]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2014 Red Hat | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
15 | #ifndef CEPH_OSD_BLUESTORE_H | |
16 | #define CEPH_OSD_BLUESTORE_H | |
17 | ||
18 | #include "acconfig.h" | |
19 | ||
20 | #include <unistd.h> | |
21 | ||
22 | #include <atomic> | |
23 | #include <mutex> | |
24 | #include <condition_variable> | |
25 | ||
26 | #include <boost/intrusive/list.hpp> | |
27 | #include <boost/intrusive/unordered_set.hpp> | |
28 | #include <boost/intrusive/set.hpp> | |
29 | #include <boost/functional/hash.hpp> | |
30 | #include <boost/dynamic_bitset.hpp> | |
31 | ||
32 | #include "include/assert.h" | |
33 | #include "include/unordered_map.h" | |
34 | #include "include/memory.h" | |
35 | #include "include/mempool.h" | |
36 | #include "common/Finisher.h" | |
37 | #include "common/perf_counters.h" | |
38 | #include "compressor/Compressor.h" | |
39 | #include "os/ObjectStore.h" | |
40 | ||
41 | #include "bluestore_types.h" | |
42 | #include "BlockDevice.h" | |
43 | #include "common/EventTrace.h" | |
44 | ||
45 | class Allocator; | |
46 | class FreelistManager; | |
47 | class BlueFS; | |
48 | ||
49 | //#define DEBUG_CACHE | |
50 | //#define DEBUG_DEFERRED | |
51 | ||
31f18b77 FG |
52 | |
53 | ||
54 | // constants for Buffer::optimize() | |
55 | #define MAX_BUFFER_SLOP_RATIO_DEN 8 // so actually 1/N | |
56 | ||
57 | ||
7c673cae FG |
58 | enum { |
59 | l_bluestore_first = 732430, | |
60 | l_bluestore_kv_flush_lat, | |
61 | l_bluestore_kv_commit_lat, | |
62 | l_bluestore_kv_lat, | |
63 | l_bluestore_state_prepare_lat, | |
64 | l_bluestore_state_aio_wait_lat, | |
65 | l_bluestore_state_io_done_lat, | |
66 | l_bluestore_state_kv_queued_lat, | |
67 | l_bluestore_state_kv_committing_lat, | |
68 | l_bluestore_state_kv_done_lat, | |
69 | l_bluestore_state_deferred_queued_lat, | |
70 | l_bluestore_state_deferred_aio_wait_lat, | |
71 | l_bluestore_state_deferred_cleanup_lat, | |
72 | l_bluestore_state_finishing_lat, | |
73 | l_bluestore_state_done_lat, | |
74 | l_bluestore_throttle_lat, | |
75 | l_bluestore_submit_lat, | |
76 | l_bluestore_commit_lat, | |
77 | l_bluestore_read_lat, | |
78 | l_bluestore_read_onode_meta_lat, | |
79 | l_bluestore_read_wait_aio_lat, | |
80 | l_bluestore_compress_lat, | |
81 | l_bluestore_decompress_lat, | |
82 | l_bluestore_csum_lat, | |
83 | l_bluestore_compress_success_count, | |
84 | l_bluestore_compress_rejected_count, | |
85 | l_bluestore_write_pad_bytes, | |
86 | l_bluestore_deferred_write_ops, | |
87 | l_bluestore_deferred_write_bytes, | |
88 | l_bluestore_write_penalty_read_ops, | |
89 | l_bluestore_allocated, | |
90 | l_bluestore_stored, | |
91 | l_bluestore_compressed, | |
92 | l_bluestore_compressed_allocated, | |
93 | l_bluestore_compressed_original, | |
94 | l_bluestore_onodes, | |
95 | l_bluestore_onode_hits, | |
96 | l_bluestore_onode_misses, | |
97 | l_bluestore_onode_shard_hits, | |
98 | l_bluestore_onode_shard_misses, | |
99 | l_bluestore_extents, | |
100 | l_bluestore_blobs, | |
101 | l_bluestore_buffers, | |
102 | l_bluestore_buffer_bytes, | |
103 | l_bluestore_buffer_hit_bytes, | |
104 | l_bluestore_buffer_miss_bytes, | |
105 | l_bluestore_write_big, | |
106 | l_bluestore_write_big_bytes, | |
107 | l_bluestore_write_big_blobs, | |
108 | l_bluestore_write_small, | |
109 | l_bluestore_write_small_bytes, | |
110 | l_bluestore_write_small_unused, | |
111 | l_bluestore_write_small_deferred, | |
112 | l_bluestore_write_small_pre_read, | |
113 | l_bluestore_write_small_new, | |
114 | l_bluestore_txc, | |
115 | l_bluestore_onode_reshard, | |
116 | l_bluestore_blob_split, | |
117 | l_bluestore_extent_compress, | |
118 | l_bluestore_gc_merged, | |
b32b8144 | 119 | l_bluestore_read_eio, |
7c673cae FG |
120 | l_bluestore_last |
121 | }; | |
122 | ||
123 | class BlueStore : public ObjectStore, | |
124 | public md_config_obs_t { | |
125 | // ----------------------------------------------------- | |
126 | // types | |
127 | public: | |
128 | // config observer | |
129 | const char** get_tracked_conf_keys() const override; | |
130 | void handle_conf_change(const struct md_config_t *conf, | |
131 | const std::set<std::string> &changed) override; | |
132 | ||
133 | void _set_csum(); | |
134 | void _set_compression(); | |
135 | void _set_throttle_params(); | |
31f18b77 | 136 | int _set_cache_sizes(); |
7c673cae FG |
137 | |
138 | class TransContext; | |
139 | ||
140 | typedef map<uint64_t, bufferlist> ready_regions_t; | |
141 | ||
142 | struct BufferSpace; | |
143 | struct Collection; | |
144 | typedef boost::intrusive_ptr<Collection> CollectionRef; | |
145 | ||
146 | struct AioContext { | |
147 | virtual void aio_finish(BlueStore *store) = 0; | |
148 | virtual ~AioContext() {} | |
149 | }; | |
150 | ||
151 | /// cached buffer | |
152 | struct Buffer { | |
153 | MEMPOOL_CLASS_HELPERS(); | |
154 | ||
155 | enum { | |
156 | STATE_EMPTY, ///< empty buffer -- used for cache history | |
157 | STATE_CLEAN, ///< clean data that is up to date | |
158 | STATE_WRITING, ///< data that is being written (io not yet complete) | |
159 | }; | |
160 | static const char *get_state_name(int s) { | |
161 | switch (s) { | |
162 | case STATE_EMPTY: return "empty"; | |
163 | case STATE_CLEAN: return "clean"; | |
164 | case STATE_WRITING: return "writing"; | |
165 | default: return "???"; | |
166 | } | |
167 | } | |
168 | enum { | |
169 | FLAG_NOCACHE = 1, ///< trim when done WRITING (do not become CLEAN) | |
170 | // NOTE: fix operator<< when you define a second flag | |
171 | }; | |
172 | static const char *get_flag_name(int s) { | |
173 | switch (s) { | |
174 | case FLAG_NOCACHE: return "nocache"; | |
175 | default: return "???"; | |
176 | } | |
177 | } | |
178 | ||
179 | BufferSpace *space; | |
180 | uint16_t state; ///< STATE_* | |
181 | uint16_t cache_private = 0; ///< opaque (to us) value used by Cache impl | |
182 | uint32_t flags; ///< FLAG_* | |
183 | uint64_t seq; | |
184 | uint32_t offset, length; | |
185 | bufferlist data; | |
186 | ||
187 | boost::intrusive::list_member_hook<> lru_item; | |
188 | boost::intrusive::list_member_hook<> state_item; | |
189 | ||
190 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, uint32_t l, | |
191 | unsigned f = 0) | |
192 | : space(space), state(s), flags(f), seq(q), offset(o), length(l) {} | |
193 | Buffer(BufferSpace *space, unsigned s, uint64_t q, uint32_t o, bufferlist& b, | |
194 | unsigned f = 0) | |
195 | : space(space), state(s), flags(f), seq(q), offset(o), | |
196 | length(b.length()), data(b) {} | |
197 | ||
198 | bool is_empty() const { | |
199 | return state == STATE_EMPTY; | |
200 | } | |
201 | bool is_clean() const { | |
202 | return state == STATE_CLEAN; | |
203 | } | |
204 | bool is_writing() const { | |
205 | return state == STATE_WRITING; | |
206 | } | |
207 | ||
208 | uint32_t end() const { | |
209 | return offset + length; | |
210 | } | |
211 | ||
212 | void truncate(uint32_t newlen) { | |
213 | assert(newlen < length); | |
214 | if (data.length()) { | |
215 | bufferlist t; | |
216 | t.substr_of(data, 0, newlen); | |
217 | data.claim(t); | |
218 | } | |
219 | length = newlen; | |
220 | } | |
31f18b77 FG |
221 | void maybe_rebuild() { |
222 | if (data.length() && | |
223 | (data.get_num_buffers() > 1 || | |
224 | data.front().wasted() > data.length() / MAX_BUFFER_SLOP_RATIO_DEN)) { | |
225 | data.rebuild(); | |
226 | } | |
227 | } | |
7c673cae FG |
228 | |
229 | void dump(Formatter *f) const { | |
230 | f->dump_string("state", get_state_name(state)); | |
231 | f->dump_unsigned("seq", seq); | |
232 | f->dump_unsigned("offset", offset); | |
233 | f->dump_unsigned("length", length); | |
234 | f->dump_unsigned("data_length", data.length()); | |
235 | } | |
236 | }; | |
237 | ||
238 | struct Cache; | |
239 | ||
240 | /// map logical extent range (object) onto buffers | |
241 | struct BufferSpace { | |
242 | typedef boost::intrusive::list< | |
243 | Buffer, | |
244 | boost::intrusive::member_hook< | |
245 | Buffer, | |
246 | boost::intrusive::list_member_hook<>, | |
247 | &Buffer::state_item> > state_list_t; | |
248 | ||
31f18b77 | 249 | mempool::bluestore_cache_other::map<uint32_t, std::unique_ptr<Buffer>> |
7c673cae FG |
250 | buffer_map; |
251 | ||
252 | // we use a bare intrusive list here instead of std::map because | |
253 | // it uses less memory and we expect this to be very small (very | |
254 | // few IOs in flight to the same Blob at the same time). | |
255 | state_list_t writing; ///< writing buffers, sorted by seq, ascending | |
256 | ||
257 | ~BufferSpace() { | |
258 | assert(buffer_map.empty()); | |
259 | assert(writing.empty()); | |
260 | } | |
261 | ||
262 | void _add_buffer(Cache* cache, Buffer *b, int level, Buffer *near) { | |
263 | cache->_audit("_add_buffer start"); | |
264 | buffer_map[b->offset].reset(b); | |
265 | if (b->is_writing()) { | |
31f18b77 | 266 | b->data.reassign_to_mempool(mempool::mempool_bluestore_writing); |
224ce89b WB |
267 | if (writing.empty() || writing.rbegin()->seq <= b->seq) { |
268 | writing.push_back(*b); | |
269 | } else { | |
270 | auto it = writing.begin(); | |
271 | while (it->seq < b->seq) { | |
272 | ++it; | |
273 | } | |
274 | ||
275 | assert(it->seq >= b->seq); | |
276 | // note that this will insert b before it | |
277 | // hence the order is maintained | |
278 | writing.insert(it, *b); | |
279 | } | |
7c673cae | 280 | } else { |
31f18b77 | 281 | b->data.reassign_to_mempool(mempool::mempool_bluestore_cache_data); |
7c673cae FG |
282 | cache->_add_buffer(b, level, near); |
283 | } | |
284 | cache->_audit("_add_buffer end"); | |
285 | } | |
286 | void _rm_buffer(Cache* cache, Buffer *b) { | |
287 | _rm_buffer(cache, buffer_map.find(b->offset)); | |
288 | } | |
31f18b77 FG |
289 | void _rm_buffer(Cache* cache, |
290 | map<uint32_t, std::unique_ptr<Buffer>>::iterator p) { | |
7c673cae FG |
291 | assert(p != buffer_map.end()); |
292 | cache->_audit("_rm_buffer start"); | |
293 | if (p->second->is_writing()) { | |
294 | writing.erase(writing.iterator_to(*p->second)); | |
295 | } else { | |
296 | cache->_rm_buffer(p->second.get()); | |
297 | } | |
298 | buffer_map.erase(p); | |
299 | cache->_audit("_rm_buffer end"); | |
300 | } | |
301 | ||
302 | map<uint32_t,std::unique_ptr<Buffer>>::iterator _data_lower_bound( | |
303 | uint32_t offset) { | |
304 | auto i = buffer_map.lower_bound(offset); | |
305 | if (i != buffer_map.begin()) { | |
306 | --i; | |
307 | if (i->first + i->second->length <= offset) | |
308 | ++i; | |
309 | } | |
310 | return i; | |
311 | } | |
312 | ||
313 | // must be called under protection of the Cache lock | |
314 | void _clear(Cache* cache); | |
315 | ||
316 | // return value is the highest cache_private of a trimmed buffer, or 0. | |
317 | int discard(Cache* cache, uint32_t offset, uint32_t length) { | |
318 | std::lock_guard<std::recursive_mutex> l(cache->lock); | |
319 | return _discard(cache, offset, length); | |
320 | } | |
321 | int _discard(Cache* cache, uint32_t offset, uint32_t length); | |
322 | ||
323 | void write(Cache* cache, uint64_t seq, uint32_t offset, bufferlist& bl, | |
324 | unsigned flags) { | |
325 | std::lock_guard<std::recursive_mutex> l(cache->lock); | |
326 | Buffer *b = new Buffer(this, Buffer::STATE_WRITING, seq, offset, bl, | |
327 | flags); | |
328 | b->cache_private = _discard(cache, offset, bl.length()); | |
329 | _add_buffer(cache, b, (flags & Buffer::FLAG_NOCACHE) ? 0 : 1, nullptr); | |
330 | } | |
331 | void finish_write(Cache* cache, uint64_t seq); | |
332 | void did_read(Cache* cache, uint32_t offset, bufferlist& bl) { | |
333 | std::lock_guard<std::recursive_mutex> l(cache->lock); | |
334 | Buffer *b = new Buffer(this, Buffer::STATE_CLEAN, 0, offset, bl); | |
335 | b->cache_private = _discard(cache, offset, bl.length()); | |
336 | _add_buffer(cache, b, 1, nullptr); | |
337 | } | |
338 | ||
339 | void read(Cache* cache, uint32_t offset, uint32_t length, | |
340 | BlueStore::ready_regions_t& res, | |
341 | interval_set<uint32_t>& res_intervals); | |
342 | ||
343 | void truncate(Cache* cache, uint32_t offset) { | |
344 | discard(cache, offset, (uint32_t)-1 - offset); | |
345 | } | |
346 | ||
347 | void split(Cache* cache, size_t pos, BufferSpace &r); | |
348 | ||
349 | void dump(Cache* cache, Formatter *f) const { | |
350 | std::lock_guard<std::recursive_mutex> l(cache->lock); | |
351 | f->open_array_section("buffers"); | |
352 | for (auto& i : buffer_map) { | |
353 | f->open_object_section("buffer"); | |
354 | assert(i.first == i.second->offset); | |
355 | i.second->dump(f); | |
356 | f->close_section(); | |
357 | } | |
358 | f->close_section(); | |
359 | } | |
360 | }; | |
361 | ||
362 | struct SharedBlobSet; | |
363 | ||
364 | /// in-memory shared blob state (incl cached buffers) | |
365 | struct SharedBlob { | |
366 | MEMPOOL_CLASS_HELPERS(); | |
367 | ||
368 | std::atomic_int nref = {0}; ///< reference count | |
369 | bool loaded = false; | |
370 | ||
371 | CollectionRef coll; | |
372 | union { | |
373 | uint64_t sbid_unloaded; ///< sbid if persistent isn't loaded | |
374 | bluestore_shared_blob_t *persistent; ///< persistent part of the shared blob if any | |
375 | }; | |
376 | BufferSpace bc; ///< buffer cache | |
377 | ||
378 | SharedBlob(Collection *_coll) : coll(_coll), sbid_unloaded(0) { | |
379 | if (get_cache()) { | |
380 | get_cache()->add_blob(); | |
381 | } | |
382 | } | |
383 | SharedBlob(uint64_t i, Collection *_coll); | |
384 | ~SharedBlob(); | |
385 | ||
386 | uint64_t get_sbid() const { | |
387 | return loaded ? persistent->sbid : sbid_unloaded; | |
388 | } | |
389 | ||
390 | friend void intrusive_ptr_add_ref(SharedBlob *b) { b->get(); } | |
391 | friend void intrusive_ptr_release(SharedBlob *b) { b->put(); } | |
392 | ||
393 | friend ostream& operator<<(ostream& out, const SharedBlob& sb); | |
394 | ||
395 | void get() { | |
396 | ++nref; | |
397 | } | |
398 | void put(); | |
399 | ||
400 | /// get logical references | |
401 | void get_ref(uint64_t offset, uint32_t length); | |
402 | ||
403 | /// put logical references, and get back any released extents | |
404 | void put_ref(uint64_t offset, uint32_t length, | |
31f18b77 | 405 | PExtentVector *r, set<SharedBlob*> *maybe_unshared_blobs); |
7c673cae FG |
406 | |
407 | friend bool operator==(const SharedBlob &l, const SharedBlob &r) { | |
408 | return l.get_sbid() == r.get_sbid(); | |
409 | } | |
410 | inline Cache* get_cache() { | |
411 | return coll ? coll->cache : nullptr; | |
412 | } | |
413 | inline SharedBlobSet* get_parent() { | |
414 | return coll ? &(coll->shared_blob_set) : nullptr; | |
415 | } | |
416 | inline bool is_loaded() const { | |
417 | return loaded; | |
418 | } | |
419 | ||
420 | }; | |
421 | typedef boost::intrusive_ptr<SharedBlob> SharedBlobRef; | |
422 | ||
423 | /// a lookup table of SharedBlobs | |
424 | struct SharedBlobSet { | |
425 | std::mutex lock; ///< protect lookup, insertion, removal | |
426 | ||
427 | // we use a bare pointer because we don't want to affect the ref | |
428 | // count | |
31f18b77 | 429 | mempool::bluestore_cache_other::unordered_map<uint64_t,SharedBlob*> sb_map; |
7c673cae FG |
430 | |
431 | SharedBlobRef lookup(uint64_t sbid) { | |
432 | std::lock_guard<std::mutex> l(lock); | |
433 | auto p = sb_map.find(sbid); | |
434 | if (p == sb_map.end()) { | |
435 | return nullptr; | |
436 | } | |
437 | return p->second; | |
438 | } | |
439 | ||
440 | void add(Collection* coll, SharedBlob *sb) { | |
441 | std::lock_guard<std::mutex> l(lock); | |
442 | sb_map[sb->get_sbid()] = sb; | |
443 | sb->coll = coll; | |
444 | } | |
445 | ||
3efd9988 | 446 | bool try_remove(SharedBlob *sb) { |
7c673cae FG |
447 | std::lock_guard<std::mutex> l(lock); |
448 | if (sb->nref == 0) { | |
449 | assert(sb->get_parent() == this); | |
450 | sb_map.erase(sb->get_sbid()); | |
451 | return true; | |
452 | } | |
453 | return false; | |
454 | } | |
455 | ||
3efd9988 FG |
456 | void remove(SharedBlob *sb) { |
457 | std::lock_guard<std::mutex> l(lock); | |
458 | assert(sb->get_parent() == this); | |
459 | sb_map.erase(sb->get_sbid()); | |
460 | } | |
461 | ||
7c673cae FG |
462 | bool empty() { |
463 | std::lock_guard<std::mutex> l(lock); | |
464 | return sb_map.empty(); | |
465 | } | |
3efd9988 FG |
466 | |
467 | void dump(CephContext *cct, int lvl); | |
7c673cae FG |
468 | }; |
469 | ||
470 | //#define CACHE_BLOB_BL // not sure if this is a win yet or not... :/ | |
471 | ||
472 | /// in-memory blob metadata and associated cached buffers (if any) | |
473 | struct Blob { | |
474 | MEMPOOL_CLASS_HELPERS(); | |
475 | ||
476 | std::atomic_int nref = {0}; ///< reference count | |
477 | int16_t id = -1; ///< id, for spanning blobs only, >= 0 | |
478 | int16_t last_encoded_id = -1; ///< (ephemeral) used during encoding only | |
479 | SharedBlobRef shared_blob; ///< shared blob state (if any) | |
480 | ||
481 | private: | |
482 | mutable bluestore_blob_t blob; ///< decoded blob metadata | |
483 | #ifdef CACHE_BLOB_BL | |
484 | mutable bufferlist blob_bl; ///< cached encoded blob, blob is dirty if empty | |
485 | #endif | |
486 | /// refs from this shard. ephemeral if id<0, persisted if spanning. | |
487 | bluestore_blob_use_tracker_t used_in_blob; | |
488 | ||
489 | public: | |
490 | ||
491 | friend void intrusive_ptr_add_ref(Blob *b) { b->get(); } | |
492 | friend void intrusive_ptr_release(Blob *b) { b->put(); } | |
493 | ||
494 | friend ostream& operator<<(ostream& out, const Blob &b); | |
495 | ||
496 | const bluestore_blob_use_tracker_t& get_blob_use_tracker() const { | |
497 | return used_in_blob; | |
498 | } | |
499 | bool is_referenced() const { | |
500 | return used_in_blob.is_not_empty(); | |
501 | } | |
502 | uint32_t get_referenced_bytes() const { | |
503 | return used_in_blob.get_referenced_bytes(); | |
504 | } | |
505 | ||
506 | bool is_spanning() const { | |
507 | return id >= 0; | |
508 | } | |
509 | ||
510 | bool can_split() const { | |
511 | std::lock_guard<std::recursive_mutex> l(shared_blob->get_cache()->lock); | |
512 | // splitting a BufferSpace writing list is too hard; don't try. | |
513 | return shared_blob->bc.writing.empty() && | |
514 | used_in_blob.can_split() && | |
515 | get_blob().can_split(); | |
516 | } | |
517 | ||
518 | bool can_split_at(uint32_t blob_offset) const { | |
519 | return used_in_blob.can_split_at(blob_offset) && | |
520 | get_blob().can_split_at(blob_offset); | |
521 | } | |
522 | ||
224ce89b | 523 | bool can_reuse_blob(uint32_t min_alloc_size, |
7c673cae FG |
524 | uint32_t target_blob_size, |
525 | uint32_t b_offset, | |
526 | uint32_t *length0); | |
527 | ||
528 | void dup(Blob& o) { | |
529 | o.shared_blob = shared_blob; | |
530 | o.blob = blob; | |
531 | #ifdef CACHE_BLOB_BL | |
532 | o.blob_bl = blob_bl; | |
533 | #endif | |
534 | } | |
535 | ||
224ce89b | 536 | inline const bluestore_blob_t& get_blob() const { |
7c673cae FG |
537 | return blob; |
538 | } | |
224ce89b | 539 | inline bluestore_blob_t& dirty_blob() { |
7c673cae FG |
540 | #ifdef CACHE_BLOB_BL |
541 | blob_bl.clear(); | |
542 | #endif | |
543 | return blob; | |
544 | } | |
545 | ||
546 | /// discard buffers for unallocated regions | |
547 | void discard_unallocated(Collection *coll); | |
548 | ||
549 | /// get logical references | |
550 | void get_ref(Collection *coll, uint32_t offset, uint32_t length); | |
551 | /// put logical references, and get back any released extents | |
552 | bool put_ref(Collection *coll, uint32_t offset, uint32_t length, | |
553 | PExtentVector *r); | |
554 | ||
555 | /// split the blob | |
556 | void split(Collection *coll, uint32_t blob_offset, Blob *o); | |
557 | ||
558 | void get() { | |
559 | ++nref; | |
560 | } | |
561 | void put() { | |
562 | if (--nref == 0) | |
563 | delete this; | |
564 | } | |
565 | ||
566 | ||
567 | #ifdef CACHE_BLOB_BL | |
568 | void _encode() const { | |
569 | if (blob_bl.length() == 0 ) { | |
570 | ::encode(blob, blob_bl); | |
571 | } else { | |
572 | assert(blob_bl.length()); | |
573 | } | |
574 | } | |
575 | void bound_encode( | |
576 | size_t& p, | |
577 | bool include_ref_map) const { | |
578 | _encode(); | |
579 | p += blob_bl.length(); | |
580 | if (include_ref_map) { | |
581 | used_in_blob.bound_encode(p); | |
582 | } | |
583 | } | |
584 | void encode( | |
585 | bufferlist::contiguous_appender& p, | |
586 | bool include_ref_map) const { | |
587 | _encode(); | |
588 | p.append(blob_bl); | |
589 | if (include_ref_map) { | |
590 | used_in_blob.encode(p); | |
591 | } | |
592 | } | |
593 | void decode( | |
594 | Collection */*coll*/, | |
595 | bufferptr::iterator& p, | |
596 | bool include_ref_map) { | |
597 | const char *start = p.get_pos(); | |
598 | denc(blob, p); | |
599 | const char *end = p.get_pos(); | |
600 | blob_bl.clear(); | |
601 | blob_bl.append(start, end - start); | |
602 | if (include_ref_map) { | |
603 | used_in_blob.decode(p); | |
604 | } | |
605 | } | |
606 | #else | |
607 | void bound_encode( | |
608 | size_t& p, | |
609 | uint64_t struct_v, | |
610 | uint64_t sbid, | |
611 | bool include_ref_map) const { | |
612 | denc(blob, p, struct_v); | |
613 | if (blob.is_shared()) { | |
614 | denc(sbid, p); | |
615 | } | |
616 | if (include_ref_map) { | |
617 | used_in_blob.bound_encode(p); | |
618 | } | |
619 | } | |
620 | void encode( | |
621 | bufferlist::contiguous_appender& p, | |
622 | uint64_t struct_v, | |
623 | uint64_t sbid, | |
624 | bool include_ref_map) const { | |
625 | denc(blob, p, struct_v); | |
626 | if (blob.is_shared()) { | |
627 | denc(sbid, p); | |
628 | } | |
629 | if (include_ref_map) { | |
630 | used_in_blob.encode(p); | |
631 | } | |
632 | } | |
633 | void decode( | |
634 | Collection *coll, | |
635 | bufferptr::iterator& p, | |
636 | uint64_t struct_v, | |
637 | uint64_t* sbid, | |
638 | bool include_ref_map); | |
639 | #endif | |
640 | }; | |
641 | typedef boost::intrusive_ptr<Blob> BlobRef; | |
31f18b77 | 642 | typedef mempool::bluestore_cache_other::map<int,BlobRef> blob_map_t; |
7c673cae FG |
643 | |
644 | /// a logical extent, pointing to (some portion of) a blob | |
645 | typedef boost::intrusive::set_base_hook<boost::intrusive::optimize_size<true> > ExtentBase; //making an alias to avoid build warnings | |
646 | struct Extent : public ExtentBase { | |
647 | MEMPOOL_CLASS_HELPERS(); | |
648 | ||
649 | uint32_t logical_offset = 0; ///< logical offset | |
650 | uint32_t blob_offset = 0; ///< blob offset | |
651 | uint32_t length = 0; ///< length | |
652 | BlobRef blob; ///< the blob with our data | |
653 | ||
654 | /// ctor for lookup only | |
655 | explicit Extent(uint32_t lo) : ExtentBase(), logical_offset(lo) { } | |
656 | /// ctor for delayed initialization (see decode_some()) | |
657 | explicit Extent() : ExtentBase() { | |
658 | } | |
659 | /// ctor for general usage | |
660 | Extent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
661 | : ExtentBase(), | |
662 | logical_offset(lo), blob_offset(o), length(l) { | |
663 | assign_blob(b); | |
664 | } | |
665 | ~Extent() { | |
666 | if (blob) { | |
667 | blob->shared_blob->get_cache()->rm_extent(); | |
668 | } | |
669 | } | |
670 | ||
671 | void assign_blob(const BlobRef& b) { | |
672 | assert(!blob); | |
673 | blob = b; | |
674 | blob->shared_blob->get_cache()->add_extent(); | |
675 | } | |
676 | ||
677 | // comparators for intrusive_set | |
678 | friend bool operator<(const Extent &a, const Extent &b) { | |
679 | return a.logical_offset < b.logical_offset; | |
680 | } | |
681 | friend bool operator>(const Extent &a, const Extent &b) { | |
682 | return a.logical_offset > b.logical_offset; | |
683 | } | |
684 | friend bool operator==(const Extent &a, const Extent &b) { | |
685 | return a.logical_offset == b.logical_offset; | |
686 | } | |
687 | ||
688 | uint32_t blob_start() const { | |
689 | return logical_offset - blob_offset; | |
690 | } | |
691 | ||
692 | uint32_t blob_end() const { | |
693 | return blob_start() + blob->get_blob().get_logical_length(); | |
694 | } | |
695 | ||
696 | uint32_t logical_end() const { | |
697 | return logical_offset + length; | |
698 | } | |
699 | ||
700 | // return true if any piece of the blob is out of | |
701 | // the given range [o, o + l]. | |
702 | bool blob_escapes_range(uint32_t o, uint32_t l) const { | |
703 | return blob_start() < o || blob_end() > o + l; | |
704 | } | |
705 | }; | |
706 | typedef boost::intrusive::set<Extent> extent_map_t; | |
707 | ||
708 | ||
709 | friend ostream& operator<<(ostream& out, const Extent& e); | |
710 | ||
711 | struct OldExtent { | |
712 | boost::intrusive::list_member_hook<> old_extent_item; | |
713 | Extent e; | |
714 | PExtentVector r; | |
715 | bool blob_empty; // flag to track the last removed extent that makes blob | |
716 | // empty - required to update compression stat properly | |
717 | OldExtent(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) | |
718 | : e(lo, o, l, b), blob_empty(false) { | |
719 | } | |
720 | static OldExtent* create(CollectionRef c, | |
721 | uint32_t lo, | |
722 | uint32_t o, | |
723 | uint32_t l, | |
724 | BlobRef& b); | |
725 | }; | |
726 | typedef boost::intrusive::list< | |
727 | OldExtent, | |
728 | boost::intrusive::member_hook< | |
729 | OldExtent, | |
730 | boost::intrusive::list_member_hook<>, | |
731 | &OldExtent::old_extent_item> > old_extent_map_t; | |
732 | ||
733 | struct Onode; | |
734 | ||
735 | /// a sharded extent map, mapping offsets to lextents to blobs | |
736 | struct ExtentMap { | |
737 | Onode *onode; | |
738 | extent_map_t extent_map; ///< map of Extents to Blobs | |
739 | blob_map_t spanning_blob_map; ///< blobs that span shards | |
740 | ||
741 | struct Shard { | |
742 | bluestore_onode_t::shard_info *shard_info = nullptr; | |
743 | unsigned extents = 0; ///< count extents in this shard | |
744 | bool loaded = false; ///< true if shard is loaded | |
745 | bool dirty = false; ///< true if shard is dirty and needs reencoding | |
746 | }; | |
31f18b77 | 747 | mempool::bluestore_cache_other::vector<Shard> shards; ///< shards |
7c673cae FG |
748 | |
749 | bufferlist inline_bl; ///< cached encoded map, if unsharded; empty=>dirty | |
750 | ||
751 | uint32_t needs_reshard_begin = 0; | |
752 | uint32_t needs_reshard_end = 0; | |
753 | ||
754 | bool needs_reshard() const { | |
755 | return needs_reshard_end > needs_reshard_begin; | |
756 | } | |
757 | void clear_needs_reshard() { | |
758 | needs_reshard_begin = needs_reshard_end = 0; | |
759 | } | |
760 | void request_reshard(uint32_t begin, uint32_t end) { | |
761 | if (begin < needs_reshard_begin) { | |
762 | needs_reshard_begin = begin; | |
763 | } | |
764 | if (end > needs_reshard_end) { | |
765 | needs_reshard_end = end; | |
766 | } | |
767 | } | |
768 | ||
769 | struct DeleteDisposer { | |
770 | void operator()(Extent *e) { delete e; } | |
771 | }; | |
772 | ||
773 | ExtentMap(Onode *o); | |
774 | ~ExtentMap() { | |
775 | extent_map.clear_and_dispose(DeleteDisposer()); | |
776 | } | |
777 | ||
778 | void clear() { | |
779 | extent_map.clear_and_dispose(DeleteDisposer()); | |
780 | shards.clear(); | |
781 | inline_bl.clear(); | |
782 | clear_needs_reshard(); | |
783 | } | |
784 | ||
785 | bool encode_some(uint32_t offset, uint32_t length, bufferlist& bl, | |
786 | unsigned *pn); | |
787 | unsigned decode_some(bufferlist& bl); | |
788 | ||
789 | void bound_encode_spanning_blobs(size_t& p); | |
790 | void encode_spanning_blobs(bufferlist::contiguous_appender& p); | |
791 | void decode_spanning_blobs(bufferptr::iterator& p); | |
792 | ||
793 | BlobRef get_spanning_blob(int id) { | |
794 | auto p = spanning_blob_map.find(id); | |
795 | assert(p != spanning_blob_map.end()); | |
796 | return p->second; | |
797 | } | |
798 | ||
799 | void update(KeyValueDB::Transaction t, bool force); | |
31f18b77 | 800 | decltype(BlueStore::Blob::id) allocate_spanning_blob_id(); |
7c673cae FG |
801 | void reshard( |
802 | KeyValueDB *db, | |
803 | KeyValueDB::Transaction t); | |
804 | ||
805 | /// initialize Shards from the onode | |
806 | void init_shards(bool loaded, bool dirty); | |
807 | ||
808 | /// return index of shard containing offset | |
809 | /// or -1 if not found | |
810 | int seek_shard(uint32_t offset) { | |
811 | size_t end = shards.size(); | |
812 | size_t mid, left = 0; | |
813 | size_t right = end; // one passed the right end | |
814 | ||
815 | while (left < right) { | |
816 | mid = left + (right - left) / 2; | |
817 | if (offset >= shards[mid].shard_info->offset) { | |
818 | size_t next = mid + 1; | |
819 | if (next >= end || offset < shards[next].shard_info->offset) | |
820 | return mid; | |
821 | //continue to search forwards | |
822 | left = next; | |
823 | } else { | |
824 | //continue to search backwards | |
825 | right = mid; | |
826 | } | |
827 | } | |
828 | ||
829 | return -1; // not found | |
830 | } | |
831 | ||
832 | /// check if a range spans a shard | |
833 | bool spans_shard(uint32_t offset, uint32_t length) { | |
834 | if (shards.empty()) { | |
835 | return false; | |
836 | } | |
837 | int s = seek_shard(offset); | |
838 | assert(s >= 0); | |
839 | if (s == (int)shards.size() - 1) { | |
840 | return false; // last shard | |
841 | } | |
842 | if (offset + length <= shards[s+1].shard_info->offset) { | |
843 | return false; | |
844 | } | |
845 | return true; | |
846 | } | |
847 | ||
848 | /// ensure that a range of the map is loaded | |
849 | void fault_range(KeyValueDB *db, | |
850 | uint32_t offset, uint32_t length); | |
851 | ||
852 | /// ensure a range of the map is marked dirty | |
31f18b77 | 853 | void dirty_range(uint32_t offset, uint32_t length); |
7c673cae | 854 | |
31f18b77 | 855 | /// for seek_lextent test |
7c673cae FG |
856 | extent_map_t::iterator find(uint64_t offset); |
857 | ||
7c673cae FG |
858 | /// seek to the first lextent including or after offset |
859 | extent_map_t::iterator seek_lextent(uint64_t offset); | |
860 | extent_map_t::const_iterator seek_lextent(uint64_t offset) const; | |
861 | ||
862 | /// add a new Extent | |
863 | void add(uint32_t lo, uint32_t o, uint32_t l, BlobRef& b) { | |
864 | extent_map.insert(*new Extent(lo, o, l, b)); | |
865 | } | |
866 | ||
867 | /// remove (and delete) an Extent | |
868 | void rm(extent_map_t::iterator p) { | |
869 | extent_map.erase_and_dispose(p, DeleteDisposer()); | |
870 | } | |
871 | ||
872 | bool has_any_lextents(uint64_t offset, uint64_t length); | |
873 | ||
874 | /// consolidate adjacent lextents in extent_map | |
875 | int compress_extent_map(uint64_t offset, uint64_t length); | |
876 | ||
877 | /// punch a logical hole. add lextents to deref to target list. | |
878 | void punch_hole(CollectionRef &c, | |
879 | uint64_t offset, uint64_t length, | |
880 | old_extent_map_t *old_extents); | |
881 | ||
882 | /// put new lextent into lextent_map overwriting existing ones if | |
883 | /// any and update references accordingly | |
884 | Extent *set_lextent(CollectionRef &c, | |
885 | uint64_t logical_offset, | |
886 | uint64_t offset, uint64_t length, | |
887 | BlobRef b, | |
888 | old_extent_map_t *old_extents); | |
889 | ||
890 | /// split a blob (and referring extents) | |
891 | BlobRef split_blob(BlobRef lb, uint32_t blob_offset, uint32_t pos); | |
892 | }; | |
893 | ||
894 | /// Compressed Blob Garbage collector | |
895 | /* | |
896 | The primary idea of the collector is to estimate a difference between | |
897 | allocation units(AU) currently present for compressed blobs and new AUs | |
898 | required to store that data uncompressed. | |
899 | Estimation is performed for protrusive extents within a logical range | |
900 | determined by a concatenation of old_extents collection and specific(current) | |
901 | write request. | |
902 | The root cause for old_extents use is the need to handle blob ref counts | |
903 | properly. Old extents still hold blob refs and hence we need to traverse | |
904 | the collection to determine if blob to be released. | |
905 | Protrusive extents are extents that fit into the blob set in action | |
906 | (ones that are below the logical range from above) but not removed totally | |
907 | due to the current write. | |
908 | E.g. for | |
909 | extent1 <loffs = 100, boffs = 100, len = 100> -> | |
910 | blob1<compressed, len_on_disk=4096, logical_len=8192> | |
911 | extent2 <loffs = 200, boffs = 200, len = 100> -> | |
912 | blob2<raw, len_on_disk=4096, llen=4096> | |
913 | extent3 <loffs = 300, boffs = 300, len = 100> -> | |
914 | blob1<compressed, len_on_disk=4096, llen=8192> | |
915 | extent4 <loffs = 4096, boffs = 0, len = 100> -> | |
916 | blob3<raw, len_on_disk=4096, llen=4096> | |
917 | write(300~100) | |
918 | protrusive extents are within the following ranges <0~300, 400~8192-400> | |
919 | In this case existing AUs that might be removed due to GC (i.e. blob1) | |
920 | use 2x4K bytes. | |
921 | And new AUs expected after GC = 0 since extent1 to be merged into blob2. | |
922 | Hence we should do a collect. | |
923 | */ | |
924 | class GarbageCollector | |
925 | { | |
926 | public: | |
927 | /// return amount of allocation units that might be saved due to GC | |
928 | int64_t estimate( | |
929 | uint64_t offset, | |
930 | uint64_t length, | |
931 | const ExtentMap& extent_map, | |
932 | const old_extent_map_t& old_extents, | |
933 | uint64_t min_alloc_size); | |
934 | ||
935 | /// return a collection of extents to perform GC on | |
936 | const vector<AllocExtent>& get_extents_to_collect() const { | |
937 | return extents_to_collect; | |
938 | } | |
939 | GarbageCollector(CephContext* _cct) : cct(_cct) {} | |
940 | ||
941 | private: | |
942 | struct BlobInfo { | |
943 | uint64_t referenced_bytes = 0; ///< amount of bytes referenced in blob | |
944 | int64_t expected_allocations = 0; ///< new alloc units required | |
945 | ///< in case of gc fulfilled | |
946 | bool collect_candidate = false; ///< indicate if blob has any extents | |
947 | ///< eligible for GC. | |
948 | extent_map_t::const_iterator first_lextent; ///< points to the first | |
949 | ///< lextent referring to | |
950 | ///< the blob if any. | |
951 | ///< collect_candidate flag | |
952 | ///< determines the validity | |
953 | extent_map_t::const_iterator last_lextent; ///< points to the last | |
954 | ///< lextent referring to | |
955 | ///< the blob if any. | |
956 | ||
957 | BlobInfo(uint64_t ref_bytes) : | |
958 | referenced_bytes(ref_bytes) { | |
959 | } | |
960 | }; | |
961 | CephContext* cct; | |
962 | map<Blob*, BlobInfo> affected_blobs; ///< compressed blobs and their ref_map | |
963 | ///< copies that are affected by the | |
964 | ///< specific write | |
965 | ||
966 | vector<AllocExtent> extents_to_collect; ///< protrusive extents that should | |
967 | ///< be collected if GC takes place | |
968 | ||
969 | boost::optional<uint64_t > used_alloc_unit; ///< last processed allocation | |
970 | ///< unit when traversing | |
971 | ///< protrusive extents. | |
972 | ///< Other extents mapped to | |
973 | ///< this AU to be ignored | |
974 | ///< (except the case where | |
975 | ///< uncompressed extent follows | |
976 | ///< compressed one - see below). | |
977 | BlobInfo* blob_info_counted = nullptr; ///< set if previous allocation unit | |
978 | ///< caused expected_allocations | |
979 | ///< counter increment at this blob. | |
980 | ///< if uncompressed extent follows | |
981 | ///< a decrement for the | |
982 | ///< expected_allocations counter | |
983 | ///< is needed | |
984 | int64_t expected_allocations = 0; ///< new alloc units required in case | |
985 | ///< of gc fulfilled | |
986 | int64_t expected_for_release = 0; ///< alloc units currently used by | |
987 | ///< compressed blobs that might | |
988 | ///< gone after GC | |
989 | uint64_t gc_start_offset; ///starting offset for GC | |
990 | uint64_t gc_end_offset; ///ending offset for GC | |
991 | ||
992 | protected: | |
993 | void process_protrusive_extents(const BlueStore::ExtentMap& extent_map, | |
994 | uint64_t start_offset, | |
995 | uint64_t end_offset, | |
996 | uint64_t start_touch_offset, | |
997 | uint64_t end_touch_offset, | |
998 | uint64_t min_alloc_size); | |
999 | }; | |
1000 | ||
1001 | struct OnodeSpace; | |
1002 | ||
1003 | /// an in-memory object | |
1004 | struct Onode { | |
1005 | MEMPOOL_CLASS_HELPERS(); | |
1006 | ||
1007 | std::atomic_int nref; ///< reference count | |
1008 | Collection *c; | |
1009 | ||
1010 | ghobject_t oid; | |
1011 | ||
1012 | /// key under PREFIX_OBJ where we are stored | |
31f18b77 | 1013 | mempool::bluestore_cache_other::string key; |
7c673cae FG |
1014 | |
1015 | boost::intrusive::list_member_hook<> lru_item; | |
1016 | ||
1017 | bluestore_onode_t onode; ///< metadata stored as value in kv store | |
1018 | bool exists; ///< true if object logically exists | |
1019 | ||
1020 | ExtentMap extent_map; | |
1021 | ||
1022 | // track txc's that have not been committed to kv store (and whose | |
1023 | // effects cannot be read via the kvdb read methods) | |
1024 | std::atomic<int> flushing_count = {0}; | |
1025 | std::mutex flush_lock; ///< protect flush_txns | |
1026 | std::condition_variable flush_cond; ///< wait here for uncommitted txns | |
1027 | ||
1028 | Onode(Collection *c, const ghobject_t& o, | |
31f18b77 | 1029 | const mempool::bluestore_cache_other::string& k) |
7c673cae FG |
1030 | : nref(0), |
1031 | c(c), | |
1032 | oid(o), | |
1033 | key(k), | |
1034 | exists(false), | |
1035 | extent_map(this) { | |
1036 | } | |
1037 | ||
1038 | void flush(); | |
1039 | void get() { | |
1040 | ++nref; | |
1041 | } | |
1042 | void put() { | |
1043 | if (--nref == 0) | |
1044 | delete this; | |
1045 | } | |
1046 | }; | |
1047 | typedef boost::intrusive_ptr<Onode> OnodeRef; | |
1048 | ||
1049 | ||
1050 | /// a cache (shard) of onodes and buffers | |
1051 | struct Cache { | |
1052 | CephContext* cct; | |
1053 | PerfCounters *logger; | |
1054 | std::recursive_mutex lock; ///< protect lru and other structures | |
1055 | ||
1056 | std::atomic<uint64_t> num_extents = {0}; | |
1057 | std::atomic<uint64_t> num_blobs = {0}; | |
1058 | ||
7c673cae FG |
1059 | static Cache *create(CephContext* cct, string type, PerfCounters *logger); |
1060 | ||
1061 | Cache(CephContext* cct) : cct(cct), logger(nullptr) {} | |
1062 | virtual ~Cache() {} | |
1063 | ||
1064 | virtual void _add_onode(OnodeRef& o, int level) = 0; | |
1065 | virtual void _rm_onode(OnodeRef& o) = 0; | |
1066 | virtual void _touch_onode(OnodeRef& o) = 0; | |
1067 | ||
1068 | virtual void _add_buffer(Buffer *b, int level, Buffer *near) = 0; | |
1069 | virtual void _rm_buffer(Buffer *b) = 0; | |
1070 | virtual void _move_buffer(Cache *src, Buffer *b) = 0; | |
1071 | virtual void _adjust_buffer_size(Buffer *b, int64_t delta) = 0; | |
1072 | virtual void _touch_buffer(Buffer *b) = 0; | |
1073 | ||
1074 | virtual uint64_t _get_num_onodes() = 0; | |
1075 | virtual uint64_t _get_buffer_bytes() = 0; | |
1076 | ||
1077 | void add_extent() { | |
1078 | ++num_extents; | |
1079 | } | |
1080 | void rm_extent() { | |
1081 | --num_extents; | |
1082 | } | |
1083 | ||
1084 | void add_blob() { | |
1085 | ++num_blobs; | |
1086 | } | |
1087 | void rm_blob() { | |
1088 | --num_blobs; | |
1089 | } | |
1090 | ||
31f18b77 FG |
1091 | void trim(uint64_t target_bytes, |
1092 | float target_meta_ratio, | |
1093 | float target_data_ratio, | |
7c673cae FG |
1094 | float bytes_per_onode); |
1095 | ||
1096 | void trim_all(); | |
1097 | ||
1098 | virtual void _trim(uint64_t onode_max, uint64_t buffer_max) = 0; | |
1099 | ||
1100 | virtual void add_stats(uint64_t *onodes, uint64_t *extents, | |
1101 | uint64_t *blobs, | |
1102 | uint64_t *buffers, | |
1103 | uint64_t *bytes) = 0; | |
1104 | ||
31f18b77 FG |
1105 | bool empty() { |
1106 | std::lock_guard<std::recursive_mutex> l(lock); | |
1107 | return _get_num_onodes() == 0 && _get_buffer_bytes() == 0; | |
1108 | } | |
1109 | ||
7c673cae FG |
1110 | #ifdef DEBUG_CACHE |
1111 | virtual void _audit(const char *s) = 0; | |
1112 | #else | |
1113 | void _audit(const char *s) { /* no-op */ } | |
1114 | #endif | |
1115 | }; | |
1116 | ||
1117 | /// simple LRU cache for onodes and buffers | |
1118 | struct LRUCache : public Cache { | |
1119 | private: | |
1120 | typedef boost::intrusive::list< | |
1121 | Onode, | |
1122 | boost::intrusive::member_hook< | |
1123 | Onode, | |
1124 | boost::intrusive::list_member_hook<>, | |
1125 | &Onode::lru_item> > onode_lru_list_t; | |
1126 | typedef boost::intrusive::list< | |
1127 | Buffer, | |
1128 | boost::intrusive::member_hook< | |
1129 | Buffer, | |
1130 | boost::intrusive::list_member_hook<>, | |
1131 | &Buffer::lru_item> > buffer_lru_list_t; | |
1132 | ||
1133 | onode_lru_list_t onode_lru; | |
1134 | ||
1135 | buffer_lru_list_t buffer_lru; | |
1136 | uint64_t buffer_size = 0; | |
1137 | ||
1138 | public: | |
1139 | LRUCache(CephContext* cct) : Cache(cct) {} | |
1140 | uint64_t _get_num_onodes() override { | |
1141 | return onode_lru.size(); | |
1142 | } | |
1143 | void _add_onode(OnodeRef& o, int level) override { | |
1144 | if (level > 0) | |
1145 | onode_lru.push_front(*o); | |
1146 | else | |
1147 | onode_lru.push_back(*o); | |
1148 | } | |
1149 | void _rm_onode(OnodeRef& o) override { | |
1150 | auto q = onode_lru.iterator_to(*o); | |
1151 | onode_lru.erase(q); | |
1152 | } | |
1153 | void _touch_onode(OnodeRef& o) override; | |
1154 | ||
1155 | uint64_t _get_buffer_bytes() override { | |
1156 | return buffer_size; | |
1157 | } | |
1158 | void _add_buffer(Buffer *b, int level, Buffer *near) override { | |
1159 | if (near) { | |
1160 | auto q = buffer_lru.iterator_to(*near); | |
1161 | buffer_lru.insert(q, *b); | |
1162 | } else if (level > 0) { | |
1163 | buffer_lru.push_front(*b); | |
1164 | } else { | |
1165 | buffer_lru.push_back(*b); | |
1166 | } | |
1167 | buffer_size += b->length; | |
1168 | } | |
1169 | void _rm_buffer(Buffer *b) override { | |
1170 | assert(buffer_size >= b->length); | |
1171 | buffer_size -= b->length; | |
1172 | auto q = buffer_lru.iterator_to(*b); | |
1173 | buffer_lru.erase(q); | |
1174 | } | |
1175 | void _move_buffer(Cache *src, Buffer *b) override { | |
1176 | src->_rm_buffer(b); | |
1177 | _add_buffer(b, 0, nullptr); | |
1178 | } | |
1179 | void _adjust_buffer_size(Buffer *b, int64_t delta) override { | |
1180 | assert((int64_t)buffer_size + delta >= 0); | |
1181 | buffer_size += delta; | |
1182 | } | |
1183 | void _touch_buffer(Buffer *b) override { | |
1184 | auto p = buffer_lru.iterator_to(*b); | |
1185 | buffer_lru.erase(p); | |
1186 | buffer_lru.push_front(*b); | |
1187 | _audit("_touch_buffer end"); | |
1188 | } | |
1189 | ||
1190 | void _trim(uint64_t onode_max, uint64_t buffer_max) override; | |
1191 | ||
1192 | void add_stats(uint64_t *onodes, uint64_t *extents, | |
1193 | uint64_t *blobs, | |
1194 | uint64_t *buffers, | |
1195 | uint64_t *bytes) override { | |
1196 | std::lock_guard<std::recursive_mutex> l(lock); | |
1197 | *onodes += onode_lru.size(); | |
1198 | *extents += num_extents; | |
1199 | *blobs += num_blobs; | |
1200 | *buffers += buffer_lru.size(); | |
1201 | *bytes += buffer_size; | |
1202 | } | |
1203 | ||
1204 | #ifdef DEBUG_CACHE | |
1205 | void _audit(const char *s) override; | |
1206 | #endif | |
1207 | }; | |
1208 | ||
1209 | // 2Q cache for buffers, LRU for onodes | |
1210 | struct TwoQCache : public Cache { | |
1211 | private: | |
1212 | // stick with LRU for onodes for now (fixme?) | |
1213 | typedef boost::intrusive::list< | |
1214 | Onode, | |
1215 | boost::intrusive::member_hook< | |
1216 | Onode, | |
1217 | boost::intrusive::list_member_hook<>, | |
1218 | &Onode::lru_item> > onode_lru_list_t; | |
1219 | typedef boost::intrusive::list< | |
1220 | Buffer, | |
1221 | boost::intrusive::member_hook< | |
1222 | Buffer, | |
1223 | boost::intrusive::list_member_hook<>, | |
1224 | &Buffer::lru_item> > buffer_list_t; | |
1225 | ||
1226 | onode_lru_list_t onode_lru; | |
1227 | ||
1228 | buffer_list_t buffer_hot; ///< "Am" hot buffers | |
1229 | buffer_list_t buffer_warm_in; ///< "A1in" newly warm buffers | |
1230 | buffer_list_t buffer_warm_out; ///< "A1out" empty buffers we've evicted | |
1231 | uint64_t buffer_bytes = 0; ///< bytes | |
1232 | ||
1233 | enum { | |
1234 | BUFFER_NEW = 0, | |
1235 | BUFFER_WARM_IN, ///< in buffer_warm_in | |
1236 | BUFFER_WARM_OUT, ///< in buffer_warm_out | |
1237 | BUFFER_HOT, ///< in buffer_hot | |
1238 | BUFFER_TYPE_MAX | |
1239 | }; | |
1240 | ||
1241 | uint64_t buffer_list_bytes[BUFFER_TYPE_MAX] = {0}; ///< bytes per type | |
1242 | ||
1243 | public: | |
1244 | TwoQCache(CephContext* cct) : Cache(cct) {} | |
1245 | uint64_t _get_num_onodes() override { | |
1246 | return onode_lru.size(); | |
1247 | } | |
1248 | void _add_onode(OnodeRef& o, int level) override { | |
1249 | if (level > 0) | |
1250 | onode_lru.push_front(*o); | |
1251 | else | |
1252 | onode_lru.push_back(*o); | |
1253 | } | |
1254 | void _rm_onode(OnodeRef& o) override { | |
1255 | auto q = onode_lru.iterator_to(*o); | |
1256 | onode_lru.erase(q); | |
1257 | } | |
1258 | void _touch_onode(OnodeRef& o) override; | |
1259 | ||
1260 | uint64_t _get_buffer_bytes() override { | |
1261 | return buffer_bytes; | |
1262 | } | |
1263 | void _add_buffer(Buffer *b, int level, Buffer *near) override; | |
1264 | void _rm_buffer(Buffer *b) override; | |
1265 | void _move_buffer(Cache *src, Buffer *b) override; | |
1266 | void _adjust_buffer_size(Buffer *b, int64_t delta) override; | |
1267 | void _touch_buffer(Buffer *b) override { | |
1268 | switch (b->cache_private) { | |
1269 | case BUFFER_WARM_IN: | |
1270 | // do nothing (somewhat counter-intuitively!) | |
1271 | break; | |
1272 | case BUFFER_WARM_OUT: | |
1273 | // move from warm_out to hot LRU | |
1274 | assert(0 == "this happens via discard hint"); | |
1275 | break; | |
1276 | case BUFFER_HOT: | |
1277 | // move to front of hot LRU | |
1278 | buffer_hot.erase(buffer_hot.iterator_to(*b)); | |
1279 | buffer_hot.push_front(*b); | |
1280 | break; | |
1281 | } | |
1282 | _audit("_touch_buffer end"); | |
1283 | } | |
1284 | ||
1285 | void _trim(uint64_t onode_max, uint64_t buffer_max) override; | |
1286 | ||
1287 | void add_stats(uint64_t *onodes, uint64_t *extents, | |
1288 | uint64_t *blobs, | |
1289 | uint64_t *buffers, | |
1290 | uint64_t *bytes) override { | |
1291 | std::lock_guard<std::recursive_mutex> l(lock); | |
1292 | *onodes += onode_lru.size(); | |
1293 | *extents += num_extents; | |
1294 | *blobs += num_blobs; | |
1295 | *buffers += buffer_hot.size() + buffer_warm_in.size(); | |
1296 | *bytes += buffer_bytes; | |
1297 | } | |
1298 | ||
1299 | #ifdef DEBUG_CACHE | |
1300 | void _audit(const char *s) override; | |
1301 | #endif | |
1302 | }; | |
1303 | ||
1304 | struct OnodeSpace { | |
1305 | private: | |
1306 | Cache *cache; | |
1307 | ||
1308 | /// forward lookups | |
31f18b77 | 1309 | mempool::bluestore_cache_other::unordered_map<ghobject_t,OnodeRef> onode_map; |
7c673cae FG |
1310 | |
1311 | friend class Collection; // for split_cache() | |
1312 | ||
1313 | public: | |
1314 | OnodeSpace(Cache *c) : cache(c) {} | |
1315 | ~OnodeSpace() { | |
1316 | clear(); | |
1317 | } | |
1318 | ||
1319 | OnodeRef add(const ghobject_t& oid, OnodeRef o); | |
1320 | OnodeRef lookup(const ghobject_t& o); | |
1321 | void remove(const ghobject_t& oid) { | |
1322 | onode_map.erase(oid); | |
1323 | } | |
1324 | void rename(OnodeRef& o, const ghobject_t& old_oid, | |
1325 | const ghobject_t& new_oid, | |
31f18b77 | 1326 | const mempool::bluestore_cache_other::string& new_okey); |
7c673cae FG |
1327 | void clear(); |
1328 | bool empty(); | |
1329 | ||
3efd9988 FG |
1330 | void dump(CephContext *cct, int lvl); |
1331 | ||
7c673cae FG |
1332 | /// return true if f true for any item |
1333 | bool map_any(std::function<bool(OnodeRef)> f); | |
1334 | }; | |
1335 | ||
1336 | struct Collection : public CollectionImpl { | |
1337 | BlueStore *store; | |
1338 | Cache *cache; ///< our cache shard | |
1339 | coll_t cid; | |
1340 | bluestore_cnode_t cnode; | |
1341 | RWLock lock; | |
1342 | ||
1343 | bool exists; | |
1344 | ||
1345 | SharedBlobSet shared_blob_set; ///< open SharedBlobs | |
1346 | ||
1347 | // cache onodes on a per-collection basis to avoid lock | |
1348 | // contention. | |
1349 | OnodeSpace onode_map; | |
1350 | ||
1351 | //pool options | |
1352 | pool_opts_t pool_opts; | |
1353 | ||
1354 | OnodeRef get_onode(const ghobject_t& oid, bool create); | |
1355 | ||
1356 | // the terminology is confusing here, sorry! | |
1357 | // | |
1358 | // blob_t shared_blob_t | |
1359 | // !shared unused -> open | |
1360 | // shared !loaded -> open + shared | |
1361 | // shared loaded -> open + shared + loaded | |
1362 | // | |
1363 | // i.e., | |
1364 | // open = SharedBlob is instantiated | |
1365 | // shared = blob_t shared flag is set; SharedBlob is hashed. | |
1366 | // loaded = SharedBlob::shared_blob_t is loaded from kv store | |
1367 | void open_shared_blob(uint64_t sbid, BlobRef b); | |
1368 | void load_shared_blob(SharedBlobRef sb); | |
1369 | void make_blob_shared(uint64_t sbid, BlobRef b); | |
31f18b77 | 1370 | uint64_t make_blob_unshared(SharedBlob *sb); |
7c673cae FG |
1371 | |
1372 | BlobRef new_blob() { | |
1373 | BlobRef b = new Blob(); | |
1374 | b->shared_blob = new SharedBlob(this); | |
1375 | return b; | |
1376 | } | |
1377 | ||
1378 | const coll_t &get_cid() override { | |
1379 | return cid; | |
1380 | } | |
1381 | ||
1382 | bool contains(const ghobject_t& oid) { | |
1383 | if (cid.is_meta()) | |
1384 | return oid.hobj.pool == -1; | |
1385 | spg_t spgid; | |
1386 | if (cid.is_pg(&spgid)) | |
1387 | return | |
1388 | spgid.pgid.contains(cnode.bits, oid) && | |
1389 | oid.shard_id == spgid.shard; | |
1390 | return false; | |
1391 | } | |
1392 | ||
1393 | void split_cache(Collection *dest); | |
7c673cae FG |
1394 | |
1395 | Collection(BlueStore *ns, Cache *ca, coll_t c); | |
1396 | }; | |
1397 | ||
1398 | class OmapIteratorImpl : public ObjectMap::ObjectMapIteratorImpl { | |
1399 | CollectionRef c; | |
1400 | OnodeRef o; | |
1401 | KeyValueDB::Iterator it; | |
1402 | string head, tail; | |
1403 | public: | |
1404 | OmapIteratorImpl(CollectionRef c, OnodeRef o, KeyValueDB::Iterator it); | |
1405 | int seek_to_first() override; | |
1406 | int upper_bound(const string &after) override; | |
1407 | int lower_bound(const string &to) override; | |
1408 | bool valid() override; | |
1409 | int next(bool validate=true) override; | |
1410 | string key() override; | |
1411 | bufferlist value() override; | |
1412 | int status() override { | |
1413 | return 0; | |
1414 | } | |
1415 | }; | |
1416 | ||
1417 | class OpSequencer; | |
1418 | typedef boost::intrusive_ptr<OpSequencer> OpSequencerRef; | |
1419 | ||
31f18b77 FG |
1420 | struct volatile_statfs{ |
1421 | enum { | |
1422 | STATFS_ALLOCATED = 0, | |
1423 | STATFS_STORED, | |
1424 | STATFS_COMPRESSED_ORIGINAL, | |
1425 | STATFS_COMPRESSED, | |
1426 | STATFS_COMPRESSED_ALLOCATED, | |
1427 | STATFS_LAST | |
1428 | }; | |
1429 | int64_t values[STATFS_LAST]; | |
1430 | volatile_statfs() { | |
1431 | memset(this, 0, sizeof(volatile_statfs)); | |
1432 | } | |
1433 | void reset() { | |
1434 | *this = volatile_statfs(); | |
1435 | } | |
1436 | volatile_statfs& operator+=(const volatile_statfs& other) { | |
1437 | for (size_t i = 0; i < STATFS_LAST; ++i) { | |
1438 | values[i] += other.values[i]; | |
1439 | } | |
1440 | return *this; | |
1441 | } | |
1442 | int64_t& allocated() { | |
1443 | return values[STATFS_ALLOCATED]; | |
1444 | } | |
1445 | int64_t& stored() { | |
1446 | return values[STATFS_STORED]; | |
1447 | } | |
1448 | int64_t& compressed_original() { | |
1449 | return values[STATFS_COMPRESSED_ORIGINAL]; | |
1450 | } | |
1451 | int64_t& compressed() { | |
1452 | return values[STATFS_COMPRESSED]; | |
1453 | } | |
1454 | int64_t& compressed_allocated() { | |
1455 | return values[STATFS_COMPRESSED_ALLOCATED]; | |
1456 | } | |
1457 | bool is_empty() { | |
1458 | return values[STATFS_ALLOCATED] == 0 && | |
1459 | values[STATFS_STORED] == 0 && | |
1460 | values[STATFS_COMPRESSED] == 0 && | |
1461 | values[STATFS_COMPRESSED_ORIGINAL] == 0 && | |
1462 | values[STATFS_COMPRESSED_ALLOCATED] == 0; | |
1463 | } | |
1464 | void decode(bufferlist::iterator& it) { | |
1465 | for (size_t i = 0; i < STATFS_LAST; i++) { | |
1466 | ::decode(values[i], it); | |
1467 | } | |
1468 | } | |
1469 | ||
1470 | void encode(bufferlist& bl) { | |
1471 | for (size_t i = 0; i < STATFS_LAST; i++) { | |
1472 | ::encode(values[i], bl); | |
1473 | } | |
1474 | } | |
1475 | }; | |
1476 | ||
7c673cae | 1477 | struct TransContext : public AioContext { |
31f18b77 FG |
1478 | MEMPOOL_CLASS_HELPERS(); |
1479 | ||
7c673cae FG |
1480 | typedef enum { |
1481 | STATE_PREPARE, | |
1482 | STATE_AIO_WAIT, | |
1483 | STATE_IO_DONE, | |
1484 | STATE_KV_QUEUED, // queued for kv_sync_thread submission | |
1485 | STATE_KV_SUBMITTED, // submitted to kv; not yet synced | |
1486 | STATE_KV_DONE, | |
1487 | STATE_DEFERRED_QUEUED, // in deferred_queue (pending or running) | |
1488 | STATE_DEFERRED_CLEANUP, // remove deferred kv record | |
1489 | STATE_DEFERRED_DONE, | |
1490 | STATE_FINISHING, | |
1491 | STATE_DONE, | |
1492 | } state_t; | |
1493 | ||
1494 | state_t state = STATE_PREPARE; | |
1495 | ||
1496 | const char *get_state_name() { | |
1497 | switch (state) { | |
1498 | case STATE_PREPARE: return "prepare"; | |
1499 | case STATE_AIO_WAIT: return "aio_wait"; | |
1500 | case STATE_IO_DONE: return "io_done"; | |
1501 | case STATE_KV_QUEUED: return "kv_queued"; | |
1502 | case STATE_KV_SUBMITTED: return "kv_submitted"; | |
1503 | case STATE_KV_DONE: return "kv_done"; | |
1504 | case STATE_DEFERRED_QUEUED: return "deferred_queued"; | |
1505 | case STATE_DEFERRED_CLEANUP: return "deferred_cleanup"; | |
1506 | case STATE_DEFERRED_DONE: return "deferred_done"; | |
1507 | case STATE_FINISHING: return "finishing"; | |
1508 | case STATE_DONE: return "done"; | |
1509 | } | |
1510 | return "???"; | |
1511 | } | |
1512 | ||
1513 | #if defined(WITH_LTTNG) && defined(WITH_EVENTTRACE) | |
1514 | const char *get_state_latency_name(int state) { | |
1515 | switch (state) { | |
1516 | case l_bluestore_state_prepare_lat: return "prepare"; | |
1517 | case l_bluestore_state_aio_wait_lat: return "aio_wait"; | |
1518 | case l_bluestore_state_io_done_lat: return "io_done"; | |
1519 | case l_bluestore_state_kv_queued_lat: return "kv_queued"; | |
1520 | case l_bluestore_state_kv_committing_lat: return "kv_committing"; | |
1521 | case l_bluestore_state_kv_done_lat: return "kv_done"; | |
1522 | case l_bluestore_state_deferred_queued_lat: return "deferred_queued"; | |
1523 | case l_bluestore_state_deferred_cleanup_lat: return "deferred_cleanup"; | |
1524 | case l_bluestore_state_finishing_lat: return "finishing"; | |
1525 | case l_bluestore_state_done_lat: return "done"; | |
1526 | } | |
1527 | return "???"; | |
1528 | } | |
1529 | #endif | |
1530 | ||
1531 | void log_state_latency(PerfCounters *logger, int state) { | |
1532 | utime_t lat, now = ceph_clock_now(); | |
1533 | lat = now - last_stamp; | |
1534 | logger->tinc(state, lat); | |
1535 | #if defined(WITH_LTTNG) && defined(WITH_EVENTTRACE) | |
1536 | if (state >= l_bluestore_state_prepare_lat && state <= l_bluestore_state_done_lat) { | |
1537 | double usecs = (now.to_nsec()-last_stamp.to_nsec())/1000; | |
1538 | OID_ELAPSED("", usecs, get_state_latency_name(state)); | |
1539 | } | |
1540 | #endif | |
1541 | last_stamp = now; | |
1542 | } | |
1543 | ||
1544 | OpSequencerRef osr; | |
1545 | boost::intrusive::list_member_hook<> sequencer_item; | |
1546 | ||
1547 | uint64_t bytes = 0, cost = 0; | |
1548 | ||
1549 | set<OnodeRef> onodes; ///< these need to be updated/written | |
1550 | set<OnodeRef> modified_objects; ///< objects we modified (and need a ref) | |
1551 | set<SharedBlobRef> shared_blobs; ///< these need to be updated/written | |
1552 | set<SharedBlobRef> shared_blobs_written; ///< update these on io completion | |
1553 | ||
1554 | KeyValueDB::Transaction t; ///< then we will commit this | |
1555 | Context *oncommit = nullptr; ///< signal on commit | |
1556 | Context *onreadable = nullptr; ///< signal on readable | |
1557 | Context *onreadable_sync = nullptr; ///< signal on readable | |
1558 | list<Context*> oncommits; ///< more commit completions | |
1559 | list<CollectionRef> removed_collections; ///< colls we removed | |
1560 | ||
1561 | boost::intrusive::list_member_hook<> deferred_queue_item; | |
1562 | bluestore_deferred_transaction_t *deferred_txn = nullptr; ///< if any | |
1563 | ||
1564 | interval_set<uint64_t> allocated, released; | |
31f18b77 | 1565 | volatile_statfs statfs_delta; |
7c673cae FG |
1566 | |
1567 | IOContext ioc; | |
1568 | bool had_ios = false; ///< true if we submitted IOs before our kv txn | |
1569 | ||
7c673cae FG |
1570 | uint64_t seq = 0; |
1571 | utime_t start; | |
1572 | utime_t last_stamp; | |
1573 | ||
1574 | uint64_t last_nid = 0; ///< if non-zero, highest new nid we allocated | |
1575 | uint64_t last_blobid = 0; ///< if non-zero, highest new blobid we allocated | |
1576 | ||
1577 | explicit TransContext(CephContext* cct, OpSequencer *o) | |
1578 | : osr(o), | |
1579 | ioc(cct, this), | |
1580 | start(ceph_clock_now()) { | |
1581 | last_stamp = start; | |
1582 | } | |
1583 | ~TransContext() { | |
1584 | delete deferred_txn; | |
1585 | } | |
1586 | ||
1587 | void write_onode(OnodeRef &o) { | |
1588 | onodes.insert(o); | |
1589 | } | |
1590 | void write_shared_blob(SharedBlobRef &sb) { | |
1591 | shared_blobs.insert(sb); | |
1592 | } | |
31f18b77 FG |
1593 | void unshare_blob(SharedBlob *sb) { |
1594 | shared_blobs.erase(sb); | |
1595 | } | |
1596 | ||
7c673cae FG |
1597 | /// note we logically modified object (when onode itself is unmodified) |
1598 | void note_modified_object(OnodeRef &o) { | |
1599 | // onode itself isn't written, though | |
1600 | modified_objects.insert(o); | |
1601 | } | |
1602 | void removed(OnodeRef& o) { | |
1603 | onodes.erase(o); | |
1604 | modified_objects.erase(o); | |
1605 | } | |
1606 | ||
1607 | void aio_finish(BlueStore *store) override { | |
1608 | store->txc_aio_finish(this); | |
1609 | } | |
1610 | }; | |
1611 | ||
1612 | typedef boost::intrusive::list< | |
1613 | TransContext, | |
1614 | boost::intrusive::member_hook< | |
1615 | TransContext, | |
1616 | boost::intrusive::list_member_hook<>, | |
1617 | &TransContext::deferred_queue_item> > deferred_queue_t; | |
1618 | ||
1619 | struct DeferredBatch : public AioContext { | |
1620 | OpSequencer *osr; | |
1621 | struct deferred_io { | |
1622 | bufferlist bl; ///< data | |
1623 | uint64_t seq; ///< deferred transaction seq | |
1624 | }; | |
1625 | map<uint64_t,deferred_io> iomap; ///< map of ios in this batch | |
1626 | deferred_queue_t txcs; ///< txcs in this batch | |
1627 | IOContext ioc; ///< our aios | |
1628 | /// bytes of pending io for each deferred seq (may be 0) | |
1629 | map<uint64_t,int> seq_bytes; | |
1630 | ||
1631 | void _discard(CephContext *cct, uint64_t offset, uint64_t length); | |
1632 | void _audit(CephContext *cct); | |
1633 | ||
1634 | DeferredBatch(CephContext *cct, OpSequencer *osr) | |
1635 | : osr(osr), ioc(cct, this) {} | |
1636 | ||
1637 | /// prepare a write | |
1638 | void prepare_write(CephContext *cct, | |
1639 | uint64_t seq, uint64_t offset, uint64_t length, | |
1640 | bufferlist::const_iterator& p); | |
1641 | ||
1642 | void aio_finish(BlueStore *store) override { | |
1643 | store->_deferred_aio_finish(osr); | |
1644 | } | |
1645 | }; | |
1646 | ||
1647 | class OpSequencer : public Sequencer_impl { | |
1648 | public: | |
1649 | std::mutex qlock; | |
1650 | std::condition_variable qcond; | |
1651 | typedef boost::intrusive::list< | |
1652 | TransContext, | |
1653 | boost::intrusive::member_hook< | |
1654 | TransContext, | |
1655 | boost::intrusive::list_member_hook<>, | |
1656 | &TransContext::sequencer_item> > q_list_t; | |
1657 | q_list_t q; ///< transactions | |
1658 | ||
1659 | boost::intrusive::list_member_hook<> deferred_osr_queue_item; | |
1660 | ||
1661 | DeferredBatch *deferred_running = nullptr; | |
1662 | DeferredBatch *deferred_pending = nullptr; | |
1663 | ||
1664 | Sequencer *parent; | |
1665 | BlueStore *store; | |
1666 | ||
1667 | uint64_t last_seq = 0; | |
1668 | ||
1669 | std::atomic_int txc_with_unstable_io = {0}; ///< num txcs with unstable io | |
1670 | ||
1671 | std::atomic_int kv_committing_serially = {0}; | |
1672 | ||
1673 | std::atomic_int kv_submitted_waiters = {0}; | |
1674 | ||
1675 | std::atomic_bool registered = {true}; ///< registered in BlueStore's osr_set | |
1676 | std::atomic_bool zombie = {false}; ///< owning Sequencer has gone away | |
1677 | ||
1678 | OpSequencer(CephContext* cct, BlueStore *store) | |
1679 | : Sequencer_impl(cct), | |
1680 | parent(NULL), store(store) { | |
1681 | store->register_osr(this); | |
1682 | } | |
1683 | ~OpSequencer() override { | |
1684 | assert(q.empty()); | |
1685 | _unregister(); | |
1686 | } | |
1687 | ||
1688 | void discard() override { | |
1689 | // Note that we may have txc's in flight when the parent Sequencer | |
1690 | // goes away. Reflect this with zombie==registered==true and let | |
1691 | // _osr_drain_all clean up later. | |
1692 | assert(!zombie); | |
1693 | zombie = true; | |
1694 | parent = nullptr; | |
1695 | bool empty; | |
1696 | { | |
1697 | std::lock_guard<std::mutex> l(qlock); | |
1698 | empty = q.empty(); | |
1699 | } | |
1700 | if (empty) { | |
1701 | _unregister(); | |
1702 | } | |
1703 | } | |
1704 | ||
1705 | void _unregister() { | |
1706 | if (registered) { | |
1707 | store->unregister_osr(this); | |
1708 | registered = false; | |
1709 | } | |
1710 | } | |
1711 | ||
1712 | void queue_new(TransContext *txc) { | |
1713 | std::lock_guard<std::mutex> l(qlock); | |
1714 | txc->seq = ++last_seq; | |
1715 | q.push_back(*txc); | |
1716 | } | |
1717 | ||
1718 | void drain() { | |
1719 | std::unique_lock<std::mutex> l(qlock); | |
1720 | while (!q.empty()) | |
1721 | qcond.wait(l); | |
1722 | } | |
1723 | ||
1724 | void drain_preceding(TransContext *txc) { | |
1725 | std::unique_lock<std::mutex> l(qlock); | |
1726 | while (!q.empty() && &q.front() != txc) | |
1727 | qcond.wait(l); | |
1728 | } | |
1729 | ||
1730 | bool _is_all_kv_submitted() { | |
1731 | // caller must hold qlock | |
1732 | if (q.empty()) { | |
1733 | return true; | |
1734 | } | |
1735 | TransContext *txc = &q.back(); | |
1736 | if (txc->state >= TransContext::STATE_KV_SUBMITTED) { | |
1737 | return true; | |
1738 | } | |
1739 | return false; | |
1740 | } | |
1741 | ||
1742 | void flush() override { | |
1743 | std::unique_lock<std::mutex> l(qlock); | |
1744 | while (true) { | |
1745 | // set flag before the check because the condition | |
1746 | // may become true outside qlock, and we need to make | |
1747 | // sure those threads see waiters and signal qcond. | |
1748 | ++kv_submitted_waiters; | |
1749 | if (_is_all_kv_submitted()) { | |
1750 | return; | |
1751 | } | |
1752 | qcond.wait(l); | |
1753 | --kv_submitted_waiters; | |
1754 | } | |
1755 | } | |
1756 | ||
1757 | bool flush_commit(Context *c) override { | |
1758 | std::lock_guard<std::mutex> l(qlock); | |
1759 | if (q.empty()) { | |
1760 | return true; | |
1761 | } | |
1762 | TransContext *txc = &q.back(); | |
1763 | if (txc->state >= TransContext::STATE_KV_DONE) { | |
1764 | return true; | |
1765 | } | |
1766 | txc->oncommits.push_back(c); | |
1767 | return false; | |
1768 | } | |
1769 | }; | |
1770 | ||
1771 | typedef boost::intrusive::list< | |
1772 | OpSequencer, | |
1773 | boost::intrusive::member_hook< | |
1774 | OpSequencer, | |
1775 | boost::intrusive::list_member_hook<>, | |
1776 | &OpSequencer::deferred_osr_queue_item> > deferred_osr_queue_t; | |
1777 | ||
1778 | struct KVSyncThread : public Thread { | |
1779 | BlueStore *store; | |
1780 | explicit KVSyncThread(BlueStore *s) : store(s) {} | |
1781 | void *entry() override { | |
1782 | store->_kv_sync_thread(); | |
1783 | return NULL; | |
1784 | } | |
1785 | }; | |
31f18b77 FG |
1786 | struct KVFinalizeThread : public Thread { |
1787 | BlueStore *store; | |
1788 | explicit KVFinalizeThread(BlueStore *s) : store(s) {} | |
1789 | void *entry() { | |
1790 | store->_kv_finalize_thread(); | |
1791 | return NULL; | |
1792 | } | |
1793 | }; | |
7c673cae FG |
1794 | |
1795 | struct DBHistogram { | |
1796 | struct value_dist { | |
1797 | uint64_t count; | |
1798 | uint32_t max_len; | |
1799 | }; | |
1800 | ||
1801 | struct key_dist { | |
1802 | uint64_t count; | |
1803 | uint32_t max_len; | |
1804 | map<int, struct value_dist> val_map; ///< slab id to count, max length of value and key | |
1805 | }; | |
1806 | ||
1807 | map<string, map<int, struct key_dist> > key_hist; | |
1808 | map<int, uint64_t> value_hist; | |
1809 | int get_key_slab(size_t sz); | |
1810 | string get_key_slab_to_range(int slab); | |
1811 | int get_value_slab(size_t sz); | |
1812 | string get_value_slab_to_range(int slab); | |
1813 | void update_hist_entry(map<string, map<int, struct key_dist> > &key_hist, | |
1814 | const string &prefix, size_t key_size, size_t value_size); | |
1815 | void dump(Formatter *f); | |
1816 | }; | |
1817 | ||
1818 | // -------------------------------------------------------- | |
1819 | // members | |
1820 | private: | |
1821 | BlueFS *bluefs = nullptr; | |
1822 | unsigned bluefs_shared_bdev = 0; ///< which bluefs bdev we are sharing | |
1823 | bool bluefs_single_shared_device = true; | |
1824 | utime_t bluefs_last_balance; | |
1825 | ||
1826 | KeyValueDB *db = nullptr; | |
1827 | BlockDevice *bdev = nullptr; | |
1828 | std::string freelist_type; | |
1829 | FreelistManager *fm = nullptr; | |
1830 | Allocator *alloc = nullptr; | |
1831 | uuid_d fsid; | |
1832 | int path_fd = -1; ///< open handle to $path | |
1833 | int fsid_fd = -1; ///< open handle (locked) to $path/fsid | |
1834 | bool mounted = false; | |
1835 | ||
1836 | RWLock coll_lock = {"BlueStore::coll_lock"}; ///< rwlock to protect coll_map | |
31f18b77 | 1837 | mempool::bluestore_cache_other::unordered_map<coll_t, CollectionRef> coll_map; |
7c673cae FG |
1838 | |
1839 | vector<Cache*> cache_shards; | |
1840 | ||
1841 | std::mutex osr_lock; ///< protect osd_set | |
1842 | std::set<OpSequencerRef> osr_set; ///< set of all OpSequencers | |
1843 | ||
1844 | std::atomic<uint64_t> nid_last = {0}; | |
1845 | std::atomic<uint64_t> nid_max = {0}; | |
1846 | std::atomic<uint64_t> blobid_last = {0}; | |
1847 | std::atomic<uint64_t> blobid_max = {0}; | |
1848 | ||
1849 | Throttle throttle_bytes; ///< submit to commit | |
1850 | Throttle throttle_deferred_bytes; ///< submit to deferred complete | |
1851 | ||
1852 | interval_set<uint64_t> bluefs_extents; ///< block extents owned by bluefs | |
1853 | interval_set<uint64_t> bluefs_extents_reclaiming; ///< currently reclaiming | |
1854 | ||
181888fb | 1855 | std::mutex deferred_lock; |
7c673cae FG |
1856 | std::atomic<uint64_t> deferred_seq = {0}; |
1857 | deferred_osr_queue_t deferred_queue; ///< osr's with deferred io pending | |
1858 | int deferred_queue_size = 0; ///< num txc's queued across all osrs | |
1859 | atomic_int deferred_aggressive = {0}; ///< aggressive wakeup of kv thread | |
181888fb | 1860 | Finisher deferred_finisher; |
7c673cae FG |
1861 | |
1862 | int m_finisher_num = 1; | |
1863 | vector<Finisher*> finishers; | |
1864 | ||
1865 | KVSyncThread kv_sync_thread; | |
1866 | std::mutex kv_lock; | |
1867 | std::condition_variable kv_cond; | |
3efd9988 | 1868 | bool _kv_only = false; |
31f18b77 | 1869 | bool kv_sync_started = false; |
7c673cae | 1870 | bool kv_stop = false; |
31f18b77 FG |
1871 | bool kv_finalize_started = false; |
1872 | bool kv_finalize_stop = false; | |
7c673cae FG |
1873 | deque<TransContext*> kv_queue; ///< ready, already submitted |
1874 | deque<TransContext*> kv_queue_unsubmitted; ///< ready, need submit by kv thread | |
1875 | deque<TransContext*> kv_committing; ///< currently syncing | |
1876 | deque<DeferredBatch*> deferred_done_queue; ///< deferred ios done | |
1877 | deque<DeferredBatch*> deferred_stable_queue; ///< deferred ios done + stable | |
1878 | ||
31f18b77 FG |
1879 | KVFinalizeThread kv_finalize_thread; |
1880 | std::mutex kv_finalize_lock; | |
1881 | std::condition_variable kv_finalize_cond; | |
1882 | deque<TransContext*> kv_committing_to_finalize; ///< pending finalization | |
1883 | deque<DeferredBatch*> deferred_stable_to_finalize; ///< pending finalization | |
1884 | ||
7c673cae FG |
1885 | PerfCounters *logger = nullptr; |
1886 | ||
1887 | std::mutex reap_lock; | |
1888 | list<CollectionRef> removed_collections; | |
1889 | ||
1890 | RWLock debug_read_error_lock = {"BlueStore::debug_read_error_lock"}; | |
1891 | set<ghobject_t> debug_data_error_objects; | |
1892 | set<ghobject_t> debug_mdata_error_objects; | |
1893 | ||
1894 | std::atomic<int> csum_type = {Checksummer::CSUM_CRC32C}; | |
1895 | ||
1896 | uint64_t block_size = 0; ///< block size of block device (power of 2) | |
1897 | uint64_t block_mask = 0; ///< mask to get just the block offset | |
1898 | size_t block_size_order = 0; ///< bits to shift to get block size | |
1899 | ||
1900 | uint64_t min_alloc_size = 0; ///< minimum allocation unit (power of 2) | |
7c673cae | 1901 | ///< bits for min_alloc_size |
224ce89b | 1902 | uint8_t min_alloc_size_order = 0; |
7c673cae FG |
1903 | static_assert(std::numeric_limits<uint8_t>::max() > |
1904 | std::numeric_limits<decltype(min_alloc_size)>::digits, | |
1905 | "not enough bits for min_alloc_size"); | |
1906 | ||
7c673cae FG |
1907 | ///< maximum allocation unit (power of 2) |
1908 | std::atomic<uint64_t> max_alloc_size = {0}; | |
1909 | ||
224ce89b WB |
1910 | ///< number threshold for forced deferred writes |
1911 | std::atomic<int> deferred_batch_ops = {0}; | |
1912 | ||
1913 | ///< size threshold for forced deferred writes | |
1914 | std::atomic<uint64_t> prefer_deferred_size = {0}; | |
1915 | ||
7c673cae FG |
1916 | ///< approx cost per io, in bytes |
1917 | std::atomic<uint64_t> throttle_cost_per_io = {0}; | |
1918 | ||
224ce89b WB |
1919 | std::atomic<Compressor::CompressionMode> comp_mode = |
1920 | {Compressor::COMP_NONE}; ///< compression mode | |
7c673cae FG |
1921 | CompressorRef compressor; |
1922 | std::atomic<uint64_t> comp_min_blob_size = {0}; | |
1923 | std::atomic<uint64_t> comp_max_blob_size = {0}; | |
1924 | ||
1925 | std::atomic<uint64_t> max_blob_size = {0}; ///< maximum blob size | |
1926 | ||
31f18b77 FG |
1927 | uint64_t kv_ios = 0; |
1928 | uint64_t kv_throttle_costs = 0; | |
1929 | ||
7c673cae | 1930 | // cache trim control |
224ce89b | 1931 | uint64_t cache_size = 0; ///< total cache size |
31f18b77 FG |
1932 | float cache_meta_ratio = 0; ///< cache ratio dedicated to metadata |
1933 | float cache_kv_ratio = 0; ///< cache ratio dedicated to kv (e.g., rocksdb) | |
1934 | float cache_data_ratio = 0; ///< cache ratio dedicated to object data | |
7c673cae | 1935 | |
31f18b77 FG |
1936 | std::mutex vstatfs_lock; |
1937 | volatile_statfs vstatfs; | |
7c673cae FG |
1938 | |
1939 | struct MempoolThread : public Thread { | |
1940 | BlueStore *store; | |
1941 | Cond cond; | |
1942 | Mutex lock; | |
1943 | bool stop = false; | |
1944 | public: | |
1945 | explicit MempoolThread(BlueStore *s) | |
1946 | : store(s), | |
1947 | lock("BlueStore::MempoolThread::lock") {} | |
1948 | void *entry() override; | |
1949 | void init() { | |
1950 | assert(stop == false); | |
1951 | create("bstore_mempool"); | |
1952 | } | |
1953 | void shutdown() { | |
1954 | lock.Lock(); | |
1955 | stop = true; | |
1956 | cond.Signal(); | |
1957 | lock.Unlock(); | |
1958 | join(); | |
1959 | } | |
1960 | } mempool_thread; | |
1961 | ||
1962 | // -------------------------------------------------------- | |
1963 | // private methods | |
1964 | ||
1965 | void _init_logger(); | |
1966 | void _shutdown_logger(); | |
1967 | int _reload_logger(); | |
1968 | ||
1969 | int _open_path(); | |
1970 | void _close_path(); | |
1971 | int _open_fsid(bool create); | |
1972 | int _lock_fsid(); | |
1973 | int _read_fsid(uuid_d *f); | |
1974 | int _write_fsid(); | |
1975 | void _close_fsid(); | |
1976 | void _set_alloc_sizes(); | |
1977 | void _set_blob_size(); | |
1978 | ||
1979 | int _open_bdev(bool create); | |
1980 | void _close_bdev(); | |
1981 | int _open_db(bool create); | |
1982 | void _close_db(); | |
1983 | int _open_fm(bool create); | |
1984 | void _close_fm(); | |
1985 | int _open_alloc(); | |
1986 | void _close_alloc(); | |
1987 | int _open_collections(int *errors=0); | |
1988 | void _close_collections(); | |
1989 | ||
1990 | int _setup_block_symlink_or_file(string name, string path, uint64_t size, | |
1991 | bool create); | |
1992 | ||
7c673cae | 1993 | public: |
3efd9988 FG |
1994 | static int _write_bdev_label(CephContext* cct, |
1995 | string path, bluestore_bdev_label_t label); | |
7c673cae FG |
1996 | static int _read_bdev_label(CephContext* cct, string path, |
1997 | bluestore_bdev_label_t *label); | |
1998 | private: | |
1999 | int _check_or_set_bdev_label(string path, uint64_t size, string desc, | |
2000 | bool create); | |
2001 | ||
2002 | int _open_super_meta(); | |
2003 | ||
224ce89b | 2004 | void _open_statfs(); |
31f18b77 | 2005 | |
7c673cae FG |
2006 | int _reconcile_bluefs_freespace(); |
2007 | int _balance_bluefs_freespace(PExtentVector *extents); | |
2008 | void _commit_bluefs_freespace(const PExtentVector& extents); | |
2009 | ||
2010 | CollectionRef _get_collection(const coll_t& cid); | |
2011 | void _queue_reap_collection(CollectionRef& c); | |
2012 | void _reap_collections(); | |
2013 | void _update_cache_logger(); | |
2014 | ||
2015 | void _assign_nid(TransContext *txc, OnodeRef o); | |
2016 | uint64_t _assign_blobid(TransContext *txc); | |
2017 | ||
2018 | void _dump_onode(OnodeRef o, int log_level=30); | |
2019 | void _dump_extent_map(ExtentMap& em, int log_level=30); | |
2020 | void _dump_transaction(Transaction *t, int log_level = 30); | |
2021 | ||
2022 | TransContext *_txc_create(OpSequencer *osr); | |
2023 | void _txc_update_store_statfs(TransContext *txc); | |
2024 | void _txc_add_transaction(TransContext *txc, Transaction *t); | |
2025 | void _txc_calc_cost(TransContext *txc); | |
2026 | void _txc_write_nodes(TransContext *txc, KeyValueDB::Transaction t); | |
2027 | void _txc_state_proc(TransContext *txc); | |
2028 | void _txc_aio_submit(TransContext *txc); | |
2029 | public: | |
2030 | void txc_aio_finish(void *p) { | |
2031 | _txc_state_proc(static_cast<TransContext*>(p)); | |
2032 | } | |
2033 | private: | |
2034 | void _txc_finish_io(TransContext *txc); | |
2035 | void _txc_finalize_kv(TransContext *txc, KeyValueDB::Transaction t); | |
2036 | void _txc_applied_kv(TransContext *txc); | |
2037 | void _txc_committed_kv(TransContext *txc); | |
2038 | void _txc_finish(TransContext *txc); | |
2039 | void _txc_release_alloc(TransContext *txc); | |
2040 | ||
2041 | void _osr_drain_preceding(TransContext *txc); | |
2042 | void _osr_drain_all(); | |
2043 | void _osr_unregister_all(); | |
2044 | ||
31f18b77 FG |
2045 | void _kv_start(); |
2046 | void _kv_stop(); | |
7c673cae | 2047 | void _kv_sync_thread(); |
31f18b77 | 2048 | void _kv_finalize_thread(); |
7c673cae FG |
2049 | |
2050 | bluestore_deferred_op_t *_get_deferred_op(TransContext *txc, OnodeRef o); | |
2051 | void _deferred_queue(TransContext *txc); | |
3efd9988 | 2052 | public: |
224ce89b | 2053 | void deferred_try_submit(); |
3efd9988 | 2054 | private: |
224ce89b | 2055 | void _deferred_submit_unlock(OpSequencer *osr); |
7c673cae FG |
2056 | void _deferred_aio_finish(OpSequencer *osr); |
2057 | int _deferred_replay(); | |
2058 | ||
2059 | public: | |
2060 | using mempool_dynamic_bitset = | |
2061 | boost::dynamic_bitset<uint64_t, | |
2062 | mempool::bluestore_fsck::pool_allocator<uint64_t>>; | |
2063 | ||
2064 | private: | |
2065 | int _fsck_check_extents( | |
2066 | const ghobject_t& oid, | |
2067 | const PExtentVector& extents, | |
2068 | bool compressed, | |
2069 | mempool_dynamic_bitset &used_blocks, | |
b32b8144 | 2070 | uint64_t granularity, |
7c673cae FG |
2071 | store_statfs_t& expected_statfs); |
2072 | ||
2073 | void _buffer_cache_write( | |
2074 | TransContext *txc, | |
2075 | BlobRef b, | |
2076 | uint64_t offset, | |
2077 | bufferlist& bl, | |
2078 | unsigned flags) { | |
2079 | b->shared_blob->bc.write(b->shared_blob->get_cache(), txc->seq, offset, bl, | |
2080 | flags); | |
2081 | txc->shared_blobs_written.insert(b->shared_blob); | |
2082 | } | |
2083 | ||
2084 | int _collection_list( | |
2085 | Collection *c, const ghobject_t& start, const ghobject_t& end, | |
2086 | int max, vector<ghobject_t> *ls, ghobject_t *next); | |
2087 | ||
2088 | template <typename T, typename F> | |
2089 | T select_option(const std::string& opt_name, T val1, F f) { | |
2090 | //NB: opt_name reserved for future use | |
2091 | boost::optional<T> val2 = f(); | |
2092 | if (val2) { | |
2093 | return *val2; | |
2094 | } | |
2095 | return val1; | |
2096 | } | |
2097 | ||
2098 | void _apply_padding(uint64_t head_pad, | |
2099 | uint64_t tail_pad, | |
7c673cae FG |
2100 | bufferlist& padded); |
2101 | ||
2102 | // -- ondisk version --- | |
2103 | public: | |
2104 | const int32_t latest_ondisk_format = 2; ///< our version | |
2105 | const int32_t min_readable_ondisk_format = 1; ///< what we can read | |
2106 | const int32_t min_compat_ondisk_format = 2; ///< who can read us | |
2107 | ||
2108 | private: | |
2109 | int32_t ondisk_format = 0; ///< value detected on mount | |
2110 | ||
2111 | int _upgrade_super(); ///< upgrade (called during open_super) | |
2112 | void _prepare_ondisk_format_super(KeyValueDB::Transaction& t); | |
2113 | ||
2114 | // --- public interface --- | |
2115 | public: | |
2116 | BlueStore(CephContext *cct, const string& path); | |
2117 | BlueStore(CephContext *cct, const string& path, uint64_t min_alloc_size); // Ctor for UT only | |
2118 | ~BlueStore() override; | |
2119 | ||
2120 | string get_type() override { | |
2121 | return "bluestore"; | |
2122 | } | |
2123 | ||
2124 | bool needs_journal() override { return false; }; | |
2125 | bool wants_journal() override { return false; }; | |
2126 | bool allows_journal() override { return false; }; | |
2127 | ||
31f18b77 | 2128 | bool is_rotational() override; |
d2e6a577 | 2129 | bool is_journal_rotational() override; |
31f18b77 | 2130 | |
224ce89b WB |
2131 | string get_default_device_class() override { |
2132 | string device_class; | |
2133 | map<string, string> metadata; | |
2134 | collect_metadata(&metadata); | |
2135 | auto it = metadata.find("bluestore_bdev_type"); | |
2136 | if (it != metadata.end()) { | |
2137 | device_class = it->second; | |
2138 | } | |
2139 | return device_class; | |
2140 | } | |
2141 | ||
7c673cae FG |
2142 | static int get_block_device_fsid(CephContext* cct, const string& path, |
2143 | uuid_d *fsid); | |
2144 | ||
2145 | bool test_mount_in_use() override; | |
2146 | ||
2147 | private: | |
2148 | int _mount(bool kv_only); | |
2149 | public: | |
2150 | int mount() override { | |
2151 | return _mount(false); | |
2152 | } | |
2153 | int umount() override; | |
2154 | ||
2155 | int start_kv_only(KeyValueDB **pdb) { | |
2156 | int r = _mount(true); | |
2157 | if (r < 0) | |
2158 | return r; | |
2159 | *pdb = db; | |
2160 | return 0; | |
2161 | } | |
2162 | ||
3efd9988 FG |
2163 | int write_meta(const std::string& key, const std::string& value) override; |
2164 | int read_meta(const std::string& key, std::string *value) override; | |
2165 | ||
2166 | ||
2167 | int fsck(bool deep) override { | |
2168 | return _fsck(deep, false); | |
2169 | } | |
2170 | int repair(bool deep) override { | |
2171 | return _fsck(deep, true); | |
2172 | } | |
2173 | int _fsck(bool deep, bool repair); | |
7c673cae FG |
2174 | |
2175 | void set_cache_shards(unsigned num) override; | |
2176 | ||
2177 | int validate_hobject_key(const hobject_t &obj) const override { | |
2178 | return 0; | |
2179 | } | |
2180 | unsigned get_max_attr_name_length() override { | |
2181 | return 256; // arbitrary; there is no real limit internally | |
2182 | } | |
2183 | ||
2184 | int mkfs() override; | |
2185 | int mkjournal() override { | |
2186 | return 0; | |
2187 | } | |
2188 | ||
2189 | void get_db_statistics(Formatter *f) override; | |
2190 | void generate_db_histogram(Formatter *f) override; | |
31f18b77 | 2191 | void _flush_cache(); |
7c673cae FG |
2192 | void flush_cache() override; |
2193 | void dump_perf_counters(Formatter *f) override { | |
2194 | f->open_object_section("perf_counters"); | |
2195 | logger->dump_formatted(f, false); | |
2196 | f->close_section(); | |
2197 | } | |
2198 | ||
2199 | void register_osr(OpSequencer *osr) { | |
2200 | std::lock_guard<std::mutex> l(osr_lock); | |
2201 | osr_set.insert(osr); | |
2202 | } | |
2203 | void unregister_osr(OpSequencer *osr) { | |
2204 | std::lock_guard<std::mutex> l(osr_lock); | |
2205 | osr_set.erase(osr); | |
2206 | } | |
2207 | ||
2208 | public: | |
2209 | int statfs(struct store_statfs_t *buf) override; | |
2210 | ||
2211 | void collect_metadata(map<string,string> *pm) override; | |
2212 | ||
2213 | bool exists(const coll_t& cid, const ghobject_t& oid) override; | |
2214 | bool exists(CollectionHandle &c, const ghobject_t& oid) override; | |
2215 | int set_collection_opts( | |
2216 | const coll_t& cid, | |
2217 | const pool_opts_t& opts) override; | |
2218 | int stat( | |
2219 | const coll_t& cid, | |
2220 | const ghobject_t& oid, | |
2221 | struct stat *st, | |
2222 | bool allow_eio = false) override; | |
2223 | int stat( | |
2224 | CollectionHandle &c, | |
2225 | const ghobject_t& oid, | |
2226 | struct stat *st, | |
2227 | bool allow_eio = false) override; | |
2228 | int read( | |
2229 | const coll_t& cid, | |
2230 | const ghobject_t& oid, | |
2231 | uint64_t offset, | |
2232 | size_t len, | |
2233 | bufferlist& bl, | |
224ce89b | 2234 | uint32_t op_flags = 0) override; |
7c673cae FG |
2235 | int read( |
2236 | CollectionHandle &c, | |
2237 | const ghobject_t& oid, | |
2238 | uint64_t offset, | |
2239 | size_t len, | |
2240 | bufferlist& bl, | |
224ce89b | 2241 | uint32_t op_flags = 0) override; |
7c673cae FG |
2242 | int _do_read( |
2243 | Collection *c, | |
2244 | OnodeRef o, | |
2245 | uint64_t offset, | |
2246 | size_t len, | |
2247 | bufferlist& bl, | |
2248 | uint32_t op_flags = 0); | |
2249 | ||
2250 | private: | |
2251 | int _fiemap(CollectionHandle &c_, const ghobject_t& oid, | |
2252 | uint64_t offset, size_t len, interval_set<uint64_t>& destset); | |
2253 | public: | |
2254 | int fiemap(const coll_t& cid, const ghobject_t& oid, | |
2255 | uint64_t offset, size_t len, bufferlist& bl) override; | |
2256 | int fiemap(CollectionHandle &c, const ghobject_t& oid, | |
2257 | uint64_t offset, size_t len, bufferlist& bl) override; | |
2258 | int fiemap(const coll_t& cid, const ghobject_t& oid, | |
2259 | uint64_t offset, size_t len, map<uint64_t, uint64_t>& destmap) override; | |
2260 | int fiemap(CollectionHandle &c, const ghobject_t& oid, | |
2261 | uint64_t offset, size_t len, map<uint64_t, uint64_t>& destmap) override; | |
2262 | ||
2263 | ||
2264 | int getattr(const coll_t& cid, const ghobject_t& oid, const char *name, | |
2265 | bufferptr& value) override; | |
2266 | int getattr(CollectionHandle &c, const ghobject_t& oid, const char *name, | |
2267 | bufferptr& value) override; | |
2268 | ||
2269 | int getattrs(const coll_t& cid, const ghobject_t& oid, | |
2270 | map<string,bufferptr>& aset) override; | |
2271 | int getattrs(CollectionHandle &c, const ghobject_t& oid, | |
2272 | map<string,bufferptr>& aset) override; | |
2273 | ||
2274 | int list_collections(vector<coll_t>& ls) override; | |
2275 | ||
2276 | CollectionHandle open_collection(const coll_t &c) override; | |
2277 | ||
2278 | bool collection_exists(const coll_t& c) override; | |
2279 | int collection_empty(const coll_t& c, bool *empty) override; | |
2280 | int collection_bits(const coll_t& c) override; | |
2281 | ||
2282 | int collection_list(const coll_t& cid, | |
2283 | const ghobject_t& start, | |
2284 | const ghobject_t& end, | |
2285 | int max, | |
2286 | vector<ghobject_t> *ls, ghobject_t *next) override; | |
2287 | int collection_list(CollectionHandle &c, | |
2288 | const ghobject_t& start, | |
2289 | const ghobject_t& end, | |
2290 | int max, | |
2291 | vector<ghobject_t> *ls, ghobject_t *next) override; | |
2292 | ||
2293 | int omap_get( | |
2294 | const coll_t& cid, ///< [in] Collection containing oid | |
2295 | const ghobject_t &oid, ///< [in] Object containing omap | |
2296 | bufferlist *header, ///< [out] omap header | |
2297 | map<string, bufferlist> *out /// < [out] Key to value map | |
2298 | ) override; | |
2299 | int omap_get( | |
2300 | CollectionHandle &c, ///< [in] Collection containing oid | |
2301 | const ghobject_t &oid, ///< [in] Object containing omap | |
2302 | bufferlist *header, ///< [out] omap header | |
2303 | map<string, bufferlist> *out /// < [out] Key to value map | |
2304 | ) override; | |
2305 | ||
2306 | /// Get omap header | |
2307 | int omap_get_header( | |
2308 | const coll_t& cid, ///< [in] Collection containing oid | |
2309 | const ghobject_t &oid, ///< [in] Object containing omap | |
2310 | bufferlist *header, ///< [out] omap header | |
2311 | bool allow_eio = false ///< [in] don't assert on eio | |
2312 | ) override; | |
2313 | int omap_get_header( | |
2314 | CollectionHandle &c, ///< [in] Collection containing oid | |
2315 | const ghobject_t &oid, ///< [in] Object containing omap | |
2316 | bufferlist *header, ///< [out] omap header | |
2317 | bool allow_eio = false ///< [in] don't assert on eio | |
2318 | ) override; | |
2319 | ||
2320 | /// Get keys defined on oid | |
2321 | int omap_get_keys( | |
2322 | const coll_t& cid, ///< [in] Collection containing oid | |
2323 | const ghobject_t &oid, ///< [in] Object containing omap | |
2324 | set<string> *keys ///< [out] Keys defined on oid | |
2325 | ) override; | |
2326 | int omap_get_keys( | |
2327 | CollectionHandle &c, ///< [in] Collection containing oid | |
2328 | const ghobject_t &oid, ///< [in] Object containing omap | |
2329 | set<string> *keys ///< [out] Keys defined on oid | |
2330 | ) override; | |
2331 | ||
2332 | /// Get key values | |
2333 | int omap_get_values( | |
2334 | const coll_t& cid, ///< [in] Collection containing oid | |
2335 | const ghobject_t &oid, ///< [in] Object containing omap | |
2336 | const set<string> &keys, ///< [in] Keys to get | |
2337 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
2338 | ) override; | |
2339 | int omap_get_values( | |
2340 | CollectionHandle &c, ///< [in] Collection containing oid | |
2341 | const ghobject_t &oid, ///< [in] Object containing omap | |
2342 | const set<string> &keys, ///< [in] Keys to get | |
2343 | map<string, bufferlist> *out ///< [out] Returned keys and values | |
2344 | ) override; | |
2345 | ||
2346 | /// Filters keys into out which are defined on oid | |
2347 | int omap_check_keys( | |
2348 | const coll_t& cid, ///< [in] Collection containing oid | |
2349 | const ghobject_t &oid, ///< [in] Object containing omap | |
2350 | const set<string> &keys, ///< [in] Keys to check | |
2351 | set<string> *out ///< [out] Subset of keys defined on oid | |
2352 | ) override; | |
2353 | int omap_check_keys( | |
2354 | CollectionHandle &c, ///< [in] Collection containing oid | |
2355 | const ghobject_t &oid, ///< [in] Object containing omap | |
2356 | const set<string> &keys, ///< [in] Keys to check | |
2357 | set<string> *out ///< [out] Subset of keys defined on oid | |
2358 | ) override; | |
2359 | ||
2360 | ObjectMap::ObjectMapIterator get_omap_iterator( | |
2361 | const coll_t& cid, ///< [in] collection | |
2362 | const ghobject_t &oid ///< [in] object | |
2363 | ) override; | |
2364 | ObjectMap::ObjectMapIterator get_omap_iterator( | |
2365 | CollectionHandle &c, ///< [in] collection | |
2366 | const ghobject_t &oid ///< [in] object | |
2367 | ) override; | |
2368 | ||
2369 | void set_fsid(uuid_d u) override { | |
2370 | fsid = u; | |
2371 | } | |
2372 | uuid_d get_fsid() override { | |
2373 | return fsid; | |
2374 | } | |
2375 | ||
2376 | uint64_t estimate_objects_overhead(uint64_t num_objects) override { | |
2377 | return num_objects * 300; //assuming per-object overhead is 300 bytes | |
2378 | } | |
2379 | ||
2380 | struct BSPerfTracker { | |
2381 | PerfCounters::avg_tracker<uint64_t> os_commit_latency; | |
2382 | PerfCounters::avg_tracker<uint64_t> os_apply_latency; | |
2383 | ||
2384 | objectstore_perf_stat_t get_cur_stats() const { | |
2385 | objectstore_perf_stat_t ret; | |
c07f9fc5 FG |
2386 | ret.os_commit_latency = os_commit_latency.current_avg(); |
2387 | ret.os_apply_latency = os_apply_latency.current_avg(); | |
7c673cae FG |
2388 | return ret; |
2389 | } | |
2390 | ||
2391 | void update_from_perfcounters(PerfCounters &logger); | |
2392 | } perf_tracker; | |
2393 | ||
2394 | objectstore_perf_stat_t get_cur_stats() override { | |
2395 | perf_tracker.update_from_perfcounters(*logger); | |
2396 | return perf_tracker.get_cur_stats(); | |
2397 | } | |
2398 | const PerfCounters* get_perf_counters() const override { | |
2399 | return logger; | |
2400 | } | |
2401 | ||
2402 | int queue_transactions( | |
2403 | Sequencer *osr, | |
2404 | vector<Transaction>& tls, | |
2405 | TrackedOpRef op = TrackedOpRef(), | |
2406 | ThreadPool::TPHandle *handle = NULL) override; | |
2407 | ||
2408 | // error injection | |
2409 | void inject_data_error(const ghobject_t& o) override { | |
2410 | RWLock::WLocker l(debug_read_error_lock); | |
2411 | debug_data_error_objects.insert(o); | |
2412 | } | |
2413 | void inject_mdata_error(const ghobject_t& o) override { | |
2414 | RWLock::WLocker l(debug_read_error_lock); | |
2415 | debug_mdata_error_objects.insert(o); | |
2416 | } | |
224ce89b WB |
2417 | void compact() override { |
2418 | assert(db); | |
2419 | db->compact(); | |
2420 | } | |
2421 | ||
7c673cae FG |
2422 | private: |
2423 | bool _debug_data_eio(const ghobject_t& o) { | |
2424 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
2425 | return false; | |
2426 | } | |
2427 | RWLock::RLocker l(debug_read_error_lock); | |
2428 | return debug_data_error_objects.count(o); | |
2429 | } | |
2430 | bool _debug_mdata_eio(const ghobject_t& o) { | |
2431 | if (!cct->_conf->bluestore_debug_inject_read_err) { | |
2432 | return false; | |
2433 | } | |
2434 | RWLock::RLocker l(debug_read_error_lock); | |
2435 | return debug_mdata_error_objects.count(o); | |
2436 | } | |
2437 | void _debug_obj_on_delete(const ghobject_t& o) { | |
2438 | if (cct->_conf->bluestore_debug_inject_read_err) { | |
2439 | RWLock::WLocker l(debug_read_error_lock); | |
2440 | debug_data_error_objects.erase(o); | |
2441 | debug_mdata_error_objects.erase(o); | |
2442 | } | |
2443 | } | |
2444 | ||
2445 | private: | |
2446 | ||
2447 | // -------------------------------------------------------- | |
2448 | // read processing internal methods | |
2449 | int _verify_csum( | |
2450 | OnodeRef& o, | |
2451 | const bluestore_blob_t* blob, | |
2452 | uint64_t blob_xoffset, | |
2453 | const bufferlist& bl, | |
2454 | uint64_t logical_offset) const; | |
2455 | int _decompress(bufferlist& source, bufferlist* result); | |
2456 | ||
2457 | ||
2458 | // -------------------------------------------------------- | |
2459 | // write ops | |
2460 | ||
2461 | struct WriteContext { | |
2462 | bool buffered = false; ///< buffered write | |
2463 | bool compress = false; ///< compressed write | |
2464 | uint64_t target_blob_size = 0; ///< target (max) blob size | |
2465 | unsigned csum_order = 0; ///< target checksum chunk order | |
2466 | ||
2467 | old_extent_map_t old_extents; ///< must deref these blobs | |
2468 | ||
2469 | struct write_item { | |
2470 | uint64_t logical_offset; ///< write logical offset | |
2471 | BlobRef b; | |
2472 | uint64_t blob_length; | |
2473 | uint64_t b_off; | |
2474 | bufferlist bl; | |
2475 | uint64_t b_off0; ///< original offset in a blob prior to padding | |
2476 | uint64_t length0; ///< original data length prior to padding | |
2477 | ||
2478 | bool mark_unused; | |
2479 | bool new_blob; ///< whether new blob was created | |
2480 | ||
3efd9988 FG |
2481 | bool compressed = false; |
2482 | bufferlist compressed_bl; | |
2483 | size_t compressed_len = 0; | |
2484 | ||
7c673cae FG |
2485 | write_item( |
2486 | uint64_t logical_offs, | |
2487 | BlobRef b, | |
2488 | uint64_t blob_len, | |
2489 | uint64_t o, | |
2490 | bufferlist& bl, | |
2491 | uint64_t o0, | |
2492 | uint64_t l0, | |
2493 | bool _mark_unused, | |
2494 | bool _new_blob) | |
2495 | : | |
2496 | logical_offset(logical_offs), | |
2497 | b(b), | |
2498 | blob_length(blob_len), | |
2499 | b_off(o), | |
2500 | bl(bl), | |
2501 | b_off0(o0), | |
2502 | length0(l0), | |
2503 | mark_unused(_mark_unused), | |
2504 | new_blob(_new_blob) {} | |
2505 | }; | |
2506 | vector<write_item> writes; ///< blobs we're writing | |
2507 | ||
2508 | /// partial clone of the context | |
2509 | void fork(const WriteContext& other) { | |
2510 | buffered = other.buffered; | |
2511 | compress = other.compress; | |
2512 | target_blob_size = other.target_blob_size; | |
2513 | csum_order = other.csum_order; | |
2514 | } | |
2515 | void write( | |
2516 | uint64_t loffs, | |
2517 | BlobRef b, | |
2518 | uint64_t blob_len, | |
2519 | uint64_t o, | |
2520 | bufferlist& bl, | |
2521 | uint64_t o0, | |
2522 | uint64_t len0, | |
2523 | bool _mark_unused, | |
2524 | bool _new_blob) { | |
2525 | writes.emplace_back(loffs, | |
2526 | b, | |
2527 | blob_len, | |
2528 | o, | |
2529 | bl, | |
2530 | o0, | |
2531 | len0, | |
2532 | _mark_unused, | |
2533 | _new_blob); | |
2534 | } | |
2535 | /// Checks for writes to the same pextent within a blob | |
2536 | bool has_conflict( | |
2537 | BlobRef b, | |
2538 | uint64_t loffs, | |
2539 | uint64_t loffs_end, | |
2540 | uint64_t min_alloc_size); | |
2541 | }; | |
2542 | ||
2543 | void _do_write_small( | |
2544 | TransContext *txc, | |
2545 | CollectionRef &c, | |
2546 | OnodeRef o, | |
2547 | uint64_t offset, uint64_t length, | |
2548 | bufferlist::iterator& blp, | |
2549 | WriteContext *wctx); | |
2550 | void _do_write_big( | |
2551 | TransContext *txc, | |
2552 | CollectionRef &c, | |
2553 | OnodeRef o, | |
2554 | uint64_t offset, uint64_t length, | |
2555 | bufferlist::iterator& blp, | |
2556 | WriteContext *wctx); | |
2557 | int _do_alloc_write( | |
2558 | TransContext *txc, | |
2559 | CollectionRef c, | |
2560 | OnodeRef o, | |
2561 | WriteContext *wctx); | |
2562 | void _wctx_finish( | |
2563 | TransContext *txc, | |
2564 | CollectionRef& c, | |
2565 | OnodeRef o, | |
31f18b77 FG |
2566 | WriteContext *wctx, |
2567 | set<SharedBlob*> *maybe_unshared_blobs=0); | |
7c673cae FG |
2568 | |
2569 | int _do_transaction(Transaction *t, | |
2570 | TransContext *txc, | |
2571 | ThreadPool::TPHandle *handle); | |
2572 | ||
2573 | int _write(TransContext *txc, | |
2574 | CollectionRef& c, | |
2575 | OnodeRef& o, | |
2576 | uint64_t offset, size_t len, | |
2577 | bufferlist& bl, | |
2578 | uint32_t fadvise_flags); | |
2579 | void _pad_zeros(bufferlist *bl, uint64_t *offset, | |
2580 | uint64_t chunk_size); | |
2581 | ||
31f18b77 FG |
2582 | void _choose_write_options(CollectionRef& c, |
2583 | OnodeRef o, | |
2584 | uint32_t fadvise_flags, | |
2585 | WriteContext *wctx); | |
2586 | ||
2587 | int _do_gc(TransContext *txc, | |
2588 | CollectionRef& c, | |
2589 | OnodeRef o, | |
2590 | const GarbageCollector& gc, | |
2591 | const WriteContext& wctx, | |
2592 | uint64_t *dirty_start, | |
2593 | uint64_t *dirty_end); | |
2594 | ||
7c673cae FG |
2595 | int _do_write(TransContext *txc, |
2596 | CollectionRef &c, | |
2597 | OnodeRef o, | |
2598 | uint64_t offset, uint64_t length, | |
2599 | bufferlist& bl, | |
2600 | uint32_t fadvise_flags); | |
2601 | void _do_write_data(TransContext *txc, | |
2602 | CollectionRef& c, | |
2603 | OnodeRef o, | |
2604 | uint64_t offset, | |
2605 | uint64_t length, | |
2606 | bufferlist& bl, | |
2607 | WriteContext *wctx); | |
2608 | ||
2609 | int _touch(TransContext *txc, | |
2610 | CollectionRef& c, | |
2611 | OnodeRef& o); | |
2612 | int _do_zero(TransContext *txc, | |
2613 | CollectionRef& c, | |
2614 | OnodeRef& o, | |
2615 | uint64_t offset, size_t len); | |
2616 | int _zero(TransContext *txc, | |
2617 | CollectionRef& c, | |
2618 | OnodeRef& o, | |
2619 | uint64_t offset, size_t len); | |
2620 | void _do_truncate(TransContext *txc, | |
2621 | CollectionRef& c, | |
2622 | OnodeRef o, | |
31f18b77 FG |
2623 | uint64_t offset, |
2624 | set<SharedBlob*> *maybe_unshared_blobs=0); | |
35e4c445 | 2625 | int _truncate(TransContext *txc, |
7c673cae FG |
2626 | CollectionRef& c, |
2627 | OnodeRef& o, | |
2628 | uint64_t offset); | |
2629 | int _remove(TransContext *txc, | |
2630 | CollectionRef& c, | |
2631 | OnodeRef& o); | |
2632 | int _do_remove(TransContext *txc, | |
2633 | CollectionRef& c, | |
2634 | OnodeRef o); | |
2635 | int _setattr(TransContext *txc, | |
2636 | CollectionRef& c, | |
2637 | OnodeRef& o, | |
2638 | const string& name, | |
2639 | bufferptr& val); | |
2640 | int _setattrs(TransContext *txc, | |
2641 | CollectionRef& c, | |
2642 | OnodeRef& o, | |
2643 | const map<string,bufferptr>& aset); | |
2644 | int _rmattr(TransContext *txc, | |
2645 | CollectionRef& c, | |
2646 | OnodeRef& o, | |
2647 | const string& name); | |
2648 | int _rmattrs(TransContext *txc, | |
2649 | CollectionRef& c, | |
2650 | OnodeRef& o); | |
2651 | void _do_omap_clear(TransContext *txc, uint64_t id); | |
2652 | int _omap_clear(TransContext *txc, | |
2653 | CollectionRef& c, | |
2654 | OnodeRef& o); | |
2655 | int _omap_setkeys(TransContext *txc, | |
2656 | CollectionRef& c, | |
2657 | OnodeRef& o, | |
2658 | bufferlist& bl); | |
2659 | int _omap_setheader(TransContext *txc, | |
2660 | CollectionRef& c, | |
2661 | OnodeRef& o, | |
2662 | bufferlist& header); | |
2663 | int _omap_rmkeys(TransContext *txc, | |
2664 | CollectionRef& c, | |
2665 | OnodeRef& o, | |
2666 | bufferlist& bl); | |
2667 | int _omap_rmkey_range(TransContext *txc, | |
2668 | CollectionRef& c, | |
2669 | OnodeRef& o, | |
2670 | const string& first, const string& last); | |
2671 | int _set_alloc_hint( | |
2672 | TransContext *txc, | |
2673 | CollectionRef& c, | |
2674 | OnodeRef& o, | |
2675 | uint64_t expected_object_size, | |
2676 | uint64_t expected_write_size, | |
2677 | uint32_t flags); | |
2678 | int _do_clone_range(TransContext *txc, | |
2679 | CollectionRef& c, | |
2680 | OnodeRef& oldo, | |
2681 | OnodeRef& newo, | |
2682 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
2683 | int _clone(TransContext *txc, | |
2684 | CollectionRef& c, | |
2685 | OnodeRef& oldo, | |
2686 | OnodeRef& newo); | |
2687 | int _clone_range(TransContext *txc, | |
2688 | CollectionRef& c, | |
2689 | OnodeRef& oldo, | |
2690 | OnodeRef& newo, | |
2691 | uint64_t srcoff, uint64_t length, uint64_t dstoff); | |
2692 | int _rename(TransContext *txc, | |
2693 | CollectionRef& c, | |
2694 | OnodeRef& oldo, | |
2695 | OnodeRef& newo, | |
2696 | const ghobject_t& new_oid); | |
2697 | int _create_collection(TransContext *txc, const coll_t &cid, | |
2698 | unsigned bits, CollectionRef *c); | |
2699 | int _remove_collection(TransContext *txc, const coll_t &cid, | |
2700 | CollectionRef *c); | |
2701 | int _split_collection(TransContext *txc, | |
2702 | CollectionRef& c, | |
2703 | CollectionRef& d, | |
2704 | unsigned bits, int rem); | |
2705 | }; | |
2706 | ||
2707 | inline ostream& operator<<(ostream& out, const BlueStore::OpSequencer& s) { | |
2708 | return out << *s.parent; | |
2709 | } | |
2710 | ||
2711 | static inline void intrusive_ptr_add_ref(BlueStore::Onode *o) { | |
2712 | o->get(); | |
2713 | } | |
2714 | static inline void intrusive_ptr_release(BlueStore::Onode *o) { | |
2715 | o->put(); | |
2716 | } | |
2717 | ||
2718 | static inline void intrusive_ptr_add_ref(BlueStore::OpSequencer *o) { | |
2719 | o->get(); | |
2720 | } | |
2721 | static inline void intrusive_ptr_release(BlueStore::OpSequencer *o) { | |
2722 | o->put(); | |
2723 | } | |
2724 | ||
2725 | #endif |