]>
Commit | Line | Data |
---|---|---|
7c673cae FG |
1 | // -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- |
2 | // vim: ts=8 sw=2 smarttab | |
3 | /* | |
4 | * Ceph - scalable distributed file system | |
5 | * | |
6 | * Copyright (C) 2015 Red Hat | |
7 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
11fdf7f2 | 15 | #include <string_view> |
94b18763 | 16 | |
7c673cae FG |
17 | #include "common/debug.h" |
18 | #include "common/errno.h" | |
19 | ||
20 | #include "messages/MClientRequestForward.h" | |
21 | #include "messages/MMDSLoadTargets.h" | |
7c673cae | 22 | #include "messages/MMDSTableRequest.h" |
7c673cae | 23 | |
9f95a23c TL |
24 | #include "mgr/MgrClient.h" |
25 | ||
7c673cae FG |
26 | #include "MDSDaemon.h" |
27 | #include "MDSMap.h" | |
28 | #include "SnapClient.h" | |
29 | #include "SnapServer.h" | |
30 | #include "MDBalancer.h" | |
91327a77 | 31 | #include "Migrator.h" |
7c673cae | 32 | #include "Locker.h" |
7c673cae FG |
33 | #include "InoTable.h" |
34 | #include "mon/MonClient.h" | |
35 | #include "common/HeartbeatMap.h" | |
36 | #include "ScrubStack.h" | |
37 | ||
38 | ||
39 | #include "MDSRank.h" | |
40 | ||
41 | #define dout_context g_ceph_context | |
42 | #define dout_subsys ceph_subsys_mds | |
43 | #undef dout_prefix | |
44 | #define dout_prefix *_dout << "mds." << whoami << '.' << incarnation << ' ' | |
9f95a23c | 45 | using TOPNSPC::common::cmd_getval; |
f64942e4 AA |
46 | class C_Flush_Journal : public MDSInternalContext { |
47 | public: | |
48 | C_Flush_Journal(MDCache *mdcache, MDLog *mdlog, MDSRank *mds, | |
49 | std::ostream *ss, Context *on_finish) | |
50 | : MDSInternalContext(mds), | |
51 | mdcache(mdcache), mdlog(mdlog), ss(ss), on_finish(on_finish), | |
52 | whoami(mds->whoami), incarnation(mds->incarnation) { | |
53 | } | |
54 | ||
55 | void send() { | |
9f95a23c | 56 | assert(ceph_mutex_is_locked(mds->mds_lock)); |
f64942e4 AA |
57 | |
58 | dout(20) << __func__ << dendl; | |
59 | ||
60 | if (mdcache->is_readonly()) { | |
61 | dout(5) << __func__ << ": read-only FS" << dendl; | |
62 | complete(-EROFS); | |
63 | return; | |
64 | } | |
65 | ||
66 | if (!mds->is_active()) { | |
67 | dout(5) << __func__ << ": MDS not active, no-op" << dendl; | |
68 | complete(0); | |
69 | return; | |
70 | } | |
71 | ||
72 | flush_mdlog(); | |
73 | } | |
74 | ||
75 | private: | |
76 | ||
77 | void flush_mdlog() { | |
78 | dout(20) << __func__ << dendl; | |
79 | ||
80 | // I need to seal off the current segment, and then mark all | |
81 | // previous segments for expiry | |
82 | mdlog->start_new_segment(); | |
83 | ||
9f95a23c | 84 | Context *ctx = new LambdaContext([this](int r) { |
f64942e4 AA |
85 | handle_flush_mdlog(r); |
86 | }); | |
87 | ||
88 | // Flush initially so that all the segments older than our new one | |
89 | // will be elegible for expiry | |
90 | mdlog->flush(); | |
91 | mdlog->wait_for_safe(new MDSInternalContextWrapper(mds, ctx)); | |
92 | } | |
93 | ||
94 | void handle_flush_mdlog(int r) { | |
95 | dout(20) << __func__ << ": r=" << r << dendl; | |
96 | ||
97 | if (r != 0) { | |
98 | *ss << "Error " << r << " (" << cpp_strerror(r) << ") while flushing journal"; | |
99 | complete(r); | |
100 | return; | |
101 | } | |
102 | ||
103 | clear_mdlog(); | |
104 | } | |
105 | ||
106 | void clear_mdlog() { | |
107 | dout(20) << __func__ << dendl; | |
108 | ||
9f95a23c | 109 | Context *ctx = new LambdaContext([this](int r) { |
f64942e4 AA |
110 | handle_clear_mdlog(r); |
111 | }); | |
112 | ||
113 | // Because we may not be the last wait_for_safe context on MDLog, | |
114 | // and subsequent contexts might wake up in the middle of our | |
115 | // later trim_all and interfere with expiry (by e.g. marking | |
116 | // dirs/dentries dirty on previous log segments), we run a second | |
117 | // wait_for_safe here. See #10368 | |
118 | mdlog->wait_for_safe(new MDSInternalContextWrapper(mds, ctx)); | |
119 | } | |
120 | ||
121 | void handle_clear_mdlog(int r) { | |
122 | dout(20) << __func__ << ": r=" << r << dendl; | |
123 | ||
124 | if (r != 0) { | |
125 | *ss << "Error " << r << " (" << cpp_strerror(r) << ") while flushing journal"; | |
126 | complete(r); | |
127 | return; | |
128 | } | |
129 | ||
130 | trim_mdlog(); | |
131 | } | |
132 | ||
133 | void trim_mdlog() { | |
134 | // Put all the old log segments into expiring or expired state | |
135 | dout(5) << __func__ << ": beginning segment expiry" << dendl; | |
136 | ||
137 | int ret = mdlog->trim_all(); | |
138 | if (ret != 0) { | |
139 | *ss << "Error " << ret << " (" << cpp_strerror(ret) << ") while trimming log"; | |
140 | complete(ret); | |
141 | return; | |
142 | } | |
143 | ||
144 | expire_segments(); | |
145 | } | |
146 | ||
147 | void expire_segments() { | |
148 | dout(20) << __func__ << dendl; | |
149 | ||
150 | // Attach contexts to wait for all expiring segments to expire | |
151 | MDSGatherBuilder *expiry_gather = new MDSGatherBuilder(g_ceph_context); | |
152 | ||
153 | const auto &expiring_segments = mdlog->get_expiring_segments(); | |
154 | for (auto p : expiring_segments) { | |
155 | p->wait_for_expiry(expiry_gather->new_sub()); | |
156 | } | |
157 | dout(5) << __func__ << ": waiting for " << expiry_gather->num_subs_created() | |
158 | << " segments to expire" << dendl; | |
159 | ||
160 | if (!expiry_gather->has_subs()) { | |
161 | trim_segments(); | |
162 | delete expiry_gather; | |
163 | return; | |
164 | } | |
165 | ||
9f95a23c | 166 | Context *ctx = new LambdaContext([this](int r) { |
f64942e4 AA |
167 | handle_expire_segments(r); |
168 | }); | |
169 | expiry_gather->set_finisher(new MDSInternalContextWrapper(mds, ctx)); | |
170 | expiry_gather->activate(); | |
171 | } | |
172 | ||
173 | void handle_expire_segments(int r) { | |
174 | dout(20) << __func__ << ": r=" << r << dendl; | |
175 | ||
176 | ceph_assert(r == 0); // MDLog is not allowed to raise errors via | |
177 | // wait_for_expiry | |
178 | trim_segments(); | |
179 | } | |
180 | ||
181 | void trim_segments() { | |
182 | dout(20) << __func__ << dendl; | |
183 | ||
9f95a23c | 184 | Context *ctx = new C_OnFinisher(new LambdaContext([this](int) { |
11fdf7f2 | 185 | std::lock_guard locker(mds->mds_lock); |
f64942e4 AA |
186 | trim_expired_segments(); |
187 | }), mds->finisher); | |
188 | ctx->complete(0); | |
189 | } | |
190 | ||
191 | void trim_expired_segments() { | |
192 | dout(5) << __func__ << ": expiry complete, expire_pos/trim_pos is now " | |
193 | << std::hex << mdlog->get_journaler()->get_expire_pos() << "/" | |
194 | << mdlog->get_journaler()->get_trimmed_pos() << dendl; | |
195 | ||
196 | // Now everyone I'm interested in is expired | |
197 | mdlog->trim_expired_segments(); | |
198 | ||
199 | dout(5) << __func__ << ": trim complete, expire_pos/trim_pos is now " | |
200 | << std::hex << mdlog->get_journaler()->get_expire_pos() << "/" | |
201 | << mdlog->get_journaler()->get_trimmed_pos() << dendl; | |
202 | ||
203 | write_journal_head(); | |
204 | } | |
205 | ||
206 | void write_journal_head() { | |
207 | dout(20) << __func__ << dendl; | |
208 | ||
9f95a23c | 209 | Context *ctx = new LambdaContext([this](int r) { |
11fdf7f2 | 210 | std::lock_guard locker(mds->mds_lock); |
f64942e4 AA |
211 | handle_write_head(r); |
212 | }); | |
213 | // Flush the journal header so that readers will start from after | |
214 | // the flushed region | |
215 | mdlog->get_journaler()->write_head(ctx); | |
216 | } | |
217 | ||
218 | void handle_write_head(int r) { | |
219 | if (r != 0) { | |
220 | *ss << "Error " << r << " (" << cpp_strerror(r) << ") while writing header"; | |
221 | } else { | |
222 | dout(5) << __func__ << ": write_head complete, all done!" << dendl; | |
223 | } | |
224 | ||
225 | complete(r); | |
226 | } | |
227 | ||
228 | void finish(int r) override { | |
229 | dout(20) << __func__ << ": r=" << r << dendl; | |
230 | on_finish->complete(r); | |
231 | } | |
232 | ||
233 | MDCache *mdcache; | |
234 | MDLog *mdlog; | |
235 | std::ostream *ss; | |
236 | Context *on_finish; | |
237 | ||
238 | // so as to use dout | |
239 | mds_rank_t whoami; | |
240 | int incarnation; | |
241 | }; | |
242 | ||
243 | class C_Drop_Cache : public MDSInternalContext { | |
244 | public: | |
245 | C_Drop_Cache(Server *server, MDCache *mdcache, MDLog *mdlog, | |
246 | MDSRank *mds, uint64_t recall_timeout, | |
247 | Formatter *f, Context *on_finish) | |
248 | : MDSInternalContext(mds), | |
249 | server(server), mdcache(mdcache), mdlog(mdlog), | |
a8e16298 TL |
250 | recall_timeout(recall_timeout), recall_start(mono_clock::now()), |
251 | f(f), on_finish(on_finish), | |
f64942e4 AA |
252 | whoami(mds->whoami), incarnation(mds->incarnation) { |
253 | } | |
254 | ||
255 | void send() { | |
256 | // not really a hard requirement here, but lets ensure this in | |
257 | // case we change the logic here. | |
9f95a23c | 258 | assert(ceph_mutex_is_locked(mds->mds_lock)); |
f64942e4 AA |
259 | |
260 | dout(20) << __func__ << dendl; | |
a8e16298 | 261 | f->open_object_section("result"); |
f64942e4 AA |
262 | recall_client_state(); |
263 | } | |
264 | ||
265 | private: | |
266 | // context which completes itself (with -ETIMEDOUT) after a specified | |
267 | // timeout or when explicitly completed, whichever comes first. Note | |
268 | // that the context does not detroy itself after completion -- it | |
269 | // needs to be explicitly freed. | |
270 | class C_ContextTimeout : public MDSInternalContext { | |
271 | public: | |
272 | C_ContextTimeout(MDSRank *mds, uint64_t timeout, Context *on_finish) | |
273 | : MDSInternalContext(mds), | |
274 | timeout(timeout), | |
f64942e4 AA |
275 | on_finish(on_finish) { |
276 | } | |
277 | ~C_ContextTimeout() { | |
278 | ceph_assert(timer_task == nullptr); | |
279 | } | |
280 | ||
281 | void start_timer() { | |
282 | if (!timeout) { | |
283 | return; | |
284 | } | |
285 | ||
9f95a23c | 286 | timer_task = new LambdaContext([this](int) { |
f64942e4 AA |
287 | timer_task = nullptr; |
288 | complete(-ETIMEDOUT); | |
289 | }); | |
290 | mds->timer.add_event_after(timeout, timer_task); | |
291 | } | |
292 | ||
293 | void finish(int r) override { | |
294 | Context *ctx = nullptr; | |
295 | { | |
11fdf7f2 | 296 | std::lock_guard locker(lock); |
f64942e4 AA |
297 | std::swap(on_finish, ctx); |
298 | } | |
299 | if (ctx != nullptr) { | |
300 | ctx->complete(r); | |
301 | } | |
302 | } | |
303 | void complete(int r) override { | |
304 | if (timer_task != nullptr) { | |
305 | mds->timer.cancel_event(timer_task); | |
306 | } | |
307 | ||
308 | finish(r); | |
309 | } | |
310 | ||
311 | uint64_t timeout; | |
9f95a23c | 312 | ceph::mutex lock = ceph::make_mutex("mds::context::timeout"); |
f64942e4 AA |
313 | Context *on_finish = nullptr; |
314 | Context *timer_task = nullptr; | |
315 | }; | |
316 | ||
11fdf7f2 TL |
317 | auto do_trim() { |
318 | auto [throttled, count] = mdcache->trim(UINT64_MAX); | |
319 | dout(10) << __func__ | |
320 | << (throttled ? " (throttled)" : "") | |
321 | << " trimmed " << count << " caps" << dendl; | |
322 | dentries_trimmed += count; | |
323 | return std::make_pair(throttled, count); | |
324 | } | |
325 | ||
f64942e4 AA |
326 | void recall_client_state() { |
327 | dout(20) << __func__ << dendl; | |
a8e16298 TL |
328 | auto now = mono_clock::now(); |
329 | auto duration = std::chrono::duration<double>(now-recall_start).count(); | |
f64942e4 AA |
330 | |
331 | MDSGatherBuilder *gather = new MDSGatherBuilder(g_ceph_context); | |
92f5a8d4 TL |
332 | auto flags = Server::RecallFlags::STEADY|Server::RecallFlags::TRIM; |
333 | auto [throttled, count] = server->recall_client_state(gather, flags); | |
a8e16298 TL |
334 | dout(10) << __func__ |
335 | << (throttled ? " (throttled)" : "") | |
336 | << " recalled " << count << " caps" << dendl; | |
337 | ||
338 | caps_recalled += count; | |
339 | if ((throttled || count > 0) && (recall_timeout == 0 || duration < recall_timeout)) { | |
11fdf7f2 | 340 | C_ContextTimeout *ctx = new C_ContextTimeout( |
9f95a23c | 341 | mds, 1, new LambdaContext([this](int r) { |
11fdf7f2 TL |
342 | recall_client_state(); |
343 | })); | |
344 | ctx->start_timer(); | |
345 | gather->set_finisher(new MDSInternalContextWrapper(mds, ctx)); | |
346 | gather->activate(); | |
347 | mdlog->flush(); /* use down-time to incrementally flush log */ | |
348 | do_trim(); /* use down-time to incrementally trim cache */ | |
a8e16298 TL |
349 | } else { |
350 | if (!gather->has_subs()) { | |
351 | delete gather; | |
352 | return handle_recall_client_state(0); | |
353 | } else if (recall_timeout > 0 && duration > recall_timeout) { | |
11fdf7f2 TL |
354 | gather->set_finisher(new C_MDSInternalNoop); |
355 | gather->activate(); | |
a8e16298 TL |
356 | return handle_recall_client_state(-ETIMEDOUT); |
357 | } else { | |
358 | uint64_t remaining = (recall_timeout == 0 ? 0 : recall_timeout-duration); | |
359 | C_ContextTimeout *ctx = new C_ContextTimeout( | |
9f95a23c | 360 | mds, remaining, new LambdaContext([this](int r) { |
a8e16298 TL |
361 | handle_recall_client_state(r); |
362 | })); | |
363 | ||
364 | ctx->start_timer(); | |
365 | gather->set_finisher(new MDSInternalContextWrapper(mds, ctx)); | |
366 | gather->activate(); | |
367 | } | |
f64942e4 | 368 | } |
f64942e4 AA |
369 | } |
370 | ||
371 | void handle_recall_client_state(int r) { | |
372 | dout(20) << __func__ << ": r=" << r << dendl; | |
373 | ||
374 | // client recall section | |
375 | f->open_object_section("client_recall"); | |
376 | f->dump_int("return_code", r); | |
377 | f->dump_string("message", cpp_strerror(r)); | |
a8e16298 | 378 | f->dump_int("recalled", caps_recalled); |
f64942e4 AA |
379 | f->close_section(); |
380 | ||
381 | // we can still continue after recall timeout | |
f64942e4 AA |
382 | flush_journal(); |
383 | } | |
384 | ||
385 | void flush_journal() { | |
386 | dout(20) << __func__ << dendl; | |
387 | ||
9f95a23c | 388 | Context *ctx = new LambdaContext([this](int r) { |
f64942e4 AA |
389 | handle_flush_journal(r); |
390 | }); | |
391 | ||
392 | C_Flush_Journal *flush_journal = new C_Flush_Journal(mdcache, mdlog, mds, &ss, ctx); | |
393 | flush_journal->send(); | |
394 | } | |
395 | ||
396 | void handle_flush_journal(int r) { | |
397 | dout(20) << __func__ << ": r=" << r << dendl; | |
398 | ||
399 | if (r != 0) { | |
400 | cmd_err(f, ss.str()); | |
401 | complete(r); | |
402 | return; | |
403 | } | |
404 | ||
405 | // journal flush section | |
406 | f->open_object_section("flush_journal"); | |
407 | f->dump_int("return_code", r); | |
408 | f->dump_string("message", ss.str()); | |
409 | f->close_section(); | |
410 | ||
a8e16298 TL |
411 | trim_cache(); |
412 | } | |
413 | ||
414 | void trim_cache() { | |
415 | dout(20) << __func__ << dendl; | |
416 | ||
11fdf7f2 | 417 | auto [throttled, count] = do_trim(); |
a8e16298 | 418 | if (throttled && count > 0) { |
9f95a23c | 419 | auto timer = new LambdaContext([this](int) { |
a8e16298 TL |
420 | trim_cache(); |
421 | }); | |
422 | mds->timer.add_event_after(1.0, timer); | |
423 | } else { | |
424 | cache_status(); | |
425 | } | |
f64942e4 AA |
426 | } |
427 | ||
428 | void cache_status() { | |
429 | dout(20) << __func__ << dendl; | |
430 | ||
a8e16298 TL |
431 | f->open_object_section("trim_cache"); |
432 | f->dump_int("trimmed", dentries_trimmed); | |
433 | f->close_section(); | |
434 | ||
f64942e4 AA |
435 | // cache status section |
436 | mdcache->cache_status(f); | |
f64942e4 AA |
437 | |
438 | complete(0); | |
439 | } | |
440 | ||
441 | void finish(int r) override { | |
442 | dout(20) << __func__ << ": r=" << r << dendl; | |
443 | ||
a8e16298 TL |
444 | auto d = std::chrono::duration<double>(mono_clock::now()-recall_start); |
445 | f->dump_float("duration", d.count()); | |
446 | ||
447 | f->close_section(); | |
f64942e4 AA |
448 | on_finish->complete(r); |
449 | } | |
450 | ||
451 | Server *server; | |
452 | MDCache *mdcache; | |
453 | MDLog *mdlog; | |
454 | uint64_t recall_timeout; | |
a8e16298 | 455 | mono_time recall_start; |
f64942e4 AA |
456 | Formatter *f; |
457 | Context *on_finish; | |
458 | ||
459 | int retval = 0; | |
460 | std::stringstream ss; | |
a8e16298 TL |
461 | uint64_t caps_recalled = 0; |
462 | uint64_t dentries_trimmed = 0; | |
f64942e4 AA |
463 | |
464 | // so as to use dout | |
465 | mds_rank_t whoami; | |
466 | int incarnation; | |
467 | ||
11fdf7f2 | 468 | void cmd_err(Formatter *f, std::string_view err) { |
f64942e4 AA |
469 | f->reset(); |
470 | f->open_object_section("result"); | |
471 | f->dump_string("error", err); | |
472 | f->close_section(); | |
473 | } | |
474 | }; | |
475 | ||
7c673cae FG |
476 | MDSRank::MDSRank( |
477 | mds_rank_t whoami_, | |
9f95a23c | 478 | ceph::mutex &mds_lock_, |
7c673cae FG |
479 | LogChannelRef &clog_, |
480 | SafeTimer &timer_, | |
481 | Beacon &beacon_, | |
11fdf7f2 | 482 | std::unique_ptr<MDSMap>& mdsmap_, |
7c673cae FG |
483 | Messenger *msgr, |
484 | MonClient *monc_, | |
9f95a23c | 485 | MgrClient *mgrc, |
7c673cae | 486 | Context *respawn_hook_, |
9f95a23c TL |
487 | Context *suicide_hook_) : |
488 | cct(msgr->cct), mds_lock(mds_lock_), clog(clog_), | |
489 | timer(timer_), mdsmap(mdsmap_), | |
7c673cae | 490 | objecter(new Objecter(g_ceph_context, msgr, monc_, nullptr, 0, 0)), |
9f95a23c | 491 | damage_table(whoami_), sessionmap(this), |
11fdf7f2 TL |
492 | op_tracker(g_ceph_context, g_conf()->mds_enable_op_tracker, |
493 | g_conf()->osd_num_op_tracker_shard), | |
9f95a23c | 494 | progress_thread(this), whoami(whoami_), |
7c673cae FG |
495 | purge_queue(g_ceph_context, whoami_, |
496 | mdsmap_->get_metadata_pool(), objecter, | |
9f95a23c TL |
497 | new LambdaContext([this](int r) { |
498 | std::lock_guard l(mds_lock); | |
499 | handle_write_error(r); | |
500 | } | |
7c673cae FG |
501 | ) |
502 | ), | |
9f95a23c TL |
503 | beacon(beacon_), |
504 | messenger(msgr), monc(monc_), mgrc(mgrc), | |
7c673cae FG |
505 | respawn_hook(respawn_hook_), |
506 | suicide_hook(suicide_hook_), | |
94b18763 | 507 | starttime(mono_clock::now()) |
7c673cae FG |
508 | { |
509 | hb = g_ceph_context->get_heartbeat_map()->add_worker("MDSRank", pthread_self()); | |
510 | ||
511 | purge_queue.update_op_limit(*mdsmap); | |
512 | ||
9f95a23c | 513 | objecter->unset_honor_pool_full(); |
7c673cae | 514 | |
9f95a23c | 515 | finisher = new Finisher(cct, "MDSRank", "MR_Finisher"); |
7c673cae FG |
516 | |
517 | mdcache = new MDCache(this, purge_queue); | |
518 | mdlog = new MDLog(this); | |
519 | balancer = new MDBalancer(this, messenger, monc); | |
520 | ||
9f95a23c | 521 | scrubstack = new ScrubStack(mdcache, clog, finisher); |
7c673cae FG |
522 | |
523 | inotable = new InoTable(this); | |
524 | snapserver = new SnapServer(this, monc); | |
525 | snapclient = new SnapClient(this); | |
526 | ||
527 | server = new Server(this); | |
528 | locker = new Locker(this, mdcache); | |
529 | ||
b32b8144 FG |
530 | op_tracker.set_complaint_and_threshold(cct->_conf->mds_op_complaint_time, |
531 | cct->_conf->mds_op_log_threshold); | |
532 | op_tracker.set_history_size_and_duration(cct->_conf->mds_op_history_size, | |
533 | cct->_conf->mds_op_history_duration); | |
9f95a23c TL |
534 | |
535 | schedule_update_timer_task(); | |
7c673cae FG |
536 | } |
537 | ||
538 | MDSRank::~MDSRank() | |
539 | { | |
540 | if (hb) { | |
541 | g_ceph_context->get_heartbeat_map()->remove_worker(hb); | |
542 | } | |
543 | ||
544 | if (scrubstack) { delete scrubstack; scrubstack = NULL; } | |
545 | if (mdcache) { delete mdcache; mdcache = NULL; } | |
546 | if (mdlog) { delete mdlog; mdlog = NULL; } | |
547 | if (balancer) { delete balancer; balancer = NULL; } | |
548 | if (inotable) { delete inotable; inotable = NULL; } | |
549 | if (snapserver) { delete snapserver; snapserver = NULL; } | |
550 | if (snapclient) { delete snapclient; snapclient = NULL; } | |
7c673cae FG |
551 | |
552 | if (server) { delete server; server = 0; } | |
553 | if (locker) { delete locker; locker = 0; } | |
554 | ||
555 | if (logger) { | |
556 | g_ceph_context->get_perfcounters_collection()->remove(logger); | |
557 | delete logger; | |
558 | logger = 0; | |
559 | } | |
560 | if (mlogger) { | |
561 | g_ceph_context->get_perfcounters_collection()->remove(mlogger); | |
562 | delete mlogger; | |
563 | mlogger = 0; | |
564 | } | |
565 | ||
566 | delete finisher; | |
567 | finisher = NULL; | |
568 | ||
569 | delete suicide_hook; | |
570 | suicide_hook = NULL; | |
571 | ||
572 | delete respawn_hook; | |
573 | respawn_hook = NULL; | |
574 | ||
575 | delete objecter; | |
576 | objecter = nullptr; | |
577 | } | |
578 | ||
579 | void MDSRankDispatcher::init() | |
580 | { | |
581 | objecter->init(); | |
582 | messenger->add_dispatcher_head(objecter); | |
583 | ||
584 | objecter->start(); | |
585 | ||
586 | update_log_config(); | |
587 | create_logger(); | |
588 | ||
589 | // Expose the OSDMap (already populated during MDS::init) to anyone | |
590 | // who is interested in it. | |
591 | handle_osd_map(); | |
592 | ||
593 | progress_thread.create("mds_rank_progr"); | |
594 | ||
595 | purge_queue.init(); | |
596 | ||
597 | finisher->start(); | |
598 | } | |
599 | ||
11fdf7f2 | 600 | void MDSRank::update_targets() |
7c673cae FG |
601 | { |
602 | // get MonMap's idea of my export_targets | |
603 | const set<mds_rank_t>& map_targets = mdsmap->get_mds_info(get_nodeid()).export_targets; | |
604 | ||
605 | dout(20) << "updating export targets, currently " << map_targets.size() << " ranks are targets" << dendl; | |
606 | ||
607 | bool send = false; | |
608 | set<mds_rank_t> new_map_targets; | |
609 | ||
610 | auto it = export_targets.begin(); | |
611 | while (it != export_targets.end()) { | |
612 | mds_rank_t rank = it->first; | |
11fdf7f2 TL |
613 | auto &counter = it->second; |
614 | dout(20) << "export target mds." << rank << " is " << counter << dendl; | |
7c673cae | 615 | |
11fdf7f2 | 616 | double val = counter.get(); |
7c673cae FG |
617 | if (val <= 0.01) { |
618 | dout(15) << "export target mds." << rank << " is no longer an export target" << dendl; | |
619 | export_targets.erase(it++); | |
620 | send = true; | |
621 | continue; | |
622 | } | |
623 | if (!map_targets.count(rank)) { | |
624 | dout(15) << "export target mds." << rank << " not in map's export_targets" << dendl; | |
625 | send = true; | |
626 | } | |
627 | new_map_targets.insert(rank); | |
628 | it++; | |
629 | } | |
630 | if (new_map_targets.size() < map_targets.size()) { | |
631 | dout(15) << "export target map holds stale targets, sending update" << dendl; | |
632 | send = true; | |
633 | } | |
634 | ||
635 | if (send) { | |
636 | dout(15) << "updating export_targets, now " << new_map_targets.size() << " ranks are targets" << dendl; | |
9f95a23c | 637 | auto m = make_message<MMDSLoadTargets>(mds_gid_t(monc->get_global_id()), new_map_targets); |
11fdf7f2 | 638 | monc->send_mon_message(m.detach()); |
7c673cae FG |
639 | } |
640 | } | |
641 | ||
11fdf7f2 | 642 | void MDSRank::hit_export_target(mds_rank_t rank, double amount) |
7c673cae | 643 | { |
11fdf7f2 | 644 | double rate = g_conf()->mds_bal_target_decay; |
7c673cae | 645 | if (amount < 0.0) { |
11fdf7f2 | 646 | amount = 100.0/g_conf()->mds_bal_target_decay; /* a good default for "i am trying to keep this export_target active" */ |
7c673cae | 647 | } |
11fdf7f2 TL |
648 | auto em = export_targets.emplace(std::piecewise_construct, std::forward_as_tuple(rank), std::forward_as_tuple(DecayRate(rate))); |
649 | auto &counter = em.first->second; | |
650 | counter.hit(amount); | |
7c673cae | 651 | if (em.second) { |
11fdf7f2 TL |
652 | dout(15) << "hit export target (new) is " << counter << dendl; |
653 | } else { | |
654 | dout(15) << "hit export target is " << counter << dendl; | |
655 | } | |
656 | } | |
657 | ||
658 | class C_MDS_MonCommand : public MDSInternalContext { | |
659 | std::string cmd; | |
660 | public: | |
661 | std::string outs; | |
662 | C_MDS_MonCommand(MDSRank *m, std::string_view c) | |
663 | : MDSInternalContext(m), cmd(c) {} | |
664 | void finish(int r) override { | |
665 | mds->_mon_command_finish(r, cmd, outs); | |
666 | } | |
667 | }; | |
668 | ||
669 | void MDSRank::_mon_command_finish(int r, std::string_view cmd, std::string_view outs) | |
670 | { | |
671 | if (r < 0) { | |
672 | dout(0) << __func__ << ": mon command " << cmd << " failed with errno " << r | |
673 | << " (" << outs << ")" << dendl; | |
7c673cae | 674 | } else { |
11fdf7f2 | 675 | dout(1) << __func__ << ": mon command " << cmd << " succeed" << dendl; |
7c673cae | 676 | } |
11fdf7f2 TL |
677 | } |
678 | ||
679 | void MDSRank::set_mdsmap_multimds_snaps_allowed() | |
680 | { | |
681 | static bool already_sent = false; | |
682 | if (already_sent) | |
683 | return; | |
684 | ||
685 | stringstream ss; | |
686 | ss << "{\"prefix\":\"fs set\", \"fs_name\":\"" << mdsmap->get_fs_name() << "\", "; | |
687 | ss << "\"var\":\"allow_multimds_snaps\", \"val\":\"true\", "; | |
688 | ss << "\"confirm\":\"--yes-i-am-really-a-mds\"}"; | |
689 | std::vector<std::string> cmd = {ss.str()}; | |
690 | ||
691 | dout(0) << __func__ << ": sending mon command: " << cmd[0] << dendl; | |
692 | ||
693 | C_MDS_MonCommand *fin = new C_MDS_MonCommand(this, cmd[0]); | |
694 | monc->start_mon_command(cmd, {}, nullptr, &fin->outs, new C_IO_Wrapper(this, fin)); | |
695 | ||
696 | already_sent = true; | |
697 | } | |
698 | ||
699 | void MDSRank::mark_base_recursively_scrubbed(inodeno_t ino) | |
700 | { | |
701 | if (mdsmap->get_tableserver() == whoami) | |
702 | snapserver->mark_base_recursively_scrubbed(ino); | |
7c673cae FG |
703 | } |
704 | ||
705 | void MDSRankDispatcher::tick() | |
706 | { | |
707 | heartbeat_reset(); | |
708 | ||
709 | if (beacon.is_laggy()) { | |
91327a77 | 710 | dout(1) << "skipping upkeep work because connection to Monitors appears laggy" << dendl; |
7c673cae FG |
711 | return; |
712 | } | |
713 | ||
714 | check_ops_in_flight(); | |
715 | ||
716 | // Wake up thread in case we use to be laggy and have waiting_for_nolaggy | |
717 | // messages to progress. | |
718 | progress_thread.signal(); | |
719 | ||
720 | // make sure mds log flushes, trims periodically | |
721 | mdlog->flush(); | |
722 | ||
91327a77 AA |
723 | // update average session uptime |
724 | sessionmap.update_average_session_age(); | |
725 | ||
7c673cae | 726 | if (is_active() || is_stopping()) { |
7c673cae FG |
727 | mdlog->trim(); // NOT during recovery! |
728 | } | |
729 | ||
7c673cae | 730 | // ... |
eafe8130 | 731 | if (is_cache_trimmable()) { |
7c673cae | 732 | server->find_idle_sessions(); |
91327a77 | 733 | server->evict_cap_revoke_non_responders(); |
7c673cae FG |
734 | locker->tick(); |
735 | } | |
736 | ||
11fdf7f2 TL |
737 | // log |
738 | if (logger) { | |
739 | logger->set(l_mds_subtrees, mdcache->num_subtrees()); | |
740 | mdcache->log_stat(); | |
741 | } | |
742 | ||
7c673cae FG |
743 | if (is_reconnect()) |
744 | server->reconnect_tick(); | |
745 | ||
746 | if (is_active()) { | |
747 | balancer->tick(); | |
748 | mdcache->find_stale_fragment_freeze(); | |
749 | mdcache->migrator->find_stale_export_freeze(); | |
11fdf7f2 TL |
750 | |
751 | if (mdsmap->get_tableserver() == whoami) { | |
7c673cae | 752 | snapserver->check_osd_map(false); |
11fdf7f2 TL |
753 | // Filesystem was created by pre-mimic mds. Allow multi-active mds after |
754 | // all old snapshots are deleted. | |
755 | if (!mdsmap->allows_multimds_snaps() && | |
756 | snapserver->can_allow_multimds_snaps()) { | |
757 | set_mdsmap_multimds_snaps_allowed(); | |
758 | } | |
759 | } | |
7c673cae FG |
760 | } |
761 | ||
762 | if (is_active() || is_stopping()) { | |
11fdf7f2 | 763 | update_targets(); |
7c673cae FG |
764 | } |
765 | ||
766 | // shut down? | |
767 | if (is_stopping()) { | |
768 | mdlog->trim(); | |
769 | if (mdcache->shutdown_pass()) { | |
770 | uint64_t pq_progress = 0 ; | |
771 | uint64_t pq_total = 0; | |
772 | size_t pq_in_flight = 0; | |
773 | if (!purge_queue.drain(&pq_progress, &pq_total, &pq_in_flight)) { | |
774 | dout(7) << "shutdown_pass=true, but still waiting for purge queue" | |
775 | << dendl; | |
776 | // This takes unbounded time, so we must indicate progress | |
777 | // to the administrator: we do it in a slightly imperfect way | |
778 | // by sending periodic (tick frequency) clog messages while | |
779 | // in this state. | |
780 | clog->info() << "MDS rank " << whoami << " waiting for purge queue (" | |
781 | << std::dec << pq_progress << "/" << pq_total << " " << pq_in_flight | |
782 | << " files purging" << ")"; | |
783 | } else { | |
784 | dout(7) << "shutdown_pass=true, finished w/ shutdown, moving to " | |
785 | "down:stopped" << dendl; | |
786 | stopping_done(); | |
787 | } | |
788 | } | |
789 | else { | |
790 | dout(7) << "shutdown_pass=false" << dendl; | |
791 | } | |
792 | } | |
793 | ||
794 | // Expose ourselves to Beacon to update health indicators | |
795 | beacon.notify_health(this); | |
796 | } | |
797 | ||
798 | void MDSRankDispatcher::shutdown() | |
799 | { | |
800 | // It should never be possible for shutdown to get called twice, because | |
801 | // anyone picking up mds_lock checks if stopping is true and drops | |
802 | // out if it is. | |
11fdf7f2 | 803 | ceph_assert(stopping == false); |
7c673cae FG |
804 | stopping = true; |
805 | ||
806 | dout(1) << __func__ << ": shutting down rank " << whoami << dendl; | |
807 | ||
92f5a8d4 TL |
808 | g_conf().remove_observer(this); |
809 | ||
7c673cae FG |
810 | timer.shutdown(); |
811 | ||
812 | // MDLog has to shut down before the finisher, because some of its | |
813 | // threads block on IOs that require finisher to complete. | |
814 | mdlog->shutdown(); | |
815 | ||
816 | // shut down cache | |
817 | mdcache->shutdown(); | |
818 | ||
819 | purge_queue.shutdown(); | |
820 | ||
9f95a23c | 821 | mds_lock.unlock(); |
7c673cae | 822 | finisher->stop(); // no flushing |
9f95a23c | 823 | mds_lock.lock(); |
7c673cae | 824 | |
31f18b77 | 825 | if (objecter->initialized) |
7c673cae FG |
826 | objecter->shutdown(); |
827 | ||
828 | monc->shutdown(); | |
829 | ||
830 | op_tracker.on_shutdown(); | |
831 | ||
832 | progress_thread.shutdown(); | |
833 | ||
834 | // release mds_lock for finisher/messenger threads (e.g. | |
835 | // MDSDaemon::ms_handle_reset called from Messenger). | |
9f95a23c | 836 | mds_lock.unlock(); |
7c673cae FG |
837 | |
838 | // shut down messenger | |
839 | messenger->shutdown(); | |
840 | ||
9f95a23c | 841 | mds_lock.lock(); |
7c673cae FG |
842 | |
843 | // Workaround unclean shutdown: HeartbeatMap will assert if | |
844 | // worker is not removed (as we do in ~MDS), but ~MDS is not | |
845 | // always called after suicide. | |
846 | if (hb) { | |
847 | g_ceph_context->get_heartbeat_map()->remove_worker(hb); | |
848 | hb = NULL; | |
849 | } | |
850 | } | |
851 | ||
852 | /** | |
853 | * Helper for simple callbacks that call a void fn with no args. | |
854 | */ | |
855 | class C_MDS_VoidFn : public MDSInternalContext | |
856 | { | |
857 | typedef void (MDSRank::*fn_ptr)(); | |
858 | protected: | |
859 | fn_ptr fn; | |
860 | public: | |
861 | C_MDS_VoidFn(MDSRank *mds_, fn_ptr fn_) | |
862 | : MDSInternalContext(mds_), fn(fn_) | |
863 | { | |
11fdf7f2 TL |
864 | ceph_assert(mds_); |
865 | ceph_assert(fn_); | |
7c673cae FG |
866 | } |
867 | ||
868 | void finish(int r) override | |
869 | { | |
870 | (mds->*fn)(); | |
871 | } | |
872 | }; | |
873 | ||
874 | int64_t MDSRank::get_metadata_pool() | |
875 | { | |
876 | return mdsmap->get_metadata_pool(); | |
877 | } | |
878 | ||
879 | MDSTableClient *MDSRank::get_table_client(int t) | |
880 | { | |
881 | switch (t) { | |
882 | case TABLE_ANCHOR: return NULL; | |
883 | case TABLE_SNAP: return snapclient; | |
884 | default: ceph_abort(); | |
885 | } | |
886 | } | |
887 | ||
888 | MDSTableServer *MDSRank::get_table_server(int t) | |
889 | { | |
890 | switch (t) { | |
891 | case TABLE_ANCHOR: return NULL; | |
892 | case TABLE_SNAP: return snapserver; | |
893 | default: ceph_abort(); | |
894 | } | |
895 | } | |
896 | ||
897 | void MDSRank::suicide() | |
898 | { | |
899 | if (suicide_hook) { | |
900 | suicide_hook->complete(0); | |
901 | suicide_hook = NULL; | |
902 | } | |
903 | } | |
904 | ||
905 | void MDSRank::respawn() | |
906 | { | |
907 | if (respawn_hook) { | |
908 | respawn_hook->complete(0); | |
909 | respawn_hook = NULL; | |
910 | } | |
911 | } | |
912 | ||
913 | void MDSRank::damaged() | |
914 | { | |
11fdf7f2 | 915 | ceph_assert(whoami != MDS_RANK_NONE); |
9f95a23c | 916 | ceph_assert(ceph_mutex_is_locked_by_me(mds_lock)); |
7c673cae | 917 | |
11fdf7f2 | 918 | beacon.set_want_state(*mdsmap, MDSMap::STATE_DAMAGED); |
7c673cae FG |
919 | monc->flush_log(); // Flush any clog error from before we were called |
920 | beacon.notify_health(this); // Include latest status in our swan song | |
11fdf7f2 | 921 | beacon.send_and_wait(g_conf()->mds_mon_shutdown_timeout); |
7c673cae FG |
922 | |
923 | // It's okay if we timed out and the mon didn't get our beacon, because | |
924 | // another daemon (or ourselves after respawn) will eventually take the | |
925 | // rank and report DAMAGED again when it hits same problem we did. | |
926 | ||
927 | respawn(); // Respawn into standby in case mon has other work for us | |
928 | } | |
929 | ||
930 | void MDSRank::damaged_unlocked() | |
931 | { | |
11fdf7f2 | 932 | std::lock_guard l(mds_lock); |
7c673cae FG |
933 | damaged(); |
934 | } | |
935 | ||
936 | void MDSRank::handle_write_error(int err) | |
937 | { | |
938 | if (err == -EBLACKLISTED) { | |
939 | derr << "we have been blacklisted (fenced), respawning..." << dendl; | |
940 | respawn(); | |
941 | return; | |
942 | } | |
943 | ||
11fdf7f2 | 944 | if (g_conf()->mds_action_on_write_error >= 2) { |
7c673cae FG |
945 | derr << "unhandled write error " << cpp_strerror(err) << ", suicide..." << dendl; |
946 | respawn(); | |
11fdf7f2 | 947 | } else if (g_conf()->mds_action_on_write_error == 1) { |
7c673cae FG |
948 | derr << "unhandled write error " << cpp_strerror(err) << ", force readonly..." << dendl; |
949 | mdcache->force_readonly(); | |
950 | } else { | |
951 | // ignore; | |
952 | derr << "unhandled write error " << cpp_strerror(err) << ", ignore..." << dendl; | |
953 | } | |
954 | } | |
955 | ||
956 | void *MDSRank::ProgressThread::entry() | |
957 | { | |
9f95a23c | 958 | std::unique_lock l(mds->mds_lock); |
7c673cae | 959 | while (true) { |
9f95a23c TL |
960 | cond.wait(l, [this] { |
961 | return (mds->stopping || | |
962 | !mds->finished_queue.empty() || | |
963 | (!mds->waiting_for_nolaggy.empty() && !mds->beacon.is_laggy())); | |
964 | }); | |
7c673cae FG |
965 | |
966 | if (mds->stopping) { | |
967 | break; | |
968 | } | |
969 | ||
970 | mds->_advance_queues(); | |
971 | } | |
972 | ||
973 | return NULL; | |
974 | } | |
975 | ||
976 | ||
977 | void MDSRank::ProgressThread::shutdown() | |
978 | { | |
9f95a23c | 979 | ceph_assert(ceph_mutex_is_locked_by_me(mds->mds_lock)); |
11fdf7f2 | 980 | ceph_assert(mds->stopping); |
7c673cae FG |
981 | |
982 | if (am_self()) { | |
983 | // Stopping is set, we will fall out of our main loop naturally | |
984 | } else { | |
985 | // Kick the thread to notice mds->stopping, and join it | |
9f95a23c TL |
986 | cond.notify_all(); |
987 | mds->mds_lock.unlock(); | |
7c673cae FG |
988 | if (is_started()) |
989 | join(); | |
9f95a23c | 990 | mds->mds_lock.lock(); |
7c673cae FG |
991 | } |
992 | } | |
993 | ||
9f95a23c | 994 | bool MDSRankDispatcher::ms_dispatch(const cref_t<Message> &m) |
7c673cae | 995 | { |
f64942e4 | 996 | if (m->get_source().is_client()) { |
11fdf7f2 | 997 | Session *session = static_cast<Session*>(m->get_connection()->get_priv().get()); |
f64942e4 AA |
998 | if (session) |
999 | session->last_seen = Session::clock::now(); | |
1000 | } | |
1001 | ||
7c673cae | 1002 | inc_dispatch_depth(); |
f64942e4 | 1003 | bool ret = _dispatch(m, true); |
7c673cae FG |
1004 | dec_dispatch_depth(); |
1005 | return ret; | |
1006 | } | |
1007 | ||
9f95a23c | 1008 | bool MDSRank::_dispatch(const cref_t<Message> &m, bool new_msg) |
7c673cae FG |
1009 | { |
1010 | if (is_stale_message(m)) { | |
7c673cae FG |
1011 | return true; |
1012 | } | |
f6b5b4d7 TL |
1013 | // do not proceed if this message cannot be handled |
1014 | if (!is_valid_message(m)) { | |
1015 | return false; | |
1016 | } | |
7c673cae FG |
1017 | |
1018 | if (beacon.is_laggy()) { | |
91327a77 | 1019 | dout(5) << " laggy, deferring " << *m << dendl; |
7c673cae FG |
1020 | waiting_for_nolaggy.push_back(m); |
1021 | } else if (new_msg && !waiting_for_nolaggy.empty()) { | |
91327a77 | 1022 | dout(5) << " there are deferred messages, deferring " << *m << dendl; |
7c673cae FG |
1023 | waiting_for_nolaggy.push_back(m); |
1024 | } else { | |
f6b5b4d7 | 1025 | handle_message(m); |
7c673cae FG |
1026 | heartbeat_reset(); |
1027 | } | |
1028 | ||
1029 | if (dispatch_depth > 1) | |
1030 | return true; | |
1031 | ||
1032 | // finish any triggered contexts | |
1033 | _advance_queues(); | |
1034 | ||
1035 | if (beacon.is_laggy()) { | |
1036 | // We've gone laggy during dispatch, don't do any | |
1037 | // more housekeeping | |
1038 | return true; | |
1039 | } | |
1040 | ||
7c673cae FG |
1041 | // hack: thrash exports |
1042 | static utime_t start; | |
1043 | utime_t now = ceph_clock_now(); | |
1044 | if (start == utime_t()) | |
1045 | start = now; | |
1046 | /*double el = now - start; | |
1047 | if (el > 30.0 && | |
1048 | el < 60.0)*/ | |
11fdf7f2 | 1049 | for (int i=0; i<g_conf()->mds_thrash_exports; i++) { |
7c673cae FG |
1050 | set<mds_rank_t> s; |
1051 | if (!is_active()) break; | |
1052 | mdsmap->get_mds_set(s, MDSMap::STATE_ACTIVE); | |
1053 | if (s.size() < 2 || CInode::count() < 10) | |
1054 | break; // need peers for this to work. | |
11fdf7f2 TL |
1055 | if (mdcache->migrator->get_num_exporting() > g_conf()->mds_thrash_exports * 5 || |
1056 | mdcache->migrator->get_export_queue_size() > g_conf()->mds_thrash_exports * 10) | |
7c673cae FG |
1057 | break; |
1058 | ||
11fdf7f2 | 1059 | dout(7) << "mds thrashing exports pass " << (i+1) << "/" << g_conf()->mds_thrash_exports << dendl; |
7c673cae FG |
1060 | |
1061 | // pick a random dir inode | |
1062 | CInode *in = mdcache->hack_pick_random_inode(); | |
1063 | ||
9f95a23c | 1064 | auto&& ls = in->get_dirfrags(); |
7c673cae | 1065 | if (!ls.empty()) { // must be an open dir. |
9f95a23c | 1066 | const auto& dir = ls[rand() % ls.size()]; |
7c673cae FG |
1067 | if (!dir->get_parent_dir()) continue; // must be linked. |
1068 | if (!dir->is_auth()) continue; // must be auth. | |
1069 | ||
1070 | mds_rank_t dest; | |
1071 | do { | |
1072 | int k = rand() % s.size(); | |
1073 | set<mds_rank_t>::iterator p = s.begin(); | |
1074 | while (k--) ++p; | |
1075 | dest = *p; | |
1076 | } while (dest == whoami); | |
1077 | mdcache->migrator->export_dir_nicely(dir,dest); | |
1078 | } | |
1079 | } | |
1080 | // hack: thrash fragments | |
11fdf7f2 | 1081 | for (int i=0; i<g_conf()->mds_thrash_fragments; i++) { |
7c673cae | 1082 | if (!is_active()) break; |
11fdf7f2 TL |
1083 | if (mdcache->get_num_fragmenting_dirs() > 5 * g_conf()->mds_thrash_fragments) break; |
1084 | dout(7) << "mds thrashing fragments pass " << (i+1) << "/" << g_conf()->mds_thrash_fragments << dendl; | |
7c673cae FG |
1085 | |
1086 | // pick a random dir inode | |
1087 | CInode *in = mdcache->hack_pick_random_inode(); | |
1088 | ||
9f95a23c | 1089 | auto&& ls = in->get_dirfrags(); |
7c673cae FG |
1090 | if (ls.empty()) continue; // must be an open dir. |
1091 | CDir *dir = ls.front(); | |
1092 | if (!dir->get_parent_dir()) continue; // must be linked. | |
1093 | if (!dir->is_auth()) continue; // must be auth. | |
1094 | frag_t fg = dir->get_frag(); | |
11fdf7f2 TL |
1095 | if ((fg == frag_t() || (rand() % (1 << fg.bits()) == 0))) { |
1096 | mdcache->split_dir(dir, 1); | |
1097 | } else { | |
1098 | balancer->queue_merge(dir); | |
7c673cae FG |
1099 | } |
1100 | } | |
1101 | ||
1102 | // hack: force hash root? | |
1103 | /* | |
1104 | if (false && | |
1105 | mdcache->get_root() && | |
1106 | mdcache->get_root()->dir && | |
1107 | !(mdcache->get_root()->dir->is_hashed() || | |
1108 | mdcache->get_root()->dir->is_hashing())) { | |
1109 | dout(0) << "hashing root" << dendl; | |
1110 | mdcache->migrator->hash_dir(mdcache->get_root()->dir); | |
1111 | } | |
1112 | */ | |
1113 | ||
c07f9fc5 FG |
1114 | update_mlogger(); |
1115 | return true; | |
1116 | } | |
1117 | ||
1118 | void MDSRank::update_mlogger() | |
1119 | { | |
7c673cae FG |
1120 | if (mlogger) { |
1121 | mlogger->set(l_mdm_ino, CInode::count()); | |
1122 | mlogger->set(l_mdm_dir, CDir::count()); | |
1123 | mlogger->set(l_mdm_dn, CDentry::count()); | |
1124 | mlogger->set(l_mdm_cap, Capability::count()); | |
7c673cae FG |
1125 | mlogger->set(l_mdm_inoa, CInode::increments()); |
1126 | mlogger->set(l_mdm_inos, CInode::decrements()); | |
1127 | mlogger->set(l_mdm_dira, CDir::increments()); | |
1128 | mlogger->set(l_mdm_dirs, CDir::decrements()); | |
1129 | mlogger->set(l_mdm_dna, CDentry::increments()); | |
1130 | mlogger->set(l_mdm_dns, CDentry::decrements()); | |
1131 | mlogger->set(l_mdm_capa, Capability::increments()); | |
1132 | mlogger->set(l_mdm_caps, Capability::decrements()); | |
7c673cae | 1133 | } |
7c673cae FG |
1134 | } |
1135 | ||
f6b5b4d7 TL |
1136 | // message types that the mds can handle |
1137 | bool MDSRank::is_valid_message(const cref_t<Message> &m) { | |
1138 | int port = m->get_type() & 0xff00; | |
1139 | int type = m->get_type(); | |
1140 | ||
1141 | if (port == MDS_PORT_CACHE || | |
1142 | port == MDS_PORT_MIGRATOR || | |
1143 | type == CEPH_MSG_CLIENT_SESSION || | |
1144 | type == CEPH_MSG_CLIENT_RECONNECT || | |
1145 | type == CEPH_MSG_CLIENT_RECLAIM || | |
1146 | type == CEPH_MSG_CLIENT_REQUEST || | |
1147 | type == MSG_MDS_SLAVE_REQUEST || | |
1148 | type == MSG_MDS_HEARTBEAT || | |
1149 | type == MSG_MDS_TABLE_REQUEST || | |
1150 | type == MSG_MDS_LOCK || | |
1151 | type == MSG_MDS_INODEFILECAPS || | |
1152 | type == CEPH_MSG_CLIENT_CAPS || | |
1153 | type == CEPH_MSG_CLIENT_CAPRELEASE || | |
1154 | type == CEPH_MSG_CLIENT_LEASE) { | |
1155 | return true; | |
1156 | } | |
1157 | ||
1158 | return false; | |
1159 | } | |
1160 | ||
7c673cae FG |
1161 | /* |
1162 | * lower priority messages we defer if we seem laggy | |
1163 | */ | |
f6b5b4d7 TL |
1164 | |
1165 | #define ALLOW_MESSAGES_FROM(peers) \ | |
1166 | do { \ | |
1167 | if (m->get_connection() && (m->get_connection()->get_peer_type() & (peers)) == 0) { \ | |
1168 | dout(0) << __FILE__ << "." << __LINE__ << ": filtered out request, peer=" << m->get_connection()->get_peer_type() \ | |
1169 | << " allowing=" << #peers << " message=" << *m << dendl; \ | |
1170 | return; \ | |
1171 | } \ | |
1172 | } while (0) | |
1173 | ||
1174 | void MDSRank::handle_message(const cref_t<Message> &m) | |
7c673cae FG |
1175 | { |
1176 | int port = m->get_type() & 0xff00; | |
1177 | ||
1178 | switch (port) { | |
1179 | case MDS_PORT_CACHE: | |
1180 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_MDS); | |
1181 | mdcache->dispatch(m); | |
1182 | break; | |
1183 | ||
1184 | case MDS_PORT_MIGRATOR: | |
1185 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_MDS); | |
1186 | mdcache->migrator->dispatch(m); | |
1187 | break; | |
1188 | ||
1189 | default: | |
1190 | switch (m->get_type()) { | |
1191 | // SERVER | |
1192 | case CEPH_MSG_CLIENT_SESSION: | |
1193 | case CEPH_MSG_CLIENT_RECONNECT: | |
11fdf7f2 | 1194 | case CEPH_MSG_CLIENT_RECLAIM: |
7c673cae FG |
1195 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_CLIENT); |
1196 | // fall-thru | |
1197 | case CEPH_MSG_CLIENT_REQUEST: | |
1198 | server->dispatch(m); | |
1199 | break; | |
1200 | case MSG_MDS_SLAVE_REQUEST: | |
1201 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_MDS); | |
1202 | server->dispatch(m); | |
1203 | break; | |
1204 | ||
1205 | case MSG_MDS_HEARTBEAT: | |
1206 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_MDS); | |
1207 | balancer->proc_message(m); | |
1208 | break; | |
1209 | ||
1210 | case MSG_MDS_TABLE_REQUEST: | |
1211 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_MDS); | |
1212 | { | |
9f95a23c | 1213 | const cref_t<MMDSTableRequest> &req = ref_cast<MMDSTableRequest>(m); |
11fdf7f2 TL |
1214 | if (req->op < 0) { |
1215 | MDSTableClient *client = get_table_client(req->table); | |
1216 | client->handle_request(req); | |
1217 | } else { | |
1218 | MDSTableServer *server = get_table_server(req->table); | |
1219 | server->handle_request(req); | |
1220 | } | |
7c673cae FG |
1221 | } |
1222 | break; | |
1223 | ||
1224 | case MSG_MDS_LOCK: | |
1225 | case MSG_MDS_INODEFILECAPS: | |
1226 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_MDS); | |
1227 | locker->dispatch(m); | |
1228 | break; | |
1229 | ||
1230 | case CEPH_MSG_CLIENT_CAPS: | |
1231 | case CEPH_MSG_CLIENT_CAPRELEASE: | |
1232 | case CEPH_MSG_CLIENT_LEASE: | |
1233 | ALLOW_MESSAGES_FROM(CEPH_ENTITY_TYPE_CLIENT); | |
1234 | locker->dispatch(m); | |
1235 | break; | |
1236 | ||
1237 | default: | |
f6b5b4d7 | 1238 | derr << "unrecognized message " << *m << dendl; |
7c673cae FG |
1239 | } |
1240 | } | |
7c673cae FG |
1241 | } |
1242 | ||
1243 | /** | |
1244 | * Advance finished_queue and waiting_for_nolaggy. | |
1245 | * | |
1246 | * Usually drain both queues, but may not drain waiting_for_nolaggy | |
1247 | * if beacon is currently laggy. | |
1248 | */ | |
1249 | void MDSRank::_advance_queues() | |
1250 | { | |
9f95a23c | 1251 | ceph_assert(ceph_mutex_is_locked_by_me(mds_lock)); |
7c673cae | 1252 | |
11fdf7f2 | 1253 | if (!finished_queue.empty()) { |
7c673cae | 1254 | dout(7) << "mds has " << finished_queue.size() << " queued contexts" << dendl; |
11fdf7f2 TL |
1255 | while (!finished_queue.empty()) { |
1256 | auto fin = finished_queue.front(); | |
1257 | finished_queue.pop_front(); | |
1258 | ||
1259 | dout(10) << " finish " << fin << dendl; | |
1260 | fin->complete(0); | |
7c673cae FG |
1261 | |
1262 | heartbeat_reset(); | |
1263 | } | |
1264 | } | |
1265 | ||
1266 | while (!waiting_for_nolaggy.empty()) { | |
1267 | // stop if we're laggy now! | |
1268 | if (beacon.is_laggy()) | |
1269 | break; | |
1270 | ||
9f95a23c | 1271 | cref_t<Message> old = waiting_for_nolaggy.front(); |
7c673cae FG |
1272 | waiting_for_nolaggy.pop_front(); |
1273 | ||
11fdf7f2 | 1274 | if (!is_stale_message(old)) { |
7c673cae | 1275 | dout(7) << " processing laggy deferred " << *old << dendl; |
f6b5b4d7 TL |
1276 | ceph_assert(is_valid_message(old)); |
1277 | handle_message(old); | |
7c673cae FG |
1278 | } |
1279 | ||
1280 | heartbeat_reset(); | |
1281 | } | |
1282 | } | |
1283 | ||
1284 | /** | |
1285 | * Call this when you take mds_lock, or periodically if you're going to | |
1286 | * hold the lock for a long time (e.g. iterating over clients/inodes) | |
1287 | */ | |
1288 | void MDSRank::heartbeat_reset() | |
1289 | { | |
1290 | // Any thread might jump into mds_lock and call us immediately | |
1291 | // after a call to suicide() completes, in which case MDSRank::hb | |
1292 | // has been freed and we are a no-op. | |
1293 | if (!hb) { | |
11fdf7f2 | 1294 | ceph_assert(stopping); |
7c673cae FG |
1295 | return; |
1296 | } | |
1297 | ||
1298 | // NB not enabling suicide grace, because the mon takes care of killing us | |
1299 | // (by blacklisting us) when we fail to send beacons, and it's simpler to | |
1300 | // only have one way of dying. | |
11fdf7f2 | 1301 | auto grace = g_conf().get_val<double>("mds_heartbeat_grace"); |
f64942e4 | 1302 | g_ceph_context->get_heartbeat_map()->reset_timeout(hb, grace, 0); |
7c673cae FG |
1303 | } |
1304 | ||
9f95a23c | 1305 | bool MDSRank::is_stale_message(const cref_t<Message> &m) const |
7c673cae FG |
1306 | { |
1307 | // from bad mds? | |
1308 | if (m->get_source().is_mds()) { | |
1309 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
11fdf7f2 TL |
1310 | bool bad = false; |
1311 | if (mdsmap->is_down(from)) { | |
1312 | bad = true; | |
1313 | } else { | |
1314 | // FIXME: this is a convoluted check. we should be maintaining a nice | |
1315 | // clean map of current ConnectionRefs for current mdses!!! | |
1316 | auto c = messenger->connect_to(CEPH_ENTITY_TYPE_MDS, | |
1317 | mdsmap->get_addrs(from)); | |
1318 | if (c != m->get_connection()) { | |
1319 | bad = true; | |
1320 | dout(5) << " mds." << from << " should be " << c << " " | |
1321 | << c->get_peer_addrs() << " but this message is " | |
1322 | << m->get_connection() << " " << m->get_source_addrs() | |
1323 | << dendl; | |
1324 | } | |
1325 | } | |
1326 | if (bad) { | |
7c673cae FG |
1327 | // bogus mds? |
1328 | if (m->get_type() == CEPH_MSG_MDS_MAP) { | |
1329 | dout(5) << "got " << *m << " from old/bad/imposter mds " << m->get_source() | |
1330 | << ", but it's an mdsmap, looking at it" << dendl; | |
1331 | } else if (m->get_type() == MSG_MDS_CACHEEXPIRE && | |
11fdf7f2 | 1332 | mdsmap->get_addrs(from) == m->get_source_addrs()) { |
7c673cae FG |
1333 | dout(5) << "got " << *m << " from down mds " << m->get_source() |
1334 | << ", but it's a cache_expire, looking at it" << dendl; | |
1335 | } else { | |
1336 | dout(5) << "got " << *m << " from down/old/bad/imposter mds " << m->get_source() | |
1337 | << ", dropping" << dendl; | |
1338 | return true; | |
1339 | } | |
1340 | } | |
1341 | } | |
1342 | return false; | |
1343 | } | |
1344 | ||
9f95a23c | 1345 | Session *MDSRank::get_session(const cref_t<Message> &m) |
94b18763 | 1346 | { |
11fdf7f2 TL |
1347 | // do not carry ref |
1348 | auto session = static_cast<Session *>(m->get_connection()->get_priv().get()); | |
94b18763 FG |
1349 | if (session) { |
1350 | dout(20) << "get_session have " << session << " " << session->info.inst | |
1351 | << " state " << session->get_state_name() << dendl; | |
28e407b8 AA |
1352 | // Check if we've imported an open session since (new sessions start closed) |
1353 | if (session->is_closed()) { | |
1354 | Session *imported_session = sessionmap.get_session(session->info.inst.name); | |
1355 | if (imported_session && imported_session != session) { | |
11fdf7f2 TL |
1356 | dout(10) << __func__ << " replacing connection bootstrap session " |
1357 | << session << " with imported session " << imported_session | |
1358 | << dendl; | |
28e407b8 AA |
1359 | imported_session->info.auth_name = session->info.auth_name; |
1360 | //assert(session->info.auth_name == imported_session->info.auth_name); | |
11fdf7f2 TL |
1361 | ceph_assert(session->info.inst == imported_session->info.inst); |
1362 | imported_session->set_connection(session->get_connection().get()); | |
28e407b8 AA |
1363 | // send out any queued messages |
1364 | while (!session->preopen_out_queue.empty()) { | |
11fdf7f2 | 1365 | imported_session->get_connection()->send_message2(std::move(session->preopen_out_queue.front())); |
28e407b8 AA |
1366 | session->preopen_out_queue.pop_front(); |
1367 | } | |
1368 | imported_session->auth_caps = session->auth_caps; | |
f64942e4 | 1369 | imported_session->last_seen = session->last_seen; |
11fdf7f2 TL |
1370 | ceph_assert(session->get_nref() == 1); |
1371 | imported_session->get_connection()->set_priv(imported_session->get()); | |
28e407b8 AA |
1372 | session = imported_session; |
1373 | } | |
1374 | } | |
94b18763 FG |
1375 | } else { |
1376 | dout(20) << "get_session dne for " << m->get_source_inst() << dendl; | |
1377 | } | |
1378 | return session; | |
1379 | } | |
7c673cae | 1380 | |
9f95a23c | 1381 | void MDSRank::send_message(const ref_t<Message>& m, const ConnectionRef& c) |
7c673cae | 1382 | { |
11fdf7f2 TL |
1383 | ceph_assert(c); |
1384 | c->send_message2(m); | |
7c673cae FG |
1385 | } |
1386 | ||
1387 | ||
9f95a23c | 1388 | void MDSRank::send_message_mds(const ref_t<Message>& m, mds_rank_t mds) |
7c673cae FG |
1389 | { |
1390 | if (!mdsmap->is_up(mds)) { | |
1391 | dout(10) << "send_message_mds mds." << mds << " not up, dropping " << *m << dendl; | |
7c673cae FG |
1392 | return; |
1393 | } | |
1394 | ||
1395 | // send mdsmap first? | |
1396 | if (mds != whoami && peer_mdsmap_epoch[mds] < mdsmap->get_epoch()) { | |
9f95a23c | 1397 | auto _m = make_message<MMDSMap>(monc->get_fsid(), *mdsmap); |
11fdf7f2 | 1398 | messenger->send_to_mds(_m.detach(), mdsmap->get_addrs(mds)); |
7c673cae FG |
1399 | peer_mdsmap_epoch[mds] = mdsmap->get_epoch(); |
1400 | } | |
1401 | ||
1402 | // send message | |
9f95a23c | 1403 | messenger->send_to_mds(ref_t<Message>(m).detach(), mdsmap->get_addrs(mds)); |
7c673cae FG |
1404 | } |
1405 | ||
9f95a23c | 1406 | void MDSRank::forward_message_mds(const cref_t<MClientRequest>& m, mds_rank_t mds) |
7c673cae | 1407 | { |
11fdf7f2 | 1408 | ceph_assert(mds != whoami); |
7c673cae | 1409 | |
11fdf7f2 TL |
1410 | /* |
1411 | * don't actually forward if non-idempotent! | |
1412 | * client has to do it. although the MDS will ignore duplicate requests, | |
1413 | * the affected metadata may migrate, in which case the new authority | |
1414 | * won't have the metareq_id in the completed request map. | |
1415 | */ | |
1416 | // NEW: always make the client resend! | |
1417 | bool client_must_resend = true; //!creq->can_forward(); | |
7c673cae | 1418 | |
11fdf7f2 TL |
1419 | // tell the client where it should go |
1420 | auto session = get_session(m); | |
9f95a23c | 1421 | auto f = make_message<MClientRequestForward>(m->get_tid(), mds, m->get_num_fwd()+1, client_must_resend); |
11fdf7f2 | 1422 | send_message_client(f, session); |
7c673cae FG |
1423 | } |
1424 | ||
9f95a23c | 1425 | void MDSRank::send_message_client_counted(const ref_t<Message>& m, client_t client) |
7c673cae | 1426 | { |
11fdf7f2 | 1427 | Session *session = sessionmap.get_session(entity_name_t::CLIENT(client.v)); |
7c673cae FG |
1428 | if (session) { |
1429 | send_message_client_counted(m, session); | |
1430 | } else { | |
1431 | dout(10) << "send_message_client_counted no session for client." << client << " " << *m << dendl; | |
1432 | } | |
1433 | } | |
1434 | ||
9f95a23c | 1435 | void MDSRank::send_message_client_counted(const ref_t<Message>& m, const ConnectionRef& connection) |
7c673cae | 1436 | { |
11fdf7f2 TL |
1437 | // do not carry ref |
1438 | auto session = static_cast<Session *>(connection->get_priv().get()); | |
7c673cae | 1439 | if (session) { |
7c673cae FG |
1440 | send_message_client_counted(m, session); |
1441 | } else { | |
1442 | dout(10) << "send_message_client_counted has no session for " << m->get_source_inst() << dendl; | |
1443 | // another Connection took over the Session | |
1444 | } | |
1445 | } | |
1446 | ||
9f95a23c | 1447 | void MDSRank::send_message_client_counted(const ref_t<Message>& m, Session* session) |
7c673cae FG |
1448 | { |
1449 | version_t seq = session->inc_push_seq(); | |
1450 | dout(10) << "send_message_client_counted " << session->info.inst.name << " seq " | |
1451 | << seq << " " << *m << dendl; | |
11fdf7f2 TL |
1452 | if (session->get_connection()) { |
1453 | session->get_connection()->send_message2(m); | |
7c673cae FG |
1454 | } else { |
1455 | session->preopen_out_queue.push_back(m); | |
1456 | } | |
1457 | } | |
1458 | ||
9f95a23c | 1459 | void MDSRank::send_message_client(const ref_t<Message>& m, Session* session) |
7c673cae FG |
1460 | { |
1461 | dout(10) << "send_message_client " << session->info.inst << " " << *m << dendl; | |
11fdf7f2 TL |
1462 | if (session->get_connection()) { |
1463 | session->get_connection()->send_message2(m); | |
7c673cae FG |
1464 | } else { |
1465 | session->preopen_out_queue.push_back(m); | |
1466 | } | |
1467 | } | |
1468 | ||
1469 | /** | |
1470 | * This is used whenever a RADOS operation has been cancelled | |
1471 | * or a RADOS client has been blacklisted, to cause the MDS and | |
1472 | * any clients to wait for this OSD epoch before using any new caps. | |
1473 | * | |
1474 | * See doc/cephfs/eviction | |
1475 | */ | |
1476 | void MDSRank::set_osd_epoch_barrier(epoch_t e) | |
1477 | { | |
1478 | dout(4) << __func__ << ": epoch=" << e << dendl; | |
1479 | osd_epoch_barrier = e; | |
1480 | } | |
1481 | ||
9f95a23c | 1482 | void MDSRank::retry_dispatch(const cref_t<Message> &m) |
7c673cae FG |
1483 | { |
1484 | inc_dispatch_depth(); | |
1485 | _dispatch(m, false); | |
1486 | dec_dispatch_depth(); | |
1487 | } | |
1488 | ||
91327a77 | 1489 | double MDSRank::get_dispatch_queue_max_age(utime_t now) const |
7c673cae | 1490 | { |
91327a77 | 1491 | return messenger->get_dispatch_queue_max_age(now); |
7c673cae FG |
1492 | } |
1493 | ||
1494 | bool MDSRank::is_daemon_stopping() const | |
1495 | { | |
1496 | return stopping; | |
1497 | } | |
1498 | ||
1499 | void MDSRank::request_state(MDSMap::DaemonState s) | |
1500 | { | |
1501 | dout(3) << "request_state " << ceph_mds_state_name(s) << dendl; | |
11fdf7f2 | 1502 | beacon.set_want_state(*mdsmap, s); |
7c673cae FG |
1503 | beacon.send(); |
1504 | } | |
1505 | ||
1506 | ||
1507 | class C_MDS_BootStart : public MDSInternalContext { | |
1508 | MDSRank::BootStep nextstep; | |
1509 | public: | |
1510 | C_MDS_BootStart(MDSRank *m, MDSRank::BootStep n) | |
1511 | : MDSInternalContext(m), nextstep(n) {} | |
1512 | void finish(int r) override { | |
1513 | mds->boot_start(nextstep, r); | |
1514 | } | |
1515 | }; | |
1516 | ||
1517 | ||
1518 | void MDSRank::boot_start(BootStep step, int r) | |
1519 | { | |
1520 | // Handle errors from previous step | |
1521 | if (r < 0) { | |
1522 | if (is_standby_replay() && (r == -EAGAIN)) { | |
1523 | dout(0) << "boot_start encountered an error EAGAIN" | |
1524 | << ", respawning since we fell behind journal" << dendl; | |
1525 | respawn(); | |
1526 | } else if (r == -EINVAL || r == -ENOENT) { | |
1527 | // Invalid or absent data, indicates damaged on-disk structures | |
1528 | clog->error() << "Error loading MDS rank " << whoami << ": " | |
1529 | << cpp_strerror(r); | |
1530 | damaged(); | |
11fdf7f2 | 1531 | ceph_assert(r == 0); // Unreachable, damaged() calls respawn() |
f64942e4 AA |
1532 | } else if (r == -EROFS) { |
1533 | dout(0) << "boot error forcing transition to read-only; MDS will try to continue" << dendl; | |
7c673cae FG |
1534 | } else { |
1535 | // Completely unexpected error, give up and die | |
1536 | dout(0) << "boot_start encountered an error, failing" << dendl; | |
1537 | suicide(); | |
1538 | return; | |
1539 | } | |
1540 | } | |
1541 | ||
11fdf7f2 | 1542 | ceph_assert(is_starting() || is_any_replay()); |
7c673cae FG |
1543 | |
1544 | switch(step) { | |
1545 | case MDS_BOOT_INITIAL: | |
1546 | { | |
1547 | mdcache->init_layouts(); | |
1548 | ||
1549 | MDSGatherBuilder gather(g_ceph_context, | |
1550 | new C_MDS_BootStart(this, MDS_BOOT_OPEN_ROOT)); | |
a8e16298 | 1551 | dout(2) << "Booting: " << step << ": opening inotable" << dendl; |
7c673cae FG |
1552 | inotable->set_rank(whoami); |
1553 | inotable->load(gather.new_sub()); | |
1554 | ||
a8e16298 | 1555 | dout(2) << "Booting: " << step << ": opening sessionmap" << dendl; |
7c673cae FG |
1556 | sessionmap.set_rank(whoami); |
1557 | sessionmap.load(gather.new_sub()); | |
1558 | ||
a8e16298 | 1559 | dout(2) << "Booting: " << step << ": opening mds log" << dendl; |
7c673cae FG |
1560 | mdlog->open(gather.new_sub()); |
1561 | ||
3efd9988 | 1562 | if (is_starting()) { |
a8e16298 | 1563 | dout(2) << "Booting: " << step << ": opening purge queue" << dendl; |
3efd9988 FG |
1564 | purge_queue.open(new C_IO_Wrapper(this, gather.new_sub())); |
1565 | } else if (!standby_replaying) { | |
a8e16298 | 1566 | dout(2) << "Booting: " << step << ": opening purge queue (async)" << dendl; |
3efd9988 | 1567 | purge_queue.open(NULL); |
11fdf7f2 TL |
1568 | dout(2) << "Booting: " << step << ": loading open file table (async)" << dendl; |
1569 | mdcache->open_file_table.load(nullptr); | |
3efd9988 FG |
1570 | } |
1571 | ||
7c673cae | 1572 | if (mdsmap->get_tableserver() == whoami) { |
a8e16298 | 1573 | dout(2) << "Booting: " << step << ": opening snap table" << dendl; |
7c673cae FG |
1574 | snapserver->set_rank(whoami); |
1575 | snapserver->load(gather.new_sub()); | |
1576 | } | |
1577 | ||
1578 | gather.activate(); | |
1579 | } | |
1580 | break; | |
1581 | case MDS_BOOT_OPEN_ROOT: | |
1582 | { | |
a8e16298 | 1583 | dout(2) << "Booting: " << step << ": loading/discovering base inodes" << dendl; |
7c673cae FG |
1584 | |
1585 | MDSGatherBuilder gather(g_ceph_context, | |
1586 | new C_MDS_BootStart(this, MDS_BOOT_PREPARE_LOG)); | |
1587 | ||
28e407b8 AA |
1588 | if (is_starting()) { |
1589 | // load mydir frag for the first log segment (creating subtree map) | |
1590 | mdcache->open_mydir_frag(gather.new_sub()); | |
1591 | } else { | |
1592 | mdcache->open_mydir_inode(gather.new_sub()); | |
1593 | } | |
7c673cae | 1594 | |
11fdf7f2 TL |
1595 | mdcache->create_global_snaprealm(); |
1596 | ||
28e407b8 AA |
1597 | if (whoami == mdsmap->get_root()) { // load root inode off disk if we are auth |
1598 | mdcache->open_root_inode(gather.new_sub()); | |
1599 | } else if (is_any_replay()) { | |
1600 | // replay. make up fake root inode to start with | |
1601 | mdcache->create_root_inode(); | |
1602 | } | |
7c673cae FG |
1603 | gather.activate(); |
1604 | } | |
1605 | break; | |
1606 | case MDS_BOOT_PREPARE_LOG: | |
1607 | if (is_any_replay()) { | |
a8e16298 | 1608 | dout(2) << "Booting: " << step << ": replaying mds log" << dendl; |
3efd9988 FG |
1609 | MDSGatherBuilder gather(g_ceph_context, |
1610 | new C_MDS_BootStart(this, MDS_BOOT_REPLAY_DONE)); | |
1611 | ||
1612 | if (!standby_replaying) { | |
a8e16298 | 1613 | dout(2) << "Booting: " << step << ": waiting for purge queue recovered" << dendl; |
3efd9988 FG |
1614 | purge_queue.wait_for_recovery(new C_IO_Wrapper(this, gather.new_sub())); |
1615 | } | |
1616 | ||
1617 | mdlog->replay(gather.new_sub()); | |
1618 | gather.activate(); | |
7c673cae | 1619 | } else { |
a8e16298 | 1620 | dout(2) << "Booting: " << step << ": positioning at end of old mds log" << dendl; |
7c673cae FG |
1621 | mdlog->append(); |
1622 | starting_done(); | |
1623 | } | |
1624 | break; | |
1625 | case MDS_BOOT_REPLAY_DONE: | |
11fdf7f2 | 1626 | ceph_assert(is_any_replay()); |
7c673cae FG |
1627 | |
1628 | // Sessiontable and inotable should be in sync after replay, validate | |
1629 | // that they are consistent. | |
1630 | validate_sessions(); | |
1631 | ||
1632 | replay_done(); | |
1633 | break; | |
1634 | } | |
1635 | } | |
1636 | ||
1637 | void MDSRank::validate_sessions() | |
1638 | { | |
9f95a23c | 1639 | ceph_assert(ceph_mutex_is_locked_by_me(mds_lock)); |
28e407b8 | 1640 | bool valid = true; |
7c673cae FG |
1641 | |
1642 | // Identify any sessions which have state inconsistent with other, | |
1643 | // after they have been loaded from rados during startup. | |
1644 | // Mitigate bugs like: http://tracker.ceph.com/issues/16842 | |
11fdf7f2 | 1645 | for (const auto &i : sessionmap.get_sessions()) { |
7c673cae FG |
1646 | Session *session = i.second; |
1647 | interval_set<inodeno_t> badones; | |
1648 | if (inotable->intersects_free(session->info.prealloc_inos, &badones)) { | |
28e407b8 AA |
1649 | clog->error() << "client " << *session |
1650 | << "loaded with preallocated inodes that are inconsistent with inotable"; | |
1651 | valid = false; | |
7c673cae FG |
1652 | } |
1653 | } | |
1654 | ||
28e407b8 AA |
1655 | if (!valid) { |
1656 | damaged(); | |
11fdf7f2 | 1657 | ceph_assert(valid); |
7c673cae FG |
1658 | } |
1659 | } | |
1660 | ||
1661 | void MDSRank::starting_done() | |
1662 | { | |
1663 | dout(3) << "starting_done" << dendl; | |
11fdf7f2 | 1664 | ceph_assert(is_starting()); |
7c673cae FG |
1665 | request_state(MDSMap::STATE_ACTIVE); |
1666 | ||
28e407b8 | 1667 | mdlog->start_new_segment(); |
11fdf7f2 TL |
1668 | |
1669 | // sync snaptable cache | |
1670 | snapclient->sync(new C_MDSInternalNoop); | |
7c673cae FG |
1671 | } |
1672 | ||
1673 | ||
1674 | void MDSRank::calc_recovery_set() | |
1675 | { | |
1676 | // initialize gather sets | |
1677 | set<mds_rank_t> rs; | |
1678 | mdsmap->get_recovery_mds_set(rs); | |
1679 | rs.erase(whoami); | |
1680 | mdcache->set_recovery_set(rs); | |
1681 | ||
1682 | dout(1) << " recovery set is " << rs << dendl; | |
1683 | } | |
1684 | ||
1685 | ||
1686 | void MDSRank::replay_start() | |
1687 | { | |
1688 | dout(1) << "replay_start" << dendl; | |
1689 | ||
1690 | if (is_standby_replay()) | |
1691 | standby_replaying = true; | |
1692 | ||
1693 | calc_recovery_set(); | |
1694 | ||
1695 | // Check if we need to wait for a newer OSD map before starting | |
1696 | Context *fin = new C_IO_Wrapper(this, new C_MDS_BootStart(this, MDS_BOOT_INITIAL)); | |
1697 | bool const ready = objecter->wait_for_map( | |
1698 | mdsmap->get_last_failure_osd_epoch(), | |
1699 | fin); | |
1700 | ||
1701 | if (ready) { | |
1702 | delete fin; | |
1703 | boot_start(); | |
1704 | } else { | |
1705 | dout(1) << " waiting for osdmap " << mdsmap->get_last_failure_osd_epoch() | |
1706 | << " (which blacklists prior instance)" << dendl; | |
1707 | } | |
1708 | } | |
1709 | ||
1710 | ||
1711 | class MDSRank::C_MDS_StandbyReplayRestartFinish : public MDSIOContext { | |
1712 | uint64_t old_read_pos; | |
1713 | public: | |
1714 | C_MDS_StandbyReplayRestartFinish(MDSRank *mds_, uint64_t old_read_pos_) : | |
1715 | MDSIOContext(mds_), old_read_pos(old_read_pos_) {} | |
1716 | void finish(int r) override { | |
1717 | mds->_standby_replay_restart_finish(r, old_read_pos); | |
1718 | } | |
91327a77 AA |
1719 | void print(ostream& out) const override { |
1720 | out << "standby_replay_restart"; | |
1721 | } | |
7c673cae FG |
1722 | }; |
1723 | ||
1724 | void MDSRank::_standby_replay_restart_finish(int r, uint64_t old_read_pos) | |
1725 | { | |
1726 | if (old_read_pos < mdlog->get_journaler()->get_trimmed_pos()) { | |
1727 | dout(0) << "standby MDS fell behind active MDS journal's expire_pos, restarting" << dendl; | |
1728 | respawn(); /* we're too far back, and this is easier than | |
1729 | trying to reset everything in the cache, etc */ | |
1730 | } else { | |
1731 | mdlog->standby_trim_segments(); | |
1732 | boot_start(MDS_BOOT_PREPARE_LOG, r); | |
1733 | } | |
1734 | } | |
1735 | ||
3efd9988 FG |
1736 | class MDSRank::C_MDS_StandbyReplayRestart : public MDSInternalContext { |
1737 | public: | |
1738 | explicit C_MDS_StandbyReplayRestart(MDSRank *m) : MDSInternalContext(m) {} | |
1739 | void finish(int r) override { | |
11fdf7f2 | 1740 | ceph_assert(!r); |
3efd9988 FG |
1741 | mds->standby_replay_restart(); |
1742 | } | |
1743 | }; | |
1744 | ||
1745 | void MDSRank::standby_replay_restart() | |
7c673cae FG |
1746 | { |
1747 | if (standby_replaying) { | |
1748 | /* Go around for another pass of replaying in standby */ | |
f64942e4 | 1749 | dout(5) << "Restarting replay as standby-replay" << dendl; |
7c673cae FG |
1750 | mdlog->get_journaler()->reread_head_and_probe( |
1751 | new C_MDS_StandbyReplayRestartFinish( | |
1752 | this, | |
1753 | mdlog->get_journaler()->get_read_pos())); | |
1754 | } else { | |
1755 | /* We are transitioning out of standby: wait for OSD map update | |
1756 | before making final pass */ | |
1757 | dout(1) << "standby_replay_restart (final takeover pass)" << dendl; | |
3efd9988 FG |
1758 | Context *fin = new C_IO_Wrapper(this, new C_MDS_StandbyReplayRestart(this)); |
1759 | bool ready = objecter->wait_for_map(mdsmap->get_last_failure_osd_epoch(), fin); | |
7c673cae FG |
1760 | if (ready) { |
1761 | delete fin; | |
1762 | mdlog->get_journaler()->reread_head_and_probe( | |
1763 | new C_MDS_StandbyReplayRestartFinish( | |
1764 | this, | |
1765 | mdlog->get_journaler()->get_read_pos())); | |
3efd9988 | 1766 | |
11fdf7f2 | 1767 | dout(1) << " opening purge_queue (async)" << dendl; |
3efd9988 | 1768 | purge_queue.open(NULL); |
11fdf7f2 TL |
1769 | dout(1) << " opening open_file_table (async)" << dendl; |
1770 | mdcache->open_file_table.load(nullptr); | |
7c673cae FG |
1771 | } else { |
1772 | dout(1) << " waiting for osdmap " << mdsmap->get_last_failure_osd_epoch() | |
1773 | << " (which blacklists prior instance)" << dendl; | |
1774 | } | |
1775 | } | |
1776 | } | |
1777 | ||
7c673cae FG |
1778 | void MDSRank::replay_done() |
1779 | { | |
f64942e4 AA |
1780 | if (!standby_replaying) { |
1781 | dout(1) << "Finished replaying journal" << dendl; | |
1782 | } else { | |
1783 | dout(5) << "Finished replaying journal as standby-replay" << dendl; | |
1784 | } | |
7c673cae FG |
1785 | |
1786 | if (is_standby_replay()) { | |
1787 | // The replay was done in standby state, and we are still in that state | |
11fdf7f2 | 1788 | ceph_assert(standby_replaying); |
7c673cae | 1789 | dout(10) << "setting replay timer" << dendl; |
11fdf7f2 | 1790 | timer.add_event_after(g_conf()->mds_replay_interval, |
7c673cae FG |
1791 | new C_MDS_StandbyReplayRestart(this)); |
1792 | return; | |
1793 | } else if (standby_replaying) { | |
1794 | // The replay was done in standby state, we have now _left_ that state | |
1795 | dout(10) << " last replay pass was as a standby; making final pass" << dendl; | |
1796 | standby_replaying = false; | |
1797 | standby_replay_restart(); | |
1798 | return; | |
1799 | } else { | |
1800 | // Replay is complete, journal read should be up to date | |
11fdf7f2 TL |
1801 | ceph_assert(mdlog->get_journaler()->get_read_pos() == mdlog->get_journaler()->get_write_pos()); |
1802 | ceph_assert(!is_standby_replay()); | |
7c673cae FG |
1803 | |
1804 | // Reformat and come back here | |
11fdf7f2 | 1805 | if (mdlog->get_journaler()->get_stream_format() < g_conf()->mds_journal_format) { |
f64942e4 | 1806 | dout(4) << "reformatting journal on standby-replay->replay transition" << dendl; |
7c673cae FG |
1807 | mdlog->reopen(new C_MDS_BootStart(this, MDS_BOOT_REPLAY_DONE)); |
1808 | return; | |
1809 | } | |
1810 | } | |
1811 | ||
1812 | dout(1) << "making mds journal writeable" << dendl; | |
1813 | mdlog->get_journaler()->set_writeable(); | |
1814 | mdlog->get_journaler()->trim_tail(); | |
1815 | ||
11fdf7f2 TL |
1816 | if (mdsmap->get_tableserver() == whoami && |
1817 | snapserver->upgrade_format()) { | |
1818 | dout(1) << "upgrading snaptable format" << dendl; | |
1819 | snapserver->save(new C_MDSInternalNoop); | |
1820 | } | |
1821 | ||
1822 | if (g_conf()->mds_wipe_sessions) { | |
7c673cae FG |
1823 | dout(1) << "wiping out client sessions" << dendl; |
1824 | sessionmap.wipe(); | |
1825 | sessionmap.save(new C_MDSInternalNoop); | |
1826 | } | |
11fdf7f2 | 1827 | if (g_conf()->mds_wipe_ino_prealloc) { |
7c673cae FG |
1828 | dout(1) << "wiping out ino prealloc from sessions" << dendl; |
1829 | sessionmap.wipe_ino_prealloc(); | |
1830 | sessionmap.save(new C_MDSInternalNoop); | |
1831 | } | |
11fdf7f2 TL |
1832 | if (g_conf()->mds_skip_ino) { |
1833 | inodeno_t i = g_conf()->mds_skip_ino; | |
7c673cae FG |
1834 | dout(1) << "skipping " << i << " inodes" << dendl; |
1835 | inotable->skip_inos(i); | |
1836 | inotable->save(new C_MDSInternalNoop); | |
1837 | } | |
1838 | ||
1839 | if (mdsmap->get_num_in_mds() == 1 && | |
1840 | mdsmap->get_num_failed_mds() == 0) { // just me! | |
1841 | dout(2) << "i am alone, moving to state reconnect" << dendl; | |
1842 | request_state(MDSMap::STATE_RECONNECT); | |
11fdf7f2 TL |
1843 | // sync snaptable cache |
1844 | snapclient->sync(new C_MDSInternalNoop); | |
7c673cae FG |
1845 | } else { |
1846 | dout(2) << "i am not alone, moving to state resolve" << dendl; | |
1847 | request_state(MDSMap::STATE_RESOLVE); | |
1848 | } | |
1849 | } | |
1850 | ||
1851 | void MDSRank::reopen_log() | |
1852 | { | |
1853 | dout(1) << "reopen_log" << dendl; | |
1854 | mdcache->rollback_uncommitted_fragments(); | |
1855 | } | |
1856 | ||
7c673cae FG |
1857 | void MDSRank::resolve_start() |
1858 | { | |
1859 | dout(1) << "resolve_start" << dendl; | |
1860 | ||
1861 | reopen_log(); | |
1862 | ||
1863 | mdcache->resolve_start(new C_MDS_VoidFn(this, &MDSRank::resolve_done)); | |
1864 | finish_contexts(g_ceph_context, waiting_for_resolve); | |
1865 | } | |
11fdf7f2 | 1866 | |
7c673cae FG |
1867 | void MDSRank::resolve_done() |
1868 | { | |
1869 | dout(1) << "resolve_done" << dendl; | |
1870 | request_state(MDSMap::STATE_RECONNECT); | |
11fdf7f2 TL |
1871 | // sync snaptable cache |
1872 | snapclient->sync(new C_MDSInternalNoop); | |
7c673cae FG |
1873 | } |
1874 | ||
1875 | void MDSRank::reconnect_start() | |
1876 | { | |
1877 | dout(1) << "reconnect_start" << dendl; | |
1878 | ||
1879 | if (last_state == MDSMap::STATE_REPLAY) { | |
1880 | reopen_log(); | |
1881 | } | |
1882 | ||
31f18b77 FG |
1883 | // Drop any blacklisted clients from the SessionMap before going |
1884 | // into reconnect, so that we don't wait for them. | |
1885 | objecter->enable_blacklist_events(); | |
1886 | std::set<entity_addr_t> blacklist; | |
1887 | epoch_t epoch = 0; | |
11fdf7f2 | 1888 | objecter->with_osdmap([&blacklist, &epoch](const OSDMap& o) { |
31f18b77 FG |
1889 | o.get_blacklist(&blacklist); |
1890 | epoch = o.get_epoch(); | |
1891 | }); | |
1892 | auto killed = server->apply_blacklist(blacklist); | |
1893 | dout(4) << "reconnect_start: killed " << killed << " blacklisted sessions (" | |
1894 | << blacklist.size() << " blacklist entries, " | |
1895 | << sessionmap.get_sessions().size() << ")" << dendl; | |
1896 | if (killed) { | |
1897 | set_osd_epoch_barrier(epoch); | |
1898 | } | |
1899 | ||
7c673cae FG |
1900 | server->reconnect_clients(new C_MDS_VoidFn(this, &MDSRank::reconnect_done)); |
1901 | finish_contexts(g_ceph_context, waiting_for_reconnect); | |
1902 | } | |
1903 | void MDSRank::reconnect_done() | |
1904 | { | |
1905 | dout(1) << "reconnect_done" << dendl; | |
1906 | request_state(MDSMap::STATE_REJOIN); // move to rejoin state | |
1907 | } | |
1908 | ||
1909 | void MDSRank::rejoin_joint_start() | |
1910 | { | |
1911 | dout(1) << "rejoin_joint_start" << dendl; | |
1912 | mdcache->rejoin_send_rejoins(); | |
1913 | } | |
1914 | void MDSRank::rejoin_start() | |
1915 | { | |
1916 | dout(1) << "rejoin_start" << dendl; | |
1917 | mdcache->rejoin_start(new C_MDS_VoidFn(this, &MDSRank::rejoin_done)); | |
a8e16298 | 1918 | finish_contexts(g_ceph_context, waiting_for_rejoin); |
7c673cae FG |
1919 | } |
1920 | void MDSRank::rejoin_done() | |
1921 | { | |
1922 | dout(1) << "rejoin_done" << dendl; | |
1923 | mdcache->show_subtrees(); | |
1924 | mdcache->show_cache(); | |
1925 | ||
e306af50 TL |
1926 | if (mdcache->is_any_uncommitted_fragment()) { |
1927 | dout(1) << " waiting for uncommitted fragments" << dendl; | |
1928 | MDSGatherBuilder gather(g_ceph_context, new C_MDS_VoidFn(this, &MDSRank::rejoin_done)); | |
1929 | mdcache->wait_for_uncommitted_fragments(gather.get()); | |
1930 | gather.activate(); | |
1931 | return; | |
1932 | } | |
1933 | ||
7c673cae FG |
1934 | // funny case: is our cache empty? no subtrees? |
1935 | if (!mdcache->is_subtrees()) { | |
1936 | if (whoami == 0) { | |
1937 | // The root should always have a subtree! | |
1938 | clog->error() << "No subtrees found for root MDS rank!"; | |
1939 | damaged(); | |
11fdf7f2 | 1940 | ceph_assert(mdcache->is_subtrees()); |
7c673cae FG |
1941 | } else { |
1942 | dout(1) << " empty cache, no subtrees, leaving cluster" << dendl; | |
1943 | request_state(MDSMap::STATE_STOPPED); | |
1944 | } | |
1945 | return; | |
1946 | } | |
1947 | ||
11fdf7f2 | 1948 | if (replay_queue.empty() && !server->get_num_pending_reclaim()) { |
7c673cae | 1949 | request_state(MDSMap::STATE_ACTIVE); |
11fdf7f2 TL |
1950 | } else { |
1951 | replaying_requests_done = replay_queue.empty(); | |
7c673cae | 1952 | request_state(MDSMap::STATE_CLIENTREPLAY); |
11fdf7f2 | 1953 | } |
7c673cae FG |
1954 | } |
1955 | ||
1956 | void MDSRank::clientreplay_start() | |
1957 | { | |
1958 | dout(1) << "clientreplay_start" << dendl; | |
1959 | finish_contexts(g_ceph_context, waiting_for_replay); // kick waiters | |
1960 | mdcache->start_files_to_recover(); | |
1961 | queue_one_replay(); | |
1962 | } | |
1963 | ||
1964 | bool MDSRank::queue_one_replay() | |
1965 | { | |
11fdf7f2 TL |
1966 | if (!replay_queue.empty()) { |
1967 | queue_waiter(replay_queue.front()); | |
1968 | replay_queue.pop_front(); | |
1969 | return true; | |
1970 | } | |
1971 | if (!replaying_requests_done) { | |
1972 | replaying_requests_done = true; | |
1973 | mdlog->flush(); | |
1974 | } | |
1975 | maybe_clientreplay_done(); | |
1976 | return false; | |
1977 | } | |
1978 | ||
1979 | void MDSRank::maybe_clientreplay_done() | |
1980 | { | |
1981 | if (is_clientreplay() && get_want_state() == MDSMap::STATE_CLIENTREPLAY) { | |
1982 | ||
1983 | // don't go to active if there are session waiting for being reclaimed | |
1984 | if (replaying_requests_done && !server->get_num_pending_reclaim()) { | |
1985 | mdlog->wait_for_safe(new C_MDS_VoidFn(this, &MDSRank::clientreplay_done)); | |
1986 | return; | |
1987 | } | |
1988 | ||
1989 | dout(1) << " still have " << replay_queue.size() + (int)!replaying_requests_done | |
1990 | << " requests need to be replayed, " << server->get_num_pending_reclaim() | |
1991 | << " sessions need to be reclaimed" << dendl; | |
7c673cae | 1992 | } |
7c673cae FG |
1993 | } |
1994 | ||
1995 | void MDSRank::clientreplay_done() | |
1996 | { | |
1997 | dout(1) << "clientreplay_done" << dendl; | |
1998 | request_state(MDSMap::STATE_ACTIVE); | |
1999 | } | |
2000 | ||
2001 | void MDSRank::active_start() | |
2002 | { | |
2003 | dout(1) << "active_start" << dendl; | |
2004 | ||
28e407b8 AA |
2005 | if (last_state == MDSMap::STATE_CREATING || |
2006 | last_state == MDSMap::STATE_STARTING) { | |
7c673cae FG |
2007 | mdcache->open_root(); |
2008 | } | |
2009 | ||
2010 | mdcache->clean_open_file_lists(); | |
2011 | mdcache->export_remaining_imported_caps(); | |
2012 | finish_contexts(g_ceph_context, waiting_for_replay); // kick waiters | |
2013 | mdcache->start_files_to_recover(); | |
2014 | ||
2015 | mdcache->reissue_all_caps(); | |
7c673cae FG |
2016 | |
2017 | finish_contexts(g_ceph_context, waiting_for_active); // kick waiters | |
2018 | } | |
2019 | ||
2020 | void MDSRank::recovery_done(int oldstate) | |
2021 | { | |
2022 | dout(1) << "recovery_done -- successful recovery!" << dendl; | |
11fdf7f2 | 2023 | ceph_assert(is_clientreplay() || is_active()); |
7c673cae FG |
2024 | |
2025 | if (oldstate == MDSMap::STATE_CREATING) | |
2026 | return; | |
2027 | ||
2028 | mdcache->start_recovered_truncates(); | |
9f95a23c | 2029 | mdcache->start_purge_inodes(); |
7c673cae FG |
2030 | mdcache->do_file_recover(); |
2031 | ||
2032 | // tell connected clients | |
2033 | //bcast_mds_map(); // not anymore, they get this from the monitor | |
2034 | ||
2035 | mdcache->populate_mydir(); | |
2036 | } | |
2037 | ||
2038 | void MDSRank::creating_done() | |
2039 | { | |
2040 | dout(1)<< "creating_done" << dendl; | |
2041 | request_state(MDSMap::STATE_ACTIVE); | |
11fdf7f2 TL |
2042 | // sync snaptable cache |
2043 | snapclient->sync(new C_MDSInternalNoop); | |
7c673cae FG |
2044 | } |
2045 | ||
2046 | void MDSRank::boot_create() | |
2047 | { | |
2048 | dout(3) << "boot_create" << dendl; | |
2049 | ||
2050 | MDSGatherBuilder fin(g_ceph_context, new C_MDS_VoidFn(this, &MDSRank::creating_done)); | |
2051 | ||
2052 | mdcache->init_layouts(); | |
2053 | ||
7c673cae FG |
2054 | inotable->set_rank(whoami); |
2055 | sessionmap.set_rank(whoami); | |
2056 | ||
2057 | // start with a fresh journal | |
2058 | dout(10) << "boot_create creating fresh journal" << dendl; | |
2059 | mdlog->create(fin.new_sub()); | |
2060 | ||
2061 | // open new journal segment, but do not journal subtree map (yet) | |
2062 | mdlog->prepare_new_segment(); | |
2063 | ||
2064 | if (whoami == mdsmap->get_root()) { | |
2065 | dout(3) << "boot_create creating fresh hierarchy" << dendl; | |
2066 | mdcache->create_empty_hierarchy(fin.get()); | |
2067 | } | |
2068 | ||
2069 | dout(3) << "boot_create creating mydir hierarchy" << dendl; | |
2070 | mdcache->create_mydir_hierarchy(fin.get()); | |
2071 | ||
11fdf7f2 TL |
2072 | dout(3) << "boot_create creating global snaprealm" << dendl; |
2073 | mdcache->create_global_snaprealm(); | |
2074 | ||
7c673cae FG |
2075 | // fixme: fake out inotable (reset, pretend loaded) |
2076 | dout(10) << "boot_create creating fresh inotable table" << dendl; | |
2077 | inotable->reset(); | |
2078 | inotable->save(fin.new_sub()); | |
2079 | ||
2080 | // write empty sessionmap | |
2081 | sessionmap.save(fin.new_sub()); | |
2082 | ||
2083 | // Create empty purge queue | |
2084 | purge_queue.create(new C_IO_Wrapper(this, fin.new_sub())); | |
2085 | ||
2086 | // initialize tables | |
2087 | if (mdsmap->get_tableserver() == whoami) { | |
2088 | dout(10) << "boot_create creating fresh snaptable" << dendl; | |
11fdf7f2 | 2089 | snapserver->set_rank(whoami); |
7c673cae FG |
2090 | snapserver->reset(); |
2091 | snapserver->save(fin.new_sub()); | |
2092 | } | |
2093 | ||
11fdf7f2 | 2094 | ceph_assert(g_conf()->mds_kill_create_at != 1); |
7c673cae FG |
2095 | |
2096 | // ok now journal it | |
2097 | mdlog->journal_segment_subtree_map(fin.new_sub()); | |
2098 | mdlog->flush(); | |
2099 | ||
31f18b77 FG |
2100 | // Usually we do this during reconnect, but creation skips that. |
2101 | objecter->enable_blacklist_events(); | |
2102 | ||
7c673cae FG |
2103 | fin.activate(); |
2104 | } | |
2105 | ||
2106 | void MDSRank::stopping_start() | |
2107 | { | |
a8e16298 | 2108 | dout(2) << "Stopping..." << dendl; |
7c673cae FG |
2109 | |
2110 | if (mdsmap->get_num_in_mds() == 1 && !sessionmap.empty()) { | |
11fdf7f2 TL |
2111 | std::vector<Session*> victims; |
2112 | const auto& sessions = sessionmap.get_sessions(); | |
2113 | for (const auto& p : sessions) { | |
2114 | if (!p.first.is_client()) { | |
2115 | continue; | |
2116 | } | |
2117 | ||
2118 | Session *s = p.second; | |
2119 | victims.push_back(s); | |
2120 | } | |
2121 | ||
2122 | dout(20) << __func__ << " matched " << victims.size() << " sessions" << dendl; | |
2123 | ceph_assert(!victims.empty()); | |
2124 | ||
2125 | C_GatherBuilder gather(g_ceph_context, new C_MDSInternalNoop); | |
2126 | for (const auto &s : victims) { | |
2127 | std::stringstream ss; | |
2128 | evict_client(s->get_client().v, false, | |
2129 | g_conf()->mds_session_blacklist_on_evict, ss, gather.new_sub()); | |
2130 | } | |
2131 | gather.activate(); | |
7c673cae FG |
2132 | } |
2133 | ||
2134 | mdcache->shutdown_start(); | |
2135 | } | |
2136 | ||
2137 | void MDSRank::stopping_done() | |
2138 | { | |
a8e16298 | 2139 | dout(2) << "Finished stopping..." << dendl; |
7c673cae FG |
2140 | |
2141 | // tell monitor we shut down cleanly. | |
2142 | request_state(MDSMap::STATE_STOPPED); | |
2143 | } | |
2144 | ||
2145 | void MDSRankDispatcher::handle_mds_map( | |
9f95a23c | 2146 | const cref_t<MMDSMap> &m, |
11fdf7f2 | 2147 | const MDSMap &oldmap) |
7c673cae FG |
2148 | { |
2149 | // I am only to be passed MDSMaps in which I hold a rank | |
11fdf7f2 | 2150 | ceph_assert(whoami != MDS_RANK_NONE); |
7c673cae FG |
2151 | |
2152 | MDSMap::DaemonState oldstate = state; | |
2153 | mds_gid_t mds_gid = mds_gid_t(monc->get_global_id()); | |
2154 | state = mdsmap->get_state_gid(mds_gid); | |
2155 | if (state != oldstate) { | |
2156 | last_state = oldstate; | |
2157 | incarnation = mdsmap->get_inc_gid(mds_gid); | |
2158 | } | |
2159 | ||
2160 | version_t epoch = m->get_epoch(); | |
2161 | ||
2162 | // note source's map version | |
2163 | if (m->get_source().is_mds() && | |
2164 | peer_mdsmap_epoch[mds_rank_t(m->get_source().num())] < epoch) { | |
2165 | dout(15) << " peer " << m->get_source() | |
2166 | << " has mdsmap epoch >= " << epoch | |
2167 | << dendl; | |
2168 | peer_mdsmap_epoch[mds_rank_t(m->get_source().num())] = epoch; | |
2169 | } | |
2170 | ||
2171 | // Validate state transitions while I hold a rank | |
2172 | if (!MDSMap::state_transition_valid(oldstate, state)) { | |
2173 | derr << "Invalid state transition " << ceph_mds_state_name(oldstate) | |
2174 | << "->" << ceph_mds_state_name(state) << dendl; | |
2175 | respawn(); | |
2176 | } | |
2177 | ||
2178 | if (oldstate != state) { | |
2179 | // update messenger. | |
2180 | if (state == MDSMap::STATE_STANDBY_REPLAY) { | |
2181 | dout(1) << "handle_mds_map i am now mds." << mds_gid << "." << incarnation | |
2182 | << " replaying mds." << whoami << "." << incarnation << dendl; | |
2183 | messenger->set_myname(entity_name_t::MDS(mds_gid)); | |
2184 | } else { | |
2185 | dout(1) << "handle_mds_map i am now mds." << whoami << "." << incarnation << dendl; | |
2186 | messenger->set_myname(entity_name_t::MDS(whoami)); | |
2187 | } | |
2188 | } | |
2189 | ||
2190 | // tell objecter my incarnation | |
2191 | if (objecter->get_client_incarnation() != incarnation) | |
2192 | objecter->set_client_incarnation(incarnation); | |
2193 | ||
9f95a23c TL |
2194 | if (mdsmap->get_min_compat_client() < ceph_release_t::max && |
2195 | oldmap.get_min_compat_client() != mdsmap->get_min_compat_client()) | |
11fdf7f2 TL |
2196 | server->update_required_client_features(); |
2197 | ||
7c673cae | 2198 | // for debug |
11fdf7f2 | 2199 | if (g_conf()->mds_dump_cache_on_map) |
7c673cae FG |
2200 | mdcache->dump_cache(); |
2201 | ||
1adf2230 AA |
2202 | cluster_degraded = mdsmap->is_degraded(); |
2203 | ||
2204 | // mdsmap and oldmap can be discontinuous. failover might happen in the missing mdsmap. | |
2205 | // the 'restart' set tracks ranks that have restarted since the old mdsmap | |
2206 | set<mds_rank_t> restart; | |
2207 | // replaying mds does not communicate with other ranks | |
2208 | if (state >= MDSMap::STATE_RESOLVE) { | |
2209 | // did someone fail? | |
2210 | // new down? | |
2211 | set<mds_rank_t> olddown, down; | |
11fdf7f2 | 2212 | oldmap.get_down_mds_set(&olddown); |
1adf2230 AA |
2213 | mdsmap->get_down_mds_set(&down); |
2214 | for (const auto& r : down) { | |
11fdf7f2 TL |
2215 | if (oldmap.have_inst(r) && olddown.count(r) == 0) { |
2216 | messenger->mark_down_addrs(oldmap.get_addrs(r)); | |
1adf2230 AA |
2217 | handle_mds_failure(r); |
2218 | } | |
2219 | } | |
2220 | ||
2221 | // did someone fail? | |
2222 | // did their addr/inst change? | |
2223 | set<mds_rank_t> up; | |
2224 | mdsmap->get_up_mds_set(up); | |
2225 | for (const auto& r : up) { | |
2226 | auto& info = mdsmap->get_info(r); | |
11fdf7f2 TL |
2227 | if (oldmap.have_inst(r)) { |
2228 | auto& oldinfo = oldmap.get_info(r); | |
1adf2230 | 2229 | if (info.inc != oldinfo.inc) { |
11fdf7f2 | 2230 | messenger->mark_down_addrs(oldinfo.get_addrs()); |
1adf2230 AA |
2231 | if (info.state == MDSMap::STATE_REPLAY || |
2232 | info.state == MDSMap::STATE_RESOLVE) { | |
2233 | restart.insert(r); | |
2234 | handle_mds_failure(r); | |
2235 | } else { | |
11fdf7f2 | 2236 | ceph_assert(info.state == MDSMap::STATE_STARTING || |
1adf2230 AA |
2237 | info.state == MDSMap::STATE_ACTIVE); |
2238 | // -> stopped (missing) -> starting -> active | |
2239 | restart.insert(r); | |
2240 | mdcache->migrator->handle_mds_failure_or_stop(r); | |
11fdf7f2 TL |
2241 | if (mdsmap->get_tableserver() == whoami) |
2242 | snapserver->handle_mds_failure_or_stop(r); | |
1adf2230 AA |
2243 | } |
2244 | } | |
2245 | } else { | |
2246 | if (info.state == MDSMap::STATE_REPLAY || | |
2247 | info.state == MDSMap::STATE_RESOLVE) { | |
2248 | // -> starting/creating (missing) -> active (missing) -> replay -> resolve | |
2249 | restart.insert(r); | |
2250 | handle_mds_failure(r); | |
2251 | } else { | |
11fdf7f2 | 2252 | ceph_assert(info.state == MDSMap::STATE_CREATING || |
1adf2230 AA |
2253 | info.state == MDSMap::STATE_STARTING || |
2254 | info.state == MDSMap::STATE_ACTIVE); | |
2255 | } | |
2256 | } | |
2257 | } | |
2258 | } | |
2259 | ||
7c673cae FG |
2260 | // did it change? |
2261 | if (oldstate != state) { | |
2262 | dout(1) << "handle_mds_map state change " | |
2263 | << ceph_mds_state_name(oldstate) << " --> " | |
2264 | << ceph_mds_state_name(state) << dendl; | |
11fdf7f2 | 2265 | beacon.set_want_state(*mdsmap, state); |
7c673cae FG |
2266 | |
2267 | if (oldstate == MDSMap::STATE_STANDBY_REPLAY) { | |
2268 | dout(10) << "Monitor activated us! Deactivating replay loop" << dendl; | |
2269 | assert (state == MDSMap::STATE_REPLAY); | |
2270 | } else { | |
2271 | // did i just recover? | |
2272 | if ((is_active() || is_clientreplay()) && | |
2273 | (oldstate == MDSMap::STATE_CREATING || | |
2274 | oldstate == MDSMap::STATE_REJOIN || | |
2275 | oldstate == MDSMap::STATE_RECONNECT)) | |
2276 | recovery_done(oldstate); | |
2277 | ||
2278 | if (is_active()) { | |
2279 | active_start(); | |
2280 | } else if (is_any_replay()) { | |
2281 | replay_start(); | |
2282 | } else if (is_resolve()) { | |
2283 | resolve_start(); | |
2284 | } else if (is_reconnect()) { | |
2285 | reconnect_start(); | |
2286 | } else if (is_rejoin()) { | |
2287 | rejoin_start(); | |
2288 | } else if (is_clientreplay()) { | |
2289 | clientreplay_start(); | |
2290 | } else if (is_creating()) { | |
2291 | boot_create(); | |
2292 | } else if (is_starting()) { | |
2293 | boot_start(); | |
2294 | } else if (is_stopping()) { | |
11fdf7f2 | 2295 | ceph_assert(oldstate == MDSMap::STATE_ACTIVE); |
7c673cae FG |
2296 | stopping_start(); |
2297 | } | |
2298 | } | |
2299 | } | |
2300 | ||
2301 | // RESOLVE | |
2302 | // is someone else newly resolving? | |
1adf2230 | 2303 | if (state >= MDSMap::STATE_RESOLVE) { |
11fdf7f2 TL |
2304 | // recover snaptable |
2305 | if (mdsmap->get_tableserver() == whoami) { | |
2306 | if (oldstate < MDSMap::STATE_RESOLVE) { | |
2307 | set<mds_rank_t> s; | |
2308 | mdsmap->get_mds_set_lower_bound(s, MDSMap::STATE_RESOLVE); | |
2309 | snapserver->finish_recovery(s); | |
2310 | } else { | |
2311 | set<mds_rank_t> old_set, new_set; | |
2312 | oldmap.get_mds_set_lower_bound(old_set, MDSMap::STATE_RESOLVE); | |
2313 | mdsmap->get_mds_set_lower_bound(new_set, MDSMap::STATE_RESOLVE); | |
2314 | for (const auto& r : new_set) { | |
2315 | if (r == whoami) | |
2316 | continue; // not me | |
2317 | if (!old_set.count(r) || restart.count(r)) { // newly so? | |
2318 | snapserver->handle_mds_recovery(r); | |
2319 | } | |
2320 | } | |
2321 | } | |
2322 | } | |
2323 | ||
2324 | if ((!oldmap.is_resolving() || !restart.empty()) && mdsmap->is_resolving()) { | |
7c673cae FG |
2325 | set<mds_rank_t> resolve; |
2326 | mdsmap->get_mds_set(resolve, MDSMap::STATE_RESOLVE); | |
2327 | dout(10) << " resolve set is " << resolve << dendl; | |
2328 | calc_recovery_set(); | |
2329 | mdcache->send_resolves(); | |
2330 | } | |
2331 | } | |
2332 | ||
2333 | // REJOIN | |
2334 | // is everybody finally rejoining? | |
1adf2230 | 2335 | if (state >= MDSMap::STATE_REJOIN) { |
7c673cae | 2336 | // did we start? |
11fdf7f2 | 2337 | if (!oldmap.is_rejoining() && mdsmap->is_rejoining()) |
7c673cae FG |
2338 | rejoin_joint_start(); |
2339 | ||
2340 | // did we finish? | |
11fdf7f2 TL |
2341 | if (g_conf()->mds_dump_cache_after_rejoin && |
2342 | oldmap.is_rejoining() && !mdsmap->is_rejoining()) | |
7c673cae FG |
2343 | mdcache->dump_cache(); // for DEBUG only |
2344 | ||
d2e6a577 FG |
2345 | if (oldstate >= MDSMap::STATE_REJOIN || |
2346 | oldstate == MDSMap::STATE_STARTING) { | |
7c673cae FG |
2347 | // ACTIVE|CLIENTREPLAY|REJOIN => we can discover from them. |
2348 | set<mds_rank_t> olddis, dis; | |
11fdf7f2 | 2349 | oldmap.get_mds_set_lower_bound(olddis, MDSMap::STATE_REJOIN); |
1adf2230 AA |
2350 | mdsmap->get_mds_set_lower_bound(dis, MDSMap::STATE_REJOIN); |
2351 | for (const auto& r : dis) { | |
2352 | if (r == whoami) | |
2353 | continue; // not me | |
2354 | if (!olddis.count(r) || restart.count(r)) { // newly so? | |
2355 | mdcache->kick_discovers(r); | |
2356 | mdcache->kick_open_ino_peers(r); | |
7c673cae | 2357 | } |
1adf2230 | 2358 | } |
7c673cae FG |
2359 | } |
2360 | } | |
2361 | ||
11fdf7f2 | 2362 | if (oldmap.is_degraded() && !cluster_degraded && state >= MDSMap::STATE_ACTIVE) { |
7c673cae FG |
2363 | dout(1) << "cluster recovered." << dendl; |
2364 | auto it = waiting_for_active_peer.find(MDS_RANK_NONE); | |
2365 | if (it != waiting_for_active_peer.end()) { | |
2366 | queue_waiters(it->second); | |
2367 | waiting_for_active_peer.erase(it); | |
2368 | } | |
2369 | } | |
2370 | ||
2371 | // did someone go active? | |
1adf2230 AA |
2372 | if (state >= MDSMap::STATE_CLIENTREPLAY && |
2373 | oldstate >= MDSMap::STATE_CLIENTREPLAY) { | |
7c673cae | 2374 | set<mds_rank_t> oldactive, active; |
11fdf7f2 | 2375 | oldmap.get_mds_set_lower_bound(oldactive, MDSMap::STATE_CLIENTREPLAY); |
1adf2230 AA |
2376 | mdsmap->get_mds_set_lower_bound(active, MDSMap::STATE_CLIENTREPLAY); |
2377 | for (const auto& r : active) { | |
2378 | if (r == whoami) | |
2379 | continue; // not me | |
2380 | if (!oldactive.count(r) || restart.count(r)) // newly so? | |
2381 | handle_mds_recovery(r); | |
7c673cae FG |
2382 | } |
2383 | } | |
2384 | ||
11fdf7f2 | 2385 | if (is_clientreplay() || is_active() || is_stopping()) { |
7c673cae FG |
2386 | // did anyone stop? |
2387 | set<mds_rank_t> oldstopped, stopped; | |
11fdf7f2 | 2388 | oldmap.get_stopped_mds_set(oldstopped); |
7c673cae | 2389 | mdsmap->get_stopped_mds_set(stopped); |
1adf2230 | 2390 | for (const auto& r : stopped) |
11fdf7f2 | 2391 | if (oldstopped.count(r) == 0) { // newly so? |
1adf2230 | 2392 | mdcache->migrator->handle_mds_failure_or_stop(r); |
11fdf7f2 TL |
2393 | if (mdsmap->get_tableserver() == whoami) |
2394 | snapserver->handle_mds_failure_or_stop(r); | |
2395 | } | |
7c673cae FG |
2396 | } |
2397 | ||
2398 | { | |
11fdf7f2 | 2399 | map<epoch_t,MDSContext::vec >::iterator p = waiting_for_mdsmap.begin(); |
7c673cae | 2400 | while (p != waiting_for_mdsmap.end() && p->first <= mdsmap->get_epoch()) { |
11fdf7f2 | 2401 | MDSContext::vec ls; |
7c673cae FG |
2402 | ls.swap(p->second); |
2403 | waiting_for_mdsmap.erase(p++); | |
91327a77 | 2404 | queue_waiters(ls); |
7c673cae FG |
2405 | } |
2406 | } | |
2407 | ||
2408 | if (is_active()) { | |
2409 | // Before going active, set OSD epoch barrier to latest (so that | |
2410 | // we don't risk handing out caps to clients with old OSD maps that | |
2411 | // might not include barriers from the previous incarnation of this MDS) | |
2412 | set_osd_epoch_barrier(objecter->with_osdmap( | |
2413 | std::mem_fn(&OSDMap::get_epoch))); | |
7c673cae | 2414 | |
11fdf7f2 TL |
2415 | /* Now check if we should hint to the OSD that a read may follow */ |
2416 | if (mdsmap->has_standby_replay(whoami)) | |
2417 | mdlog->set_write_iohint(0); | |
2418 | else | |
2419 | mdlog->set_write_iohint(CEPH_OSD_OP_FLAG_FADVISE_DONTNEED); | |
7c673cae FG |
2420 | } |
2421 | ||
11fdf7f2 | 2422 | if (oldmap.get_max_mds() != mdsmap->get_max_mds()) { |
7c673cae FG |
2423 | purge_queue.update_op_limit(*mdsmap); |
2424 | } | |
eafe8130 | 2425 | |
9f95a23c TL |
2426 | if (mdsmap->get_inline_data_enabled() && !oldmap.get_inline_data_enabled()) |
2427 | dout(0) << "WARNING: inline_data support has been deprecated and will be removed in a future release" << dendl; | |
2428 | ||
92f5a8d4 TL |
2429 | if (scrubstack->is_scrubbing()) { |
2430 | if (mdsmap->get_max_mds() > 1) { | |
2431 | auto c = new C_MDSInternalNoop; | |
2432 | scrubstack->scrub_abort(c); | |
2433 | } | |
2434 | } | |
f6b5b4d7 | 2435 | mdcache->handle_mdsmap(*mdsmap, oldmap); |
7c673cae FG |
2436 | } |
2437 | ||
2438 | void MDSRank::handle_mds_recovery(mds_rank_t who) | |
2439 | { | |
2440 | dout(5) << "handle_mds_recovery mds." << who << dendl; | |
2441 | ||
2442 | mdcache->handle_mds_recovery(who); | |
2443 | ||
7c673cae FG |
2444 | queue_waiters(waiting_for_active_peer[who]); |
2445 | waiting_for_active_peer.erase(who); | |
2446 | } | |
2447 | ||
2448 | void MDSRank::handle_mds_failure(mds_rank_t who) | |
2449 | { | |
2450 | if (who == whoami) { | |
2451 | dout(5) << "handle_mds_failure for myself; not doing anything" << dendl; | |
2452 | return; | |
2453 | } | |
2454 | dout(5) << "handle_mds_failure mds." << who << dendl; | |
2455 | ||
2456 | mdcache->handle_mds_failure(who); | |
2457 | ||
11fdf7f2 TL |
2458 | if (mdsmap->get_tableserver() == whoami) |
2459 | snapserver->handle_mds_failure_or_stop(who); | |
2460 | ||
7c673cae FG |
2461 | snapclient->handle_mds_failure(who); |
2462 | } | |
2463 | ||
9f95a23c TL |
2464 | void MDSRankDispatcher::handle_asok_command( |
2465 | std::string_view command, | |
2466 | const cmdmap_t& cmdmap, | |
2467 | Formatter *f, | |
2468 | const bufferlist &inbl, | |
2469 | std::function<void(int,const std::string&,bufferlist&)> on_finish) | |
7c673cae | 2470 | { |
9f95a23c TL |
2471 | int r = 0; |
2472 | stringstream ss; | |
2473 | bufferlist outbl; | |
7c673cae | 2474 | if (command == "dump_ops_in_flight" || |
9f95a23c | 2475 | command == "ops") { |
7c673cae | 2476 | if (!op_tracker.dump_ops_in_flight(f)) { |
9f95a23c | 2477 | ss << "op_tracker disabled; set mds_enable_op_tracker=true to enable"; |
7c673cae FG |
2478 | } |
2479 | } else if (command == "dump_blocked_ops") { | |
2480 | if (!op_tracker.dump_ops_in_flight(f, true)) { | |
9f95a23c | 2481 | ss << "op_tracker disabled; set mds_enable_op_tracker=true to enable"; |
7c673cae FG |
2482 | } |
2483 | } else if (command == "dump_historic_ops") { | |
2484 | if (!op_tracker.dump_historic_ops(f)) { | |
9f95a23c | 2485 | ss << "op_tracker disabled; set mds_enable_op_tracker=true to enable"; |
7c673cae FG |
2486 | } |
2487 | } else if (command == "dump_historic_ops_by_duration") { | |
2488 | if (!op_tracker.dump_historic_ops(f, true)) { | |
9f95a23c | 2489 | ss << "op_tracker disabled; set mds_enable_op_tracker=true to enable"; |
7c673cae FG |
2490 | } |
2491 | } else if (command == "osdmap barrier") { | |
2492 | int64_t target_epoch = 0; | |
9f95a23c | 2493 | bool got_val = cmd_getval(cmdmap, "target_epoch", target_epoch); |
7c673cae FG |
2494 | |
2495 | if (!got_val) { | |
2496 | ss << "no target epoch given"; | |
9f95a23c TL |
2497 | r = -EINVAL; |
2498 | goto out; | |
2499 | } | |
2500 | { | |
2501 | std::lock_guard l(mds_lock); | |
2502 | set_osd_epoch_barrier(target_epoch); | |
7c673cae | 2503 | } |
7c673cae FG |
2504 | C_SaferCond cond; |
2505 | bool already_got = objecter->wait_for_map(target_epoch, &cond); | |
2506 | if (!already_got) { | |
2507 | dout(4) << __func__ << ": waiting for OSD epoch " << target_epoch << dendl; | |
2508 | cond.wait(); | |
2509 | } | |
9f95a23c TL |
2510 | } else if (command == "session ls" || |
2511 | command == "client ls") { | |
11fdf7f2 | 2512 | std::lock_guard l(mds_lock); |
9f95a23c TL |
2513 | std::vector<std::string> filter_args; |
2514 | cmd_getval(cmdmap, "filters", filter_args); | |
2515 | SessionFilter filter; | |
2516 | r = filter.parse(filter_args, &ss); | |
2517 | if (r != 0) { | |
2518 | goto out; | |
2519 | } | |
2520 | dump_sessions(filter, f); | |
2521 | } else if (command == "session evict" || | |
2522 | command == "client evict") { | |
2523 | std::lock_guard l(mds_lock); | |
2524 | std::vector<std::string> filter_args; | |
2525 | cmd_getval(cmdmap, "filters", filter_args); | |
7c673cae | 2526 | |
9f95a23c TL |
2527 | SessionFilter filter; |
2528 | r = filter.parse(filter_args, &ss); | |
2529 | if (r != 0) { | |
2530 | r = -EINVAL; | |
2531 | goto out; | |
2532 | } | |
2533 | evict_clients(filter, on_finish); | |
2534 | return; | |
2535 | } else if (command == "session kill") { | |
7c673cae | 2536 | std::string client_id; |
9f95a23c | 2537 | if (!cmd_getval(cmdmap, "client_id", client_id)) { |
7c673cae | 2538 | ss << "Invalid client_id specified"; |
9f95a23c TL |
2539 | r = -ENOENT; |
2540 | goto out; | |
7c673cae | 2541 | } |
9f95a23c | 2542 | std::lock_guard l(mds_lock); |
31f18b77 | 2543 | bool evicted = evict_client(strtol(client_id.c_str(), 0, 10), true, |
9f95a23c | 2544 | g_conf()->mds_session_blacklist_on_evict, ss); |
31f18b77 | 2545 | if (!evicted) { |
9f95a23c TL |
2546 | dout(15) << ss.str() << dendl; |
2547 | r = -ENOENT; | |
7c673cae | 2548 | } |
9f95a23c TL |
2549 | } else if (command == "session config" || |
2550 | command == "client config") { | |
92f5a8d4 TL |
2551 | int64_t client_id; |
2552 | std::string option; | |
2553 | std::string value; | |
2554 | ||
9f95a23c TL |
2555 | cmd_getval(cmdmap, "client_id", client_id); |
2556 | cmd_getval(cmdmap, "option", option); | |
2557 | bool got_value = cmd_getval(cmdmap, "value", value); | |
92f5a8d4 | 2558 | |
9f95a23c TL |
2559 | std::lock_guard l(mds_lock); |
2560 | r = config_client(client_id, !got_value, option, value, ss); | |
2561 | } else if (command == "scrub start" || | |
2562 | command == "scrub_start") { | |
7c673cae | 2563 | string path; |
9f95a23c | 2564 | string tag; |
7c673cae | 2565 | vector<string> scrubop_vec; |
9f95a23c TL |
2566 | cmd_getval(cmdmap, "scrubops", scrubop_vec); |
2567 | cmd_getval(cmdmap, "path", path); | |
2568 | cmd_getval(cmdmap, "tag", tag); | |
11fdf7f2 | 2569 | |
92f5a8d4 TL |
2570 | /* Multiple MDS scrub is not currently supported. See also: https://tracker.ceph.com/issues/12274 */ |
2571 | if (mdsmap->get_max_mds() > 1) { | |
2572 | ss << "Scrub is not currently supported for multiple active MDS. Please reduce max_mds to 1 and then scrub."; | |
9f95a23c TL |
2573 | r = -EINVAL; |
2574 | goto out; | |
2575 | } | |
2576 | ||
2577 | finisher->queue( | |
2578 | new LambdaContext( | |
2579 | [this, on_finish, f, path, tag, scrubop_vec](int r) { | |
2580 | command_scrub_start( | |
2581 | f, path, tag, scrubop_vec, | |
2582 | new LambdaContext( | |
2583 | [on_finish](int r) { | |
2584 | bufferlist outbl; | |
2585 | on_finish(r, {}, outbl); | |
2586 | })); | |
2587 | })); | |
2588 | return; | |
2589 | } else if (command == "scrub abort") { | |
2590 | finisher->queue( | |
2591 | new LambdaContext( | |
2592 | [this, on_finish, f](int r) { | |
2593 | command_scrub_abort( | |
2594 | f, | |
2595 | new LambdaContext( | |
2596 | [on_finish, f](int r) { | |
2597 | bufferlist outbl; | |
2598 | f->open_object_section("result"); | |
2599 | f->dump_int("return_code", r); | |
2600 | f->close_section(); | |
2601 | on_finish(r, {}, outbl); | |
2602 | })); | |
2603 | })); | |
2604 | return; | |
2605 | } else if (command == "scrub pause") { | |
2606 | finisher->queue( | |
2607 | new LambdaContext( | |
2608 | [this, on_finish, f](int r) { | |
2609 | command_scrub_pause( | |
2610 | f, | |
2611 | new LambdaContext( | |
2612 | [on_finish, f](int r) { | |
2613 | bufferlist outbl; | |
2614 | f->open_object_section("result"); | |
2615 | f->dump_int("return_code", r); | |
2616 | f->close_section(); | |
2617 | on_finish(r, {}, outbl); | |
2618 | })); | |
2619 | })); | |
2620 | return; | |
2621 | } else if (command == "scrub resume") { | |
2622 | command_scrub_resume(f); | |
2623 | } else if (command == "scrub status") { | |
2624 | command_scrub_status(f); | |
7c673cae FG |
2625 | } else if (command == "tag path") { |
2626 | string path; | |
9f95a23c | 2627 | cmd_getval(cmdmap, "path", path); |
7c673cae | 2628 | string tag; |
9f95a23c | 2629 | cmd_getval(cmdmap, "tag", tag); |
7c673cae FG |
2630 | command_tag_path(f, path, tag); |
2631 | } else if (command == "flush_path") { | |
2632 | string path; | |
9f95a23c | 2633 | cmd_getval(cmdmap, "path", path); |
7c673cae FG |
2634 | command_flush_path(f, path); |
2635 | } else if (command == "flush journal") { | |
2636 | command_flush_journal(f); | |
2637 | } else if (command == "get subtrees") { | |
2638 | command_get_subtrees(f); | |
2639 | } else if (command == "export dir") { | |
2640 | string path; | |
9f95a23c | 2641 | if(!cmd_getval(cmdmap, "path", path)) { |
7c673cae | 2642 | ss << "malformed path"; |
9f95a23c TL |
2643 | r = -EINVAL; |
2644 | goto out; | |
7c673cae FG |
2645 | } |
2646 | int64_t rank; | |
9f95a23c | 2647 | if(!cmd_getval(cmdmap, "rank", rank)) { |
7c673cae | 2648 | ss << "malformed rank"; |
9f95a23c TL |
2649 | r = -EINVAL; |
2650 | goto out; | |
7c673cae FG |
2651 | } |
2652 | command_export_dir(f, path, (mds_rank_t)rank); | |
2653 | } else if (command == "dump cache") { | |
11fdf7f2 | 2654 | std::lock_guard l(mds_lock); |
7c673cae | 2655 | string path; |
9f95a23c | 2656 | if (!cmd_getval(cmdmap, "path", path)) { |
31f18b77 | 2657 | r = mdcache->dump_cache(f); |
7c673cae | 2658 | } else { |
31f18b77 FG |
2659 | r = mdcache->dump_cache(path); |
2660 | } | |
9f95a23c TL |
2661 | } else if (command == "cache drop") { |
2662 | int64_t timeout = 0; | |
2663 | cmd_getval(cmdmap, "timeout", timeout); | |
2664 | finisher->queue( | |
2665 | new LambdaContext( | |
2666 | [this, on_finish, f, timeout](int r) { | |
2667 | command_cache_drop( | |
2668 | timeout, f, | |
2669 | new LambdaContext( | |
2670 | [on_finish](int r) { | |
2671 | bufferlist outbl; | |
2672 | on_finish(r, {}, outbl); | |
2673 | })); | |
2674 | })); | |
2675 | return; | |
181888fb | 2676 | } else if (command == "cache status") { |
11fdf7f2 | 2677 | std::lock_guard l(mds_lock); |
f64942e4 | 2678 | mdcache->cache_status(f); |
7c673cae | 2679 | } else if (command == "dump tree") { |
11fdf7f2 | 2680 | command_dump_tree(cmdmap, ss, f); |
28e407b8 | 2681 | } else if (command == "dump loads") { |
11fdf7f2 | 2682 | std::lock_guard l(mds_lock); |
9f95a23c | 2683 | r = balancer->dump_loads(f); |
11fdf7f2 TL |
2684 | } else if (command == "dump snaps") { |
2685 | std::lock_guard l(mds_lock); | |
2686 | string server; | |
9f95a23c | 2687 | cmd_getval(cmdmap, "server", server); |
11fdf7f2 TL |
2688 | if (server == "--server") { |
2689 | if (mdsmap->get_tableserver() == whoami) { | |
2690 | snapserver->dump(f); | |
2691 | } else { | |
9f95a23c | 2692 | r = -EXDEV; |
11fdf7f2 TL |
2693 | ss << "Not snapserver"; |
2694 | } | |
2695 | } else { | |
9f95a23c | 2696 | r = snapclient->dump_cache(f); |
11fdf7f2 | 2697 | } |
7c673cae | 2698 | } else if (command == "force_readonly") { |
11fdf7f2 | 2699 | std::lock_guard l(mds_lock); |
7c673cae FG |
2700 | mdcache->force_readonly(); |
2701 | } else if (command == "dirfrag split") { | |
2702 | command_dirfrag_split(cmdmap, ss); | |
2703 | } else if (command == "dirfrag merge") { | |
2704 | command_dirfrag_merge(cmdmap, ss); | |
2705 | } else if (command == "dirfrag ls") { | |
2706 | command_dirfrag_ls(cmdmap, ss, f); | |
11fdf7f2 TL |
2707 | } else if (command == "openfiles ls") { |
2708 | command_openfiles_ls(f); | |
2709 | } else if (command == "dump inode") { | |
2710 | command_dump_inode(f, cmdmap, ss); | |
9f95a23c TL |
2711 | } else if (command == "damage ls") { |
2712 | std::lock_guard l(mds_lock); | |
2713 | damage_table.dump(f); | |
2714 | } else if (command == "damage rm") { | |
2715 | std::lock_guard l(mds_lock); | |
2716 | damage_entry_id_t id = 0; | |
2717 | if (!cmd_getval(cmdmap, "damage_id", (int64_t&)id)) { | |
2718 | r = -EINVAL; | |
2719 | goto out; | |
2720 | } | |
2721 | damage_table.erase(id); | |
7c673cae | 2722 | } else { |
9f95a23c | 2723 | r = -ENOSYS; |
7c673cae | 2724 | } |
9f95a23c TL |
2725 | out: |
2726 | on_finish(r, ss.str(), outbl); | |
7c673cae FG |
2727 | } |
2728 | ||
7c673cae FG |
2729 | /** |
2730 | * This function drops the mds_lock, so don't do anything with | |
2731 | * MDSRank after calling it (we could have gone into shutdown): just | |
2732 | * send your result back to the calling client and finish. | |
2733 | */ | |
9f95a23c TL |
2734 | void MDSRankDispatcher::evict_clients( |
2735 | const SessionFilter &filter, | |
2736 | std::function<void(int,const std::string&,bufferlist&)> on_finish) | |
7c673cae | 2737 | { |
9f95a23c | 2738 | bufferlist outbl; |
7c673cae | 2739 | if (is_any_replay()) { |
9f95a23c | 2740 | on_finish(-EAGAIN, "MDS is replaying log", outbl); |
7c673cae FG |
2741 | return; |
2742 | } | |
2743 | ||
11fdf7f2 TL |
2744 | std::vector<Session*> victims; |
2745 | const auto& sessions = sessionmap.get_sessions(); | |
2746 | for (const auto& p : sessions) { | |
7c673cae FG |
2747 | if (!p.first.is_client()) { |
2748 | continue; | |
2749 | } | |
2750 | ||
2751 | Session *s = p.second; | |
2752 | ||
9f95a23c TL |
2753 | if (filter.match(*s, std::bind(&Server::waiting_for_reconnect, server, |
2754 | std::placeholders::_1))) { | |
7c673cae FG |
2755 | victims.push_back(s); |
2756 | } | |
2757 | } | |
2758 | ||
2759 | dout(20) << __func__ << " matched " << victims.size() << " sessions" << dendl; | |
2760 | ||
2761 | if (victims.empty()) { | |
9f95a23c | 2762 | on_finish(0, {}, outbl); |
7c673cae FG |
2763 | return; |
2764 | } | |
2765 | ||
9f95a23c TL |
2766 | C_GatherBuilder gather(g_ceph_context, |
2767 | new LambdaContext([on_finish](int r) { | |
2768 | bufferlist bl; | |
2769 | on_finish(r, {}, bl); | |
2770 | })); | |
7c673cae | 2771 | for (const auto s : victims) { |
31f18b77 | 2772 | std::stringstream ss; |
11fdf7f2 TL |
2773 | evict_client(s->get_client().v, false, |
2774 | g_conf()->mds_session_blacklist_on_evict, ss, gather.new_sub()); | |
7c673cae FG |
2775 | } |
2776 | gather.activate(); | |
2777 | } | |
2778 | ||
2779 | void MDSRankDispatcher::dump_sessions(const SessionFilter &filter, Formatter *f) const | |
2780 | { | |
2781 | // Dump sessions, decorated with recovery/replay status | |
2782 | f->open_array_section("sessions"); | |
92f5a8d4 TL |
2783 | for (auto& [name, s] : sessionmap.get_sessions()) { |
2784 | if (!name.is_client()) { | |
7c673cae FG |
2785 | continue; |
2786 | } | |
2787 | ||
7c673cae FG |
2788 | if (!filter.match(*s, std::bind(&Server::waiting_for_reconnect, server, std::placeholders::_1))) { |
2789 | continue; | |
2790 | } | |
2791 | ||
92f5a8d4 | 2792 | f->dump_object("session", *s); |
7c673cae | 2793 | } |
92f5a8d4 | 2794 | f->close_section(); // sessions |
7c673cae FG |
2795 | } |
2796 | ||
11fdf7f2 TL |
2797 | void MDSRank::command_scrub_start(Formatter *f, |
2798 | std::string_view path, std::string_view tag, | |
2799 | const vector<string>& scrubop_vec, Context *on_finish) | |
7c673cae FG |
2800 | { |
2801 | bool force = false; | |
2802 | bool recursive = false; | |
2803 | bool repair = false; | |
11fdf7f2 TL |
2804 | for (auto &op : scrubop_vec) { |
2805 | if (op == "force") | |
7c673cae | 2806 | force = true; |
11fdf7f2 | 2807 | else if (op == "recursive") |
7c673cae | 2808 | recursive = true; |
11fdf7f2 | 2809 | else if (op == "repair") |
7c673cae FG |
2810 | repair = true; |
2811 | } | |
11fdf7f2 TL |
2812 | |
2813 | std::lock_guard l(mds_lock); | |
2814 | mdcache->enqueue_scrub(path, tag, force, recursive, repair, f, on_finish); | |
7c673cae FG |
2815 | // scrub_dentry() finishers will dump the data for us; we're done! |
2816 | } | |
2817 | ||
2818 | void MDSRank::command_tag_path(Formatter *f, | |
11fdf7f2 | 2819 | std::string_view path, std::string_view tag) |
7c673cae FG |
2820 | { |
2821 | C_SaferCond scond; | |
2822 | { | |
11fdf7f2 | 2823 | std::lock_guard l(mds_lock); |
7c673cae FG |
2824 | mdcache->enqueue_scrub(path, tag, true, true, false, f, &scond); |
2825 | } | |
2826 | scond.wait(); | |
2827 | } | |
2828 | ||
11fdf7f2 TL |
2829 | void MDSRank::command_scrub_abort(Formatter *f, Context *on_finish) { |
2830 | std::lock_guard l(mds_lock); | |
2831 | scrubstack->scrub_abort(on_finish); | |
2832 | } | |
2833 | ||
2834 | void MDSRank::command_scrub_pause(Formatter *f, Context *on_finish) { | |
2835 | std::lock_guard l(mds_lock); | |
2836 | scrubstack->scrub_pause(on_finish); | |
2837 | } | |
2838 | ||
2839 | void MDSRank::command_scrub_resume(Formatter *f) { | |
9f95a23c | 2840 | std::lock_guard l(mds_lock); |
11fdf7f2 TL |
2841 | int r = scrubstack->scrub_resume(); |
2842 | ||
2843 | f->open_object_section("result"); | |
2844 | f->dump_int("return_code", r); | |
2845 | f->close_section(); | |
2846 | } | |
2847 | ||
2848 | void MDSRank::command_scrub_status(Formatter *f) { | |
9f95a23c | 2849 | std::lock_guard l(mds_lock); |
11fdf7f2 TL |
2850 | scrubstack->scrub_status(f); |
2851 | } | |
2852 | ||
2853 | void MDSRank::command_flush_path(Formatter *f, std::string_view path) | |
7c673cae FG |
2854 | { |
2855 | C_SaferCond scond; | |
2856 | { | |
11fdf7f2 | 2857 | std::lock_guard l(mds_lock); |
7c673cae FG |
2858 | mdcache->flush_dentry(path, &scond); |
2859 | } | |
2860 | int r = scond.wait(); | |
2861 | f->open_object_section("results"); | |
2862 | f->dump_int("return_code", r); | |
2863 | f->close_section(); // results | |
2864 | } | |
2865 | ||
f64942e4 AA |
2866 | // synchronous wrapper around "journal flush" asynchronous context |
2867 | // execution. | |
2868 | void MDSRank::command_flush_journal(Formatter *f) { | |
2869 | ceph_assert(f != NULL); | |
7c673cae | 2870 | |
f64942e4 | 2871 | C_SaferCond cond; |
7c673cae | 2872 | std::stringstream ss; |
7c673cae | 2873 | { |
11fdf7f2 | 2874 | std::lock_guard locker(mds_lock); |
f64942e4 AA |
2875 | C_Flush_Journal *flush_journal = new C_Flush_Journal(mdcache, mdlog, this, &ss, &cond); |
2876 | flush_journal->send(); | |
7c673cae | 2877 | } |
f64942e4 | 2878 | int r = cond.wait(); |
7c673cae | 2879 | |
f64942e4 AA |
2880 | f->open_object_section("result"); |
2881 | f->dump_string("message", ss.str()); | |
2882 | f->dump_int("return_code", r); | |
2883 | f->close_section(); | |
7c673cae FG |
2884 | } |
2885 | ||
7c673cae FG |
2886 | void MDSRank::command_get_subtrees(Formatter *f) |
2887 | { | |
11fdf7f2 TL |
2888 | ceph_assert(f != NULL); |
2889 | std::lock_guard l(mds_lock); | |
7c673cae | 2890 | |
11fdf7f2 TL |
2891 | std::vector<CDir*> subtrees; |
2892 | mdcache->get_subtrees(subtrees); | |
7c673cae FG |
2893 | |
2894 | f->open_array_section("subtrees"); | |
11fdf7f2 | 2895 | for (const auto& dir : subtrees) { |
7c673cae FG |
2896 | f->open_object_section("subtree"); |
2897 | { | |
2898 | f->dump_bool("is_auth", dir->is_auth()); | |
2899 | f->dump_int("auth_first", dir->get_dir_auth().first); | |
2900 | f->dump_int("auth_second", dir->get_dir_auth().second); | |
f6b5b4d7 TL |
2901 | f->dump_int("export_pin", dir->inode->get_export_pin(false, false)); |
2902 | f->dump_bool("distributed_ephemeral_pin", dir->inode->is_ephemeral_dist()); | |
2903 | f->dump_bool("random_ephemeral_pin", dir->inode->is_ephemeral_rand()); | |
2904 | f->dump_int("ephemeral_pin", mdcache->hash_into_rank_bucket(dir->inode->ino())); | |
7c673cae FG |
2905 | f->open_object_section("dir"); |
2906 | dir->dump(f); | |
2907 | f->close_section(); | |
2908 | } | |
2909 | f->close_section(); | |
2910 | } | |
2911 | f->close_section(); | |
2912 | } | |
2913 | ||
2914 | ||
2915 | void MDSRank::command_export_dir(Formatter *f, | |
11fdf7f2 | 2916 | std::string_view path, |
7c673cae FG |
2917 | mds_rank_t target) |
2918 | { | |
2919 | int r = _command_export_dir(path, target); | |
2920 | f->open_object_section("results"); | |
2921 | f->dump_int("return_code", r); | |
2922 | f->close_section(); // results | |
2923 | } | |
2924 | ||
2925 | int MDSRank::_command_export_dir( | |
11fdf7f2 | 2926 | std::string_view path, |
7c673cae FG |
2927 | mds_rank_t target) |
2928 | { | |
11fdf7f2 | 2929 | std::lock_guard l(mds_lock); |
94b18763 | 2930 | filepath fp(path); |
7c673cae FG |
2931 | |
2932 | if (target == whoami || !mdsmap->is_up(target) || !mdsmap->is_in(target)) { | |
2933 | derr << "bad MDS target " << target << dendl; | |
2934 | return -ENOENT; | |
2935 | } | |
2936 | ||
2937 | CInode *in = mdcache->cache_traverse(fp); | |
2938 | if (!in) { | |
2939 | derr << "Bath path '" << path << "'" << dendl; | |
2940 | return -ENOENT; | |
2941 | } | |
2942 | CDir *dir = in->get_dirfrag(frag_t()); | |
2943 | if (!dir || !(dir->is_auth())) { | |
2944 | derr << "bad export_dir path dirfrag frag_t() or dir not auth" << dendl; | |
2945 | return -EINVAL; | |
2946 | } | |
2947 | ||
2948 | mdcache->migrator->export_dir(dir, target); | |
2949 | return 0; | |
2950 | } | |
2951 | ||
11fdf7f2 TL |
2952 | void MDSRank::command_dump_tree(const cmdmap_t &cmdmap, std::ostream &ss, Formatter *f) |
2953 | { | |
2954 | std::string root; | |
2955 | int64_t depth; | |
9f95a23c TL |
2956 | cmd_getval(cmdmap, "root", root); |
2957 | if (!cmd_getval(cmdmap, "depth", depth)) | |
11fdf7f2 TL |
2958 | depth = -1; |
2959 | std::lock_guard l(mds_lock); | |
2960 | CInode *in = mdcache->cache_traverse(filepath(root.c_str())); | |
2961 | if (!in) { | |
2962 | ss << "root inode is not in cache"; | |
2963 | return; | |
2964 | } | |
2965 | f->open_array_section("inodes"); | |
2966 | mdcache->dump_tree(in, 0, depth, f); | |
2967 | f->close_section(); | |
2968 | } | |
2969 | ||
7c673cae FG |
2970 | CDir *MDSRank::_command_dirfrag_get( |
2971 | const cmdmap_t &cmdmap, | |
2972 | std::ostream &ss) | |
2973 | { | |
2974 | std::string path; | |
9f95a23c | 2975 | bool got = cmd_getval(cmdmap, "path", path); |
7c673cae FG |
2976 | if (!got) { |
2977 | ss << "missing path argument"; | |
2978 | return NULL; | |
2979 | } | |
2980 | ||
2981 | std::string frag_str; | |
9f95a23c | 2982 | if (!cmd_getval(cmdmap, "frag", frag_str)) { |
7c673cae FG |
2983 | ss << "missing frag argument"; |
2984 | return NULL; | |
2985 | } | |
2986 | ||
2987 | CInode *in = mdcache->cache_traverse(filepath(path.c_str())); | |
2988 | if (!in) { | |
2989 | // TODO really we should load something in if it's not in cache, | |
2990 | // but the infrastructure is harder, and we might still be unable | |
2991 | // to act on it if someone else is auth. | |
2992 | ss << "directory '" << path << "' inode not in cache"; | |
2993 | return NULL; | |
2994 | } | |
2995 | ||
2996 | frag_t fg; | |
2997 | ||
2998 | if (!fg.parse(frag_str.c_str())) { | |
2999 | ss << "frag " << frag_str << " failed to parse"; | |
3000 | return NULL; | |
3001 | } | |
3002 | ||
3003 | CDir *dir = in->get_dirfrag(fg); | |
3004 | if (!dir) { | |
11fdf7f2 | 3005 | ss << "frag " << in->ino() << "/" << fg << " not in cache (" |
7c673cae FG |
3006 | "use `dirfrag ls` to see if it should exist)"; |
3007 | return NULL; | |
3008 | } | |
3009 | ||
3010 | if (!dir->is_auth()) { | |
3011 | ss << "frag " << dir->dirfrag() << " not auth (auth = " | |
3012 | << dir->authority() << ")"; | |
3013 | return NULL; | |
3014 | } | |
3015 | ||
3016 | return dir; | |
3017 | } | |
3018 | ||
3019 | bool MDSRank::command_dirfrag_split( | |
3020 | cmdmap_t cmdmap, | |
3021 | std::ostream &ss) | |
3022 | { | |
11fdf7f2 | 3023 | std::lock_guard l(mds_lock); |
7c673cae | 3024 | int64_t by = 0; |
9f95a23c | 3025 | if (!cmd_getval(cmdmap, "bits", by)) { |
7c673cae FG |
3026 | ss << "missing bits argument"; |
3027 | return false; | |
3028 | } | |
3029 | ||
3030 | if (by <= 0) { | |
3031 | ss << "must split by >0 bits"; | |
3032 | return false; | |
3033 | } | |
3034 | ||
3035 | CDir *dir = _command_dirfrag_get(cmdmap, ss); | |
3036 | if (!dir) { | |
3037 | return false; | |
3038 | } | |
3039 | ||
3040 | mdcache->split_dir(dir, by); | |
3041 | ||
3042 | return true; | |
3043 | } | |
3044 | ||
3045 | bool MDSRank::command_dirfrag_merge( | |
3046 | cmdmap_t cmdmap, | |
3047 | std::ostream &ss) | |
3048 | { | |
11fdf7f2 | 3049 | std::lock_guard l(mds_lock); |
7c673cae | 3050 | std::string path; |
9f95a23c | 3051 | bool got = cmd_getval(cmdmap, "path", path); |
7c673cae FG |
3052 | if (!got) { |
3053 | ss << "missing path argument"; | |
3054 | return false; | |
3055 | } | |
3056 | ||
3057 | std::string frag_str; | |
9f95a23c | 3058 | if (!cmd_getval(cmdmap, "frag", frag_str)) { |
7c673cae FG |
3059 | ss << "missing frag argument"; |
3060 | return false; | |
3061 | } | |
3062 | ||
3063 | CInode *in = mdcache->cache_traverse(filepath(path.c_str())); | |
3064 | if (!in) { | |
3065 | ss << "directory '" << path << "' inode not in cache"; | |
3066 | return false; | |
3067 | } | |
3068 | ||
3069 | frag_t fg; | |
3070 | if (!fg.parse(frag_str.c_str())) { | |
3071 | ss << "frag " << frag_str << " failed to parse"; | |
3072 | return false; | |
3073 | } | |
3074 | ||
3075 | mdcache->merge_dir(in, fg); | |
3076 | ||
3077 | return true; | |
3078 | } | |
3079 | ||
3080 | bool MDSRank::command_dirfrag_ls( | |
3081 | cmdmap_t cmdmap, | |
3082 | std::ostream &ss, | |
3083 | Formatter *f) | |
3084 | { | |
11fdf7f2 | 3085 | std::lock_guard l(mds_lock); |
7c673cae | 3086 | std::string path; |
9f95a23c | 3087 | bool got = cmd_getval(cmdmap, "path", path); |
7c673cae FG |
3088 | if (!got) { |
3089 | ss << "missing path argument"; | |
3090 | return false; | |
3091 | } | |
3092 | ||
3093 | CInode *in = mdcache->cache_traverse(filepath(path.c_str())); | |
3094 | if (!in) { | |
3095 | ss << "directory inode not in cache"; | |
3096 | return false; | |
3097 | } | |
3098 | ||
3099 | f->open_array_section("frags"); | |
11fdf7f2 | 3100 | frag_vec_t leaves; |
7c673cae FG |
3101 | // NB using get_leaves_under instead of get_dirfrags to give |
3102 | // you the list of what dirfrags may exist, not which are in cache | |
11fdf7f2 TL |
3103 | in->dirfragtree.get_leaves_under(frag_t(), leaves); |
3104 | for (const auto& leaf : leaves) { | |
7c673cae | 3105 | f->open_object_section("frag"); |
11fdf7f2 TL |
3106 | f->dump_int("value", leaf.value()); |
3107 | f->dump_int("bits", leaf.bits()); | |
3108 | CachedStackStringStream css; | |
3109 | *css << std::hex << leaf.value() << "/" << std::dec << leaf.bits(); | |
3110 | f->dump_string("str", css->strv()); | |
7c673cae FG |
3111 | f->close_section(); |
3112 | } | |
3113 | f->close_section(); | |
3114 | ||
3115 | return true; | |
3116 | } | |
3117 | ||
11fdf7f2 TL |
3118 | void MDSRank::command_openfiles_ls(Formatter *f) |
3119 | { | |
3120 | std::lock_guard l(mds_lock); | |
3121 | mdcache->dump_openfiles(f); | |
3122 | } | |
3123 | ||
3124 | void MDSRank::command_dump_inode(Formatter *f, const cmdmap_t &cmdmap, std::ostream &ss) | |
3125 | { | |
3126 | std::lock_guard l(mds_lock); | |
3127 | int64_t number; | |
9f95a23c | 3128 | bool got = cmd_getval(cmdmap, "number", number); |
11fdf7f2 TL |
3129 | if (!got) { |
3130 | ss << "missing inode number"; | |
3131 | return; | |
3132 | } | |
3133 | ||
3134 | bool success = mdcache->dump_inode(f, number); | |
3135 | if (!success) { | |
3136 | ss << "dump inode failed, wrong inode number or the inode is not cached"; | |
3137 | } | |
3138 | } | |
3139 | ||
7c673cae FG |
3140 | void MDSRank::dump_status(Formatter *f) const |
3141 | { | |
3142 | if (state == MDSMap::STATE_REPLAY || | |
3143 | state == MDSMap::STATE_STANDBY_REPLAY) { | |
3144 | mdlog->dump_replay_status(f); | |
3145 | } else if (state == MDSMap::STATE_RESOLVE) { | |
3146 | mdcache->dump_resolve_status(f); | |
3147 | } else if (state == MDSMap::STATE_RECONNECT) { | |
3148 | server->dump_reconnect_status(f); | |
3149 | } else if (state == MDSMap::STATE_REJOIN) { | |
3150 | mdcache->dump_rejoin_status(f); | |
3151 | } else if (state == MDSMap::STATE_CLIENTREPLAY) { | |
3152 | dump_clientreplay_status(f); | |
3153 | } | |
94b18763 | 3154 | f->dump_float("rank_uptime", get_uptime().count()); |
7c673cae FG |
3155 | } |
3156 | ||
3157 | void MDSRank::dump_clientreplay_status(Formatter *f) const | |
3158 | { | |
3159 | f->open_object_section("clientreplay_status"); | |
3160 | f->dump_unsigned("clientreplay_queue", replay_queue.size()); | |
3161 | f->dump_unsigned("active_replay", mdcache->get_num_client_requests()); | |
3162 | f->close_section(); | |
3163 | } | |
3164 | ||
3165 | void MDSRankDispatcher::update_log_config() | |
3166 | { | |
3167 | map<string,string> log_to_monitors; | |
3168 | map<string,string> log_to_syslog; | |
3169 | map<string,string> log_channel; | |
3170 | map<string,string> log_prio; | |
3171 | map<string,string> log_to_graylog; | |
3172 | map<string,string> log_to_graylog_host; | |
3173 | map<string,string> log_to_graylog_port; | |
3174 | uuid_d fsid; | |
3175 | string host; | |
3176 | ||
3177 | if (parse_log_client_options(g_ceph_context, log_to_monitors, log_to_syslog, | |
3178 | log_channel, log_prio, log_to_graylog, | |
3179 | log_to_graylog_host, log_to_graylog_port, | |
3180 | fsid, host) == 0) | |
3181 | clog->update_config(log_to_monitors, log_to_syslog, | |
3182 | log_channel, log_prio, log_to_graylog, | |
3183 | log_to_graylog_host, log_to_graylog_port, | |
3184 | fsid, host); | |
3185 | dout(10) << __func__ << " log_to_monitors " << log_to_monitors << dendl; | |
3186 | } | |
3187 | ||
3188 | void MDSRank::create_logger() | |
3189 | { | |
3190 | dout(10) << "create_logger" << dendl; | |
3191 | { | |
3192 | PerfCountersBuilder mds_plb(g_ceph_context, "mds", l_mds_first, l_mds_last); | |
3193 | ||
91327a77 AA |
3194 | // super useful (high prio) perf stats |
3195 | mds_plb.add_u64_counter(l_mds_request, "request", "Requests", "req", | |
3196 | PerfCountersBuilder::PRIO_CRITICAL); | |
3197 | mds_plb.add_time_avg(l_mds_reply_latency, "reply_latency", "Reply latency", "rlat", | |
3198 | PerfCountersBuilder::PRIO_CRITICAL); | |
3199 | mds_plb.add_u64(l_mds_inodes, "inodes", "Inodes", "inos", | |
3200 | PerfCountersBuilder::PRIO_CRITICAL); | |
3201 | mds_plb.add_u64_counter(l_mds_forward, "forward", "Forwarding request", "fwd", | |
3202 | PerfCountersBuilder::PRIO_INTERESTING); | |
3203 | mds_plb.add_u64(l_mds_caps, "caps", "Capabilities", "caps", | |
3204 | PerfCountersBuilder::PRIO_INTERESTING); | |
3205 | mds_plb.add_u64_counter(l_mds_exported_inodes, "exported_inodes", "Exported inodes", | |
3206 | "exi", PerfCountersBuilder::PRIO_INTERESTING); | |
3207 | mds_plb.add_u64_counter(l_mds_imported_inodes, "imported_inodes", "Imported inodes", | |
3208 | "imi", PerfCountersBuilder::PRIO_INTERESTING); | |
3209 | ||
3210 | // useful dir/inode/subtree stats | |
3211 | mds_plb.set_prio_default(PerfCountersBuilder::PRIO_USEFUL); | |
eafe8130 TL |
3212 | mds_plb.add_u64(l_mds_root_rfiles, "root_rfiles", "root inode rfiles"); |
3213 | mds_plb.add_u64(l_mds_root_rbytes, "root_rbytes", "root inode rbytes"); | |
3214 | mds_plb.add_u64(l_mds_root_rsnaps, "root_rsnaps", "root inode rsnaps"); | |
7c673cae FG |
3215 | mds_plb.add_u64_counter(l_mds_dir_fetch, "dir_fetch", "Directory fetch"); |
3216 | mds_plb.add_u64_counter(l_mds_dir_commit, "dir_commit", "Directory commit"); | |
3217 | mds_plb.add_u64_counter(l_mds_dir_split, "dir_split", "Directory split"); | |
3218 | mds_plb.add_u64_counter(l_mds_dir_merge, "dir_merge", "Directory merge"); | |
7c673cae FG |
3219 | mds_plb.add_u64(l_mds_inodes_pinned, "inodes_pinned", "Inodes pinned"); |
3220 | mds_plb.add_u64(l_mds_inodes_expired, "inodes_expired", "Inodes expired"); | |
91327a77 AA |
3221 | mds_plb.add_u64(l_mds_inodes_with_caps, "inodes_with_caps", |
3222 | "Inodes with capabilities"); | |
7c673cae | 3223 | mds_plb.add_u64(l_mds_subtrees, "subtrees", "Subtrees"); |
91327a77 | 3224 | mds_plb.add_u64(l_mds_load_cent, "load_cent", "Load per cent"); |
11fdf7f2 TL |
3225 | mds_plb.add_u64_counter(l_mds_openino_dir_fetch, "openino_dir_fetch", |
3226 | "OpenIno incomplete directory fetchings"); | |
7c673cae | 3227 | |
91327a77 AA |
3228 | // low prio stats |
3229 | mds_plb.set_prio_default(PerfCountersBuilder::PRIO_DEBUGONLY); | |
3230 | mds_plb.add_u64_counter(l_mds_reply, "reply", "Replies"); | |
3231 | mds_plb.add_u64(l_mds_inodes_top, "inodes_top", "Inodes on top"); | |
3232 | mds_plb.add_u64(l_mds_inodes_bottom, "inodes_bottom", "Inodes on bottom"); | |
3233 | mds_plb.add_u64( | |
3234 | l_mds_inodes_pin_tail, "inodes_pin_tail", "Inodes on pin tail"); | |
7c673cae FG |
3235 | mds_plb.add_u64_counter(l_mds_traverse, "traverse", "Traverses"); |
3236 | mds_plb.add_u64_counter(l_mds_traverse_hit, "traverse_hit", "Traverse hits"); | |
3237 | mds_plb.add_u64_counter(l_mds_traverse_forward, "traverse_forward", | |
91327a77 | 3238 | "Traverse forwards"); |
7c673cae | 3239 | mds_plb.add_u64_counter(l_mds_traverse_discover, "traverse_discover", |
91327a77 | 3240 | "Traverse directory discovers"); |
7c673cae | 3241 | mds_plb.add_u64_counter(l_mds_traverse_dir_fetch, "traverse_dir_fetch", |
91327a77 | 3242 | "Traverse incomplete directory content fetchings"); |
7c673cae | 3243 | mds_plb.add_u64_counter(l_mds_traverse_remote_ino, "traverse_remote_ino", |
91327a77 | 3244 | "Traverse remote dentries"); |
7c673cae | 3245 | mds_plb.add_u64_counter(l_mds_traverse_lock, "traverse_lock", |
91327a77 | 3246 | "Traverse locks"); |
7c673cae | 3247 | mds_plb.add_u64(l_mds_dispatch_queue_len, "q", "Dispatch queue length"); |
7c673cae | 3248 | mds_plb.add_u64_counter(l_mds_exported, "exported", "Exports"); |
7c673cae | 3249 | mds_plb.add_u64_counter(l_mds_imported, "imported", "Imports"); |
11fdf7f2 TL |
3250 | mds_plb.add_u64_counter(l_mds_openino_backtrace_fetch, "openino_backtrace_fetch", |
3251 | "OpenIno backtrace fetchings"); | |
3252 | mds_plb.add_u64_counter(l_mds_openino_peer_discover, "openino_peer_discover", | |
3253 | "OpenIno peer inode discovers"); | |
91327a77 | 3254 | |
7c673cae FG |
3255 | logger = mds_plb.create_perf_counters(); |
3256 | g_ceph_context->get_perfcounters_collection()->add(logger); | |
3257 | } | |
3258 | ||
3259 | { | |
3260 | PerfCountersBuilder mdm_plb(g_ceph_context, "mds_mem", l_mdm_first, l_mdm_last); | |
b32b8144 FG |
3261 | mdm_plb.add_u64(l_mdm_ino, "ino", "Inodes", "ino", |
3262 | PerfCountersBuilder::PRIO_INTERESTING); | |
91327a77 AA |
3263 | mdm_plb.add_u64(l_mdm_dn, "dn", "Dentries", "dn", |
3264 | PerfCountersBuilder::PRIO_INTERESTING); | |
3265 | ||
3266 | mdm_plb.set_prio_default(PerfCountersBuilder::PRIO_USEFUL); | |
7c673cae FG |
3267 | mdm_plb.add_u64_counter(l_mdm_inoa, "ino+", "Inodes opened"); |
3268 | mdm_plb.add_u64_counter(l_mdm_inos, "ino-", "Inodes closed"); | |
3269 | mdm_plb.add_u64(l_mdm_dir, "dir", "Directories"); | |
3270 | mdm_plb.add_u64_counter(l_mdm_dira, "dir+", "Directories opened"); | |
3271 | mdm_plb.add_u64_counter(l_mdm_dirs, "dir-", "Directories closed"); | |
7c673cae FG |
3272 | mdm_plb.add_u64_counter(l_mdm_dna, "dn+", "Dentries opened"); |
3273 | mdm_plb.add_u64_counter(l_mdm_dns, "dn-", "Dentries closed"); | |
3274 | mdm_plb.add_u64(l_mdm_cap, "cap", "Capabilities"); | |
3275 | mdm_plb.add_u64_counter(l_mdm_capa, "cap+", "Capabilities added"); | |
3276 | mdm_plb.add_u64_counter(l_mdm_caps, "cap-", "Capabilities removed"); | |
7c673cae | 3277 | mdm_plb.add_u64(l_mdm_heap, "heap", "Heap size"); |
91327a77 AA |
3278 | |
3279 | mdm_plb.set_prio_default(PerfCountersBuilder::PRIO_DEBUGONLY); | |
3280 | mdm_plb.add_u64(l_mdm_rss, "rss", "RSS"); | |
3281 | ||
7c673cae FG |
3282 | mlogger = mdm_plb.create_perf_counters(); |
3283 | g_ceph_context->get_perfcounters_collection()->add(mlogger); | |
3284 | } | |
3285 | ||
3286 | mdlog->create_logger(); | |
3287 | server->create_logger(); | |
3288 | purge_queue.create_logger(); | |
3289 | sessionmap.register_perfcounters(); | |
3290 | mdcache->register_perfcounters(); | |
3291 | } | |
3292 | ||
3293 | void MDSRank::check_ops_in_flight() | |
3294 | { | |
11fdf7f2 | 3295 | string summary; |
7c673cae FG |
3296 | vector<string> warnings; |
3297 | int slow = 0; | |
11fdf7f2 TL |
3298 | if (op_tracker.check_ops_in_flight(&summary, warnings, &slow)) { |
3299 | clog->warn() << summary; | |
3300 | for (const auto& warning : warnings) { | |
3301 | clog->warn() << warning; | |
7c673cae FG |
3302 | } |
3303 | } | |
3304 | ||
3305 | // set mds slow request count | |
3306 | mds_slow_req_count = slow; | |
3307 | return; | |
3308 | } | |
3309 | ||
3310 | void MDSRankDispatcher::handle_osd_map() | |
3311 | { | |
11fdf7f2 TL |
3312 | if (is_active() && |
3313 | mdsmap->get_tableserver() == whoami) { | |
7c673cae FG |
3314 | snapserver->check_osd_map(true); |
3315 | } | |
3316 | ||
3317 | server->handle_osd_map(); | |
3318 | ||
3319 | purge_queue.update_op_limit(*mdsmap); | |
3320 | ||
31f18b77 FG |
3321 | std::set<entity_addr_t> newly_blacklisted; |
3322 | objecter->consume_blacklist_events(&newly_blacklisted); | |
3323 | auto epoch = objecter->with_osdmap([](const OSDMap &o){return o.get_epoch();}); | |
3324 | dout(4) << "handle_osd_map epoch " << epoch << ", " | |
3325 | << newly_blacklisted.size() << " new blacklist entries" << dendl; | |
3326 | auto victims = server->apply_blacklist(newly_blacklisted); | |
3327 | if (victims) { | |
3328 | set_osd_epoch_barrier(epoch); | |
3329 | } | |
3330 | ||
3331 | ||
7c673cae FG |
3332 | // By default the objecter only requests OSDMap updates on use, |
3333 | // we would like to always receive the latest maps in order to | |
3334 | // apply policy based on the FULL flag. | |
3335 | objecter->maybe_request_map(); | |
3336 | } | |
3337 | ||
92f5a8d4 TL |
3338 | int MDSRank::config_client(int64_t session_id, bool remove, |
3339 | const std::string& option, const std::string& value, | |
3340 | std::ostream& ss) | |
3341 | { | |
3342 | Session *session = sessionmap.get_session(entity_name_t(CEPH_ENTITY_TYPE_CLIENT, session_id)); | |
3343 | if (!session) { | |
3344 | ss << "session " << session_id << " not in sessionmap!"; | |
3345 | return -ENOENT; | |
3346 | } | |
3347 | ||
3348 | if (option == "timeout") { | |
3349 | if (remove) { | |
3350 | auto it = session->info.client_metadata.find("timeout"); | |
3351 | if (it == session->info.client_metadata.end()) { | |
3352 | ss << "Nonexistent config: " << option; | |
3353 | return -ENODATA; | |
3354 | } | |
3355 | session->info.client_metadata.erase(it); | |
3356 | } else { | |
3357 | char *end; | |
3358 | strtoul(value.c_str(), &end, 0); | |
3359 | if (*end) { | |
3360 | ss << "Invalid config for timeout: " << value; | |
3361 | return -EINVAL; | |
3362 | } | |
3363 | session->info.client_metadata[option] = value; | |
3364 | } | |
3365 | //sessionmap._mark_dirty(session, true); | |
3366 | } else { | |
3367 | ss << "Invalid config option: " << option; | |
3368 | return -EINVAL; | |
3369 | } | |
3370 | ||
3371 | return 0; | |
3372 | } | |
3373 | ||
31f18b77 | 3374 | bool MDSRank::evict_client(int64_t session_id, |
11fdf7f2 | 3375 | bool wait, bool blacklist, std::ostream& err_ss, |
31f18b77 | 3376 | Context *on_killed) |
7c673cae | 3377 | { |
9f95a23c | 3378 | ceph_assert(ceph_mutex_is_locked_by_me(mds_lock)); |
31f18b77 FG |
3379 | |
3380 | // Mutually exclusive args | |
11fdf7f2 | 3381 | ceph_assert(!(wait && on_killed != nullptr)); |
31f18b77 | 3382 | |
7c673cae FG |
3383 | if (is_any_replay()) { |
3384 | err_ss << "MDS is replaying log"; | |
3385 | return false; | |
3386 | } | |
3387 | ||
31f18b77 FG |
3388 | Session *session = sessionmap.get_session( |
3389 | entity_name_t(CEPH_ENTITY_TYPE_CLIENT, session_id)); | |
7c673cae FG |
3390 | if (!session) { |
3391 | err_ss << "session " << session_id << " not in sessionmap!"; | |
3392 | return false; | |
3393 | } | |
7c673cae | 3394 | |
a8e16298 TL |
3395 | auto& addr = session->info.inst.addr; |
3396 | { | |
11fdf7f2 TL |
3397 | CachedStackStringStream css; |
3398 | *css << "Evicting " << (blacklist ? "(and blacklisting) " : "") | |
3399 | << "client session " << session_id << " (" << addr << ")"; | |
3400 | dout(1) << css->strv() << dendl; | |
3401 | clog->info() << css->strv(); | |
a8e16298 TL |
3402 | } |
3403 | ||
31f18b77 FG |
3404 | dout(4) << "Preparing blacklist command... (wait=" << wait << ")" << dendl; |
3405 | stringstream ss; | |
3406 | ss << "{\"prefix\":\"osd blacklist\", \"blacklistop\":\"add\","; | |
3407 | ss << "\"addr\":\""; | |
a8e16298 | 3408 | ss << addr; |
31f18b77 FG |
3409 | ss << "\"}"; |
3410 | std::string tmp = ss.str(); | |
3411 | std::vector<std::string> cmd = {tmp}; | |
3412 | ||
91327a77 | 3413 | auto kill_client_session = [this, session_id, wait, on_killed](){ |
9f95a23c | 3414 | ceph_assert(ceph_mutex_is_locked_by_me(mds_lock)); |
31f18b77 FG |
3415 | Session *session = sessionmap.get_session( |
3416 | entity_name_t(CEPH_ENTITY_TYPE_CLIENT, session_id)); | |
3417 | if (session) { | |
91327a77 | 3418 | if (on_killed || !wait) { |
31f18b77 FG |
3419 | server->kill_session(session, on_killed); |
3420 | } else { | |
3421 | C_SaferCond on_safe; | |
3422 | server->kill_session(session, &on_safe); | |
3423 | ||
9f95a23c | 3424 | mds_lock.unlock(); |
31f18b77 | 3425 | on_safe.wait(); |
9f95a23c | 3426 | mds_lock.lock(); |
31f18b77 FG |
3427 | } |
3428 | } else { | |
3429 | dout(1) << "session " << session_id << " was removed while we waited " | |
3430 | "for blacklist" << dendl; | |
3431 | ||
3432 | // Even though it wasn't us that removed it, kick our completion | |
3433 | // as the session has been removed. | |
3434 | if (on_killed) { | |
3435 | on_killed->complete(0); | |
3436 | } | |
3437 | } | |
3438 | }; | |
3439 | ||
91327a77 | 3440 | auto apply_blacklist = [this, cmd](std::function<void ()> fn){ |
9f95a23c | 3441 | ceph_assert(ceph_mutex_is_locked_by_me(mds_lock)); |
31f18b77 | 3442 | |
9f95a23c | 3443 | Context *on_blacklist_done = new LambdaContext([this, fn](int r) { |
31f18b77 FG |
3444 | objecter->wait_for_latest_osdmap( |
3445 | new C_OnFinisher( | |
9f95a23c | 3446 | new LambdaContext([this, fn](int r) { |
11fdf7f2 | 3447 | std::lock_guard l(mds_lock); |
31f18b77 FG |
3448 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ |
3449 | return o.get_epoch(); | |
3450 | }); | |
3451 | ||
3452 | set_osd_epoch_barrier(epoch); | |
3453 | ||
3454 | fn(); | |
3455 | }), finisher) | |
3456 | ); | |
3457 | }); | |
3458 | ||
3459 | dout(4) << "Sending mon blacklist command: " << cmd[0] << dendl; | |
3460 | monc->start_mon_command(cmd, {}, nullptr, nullptr, on_blacklist_done); | |
3461 | }; | |
3462 | ||
31f18b77 FG |
3463 | if (wait) { |
3464 | if (blacklist) { | |
91327a77 AA |
3465 | C_SaferCond inline_ctx; |
3466 | apply_blacklist([&inline_ctx](){inline_ctx.complete(0);}); | |
9f95a23c | 3467 | mds_lock.unlock(); |
91327a77 | 3468 | inline_ctx.wait(); |
9f95a23c | 3469 | mds_lock.lock(); |
31f18b77 FG |
3470 | } |
3471 | ||
3472 | // We dropped mds_lock, so check that session still exists | |
3473 | session = sessionmap.get_session(entity_name_t(CEPH_ENTITY_TYPE_CLIENT, | |
11fdf7f2 | 3474 | session_id)); |
31f18b77 FG |
3475 | if (!session) { |
3476 | dout(1) << "session " << session_id << " was removed while we waited " | |
3477 | "for blacklist" << dendl; | |
3478 | return true; | |
3479 | } | |
91327a77 | 3480 | kill_client_session(); |
7c673cae | 3481 | } else { |
31f18b77 | 3482 | if (blacklist) { |
91327a77 | 3483 | apply_blacklist(kill_client_session); |
31f18b77 | 3484 | } else { |
91327a77 | 3485 | kill_client_session(); |
31f18b77 | 3486 | } |
7c673cae | 3487 | } |
31f18b77 | 3488 | |
7c673cae FG |
3489 | return true; |
3490 | } | |
3491 | ||
3492 | void MDSRank::bcast_mds_map() | |
3493 | { | |
3494 | dout(7) << "bcast_mds_map " << mdsmap->get_epoch() << dendl; | |
3495 | ||
3496 | // share the map with mounted clients | |
3497 | set<Session*> clients; | |
3498 | sessionmap.get_client_session_set(clients); | |
11fdf7f2 | 3499 | for (const auto &session : clients) { |
9f95a23c | 3500 | auto m = make_message<MMDSMap>(monc->get_fsid(), *mdsmap); |
11fdf7f2 TL |
3501 | session->get_connection()->send_message2(std::move(m)); |
3502 | } | |
7c673cae FG |
3503 | last_client_mdsmap_bcast = mdsmap->get_epoch(); |
3504 | } | |
3505 | ||
3506 | MDSRankDispatcher::MDSRankDispatcher( | |
3507 | mds_rank_t whoami_, | |
9f95a23c | 3508 | ceph::mutex &mds_lock_, |
7c673cae FG |
3509 | LogChannelRef &clog_, |
3510 | SafeTimer &timer_, | |
3511 | Beacon &beacon_, | |
11fdf7f2 | 3512 | std::unique_ptr<MDSMap> &mdsmap_, |
7c673cae FG |
3513 | Messenger *msgr, |
3514 | MonClient *monc_, | |
9f95a23c | 3515 | MgrClient *mgrc, |
7c673cae FG |
3516 | Context *respawn_hook_, |
3517 | Context *suicide_hook_) | |
3518 | : MDSRank(whoami_, mds_lock_, clog_, timer_, beacon_, mdsmap_, | |
9f95a23c | 3519 | msgr, monc_, mgrc, respawn_hook_, suicide_hook_) |
92f5a8d4 | 3520 | { |
9f95a23c | 3521 | g_conf().add_observer(this); |
7c673cae FG |
3522 | } |
3523 | ||
f64942e4 AA |
3524 | void MDSRank::command_cache_drop(uint64_t timeout, Formatter *f, Context *on_finish) { |
3525 | dout(20) << __func__ << dendl; | |
3526 | ||
11fdf7f2 | 3527 | std::lock_guard locker(mds_lock); |
f64942e4 AA |
3528 | C_Drop_Cache *request = new C_Drop_Cache(server, mdcache, mdlog, this, |
3529 | timeout, f, on_finish); | |
3530 | request->send(); | |
3531 | } | |
3532 | ||
7c673cae FG |
3533 | epoch_t MDSRank::get_osd_epoch() const |
3534 | { | |
3535 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
3536 | } | |
3537 | ||
92f5a8d4 TL |
3538 | const char** MDSRankDispatcher::get_tracked_conf_keys() const |
3539 | { | |
3540 | static const char* KEYS[] = { | |
3541 | "clog_to_graylog", | |
3542 | "clog_to_graylog_host", | |
3543 | "clog_to_graylog_port", | |
3544 | "clog_to_monitors", | |
3545 | "clog_to_syslog", | |
3546 | "clog_to_syslog_facility", | |
3547 | "clog_to_syslog_level", | |
3548 | "fsid", | |
3549 | "host", | |
3550 | "mds_bal_fragment_dirs", | |
3551 | "mds_bal_fragment_interval", | |
3552 | "mds_cache_memory_limit", | |
3553 | "mds_cache_mid", | |
3554 | "mds_cache_reservation", | |
92f5a8d4 TL |
3555 | "mds_cache_trim_decay_rate", |
3556 | "mds_cap_revoke_eviction_timeout", | |
3557 | "mds_dump_cache_threshold_file", | |
3558 | "mds_dump_cache_threshold_formatter", | |
3559 | "mds_enable_op_tracker", | |
f6b5b4d7 TL |
3560 | "mds_export_ephemeral_random", |
3561 | "mds_export_ephemeral_random_max", | |
3562 | "mds_export_ephemeral_distributed", | |
92f5a8d4 TL |
3563 | "mds_health_cache_threshold", |
3564 | "mds_inject_migrator_session_race", | |
3565 | "mds_log_pause", | |
3566 | "mds_max_export_size", | |
3567 | "mds_max_purge_files", | |
9f95a23c | 3568 | "mds_forward_all_requests_to_auth", |
92f5a8d4 TL |
3569 | "mds_max_purge_ops", |
3570 | "mds_max_purge_ops_per_pg", | |
9f95a23c | 3571 | "mds_max_snaps_per_dir", |
92f5a8d4 TL |
3572 | "mds_op_complaint_time", |
3573 | "mds_op_history_duration", | |
3574 | "mds_op_history_size", | |
3575 | "mds_op_log_threshold", | |
3576 | "mds_recall_max_decay_rate", | |
3577 | "mds_recall_warning_decay_rate", | |
3578 | "mds_request_load_average_decay_rate", | |
3579 | "mds_session_cache_liveness_decay_rate", | |
3580 | "mds_replay_unsafe_with_closed_session", | |
3581 | NULL | |
3582 | }; | |
3583 | return KEYS; | |
3584 | } | |
3585 | ||
3586 | void MDSRankDispatcher::handle_conf_change(const ConfigProxy& conf, const std::set<std::string>& changed) | |
3587 | { | |
3588 | // XXX with or without mds_lock! | |
3589 | ||
3590 | if (changed.count("mds_op_complaint_time") || changed.count("mds_op_log_threshold")) { | |
3591 | op_tracker.set_complaint_and_threshold(conf->mds_op_complaint_time, conf->mds_op_log_threshold); | |
3592 | } | |
3593 | if (changed.count("mds_op_history_size") || changed.count("mds_op_history_duration")) { | |
3594 | op_tracker.set_history_size_and_duration(conf->mds_op_history_size, conf->mds_op_history_duration); | |
3595 | } | |
3596 | if (changed.count("mds_enable_op_tracker")) { | |
3597 | op_tracker.set_tracking(conf->mds_enable_op_tracker); | |
3598 | } | |
3599 | if (changed.count("clog_to_monitors") || | |
3600 | changed.count("clog_to_syslog") || | |
3601 | changed.count("clog_to_syslog_level") || | |
3602 | changed.count("clog_to_syslog_facility") || | |
3603 | changed.count("clog_to_graylog") || | |
3604 | changed.count("clog_to_graylog_host") || | |
3605 | changed.count("clog_to_graylog_port") || | |
3606 | changed.count("host") || | |
3607 | changed.count("fsid")) { | |
3608 | update_log_config(); | |
3609 | } | |
3610 | ||
9f95a23c | 3611 | finisher->queue(new LambdaContext([this, changed](int) { |
92f5a8d4 TL |
3612 | std::scoped_lock lock(mds_lock); |
3613 | ||
f6b5b4d7 TL |
3614 | dout(10) << "flushing conf change to components: " << changed << dendl; |
3615 | ||
92f5a8d4 TL |
3616 | if (changed.count("mds_log_pause") && !g_conf()->mds_log_pause) { |
3617 | mdlog->kick_submitter(); | |
3618 | } | |
3619 | sessionmap.handle_conf_change(changed); | |
3620 | server->handle_conf_change(changed); | |
3621 | mdcache->handle_conf_change(changed, *mdsmap); | |
3622 | purge_queue.handle_conf_change(changed, *mdsmap); | |
3623 | })); | |
3624 | } | |
9f95a23c TL |
3625 | |
3626 | void MDSRank::get_task_status(std::map<std::string, std::string> *status) { | |
3627 | dout(20) << __func__ << dendl; | |
3628 | ||
3629 | // scrub summary for now.. | |
3630 | std::string_view scrub_summary = scrubstack->scrub_summary(); | |
3631 | status->emplace(SCRUB_STATUS_KEY, std::move(scrub_summary)); | |
3632 | } | |
3633 | ||
3634 | void MDSRank::schedule_update_timer_task() { | |
3635 | dout(20) << __func__ << dendl; | |
3636 | ||
3637 | timer.add_event_after(g_conf().get_val<double>("mds_task_status_update_interval"), | |
3638 | new LambdaContext([this](int) { | |
3639 | send_task_status(); | |
3640 | })); | |
3641 | } | |
3642 | ||
3643 | void MDSRank::send_task_status() { | |
3644 | std::map<std::string, std::string> status; | |
3645 | get_task_status(&status); | |
3646 | ||
3647 | if (!status.empty()) { | |
3648 | dout(20) << __func__ << ": updating " << status.size() << " status keys" << dendl; | |
3649 | ||
3650 | int r = mgrc->service_daemon_update_task_status(std::move(status)); | |
3651 | if (r < 0) { | |
3652 | derr << ": failed to update service daemon status: " << cpp_strerror(r) << dendl; | |
3653 | } | |
3654 | } | |
3655 | ||
3656 | schedule_update_timer_task(); | |
3657 | } |