]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
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 | #ifndef CEPH_OSD_BLUESTORE_H | |
16 | #define CEPH_OSD_BLUESTORE_H | |
17 | ||
18 | #include "acconfig.h" | |
19 | ||
20 | #include <unistd.h> | |
21 | ||
22 | #include <atomic> | |
23 | #include <mutex> | |
24 | #include <condition_variable> | |
25 | ||
26 | #include <boost/intrusive/list.hpp> | |
27 | #include <boost/intrusive/unordered_set.hpp> | |
28 | #include <boost/intrusive/set.hpp> | |
29 | #include <boost/functional/hash.hpp> | |
30 | #include <boost/dynamic_bitset.hpp> | |
31 | ||
11fdf7f2 | 32 | #include "include/ceph_assert.h" |
7c673cae | 33 | #include "include/unordered_map.h" |
7c673cae | 34 | #include "include/mempool.h" |
11fdf7f2 | 35 | #include "common/bloom_filter.hpp" |
7c673cae | 36 | #include "common/Finisher.h" |
11fdf7f2 | 37 | #include "common/Throttle.h" |
7c673cae | 38 | #include "common/perf_counters.h" |
91327a77 | 39 | #include "common/PriorityCache.h" |
7c673cae FG |
40 | #include "compressor/Compressor.h" |
41 | #include "os/ObjectStore.h" | |
42 | ||
43 | #include "bluestore_types.h" | |
44 | #include "BlockDevice.h" | |
11fdf7f2 | 45 | #include "BlueFS.h" |
7c673cae FG |
46 | #include "common/EventTrace.h" |
47 | ||
48 | class Allocator; | |
49 | class FreelistManager; | |
11fdf7f2 | 50 | class BlueStoreRepairer; |
7c673cae FG |
51 | |
52 | //#define DEBUG_CACHE | |
53 | //#define DEBUG_DEFERRED | |
54 | ||
31f18b77 FG |
55 | |
56 | ||
57 | // constants for Buffer::optimize() | |
58 | #define MAX_BUFFER_SLOP_RATIO_DEN 8 // so actually 1/N | |
59 | ||
60 | ||
7c673cae FG |
61 | enum { |
62 | l_bluestore_first = 732430, | |
63 | l_bluestore_kv_flush_lat, | |
64 | l_bluestore_kv_commit_lat, | |
11fdf7f2 TL |
65 | l_bluestore_kv_sync_lat, |
66 | l_bluestore_kv_final_lat, | |
7c673cae FG |
67 | l_bluestore_state_prepare_lat, |
68 | l_bluestore_state_aio_wait_lat, | |
69 | l_bluestore_state_io_done_lat, | |
70 | l_bluestore_state_kv_queued_lat, | |
71 | l_bluestore_state_kv_committing_lat, | |
72 | l_bluestore_state_kv_done_lat, | |
73 | l_bluestore_state_deferred_queued_lat, | |
74 | l_bluestore_state_deferred_aio_wait_lat, | |
75 | l_bluestore_state_deferred_cleanup_lat, | |
76 | l_bluestore_state_finishing_lat, | |
77 | l_bluestore_state_done_lat, | |
78 | l_bluestore_throttle_lat, | |
79 | l_bluestore_submit_lat, | |
80 | l_bluestore_commit_lat, | |
81 | l_bluestore_read_lat, | |
82 | l_bluestore_read_onode_meta_lat, | |
83 | l_bluestore_read_wait_aio_lat, | |
84 | l_bluestore_compress_lat, | |
85 | l_bluestore_decompress_lat, | |
86 | l_bluestore_csum_lat, | |
87 | l_bluestore_compress_success_count, | |
88 | l_bluestore_compress_rejected_count, | |
89 | l_bluestore_write_pad_bytes, | |
90 | l_bluestore_deferred_write_ops, | |
91 | l_bluestore_deferred_write_bytes, | |
92 | l_bluestore_write_penalty_read_ops, | |
93 | l_bluestore_allocated, | |
94 | l_bluestore_stored, | |
95 | l_bluestore_compressed, | |
96 | l_bluestore_compressed_allocated, | |
97 | l_bluestore_compressed_original, | |
98 | l_bluestore_onodes, | |
99 | l_bluestore_onode_hits, | |
100 | l_bluestore_onode_misses, | |
101 | l_bluestore_onode_shard_hits, | |
102 | l_bluestore_onode_shard_misses, | |
103 | l_bluestore_extents, | |
104 | l_bluestore_blobs, | |
105 | l_bluestore_buffers, | |
106 | l_bluestore_buffer_bytes, | |
107 | l_bluestore_buffer_hit_bytes, | |
108 | l_bluestore_buffer_miss_bytes, | |
109 | l_bluestore_write_big, | |
110 | l_bluestore_write_big_bytes, | |
111 | l_bluestore_write_big_blobs, | |
112 | l_bluestore_write_small, | |
113 | l_bluestore_write_small_bytes, | |
114 | l_bluestore_write_small_unused, | |
115 | l_bluestore_write_small_deferred, | |
116 | l_bluestore_write_small_pre_read, | |
117 | l_bluestore_write_small_new, | |
118 | l_bluestore_txc, | |
119 | l_bluestore_onode_reshard, | |
120 | l_bluestore_blob_split, | |
121 | l_bluestore_extent_compress, | |
122 | l_bluestore_gc_merged, | |
b32b8144 | 123 | l_bluestore_read_eio, |
f64942e4 | 124 | l_bluestore_reads_with_retries, |
a8e16298 | 125 | l_bluestore_fragmentation, |
11fdf7f2 TL |
126 | l_bluestore_omap_seek_to_first_lat, |
127 | l_bluestore_omap_upper_bound_lat, | |
128 | l_bluestore_omap_lower_bound_lat, | |
129 | l_bluestore_omap_next_lat, | |
7c673cae FG |
130 | l_bluestore_last |
131 | }; | |
132 | ||
11fdf7f2 TL |
133 | #define META_POOL_ID ((uint64_t)-1ull) |
134 | ||
7c673cae | 135 | class BlueStore : public ObjectStore, |
11fdf7f2 | 136 | public BlueFSDeviceExpander, |
7c673cae FG |
137 | public md_config_obs_t { |
138 | // ----------------------------------------------------- | |
139 | // types | |
140 | public: | |
141 | // config observer | |
142 | const char** get_tracked_conf_keys() const override; | |
11fdf7f2 TL |
143 | void handle_conf_change(const ConfigProxy& conf, |
144 | const std::set<std::string> &changed) override; | |
145 | ||
146 | //handler for discard event | |
147 | void handle_discard(interval_set<uint64_t>& to_release); | |
7c673cae FG |
148 | |
149 | void _set_csum(); | |
150 | void _set_compression(); | |
151 | void _set_throttle_params(); | |
31f18b77 | 152 | int _set_cache_sizes(); |
7c673cae FG |
153 | |
154 | class TransContext; | |
155 | ||
156 | typedef map<uint64_t, bufferlist> ready_regions_t; | |
157 | ||
158 | struct BufferSpace; | |
159 | struct Collection; | |
160 | typedef boost::intrusive_ptr<Collection> CollectionRef; | |
161 | ||
162 | struct AioContext { | |
163 | virtual void aio_finish(BlueStore *store) = 0; | |
164 | virtual ~AioContext() {} | |
165 | }; | |
166 | ||
167 | /// cached buffer | |
168 | struct Buffer { | |
169 | MEMPOOL_CLASS_HELPERS(); | |
170 | ||
171 | enum { | |
172 | STATE_EMPTY, ///< empty buffer -- used for cache history | |
173 | STATE_CLEAN, ///< clean data that is up to date | |
174 | STATE_WRITING, ///< data that is being written (io not yet complete) | |
175 | }; | |
176 | static const char *get_state_name(int s) { | |
177 | switch (s) { | |
178 | case STATE_EMPTY: return "empty"; | |
179 | case STATE_CLEAN: return "clean"; | |
180 | case STATE_WRITING: return "writing"; | |
181 | default: return "???"; | |
182 | } | |
183 | } | |
184 | enum { | |
185 | FLAG_NOCACHE = 1, ///< trim when done WRITING (do not become CLEAN) | |
186 | // NOTE: fix operator<< when you define a second flag | |
187 | }; | |
188 | static const char *get_flag_name(int s) { | |
189 | switch (s) { | |
190 | case FLAG_NOCACHE: return "nocache"; | |
191 | default: return "???"; | |
192 | } | |
193 | } | |
194 | ||
195 | BufferSpace *space; | |
196 | uint16_t state; ///< STATE_* | |
197 | uint16_t cache_private = 0; ///< opaque (to us) value used by Cache impl | |
198 | uint32_t flags; ///< FLAG_* | |
199 | uint64_t seq; | |
200 | uint32_t offset, length; | |
201 | bufferlist data; | |
202 | ||
203 | boost::intrusive::list_member_hook<> lru_item; | |
204 | boost::intrusive::list_member_hook<> state_item; | |
205 | ||
206 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, uint32_t l, | |
207 | unsigned f = 0) | |
208 | : space(space), state(s), flags(f), seq(q), offset(o), length(l) {} | |
209 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, bufferlist& b, | |
210 | unsigned f = 0) | |
211 | : space(space), state(s), flags(f), seq(q), offset(o), | |
212 | length(b.length()), data(b) {} | |
213 | ||
214 | bool is_empty() const { | |
215 | return state == STATE_EMPTY; | |
216 | } | |
217 | bool is_clean() const { | |
218 | return state == STATE_CLEAN; | |
219 | } | |
220 | bool is_writing() const { | |
221 | return state == STATE_WRITING; | |
222 | } | |
223 | ||
224 | uint32_t end() const { | |
225 | return offset + length; | |
226 | } | |
227 | ||
228 | void truncate(uint32_t newlen) { | |
11fdf7f2 | 229 | ceph_assert(newlen < length); |
7c673cae FG |
230 | if (data.length()) { |
231 | bufferlist t; | |
232 | t.substr_of(data, 0, newlen); | |
233 | data.claim(t); | |
234 | } | |
235 | length = newlen; | |
236 | } | |
31f18b77 FG |
237 | void maybe_rebuild() { |
238 | if (data.length() && | |
239 | (data.get_num_buffers() > 1 || | |
240 | data.front().wasted() > data.length() / MAX_BUFFER_SLOP_RATIO_DEN)) { | |
241 | data.rebuild(); | |
242 | } | |
243 | } | |
7c673cae FG |
244 | |
245 | void dump(Formatter *f) const { | |
246 | f->dump_string("state", get_state_name(state)); | |
247 | f->dump_unsigned("seq", seq); | |
248 | f->dump_unsigned("offset", offset); | |
249 | f->dump_unsigned("length", length); | |
250 | f->dump_unsigned("data_length", data.length()); | |
251 | } | |
252 | }; | |
253 | ||
254 | struct Cache; | |
255 | ||
256 | /// map logical extent range (object) onto buffers | |
257 | struct BufferSpace { | |
91327a77 AA |
258 | enum { |
259 | BYPASS_CLEAN_CACHE = 0x1, // bypass clean cache | |
260 | }; | |
261 | ||
7c673cae FG |
262 | typedef boost::intrusive::list< |
263 | Buffer, | |
264 | boost::intrusive::member_hook< | |
265 | Buffer, | |
266 | boost::intrusive::list_member_hook<>, | |
267 | &Buffer::state_item> > state_list_t; | |
268 | ||
31f18b77 | 269 | mempool::bluestore_cache_other::map<uint32_t, std::unique_ptr<Buffer>> |
7c673cae FG |
270 | buffer_map; |
271 | ||
272 | // we use a bare intrusive list here instead of std::map because | |
273 | // it uses less memory and we expect this to be very small (very | |
274 | // few IOs in flight to the same Blob at the same time). | |
275 | state_list_t writing; ///< writing buffers, sorted by seq, ascending | |
276 | ||
277 | ~BufferSpace() { | |
11fdf7f2 TL |
278 | ceph_assert(buffer_map.empty()); |
279 | ceph_assert(writing.empty()); | |
7c673cae FG |
280 | } |
281 | ||
282 | void _add_buffer(Cache* cache, Buffer *b, int level, Buffer *near) { | |
283 | cache->_audit("_add_buffer start"); | |
284 | buffer_map[b->offset].reset(b); | |
285 | if (b->is_writing()) { | |
31f18b77 | 286 | b->data.reassign_to_mempool(mempool::mempool_bluestore_writing); |
224ce89b WB |
287 | if (writing.empty() || writing.rbegin()->seq <= b->seq) { |
288 | writing.push_back(*b); | |
289 | } else { | |
290 | auto it = writing.begin(); | |
291 | while (it->seq < b->seq) { | |
292 | ++it; | |
293 | } | |
294 | ||
11fdf7f2 | 295 | ceph_assert(it->seq >= b->seq); |
224ce89b WB |
296 | // note that this will insert b before it |
297 | // hence the order is maintained | |
298 | writing.insert(it, *b); | |
299 | } | |
7c673cae | 300 | } else { |
31f18b77 | 301 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); |
7c673cae FG |
302 | cache->_add_buffer(b, level, near); |
303 | } | |
304 | cache->_audit("_add_buffer end"); | |
305 | } | |
306 | void _rm_buffer(Cache* cache, Buffer *b) { | |
307 | _rm_buffer(cache, buffer_map.find(b->offset)); | |
308 | } | |
31f18b77 FG |
309 | void _rm_buffer(Cache* cache, |
310 | map<uint32_t, std::unique_ptr<Buffer>>::iterator p) { | |
11fdf7f2 | 311 | ceph_assert(p != buffer_map.end()); |
7c673cae FG |
312 | cache->_audit("_rm_buffer start"); |
313 | if (p->second->is_writing()) { | |
314 | writing.erase(writing.iterator_to(*p->second)); | |
315 | } else { | |
316 | cache->_rm_buffer(p->second.get()); | |
317 | } | |
318 | buffer_map.erase(p); | |
319 | cache->_audit("_rm_buffer end"); | |
320 | } | |
321 | ||
322 | map<uint32_t,std::unique_ptr<Buffer>>::iterator _data_lower_bound( | |
323 | uint32_t offset) { | |
324 | auto i = buffer_map.lower_bound(offset); | |
325 | if (i != buffer_map.begin()) { | |
326 | --i; | |
327 | if (i->first + i->second->length <= offset) | |
328 | ++i; | |
329 | } | |
330 | return i; | |
331 | } | |
332 | ||
333 | // must be called under protection of the Cache lock | |
334 | void _clear(Cache* cache); | |
335 | ||
336 | // return value is the highest cache_private of a trimmed buffer, or 0. | |
337 | int discard(Cache* cache, uint32_t offset, uint32_t length) { | |
11fdf7f2 | 338 | std::lock_guard l(cache->lock); |
7c673cae FG |
339 | return _discard(cache, offset, length); |
340 | } | |
341 | int _discard(Cache* cache, uint32_t offset, uint32_t length); | |
342 | ||
343 | void write(Cache* cache, uint64_t seq, uint32_t offset, bufferlist& bl, | |
344 | unsigned flags) { | |
11fdf7f2 | 345 | std::lock_guard l(cache->lock); |
7c673cae FG |
346 | Buffer *b = new Buffer(this, Buffer::STATE_WRITING, seq, offset, bl, |
347 | flags); | |
348 | b->cache_private = _discard(cache, offset, bl.length()); | |
349 | _add_buffer(cache, b, (flags & Buffer::FLAG_NOCACHE) ? 0 : 1, nullptr); | |
350 | } | |
f64942e4 | 351 | void _finish_write(Cache* cache, uint64_t seq); |
7c673cae | 352 | void did_read(Cache* cache, uint32_t offset, bufferlist& bl) { |
11fdf7f2 | 353 | std::lock_guard l(cache->lock); |
7c673cae FG |
354 | Buffer *b = new Buffer(this, Buffer::STATE_CLEAN, 0, offset, bl); |
355 | b->cache_private = _discard(cache, offset, bl.length()); | |
356 | _add_buffer(cache, b, 1, nullptr); | |
357 | } | |
358 | ||
359 | void read(Cache* cache, uint32_t offset, uint32_t length, | |
360 | BlueStore::ready_regions_t& res, | |
91327a77 AA |
361 | interval_set<uint32_t>& res_intervals, |
362 | int flags = 0); | |
7c673cae FG |
363 | |
364 | void truncate(Cache* cache, uint32_t offset) { | |
365 | discard(cache, offset, (uint32_t)-1 - offset); | |
366 | } | |
367 | ||
368 | void split(Cache* cache, size_t pos, BufferSpace &r); | |
369 | ||
370 | void dump(Cache* cache, Formatter *f) const { | |
11fdf7f2 | 371 | std::lock_guard l(cache->lock); |
7c673cae FG |
372 | f->open_array_section("buffers"); |
373 | for (auto& i : buffer_map) { | |
374 | f->open_object_section("buffer"); | |
11fdf7f2 | 375 | ceph_assert(i.first == i.second->offset); |
7c673cae FG |
376 | i.second->dump(f); |
377 | f->close_section(); | |
378 | } | |
379 | f->close_section(); | |
380 | } | |
381 | }; | |
382 | ||
383 | struct SharedBlobSet; | |
384 | ||
385 | /// in-memory shared blob state (incl cached buffers) | |
386 | struct SharedBlob { | |
387 | MEMPOOL_CLASS_HELPERS(); | |
388 | ||
389 | std::atomic_int nref = {0}; ///< reference count | |
390 | bool loaded = false; | |
391 | ||
392 | CollectionRef coll; | |
393 | union { | |
394 | uint64_t sbid_unloaded; ///< sbid if persistent isn't loaded | |
395 | bluestore_shared_blob_t *persistent; ///< persistent part of the shared blob if any | |
396 | }; | |
397 | BufferSpace bc; ///< buffer cache | |
398 | ||
399 | SharedBlob(Collection *_coll) : coll(_coll), sbid_unloaded(0) { | |
400 | if (get_cache()) { | |
401 | get_cache()->add_blob(); | |
402 | } | |
403 | } | |
404 | SharedBlob(uint64_t i, Collection *_coll); | |
405 | ~SharedBlob(); | |
406 | ||
407 | uint64_t get_sbid() const { | |
408 | return loaded ? persistent->sbid : sbid_unloaded; | |
409 | } | |
410 | ||
411 | friend void intrusive_ptr_add_ref(SharedBlob *b) { b->get(); } | |
412 | friend void intrusive_ptr_release(SharedBlob *b) { b->put(); } | |
413 | ||
414 | friend ostream& operator<<(ostream& out, const SharedBlob& sb); | |
415 | ||
416 | void get() { | |
417 | ++nref; | |
418 | } | |
419 | void put(); | |
420 | ||
421 | /// get logical references | |
422 | void get_ref(uint64_t offset, uint32_t length); | |
423 | ||
424 | /// put logical references, and get back any released extents | |
425 | void put_ref(uint64_t offset, uint32_t length, | |
11fdf7f2 | 426 | PExtentVector *r, bool *unshare); |
7c673cae | 427 | |
f64942e4 AA |
428 | void finish_write(uint64_t seq); |
429 | ||
7c673cae FG |
430 | friend bool operator==(const SharedBlob &l, const SharedBlob &r) { |
431 | return l.get_sbid() == r.get_sbid(); | |
432 | } | |
433 | inline Cache* get_cache() { | |
434 | return coll ? coll->cache : nullptr; | |
435 | } | |
436 | inline SharedBlobSet* get_parent() { | |
437 | return coll ? &(coll->shared_blob_set) : nullptr; | |
438 | } | |
439 | inline bool is_loaded() const { | |
440 | return loaded; | |
441 | } | |
442 | ||
443 | }; | |
444 | typedef boost::intrusive_ptr<SharedBlob> SharedBlobRef; | |
445 | ||
446 | /// a lookup table of SharedBlobs | |
447 | struct SharedBlobSet { | |
11fdf7f2 TL |
448 | /// protect lookup, insertion, removal |
449 | ceph::mutex lock = ceph::make_mutex("BlueStore::SharedBlobSet::lock"); | |
7c673cae FG |
450 | |
451 | // we use a bare pointer because we don't want to affect the ref | |
452 | // count | |
31f18b77 | 453 | mempool::bluestore_cache_other::unordered_map<uint64_t,SharedBlob*> sb_map; |
7c673cae FG |
454 | |
455 | SharedBlobRef lookup(uint64_t sbid) { | |
11fdf7f2 | 456 | std::lock_guard l(lock); |
7c673cae | 457 | auto p = sb_map.find(sbid); |
28e407b8 AA |
458 | if (p == sb_map.end() || |
459 | p->second->nref == 0) { | |
7c673cae FG |
460 | return nullptr; |
461 | } | |
462 | return p->second; | |
463 | } | |
464 | ||
465 | void add(Collection* coll, SharedBlob *sb) { | |
11fdf7f2 | 466 | std::lock_guard l(lock); |
7c673cae FG |
467 | sb_map[sb->get_sbid()] = sb; |
468 | sb->coll = coll; | |
469 | } | |
470 | ||
91327a77 | 471 | bool remove(SharedBlob *sb, bool verify_nref_is_zero=false) { |
11fdf7f2 TL |
472 | std::lock_guard l(lock); |
473 | ceph_assert(sb->get_parent() == this); | |
91327a77 AA |
474 | if (verify_nref_is_zero && sb->nref != 0) { |
475 | return false; | |
476 | } | |
28e407b8 AA |
477 | // only remove if it still points to us |
478 | auto p = sb_map.find(sb->get_sbid()); | |
479 | if (p != sb_map.end() && | |
480 | p->second == sb) { | |
481 | sb_map.erase(p); | |
482 | } | |
91327a77 | 483 | return true; |
3efd9988 FG |
484 | } |
485 | ||
7c673cae | 486 | bool empty() { |
11fdf7f2 | 487 | std::lock_guard l(lock); |
7c673cae FG |
488 | return sb_map.empty(); |
489 | } | |
3efd9988 | 490 | |
11fdf7f2 TL |
491 | template <int LogLevelV> |
492 | void dump(CephContext *cct); | |
7c673cae FG |
493 | }; |
494 | ||
495 | //#define CACHE_BLOB_BL // not sure if this is a win yet or not... :/ | |
496 | ||
497 | /// in-memory blob metadata and associated cached buffers (if any) | |
498 | struct Blob { | |
499 | MEMPOOL_CLASS_HELPERS(); | |
500 | ||
501 | std::atomic_int nref = {0}; ///< reference count | |
502 | int16_t id = -1; ///< id, for spanning blobs only, >= 0 | |
503 | int16_t last_encoded_id = -1; ///< (ephemeral) used during encoding only | |
504 | SharedBlobRef shared_blob; ///< shared blob state (if any) | |
505 | ||
506 | private: | |
507 | mutable bluestore_blob_t blob; ///< decoded blob metadata | |
508 | #ifdef CACHE_BLOB_BL | |
509 | mutable bufferlist blob_bl; ///< cached encoded blob, blob is dirty if empty | |
510 | #endif | |
511 | /// refs from this shard. ephemeral if id<0, persisted if spanning. | |
512 | bluestore_blob_use_tracker_t used_in_blob; | |
513 | ||
514 | public: | |
515 | ||
516 | friend void intrusive_ptr_add_ref(Blob *b) { b->get(); } | |
517 | friend void intrusive_ptr_release(Blob *b) { b->put(); } | |
518 | ||
519 | friend ostream& operator<<(ostream& out, const Blob &b); | |
520 | ||
521 | const bluestore_blob_use_tracker_t& get_blob_use_tracker() const { | |
522 | return used_in_blob; | |
523 | } | |
524 | bool is_referenced() const { | |
525 | return used_in_blob.is_not_empty(); | |
526 | } | |
527 | uint32_t get_referenced_bytes() const { | |
528 | return used_in_blob.get_referenced_bytes(); | |
529 | } | |
530 | ||
531 | bool is_spanning() const { | |
532 | return id >= 0; | |
533 | } | |
534 | ||
535 | bool can_split() const { | |
11fdf7f2 | 536 | std::lock_guard l(shared_blob->get_cache()->lock); |
7c673cae FG |
537 | // splitting a BufferSpace writing list is too hard; don't try. |
538 | return shared_blob->bc.writing.empty() && | |
539 | used_in_blob.can_split() && | |
540 | get_blob().can_split(); | |
541 | } | |
542 | ||
543 | bool can_split_at(uint32_t blob_offset) const { | |
544 | return used_in_blob.can_split_at(blob_offset) && | |
545 | get_blob().can_split_at(blob_offset); | |
546 | } | |
547 | ||
224ce89b | 548 | bool can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
549 | uint32_t target_blob_size, |
550 | uint32_t b_offset, | |
551 | uint32_t *length0); | |
552 | ||
553 | void dup(Blob& o) { | |
554 | o.shared_blob = shared_blob; | |
555 | o.blob = blob; | |
556 | #ifdef CACHE_BLOB_BL | |
557 | o.blob_bl = blob_bl; | |
558 | #endif | |
559 | } | |
560 | ||
224ce89b | 561 | inline const bluestore_blob_t& get_blob() const { |
7c673cae FG |
562 | return blob; |
563 | } | |
224ce89b | 564 | inline bluestore_blob_t& dirty_blob() { |
7c673cae FG |
565 | #ifdef CACHE_BLOB_BL |
566 | blob_bl.clear(); | |
567 | #endif | |
568 | return blob; | |
569 | } | |
570 | ||
571 | /// discard buffers for unallocated regions | |
572 | void discard_unallocated(Collection *coll); | |
573 | ||
574 | /// get logical references | |
575 | void get_ref(Collection *coll, uint32_t offset, uint32_t length); | |
576 | /// put logical references, and get back any released extents | |
577 | bool put_ref(Collection *coll, uint32_t offset, uint32_t length, | |
578 | PExtentVector *r); | |
579 | ||
580 | /// split the blob | |
581 | void split(Collection *coll, uint32_t blob_offset, Blob *o); | |
582 | ||
583 | void get() { | |
584 | ++nref; | |
585 | } | |
586 | void put() { | |
587 | if (--nref == 0) | |
588 | delete this; | |
589 | } | |
590 | ||
591 | ||
592 | #ifdef CACHE_BLOB_BL | |
593 | void _encode() const { | |
594 | if (blob_bl.length() == 0 ) { | |
11fdf7f2 | 595 | encode(blob, blob_bl); |
7c673cae | 596 | } else { |
11fdf7f2 | 597 | ceph_assert(blob_bl.length()); |
7c673cae FG |
598 | } |
599 | } | |
600 | void bound_encode( | |
601 | size_t& p, | |
602 | bool include_ref_map) const { | |
603 | _encode(); | |
604 | p += blob_bl.length(); | |
605 | if (include_ref_map) { | |
606 | used_in_blob.bound_encode(p); | |
607 | } | |
608 | } | |
609 | void encode( | |
610 | bufferlist::contiguous_appender& p, | |
611 | bool include_ref_map) const { | |
612 | _encode(); | |
613 | p.append(blob_bl); | |
614 | if (include_ref_map) { | |
615 | used_in_blob.encode(p); | |
616 | } | |
617 | } | |
618 | void decode( | |
619 | Collection */*coll*/, | |
11fdf7f2 | 620 | bufferptr::const_iterator& p, |
7c673cae FG |
621 | bool include_ref_map) { |
622 | const char *start = p.get_pos(); | |
623 | denc(blob, p); | |
624 | const char *end = p.get_pos(); | |
625 | blob_bl.clear(); | |
626 | blob_bl.append(start, end - start); | |
627 | if (include_ref_map) { | |
628 | used_in_blob.decode(p); | |
629 | } | |
630 | } | |
631 | #else | |
632 | void bound_encode( | |
633 | size_t& p, | |
634 | uint64_t struct_v, | |
635 | uint64_t sbid, | |
636 | bool include_ref_map) const { | |
637 | denc(blob, p, struct_v); | |
638 | if (blob.is_shared()) { | |
639 | denc(sbid, p); | |
640 | } | |
641 | if (include_ref_map) { | |
642 | used_in_blob.bound_encode(p); | |
643 | } | |
644 | } | |
645 | void encode( | |
646 | bufferlist::contiguous_appender& p, | |
647 | uint64_t struct_v, | |
648 | uint64_t sbid, | |
649 | bool include_ref_map) const { | |
650 | denc(blob, p, struct_v); | |
651 | if (blob.is_shared()) { | |
652 | denc(sbid, p); | |
653 | } | |
654 | if (include_ref_map) { | |
655 | used_in_blob.encode(p); | |
656 | } | |
657 | } | |
658 | void decode( | |
659 | Collection *coll, | |
11fdf7f2 | 660 | bufferptr::const_iterator& p, |
7c673cae FG |
661 | uint64_t struct_v, |
662 | uint64_t* sbid, | |
663 | bool include_ref_map); | |
664 | #endif | |
665 | }; | |
666 | typedef boost::intrusive_ptr<Blob> BlobRef; | |
31f18b77 | 667 | typedef mempool::bluestore_cache_other::map<int,BlobRef> blob_map_t; |
7c673cae FG |
668 | |
669 | /// a logical extent, pointing to (some portion of) a blob | |
670 | typedef boost::intrusive::set_base_hook<boost::intrusive::optimize_size<true> > ExtentBase; //making an alias to avoid build warnings | |
671 | struct Extent : public ExtentBase { | |
672 | MEMPOOL_CLASS_HELPERS(); | |
673 | ||
674 | uint32_t logical_offset = 0; ///< logical offset | |
675 | uint32_t blob_offset = 0; ///< blob offset | |
676 | uint32_t length = 0; ///< length | |
677 | BlobRef blob; ///< the blob with our data | |
678 | ||
679 | /// ctor for lookup only | |
680 | explicit Extent(uint32_t lo) : ExtentBase(), logical_offset(lo) { } | |
681 | /// ctor for delayed initialization (see decode_some()) | |
682 | explicit Extent() : ExtentBase() { | |
683 | } | |
684 | /// ctor for general usage | |
685 | Extent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
686 | : ExtentBase(), | |
687 | logical_offset(lo), blob_offset(o), length(l) { | |
688 | assign_blob(b); | |
689 | } | |
690 | ~Extent() { | |
691 | if (blob) { | |
692 | blob->shared_blob->get_cache()->rm_extent(); | |
693 | } | |
694 | } | |
695 | ||
696 | void assign_blob(const BlobRef& b) { | |
11fdf7f2 | 697 | ceph_assert(!blob); |
7c673cae FG |
698 | blob = b; |
699 | blob->shared_blob->get_cache()->add_extent(); | |
700 | } | |
701 | ||
702 | // comparators for intrusive_set | |
703 | friend bool operator<(const Extent &a, const Extent &b) { | |
704 | return a.logical_offset < b.logical_offset; | |
705 | } | |
706 | friend bool operator>(const Extent &a, const Extent &b) { | |
707 | return a.logical_offset > b.logical_offset; | |
708 | } | |
709 | friend bool operator==(const Extent &a, const Extent &b) { | |
710 | return a.logical_offset == b.logical_offset; | |
711 | } | |
712 | ||
713 | uint32_t blob_start() const { | |
714 | return logical_offset - blob_offset; | |
715 | } | |
716 | ||
717 | uint32_t blob_end() const { | |
718 | return blob_start() + blob->get_blob().get_logical_length(); | |
719 | } | |
720 | ||
721 | uint32_t logical_end() const { | |
722 | return logical_offset + length; | |
723 | } | |
724 | ||
725 | // return true if any piece of the blob is out of | |
726 | // the given range [o, o + l]. | |
727 | bool blob_escapes_range(uint32_t o, uint32_t l) const { | |
728 | return blob_start() < o || blob_end() > o + l; | |
729 | } | |
730 | }; | |
731 | typedef boost::intrusive::set<Extent> extent_map_t; | |
732 | ||
733 | ||
734 | friend ostream& operator<<(ostream& out, const Extent& e); | |
735 | ||
736 | struct OldExtent { | |
737 | boost::intrusive::list_member_hook<> old_extent_item; | |
738 | Extent e; | |
739 | PExtentVector r; | |
740 | bool blob_empty; // flag to track the last removed extent that makes blob | |
741 | // empty - required to update compression stat properly | |
742 | OldExtent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
743 | : e(lo, o, l, b), blob_empty(false) { | |
744 | } | |
745 | static OldExtent* create(CollectionRef c, | |
746 | uint32_t lo, | |
747 | uint32_t o, | |
748 | uint32_t l, | |
749 | BlobRef& b); | |
750 | }; | |
751 | typedef boost::intrusive::list< | |
752 | OldExtent, | |
753 | boost::intrusive::member_hook< | |
754 | OldExtent, | |
755 | boost::intrusive::list_member_hook<>, | |
756 | &OldExtent::old_extent_item> > old_extent_map_t; | |
757 | ||
758 | struct Onode; | |
759 | ||
760 | /// a sharded extent map, mapping offsets to lextents to blobs | |
761 | struct ExtentMap { | |
762 | Onode *onode; | |
763 | extent_map_t extent_map; ///< map of Extents to Blobs | |
764 | blob_map_t spanning_blob_map; ///< blobs that span shards | |
11fdf7f2 | 765 | typedef boost::intrusive_ptr<Onode> OnodeRef; |
7c673cae FG |
766 | |
767 | struct Shard { | |
768 | bluestore_onode_t::shard_info *shard_info = nullptr; | |
769 | unsigned extents = 0; ///< count extents in this shard | |
770 | bool loaded = false; ///< true if shard is loaded | |
771 | bool dirty = false; ///< true if shard is dirty and needs reencoding | |
772 | }; | |
31f18b77 | 773 | mempool::bluestore_cache_other::vector<Shard> shards; ///< shards |
7c673cae FG |
774 | |
775 | bufferlist inline_bl; ///< cached encoded map, if unsharded; empty=>dirty | |
776 | ||
777 | uint32_t needs_reshard_begin = 0; | |
778 | uint32_t needs_reshard_end = 0; | |
779 | ||
11fdf7f2 TL |
780 | void dup(BlueStore* b, TransContext*, CollectionRef&, OnodeRef&, OnodeRef&, |
781 | uint64_t&, uint64_t&, uint64_t&); | |
782 | ||
7c673cae FG |
783 | bool needs_reshard() const { |
784 | return needs_reshard_end > needs_reshard_begin; | |
785 | } | |
786 | void clear_needs_reshard() { | |
787 | needs_reshard_begin = needs_reshard_end = 0; | |
788 | } | |
789 | void request_reshard(uint32_t begin, uint32_t end) { | |
790 | if (begin < needs_reshard_begin) { | |
791 | needs_reshard_begin = begin; | |
792 | } | |
793 | if (end > needs_reshard_end) { | |
794 | needs_reshard_end = end; | |
795 | } | |
796 | } | |
797 | ||
798 | struct DeleteDisposer { | |
799 | void operator()(Extent *e) { delete e; } | |
800 | }; | |
801 | ||
802 | ExtentMap(Onode *o); | |
803 | ~ExtentMap() { | |
804 | extent_map.clear_and_dispose(DeleteDisposer()); | |
805 | } | |
806 | ||
807 | void clear() { | |
808 | extent_map.clear_and_dispose(DeleteDisposer()); | |
809 | shards.clear(); | |
810 | inline_bl.clear(); | |
811 | clear_needs_reshard(); | |
812 | } | |
813 | ||
814 | bool encode_some(uint32_t offset, uint32_t length, bufferlist& bl, | |
815 | unsigned *pn); | |
816 | unsigned decode_some(bufferlist& bl); | |
817 | ||
818 | void bound_encode_spanning_blobs(size_t& p); | |
819 | void encode_spanning_blobs(bufferlist::contiguous_appender& p); | |
11fdf7f2 | 820 | void decode_spanning_blobs(bufferptr::const_iterator& p); |
7c673cae FG |
821 | |
822 | BlobRef get_spanning_blob(int id) { | |
823 | auto p = spanning_blob_map.find(id); | |
11fdf7f2 | 824 | ceph_assert(p != spanning_blob_map.end()); |
7c673cae FG |
825 | return p->second; |
826 | } | |
827 | ||
828 | void update(KeyValueDB::Transaction t, bool force); | |
31f18b77 | 829 | decltype(BlueStore::Blob::id) allocate_spanning_blob_id(); |
7c673cae FG |
830 | void reshard( |
831 | KeyValueDB *db, | |
832 | KeyValueDB::Transaction t); | |
833 | ||
834 | /// initialize Shards from the onode | |
835 | void init_shards(bool loaded, bool dirty); | |
836 | ||
837 | /// return index of shard containing offset | |
838 | /// or -1 if not found | |
839 | int seek_shard(uint32_t offset) { | |
840 | size_t end = shards.size(); | |
841 | size_t mid, left = 0; | |
842 | size_t right = end; // one passed the right end | |
843 | ||
844 | while (left < right) { | |
845 | mid = left + (right - left) / 2; | |
846 | if (offset >= shards[mid].shard_info->offset) { | |
847 | size_t next = mid + 1; | |
848 | if (next >= end || offset < shards[next].shard_info->offset) | |
849 | return mid; | |
850 | //continue to search forwards | |
851 | left = next; | |
852 | } else { | |
853 | //continue to search backwards | |
854 | right = mid; | |
855 | } | |
856 | } | |
857 | ||
858 | return -1; // not found | |
859 | } | |
860 | ||
861 | /// check if a range spans a shard | |
862 | bool spans_shard(uint32_t offset, uint32_t length) { | |
863 | if (shards.empty()) { | |
864 | return false; | |
865 | } | |
866 | int s = seek_shard(offset); | |
11fdf7f2 | 867 | ceph_assert(s >= 0); |
7c673cae FG |
868 | if (s == (int)shards.size() - 1) { |
869 | return false; // last shard | |
870 | } | |
871 | if (offset + length <= shards[s+1].shard_info->offset) { | |
872 | return false; | |
873 | } | |
874 | return true; | |
875 | } | |
876 | ||
877 | /// ensure that a range of the map is loaded | |
878 | void fault_range(KeyValueDB *db, | |
879 | uint32_t offset, uint32_t length); | |
880 | ||
881 | /// ensure a range of the map is marked dirty | |
31f18b77 | 882 | void dirty_range(uint32_t offset, uint32_t length); |
7c673cae | 883 | |
31f18b77 | 884 | /// for seek_lextent test |
7c673cae FG |
885 | extent_map_t::iterator find(uint64_t offset); |
886 | ||
7c673cae FG |
887 | /// seek to the first lextent including or after offset |
888 | extent_map_t::iterator seek_lextent(uint64_t offset); | |
889 | extent_map_t::const_iterator seek_lextent(uint64_t offset) const; | |
890 | ||
891 | /// add a new Extent | |
892 | void add(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) { | |
893 | extent_map.insert(*new Extent(lo, o, l, b)); | |
894 | } | |
895 | ||
896 | /// remove (and delete) an Extent | |
897 | void rm(extent_map_t::iterator p) { | |
898 | extent_map.erase_and_dispose(p, DeleteDisposer()); | |
899 | } | |
900 | ||
901 | bool has_any_lextents(uint64_t offset, uint64_t length); | |
902 | ||
903 | /// consolidate adjacent lextents in extent_map | |
904 | int compress_extent_map(uint64_t offset, uint64_t length); | |
905 | ||
906 | /// punch a logical hole. add lextents to deref to target list. | |
907 | void punch_hole(CollectionRef &c, | |
908 | uint64_t offset, uint64_t length, | |
909 | old_extent_map_t *old_extents); | |
910 | ||
911 | /// put new lextent into lextent_map overwriting existing ones if | |
912 | /// any and update references accordingly | |
913 | Extent *set_lextent(CollectionRef &c, | |
914 | uint64_t logical_offset, | |
915 | uint64_t offset, uint64_t length, | |
916 | BlobRef b, | |
917 | old_extent_map_t *old_extents); | |
918 | ||
919 | /// split a blob (and referring extents) | |
920 | BlobRef split_blob(BlobRef lb, uint32_t blob_offset, uint32_t pos); | |
921 | }; | |
922 | ||
923 | /// Compressed Blob Garbage collector | |
924 | /* | |
925 | The primary idea of the collector is to estimate a difference between | |
926 | allocation units(AU) currently present for compressed blobs and new AUs | |
927 | required to store that data uncompressed. | |
928 | Estimation is performed for protrusive extents within a logical range | |
929 | determined by a concatenation of old_extents collection and specific(current) | |
930 | write request. | |
931 | The root cause for old_extents use is the need to handle blob ref counts | |
932 | properly. Old extents still hold blob refs and hence we need to traverse | |
933 | the collection to determine if blob to be released. | |
934 | Protrusive extents are extents that fit into the blob set in action | |
935 | (ones that are below the logical range from above) but not removed totally | |
936 | due to the current write. | |
937 | E.g. for | |
938 | extent1 <loffs = 100, boffs = 100, len = 100> -> | |
939 | blob1<compressed, len_on_disk=4096, logical_len=8192> | |
940 | extent2 <loffs = 200, boffs = 200, len = 100> -> | |
941 | blob2<raw, len_on_disk=4096, llen=4096> | |
942 | extent3 <loffs = 300, boffs = 300, len = 100> -> | |
943 | blob1<compressed, len_on_disk=4096, llen=8192> | |
944 | extent4 <loffs = 4096, boffs = 0, len = 100> -> | |
945 | blob3<raw, len_on_disk=4096, llen=4096> | |
946 | write(300~100) | |
947 | protrusive extents are within the following ranges <0~300, 400~8192-400> | |
948 | In this case existing AUs that might be removed due to GC (i.e. blob1) | |
949 | use 2x4K bytes. | |
950 | And new AUs expected after GC = 0 since extent1 to be merged into blob2. | |
951 | Hence we should do a collect. | |
952 | */ | |
953 | class GarbageCollector | |
954 | { | |
955 | public: | |
956 | /// return amount of allocation units that might be saved due to GC | |
957 | int64_t estimate( | |
958 | uint64_t offset, | |
959 | uint64_t length, | |
960 | const ExtentMap& extent_map, | |
961 | const old_extent_map_t& old_extents, | |
962 | uint64_t min_alloc_size); | |
963 | ||
964 | /// return a collection of extents to perform GC on | |
a8e16298 | 965 | const vector<bluestore_pextent_t>& get_extents_to_collect() const { |
7c673cae FG |
966 | return extents_to_collect; |
967 | } | |
968 | GarbageCollector(CephContext* _cct) : cct(_cct) {} | |
969 | ||
970 | private: | |
971 | struct BlobInfo { | |
972 | uint64_t referenced_bytes = 0; ///< amount of bytes referenced in blob | |
973 | int64_t expected_allocations = 0; ///< new alloc units required | |
974 | ///< in case of gc fulfilled | |
975 | bool collect_candidate = false; ///< indicate if blob has any extents | |
976 | ///< eligible for GC. | |
977 | extent_map_t::const_iterator first_lextent; ///< points to the first | |
978 | ///< lextent referring to | |
979 | ///< the blob if any. | |
980 | ///< collect_candidate flag | |
981 | ///< determines the validity | |
982 | extent_map_t::const_iterator last_lextent; ///< points to the last | |
983 | ///< lextent referring to | |
984 | ///< the blob if any. | |
985 | ||
986 | BlobInfo(uint64_t ref_bytes) : | |
987 | referenced_bytes(ref_bytes) { | |
988 | } | |
989 | }; | |
990 | CephContext* cct; | |
991 | map<Blob*, BlobInfo> affected_blobs; ///< compressed blobs and their ref_map | |
992 | ///< copies that are affected by the | |
993 | ///< specific write | |
994 | ||
a8e16298 TL |
995 | ///< protrusive extents that should be collected if GC takes place |
996 | vector<bluestore_pextent_t> extents_to_collect; | |
7c673cae FG |
997 | |
998 | boost::optional<uint64_t > used_alloc_unit; ///< last processed allocation | |
999 | ///< unit when traversing | |
1000 | ///< protrusive extents. | |
1001 | ///< Other extents mapped to | |
1002 | ///< this AU to be ignored | |
1003 | ///< (except the case where | |
1004 | ///< uncompressed extent follows | |
1005 | ///< compressed one - see below). | |
1006 | BlobInfo* blob_info_counted = nullptr; ///< set if previous allocation unit | |
1007 | ///< caused expected_allocations | |
1008 | ///< counter increment at this blob. | |
1009 | ///< if uncompressed extent follows | |
1010 | ///< a decrement for the | |
1011 | ///< expected_allocations counter | |
1012 | ///< is needed | |
1013 | int64_t expected_allocations = 0; ///< new alloc units required in case | |
1014 | ///< of gc fulfilled | |
1015 | int64_t expected_for_release = 0; ///< alloc units currently used by | |
1016 | ///< compressed blobs that might | |
1017 | ///< gone after GC | |
11fdf7f2 TL |
1018 | uint64_t gc_start_offset = 0; ///starting offset for GC |
1019 | uint64_t gc_end_offset = 0; ///ending offset for GC | |
7c673cae FG |
1020 | |
1021 | protected: | |
1022 | void process_protrusive_extents(const BlueStore::ExtentMap& extent_map, | |
1023 | uint64_t start_offset, | |
1024 | uint64_t end_offset, | |
1025 | uint64_t start_touch_offset, | |
1026 | uint64_t end_touch_offset, | |
1027 | uint64_t min_alloc_size); | |
1028 | }; | |
1029 | ||
1030 | struct OnodeSpace; | |
1031 | ||
1032 | /// an in-memory object | |
1033 | struct Onode { | |
1034 | MEMPOOL_CLASS_HELPERS(); | |
1035 | ||
1036 | std::atomic_int nref; ///< reference count | |
1037 | Collection *c; | |
1038 | ||
1039 | ghobject_t oid; | |
1040 | ||
1041 | /// key under PREFIX_OBJ where we are stored | |
31f18b77 | 1042 | mempool::bluestore_cache_other::string key; |
7c673cae FG |
1043 | |
1044 | boost::intrusive::list_member_hook<> lru_item; | |
1045 | ||
1046 | bluestore_onode_t onode; ///< metadata stored as value in kv store | |
1047 | bool exists; ///< true if object logically exists | |
1048 | ||
1049 | ExtentMap extent_map; | |
1050 | ||
1051 | // track txc's that have not been committed to kv store (and whose | |
1052 | // effects cannot be read via the kvdb read methods) | |
1053 | std::atomic<int> flushing_count = {0}; | |
11fdf7f2 TL |
1054 | /// protect flush_txns |
1055 | ceph::mutex flush_lock = ceph::make_mutex("BlueStore::Onode::flush_lock"); | |
1056 | ceph::condition_variable flush_cond; ///< wait here for uncommitted txns | |
7c673cae FG |
1057 | |
1058 | Onode(Collection *c, const ghobject_t& o, | |
31f18b77 | 1059 | const mempool::bluestore_cache_other::string& k) |
7c673cae FG |
1060 | : nref(0), |
1061 | c(c), | |
1062 | oid(o), | |
1063 | key(k), | |
1064 | exists(false), | |
1065 | extent_map(this) { | |
1066 | } | |
1067 | ||
1068 | void flush(); | |
1069 | void get() { | |
1070 | ++nref; | |
1071 | } | |
1072 | void put() { | |
1073 | if (--nref == 0) | |
1074 | delete this; | |
1075 | } | |
1076 | }; | |
1077 | typedef boost::intrusive_ptr<Onode> OnodeRef; | |
1078 | ||
1079 | ||
1080 | /// a cache (shard) of onodes and buffers | |
1081 | struct Cache { | |
1082 | CephContext* cct; | |
1083 | PerfCounters *logger; | |
11fdf7f2 TL |
1084 | |
1085 | /// protect lru and other structures | |
1086 | ceph::recursive_mutex lock = { | |
1087 | ceph::make_recursive_mutex("BlueStore::Cache::lock") }; | |
7c673cae FG |
1088 | |
1089 | std::atomic<uint64_t> num_extents = {0}; | |
1090 | std::atomic<uint64_t> num_blobs = {0}; | |
1091 | ||
7c673cae FG |
1092 | static Cache *create(CephContext* cct, string type, PerfCounters *logger); |
1093 | ||
1094 | Cache(CephContext* cct) : cct(cct), logger(nullptr) {} | |
1095 | virtual ~Cache() {} | |
1096 | ||
1097 | virtual void _add_onode(OnodeRef& o, int level) = 0; | |
1098 | virtual void _rm_onode(OnodeRef& o) = 0; | |
1099 | virtual void _touch_onode(OnodeRef& o) = 0; | |
1100 | ||
1101 | virtual void _add_buffer(Buffer *b, int level, Buffer *near) = 0; | |
1102 | virtual void _rm_buffer(Buffer *b) = 0; | |
1103 | virtual void _move_buffer(Cache *src, Buffer *b) = 0; | |
1104 | virtual void _adjust_buffer_size(Buffer *b, int64_t delta) = 0; | |
1105 | virtual void _touch_buffer(Buffer *b) = 0; | |
1106 | ||
1107 | virtual uint64_t _get_num_onodes() = 0; | |
1108 | virtual uint64_t _get_buffer_bytes() = 0; | |
1109 | ||
1110 | void add_extent() { | |
1111 | ++num_extents; | |
1112 | } | |
1113 | void rm_extent() { | |
1114 | --num_extents; | |
1115 | } | |
1116 | ||
1117 | void add_blob() { | |
1118 | ++num_blobs; | |
1119 | } | |
1120 | void rm_blob() { | |
1121 | --num_blobs; | |
1122 | } | |
1123 | ||
91327a77 | 1124 | void trim(uint64_t onode_max, uint64_t buffer_max); |
7c673cae FG |
1125 | |
1126 | void trim_all(); | |
1127 | ||
1128 | virtual void _trim(uint64_t onode_max, uint64_t buffer_max) = 0; | |
1129 | ||
1130 | virtual void add_stats(uint64_t *onodes, uint64_t *extents, | |
1131 | uint64_t *blobs, | |
1132 | uint64_t *buffers, | |
1133 | uint64_t *bytes) = 0; | |
1134 | ||
31f18b77 | 1135 | bool empty() { |
11fdf7f2 | 1136 | std::lock_guard l(lock); |
31f18b77 FG |
1137 | return _get_num_onodes() == 0 && _get_buffer_bytes() == 0; |
1138 | } | |
1139 | ||
7c673cae FG |
1140 | #ifdef DEBUG_CACHE |
1141 | virtual void _audit(const char *s) = 0; | |
1142 | #else | |
1143 | void _audit(const char *s) { /* no-op */ } | |
1144 | #endif | |
1145 | }; | |
1146 | ||
1147 | /// simple LRU cache for onodes and buffers | |
1148 | struct LRUCache : public Cache { | |
1149 | private: | |
1150 | typedef boost::intrusive::list< | |
1151 | Onode, | |
1152 | boost::intrusive::member_hook< | |
1153 | Onode, | |
1154 | boost::intrusive::list_member_hook<>, | |
1155 | &Onode::lru_item> > onode_lru_list_t; | |
1156 | typedef boost::intrusive::list< | |
1157 | Buffer, | |
1158 | boost::intrusive::member_hook< | |
1159 | Buffer, | |
1160 | boost::intrusive::list_member_hook<>, | |
1161 | &Buffer::lru_item> > buffer_lru_list_t; | |
1162 | ||
1163 | onode_lru_list_t onode_lru; | |
1164 | ||
1165 | buffer_lru_list_t buffer_lru; | |
1166 | uint64_t buffer_size = 0; | |
1167 | ||
1168 | public: | |
1169 | LRUCache(CephContext* cct) : Cache(cct) {} | |
1170 | uint64_t _get_num_onodes() override { | |
1171 | return onode_lru.size(); | |
1172 | } | |
1173 | void _add_onode(OnodeRef& o, int level) override { | |
1174 | if (level > 0) | |
1175 | onode_lru.push_front(*o); | |
1176 | else | |
1177 | onode_lru.push_back(*o); | |
1178 | } | |
1179 | void _rm_onode(OnodeRef& o) override { | |
1180 | auto q = onode_lru.iterator_to(*o); | |
1181 | onode_lru.erase(q); | |
1182 | } | |
1183 | void _touch_onode(OnodeRef& o) override; | |
1184 | ||
1185 | uint64_t _get_buffer_bytes() override { | |
1186 | return buffer_size; | |
1187 | } | |
1188 | void _add_buffer(Buffer *b, int level, Buffer *near) override { | |
1189 | if (near) { | |
1190 | auto q = buffer_lru.iterator_to(*near); | |
1191 | buffer_lru.insert(q, *b); | |
1192 | } else if (level > 0) { | |
1193 | buffer_lru.push_front(*b); | |
1194 | } else { | |
1195 | buffer_lru.push_back(*b); | |
1196 | } | |
1197 | buffer_size += b->length; | |
1198 | } | |
1199 | void _rm_buffer(Buffer *b) override { | |
11fdf7f2 | 1200 | ceph_assert(buffer_size >= b->length); |
7c673cae FG |
1201 | buffer_size -= b->length; |
1202 | auto q = buffer_lru.iterator_to(*b); | |
1203 | buffer_lru.erase(q); | |
1204 | } | |
1205 | void _move_buffer(Cache *src, Buffer *b) override { | |
1206 | src->_rm_buffer(b); | |
1207 | _add_buffer(b, 0, nullptr); | |
1208 | } | |
1209 | void _adjust_buffer_size(Buffer *b, int64_t delta) override { | |
11fdf7f2 | 1210 | ceph_assert((int64_t)buffer_size + delta >= 0); |
7c673cae FG |
1211 | buffer_size += delta; |
1212 | } | |
1213 | void _touch_buffer(Buffer *b) override { | |
1214 | auto p = buffer_lru.iterator_to(*b); | |
1215 | buffer_lru.erase(p); | |
1216 | buffer_lru.push_front(*b); | |
1217 | _audit("_touch_buffer end"); | |
1218 | } | |
1219 | ||
1220 | void _trim(uint64_t onode_max, uint64_t buffer_max) override; | |
1221 | ||
1222 | void add_stats(uint64_t *onodes, uint64_t *extents, | |
1223 | uint64_t *blobs, | |
1224 | uint64_t *buffers, | |
1225 | uint64_t *bytes) override { | |
11fdf7f2 | 1226 | std::lock_guard l(lock); |
7c673cae FG |
1227 | *onodes += onode_lru.size(); |
1228 | *extents += num_extents; | |
1229 | *blobs += num_blobs; | |
1230 | *buffers += buffer_lru.size(); | |
1231 | *bytes += buffer_size; | |
1232 | } | |
1233 | ||
1234 | #ifdef DEBUG_CACHE | |
1235 | void _audit(const char *s) override; | |
1236 | #endif | |
1237 | }; | |
1238 | ||
1239 | // 2Q cache for buffers, LRU for onodes | |
1240 | struct TwoQCache : public Cache { | |
1241 | private: | |
1242 | // stick with LRU for onodes for now (fixme?) | |
1243 | typedef boost::intrusive::list< | |
1244 | Onode, | |
1245 | boost::intrusive::member_hook< | |
1246 | Onode, | |
1247 | boost::intrusive::list_member_hook<>, | |
1248 | &Onode::lru_item> > onode_lru_list_t; | |
1249 | typedef boost::intrusive::list< | |
1250 | Buffer, | |
1251 | boost::intrusive::member_hook< | |
1252 | Buffer, | |
1253 | boost::intrusive::list_member_hook<>, | |
1254 | &Buffer::lru_item> > buffer_list_t; | |
1255 | ||
1256 | onode_lru_list_t onode_lru; | |
1257 | ||
1258 | buffer_list_t buffer_hot; ///< "Am" hot buffers | |
1259 | buffer_list_t buffer_warm_in; ///< "A1in" newly warm buffers | |
1260 | buffer_list_t buffer_warm_out; ///< "A1out" empty buffers we've evicted | |
1261 | uint64_t buffer_bytes = 0; ///< bytes | |
1262 | ||
1263 | enum { | |
1264 | BUFFER_NEW = 0, | |
1265 | BUFFER_WARM_IN, ///< in buffer_warm_in | |
1266 | BUFFER_WARM_OUT, ///< in buffer_warm_out | |
1267 | BUFFER_HOT, ///< in buffer_hot | |
1268 | BUFFER_TYPE_MAX | |
1269 | }; | |
1270 | ||
1271 | uint64_t buffer_list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type | |
1272 | ||
1273 | public: | |
1274 | TwoQCache(CephContext* cct) : Cache(cct) {} | |
1275 | uint64_t _get_num_onodes() override { | |
1276 | return onode_lru.size(); | |
1277 | } | |
1278 | void _add_onode(OnodeRef& o, int level) override { | |
1279 | if (level > 0) | |
1280 | onode_lru.push_front(*o); | |
1281 | else | |
1282 | onode_lru.push_back(*o); | |
1283 | } | |
1284 | void _rm_onode(OnodeRef& o) override { | |
1285 | auto q = onode_lru.iterator_to(*o); | |
1286 | onode_lru.erase(q); | |
1287 | } | |
1288 | void _touch_onode(OnodeRef& o) override; | |
1289 | ||
1290 | uint64_t _get_buffer_bytes() override { | |
1291 | return buffer_bytes; | |
1292 | } | |
1293 | void _add_buffer(Buffer *b, int level, Buffer *near) override; | |
1294 | void _rm_buffer(Buffer *b) override; | |
1295 | void _move_buffer(Cache *src, Buffer *b) override; | |
1296 | void _adjust_buffer_size(Buffer *b, int64_t delta) override; | |
1297 | void _touch_buffer(Buffer *b) override { | |
1298 | switch (b->cache_private) { | |
1299 | case BUFFER_WARM_IN: | |
1300 | // do nothing (somewhat counter-intuitively!) | |
1301 | break; | |
1302 | case BUFFER_WARM_OUT: | |
1303 | // move from warm_out to hot LRU | |
11fdf7f2 | 1304 | ceph_abort_msg("this happens via discard hint"); |
7c673cae FG |
1305 | break; |
1306 | case BUFFER_HOT: | |
1307 | // move to front of hot LRU | |
1308 | buffer_hot.erase(buffer_hot.iterator_to(*b)); | |
1309 | buffer_hot.push_front(*b); | |
1310 | break; | |
1311 | } | |
1312 | _audit("_touch_buffer end"); | |
1313 | } | |
1314 | ||
1315 | void _trim(uint64_t onode_max, uint64_t buffer_max) override; | |
1316 | ||
1317 | void add_stats(uint64_t *onodes, uint64_t *extents, | |
1318 | uint64_t *blobs, | |
1319 | uint64_t *buffers, | |
1320 | uint64_t *bytes) override { | |
11fdf7f2 | 1321 | std::lock_guard l(lock); |
7c673cae FG |
1322 | *onodes += onode_lru.size(); |
1323 | *extents += num_extents; | |
1324 | *blobs += num_blobs; | |
1325 | *buffers += buffer_hot.size() + buffer_warm_in.size(); | |
1326 | *bytes += buffer_bytes; | |
1327 | } | |
1328 | ||
1329 | #ifdef DEBUG_CACHE | |
1330 | void _audit(const char *s) override; | |
1331 | #endif | |
1332 | }; | |
1333 | ||
1334 | struct OnodeSpace { | |
1335 | private: | |
1336 | Cache *cache; | |
1337 | ||
1338 | /// forward lookups | |
31f18b77 | 1339 | mempool::bluestore_cache_other::unordered_map<ghobject_t,OnodeRef> onode_map; |
7c673cae FG |
1340 | |
1341 | friend class Collection; // for split_cache() | |
1342 | ||
1343 | public: | |
1344 | OnodeSpace(Cache *c) : cache(c) {} | |
1345 | ~OnodeSpace() { | |
1346 | clear(); | |
1347 | } | |
1348 | ||
1349 | OnodeRef add(const ghobject_t& oid, OnodeRef o); | |
1350 | OnodeRef lookup(const ghobject_t& o); | |
1351 | void remove(const ghobject_t& oid) { | |
1352 | onode_map.erase(oid); | |
1353 | } | |
1354 | void rename(OnodeRef& o, const ghobject_t& old_oid, | |
1355 | const ghobject_t& new_oid, | |
31f18b77 | 1356 | const mempool::bluestore_cache_other::string& new_okey); |
7c673cae FG |
1357 | void clear(); |
1358 | bool empty(); | |
1359 | ||
11fdf7f2 TL |
1360 | template <int LogLevelV> |
1361 | void dump(CephContext *cct); | |
3efd9988 | 1362 | |
7c673cae FG |
1363 | /// return true if f true for any item |
1364 | bool map_any(std::function<bool(OnodeRef)> f); | |
1365 | }; | |
1366 | ||
11fdf7f2 TL |
1367 | class OpSequencer; |
1368 | typedef boost::intrusive_ptr<OpSequencer> OpSequencerRef; | |
1369 | ||
7c673cae FG |
1370 | struct Collection : public CollectionImpl { |
1371 | BlueStore *store; | |
11fdf7f2 | 1372 | OpSequencerRef osr; |
7c673cae | 1373 | Cache *cache; ///< our cache shard |
7c673cae FG |
1374 | bluestore_cnode_t cnode; |
1375 | RWLock lock; | |
1376 | ||
1377 | bool exists; | |
1378 | ||
1379 | SharedBlobSet shared_blob_set; ///< open SharedBlobs | |
1380 | ||
1381 | // cache onodes on a per-collection basis to avoid lock | |
1382 | // contention. | |
1383 | OnodeSpace onode_map; | |
1384 | ||
1385 | //pool options | |
1386 | pool_opts_t pool_opts; | |
11fdf7f2 | 1387 | ContextQueue *commit_queue; |
7c673cae FG |
1388 | |
1389 | OnodeRef get_onode(const ghobject_t& oid, bool create); | |
1390 | ||
1391 | // the terminology is confusing here, sorry! | |
1392 | // | |
1393 | // blob_t shared_blob_t | |
1394 | // !shared unused -> open | |
1395 | // shared !loaded -> open + shared | |
1396 | // shared loaded -> open + shared + loaded | |
1397 | // | |
1398 | // i.e., | |
1399 | // open = SharedBlob is instantiated | |
1400 | // shared = blob_t shared flag is set; SharedBlob is hashed. | |
1401 | // loaded = SharedBlob::shared_blob_t is loaded from kv store | |
1402 | void open_shared_blob(uint64_t sbid, BlobRef b); | |
1403 | void load_shared_blob(SharedBlobRef sb); | |
1404 | void make_blob_shared(uint64_t sbid, BlobRef b); | |
31f18b77 | 1405 | uint64_t make_blob_unshared(SharedBlob *sb); |
7c673cae FG |
1406 | |
1407 | BlobRef new_blob() { | |
1408 | BlobRef b = new Blob(); | |
1409 | b->shared_blob = new SharedBlob(this); | |
1410 | return b; | |
1411 | } | |
1412 | ||
7c673cae FG |
1413 | bool contains(const ghobject_t& oid) { |
1414 | if (cid.is_meta()) | |
1415 | return oid.hobj.pool == -1; | |
1416 | spg_t spgid; | |
1417 | if (cid.is_pg(&spgid)) | |
1418 | return | |
1419 | spgid.pgid.contains(cnode.bits, oid) && | |
1420 | oid.shard_id == spgid.shard; | |
1421 | return false; | |
1422 | } | |
1423 | ||
1424 | void split_cache(Collection *dest); | |
7c673cae | 1425 | |
11fdf7f2 TL |
1426 | bool flush_commit(Context *c) override; |
1427 | void flush() override; | |
1428 | void flush_all_but_last(); | |
1429 | ||
7c673cae FG |
1430 | Collection(BlueStore *ns, Cache *ca, coll_t c); |
1431 | }; | |
1432 | ||
1433 | class OmapIteratorImpl : public ObjectMap::ObjectMapIteratorImpl { | |
1434 | CollectionRef c; | |
1435 | OnodeRef o; | |
1436 | KeyValueDB::Iterator it; | |
1437 | string head, tail; | |
11fdf7f2 TL |
1438 | |
1439 | string _stringify() const; | |
1440 | ||
7c673cae FG |
1441 | public: |
1442 | OmapIteratorImpl(CollectionRef c, OnodeRef o, KeyValueDB::Iterator it); | |
1443 | int seek_to_first() override; | |
1444 | int upper_bound(const string &after) override; | |
1445 | int lower_bound(const string &to) override; | |
1446 | bool valid() override; | |
11fdf7f2 | 1447 | int next() override; |
7c673cae FG |
1448 | string key() override; |
1449 | bufferlist value() override; | |
1450 | int status() override { | |
1451 | return 0; | |
1452 | } | |
1453 | }; | |
1454 | ||
31f18b77 FG |
1455 | struct volatile_statfs{ |
1456 | enum { | |
1457 | STATFS_ALLOCATED = 0, | |
1458 | STATFS_STORED, | |
1459 | STATFS_COMPRESSED_ORIGINAL, | |
1460 | STATFS_COMPRESSED, | |
1461 | STATFS_COMPRESSED_ALLOCATED, | |
1462 | STATFS_LAST | |
1463 | }; | |
1464 | int64_t values[STATFS_LAST]; | |
1465 | volatile_statfs() { | |
1466 | memset(this, 0, sizeof(volatile_statfs)); | |
1467 | } | |
1468 | void reset() { | |
1469 | *this = volatile_statfs(); | |
1470 | } | |
11fdf7f2 TL |
1471 | void publish(store_statfs_t* buf) const { |
1472 | buf->allocated = allocated(); | |
1473 | buf->data_stored = stored(); | |
1474 | buf->data_compressed = compressed(); | |
1475 | buf->data_compressed_original = compressed_original(); | |
1476 | buf->data_compressed_allocated = compressed_allocated(); | |
1477 | } | |
1478 | ||
31f18b77 FG |
1479 | volatile_statfs& operator+=(const volatile_statfs& other) { |
1480 | for (size_t i = 0; i < STATFS_LAST; ++i) { | |
1481 | values[i] += other.values[i]; | |
1482 | } | |
1483 | return *this; | |
1484 | } | |
1485 | int64_t& allocated() { | |
1486 | return values[STATFS_ALLOCATED]; | |
1487 | } | |
1488 | int64_t& stored() { | |
1489 | return values[STATFS_STORED]; | |
1490 | } | |
1491 | int64_t& compressed_original() { | |
1492 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1493 | } | |
1494 | int64_t& compressed() { | |
1495 | return values[STATFS_COMPRESSED]; | |
1496 | } | |
1497 | int64_t& compressed_allocated() { | |
1498 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1499 | } | |
11fdf7f2 TL |
1500 | int64_t allocated() const { |
1501 | return values[STATFS_ALLOCATED]; | |
1502 | } | |
1503 | int64_t stored() const { | |
1504 | return values[STATFS_STORED]; | |
1505 | } | |
1506 | int64_t compressed_original() const { | |
1507 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1508 | } | |
1509 | int64_t compressed() const { | |
1510 | return values[STATFS_COMPRESSED]; | |
1511 | } | |
1512 | int64_t compressed_allocated() const { | |
1513 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1514 | } | |
1515 | volatile_statfs& operator=(const store_statfs_t& st) { | |
1516 | values[STATFS_ALLOCATED] = st.allocated; | |
1517 | values[STATFS_STORED] = st.data_stored; | |
1518 | values[STATFS_COMPRESSED_ORIGINAL] = st.data_compressed_original; | |
1519 | values[STATFS_COMPRESSED] = st.data_compressed; | |
1520 | values[STATFS_COMPRESSED_ALLOCATED] = st.data_compressed_allocated; | |
1521 | return *this; | |
1522 | } | |
31f18b77 FG |
1523 | bool is_empty() { |
1524 | return values[STATFS_ALLOCATED] == 0 && | |
1525 | values[STATFS_STORED] == 0 && | |
1526 | values[STATFS_COMPRESSED] == 0 && | |
1527 | values[STATFS_COMPRESSED_ORIGINAL] == 0 && | |
1528 | values[STATFS_COMPRESSED_ALLOCATED] == 0; | |
1529 | } | |
11fdf7f2 TL |
1530 | void decode(bufferlist::const_iterator& it) { |
1531 | using ceph::decode; | |
31f18b77 | 1532 | for (size_t i = 0; i < STATFS_LAST; i++) { |
11fdf7f2 | 1533 | decode(values[i], it); |
31f18b77 FG |
1534 | } |
1535 | } | |
1536 | ||
1537 | void encode(bufferlist& bl) { | |
11fdf7f2 | 1538 | using ceph::encode; |
31f18b77 | 1539 | for (size_t i = 0; i < STATFS_LAST; i++) { |
11fdf7f2 | 1540 | encode(values[i], bl); |
31f18b77 FG |
1541 | } |
1542 | } | |
1543 | }; | |
1544 | ||
11fdf7f2 | 1545 | struct TransContext final : public AioContext { |
31f18b77 FG |
1546 | MEMPOOL_CLASS_HELPERS(); |
1547 | ||
7c673cae FG |
1548 | typedef enum { |
1549 | STATE_PREPARE, | |
1550 | STATE_AIO_WAIT, | |
1551 | STATE_IO_DONE, | |
1552 | STATE_KV_QUEUED, // queued for kv_sync_thread submission | |
1553 | STATE_KV_SUBMITTED, // submitted to kv; not yet synced | |
1554 | STATE_KV_DONE, | |
1555 | STATE_DEFERRED_QUEUED, // in deferred_queue (pending or running) | |
1556 | STATE_DEFERRED_CLEANUP, // remove deferred kv record | |
1557 | STATE_DEFERRED_DONE, | |
1558 | STATE_FINISHING, | |
1559 | STATE_DONE, | |
1560 | } state_t; | |
1561 | ||
1562 | state_t state = STATE_PREPARE; | |
1563 | ||
1564 | const char *get_state_name() { | |
1565 | switch (state) { | |
1566 | case STATE_PREPARE: return "prepare"; | |
1567 | case STATE_AIO_WAIT: return "aio_wait"; | |
1568 | case STATE_IO_DONE: return "io_done"; | |
1569 | case STATE_KV_QUEUED: return "kv_queued"; | |
1570 | case STATE_KV_SUBMITTED: return "kv_submitted"; | |
1571 | case STATE_KV_DONE: return "kv_done"; | |
1572 | case STATE_DEFERRED_QUEUED: return "deferred_queued"; | |
1573 | case STATE_DEFERRED_CLEANUP: return "deferred_cleanup"; | |
1574 | case STATE_DEFERRED_DONE: return "deferred_done"; | |
1575 | case STATE_FINISHING: return "finishing"; | |
1576 | case STATE_DONE: return "done"; | |
1577 | } | |
1578 | return "???"; | |
1579 | } | |
1580 | ||
1581 | #if defined(WITH_LTTNG) && defined(WITH_EVENTTRACE) | |
1582 | const char *get_state_latency_name(int state) { | |
1583 | switch (state) { | |
1584 | case l_bluestore_state_prepare_lat: return "prepare"; | |
1585 | case l_bluestore_state_aio_wait_lat: return "aio_wait"; | |
1586 | case l_bluestore_state_io_done_lat: return "io_done"; | |
1587 | case l_bluestore_state_kv_queued_lat: return "kv_queued"; | |
1588 | case l_bluestore_state_kv_committing_lat: return "kv_committing"; | |
1589 | case l_bluestore_state_kv_done_lat: return "kv_done"; | |
1590 | case l_bluestore_state_deferred_queued_lat: return "deferred_queued"; | |
1591 | case l_bluestore_state_deferred_cleanup_lat: return "deferred_cleanup"; | |
1592 | case l_bluestore_state_finishing_lat: return "finishing"; | |
1593 | case l_bluestore_state_done_lat: return "done"; | |
1594 | } | |
1595 | return "???"; | |
1596 | } | |
1597 | #endif | |
1598 | ||
11fdf7f2 | 1599 | utime_t log_state_latency(PerfCounters *logger, int state) { |
7c673cae FG |
1600 | utime_t lat, now = ceph_clock_now(); |
1601 | lat = now - last_stamp; | |
1602 | logger->tinc(state, lat); | |
1603 | #if defined(WITH_LTTNG) && defined(WITH_EVENTTRACE) | |
1604 | if (state >= l_bluestore_state_prepare_lat && state <= l_bluestore_state_done_lat) { | |
1605 | double usecs = (now.to_nsec()-last_stamp.to_nsec())/1000; | |
1606 | OID_ELAPSED("", usecs, get_state_latency_name(state)); | |
1607 | } | |
1608 | #endif | |
1609 | last_stamp = now; | |
11fdf7f2 | 1610 | return lat; |
7c673cae FG |
1611 | } |
1612 | ||
11fdf7f2 TL |
1613 | CollectionRef ch; |
1614 | OpSequencerRef osr; // this should be ch->osr | |
7c673cae FG |
1615 | boost::intrusive::list_member_hook<> sequencer_item; |
1616 | ||
1617 | uint64_t bytes = 0, cost = 0; | |
1618 | ||
1619 | set<OnodeRef> onodes; ///< these need to be updated/written | |
1620 | set<OnodeRef> modified_objects; ///< objects we modified (and need a ref) | |
1621 | set<SharedBlobRef> shared_blobs; ///< these need to be updated/written | |
1622 | set<SharedBlobRef> shared_blobs_written; ///< update these on io completion | |
1623 | ||
1624 | KeyValueDB::Transaction t; ///< then we will commit this | |
7c673cae FG |
1625 | list<Context*> oncommits; ///< more commit completions |
1626 | list<CollectionRef> removed_collections; ///< colls we removed | |
1627 | ||
1628 | boost::intrusive::list_member_hook<> deferred_queue_item; | |
1629 | bluestore_deferred_transaction_t *deferred_txn = nullptr; ///< if any | |
1630 | ||
1631 | interval_set<uint64_t> allocated, released; | |
11fdf7f2 TL |
1632 | volatile_statfs statfs_delta; ///< overall store statistics delta |
1633 | uint64_t osd_pool_id = META_POOL_ID; ///< osd pool id we're operating on | |
1634 | ||
7c673cae FG |
1635 | IOContext ioc; |
1636 | bool had_ios = false; ///< true if we submitted IOs before our kv txn | |
1637 | ||
7c673cae FG |
1638 | uint64_t seq = 0; |
1639 | utime_t start; | |
1640 | utime_t last_stamp; | |
1641 | ||
1642 | uint64_t last_nid = 0; ///< if non-zero, highest new nid we allocated | |
1643 | uint64_t last_blobid = 0; ///< if non-zero, highest new blobid we allocated | |
1644 | ||
11fdf7f2 TL |
1645 | explicit TransContext(CephContext* cct, Collection *c, OpSequencer *o, |
1646 | list<Context*> *on_commits) | |
1647 | : ch(c), | |
1648 | osr(o), | |
7c673cae FG |
1649 | ioc(cct, this), |
1650 | start(ceph_clock_now()) { | |
1651 | last_stamp = start; | |
11fdf7f2 TL |
1652 | if (on_commits) { |
1653 | oncommits.swap(*on_commits); | |
1654 | } | |
7c673cae FG |
1655 | } |
1656 | ~TransContext() { | |
1657 | delete deferred_txn; | |
1658 | } | |
1659 | ||
1660 | void write_onode(OnodeRef &o) { | |
1661 | onodes.insert(o); | |
1662 | } | |
1663 | void write_shared_blob(SharedBlobRef &sb) { | |
1664 | shared_blobs.insert(sb); | |
1665 | } | |
31f18b77 FG |
1666 | void unshare_blob(SharedBlob *sb) { |
1667 | shared_blobs.erase(sb); | |
1668 | } | |
1669 | ||
7c673cae FG |
1670 | /// note we logically modified object (when onode itself is unmodified) |
1671 | void note_modified_object(OnodeRef &o) { | |
1672 | // onode itself isn't written, though | |
1673 | modified_objects.insert(o); | |
1674 | } | |
a8e16298 | 1675 | void note_removed_object(OnodeRef& o) { |
7c673cae | 1676 | onodes.erase(o); |
a8e16298 | 1677 | modified_objects.insert(o); |
7c673cae FG |
1678 | } |
1679 | ||
1680 | void aio_finish(BlueStore *store) override { | |
1681 | store->txc_aio_finish(this); | |
1682 | } | |
1683 | }; | |
1684 | ||
1685 | typedef boost::intrusive::list< | |
1686 | TransContext, | |
1687 | boost::intrusive::member_hook< | |
1688 | TransContext, | |
1689 | boost::intrusive::list_member_hook<>, | |
1690 | &TransContext::deferred_queue_item> > deferred_queue_t; | |
1691 | ||
11fdf7f2 | 1692 | struct DeferredBatch final : public AioContext { |
7c673cae FG |
1693 | OpSequencer *osr; |
1694 | struct deferred_io { | |
1695 | bufferlist bl; ///< data | |
1696 | uint64_t seq; ///< deferred transaction seq | |
1697 | }; | |
1698 | map<uint64_t,deferred_io> iomap; ///< map of ios in this batch | |
1699 | deferred_queue_t txcs; ///< txcs in this batch | |
1700 | IOContext ioc; ///< our aios | |
1701 | /// bytes of pending io for each deferred seq (may be 0) | |
1702 | map<uint64_t,int> seq_bytes; | |
1703 | ||
1704 | void _discard(CephContext *cct, uint64_t offset, uint64_t length); | |
1705 | void _audit(CephContext *cct); | |
1706 | ||
1707 | DeferredBatch(CephContext *cct, OpSequencer *osr) | |
1708 | : osr(osr), ioc(cct, this) {} | |
1709 | ||
1710 | /// prepare a write | |
1711 | void prepare_write(CephContext *cct, | |
1712 | uint64_t seq, uint64_t offset, uint64_t length, | |
1713 | bufferlist::const_iterator& p); | |
1714 | ||
1715 | void aio_finish(BlueStore *store) override { | |
1716 | store->_deferred_aio_finish(osr); | |
1717 | } | |
1718 | }; | |
1719 | ||
11fdf7f2 | 1720 | class OpSequencer : public RefCountedObject { |
7c673cae | 1721 | public: |
11fdf7f2 TL |
1722 | ceph::mutex qlock = ceph::make_mutex("BlueStore::OpSequencer::qlock"); |
1723 | ceph::condition_variable qcond; | |
7c673cae FG |
1724 | typedef boost::intrusive::list< |
1725 | TransContext, | |
1726 | boost::intrusive::member_hook< | |
1727 | TransContext, | |
1728 | boost::intrusive::list_member_hook<>, | |
1729 | &TransContext::sequencer_item> > q_list_t; | |
1730 | q_list_t q; ///< transactions | |
1731 | ||
1732 | boost::intrusive::list_member_hook<> deferred_osr_queue_item; | |
1733 | ||
1734 | DeferredBatch *deferred_running = nullptr; | |
1735 | DeferredBatch *deferred_pending = nullptr; | |
1736 | ||
7c673cae | 1737 | BlueStore *store; |
11fdf7f2 | 1738 | coll_t cid; |
7c673cae FG |
1739 | |
1740 | uint64_t last_seq = 0; | |
1741 | ||
1742 | std::atomic_int txc_with_unstable_io = {0}; ///< num txcs with unstable io | |
1743 | ||
1744 | std::atomic_int kv_committing_serially = {0}; | |
1745 | ||
1746 | std::atomic_int kv_submitted_waiters = {0}; | |
1747 | ||
11fdf7f2 | 1748 | std::atomic_bool zombie = {false}; ///< in zombie_osr set (collection going away) |
7c673cae | 1749 | |
11fdf7f2 TL |
1750 | OpSequencer(BlueStore *store, const coll_t& c) |
1751 | : RefCountedObject(store->cct, 0), | |
1752 | store(store), cid(c) { | |
7c673cae | 1753 | } |
11fdf7f2 TL |
1754 | ~OpSequencer() { |
1755 | ceph_assert(q.empty()); | |
7c673cae FG |
1756 | } |
1757 | ||
1758 | void queue_new(TransContext *txc) { | |
11fdf7f2 | 1759 | std::lock_guard l(qlock); |
7c673cae FG |
1760 | txc->seq = ++last_seq; |
1761 | q.push_back(*txc); | |
1762 | } | |
1763 | ||
1764 | void drain() { | |
11fdf7f2 | 1765 | std::unique_lock l(qlock); |
7c673cae FG |
1766 | while (!q.empty()) |
1767 | qcond.wait(l); | |
1768 | } | |
1769 | ||
1770 | void drain_preceding(TransContext *txc) { | |
11fdf7f2 | 1771 | std::unique_lock l(qlock); |
7c673cae FG |
1772 | while (!q.empty() && &q.front() != txc) |
1773 | qcond.wait(l); | |
1774 | } | |
1775 | ||
1776 | bool _is_all_kv_submitted() { | |
11fdf7f2 TL |
1777 | // caller must hold qlock & q.empty() must not empty |
1778 | ceph_assert(!q.empty()); | |
7c673cae FG |
1779 | TransContext *txc = &q.back(); |
1780 | if (txc->state >= TransContext::STATE_KV_SUBMITTED) { | |
1781 | return true; | |
1782 | } | |
1783 | return false; | |
1784 | } | |
1785 | ||
11fdf7f2 TL |
1786 | void flush() { |
1787 | std::unique_lock l(qlock); | |
1788 | while (true) { | |
1789 | // set flag before the check because the condition | |
1790 | // may become true outside qlock, and we need to make | |
1791 | // sure those threads see waiters and signal qcond. | |
1792 | ++kv_submitted_waiters; | |
1793 | if (q.empty() || _is_all_kv_submitted()) { | |
1794 | --kv_submitted_waiters; | |
1795 | return; | |
1796 | } | |
1797 | qcond.wait(l); | |
1798 | --kv_submitted_waiters; | |
1799 | } | |
1800 | } | |
1801 | ||
1802 | void flush_all_but_last() { | |
1803 | std::unique_lock l(qlock); | |
1804 | assert (q.size() >= 1); | |
7c673cae FG |
1805 | while (true) { |
1806 | // set flag before the check because the condition | |
1807 | // may become true outside qlock, and we need to make | |
1808 | // sure those threads see waiters and signal qcond. | |
1809 | ++kv_submitted_waiters; | |
11fdf7f2 TL |
1810 | if (q.size() <= 1) { |
1811 | --kv_submitted_waiters; | |
7c673cae | 1812 | return; |
11fdf7f2 TL |
1813 | } else { |
1814 | auto it = q.rbegin(); | |
1815 | it++; | |
1816 | if (it->state >= TransContext::STATE_KV_SUBMITTED) { | |
1817 | return; | |
1818 | } | |
7c673cae FG |
1819 | } |
1820 | qcond.wait(l); | |
1821 | --kv_submitted_waiters; | |
1822 | } | |
1823 | } | |
1824 | ||
11fdf7f2 TL |
1825 | bool flush_commit(Context *c) { |
1826 | std::lock_guard l(qlock); | |
7c673cae FG |
1827 | if (q.empty()) { |
1828 | return true; | |
1829 | } | |
1830 | TransContext *txc = &q.back(); | |
1831 | if (txc->state >= TransContext::STATE_KV_DONE) { | |
1832 | return true; | |
1833 | } | |
1834 | txc->oncommits.push_back(c); | |
1835 | return false; | |
1836 | } | |
1837 | }; | |
1838 | ||
1839 | typedef boost::intrusive::list< | |
1840 | OpSequencer, | |
1841 | boost::intrusive::member_hook< | |
1842 | OpSequencer, | |
1843 | boost::intrusive::list_member_hook<>, | |
1844 | &OpSequencer::deferred_osr_queue_item> > deferred_osr_queue_t; | |
1845 | ||
1846 | struct KVSyncThread : public Thread { | |
1847 | BlueStore *store; | |
1848 | explicit KVSyncThread(BlueStore *s) : store(s) {} | |
1849 | void *entry() override { | |
1850 | store->_kv_sync_thread(); | |
1851 | return NULL; | |
1852 | } | |
1853 | }; | |
31f18b77 FG |
1854 | struct KVFinalizeThread : public Thread { |
1855 | BlueStore *store; | |
1856 | explicit KVFinalizeThread(BlueStore *s) : store(s) {} | |
1857 | void *entry() { | |
1858 | store->_kv_finalize_thread(); | |
1859 | return NULL; | |
1860 | } | |
1861 | }; | |
7c673cae FG |
1862 | |
1863 | struct DBHistogram { | |
1864 | struct value_dist { | |
1865 | uint64_t count; | |
1866 | uint32_t max_len; | |
1867 | }; | |
1868 | ||
1869 | struct key_dist { | |
1870 | uint64_t count; | |
1871 | uint32_t max_len; | |
1872 | map<int, struct value_dist> val_map; ///< slab id to count, max length of value and key | |
1873 | }; | |
1874 | ||
1875 | map<string, map<int, struct key_dist> > key_hist; | |
1876 | map<int, uint64_t> value_hist; | |
1877 | int get_key_slab(size_t sz); | |
1878 | string get_key_slab_to_range(int slab); | |
1879 | int get_value_slab(size_t sz); | |
1880 | string get_value_slab_to_range(int slab); | |
1881 | void update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
1882 | const string &prefix, size_t key_size, size_t value_size); | |
1883 | void dump(Formatter *f); | |
1884 | }; | |
1885 | ||
1886 | // -------------------------------------------------------- | |
1887 | // members | |
1888 | private: | |
1889 | BlueFS *bluefs = nullptr; | |
1890 | unsigned bluefs_shared_bdev = 0; ///< which bluefs bdev we are sharing | |
1891 | bool bluefs_single_shared_device = true; | |
11fdf7f2 TL |
1892 | mono_time bluefs_last_balance; |
1893 | utime_t next_dump_on_bluefs_alloc_failure; | |
7c673cae FG |
1894 | |
1895 | KeyValueDB *db = nullptr; | |
1896 | BlockDevice *bdev = nullptr; | |
1897 | std::string freelist_type; | |
1898 | FreelistManager *fm = nullptr; | |
1899 | Allocator *alloc = nullptr; | |
1900 | uuid_d fsid; | |
1901 | int path_fd = -1; ///< open handle to $path | |
1902 | int fsid_fd = -1; ///< open handle (locked) to $path/fsid | |
1903 | bool mounted = false; | |
1904 | ||
1905 | RWLock coll_lock = {"BlueStore::coll_lock"}; ///< rwlock to protect coll_map | |
31f18b77 | 1906 | mempool::bluestore_cache_other::unordered_map<coll_t, CollectionRef> coll_map; |
11fdf7f2 | 1907 | map<coll_t,CollectionRef> new_coll_map; |
7c673cae FG |
1908 | |
1909 | vector<Cache*> cache_shards; | |
1910 | ||
11fdf7f2 TL |
1911 | /// protect zombie_osr_set |
1912 | ceph::mutex zombie_osr_lock = ceph::make_mutex("BlueStore::zombie_osr_lock"); | |
1913 | std::map<coll_t,OpSequencerRef> zombie_osr_set; ///< set of OpSequencers for deleted collections | |
7c673cae FG |
1914 | |
1915 | std::atomic<uint64_t> nid_last = {0}; | |
1916 | std::atomic<uint64_t> nid_max = {0}; | |
1917 | std::atomic<uint64_t> blobid_last = {0}; | |
1918 | std::atomic<uint64_t> blobid_max = {0}; | |
1919 | ||
1920 | Throttle throttle_bytes; ///< submit to commit | |
1921 | Throttle throttle_deferred_bytes; ///< submit to deferred complete | |
1922 | ||
1923 | interval_set<uint64_t> bluefs_extents; ///< block extents owned by bluefs | |
1924 | interval_set<uint64_t> bluefs_extents_reclaiming; ///< currently reclaiming | |
1925 | ||
11fdf7f2 | 1926 | ceph::mutex deferred_lock = ceph::make_mutex("BlueStore::deferred_lock"); |
7c673cae FG |
1927 | std::atomic<uint64_t> deferred_seq = {0}; |
1928 | deferred_osr_queue_t deferred_queue; ///< osr's with deferred io pending | |
1929 | int deferred_queue_size = 0; ///< num txc's queued across all osrs | |
1930 | atomic_int deferred_aggressive = {0}; ///< aggressive wakeup of kv thread | |
11fdf7f2 | 1931 | Finisher deferred_finisher, finisher; |
7c673cae FG |
1932 | |
1933 | KVSyncThread kv_sync_thread; | |
11fdf7f2 TL |
1934 | ceph::mutex kv_lock = ceph::make_mutex("BlueStore::kv_lock"); |
1935 | ceph::condition_variable kv_cond; | |
3efd9988 | 1936 | bool _kv_only = false; |
31f18b77 | 1937 | bool kv_sync_started = false; |
7c673cae | 1938 | bool kv_stop = false; |
31f18b77 FG |
1939 | bool kv_finalize_started = false; |
1940 | bool kv_finalize_stop = false; | |
7c673cae FG |
1941 | deque<TransContext*> kv_queue; ///< ready, already submitted |
1942 | deque<TransContext*> kv_queue_unsubmitted; ///< ready, need submit by kv thread | |
1943 | deque<TransContext*> kv_committing; ///< currently syncing | |
1944 | deque<DeferredBatch*> deferred_done_queue; ///< deferred ios done | |
7c673cae | 1945 | |
31f18b77 | 1946 | KVFinalizeThread kv_finalize_thread; |
11fdf7f2 TL |
1947 | ceph::mutex kv_finalize_lock = ceph::make_mutex("BlueStore::kv_finalize_lock"); |
1948 | ceph::condition_variable kv_finalize_cond; | |
31f18b77 FG |
1949 | deque<TransContext*> kv_committing_to_finalize; ///< pending finalization |
1950 | deque<DeferredBatch*> deferred_stable_to_finalize; ///< pending finalization | |
1951 | ||
7c673cae FG |
1952 | PerfCounters *logger = nullptr; |
1953 | ||
7c673cae FG |
1954 | list<CollectionRef> removed_collections; |
1955 | ||
1956 | RWLock debug_read_error_lock = {"BlueStore::debug_read_error_lock"}; | |
1957 | set<ghobject_t> debug_data_error_objects; | |
1958 | set<ghobject_t> debug_mdata_error_objects; | |
1959 | ||
1960 | std::atomic<int> csum_type = {Checksummer::CSUM_CRC32C}; | |
1961 | ||
1962 | uint64_t block_size = 0; ///< block size of block device (power of 2) | |
1963 | uint64_t block_mask = 0; ///< mask to get just the block offset | |
1964 | size_t block_size_order = 0; ///< bits to shift to get block size | |
1965 | ||
1966 | uint64_t min_alloc_size = 0; ///< minimum allocation unit (power of 2) | |
7c673cae | 1967 | ///< bits for min_alloc_size |
224ce89b | 1968 | uint8_t min_alloc_size_order = 0; |
7c673cae FG |
1969 | static_assert(std::numeric_limits<uint8_t>::max() > |
1970 | std::numeric_limits<decltype(min_alloc_size)>::digits, | |
1971 | "not enough bits for min_alloc_size"); | |
1972 | ||
7c673cae FG |
1973 | ///< maximum allocation unit (power of 2) |
1974 | std::atomic<uint64_t> max_alloc_size = {0}; | |
1975 | ||
224ce89b WB |
1976 | ///< number threshold for forced deferred writes |
1977 | std::atomic<int> deferred_batch_ops = {0}; | |
1978 | ||
1979 | ///< size threshold for forced deferred writes | |
1980 | std::atomic<uint64_t> prefer_deferred_size = {0}; | |
1981 | ||
7c673cae FG |
1982 | ///< approx cost per io, in bytes |
1983 | std::atomic<uint64_t> throttle_cost_per_io = {0}; | |
1984 | ||
224ce89b WB |
1985 | std::atomic<Compressor::CompressionMode> comp_mode = |
1986 | {Compressor::COMP_NONE}; ///< compression mode | |
7c673cae FG |
1987 | CompressorRef compressor; |
1988 | std::atomic<uint64_t> comp_min_blob_size = {0}; | |
1989 | std::atomic<uint64_t> comp_max_blob_size = {0}; | |
1990 | ||
1991 | std::atomic<uint64_t> max_blob_size = {0}; ///< maximum blob size | |
1992 | ||
31f18b77 FG |
1993 | uint64_t kv_ios = 0; |
1994 | uint64_t kv_throttle_costs = 0; | |
1995 | ||
7c673cae | 1996 | // cache trim control |
91327a77 AA |
1997 | uint64_t cache_size = 0; ///< total cache size |
1998 | double cache_meta_ratio = 0; ///< cache ratio dedicated to metadata | |
1999 | double cache_kv_ratio = 0; ///< cache ratio dedicated to kv (e.g., rocksdb) | |
2000 | double cache_data_ratio = 0; ///< cache ratio dedicated to object data | |
2001 | bool cache_autotune = false; ///< cache autotune setting | |
91327a77 AA |
2002 | double cache_autotune_interval = 0; ///< time to wait between cache rebalancing |
2003 | uint64_t osd_memory_target = 0; ///< OSD memory target when autotuning cache | |
2004 | uint64_t osd_memory_base = 0; ///< OSD base memory when autotuning cache | |
2005 | double osd_memory_expected_fragmentation = 0; ///< expected memory fragmentation | |
11fdf7f2 | 2006 | uint64_t osd_memory_cache_min = 0; ///< Min memory to assign when autotuning cache |
91327a77 | 2007 | double osd_memory_cache_resize_interval = 0; ///< Time to wait between cache resizing |
11fdf7f2 TL |
2008 | |
2009 | typedef map<uint64_t, volatile_statfs> osd_pools_map; | |
2010 | ||
2011 | ceph::mutex vstatfs_lock = ceph::make_mutex("BlueStore::vstatfs_lock"); | |
31f18b77 | 2012 | volatile_statfs vstatfs; |
11fdf7f2 TL |
2013 | osd_pools_map osd_pools; // protected by vstatfs_lock as well |
2014 | ||
2015 | bool per_pool_stat_collection = true; | |
7c673cae FG |
2016 | |
2017 | struct MempoolThread : public Thread { | |
91327a77 | 2018 | public: |
7c673cae | 2019 | BlueStore *store; |
91327a77 | 2020 | |
11fdf7f2 TL |
2021 | ceph::condition_variable cond; |
2022 | ceph::mutex lock = ceph::make_mutex("BlueStore::MempoolThread::lock"); | |
7c673cae | 2023 | bool stop = false; |
91327a77 | 2024 | uint64_t autotune_cache_size = 0; |
11fdf7f2 | 2025 | std::shared_ptr<PriorityCache::PriCache> binned_kv_cache = nullptr; |
91327a77 AA |
2026 | |
2027 | struct MempoolCache : public PriorityCache::PriCache { | |
2028 | BlueStore *store; | |
11fdf7f2 TL |
2029 | int64_t cache_bytes[PriorityCache::Priority::LAST+1] = {0}; |
2030 | int64_t committed_bytes = 0; | |
91327a77 AA |
2031 | double cache_ratio = 0; |
2032 | ||
2033 | MempoolCache(BlueStore *s) : store(s) {}; | |
2034 | ||
2035 | virtual uint64_t _get_used_bytes() const = 0; | |
2036 | ||
2037 | virtual int64_t request_cache_bytes( | |
11fdf7f2 | 2038 | PriorityCache::Priority pri, uint64_t total_cache) const { |
91327a77 AA |
2039 | int64_t assigned = get_cache_bytes(pri); |
2040 | ||
2041 | switch (pri) { | |
2042 | // All cache items are currently shoved into the LAST priority | |
2043 | case PriorityCache::Priority::LAST: | |
2044 | { | |
11fdf7f2 | 2045 | int64_t request = _get_used_bytes(); |
91327a77 AA |
2046 | return(request > assigned) ? request - assigned : 0; |
2047 | } | |
2048 | default: | |
2049 | break; | |
2050 | } | |
2051 | return -EOPNOTSUPP; | |
2052 | } | |
2053 | ||
2054 | virtual int64_t get_cache_bytes(PriorityCache::Priority pri) const { | |
2055 | return cache_bytes[pri]; | |
2056 | } | |
2057 | virtual int64_t get_cache_bytes() const { | |
2058 | int64_t total = 0; | |
2059 | ||
2060 | for (int i = 0; i < PriorityCache::Priority::LAST + 1; i++) { | |
2061 | PriorityCache::Priority pri = static_cast<PriorityCache::Priority>(i); | |
2062 | total += get_cache_bytes(pri); | |
2063 | } | |
2064 | return total; | |
2065 | } | |
2066 | virtual void set_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
2067 | cache_bytes[pri] = bytes; | |
2068 | } | |
2069 | virtual void add_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
2070 | cache_bytes[pri] += bytes; | |
2071 | } | |
11fdf7f2 TL |
2072 | virtual int64_t commit_cache_size(uint64_t total_cache) { |
2073 | committed_bytes = PriorityCache::get_chunk( | |
2074 | get_cache_bytes(), total_cache); | |
2075 | return committed_bytes; | |
2076 | } | |
2077 | virtual int64_t get_committed_size() const { | |
2078 | return committed_bytes; | |
91327a77 AA |
2079 | } |
2080 | virtual double get_cache_ratio() const { | |
2081 | return cache_ratio; | |
2082 | } | |
2083 | virtual void set_cache_ratio(double ratio) { | |
2084 | cache_ratio = ratio; | |
2085 | } | |
2086 | virtual string get_cache_name() const = 0; | |
2087 | }; | |
2088 | ||
2089 | struct MetaCache : public MempoolCache { | |
2090 | MetaCache(BlueStore *s) : MempoolCache(s) {}; | |
2091 | ||
2092 | virtual uint64_t _get_used_bytes() const { | |
2093 | return mempool::bluestore_cache_other::allocated_bytes() + | |
2094 | mempool::bluestore_cache_onode::allocated_bytes(); | |
2095 | } | |
2096 | ||
2097 | virtual string get_cache_name() const { | |
2098 | return "BlueStore Meta Cache"; | |
2099 | } | |
2100 | ||
2101 | uint64_t _get_num_onodes() const { | |
2102 | uint64_t onode_num = | |
2103 | mempool::bluestore_cache_onode::allocated_items(); | |
2104 | return (2 > onode_num) ? 2 : onode_num; | |
2105 | } | |
2106 | ||
2107 | double get_bytes_per_onode() const { | |
2108 | return (double)_get_used_bytes() / (double)_get_num_onodes(); | |
2109 | } | |
11fdf7f2 TL |
2110 | }; |
2111 | std::shared_ptr<MetaCache> meta_cache; | |
91327a77 AA |
2112 | |
2113 | struct DataCache : public MempoolCache { | |
2114 | DataCache(BlueStore *s) : MempoolCache(s) {}; | |
2115 | ||
2116 | virtual uint64_t _get_used_bytes() const { | |
2117 | uint64_t bytes = 0; | |
2118 | for (auto i : store->cache_shards) { | |
2119 | bytes += i->_get_buffer_bytes(); | |
2120 | } | |
2121 | return bytes; | |
2122 | } | |
2123 | virtual string get_cache_name() const { | |
2124 | return "BlueStore Data Cache"; | |
2125 | } | |
11fdf7f2 TL |
2126 | }; |
2127 | std::shared_ptr<DataCache> data_cache; | |
91327a77 | 2128 | |
7c673cae FG |
2129 | public: |
2130 | explicit MempoolThread(BlueStore *s) | |
2131 | : store(s), | |
11fdf7f2 TL |
2132 | meta_cache(new MetaCache(s)), |
2133 | data_cache(new DataCache(s)) {} | |
91327a77 | 2134 | |
7c673cae FG |
2135 | void *entry() override; |
2136 | void init() { | |
11fdf7f2 | 2137 | ceph_assert(stop == false); |
7c673cae FG |
2138 | create("bstore_mempool"); |
2139 | } | |
2140 | void shutdown() { | |
11fdf7f2 | 2141 | lock.lock(); |
7c673cae | 2142 | stop = true; |
11fdf7f2 TL |
2143 | cond.notify_all(); |
2144 | lock.unlock(); | |
7c673cae FG |
2145 | join(); |
2146 | } | |
91327a77 AA |
2147 | |
2148 | private: | |
2149 | void _adjust_cache_settings(); | |
2150 | void _trim_shards(bool interval_stats); | |
2151 | void _tune_cache_size(bool interval_stats); | |
11fdf7f2 TL |
2152 | void _balance_cache( |
2153 | const std::list<std::shared_ptr<PriorityCache::PriCache>>& caches); | |
2154 | void _balance_cache_pri( | |
2155 | int64_t *mem_avail, | |
2156 | const std::list<std::shared_ptr<PriorityCache::PriCache>>& caches, | |
2157 | PriorityCache::Priority pri); | |
7c673cae FG |
2158 | } mempool_thread; |
2159 | ||
2160 | // -------------------------------------------------------- | |
2161 | // private methods | |
2162 | ||
2163 | void _init_logger(); | |
2164 | void _shutdown_logger(); | |
2165 | int _reload_logger(); | |
2166 | ||
2167 | int _open_path(); | |
2168 | void _close_path(); | |
2169 | int _open_fsid(bool create); | |
2170 | int _lock_fsid(); | |
2171 | int _read_fsid(uuid_d *f); | |
2172 | int _write_fsid(); | |
2173 | void _close_fsid(); | |
2174 | void _set_alloc_sizes(); | |
2175 | void _set_blob_size(); | |
1adf2230 | 2176 | void _set_finisher_num(); |
7c673cae FG |
2177 | |
2178 | int _open_bdev(bool create); | |
11fdf7f2 TL |
2179 | // Verifies if disk space is enough for reserved + min bluefs |
2180 | // and alters the latter if needed. | |
2181 | // Depends on min_alloc_size hence should be called after | |
2182 | // its initialization (and outside of _open_bdev) | |
2183 | void _validate_bdev(); | |
7c673cae | 2184 | void _close_bdev(); |
11fdf7f2 TL |
2185 | |
2186 | int _minimal_open_bluefs(bool create); | |
2187 | void _minimal_close_bluefs(); | |
2188 | int _open_bluefs(bool create); | |
2189 | void _close_bluefs(); | |
2190 | ||
2191 | // Limited (u)mount intended for BlueFS operations only | |
2192 | int _mount_for_bluefs(); | |
2193 | void _umount_for_bluefs(); | |
2194 | ||
2195 | ||
2196 | int _is_bluefs(bool create, bool* ret); | |
2197 | /* | |
2198 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
2199 | * in the proper order | |
2200 | */ | |
2201 | int _open_db_and_around(bool read_only); | |
2202 | void _close_db_and_around(); | |
2203 | ||
2204 | // updates legacy bluefs related recs in DB to a state valid for | |
2205 | // downgrades from nautilus. | |
2206 | void _sync_bluefs_and_fm(); | |
2207 | ||
2208 | /* | |
2209 | * @warning to_repair_db means that we open this db to repair it, will not | |
2210 | * hold the rocksdb's file lock. | |
2211 | */ | |
2212 | int _open_db(bool create, | |
2213 | bool to_repair_db=false, | |
2214 | bool read_only = false); | |
7c673cae | 2215 | void _close_db(); |
11fdf7f2 | 2216 | int _open_fm(KeyValueDB::Transaction t); |
7c673cae FG |
2217 | void _close_fm(); |
2218 | int _open_alloc(); | |
2219 | void _close_alloc(); | |
2220 | int _open_collections(int *errors=0); | |
2221 | void _close_collections(); | |
2222 | ||
2223 | int _setup_block_symlink_or_file(string name, string path, uint64_t size, | |
2224 | bool create); | |
2225 | ||
7c673cae | 2226 | public: |
3efd9988 FG |
2227 | static int _write_bdev_label(CephContext* cct, |
2228 | string path, bluestore_bdev_label_t label); | |
7c673cae FG |
2229 | static int _read_bdev_label(CephContext* cct, string path, |
2230 | bluestore_bdev_label_t *label); | |
2231 | private: | |
2232 | int _check_or_set_bdev_label(string path, uint64_t size, string desc, | |
2233 | bool create); | |
2234 | ||
2235 | int _open_super_meta(); | |
2236 | ||
224ce89b | 2237 | void _open_statfs(); |
11fdf7f2 | 2238 | void _get_statfs_overall(struct store_statfs_t *buf); |
31f18b77 | 2239 | |
11fdf7f2 TL |
2240 | void _dump_alloc_on_failure(); |
2241 | ||
2242 | int64_t _get_bluefs_size_delta(uint64_t bluefs_free, uint64_t bluefs_total); | |
2243 | int _balance_bluefs_freespace(); | |
7c673cae FG |
2244 | |
2245 | CollectionRef _get_collection(const coll_t& cid); | |
2246 | void _queue_reap_collection(CollectionRef& c); | |
2247 | void _reap_collections(); | |
2248 | void _update_cache_logger(); | |
2249 | ||
2250 | void _assign_nid(TransContext *txc, OnodeRef o); | |
2251 | uint64_t _assign_blobid(TransContext *txc); | |
2252 | ||
81eedcae TL |
2253 | template <int LogLevelV> |
2254 | friend void _dump_onode(CephContext *cct, const Onode& o); | |
2255 | template <int LogLevelV> | |
2256 | friend void _dump_extent_map(CephContext *cct, const ExtentMap& em); | |
2257 | template <int LogLevelV> | |
2258 | friend void _dump_transaction(CephContext *cct, Transaction *t); | |
7c673cae | 2259 | |
11fdf7f2 TL |
2260 | TransContext *_txc_create(Collection *c, OpSequencer *osr, |
2261 | list<Context*> *on_commits); | |
7c673cae FG |
2262 | void _txc_update_store_statfs(TransContext *txc); |
2263 | void _txc_add_transaction(TransContext *txc, Transaction *t); | |
2264 | void _txc_calc_cost(TransContext *txc); | |
2265 | void _txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t); | |
2266 | void _txc_state_proc(TransContext *txc); | |
2267 | void _txc_aio_submit(TransContext *txc); | |
2268 | public: | |
2269 | void txc_aio_finish(void *p) { | |
2270 | _txc_state_proc(static_cast<TransContext*>(p)); | |
2271 | } | |
2272 | private: | |
2273 | void _txc_finish_io(TransContext *txc); | |
2274 | void _txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t); | |
2275 | void _txc_applied_kv(TransContext *txc); | |
2276 | void _txc_committed_kv(TransContext *txc); | |
2277 | void _txc_finish(TransContext *txc); | |
2278 | void _txc_release_alloc(TransContext *txc); | |
2279 | ||
11fdf7f2 TL |
2280 | void _osr_attach(Collection *c); |
2281 | void _osr_register_zombie(OpSequencer *osr); | |
2282 | void _osr_drain(OpSequencer *osr); | |
7c673cae FG |
2283 | void _osr_drain_preceding(TransContext *txc); |
2284 | void _osr_drain_all(); | |
7c673cae | 2285 | |
31f18b77 FG |
2286 | void _kv_start(); |
2287 | void _kv_stop(); | |
7c673cae | 2288 | void _kv_sync_thread(); |
31f18b77 | 2289 | void _kv_finalize_thread(); |
7c673cae FG |
2290 | |
2291 | bluestore_deferred_op_t *_get_deferred_op(TransContext *txc, OnodeRef o); | |
2292 | void _deferred_queue(TransContext *txc); | |
3efd9988 | 2293 | public: |
224ce89b | 2294 | void deferred_try_submit(); |
3efd9988 | 2295 | private: |
224ce89b | 2296 | void _deferred_submit_unlock(OpSequencer *osr); |
7c673cae FG |
2297 | void _deferred_aio_finish(OpSequencer *osr); |
2298 | int _deferred_replay(); | |
2299 | ||
2300 | public: | |
2301 | using mempool_dynamic_bitset = | |
2302 | boost::dynamic_bitset<uint64_t, | |
2303 | mempool::bluestore_fsck::pool_allocator<uint64_t>>; | |
2304 | ||
2305 | private: | |
2306 | int _fsck_check_extents( | |
11fdf7f2 | 2307 | const coll_t& cid, |
7c673cae FG |
2308 | const ghobject_t& oid, |
2309 | const PExtentVector& extents, | |
2310 | bool compressed, | |
2311 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 2312 | uint64_t granularity, |
11fdf7f2 | 2313 | BlueStoreRepairer* repairer, |
7c673cae FG |
2314 | store_statfs_t& expected_statfs); |
2315 | ||
11fdf7f2 TL |
2316 | using per_pool_statfs = |
2317 | mempool::bluestore_fsck::map<uint64_t, store_statfs_t>; | |
2318 | void _fsck_check_pool_statfs( | |
2319 | per_pool_statfs& expected_pool_statfs, | |
2320 | bool need_per_pool_stats, | |
2321 | int& errors, | |
2322 | BlueStoreRepairer* repairer); | |
2323 | ||
7c673cae FG |
2324 | void _buffer_cache_write( |
2325 | TransContext *txc, | |
2326 | BlobRef b, | |
2327 | uint64_t offset, | |
2328 | bufferlist& bl, | |
2329 | unsigned flags) { | |
2330 | b->shared_blob->bc.write(b->shared_blob->get_cache(), txc->seq, offset, bl, | |
2331 | flags); | |
2332 | txc->shared_blobs_written.insert(b->shared_blob); | |
2333 | } | |
2334 | ||
2335 | int _collection_list( | |
2336 | Collection *c, const ghobject_t& start, const ghobject_t& end, | |
2337 | int max, vector<ghobject_t> *ls, ghobject_t *next); | |
2338 | ||
2339 | template <typename T, typename F> | |
2340 | T select_option(const std::string& opt_name, T val1, F f) { | |
2341 | //NB: opt_name reserved for future use | |
2342 | boost::optional<T> val2 = f(); | |
2343 | if (val2) { | |
2344 | return *val2; | |
2345 | } | |
2346 | return val1; | |
2347 | } | |
2348 | ||
2349 | void _apply_padding(uint64_t head_pad, | |
2350 | uint64_t tail_pad, | |
7c673cae FG |
2351 | bufferlist& padded); |
2352 | ||
11fdf7f2 TL |
2353 | void _record_onode(OnodeRef &o, KeyValueDB::Transaction &txn); |
2354 | ||
7c673cae FG |
2355 | // -- ondisk version --- |
2356 | public: | |
2357 | const int32_t latest_ondisk_format = 2; ///< our version | |
2358 | const int32_t min_readable_ondisk_format = 1; ///< what we can read | |
2359 | const int32_t min_compat_ondisk_format = 2; ///< who can read us | |
2360 | ||
2361 | private: | |
2362 | int32_t ondisk_format = 0; ///< value detected on mount | |
2363 | ||
2364 | int _upgrade_super(); ///< upgrade (called during open_super) | |
11fdf7f2 | 2365 | uint64_t _get_ondisk_reserved() const; |
7c673cae FG |
2366 | void _prepare_ondisk_format_super(KeyValueDB::Transaction& t); |
2367 | ||
2368 | // --- public interface --- | |
2369 | public: | |
2370 | BlueStore(CephContext *cct, const string& path); | |
2371 | BlueStore(CephContext *cct, const string& path, uint64_t min_alloc_size); // Ctor for UT only | |
2372 | ~BlueStore() override; | |
2373 | ||
2374 | string get_type() override { | |
2375 | return "bluestore"; | |
2376 | } | |
2377 | ||
2378 | bool needs_journal() override { return false; }; | |
2379 | bool wants_journal() override { return false; }; | |
2380 | bool allows_journal() override { return false; }; | |
2381 | ||
11fdf7f2 TL |
2382 | int get_devices(set<string> *ls) override; |
2383 | ||
31f18b77 | 2384 | bool is_rotational() override; |
d2e6a577 | 2385 | bool is_journal_rotational() override; |
31f18b77 | 2386 | |
224ce89b WB |
2387 | string get_default_device_class() override { |
2388 | string device_class; | |
2389 | map<string, string> metadata; | |
2390 | collect_metadata(&metadata); | |
2391 | auto it = metadata.find("bluestore_bdev_type"); | |
2392 | if (it != metadata.end()) { | |
2393 | device_class = it->second; | |
2394 | } | |
2395 | return device_class; | |
2396 | } | |
2397 | ||
11fdf7f2 TL |
2398 | int get_numa_node( |
2399 | int *numa_node, | |
2400 | set<int> *nodes, | |
2401 | set<string> *failed) override; | |
2402 | ||
7c673cae FG |
2403 | static int get_block_device_fsid(CephContext* cct, const string& path, |
2404 | uuid_d *fsid); | |
2405 | ||
2406 | bool test_mount_in_use() override; | |
2407 | ||
2408 | private: | |
11fdf7f2 | 2409 | int _mount(bool kv_only, bool open_db=true); |
7c673cae FG |
2410 | public: |
2411 | int mount() override { | |
2412 | return _mount(false); | |
2413 | } | |
2414 | int umount() override; | |
2415 | ||
11fdf7f2 TL |
2416 | int start_kv_only(KeyValueDB **pdb, bool open_db=true) { |
2417 | int r = _mount(true, open_db); | |
7c673cae FG |
2418 | if (r < 0) |
2419 | return r; | |
2420 | *pdb = db; | |
2421 | return 0; | |
2422 | } | |
2423 | ||
3efd9988 FG |
2424 | int write_meta(const std::string& key, const std::string& value) override; |
2425 | int read_meta(const std::string& key, std::string *value) override; | |
2426 | ||
2427 | ||
2428 | int fsck(bool deep) override { | |
2429 | return _fsck(deep, false); | |
2430 | } | |
2431 | int repair(bool deep) override { | |
2432 | return _fsck(deep, true); | |
2433 | } | |
2434 | int _fsck(bool deep, bool repair); | |
7c673cae FG |
2435 | |
2436 | void set_cache_shards(unsigned num) override; | |
11fdf7f2 TL |
2437 | void dump_cache_stats(Formatter *f) override { |
2438 | int onode_count = 0, buffers_bytes = 0; | |
2439 | for (auto i: cache_shards) { | |
2440 | onode_count += i->_get_num_onodes(); | |
2441 | buffers_bytes += i->_get_buffer_bytes(); | |
2442 | } | |
2443 | f->dump_int("bluestore_onode", onode_count); | |
2444 | f->dump_int("bluestore_buffers", buffers_bytes); | |
2445 | } | |
2446 | void dump_cache_stats(ostream& ss) override { | |
2447 | int onode_count = 0, buffers_bytes = 0; | |
2448 | for (auto i: cache_shards) { | |
2449 | onode_count += i->_get_num_onodes(); | |
2450 | buffers_bytes += i->_get_buffer_bytes(); | |
2451 | } | |
2452 | ss << "bluestore_onode: " << onode_count; | |
2453 | ss << "bluestore_buffers: " << buffers_bytes; | |
2454 | } | |
7c673cae FG |
2455 | |
2456 | int validate_hobject_key(const hobject_t &obj) const override { | |
2457 | return 0; | |
2458 | } | |
2459 | unsigned get_max_attr_name_length() override { | |
2460 | return 256; // arbitrary; there is no real limit internally | |
2461 | } | |
2462 | ||
2463 | int mkfs() override; | |
2464 | int mkjournal() override { | |
2465 | return 0; | |
2466 | } | |
2467 | ||
2468 | void get_db_statistics(Formatter *f) override; | |
2469 | void generate_db_histogram(Formatter *f) override; | |
31f18b77 | 2470 | void _flush_cache(); |
11fdf7f2 | 2471 | int flush_cache(ostream *os = NULL) override; |
7c673cae FG |
2472 | void dump_perf_counters(Formatter *f) override { |
2473 | f->open_object_section("perf_counters"); | |
2474 | logger->dump_formatted(f, false); | |
2475 | f->close_section(); | |
2476 | } | |
2477 | ||
11fdf7f2 TL |
2478 | int add_new_bluefs_device(int id, const string& path); |
2479 | int migrate_to_existing_bluefs_device(const set<int>& devs_source, | |
2480 | int id); | |
2481 | int migrate_to_new_bluefs_device(const set<int>& devs_source, | |
2482 | int id, | |
2483 | const string& path); | |
2484 | int expand_devices(ostream& out); | |
2485 | string get_device_path(unsigned id); | |
7c673cae FG |
2486 | |
2487 | public: | |
11fdf7f2 TL |
2488 | int statfs(struct store_statfs_t *buf, |
2489 | osd_alert_list_t* alerts = nullptr) override; | |
2490 | int pool_statfs(uint64_t pool_id, struct store_statfs_t *buf) override; | |
7c673cae FG |
2491 | |
2492 | void collect_metadata(map<string,string> *pm) override; | |
2493 | ||
7c673cae FG |
2494 | bool exists(CollectionHandle &c, const ghobject_t& oid) override; |
2495 | int set_collection_opts( | |
11fdf7f2 | 2496 | CollectionHandle& c, |
7c673cae | 2497 | const pool_opts_t& opts) override; |
7c673cae FG |
2498 | int stat( |
2499 | CollectionHandle &c, | |
2500 | const ghobject_t& oid, | |
2501 | struct stat *st, | |
2502 | bool allow_eio = false) override; | |
7c673cae FG |
2503 | int read( |
2504 | CollectionHandle &c, | |
2505 | const ghobject_t& oid, | |
2506 | uint64_t offset, | |
2507 | size_t len, | |
2508 | bufferlist& bl, | |
224ce89b | 2509 | uint32_t op_flags = 0) override; |
7c673cae FG |
2510 | int _do_read( |
2511 | Collection *c, | |
2512 | OnodeRef o, | |
2513 | uint64_t offset, | |
2514 | size_t len, | |
2515 | bufferlist& bl, | |
f64942e4 AA |
2516 | uint32_t op_flags = 0, |
2517 | uint64_t retry_count = 0); | |
7c673cae FG |
2518 | |
2519 | private: | |
2520 | int _fiemap(CollectionHandle &c_, const ghobject_t& oid, | |
2521 | uint64_t offset, size_t len, interval_set<uint64_t>& destset); | |
2522 | public: | |
7c673cae FG |
2523 | int fiemap(CollectionHandle &c, const ghobject_t& oid, |
2524 | uint64_t offset, size_t len, bufferlist& bl) override; | |
7c673cae FG |
2525 | int fiemap(CollectionHandle &c, const ghobject_t& oid, |
2526 | uint64_t offset, size_t len, map<uint64_t, uint64_t>& destmap) override; | |
2527 | ||
2528 | ||
7c673cae FG |
2529 | int getattr(CollectionHandle &c, const ghobject_t& oid, const char *name, |
2530 | bufferptr& value) override; | |
2531 | ||
7c673cae FG |
2532 | int getattrs(CollectionHandle &c, const ghobject_t& oid, |
2533 | map<string,bufferptr>& aset) override; | |
2534 | ||
2535 | int list_collections(vector<coll_t>& ls) override; | |
2536 | ||
2537 | CollectionHandle open_collection(const coll_t &c) override; | |
11fdf7f2 TL |
2538 | CollectionHandle create_new_collection(const coll_t& cid) override; |
2539 | void set_collection_commit_queue(const coll_t& cid, | |
2540 | ContextQueue *commit_queue) override; | |
7c673cae FG |
2541 | |
2542 | bool collection_exists(const coll_t& c) override; | |
11fdf7f2 TL |
2543 | int collection_empty(CollectionHandle& c, bool *empty) override; |
2544 | int collection_bits(CollectionHandle& c) override; | |
7c673cae | 2545 | |
7c673cae FG |
2546 | int collection_list(CollectionHandle &c, |
2547 | const ghobject_t& start, | |
2548 | const ghobject_t& end, | |
2549 | int max, | |
2550 | vector<ghobject_t> *ls, ghobject_t *next) override; | |
2551 | ||
7c673cae FG |
2552 | int omap_get( |
2553 | CollectionHandle &c, ///< [in] Collection containing oid | |
2554 | const ghobject_t &oid, ///< [in] Object containing omap | |
2555 | bufferlist *header, ///< [out] omap header | |
2556 | map<string, bufferlist> *out /// < [out] Key to value map | |
2557 | ) override; | |
2558 | ||
2559 | /// Get omap header | |
7c673cae FG |
2560 | int omap_get_header( |
2561 | CollectionHandle &c, ///< [in] Collection containing oid | |
2562 | const ghobject_t &oid, ///< [in] Object containing omap | |
2563 | bufferlist *header, ///< [out] omap header | |
2564 | bool allow_eio = false ///< [in] don't assert on eio | |
2565 | ) override; | |
2566 | ||
2567 | /// Get keys defined on oid | |
7c673cae FG |
2568 | int omap_get_keys( |
2569 | CollectionHandle &c, ///< [in] Collection containing oid | |
2570 | const ghobject_t &oid, ///< [in] Object containing omap | |
2571 | set<string> *keys ///< [out] Keys defined on oid | |
2572 | ) override; | |
2573 | ||
2574 | /// Get key values | |
7c673cae FG |
2575 | int omap_get_values( |
2576 | CollectionHandle &c, ///< [in] Collection containing oid | |
2577 | const ghobject_t &oid, ///< [in] Object containing omap | |
2578 | const set<string> &keys, ///< [in] Keys to get | |
2579 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
2580 | ) override; | |
2581 | ||
2582 | /// Filters keys into out which are defined on oid | |
7c673cae FG |
2583 | int omap_check_keys( |
2584 | CollectionHandle &c, ///< [in] Collection containing oid | |
2585 | const ghobject_t &oid, ///< [in] Object containing omap | |
2586 | const set<string> &keys, ///< [in] Keys to check | |
2587 | set<string> *out ///< [out] Subset of keys defined on oid | |
2588 | ) override; | |
2589 | ||
7c673cae FG |
2590 | ObjectMap::ObjectMapIterator get_omap_iterator( |
2591 | CollectionHandle &c, ///< [in] collection | |
2592 | const ghobject_t &oid ///< [in] object | |
2593 | ) override; | |
2594 | ||
2595 | void set_fsid(uuid_d u) override { | |
2596 | fsid = u; | |
2597 | } | |
2598 | uuid_d get_fsid() override { | |
2599 | return fsid; | |
2600 | } | |
2601 | ||
2602 | uint64_t estimate_objects_overhead(uint64_t num_objects) override { | |
2603 | return num_objects * 300; //assuming per-object overhead is 300 bytes | |
2604 | } | |
2605 | ||
2606 | struct BSPerfTracker { | |
11fdf7f2 TL |
2607 | PerfCounters::avg_tracker<uint64_t> os_commit_latency_ns; |
2608 | PerfCounters::avg_tracker<uint64_t> os_apply_latency_ns; | |
7c673cae FG |
2609 | |
2610 | objectstore_perf_stat_t get_cur_stats() const { | |
2611 | objectstore_perf_stat_t ret; | |
11fdf7f2 TL |
2612 | ret.os_commit_latency_ns = os_commit_latency_ns.current_avg(); |
2613 | ret.os_apply_latency_ns = os_apply_latency_ns.current_avg(); | |
7c673cae FG |
2614 | return ret; |
2615 | } | |
2616 | ||
2617 | void update_from_perfcounters(PerfCounters &logger); | |
2618 | } perf_tracker; | |
2619 | ||
2620 | objectstore_perf_stat_t get_cur_stats() override { | |
2621 | perf_tracker.update_from_perfcounters(*logger); | |
2622 | return perf_tracker.get_cur_stats(); | |
2623 | } | |
2624 | const PerfCounters* get_perf_counters() const override { | |
2625 | return logger; | |
2626 | } | |
2627 | ||
2628 | int queue_transactions( | |
11fdf7f2 | 2629 | CollectionHandle& ch, |
7c673cae FG |
2630 | vector<Transaction>& tls, |
2631 | TrackedOpRef op = TrackedOpRef(), | |
2632 | ThreadPool::TPHandle *handle = NULL) override; | |
2633 | ||
2634 | // error injection | |
2635 | void inject_data_error(const ghobject_t& o) override { | |
2636 | RWLock::WLocker l(debug_read_error_lock); | |
2637 | debug_data_error_objects.insert(o); | |
2638 | } | |
2639 | void inject_mdata_error(const ghobject_t& o) override { | |
2640 | RWLock::WLocker l(debug_read_error_lock); | |
2641 | debug_mdata_error_objects.insert(o); | |
2642 | } | |
11fdf7f2 TL |
2643 | |
2644 | /// methods to inject various errors fsck can repair | |
2645 | void inject_broken_shared_blob_key(const string& key, | |
2646 | const bufferlist& bl); | |
2647 | void inject_leaked(uint64_t len); | |
2648 | void inject_false_free(coll_t cid, ghobject_t oid); | |
2649 | void inject_statfs(const string& key, const store_statfs_t& new_statfs); | |
2650 | void inject_misreference(coll_t cid1, ghobject_t oid1, | |
2651 | coll_t cid2, ghobject_t oid2, | |
2652 | uint64_t offset); | |
2653 | ||
224ce89b | 2654 | void compact() override { |
11fdf7f2 | 2655 | ceph_assert(db); |
224ce89b WB |
2656 | db->compact(); |
2657 | } | |
28e407b8 AA |
2658 | bool has_builtin_csum() const override { |
2659 | return true; | |
2660 | } | |
2661 | ||
11fdf7f2 TL |
2662 | /* |
2663 | Allocate space for BlueFS from slow device. | |
2664 | Either automatically applies allocated extents to underlying | |
2665 | BlueFS (extents == nullptr) or just return them (non-null extents) provided | |
2666 | */ | |
2667 | int allocate_bluefs_freespace( | |
2668 | uint64_t min_size, | |
2669 | uint64_t size, | |
2670 | PExtentVector* extents); | |
2671 | ||
2672 | void log_latency_fn(int idx, | |
2673 | const ceph::timespan& lat, | |
2674 | std::function<string (const ceph::timespan& lat)> fn); | |
2675 | ||
7c673cae FG |
2676 | private: |
2677 | bool _debug_data_eio(const ghobject_t& o) { | |
2678 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
2679 | return false; | |
2680 | } | |
2681 | RWLock::RLocker l(debug_read_error_lock); | |
2682 | return debug_data_error_objects.count(o); | |
2683 | } | |
2684 | bool _debug_mdata_eio(const ghobject_t& o) { | |
2685 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
2686 | return false; | |
2687 | } | |
2688 | RWLock::RLocker l(debug_read_error_lock); | |
2689 | return debug_mdata_error_objects.count(o); | |
2690 | } | |
2691 | void _debug_obj_on_delete(const ghobject_t& o) { | |
2692 | if (cct->_conf->bluestore_debug_inject_read_err) { | |
2693 | RWLock::WLocker l(debug_read_error_lock); | |
2694 | debug_data_error_objects.erase(o); | |
2695 | debug_mdata_error_objects.erase(o); | |
2696 | } | |
2697 | } | |
11fdf7f2 TL |
2698 | private: |
2699 | ceph::mutex qlock = ceph::make_mutex("BlueStore::Alerts::qlock"); | |
2700 | string failed_cmode; | |
2701 | set<string> failed_compressors; | |
2702 | string spillover_alert; | |
81eedcae TL |
2703 | string legacy_statfs_alert; |
2704 | string disk_size_mismatch_alert; | |
11fdf7f2 TL |
2705 | |
2706 | void _log_alerts(osd_alert_list_t& alerts); | |
2707 | bool _set_compression_alert(bool cmode, const char* s) { | |
2708 | std::lock_guard l(qlock); | |
2709 | if (cmode) { | |
2710 | bool ret = failed_cmode.empty(); | |
2711 | failed_cmode = s; | |
2712 | return ret; | |
2713 | } | |
2714 | return failed_compressors.emplace(s).second; | |
2715 | } | |
2716 | void _clear_compression_alert() { | |
2717 | std::lock_guard l(qlock); | |
2718 | failed_compressors.clear(); | |
2719 | failed_cmode.clear(); | |
2720 | } | |
2721 | ||
2722 | void _set_spillover_alert(const string& s) { | |
2723 | std::lock_guard l(qlock); | |
2724 | spillover_alert = s; | |
2725 | } | |
2726 | void _clear_spillover_alert() { | |
2727 | std::lock_guard l(qlock); | |
2728 | spillover_alert.clear(); | |
2729 | } | |
7c673cae | 2730 | |
81eedcae TL |
2731 | void _check_legacy_statfs_alert(); |
2732 | void _set_disk_size_mismatch_alert(const string& s) { | |
2733 | std::lock_guard l(qlock); | |
2734 | disk_size_mismatch_alert = s; | |
2735 | } | |
2736 | ||
7c673cae FG |
2737 | private: |
2738 | ||
2739 | // -------------------------------------------------------- | |
2740 | // read processing internal methods | |
2741 | int _verify_csum( | |
2742 | OnodeRef& o, | |
2743 | const bluestore_blob_t* blob, | |
2744 | uint64_t blob_xoffset, | |
2745 | const bufferlist& bl, | |
2746 | uint64_t logical_offset) const; | |
2747 | int _decompress(bufferlist& source, bufferlist* result); | |
2748 | ||
2749 | ||
2750 | // -------------------------------------------------------- | |
2751 | // write ops | |
2752 | ||
2753 | struct WriteContext { | |
2754 | bool buffered = false; ///< buffered write | |
2755 | bool compress = false; ///< compressed write | |
2756 | uint64_t target_blob_size = 0; ///< target (max) blob size | |
2757 | unsigned csum_order = 0; ///< target checksum chunk order | |
2758 | ||
2759 | old_extent_map_t old_extents; ///< must deref these blobs | |
2760 | ||
2761 | struct write_item { | |
2762 | uint64_t logical_offset; ///< write logical offset | |
2763 | BlobRef b; | |
2764 | uint64_t blob_length; | |
2765 | uint64_t b_off; | |
2766 | bufferlist bl; | |
2767 | uint64_t b_off0; ///< original offset in a blob prior to padding | |
2768 | uint64_t length0; ///< original data length prior to padding | |
2769 | ||
2770 | bool mark_unused; | |
2771 | bool new_blob; ///< whether new blob was created | |
2772 | ||
3efd9988 FG |
2773 | bool compressed = false; |
2774 | bufferlist compressed_bl; | |
2775 | size_t compressed_len = 0; | |
2776 | ||
7c673cae FG |
2777 | write_item( |
2778 | uint64_t logical_offs, | |
2779 | BlobRef b, | |
2780 | uint64_t blob_len, | |
2781 | uint64_t o, | |
2782 | bufferlist& bl, | |
2783 | uint64_t o0, | |
2784 | uint64_t l0, | |
2785 | bool _mark_unused, | |
2786 | bool _new_blob) | |
2787 | : | |
2788 | logical_offset(logical_offs), | |
2789 | b(b), | |
2790 | blob_length(blob_len), | |
2791 | b_off(o), | |
2792 | bl(bl), | |
2793 | b_off0(o0), | |
2794 | length0(l0), | |
2795 | mark_unused(_mark_unused), | |
2796 | new_blob(_new_blob) {} | |
2797 | }; | |
2798 | vector<write_item> writes; ///< blobs we're writing | |
2799 | ||
2800 | /// partial clone of the context | |
2801 | void fork(const WriteContext& other) { | |
2802 | buffered = other.buffered; | |
2803 | compress = other.compress; | |
2804 | target_blob_size = other.target_blob_size; | |
2805 | csum_order = other.csum_order; | |
2806 | } | |
2807 | void write( | |
2808 | uint64_t loffs, | |
2809 | BlobRef b, | |
2810 | uint64_t blob_len, | |
2811 | uint64_t o, | |
2812 | bufferlist& bl, | |
2813 | uint64_t o0, | |
2814 | uint64_t len0, | |
2815 | bool _mark_unused, | |
2816 | bool _new_blob) { | |
2817 | writes.emplace_back(loffs, | |
2818 | b, | |
2819 | blob_len, | |
2820 | o, | |
2821 | bl, | |
2822 | o0, | |
2823 | len0, | |
2824 | _mark_unused, | |
2825 | _new_blob); | |
2826 | } | |
2827 | /// Checks for writes to the same pextent within a blob | |
2828 | bool has_conflict( | |
2829 | BlobRef b, | |
2830 | uint64_t loffs, | |
2831 | uint64_t loffs_end, | |
2832 | uint64_t min_alloc_size); | |
2833 | }; | |
2834 | ||
2835 | void _do_write_small( | |
2836 | TransContext *txc, | |
2837 | CollectionRef &c, | |
2838 | OnodeRef o, | |
2839 | uint64_t offset, uint64_t length, | |
2840 | bufferlist::iterator& blp, | |
2841 | WriteContext *wctx); | |
2842 | void _do_write_big( | |
2843 | TransContext *txc, | |
2844 | CollectionRef &c, | |
2845 | OnodeRef o, | |
2846 | uint64_t offset, uint64_t length, | |
2847 | bufferlist::iterator& blp, | |
2848 | WriteContext *wctx); | |
2849 | int _do_alloc_write( | |
2850 | TransContext *txc, | |
2851 | CollectionRef c, | |
2852 | OnodeRef o, | |
2853 | WriteContext *wctx); | |
2854 | void _wctx_finish( | |
2855 | TransContext *txc, | |
2856 | CollectionRef& c, | |
2857 | OnodeRef o, | |
31f18b77 FG |
2858 | WriteContext *wctx, |
2859 | set<SharedBlob*> *maybe_unshared_blobs=0); | |
7c673cae | 2860 | |
7c673cae FG |
2861 | int _write(TransContext *txc, |
2862 | CollectionRef& c, | |
2863 | OnodeRef& o, | |
2864 | uint64_t offset, size_t len, | |
2865 | bufferlist& bl, | |
2866 | uint32_t fadvise_flags); | |
2867 | void _pad_zeros(bufferlist *bl, uint64_t *offset, | |
2868 | uint64_t chunk_size); | |
2869 | ||
31f18b77 FG |
2870 | void _choose_write_options(CollectionRef& c, |
2871 | OnodeRef o, | |
2872 | uint32_t fadvise_flags, | |
2873 | WriteContext *wctx); | |
2874 | ||
2875 | int _do_gc(TransContext *txc, | |
2876 | CollectionRef& c, | |
2877 | OnodeRef o, | |
2878 | const GarbageCollector& gc, | |
2879 | const WriteContext& wctx, | |
2880 | uint64_t *dirty_start, | |
2881 | uint64_t *dirty_end); | |
2882 | ||
7c673cae FG |
2883 | int _do_write(TransContext *txc, |
2884 | CollectionRef &c, | |
2885 | OnodeRef o, | |
2886 | uint64_t offset, uint64_t length, | |
2887 | bufferlist& bl, | |
2888 | uint32_t fadvise_flags); | |
2889 | void _do_write_data(TransContext *txc, | |
2890 | CollectionRef& c, | |
2891 | OnodeRef o, | |
2892 | uint64_t offset, | |
2893 | uint64_t length, | |
2894 | bufferlist& bl, | |
2895 | WriteContext *wctx); | |
2896 | ||
2897 | int _touch(TransContext *txc, | |
2898 | CollectionRef& c, | |
2899 | OnodeRef& o); | |
2900 | int _do_zero(TransContext *txc, | |
2901 | CollectionRef& c, | |
2902 | OnodeRef& o, | |
2903 | uint64_t offset, size_t len); | |
2904 | int _zero(TransContext *txc, | |
2905 | CollectionRef& c, | |
2906 | OnodeRef& o, | |
2907 | uint64_t offset, size_t len); | |
2908 | void _do_truncate(TransContext *txc, | |
2909 | CollectionRef& c, | |
2910 | OnodeRef o, | |
31f18b77 FG |
2911 | uint64_t offset, |
2912 | set<SharedBlob*> *maybe_unshared_blobs=0); | |
35e4c445 | 2913 | int _truncate(TransContext *txc, |
7c673cae FG |
2914 | CollectionRef& c, |
2915 | OnodeRef& o, | |
2916 | uint64_t offset); | |
2917 | int _remove(TransContext *txc, | |
2918 | CollectionRef& c, | |
2919 | OnodeRef& o); | |
2920 | int _do_remove(TransContext *txc, | |
2921 | CollectionRef& c, | |
2922 | OnodeRef o); | |
2923 | int _setattr(TransContext *txc, | |
2924 | CollectionRef& c, | |
2925 | OnodeRef& o, | |
2926 | const string& name, | |
2927 | bufferptr& val); | |
2928 | int _setattrs(TransContext *txc, | |
2929 | CollectionRef& c, | |
2930 | OnodeRef& o, | |
2931 | const map<string,bufferptr>& aset); | |
2932 | int _rmattr(TransContext *txc, | |
2933 | CollectionRef& c, | |
2934 | OnodeRef& o, | |
2935 | const string& name); | |
2936 | int _rmattrs(TransContext *txc, | |
2937 | CollectionRef& c, | |
2938 | OnodeRef& o); | |
11fdf7f2 | 2939 | void _do_omap_clear(TransContext *txc, const string& prefix, uint64_t id); |
7c673cae FG |
2940 | int _omap_clear(TransContext *txc, |
2941 | CollectionRef& c, | |
2942 | OnodeRef& o); | |
2943 | int _omap_setkeys(TransContext *txc, | |
2944 | CollectionRef& c, | |
2945 | OnodeRef& o, | |
2946 | bufferlist& bl); | |
2947 | int _omap_setheader(TransContext *txc, | |
2948 | CollectionRef& c, | |
2949 | OnodeRef& o, | |
2950 | bufferlist& header); | |
2951 | int _omap_rmkeys(TransContext *txc, | |
2952 | CollectionRef& c, | |
2953 | OnodeRef& o, | |
2954 | bufferlist& bl); | |
2955 | int _omap_rmkey_range(TransContext *txc, | |
2956 | CollectionRef& c, | |
2957 | OnodeRef& o, | |
2958 | const string& first, const string& last); | |
2959 | int _set_alloc_hint( | |
2960 | TransContext *txc, | |
2961 | CollectionRef& c, | |
2962 | OnodeRef& o, | |
2963 | uint64_t expected_object_size, | |
2964 | uint64_t expected_write_size, | |
2965 | uint32_t flags); | |
2966 | int _do_clone_range(TransContext *txc, | |
2967 | CollectionRef& c, | |
2968 | OnodeRef& oldo, | |
2969 | OnodeRef& newo, | |
2970 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
2971 | int _clone(TransContext *txc, | |
2972 | CollectionRef& c, | |
2973 | OnodeRef& oldo, | |
2974 | OnodeRef& newo); | |
2975 | int _clone_range(TransContext *txc, | |
2976 | CollectionRef& c, | |
2977 | OnodeRef& oldo, | |
2978 | OnodeRef& newo, | |
2979 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
2980 | int _rename(TransContext *txc, | |
2981 | CollectionRef& c, | |
2982 | OnodeRef& oldo, | |
2983 | OnodeRef& newo, | |
2984 | const ghobject_t& new_oid); | |
2985 | int _create_collection(TransContext *txc, const coll_t &cid, | |
2986 | unsigned bits, CollectionRef *c); | |
2987 | int _remove_collection(TransContext *txc, const coll_t &cid, | |
2988 | CollectionRef *c); | |
11fdf7f2 | 2989 | void _do_remove_collection(TransContext *txc, CollectionRef *c); |
7c673cae FG |
2990 | int _split_collection(TransContext *txc, |
2991 | CollectionRef& c, | |
2992 | CollectionRef& d, | |
2993 | unsigned bits, int rem); | |
11fdf7f2 TL |
2994 | int _merge_collection(TransContext *txc, |
2995 | CollectionRef *c, | |
2996 | CollectionRef& d, | |
2997 | unsigned bits); | |
2998 | ||
2999 | private: | |
3000 | std::atomic<uint64_t> out_of_sync_fm = {0}; | |
3001 | // -------------------------------------------------------- | |
3002 | // BlueFSDeviceExpander implementation | |
3003 | uint64_t get_recommended_expansion_delta(uint64_t bluefs_free, | |
3004 | uint64_t bluefs_total) override { | |
3005 | auto delta = _get_bluefs_size_delta(bluefs_free, bluefs_total); | |
3006 | return delta > 0 ? delta : 0; | |
3007 | } | |
3008 | int allocate_freespace( | |
3009 | uint64_t min_size, | |
3010 | uint64_t size, | |
3011 | PExtentVector& extents) override { | |
3012 | return allocate_bluefs_freespace(min_size, size, &extents); | |
3013 | }; | |
7c673cae FG |
3014 | }; |
3015 | ||
11fdf7f2 TL |
3016 | inline ostream& operator<<(ostream& out, const BlueStore::volatile_statfs& s) { |
3017 | return out | |
3018 | << " allocated:" | |
3019 | << s.values[BlueStore::volatile_statfs::STATFS_ALLOCATED] | |
3020 | << " stored:" | |
3021 | << s.values[BlueStore::volatile_statfs::STATFS_STORED] | |
3022 | << " compressed:" | |
3023 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED] | |
3024 | << " compressed_orig:" | |
3025 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED_ORIGINAL] | |
3026 | << " compressed_alloc:" | |
3027 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED_ALLOCATED]; | |
7c673cae FG |
3028 | } |
3029 | ||
3030 | static inline void intrusive_ptr_add_ref(BlueStore::Onode *o) { | |
3031 | o->get(); | |
3032 | } | |
3033 | static inline void intrusive_ptr_release(BlueStore::Onode *o) { | |
3034 | o->put(); | |
3035 | } | |
3036 | ||
3037 | static inline void intrusive_ptr_add_ref(BlueStore::OpSequencer *o) { | |
3038 | o->get(); | |
3039 | } | |
3040 | static inline void intrusive_ptr_release(BlueStore::OpSequencer *o) { | |
3041 | o->put(); | |
3042 | } | |
3043 | ||
11fdf7f2 TL |
3044 | class BlueStoreRepairer |
3045 | { | |
3046 | public: | |
3047 | // to simplify future potential migration to mempools | |
3048 | using fsck_interval = interval_set<uint64_t>; | |
3049 | ||
3050 | // Structure to track what pextents are used for specific cid/oid. | |
3051 | // Similar to Bloom filter positive and false-positive matches are | |
3052 | // possible only. | |
3053 | // Maintains two lists of bloom filters for both cids and oids | |
3054 | // where each list entry is a BF for specific disk pextent | |
3055 | // The length of the extent per filter is measured on init. | |
3056 | // Allows to filter out 'uninteresting' pextents to speadup subsequent | |
3057 | // 'is_used' access. | |
3058 | struct StoreSpaceTracker { | |
3059 | const uint64_t BLOOM_FILTER_SALT_COUNT = 2; | |
3060 | const uint64_t BLOOM_FILTER_TABLE_SIZE = 32; // bytes per single filter | |
3061 | const uint64_t BLOOM_FILTER_EXPECTED_COUNT = 16; // arbitrary selected | |
3062 | static const uint64_t DEF_MEM_CAP = 128 * 1024 * 1024; | |
3063 | ||
3064 | typedef mempool::bluestore_fsck::vector<bloom_filter> bloom_vector; | |
3065 | bloom_vector collections_bfs; | |
3066 | bloom_vector objects_bfs; | |
3067 | ||
3068 | bool was_filtered_out = false; | |
3069 | uint64_t granularity = 0; // extent length for a single filter | |
3070 | ||
3071 | StoreSpaceTracker() { | |
3072 | } | |
3073 | StoreSpaceTracker(const StoreSpaceTracker& from) : | |
3074 | collections_bfs(from.collections_bfs), | |
3075 | objects_bfs(from.objects_bfs), | |
3076 | granularity(from.granularity) { | |
3077 | } | |
3078 | ||
3079 | void init(uint64_t total, | |
3080 | uint64_t min_alloc_size, | |
3081 | uint64_t mem_cap = DEF_MEM_CAP) { | |
3082 | ceph_assert(!granularity); // not initialized yet | |
3083 | ceph_assert(min_alloc_size && isp2(min_alloc_size)); | |
3084 | ceph_assert(mem_cap); | |
3085 | ||
3086 | total = round_up_to(total, min_alloc_size); | |
3087 | granularity = total * BLOOM_FILTER_TABLE_SIZE * 2 / mem_cap; | |
3088 | ||
3089 | if (!granularity) { | |
3090 | granularity = min_alloc_size; | |
3091 | } else { | |
3092 | granularity = round_up_to(granularity, min_alloc_size); | |
3093 | } | |
3094 | ||
3095 | uint64_t entries = round_up_to(total, granularity) / granularity; | |
3096 | collections_bfs.resize(entries, | |
3097 | bloom_filter(BLOOM_FILTER_SALT_COUNT, | |
3098 | BLOOM_FILTER_TABLE_SIZE, | |
3099 | 0, | |
3100 | BLOOM_FILTER_EXPECTED_COUNT)); | |
3101 | objects_bfs.resize(entries, | |
3102 | bloom_filter(BLOOM_FILTER_SALT_COUNT, | |
3103 | BLOOM_FILTER_TABLE_SIZE, | |
3104 | 0, | |
3105 | BLOOM_FILTER_EXPECTED_COUNT)); | |
3106 | } | |
3107 | inline uint32_t get_hash(const coll_t& cid) const { | |
3108 | return cid.hash_to_shard(1); | |
3109 | } | |
3110 | inline void set_used(uint64_t offset, uint64_t len, | |
3111 | const coll_t& cid, const ghobject_t& oid) { | |
3112 | ceph_assert(granularity); // initialized | |
3113 | ||
3114 | // can't call this func after filter_out has been applied | |
3115 | ceph_assert(!was_filtered_out); | |
3116 | if (!len) { | |
3117 | return; | |
3118 | } | |
3119 | auto pos = offset / granularity; | |
3120 | auto end_pos = (offset + len - 1) / granularity; | |
3121 | while (pos <= end_pos) { | |
3122 | collections_bfs[pos].insert(get_hash(cid)); | |
3123 | objects_bfs[pos].insert(oid.hobj.get_hash()); | |
3124 | ++pos; | |
3125 | } | |
3126 | } | |
3127 | // filter-out entries unrelated to the specified(broken) extents. | |
3128 | // 'is_used' calls are permitted after that only | |
3129 | size_t filter_out(const fsck_interval& extents); | |
3130 | ||
3131 | // determines if collection's present after filtering-out | |
3132 | inline bool is_used(const coll_t& cid) const { | |
3133 | ceph_assert(was_filtered_out); | |
3134 | for(auto& bf : collections_bfs) { | |
3135 | if (bf.contains(get_hash(cid))) { | |
3136 | return true; | |
3137 | } | |
3138 | } | |
3139 | return false; | |
3140 | } | |
3141 | // determines if object's present after filtering-out | |
3142 | inline bool is_used(const ghobject_t& oid) const { | |
3143 | ceph_assert(was_filtered_out); | |
3144 | for(auto& bf : objects_bfs) { | |
3145 | if (bf.contains(oid.hobj.get_hash())) { | |
3146 | return true; | |
3147 | } | |
3148 | } | |
3149 | return false; | |
3150 | } | |
3151 | // determines if collection's present before filtering-out | |
3152 | inline bool is_used(const coll_t& cid, uint64_t offs) const { | |
3153 | ceph_assert(granularity); // initialized | |
3154 | ceph_assert(!was_filtered_out); | |
3155 | auto &bf = collections_bfs[offs / granularity]; | |
3156 | if (bf.contains(get_hash(cid))) { | |
3157 | return true; | |
3158 | } | |
3159 | return false; | |
3160 | } | |
3161 | // determines if object's present before filtering-out | |
3162 | inline bool is_used(const ghobject_t& oid, uint64_t offs) const { | |
3163 | ceph_assert(granularity); // initialized | |
3164 | ceph_assert(!was_filtered_out); | |
3165 | auto &bf = objects_bfs[offs / granularity]; | |
3166 | if (bf.contains(oid.hobj.get_hash())) { | |
3167 | return true; | |
3168 | } | |
3169 | return false; | |
3170 | } | |
3171 | }; | |
3172 | public: | |
3173 | ||
3174 | bool remove_key(KeyValueDB *db, const string& prefix, const string& key); | |
3175 | bool fix_shared_blob(KeyValueDB *db, | |
3176 | uint64_t sbid, | |
3177 | const bufferlist* bl); | |
3178 | bool fix_statfs(KeyValueDB *db, const string& key, | |
3179 | const store_statfs_t& new_statfs); | |
3180 | ||
3181 | bool fix_leaked(KeyValueDB *db, | |
3182 | FreelistManager* fm, | |
3183 | uint64_t offset, uint64_t len); | |
3184 | bool fix_false_free(KeyValueDB *db, | |
3185 | FreelistManager* fm, | |
3186 | uint64_t offset, uint64_t len); | |
3187 | bool fix_bluefs_extents(std::atomic<uint64_t>& out_of_sync_flag); | |
3188 | ||
3189 | void init(uint64_t total_space, uint64_t lres_tracking_unit_size); | |
3190 | ||
3191 | bool preprocess_misreference(KeyValueDB *db); | |
3192 | ||
3193 | unsigned apply(KeyValueDB* db); | |
3194 | ||
3195 | void note_misreference(uint64_t offs, uint64_t len, bool inc_error) { | |
3196 | misreferenced_extents.union_insert(offs, len); | |
3197 | if (inc_error) { | |
3198 | ++to_repair_cnt; | |
3199 | } | |
3200 | } | |
3201 | ||
3202 | StoreSpaceTracker& get_space_usage_tracker() { | |
3203 | return space_usage_tracker; | |
3204 | } | |
3205 | const fsck_interval& get_misreferences() const { | |
3206 | return misreferenced_extents; | |
3207 | } | |
3208 | KeyValueDB::Transaction get_fix_misreferences_txn() { | |
3209 | return fix_misreferences_txn; | |
3210 | } | |
3211 | ||
3212 | private: | |
3213 | unsigned to_repair_cnt = 0; | |
3214 | KeyValueDB::Transaction fix_fm_leaked_txn; | |
3215 | KeyValueDB::Transaction fix_fm_false_free_txn; | |
3216 | KeyValueDB::Transaction remove_key_txn; | |
3217 | KeyValueDB::Transaction fix_statfs_txn; | |
3218 | KeyValueDB::Transaction fix_shared_blob_txn; | |
3219 | ||
3220 | KeyValueDB::Transaction fix_misreferences_txn; | |
3221 | ||
3222 | StoreSpaceTracker space_usage_tracker; | |
3223 | ||
3224 | // non-shared extents with multiple references | |
3225 | fsck_interval misreferenced_extents; | |
3226 | ||
3227 | }; | |
7c673cae | 3228 | #endif |