]>
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 | #include "boost/algorithm/string.hpp" | |
9f95a23c | 5 | #include "bluestore_common.h" |
7c673cae FG |
6 | #include "BlueFS.h" |
7 | ||
8 | #include "common/debug.h" | |
9 | #include "common/errno.h" | |
10 | #include "common/perf_counters.h" | |
11 | #include "BlockDevice.h" | |
12 | #include "Allocator.h" | |
11fdf7f2 | 13 | #include "include/ceph_assert.h" |
eafe8130 | 14 | #include "common/admin_socket.h" |
7c673cae FG |
15 | |
16 | #define dout_context cct | |
17 | #define dout_subsys ceph_subsys_bluefs | |
18 | #undef dout_prefix | |
19 | #define dout_prefix *_dout << "bluefs " | |
9f95a23c | 20 | using TOPNSPC::common::cmd_getval; |
7c673cae FG |
21 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueFS::File, bluefs_file, bluefs); |
22 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueFS::Dir, bluefs_dir, bluefs); | |
f91f0fd5 | 23 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueFS::FileWriter, bluefs_file_writer, bluefs_file_writer); |
7c673cae | 24 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueFS::FileReaderBuffer, |
f91f0fd5 TL |
25 | bluefs_file_reader_buffer, bluefs_file_reader); |
26 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueFS::FileReader, bluefs_file_reader, bluefs_file_reader); | |
7c673cae FG |
27 | MEMPOOL_DEFINE_OBJECT_FACTORY(BlueFS::FileLock, bluefs_file_lock, bluefs); |
28 | ||
11fdf7f2 TL |
29 | static void wal_discard_cb(void *priv, void* priv2) { |
30 | BlueFS *bluefs = static_cast<BlueFS*>(priv); | |
31 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
32 | bluefs->handle_discard(BlueFS::BDEV_WAL, *tmp); | |
33 | } | |
34 | ||
35 | static void db_discard_cb(void *priv, void* priv2) { | |
36 | BlueFS *bluefs = static_cast<BlueFS*>(priv); | |
37 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
38 | bluefs->handle_discard(BlueFS::BDEV_DB, *tmp); | |
39 | } | |
40 | ||
41 | static void slow_discard_cb(void *priv, void* priv2) { | |
42 | BlueFS *bluefs = static_cast<BlueFS*>(priv); | |
43 | interval_set<uint64_t> *tmp = static_cast<interval_set<uint64_t>*>(priv2); | |
44 | bluefs->handle_discard(BlueFS::BDEV_SLOW, *tmp); | |
45 | } | |
7c673cae | 46 | |
eafe8130 TL |
47 | class BlueFS::SocketHook : public AdminSocketHook { |
48 | BlueFS* bluefs; | |
49 | public: | |
50 | static BlueFS::SocketHook* create(BlueFS* bluefs) | |
51 | { | |
52 | BlueFS::SocketHook* hook = nullptr; | |
53 | AdminSocket* admin_socket = bluefs->cct->get_admin_socket(); | |
54 | if (admin_socket) { | |
55 | hook = new BlueFS::SocketHook(bluefs); | |
9f95a23c | 56 | int r = admin_socket->register_command("bluestore bluefs available " |
eafe8130 TL |
57 | "name=alloc_size,type=CephInt,req=false", |
58 | hook, | |
59 | "Report available space for bluefs. " | |
60 | "If alloc_size set, make simulation."); | |
61 | if (r != 0) { | |
62 | ldout(bluefs->cct, 1) << __func__ << " cannot register SocketHook" << dendl; | |
63 | delete hook; | |
64 | hook = nullptr; | |
9f95a23c | 65 | } else { |
f6b5b4d7 | 66 | r = admin_socket->register_command("bluefs stats", |
9f95a23c TL |
67 | hook, |
68 | "Dump internal statistics for bluefs." | |
69 | ""); | |
70 | ceph_assert(r == 0); | |
eafe8130 TL |
71 | } |
72 | } | |
73 | return hook; | |
74 | } | |
75 | ||
76 | ~SocketHook() { | |
77 | AdminSocket* admin_socket = bluefs->cct->get_admin_socket(); | |
9f95a23c | 78 | admin_socket->unregister_commands(this); |
eafe8130 TL |
79 | } |
80 | private: | |
81 | SocketHook(BlueFS* bluefs) : | |
82 | bluefs(bluefs) {} | |
9f95a23c TL |
83 | int call(std::string_view command, const cmdmap_t& cmdmap, |
84 | Formatter *f, | |
85 | std::ostream& errss, | |
86 | bufferlist& out) override { | |
87 | if (command == "bluestore bluefs available") { | |
88 | int64_t alloc_size = 0; | |
89 | cmd_getval(cmdmap, "alloc_size", alloc_size); | |
90 | if ((alloc_size & (alloc_size - 1)) != 0) { | |
91 | errss << "Invalid allocation size:'" << alloc_size << std::endl; | |
92 | return -EINVAL; | |
93 | } | |
94 | if (alloc_size == 0) | |
95 | alloc_size = bluefs->cct->_conf->bluefs_alloc_size; | |
96 | f->open_object_section("bluefs_available_space"); | |
97 | for (unsigned dev = BDEV_WAL; dev <= BDEV_SLOW; dev++) { | |
98 | if (bluefs->bdev[dev]) { | |
99 | f->open_object_section("dev"); | |
100 | f->dump_string("device", bluefs->get_device_name(dev)); | |
101 | ceph_assert(bluefs->alloc[dev]); | |
102 | f->dump_int("free", bluefs->alloc[dev]->get_free()); | |
103 | f->close_section(); | |
104 | } | |
105 | } | |
106 | size_t extra_space = 0; | |
107 | if (bluefs->slow_dev_expander) { | |
108 | extra_space = bluefs->slow_dev_expander->available_freespace(alloc_size); | |
eafe8130 | 109 | } |
9f95a23c TL |
110 | f->dump_int("available_from_bluestore", extra_space); |
111 | f->close_section(); | |
112 | } else if (command == "bluefs stats") { | |
113 | std::stringstream ss; | |
114 | bluefs->dump_block_extents(ss); | |
115 | bluefs->dump_volume_selector(ss); | |
eafe8130 | 116 | out.append(ss); |
9f95a23c TL |
117 | } else { |
118 | errss << "Invalid command" << std::endl; | |
119 | return -ENOSYS; | |
eafe8130 | 120 | } |
9f95a23c TL |
121 | return 0; |
122 | } | |
eafe8130 TL |
123 | }; |
124 | ||
7c673cae FG |
125 | BlueFS::BlueFS(CephContext* cct) |
126 | : cct(cct), | |
127 | bdev(MAX_BDEV), | |
128 | ioc(MAX_BDEV), | |
11fdf7f2 | 129 | block_all(MAX_BDEV) |
7c673cae | 130 | { |
11fdf7f2 TL |
131 | discard_cb[BDEV_WAL] = wal_discard_cb; |
132 | discard_cb[BDEV_DB] = db_discard_cb; | |
133 | discard_cb[BDEV_SLOW] = slow_discard_cb; | |
eafe8130 | 134 | asok_hook = SocketHook::create(this); |
7c673cae FG |
135 | } |
136 | ||
137 | BlueFS::~BlueFS() | |
138 | { | |
eafe8130 | 139 | delete asok_hook; |
7c673cae FG |
140 | for (auto p : ioc) { |
141 | if (p) | |
142 | p->aio_wait(); | |
143 | } | |
144 | for (auto p : bdev) { | |
145 | if (p) { | |
146 | p->close(); | |
147 | delete p; | |
148 | } | |
149 | } | |
150 | for (auto p : ioc) { | |
151 | delete p; | |
152 | } | |
153 | } | |
154 | ||
155 | void BlueFS::_init_logger() | |
156 | { | |
157 | PerfCountersBuilder b(cct, "bluefs", | |
158 | l_bluefs_first, l_bluefs_last); | |
159 | b.add_u64_counter(l_bluefs_gift_bytes, "gift_bytes", | |
11fdf7f2 | 160 | "Bytes gifted from BlueStore", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae | 161 | b.add_u64_counter(l_bluefs_reclaim_bytes, "reclaim_bytes", |
11fdf7f2 | 162 | "Bytes reclaimed by BlueStore", NULL, 0, unit_t(UNIT_BYTES)); |
7c673cae FG |
163 | b.add_u64(l_bluefs_db_total_bytes, "db_total_bytes", |
164 | "Total bytes (main db device)", | |
11fdf7f2 | 165 | "b", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
166 | b.add_u64(l_bluefs_db_used_bytes, "db_used_bytes", |
167 | "Used bytes (main db device)", | |
11fdf7f2 | 168 | "u", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
169 | b.add_u64(l_bluefs_wal_total_bytes, "wal_total_bytes", |
170 | "Total bytes (wal device)", | |
11fdf7f2 | 171 | "walb", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
172 | b.add_u64(l_bluefs_wal_used_bytes, "wal_used_bytes", |
173 | "Used bytes (wal device)", | |
11fdf7f2 | 174 | "walu", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
175 | b.add_u64(l_bluefs_slow_total_bytes, "slow_total_bytes", |
176 | "Total bytes (slow device)", | |
11fdf7f2 | 177 | "slob", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
178 | b.add_u64(l_bluefs_slow_used_bytes, "slow_used_bytes", |
179 | "Used bytes (slow device)", | |
11fdf7f2 | 180 | "slou", PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); |
7c673cae FG |
181 | b.add_u64(l_bluefs_num_files, "num_files", "File count", |
182 | "f", PerfCountersBuilder::PRIO_USEFUL); | |
183 | b.add_u64(l_bluefs_log_bytes, "log_bytes", "Size of the metadata log", | |
11fdf7f2 | 184 | "jlen", PerfCountersBuilder::PRIO_INTERESTING, unit_t(UNIT_BYTES)); |
7c673cae FG |
185 | b.add_u64_counter(l_bluefs_log_compactions, "log_compactions", |
186 | "Compactions of the metadata log"); | |
187 | b.add_u64_counter(l_bluefs_logged_bytes, "logged_bytes", | |
188 | "Bytes written to the metadata log", "j", | |
11fdf7f2 | 189 | PerfCountersBuilder::PRIO_CRITICAL, unit_t(UNIT_BYTES)); |
7c673cae FG |
190 | b.add_u64_counter(l_bluefs_files_written_wal, "files_written_wal", |
191 | "Files written to WAL"); | |
192 | b.add_u64_counter(l_bluefs_files_written_sst, "files_written_sst", | |
193 | "Files written to SSTs"); | |
194 | b.add_u64_counter(l_bluefs_bytes_written_wal, "bytes_written_wal", | |
195 | "Bytes written to WAL", "wal", | |
196 | PerfCountersBuilder::PRIO_CRITICAL); | |
197 | b.add_u64_counter(l_bluefs_bytes_written_sst, "bytes_written_sst", | |
198 | "Bytes written to SSTs", "sst", | |
11fdf7f2 TL |
199 | PerfCountersBuilder::PRIO_CRITICAL, unit_t(UNIT_BYTES)); |
200 | b.add_u64_counter(l_bluefs_bytes_written_slow, "bytes_written_slow", | |
201 | "Bytes written to WAL/SSTs at slow device", NULL, | |
202 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
203 | b.add_u64_counter(l_bluefs_max_bytes_wal, "max_bytes_wal", | |
204 | "Maximum bytes allocated from WAL"); | |
205 | b.add_u64_counter(l_bluefs_max_bytes_db, "max_bytes_db", | |
206 | "Maximum bytes allocated from DB"); | |
207 | b.add_u64_counter(l_bluefs_max_bytes_slow, "max_bytes_slow", | |
208 | "Maximum bytes allocated from SLOW"); | |
494da23a TL |
209 | |
210 | b.add_u64_counter(l_bluefs_read_random_count, "read_random_count", | |
211 | "random read requests processed"); | |
212 | b.add_u64_counter(l_bluefs_read_random_bytes, "read_random_bytes", | |
213 | "Bytes requested in random read mode", NULL, | |
214 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
215 | b.add_u64_counter(l_bluefs_read_random_disk_count, "read_random_disk_count", | |
216 | "random reads requests going to disk"); | |
217 | b.add_u64_counter(l_bluefs_read_random_disk_bytes, "read_random_disk_bytes", | |
218 | "Bytes read from disk in random read mode", NULL, | |
219 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
220 | b.add_u64_counter(l_bluefs_read_random_buffer_count, "read_random_buffer_count", | |
221 | "random read requests processed using prefetch buffer"); | |
222 | b.add_u64_counter(l_bluefs_read_random_buffer_bytes, "read_random_buffer_bytes", | |
223 | "Bytes read from prefetch buffer in random read mode", NULL, | |
224 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
225 | ||
226 | b.add_u64_counter(l_bluefs_read_count, "read_count", | |
227 | "buffered read requests processed"); | |
228 | b.add_u64_counter(l_bluefs_read_bytes, "read_bytes", | |
229 | "Bytes requested in buffered read mode", NULL, | |
230 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
231 | ||
232 | b.add_u64_counter(l_bluefs_read_prefetch_count, "read_prefetch_count", | |
233 | "prefetch read requests processed"); | |
234 | b.add_u64_counter(l_bluefs_read_prefetch_bytes, "read_prefetch_bytes", | |
235 | "Bytes requested in prefetch read mode", NULL, | |
236 | PerfCountersBuilder::PRIO_USEFUL, unit_t(UNIT_BYTES)); | |
237 | ||
7c673cae FG |
238 | logger = b.create_perf_counters(); |
239 | cct->get_perfcounters_collection()->add(logger); | |
240 | } | |
241 | ||
242 | void BlueFS::_shutdown_logger() | |
243 | { | |
244 | cct->get_perfcounters_collection()->remove(logger); | |
245 | delete logger; | |
246 | } | |
247 | ||
248 | void BlueFS::_update_logger_stats() | |
249 | { | |
250 | // we must be holding the lock | |
251 | logger->set(l_bluefs_num_files, file_map.size()); | |
252 | logger->set(l_bluefs_log_bytes, log_writer->file->fnode.size); | |
253 | ||
254 | if (alloc[BDEV_WAL]) { | |
11fdf7f2 | 255 | logger->set(l_bluefs_wal_total_bytes, block_all[BDEV_WAL].size()); |
7c673cae | 256 | logger->set(l_bluefs_wal_used_bytes, |
11fdf7f2 | 257 | block_all[BDEV_WAL].size() - alloc[BDEV_WAL]->get_free()); |
7c673cae FG |
258 | } |
259 | if (alloc[BDEV_DB]) { | |
11fdf7f2 | 260 | logger->set(l_bluefs_db_total_bytes, block_all[BDEV_DB].size()); |
7c673cae | 261 | logger->set(l_bluefs_db_used_bytes, |
11fdf7f2 | 262 | block_all[BDEV_DB].size() - alloc[BDEV_DB]->get_free()); |
7c673cae FG |
263 | } |
264 | if (alloc[BDEV_SLOW]) { | |
11fdf7f2 | 265 | logger->set(l_bluefs_slow_total_bytes, block_all[BDEV_SLOW].size()); |
7c673cae | 266 | logger->set(l_bluefs_slow_used_bytes, |
11fdf7f2 | 267 | block_all[BDEV_SLOW].size() - alloc[BDEV_SLOW]->get_free()); |
7c673cae FG |
268 | } |
269 | } | |
270 | ||
11fdf7f2 TL |
271 | int BlueFS::add_block_device(unsigned id, const string& path, bool trim, |
272 | bool shared_with_bluestore) | |
7c673cae FG |
273 | { |
274 | dout(10) << __func__ << " bdev " << id << " path " << path << dendl; | |
11fdf7f2 TL |
275 | ceph_assert(id < bdev.size()); |
276 | ceph_assert(bdev[id] == NULL); | |
277 | BlockDevice *b = BlockDevice::create(cct, path, NULL, NULL, | |
278 | discard_cb[id], static_cast<void*>(this)); | |
279 | if (shared_with_bluestore) { | |
280 | b->set_no_exclusive_lock(); | |
281 | } | |
7c673cae FG |
282 | int r = b->open(path); |
283 | if (r < 0) { | |
284 | delete b; | |
285 | return r; | |
286 | } | |
11fdf7f2 TL |
287 | if (trim) { |
288 | b->discard(0, b->get_size()); | |
289 | } | |
290 | ||
7c673cae | 291 | dout(1) << __func__ << " bdev " << id << " path " << path |
1adf2230 | 292 | << " size " << byte_u_t(b->get_size()) << dendl; |
7c673cae FG |
293 | bdev[id] = b; |
294 | ioc[id] = new IOContext(cct, NULL); | |
295 | return 0; | |
296 | } | |
297 | ||
298 | bool BlueFS::bdev_support_label(unsigned id) | |
299 | { | |
11fdf7f2 TL |
300 | ceph_assert(id < bdev.size()); |
301 | ceph_assert(bdev[id]); | |
7c673cae FG |
302 | return bdev[id]->supported_bdev_label(); |
303 | } | |
304 | ||
305 | uint64_t BlueFS::get_block_device_size(unsigned id) | |
306 | { | |
307 | if (id < bdev.size() && bdev[id]) | |
308 | return bdev[id]->get_size(); | |
309 | return 0; | |
310 | } | |
311 | ||
1911f103 TL |
312 | void BlueFS::_add_block_extent(unsigned id, uint64_t offset, uint64_t length, |
313 | bool skip) | |
7c673cae | 314 | { |
7c673cae | 315 | dout(1) << __func__ << " bdev " << id |
11fdf7f2 | 316 | << " 0x" << std::hex << offset << "~" << length << std::dec |
1911f103 | 317 | << " skip " << skip |
7c673cae | 318 | << dendl; |
11fdf7f2 TL |
319 | |
320 | ceph_assert(id < bdev.size()); | |
321 | ceph_assert(bdev[id]); | |
322 | ceph_assert(bdev[id]->get_size() >= offset + length); | |
7c673cae | 323 | block_all[id].insert(offset, length); |
7c673cae FG |
324 | |
325 | if (id < alloc.size() && alloc[id]) { | |
1911f103 TL |
326 | if (!skip) |
327 | log_t.op_alloc_add(id, offset, length); | |
328 | ||
7c673cae FG |
329 | alloc[id]->init_add_free(offset, length); |
330 | } | |
331 | ||
332 | if (logger) | |
333 | logger->inc(l_bluefs_gift_bytes, length); | |
334 | dout(10) << __func__ << " done" << dendl; | |
335 | } | |
336 | ||
337 | int BlueFS::reclaim_blocks(unsigned id, uint64_t want, | |
a8e16298 | 338 | PExtentVector *extents) |
7c673cae | 339 | { |
11fdf7f2 | 340 | std::unique_lock l(lock); |
7c673cae FG |
341 | dout(1) << __func__ << " bdev " << id |
342 | << " want 0x" << std::hex << want << std::dec << dendl; | |
11fdf7f2 TL |
343 | ceph_assert(id < alloc.size()); |
344 | ceph_assert(alloc[id]); | |
9f95a23c TL |
345 | int64_t got = 0; |
346 | ||
347 | interval_set<uint64_t> granular; | |
348 | while (want > 0 && !block_unused_too_granular[id].empty()) { | |
349 | auto p = block_unused_too_granular[id].begin(); | |
350 | dout(20) << __func__ << " unused " << (int)id << ":" | |
351 | << std::hex << p.get_start() << "~" << p.get_len() << dendl; | |
352 | extents->push_back({p.get_start(), p.get_len()}); | |
353 | granular.insert(p.get_start(), p.get_len()); | |
354 | if (want >= p.get_len()) { | |
355 | want -= p.get_len(); | |
356 | } else { | |
357 | want = 0; | |
358 | } | |
359 | got += p.get_len(); | |
360 | block_unused_too_granular[id].erase(p); | |
7c673cae FG |
361 | } |
362 | ||
9f95a23c TL |
363 | if (want > 0) { |
364 | got += alloc[id]->allocate(want, alloc_size[id], 0, extents); | |
365 | ceph_assert(got != 0); | |
366 | if (got < 0) { | |
367 | derr << __func__ << " failed to allocate space to return to bluestore" | |
368 | << dendl; | |
369 | alloc[id]->dump(); | |
370 | block_unused_too_granular[id].insert(granular); | |
371 | return got; | |
372 | } | |
7c673cae | 373 | |
9f95a23c TL |
374 | for (auto& p : *extents) { |
375 | block_all[id].erase(p.offset, p.length); | |
376 | log_t.op_alloc_rm(id, p.offset, p.length); | |
377 | } | |
378 | ||
379 | flush_bdev(); | |
380 | int r = _flush_and_sync_log(l); | |
381 | ceph_assert(r == 0); | |
382 | } | |
7c673cae | 383 | |
11fdf7f2 | 384 | logger->inc(l_bluefs_reclaim_bytes, got); |
7c673cae FG |
385 | dout(1) << __func__ << " bdev " << id << " want 0x" << std::hex << want |
386 | << " got " << *extents << dendl; | |
387 | return 0; | |
388 | } | |
389 | ||
11fdf7f2 | 390 | void BlueFS::handle_discard(unsigned id, interval_set<uint64_t>& to_release) |
7c673cae | 391 | { |
11fdf7f2 TL |
392 | dout(10) << __func__ << " bdev " << id << dendl; |
393 | ceph_assert(alloc[id]); | |
394 | alloc[id]->release(to_release); | |
395 | } | |
396 | ||
397 | uint64_t BlueFS::get_used() | |
398 | { | |
399 | std::lock_guard l(lock); | |
400 | uint64_t used = 0; | |
401 | for (unsigned id = 0; id < MAX_BDEV; ++id) { | |
402 | if (alloc[id]) { | |
403 | used += block_all[id].size() - alloc[id]->get_free(); | |
404 | } | |
7c673cae | 405 | } |
11fdf7f2 | 406 | return used; |
7c673cae FG |
407 | } |
408 | ||
409 | uint64_t BlueFS::get_total(unsigned id) | |
410 | { | |
11fdf7f2 TL |
411 | std::lock_guard l(lock); |
412 | ceph_assert(id < block_all.size()); | |
413 | return block_all[id].size(); | |
7c673cae FG |
414 | } |
415 | ||
416 | uint64_t BlueFS::get_free(unsigned id) | |
417 | { | |
11fdf7f2 TL |
418 | std::lock_guard l(lock); |
419 | ceph_assert(id < alloc.size()); | |
7c673cae FG |
420 | return alloc[id]->get_free(); |
421 | } | |
422 | ||
423 | void BlueFS::dump_perf_counters(Formatter *f) | |
424 | { | |
425 | f->open_object_section("bluefs_perf_counters"); | |
426 | logger->dump_formatted(f,0); | |
427 | f->close_section(); | |
428 | } | |
429 | ||
3efd9988 FG |
430 | void BlueFS::dump_block_extents(ostream& out) |
431 | { | |
432 | for (unsigned i = 0; i < MAX_BDEV; ++i) { | |
433 | if (!bdev[i]) { | |
434 | continue; | |
435 | } | |
11fdf7f2 TL |
436 | auto owned = get_total(i); |
437 | auto free = get_free(i); | |
1911f103 | 438 | |
11fdf7f2 TL |
439 | out << i << " : device size 0x" << std::hex << bdev[i]->get_size() |
440 | << " : own 0x" << block_all[i] | |
441 | << " = 0x" << owned | |
442 | << " : using 0x" << owned - free | |
1911f103 TL |
443 | << std::dec << "(" << byte_u_t(owned - free) << ")"; |
444 | if (i == _get_slow_device_id()) { | |
445 | ceph_assert(slow_dev_expander); | |
446 | ceph_assert(alloc[i]); | |
447 | free = slow_dev_expander->available_freespace(alloc_size[i]); | |
448 | out << std::hex | |
449 | << " : bluestore has 0x" << free | |
450 | << std::dec << "(" << byte_u_t(free) << ") available"; | |
451 | } | |
452 | out << "\n"; | |
3efd9988 FG |
453 | } |
454 | } | |
7c673cae FG |
455 | |
456 | void BlueFS::get_usage(vector<pair<uint64_t,uint64_t>> *usage) | |
457 | { | |
11fdf7f2 | 458 | std::lock_guard l(lock); |
7c673cae FG |
459 | usage->resize(bdev.size()); |
460 | for (unsigned id = 0; id < bdev.size(); ++id) { | |
461 | if (!bdev[id]) { | |
462 | (*usage)[id] = make_pair(0, 0); | |
463 | continue; | |
464 | } | |
465 | (*usage)[id].first = alloc[id]->get_free(); | |
11fdf7f2 | 466 | (*usage)[id].second = block_all[id].size(); |
7c673cae | 467 | uint64_t used = |
11fdf7f2 | 468 | (block_all[id].size() - (*usage)[id].first) * 100 / block_all[id].size(); |
7c673cae FG |
469 | dout(10) << __func__ << " bdev " << id |
470 | << " free " << (*usage)[id].first | |
1adf2230 | 471 | << " (" << byte_u_t((*usage)[id].first) << ")" |
7c673cae | 472 | << " / " << (*usage)[id].second |
1adf2230 | 473 | << " (" << byte_u_t((*usage)[id].second) << ")" |
7c673cae FG |
474 | << ", used " << used << "%" |
475 | << dendl; | |
476 | } | |
477 | } | |
478 | ||
479 | int BlueFS::get_block_extents(unsigned id, interval_set<uint64_t> *extents) | |
480 | { | |
11fdf7f2 | 481 | std::lock_guard l(lock); |
7c673cae FG |
482 | dout(10) << __func__ << " bdev " << id << dendl; |
483 | if (id >= block_all.size()) | |
484 | return -EINVAL; | |
485 | *extents = block_all[id]; | |
486 | return 0; | |
487 | } | |
488 | ||
9f95a23c | 489 | int BlueFS::mkfs(uuid_d osd_uuid, const bluefs_layout_t& layout) |
7c673cae | 490 | { |
11fdf7f2 | 491 | std::unique_lock l(lock); |
7c673cae FG |
492 | dout(1) << __func__ |
493 | << " osd_uuid " << osd_uuid | |
494 | << dendl; | |
495 | ||
9f95a23c TL |
496 | // set volume selector if not provided before/outside |
497 | if (vselector == nullptr) { | |
498 | vselector.reset( | |
499 | new OriginalVolumeSelector( | |
500 | get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
501 | get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
502 | get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100)); | |
503 | } | |
504 | ||
7c673cae FG |
505 | _init_alloc(); |
506 | _init_logger(); | |
507 | ||
508 | super.version = 1; | |
509 | super.block_size = bdev[BDEV_DB]->get_block_size(); | |
510 | super.osd_uuid = osd_uuid; | |
511 | super.uuid.generate_random(); | |
512 | dout(1) << __func__ << " uuid " << super.uuid << dendl; | |
513 | ||
514 | // init log | |
9f95a23c | 515 | FileRef log_file = ceph::make_ref<File>(); |
7c673cae | 516 | log_file->fnode.ino = 1; |
f6b5b4d7 | 517 | log_file->vselector_hint = vselector->get_hint_for_log(); |
7c673cae | 518 | int r = _allocate( |
9f95a23c | 519 | vselector->select_prefer_bdev(log_file->vselector_hint), |
7c673cae | 520 | cct->_conf->bluefs_max_log_runway, |
94b18763 | 521 | &log_file->fnode); |
9f95a23c | 522 | vselector->add_usage(log_file->vselector_hint, log_file->fnode); |
11fdf7f2 | 523 | ceph_assert(r == 0); |
7c673cae FG |
524 | log_writer = _create_writer(log_file); |
525 | ||
526 | // initial txn | |
527 | log_t.op_init(); | |
528 | for (unsigned bdev = 0; bdev < MAX_BDEV; ++bdev) { | |
529 | interval_set<uint64_t>& p = block_all[bdev]; | |
530 | if (p.empty()) | |
531 | continue; | |
532 | for (interval_set<uint64_t>::iterator q = p.begin(); q != p.end(); ++q) { | |
533 | dout(20) << __func__ << " op_alloc_add " << bdev << " 0x" | |
534 | << std::hex << q.get_start() << "~" << q.get_len() << std::dec | |
535 | << dendl; | |
536 | log_t.op_alloc_add(bdev, q.get_start(), q.get_len()); | |
537 | } | |
538 | } | |
539 | _flush_and_sync_log(l); | |
540 | ||
541 | // write supers | |
542 | super.log_fnode = log_file->fnode; | |
9f95a23c | 543 | super.memorized_layout = layout; |
11fdf7f2 | 544 | _write_super(BDEV_DB); |
7c673cae FG |
545 | flush_bdev(); |
546 | ||
547 | // clean up | |
548 | super = bluefs_super_t(); | |
549 | _close_writer(log_writer); | |
550 | log_writer = NULL; | |
551 | block_all.clear(); | |
9f95a23c | 552 | vselector.reset(nullptr); |
7c673cae FG |
553 | _stop_alloc(); |
554 | _shutdown_logger(); | |
555 | ||
556 | dout(10) << __func__ << " success" << dendl; | |
557 | return 0; | |
558 | } | |
559 | ||
560 | void BlueFS::_init_alloc() | |
561 | { | |
562 | dout(20) << __func__ << dendl; | |
563 | alloc.resize(MAX_BDEV); | |
eafe8130 | 564 | alloc_size.resize(MAX_BDEV, 0); |
7c673cae | 565 | pending_release.resize(MAX_BDEV); |
9f95a23c | 566 | block_unused_too_granular.resize(MAX_BDEV); |
eafe8130 TL |
567 | |
568 | if (bdev[BDEV_WAL]) { | |
569 | alloc_size[BDEV_WAL] = cct->_conf->bluefs_alloc_size; | |
570 | } | |
571 | if (bdev[BDEV_SLOW]) { | |
572 | alloc_size[BDEV_DB] = cct->_conf->bluefs_alloc_size; | |
573 | alloc_size[BDEV_SLOW] = cct->_conf->bluefs_shared_alloc_size; | |
574 | } else { | |
575 | alloc_size[BDEV_DB] = cct->_conf->bluefs_shared_alloc_size; | |
576 | } | |
577 | // new wal and db devices are never shared | |
578 | if (bdev[BDEV_NEWWAL]) { | |
579 | alloc_size[BDEV_NEWWAL] = cct->_conf->bluefs_alloc_size; | |
580 | } | |
581 | if (bdev[BDEV_NEWDB]) { | |
582 | alloc_size[BDEV_NEWDB] = cct->_conf->bluefs_alloc_size; | |
583 | } | |
584 | ||
7c673cae FG |
585 | for (unsigned id = 0; id < bdev.size(); ++id) { |
586 | if (!bdev[id]) { | |
587 | continue; | |
588 | } | |
11fdf7f2 | 589 | ceph_assert(bdev[id]->get_size()); |
eafe8130 TL |
590 | std::string name = "bluefs-"; |
591 | const char* devnames[] = {"wal","db","slow"}; | |
592 | if (id <= BDEV_SLOW) | |
593 | name += devnames[id]; | |
594 | else | |
595 | name += to_string(uintptr_t(this)); | |
596 | ceph_assert(alloc_size[id]); | |
597 | dout(1) << __func__ << " id " << id | |
598 | << " alloc_size 0x" << std::hex << alloc_size[id] | |
599 | << " size 0x" << bdev[id]->get_size() << std::dec << dendl; | |
7c673cae FG |
600 | alloc[id] = Allocator::create(cct, cct->_conf->bluefs_allocator, |
601 | bdev[id]->get_size(), | |
eafe8130 | 602 | alloc_size[id], name); |
7c673cae FG |
603 | interval_set<uint64_t>& p = block_all[id]; |
604 | for (interval_set<uint64_t>::iterator q = p.begin(); q != p.end(); ++q) { | |
605 | alloc[id]->init_add_free(q.get_start(), q.get_len()); | |
606 | } | |
607 | } | |
608 | } | |
609 | ||
610 | void BlueFS::_stop_alloc() | |
611 | { | |
612 | dout(20) << __func__ << dendl; | |
11fdf7f2 TL |
613 | for (auto p : bdev) { |
614 | if (p) | |
615 | p->discard_drain(); | |
616 | } | |
617 | ||
7c673cae FG |
618 | for (auto p : alloc) { |
619 | if (p != nullptr) { | |
620 | p->shutdown(); | |
621 | delete p; | |
622 | } | |
623 | } | |
624 | alloc.clear(); | |
9f95a23c | 625 | block_unused_too_granular.clear(); |
7c673cae FG |
626 | } |
627 | ||
628 | int BlueFS::mount() | |
629 | { | |
630 | dout(1) << __func__ << dendl; | |
631 | ||
632 | int r = _open_super(); | |
633 | if (r < 0) { | |
634 | derr << __func__ << " failed to open super: " << cpp_strerror(r) << dendl; | |
635 | goto out; | |
636 | } | |
637 | ||
9f95a23c TL |
638 | // set volume selector if not provided before/outside |
639 | if (vselector == nullptr) { | |
640 | vselector.reset( | |
641 | new OriginalVolumeSelector( | |
642 | get_block_device_size(BlueFS::BDEV_WAL) * 95 / 100, | |
643 | get_block_device_size(BlueFS::BDEV_DB) * 95 / 100, | |
644 | get_block_device_size(BlueFS::BDEV_SLOW) * 95 / 100)); | |
645 | } | |
646 | ||
7c673cae FG |
647 | block_all.clear(); |
648 | block_all.resize(MAX_BDEV); | |
7c673cae | 649 | _init_alloc(); |
494da23a | 650 | _init_logger(); |
7c673cae | 651 | |
11fdf7f2 | 652 | r = _replay(false, false); |
7c673cae FG |
653 | if (r < 0) { |
654 | derr << __func__ << " failed to replay log: " << cpp_strerror(r) << dendl; | |
655 | _stop_alloc(); | |
656 | goto out; | |
657 | } | |
658 | ||
659 | // init freelist | |
660 | for (auto& p : file_map) { | |
661 | dout(30) << __func__ << " noting alloc for " << p.second->fnode << dendl; | |
662 | for (auto& q : p.second->fnode.extents) { | |
663 | alloc[q.bdev]->init_rm_free(q.offset, q.length); | |
664 | } | |
665 | } | |
666 | ||
667 | // set up the log for future writes | |
668 | log_writer = _create_writer(_get_file(1)); | |
11fdf7f2 | 669 | ceph_assert(log_writer->file->fnode.ino == 1); |
7c673cae FG |
670 | log_writer->pos = log_writer->file->fnode.size; |
671 | dout(10) << __func__ << " log write pos set to 0x" | |
672 | << std::hex << log_writer->pos << std::dec | |
673 | << dendl; | |
674 | ||
7c673cae FG |
675 | return 0; |
676 | ||
677 | out: | |
678 | super = bluefs_super_t(); | |
679 | return r; | |
680 | } | |
681 | ||
9f95a23c TL |
682 | int BlueFS::maybe_verify_layout(const bluefs_layout_t& layout) const |
683 | { | |
684 | if (super.memorized_layout) { | |
685 | if (layout == *super.memorized_layout) { | |
686 | dout(10) << __func__ << " bluefs layout verified positively" << dendl; | |
687 | } else { | |
688 | derr << __func__ << " memorized layout doesn't fit current one" << dendl; | |
689 | return -EIO; | |
690 | } | |
691 | } else { | |
692 | dout(10) << __func__ << " no memorized_layout in bluefs superblock" | |
693 | << dendl; | |
694 | } | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
1911f103 | 699 | void BlueFS::umount(bool avoid_compact) |
7c673cae FG |
700 | { |
701 | dout(1) << __func__ << dendl; | |
702 | ||
1911f103 | 703 | sync_metadata(avoid_compact); |
7c673cae FG |
704 | |
705 | _close_writer(log_writer); | |
706 | log_writer = NULL; | |
707 | ||
9f95a23c | 708 | vselector.reset(nullptr); |
7c673cae FG |
709 | _stop_alloc(); |
710 | file_map.clear(); | |
711 | dir_map.clear(); | |
712 | super = bluefs_super_t(); | |
713 | log_t.clear(); | |
714 | _shutdown_logger(); | |
715 | } | |
716 | ||
9f95a23c | 717 | int BlueFS::prepare_new_device(int id, const bluefs_layout_t& layout) |
7c673cae | 718 | { |
11fdf7f2 TL |
719 | dout(1) << __func__ << dendl; |
720 | ||
721 | if(id == BDEV_NEWDB) { | |
722 | int new_log_dev_cur = BDEV_WAL; | |
723 | int new_log_dev_next = BDEV_WAL; | |
724 | if (!bdev[BDEV_WAL]) { | |
725 | new_log_dev_cur = BDEV_NEWDB; | |
726 | new_log_dev_next = BDEV_DB; | |
727 | } | |
9f95a23c | 728 | _rewrite_log_and_layout_sync(false, |
11fdf7f2 TL |
729 | BDEV_NEWDB, |
730 | new_log_dev_cur, | |
731 | new_log_dev_next, | |
9f95a23c TL |
732 | RENAME_DB2SLOW, |
733 | layout); | |
11fdf7f2 TL |
734 | //} |
735 | } else if(id == BDEV_NEWWAL) { | |
9f95a23c TL |
736 | _rewrite_log_and_layout_sync(false, |
737 | BDEV_DB, | |
738 | BDEV_NEWWAL, | |
739 | BDEV_WAL, | |
740 | REMOVE_WAL, | |
741 | layout); | |
11fdf7f2 TL |
742 | } else { |
743 | assert(false); | |
744 | } | |
745 | return 0; | |
746 | } | |
747 | ||
748 | void BlueFS::collect_metadata(map<string,string> *pm, unsigned skip_bdev_id) | |
749 | { | |
750 | if (skip_bdev_id != BDEV_DB && bdev[BDEV_DB]) | |
7c673cae FG |
751 | bdev[BDEV_DB]->collect_metadata("bluefs_db_", pm); |
752 | if (bdev[BDEV_WAL]) | |
753 | bdev[BDEV_WAL]->collect_metadata("bluefs_wal_", pm); | |
11fdf7f2 TL |
754 | } |
755 | ||
756 | void BlueFS::get_devices(set<string> *ls) | |
757 | { | |
758 | for (unsigned i = 0; i < MAX_BDEV; ++i) { | |
759 | if (bdev[i]) { | |
760 | bdev[i]->get_devices(ls); | |
761 | } | |
762 | } | |
7c673cae FG |
763 | } |
764 | ||
765 | int BlueFS::fsck() | |
766 | { | |
11fdf7f2 | 767 | std::lock_guard l(lock); |
7c673cae FG |
768 | dout(1) << __func__ << dendl; |
769 | // hrm, i think we check everything on mount... | |
770 | return 0; | |
771 | } | |
772 | ||
11fdf7f2 | 773 | int BlueFS::_write_super(int dev) |
7c673cae FG |
774 | { |
775 | // build superblock | |
776 | bufferlist bl; | |
11fdf7f2 | 777 | encode(super, bl); |
7c673cae | 778 | uint32_t crc = bl.crc32c(-1); |
11fdf7f2 | 779 | encode(crc, bl); |
7c673cae FG |
780 | dout(10) << __func__ << " super block length(encoded): " << bl.length() << dendl; |
781 | dout(10) << __func__ << " superblock " << super.version << dendl; | |
782 | dout(10) << __func__ << " log_fnode " << super.log_fnode << dendl; | |
9f95a23c | 783 | ceph_assert_always(bl.length() <= get_super_length()); |
7c673cae FG |
784 | bl.append_zero(get_super_length() - bl.length()); |
785 | ||
11fdf7f2 | 786 | bdev[dev]->write(get_super_offset(), bl, false, WRITE_LIFE_SHORT); |
7c673cae FG |
787 | dout(20) << __func__ << " v " << super.version |
788 | << " crc 0x" << std::hex << crc | |
789 | << " offset 0x" << get_super_offset() << std::dec | |
790 | << dendl; | |
791 | return 0; | |
792 | } | |
793 | ||
794 | int BlueFS::_open_super() | |
795 | { | |
796 | dout(10) << __func__ << dendl; | |
797 | ||
798 | bufferlist bl; | |
799 | uint32_t expected_crc, crc; | |
800 | int r; | |
801 | ||
802 | // always the second block | |
803 | r = bdev[BDEV_DB]->read(get_super_offset(), get_super_length(), | |
804 | &bl, ioc[BDEV_DB], false); | |
805 | if (r < 0) | |
806 | return r; | |
807 | ||
11fdf7f2 TL |
808 | auto p = bl.cbegin(); |
809 | decode(super, p); | |
7c673cae FG |
810 | { |
811 | bufferlist t; | |
812 | t.substr_of(bl, 0, p.get_off()); | |
813 | crc = t.crc32c(-1); | |
814 | } | |
11fdf7f2 | 815 | decode(expected_crc, p); |
7c673cae FG |
816 | if (crc != expected_crc) { |
817 | derr << __func__ << " bad crc on superblock, expected 0x" | |
818 | << std::hex << expected_crc << " != actual 0x" << crc << std::dec | |
819 | << dendl; | |
820 | return -EIO; | |
821 | } | |
822 | dout(10) << __func__ << " superblock " << super.version << dendl; | |
823 | dout(10) << __func__ << " log_fnode " << super.log_fnode << dendl; | |
824 | return 0; | |
825 | } | |
826 | ||
9f95a23c TL |
827 | int BlueFS::_check_new_allocations(const bluefs_fnode_t& fnode, |
828 | size_t dev_count, | |
829 | boost::dynamic_bitset<uint64_t>* owned_blocks, | |
830 | boost::dynamic_bitset<uint64_t>* used_blocks) | |
831 | { | |
832 | auto& fnode_extents = fnode.extents; | |
833 | for (auto e : fnode_extents) { | |
834 | auto id = e.bdev; | |
835 | bool fail = false; | |
836 | ceph_assert(id < dev_count); | |
837 | apply_for_bitset_range(e.offset, e.length, alloc_size[id], owned_blocks[id], | |
838 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
839 | if (!bs.test(pos)) { | |
840 | fail = true; | |
841 | } | |
842 | } | |
843 | ); | |
844 | if (fail) { | |
845 | derr << __func__ << " invalid extent " << int(id) | |
846 | << ": 0x" << std::hex << e.offset << "~" << e.length | |
847 | << std::dec | |
848 | << ": wasn't given but allocated for ino " << fnode.ino | |
849 | << dendl; | |
850 | return -EFAULT; | |
851 | } | |
852 | ||
853 | apply_for_bitset_range(e.offset, e.length, alloc_size[id], used_blocks[id], | |
854 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
855 | if (bs.test(pos)) { | |
856 | fail = true; | |
857 | } | |
858 | bs.set(pos); | |
859 | } | |
860 | ); | |
861 | if (fail) { | |
862 | derr << __func__ << " invalid extent " << int(e.bdev) | |
863 | << ": 0x" << std::hex << e.offset << "~" << e.length | |
864 | << std::dec << ": duplicate reference, ino " << fnode.ino | |
865 | << dendl; | |
866 | return -EFAULT; | |
867 | } | |
868 | } | |
869 | return 0; | |
870 | } | |
871 | ||
872 | int BlueFS::_adjust_granularity( | |
873 | __u8 id, uint64_t *offset, uint64_t *length, bool alloc) | |
874 | { | |
875 | const char *op = alloc ? "op_alloc_add" : "op_alloc_rm"; | |
876 | auto oldo = *offset; | |
877 | auto oldl = *length; | |
878 | if (*offset & (alloc_size[id] - 1)) { | |
879 | *offset &= ~(alloc_size[id] - 1); | |
880 | *offset += alloc_size[id]; | |
881 | if (*length > *offset - oldo) { | |
882 | if (alloc) { | |
883 | block_unused_too_granular[id].insert(oldo, *offset - oldo); | |
884 | } else { | |
885 | block_unused_too_granular[id].erase(oldo, *offset - oldo); | |
886 | } | |
887 | *length -= (*offset - oldo); | |
888 | } else { | |
889 | if (alloc) { | |
890 | block_unused_too_granular[id].insert(oldo, *length); | |
891 | } else { | |
892 | block_unused_too_granular[id].erase(oldo, *length); | |
893 | } | |
894 | *length = 0; | |
895 | } | |
896 | } | |
897 | if (*length & (alloc_size[id] - 1)) { | |
898 | *length &= ~(alloc_size[id] - 1); | |
899 | if (alloc) { | |
900 | block_unused_too_granular[id].insert( | |
901 | *offset + *length, | |
902 | oldo + oldl - *offset - *length); | |
903 | } else { | |
904 | block_unused_too_granular[id].erase( | |
905 | *offset + *length, | |
906 | oldo + oldl - *offset - *length); | |
907 | } | |
908 | } | |
909 | if (oldo != *offset || oldl != *length) { | |
910 | dout(10) << __func__ << " " << op << " " | |
911 | << (int)id << ":" << std::hex << oldo << "~" << oldl | |
912 | << " -> " << (int)id << ":" << *offset << "~" << *length << dendl; | |
913 | } | |
914 | return 0; | |
915 | } | |
916 | ||
917 | int BlueFS::_verify_alloc_granularity( | |
918 | __u8 id, uint64_t offset, uint64_t length, const char *op) | |
919 | { | |
920 | if ((offset & (alloc_size[id] - 1)) || | |
921 | (length & (alloc_size[id] - 1))) { | |
922 | derr << __func__ << " " << op << " of " << (int)id | |
923 | << ":0x" << std::hex << offset << "~" << length << std::dec | |
924 | << " does not align to alloc_size 0x" | |
925 | << std::hex << alloc_size[id] << std::dec << dendl; | |
926 | // be helpful | |
927 | auto need = alloc_size[id]; | |
928 | while (need && ((offset & (need - 1)) || | |
929 | (length & (need - 1)))) { | |
930 | need >>= 1; | |
931 | } | |
932 | if (need) { | |
933 | const char *which; | |
934 | if (id == BDEV_SLOW || | |
935 | (id == BDEV_DB && !bdev[BDEV_SLOW])) { | |
936 | which = "bluefs_shared_alloc_size"; | |
937 | } else { | |
938 | which = "bluefs_alloc_size"; | |
939 | } | |
940 | derr << "work-around by setting " << which << " = " << need | |
941 | << " for this OSD" << dendl; | |
942 | } | |
943 | return -EFAULT; | |
944 | } | |
945 | return 0; | |
946 | } | |
947 | ||
11fdf7f2 | 948 | int BlueFS::_replay(bool noop, bool to_stdout) |
7c673cae FG |
949 | { |
950 | dout(10) << __func__ << (noop ? " NO-OP" : "") << dendl; | |
951 | ino_last = 1; // by the log | |
952 | log_seq = 0; | |
953 | ||
954 | FileRef log_file; | |
11fdf7f2 | 955 | log_file = _get_file(1); |
9f95a23c TL |
956 | |
957 | // sanity check | |
958 | for (auto& a : block_unused_too_granular) { | |
959 | ceph_assert(a.empty()); | |
960 | } | |
961 | ||
11fdf7f2 TL |
962 | if (!noop) { |
963 | log_file->fnode = super.log_fnode; | |
9f95a23c | 964 | log_file->vselector_hint = |
f6b5b4d7 | 965 | vselector->get_hint_for_log(); |
7c673cae | 966 | } else { |
11fdf7f2 TL |
967 | // do not use fnode from superblock in 'noop' mode - log_file's one should |
968 | // be fine and up-to-date | |
969 | ceph_assert(log_file->fnode.ino == 1); | |
970 | ceph_assert(log_file->fnode.extents.size() != 0); | |
7c673cae | 971 | } |
7c673cae | 972 | dout(10) << __func__ << " log_fnode " << super.log_fnode << dendl; |
11fdf7f2 TL |
973 | if (unlikely(to_stdout)) { |
974 | std::cout << " log_fnode " << super.log_fnode << std::endl; | |
975 | } | |
7c673cae FG |
976 | |
977 | FileReader *log_reader = new FileReader( | |
978 | log_file, cct->_conf->bluefs_max_prefetch, | |
979 | false, // !random | |
980 | true); // ignore eof | |
9f95a23c TL |
981 | |
982 | bool seen_recs = false; | |
983 | ||
984 | boost::dynamic_bitset<uint64_t> used_blocks[MAX_BDEV]; | |
985 | boost::dynamic_bitset<uint64_t> owned_blocks[MAX_BDEV]; | |
986 | ||
987 | if (cct->_conf->bluefs_log_replay_check_allocations) { | |
988 | for (size_t i = 0; i < MAX_BDEV; ++i) { | |
989 | if (alloc_size[i] != 0 && bdev[i] != nullptr) { | |
990 | used_blocks[i].resize(round_up_to(bdev[i]->get_size(), alloc_size[i]) / alloc_size[i]); | |
991 | owned_blocks[i].resize(round_up_to(bdev[i]->get_size(), alloc_size[i]) / alloc_size[i]); | |
992 | } | |
993 | } | |
994 | } | |
995 | ||
996 | bool first_log_check = true; | |
997 | ||
7c673cae | 998 | while (true) { |
11fdf7f2 | 999 | ceph_assert((log_reader->buf.pos & ~super.block_mask()) == 0); |
7c673cae FG |
1000 | uint64_t pos = log_reader->buf.pos; |
1001 | uint64_t read_pos = pos; | |
1002 | bufferlist bl; | |
1003 | { | |
1004 | int r = _read(log_reader, &log_reader->buf, read_pos, super.block_size, | |
1005 | &bl, NULL); | |
f6b5b4d7 TL |
1006 | if (r != (int)super.block_size && cct->_conf->bluefs_replay_recovery) { |
1007 | r += do_replay_recovery_read(log_reader, pos, read_pos + r, super.block_size - r, &bl); | |
1008 | } | |
1009 | assert(r == (int)super.block_size); | |
7c673cae FG |
1010 | read_pos += r; |
1011 | } | |
1012 | uint64_t more = 0; | |
1013 | uint64_t seq; | |
1014 | uuid_d uuid; | |
1015 | { | |
11fdf7f2 | 1016 | auto p = bl.cbegin(); |
7c673cae FG |
1017 | __u8 a, b; |
1018 | uint32_t len; | |
11fdf7f2 TL |
1019 | decode(a, p); |
1020 | decode(b, p); | |
1021 | decode(len, p); | |
1022 | decode(uuid, p); | |
1023 | decode(seq, p); | |
7c673cae | 1024 | if (len + 6 > bl.length()) { |
11fdf7f2 | 1025 | more = round_up_to(len + 6 - bl.length(), super.block_size); |
7c673cae FG |
1026 | } |
1027 | } | |
1028 | if (uuid != super.uuid) { | |
9f95a23c TL |
1029 | if (seen_recs) { |
1030 | dout(10) << __func__ << " 0x" << std::hex << pos << std::dec | |
1031 | << ": stop: uuid " << uuid << " != super.uuid " << super.uuid | |
1032 | << dendl; | |
1033 | } else { | |
1034 | derr << __func__ << " 0x" << std::hex << pos << std::dec | |
1035 | << ": stop: uuid " << uuid << " != super.uuid " << super.uuid | |
1036 | << ", block dump: \n"; | |
1037 | bufferlist t; | |
1038 | t.substr_of(bl, 0, super.block_size); | |
1039 | t.hexdump(*_dout); | |
1040 | *_dout << dendl; | |
1041 | } | |
7c673cae FG |
1042 | break; |
1043 | } | |
1044 | if (seq != log_seq + 1) { | |
9f95a23c TL |
1045 | if (seen_recs) { |
1046 | dout(10) << __func__ << " 0x" << std::hex << pos << std::dec | |
1047 | << ": stop: seq " << seq << " != expected " << log_seq + 1 | |
1048 | << dendl;; | |
1049 | } else { | |
1050 | derr << __func__ << " 0x" << std::hex << pos << std::dec | |
1051 | << ": stop: seq " << seq << " != expected " << log_seq + 1 | |
1052 | << dendl;; | |
1053 | } | |
7c673cae FG |
1054 | break; |
1055 | } | |
1056 | if (more) { | |
1057 | dout(20) << __func__ << " need 0x" << std::hex << more << std::dec | |
1058 | << " more bytes" << dendl; | |
1059 | bufferlist t; | |
1060 | int r = _read(log_reader, &log_reader->buf, read_pos, more, &t, NULL); | |
1061 | if (r < (int)more) { | |
f6b5b4d7 TL |
1062 | dout(10) << __func__ << " 0x" << std::hex << pos |
1063 | << ": stop: len is 0x" << bl.length() + more << std::dec | |
1064 | << ", which is past eof" << dendl; | |
1065 | if (cct->_conf->bluefs_replay_recovery) { | |
1066 | //try to search for more data | |
1067 | r += do_replay_recovery_read(log_reader, pos, read_pos + r, more - r, &t); | |
1068 | if (r < (int)more) { | |
1069 | //in normal mode we must read r==more, for recovery it is too strict | |
1070 | break; | |
1071 | } | |
1072 | } | |
7c673cae | 1073 | } |
11fdf7f2 | 1074 | ceph_assert(r == (int)more); |
7c673cae FG |
1075 | bl.claim_append(t); |
1076 | read_pos += r; | |
1077 | } | |
9f95a23c | 1078 | seen_recs = true; |
7c673cae FG |
1079 | bluefs_transaction_t t; |
1080 | try { | |
11fdf7f2 TL |
1081 | auto p = bl.cbegin(); |
1082 | decode(t, p); | |
7c673cae FG |
1083 | } |
1084 | catch (buffer::error& e) { | |
9f95a23c TL |
1085 | derr << __func__ << " 0x" << std::hex << pos << std::dec |
1086 | << ": stop: failed to decode: " << e.what() | |
1087 | << dendl; | |
7c673cae FG |
1088 | delete log_reader; |
1089 | return -EIO; | |
1090 | } | |
11fdf7f2 | 1091 | ceph_assert(seq == t.seq); |
7c673cae FG |
1092 | dout(10) << __func__ << " 0x" << std::hex << pos << std::dec |
1093 | << ": " << t << dendl; | |
11fdf7f2 TL |
1094 | if (unlikely(to_stdout)) { |
1095 | std::cout << " 0x" << std::hex << pos << std::dec | |
1096 | << ": " << t << std::endl; | |
1097 | } | |
7c673cae | 1098 | |
11fdf7f2 | 1099 | auto p = t.op_bl.cbegin(); |
7c673cae FG |
1100 | while (!p.end()) { |
1101 | __u8 op; | |
11fdf7f2 | 1102 | decode(op, p); |
7c673cae FG |
1103 | switch (op) { |
1104 | ||
1105 | case bluefs_transaction_t::OP_INIT: | |
1106 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec | |
1107 | << ": op_init" << dendl; | |
11fdf7f2 TL |
1108 | if (unlikely(to_stdout)) { |
1109 | std::cout << " 0x" << std::hex << pos << std::dec | |
1110 | << ": op_init" << std::endl; | |
1111 | } | |
1112 | ||
1113 | ceph_assert(t.seq == 1); | |
7c673cae FG |
1114 | break; |
1115 | ||
1116 | case bluefs_transaction_t::OP_JUMP: | |
1117 | { | |
1118 | uint64_t next_seq; | |
1119 | uint64_t offset; | |
11fdf7f2 TL |
1120 | decode(next_seq, p); |
1121 | decode(offset, p); | |
7c673cae FG |
1122 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1123 | << ": op_jump seq " << next_seq | |
1124 | << " offset 0x" << std::hex << offset << std::dec << dendl; | |
11fdf7f2 TL |
1125 | if (unlikely(to_stdout)) { |
1126 | std::cout << " 0x" << std::hex << pos << std::dec | |
1127 | << ": op_jump seq " << next_seq | |
1128 | << " offset 0x" << std::hex << offset << std::dec | |
1129 | << std::endl; | |
1130 | } | |
1131 | ||
1132 | ceph_assert(next_seq >= log_seq); | |
7c673cae FG |
1133 | log_seq = next_seq - 1; // we will increment it below |
1134 | uint64_t skip = offset - read_pos; | |
1135 | if (skip) { | |
1136 | bufferlist junk; | |
1137 | int r = _read(log_reader, &log_reader->buf, read_pos, skip, &junk, | |
1138 | NULL); | |
1139 | if (r != (int)skip) { | |
1140 | dout(10) << __func__ << " 0x" << std::hex << read_pos | |
1141 | << ": stop: failed to skip to " << offset | |
1142 | << std::dec << dendl; | |
11fdf7f2 | 1143 | ceph_abort_msg("problem with op_jump"); |
7c673cae FG |
1144 | } |
1145 | } | |
1146 | } | |
1147 | break; | |
1148 | ||
1149 | case bluefs_transaction_t::OP_JUMP_SEQ: | |
1150 | { | |
1151 | uint64_t next_seq; | |
11fdf7f2 | 1152 | decode(next_seq, p); |
7c673cae FG |
1153 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1154 | << ": op_jump_seq " << next_seq << dendl; | |
11fdf7f2 TL |
1155 | if (unlikely(to_stdout)) { |
1156 | std::cout << " 0x" << std::hex << pos << std::dec | |
1157 | << ": op_jump_seq " << next_seq << std::endl; | |
1158 | } | |
1159 | ||
1160 | ceph_assert(next_seq >= log_seq); | |
7c673cae FG |
1161 | log_seq = next_seq - 1; // we will increment it below |
1162 | } | |
1163 | break; | |
1164 | ||
1165 | case bluefs_transaction_t::OP_ALLOC_ADD: | |
1166 | { | |
1167 | __u8 id; | |
1168 | uint64_t offset, length; | |
11fdf7f2 TL |
1169 | decode(id, p); |
1170 | decode(offset, p); | |
1171 | decode(length, p); | |
7c673cae FG |
1172 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1173 | << ": op_alloc_add " << " " << (int)id | |
1174 | << ":0x" << std::hex << offset << "~" << length << std::dec | |
1175 | << dendl; | |
11fdf7f2 TL |
1176 | if (unlikely(to_stdout)) { |
1177 | std::cout << " 0x" << std::hex << pos << std::dec | |
1178 | << ": op_alloc_add " << " " << (int)id | |
1179 | << ":0x" << std::hex << offset << "~" << length << std::dec | |
1180 | << std::endl; | |
1181 | } | |
7c673cae FG |
1182 | if (!noop) { |
1183 | block_all[id].insert(offset, length); | |
9f95a23c TL |
1184 | _adjust_granularity(id, &offset, &length, true); |
1185 | if (length) { | |
1186 | alloc[id]->init_add_free(offset, length); | |
1187 | } | |
1188 | ||
1189 | if (cct->_conf->bluefs_log_replay_check_allocations) { | |
1190 | bool fail = false; | |
1191 | apply_for_bitset_range(offset, length, alloc_size[id], owned_blocks[id], | |
1192 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1193 | if (bs.test(pos)) { | |
1194 | fail = true; | |
1195 | } else { | |
1196 | bs.set(pos); | |
1197 | } | |
1198 | } | |
1199 | ); | |
1200 | if (fail) { | |
1201 | derr << __func__ << " invalid extent " << (int)id | |
1202 | << ": 0x" << std::hex << offset << "~" << length | |
1203 | << std::dec << ": already given" << dendl; | |
1204 | return -EFAULT; | |
1205 | } | |
1206 | apply_for_bitset_range(offset, length, alloc_size[id], used_blocks[id], | |
1207 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1208 | if (bs.test(pos)) { | |
1209 | fail = true; | |
1210 | } | |
1211 | } | |
1212 | ); | |
1213 | if (fail) { | |
1214 | derr << __func__ << " invalid extent " << int(id) | |
1215 | << ": 0x" << std::hex << offset << "~" << length | |
1216 | << std::dec << ": already in use" << dendl; | |
1217 | return -EFAULT; | |
1218 | } | |
1219 | } | |
7c673cae FG |
1220 | } |
1221 | } | |
1222 | break; | |
1223 | ||
1224 | case bluefs_transaction_t::OP_ALLOC_RM: | |
1225 | { | |
1226 | __u8 id; | |
1227 | uint64_t offset, length; | |
11fdf7f2 TL |
1228 | decode(id, p); |
1229 | decode(offset, p); | |
1230 | decode(length, p); | |
7c673cae FG |
1231 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1232 | << ": op_alloc_rm " << " " << (int)id | |
1233 | << ":0x" << std::hex << offset << "~" << length << std::dec | |
1234 | << dendl; | |
11fdf7f2 TL |
1235 | if (unlikely(to_stdout)) { |
1236 | std::cout << " 0x" << std::hex << pos << std::dec | |
1237 | << ": op_alloc_rm " << " " << (int)id | |
1238 | << ":0x" << std::hex << offset << "~" << length << std::dec | |
1239 | << std::endl; | |
1240 | } | |
7c673cae FG |
1241 | if (!noop) { |
1242 | block_all[id].erase(offset, length); | |
9f95a23c TL |
1243 | _adjust_granularity(id, &offset, &length, false); |
1244 | if (length) { | |
1245 | alloc[id]->init_rm_free(offset, length); | |
1246 | } | |
1247 | if (cct->_conf->bluefs_log_replay_check_allocations) { | |
1248 | bool fail = false; | |
1249 | apply_for_bitset_range(offset, length, alloc_size[id], owned_blocks[id], | |
1250 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1251 | if (!bs.test(pos)) { | |
1252 | fail = true; | |
1253 | } else { | |
1254 | bs.reset(pos); | |
1255 | } | |
1256 | } | |
1257 | ); | |
1258 | if (fail) { | |
1259 | derr << __func__ << " invalid extent " << int(id) | |
1260 | << ": 0x" << std::hex << offset << "~" << length | |
1261 | << std::dec << ": wasn't given" << dendl; | |
1262 | return -EFAULT; | |
1263 | } | |
1264 | ||
1265 | apply_for_bitset_range(offset, length, alloc_size[id], used_blocks[id], | |
1266 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1267 | if (bs.test(pos)) { | |
1268 | fail = true; | |
1269 | } | |
1270 | } | |
1271 | ); | |
1272 | if (fail) { | |
1273 | derr << __func__ << " invalid extent " << (int)id | |
1274 | << ": 0x" << std::hex << offset << "~" << length | |
1275 | << std::dec << ": still in use" << dendl; | |
1276 | return -EFAULT; | |
1277 | } | |
1278 | } | |
1279 | } | |
7c673cae FG |
1280 | } |
1281 | break; | |
1282 | ||
1283 | case bluefs_transaction_t::OP_DIR_LINK: | |
1284 | { | |
1285 | string dirname, filename; | |
1286 | uint64_t ino; | |
11fdf7f2 TL |
1287 | decode(dirname, p); |
1288 | decode(filename, p); | |
1289 | decode(ino, p); | |
7c673cae FG |
1290 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1291 | << ": op_dir_link " << " " << dirname << "/" << filename | |
1292 | << " to " << ino | |
1293 | << dendl; | |
11fdf7f2 TL |
1294 | if (unlikely(to_stdout)) { |
1295 | std::cout << " 0x" << std::hex << pos << std::dec | |
1296 | << ": op_dir_link " << " " << dirname << "/" << filename | |
1297 | << " to " << ino | |
1298 | << std::endl; | |
1299 | } | |
1300 | ||
7c673cae FG |
1301 | if (!noop) { |
1302 | FileRef file = _get_file(ino); | |
11fdf7f2 | 1303 | ceph_assert(file->fnode.ino); |
7c673cae | 1304 | map<string,DirRef>::iterator q = dir_map.find(dirname); |
11fdf7f2 | 1305 | ceph_assert(q != dir_map.end()); |
7c673cae | 1306 | map<string,FileRef>::iterator r = q->second->file_map.find(filename); |
11fdf7f2 | 1307 | ceph_assert(r == q->second->file_map.end()); |
9f95a23c TL |
1308 | |
1309 | vselector->sub_usage(file->vselector_hint, file->fnode); | |
1310 | file->vselector_hint = | |
1311 | vselector->get_hint_by_dir(dirname); | |
1312 | vselector->add_usage(file->vselector_hint, file->fnode); | |
1313 | ||
7c673cae FG |
1314 | q->second->file_map[filename] = file; |
1315 | ++file->refs; | |
1316 | } | |
1317 | } | |
1318 | break; | |
1319 | ||
1320 | case bluefs_transaction_t::OP_DIR_UNLINK: | |
1321 | { | |
1322 | string dirname, filename; | |
11fdf7f2 TL |
1323 | decode(dirname, p); |
1324 | decode(filename, p); | |
7c673cae FG |
1325 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1326 | << ": op_dir_unlink " << " " << dirname << "/" << filename | |
1327 | << dendl; | |
11fdf7f2 TL |
1328 | if (unlikely(to_stdout)) { |
1329 | std::cout << " 0x" << std::hex << pos << std::dec | |
1330 | << ": op_dir_unlink " << " " << dirname << "/" << filename | |
1331 | << std::endl; | |
1332 | } | |
1333 | ||
7c673cae FG |
1334 | if (!noop) { |
1335 | map<string,DirRef>::iterator q = dir_map.find(dirname); | |
11fdf7f2 | 1336 | ceph_assert(q != dir_map.end()); |
7c673cae | 1337 | map<string,FileRef>::iterator r = q->second->file_map.find(filename); |
11fdf7f2 TL |
1338 | ceph_assert(r != q->second->file_map.end()); |
1339 | ceph_assert(r->second->refs > 0); | |
7c673cae FG |
1340 | --r->second->refs; |
1341 | q->second->file_map.erase(r); | |
1342 | } | |
1343 | } | |
1344 | break; | |
1345 | ||
1346 | case bluefs_transaction_t::OP_DIR_CREATE: | |
1347 | { | |
1348 | string dirname; | |
11fdf7f2 | 1349 | decode(dirname, p); |
7c673cae FG |
1350 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1351 | << ": op_dir_create " << dirname << dendl; | |
11fdf7f2 TL |
1352 | if (unlikely(to_stdout)) { |
1353 | std::cout << " 0x" << std::hex << pos << std::dec | |
1354 | << ": op_dir_create " << dirname << std::endl; | |
1355 | } | |
1356 | ||
7c673cae FG |
1357 | if (!noop) { |
1358 | map<string,DirRef>::iterator q = dir_map.find(dirname); | |
11fdf7f2 | 1359 | ceph_assert(q == dir_map.end()); |
9f95a23c | 1360 | dir_map[dirname] = ceph::make_ref<Dir>(); |
7c673cae FG |
1361 | } |
1362 | } | |
1363 | break; | |
1364 | ||
1365 | case bluefs_transaction_t::OP_DIR_REMOVE: | |
1366 | { | |
1367 | string dirname; | |
11fdf7f2 | 1368 | decode(dirname, p); |
7c673cae FG |
1369 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1370 | << ": op_dir_remove " << dirname << dendl; | |
11fdf7f2 TL |
1371 | if (unlikely(to_stdout)) { |
1372 | std::cout << " 0x" << std::hex << pos << std::dec | |
1373 | << ": op_dir_remove " << dirname << std::endl; | |
1374 | } | |
1375 | ||
7c673cae FG |
1376 | if (!noop) { |
1377 | map<string,DirRef>::iterator q = dir_map.find(dirname); | |
11fdf7f2 TL |
1378 | ceph_assert(q != dir_map.end()); |
1379 | ceph_assert(q->second->file_map.empty()); | |
7c673cae FG |
1380 | dir_map.erase(q); |
1381 | } | |
1382 | } | |
1383 | break; | |
1384 | ||
1385 | case bluefs_transaction_t::OP_FILE_UPDATE: | |
1386 | { | |
1387 | bluefs_fnode_t fnode; | |
11fdf7f2 | 1388 | decode(fnode, p); |
7c673cae | 1389 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
9f95a23c | 1390 | << ": op_file_update " << " " << fnode << " " << dendl; |
11fdf7f2 TL |
1391 | if (unlikely(to_stdout)) { |
1392 | std::cout << " 0x" << std::hex << pos << std::dec | |
1393 | << ": op_file_update " << " " << fnode << std::endl; | |
1394 | } | |
9f95a23c | 1395 | if (!noop) { |
7c673cae | 1396 | FileRef f = _get_file(fnode.ino); |
9f95a23c TL |
1397 | if (cct->_conf->bluefs_log_replay_check_allocations) { |
1398 | // check initial log layout | |
1399 | if (first_log_check) { | |
1400 | first_log_check = false; | |
1401 | int r = _check_new_allocations(log_file->fnode, | |
1402 | MAX_BDEV, owned_blocks, used_blocks); | |
1403 | if (r < 0) { | |
1404 | return r; | |
1405 | } | |
1406 | } | |
1407 | ||
1408 | auto& fnode_extents = f->fnode.extents; | |
1409 | for (auto e : fnode_extents) { | |
1410 | auto id = e.bdev; | |
1411 | if (int r = _verify_alloc_granularity(id, e.offset, e.length, | |
1412 | "OP_FILE_UPDATE"); r < 0) { | |
1413 | return r; | |
1414 | } | |
1415 | apply_for_bitset_range(e.offset, e.length, alloc_size[id], | |
1416 | used_blocks[id], | |
1417 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1418 | ceph_assert(bs.test(pos)); | |
1419 | bs.reset(pos); | |
1420 | } | |
1421 | ); | |
1422 | } | |
1423 | } | |
1424 | ||
1425 | if (fnode.ino != 1) { | |
1426 | vselector->sub_usage(f->vselector_hint, f->fnode); | |
1427 | } | |
1428 | f->fnode = fnode; | |
1429 | if (fnode.ino != 1) { | |
1430 | vselector->add_usage(f->vselector_hint, f->fnode); | |
1431 | } | |
1432 | ||
7c673cae FG |
1433 | if (fnode.ino > ino_last) { |
1434 | ino_last = fnode.ino; | |
1435 | } | |
9f95a23c TL |
1436 | if (cct->_conf->bluefs_log_replay_check_allocations) { |
1437 | int r = _check_new_allocations(f->fnode, | |
1438 | MAX_BDEV, owned_blocks, used_blocks); | |
1439 | if (r < 0) { | |
1440 | return r; | |
1441 | } | |
1442 | } | |
7c673cae | 1443 | } |
9f95a23c | 1444 | } |
7c673cae FG |
1445 | break; |
1446 | ||
1447 | case bluefs_transaction_t::OP_FILE_REMOVE: | |
1448 | { | |
1449 | uint64_t ino; | |
11fdf7f2 | 1450 | decode(ino, p); |
7c673cae FG |
1451 | dout(20) << __func__ << " 0x" << std::hex << pos << std::dec |
1452 | << ": op_file_remove " << ino << dendl; | |
11fdf7f2 TL |
1453 | if (unlikely(to_stdout)) { |
1454 | std::cout << " 0x" << std::hex << pos << std::dec | |
1455 | << ": op_file_remove " << ino << std::endl; | |
1456 | } | |
1457 | ||
9f95a23c TL |
1458 | if (!noop) { |
1459 | auto p = file_map.find(ino); | |
1460 | ceph_assert(p != file_map.end()); | |
1461 | vselector->sub_usage(p->second->vselector_hint, p->second->fnode); | |
1462 | if (cct->_conf->bluefs_log_replay_check_allocations) { | |
1463 | auto& fnode_extents = p->second->fnode.extents; | |
1464 | for (auto e : fnode_extents) { | |
1465 | auto id = e.bdev; | |
1466 | bool fail = false; | |
1467 | apply_for_bitset_range(e.offset, e.length, alloc_size[id], owned_blocks[id], | |
1468 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1469 | if (!bs.test(pos)) { | |
1470 | fail = true; | |
1471 | } | |
1472 | } | |
1473 | ); | |
1474 | if (fail) { | |
1475 | derr << __func__ << " invalid extent " << int(id) | |
1476 | << ": 0x" << std::hex << e.offset << "~" << e.length | |
1477 | << std::dec | |
1478 | << ": wasn't given but is allocated for removed ino " << ino | |
1479 | << dendl; | |
1480 | return -EFAULT; | |
1481 | } | |
1482 | ||
1483 | apply_for_bitset_range(e.offset, e.length, alloc_size[id], used_blocks[id], | |
1484 | [&](uint64_t pos, boost::dynamic_bitset<uint64_t> &bs) { | |
1485 | if (!bs.test(pos)) { | |
1486 | fail = true; | |
1487 | } | |
1488 | bs.reset(pos); | |
1489 | } | |
1490 | ); | |
1491 | if (fail) { | |
1492 | derr << __func__ << " invalid extent " << int(id) | |
1493 | << ": 0x" << std::hex << e.offset << "~" << e.length | |
1494 | << std::dec | |
1495 | << ": not in use but is allocated for removed ino " << ino | |
1496 | << dendl; | |
1497 | return -EFAULT; | |
1498 | } | |
1499 | } | |
1500 | } | |
1501 | file_map.erase(p); | |
1502 | } | |
1503 | } | |
7c673cae FG |
1504 | break; |
1505 | ||
1506 | default: | |
1507 | derr << __func__ << " 0x" << std::hex << pos << std::dec | |
1508 | << ": stop: unrecognized op " << (int)op << dendl; | |
1509 | delete log_reader; | |
1510 | return -EIO; | |
1511 | } | |
1512 | } | |
11fdf7f2 | 1513 | ceph_assert(p.end()); |
7c673cae FG |
1514 | |
1515 | // we successfully replayed the transaction; bump the seq and log size | |
1516 | ++log_seq; | |
1517 | log_file->fnode.size = log_reader->buf.pos; | |
1518 | } | |
9f95a23c TL |
1519 | vselector->add_usage(log_file->vselector_hint, log_file->fnode); |
1520 | ||
1521 | if (!noop && first_log_check && | |
1522 | cct->_conf->bluefs_log_replay_check_allocations) { | |
1523 | int r = _check_new_allocations(log_file->fnode, | |
1524 | MAX_BDEV, owned_blocks, used_blocks); | |
1525 | if (r < 0) { | |
1526 | return r; | |
1527 | } | |
1528 | } | |
7c673cae FG |
1529 | |
1530 | dout(10) << __func__ << " log file size was 0x" | |
1531 | << std::hex << log_file->fnode.size << std::dec << dendl; | |
11fdf7f2 TL |
1532 | if (unlikely(to_stdout)) { |
1533 | std::cout << " log file size was 0x" | |
1534 | << std::hex << log_file->fnode.size << std::dec << std::endl; | |
1535 | } | |
1536 | ||
7c673cae FG |
1537 | delete log_reader; |
1538 | ||
1539 | if (!noop) { | |
1540 | // verify file link counts are all >0 | |
1541 | for (auto& p : file_map) { | |
1542 | if (p.second->refs == 0 && | |
1543 | p.second->fnode.ino > 1) { | |
1544 | derr << __func__ << " file with link count 0: " << p.second->fnode | |
1545 | << dendl; | |
1546 | return -EIO; | |
1547 | } | |
1548 | } | |
1549 | } | |
1550 | ||
9f95a23c TL |
1551 | for (unsigned id = 0; id < MAX_BDEV; ++id) { |
1552 | dout(10) << __func__ << " block_unused_too_granular " << id << ": " | |
1553 | << block_unused_too_granular[id] << dendl; | |
1554 | } | |
7c673cae FG |
1555 | dout(10) << __func__ << " done" << dendl; |
1556 | return 0; | |
1557 | } | |
1558 | ||
11fdf7f2 TL |
1559 | int BlueFS::log_dump() |
1560 | { | |
1561 | // only dump log file's content | |
1562 | int r = _replay(true, true); | |
1563 | if (r < 0) { | |
1564 | derr << __func__ << " failed to replay log: " << cpp_strerror(r) << dendl; | |
1565 | return r; | |
1566 | } | |
1567 | ||
1568 | return 0; | |
1569 | } | |
1570 | ||
1571 | int BlueFS::device_migrate_to_existing( | |
1572 | CephContext *cct, | |
1573 | const set<int>& devs_source, | |
9f95a23c TL |
1574 | int dev_target, |
1575 | const bluefs_layout_t& layout) | |
11fdf7f2 TL |
1576 | { |
1577 | vector<byte> buf; | |
1578 | bool buffered = cct->_conf->bluefs_buffered_io; | |
1579 | ||
eafe8130 TL |
1580 | dout(10) << __func__ << " devs_source " << devs_source |
1581 | << " dev_target " << dev_target << dendl; | |
11fdf7f2 TL |
1582 | assert(dev_target < (int)MAX_BDEV); |
1583 | ||
1584 | int flags = 0; | |
1585 | flags |= devs_source.count(BDEV_DB) ? | |
1586 | (REMOVE_DB | RENAME_SLOW2DB) : 0; | |
1587 | flags |= devs_source.count(BDEV_WAL) ? REMOVE_WAL : 0; | |
1588 | int dev_target_new = dev_target; | |
1589 | ||
1590 | // Slow device without separate DB one is addressed via BDEV_DB | |
1591 | // Hence need renaming. | |
1592 | if ((flags & REMOVE_DB) && dev_target == BDEV_SLOW) { | |
1593 | dev_target_new = BDEV_DB; | |
1594 | dout(0) << __func__ << " super to be written to " << dev_target << dendl; | |
1595 | } | |
1596 | ||
9f95a23c | 1597 | for (auto& [ino, file_ref] : file_map) { |
11fdf7f2 | 1598 | //do not copy log |
9f95a23c | 1599 | if (file_ref->fnode.ino == 1) { |
11fdf7f2 TL |
1600 | continue; |
1601 | } | |
9f95a23c | 1602 | dout(10) << __func__ << " " << ino << " " << file_ref->fnode << dendl; |
eafe8130 | 1603 | |
9f95a23c | 1604 | auto& fnode_extents = file_ref->fnode.extents; |
11fdf7f2 | 1605 | |
9f95a23c TL |
1606 | bool rewrite = std::any_of( |
1607 | fnode_extents.begin(), | |
1608 | fnode_extents.end(), | |
1609 | [=](auto& ext) { | |
1610 | return ext.bdev != dev_target && devs_source.count(ext.bdev); | |
1611 | }); | |
eafe8130 TL |
1612 | if (rewrite) { |
1613 | dout(10) << __func__ << " migrating" << dendl; | |
1614 | ||
1615 | // read entire file | |
1616 | bufferlist bl; | |
1617 | for (auto old_ext : fnode_extents) { | |
1618 | buf.resize(old_ext.length); | |
1619 | int r = bdev[old_ext.bdev]->read_random( | |
1620 | old_ext.offset, | |
1621 | old_ext.length, | |
1622 | (char*)&buf.at(0), | |
1623 | buffered); | |
1624 | if (r != 0) { | |
1625 | derr << __func__ << " failed to read 0x" << std::hex | |
1626 | << old_ext.offset << "~" << old_ext.length << std::dec | |
1627 | << " from " << (int)dev_target << dendl; | |
1628 | return -EIO; | |
1629 | } | |
1630 | bl.append((char*)&buf[0], old_ext.length); | |
1631 | } | |
11fdf7f2 | 1632 | |
eafe8130 TL |
1633 | // write entire file |
1634 | PExtentVector extents; | |
1635 | auto l = _allocate_without_fallback(dev_target, bl.length(), &extents); | |
1636 | if (l < 0) { | |
1637 | derr << __func__ << " unable to allocate len 0x" << std::hex | |
1638 | << bl.length() << std::dec << " from " << (int)dev_target | |
1639 | << ": " << cpp_strerror(l) << dendl; | |
1640 | return -ENOSPC; | |
1641 | } | |
11fdf7f2 | 1642 | |
eafe8130 TL |
1643 | uint64_t off = 0; |
1644 | for (auto& i : extents) { | |
1645 | bufferlist cur; | |
1646 | uint64_t cur_len = std::min<uint64_t>(i.length, bl.length() - off); | |
1647 | ceph_assert(cur_len > 0); | |
1648 | cur.substr_of(bl, off, cur_len); | |
1649 | int r = bdev[dev_target]->write(i.offset, cur, buffered); | |
1650 | ceph_assert(r == 0); | |
1651 | off += cur_len; | |
1652 | } | |
1653 | ||
1654 | // release old extents | |
1655 | for (auto old_ext : fnode_extents) { | |
1656 | PExtentVector to_release; | |
1657 | to_release.emplace_back(old_ext.offset, old_ext.length); | |
1658 | alloc[old_ext.bdev]->release(to_release); | |
1659 | } | |
1660 | ||
1661 | // update fnode | |
1662 | fnode_extents.clear(); | |
1663 | for (auto& i : extents) { | |
1664 | fnode_extents.emplace_back(dev_target_new, i.offset, i.length); | |
1665 | } | |
1666 | } else { | |
9f95a23c TL |
1667 | for (auto& ext : fnode_extents) { |
1668 | if (dev_target != dev_target_new && ext.bdev == dev_target) { | |
eafe8130 | 1669 | dout(20) << __func__ << " " << " ... adjusting extent 0x" |
9f95a23c | 1670 | << std::hex << ext.offset << std::dec |
eafe8130 TL |
1671 | << " bdev " << dev_target << " -> " << dev_target_new |
1672 | << dendl; | |
9f95a23c | 1673 | ext.bdev = dev_target_new; |
11fdf7f2 | 1674 | } |
11fdf7f2 TL |
1675 | } |
1676 | } | |
11fdf7f2 TL |
1677 | } |
1678 | // new logging device in the current naming scheme | |
1679 | int new_log_dev_cur = bdev[BDEV_WAL] ? | |
1680 | BDEV_WAL : | |
1681 | bdev[BDEV_DB] ? BDEV_DB : BDEV_SLOW; | |
1682 | ||
1683 | // new logging device in new naming scheme | |
1684 | int new_log_dev_next = new_log_dev_cur; | |
1685 | ||
1686 | if (devs_source.count(new_log_dev_cur)) { | |
1687 | // SLOW device is addressed via BDEV_DB too hence either WAL or DB | |
1688 | new_log_dev_next = (flags & REMOVE_WAL) || !bdev[BDEV_WAL] ? | |
1689 | BDEV_DB : | |
1690 | BDEV_WAL; | |
1691 | ||
1692 | dout(0) << __func__ << " log moved from " << new_log_dev_cur | |
1693 | << " to " << new_log_dev_next << dendl; | |
1694 | ||
1695 | new_log_dev_cur = | |
1696 | (flags & REMOVE_DB) && new_log_dev_next == BDEV_DB ? | |
1697 | BDEV_SLOW : | |
1698 | new_log_dev_next; | |
1699 | } | |
1700 | ||
9f95a23c | 1701 | _rewrite_log_and_layout_sync( |
11fdf7f2 TL |
1702 | false, |
1703 | (flags & REMOVE_DB) ? BDEV_SLOW : BDEV_DB, | |
1704 | new_log_dev_cur, | |
1705 | new_log_dev_next, | |
9f95a23c TL |
1706 | flags, |
1707 | layout); | |
11fdf7f2 TL |
1708 | return 0; |
1709 | } | |
1710 | ||
1711 | int BlueFS::device_migrate_to_new( | |
1712 | CephContext *cct, | |
1713 | const set<int>& devs_source, | |
9f95a23c TL |
1714 | int dev_target, |
1715 | const bluefs_layout_t& layout) | |
11fdf7f2 TL |
1716 | { |
1717 | vector<byte> buf; | |
1718 | bool buffered = cct->_conf->bluefs_buffered_io; | |
1719 | ||
eafe8130 TL |
1720 | dout(10) << __func__ << " devs_source " << devs_source |
1721 | << " dev_target " << dev_target << dendl; | |
11fdf7f2 TL |
1722 | assert(dev_target == (int)BDEV_NEWDB || (int)BDEV_NEWWAL); |
1723 | ||
1724 | int flags = 0; | |
1725 | ||
1726 | flags |= devs_source.count(BDEV_DB) ? | |
1727 | (!bdev[BDEV_SLOW] ? RENAME_DB2SLOW: REMOVE_DB) : | |
1728 | 0; | |
1729 | flags |= devs_source.count(BDEV_WAL) ? REMOVE_WAL : 0; | |
9f95a23c | 1730 | int dev_target_new = dev_target; //FIXME: remove, makes no sense |
11fdf7f2 TL |
1731 | |
1732 | for (auto& p : file_map) { | |
1733 | //do not copy log | |
1734 | if (p.second->fnode.ino == 1) { | |
1735 | continue; | |
1736 | } | |
eafe8130 TL |
1737 | dout(10) << __func__ << " " << p.first << " " << p.second->fnode << dendl; |
1738 | ||
11fdf7f2 TL |
1739 | auto& fnode_extents = p.second->fnode.extents; |
1740 | ||
eafe8130 | 1741 | bool rewrite = false; |
11fdf7f2 | 1742 | for (auto ext_it = fnode_extents.begin(); |
eafe8130 TL |
1743 | ext_it != p.second->fnode.extents.end(); |
1744 | ++ext_it) { | |
11fdf7f2 | 1745 | if (ext_it->bdev != dev_target && devs_source.count(ext_it->bdev)) { |
eafe8130 TL |
1746 | rewrite = true; |
1747 | break; | |
1748 | } | |
1749 | } | |
1750 | if (rewrite) { | |
1751 | dout(10) << __func__ << " migrating" << dendl; | |
1752 | ||
1753 | // read entire file | |
1754 | bufferlist bl; | |
1755 | for (auto old_ext : fnode_extents) { | |
1756 | buf.resize(old_ext.length); | |
1757 | int r = bdev[old_ext.bdev]->read_random( | |
1758 | old_ext.offset, | |
1759 | old_ext.length, | |
1760 | (char*)&buf.at(0), | |
1761 | buffered); | |
1762 | if (r != 0) { | |
1763 | derr << __func__ << " failed to read 0x" << std::hex | |
1764 | << old_ext.offset << "~" << old_ext.length << std::dec | |
1765 | << " from " << (int)dev_target << dendl; | |
1766 | return -EIO; | |
11fdf7f2 | 1767 | } |
eafe8130 TL |
1768 | bl.append((char*)&buf[0], old_ext.length); |
1769 | } | |
1770 | ||
1771 | // write entire file | |
1772 | PExtentVector extents; | |
1773 | auto l = _allocate_without_fallback(dev_target, bl.length(), &extents); | |
1774 | if (l < 0) { | |
1775 | derr << __func__ << " unable to allocate len 0x" << std::hex | |
1776 | << bl.length() << std::dec << " from " << (int)dev_target | |
1777 | << ": " << cpp_strerror(l) << dendl; | |
1778 | return -ENOSPC; | |
1779 | } | |
1780 | ||
1781 | uint64_t off = 0; | |
1782 | for (auto& i : extents) { | |
1783 | bufferlist cur; | |
1784 | uint64_t cur_len = std::min<uint64_t>(i.length, bl.length() - off); | |
1785 | ceph_assert(cur_len > 0); | |
1786 | cur.substr_of(bl, off, cur_len); | |
1787 | int r = bdev[dev_target]->write(i.offset, cur, buffered); | |
1788 | ceph_assert(r == 0); | |
1789 | off += cur_len; | |
1790 | } | |
1791 | ||
1792 | // release old extents | |
1793 | for (auto old_ext : fnode_extents) { | |
1794 | PExtentVector to_release; | |
1795 | to_release.emplace_back(old_ext.offset, old_ext.length); | |
1796 | alloc[old_ext.bdev]->release(to_release); | |
1797 | } | |
1798 | ||
1799 | // update fnode | |
1800 | fnode_extents.clear(); | |
1801 | for (auto& i : extents) { | |
1802 | fnode_extents.emplace_back(dev_target_new, i.offset, i.length); | |
11fdf7f2 TL |
1803 | } |
1804 | } | |
11fdf7f2 TL |
1805 | } |
1806 | // new logging device in the current naming scheme | |
1807 | int new_log_dev_cur = | |
1808 | bdev[BDEV_NEWWAL] ? | |
1809 | BDEV_NEWWAL : | |
1810 | bdev[BDEV_WAL] && !(flags & REMOVE_WAL) ? | |
1811 | BDEV_WAL : | |
1812 | bdev[BDEV_NEWDB] ? | |
1813 | BDEV_NEWDB : | |
1814 | bdev[BDEV_DB] && !(flags & REMOVE_DB)? | |
1815 | BDEV_DB : | |
1816 | BDEV_SLOW; | |
1817 | ||
1818 | // new logging device in new naming scheme | |
1819 | int new_log_dev_next = | |
1820 | new_log_dev_cur == BDEV_NEWWAL ? | |
1821 | BDEV_WAL : | |
1822 | new_log_dev_cur == BDEV_NEWDB ? | |
1823 | BDEV_DB : | |
1824 | new_log_dev_cur; | |
1825 | ||
1826 | int super_dev = | |
1827 | dev_target == BDEV_NEWDB ? | |
1828 | BDEV_NEWDB : | |
1829 | bdev[BDEV_DB] ? | |
1830 | BDEV_DB : | |
1831 | BDEV_SLOW; | |
1832 | ||
9f95a23c | 1833 | _rewrite_log_and_layout_sync( |
11fdf7f2 TL |
1834 | false, |
1835 | super_dev, | |
1836 | new_log_dev_cur, | |
1837 | new_log_dev_next, | |
9f95a23c TL |
1838 | flags, |
1839 | layout); | |
11fdf7f2 TL |
1840 | return 0; |
1841 | } | |
1842 | ||
7c673cae FG |
1843 | BlueFS::FileRef BlueFS::_get_file(uint64_t ino) |
1844 | { | |
1845 | auto p = file_map.find(ino); | |
1846 | if (p == file_map.end()) { | |
9f95a23c | 1847 | FileRef f = ceph::make_ref<File>(); |
7c673cae FG |
1848 | file_map[ino] = f; |
1849 | dout(30) << __func__ << " ino " << ino << " = " << f | |
1850 | << " (new)" << dendl; | |
1851 | return f; | |
1852 | } else { | |
1853 | dout(30) << __func__ << " ino " << ino << " = " << p->second << dendl; | |
1854 | return p->second; | |
1855 | } | |
1856 | } | |
1857 | ||
1858 | void BlueFS::_drop_link(FileRef file) | |
1859 | { | |
1860 | dout(20) << __func__ << " had refs " << file->refs | |
1861 | << " on " << file->fnode << dendl; | |
11fdf7f2 | 1862 | ceph_assert(file->refs > 0); |
7c673cae FG |
1863 | --file->refs; |
1864 | if (file->refs == 0) { | |
1865 | dout(20) << __func__ << " destroying " << file->fnode << dendl; | |
11fdf7f2 | 1866 | ceph_assert(file->num_reading.load() == 0); |
9f95a23c | 1867 | vselector->sub_usage(file->vselector_hint, file->fnode); |
7c673cae FG |
1868 | log_t.op_file_remove(file->fnode.ino); |
1869 | for (auto& r : file->fnode.extents) { | |
1870 | pending_release[r.bdev].insert(r.offset, r.length); | |
1871 | } | |
1872 | file_map.erase(file->fnode.ino); | |
1873 | file->deleted = true; | |
94b18763 | 1874 | |
7c673cae | 1875 | if (file->dirty_seq) { |
11fdf7f2 TL |
1876 | ceph_assert(file->dirty_seq > log_seq_stable); |
1877 | ceph_assert(dirty_files.count(file->dirty_seq)); | |
7c673cae FG |
1878 | auto it = dirty_files[file->dirty_seq].iterator_to(*file); |
1879 | dirty_files[file->dirty_seq].erase(it); | |
1880 | file->dirty_seq = 0; | |
1881 | } | |
1882 | } | |
1883 | } | |
1884 | ||
1885 | int BlueFS::_read_random( | |
1886 | FileReader *h, ///< [in] read from here | |
1887 | uint64_t off, ///< [in] offset | |
9f95a23c | 1888 | uint64_t len, ///< [in] this many bytes |
7c673cae FG |
1889 | char *out) ///< [out] optional: or copy it here |
1890 | { | |
494da23a TL |
1891 | auto* buf = &h->buf; |
1892 | ||
1893 | int ret = 0; | |
7c673cae FG |
1894 | dout(10) << __func__ << " h " << h |
1895 | << " 0x" << std::hex << off << "~" << len << std::dec | |
1896 | << " from " << h->file->fnode << dendl; | |
1897 | ||
1898 | ++h->file->num_reading; | |
1899 | ||
1900 | if (!h->ignore_eof && | |
1901 | off + len > h->file->fnode.size) { | |
1902 | if (off > h->file->fnode.size) | |
1903 | len = 0; | |
1904 | else | |
1905 | len = h->file->fnode.size - off; | |
1906 | dout(20) << __func__ << " reaching (or past) eof, len clipped to 0x" | |
1907 | << std::hex << len << std::dec << dendl; | |
1908 | } | |
494da23a TL |
1909 | logger->inc(l_bluefs_read_random_count, 1); |
1910 | logger->inc(l_bluefs_read_random_bytes, len); | |
7c673cae | 1911 | |
494da23a | 1912 | std::shared_lock s_lock(h->lock); |
f91f0fd5 | 1913 | buf->bl.reassign_to_mempool(mempool::mempool_bluefs_file_reader); |
7c673cae | 1914 | while (len > 0) { |
494da23a TL |
1915 | if (off < buf->bl_off || off >= buf->get_buf_end()) { |
1916 | s_lock.unlock(); | |
1917 | uint64_t x_off = 0; | |
1918 | auto p = h->file->fnode.seek(off, &x_off); | |
f6b5b4d7 | 1919 | ceph_assert(p != h->file->fnode.extents.end()); |
9f95a23c | 1920 | uint64_t l = std::min(p->length - x_off, len); |
494da23a TL |
1921 | dout(20) << __func__ << " read random 0x" |
1922 | << std::hex << x_off << "~" << l << std::dec | |
1923 | << " of " << *p << dendl; | |
1924 | int r = bdev[p->bdev]->read_random(p->offset + x_off, l, out, | |
1925 | cct->_conf->bluefs_buffered_io); | |
1926 | ceph_assert(r == 0); | |
1927 | off += l; | |
1928 | len -= l; | |
1929 | ret += l; | |
1930 | out += l; | |
1931 | ||
1932 | logger->inc(l_bluefs_read_random_disk_count, 1); | |
1933 | logger->inc(l_bluefs_read_random_disk_bytes, l); | |
1934 | if (len > 0) { | |
1935 | s_lock.lock(); | |
1936 | } | |
1937 | } else { | |
1938 | auto left = buf->get_buf_remaining(off); | |
1939 | int r = std::min(len, left); | |
1940 | logger->inc(l_bluefs_read_random_buffer_count, 1); | |
1941 | logger->inc(l_bluefs_read_random_buffer_bytes, r); | |
1942 | dout(20) << __func__ << " left 0x" << std::hex << left | |
1943 | << " 0x" << off << "~" << len << std::dec | |
1944 | << dendl; | |
1945 | ||
1946 | if (out) { | |
1947 | // NOTE: h->bl is normally a contiguous buffer so c_str() is free. | |
1948 | memcpy(out, buf->bl.c_str() + off - buf->bl_off, r); | |
1949 | out += r; | |
1950 | } | |
7c673cae | 1951 | |
494da23a TL |
1952 | dout(30) << __func__ << " result chunk (0x" |
1953 | << std::hex << r << std::dec << " bytes):\n"; | |
1954 | bufferlist t; | |
1955 | t.substr_of(buf->bl, off - buf->bl_off, r); | |
1956 | t.hexdump(*_dout); | |
1957 | *_dout << dendl; | |
1958 | ||
1959 | off += r; | |
1960 | len -= r; | |
1961 | ret += r; | |
1962 | buf->pos += r; | |
1963 | } | |
1964 | } | |
7c673cae FG |
1965 | dout(20) << __func__ << " got " << ret << dendl; |
1966 | --h->file->num_reading; | |
1967 | return ret; | |
1968 | } | |
1969 | ||
1970 | int BlueFS::_read( | |
1971 | FileReader *h, ///< [in] read from here | |
1972 | FileReaderBuffer *buf, ///< [in] reader state | |
1973 | uint64_t off, ///< [in] offset | |
1974 | size_t len, ///< [in] this many bytes | |
1975 | bufferlist *outbl, ///< [out] optional: reference the result here | |
1976 | char *out) ///< [out] optional: or copy it here | |
1977 | { | |
494da23a | 1978 | bool prefetch = !outbl && !out; |
7c673cae FG |
1979 | dout(10) << __func__ << " h " << h |
1980 | << " 0x" << std::hex << off << "~" << len << std::dec | |
494da23a TL |
1981 | << " from " << h->file->fnode |
1982 | << (prefetch ? " prefetch" : "") | |
1983 | << dendl; | |
7c673cae FG |
1984 | |
1985 | ++h->file->num_reading; | |
1986 | ||
1987 | if (!h->ignore_eof && | |
1988 | off + len > h->file->fnode.size) { | |
1989 | if (off > h->file->fnode.size) | |
1990 | len = 0; | |
1991 | else | |
1992 | len = h->file->fnode.size - off; | |
1993 | dout(20) << __func__ << " reaching (or past) eof, len clipped to 0x" | |
1994 | << std::hex << len << std::dec << dendl; | |
1995 | } | |
494da23a TL |
1996 | logger->inc(l_bluefs_read_count, 1); |
1997 | logger->inc(l_bluefs_read_bytes, len); | |
1998 | if (prefetch) { | |
1999 | logger->inc(l_bluefs_read_prefetch_count, 1); | |
2000 | logger->inc(l_bluefs_read_prefetch_bytes, len); | |
2001 | } | |
2002 | ||
7c673cae FG |
2003 | if (outbl) |
2004 | outbl->clear(); | |
2005 | ||
2006 | int ret = 0; | |
494da23a | 2007 | std::shared_lock s_lock(h->lock); |
7c673cae FG |
2008 | while (len > 0) { |
2009 | size_t left; | |
2010 | if (off < buf->bl_off || off >= buf->get_buf_end()) { | |
494da23a TL |
2011 | s_lock.unlock(); |
2012 | std::unique_lock u_lock(h->lock); | |
f91f0fd5 | 2013 | buf->bl.reassign_to_mempool(mempool::mempool_bluefs_file_reader); |
494da23a TL |
2014 | if (off < buf->bl_off || off >= buf->get_buf_end()) { |
2015 | // if precondition hasn't changed during locking upgrade. | |
2016 | buf->bl.clear(); | |
2017 | buf->bl_off = off & super.block_mask(); | |
2018 | uint64_t x_off = 0; | |
2019 | auto p = h->file->fnode.seek(buf->bl_off, &x_off); | |
f6b5b4d7 TL |
2020 | if (p == h->file->fnode.extents.end()) { |
2021 | dout(5) << __func__ << " reading less then required " | |
2022 | << ret << "<" << ret + len << dendl; | |
2023 | break; | |
2024 | } | |
2025 | ||
494da23a TL |
2026 | uint64_t want = round_up_to(len + (off & ~super.block_mask()), |
2027 | super.block_size); | |
2028 | want = std::max(want, buf->max_prefetch); | |
2029 | uint64_t l = std::min(p->length - x_off, want); | |
2030 | uint64_t eof_offset = round_up_to(h->file->fnode.size, super.block_size); | |
2031 | if (!h->ignore_eof && | |
2032 | buf->bl_off + l > eof_offset) { | |
2033 | l = eof_offset - buf->bl_off; | |
2034 | } | |
2035 | dout(20) << __func__ << " fetching 0x" | |
2036 | << std::hex << x_off << "~" << l << std::dec | |
2037 | << " of " << *p << dendl; | |
2038 | int r = bdev[p->bdev]->read(p->offset + x_off, l, &buf->bl, ioc[p->bdev], | |
2039 | cct->_conf->bluefs_buffered_io); | |
2040 | ceph_assert(r == 0); | |
7c673cae | 2041 | } |
494da23a TL |
2042 | u_lock.unlock(); |
2043 | s_lock.lock(); | |
2044 | // we should recheck if buffer is valid after lock downgrade | |
2045 | continue; | |
7c673cae FG |
2046 | } |
2047 | left = buf->get_buf_remaining(off); | |
2048 | dout(20) << __func__ << " left 0x" << std::hex << left | |
2049 | << " len 0x" << len << std::dec << dendl; | |
2050 | ||
11fdf7f2 | 2051 | int r = std::min(len, left); |
7c673cae FG |
2052 | if (outbl) { |
2053 | bufferlist t; | |
2054 | t.substr_of(buf->bl, off - buf->bl_off, r); | |
2055 | outbl->claim_append(t); | |
2056 | } | |
2057 | if (out) { | |
2058 | // NOTE: h->bl is normally a contiguous buffer so c_str() is free. | |
2059 | memcpy(out, buf->bl.c_str() + off - buf->bl_off, r); | |
2060 | out += r; | |
2061 | } | |
2062 | ||
2063 | dout(30) << __func__ << " result chunk (0x" | |
2064 | << std::hex << r << std::dec << " bytes):\n"; | |
2065 | bufferlist t; | |
2066 | t.substr_of(buf->bl, off - buf->bl_off, r); | |
2067 | t.hexdump(*_dout); | |
2068 | *_dout << dendl; | |
2069 | ||
2070 | off += r; | |
2071 | len -= r; | |
2072 | ret += r; | |
2073 | buf->pos += r; | |
2074 | } | |
2075 | ||
2076 | dout(20) << __func__ << " got " << ret << dendl; | |
11fdf7f2 | 2077 | ceph_assert(!outbl || (int)outbl->length() == ret); |
7c673cae FG |
2078 | --h->file->num_reading; |
2079 | return ret; | |
2080 | } | |
2081 | ||
2082 | void BlueFS::_invalidate_cache(FileRef f, uint64_t offset, uint64_t length) | |
2083 | { | |
2084 | dout(10) << __func__ << " file " << f->fnode | |
2085 | << " 0x" << std::hex << offset << "~" << length << std::dec | |
2086 | << dendl; | |
2087 | if (offset & ~super.block_mask()) { | |
2088 | offset &= super.block_mask(); | |
11fdf7f2 | 2089 | length = round_up_to(length, super.block_size); |
7c673cae FG |
2090 | } |
2091 | uint64_t x_off = 0; | |
2092 | auto p = f->fnode.seek(offset, &x_off); | |
2093 | while (length > 0 && p != f->fnode.extents.end()) { | |
11fdf7f2 | 2094 | uint64_t x_len = std::min(p->length - x_off, length); |
7c673cae FG |
2095 | bdev[p->bdev]->invalidate_cache(p->offset + x_off, x_len); |
2096 | dout(20) << __func__ << " 0x" << std::hex << x_off << "~" << x_len | |
2097 | << std:: dec << " of " << *p << dendl; | |
2098 | offset += x_len; | |
2099 | length -= x_len; | |
2100 | } | |
2101 | } | |
2102 | ||
2103 | uint64_t BlueFS::_estimate_log_size() | |
2104 | { | |
2105 | int avg_dir_size = 40; // fixme | |
2106 | int avg_file_size = 12; | |
2107 | uint64_t size = 4096 * 2; | |
2108 | size += file_map.size() * (1 + sizeof(bluefs_fnode_t)); | |
2109 | for (auto& p : block_all) | |
2110 | size += p.num_intervals() * (1 + 1 + sizeof(uint64_t) * 2); | |
2111 | size += dir_map.size() + (1 + avg_dir_size); | |
2112 | size += file_map.size() * (1 + avg_dir_size + avg_file_size); | |
11fdf7f2 | 2113 | return round_up_to(size, super.block_size); |
7c673cae FG |
2114 | } |
2115 | ||
2116 | void BlueFS::compact_log() | |
2117 | { | |
f6b5b4d7 TL |
2118 | std::unique_lock<ceph::mutex> l(lock); |
2119 | if (!cct->_conf->bluefs_replay_recovery_disable_compact) { | |
2120 | if (cct->_conf->bluefs_compact_log_sync) { | |
2121 | _compact_log_sync(); | |
2122 | } else { | |
2123 | _compact_log_async(l); | |
2124 | } | |
7c673cae FG |
2125 | } |
2126 | } | |
2127 | ||
2128 | bool BlueFS::_should_compact_log() | |
2129 | { | |
2130 | uint64_t current = log_writer->file->fnode.size; | |
2131 | uint64_t expected = _estimate_log_size(); | |
2132 | float ratio = (float)current / (float)expected; | |
2133 | dout(10) << __func__ << " current 0x" << std::hex << current | |
2134 | << " expected " << expected << std::dec | |
2135 | << " ratio " << ratio | |
2136 | << (new_log ? " (async compaction in progress)" : "") | |
2137 | << dendl; | |
2138 | if (new_log || | |
2139 | current < cct->_conf->bluefs_log_compact_min_size || | |
2140 | ratio < cct->_conf->bluefs_log_compact_min_ratio) { | |
2141 | return false; | |
2142 | } | |
2143 | return true; | |
2144 | } | |
2145 | ||
11fdf7f2 TL |
2146 | void BlueFS::_compact_log_dump_metadata(bluefs_transaction_t *t, |
2147 | int flags) | |
7c673cae FG |
2148 | { |
2149 | t->seq = 1; | |
2150 | t->uuid = super.uuid; | |
2151 | dout(20) << __func__ << " op_init" << dendl; | |
2152 | ||
2153 | t->op_init(); | |
2154 | for (unsigned bdev = 0; bdev < MAX_BDEV; ++bdev) { | |
2155 | interval_set<uint64_t>& p = block_all[bdev]; | |
2156 | for (interval_set<uint64_t>::iterator q = p.begin(); q != p.end(); ++q) { | |
11fdf7f2 TL |
2157 | auto bdev_new = bdev; |
2158 | if ((flags & REMOVE_WAL) && bdev == BDEV_WAL) { | |
2159 | continue; | |
2160 | } | |
2161 | if ((flags & REMOVE_DB) && bdev == BDEV_DB) { | |
2162 | continue; | |
2163 | } | |
2164 | if ((flags & RENAME_SLOW2DB) && bdev == BDEV_SLOW) { | |
2165 | bdev_new = BDEV_DB; | |
2166 | } | |
2167 | if ((flags & RENAME_DB2SLOW) && bdev == BDEV_DB) { | |
2168 | bdev_new = BDEV_SLOW; | |
2169 | } | |
2170 | if (bdev == BDEV_NEWDB) { | |
2171 | // REMOVE_DB xor RENAME_DB | |
2172 | ceph_assert(!(flags & REMOVE_DB) != !(flags & RENAME_DB2SLOW)); | |
2173 | ceph_assert(!(flags & RENAME_SLOW2DB)); | |
2174 | bdev_new = BDEV_DB; | |
2175 | } | |
2176 | if (bdev == BDEV_NEWWAL) { | |
2177 | ceph_assert(flags & REMOVE_WAL); | |
2178 | bdev_new = BDEV_WAL; | |
2179 | } | |
2180 | dout(20) << __func__ << " op_alloc_add " << bdev_new << " 0x" | |
7c673cae FG |
2181 | << std::hex << q.get_start() << "~" << q.get_len() << std::dec |
2182 | << dendl; | |
11fdf7f2 | 2183 | t->op_alloc_add(bdev_new, q.get_start(), q.get_len()); |
7c673cae FG |
2184 | } |
2185 | } | |
9f95a23c TL |
2186 | for (auto& [ino, file_ref] : file_map) { |
2187 | if (ino == 1) | |
7c673cae | 2188 | continue; |
9f95a23c | 2189 | ceph_assert(ino > 1); |
11fdf7f2 | 2190 | |
9f95a23c | 2191 | for(auto& e : file_ref->fnode.extents) { |
11fdf7f2 TL |
2192 | auto bdev = e.bdev; |
2193 | auto bdev_new = bdev; | |
2194 | ceph_assert(!((flags & REMOVE_WAL) && bdev == BDEV_WAL)); | |
2195 | if ((flags & RENAME_SLOW2DB) && bdev == BDEV_SLOW) { | |
2196 | bdev_new = BDEV_DB; | |
2197 | } | |
2198 | if ((flags & RENAME_DB2SLOW) && bdev == BDEV_DB) { | |
2199 | bdev_new = BDEV_SLOW; | |
2200 | } | |
2201 | if (bdev == BDEV_NEWDB) { | |
2202 | // REMOVE_DB xor RENAME_DB | |
2203 | ceph_assert(!(flags & REMOVE_DB) != !(flags & RENAME_DB2SLOW)); | |
2204 | ceph_assert(!(flags & RENAME_SLOW2DB)); | |
2205 | bdev_new = BDEV_DB; | |
2206 | } | |
2207 | if (bdev == BDEV_NEWWAL) { | |
2208 | ceph_assert(flags & REMOVE_WAL); | |
2209 | bdev_new = BDEV_WAL; | |
2210 | } | |
2211 | e.bdev = bdev_new; | |
2212 | } | |
9f95a23c TL |
2213 | dout(20) << __func__ << " op_file_update " << file_ref->fnode << dendl; |
2214 | t->op_file_update(file_ref->fnode); | |
7c673cae | 2215 | } |
9f95a23c TL |
2216 | for (auto& [path, dir_ref] : dir_map) { |
2217 | dout(20) << __func__ << " op_dir_create " << path << dendl; | |
2218 | t->op_dir_create(path); | |
2219 | for (auto& [fname, file_ref] : dir_ref->file_map) { | |
2220 | dout(20) << __func__ << " op_dir_link " << path << "/" << fname | |
2221 | << " to " << file_ref->fnode.ino << dendl; | |
2222 | t->op_dir_link(path, fname, file_ref->fnode.ino); | |
7c673cae FG |
2223 | } |
2224 | } | |
2225 | } | |
2226 | ||
2227 | void BlueFS::_compact_log_sync() | |
2228 | { | |
2229 | dout(10) << __func__ << dendl; | |
9f95a23c TL |
2230 | auto prefer_bdev = |
2231 | vselector->select_prefer_bdev(log_writer->file->vselector_hint); | |
2232 | _rewrite_log_and_layout_sync(true, | |
11fdf7f2 | 2233 | BDEV_DB, |
9f95a23c TL |
2234 | prefer_bdev, |
2235 | prefer_bdev, | |
2236 | 0, | |
2237 | super.memorized_layout); | |
11fdf7f2 TL |
2238 | logger->inc(l_bluefs_log_compactions); |
2239 | } | |
2240 | ||
9f95a23c TL |
2241 | void BlueFS::_rewrite_log_and_layout_sync(bool allocate_with_fallback, |
2242 | int super_dev, | |
2243 | int log_dev, | |
2244 | int log_dev_new, | |
2245 | int flags, | |
2246 | std::optional<bluefs_layout_t> layout) | |
11fdf7f2 | 2247 | { |
7c673cae FG |
2248 | File *log_file = log_writer->file.get(); |
2249 | ||
2250 | // clear out log (be careful who calls us!!!) | |
2251 | log_t.clear(); | |
2252 | ||
11fdf7f2 TL |
2253 | dout(20) << __func__ << " super_dev:" << super_dev |
2254 | << " log_dev:" << log_dev | |
2255 | << " log_dev_new:" << log_dev_new | |
2256 | << " flags:" << flags | |
2257 | << dendl; | |
7c673cae | 2258 | bluefs_transaction_t t; |
11fdf7f2 | 2259 | _compact_log_dump_metadata(&t, flags); |
7c673cae FG |
2260 | |
2261 | dout(20) << __func__ << " op_jump_seq " << log_seq << dendl; | |
2262 | t.op_jump_seq(log_seq); | |
2263 | ||
2264 | bufferlist bl; | |
11fdf7f2 | 2265 | encode(t, bl); |
7c673cae FG |
2266 | _pad_bl(bl); |
2267 | ||
2268 | uint64_t need = bl.length() + cct->_conf->bluefs_max_log_runway; | |
2269 | dout(20) << __func__ << " need " << need << dendl; | |
2270 | ||
494da23a | 2271 | bluefs_fnode_t old_fnode; |
11fdf7f2 | 2272 | int r; |
494da23a | 2273 | log_file->fnode.swap_extents(old_fnode); |
11fdf7f2 TL |
2274 | if (allocate_with_fallback) { |
2275 | r = _allocate(log_dev, need, &log_file->fnode); | |
2276 | ceph_assert(r == 0); | |
2277 | } else { | |
2278 | PExtentVector extents; | |
2279 | r = _allocate_without_fallback(log_dev, | |
2280 | need, | |
2281 | &extents); | |
2282 | ceph_assert(r == 0); | |
2283 | for (auto& p : extents) { | |
2284 | log_file->fnode.append_extent( | |
2285 | bluefs_extent_t(log_dev, p.offset, p.length)); | |
2286 | } | |
7c673cae FG |
2287 | } |
2288 | ||
2289 | _close_writer(log_writer); | |
2290 | ||
2291 | log_file->fnode.size = bl.length(); | |
9f95a23c TL |
2292 | vselector->sub_usage(log_file->vselector_hint, old_fnode); |
2293 | vselector->add_usage(log_file->vselector_hint, log_file->fnode); | |
2294 | ||
7c673cae FG |
2295 | log_writer = _create_writer(log_file); |
2296 | log_writer->append(bl); | |
11fdf7f2 TL |
2297 | r = _flush(log_writer, true); |
2298 | ceph_assert(r == 0); | |
2299 | #ifdef HAVE_LIBAIO | |
2300 | if (!cct->_conf->bluefs_sync_write) { | |
2301 | list<aio_t> completed_ios; | |
2302 | _claim_completed_aios(log_writer, &completed_ios); | |
2303 | wait_for_aio(log_writer); | |
2304 | completed_ios.clear(); | |
2305 | } | |
2306 | #endif | |
224ce89b | 2307 | flush_bdev(); |
224ce89b | 2308 | |
9f95a23c | 2309 | super.memorized_layout = layout; |
7c673cae | 2310 | super.log_fnode = log_file->fnode; |
11fdf7f2 TL |
2311 | // rename device if needed |
2312 | if (log_dev != log_dev_new) { | |
2313 | dout(10) << __func__ << " renaming log extents to " << log_dev_new << dendl; | |
2314 | for (auto& p : super.log_fnode.extents) { | |
2315 | p.bdev = log_dev_new; | |
2316 | } | |
2317 | } | |
2318 | dout(10) << __func__ << " writing super, log fnode: " << super.log_fnode << dendl; | |
2319 | ||
7c673cae | 2320 | ++super.version; |
11fdf7f2 | 2321 | _write_super(super_dev); |
7c673cae FG |
2322 | flush_bdev(); |
2323 | ||
494da23a TL |
2324 | dout(10) << __func__ << " release old log extents " << old_fnode.extents << dendl; |
2325 | for (auto& r : old_fnode.extents) { | |
7c673cae FG |
2326 | pending_release[r.bdev].insert(r.offset, r.length); |
2327 | } | |
7c673cae FG |
2328 | } |
2329 | ||
2330 | /* | |
2331 | * 1. Allocate a new extent to continue the log, and then log an event | |
2332 | * that jumps the log write position to the new extent. At this point, the | |
2333 | * old extent(s) won't be written to, and reflect everything to compact. | |
2334 | * New events will be written to the new region that we'll keep. | |
2335 | * | |
2336 | * 2. While still holding the lock, encode a bufferlist that dumps all of the | |
2337 | * in-memory fnodes and names. This will become the new beginning of the | |
2338 | * log. The last event will jump to the log continuation extent from #1. | |
2339 | * | |
2340 | * 3. Queue a write to a new extent for the new beginnging of the log. | |
2341 | * | |
2342 | * 4. Drop lock and wait | |
2343 | * | |
2344 | * 5. Retake the lock. | |
2345 | * | |
2346 | * 6. Update the log_fnode to splice in the new beginning. | |
2347 | * | |
2348 | * 7. Write the new superblock. | |
2349 | * | |
2350 | * 8. Release the old log space. Clean up. | |
2351 | */ | |
11fdf7f2 | 2352 | void BlueFS::_compact_log_async(std::unique_lock<ceph::mutex>& l) |
7c673cae FG |
2353 | { |
2354 | dout(10) << __func__ << dendl; | |
2355 | File *log_file = log_writer->file.get(); | |
11fdf7f2 TL |
2356 | ceph_assert(!new_log); |
2357 | ceph_assert(!new_log_writer); | |
7c673cae | 2358 | |
181888fb FG |
2359 | // create a new log [writer] so that we know compaction is in progress |
2360 | // (see _should_compact_log) | |
9f95a23c | 2361 | new_log = ceph::make_ref<File>(); |
181888fb FG |
2362 | new_log->fnode.ino = 0; // so that _flush_range won't try to log the fnode |
2363 | ||
3efd9988 FG |
2364 | // 0. wait for any racing flushes to complete. (We do not want to block |
2365 | // in _flush_sync_log with jump_to set or else a racing thread might flush | |
2366 | // our entries and our jump_to update won't be correct.) | |
2367 | while (log_flushing) { | |
2368 | dout(10) << __func__ << " log is currently flushing, waiting" << dendl; | |
2369 | log_cond.wait(l); | |
2370 | } | |
2371 | ||
9f95a23c TL |
2372 | vselector->sub_usage(log_file->vselector_hint, log_file->fnode); |
2373 | ||
7c673cae FG |
2374 | // 1. allocate new log space and jump to it. |
2375 | old_log_jump_to = log_file->fnode.get_allocated(); | |
7c673cae | 2376 | dout(10) << __func__ << " old_log_jump_to 0x" << std::hex << old_log_jump_to |
11fdf7f2 | 2377 | << " need 0x" << (old_log_jump_to + cct->_conf->bluefs_max_log_runway) << std::dec << dendl; |
9f95a23c TL |
2378 | int r = _allocate(vselector->select_prefer_bdev(log_file->vselector_hint), |
2379 | cct->_conf->bluefs_max_log_runway, | |
2380 | &log_file->fnode); | |
11fdf7f2 | 2381 | ceph_assert(r == 0); |
9f95a23c TL |
2382 | //adjust usage as flush below will need it |
2383 | vselector->add_usage(log_file->vselector_hint, log_file->fnode); | |
7c673cae FG |
2384 | dout(10) << __func__ << " log extents " << log_file->fnode.extents << dendl; |
2385 | ||
2386 | // update the log file change and log a jump to the offset where we want to | |
2387 | // write the new entries | |
2388 | log_t.op_file_update(log_file->fnode); | |
2389 | log_t.op_jump(log_seq, old_log_jump_to); | |
2390 | ||
2391 | flush_bdev(); // FIXME? | |
2392 | ||
2393 | _flush_and_sync_log(l, 0, old_log_jump_to); | |
2394 | ||
2395 | // 2. prepare compacted log | |
2396 | bluefs_transaction_t t; | |
224ce89b WB |
2397 | //avoid record two times in log_t and _compact_log_dump_metadata. |
2398 | log_t.clear(); | |
11fdf7f2 | 2399 | _compact_log_dump_metadata(&t, 0); |
7c673cae | 2400 | |
eafe8130 TL |
2401 | uint64_t max_alloc_size = std::max(alloc_size[BDEV_WAL], |
2402 | std::max(alloc_size[BDEV_DB], | |
2403 | alloc_size[BDEV_SLOW])); | |
2404 | ||
7c673cae | 2405 | // conservative estimate for final encoded size |
11fdf7f2 | 2406 | new_log_jump_to = round_up_to(t.op_bl.length() + super.block_size * 2, |
eafe8130 | 2407 | max_alloc_size); |
7c673cae FG |
2408 | t.op_jump(log_seq, new_log_jump_to); |
2409 | ||
11fdf7f2 | 2410 | // allocate |
9f95a23c | 2411 | //FIXME: check if we want DB here? |
11fdf7f2 TL |
2412 | r = _allocate(BlueFS::BDEV_DB, new_log_jump_to, |
2413 | &new_log->fnode); | |
2414 | ceph_assert(r == 0); | |
2415 | ||
2416 | // we might have some more ops in log_t due to _allocate call | |
2417 | t.claim_ops(log_t); | |
2418 | ||
7c673cae | 2419 | bufferlist bl; |
11fdf7f2 | 2420 | encode(t, bl); |
7c673cae FG |
2421 | _pad_bl(bl); |
2422 | ||
2423 | dout(10) << __func__ << " new_log_jump_to 0x" << std::hex << new_log_jump_to | |
2424 | << std::dec << dendl; | |
2425 | ||
7c673cae FG |
2426 | new_log_writer = _create_writer(new_log); |
2427 | new_log_writer->append(bl); | |
2428 | ||
2429 | // 3. flush | |
2430 | r = _flush(new_log_writer, true); | |
11fdf7f2 | 2431 | ceph_assert(r == 0); |
7c673cae FG |
2432 | |
2433 | // 4. wait | |
11fdf7f2 | 2434 | _flush_bdev_safely(new_log_writer); |
7c673cae | 2435 | |
11fdf7f2 | 2436 | // 5. update our log fnode |
7c673cae | 2437 | // discard first old_log_jump_to extents |
9f95a23c | 2438 | |
7c673cae FG |
2439 | dout(10) << __func__ << " remove 0x" << std::hex << old_log_jump_to << std::dec |
2440 | << " of " << log_file->fnode.extents << dendl; | |
2441 | uint64_t discarded = 0; | |
2442 | mempool::bluefs::vector<bluefs_extent_t> old_extents; | |
2443 | while (discarded < old_log_jump_to) { | |
11fdf7f2 | 2444 | ceph_assert(!log_file->fnode.extents.empty()); |
7c673cae FG |
2445 | bluefs_extent_t& e = log_file->fnode.extents.front(); |
2446 | bluefs_extent_t temp = e; | |
2447 | if (discarded + e.length <= old_log_jump_to) { | |
2448 | dout(10) << __func__ << " remove old log extent " << e << dendl; | |
2449 | discarded += e.length; | |
94b18763 | 2450 | log_file->fnode.pop_front_extent(); |
7c673cae FG |
2451 | } else { |
2452 | dout(10) << __func__ << " remove front of old log extent " << e << dendl; | |
2453 | uint64_t drop = old_log_jump_to - discarded; | |
2454 | temp.length = drop; | |
2455 | e.offset += drop; | |
2456 | e.length -= drop; | |
2457 | discarded += drop; | |
2458 | dout(10) << __func__ << " kept " << e << " removed " << temp << dendl; | |
2459 | } | |
2460 | old_extents.push_back(temp); | |
2461 | } | |
94b18763 FG |
2462 | auto from = log_file->fnode.extents.begin(); |
2463 | auto to = log_file->fnode.extents.end(); | |
2464 | while (from != to) { | |
2465 | new_log->fnode.append_extent(*from); | |
2466 | ++from; | |
2467 | } | |
7c673cae | 2468 | |
9f95a23c TL |
2469 | vselector->sub_usage(log_file->vselector_hint, log_file->fnode); |
2470 | ||
7c673cae | 2471 | // clear the extents from old log file, they are added to new log |
94b18763 | 2472 | log_file->fnode.clear_extents(); |
7c673cae | 2473 | // swap the log files. New log file is the log file now. |
94b18763 FG |
2474 | new_log->fnode.swap_extents(log_file->fnode); |
2475 | ||
7c673cae FG |
2476 | log_writer->pos = log_writer->file->fnode.size = |
2477 | log_writer->pos - old_log_jump_to + new_log_jump_to; | |
2478 | ||
9f95a23c TL |
2479 | vselector->add_usage(log_file->vselector_hint, log_file->fnode); |
2480 | ||
11fdf7f2 | 2481 | // 6. write the super block to reflect the changes |
7c673cae FG |
2482 | dout(10) << __func__ << " writing super" << dendl; |
2483 | super.log_fnode = log_file->fnode; | |
2484 | ++super.version; | |
11fdf7f2 | 2485 | _write_super(BDEV_DB); |
7c673cae FG |
2486 | |
2487 | lock.unlock(); | |
2488 | flush_bdev(); | |
2489 | lock.lock(); | |
2490 | ||
11fdf7f2 | 2491 | // 7. release old space |
7c673cae FG |
2492 | dout(10) << __func__ << " release old log extents " << old_extents << dendl; |
2493 | for (auto& r : old_extents) { | |
2494 | pending_release[r.bdev].insert(r.offset, r.length); | |
2495 | } | |
2496 | ||
2497 | // delete the new log, remove from the dirty files list | |
2498 | _close_writer(new_log_writer); | |
2499 | if (new_log->dirty_seq) { | |
11fdf7f2 | 2500 | ceph_assert(dirty_files.count(new_log->dirty_seq)); |
7c673cae FG |
2501 | auto it = dirty_files[new_log->dirty_seq].iterator_to(*new_log); |
2502 | dirty_files[new_log->dirty_seq].erase(it); | |
2503 | } | |
2504 | new_log_writer = nullptr; | |
2505 | new_log = nullptr; | |
2506 | log_cond.notify_all(); | |
2507 | ||
2508 | dout(10) << __func__ << " log extents " << log_file->fnode.extents << dendl; | |
2509 | logger->inc(l_bluefs_log_compactions); | |
2510 | } | |
2511 | ||
2512 | void BlueFS::_pad_bl(bufferlist& bl) | |
2513 | { | |
2514 | uint64_t partial = bl.length() % super.block_size; | |
2515 | if (partial) { | |
2516 | dout(10) << __func__ << " padding with 0x" << std::hex | |
2517 | << super.block_size - partial << " zeros" << std::dec << dendl; | |
2518 | bl.append_zero(super.block_size - partial); | |
2519 | } | |
2520 | } | |
2521 | ||
7c673cae | 2522 | |
11fdf7f2 | 2523 | int BlueFS::_flush_and_sync_log(std::unique_lock<ceph::mutex>& l, |
7c673cae FG |
2524 | uint64_t want_seq, |
2525 | uint64_t jump_to) | |
2526 | { | |
2527 | while (log_flushing) { | |
2528 | dout(10) << __func__ << " want_seq " << want_seq | |
2529 | << " log is currently flushing, waiting" << dendl; | |
11fdf7f2 | 2530 | ceph_assert(!jump_to); |
7c673cae FG |
2531 | log_cond.wait(l); |
2532 | } | |
2533 | if (want_seq && want_seq <= log_seq_stable) { | |
2534 | dout(10) << __func__ << " want_seq " << want_seq << " <= log_seq_stable " | |
2535 | << log_seq_stable << ", done" << dendl; | |
11fdf7f2 | 2536 | ceph_assert(!jump_to); |
7c673cae FG |
2537 | return 0; |
2538 | } | |
2539 | if (log_t.empty() && dirty_files.empty()) { | |
2540 | dout(10) << __func__ << " want_seq " << want_seq | |
2541 | << " " << log_t << " not dirty, dirty_files empty, no-op" << dendl; | |
11fdf7f2 | 2542 | ceph_assert(!jump_to); |
7c673cae FG |
2543 | return 0; |
2544 | } | |
2545 | ||
a8e16298 TL |
2546 | vector<interval_set<uint64_t>> to_release(pending_release.size()); |
2547 | to_release.swap(pending_release); | |
2548 | ||
7c673cae | 2549 | uint64_t seq = log_t.seq = ++log_seq; |
11fdf7f2 | 2550 | ceph_assert(want_seq == 0 || want_seq <= seq); |
7c673cae FG |
2551 | log_t.uuid = super.uuid; |
2552 | ||
2553 | // log dirty files | |
2554 | auto lsi = dirty_files.find(seq); | |
2555 | if (lsi != dirty_files.end()) { | |
2556 | dout(20) << __func__ << " " << lsi->second.size() << " dirty_files" << dendl; | |
2557 | for (auto &f : lsi->second) { | |
2558 | dout(20) << __func__ << " op_file_update " << f.fnode << dendl; | |
2559 | log_t.op_file_update(f.fnode); | |
2560 | } | |
2561 | } | |
2562 | ||
2563 | dout(10) << __func__ << " " << log_t << dendl; | |
11fdf7f2 | 2564 | ceph_assert(!log_t.empty()); |
7c673cae FG |
2565 | |
2566 | // allocate some more space (before we run out)? | |
2567 | int64_t runway = log_writer->file->fnode.get_allocated() - | |
2568 | log_writer->get_effective_write_pos(); | |
f6b5b4d7 | 2569 | bool just_expanded_log = false; |
7c673cae FG |
2570 | if (runway < (int64_t)cct->_conf->bluefs_min_log_runway) { |
2571 | dout(10) << __func__ << " allocating more log runway (0x" | |
2572 | << std::hex << runway << std::dec << " remaining)" << dendl; | |
2573 | while (new_log_writer) { | |
2574 | dout(10) << __func__ << " waiting for async compaction" << dendl; | |
2575 | log_cond.wait(l); | |
2576 | } | |
9f95a23c TL |
2577 | vselector->sub_usage(log_writer->file->vselector_hint, log_writer->file->fnode); |
2578 | int r = _allocate( | |
2579 | vselector->select_prefer_bdev(log_writer->file->vselector_hint), | |
2580 | cct->_conf->bluefs_max_log_runway, | |
2581 | &log_writer->file->fnode); | |
11fdf7f2 | 2582 | ceph_assert(r == 0); |
9f95a23c | 2583 | vselector->add_usage(log_writer->file->vselector_hint, log_writer->file->fnode); |
7c673cae | 2584 | log_t.op_file_update(log_writer->file->fnode); |
f6b5b4d7 | 2585 | just_expanded_log = true; |
7c673cae FG |
2586 | } |
2587 | ||
2588 | bufferlist bl; | |
11fdf7f2 TL |
2589 | bl.reserve(super.block_size); |
2590 | encode(log_t, bl); | |
7c673cae | 2591 | // pad to block boundary |
11fdf7f2 TL |
2592 | size_t realign = super.block_size - (bl.length() % super.block_size); |
2593 | if (realign && realign != super.block_size) | |
2594 | bl.append_zero(realign); | |
2595 | ||
7c673cae FG |
2596 | logger->inc(l_bluefs_logged_bytes, bl.length()); |
2597 | ||
f6b5b4d7 TL |
2598 | if (just_expanded_log) { |
2599 | ceph_assert(bl.length() <= runway); // if we write this, we will have an unrecoverable data loss | |
2600 | } | |
2601 | ||
7c673cae FG |
2602 | log_writer->append(bl); |
2603 | ||
2604 | log_t.clear(); | |
2605 | log_t.seq = 0; // just so debug output is less confusing | |
2606 | log_flushing = true; | |
2607 | ||
2608 | int r = _flush(log_writer, true); | |
11fdf7f2 | 2609 | ceph_assert(r == 0); |
7c673cae FG |
2610 | |
2611 | if (jump_to) { | |
2612 | dout(10) << __func__ << " jumping log offset from 0x" << std::hex | |
2613 | << log_writer->pos << " -> 0x" << jump_to << std::dec << dendl; | |
2614 | log_writer->pos = jump_to; | |
9f95a23c | 2615 | vselector->sub_usage(log_writer->file->vselector_hint, log_writer->file->fnode.size); |
7c673cae | 2616 | log_writer->file->fnode.size = jump_to; |
9f95a23c | 2617 | vselector->add_usage(log_writer->file->vselector_hint, log_writer->file->fnode.size); |
7c673cae FG |
2618 | } |
2619 | ||
2620 | _flush_bdev_safely(log_writer); | |
2621 | ||
2622 | log_flushing = false; | |
2623 | log_cond.notify_all(); | |
2624 | ||
2625 | // clean dirty files | |
2626 | if (seq > log_seq_stable) { | |
2627 | log_seq_stable = seq; | |
2628 | dout(20) << __func__ << " log_seq_stable " << log_seq_stable << dendl; | |
2629 | ||
2630 | auto p = dirty_files.begin(); | |
2631 | while (p != dirty_files.end()) { | |
2632 | if (p->first > log_seq_stable) { | |
2633 | dout(20) << __func__ << " done cleaning up dirty files" << dendl; | |
2634 | break; | |
2635 | } | |
2636 | ||
2637 | auto l = p->second.begin(); | |
2638 | while (l != p->second.end()) { | |
2639 | File *file = &*l; | |
11fdf7f2 TL |
2640 | ceph_assert(file->dirty_seq > 0); |
2641 | ceph_assert(file->dirty_seq <= log_seq_stable); | |
7c673cae FG |
2642 | dout(20) << __func__ << " cleaned file " << file->fnode << dendl; |
2643 | file->dirty_seq = 0; | |
2644 | p->second.erase(l++); | |
2645 | } | |
2646 | ||
11fdf7f2 | 2647 | ceph_assert(p->second.empty()); |
7c673cae FG |
2648 | dirty_files.erase(p++); |
2649 | } | |
2650 | } else { | |
2651 | dout(20) << __func__ << " log_seq_stable " << log_seq_stable | |
2652 | << " already >= out seq " << seq | |
2653 | << ", we lost a race against another log flush, done" << dendl; | |
2654 | } | |
a8e16298 TL |
2655 | |
2656 | for (unsigned i = 0; i < to_release.size(); ++i) { | |
2657 | if (!to_release[i].empty()) { | |
2658 | /* OK, now we have the guarantee alloc[i] won't be null. */ | |
11fdf7f2 TL |
2659 | int r = 0; |
2660 | if (cct->_conf->bdev_enable_discard && cct->_conf->bdev_async_discard) { | |
2661 | r = bdev[i]->queue_discard(to_release[i]); | |
2662 | if (r == 0) | |
2663 | continue; | |
2664 | } else if (cct->_conf->bdev_enable_discard) { | |
2665 | for (auto p = to_release[i].begin(); p != to_release[i].end(); ++p) { | |
2666 | bdev[i]->discard(p.get_start(), p.get_len()); | |
2667 | } | |
2668 | } | |
a8e16298 TL |
2669 | alloc[i]->release(to_release[i]); |
2670 | } | |
2671 | } | |
2672 | ||
7c673cae FG |
2673 | _update_logger_stats(); |
2674 | ||
2675 | return 0; | |
2676 | } | |
2677 | ||
2678 | int BlueFS::_flush_range(FileWriter *h, uint64_t offset, uint64_t length) | |
2679 | { | |
2680 | dout(10) << __func__ << " " << h << " pos 0x" << std::hex << h->pos | |
2681 | << " 0x" << offset << "~" << length << std::dec | |
2682 | << " to " << h->file->fnode << dendl; | |
11fdf7f2 TL |
2683 | ceph_assert(!h->file->deleted); |
2684 | ceph_assert(h->file->num_readers.load() == 0); | |
7c673cae FG |
2685 | |
2686 | h->buffer_appender.flush(); | |
2687 | ||
2688 | bool buffered; | |
2689 | if (h->file->fnode.ino == 1) | |
2690 | buffered = false; | |
2691 | else | |
2692 | buffered = cct->_conf->bluefs_buffered_io; | |
2693 | ||
2694 | if (offset + length <= h->pos) | |
2695 | return 0; | |
2696 | if (offset < h->pos) { | |
2697 | length -= h->pos - offset; | |
2698 | offset = h->pos; | |
2699 | dout(10) << " still need 0x" | |
2700 | << std::hex << offset << "~" << length << std::dec | |
2701 | << dendl; | |
2702 | } | |
11fdf7f2 | 2703 | ceph_assert(offset <= h->file->fnode.size); |
7c673cae FG |
2704 | |
2705 | uint64_t allocated = h->file->fnode.get_allocated(); | |
9f95a23c | 2706 | vselector->sub_usage(h->file->vselector_hint, h->file->fnode); |
7c673cae FG |
2707 | // do not bother to dirty the file if we are overwriting |
2708 | // previously allocated extents. | |
2709 | bool must_dirty = false; | |
2710 | if (allocated < offset + length) { | |
2711 | // we should never run out of log space here; see the min runway check | |
2712 | // in _flush_and_sync_log. | |
11fdf7f2 | 2713 | ceph_assert(h->file->fnode.ino != 1); |
9f95a23c | 2714 | int r = _allocate(vselector->select_prefer_bdev(h->file->vselector_hint), |
7c673cae | 2715 | offset + length - allocated, |
94b18763 | 2716 | &h->file->fnode); |
7c673cae FG |
2717 | if (r < 0) { |
2718 | derr << __func__ << " allocated: 0x" << std::hex << allocated | |
2719 | << " offset: 0x" << offset << " length: 0x" << length << std::dec | |
2720 | << dendl; | |
9f95a23c | 2721 | vselector->add_usage(h->file->vselector_hint, h->file->fnode); // undo |
11fdf7f2 | 2722 | ceph_abort_msg("bluefs enospc"); |
7c673cae FG |
2723 | return r; |
2724 | } | |
7c673cae FG |
2725 | must_dirty = true; |
2726 | } | |
2727 | if (h->file->fnode.size < offset + length) { | |
2728 | h->file->fnode.size = offset + length; | |
2729 | if (h->file->fnode.ino > 1) { | |
2730 | // we do not need to dirty the log file (or it's compacting | |
2731 | // replacement) when the file size changes because replay is | |
2732 | // smart enough to discover it on its own. | |
2733 | must_dirty = true; | |
2734 | } | |
2735 | } | |
2736 | if (must_dirty) { | |
2737 | h->file->fnode.mtime = ceph_clock_now(); | |
11fdf7f2 | 2738 | ceph_assert(h->file->fnode.ino >= 1); |
7c673cae FG |
2739 | if (h->file->dirty_seq == 0) { |
2740 | h->file->dirty_seq = log_seq + 1; | |
2741 | dirty_files[h->file->dirty_seq].push_back(*h->file); | |
2742 | dout(20) << __func__ << " dirty_seq = " << log_seq + 1 | |
2743 | << " (was clean)" << dendl; | |
2744 | } else { | |
2745 | if (h->file->dirty_seq != log_seq + 1) { | |
2746 | // need re-dirty, erase from list first | |
11fdf7f2 | 2747 | ceph_assert(dirty_files.count(h->file->dirty_seq)); |
7c673cae FG |
2748 | auto it = dirty_files[h->file->dirty_seq].iterator_to(*h->file); |
2749 | dirty_files[h->file->dirty_seq].erase(it); | |
2750 | h->file->dirty_seq = log_seq + 1; | |
2751 | dirty_files[h->file->dirty_seq].push_back(*h->file); | |
2752 | dout(20) << __func__ << " dirty_seq = " << log_seq + 1 | |
2753 | << " (was " << h->file->dirty_seq << ")" << dendl; | |
2754 | } else { | |
2755 | dout(20) << __func__ << " dirty_seq = " << log_seq + 1 | |
2756 | << " (unchanged, do nothing) " << dendl; | |
2757 | } | |
2758 | } | |
2759 | } | |
2760 | dout(20) << __func__ << " file now " << h->file->fnode << dendl; | |
2761 | ||
2762 | uint64_t x_off = 0; | |
2763 | auto p = h->file->fnode.seek(offset, &x_off); | |
11fdf7f2 | 2764 | ceph_assert(p != h->file->fnode.extents.end()); |
7c673cae FG |
2765 | dout(20) << __func__ << " in " << *p << " x_off 0x" |
2766 | << std::hex << x_off << std::dec << dendl; | |
2767 | ||
2768 | unsigned partial = x_off & ~super.block_mask(); | |
2769 | bufferlist bl; | |
2770 | if (partial) { | |
2771 | dout(20) << __func__ << " using partial tail 0x" | |
2772 | << std::hex << partial << std::dec << dendl; | |
11fdf7f2 | 2773 | ceph_assert(h->tail_block.length() == partial); |
31f18b77 | 2774 | bl.claim_append_piecewise(h->tail_block); |
7c673cae FG |
2775 | x_off -= partial; |
2776 | offset -= partial; | |
2777 | length += partial; | |
2778 | dout(20) << __func__ << " waiting for previous aio to complete" << dendl; | |
2779 | for (auto p : h->iocv) { | |
2780 | if (p) { | |
2781 | p->aio_wait(); | |
2782 | } | |
2783 | } | |
2784 | } | |
f91f0fd5 | 2785 | if (length == partial + h->buffer.length()) { |
9f95a23c | 2786 | /* in case of inital allocation and need to zero, limited flush is unacceptable */ |
31f18b77 | 2787 | bl.claim_append_piecewise(h->buffer); |
7c673cae FG |
2788 | } else { |
2789 | bufferlist t; | |
31f18b77 FG |
2790 | h->buffer.splice(0, length, &t); |
2791 | bl.claim_append_piecewise(t); | |
7c673cae FG |
2792 | t.substr_of(h->buffer, length, h->buffer.length() - length); |
2793 | h->buffer.swap(t); | |
2794 | dout(20) << " leaving 0x" << std::hex << h->buffer.length() << std::dec | |
2795 | << " unflushed" << dendl; | |
2796 | } | |
11fdf7f2 | 2797 | ceph_assert(bl.length() == length); |
7c673cae | 2798 | |
9f95a23c TL |
2799 | h->pos = offset + length; |
2800 | ||
2801 | unsigned tail = bl.length() & ~super.block_mask(); | |
2802 | if (tail) { | |
2803 | dout(20) << __func__ << " caching tail of 0x" | |
2804 | << std::hex << tail | |
2805 | << " and padding block with 0x" << (super.block_size - tail) | |
2806 | << std::dec << dendl; | |
2807 | h->tail_block.substr_of(bl, bl.length() - tail, tail); | |
2808 | bl.append_zero(super.block_size - tail); | |
2809 | length += super.block_size - tail; | |
2810 | } else { | |
2811 | h->tail_block.clear(); | |
2812 | } | |
9f95a23c TL |
2813 | ceph_assert(bl.length() == length); |
2814 | ||
7c673cae FG |
2815 | switch (h->writer_type) { |
2816 | case WRITER_WAL: | |
2817 | logger->inc(l_bluefs_bytes_written_wal, length); | |
2818 | break; | |
2819 | case WRITER_SST: | |
2820 | logger->inc(l_bluefs_bytes_written_sst, length); | |
2821 | break; | |
2822 | } | |
2823 | ||
2824 | dout(30) << "dump:\n"; | |
2825 | bl.hexdump(*_dout); | |
2826 | *_dout << dendl; | |
2827 | ||
7c673cae | 2828 | uint64_t bloff = 0; |
11fdf7f2 | 2829 | uint64_t bytes_written_slow = 0; |
7c673cae | 2830 | while (length > 0) { |
11fdf7f2 | 2831 | uint64_t x_len = std::min(p->length - x_off, length); |
7c673cae FG |
2832 | bufferlist t; |
2833 | t.substr_of(bl, bloff, x_len); | |
7c673cae | 2834 | if (cct->_conf->bluefs_sync_write) { |
11fdf7f2 | 2835 | bdev[p->bdev]->write(p->offset + x_off, t, buffered, h->write_hint); |
7c673cae | 2836 | } else { |
11fdf7f2 TL |
2837 | bdev[p->bdev]->aio_write(p->offset + x_off, t, h->iocv[p->bdev], buffered, h->write_hint); |
2838 | } | |
2839 | h->dirty_devs[p->bdev] = true; | |
2840 | if (p->bdev == BDEV_SLOW) { | |
2841 | bytes_written_slow += t.length(); | |
7c673cae | 2842 | } |
11fdf7f2 | 2843 | |
7c673cae FG |
2844 | bloff += x_len; |
2845 | length -= x_len; | |
2846 | ++p; | |
2847 | x_off = 0; | |
2848 | } | |
11fdf7f2 | 2849 | logger->inc(l_bluefs_bytes_written_slow, bytes_written_slow); |
7c673cae FG |
2850 | for (unsigned i = 0; i < MAX_BDEV; ++i) { |
2851 | if (bdev[i]) { | |
11fdf7f2 | 2852 | if (h->iocv[i] && h->iocv[i]->has_pending_aios()) { |
7c673cae FG |
2853 | bdev[i]->aio_submit(h->iocv[i]); |
2854 | } | |
2855 | } | |
2856 | } | |
9f95a23c | 2857 | vselector->add_usage(h->file->vselector_hint, h->file->fnode); |
7c673cae FG |
2858 | dout(20) << __func__ << " h " << h << " pos now 0x" |
2859 | << std::hex << h->pos << std::dec << dendl; | |
2860 | return 0; | |
2861 | } | |
2862 | ||
11fdf7f2 | 2863 | #ifdef HAVE_LIBAIO |
7c673cae FG |
2864 | // we need to retire old completed aios so they don't stick around in |
2865 | // memory indefinitely (along with their bufferlist refs). | |
2866 | void BlueFS::_claim_completed_aios(FileWriter *h, list<aio_t> *ls) | |
2867 | { | |
2868 | for (auto p : h->iocv) { | |
2869 | if (p) { | |
2870 | ls->splice(ls->end(), p->running_aios); | |
2871 | } | |
2872 | } | |
2873 | dout(10) << __func__ << " got " << ls->size() << " aios" << dendl; | |
2874 | } | |
2875 | ||
2876 | void BlueFS::wait_for_aio(FileWriter *h) | |
2877 | { | |
2878 | // NOTE: this is safe to call without a lock, as long as our reference is | |
2879 | // stable. | |
2880 | dout(10) << __func__ << " " << h << dendl; | |
2881 | utime_t start = ceph_clock_now(); | |
2882 | for (auto p : h->iocv) { | |
2883 | if (p) { | |
2884 | p->aio_wait(); | |
2885 | } | |
2886 | } | |
11fdf7f2 | 2887 | dout(10) << __func__ << " " << h << " done in " << (ceph_clock_now() - start) << dendl; |
7c673cae | 2888 | } |
11fdf7f2 | 2889 | #endif |
7c673cae | 2890 | |
f6b5b4d7 TL |
2891 | int BlueFS::_flush(FileWriter *h, bool force, std::unique_lock<ceph::mutex>& l) |
2892 | { | |
2893 | bool flushed = false; | |
2894 | int r = _flush(h, force, &flushed); | |
2895 | if (r == 0 && flushed) { | |
2896 | _maybe_compact_log(l); | |
2897 | } | |
2898 | return r; | |
2899 | } | |
2900 | ||
2901 | int BlueFS::_flush(FileWriter *h, bool force, bool *flushed) | |
7c673cae FG |
2902 | { |
2903 | h->buffer_appender.flush(); | |
2904 | uint64_t length = h->buffer.length(); | |
2905 | uint64_t offset = h->pos; | |
f6b5b4d7 TL |
2906 | if (flushed) { |
2907 | *flushed = false; | |
2908 | } | |
7c673cae FG |
2909 | if (!force && |
2910 | length < cct->_conf->bluefs_min_flush_size) { | |
2911 | dout(10) << __func__ << " " << h << " ignoring, length " << length | |
2912 | << " < min_flush_size " << cct->_conf->bluefs_min_flush_size | |
2913 | << dendl; | |
2914 | return 0; | |
2915 | } | |
2916 | if (length == 0) { | |
2917 | dout(10) << __func__ << " " << h << " no dirty data on " | |
2918 | << h->file->fnode << dendl; | |
2919 | return 0; | |
2920 | } | |
2921 | dout(10) << __func__ << " " << h << " 0x" | |
2922 | << std::hex << offset << "~" << length << std::dec | |
2923 | << " to " << h->file->fnode << dendl; | |
11fdf7f2 | 2924 | ceph_assert(h->pos <= h->file->fnode.size); |
f6b5b4d7 TL |
2925 | int r = _flush_range(h, offset, length); |
2926 | if (flushed) { | |
2927 | *flushed = true; | |
2928 | } | |
2929 | return r; | |
7c673cae FG |
2930 | } |
2931 | ||
2932 | int BlueFS::_truncate(FileWriter *h, uint64_t offset) | |
2933 | { | |
2934 | dout(10) << __func__ << " 0x" << std::hex << offset << std::dec | |
2935 | << " file " << h->file->fnode << dendl; | |
2936 | if (h->file->deleted) { | |
2937 | dout(10) << __func__ << " deleted, no-op" << dendl; | |
2938 | return 0; | |
2939 | } | |
2940 | ||
2941 | // we never truncate internal log files | |
11fdf7f2 | 2942 | ceph_assert(h->file->fnode.ino > 1); |
7c673cae FG |
2943 | |
2944 | h->buffer_appender.flush(); | |
2945 | ||
2946 | // truncate off unflushed data? | |
2947 | if (h->pos < offset && | |
2948 | h->pos + h->buffer.length() > offset) { | |
2949 | bufferlist t; | |
2950 | dout(20) << __func__ << " tossing out last " << offset - h->pos | |
2951 | << " unflushed bytes" << dendl; | |
2952 | t.substr_of(h->buffer, 0, offset - h->pos); | |
2953 | h->buffer.swap(t); | |
11fdf7f2 | 2954 | ceph_abort_msg("actually this shouldn't happen"); |
7c673cae FG |
2955 | } |
2956 | if (h->buffer.length()) { | |
2957 | int r = _flush(h, true); | |
2958 | if (r < 0) | |
2959 | return r; | |
2960 | } | |
2961 | if (offset == h->file->fnode.size) { | |
2962 | return 0; // no-op! | |
2963 | } | |
2964 | if (offset > h->file->fnode.size) { | |
11fdf7f2 | 2965 | ceph_abort_msg("truncate up not supported"); |
7c673cae | 2966 | } |
11fdf7f2 | 2967 | ceph_assert(h->file->fnode.size >= offset); |
9f95a23c | 2968 | vselector->sub_usage(h->file->vselector_hint, h->file->fnode.size); |
7c673cae | 2969 | h->file->fnode.size = offset; |
9f95a23c | 2970 | vselector->add_usage(h->file->vselector_hint, h->file->fnode.size); |
7c673cae FG |
2971 | log_t.op_file_update(h->file->fnode); |
2972 | return 0; | |
2973 | } | |
2974 | ||
11fdf7f2 | 2975 | int BlueFS::_fsync(FileWriter *h, std::unique_lock<ceph::mutex>& l) |
7c673cae FG |
2976 | { |
2977 | dout(10) << __func__ << " " << h << " " << h->file->fnode << dendl; | |
2978 | int r = _flush(h, true); | |
2979 | if (r < 0) | |
2980 | return r; | |
2981 | uint64_t old_dirty_seq = h->file->dirty_seq; | |
2982 | ||
2983 | _flush_bdev_safely(h); | |
2984 | ||
2985 | if (old_dirty_seq) { | |
2986 | uint64_t s = log_seq; | |
2987 | dout(20) << __func__ << " file metadata was dirty (" << old_dirty_seq | |
2988 | << ") on " << h->file->fnode << ", flushing log" << dendl; | |
2989 | _flush_and_sync_log(l, old_dirty_seq); | |
11fdf7f2 | 2990 | ceph_assert(h->file->dirty_seq == 0 || // cleaned |
7c673cae FG |
2991 | h->file->dirty_seq > s); // or redirtied by someone else |
2992 | } | |
2993 | return 0; | |
2994 | } | |
2995 | ||
2996 | void BlueFS::_flush_bdev_safely(FileWriter *h) | |
2997 | { | |
11fdf7f2 TL |
2998 | std::array<bool, MAX_BDEV> flush_devs = h->dirty_devs; |
2999 | h->dirty_devs.fill(false); | |
3000 | #ifdef HAVE_LIBAIO | |
7c673cae FG |
3001 | if (!cct->_conf->bluefs_sync_write) { |
3002 | list<aio_t> completed_ios; | |
3003 | _claim_completed_aios(h, &completed_ios); | |
3004 | lock.unlock(); | |
3005 | wait_for_aio(h); | |
3006 | completed_ios.clear(); | |
11fdf7f2 | 3007 | flush_bdev(flush_devs); |
7c673cae | 3008 | lock.lock(); |
11fdf7f2 TL |
3009 | } else |
3010 | #endif | |
3011 | { | |
7c673cae | 3012 | lock.unlock(); |
11fdf7f2 | 3013 | flush_bdev(flush_devs); |
7c673cae FG |
3014 | lock.lock(); |
3015 | } | |
3016 | } | |
3017 | ||
11fdf7f2 TL |
3018 | void BlueFS::flush_bdev(std::array<bool, MAX_BDEV>& dirty_bdevs) |
3019 | { | |
3020 | // NOTE: this is safe to call without a lock. | |
3021 | dout(20) << __func__ << dendl; | |
3022 | for (unsigned i = 0; i < MAX_BDEV; i++) { | |
3023 | if (dirty_bdevs[i]) | |
3024 | bdev[i]->flush(); | |
3025 | } | |
3026 | } | |
3027 | ||
7c673cae FG |
3028 | void BlueFS::flush_bdev() |
3029 | { | |
3030 | // NOTE: this is safe to call without a lock. | |
3031 | dout(20) << __func__ << dendl; | |
3032 | for (auto p : bdev) { | |
3033 | if (p) | |
3034 | p->flush(); | |
3035 | } | |
3036 | } | |
3037 | ||
eafe8130 TL |
3038 | const char* BlueFS::get_device_name(unsigned id) |
3039 | { | |
3040 | if (id >= MAX_BDEV) return "BDEV_INV"; | |
3041 | const char* names[] = {"BDEV_WAL", "BDEV_DB", "BDEV_SLOW", "BDEV_NEWWAL", "BDEV_NEWDB"}; | |
3042 | return names[id]; | |
3043 | } | |
3044 | ||
11fdf7f2 TL |
3045 | int BlueFS::_expand_slow_device(uint64_t need, PExtentVector& extents) |
3046 | { | |
3047 | int r = -ENOSPC; | |
3048 | if (slow_dev_expander) { | |
1911f103 | 3049 | auto id = _get_slow_device_id(); |
eafe8130 | 3050 | auto min_alloc_size = alloc_size[id]; |
1911f103 | 3051 | ceph_assert(id <= alloc.size() && alloc[id]); |
11fdf7f2 TL |
3052 | auto min_need = round_up_to(need, min_alloc_size); |
3053 | need = std::max(need, | |
3054 | slow_dev_expander->get_recommended_expansion_delta( | |
3055 | alloc[id]->get_free(), block_all[id].size())); | |
3056 | ||
3057 | need = round_up_to(need, min_alloc_size); | |
3058 | dout(10) << __func__ << " expanding slow device by 0x" | |
3059 | << std::hex << need << std::dec | |
3060 | << dendl; | |
3061 | r = slow_dev_expander->allocate_freespace(min_need, need, extents); | |
3062 | } | |
3063 | return r; | |
3064 | } | |
3065 | ||
3066 | int BlueFS::_allocate_without_fallback(uint8_t id, uint64_t len, | |
3067 | PExtentVector* extents) | |
3068 | { | |
3069 | dout(10) << __func__ << " len 0x" << std::hex << len << std::dec | |
3070 | << " from " << (int)id << dendl; | |
3071 | assert(id < alloc.size()); | |
11fdf7f2 TL |
3072 | if (!alloc[id]) { |
3073 | return -ENOENT; | |
3074 | } | |
3075 | extents->reserve(4); // 4 should be (more than) enough for most allocations | |
eafe8130 TL |
3076 | uint64_t min_alloc_size = alloc_size[id]; |
3077 | uint64_t left = round_up_to(len, min_alloc_size); | |
11fdf7f2 | 3078 | int64_t alloc_len = alloc[id]->allocate(left, min_alloc_size, 0, extents); |
eafe8130 TL |
3079 | if (alloc_len < 0 || alloc_len < (int64_t)left) { |
3080 | if (alloc_len > 0) { | |
11fdf7f2 TL |
3081 | alloc[id]->release(*extents); |
3082 | } | |
3083 | if (bdev[id]) | |
3084 | derr << __func__ << " failed to allocate 0x" << std::hex << left | |
3085 | << " on bdev " << (int)id | |
3086 | << ", free 0x" << alloc[id]->get_free() << std::dec << dendl; | |
3087 | else | |
3088 | derr << __func__ << " failed to allocate 0x" << std::hex << left | |
3089 | << " on bdev " << (int)id << ", dne" << std::dec << dendl; | |
3090 | if (alloc[id]) | |
3091 | alloc[id]->dump(); | |
3092 | return -ENOSPC; | |
3093 | } | |
3094 | ||
3095 | return 0; | |
3096 | } | |
3097 | ||
7c673cae | 3098 | int BlueFS::_allocate(uint8_t id, uint64_t len, |
94b18763 | 3099 | bluefs_fnode_t* node) |
7c673cae FG |
3100 | { |
3101 | dout(10) << __func__ << " len 0x" << std::hex << len << std::dec | |
3102 | << " from " << (int)id << dendl; | |
11fdf7f2 | 3103 | ceph_assert(id < alloc.size()); |
b32b8144 | 3104 | int64_t alloc_len = 0; |
a8e16298 | 3105 | PExtentVector extents; |
11fdf7f2 | 3106 | uint64_t hint = 0; |
7c673cae | 3107 | if (alloc[id]) { |
94b18763 FG |
3108 | if (!node->extents.empty() && node->extents.back().bdev == id) { |
3109 | hint = node->extents.back().end(); | |
11fdf7f2 | 3110 | } |
b32b8144 | 3111 | extents.reserve(4); // 4 should be (more than) enough for most allocations |
eafe8130 TL |
3112 | alloc_len = alloc[id]->allocate(round_up_to(len, alloc_size[id]), |
3113 | alloc_size[id], hint, &extents); | |
b32b8144 | 3114 | } |
eafe8130 TL |
3115 | if (!alloc[id] || |
3116 | alloc_len < 0 || | |
3117 | alloc_len < (int64_t)round_up_to(len, alloc_size[id])) { | |
11fdf7f2 | 3118 | if (alloc_len > 0) { |
a8e16298 | 3119 | alloc[id]->release(extents); |
b32b8144 | 3120 | } |
7c673cae FG |
3121 | if (id != BDEV_SLOW) { |
3122 | if (bdev[id]) { | |
eafe8130 | 3123 | dout(1) << __func__ << " failed to allocate 0x" << std::hex << len |
7c673cae FG |
3124 | << " on bdev " << (int)id |
3125 | << ", free 0x" << alloc[id]->get_free() | |
3126 | << "; fallback to bdev " << (int)id + 1 | |
3127 | << std::dec << dendl; | |
3128 | } | |
94b18763 | 3129 | return _allocate(id + 1, len, node); |
7c673cae | 3130 | } |
eafe8130 | 3131 | dout(1) << __func__ << " unable to allocate 0x" << std::hex << len |
11fdf7f2 TL |
3132 | << " on bdev " << (int)id << ", free 0x" |
3133 | << (alloc[id] ? alloc[id]->get_free() : (uint64_t)-1) | |
3134 | << "; fallback to slow device expander " | |
3135 | << std::dec << dendl; | |
3136 | extents.clear(); | |
eafe8130 | 3137 | if (_expand_slow_device(len, extents) == 0) { |
11fdf7f2 TL |
3138 | id = _get_slow_device_id(); |
3139 | for (auto& e : extents) { | |
3140 | _add_block_extent(id, e.offset, e.length); | |
3141 | } | |
3142 | extents.clear(); | |
3143 | auto* last_alloc = alloc[id]; | |
3144 | ceph_assert(last_alloc); | |
3145 | // try again | |
eafe8130 TL |
3146 | alloc_len = last_alloc->allocate(round_up_to(len, alloc_size[id]), |
3147 | alloc_size[id], hint, &extents); | |
3148 | if (alloc_len < 0 || alloc_len < (int64_t)len) { | |
11fdf7f2 TL |
3149 | if (alloc_len > 0) { |
3150 | last_alloc->release(extents); | |
3151 | } | |
eafe8130 | 3152 | derr << __func__ << " failed to allocate 0x" << std::hex << len |
11fdf7f2 TL |
3153 | << " on bdev " << (int)id |
3154 | << ", free 0x" << last_alloc->get_free() << std::dec << dendl; | |
3155 | return -ENOSPC; | |
3156 | } | |
3157 | } else { | |
3158 | derr << __func__ << " failed to expand slow device to fit +0x" | |
eafe8130 | 3159 | << std::hex << len << std::dec |
11fdf7f2 TL |
3160 | << dendl; |
3161 | return -ENOSPC; | |
3162 | } | |
3163 | } else { | |
3164 | uint64_t total_allocated = | |
3165 | block_all[id].size() - alloc[id]->get_free(); | |
3166 | if (max_bytes[id] < total_allocated) { | |
3167 | logger->set(max_bytes_pcounters[id], total_allocated); | |
3168 | max_bytes[id] = total_allocated; | |
3169 | } | |
7c673cae FG |
3170 | } |
3171 | ||
3172 | for (auto& p : extents) { | |
94b18763 | 3173 | node->append_extent(bluefs_extent_t(id, p.offset, p.length)); |
7c673cae FG |
3174 | } |
3175 | ||
3176 | return 0; | |
3177 | } | |
3178 | ||
3179 | int BlueFS::_preallocate(FileRef f, uint64_t off, uint64_t len) | |
3180 | { | |
3181 | dout(10) << __func__ << " file " << f->fnode << " 0x" | |
3182 | << std::hex << off << "~" << len << std::dec << dendl; | |
3183 | if (f->deleted) { | |
3184 | dout(10) << __func__ << " deleted, no-op" << dendl; | |
3185 | return 0; | |
3186 | } | |
11fdf7f2 | 3187 | ceph_assert(f->fnode.ino > 1); |
7c673cae FG |
3188 | uint64_t allocated = f->fnode.get_allocated(); |
3189 | if (off + len > allocated) { | |
3190 | uint64_t want = off + len - allocated; | |
9f95a23c TL |
3191 | vselector->sub_usage(f->vselector_hint, f->fnode); |
3192 | ||
3193 | int r = _allocate(vselector->select_prefer_bdev(f->vselector_hint), | |
3194 | want, | |
3195 | &f->fnode); | |
3196 | vselector->add_usage(f->vselector_hint, f->fnode); | |
7c673cae FG |
3197 | if (r < 0) |
3198 | return r; | |
7c673cae FG |
3199 | log_t.op_file_update(f->fnode); |
3200 | } | |
3201 | return 0; | |
3202 | } | |
3203 | ||
1911f103 | 3204 | void BlueFS::sync_metadata(bool avoid_compact) |
7c673cae | 3205 | { |
f6b5b4d7 | 3206 | std::unique_lock<ceph::mutex> l(lock); |
9f95a23c | 3207 | if (log_t.empty() && dirty_files.empty()) { |
7c673cae | 3208 | dout(10) << __func__ << " - no pending log events" << dendl; |
11fdf7f2 TL |
3209 | } else { |
3210 | dout(10) << __func__ << dendl; | |
3211 | utime_t start = ceph_clock_now(); | |
3212 | flush_bdev(); // FIXME? | |
3213 | _flush_and_sync_log(l); | |
3214 | dout(10) << __func__ << " done in " << (ceph_clock_now() - start) << dendl; | |
7c673cae | 3215 | } |
7c673cae | 3216 | |
f6b5b4d7 TL |
3217 | if (!avoid_compact) { |
3218 | _maybe_compact_log(l); | |
3219 | } | |
3220 | } | |
3221 | ||
3222 | void BlueFS::_maybe_compact_log(std::unique_lock<ceph::mutex>& l) | |
3223 | { | |
3224 | if (!cct->_conf->bluefs_replay_recovery_disable_compact && | |
3225 | _should_compact_log()) { | |
7c673cae FG |
3226 | if (cct->_conf->bluefs_compact_log_sync) { |
3227 | _compact_log_sync(); | |
3228 | } else { | |
3229 | _compact_log_async(l); | |
3230 | } | |
3231 | } | |
7c673cae FG |
3232 | } |
3233 | ||
3234 | int BlueFS::open_for_write( | |
3235 | const string& dirname, | |
3236 | const string& filename, | |
3237 | FileWriter **h, | |
3238 | bool overwrite) | |
3239 | { | |
11fdf7f2 | 3240 | std::lock_guard l(lock); |
7c673cae FG |
3241 | dout(10) << __func__ << " " << dirname << "/" << filename << dendl; |
3242 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3243 | DirRef dir; | |
3244 | if (p == dir_map.end()) { | |
3245 | // implicitly create the dir | |
3246 | dout(20) << __func__ << " dir " << dirname | |
3247 | << " does not exist" << dendl; | |
3248 | return -ENOENT; | |
3249 | } else { | |
3250 | dir = p->second; | |
3251 | } | |
3252 | ||
3253 | FileRef file; | |
3254 | bool create = false; | |
f6b5b4d7 | 3255 | bool truncate = false; |
7c673cae FG |
3256 | map<string,FileRef>::iterator q = dir->file_map.find(filename); |
3257 | if (q == dir->file_map.end()) { | |
3258 | if (overwrite) { | |
3259 | dout(20) << __func__ << " dir " << dirname << " (" << dir | |
3260 | << ") file " << filename | |
3261 | << " does not exist" << dendl; | |
3262 | return -ENOENT; | |
3263 | } | |
9f95a23c | 3264 | file = ceph::make_ref<File>(); |
7c673cae FG |
3265 | file->fnode.ino = ++ino_last; |
3266 | file_map[ino_last] = file; | |
3267 | dir->file_map[filename] = file; | |
3268 | ++file->refs; | |
3269 | create = true; | |
3270 | } else { | |
3271 | // overwrite existing file? | |
3272 | file = q->second; | |
3273 | if (overwrite) { | |
3274 | dout(20) << __func__ << " dir " << dirname << " (" << dir | |
3275 | << ") file " << filename | |
3276 | << " already exists, overwrite in place" << dendl; | |
3277 | } else { | |
3278 | dout(20) << __func__ << " dir " << dirname << " (" << dir | |
3279 | << ") file " << filename | |
3280 | << " already exists, truncate + overwrite" << dendl; | |
9f95a23c | 3281 | vselector->sub_usage(file->vselector_hint, file->fnode); |
7c673cae FG |
3282 | file->fnode.size = 0; |
3283 | for (auto& p : file->fnode.extents) { | |
3284 | pending_release[p.bdev].insert(p.offset, p.length); | |
3285 | } | |
f6b5b4d7 | 3286 | truncate = true; |
94b18763 FG |
3287 | |
3288 | file->fnode.clear_extents(); | |
7c673cae FG |
3289 | } |
3290 | } | |
11fdf7f2 | 3291 | ceph_assert(file->fnode.ino > 1); |
7c673cae FG |
3292 | |
3293 | file->fnode.mtime = ceph_clock_now(); | |
9f95a23c | 3294 | file->vselector_hint = vselector->get_hint_by_dir(dirname); |
f6b5b4d7 TL |
3295 | if (create || truncate) { |
3296 | vselector->add_usage(file->vselector_hint, file->fnode); // update file count | |
3297 | } | |
9f95a23c | 3298 | |
7c673cae | 3299 | dout(20) << __func__ << " mapping " << dirname << "/" << filename |
9f95a23c TL |
3300 | << " vsel_hint " << file->vselector_hint |
3301 | << dendl; | |
7c673cae FG |
3302 | |
3303 | log_t.op_file_update(file->fnode); | |
3304 | if (create) | |
3305 | log_t.op_dir_link(dirname, filename, file->fnode.ino); | |
3306 | ||
3307 | *h = _create_writer(file); | |
3308 | ||
3309 | if (boost::algorithm::ends_with(filename, ".log")) { | |
3310 | (*h)->writer_type = BlueFS::WRITER_WAL; | |
3311 | if (logger && !overwrite) { | |
3312 | logger->inc(l_bluefs_files_written_wal); | |
3313 | } | |
3314 | } else if (boost::algorithm::ends_with(filename, ".sst")) { | |
3315 | (*h)->writer_type = BlueFS::WRITER_SST; | |
3316 | if (logger) { | |
3317 | logger->inc(l_bluefs_files_written_sst); | |
3318 | } | |
3319 | } | |
3320 | ||
3321 | dout(10) << __func__ << " h " << *h << " on " << file->fnode << dendl; | |
3322 | return 0; | |
3323 | } | |
3324 | ||
3325 | BlueFS::FileWriter *BlueFS::_create_writer(FileRef f) | |
3326 | { | |
3327 | FileWriter *w = new FileWriter(f); | |
3328 | for (unsigned i = 0; i < MAX_BDEV; ++i) { | |
3329 | if (bdev[i]) { | |
3330 | w->iocv[i] = new IOContext(cct, NULL); | |
7c673cae FG |
3331 | } |
3332 | } | |
3333 | return w; | |
3334 | } | |
3335 | ||
3336 | void BlueFS::_close_writer(FileWriter *h) | |
3337 | { | |
3338 | dout(10) << __func__ << " " << h << " type " << h->writer_type << dendl; | |
f91f0fd5 | 3339 | h->buffer.reassign_to_mempool(mempool::mempool_bluefs_file_writer); |
7c673cae FG |
3340 | for (unsigned i=0; i<MAX_BDEV; ++i) { |
3341 | if (bdev[i]) { | |
11fdf7f2 TL |
3342 | if (h->iocv[i]) { |
3343 | h->iocv[i]->aio_wait(); | |
3344 | bdev[i]->queue_reap_ioc(h->iocv[i]); | |
3345 | } | |
7c673cae FG |
3346 | } |
3347 | } | |
3348 | delete h; | |
3349 | } | |
3350 | ||
3351 | int BlueFS::open_for_read( | |
3352 | const string& dirname, | |
3353 | const string& filename, | |
3354 | FileReader **h, | |
3355 | bool random) | |
3356 | { | |
11fdf7f2 | 3357 | std::lock_guard l(lock); |
7c673cae FG |
3358 | dout(10) << __func__ << " " << dirname << "/" << filename |
3359 | << (random ? " (random)":" (sequential)") << dendl; | |
3360 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3361 | if (p == dir_map.end()) { | |
3362 | dout(20) << __func__ << " dir " << dirname << " not found" << dendl; | |
3363 | return -ENOENT; | |
3364 | } | |
3365 | DirRef dir = p->second; | |
3366 | ||
3367 | map<string,FileRef>::iterator q = dir->file_map.find(filename); | |
3368 | if (q == dir->file_map.end()) { | |
3369 | dout(20) << __func__ << " dir " << dirname << " (" << dir | |
3370 | << ") file " << filename | |
3371 | << " not found" << dendl; | |
3372 | return -ENOENT; | |
3373 | } | |
3374 | File *file = q->second.get(); | |
3375 | ||
3376 | *h = new FileReader(file, random ? 4096 : cct->_conf->bluefs_max_prefetch, | |
3377 | random, false); | |
3378 | dout(10) << __func__ << " h " << *h << " on " << file->fnode << dendl; | |
3379 | return 0; | |
3380 | } | |
3381 | ||
3382 | int BlueFS::rename( | |
3383 | const string& old_dirname, const string& old_filename, | |
3384 | const string& new_dirname, const string& new_filename) | |
3385 | { | |
11fdf7f2 | 3386 | std::lock_guard l(lock); |
7c673cae FG |
3387 | dout(10) << __func__ << " " << old_dirname << "/" << old_filename |
3388 | << " -> " << new_dirname << "/" << new_filename << dendl; | |
3389 | map<string,DirRef>::iterator p = dir_map.find(old_dirname); | |
3390 | if (p == dir_map.end()) { | |
3391 | dout(20) << __func__ << " dir " << old_dirname << " not found" << dendl; | |
3392 | return -ENOENT; | |
3393 | } | |
3394 | DirRef old_dir = p->second; | |
3395 | map<string,FileRef>::iterator q = old_dir->file_map.find(old_filename); | |
3396 | if (q == old_dir->file_map.end()) { | |
3397 | dout(20) << __func__ << " dir " << old_dirname << " (" << old_dir | |
3398 | << ") file " << old_filename | |
3399 | << " not found" << dendl; | |
3400 | return -ENOENT; | |
3401 | } | |
3402 | FileRef file = q->second; | |
3403 | ||
3404 | p = dir_map.find(new_dirname); | |
3405 | if (p == dir_map.end()) { | |
3406 | dout(20) << __func__ << " dir " << new_dirname << " not found" << dendl; | |
3407 | return -ENOENT; | |
3408 | } | |
3409 | DirRef new_dir = p->second; | |
3410 | q = new_dir->file_map.find(new_filename); | |
3411 | if (q != new_dir->file_map.end()) { | |
3412 | dout(20) << __func__ << " dir " << new_dirname << " (" << old_dir | |
3413 | << ") file " << new_filename | |
3414 | << " already exists, unlinking" << dendl; | |
11fdf7f2 | 3415 | ceph_assert(q->second != file); |
7c673cae FG |
3416 | log_t.op_dir_unlink(new_dirname, new_filename); |
3417 | _drop_link(q->second); | |
3418 | } | |
3419 | ||
3420 | dout(10) << __func__ << " " << new_dirname << "/" << new_filename << " " | |
3421 | << " " << file->fnode << dendl; | |
3422 | ||
3423 | new_dir->file_map[new_filename] = file; | |
3424 | old_dir->file_map.erase(old_filename); | |
3425 | ||
3426 | log_t.op_dir_link(new_dirname, new_filename, file->fnode.ino); | |
3427 | log_t.op_dir_unlink(old_dirname, old_filename); | |
3428 | return 0; | |
3429 | } | |
3430 | ||
3431 | int BlueFS::mkdir(const string& dirname) | |
3432 | { | |
11fdf7f2 | 3433 | std::lock_guard l(lock); |
7c673cae FG |
3434 | dout(10) << __func__ << " " << dirname << dendl; |
3435 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3436 | if (p != dir_map.end()) { | |
3437 | dout(20) << __func__ << " dir " << dirname << " exists" << dendl; | |
3438 | return -EEXIST; | |
3439 | } | |
9f95a23c | 3440 | dir_map[dirname] = ceph::make_ref<Dir>(); |
7c673cae FG |
3441 | log_t.op_dir_create(dirname); |
3442 | return 0; | |
3443 | } | |
3444 | ||
3445 | int BlueFS::rmdir(const string& dirname) | |
3446 | { | |
11fdf7f2 | 3447 | std::lock_guard l(lock); |
7c673cae FG |
3448 | dout(10) << __func__ << " " << dirname << dendl; |
3449 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3450 | if (p == dir_map.end()) { | |
3451 | dout(20) << __func__ << " dir " << dirname << " does not exist" << dendl; | |
3452 | return -ENOENT; | |
3453 | } | |
3454 | DirRef dir = p->second; | |
3455 | if (!dir->file_map.empty()) { | |
3456 | dout(20) << __func__ << " dir " << dirname << " not empty" << dendl; | |
3457 | return -ENOTEMPTY; | |
3458 | } | |
3459 | dir_map.erase(dirname); | |
3460 | log_t.op_dir_remove(dirname); | |
3461 | return 0; | |
3462 | } | |
3463 | ||
3464 | bool BlueFS::dir_exists(const string& dirname) | |
3465 | { | |
11fdf7f2 | 3466 | std::lock_guard l(lock); |
7c673cae FG |
3467 | map<string,DirRef>::iterator p = dir_map.find(dirname); |
3468 | bool exists = p != dir_map.end(); | |
3469 | dout(10) << __func__ << " " << dirname << " = " << (int)exists << dendl; | |
3470 | return exists; | |
3471 | } | |
3472 | ||
3473 | int BlueFS::stat(const string& dirname, const string& filename, | |
3474 | uint64_t *size, utime_t *mtime) | |
3475 | { | |
11fdf7f2 | 3476 | std::lock_guard l(lock); |
7c673cae FG |
3477 | dout(10) << __func__ << " " << dirname << "/" << filename << dendl; |
3478 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3479 | if (p == dir_map.end()) { | |
3480 | dout(20) << __func__ << " dir " << dirname << " not found" << dendl; | |
3481 | return -ENOENT; | |
3482 | } | |
3483 | DirRef dir = p->second; | |
3484 | map<string,FileRef>::iterator q = dir->file_map.find(filename); | |
3485 | if (q == dir->file_map.end()) { | |
3486 | dout(20) << __func__ << " dir " << dirname << " (" << dir | |
3487 | << ") file " << filename | |
3488 | << " not found" << dendl; | |
3489 | return -ENOENT; | |
3490 | } | |
3491 | File *file = q->second.get(); | |
3492 | dout(10) << __func__ << " " << dirname << "/" << filename | |
3493 | << " " << file->fnode << dendl; | |
3494 | if (size) | |
3495 | *size = file->fnode.size; | |
3496 | if (mtime) | |
3497 | *mtime = file->fnode.mtime; | |
3498 | return 0; | |
3499 | } | |
3500 | ||
3501 | int BlueFS::lock_file(const string& dirname, const string& filename, | |
3502 | FileLock **plock) | |
3503 | { | |
11fdf7f2 | 3504 | std::lock_guard l(lock); |
7c673cae FG |
3505 | dout(10) << __func__ << " " << dirname << "/" << filename << dendl; |
3506 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3507 | if (p == dir_map.end()) { | |
3508 | dout(20) << __func__ << " dir " << dirname << " not found" << dendl; | |
3509 | return -ENOENT; | |
3510 | } | |
3511 | DirRef dir = p->second; | |
3512 | map<string,FileRef>::iterator q = dir->file_map.find(filename); | |
9f95a23c | 3513 | FileRef file; |
7c673cae FG |
3514 | if (q == dir->file_map.end()) { |
3515 | dout(20) << __func__ << " dir " << dirname << " (" << dir | |
3516 | << ") file " << filename | |
3517 | << " not found, creating" << dendl; | |
9f95a23c | 3518 | file = ceph::make_ref<File>(); |
7c673cae FG |
3519 | file->fnode.ino = ++ino_last; |
3520 | file->fnode.mtime = ceph_clock_now(); | |
3521 | file_map[ino_last] = file; | |
3522 | dir->file_map[filename] = file; | |
3523 | ++file->refs; | |
3524 | log_t.op_file_update(file->fnode); | |
3525 | log_t.op_dir_link(dirname, filename, file->fnode.ino); | |
3526 | } else { | |
9f95a23c | 3527 | file = q->second; |
7c673cae FG |
3528 | if (file->locked) { |
3529 | dout(10) << __func__ << " already locked" << dendl; | |
11fdf7f2 | 3530 | return -ENOLCK; |
7c673cae FG |
3531 | } |
3532 | } | |
3533 | file->locked = true; | |
3534 | *plock = new FileLock(file); | |
3535 | dout(10) << __func__ << " locked " << file->fnode | |
3536 | << " with " << *plock << dendl; | |
3537 | return 0; | |
3538 | } | |
3539 | ||
3540 | int BlueFS::unlock_file(FileLock *fl) | |
3541 | { | |
11fdf7f2 | 3542 | std::lock_guard l(lock); |
7c673cae | 3543 | dout(10) << __func__ << " " << fl << " on " << fl->file->fnode << dendl; |
11fdf7f2 | 3544 | ceph_assert(fl->file->locked); |
7c673cae FG |
3545 | fl->file->locked = false; |
3546 | delete fl; | |
3547 | return 0; | |
3548 | } | |
3549 | ||
3550 | int BlueFS::readdir(const string& dirname, vector<string> *ls) | |
3551 | { | |
11fdf7f2 | 3552 | std::lock_guard l(lock); |
7c673cae FG |
3553 | dout(10) << __func__ << " " << dirname << dendl; |
3554 | if (dirname.empty()) { | |
3555 | // list dirs | |
3556 | ls->reserve(dir_map.size() + 2); | |
3557 | for (auto& q : dir_map) { | |
3558 | ls->push_back(q.first); | |
3559 | } | |
3560 | } else { | |
3561 | // list files in dir | |
3562 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3563 | if (p == dir_map.end()) { | |
3564 | dout(20) << __func__ << " dir " << dirname << " not found" << dendl; | |
3565 | return -ENOENT; | |
3566 | } | |
3567 | DirRef dir = p->second; | |
3568 | ls->reserve(dir->file_map.size() + 2); | |
3569 | for (auto& q : dir->file_map) { | |
3570 | ls->push_back(q.first); | |
3571 | } | |
3572 | } | |
3573 | ls->push_back("."); | |
3574 | ls->push_back(".."); | |
3575 | return 0; | |
3576 | } | |
3577 | ||
3578 | int BlueFS::unlink(const string& dirname, const string& filename) | |
3579 | { | |
11fdf7f2 | 3580 | std::lock_guard l(lock); |
7c673cae FG |
3581 | dout(10) << __func__ << " " << dirname << "/" << filename << dendl; |
3582 | map<string,DirRef>::iterator p = dir_map.find(dirname); | |
3583 | if (p == dir_map.end()) { | |
3584 | dout(20) << __func__ << " dir " << dirname << " not found" << dendl; | |
3585 | return -ENOENT; | |
3586 | } | |
3587 | DirRef dir = p->second; | |
3588 | map<string,FileRef>::iterator q = dir->file_map.find(filename); | |
3589 | if (q == dir->file_map.end()) { | |
3590 | dout(20) << __func__ << " file " << dirname << "/" << filename | |
3591 | << " not found" << dendl; | |
3592 | return -ENOENT; | |
3593 | } | |
3594 | FileRef file = q->second; | |
3595 | if (file->locked) { | |
3596 | dout(20) << __func__ << " file " << dirname << "/" << filename | |
3597 | << " is locked" << dendl; | |
3598 | return -EBUSY; | |
3599 | } | |
3600 | dir->file_map.erase(filename); | |
3601 | log_t.op_dir_unlink(dirname, filename); | |
3602 | _drop_link(file); | |
3603 | return 0; | |
3604 | } | |
d2e6a577 FG |
3605 | |
3606 | bool BlueFS::wal_is_rotational() | |
3607 | { | |
94b18763 FG |
3608 | if (bdev[BDEV_WAL]) { |
3609 | return bdev[BDEV_WAL]->is_rotational(); | |
3610 | } else if (bdev[BDEV_DB]) { | |
3611 | return bdev[BDEV_DB]->is_rotational(); | |
3612 | } | |
3613 | return bdev[BDEV_SLOW]->is_rotational(); | |
d2e6a577 | 3614 | } |
9f95a23c | 3615 | |
f6b5b4d7 TL |
3616 | /* |
3617 | Algorithm. | |
3618 | do_replay_recovery_read is used when bluefs log abruptly ends, but it seems that more data should be there. | |
3619 | Idea is to search disk for definiton of extents that will be accompanied with bluefs log in future, | |
3620 | and try if using it will produce healthy bluefs transaction. | |
3621 | We encode already known bluefs log extents and search disk for these bytes. | |
3622 | When we find it, we decode following bytes as extent. | |
3623 | We read that whole extent and then check if merged with existing log part gives a proper bluefs transaction. | |
3624 | */ | |
3625 | int BlueFS::do_replay_recovery_read(FileReader *log_reader, | |
3626 | size_t replay_pos, | |
3627 | size_t read_offset, | |
3628 | size_t read_len, | |
3629 | bufferlist* bl) { | |
3630 | dout(1) << __func__ << " replay_pos=0x" << std::hex << replay_pos << | |
3631 | " needs 0x" << read_offset << "~" << read_len << std::dec << dendl; | |
3632 | ||
3633 | bluefs_fnode_t& log_fnode = log_reader->file->fnode; | |
3634 | bufferlist bin_extents; | |
3635 | ceph::encode(log_fnode.extents, bin_extents); | |
3636 | dout(2) << __func__ << " log file encoded extents length = " << bin_extents.length() << dendl; | |
3637 | ||
3638 | // cannot process if too small to effectively search | |
3639 | ceph_assert(bin_extents.length() >= 32); | |
3640 | bufferlist last_32; | |
3641 | last_32.substr_of(bin_extents, bin_extents.length() - 32, 32); | |
3642 | ||
3643 | //read fixed part from replay_pos to end of bluefs_log extents | |
3644 | bufferlist fixed; | |
3645 | uint64_t e_off = 0; | |
3646 | auto e = log_fnode.seek(replay_pos, &e_off); | |
3647 | ceph_assert(e != log_fnode.extents.end()); | |
3648 | int r = bdev[e->bdev]->read(e->offset + e_off, e->length - e_off, &fixed, ioc[e->bdev], | |
3649 | cct->_conf->bluefs_buffered_io); | |
3650 | ceph_assert(r == 0); | |
3651 | //capture dev of last good extent | |
3652 | uint8_t last_e_dev = e->bdev; | |
3653 | uint64_t last_e_off = e->offset; | |
3654 | ++e; | |
3655 | while (e != log_fnode.extents.end()) { | |
3656 | r = bdev[e->bdev]->read(e->offset, e->length, &fixed, ioc[e->bdev], | |
3657 | cct->_conf->bluefs_buffered_io); | |
3658 | ceph_assert(r == 0); | |
3659 | last_e_dev = e->bdev; | |
3660 | ++e; | |
3661 | } | |
3662 | ceph_assert(replay_pos + fixed.length() == read_offset); | |
3663 | ||
3664 | dout(2) << __func__ << " valid data in log = " << fixed.length() << dendl; | |
3665 | ||
3666 | struct compare { | |
3667 | bool operator()(const bluefs_extent_t& a, const bluefs_extent_t& b) const { | |
3668 | if (a.bdev < b.bdev) return true; | |
3669 | if (a.offset < b.offset) return true; | |
3670 | return a.length < b.length; | |
3671 | } | |
3672 | }; | |
3673 | std::set<bluefs_extent_t, compare> extents_rejected; | |
3674 | for (int dcnt = 0; dcnt < 3; dcnt++) { | |
3675 | uint8_t dev = (last_e_dev + dcnt) % MAX_BDEV; | |
3676 | if (bdev[dev] == nullptr) continue; | |
3677 | dout(2) << __func__ << " processing " << get_device_name(dev) << dendl; | |
3678 | interval_set<uint64_t> disk_regions; | |
3679 | disk_regions.insert(0, bdev[dev]->get_size()); | |
3680 | for (auto f : file_map) { | |
3681 | auto& e = f.second->fnode.extents; | |
3682 | for (auto& p : e) { | |
3683 | if (p.bdev == dev) { | |
3684 | disk_regions.erase(p.offset, p.length); | |
3685 | } | |
3686 | } | |
3687 | } | |
3688 | size_t disk_regions_count = disk_regions.num_intervals(); | |
3689 | dout(5) << __func__ << " " << disk_regions_count << " regions to scan on " << get_device_name(dev) << dendl; | |
3690 | ||
3691 | auto reg = disk_regions.lower_bound(last_e_off); | |
3692 | //for all except first, start from beginning | |
3693 | last_e_off = 0; | |
3694 | if (reg == disk_regions.end()) { | |
3695 | reg = disk_regions.begin(); | |
3696 | } | |
3697 | const uint64_t chunk_size = 4 * 1024 * 1024; | |
3698 | const uint64_t page_size = 4096; | |
3699 | const uint64_t max_extent_size = 16; | |
3700 | uint64_t overlay_size = last_32.length() + max_extent_size; | |
3701 | for (size_t i = 0; i < disk_regions_count; reg++, i++) { | |
3702 | if (reg == disk_regions.end()) { | |
3703 | reg = disk_regions.begin(); | |
3704 | } | |
3705 | uint64_t pos = reg.get_start(); | |
3706 | uint64_t len = reg.get_len(); | |
3707 | ||
3708 | std::unique_ptr<char[]> raw_data_p{new char[page_size + chunk_size]}; | |
3709 | char* raw_data = raw_data_p.get(); | |
3710 | memset(raw_data, 0, page_size); | |
3711 | ||
3712 | while (len > last_32.length()) { | |
3713 | uint64_t chunk_len = len > chunk_size ? chunk_size : len; | |
3714 | dout(5) << __func__ << " read " | |
3715 | << get_device_name(dev) << ":0x" << std::hex << pos << "+" << chunk_len << std::dec << dendl; | |
3716 | r = bdev[dev]->read_random(pos, chunk_len, raw_data + page_size, cct->_conf->bluefs_buffered_io); | |
3717 | ceph_assert(r == 0); | |
3718 | ||
3719 | //search for fixed_last_32 | |
3720 | char* chunk_b = raw_data + page_size; | |
3721 | char* chunk_e = chunk_b + chunk_len; | |
3722 | ||
3723 | char* search_b = chunk_b - overlay_size; | |
3724 | char* search_e = chunk_e; | |
3725 | ||
3726 | for (char* sp = search_b; ; sp += last_32.length()) { | |
3727 | sp = (char*)memmem(sp, search_e - sp, last_32.c_str(), last_32.length()); | |
3728 | if (sp == nullptr) { | |
3729 | break; | |
3730 | } | |
3731 | ||
3732 | char* n = sp + last_32.length(); | |
3733 | dout(5) << __func__ << " checking location 0x" << std::hex << pos + (n - chunk_b) << std::dec << dendl; | |
3734 | bufferlist test; | |
3735 | test.append(n, std::min<size_t>(max_extent_size, chunk_e - n)); | |
3736 | bluefs_extent_t ne; | |
3737 | try { | |
3738 | bufferlist::const_iterator p = test.begin(); | |
3739 | ceph::decode(ne, p); | |
3740 | } catch (buffer::error& e) { | |
3741 | continue; | |
3742 | } | |
3743 | if (extents_rejected.count(ne) != 0) { | |
3744 | dout(5) << __func__ << " extent " << ne << " already refected" <<dendl; | |
3745 | continue; | |
3746 | } | |
3747 | //insert as rejected already. if we succeed, it wouldn't make difference. | |
3748 | extents_rejected.insert(ne); | |
3749 | ||
3750 | if (ne.bdev >= MAX_BDEV || | |
3751 | bdev[ne.bdev] == nullptr || | |
3752 | ne.length > 16 * 1024 * 1024 || | |
3753 | (ne.length & 4095) != 0 || | |
3754 | ne.offset + ne.length > bdev[ne.bdev]->get_size() || | |
3755 | (ne.offset & 4095) != 0) { | |
3756 | dout(5) << __func__ << " refusing extent " << ne << dendl; | |
3757 | continue; | |
3758 | } | |
3759 | dout(5) << __func__ << " checking extent " << ne << dendl; | |
3760 | ||
3761 | //read candidate extent - whole | |
3762 | bufferlist candidate; | |
3763 | candidate.append(fixed); | |
3764 | r = bdev[ne.bdev]->read(ne.offset, ne.length, &candidate, ioc[ne.bdev], | |
3765 | cct->_conf->bluefs_buffered_io); | |
3766 | ceph_assert(r == 0); | |
3767 | ||
3768 | //check if transaction & crc is ok | |
3769 | bluefs_transaction_t t; | |
3770 | try { | |
3771 | bufferlist::const_iterator p = candidate.cbegin(); | |
3772 | decode(t, p); | |
3773 | } | |
3774 | catch (buffer::error& e) { | |
3775 | dout(5) << __func__ << " failed match" << dendl; | |
3776 | continue; | |
3777 | } | |
3778 | ||
3779 | //success, it seems a probable candidate | |
3780 | uint64_t l = std::min<uint64_t>(ne.length, read_len); | |
3781 | //trim to required size | |
3782 | bufferlist requested_read; | |
3783 | requested_read.substr_of(candidate, fixed.length(), l); | |
3784 | bl->append(requested_read); | |
3785 | dout(5) << __func__ << " successful extension of log " << l << "/" << read_len << dendl; | |
3786 | log_fnode.append_extent(ne); | |
3787 | log_fnode.recalc_allocated(); | |
3788 | log_reader->buf.pos += l; | |
3789 | return l; | |
3790 | } | |
3791 | //save overlay for next search | |
3792 | memcpy(search_b, chunk_e - overlay_size, overlay_size); | |
3793 | pos += chunk_len; | |
3794 | len -= chunk_len; | |
3795 | } | |
3796 | } | |
3797 | } | |
3798 | return 0; | |
3799 | } | |
3800 | ||
9f95a23c TL |
3801 | void BlueFS::debug_inject_duplicate_gift(unsigned id, |
3802 | uint64_t offset, | |
3803 | uint64_t len) | |
3804 | { | |
3805 | dout(0) << __func__ << dendl; | |
3806 | if (id < alloc.size() && alloc[id]) { | |
3807 | alloc[id]->init_add_free(offset, len); | |
3808 | } | |
3809 | } | |
3810 | ||
3811 | // =============================================== | |
3812 | // OriginalVolumeSelector | |
3813 | ||
f6b5b4d7 TL |
3814 | void* OriginalVolumeSelector::get_hint_for_log() const { |
3815 | return reinterpret_cast<void*>(BlueFS::BDEV_WAL); | |
9f95a23c TL |
3816 | } |
3817 | void* OriginalVolumeSelector::get_hint_by_dir(const string& dirname) const { | |
3818 | uint8_t res = BlueFS::BDEV_DB; | |
3819 | if (dirname.length() > 5) { | |
3820 | // the "db.slow" and "db.wal" directory names are hard-coded at | |
3821 | // match up with bluestore. the slow device is always the second | |
3822 | // one (when a dedicated block.db device is present and used at | |
3823 | // bdev 0). the wal device is always last. | |
3824 | if (boost::algorithm::ends_with(dirname, ".slow")) { | |
3825 | res = BlueFS::BDEV_SLOW; | |
3826 | } | |
3827 | else if (boost::algorithm::ends_with(dirname, ".wal")) { | |
3828 | res = BlueFS::BDEV_WAL; | |
3829 | } | |
3830 | } | |
3831 | return reinterpret_cast<void*>(res); | |
3832 | } | |
3833 | ||
3834 | uint8_t OriginalVolumeSelector::select_prefer_bdev(void* hint) | |
3835 | { | |
3836 | return (uint8_t)(reinterpret_cast<uint64_t>(hint)); | |
3837 | } | |
3838 | ||
3839 | void OriginalVolumeSelector::get_paths(const std::string& base, paths& res) const | |
3840 | { | |
3841 | res.emplace_back(base, db_total); | |
3842 | res.emplace_back(base + ".slow", slow_total); | |
3843 | } | |
3844 | ||
3845 | #undef dout_prefix | |
3846 | #define dout_prefix *_dout << "OriginalVolumeSelector: " | |
3847 | ||
3848 | void OriginalVolumeSelector::dump(ostream& sout) { | |
3849 | sout<< "wal_total:" << wal_total | |
3850 | << ", db_total:" << db_total | |
3851 | << ", slow_total:" << slow_total | |
3852 | << std::endl; | |
3853 | } |