]>
Commit | Line | Data |
---|---|---|
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 <chrono> | |
24 | #include <ratio> | |
25 | #include <mutex> | |
26 | #include <condition_variable> | |
27 | ||
28 | #include <boost/intrusive/list.hpp> | |
29 | #include <boost/intrusive/unordered_set.hpp> | |
30 | #include <boost/intrusive/set.hpp> | |
31 | #include <boost/functional/hash.hpp> | |
32 | #include <boost/dynamic_bitset.hpp> | |
33 | #include <boost/circular_buffer.hpp> | |
34 | ||
35 | #include "include/cpp-btree/btree_set.h" | |
36 | ||
37 | #include "include/ceph_assert.h" | |
38 | #include "include/interval_set.h" | |
39 | #include "include/unordered_map.h" | |
40 | #include "include/mempool.h" | |
41 | #include "include/hash.h" | |
42 | #include "common/bloom_filter.hpp" | |
43 | #include "common/Finisher.h" | |
44 | #include "common/ceph_mutex.h" | |
45 | #include "common/Throttle.h" | |
46 | #include "common/perf_counters.h" | |
47 | #include "common/PriorityCache.h" | |
48 | #include "compressor/Compressor.h" | |
49 | #include "os/ObjectStore.h" | |
50 | ||
51 | #include "bluestore_types.h" | |
52 | #include "BlueFS.h" | |
53 | #include "common/EventTrace.h" | |
54 | ||
55 | #ifdef WITH_BLKIN | |
56 | #include "common/zipkin_trace.h" | |
57 | #endif | |
58 | ||
59 | class Allocator; | |
60 | class FreelistManager; | |
61 | class BlueStoreRepairer; | |
62 | ||
63 | //#define DEBUG_CACHE | |
64 | //#define DEBUG_DEFERRED | |
65 | ||
66 | ||
67 | ||
68 | // constants for Buffer::optimize() | |
69 | #define MAX_BUFFER_SLOP_RATIO_DEN 8 // so actually 1/N | |
70 | ||
71 | ||
72 | enum { | |
73 | l_bluestore_first = 732430, | |
74 | l_bluestore_kv_flush_lat, | |
75 | l_bluestore_kv_commit_lat, | |
76 | l_bluestore_kv_sync_lat, | |
77 | l_bluestore_kv_final_lat, | |
78 | l_bluestore_state_prepare_lat, | |
79 | l_bluestore_state_aio_wait_lat, | |
80 | l_bluestore_state_io_done_lat, | |
81 | l_bluestore_state_kv_queued_lat, | |
82 | l_bluestore_state_kv_committing_lat, | |
83 | l_bluestore_state_kv_done_lat, | |
84 | l_bluestore_state_deferred_queued_lat, | |
85 | l_bluestore_state_deferred_aio_wait_lat, | |
86 | l_bluestore_state_deferred_cleanup_lat, | |
87 | l_bluestore_state_finishing_lat, | |
88 | l_bluestore_state_done_lat, | |
89 | l_bluestore_throttle_lat, | |
90 | l_bluestore_submit_lat, | |
91 | l_bluestore_commit_lat, | |
92 | l_bluestore_read_lat, | |
93 | l_bluestore_read_onode_meta_lat, | |
94 | l_bluestore_read_wait_aio_lat, | |
95 | l_bluestore_compress_lat, | |
96 | l_bluestore_decompress_lat, | |
97 | l_bluestore_csum_lat, | |
98 | l_bluestore_compress_success_count, | |
99 | l_bluestore_compress_rejected_count, | |
100 | l_bluestore_write_pad_bytes, | |
101 | l_bluestore_deferred_write_ops, | |
102 | l_bluestore_deferred_write_bytes, | |
103 | l_bluestore_write_penalty_read_ops, | |
104 | l_bluestore_allocated, | |
105 | l_bluestore_stored, | |
106 | l_bluestore_compressed, | |
107 | l_bluestore_compressed_allocated, | |
108 | l_bluestore_compressed_original, | |
109 | l_bluestore_onodes, | |
110 | l_bluestore_pinned_onodes, | |
111 | l_bluestore_onode_hits, | |
112 | l_bluestore_onode_misses, | |
113 | l_bluestore_onode_shard_hits, | |
114 | l_bluestore_onode_shard_misses, | |
115 | l_bluestore_extents, | |
116 | l_bluestore_blobs, | |
117 | l_bluestore_buffers, | |
118 | l_bluestore_buffer_bytes, | |
119 | l_bluestore_buffer_hit_bytes, | |
120 | l_bluestore_buffer_miss_bytes, | |
121 | l_bluestore_write_big, | |
122 | l_bluestore_write_big_bytes, | |
123 | l_bluestore_write_big_blobs, | |
124 | l_bluestore_write_big_deferred, | |
125 | l_bluestore_write_small, | |
126 | l_bluestore_write_small_bytes, | |
127 | l_bluestore_write_small_unused, | |
128 | l_bluestore_write_deferred, | |
129 | l_bluestore_write_small_pre_read, | |
130 | l_bluestore_write_new, | |
131 | l_bluestore_txc, | |
132 | l_bluestore_onode_reshard, | |
133 | l_bluestore_blob_split, | |
134 | l_bluestore_extent_compress, | |
135 | l_bluestore_gc_merged, | |
136 | l_bluestore_read_eio, | |
137 | l_bluestore_reads_with_retries, | |
138 | l_bluestore_fragmentation, | |
139 | l_bluestore_omap_seek_to_first_lat, | |
140 | l_bluestore_omap_upper_bound_lat, | |
141 | l_bluestore_omap_lower_bound_lat, | |
142 | l_bluestore_omap_next_lat, | |
143 | l_bluestore_omap_get_keys_lat, | |
144 | l_bluestore_omap_get_values_lat, | |
145 | l_bluestore_clist_lat, | |
146 | l_bluestore_remove_lat, | |
147 | l_bluestore_last | |
148 | }; | |
149 | ||
150 | #define META_POOL_ID ((uint64_t)-1ull) | |
151 | ||
152 | class BlueStore : public ObjectStore, | |
153 | public md_config_obs_t { | |
154 | // ----------------------------------------------------- | |
155 | // types | |
156 | public: | |
157 | // config observer | |
158 | const char** get_tracked_conf_keys() const override; | |
159 | void handle_conf_change(const ConfigProxy& conf, | |
160 | const std::set<std::string> &changed) override; | |
161 | ||
162 | //handler for discard event | |
163 | void handle_discard(interval_set<uint64_t>& to_release); | |
164 | ||
165 | void _set_csum(); | |
166 | void _set_compression(); | |
167 | void _set_throttle_params(); | |
168 | int _set_cache_sizes(); | |
169 | void _set_max_defer_interval() { | |
170 | max_defer_interval = | |
171 | cct->_conf.get_val<double>("bluestore_max_defer_interval"); | |
172 | } | |
173 | ||
174 | struct TransContext; | |
175 | ||
176 | typedef std::map<uint64_t, ceph::buffer::list> ready_regions_t; | |
177 | ||
178 | ||
179 | struct BufferSpace; | |
180 | struct Collection; | |
181 | typedef boost::intrusive_ptr<Collection> CollectionRef; | |
182 | ||
183 | struct AioContext { | |
184 | virtual void aio_finish(BlueStore *store) = 0; | |
185 | virtual ~AioContext() {} | |
186 | }; | |
187 | ||
188 | /// cached buffer | |
189 | struct Buffer { | |
190 | MEMPOOL_CLASS_HELPERS(); | |
191 | ||
192 | enum { | |
193 | STATE_EMPTY, ///< empty buffer -- used for cache history | |
194 | STATE_CLEAN, ///< clean data that is up to date | |
195 | STATE_WRITING, ///< data that is being written (io not yet complete) | |
196 | }; | |
197 | static const char *get_state_name(int s) { | |
198 | switch (s) { | |
199 | case STATE_EMPTY: return "empty"; | |
200 | case STATE_CLEAN: return "clean"; | |
201 | case STATE_WRITING: return "writing"; | |
202 | default: return "???"; | |
203 | } | |
204 | } | |
205 | enum { | |
206 | FLAG_NOCACHE = 1, ///< trim when done WRITING (do not become CLEAN) | |
207 | // NOTE: fix operator<< when you define a second flag | |
208 | }; | |
209 | static const char *get_flag_name(int s) { | |
210 | switch (s) { | |
211 | case FLAG_NOCACHE: return "nocache"; | |
212 | default: return "???"; | |
213 | } | |
214 | } | |
215 | ||
216 | BufferSpace *space; | |
217 | uint16_t state; ///< STATE_* | |
218 | uint16_t cache_private = 0; ///< opaque (to us) value used by Cache impl | |
219 | uint32_t flags; ///< FLAG_* | |
220 | uint64_t seq; | |
221 | uint32_t offset, length; | |
222 | ceph::buffer::list data; | |
223 | ||
224 | boost::intrusive::list_member_hook<> lru_item; | |
225 | boost::intrusive::list_member_hook<> state_item; | |
226 | ||
227 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, uint32_t l, | |
228 | unsigned f = 0) | |
229 | : space(space), state(s), flags(f), seq(q), offset(o), length(l) {} | |
230 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, ceph::buffer::list& b, | |
231 | unsigned f = 0) | |
232 | : space(space), state(s), flags(f), seq(q), offset(o), | |
233 | length(b.length()), data(b) {} | |
234 | ||
235 | bool is_empty() const { | |
236 | return state == STATE_EMPTY; | |
237 | } | |
238 | bool is_clean() const { | |
239 | return state == STATE_CLEAN; | |
240 | } | |
241 | bool is_writing() const { | |
242 | return state == STATE_WRITING; | |
243 | } | |
244 | ||
245 | uint32_t end() const { | |
246 | return offset + length; | |
247 | } | |
248 | ||
249 | void truncate(uint32_t newlen) { | |
250 | ceph_assert(newlen < length); | |
251 | if (data.length()) { | |
252 | ceph::buffer::list t; | |
253 | t.substr_of(data, 0, newlen); | |
254 | data = std::move(t); | |
255 | } | |
256 | length = newlen; | |
257 | } | |
258 | void maybe_rebuild() { | |
259 | if (data.length() && | |
260 | (data.get_num_buffers() > 1 || | |
261 | data.front().wasted() > data.length() / MAX_BUFFER_SLOP_RATIO_DEN)) { | |
262 | data.rebuild(); | |
263 | } | |
264 | } | |
265 | ||
266 | void dump(ceph::Formatter *f) const { | |
267 | f->dump_string("state", get_state_name(state)); | |
268 | f->dump_unsigned("seq", seq); | |
269 | f->dump_unsigned("offset", offset); | |
270 | f->dump_unsigned("length", length); | |
271 | f->dump_unsigned("data_length", data.length()); | |
272 | } | |
273 | }; | |
274 | ||
275 | struct BufferCacheShard; | |
276 | ||
277 | /// map logical extent range (object) onto buffers | |
278 | struct BufferSpace { | |
279 | enum { | |
280 | BYPASS_CLEAN_CACHE = 0x1, // bypass clean cache | |
281 | }; | |
282 | ||
283 | typedef boost::intrusive::list< | |
284 | Buffer, | |
285 | boost::intrusive::member_hook< | |
286 | Buffer, | |
287 | boost::intrusive::list_member_hook<>, | |
288 | &Buffer::state_item> > state_list_t; | |
289 | ||
290 | mempool::bluestore_cache_meta::map<uint32_t, std::unique_ptr<Buffer>> | |
291 | buffer_map; | |
292 | ||
293 | // we use a bare intrusive list here instead of std::map because | |
294 | // it uses less memory and we expect this to be very small (very | |
295 | // few IOs in flight to the same Blob at the same time). | |
296 | state_list_t writing; ///< writing buffers, sorted by seq, ascending | |
297 | ||
298 | ~BufferSpace() { | |
299 | ceph_assert(buffer_map.empty()); | |
300 | ceph_assert(writing.empty()); | |
301 | } | |
302 | ||
303 | void _add_buffer(BufferCacheShard* cache, Buffer* b, int level, Buffer* near) { | |
304 | cache->_audit("_add_buffer start"); | |
305 | buffer_map[b->offset].reset(b); | |
306 | if (b->is_writing()) { | |
307 | // we might get already cached data for which resetting mempool is inppropriate | |
308 | // hence calling try_assign_to_mempool | |
309 | b->data.try_assign_to_mempool(mempool::mempool_bluestore_writing); | |
310 | if (writing.empty() || writing.rbegin()->seq <= b->seq) { | |
311 | writing.push_back(*b); | |
312 | } else { | |
313 | auto it = writing.begin(); | |
314 | while (it->seq < b->seq) { | |
315 | ++it; | |
316 | } | |
317 | ||
318 | ceph_assert(it->seq >= b->seq); | |
319 | // note that this will insert b before it | |
320 | // hence the order is maintained | |
321 | writing.insert(it, *b); | |
322 | } | |
323 | } else { | |
324 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); | |
325 | cache->_add(b, level, near); | |
326 | } | |
327 | cache->_audit("_add_buffer end"); | |
328 | } | |
329 | void _rm_buffer(BufferCacheShard* cache, Buffer *b) { | |
330 | _rm_buffer(cache, buffer_map.find(b->offset)); | |
331 | } | |
332 | void _rm_buffer(BufferCacheShard* cache, | |
333 | std::map<uint32_t, std::unique_ptr<Buffer>>::iterator p) { | |
334 | ceph_assert(p != buffer_map.end()); | |
335 | cache->_audit("_rm_buffer start"); | |
336 | if (p->second->is_writing()) { | |
337 | writing.erase(writing.iterator_to(*p->second)); | |
338 | } else { | |
339 | cache->_rm(p->second.get()); | |
340 | } | |
341 | buffer_map.erase(p); | |
342 | cache->_audit("_rm_buffer end"); | |
343 | } | |
344 | ||
345 | std::map<uint32_t,std::unique_ptr<Buffer>>::iterator _data_lower_bound( | |
346 | uint32_t offset) { | |
347 | auto i = buffer_map.lower_bound(offset); | |
348 | if (i != buffer_map.begin()) { | |
349 | --i; | |
350 | if (i->first + i->second->length <= offset) | |
351 | ++i; | |
352 | } | |
353 | return i; | |
354 | } | |
355 | ||
356 | // must be called under protection of the Cache lock | |
357 | void _clear(BufferCacheShard* cache); | |
358 | ||
359 | // return value is the highest cache_private of a trimmed buffer, or 0. | |
360 | int discard(BufferCacheShard* cache, uint32_t offset, uint32_t length) { | |
361 | std::lock_guard l(cache->lock); | |
362 | int ret = _discard(cache, offset, length); | |
363 | cache->_trim(); | |
364 | return ret; | |
365 | } | |
366 | int _discard(BufferCacheShard* cache, uint32_t offset, uint32_t length); | |
367 | ||
368 | void write(BufferCacheShard* cache, uint64_t seq, uint32_t offset, ceph::buffer::list& bl, | |
369 | unsigned flags) { | |
370 | std::lock_guard l(cache->lock); | |
371 | Buffer *b = new Buffer(this, Buffer::STATE_WRITING, seq, offset, bl, | |
372 | flags); | |
373 | b->cache_private = _discard(cache, offset, bl.length()); | |
374 | _add_buffer(cache, b, (flags & Buffer::FLAG_NOCACHE) ? 0 : 1, nullptr); | |
375 | cache->_trim(); | |
376 | } | |
377 | void _finish_write(BufferCacheShard* cache, uint64_t seq); | |
378 | void did_read(BufferCacheShard* cache, uint32_t offset, ceph::buffer::list& bl) { | |
379 | std::lock_guard l(cache->lock); | |
380 | Buffer *b = new Buffer(this, Buffer::STATE_CLEAN, 0, offset, bl); | |
381 | b->cache_private = _discard(cache, offset, bl.length()); | |
382 | _add_buffer(cache, b, 1, nullptr); | |
383 | cache->_trim(); | |
384 | } | |
385 | ||
386 | void read(BufferCacheShard* cache, uint32_t offset, uint32_t length, | |
387 | BlueStore::ready_regions_t& res, | |
388 | interval_set<uint32_t>& res_intervals, | |
389 | int flags = 0); | |
390 | ||
391 | void truncate(BufferCacheShard* cache, uint32_t offset) { | |
392 | discard(cache, offset, (uint32_t)-1 - offset); | |
393 | } | |
394 | ||
395 | void split(BufferCacheShard* cache, size_t pos, BufferSpace &r); | |
396 | ||
397 | void dump(BufferCacheShard* cache, ceph::Formatter *f) const { | |
398 | std::lock_guard l(cache->lock); | |
399 | f->open_array_section("buffers"); | |
400 | for (auto& i : buffer_map) { | |
401 | f->open_object_section("buffer"); | |
402 | ceph_assert(i.first == i.second->offset); | |
403 | i.second->dump(f); | |
404 | f->close_section(); | |
405 | } | |
406 | f->close_section(); | |
407 | } | |
408 | }; | |
409 | ||
410 | struct SharedBlobSet; | |
411 | ||
412 | /// in-memory shared blob state (incl cached buffers) | |
413 | struct SharedBlob { | |
414 | MEMPOOL_CLASS_HELPERS(); | |
415 | ||
416 | std::atomic_int nref = {0}; ///< reference count | |
417 | bool loaded = false; | |
418 | ||
419 | CollectionRef coll; | |
420 | union { | |
421 | uint64_t sbid_unloaded; ///< sbid if persistent isn't loaded | |
422 | bluestore_shared_blob_t *persistent; ///< persistent part of the shared blob if any | |
423 | }; | |
424 | BufferSpace bc; ///< buffer cache | |
425 | ||
426 | SharedBlob(Collection *_coll) : coll(_coll), sbid_unloaded(0) { | |
427 | if (get_cache()) { | |
428 | get_cache()->add_blob(); | |
429 | } | |
430 | } | |
431 | SharedBlob(uint64_t i, Collection *_coll); | |
432 | ~SharedBlob(); | |
433 | ||
434 | uint64_t get_sbid() const { | |
435 | return loaded ? persistent->sbid : sbid_unloaded; | |
436 | } | |
437 | ||
438 | friend void intrusive_ptr_add_ref(SharedBlob *b) { b->get(); } | |
439 | friend void intrusive_ptr_release(SharedBlob *b) { b->put(); } | |
440 | ||
441 | void dump(ceph::Formatter* f) const; | |
442 | friend std::ostream& operator<<(std::ostream& out, const SharedBlob& sb); | |
443 | ||
444 | void get() { | |
445 | ++nref; | |
446 | } | |
447 | void put(); | |
448 | ||
449 | /// get logical references | |
450 | void get_ref(uint64_t offset, uint32_t length); | |
451 | ||
452 | /// put logical references, and get back any released extents | |
453 | void put_ref(uint64_t offset, uint32_t length, | |
454 | PExtentVector *r, bool *unshare); | |
455 | ||
456 | void finish_write(uint64_t seq); | |
457 | ||
458 | friend bool operator==(const SharedBlob &l, const SharedBlob &r) { | |
459 | return l.get_sbid() == r.get_sbid(); | |
460 | } | |
461 | inline BufferCacheShard* get_cache() { | |
462 | return coll ? coll->cache : nullptr; | |
463 | } | |
464 | inline SharedBlobSet* get_parent() { | |
465 | return coll ? &(coll->shared_blob_set) : nullptr; | |
466 | } | |
467 | inline bool is_loaded() const { | |
468 | return loaded; | |
469 | } | |
470 | ||
471 | }; | |
472 | typedef boost::intrusive_ptr<SharedBlob> SharedBlobRef; | |
473 | ||
474 | /// a lookup table of SharedBlobs | |
475 | struct SharedBlobSet { | |
476 | /// protect lookup, insertion, removal | |
477 | ceph::mutex lock = ceph::make_mutex("BlueStore::SharedBlobSet::lock"); | |
478 | ||
479 | // we use a bare pointer because we don't want to affect the ref | |
480 | // count | |
481 | mempool::bluestore_cache_meta::unordered_map<uint64_t,SharedBlob*> sb_map; | |
482 | ||
483 | SharedBlobRef lookup(uint64_t sbid) { | |
484 | std::lock_guard l(lock); | |
485 | auto p = sb_map.find(sbid); | |
486 | if (p == sb_map.end() || | |
487 | p->second->nref == 0) { | |
488 | return nullptr; | |
489 | } | |
490 | return p->second; | |
491 | } | |
492 | ||
493 | void add(Collection* coll, SharedBlob *sb) { | |
494 | std::lock_guard l(lock); | |
495 | sb_map[sb->get_sbid()] = sb; | |
496 | sb->coll = coll; | |
497 | } | |
498 | ||
499 | bool remove(SharedBlob *sb, bool verify_nref_is_zero=false) { | |
500 | std::lock_guard l(lock); | |
501 | ceph_assert(sb->get_parent() == this); | |
502 | if (verify_nref_is_zero && sb->nref != 0) { | |
503 | return false; | |
504 | } | |
505 | // only remove if it still points to us | |
506 | auto p = sb_map.find(sb->get_sbid()); | |
507 | if (p != sb_map.end() && | |
508 | p->second == sb) { | |
509 | sb_map.erase(p); | |
510 | } | |
511 | return true; | |
512 | } | |
513 | ||
514 | bool empty() { | |
515 | std::lock_guard l(lock); | |
516 | return sb_map.empty(); | |
517 | } | |
518 | ||
519 | template <int LogLevelV> | |
520 | void dump(CephContext *cct); | |
521 | }; | |
522 | ||
523 | //#define CACHE_BLOB_BL // not sure if this is a win yet or not... :/ | |
524 | ||
525 | /// in-memory blob metadata and associated cached buffers (if any) | |
526 | struct Blob { | |
527 | MEMPOOL_CLASS_HELPERS(); | |
528 | ||
529 | std::atomic_int nref = {0}; ///< reference count | |
530 | int16_t id = -1; ///< id, for spanning blobs only, >= 0 | |
531 | int16_t last_encoded_id = -1; ///< (ephemeral) used during encoding only | |
532 | SharedBlobRef shared_blob; ///< shared blob state (if any) | |
533 | ||
534 | private: | |
535 | mutable bluestore_blob_t blob; ///< decoded blob metadata | |
536 | #ifdef CACHE_BLOB_BL | |
537 | mutable ceph::buffer::list blob_bl; ///< cached encoded blob, blob is dirty if empty | |
538 | #endif | |
539 | /// refs from this shard. ephemeral if id<0, persisted if spanning. | |
540 | bluestore_blob_use_tracker_t used_in_blob; | |
541 | ||
542 | public: | |
543 | ||
544 | friend void intrusive_ptr_add_ref(Blob *b) { b->get(); } | |
545 | friend void intrusive_ptr_release(Blob *b) { b->put(); } | |
546 | ||
547 | void dump(ceph::Formatter* f) const; | |
548 | friend std::ostream& operator<<(std::ostream& out, const Blob &b); | |
549 | ||
550 | const bluestore_blob_use_tracker_t& get_blob_use_tracker() const { | |
551 | return used_in_blob; | |
552 | } | |
553 | bool is_referenced() const { | |
554 | return used_in_blob.is_not_empty(); | |
555 | } | |
556 | uint32_t get_referenced_bytes() const { | |
557 | return used_in_blob.get_referenced_bytes(); | |
558 | } | |
559 | ||
560 | bool is_spanning() const { | |
561 | return id >= 0; | |
562 | } | |
563 | ||
564 | bool can_split() const { | |
565 | std::lock_guard l(shared_blob->get_cache()->lock); | |
566 | // splitting a BufferSpace writing list is too hard; don't try. | |
567 | return shared_blob->bc.writing.empty() && | |
568 | used_in_blob.can_split() && | |
569 | get_blob().can_split(); | |
570 | } | |
571 | ||
572 | bool can_split_at(uint32_t blob_offset) const { | |
573 | return used_in_blob.can_split_at(blob_offset) && | |
574 | get_blob().can_split_at(blob_offset); | |
575 | } | |
576 | ||
577 | bool can_reuse_blob(uint32_t min_alloc_size, | |
578 | uint32_t target_blob_size, | |
579 | uint32_t b_offset, | |
580 | uint32_t *length0); | |
581 | ||
582 | void dup(Blob& o) { | |
583 | o.shared_blob = shared_blob; | |
584 | o.blob = blob; | |
585 | #ifdef CACHE_BLOB_BL | |
586 | o.blob_bl = blob_bl; | |
587 | #endif | |
588 | } | |
589 | ||
590 | inline const bluestore_blob_t& get_blob() const { | |
591 | return blob; | |
592 | } | |
593 | inline bluestore_blob_t& dirty_blob() { | |
594 | #ifdef CACHE_BLOB_BL | |
595 | blob_bl.clear(); | |
596 | #endif | |
597 | return blob; | |
598 | } | |
599 | ||
600 | /// discard buffers for unallocated regions | |
601 | void discard_unallocated(Collection *coll); | |
602 | ||
603 | /// get logical references | |
604 | void get_ref(Collection *coll, uint32_t offset, uint32_t length); | |
605 | /// put logical references, and get back any released extents | |
606 | bool put_ref(Collection *coll, uint32_t offset, uint32_t length, | |
607 | PExtentVector *r); | |
608 | ||
609 | /// split the blob | |
610 | void split(Collection *coll, uint32_t blob_offset, Blob *o); | |
611 | ||
612 | void get() { | |
613 | ++nref; | |
614 | } | |
615 | void put() { | |
616 | if (--nref == 0) | |
617 | delete this; | |
618 | } | |
619 | ||
620 | ||
621 | #ifdef CACHE_BLOB_BL | |
622 | void _encode() const { | |
623 | if (blob_bl.length() == 0 ) { | |
624 | encode(blob, blob_bl); | |
625 | } else { | |
626 | ceph_assert(blob_bl.length()); | |
627 | } | |
628 | } | |
629 | void bound_encode( | |
630 | size_t& p, | |
631 | bool include_ref_map) const { | |
632 | _encode(); | |
633 | p += blob_bl.length(); | |
634 | if (include_ref_map) { | |
635 | used_in_blob.bound_encode(p); | |
636 | } | |
637 | } | |
638 | void encode( | |
639 | ceph::buffer::list::contiguous_appender& p, | |
640 | bool include_ref_map) const { | |
641 | _encode(); | |
642 | p.append(blob_bl); | |
643 | if (include_ref_map) { | |
644 | used_in_blob.encode(p); | |
645 | } | |
646 | } | |
647 | void decode( | |
648 | Collection */*coll*/, | |
649 | ceph::buffer::ptr::const_iterator& p, | |
650 | bool include_ref_map) { | |
651 | const char *start = p.get_pos(); | |
652 | denc(blob, p); | |
653 | const char *end = p.get_pos(); | |
654 | blob_bl.clear(); | |
655 | blob_bl.append(start, end - start); | |
656 | if (include_ref_map) { | |
657 | used_in_blob.decode(p); | |
658 | } | |
659 | } | |
660 | #else | |
661 | void bound_encode( | |
662 | size_t& p, | |
663 | uint64_t struct_v, | |
664 | uint64_t sbid, | |
665 | bool include_ref_map) const { | |
666 | denc(blob, p, struct_v); | |
667 | if (blob.is_shared()) { | |
668 | denc(sbid, p); | |
669 | } | |
670 | if (include_ref_map) { | |
671 | used_in_blob.bound_encode(p); | |
672 | } | |
673 | } | |
674 | void encode( | |
675 | ceph::buffer::list::contiguous_appender& p, | |
676 | uint64_t struct_v, | |
677 | uint64_t sbid, | |
678 | bool include_ref_map) const { | |
679 | denc(blob, p, struct_v); | |
680 | if (blob.is_shared()) { | |
681 | denc(sbid, p); | |
682 | } | |
683 | if (include_ref_map) { | |
684 | used_in_blob.encode(p); | |
685 | } | |
686 | } | |
687 | void decode( | |
688 | Collection *coll, | |
689 | ceph::buffer::ptr::const_iterator& p, | |
690 | uint64_t struct_v, | |
691 | uint64_t* sbid, | |
692 | bool include_ref_map); | |
693 | #endif | |
694 | }; | |
695 | typedef boost::intrusive_ptr<Blob> BlobRef; | |
696 | typedef mempool::bluestore_cache_meta::map<int,BlobRef> blob_map_t; | |
697 | ||
698 | /// a logical extent, pointing to (some portion of) a blob | |
699 | typedef boost::intrusive::set_base_hook<boost::intrusive::optimize_size<true> > ExtentBase; //making an alias to avoid build warnings | |
700 | struct Extent : public ExtentBase { | |
701 | MEMPOOL_CLASS_HELPERS(); | |
702 | ||
703 | uint32_t logical_offset = 0; ///< logical offset | |
704 | uint32_t blob_offset = 0; ///< blob offset | |
705 | uint32_t length = 0; ///< length | |
706 | BlobRef blob; ///< the blob with our data | |
707 | ||
708 | /// ctor for lookup only | |
709 | explicit Extent(uint32_t lo) : ExtentBase(), logical_offset(lo) { } | |
710 | /// ctor for delayed initialization (see decode_some()) | |
711 | explicit Extent() : ExtentBase() { | |
712 | } | |
713 | /// ctor for general usage | |
714 | Extent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
715 | : ExtentBase(), | |
716 | logical_offset(lo), blob_offset(o), length(l) { | |
717 | assign_blob(b); | |
718 | } | |
719 | ~Extent() { | |
720 | if (blob) { | |
721 | blob->shared_blob->get_cache()->rm_extent(); | |
722 | } | |
723 | } | |
724 | ||
725 | void dump(ceph::Formatter* f) const; | |
726 | ||
727 | void assign_blob(const BlobRef& b) { | |
728 | ceph_assert(!blob); | |
729 | blob = b; | |
730 | blob->shared_blob->get_cache()->add_extent(); | |
731 | } | |
732 | ||
733 | // comparators for intrusive_set | |
734 | friend bool operator<(const Extent &a, const Extent &b) { | |
735 | return a.logical_offset < b.logical_offset; | |
736 | } | |
737 | friend bool operator>(const Extent &a, const Extent &b) { | |
738 | return a.logical_offset > b.logical_offset; | |
739 | } | |
740 | friend bool operator==(const Extent &a, const Extent &b) { | |
741 | return a.logical_offset == b.logical_offset; | |
742 | } | |
743 | ||
744 | uint32_t blob_start() const { | |
745 | return logical_offset - blob_offset; | |
746 | } | |
747 | ||
748 | uint32_t blob_end() const { | |
749 | return blob_start() + blob->get_blob().get_logical_length(); | |
750 | } | |
751 | ||
752 | uint32_t logical_end() const { | |
753 | return logical_offset + length; | |
754 | } | |
755 | ||
756 | // return true if any piece of the blob is out of | |
757 | // the given range [o, o + l]. | |
758 | bool blob_escapes_range(uint32_t o, uint32_t l) const { | |
759 | return blob_start() < o || blob_end() > o + l; | |
760 | } | |
761 | }; | |
762 | typedef boost::intrusive::set<Extent> extent_map_t; | |
763 | ||
764 | ||
765 | friend std::ostream& operator<<(std::ostream& out, const Extent& e); | |
766 | ||
767 | struct OldExtent { | |
768 | boost::intrusive::list_member_hook<> old_extent_item; | |
769 | Extent e; | |
770 | PExtentVector r; | |
771 | bool blob_empty; // flag to track the last removed extent that makes blob | |
772 | // empty - required to update compression stat properly | |
773 | OldExtent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
774 | : e(lo, o, l, b), blob_empty(false) { | |
775 | } | |
776 | static OldExtent* create(CollectionRef c, | |
777 | uint32_t lo, | |
778 | uint32_t o, | |
779 | uint32_t l, | |
780 | BlobRef& b); | |
781 | }; | |
782 | typedef boost::intrusive::list< | |
783 | OldExtent, | |
784 | boost::intrusive::member_hook< | |
785 | OldExtent, | |
786 | boost::intrusive::list_member_hook<>, | |
787 | &OldExtent::old_extent_item> > old_extent_map_t; | |
788 | ||
789 | struct Onode; | |
790 | ||
791 | /// a sharded extent map, mapping offsets to lextents to blobs | |
792 | struct ExtentMap { | |
793 | Onode *onode; | |
794 | extent_map_t extent_map; ///< map of Extents to Blobs | |
795 | blob_map_t spanning_blob_map; ///< blobs that span shards | |
796 | typedef boost::intrusive_ptr<Onode> OnodeRef; | |
797 | ||
798 | struct Shard { | |
799 | bluestore_onode_t::shard_info *shard_info = nullptr; | |
800 | unsigned extents = 0; ///< count extents in this shard | |
801 | bool loaded = false; ///< true if shard is loaded | |
802 | bool dirty = false; ///< true if shard is dirty and needs reencoding | |
803 | }; | |
804 | mempool::bluestore_cache_meta::vector<Shard> shards; ///< shards | |
805 | ||
806 | ceph::buffer::list inline_bl; ///< cached encoded map, if unsharded; empty=>dirty | |
807 | ||
808 | uint32_t needs_reshard_begin = 0; | |
809 | uint32_t needs_reshard_end = 0; | |
810 | ||
811 | void dup(BlueStore* b, TransContext*, CollectionRef&, OnodeRef&, OnodeRef&, | |
812 | uint64_t&, uint64_t&, uint64_t&); | |
813 | ||
814 | bool needs_reshard() const { | |
815 | return needs_reshard_end > needs_reshard_begin; | |
816 | } | |
817 | void clear_needs_reshard() { | |
818 | needs_reshard_begin = needs_reshard_end = 0; | |
819 | } | |
820 | void request_reshard(uint32_t begin, uint32_t end) { | |
821 | if (begin < needs_reshard_begin) { | |
822 | needs_reshard_begin = begin; | |
823 | } | |
824 | if (end > needs_reshard_end) { | |
825 | needs_reshard_end = end; | |
826 | } | |
827 | } | |
828 | ||
829 | struct DeleteDisposer { | |
830 | void operator()(Extent *e) { delete e; } | |
831 | }; | |
832 | ||
833 | ExtentMap(Onode *o); | |
834 | ~ExtentMap() { | |
835 | extent_map.clear_and_dispose(DeleteDisposer()); | |
836 | } | |
837 | ||
838 | void clear() { | |
839 | extent_map.clear_and_dispose(DeleteDisposer()); | |
840 | shards.clear(); | |
841 | inline_bl.clear(); | |
842 | clear_needs_reshard(); | |
843 | } | |
844 | ||
845 | void dump(ceph::Formatter* f) const; | |
846 | ||
847 | bool encode_some(uint32_t offset, uint32_t length, ceph::buffer::list& bl, | |
848 | unsigned *pn); | |
849 | unsigned decode_some(ceph::buffer::list& bl); | |
850 | ||
851 | void bound_encode_spanning_blobs(size_t& p); | |
852 | void encode_spanning_blobs(ceph::buffer::list::contiguous_appender& p); | |
853 | void decode_spanning_blobs(ceph::buffer::ptr::const_iterator& p); | |
854 | ||
855 | BlobRef get_spanning_blob(int id) { | |
856 | auto p = spanning_blob_map.find(id); | |
857 | ceph_assert(p != spanning_blob_map.end()); | |
858 | return p->second; | |
859 | } | |
860 | ||
861 | void update(KeyValueDB::Transaction t, bool force); | |
862 | decltype(BlueStore::Blob::id) allocate_spanning_blob_id(); | |
863 | void reshard( | |
864 | KeyValueDB *db, | |
865 | KeyValueDB::Transaction t); | |
866 | ||
867 | /// initialize Shards from the onode | |
868 | void init_shards(bool loaded, bool dirty); | |
869 | ||
870 | /// return index of shard containing offset | |
871 | /// or -1 if not found | |
872 | int seek_shard(uint32_t offset) { | |
873 | size_t end = shards.size(); | |
874 | size_t mid, left = 0; | |
875 | size_t right = end; // one passed the right end | |
876 | ||
877 | while (left < right) { | |
878 | mid = left + (right - left) / 2; | |
879 | if (offset >= shards[mid].shard_info->offset) { | |
880 | size_t next = mid + 1; | |
881 | if (next >= end || offset < shards[next].shard_info->offset) | |
882 | return mid; | |
883 | //continue to search forwards | |
884 | left = next; | |
885 | } else { | |
886 | //continue to search backwards | |
887 | right = mid; | |
888 | } | |
889 | } | |
890 | ||
891 | return -1; // not found | |
892 | } | |
893 | ||
894 | /// check if a range spans a shard | |
895 | bool spans_shard(uint32_t offset, uint32_t length) { | |
896 | if (shards.empty()) { | |
897 | return false; | |
898 | } | |
899 | int s = seek_shard(offset); | |
900 | ceph_assert(s >= 0); | |
901 | if (s == (int)shards.size() - 1) { | |
902 | return false; // last shard | |
903 | } | |
904 | if (offset + length <= shards[s+1].shard_info->offset) { | |
905 | return false; | |
906 | } | |
907 | return true; | |
908 | } | |
909 | ||
910 | /// ensure that a range of the map is loaded | |
911 | void fault_range(KeyValueDB *db, | |
912 | uint32_t offset, uint32_t length); | |
913 | ||
914 | /// ensure a range of the map is marked dirty | |
915 | void dirty_range(uint32_t offset, uint32_t length); | |
916 | ||
917 | /// for seek_lextent test | |
918 | extent_map_t::iterator find(uint64_t offset); | |
919 | ||
920 | /// seek to the first lextent including or after offset | |
921 | extent_map_t::iterator seek_lextent(uint64_t offset); | |
922 | extent_map_t::const_iterator seek_lextent(uint64_t offset) const; | |
923 | ||
924 | /// add a new Extent | |
925 | void add(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) { | |
926 | extent_map.insert(*new Extent(lo, o, l, b)); | |
927 | } | |
928 | ||
929 | /// remove (and delete) an Extent | |
930 | void rm(extent_map_t::iterator p) { | |
931 | extent_map.erase_and_dispose(p, DeleteDisposer()); | |
932 | } | |
933 | ||
934 | bool has_any_lextents(uint64_t offset, uint64_t length); | |
935 | ||
936 | /// consolidate adjacent lextents in extent_map | |
937 | int compress_extent_map(uint64_t offset, uint64_t length); | |
938 | ||
939 | /// punch a logical hole. add lextents to deref to target list. | |
940 | void punch_hole(CollectionRef &c, | |
941 | uint64_t offset, uint64_t length, | |
942 | old_extent_map_t *old_extents); | |
943 | ||
944 | /// put new lextent into lextent_map overwriting existing ones if | |
945 | /// any and update references accordingly | |
946 | Extent *set_lextent(CollectionRef &c, | |
947 | uint64_t logical_offset, | |
948 | uint64_t offset, uint64_t length, | |
949 | BlobRef b, | |
950 | old_extent_map_t *old_extents); | |
951 | ||
952 | /// split a blob (and referring extents) | |
953 | BlobRef split_blob(BlobRef lb, uint32_t blob_offset, uint32_t pos); | |
954 | }; | |
955 | ||
956 | /// Compressed Blob Garbage collector | |
957 | /* | |
958 | The primary idea of the collector is to estimate a difference between | |
959 | allocation units(AU) currently present for compressed blobs and new AUs | |
960 | required to store that data uncompressed. | |
961 | Estimation is performed for protrusive extents within a logical range | |
962 | determined by a concatenation of old_extents collection and specific(current) | |
963 | write request. | |
964 | The root cause for old_extents use is the need to handle blob ref counts | |
965 | properly. Old extents still hold blob refs and hence we need to traverse | |
966 | the collection to determine if blob to be released. | |
967 | Protrusive extents are extents that fit into the blob std::set in action | |
968 | (ones that are below the logical range from above) but not removed totally | |
969 | due to the current write. | |
970 | E.g. for | |
971 | extent1 <loffs = 100, boffs = 100, len = 100> -> | |
972 | blob1<compressed, len_on_disk=4096, logical_len=8192> | |
973 | extent2 <loffs = 200, boffs = 200, len = 100> -> | |
974 | blob2<raw, len_on_disk=4096, llen=4096> | |
975 | extent3 <loffs = 300, boffs = 300, len = 100> -> | |
976 | blob1<compressed, len_on_disk=4096, llen=8192> | |
977 | extent4 <loffs = 4096, boffs = 0, len = 100> -> | |
978 | blob3<raw, len_on_disk=4096, llen=4096> | |
979 | write(300~100) | |
980 | protrusive extents are within the following ranges <0~300, 400~8192-400> | |
981 | In this case existing AUs that might be removed due to GC (i.e. blob1) | |
982 | use 2x4K bytes. | |
983 | And new AUs expected after GC = 0 since extent1 to be merged into blob2. | |
984 | Hence we should do a collect. | |
985 | */ | |
986 | class GarbageCollector | |
987 | { | |
988 | public: | |
989 | /// return amount of allocation units that might be saved due to GC | |
990 | int64_t estimate( | |
991 | uint64_t offset, | |
992 | uint64_t length, | |
993 | const ExtentMap& extent_map, | |
994 | const old_extent_map_t& old_extents, | |
995 | uint64_t min_alloc_size); | |
996 | ||
997 | /// return a collection of extents to perform GC on | |
998 | const interval_set<uint64_t>& get_extents_to_collect() const { | |
999 | return extents_to_collect; | |
1000 | } | |
1001 | GarbageCollector(CephContext* _cct) : cct(_cct) {} | |
1002 | ||
1003 | private: | |
1004 | struct BlobInfo { | |
1005 | uint64_t referenced_bytes = 0; ///< amount of bytes referenced in blob | |
1006 | int64_t expected_allocations = 0; ///< new alloc units required | |
1007 | ///< in case of gc fulfilled | |
1008 | bool collect_candidate = false; ///< indicate if blob has any extents | |
1009 | ///< eligible for GC. | |
1010 | extent_map_t::const_iterator first_lextent; ///< points to the first | |
1011 | ///< lextent referring to | |
1012 | ///< the blob if any. | |
1013 | ///< collect_candidate flag | |
1014 | ///< determines the validity | |
1015 | extent_map_t::const_iterator last_lextent; ///< points to the last | |
1016 | ///< lextent referring to | |
1017 | ///< the blob if any. | |
1018 | ||
1019 | BlobInfo(uint64_t ref_bytes) : | |
1020 | referenced_bytes(ref_bytes) { | |
1021 | } | |
1022 | }; | |
1023 | CephContext* cct; | |
1024 | std::map<Blob*, BlobInfo> affected_blobs; ///< compressed blobs and their ref_map | |
1025 | ///< copies that are affected by the | |
1026 | ///< specific write | |
1027 | ||
1028 | ///< protrusive extents that should be collected if GC takes place | |
1029 | interval_set<uint64_t> extents_to_collect; | |
1030 | ||
1031 | boost::optional<uint64_t > used_alloc_unit; ///< last processed allocation | |
1032 | ///< unit when traversing | |
1033 | ///< protrusive extents. | |
1034 | ///< Other extents mapped to | |
1035 | ///< this AU to be ignored | |
1036 | ///< (except the case where | |
1037 | ///< uncompressed extent follows | |
1038 | ///< compressed one - see below). | |
1039 | BlobInfo* blob_info_counted = nullptr; ///< std::set if previous allocation unit | |
1040 | ///< caused expected_allocations | |
1041 | ///< counter increment at this blob. | |
1042 | ///< if uncompressed extent follows | |
1043 | ///< a decrement for the | |
1044 | ///< expected_allocations counter | |
1045 | ///< is needed | |
1046 | int64_t expected_allocations = 0; ///< new alloc units required in case | |
1047 | ///< of gc fulfilled | |
1048 | int64_t expected_for_release = 0; ///< alloc units currently used by | |
1049 | ///< compressed blobs that might | |
1050 | ///< gone after GC | |
1051 | ||
1052 | protected: | |
1053 | void process_protrusive_extents(const BlueStore::ExtentMap& extent_map, | |
1054 | uint64_t start_offset, | |
1055 | uint64_t end_offset, | |
1056 | uint64_t start_touch_offset, | |
1057 | uint64_t end_touch_offset, | |
1058 | uint64_t min_alloc_size); | |
1059 | }; | |
1060 | ||
1061 | struct OnodeSpace; | |
1062 | /// an in-memory object | |
1063 | struct Onode { | |
1064 | MEMPOOL_CLASS_HELPERS(); | |
1065 | ||
1066 | std::atomic_int nref; ///< reference count | |
1067 | Collection *c; | |
1068 | ghobject_t oid; | |
1069 | ||
1070 | /// key under PREFIX_OBJ where we are stored | |
1071 | mempool::bluestore_cache_meta::string key; | |
1072 | ||
1073 | boost::intrusive::list_member_hook<> lru_item; | |
1074 | ||
1075 | bluestore_onode_t onode; ///< metadata stored as value in kv store | |
1076 | bool exists; ///< true if object logically exists | |
1077 | bool cached; ///< Onode is logically in the cache | |
1078 | /// (it can be pinned and hence physically out | |
1079 | /// of it at the moment though) | |
1080 | std::atomic_bool pinned; ///< Onode is pinned | |
1081 | /// (or should be pinned when cached) | |
1082 | ExtentMap extent_map; | |
1083 | ||
1084 | // track txc's that have not been committed to kv store (and whose | |
1085 | // effects cannot be read via the kvdb read methods) | |
1086 | std::atomic<int> flushing_count = {0}; | |
1087 | std::atomic<int> waiting_count = {0}; | |
1088 | /// protect flush_txns | |
1089 | ceph::mutex flush_lock = ceph::make_mutex("BlueStore::Onode::flush_lock"); | |
1090 | ceph::condition_variable flush_cond; ///< wait here for uncommitted txns | |
1091 | ||
1092 | Onode(Collection *c, const ghobject_t& o, | |
1093 | const mempool::bluestore_cache_meta::string& k) | |
1094 | : nref(0), | |
1095 | c(c), | |
1096 | oid(o), | |
1097 | key(k), | |
1098 | exists(false), | |
1099 | cached(false), | |
1100 | pinned(false), | |
1101 | extent_map(this) { | |
1102 | } | |
1103 | Onode(Collection* c, const ghobject_t& o, | |
1104 | const std::string& k) | |
1105 | : nref(0), | |
1106 | c(c), | |
1107 | oid(o), | |
1108 | key(k), | |
1109 | exists(false), | |
1110 | cached(false), | |
1111 | pinned(false), | |
1112 | extent_map(this) { | |
1113 | } | |
1114 | Onode(Collection* c, const ghobject_t& o, | |
1115 | const char* k) | |
1116 | : nref(0), | |
1117 | c(c), | |
1118 | oid(o), | |
1119 | key(k), | |
1120 | exists(false), | |
1121 | cached(false), | |
1122 | pinned(false), | |
1123 | extent_map(this) { | |
1124 | } | |
1125 | ||
1126 | static Onode* decode( | |
1127 | CollectionRef c, | |
1128 | const ghobject_t& oid, | |
1129 | const std::string& key, | |
1130 | const ceph::buffer::list& v); | |
1131 | ||
1132 | void dump(ceph::Formatter* f) const; | |
1133 | ||
1134 | void flush(); | |
1135 | void get(); | |
1136 | void put(); | |
1137 | ||
1138 | inline bool put_cache() { | |
1139 | ceph_assert(!cached); | |
1140 | cached = true; | |
1141 | return !pinned; | |
1142 | } | |
1143 | inline bool pop_cache() { | |
1144 | ceph_assert(cached); | |
1145 | cached = false; | |
1146 | return !pinned; | |
1147 | } | |
1148 | ||
1149 | const std::string& get_omap_prefix(); | |
1150 | void get_omap_header(std::string *out); | |
1151 | void get_omap_key(const std::string& key, std::string *out); | |
1152 | void rewrite_omap_key(const std::string& old, std::string *out); | |
1153 | void get_omap_tail(std::string *out); | |
1154 | void decode_omap_key(const std::string& key, std::string *user_key); | |
1155 | ||
1156 | // Return the offset of an object on disk. This function is intended *only* | |
1157 | // for use with zoned storage devices because in these devices, the objects | |
1158 | // are laid out contiguously on disk, which is not the case in general. | |
1159 | // Also, it should always be called after calling extent_map.fault_range(), | |
1160 | // so that the extent map is loaded. | |
1161 | int64_t zoned_get_ondisk_starting_offset() const { | |
1162 | return extent_map.extent_map.begin()->blob-> | |
1163 | get_blob().calc_offset(0, nullptr); | |
1164 | } | |
1165 | }; | |
1166 | typedef boost::intrusive_ptr<Onode> OnodeRef; | |
1167 | ||
1168 | /// A generic Cache Shard | |
1169 | struct CacheShard { | |
1170 | CephContext *cct; | |
1171 | PerfCounters *logger; | |
1172 | ||
1173 | /// protect lru and other structures | |
1174 | ceph::recursive_mutex lock = { | |
1175 | ceph::make_recursive_mutex("BlueStore::CacheShard::lock") }; | |
1176 | ||
1177 | std::atomic<uint64_t> max = {0}; | |
1178 | std::atomic<uint64_t> num = {0}; | |
1179 | ||
1180 | CacheShard(CephContext* cct) : cct(cct), logger(nullptr) {} | |
1181 | virtual ~CacheShard() {} | |
1182 | ||
1183 | void set_max(uint64_t max_) { | |
1184 | max = max_; | |
1185 | } | |
1186 | ||
1187 | uint64_t _get_num() { | |
1188 | return num; | |
1189 | } | |
1190 | ||
1191 | virtual void _trim_to(uint64_t new_size) = 0; | |
1192 | void _trim() { | |
1193 | if (cct->_conf->objectstore_blackhole) { | |
1194 | // do not trim if we are throwing away IOs a layer down | |
1195 | return; | |
1196 | } | |
1197 | _trim_to(max); | |
1198 | } | |
1199 | ||
1200 | void trim() { | |
1201 | std::lock_guard l(lock); | |
1202 | _trim(); | |
1203 | } | |
1204 | void flush() { | |
1205 | std::lock_guard l(lock); | |
1206 | // we should not be shutting down after the blackhole is enabled | |
1207 | assert(!cct->_conf->objectstore_blackhole); | |
1208 | _trim_to(0); | |
1209 | } | |
1210 | ||
1211 | #ifdef DEBUG_CACHE | |
1212 | virtual void _audit(const char *s) = 0; | |
1213 | #else | |
1214 | void _audit(const char *s) { /* no-op */ } | |
1215 | #endif | |
1216 | }; | |
1217 | ||
1218 | /// A Generic onode Cache Shard | |
1219 | struct OnodeCacheShard : public CacheShard { | |
1220 | std::atomic<uint64_t> num_pinned = {0}; | |
1221 | ||
1222 | std::array<std::pair<ghobject_t, ceph::mono_clock::time_point>, 64> dumped_onodes; | |
1223 | ||
1224 | virtual void _pin(Onode* o) = 0; | |
1225 | virtual void _unpin(Onode* o) = 0; | |
1226 | ||
1227 | public: | |
1228 | OnodeCacheShard(CephContext* cct) : CacheShard(cct) {} | |
1229 | static OnodeCacheShard *create(CephContext* cct, std::string type, | |
1230 | PerfCounters *logger); | |
1231 | virtual void _add(Onode* o, int level) = 0; | |
1232 | virtual void _rm(Onode* o) = 0; | |
1233 | virtual void _unpin_and_rm(Onode* o) = 0; | |
1234 | ||
1235 | virtual void move_pinned(OnodeCacheShard *to, Onode *o) = 0; | |
1236 | virtual void add_stats(uint64_t *onodes, uint64_t *pinned_onodes) = 0; | |
1237 | bool empty() { | |
1238 | return _get_num() == 0; | |
1239 | } | |
1240 | }; | |
1241 | ||
1242 | /// A Generic buffer Cache Shard | |
1243 | struct BufferCacheShard : public CacheShard { | |
1244 | std::atomic<uint64_t> num_extents = {0}; | |
1245 | std::atomic<uint64_t> num_blobs = {0}; | |
1246 | uint64_t buffer_bytes = 0; | |
1247 | ||
1248 | public: | |
1249 | BufferCacheShard(CephContext* cct) : CacheShard(cct) {} | |
1250 | static BufferCacheShard *create(CephContext* cct, std::string type, | |
1251 | PerfCounters *logger); | |
1252 | virtual void _add(Buffer *b, int level, Buffer *near) = 0; | |
1253 | virtual void _rm(Buffer *b) = 0; | |
1254 | virtual void _move(BufferCacheShard *src, Buffer *b) = 0; | |
1255 | virtual void _touch(Buffer *b) = 0; | |
1256 | virtual void _adjust_size(Buffer *b, int64_t delta) = 0; | |
1257 | ||
1258 | uint64_t _get_bytes() { | |
1259 | return buffer_bytes; | |
1260 | } | |
1261 | ||
1262 | void add_extent() { | |
1263 | ++num_extents; | |
1264 | } | |
1265 | void rm_extent() { | |
1266 | --num_extents; | |
1267 | } | |
1268 | ||
1269 | void add_blob() { | |
1270 | ++num_blobs; | |
1271 | } | |
1272 | void rm_blob() { | |
1273 | --num_blobs; | |
1274 | } | |
1275 | ||
1276 | virtual void add_stats(uint64_t *extents, | |
1277 | uint64_t *blobs, | |
1278 | uint64_t *buffers, | |
1279 | uint64_t *bytes) = 0; | |
1280 | ||
1281 | bool empty() { | |
1282 | std::lock_guard l(lock); | |
1283 | return _get_bytes() == 0; | |
1284 | } | |
1285 | }; | |
1286 | ||
1287 | struct OnodeSpace { | |
1288 | OnodeCacheShard *cache; | |
1289 | ||
1290 | private: | |
1291 | /// forward lookups | |
1292 | mempool::bluestore_cache_meta::unordered_map<ghobject_t,OnodeRef> onode_map; | |
1293 | ||
1294 | friend struct Collection; // for split_cache() | |
1295 | friend struct Onode; // for put() | |
1296 | friend struct LruOnodeCacheShard; | |
1297 | void _remove(const ghobject_t& oid); | |
1298 | public: | |
1299 | OnodeSpace(OnodeCacheShard *c) : cache(c) {} | |
1300 | ~OnodeSpace() { | |
1301 | clear(); | |
1302 | } | |
1303 | ||
1304 | OnodeRef add(const ghobject_t& oid, OnodeRef& o); | |
1305 | OnodeRef lookup(const ghobject_t& o); | |
1306 | void rename(OnodeRef& o, const ghobject_t& old_oid, | |
1307 | const ghobject_t& new_oid, | |
1308 | const mempool::bluestore_cache_meta::string& new_okey); | |
1309 | void clear(); | |
1310 | bool empty(); | |
1311 | ||
1312 | template <int LogLevelV> | |
1313 | void dump(CephContext *cct); | |
1314 | ||
1315 | /// return true if f true for any item | |
1316 | bool map_any(std::function<bool(Onode*)> f); | |
1317 | }; | |
1318 | ||
1319 | class OpSequencer; | |
1320 | using OpSequencerRef = ceph::ref_t<OpSequencer>; | |
1321 | ||
1322 | struct Collection : public CollectionImpl { | |
1323 | BlueStore *store; | |
1324 | OpSequencerRef osr; | |
1325 | BufferCacheShard *cache; ///< our cache shard | |
1326 | bluestore_cnode_t cnode; | |
1327 | ceph::shared_mutex lock = | |
1328 | ceph::make_shared_mutex("BlueStore::Collection::lock", true, false); | |
1329 | ||
1330 | bool exists; | |
1331 | ||
1332 | SharedBlobSet shared_blob_set; ///< open SharedBlobs | |
1333 | ||
1334 | // cache onodes on a per-collection basis to avoid lock | |
1335 | // contention. | |
1336 | OnodeSpace onode_map; | |
1337 | ||
1338 | //pool options | |
1339 | pool_opts_t pool_opts; | |
1340 | ContextQueue *commit_queue; | |
1341 | ||
1342 | OnodeCacheShard* get_onode_cache() const { | |
1343 | return onode_map.cache; | |
1344 | } | |
1345 | OnodeRef get_onode(const ghobject_t& oid, bool create, bool is_createop=false); | |
1346 | ||
1347 | // the terminology is confusing here, sorry! | |
1348 | // | |
1349 | // blob_t shared_blob_t | |
1350 | // !shared unused -> open | |
1351 | // shared !loaded -> open + shared | |
1352 | // shared loaded -> open + shared + loaded | |
1353 | // | |
1354 | // i.e., | |
1355 | // open = SharedBlob is instantiated | |
1356 | // shared = blob_t shared flag is std::set; SharedBlob is hashed. | |
1357 | // loaded = SharedBlob::shared_blob_t is loaded from kv store | |
1358 | void open_shared_blob(uint64_t sbid, BlobRef b); | |
1359 | void load_shared_blob(SharedBlobRef sb); | |
1360 | void make_blob_shared(uint64_t sbid, BlobRef b); | |
1361 | uint64_t make_blob_unshared(SharedBlob *sb); | |
1362 | ||
1363 | BlobRef new_blob() { | |
1364 | BlobRef b = new Blob(); | |
1365 | b->shared_blob = new SharedBlob(this); | |
1366 | return b; | |
1367 | } | |
1368 | ||
1369 | bool contains(const ghobject_t& oid) { | |
1370 | if (cid.is_meta()) | |
1371 | return oid.hobj.pool == -1; | |
1372 | spg_t spgid; | |
1373 | if (cid.is_pg(&spgid)) | |
1374 | return | |
1375 | spgid.pgid.contains(cnode.bits, oid) && | |
1376 | oid.shard_id == spgid.shard; | |
1377 | return false; | |
1378 | } | |
1379 | ||
1380 | int64_t pool() const { | |
1381 | return cid.pool(); | |
1382 | } | |
1383 | ||
1384 | void split_cache(Collection *dest); | |
1385 | ||
1386 | bool flush_commit(Context *c) override; | |
1387 | void flush() override; | |
1388 | void flush_all_but_last(); | |
1389 | ||
1390 | Collection(BlueStore *ns, OnodeCacheShard *oc, BufferCacheShard *bc, coll_t c); | |
1391 | }; | |
1392 | ||
1393 | class OmapIteratorImpl : public ObjectMap::ObjectMapIteratorImpl { | |
1394 | CollectionRef c; | |
1395 | OnodeRef o; | |
1396 | KeyValueDB::Iterator it; | |
1397 | std::string head, tail; | |
1398 | ||
1399 | std::string _stringify() const; | |
1400 | ||
1401 | public: | |
1402 | OmapIteratorImpl(CollectionRef c, OnodeRef o, KeyValueDB::Iterator it); | |
1403 | int seek_to_first() override; | |
1404 | int upper_bound(const std::string &after) override; | |
1405 | int lower_bound(const std::string &to) override; | |
1406 | bool valid() override; | |
1407 | int next() override; | |
1408 | std::string key() override; | |
1409 | ceph::buffer::list value() override; | |
1410 | std::string tail_key() override { | |
1411 | return tail; | |
1412 | } | |
1413 | ||
1414 | int status() override { | |
1415 | return 0; | |
1416 | } | |
1417 | }; | |
1418 | ||
1419 | struct volatile_statfs{ | |
1420 | enum { | |
1421 | STATFS_ALLOCATED = 0, | |
1422 | STATFS_STORED, | |
1423 | STATFS_COMPRESSED_ORIGINAL, | |
1424 | STATFS_COMPRESSED, | |
1425 | STATFS_COMPRESSED_ALLOCATED, | |
1426 | STATFS_LAST | |
1427 | }; | |
1428 | int64_t values[STATFS_LAST]; | |
1429 | volatile_statfs() { | |
1430 | memset(this, 0, sizeof(volatile_statfs)); | |
1431 | } | |
1432 | void reset() { | |
1433 | *this = volatile_statfs(); | |
1434 | } | |
1435 | void publish(store_statfs_t* buf) const { | |
1436 | buf->allocated = allocated(); | |
1437 | buf->data_stored = stored(); | |
1438 | buf->data_compressed = compressed(); | |
1439 | buf->data_compressed_original = compressed_original(); | |
1440 | buf->data_compressed_allocated = compressed_allocated(); | |
1441 | } | |
1442 | ||
1443 | volatile_statfs& operator+=(const volatile_statfs& other) { | |
1444 | for (size_t i = 0; i < STATFS_LAST; ++i) { | |
1445 | values[i] += other.values[i]; | |
1446 | } | |
1447 | return *this; | |
1448 | } | |
1449 | int64_t& allocated() { | |
1450 | return values[STATFS_ALLOCATED]; | |
1451 | } | |
1452 | int64_t& stored() { | |
1453 | return values[STATFS_STORED]; | |
1454 | } | |
1455 | int64_t& compressed_original() { | |
1456 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1457 | } | |
1458 | int64_t& compressed() { | |
1459 | return values[STATFS_COMPRESSED]; | |
1460 | } | |
1461 | int64_t& compressed_allocated() { | |
1462 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1463 | } | |
1464 | int64_t allocated() const { | |
1465 | return values[STATFS_ALLOCATED]; | |
1466 | } | |
1467 | int64_t stored() const { | |
1468 | return values[STATFS_STORED]; | |
1469 | } | |
1470 | int64_t compressed_original() const { | |
1471 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1472 | } | |
1473 | int64_t compressed() const { | |
1474 | return values[STATFS_COMPRESSED]; | |
1475 | } | |
1476 | int64_t compressed_allocated() const { | |
1477 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1478 | } | |
1479 | volatile_statfs& operator=(const store_statfs_t& st) { | |
1480 | values[STATFS_ALLOCATED] = st.allocated; | |
1481 | values[STATFS_STORED] = st.data_stored; | |
1482 | values[STATFS_COMPRESSED_ORIGINAL] = st.data_compressed_original; | |
1483 | values[STATFS_COMPRESSED] = st.data_compressed; | |
1484 | values[STATFS_COMPRESSED_ALLOCATED] = st.data_compressed_allocated; | |
1485 | return *this; | |
1486 | } | |
1487 | bool is_empty() { | |
1488 | return values[STATFS_ALLOCATED] == 0 && | |
1489 | values[STATFS_STORED] == 0 && | |
1490 | values[STATFS_COMPRESSED] == 0 && | |
1491 | values[STATFS_COMPRESSED_ORIGINAL] == 0 && | |
1492 | values[STATFS_COMPRESSED_ALLOCATED] == 0; | |
1493 | } | |
1494 | void decode(ceph::buffer::list::const_iterator& it) { | |
1495 | using ceph::decode; | |
1496 | for (size_t i = 0; i < STATFS_LAST; i++) { | |
1497 | decode(values[i], it); | |
1498 | } | |
1499 | } | |
1500 | ||
1501 | void encode(ceph::buffer::list& bl) { | |
1502 | using ceph::encode; | |
1503 | for (size_t i = 0; i < STATFS_LAST; i++) { | |
1504 | encode(values[i], bl); | |
1505 | } | |
1506 | } | |
1507 | }; | |
1508 | ||
1509 | struct TransContext final : public AioContext { | |
1510 | MEMPOOL_CLASS_HELPERS(); | |
1511 | ||
1512 | typedef enum { | |
1513 | STATE_PREPARE, | |
1514 | STATE_AIO_WAIT, | |
1515 | STATE_IO_DONE, | |
1516 | STATE_KV_QUEUED, // queued for kv_sync_thread submission | |
1517 | STATE_KV_SUBMITTED, // submitted to kv; not yet synced | |
1518 | STATE_KV_DONE, | |
1519 | STATE_DEFERRED_QUEUED, // in deferred_queue (pending or running) | |
1520 | STATE_DEFERRED_CLEANUP, // remove deferred kv record | |
1521 | STATE_DEFERRED_DONE, | |
1522 | STATE_FINISHING, | |
1523 | STATE_DONE, | |
1524 | } state_t; | |
1525 | ||
1526 | const char *get_state_name() { | |
1527 | switch (state) { | |
1528 | case STATE_PREPARE: return "prepare"; | |
1529 | case STATE_AIO_WAIT: return "aio_wait"; | |
1530 | case STATE_IO_DONE: return "io_done"; | |
1531 | case STATE_KV_QUEUED: return "kv_queued"; | |
1532 | case STATE_KV_SUBMITTED: return "kv_submitted"; | |
1533 | case STATE_KV_DONE: return "kv_done"; | |
1534 | case STATE_DEFERRED_QUEUED: return "deferred_queued"; | |
1535 | case STATE_DEFERRED_CLEANUP: return "deferred_cleanup"; | |
1536 | case STATE_DEFERRED_DONE: return "deferred_done"; | |
1537 | case STATE_FINISHING: return "finishing"; | |
1538 | case STATE_DONE: return "done"; | |
1539 | } | |
1540 | return "???"; | |
1541 | } | |
1542 | ||
1543 | #if defined(WITH_LTTNG) | |
1544 | const char *get_state_latency_name(int state) { | |
1545 | switch (state) { | |
1546 | case l_bluestore_state_prepare_lat: return "prepare"; | |
1547 | case l_bluestore_state_aio_wait_lat: return "aio_wait"; | |
1548 | case l_bluestore_state_io_done_lat: return "io_done"; | |
1549 | case l_bluestore_state_kv_queued_lat: return "kv_queued"; | |
1550 | case l_bluestore_state_kv_committing_lat: return "kv_committing"; | |
1551 | case l_bluestore_state_kv_done_lat: return "kv_done"; | |
1552 | case l_bluestore_state_deferred_queued_lat: return "deferred_queued"; | |
1553 | case l_bluestore_state_deferred_cleanup_lat: return "deferred_cleanup"; | |
1554 | case l_bluestore_state_finishing_lat: return "finishing"; | |
1555 | case l_bluestore_state_done_lat: return "done"; | |
1556 | } | |
1557 | return "???"; | |
1558 | } | |
1559 | #endif | |
1560 | ||
1561 | inline void set_state(state_t s) { | |
1562 | state = s; | |
1563 | #ifdef WITH_BLKIN | |
1564 | if (trace) { | |
1565 | trace.event(get_state_name()); | |
1566 | } | |
1567 | #endif | |
1568 | } | |
1569 | inline state_t get_state() { | |
1570 | return state; | |
1571 | } | |
1572 | ||
1573 | CollectionRef ch; | |
1574 | OpSequencerRef osr; // this should be ch->osr | |
1575 | boost::intrusive::list_member_hook<> sequencer_item; | |
1576 | ||
1577 | uint64_t bytes = 0, ios = 0, cost = 0; | |
1578 | ||
1579 | std::set<OnodeRef> onodes; ///< these need to be updated/written | |
1580 | std::set<OnodeRef> modified_objects; ///< objects we modified (and need a ref) | |
1581 | ||
1582 | // A map from onode to a vector of object offset. For new objects created | |
1583 | // in the transaction we append the new offset to the vector, for | |
1584 | // overwritten objects we append the negative of the previous ondisk offset | |
1585 | // followed by the new offset, and for truncated objects we append the | |
1586 | // negative of the previous ondisk offset. We need to maintain a vector of | |
1587 | // offsets because *within the same transaction* an object may be truncated | |
1588 | // and then written again, or an object may be overwritten multiple times to | |
1589 | // different zones. See update_cleaning_metadata function for how this map | |
1590 | // is used. | |
1591 | std::map<OnodeRef, std::vector<int64_t>> zoned_onode_to_offset_map; | |
1592 | ||
1593 | std::set<SharedBlobRef> shared_blobs; ///< these need to be updated/written | |
1594 | std::set<SharedBlobRef> shared_blobs_written; ///< update these on io completion | |
1595 | ||
1596 | KeyValueDB::Transaction t; ///< then we will commit this | |
1597 | std::list<Context*> oncommits; ///< more commit completions | |
1598 | std::list<CollectionRef> removed_collections; ///< colls we removed | |
1599 | ||
1600 | boost::intrusive::list_member_hook<> deferred_queue_item; | |
1601 | bluestore_deferred_transaction_t *deferred_txn = nullptr; ///< if any | |
1602 | ||
1603 | interval_set<uint64_t> allocated, released; | |
1604 | volatile_statfs statfs_delta; ///< overall store statistics delta | |
1605 | uint64_t osd_pool_id = META_POOL_ID; ///< osd pool id we're operating on | |
1606 | ||
1607 | IOContext ioc; | |
1608 | bool had_ios = false; ///< true if we submitted IOs before our kv txn | |
1609 | ||
1610 | uint64_t seq = 0; | |
1611 | ceph::mono_clock::time_point start; | |
1612 | ceph::mono_clock::time_point last_stamp; | |
1613 | ||
1614 | uint64_t last_nid = 0; ///< if non-zero, highest new nid we allocated | |
1615 | uint64_t last_blobid = 0; ///< if non-zero, highest new blobid we allocated | |
1616 | ||
1617 | #if defined(WITH_LTTNG) | |
1618 | bool tracing = false; | |
1619 | #endif | |
1620 | ||
1621 | #ifdef WITH_BLKIN | |
1622 | ZTracer::Trace trace; | |
1623 | #endif | |
1624 | ||
1625 | explicit TransContext(CephContext* cct, Collection *c, OpSequencer *o, | |
1626 | std::list<Context*> *on_commits) | |
1627 | : ch(c), | |
1628 | osr(o), | |
1629 | ioc(cct, this), | |
1630 | start(ceph::mono_clock::now()) { | |
1631 | last_stamp = start; | |
1632 | if (on_commits) { | |
1633 | oncommits.swap(*on_commits); | |
1634 | } | |
1635 | } | |
1636 | ~TransContext() { | |
1637 | #ifdef WITH_BLKIN | |
1638 | if (trace) { | |
1639 | trace.event("txc destruct"); | |
1640 | } | |
1641 | #endif | |
1642 | delete deferred_txn; | |
1643 | } | |
1644 | ||
1645 | void write_onode(OnodeRef &o) { | |
1646 | onodes.insert(o); | |
1647 | } | |
1648 | void write_shared_blob(SharedBlobRef &sb) { | |
1649 | shared_blobs.insert(sb); | |
1650 | } | |
1651 | void unshare_blob(SharedBlob *sb) { | |
1652 | shared_blobs.erase(sb); | |
1653 | } | |
1654 | ||
1655 | /// note we logically modified object (when onode itself is unmodified) | |
1656 | void note_modified_object(OnodeRef &o) { | |
1657 | // onode itself isn't written, though | |
1658 | modified_objects.insert(o); | |
1659 | } | |
1660 | void note_removed_object(OnodeRef& o) { | |
1661 | modified_objects.insert(o); | |
1662 | onodes.erase(o); | |
1663 | } | |
1664 | ||
1665 | void zoned_note_new_object(OnodeRef &o) { | |
1666 | auto [_, ok] = zoned_onode_to_offset_map.emplace( | |
1667 | std::pair<OnodeRef, std::vector<int64_t>>(o, {o->zoned_get_ondisk_starting_offset()})); | |
1668 | ceph_assert(ok); | |
1669 | } | |
1670 | ||
1671 | void zoned_note_updated_object(OnodeRef &o, int64_t prev_offset) { | |
1672 | int64_t new_offset = o->zoned_get_ondisk_starting_offset(); | |
1673 | auto [it, ok] = zoned_onode_to_offset_map.emplace( | |
1674 | std::pair<OnodeRef, std::vector<int64_t>>(o, {-prev_offset, new_offset})); | |
1675 | if (!ok) { | |
1676 | it->second.push_back(-prev_offset); | |
1677 | it->second.push_back(new_offset); | |
1678 | } | |
1679 | } | |
1680 | ||
1681 | void zoned_note_truncated_object(OnodeRef &o, int64_t offset) { | |
1682 | auto [it, ok] = zoned_onode_to_offset_map.emplace( | |
1683 | std::pair<OnodeRef, std::vector<int64_t>>(o, {-offset})); | |
1684 | if (!ok) { | |
1685 | it->second.push_back(-offset); | |
1686 | } | |
1687 | } | |
1688 | ||
1689 | void aio_finish(BlueStore *store) override { | |
1690 | store->txc_aio_finish(this); | |
1691 | } | |
1692 | private: | |
1693 | state_t state = STATE_PREPARE; | |
1694 | }; | |
1695 | ||
1696 | class BlueStoreThrottle { | |
1697 | #if defined(WITH_LTTNG) | |
1698 | const std::chrono::time_point<ceph::mono_clock> time_base = ceph::mono_clock::now(); | |
1699 | ||
1700 | // Time of last chosen io (microseconds) | |
1701 | std::atomic<uint64_t> previous_emitted_tp_time_mono_mcs = {0}; | |
1702 | std::atomic<uint64_t> ios_started_since_last_traced = {0}; | |
1703 | std::atomic<uint64_t> ios_completed_since_last_traced = {0}; | |
1704 | ||
1705 | std::atomic_uint pending_kv_ios = {0}; | |
1706 | std::atomic_uint pending_deferred_ios = {0}; | |
1707 | ||
1708 | // Min period between trace points (microseconds) | |
1709 | std::atomic<uint64_t> trace_period_mcs = {0}; | |
1710 | ||
1711 | bool should_trace( | |
1712 | uint64_t *started, | |
1713 | uint64_t *completed) { | |
1714 | uint64_t min_period_mcs = trace_period_mcs.load( | |
1715 | std::memory_order_relaxed); | |
1716 | ||
1717 | if (min_period_mcs == 0) { | |
1718 | *started = 1; | |
1719 | *completed = ios_completed_since_last_traced.exchange(0); | |
1720 | return true; | |
1721 | } else { | |
1722 | ios_started_since_last_traced++; | |
1723 | auto now_mcs = ceph::to_microseconds<uint64_t>( | |
1724 | ceph::mono_clock::now() - time_base); | |
1725 | uint64_t previous_mcs = previous_emitted_tp_time_mono_mcs; | |
1726 | uint64_t period_mcs = now_mcs - previous_mcs; | |
1727 | if (period_mcs > min_period_mcs) { | |
1728 | if (previous_emitted_tp_time_mono_mcs.compare_exchange_strong( | |
1729 | previous_mcs, now_mcs)) { | |
1730 | // This would be racy at a sufficiently extreme trace rate, but isn't | |
1731 | // worth the overhead of doing it more carefully. | |
1732 | *started = ios_started_since_last_traced.exchange(0); | |
1733 | *completed = ios_completed_since_last_traced.exchange(0); | |
1734 | return true; | |
1735 | } | |
1736 | } | |
1737 | return false; | |
1738 | } | |
1739 | } | |
1740 | #endif | |
1741 | ||
1742 | #if defined(WITH_LTTNG) | |
1743 | void emit_initial_tracepoint( | |
1744 | KeyValueDB &db, | |
1745 | TransContext &txc, | |
1746 | ceph::mono_clock::time_point); | |
1747 | #else | |
1748 | void emit_initial_tracepoint( | |
1749 | KeyValueDB &db, | |
1750 | TransContext &txc, | |
1751 | ceph::mono_clock::time_point) {} | |
1752 | #endif | |
1753 | ||
1754 | Throttle throttle_bytes; ///< submit to commit | |
1755 | Throttle throttle_deferred_bytes; ///< submit to deferred complete | |
1756 | ||
1757 | public: | |
1758 | BlueStoreThrottle(CephContext *cct) : | |
1759 | throttle_bytes(cct, "bluestore_throttle_bytes", 0), | |
1760 | throttle_deferred_bytes(cct, "bluestore_throttle_deferred_bytes", 0) | |
1761 | { | |
1762 | reset_throttle(cct->_conf); | |
1763 | } | |
1764 | ||
1765 | #if defined(WITH_LTTNG) | |
1766 | void complete_kv(TransContext &txc); | |
1767 | void complete(TransContext &txc); | |
1768 | #else | |
1769 | void complete_kv(TransContext &txc) {} | |
1770 | void complete(TransContext &txc) {} | |
1771 | #endif | |
1772 | ||
1773 | ceph::mono_clock::duration log_state_latency( | |
1774 | TransContext &txc, PerfCounters *logger, int state); | |
1775 | bool try_start_transaction( | |
1776 | KeyValueDB &db, | |
1777 | TransContext &txc, | |
1778 | ceph::mono_clock::time_point); | |
1779 | void finish_start_transaction( | |
1780 | KeyValueDB &db, | |
1781 | TransContext &txc, | |
1782 | ceph::mono_clock::time_point); | |
1783 | void release_kv_throttle(uint64_t cost) { | |
1784 | throttle_bytes.put(cost); | |
1785 | } | |
1786 | void release_deferred_throttle(uint64_t cost) { | |
1787 | throttle_deferred_bytes.put(cost); | |
1788 | } | |
1789 | bool should_submit_deferred() { | |
1790 | return throttle_deferred_bytes.past_midpoint(); | |
1791 | } | |
1792 | void reset_throttle(const ConfigProxy &conf) { | |
1793 | throttle_bytes.reset_max(conf->bluestore_throttle_bytes); | |
1794 | throttle_deferred_bytes.reset_max( | |
1795 | conf->bluestore_throttle_bytes + | |
1796 | conf->bluestore_throttle_deferred_bytes); | |
1797 | #if defined(WITH_LTTNG) | |
1798 | double rate = conf.get_val<double>("bluestore_throttle_trace_rate"); | |
1799 | trace_period_mcs = rate > 0 ? floor((1/rate) * 1000000.0) : 0; | |
1800 | #endif | |
1801 | } | |
1802 | } throttle; | |
1803 | ||
1804 | typedef boost::intrusive::list< | |
1805 | TransContext, | |
1806 | boost::intrusive::member_hook< | |
1807 | TransContext, | |
1808 | boost::intrusive::list_member_hook<>, | |
1809 | &TransContext::deferred_queue_item> > deferred_queue_t; | |
1810 | ||
1811 | struct DeferredBatch final : public AioContext { | |
1812 | OpSequencer *osr; | |
1813 | struct deferred_io { | |
1814 | ceph::buffer::list bl; ///< data | |
1815 | uint64_t seq; ///< deferred transaction seq | |
1816 | }; | |
1817 | std::map<uint64_t,deferred_io> iomap; ///< map of ios in this batch | |
1818 | deferred_queue_t txcs; ///< txcs in this batch | |
1819 | IOContext ioc; ///< our aios | |
1820 | /// bytes of pending io for each deferred seq (may be 0) | |
1821 | std::map<uint64_t,int> seq_bytes; | |
1822 | ||
1823 | void _discard(CephContext *cct, uint64_t offset, uint64_t length); | |
1824 | void _audit(CephContext *cct); | |
1825 | ||
1826 | DeferredBatch(CephContext *cct, OpSequencer *osr) | |
1827 | : osr(osr), ioc(cct, this) {} | |
1828 | ||
1829 | /// prepare a write | |
1830 | void prepare_write(CephContext *cct, | |
1831 | uint64_t seq, uint64_t offset, uint64_t length, | |
1832 | ceph::buffer::list::const_iterator& p); | |
1833 | ||
1834 | void aio_finish(BlueStore *store) override { | |
1835 | store->_deferred_aio_finish(osr); | |
1836 | } | |
1837 | }; | |
1838 | ||
1839 | class OpSequencer : public RefCountedObject { | |
1840 | public: | |
1841 | ceph::mutex qlock = ceph::make_mutex("BlueStore::OpSequencer::qlock"); | |
1842 | ceph::condition_variable qcond; | |
1843 | typedef boost::intrusive::list< | |
1844 | TransContext, | |
1845 | boost::intrusive::member_hook< | |
1846 | TransContext, | |
1847 | boost::intrusive::list_member_hook<>, | |
1848 | &TransContext::sequencer_item> > q_list_t; | |
1849 | q_list_t q; ///< transactions | |
1850 | ||
1851 | boost::intrusive::list_member_hook<> deferred_osr_queue_item; | |
1852 | ||
1853 | DeferredBatch *deferred_running = nullptr; | |
1854 | DeferredBatch *deferred_pending = nullptr; | |
1855 | ||
1856 | ceph::mutex deferred_lock = ceph::make_mutex("BlueStore::OpSequencer::deferred_lock"); | |
1857 | ||
1858 | BlueStore *store; | |
1859 | coll_t cid; | |
1860 | ||
1861 | uint64_t last_seq = 0; | |
1862 | ||
1863 | std::atomic_int txc_with_unstable_io = {0}; ///< num txcs with unstable io | |
1864 | ||
1865 | std::atomic_int kv_committing_serially = {0}; | |
1866 | ||
1867 | std::atomic_int kv_submitted_waiters = {0}; | |
1868 | ||
1869 | std::atomic_bool zombie = {false}; ///< in zombie_osr std::set (collection going away) | |
1870 | ||
1871 | const uint32_t sequencer_id; | |
1872 | ||
1873 | uint32_t get_sequencer_id() const { | |
1874 | return sequencer_id; | |
1875 | } | |
1876 | ||
1877 | void queue_new(TransContext *txc) { | |
1878 | std::lock_guard l(qlock); | |
1879 | txc->seq = ++last_seq; | |
1880 | q.push_back(*txc); | |
1881 | } | |
1882 | ||
1883 | void drain() { | |
1884 | std::unique_lock l(qlock); | |
1885 | while (!q.empty()) | |
1886 | qcond.wait(l); | |
1887 | } | |
1888 | ||
1889 | void drain_preceding(TransContext *txc) { | |
1890 | std::unique_lock l(qlock); | |
1891 | while (&q.front() != txc) | |
1892 | qcond.wait(l); | |
1893 | } | |
1894 | ||
1895 | bool _is_all_kv_submitted() { | |
1896 | // caller must hold qlock & q.empty() must not empty | |
1897 | ceph_assert(!q.empty()); | |
1898 | TransContext *txc = &q.back(); | |
1899 | if (txc->get_state() >= TransContext::STATE_KV_SUBMITTED) { | |
1900 | return true; | |
1901 | } | |
1902 | return false; | |
1903 | } | |
1904 | ||
1905 | void flush() { | |
1906 | std::unique_lock l(qlock); | |
1907 | while (true) { | |
1908 | // std::set flag before the check because the condition | |
1909 | // may become true outside qlock, and we need to make | |
1910 | // sure those threads see waiters and signal qcond. | |
1911 | ++kv_submitted_waiters; | |
1912 | if (q.empty() || _is_all_kv_submitted()) { | |
1913 | --kv_submitted_waiters; | |
1914 | return; | |
1915 | } | |
1916 | qcond.wait(l); | |
1917 | --kv_submitted_waiters; | |
1918 | } | |
1919 | } | |
1920 | ||
1921 | void flush_all_but_last() { | |
1922 | std::unique_lock l(qlock); | |
1923 | assert (q.size() >= 1); | |
1924 | while (true) { | |
1925 | // std::set flag before the check because the condition | |
1926 | // may become true outside qlock, and we need to make | |
1927 | // sure those threads see waiters and signal qcond. | |
1928 | ++kv_submitted_waiters; | |
1929 | if (q.size() <= 1) { | |
1930 | --kv_submitted_waiters; | |
1931 | return; | |
1932 | } else { | |
1933 | auto it = q.rbegin(); | |
1934 | it++; | |
1935 | if (it->get_state() >= TransContext::STATE_KV_SUBMITTED) { | |
1936 | --kv_submitted_waiters; | |
1937 | return; | |
1938 | } | |
1939 | } | |
1940 | qcond.wait(l); | |
1941 | --kv_submitted_waiters; | |
1942 | } | |
1943 | } | |
1944 | ||
1945 | bool flush_commit(Context *c) { | |
1946 | std::lock_guard l(qlock); | |
1947 | if (q.empty()) { | |
1948 | return true; | |
1949 | } | |
1950 | TransContext *txc = &q.back(); | |
1951 | if (txc->get_state() >= TransContext::STATE_KV_DONE) { | |
1952 | return true; | |
1953 | } | |
1954 | txc->oncommits.push_back(c); | |
1955 | return false; | |
1956 | } | |
1957 | private: | |
1958 | FRIEND_MAKE_REF(OpSequencer); | |
1959 | OpSequencer(BlueStore *store, uint32_t sequencer_id, const coll_t& c) | |
1960 | : RefCountedObject(store->cct), | |
1961 | store(store), cid(c), sequencer_id(sequencer_id) { | |
1962 | } | |
1963 | ~OpSequencer() { | |
1964 | ceph_assert(q.empty()); | |
1965 | } | |
1966 | }; | |
1967 | ||
1968 | typedef boost::intrusive::list< | |
1969 | OpSequencer, | |
1970 | boost::intrusive::member_hook< | |
1971 | OpSequencer, | |
1972 | boost::intrusive::list_member_hook<>, | |
1973 | &OpSequencer::deferred_osr_queue_item> > deferred_osr_queue_t; | |
1974 | ||
1975 | struct KVSyncThread : public Thread { | |
1976 | BlueStore *store; | |
1977 | explicit KVSyncThread(BlueStore *s) : store(s) {} | |
1978 | void *entry() override { | |
1979 | store->_kv_sync_thread(); | |
1980 | return NULL; | |
1981 | } | |
1982 | }; | |
1983 | struct KVFinalizeThread : public Thread { | |
1984 | BlueStore *store; | |
1985 | explicit KVFinalizeThread(BlueStore *s) : store(s) {} | |
1986 | void *entry() override { | |
1987 | store->_kv_finalize_thread(); | |
1988 | return NULL; | |
1989 | } | |
1990 | }; | |
1991 | struct ZonedCleanerThread : public Thread { | |
1992 | BlueStore *store; | |
1993 | explicit ZonedCleanerThread(BlueStore *s) : store(s) {} | |
1994 | void *entry() override { | |
1995 | store->_zoned_cleaner_thread(); | |
1996 | return nullptr; | |
1997 | } | |
1998 | }; | |
1999 | ||
2000 | struct DBHistogram { | |
2001 | struct value_dist { | |
2002 | uint64_t count; | |
2003 | uint32_t max_len; | |
2004 | }; | |
2005 | ||
2006 | struct key_dist { | |
2007 | uint64_t count; | |
2008 | uint32_t max_len; | |
2009 | std::map<int, struct value_dist> val_map; ///< slab id to count, max length of value and key | |
2010 | }; | |
2011 | ||
2012 | std::map<std::string, std::map<int, struct key_dist> > key_hist; | |
2013 | std::map<int, uint64_t> value_hist; | |
2014 | int get_key_slab(size_t sz); | |
2015 | std::string get_key_slab_to_range(int slab); | |
2016 | int get_value_slab(size_t sz); | |
2017 | std::string get_value_slab_to_range(int slab); | |
2018 | void update_hist_entry(std::map<std::string, std::map<int, struct key_dist> > &key_hist, | |
2019 | const std::string &prefix, size_t key_size, size_t value_size); | |
2020 | void dump(ceph::Formatter *f); | |
2021 | }; | |
2022 | ||
2023 | struct BigDeferredWriteContext { | |
2024 | uint64_t off = 0; // original logical offset | |
2025 | uint32_t b_off = 0; // blob relative offset | |
2026 | uint32_t used = 0; | |
2027 | uint64_t head_read = 0; | |
2028 | uint64_t tail_read = 0; | |
2029 | BlobRef blob_ref; | |
2030 | uint64_t blob_start = 0; | |
2031 | PExtentVector res_extents; | |
2032 | ||
2033 | inline uint64_t blob_aligned_len() const { | |
2034 | return used + head_read + tail_read; | |
2035 | } | |
2036 | ||
2037 | bool can_defer(BlueStore::extent_map_t::iterator ep, | |
2038 | uint64_t prefer_deferred_size, | |
2039 | uint64_t block_size, | |
2040 | uint64_t offset, | |
2041 | uint64_t l); | |
2042 | bool apply_defer(); | |
2043 | }; | |
2044 | ||
2045 | // -------------------------------------------------------- | |
2046 | // members | |
2047 | private: | |
2048 | BlueFS *bluefs = nullptr; | |
2049 | bluefs_layout_t bluefs_layout; | |
2050 | utime_t next_dump_on_bluefs_alloc_failure; | |
2051 | ||
2052 | KeyValueDB *db = nullptr; | |
2053 | BlockDevice *bdev = nullptr; | |
2054 | std::string freelist_type; | |
2055 | FreelistManager *fm = nullptr; | |
2056 | ||
2057 | bluefs_shared_alloc_context_t shared_alloc; | |
2058 | ||
2059 | uuid_d fsid; | |
2060 | int path_fd = -1; ///< open handle to $path | |
2061 | int fsid_fd = -1; ///< open handle (locked) to $path/fsid | |
2062 | bool mounted = false; | |
2063 | ||
2064 | ceph::shared_mutex coll_lock = ceph::make_shared_mutex("BlueStore::coll_lock"); ///< rwlock to protect coll_map | |
2065 | mempool::bluestore_cache_other::unordered_map<coll_t, CollectionRef> coll_map; | |
2066 | bool collections_had_errors = false; | |
2067 | std::map<coll_t,CollectionRef> new_coll_map; | |
2068 | ||
2069 | std::vector<OnodeCacheShard*> onode_cache_shards; | |
2070 | std::vector<BufferCacheShard*> buffer_cache_shards; | |
2071 | ||
2072 | /// protect zombie_osr_set | |
2073 | ceph::mutex zombie_osr_lock = ceph::make_mutex("BlueStore::zombie_osr_lock"); | |
2074 | uint32_t next_sequencer_id = 0; | |
2075 | std::map<coll_t,OpSequencerRef> zombie_osr_set; ///< std::set of OpSequencers for deleted collections | |
2076 | ||
2077 | std::atomic<uint64_t> nid_last = {0}; | |
2078 | std::atomic<uint64_t> nid_max = {0}; | |
2079 | std::atomic<uint64_t> blobid_last = {0}; | |
2080 | std::atomic<uint64_t> blobid_max = {0}; | |
2081 | ||
2082 | ceph::mutex deferred_lock = ceph::make_mutex("BlueStore::deferred_lock"); | |
2083 | ceph::mutex atomic_alloc_and_submit_lock = | |
2084 | ceph::make_mutex("BlueStore::atomic_alloc_and_submit_lock"); | |
2085 | std::atomic<uint64_t> deferred_seq = {0}; | |
2086 | deferred_osr_queue_t deferred_queue; ///< osr's with deferred io pending | |
2087 | std::atomic_int deferred_queue_size = {0}; ///< num txc's queued across all osrs | |
2088 | std::atomic_int deferred_aggressive = {0}; ///< aggressive wakeup of kv thread | |
2089 | Finisher finisher; | |
2090 | utime_t deferred_last_submitted = utime_t(); | |
2091 | ||
2092 | KVSyncThread kv_sync_thread; | |
2093 | ceph::mutex kv_lock = ceph::make_mutex("BlueStore::kv_lock"); | |
2094 | ceph::condition_variable kv_cond; | |
2095 | bool _kv_only = false; | |
2096 | bool kv_sync_started = false; | |
2097 | bool kv_stop = false; | |
2098 | bool kv_finalize_started = false; | |
2099 | bool kv_finalize_stop = false; | |
2100 | std::deque<TransContext*> kv_queue; ///< ready, already submitted | |
2101 | std::deque<TransContext*> kv_queue_unsubmitted; ///< ready, need submit by kv thread | |
2102 | std::deque<TransContext*> kv_committing; ///< currently syncing | |
2103 | std::deque<DeferredBatch*> deferred_done_queue; ///< deferred ios done | |
2104 | bool kv_sync_in_progress = false; | |
2105 | ||
2106 | KVFinalizeThread kv_finalize_thread; | |
2107 | ceph::mutex kv_finalize_lock = ceph::make_mutex("BlueStore::kv_finalize_lock"); | |
2108 | ceph::condition_variable kv_finalize_cond; | |
2109 | std::deque<TransContext*> kv_committing_to_finalize; ///< pending finalization | |
2110 | std::deque<DeferredBatch*> deferred_stable_to_finalize; ///< pending finalization | |
2111 | bool kv_finalize_in_progress = false; | |
2112 | ||
2113 | ZonedCleanerThread zoned_cleaner_thread; | |
2114 | ceph::mutex zoned_cleaner_lock = ceph::make_mutex("BlueStore::zoned_cleaner_lock"); | |
2115 | ceph::condition_variable zoned_cleaner_cond; | |
2116 | bool zoned_cleaner_started = false; | |
2117 | bool zoned_cleaner_stop = false; | |
2118 | std::deque<uint64_t> zoned_cleaner_queue; | |
2119 | ||
2120 | PerfCounters *logger = nullptr; | |
2121 | ||
2122 | std::list<CollectionRef> removed_collections; | |
2123 | ||
2124 | ceph::shared_mutex debug_read_error_lock = | |
2125 | ceph::make_shared_mutex("BlueStore::debug_read_error_lock"); | |
2126 | std::set<ghobject_t> debug_data_error_objects; | |
2127 | std::set<ghobject_t> debug_mdata_error_objects; | |
2128 | ||
2129 | std::atomic<int> csum_type = {Checksummer::CSUM_CRC32C}; | |
2130 | ||
2131 | uint64_t block_size = 0; ///< block size of block device (power of 2) | |
2132 | uint64_t block_mask = 0; ///< mask to get just the block offset | |
2133 | size_t block_size_order = 0; ///< bits to shift to get block size | |
2134 | ||
2135 | uint64_t min_alloc_size; ///< minimum allocation unit (power of 2) | |
2136 | ///< bits for min_alloc_size | |
2137 | uint8_t min_alloc_size_order = 0; | |
2138 | static_assert(std::numeric_limits<uint8_t>::max() > | |
2139 | std::numeric_limits<decltype(min_alloc_size)>::digits, | |
2140 | "not enough bits for min_alloc_size"); | |
2141 | ||
2142 | enum { | |
2143 | // Please preserve the order since it's DB persistent | |
2144 | OMAP_BULK = 0, | |
2145 | OMAP_PER_POOL = 1, | |
2146 | OMAP_PER_PG = 2, | |
2147 | } per_pool_omap = OMAP_BULK; | |
2148 | ||
2149 | ///< maximum allocation unit (power of 2) | |
2150 | std::atomic<uint64_t> max_alloc_size = {0}; | |
2151 | ||
2152 | ///< number threshold for forced deferred writes | |
2153 | std::atomic<int> deferred_batch_ops = {0}; | |
2154 | ||
2155 | ///< size threshold for forced deferred writes | |
2156 | std::atomic<uint64_t> prefer_deferred_size = {0}; | |
2157 | ||
2158 | ///< approx cost per io, in bytes | |
2159 | std::atomic<uint64_t> throttle_cost_per_io = {0}; | |
2160 | ||
2161 | std::atomic<Compressor::CompressionMode> comp_mode = | |
2162 | {Compressor::COMP_NONE}; ///< compression mode | |
2163 | CompressorRef compressor; | |
2164 | std::atomic<uint64_t> comp_min_blob_size = {0}; | |
2165 | std::atomic<uint64_t> comp_max_blob_size = {0}; | |
2166 | ||
2167 | std::atomic<uint64_t> max_blob_size = {0}; ///< maximum blob size | |
2168 | ||
2169 | uint64_t kv_ios = 0; | |
2170 | uint64_t kv_throttle_costs = 0; | |
2171 | ||
2172 | // cache trim control | |
2173 | uint64_t cache_size = 0; ///< total cache size | |
2174 | double cache_meta_ratio = 0; ///< cache ratio dedicated to metadata | |
2175 | double cache_kv_ratio = 0; ///< cache ratio dedicated to kv (e.g., rocksdb) | |
2176 | double cache_kv_onode_ratio = 0; ///< cache ratio dedicated to kv onodes (e.g., rocksdb onode CF) | |
2177 | double cache_data_ratio = 0; ///< cache ratio dedicated to object data | |
2178 | bool cache_autotune = false; ///< cache autotune setting | |
2179 | double cache_autotune_interval = 0; ///< time to wait between cache rebalancing | |
2180 | uint64_t osd_memory_target = 0; ///< OSD memory target when autotuning cache | |
2181 | uint64_t osd_memory_base = 0; ///< OSD base memory when autotuning cache | |
2182 | double osd_memory_expected_fragmentation = 0; ///< expected memory fragmentation | |
2183 | uint64_t osd_memory_cache_min = 0; ///< Min memory to assign when autotuning cache | |
2184 | double osd_memory_cache_resize_interval = 0; ///< Time to wait between cache resizing | |
2185 | double max_defer_interval = 0; ///< Time to wait between last deferred submit | |
2186 | std::atomic<uint32_t> config_changed = {0}; ///< Counter to determine if there is a configuration change. | |
2187 | ||
2188 | typedef std::map<uint64_t, volatile_statfs> osd_pools_map; | |
2189 | ||
2190 | ceph::mutex vstatfs_lock = ceph::make_mutex("BlueStore::vstatfs_lock"); | |
2191 | volatile_statfs vstatfs; | |
2192 | osd_pools_map osd_pools; // protected by vstatfs_lock as well | |
2193 | ||
2194 | bool per_pool_stat_collection = true; | |
2195 | ||
2196 | struct MempoolThread : public Thread { | |
2197 | public: | |
2198 | BlueStore *store; | |
2199 | ||
2200 | ceph::condition_variable cond; | |
2201 | ceph::mutex lock = ceph::make_mutex("BlueStore::MempoolThread::lock"); | |
2202 | bool stop = false; | |
2203 | std::shared_ptr<PriorityCache::PriCache> binned_kv_cache = nullptr; | |
2204 | std::shared_ptr<PriorityCache::PriCache> binned_kv_onode_cache = nullptr; | |
2205 | std::shared_ptr<PriorityCache::Manager> pcm = nullptr; | |
2206 | ||
2207 | struct MempoolCache : public PriorityCache::PriCache { | |
2208 | BlueStore *store; | |
2209 | int64_t cache_bytes[PriorityCache::Priority::LAST+1] = {0}; | |
2210 | int64_t committed_bytes = 0; | |
2211 | double cache_ratio = 0; | |
2212 | ||
2213 | MempoolCache(BlueStore *s) : store(s) {}; | |
2214 | ||
2215 | virtual uint64_t _get_used_bytes() const = 0; | |
2216 | ||
2217 | virtual int64_t request_cache_bytes( | |
2218 | PriorityCache::Priority pri, uint64_t total_cache) const { | |
2219 | int64_t assigned = get_cache_bytes(pri); | |
2220 | ||
2221 | switch (pri) { | |
2222 | // All cache items are currently shoved into the PRI1 priority | |
2223 | case PriorityCache::Priority::PRI1: | |
2224 | { | |
2225 | int64_t request = _get_used_bytes(); | |
2226 | return(request > assigned) ? request - assigned : 0; | |
2227 | } | |
2228 | default: | |
2229 | break; | |
2230 | } | |
2231 | return -EOPNOTSUPP; | |
2232 | } | |
2233 | ||
2234 | virtual int64_t get_cache_bytes(PriorityCache::Priority pri) const { | |
2235 | return cache_bytes[pri]; | |
2236 | } | |
2237 | virtual int64_t get_cache_bytes() const { | |
2238 | int64_t total = 0; | |
2239 | ||
2240 | for (int i = 0; i < PriorityCache::Priority::LAST + 1; i++) { | |
2241 | PriorityCache::Priority pri = static_cast<PriorityCache::Priority>(i); | |
2242 | total += get_cache_bytes(pri); | |
2243 | } | |
2244 | return total; | |
2245 | } | |
2246 | virtual void set_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
2247 | cache_bytes[pri] = bytes; | |
2248 | } | |
2249 | virtual void add_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
2250 | cache_bytes[pri] += bytes; | |
2251 | } | |
2252 | virtual int64_t commit_cache_size(uint64_t total_cache) { | |
2253 | committed_bytes = PriorityCache::get_chunk( | |
2254 | get_cache_bytes(), total_cache); | |
2255 | return committed_bytes; | |
2256 | } | |
2257 | virtual int64_t get_committed_size() const { | |
2258 | return committed_bytes; | |
2259 | } | |
2260 | virtual double get_cache_ratio() const { | |
2261 | return cache_ratio; | |
2262 | } | |
2263 | virtual void set_cache_ratio(double ratio) { | |
2264 | cache_ratio = ratio; | |
2265 | } | |
2266 | virtual std::string get_cache_name() const = 0; | |
2267 | }; | |
2268 | ||
2269 | struct MetaCache : public MempoolCache { | |
2270 | MetaCache(BlueStore *s) : MempoolCache(s) {}; | |
2271 | ||
2272 | virtual uint64_t _get_used_bytes() const { | |
2273 | return mempool::bluestore_Buffer::allocated_bytes() + | |
2274 | mempool::bluestore_Blob::allocated_bytes() + | |
2275 | mempool::bluestore_Extent::allocated_bytes() + | |
2276 | mempool::bluestore_cache_meta::allocated_bytes() + | |
2277 | mempool::bluestore_cache_other::allocated_bytes() + | |
2278 | mempool::bluestore_cache_onode::allocated_bytes() + | |
2279 | mempool::bluestore_SharedBlob::allocated_bytes() + | |
2280 | mempool::bluestore_inline_bl::allocated_bytes(); | |
2281 | } | |
2282 | ||
2283 | virtual std::string get_cache_name() const { | |
2284 | return "BlueStore Meta Cache"; | |
2285 | } | |
2286 | ||
2287 | uint64_t _get_num_onodes() const { | |
2288 | uint64_t onode_num = | |
2289 | mempool::bluestore_cache_onode::allocated_items(); | |
2290 | return (2 > onode_num) ? 2 : onode_num; | |
2291 | } | |
2292 | ||
2293 | double get_bytes_per_onode() const { | |
2294 | return (double)_get_used_bytes() / (double)_get_num_onodes(); | |
2295 | } | |
2296 | }; | |
2297 | std::shared_ptr<MetaCache> meta_cache; | |
2298 | ||
2299 | struct DataCache : public MempoolCache { | |
2300 | DataCache(BlueStore *s) : MempoolCache(s) {}; | |
2301 | ||
2302 | virtual uint64_t _get_used_bytes() const { | |
2303 | uint64_t bytes = 0; | |
2304 | for (auto i : store->buffer_cache_shards) { | |
2305 | bytes += i->_get_bytes(); | |
2306 | } | |
2307 | return bytes; | |
2308 | } | |
2309 | virtual std::string get_cache_name() const { | |
2310 | return "BlueStore Data Cache"; | |
2311 | } | |
2312 | }; | |
2313 | std::shared_ptr<DataCache> data_cache; | |
2314 | ||
2315 | public: | |
2316 | explicit MempoolThread(BlueStore *s) | |
2317 | : store(s), | |
2318 | meta_cache(new MetaCache(s)), | |
2319 | data_cache(new DataCache(s)) {} | |
2320 | ||
2321 | void *entry() override; | |
2322 | void init() { | |
2323 | ceph_assert(stop == false); | |
2324 | create("bstore_mempool"); | |
2325 | } | |
2326 | void shutdown() { | |
2327 | lock.lock(); | |
2328 | stop = true; | |
2329 | cond.notify_all(); | |
2330 | lock.unlock(); | |
2331 | join(); | |
2332 | } | |
2333 | ||
2334 | private: | |
2335 | void _adjust_cache_settings(); | |
2336 | void _update_cache_settings(); | |
2337 | void _resize_shards(bool interval_stats); | |
2338 | } mempool_thread; | |
2339 | ||
2340 | #ifdef WITH_BLKIN | |
2341 | ZTracer::Endpoint trace_endpoint {"0.0.0.0", 0, "BlueStore"}; | |
2342 | #endif | |
2343 | ||
2344 | // -------------------------------------------------------- | |
2345 | // private methods | |
2346 | ||
2347 | void _init_logger(); | |
2348 | void _shutdown_logger(); | |
2349 | int _reload_logger(); | |
2350 | ||
2351 | int _open_path(); | |
2352 | void _close_path(); | |
2353 | int _open_fsid(bool create); | |
2354 | int _lock_fsid(); | |
2355 | int _read_fsid(uuid_d *f); | |
2356 | int _write_fsid(); | |
2357 | void _close_fsid(); | |
2358 | void _set_alloc_sizes(); | |
2359 | void _set_blob_size(); | |
2360 | void _set_finisher_num(); | |
2361 | void _set_per_pool_omap(); | |
2362 | void _update_osd_memory_options(); | |
2363 | ||
2364 | int _open_bdev(bool create); | |
2365 | // Verifies if disk space is enough for reserved + min bluefs | |
2366 | // and alters the latter if needed. | |
2367 | // Depends on min_alloc_size hence should be called after | |
2368 | // its initialization (and outside of _open_bdev) | |
2369 | void _validate_bdev(); | |
2370 | void _close_bdev(); | |
2371 | ||
2372 | int _minimal_open_bluefs(bool create); | |
2373 | void _minimal_close_bluefs(); | |
2374 | int _open_bluefs(bool create, bool read_only); | |
2375 | void _close_bluefs(bool cold_close); | |
2376 | ||
2377 | int _is_bluefs(bool create, bool* ret); | |
2378 | /* | |
2379 | * opens both DB and dependant super_meta, FreelistManager and allocator | |
2380 | * in the proper order | |
2381 | */ | |
2382 | int _open_db_and_around(bool read_only, bool to_repair = false); | |
2383 | void _close_db_and_around(bool read_only); | |
2384 | ||
2385 | int _prepare_db_environment(bool create, bool read_only, | |
2386 | std::string* kv_dir, std::string* kv_backend); | |
2387 | ||
2388 | /* | |
2389 | * @warning to_repair_db means that we open this db to repair it, will not | |
2390 | * hold the rocksdb's file lock. | |
2391 | */ | |
2392 | int _open_db(bool create, | |
2393 | bool to_repair_db=false, | |
2394 | bool read_only = false); | |
2395 | void _close_db(bool read_only); | |
2396 | int _open_fm(KeyValueDB::Transaction t, bool read_only); | |
2397 | void _close_fm(); | |
2398 | int _write_out_fm_meta(uint64_t target_size); | |
2399 | int _create_alloc(); | |
2400 | int _init_alloc(); | |
2401 | void _close_alloc(); | |
2402 | int _open_collections(); | |
2403 | void _fsck_collections(int64_t* errors); | |
2404 | void _close_collections(); | |
2405 | ||
2406 | int _setup_block_symlink_or_file(std::string name, std::string path, uint64_t size, | |
2407 | bool create); | |
2408 | ||
2409 | // Functions related to zoned storage. | |
2410 | uint64_t _zoned_piggyback_device_parameters_onto(uint64_t min_alloc_size); | |
2411 | int _zoned_check_config_settings(); | |
2412 | void _zoned_update_cleaning_metadata(TransContext *txc); | |
2413 | std::string _zoned_get_prefix(uint64_t offset); | |
2414 | ||
2415 | public: | |
2416 | utime_t get_deferred_last_submitted() { | |
2417 | std::lock_guard l(deferred_lock); | |
2418 | return deferred_last_submitted; | |
2419 | } | |
2420 | ||
2421 | static int _write_bdev_label(CephContext* cct, | |
2422 | std::string path, bluestore_bdev_label_t label); | |
2423 | static int _read_bdev_label(CephContext* cct, std::string path, | |
2424 | bluestore_bdev_label_t *label); | |
2425 | private: | |
2426 | int _check_or_set_bdev_label(std::string path, uint64_t size, std::string desc, | |
2427 | bool create); | |
2428 | int _set_bdev_label_size(const string& path, uint64_t size); | |
2429 | ||
2430 | int _open_super_meta(); | |
2431 | ||
2432 | void _open_statfs(); | |
2433 | void _get_statfs_overall(struct store_statfs_t *buf); | |
2434 | ||
2435 | void _dump_alloc_on_failure(); | |
2436 | ||
2437 | CollectionRef _get_collection(const coll_t& cid); | |
2438 | void _queue_reap_collection(CollectionRef& c); | |
2439 | void _reap_collections(); | |
2440 | void _update_cache_logger(); | |
2441 | ||
2442 | void _assign_nid(TransContext *txc, OnodeRef o); | |
2443 | uint64_t _assign_blobid(TransContext *txc); | |
2444 | ||
2445 | template <int LogLevelV> | |
2446 | friend void _dump_onode(CephContext *cct, const Onode& o); | |
2447 | template <int LogLevelV> | |
2448 | friend void _dump_extent_map(CephContext *cct, const ExtentMap& em); | |
2449 | template <int LogLevelV> | |
2450 | friend void _dump_transaction(CephContext *cct, Transaction *t); | |
2451 | ||
2452 | TransContext *_txc_create(Collection *c, OpSequencer *osr, | |
2453 | std::list<Context*> *on_commits, | |
2454 | TrackedOpRef osd_op=TrackedOpRef()); | |
2455 | void _txc_update_store_statfs(TransContext *txc); | |
2456 | void _txc_add_transaction(TransContext *txc, Transaction *t); | |
2457 | void _txc_calc_cost(TransContext *txc); | |
2458 | void _txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t); | |
2459 | void _txc_state_proc(TransContext *txc); | |
2460 | void _txc_aio_submit(TransContext *txc); | |
2461 | public: | |
2462 | void txc_aio_finish(void *p) { | |
2463 | _txc_state_proc(static_cast<TransContext*>(p)); | |
2464 | } | |
2465 | private: | |
2466 | void _txc_finish_io(TransContext *txc); | |
2467 | void _txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t); | |
2468 | void _txc_apply_kv(TransContext *txc, bool sync_submit_transaction); | |
2469 | void _txc_committed_kv(TransContext *txc); | |
2470 | void _txc_finish(TransContext *txc); | |
2471 | void _txc_release_alloc(TransContext *txc); | |
2472 | ||
2473 | void _osr_attach(Collection *c); | |
2474 | void _osr_register_zombie(OpSequencer *osr); | |
2475 | void _osr_drain(OpSequencer *osr); | |
2476 | void _osr_drain_preceding(TransContext *txc); | |
2477 | void _osr_drain_all(); | |
2478 | ||
2479 | void _kv_start(); | |
2480 | void _kv_stop(); | |
2481 | void _kv_sync_thread(); | |
2482 | void _kv_finalize_thread(); | |
2483 | ||
2484 | void _zoned_cleaner_start(); | |
2485 | void _zoned_cleaner_stop(); | |
2486 | void _zoned_cleaner_thread(); | |
2487 | void _zoned_clean_zone(uint64_t zone_num); | |
2488 | ||
2489 | bluestore_deferred_op_t *_get_deferred_op(TransContext *txc); | |
2490 | void _deferred_queue(TransContext *txc); | |
2491 | public: | |
2492 | void deferred_try_submit(); | |
2493 | private: | |
2494 | void _deferred_submit_unlock(OpSequencer *osr); | |
2495 | void _deferred_aio_finish(OpSequencer *osr); | |
2496 | int _deferred_replay(); | |
2497 | ||
2498 | public: | |
2499 | using mempool_dynamic_bitset = | |
2500 | boost::dynamic_bitset<uint64_t, | |
2501 | mempool::bluestore_fsck::pool_allocator<uint64_t>>; | |
2502 | using per_pool_statfs = | |
2503 | mempool::bluestore_fsck::map<uint64_t, store_statfs_t>; | |
2504 | ||
2505 | enum FSCKDepth { | |
2506 | FSCK_REGULAR, | |
2507 | FSCK_DEEP, | |
2508 | FSCK_SHALLOW | |
2509 | }; | |
2510 | enum { | |
2511 | MAX_FSCK_ERROR_LINES = 100, | |
2512 | }; | |
2513 | ||
2514 | private: | |
2515 | int _fsck_check_extents( | |
2516 | const coll_t& cid, | |
2517 | const ghobject_t& oid, | |
2518 | const PExtentVector& extents, | |
2519 | bool compressed, | |
2520 | mempool_dynamic_bitset &used_blocks, | |
2521 | uint64_t granularity, | |
2522 | BlueStoreRepairer* repairer, | |
2523 | store_statfs_t& expected_statfs, | |
2524 | FSCKDepth depth); | |
2525 | ||
2526 | void _fsck_check_pool_statfs( | |
2527 | per_pool_statfs& expected_pool_statfs, | |
2528 | int64_t& errors, | |
2529 | int64_t &warnings, | |
2530 | BlueStoreRepairer* repairer); | |
2531 | ||
2532 | int _fsck(FSCKDepth depth, bool repair); | |
2533 | int _fsck_on_open(BlueStore::FSCKDepth depth, bool repair); | |
2534 | ||
2535 | void _buffer_cache_write( | |
2536 | TransContext *txc, | |
2537 | BlobRef b, | |
2538 | uint64_t offset, | |
2539 | ceph::buffer::list& bl, | |
2540 | unsigned flags) { | |
2541 | b->shared_blob->bc.write(b->shared_blob->get_cache(), txc->seq, offset, bl, | |
2542 | flags); | |
2543 | txc->shared_blobs_written.insert(b->shared_blob); | |
2544 | } | |
2545 | ||
2546 | int _collection_list( | |
2547 | Collection *c, const ghobject_t& start, const ghobject_t& end, | |
2548 | int max, bool legacy, std::vector<ghobject_t> *ls, ghobject_t *next); | |
2549 | ||
2550 | template <typename T, typename F> | |
2551 | T select_option(const std::string& opt_name, T val1, F f) { | |
2552 | //NB: opt_name reserved for future use | |
2553 | boost::optional<T> val2 = f(); | |
2554 | if (val2) { | |
2555 | return *val2; | |
2556 | } | |
2557 | return val1; | |
2558 | } | |
2559 | ||
2560 | void _apply_padding(uint64_t head_pad, | |
2561 | uint64_t tail_pad, | |
2562 | ceph::buffer::list& padded); | |
2563 | ||
2564 | void _record_onode(OnodeRef &o, KeyValueDB::Transaction &txn); | |
2565 | ||
2566 | // -- ondisk version --- | |
2567 | public: | |
2568 | const int32_t latest_ondisk_format = 4; ///< our version | |
2569 | const int32_t min_readable_ondisk_format = 1; ///< what we can read | |
2570 | const int32_t min_compat_ondisk_format = 3; ///< who can read us | |
2571 | ||
2572 | private: | |
2573 | int32_t ondisk_format = 0; ///< value detected on mount | |
2574 | ||
2575 | int _upgrade_super(); ///< upgrade (called during open_super) | |
2576 | uint64_t _get_ondisk_reserved() const; | |
2577 | void _prepare_ondisk_format_super(KeyValueDB::Transaction& t); | |
2578 | ||
2579 | // --- public interface --- | |
2580 | public: | |
2581 | BlueStore(CephContext *cct, const std::string& path); | |
2582 | BlueStore(CephContext *cct, const std::string& path, uint64_t min_alloc_size); // Ctor for UT only | |
2583 | ~BlueStore() override; | |
2584 | ||
2585 | std::string get_type() override { | |
2586 | return "bluestore"; | |
2587 | } | |
2588 | ||
2589 | bool needs_journal() override { return false; }; | |
2590 | bool wants_journal() override { return false; }; | |
2591 | bool allows_journal() override { return false; }; | |
2592 | ||
2593 | uint64_t get_min_alloc_size() const override { | |
2594 | return min_alloc_size; | |
2595 | } | |
2596 | ||
2597 | int get_devices(std::set<std::string> *ls) override; | |
2598 | ||
2599 | bool is_rotational() override; | |
2600 | bool is_journal_rotational() override; | |
2601 | ||
2602 | std::string get_default_device_class() override { | |
2603 | std::string device_class; | |
2604 | std::map<std::string, std::string> metadata; | |
2605 | collect_metadata(&metadata); | |
2606 | auto it = metadata.find("bluestore_bdev_type"); | |
2607 | if (it != metadata.end()) { | |
2608 | device_class = it->second; | |
2609 | } | |
2610 | return device_class; | |
2611 | } | |
2612 | ||
2613 | int get_numa_node( | |
2614 | int *numa_node, | |
2615 | std::set<int> *nodes, | |
2616 | std::set<std::string> *failed) override; | |
2617 | ||
2618 | static int get_block_device_fsid(CephContext* cct, const std::string& path, | |
2619 | uuid_d *fsid); | |
2620 | ||
2621 | bool test_mount_in_use() override; | |
2622 | ||
2623 | private: | |
2624 | int _mount(); | |
2625 | public: | |
2626 | int mount() override { | |
2627 | return _mount(); | |
2628 | } | |
2629 | int umount() override; | |
2630 | ||
2631 | int open_db_environment(KeyValueDB **pdb, bool to_repair); | |
2632 | int close_db_environment(); | |
2633 | ||
2634 | int write_meta(const std::string& key, const std::string& value) override; | |
2635 | int read_meta(const std::string& key, std::string *value) override; | |
2636 | ||
2637 | // open in read-only and limited mode | |
2638 | int cold_open(); | |
2639 | int cold_close(); | |
2640 | ||
2641 | int fsck(bool deep) override { | |
2642 | return _fsck(deep ? FSCK_DEEP : FSCK_REGULAR, false); | |
2643 | } | |
2644 | int repair(bool deep) override { | |
2645 | return _fsck(deep ? FSCK_DEEP : FSCK_REGULAR, true); | |
2646 | } | |
2647 | int quick_fix() override { | |
2648 | return _fsck(FSCK_SHALLOW, true); | |
2649 | } | |
2650 | ||
2651 | void set_cache_shards(unsigned num) override; | |
2652 | void dump_cache_stats(ceph::Formatter *f) override { | |
2653 | int onode_count = 0, buffers_bytes = 0; | |
2654 | for (auto i: onode_cache_shards) { | |
2655 | onode_count += i->_get_num(); | |
2656 | } | |
2657 | for (auto i: buffer_cache_shards) { | |
2658 | buffers_bytes += i->_get_bytes(); | |
2659 | } | |
2660 | f->dump_int("bluestore_onode", onode_count); | |
2661 | f->dump_int("bluestore_buffers", buffers_bytes); | |
2662 | } | |
2663 | void dump_cache_stats(std::ostream& ss) override { | |
2664 | int onode_count = 0, buffers_bytes = 0; | |
2665 | for (auto i: onode_cache_shards) { | |
2666 | onode_count += i->_get_num(); | |
2667 | } | |
2668 | for (auto i: buffer_cache_shards) { | |
2669 | buffers_bytes += i->_get_bytes(); | |
2670 | } | |
2671 | ss << "bluestore_onode: " << onode_count; | |
2672 | ss << "bluestore_buffers: " << buffers_bytes; | |
2673 | } | |
2674 | ||
2675 | int validate_hobject_key(const hobject_t &obj) const override { | |
2676 | return 0; | |
2677 | } | |
2678 | unsigned get_max_attr_name_length() override { | |
2679 | return 256; // arbitrary; there is no real limit internally | |
2680 | } | |
2681 | ||
2682 | int mkfs() override; | |
2683 | int mkjournal() override { | |
2684 | return 0; | |
2685 | } | |
2686 | ||
2687 | void get_db_statistics(ceph::Formatter *f) override; | |
2688 | void generate_db_histogram(ceph::Formatter *f) override; | |
2689 | void _shutdown_cache(); | |
2690 | int flush_cache(std::ostream *os = NULL) override; | |
2691 | void dump_perf_counters(ceph::Formatter *f) override { | |
2692 | f->open_object_section("perf_counters"); | |
2693 | logger->dump_formatted(f, false); | |
2694 | f->close_section(); | |
2695 | } | |
2696 | ||
2697 | int add_new_bluefs_device(int id, const std::string& path); | |
2698 | int migrate_to_existing_bluefs_device(const std::set<int>& devs_source, | |
2699 | int id); | |
2700 | int migrate_to_new_bluefs_device(const std::set<int>& devs_source, | |
2701 | int id, | |
2702 | const std::string& path); | |
2703 | int expand_devices(std::ostream& out); | |
2704 | std::string get_device_path(unsigned id); | |
2705 | ||
2706 | int dump_bluefs_sizes(ostream& out); | |
2707 | ||
2708 | public: | |
2709 | int statfs(struct store_statfs_t *buf, | |
2710 | osd_alert_list_t* alerts = nullptr) override; | |
2711 | int pool_statfs(uint64_t pool_id, struct store_statfs_t *buf, | |
2712 | bool *per_pool_omap) override; | |
2713 | ||
2714 | void collect_metadata(std::map<std::string,std::string> *pm) override; | |
2715 | ||
2716 | bool exists(CollectionHandle &c, const ghobject_t& oid) override; | |
2717 | int set_collection_opts( | |
2718 | CollectionHandle& c, | |
2719 | const pool_opts_t& opts) override; | |
2720 | int stat( | |
2721 | CollectionHandle &c, | |
2722 | const ghobject_t& oid, | |
2723 | struct stat *st, | |
2724 | bool allow_eio = false) override; | |
2725 | int read( | |
2726 | CollectionHandle &c, | |
2727 | const ghobject_t& oid, | |
2728 | uint64_t offset, | |
2729 | size_t len, | |
2730 | ceph::buffer::list& bl, | |
2731 | uint32_t op_flags = 0) override; | |
2732 | ||
2733 | private: | |
2734 | ||
2735 | // -------------------------------------------------------- | |
2736 | // intermediate data structures used while reading | |
2737 | struct region_t { | |
2738 | uint64_t logical_offset; | |
2739 | uint64_t blob_xoffset; //region offset within the blob | |
2740 | uint64_t length; | |
2741 | ||
2742 | // used later in read process | |
2743 | uint64_t front = 0; | |
2744 | ||
2745 | region_t(uint64_t offset, uint64_t b_offs, uint64_t len, uint64_t front = 0) | |
2746 | : logical_offset(offset), | |
2747 | blob_xoffset(b_offs), | |
2748 | length(len), | |
2749 | front(front){} | |
2750 | region_t(const region_t& from) | |
2751 | : logical_offset(from.logical_offset), | |
2752 | blob_xoffset(from.blob_xoffset), | |
2753 | length(from.length), | |
2754 | front(from.front){} | |
2755 | ||
2756 | friend std::ostream& operator<<(std::ostream& out, const region_t& r) { | |
2757 | return out << "0x" << std::hex << r.logical_offset << ":" | |
2758 | << r.blob_xoffset << "~" << r.length << std::dec; | |
2759 | } | |
2760 | }; | |
2761 | ||
2762 | // merged blob read request | |
2763 | struct read_req_t { | |
2764 | uint64_t r_off = 0; | |
2765 | uint64_t r_len = 0; | |
2766 | ceph::buffer::list bl; | |
2767 | std::list<region_t> regs; // original read regions | |
2768 | ||
2769 | read_req_t(uint64_t off, uint64_t len) : r_off(off), r_len(len) {} | |
2770 | ||
2771 | friend std::ostream& operator<<(std::ostream& out, const read_req_t& r) { | |
2772 | out << "{<0x" << std::hex << r.r_off << ", 0x" << r.r_len << "> : ["; | |
2773 | for (const auto& reg : r.regs) | |
2774 | out << reg; | |
2775 | return out << "]}" << std::dec; | |
2776 | } | |
2777 | }; | |
2778 | ||
2779 | typedef std::list<read_req_t> regions2read_t; | |
2780 | typedef std::map<BlueStore::BlobRef, regions2read_t> blobs2read_t; | |
2781 | ||
2782 | void _read_cache( | |
2783 | OnodeRef o, | |
2784 | uint64_t offset, | |
2785 | size_t length, | |
2786 | int read_cache_policy, | |
2787 | ready_regions_t& ready_regions, | |
2788 | blobs2read_t& blobs2read); | |
2789 | ||
2790 | ||
2791 | int _prepare_read_ioc( | |
2792 | blobs2read_t& blobs2read, | |
2793 | std::vector<ceph::buffer::list>* compressed_blob_bls, | |
2794 | IOContext* ioc); | |
2795 | ||
2796 | int _generate_read_result_bl( | |
2797 | OnodeRef o, | |
2798 | uint64_t offset, | |
2799 | size_t length, | |
2800 | ready_regions_t& ready_regions, | |
2801 | std::vector<ceph::buffer::list>& compressed_blob_bls, | |
2802 | blobs2read_t& blobs2read, | |
2803 | bool buffered, | |
2804 | bool* csum_error, | |
2805 | ceph::buffer::list& bl); | |
2806 | ||
2807 | int _do_read( | |
2808 | Collection *c, | |
2809 | OnodeRef o, | |
2810 | uint64_t offset, | |
2811 | size_t len, | |
2812 | ceph::buffer::list& bl, | |
2813 | uint32_t op_flags = 0, | |
2814 | uint64_t retry_count = 0); | |
2815 | ||
2816 | int _do_readv( | |
2817 | Collection *c, | |
2818 | OnodeRef o, | |
2819 | const interval_set<uint64_t>& m, | |
2820 | ceph::buffer::list& bl, | |
2821 | uint32_t op_flags = 0, | |
2822 | uint64_t retry_count = 0); | |
2823 | ||
2824 | int _fiemap(CollectionHandle &c_, const ghobject_t& oid, | |
2825 | uint64_t offset, size_t len, interval_set<uint64_t>& destset); | |
2826 | public: | |
2827 | int fiemap(CollectionHandle &c, const ghobject_t& oid, | |
2828 | uint64_t offset, size_t len, ceph::buffer::list& bl) override; | |
2829 | int fiemap(CollectionHandle &c, const ghobject_t& oid, | |
2830 | uint64_t offset, size_t len, std::map<uint64_t, uint64_t>& destmap) override; | |
2831 | ||
2832 | int readv( | |
2833 | CollectionHandle &c_, | |
2834 | const ghobject_t& oid, | |
2835 | interval_set<uint64_t>& m, | |
2836 | ceph::buffer::list& bl, | |
2837 | uint32_t op_flags) override; | |
2838 | ||
2839 | int dump_onode(CollectionHandle &c, const ghobject_t& oid, | |
2840 | const std::string& section_name, ceph::Formatter *f) override; | |
2841 | ||
2842 | int getattr(CollectionHandle &c, const ghobject_t& oid, const char *name, | |
2843 | ceph::buffer::ptr& value) override; | |
2844 | ||
2845 | int getattrs(CollectionHandle &c, const ghobject_t& oid, | |
2846 | std::map<std::string,ceph::buffer::ptr>& aset) override; | |
2847 | ||
2848 | int list_collections(std::vector<coll_t>& ls) override; | |
2849 | ||
2850 | CollectionHandle open_collection(const coll_t &c) override; | |
2851 | CollectionHandle create_new_collection(const coll_t& cid) override; | |
2852 | void set_collection_commit_queue(const coll_t& cid, | |
2853 | ContextQueue *commit_queue) override; | |
2854 | ||
2855 | bool collection_exists(const coll_t& c) override; | |
2856 | int collection_empty(CollectionHandle& c, bool *empty) override; | |
2857 | int collection_bits(CollectionHandle& c) override; | |
2858 | ||
2859 | int collection_list(CollectionHandle &c, | |
2860 | const ghobject_t& start, | |
2861 | const ghobject_t& end, | |
2862 | int max, | |
2863 | std::vector<ghobject_t> *ls, ghobject_t *next) override; | |
2864 | ||
2865 | int collection_list_legacy(CollectionHandle &c, | |
2866 | const ghobject_t& start, | |
2867 | const ghobject_t& end, | |
2868 | int max, | |
2869 | std::vector<ghobject_t> *ls, | |
2870 | ghobject_t *next) override; | |
2871 | ||
2872 | int omap_get( | |
2873 | CollectionHandle &c, ///< [in] Collection containing oid | |
2874 | const ghobject_t &oid, ///< [in] Object containing omap | |
2875 | ceph::buffer::list *header, ///< [out] omap header | |
2876 | std::map<std::string, ceph::buffer::list> *out /// < [out] Key to value map | |
2877 | ) override; | |
2878 | int _omap_get( | |
2879 | Collection *c, ///< [in] Collection containing oid | |
2880 | const ghobject_t &oid, ///< [in] Object containing omap | |
2881 | ceph::buffer::list *header, ///< [out] omap header | |
2882 | std::map<std::string, ceph::buffer::list> *out /// < [out] Key to value map | |
2883 | ); | |
2884 | int _onode_omap_get( | |
2885 | const OnodeRef &o, ///< [in] Object containing omap | |
2886 | ceph::buffer::list *header, ///< [out] omap header | |
2887 | std::map<std::string, ceph::buffer::list> *out /// < [out] Key to value map | |
2888 | ); | |
2889 | ||
2890 | ||
2891 | /// Get omap header | |
2892 | int omap_get_header( | |
2893 | CollectionHandle &c, ///< [in] Collection containing oid | |
2894 | const ghobject_t &oid, ///< [in] Object containing omap | |
2895 | ceph::buffer::list *header, ///< [out] omap header | |
2896 | bool allow_eio = false ///< [in] don't assert on eio | |
2897 | ) override; | |
2898 | ||
2899 | /// Get keys defined on oid | |
2900 | int omap_get_keys( | |
2901 | CollectionHandle &c, ///< [in] Collection containing oid | |
2902 | const ghobject_t &oid, ///< [in] Object containing omap | |
2903 | std::set<std::string> *keys ///< [out] Keys defined on oid | |
2904 | ) override; | |
2905 | ||
2906 | /// Get key values | |
2907 | int omap_get_values( | |
2908 | CollectionHandle &c, ///< [in] Collection containing oid | |
2909 | const ghobject_t &oid, ///< [in] Object containing omap | |
2910 | const std::set<std::string> &keys, ///< [in] Keys to get | |
2911 | std::map<std::string, ceph::buffer::list> *out ///< [out] Returned keys and values | |
2912 | ) override; | |
2913 | ||
2914 | #ifdef WITH_SEASTAR | |
2915 | int omap_get_values( | |
2916 | CollectionHandle &c, ///< [in] Collection containing oid | |
2917 | const ghobject_t &oid, ///< [in] Object containing omap | |
2918 | const std::optional<std::string> &start_after, ///< [in] Keys to get | |
2919 | std::map<std::string, ceph::buffer::list> *out ///< [out] Returned keys and values | |
2920 | ) override; | |
2921 | #endif | |
2922 | ||
2923 | /// Filters keys into out which are defined on oid | |
2924 | int omap_check_keys( | |
2925 | CollectionHandle &c, ///< [in] Collection containing oid | |
2926 | const ghobject_t &oid, ///< [in] Object containing omap | |
2927 | const std::set<std::string> &keys, ///< [in] Keys to check | |
2928 | std::set<std::string> *out ///< [out] Subset of keys defined on oid | |
2929 | ) override; | |
2930 | ||
2931 | ObjectMap::ObjectMapIterator get_omap_iterator( | |
2932 | CollectionHandle &c, ///< [in] collection | |
2933 | const ghobject_t &oid ///< [in] object | |
2934 | ) override; | |
2935 | ||
2936 | void set_fsid(uuid_d u) override { | |
2937 | fsid = u; | |
2938 | } | |
2939 | uuid_d get_fsid() override { | |
2940 | return fsid; | |
2941 | } | |
2942 | ||
2943 | uint64_t estimate_objects_overhead(uint64_t num_objects) override { | |
2944 | return num_objects * 300; //assuming per-object overhead is 300 bytes | |
2945 | } | |
2946 | ||
2947 | struct BSPerfTracker { | |
2948 | PerfCounters::avg_tracker<uint64_t> os_commit_latency_ns; | |
2949 | PerfCounters::avg_tracker<uint64_t> os_apply_latency_ns; | |
2950 | ||
2951 | objectstore_perf_stat_t get_cur_stats() const { | |
2952 | objectstore_perf_stat_t ret; | |
2953 | ret.os_commit_latency_ns = os_commit_latency_ns.current_avg(); | |
2954 | ret.os_apply_latency_ns = os_apply_latency_ns.current_avg(); | |
2955 | return ret; | |
2956 | } | |
2957 | ||
2958 | void update_from_perfcounters(PerfCounters &logger); | |
2959 | } perf_tracker; | |
2960 | ||
2961 | objectstore_perf_stat_t get_cur_stats() override { | |
2962 | perf_tracker.update_from_perfcounters(*logger); | |
2963 | return perf_tracker.get_cur_stats(); | |
2964 | } | |
2965 | const PerfCounters* get_perf_counters() const override { | |
2966 | return logger; | |
2967 | } | |
2968 | const PerfCounters* get_bluefs_perf_counters() const { | |
2969 | return bluefs->get_perf_counters(); | |
2970 | } | |
2971 | KeyValueDB* get_kv() { | |
2972 | return db; | |
2973 | } | |
2974 | ||
2975 | int queue_transactions( | |
2976 | CollectionHandle& ch, | |
2977 | std::vector<Transaction>& tls, | |
2978 | TrackedOpRef op = TrackedOpRef(), | |
2979 | ThreadPool::TPHandle *handle = NULL) override; | |
2980 | ||
2981 | // error injection | |
2982 | void inject_data_error(const ghobject_t& o) override { | |
2983 | std::unique_lock l(debug_read_error_lock); | |
2984 | debug_data_error_objects.insert(o); | |
2985 | } | |
2986 | void inject_mdata_error(const ghobject_t& o) override { | |
2987 | std::unique_lock l(debug_read_error_lock); | |
2988 | debug_mdata_error_objects.insert(o); | |
2989 | } | |
2990 | ||
2991 | /// methods to inject various errors fsck can repair | |
2992 | void inject_broken_shared_blob_key(const std::string& key, | |
2993 | const ceph::buffer::list& bl); | |
2994 | void inject_leaked(uint64_t len); | |
2995 | void inject_false_free(coll_t cid, ghobject_t oid); | |
2996 | void inject_statfs(const std::string& key, const store_statfs_t& new_statfs); | |
2997 | void inject_global_statfs(const store_statfs_t& new_statfs); | |
2998 | void inject_misreference(coll_t cid1, ghobject_t oid1, | |
2999 | coll_t cid2, ghobject_t oid2, | |
3000 | uint64_t offset); | |
3001 | void inject_zombie_spanning_blob(coll_t cid, ghobject_t oid, int16_t blob_id); | |
3002 | // resets global per_pool_omap in DB | |
3003 | void inject_legacy_omap(); | |
3004 | // resets per_pool_omap | pgmeta_omap for onode | |
3005 | void inject_legacy_omap(coll_t cid, ghobject_t oid); | |
3006 | ||
3007 | void compact() override { | |
3008 | ceph_assert(db); | |
3009 | db->compact(); | |
3010 | } | |
3011 | bool has_builtin_csum() const override { | |
3012 | return true; | |
3013 | } | |
3014 | ||
3015 | inline void log_latency(const char* name, | |
3016 | int idx, | |
3017 | const ceph::timespan& lat, | |
3018 | double lat_threshold, | |
3019 | const char* info = "") const; | |
3020 | ||
3021 | inline void log_latency_fn(const char* name, | |
3022 | int idx, | |
3023 | const ceph::timespan& lat, | |
3024 | double lat_threshold, | |
3025 | std::function<std::string (const ceph::timespan& lat)> fn) const; | |
3026 | ||
3027 | private: | |
3028 | bool _debug_data_eio(const ghobject_t& o) { | |
3029 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
3030 | return false; | |
3031 | } | |
3032 | std::shared_lock l(debug_read_error_lock); | |
3033 | return debug_data_error_objects.count(o); | |
3034 | } | |
3035 | bool _debug_mdata_eio(const ghobject_t& o) { | |
3036 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
3037 | return false; | |
3038 | } | |
3039 | std::shared_lock l(debug_read_error_lock); | |
3040 | return debug_mdata_error_objects.count(o); | |
3041 | } | |
3042 | void _debug_obj_on_delete(const ghobject_t& o) { | |
3043 | if (cct->_conf->bluestore_debug_inject_read_err) { | |
3044 | std::unique_lock l(debug_read_error_lock); | |
3045 | debug_data_error_objects.erase(o); | |
3046 | debug_mdata_error_objects.erase(o); | |
3047 | } | |
3048 | } | |
3049 | private: | |
3050 | ceph::mutex qlock = ceph::make_mutex("BlueStore::Alerts::qlock"); | |
3051 | std::string failed_cmode; | |
3052 | std::set<std::string> failed_compressors; | |
3053 | std::string spillover_alert; | |
3054 | std::string legacy_statfs_alert; | |
3055 | std::string no_per_pool_omap_alert; | |
3056 | std::string no_per_pg_omap_alert; | |
3057 | std::string disk_size_mismatch_alert; | |
3058 | std::string spurious_read_errors_alert; | |
3059 | ||
3060 | void _log_alerts(osd_alert_list_t& alerts); | |
3061 | bool _set_compression_alert(bool cmode, const char* s) { | |
3062 | std::lock_guard l(qlock); | |
3063 | if (cmode) { | |
3064 | bool ret = failed_cmode.empty(); | |
3065 | failed_cmode = s; | |
3066 | return ret; | |
3067 | } | |
3068 | return failed_compressors.emplace(s).second; | |
3069 | } | |
3070 | void _clear_compression_alert() { | |
3071 | std::lock_guard l(qlock); | |
3072 | failed_compressors.clear(); | |
3073 | failed_cmode.clear(); | |
3074 | } | |
3075 | ||
3076 | void _set_spillover_alert(const std::string& s) { | |
3077 | std::lock_guard l(qlock); | |
3078 | spillover_alert = s; | |
3079 | } | |
3080 | void _clear_spillover_alert() { | |
3081 | std::lock_guard l(qlock); | |
3082 | spillover_alert.clear(); | |
3083 | } | |
3084 | ||
3085 | void _check_legacy_statfs_alert(); | |
3086 | void _check_no_per_pg_or_pool_omap_alert(); | |
3087 | void _set_disk_size_mismatch_alert(const std::string& s) { | |
3088 | std::lock_guard l(qlock); | |
3089 | disk_size_mismatch_alert = s; | |
3090 | } | |
3091 | void _set_spurious_read_errors_alert(const string& s) { | |
3092 | std::lock_guard l(qlock); | |
3093 | spurious_read_errors_alert = s; | |
3094 | } | |
3095 | ||
3096 | private: | |
3097 | ||
3098 | // -------------------------------------------------------- | |
3099 | // read processing internal methods | |
3100 | int _verify_csum( | |
3101 | OnodeRef& o, | |
3102 | const bluestore_blob_t* blob, | |
3103 | uint64_t blob_xoffset, | |
3104 | const ceph::buffer::list& bl, | |
3105 | uint64_t logical_offset) const; | |
3106 | int _decompress(ceph::buffer::list& source, ceph::buffer::list* result); | |
3107 | ||
3108 | ||
3109 | // -------------------------------------------------------- | |
3110 | // write ops | |
3111 | ||
3112 | struct WriteContext { | |
3113 | bool buffered = false; ///< buffered write | |
3114 | bool compress = false; ///< compressed write | |
3115 | uint64_t target_blob_size = 0; ///< target (max) blob size | |
3116 | unsigned csum_order = 0; ///< target checksum chunk order | |
3117 | ||
3118 | old_extent_map_t old_extents; ///< must deref these blobs | |
3119 | interval_set<uint64_t> extents_to_gc; ///< extents for garbage collection | |
3120 | ||
3121 | struct write_item { | |
3122 | uint64_t logical_offset; ///< write logical offset | |
3123 | BlobRef b; | |
3124 | uint64_t blob_length; | |
3125 | uint64_t b_off; | |
3126 | ceph::buffer::list bl; | |
3127 | uint64_t b_off0; ///< original offset in a blob prior to padding | |
3128 | uint64_t length0; ///< original data length prior to padding | |
3129 | ||
3130 | bool mark_unused; | |
3131 | bool new_blob; ///< whether new blob was created | |
3132 | ||
3133 | bool compressed = false; | |
3134 | ceph::buffer::list compressed_bl; | |
3135 | size_t compressed_len = 0; | |
3136 | ||
3137 | write_item( | |
3138 | uint64_t logical_offs, | |
3139 | BlobRef b, | |
3140 | uint64_t blob_len, | |
3141 | uint64_t o, | |
3142 | ceph::buffer::list& bl, | |
3143 | uint64_t o0, | |
3144 | uint64_t l0, | |
3145 | bool _mark_unused, | |
3146 | bool _new_blob) | |
3147 | : | |
3148 | logical_offset(logical_offs), | |
3149 | b(b), | |
3150 | blob_length(blob_len), | |
3151 | b_off(o), | |
3152 | bl(bl), | |
3153 | b_off0(o0), | |
3154 | length0(l0), | |
3155 | mark_unused(_mark_unused), | |
3156 | new_blob(_new_blob) {} | |
3157 | }; | |
3158 | std::vector<write_item> writes; ///< blobs we're writing | |
3159 | ||
3160 | /// partial clone of the context | |
3161 | void fork(const WriteContext& other) { | |
3162 | buffered = other.buffered; | |
3163 | compress = other.compress; | |
3164 | target_blob_size = other.target_blob_size; | |
3165 | csum_order = other.csum_order; | |
3166 | } | |
3167 | void write( | |
3168 | uint64_t loffs, | |
3169 | BlobRef b, | |
3170 | uint64_t blob_len, | |
3171 | uint64_t o, | |
3172 | ceph::buffer::list& bl, | |
3173 | uint64_t o0, | |
3174 | uint64_t len0, | |
3175 | bool _mark_unused, | |
3176 | bool _new_blob) { | |
3177 | writes.emplace_back(loffs, | |
3178 | b, | |
3179 | blob_len, | |
3180 | o, | |
3181 | bl, | |
3182 | o0, | |
3183 | len0, | |
3184 | _mark_unused, | |
3185 | _new_blob); | |
3186 | } | |
3187 | /// Checks for writes to the same pextent within a blob | |
3188 | bool has_conflict( | |
3189 | BlobRef b, | |
3190 | uint64_t loffs, | |
3191 | uint64_t loffs_end, | |
3192 | uint64_t min_alloc_size); | |
3193 | }; | |
3194 | ||
3195 | void _do_write_small( | |
3196 | TransContext *txc, | |
3197 | CollectionRef &c, | |
3198 | OnodeRef o, | |
3199 | uint64_t offset, uint64_t length, | |
3200 | ceph::buffer::list::iterator& blp, | |
3201 | WriteContext *wctx); | |
3202 | void _do_write_big_apply_deferred( | |
3203 | TransContext* txc, | |
3204 | CollectionRef& c, | |
3205 | OnodeRef o, | |
3206 | BigDeferredWriteContext& dctx, | |
3207 | bufferlist::iterator& blp, | |
3208 | WriteContext* wctx); | |
3209 | void _do_write_big( | |
3210 | TransContext *txc, | |
3211 | CollectionRef &c, | |
3212 | OnodeRef o, | |
3213 | uint64_t offset, uint64_t length, | |
3214 | ceph::buffer::list::iterator& blp, | |
3215 | WriteContext *wctx); | |
3216 | int _do_alloc_write( | |
3217 | TransContext *txc, | |
3218 | CollectionRef c, | |
3219 | OnodeRef o, | |
3220 | WriteContext *wctx); | |
3221 | void _wctx_finish( | |
3222 | TransContext *txc, | |
3223 | CollectionRef& c, | |
3224 | OnodeRef o, | |
3225 | WriteContext *wctx, | |
3226 | std::set<SharedBlob*> *maybe_unshared_blobs=0); | |
3227 | ||
3228 | int _write(TransContext *txc, | |
3229 | CollectionRef& c, | |
3230 | OnodeRef& o, | |
3231 | uint64_t offset, size_t len, | |
3232 | ceph::buffer::list& bl, | |
3233 | uint32_t fadvise_flags); | |
3234 | void _pad_zeros(ceph::buffer::list *bl, uint64_t *offset, | |
3235 | uint64_t chunk_size); | |
3236 | ||
3237 | void _choose_write_options(CollectionRef& c, | |
3238 | OnodeRef o, | |
3239 | uint32_t fadvise_flags, | |
3240 | WriteContext *wctx); | |
3241 | ||
3242 | int _do_gc(TransContext *txc, | |
3243 | CollectionRef& c, | |
3244 | OnodeRef o, | |
3245 | const WriteContext& wctx, | |
3246 | uint64_t *dirty_start, | |
3247 | uint64_t *dirty_end); | |
3248 | ||
3249 | int _do_write(TransContext *txc, | |
3250 | CollectionRef &c, | |
3251 | OnodeRef o, | |
3252 | uint64_t offset, uint64_t length, | |
3253 | ceph::buffer::list& bl, | |
3254 | uint32_t fadvise_flags); | |
3255 | void _do_write_data(TransContext *txc, | |
3256 | CollectionRef& c, | |
3257 | OnodeRef o, | |
3258 | uint64_t offset, | |
3259 | uint64_t length, | |
3260 | ceph::buffer::list& bl, | |
3261 | WriteContext *wctx); | |
3262 | ||
3263 | int _touch(TransContext *txc, | |
3264 | CollectionRef& c, | |
3265 | OnodeRef& o); | |
3266 | int _do_zero(TransContext *txc, | |
3267 | CollectionRef& c, | |
3268 | OnodeRef& o, | |
3269 | uint64_t offset, size_t len); | |
3270 | int _zero(TransContext *txc, | |
3271 | CollectionRef& c, | |
3272 | OnodeRef& o, | |
3273 | uint64_t offset, size_t len); | |
3274 | void _do_truncate(TransContext *txc, | |
3275 | CollectionRef& c, | |
3276 | OnodeRef o, | |
3277 | uint64_t offset, | |
3278 | std::set<SharedBlob*> *maybe_unshared_blobs=0); | |
3279 | int _truncate(TransContext *txc, | |
3280 | CollectionRef& c, | |
3281 | OnodeRef& o, | |
3282 | uint64_t offset); | |
3283 | int _remove(TransContext *txc, | |
3284 | CollectionRef& c, | |
3285 | OnodeRef& o); | |
3286 | int _do_remove(TransContext *txc, | |
3287 | CollectionRef& c, | |
3288 | OnodeRef o); | |
3289 | int _setattr(TransContext *txc, | |
3290 | CollectionRef& c, | |
3291 | OnodeRef& o, | |
3292 | const std::string& name, | |
3293 | ceph::buffer::ptr& val); | |
3294 | int _setattrs(TransContext *txc, | |
3295 | CollectionRef& c, | |
3296 | OnodeRef& o, | |
3297 | const std::map<std::string,ceph::buffer::ptr>& aset); | |
3298 | int _rmattr(TransContext *txc, | |
3299 | CollectionRef& c, | |
3300 | OnodeRef& o, | |
3301 | const std::string& name); | |
3302 | int _rmattrs(TransContext *txc, | |
3303 | CollectionRef& c, | |
3304 | OnodeRef& o); | |
3305 | void _do_omap_clear(TransContext *txc, OnodeRef &o); | |
3306 | int _omap_clear(TransContext *txc, | |
3307 | CollectionRef& c, | |
3308 | OnodeRef& o); | |
3309 | int _omap_setkeys(TransContext *txc, | |
3310 | CollectionRef& c, | |
3311 | OnodeRef& o, | |
3312 | ceph::buffer::list& bl); | |
3313 | int _omap_setheader(TransContext *txc, | |
3314 | CollectionRef& c, | |
3315 | OnodeRef& o, | |
3316 | ceph::buffer::list& header); | |
3317 | int _omap_rmkeys(TransContext *txc, | |
3318 | CollectionRef& c, | |
3319 | OnodeRef& o, | |
3320 | ceph::buffer::list& bl); | |
3321 | int _omap_rmkey_range(TransContext *txc, | |
3322 | CollectionRef& c, | |
3323 | OnodeRef& o, | |
3324 | const std::string& first, const std::string& last); | |
3325 | int _set_alloc_hint( | |
3326 | TransContext *txc, | |
3327 | CollectionRef& c, | |
3328 | OnodeRef& o, | |
3329 | uint64_t expected_object_size, | |
3330 | uint64_t expected_write_size, | |
3331 | uint32_t flags); | |
3332 | int _do_clone_range(TransContext *txc, | |
3333 | CollectionRef& c, | |
3334 | OnodeRef& oldo, | |
3335 | OnodeRef& newo, | |
3336 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
3337 | int _clone(TransContext *txc, | |
3338 | CollectionRef& c, | |
3339 | OnodeRef& oldo, | |
3340 | OnodeRef& newo); | |
3341 | int _clone_range(TransContext *txc, | |
3342 | CollectionRef& c, | |
3343 | OnodeRef& oldo, | |
3344 | OnodeRef& newo, | |
3345 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
3346 | int _rename(TransContext *txc, | |
3347 | CollectionRef& c, | |
3348 | OnodeRef& oldo, | |
3349 | OnodeRef& newo, | |
3350 | const ghobject_t& new_oid); | |
3351 | int _create_collection(TransContext *txc, const coll_t &cid, | |
3352 | unsigned bits, CollectionRef *c); | |
3353 | int _remove_collection(TransContext *txc, const coll_t &cid, | |
3354 | CollectionRef *c); | |
3355 | void _do_remove_collection(TransContext *txc, CollectionRef *c); | |
3356 | int _split_collection(TransContext *txc, | |
3357 | CollectionRef& c, | |
3358 | CollectionRef& d, | |
3359 | unsigned bits, int rem); | |
3360 | int _merge_collection(TransContext *txc, | |
3361 | CollectionRef *c, | |
3362 | CollectionRef& d, | |
3363 | unsigned bits); | |
3364 | ||
3365 | void _collect_allocation_stats(uint64_t need, uint32_t alloc_size, | |
3366 | size_t extents); | |
3367 | void _record_allocation_stats(); | |
3368 | private: | |
3369 | uint64_t probe_count = 0; | |
3370 | std::atomic<uint64_t> alloc_stats_count = {0}; | |
3371 | std::atomic<uint64_t> alloc_stats_fragments = { 0 }; | |
3372 | std::atomic<uint64_t> alloc_stats_size = { 0 }; | |
3373 | // | |
3374 | std::array<std::tuple<uint64_t, uint64_t, uint64_t>, 5> alloc_stats_history = | |
3375 | { std::make_tuple(0ul, 0ul, 0ul) }; | |
3376 | ||
3377 | inline bool _use_rotational_settings(); | |
3378 | ||
3379 | public: | |
3380 | struct sb_info_t { | |
3381 | coll_t cid; | |
3382 | int64_t pool_id = INT64_MIN; | |
3383 | std::list<ghobject_t> oids; | |
3384 | BlueStore::SharedBlobRef sb; | |
3385 | bluestore_extent_ref_map_t ref_map; | |
3386 | bool compressed = false; | |
3387 | bool passed = false; | |
3388 | bool updated = false; | |
3389 | }; | |
3390 | typedef btree::btree_set< | |
3391 | uint64_t, std::less<uint64_t>, | |
3392 | mempool::bluestore_fsck::pool_allocator<uint64_t>> uint64_t_btree_t; | |
3393 | ||
3394 | typedef mempool::bluestore_fsck::map<uint64_t, sb_info_t> sb_info_map_t; | |
3395 | struct FSCK_ObjectCtx { | |
3396 | int64_t& errors; | |
3397 | int64_t& warnings; | |
3398 | uint64_t& num_objects; | |
3399 | uint64_t& num_extents; | |
3400 | uint64_t& num_blobs; | |
3401 | uint64_t& num_sharded_objects; | |
3402 | uint64_t& num_spanning_blobs; | |
3403 | ||
3404 | mempool_dynamic_bitset* used_blocks; | |
3405 | uint64_t_btree_t* used_omap_head; | |
3406 | ||
3407 | ceph::mutex* sb_info_lock; | |
3408 | sb_info_map_t& sb_info; | |
3409 | ||
3410 | store_statfs_t& expected_store_statfs; | |
3411 | per_pool_statfs& expected_pool_statfs; | |
3412 | BlueStoreRepairer* repairer; | |
3413 | ||
3414 | FSCK_ObjectCtx(int64_t& e, | |
3415 | int64_t& w, | |
3416 | uint64_t& _num_objects, | |
3417 | uint64_t& _num_extents, | |
3418 | uint64_t& _num_blobs, | |
3419 | uint64_t& _num_sharded_objects, | |
3420 | uint64_t& _num_spanning_blobs, | |
3421 | mempool_dynamic_bitset* _ub, | |
3422 | uint64_t_btree_t* _used_omap_head, | |
3423 | ceph::mutex* _sb_info_lock, | |
3424 | sb_info_map_t& _sb_info, | |
3425 | store_statfs_t& _store_statfs, | |
3426 | per_pool_statfs& _pool_statfs, | |
3427 | BlueStoreRepairer* _repairer) : | |
3428 | errors(e), | |
3429 | warnings(w), | |
3430 | num_objects(_num_objects), | |
3431 | num_extents(_num_extents), | |
3432 | num_blobs(_num_blobs), | |
3433 | num_sharded_objects(_num_sharded_objects), | |
3434 | num_spanning_blobs(_num_spanning_blobs), | |
3435 | used_blocks(_ub), | |
3436 | used_omap_head(_used_omap_head), | |
3437 | sb_info_lock(_sb_info_lock), | |
3438 | sb_info(_sb_info), | |
3439 | expected_store_statfs(_store_statfs), | |
3440 | expected_pool_statfs(_pool_statfs), | |
3441 | repairer(_repairer) { | |
3442 | } | |
3443 | }; | |
3444 | ||
3445 | OnodeRef fsck_check_objects_shallow( | |
3446 | FSCKDepth depth, | |
3447 | int64_t pool_id, | |
3448 | CollectionRef c, | |
3449 | const ghobject_t& oid, | |
3450 | const std::string& key, | |
3451 | const ceph::buffer::list& value, | |
3452 | mempool::bluestore_fsck::list<std::string>* expecting_shards, | |
3453 | std::map<BlobRef, bluestore_blob_t::unused_t>* referenced, | |
3454 | const BlueStore::FSCK_ObjectCtx& ctx); | |
3455 | ||
3456 | private: | |
3457 | void _fsck_check_object_omap(FSCKDepth depth, | |
3458 | OnodeRef& o, | |
3459 | const BlueStore::FSCK_ObjectCtx& ctx); | |
3460 | ||
3461 | void _fsck_check_objects(FSCKDepth depth, | |
3462 | FSCK_ObjectCtx& ctx); | |
3463 | }; | |
3464 | ||
3465 | inline std::ostream& operator<<(std::ostream& out, const BlueStore::volatile_statfs& s) { | |
3466 | return out | |
3467 | << " allocated:" | |
3468 | << s.values[BlueStore::volatile_statfs::STATFS_ALLOCATED] | |
3469 | << " stored:" | |
3470 | << s.values[BlueStore::volatile_statfs::STATFS_STORED] | |
3471 | << " compressed:" | |
3472 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED] | |
3473 | << " compressed_orig:" | |
3474 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED_ORIGINAL] | |
3475 | << " compressed_alloc:" | |
3476 | << s.values[BlueStore::volatile_statfs::STATFS_COMPRESSED_ALLOCATED]; | |
3477 | } | |
3478 | ||
3479 | static inline void intrusive_ptr_add_ref(BlueStore::Onode *o) { | |
3480 | o->get(); | |
3481 | } | |
3482 | static inline void intrusive_ptr_release(BlueStore::Onode *o) { | |
3483 | o->put(); | |
3484 | } | |
3485 | ||
3486 | static inline void intrusive_ptr_add_ref(BlueStore::OpSequencer *o) { | |
3487 | o->get(); | |
3488 | } | |
3489 | static inline void intrusive_ptr_release(BlueStore::OpSequencer *o) { | |
3490 | o->put(); | |
3491 | } | |
3492 | ||
3493 | class BlueStoreRepairer | |
3494 | { | |
3495 | public: | |
3496 | // to simplify future potential migration to mempools | |
3497 | using fsck_interval = interval_set<uint64_t>; | |
3498 | ||
3499 | // Structure to track what pextents are used for specific cid/oid. | |
3500 | // Similar to Bloom filter positive and false-positive matches are | |
3501 | // possible only. | |
3502 | // Maintains two lists of bloom filters for both cids and oids | |
3503 | // where each list entry is a BF for specific disk pextent | |
3504 | // The length of the extent per filter is measured on init. | |
3505 | // Allows to filter out 'uninteresting' pextents to speadup subsequent | |
3506 | // 'is_used' access. | |
3507 | struct StoreSpaceTracker { | |
3508 | const uint64_t BLOOM_FILTER_SALT_COUNT = 2; | |
3509 | const uint64_t BLOOM_FILTER_TABLE_SIZE = 32; // bytes per single filter | |
3510 | const uint64_t BLOOM_FILTER_EXPECTED_COUNT = 16; // arbitrary selected | |
3511 | static const uint64_t DEF_MEM_CAP = 128 * 1024 * 1024; | |
3512 | ||
3513 | typedef mempool::bluestore_fsck::vector<bloom_filter> bloom_vector; | |
3514 | bloom_vector collections_bfs; | |
3515 | bloom_vector objects_bfs; | |
3516 | ||
3517 | bool was_filtered_out = false; | |
3518 | uint64_t granularity = 0; // extent length for a single filter | |
3519 | ||
3520 | StoreSpaceTracker() { | |
3521 | } | |
3522 | StoreSpaceTracker(const StoreSpaceTracker& from) : | |
3523 | collections_bfs(from.collections_bfs), | |
3524 | objects_bfs(from.objects_bfs), | |
3525 | granularity(from.granularity) { | |
3526 | } | |
3527 | ||
3528 | void init(uint64_t total, | |
3529 | uint64_t min_alloc_size, | |
3530 | uint64_t mem_cap = DEF_MEM_CAP) { | |
3531 | ceph_assert(!granularity); // not initialized yet | |
3532 | ceph_assert(min_alloc_size && isp2(min_alloc_size)); | |
3533 | ceph_assert(mem_cap); | |
3534 | ||
3535 | total = round_up_to(total, min_alloc_size); | |
3536 | granularity = total * BLOOM_FILTER_TABLE_SIZE * 2 / mem_cap; | |
3537 | ||
3538 | if (!granularity) { | |
3539 | granularity = min_alloc_size; | |
3540 | } else { | |
3541 | granularity = round_up_to(granularity, min_alloc_size); | |
3542 | } | |
3543 | ||
3544 | uint64_t entries = round_up_to(total, granularity) / granularity; | |
3545 | collections_bfs.resize(entries, | |
3546 | bloom_filter(BLOOM_FILTER_SALT_COUNT, | |
3547 | BLOOM_FILTER_TABLE_SIZE, | |
3548 | 0, | |
3549 | BLOOM_FILTER_EXPECTED_COUNT)); | |
3550 | objects_bfs.resize(entries, | |
3551 | bloom_filter(BLOOM_FILTER_SALT_COUNT, | |
3552 | BLOOM_FILTER_TABLE_SIZE, | |
3553 | 0, | |
3554 | BLOOM_FILTER_EXPECTED_COUNT)); | |
3555 | } | |
3556 | inline uint32_t get_hash(const coll_t& cid) const { | |
3557 | return cid.hash_to_shard(1); | |
3558 | } | |
3559 | inline void set_used(uint64_t offset, uint64_t len, | |
3560 | const coll_t& cid, const ghobject_t& oid) { | |
3561 | ceph_assert(granularity); // initialized | |
3562 | ||
3563 | // can't call this func after filter_out has been applied | |
3564 | ceph_assert(!was_filtered_out); | |
3565 | if (!len) { | |
3566 | return; | |
3567 | } | |
3568 | auto pos = offset / granularity; | |
3569 | auto end_pos = (offset + len - 1) / granularity; | |
3570 | while (pos <= end_pos) { | |
3571 | collections_bfs[pos].insert(get_hash(cid)); | |
3572 | objects_bfs[pos].insert(oid.hobj.get_hash()); | |
3573 | ++pos; | |
3574 | } | |
3575 | } | |
3576 | // filter-out entries unrelated to the specified(broken) extents. | |
3577 | // 'is_used' calls are permitted after that only | |
3578 | size_t filter_out(const fsck_interval& extents); | |
3579 | ||
3580 | // determines if collection's present after filtering-out | |
3581 | inline bool is_used(const coll_t& cid) const { | |
3582 | ceph_assert(was_filtered_out); | |
3583 | for(auto& bf : collections_bfs) { | |
3584 | if (bf.contains(get_hash(cid))) { | |
3585 | return true; | |
3586 | } | |
3587 | } | |
3588 | return false; | |
3589 | } | |
3590 | // determines if object's present after filtering-out | |
3591 | inline bool is_used(const ghobject_t& oid) const { | |
3592 | ceph_assert(was_filtered_out); | |
3593 | for(auto& bf : objects_bfs) { | |
3594 | if (bf.contains(oid.hobj.get_hash())) { | |
3595 | return true; | |
3596 | } | |
3597 | } | |
3598 | return false; | |
3599 | } | |
3600 | // determines if collection's present before filtering-out | |
3601 | inline bool is_used(const coll_t& cid, uint64_t offs) const { | |
3602 | ceph_assert(granularity); // initialized | |
3603 | ceph_assert(!was_filtered_out); | |
3604 | auto &bf = collections_bfs[offs / granularity]; | |
3605 | if (bf.contains(get_hash(cid))) { | |
3606 | return true; | |
3607 | } | |
3608 | return false; | |
3609 | } | |
3610 | // determines if object's present before filtering-out | |
3611 | inline bool is_used(const ghobject_t& oid, uint64_t offs) const { | |
3612 | ceph_assert(granularity); // initialized | |
3613 | ceph_assert(!was_filtered_out); | |
3614 | auto &bf = objects_bfs[offs / granularity]; | |
3615 | if (bf.contains(oid.hobj.get_hash())) { | |
3616 | return true; | |
3617 | } | |
3618 | return false; | |
3619 | } | |
3620 | }; | |
3621 | public: | |
3622 | void fix_per_pool_omap(KeyValueDB *db, int); | |
3623 | bool remove_key(KeyValueDB *db, const std::string& prefix, const std::string& key); | |
3624 | bool fix_shared_blob(KeyValueDB *db, | |
3625 | uint64_t sbid, | |
3626 | const ceph::buffer::list* bl); | |
3627 | bool fix_statfs(KeyValueDB *db, const std::string& key, | |
3628 | const store_statfs_t& new_statfs); | |
3629 | ||
3630 | bool fix_leaked(KeyValueDB *db, | |
3631 | FreelistManager* fm, | |
3632 | uint64_t offset, uint64_t len); | |
3633 | bool fix_false_free(KeyValueDB *db, | |
3634 | FreelistManager* fm, | |
3635 | uint64_t offset, uint64_t len); | |
3636 | KeyValueDB::Transaction fix_spanning_blobs(KeyValueDB* db); | |
3637 | ||
3638 | void init(uint64_t total_space, uint64_t lres_tracking_unit_size); | |
3639 | ||
3640 | bool preprocess_misreference(KeyValueDB *db); | |
3641 | ||
3642 | unsigned apply(KeyValueDB* db); | |
3643 | ||
3644 | void note_misreference(uint64_t offs, uint64_t len, bool inc_error) { | |
3645 | misreferenced_extents.union_insert(offs, len); | |
3646 | if (inc_error) { | |
3647 | ++to_repair_cnt; | |
3648 | } | |
3649 | } | |
3650 | // In fact this is the only repairer's method which is thread-safe!! | |
3651 | void inc_repaired() { | |
3652 | ++to_repair_cnt; | |
3653 | } | |
3654 | ||
3655 | StoreSpaceTracker& get_space_usage_tracker() { | |
3656 | return space_usage_tracker; | |
3657 | } | |
3658 | const fsck_interval& get_misreferences() const { | |
3659 | return misreferenced_extents; | |
3660 | } | |
3661 | KeyValueDB::Transaction get_fix_misreferences_txn() { | |
3662 | return fix_misreferences_txn; | |
3663 | } | |
3664 | ||
3665 | private: | |
3666 | std::atomic<unsigned> to_repair_cnt = { 0 }; | |
3667 | KeyValueDB::Transaction fix_per_pool_omap_txn; | |
3668 | KeyValueDB::Transaction fix_fm_leaked_txn; | |
3669 | KeyValueDB::Transaction fix_fm_false_free_txn; | |
3670 | KeyValueDB::Transaction remove_key_txn; | |
3671 | KeyValueDB::Transaction fix_statfs_txn; | |
3672 | KeyValueDB::Transaction fix_shared_blob_txn; | |
3673 | ||
3674 | KeyValueDB::Transaction fix_misreferences_txn; | |
3675 | KeyValueDB::Transaction fix_onode_txn; | |
3676 | ||
3677 | StoreSpaceTracker space_usage_tracker; | |
3678 | ||
3679 | // non-shared extents with multiple references | |
3680 | fsck_interval misreferenced_extents; | |
3681 | ||
3682 | }; | |
3683 | ||
3684 | class RocksDBBlueFSVolumeSelector : public BlueFSVolumeSelector | |
3685 | { | |
3686 | template <class T, size_t MaxX, size_t MaxY> | |
3687 | class matrix_2d { | |
3688 | T values[MaxX][MaxY]; | |
3689 | public: | |
3690 | matrix_2d() { | |
3691 | clear(); | |
3692 | } | |
3693 | T& at(size_t x, size_t y) { | |
3694 | ceph_assert(x < MaxX); | |
3695 | ceph_assert(y < MaxY); | |
3696 | ||
3697 | return values[x][y]; | |
3698 | } | |
3699 | size_t get_max_x() const { | |
3700 | return MaxX; | |
3701 | } | |
3702 | size_t get_max_y() const { | |
3703 | return MaxY; | |
3704 | } | |
3705 | void clear() { | |
3706 | memset(values, 0, sizeof(values)); | |
3707 | } | |
3708 | }; | |
3709 | ||
3710 | enum { | |
3711 | // use 0/nullptr as unset indication | |
3712 | LEVEL_FIRST = 1, | |
3713 | LEVEL_LOG = LEVEL_FIRST, // BlueFS log | |
3714 | LEVEL_WAL, | |
3715 | LEVEL_DB, | |
3716 | LEVEL_SLOW, | |
3717 | LEVEL_MAX | |
3718 | }; | |
3719 | // add +1 row for corresponding per-device totals | |
3720 | // add +1 column for per-level actual (taken from file size) total | |
3721 | typedef matrix_2d<uint64_t, BlueFS::MAX_BDEV + 1, LEVEL_MAX - LEVEL_FIRST + 1> per_level_per_dev_usage_t; | |
3722 | ||
3723 | per_level_per_dev_usage_t per_level_per_dev_usage; | |
3724 | // file count per level, add +1 to keep total file count | |
3725 | uint64_t per_level_files[LEVEL_MAX - LEVEL_FIRST + 1] = { 0 }; | |
3726 | ||
3727 | // Note: maximum per-device totals below might be smaller than corresponding | |
3728 | // perf counters by up to a single alloc unit (1M) due to superblock extent. | |
3729 | // The later is not accounted here. | |
3730 | per_level_per_dev_usage_t per_level_per_dev_max; | |
3731 | ||
3732 | uint64_t l_totals[LEVEL_MAX - LEVEL_FIRST]; | |
3733 | uint64_t db_avail4slow = 0; | |
3734 | enum { | |
3735 | OLD_POLICY, | |
3736 | USE_SOME_EXTRA | |
3737 | }; | |
3738 | ||
3739 | public: | |
3740 | RocksDBBlueFSVolumeSelector( | |
3741 | uint64_t _wal_total, | |
3742 | uint64_t _db_total, | |
3743 | uint64_t _slow_total, | |
3744 | uint64_t _level0_size, | |
3745 | uint64_t _level_base, | |
3746 | uint64_t _level_multiplier, | |
3747 | double reserved_factor, | |
3748 | uint64_t reserved, | |
3749 | bool new_pol) | |
3750 | { | |
3751 | l_totals[LEVEL_LOG - LEVEL_FIRST] = 0; // not used at the moment | |
3752 | l_totals[LEVEL_WAL - LEVEL_FIRST] = _wal_total; | |
3753 | l_totals[LEVEL_DB - LEVEL_FIRST] = _db_total; | |
3754 | l_totals[LEVEL_SLOW - LEVEL_FIRST] = _slow_total; | |
3755 | ||
3756 | if (!new_pol) { | |
3757 | return; | |
3758 | } | |
3759 | ||
3760 | // Calculating how much extra space is available at DB volume. | |
3761 | // Depending on the presence of explicit reserved size specification it might be either | |
3762 | // * DB volume size - reserved | |
3763 | // or | |
3764 | // * DB volume size - sum_max_level_size(0, L-1) - max_level_size(L) * reserved_factor | |
3765 | if (!reserved) { | |
3766 | uint64_t prev_levels = _level0_size; | |
3767 | uint64_t cur_level = _level_base; | |
3768 | uint64_t cur_threshold = 0; | |
3769 | do { | |
3770 | uint64_t next_level = cur_level * _level_multiplier; | |
3771 | uint64_t next_threshold = prev_levels + cur_level + next_level * reserved_factor; | |
3772 | if (_db_total <= next_threshold) { | |
3773 | db_avail4slow = cur_threshold ? _db_total - cur_threshold : 0; | |
3774 | break; | |
3775 | } else { | |
3776 | prev_levels += cur_level; | |
3777 | cur_level = next_level; | |
3778 | cur_threshold = next_threshold; | |
3779 | } | |
3780 | } while (true); | |
3781 | } else { | |
3782 | db_avail4slow = _db_total - reserved; | |
3783 | } | |
3784 | } | |
3785 | ||
3786 | void* get_hint_for_log() const override { | |
3787 | return reinterpret_cast<void*>(LEVEL_LOG); | |
3788 | } | |
3789 | void* get_hint_by_dir(const std::string& dirname) const override; | |
3790 | ||
3791 | void add_usage(void* hint, const bluefs_fnode_t& fnode) override { | |
3792 | if (hint == nullptr) | |
3793 | return; | |
3794 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3795 | for (auto& p : fnode.extents) { | |
3796 | auto& cur = per_level_per_dev_usage.at(p.bdev, pos); | |
3797 | auto& max = per_level_per_dev_max.at(p.bdev, pos); | |
3798 | cur += p.length; | |
3799 | if (cur > max) { | |
3800 | max = cur; | |
3801 | } | |
3802 | { | |
3803 | //update per-device totals | |
3804 | auto& cur = per_level_per_dev_usage.at(p.bdev, LEVEL_MAX - LEVEL_FIRST); | |
3805 | auto& max = per_level_per_dev_max.at(p.bdev, LEVEL_MAX - LEVEL_FIRST); | |
3806 | cur += p.length; | |
3807 | if (cur > max) { | |
3808 | max = cur; | |
3809 | } | |
3810 | } | |
3811 | } | |
3812 | { | |
3813 | //update per-level actual totals | |
3814 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3815 | auto& max = per_level_per_dev_max.at(BlueFS::MAX_BDEV, pos); | |
3816 | cur += fnode.size; | |
3817 | if (cur > max) { | |
3818 | max = cur; | |
3819 | } | |
3820 | } | |
3821 | ++per_level_files[pos]; | |
3822 | ++per_level_files[LEVEL_MAX - LEVEL_FIRST]; | |
3823 | } | |
3824 | void sub_usage(void* hint, const bluefs_fnode_t& fnode) override { | |
3825 | if (hint == nullptr) | |
3826 | return; | |
3827 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3828 | for (auto& p : fnode.extents) { | |
3829 | auto& cur = per_level_per_dev_usage.at(p.bdev, pos); | |
3830 | ceph_assert(cur >= p.length); | |
3831 | cur -= p.length; | |
3832 | ||
3833 | //update per-device totals | |
3834 | auto& cur2 = per_level_per_dev_usage.at(p.bdev, LEVEL_MAX - LEVEL_FIRST); | |
3835 | ceph_assert(cur2 >= p.length); | |
3836 | cur2 -= p.length; | |
3837 | } | |
3838 | //update per-level actual totals | |
3839 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3840 | ceph_assert(cur >= fnode.size); | |
3841 | cur -= fnode.size; | |
3842 | ceph_assert(per_level_files[pos] > 0); | |
3843 | --per_level_files[pos]; | |
3844 | ceph_assert(per_level_files[LEVEL_MAX - LEVEL_FIRST] > 0); | |
3845 | --per_level_files[LEVEL_MAX - LEVEL_FIRST]; | |
3846 | } | |
3847 | void add_usage(void* hint, uint64_t fsize) override { | |
3848 | if (hint == nullptr) | |
3849 | return; | |
3850 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3851 | //update per-level actual totals | |
3852 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3853 | auto& max = per_level_per_dev_max.at(BlueFS::MAX_BDEV, pos); | |
3854 | cur += fsize; | |
3855 | if (cur > max) { | |
3856 | max = cur; | |
3857 | } | |
3858 | } | |
3859 | void sub_usage(void* hint, uint64_t fsize) override { | |
3860 | if (hint == nullptr) | |
3861 | return; | |
3862 | size_t pos = (size_t)hint - LEVEL_FIRST; | |
3863 | //update per-level actual totals | |
3864 | auto& cur = per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos); | |
3865 | ceph_assert(cur >= fsize); | |
3866 | per_level_per_dev_usage.at(BlueFS::MAX_BDEV, pos) -= fsize; | |
3867 | } | |
3868 | ||
3869 | uint8_t select_prefer_bdev(void* h) override; | |
3870 | void get_paths( | |
3871 | const std::string& base, | |
3872 | BlueFSVolumeSelector::paths& res) const override; | |
3873 | ||
3874 | void dump(std::ostream& sout) override; | |
3875 | }; | |
3876 | ||
3877 | #endif |