stop) |
11848 | break; | |
11849 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 11850 | kv_sync_in_progress = false; |
11fdf7f2 | 11851 | kv_cond.wait(l); |
7c673cae FG |
11852 | dout(20) << __func__ << " wake" << dendl; |
11853 | } else { | |
11854 | deque<TransContext*> kv_submitting; | |
11855 | deque<DeferredBatch*> deferred_done, deferred_stable; | |
31f18b77 FG |
11856 | uint64_t aios = 0, costs = 0; |
11857 | ||
7c673cae FG |
11858 | dout(20) << __func__ << " committing " << kv_queue.size() |
11859 | << " submitting " << kv_queue_unsubmitted.size() | |
11860 | << " deferred done " << deferred_done_queue.size() | |
11861 | << " stable " << deferred_stable_queue.size() | |
11862 | << dendl; | |
11863 | kv_committing.swap(kv_queue); | |
11864 | kv_submitting.swap(kv_queue_unsubmitted); | |
11865 | deferred_done.swap(deferred_done_queue); | |
11866 | deferred_stable.swap(deferred_stable_queue); | |
31f18b77 FG |
11867 | aios = kv_ios; |
11868 | costs = kv_throttle_costs; | |
11869 | kv_ios = 0; | |
11870 | kv_throttle_costs = 0; | |
7c673cae FG |
11871 | l.unlock(); |
11872 | ||
11873 | dout(30) << __func__ << " committing " << kv_committing << dendl; | |
11874 | dout(30) << __func__ << " submitting " << kv_submitting << dendl; | |
11875 | dout(30) << __func__ << " deferred_done " << deferred_done << dendl; | |
11876 | dout(30) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
11877 | ||
11fdf7f2 TL |
11878 | auto start = mono_clock::now(); |
11879 | ||
7c673cae FG |
11880 | bool force_flush = false; |
11881 | // if bluefs is sharing the same device as data (only), then we | |
11882 | // can rely on the bluefs commit to flush the device and make | |
11883 | // deferred aios stable. that means that if we do have done deferred | |
11884 | // txcs AND we are not on a single device, we need to force a flush. | |
9f95a23c | 11885 | if (bluefs && bluefs_layout.single_shared_device()) { |
31f18b77 | 11886 | if (aios) { |
7c673cae | 11887 | force_flush = true; |
11fdf7f2 | 11888 | } else if (kv_committing.empty() && deferred_stable.empty()) { |
7c673cae FG |
11889 | force_flush = true; // there's nothing else to commit! |
11890 | } else if (deferred_aggressive) { | |
11891 | force_flush = true; | |
11892 | } | |
11fdf7f2 TL |
11893 | } else { |
11894 | if (aios || !deferred_done.empty()) { | |
11895 | force_flush = true; | |
11896 | } else { | |
11897 | dout(20) << __func__ << " skipping flush (no aios, no deferred_done)" << dendl; | |
11898 | } | |
11899 | } | |
7c673cae FG |
11900 | |
11901 | if (force_flush) { | |
31f18b77 | 11902 | dout(20) << __func__ << " num_aios=" << aios |
7c673cae FG |
11903 | << " force_flush=" << (int)force_flush |
11904 | << ", flushing, deferred done->stable" << dendl; | |
11905 | // flush/barrier on block device | |
11906 | bdev->flush(); | |
11907 | ||
11908 | // if we flush then deferred done are now deferred stable | |
11909 | deferred_stable.insert(deferred_stable.end(), deferred_done.begin(), | |
11910 | deferred_done.end()); | |
11911 | deferred_done.clear(); | |
11912 | } | |
11fdf7f2 | 11913 | auto after_flush = mono_clock::now(); |
7c673cae FG |
11914 | |
11915 | // we will use one final transaction to force a sync | |
11916 | KeyValueDB::Transaction synct = db->get_transaction(); | |
11917 | ||
11918 | // increase {nid,blobid}_max? note that this covers both the | |
11919 | // case where we are approaching the max and the case we passed | |
11920 | // it. in either case, we increase the max in the earlier txn | |
11921 | // we submit. | |
11922 | uint64_t new_nid_max = 0, new_blobid_max = 0; | |
11923 | if (nid_last + cct->_conf->bluestore_nid_prealloc/2 > nid_max) { | |
11924 | KeyValueDB::Transaction t = | |
11925 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
11926 | new_nid_max = nid_last + cct->_conf->bluestore_nid_prealloc; | |
11927 | bufferlist bl; | |
11fdf7f2 | 11928 | encode(new_nid_max, bl); |
7c673cae FG |
11929 | t->set(PREFIX_SUPER, "nid_max", bl); |
11930 | dout(10) << __func__ << " new_nid_max " << new_nid_max << dendl; | |
11931 | } | |
11932 | if (blobid_last + cct->_conf->bluestore_blobid_prealloc/2 > blobid_max) { | |
11933 | KeyValueDB::Transaction t = | |
11934 | kv_submitting.empty() ? synct : kv_submitting.front()->t; | |
11935 | new_blobid_max = blobid_last + cct->_conf->bluestore_blobid_prealloc; | |
11936 | bufferlist bl; | |
11fdf7f2 | 11937 | encode(new_blobid_max, bl); |
7c673cae FG |
11938 | t->set(PREFIX_SUPER, "blobid_max", bl); |
11939 | dout(10) << __func__ << " new_blobid_max " << new_blobid_max << dendl; | |
11940 | } | |
c07f9fc5 FG |
11941 | |
11942 | for (auto txc : kv_committing) { | |
9f95a23c | 11943 | throttle.log_state_latency(*txc, logger, l_bluestore_state_kv_queued_lat); |
c07f9fc5 | 11944 | if (txc->state == TransContext::STATE_KV_QUEUED) { |
9f95a23c | 11945 | _txc_apply_kv(txc, false); |
c07f9fc5 | 11946 | --txc->osr->kv_committing_serially; |
c07f9fc5 | 11947 | } else { |
11fdf7f2 | 11948 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 11949 | } |
7c673cae FG |
11950 | if (txc->had_ios) { |
11951 | --txc->osr->txc_with_unstable_io; | |
11952 | } | |
7c673cae FG |
11953 | } |
11954 | ||
31f18b77 FG |
11955 | // release throttle *before* we commit. this allows new ops |
11956 | // to be prepared and enter pipeline while we are waiting on | |
11957 | // the kv commit sync/flush. then hopefully on the next | |
11958 | // iteration there will already be ops awake. otherwise, we | |
11959 | // end up going to sleep, and then wake up when the very first | |
11960 | // transaction is ready for commit. | |
9f95a23c | 11961 | throttle.release_kv_throttle(costs); |
31f18b77 | 11962 | |
7c673cae FG |
11963 | if (bluefs && |
11964 | after_flush - bluefs_last_balance > | |
11fdf7f2 | 11965 | ceph::make_timespan(cct->_conf->bluestore_bluefs_balance_interval)) { |
7c673cae | 11966 | bluefs_last_balance = after_flush; |
11fdf7f2 TL |
11967 | int r = _balance_bluefs_freespace(); |
11968 | ceph_assert(r >= 0); | |
7c673cae FG |
11969 | } |
11970 | ||
11971 | // cleanup sync deferred keys | |
11972 | for (auto b : deferred_stable) { | |
11973 | for (auto& txc : b->txcs) { | |
11974 | bluestore_deferred_transaction_t& wt = *txc.deferred_txn; | |
11fdf7f2 | 11975 | ceph_assert(wt.released.empty()); // only kraken did this |
7c673cae FG |
11976 | string key; |
11977 | get_deferred_key(wt.seq, &key); | |
11978 | synct->rm_single_key(PREFIX_DEFERRED, key); | |
11979 | } | |
11980 | } | |
11981 | ||
9f95a23c TL |
11982 | #if defined(WITH_LTTNG) |
11983 | auto sync_start = mono_clock::now(); | |
11984 | #endif | |
7c673cae | 11985 | // submit synct synchronously (block and wait for it to commit) |
31f18b77 | 11986 | int r = cct->_conf->bluestore_debug_omit_kv_commit ? 0 : db->submit_transaction_sync(synct); |
11fdf7f2 TL |
11987 | ceph_assert(r == 0); |
11988 | ||
9f95a23c TL |
11989 | int committing_size = kv_committing.size(); |
11990 | int deferred_size = deferred_stable.size(); | |
11991 | ||
11992 | #if defined(WITH_LTTNG) | |
11993 | double sync_latency = ceph::to_seconds<double>(mono_clock::now() - sync_start); | |
11994 | for (auto txc: kv_committing) { | |
11995 | if (txc->tracing) { | |
11996 | tracepoint( | |
11997 | bluestore, | |
11998 | transaction_kv_sync_latency, | |
11999 | txc->osr->get_sequencer_id(), | |
12000 | txc->seq, | |
12001 | kv_committing.size(), | |
12002 | deferred_done.size(), | |
12003 | deferred_stable.size(), | |
12004 | sync_latency); | |
12005 | } | |
12006 | } | |
12007 | #endif | |
12008 | ||
11fdf7f2 | 12009 | { |
9f95a23c | 12010 | std::unique_lock m{kv_finalize_lock}; |
11fdf7f2 TL |
12011 | if (kv_committing_to_finalize.empty()) { |
12012 | kv_committing_to_finalize.swap(kv_committing); | |
12013 | } else { | |
12014 | kv_committing_to_finalize.insert( | |
12015 | kv_committing_to_finalize.end(), | |
12016 | kv_committing.begin(), | |
12017 | kv_committing.end()); | |
12018 | kv_committing.clear(); | |
12019 | } | |
12020 | if (deferred_stable_to_finalize.empty()) { | |
12021 | deferred_stable_to_finalize.swap(deferred_stable); | |
12022 | } else { | |
12023 | deferred_stable_to_finalize.insert( | |
12024 | deferred_stable_to_finalize.end(), | |
12025 | deferred_stable.begin(), | |
12026 | deferred_stable.end()); | |
12027 | deferred_stable.clear(); | |
12028 | } | |
9f95a23c TL |
12029 | if (!kv_finalize_in_progress) { |
12030 | kv_finalize_in_progress = true; | |
12031 | kv_finalize_cond.notify_one(); | |
12032 | } | |
11fdf7f2 | 12033 | } |
7c673cae FG |
12034 | |
12035 | if (new_nid_max) { | |
12036 | nid_max = new_nid_max; | |
12037 | dout(10) << __func__ << " nid_max now " << nid_max << dendl; | |
12038 | } | |
12039 | if (new_blobid_max) { | |
12040 | blobid_max = new_blobid_max; | |
12041 | dout(10) << __func__ << " blobid_max now " << blobid_max << dendl; | |
12042 | } | |
12043 | ||
224ce89b | 12044 | { |
11fdf7f2 TL |
12045 | auto finish = mono_clock::now(); |
12046 | ceph::timespan dur_flush = after_flush - start; | |
12047 | ceph::timespan dur_kv = finish - after_flush; | |
12048 | ceph::timespan dur = finish - start; | |
9f95a23c TL |
12049 | dout(20) << __func__ << " committed " << committing_size |
12050 | << " cleaned " << deferred_size | |
224ce89b WB |
12051 | << " in " << dur |
12052 | << " (" << dur_flush << " flush + " << dur_kv << " kv commit)" | |
12053 | << dendl; | |
494da23a TL |
12054 | log_latency("kv_flush", |
12055 | l_bluestore_kv_flush_lat, | |
12056 | dur_flush, | |
12057 | cct->_conf->bluestore_log_op_age); | |
12058 | log_latency("kv_commit", | |
12059 | l_bluestore_kv_commit_lat, | |
12060 | dur_kv, | |
12061 | cct->_conf->bluestore_log_op_age); | |
12062 | log_latency("kv_sync", | |
12063 | l_bluestore_kv_sync_lat, | |
12064 | dur, | |
12065 | cct->_conf->bluestore_log_op_age); | |
7c673cae | 12066 | } |
31f18b77 FG |
12067 | |
12068 | if (bluefs) { | |
11fdf7f2 TL |
12069 | if (!bluefs_extents_reclaiming.empty()) { |
12070 | dout(0) << __func__ << " releasing old bluefs 0x" << std::hex | |
12071 | << bluefs_extents_reclaiming << std::dec << dendl; | |
81eedcae TL |
12072 | int r = 0; |
12073 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
12074 | r = bdev->queue_discard(bluefs_extents_reclaiming); | |
12075 | if (r == 0) { | |
12076 | goto clear; | |
12077 | } | |
12078 | } else if (cct->_conf->bdev_enable_discard) { | |
12079 | for (auto p = bluefs_extents_reclaiming.begin(); p != bluefs_extents_reclaiming.end(); ++p) { | |
12080 | bdev->discard(p.get_start(), p.get_len()); | |
12081 | } | |
12082 | } | |
12083 | ||
11fdf7f2 | 12084 | alloc->release(bluefs_extents_reclaiming); |
81eedcae | 12085 | clear: |
11fdf7f2 | 12086 | bluefs_extents_reclaiming.clear(); |
31f18b77 | 12087 | } |
31f18b77 FG |
12088 | } |
12089 | ||
12090 | l.lock(); | |
12091 | // previously deferred "done" are now "stable" by virtue of this | |
12092 | // commit cycle. | |
12093 | deferred_stable_queue.swap(deferred_done); | |
12094 | } | |
12095 | } | |
12096 | dout(10) << __func__ << " finish" << dendl; | |
12097 | kv_sync_started = false; | |
12098 | } | |
12099 | ||
12100 | void BlueStore::_kv_finalize_thread() | |
12101 | { | |
12102 | deque<TransContext*> kv_committed; | |
12103 | deque<DeferredBatch*> deferred_stable; | |
12104 | dout(10) << __func__ << " start" << dendl; | |
11fdf7f2 TL |
12105 | std::unique_lock l(kv_finalize_lock); |
12106 | ceph_assert(!kv_finalize_started); | |
31f18b77 FG |
12107 | kv_finalize_started = true; |
12108 | kv_finalize_cond.notify_all(); | |
12109 | while (true) { | |
11fdf7f2 TL |
12110 | ceph_assert(kv_committed.empty()); |
12111 | ceph_assert(deferred_stable.empty()); | |
31f18b77 FG |
12112 | if (kv_committing_to_finalize.empty() && |
12113 | deferred_stable_to_finalize.empty()) { | |
12114 | if (kv_finalize_stop) | |
12115 | break; | |
12116 | dout(20) << __func__ << " sleep" << dendl; | |
9f95a23c | 12117 | kv_finalize_in_progress = false; |
31f18b77 FG |
12118 | kv_finalize_cond.wait(l); |
12119 | dout(20) << __func__ << " wake" << dendl; | |
12120 | } else { | |
12121 | kv_committed.swap(kv_committing_to_finalize); | |
12122 | deferred_stable.swap(deferred_stable_to_finalize); | |
12123 | l.unlock(); | |
12124 | dout(20) << __func__ << " kv_committed " << kv_committed << dendl; | |
12125 | dout(20) << __func__ << " deferred_stable " << deferred_stable << dendl; | |
12126 | ||
11fdf7f2 TL |
12127 | auto start = mono_clock::now(); |
12128 | ||
31f18b77 FG |
12129 | while (!kv_committed.empty()) { |
12130 | TransContext *txc = kv_committed.front(); | |
11fdf7f2 | 12131 | ceph_assert(txc->state == TransContext::STATE_KV_SUBMITTED); |
7c673cae | 12132 | _txc_state_proc(txc); |
31f18b77 | 12133 | kv_committed.pop_front(); |
7c673cae | 12134 | } |
31f18b77 | 12135 | |
7c673cae FG |
12136 | for (auto b : deferred_stable) { |
12137 | auto p = b->txcs.begin(); | |
12138 | while (p != b->txcs.end()) { | |
12139 | TransContext *txc = &*p; | |
12140 | p = b->txcs.erase(p); // unlink here because | |
12141 | _txc_state_proc(txc); // this may destroy txc | |
12142 | } | |
12143 | delete b; | |
12144 | } | |
31f18b77 | 12145 | deferred_stable.clear(); |
7c673cae FG |
12146 | |
12147 | if (!deferred_aggressive) { | |
31f18b77 | 12148 | if (deferred_queue_size >= deferred_batch_ops.load() || |
9f95a23c | 12149 | throttle.should_submit_deferred()) { |
224ce89b | 12150 | deferred_try_submit(); |
7c673cae FG |
12151 | } |
12152 | } | |
12153 | ||
12154 | // this is as good a place as any ... | |
12155 | _reap_collections(); | |
12156 | ||
11fdf7f2 | 12157 | logger->set(l_bluestore_fragmentation, |
9f95a23c | 12158 | (uint64_t)(alloc->get_fragmentation() * 1000)); |
11fdf7f2 | 12159 | |
494da23a TL |
12160 | log_latency("kv_final", |
12161 | l_bluestore_kv_final_lat, | |
12162 | mono_clock::now() - start, | |
12163 | cct->_conf->bluestore_log_op_age); | |
11fdf7f2 | 12164 | |
7c673cae | 12165 | l.lock(); |
7c673cae FG |
12166 | } |
12167 | } | |
12168 | dout(10) << __func__ << " finish" << dendl; | |
31f18b77 | 12169 | kv_finalize_started = false; |
7c673cae FG |
12170 | } |
12171 | ||
12172 | bluestore_deferred_op_t *BlueStore::_get_deferred_op( | |
9f95a23c | 12173 | TransContext *txc) |
7c673cae FG |
12174 | { |
12175 | if (!txc->deferred_txn) { | |
12176 | txc->deferred_txn = new bluestore_deferred_transaction_t; | |
12177 | } | |
12178 | txc->deferred_txn->ops.push_back(bluestore_deferred_op_t()); | |
12179 | return &txc->deferred_txn->ops.back(); | |
12180 | } | |
12181 | ||
12182 | void BlueStore::_deferred_queue(TransContext *txc) | |
12183 | { | |
12184 | dout(20) << __func__ << " txc " << txc << " osr " << txc->osr << dendl; | |
224ce89b | 12185 | deferred_lock.lock(); |
7c673cae FG |
12186 | if (!txc->osr->deferred_pending && |
12187 | !txc->osr->deferred_running) { | |
12188 | deferred_queue.push_back(*txc->osr); | |
12189 | } | |
12190 | if (!txc->osr->deferred_pending) { | |
12191 | txc->osr->deferred_pending = new DeferredBatch(cct, txc->osr.get()); | |
12192 | } | |
12193 | ++deferred_queue_size; | |
12194 | txc->osr->deferred_pending->txcs.push_back(*txc); | |
12195 | bluestore_deferred_transaction_t& wt = *txc->deferred_txn; | |
12196 | for (auto opi = wt.ops.begin(); opi != wt.ops.end(); ++opi) { | |
12197 | const auto& op = *opi; | |
11fdf7f2 | 12198 | ceph_assert(op.op == bluestore_deferred_op_t::OP_WRITE); |
7c673cae FG |
12199 | bufferlist::const_iterator p = op.data.begin(); |
12200 | for (auto e : op.extents) { | |
12201 | txc->osr->deferred_pending->prepare_write( | |
12202 | cct, wt.seq, e.offset, e.length, p); | |
12203 | } | |
12204 | } | |
12205 | if (deferred_aggressive && | |
12206 | !txc->osr->deferred_running) { | |
224ce89b WB |
12207 | _deferred_submit_unlock(txc->osr.get()); |
12208 | } else { | |
12209 | deferred_lock.unlock(); | |
7c673cae FG |
12210 | } |
12211 | } | |
12212 | ||
224ce89b | 12213 | void BlueStore::deferred_try_submit() |
7c673cae FG |
12214 | { |
12215 | dout(20) << __func__ << " " << deferred_queue.size() << " osrs, " | |
12216 | << deferred_queue_size << " txcs" << dendl; | |
11fdf7f2 | 12217 | std::lock_guard l(deferred_lock); |
224ce89b WB |
12218 | vector<OpSequencerRef> osrs; |
12219 | osrs.reserve(deferred_queue.size()); | |
7c673cae | 12220 | for (auto& osr : deferred_queue) { |
224ce89b WB |
12221 | osrs.push_back(&osr); |
12222 | } | |
12223 | for (auto& osr : osrs) { | |
181888fb FG |
12224 | if (osr->deferred_pending) { |
12225 | if (!osr->deferred_running) { | |
12226 | _deferred_submit_unlock(osr.get()); | |
12227 | deferred_lock.lock(); | |
12228 | } else { | |
12229 | dout(20) << __func__ << " osr " << osr << " already has running" | |
12230 | << dendl; | |
12231 | } | |
12232 | } else { | |
12233 | dout(20) << __func__ << " osr " << osr << " has no pending" << dendl; | |
7c673cae FG |
12234 | } |
12235 | } | |
9f95a23c TL |
12236 | |
12237 | deferred_last_submitted = ceph_clock_now(); | |
7c673cae FG |
12238 | } |
12239 | ||
224ce89b | 12240 | void BlueStore::_deferred_submit_unlock(OpSequencer *osr) |
7c673cae FG |
12241 | { |
12242 | dout(10) << __func__ << " osr " << osr | |
12243 | << " " << osr->deferred_pending->iomap.size() << " ios pending " | |
12244 | << dendl; | |
11fdf7f2 TL |
12245 | ceph_assert(osr->deferred_pending); |
12246 | ceph_assert(!osr->deferred_running); | |
7c673cae FG |
12247 | |
12248 | auto b = osr->deferred_pending; | |
12249 | deferred_queue_size -= b->seq_bytes.size(); | |
11fdf7f2 | 12250 | ceph_assert(deferred_queue_size >= 0); |
7c673cae FG |
12251 | |
12252 | osr->deferred_running = osr->deferred_pending; | |
12253 | osr->deferred_pending = nullptr; | |
12254 | ||
11fdf7f2 TL |
12255 | deferred_lock.unlock(); |
12256 | ||
12257 | for (auto& txc : b->txcs) { | |
9f95a23c | 12258 | throttle.log_state_latency(txc, logger, l_bluestore_state_deferred_queued_lat); |
11fdf7f2 | 12259 | } |
7c673cae FG |
12260 | uint64_t start = 0, pos = 0; |
12261 | bufferlist bl; | |
12262 | auto i = b->iomap.begin(); | |
12263 | while (true) { | |
12264 | if (i == b->iomap.end() || i->first != pos) { | |
12265 | if (bl.length()) { | |
12266 | dout(20) << __func__ << " write 0x" << std::hex | |
12267 | << start << "~" << bl.length() | |
12268 | << " crc " << bl.crc32c(-1) << std::dec << dendl; | |
11fdf7f2 | 12269 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
12270 | logger->inc(l_bluestore_deferred_write_ops); |
12271 | logger->inc(l_bluestore_deferred_write_bytes, bl.length()); | |
12272 | int r = bdev->aio_write(start, bl, &b->ioc, false); | |
11fdf7f2 | 12273 | ceph_assert(r == 0); |
7c673cae FG |
12274 | } |
12275 | } | |
12276 | if (i == b->iomap.end()) { | |
12277 | break; | |
12278 | } | |
12279 | start = 0; | |
12280 | pos = i->first; | |
12281 | bl.clear(); | |
12282 | } | |
12283 | dout(20) << __func__ << " seq " << i->second.seq << " 0x" | |
12284 | << std::hex << pos << "~" << i->second.bl.length() << std::dec | |
12285 | << dendl; | |
12286 | if (!bl.length()) { | |
12287 | start = pos; | |
12288 | } | |
12289 | pos += i->second.bl.length(); | |
12290 | bl.claim_append(i->second.bl); | |
12291 | ++i; | |
12292 | } | |
224ce89b | 12293 | |
7c673cae FG |
12294 | bdev->aio_submit(&b->ioc); |
12295 | } | |
12296 | ||
3efd9988 FG |
12297 | struct C_DeferredTrySubmit : public Context { |
12298 | BlueStore *store; | |
12299 | C_DeferredTrySubmit(BlueStore *s) : store(s) {} | |
12300 | void finish(int r) { | |
12301 | store->deferred_try_submit(); | |
12302 | } | |
12303 | }; | |
12304 | ||
7c673cae FG |
12305 | void BlueStore::_deferred_aio_finish(OpSequencer *osr) |
12306 | { | |
12307 | dout(10) << __func__ << " osr " << osr << dendl; | |
11fdf7f2 | 12308 | ceph_assert(osr->deferred_running); |
7c673cae FG |
12309 | DeferredBatch *b = osr->deferred_running; |
12310 | ||
12311 | { | |
9f95a23c | 12312 | deferred_lock.lock(); |
11fdf7f2 | 12313 | ceph_assert(osr->deferred_running == b); |
7c673cae FG |
12314 | osr->deferred_running = nullptr; |
12315 | if (!osr->deferred_pending) { | |
181888fb | 12316 | dout(20) << __func__ << " dequeueing" << dendl; |
7c673cae FG |
12317 | auto q = deferred_queue.iterator_to(*osr); |
12318 | deferred_queue.erase(q); | |
9f95a23c | 12319 | deferred_lock.unlock(); |
181888fb | 12320 | } else { |
9f95a23c TL |
12321 | deferred_lock.unlock(); |
12322 | if (deferred_aggressive) { | |
12323 | dout(20) << __func__ << " queuing async deferred_try_submit" << dendl; | |
12324 | finisher.queue(new C_DeferredTrySubmit(this)); | |
12325 | } else { | |
12326 | dout(20) << __func__ << " leaving queued, more pending" << dendl; | |
12327 | } | |
7c673cae FG |
12328 | } |
12329 | } | |
12330 | ||
12331 | { | |
31f18b77 | 12332 | uint64_t costs = 0; |
11fdf7f2 | 12333 | { |
11fdf7f2 TL |
12334 | for (auto& i : b->txcs) { |
12335 | TransContext *txc = &i; | |
9f95a23c | 12336 | throttle.log_state_latency(*txc, logger, l_bluestore_state_deferred_aio_wait_lat); |
11fdf7f2 TL |
12337 | txc->state = TransContext::STATE_DEFERRED_CLEANUP; |
12338 | costs += txc->cost; | |
12339 | } | |
7c673cae | 12340 | } |
9f95a23c | 12341 | throttle.release_deferred_throttle(costs); |
7c673cae FG |
12342 | } |
12343 | ||
9f95a23c | 12344 | { |
11fdf7f2 | 12345 | std::lock_guard l(kv_lock); |
9f95a23c TL |
12346 | deferred_done_queue.emplace_back(b); |
12347 | ||
12348 | // in the normal case, do not bother waking up the kv thread; it will | |
12349 | // catch us on the next commit anyway. | |
12350 | if (deferred_aggressive && !kv_sync_in_progress) { | |
12351 | kv_sync_in_progress = true; | |
12352 | kv_cond.notify_one(); | |
12353 | } | |
7c673cae FG |
12354 | } |
12355 | } | |
12356 | ||
12357 | int BlueStore::_deferred_replay() | |
12358 | { | |
12359 | dout(10) << __func__ << " start" << dendl; | |
7c673cae FG |
12360 | int count = 0; |
12361 | int r = 0; | |
11fdf7f2 TL |
12362 | CollectionRef ch = _get_collection(coll_t::meta()); |
12363 | bool fake_ch = false; | |
12364 | if (!ch) { | |
12365 | // hmm, replaying initial mkfs? | |
12366 | ch = static_cast<Collection*>(create_new_collection(coll_t::meta()).get()); | |
12367 | fake_ch = true; | |
12368 | } | |
12369 | OpSequencer *osr = static_cast<OpSequencer*>(ch->osr.get()); | |
7c673cae FG |
12370 | KeyValueDB::Iterator it = db->get_iterator(PREFIX_DEFERRED); |
12371 | for (it->lower_bound(string()); it->valid(); it->next(), ++count) { | |
12372 | dout(20) << __func__ << " replay " << pretty_binary_string(it->key()) | |
12373 | << dendl; | |
12374 | bluestore_deferred_transaction_t *deferred_txn = | |
12375 | new bluestore_deferred_transaction_t; | |
12376 | bufferlist bl = it->value(); | |
11fdf7f2 | 12377 | auto p = bl.cbegin(); |
7c673cae | 12378 | try { |
11fdf7f2 | 12379 | decode(*deferred_txn, p); |
7c673cae FG |
12380 | } catch (buffer::error& e) { |
12381 | derr << __func__ << " failed to decode deferred txn " | |
12382 | << pretty_binary_string(it->key()) << dendl; | |
12383 | delete deferred_txn; | |
12384 | r = -EIO; | |
12385 | goto out; | |
12386 | } | |
11fdf7f2 | 12387 | TransContext *txc = _txc_create(ch.get(), osr, nullptr); |
7c673cae FG |
12388 | txc->deferred_txn = deferred_txn; |
12389 | txc->state = TransContext::STATE_KV_DONE; | |
12390 | _txc_state_proc(txc); | |
12391 | } | |
12392 | out: | |
12393 | dout(20) << __func__ << " draining osr" << dendl; | |
11fdf7f2 | 12394 | _osr_register_zombie(osr); |
7c673cae | 12395 | _osr_drain_all(); |
11fdf7f2 TL |
12396 | if (fake_ch) { |
12397 | new_coll_map.clear(); | |
12398 | } | |
7c673cae FG |
12399 | dout(10) << __func__ << " completed " << count << " events" << dendl; |
12400 | return r; | |
12401 | } | |
12402 | ||
12403 | // --------------------------- | |
12404 | // transactions | |
12405 | ||
12406 | int BlueStore::queue_transactions( | |
11fdf7f2 TL |
12407 | CollectionHandle& ch, |
12408 | vector<Transaction>& tls, | |
12409 | TrackedOpRef op, | |
12410 | ThreadPool::TPHandle *handle) | |
12411 | { | |
12412 | FUNCTRACE(cct); | |
12413 | list<Context *> on_applied, on_commit, on_applied_sync; | |
7c673cae | 12414 | ObjectStore::Transaction::collect_contexts( |
11fdf7f2 | 12415 | tls, &on_applied, &on_commit, &on_applied_sync); |
7c673cae | 12416 | |
11fdf7f2 TL |
12417 | auto start = mono_clock::now(); |
12418 | ||
12419 | Collection *c = static_cast<Collection*>(ch.get()); | |
12420 | OpSequencer *osr = c->osr.get(); | |
12421 | dout(10) << __func__ << " ch " << c << " " << c->cid << dendl; | |
7c673cae FG |
12422 | |
12423 | // prepare | |
11fdf7f2 TL |
12424 | TransContext *txc = _txc_create(static_cast<Collection*>(ch.get()), osr, |
12425 | &on_commit); | |
7c673cae FG |
12426 | |
12427 | for (vector<Transaction>::iterator p = tls.begin(); p != tls.end(); ++p) { | |
7c673cae FG |
12428 | txc->bytes += (*p).get_num_bytes(); |
12429 | _txc_add_transaction(txc, &(*p)); | |
12430 | } | |
12431 | _txc_calc_cost(txc); | |
12432 | ||
12433 | _txc_write_nodes(txc, txc->t); | |
12434 | ||
12435 | // journal deferred items | |
12436 | if (txc->deferred_txn) { | |
12437 | txc->deferred_txn->seq = ++deferred_seq; | |
12438 | bufferlist bl; | |
11fdf7f2 | 12439 | encode(*txc->deferred_txn, bl); |
7c673cae FG |
12440 | string key; |
12441 | get_deferred_key(txc->deferred_txn->seq, &key); | |
12442 | txc->t->set(PREFIX_DEFERRED, key, bl); | |
12443 | } | |
12444 | ||
12445 | _txc_finalize_kv(txc, txc->t); | |
12446 | if (handle) | |
12447 | handle->suspend_tp_timeout(); | |
12448 | ||
11fdf7f2 | 12449 | auto tstart = mono_clock::now(); |
9f95a23c TL |
12450 | |
12451 | if (!throttle.try_start_transaction( | |
12452 | *db, | |
12453 | *txc, | |
12454 | tstart)) { | |
7c673cae | 12455 | // ensure we do not block here because of deferred writes |
9f95a23c TL |
12456 | dout(10) << __func__ << " failed get throttle_deferred_bytes, aggressive" |
12457 | << dendl; | |
12458 | ++deferred_aggressive; | |
12459 | deferred_try_submit(); | |
12460 | { | |
12461 | // wake up any previously finished deferred events | |
12462 | std::lock_guard l(kv_lock); | |
12463 | if (!kv_sync_in_progress) { | |
12464 | kv_sync_in_progress = true; | |
3efd9988 FG |
12465 | kv_cond.notify_one(); |
12466 | } | |
9f95a23c TL |
12467 | } |
12468 | throttle.finish_start_transaction(*db, *txc, tstart); | |
12469 | --deferred_aggressive; | |
7c673cae | 12470 | } |
11fdf7f2 | 12471 | auto tend = mono_clock::now(); |
7c673cae FG |
12472 | |
12473 | if (handle) | |
12474 | handle->reset_tp_timeout(); | |
12475 | ||
12476 | logger->inc(l_bluestore_txc); | |
12477 | ||
12478 | // execute (start) | |
12479 | _txc_state_proc(txc); | |
12480 | ||
11fdf7f2 TL |
12481 | // we're immediately readable (unlike FileStore) |
12482 | for (auto c : on_applied_sync) { | |
12483 | c->complete(0); | |
12484 | } | |
12485 | if (!on_applied.empty()) { | |
12486 | if (c->commit_queue) { | |
12487 | c->commit_queue->queue(on_applied); | |
12488 | } else { | |
12489 | finisher.queue(on_applied); | |
12490 | } | |
12491 | } | |
12492 | ||
494da23a TL |
12493 | log_latency("submit_transact", |
12494 | l_bluestore_submit_lat, | |
12495 | mono_clock::now() - start, | |
12496 | cct->_conf->bluestore_log_op_age); | |
12497 | log_latency("throttle_transact", | |
12498 | l_bluestore_throttle_lat, | |
12499 | tend - tstart, | |
12500 | cct->_conf->bluestore_log_op_age); | |
7c673cae FG |
12501 | return 0; |
12502 | } | |
12503 | ||
12504 | void BlueStore::_txc_aio_submit(TransContext *txc) | |
12505 | { | |
12506 | dout(10) << __func__ << " txc " << txc << dendl; | |
12507 | bdev->aio_submit(&txc->ioc); | |
12508 | } | |
12509 | ||
12510 | void BlueStore::_txc_add_transaction(TransContext *txc, Transaction *t) | |
12511 | { | |
12512 | Transaction::iterator i = t->begin(); | |
12513 | ||
81eedcae | 12514 | _dump_transaction<30>(cct, t); |
7c673cae FG |
12515 | |
12516 | vector<CollectionRef> cvec(i.colls.size()); | |
12517 | unsigned j = 0; | |
12518 | for (vector<coll_t>::iterator p = i.colls.begin(); p != i.colls.end(); | |
12519 | ++p, ++j) { | |
12520 | cvec[j] = _get_collection(*p); | |
7c673cae | 12521 | } |
11fdf7f2 | 12522 | |
7c673cae FG |
12523 | vector<OnodeRef> ovec(i.objects.size()); |
12524 | ||
12525 | for (int pos = 0; i.have_op(); ++pos) { | |
12526 | Transaction::Op *op = i.decode_op(); | |
12527 | int r = 0; | |
12528 | ||
12529 | // no coll or obj | |
12530 | if (op->op == Transaction::OP_NOP) | |
12531 | continue; | |
12532 | ||
11fdf7f2 | 12533 | |
7c673cae FG |
12534 | // collection operations |
12535 | CollectionRef &c = cvec[op->cid]; | |
11fdf7f2 TL |
12536 | |
12537 | // initialize osd_pool_id and do a smoke test that all collections belong | |
12538 | // to the same pool | |
12539 | spg_t pgid; | |
12540 | if (!!c ? c->cid.is_pg(&pgid) : false) { | |
12541 | ceph_assert(txc->osd_pool_id == META_POOL_ID || | |
12542 | txc->osd_pool_id == pgid.pool()); | |
12543 | txc->osd_pool_id = pgid.pool(); | |
12544 | } | |
12545 | ||
7c673cae FG |
12546 | switch (op->op) { |
12547 | case Transaction::OP_RMCOLL: | |
12548 | { | |
12549 | const coll_t &cid = i.get_cid(op->cid); | |
12550 | r = _remove_collection(txc, cid, &c); | |
12551 | if (!r) | |
12552 | continue; | |
12553 | } | |
12554 | break; | |
12555 | ||
12556 | case Transaction::OP_MKCOLL: | |
12557 | { | |
11fdf7f2 | 12558 | ceph_assert(!c); |
7c673cae FG |
12559 | const coll_t &cid = i.get_cid(op->cid); |
12560 | r = _create_collection(txc, cid, op->split_bits, &c); | |
12561 | if (!r) | |
12562 | continue; | |
12563 | } | |
12564 | break; | |
12565 | ||
12566 | case Transaction::OP_SPLIT_COLLECTION: | |
11fdf7f2 | 12567 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12568 | break; |
12569 | ||
12570 | case Transaction::OP_SPLIT_COLLECTION2: | |
12571 | { | |
12572 | uint32_t bits = op->split_bits; | |
12573 | uint32_t rem = op->split_rem; | |
12574 | r = _split_collection(txc, c, cvec[op->dest_cid], bits, rem); | |
12575 | if (!r) | |
12576 | continue; | |
12577 | } | |
12578 | break; | |
12579 | ||
11fdf7f2 TL |
12580 | case Transaction::OP_MERGE_COLLECTION: |
12581 | { | |
12582 | uint32_t bits = op->split_bits; | |
12583 | r = _merge_collection(txc, &c, cvec[op->dest_cid], bits); | |
12584 | if (!r) | |
12585 | continue; | |
12586 | } | |
12587 | break; | |
12588 | ||
7c673cae FG |
12589 | case Transaction::OP_COLL_HINT: |
12590 | { | |
12591 | uint32_t type = op->hint_type; | |
12592 | bufferlist hint; | |
12593 | i.decode_bl(hint); | |
11fdf7f2 | 12594 | auto hiter = hint.cbegin(); |
7c673cae FG |
12595 | if (type == Transaction::COLL_HINT_EXPECTED_NUM_OBJECTS) { |
12596 | uint32_t pg_num; | |
12597 | uint64_t num_objs; | |
11fdf7f2 TL |
12598 | decode(pg_num, hiter); |
12599 | decode(num_objs, hiter); | |
7c673cae FG |
12600 | dout(10) << __func__ << " collection hint objects is a no-op, " |
12601 | << " pg_num " << pg_num << " num_objects " << num_objs | |
12602 | << dendl; | |
12603 | } else { | |
12604 | // Ignore the hint | |
12605 | dout(10) << __func__ << " unknown collection hint " << type << dendl; | |
12606 | } | |
12607 | continue; | |
12608 | } | |
12609 | break; | |
12610 | ||
12611 | case Transaction::OP_COLL_SETATTR: | |
12612 | r = -EOPNOTSUPP; | |
12613 | break; | |
12614 | ||
12615 | case Transaction::OP_COLL_RMATTR: | |
12616 | r = -EOPNOTSUPP; | |
12617 | break; | |
12618 | ||
12619 | case Transaction::OP_COLL_RENAME: | |
11fdf7f2 | 12620 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12621 | break; |
12622 | } | |
12623 | if (r < 0) { | |
12624 | derr << __func__ << " error " << cpp_strerror(r) | |
12625 | << " not handled on operation " << op->op | |
12626 | << " (op " << pos << ", counting from 0)" << dendl; | |
81eedcae | 12627 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12628 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12629 | } |
12630 | ||
12631 | // these operations implicity create the object | |
12632 | bool create = false; | |
12633 | if (op->op == Transaction::OP_TOUCH || | |
9f95a23c | 12634 | op->op == Transaction::OP_CREATE || |
7c673cae FG |
12635 | op->op == Transaction::OP_WRITE || |
12636 | op->op == Transaction::OP_ZERO) { | |
12637 | create = true; | |
12638 | } | |
12639 | ||
12640 | // object operations | |
9f95a23c | 12641 | std::unique_lock l(c->lock); |
7c673cae FG |
12642 | OnodeRef &o = ovec[op->oid]; |
12643 | if (!o) { | |
12644 | ghobject_t oid = i.get_oid(op->oid); | |
9f95a23c | 12645 | o = c->get_onode(oid, create, op->op == Transaction::OP_CREATE); |
7c673cae FG |
12646 | } |
12647 | if (!create && (!o || !o->exists)) { | |
12648 | dout(10) << __func__ << " op " << op->op << " got ENOENT on " | |
12649 | << i.get_oid(op->oid) << dendl; | |
12650 | r = -ENOENT; | |
12651 | goto endop; | |
12652 | } | |
12653 | ||
12654 | switch (op->op) { | |
9f95a23c | 12655 | case Transaction::OP_CREATE: |
7c673cae FG |
12656 | case Transaction::OP_TOUCH: |
12657 | r = _touch(txc, c, o); | |
12658 | break; | |
12659 | ||
12660 | case Transaction::OP_WRITE: | |
12661 | { | |
12662 | uint64_t off = op->off; | |
12663 | uint64_t len = op->len; | |
12664 | uint32_t fadvise_flags = i.get_fadvise_flags(); | |
12665 | bufferlist bl; | |
12666 | i.decode_bl(bl); | |
12667 | r = _write(txc, c, o, off, len, bl, fadvise_flags); | |
12668 | } | |
12669 | break; | |
12670 | ||
12671 | case Transaction::OP_ZERO: | |
12672 | { | |
12673 | uint64_t off = op->off; | |
12674 | uint64_t len = op->len; | |
12675 | r = _zero(txc, c, o, off, len); | |
12676 | } | |
12677 | break; | |
12678 | ||
12679 | case Transaction::OP_TRIMCACHE: | |
12680 | { | |
12681 | // deprecated, no-op | |
12682 | } | |
12683 | break; | |
12684 | ||
12685 | case Transaction::OP_TRUNCATE: | |
12686 | { | |
12687 | uint64_t off = op->off; | |
35e4c445 | 12688 | r = _truncate(txc, c, o, off); |
7c673cae FG |
12689 | } |
12690 | break; | |
12691 | ||
12692 | case Transaction::OP_REMOVE: | |
12693 | { | |
12694 | r = _remove(txc, c, o); | |
12695 | } | |
12696 | break; | |
12697 | ||
12698 | case Transaction::OP_SETATTR: | |
12699 | { | |
12700 | string name = i.decode_string(); | |
12701 | bufferptr bp; | |
12702 | i.decode_bp(bp); | |
12703 | r = _setattr(txc, c, o, name, bp); | |
12704 | } | |
12705 | break; | |
12706 | ||
12707 | case Transaction::OP_SETATTRS: | |
12708 | { | |
12709 | map<string, bufferptr> aset; | |
12710 | i.decode_attrset(aset); | |
12711 | r = _setattrs(txc, c, o, aset); | |
12712 | } | |
12713 | break; | |
12714 | ||
12715 | case Transaction::OP_RMATTR: | |
12716 | { | |
12717 | string name = i.decode_string(); | |
12718 | r = _rmattr(txc, c, o, name); | |
12719 | } | |
12720 | break; | |
12721 | ||
12722 | case Transaction::OP_RMATTRS: | |
12723 | { | |
12724 | r = _rmattrs(txc, c, o); | |
12725 | } | |
12726 | break; | |
12727 | ||
12728 | case Transaction::OP_CLONE: | |
12729 | { | |
12730 | OnodeRef& no = ovec[op->dest_oid]; | |
12731 | if (!no) { | |
12732 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12733 | no = c->get_onode(noid, true); | |
12734 | } | |
12735 | r = _clone(txc, c, o, no); | |
12736 | } | |
12737 | break; | |
12738 | ||
12739 | case Transaction::OP_CLONERANGE: | |
11fdf7f2 | 12740 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12741 | break; |
12742 | ||
12743 | case Transaction::OP_CLONERANGE2: | |
12744 | { | |
12745 | OnodeRef& no = ovec[op->dest_oid]; | |
12746 | if (!no) { | |
12747 | const ghobject_t& noid = i.get_oid(op->dest_oid); | |
12748 | no = c->get_onode(noid, true); | |
12749 | } | |
12750 | uint64_t srcoff = op->off; | |
12751 | uint64_t len = op->len; | |
12752 | uint64_t dstoff = op->dest_off; | |
12753 | r = _clone_range(txc, c, o, no, srcoff, len, dstoff); | |
12754 | } | |
12755 | break; | |
12756 | ||
12757 | case Transaction::OP_COLL_ADD: | |
11fdf7f2 | 12758 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12759 | break; |
12760 | ||
12761 | case Transaction::OP_COLL_REMOVE: | |
11fdf7f2 | 12762 | ceph_abort_msg("not implemented"); |
7c673cae FG |
12763 | break; |
12764 | ||
12765 | case Transaction::OP_COLL_MOVE: | |
11fdf7f2 | 12766 | ceph_abort_msg("deprecated"); |
7c673cae FG |
12767 | break; |
12768 | ||
12769 | case Transaction::OP_COLL_MOVE_RENAME: | |
12770 | case Transaction::OP_TRY_RENAME: | |
12771 | { | |
11fdf7f2 | 12772 | ceph_assert(op->cid == op->dest_cid); |
7c673cae FG |
12773 | const ghobject_t& noid = i.get_oid(op->dest_oid); |
12774 | OnodeRef& no = ovec[op->dest_oid]; | |
12775 | if (!no) { | |
12776 | no = c->get_onode(noid, false); | |
12777 | } | |
12778 | r = _rename(txc, c, o, no, noid); | |
12779 | } | |
12780 | break; | |
12781 | ||
12782 | case Transaction::OP_OMAP_CLEAR: | |
12783 | { | |
12784 | r = _omap_clear(txc, c, o); | |
12785 | } | |
12786 | break; | |
12787 | case Transaction::OP_OMAP_SETKEYS: | |
12788 | { | |
12789 | bufferlist aset_bl; | |
12790 | i.decode_attrset_bl(&aset_bl); | |
12791 | r = _omap_setkeys(txc, c, o, aset_bl); | |
12792 | } | |
12793 | break; | |
12794 | case Transaction::OP_OMAP_RMKEYS: | |
12795 | { | |
12796 | bufferlist keys_bl; | |
12797 | i.decode_keyset_bl(&keys_bl); | |
12798 | r = _omap_rmkeys(txc, c, o, keys_bl); | |
12799 | } | |
12800 | break; | |
12801 | case Transaction::OP_OMAP_RMKEYRANGE: | |
12802 | { | |
12803 | string first, last; | |
12804 | first = i.decode_string(); | |
12805 | last = i.decode_string(); | |
12806 | r = _omap_rmkey_range(txc, c, o, first, last); | |
12807 | } | |
12808 | break; | |
12809 | case Transaction::OP_OMAP_SETHEADER: | |
12810 | { | |
12811 | bufferlist bl; | |
12812 | i.decode_bl(bl); | |
12813 | r = _omap_setheader(txc, c, o, bl); | |
12814 | } | |
12815 | break; | |
12816 | ||
12817 | case Transaction::OP_SETALLOCHINT: | |
12818 | { | |
12819 | r = _set_alloc_hint(txc, c, o, | |
12820 | op->expected_object_size, | |
12821 | op->expected_write_size, | |
12822 | op->alloc_hint_flags); | |
12823 | } | |
12824 | break; | |
12825 | ||
12826 | default: | |
11fdf7f2 | 12827 | derr << __func__ << " bad op " << op->op << dendl; |
7c673cae FG |
12828 | ceph_abort(); |
12829 | } | |
12830 | ||
12831 | endop: | |
12832 | if (r < 0) { | |
12833 | bool ok = false; | |
12834 | ||
12835 | if (r == -ENOENT && !(op->op == Transaction::OP_CLONERANGE || | |
12836 | op->op == Transaction::OP_CLONE || | |
12837 | op->op == Transaction::OP_CLONERANGE2 || | |
12838 | op->op == Transaction::OP_COLL_ADD || | |
12839 | op->op == Transaction::OP_SETATTR || | |
12840 | op->op == Transaction::OP_SETATTRS || | |
12841 | op->op == Transaction::OP_RMATTR || | |
12842 | op->op == Transaction::OP_OMAP_SETKEYS || | |
12843 | op->op == Transaction::OP_OMAP_RMKEYS || | |
12844 | op->op == Transaction::OP_OMAP_RMKEYRANGE || | |
12845 | op->op == Transaction::OP_OMAP_SETHEADER)) | |
12846 | // -ENOENT is usually okay | |
12847 | ok = true; | |
12848 | if (r == -ENODATA) | |
12849 | ok = true; | |
12850 | ||
12851 | if (!ok) { | |
12852 | const char *msg = "unexpected error code"; | |
12853 | ||
12854 | if (r == -ENOENT && (op->op == Transaction::OP_CLONERANGE || | |
12855 | op->op == Transaction::OP_CLONE || | |
12856 | op->op == Transaction::OP_CLONERANGE2)) | |
12857 | msg = "ENOENT on clone suggests osd bug"; | |
12858 | ||
12859 | if (r == -ENOSPC) | |
12860 | // For now, if we hit _any_ ENOSPC, crash, before we do any damage | |
12861 | // by partially applying transactions. | |
12862 | msg = "ENOSPC from bluestore, misconfigured cluster"; | |
12863 | ||
12864 | if (r == -ENOTEMPTY) { | |
12865 | msg = "ENOTEMPTY suggests garbage data in osd data dir"; | |
12866 | } | |
12867 | ||
12868 | derr << __func__ << " error " << cpp_strerror(r) | |
12869 | << " not handled on operation " << op->op | |
12870 | << " (op " << pos << ", counting from 0)" | |
12871 | << dendl; | |
12872 | derr << msg << dendl; | |
81eedcae | 12873 | _dump_transaction<0>(cct, t); |
11fdf7f2 | 12874 | ceph_abort_msg("unexpected error"); |
7c673cae FG |
12875 | } |
12876 | } | |
12877 | } | |
12878 | } | |
12879 | ||
12880 | ||
12881 | ||
12882 | // ----------------- | |
12883 | // write operations | |
12884 | ||
12885 | int BlueStore::_touch(TransContext *txc, | |
12886 | CollectionRef& c, | |
12887 | OnodeRef &o) | |
12888 | { | |
12889 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
12890 | int r = 0; | |
7c673cae FG |
12891 | _assign_nid(txc, o); |
12892 | txc->write_onode(o); | |
12893 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
12894 | return r; | |
12895 | } | |
12896 | ||
7c673cae FG |
12897 | void BlueStore::_pad_zeros( |
12898 | bufferlist *bl, uint64_t *offset, | |
12899 | uint64_t chunk_size) | |
12900 | { | |
12901 | auto length = bl->length(); | |
12902 | dout(30) << __func__ << " 0x" << std::hex << *offset << "~" << length | |
12903 | << " chunk_size 0x" << chunk_size << std::dec << dendl; | |
12904 | dout(40) << "before:\n"; | |
12905 | bl->hexdump(*_dout); | |
12906 | *_dout << dendl; | |
12907 | // front | |
12908 | size_t front_pad = *offset % chunk_size; | |
12909 | size_t back_pad = 0; | |
12910 | size_t pad_count = 0; | |
12911 | if (front_pad) { | |
11fdf7f2 TL |
12912 | size_t front_copy = std::min<uint64_t>(chunk_size - front_pad, length); |
12913 | bufferptr z = buffer::create_small_page_aligned(chunk_size); | |
224ce89b | 12914 | z.zero(0, front_pad, false); |
7c673cae | 12915 | pad_count += front_pad; |
9f95a23c | 12916 | bl->begin().copy(front_copy, z.c_str() + front_pad); |
7c673cae FG |
12917 | if (front_copy + front_pad < chunk_size) { |
12918 | back_pad = chunk_size - (length + front_pad); | |
224ce89b | 12919 | z.zero(front_pad + length, back_pad, false); |
7c673cae FG |
12920 | pad_count += back_pad; |
12921 | } | |
12922 | bufferlist old, t; | |
12923 | old.swap(*bl); | |
12924 | t.substr_of(old, front_copy, length - front_copy); | |
12925 | bl->append(z); | |
12926 | bl->claim_append(t); | |
12927 | *offset -= front_pad; | |
224ce89b | 12928 | length += pad_count; |
7c673cae FG |
12929 | } |
12930 | ||
12931 | // back | |
12932 | uint64_t end = *offset + length; | |
12933 | unsigned back_copy = end % chunk_size; | |
12934 | if (back_copy) { | |
11fdf7f2 | 12935 | ceph_assert(back_pad == 0); |
7c673cae | 12936 | back_pad = chunk_size - back_copy; |
11fdf7f2 | 12937 | ceph_assert(back_copy <= length); |
7c673cae | 12938 | bufferptr tail(chunk_size); |
9f95a23c | 12939 | bl->begin(length - back_copy).copy(back_copy, tail.c_str()); |
224ce89b | 12940 | tail.zero(back_copy, back_pad, false); |
7c673cae FG |
12941 | bufferlist old; |
12942 | old.swap(*bl); | |
12943 | bl->substr_of(old, 0, length - back_copy); | |
12944 | bl->append(tail); | |
12945 | length += back_pad; | |
12946 | pad_count += back_pad; | |
12947 | } | |
12948 | dout(20) << __func__ << " pad 0x" << std::hex << front_pad << " + 0x" | |
12949 | << back_pad << " on front/back, now 0x" << *offset << "~" | |
12950 | << length << std::dec << dendl; | |
12951 | dout(40) << "after:\n"; | |
12952 | bl->hexdump(*_dout); | |
12953 | *_dout << dendl; | |
12954 | if (pad_count) | |
12955 | logger->inc(l_bluestore_write_pad_bytes, pad_count); | |
11fdf7f2 | 12956 | ceph_assert(bl->length() == length); |
7c673cae FG |
12957 | } |
12958 | ||
12959 | void BlueStore::_do_write_small( | |
12960 | TransContext *txc, | |
12961 | CollectionRef &c, | |
12962 | OnodeRef o, | |
12963 | uint64_t offset, uint64_t length, | |
12964 | bufferlist::iterator& blp, | |
12965 | WriteContext *wctx) | |
12966 | { | |
12967 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
12968 | << std::dec << dendl; | |
11fdf7f2 | 12969 | ceph_assert(length < min_alloc_size); |
7c673cae FG |
12970 | uint64_t end_offs = offset + length; |
12971 | ||
12972 | logger->inc(l_bluestore_write_small); | |
12973 | logger->inc(l_bluestore_write_small_bytes, length); | |
12974 | ||
12975 | bufferlist bl; | |
12976 | blp.copy(length, bl); | |
12977 | ||
81eedcae TL |
12978 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
12979 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
12980 | uint32_t alloc_len = min_alloc_size; | |
12981 | auto offset0 = p2align<uint64_t>(offset, alloc_len); | |
12982 | ||
12983 | bool any_change; | |
12984 | ||
12985 | // search suitable extent in both forward and reverse direction in | |
12986 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
12987 | // then check if blob can be reused via can_reuse_blob func or apply | |
12988 | // direct/deferred write (the latter for extents including or higher | |
12989 | // than 'offset' only). | |
12990 | o->extent_map.fault_range(db, min_off, offset + max_bsize - min_off); | |
12991 | ||
7c673cae FG |
12992 | // Look for an existing mutable blob we can use. |
12993 | auto begin = o->extent_map.extent_map.begin(); | |
12994 | auto end = o->extent_map.extent_map.end(); | |
12995 | auto ep = o->extent_map.seek_lextent(offset); | |
12996 | if (ep != begin) { | |
12997 | --ep; | |
12998 | if (ep->blob_end() <= offset) { | |
12999 | ++ep; | |
13000 | } | |
13001 | } | |
13002 | auto prev_ep = ep; | |
13003 | if (prev_ep != begin) { | |
13004 | --prev_ep; | |
13005 | } else { | |
13006 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
13007 | } | |
13008 | ||
eafe8130 TL |
13009 | boost::container::flat_set<const bluestore_blob_t*> inspected_blobs; |
13010 | // We don't want to have more blobs than min alloc units fit | |
13011 | // into 2 max blobs | |
13012 | size_t blob_threshold = max_blob_size / min_alloc_size * 2 + 1; | |
13013 | bool above_blob_threshold = false; | |
13014 | ||
13015 | inspected_blobs.reserve(blob_threshold); | |
13016 | ||
13017 | uint64_t max_off = 0; | |
13018 | auto start_ep = ep; | |
13019 | auto end_ep = ep; // exclusively | |
7c673cae FG |
13020 | do { |
13021 | any_change = false; | |
13022 | ||
13023 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13024 | BlobRef b = ep->blob; | |
eafe8130 TL |
13025 | if (!above_blob_threshold) { |
13026 | inspected_blobs.insert(&b->get_blob()); | |
13027 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13028 | } | |
13029 | max_off = ep->logical_end(); | |
7c673cae | 13030 | auto bstart = ep->blob_start(); |
eafe8130 | 13031 | |
7c673cae FG |
13032 | dout(20) << __func__ << " considering " << *b |
13033 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
13034 | if (bstart >= end_offs) { | |
13035 | dout(20) << __func__ << " ignoring distant " << *b << dendl; | |
13036 | } else if (!b->get_blob().is_mutable()) { | |
13037 | dout(20) << __func__ << " ignoring immutable " << *b << dendl; | |
13038 | } else if (ep->logical_offset % min_alloc_size != | |
13039 | ep->blob_offset % min_alloc_size) { | |
13040 | dout(20) << __func__ << " ignoring offset-skewed " << *b << dendl; | |
13041 | } else { | |
13042 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13043 | // can we pad our head/tail out with zeros? | |
13044 | uint64_t head_pad, tail_pad; | |
11fdf7f2 TL |
13045 | head_pad = p2phase(offset, chunk_size); |
13046 | tail_pad = p2nphase(end_offs, chunk_size); | |
7c673cae FG |
13047 | if (head_pad || tail_pad) { |
13048 | o->extent_map.fault_range(db, offset - head_pad, | |
13049 | end_offs - offset + head_pad + tail_pad); | |
13050 | } | |
13051 | if (head_pad && | |
13052 | o->extent_map.has_any_lextents(offset - head_pad, chunk_size)) { | |
13053 | head_pad = 0; | |
13054 | } | |
13055 | if (tail_pad && o->extent_map.has_any_lextents(end_offs, tail_pad)) { | |
13056 | tail_pad = 0; | |
13057 | } | |
13058 | ||
13059 | uint64_t b_off = offset - head_pad - bstart; | |
13060 | uint64_t b_len = length + head_pad + tail_pad; | |
13061 | ||
13062 | // direct write into unused blocks of an existing mutable blob? | |
13063 | if ((b_off % chunk_size == 0 && b_len % chunk_size == 0) && | |
13064 | b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13065 | b->get_blob().is_unused(b_off, b_len) && | |
13066 | b->get_blob().is_allocated(b_off, b_len)) { | |
224ce89b | 13067 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13068 | |
13069 | dout(20) << __func__ << " write to unused 0x" << std::hex | |
13070 | << b_off << "~" << b_len | |
13071 | << " pad 0x" << head_pad << " + 0x" << tail_pad | |
13072 | << std::dec << " of mutable " << *b << dendl; | |
224ce89b | 13073 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13074 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13075 | ||
11fdf7f2 | 13076 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13077 | if (b_len <= prefer_deferred_size) { |
13078 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13079 | << b_len << std::dec << " unused write via deferred" << dendl; | |
9f95a23c | 13080 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13081 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13082 | b->get_blob().map( | |
13083 | b_off, b_len, | |
13084 | [&](uint64_t offset, uint64_t length) { | |
13085 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13086 | return 0; | |
13087 | }); | |
224ce89b | 13088 | op->data = bl; |
7c673cae FG |
13089 | } else { |
13090 | b->get_blob().map_bl( | |
224ce89b | 13091 | b_off, bl, |
7c673cae FG |
13092 | [&](uint64_t offset, bufferlist& t) { |
13093 | bdev->aio_write(offset, t, | |
13094 | &txc->ioc, wctx->buffered); | |
13095 | }); | |
13096 | } | |
13097 | } | |
224ce89b | 13098 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae FG |
13099 | dout(20) << __func__ << " lex old " << *ep << dendl; |
13100 | Extent *le = o->extent_map.set_lextent(c, offset, b_off + head_pad, length, | |
13101 | b, | |
13102 | &wctx->old_extents); | |
13103 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13104 | txc->statfs_delta.stored() += le->length; | |
13105 | dout(20) << __func__ << " lex " << *le << dendl; | |
13106 | logger->inc(l_bluestore_write_small_unused); | |
13107 | return; | |
13108 | } | |
13109 | // read some data to fill out the chunk? | |
11fdf7f2 TL |
13110 | uint64_t head_read = p2phase(b_off, chunk_size); |
13111 | uint64_t tail_read = p2nphase(b_off + b_len, chunk_size); | |
7c673cae FG |
13112 | if ((head_read || tail_read) && |
13113 | (b->get_blob().get_ondisk_length() >= b_off + b_len + tail_read) && | |
13114 | head_read + tail_read < min_alloc_size) { | |
13115 | b_off -= head_read; | |
13116 | b_len += head_read + tail_read; | |
13117 | ||
13118 | } else { | |
13119 | head_read = tail_read = 0; | |
13120 | } | |
13121 | ||
13122 | // chunk-aligned deferred overwrite? | |
13123 | if (b->get_blob().get_ondisk_length() >= b_off + b_len && | |
13124 | b_off % chunk_size == 0 && | |
13125 | b_len % chunk_size == 0 && | |
13126 | b->get_blob().is_allocated(b_off, b_len)) { | |
13127 | ||
224ce89b | 13128 | _apply_padding(head_pad, tail_pad, bl); |
7c673cae FG |
13129 | |
13130 | dout(20) << __func__ << " reading head 0x" << std::hex << head_read | |
13131 | << " and tail 0x" << tail_read << std::dec << dendl; | |
13132 | if (head_read) { | |
13133 | bufferlist head_bl; | |
13134 | int r = _do_read(c.get(), o, offset - head_pad - head_read, head_read, | |
13135 | head_bl, 0); | |
11fdf7f2 | 13136 | ceph_assert(r >= 0 && r <= (int)head_read); |
7c673cae FG |
13137 | size_t zlen = head_read - r; |
13138 | if (zlen) { | |
13139 | head_bl.append_zero(zlen); | |
13140 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13141 | } | |
11fdf7f2 TL |
13142 | head_bl.claim_append(bl); |
13143 | bl.swap(head_bl); | |
7c673cae FG |
13144 | logger->inc(l_bluestore_write_penalty_read_ops); |
13145 | } | |
13146 | if (tail_read) { | |
13147 | bufferlist tail_bl; | |
13148 | int r = _do_read(c.get(), o, offset + length + tail_pad, tail_read, | |
13149 | tail_bl, 0); | |
11fdf7f2 | 13150 | ceph_assert(r >= 0 && r <= (int)tail_read); |
7c673cae FG |
13151 | size_t zlen = tail_read - r; |
13152 | if (zlen) { | |
13153 | tail_bl.append_zero(zlen); | |
13154 | logger->inc(l_bluestore_write_pad_bytes, zlen); | |
13155 | } | |
224ce89b | 13156 | bl.claim_append(tail_bl); |
7c673cae FG |
13157 | logger->inc(l_bluestore_write_penalty_read_ops); |
13158 | } | |
13159 | logger->inc(l_bluestore_write_small_pre_read); | |
13160 | ||
224ce89b | 13161 | _buffer_cache_write(txc, b, b_off, bl, |
7c673cae FG |
13162 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); |
13163 | ||
7c673cae | 13164 | if (b->get_blob().csum_type) { |
224ce89b | 13165 | b->dirty_blob().calc_csum(b_off, bl); |
7c673cae | 13166 | } |
11fdf7f2 TL |
13167 | |
13168 | if (!g_conf()->bluestore_debug_omit_block_device_write) { | |
9f95a23c | 13169 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
11fdf7f2 TL |
13170 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13171 | int r = b->get_blob().map( | |
13172 | b_off, b_len, | |
13173 | [&](uint64_t offset, uint64_t length) { | |
13174 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13175 | return 0; | |
13176 | }); | |
13177 | ceph_assert(r == 0); | |
13178 | op->data.claim(bl); | |
13179 | dout(20) << __func__ << " deferred write 0x" << std::hex << b_off << "~" | |
13180 | << b_len << std::dec << " of mutable " << *b | |
13181 | << " at " << op->extents << dendl; | |
13182 | } | |
13183 | ||
7c673cae FG |
13184 | Extent *le = o->extent_map.set_lextent(c, offset, offset - bstart, length, |
13185 | b, &wctx->old_extents); | |
13186 | b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13187 | txc->statfs_delta.stored() += le->length; | |
13188 | dout(20) << __func__ << " lex " << *le << dendl; | |
13189 | logger->inc(l_bluestore_write_small_deferred); | |
13190 | return; | |
13191 | } | |
224ce89b WB |
13192 | // try to reuse blob if we can |
13193 | if (b->can_reuse_blob(min_alloc_size, | |
7c673cae FG |
13194 | max_bsize, |
13195 | offset0 - bstart, | |
13196 | &alloc_len)) { | |
11fdf7f2 | 13197 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13198 | // fit into reused blob |
13199 | // Need to check for pending writes desiring to | |
13200 | // reuse the same pextent. The rationale is that during GC two chunks | |
13201 | // from garbage blobs(compressed?) can share logical space within the same | |
13202 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13203 | // Hence the second write will fail in an attempt to reuse blob at | |
13204 | // do_alloc_write(). | |
13205 | if (!wctx->has_conflict(b, | |
13206 | offset0, | |
13207 | offset0 + alloc_len, | |
13208 | min_alloc_size)) { | |
13209 | ||
13210 | // we can't reuse pad_head/pad_tail since they might be truncated | |
13211 | // due to existent extents | |
13212 | uint64_t b_off = offset - bstart; | |
13213 | uint64_t b_off0 = b_off; | |
13214 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13215 | ||
13216 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13217 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13218 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13219 | << std::dec << dendl; |
13220 | ||
13221 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13222 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13223 | false, false); | |
13224 | logger->inc(l_bluestore_write_small_unused); | |
13225 | return; | |
13226 | } | |
13227 | } | |
13228 | } | |
13229 | ++ep; | |
eafe8130 | 13230 | end_ep = ep; |
7c673cae FG |
13231 | any_change = true; |
13232 | } // if (ep != end && ep->logical_offset < offset + max_bsize) | |
13233 | ||
13234 | // check extent for reuse in reverse order | |
13235 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
13236 | BlobRef b = prev_ep->blob; | |
eafe8130 TL |
13237 | if (!above_blob_threshold) { |
13238 | inspected_blobs.insert(&b->get_blob()); | |
13239 | above_blob_threshold = inspected_blobs.size() >= blob_threshold; | |
13240 | } | |
13241 | start_ep = prev_ep; | |
7c673cae FG |
13242 | auto bstart = prev_ep->blob_start(); |
13243 | dout(20) << __func__ << " considering " << *b | |
13244 | << " bstart 0x" << std::hex << bstart << std::dec << dendl; | |
224ce89b | 13245 | if (b->can_reuse_blob(min_alloc_size, |
7c673cae FG |
13246 | max_bsize, |
13247 | offset0 - bstart, | |
13248 | &alloc_len)) { | |
11fdf7f2 | 13249 | ceph_assert(alloc_len == min_alloc_size); // expecting data always |
7c673cae FG |
13250 | // fit into reused blob |
13251 | // Need to check for pending writes desiring to | |
13252 | // reuse the same pextent. The rationale is that during GC two chunks | |
13253 | // from garbage blobs(compressed?) can share logical space within the same | |
13254 | // AU. That's in turn might be caused by unaligned len in clone_range2. | |
13255 | // Hence the second write will fail in an attempt to reuse blob at | |
13256 | // do_alloc_write(). | |
13257 | if (!wctx->has_conflict(b, | |
13258 | offset0, | |
13259 | offset0 + alloc_len, | |
13260 | min_alloc_size)) { | |
13261 | ||
13262 | uint64_t chunk_size = b->get_blob().get_chunk_size(block_size); | |
13263 | uint64_t b_off = offset - bstart; | |
13264 | uint64_t b_off0 = b_off; | |
13265 | _pad_zeros(&bl, &b_off0, chunk_size); | |
13266 | ||
13267 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b WB |
13268 | << " (0x" << b_off0 << "~" << bl.length() << ")" |
13269 | << " (0x" << b_off << "~" << length << ")" | |
7c673cae FG |
13270 | << std::dec << dendl; |
13271 | ||
13272 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
13273 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, | |
13274 | false, false); | |
13275 | logger->inc(l_bluestore_write_small_unused); | |
13276 | return; | |
13277 | } | |
13278 | } | |
13279 | if (prev_ep != begin) { | |
13280 | --prev_ep; | |
13281 | any_change = true; | |
13282 | } else { | |
13283 | prev_ep = end; // to avoid useless first extent re-check | |
13284 | } | |
13285 | } // if (prev_ep != end && prev_ep->logical_offset >= min_off) | |
13286 | } while (any_change); | |
13287 | ||
eafe8130 TL |
13288 | if (above_blob_threshold) { |
13289 | dout(10) << __func__ << " request GC, blobs >= " << inspected_blobs.size() | |
13290 | << " " << std::hex << min_off << "~" << max_off << std::dec | |
13291 | << dendl; | |
13292 | ceph_assert(start_ep != end_ep); | |
13293 | for (auto ep = start_ep; ep != end_ep; ++ep) { | |
13294 | dout(20) << __func__ << " inserting for GC " | |
13295 | << std::hex << ep->logical_offset << "~" << ep->length | |
13296 | << std::dec << dendl; | |
13297 | ||
13298 | wctx->extents_to_gc.union_insert(ep->logical_offset, ep->length); | |
13299 | } | |
13300 | // insert newly written extent to GC | |
13301 | wctx->extents_to_gc.union_insert(offset, length); | |
13302 | dout(20) << __func__ << " inserting (last) for GC " | |
13303 | << std::hex << offset << "~" << length | |
13304 | << std::dec << dendl; | |
13305 | } | |
7c673cae | 13306 | // new blob. |
7c673cae | 13307 | BlobRef b = c->new_blob(); |
11fdf7f2 | 13308 | uint64_t b_off = p2phase<uint64_t>(offset, alloc_len); |
7c673cae FG |
13309 | uint64_t b_off0 = b_off; |
13310 | _pad_zeros(&bl, &b_off0, block_size); | |
13311 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
1911f103 TL |
13312 | wctx->write(offset, b, alloc_len, b_off0, bl, b_off, length, |
13313 | min_alloc_size != block_size, // use 'unused' bitmap when alloc granularity | |
13314 | // doesn't match disk one only | |
13315 | true); | |
7c673cae FG |
13316 | |
13317 | return; | |
13318 | } | |
13319 | ||
13320 | void BlueStore::_do_write_big( | |
13321 | TransContext *txc, | |
13322 | CollectionRef &c, | |
13323 | OnodeRef o, | |
13324 | uint64_t offset, uint64_t length, | |
13325 | bufferlist::iterator& blp, | |
13326 | WriteContext *wctx) | |
13327 | { | |
13328 | dout(10) << __func__ << " 0x" << std::hex << offset << "~" << length | |
13329 | << " target_blob_size 0x" << wctx->target_blob_size << std::dec | |
13330 | << " compress " << (int)wctx->compress | |
13331 | << dendl; | |
13332 | logger->inc(l_bluestore_write_big); | |
13333 | logger->inc(l_bluestore_write_big_bytes, length); | |
13334 | o->extent_map.punch_hole(c, offset, length, &wctx->old_extents); | |
11fdf7f2 | 13335 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae FG |
13336 | while (length > 0) { |
13337 | bool new_blob = false; | |
11fdf7f2 | 13338 | uint32_t l = std::min(max_bsize, length); |
7c673cae FG |
13339 | BlobRef b; |
13340 | uint32_t b_off = 0; | |
13341 | ||
13342 | //attempting to reuse existing blob | |
13343 | if (!wctx->compress) { | |
13344 | // look for an existing mutable blob we can reuse | |
13345 | auto begin = o->extent_map.extent_map.begin(); | |
13346 | auto end = o->extent_map.extent_map.end(); | |
13347 | auto ep = o->extent_map.seek_lextent(offset); | |
13348 | auto prev_ep = ep; | |
13349 | if (prev_ep != begin) { | |
13350 | --prev_ep; | |
13351 | } else { | |
13352 | prev_ep = end; // to avoid this extent check as it's a duplicate | |
13353 | } | |
13354 | auto min_off = offset >= max_bsize ? offset - max_bsize : 0; | |
13355 | // search suitable extent in both forward and reverse direction in | |
13356 | // [offset - target_max_blob_size, offset + target_max_blob_size] range | |
224ce89b | 13357 | // then check if blob can be reused via can_reuse_blob func. |
7c673cae FG |
13358 | bool any_change; |
13359 | do { | |
13360 | any_change = false; | |
13361 | if (ep != end && ep->logical_offset < offset + max_bsize) { | |
13362 | if (offset >= ep->blob_start() && | |
224ce89b | 13363 | ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13364 | offset - ep->blob_start(), |
13365 | &l)) { | |
13366 | b = ep->blob; | |
13367 | b_off = offset - ep->blob_start(); | |
13368 | prev_ep = end; // to avoid check below | |
13369 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13370 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13371 | } else { |
13372 | ++ep; | |
13373 | any_change = true; | |
13374 | } | |
13375 | } | |
13376 | ||
13377 | if (prev_ep != end && prev_ep->logical_offset >= min_off) { | |
224ce89b | 13378 | if (prev_ep->blob->can_reuse_blob(min_alloc_size, max_bsize, |
7c673cae FG |
13379 | offset - prev_ep->blob_start(), |
13380 | &l)) { | |
13381 | b = prev_ep->blob; | |
13382 | b_off = offset - prev_ep->blob_start(); | |
13383 | dout(20) << __func__ << " reuse blob " << *b << std::hex | |
224ce89b | 13384 | << " (0x" << b_off << "~" << l << ")" << std::dec << dendl; |
7c673cae FG |
13385 | } else if (prev_ep != begin) { |
13386 | --prev_ep; | |
13387 | any_change = true; | |
13388 | } else { | |
13389 | prev_ep = end; // to avoid useless first extent re-check | |
13390 | } | |
13391 | } | |
13392 | } while (b == nullptr && any_change); | |
13393 | } | |
13394 | if (b == nullptr) { | |
13395 | b = c->new_blob(); | |
13396 | b_off = 0; | |
13397 | new_blob = true; | |
13398 | } | |
13399 | ||
13400 | bufferlist t; | |
13401 | blp.copy(l, t); | |
13402 | wctx->write(offset, b, l, b_off, t, b_off, l, false, new_blob); | |
13403 | offset += l; | |
13404 | length -= l; | |
13405 | logger->inc(l_bluestore_write_big_blobs); | |
13406 | } | |
13407 | } | |
13408 | ||
13409 | int BlueStore::_do_alloc_write( | |
13410 | TransContext *txc, | |
13411 | CollectionRef coll, | |
13412 | OnodeRef o, | |
13413 | WriteContext *wctx) | |
13414 | { | |
13415 | dout(20) << __func__ << " txc " << txc | |
13416 | << " " << wctx->writes.size() << " blobs" | |
13417 | << dendl; | |
3efd9988 FG |
13418 | if (wctx->writes.empty()) { |
13419 | return 0; | |
7c673cae FG |
13420 | } |
13421 | ||
7c673cae FG |
13422 | CompressorRef c; |
13423 | double crr = 0; | |
13424 | if (wctx->compress) { | |
13425 | c = select_option( | |
13426 | "compression_algorithm", | |
13427 | compressor, | |
13428 | [&]() { | |
13429 | string val; | |
13430 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_ALGORITHM, &val)) { | |
13431 | CompressorRef cp = compressor; | |
13432 | if (!cp || cp->get_type_name() != val) { | |
13433 | cp = Compressor::create(cct, val); | |
11fdf7f2 TL |
13434 | if (!cp) { |
13435 | if (_set_compression_alert(false, val.c_str())) { | |
13436 | derr << __func__ << " unable to initialize " << val.c_str() | |
13437 | << " compressor" << dendl; | |
13438 | } | |
13439 | } | |
7c673cae FG |
13440 | } |
13441 | return boost::optional<CompressorRef>(cp); | |
13442 | } | |
13443 | return boost::optional<CompressorRef>(); | |
13444 | } | |
13445 | ); | |
13446 | ||
13447 | crr = select_option( | |
13448 | "compression_required_ratio", | |
13449 | cct->_conf->bluestore_compression_required_ratio, | |
13450 | [&]() { | |
13451 | double val; | |
3efd9988 | 13452 | if (coll->pool_opts.get(pool_opts_t::COMPRESSION_REQUIRED_RATIO, &val)) { |
7c673cae FG |
13453 | return boost::optional<double>(val); |
13454 | } | |
13455 | return boost::optional<double>(); | |
13456 | } | |
13457 | ); | |
13458 | } | |
13459 | ||
13460 | // checksum | |
11fdf7f2 | 13461 | int64_t csum = csum_type.load(); |
7c673cae FG |
13462 | csum = select_option( |
13463 | "csum_type", | |
13464 | csum, | |
13465 | [&]() { | |
11fdf7f2 | 13466 | int64_t val; |
3efd9988 | 13467 | if (coll->pool_opts.get(pool_opts_t::CSUM_TYPE, &val)) { |
11fdf7f2 | 13468 | return boost::optional<int64_t>(val); |
7c673cae | 13469 | } |
11fdf7f2 | 13470 | return boost::optional<int64_t>(); |
7c673cae FG |
13471 | } |
13472 | ); | |
13473 | ||
3efd9988 FG |
13474 | // compress (as needed) and calc needed space |
13475 | uint64_t need = 0; | |
11fdf7f2 | 13476 | auto max_bsize = std::max(wctx->target_blob_size, min_alloc_size); |
7c673cae | 13477 | for (auto& wi : wctx->writes) { |
3efd9988 | 13478 | if (c && wi.blob_length > min_alloc_size) { |
11fdf7f2 | 13479 | auto start = mono_clock::now(); |
7c673cae FG |
13480 | |
13481 | // compress | |
11fdf7f2 TL |
13482 | ceph_assert(wi.b_off == 0); |
13483 | ceph_assert(wi.blob_length == wi.bl.length()); | |
3efd9988 | 13484 | |
7c673cae FG |
13485 | // FIXME: memory alignment here is bad |
13486 | bufferlist t; | |
3efd9988 | 13487 | int r = c->compress(wi.bl, t); |
3efd9988 | 13488 | uint64_t want_len_raw = wi.blob_length * crr; |
11fdf7f2 | 13489 | uint64_t want_len = p2roundup(want_len_raw, min_alloc_size); |
a8e16298 TL |
13490 | bool rejected = false; |
13491 | uint64_t compressed_len = t.length(); | |
13492 | // do an approximate (fast) estimation for resulting blob size | |
13493 | // that doesn't take header overhead into account | |
11fdf7f2 | 13494 | uint64_t result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13495 | if (r == 0 && result_len <= want_len && result_len < wi.blob_length) { |
13496 | bluestore_compression_header_t chdr; | |
13497 | chdr.type = c->get_type(); | |
13498 | chdr.length = t.length(); | |
13499 | encode(chdr, wi.compressed_bl); | |
13500 | wi.compressed_bl.claim_append(t); | |
13501 | ||
13502 | compressed_len = wi.compressed_bl.length(); | |
11fdf7f2 | 13503 | result_len = p2roundup(compressed_len, min_alloc_size); |
a8e16298 TL |
13504 | if (result_len <= want_len && result_len < wi.blob_length) { |
13505 | // Cool. We compressed at least as much as we were hoping to. | |
13506 | // pad out to min_alloc_size | |
13507 | wi.compressed_bl.append_zero(result_len - compressed_len); | |
13508 | wi.compressed_len = compressed_len; | |
13509 | wi.compressed = true; | |
13510 | logger->inc(l_bluestore_write_pad_bytes, result_len - compressed_len); | |
13511 | dout(20) << __func__ << std::hex << " compressed 0x" << wi.blob_length | |
13512 | << " -> 0x" << compressed_len << " => 0x" << result_len | |
13513 | << " with " << c->get_type() | |
13514 | << std::dec << dendl; | |
13515 | txc->statfs_delta.compressed() += compressed_len; | |
13516 | txc->statfs_delta.compressed_original() += wi.blob_length; | |
13517 | txc->statfs_delta.compressed_allocated() += result_len; | |
13518 | logger->inc(l_bluestore_compress_success_count); | |
13519 | need += result_len; | |
13520 | } else { | |
13521 | rejected = true; | |
13522 | } | |
13523 | } else if (r != 0) { | |
13524 | dout(5) << __func__ << std::hex << " 0x" << wi.blob_length | |
13525 | << " bytes compressed using " << c->get_type_name() | |
13526 | << std::dec | |
13527 | << " failed with errcode = " << r | |
13528 | << ", leaving uncompressed" | |
13529 | << dendl; | |
13530 | logger->inc(l_bluestore_compress_rejected_count); | |
13531 | need += wi.blob_length; | |
7c673cae | 13532 | } else { |
a8e16298 TL |
13533 | rejected = true; |
13534 | } | |
13535 | ||
13536 | if (rejected) { | |
3efd9988 | 13537 | dout(20) << __func__ << std::hex << " 0x" << wi.blob_length |
a8e16298 | 13538 | << " compressed to 0x" << compressed_len << " -> 0x" << result_len |
3efd9988 FG |
13539 | << " with " << c->get_type() |
13540 | << ", which is more than required 0x" << want_len_raw | |
7c673cae | 13541 | << " -> 0x" << want_len |
3efd9988 FG |
13542 | << ", leaving uncompressed" |
13543 | << std::dec << dendl; | |
13544 | logger->inc(l_bluestore_compress_rejected_count); | |
13545 | need += wi.blob_length; | |
7c673cae | 13546 | } |
494da23a TL |
13547 | log_latency("compress@_do_alloc_write", |
13548 | l_bluestore_compress_lat, | |
13549 | mono_clock::now() - start, | |
13550 | cct->_conf->bluestore_log_op_age ); | |
3efd9988 FG |
13551 | } else { |
13552 | need += wi.blob_length; | |
7c673cae | 13553 | } |
3efd9988 | 13554 | } |
a8e16298 | 13555 | PExtentVector prealloc; |
3efd9988 | 13556 | prealloc.reserve(2 * wctx->writes.size());; |
11fdf7f2 | 13557 | int64_t prealloc_left = 0; |
3efd9988 FG |
13558 | prealloc_left = alloc->allocate( |
13559 | need, min_alloc_size, need, | |
13560 | 0, &prealloc); | |
eafe8130 | 13561 | if (prealloc_left < 0 || prealloc_left < (int64_t)need) { |
11fdf7f2 | 13562 | derr << __func__ << " failed to allocate 0x" << std::hex << need |
eafe8130 | 13563 | << " allocated 0x " << (prealloc_left < 0 ? 0 : prealloc_left) |
11fdf7f2 TL |
13564 | << " min_alloc_size 0x" << min_alloc_size |
13565 | << " available 0x " << alloc->get_free() | |
13566 | << std::dec << dendl; | |
13567 | if (prealloc.size()) { | |
13568 | alloc->release(prealloc); | |
13569 | } | |
a8e16298 TL |
13570 | return -ENOSPC; |
13571 | } | |
9f95a23c | 13572 | _collect_allocation_stats(need, min_alloc_size, prealloc.size()); |
a8e16298 | 13573 | |
3efd9988 FG |
13574 | dout(20) << __func__ << " prealloc " << prealloc << dendl; |
13575 | auto prealloc_pos = prealloc.begin(); | |
13576 | ||
13577 | for (auto& wi : wctx->writes) { | |
13578 | BlobRef b = wi.b; | |
13579 | bluestore_blob_t& dblob = b->dirty_blob(); | |
13580 | uint64_t b_off = wi.b_off; | |
13581 | bufferlist *l = &wi.bl; | |
13582 | uint64_t final_length = wi.blob_length; | |
13583 | uint64_t csum_length = wi.blob_length; | |
3efd9988 FG |
13584 | if (wi.compressed) { |
13585 | final_length = wi.compressed_bl.length(); | |
13586 | csum_length = final_length; | |
3efd9988 FG |
13587 | l = &wi.compressed_bl; |
13588 | dblob.set_compressed(wi.blob_length, wi.compressed_len); | |
13589 | } else if (wi.new_blob) { | |
7c673cae | 13590 | // initialize newly created blob only |
11fdf7f2 TL |
13591 | ceph_assert(dblob.is_mutable()); |
13592 | unsigned csum_order; | |
7c673cae FG |
13593 | if (l->length() != wi.blob_length) { |
13594 | // hrm, maybe we could do better here, but let's not bother. | |
13595 | dout(20) << __func__ << " forcing csum_order to block_size_order " | |
13596 | << block_size_order << dendl; | |
31f18b77 | 13597 | csum_order = block_size_order; |
7c673cae FG |
13598 | } else { |
13599 | csum_order = std::min(wctx->csum_order, ctz(l->length())); | |
13600 | } | |
13601 | // try to align blob with max_blob_size to improve | |
13602 | // its reuse ratio, e.g. in case of reverse write | |
13603 | uint32_t suggested_boff = | |
13604 | (wi.logical_offset - (wi.b_off0 - wi.b_off)) % max_bsize; | |
13605 | if ((suggested_boff % (1 << csum_order)) == 0 && | |
13606 | suggested_boff + final_length <= max_bsize && | |
13607 | suggested_boff > b_off) { | |
181888fb | 13608 | dout(20) << __func__ << " forcing blob_offset to 0x" |
7c673cae | 13609 | << std::hex << suggested_boff << std::dec << dendl; |
11fdf7f2 | 13610 | ceph_assert(suggested_boff >= b_off); |
7c673cae FG |
13611 | csum_length += suggested_boff - b_off; |
13612 | b_off = suggested_boff; | |
13613 | } | |
181888fb FG |
13614 | if (csum != Checksummer::CSUM_NONE) { |
13615 | dout(20) << __func__ << " initialize csum setting for new blob " << *b | |
13616 | << " csum_type " << Checksummer::get_csum_type_string(csum) | |
13617 | << " csum_order " << csum_order | |
13618 | << " csum_length 0x" << std::hex << csum_length << std::dec | |
13619 | << dendl; | |
13620 | dblob.init_csum(csum, csum_order, csum_length); | |
13621 | } | |
7c673cae FG |
13622 | } |
13623 | ||
a8e16298 | 13624 | PExtentVector extents; |
3efd9988 FG |
13625 | int64_t left = final_length; |
13626 | while (left > 0) { | |
11fdf7f2 | 13627 | ceph_assert(prealloc_left > 0); |
3efd9988 FG |
13628 | if (prealloc_pos->length <= left) { |
13629 | prealloc_left -= prealloc_pos->length; | |
13630 | left -= prealloc_pos->length; | |
13631 | txc->statfs_delta.allocated() += prealloc_pos->length; | |
13632 | extents.push_back(*prealloc_pos); | |
13633 | ++prealloc_pos; | |
13634 | } else { | |
13635 | extents.emplace_back(prealloc_pos->offset, left); | |
13636 | prealloc_pos->offset += left; | |
13637 | prealloc_pos->length -= left; | |
13638 | prealloc_left -= left; | |
13639 | txc->statfs_delta.allocated() += left; | |
13640 | left = 0; | |
13641 | break; | |
13642 | } | |
13643 | } | |
7c673cae | 13644 | for (auto& p : extents) { |
3efd9988 | 13645 | txc->allocated.insert(p.offset, p.length); |
7c673cae | 13646 | } |
11fdf7f2 | 13647 | dblob.allocated(p2align(b_off, min_alloc_size), final_length, extents); |
7c673cae | 13648 | |
181888fb FG |
13649 | dout(20) << __func__ << " blob " << *b << dendl; |
13650 | if (dblob.has_csum()) { | |
7c673cae FG |
13651 | dblob.calc_csum(b_off, *l); |
13652 | } | |
181888fb | 13653 | |
7c673cae | 13654 | if (wi.mark_unused) { |
1911f103 | 13655 | ceph_assert(!dblob.is_compressed()); |
7c673cae FG |
13656 | auto b_end = b_off + wi.bl.length(); |
13657 | if (b_off) { | |
13658 | dblob.add_unused(0, b_off); | |
13659 | } | |
1911f103 TL |
13660 | uint64_t llen = dblob.get_logical_length(); |
13661 | if (b_end < llen) { | |
13662 | dblob.add_unused(b_end, llen - b_end); | |
7c673cae FG |
13663 | } |
13664 | } | |
13665 | ||
13666 | Extent *le = o->extent_map.set_lextent(coll, wi.logical_offset, | |
13667 | b_off + (wi.b_off0 - wi.b_off), | |
13668 | wi.length0, | |
13669 | wi.b, | |
13670 | nullptr); | |
13671 | wi.b->dirty_blob().mark_used(le->blob_offset, le->length); | |
13672 | txc->statfs_delta.stored() += le->length; | |
13673 | dout(20) << __func__ << " lex " << *le << dendl; | |
13674 | _buffer_cache_write(txc, wi.b, b_off, wi.bl, | |
13675 | wctx->buffered ? 0 : Buffer::FLAG_NOCACHE); | |
13676 | ||
13677 | // queue io | |
11fdf7f2 | 13678 | if (!g_conf()->bluestore_debug_omit_block_device_write) { |
7c673cae FG |
13679 | if (l->length() <= prefer_deferred_size.load()) { |
13680 | dout(20) << __func__ << " deferring small 0x" << std::hex | |
13681 | << l->length() << std::dec << " write via deferred" << dendl; | |
9f95a23c | 13682 | bluestore_deferred_op_t *op = _get_deferred_op(txc); |
7c673cae FG |
13683 | op->op = bluestore_deferred_op_t::OP_WRITE; |
13684 | int r = b->get_blob().map( | |
13685 | b_off, l->length(), | |
13686 | [&](uint64_t offset, uint64_t length) { | |
13687 | op->extents.emplace_back(bluestore_pextent_t(offset, length)); | |
13688 | return 0; | |
13689 | }); | |
11fdf7f2 | 13690 | ceph_assert(r == 0); |
7c673cae | 13691 | op->data = *l; |
81eedcae | 13692 | logger->inc(l_bluestore_write_small_deferred); |
7c673cae FG |
13693 | } else { |
13694 | b->get_blob().map_bl( | |
13695 | b_off, *l, | |
13696 | [&](uint64_t offset, bufferlist& t) { | |
13697 | bdev->aio_write(offset, t, &txc->ioc, false); | |
13698 | }); | |
81eedcae | 13699 | logger->inc(l_bluestore_write_small_new); |
7c673cae FG |
13700 | } |
13701 | } | |
13702 | } | |
11fdf7f2 TL |
13703 | ceph_assert(prealloc_pos == prealloc.end()); |
13704 | ceph_assert(prealloc_left == 0); | |
7c673cae FG |
13705 | return 0; |
13706 | } | |
13707 | ||
13708 | void BlueStore::_wctx_finish( | |
13709 | TransContext *txc, | |
13710 | CollectionRef& c, | |
13711 | OnodeRef o, | |
31f18b77 FG |
13712 | WriteContext *wctx, |
13713 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
13714 | { |
13715 | auto oep = wctx->old_extents.begin(); | |
13716 | while (oep != wctx->old_extents.end()) { | |
13717 | auto &lo = *oep; | |
13718 | oep = wctx->old_extents.erase(oep); | |
13719 | dout(20) << __func__ << " lex_old " << lo.e << dendl; | |
13720 | BlobRef b = lo.e.blob; | |
13721 | const bluestore_blob_t& blob = b->get_blob(); | |
13722 | if (blob.is_compressed()) { | |
13723 | if (lo.blob_empty) { | |
13724 | txc->statfs_delta.compressed() -= blob.get_compressed_payload_length(); | |
13725 | } | |
13726 | txc->statfs_delta.compressed_original() -= lo.e.length; | |
13727 | } | |
13728 | auto& r = lo.r; | |
13729 | txc->statfs_delta.stored() -= lo.e.length; | |
13730 | if (!r.empty()) { | |
13731 | dout(20) << __func__ << " blob release " << r << dendl; | |
13732 | if (blob.is_shared()) { | |
13733 | PExtentVector final; | |
13734 | c->load_shared_blob(b->shared_blob); | |
11fdf7f2 TL |
13735 | bool unshare = false; |
13736 | bool* unshare_ptr = | |
13737 | !maybe_unshared_blobs || b->is_referenced() ? nullptr : &unshare; | |
7c673cae | 13738 | for (auto e : r) { |
31f18b77 FG |
13739 | b->shared_blob->put_ref( |
13740 | e.offset, e.length, &final, | |
11fdf7f2 TL |
13741 | unshare_ptr); |
13742 | } | |
13743 | if (unshare) { | |
13744 | ceph_assert(maybe_unshared_blobs); | |
13745 | maybe_unshared_blobs->insert(b->shared_blob.get()); | |
7c673cae FG |
13746 | } |
13747 | dout(20) << __func__ << " shared_blob release " << final | |
13748 | << " from " << *b->shared_blob << dendl; | |
13749 | txc->write_shared_blob(b->shared_blob); | |
13750 | r.clear(); | |
13751 | r.swap(final); | |
13752 | } | |
13753 | } | |
13754 | // we can't invalidate our logical extents as we drop them because | |
13755 | // other lextents (either in our onode or others) may still | |
13756 | // reference them. but we can throw out anything that is no | |
13757 | // longer allocated. Note that this will leave behind edge bits | |
13758 | // that are no longer referenced but not deallocated (until they | |
13759 | // age out of the cache naturally). | |
13760 | b->discard_unallocated(c.get()); | |
13761 | for (auto e : r) { | |
13762 | dout(20) << __func__ << " release " << e << dendl; | |
13763 | txc->released.insert(e.offset, e.length); | |
13764 | txc->statfs_delta.allocated() -= e.length; | |
13765 | if (blob.is_compressed()) { | |
13766 | txc->statfs_delta.compressed_allocated() -= e.length; | |
13767 | } | |
13768 | } | |
9f95a23c TL |
13769 | |
13770 | if (b->is_spanning() && !b->is_referenced() && lo.blob_empty) { | |
7c673cae FG |
13771 | dout(20) << __func__ << " spanning_blob_map removing empty " << *b |
13772 | << dendl; | |
13773 | o->extent_map.spanning_blob_map.erase(b->id); | |
13774 | } | |
9f95a23c | 13775 | delete &lo; |
7c673cae FG |
13776 | } |
13777 | } | |
13778 | ||
13779 | void BlueStore::_do_write_data( | |
13780 | TransContext *txc, | |
13781 | CollectionRef& c, | |
13782 | OnodeRef o, | |
13783 | uint64_t offset, | |
13784 | uint64_t length, | |
13785 | bufferlist& bl, | |
13786 | WriteContext *wctx) | |
13787 | { | |
13788 | uint64_t end = offset + length; | |
13789 | bufferlist::iterator p = bl.begin(); | |
13790 | ||
13791 | if (offset / min_alloc_size == (end - 1) / min_alloc_size && | |
13792 | (length != min_alloc_size)) { | |
13793 | // we fall within the same block | |
13794 | _do_write_small(txc, c, o, offset, length, p, wctx); | |
13795 | } else { | |
13796 | uint64_t head_offset, head_length; | |
13797 | uint64_t middle_offset, middle_length; | |
13798 | uint64_t tail_offset, tail_length; | |
13799 | ||
13800 | head_offset = offset; | |
11fdf7f2 | 13801 | head_length = p2nphase(offset, min_alloc_size); |
7c673cae | 13802 | |
11fdf7f2 TL |
13803 | tail_offset = p2align(end, min_alloc_size); |
13804 | tail_length = p2phase(end, min_alloc_size); | |
7c673cae FG |
13805 | |
13806 | middle_offset = head_offset + head_length; | |
13807 | middle_length = length - head_length - tail_length; | |
13808 | ||
13809 | if (head_length) { | |
13810 | _do_write_small(txc, c, o, head_offset, head_length, p, wctx); | |
13811 | } | |
13812 | ||
13813 | if (middle_length) { | |
13814 | _do_write_big(txc, c, o, middle_offset, middle_length, p, wctx); | |
13815 | } | |
13816 | ||
13817 | if (tail_length) { | |
13818 | _do_write_small(txc, c, o, tail_offset, tail_length, p, wctx); | |
13819 | } | |
13820 | } | |
13821 | } | |
13822 | ||
31f18b77 FG |
13823 | void BlueStore::_choose_write_options( |
13824 | CollectionRef& c, | |
13825 | OnodeRef o, | |
13826 | uint32_t fadvise_flags, | |
13827 | WriteContext *wctx) | |
7c673cae | 13828 | { |
7c673cae FG |
13829 | if (fadvise_flags & CEPH_OSD_OP_FLAG_FADVISE_WILLNEED) { |
13830 | dout(20) << __func__ << " will do buffered write" << dendl; | |
31f18b77 | 13831 | wctx->buffered = true; |
7c673cae FG |
13832 | } else if (cct->_conf->bluestore_default_buffered_write && |
13833 | (fadvise_flags & (CEPH_OSD_OP_FLAG_FADVISE_DONTNEED | | |
13834 | CEPH_OSD_OP_FLAG_FADVISE_NOCACHE)) == 0) { | |
13835 | dout(20) << __func__ << " defaulting to buffered write" << dendl; | |
31f18b77 | 13836 | wctx->buffered = true; |
7c673cae FG |
13837 | } |
13838 | ||
31f18b77 FG |
13839 | // apply basic csum block size |
13840 | wctx->csum_order = block_size_order; | |
7c673cae FG |
13841 | |
13842 | // compression parameters | |
13843 | unsigned alloc_hints = o->onode.alloc_hint_flags; | |
13844 | auto cm = select_option( | |
13845 | "compression_mode", | |
31f18b77 | 13846 | comp_mode.load(), |
7c673cae FG |
13847 | [&]() { |
13848 | string val; | |
11fdf7f2 | 13849 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MODE, &val)) { |
31f18b77 FG |
13850 | return boost::optional<Compressor::CompressionMode>( |
13851 | Compressor::get_comp_mode_type(val)); | |
7c673cae FG |
13852 | } |
13853 | return boost::optional<Compressor::CompressionMode>(); | |
13854 | } | |
13855 | ); | |
31f18b77 FG |
13856 | |
13857 | wctx->compress = (cm != Compressor::COMP_NONE) && | |
7c673cae FG |
13858 | ((cm == Compressor::COMP_FORCE) || |
13859 | (cm == Compressor::COMP_AGGRESSIVE && | |
13860 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_INCOMPRESSIBLE) == 0) || | |
13861 | (cm == Compressor::COMP_PASSIVE && | |
13862 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_COMPRESSIBLE))); | |
13863 | ||
13864 | if ((alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_SEQUENTIAL_READ) && | |
13865 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_READ) == 0 && | |
31f18b77 FG |
13866 | (alloc_hints & (CEPH_OSD_ALLOC_HINT_FLAG_IMMUTABLE | |
13867 | CEPH_OSD_ALLOC_HINT_FLAG_APPEND_ONLY)) && | |
7c673cae | 13868 | (alloc_hints & CEPH_OSD_ALLOC_HINT_FLAG_RANDOM_WRITE) == 0) { |
31f18b77 | 13869 | |
7c673cae | 13870 | dout(20) << __func__ << " will prefer large blob and csum sizes" << dendl; |
31f18b77 | 13871 | |
7c673cae | 13872 | if (o->onode.expected_write_size) { |
224ce89b | 13873 | wctx->csum_order = std::max(min_alloc_size_order, |
31f18b77 | 13874 | (uint8_t)ctz(o->onode.expected_write_size)); |
7c673cae | 13875 | } else { |
224ce89b | 13876 | wctx->csum_order = min_alloc_size_order; |
7c673cae FG |
13877 | } |
13878 | ||
31f18b77 FG |
13879 | if (wctx->compress) { |
13880 | wctx->target_blob_size = select_option( | |
7c673cae | 13881 | "compression_max_blob_size", |
31f18b77 | 13882 | comp_max_blob_size.load(), |
7c673cae | 13883 | [&]() { |
11fdf7f2 TL |
13884 | int64_t val; |
13885 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MAX_BLOB_SIZE, &val)) { | |
7c673cae FG |
13886 | return boost::optional<uint64_t>((uint64_t)val); |
13887 | } | |
13888 | return boost::optional<uint64_t>(); | |
13889 | } | |
13890 | ); | |
13891 | } | |
13892 | } else { | |
31f18b77 FG |
13893 | if (wctx->compress) { |
13894 | wctx->target_blob_size = select_option( | |
7c673cae | 13895 | "compression_min_blob_size", |
31f18b77 | 13896 | comp_min_blob_size.load(), |
7c673cae | 13897 | [&]() { |
11fdf7f2 TL |
13898 | int64_t val; |
13899 | if (c->pool_opts.get(pool_opts_t::COMPRESSION_MIN_BLOB_SIZE, &val)) { | |
7c673cae FG |
13900 | return boost::optional<uint64_t>((uint64_t)val); |
13901 | } | |
13902 | return boost::optional<uint64_t>(); | |
13903 | } | |
13904 | ); | |
13905 | } | |
13906 | } | |
31f18b77 | 13907 | |
7c673cae | 13908 | uint64_t max_bsize = max_blob_size.load(); |
31f18b77 FG |
13909 | if (wctx->target_blob_size == 0 || wctx->target_blob_size > max_bsize) { |
13910 | wctx->target_blob_size = max_bsize; | |
7c673cae | 13911 | } |
31f18b77 | 13912 | |
7c673cae FG |
13913 | // set the min blob size floor at 2x the min_alloc_size, or else we |
13914 | // won't be able to allocate a smaller extent for the compressed | |
13915 | // data. | |
31f18b77 FG |
13916 | if (wctx->compress && |
13917 | wctx->target_blob_size < min_alloc_size * 2) { | |
13918 | wctx->target_blob_size = min_alloc_size * 2; | |
7c673cae | 13919 | } |
31f18b77 FG |
13920 | |
13921 | dout(20) << __func__ << " prefer csum_order " << wctx->csum_order | |
13922 | << " target_blob_size 0x" << std::hex << wctx->target_blob_size | |
f64942e4 AA |
13923 | << " compress=" << (int)wctx->compress |
13924 | << " buffered=" << (int)wctx->buffered | |
31f18b77 FG |
13925 | << std::dec << dendl; |
13926 | } | |
13927 | ||
13928 | int BlueStore::_do_gc( | |
13929 | TransContext *txc, | |
13930 | CollectionRef& c, | |
13931 | OnodeRef o, | |
31f18b77 FG |
13932 | const WriteContext& wctx, |
13933 | uint64_t *dirty_start, | |
13934 | uint64_t *dirty_end) | |
13935 | { | |
31f18b77 | 13936 | |
1adf2230 | 13937 | bool dirty_range_updated = false; |
31f18b77 | 13938 | WriteContext wctx_gc; |
7c673cae | 13939 | wctx_gc.fork(wctx); // make a clone for garbage collection |
7c673cae | 13940 | |
eafe8130 | 13941 | auto & extents_to_collect = wctx.extents_to_gc; |
31f18b77 FG |
13942 | for (auto it = extents_to_collect.begin(); |
13943 | it != extents_to_collect.end(); | |
13944 | ++it) { | |
13945 | bufferlist bl; | |
eafe8130 TL |
13946 | auto offset = (*it).first; |
13947 | auto length = (*it).second; | |
13948 | dout(20) << __func__ << " processing " << std::hex | |
13949 | << offset << "~" << length << std::dec | |
13950 | << dendl; | |
13951 | int r = _do_read(c.get(), o, offset, length, bl, 0); | |
13952 | ceph_assert(r == (int)length); | |
31f18b77 | 13953 | |
eafe8130 TL |
13954 | _do_write_data(txc, c, o, offset, length, bl, &wctx_gc); |
13955 | logger->inc(l_bluestore_gc_merged, length); | |
31f18b77 | 13956 | |
eafe8130 TL |
13957 | if (*dirty_start > offset) { |
13958 | *dirty_start = offset; | |
1adf2230 | 13959 | dirty_range_updated = true; |
31f18b77 FG |
13960 | } |
13961 | ||
eafe8130 TL |
13962 | if (*dirty_end < offset + length) { |
13963 | *dirty_end = offset + length; | |
1adf2230 | 13964 | dirty_range_updated = true; |
31f18b77 FG |
13965 | } |
13966 | } | |
1adf2230 AA |
13967 | if (dirty_range_updated) { |
13968 | o->extent_map.fault_range(db, *dirty_start, *dirty_end); | |
13969 | } | |
31f18b77 FG |
13970 | |
13971 | dout(30) << __func__ << " alloc write" << dendl; | |
13972 | int r = _do_alloc_write(txc, c, o, &wctx_gc); | |
13973 | if (r < 0) { | |
13974 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
13975 | << dendl; | |
13976 | return r; | |
13977 | } | |
13978 | ||
13979 | _wctx_finish(txc, c, o, &wctx_gc); | |
13980 | return 0; | |
13981 | } | |
13982 | ||
13983 | int BlueStore::_do_write( | |
13984 | TransContext *txc, | |
13985 | CollectionRef& c, | |
13986 | OnodeRef o, | |
13987 | uint64_t offset, | |
13988 | uint64_t length, | |
13989 | bufferlist& bl, | |
13990 | uint32_t fadvise_flags) | |
13991 | { | |
13992 | int r = 0; | |
13993 | ||
13994 | dout(20) << __func__ | |
13995 | << " " << o->oid | |
13996 | << " 0x" << std::hex << offset << "~" << length | |
13997 | << " - have 0x" << o->onode.size | |
13998 | << " (" << std::dec << o->onode.size << ")" | |
13999 | << " bytes" | |
14000 | << " fadvise_flags 0x" << std::hex << fadvise_flags << std::dec | |
14001 | << dendl; | |
81eedcae | 14002 | _dump_onode<30>(cct, *o); |
31f18b77 FG |
14003 | |
14004 | if (length == 0) { | |
14005 | return 0; | |
14006 | } | |
14007 | ||
14008 | uint64_t end = offset + length; | |
14009 | ||
14010 | GarbageCollector gc(c->store->cct); | |
eafe8130 | 14011 | int64_t benefit = 0; |
31f18b77 FG |
14012 | auto dirty_start = offset; |
14013 | auto dirty_end = end; | |
14014 | ||
14015 | WriteContext wctx; | |
14016 | _choose_write_options(c, o, fadvise_flags, &wctx); | |
7c673cae FG |
14017 | o->extent_map.fault_range(db, offset, length); |
14018 | _do_write_data(txc, c, o, offset, length, bl, &wctx); | |
7c673cae FG |
14019 | r = _do_alloc_write(txc, c, o, &wctx); |
14020 | if (r < 0) { | |
14021 | derr << __func__ << " _do_alloc_write failed with " << cpp_strerror(r) | |
14022 | << dendl; | |
14023 | goto out; | |
14024 | } | |
14025 | ||
eafe8130 TL |
14026 | if (wctx.extents_to_gc.empty() || |
14027 | wctx.extents_to_gc.range_start() > offset || | |
14028 | wctx.extents_to_gc.range_end() < offset + length) { | |
14029 | benefit = gc.estimate(offset, | |
14030 | length, | |
14031 | o->extent_map, | |
14032 | wctx.old_extents, | |
14033 | min_alloc_size); | |
14034 | } | |
14035 | ||
31f18b77 FG |
14036 | // NB: _wctx_finish() will empty old_extents |
14037 | // so we must do gc estimation before that | |
7c673cae FG |
14038 | _wctx_finish(txc, c, o, &wctx); |
14039 | if (end > o->onode.size) { | |
14040 | dout(20) << __func__ << " extending size to 0x" << std::hex << end | |
31f18b77 | 14041 | << std::dec << dendl; |
7c673cae FG |
14042 | o->onode.size = end; |
14043 | } | |
14044 | ||
11fdf7f2 | 14045 | if (benefit >= g_conf()->bluestore_gc_enable_total_threshold) { |
eafe8130 TL |
14046 | wctx.extents_to_gc.union_of(gc.get_extents_to_collect()); |
14047 | dout(20) << __func__ | |
14048 | << " perform garbage collection for compressed extents, " | |
14049 | << "expected benefit = " << benefit << " AUs" << dendl; | |
14050 | } | |
14051 | if (!wctx.extents_to_gc.empty()) { | |
14052 | dout(20) << __func__ << " perform garbage collection" << dendl; | |
14053 | ||
14054 | r = _do_gc(txc, c, o, | |
14055 | wctx, | |
14056 | &dirty_start, &dirty_end); | |
14057 | if (r < 0) { | |
14058 | derr << __func__ << " _do_gc failed with " << cpp_strerror(r) | |
14059 | << dendl; | |
14060 | goto out; | |
7c673cae | 14061 | } |
eafe8130 TL |
14062 | dout(20)<<__func__<<" gc range is " << std::hex << dirty_start |
14063 | << "~" << dirty_end - dirty_start << std::dec << dendl; | |
7c673cae | 14064 | } |
7c673cae | 14065 | o->extent_map.compress_extent_map(dirty_start, dirty_end - dirty_start); |
31f18b77 FG |
14066 | o->extent_map.dirty_range(dirty_start, dirty_end - dirty_start); |
14067 | ||
7c673cae FG |
14068 | r = 0; |
14069 | ||
14070 | out: | |
14071 | return r; | |
14072 | } | |
14073 | ||
14074 | int BlueStore::_write(TransContext *txc, | |
14075 | CollectionRef& c, | |
14076 | OnodeRef& o, | |
31f18b77 FG |
14077 | uint64_t offset, size_t length, |
14078 | bufferlist& bl, | |
14079 | uint32_t fadvise_flags) | |
7c673cae FG |
14080 | { |
14081 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14082 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14083 | << dendl; | |
35e4c445 FG |
14084 | int r = 0; |
14085 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14086 | r = -E2BIG; | |
14087 | } else { | |
14088 | _assign_nid(txc, o); | |
14089 | r = _do_write(txc, c, o, offset, length, bl, fadvise_flags); | |
14090 | txc->write_onode(o); | |
14091 | } | |
7c673cae FG |
14092 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14093 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14094 | << " = " << r << dendl; | |
14095 | return r; | |
14096 | } | |
14097 | ||
14098 | int BlueStore::_zero(TransContext *txc, | |
14099 | CollectionRef& c, | |
14100 | OnodeRef& o, | |
14101 | uint64_t offset, size_t length) | |
14102 | { | |
14103 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14104 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14105 | << dendl; | |
35e4c445 FG |
14106 | int r = 0; |
14107 | if (offset + length >= OBJECT_MAX_SIZE) { | |
14108 | r = -E2BIG; | |
14109 | } else { | |
14110 | _assign_nid(txc, o); | |
14111 | r = _do_zero(txc, c, o, offset, length); | |
14112 | } | |
7c673cae FG |
14113 | dout(10) << __func__ << " " << c->cid << " " << o->oid |
14114 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14115 | << " = " << r << dendl; | |
14116 | return r; | |
14117 | } | |
14118 | ||
14119 | int BlueStore::_do_zero(TransContext *txc, | |
14120 | CollectionRef& c, | |
14121 | OnodeRef& o, | |
14122 | uint64_t offset, size_t length) | |
14123 | { | |
14124 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14125 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14126 | << dendl; | |
14127 | int r = 0; | |
14128 | ||
81eedcae | 14129 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14130 | |
14131 | WriteContext wctx; | |
14132 | o->extent_map.fault_range(db, offset, length); | |
14133 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 | 14134 | o->extent_map.dirty_range(offset, length); |
7c673cae FG |
14135 | _wctx_finish(txc, c, o, &wctx); |
14136 | ||
b32b8144 | 14137 | if (length > 0 && offset + length > o->onode.size) { |
7c673cae FG |
14138 | o->onode.size = offset + length; |
14139 | dout(20) << __func__ << " extending size to " << offset + length | |
14140 | << dendl; | |
14141 | } | |
14142 | txc->write_onode(o); | |
14143 | ||
14144 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14145 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
14146 | << " = " << r << dendl; | |
14147 | return r; | |
14148 | } | |
14149 | ||
14150 | void BlueStore::_do_truncate( | |
31f18b77 FG |
14151 | TransContext *txc, CollectionRef& c, OnodeRef o, uint64_t offset, |
14152 | set<SharedBlob*> *maybe_unshared_blobs) | |
7c673cae FG |
14153 | { |
14154 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14155 | << " 0x" << std::hex << offset << std::dec << dendl; | |
14156 | ||
81eedcae | 14157 | _dump_onode<30>(cct, *o); |
7c673cae FG |
14158 | |
14159 | if (offset == o->onode.size) | |
31f18b77 | 14160 | return; |
7c673cae FG |
14161 | |
14162 | if (offset < o->onode.size) { | |
14163 | WriteContext wctx; | |
14164 | uint64_t length = o->onode.size - offset; | |
14165 | o->extent_map.fault_range(db, offset, length); | |
14166 | o->extent_map.punch_hole(c, offset, length, &wctx.old_extents); | |
31f18b77 FG |
14167 | o->extent_map.dirty_range(offset, length); |
14168 | _wctx_finish(txc, c, o, &wctx, maybe_unshared_blobs); | |
7c673cae FG |
14169 | |
14170 | // if we have shards past EOF, ask for a reshard | |
14171 | if (!o->onode.extent_map_shards.empty() && | |
14172 | o->onode.extent_map_shards.back().offset >= offset) { | |
14173 | dout(10) << __func__ << " request reshard past EOF" << dendl; | |
14174 | if (offset) { | |
14175 | o->extent_map.request_reshard(offset - 1, offset + length); | |
14176 | } else { | |
14177 | o->extent_map.request_reshard(0, length); | |
14178 | } | |
14179 | } | |
14180 | } | |
14181 | ||
14182 | o->onode.size = offset; | |
14183 | ||
14184 | txc->write_onode(o); | |
14185 | } | |
14186 | ||
35e4c445 | 14187 | int BlueStore::_truncate(TransContext *txc, |
7c673cae FG |
14188 | CollectionRef& c, |
14189 | OnodeRef& o, | |
14190 | uint64_t offset) | |
14191 | { | |
14192 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14193 | << " 0x" << std::hex << offset << std::dec | |
14194 | << dendl; | |
35e4c445 FG |
14195 | int r = 0; |
14196 | if (offset >= OBJECT_MAX_SIZE) { | |
14197 | r = -E2BIG; | |
14198 | } else { | |
14199 | _do_truncate(txc, c, o, offset); | |
14200 | } | |
14201 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14202 | << " 0x" << std::hex << offset << std::dec | |
14203 | << " = " << r << dendl; | |
14204 | return r; | |
7c673cae FG |
14205 | } |
14206 | ||
14207 | int BlueStore::_do_remove( | |
14208 | TransContext *txc, | |
14209 | CollectionRef& c, | |
14210 | OnodeRef o) | |
14211 | { | |
31f18b77 | 14212 | set<SharedBlob*> maybe_unshared_blobs; |
224ce89b WB |
14213 | bool is_gen = !o->oid.is_no_gen(); |
14214 | _do_truncate(txc, c, o, 0, is_gen ? &maybe_unshared_blobs : nullptr); | |
7c673cae FG |
14215 | if (o->onode.has_omap()) { |
14216 | o->flush(); | |
9f95a23c | 14217 | _do_omap_clear(txc, o); |
7c673cae FG |
14218 | } |
14219 | o->exists = false; | |
14220 | string key; | |
14221 | for (auto &s : o->extent_map.shards) { | |
14222 | dout(20) << __func__ << " removing shard 0x" << std::hex | |
14223 | << s.shard_info->offset << std::dec << dendl; | |
14224 | generate_extent_shard_key_and_apply(o->key, s.shard_info->offset, &key, | |
14225 | [&](const string& final_key) { | |
14226 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14227 | } | |
14228 | ); | |
14229 | } | |
14230 | txc->t->rmkey(PREFIX_OBJ, o->key.c_str(), o->key.size()); | |
a8e16298 | 14231 | txc->note_removed_object(o); |
7c673cae FG |
14232 | o->extent_map.clear(); |
14233 | o->onode = bluestore_onode_t(); | |
14234 | _debug_obj_on_delete(o->oid); | |
31f18b77 | 14235 | |
224ce89b WB |
14236 | if (!is_gen || maybe_unshared_blobs.empty()) { |
14237 | return 0; | |
14238 | } | |
31f18b77 | 14239 | |
224ce89b WB |
14240 | // see if we can unshare blobs still referenced by the head |
14241 | dout(10) << __func__ << " gen and maybe_unshared_blobs " | |
14242 | << maybe_unshared_blobs << dendl; | |
14243 | ghobject_t nogen = o->oid; | |
14244 | nogen.generation = ghobject_t::NO_GEN; | |
14245 | OnodeRef h = c->onode_map.lookup(nogen); | |
14246 | ||
14247 | if (!h || !h->exists) { | |
14248 | return 0; | |
14249 | } | |
14250 | ||
14251 | dout(20) << __func__ << " checking for unshareable blobs on " << h | |
14252 | << " " << h->oid << dendl; | |
14253 | map<SharedBlob*,bluestore_extent_ref_map_t> expect; | |
14254 | for (auto& e : h->extent_map.extent_map) { | |
14255 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14256 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14257 | if (b.is_shared() && | |
14258 | sb->loaded && | |
14259 | maybe_unshared_blobs.count(sb)) { | |
3efd9988 FG |
14260 | if (b.is_compressed()) { |
14261 | expect[sb].get(0, b.get_ondisk_length()); | |
14262 | } else { | |
14263 | b.map(e.blob_offset, e.length, [&](uint64_t off, uint64_t len) { | |
14264 | expect[sb].get(off, len); | |
14265 | return 0; | |
14266 | }); | |
14267 | } | |
224ce89b WB |
14268 | } |
14269 | } | |
31f18b77 | 14270 | |
224ce89b WB |
14271 | vector<SharedBlob*> unshared_blobs; |
14272 | unshared_blobs.reserve(maybe_unshared_blobs.size()); | |
14273 | for (auto& p : expect) { | |
14274 | dout(20) << " ? " << *p.first << " vs " << p.second << dendl; | |
14275 | if (p.first->persistent->ref_map == p.second) { | |
14276 | SharedBlob *sb = p.first; | |
14277 | dout(20) << __func__ << " unsharing " << *sb << dendl; | |
14278 | unshared_blobs.push_back(sb); | |
14279 | txc->unshare_blob(sb); | |
14280 | uint64_t sbid = c->make_blob_unshared(sb); | |
14281 | string key; | |
14282 | get_shared_blob_key(sbid, &key); | |
14283 | txc->t->rmkey(PREFIX_SHARED_BLOB, key); | |
14284 | } | |
14285 | } | |
14286 | ||
14287 | if (unshared_blobs.empty()) { | |
14288 | return 0; | |
14289 | } | |
14290 | ||
224ce89b WB |
14291 | for (auto& e : h->extent_map.extent_map) { |
14292 | const bluestore_blob_t& b = e.blob->get_blob(); | |
14293 | SharedBlob *sb = e.blob->shared_blob.get(); | |
14294 | if (b.is_shared() && | |
14295 | std::find(unshared_blobs.begin(), unshared_blobs.end(), | |
14296 | sb) != unshared_blobs.end()) { | |
14297 | dout(20) << __func__ << " unsharing " << e << dendl; | |
14298 | bluestore_blob_t& blob = e.blob->dirty_blob(); | |
14299 | blob.clear_flag(bluestore_blob_t::FLAG_SHARED); | |
d2e6a577 | 14300 | h->extent_map.dirty_range(e.logical_offset, 1); |
31f18b77 FG |
14301 | } |
14302 | } | |
224ce89b WB |
14303 | txc->write_onode(h); |
14304 | ||
7c673cae FG |
14305 | return 0; |
14306 | } | |
14307 | ||
14308 | int BlueStore::_remove(TransContext *txc, | |
14309 | CollectionRef& c, | |
14310 | OnodeRef &o) | |
14311 | { | |
11fdf7f2 TL |
14312 | dout(15) << __func__ << " " << c->cid << " " << o->oid |
14313 | << " onode " << o.get() | |
14314 | << " txc "<< txc << dendl; | |
7c673cae FG |
14315 | int r = _do_remove(txc, c, o); |
14316 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14317 | return r; | |
14318 | } | |
14319 | ||
14320 | int BlueStore::_setattr(TransContext *txc, | |
14321 | CollectionRef& c, | |
14322 | OnodeRef& o, | |
14323 | const string& name, | |
14324 | bufferptr& val) | |
14325 | { | |
14326 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14327 | << " " << name << " (" << val.length() << " bytes)" | |
14328 | << dendl; | |
14329 | int r = 0; | |
3efd9988 FG |
14330 | if (val.is_partial()) { |
14331 | auto& b = o->onode.attrs[name.c_str()] = bufferptr(val.c_str(), | |
14332 | val.length()); | |
14333 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
14334 | } else { | |
14335 | auto& b = o->onode.attrs[name.c_str()] = val; | |
14336 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
14337 | } | |
7c673cae FG |
14338 | txc->write_onode(o); |
14339 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14340 | << " " << name << " (" << val.length() << " bytes)" | |
14341 | << " = " << r << dendl; | |
14342 | return r; | |
14343 | } | |
14344 | ||
14345 | int BlueStore::_setattrs(TransContext *txc, | |
14346 | CollectionRef& c, | |
14347 | OnodeRef& o, | |
14348 | const map<string,bufferptr>& aset) | |
14349 | { | |
14350 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14351 | << " " << aset.size() << " keys" | |
14352 | << dendl; | |
14353 | int r = 0; | |
14354 | for (map<string,bufferptr>::const_iterator p = aset.begin(); | |
14355 | p != aset.end(); ++p) { | |
3efd9988 FG |
14356 | if (p->second.is_partial()) { |
14357 | auto& b = o->onode.attrs[p->first.c_str()] = | |
7c673cae | 14358 | bufferptr(p->second.c_str(), p->second.length()); |
3efd9988 FG |
14359 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); |
14360 | } else { | |
14361 | auto& b = o->onode.attrs[p->first.c_str()] = p->second; | |
14362 | b.reassign_to_mempool(mempool::mempool_bluestore_cache_other); | |
14363 | } | |
7c673cae FG |
14364 | } |
14365 | txc->write_onode(o); | |
14366 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14367 | << " " << aset.size() << " keys" | |
14368 | << " = " << r << dendl; | |
14369 | return r; | |
14370 | } | |
14371 | ||
14372 | ||
14373 | int BlueStore::_rmattr(TransContext *txc, | |
14374 | CollectionRef& c, | |
14375 | OnodeRef& o, | |
14376 | const string& name) | |
14377 | { | |
14378 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14379 | << " " << name << dendl; | |
14380 | int r = 0; | |
14381 | auto it = o->onode.attrs.find(name.c_str()); | |
14382 | if (it == o->onode.attrs.end()) | |
14383 | goto out; | |
14384 | ||
14385 | o->onode.attrs.erase(it); | |
14386 | txc->write_onode(o); | |
14387 | ||
14388 | out: | |
14389 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14390 | << " " << name << " = " << r << dendl; | |
14391 | return r; | |
14392 | } | |
14393 | ||
14394 | int BlueStore::_rmattrs(TransContext *txc, | |
14395 | CollectionRef& c, | |
14396 | OnodeRef& o) | |
14397 | { | |
14398 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14399 | int r = 0; | |
14400 | ||
14401 | if (o->onode.attrs.empty()) | |
14402 | goto out; | |
14403 | ||
14404 | o->onode.attrs.clear(); | |
14405 | txc->write_onode(o); | |
14406 | ||
14407 | out: | |
14408 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14409 | return r; | |
14410 | } | |
14411 | ||
9f95a23c | 14412 | void BlueStore::_do_omap_clear(TransContext *txc, OnodeRef& o) |
7c673cae | 14413 | { |
9f95a23c | 14414 | const string& omap_prefix = o->get_omap_prefix(); |
7c673cae | 14415 | string prefix, tail; |
9f95a23c TL |
14416 | o->get_omap_header(&prefix); |
14417 | o->get_omap_tail(&tail); | |
11fdf7f2 | 14418 | txc->t->rm_range_keys(omap_prefix, prefix, tail); |
494da23a | 14419 | txc->t->rmkey(omap_prefix, tail); |
11fdf7f2 TL |
14420 | dout(20) << __func__ << " remove range start: " |
14421 | << pretty_binary_string(prefix) << " end: " | |
14422 | << pretty_binary_string(tail) << dendl; | |
7c673cae FG |
14423 | } |
14424 | ||
14425 | int BlueStore::_omap_clear(TransContext *txc, | |
14426 | CollectionRef& c, | |
14427 | OnodeRef& o) | |
14428 | { | |
14429 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14430 | int r = 0; | |
14431 | if (o->onode.has_omap()) { | |
14432 | o->flush(); | |
9f95a23c | 14433 | _do_omap_clear(txc, o); |
7c673cae FG |
14434 | o->onode.clear_omap_flag(); |
14435 | txc->write_onode(o); | |
14436 | } | |
14437 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14438 | return r; | |
14439 | } | |
14440 | ||
14441 | int BlueStore::_omap_setkeys(TransContext *txc, | |
14442 | CollectionRef& c, | |
14443 | OnodeRef& o, | |
14444 | bufferlist &bl) | |
14445 | { | |
14446 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14447 | int r; | |
11fdf7f2 | 14448 | auto p = bl.cbegin(); |
7c673cae FG |
14449 | __u32 num; |
14450 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14451 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14452 | o->onode.set_omap_flags_pgmeta(); |
14453 | } else { | |
14454 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14455 | } |
7c673cae | 14456 | txc->write_onode(o); |
494da23a | 14457 | |
9f95a23c | 14458 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14459 | string key_tail; |
14460 | bufferlist tail; | |
9f95a23c | 14461 | o->get_omap_tail(&key_tail); |
494da23a | 14462 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14463 | } else { |
14464 | txc->note_modified_object(o); | |
14465 | } | |
9f95a23c | 14466 | const string& prefix = o->get_omap_prefix(); |
7c673cae | 14467 | string final_key; |
9f95a23c TL |
14468 | o->get_omap_key(string(), &final_key); |
14469 | size_t base_key_len = final_key.size(); | |
11fdf7f2 | 14470 | decode(num, p); |
7c673cae FG |
14471 | while (num--) { |
14472 | string key; | |
14473 | bufferlist value; | |
11fdf7f2 TL |
14474 | decode(key, p); |
14475 | decode(value, p); | |
9f95a23c | 14476 | final_key.resize(base_key_len); // keep prefix |
7c673cae | 14477 | final_key += key; |
11fdf7f2 | 14478 | dout(20) << __func__ << " " << pretty_binary_string(final_key) |
7c673cae | 14479 | << " <- " << key << dendl; |
11fdf7f2 | 14480 | txc->t->set(prefix, final_key, value); |
7c673cae FG |
14481 | } |
14482 | r = 0; | |
14483 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14484 | return r; | |
14485 | } | |
14486 | ||
14487 | int BlueStore::_omap_setheader(TransContext *txc, | |
14488 | CollectionRef& c, | |
14489 | OnodeRef &o, | |
14490 | bufferlist& bl) | |
14491 | { | |
14492 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14493 | int r; | |
14494 | string key; | |
14495 | if (!o->onode.has_omap()) { | |
11fdf7f2 | 14496 | if (o->oid.is_pgmeta()) { |
9f95a23c TL |
14497 | o->onode.set_omap_flags_pgmeta(); |
14498 | } else { | |
14499 | o->onode.set_omap_flags(); | |
11fdf7f2 | 14500 | } |
7c673cae | 14501 | txc->write_onode(o); |
494da23a | 14502 | |
9f95a23c | 14503 | const string& prefix = o->get_omap_prefix(); |
494da23a TL |
14504 | string key_tail; |
14505 | bufferlist tail; | |
9f95a23c | 14506 | o->get_omap_tail(&key_tail); |
494da23a | 14507 | txc->t->set(prefix, key_tail, tail); |
7c673cae FG |
14508 | } else { |
14509 | txc->note_modified_object(o); | |
14510 | } | |
9f95a23c TL |
14511 | const string& prefix = o->get_omap_prefix(); |
14512 | o->get_omap_header(&key); | |
11fdf7f2 | 14513 | txc->t->set(prefix, key, bl); |
7c673cae FG |
14514 | r = 0; |
14515 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14516 | return r; | |
14517 | } | |
14518 | ||
14519 | int BlueStore::_omap_rmkeys(TransContext *txc, | |
14520 | CollectionRef& c, | |
14521 | OnodeRef& o, | |
14522 | bufferlist& bl) | |
14523 | { | |
14524 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
14525 | int r = 0; | |
11fdf7f2 | 14526 | auto p = bl.cbegin(); |
7c673cae FG |
14527 | __u32 num; |
14528 | string final_key; | |
14529 | ||
14530 | if (!o->onode.has_omap()) { | |
14531 | goto out; | |
14532 | } | |
11fdf7f2 | 14533 | { |
9f95a23c TL |
14534 | const string& prefix = o->get_omap_prefix(); |
14535 | o->get_omap_key(string(), &final_key); | |
14536 | size_t base_key_len = final_key.size(); | |
11fdf7f2 TL |
14537 | decode(num, p); |
14538 | while (num--) { | |
14539 | string key; | |
14540 | decode(key, p); | |
9f95a23c | 14541 | final_key.resize(base_key_len); // keep prefix |
11fdf7f2 TL |
14542 | final_key += key; |
14543 | dout(20) << __func__ << " rm " << pretty_binary_string(final_key) | |
14544 | << " <- " << key << dendl; | |
14545 | txc->t->rmkey(prefix, final_key); | |
14546 | } | |
7c673cae FG |
14547 | } |
14548 | txc->note_modified_object(o); | |
14549 | ||
14550 | out: | |
14551 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14552 | return r; | |
14553 | } | |
14554 | ||
14555 | int BlueStore::_omap_rmkey_range(TransContext *txc, | |
14556 | CollectionRef& c, | |
14557 | OnodeRef& o, | |
14558 | const string& first, const string& last) | |
14559 | { | |
14560 | dout(15) << __func__ << " " << c->cid << " " << o->oid << dendl; | |
7c673cae FG |
14561 | string key_first, key_last; |
14562 | int r = 0; | |
14563 | if (!o->onode.has_omap()) { | |
14564 | goto out; | |
14565 | } | |
11fdf7f2 | 14566 | { |
9f95a23c | 14567 | const string& prefix = o->get_omap_prefix(); |
11fdf7f2 | 14568 | o->flush(); |
9f95a23c TL |
14569 | o->get_omap_key(first, &key_first); |
14570 | o->get_omap_key(last, &key_last); | |
11fdf7f2 TL |
14571 | txc->t->rm_range_keys(prefix, key_first, key_last); |
14572 | dout(20) << __func__ << " remove range start: " | |
14573 | << pretty_binary_string(key_first) << " end: " | |
14574 | << pretty_binary_string(key_last) << dendl; | |
7c673cae FG |
14575 | } |
14576 | txc->note_modified_object(o); | |
14577 | ||
14578 | out: | |
14579 | dout(10) << __func__ << " " << c->cid << " " << o->oid << " = " << r << dendl; | |
14580 | return r; | |
14581 | } | |
14582 | ||
14583 | int BlueStore::_set_alloc_hint( | |
14584 | TransContext *txc, | |
14585 | CollectionRef& c, | |
14586 | OnodeRef& o, | |
14587 | uint64_t expected_object_size, | |
14588 | uint64_t expected_write_size, | |
14589 | uint32_t flags) | |
14590 | { | |
14591 | dout(15) << __func__ << " " << c->cid << " " << o->oid | |
14592 | << " object_size " << expected_object_size | |
14593 | << " write_size " << expected_write_size | |
14594 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
14595 | << dendl; | |
14596 | int r = 0; | |
14597 | o->onode.expected_object_size = expected_object_size; | |
14598 | o->onode.expected_write_size = expected_write_size; | |
14599 | o->onode.alloc_hint_flags = flags; | |
14600 | txc->write_onode(o); | |
14601 | dout(10) << __func__ << " " << c->cid << " " << o->oid | |
14602 | << " object_size " << expected_object_size | |
14603 | << " write_size " << expected_write_size | |
14604 | << " flags " << ceph_osd_alloc_hint_flag_string(flags) | |
14605 | << " = " << r << dendl; | |
14606 | return r; | |
14607 | } | |
14608 | ||
14609 | int BlueStore::_clone(TransContext *txc, | |
14610 | CollectionRef& c, | |
14611 | OnodeRef& oldo, | |
14612 | OnodeRef& newo) | |
14613 | { | |
14614 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14615 | << newo->oid << dendl; | |
14616 | int r = 0; | |
14617 | if (oldo->oid.hobj.get_hash() != newo->oid.hobj.get_hash()) { | |
14618 | derr << __func__ << " mismatched hash on " << oldo->oid | |
14619 | << " and " << newo->oid << dendl; | |
14620 | return -EINVAL; | |
14621 | } | |
14622 | ||
7c673cae FG |
14623 | _assign_nid(txc, newo); |
14624 | ||
14625 | // clone data | |
14626 | oldo->flush(); | |
14627 | _do_truncate(txc, c, newo, 0); | |
14628 | if (cct->_conf->bluestore_clone_cow) { | |
14629 | _do_clone_range(txc, c, oldo, newo, 0, oldo->onode.size, 0); | |
14630 | } else { | |
14631 | bufferlist bl; | |
14632 | r = _do_read(c.get(), oldo, 0, oldo->onode.size, bl, 0); | |
14633 | if (r < 0) | |
14634 | goto out; | |
14635 | r = _do_write(txc, c, newo, 0, oldo->onode.size, bl, 0); | |
14636 | if (r < 0) | |
14637 | goto out; | |
14638 | } | |
14639 | ||
14640 | // clone attrs | |
14641 | newo->onode.attrs = oldo->onode.attrs; | |
14642 | ||
14643 | // clone omap | |
14644 | if (newo->onode.has_omap()) { | |
14645 | dout(20) << __func__ << " clearing old omap data" << dendl; | |
14646 | newo->flush(); | |
9f95a23c | 14647 | _do_omap_clear(txc, newo); |
494da23a | 14648 | newo->onode.clear_omap_flag(); |
7c673cae FG |
14649 | } |
14650 | if (oldo->onode.has_omap()) { | |
14651 | dout(20) << __func__ << " copying omap data" << dendl; | |
494da23a | 14652 | if (newo->oid.is_pgmeta()) { |
9f95a23c TL |
14653 | newo->onode.set_omap_flags_pgmeta(); |
14654 | } else { | |
14655 | newo->onode.set_omap_flags(); | |
7c673cae | 14656 | } |
9f95a23c | 14657 | const string& prefix = newo->get_omap_prefix(); |
11fdf7f2 | 14658 | KeyValueDB::Iterator it = db->get_iterator(prefix); |
7c673cae | 14659 | string head, tail; |
9f95a23c TL |
14660 | oldo->get_omap_header(&head); |
14661 | oldo->get_omap_tail(&tail); | |
7c673cae FG |
14662 | it->lower_bound(head); |
14663 | while (it->valid()) { | |
14664 | if (it->key() >= tail) { | |
14665 | dout(30) << __func__ << " reached tail" << dendl; | |
14666 | break; | |
14667 | } else { | |
14668 | dout(30) << __func__ << " got header/data " | |
14669 | << pretty_binary_string(it->key()) << dendl; | |
14670 | string key; | |
9f95a23c | 14671 | newo->rewrite_omap_key(it->key(), &key); |
11fdf7f2 | 14672 | txc->t->set(prefix, key, it->value()); |
7c673cae FG |
14673 | } |
14674 | it->next(); | |
14675 | } | |
494da23a TL |
14676 | string new_tail; |
14677 | bufferlist new_tail_value; | |
9f95a23c | 14678 | newo->get_omap_tail(&new_tail); |
494da23a | 14679 | txc->t->set(prefix, new_tail, new_tail_value); |
7c673cae FG |
14680 | } |
14681 | ||
14682 | txc->write_onode(newo); | |
14683 | r = 0; | |
14684 | ||
14685 | out: | |
14686 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14687 | << newo->oid << " = " << r << dendl; | |
14688 | return r; | |
14689 | } | |
14690 | ||
14691 | int BlueStore::_do_clone_range( | |
14692 | TransContext *txc, | |
14693 | CollectionRef& c, | |
14694 | OnodeRef& oldo, | |
14695 | OnodeRef& newo, | |
224ce89b WB |
14696 | uint64_t srcoff, |
14697 | uint64_t length, | |
14698 | uint64_t dstoff) | |
7c673cae FG |
14699 | { |
14700 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14701 | << newo->oid | |
14702 | << " 0x" << std::hex << srcoff << "~" << length << " -> " | |
14703 | << " 0x" << dstoff << "~" << length << std::dec << dendl; | |
14704 | oldo->extent_map.fault_range(db, srcoff, length); | |
14705 | newo->extent_map.fault_range(db, dstoff, length); | |
81eedcae TL |
14706 | _dump_onode<30>(cct, *oldo); |
14707 | _dump_onode<30>(cct, *newo); | |
7c673cae | 14708 | |
11fdf7f2 | 14709 | oldo->extent_map.dup(this, txc, c, oldo, newo, srcoff, length, dstoff); |
81eedcae TL |
14710 | _dump_onode<30>(cct, *oldo); |
14711 | _dump_onode<30>(cct, *newo); | |
7c673cae FG |
14712 | return 0; |
14713 | } | |
14714 | ||
14715 | int BlueStore::_clone_range(TransContext *txc, | |
14716 | CollectionRef& c, | |
14717 | OnodeRef& oldo, | |
14718 | OnodeRef& newo, | |
14719 | uint64_t srcoff, uint64_t length, uint64_t dstoff) | |
14720 | { | |
14721 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14722 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
14723 | << " to offset 0x" << dstoff << std::dec << dendl; | |
14724 | int r = 0; | |
14725 | ||
35e4c445 FG |
14726 | if (srcoff + length >= OBJECT_MAX_SIZE || |
14727 | dstoff + length >= OBJECT_MAX_SIZE) { | |
14728 | r = -E2BIG; | |
14729 | goto out; | |
14730 | } | |
7c673cae FG |
14731 | if (srcoff + length > oldo->onode.size) { |
14732 | r = -EINVAL; | |
14733 | goto out; | |
14734 | } | |
14735 | ||
7c673cae FG |
14736 | _assign_nid(txc, newo); |
14737 | ||
14738 | if (length > 0) { | |
14739 | if (cct->_conf->bluestore_clone_cow) { | |
14740 | _do_zero(txc, c, newo, dstoff, length); | |
14741 | _do_clone_range(txc, c, oldo, newo, srcoff, length, dstoff); | |
14742 | } else { | |
14743 | bufferlist bl; | |
14744 | r = _do_read(c.get(), oldo, srcoff, length, bl, 0); | |
14745 | if (r < 0) | |
14746 | goto out; | |
14747 | r = _do_write(txc, c, newo, dstoff, bl.length(), bl, 0); | |
14748 | if (r < 0) | |
14749 | goto out; | |
14750 | } | |
14751 | } | |
14752 | ||
14753 | txc->write_onode(newo); | |
14754 | r = 0; | |
14755 | ||
14756 | out: | |
14757 | dout(10) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14758 | << newo->oid << " from 0x" << std::hex << srcoff << "~" << length | |
14759 | << " to offset 0x" << dstoff << std::dec | |
14760 | << " = " << r << dendl; | |
14761 | return r; | |
14762 | } | |
14763 | ||
14764 | int BlueStore::_rename(TransContext *txc, | |
14765 | CollectionRef& c, | |
14766 | OnodeRef& oldo, | |
14767 | OnodeRef& newo, | |
14768 | const ghobject_t& new_oid) | |
14769 | { | |
14770 | dout(15) << __func__ << " " << c->cid << " " << oldo->oid << " -> " | |
14771 | << new_oid << dendl; | |
14772 | int r; | |
14773 | ghobject_t old_oid = oldo->oid; | |
31f18b77 | 14774 | mempool::bluestore_cache_other::string new_okey; |
7c673cae FG |
14775 | |
14776 | if (newo) { | |
14777 | if (newo->exists) { | |
14778 | r = -EEXIST; | |
14779 | goto out; | |
14780 | } | |
11fdf7f2 | 14781 | ceph_assert(txc->onodes.count(newo) == 0); |
7c673cae FG |
14782 | } |
14783 | ||
14784 | txc->t->rmkey(PREFIX_OBJ, oldo->key.c_str(), oldo->key.size()); | |
14785 | ||
14786 | // rewrite shards | |
14787 | { | |
14788 | oldo->extent_map.fault_range(db, 0, oldo->onode.size); | |
14789 | get_object_key(cct, new_oid, &new_okey); | |
14790 | string key; | |
14791 | for (auto &s : oldo->extent_map.shards) { | |
14792 | generate_extent_shard_key_and_apply(oldo->key, s.shard_info->offset, &key, | |
14793 | [&](const string& final_key) { | |
14794 | txc->t->rmkey(PREFIX_OBJ, final_key); | |
14795 | } | |
14796 | ); | |
14797 | s.dirty = true; | |
14798 | } | |
14799 | } | |
14800 | ||
14801 | newo = oldo; | |
14802 | txc->write_onode(newo); | |
14803 | ||
14804 | // this adjusts oldo->{oid,key}, and reset oldo to a fresh empty | |
14805 | // Onode in the old slot | |
14806 | c->onode_map.rename(oldo, old_oid, new_oid, new_okey); | |
14807 | r = 0; | |
14808 | ||
f64942e4 AA |
14809 | // hold a ref to new Onode in old name position, to ensure we don't drop |
14810 | // it from the cache before this txc commits (or else someone may come along | |
14811 | // and read newo's metadata via the old name). | |
14812 | txc->note_modified_object(oldo); | |
14813 | ||
7c673cae FG |
14814 | out: |
14815 | dout(10) << __func__ << " " << c->cid << " " << old_oid << " -> " | |
14816 | << new_oid << " = " << r << dendl; | |
14817 | return r; | |
14818 | } | |
14819 | ||
14820 | // collections | |
14821 | ||
14822 | int BlueStore::_create_collection( | |
14823 | TransContext *txc, | |
14824 | const coll_t &cid, | |
14825 | unsigned bits, | |
14826 | CollectionRef *c) | |
14827 | { | |
14828 | dout(15) << __func__ << " " << cid << " bits " << bits << dendl; | |
14829 | int r; | |
14830 | bufferlist bl; | |
14831 | ||
14832 | { | |
9f95a23c | 14833 | std::unique_lock l(coll_lock); |
7c673cae FG |
14834 | if (*c) { |
14835 | r = -EEXIST; | |
14836 | goto out; | |
14837 | } | |
11fdf7f2 TL |
14838 | auto p = new_coll_map.find(cid); |
14839 | ceph_assert(p != new_coll_map.end()); | |
14840 | *c = p->second; | |
7c673cae FG |
14841 | (*c)->cnode.bits = bits; |
14842 | coll_map[cid] = *c; | |
11fdf7f2 | 14843 | new_coll_map.erase(p); |
7c673cae | 14844 | } |
11fdf7f2 | 14845 | encode((*c)->cnode, bl); |
7c673cae FG |
14846 | txc->t->set(PREFIX_COLL, stringify(cid), bl); |
14847 | r = 0; | |
14848 | ||
14849 | out: | |
14850 | dout(10) << __func__ << " " << cid << " bits " << bits << " = " << r << dendl; | |
14851 | return r; | |
14852 | } | |
14853 | ||
14854 | int BlueStore::_remove_collection(TransContext *txc, const coll_t &cid, | |
14855 | CollectionRef *c) | |
14856 | { | |
14857 | dout(15) << __func__ << " " << cid << dendl; | |
14858 | int r; | |
14859 | ||
11fdf7f2 | 14860 | (*c)->flush_all_but_last(); |
7c673cae | 14861 | { |
9f95a23c | 14862 | std::unique_lock l(coll_lock); |
7c673cae FG |
14863 | if (!*c) { |
14864 | r = -ENOENT; | |
14865 | goto out; | |
14866 | } | |
14867 | size_t nonexistent_count = 0; | |
11fdf7f2 | 14868 | ceph_assert((*c)->exists); |
7c673cae FG |
14869 | if ((*c)->onode_map.map_any([&](OnodeRef o) { |
14870 | if (o->exists) { | |
494da23a TL |
14871 | dout(1) << __func__ << " " << o->oid << " " << o |
14872 | << " exists in onode_map" << dendl; | |
7c673cae FG |
14873 | return true; |
14874 | } | |
14875 | ++nonexistent_count; | |
14876 | return false; | |
14877 | })) { | |
14878 | r = -ENOTEMPTY; | |
14879 | goto out; | |
14880 | } | |
14881 | ||
14882 | vector<ghobject_t> ls; | |
14883 | ghobject_t next; | |
14884 | // Enumerate onodes in db, up to nonexistent_count + 1 | |
14885 | // then check if all of them are marked as non-existent. | |
11fdf7f2 | 14886 | // Bypass the check if (next != ghobject_t::get_max()) |
7c673cae FG |
14887 | r = _collection_list(c->get(), ghobject_t(), ghobject_t::get_max(), |
14888 | nonexistent_count + 1, &ls, &next); | |
14889 | if (r >= 0) { | |
11fdf7f2 TL |
14890 | // If true mean collecton has more objects than nonexistent_count, |
14891 | // so bypass check. | |
14892 | bool exists = (!next.is_max()); | |
7c673cae FG |
14893 | for (auto it = ls.begin(); !exists && it < ls.end(); ++it) { |
14894 | dout(10) << __func__ << " oid " << *it << dendl; | |
14895 | auto onode = (*c)->onode_map.lookup(*it); | |
14896 | exists = !onode || onode->exists; | |
14897 | if (exists) { | |
494da23a TL |
14898 | dout(1) << __func__ << " " << *it |
14899 | << " exists in db, " | |
14900 | << (!onode ? "not present in ram" : "present in ram") | |
14901 | << dendl; | |
7c673cae FG |
14902 | } |
14903 | } | |
14904 | if (!exists) { | |
11fdf7f2 | 14905 | _do_remove_collection(txc, c); |
7c673cae FG |
14906 | r = 0; |
14907 | } else { | |
14908 | dout(10) << __func__ << " " << cid | |
14909 | << " is non-empty" << dendl; | |
14910 | r = -ENOTEMPTY; | |
14911 | } | |
14912 | } | |
14913 | } | |
14914 | ||
14915 | out: | |
14916 | dout(10) << __func__ << " " << cid << " = " << r << dendl; | |
14917 | return r; | |
14918 | } | |
14919 | ||
11fdf7f2 TL |
14920 | void BlueStore::_do_remove_collection(TransContext *txc, |
14921 | CollectionRef *c) | |
14922 | { | |
14923 | coll_map.erase((*c)->cid); | |
14924 | txc->removed_collections.push_back(*c); | |
14925 | (*c)->exists = false; | |
14926 | _osr_register_zombie((*c)->osr.get()); | |
14927 | txc->t->rmkey(PREFIX_COLL, stringify((*c)->cid)); | |
14928 | c->reset(); | |
14929 | } | |
14930 | ||
7c673cae FG |
14931 | int BlueStore::_split_collection(TransContext *txc, |
14932 | CollectionRef& c, | |
14933 | CollectionRef& d, | |
14934 | unsigned bits, int rem) | |
14935 | { | |
14936 | dout(15) << __func__ << " " << c->cid << " to " << d->cid << " " | |
14937 | << " bits " << bits << dendl; | |
9f95a23c TL |
14938 | std::unique_lock l(c->lock); |
14939 | std::unique_lock l2(d->lock); | |
7c673cae FG |
14940 | int r; |
14941 | ||
14942 | // flush all previous deferred writes on this sequencer. this is a bit | |
14943 | // heavyweight, but we need to make sure all deferred writes complete | |
14944 | // before we split as the new collection's sequencer may need to order | |
14945 | // this after those writes, and we don't bother with the complexity of | |
14946 | // moving those TransContexts over to the new osr. | |
14947 | _osr_drain_preceding(txc); | |
14948 | ||
14949 | // move any cached items (onodes and referenced shared blobs) that will | |
14950 | // belong to the child collection post-split. leave everything else behind. | |
14951 | // this may include things that don't strictly belong to the now-smaller | |
14952 | // parent split, but the OSD will always send us a split for every new | |
14953 | // child. | |
14954 | ||
14955 | spg_t pgid, dest_pgid; | |
14956 | bool is_pg = c->cid.is_pg(&pgid); | |
11fdf7f2 | 14957 | ceph_assert(is_pg); |
7c673cae | 14958 | is_pg = d->cid.is_pg(&dest_pgid); |
11fdf7f2 | 14959 | ceph_assert(is_pg); |
7c673cae FG |
14960 | |
14961 | // the destination should initially be empty. | |
11fdf7f2 TL |
14962 | ceph_assert(d->onode_map.empty()); |
14963 | ceph_assert(d->shared_blob_set.empty()); | |
14964 | ceph_assert(d->cnode.bits == bits); | |
7c673cae FG |
14965 | |
14966 | c->split_cache(d.get()); | |
14967 | ||
14968 | // adjust bits. note that this will be redundant for all but the first | |
14969 | // split call for this parent (first child). | |
14970 | c->cnode.bits = bits; | |
11fdf7f2 | 14971 | ceph_assert(d->cnode.bits == bits); |
7c673cae FG |
14972 | r = 0; |
14973 | ||
14974 | bufferlist bl; | |
11fdf7f2 | 14975 | encode(c->cnode, bl); |
7c673cae FG |
14976 | txc->t->set(PREFIX_COLL, stringify(c->cid), bl); |
14977 | ||
14978 | dout(10) << __func__ << " " << c->cid << " to " << d->cid << " " | |
14979 | << " bits " << bits << " = " << r << dendl; | |
14980 | return r; | |
14981 | } | |
14982 | ||
11fdf7f2 TL |
14983 | int BlueStore::_merge_collection( |
14984 | TransContext *txc, | |
14985 | CollectionRef *c, | |
14986 | CollectionRef& d, | |
14987 | unsigned bits) | |
14988 | { | |
14989 | dout(15) << __func__ << " " << (*c)->cid << " to " << d->cid | |
14990 | << " bits " << bits << dendl; | |
9f95a23c TL |
14991 | std::unique_lock l((*c)->lock); |
14992 | std::unique_lock l2(d->lock); | |
11fdf7f2 TL |
14993 | int r; |
14994 | ||
14995 | coll_t cid = (*c)->cid; | |
14996 | ||
14997 | // flush all previous deferred writes on the source collection to ensure | |
14998 | // that all deferred writes complete before we merge as the target collection's | |
14999 | // sequencer may need to order new ops after those writes. | |
15000 | ||
15001 | _osr_drain((*c)->osr.get()); | |
15002 | ||
15003 | // move any cached items (onodes and referenced shared blobs) that will | |
15004 | // belong to the child collection post-split. leave everything else behind. | |
15005 | // this may include things that don't strictly belong to the now-smaller | |
15006 | // parent split, but the OSD will always send us a split for every new | |
15007 | // child. | |
15008 | ||
15009 | spg_t pgid, dest_pgid; | |
15010 | bool is_pg = cid.is_pg(&pgid); | |
15011 | ceph_assert(is_pg); | |
15012 | is_pg = d->cid.is_pg(&dest_pgid); | |
15013 | ceph_assert(is_pg); | |
15014 | ||
15015 | // adjust bits. note that this will be redundant for all but the first | |
15016 | // merge call for the parent/target. | |
15017 | d->cnode.bits = bits; | |
15018 | ||
15019 | // behavior depends on target (d) bits, so this after that is updated. | |
15020 | (*c)->split_cache(d.get()); | |
15021 | ||
15022 | // remove source collection | |
15023 | { | |
9f95a23c | 15024 | std::unique_lock l3(coll_lock); |
11fdf7f2 TL |
15025 | _do_remove_collection(txc, c); |
15026 | } | |
15027 | ||
15028 | r = 0; | |
15029 | ||
15030 | bufferlist bl; | |
15031 | encode(d->cnode, bl); | |
15032 | txc->t->set(PREFIX_COLL, stringify(d->cid), bl); | |
15033 | ||
15034 | dout(10) << __func__ << " " << cid << " to " << d->cid << " " | |
15035 | << " bits " << bits << " = " << r << dendl; | |
15036 | return r; | |
15037 | } | |
15038 | ||
494da23a TL |
15039 | void BlueStore::log_latency( |
15040 | const char* name, | |
15041 | int idx, | |
15042 | const ceph::timespan& l, | |
15043 | double lat_threshold, | |
15044 | const char* info) const | |
15045 | { | |
15046 | logger->tinc(idx, l); | |
15047 | if (lat_threshold > 0.0 && | |
15048 | l >= make_timespan(lat_threshold)) { | |
15049 | dout(0) << __func__ << " slow operation observed for " << name | |
15050 | << ", latency = " << l | |
15051 | << info | |
15052 | << dendl; | |
15053 | } | |
15054 | } | |
15055 | ||
11fdf7f2 | 15056 | void BlueStore::log_latency_fn( |
494da23a | 15057 | const char* name, |
11fdf7f2 TL |
15058 | int idx, |
15059 | const ceph::timespan& l, | |
494da23a TL |
15060 | double lat_threshold, |
15061 | std::function<string (const ceph::timespan& lat)> fn) const | |
11fdf7f2 | 15062 | { |
494da23a TL |
15063 | logger->tinc(idx, l); |
15064 | if (lat_threshold > 0.0 && | |
15065 | l >= make_timespan(lat_threshold)) { | |
15066 | dout(0) << __func__ << " slow operation observed for " << name | |
15067 | << ", latency = " << l | |
15068 | << fn(l) | |
15069 | << dendl; | |
15070 | } | |
11fdf7f2 TL |
15071 | } |
15072 | ||
9f95a23c TL |
15073 | #if defined(WITH_LTTNG) |
15074 | void BlueStore::BlueStoreThrottle::emit_initial_tracepoint( | |
15075 | KeyValueDB &db, | |
15076 | TransContext &txc, | |
15077 | mono_clock::time_point start_throttle_acquire) | |
15078 | { | |
15079 | pending_kv_ios += txc.ios; | |
15080 | if (txc.deferred_txn) { | |
15081 | pending_deferred_ios += txc.ios; | |
15082 | } | |
15083 | ||
15084 | uint64_t started = 0; | |
15085 | uint64_t completed = 0; | |
15086 | if (should_trace(&started, &completed)) { | |
15087 | txc.tracing = true; | |
15088 | uint64_t rocksdb_base_level, | |
15089 | rocksdb_estimate_pending_compaction_bytes, | |
15090 | rocksdb_cur_size_all_mem_tables, | |
15091 | rocksdb_compaction_pending, | |
15092 | rocksdb_mem_table_flush_pending, | |
15093 | rocksdb_num_running_compactions, | |
15094 | rocksdb_num_running_flushes, | |
15095 | rocksdb_actual_delayed_write_rate; | |
15096 | db.get_property( | |
15097 | "rocksdb.base-level", | |
15098 | &rocksdb_base_level); | |
15099 | db.get_property( | |
15100 | "rocksdb.estimate-pending-compaction-bytes", | |
15101 | &rocksdb_estimate_pending_compaction_bytes); | |
15102 | db.get_property( | |
15103 | "rocksdb.cur-size-all-mem-tables", | |
15104 | &rocksdb_cur_size_all_mem_tables); | |
15105 | db.get_property( | |
15106 | "rocksdb.compaction-pending", | |
15107 | &rocksdb_compaction_pending); | |
15108 | db.get_property( | |
15109 | "rocksdb.mem-table-flush-pending", | |
15110 | &rocksdb_mem_table_flush_pending); | |
15111 | db.get_property( | |
15112 | "rocksdb.num-running-compactions", | |
15113 | &rocksdb_num_running_compactions); | |
15114 | db.get_property( | |
15115 | "rocksdb.num-running-flushes", | |
15116 | &rocksdb_num_running_flushes); | |
15117 | db.get_property( | |
15118 | "rocksdb.actual-delayed-write-rate", | |
15119 | &rocksdb_actual_delayed_write_rate); | |
15120 | ||
15121 | ||
15122 | tracepoint( | |
15123 | bluestore, | |
15124 | transaction_initial_state, | |
15125 | txc.osr->get_sequencer_id(), | |
15126 | txc.seq, | |
15127 | throttle_bytes.get_current(), | |
15128 | throttle_deferred_bytes.get_current(), | |
15129 | pending_kv_ios, | |
15130 | pending_deferred_ios, | |
15131 | started, | |
15132 | completed, | |
15133 | ceph::to_seconds<double>(mono_clock::now() - start_throttle_acquire)); | |
15134 | ||
15135 | tracepoint( | |
15136 | bluestore, | |
15137 | transaction_initial_state_rocksdb, | |
15138 | txc.osr->get_sequencer_id(), | |
15139 | txc.seq, | |
15140 | rocksdb_base_level, | |
15141 | rocksdb_estimate_pending_compaction_bytes, | |
15142 | rocksdb_cur_size_all_mem_tables, | |
15143 | rocksdb_compaction_pending, | |
15144 | rocksdb_mem_table_flush_pending, | |
15145 | rocksdb_num_running_compactions, | |
15146 | rocksdb_num_running_flushes, | |
15147 | rocksdb_actual_delayed_write_rate); | |
15148 | } | |
15149 | } | |
15150 | #endif | |
15151 | ||
15152 | mono_clock::duration BlueStore::BlueStoreThrottle::log_state_latency( | |
15153 | TransContext &txc, PerfCounters *logger, int state) | |
15154 | { | |
15155 | mono_clock::time_point now = mono_clock::now(); | |
15156 | mono_clock::duration lat = now - txc.last_stamp; | |
15157 | logger->tinc(state, lat); | |
15158 | #if defined(WITH_LTTNG) | |
15159 | if (txc.tracing && | |
15160 | state >= l_bluestore_state_prepare_lat && | |
15161 | state <= l_bluestore_state_done_lat) { | |
15162 | OID_ELAPSED("", lat.to_nsec() / 1000.0, txc.get_state_latency_name(state)); | |
15163 | tracepoint( | |
15164 | bluestore, | |
15165 | transaction_state_duration, | |
15166 | txc.osr->get_sequencer_id(), | |
15167 | txc.seq, | |
15168 | state, | |
15169 | ceph::to_seconds<double>(lat)); | |
15170 | } | |
15171 | #endif | |
15172 | txc.last_stamp = now; | |
15173 | return lat; | |
15174 | } | |
15175 | ||
15176 | bool BlueStore::BlueStoreThrottle::try_start_transaction( | |
15177 | KeyValueDB &db, | |
15178 | TransContext &txc, | |
15179 | mono_clock::time_point start_throttle_acquire) | |
15180 | { | |
15181 | throttle_bytes.get(txc.cost); | |
15182 | ||
15183 | if (!txc.deferred_txn || throttle_deferred_bytes.get_or_fail(txc.cost)) { | |
15184 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15185 | return true; | |
15186 | } else { | |
15187 | return false; | |
15188 | } | |
15189 | } | |
15190 | ||
15191 | void BlueStore::BlueStoreThrottle::finish_start_transaction( | |
15192 | KeyValueDB &db, | |
15193 | TransContext &txc, | |
15194 | mono_clock::time_point start_throttle_acquire) | |
15195 | { | |
15196 | ceph_assert(txc.deferred_txn); | |
15197 | throttle_deferred_bytes.get(txc.cost); | |
15198 | emit_initial_tracepoint(db, txc, start_throttle_acquire); | |
15199 | } | |
15200 | ||
15201 | #if defined(WITH_LTTNG) | |
15202 | void BlueStore::BlueStoreThrottle::complete_kv(TransContext &txc) | |
15203 | { | |
15204 | pending_kv_ios -= 1; | |
15205 | ios_completed_since_last_traced++; | |
15206 | if (txc.tracing) { | |
15207 | tracepoint( | |
15208 | bluestore, | |
15209 | transaction_commit_latency, | |
15210 | txc.osr->get_sequencer_id(), | |
15211 | txc.seq, | |
15212 | ceph::to_seconds<double>(mono_clock::now() - txc.start)); | |
15213 | } | |
15214 | } | |
15215 | #endif | |
15216 | ||
15217 | #if defined(WITH_LTTNG) | |
15218 | void BlueStore::BlueStoreThrottle::complete(TransContext &txc) | |
15219 | { | |
15220 | if (txc.deferred_txn) { | |
15221 | pending_deferred_ios -= 1; | |
15222 | } | |
15223 | if (txc.tracing) { | |
15224 | mono_clock::time_point now = mono_clock::now(); | |
15225 | mono_clock::duration lat = now - txc.start; | |
15226 | tracepoint( | |
15227 | bluestore, | |
15228 | transaction_total_duration, | |
15229 | txc.osr->get_sequencer_id(), | |
15230 | txc.seq, | |
15231 | ceph::to_seconds<double>(lat)); | |
15232 | } | |
15233 | } | |
15234 | #endif | |
11fdf7f2 | 15235 | |
7c673cae FG |
15236 | // DB key value Histogram |
15237 | #define KEY_SLAB 32 | |
15238 | #define VALUE_SLAB 64 | |
15239 | ||
15240 | const string prefix_onode = "o"; | |
15241 | const string prefix_onode_shard = "x"; | |
15242 | const string prefix_other = "Z"; | |
15243 | ||
15244 | int BlueStore::DBHistogram::get_key_slab(size_t sz) | |
15245 | { | |
15246 | return (sz/KEY_SLAB); | |
15247 | } | |
15248 | ||
15249 | string BlueStore::DBHistogram::get_key_slab_to_range(int slab) | |
15250 | { | |
15251 | int lower_bound = slab * KEY_SLAB; | |
15252 | int upper_bound = (slab + 1) * KEY_SLAB; | |
15253 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15254 | return ret; | |
15255 | } | |
15256 | ||
15257 | int BlueStore::DBHistogram::get_value_slab(size_t sz) | |
15258 | { | |
15259 | return (sz/VALUE_SLAB); | |
15260 | } | |
15261 | ||
15262 | string BlueStore::DBHistogram::get_value_slab_to_range(int slab) | |
15263 | { | |
15264 | int lower_bound = slab * VALUE_SLAB; | |
15265 | int upper_bound = (slab + 1) * VALUE_SLAB; | |
15266 | string ret = "[" + stringify(lower_bound) + "," + stringify(upper_bound) + ")"; | |
15267 | return ret; | |
15268 | } | |
15269 | ||
15270 | void BlueStore::DBHistogram::update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
15271 | const string &prefix, size_t key_size, size_t value_size) | |
15272 | { | |
15273 | uint32_t key_slab = get_key_slab(key_size); | |
15274 | uint32_t value_slab = get_value_slab(value_size); | |
15275 | key_hist[prefix][key_slab].count++; | |
11fdf7f2 TL |
15276 | key_hist[prefix][key_slab].max_len = |
15277 | std::max<size_t>(key_size, key_hist[prefix][key_slab].max_len); | |
7c673cae FG |
15278 | key_hist[prefix][key_slab].val_map[value_slab].count++; |
15279 | key_hist[prefix][key_slab].val_map[value_slab].max_len = | |
11fdf7f2 TL |
15280 | std::max<size_t>(value_size, |
15281 | key_hist[prefix][key_slab].val_map[value_slab].max_len); | |
7c673cae FG |
15282 | } |
15283 | ||
15284 | void BlueStore::DBHistogram::dump(Formatter *f) | |
15285 | { | |
15286 | f->open_object_section("rocksdb_value_distribution"); | |
15287 | for (auto i : value_hist) { | |
15288 | f->dump_unsigned(get_value_slab_to_range(i.first).data(), i.second); | |
15289 | } | |
15290 | f->close_section(); | |
15291 | ||
15292 | f->open_object_section("rocksdb_key_value_histogram"); | |
15293 | for (auto i : key_hist) { | |
15294 | f->dump_string("prefix", i.first); | |
15295 | f->open_object_section("key_hist"); | |
15296 | for ( auto k : i.second) { | |
15297 | f->dump_unsigned(get_key_slab_to_range(k.first).data(), k.second.count); | |
15298 | f->dump_unsigned("max_len", k.second.max_len); | |
15299 | f->open_object_section("value_hist"); | |
15300 | for ( auto j : k.second.val_map) { | |
15301 | f->dump_unsigned(get_value_slab_to_range(j.first).data(), j.second.count); | |
15302 | f->dump_unsigned("max_len", j.second.max_len); | |
15303 | } | |
15304 | f->close_section(); | |
15305 | } | |
15306 | f->close_section(); | |
15307 | } | |
15308 | f->close_section(); | |
15309 | } | |
15310 | ||
15311 | //Itrerates through the db and collects the stats | |
15312 | void BlueStore::generate_db_histogram(Formatter *f) | |
15313 | { | |
15314 | //globals | |
15315 | uint64_t num_onodes = 0; | |
15316 | uint64_t num_shards = 0; | |
15317 | uint64_t num_super = 0; | |
15318 | uint64_t num_coll = 0; | |
15319 | uint64_t num_omap = 0; | |
11fdf7f2 | 15320 | uint64_t num_pgmeta_omap = 0; |
7c673cae FG |
15321 | uint64_t num_deferred = 0; |
15322 | uint64_t num_alloc = 0; | |
15323 | uint64_t num_stat = 0; | |
15324 | uint64_t num_others = 0; | |
15325 | uint64_t num_shared_shards = 0; | |
15326 | size_t max_key_size =0, max_value_size = 0; | |
15327 | uint64_t total_key_size = 0, total_value_size = 0; | |
15328 | size_t key_size = 0, value_size = 0; | |
15329 | DBHistogram hist; | |
15330 | ||
11fdf7f2 | 15331 | auto start = coarse_mono_clock::now(); |
7c673cae | 15332 | |
11fdf7f2 | 15333 | KeyValueDB::WholeSpaceIterator iter = db->get_wholespace_iterator(); |
7c673cae FG |
15334 | iter->seek_to_first(); |
15335 | while (iter->valid()) { | |
15336 | dout(30) << __func__ << " Key: " << iter->key() << dendl; | |
15337 | key_size = iter->key_size(); | |
15338 | value_size = iter->value_size(); | |
15339 | hist.value_hist[hist.get_value_slab(value_size)]++; | |
11fdf7f2 TL |
15340 | max_key_size = std::max(max_key_size, key_size); |
15341 | max_value_size = std::max(max_value_size, value_size); | |
7c673cae FG |
15342 | total_key_size += key_size; |
15343 | total_value_size += value_size; | |
15344 | ||
15345 | pair<string,string> key(iter->raw_key()); | |
15346 | ||
15347 | if (key.first == PREFIX_SUPER) { | |
15348 | hist.update_hist_entry(hist.key_hist, PREFIX_SUPER, key_size, value_size); | |
15349 | num_super++; | |
15350 | } else if (key.first == PREFIX_STAT) { | |
15351 | hist.update_hist_entry(hist.key_hist, PREFIX_STAT, key_size, value_size); | |
15352 | num_stat++; | |
15353 | } else if (key.first == PREFIX_COLL) { | |
15354 | hist.update_hist_entry(hist.key_hist, PREFIX_COLL, key_size, value_size); | |
15355 | num_coll++; | |
15356 | } else if (key.first == PREFIX_OBJ) { | |
15357 | if (key.second.back() == ONODE_KEY_SUFFIX) { | |
15358 | hist.update_hist_entry(hist.key_hist, prefix_onode, key_size, value_size); | |
15359 | num_onodes++; | |
15360 | } else { | |
15361 | hist.update_hist_entry(hist.key_hist, prefix_onode_shard, key_size, value_size); | |
15362 | num_shards++; | |
15363 | } | |
15364 | } else if (key.first == PREFIX_OMAP) { | |
15365 | hist.update_hist_entry(hist.key_hist, PREFIX_OMAP, key_size, value_size); | |
15366 | num_omap++; | |
11fdf7f2 TL |
15367 | } else if (key.first == PREFIX_PGMETA_OMAP) { |
15368 | hist.update_hist_entry(hist.key_hist, PREFIX_PGMETA_OMAP, key_size, value_size); | |
15369 | num_pgmeta_omap++; | |
7c673cae FG |
15370 | } else if (key.first == PREFIX_DEFERRED) { |
15371 | hist.update_hist_entry(hist.key_hist, PREFIX_DEFERRED, key_size, value_size); | |
15372 | num_deferred++; | |
11fdf7f2 | 15373 | } else if (key.first == PREFIX_ALLOC || key.first == PREFIX_ALLOC_BITMAP) { |
7c673cae FG |
15374 | hist.update_hist_entry(hist.key_hist, PREFIX_ALLOC, key_size, value_size); |
15375 | num_alloc++; | |
15376 | } else if (key.first == PREFIX_SHARED_BLOB) { | |
15377 | hist.update_hist_entry(hist.key_hist, PREFIX_SHARED_BLOB, key_size, value_size); | |
15378 | num_shared_shards++; | |
15379 | } else { | |
15380 | hist.update_hist_entry(hist.key_hist, prefix_other, key_size, value_size); | |
15381 | num_others++; | |
15382 | } | |
15383 | iter->next(); | |
15384 | } | |
15385 | ||
11fdf7f2 | 15386 | ceph::timespan duration = coarse_mono_clock::now() - start; |
7c673cae FG |
15387 | f->open_object_section("rocksdb_key_value_stats"); |
15388 | f->dump_unsigned("num_onodes", num_onodes); | |
15389 | f->dump_unsigned("num_shards", num_shards); | |
15390 | f->dump_unsigned("num_super", num_super); | |
15391 | f->dump_unsigned("num_coll", num_coll); | |
15392 | f->dump_unsigned("num_omap", num_omap); | |
11fdf7f2 | 15393 | f->dump_unsigned("num_pgmeta_omap", num_pgmeta_omap); |
7c673cae FG |
15394 | f->dump_unsigned("num_deferred", num_deferred); |
15395 | f->dump_unsigned("num_alloc", num_alloc); | |
15396 | f->dump_unsigned("num_stat", num_stat); | |
15397 | f->dump_unsigned("num_shared_shards", num_shared_shards); | |
15398 | f->dump_unsigned("num_others", num_others); | |
15399 | f->dump_unsigned("max_key_size", max_key_size); | |
15400 | f->dump_unsigned("max_value_size", max_value_size); | |
15401 | f->dump_unsigned("total_key_size", total_key_size); | |
15402 | f->dump_unsigned("total_value_size", total_value_size); | |
15403 | f->close_section(); | |
15404 | ||
15405 | hist.dump(f); | |
15406 | ||
15407 | dout(20) << __func__ << " finished in " << duration << " seconds" << dendl; | |
15408 | ||
15409 | } | |
15410 | ||
f6b5b4d7 | 15411 | void BlueStore::_shutdown_cache() |
7c673cae FG |
15412 | { |
15413 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15414 | for (auto i : buffer_cache_shards) { |
15415 | i->flush(); | |
11fdf7f2 | 15416 | ceph_assert(i->empty()); |
7c673cae FG |
15417 | } |
15418 | for (auto& p : coll_map) { | |
f6b5b4d7 | 15419 | p.second->onode_map.clear(); |
3efd9988 FG |
15420 | if (!p.second->shared_blob_set.empty()) { |
15421 | derr << __func__ << " stray shared blobs on " << p.first << dendl; | |
11fdf7f2 | 15422 | p.second->shared_blob_set.dump<0>(cct); |
3efd9988 | 15423 | } |
11fdf7f2 TL |
15424 | ceph_assert(p.second->onode_map.empty()); |
15425 | ceph_assert(p.second->shared_blob_set.empty()); | |
7c673cae FG |
15426 | } |
15427 | coll_map.clear(); | |
f6b5b4d7 TL |
15428 | for (auto i : onode_cache_shards) { |
15429 | ceph_assert(i->empty()); | |
15430 | } | |
7c673cae FG |
15431 | } |
15432 | ||
31f18b77 FG |
15433 | // For external caller. |
15434 | // We use a best-effort policy instead, e.g., | |
15435 | // we don't care if there are still some pinned onodes/data in the cache | |
15436 | // after this command is completed. | |
11fdf7f2 | 15437 | int BlueStore::flush_cache(ostream *os) |
31f18b77 FG |
15438 | { |
15439 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
15440 | for (auto i : onode_cache_shards) { |
15441 | i->flush(); | |
15442 | } | |
15443 | for (auto i : buffer_cache_shards) { | |
15444 | i->flush(); | |
31f18b77 | 15445 | } |
11fdf7f2 TL |
15446 | |
15447 | return 0; | |
31f18b77 FG |
15448 | } |
15449 | ||
7c673cae FG |
15450 | void BlueStore::_apply_padding(uint64_t head_pad, |
15451 | uint64_t tail_pad, | |
7c673cae FG |
15452 | bufferlist& padded) |
15453 | { | |
7c673cae | 15454 | if (head_pad) { |
224ce89b | 15455 | padded.prepend_zero(head_pad); |
7c673cae FG |
15456 | } |
15457 | if (tail_pad) { | |
15458 | padded.append_zero(tail_pad); | |
15459 | } | |
15460 | if (head_pad || tail_pad) { | |
15461 | dout(20) << __func__ << " can pad head 0x" << std::hex << head_pad | |
15462 | << " tail 0x" << tail_pad << std::dec << dendl; | |
15463 | logger->inc(l_bluestore_write_pad_bytes, head_pad + tail_pad); | |
15464 | } | |
15465 | } | |
15466 | ||
11fdf7f2 TL |
15467 | void BlueStore::_record_onode(OnodeRef &o, KeyValueDB::Transaction &txn) |
15468 | { | |
15469 | // finalize extent_map shards | |
15470 | o->extent_map.update(txn, false); | |
15471 | if (o->extent_map.needs_reshard()) { | |
15472 | o->extent_map.reshard(db, txn); | |
15473 | o->extent_map.update(txn, true); | |
15474 | if (o->extent_map.needs_reshard()) { | |
15475 | dout(20) << __func__ << " warning: still wants reshard, check options?" | |
15476 | << dendl; | |
15477 | o->extent_map.clear_needs_reshard(); | |
15478 | } | |
15479 | logger->inc(l_bluestore_onode_reshard); | |
15480 | } | |
15481 | ||
15482 | // bound encode | |
15483 | size_t bound = 0; | |
15484 | denc(o->onode, bound); | |
15485 | o->extent_map.bound_encode_spanning_blobs(bound); | |
15486 | if (o->onode.extent_map_shards.empty()) { | |
15487 | denc(o->extent_map.inline_bl, bound); | |
15488 | } | |
15489 | ||
15490 | // encode | |
15491 | bufferlist bl; | |
15492 | unsigned onode_part, blob_part, extent_part; | |
15493 | { | |
15494 | auto p = bl.get_contiguous_appender(bound, true); | |
15495 | denc(o->onode, p); | |
15496 | onode_part = p.get_logical_offset(); | |
15497 | o->extent_map.encode_spanning_blobs(p); | |
15498 | blob_part = p.get_logical_offset() - onode_part; | |
15499 | if (o->onode.extent_map_shards.empty()) { | |
15500 | denc(o->extent_map.inline_bl, p); | |
15501 | } | |
15502 | extent_part = p.get_logical_offset() - onode_part - blob_part; | |
15503 | } | |
15504 | ||
15505 | dout(20) << __func__ << " onode " << o->oid << " is " << bl.length() | |
15506 | << " (" << onode_part << " bytes onode + " | |
15507 | << blob_part << " bytes spanning blobs + " | |
15508 | << extent_part << " bytes inline extents)" | |
15509 | << dendl; | |
15510 | ||
15511 | ||
15512 | txn->set(PREFIX_OBJ, o->key.c_str(), o->key.size(), bl); | |
15513 | } | |
15514 | ||
15515 | void BlueStore::_log_alerts(osd_alert_list_t& alerts) | |
15516 | { | |
15517 | std::lock_guard l(qlock); | |
15518 | ||
81eedcae TL |
15519 | if (!disk_size_mismatch_alert.empty()) { |
15520 | alerts.emplace( | |
15521 | "BLUESTORE_DISK_SIZE_MISMATCH", | |
15522 | disk_size_mismatch_alert); | |
15523 | } | |
15524 | if (!legacy_statfs_alert.empty()) { | |
15525 | alerts.emplace( | |
15526 | "BLUESTORE_LEGACY_STATFS", | |
15527 | legacy_statfs_alert); | |
15528 | } | |
11fdf7f2 TL |
15529 | if (!spillover_alert.empty() && |
15530 | cct->_conf->bluestore_warn_on_bluefs_spillover) { | |
15531 | alerts.emplace( | |
15532 | "BLUEFS_SPILLOVER", | |
15533 | spillover_alert); | |
15534 | } | |
9f95a23c TL |
15535 | if (!no_per_pool_omap_alert.empty()) { |
15536 | alerts.emplace( | |
15537 | "BLUESTORE_NO_PER_POOL_OMAP", | |
15538 | no_per_pool_omap_alert); | |
15539 | } | |
11fdf7f2 TL |
15540 | string s0(failed_cmode); |
15541 | ||
15542 | if (!failed_compressors.empty()) { | |
15543 | if (!s0.empty()) { | |
15544 | s0 += ", "; | |
15545 | } | |
15546 | s0 += "unable to load:"; | |
15547 | bool first = true; | |
15548 | for (auto& s : failed_compressors) { | |
15549 | if (first) { | |
15550 | first = false; | |
15551 | } else { | |
15552 | s0 += ", "; | |
15553 | } | |
15554 | s0 += s; | |
15555 | } | |
15556 | alerts.emplace( | |
15557 | "BLUESTORE_NO_COMPRESSION", | |
15558 | s0); | |
15559 | } | |
15560 | } | |
15561 | ||
9f95a23c TL |
15562 | void BlueStore::_collect_allocation_stats(uint64_t need, uint32_t alloc_size, |
15563 | size_t extents) | |
15564 | { | |
15565 | alloc_stats_count++; | |
15566 | alloc_stats_fragments += extents; | |
15567 | alloc_stats_size += need; | |
15568 | } | |
15569 | ||
15570 | void BlueStore::_record_allocation_stats() | |
15571 | { | |
15572 | // don't care about data consistency, | |
15573 | // fields can be partially modified while making the tuple | |
15574 | auto t0 = std::make_tuple( | |
15575 | alloc_stats_count.exchange(0), | |
15576 | alloc_stats_fragments.exchange(0), | |
15577 | alloc_stats_size.exchange(0)); | |
15578 | ||
15579 | dout(0) << " allocation stats probe " | |
15580 | << probe_count << ":" | |
15581 | << " cnt: " << std::get<0>(t0) | |
15582 | << " frags: " << std::get<1>(t0) | |
15583 | << " size: " << std::get<2>(t0) | |
15584 | << dendl; | |
15585 | ||
15586 | ||
15587 | // | |
15588 | // Keep the history for probes from the power-of-two sequence: | |
15589 | // -1, -2, -4, -8, -16 | |
15590 | // | |
15591 | size_t base = 1; | |
15592 | for (auto& t : alloc_stats_history) { | |
15593 | dout(0) << " probe -" | |
15594 | << base + (probe_count % base) << ": " | |
15595 | << std::get<0>(t) | |
15596 | << ", " << std::get<1>(t) | |
15597 | << ", " << std::get<2>(t) | |
15598 | << dendl; | |
15599 | base <<= 1; | |
15600 | } | |
15601 | dout(0) << "------------" << dendl; | |
15602 | ||
15603 | auto prev = probe_count++; | |
15604 | auto mask = (1 << alloc_stats_history.size()) - 1; | |
15605 | probe_count &= mask; | |
15606 | ||
15607 | for (size_t i = cbits(prev ^ probe_count) - 1; i > 0 ; --i) { | |
15608 | alloc_stats_history[i] = alloc_stats_history[i - 1]; | |
15609 | } | |
15610 | alloc_stats_history[0].swap(t0); | |
15611 | } | |
15612 | ||
7c673cae | 15613 | // =========================================== |
11fdf7f2 TL |
15614 | // BlueStoreRepairer |
15615 | ||
15616 | size_t BlueStoreRepairer::StoreSpaceTracker::filter_out( | |
15617 | const interval_set<uint64_t>& extents) | |
15618 | { | |
15619 | ceph_assert(granularity); // initialized | |
15620 | // can't call for the second time | |
15621 | ceph_assert(!was_filtered_out); | |
15622 | ceph_assert(collections_bfs.size() == objects_bfs.size()); | |
15623 | ||
15624 | uint64_t prev_pos = 0; | |
15625 | uint64_t npos = collections_bfs.size(); | |
15626 | ||
15627 | bloom_vector collections_reduced; | |
15628 | bloom_vector objects_reduced; | |
15629 | ||
15630 | for (auto e : extents) { | |
15631 | if (e.second == 0) { | |
15632 | continue; | |
15633 | } | |
15634 | uint64_t pos = max(e.first / granularity, prev_pos); | |
15635 | uint64_t end_pos = 1 + (e.first + e.second - 1) / granularity; | |
15636 | while (pos != npos && pos < end_pos) { | |
15637 | ceph_assert( collections_bfs[pos].element_count() == | |
15638 | objects_bfs[pos].element_count()); | |
15639 | if (collections_bfs[pos].element_count()) { | |
15640 | collections_reduced.push_back(std::move(collections_bfs[pos])); | |
15641 | objects_reduced.push_back(std::move(objects_bfs[pos])); | |
15642 | } | |
15643 | ++pos; | |
15644 | } | |
15645 | prev_pos = end_pos; | |
15646 | } | |
15647 | collections_reduced.swap(collections_bfs); | |
15648 | objects_reduced.swap(objects_bfs); | |
15649 | was_filtered_out = true; | |
15650 | return collections_bfs.size(); | |
15651 | } | |
15652 | ||
15653 | bool BlueStoreRepairer::remove_key(KeyValueDB *db, | |
15654 | const string& prefix, | |
15655 | const string& key) | |
15656 | { | |
15657 | if (!remove_key_txn) { | |
15658 | remove_key_txn = db->get_transaction(); | |
15659 | } | |
15660 | ++to_repair_cnt; | |
15661 | remove_key_txn->rmkey(prefix, key); | |
15662 | ||
15663 | return true; | |
15664 | } | |
15665 | ||
9f95a23c TL |
15666 | void BlueStoreRepairer::fix_per_pool_omap(KeyValueDB *db) |
15667 | { | |
15668 | fix_per_pool_omap_txn = db->get_transaction(); | |
15669 | ++to_repair_cnt; | |
15670 | bufferlist bl; | |
15671 | bl.append("1"); | |
15672 | fix_per_pool_omap_txn->set(PREFIX_SUPER, "per_pool_omap", bl); | |
15673 | } | |
15674 | ||
11fdf7f2 TL |
15675 | bool BlueStoreRepairer::fix_shared_blob( |
15676 | KeyValueDB *db, | |
15677 | uint64_t sbid, | |
15678 | const bufferlist* bl) | |
15679 | { | |
15680 | KeyValueDB::Transaction txn; | |
15681 | if (fix_misreferences_txn) { // reuse this txn | |
15682 | txn = fix_misreferences_txn; | |
15683 | } else { | |
15684 | if (!fix_shared_blob_txn) { | |
15685 | fix_shared_blob_txn = db->get_transaction(); | |
15686 | } | |
15687 | txn = fix_shared_blob_txn; | |
15688 | } | |
15689 | string key; | |
15690 | get_shared_blob_key(sbid, &key); | |
15691 | ||
15692 | ++to_repair_cnt; | |
15693 | if (bl) { | |
15694 | txn->set(PREFIX_SHARED_BLOB, key, *bl); | |
15695 | } else { | |
15696 | txn->rmkey(PREFIX_SHARED_BLOB, key); | |
15697 | } | |
15698 | return true; | |
15699 | } | |
15700 | ||
15701 | bool BlueStoreRepairer::fix_statfs(KeyValueDB *db, | |
15702 | const string& key, | |
15703 | const store_statfs_t& new_statfs) | |
15704 | { | |
15705 | if (!fix_statfs_txn) { | |
15706 | fix_statfs_txn = db->get_transaction(); | |
15707 | } | |
15708 | BlueStore::volatile_statfs vstatfs; | |
15709 | vstatfs = new_statfs; | |
15710 | bufferlist bl; | |
15711 | vstatfs.encode(bl); | |
15712 | ++to_repair_cnt; | |
15713 | fix_statfs_txn->set(PREFIX_STAT, key, bl); | |
15714 | return true; | |
15715 | } | |
15716 | ||
15717 | bool BlueStoreRepairer::fix_leaked(KeyValueDB *db, | |
15718 | FreelistManager* fm, | |
15719 | uint64_t offset, uint64_t len) | |
15720 | { | |
15721 | if (!fix_fm_leaked_txn) { | |
15722 | fix_fm_leaked_txn = db->get_transaction(); | |
15723 | } | |
15724 | ++to_repair_cnt; | |
15725 | fm->release(offset, len, fix_fm_leaked_txn); | |
15726 | return true; | |
15727 | } | |
15728 | bool BlueStoreRepairer::fix_false_free(KeyValueDB *db, | |
15729 | FreelistManager* fm, | |
15730 | uint64_t offset, uint64_t len) | |
15731 | { | |
15732 | if (!fix_fm_false_free_txn) { | |
15733 | fix_fm_false_free_txn = db->get_transaction(); | |
15734 | } | |
15735 | ++to_repair_cnt; | |
15736 | fm->allocate(offset, len, fix_fm_false_free_txn); | |
15737 | return true; | |
15738 | } | |
15739 | ||
15740 | bool BlueStoreRepairer::fix_bluefs_extents(std::atomic<uint64_t>& out_of_sync_flag) | |
15741 | { | |
15742 | // this is just a stub to count num of repairs properly, | |
15743 | // actual repair happens in BlueStore::_close_db_and_around() | |
15744 | // while doing _sync_bluefs_and_fm | |
15745 | ++out_of_sync_flag; | |
15746 | ++to_repair_cnt; | |
15747 | return true; | |
15748 | } | |
15749 | ||
15750 | bool BlueStoreRepairer::preprocess_misreference(KeyValueDB *db) | |
15751 | { | |
15752 | if (misreferenced_extents.size()) { | |
15753 | size_t n = space_usage_tracker.filter_out(misreferenced_extents); | |
15754 | ceph_assert(n > 0); | |
15755 | if (!fix_misreferences_txn) { | |
15756 | fix_misreferences_txn = db->get_transaction(); | |
15757 | } | |
15758 | return true; | |
15759 | } | |
15760 | return false; | |
15761 | } | |
15762 | ||
15763 | unsigned BlueStoreRepairer::apply(KeyValueDB* db) | |
15764 | { | |
9f95a23c TL |
15765 | if (fix_per_pool_omap_txn) { |
15766 | db->submit_transaction_sync(fix_per_pool_omap_txn); | |
15767 | fix_per_pool_omap_txn = nullptr; | |
15768 | } | |
11fdf7f2 TL |
15769 | if (fix_fm_leaked_txn) { |
15770 | db->submit_transaction_sync(fix_fm_leaked_txn); | |
15771 | fix_fm_leaked_txn = nullptr; | |
15772 | } | |
15773 | if (fix_fm_false_free_txn) { | |
15774 | db->submit_transaction_sync(fix_fm_false_free_txn); | |
15775 | fix_fm_false_free_txn = nullptr; | |
15776 | } | |
15777 | if (remove_key_txn) { | |
15778 | db->submit_transaction_sync(remove_key_txn); | |
15779 | remove_key_txn = nullptr; | |
15780 | } | |
15781 | if (fix_misreferences_txn) { | |
15782 | db->submit_transaction_sync(fix_misreferences_txn); | |
15783 | fix_misreferences_txn = nullptr; | |
15784 | } | |
15785 | if (fix_shared_blob_txn) { | |
15786 | db->submit_transaction_sync(fix_shared_blob_txn); | |
15787 | fix_shared_blob_txn = nullptr; | |
15788 | } | |
15789 | ||
15790 | if (fix_statfs_txn) { | |
15791 | db->submit_transaction_sync(fix_statfs_txn); | |
15792 | fix_statfs_txn = nullptr; | |
15793 | } | |
15794 | unsigned repaired = to_repair_cnt; | |
15795 | to_repair_cnt = 0; | |
15796 | return repaired; | |
15797 | } | |
15798 | ||
15799 | // ======================================================= | |
9f95a23c TL |
15800 | // RocksDBBlueFSVolumeSelector |
15801 | ||
15802 | uint8_t RocksDBBlueFSVolumeSelector::select_prefer_bdev(void* h) { | |
15803 | ceph_assert(h != nullptr); | |
15804 | uint64_t hint = reinterpret_cast<uint64_t>(h); | |
15805 | uint8_t res; | |
15806 | switch (hint) { | |
15807 | case LEVEL_SLOW: | |
15808 | res = BlueFS::BDEV_SLOW; | |
15809 | if (db_avail4slow > 0) { | |
15810 | // considering statically available db space vs. | |
15811 | // - observed maximums on DB dev for DB/WAL/UNSORTED data | |
15812 | // - observed maximum spillovers | |
15813 | uint64_t max_db_use = 0; // max db usage we potentially observed | |
f6b5b4d7 | 15814 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_LOG - LEVEL_FIRST); |
9f95a23c TL |
15815 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_WAL - LEVEL_FIRST); |
15816 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_DB, LEVEL_DB - LEVEL_FIRST); | |
15817 | // this could go to db hence using it in the estimation | |
15818 | max_db_use += per_level_per_dev_max.at(BlueFS::BDEV_SLOW, LEVEL_DB - LEVEL_FIRST); | |
15819 | ||
15820 | auto db_total = l_totals[LEVEL_DB - LEVEL_FIRST]; | |
15821 | uint64_t avail = min( | |
15822 | db_avail4slow, | |
15823 | max_db_use < db_total ? db_total - max_db_use : 0); | |
15824 | ||
15825 | // considering current DB dev usage for SLOW data | |
15826 | if (avail > per_level_per_dev_usage.at(BlueFS::BDEV_DB, LEVEL_SLOW - LEVEL_FIRST)) { | |
15827 | res = BlueFS::BDEV_DB; | |
15828 | } | |
15829 | } | |
15830 | break; | |
f6b5b4d7 | 15831 | case LEVEL_LOG: |
9f95a23c TL |
15832 | case LEVEL_WAL: |
15833 | res = BlueFS::BDEV_WAL; | |
15834 | break; | |
15835 | case LEVEL_DB: | |
15836 | default: | |
15837 | res = BlueFS::BDEV_DB; | |
15838 | break; | |
15839 | } | |
15840 | return res; | |
15841 | } | |
15842 | ||
15843 | void RocksDBBlueFSVolumeSelector::get_paths(const std::string& base, paths& res) const | |
15844 | { | |
15845 | res.emplace_back(base, l_totals[LEVEL_DB - LEVEL_FIRST]); | |
15846 | res.emplace_back(base + ".slow", l_totals[LEVEL_SLOW - LEVEL_FIRST]); | |
15847 | } | |
15848 | ||
15849 | void* RocksDBBlueFSVolumeSelector::get_hint_by_dir(const string& dirname) const { | |
15850 | uint8_t res = LEVEL_DB; | |
15851 | if (dirname.length() > 5) { | |
15852 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
15853 | // match up with bluestore. the slow device is always the second | |
15854 | // one (when a dedicated block.db device is present and used at | |
15855 | // bdev 0). the wal device is always last. | |
15856 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
15857 | res = LEVEL_SLOW; | |
15858 | } | |
15859 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
15860 | res = LEVEL_WAL; | |
15861 | } | |
15862 | } | |
15863 | return reinterpret_cast<void*>(res); | |
15864 | } | |
15865 | ||
15866 | void RocksDBBlueFSVolumeSelector::dump(ostream& sout) { | |
15867 | auto max_x = per_level_per_dev_usage.get_max_x(); | |
15868 | auto max_y = per_level_per_dev_usage.get_max_y(); | |
15869 | sout << "RocksDBBlueFSVolumeSelector: wal_total:" << l_totals[LEVEL_WAL - LEVEL_FIRST] | |
15870 | << ", db_total:" << l_totals[LEVEL_DB - LEVEL_FIRST] | |
15871 | << ", slow_total:" << l_totals[LEVEL_SLOW - LEVEL_FIRST] | |
15872 | << ", db_avail:" << db_avail4slow << std::endl | |
15873 | << "Usage matrix:" << std::endl; | |
f6b5b4d7 | 15874 | constexpr std::array<const char*, 8> names{ { |
9f95a23c TL |
15875 | "DEV/LEV", |
15876 | "WAL", | |
15877 | "DB", | |
15878 | "SLOW", | |
15879 | "*", | |
15880 | "*", | |
f6b5b4d7 TL |
15881 | "REAL", |
15882 | "FILES", | |
9f95a23c TL |
15883 | } }; |
15884 | const size_t width = 12; | |
15885 | for (size_t i = 0; i < names.size(); ++i) { | |
15886 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15887 | sout.width(width); | |
15888 | sout << names[i]; | |
15889 | } | |
15890 | sout << std::endl; | |
15891 | for (size_t l = 0; l < max_y; l++) { | |
15892 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15893 | sout.width(width); | |
15894 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
15895 | case LEVEL_LOG: |
15896 | sout << "LOG"; break; | |
9f95a23c TL |
15897 | case LEVEL_WAL: |
15898 | sout << "WAL"; break; | |
15899 | case LEVEL_DB: | |
15900 | sout << "DB"; break; | |
15901 | case LEVEL_SLOW: | |
15902 | sout << "SLOW"; break; | |
15903 | case LEVEL_MAX: | |
15904 | sout << "TOTALS"; break; | |
15905 | } | |
f6b5b4d7 | 15906 | for (size_t d = 0; d < max_x; d++) { |
9f95a23c TL |
15907 | sout.setf(std::ios::left, std::ios::adjustfield); |
15908 | sout.width(width); | |
15909 | sout << stringify(byte_u_t(per_level_per_dev_usage.at(d, l))); | |
15910 | } | |
15911 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15912 | sout.width(width); | |
f6b5b4d7 | 15913 | sout << stringify(per_level_files[l]) << std::endl; |
9f95a23c TL |
15914 | } |
15915 | ceph_assert(max_x == per_level_per_dev_max.get_max_x()); | |
15916 | ceph_assert(max_y == per_level_per_dev_max.get_max_y()); | |
15917 | sout << "MAXIMUMS:" << std::endl; | |
15918 | for (size_t l = 0; l < max_y; l++) { | |
15919 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15920 | sout.width(width); | |
15921 | switch (l + LEVEL_FIRST) { | |
f6b5b4d7 TL |
15922 | case LEVEL_LOG: |
15923 | sout << "LOG"; break; | |
9f95a23c TL |
15924 | case LEVEL_WAL: |
15925 | sout << "WAL"; break; | |
15926 | case LEVEL_DB: | |
15927 | sout << "DB"; break; | |
15928 | case LEVEL_SLOW: | |
15929 | sout << "SLOW"; break; | |
15930 | case LEVEL_MAX: | |
15931 | sout << "TOTALS"; break; | |
15932 | } | |
15933 | for (size_t d = 0; d < max_x - 1; d++) { | |
15934 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15935 | sout.width(width); | |
15936 | sout << stringify(byte_u_t(per_level_per_dev_max.at(d, l))); | |
15937 | } | |
15938 | sout.setf(std::ios::left, std::ios::adjustfield); | |
15939 | sout.width(width); | |
15940 | sout << stringify(byte_u_t(per_level_per_dev_max.at(max_x - 1, l))); | |
15941 | if (l < max_y - 1) { | |
15942 | sout << std::endl; | |
15943 | } | |
15944 | } | |
15945 | } | |
11fdf7f2 | 15946 | |
9f95a23c | 15947 | // ======================================================= |