]>
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) 2004-2006 Sage Weil <sage@newdream.net> | |
7 | * Copyright (C) 2013,2014 Cloudwatt <libre.licensing@cloudwatt.com> | |
8 | * Copyright (C) 2014 Red Hat <contact@redhat.com> | |
9 | * | |
10 | * Author: Loic Dachary <loic@dachary.org> | |
11 | * | |
12 | * This is free software; you can redistribute it and/or | |
13 | * modify it under the terms of the GNU Lesser General Public | |
14 | * License version 2.1, as published by the Free Software | |
15 | * Foundation. See file COPYING. | |
16 | * | |
17 | */ | |
18 | ||
19 | #include <algorithm> | |
224ce89b | 20 | #include <boost/algorithm/string.hpp> |
11fdf7f2 | 21 | #include <experimental/iterator> |
224ce89b | 22 | #include <locale> |
7c673cae FG |
23 | #include <sstream> |
24 | ||
31f18b77 FG |
25 | #include "mon/OSDMonitor.h" |
26 | #include "mon/Monitor.h" | |
27 | #include "mon/MDSMonitor.h" | |
31f18b77 FG |
28 | #include "mon/MgrStatMonitor.h" |
29 | #include "mon/AuthMonitor.h" | |
30 | #include "mon/ConfigKeyService.h" | |
7c673cae | 31 | |
31f18b77 FG |
32 | #include "mon/MonitorDBStore.h" |
33 | #include "mon/Session.h" | |
7c673cae FG |
34 | |
35 | #include "crush/CrushWrapper.h" | |
36 | #include "crush/CrushTester.h" | |
37 | #include "crush/CrushTreeDumper.h" | |
38 | ||
39 | #include "messages/MOSDBeacon.h" | |
40 | #include "messages/MOSDFailure.h" | |
41 | #include "messages/MOSDMarkMeDown.h" | |
42 | #include "messages/MOSDFull.h" | |
43 | #include "messages/MOSDMap.h" | |
44 | #include "messages/MMonGetOSDMap.h" | |
45 | #include "messages/MOSDBoot.h" | |
46 | #include "messages/MOSDAlive.h" | |
47 | #include "messages/MPoolOp.h" | |
48 | #include "messages/MPoolOpReply.h" | |
49 | #include "messages/MOSDPGCreate.h" | |
11fdf7f2 | 50 | #include "messages/MOSDPGCreate2.h" |
7c673cae FG |
51 | #include "messages/MOSDPGCreated.h" |
52 | #include "messages/MOSDPGTemp.h" | |
11fdf7f2 | 53 | #include "messages/MOSDPGReadyToMerge.h" |
7c673cae FG |
54 | #include "messages/MMonCommand.h" |
55 | #include "messages/MRemoveSnaps.h" | |
56 | #include "messages/MOSDScrub.h" | |
57 | #include "messages/MRoute.h" | |
58 | ||
59 | #include "common/TextTable.h" | |
60 | #include "common/Timer.h" | |
61 | #include "common/ceph_argparse.h" | |
62 | #include "common/perf_counters.h" | |
eafe8130 | 63 | #include "common/PriorityCache.h" |
7c673cae | 64 | #include "common/strtol.h" |
11fdf7f2 | 65 | #include "common/numa.h" |
7c673cae FG |
66 | |
67 | #include "common/config.h" | |
68 | #include "common/errno.h" | |
69 | ||
70 | #include "erasure-code/ErasureCodePlugin.h" | |
71 | #include "compressor/Compressor.h" | |
72 | #include "common/Checksummer.h" | |
73 | ||
74 | #include "include/compat.h" | |
11fdf7f2 | 75 | #include "include/ceph_assert.h" |
7c673cae FG |
76 | #include "include/stringify.h" |
77 | #include "include/util.h" | |
78 | #include "common/cmdparse.h" | |
79 | #include "include/str_list.h" | |
80 | #include "include/str_map.h" | |
224ce89b | 81 | #include "include/scope_guard.h" |
eafe8130 | 82 | #include "perfglue/heap_profiler.h" |
7c673cae | 83 | |
28e407b8 AA |
84 | #include "auth/cephx/CephxKeyServer.h" |
85 | #include "osd/OSDCap.h" | |
86 | ||
7c673cae FG |
87 | #include "json_spirit/json_spirit_reader.h" |
88 | ||
c07f9fc5 FG |
89 | #include <boost/algorithm/string/predicate.hpp> |
90 | ||
7c673cae | 91 | #define dout_subsys ceph_subsys_mon |
3efd9988 FG |
92 | static const string OSD_PG_CREATING_PREFIX("osd_pg_creating"); |
93 | static const string OSD_METADATA_PREFIX("osd_metadata"); | |
11fdf7f2 | 94 | static const string OSD_SNAP_PREFIX("osd_snap"); |
7c673cae | 95 | |
c07f9fc5 FG |
96 | namespace { |
97 | ||
eafe8130 TL |
98 | struct OSDMemCache : public PriorityCache::PriCache { |
99 | OSDMonitor *osdmon; | |
100 | int64_t cache_bytes[PriorityCache::Priority::LAST+1] = {0}; | |
101 | int64_t committed_bytes = 0; | |
102 | double cache_ratio = 0; | |
103 | ||
104 | OSDMemCache(OSDMonitor *m) : osdmon(m) {}; | |
105 | ||
106 | virtual uint64_t _get_used_bytes() const = 0; | |
107 | ||
108 | virtual int64_t request_cache_bytes( | |
109 | PriorityCache::Priority pri, uint64_t total_cache) const { | |
110 | int64_t assigned = get_cache_bytes(pri); | |
111 | ||
112 | switch (pri) { | |
113 | // All cache items are currently set to have PRI1 priority | |
114 | case PriorityCache::Priority::PRI1: | |
115 | { | |
116 | int64_t request = _get_used_bytes(); | |
117 | return (request > assigned) ? request - assigned : 0; | |
118 | } | |
119 | default: | |
120 | break; | |
121 | } | |
122 | return -EOPNOTSUPP; | |
123 | } | |
124 | ||
125 | virtual int64_t get_cache_bytes(PriorityCache::Priority pri) const { | |
126 | return cache_bytes[pri]; | |
127 | } | |
128 | ||
129 | virtual int64_t get_cache_bytes() const { | |
130 | int64_t total = 0; | |
131 | ||
132 | for (int i = 0; i < PriorityCache::Priority::LAST + 1; i++) { | |
133 | PriorityCache::Priority pri = static_cast<PriorityCache::Priority>(i); | |
134 | total += get_cache_bytes(pri); | |
135 | } | |
136 | return total; | |
137 | } | |
138 | ||
139 | virtual void set_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
140 | cache_bytes[pri] = bytes; | |
141 | } | |
142 | virtual void add_cache_bytes(PriorityCache::Priority pri, int64_t bytes) { | |
143 | cache_bytes[pri] += bytes; | |
144 | } | |
145 | virtual int64_t commit_cache_size(uint64_t total_cache) { | |
146 | committed_bytes = PriorityCache::get_chunk( | |
147 | get_cache_bytes(), total_cache); | |
148 | return committed_bytes; | |
149 | } | |
150 | virtual int64_t get_committed_size() const { | |
151 | return committed_bytes; | |
152 | } | |
153 | virtual double get_cache_ratio() const { | |
154 | return cache_ratio; | |
155 | } | |
156 | virtual void set_cache_ratio(double ratio) { | |
157 | cache_ratio = ratio; | |
158 | } | |
159 | virtual string get_cache_name() const = 0; | |
160 | }; | |
161 | ||
162 | struct IncCache : public OSDMemCache { | |
163 | IncCache(OSDMonitor *m) : OSDMemCache(m) {}; | |
164 | ||
165 | virtual uint64_t _get_used_bytes() const { | |
166 | return osdmon->inc_osd_cache.get_bytes(); | |
167 | } | |
168 | ||
169 | virtual string get_cache_name() const { | |
170 | return "OSDMap Inc Cache"; | |
171 | } | |
172 | ||
173 | uint64_t _get_num_osdmaps() const { | |
174 | return osdmon->inc_osd_cache.get_size(); | |
175 | } | |
176 | }; | |
177 | ||
178 | struct FullCache : public OSDMemCache { | |
179 | FullCache(OSDMonitor *m) : OSDMemCache(m) {}; | |
180 | ||
181 | virtual uint64_t _get_used_bytes() const { | |
182 | return osdmon->full_osd_cache.get_bytes(); | |
183 | } | |
184 | ||
185 | virtual string get_cache_name() const { | |
186 | return "OSDMap Full Cache"; | |
187 | } | |
188 | ||
189 | uint64_t _get_num_osdmaps() const { | |
190 | return osdmon->full_osd_cache.get_size(); | |
191 | } | |
192 | }; | |
193 | ||
194 | std::shared_ptr<IncCache> inc_cache; | |
195 | std::shared_ptr<FullCache> full_cache; | |
196 | ||
c07f9fc5 FG |
197 | const uint32_t MAX_POOL_APPLICATIONS = 4; |
198 | const uint32_t MAX_POOL_APPLICATION_KEYS = 64; | |
199 | const uint32_t MAX_POOL_APPLICATION_LENGTH = 128; | |
200 | ||
28e407b8 AA |
201 | bool is_osd_writable(const OSDCapGrant& grant, const std::string* pool_name) { |
202 | // Note: this doesn't include support for the application tag match | |
203 | if ((grant.spec.allow & OSD_CAP_W) != 0) { | |
204 | auto& match = grant.match; | |
205 | if (match.is_match_all()) { | |
206 | return true; | |
11fdf7f2 | 207 | } else if (pool_name != nullptr && |
28e407b8 AA |
208 | !match.pool_namespace.pool_name.empty() && |
209 | match.pool_namespace.pool_name == *pool_name) { | |
210 | return true; | |
211 | } | |
212 | } | |
213 | return false; | |
214 | } | |
215 | ||
216 | bool is_unmanaged_snap_op_permitted(CephContext* cct, | |
217 | const KeyServer& key_server, | |
218 | const EntityName& entity_name, | |
219 | const MonCap& mon_caps, | |
11fdf7f2 | 220 | const entity_addr_t& peer_socket_addr, |
28e407b8 AA |
221 | const std::string* pool_name) |
222 | { | |
223 | typedef std::map<std::string, std::string> CommandArgs; | |
224 | ||
11fdf7f2 TL |
225 | if (mon_caps.is_capable( |
226 | cct, CEPH_ENTITY_TYPE_MON, | |
227 | entity_name, "osd", | |
228 | "osd pool op unmanaged-snap", | |
229 | (pool_name == nullptr ? | |
230 | CommandArgs{} /* pool DNE, require unrestricted cap */ : | |
231 | CommandArgs{{"poolname", *pool_name}}), | |
232 | false, true, false, | |
233 | peer_socket_addr)) { | |
28e407b8 AA |
234 | return true; |
235 | } | |
236 | ||
237 | AuthCapsInfo caps_info; | |
238 | if (!key_server.get_service_caps(entity_name, CEPH_ENTITY_TYPE_OSD, | |
239 | caps_info)) { | |
240 | dout(10) << "unable to locate OSD cap data for " << entity_name | |
241 | << " in auth db" << dendl; | |
242 | return false; | |
243 | } | |
244 | ||
245 | string caps_str; | |
246 | if (caps_info.caps.length() > 0) { | |
11fdf7f2 | 247 | auto p = caps_info.caps.cbegin(); |
28e407b8 AA |
248 | try { |
249 | decode(caps_str, p); | |
250 | } catch (const buffer::error &err) { | |
251 | derr << "corrupt OSD cap data for " << entity_name << " in auth db" | |
252 | << dendl; | |
253 | return false; | |
254 | } | |
255 | } | |
256 | ||
257 | OSDCap osd_cap; | |
258 | if (!osd_cap.parse(caps_str, nullptr)) { | |
259 | dout(10) << "unable to parse OSD cap data for " << entity_name | |
260 | << " in auth db" << dendl; | |
261 | return false; | |
262 | } | |
263 | ||
264 | // if the entity has write permissions in one or all pools, permit | |
265 | // usage of unmanaged-snapshots | |
266 | if (osd_cap.allow_all()) { | |
267 | return true; | |
268 | } | |
269 | ||
270 | for (auto& grant : osd_cap.grants) { | |
271 | if (grant.profile.is_valid()) { | |
272 | for (auto& profile_grant : grant.profile_grants) { | |
273 | if (is_osd_writable(profile_grant, pool_name)) { | |
274 | return true; | |
275 | } | |
276 | } | |
277 | } else if (is_osd_writable(grant, pool_name)) { | |
278 | return true; | |
279 | } | |
280 | } | |
281 | ||
282 | return false; | |
283 | } | |
284 | ||
c07f9fc5 FG |
285 | } // anonymous namespace |
286 | ||
7c673cae FG |
287 | void LastEpochClean::Lec::report(ps_t ps, epoch_t last_epoch_clean) |
288 | { | |
289 | if (epoch_by_pg.size() <= ps) { | |
290 | epoch_by_pg.resize(ps + 1, 0); | |
291 | } | |
292 | const auto old_lec = epoch_by_pg[ps]; | |
293 | if (old_lec >= last_epoch_clean) { | |
294 | // stale lec | |
295 | return; | |
296 | } | |
297 | epoch_by_pg[ps] = last_epoch_clean; | |
298 | if (last_epoch_clean < floor) { | |
299 | floor = last_epoch_clean; | |
300 | } else if (last_epoch_clean > floor) { | |
301 | if (old_lec == floor) { | |
302 | // probably should increase floor? | |
303 | auto new_floor = std::min_element(std::begin(epoch_by_pg), | |
304 | std::end(epoch_by_pg)); | |
305 | floor = *new_floor; | |
306 | } | |
307 | } | |
308 | if (ps != next_missing) { | |
309 | return; | |
310 | } | |
311 | for (; next_missing < epoch_by_pg.size(); next_missing++) { | |
312 | if (epoch_by_pg[next_missing] == 0) { | |
313 | break; | |
314 | } | |
315 | } | |
316 | } | |
317 | ||
318 | void LastEpochClean::remove_pool(uint64_t pool) | |
319 | { | |
320 | report_by_pool.erase(pool); | |
321 | } | |
322 | ||
323 | void LastEpochClean::report(const pg_t& pg, epoch_t last_epoch_clean) | |
324 | { | |
325 | auto& lec = report_by_pool[pg.pool()]; | |
326 | return lec.report(pg.ps(), last_epoch_clean); | |
327 | } | |
328 | ||
329 | epoch_t LastEpochClean::get_lower_bound(const OSDMap& latest) const | |
330 | { | |
331 | auto floor = latest.get_epoch(); | |
332 | for (auto& pool : latest.get_pools()) { | |
333 | auto reported = report_by_pool.find(pool.first); | |
334 | if (reported == report_by_pool.end()) { | |
335 | return 0; | |
336 | } | |
337 | if (reported->second.next_missing < pool.second.get_pg_num()) { | |
338 | return 0; | |
339 | } | |
340 | if (reported->second.floor < floor) { | |
341 | floor = reported->second.floor; | |
342 | } | |
343 | } | |
344 | return floor; | |
345 | } | |
346 | ||
347 | ||
11fdf7f2 TL |
348 | class C_UpdateCreatingPGs : public Context { |
349 | public: | |
7c673cae FG |
350 | OSDMonitor *osdmon; |
351 | utime_t start; | |
352 | epoch_t epoch; | |
353 | C_UpdateCreatingPGs(OSDMonitor *osdmon, epoch_t e) : | |
354 | osdmon(osdmon), start(ceph_clock_now()), epoch(e) {} | |
355 | void finish(int r) override { | |
356 | if (r >= 0) { | |
357 | utime_t end = ceph_clock_now(); | |
358 | dout(10) << "osdmap epoch " << epoch << " mapping took " | |
359 | << (end - start) << " seconds" << dendl; | |
360 | osdmon->update_creating_pgs(); | |
361 | osdmon->check_pg_creates_subs(); | |
362 | } | |
363 | } | |
364 | }; | |
365 | ||
366 | #undef dout_prefix | |
367 | #define dout_prefix _prefix(_dout, mon, osdmap) | |
368 | static ostream& _prefix(std::ostream *_dout, Monitor *mon, const OSDMap& osdmap) { | |
369 | return *_dout << "mon." << mon->name << "@" << mon->rank | |
370 | << "(" << mon->get_state_name() | |
371 | << ").osd e" << osdmap.get_epoch() << " "; | |
372 | } | |
373 | ||
374 | OSDMonitor::OSDMonitor( | |
375 | CephContext *cct, | |
376 | Monitor *mn, | |
377 | Paxos *p, | |
378 | const string& service_name) | |
379 | : PaxosService(mn, p, service_name), | |
380 | cct(cct), | |
11fdf7f2 TL |
381 | inc_osd_cache(g_conf()->mon_osd_cache_size), |
382 | full_osd_cache(g_conf()->mon_osd_cache_size), | |
383 | has_osdmap_manifest(false), | |
384 | mapper(mn->cct, &mn->cpu_tp) | |
eafe8130 TL |
385 | { |
386 | inc_cache = std::make_shared<IncCache>(this); | |
387 | full_cache = std::make_shared<FullCache>(this); | |
388 | cct->_conf.add_observer(this); | |
389 | int r = _set_cache_sizes(); | |
390 | if (r < 0) { | |
391 | derr << __func__ << " using default osd cache size - mon_osd_cache_size (" | |
392 | << g_conf()->mon_osd_cache_size | |
393 | << ") without priority cache management" | |
394 | << dendl; | |
395 | } | |
396 | } | |
397 | ||
398 | const char **OSDMonitor::get_tracked_conf_keys() const | |
399 | { | |
400 | static const char* KEYS[] = { | |
401 | "mon_memory_target", | |
402 | "mon_memory_autotune", | |
403 | "rocksdb_cache_size", | |
404 | NULL | |
405 | }; | |
406 | return KEYS; | |
407 | } | |
408 | ||
409 | void OSDMonitor::handle_conf_change(const ConfigProxy& conf, | |
410 | const std::set<std::string> &changed) | |
411 | { | |
412 | dout(10) << __func__ << " " << changed << dendl; | |
413 | ||
414 | if (changed.count("mon_memory_autotune")) { | |
415 | _set_cache_autotuning(); | |
416 | } | |
417 | if (changed.count("mon_memory_target") || | |
418 | changed.count("rocksdb_cache_size")) { | |
419 | int r = _update_mon_cache_settings(); | |
420 | if (r < 0) { | |
421 | derr << __func__ << " mon_memory_target:" | |
422 | << g_conf()->mon_memory_target | |
423 | << " rocksdb_cache_size:" | |
424 | << g_conf()->rocksdb_cache_size | |
425 | << ". Invalid size provided." | |
426 | << dendl; | |
427 | } | |
428 | } | |
429 | } | |
430 | ||
431 | void OSDMonitor::_set_cache_autotuning() | |
432 | { | |
433 | if (!g_conf()->mon_memory_autotune && pcm != nullptr) { | |
434 | // Disable cache autotuning | |
435 | std::lock_guard l(balancer_lock); | |
436 | pcm = nullptr; | |
437 | } | |
438 | ||
439 | if (g_conf()->mon_memory_autotune && pcm == nullptr) { | |
440 | int r = register_cache_with_pcm(); | |
441 | if (r < 0) { | |
442 | dout(10) << __func__ | |
443 | << " Error while registering osdmon caches with pcm." | |
444 | << " Cache auto tuning not enabled." | |
445 | << dendl; | |
446 | mon_memory_autotune = false; | |
447 | } else { | |
448 | mon_memory_autotune = true; | |
449 | } | |
450 | } | |
451 | } | |
452 | ||
453 | int OSDMonitor::_update_mon_cache_settings() | |
454 | { | |
455 | if (g_conf()->mon_memory_target <= 0 || | |
456 | g_conf()->mon_memory_target < mon_memory_min || | |
457 | g_conf()->rocksdb_cache_size <= 0) { | |
458 | return -EINVAL; | |
459 | } | |
460 | ||
461 | uint64_t old_mon_memory_target = mon_memory_target; | |
462 | uint64_t old_rocksdb_cache_size = rocksdb_cache_size; | |
463 | ||
464 | // Set the new pcm memory cache sizes | |
465 | mon_memory_target = g_conf()->mon_memory_target; | |
466 | rocksdb_cache_size = g_conf()->rocksdb_cache_size; | |
467 | ||
468 | uint64_t base = mon_memory_base; | |
469 | double fragmentation = mon_memory_fragmentation; | |
470 | uint64_t target = mon_memory_target; | |
471 | uint64_t min = mon_memory_min; | |
472 | uint64_t max = min; | |
473 | ||
474 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
475 | if (ltarget > base + min) { | |
476 | max = ltarget - base; | |
477 | } | |
478 | ||
479 | int r = _set_cache_ratios(); | |
480 | if (r < 0) { | |
481 | derr << __func__ << " Cache ratios for pcm could not be set." | |
482 | << " Review the kv (rocksdb) and mon_memory_target sizes." | |
483 | << dendl; | |
484 | mon_memory_target = old_mon_memory_target; | |
485 | rocksdb_cache_size = old_rocksdb_cache_size; | |
486 | return -EINVAL; | |
487 | } | |
488 | ||
489 | if (mon_memory_autotune && pcm != nullptr) { | |
490 | std::lock_guard l(balancer_lock); | |
491 | // set pcm cache levels | |
492 | pcm->set_target_memory(target); | |
493 | pcm->set_min_memory(min); | |
494 | pcm->set_max_memory(max); | |
495 | // tune memory based on new values | |
496 | pcm->tune_memory(); | |
497 | pcm->balance(); | |
498 | _set_new_cache_sizes(); | |
499 | dout(10) << __func__ << " Updated mon cache setting." | |
500 | << " target: " << target | |
501 | << " min: " << min | |
502 | << " max: " << max | |
503 | << dendl; | |
504 | } | |
505 | return 0; | |
506 | } | |
507 | ||
508 | int OSDMonitor::_set_cache_sizes() | |
509 | { | |
510 | if (g_conf()->mon_memory_autotune) { | |
511 | // set the new osdmon cache targets to be managed by pcm | |
512 | mon_osd_cache_size = g_conf()->mon_osd_cache_size; | |
513 | rocksdb_cache_size = g_conf()->rocksdb_cache_size; | |
514 | mon_memory_base = cct->_conf.get_val<Option::size_t>("osd_memory_base"); | |
515 | mon_memory_fragmentation = cct->_conf.get_val<double>("osd_memory_expected_fragmentation"); | |
516 | mon_memory_target = g_conf()->mon_memory_target; | |
517 | mon_memory_min = g_conf()->mon_osd_cache_size_min; | |
518 | if (mon_memory_target <= 0 || mon_memory_min <= 0) { | |
519 | derr << __func__ << " mon_memory_target:" << mon_memory_target | |
520 | << " mon_memory_min:" << mon_memory_min | |
521 | << ". Invalid size option(s) provided." | |
522 | << dendl; | |
523 | return -EINVAL; | |
524 | } | |
525 | // Set the initial inc and full LRU cache sizes | |
526 | inc_osd_cache.set_bytes(mon_memory_min); | |
527 | full_osd_cache.set_bytes(mon_memory_min); | |
528 | mon_memory_autotune = g_conf()->mon_memory_autotune; | |
529 | } | |
530 | return 0; | |
531 | } | |
7c673cae FG |
532 | |
533 | bool OSDMonitor::_have_pending_crush() | |
534 | { | |
535 | return pending_inc.crush.length() > 0; | |
536 | } | |
537 | ||
538 | CrushWrapper &OSDMonitor::_get_stable_crush() | |
539 | { | |
540 | return *osdmap.crush; | |
541 | } | |
542 | ||
543 | void OSDMonitor::_get_pending_crush(CrushWrapper& newcrush) | |
544 | { | |
545 | bufferlist bl; | |
546 | if (pending_inc.crush.length()) | |
547 | bl = pending_inc.crush; | |
548 | else | |
549 | osdmap.crush->encode(bl, CEPH_FEATURES_SUPPORTED_DEFAULT); | |
550 | ||
11fdf7f2 | 551 | auto p = bl.cbegin(); |
7c673cae FG |
552 | newcrush.decode(p); |
553 | } | |
554 | ||
555 | void OSDMonitor::create_initial() | |
556 | { | |
557 | dout(10) << "create_initial for " << mon->monmap->fsid << dendl; | |
558 | ||
559 | OSDMap newmap; | |
560 | ||
561 | bufferlist bl; | |
562 | mon->store->get("mkfs", "osdmap", bl); | |
563 | ||
564 | if (bl.length()) { | |
565 | newmap.decode(bl); | |
566 | newmap.set_fsid(mon->monmap->fsid); | |
567 | } else { | |
11fdf7f2 | 568 | newmap.build_simple(cct, 0, mon->monmap->fsid, 0); |
7c673cae FG |
569 | } |
570 | newmap.set_epoch(1); | |
571 | newmap.created = newmap.modified = ceph_clock_now(); | |
572 | ||
573 | // new clusters should sort bitwise by default. | |
574 | newmap.set_flag(CEPH_OSDMAP_SORTBITWISE); | |
575 | ||
11fdf7f2 TL |
576 | newmap.flags |= |
577 | CEPH_OSDMAP_RECOVERY_DELETES | | |
578 | CEPH_OSDMAP_PURGED_SNAPDIRS | | |
579 | CEPH_OSDMAP_PGLOG_HARDLIMIT; | |
580 | newmap.full_ratio = g_conf()->mon_osd_full_ratio; | |
581 | if (newmap.full_ratio > 1.0) newmap.full_ratio /= 100; | |
582 | newmap.backfillfull_ratio = g_conf()->mon_osd_backfillfull_ratio; | |
583 | if (newmap.backfillfull_ratio > 1.0) newmap.backfillfull_ratio /= 100; | |
584 | newmap.nearfull_ratio = g_conf()->mon_osd_nearfull_ratio; | |
585 | if (newmap.nearfull_ratio > 1.0) newmap.nearfull_ratio /= 100; | |
586 | ||
7c673cae | 587 | // new cluster should require latest by default |
11fdf7f2 TL |
588 | if (g_conf().get_val<bool>("mon_debug_no_require_nautilus")) { |
589 | if (g_conf()->mon_debug_no_require_mimic) { | |
590 | derr << __func__ << " mon_debug_no_require_mimic=true and nautilus=true" << dendl; | |
591 | newmap.require_osd_release = CEPH_RELEASE_LUMINOUS; | |
592 | } else { | |
593 | derr << __func__ << " mon_debug_no_require_nautilus=true" << dendl; | |
594 | newmap.require_osd_release = CEPH_RELEASE_MIMIC; | |
595 | } | |
31f18b77 | 596 | } else { |
11fdf7f2 | 597 | newmap.require_osd_release = CEPH_RELEASE_NAUTILUS; |
31f18b77 | 598 | int r = ceph_release_from_name( |
11fdf7f2 | 599 | g_conf()->mon_osd_initial_require_min_compat_client.c_str()); |
31f18b77 | 600 | if (r <= 0) { |
11fdf7f2 | 601 | ceph_abort_msg("mon_osd_initial_require_min_compat_client is not valid"); |
31f18b77 FG |
602 | } |
603 | newmap.require_min_compat_client = r; | |
7c673cae FG |
604 | } |
605 | ||
606 | // encode into pending incremental | |
28e407b8 | 607 | uint64_t features = newmap.get_encoding_features(); |
7c673cae | 608 | newmap.encode(pending_inc.fullmap, |
28e407b8 | 609 | features | CEPH_FEATURE_RESERVED); |
7c673cae FG |
610 | pending_inc.full_crc = newmap.get_crc(); |
611 | dout(20) << " full crc " << pending_inc.full_crc << dendl; | |
612 | } | |
613 | ||
11fdf7f2 | 614 | void OSDMonitor::get_store_prefixes(std::set<string>& s) const |
7c673cae FG |
615 | { |
616 | s.insert(service_name); | |
617 | s.insert(OSD_PG_CREATING_PREFIX); | |
3efd9988 | 618 | s.insert(OSD_METADATA_PREFIX); |
11fdf7f2 | 619 | s.insert(OSD_SNAP_PREFIX); |
7c673cae FG |
620 | } |
621 | ||
622 | void OSDMonitor::update_from_paxos(bool *need_bootstrap) | |
623 | { | |
11fdf7f2 TL |
624 | // we really don't care if the version has been updated, because we may |
625 | // have trimmed without having increased the last committed; yet, we may | |
626 | // need to update the in-memory manifest. | |
627 | load_osdmap_manifest(); | |
628 | ||
7c673cae FG |
629 | version_t version = get_last_committed(); |
630 | if (version == osdmap.epoch) | |
631 | return; | |
11fdf7f2 | 632 | ceph_assert(version > osdmap.epoch); |
7c673cae FG |
633 | |
634 | dout(15) << "update_from_paxos paxos e " << version | |
635 | << ", my e " << osdmap.epoch << dendl; | |
636 | ||
31f18b77 FG |
637 | if (mapping_job) { |
638 | if (!mapping_job->is_done()) { | |
639 | dout(1) << __func__ << " mapping job " | |
640 | << mapping_job.get() << " did not complete, " | |
641 | << mapping_job->shards << " left, canceling" << dendl; | |
642 | mapping_job->abort(); | |
643 | } | |
644 | mapping_job.reset(); | |
645 | } | |
7c673cae | 646 | |
224ce89b WB |
647 | load_health(); |
648 | ||
7c673cae FG |
649 | /* |
650 | * We will possibly have a stashed latest that *we* wrote, and we will | |
651 | * always be sure to have the oldest full map in the first..last range | |
652 | * due to encode_trim_extra(), which includes the oldest full map in the trim | |
653 | * transaction. | |
654 | * | |
655 | * encode_trim_extra() does not however write the full map's | |
656 | * version to 'full_latest'. This is only done when we are building the | |
657 | * full maps from the incremental versions. But don't panic! We make sure | |
658 | * that the following conditions find whichever full map version is newer. | |
659 | */ | |
660 | version_t latest_full = get_version_latest_full(); | |
661 | if (latest_full == 0 && get_first_committed() > 1) | |
662 | latest_full = get_first_committed(); | |
663 | ||
664 | if (get_first_committed() > 1 && | |
665 | latest_full < get_first_committed()) { | |
666 | // the monitor could be just sync'ed with its peer, and the latest_full key | |
667 | // is not encoded in the paxos commits in encode_pending(), so we need to | |
668 | // make sure we get it pointing to a proper version. | |
669 | version_t lc = get_last_committed(); | |
670 | version_t fc = get_first_committed(); | |
671 | ||
672 | dout(10) << __func__ << " looking for valid full map in interval" | |
673 | << " [" << fc << ", " << lc << "]" << dendl; | |
674 | ||
675 | latest_full = 0; | |
676 | for (version_t v = lc; v >= fc; v--) { | |
677 | string full_key = "full_" + stringify(v); | |
678 | if (mon->store->exists(get_service_name(), full_key)) { | |
679 | dout(10) << __func__ << " found latest full map v " << v << dendl; | |
680 | latest_full = v; | |
681 | break; | |
682 | } | |
683 | } | |
684 | ||
11fdf7f2 | 685 | ceph_assert(latest_full > 0); |
7c673cae FG |
686 | auto t(std::make_shared<MonitorDBStore::Transaction>()); |
687 | put_version_latest_full(t, latest_full); | |
688 | mon->store->apply_transaction(t); | |
689 | dout(10) << __func__ << " updated the on-disk full map version to " | |
690 | << latest_full << dendl; | |
691 | } | |
692 | ||
693 | if ((latest_full > 0) && (latest_full > osdmap.epoch)) { | |
694 | bufferlist latest_bl; | |
695 | get_version_full(latest_full, latest_bl); | |
11fdf7f2 | 696 | ceph_assert(latest_bl.length() != 0); |
7c673cae | 697 | dout(7) << __func__ << " loading latest full map e" << latest_full << dendl; |
11fdf7f2 | 698 | osdmap = OSDMap(); |
7c673cae FG |
699 | osdmap.decode(latest_bl); |
700 | } | |
701 | ||
11fdf7f2 TL |
702 | bufferlist bl; |
703 | if (!mon->store->get(OSD_PG_CREATING_PREFIX, "creating", bl)) { | |
704 | auto p = bl.cbegin(); | |
705 | std::lock_guard<std::mutex> l(creating_pgs_lock); | |
706 | creating_pgs.decode(p); | |
707 | dout(7) << __func__ << " loading creating_pgs last_scan_epoch " | |
708 | << creating_pgs.last_scan_epoch | |
709 | << " with " << creating_pgs.pgs.size() << " pgs" << dendl; | |
31f18b77 | 710 | } else { |
11fdf7f2 TL |
711 | dout(1) << __func__ << " missing creating pgs; upgrade from post-kraken?" |
712 | << dendl; | |
31f18b77 FG |
713 | } |
714 | ||
7c673cae FG |
715 | // walk through incrementals |
716 | MonitorDBStore::TransactionRef t; | |
717 | size_t tx_size = 0; | |
718 | while (version > osdmap.epoch) { | |
719 | bufferlist inc_bl; | |
720 | int err = get_version(osdmap.epoch+1, inc_bl); | |
11fdf7f2 TL |
721 | ceph_assert(err == 0); |
722 | ceph_assert(inc_bl.length()); | |
eafe8130 TL |
723 | // set priority cache manager levels if the osdmap is |
724 | // being populated for the first time. | |
725 | if (mon_memory_autotune && pcm == nullptr) { | |
726 | int r = register_cache_with_pcm(); | |
727 | if (r < 0) { | |
728 | dout(10) << __func__ | |
729 | << " Error while registering osdmon caches with pcm." | |
730 | << " Proceeding without cache auto tuning." | |
731 | << dendl; | |
732 | } | |
733 | } | |
7c673cae FG |
734 | |
735 | dout(7) << "update_from_paxos applying incremental " << osdmap.epoch+1 | |
736 | << dendl; | |
737 | OSDMap::Incremental inc(inc_bl); | |
738 | err = osdmap.apply_incremental(inc); | |
11fdf7f2 | 739 | ceph_assert(err == 0); |
7c673cae FG |
740 | |
741 | if (!t) | |
742 | t.reset(new MonitorDBStore::Transaction); | |
743 | ||
744 | // Write out the full map for all past epochs. Encode the full | |
745 | // map with the same features as the incremental. If we don't | |
746 | // know, use the quorum features. If we don't know those either, | |
747 | // encode with all features. | |
748 | uint64_t f = inc.encode_features; | |
749 | if (!f) | |
750 | f = mon->get_quorum_con_features(); | |
751 | if (!f) | |
752 | f = -1; | |
753 | bufferlist full_bl; | |
754 | osdmap.encode(full_bl, f | CEPH_FEATURE_RESERVED); | |
755 | tx_size += full_bl.length(); | |
756 | ||
757 | bufferlist orig_full_bl; | |
758 | get_version_full(osdmap.epoch, orig_full_bl); | |
759 | if (orig_full_bl.length()) { | |
760 | // the primary provided the full map | |
11fdf7f2 | 761 | ceph_assert(inc.have_crc); |
7c673cae FG |
762 | if (inc.full_crc != osdmap.crc) { |
763 | // This will happen if the mons were running mixed versions in | |
764 | // the past or some other circumstance made the full encoded | |
765 | // maps divergent. Reloading here will bring us back into | |
766 | // sync with the primary for this and all future maps. OSDs | |
767 | // will also be brought back into sync when they discover the | |
768 | // crc mismatch and request a full map from a mon. | |
769 | derr << __func__ << " full map CRC mismatch, resetting to canonical" | |
770 | << dendl; | |
11fdf7f2 TL |
771 | |
772 | dout(20) << __func__ << " my (bad) full osdmap:\n"; | |
773 | JSONFormatter jf(true); | |
774 | jf.dump_object("osdmap", osdmap); | |
775 | jf.flush(*_dout); | |
776 | *_dout << "\nhexdump:\n"; | |
777 | full_bl.hexdump(*_dout); | |
778 | *_dout << dendl; | |
779 | ||
7c673cae FG |
780 | osdmap = OSDMap(); |
781 | osdmap.decode(orig_full_bl); | |
11fdf7f2 TL |
782 | |
783 | dout(20) << __func__ << " canonical full osdmap:\n"; | |
784 | JSONFormatter jf(true); | |
785 | jf.dump_object("osdmap", osdmap); | |
786 | jf.flush(*_dout); | |
787 | *_dout << "\nhexdump:\n"; | |
788 | orig_full_bl.hexdump(*_dout); | |
789 | *_dout << dendl; | |
7c673cae FG |
790 | } |
791 | } else { | |
11fdf7f2 | 792 | ceph_assert(!inc.have_crc); |
7c673cae FG |
793 | put_version_full(t, osdmap.epoch, full_bl); |
794 | } | |
795 | put_version_latest_full(t, osdmap.epoch); | |
796 | ||
797 | // share | |
798 | dout(1) << osdmap << dendl; | |
799 | ||
800 | if (osdmap.epoch == 1) { | |
801 | t->erase("mkfs", "osdmap"); | |
802 | } | |
803 | ||
11fdf7f2 | 804 | if (tx_size > g_conf()->mon_sync_max_payload_size*2) { |
7c673cae FG |
805 | mon->store->apply_transaction(t); |
806 | t = MonitorDBStore::TransactionRef(); | |
807 | tx_size = 0; | |
808 | } | |
11fdf7f2 TL |
809 | for (const auto &osd_state : inc.new_state) { |
810 | if (osd_state.second & CEPH_OSD_UP) { | |
811 | // could be marked up *or* down, but we're too lazy to check which | |
812 | last_osd_report.erase(osd_state.first); | |
813 | } | |
814 | if (osd_state.second & CEPH_OSD_EXISTS) { | |
815 | // could be created *or* destroyed, but we can safely drop it | |
816 | osd_epochs.erase(osd_state.first); | |
7c673cae FG |
817 | } |
818 | } | |
819 | } | |
820 | ||
821 | if (t) { | |
822 | mon->store->apply_transaction(t); | |
823 | } | |
824 | ||
825 | for (int o = 0; o < osdmap.get_max_osd(); o++) { | |
826 | if (osdmap.is_out(o)) | |
827 | continue; | |
828 | auto found = down_pending_out.find(o); | |
829 | if (osdmap.is_down(o)) { | |
830 | // populate down -> out map | |
831 | if (found == down_pending_out.end()) { | |
832 | dout(10) << " adding osd." << o << " to down_pending_out map" << dendl; | |
833 | down_pending_out[o] = ceph_clock_now(); | |
834 | } | |
835 | } else { | |
836 | if (found != down_pending_out.end()) { | |
837 | dout(10) << " removing osd." << o << " from down_pending_out map" << dendl; | |
838 | down_pending_out.erase(found); | |
839 | } | |
840 | } | |
841 | } | |
842 | // XXX: need to trim MonSession connected with a osd whose id > max_osd? | |
843 | ||
7c673cae FG |
844 | check_osdmap_subs(); |
845 | check_pg_creates_subs(); | |
846 | ||
847 | share_map_with_random_osd(); | |
848 | update_logger(); | |
7c673cae FG |
849 | process_failures(); |
850 | ||
851 | // make sure our feature bits reflect the latest map | |
852 | update_msgr_features(); | |
853 | ||
854 | if (!mon->is_leader()) { | |
855 | // will be called by on_active() on the leader, avoid doing so twice | |
856 | start_mapping(); | |
857 | } | |
858 | } | |
859 | ||
eafe8130 TL |
860 | int OSDMonitor::register_cache_with_pcm() |
861 | { | |
862 | if (mon_memory_target <= 0 || mon_memory_min <= 0) { | |
863 | derr << __func__ << " Invalid memory size specified for mon caches." | |
864 | << " Caches will not be auto-tuned." | |
865 | << dendl; | |
866 | return -EINVAL; | |
867 | } | |
868 | uint64_t base = mon_memory_base; | |
869 | double fragmentation = mon_memory_fragmentation; | |
870 | // For calculating total target memory, consider rocksdb cache size. | |
871 | uint64_t target = mon_memory_target; | |
872 | uint64_t min = mon_memory_min; | |
873 | uint64_t max = min; | |
874 | ||
875 | // Apply the same logic as in bluestore to set the max amount | |
876 | // of memory to use for cache. Assume base memory for OSDMaps | |
877 | // and then add in some overhead for fragmentation. | |
878 | uint64_t ltarget = (1.0 - fragmentation) * target; | |
879 | if (ltarget > base + min) { | |
880 | max = ltarget - base; | |
881 | } | |
882 | ||
883 | rocksdb_binned_kv_cache = mon->store->get_priority_cache(); | |
884 | if (!rocksdb_binned_kv_cache) { | |
885 | derr << __func__ << " not using rocksdb" << dendl; | |
886 | return -EINVAL; | |
887 | } | |
888 | ||
889 | int r = _set_cache_ratios(); | |
890 | if (r < 0) { | |
891 | derr << __func__ << " Cache ratios for pcm could not be set." | |
892 | << " Review the kv (rocksdb) and mon_memory_target sizes." | |
893 | << dendl; | |
894 | return -EINVAL; | |
895 | } | |
896 | ||
897 | pcm = std::make_shared<PriorityCache::Manager>( | |
898 | cct, min, max, target, true); | |
899 | pcm->insert("kv", rocksdb_binned_kv_cache, true); | |
900 | pcm->insert("inc", inc_cache, true); | |
901 | pcm->insert("full", full_cache, true); | |
902 | dout(10) << __func__ << " pcm target: " << target | |
903 | << " pcm max: " << max | |
904 | << " pcm min: " << min | |
905 | << " inc_osd_cache size: " << inc_osd_cache.get_size() | |
906 | << dendl; | |
907 | return 0; | |
908 | } | |
909 | ||
910 | int OSDMonitor::_set_cache_ratios() | |
911 | { | |
912 | double old_cache_kv_ratio = cache_kv_ratio; | |
913 | ||
914 | // Set the cache ratios for kv(rocksdb), inc and full caches | |
915 | cache_kv_ratio = (double)rocksdb_cache_size / (double)mon_memory_target; | |
916 | if (cache_kv_ratio >= 1.0) { | |
917 | derr << __func__ << " Cache kv ratio (" << cache_kv_ratio | |
918 | << ") must be in range [0,<1.0]." | |
919 | << dendl; | |
920 | cache_kv_ratio = old_cache_kv_ratio; | |
921 | return -EINVAL; | |
922 | } | |
923 | rocksdb_binned_kv_cache->set_cache_ratio(cache_kv_ratio); | |
924 | cache_inc_ratio = cache_full_ratio = (1.0 - cache_kv_ratio) / 2; | |
925 | inc_cache->set_cache_ratio(cache_inc_ratio); | |
926 | full_cache->set_cache_ratio(cache_full_ratio); | |
927 | ||
928 | dout(10) << __func__ << " kv ratio " << cache_kv_ratio | |
929 | << " inc ratio " << cache_inc_ratio | |
930 | << " full ratio " << cache_full_ratio | |
931 | << dendl; | |
932 | return 0; | |
933 | } | |
934 | ||
7c673cae FG |
935 | void OSDMonitor::start_mapping() |
936 | { | |
937 | // initiate mapping job | |
938 | if (mapping_job) { | |
939 | dout(10) << __func__ << " canceling previous mapping_job " << mapping_job.get() | |
940 | << dendl; | |
941 | mapping_job->abort(); | |
942 | } | |
224ce89b WB |
943 | if (!osdmap.get_pools().empty()) { |
944 | auto fin = new C_UpdateCreatingPGs(this, osdmap.get_epoch()); | |
945 | mapping_job = mapping.start_update(osdmap, mapper, | |
11fdf7f2 | 946 | g_conf()->mon_osd_mapping_pgs_per_chunk); |
224ce89b WB |
947 | dout(10) << __func__ << " started mapping job " << mapping_job.get() |
948 | << " at " << fin->start << dendl; | |
949 | mapping_job->set_finish_event(fin); | |
950 | } else { | |
951 | dout(10) << __func__ << " no pools, no mapping job" << dendl; | |
952 | mapping_job = nullptr; | |
953 | } | |
7c673cae FG |
954 | } |
955 | ||
956 | void OSDMonitor::update_msgr_features() | |
957 | { | |
958 | set<int> types; | |
959 | types.insert((int)entity_name_t::TYPE_OSD); | |
960 | types.insert((int)entity_name_t::TYPE_CLIENT); | |
961 | types.insert((int)entity_name_t::TYPE_MDS); | |
962 | types.insert((int)entity_name_t::TYPE_MON); | |
963 | for (set<int>::iterator q = types.begin(); q != types.end(); ++q) { | |
964 | uint64_t mask; | |
965 | uint64_t features = osdmap.get_features(*q, &mask); | |
966 | if ((mon->messenger->get_policy(*q).features_required & mask) != features) { | |
967 | dout(0) << "crush map has features " << features << ", adjusting msgr requires" << dendl; | |
11fdf7f2 | 968 | ceph::net::Policy p = mon->messenger->get_policy(*q); |
7c673cae FG |
969 | p.features_required = (p.features_required & ~mask) | features; |
970 | mon->messenger->set_policy(*q, p); | |
971 | } | |
972 | } | |
973 | } | |
974 | ||
975 | void OSDMonitor::on_active() | |
976 | { | |
977 | update_logger(); | |
978 | ||
979 | if (mon->is_leader()) { | |
224ce89b | 980 | mon->clog->debug() << "osdmap " << osdmap; |
81eedcae TL |
981 | if (!priority_convert) { |
982 | // Only do this once at start-up | |
983 | convert_pool_priorities(); | |
984 | priority_convert = true; | |
985 | } | |
7c673cae FG |
986 | } else { |
987 | list<MonOpRequestRef> ls; | |
988 | take_all_failures(ls); | |
989 | while (!ls.empty()) { | |
990 | MonOpRequestRef op = ls.front(); | |
991 | op->mark_osdmon_event(__func__); | |
992 | dispatch(op); | |
993 | ls.pop_front(); | |
994 | } | |
995 | } | |
996 | start_mapping(); | |
997 | } | |
998 | ||
999 | void OSDMonitor::on_restart() | |
1000 | { | |
1001 | last_osd_report.clear(); | |
1002 | } | |
1003 | ||
1004 | void OSDMonitor::on_shutdown() | |
1005 | { | |
1006 | dout(10) << __func__ << dendl; | |
1007 | if (mapping_job) { | |
1008 | dout(10) << __func__ << " canceling previous mapping_job " << mapping_job.get() | |
1009 | << dendl; | |
1010 | mapping_job->abort(); | |
1011 | } | |
1012 | ||
1013 | // discard failure info, waiters | |
1014 | list<MonOpRequestRef> ls; | |
1015 | take_all_failures(ls); | |
1016 | ls.clear(); | |
1017 | } | |
1018 | ||
1019 | void OSDMonitor::update_logger() | |
1020 | { | |
1021 | dout(10) << "update_logger" << dendl; | |
1022 | ||
1023 | mon->cluster_logger->set(l_cluster_num_osd, osdmap.get_num_osds()); | |
1024 | mon->cluster_logger->set(l_cluster_num_osd_up, osdmap.get_num_up_osds()); | |
1025 | mon->cluster_logger->set(l_cluster_num_osd_in, osdmap.get_num_in_osds()); | |
1026 | mon->cluster_logger->set(l_cluster_osd_epoch, osdmap.get_epoch()); | |
1027 | } | |
1028 | ||
7c673cae FG |
1029 | void OSDMonitor::create_pending() |
1030 | { | |
1031 | pending_inc = OSDMap::Incremental(osdmap.epoch+1); | |
1032 | pending_inc.fsid = mon->monmap->fsid; | |
11fdf7f2 TL |
1033 | pending_metadata.clear(); |
1034 | pending_metadata_rm.clear(); | |
7c673cae FG |
1035 | |
1036 | dout(10) << "create_pending e " << pending_inc.epoch << dendl; | |
1037 | ||
11fdf7f2 TL |
1038 | // safety checks (this shouldn't really happen) |
1039 | { | |
1040 | if (osdmap.backfillfull_ratio <= 0) { | |
1041 | pending_inc.new_backfillfull_ratio = g_conf()->mon_osd_backfillfull_ratio; | |
1042 | if (pending_inc.new_backfillfull_ratio > 1.0) | |
1043 | pending_inc.new_backfillfull_ratio /= 100; | |
1044 | dout(1) << __func__ << " setting backfillfull_ratio = " | |
1045 | << pending_inc.new_backfillfull_ratio << dendl; | |
7c673cae | 1046 | } |
7c673cae | 1047 | if (osdmap.full_ratio <= 0) { |
11fdf7f2 | 1048 | pending_inc.new_full_ratio = g_conf()->mon_osd_full_ratio; |
7c673cae FG |
1049 | if (pending_inc.new_full_ratio > 1.0) |
1050 | pending_inc.new_full_ratio /= 100; | |
1051 | dout(1) << __func__ << " setting full_ratio = " | |
1052 | << pending_inc.new_full_ratio << dendl; | |
1053 | } | |
1054 | if (osdmap.nearfull_ratio <= 0) { | |
11fdf7f2 | 1055 | pending_inc.new_nearfull_ratio = g_conf()->mon_osd_nearfull_ratio; |
7c673cae FG |
1056 | if (pending_inc.new_nearfull_ratio > 1.0) |
1057 | pending_inc.new_nearfull_ratio /= 100; | |
1058 | dout(1) << __func__ << " setting nearfull_ratio = " | |
1059 | << pending_inc.new_nearfull_ratio << dendl; | |
1060 | } | |
1061 | } | |
3efd9988 FG |
1062 | |
1063 | // Rewrite CRUSH rule IDs if they are using legacy "ruleset" | |
1064 | // structure. | |
1065 | if (osdmap.crush->has_legacy_rule_ids()) { | |
1066 | CrushWrapper newcrush; | |
1067 | _get_pending_crush(newcrush); | |
1068 | ||
1069 | // First, for all pools, work out which rule they really used | |
1070 | // by resolving ruleset to rule. | |
1071 | for (const auto &i : osdmap.get_pools()) { | |
1072 | const auto pool_id = i.first; | |
1073 | const auto &pool = i.second; | |
1074 | int new_rule_id = newcrush.find_rule(pool.crush_rule, | |
1075 | pool.type, pool.size); | |
1076 | ||
1077 | dout(1) << __func__ << " rewriting pool " | |
1078 | << osdmap.get_pool_name(pool_id) << " crush ruleset " | |
1079 | << pool.crush_rule << " -> rule id " << new_rule_id << dendl; | |
1080 | if (pending_inc.new_pools.count(pool_id) == 0) { | |
1081 | pending_inc.new_pools[pool_id] = pool; | |
1082 | } | |
1083 | pending_inc.new_pools[pool_id].crush_rule = new_rule_id; | |
1084 | } | |
1085 | ||
1086 | // Now, go ahead and renumber all the rules so that their | |
1087 | // rule_id field corresponds to their position in the array | |
1088 | auto old_to_new = newcrush.renumber_rules(); | |
1089 | dout(1) << __func__ << " Rewrote " << old_to_new << " crush IDs:" << dendl; | |
1090 | for (const auto &i : old_to_new) { | |
1091 | dout(1) << __func__ << " " << i.first << " -> " << i.second << dendl; | |
1092 | } | |
1093 | pending_inc.crush.clear(); | |
1094 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
1095 | } | |
7c673cae FG |
1096 | } |
1097 | ||
1098 | creating_pgs_t | |
94b18763 FG |
1099 | OSDMonitor::update_pending_pgs(const OSDMap::Incremental& inc, |
1100 | const OSDMap& nextmap) | |
7c673cae | 1101 | { |
31f18b77 | 1102 | dout(10) << __func__ << dendl; |
7c673cae FG |
1103 | creating_pgs_t pending_creatings; |
1104 | { | |
1105 | std::lock_guard<std::mutex> l(creating_pgs_lock); | |
1106 | pending_creatings = creating_pgs; | |
1107 | } | |
31f18b77 FG |
1108 | // check for new or old pools |
1109 | if (pending_creatings.last_scan_epoch < inc.epoch) { | |
31f18b77 FG |
1110 | unsigned queued = 0; |
1111 | queued += scan_for_creating_pgs(osdmap.get_pools(), | |
1112 | inc.old_pools, | |
1113 | inc.modified, | |
1114 | &pending_creatings); | |
1115 | queued += scan_for_creating_pgs(inc.new_pools, | |
1116 | inc.old_pools, | |
1117 | inc.modified, | |
1118 | &pending_creatings); | |
1119 | dout(10) << __func__ << " " << queued << " pools queued" << dendl; | |
1120 | for (auto deleted_pool : inc.old_pools) { | |
1121 | auto removed = pending_creatings.remove_pool(deleted_pool); | |
1122 | dout(10) << __func__ << " " << removed | |
1123 | << " pg removed because containing pool deleted: " | |
1124 | << deleted_pool << dendl; | |
1125 | last_epoch_clean.remove_pool(deleted_pool); | |
1126 | } | |
1127 | // pgmon updates its creating_pgs in check_osd_map() which is called by | |
1128 | // on_active() and check_osd_map() could be delayed if lease expires, so its | |
1129 | // creating_pgs could be stale in comparison with the one of osdmon. let's | |
1130 | // trim them here. otherwise, they will be added back after being erased. | |
1131 | unsigned removed = 0; | |
1132 | for (auto& pg : pending_created_pgs) { | |
1133 | dout(20) << __func__ << " noting created pg " << pg << dendl; | |
1134 | pending_creatings.created_pools.insert(pg.pool()); | |
1135 | removed += pending_creatings.pgs.erase(pg); | |
1136 | } | |
1137 | pending_created_pgs.clear(); | |
1138 | dout(10) << __func__ << " " << removed | |
1139 | << " pgs removed because they're created" << dendl; | |
1140 | pending_creatings.last_scan_epoch = osdmap.get_epoch(); | |
1141 | } | |
1142 | ||
94b18763 FG |
1143 | // filter out any pgs that shouldn't exist. |
1144 | { | |
1145 | auto i = pending_creatings.pgs.begin(); | |
1146 | while (i != pending_creatings.pgs.end()) { | |
1147 | if (!nextmap.pg_exists(i->first)) { | |
1148 | dout(10) << __func__ << " removing pg " << i->first | |
1149 | << " which should not exist" << dendl; | |
1150 | i = pending_creatings.pgs.erase(i); | |
1151 | } else { | |
1152 | ++i; | |
1153 | } | |
1154 | } | |
1155 | } | |
1156 | ||
31f18b77 | 1157 | // process queue |
11fdf7f2 | 1158 | unsigned max = std::max<int64_t>(1, g_conf()->mon_osd_max_creating_pgs); |
31f18b77 FG |
1159 | const auto total = pending_creatings.pgs.size(); |
1160 | while (pending_creatings.pgs.size() < max && | |
1161 | !pending_creatings.queue.empty()) { | |
1162 | auto p = pending_creatings.queue.begin(); | |
1163 | int64_t poolid = p->first; | |
1164 | dout(10) << __func__ << " pool " << poolid | |
1165 | << " created " << p->second.created | |
1166 | << " modified " << p->second.modified | |
1167 | << " [" << p->second.start << "-" << p->second.end << ")" | |
1168 | << dendl; | |
11fdf7f2 TL |
1169 | int64_t n = std::min<int64_t>(max - pending_creatings.pgs.size(), |
1170 | p->second.end - p->second.start); | |
31f18b77 FG |
1171 | ps_t first = p->second.start; |
1172 | ps_t end = first + n; | |
1173 | for (ps_t ps = first; ps < end; ++ps) { | |
1174 | const pg_t pgid{ps, static_cast<uint64_t>(poolid)}; | |
1175 | // NOTE: use the *current* epoch as the PG creation epoch so that the | |
1176 | // OSD does not have to generate a long set of PastIntervals. | |
1177 | pending_creatings.pgs.emplace(pgid, make_pair(inc.epoch, | |
1178 | p->second.modified)); | |
1179 | dout(10) << __func__ << " adding " << pgid << dendl; | |
1180 | } | |
1181 | p->second.start = end; | |
1182 | if (p->second.done()) { | |
1183 | dout(10) << __func__ << " done with queue for " << poolid << dendl; | |
1184 | pending_creatings.queue.erase(p); | |
1185 | } else { | |
1186 | dout(10) << __func__ << " pool " << poolid | |
1187 | << " now [" << p->second.start << "-" << p->second.end << ")" | |
1188 | << dendl; | |
1189 | } | |
1190 | } | |
1191 | dout(10) << __func__ << " queue remaining: " << pending_creatings.queue.size() | |
1192 | << " pools" << dendl; | |
c07f9fc5 FG |
1193 | dout(10) << __func__ |
1194 | << " " << (pending_creatings.pgs.size() - total) | |
1195 | << "/" << pending_creatings.pgs.size() | |
31f18b77 | 1196 | << " pgs added from queued pools" << dendl; |
7c673cae FG |
1197 | return pending_creatings; |
1198 | } | |
1199 | ||
1200 | void OSDMonitor::maybe_prime_pg_temp() | |
1201 | { | |
1202 | bool all = false; | |
1203 | if (pending_inc.crush.length()) { | |
1204 | dout(10) << __func__ << " new crush map, all" << dendl; | |
1205 | all = true; | |
1206 | } | |
1207 | ||
1208 | if (!pending_inc.new_up_client.empty()) { | |
1209 | dout(10) << __func__ << " new up osds, all" << dendl; | |
1210 | all = true; | |
1211 | } | |
1212 | ||
1213 | // check for interesting OSDs | |
1214 | set<int> osds; | |
31f18b77 | 1215 | for (auto p = pending_inc.new_state.begin(); |
7c673cae FG |
1216 | !all && p != pending_inc.new_state.end(); |
1217 | ++p) { | |
1218 | if ((p->second & CEPH_OSD_UP) && | |
1219 | osdmap.is_up(p->first)) { | |
1220 | osds.insert(p->first); | |
1221 | } | |
1222 | } | |
1223 | for (map<int32_t,uint32_t>::iterator p = pending_inc.new_weight.begin(); | |
1224 | !all && p != pending_inc.new_weight.end(); | |
1225 | ++p) { | |
1226 | if (p->second < osdmap.get_weight(p->first)) { | |
1227 | // weight reduction | |
1228 | osds.insert(p->first); | |
1229 | } else { | |
1230 | dout(10) << __func__ << " osd." << p->first << " weight increase, all" | |
1231 | << dendl; | |
1232 | all = true; | |
1233 | } | |
1234 | } | |
1235 | ||
1236 | if (!all && osds.empty()) | |
1237 | return; | |
1238 | ||
1239 | if (!all) { | |
1240 | unsigned estimate = | |
1241 | mapping.get_osd_acting_pgs(*osds.begin()).size() * osds.size(); | |
1242 | if (estimate > mapping.get_num_pgs() * | |
11fdf7f2 | 1243 | g_conf()->mon_osd_prime_pg_temp_max_estimate) { |
7c673cae FG |
1244 | dout(10) << __func__ << " estimate " << estimate << " pgs on " |
1245 | << osds.size() << " osds >= " | |
11fdf7f2 | 1246 | << g_conf()->mon_osd_prime_pg_temp_max_estimate << " of total " |
7c673cae FG |
1247 | << mapping.get_num_pgs() << " pgs, all" |
1248 | << dendl; | |
1249 | all = true; | |
1250 | } else { | |
1251 | dout(10) << __func__ << " estimate " << estimate << " pgs on " | |
1252 | << osds.size() << " osds" << dendl; | |
1253 | } | |
1254 | } | |
1255 | ||
1256 | OSDMap next; | |
1257 | next.deepish_copy_from(osdmap); | |
1258 | next.apply_incremental(pending_inc); | |
1259 | ||
224ce89b WB |
1260 | if (next.get_pools().empty()) { |
1261 | dout(10) << __func__ << " no pools, no pg_temp priming" << dendl; | |
1262 | } else if (all) { | |
7c673cae | 1263 | PrimeTempJob job(next, this); |
494da23a | 1264 | mapper.queue(&job, g_conf()->mon_osd_mapping_pgs_per_chunk, {}); |
11fdf7f2 | 1265 | if (job.wait_for(g_conf()->mon_osd_prime_pg_temp_max_time)) { |
7c673cae FG |
1266 | dout(10) << __func__ << " done in " << job.get_duration() << dendl; |
1267 | } else { | |
1268 | dout(10) << __func__ << " did not finish in " | |
11fdf7f2 | 1269 | << g_conf()->mon_osd_prime_pg_temp_max_time |
7c673cae FG |
1270 | << ", stopping" << dendl; |
1271 | job.abort(); | |
1272 | } | |
1273 | } else { | |
1274 | dout(10) << __func__ << " " << osds.size() << " interesting osds" << dendl; | |
1275 | utime_t stop = ceph_clock_now(); | |
11fdf7f2 | 1276 | stop += g_conf()->mon_osd_prime_pg_temp_max_time; |
7c673cae FG |
1277 | const int chunk = 1000; |
1278 | int n = chunk; | |
1279 | std::unordered_set<pg_t> did_pgs; | |
1280 | for (auto osd : osds) { | |
1281 | auto& pgs = mapping.get_osd_acting_pgs(osd); | |
1282 | dout(20) << __func__ << " osd." << osd << " " << pgs << dendl; | |
1283 | for (auto pgid : pgs) { | |
1284 | if (!did_pgs.insert(pgid).second) { | |
1285 | continue; | |
1286 | } | |
1287 | prime_pg_temp(next, pgid); | |
1288 | if (--n <= 0) { | |
1289 | n = chunk; | |
1290 | if (ceph_clock_now() > stop) { | |
1291 | dout(10) << __func__ << " consumed more than " | |
11fdf7f2 | 1292 | << g_conf()->mon_osd_prime_pg_temp_max_time |
7c673cae FG |
1293 | << " seconds, stopping" |
1294 | << dendl; | |
1295 | return; | |
1296 | } | |
1297 | } | |
1298 | } | |
1299 | } | |
1300 | } | |
1301 | } | |
1302 | ||
1303 | void OSDMonitor::prime_pg_temp( | |
1304 | const OSDMap& next, | |
1305 | pg_t pgid) | |
1306 | { | |
11fdf7f2 TL |
1307 | // TODO: remove this creating_pgs direct access? |
1308 | if (creating_pgs.pgs.count(pgid)) { | |
1309 | return; | |
7c673cae FG |
1310 | } |
1311 | if (!osdmap.pg_exists(pgid)) { | |
1312 | return; | |
1313 | } | |
1314 | ||
1315 | vector<int> up, acting; | |
1316 | mapping.get(pgid, &up, nullptr, &acting, nullptr); | |
1317 | ||
1318 | vector<int> next_up, next_acting; | |
1319 | int next_up_primary, next_acting_primary; | |
1320 | next.pg_to_up_acting_osds(pgid, &next_up, &next_up_primary, | |
1321 | &next_acting, &next_acting_primary); | |
f64942e4 AA |
1322 | if (acting == next_acting && |
1323 | !(up != acting && next_up == next_acting)) | |
7c673cae FG |
1324 | return; // no change since last epoch |
1325 | ||
1326 | if (acting.empty()) | |
1327 | return; // if previously empty now we can be no worse off | |
1328 | const pg_pool_t *pool = next.get_pg_pool(pgid.pool()); | |
1329 | if (pool && acting.size() < pool->min_size) | |
1330 | return; // can be no worse off than before | |
1331 | ||
c07f9fc5 FG |
1332 | if (next_up == next_acting) { |
1333 | acting.clear(); | |
11fdf7f2 TL |
1334 | dout(20) << __func__ << " next_up == next_acting now, clear pg_temp" |
1335 | << dendl; | |
c07f9fc5 FG |
1336 | } |
1337 | ||
7c673cae FG |
1338 | dout(20) << __func__ << " " << pgid << " " << up << "/" << acting |
1339 | << " -> " << next_up << "/" << next_acting | |
1340 | << ", priming " << acting | |
1341 | << dendl; | |
1342 | { | |
11fdf7f2 | 1343 | std::lock_guard l(prime_pg_temp_lock); |
7c673cae FG |
1344 | // do not touch a mapping if a change is pending |
1345 | pending_inc.new_pg_temp.emplace( | |
1346 | pgid, | |
1347 | mempool::osdmap::vector<int>(acting.begin(), acting.end())); | |
1348 | } | |
1349 | } | |
1350 | ||
1351 | /** | |
1352 | * @note receiving a transaction in this function gives a fair amount of | |
1353 | * freedom to the service implementation if it does need it. It shouldn't. | |
1354 | */ | |
1355 | void OSDMonitor::encode_pending(MonitorDBStore::TransactionRef t) | |
1356 | { | |
1357 | dout(10) << "encode_pending e " << pending_inc.epoch | |
1358 | << dendl; | |
1359 | ||
11fdf7f2 TL |
1360 | if (do_prune(t)) { |
1361 | dout(1) << __func__ << " osdmap full prune encoded e" | |
1362 | << pending_inc.epoch << dendl; | |
1363 | } | |
1364 | ||
7c673cae FG |
1365 | // finalize up pending_inc |
1366 | pending_inc.modified = ceph_clock_now(); | |
1367 | ||
11fdf7f2 TL |
1368 | int r = pending_inc.propagate_snaps_to_tiers(cct, osdmap); |
1369 | ceph_assert(r == 0); | |
7c673cae FG |
1370 | |
1371 | if (mapping_job) { | |
1372 | if (!mapping_job->is_done()) { | |
1373 | dout(1) << __func__ << " skipping prime_pg_temp; mapping job " | |
1374 | << mapping_job.get() << " did not complete, " | |
1375 | << mapping_job->shards << " left" << dendl; | |
1376 | mapping_job->abort(); | |
1377 | } else if (mapping.get_epoch() < osdmap.get_epoch()) { | |
1378 | dout(1) << __func__ << " skipping prime_pg_temp; mapping job " | |
1379 | << mapping_job.get() << " is prior epoch " | |
1380 | << mapping.get_epoch() << dendl; | |
1381 | } else { | |
11fdf7f2 | 1382 | if (g_conf()->mon_osd_prime_pg_temp) { |
7c673cae FG |
1383 | maybe_prime_pg_temp(); |
1384 | } | |
1385 | } | |
11fdf7f2 | 1386 | } else if (g_conf()->mon_osd_prime_pg_temp) { |
7c673cae FG |
1387 | dout(1) << __func__ << " skipping prime_pg_temp; mapping job did not start" |
1388 | << dendl; | |
1389 | } | |
1390 | mapping_job.reset(); | |
1391 | ||
c07f9fc5 FG |
1392 | // ensure we don't have blank new_state updates. these are interrpeted as |
1393 | // CEPH_OSD_UP (and almost certainly not what we want!). | |
1394 | auto p = pending_inc.new_state.begin(); | |
1395 | while (p != pending_inc.new_state.end()) { | |
1396 | if (p->second == 0) { | |
1397 | dout(10) << "new_state for osd." << p->first << " is 0, removing" << dendl; | |
1398 | p = pending_inc.new_state.erase(p); | |
1399 | } else { | |
11fdf7f2 TL |
1400 | if (p->second & CEPH_OSD_UP) { |
1401 | pending_inc.new_last_up_change = pending_inc.modified; | |
1402 | } | |
c07f9fc5 FG |
1403 | ++p; |
1404 | } | |
1405 | } | |
11fdf7f2 TL |
1406 | if (!pending_inc.new_up_client.empty()) { |
1407 | pending_inc.new_last_up_change = pending_inc.modified; | |
1408 | } | |
1409 | for (auto& i : pending_inc.new_weight) { | |
1410 | if (i.first > osdmap.max_osd) { | |
1411 | if (i.second) { | |
1412 | // new osd is already marked in | |
1413 | pending_inc.new_last_in_change = pending_inc.modified; | |
1414 | } | |
1415 | } else if (!!i.second != !!osdmap.osd_weight[i.first]) { | |
1416 | // existing osd marked in or out | |
1417 | pending_inc.new_last_in_change = pending_inc.modified; | |
1418 | } | |
1419 | } | |
7c673cae FG |
1420 | |
1421 | { | |
1422 | OSDMap tmp; | |
1423 | tmp.deepish_copy_from(osdmap); | |
1424 | tmp.apply_incremental(pending_inc); | |
1425 | ||
11fdf7f2 TL |
1426 | // clean pg_temp mappings |
1427 | OSDMap::clean_temps(cct, osdmap, tmp, &pending_inc); | |
1428 | ||
1429 | // clean inappropriate pg_upmap/pg_upmap_items (if any) | |
494da23a TL |
1430 | { |
1431 | // check every upmapped pg for now | |
1432 | // until we could reliably identify certain cases to ignore, | |
1433 | // which is obviously the hard part TBD.. | |
1434 | vector<pg_t> pgs_to_check; | |
1435 | tmp.get_upmap_pgs(&pgs_to_check); | |
1436 | if (pgs_to_check.size() < g_conf()->mon_clean_pg_upmaps_per_chunk * 2) { | |
1437 | // not enough pgs, do it inline | |
1438 | tmp.clean_pg_upmaps(cct, &pending_inc); | |
1439 | } else { | |
1440 | CleanUpmapJob job(cct, tmp, pending_inc); | |
1441 | mapper.queue(&job, g_conf()->mon_clean_pg_upmaps_per_chunk, pgs_to_check); | |
1442 | job.wait(); | |
1443 | } | |
1444 | } | |
11fdf7f2 TL |
1445 | |
1446 | // update creating pgs first so that we can remove the created pgid and | |
1447 | // process the pool flag removal below in the same osdmap epoch. | |
1448 | auto pending_creatings = update_pending_pgs(pending_inc, tmp); | |
1449 | bufferlist creatings_bl; | |
1450 | encode(pending_creatings, creatings_bl); | |
1451 | t->put(OSD_PG_CREATING_PREFIX, "creating", creatings_bl); | |
1452 | ||
1453 | // remove any old (or incompat) POOL_CREATING flags | |
1454 | for (auto& i : tmp.get_pools()) { | |
1455 | if (tmp.require_osd_release < CEPH_RELEASE_NAUTILUS) { | |
1456 | // pre-nautilus OSDMaps shouldn't get this flag. | |
1457 | if (pending_inc.new_pools.count(i.first)) { | |
1458 | pending_inc.new_pools[i.first].flags &= ~pg_pool_t::FLAG_CREATING; | |
1459 | } | |
1460 | } | |
1461 | if (i.second.has_flag(pg_pool_t::FLAG_CREATING) && | |
1462 | !pending_creatings.still_creating_pool(i.first)) { | |
1463 | dout(10) << __func__ << " done creating pool " << i.first | |
1464 | << ", clearing CREATING flag" << dendl; | |
1465 | if (pending_inc.new_pools.count(i.first) == 0) { | |
1466 | pending_inc.new_pools[i.first] = i.second; | |
1467 | } | |
1468 | pending_inc.new_pools[i.first].flags &= ~pg_pool_t::FLAG_CREATING; | |
3efd9988 | 1469 | } |
11fdf7f2 TL |
1470 | } |
1471 | ||
1472 | // remove any legacy osdmap nearfull/full flags | |
1473 | { | |
1474 | if (tmp.test_flag(CEPH_OSDMAP_FULL | CEPH_OSDMAP_NEARFULL)) { | |
1475 | dout(10) << __func__ << " clearing legacy osdmap nearfull/full flag" | |
1476 | << dendl; | |
1477 | remove_flag(CEPH_OSDMAP_NEARFULL); | |
1478 | remove_flag(CEPH_OSDMAP_FULL); | |
1479 | } | |
1480 | } | |
1481 | // collect which pools are currently affected by | |
1482 | // the near/backfill/full osd(s), | |
1483 | // and set per-pool near/backfill/full flag instead | |
1484 | set<int64_t> full_pool_ids; | |
1485 | set<int64_t> backfillfull_pool_ids; | |
1486 | set<int64_t> nearfull_pool_ids; | |
1487 | tmp.get_full_pools(cct, | |
1488 | &full_pool_ids, | |
1489 | &backfillfull_pool_ids, | |
3efd9988 | 1490 | &nearfull_pool_ids); |
11fdf7f2 TL |
1491 | if (full_pool_ids.empty() || |
1492 | backfillfull_pool_ids.empty() || | |
1493 | nearfull_pool_ids.empty()) { | |
1494 | // normal case - no nearfull, backfillfull or full osds | |
3efd9988 FG |
1495 | // try cancel any improper nearfull/backfillfull/full pool |
1496 | // flags first | |
11fdf7f2 TL |
1497 | for (auto &pool: tmp.get_pools()) { |
1498 | auto p = pool.first; | |
1499 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_NEARFULL) && | |
1500 | nearfull_pool_ids.empty()) { | |
1501 | dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p] | |
1502 | << "'s nearfull flag" << dendl; | |
1503 | if (pending_inc.new_pools.count(p) == 0) { | |
1504 | // load original pool info first! | |
1505 | pending_inc.new_pools[p] = pool.second; | |
1506 | } | |
1507 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL; | |
1508 | } | |
1509 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_BACKFILLFULL) && | |
1510 | backfillfull_pool_ids.empty()) { | |
1511 | dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p] | |
1512 | << "'s backfillfull flag" << dendl; | |
1513 | if (pending_inc.new_pools.count(p) == 0) { | |
1514 | pending_inc.new_pools[p] = pool.second; | |
1515 | } | |
1516 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_BACKFILLFULL; | |
1517 | } | |
1518 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL) && | |
1519 | full_pool_ids.empty()) { | |
1520 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_QUOTA)) { | |
1521 | // set by EQUOTA, skipping | |
1522 | continue; | |
1523 | } | |
1524 | dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p] | |
1525 | << "'s full flag" << dendl; | |
1526 | if (pending_inc.new_pools.count(p) == 0) { | |
1527 | pending_inc.new_pools[p] = pool.second; | |
1528 | } | |
1529 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_FULL; | |
1530 | } | |
3efd9988 | 1531 | } |
11fdf7f2 TL |
1532 | } |
1533 | if (!full_pool_ids.empty()) { | |
1534 | dout(10) << __func__ << " marking pool(s) " << full_pool_ids | |
1535 | << " as full" << dendl; | |
1536 | for (auto &p: full_pool_ids) { | |
1537 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL)) { | |
1538 | continue; | |
1539 | } | |
1540 | if (pending_inc.new_pools.count(p) == 0) { | |
1541 | pending_inc.new_pools[p] = tmp.pools[p]; | |
1542 | } | |
1543 | pending_inc.new_pools[p].flags |= pg_pool_t::FLAG_FULL; | |
1544 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_BACKFILLFULL; | |
1545 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL; | |
1546 | } | |
1547 | // cancel FLAG_FULL for pools which are no longer full too | |
1548 | for (auto &pool: tmp.get_pools()) { | |
1549 | auto p = pool.first; | |
1550 | if (full_pool_ids.count(p)) { | |
1551 | // skip pools we have just marked as full above | |
1552 | continue; | |
1553 | } | |
1554 | if (!tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL) || | |
1555 | tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_QUOTA)) { | |
1556 | // don't touch if currently is not full | |
1557 | // or is running out of quota (and hence considered as full) | |
1558 | continue; | |
1559 | } | |
1560 | dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p] | |
1561 | << "'s full flag" << dendl; | |
1562 | if (pending_inc.new_pools.count(p) == 0) { | |
1563 | pending_inc.new_pools[p] = pool.second; | |
1564 | } | |
1565 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_FULL; | |
3efd9988 | 1566 | } |
11fdf7f2 TL |
1567 | } |
1568 | if (!backfillfull_pool_ids.empty()) { | |
1569 | for (auto &p: backfillfull_pool_ids) { | |
1570 | if (full_pool_ids.count(p)) { | |
1571 | // skip pools we have already considered as full above | |
1572 | continue; | |
1573 | } | |
1574 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_QUOTA)) { | |
1575 | // make sure FLAG_FULL is truly set, so we are safe not | |
1576 | // to set a extra (redundant) FLAG_BACKFILLFULL flag | |
1577 | ceph_assert(tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL)); | |
1578 | continue; | |
1579 | } | |
1580 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_BACKFILLFULL)) { | |
1581 | // don't bother if pool is already marked as backfillfull | |
1582 | continue; | |
1583 | } | |
1584 | dout(10) << __func__ << " marking pool '" << tmp.pool_name[p] | |
1585 | << "'s as backfillfull" << dendl; | |
1586 | if (pending_inc.new_pools.count(p) == 0) { | |
1587 | pending_inc.new_pools[p] = tmp.pools[p]; | |
1588 | } | |
1589 | pending_inc.new_pools[p].flags |= pg_pool_t::FLAG_BACKFILLFULL; | |
1590 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL; | |
1591 | } | |
1592 | // cancel FLAG_BACKFILLFULL for pools | |
1593 | // which are no longer backfillfull too | |
1594 | for (auto &pool: tmp.get_pools()) { | |
1595 | auto p = pool.first; | |
1596 | if (full_pool_ids.count(p) || backfillfull_pool_ids.count(p)) { | |
1597 | // skip pools we have just marked as backfillfull/full above | |
1598 | continue; | |
1599 | } | |
1600 | if (!tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_BACKFILLFULL)) { | |
1601 | // and don't touch if currently is not backfillfull | |
1602 | continue; | |
1603 | } | |
1604 | dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p] | |
1605 | << "'s backfillfull flag" << dendl; | |
1606 | if (pending_inc.new_pools.count(p) == 0) { | |
1607 | pending_inc.new_pools[p] = pool.second; | |
1608 | } | |
1609 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_BACKFILLFULL; | |
3efd9988 | 1610 | } |
11fdf7f2 TL |
1611 | } |
1612 | if (!nearfull_pool_ids.empty()) { | |
1613 | for (auto &p: nearfull_pool_ids) { | |
1614 | if (full_pool_ids.count(p) || backfillfull_pool_ids.count(p)) { | |
1615 | continue; | |
1616 | } | |
1617 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL_QUOTA)) { | |
1618 | // make sure FLAG_FULL is truly set, so we are safe not | |
1619 | // to set a extra (redundant) FLAG_NEARFULL flag | |
1620 | ceph_assert(tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_FULL)); | |
1621 | continue; | |
1622 | } | |
1623 | if (tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_NEARFULL)) { | |
1624 | // don't bother if pool is already marked as nearfull | |
1625 | continue; | |
1626 | } | |
1627 | dout(10) << __func__ << " marking pool '" << tmp.pool_name[p] | |
1628 | << "'s as nearfull" << dendl; | |
1629 | if (pending_inc.new_pools.count(p) == 0) { | |
1630 | pending_inc.new_pools[p] = tmp.pools[p]; | |
1631 | } | |
1632 | pending_inc.new_pools[p].flags |= pg_pool_t::FLAG_NEARFULL; | |
1633 | } | |
1634 | // cancel FLAG_NEARFULL for pools | |
1635 | // which are no longer nearfull too | |
1636 | for (auto &pool: tmp.get_pools()) { | |
1637 | auto p = pool.first; | |
1638 | if (full_pool_ids.count(p) || | |
1639 | backfillfull_pool_ids.count(p) || | |
1640 | nearfull_pool_ids.count(p)) { | |
1641 | // skip pools we have just marked as | |
1642 | // nearfull/backfillfull/full above | |
1643 | continue; | |
1644 | } | |
1645 | if (!tmp.get_pg_pool(p)->has_flag(pg_pool_t::FLAG_NEARFULL)) { | |
1646 | // and don't touch if currently is not nearfull | |
1647 | continue; | |
1648 | } | |
1649 | dout(10) << __func__ << " clearing pool '" << tmp.pool_name[p] | |
1650 | << "'s nearfull flag" << dendl; | |
1651 | if (pending_inc.new_pools.count(p) == 0) { | |
1652 | pending_inc.new_pools[p] = pool.second; | |
1653 | } | |
1654 | pending_inc.new_pools[p].flags &= ~pg_pool_t::FLAG_NEARFULL; | |
7c673cae | 1655 | } |
11fdf7f2 | 1656 | } |
7c673cae | 1657 | |
11fdf7f2 TL |
1658 | // min_compat_client? |
1659 | if (tmp.require_min_compat_client == 0) { | |
1660 | auto mv = tmp.get_min_compat_client(); | |
1661 | dout(1) << __func__ << " setting require_min_compat_client to currently " | |
1662 | << "required " << ceph_release_name(mv) << dendl; | |
1663 | mon->clog->info() << "setting require_min_compat_client to currently " | |
1664 | << "required " << ceph_release_name(mv); | |
1665 | pending_inc.new_require_min_compat_client = mv; | |
1666 | } | |
1667 | ||
1668 | // upgrade to mimic? | |
1669 | if (osdmap.require_osd_release < CEPH_RELEASE_MIMIC && | |
1670 | tmp.require_osd_release >= CEPH_RELEASE_MIMIC) { | |
1671 | dout(10) << __func__ << " first mimic+ epoch" << dendl; | |
1672 | // record this epoch as the deletion for all legacy removed_snaps | |
1673 | for (auto& p : tmp.get_pools()) { | |
1674 | // update every pool | |
1675 | if (pending_inc.new_pools.count(p.first) == 0) { | |
1676 | pending_inc.new_pools[p.first] = p.second; | |
1677 | } | |
1678 | auto& pi = pending_inc.new_pools[p.first]; | |
1679 | if (pi.snap_seq == 0) { | |
1680 | // no snaps on this pool | |
1681 | continue; | |
1682 | } | |
1683 | if ((pi.flags & (pg_pool_t::FLAG_SELFMANAGED_SNAPS | | |
1684 | pg_pool_t::FLAG_POOL_SNAPS)) == 0) { | |
1685 | if (!pi.removed_snaps.empty()) { | |
1686 | pi.flags |= pg_pool_t::FLAG_SELFMANAGED_SNAPS; | |
1687 | } else { | |
1688 | pi.flags |= pg_pool_t::FLAG_POOL_SNAPS; | |
1689 | } | |
1690 | } | |
1691 | ||
1692 | // Make all previously removed snaps appear to be removed in this | |
1693 | // epoch. this populates removed_snaps_queue. The OSD will subtract | |
1694 | // off its purged_snaps, as before, and this set will shrink over the | |
1695 | // following epochs as the purged snaps are reported back through the | |
1696 | // mgr. | |
1697 | OSDMap::snap_interval_set_t removed; | |
1698 | if (!p.second.removed_snaps.empty()) { | |
1699 | // different flavor of interval_set :( | |
1700 | for (auto q = p.second.removed_snaps.begin(); | |
1701 | q != p.second.removed_snaps.end(); | |
1702 | ++q) { | |
1703 | removed.insert(q.get_start(), q.get_len()); | |
1704 | } | |
1705 | } else { | |
1706 | for (snapid_t s = 1; s <= pi.get_snap_seq(); s = s + 1) { | |
1707 | if (pi.snaps.count(s) == 0) { | |
1708 | removed.insert(s); | |
224ce89b WB |
1709 | } |
1710 | } | |
11fdf7f2 TL |
1711 | } |
1712 | pending_inc.new_removed_snaps[p.first].union_of(removed); | |
1713 | ||
1714 | dout(10) << __func__ << " converting pool " << p.first | |
1715 | << " with " << p.second.removed_snaps.size() | |
1716 | << " legacy removed_snaps" << dendl; | |
1717 | string k = make_snap_epoch_key(p.first, pending_inc.epoch); | |
1718 | bufferlist v; | |
1719 | encode(p.second.removed_snaps, v); | |
1720 | t->put(OSD_SNAP_PREFIX, k, v); | |
1721 | for (auto q = p.second.removed_snaps.begin(); | |
1722 | q != p.second.removed_snaps.end(); | |
1723 | ++q) { | |
1724 | bufferlist v; | |
1725 | string k = make_snap_key_value(p.first, q.get_start(), | |
1726 | q.get_len(), pending_inc.epoch, &v); | |
1727 | t->put(OSD_SNAP_PREFIX, k, v); | |
1728 | } | |
1729 | } | |
1730 | } | |
1731 | if (osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS && | |
1732 | tmp.require_osd_release >= CEPH_RELEASE_NAUTILUS) { | |
1733 | dout(10) << __func__ << " first nautilus+ epoch" << dendl; | |
1734 | // add creating flags? | |
1735 | for (auto& i : tmp.get_pools()) { | |
1736 | if (pending_creatings.still_creating_pool(i.first)) { | |
1737 | dout(10) << __func__ << " adding CREATING flag to pool " << i.first | |
1738 | << dendl; | |
1739 | if (pending_inc.new_pools.count(i.first) == 0) { | |
1740 | pending_inc.new_pools[i.first] = i.second; | |
224ce89b | 1741 | } |
11fdf7f2 | 1742 | pending_inc.new_pools[i.first].flags |= pg_pool_t::FLAG_CREATING; |
224ce89b | 1743 | } |
11fdf7f2 TL |
1744 | } |
1745 | // adjust blacklist items to all be TYPE_ANY | |
1746 | for (auto& i : tmp.blacklist) { | |
1747 | auto a = i.first; | |
1748 | a.set_type(entity_addr_t::TYPE_ANY); | |
1749 | pending_inc.new_blacklist[a] = i.second; | |
1750 | pending_inc.old_blacklist.push_back(i.first); | |
224ce89b | 1751 | } |
7c673cae FG |
1752 | } |
1753 | } | |
1754 | ||
1755 | // tell me about it | |
31f18b77 | 1756 | for (auto i = pending_inc.new_state.begin(); |
7c673cae FG |
1757 | i != pending_inc.new_state.end(); |
1758 | ++i) { | |
1759 | int s = i->second ? i->second : CEPH_OSD_UP; | |
1760 | if (s & CEPH_OSD_UP) | |
1761 | dout(2) << " osd." << i->first << " DOWN" << dendl; | |
1762 | if (s & CEPH_OSD_EXISTS) | |
1763 | dout(2) << " osd." << i->first << " DNE" << dendl; | |
1764 | } | |
11fdf7f2 | 1765 | for (auto i = pending_inc.new_up_client.begin(); |
7c673cae FG |
1766 | i != pending_inc.new_up_client.end(); |
1767 | ++i) { | |
1768 | //FIXME: insert cluster addresses too | |
1769 | dout(2) << " osd." << i->first << " UP " << i->second << dendl; | |
1770 | } | |
1771 | for (map<int32_t,uint32_t>::iterator i = pending_inc.new_weight.begin(); | |
1772 | i != pending_inc.new_weight.end(); | |
1773 | ++i) { | |
1774 | if (i->second == CEPH_OSD_OUT) { | |
1775 | dout(2) << " osd." << i->first << " OUT" << dendl; | |
1776 | } else if (i->second == CEPH_OSD_IN) { | |
1777 | dout(2) << " osd." << i->first << " IN" << dendl; | |
1778 | } else { | |
1779 | dout(2) << " osd." << i->first << " WEIGHT " << hex << i->second << dec << dendl; | |
1780 | } | |
1781 | } | |
1782 | ||
1783 | // features for osdmap and its incremental | |
28e407b8 | 1784 | uint64_t features; |
7c673cae FG |
1785 | |
1786 | // encode full map and determine its crc | |
1787 | OSDMap tmp; | |
1788 | { | |
1789 | tmp.deepish_copy_from(osdmap); | |
1790 | tmp.apply_incremental(pending_inc); | |
1791 | ||
1792 | // determine appropriate features | |
28e407b8 AA |
1793 | features = tmp.get_encoding_features(); |
1794 | dout(10) << __func__ << " encoding full map with " | |
1795 | << ceph_release_name(tmp.require_osd_release) | |
1796 | << " features " << features << dendl; | |
1797 | ||
1798 | // the features should be a subset of the mon quorum's features! | |
11fdf7f2 | 1799 | ceph_assert((features & ~mon->get_quorum_con_features()) == 0); |
7c673cae FG |
1800 | |
1801 | bufferlist fullbl; | |
11fdf7f2 | 1802 | encode(tmp, fullbl, features | CEPH_FEATURE_RESERVED); |
7c673cae FG |
1803 | pending_inc.full_crc = tmp.get_crc(); |
1804 | ||
1805 | // include full map in the txn. note that old monitors will | |
1806 | // overwrite this. new ones will now skip the local full map | |
1807 | // encode and reload from this. | |
1808 | put_version_full(t, pending_inc.epoch, fullbl); | |
1809 | } | |
1810 | ||
1811 | // encode | |
11fdf7f2 TL |
1812 | ceph_assert(get_last_committed() + 1 == pending_inc.epoch); |
1813 | bufferlist bl; | |
1814 | encode(pending_inc, bl, features | CEPH_FEATURE_RESERVED); | |
7c673cae FG |
1815 | |
1816 | dout(20) << " full_crc " << tmp.get_crc() | |
1817 | << " inc_crc " << pending_inc.inc_crc << dendl; | |
1818 | ||
1819 | /* put everything in the transaction */ | |
1820 | put_version(t, pending_inc.epoch, bl); | |
1821 | put_last_committed(t, pending_inc.epoch); | |
1822 | ||
1823 | // metadata, too! | |
1824 | for (map<int,bufferlist>::iterator p = pending_metadata.begin(); | |
1825 | p != pending_metadata.end(); | |
1826 | ++p) | |
1827 | t->put(OSD_METADATA_PREFIX, stringify(p->first), p->second); | |
1828 | for (set<int>::iterator p = pending_metadata_rm.begin(); | |
1829 | p != pending_metadata_rm.end(); | |
1830 | ++p) | |
1831 | t->erase(OSD_METADATA_PREFIX, stringify(*p)); | |
1832 | pending_metadata.clear(); | |
1833 | pending_metadata_rm.clear(); | |
1834 | ||
11fdf7f2 TL |
1835 | // removed_snaps |
1836 | if (tmp.require_osd_release >= CEPH_RELEASE_MIMIC) { | |
1837 | for (auto& i : pending_inc.new_removed_snaps) { | |
1838 | { | |
1839 | // all snaps removed this epoch | |
1840 | string k = make_snap_epoch_key(i.first, pending_inc.epoch); | |
1841 | bufferlist v; | |
1842 | encode(i.second, v); | |
1843 | t->put(OSD_SNAP_PREFIX, k, v); | |
1844 | } | |
1845 | for (auto q = i.second.begin(); | |
1846 | q != i.second.end(); | |
1847 | ++q) { | |
1848 | bufferlist v; | |
1849 | string k = make_snap_key_value(i.first, q.get_start(), | |
1850 | q.get_len(), pending_inc.epoch, &v); | |
1851 | t->put(OSD_SNAP_PREFIX, k, v); | |
1852 | } | |
1853 | } | |
1854 | for (auto& i : pending_inc.new_purged_snaps) { | |
1855 | for (auto q = i.second.begin(); | |
1856 | q != i.second.end(); | |
1857 | ++q) { | |
1858 | bufferlist v; | |
1859 | string k = make_snap_purged_key_value(i.first, q.get_start(), | |
1860 | q.get_len(), pending_inc.epoch, | |
1861 | &v); | |
1862 | t->put(OSD_SNAP_PREFIX, k, v); | |
1863 | } | |
7c673cae | 1864 | } |
7c673cae | 1865 | } |
224ce89b WB |
1866 | |
1867 | // health | |
1868 | health_check_map_t next; | |
1869 | tmp.check_health(&next); | |
1870 | encode_health(next, t); | |
7c673cae FG |
1871 | } |
1872 | ||
7c673cae FG |
1873 | int OSDMonitor::load_metadata(int osd, map<string, string>& m, ostream *err) |
1874 | { | |
1875 | bufferlist bl; | |
1876 | int r = mon->store->get(OSD_METADATA_PREFIX, stringify(osd), bl); | |
1877 | if (r < 0) | |
1878 | return r; | |
1879 | try { | |
11fdf7f2 TL |
1880 | auto p = bl.cbegin(); |
1881 | decode(m, p); | |
7c673cae FG |
1882 | } |
1883 | catch (buffer::error& e) { | |
1884 | if (err) | |
1885 | *err << "osd." << osd << " metadata is corrupt"; | |
1886 | return -EIO; | |
1887 | } | |
1888 | return 0; | |
1889 | } | |
1890 | ||
c07f9fc5 | 1891 | void OSDMonitor::count_metadata(const string& field, map<string,int> *out) |
31f18b77 | 1892 | { |
31f18b77 FG |
1893 | for (int osd = 0; osd < osdmap.get_max_osd(); ++osd) { |
1894 | if (osdmap.is_up(osd)) { | |
1895 | map<string,string> meta; | |
1896 | load_metadata(osd, meta, nullptr); | |
1897 | auto p = meta.find(field); | |
1898 | if (p == meta.end()) { | |
c07f9fc5 | 1899 | (*out)["unknown"]++; |
31f18b77 | 1900 | } else { |
c07f9fc5 | 1901 | (*out)[p->second]++; |
31f18b77 FG |
1902 | } |
1903 | } | |
1904 | } | |
c07f9fc5 FG |
1905 | } |
1906 | ||
1907 | void OSDMonitor::count_metadata(const string& field, Formatter *f) | |
1908 | { | |
1909 | map<string,int> by_val; | |
1910 | count_metadata(field, &by_val); | |
31f18b77 FG |
1911 | f->open_object_section(field.c_str()); |
1912 | for (auto& p : by_val) { | |
1913 | f->dump_int(p.first.c_str(), p.second); | |
1914 | } | |
1915 | f->close_section(); | |
1916 | } | |
1917 | ||
7c673cae FG |
1918 | int OSDMonitor::get_osd_objectstore_type(int osd, string *type) |
1919 | { | |
1920 | map<string, string> metadata; | |
1921 | int r = load_metadata(osd, metadata, nullptr); | |
1922 | if (r < 0) | |
1923 | return r; | |
1924 | ||
1925 | auto it = metadata.find("osd_objectstore"); | |
1926 | if (it == metadata.end()) | |
1927 | return -ENOENT; | |
1928 | *type = it->second; | |
1929 | return 0; | |
1930 | } | |
1931 | ||
1932 | bool OSDMonitor::is_pool_currently_all_bluestore(int64_t pool_id, | |
1933 | const pg_pool_t &pool, | |
1934 | ostream *err) | |
1935 | { | |
1936 | // just check a few pgs for efficiency - this can't give a guarantee anyway, | |
1937 | // since filestore osds could always join the pool later | |
1938 | set<int> checked_osds; | |
11fdf7f2 | 1939 | for (unsigned ps = 0; ps < std::min(8u, pool.get_pg_num()); ++ps) { |
7c673cae | 1940 | vector<int> up, acting; |
11fdf7f2 | 1941 | pg_t pgid(ps, pool_id); |
7c673cae FG |
1942 | osdmap.pg_to_up_acting_osds(pgid, up, acting); |
1943 | for (int osd : up) { | |
1944 | if (checked_osds.find(osd) != checked_osds.end()) | |
1945 | continue; | |
1946 | string objectstore_type; | |
1947 | int r = get_osd_objectstore_type(osd, &objectstore_type); | |
1948 | // allow with missing metadata, e.g. due to an osd never booting yet | |
1949 | if (r < 0 || objectstore_type == "bluestore") { | |
1950 | checked_osds.insert(osd); | |
1951 | continue; | |
1952 | } | |
1953 | *err << "osd." << osd << " uses " << objectstore_type; | |
1954 | return false; | |
1955 | } | |
1956 | } | |
1957 | return true; | |
1958 | } | |
1959 | ||
1960 | int OSDMonitor::dump_osd_metadata(int osd, Formatter *f, ostream *err) | |
1961 | { | |
1962 | map<string,string> m; | |
1963 | if (int r = load_metadata(osd, m, err)) | |
1964 | return r; | |
1965 | for (map<string,string>::iterator p = m.begin(); p != m.end(); ++p) | |
1966 | f->dump_string(p->first.c_str(), p->second); | |
1967 | return 0; | |
1968 | } | |
1969 | ||
1970 | void OSDMonitor::print_nodes(Formatter *f) | |
1971 | { | |
1972 | // group OSDs by their hosts | |
1973 | map<string, list<int> > osds; // hostname => osd | |
1974 | for (int osd = 0; osd < osdmap.get_max_osd(); osd++) { | |
1975 | map<string, string> m; | |
1976 | if (load_metadata(osd, m, NULL)) { | |
1977 | continue; | |
1978 | } | |
1979 | map<string, string>::iterator hostname = m.find("hostname"); | |
1980 | if (hostname == m.end()) { | |
1981 | // not likely though | |
1982 | continue; | |
1983 | } | |
1984 | osds[hostname->second].push_back(osd); | |
1985 | } | |
1986 | ||
1987 | dump_services(f, osds, "osd"); | |
1988 | } | |
1989 | ||
1990 | void OSDMonitor::share_map_with_random_osd() | |
1991 | { | |
1992 | if (osdmap.get_num_up_osds() == 0) { | |
1993 | dout(10) << __func__ << " no up osds, don't share with anyone" << dendl; | |
1994 | return; | |
1995 | } | |
1996 | ||
1997 | MonSession *s = mon->session_map.get_random_osd_session(&osdmap); | |
1998 | if (!s) { | |
1999 | dout(10) << __func__ << " no up osd on our session map" << dendl; | |
2000 | return; | |
2001 | } | |
2002 | ||
11fdf7f2 TL |
2003 | dout(10) << "committed, telling random " << s->name |
2004 | << " all about it" << dendl; | |
28e407b8 AA |
2005 | |
2006 | // get feature of the peer | |
2007 | // use quorum_con_features, if it's an anonymous connection. | |
2008 | uint64_t features = s->con_features ? s->con_features : | |
2009 | mon->get_quorum_con_features(); | |
7c673cae | 2010 | // whatev, they'll request more if they need it |
28e407b8 | 2011 | MOSDMap *m = build_incremental(osdmap.get_epoch() - 1, osdmap.get_epoch(), features); |
7c673cae FG |
2012 | s->con->send_message(m); |
2013 | // NOTE: do *not* record osd has up to this epoch (as we do | |
2014 | // elsewhere) as they may still need to request older values. | |
2015 | } | |
2016 | ||
11fdf7f2 | 2017 | version_t OSDMonitor::get_trim_to() const |
7c673cae | 2018 | { |
31f18b77 FG |
2019 | if (mon->get_quorum().empty()) { |
2020 | dout(10) << __func__ << ": quorum not formed" << dendl; | |
2021 | return 0; | |
2022 | } | |
7c673cae | 2023 | |
11fdf7f2 TL |
2024 | { |
2025 | std::lock_guard<std::mutex> l(creating_pgs_lock); | |
2026 | if (!creating_pgs.pgs.empty()) { | |
7c673cae FG |
2027 | return 0; |
2028 | } | |
7c673cae | 2029 | } |
11fdf7f2 TL |
2030 | |
2031 | if (g_conf().get_val<bool>("mon_debug_block_osdmap_trim")) { | |
2032 | dout(0) << __func__ | |
2033 | << " blocking osdmap trim" | |
2034 | " ('mon_debug_block_osdmap_trim' set to 'true')" | |
2035 | << dendl; | |
2036 | return 0; | |
2037 | } | |
2038 | ||
7c673cae | 2039 | { |
11fdf7f2 | 2040 | epoch_t floor = get_min_last_epoch_clean(); |
7c673cae | 2041 | dout(10) << " min_last_epoch_clean " << floor << dendl; |
11fdf7f2 TL |
2042 | if (g_conf()->mon_osd_force_trim_to > 0 && |
2043 | g_conf()->mon_osd_force_trim_to < (int)get_last_committed()) { | |
2044 | floor = g_conf()->mon_osd_force_trim_to; | |
7c673cae FG |
2045 | dout(10) << " explicit mon_osd_force_trim_to = " << floor << dendl; |
2046 | } | |
11fdf7f2 | 2047 | unsigned min = g_conf()->mon_min_osdmap_epochs; |
7c673cae FG |
2048 | if (floor + min > get_last_committed()) { |
2049 | if (min < get_last_committed()) | |
2050 | floor = get_last_committed() - min; | |
2051 | else | |
2052 | floor = 0; | |
2053 | } | |
2054 | if (floor > get_first_committed()) | |
2055 | return floor; | |
2056 | } | |
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | epoch_t OSDMonitor::get_min_last_epoch_clean() const | |
2061 | { | |
2062 | auto floor = last_epoch_clean.get_lower_bound(osdmap); | |
2063 | // also scan osd epochs | |
2064 | // don't trim past the oldest reported osd epoch | |
2065 | for (auto& osd_epoch : osd_epochs) { | |
2066 | if (osd_epoch.second < floor) { | |
2067 | floor = osd_epoch.second; | |
2068 | } | |
2069 | } | |
2070 | return floor; | |
2071 | } | |
2072 | ||
2073 | void OSDMonitor::encode_trim_extra(MonitorDBStore::TransactionRef tx, | |
2074 | version_t first) | |
2075 | { | |
2076 | dout(10) << __func__ << " including full map for e " << first << dendl; | |
2077 | bufferlist bl; | |
2078 | get_version_full(first, bl); | |
2079 | put_version_full(tx, first, bl); | |
11fdf7f2 TL |
2080 | |
2081 | if (has_osdmap_manifest && | |
2082 | first > osdmap_manifest.get_first_pinned()) { | |
2083 | _prune_update_trimmed(tx, first); | |
2084 | } | |
7c673cae FG |
2085 | } |
2086 | ||
11fdf7f2 TL |
2087 | |
2088 | /* full osdmap prune | |
2089 | * | |
2090 | * for more information, please refer to doc/dev/mon-osdmap-prune.rst | |
2091 | */ | |
2092 | ||
2093 | void OSDMonitor::load_osdmap_manifest() | |
2094 | { | |
2095 | bool store_has_manifest = | |
2096 | mon->store->exists(get_service_name(), "osdmap_manifest"); | |
2097 | ||
2098 | if (!store_has_manifest) { | |
2099 | if (!has_osdmap_manifest) { | |
2100 | return; | |
2101 | } | |
2102 | ||
2103 | dout(20) << __func__ | |
2104 | << " dropping osdmap manifest from memory." << dendl; | |
2105 | osdmap_manifest = osdmap_manifest_t(); | |
2106 | has_osdmap_manifest = false; | |
2107 | return; | |
2108 | } | |
2109 | ||
2110 | dout(20) << __func__ | |
2111 | << " osdmap manifest detected in store; reload." << dendl; | |
2112 | ||
2113 | bufferlist manifest_bl; | |
2114 | int r = get_value("osdmap_manifest", manifest_bl); | |
2115 | if (r < 0) { | |
2116 | derr << __func__ << " unable to read osdmap version manifest" << dendl; | |
2117 | ceph_abort_msg("error reading manifest"); | |
2118 | } | |
2119 | osdmap_manifest.decode(manifest_bl); | |
2120 | has_osdmap_manifest = true; | |
2121 | ||
2122 | dout(10) << __func__ << " store osdmap manifest pinned (" | |
2123 | << osdmap_manifest.get_first_pinned() | |
2124 | << " .. " | |
2125 | << osdmap_manifest.get_last_pinned() | |
2126 | << ")" | |
2127 | << dendl; | |
2128 | } | |
2129 | ||
2130 | bool OSDMonitor::should_prune() const | |
2131 | { | |
2132 | version_t first = get_first_committed(); | |
2133 | version_t last = get_last_committed(); | |
2134 | version_t min_osdmap_epochs = | |
2135 | g_conf().get_val<int64_t>("mon_min_osdmap_epochs"); | |
2136 | version_t prune_min = | |
2137 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_min"); | |
2138 | version_t prune_interval = | |
2139 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_interval"); | |
2140 | version_t last_pinned = osdmap_manifest.get_last_pinned(); | |
2141 | version_t last_to_pin = last - min_osdmap_epochs; | |
2142 | ||
2143 | // Make it or break it constraints. | |
2144 | // | |
2145 | // If any of these conditions fails, we will not prune, regardless of | |
2146 | // whether we have an on-disk manifest with an on-going pruning state. | |
2147 | // | |
2148 | if ((last - first) <= min_osdmap_epochs) { | |
2149 | // between the first and last committed epochs, we don't have | |
2150 | // enough epochs to trim, much less to prune. | |
2151 | dout(10) << __func__ | |
2152 | << " currently holding only " << (last - first) | |
2153 | << " epochs (min osdmap epochs: " << min_osdmap_epochs | |
2154 | << "); do not prune." | |
2155 | << dendl; | |
2156 | return false; | |
2157 | ||
2158 | } else if ((last_to_pin - first) < prune_min) { | |
2159 | // between the first committed epoch and the last epoch we would prune, | |
2160 | // we simply don't have enough versions over the minimum to prune maps. | |
2161 | dout(10) << __func__ | |
2162 | << " could only prune " << (last_to_pin - first) | |
2163 | << " epochs (" << first << ".." << last_to_pin << "), which" | |
2164 | " is less than the required minimum (" << prune_min << ")" | |
2165 | << dendl; | |
2166 | return false; | |
2167 | ||
2168 | } else if (has_osdmap_manifest && last_pinned >= last_to_pin) { | |
2169 | dout(10) << __func__ | |
2170 | << " we have pruned as far as we can; do not prune." | |
2171 | << dendl; | |
2172 | return false; | |
2173 | ||
2174 | } else if (last_pinned + prune_interval > last_to_pin) { | |
2175 | dout(10) << __func__ | |
2176 | << " not enough epochs to form an interval (last pinned: " | |
2177 | << last_pinned << ", last to pin: " | |
2178 | << last_to_pin << ", interval: " << prune_interval << ")" | |
2179 | << dendl; | |
2180 | return false; | |
2181 | } | |
2182 | ||
2183 | dout(15) << __func__ | |
2184 | << " should prune (" << last_pinned << ".." << last_to_pin << ")" | |
2185 | << " lc (" << first << ".." << last << ")" | |
2186 | << dendl; | |
2187 | return true; | |
2188 | } | |
2189 | ||
2190 | void OSDMonitor::_prune_update_trimmed( | |
2191 | MonitorDBStore::TransactionRef tx, | |
2192 | version_t first) | |
2193 | { | |
2194 | dout(10) << __func__ | |
2195 | << " first " << first | |
2196 | << " last_pinned " << osdmap_manifest.get_last_pinned() | |
2197 | << " last_pinned " << osdmap_manifest.get_last_pinned() | |
2198 | << dendl; | |
2199 | ||
2200 | osdmap_manifest_t manifest = osdmap_manifest; | |
2201 | ||
2202 | if (!manifest.is_pinned(first)) { | |
2203 | manifest.pin(first); | |
2204 | } | |
2205 | ||
2206 | set<version_t>::iterator p_end = manifest.pinned.find(first); | |
2207 | set<version_t>::iterator p = manifest.pinned.begin(); | |
2208 | manifest.pinned.erase(p, p_end); | |
2209 | ceph_assert(manifest.get_first_pinned() == first); | |
2210 | ||
2211 | if (manifest.get_last_pinned() == first+1 || | |
2212 | manifest.pinned.size() == 1) { | |
2213 | // we reached the end of the line, as pinned maps go; clean up our | |
2214 | // manifest, and let `should_prune()` decide whether we should prune | |
2215 | // again. | |
2216 | tx->erase(get_service_name(), "osdmap_manifest"); | |
2217 | return; | |
2218 | } | |
2219 | ||
2220 | bufferlist bl; | |
2221 | manifest.encode(bl); | |
2222 | tx->put(get_service_name(), "osdmap_manifest", bl); | |
2223 | } | |
2224 | ||
2225 | void OSDMonitor::prune_init(osdmap_manifest_t& manifest) | |
2226 | { | |
2227 | dout(1) << __func__ << dendl; | |
2228 | ||
2229 | version_t pin_first; | |
2230 | ||
2231 | // verify constrainsts on stable in-memory state | |
2232 | if (!has_osdmap_manifest) { | |
2233 | // we must have never pruned, OR if we pruned the state must no longer | |
2234 | // be relevant (i.e., the state must have been removed alongside with | |
2235 | // the trim that *must* have removed past the last pinned map in a | |
2236 | // previous prune). | |
2237 | ceph_assert(osdmap_manifest.pinned.empty()); | |
2238 | ceph_assert(!mon->store->exists(get_service_name(), "osdmap_manifest")); | |
2239 | pin_first = get_first_committed(); | |
2240 | ||
2241 | } else { | |
2242 | // we must have pruned in the past AND its state is still relevant | |
2243 | // (i.e., even if we trimmed, we still hold pinned maps in the manifest, | |
2244 | // and thus we still hold a manifest in the store). | |
2245 | ceph_assert(!osdmap_manifest.pinned.empty()); | |
2246 | ceph_assert(osdmap_manifest.get_first_pinned() == get_first_committed()); | |
2247 | ceph_assert(osdmap_manifest.get_last_pinned() < get_last_committed()); | |
2248 | ||
2249 | dout(10) << __func__ | |
2250 | << " first_pinned " << osdmap_manifest.get_first_pinned() | |
2251 | << " last_pinned " << osdmap_manifest.get_last_pinned() | |
2252 | << dendl; | |
2253 | ||
2254 | pin_first = osdmap_manifest.get_last_pinned(); | |
2255 | } | |
2256 | ||
2257 | manifest.pin(pin_first); | |
2258 | } | |
2259 | ||
2260 | bool OSDMonitor::_prune_sanitize_options() const | |
2261 | { | |
2262 | uint64_t prune_interval = | |
2263 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_interval"); | |
2264 | uint64_t prune_min = | |
2265 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_min"); | |
2266 | uint64_t txsize = | |
2267 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_txsize"); | |
2268 | ||
2269 | bool r = true; | |
2270 | ||
2271 | if (prune_interval == 0) { | |
2272 | derr << __func__ | |
2273 | << " prune is enabled BUT prune interval is zero; abort." | |
2274 | << dendl; | |
2275 | r = false; | |
2276 | } else if (prune_interval == 1) { | |
2277 | derr << __func__ | |
2278 | << " prune interval is equal to one, which essentially means" | |
2279 | " no pruning; abort." | |
2280 | << dendl; | |
2281 | r = false; | |
2282 | } | |
2283 | if (prune_min == 0) { | |
2284 | derr << __func__ | |
2285 | << " prune is enabled BUT prune min is zero; abort." | |
2286 | << dendl; | |
2287 | r = false; | |
2288 | } | |
2289 | if (prune_interval > prune_min) { | |
2290 | derr << __func__ | |
2291 | << " impossible to ascertain proper prune interval because" | |
2292 | << " it is greater than the minimum prune epochs" | |
2293 | << " (min: " << prune_min << ", interval: " << prune_interval << ")" | |
2294 | << dendl; | |
2295 | r = false; | |
2296 | } | |
2297 | ||
2298 | if (txsize < prune_interval - 1) { | |
2299 | derr << __func__ | |
2300 | << "'mon_osdmap_full_prune_txsize' (" << txsize | |
2301 | << ") < 'mon_osdmap_full_prune_interval-1' (" << prune_interval - 1 | |
2302 | << "); abort." << dendl; | |
2303 | r = false; | |
2304 | } | |
2305 | return r; | |
2306 | } | |
2307 | ||
2308 | bool OSDMonitor::is_prune_enabled() const { | |
2309 | return g_conf().get_val<bool>("mon_osdmap_full_prune_enabled"); | |
2310 | } | |
2311 | ||
2312 | bool OSDMonitor::is_prune_supported() const { | |
2313 | return mon->get_required_mon_features().contains_any( | |
2314 | ceph::features::mon::FEATURE_OSDMAP_PRUNE); | |
2315 | } | |
2316 | ||
2317 | /** do_prune | |
2318 | * | |
2319 | * @returns true if has side-effects; false otherwise. | |
2320 | */ | |
2321 | bool OSDMonitor::do_prune(MonitorDBStore::TransactionRef tx) | |
2322 | { | |
2323 | bool enabled = is_prune_enabled(); | |
2324 | ||
2325 | dout(1) << __func__ << " osdmap full prune " | |
2326 | << ( enabled ? "enabled" : "disabled") | |
2327 | << dendl; | |
2328 | ||
2329 | if (!enabled || !_prune_sanitize_options() || !should_prune()) { | |
2330 | return false; | |
2331 | } | |
2332 | ||
2333 | // we are beyond the minimum prune versions, we need to remove maps because | |
2334 | // otherwise the store will grow unbounded and we may end up having issues | |
2335 | // with available disk space or store hangs. | |
2336 | ||
2337 | // we will not pin all versions. We will leave a buffer number of versions. | |
2338 | // this allows us the monitor to trim maps without caring too much about | |
2339 | // pinned maps, and then allow us to use another ceph-mon without these | |
2340 | // capabilities, without having to repair the store. | |
2341 | ||
2342 | osdmap_manifest_t manifest = osdmap_manifest; | |
2343 | ||
2344 | version_t first = get_first_committed(); | |
2345 | version_t last = get_last_committed(); | |
2346 | ||
2347 | version_t last_to_pin = last - g_conf()->mon_min_osdmap_epochs; | |
2348 | version_t last_pinned = manifest.get_last_pinned(); | |
2349 | uint64_t prune_interval = | |
2350 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_interval"); | |
2351 | uint64_t txsize = | |
2352 | g_conf().get_val<uint64_t>("mon_osdmap_full_prune_txsize"); | |
2353 | ||
2354 | prune_init(manifest); | |
2355 | ||
2356 | // we need to get rid of some osdmaps | |
2357 | ||
2358 | dout(5) << __func__ | |
2359 | << " lc (" << first << " .. " << last << ")" | |
2360 | << " last_pinned " << last_pinned | |
2361 | << " interval " << prune_interval | |
2362 | << " last_to_pin " << last_to_pin | |
2363 | << dendl; | |
2364 | ||
2365 | // We will be erasing maps as we go. | |
2366 | // | |
2367 | // We will erase all maps between `last_pinned` and the `next_to_pin`. | |
2368 | // | |
2369 | // If `next_to_pin` happens to be greater than `last_to_pin`, then | |
2370 | // we stop pruning. We could prune the maps between `next_to_pin` and | |
2371 | // `last_to_pin`, but by not doing it we end up with neater pruned | |
2372 | // intervals, aligned with `prune_interval`. Besides, this should not be a | |
2373 | // problem as long as `prune_interval` is set to a sane value, instead of | |
2374 | // hundreds or thousands of maps. | |
2375 | ||
2376 | auto map_exists = [this](version_t v) { | |
2377 | string k = mon->store->combine_strings("full", v); | |
2378 | return mon->store->exists(get_service_name(), k); | |
2379 | }; | |
2380 | ||
2381 | // 'interval' represents the number of maps from the last pinned | |
2382 | // i.e., if we pinned version 1 and have an interval of 10, we're pinning | |
2383 | // version 11 next; all intermediate versions will be removed. | |
2384 | // | |
2385 | // 'txsize' represents the maximum number of versions we'll be removing in | |
2386 | // this iteration. If 'txsize' is large enough to perform multiple passes | |
2387 | // pinning and removing maps, we will do so; if not, we'll do at least one | |
2388 | // pass. We are quite relaxed about honouring 'txsize', but we'll always | |
2389 | // ensure that we never go *over* the maximum. | |
2390 | ||
2391 | // e.g., if we pin 1 and 11, we're removing versions [2..10]; i.e., 9 maps. | |
2392 | uint64_t removal_interval = prune_interval - 1; | |
2393 | ||
2394 | if (txsize < removal_interval) { | |
2395 | dout(5) << __func__ | |
2396 | << " setting txsize to removal interval size (" | |
2397 | << removal_interval << " versions" | |
2398 | << dendl; | |
2399 | txsize = removal_interval; | |
2400 | } | |
2401 | ceph_assert(removal_interval > 0); | |
2402 | ||
2403 | uint64_t num_pruned = 0; | |
2404 | while (num_pruned + removal_interval <= txsize) { | |
2405 | last_pinned = manifest.get_last_pinned(); | |
2406 | ||
2407 | if (last_pinned + prune_interval > last_to_pin) { | |
2408 | break; | |
2409 | } | |
2410 | ceph_assert(last_pinned < last_to_pin); | |
2411 | ||
2412 | version_t next_pinned = last_pinned + prune_interval; | |
2413 | ceph_assert(next_pinned <= last_to_pin); | |
2414 | manifest.pin(next_pinned); | |
2415 | ||
2416 | dout(20) << __func__ | |
2417 | << " last_pinned " << last_pinned | |
2418 | << " next_pinned " << next_pinned | |
2419 | << " num_pruned " << num_pruned | |
2420 | << " removal interval (" << (last_pinned+1) | |
2421 | << ".." << (next_pinned-1) << ")" | |
2422 | << " txsize " << txsize << dendl; | |
2423 | ||
2424 | ceph_assert(map_exists(last_pinned)); | |
2425 | ceph_assert(map_exists(next_pinned)); | |
2426 | ||
2427 | for (version_t v = last_pinned+1; v < next_pinned; ++v) { | |
2428 | ceph_assert(!manifest.is_pinned(v)); | |
2429 | ||
2430 | dout(20) << __func__ << " pruning full osdmap e" << v << dendl; | |
2431 | string full_key = mon->store->combine_strings("full", v); | |
2432 | tx->erase(get_service_name(), full_key); | |
2433 | ++num_pruned; | |
2434 | } | |
2435 | } | |
2436 | ||
2437 | ceph_assert(num_pruned > 0); | |
2438 | ||
2439 | bufferlist bl; | |
2440 | manifest.encode(bl); | |
2441 | tx->put(get_service_name(), "osdmap_manifest", bl); | |
2442 | ||
2443 | return true; | |
2444 | } | |
2445 | ||
2446 | ||
7c673cae FG |
2447 | // ------------- |
2448 | ||
2449 | bool OSDMonitor::preprocess_query(MonOpRequestRef op) | |
2450 | { | |
2451 | op->mark_osdmon_event(__func__); | |
2452 | Message *m = op->get_req(); | |
2453 | dout(10) << "preprocess_query " << *m << " from " << m->get_orig_source_inst() << dendl; | |
2454 | ||
2455 | switch (m->get_type()) { | |
2456 | // READs | |
2457 | case MSG_MON_COMMAND: | |
f64942e4 AA |
2458 | try { |
2459 | return preprocess_command(op); | |
11fdf7f2 | 2460 | } catch (const bad_cmd_get& e) { |
f64942e4 AA |
2461 | bufferlist bl; |
2462 | mon->reply_command(op, -EINVAL, e.what(), bl, get_last_committed()); | |
2463 | return true; | |
2464 | } | |
7c673cae FG |
2465 | case CEPH_MSG_MON_GET_OSDMAP: |
2466 | return preprocess_get_osdmap(op); | |
2467 | ||
2468 | // damp updates | |
2469 | case MSG_OSD_MARK_ME_DOWN: | |
2470 | return preprocess_mark_me_down(op); | |
2471 | case MSG_OSD_FULL: | |
2472 | return preprocess_full(op); | |
2473 | case MSG_OSD_FAILURE: | |
2474 | return preprocess_failure(op); | |
2475 | case MSG_OSD_BOOT: | |
2476 | return preprocess_boot(op); | |
2477 | case MSG_OSD_ALIVE: | |
2478 | return preprocess_alive(op); | |
2479 | case MSG_OSD_PG_CREATED: | |
2480 | return preprocess_pg_created(op); | |
11fdf7f2 TL |
2481 | case MSG_OSD_PG_READY_TO_MERGE: |
2482 | return preprocess_pg_ready_to_merge(op); | |
7c673cae FG |
2483 | case MSG_OSD_PGTEMP: |
2484 | return preprocess_pgtemp(op); | |
2485 | case MSG_OSD_BEACON: | |
2486 | return preprocess_beacon(op); | |
2487 | ||
2488 | case CEPH_MSG_POOLOP: | |
2489 | return preprocess_pool_op(op); | |
2490 | ||
2491 | case MSG_REMOVE_SNAPS: | |
2492 | return preprocess_remove_snaps(op); | |
2493 | ||
2494 | default: | |
2495 | ceph_abort(); | |
2496 | return true; | |
2497 | } | |
2498 | } | |
2499 | ||
2500 | bool OSDMonitor::prepare_update(MonOpRequestRef op) | |
2501 | { | |
2502 | op->mark_osdmon_event(__func__); | |
2503 | Message *m = op->get_req(); | |
2504 | dout(7) << "prepare_update " << *m << " from " << m->get_orig_source_inst() << dendl; | |
2505 | ||
2506 | switch (m->get_type()) { | |
2507 | // damp updates | |
2508 | case MSG_OSD_MARK_ME_DOWN: | |
2509 | return prepare_mark_me_down(op); | |
2510 | case MSG_OSD_FULL: | |
2511 | return prepare_full(op); | |
2512 | case MSG_OSD_FAILURE: | |
2513 | return prepare_failure(op); | |
2514 | case MSG_OSD_BOOT: | |
2515 | return prepare_boot(op); | |
2516 | case MSG_OSD_ALIVE: | |
2517 | return prepare_alive(op); | |
2518 | case MSG_OSD_PG_CREATED: | |
2519 | return prepare_pg_created(op); | |
2520 | case MSG_OSD_PGTEMP: | |
2521 | return prepare_pgtemp(op); | |
11fdf7f2 TL |
2522 | case MSG_OSD_PG_READY_TO_MERGE: |
2523 | return prepare_pg_ready_to_merge(op); | |
7c673cae FG |
2524 | case MSG_OSD_BEACON: |
2525 | return prepare_beacon(op); | |
2526 | ||
2527 | case MSG_MON_COMMAND: | |
f64942e4 AA |
2528 | try { |
2529 | return prepare_command(op); | |
11fdf7f2 | 2530 | } catch (const bad_cmd_get& e) { |
f64942e4 AA |
2531 | bufferlist bl; |
2532 | mon->reply_command(op, -EINVAL, e.what(), bl, get_last_committed()); | |
2533 | return true; | |
2534 | } | |
7c673cae FG |
2535 | |
2536 | case CEPH_MSG_POOLOP: | |
2537 | return prepare_pool_op(op); | |
2538 | ||
2539 | case MSG_REMOVE_SNAPS: | |
2540 | return prepare_remove_snaps(op); | |
2541 | ||
2542 | ||
2543 | default: | |
2544 | ceph_abort(); | |
2545 | } | |
2546 | ||
2547 | return false; | |
2548 | } | |
2549 | ||
2550 | bool OSDMonitor::should_propose(double& delay) | |
2551 | { | |
2552 | dout(10) << "should_propose" << dendl; | |
2553 | ||
2554 | // if full map, propose immediately! any subsequent changes will be clobbered. | |
2555 | if (pending_inc.fullmap.length()) | |
2556 | return true; | |
2557 | ||
2558 | // adjust osd weights? | |
2559 | if (!osd_weight.empty() && | |
2560 | osd_weight.size() == (unsigned)osdmap.get_max_osd()) { | |
2561 | dout(0) << " adjusting osd weights based on " << osd_weight << dendl; | |
2562 | osdmap.adjust_osd_weights(osd_weight, pending_inc); | |
2563 | delay = 0.0; | |
2564 | osd_weight.clear(); | |
2565 | return true; | |
2566 | } | |
2567 | ||
7c673cae FG |
2568 | return PaxosService::should_propose(delay); |
2569 | } | |
2570 | ||
2571 | ||
2572 | ||
2573 | // --------------------------- | |
2574 | // READs | |
2575 | ||
2576 | bool OSDMonitor::preprocess_get_osdmap(MonOpRequestRef op) | |
2577 | { | |
2578 | op->mark_osdmon_event(__func__); | |
2579 | MMonGetOSDMap *m = static_cast<MMonGetOSDMap*>(op->get_req()); | |
28e407b8 AA |
2580 | |
2581 | uint64_t features = mon->get_quorum_con_features(); | |
11fdf7f2 TL |
2582 | if (op->get_session() && op->get_session()->con_features) |
2583 | features = op->get_session()->con_features; | |
28e407b8 | 2584 | |
7c673cae | 2585 | dout(10) << __func__ << " " << *m << dendl; |
28e407b8 | 2586 | MOSDMap *reply = new MOSDMap(mon->monmap->fsid, features); |
7c673cae FG |
2587 | epoch_t first = get_first_committed(); |
2588 | epoch_t last = osdmap.get_epoch(); | |
11fdf7f2 TL |
2589 | int max = g_conf()->osd_map_message_max; |
2590 | ssize_t max_bytes = g_conf()->osd_map_message_max_bytes; | |
2591 | for (epoch_t e = std::max(first, m->get_full_first()); | |
2592 | e <= std::min(last, m->get_full_last()) && max > 0 && max_bytes > 0; | |
7c673cae | 2593 | ++e, --max) { |
11fdf7f2 TL |
2594 | bufferlist& bl = reply->maps[e]; |
2595 | int r = get_version_full(e, features, bl); | |
2596 | ceph_assert(r >= 0); | |
2597 | max_bytes -= bl.length(); | |
7c673cae | 2598 | } |
11fdf7f2 TL |
2599 | for (epoch_t e = std::max(first, m->get_inc_first()); |
2600 | e <= std::min(last, m->get_inc_last()) && max > 0 && max_bytes > 0; | |
7c673cae | 2601 | ++e, --max) { |
11fdf7f2 TL |
2602 | bufferlist& bl = reply->incremental_maps[e]; |
2603 | int r = get_version(e, features, bl); | |
2604 | ceph_assert(r >= 0); | |
2605 | max_bytes -= bl.length(); | |
7c673cae FG |
2606 | } |
2607 | reply->oldest_map = first; | |
2608 | reply->newest_map = last; | |
2609 | mon->send_reply(op, reply); | |
2610 | return true; | |
2611 | } | |
2612 | ||
2613 | ||
2614 | // --------------------------- | |
2615 | // UPDATEs | |
2616 | ||
2617 | // failure -- | |
2618 | ||
11fdf7f2 | 2619 | bool OSDMonitor::check_source(MonOpRequestRef op, uuid_d fsid) { |
7c673cae | 2620 | // check permissions |
11fdf7f2 | 2621 | MonSession *session = op->get_session(); |
7c673cae FG |
2622 | if (!session) |
2623 | return true; | |
2624 | if (!session->is_capable("osd", MON_CAP_X)) { | |
2625 | dout(0) << "got MOSDFailure from entity with insufficient caps " | |
2626 | << session->caps << dendl; | |
2627 | return true; | |
2628 | } | |
2629 | if (fsid != mon->monmap->fsid) { | |
2630 | dout(0) << "check_source: on fsid " << fsid | |
2631 | << " != " << mon->monmap->fsid << dendl; | |
2632 | return true; | |
2633 | } | |
2634 | return false; | |
2635 | } | |
2636 | ||
2637 | ||
2638 | bool OSDMonitor::preprocess_failure(MonOpRequestRef op) | |
2639 | { | |
2640 | op->mark_osdmon_event(__func__); | |
2641 | MOSDFailure *m = static_cast<MOSDFailure*>(op->get_req()); | |
2642 | // who is target_osd | |
11fdf7f2 | 2643 | int badboy = m->get_target_osd(); |
7c673cae FG |
2644 | |
2645 | // check permissions | |
11fdf7f2 | 2646 | if (check_source(op, m->fsid)) |
7c673cae FG |
2647 | goto didit; |
2648 | ||
2649 | // first, verify the reporting host is valid | |
2650 | if (m->get_orig_source().is_osd()) { | |
2651 | int from = m->get_orig_source().num(); | |
2652 | if (!osdmap.exists(from) || | |
11fdf7f2 | 2653 | !osdmap.get_addrs(from).legacy_equals(m->get_orig_source_addrs()) || |
7c673cae | 2654 | (osdmap.is_down(from) && m->if_osd_failed())) { |
11fdf7f2 TL |
2655 | dout(5) << "preprocess_failure from dead osd." << from |
2656 | << ", ignoring" << dendl; | |
7c673cae FG |
2657 | send_incremental(op, m->get_epoch()+1); |
2658 | goto didit; | |
2659 | } | |
2660 | } | |
2661 | ||
2662 | ||
2663 | // weird? | |
2664 | if (osdmap.is_down(badboy)) { | |
11fdf7f2 TL |
2665 | dout(5) << "preprocess_failure dne(/dup?): osd." << m->get_target_osd() |
2666 | << " " << m->get_target_addrs() | |
2667 | << ", from " << m->get_orig_source() << dendl; | |
7c673cae FG |
2668 | if (m->get_epoch() < osdmap.get_epoch()) |
2669 | send_incremental(op, m->get_epoch()+1); | |
2670 | goto didit; | |
2671 | } | |
11fdf7f2 TL |
2672 | if (osdmap.get_addrs(badboy) != m->get_target_addrs()) { |
2673 | dout(5) << "preprocess_failure wrong osd: report osd." << m->get_target_osd() | |
2674 | << " " << m->get_target_addrs() | |
2675 | << " != map's " << osdmap.get_addrs(badboy) | |
2676 | << ", from " << m->get_orig_source() << dendl; | |
7c673cae FG |
2677 | if (m->get_epoch() < osdmap.get_epoch()) |
2678 | send_incremental(op, m->get_epoch()+1); | |
2679 | goto didit; | |
2680 | } | |
2681 | ||
2682 | // already reported? | |
2683 | if (osdmap.is_down(badboy) || | |
2684 | osdmap.get_up_from(badboy) > m->get_epoch()) { | |
11fdf7f2 TL |
2685 | dout(5) << "preprocess_failure dup/old: osd." << m->get_target_osd() |
2686 | << " " << m->get_target_addrs() | |
2687 | << ", from " << m->get_orig_source() << dendl; | |
7c673cae FG |
2688 | if (m->get_epoch() < osdmap.get_epoch()) |
2689 | send_incremental(op, m->get_epoch()+1); | |
2690 | goto didit; | |
2691 | } | |
2692 | ||
2693 | if (!can_mark_down(badboy)) { | |
11fdf7f2 TL |
2694 | dout(5) << "preprocess_failure ignoring report of osd." |
2695 | << m->get_target_osd() << " " << m->get_target_addrs() | |
2696 | << " from " << m->get_orig_source() << dendl; | |
7c673cae FG |
2697 | goto didit; |
2698 | } | |
2699 | ||
11fdf7f2 TL |
2700 | dout(10) << "preprocess_failure new: osd." << m->get_target_osd() |
2701 | << " " << m->get_target_addrs() | |
2702 | << ", from " << m->get_orig_source() << dendl; | |
7c673cae FG |
2703 | return false; |
2704 | ||
2705 | didit: | |
28e407b8 | 2706 | mon->no_reply(op); |
7c673cae FG |
2707 | return true; |
2708 | } | |
2709 | ||
2710 | class C_AckMarkedDown : public C_MonOp { | |
2711 | OSDMonitor *osdmon; | |
2712 | public: | |
2713 | C_AckMarkedDown( | |
2714 | OSDMonitor *osdmon, | |
2715 | MonOpRequestRef op) | |
2716 | : C_MonOp(op), osdmon(osdmon) {} | |
2717 | ||
eafe8130 TL |
2718 | void _finish(int r) override { |
2719 | if (r == 0) { | |
2720 | MOSDMarkMeDown *m = static_cast<MOSDMarkMeDown*>(op->get_req()); | |
2721 | osdmon->mon->send_reply( | |
2722 | op, | |
2723 | new MOSDMarkMeDown( | |
2724 | m->fsid, | |
2725 | m->target_osd, | |
2726 | m->target_addrs, | |
2727 | m->get_epoch(), | |
2728 | false)); // ACK itself does not request an ack | |
2729 | } else if (r == -EAGAIN) { | |
2730 | osdmon->dispatch(op); | |
2731 | } else { | |
2732 | ceph_abort_msgf("C_AckMarkedDown: unknown result %d", r); | |
2733 | } | |
7c673cae FG |
2734 | } |
2735 | ~C_AckMarkedDown() override { | |
2736 | } | |
2737 | }; | |
2738 | ||
2739 | bool OSDMonitor::preprocess_mark_me_down(MonOpRequestRef op) | |
2740 | { | |
2741 | op->mark_osdmon_event(__func__); | |
2742 | MOSDMarkMeDown *m = static_cast<MOSDMarkMeDown*>(op->get_req()); | |
11fdf7f2 | 2743 | int from = m->target_osd; |
7c673cae FG |
2744 | |
2745 | // check permissions | |
11fdf7f2 | 2746 | if (check_source(op, m->fsid)) |
7c673cae FG |
2747 | goto reply; |
2748 | ||
2749 | // first, verify the reporting host is valid | |
2750 | if (!m->get_orig_source().is_osd()) | |
2751 | goto reply; | |
2752 | ||
2753 | if (!osdmap.exists(from) || | |
2754 | osdmap.is_down(from) || | |
11fdf7f2 | 2755 | osdmap.get_addrs(from) != m->target_addrs) { |
7c673cae FG |
2756 | dout(5) << "preprocess_mark_me_down from dead osd." |
2757 | << from << ", ignoring" << dendl; | |
2758 | send_incremental(op, m->get_epoch()+1); | |
2759 | goto reply; | |
2760 | } | |
2761 | ||
2762 | // no down might be set | |
11fdf7f2 | 2763 | if (!can_mark_down(from)) |
7c673cae FG |
2764 | goto reply; |
2765 | ||
11fdf7f2 TL |
2766 | dout(10) << "MOSDMarkMeDown for: " << m->get_orig_source() |
2767 | << " " << m->target_addrs << dendl; | |
7c673cae FG |
2768 | return false; |
2769 | ||
2770 | reply: | |
2771 | if (m->request_ack) { | |
2772 | Context *c(new C_AckMarkedDown(this, op)); | |
2773 | c->complete(0); | |
2774 | } | |
2775 | return true; | |
2776 | } | |
2777 | ||
2778 | bool OSDMonitor::prepare_mark_me_down(MonOpRequestRef op) | |
2779 | { | |
2780 | op->mark_osdmon_event(__func__); | |
2781 | MOSDMarkMeDown *m = static_cast<MOSDMarkMeDown*>(op->get_req()); | |
11fdf7f2 | 2782 | int target_osd = m->target_osd; |
7c673cae | 2783 | |
11fdf7f2 TL |
2784 | ceph_assert(osdmap.is_up(target_osd)); |
2785 | ceph_assert(osdmap.get_addrs(target_osd) == m->target_addrs); | |
7c673cae FG |
2786 | |
2787 | mon->clog->info() << "osd." << target_osd << " marked itself down"; | |
2788 | pending_inc.new_state[target_osd] = CEPH_OSD_UP; | |
2789 | if (m->request_ack) | |
2790 | wait_for_finished_proposal(op, new C_AckMarkedDown(this, op)); | |
2791 | return true; | |
2792 | } | |
2793 | ||
2794 | bool OSDMonitor::can_mark_down(int i) | |
2795 | { | |
31f18b77 FG |
2796 | if (osdmap.is_nodown(i)) { |
2797 | dout(5) << __func__ << " osd." << i << " is marked as nodown, " | |
2798 | << "will not mark it down" << dendl; | |
7c673cae FG |
2799 | return false; |
2800 | } | |
31f18b77 | 2801 | |
7c673cae FG |
2802 | int num_osds = osdmap.get_num_osds(); |
2803 | if (num_osds == 0) { | |
31f18b77 | 2804 | dout(5) << __func__ << " no osds" << dendl; |
7c673cae FG |
2805 | return false; |
2806 | } | |
2807 | int up = osdmap.get_num_up_osds() - pending_inc.get_net_marked_down(&osdmap); | |
2808 | float up_ratio = (float)up / (float)num_osds; | |
11fdf7f2 | 2809 | if (up_ratio < g_conf()->mon_osd_min_up_ratio) { |
31f18b77 | 2810 | dout(2) << __func__ << " current up_ratio " << up_ratio << " < min " |
11fdf7f2 | 2811 | << g_conf()->mon_osd_min_up_ratio |
7c673cae FG |
2812 | << ", will not mark osd." << i << " down" << dendl; |
2813 | return false; | |
2814 | } | |
2815 | return true; | |
2816 | } | |
2817 | ||
2818 | bool OSDMonitor::can_mark_up(int i) | |
2819 | { | |
31f18b77 FG |
2820 | if (osdmap.is_noup(i)) { |
2821 | dout(5) << __func__ << " osd." << i << " is marked as noup, " | |
2822 | << "will not mark it up" << dendl; | |
7c673cae FG |
2823 | return false; |
2824 | } | |
31f18b77 | 2825 | |
7c673cae FG |
2826 | return true; |
2827 | } | |
2828 | ||
2829 | /** | |
2830 | * @note the parameter @p i apparently only exists here so we can output the | |
2831 | * osd's id on messages. | |
2832 | */ | |
2833 | bool OSDMonitor::can_mark_out(int i) | |
2834 | { | |
31f18b77 FG |
2835 | if (osdmap.is_noout(i)) { |
2836 | dout(5) << __func__ << " osd." << i << " is marked as noout, " | |
2837 | << "will not mark it out" << dendl; | |
2838 | return false; | |
2839 | } | |
2840 | ||
7c673cae FG |
2841 | int num_osds = osdmap.get_num_osds(); |
2842 | if (num_osds == 0) { | |
2843 | dout(5) << __func__ << " no osds" << dendl; | |
2844 | return false; | |
2845 | } | |
2846 | int in = osdmap.get_num_in_osds() - pending_inc.get_net_marked_out(&osdmap); | |
2847 | float in_ratio = (float)in / (float)num_osds; | |
11fdf7f2 | 2848 | if (in_ratio < g_conf()->mon_osd_min_in_ratio) { |
7c673cae FG |
2849 | if (i >= 0) |
2850 | dout(5) << __func__ << " current in_ratio " << in_ratio << " < min " | |
11fdf7f2 | 2851 | << g_conf()->mon_osd_min_in_ratio |
7c673cae FG |
2852 | << ", will not mark osd." << i << " out" << dendl; |
2853 | else | |
2854 | dout(5) << __func__ << " current in_ratio " << in_ratio << " < min " | |
11fdf7f2 | 2855 | << g_conf()->mon_osd_min_in_ratio |
7c673cae FG |
2856 | << ", will not mark osds out" << dendl; |
2857 | return false; | |
2858 | } | |
2859 | ||
2860 | return true; | |
2861 | } | |
2862 | ||
2863 | bool OSDMonitor::can_mark_in(int i) | |
2864 | { | |
31f18b77 FG |
2865 | if (osdmap.is_noin(i)) { |
2866 | dout(5) << __func__ << " osd." << i << " is marked as noin, " | |
2867 | << "will not mark it in" << dendl; | |
7c673cae FG |
2868 | return false; |
2869 | } | |
31f18b77 | 2870 | |
7c673cae FG |
2871 | return true; |
2872 | } | |
2873 | ||
2874 | bool OSDMonitor::check_failures(utime_t now) | |
2875 | { | |
2876 | bool found_failure = false; | |
2877 | for (map<int,failure_info_t>::iterator p = failure_info.begin(); | |
2878 | p != failure_info.end(); | |
2879 | ++p) { | |
2880 | if (can_mark_down(p->first)) { | |
2881 | found_failure |= check_failure(now, p->first, p->second); | |
2882 | } | |
2883 | } | |
2884 | return found_failure; | |
2885 | } | |
2886 | ||
2887 | bool OSDMonitor::check_failure(utime_t now, int target_osd, failure_info_t& fi) | |
2888 | { | |
2889 | // already pending failure? | |
2890 | if (pending_inc.new_state.count(target_osd) && | |
2891 | pending_inc.new_state[target_osd] & CEPH_OSD_UP) { | |
2892 | dout(10) << " already pending failure" << dendl; | |
2893 | return true; | |
2894 | } | |
2895 | ||
2896 | set<string> reporters_by_subtree; | |
11fdf7f2 TL |
2897 | auto reporter_subtree_level = g_conf().get_val<string>("mon_osd_reporter_subtree_level"); |
2898 | utime_t orig_grace(g_conf()->osd_heartbeat_grace, 0); | |
7c673cae FG |
2899 | utime_t max_failed_since = fi.get_failed_since(); |
2900 | utime_t failed_for = now - max_failed_since; | |
2901 | ||
2902 | utime_t grace = orig_grace; | |
2903 | double my_grace = 0, peer_grace = 0; | |
2904 | double decay_k = 0; | |
11fdf7f2 TL |
2905 | if (g_conf()->mon_osd_adjust_heartbeat_grace) { |
2906 | double halflife = (double)g_conf()->mon_osd_laggy_halflife; | |
7c673cae FG |
2907 | decay_k = ::log(.5) / halflife; |
2908 | ||
2909 | // scale grace period based on historical probability of 'lagginess' | |
2910 | // (false positive failures due to slowness). | |
2911 | const osd_xinfo_t& xi = osdmap.get_xinfo(target_osd); | |
2912 | double decay = exp((double)failed_for * decay_k); | |
2913 | dout(20) << " halflife " << halflife << " decay_k " << decay_k | |
2914 | << " failed_for " << failed_for << " decay " << decay << dendl; | |
2915 | my_grace = decay * (double)xi.laggy_interval * xi.laggy_probability; | |
2916 | grace += my_grace; | |
2917 | } | |
2918 | ||
2919 | // consider the peers reporting a failure a proxy for a potential | |
2920 | // 'subcluster' over the overall cluster that is similarly | |
2921 | // laggy. this is clearly not true in all cases, but will sometimes | |
2922 | // help us localize the grace correction to a subset of the system | |
2923 | // (say, a rack with a bad switch) that is unhappy. | |
11fdf7f2 | 2924 | ceph_assert(fi.reporters.size()); |
eafe8130 | 2925 | for (auto p = fi.reporters.begin(); p != fi.reporters.end();) { |
7c673cae FG |
2926 | // get the parent bucket whose type matches with "reporter_subtree_level". |
2927 | // fall back to OSD if the level doesn't exist. | |
eafe8130 TL |
2928 | if (osdmap.exists(p->first)) { |
2929 | auto reporter_loc = osdmap.crush->get_full_location(p->first); | |
2930 | if (auto iter = reporter_loc.find(reporter_subtree_level); | |
2931 | iter == reporter_loc.end()) { | |
2932 | reporters_by_subtree.insert("osd." + to_string(p->first)); | |
2933 | } else { | |
2934 | reporters_by_subtree.insert(iter->second); | |
2935 | } | |
2936 | if (g_conf()->mon_osd_adjust_heartbeat_grace) { | |
2937 | const osd_xinfo_t& xi = osdmap.get_xinfo(p->first); | |
2938 | utime_t elapsed = now - xi.down_stamp; | |
2939 | double decay = exp((double)elapsed * decay_k); | |
2940 | peer_grace += decay * (double)xi.laggy_interval * xi.laggy_probability; | |
2941 | } | |
2942 | ++p; | |
7c673cae | 2943 | } else { |
eafe8130 TL |
2944 | fi.cancel_report(p->first);; |
2945 | p = fi.reporters.erase(p); | |
7c673cae FG |
2946 | } |
2947 | } | |
2948 | ||
11fdf7f2 | 2949 | if (g_conf()->mon_osd_adjust_heartbeat_grace) { |
7c673cae FG |
2950 | peer_grace /= (double)fi.reporters.size(); |
2951 | grace += peer_grace; | |
2952 | } | |
2953 | ||
2954 | dout(10) << " osd." << target_osd << " has " | |
2955 | << fi.reporters.size() << " reporters, " | |
2956 | << grace << " grace (" << orig_grace << " + " << my_grace | |
2957 | << " + " << peer_grace << "), max_failed_since " << max_failed_since | |
2958 | << dendl; | |
2959 | ||
2960 | if (failed_for >= grace && | |
11fdf7f2 | 2961 | reporters_by_subtree.size() >= g_conf().get_val<uint64_t>("mon_osd_min_down_reporters")) { |
7c673cae FG |
2962 | dout(1) << " we have enough reporters to mark osd." << target_osd |
2963 | << " down" << dendl; | |
2964 | pending_inc.new_state[target_osd] = CEPH_OSD_UP; | |
2965 | ||
31f18b77 FG |
2966 | mon->clog->info() << "osd." << target_osd << " failed (" |
2967 | << osdmap.crush->get_full_location_ordered_string( | |
2968 | target_osd) | |
2969 | << ") (" | |
2970 | << (int)reporters_by_subtree.size() | |
2971 | << " reporters from different " | |
7c673cae FG |
2972 | << reporter_subtree_level << " after " |
2973 | << failed_for << " >= grace " << grace << ")"; | |
2974 | return true; | |
2975 | } | |
2976 | return false; | |
2977 | } | |
2978 | ||
224ce89b | 2979 | void OSDMonitor::force_failure(int target_osd, int by) |
7c673cae FG |
2980 | { |
2981 | // already pending failure? | |
2982 | if (pending_inc.new_state.count(target_osd) && | |
2983 | pending_inc.new_state[target_osd] & CEPH_OSD_UP) { | |
2984 | dout(10) << " already pending failure" << dendl; | |
2985 | return; | |
2986 | } | |
2987 | ||
2988 | dout(1) << " we're forcing failure of osd." << target_osd << dendl; | |
2989 | pending_inc.new_state[target_osd] = CEPH_OSD_UP; | |
2990 | ||
31f18b77 FG |
2991 | mon->clog->info() << "osd." << target_osd << " failed (" |
2992 | << osdmap.crush->get_full_location_ordered_string(target_osd) | |
2993 | << ") (connection refused reported by osd." << by << ")"; | |
7c673cae FG |
2994 | return; |
2995 | } | |
2996 | ||
2997 | bool OSDMonitor::prepare_failure(MonOpRequestRef op) | |
2998 | { | |
2999 | op->mark_osdmon_event(__func__); | |
3000 | MOSDFailure *m = static_cast<MOSDFailure*>(op->get_req()); | |
11fdf7f2 TL |
3001 | dout(1) << "prepare_failure osd." << m->get_target_osd() |
3002 | << " " << m->get_target_addrs() | |
3003 | << " from " << m->get_orig_source() | |
7c673cae FG |
3004 | << " is reporting failure:" << m->if_osd_failed() << dendl; |
3005 | ||
11fdf7f2 | 3006 | int target_osd = m->get_target_osd(); |
7c673cae | 3007 | int reporter = m->get_orig_source().num(); |
11fdf7f2 TL |
3008 | ceph_assert(osdmap.is_up(target_osd)); |
3009 | ceph_assert(osdmap.get_addrs(target_osd) == m->get_target_addrs()); | |
7c673cae | 3010 | |
eafe8130 TL |
3011 | mon->no_reply(op); |
3012 | ||
7c673cae FG |
3013 | if (m->if_osd_failed()) { |
3014 | // calculate failure time | |
3015 | utime_t now = ceph_clock_now(); | |
3016 | utime_t failed_since = | |
3017 | m->get_recv_stamp() - utime_t(m->failed_for, 0); | |
3018 | ||
3019 | // add a report | |
3020 | if (m->is_immediate()) { | |
11fdf7f2 TL |
3021 | mon->clog->debug() << "osd." << m->get_target_osd() |
3022 | << " reported immediately failed by " | |
3023 | << m->get_orig_source(); | |
224ce89b | 3024 | force_failure(target_osd, reporter); |
7c673cae FG |
3025 | return true; |
3026 | } | |
11fdf7f2 TL |
3027 | mon->clog->debug() << "osd." << m->get_target_osd() << " reported failed by " |
3028 | << m->get_orig_source(); | |
7c673cae FG |
3029 | |
3030 | failure_info_t& fi = failure_info[target_osd]; | |
3031 | MonOpRequestRef old_op = fi.add_report(reporter, failed_since, op); | |
3032 | if (old_op) { | |
3033 | mon->no_reply(old_op); | |
3034 | } | |
3035 | ||
3036 | return check_failure(now, target_osd, fi); | |
3037 | } else { | |
3038 | // remove the report | |
11fdf7f2 TL |
3039 | mon->clog->debug() << "osd." << m->get_target_osd() |
3040 | << " failure report canceled by " | |
3041 | << m->get_orig_source(); | |
7c673cae FG |
3042 | if (failure_info.count(target_osd)) { |
3043 | failure_info_t& fi = failure_info[target_osd]; | |
3044 | MonOpRequestRef report_op = fi.cancel_report(reporter); | |
3045 | if (report_op) { | |
3046 | mon->no_reply(report_op); | |
3047 | } | |
3048 | if (fi.reporters.empty()) { | |
3049 | dout(10) << " removing last failure_info for osd." << target_osd | |
3050 | << dendl; | |
3051 | failure_info.erase(target_osd); | |
3052 | } else { | |
3053 | dout(10) << " failure_info for osd." << target_osd << " now " | |
3054 | << fi.reporters.size() << " reporters" << dendl; | |
3055 | } | |
3056 | } else { | |
3057 | dout(10) << " no failure_info for osd." << target_osd << dendl; | |
3058 | } | |
7c673cae FG |
3059 | } |
3060 | ||
3061 | return false; | |
3062 | } | |
3063 | ||
3064 | void OSDMonitor::process_failures() | |
3065 | { | |
3066 | map<int,failure_info_t>::iterator p = failure_info.begin(); | |
3067 | while (p != failure_info.end()) { | |
3068 | if (osdmap.is_up(p->first)) { | |
3069 | ++p; | |
3070 | } else { | |
3071 | dout(10) << "process_failures osd." << p->first << dendl; | |
3072 | list<MonOpRequestRef> ls; | |
3073 | p->second.take_report_messages(ls); | |
3074 | failure_info.erase(p++); | |
3075 | ||
3076 | while (!ls.empty()) { | |
3077 | MonOpRequestRef o = ls.front(); | |
3078 | if (o) { | |
3079 | o->mark_event(__func__); | |
3080 | MOSDFailure *m = o->get_req<MOSDFailure>(); | |
3081 | send_latest(o, m->get_epoch()); | |
28e407b8 | 3082 | mon->no_reply(o); |
7c673cae FG |
3083 | } |
3084 | ls.pop_front(); | |
3085 | } | |
3086 | } | |
3087 | } | |
3088 | } | |
3089 | ||
3090 | void OSDMonitor::take_all_failures(list<MonOpRequestRef>& ls) | |
3091 | { | |
3092 | dout(10) << __func__ << " on " << failure_info.size() << " osds" << dendl; | |
3093 | ||
3094 | for (map<int,failure_info_t>::iterator p = failure_info.begin(); | |
3095 | p != failure_info.end(); | |
3096 | ++p) { | |
3097 | p->second.take_report_messages(ls); | |
3098 | } | |
3099 | failure_info.clear(); | |
3100 | } | |
3101 | ||
3102 | ||
3103 | // boot -- | |
3104 | ||
3105 | bool OSDMonitor::preprocess_boot(MonOpRequestRef op) | |
3106 | { | |
3107 | op->mark_osdmon_event(__func__); | |
3108 | MOSDBoot *m = static_cast<MOSDBoot*>(op->get_req()); | |
3109 | int from = m->get_orig_source_inst().name.num(); | |
3110 | ||
3111 | // check permissions, ignore if failed (no response expected) | |
11fdf7f2 | 3112 | MonSession *session = op->get_session(); |
7c673cae FG |
3113 | if (!session) |
3114 | goto ignore; | |
3115 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3116 | dout(0) << "got preprocess_boot message from entity with insufficient caps" | |
11fdf7f2 | 3117 | << session->caps << dendl; |
7c673cae FG |
3118 | goto ignore; |
3119 | } | |
3120 | ||
11fdf7f2 TL |
3121 | if (m->sb.cluster_fsid != mon->monmap->fsid) { |
3122 | dout(0) << "preprocess_boot on fsid " << m->sb.cluster_fsid | |
3123 | << " != " << mon->monmap->fsid << dendl; | |
7c673cae FG |
3124 | goto ignore; |
3125 | } | |
3126 | ||
11fdf7f2 TL |
3127 | if (m->get_orig_source_inst().addr.is_blank_ip()) { |
3128 | dout(0) << "preprocess_boot got blank addr for " << m->get_orig_source_inst() << dendl; | |
7c673cae FG |
3129 | goto ignore; |
3130 | } | |
3131 | ||
11fdf7f2 | 3132 | ceph_assert(m->get_orig_source_inst().name.is_osd()); |
7c673cae | 3133 | |
11fdf7f2 TL |
3134 | // force all osds to have gone through luminous prior to upgrade to nautilus |
3135 | { | |
3136 | vector<string> missing; | |
3137 | if (!HAVE_FEATURE(m->osd_features, SERVER_LUMINOUS)) { | |
3138 | missing.push_back("CEPH_FEATURE_SERVER_LUMINOUS"); | |
3139 | } | |
3140 | if (!HAVE_FEATURE(m->osd_features, SERVER_JEWEL)) { | |
3141 | missing.push_back("CEPH_FEATURE_SERVER_JEWEL"); | |
3142 | } | |
3143 | if (!HAVE_FEATURE(m->osd_features, SERVER_KRAKEN)) { | |
3144 | missing.push_back("CEPH_FEATURE_SERVER_KRAKEN"); | |
3145 | } | |
3146 | if (!HAVE_FEATURE(m->osd_features, OSD_RECOVERY_DELETES)) { | |
3147 | missing.push_back("CEPH_FEATURE_OSD_RECOVERY_DELETES"); | |
3148 | } | |
7c673cae | 3149 | |
11fdf7f2 TL |
3150 | if (!missing.empty()) { |
3151 | using std::experimental::make_ostream_joiner; | |
7c673cae | 3152 | |
11fdf7f2 TL |
3153 | stringstream ss; |
3154 | copy(begin(missing), end(missing), make_ostream_joiner(ss, ";")); | |
c07f9fc5 | 3155 | |
11fdf7f2 TL |
3156 | mon->clog->info() << "disallowing boot of OSD " |
3157 | << m->get_orig_source_inst() | |
3158 | << " because the osd lacks " << ss.str(); | |
7c673cae FG |
3159 | goto ignore; |
3160 | } | |
3161 | } | |
3162 | ||
11fdf7f2 TL |
3163 | // make sure upgrades stop at nautilus |
3164 | if (HAVE_FEATURE(m->osd_features, SERVER_O) && | |
3165 | osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS) { | |
3166 | mon->clog->info() << "disallowing boot of post-nautilus OSD " | |
7c673cae | 3167 | << m->get_orig_source_inst() |
11fdf7f2 | 3168 | << " because require_osd_release < nautilus"; |
7c673cae FG |
3169 | goto ignore; |
3170 | } | |
3171 | ||
f64942e4 AA |
3172 | // The release check here is required because for OSD_PGLOG_HARDLIMIT, |
3173 | // we are reusing a jewel feature bit that was retired in luminous. | |
3174 | if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS && | |
3175 | osdmap.test_flag(CEPH_OSDMAP_PGLOG_HARDLIMIT) && | |
3176 | !(m->osd_features & CEPH_FEATURE_OSD_PGLOG_HARDLIMIT)) { | |
3177 | mon->clog->info() << "disallowing boot of OSD " | |
3178 | << m->get_orig_source_inst() | |
3179 | << " because 'pglog_hardlimit' osdmap flag is set and OSD lacks the OSD_PGLOG_HARDLIMIT feature"; | |
3180 | goto ignore; | |
3181 | } | |
3182 | ||
7c673cae FG |
3183 | // already booted? |
3184 | if (osdmap.is_up(from) && | |
11fdf7f2 TL |
3185 | osdmap.get_addrs(from).legacy_equals(m->get_orig_source_addrs()) && |
3186 | osdmap.get_cluster_addrs(from).legacy_equals(m->cluster_addrs)) { | |
7c673cae | 3187 | // yup. |
11fdf7f2 TL |
3188 | dout(7) << "preprocess_boot dup from " << m->get_orig_source() |
3189 | << " " << m->get_orig_source_addrs() | |
3190 | << " =~ " << osdmap.get_addrs(from) << dendl; | |
7c673cae FG |
3191 | _booted(op, false); |
3192 | return true; | |
3193 | } | |
3194 | ||
3195 | if (osdmap.exists(from) && | |
3196 | !osdmap.get_uuid(from).is_zero() && | |
3197 | osdmap.get_uuid(from) != m->sb.osd_fsid) { | |
3198 | dout(7) << __func__ << " from " << m->get_orig_source_inst() | |
3199 | << " clashes with existing osd: different fsid" | |
3200 | << " (ours: " << osdmap.get_uuid(from) | |
3201 | << " ; theirs: " << m->sb.osd_fsid << ")" << dendl; | |
3202 | goto ignore; | |
3203 | } | |
3204 | ||
3205 | if (osdmap.exists(from) && | |
3206 | osdmap.get_info(from).up_from > m->version && | |
11fdf7f2 TL |
3207 | osdmap.get_most_recent_addrs(from).legacy_equals( |
3208 | m->get_orig_source_addrs())) { | |
7c673cae FG |
3209 | dout(7) << "prepare_boot msg from before last up_from, ignoring" << dendl; |
3210 | send_latest(op, m->sb.current_epoch+1); | |
3211 | return true; | |
3212 | } | |
3213 | ||
3214 | // noup? | |
3215 | if (!can_mark_up(from)) { | |
3216 | dout(7) << "preprocess_boot ignoring boot from " << m->get_orig_source_inst() << dendl; | |
3217 | send_latest(op, m->sb.current_epoch+1); | |
3218 | return true; | |
3219 | } | |
3220 | ||
3221 | dout(10) << "preprocess_boot from " << m->get_orig_source_inst() << dendl; | |
3222 | return false; | |
3223 | ||
3224 | ignore: | |
3225 | return true; | |
3226 | } | |
3227 | ||
3228 | bool OSDMonitor::prepare_boot(MonOpRequestRef op) | |
3229 | { | |
3230 | op->mark_osdmon_event(__func__); | |
3231 | MOSDBoot *m = static_cast<MOSDBoot*>(op->get_req()); | |
11fdf7f2 TL |
3232 | dout(7) << __func__ << " from " << m->get_source() |
3233 | << " sb " << m->sb | |
3234 | << " client_addrs" << m->get_connection()->get_peer_addrs() | |
3235 | << " cluster_addrs " << m->cluster_addrs | |
3236 | << " hb_back_addrs " << m->hb_back_addrs | |
3237 | << " hb_front_addrs " << m->hb_front_addrs | |
7c673cae FG |
3238 | << dendl; |
3239 | ||
11fdf7f2 | 3240 | ceph_assert(m->get_orig_source().is_osd()); |
7c673cae FG |
3241 | int from = m->get_orig_source().num(); |
3242 | ||
3243 | // does this osd exist? | |
3244 | if (from >= osdmap.get_max_osd()) { | |
3245 | dout(1) << "boot from osd." << from << " >= max_osd " | |
3246 | << osdmap.get_max_osd() << dendl; | |
3247 | return false; | |
3248 | } | |
3249 | ||
3250 | int oldstate = osdmap.exists(from) ? osdmap.get_state(from) : CEPH_OSD_NEW; | |
3251 | if (pending_inc.new_state.count(from)) | |
3252 | oldstate ^= pending_inc.new_state[from]; | |
3253 | ||
3254 | // already up? mark down first? | |
3255 | if (osdmap.is_up(from)) { | |
11fdf7f2 TL |
3256 | dout(7) << __func__ << " was up, first marking down osd." << from << " " |
3257 | << osdmap.get_addrs(from) << dendl; | |
7c673cae | 3258 | // preprocess should have caught these; if not, assert. |
11fdf7f2 TL |
3259 | ceph_assert(!osdmap.get_addrs(from).legacy_equals( |
3260 | m->get_orig_source_addrs()) || | |
3261 | !osdmap.get_cluster_addrs(from).legacy_equals(m->cluster_addrs)); | |
3262 | ceph_assert(osdmap.get_uuid(from) == m->sb.osd_fsid); | |
7c673cae FG |
3263 | |
3264 | if (pending_inc.new_state.count(from) == 0 || | |
3265 | (pending_inc.new_state[from] & CEPH_OSD_UP) == 0) { | |
3266 | // mark previous guy down | |
3267 | pending_inc.new_state[from] = CEPH_OSD_UP; | |
3268 | } | |
3269 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
3270 | } else if (pending_inc.new_up_client.count(from)) { | |
3271 | // already prepared, just wait | |
3272 | dout(7) << __func__ << " already prepared, waiting on " | |
3273 | << m->get_orig_source_addr() << dendl; | |
3274 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
3275 | } else { | |
3276 | // mark new guy up. | |
11fdf7f2 TL |
3277 | pending_inc.new_up_client[from] = m->get_orig_source_addrs(); |
3278 | pending_inc.new_up_cluster[from] = m->cluster_addrs; | |
3279 | pending_inc.new_hb_back_up[from] = m->hb_back_addrs; | |
3280 | pending_inc.new_hb_front_up[from] = m->hb_front_addrs; | |
7c673cae FG |
3281 | |
3282 | down_pending_out.erase(from); // if any | |
3283 | ||
3284 | if (m->sb.weight) | |
3285 | osd_weight[from] = m->sb.weight; | |
3286 | ||
3287 | // set uuid? | |
3288 | dout(10) << " setting osd." << from << " uuid to " << m->sb.osd_fsid | |
3289 | << dendl; | |
3290 | if (!osdmap.exists(from) || osdmap.get_uuid(from) != m->sb.osd_fsid) { | |
3291 | // preprocess should have caught this; if not, assert. | |
11fdf7f2 | 3292 | ceph_assert(!osdmap.exists(from) || osdmap.get_uuid(from).is_zero()); |
7c673cae FG |
3293 | pending_inc.new_uuid[from] = m->sb.osd_fsid; |
3294 | } | |
3295 | ||
3296 | // fresh osd? | |
3297 | if (m->sb.newest_map == 0 && osdmap.exists(from)) { | |
3298 | const osd_info_t& i = osdmap.get_info(from); | |
3299 | if (i.up_from > i.lost_at) { | |
3300 | dout(10) << " fresh osd; marking lost_at too" << dendl; | |
3301 | pending_inc.new_lost[from] = osdmap.get_epoch(); | |
3302 | } | |
3303 | } | |
3304 | ||
3305 | // metadata | |
3306 | bufferlist osd_metadata; | |
11fdf7f2 | 3307 | encode(m->metadata, osd_metadata); |
7c673cae | 3308 | pending_metadata[from] = osd_metadata; |
31f18b77 | 3309 | pending_metadata_rm.erase(from); |
7c673cae FG |
3310 | |
3311 | // adjust last clean unmount epoch? | |
3312 | const osd_info_t& info = osdmap.get_info(from); | |
3313 | dout(10) << " old osd_info: " << info << dendl; | |
3314 | if (m->sb.mounted > info.last_clean_begin || | |
3315 | (m->sb.mounted == info.last_clean_begin && | |
3316 | m->sb.clean_thru > info.last_clean_end)) { | |
3317 | epoch_t begin = m->sb.mounted; | |
3318 | epoch_t end = m->sb.clean_thru; | |
3319 | ||
3320 | dout(10) << __func__ << " osd." << from << " last_clean_interval " | |
3321 | << "[" << info.last_clean_begin << "," << info.last_clean_end | |
3322 | << ") -> [" << begin << "-" << end << ")" | |
3323 | << dendl; | |
3324 | pending_inc.new_last_clean_interval[from] = | |
3325 | pair<epoch_t,epoch_t>(begin, end); | |
3326 | } | |
3327 | ||
3328 | osd_xinfo_t xi = osdmap.get_xinfo(from); | |
3329 | if (m->boot_epoch == 0) { | |
11fdf7f2 TL |
3330 | xi.laggy_probability *= (1.0 - g_conf()->mon_osd_laggy_weight); |
3331 | xi.laggy_interval *= (1.0 - g_conf()->mon_osd_laggy_weight); | |
7c673cae FG |
3332 | dout(10) << " not laggy, new xi " << xi << dendl; |
3333 | } else { | |
3334 | if (xi.down_stamp.sec()) { | |
3335 | int interval = ceph_clock_now().sec() - | |
3336 | xi.down_stamp.sec(); | |
11fdf7f2 TL |
3337 | if (g_conf()->mon_osd_laggy_max_interval && |
3338 | (interval > g_conf()->mon_osd_laggy_max_interval)) { | |
3339 | interval = g_conf()->mon_osd_laggy_max_interval; | |
7c673cae FG |
3340 | } |
3341 | xi.laggy_interval = | |
11fdf7f2 TL |
3342 | interval * g_conf()->mon_osd_laggy_weight + |
3343 | xi.laggy_interval * (1.0 - g_conf()->mon_osd_laggy_weight); | |
7c673cae FG |
3344 | } |
3345 | xi.laggy_probability = | |
11fdf7f2 TL |
3346 | g_conf()->mon_osd_laggy_weight + |
3347 | xi.laggy_probability * (1.0 - g_conf()->mon_osd_laggy_weight); | |
7c673cae FG |
3348 | dout(10) << " laggy, now xi " << xi << dendl; |
3349 | } | |
3350 | ||
3351 | // set features shared by the osd | |
3352 | if (m->osd_features) | |
3353 | xi.features = m->osd_features; | |
3354 | else | |
3355 | xi.features = m->get_connection()->get_features(); | |
3356 | ||
3357 | // mark in? | |
11fdf7f2 | 3358 | if ((g_conf()->mon_osd_auto_mark_auto_out_in && |
7c673cae | 3359 | (oldstate & CEPH_OSD_AUTOOUT)) || |
11fdf7f2 TL |
3360 | (g_conf()->mon_osd_auto_mark_new_in && (oldstate & CEPH_OSD_NEW)) || |
3361 | (g_conf()->mon_osd_auto_mark_in)) { | |
7c673cae FG |
3362 | if (can_mark_in(from)) { |
3363 | if (osdmap.osd_xinfo[from].old_weight > 0) { | |
3364 | pending_inc.new_weight[from] = osdmap.osd_xinfo[from].old_weight; | |
3365 | xi.old_weight = 0; | |
3366 | } else { | |
3367 | pending_inc.new_weight[from] = CEPH_OSD_IN; | |
3368 | } | |
3369 | } else { | |
3370 | dout(7) << __func__ << " NOIN set, will not mark in " | |
3371 | << m->get_orig_source_addr() << dendl; | |
3372 | } | |
3373 | } | |
3374 | ||
3375 | pending_inc.new_xinfo[from] = xi; | |
3376 | ||
3377 | // wait | |
3378 | wait_for_finished_proposal(op, new C_Booted(this, op)); | |
3379 | } | |
3380 | return true; | |
3381 | } | |
3382 | ||
3383 | void OSDMonitor::_booted(MonOpRequestRef op, bool logit) | |
3384 | { | |
3385 | op->mark_osdmon_event(__func__); | |
3386 | MOSDBoot *m = static_cast<MOSDBoot*>(op->get_req()); | |
3387 | dout(7) << "_booted " << m->get_orig_source_inst() | |
3388 | << " w " << m->sb.weight << " from " << m->sb.current_epoch << dendl; | |
3389 | ||
3390 | if (logit) { | |
11fdf7f2 TL |
3391 | mon->clog->info() << m->get_source() << " " << m->get_orig_source_addrs() |
3392 | << " boot"; | |
7c673cae FG |
3393 | } |
3394 | ||
3395 | send_latest(op, m->sb.current_epoch+1); | |
3396 | } | |
3397 | ||
3398 | ||
3399 | // ------------- | |
3400 | // full | |
3401 | ||
3402 | bool OSDMonitor::preprocess_full(MonOpRequestRef op) | |
3403 | { | |
3404 | op->mark_osdmon_event(__func__); | |
3405 | MOSDFull *m = static_cast<MOSDFull*>(op->get_req()); | |
3406 | int from = m->get_orig_source().num(); | |
3407 | set<string> state; | |
3408 | unsigned mask = CEPH_OSD_NEARFULL | CEPH_OSD_BACKFILLFULL | CEPH_OSD_FULL; | |
3409 | ||
3410 | // check permissions, ignore if failed | |
11fdf7f2 | 3411 | MonSession *session = op->get_session(); |
7c673cae FG |
3412 | if (!session) |
3413 | goto ignore; | |
3414 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3415 | dout(0) << "MOSDFull from entity with insufficient privileges:" | |
3416 | << session->caps << dendl; | |
3417 | goto ignore; | |
3418 | } | |
3419 | ||
3420 | // ignore a full message from the osd instance that already went down | |
3421 | if (!osdmap.exists(from)) { | |
3422 | dout(7) << __func__ << " ignoring full message from nonexistent " | |
3423 | << m->get_orig_source_inst() << dendl; | |
3424 | goto ignore; | |
3425 | } | |
3426 | if ((!osdmap.is_up(from) && | |
11fdf7f2 TL |
3427 | osdmap.get_most_recent_addrs(from).legacy_equals( |
3428 | m->get_orig_source_addrs())) || | |
7c673cae | 3429 | (osdmap.is_up(from) && |
11fdf7f2 | 3430 | !osdmap.get_addrs(from).legacy_equals(m->get_orig_source_addrs()))) { |
7c673cae FG |
3431 | dout(7) << __func__ << " ignoring full message from down " |
3432 | << m->get_orig_source_inst() << dendl; | |
3433 | goto ignore; | |
3434 | } | |
3435 | ||
3436 | OSDMap::calc_state_set(osdmap.get_state(from), state); | |
3437 | ||
3438 | if ((osdmap.get_state(from) & mask) == m->state) { | |
3439 | dout(7) << __func__ << " state already " << state << " for osd." << from | |
3440 | << " " << m->get_orig_source_inst() << dendl; | |
3441 | _reply_map(op, m->version); | |
3442 | goto ignore; | |
3443 | } | |
3444 | ||
3445 | dout(10) << __func__ << " want state " << state << " for osd." << from | |
3446 | << " " << m->get_orig_source_inst() << dendl; | |
3447 | return false; | |
3448 | ||
3449 | ignore: | |
3450 | return true; | |
3451 | } | |
3452 | ||
3453 | bool OSDMonitor::prepare_full(MonOpRequestRef op) | |
3454 | { | |
3455 | op->mark_osdmon_event(__func__); | |
3456 | const MOSDFull *m = static_cast<MOSDFull*>(op->get_req()); | |
3457 | const int from = m->get_orig_source().num(); | |
3458 | ||
3459 | const unsigned mask = CEPH_OSD_NEARFULL | CEPH_OSD_BACKFILLFULL | CEPH_OSD_FULL; | |
3460 | const unsigned want_state = m->state & mask; // safety first | |
3461 | ||
3462 | unsigned cur_state = osdmap.get_state(from); | |
3463 | auto p = pending_inc.new_state.find(from); | |
3464 | if (p != pending_inc.new_state.end()) { | |
3465 | cur_state ^= p->second; | |
3466 | } | |
3467 | cur_state &= mask; | |
3468 | ||
3469 | set<string> want_state_set, cur_state_set; | |
3470 | OSDMap::calc_state_set(want_state, want_state_set); | |
3471 | OSDMap::calc_state_set(cur_state, cur_state_set); | |
3472 | ||
3473 | if (cur_state != want_state) { | |
3474 | if (p != pending_inc.new_state.end()) { | |
3475 | p->second &= ~mask; | |
3476 | } else { | |
3477 | pending_inc.new_state[from] = 0; | |
3478 | } | |
3479 | pending_inc.new_state[from] |= (osdmap.get_state(from) & mask) ^ want_state; | |
3480 | dout(7) << __func__ << " osd." << from << " " << cur_state_set | |
3481 | << " -> " << want_state_set << dendl; | |
3482 | } else { | |
3483 | dout(7) << __func__ << " osd." << from << " " << cur_state_set | |
3484 | << " = wanted " << want_state_set << ", just waiting" << dendl; | |
3485 | } | |
3486 | ||
3487 | wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->version)); | |
3488 | return true; | |
3489 | } | |
3490 | ||
3491 | // ------------- | |
3492 | // alive | |
3493 | ||
3494 | bool OSDMonitor::preprocess_alive(MonOpRequestRef op) | |
3495 | { | |
3496 | op->mark_osdmon_event(__func__); | |
3497 | MOSDAlive *m = static_cast<MOSDAlive*>(op->get_req()); | |
3498 | int from = m->get_orig_source().num(); | |
3499 | ||
3500 | // check permissions, ignore if failed | |
11fdf7f2 | 3501 | MonSession *session = op->get_session(); |
7c673cae FG |
3502 | if (!session) |
3503 | goto ignore; | |
3504 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3505 | dout(0) << "attempt to send MOSDAlive from entity with insufficient privileges:" | |
3506 | << session->caps << dendl; | |
3507 | goto ignore; | |
3508 | } | |
3509 | ||
3510 | if (!osdmap.is_up(from) || | |
11fdf7f2 TL |
3511 | !osdmap.get_addrs(from).legacy_equals(m->get_orig_source_addrs())) { |
3512 | dout(7) << "preprocess_alive ignoring alive message from down " | |
3513 | << m->get_orig_source() << " " << m->get_orig_source_addrs() | |
3514 | << dendl; | |
7c673cae FG |
3515 | goto ignore; |
3516 | } | |
3517 | ||
3518 | if (osdmap.get_up_thru(from) >= m->want) { | |
3519 | // yup. | |
3520 | dout(7) << "preprocess_alive want up_thru " << m->want << " dup from " << m->get_orig_source_inst() << dendl; | |
3521 | _reply_map(op, m->version); | |
3522 | return true; | |
3523 | } | |
3524 | ||
3525 | dout(10) << "preprocess_alive want up_thru " << m->want | |
3526 | << " from " << m->get_orig_source_inst() << dendl; | |
3527 | return false; | |
3528 | ||
3529 | ignore: | |
3530 | return true; | |
3531 | } | |
3532 | ||
3533 | bool OSDMonitor::prepare_alive(MonOpRequestRef op) | |
3534 | { | |
3535 | op->mark_osdmon_event(__func__); | |
3536 | MOSDAlive *m = static_cast<MOSDAlive*>(op->get_req()); | |
3537 | int from = m->get_orig_source().num(); | |
3538 | ||
3539 | if (0) { // we probably don't care much about these | |
3540 | mon->clog->debug() << m->get_orig_source_inst() << " alive"; | |
3541 | } | |
3542 | ||
3543 | dout(7) << "prepare_alive want up_thru " << m->want << " have " << m->version | |
3544 | << " from " << m->get_orig_source_inst() << dendl; | |
3545 | ||
3546 | update_up_thru(from, m->version); // set to the latest map the OSD has | |
3547 | wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->version)); | |
3548 | return true; | |
3549 | } | |
3550 | ||
3551 | void OSDMonitor::_reply_map(MonOpRequestRef op, epoch_t e) | |
3552 | { | |
3553 | op->mark_osdmon_event(__func__); | |
3554 | dout(7) << "_reply_map " << e | |
3555 | << " from " << op->get_req()->get_orig_source_inst() | |
3556 | << dendl; | |
3557 | send_latest(op, e); | |
3558 | } | |
3559 | ||
3560 | // pg_created | |
3561 | bool OSDMonitor::preprocess_pg_created(MonOpRequestRef op) | |
3562 | { | |
3563 | op->mark_osdmon_event(__func__); | |
3564 | auto m = static_cast<MOSDPGCreated*>(op->get_req()); | |
3565 | dout(10) << __func__ << " " << *m << dendl; | |
11fdf7f2 | 3566 | auto session = op->get_session(); |
94b18763 | 3567 | mon->no_reply(op); |
7c673cae FG |
3568 | if (!session) { |
3569 | dout(10) << __func__ << ": no monitor session!" << dendl; | |
3570 | return true; | |
3571 | } | |
3572 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3573 | derr << __func__ << " received from entity " | |
3574 | << "with insufficient privileges " << session->caps << dendl; | |
3575 | return true; | |
3576 | } | |
3577 | // always forward the "created!" to the leader | |
3578 | return false; | |
3579 | } | |
3580 | ||
3581 | bool OSDMonitor::prepare_pg_created(MonOpRequestRef op) | |
3582 | { | |
3583 | op->mark_osdmon_event(__func__); | |
3584 | auto m = static_cast<MOSDPGCreated*>(op->get_req()); | |
3585 | dout(10) << __func__ << " " << *m << dendl; | |
3586 | auto src = m->get_orig_source(); | |
3587 | auto from = src.num(); | |
3588 | if (!src.is_osd() || | |
3589 | !mon->osdmon()->osdmap.is_up(from) || | |
11fdf7f2 TL |
3590 | !mon->osdmon()->osdmap.get_addrs(from).legacy_equals( |
3591 | m->get_orig_source_addrs())) { | |
7c673cae FG |
3592 | dout(1) << __func__ << " ignoring stats from non-active osd." << dendl; |
3593 | return false; | |
3594 | } | |
3595 | pending_created_pgs.push_back(m->pgid); | |
3596 | return true; | |
3597 | } | |
3598 | ||
11fdf7f2 TL |
3599 | bool OSDMonitor::preprocess_pg_ready_to_merge(MonOpRequestRef op) |
3600 | { | |
3601 | op->mark_osdmon_event(__func__); | |
3602 | auto m = static_cast<MOSDPGReadyToMerge*>(op->get_req()); | |
3603 | dout(10) << __func__ << " " << *m << dendl; | |
3604 | const pg_pool_t *pi; | |
3605 | auto session = op->get_session(); | |
3606 | if (!session) { | |
3607 | dout(10) << __func__ << ": no monitor session!" << dendl; | |
3608 | goto ignore; | |
3609 | } | |
3610 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3611 | derr << __func__ << " received from entity " | |
3612 | << "with insufficient privileges " << session->caps << dendl; | |
3613 | goto ignore; | |
3614 | } | |
3615 | pi = osdmap.get_pg_pool(m->pgid.pool()); | |
3616 | if (!pi) { | |
3617 | derr << __func__ << " pool for " << m->pgid << " dne" << dendl; | |
3618 | goto ignore; | |
3619 | } | |
3620 | if (pi->get_pg_num() <= m->pgid.ps()) { | |
3621 | dout(20) << " pg_num " << pi->get_pg_num() << " already < " << m->pgid << dendl; | |
3622 | goto ignore; | |
3623 | } | |
3624 | if (pi->get_pg_num() != m->pgid.ps() + 1) { | |
3625 | derr << " OSD trying to merge wrong pgid " << m->pgid << dendl; | |
3626 | goto ignore; | |
3627 | } | |
3628 | if (pi->get_pg_num_pending() > m->pgid.ps()) { | |
3629 | dout(20) << " pg_num_pending " << pi->get_pg_num_pending() << " > " << m->pgid << dendl; | |
3630 | goto ignore; | |
3631 | } | |
3632 | return false; | |
3633 | ||
3634 | ignore: | |
3635 | mon->no_reply(op); | |
3636 | return true; | |
3637 | } | |
3638 | ||
3639 | bool OSDMonitor::prepare_pg_ready_to_merge(MonOpRequestRef op) | |
3640 | { | |
3641 | op->mark_osdmon_event(__func__); | |
3642 | auto m = static_cast<MOSDPGReadyToMerge*>(op->get_req()); | |
3643 | dout(10) << __func__ << " " << *m << dendl; | |
3644 | pg_pool_t p; | |
3645 | if (pending_inc.new_pools.count(m->pgid.pool())) | |
3646 | p = pending_inc.new_pools[m->pgid.pool()]; | |
3647 | else | |
3648 | p = *osdmap.get_pg_pool(m->pgid.pool()); | |
3649 | if (p.get_pg_num() != m->pgid.ps() + 1 || | |
3650 | p.get_pg_num_pending() > m->pgid.ps()) { | |
3651 | dout(10) << __func__ | |
3652 | << " race with concurrent pg_num[_pending] update, will retry" | |
3653 | << dendl; | |
3654 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
3655 | return true; | |
3656 | } | |
3657 | ||
3658 | if (m->ready) { | |
3659 | p.dec_pg_num(m->pgid, | |
3660 | pending_inc.epoch, | |
3661 | m->source_version, | |
3662 | m->target_version, | |
3663 | m->last_epoch_started, | |
3664 | m->last_epoch_clean); | |
3665 | p.last_change = pending_inc.epoch; | |
3666 | } else { | |
3667 | // back off the merge attempt! | |
3668 | p.set_pg_num_pending(p.get_pg_num()); | |
3669 | } | |
3670 | ||
3671 | // force pre-nautilus clients to resend their ops, since they | |
3672 | // don't understand pg_num_pending changes form a new interval | |
3673 | p.last_force_op_resend_prenautilus = pending_inc.epoch; | |
3674 | ||
3675 | pending_inc.new_pools[m->pgid.pool()] = p; | |
3676 | ||
3677 | auto prob = g_conf().get_val<double>("mon_inject_pg_merge_bounce_probability"); | |
3678 | if (m->ready && | |
3679 | prob > 0 && | |
3680 | prob > (double)(rand() % 1000)/1000.0) { | |
3681 | derr << __func__ << " injecting pg merge pg_num bounce" << dendl; | |
3682 | auto n = new MMonCommand(mon->monmap->get_fsid()); | |
3683 | n->set_connection(m->get_connection()); | |
3684 | n->cmd = { "{\"prefix\":\"osd pool set\", \"pool\": \"" + | |
3685 | osdmap.get_pool_name(m->pgid.pool()) + | |
3686 | "\", \"var\": \"pg_num_actual\", \"val\": \"" + | |
3687 | stringify(m->pgid.ps() + 1) + "\"}" }; | |
3688 | MonOpRequestRef nop = mon->op_tracker.create_request<MonOpRequest>(n); | |
3689 | nop->set_type_service(); | |
3690 | wait_for_finished_proposal(op, new C_RetryMessage(this, nop)); | |
3691 | } else { | |
3692 | wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->version)); | |
3693 | } | |
3694 | return true; | |
3695 | } | |
3696 | ||
3697 | ||
7c673cae FG |
3698 | // ------------- |
3699 | // pg_temp changes | |
3700 | ||
3701 | bool OSDMonitor::preprocess_pgtemp(MonOpRequestRef op) | |
3702 | { | |
3703 | MOSDPGTemp *m = static_cast<MOSDPGTemp*>(op->get_req()); | |
3704 | dout(10) << "preprocess_pgtemp " << *m << dendl; | |
3705 | mempool::osdmap::vector<int> empty; | |
3706 | int from = m->get_orig_source().num(); | |
3707 | size_t ignore_cnt = 0; | |
3708 | ||
3709 | // check caps | |
11fdf7f2 | 3710 | MonSession *session = op->get_session(); |
7c673cae FG |
3711 | if (!session) |
3712 | goto ignore; | |
3713 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3714 | dout(0) << "attempt to send MOSDPGTemp from entity with insufficient caps " | |
3715 | << session->caps << dendl; | |
3716 | goto ignore; | |
3717 | } | |
3718 | ||
3719 | if (!osdmap.is_up(from) || | |
11fdf7f2 TL |
3720 | !osdmap.get_addrs(from).legacy_equals(m->get_orig_source_addrs())) { |
3721 | dout(7) << "ignoring pgtemp message from down " | |
3722 | << m->get_orig_source() << " " << m->get_orig_source_addrs() | |
3723 | << dendl; | |
7c673cae FG |
3724 | goto ignore; |
3725 | } | |
3726 | ||
3efd9988 FG |
3727 | if (m->forced) { |
3728 | return false; | |
3729 | } | |
3730 | ||
7c673cae FG |
3731 | for (auto p = m->pg_temp.begin(); p != m->pg_temp.end(); ++p) { |
3732 | dout(20) << " " << p->first | |
31f18b77 | 3733 | << (osdmap.pg_temp->count(p->first) ? osdmap.pg_temp->get(p->first) : empty) |
7c673cae FG |
3734 | << " -> " << p->second << dendl; |
3735 | ||
3736 | // does the pool exist? | |
3737 | if (!osdmap.have_pg_pool(p->first.pool())) { | |
3738 | /* | |
3739 | * 1. If the osdmap does not have the pool, it means the pool has been | |
3740 | * removed in-between the osd sending this message and us handling it. | |
3741 | * 2. If osdmap doesn't have the pool, it is safe to assume the pool does | |
3742 | * not exist in the pending either, as the osds would not send a | |
3743 | * message about a pool they know nothing about (yet). | |
3744 | * 3. However, if the pool does exist in the pending, then it must be a | |
3745 | * new pool, and not relevant to this message (see 1). | |
3746 | */ | |
3747 | dout(10) << __func__ << " ignore " << p->first << " -> " << p->second | |
3748 | << ": pool has been removed" << dendl; | |
3749 | ignore_cnt++; | |
3750 | continue; | |
3751 | } | |
3752 | ||
3753 | int acting_primary = -1; | |
3754 | osdmap.pg_to_up_acting_osds( | |
3755 | p->first, nullptr, nullptr, nullptr, &acting_primary); | |
3756 | if (acting_primary != from) { | |
3757 | /* If the source isn't the primary based on the current osdmap, we know | |
3758 | * that the interval changed and that we can discard this message. | |
3759 | * Indeed, we must do so to avoid 16127 since we can't otherwise determine | |
3760 | * which of two pg temp mappings on the same pg is more recent. | |
3761 | */ | |
3762 | dout(10) << __func__ << " ignore " << p->first << " -> " << p->second | |
3763 | << ": primary has changed" << dendl; | |
3764 | ignore_cnt++; | |
3765 | continue; | |
3766 | } | |
3767 | ||
3768 | // removal? | |
3769 | if (p->second.empty() && (osdmap.pg_temp->count(p->first) || | |
3770 | osdmap.primary_temp->count(p->first))) | |
3771 | return false; | |
3772 | // change? | |
3773 | // NOTE: we assume that this will clear pg_primary, so consider | |
3774 | // an existing pg_primary field to imply a change | |
3775 | if (p->second.size() && | |
3776 | (osdmap.pg_temp->count(p->first) == 0 || | |
11fdf7f2 | 3777 | osdmap.pg_temp->get(p->first) != p->second || |
7c673cae FG |
3778 | osdmap.primary_temp->count(p->first))) |
3779 | return false; | |
3780 | } | |
3781 | ||
3782 | // should we ignore all the pgs? | |
3783 | if (ignore_cnt == m->pg_temp.size()) | |
3784 | goto ignore; | |
3785 | ||
3786 | dout(7) << "preprocess_pgtemp e" << m->map_epoch << " no changes from " << m->get_orig_source_inst() << dendl; | |
3787 | _reply_map(op, m->map_epoch); | |
3788 | return true; | |
3789 | ||
3790 | ignore: | |
3791 | return true; | |
3792 | } | |
3793 | ||
3794 | void OSDMonitor::update_up_thru(int from, epoch_t up_thru) | |
3795 | { | |
3796 | epoch_t old_up_thru = osdmap.get_up_thru(from); | |
3797 | auto ut = pending_inc.new_up_thru.find(from); | |
3798 | if (ut != pending_inc.new_up_thru.end()) { | |
3799 | old_up_thru = ut->second; | |
3800 | } | |
3801 | if (up_thru > old_up_thru) { | |
3802 | // set up_thru too, so the osd doesn't have to ask again | |
3803 | pending_inc.new_up_thru[from] = up_thru; | |
3804 | } | |
3805 | } | |
3806 | ||
3807 | bool OSDMonitor::prepare_pgtemp(MonOpRequestRef op) | |
3808 | { | |
3809 | op->mark_osdmon_event(__func__); | |
3810 | MOSDPGTemp *m = static_cast<MOSDPGTemp*>(op->get_req()); | |
3811 | int from = m->get_orig_source().num(); | |
3812 | dout(7) << "prepare_pgtemp e" << m->map_epoch << " from " << m->get_orig_source_inst() << dendl; | |
3813 | for (map<pg_t,vector<int32_t> >::iterator p = m->pg_temp.begin(); p != m->pg_temp.end(); ++p) { | |
3814 | uint64_t pool = p->first.pool(); | |
3815 | if (pending_inc.old_pools.count(pool)) { | |
3816 | dout(10) << __func__ << " ignore " << p->first << " -> " << p->second | |
3817 | << ": pool pending removal" << dendl; | |
3818 | continue; | |
3819 | } | |
3820 | if (!osdmap.have_pg_pool(pool)) { | |
3821 | dout(10) << __func__ << " ignore " << p->first << " -> " << p->second | |
3822 | << ": pool has been removed" << dendl; | |
3823 | continue; | |
3824 | } | |
3825 | pending_inc.new_pg_temp[p->first] = | |
3826 | mempool::osdmap::vector<int>(p->second.begin(), p->second.end()); | |
3827 | ||
3828 | // unconditionally clear pg_primary (until this message can encode | |
3829 | // a change for that, too.. at which point we need to also fix | |
3830 | // preprocess_pg_temp) | |
3831 | if (osdmap.primary_temp->count(p->first) || | |
3832 | pending_inc.new_primary_temp.count(p->first)) | |
3833 | pending_inc.new_primary_temp[p->first] = -1; | |
3834 | } | |
3835 | ||
3836 | // set up_thru too, so the osd doesn't have to ask again | |
3837 | update_up_thru(from, m->map_epoch); | |
3838 | ||
3839 | wait_for_finished_proposal(op, new C_ReplyMap(this, op, m->map_epoch)); | |
3840 | return true; | |
3841 | } | |
3842 | ||
3843 | ||
3844 | // --- | |
3845 | ||
3846 | bool OSDMonitor::preprocess_remove_snaps(MonOpRequestRef op) | |
3847 | { | |
3848 | op->mark_osdmon_event(__func__); | |
3849 | MRemoveSnaps *m = static_cast<MRemoveSnaps*>(op->get_req()); | |
3850 | dout(7) << "preprocess_remove_snaps " << *m << dendl; | |
3851 | ||
3852 | // check privilege, ignore if failed | |
11fdf7f2 | 3853 | MonSession *session = op->get_session(); |
f64942e4 | 3854 | mon->no_reply(op); |
7c673cae FG |
3855 | if (!session) |
3856 | goto ignore; | |
3857 | if (!session->caps.is_capable( | |
11fdf7f2 | 3858 | cct, |
7c673cae FG |
3859 | CEPH_ENTITY_TYPE_MON, |
3860 | session->entity_name, | |
11fdf7f2 TL |
3861 | "osd", "osd pool rmsnap", {}, true, true, false, |
3862 | session->get_peer_socket_addr())) { | |
7c673cae FG |
3863 | dout(0) << "got preprocess_remove_snaps from entity with insufficient caps " |
3864 | << session->caps << dendl; | |
3865 | goto ignore; | |
3866 | } | |
3867 | ||
3868 | for (map<int, vector<snapid_t> >::iterator q = m->snaps.begin(); | |
3869 | q != m->snaps.end(); | |
3870 | ++q) { | |
3871 | if (!osdmap.have_pg_pool(q->first)) { | |
3872 | dout(10) << " ignoring removed_snaps " << q->second << " on non-existent pool " << q->first << dendl; | |
3873 | continue; | |
3874 | } | |
3875 | const pg_pool_t *pi = osdmap.get_pg_pool(q->first); | |
3876 | for (vector<snapid_t>::iterator p = q->second.begin(); | |
3877 | p != q->second.end(); | |
3878 | ++p) { | |
3879 | if (*p > pi->get_snap_seq() || | |
3880 | !pi->removed_snaps.contains(*p)) | |
3881 | return false; | |
3882 | } | |
3883 | } | |
3884 | ||
3885 | ignore: | |
3886 | return true; | |
3887 | } | |
3888 | ||
3889 | bool OSDMonitor::prepare_remove_snaps(MonOpRequestRef op) | |
3890 | { | |
3891 | op->mark_osdmon_event(__func__); | |
3892 | MRemoveSnaps *m = static_cast<MRemoveSnaps*>(op->get_req()); | |
3893 | dout(7) << "prepare_remove_snaps " << *m << dendl; | |
3894 | ||
3895 | for (map<int, vector<snapid_t> >::iterator p = m->snaps.begin(); | |
3896 | p != m->snaps.end(); | |
3897 | ++p) { | |
3898 | ||
3899 | if (!osdmap.have_pg_pool(p->first)) { | |
3900 | dout(10) << " ignoring removed_snaps " << p->second << " on non-existent pool " << p->first << dendl; | |
3901 | continue; | |
3902 | } | |
3903 | ||
3904 | pg_pool_t& pi = osdmap.pools[p->first]; | |
3905 | for (vector<snapid_t>::iterator q = p->second.begin(); | |
3906 | q != p->second.end(); | |
3907 | ++q) { | |
3908 | if (!pi.removed_snaps.contains(*q) && | |
3909 | (!pending_inc.new_pools.count(p->first) || | |
3910 | !pending_inc.new_pools[p->first].removed_snaps.contains(*q))) { | |
3911 | pg_pool_t *newpi = pending_inc.get_new_pool(p->first, &pi); | |
3912 | newpi->removed_snaps.insert(*q); | |
11fdf7f2 | 3913 | newpi->flags |= pg_pool_t::FLAG_SELFMANAGED_SNAPS; |
7c673cae FG |
3914 | dout(10) << " pool " << p->first << " removed_snaps added " << *q |
3915 | << " (now " << newpi->removed_snaps << ")" << dendl; | |
3916 | if (*q > newpi->get_snap_seq()) { | |
11fdf7f2 TL |
3917 | dout(10) << " pool " << p->first << " snap_seq " |
3918 | << newpi->get_snap_seq() << " -> " << *q << dendl; | |
7c673cae FG |
3919 | newpi->set_snap_seq(*q); |
3920 | } | |
3921 | newpi->set_snap_epoch(pending_inc.epoch); | |
11fdf7f2 | 3922 | pending_inc.new_removed_snaps[p->first].insert(*q); |
7c673cae FG |
3923 | } |
3924 | } | |
3925 | } | |
3926 | return true; | |
3927 | } | |
3928 | ||
3929 | // osd beacon | |
3930 | bool OSDMonitor::preprocess_beacon(MonOpRequestRef op) | |
3931 | { | |
3932 | op->mark_osdmon_event(__func__); | |
7c673cae | 3933 | // check caps |
11fdf7f2 | 3934 | auto session = op->get_session(); |
94b18763 | 3935 | mon->no_reply(op); |
7c673cae FG |
3936 | if (!session) { |
3937 | dout(10) << __func__ << " no monitor session!" << dendl; | |
3938 | return true; | |
3939 | } | |
3940 | if (!session->is_capable("osd", MON_CAP_X)) { | |
3941 | derr << __func__ << " received from entity " | |
3942 | << "with insufficient privileges " << session->caps << dendl; | |
3943 | return true; | |
3944 | } | |
3945 | // Always forward the beacon to the leader, even if they are the same as | |
3946 | // the old one. The leader will mark as down osds that haven't sent | |
3947 | // beacon for a few minutes. | |
3948 | return false; | |
3949 | } | |
3950 | ||
3951 | bool OSDMonitor::prepare_beacon(MonOpRequestRef op) | |
3952 | { | |
3953 | op->mark_osdmon_event(__func__); | |
3954 | const auto beacon = static_cast<MOSDBeacon*>(op->get_req()); | |
3955 | const auto src = beacon->get_orig_source(); | |
3956 | dout(10) << __func__ << " " << *beacon | |
3957 | << " from " << src << dendl; | |
3958 | int from = src.num(); | |
3959 | ||
3960 | if (!src.is_osd() || | |
3961 | !osdmap.is_up(from) || | |
11fdf7f2 TL |
3962 | !osdmap.get_addrs(from).legacy_equals(beacon->get_orig_source_addrs())) { |
3963 | if (src.is_osd() && !osdmap.is_up(from)) { | |
3964 | // share some new maps with this guy in case it may not be | |
3965 | // aware of its own deadness... | |
3966 | send_latest(op, beacon->version+1); | |
3967 | } | |
3968 | dout(1) << " ignoring beacon from non-active osd." << from << dendl; | |
7c673cae FG |
3969 | return false; |
3970 | } | |
3971 | ||
3972 | last_osd_report[from] = ceph_clock_now(); | |
3973 | osd_epochs[from] = beacon->version; | |
3974 | ||
3975 | for (const auto& pg : beacon->pgs) { | |
3976 | last_epoch_clean.report(pg, beacon->min_last_epoch_clean); | |
3977 | } | |
3978 | return false; | |
3979 | } | |
3980 | ||
3981 | // --------------- | |
3982 | // map helpers | |
3983 | ||
3984 | void OSDMonitor::send_latest(MonOpRequestRef op, epoch_t start) | |
3985 | { | |
3986 | op->mark_osdmon_event(__func__); | |
3987 | dout(5) << "send_latest to " << op->get_req()->get_orig_source_inst() | |
3988 | << " start " << start << dendl; | |
3989 | if (start == 0) | |
3990 | send_full(op); | |
3991 | else | |
3992 | send_incremental(op, start); | |
3993 | } | |
3994 | ||
3995 | ||
28e407b8 | 3996 | MOSDMap *OSDMonitor::build_latest_full(uint64_t features) |
7c673cae | 3997 | { |
28e407b8 AA |
3998 | MOSDMap *r = new MOSDMap(mon->monmap->fsid, features); |
3999 | get_version_full(osdmap.get_epoch(), features, r->maps[osdmap.get_epoch()]); | |
7c673cae FG |
4000 | r->oldest_map = get_first_committed(); |
4001 | r->newest_map = osdmap.get_epoch(); | |
4002 | return r; | |
4003 | } | |
4004 | ||
28e407b8 | 4005 | MOSDMap *OSDMonitor::build_incremental(epoch_t from, epoch_t to, uint64_t features) |
7c673cae | 4006 | { |
11fdf7f2 TL |
4007 | dout(10) << "build_incremental [" << from << ".." << to << "] with features " |
4008 | << std::hex << features << std::dec << dendl; | |
28e407b8 | 4009 | MOSDMap *m = new MOSDMap(mon->monmap->fsid, features); |
7c673cae FG |
4010 | m->oldest_map = get_first_committed(); |
4011 | m->newest_map = osdmap.get_epoch(); | |
4012 | ||
4013 | for (epoch_t e = to; e >= from && e > 0; e--) { | |
4014 | bufferlist bl; | |
28e407b8 | 4015 | int err = get_version(e, features, bl); |
7c673cae | 4016 | if (err == 0) { |
11fdf7f2 | 4017 | ceph_assert(bl.length()); |
7c673cae FG |
4018 | // if (get_version(e, bl) > 0) { |
4019 | dout(20) << "build_incremental inc " << e << " " | |
4020 | << bl.length() << " bytes" << dendl; | |
4021 | m->incremental_maps[e] = bl; | |
4022 | } else { | |
11fdf7f2 TL |
4023 | ceph_assert(err == -ENOENT); |
4024 | ceph_assert(!bl.length()); | |
28e407b8 | 4025 | get_version_full(e, features, bl); |
7c673cae FG |
4026 | if (bl.length() > 0) { |
4027 | //else if (get_version("full", e, bl) > 0) { | |
4028 | dout(20) << "build_incremental full " << e << " " | |
4029 | << bl.length() << " bytes" << dendl; | |
4030 | m->maps[e] = bl; | |
4031 | } else { | |
4032 | ceph_abort(); // we should have all maps. | |
4033 | } | |
4034 | } | |
4035 | } | |
4036 | return m; | |
4037 | } | |
4038 | ||
4039 | void OSDMonitor::send_full(MonOpRequestRef op) | |
4040 | { | |
4041 | op->mark_osdmon_event(__func__); | |
4042 | dout(5) << "send_full to " << op->get_req()->get_orig_source_inst() << dendl; | |
28e407b8 | 4043 | mon->send_reply(op, build_latest_full(op->get_session()->con_features)); |
7c673cae FG |
4044 | } |
4045 | ||
4046 | void OSDMonitor::send_incremental(MonOpRequestRef op, epoch_t first) | |
4047 | { | |
4048 | op->mark_osdmon_event(__func__); | |
4049 | ||
4050 | MonSession *s = op->get_session(); | |
11fdf7f2 | 4051 | ceph_assert(s); |
7c673cae | 4052 | |
11fdf7f2 | 4053 | if (s->proxy_con) { |
7c673cae FG |
4054 | // oh, we can tell the other mon to do it |
4055 | dout(10) << __func__ << " asking proxying mon to send_incremental from " | |
4056 | << first << dendl; | |
4057 | MRoute *r = new MRoute(s->proxy_tid, NULL); | |
4058 | r->send_osdmap_first = first; | |
4059 | s->proxy_con->send_message(r); | |
4060 | op->mark_event("reply: send routed send_osdmap_first reply"); | |
4061 | } else { | |
4062 | // do it ourselves | |
4063 | send_incremental(first, s, false, op); | |
4064 | } | |
4065 | } | |
4066 | ||
4067 | void OSDMonitor::send_incremental(epoch_t first, | |
4068 | MonSession *session, | |
4069 | bool onetime, | |
4070 | MonOpRequestRef req) | |
4071 | { | |
4072 | dout(5) << "send_incremental [" << first << ".." << osdmap.get_epoch() << "]" | |
11fdf7f2 | 4073 | << " to " << session->name << dendl; |
7c673cae | 4074 | |
28e407b8 AA |
4075 | // get feature of the peer |
4076 | // use quorum_con_features, if it's an anonymous connection. | |
4077 | uint64_t features = session->con_features ? session->con_features : | |
4078 | mon->get_quorum_con_features(); | |
4079 | ||
7c673cae | 4080 | if (first <= session->osd_epoch) { |
11fdf7f2 | 4081 | dout(10) << __func__ << " " << session->name << " should already have epoch " |
7c673cae FG |
4082 | << session->osd_epoch << dendl; |
4083 | first = session->osd_epoch + 1; | |
4084 | } | |
4085 | ||
4086 | if (first < get_first_committed()) { | |
11fdf7f2 TL |
4087 | MOSDMap *m = new MOSDMap(osdmap.get_fsid(), features); |
4088 | m->oldest_map = get_first_committed(); | |
4089 | m->newest_map = osdmap.get_epoch(); | |
4090 | ||
4091 | // share removed snaps during the gap | |
4092 | get_removed_snaps_range(first, m->oldest_map, &m->gap_removed_snaps); | |
4093 | ||
7c673cae FG |
4094 | first = get_first_committed(); |
4095 | bufferlist bl; | |
28e407b8 | 4096 | int err = get_version_full(first, features, bl); |
11fdf7f2 TL |
4097 | ceph_assert(err == 0); |
4098 | ceph_assert(bl.length()); | |
7c673cae FG |
4099 | dout(20) << "send_incremental starting with base full " |
4100 | << first << " " << bl.length() << " bytes" << dendl; | |
7c673cae FG |
4101 | m->maps[first] = bl; |
4102 | ||
4103 | if (req) { | |
4104 | mon->send_reply(req, m); | |
4105 | session->osd_epoch = first; | |
4106 | return; | |
4107 | } else { | |
4108 | session->con->send_message(m); | |
4109 | session->osd_epoch = first; | |
4110 | } | |
4111 | first++; | |
4112 | } | |
4113 | ||
4114 | while (first <= osdmap.get_epoch()) { | |
11fdf7f2 | 4115 | epoch_t last = std::min<epoch_t>(first + g_conf()->osd_map_message_max - 1, |
28e407b8 AA |
4116 | osdmap.get_epoch()); |
4117 | MOSDMap *m = build_incremental(first, last, features); | |
7c673cae FG |
4118 | |
4119 | if (req) { | |
4120 | // send some maps. it may not be all of them, but it will get them | |
4121 | // started. | |
4122 | mon->send_reply(req, m); | |
4123 | } else { | |
4124 | session->con->send_message(m); | |
4125 | first = last + 1; | |
4126 | } | |
4127 | session->osd_epoch = last; | |
4128 | if (onetime || req) | |
4129 | break; | |
4130 | } | |
4131 | } | |
4132 | ||
11fdf7f2 TL |
4133 | void OSDMonitor::get_removed_snaps_range( |
4134 | epoch_t start, epoch_t end, | |
4135 | mempool::osdmap::map<int64_t,OSDMap::snap_interval_set_t> *gap_removed_snaps) | |
4136 | { | |
4137 | // we only care about pools that exist now. | |
4138 | for (auto& p : osdmap.get_pools()) { | |
4139 | auto& t = (*gap_removed_snaps)[p.first]; | |
4140 | for (epoch_t epoch = start; epoch < end; ++epoch) { | |
4141 | string k = make_snap_epoch_key(p.first, epoch); | |
4142 | bufferlist v; | |
4143 | mon->store->get(OSD_SNAP_PREFIX, k, v); | |
4144 | if (v.length()) { | |
4145 | auto q = v.cbegin(); | |
4146 | OSDMap::snap_interval_set_t snaps; | |
4147 | decode(snaps, q); | |
4148 | t.union_of(snaps); | |
4149 | } | |
4150 | } | |
4151 | dout(10) << __func__ << " " << p.first << " " << t << dendl; | |
4152 | } | |
4153 | } | |
4154 | ||
7c673cae FG |
4155 | int OSDMonitor::get_version(version_t ver, bufferlist& bl) |
4156 | { | |
28e407b8 AA |
4157 | return get_version(ver, mon->get_quorum_con_features(), bl); |
4158 | } | |
4159 | ||
4160 | void OSDMonitor::reencode_incremental_map(bufferlist& bl, uint64_t features) | |
4161 | { | |
4162 | OSDMap::Incremental inc; | |
11fdf7f2 | 4163 | auto q = bl.cbegin(); |
28e407b8 AA |
4164 | inc.decode(q); |
4165 | // always encode with subset of osdmap's canonical features | |
4166 | uint64_t f = features & inc.encode_features; | |
4167 | dout(20) << __func__ << " " << inc.epoch << " with features " << f | |
4168 | << dendl; | |
4169 | bl.clear(); | |
4170 | if (inc.fullmap.length()) { | |
4171 | // embedded full map? | |
4172 | OSDMap m; | |
4173 | m.decode(inc.fullmap); | |
4174 | inc.fullmap.clear(); | |
4175 | m.encode(inc.fullmap, f | CEPH_FEATURE_RESERVED); | |
4176 | } | |
4177 | if (inc.crush.length()) { | |
4178 | // embedded crush map | |
4179 | CrushWrapper c; | |
11fdf7f2 | 4180 | auto p = inc.crush.cbegin(); |
28e407b8 AA |
4181 | c.decode(p); |
4182 | inc.crush.clear(); | |
4183 | c.encode(inc.crush, f); | |
4184 | } | |
4185 | inc.encode(bl, f | CEPH_FEATURE_RESERVED); | |
4186 | } | |
4187 | ||
4188 | void OSDMonitor::reencode_full_map(bufferlist& bl, uint64_t features) | |
4189 | { | |
4190 | OSDMap m; | |
11fdf7f2 | 4191 | auto q = bl.cbegin(); |
28e407b8 AA |
4192 | m.decode(q); |
4193 | // always encode with subset of osdmap's canonical features | |
4194 | uint64_t f = features & m.get_encoding_features(); | |
4195 | dout(20) << __func__ << " " << m.get_epoch() << " with features " << f | |
4196 | << dendl; | |
4197 | bl.clear(); | |
4198 | m.encode(bl, f | CEPH_FEATURE_RESERVED); | |
4199 | } | |
4200 | ||
4201 | int OSDMonitor::get_version(version_t ver, uint64_t features, bufferlist& bl) | |
4202 | { | |
4203 | uint64_t significant_features = OSDMap::get_significant_features(features); | |
4204 | if (inc_osd_cache.lookup({ver, significant_features}, &bl)) { | |
4205 | return 0; | |
4206 | } | |
4207 | int ret = PaxosService::get_version(ver, bl); | |
4208 | if (ret < 0) { | |
7c673cae | 4209 | return ret; |
28e407b8 AA |
4210 | } |
4211 | // NOTE: this check is imprecise; the OSDMap encoding features may | |
4212 | // be a subset of the latest mon quorum features, but worst case we | |
4213 | // reencode once and then cache the (identical) result under both | |
4214 | // feature masks. | |
4215 | if (significant_features != | |
4216 | OSDMap::get_significant_features(mon->get_quorum_con_features())) { | |
4217 | reencode_incremental_map(bl, features); | |
4218 | } | |
eafe8130 | 4219 | inc_osd_cache.add_bytes({ver, significant_features}, bl); |
28e407b8 | 4220 | return 0; |
7c673cae FG |
4221 | } |
4222 | ||
11fdf7f2 TL |
4223 | int OSDMonitor::get_inc(version_t ver, OSDMap::Incremental& inc) |
4224 | { | |
4225 | bufferlist inc_bl; | |
4226 | int err = get_version(ver, inc_bl); | |
4227 | ceph_assert(err == 0); | |
4228 | ceph_assert(inc_bl.length()); | |
4229 | ||
4230 | auto p = inc_bl.cbegin(); | |
4231 | inc.decode(p); | |
4232 | dout(10) << __func__ << " " | |
4233 | << " epoch " << inc.epoch | |
4234 | << " inc_crc " << inc.inc_crc | |
4235 | << " full_crc " << inc.full_crc | |
4236 | << " encode_features " << inc.encode_features << dendl; | |
4237 | return 0; | |
4238 | } | |
4239 | ||
4240 | int OSDMonitor::get_full_from_pinned_map(version_t ver, bufferlist& bl) | |
4241 | { | |
4242 | dout(10) << __func__ << " ver " << ver << dendl; | |
4243 | ||
4244 | version_t closest_pinned = osdmap_manifest.get_lower_closest_pinned(ver); | |
4245 | if (closest_pinned == 0) { | |
4246 | return -ENOENT; | |
4247 | } | |
4248 | if (closest_pinned > ver) { | |
4249 | dout(0) << __func__ << " pinned: " << osdmap_manifest.pinned << dendl; | |
4250 | } | |
4251 | ceph_assert(closest_pinned <= ver); | |
4252 | ||
4253 | dout(10) << __func__ << " closest pinned ver " << closest_pinned << dendl; | |
4254 | ||
4255 | // get osdmap incremental maps and apply on top of this one. | |
4256 | bufferlist osdm_bl; | |
4257 | bool has_cached_osdmap = false; | |
4258 | for (version_t v = ver-1; v >= closest_pinned; --v) { | |
4259 | if (full_osd_cache.lookup({v, mon->get_quorum_con_features()}, | |
4260 | &osdm_bl)) { | |
4261 | dout(10) << __func__ << " found map in cache ver " << v << dendl; | |
4262 | closest_pinned = v; | |
4263 | has_cached_osdmap = true; | |
4264 | break; | |
4265 | } | |
4266 | } | |
4267 | ||
4268 | if (!has_cached_osdmap) { | |
4269 | int err = PaxosService::get_version_full(closest_pinned, osdm_bl); | |
4270 | if (err != 0) { | |
4271 | derr << __func__ << " closest pinned map ver " << closest_pinned | |
4272 | << " not available! error: " << cpp_strerror(err) << dendl; | |
4273 | } | |
4274 | ceph_assert(err == 0); | |
4275 | } | |
4276 | ||
4277 | ceph_assert(osdm_bl.length()); | |
4278 | ||
4279 | OSDMap osdm; | |
4280 | osdm.decode(osdm_bl); | |
4281 | ||
4282 | dout(10) << __func__ << " loaded osdmap epoch " << closest_pinned | |
4283 | << " e" << osdm.epoch | |
4284 | << " crc " << osdm.get_crc() | |
4285 | << " -- applying incremental maps." << dendl; | |
4286 | ||
4287 | uint64_t encode_features = 0; | |
4288 | for (version_t v = closest_pinned + 1; v <= ver; ++v) { | |
4289 | dout(20) << __func__ << " applying inc epoch " << v << dendl; | |
4290 | ||
4291 | OSDMap::Incremental inc; | |
4292 | int err = get_inc(v, inc); | |
4293 | ceph_assert(err == 0); | |
4294 | ||
4295 | encode_features = inc.encode_features; | |
4296 | ||
4297 | err = osdm.apply_incremental(inc); | |
4298 | ceph_assert(err == 0); | |
4299 | ||
4300 | // this block performs paranoid checks on map retrieval | |
4301 | if (g_conf().get_val<bool>("mon_debug_extra_checks") && | |
4302 | inc.full_crc != 0) { | |
4303 | ||
4304 | uint64_t f = encode_features; | |
4305 | if (!f) { | |
4306 | f = (mon->quorum_con_features ? mon->quorum_con_features : -1); | |
4307 | } | |
4308 | ||
4309 | // encode osdmap to force calculating crcs | |
4310 | bufferlist tbl; | |
4311 | osdm.encode(tbl, f | CEPH_FEATURE_RESERVED); | |
4312 | // decode osdmap to compare crcs with what's expected by incremental | |
4313 | OSDMap tosdm; | |
4314 | tosdm.decode(tbl); | |
4315 | ||
4316 | if (tosdm.get_crc() != inc.full_crc) { | |
4317 | derr << __func__ | |
4318 | << " osdmap crc mismatch! (osdmap crc " << tosdm.get_crc() | |
4319 | << ", expected " << inc.full_crc << ")" << dendl; | |
4320 | ceph_abort_msg("osdmap crc mismatch"); | |
4321 | } | |
4322 | } | |
4323 | ||
4324 | // note: we cannot add the recently computed map to the cache, as is, | |
4325 | // because we have not encoded the map into a bl. | |
4326 | } | |
4327 | ||
4328 | if (!encode_features) { | |
4329 | dout(10) << __func__ | |
4330 | << " last incremental map didn't have features;" | |
4331 | << " defaulting to quorum's or all" << dendl; | |
4332 | encode_features = | |
4333 | (mon->quorum_con_features ? mon->quorum_con_features : -1); | |
4334 | } | |
4335 | osdm.encode(bl, encode_features | CEPH_FEATURE_RESERVED); | |
4336 | ||
4337 | return 0; | |
4338 | } | |
4339 | ||
7c673cae FG |
4340 | int OSDMonitor::get_version_full(version_t ver, bufferlist& bl) |
4341 | { | |
28e407b8 AA |
4342 | return get_version_full(ver, mon->get_quorum_con_features(), bl); |
4343 | } | |
4344 | ||
4345 | int OSDMonitor::get_version_full(version_t ver, uint64_t features, | |
4346 | bufferlist& bl) | |
4347 | { | |
4348 | uint64_t significant_features = OSDMap::get_significant_features(features); | |
4349 | if (full_osd_cache.lookup({ver, significant_features}, &bl)) { | |
4350 | return 0; | |
4351 | } | |
4352 | int ret = PaxosService::get_version_full(ver, bl); | |
11fdf7f2 TL |
4353 | if (ret == -ENOENT) { |
4354 | // build map? | |
4355 | ret = get_full_from_pinned_map(ver, bl); | |
4356 | } | |
28e407b8 | 4357 | if (ret < 0) { |
7c673cae | 4358 | return ret; |
28e407b8 AA |
4359 | } |
4360 | // NOTE: this check is imprecise; the OSDMap encoding features may | |
4361 | // be a subset of the latest mon quorum features, but worst case we | |
4362 | // reencode once and then cache the (identical) result under both | |
4363 | // feature masks. | |
4364 | if (significant_features != | |
4365 | OSDMap::get_significant_features(mon->get_quorum_con_features())) { | |
4366 | reencode_full_map(bl, features); | |
4367 | } | |
eafe8130 | 4368 | full_osd_cache.add_bytes({ver, significant_features}, bl); |
28e407b8 | 4369 | return 0; |
7c673cae FG |
4370 | } |
4371 | ||
11fdf7f2 TL |
4372 | epoch_t OSDMonitor::blacklist(const entity_addrvec_t& av, utime_t until) |
4373 | { | |
4374 | dout(10) << "blacklist " << av << " until " << until << dendl; | |
4375 | for (auto a : av.v) { | |
4376 | if (osdmap.require_osd_release >= CEPH_RELEASE_NAUTILUS) { | |
4377 | a.set_type(entity_addr_t::TYPE_ANY); | |
4378 | } else { | |
4379 | a.set_type(entity_addr_t::TYPE_LEGACY); | |
4380 | } | |
4381 | pending_inc.new_blacklist[a] = until; | |
4382 | } | |
4383 | return pending_inc.epoch; | |
4384 | } | |
4385 | ||
4386 | epoch_t OSDMonitor::blacklist(entity_addr_t a, utime_t until) | |
7c673cae | 4387 | { |
11fdf7f2 TL |
4388 | if (osdmap.require_osd_release >= CEPH_RELEASE_NAUTILUS) { |
4389 | a.set_type(entity_addr_t::TYPE_ANY); | |
4390 | } else { | |
4391 | a.set_type(entity_addr_t::TYPE_LEGACY); | |
4392 | } | |
7c673cae FG |
4393 | dout(10) << "blacklist " << a << " until " << until << dendl; |
4394 | pending_inc.new_blacklist[a] = until; | |
4395 | return pending_inc.epoch; | |
4396 | } | |
4397 | ||
4398 | ||
4399 | void OSDMonitor::check_osdmap_subs() | |
4400 | { | |
4401 | dout(10) << __func__ << dendl; | |
4402 | if (!osdmap.get_epoch()) { | |
4403 | return; | |
4404 | } | |
4405 | auto osdmap_subs = mon->session_map.subs.find("osdmap"); | |
4406 | if (osdmap_subs == mon->session_map.subs.end()) { | |
4407 | return; | |
4408 | } | |
4409 | auto p = osdmap_subs->second->begin(); | |
4410 | while (!p.end()) { | |
4411 | auto sub = *p; | |
4412 | ++p; | |
4413 | check_osdmap_sub(sub); | |
4414 | } | |
4415 | } | |
4416 | ||
4417 | void OSDMonitor::check_osdmap_sub(Subscription *sub) | |
4418 | { | |
4419 | dout(10) << __func__ << " " << sub << " next " << sub->next | |
4420 | << (sub->onetime ? " (onetime)":" (ongoing)") << dendl; | |
4421 | if (sub->next <= osdmap.get_epoch()) { | |
4422 | if (sub->next >= 1) | |
4423 | send_incremental(sub->next, sub->session, sub->incremental_onetime); | |
4424 | else | |
28e407b8 | 4425 | sub->session->con->send_message(build_latest_full(sub->session->con_features)); |
7c673cae FG |
4426 | if (sub->onetime) |
4427 | mon->session_map.remove_sub(sub); | |
4428 | else | |
4429 | sub->next = osdmap.get_epoch() + 1; | |
4430 | } | |
4431 | } | |
4432 | ||
4433 | void OSDMonitor::check_pg_creates_subs() | |
4434 | { | |
7c673cae FG |
4435 | if (!osdmap.get_num_up_osds()) { |
4436 | return; | |
4437 | } | |
11fdf7f2 | 4438 | ceph_assert(osdmap.get_up_osd_features() & CEPH_FEATURE_MON_STATEFUL_SUB); |
7c673cae FG |
4439 | mon->with_session_map([this](const MonSessionMap& session_map) { |
4440 | auto pg_creates_subs = session_map.subs.find("osd_pg_creates"); | |
4441 | if (pg_creates_subs == session_map.subs.end()) { | |
4442 | return; | |
4443 | } | |
4444 | for (auto sub : *pg_creates_subs->second) { | |
4445 | check_pg_creates_sub(sub); | |
4446 | } | |
4447 | }); | |
4448 | } | |
4449 | ||
4450 | void OSDMonitor::check_pg_creates_sub(Subscription *sub) | |
4451 | { | |
11fdf7f2 TL |
4452 | dout(20) << __func__ << " .. " << sub->session->name << dendl; |
4453 | ceph_assert(sub->type == "osd_pg_creates"); | |
7c673cae FG |
4454 | // only send these if the OSD is up. we will check_subs() when they do |
4455 | // come up so they will get the creates then. | |
11fdf7f2 TL |
4456 | if (sub->session->name.is_osd() && |
4457 | mon->osdmon()->osdmap.is_up(sub->session->name.num())) { | |
4458 | sub->next = send_pg_creates(sub->session->name.num(), | |
7c673cae FG |
4459 | sub->session->con.get(), |
4460 | sub->next); | |
4461 | } | |
4462 | } | |
4463 | ||
c07f9fc5 | 4464 | void OSDMonitor::do_application_enable(int64_t pool_id, |
11fdf7f2 TL |
4465 | const std::string &app_name, |
4466 | const std::string &app_key, | |
4467 | const std::string &app_value) | |
c07f9fc5 | 4468 | { |
11fdf7f2 | 4469 | ceph_assert(paxos->is_plugged() && is_writeable()); |
c07f9fc5 FG |
4470 | |
4471 | dout(20) << __func__ << ": pool_id=" << pool_id << ", app_name=" << app_name | |
4472 | << dendl; | |
4473 | ||
11fdf7f2 | 4474 | ceph_assert(osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS); |
35e4c445 | 4475 | |
c07f9fc5 | 4476 | auto pp = osdmap.get_pg_pool(pool_id); |
11fdf7f2 | 4477 | ceph_assert(pp != nullptr); |
c07f9fc5 FG |
4478 | |
4479 | pg_pool_t p = *pp; | |
4480 | if (pending_inc.new_pools.count(pool_id)) { | |
4481 | p = pending_inc.new_pools[pool_id]; | |
4482 | } | |
4483 | ||
11fdf7f2 TL |
4484 | if (app_key.empty()) { |
4485 | p.application_metadata.insert({app_name, {}}); | |
4486 | } else { | |
4487 | p.application_metadata.insert({app_name, {{app_key, app_value}}}); | |
4488 | } | |
c07f9fc5 FG |
4489 | p.last_change = pending_inc.epoch; |
4490 | pending_inc.new_pools[pool_id] = p; | |
4491 | } | |
4492 | ||
494da23a TL |
4493 | void OSDMonitor::do_set_pool_opt(int64_t pool_id, |
4494 | pool_opts_t::key_t opt, | |
4495 | pool_opts_t::value_t val) | |
4496 | { | |
4497 | auto p = pending_inc.new_pools.try_emplace( | |
4498 | pool_id, *osdmap.get_pg_pool(pool_id)); | |
4499 | p.first->second.opts.set(opt, val); | |
4500 | } | |
4501 | ||
31f18b77 | 4502 | unsigned OSDMonitor::scan_for_creating_pgs( |
7c673cae FG |
4503 | const mempool::osdmap::map<int64_t,pg_pool_t>& pools, |
4504 | const mempool::osdmap::set<int64_t>& removed_pools, | |
4505 | utime_t modified, | |
4506 | creating_pgs_t* creating_pgs) const | |
4507 | { | |
31f18b77 | 4508 | unsigned queued = 0; |
7c673cae FG |
4509 | for (auto& p : pools) { |
4510 | int64_t poolid = p.first; | |
11fdf7f2 TL |
4511 | if (creating_pgs->created_pools.count(poolid)) { |
4512 | dout(10) << __func__ << " already created " << poolid << dendl; | |
4513 | continue; | |
4514 | } | |
7c673cae | 4515 | const pg_pool_t& pool = p.second; |
31f18b77 | 4516 | int ruleno = osdmap.crush->find_rule(pool.get_crush_rule(), |
7c673cae FG |
4517 | pool.get_type(), pool.get_size()); |
4518 | if (ruleno < 0 || !osdmap.crush->rule_exists(ruleno)) | |
4519 | continue; | |
4520 | ||
4521 | const auto last_scan_epoch = creating_pgs->last_scan_epoch; | |
4522 | const auto created = pool.get_last_change(); | |
4523 | if (last_scan_epoch && created <= last_scan_epoch) { | |
4524 | dout(10) << __func__ << " no change in pool " << poolid | |
4525 | << " " << pool << dendl; | |
4526 | continue; | |
4527 | } | |
4528 | if (removed_pools.count(poolid)) { | |
4529 | dout(10) << __func__ << " pool is being removed: " << poolid | |
4530 | << " " << pool << dendl; | |
4531 | continue; | |
4532 | } | |
31f18b77 | 4533 | dout(10) << __func__ << " queueing pool create for " << poolid |
7c673cae | 4534 | << " " << pool << dendl; |
11fdf7f2 TL |
4535 | creating_pgs->create_pool(poolid, pool.get_pg_num(), |
4536 | created, modified); | |
4537 | queued++; | |
7c673cae | 4538 | } |
31f18b77 | 4539 | return queued; |
7c673cae FG |
4540 | } |
4541 | ||
4542 | void OSDMonitor::update_creating_pgs() | |
4543 | { | |
31f18b77 FG |
4544 | dout(10) << __func__ << " " << creating_pgs.pgs.size() << " pgs creating, " |
4545 | << creating_pgs.queue.size() << " pools in queue" << dendl; | |
7c673cae FG |
4546 | decltype(creating_pgs_by_osd_epoch) new_pgs_by_osd_epoch; |
4547 | std::lock_guard<std::mutex> l(creating_pgs_lock); | |
c07f9fc5 | 4548 | for (const auto& pg : creating_pgs.pgs) { |
7c673cae FG |
4549 | int acting_primary = -1; |
4550 | auto pgid = pg.first; | |
94b18763 FG |
4551 | if (!osdmap.pg_exists(pgid)) { |
4552 | dout(20) << __func__ << " ignoring " << pgid << " which should not exist" | |
4553 | << dendl; | |
4554 | continue; | |
4555 | } | |
7c673cae | 4556 | auto mapped = pg.second.first; |
c07f9fc5 | 4557 | dout(20) << __func__ << " looking up " << pgid << "@" << mapped << dendl; |
11fdf7f2 TL |
4558 | spg_t spgid(pgid); |
4559 | mapping.get_primary_and_shard(pgid, &acting_primary, &spgid); | |
7c673cae FG |
4560 | // check the previous creating_pgs, look for the target to whom the pg was |
4561 | // previously mapped | |
4562 | for (const auto& pgs_by_epoch : creating_pgs_by_osd_epoch) { | |
4563 | const auto last_acting_primary = pgs_by_epoch.first; | |
4564 | for (auto& pgs: pgs_by_epoch.second) { | |
11fdf7f2 | 4565 | if (pgs.second.count(spgid)) { |
7c673cae FG |
4566 | if (last_acting_primary == acting_primary) { |
4567 | mapped = pgs.first; | |
4568 | } else { | |
4569 | dout(20) << __func__ << " " << pgid << " " | |
4570 | << " acting_primary:" << last_acting_primary | |
4571 | << " -> " << acting_primary << dendl; | |
4572 | // note epoch if the target of the create message changed. | |
4573 | mapped = mapping.get_epoch(); | |
4574 | } | |
4575 | break; | |
31f18b77 FG |
4576 | } else { |
4577 | // newly creating | |
4578 | mapped = mapping.get_epoch(); | |
4579 | } | |
7c673cae FG |
4580 | } |
4581 | } | |
4582 | dout(10) << __func__ << " will instruct osd." << acting_primary | |
c07f9fc5 | 4583 | << " to create " << pgid << "@" << mapped << dendl; |
11fdf7f2 | 4584 | new_pgs_by_osd_epoch[acting_primary][mapped].insert(spgid); |
7c673cae FG |
4585 | } |
4586 | creating_pgs_by_osd_epoch = std::move(new_pgs_by_osd_epoch); | |
4587 | creating_pgs_epoch = mapping.get_epoch(); | |
4588 | } | |
4589 | ||
c07f9fc5 | 4590 | epoch_t OSDMonitor::send_pg_creates(int osd, Connection *con, epoch_t next) const |
7c673cae FG |
4591 | { |
4592 | dout(30) << __func__ << " osd." << osd << " next=" << next | |
4593 | << " " << creating_pgs_by_osd_epoch << dendl; | |
4594 | std::lock_guard<std::mutex> l(creating_pgs_lock); | |
b5b8bbf5 FG |
4595 | if (creating_pgs_epoch <= creating_pgs.last_scan_epoch) { |
4596 | dout(20) << __func__ | |
4597 | << " not using stale creating_pgs@" << creating_pgs_epoch << dendl; | |
4598 | // the subscribers will be updated when the mapping is completed anyway | |
4599 | return next; | |
4600 | } | |
7c673cae FG |
4601 | auto creating_pgs_by_epoch = creating_pgs_by_osd_epoch.find(osd); |
4602 | if (creating_pgs_by_epoch == creating_pgs_by_osd_epoch.end()) | |
4603 | return next; | |
11fdf7f2 TL |
4604 | ceph_assert(!creating_pgs_by_epoch->second.empty()); |
4605 | ||
4606 | MOSDPGCreate *oldm = nullptr; // for pre-mimic OSD compat | |
4607 | MOSDPGCreate2 *m = nullptr; | |
4608 | ||
4609 | bool old = osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS; | |
7c673cae | 4610 | |
7c673cae FG |
4611 | epoch_t last = 0; |
4612 | for (auto epoch_pgs = creating_pgs_by_epoch->second.lower_bound(next); | |
4613 | epoch_pgs != creating_pgs_by_epoch->second.end(); ++epoch_pgs) { | |
4614 | auto epoch = epoch_pgs->first; | |
4615 | auto& pgs = epoch_pgs->second; | |
4616 | dout(20) << __func__ << " osd." << osd << " from " << next | |
4617 | << " : epoch " << epoch << " " << pgs.size() << " pgs" << dendl; | |
4618 | last = epoch; | |
4619 | for (auto& pg : pgs) { | |
7c673cae FG |
4620 | // Need the create time from the monitor using its clock to set |
4621 | // last_scrub_stamp upon pg creation. | |
11fdf7f2 TL |
4622 | auto create = creating_pgs.pgs.find(pg.pgid); |
4623 | ceph_assert(create != creating_pgs.pgs.end()); | |
4624 | if (old) { | |
4625 | if (!oldm) { | |
4626 | oldm = new MOSDPGCreate(creating_pgs_epoch); | |
4627 | } | |
4628 | oldm->mkpg.emplace(pg.pgid, | |
4629 | pg_create_t{create->second.first, pg.pgid, 0}); | |
4630 | oldm->ctimes.emplace(pg.pgid, create->second.second); | |
4631 | } else { | |
4632 | if (!m) { | |
4633 | m = new MOSDPGCreate2(creating_pgs_epoch); | |
4634 | } | |
4635 | m->pgs.emplace(pg, create->second); | |
4636 | } | |
7c673cae | 4637 | dout(20) << __func__ << " will create " << pg |
c07f9fc5 | 4638 | << " at " << create->second.first << dendl; |
7c673cae FG |
4639 | } |
4640 | } | |
11fdf7f2 TL |
4641 | if (m) { |
4642 | con->send_message(m); | |
4643 | } else if (oldm) { | |
4644 | con->send_message(oldm); | |
4645 | } else { | |
7c673cae FG |
4646 | dout(20) << __func__ << " osd." << osd << " from " << next |
4647 | << " has nothing to send" << dendl; | |
4648 | return next; | |
4649 | } | |
11fdf7f2 | 4650 | |
7c673cae FG |
4651 | // sub is current through last + 1 |
4652 | return last + 1; | |
4653 | } | |
4654 | ||
4655 | // TICK | |
4656 | ||
4657 | ||
4658 | void OSDMonitor::tick() | |
4659 | { | |
4660 | if (!is_active()) return; | |
4661 | ||
4662 | dout(10) << osdmap << dendl; | |
4663 | ||
11fdf7f2 TL |
4664 | // always update osdmap manifest, regardless of being the leader. |
4665 | load_osdmap_manifest(); | |
4666 | ||
7c673cae FG |
4667 | if (!mon->is_leader()) return; |
4668 | ||
4669 | bool do_propose = false; | |
4670 | utime_t now = ceph_clock_now(); | |
4671 | ||
11fdf7f2 | 4672 | if (handle_osd_timeouts(now, last_osd_report)) { |
181888fb FG |
4673 | do_propose = true; |
4674 | } | |
7c673cae FG |
4675 | |
4676 | // mark osds down? | |
11fdf7f2 | 4677 | if (check_failures(now)) { |
7c673cae | 4678 | do_propose = true; |
11fdf7f2 TL |
4679 | } |
4680 | ||
4681 | // Force a proposal if we need to prune; pruning is performed on | |
4682 | // ``encode_pending()``, hence why we need to regularly trigger a proposal | |
4683 | // even if there's nothing going on. | |
4684 | if (is_prune_enabled() && should_prune()) { | |
4685 | do_propose = true; | |
4686 | } | |
7c673cae FG |
4687 | |
4688 | // mark down osds out? | |
4689 | ||
4690 | /* can_mark_out() checks if we can mark osds as being out. The -1 has no | |
4691 | * influence at all. The decision is made based on the ratio of "in" osds, | |
4692 | * and the function returns false if this ratio is lower that the minimum | |
11fdf7f2 | 4693 | * ratio set by g_conf()->mon_osd_min_in_ratio. So it's not really up to us. |
7c673cae FG |
4694 | */ |
4695 | if (can_mark_out(-1)) { | |
11fdf7f2 TL |
4696 | string down_out_subtree_limit = g_conf().get_val<string>( |
4697 | "mon_osd_down_out_subtree_limit"); | |
7c673cae FG |
4698 | set<int> down_cache; // quick cache of down subtrees |
4699 | ||
4700 | map<int,utime_t>::iterator i = down_pending_out.begin(); | |
4701 | while (i != down_pending_out.end()) { | |
4702 | int o = i->first; | |
4703 | utime_t down = now; | |
4704 | down -= i->second; | |
4705 | ++i; | |
4706 | ||
4707 | if (osdmap.is_down(o) && | |
4708 | osdmap.is_in(o) && | |
4709 | can_mark_out(o)) { | |
11fdf7f2 | 4710 | utime_t orig_grace(g_conf()->mon_osd_down_out_interval, 0); |
7c673cae FG |
4711 | utime_t grace = orig_grace; |
4712 | double my_grace = 0.0; | |
4713 | ||
11fdf7f2 | 4714 | if (g_conf()->mon_osd_adjust_down_out_interval) { |
7c673cae FG |
4715 | // scale grace period the same way we do the heartbeat grace. |
4716 | const osd_xinfo_t& xi = osdmap.get_xinfo(o); | |
11fdf7f2 | 4717 | double halflife = (double)g_conf()->mon_osd_laggy_halflife; |
7c673cae FG |
4718 | double decay_k = ::log(.5) / halflife; |
4719 | double decay = exp((double)down * decay_k); | |
4720 | dout(20) << "osd." << o << " laggy halflife " << halflife << " decay_k " << decay_k | |
4721 | << " down for " << down << " decay " << decay << dendl; | |
4722 | my_grace = decay * (double)xi.laggy_interval * xi.laggy_probability; | |
4723 | grace += my_grace; | |
4724 | } | |
4725 | ||
4726 | // is this an entire large subtree down? | |
11fdf7f2 TL |
4727 | if (down_out_subtree_limit.length()) { |
4728 | int type = osdmap.crush->get_type_id(down_out_subtree_limit); | |
7c673cae | 4729 | if (type > 0) { |
11fdf7f2 TL |
4730 | if (osdmap.containing_subtree_is_down(cct, o, type, &down_cache)) { |
4731 | dout(10) << "tick entire containing " << down_out_subtree_limit | |
4732 | << " subtree for osd." << o | |
4733 | << " is down; resetting timer" << dendl; | |
7c673cae FG |
4734 | // reset timer, too. |
4735 | down_pending_out[o] = now; | |
4736 | continue; | |
4737 | } | |
4738 | } | |
4739 | } | |
4740 | ||
c07f9fc5 | 4741 | bool down_out = !osdmap.is_destroyed(o) && |
11fdf7f2 | 4742 | g_conf()->mon_osd_down_out_interval > 0 && down.sec() >= grace; |
c07f9fc5 | 4743 | bool destroyed_out = osdmap.is_destroyed(o) && |
11fdf7f2 | 4744 | g_conf()->mon_osd_destroyed_out_interval > 0 && |
c07f9fc5 FG |
4745 | // this is not precise enough as we did not make a note when this osd |
4746 | // was marked as destroyed, but let's not bother with that | |
4747 | // complexity for now. | |
11fdf7f2 | 4748 | down.sec() >= g_conf()->mon_osd_destroyed_out_interval; |
c07f9fc5 | 4749 | if (down_out || destroyed_out) { |
7c673cae FG |
4750 | dout(10) << "tick marking osd." << o << " OUT after " << down |
4751 | << " sec (target " << grace << " = " << orig_grace << " + " << my_grace << ")" << dendl; | |
4752 | pending_inc.new_weight[o] = CEPH_OSD_OUT; | |
4753 | ||
4754 | // set the AUTOOUT bit. | |
4755 | if (pending_inc.new_state.count(o) == 0) | |
4756 | pending_inc.new_state[o] = 0; | |
4757 | pending_inc.new_state[o] |= CEPH_OSD_AUTOOUT; | |
4758 | ||
4759 | // remember previous weight | |
4760 | if (pending_inc.new_xinfo.count(o) == 0) | |
4761 | pending_inc.new_xinfo[o] = osdmap.osd_xinfo[o]; | |
4762 | pending_inc.new_xinfo[o].old_weight = osdmap.osd_weight[o]; | |
4763 | ||
4764 | do_propose = true; | |
4765 | ||
224ce89b WB |
4766 | mon->clog->info() << "Marking osd." << o << " out (has been down for " |
4767 | << int(down.sec()) << " seconds)"; | |
7c673cae FG |
4768 | } else |
4769 | continue; | |
4770 | } | |
4771 | ||
4772 | down_pending_out.erase(o); | |
4773 | } | |
4774 | } else { | |
4775 | dout(10) << "tick NOOUT flag set, not checking down osds" << dendl; | |
4776 | } | |
4777 | ||
4778 | // expire blacklisted items? | |
4779 | for (ceph::unordered_map<entity_addr_t,utime_t>::iterator p = osdmap.blacklist.begin(); | |
4780 | p != osdmap.blacklist.end(); | |
4781 | ++p) { | |
4782 | if (p->second < now) { | |
4783 | dout(10) << "expiring blacklist item " << p->first << " expired " << p->second << " < now " << now << dendl; | |
4784 | pending_inc.old_blacklist.push_back(p->first); | |
4785 | do_propose = true; | |
4786 | } | |
4787 | } | |
4788 | ||
11fdf7f2 TL |
4789 | if (try_prune_purged_snaps()) { |
4790 | do_propose = true; | |
7c673cae FG |
4791 | } |
4792 | ||
4793 | if (update_pools_status()) | |
4794 | do_propose = true; | |
4795 | ||
4796 | if (do_propose || | |
4797 | !pending_inc.new_pg_temp.empty()) // also propose if we adjusted pg_temp | |
4798 | propose_pending(); | |
eafe8130 TL |
4799 | |
4800 | { | |
4801 | std::lock_guard l(balancer_lock); | |
4802 | if (ceph_using_tcmalloc() && mon_memory_autotune && pcm != nullptr) { | |
4803 | pcm->tune_memory(); | |
4804 | pcm->balance(); | |
4805 | _set_new_cache_sizes(); | |
4806 | dout(10) << "tick balancer " | |
4807 | << " inc cache_bytes: " << inc_cache->get_cache_bytes() | |
4808 | << " inc comtd_bytes: " << inc_cache->get_committed_size() | |
4809 | << " inc used_bytes: " << inc_cache->_get_used_bytes() | |
4810 | << " inc num_osdmaps: " << inc_cache->_get_num_osdmaps() | |
4811 | << dendl; | |
4812 | dout(10) << "tick balancer " | |
4813 | << " full cache_bytes: " << full_cache->get_cache_bytes() | |
4814 | << " full comtd_bytes: " << full_cache->get_committed_size() | |
4815 | << " full used_bytes: " << full_cache->_get_used_bytes() | |
4816 | << " full num_osdmaps: " << full_cache->_get_num_osdmaps() | |
4817 | << dendl; | |
4818 | } | |
4819 | } | |
4820 | } | |
4821 | ||
4822 | void OSDMonitor::_set_new_cache_sizes() | |
4823 | { | |
4824 | uint64_t cache_size = 0; | |
4825 | int64_t inc_alloc = 0; | |
4826 | int64_t full_alloc = 0; | |
4827 | int64_t kv_alloc = 0; | |
4828 | ||
4829 | if (pcm != nullptr && rocksdb_binned_kv_cache != nullptr) { | |
4830 | cache_size = pcm->get_tuned_mem(); | |
4831 | inc_alloc = inc_cache->get_committed_size(); | |
4832 | full_alloc = full_cache->get_committed_size(); | |
4833 | kv_alloc = rocksdb_binned_kv_cache->get_committed_size(); | |
4834 | } | |
4835 | ||
4836 | inc_osd_cache.set_bytes(inc_alloc); | |
4837 | full_osd_cache.set_bytes(full_alloc); | |
4838 | ||
4839 | dout(10) << __func__ << " cache_size:" << cache_size | |
4840 | << " inc_alloc: " << inc_alloc | |
4841 | << " full_alloc: " << full_alloc | |
4842 | << " kv_alloc: " << kv_alloc | |
4843 | << dendl; | |
7c673cae FG |
4844 | } |
4845 | ||
4846 | bool OSDMonitor::handle_osd_timeouts(const utime_t &now, | |
4847 | std::map<int,utime_t> &last_osd_report) | |
4848 | { | |
11fdf7f2 | 4849 | utime_t timeo(g_conf()->mon_osd_report_timeout, 0); |
7c673cae FG |
4850 | if (now - mon->get_leader_since() < timeo) { |
4851 | // We haven't been the leader for long enough to consider OSD timeouts | |
4852 | return false; | |
4853 | } | |
4854 | ||
4855 | int max_osd = osdmap.get_max_osd(); | |
4856 | bool new_down = false; | |
4857 | ||
4858 | for (int i=0; i < max_osd; ++i) { | |
4859 | dout(30) << __func__ << ": checking up on osd " << i << dendl; | |
c07f9fc5 FG |
4860 | if (!osdmap.exists(i)) { |
4861 | last_osd_report.erase(i); // if any | |
4862 | continue; | |
4863 | } | |
7c673cae FG |
4864 | if (!osdmap.is_up(i)) |
4865 | continue; | |
4866 | const std::map<int,utime_t>::const_iterator t = last_osd_report.find(i); | |
4867 | if (t == last_osd_report.end()) { | |
4868 | // it wasn't in the map; start the timer. | |
4869 | last_osd_report[i] = now; | |
4870 | } else if (can_mark_down(i)) { | |
4871 | utime_t diff = now - t->second; | |
4872 | if (diff > timeo) { | |
31f18b77 FG |
4873 | mon->clog->info() << "osd." << i << " marked down after no beacon for " |
4874 | << diff << " seconds"; | |
4875 | derr << "no beacon from osd." << i << " since " << t->second | |
4876 | << ", " << diff << " seconds ago. marking down" << dendl; | |
7c673cae FG |
4877 | pending_inc.new_state[i] = CEPH_OSD_UP; |
4878 | new_down = true; | |
4879 | } | |
4880 | } | |
4881 | } | |
4882 | return new_down; | |
4883 | } | |
4884 | ||
11fdf7f2 TL |
4885 | static void dump_cpu_list(Formatter *f, const char *name, |
4886 | const string& strlist) | |
7c673cae | 4887 | { |
11fdf7f2 TL |
4888 | cpu_set_t cpu_set; |
4889 | size_t cpu_set_size; | |
4890 | if (parse_cpu_set_list(strlist.c_str(), &cpu_set_size, &cpu_set) < 0) { | |
4891 | return; | |
4892 | } | |
4893 | set<int> cpus = cpu_set_to_set(cpu_set_size, &cpu_set); | |
4894 | f->open_array_section(name); | |
4895 | for (auto cpu : cpus) { | |
4896 | f->dump_int("cpu", cpu); | |
7c673cae | 4897 | } |
11fdf7f2 | 4898 | f->close_section(); |
7c673cae FG |
4899 | } |
4900 | ||
4901 | void OSDMonitor::dump_info(Formatter *f) | |
4902 | { | |
4903 | f->open_object_section("osdmap"); | |
4904 | osdmap.dump(f); | |
4905 | f->close_section(); | |
4906 | ||
4907 | f->open_array_section("osd_metadata"); | |
4908 | for (int i=0; i<osdmap.get_max_osd(); ++i) { | |
4909 | if (osdmap.exists(i)) { | |
4910 | f->open_object_section("osd"); | |
4911 | f->dump_unsigned("id", i); | |
4912 | dump_osd_metadata(i, f, NULL); | |
4913 | f->close_section(); | |
4914 | } | |
4915 | } | |
4916 | f->close_section(); | |
4917 | ||
4918 | f->dump_unsigned("osdmap_first_committed", get_first_committed()); | |
4919 | f->dump_unsigned("osdmap_last_committed", get_last_committed()); | |
4920 | ||
4921 | f->open_object_section("crushmap"); | |
4922 | osdmap.crush->dump(f); | |
4923 | f->close_section(); | |
11fdf7f2 TL |
4924 | |
4925 | if (has_osdmap_manifest) { | |
4926 | f->open_object_section("osdmap_manifest"); | |
4927 | osdmap_manifest.dump(f); | |
4928 | f->close_section(); | |
4929 | } | |
7c673cae FG |
4930 | } |
4931 | ||
4932 | namespace { | |
4933 | enum osd_pool_get_choices { | |
11fdf7f2 | 4934 | SIZE, MIN_SIZE, |
28e407b8 | 4935 | PG_NUM, PGP_NUM, CRUSH_RULE, HASHPSPOOL, EC_OVERWRITES, |
7c673cae FG |
4936 | NODELETE, NOPGCHANGE, NOSIZECHANGE, |
4937 | WRITE_FADVISE_DONTNEED, NOSCRUB, NODEEP_SCRUB, | |
4938 | HIT_SET_TYPE, HIT_SET_PERIOD, HIT_SET_COUNT, HIT_SET_FPP, | |
11fdf7f2 | 4939 | USE_GMT_HITSET, TARGET_MAX_OBJECTS, TARGET_MAX_BYTES, |
7c673cae FG |
4940 | CACHE_TARGET_DIRTY_RATIO, CACHE_TARGET_DIRTY_HIGH_RATIO, |
4941 | CACHE_TARGET_FULL_RATIO, | |
4942 | CACHE_MIN_FLUSH_AGE, CACHE_MIN_EVICT_AGE, | |
4943 | ERASURE_CODE_PROFILE, MIN_READ_RECENCY_FOR_PROMOTE, | |
4944 | MIN_WRITE_RECENCY_FOR_PROMOTE, FAST_READ, | |
4945 | HIT_SET_GRADE_DECAY_RATE, HIT_SET_SEARCH_LAST_N, | |
4946 | SCRUB_MIN_INTERVAL, SCRUB_MAX_INTERVAL, DEEP_SCRUB_INTERVAL, | |
4947 | RECOVERY_PRIORITY, RECOVERY_OP_PRIORITY, SCRUB_PRIORITY, | |
4948 | COMPRESSION_MODE, COMPRESSION_ALGORITHM, COMPRESSION_REQUIRED_RATIO, | |
4949 | COMPRESSION_MAX_BLOB_SIZE, COMPRESSION_MIN_BLOB_SIZE, | |
11fdf7f2 TL |
4950 | CSUM_TYPE, CSUM_MAX_BLOCK, CSUM_MIN_BLOCK, FINGERPRINT_ALGORITHM, |
4951 | PG_AUTOSCALE_MODE, PG_NUM_MIN, TARGET_SIZE_BYTES, TARGET_SIZE_RATIO, | |
4952 | PG_AUTOSCALE_BIAS }; | |
7c673cae FG |
4953 | |
4954 | std::set<osd_pool_get_choices> | |
4955 | subtract_second_from_first(const std::set<osd_pool_get_choices>& first, | |
4956 | const std::set<osd_pool_get_choices>& second) | |
4957 | { | |
4958 | std::set<osd_pool_get_choices> result; | |
4959 | std::set_difference(first.begin(), first.end(), | |
4960 | second.begin(), second.end(), | |
4961 | std::inserter(result, result.end())); | |
4962 | return result; | |
4963 | } | |
4964 | } | |
4965 | ||
4966 | ||
4967 | bool OSDMonitor::preprocess_command(MonOpRequestRef op) | |
4968 | { | |
4969 | op->mark_osdmon_event(__func__); | |
4970 | MMonCommand *m = static_cast<MMonCommand*>(op->get_req()); | |
4971 | int r = 0; | |
4972 | bufferlist rdata; | |
4973 | stringstream ss, ds; | |
4974 | ||
11fdf7f2 | 4975 | cmdmap_t cmdmap; |
7c673cae FG |
4976 | if (!cmdmap_from_json(m->cmd, &cmdmap, ss)) { |
4977 | string rs = ss.str(); | |
4978 | mon->reply_command(op, -EINVAL, rs, get_last_committed()); | |
4979 | return true; | |
4980 | } | |
4981 | ||
11fdf7f2 | 4982 | MonSession *session = op->get_session(); |
7c673cae | 4983 | if (!session) { |
11fdf7f2 | 4984 | derr << __func__ << " no session" << dendl; |
7c673cae FG |
4985 | mon->reply_command(op, -EACCES, "access denied", get_last_committed()); |
4986 | return true; | |
4987 | } | |
4988 | ||
4989 | string prefix; | |
11fdf7f2 | 4990 | cmd_getval(cct, cmdmap, "prefix", prefix); |
7c673cae FG |
4991 | |
4992 | string format; | |
11fdf7f2 | 4993 | cmd_getval(cct, cmdmap, "format", format, string("plain")); |
7c673cae FG |
4994 | boost::scoped_ptr<Formatter> f(Formatter::create(format)); |
4995 | ||
4996 | if (prefix == "osd stat") { | |
11fdf7f2 | 4997 | osdmap.print_summary(f.get(), ds, "", true); |
7c673cae FG |
4998 | if (f) |
4999 | f->flush(rdata); | |
5000 | else | |
5001 | rdata.append(ds); | |
5002 | } | |
7c673cae FG |
5003 | else if (prefix == "osd dump" || |
5004 | prefix == "osd tree" || | |
11fdf7f2 | 5005 | prefix == "osd tree-from" || |
7c673cae FG |
5006 | prefix == "osd ls" || |
5007 | prefix == "osd getmap" || | |
31f18b77 FG |
5008 | prefix == "osd getcrushmap" || |
5009 | prefix == "osd ls-tree") { | |
7c673cae FG |
5010 | string val; |
5011 | ||
5012 | epoch_t epoch = 0; | |
5013 | int64_t epochnum; | |
11fdf7f2 | 5014 | cmd_getval(cct, cmdmap, "epoch", epochnum, (int64_t)osdmap.get_epoch()); |
7c673cae FG |
5015 | epoch = epochnum; |
5016 | ||
5017 | bufferlist osdmap_bl; | |
5018 | int err = get_version_full(epoch, osdmap_bl); | |
5019 | if (err == -ENOENT) { | |
5020 | r = -ENOENT; | |
5021 | ss << "there is no map for epoch " << epoch; | |
5022 | goto reply; | |
5023 | } | |
11fdf7f2 TL |
5024 | ceph_assert(err == 0); |
5025 | ceph_assert(osdmap_bl.length()); | |
7c673cae FG |
5026 | |
5027 | OSDMap *p; | |
5028 | if (epoch == osdmap.get_epoch()) { | |
5029 | p = &osdmap; | |
5030 | } else { | |
5031 | p = new OSDMap; | |
5032 | p->decode(osdmap_bl); | |
5033 | } | |
5034 | ||
224ce89b WB |
5035 | auto sg = make_scope_guard([&] { |
5036 | if (p != &osdmap) { | |
5037 | delete p; | |
5038 | } | |
5039 | }); | |
5040 | ||
7c673cae FG |
5041 | if (prefix == "osd dump") { |
5042 | stringstream ds; | |
5043 | if (f) { | |
5044 | f->open_object_section("osdmap"); | |
5045 | p->dump(f.get()); | |
5046 | f->close_section(); | |
5047 | f->flush(ds); | |
5048 | } else { | |
5049 | p->print(ds); | |
5050 | } | |
5051 | rdata.append(ds); | |
5052 | if (!f) | |
5053 | ds << " "; | |
5054 | } else if (prefix == "osd ls") { | |
5055 | if (f) { | |
5056 | f->open_array_section("osds"); | |
5057 | for (int i = 0; i < osdmap.get_max_osd(); i++) { | |
5058 | if (osdmap.exists(i)) { | |
5059 | f->dump_int("osd", i); | |
5060 | } | |
5061 | } | |
5062 | f->close_section(); | |
5063 | f->flush(ds); | |
5064 | } else { | |
5065 | bool first = true; | |
5066 | for (int i = 0; i < osdmap.get_max_osd(); i++) { | |
5067 | if (osdmap.exists(i)) { | |
5068 | if (!first) | |
5069 | ds << "\n"; | |
5070 | first = false; | |
5071 | ds << i; | |
5072 | } | |
5073 | } | |
5074 | } | |
5075 | rdata.append(ds); | |
11fdf7f2 TL |
5076 | } else if (prefix == "osd tree" || prefix == "osd tree-from") { |
5077 | string bucket; | |
5078 | if (prefix == "osd tree-from") { | |
5079 | cmd_getval(cct, cmdmap, "bucket", bucket); | |
5080 | if (!osdmap.crush->name_exists(bucket)) { | |
5081 | ss << "bucket '" << bucket << "' does not exist"; | |
5082 | r = -ENOENT; | |
5083 | goto reply; | |
5084 | } | |
5085 | int id = osdmap.crush->get_item_id(bucket); | |
5086 | if (id >= 0) { | |
5087 | ss << "\"" << bucket << "\" is not a bucket"; | |
5088 | r = -EINVAL; | |
5089 | goto reply; | |
5090 | } | |
5091 | } | |
5092 | ||
31f18b77 | 5093 | vector<string> states; |
11fdf7f2 | 5094 | cmd_getval(cct, cmdmap, "states", states); |
31f18b77 FG |
5095 | unsigned filter = 0; |
5096 | for (auto& s : states) { | |
5097 | if (s == "up") { | |
5098 | filter |= OSDMap::DUMP_UP; | |
5099 | } else if (s == "down") { | |
5100 | filter |= OSDMap::DUMP_DOWN; | |
5101 | } else if (s == "in") { | |
5102 | filter |= OSDMap::DUMP_IN; | |
5103 | } else if (s == "out") { | |
5104 | filter |= OSDMap::DUMP_OUT; | |
c07f9fc5 FG |
5105 | } else if (s == "destroyed") { |
5106 | filter |= OSDMap::DUMP_DESTROYED; | |
31f18b77 FG |
5107 | } else { |
5108 | ss << "unrecognized state '" << s << "'"; | |
5109 | r = -EINVAL; | |
5110 | goto reply; | |
5111 | } | |
5112 | } | |
5113 | if ((filter & (OSDMap::DUMP_IN|OSDMap::DUMP_OUT)) == | |
c07f9fc5 FG |
5114 | (OSDMap::DUMP_IN|OSDMap::DUMP_OUT)) { |
5115 | ss << "cannot specify both 'in' and 'out'"; | |
5116 | r = -EINVAL; | |
5117 | goto reply; | |
5118 | } | |
5119 | if (((filter & (OSDMap::DUMP_UP|OSDMap::DUMP_DOWN)) == | |
5120 | (OSDMap::DUMP_UP|OSDMap::DUMP_DOWN)) || | |
5121 | ((filter & (OSDMap::DUMP_UP|OSDMap::DUMP_DESTROYED)) == | |
5122 | (OSDMap::DUMP_UP|OSDMap::DUMP_DESTROYED)) || | |
5123 | ((filter & (OSDMap::DUMP_DOWN|OSDMap::DUMP_DESTROYED)) == | |
5124 | (OSDMap::DUMP_DOWN|OSDMap::DUMP_DESTROYED))) { | |
5125 | ss << "can specify only one of 'up', 'down' and 'destroyed'"; | |
31f18b77 FG |
5126 | r = -EINVAL; |
5127 | goto reply; | |
5128 | } | |
7c673cae FG |
5129 | if (f) { |
5130 | f->open_object_section("tree"); | |
11fdf7f2 | 5131 | p->print_tree(f.get(), NULL, filter, bucket); |
7c673cae FG |
5132 | f->close_section(); |
5133 | f->flush(ds); | |
5134 | } else { | |
11fdf7f2 | 5135 | p->print_tree(NULL, &ds, filter, bucket); |
7c673cae FG |
5136 | } |
5137 | rdata.append(ds); | |
5138 | } else if (prefix == "osd getmap") { | |
5139 | rdata.append(osdmap_bl); | |
5140 | ss << "got osdmap epoch " << p->get_epoch(); | |
5141 | } else if (prefix == "osd getcrushmap") { | |
5142 | p->crush->encode(rdata, mon->get_quorum_con_features()); | |
31f18b77 FG |
5143 | ss << p->get_crush_version(); |
5144 | } else if (prefix == "osd ls-tree") { | |
5145 | string bucket_name; | |
11fdf7f2 | 5146 | cmd_getval(cct, cmdmap, "name", bucket_name); |
31f18b77 FG |
5147 | set<int> osds; |
5148 | r = p->get_osds_by_bucket_name(bucket_name, &osds); | |
5149 | if (r == -ENOENT) { | |
5150 | ss << "\"" << bucket_name << "\" does not exist"; | |
5151 | goto reply; | |
5152 | } else if (r < 0) { | |
5153 | ss << "can not parse bucket name:\"" << bucket_name << "\""; | |
5154 | goto reply; | |
5155 | } | |
5156 | ||
5157 | if (f) { | |
5158 | f->open_array_section("osds"); | |
5159 | for (auto &i : osds) { | |
5160 | if (osdmap.exists(i)) { | |
5161 | f->dump_int("osd", i); | |
5162 | } | |
5163 | } | |
5164 | f->close_section(); | |
5165 | f->flush(ds); | |
5166 | } else { | |
5167 | bool first = true; | |
5168 | for (auto &i : osds) { | |
5169 | if (osdmap.exists(i)) { | |
5170 | if (!first) | |
5171 | ds << "\n"; | |
5172 | first = false; | |
5173 | ds << i; | |
5174 | } | |
5175 | } | |
5176 | } | |
5177 | ||
5178 | rdata.append(ds); | |
7c673cae | 5179 | } |
7c673cae FG |
5180 | } else if (prefix == "osd getmaxosd") { |
5181 | if (f) { | |
5182 | f->open_object_section("getmaxosd"); | |
5183 | f->dump_unsigned("epoch", osdmap.get_epoch()); | |
5184 | f->dump_int("max_osd", osdmap.get_max_osd()); | |
5185 | f->close_section(); | |
5186 | f->flush(rdata); | |
5187 | } else { | |
5188 | ds << "max_osd = " << osdmap.get_max_osd() << " in epoch " << osdmap.get_epoch(); | |
5189 | rdata.append(ds); | |
5190 | } | |
5191 | } else if (prefix == "osd utilization") { | |
5192 | string out; | |
5193 | osdmap.summarize_mapping_stats(NULL, NULL, &out, f.get()); | |
5194 | if (f) | |
5195 | f->flush(rdata); | |
5196 | else | |
5197 | rdata.append(out); | |
5198 | r = 0; | |
5199 | goto reply; | |
5200 | } else if (prefix == "osd find") { | |
5201 | int64_t osd; | |
11fdf7f2 | 5202 | if (!cmd_getval(cct, cmdmap, "id", osd)) { |
7c673cae FG |
5203 | ss << "unable to parse osd id value '" |
5204 | << cmd_vartype_stringify(cmdmap["id"]) << "'"; | |
5205 | r = -EINVAL; | |
5206 | goto reply; | |
5207 | } | |
5208 | if (!osdmap.exists(osd)) { | |
5209 | ss << "osd." << osd << " does not exist"; | |
5210 | r = -ENOENT; | |
5211 | goto reply; | |
5212 | } | |
5213 | string format; | |
11fdf7f2 | 5214 | cmd_getval(cct, cmdmap, "format", format); |
7c673cae FG |
5215 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); |
5216 | f->open_object_section("osd_location"); | |
5217 | f->dump_int("osd", osd); | |
11fdf7f2 | 5218 | f->dump_object("addrs", osdmap.get_addrs(osd)); |
f64942e4 | 5219 | f->dump_stream("osd_fsid") << osdmap.get_uuid(osd); |
11fdf7f2 TL |
5220 | |
5221 | // try to identify host, pod/container name, etc. | |
5222 | map<string,string> m; | |
5223 | load_metadata(osd, m, nullptr); | |
5224 | if (auto p = m.find("hostname"); p != m.end()) { | |
5225 | f->dump_string("host", p->second); | |
5226 | } | |
5227 | for (auto& k : { | |
5228 | "pod_name", "pod_namespace", // set by rook | |
5229 | "container_name" // set by ceph-ansible | |
5230 | }) { | |
5231 | if (auto p = m.find(k); p != m.end()) { | |
5232 | f->dump_string(k, p->second); | |
5233 | } | |
5234 | } | |
5235 | ||
5236 | // crush is helpful too | |
7c673cae FG |
5237 | f->open_object_section("crush_location"); |
5238 | map<string,string> loc = osdmap.crush->get_full_location(osd); | |
5239 | for (map<string,string>::iterator p = loc.begin(); p != loc.end(); ++p) | |
5240 | f->dump_string(p->first.c_str(), p->second); | |
5241 | f->close_section(); | |
5242 | f->close_section(); | |
5243 | f->flush(rdata); | |
5244 | } else if (prefix == "osd metadata") { | |
5245 | int64_t osd = -1; | |
5246 | if (cmd_vartype_stringify(cmdmap["id"]).size() && | |
11fdf7f2 | 5247 | !cmd_getval(cct, cmdmap, "id", osd)) { |
7c673cae FG |
5248 | ss << "unable to parse osd id value '" |
5249 | << cmd_vartype_stringify(cmdmap["id"]) << "'"; | |
5250 | r = -EINVAL; | |
5251 | goto reply; | |
5252 | } | |
5253 | if (osd >= 0 && !osdmap.exists(osd)) { | |
5254 | ss << "osd." << osd << " does not exist"; | |
5255 | r = -ENOENT; | |
5256 | goto reply; | |
5257 | } | |
5258 | string format; | |
11fdf7f2 | 5259 | cmd_getval(cct, cmdmap, "format", format); |
7c673cae FG |
5260 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); |
5261 | if (osd >= 0) { | |
5262 | f->open_object_section("osd_metadata"); | |
5263 | f->dump_unsigned("id", osd); | |
5264 | r = dump_osd_metadata(osd, f.get(), &ss); | |
5265 | if (r < 0) | |
5266 | goto reply; | |
5267 | f->close_section(); | |
5268 | } else { | |
5269 | r = 0; | |
5270 | f->open_array_section("osd_metadata"); | |
5271 | for (int i=0; i<osdmap.get_max_osd(); ++i) { | |
5272 | if (osdmap.exists(i)) { | |
5273 | f->open_object_section("osd"); | |
5274 | f->dump_unsigned("id", i); | |
5275 | r = dump_osd_metadata(i, f.get(), NULL); | |
5276 | if (r == -EINVAL || r == -ENOENT) { | |
5277 | // Drop error, continue to get other daemons' metadata | |
5278 | dout(4) << "No metadata for osd." << i << dendl; | |
5279 | r = 0; | |
5280 | } else if (r < 0) { | |
5281 | // Unexpected error | |
5282 | goto reply; | |
5283 | } | |
5284 | f->close_section(); | |
5285 | } | |
5286 | } | |
5287 | f->close_section(); | |
5288 | } | |
5289 | f->flush(rdata); | |
31f18b77 FG |
5290 | } else if (prefix == "osd versions") { |
5291 | if (!f) | |
5292 | f.reset(Formatter::create("json-pretty")); | |
5293 | count_metadata("ceph_version", f.get()); | |
5294 | f->flush(rdata); | |
5295 | r = 0; | |
5296 | } else if (prefix == "osd count-metadata") { | |
5297 | if (!f) | |
5298 | f.reset(Formatter::create("json-pretty")); | |
5299 | string field; | |
11fdf7f2 | 5300 | cmd_getval(cct, cmdmap, "property", field); |
31f18b77 FG |
5301 | count_metadata(field, f.get()); |
5302 | f->flush(rdata); | |
5303 | r = 0; | |
11fdf7f2 TL |
5304 | } else if (prefix == "osd numa-status") { |
5305 | TextTable tbl; | |
5306 | if (f) { | |
5307 | f->open_array_section("osds"); | |
5308 | } else { | |
5309 | tbl.define_column("OSD", TextTable::LEFT, TextTable::RIGHT); | |
5310 | tbl.define_column("HOST", TextTable::LEFT, TextTable::LEFT); | |
5311 | tbl.define_column("NETWORK", TextTable::RIGHT, TextTable::RIGHT); | |
5312 | tbl.define_column("STORAGE", TextTable::RIGHT, TextTable::RIGHT); | |
5313 | tbl.define_column("AFFINITY", TextTable::RIGHT, TextTable::RIGHT); | |
5314 | tbl.define_column("CPUS", TextTable::LEFT, TextTable::LEFT); | |
5315 | } | |
5316 | for (int i=0; i<osdmap.get_max_osd(); ++i) { | |
5317 | if (osdmap.exists(i)) { | |
5318 | map<string,string> m; | |
5319 | ostringstream err; | |
5320 | if (load_metadata(i, m, &err) < 0) { | |
5321 | continue; | |
5322 | } | |
5323 | string host; | |
5324 | auto p = m.find("hostname"); | |
5325 | if (p != m.end()) { | |
5326 | host = p->second; | |
5327 | } | |
5328 | if (f) { | |
5329 | f->open_object_section("osd"); | |
5330 | f->dump_int("osd", i); | |
5331 | f->dump_string("host", host); | |
5332 | for (auto n : { "network_numa_node", "objectstore_numa_node", | |
5333 | "numa_node" }) { | |
5334 | p = m.find(n); | |
5335 | if (p != m.end()) { | |
5336 | f->dump_int(n, atoi(p->second.c_str())); | |
5337 | } | |
5338 | } | |
5339 | for (auto n : { "network_numa_nodes", "objectstore_numa_nodes" }) { | |
5340 | p = m.find(n); | |
5341 | if (p != m.end()) { | |
5342 | list<string> ls = get_str_list(p->second, ","); | |
5343 | f->open_array_section(n); | |
5344 | for (auto node : ls) { | |
5345 | f->dump_int("node", atoi(node.c_str())); | |
5346 | } | |
5347 | f->close_section(); | |
5348 | } | |
5349 | } | |
5350 | for (auto n : { "numa_node_cpus" }) { | |
5351 | p = m.find(n); | |
5352 | if (p != m.end()) { | |
5353 | dump_cpu_list(f.get(), n, p->second); | |
5354 | } | |
5355 | } | |
5356 | f->close_section(); | |
5357 | } else { | |
5358 | tbl << i; | |
5359 | tbl << host; | |
5360 | p = m.find("network_numa_nodes"); | |
5361 | if (p != m.end()) { | |
5362 | tbl << p->second; | |
5363 | } else { | |
5364 | tbl << "-"; | |
5365 | } | |
5366 | p = m.find("objectstore_numa_nodes"); | |
5367 | if (p != m.end()) { | |
5368 | tbl << p->second; | |
5369 | } else { | |
5370 | tbl << "-"; | |
5371 | } | |
5372 | p = m.find("numa_node"); | |
5373 | auto q = m.find("numa_node_cpus"); | |
5374 | if (p != m.end() && q != m.end()) { | |
5375 | tbl << p->second; | |
5376 | tbl << q->second; | |
5377 | } else { | |
5378 | tbl << "-"; | |
5379 | tbl << "-"; | |
5380 | } | |
5381 | tbl << TextTable::endrow; | |
5382 | } | |
5383 | } | |
5384 | } | |
5385 | if (f) { | |
5386 | f->close_section(); | |
5387 | f->flush(rdata); | |
5388 | } else { | |
5389 | rdata.append(stringify(tbl)); | |
5390 | } | |
7c673cae FG |
5391 | } else if (prefix == "osd map") { |
5392 | string poolstr, objstr, namespacestr; | |
11fdf7f2 TL |
5393 | cmd_getval(cct, cmdmap, "pool", poolstr); |
5394 | cmd_getval(cct, cmdmap, "object", objstr); | |
5395 | cmd_getval(cct, cmdmap, "nspace", namespacestr); | |
7c673cae FG |
5396 | |
5397 | int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str()); | |
5398 | if (pool < 0) { | |
5399 | ss << "pool " << poolstr << " does not exist"; | |
5400 | r = -ENOENT; | |
5401 | goto reply; | |
5402 | } | |
5403 | object_locator_t oloc(pool, namespacestr); | |
5404 | object_t oid(objstr); | |
5405 | pg_t pgid = osdmap.object_locator_to_pg(oid, oloc); | |
5406 | pg_t mpgid = osdmap.raw_pg_to_pg(pgid); | |
5407 | vector<int> up, acting; | |
5408 | int up_p, acting_p; | |
5409 | osdmap.pg_to_up_acting_osds(mpgid, &up, &up_p, &acting, &acting_p); | |
5410 | ||
5411 | string fullobjname; | |
5412 | if (!namespacestr.empty()) | |
5413 | fullobjname = namespacestr + string("/") + oid.name; | |
5414 | else | |
5415 | fullobjname = oid.name; | |
5416 | if (f) { | |
5417 | f->open_object_section("osd_map"); | |
5418 | f->dump_unsigned("epoch", osdmap.get_epoch()); | |
5419 | f->dump_string("pool", poolstr); | |
5420 | f->dump_int("pool_id", pool); | |
5421 | f->dump_stream("objname") << fullobjname; | |
5422 | f->dump_stream("raw_pgid") << pgid; | |
5423 | f->dump_stream("pgid") << mpgid; | |
5424 | f->open_array_section("up"); | |
5425 | for (vector<int>::iterator p = up.begin(); p != up.end(); ++p) | |
5426 | f->dump_int("osd", *p); | |
5427 | f->close_section(); | |
5428 | f->dump_int("up_primary", up_p); | |
5429 | f->open_array_section("acting"); | |
5430 | for (vector<int>::iterator p = acting.begin(); p != acting.end(); ++p) | |
5431 | f->dump_int("osd", *p); | |
5432 | f->close_section(); | |
5433 | f->dump_int("acting_primary", acting_p); | |
5434 | f->close_section(); // osd_map | |
5435 | f->flush(rdata); | |
5436 | } else { | |
5437 | ds << "osdmap e" << osdmap.get_epoch() | |
5438 | << " pool '" << poolstr << "' (" << pool << ")" | |
5439 | << " object '" << fullobjname << "' ->" | |
5440 | << " pg " << pgid << " (" << mpgid << ")" | |
5441 | << " -> up (" << pg_vector_string(up) << ", p" << up_p << ") acting (" | |
5442 | << pg_vector_string(acting) << ", p" << acting_p << ")"; | |
5443 | rdata.append(ds); | |
5444 | } | |
5445 | ||
5446 | } else if (prefix == "pg map") { | |
5447 | pg_t pgid; | |
5448 | string pgidstr; | |
11fdf7f2 | 5449 | cmd_getval(cct, cmdmap, "pgid", pgidstr); |
7c673cae FG |
5450 | if (!pgid.parse(pgidstr.c_str())) { |
5451 | ss << "invalid pgid '" << pgidstr << "'"; | |
5452 | r = -EINVAL; | |
5453 | goto reply; | |
5454 | } | |
5455 | vector<int> up, acting; | |
5456 | if (!osdmap.have_pg_pool(pgid.pool())) { | |
5457 | ss << "pg '" << pgidstr << "' does not exist"; | |
5458 | r = -ENOENT; | |
5459 | goto reply; | |
5460 | } | |
5461 | pg_t mpgid = osdmap.raw_pg_to_pg(pgid); | |
5462 | osdmap.pg_to_up_acting_osds(pgid, up, acting); | |
5463 | if (f) { | |
5464 | f->open_object_section("pg_map"); | |
5465 | f->dump_unsigned("epoch", osdmap.get_epoch()); | |
5466 | f->dump_stream("raw_pgid") << pgid; | |
5467 | f->dump_stream("pgid") << mpgid; | |
5468 | f->open_array_section("up"); | |
5469 | for (auto osd : up) { | |
5470 | f->dump_int("up_osd", osd); | |
5471 | } | |
5472 | f->close_section(); | |
5473 | f->open_array_section("acting"); | |
5474 | for (auto osd : acting) { | |
5475 | f->dump_int("acting_osd", osd); | |
5476 | } | |
5477 | f->close_section(); | |
5478 | f->close_section(); | |
5479 | f->flush(rdata); | |
5480 | } else { | |
5481 | ds << "osdmap e" << osdmap.get_epoch() | |
5482 | << " pg " << pgid << " (" << mpgid << ")" | |
5483 | << " -> up " << up << " acting " << acting; | |
5484 | rdata.append(ds); | |
5485 | } | |
5486 | goto reply; | |
5487 | ||
7c673cae | 5488 | } else if (prefix == "osd lspools") { |
7c673cae FG |
5489 | if (f) |
5490 | f->open_array_section("pools"); | |
5491 | for (map<int64_t, pg_pool_t>::iterator p = osdmap.pools.begin(); | |
5492 | p != osdmap.pools.end(); | |
5493 | ++p) { | |
11fdf7f2 TL |
5494 | if (f) { |
5495 | f->open_object_section("pool"); | |
5496 | f->dump_int("poolnum", p->first); | |
5497 | f->dump_string("poolname", osdmap.pool_name[p->first]); | |
5498 | f->close_section(); | |
5499 | } else { | |
5500 | ds << p->first << ' ' << osdmap.pool_name[p->first]; | |
5501 | if (next(p) != osdmap.pools.end()) { | |
5502 | ds << '\n'; | |
7c673cae FG |
5503 | } |
5504 | } | |
5505 | } | |
5506 | if (f) { | |
5507 | f->close_section(); | |
5508 | f->flush(ds); | |
5509 | } | |
5510 | rdata.append(ds); | |
5511 | } else if (prefix == "osd blacklist ls") { | |
5512 | if (f) | |
5513 | f->open_array_section("blacklist"); | |
5514 | ||
5515 | for (ceph::unordered_map<entity_addr_t,utime_t>::iterator p = osdmap.blacklist.begin(); | |
5516 | p != osdmap.blacklist.end(); | |
5517 | ++p) { | |
5518 | if (f) { | |
5519 | f->open_object_section("entry"); | |
11fdf7f2 | 5520 | f->dump_string("addr", p->first.get_legacy_str()); |
7c673cae FG |
5521 | f->dump_stream("until") << p->second; |
5522 | f->close_section(); | |
5523 | } else { | |
5524 | stringstream ss; | |
5525 | string s; | |
5526 | ss << p->first << " " << p->second; | |
5527 | getline(ss, s); | |
5528 | s += "\n"; | |
5529 | rdata.append(s); | |
5530 | } | |
5531 | } | |
5532 | if (f) { | |
5533 | f->close_section(); | |
5534 | f->flush(rdata); | |
5535 | } | |
5536 | ss << "listed " << osdmap.blacklist.size() << " entries"; | |
5537 | ||
5538 | } else if (prefix == "osd pool ls") { | |
5539 | string detail; | |
11fdf7f2 | 5540 | cmd_getval(cct, cmdmap, "detail", detail); |
7c673cae FG |
5541 | if (!f && detail == "detail") { |
5542 | ostringstream ss; | |
5543 | osdmap.print_pools(ss); | |
5544 | rdata.append(ss.str()); | |
5545 | } else { | |
5546 | if (f) | |
5547 | f->open_array_section("pools"); | |
5548 | for (map<int64_t,pg_pool_t>::const_iterator it = osdmap.get_pools().begin(); | |
5549 | it != osdmap.get_pools().end(); | |
5550 | ++it) { | |
5551 | if (f) { | |
5552 | if (detail == "detail") { | |
5553 | f->open_object_section("pool"); | |
eafe8130 | 5554 | f->dump_int("pool_id", it->first); |
7c673cae FG |
5555 | f->dump_string("pool_name", osdmap.get_pool_name(it->first)); |
5556 | it->second.dump(f.get()); | |
5557 | f->close_section(); | |
5558 | } else { | |
5559 | f->dump_string("pool_name", osdmap.get_pool_name(it->first)); | |
5560 | } | |
5561 | } else { | |
5562 | rdata.append(osdmap.get_pool_name(it->first) + "\n"); | |
5563 | } | |
5564 | } | |
5565 | if (f) { | |
5566 | f->close_section(); | |
5567 | f->flush(rdata); | |
5568 | } | |
5569 | } | |
5570 | ||
5571 | } else if (prefix == "osd crush get-tunable") { | |
5572 | string tunable; | |
11fdf7f2 | 5573 | cmd_getval(cct, cmdmap, "tunable", tunable); |
7c673cae FG |
5574 | ostringstream rss; |
5575 | if (f) | |
5576 | f->open_object_section("tunable"); | |
5577 | if (tunable == "straw_calc_version") { | |
5578 | if (f) | |
5579 | f->dump_int(tunable.c_str(), osdmap.crush->get_straw_calc_version()); | |
5580 | else | |
5581 | rss << osdmap.crush->get_straw_calc_version() << "\n"; | |
5582 | } else { | |
5583 | r = -EINVAL; | |
5584 | goto reply; | |
5585 | } | |
5586 | if (f) { | |
5587 | f->close_section(); | |
5588 | f->flush(rdata); | |
5589 | } else { | |
5590 | rdata.append(rss.str()); | |
5591 | } | |
5592 | r = 0; | |
5593 | ||
5594 | } else if (prefix == "osd pool get") { | |
5595 | string poolstr; | |
11fdf7f2 | 5596 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
5597 | int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str()); |
5598 | if (pool < 0) { | |
5599 | ss << "unrecognized pool '" << poolstr << "'"; | |
5600 | r = -ENOENT; | |
5601 | goto reply; | |
5602 | } | |
5603 | ||
5604 | const pg_pool_t *p = osdmap.get_pg_pool(pool); | |
5605 | string var; | |
11fdf7f2 | 5606 | cmd_getval(cct, cmdmap, "var", var); |
7c673cae FG |
5607 | |
5608 | typedef std::map<std::string, osd_pool_get_choices> choices_map_t; | |
5609 | const choices_map_t ALL_CHOICES = { | |
5610 | {"size", SIZE}, | |
5611 | {"min_size", MIN_SIZE}, | |
7c673cae | 5612 | {"pg_num", PG_NUM}, {"pgp_num", PGP_NUM}, |
28e407b8 AA |
5613 | {"crush_rule", CRUSH_RULE}, {"hashpspool", HASHPSPOOL}, |
5614 | {"allow_ec_overwrites", EC_OVERWRITES}, {"nodelete", NODELETE}, | |
7c673cae FG |
5615 | {"nopgchange", NOPGCHANGE}, {"nosizechange", NOSIZECHANGE}, |
5616 | {"noscrub", NOSCRUB}, {"nodeep-scrub", NODEEP_SCRUB}, | |
5617 | {"write_fadvise_dontneed", WRITE_FADVISE_DONTNEED}, | |
5618 | {"hit_set_type", HIT_SET_TYPE}, {"hit_set_period", HIT_SET_PERIOD}, | |
5619 | {"hit_set_count", HIT_SET_COUNT}, {"hit_set_fpp", HIT_SET_FPP}, | |
5620 | {"use_gmt_hitset", USE_GMT_HITSET}, | |
11fdf7f2 | 5621 | {"target_max_objects", TARGET_MAX_OBJECTS}, |
7c673cae FG |
5622 | {"target_max_bytes", TARGET_MAX_BYTES}, |
5623 | {"cache_target_dirty_ratio", CACHE_TARGET_DIRTY_RATIO}, | |
5624 | {"cache_target_dirty_high_ratio", CACHE_TARGET_DIRTY_HIGH_RATIO}, | |
5625 | {"cache_target_full_ratio", CACHE_TARGET_FULL_RATIO}, | |
5626 | {"cache_min_flush_age", CACHE_MIN_FLUSH_AGE}, | |
5627 | {"cache_min_evict_age", CACHE_MIN_EVICT_AGE}, | |
5628 | {"erasure_code_profile", ERASURE_CODE_PROFILE}, | |
5629 | {"min_read_recency_for_promote", MIN_READ_RECENCY_FOR_PROMOTE}, | |
5630 | {"min_write_recency_for_promote", MIN_WRITE_RECENCY_FOR_PROMOTE}, | |
5631 | {"fast_read", FAST_READ}, | |
5632 | {"hit_set_grade_decay_rate", HIT_SET_GRADE_DECAY_RATE}, | |
5633 | {"hit_set_search_last_n", HIT_SET_SEARCH_LAST_N}, | |
5634 | {"scrub_min_interval", SCRUB_MIN_INTERVAL}, | |
5635 | {"scrub_max_interval", SCRUB_MAX_INTERVAL}, | |
5636 | {"deep_scrub_interval", DEEP_SCRUB_INTERVAL}, | |
5637 | {"recovery_priority", RECOVERY_PRIORITY}, | |
5638 | {"recovery_op_priority", RECOVERY_OP_PRIORITY}, | |
5639 | {"scrub_priority", SCRUB_PRIORITY}, | |
5640 | {"compression_mode", COMPRESSION_MODE}, | |
5641 | {"compression_algorithm", COMPRESSION_ALGORITHM}, | |
5642 | {"compression_required_ratio", COMPRESSION_REQUIRED_RATIO}, | |
5643 | {"compression_max_blob_size", COMPRESSION_MAX_BLOB_SIZE}, | |
5644 | {"compression_min_blob_size", COMPRESSION_MIN_BLOB_SIZE}, | |
5645 | {"csum_type", CSUM_TYPE}, | |
5646 | {"csum_max_block", CSUM_MAX_BLOCK}, | |
5647 | {"csum_min_block", CSUM_MIN_BLOCK}, | |
11fdf7f2 TL |
5648 | {"fingerprint_algorithm", FINGERPRINT_ALGORITHM}, |
5649 | {"pg_autoscale_mode", PG_AUTOSCALE_MODE}, | |
5650 | {"pg_num_min", PG_NUM_MIN}, | |
5651 | {"target_size_bytes", TARGET_SIZE_BYTES}, | |
5652 | {"target_size_ratio", TARGET_SIZE_RATIO}, | |
5653 | {"pg_autoscale_bias", PG_AUTOSCALE_BIAS}, | |
7c673cae FG |
5654 | }; |
5655 | ||
5656 | typedef std::set<osd_pool_get_choices> choices_set_t; | |
5657 | ||
5658 | const choices_set_t ONLY_TIER_CHOICES = { | |
5659 | HIT_SET_TYPE, HIT_SET_PERIOD, HIT_SET_COUNT, HIT_SET_FPP, | |
5660 | TARGET_MAX_OBJECTS, TARGET_MAX_BYTES, CACHE_TARGET_FULL_RATIO, | |
5661 | CACHE_TARGET_DIRTY_RATIO, CACHE_TARGET_DIRTY_HIGH_RATIO, | |
5662 | CACHE_MIN_FLUSH_AGE, CACHE_MIN_EVICT_AGE, | |
5663 | MIN_READ_RECENCY_FOR_PROMOTE, | |
c07f9fc5 | 5664 | MIN_WRITE_RECENCY_FOR_PROMOTE, |
7c673cae FG |
5665 | HIT_SET_GRADE_DECAY_RATE, HIT_SET_SEARCH_LAST_N |
5666 | }; | |
5667 | const choices_set_t ONLY_ERASURE_CHOICES = { | |
28e407b8 | 5668 | EC_OVERWRITES, ERASURE_CODE_PROFILE |
7c673cae FG |
5669 | }; |
5670 | ||
5671 | choices_set_t selected_choices; | |
5672 | if (var == "all") { | |
5673 | for(choices_map_t::const_iterator it = ALL_CHOICES.begin(); | |
5674 | it != ALL_CHOICES.end(); ++it) { | |
5675 | selected_choices.insert(it->second); | |
5676 | } | |
5677 | ||
5678 | if(!p->is_tier()) { | |
5679 | selected_choices = subtract_second_from_first(selected_choices, | |
5680 | ONLY_TIER_CHOICES); | |
5681 | } | |
5682 | ||
5683 | if(!p->is_erasure()) { | |
5684 | selected_choices = subtract_second_from_first(selected_choices, | |
5685 | ONLY_ERASURE_CHOICES); | |
5686 | } | |
5687 | } else /* var != "all" */ { | |
5688 | choices_map_t::const_iterator found = ALL_CHOICES.find(var); | |
5689 | osd_pool_get_choices selected = found->second; | |
5690 | ||
5691 | if (!p->is_tier() && | |
5692 | ONLY_TIER_CHOICES.find(selected) != ONLY_TIER_CHOICES.end()) { | |
5693 | ss << "pool '" << poolstr | |
5694 | << "' is not a tier pool: variable not applicable"; | |
5695 | r = -EACCES; | |
5696 | goto reply; | |
5697 | } | |
5698 | ||
5699 | if (!p->is_erasure() && | |
5700 | ONLY_ERASURE_CHOICES.find(selected) | |
5701 | != ONLY_ERASURE_CHOICES.end()) { | |
5702 | ss << "pool '" << poolstr | |
5703 | << "' is not a erasure pool: variable not applicable"; | |
5704 | r = -EACCES; | |
5705 | goto reply; | |
5706 | } | |
5707 | ||
94b18763 FG |
5708 | if (pool_opts_t::is_opt_name(var) && |
5709 | !p->opts.is_set(pool_opts_t::get_opt_desc(var).key)) { | |
5710 | ss << "option '" << var << "' is not set on pool '" << poolstr << "'"; | |
5711 | r = -ENOENT; | |
5712 | goto reply; | |
5713 | } | |
5714 | ||
7c673cae FG |
5715 | selected_choices.insert(selected); |
5716 | } | |
5717 | ||
5718 | if (f) { | |
94b18763 FG |
5719 | f->open_object_section("pool"); |
5720 | f->dump_string("pool", poolstr); | |
5721 | f->dump_int("pool_id", pool); | |
7c673cae FG |
5722 | for(choices_set_t::const_iterator it = selected_choices.begin(); |
5723 | it != selected_choices.end(); ++it) { | |
5724 | choices_map_t::const_iterator i; | |
c07f9fc5 FG |
5725 | for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) { |
5726 | if (i->second == *it) { | |
5727 | break; | |
5728 | } | |
5729 | } | |
11fdf7f2 | 5730 | ceph_assert(i != ALL_CHOICES.end()); |
7c673cae FG |
5731 | switch(*it) { |
5732 | case PG_NUM: | |
5733 | f->dump_int("pg_num", p->get_pg_num()); | |
5734 | break; | |
5735 | case PGP_NUM: | |
5736 | f->dump_int("pgp_num", p->get_pgp_num()); | |
5737 | break; | |
7c673cae FG |
5738 | case SIZE: |
5739 | f->dump_int("size", p->get_size()); | |
5740 | break; | |
5741 | case MIN_SIZE: | |
5742 | f->dump_int("min_size", p->get_min_size()); | |
5743 | break; | |
7c673cae | 5744 | case CRUSH_RULE: |
31f18b77 | 5745 | if (osdmap.crush->rule_exists(p->get_crush_rule())) { |
7c673cae | 5746 | f->dump_string("crush_rule", osdmap.crush->get_rule_name( |
31f18b77 | 5747 | p->get_crush_rule())); |
7c673cae | 5748 | } else { |
31f18b77 | 5749 | f->dump_string("crush_rule", stringify(p->get_crush_rule())); |
7c673cae FG |
5750 | } |
5751 | break; | |
28e407b8 AA |
5752 | case EC_OVERWRITES: |
5753 | f->dump_bool("allow_ec_overwrites", | |
5754 | p->has_flag(pg_pool_t::FLAG_EC_OVERWRITES)); | |
5755 | break; | |
11fdf7f2 TL |
5756 | case PG_AUTOSCALE_MODE: |
5757 | f->dump_string("pg_autoscale_mode", | |
5758 | pg_pool_t::get_pg_autoscale_mode_name( | |
5759 | p->pg_autoscale_mode)); | |
5760 | break; | |
7c673cae FG |
5761 | case HASHPSPOOL: |
5762 | case NODELETE: | |
5763 | case NOPGCHANGE: | |
5764 | case NOSIZECHANGE: | |
5765 | case WRITE_FADVISE_DONTNEED: | |
5766 | case NOSCRUB: | |
5767 | case NODEEP_SCRUB: | |
94b18763 FG |
5768 | f->dump_bool(i->first.c_str(), |
5769 | p->has_flag(pg_pool_t::get_flag_by_name(i->first))); | |
7c673cae FG |
5770 | break; |
5771 | case HIT_SET_PERIOD: | |
5772 | f->dump_int("hit_set_period", p->hit_set_period); | |
5773 | break; | |
5774 | case HIT_SET_COUNT: | |
5775 | f->dump_int("hit_set_count", p->hit_set_count); | |
5776 | break; | |
5777 | case HIT_SET_TYPE: | |
5778 | f->dump_string("hit_set_type", | |
5779 | HitSet::get_type_name(p->hit_set_params.get_type())); | |
5780 | break; | |
5781 | case HIT_SET_FPP: | |
5782 | { | |
5783 | if (p->hit_set_params.get_type() == HitSet::TYPE_BLOOM) { | |
5784 | BloomHitSet::Params *bloomp = | |
5785 | static_cast<BloomHitSet::Params*>(p->hit_set_params.impl.get()); | |
5786 | f->dump_float("hit_set_fpp", bloomp->get_fpp()); | |
5787 | } else if(var != "all") { | |
5788 | f->close_section(); | |
5789 | ss << "hit set is not of type Bloom; " << | |
5790 | "invalid to get a false positive rate!"; | |
5791 | r = -EINVAL; | |
5792 | goto reply; | |
5793 | } | |
5794 | } | |
5795 | break; | |
5796 | case USE_GMT_HITSET: | |
5797 | f->dump_bool("use_gmt_hitset", p->use_gmt_hitset); | |
5798 | break; | |
5799 | case TARGET_MAX_OBJECTS: | |
5800 | f->dump_unsigned("target_max_objects", p->target_max_objects); | |
5801 | break; | |
5802 | case TARGET_MAX_BYTES: | |
5803 | f->dump_unsigned("target_max_bytes", p->target_max_bytes); | |
5804 | break; | |
5805 | case CACHE_TARGET_DIRTY_RATIO: | |
5806 | f->dump_unsigned("cache_target_dirty_ratio_micro", | |
5807 | p->cache_target_dirty_ratio_micro); | |
5808 | f->dump_float("cache_target_dirty_ratio", | |
5809 | ((float)p->cache_target_dirty_ratio_micro/1000000)); | |
5810 | break; | |
5811 | case CACHE_TARGET_DIRTY_HIGH_RATIO: | |
5812 | f->dump_unsigned("cache_target_dirty_high_ratio_micro", | |
5813 | p->cache_target_dirty_high_ratio_micro); | |
5814 | f->dump_float("cache_target_dirty_high_ratio", | |
5815 | ((float)p->cache_target_dirty_high_ratio_micro/1000000)); | |
5816 | break; | |
5817 | case CACHE_TARGET_FULL_RATIO: | |
5818 | f->dump_unsigned("cache_target_full_ratio_micro", | |
5819 | p->cache_target_full_ratio_micro); | |
5820 | f->dump_float("cache_target_full_ratio", | |
5821 | ((float)p->cache_target_full_ratio_micro/1000000)); | |
5822 | break; | |
5823 | case CACHE_MIN_FLUSH_AGE: | |
5824 | f->dump_unsigned("cache_min_flush_age", p->cache_min_flush_age); | |
5825 | break; | |
5826 | case CACHE_MIN_EVICT_AGE: | |
5827 | f->dump_unsigned("cache_min_evict_age", p->cache_min_evict_age); | |
5828 | break; | |
5829 | case ERASURE_CODE_PROFILE: | |
5830 | f->dump_string("erasure_code_profile", p->erasure_code_profile); | |
5831 | break; | |
5832 | case MIN_READ_RECENCY_FOR_PROMOTE: | |
5833 | f->dump_int("min_read_recency_for_promote", | |
5834 | p->min_read_recency_for_promote); | |
5835 | break; | |
5836 | case MIN_WRITE_RECENCY_FOR_PROMOTE: | |
5837 | f->dump_int("min_write_recency_for_promote", | |
5838 | p->min_write_recency_for_promote); | |
5839 | break; | |
5840 | case FAST_READ: | |
5841 | f->dump_int("fast_read", p->fast_read); | |
5842 | break; | |
5843 | case HIT_SET_GRADE_DECAY_RATE: | |
5844 | f->dump_int("hit_set_grade_decay_rate", | |
5845 | p->hit_set_grade_decay_rate); | |
5846 | break; | |
5847 | case HIT_SET_SEARCH_LAST_N: | |
5848 | f->dump_int("hit_set_search_last_n", | |
5849 | p->hit_set_search_last_n); | |
5850 | break; | |
5851 | case SCRUB_MIN_INTERVAL: | |
5852 | case SCRUB_MAX_INTERVAL: | |
5853 | case DEEP_SCRUB_INTERVAL: | |
5854 | case RECOVERY_PRIORITY: | |
5855 | case RECOVERY_OP_PRIORITY: | |
5856 | case SCRUB_PRIORITY: | |
5857 | case COMPRESSION_MODE: | |
5858 | case COMPRESSION_ALGORITHM: | |
5859 | case COMPRESSION_REQUIRED_RATIO: | |
5860 | case COMPRESSION_MAX_BLOB_SIZE: | |
5861 | case COMPRESSION_MIN_BLOB_SIZE: | |
5862 | case CSUM_TYPE: | |
5863 | case CSUM_MAX_BLOCK: | |
5864 | case CSUM_MIN_BLOCK: | |
11fdf7f2 TL |
5865 | case FINGERPRINT_ALGORITHM: |
5866 | case PG_NUM_MIN: | |
5867 | case TARGET_SIZE_BYTES: | |
5868 | case TARGET_SIZE_RATIO: | |
5869 | case PG_AUTOSCALE_BIAS: | |
c07f9fc5 FG |
5870 | pool_opts_t::key_t key = pool_opts_t::get_opt_desc(i->first).key; |
5871 | if (p->opts.is_set(key)) { | |
c07f9fc5 | 5872 | if(*it == CSUM_TYPE) { |
11fdf7f2 | 5873 | int64_t val; |
c07f9fc5 FG |
5874 | p->opts.get(pool_opts_t::CSUM_TYPE, &val); |
5875 | f->dump_string(i->first.c_str(), Checksummer::get_csum_type_string(val)); | |
5876 | } else { | |
5877 | p->opts.dump(i->first, f.get()); | |
5878 | } | |
94b18763 | 5879 | } |
7c673cae FG |
5880 | break; |
5881 | } | |
7c673cae | 5882 | } |
94b18763 FG |
5883 | f->close_section(); |
5884 | f->flush(rdata); | |
7c673cae FG |
5885 | } else /* !f */ { |
5886 | for(choices_set_t::const_iterator it = selected_choices.begin(); | |
5887 | it != selected_choices.end(); ++it) { | |
5888 | choices_map_t::const_iterator i; | |
5889 | switch(*it) { | |
5890 | case PG_NUM: | |
5891 | ss << "pg_num: " << p->get_pg_num() << "\n"; | |
5892 | break; | |
5893 | case PGP_NUM: | |
5894 | ss << "pgp_num: " << p->get_pgp_num() << "\n"; | |
5895 | break; | |
7c673cae FG |
5896 | case SIZE: |
5897 | ss << "size: " << p->get_size() << "\n"; | |
5898 | break; | |
5899 | case MIN_SIZE: | |
5900 | ss << "min_size: " << p->get_min_size() << "\n"; | |
5901 | break; | |
7c673cae | 5902 | case CRUSH_RULE: |
31f18b77 | 5903 | if (osdmap.crush->rule_exists(p->get_crush_rule())) { |
7c673cae | 5904 | ss << "crush_rule: " << osdmap.crush->get_rule_name( |
31f18b77 | 5905 | p->get_crush_rule()) << "\n"; |
7c673cae | 5906 | } else { |
31f18b77 | 5907 | ss << "crush_rule: " << p->get_crush_rule() << "\n"; |
7c673cae FG |
5908 | } |
5909 | break; | |
11fdf7f2 TL |
5910 | case PG_AUTOSCALE_MODE: |
5911 | ss << "pg_autoscale_mode: " << pg_pool_t::get_pg_autoscale_mode_name( | |
5912 | p->pg_autoscale_mode) <<"\n"; | |
5913 | break; | |
7c673cae FG |
5914 | case HIT_SET_PERIOD: |
5915 | ss << "hit_set_period: " << p->hit_set_period << "\n"; | |
5916 | break; | |
5917 | case HIT_SET_COUNT: | |
5918 | ss << "hit_set_count: " << p->hit_set_count << "\n"; | |
5919 | break; | |
5920 | case HIT_SET_TYPE: | |
5921 | ss << "hit_set_type: " << | |
5922 | HitSet::get_type_name(p->hit_set_params.get_type()) << "\n"; | |
5923 | break; | |
5924 | case HIT_SET_FPP: | |
5925 | { | |
5926 | if (p->hit_set_params.get_type() == HitSet::TYPE_BLOOM) { | |
5927 | BloomHitSet::Params *bloomp = | |
5928 | static_cast<BloomHitSet::Params*>(p->hit_set_params.impl.get()); | |
5929 | ss << "hit_set_fpp: " << bloomp->get_fpp() << "\n"; | |
5930 | } else if(var != "all") { | |
5931 | ss << "hit set is not of type Bloom; " << | |
5932 | "invalid to get a false positive rate!"; | |
5933 | r = -EINVAL; | |
5934 | goto reply; | |
5935 | } | |
5936 | } | |
5937 | break; | |
5938 | case USE_GMT_HITSET: | |
5939 | ss << "use_gmt_hitset: " << p->use_gmt_hitset << "\n"; | |
5940 | break; | |
5941 | case TARGET_MAX_OBJECTS: | |
5942 | ss << "target_max_objects: " << p->target_max_objects << "\n"; | |
5943 | break; | |
5944 | case TARGET_MAX_BYTES: | |
5945 | ss << "target_max_bytes: " << p->target_max_bytes << "\n"; | |
5946 | break; | |
5947 | case CACHE_TARGET_DIRTY_RATIO: | |
5948 | ss << "cache_target_dirty_ratio: " | |
5949 | << ((float)p->cache_target_dirty_ratio_micro/1000000) << "\n"; | |
5950 | break; | |
5951 | case CACHE_TARGET_DIRTY_HIGH_RATIO: | |
5952 | ss << "cache_target_dirty_high_ratio: " | |
5953 | << ((float)p->cache_target_dirty_high_ratio_micro/1000000) << "\n"; | |
5954 | break; | |
5955 | case CACHE_TARGET_FULL_RATIO: | |
5956 | ss << "cache_target_full_ratio: " | |
5957 | << ((float)p->cache_target_full_ratio_micro/1000000) << "\n"; | |
5958 | break; | |
5959 | case CACHE_MIN_FLUSH_AGE: | |
5960 | ss << "cache_min_flush_age: " << p->cache_min_flush_age << "\n"; | |
5961 | break; | |
5962 | case CACHE_MIN_EVICT_AGE: | |
5963 | ss << "cache_min_evict_age: " << p->cache_min_evict_age << "\n"; | |
5964 | break; | |
5965 | case ERASURE_CODE_PROFILE: | |
5966 | ss << "erasure_code_profile: " << p->erasure_code_profile << "\n"; | |
5967 | break; | |
5968 | case MIN_READ_RECENCY_FOR_PROMOTE: | |
5969 | ss << "min_read_recency_for_promote: " << | |
5970 | p->min_read_recency_for_promote << "\n"; | |
5971 | break; | |
5972 | case HIT_SET_GRADE_DECAY_RATE: | |
5973 | ss << "hit_set_grade_decay_rate: " << | |
5974 | p->hit_set_grade_decay_rate << "\n"; | |
5975 | break; | |
5976 | case HIT_SET_SEARCH_LAST_N: | |
5977 | ss << "hit_set_search_last_n: " << | |
5978 | p->hit_set_search_last_n << "\n"; | |
5979 | break; | |
28e407b8 AA |
5980 | case EC_OVERWRITES: |
5981 | ss << "allow_ec_overwrites: " << | |
5982 | (p->has_flag(pg_pool_t::FLAG_EC_OVERWRITES) ? "true" : "false") << | |
5983 | "\n"; | |
5984 | break; | |
7c673cae FG |
5985 | case HASHPSPOOL: |
5986 | case NODELETE: | |
5987 | case NOPGCHANGE: | |
5988 | case NOSIZECHANGE: | |
5989 | case WRITE_FADVISE_DONTNEED: | |
5990 | case NOSCRUB: | |
5991 | case NODEEP_SCRUB: | |
5992 | for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) { | |
5993 | if (i->second == *it) | |
5994 | break; | |
5995 | } | |
11fdf7f2 | 5996 | ceph_assert(i != ALL_CHOICES.end()); |
7c673cae FG |
5997 | ss << i->first << ": " << |
5998 | (p->has_flag(pg_pool_t::get_flag_by_name(i->first)) ? | |
5999 | "true" : "false") << "\n"; | |
6000 | break; | |
6001 | case MIN_WRITE_RECENCY_FOR_PROMOTE: | |
6002 | ss << "min_write_recency_for_promote: " << | |
6003 | p->min_write_recency_for_promote << "\n"; | |
6004 | break; | |
6005 | case FAST_READ: | |
6006 | ss << "fast_read: " << p->fast_read << "\n"; | |
6007 | break; | |
6008 | case SCRUB_MIN_INTERVAL: | |
6009 | case SCRUB_MAX_INTERVAL: | |
6010 | case DEEP_SCRUB_INTERVAL: | |
6011 | case RECOVERY_PRIORITY: | |
6012 | case RECOVERY_OP_PRIORITY: | |
6013 | case SCRUB_PRIORITY: | |
6014 | case COMPRESSION_MODE: | |
6015 | case COMPRESSION_ALGORITHM: | |
6016 | case COMPRESSION_REQUIRED_RATIO: | |
6017 | case COMPRESSION_MAX_BLOB_SIZE: | |
6018 | case COMPRESSION_MIN_BLOB_SIZE: | |
6019 | case CSUM_TYPE: | |
6020 | case CSUM_MAX_BLOCK: | |
6021 | case CSUM_MIN_BLOCK: | |
11fdf7f2 TL |
6022 | case FINGERPRINT_ALGORITHM: |
6023 | case PG_NUM_MIN: | |
6024 | case TARGET_SIZE_BYTES: | |
6025 | case TARGET_SIZE_RATIO: | |
6026 | case PG_AUTOSCALE_BIAS: | |
7c673cae FG |
6027 | for (i = ALL_CHOICES.begin(); i != ALL_CHOICES.end(); ++i) { |
6028 | if (i->second == *it) | |
6029 | break; | |
6030 | } | |
11fdf7f2 | 6031 | ceph_assert(i != ALL_CHOICES.end()); |
7c673cae FG |
6032 | { |
6033 | pool_opts_t::key_t key = pool_opts_t::get_opt_desc(i->first).key; | |
6034 | if (p->opts.is_set(key)) { | |
6035 | if(key == pool_opts_t::CSUM_TYPE) { | |
11fdf7f2 | 6036 | int64_t val; |
7c673cae FG |
6037 | p->opts.get(key, &val); |
6038 | ss << i->first << ": " << Checksummer::get_csum_type_string(val) << "\n"; | |
6039 | } else { | |
6040 | ss << i->first << ": " << p->opts.get(key) << "\n"; | |
6041 | } | |
6042 | } | |
6043 | } | |
6044 | break; | |
6045 | } | |
6046 | rdata.append(ss.str()); | |
6047 | ss.str(""); | |
6048 | } | |
6049 | } | |
6050 | r = 0; | |
7c673cae FG |
6051 | } else if (prefix == "osd pool get-quota") { |
6052 | string pool_name; | |
11fdf7f2 | 6053 | cmd_getval(cct, cmdmap, "pool", pool_name); |
7c673cae FG |
6054 | |
6055 | int64_t poolid = osdmap.lookup_pg_pool_name(pool_name); | |
6056 | if (poolid < 0) { | |
11fdf7f2 | 6057 | ceph_assert(poolid == -ENOENT); |
7c673cae FG |
6058 | ss << "unrecognized pool '" << pool_name << "'"; |
6059 | r = -ENOENT; | |
6060 | goto reply; | |
6061 | } | |
6062 | const pg_pool_t *p = osdmap.get_pg_pool(poolid); | |
6063 | ||
6064 | if (f) { | |
6065 | f->open_object_section("pool_quotas"); | |
6066 | f->dump_string("pool_name", pool_name); | |
6067 | f->dump_unsigned("pool_id", poolid); | |
6068 | f->dump_unsigned("quota_max_objects", p->quota_max_objects); | |
6069 | f->dump_unsigned("quota_max_bytes", p->quota_max_bytes); | |
6070 | f->close_section(); | |
6071 | f->flush(rdata); | |
6072 | } else { | |
6073 | stringstream rs; | |
6074 | rs << "quotas for pool '" << pool_name << "':\n" | |
6075 | << " max objects: "; | |
6076 | if (p->quota_max_objects == 0) | |
6077 | rs << "N/A"; | |
6078 | else | |
1adf2230 | 6079 | rs << si_u_t(p->quota_max_objects) << " objects"; |
7c673cae FG |
6080 | rs << "\n" |
6081 | << " max bytes : "; | |
6082 | if (p->quota_max_bytes == 0) | |
6083 | rs << "N/A"; | |
6084 | else | |
1adf2230 | 6085 | rs << byte_u_t(p->quota_max_bytes); |
7c673cae FG |
6086 | rdata.append(rs.str()); |
6087 | } | |
6088 | rdata.append("\n"); | |
6089 | r = 0; | |
6090 | } else if (prefix == "osd crush rule list" || | |
6091 | prefix == "osd crush rule ls") { | |
c07f9fc5 FG |
6092 | if (f) { |
6093 | f->open_array_section("rules"); | |
6094 | osdmap.crush->list_rules(f.get()); | |
6095 | f->close_section(); | |
6096 | f->flush(rdata); | |
6097 | } else { | |
6098 | ostringstream ss; | |
6099 | osdmap.crush->list_rules(&ss); | |
6100 | rdata.append(ss.str()); | |
6101 | } | |
b5b8bbf5 FG |
6102 | } else if (prefix == "osd crush rule ls-by-class") { |
6103 | string class_name; | |
11fdf7f2 | 6104 | cmd_getval(cct, cmdmap, "class", class_name); |
b5b8bbf5 FG |
6105 | if (class_name.empty()) { |
6106 | ss << "no class specified"; | |
6107 | r = -EINVAL; | |
6108 | goto reply; | |
6109 | } | |
6110 | set<int> rules; | |
6111 | r = osdmap.crush->get_rules_by_class(class_name, &rules); | |
6112 | if (r < 0) { | |
6113 | ss << "failed to get rules by class '" << class_name << "'"; | |
6114 | goto reply; | |
6115 | } | |
6116 | if (f) { | |
6117 | f->open_array_section("rules"); | |
6118 | for (auto &rule: rules) { | |
6119 | f->dump_string("name", osdmap.crush->get_rule_name(rule)); | |
6120 | } | |
6121 | f->close_section(); | |
6122 | f->flush(rdata); | |
6123 | } else { | |
6124 | ostringstream rs; | |
6125 | for (auto &rule: rules) { | |
6126 | rs << osdmap.crush->get_rule_name(rule) << "\n"; | |
6127 | } | |
6128 | rdata.append(rs.str()); | |
6129 | } | |
7c673cae FG |
6130 | } else if (prefix == "osd crush rule dump") { |
6131 | string name; | |
11fdf7f2 | 6132 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae | 6133 | string format; |
11fdf7f2 | 6134 | cmd_getval(cct, cmdmap, "format", format); |
7c673cae FG |
6135 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); |
6136 | if (name == "") { | |
6137 | f->open_array_section("rules"); | |
6138 | osdmap.crush->dump_rules(f.get()); | |
6139 | f->close_section(); | |
6140 | } else { | |
6141 | int ruleno = osdmap.crush->get_rule_id(name); | |
6142 | if (ruleno < 0) { | |
31f18b77 | 6143 | ss << "unknown crush rule '" << name << "'"; |
7c673cae FG |
6144 | r = ruleno; |
6145 | goto reply; | |
6146 | } | |
6147 | osdmap.crush->dump_rule(ruleno, f.get()); | |
6148 | } | |
6149 | ostringstream rs; | |
6150 | f->flush(rs); | |
6151 | rs << "\n"; | |
6152 | rdata.append(rs.str()); | |
6153 | } else if (prefix == "osd crush dump") { | |
6154 | string format; | |
11fdf7f2 | 6155 | cmd_getval(cct, cmdmap, "format", format); |
7c673cae FG |
6156 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); |
6157 | f->open_object_section("crush_map"); | |
6158 | osdmap.crush->dump(f.get()); | |
6159 | f->close_section(); | |
6160 | ostringstream rs; | |
6161 | f->flush(rs); | |
6162 | rs << "\n"; | |
6163 | rdata.append(rs.str()); | |
6164 | } else if (prefix == "osd crush show-tunables") { | |
6165 | string format; | |
11fdf7f2 | 6166 | cmd_getval(cct, cmdmap, "format", format); |
7c673cae FG |
6167 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); |
6168 | f->open_object_section("crush_map_tunables"); | |
6169 | osdmap.crush->dump_tunables(f.get()); | |
6170 | f->close_section(); | |
6171 | ostringstream rs; | |
6172 | f->flush(rs); | |
6173 | rs << "\n"; | |
6174 | rdata.append(rs.str()); | |
6175 | } else if (prefix == "osd crush tree") { | |
c07f9fc5 | 6176 | string shadow; |
11fdf7f2 | 6177 | cmd_getval(cct, cmdmap, "shadow", shadow); |
c07f9fc5 FG |
6178 | bool show_shadow = shadow == "--show-shadow"; |
6179 | boost::scoped_ptr<Formatter> f(Formatter::create(format)); | |
6180 | if (f) { | |
91327a77 | 6181 | f->open_object_section("crush_tree"); |
c07f9fc5 FG |
6182 | osdmap.crush->dump_tree(nullptr, |
6183 | f.get(), | |
6184 | osdmap.get_pool_names(), | |
6185 | show_shadow); | |
91327a77 | 6186 | f->close_section(); |
c07f9fc5 FG |
6187 | f->flush(rdata); |
6188 | } else { | |
6189 | ostringstream ss; | |
6190 | osdmap.crush->dump_tree(&ss, | |
6191 | nullptr, | |
6192 | osdmap.get_pool_names(), | |
6193 | show_shadow); | |
6194 | rdata.append(ss.str()); | |
6195 | } | |
d2e6a577 FG |
6196 | } else if (prefix == "osd crush ls") { |
6197 | string name; | |
11fdf7f2 | 6198 | if (!cmd_getval(cct, cmdmap, "node", name)) { |
d2e6a577 FG |
6199 | ss << "no node specified"; |
6200 | r = -EINVAL; | |
6201 | goto reply; | |
6202 | } | |
6203 | if (!osdmap.crush->name_exists(name)) { | |
6204 | ss << "node '" << name << "' does not exist"; | |
6205 | r = -ENOENT; | |
6206 | goto reply; | |
6207 | } | |
6208 | int id = osdmap.crush->get_item_id(name); | |
6209 | list<int> result; | |
6210 | if (id >= 0) { | |
6211 | result.push_back(id); | |
6212 | } else { | |
6213 | int num = osdmap.crush->get_bucket_size(id); | |
6214 | for (int i = 0; i < num; ++i) { | |
6215 | result.push_back(osdmap.crush->get_bucket_item(id, i)); | |
6216 | } | |
6217 | } | |
6218 | if (f) { | |
6219 | f->open_array_section("items"); | |
6220 | for (auto i : result) { | |
6221 | f->dump_string("item", osdmap.crush->get_item_name(i)); | |
6222 | } | |
6223 | f->close_section(); | |
6224 | f->flush(rdata); | |
6225 | } else { | |
6226 | ostringstream ss; | |
6227 | for (auto i : result) { | |
6228 | ss << osdmap.crush->get_item_name(i) << "\n"; | |
6229 | } | |
6230 | rdata.append(ss.str()); | |
6231 | } | |
6232 | r = 0; | |
7c673cae FG |
6233 | } else if (prefix == "osd crush class ls") { |
6234 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", "json-pretty")); | |
6235 | f->open_array_section("crush_classes"); | |
6236 | for (auto i : osdmap.crush->class_name) | |
6237 | f->dump_string("class", i.second); | |
6238 | f->close_section(); | |
6239 | f->flush(rdata); | |
224ce89b WB |
6240 | } else if (prefix == "osd crush class ls-osd") { |
6241 | string name; | |
11fdf7f2 | 6242 | cmd_getval(cct, cmdmap, "class", name); |
224ce89b WB |
6243 | set<int> osds; |
6244 | osdmap.crush->get_devices_by_class(name, &osds); | |
b5b8bbf5 FG |
6245 | if (f) { |
6246 | f->open_array_section("osds"); | |
6247 | for (auto &osd: osds) | |
6248 | f->dump_int("osd", osd); | |
6249 | f->close_section(); | |
6250 | f->flush(rdata); | |
6251 | } else { | |
6252 | bool first = true; | |
6253 | for (auto &osd : osds) { | |
6254 | if (!first) | |
6255 | ds << "\n"; | |
6256 | first = false; | |
6257 | ds << osd; | |
6258 | } | |
6259 | rdata.append(ds); | |
6260 | } | |
11fdf7f2 TL |
6261 | } else if (prefix == "osd crush get-device-class") { |
6262 | vector<string> idvec; | |
6263 | cmd_getval(cct, cmdmap, "ids", idvec); | |
6264 | map<int, string> class_by_osd; | |
6265 | for (auto& id : idvec) { | |
6266 | ostringstream ts; | |
6267 | long osd = parse_osd_id(id.c_str(), &ts); | |
6268 | if (osd < 0) { | |
6269 | ss << "unable to parse osd id:'" << id << "'"; | |
6270 | r = -EINVAL; | |
6271 | goto reply; | |
6272 | } | |
6273 | auto device_class = osdmap.crush->get_item_class(osd); | |
6274 | if (device_class) | |
6275 | class_by_osd[osd] = device_class; | |
6276 | else | |
6277 | class_by_osd[osd] = ""; // no class | |
6278 | } | |
6279 | if (f) { | |
6280 | f->open_array_section("osd_device_classes"); | |
6281 | for (auto& i : class_by_osd) { | |
6282 | f->open_object_section("osd_device_class"); | |
6283 | f->dump_int("osd", i.first); | |
6284 | f->dump_string("device_class", i.second); | |
6285 | f->close_section(); | |
6286 | } | |
6287 | f->close_section(); | |
6288 | f->flush(rdata); | |
6289 | } else { | |
6290 | if (class_by_osd.size() == 1) { | |
6291 | // for single input, make a clean output | |
6292 | ds << class_by_osd.begin()->second; | |
6293 | } else { | |
6294 | // note that we do not group osds by class here | |
6295 | for (auto it = class_by_osd.begin(); | |
6296 | it != class_by_osd.end(); | |
6297 | it++) { | |
6298 | ds << "osd." << it->first << ' ' << it->second; | |
6299 | if (next(it) != class_by_osd.end()) | |
6300 | ds << '\n'; | |
6301 | } | |
6302 | } | |
6303 | rdata.append(ds); | |
6304 | } | |
7c673cae FG |
6305 | } else if (prefix == "osd erasure-code-profile ls") { |
6306 | const auto &profiles = osdmap.get_erasure_code_profiles(); | |
6307 | if (f) | |
6308 | f->open_array_section("erasure-code-profiles"); | |
6309 | for (auto i = profiles.begin(); i != profiles.end(); ++i) { | |
6310 | if (f) | |
6311 | f->dump_string("profile", i->first.c_str()); | |
6312 | else | |
6313 | rdata.append(i->first + "\n"); | |
6314 | } | |
6315 | if (f) { | |
6316 | f->close_section(); | |
6317 | ostringstream rs; | |
6318 | f->flush(rs); | |
6319 | rs << "\n"; | |
6320 | rdata.append(rs.str()); | |
6321 | } | |
c07f9fc5 FG |
6322 | } else if (prefix == "osd crush weight-set ls") { |
6323 | boost::scoped_ptr<Formatter> f(Formatter::create(format)); | |
6324 | if (f) { | |
6325 | f->open_array_section("weight_sets"); | |
6326 | if (osdmap.crush->have_choose_args(CrushWrapper::DEFAULT_CHOOSE_ARGS)) { | |
6327 | f->dump_string("pool", "(compat)"); | |
6328 | } | |
6329 | for (auto& i : osdmap.crush->choose_args) { | |
6330 | if (i.first >= 0) { | |
6331 | f->dump_string("pool", osdmap.get_pool_name(i.first)); | |
6332 | } | |
6333 | } | |
6334 | f->close_section(); | |
6335 | f->flush(rdata); | |
6336 | } else { | |
6337 | ostringstream rs; | |
6338 | if (osdmap.crush->have_choose_args(CrushWrapper::DEFAULT_CHOOSE_ARGS)) { | |
6339 | rs << "(compat)\n"; | |
6340 | } | |
6341 | for (auto& i : osdmap.crush->choose_args) { | |
6342 | if (i.first >= 0) { | |
6343 | rs << osdmap.get_pool_name(i.first) << "\n"; | |
6344 | } | |
6345 | } | |
6346 | rdata.append(rs.str()); | |
6347 | } | |
6348 | } else if (prefix == "osd crush weight-set dump") { | |
6349 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", | |
6350 | "json-pretty")); | |
6351 | osdmap.crush->dump_choose_args(f.get()); | |
6352 | f->flush(rdata); | |
7c673cae FG |
6353 | } else if (prefix == "osd erasure-code-profile get") { |
6354 | string name; | |
11fdf7f2 | 6355 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae FG |
6356 | if (!osdmap.has_erasure_code_profile(name)) { |
6357 | ss << "unknown erasure code profile '" << name << "'"; | |
6358 | r = -ENOENT; | |
6359 | goto reply; | |
6360 | } | |
6361 | const map<string,string> &profile = osdmap.get_erasure_code_profile(name); | |
6362 | if (f) | |
6363 | f->open_object_section("profile"); | |
6364 | for (map<string,string>::const_iterator i = profile.begin(); | |
6365 | i != profile.end(); | |
6366 | ++i) { | |
6367 | if (f) | |
6368 | f->dump_string(i->first.c_str(), i->second.c_str()); | |
6369 | else | |
6370 | rdata.append(i->first + "=" + i->second + "\n"); | |
6371 | } | |
6372 | if (f) { | |
6373 | f->close_section(); | |
6374 | ostringstream rs; | |
6375 | f->flush(rs); | |
6376 | rs << "\n"; | |
6377 | rdata.append(rs.str()); | |
6378 | } | |
181888fb FG |
6379 | } else if (prefix == "osd pool application get") { |
6380 | boost::scoped_ptr<Formatter> f(Formatter::create(format, "json-pretty", | |
6381 | "json-pretty")); | |
6382 | string pool_name; | |
11fdf7f2 | 6383 | cmd_getval(cct, cmdmap, "pool", pool_name); |
181888fb | 6384 | string app; |
11fdf7f2 | 6385 | cmd_getval(cct, cmdmap, "app", app); |
181888fb | 6386 | string key; |
11fdf7f2 | 6387 | cmd_getval(cct, cmdmap, "key", key); |
181888fb FG |
6388 | |
6389 | if (pool_name.empty()) { | |
6390 | // all | |
6391 | f->open_object_section("pools"); | |
6392 | for (const auto &pool : osdmap.pools) { | |
6393 | std::string name("<unknown>"); | |
6394 | const auto &pni = osdmap.pool_name.find(pool.first); | |
6395 | if (pni != osdmap.pool_name.end()) | |
6396 | name = pni->second; | |
6397 | f->open_object_section(name.c_str()); | |
6398 | for (auto &app_pair : pool.second.application_metadata) { | |
6399 | f->open_object_section(app_pair.first.c_str()); | |
6400 | for (auto &kv_pair : app_pair.second) { | |
6401 | f->dump_string(kv_pair.first.c_str(), kv_pair.second); | |
6402 | } | |
6403 | f->close_section(); | |
6404 | } | |
6405 | f->close_section(); // name | |
6406 | } | |
6407 | f->close_section(); // pools | |
6408 | f->flush(rdata); | |
6409 | } else { | |
6410 | int64_t pool = osdmap.lookup_pg_pool_name(pool_name.c_str()); | |
6411 | if (pool < 0) { | |
6412 | ss << "unrecognized pool '" << pool_name << "'"; | |
6413 | r = -ENOENT; | |
6414 | goto reply; | |
6415 | } | |
6416 | auto p = osdmap.get_pg_pool(pool); | |
6417 | // filter by pool | |
6418 | if (app.empty()) { | |
6419 | f->open_object_section(pool_name.c_str()); | |
6420 | for (auto &app_pair : p->application_metadata) { | |
6421 | f->open_object_section(app_pair.first.c_str()); | |
6422 | for (auto &kv_pair : app_pair.second) { | |
6423 | f->dump_string(kv_pair.first.c_str(), kv_pair.second); | |
6424 | } | |
6425 | f->close_section(); // application | |
6426 | } | |
6427 | f->close_section(); // pool_name | |
6428 | f->flush(rdata); | |
6429 | goto reply; | |
6430 | } | |
6431 | ||
6432 | auto app_it = p->application_metadata.find(app); | |
6433 | if (app_it == p->application_metadata.end()) { | |
6434 | ss << "pool '" << pool_name << "' has no application '" << app << "'"; | |
6435 | r = -ENOENT; | |
6436 | goto reply; | |
6437 | } | |
6438 | // filter by pool + app | |
6439 | if (key.empty()) { | |
6440 | f->open_object_section(app_it->first.c_str()); | |
6441 | for (auto &kv_pair : app_it->second) { | |
6442 | f->dump_string(kv_pair.first.c_str(), kv_pair.second); | |
6443 | } | |
6444 | f->close_section(); // application | |
6445 | f->flush(rdata); | |
6446 | goto reply; | |
6447 | } | |
6448 | // filter by pool + app + key | |
6449 | auto key_it = app_it->second.find(key); | |
6450 | if (key_it == app_it->second.end()) { | |
6451 | ss << "application '" << app << "' on pool '" << pool_name | |
6452 | << "' does not have key '" << key << "'"; | |
6453 | r = -ENOENT; | |
6454 | goto reply; | |
6455 | } | |
6456 | ss << key_it->second << "\n"; | |
6457 | rdata.append(ss.str()); | |
6458 | ss.str(""); | |
6459 | } | |
11fdf7f2 TL |
6460 | } else if (prefix == "osd get-require-min-compat-client") { |
6461 | ss << ceph_release_name(osdmap.require_min_compat_client) << std::endl; | |
6462 | rdata.append(ss.str()); | |
6463 | ss.str(""); | |
6464 | goto reply; | |
6465 | } else if (prefix == "osd pool application enable" || | |
6466 | prefix == "osd pool application disable" || | |
6467 | prefix == "osd pool application set" || | |
6468 | prefix == "osd pool application rm") { | |
6469 | bool changed = false; | |
6470 | r = preprocess_command_pool_application(prefix, cmdmap, ss, &changed); | |
6471 | if (r != 0) { | |
6472 | // Error, reply. | |
6473 | goto reply; | |
6474 | } else if (changed) { | |
6475 | // Valid mutation, proceed to prepare phase | |
6476 | return false; | |
6477 | } else { | |
6478 | // Idempotent case, reply | |
6479 | goto reply; | |
6480 | } | |
7c673cae FG |
6481 | } else { |
6482 | // try prepare update | |
6483 | return false; | |
6484 | } | |
6485 | ||
6486 | reply: | |
6487 | string rs; | |
6488 | getline(ss, rs); | |
6489 | mon->reply_command(op, r, rs, rdata, get_last_committed()); | |
6490 | return true; | |
6491 | } | |
6492 | ||
3efd9988 FG |
6493 | void OSDMonitor::set_pool_flags(int64_t pool_id, uint64_t flags) |
6494 | { | |
6495 | pg_pool_t *pool = pending_inc.get_new_pool(pool_id, | |
6496 | osdmap.get_pg_pool(pool_id)); | |
11fdf7f2 | 6497 | ceph_assert(pool); |
3efd9988 FG |
6498 | pool->set_flag(flags); |
6499 | } | |
6500 | ||
6501 | void OSDMonitor::clear_pool_flags(int64_t pool_id, uint64_t flags) | |
7c673cae | 6502 | { |
3efd9988 FG |
6503 | pg_pool_t *pool = pending_inc.get_new_pool(pool_id, |
6504 | osdmap.get_pg_pool(pool_id)); | |
11fdf7f2 | 6505 | ceph_assert(pool); |
3efd9988 | 6506 | pool->unset_flag(flags); |
7c673cae FG |
6507 | } |
6508 | ||
11fdf7f2 TL |
6509 | string OSDMonitor::make_snap_epoch_key(int64_t pool, epoch_t epoch) |
6510 | { | |
6511 | char k[80]; | |
6512 | snprintf(k, sizeof(k), "removed_epoch_%llu_%08lx", | |
6513 | (unsigned long long)pool, (unsigned long)epoch); | |
6514 | return k; | |
6515 | } | |
6516 | ||
6517 | string OSDMonitor::make_snap_key(int64_t pool, snapid_t snap) | |
6518 | { | |
6519 | char k[80]; | |
6520 | snprintf(k, sizeof(k), "removed_snap_%llu_%016llx", | |
6521 | (unsigned long long)pool, (unsigned long long)snap); | |
6522 | return k; | |
6523 | } | |
6524 | ||
6525 | ||
6526 | string OSDMonitor::make_snap_key_value( | |
6527 | int64_t pool, snapid_t snap, snapid_t num, | |
6528 | epoch_t epoch, bufferlist *v) | |
6529 | { | |
6530 | // encode the *last* epoch in the key so that we can use forward | |
6531 | // iteration only to search for an epoch in an interval. | |
6532 | encode(snap, *v); | |
6533 | encode(snap + num, *v); | |
6534 | encode(epoch, *v); | |
6535 | return make_snap_key(pool, snap + num - 1); | |
6536 | } | |
6537 | ||
6538 | string OSDMonitor::make_snap_purged_key(int64_t pool, snapid_t snap) | |
6539 | { | |
6540 | char k[80]; | |
6541 | snprintf(k, sizeof(k), "purged_snap_%llu_%016llx", | |
6542 | (unsigned long long)pool, (unsigned long long)snap); | |
6543 | return k; | |
6544 | } | |
6545 | string OSDMonitor::make_snap_purged_key_value( | |
6546 | int64_t pool, snapid_t snap, snapid_t num, | |
6547 | epoch_t epoch, bufferlist *v) | |
6548 | { | |
6549 | // encode the *last* epoch in the key so that we can use forward | |
6550 | // iteration only to search for an epoch in an interval. | |
6551 | encode(snap, *v); | |
6552 | encode(snap + num, *v); | |
6553 | encode(epoch, *v); | |
6554 | return make_snap_purged_key(pool, snap + num - 1); | |
6555 | } | |
6556 | ||
6557 | int OSDMonitor::lookup_pruned_snap(int64_t pool, snapid_t snap, | |
6558 | snapid_t *begin, snapid_t *end) | |
6559 | { | |
6560 | string k = make_snap_key(pool, snap); | |
6561 | auto it = mon->store->get_iterator(OSD_SNAP_PREFIX); | |
6562 | it->lower_bound(k); | |
6563 | if (!it->valid()) { | |
6564 | return -ENOENT; | |
6565 | } | |
6566 | if (it->key().find(OSD_SNAP_PREFIX) != 0) { | |
6567 | return -ENOENT; | |
6568 | } | |
6569 | bufferlist v = it->value(); | |
6570 | auto p = v.cbegin(); | |
6571 | decode(*begin, p); | |
6572 | decode(*end, p); | |
6573 | if (snap < *begin || snap >= *end) { | |
6574 | return -ENOENT; | |
6575 | } | |
6576 | return 0; | |
6577 | } | |
6578 | ||
6579 | bool OSDMonitor::try_prune_purged_snaps() | |
6580 | { | |
6581 | if (!mon->mgrstatmon()->is_readable()) { | |
6582 | return false; | |
6583 | } | |
6584 | if (osdmap.require_osd_release < CEPH_RELEASE_MIMIC) { | |
6585 | return false; | |
6586 | } | |
6587 | if (!pending_inc.new_purged_snaps.empty()) { | |
6588 | return false; // we already pruned for this epoch | |
6589 | } | |
6590 | ||
6591 | unsigned max_prune = cct->_conf.get_val<uint64_t>( | |
6592 | "mon_max_snap_prune_per_epoch"); | |
6593 | if (!max_prune) { | |
6594 | max_prune = 100000; | |
6595 | } | |
6596 | dout(10) << __func__ << " max_prune " << max_prune << dendl; | |
6597 | ||
6598 | unsigned actually_pruned = 0; | |
6599 | auto& purged_snaps = mon->mgrstatmon()->get_digest().purged_snaps; | |
6600 | for (auto& p : osdmap.get_pools()) { | |
6601 | auto q = purged_snaps.find(p.first); | |
6602 | if (q == purged_snaps.end()) { | |
6603 | continue; | |
6604 | } | |
6605 | auto& purged = q->second; | |
6606 | if (purged.empty()) { | |
6607 | dout(20) << __func__ << " " << p.first << " nothing purged" << dendl; | |
6608 | continue; | |
6609 | } | |
6610 | dout(20) << __func__ << " pool " << p.first << " purged " << purged << dendl; | |
6611 | OSDMap::snap_interval_set_t to_prune; | |
6612 | unsigned maybe_pruned = actually_pruned; | |
6613 | for (auto i = purged.begin(); i != purged.end(); ++i) { | |
6614 | snapid_t begin = i.get_start(); | |
6615 | auto end = i.get_start() + i.get_len(); | |
6616 | snapid_t pbegin = 0, pend = 0; | |
6617 | int r = lookup_pruned_snap(p.first, begin, &pbegin, &pend); | |
6618 | if (r == 0) { | |
6619 | // already purged. | |
6620 | // be a bit aggressive about backing off here, because the mon may | |
6621 | // do a lot of work going through this set, and if we know the | |
6622 | // purged set from the OSDs is at least *partly* stale we may as | |
6623 | // well wait for it to be fresh. | |
6624 | dout(20) << __func__ << " we've already pruned " << pbegin | |
6625 | << "~" << (pend - pbegin) << dendl; | |
6626 | break; // next pool | |
6627 | } | |
6628 | if (pbegin && pbegin < end) { | |
6629 | // the tail of [begin,end) is purged; shorten the range | |
6630 | ceph_assert(pbegin > begin); | |
6631 | end = pbegin; | |
6632 | } | |
6633 | to_prune.insert(begin, end - begin); | |
6634 | maybe_pruned += end - begin; | |
6635 | if (maybe_pruned >= max_prune) { | |
6636 | break; | |
6637 | } | |
6638 | } | |
6639 | if (!to_prune.empty()) { | |
6640 | // PGs may still be reporting things as purged that we have already | |
6641 | // pruned from removed_snaps_queue. | |
6642 | OSDMap::snap_interval_set_t actual; | |
6643 | auto r = osdmap.removed_snaps_queue.find(p.first); | |
6644 | if (r != osdmap.removed_snaps_queue.end()) { | |
6645 | actual.intersection_of(to_prune, r->second); | |
6646 | } | |
6647 | actually_pruned += actual.size(); | |
6648 | dout(10) << __func__ << " pool " << p.first << " reports pruned " << to_prune | |
6649 | << ", actual pruned " << actual << dendl; | |
6650 | if (!actual.empty()) { | |
6651 | pending_inc.new_purged_snaps[p.first].swap(actual); | |
6652 | } | |
6653 | } | |
6654 | if (actually_pruned >= max_prune) { | |
6655 | break; | |
6656 | } | |
6657 | } | |
6658 | dout(10) << __func__ << " actually pruned " << actually_pruned << dendl; | |
6659 | return !!actually_pruned; | |
6660 | } | |
6661 | ||
7c673cae FG |
6662 | bool OSDMonitor::update_pools_status() |
6663 | { | |
11fdf7f2 | 6664 | if (!mon->mgrstatmon()->is_readable()) |
7c673cae FG |
6665 | return false; |
6666 | ||
6667 | bool ret = false; | |
6668 | ||
6669 | auto& pools = osdmap.get_pools(); | |
6670 | for (auto it = pools.begin(); it != pools.end(); ++it) { | |
11fdf7f2 | 6671 | const pool_stat_t *pstat = mon->mgrstatmon()->get_pool_stat(it->first); |
31f18b77 | 6672 | if (!pstat) |
7c673cae | 6673 | continue; |
31f18b77 | 6674 | const object_stat_sum_t& sum = pstat->stats.sum; |
7c673cae FG |
6675 | const pg_pool_t &pool = it->second; |
6676 | const string& pool_name = osdmap.get_pool_name(it->first); | |
6677 | ||
6678 | bool pool_is_full = | |
6679 | (pool.quota_max_bytes > 0 && (uint64_t)sum.num_bytes >= pool.quota_max_bytes) || | |
6680 | (pool.quota_max_objects > 0 && (uint64_t)sum.num_objects >= pool.quota_max_objects); | |
6681 | ||
11fdf7f2 | 6682 | if (pool.has_flag(pg_pool_t::FLAG_FULL_QUOTA)) { |
7c673cae FG |
6683 | if (pool_is_full) |
6684 | continue; | |
6685 | ||
6686 | mon->clog->info() << "pool '" << pool_name | |
3efd9988 FG |
6687 | << "' no longer out of quota; removing NO_QUOTA flag"; |
6688 | // below we cancel FLAG_FULL too, we'll set it again in | |
6689 | // OSDMonitor::encode_pending if it still fails the osd-full checking. | |
6690 | clear_pool_flags(it->first, | |
11fdf7f2 | 6691 | pg_pool_t::FLAG_FULL_QUOTA | pg_pool_t::FLAG_FULL); |
7c673cae FG |
6692 | ret = true; |
6693 | } else { | |
6694 | if (!pool_is_full) | |
6695 | continue; | |
6696 | ||
6697 | if (pool.quota_max_bytes > 0 && | |
6698 | (uint64_t)sum.num_bytes >= pool.quota_max_bytes) { | |
6699 | mon->clog->warn() << "pool '" << pool_name << "' is full" | |
6700 | << " (reached quota's max_bytes: " | |
1adf2230 | 6701 | << byte_u_t(pool.quota_max_bytes) << ")"; |
7c673cae FG |
6702 | } |
6703 | if (pool.quota_max_objects > 0 && | |
6704 | (uint64_t)sum.num_objects >= pool.quota_max_objects) { | |
6705 | mon->clog->warn() << "pool '" << pool_name << "' is full" | |
6706 | << " (reached quota's max_objects: " | |
6707 | << pool.quota_max_objects << ")"; | |
6708 | } | |
11fdf7f2 | 6709 | // set both FLAG_FULL_QUOTA and FLAG_FULL |
3efd9988 FG |
6710 | // note that below we try to cancel FLAG_BACKFILLFULL/NEARFULL too |
6711 | // since FLAG_FULL should always take precedence | |
6712 | set_pool_flags(it->first, | |
11fdf7f2 | 6713 | pg_pool_t::FLAG_FULL_QUOTA | pg_pool_t::FLAG_FULL); |
3efd9988 FG |
6714 | clear_pool_flags(it->first, |
6715 | pg_pool_t::FLAG_NEARFULL | | |
6716 | pg_pool_t::FLAG_BACKFILLFULL); | |
7c673cae FG |
6717 | ret = true; |
6718 | } | |
6719 | } | |
6720 | return ret; | |
6721 | } | |
6722 | ||
7c673cae FG |
6723 | int OSDMonitor::prepare_new_pool(MonOpRequestRef op) |
6724 | { | |
6725 | op->mark_osdmon_event(__func__); | |
6726 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); | |
6727 | dout(10) << "prepare_new_pool from " << m->get_connection() << dendl; | |
11fdf7f2 | 6728 | MonSession *session = op->get_session(); |
7c673cae FG |
6729 | if (!session) |
6730 | return -EPERM; | |
6731 | string erasure_code_profile; | |
6732 | stringstream ss; | |
31f18b77 | 6733 | string rule_name; |
94b18763 | 6734 | int ret = 0; |
11fdf7f2 TL |
6735 | ret = prepare_new_pool(m->name, m->crush_rule, rule_name, |
6736 | 0, 0, 0, 0, 0, 0.0, | |
6737 | erasure_code_profile, | |
6738 | pg_pool_t::TYPE_REPLICATED, 0, FAST_READ_OFF, &ss); | |
94b18763 FG |
6739 | |
6740 | if (ret < 0) { | |
6741 | dout(10) << __func__ << " got " << ret << " " << ss.str() << dendl; | |
6742 | } | |
6743 | return ret; | |
7c673cae FG |
6744 | } |
6745 | ||
6746 | int OSDMonitor::crush_rename_bucket(const string& srcname, | |
6747 | const string& dstname, | |
6748 | ostream *ss) | |
6749 | { | |
6750 | int ret; | |
6751 | // | |
6752 | // Avoid creating a pending crush if it does not already exists and | |
6753 | // the rename would fail. | |
6754 | // | |
6755 | if (!_have_pending_crush()) { | |
6756 | ret = _get_stable_crush().can_rename_bucket(srcname, | |
6757 | dstname, | |
6758 | ss); | |
6759 | if (ret) | |
6760 | return ret; | |
6761 | } | |
6762 | ||
6763 | CrushWrapper newcrush; | |
6764 | _get_pending_crush(newcrush); | |
6765 | ||
6766 | ret = newcrush.rename_bucket(srcname, | |
6767 | dstname, | |
6768 | ss); | |
6769 | if (ret) | |
6770 | return ret; | |
6771 | ||
6772 | pending_inc.crush.clear(); | |
6773 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
6774 | *ss << "renamed bucket " << srcname << " into " << dstname; | |
6775 | return 0; | |
6776 | } | |
6777 | ||
6778 | void OSDMonitor::check_legacy_ec_plugin(const string& plugin, const string& profile) const | |
6779 | { | |
6780 | string replacement = ""; | |
6781 | ||
6782 | if (plugin == "jerasure_generic" || | |
6783 | plugin == "jerasure_sse3" || | |
6784 | plugin == "jerasure_sse4" || | |
6785 | plugin == "jerasure_neon") { | |
6786 | replacement = "jerasure"; | |
6787 | } else if (plugin == "shec_generic" || | |
6788 | plugin == "shec_sse3" || | |
6789 | plugin == "shec_sse4" || | |
6790 | plugin == "shec_neon") { | |
6791 | replacement = "shec"; | |
6792 | } | |
6793 | ||
6794 | if (replacement != "") { | |
6795 | dout(0) << "WARNING: erasure coding profile " << profile << " uses plugin " | |
6796 | << plugin << " that has been deprecated. Please use " | |
6797 | << replacement << " instead." << dendl; | |
6798 | } | |
6799 | } | |
6800 | ||
6801 | int OSDMonitor::normalize_profile(const string& profilename, | |
6802 | ErasureCodeProfile &profile, | |
6803 | bool force, | |
6804 | ostream *ss) | |
6805 | { | |
6806 | ErasureCodeInterfaceRef erasure_code; | |
6807 | ErasureCodePluginRegistry &instance = ErasureCodePluginRegistry::instance(); | |
6808 | ErasureCodeProfile::const_iterator plugin = profile.find("plugin"); | |
6809 | check_legacy_ec_plugin(plugin->second, profilename); | |
6810 | int err = instance.factory(plugin->second, | |
11fdf7f2 | 6811 | g_conf().get_val<std::string>("erasure_code_dir"), |
7c673cae FG |
6812 | profile, &erasure_code, ss); |
6813 | if (err) { | |
6814 | return err; | |
6815 | } | |
6816 | ||
6817 | err = erasure_code->init(profile, ss); | |
6818 | if (err) { | |
6819 | return err; | |
6820 | } | |
6821 | ||
6822 | auto it = profile.find("stripe_unit"); | |
6823 | if (it != profile.end()) { | |
6824 | string err_str; | |
1adf2230 | 6825 | uint32_t stripe_unit = strict_iecstrtoll(it->second.c_str(), &err_str); |
7c673cae FG |
6826 | if (!err_str.empty()) { |
6827 | *ss << "could not parse stripe_unit '" << it->second | |
6828 | << "': " << err_str << std::endl; | |
6829 | return -EINVAL; | |
6830 | } | |
6831 | uint32_t data_chunks = erasure_code->get_data_chunk_count(); | |
6832 | uint32_t chunk_size = erasure_code->get_chunk_size(stripe_unit * data_chunks); | |
6833 | if (chunk_size != stripe_unit) { | |
6834 | *ss << "stripe_unit " << stripe_unit << " does not match ec profile " | |
6835 | << "alignment. Would be padded to " << chunk_size | |
6836 | << std::endl; | |
6837 | return -EINVAL; | |
6838 | } | |
6839 | if ((stripe_unit % 4096) != 0 && !force) { | |
6840 | *ss << "stripe_unit should be a multiple of 4096 bytes for best performance." | |
6841 | << "use --force to override this check" << std::endl; | |
6842 | return -EINVAL; | |
6843 | } | |
6844 | } | |
6845 | return 0; | |
6846 | } | |
6847 | ||
31f18b77 | 6848 | int OSDMonitor::crush_rule_create_erasure(const string &name, |
7c673cae | 6849 | const string &profile, |
31f18b77 | 6850 | int *rule, |
7c673cae FG |
6851 | ostream *ss) |
6852 | { | |
6853 | int ruleid = osdmap.crush->get_rule_id(name); | |
6854 | if (ruleid != -ENOENT) { | |
31f18b77 | 6855 | *rule = osdmap.crush->get_rule_mask_ruleset(ruleid); |
7c673cae FG |
6856 | return -EEXIST; |
6857 | } | |
6858 | ||
6859 | CrushWrapper newcrush; | |
6860 | _get_pending_crush(newcrush); | |
6861 | ||
6862 | ruleid = newcrush.get_rule_id(name); | |
6863 | if (ruleid != -ENOENT) { | |
31f18b77 | 6864 | *rule = newcrush.get_rule_mask_ruleset(ruleid); |
7c673cae FG |
6865 | return -EALREADY; |
6866 | } else { | |
6867 | ErasureCodeInterfaceRef erasure_code; | |
6868 | int err = get_erasure_code(profile, &erasure_code, ss); | |
6869 | if (err) { | |
6870 | *ss << "failed to load plugin using profile " << profile << std::endl; | |
6871 | return err; | |
6872 | } | |
6873 | ||
224ce89b | 6874 | err = erasure_code->create_rule(name, newcrush, ss); |
7c673cae FG |
6875 | erasure_code.reset(); |
6876 | if (err < 0) | |
6877 | return err; | |
31f18b77 | 6878 | *rule = err; |
7c673cae FG |
6879 | pending_inc.crush.clear(); |
6880 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
6881 | return 0; | |
6882 | } | |
6883 | } | |
6884 | ||
6885 | int OSDMonitor::get_erasure_code(const string &erasure_code_profile, | |
6886 | ErasureCodeInterfaceRef *erasure_code, | |
6887 | ostream *ss) const | |
6888 | { | |
6889 | if (pending_inc.has_erasure_code_profile(erasure_code_profile)) | |
6890 | return -EAGAIN; | |
6891 | ErasureCodeProfile profile = | |
6892 | osdmap.get_erasure_code_profile(erasure_code_profile); | |
6893 | ErasureCodeProfile::const_iterator plugin = | |
6894 | profile.find("plugin"); | |
6895 | if (plugin == profile.end()) { | |
6896 | *ss << "cannot determine the erasure code plugin" | |
6897 | << " because there is no 'plugin' entry in the erasure_code_profile " | |
6898 | << profile << std::endl; | |
6899 | return -EINVAL; | |
6900 | } | |
6901 | check_legacy_ec_plugin(plugin->second, erasure_code_profile); | |
6902 | ErasureCodePluginRegistry &instance = ErasureCodePluginRegistry::instance(); | |
6903 | return instance.factory(plugin->second, | |
11fdf7f2 | 6904 | g_conf().get_val<std::string>("erasure_code_dir"), |
7c673cae FG |
6905 | profile, erasure_code, ss); |
6906 | } | |
6907 | ||
6908 | int OSDMonitor::check_cluster_features(uint64_t features, | |
6909 | stringstream &ss) | |
6910 | { | |
6911 | stringstream unsupported_ss; | |
6912 | int unsupported_count = 0; | |
6913 | if ((mon->get_quorum_con_features() & features) != features) { | |
6914 | unsupported_ss << "the monitor cluster"; | |
6915 | ++unsupported_count; | |
6916 | } | |
6917 | ||
6918 | set<int32_t> up_osds; | |
6919 | osdmap.get_up_osds(up_osds); | |
6920 | for (set<int32_t>::iterator it = up_osds.begin(); | |
6921 | it != up_osds.end(); ++it) { | |
6922 | const osd_xinfo_t &xi = osdmap.get_xinfo(*it); | |
6923 | if ((xi.features & features) != features) { | |
6924 | if (unsupported_count > 0) | |
6925 | unsupported_ss << ", "; | |
6926 | unsupported_ss << "osd." << *it; | |
6927 | unsupported_count ++; | |
6928 | } | |
6929 | } | |
6930 | ||
6931 | if (unsupported_count > 0) { | |
6932 | ss << "features " << features << " unsupported by: " | |
6933 | << unsupported_ss.str(); | |
6934 | return -ENOTSUP; | |
6935 | } | |
6936 | ||
6937 | // check pending osd state, too! | |
6938 | for (map<int32_t,osd_xinfo_t>::const_iterator p = | |
6939 | pending_inc.new_xinfo.begin(); | |
6940 | p != pending_inc.new_xinfo.end(); ++p) { | |
6941 | const osd_xinfo_t &xi = p->second; | |
6942 | if ((xi.features & features) != features) { | |
6943 | dout(10) << __func__ << " pending osd." << p->first | |
6944 | << " features are insufficient; retry" << dendl; | |
6945 | return -EAGAIN; | |
6946 | } | |
6947 | } | |
6948 | ||
6949 | return 0; | |
6950 | } | |
6951 | ||
6952 | bool OSDMonitor::validate_crush_against_features(const CrushWrapper *newcrush, | |
6953 | stringstream& ss) | |
6954 | { | |
6955 | OSDMap::Incremental new_pending = pending_inc; | |
11fdf7f2 | 6956 | encode(*newcrush, new_pending.crush, mon->get_quorum_con_features()); |
7c673cae FG |
6957 | OSDMap newmap; |
6958 | newmap.deepish_copy_from(osdmap); | |
6959 | newmap.apply_incremental(new_pending); | |
6960 | ||
6961 | // client compat | |
31f18b77 | 6962 | if (newmap.require_min_compat_client > 0) { |
7c673cae | 6963 | auto mv = newmap.get_min_compat_client(); |
31f18b77 FG |
6964 | if (mv > newmap.require_min_compat_client) { |
6965 | ss << "new crush map requires client version " << ceph_release_name(mv) | |
7c673cae | 6966 | << " but require_min_compat_client is " |
31f18b77 | 6967 | << ceph_release_name(newmap.require_min_compat_client); |
7c673cae FG |
6968 | return false; |
6969 | } | |
6970 | } | |
6971 | ||
6972 | // osd compat | |
6973 | uint64_t features = | |
6974 | newmap.get_features(CEPH_ENTITY_TYPE_MON, NULL) | | |
6975 | newmap.get_features(CEPH_ENTITY_TYPE_OSD, NULL); | |
6976 | stringstream features_ss; | |
6977 | int r = check_cluster_features(features, features_ss); | |
6978 | if (r) { | |
6979 | ss << "Could not change CRUSH: " << features_ss.str(); | |
6980 | return false; | |
6981 | } | |
6982 | ||
6983 | return true; | |
6984 | } | |
6985 | ||
6986 | bool OSDMonitor::erasure_code_profile_in_use( | |
6987 | const mempool::osdmap::map<int64_t, pg_pool_t> &pools, | |
6988 | const string &profile, | |
6989 | ostream *ss) | |
6990 | { | |
6991 | bool found = false; | |
6992 | for (map<int64_t, pg_pool_t>::const_iterator p = pools.begin(); | |
6993 | p != pools.end(); | |
6994 | ++p) { | |
11fdf7f2 | 6995 | if (p->second.erasure_code_profile == profile && p->second.is_erasure()) { |
7c673cae FG |
6996 | *ss << osdmap.pool_name[p->first] << " "; |
6997 | found = true; | |
6998 | } | |
6999 | } | |
7000 | if (found) { | |
7001 | *ss << "pool(s) are using the erasure code profile '" << profile << "'"; | |
7002 | } | |
7003 | return found; | |
7004 | } | |
7005 | ||
7006 | int OSDMonitor::parse_erasure_code_profile(const vector<string> &erasure_code_profile, | |
7007 | map<string,string> *erasure_code_profile_map, | |
7008 | ostream *ss) | |
7009 | { | |
11fdf7f2 TL |
7010 | int r = g_conf().with_val<string>("osd_pool_default_erasure_code_profile", |
7011 | get_json_str_map, | |
7012 | *ss, | |
7013 | erasure_code_profile_map, | |
7014 | true); | |
7c673cae FG |
7015 | if (r) |
7016 | return r; | |
11fdf7f2 | 7017 | ceph_assert((*erasure_code_profile_map).count("plugin")); |
7c673cae FG |
7018 | string default_plugin = (*erasure_code_profile_map)["plugin"]; |
7019 | map<string,string> user_map; | |
7020 | for (vector<string>::const_iterator i = erasure_code_profile.begin(); | |
7021 | i != erasure_code_profile.end(); | |
7022 | ++i) { | |
7023 | size_t equal = i->find('='); | |
7024 | if (equal == string::npos) { | |
7025 | user_map[*i] = string(); | |
7026 | (*erasure_code_profile_map)[*i] = string(); | |
7027 | } else { | |
11fdf7f2 | 7028 | const string key = i->substr(0, equal); |
7c673cae FG |
7029 | equal++; |
7030 | const string value = i->substr(equal); | |
11fdf7f2 TL |
7031 | if (key.find("ruleset-") == 0) { |
7032 | *ss << "property '" << key << "' is no longer supported; try " | |
7033 | << "'crush-" << key.substr(8) << "' instead"; | |
7034 | return -EINVAL; | |
3efd9988 | 7035 | } |
7c673cae FG |
7036 | user_map[key] = value; |
7037 | (*erasure_code_profile_map)[key] = value; | |
7038 | } | |
7039 | } | |
7040 | ||
7041 | if (user_map.count("plugin") && user_map["plugin"] != default_plugin) | |
7042 | (*erasure_code_profile_map) = user_map; | |
7043 | ||
7044 | return 0; | |
7045 | } | |
7046 | ||
7047 | int OSDMonitor::prepare_pool_size(const unsigned pool_type, | |
7048 | const string &erasure_code_profile, | |
11fdf7f2 | 7049 | uint8_t repl_size, |
7c673cae FG |
7050 | unsigned *size, unsigned *min_size, |
7051 | ostream *ss) | |
7052 | { | |
7053 | int err = 0; | |
7054 | switch (pool_type) { | |
7055 | case pg_pool_t::TYPE_REPLICATED: | |
11fdf7f2 TL |
7056 | if (repl_size == 0) { |
7057 | repl_size = g_conf().get_val<uint64_t>("osd_pool_default_size"); | |
7058 | } | |
7059 | *size = repl_size; | |
7060 | *min_size = g_conf().get_osd_pool_default_min_size(repl_size); | |
7c673cae FG |
7061 | break; |
7062 | case pg_pool_t::TYPE_ERASURE: | |
7063 | { | |
7064 | ErasureCodeInterfaceRef erasure_code; | |
7065 | err = get_erasure_code(erasure_code_profile, &erasure_code, ss); | |
7066 | if (err == 0) { | |
7067 | *size = erasure_code->get_chunk_count(); | |
11fdf7f2 TL |
7068 | *min_size = |
7069 | erasure_code->get_data_chunk_count() + | |
7070 | std::min<int>(1, erasure_code->get_coding_chunk_count() - 1); | |
7071 | assert(*min_size <= *size); | |
7072 | assert(*min_size >= erasure_code->get_data_chunk_count()); | |
7c673cae FG |
7073 | } |
7074 | } | |
7075 | break; | |
7076 | default: | |
7077 | *ss << "prepare_pool_size: " << pool_type << " is not a known pool type"; | |
7078 | err = -EINVAL; | |
7079 | break; | |
7080 | } | |
7081 | return err; | |
7082 | } | |
7083 | ||
7084 | int OSDMonitor::prepare_pool_stripe_width(const unsigned pool_type, | |
7085 | const string &erasure_code_profile, | |
7086 | uint32_t *stripe_width, | |
7087 | ostream *ss) | |
7088 | { | |
7089 | int err = 0; | |
7090 | switch (pool_type) { | |
7091 | case pg_pool_t::TYPE_REPLICATED: | |
7092 | // ignored | |
7093 | break; | |
7094 | case pg_pool_t::TYPE_ERASURE: | |
7095 | { | |
7096 | ErasureCodeProfile profile = | |
7097 | osdmap.get_erasure_code_profile(erasure_code_profile); | |
7098 | ErasureCodeInterfaceRef erasure_code; | |
7099 | err = get_erasure_code(erasure_code_profile, &erasure_code, ss); | |
7100 | if (err) | |
7101 | break; | |
7102 | uint32_t data_chunks = erasure_code->get_data_chunk_count(); | |
11fdf7f2 | 7103 | uint32_t stripe_unit = g_conf().get_val<Option::size_t>("osd_pool_erasure_code_stripe_unit"); |
7c673cae FG |
7104 | auto it = profile.find("stripe_unit"); |
7105 | if (it != profile.end()) { | |
7106 | string err_str; | |
1adf2230 | 7107 | stripe_unit = strict_iecstrtoll(it->second.c_str(), &err_str); |
11fdf7f2 | 7108 | ceph_assert(err_str.empty()); |
7c673cae FG |
7109 | } |
7110 | *stripe_width = data_chunks * | |
7111 | erasure_code->get_chunk_size(stripe_unit * data_chunks); | |
7112 | } | |
7113 | break; | |
7114 | default: | |
7115 | *ss << "prepare_pool_stripe_width: " | |
7116 | << pool_type << " is not a known pool type"; | |
7117 | err = -EINVAL; | |
7118 | break; | |
7119 | } | |
7120 | return err; | |
7121 | } | |
7122 | ||
31f18b77 | 7123 | int OSDMonitor::prepare_pool_crush_rule(const unsigned pool_type, |
224ce89b WB |
7124 | const string &erasure_code_profile, |
7125 | const string &rule_name, | |
7126 | int *crush_rule, | |
7127 | ostream *ss) | |
7c673cae FG |
7128 | { |
7129 | ||
31f18b77 | 7130 | if (*crush_rule < 0) { |
7c673cae FG |
7131 | switch (pool_type) { |
7132 | case pg_pool_t::TYPE_REPLICATED: | |
7133 | { | |
31f18b77 | 7134 | if (rule_name == "") { |
224ce89b | 7135 | // Use default rule |
11fdf7f2 | 7136 | *crush_rule = osdmap.crush->get_osd_pool_default_crush_replicated_ruleset(cct); |
31f18b77 FG |
7137 | if (*crush_rule < 0) { |
7138 | // Errors may happen e.g. if no valid rule is available | |
7139 | *ss << "No suitable CRUSH rule exists, check " | |
7c673cae FG |
7140 | << "'osd pool default crush *' config options"; |
7141 | return -ENOENT; | |
7142 | } | |
7143 | } else { | |
31f18b77 | 7144 | return get_crush_rule(rule_name, crush_rule, ss); |
7c673cae FG |
7145 | } |
7146 | } | |
7147 | break; | |
7148 | case pg_pool_t::TYPE_ERASURE: | |
7149 | { | |
31f18b77 | 7150 | int err = crush_rule_create_erasure(rule_name, |
7c673cae | 7151 | erasure_code_profile, |
31f18b77 | 7152 | crush_rule, ss); |
7c673cae FG |
7153 | switch (err) { |
7154 | case -EALREADY: | |
31f18b77 FG |
7155 | dout(20) << "prepare_pool_crush_rule: rule " |
7156 | << rule_name << " try again" << dendl; | |
7c673cae FG |
7157 | // fall through |
7158 | case 0: | |
7159 | // need to wait for the crush rule to be proposed before proceeding | |
7160 | err = -EAGAIN; | |
7161 | break; | |
7162 | case -EEXIST: | |
7163 | err = 0; | |
7164 | break; | |
7165 | } | |
7166 | return err; | |
7167 | } | |
7168 | break; | |
7169 | default: | |
31f18b77 | 7170 | *ss << "prepare_pool_crush_rule: " << pool_type |
7c673cae FG |
7171 | << " is not a known pool type"; |
7172 | return -EINVAL; | |
7173 | break; | |
7174 | } | |
7175 | } else { | |
31f18b77 FG |
7176 | if (!osdmap.crush->ruleset_exists(*crush_rule)) { |
7177 | *ss << "CRUSH rule " << *crush_rule << " not found"; | |
7c673cae FG |
7178 | return -ENOENT; |
7179 | } | |
7180 | } | |
7181 | ||
7182 | return 0; | |
7183 | } | |
7184 | ||
31f18b77 | 7185 | int OSDMonitor::get_crush_rule(const string &rule_name, |
224ce89b WB |
7186 | int *crush_rule, |
7187 | ostream *ss) | |
7c673cae FG |
7188 | { |
7189 | int ret; | |
31f18b77 | 7190 | ret = osdmap.crush->get_rule_id(rule_name); |
7c673cae FG |
7191 | if (ret != -ENOENT) { |
7192 | // found it, use it | |
31f18b77 | 7193 | *crush_rule = ret; |
7c673cae FG |
7194 | } else { |
7195 | CrushWrapper newcrush; | |
7196 | _get_pending_crush(newcrush); | |
7197 | ||
31f18b77 | 7198 | ret = newcrush.get_rule_id(rule_name); |
7c673cae FG |
7199 | if (ret != -ENOENT) { |
7200 | // found it, wait for it to be proposed | |
31f18b77 | 7201 | dout(20) << __func__ << ": rule " << rule_name |
7c673cae FG |
7202 | << " try again" << dendl; |
7203 | return -EAGAIN; | |
7204 | } else { | |
224ce89b | 7205 | // Cannot find it , return error |
31f18b77 | 7206 | *ss << "specified rule " << rule_name << " doesn't exist"; |
7c673cae FG |
7207 | return ret; |
7208 | } | |
7209 | } | |
7210 | return 0; | |
7211 | } | |
7212 | ||
3efd9988 FG |
7213 | int OSDMonitor::check_pg_num(int64_t pool, int pg_num, int size, ostream *ss) |
7214 | { | |
11fdf7f2 | 7215 | auto max_pgs_per_osd = g_conf().get_val<uint64_t>("mon_max_pg_per_osd"); |
3efd9988 FG |
7216 | auto num_osds = std::max(osdmap.get_num_in_osds(), 3u); // assume min cluster size 3 |
7217 | auto max_pgs = max_pgs_per_osd * num_osds; | |
7218 | uint64_t projected = 0; | |
7219 | if (pool < 0) { | |
7220 | projected += pg_num * size; | |
7221 | } | |
7222 | for (const auto& i : osdmap.get_pools()) { | |
7223 | if (i.first == pool) { | |
7224 | projected += pg_num * size; | |
7225 | } else { | |
11fdf7f2 | 7226 | projected += i.second.get_pg_num_target() * i.second.get_size(); |
3efd9988 FG |
7227 | } |
7228 | } | |
7229 | if (projected > max_pgs) { | |
7230 | if (pool >= 0) { | |
7231 | *ss << "pool id " << pool; | |
7232 | } | |
7233 | *ss << " pg_num " << pg_num << " size " << size | |
7234 | << " would mean " << projected | |
7235 | << " total pgs, which exceeds max " << max_pgs | |
7236 | << " (mon_max_pg_per_osd " << max_pgs_per_osd | |
7237 | << " * num_in_osds " << num_osds << ")"; | |
7238 | return -ERANGE; | |
7239 | } | |
7240 | return 0; | |
7241 | } | |
7242 | ||
7c673cae FG |
7243 | /** |
7244 | * @param name The name of the new pool | |
31f18b77 FG |
7245 | * @param crush_rule The crush rule to use. If <0, will use the system default |
7246 | * @param crush_rule_name The crush rule to use, if crush_rulset <0 | |
7c673cae FG |
7247 | * @param pg_num The pg_num to use. If set to 0, will use the system default |
7248 | * @param pgp_num The pgp_num to use. If set to 0, will use the system default | |
11fdf7f2 | 7249 | * @param repl_size Replication factor, or 0 for default |
7c673cae FG |
7250 | * @param erasure_code_profile The profile name in OSDMap to be used for erasure code |
7251 | * @param pool_type TYPE_ERASURE, or TYPE_REP | |
7252 | * @param expected_num_objects expected number of objects on the pool | |
7253 | * @param fast_read fast read type. | |
7254 | * @param ss human readable error message, if any. | |
7255 | * | |
7256 | * @return 0 on success, negative errno on failure. | |
7257 | */ | |
11fdf7f2 | 7258 | int OSDMonitor::prepare_new_pool(string& name, |
31f18b77 FG |
7259 | int crush_rule, |
7260 | const string &crush_rule_name, | |
7c673cae | 7261 | unsigned pg_num, unsigned pgp_num, |
11fdf7f2 TL |
7262 | unsigned pg_num_min, |
7263 | const uint64_t repl_size, | |
7264 | const uint64_t target_size_bytes, | |
7265 | const float target_size_ratio, | |
7c673cae FG |
7266 | const string &erasure_code_profile, |
7267 | const unsigned pool_type, | |
7268 | const uint64_t expected_num_objects, | |
7269 | FastReadType fast_read, | |
7270 | ostream *ss) | |
7271 | { | |
7272 | if (name.length() == 0) | |
7273 | return -EINVAL; | |
7274 | if (pg_num == 0) | |
11fdf7f2 | 7275 | pg_num = g_conf().get_val<uint64_t>("osd_pool_default_pg_num"); |
7c673cae | 7276 | if (pgp_num == 0) |
11fdf7f2 TL |
7277 | pgp_num = g_conf().get_val<uint64_t>("osd_pool_default_pgp_num"); |
7278 | if (!pgp_num) | |
7279 | pgp_num = pg_num; | |
7280 | if (pg_num > g_conf().get_val<uint64_t>("mon_max_pool_pg_num")) { | |
7c673cae | 7281 | *ss << "'pg_num' must be greater than 0 and less than or equal to " |
11fdf7f2 | 7282 | << g_conf().get_val<uint64_t>("mon_max_pool_pg_num") |
7c673cae FG |
7283 | << " (you may adjust 'mon max pool pg num' for higher values)"; |
7284 | return -ERANGE; | |
7285 | } | |
7286 | if (pgp_num > pg_num) { | |
7287 | *ss << "'pgp_num' must be greater than 0 and lower or equal than 'pg_num'" | |
7288 | << ", which in this case is " << pg_num; | |
7289 | return -ERANGE; | |
7290 | } | |
7291 | if (pool_type == pg_pool_t::TYPE_REPLICATED && fast_read == FAST_READ_ON) { | |
7292 | *ss << "'fast_read' can only apply to erasure coding pool"; | |
7293 | return -EINVAL; | |
7294 | } | |
7295 | int r; | |
31f18b77 FG |
7296 | r = prepare_pool_crush_rule(pool_type, erasure_code_profile, |
7297 | crush_rule_name, &crush_rule, ss); | |
7c673cae | 7298 | if (r) { |
94b18763 | 7299 | dout(10) << "prepare_pool_crush_rule returns " << r << dendl; |
7c673cae FG |
7300 | return r; |
7301 | } | |
11fdf7f2 | 7302 | if (g_conf()->mon_osd_crush_smoke_test) { |
224ce89b WB |
7303 | CrushWrapper newcrush; |
7304 | _get_pending_crush(newcrush); | |
7305 | ostringstream err; | |
7306 | CrushTester tester(newcrush, err); | |
b5b8bbf5 | 7307 | tester.set_min_x(0); |
224ce89b WB |
7308 | tester.set_max_x(50); |
7309 | tester.set_rule(crush_rule); | |
b5b8bbf5 | 7310 | auto start = ceph::coarse_mono_clock::now(); |
11fdf7f2 | 7311 | r = tester.test_with_fork(g_conf()->mon_lease); |
b5b8bbf5 | 7312 | auto duration = ceph::coarse_mono_clock::now() - start; |
224ce89b | 7313 | if (r < 0) { |
94b18763 | 7314 | dout(10) << "tester.test_with_fork returns " << r |
224ce89b WB |
7315 | << ": " << err.str() << dendl; |
7316 | *ss << "crush test failed with " << r << ": " << err.str(); | |
7317 | return r; | |
7318 | } | |
181888fb | 7319 | dout(10) << __func__ << " crush smoke test duration: " |
b5b8bbf5 | 7320 | << duration << dendl; |
7c673cae FG |
7321 | } |
7322 | unsigned size, min_size; | |
11fdf7f2 TL |
7323 | r = prepare_pool_size(pool_type, erasure_code_profile, repl_size, |
7324 | &size, &min_size, ss); | |
7c673cae | 7325 | if (r) { |
94b18763 | 7326 | dout(10) << "prepare_pool_size returns " << r << dendl; |
7c673cae FG |
7327 | return r; |
7328 | } | |
3efd9988 FG |
7329 | r = check_pg_num(-1, pg_num, size, ss); |
7330 | if (r) { | |
94b18763 | 7331 | dout(10) << "check_pg_num returns " << r << dendl; |
3efd9988 FG |
7332 | return r; |
7333 | } | |
7c673cae | 7334 | |
31f18b77 | 7335 | if (!osdmap.crush->check_crush_rule(crush_rule, pool_type, size, *ss)) { |
7c673cae FG |
7336 | return -EINVAL; |
7337 | } | |
7338 | ||
7339 | uint32_t stripe_width = 0; | |
7340 | r = prepare_pool_stripe_width(pool_type, erasure_code_profile, &stripe_width, ss); | |
7341 | if (r) { | |
94b18763 | 7342 | dout(10) << "prepare_pool_stripe_width returns " << r << dendl; |
7c673cae FG |
7343 | return r; |
7344 | } | |
7345 | ||
7346 | bool fread = false; | |
7347 | if (pool_type == pg_pool_t::TYPE_ERASURE) { | |
7348 | switch (fast_read) { | |
7349 | case FAST_READ_OFF: | |
7350 | fread = false; | |
7351 | break; | |
7352 | case FAST_READ_ON: | |
7353 | fread = true; | |
7354 | break; | |
7355 | case FAST_READ_DEFAULT: | |
11fdf7f2 | 7356 | fread = g_conf()->osd_pool_default_ec_fast_read; |
7c673cae FG |
7357 | break; |
7358 | default: | |
7359 | *ss << "invalid fast_read setting: " << fast_read; | |
7360 | return -EINVAL; | |
7361 | } | |
7362 | } | |
7363 | ||
7364 | for (map<int64_t,string>::iterator p = pending_inc.new_pool_names.begin(); | |
7365 | p != pending_inc.new_pool_names.end(); | |
7366 | ++p) { | |
7367 | if (p->second == name) | |
7368 | return 0; | |
7369 | } | |
7370 | ||
7371 | if (-1 == pending_inc.new_pool_max) | |
7372 | pending_inc.new_pool_max = osdmap.pool_max; | |
7373 | int64_t pool = ++pending_inc.new_pool_max; | |
7374 | pg_pool_t empty; | |
7375 | pg_pool_t *pi = pending_inc.get_new_pool(pool, &empty); | |
11fdf7f2 | 7376 | pi->create_time = ceph_clock_now(); |
7c673cae FG |
7377 | pi->type = pool_type; |
7378 | pi->fast_read = fread; | |
11fdf7f2 TL |
7379 | pi->flags = g_conf()->osd_pool_default_flags; |
7380 | if (g_conf()->osd_pool_default_flag_hashpspool) | |
7c673cae | 7381 | pi->set_flag(pg_pool_t::FLAG_HASHPSPOOL); |
11fdf7f2 | 7382 | if (g_conf()->osd_pool_default_flag_nodelete) |
7c673cae | 7383 | pi->set_flag(pg_pool_t::FLAG_NODELETE); |
11fdf7f2 | 7384 | if (g_conf()->osd_pool_default_flag_nopgchange) |
7c673cae | 7385 | pi->set_flag(pg_pool_t::FLAG_NOPGCHANGE); |
11fdf7f2 | 7386 | if (g_conf()->osd_pool_default_flag_nosizechange) |
7c673cae | 7387 | pi->set_flag(pg_pool_t::FLAG_NOSIZECHANGE); |
11fdf7f2 TL |
7388 | pi->set_flag(pg_pool_t::FLAG_CREATING); |
7389 | if (g_conf()->osd_pool_use_gmt_hitset) | |
7c673cae FG |
7390 | pi->use_gmt_hitset = true; |
7391 | else | |
7392 | pi->use_gmt_hitset = false; | |
7393 | ||
7394 | pi->size = size; | |
7395 | pi->min_size = min_size; | |
31f18b77 | 7396 | pi->crush_rule = crush_rule; |
7c673cae FG |
7397 | pi->expected_num_objects = expected_num_objects; |
7398 | pi->object_hash = CEPH_STR_HASH_RJENKINS; | |
11fdf7f2 TL |
7399 | |
7400 | { | |
7401 | auto m = pg_pool_t::get_pg_autoscale_mode_by_name( | |
7402 | g_conf().get_val<string>("osd_pool_default_pg_autoscale_mode")); | |
7403 | pi->pg_autoscale_mode = m >= 0 ? m : 0; | |
7404 | } | |
7405 | auto max = g_conf().get_val<int64_t>("mon_osd_max_initial_pgs"); | |
7406 | pi->set_pg_num( | |
7407 | max > 0 ? std::min<uint64_t>(pg_num, std::max<int64_t>(1, max)) | |
7408 | : pg_num); | |
7409 | pi->set_pg_num_pending(pi->get_pg_num()); | |
7410 | pi->set_pg_num_target(pg_num); | |
7411 | pi->set_pgp_num(pi->get_pg_num()); | |
7412 | pi->set_pgp_num_target(pgp_num); | |
7413 | if (osdmap.require_osd_release >= CEPH_RELEASE_NAUTILUS && | |
7414 | pg_num_min) { | |
7415 | pi->opts.set(pool_opts_t::PG_NUM_MIN, static_cast<int64_t>(pg_num_min)); | |
7416 | } | |
7417 | ||
7c673cae | 7418 | pi->last_change = pending_inc.epoch; |
11fdf7f2 TL |
7419 | pi->auid = 0; |
7420 | ||
7421 | if (pool_type == pg_pool_t::TYPE_ERASURE) { | |
7422 | pi->erasure_code_profile = erasure_code_profile; | |
7423 | } else { | |
7424 | pi->erasure_code_profile = ""; | |
7425 | } | |
7c673cae | 7426 | pi->stripe_width = stripe_width; |
11fdf7f2 TL |
7427 | |
7428 | if (osdmap.require_osd_release >= CEPH_RELEASE_NAUTILUS && | |
7429 | target_size_bytes) { | |
7430 | // only store for nautilus+ because TARGET_SIZE_BYTES may be | |
7431 | // larger than int32_t max. | |
7432 | pi->opts.set(pool_opts_t::TARGET_SIZE_BYTES, static_cast<int64_t>(target_size_bytes)); | |
7433 | } | |
7434 | if (target_size_ratio > 0.0 && | |
7435 | osdmap.require_osd_release >= CEPH_RELEASE_NAUTILUS) { | |
7436 | // only store for nautilus+, just to be consistent and tidy. | |
7437 | pi->opts.set(pool_opts_t::TARGET_SIZE_RATIO, target_size_ratio); | |
7438 | } | |
7439 | ||
7c673cae | 7440 | pi->cache_target_dirty_ratio_micro = |
11fdf7f2 | 7441 | g_conf()->osd_pool_default_cache_target_dirty_ratio * 1000000; |
7c673cae | 7442 | pi->cache_target_dirty_high_ratio_micro = |
11fdf7f2 | 7443 | g_conf()->osd_pool_default_cache_target_dirty_high_ratio * 1000000; |
7c673cae | 7444 | pi->cache_target_full_ratio_micro = |
11fdf7f2 TL |
7445 | g_conf()->osd_pool_default_cache_target_full_ratio * 1000000; |
7446 | pi->cache_min_flush_age = g_conf()->osd_pool_default_cache_min_flush_age; | |
7447 | pi->cache_min_evict_age = g_conf()->osd_pool_default_cache_min_evict_age; | |
7448 | ||
7c673cae FG |
7449 | pending_inc.new_pool_names[pool] = name; |
7450 | return 0; | |
7451 | } | |
7452 | ||
7453 | bool OSDMonitor::prepare_set_flag(MonOpRequestRef op, int flag) | |
7454 | { | |
7455 | op->mark_osdmon_event(__func__); | |
7456 | ostringstream ss; | |
7457 | if (pending_inc.new_flags < 0) | |
7458 | pending_inc.new_flags = osdmap.get_flags(); | |
7459 | pending_inc.new_flags |= flag; | |
7460 | ss << OSDMap::get_flag_string(flag) << " is set"; | |
7461 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
7462 | get_last_committed() + 1)); | |
7463 | return true; | |
7464 | } | |
7465 | ||
7466 | bool OSDMonitor::prepare_unset_flag(MonOpRequestRef op, int flag) | |
7467 | { | |
7468 | op->mark_osdmon_event(__func__); | |
7469 | ostringstream ss; | |
7470 | if (pending_inc.new_flags < 0) | |
7471 | pending_inc.new_flags = osdmap.get_flags(); | |
7472 | pending_inc.new_flags &= ~flag; | |
7473 | ss << OSDMap::get_flag_string(flag) << " is unset"; | |
7474 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
7475 | get_last_committed() + 1)); | |
7476 | return true; | |
7477 | } | |
7478 | ||
11fdf7f2 | 7479 | int OSDMonitor::prepare_command_pool_set(const cmdmap_t& cmdmap, |
7c673cae FG |
7480 | stringstream& ss) |
7481 | { | |
7482 | string poolstr; | |
11fdf7f2 | 7483 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
7484 | int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str()); |
7485 | if (pool < 0) { | |
7486 | ss << "unrecognized pool '" << poolstr << "'"; | |
7487 | return -ENOENT; | |
7488 | } | |
7489 | string var; | |
11fdf7f2 | 7490 | cmd_getval(cct, cmdmap, "var", var); |
7c673cae FG |
7491 | |
7492 | pg_pool_t p = *osdmap.get_pg_pool(pool); | |
7493 | if (pending_inc.new_pools.count(pool)) | |
7494 | p = pending_inc.new_pools[pool]; | |
7495 | ||
7496 | // accept val as a json string in the normal case (current | |
7497 | // generation monitor). parse out int or float values from the | |
7498 | // string as needed. however, if it is not a string, try to pull | |
7499 | // out an int, in case an older monitor with an older json schema is | |
7500 | // forwarding a request. | |
7501 | string val; | |
7502 | string interr, floaterr; | |
7503 | int64_t n = 0; | |
7504 | double f = 0; | |
7505 | int64_t uf = 0; // micro-f | |
11fdf7f2 | 7506 | cmd_getval(cct, cmdmap, "val", val); |
f64942e4 AA |
7507 | |
7508 | // parse string as both int and float; different fields use different types. | |
7509 | n = strict_strtoll(val.c_str(), 10, &interr); | |
7510 | f = strict_strtod(val.c_str(), &floaterr); | |
7511 | uf = llrintl(f * (double)1000000.0); | |
7c673cae FG |
7512 | |
7513 | if (!p.is_tier() && | |
7514 | (var == "hit_set_type" || var == "hit_set_period" || | |
7515 | var == "hit_set_count" || var == "hit_set_fpp" || | |
7516 | var == "target_max_objects" || var == "target_max_bytes" || | |
7517 | var == "cache_target_full_ratio" || var == "cache_target_dirty_ratio" || | |
7518 | var == "cache_target_dirty_high_ratio" || var == "use_gmt_hitset" || | |
7519 | var == "cache_min_flush_age" || var == "cache_min_evict_age" || | |
7520 | var == "hit_set_grade_decay_rate" || var == "hit_set_search_last_n" || | |
7521 | var == "min_read_recency_for_promote" || var == "min_write_recency_for_promote")) { | |
7522 | return -EACCES; | |
7523 | } | |
7524 | ||
7525 | if (var == "size") { | |
7526 | if (p.has_flag(pg_pool_t::FLAG_NOSIZECHANGE)) { | |
7527 | ss << "pool size change is disabled; you must unset nosizechange flag for the pool first"; | |
7528 | return -EPERM; | |
7529 | } | |
7530 | if (p.type == pg_pool_t::TYPE_ERASURE) { | |
7531 | ss << "can not change the size of an erasure-coded pool"; | |
7532 | return -ENOTSUP; | |
7533 | } | |
7534 | if (interr.length()) { | |
7535 | ss << "error parsing integer value '" << val << "': " << interr; | |
7536 | return -EINVAL; | |
7537 | } | |
7538 | if (n <= 0 || n > 10) { | |
7539 | ss << "pool size must be between 1 and 10"; | |
7540 | return -EINVAL; | |
7541 | } | |
eafe8130 TL |
7542 | if (!osdmap.crush->check_crush_rule(p.get_crush_rule(), p.type, n, ss)) { |
7543 | return -EINVAL; | |
7544 | } | |
3efd9988 FG |
7545 | int r = check_pg_num(pool, p.get_pg_num(), n, &ss); |
7546 | if (r < 0) { | |
7547 | return r; | |
7548 | } | |
7c673cae FG |
7549 | p.size = n; |
7550 | if (n < p.min_size) | |
7551 | p.min_size = n; | |
7552 | } else if (var == "min_size") { | |
7553 | if (p.has_flag(pg_pool_t::FLAG_NOSIZECHANGE)) { | |
7554 | ss << "pool min size change is disabled; you must unset nosizechange flag for the pool first"; | |
7555 | return -EPERM; | |
7556 | } | |
7557 | if (interr.length()) { | |
7558 | ss << "error parsing integer value '" << val << "': " << interr; | |
7559 | return -EINVAL; | |
7560 | } | |
7561 | ||
7562 | if (p.type != pg_pool_t::TYPE_ERASURE) { | |
7563 | if (n < 1 || n > p.size) { | |
494da23a | 7564 | ss << "pool min_size must be between 1 and size, which is set to " << (int)p.size; |
7c673cae FG |
7565 | return -EINVAL; |
7566 | } | |
7567 | } else { | |
7568 | ErasureCodeInterfaceRef erasure_code; | |
7569 | int k; | |
7570 | stringstream tmp; | |
7571 | int err = get_erasure_code(p.erasure_code_profile, &erasure_code, &tmp); | |
7572 | if (err == 0) { | |
7573 | k = erasure_code->get_data_chunk_count(); | |
7574 | } else { | |
b32b8144 | 7575 | ss << __func__ << " get_erasure_code failed: " << tmp.str(); |
7c673cae FG |
7576 | return err; |
7577 | } | |
7578 | ||
7579 | if (n < k || n > p.size) { | |
494da23a | 7580 | ss << "pool min_size must be between " << k << " and size, which is set to " << (int)p.size; |
7c673cae FG |
7581 | return -EINVAL; |
7582 | } | |
7583 | } | |
7584 | p.min_size = n; | |
11fdf7f2 | 7585 | } else if (var == "pg_num_actual") { |
7c673cae FG |
7586 | if (interr.length()) { |
7587 | ss << "error parsing integer value '" << val << "': " << interr; | |
7588 | return -EINVAL; | |
7589 | } | |
11fdf7f2 TL |
7590 | if (n == (int)p.get_pg_num()) { |
7591 | return 0; | |
7592 | } | |
7593 | if (static_cast<uint64_t>(n) > g_conf().get_val<uint64_t>("mon_max_pool_pg_num")) { | |
7594 | ss << "'pg_num' must be greater than 0 and less than or equal to " | |
7595 | << g_conf().get_val<uint64_t>("mon_max_pool_pg_num") | |
7596 | << " (you may adjust 'mon max pool pg num' for higher values)"; | |
7597 | return -ERANGE; | |
7598 | } | |
7599 | if (p.has_flag(pg_pool_t::FLAG_CREATING)) { | |
7600 | ss << "cannot adjust pg_num while initial PGs are being created"; | |
7601 | return -EBUSY; | |
7602 | } | |
7603 | if (n > (int)p.get_pg_num()) { | |
7604 | if (p.get_pg_num() != p.get_pg_num_pending()) { | |
7605 | // force pre-nautilus clients to resend their ops, since they | |
7606 | // don't understand pg_num_pending changes form a new interval | |
7607 | p.last_force_op_resend_prenautilus = pending_inc.epoch; | |
7608 | } | |
7609 | p.set_pg_num(n); | |
7610 | } else { | |
7611 | if (osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS) { | |
7612 | ss << "nautilus OSDs are required to adjust pg_num_pending"; | |
7613 | return -EPERM; | |
7614 | } | |
7615 | if (n < (int)p.get_pgp_num()) { | |
7616 | ss << "specified pg_num " << n << " < pgp_num " << p.get_pgp_num(); | |
7617 | return -EINVAL; | |
7618 | } | |
7619 | if (n < (int)p.get_pg_num() - 1) { | |
7620 | ss << "specified pg_num " << n << " < pg_num (" << p.get_pg_num() | |
7621 | << ") - 1; only single pg decrease is currently supported"; | |
7622 | return -EINVAL; | |
7623 | } | |
7624 | p.set_pg_num_pending(n); | |
7625 | // force pre-nautilus clients to resend their ops, since they | |
7626 | // don't understand pg_num_pending changes form a new interval | |
7627 | p.last_force_op_resend_prenautilus = pending_inc.epoch; | |
7c673cae | 7628 | } |
11fdf7f2 TL |
7629 | // force pre-luminous clients to resend their ops, since they |
7630 | // don't understand that split PGs now form a new interval. | |
7631 | p.last_force_op_resend_preluminous = pending_inc.epoch; | |
7c673cae FG |
7632 | } else if (var == "pg_num") { |
7633 | if (p.has_flag(pg_pool_t::FLAG_NOPGCHANGE)) { | |
7634 | ss << "pool pg_num change is disabled; you must unset nopgchange flag for the pool first"; | |
7635 | return -EPERM; | |
7636 | } | |
7637 | if (interr.length()) { | |
7638 | ss << "error parsing integer value '" << val << "': " << interr; | |
7639 | return -EINVAL; | |
7640 | } | |
11fdf7f2 | 7641 | if (n == (int)p.get_pg_num_target()) { |
7c673cae FG |
7642 | return 0; |
7643 | } | |
11fdf7f2 TL |
7644 | if (n <= 0 || static_cast<uint64_t>(n) > |
7645 | g_conf().get_val<uint64_t>("mon_max_pool_pg_num")) { | |
c07f9fc5 | 7646 | ss << "'pg_num' must be greater than 0 and less than or equal to " |
11fdf7f2 | 7647 | << g_conf().get_val<uint64_t>("mon_max_pool_pg_num") |
c07f9fc5 FG |
7648 | << " (you may adjust 'mon max pool pg num' for higher values)"; |
7649 | return -ERANGE; | |
7650 | } | |
11fdf7f2 TL |
7651 | if (n > (int)p.get_pg_num_target()) { |
7652 | int r = check_pg_num(pool, n, p.get_size(), &ss); | |
7653 | if (r) { | |
7654 | return r; | |
7655 | } | |
7656 | bool force = false; | |
7657 | cmd_getval(cct,cmdmap, "yes_i_really_mean_it", force); | |
7658 | if (p.cache_mode != pg_pool_t::CACHEMODE_NONE && !force) { | |
7659 | ss << "splits in cache pools must be followed by scrubs and leave sufficient free space to avoid overfilling. use --yes-i-really-mean-it to force."; | |
7660 | return -EPERM; | |
7661 | } | |
7662 | } else { | |
7663 | if (osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS) { | |
7664 | ss << "nautilus OSDs are required to decrease pg_num"; | |
7665 | return -EPERM; | |
7666 | } | |
7c673cae | 7667 | } |
494da23a TL |
7668 | if (osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS) { |
7669 | // pre-nautilus osdmap format; increase pg_num directly | |
7670 | assert(n > (int)p.get_pg_num()); | |
7671 | // force pre-nautilus clients to resend their ops, since they | |
7672 | // don't understand pg_num_target changes form a new interval | |
7673 | p.last_force_op_resend_prenautilus = pending_inc.epoch; | |
7674 | // force pre-luminous clients to resend their ops, since they | |
7675 | // don't understand that split PGs now form a new interval. | |
7676 | p.last_force_op_resend_preluminous = pending_inc.epoch; | |
7677 | p.set_pg_num(n); | |
7678 | } else { | |
7679 | // set targets; mgr will adjust pg_num_actual and pgp_num later. | |
7680 | // make pgp_num track pg_num if it already matches. if it is set | |
7681 | // differently, leave it different and let the user control it | |
7682 | // manually. | |
7683 | if (p.get_pg_num_target() == p.get_pgp_num_target()) { | |
7684 | p.set_pgp_num_target(n); | |
7685 | } | |
7686 | p.set_pg_num_target(n); | |
7c673cae | 7687 | } |
11fdf7f2 | 7688 | } else if (var == "pgp_num_actual") { |
7c673cae FG |
7689 | if (p.has_flag(pg_pool_t::FLAG_NOPGCHANGE)) { |
7690 | ss << "pool pgp_num change is disabled; you must unset nopgchange flag for the pool first"; | |
7691 | return -EPERM; | |
7692 | } | |
7693 | if (interr.length()) { | |
7694 | ss << "error parsing integer value '" << val << "': " << interr; | |
7695 | return -EINVAL; | |
7696 | } | |
7697 | if (n <= 0) { | |
7698 | ss << "specified pgp_num must > 0, but you set to " << n; | |
7699 | return -EINVAL; | |
7700 | } | |
7701 | if (n > (int)p.get_pg_num()) { | |
7702 | ss << "specified pgp_num " << n << " > pg_num " << p.get_pg_num(); | |
7703 | return -EINVAL; | |
7704 | } | |
11fdf7f2 TL |
7705 | if (n > (int)p.get_pg_num_pending()) { |
7706 | ss << "specified pgp_num " << n | |
7707 | << " > pg_num_pending " << p.get_pg_num_pending(); | |
7708 | return -EINVAL; | |
7709 | } | |
7c673cae | 7710 | p.set_pgp_num(n); |
11fdf7f2 TL |
7711 | } else if (var == "pgp_num") { |
7712 | if (p.has_flag(pg_pool_t::FLAG_NOPGCHANGE)) { | |
7713 | ss << "pool pgp_num change is disabled; you must unset nopgchange flag for the pool first"; | |
7714 | return -EPERM; | |
7715 | } | |
7716 | if (interr.length()) { | |
7717 | ss << "error parsing integer value '" << val << "': " << interr; | |
7718 | return -EINVAL; | |
7719 | } | |
7720 | if (n <= 0) { | |
7721 | ss << "specified pgp_num must > 0, but you set to " << n; | |
7722 | return -EINVAL; | |
7723 | } | |
7724 | if (n > (int)p.get_pg_num_target()) { | |
7725 | ss << "specified pgp_num " << n << " > pg_num " << p.get_pg_num_target(); | |
7726 | return -EINVAL; | |
7727 | } | |
494da23a TL |
7728 | if (osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS) { |
7729 | // pre-nautilus osdmap format; increase pgp_num directly | |
7730 | p.set_pgp_num(n); | |
7731 | } else { | |
7732 | p.set_pgp_num_target(n); | |
7733 | } | |
11fdf7f2 TL |
7734 | } else if (var == "pg_autoscale_mode") { |
7735 | n = pg_pool_t::get_pg_autoscale_mode_by_name(val); | |
7736 | if (n < 0) { | |
7737 | ss << "specified invalid mode " << val; | |
7738 | return -EINVAL; | |
7739 | } | |
494da23a TL |
7740 | if (osdmap.require_osd_release < CEPH_RELEASE_NAUTILUS) { |
7741 | ss << "must set require_osd_release to nautilus or later before setting pg_autoscale_mode"; | |
7742 | return -EINVAL; | |
7743 | } | |
11fdf7f2 | 7744 | p.pg_autoscale_mode = n; |
7c673cae FG |
7745 | } else if (var == "crush_rule") { |
7746 | int id = osdmap.crush->get_rule_id(val); | |
7747 | if (id == -ENOENT) { | |
7748 | ss << "crush rule " << val << " does not exist"; | |
7749 | return -ENOENT; | |
7750 | } | |
7751 | if (id < 0) { | |
7752 | ss << cpp_strerror(id); | |
7753 | return -ENOENT; | |
7754 | } | |
7755 | if (!osdmap.crush->check_crush_rule(id, p.get_type(), p.get_size(), ss)) { | |
7756 | return -EINVAL; | |
7757 | } | |
31f18b77 | 7758 | p.crush_rule = id; |
7c673cae FG |
7759 | } else if (var == "nodelete" || var == "nopgchange" || |
7760 | var == "nosizechange" || var == "write_fadvise_dontneed" || | |
7761 | var == "noscrub" || var == "nodeep-scrub") { | |
7762 | uint64_t flag = pg_pool_t::get_flag_by_name(var); | |
7763 | // make sure we only compare against 'n' if we didn't receive a string | |
7764 | if (val == "true" || (interr.empty() && n == 1)) { | |
7765 | p.set_flag(flag); | |
7766 | } else if (val == "false" || (interr.empty() && n == 0)) { | |
7767 | p.unset_flag(flag); | |
7768 | } else { | |
7769 | ss << "expecting value 'true', 'false', '0', or '1'"; | |
7770 | return -EINVAL; | |
7771 | } | |
7772 | } else if (var == "hashpspool") { | |
7773 | uint64_t flag = pg_pool_t::get_flag_by_name(var); | |
11fdf7f2 TL |
7774 | bool force = false; |
7775 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", force); | |
7776 | ||
7777 | if (!force) { | |
7c673cae FG |
7778 | ss << "are you SURE? this will remap all placement groups in this pool," |
7779 | " this triggers large data movement," | |
7780 | " pass --yes-i-really-mean-it if you really do."; | |
7781 | return -EPERM; | |
7782 | } | |
7783 | // make sure we only compare against 'n' if we didn't receive a string | |
7784 | if (val == "true" || (interr.empty() && n == 1)) { | |
7785 | p.set_flag(flag); | |
7786 | } else if (val == "false" || (interr.empty() && n == 0)) { | |
7787 | p.unset_flag(flag); | |
7788 | } else { | |
7789 | ss << "expecting value 'true', 'false', '0', or '1'"; | |
7790 | return -EINVAL; | |
7791 | } | |
7792 | } else if (var == "hit_set_type") { | |
7793 | if (val == "none") | |
7794 | p.hit_set_params = HitSet::Params(); | |
7795 | else { | |
7796 | int err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss); | |
7797 | if (err) | |
7798 | return err; | |
7799 | if (val == "bloom") { | |
7800 | BloomHitSet::Params *bsp = new BloomHitSet::Params; | |
11fdf7f2 | 7801 | bsp->set_fpp(g_conf().get_val<double>("osd_pool_default_hit_set_bloom_fpp")); |
7c673cae FG |
7802 | p.hit_set_params = HitSet::Params(bsp); |
7803 | } else if (val == "explicit_hash") | |
7804 | p.hit_set_params = HitSet::Params(new ExplicitHashHitSet::Params); | |
7805 | else if (val == "explicit_object") | |
7806 | p.hit_set_params = HitSet::Params(new ExplicitObjectHitSet::Params); | |
7807 | else { | |
7808 | ss << "unrecognized hit_set type '" << val << "'"; | |
7809 | return -EINVAL; | |
7810 | } | |
7811 | } | |
7812 | } else if (var == "hit_set_period") { | |
7813 | if (interr.length()) { | |
7814 | ss << "error parsing integer value '" << val << "': " << interr; | |
7815 | return -EINVAL; | |
11fdf7f2 TL |
7816 | } else if (n < 0) { |
7817 | ss << "hit_set_period should be non-negative"; | |
7818 | return -EINVAL; | |
7c673cae FG |
7819 | } |
7820 | p.hit_set_period = n; | |
7821 | } else if (var == "hit_set_count") { | |
7822 | if (interr.length()) { | |
7823 | ss << "error parsing integer value '" << val << "': " << interr; | |
7824 | return -EINVAL; | |
11fdf7f2 TL |
7825 | } else if (n < 0) { |
7826 | ss << "hit_set_count should be non-negative"; | |
7827 | return -EINVAL; | |
7c673cae FG |
7828 | } |
7829 | p.hit_set_count = n; | |
7830 | } else if (var == "hit_set_fpp") { | |
7831 | if (floaterr.length()) { | |
7832 | ss << "error parsing floating point value '" << val << "': " << floaterr; | |
7833 | return -EINVAL; | |
11fdf7f2 TL |
7834 | } else if (f < 0 || f > 1.0) { |
7835 | ss << "hit_set_fpp should be in the range 0..1"; | |
7836 | return -EINVAL; | |
7c673cae FG |
7837 | } |
7838 | if (p.hit_set_params.get_type() != HitSet::TYPE_BLOOM) { | |
7839 | ss << "hit set is not of type Bloom; invalid to set a false positive rate!"; | |
7840 | return -EINVAL; | |
7841 | } | |
7842 | BloomHitSet::Params *bloomp = static_cast<BloomHitSet::Params*>(p.hit_set_params.impl.get()); | |
7843 | bloomp->set_fpp(f); | |
7844 | } else if (var == "use_gmt_hitset") { | |
7845 | if (val == "true" || (interr.empty() && n == 1)) { | |
7c673cae FG |
7846 | p.use_gmt_hitset = true; |
7847 | } else { | |
7848 | ss << "expecting value 'true' or '1'"; | |
7849 | return -EINVAL; | |
7850 | } | |
7851 | } else if (var == "allow_ec_overwrites") { | |
7852 | if (!p.is_erasure()) { | |
7853 | ss << "ec overwrites can only be enabled for an erasure coded pool"; | |
7854 | return -EINVAL; | |
7855 | } | |
224ce89b | 7856 | stringstream err; |
11fdf7f2 | 7857 | if (!g_conf()->mon_debug_no_require_bluestore_for_ec_overwrites && |
224ce89b WB |
7858 | !is_pool_currently_all_bluestore(pool, p, &err)) { |
7859 | ss << "pool must only be stored on bluestore for scrubbing to work: " << err.str(); | |
7860 | return -EINVAL; | |
7861 | } | |
7c673cae FG |
7862 | if (val == "true" || (interr.empty() && n == 1)) { |
7863 | p.flags |= pg_pool_t::FLAG_EC_OVERWRITES; | |
7864 | } else if (val == "false" || (interr.empty() && n == 0)) { | |
7865 | ss << "ec overwrites cannot be disabled once enabled"; | |
7866 | return -EINVAL; | |
7867 | } else { | |
7868 | ss << "expecting value 'true', 'false', '0', or '1'"; | |
7869 | return -EINVAL; | |
7870 | } | |
7c673cae FG |
7871 | } else if (var == "target_max_objects") { |
7872 | if (interr.length()) { | |
7873 | ss << "error parsing int '" << val << "': " << interr; | |
7874 | return -EINVAL; | |
7875 | } | |
7876 | p.target_max_objects = n; | |
7877 | } else if (var == "target_max_bytes") { | |
7878 | if (interr.length()) { | |
7879 | ss << "error parsing int '" << val << "': " << interr; | |
7880 | return -EINVAL; | |
7881 | } | |
7882 | p.target_max_bytes = n; | |
7883 | } else if (var == "cache_target_dirty_ratio") { | |
7884 | if (floaterr.length()) { | |
7885 | ss << "error parsing float '" << val << "': " << floaterr; | |
7886 | return -EINVAL; | |
7887 | } | |
7888 | if (f < 0 || f > 1.0) { | |
7889 | ss << "value must be in the range 0..1"; | |
7890 | return -ERANGE; | |
7891 | } | |
7892 | p.cache_target_dirty_ratio_micro = uf; | |
7893 | } else if (var == "cache_target_dirty_high_ratio") { | |
7894 | if (floaterr.length()) { | |
7895 | ss << "error parsing float '" << val << "': " << floaterr; | |
7896 | return -EINVAL; | |
7897 | } | |
7898 | if (f < 0 || f > 1.0) { | |
7899 | ss << "value must be in the range 0..1"; | |
7900 | return -ERANGE; | |
7901 | } | |
7902 | p.cache_target_dirty_high_ratio_micro = uf; | |
7903 | } else if (var == "cache_target_full_ratio") { | |
7904 | if (floaterr.length()) { | |
7905 | ss << "error parsing float '" << val << "': " << floaterr; | |
7906 | return -EINVAL; | |
7907 | } | |
7908 | if (f < 0 || f > 1.0) { | |
7909 | ss << "value must be in the range 0..1"; | |
7910 | return -ERANGE; | |
7911 | } | |
7912 | p.cache_target_full_ratio_micro = uf; | |
7913 | } else if (var == "cache_min_flush_age") { | |
7914 | if (interr.length()) { | |
7915 | ss << "error parsing int '" << val << "': " << interr; | |
7916 | return -EINVAL; | |
7917 | } | |
7918 | p.cache_min_flush_age = n; | |
7919 | } else if (var == "cache_min_evict_age") { | |
7920 | if (interr.length()) { | |
7921 | ss << "error parsing int '" << val << "': " << interr; | |
7922 | return -EINVAL; | |
7923 | } | |
7924 | p.cache_min_evict_age = n; | |
7925 | } else if (var == "min_read_recency_for_promote") { | |
7926 | if (interr.length()) { | |
7927 | ss << "error parsing integer value '" << val << "': " << interr; | |
7928 | return -EINVAL; | |
7929 | } | |
7930 | p.min_read_recency_for_promote = n; | |
7931 | } else if (var == "hit_set_grade_decay_rate") { | |
7932 | if (interr.length()) { | |
7933 | ss << "error parsing integer value '" << val << "': " << interr; | |
7934 | return -EINVAL; | |
7935 | } | |
7936 | if (n > 100 || n < 0) { | |
7937 | ss << "value out of range,valid range is 0 - 100"; | |
7938 | return -EINVAL; | |
7939 | } | |
7940 | p.hit_set_grade_decay_rate = n; | |
7941 | } else if (var == "hit_set_search_last_n") { | |
7942 | if (interr.length()) { | |
7943 | ss << "error parsing integer value '" << val << "': " << interr; | |
7944 | return -EINVAL; | |
7945 | } | |
7946 | if (n > p.hit_set_count || n < 0) { | |
7947 | ss << "value out of range,valid range is 0 - hit_set_count"; | |
7948 | return -EINVAL; | |
7949 | } | |
7950 | p.hit_set_search_last_n = n; | |
7951 | } else if (var == "min_write_recency_for_promote") { | |
7952 | if (interr.length()) { | |
7953 | ss << "error parsing integer value '" << val << "': " << interr; | |
7954 | return -EINVAL; | |
7955 | } | |
7956 | p.min_write_recency_for_promote = n; | |
7957 | } else if (var == "fast_read") { | |
7958 | if (p.is_replicated()) { | |
7959 | ss << "fast read is not supported in replication pool"; | |
7960 | return -EINVAL; | |
7961 | } | |
7962 | if (val == "true" || (interr.empty() && n == 1)) { | |
7963 | p.fast_read = true; | |
7964 | } else if (val == "false" || (interr.empty() && n == 0)) { | |
7965 | p.fast_read = false; | |
7966 | } else { | |
7967 | ss << "expecting value 'true', 'false', '0', or '1'"; | |
7968 | return -EINVAL; | |
7969 | } | |
7970 | } else if (pool_opts_t::is_opt_name(var)) { | |
224ce89b | 7971 | bool unset = val == "unset"; |
7c673cae | 7972 | if (var == "compression_mode") { |
224ce89b WB |
7973 | if (!unset) { |
7974 | auto cmode = Compressor::get_comp_mode_type(val); | |
7975 | if (!cmode) { | |
7976 | ss << "unrecognized compression mode '" << val << "'"; | |
7977 | return -EINVAL; | |
7978 | } | |
7c673cae FG |
7979 | } |
7980 | } else if (var == "compression_algorithm") { | |
224ce89b WB |
7981 | if (!unset) { |
7982 | auto alg = Compressor::get_comp_alg_type(val); | |
7983 | if (!alg) { | |
7984 | ss << "unrecognized compression_algorithm '" << val << "'"; | |
7985 | return -EINVAL; | |
7986 | } | |
7c673cae FG |
7987 | } |
7988 | } else if (var == "compression_required_ratio") { | |
7989 | if (floaterr.length()) { | |
7990 | ss << "error parsing float value '" << val << "': " << floaterr; | |
7991 | return -EINVAL; | |
7992 | } | |
224ce89b | 7993 | if (f < 0 || f > 1) { |
7c673cae | 7994 | ss << "compression_required_ratio is out of range (0-1): '" << val << "'"; |
224ce89b | 7995 | return -EINVAL; |
7c673cae FG |
7996 | } |
7997 | } else if (var == "csum_type") { | |
224ce89b | 7998 | auto t = unset ? 0 : Checksummer::get_csum_string_type(val); |
7c673cae FG |
7999 | if (t < 0 ) { |
8000 | ss << "unrecognized csum_type '" << val << "'"; | |
224ce89b | 8001 | return -EINVAL; |
7c673cae FG |
8002 | } |
8003 | //preserve csum_type numeric value | |
8004 | n = t; | |
8005 | interr.clear(); | |
8006 | } else if (var == "compression_max_blob_size" || | |
8007 | var == "compression_min_blob_size" || | |
8008 | var == "csum_max_block" || | |
8009 | var == "csum_min_block") { | |
8010 | if (interr.length()) { | |
8011 | ss << "error parsing int value '" << val << "': " << interr; | |
8012 | return -EINVAL; | |
8013 | } | |
11fdf7f2 TL |
8014 | } else if (var == "fingerprint_algorithm") { |
8015 | if (!unset) { | |
8016 | auto alg = pg_pool_t::get_fingerprint_from_str(val); | |
8017 | if (!alg) { | |
8018 | ss << "unrecognized fingerprint_algorithm '" << val << "'"; | |
8019 | return -EINVAL; | |
8020 | } | |
8021 | } | |
8022 | } else if (var == "pg_num_min") { | |
8023 | if (interr.length()) { | |
8024 | ss << "error parsing int value '" << val << "': " << interr; | |
8025 | return -EINVAL; | |
8026 | } | |
8027 | if (n > (int)p.get_pg_num_target()) { | |
8028 | ss << "specified pg_num_min " << n | |
8029 | << " > pg_num " << p.get_pg_num_target(); | |
8030 | return -EINVAL; | |
8031 | } | |
8032 | } else if (var == "recovery_priority") { | |
8033 | if (interr.length()) { | |
8034 | ss << "error parsing int value '" << val << "': " << interr; | |
8035 | return -EINVAL; | |
8036 | } | |
81eedcae TL |
8037 | if (!g_conf()->debug_allow_any_pool_priority) { |
8038 | if (n > OSD_POOL_PRIORITY_MAX || n < OSD_POOL_PRIORITY_MIN) { | |
8039 | ss << "pool recovery_priority must be between " << OSD_POOL_PRIORITY_MIN | |
8040 | << " and " << OSD_POOL_PRIORITY_MAX; | |
8041 | return -EINVAL; | |
8042 | } | |
11fdf7f2 TL |
8043 | } |
8044 | } else if (var == "pg_autoscale_bias") { | |
8045 | if (f < 0.0 || f > 1000.0) { | |
8046 | ss << "pg_autoscale_bias must be between 0 and 1000"; | |
8047 | return -EINVAL; | |
8048 | } | |
7c673cae FG |
8049 | } |
8050 | ||
8051 | pool_opts_t::opt_desc_t desc = pool_opts_t::get_opt_desc(var); | |
8052 | switch (desc.type) { | |
8053 | case pool_opts_t::STR: | |
224ce89b | 8054 | if (unset) { |
7c673cae FG |
8055 | p.opts.unset(desc.key); |
8056 | } else { | |
8057 | p.opts.set(desc.key, static_cast<std::string>(val)); | |
8058 | } | |
8059 | break; | |
8060 | case pool_opts_t::INT: | |
8061 | if (interr.length()) { | |
8062 | ss << "error parsing integer value '" << val << "': " << interr; | |
8063 | return -EINVAL; | |
8064 | } | |
8065 | if (n == 0) { | |
8066 | p.opts.unset(desc.key); | |
8067 | } else { | |
11fdf7f2 | 8068 | p.opts.set(desc.key, static_cast<int64_t>(n)); |
7c673cae FG |
8069 | } |
8070 | break; | |
8071 | case pool_opts_t::DOUBLE: | |
8072 | if (floaterr.length()) { | |
8073 | ss << "error parsing floating point value '" << val << "': " << floaterr; | |
8074 | return -EINVAL; | |
8075 | } | |
8076 | if (f == 0) { | |
8077 | p.opts.unset(desc.key); | |
8078 | } else { | |
8079 | p.opts.set(desc.key, static_cast<double>(f)); | |
8080 | } | |
8081 | break; | |
8082 | default: | |
11fdf7f2 | 8083 | ceph_assert(!"unknown type"); |
7c673cae FG |
8084 | } |
8085 | } else { | |
8086 | ss << "unrecognized variable '" << var << "'"; | |
8087 | return -EINVAL; | |
8088 | } | |
224ce89b WB |
8089 | if (val != "unset") { |
8090 | ss << "set pool " << pool << " " << var << " to " << val; | |
8091 | } else { | |
8092 | ss << "unset pool " << pool << " " << var; | |
8093 | } | |
7c673cae FG |
8094 | p.last_change = pending_inc.epoch; |
8095 | pending_inc.new_pools[pool] = p; | |
8096 | return 0; | |
8097 | } | |
8098 | ||
c07f9fc5 | 8099 | int OSDMonitor::prepare_command_pool_application(const string &prefix, |
11fdf7f2 | 8100 | const cmdmap_t& cmdmap, |
c07f9fc5 | 8101 | stringstream& ss) |
11fdf7f2 TL |
8102 | { |
8103 | return _command_pool_application(prefix, cmdmap, ss, nullptr, true); | |
8104 | } | |
8105 | ||
8106 | int OSDMonitor::preprocess_command_pool_application(const string &prefix, | |
8107 | const cmdmap_t& cmdmap, | |
8108 | stringstream& ss, | |
8109 | bool *modified) | |
8110 | { | |
8111 | return _command_pool_application(prefix, cmdmap, ss, modified, false); | |
8112 | } | |
8113 | ||
8114 | ||
8115 | /** | |
8116 | * Common logic for preprocess and prepare phases of pool application | |
8117 | * tag commands. In preprocess mode we're only detecting invalid | |
8118 | * commands, and determining whether it was a modification or a no-op. | |
8119 | * In prepare mode we're actually updating the pending state. | |
8120 | */ | |
8121 | int OSDMonitor::_command_pool_application(const string &prefix, | |
8122 | const cmdmap_t& cmdmap, | |
8123 | stringstream& ss, | |
8124 | bool *modified, | |
8125 | bool preparing) | |
c07f9fc5 FG |
8126 | { |
8127 | string pool_name; | |
11fdf7f2 | 8128 | cmd_getval(cct, cmdmap, "pool", pool_name); |
c07f9fc5 FG |
8129 | int64_t pool = osdmap.lookup_pg_pool_name(pool_name.c_str()); |
8130 | if (pool < 0) { | |
8131 | ss << "unrecognized pool '" << pool_name << "'"; | |
8132 | return -ENOENT; | |
8133 | } | |
8134 | ||
8135 | pg_pool_t p = *osdmap.get_pg_pool(pool); | |
11fdf7f2 TL |
8136 | if (preparing) { |
8137 | if (pending_inc.new_pools.count(pool)) { | |
8138 | p = pending_inc.new_pools[pool]; | |
8139 | } | |
c07f9fc5 FG |
8140 | } |
8141 | ||
8142 | string app; | |
11fdf7f2 | 8143 | cmd_getval(cct, cmdmap, "app", app); |
c07f9fc5 FG |
8144 | bool app_exists = (p.application_metadata.count(app) > 0); |
8145 | ||
11fdf7f2 TL |
8146 | string key; |
8147 | cmd_getval(cct, cmdmap, "key", key); | |
8148 | if (key == "all") { | |
8149 | ss << "key cannot be 'all'"; | |
8150 | return -EINVAL; | |
8151 | } | |
8152 | ||
8153 | string value; | |
8154 | cmd_getval(cct, cmdmap, "value", value); | |
8155 | if (value == "all") { | |
8156 | ss << "value cannot be 'all'"; | |
8157 | return -EINVAL; | |
8158 | } | |
8159 | ||
c07f9fc5 FG |
8160 | if (boost::algorithm::ends_with(prefix, "enable")) { |
8161 | if (app.empty()) { | |
8162 | ss << "application name must be provided"; | |
8163 | return -EINVAL; | |
8164 | } | |
8165 | ||
8166 | if (p.is_tier()) { | |
8167 | ss << "application must be enabled on base tier"; | |
8168 | return -EINVAL; | |
8169 | } | |
8170 | ||
11fdf7f2 TL |
8171 | bool force = false; |
8172 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", force); | |
c07f9fc5 | 8173 | |
11fdf7f2 | 8174 | if (!app_exists && !p.application_metadata.empty() && !force) { |
c07f9fc5 FG |
8175 | ss << "Are you SURE? Pool '" << pool_name << "' already has an enabled " |
8176 | << "application; pass --yes-i-really-mean-it to proceed anyway"; | |
8177 | return -EPERM; | |
8178 | } | |
8179 | ||
8180 | if (!app_exists && p.application_metadata.size() >= MAX_POOL_APPLICATIONS) { | |
8181 | ss << "too many enabled applications on pool '" << pool_name << "'; " | |
8182 | << "max " << MAX_POOL_APPLICATIONS; | |
8183 | return -EINVAL; | |
8184 | } | |
8185 | ||
8186 | if (app.length() > MAX_POOL_APPLICATION_LENGTH) { | |
8187 | ss << "application name '" << app << "' too long; max length " | |
8188 | << MAX_POOL_APPLICATION_LENGTH; | |
8189 | return -EINVAL; | |
8190 | } | |
8191 | ||
8192 | if (!app_exists) { | |
8193 | p.application_metadata[app] = {}; | |
8194 | } | |
8195 | ss << "enabled application '" << app << "' on pool '" << pool_name << "'"; | |
8196 | ||
8197 | } else if (boost::algorithm::ends_with(prefix, "disable")) { | |
11fdf7f2 TL |
8198 | bool force = false; |
8199 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", force); | |
c07f9fc5 | 8200 | |
11fdf7f2 | 8201 | if (!force) { |
c07f9fc5 FG |
8202 | ss << "Are you SURE? Disabling an application within a pool might result " |
8203 | << "in loss of application functionality; pass " | |
8204 | << "--yes-i-really-mean-it to proceed anyway"; | |
8205 | return -EPERM; | |
8206 | } | |
8207 | ||
8208 | if (!app_exists) { | |
8209 | ss << "application '" << app << "' is not enabled on pool '" << pool_name | |
8210 | << "'"; | |
8211 | return 0; // idempotent | |
8212 | } | |
8213 | ||
8214 | p.application_metadata.erase(app); | |
8215 | ss << "disable application '" << app << "' on pool '" << pool_name << "'"; | |
8216 | ||
8217 | } else if (boost::algorithm::ends_with(prefix, "set")) { | |
8218 | if (p.is_tier()) { | |
8219 | ss << "application metadata must be set on base tier"; | |
8220 | return -EINVAL; | |
8221 | } | |
8222 | ||
8223 | if (!app_exists) { | |
8224 | ss << "application '" << app << "' is not enabled on pool '" << pool_name | |
8225 | << "'"; | |
8226 | return -ENOENT; | |
8227 | } | |
8228 | ||
8229 | string key; | |
11fdf7f2 | 8230 | cmd_getval(cct, cmdmap, "key", key); |
c07f9fc5 FG |
8231 | |
8232 | if (key.empty()) { | |
8233 | ss << "key must be provided"; | |
8234 | return -EINVAL; | |
8235 | } | |
8236 | ||
8237 | auto &app_keys = p.application_metadata[app]; | |
8238 | if (app_keys.count(key) == 0 && | |
8239 | app_keys.size() >= MAX_POOL_APPLICATION_KEYS) { | |
8240 | ss << "too many keys set for application '" << app << "' on pool '" | |
8241 | << pool_name << "'; max " << MAX_POOL_APPLICATION_KEYS; | |
8242 | return -EINVAL; | |
8243 | } | |
8244 | ||
8245 | if (key.length() > MAX_POOL_APPLICATION_LENGTH) { | |
8246 | ss << "key '" << app << "' too long; max length " | |
8247 | << MAX_POOL_APPLICATION_LENGTH; | |
8248 | return -EINVAL; | |
8249 | } | |
8250 | ||
8251 | string value; | |
11fdf7f2 | 8252 | cmd_getval(cct, cmdmap, "value", value); |
c07f9fc5 FG |
8253 | if (value.length() > MAX_POOL_APPLICATION_LENGTH) { |
8254 | ss << "value '" << value << "' too long; max length " | |
8255 | << MAX_POOL_APPLICATION_LENGTH; | |
8256 | return -EINVAL; | |
8257 | } | |
8258 | ||
8259 | p.application_metadata[app][key] = value; | |
8260 | ss << "set application '" << app << "' key '" << key << "' to '" | |
8261 | << value << "' on pool '" << pool_name << "'"; | |
8262 | } else if (boost::algorithm::ends_with(prefix, "rm")) { | |
8263 | if (!app_exists) { | |
8264 | ss << "application '" << app << "' is not enabled on pool '" << pool_name | |
8265 | << "'"; | |
8266 | return -ENOENT; | |
8267 | } | |
8268 | ||
8269 | string key; | |
11fdf7f2 | 8270 | cmd_getval(cct, cmdmap, "key", key); |
c07f9fc5 FG |
8271 | auto it = p.application_metadata[app].find(key); |
8272 | if (it == p.application_metadata[app].end()) { | |
8273 | ss << "application '" << app << "' on pool '" << pool_name | |
8274 | << "' does not have key '" << key << "'"; | |
8275 | return 0; // idempotent | |
8276 | } | |
8277 | ||
8278 | p.application_metadata[app].erase(it); | |
8279 | ss << "removed application '" << app << "' key '" << key << "' on pool '" | |
8280 | << pool_name << "'"; | |
8281 | } else { | |
11fdf7f2 TL |
8282 | ceph_abort(); |
8283 | } | |
8284 | ||
8285 | if (preparing) { | |
8286 | p.last_change = pending_inc.epoch; | |
8287 | pending_inc.new_pools[pool] = p; | |
8288 | } | |
8289 | ||
8290 | // Because we fell through this far, we didn't hit no-op cases, | |
8291 | // so pool was definitely modified | |
8292 | if (modified != nullptr) { | |
8293 | *modified = true; | |
c07f9fc5 FG |
8294 | } |
8295 | ||
c07f9fc5 FG |
8296 | return 0; |
8297 | } | |
8298 | ||
31f18b77 FG |
8299 | int OSDMonitor::_prepare_command_osd_crush_remove( |
8300 | CrushWrapper &newcrush, | |
8301 | int32_t id, | |
8302 | int32_t ancestor, | |
8303 | bool has_ancestor, | |
8304 | bool unlink_only) | |
8305 | { | |
8306 | int err = 0; | |
8307 | ||
8308 | if (has_ancestor) { | |
11fdf7f2 | 8309 | err = newcrush.remove_item_under(cct, id, ancestor, |
31f18b77 FG |
8310 | unlink_only); |
8311 | } else { | |
11fdf7f2 | 8312 | err = newcrush.remove_item(cct, id, unlink_only); |
31f18b77 FG |
8313 | } |
8314 | return err; | |
8315 | } | |
8316 | ||
8317 | void OSDMonitor::do_osd_crush_remove(CrushWrapper& newcrush) | |
8318 | { | |
8319 | pending_inc.crush.clear(); | |
8320 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
8321 | } | |
8322 | ||
8323 | int OSDMonitor::prepare_command_osd_crush_remove( | |
8324 | CrushWrapper &newcrush, | |
8325 | int32_t id, | |
8326 | int32_t ancestor, | |
8327 | bool has_ancestor, | |
8328 | bool unlink_only) | |
8329 | { | |
8330 | int err = _prepare_command_osd_crush_remove( | |
8331 | newcrush, id, ancestor, | |
8332 | has_ancestor, unlink_only); | |
8333 | ||
8334 | if (err < 0) | |
8335 | return err; | |
8336 | ||
11fdf7f2 | 8337 | ceph_assert(err == 0); |
31f18b77 FG |
8338 | do_osd_crush_remove(newcrush); |
8339 | ||
8340 | return 0; | |
8341 | } | |
8342 | ||
8343 | int OSDMonitor::prepare_command_osd_remove(int32_t id) | |
8344 | { | |
8345 | if (osdmap.is_up(id)) { | |
8346 | return -EBUSY; | |
8347 | } | |
8348 | ||
8349 | pending_inc.new_state[id] = osdmap.get_state(id); | |
8350 | pending_inc.new_uuid[id] = uuid_d(); | |
8351 | pending_metadata_rm.insert(id); | |
8352 | pending_metadata.erase(id); | |
8353 | ||
8354 | return 0; | |
8355 | } | |
8356 | ||
8357 | int32_t OSDMonitor::_allocate_osd_id(int32_t* existing_id) | |
8358 | { | |
11fdf7f2 | 8359 | ceph_assert(existing_id); |
31f18b77 FG |
8360 | *existing_id = -1; |
8361 | ||
8362 | for (int32_t i = 0; i < osdmap.get_max_osd(); ++i) { | |
8363 | if (!osdmap.exists(i) && | |
8364 | pending_inc.new_up_client.count(i) == 0 && | |
8365 | (pending_inc.new_state.count(i) == 0 || | |
8366 | (pending_inc.new_state[i] & CEPH_OSD_EXISTS) == 0)) { | |
8367 | *existing_id = i; | |
8368 | return -1; | |
8369 | } | |
8370 | } | |
8371 | ||
8372 | if (pending_inc.new_max_osd < 0) { | |
8373 | return osdmap.get_max_osd(); | |
8374 | } | |
8375 | return pending_inc.new_max_osd; | |
8376 | } | |
8377 | ||
8378 | void OSDMonitor::do_osd_create( | |
8379 | const int32_t id, | |
8380 | const uuid_d& uuid, | |
3a9019d9 | 8381 | const string& device_class, |
31f18b77 FG |
8382 | int32_t* new_id) |
8383 | { | |
8384 | dout(10) << __func__ << " uuid " << uuid << dendl; | |
11fdf7f2 | 8385 | ceph_assert(new_id); |
31f18b77 FG |
8386 | |
8387 | // We presume validation has been performed prior to calling this | |
8388 | // function. We assert with prejudice. | |
8389 | ||
8390 | int32_t allocated_id = -1; // declare here so we can jump | |
8391 | int32_t existing_id = -1; | |
8392 | if (!uuid.is_zero()) { | |
8393 | existing_id = osdmap.identify_osd(uuid); | |
8394 | if (existing_id >= 0) { | |
11fdf7f2 | 8395 | ceph_assert(id < 0 || id == existing_id); |
31f18b77 FG |
8396 | *new_id = existing_id; |
8397 | goto out; | |
8398 | } else if (id >= 0) { | |
8399 | // uuid does not exist, and id has been provided, so just create | |
8400 | // the new osd.id | |
8401 | *new_id = id; | |
8402 | goto out; | |
8403 | } | |
8404 | } | |
8405 | ||
8406 | // allocate a new id | |
8407 | allocated_id = _allocate_osd_id(&existing_id); | |
8408 | dout(10) << __func__ << " allocated id " << allocated_id | |
8409 | << " existing id " << existing_id << dendl; | |
8410 | if (existing_id >= 0) { | |
11fdf7f2 TL |
8411 | ceph_assert(existing_id < osdmap.get_max_osd()); |
8412 | ceph_assert(allocated_id < 0); | |
31f18b77 FG |
8413 | pending_inc.new_weight[existing_id] = CEPH_OSD_OUT; |
8414 | *new_id = existing_id; | |
31f18b77 | 8415 | } else if (allocated_id >= 0) { |
11fdf7f2 | 8416 | ceph_assert(existing_id < 0); |
31f18b77 FG |
8417 | // raise max_osd |
8418 | if (pending_inc.new_max_osd < 0) { | |
8419 | pending_inc.new_max_osd = osdmap.get_max_osd() + 1; | |
8420 | } else { | |
8421 | ++pending_inc.new_max_osd; | |
8422 | } | |
8423 | *new_id = pending_inc.new_max_osd - 1; | |
11fdf7f2 | 8424 | ceph_assert(*new_id == allocated_id); |
31f18b77 | 8425 | } else { |
11fdf7f2 | 8426 | ceph_abort_msg("unexpected condition"); |
31f18b77 FG |
8427 | } |
8428 | ||
8429 | out: | |
3a9019d9 FG |
8430 | if (device_class.size()) { |
8431 | CrushWrapper newcrush; | |
8432 | _get_pending_crush(newcrush); | |
8433 | if (newcrush.get_max_devices() < *new_id + 1) { | |
8434 | newcrush.set_max_devices(*new_id + 1); | |
8435 | } | |
8436 | string name = string("osd.") + stringify(*new_id); | |
8437 | if (!newcrush.item_exists(*new_id)) { | |
8438 | newcrush.set_item_name(*new_id, name); | |
8439 | } | |
8440 | ostringstream ss; | |
8441 | int r = newcrush.update_device_class(*new_id, device_class, name, &ss); | |
8442 | if (r < 0) { | |
8443 | derr << __func__ << " failed to set " << name << " device_class " | |
8444 | << device_class << ": " << cpp_strerror(r) << " - " << ss.str() | |
8445 | << dendl; | |
8446 | // non-fatal... this might be a replay and we want to be idempotent. | |
8447 | } else { | |
8448 | dout(20) << __func__ << " set " << name << " device_class " << device_class | |
8449 | << dendl; | |
8450 | pending_inc.crush.clear(); | |
8451 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
8452 | } | |
8453 | } else { | |
8454 | dout(20) << __func__ << " no device_class" << dendl; | |
8455 | } | |
8456 | ||
31f18b77 FG |
8457 | dout(10) << __func__ << " using id " << *new_id << dendl; |
8458 | if (osdmap.get_max_osd() <= *new_id && pending_inc.new_max_osd <= *new_id) { | |
8459 | pending_inc.new_max_osd = *new_id + 1; | |
8460 | } | |
8461 | ||
8462 | pending_inc.new_state[*new_id] |= CEPH_OSD_EXISTS | CEPH_OSD_NEW; | |
8463 | if (!uuid.is_zero()) | |
8464 | pending_inc.new_uuid[*new_id] = uuid; | |
8465 | } | |
8466 | ||
8467 | int OSDMonitor::validate_osd_create( | |
8468 | const int32_t id, | |
8469 | const uuid_d& uuid, | |
8470 | const bool check_osd_exists, | |
8471 | int32_t* existing_id, | |
8472 | stringstream& ss) | |
8473 | { | |
8474 | ||
8475 | dout(10) << __func__ << " id " << id << " uuid " << uuid | |
8476 | << " check_osd_exists " << check_osd_exists << dendl; | |
8477 | ||
11fdf7f2 | 8478 | ceph_assert(existing_id); |
31f18b77 FG |
8479 | |
8480 | if (id < 0 && uuid.is_zero()) { | |
8481 | // we have nothing to validate | |
8482 | *existing_id = -1; | |
8483 | return 0; | |
8484 | } else if (uuid.is_zero()) { | |
8485 | // we have an id but we will ignore it - because that's what | |
8486 | // `osd create` does. | |
8487 | return 0; | |
8488 | } | |
8489 | ||
8490 | /* | |
8491 | * This function will be used to validate whether we are able to | |
8492 | * create a new osd when the `uuid` is specified. | |
8493 | * | |
8494 | * It will be used by both `osd create` and `osd new`, as the checks | |
8495 | * are basically the same when it pertains to osd id and uuid validation. | |
8496 | * However, `osd create` presumes an `uuid` is optional, for legacy | |
8497 | * reasons, while `osd new` requires the `uuid` to be provided. This | |
8498 | * means that `osd create` will not be idempotent if an `uuid` is not | |
8499 | * provided, but we will always guarantee the idempotency of `osd new`. | |
8500 | */ | |
8501 | ||
11fdf7f2 | 8502 | ceph_assert(!uuid.is_zero()); |
31f18b77 FG |
8503 | if (pending_inc.identify_osd(uuid) >= 0) { |
8504 | // osd is about to exist | |
8505 | return -EAGAIN; | |
8506 | } | |
8507 | ||
8508 | int32_t i = osdmap.identify_osd(uuid); | |
8509 | if (i >= 0) { | |
8510 | // osd already exists | |
8511 | if (id >= 0 && i != id) { | |
8512 | ss << "uuid " << uuid << " already in use for different id " << i; | |
8513 | return -EEXIST; | |
8514 | } | |
8515 | // return a positive errno to distinguish between a blocking error | |
8516 | // and an error we consider to not be a problem (i.e., this would be | |
8517 | // an idempotent operation). | |
8518 | *existing_id = i; | |
8519 | return EEXIST; | |
8520 | } | |
8521 | // i < 0 | |
8522 | if (id >= 0) { | |
8523 | if (pending_inc.new_state.count(id)) { | |
8524 | // osd is about to exist | |
8525 | return -EAGAIN; | |
8526 | } | |
8527 | // we may not care if an osd exists if we are recreating a previously | |
8528 | // destroyed osd. | |
8529 | if (check_osd_exists && osdmap.exists(id)) { | |
8530 | ss << "id " << id << " already in use and does not match uuid " | |
8531 | << uuid; | |
8532 | return -EINVAL; | |
8533 | } | |
8534 | } | |
8535 | return 0; | |
8536 | } | |
8537 | ||
8538 | int OSDMonitor::prepare_command_osd_create( | |
8539 | const int32_t id, | |
8540 | const uuid_d& uuid, | |
8541 | int32_t* existing_id, | |
8542 | stringstream& ss) | |
8543 | { | |
8544 | dout(10) << __func__ << " id " << id << " uuid " << uuid << dendl; | |
11fdf7f2 | 8545 | ceph_assert(existing_id); |
b5b8bbf5 FG |
8546 | if (osdmap.is_destroyed(id)) { |
8547 | ss << "ceph osd create has been deprecated. Please use ceph osd new " | |
8548 | "instead."; | |
8549 | return -EINVAL; | |
8550 | } | |
31f18b77 FG |
8551 | |
8552 | if (uuid.is_zero()) { | |
8553 | dout(10) << __func__ << " no uuid; assuming legacy `osd create`" << dendl; | |
8554 | } | |
8555 | ||
8556 | return validate_osd_create(id, uuid, true, existing_id, ss); | |
8557 | } | |
8558 | ||
8559 | int OSDMonitor::prepare_command_osd_new( | |
8560 | MonOpRequestRef op, | |
11fdf7f2 | 8561 | const cmdmap_t& cmdmap, |
3a9019d9 | 8562 | const map<string,string>& params, |
31f18b77 FG |
8563 | stringstream &ss, |
8564 | Formatter *f) | |
8565 | { | |
8566 | uuid_d uuid; | |
8567 | string uuidstr; | |
8568 | int64_t id = -1; | |
8569 | ||
11fdf7f2 | 8570 | ceph_assert(paxos->is_plugged()); |
31f18b77 FG |
8571 | |
8572 | dout(10) << __func__ << " " << op << dendl; | |
8573 | ||
8574 | /* validate command. abort now if something's wrong. */ | |
8575 | ||
8576 | /* `osd new` will expect a `uuid` to be supplied; `id` is optional. | |
8577 | * | |
8578 | * If `id` is not specified, we will identify any existing osd based | |
8579 | * on `uuid`. Operation will be idempotent iff secrets match. | |
8580 | * | |
8581 | * If `id` is specified, we will identify any existing osd based on | |
8582 | * `uuid` and match against `id`. If they match, operation will be | |
8583 | * idempotent iff secrets match. | |
8584 | * | |
8585 | * `-i secrets.json` will be optional. If supplied, will be used | |
8586 | * to check for idempotency when `id` and `uuid` match. | |
8587 | * | |
8588 | * If `id` is not specified, and `uuid` does not exist, an id will | |
8589 | * be found or allocated for the osd. | |
8590 | * | |
8591 | * If `id` is specified, and the osd has been previously marked | |
8592 | * as destroyed, then the `id` will be reused. | |
8593 | */ | |
11fdf7f2 | 8594 | if (!cmd_getval(cct, cmdmap, "uuid", uuidstr)) { |
31f18b77 FG |
8595 | ss << "requires the OSD's UUID to be specified."; |
8596 | return -EINVAL; | |
8597 | } else if (!uuid.parse(uuidstr.c_str())) { | |
8598 | ss << "invalid UUID value '" << uuidstr << "'."; | |
8599 | return -EINVAL; | |
8600 | } | |
8601 | ||
11fdf7f2 | 8602 | if (cmd_getval(cct, cmdmap, "id", id) && |
31f18b77 FG |
8603 | (id < 0)) { |
8604 | ss << "invalid OSD id; must be greater or equal than zero."; | |
8605 | return -EINVAL; | |
8606 | } | |
8607 | ||
8608 | // are we running an `osd create`-like command, or recreating | |
8609 | // a previously destroyed osd? | |
8610 | ||
8611 | bool is_recreate_destroyed = (id >= 0 && osdmap.is_destroyed(id)); | |
8612 | ||
8613 | // we will care about `id` to assess whether osd is `destroyed`, or | |
8614 | // to create a new osd. | |
8615 | // we will need an `id` by the time we reach auth. | |
8616 | ||
8617 | int32_t existing_id = -1; | |
8618 | int err = validate_osd_create(id, uuid, !is_recreate_destroyed, | |
8619 | &existing_id, ss); | |
8620 | ||
8621 | bool may_be_idempotent = false; | |
8622 | if (err == EEXIST) { | |
8623 | // this is idempotent from the osdmon's point-of-view | |
8624 | may_be_idempotent = true; | |
11fdf7f2 | 8625 | ceph_assert(existing_id >= 0); |
31f18b77 FG |
8626 | id = existing_id; |
8627 | } else if (err < 0) { | |
8628 | return err; | |
8629 | } | |
8630 | ||
8631 | if (!may_be_idempotent) { | |
8632 | // idempotency is out of the window. We are either creating a new | |
8633 | // osd or recreating a destroyed osd. | |
8634 | // | |
8635 | // We now need to figure out if we have an `id` (and if it's valid), | |
8636 | // of find an `id` if we don't have one. | |
8637 | ||
8638 | // NOTE: we need to consider the case where the `id` is specified for | |
8639 | // `osd create`, and we must honor it. So this means checking if | |
8640 | // the `id` is destroyed, and if so assume the destroy; otherwise, | |
8641 | // check if it `exists` - in which case we complain about not being | |
8642 | // `destroyed`. In the end, if nothing fails, we must allow the | |
8643 | // creation, so that we are compatible with `create`. | |
8644 | if (id >= 0 && osdmap.exists(id) && !osdmap.is_destroyed(id)) { | |
8645 | dout(10) << __func__ << " osd." << id << " isn't destroyed" << dendl; | |
8646 | ss << "OSD " << id << " has not yet been destroyed"; | |
8647 | return -EINVAL; | |
8648 | } else if (id < 0) { | |
8649 | // find an `id` | |
8650 | id = _allocate_osd_id(&existing_id); | |
8651 | if (id < 0) { | |
11fdf7f2 | 8652 | ceph_assert(existing_id >= 0); |
31f18b77 FG |
8653 | id = existing_id; |
8654 | } | |
8655 | dout(10) << __func__ << " found id " << id << " to use" << dendl; | |
8656 | } else if (id >= 0 && osdmap.is_destroyed(id)) { | |
8657 | dout(10) << __func__ << " recreating osd." << id << dendl; | |
8658 | } else { | |
8659 | dout(10) << __func__ << " creating new osd." << id << dendl; | |
8660 | } | |
8661 | } else { | |
11fdf7f2 TL |
8662 | ceph_assert(id >= 0); |
8663 | ceph_assert(osdmap.exists(id)); | |
31f18b77 FG |
8664 | } |
8665 | ||
8666 | // we are now able to either create a brand new osd or reuse an existing | |
8667 | // osd that has been previously destroyed. | |
8668 | ||
8669 | dout(10) << __func__ << " id " << id << " uuid " << uuid << dendl; | |
8670 | ||
3a9019d9 | 8671 | if (may_be_idempotent && params.empty()) { |
31f18b77 | 8672 | // nothing to do, really. |
3a9019d9 | 8673 | dout(10) << __func__ << " idempotent and no params -- no op." << dendl; |
11fdf7f2 | 8674 | ceph_assert(id >= 0); |
31f18b77 FG |
8675 | if (f) { |
8676 | f->open_object_section("created_osd"); | |
8677 | f->dump_int("osdid", id); | |
8678 | f->close_section(); | |
8679 | } else { | |
8680 | ss << id; | |
8681 | } | |
8682 | return EEXIST; | |
8683 | } | |
8684 | ||
3a9019d9 FG |
8685 | string device_class; |
8686 | auto p = params.find("crush_device_class"); | |
8687 | if (p != params.end()) { | |
8688 | device_class = p->second; | |
8689 | dout(20) << __func__ << " device_class will be " << device_class << dendl; | |
8690 | } | |
31f18b77 FG |
8691 | string cephx_secret, lockbox_secret, dmcrypt_key; |
8692 | bool has_lockbox = false; | |
3a9019d9 FG |
8693 | bool has_secrets = params.count("cephx_secret") |
8694 | || params.count("cephx_lockbox_secret") | |
8695 | || params.count("dmcrypt_key"); | |
31f18b77 FG |
8696 | |
8697 | ConfigKeyService *svc = nullptr; | |
8698 | AuthMonitor::auth_entity_t cephx_entity, lockbox_entity; | |
8699 | ||
8700 | if (has_secrets) { | |
3a9019d9 | 8701 | if (params.count("cephx_secret") == 0) { |
31f18b77 FG |
8702 | ss << "requires a cephx secret."; |
8703 | return -EINVAL; | |
8704 | } | |
3a9019d9 | 8705 | cephx_secret = params.at("cephx_secret"); |
31f18b77 | 8706 | |
3a9019d9 FG |
8707 | bool has_lockbox_secret = (params.count("cephx_lockbox_secret") > 0); |
8708 | bool has_dmcrypt_key = (params.count("dmcrypt_key") > 0); | |
31f18b77 FG |
8709 | |
8710 | dout(10) << __func__ << " has lockbox " << has_lockbox_secret | |
8711 | << " dmcrypt " << has_dmcrypt_key << dendl; | |
8712 | ||
8713 | if (has_lockbox_secret && has_dmcrypt_key) { | |
8714 | has_lockbox = true; | |
3a9019d9 FG |
8715 | lockbox_secret = params.at("cephx_lockbox_secret"); |
8716 | dmcrypt_key = params.at("dmcrypt_key"); | |
31f18b77 FG |
8717 | } else if (!has_lockbox_secret != !has_dmcrypt_key) { |
8718 | ss << "requires both a cephx lockbox secret and a dm-crypt key."; | |
8719 | return -EINVAL; | |
8720 | } | |
8721 | ||
8722 | dout(10) << __func__ << " validate secrets using osd id " << id << dendl; | |
8723 | ||
8724 | err = mon->authmon()->validate_osd_new(id, uuid, | |
8725 | cephx_secret, | |
8726 | lockbox_secret, | |
8727 | cephx_entity, | |
8728 | lockbox_entity, | |
8729 | ss); | |
8730 | if (err < 0) { | |
8731 | return err; | |
8732 | } else if (may_be_idempotent && err != EEXIST) { | |
8733 | // for this to be idempotent, `id` should already be >= 0; no need | |
8734 | // to use validate_id. | |
11fdf7f2 | 8735 | ceph_assert(id >= 0); |
31f18b77 FG |
8736 | ss << "osd." << id << " exists but secrets do not match"; |
8737 | return -EEXIST; | |
8738 | } | |
8739 | ||
8740 | if (has_lockbox) { | |
8741 | svc = (ConfigKeyService*)mon->config_key_service; | |
8742 | err = svc->validate_osd_new(uuid, dmcrypt_key, ss); | |
8743 | if (err < 0) { | |
8744 | return err; | |
8745 | } else if (may_be_idempotent && err != EEXIST) { | |
11fdf7f2 | 8746 | ceph_assert(id >= 0); |
31f18b77 FG |
8747 | ss << "osd." << id << " exists but dm-crypt key does not match."; |
8748 | return -EEXIST; | |
8749 | } | |
8750 | } | |
8751 | } | |
11fdf7f2 TL |
8752 | ceph_assert(!has_secrets || !cephx_secret.empty()); |
8753 | ceph_assert(!has_lockbox || !lockbox_secret.empty()); | |
31f18b77 FG |
8754 | |
8755 | if (may_be_idempotent) { | |
8756 | // we have nothing to do for either the osdmon or the authmon, | |
8757 | // and we have no lockbox - so the config key service will not be | |
8758 | // touched. This is therefore an idempotent operation, and we can | |
8759 | // just return right away. | |
8760 | dout(10) << __func__ << " idempotent -- no op." << dendl; | |
11fdf7f2 | 8761 | ceph_assert(id >= 0); |
31f18b77 FG |
8762 | if (f) { |
8763 | f->open_object_section("created_osd"); | |
8764 | f->dump_int("osdid", id); | |
8765 | f->close_section(); | |
8766 | } else { | |
8767 | ss << id; | |
8768 | } | |
8769 | return EEXIST; | |
8770 | } | |
11fdf7f2 | 8771 | ceph_assert(!may_be_idempotent); |
31f18b77 FG |
8772 | |
8773 | // perform updates. | |
8774 | if (has_secrets) { | |
11fdf7f2 TL |
8775 | ceph_assert(!cephx_secret.empty()); |
8776 | ceph_assert((lockbox_secret.empty() && dmcrypt_key.empty()) || | |
31f18b77 FG |
8777 | (!lockbox_secret.empty() && !dmcrypt_key.empty())); |
8778 | ||
8779 | err = mon->authmon()->do_osd_new(cephx_entity, | |
8780 | lockbox_entity, | |
8781 | has_lockbox); | |
11fdf7f2 | 8782 | ceph_assert(0 == err); |
31f18b77 FG |
8783 | |
8784 | if (has_lockbox) { | |
11fdf7f2 | 8785 | ceph_assert(nullptr != svc); |
31f18b77 FG |
8786 | svc->do_osd_new(uuid, dmcrypt_key); |
8787 | } | |
8788 | } | |
8789 | ||
8790 | if (is_recreate_destroyed) { | |
11fdf7f2 TL |
8791 | ceph_assert(id >= 0); |
8792 | ceph_assert(osdmap.is_destroyed(id)); | |
31f18b77 | 8793 | pending_inc.new_weight[id] = CEPH_OSD_OUT; |
11fdf7f2 TL |
8794 | pending_inc.new_state[id] |= CEPH_OSD_DESTROYED; |
8795 | if ((osdmap.get_state(id) & CEPH_OSD_NEW) == 0) { | |
8796 | pending_inc.new_state[id] |= CEPH_OSD_NEW; | |
8797 | } | |
c07f9fc5 FG |
8798 | if (osdmap.get_state(id) & CEPH_OSD_UP) { |
8799 | // due to http://tracker.ceph.com/issues/20751 some clusters may | |
8800 | // have UP set for non-existent OSDs; make sure it is cleared | |
8801 | // for a newly created osd. | |
8802 | pending_inc.new_state[id] |= CEPH_OSD_UP; | |
8803 | } | |
31f18b77 FG |
8804 | pending_inc.new_uuid[id] = uuid; |
8805 | } else { | |
11fdf7f2 | 8806 | ceph_assert(id >= 0); |
31f18b77 | 8807 | int32_t new_id = -1; |
3a9019d9 | 8808 | do_osd_create(id, uuid, device_class, &new_id); |
11fdf7f2 TL |
8809 | ceph_assert(new_id >= 0); |
8810 | ceph_assert(id == new_id); | |
31f18b77 FG |
8811 | } |
8812 | ||
8813 | if (f) { | |
8814 | f->open_object_section("created_osd"); | |
8815 | f->dump_int("osdid", id); | |
8816 | f->close_section(); | |
8817 | } else { | |
8818 | ss << id; | |
8819 | } | |
8820 | ||
8821 | return 0; | |
8822 | } | |
8823 | ||
7c673cae FG |
8824 | bool OSDMonitor::prepare_command(MonOpRequestRef op) |
8825 | { | |
8826 | op->mark_osdmon_event(__func__); | |
8827 | MMonCommand *m = static_cast<MMonCommand*>(op->get_req()); | |
8828 | stringstream ss; | |
11fdf7f2 | 8829 | cmdmap_t cmdmap; |
7c673cae FG |
8830 | if (!cmdmap_from_json(m->cmd, &cmdmap, ss)) { |
8831 | string rs = ss.str(); | |
8832 | mon->reply_command(op, -EINVAL, rs, get_last_committed()); | |
8833 | return true; | |
8834 | } | |
8835 | ||
11fdf7f2 | 8836 | MonSession *session = op->get_session(); |
7c673cae | 8837 | if (!session) { |
11fdf7f2 | 8838 | derr << __func__ << " no session" << dendl; |
7c673cae FG |
8839 | mon->reply_command(op, -EACCES, "access denied", get_last_committed()); |
8840 | return true; | |
8841 | } | |
8842 | ||
8843 | return prepare_command_impl(op, cmdmap); | |
8844 | } | |
8845 | ||
8846 | static int parse_reweights(CephContext *cct, | |
11fdf7f2 | 8847 | const cmdmap_t& cmdmap, |
7c673cae FG |
8848 | const OSDMap& osdmap, |
8849 | map<int32_t, uint32_t>* weights) | |
8850 | { | |
8851 | string weights_str; | |
11fdf7f2 | 8852 | if (!cmd_getval(cct, cmdmap, "weights", weights_str)) { |
7c673cae FG |
8853 | return -EINVAL; |
8854 | } | |
8855 | std::replace(begin(weights_str), end(weights_str), '\'', '"'); | |
8856 | json_spirit::mValue json_value; | |
8857 | if (!json_spirit::read(weights_str, json_value)) { | |
8858 | return -EINVAL; | |
8859 | } | |
8860 | if (json_value.type() != json_spirit::obj_type) { | |
8861 | return -EINVAL; | |
8862 | } | |
8863 | const auto obj = json_value.get_obj(); | |
8864 | try { | |
8865 | for (auto& osd_weight : obj) { | |
8866 | auto osd_id = std::stoi(osd_weight.first); | |
8867 | if (!osdmap.exists(osd_id)) { | |
8868 | return -ENOENT; | |
8869 | } | |
8870 | if (osd_weight.second.type() != json_spirit::str_type) { | |
8871 | return -EINVAL; | |
8872 | } | |
8873 | auto weight = std::stoul(osd_weight.second.get_str()); | |
8874 | weights->insert({osd_id, weight}); | |
8875 | } | |
8876 | } catch (const std::logic_error& e) { | |
8877 | return -EINVAL; | |
8878 | } | |
8879 | return 0; | |
8880 | } | |
8881 | ||
31f18b77 FG |
8882 | int OSDMonitor::prepare_command_osd_destroy( |
8883 | int32_t id, | |
8884 | stringstream& ss) | |
8885 | { | |
11fdf7f2 | 8886 | ceph_assert(paxos->is_plugged()); |
31f18b77 FG |
8887 | |
8888 | // we check if the osd exists for the benefit of `osd purge`, which may | |
8889 | // have previously removed the osd. If the osd does not exist, return | |
8890 | // -ENOENT to convey this, and let the caller deal with it. | |
8891 | // | |
8892 | // we presume that all auth secrets and config keys were removed prior | |
8893 | // to this command being called. if they exist by now, we also assume | |
8894 | // they must have been created by some other command and do not pertain | |
8895 | // to this non-existent osd. | |
8896 | if (!osdmap.exists(id)) { | |
8897 | dout(10) << __func__ << " osd." << id << " does not exist." << dendl; | |
8898 | return -ENOENT; | |
8899 | } | |
8900 | ||
8901 | uuid_d uuid = osdmap.get_uuid(id); | |
8902 | dout(10) << __func__ << " destroying osd." << id | |
8903 | << " uuid " << uuid << dendl; | |
8904 | ||
8905 | // if it has been destroyed, we assume our work here is done. | |
8906 | if (osdmap.is_destroyed(id)) { | |
8907 | ss << "destroyed osd." << id; | |
8908 | return 0; | |
8909 | } | |
8910 | ||
8911 | EntityName cephx_entity, lockbox_entity; | |
8912 | bool idempotent_auth = false, idempotent_cks = false; | |
8913 | ||
8914 | int err = mon->authmon()->validate_osd_destroy(id, uuid, | |
8915 | cephx_entity, | |
8916 | lockbox_entity, | |
8917 | ss); | |
8918 | if (err < 0) { | |
8919 | if (err == -ENOENT) { | |
8920 | idempotent_auth = true; | |
31f18b77 FG |
8921 | } else { |
8922 | return err; | |
8923 | } | |
8924 | } | |
8925 | ||
8926 | ConfigKeyService *svc = (ConfigKeyService*)mon->config_key_service; | |
8927 | err = svc->validate_osd_destroy(id, uuid); | |
8928 | if (err < 0) { | |
11fdf7f2 | 8929 | ceph_assert(err == -ENOENT); |
31f18b77 FG |
8930 | err = 0; |
8931 | idempotent_cks = true; | |
8932 | } | |
8933 | ||
8934 | if (!idempotent_auth) { | |
8935 | err = mon->authmon()->do_osd_destroy(cephx_entity, lockbox_entity); | |
11fdf7f2 | 8936 | ceph_assert(0 == err); |
31f18b77 FG |
8937 | } |
8938 | ||
8939 | if (!idempotent_cks) { | |
8940 | svc->do_osd_destroy(id, uuid); | |
8941 | } | |
8942 | ||
8943 | pending_inc.new_state[id] = CEPH_OSD_DESTROYED; | |
8944 | pending_inc.new_uuid[id] = uuid_d(); | |
8945 | ||
8946 | // we can only propose_pending() once per service, otherwise we'll be | |
8947 | // defying PaxosService and all laws of nature. Therefore, as we may | |
8948 | // be used during 'osd purge', let's keep the caller responsible for | |
8949 | // proposing. | |
11fdf7f2 | 8950 | ceph_assert(err == 0); |
31f18b77 FG |
8951 | return 0; |
8952 | } | |
8953 | ||
8954 | int OSDMonitor::prepare_command_osd_purge( | |
8955 | int32_t id, | |
8956 | stringstream& ss) | |
8957 | { | |
11fdf7f2 | 8958 | ceph_assert(paxos->is_plugged()); |
31f18b77 FG |
8959 | dout(10) << __func__ << " purging osd." << id << dendl; |
8960 | ||
11fdf7f2 | 8961 | ceph_assert(!osdmap.is_up(id)); |
31f18b77 FG |
8962 | |
8963 | /* | |
8964 | * This may look a bit weird, but this is what's going to happen: | |
8965 | * | |
8966 | * 1. we make sure that removing from crush works | |
8967 | * 2. we call `prepare_command_osd_destroy()`. If it returns an | |
8968 | * error, then we abort the whole operation, as no updates | |
8969 | * have been made. However, we this function will have | |
8970 | * side-effects, thus we need to make sure that all operations | |
8971 | * performed henceforth will *always* succeed. | |
8972 | * 3. we call `prepare_command_osd_remove()`. Although this | |
8973 | * function can return an error, it currently only checks if the | |
8974 | * osd is up - and we have made sure that it is not so, so there | |
8975 | * is no conflict, and it is effectively an update. | |
8976 | * 4. finally, we call `do_osd_crush_remove()`, which will perform | |
8977 | * the crush update we delayed from before. | |
8978 | */ | |
8979 | ||
8980 | CrushWrapper newcrush; | |
8981 | _get_pending_crush(newcrush); | |
8982 | ||
8983 | bool may_be_idempotent = false; | |
8984 | ||
8985 | int err = _prepare_command_osd_crush_remove(newcrush, id, 0, false, false); | |
8986 | if (err == -ENOENT) { | |
8987 | err = 0; | |
8988 | may_be_idempotent = true; | |
8989 | } else if (err < 0) { | |
8990 | ss << "error removing osd." << id << " from crush"; | |
8991 | return err; | |
8992 | } | |
8993 | ||
8994 | // no point destroying the osd again if it has already been marked destroyed | |
8995 | if (!osdmap.is_destroyed(id)) { | |
8996 | err = prepare_command_osd_destroy(id, ss); | |
8997 | if (err < 0) { | |
8998 | if (err == -ENOENT) { | |
8999 | err = 0; | |
9000 | } else { | |
9001 | return err; | |
9002 | } | |
9003 | } else { | |
9004 | may_be_idempotent = false; | |
9005 | } | |
9006 | } | |
11fdf7f2 | 9007 | ceph_assert(0 == err); |
31f18b77 FG |
9008 | |
9009 | if (may_be_idempotent && !osdmap.exists(id)) { | |
9010 | dout(10) << __func__ << " osd." << id << " does not exist and " | |
9011 | << "we are idempotent." << dendl; | |
9012 | return -ENOENT; | |
9013 | } | |
9014 | ||
9015 | err = prepare_command_osd_remove(id); | |
9016 | // we should not be busy, as we should have made sure this id is not up. | |
11fdf7f2 | 9017 | ceph_assert(0 == err); |
31f18b77 FG |
9018 | |
9019 | do_osd_crush_remove(newcrush); | |
9020 | return 0; | |
9021 | } | |
9022 | ||
7c673cae | 9023 | bool OSDMonitor::prepare_command_impl(MonOpRequestRef op, |
11fdf7f2 | 9024 | const cmdmap_t& cmdmap) |
7c673cae FG |
9025 | { |
9026 | op->mark_osdmon_event(__func__); | |
9027 | MMonCommand *m = static_cast<MMonCommand*>(op->get_req()); | |
9028 | bool ret = false; | |
9029 | stringstream ss; | |
9030 | string rs; | |
9031 | bufferlist rdata; | |
9032 | int err = 0; | |
9033 | ||
9034 | string format; | |
11fdf7f2 | 9035 | cmd_getval(cct, cmdmap, "format", format, string("plain")); |
7c673cae FG |
9036 | boost::scoped_ptr<Formatter> f(Formatter::create(format)); |
9037 | ||
9038 | string prefix; | |
11fdf7f2 | 9039 | cmd_getval(cct, cmdmap, "prefix", prefix); |
7c673cae FG |
9040 | |
9041 | int64_t osdid; | |
11fdf7f2 | 9042 | string osd_name; |
b32b8144 FG |
9043 | bool osdid_present = false; |
9044 | if (prefix != "osd pg-temp" && | |
9045 | prefix != "osd pg-upmap" && | |
9046 | prefix != "osd pg-upmap-items") { // avoid commands with non-int id arg | |
11fdf7f2 | 9047 | osdid_present = cmd_getval(cct, cmdmap, "id", osdid); |
b32b8144 | 9048 | } |
7c673cae FG |
9049 | if (osdid_present) { |
9050 | ostringstream oss; | |
9051 | oss << "osd." << osdid; | |
11fdf7f2 | 9052 | osd_name = oss.str(); |
7c673cae FG |
9053 | } |
9054 | ||
9055 | // Even if there's a pending state with changes that could affect | |
9056 | // a command, considering that said state isn't yet committed, we | |
9057 | // just don't care about those changes if the command currently being | |
9058 | // handled acts as a no-op against the current committed state. | |
9059 | // In a nutshell, we assume this command happens *before*. | |
9060 | // | |
9061 | // Let me make this clearer: | |
9062 | // | |
9063 | // - If we have only one client, and that client issues some | |
9064 | // operation that would conflict with this operation but is | |
9065 | // still on the pending state, then we would be sure that said | |
9066 | // operation wouldn't have returned yet, so the client wouldn't | |
9067 | // issue this operation (unless the client didn't wait for the | |
9068 | // operation to finish, and that would be the client's own fault). | |
9069 | // | |
9070 | // - If we have more than one client, each client will observe | |
9071 | // whatever is the state at the moment of the commit. So, if we | |
9072 | // have two clients, one issuing an unlink and another issuing a | |
9073 | // link, and if the link happens while the unlink is still on the | |
9074 | // pending state, from the link's point-of-view this is a no-op. | |
9075 | // If different clients are issuing conflicting operations and | |
9076 | // they care about that, then the clients should make sure they | |
9077 | // enforce some kind of concurrency mechanism -- from our | |
9078 | // perspective that's what Douglas Adams would call an SEP. | |
9079 | // | |
9080 | // This should be used as a general guideline for most commands handled | |
9081 | // in this function. Adapt as you see fit, but please bear in mind that | |
9082 | // this is the expected behavior. | |
9083 | ||
9084 | ||
9085 | if (prefix == "osd setcrushmap" || | |
9086 | (prefix == "osd crush set" && !osdid_present)) { | |
31f18b77 FG |
9087 | if (pending_inc.crush.length()) { |
9088 | dout(10) << __func__ << " waiting for pending crush update " << dendl; | |
9089 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
9090 | return true; | |
9091 | } | |
7c673cae FG |
9092 | dout(10) << "prepare_command setting new crush map" << dendl; |
9093 | bufferlist data(m->get_data()); | |
9094 | CrushWrapper crush; | |
9095 | try { | |
11fdf7f2 | 9096 | auto bl = data.cbegin(); |
7c673cae FG |
9097 | crush.decode(bl); |
9098 | } | |
9099 | catch (const std::exception &e) { | |
9100 | err = -EINVAL; | |
9101 | ss << "Failed to parse crushmap: " << e.what(); | |
9102 | goto reply; | |
9103 | } | |
31f18b77 FG |
9104 | |
9105 | int64_t prior_version = 0; | |
11fdf7f2 | 9106 | if (cmd_getval(cct, cmdmap, "prior_version", prior_version)) { |
31f18b77 FG |
9107 | if (prior_version == osdmap.get_crush_version() - 1) { |
9108 | // see if we are a resend of the last update. this is imperfect | |
9109 | // (multiple racing updaters may not both get reliable success) | |
9110 | // but we expect crush updaters (via this interface) to be rare-ish. | |
9111 | bufferlist current, proposed; | |
9112 | osdmap.crush->encode(current, mon->get_quorum_con_features()); | |
9113 | crush.encode(proposed, mon->get_quorum_con_features()); | |
9114 | if (current.contents_equal(proposed)) { | |
9115 | dout(10) << __func__ | |
9116 | << " proposed matches current and version equals previous" | |
9117 | << dendl; | |
9118 | err = 0; | |
9119 | ss << osdmap.get_crush_version(); | |
9120 | goto reply; | |
9121 | } | |
9122 | } | |
9123 | if (prior_version != osdmap.get_crush_version()) { | |
9124 | err = -EPERM; | |
9125 | ss << "prior_version " << prior_version << " != crush version " | |
9126 | << osdmap.get_crush_version(); | |
9127 | goto reply; | |
9128 | } | |
9129 | } | |
7c673cae | 9130 | |
3efd9988 | 9131 | if (crush.has_legacy_rule_ids()) { |
31f18b77 FG |
9132 | err = -EINVAL; |
9133 | ss << "crush maps with ruleset != ruleid are no longer allowed"; | |
9134 | goto reply; | |
9135 | } | |
7c673cae FG |
9136 | if (!validate_crush_against_features(&crush, ss)) { |
9137 | err = -EINVAL; | |
9138 | goto reply; | |
9139 | } | |
31f18b77 | 9140 | |
3efd9988 FG |
9141 | err = osdmap.validate_crush_rules(&crush, &ss); |
9142 | if (err < 0) { | |
9143 | goto reply; | |
7c673cae FG |
9144 | } |
9145 | ||
11fdf7f2 | 9146 | if (g_conf()->mon_osd_crush_smoke_test) { |
224ce89b WB |
9147 | // sanity check: test some inputs to make sure this map isn't |
9148 | // totally broken | |
9149 | dout(10) << " testing map" << dendl; | |
9150 | stringstream ess; | |
9151 | CrushTester tester(crush, ess); | |
b5b8bbf5 | 9152 | tester.set_min_x(0); |
224ce89b | 9153 | tester.set_max_x(50); |
b5b8bbf5 | 9154 | auto start = ceph::coarse_mono_clock::now(); |
11fdf7f2 | 9155 | int r = tester.test_with_fork(g_conf()->mon_lease); |
b5b8bbf5 | 9156 | auto duration = ceph::coarse_mono_clock::now() - start; |
224ce89b WB |
9157 | if (r < 0) { |
9158 | dout(10) << " tester.test_with_fork returns " << r | |
9159 | << ": " << ess.str() << dendl; | |
9160 | ss << "crush smoke test failed with " << r << ": " << ess.str(); | |
9161 | err = r; | |
9162 | goto reply; | |
9163 | } | |
b5b8bbf5 FG |
9164 | dout(10) << __func__ << " crush somke test duration: " |
9165 | << duration << ", result: " << ess.str() << dendl; | |
7c673cae FG |
9166 | } |
9167 | ||
7c673cae | 9168 | pending_inc.crush = data; |
31f18b77 | 9169 | ss << osdmap.get_crush_version() + 1; |
7c673cae FG |
9170 | goto update; |
9171 | ||
3efd9988 FG |
9172 | } else if (prefix == "osd crush set-all-straw-buckets-to-straw2") { |
9173 | CrushWrapper newcrush; | |
9174 | _get_pending_crush(newcrush); | |
9175 | for (int b = 0; b < newcrush.get_max_buckets(); ++b) { | |
9176 | int bid = -1 - b; | |
9177 | if (newcrush.bucket_exists(bid) && | |
11fdf7f2 | 9178 | newcrush.get_bucket_alg(bid) == CRUSH_BUCKET_STRAW) { |
3efd9988 FG |
9179 | dout(20) << " bucket " << bid << " is straw, can convert" << dendl; |
9180 | newcrush.bucket_set_alg(bid, CRUSH_BUCKET_STRAW2); | |
9181 | } | |
9182 | } | |
9183 | if (!validate_crush_against_features(&newcrush, ss)) { | |
9184 | err = -EINVAL; | |
9185 | goto reply; | |
9186 | } | |
9187 | pending_inc.crush.clear(); | |
9188 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9189 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
9190 | get_last_committed() + 1)); | |
9191 | return true; | |
7c673cae | 9192 | } else if (prefix == "osd crush set-device-class") { |
7c673cae | 9193 | string device_class; |
11fdf7f2 | 9194 | if (!cmd_getval(cct, cmdmap, "class", device_class)) { |
7c673cae FG |
9195 | err = -EINVAL; // no value! |
9196 | goto reply; | |
9197 | } | |
9198 | ||
224ce89b WB |
9199 | bool stop = false; |
9200 | vector<string> idvec; | |
11fdf7f2 | 9201 | cmd_getval(cct, cmdmap, "ids", idvec); |
7c673cae FG |
9202 | CrushWrapper newcrush; |
9203 | _get_pending_crush(newcrush); | |
224ce89b WB |
9204 | set<int> updated; |
9205 | for (unsigned j = 0; j < idvec.size() && !stop; j++) { | |
9206 | set<int> osds; | |
9207 | // wildcard? | |
9208 | if (j == 0 && | |
9209 | (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) { | |
9210 | osdmap.get_all_osds(osds); | |
9211 | stop = true; | |
9212 | } else { | |
9213 | // try traditional single osd way | |
9214 | long osd = parse_osd_id(idvec[j].c_str(), &ss); | |
9215 | if (osd < 0) { | |
9216 | // ss has reason for failure | |
9217 | ss << ", unable to parse osd id:\"" << idvec[j] << "\". "; | |
9218 | err = -EINVAL; | |
9219 | continue; | |
9220 | } | |
9221 | osds.insert(osd); | |
9222 | } | |
7c673cae | 9223 | |
224ce89b WB |
9224 | for (auto &osd : osds) { |
9225 | if (!osdmap.exists(osd)) { | |
9226 | ss << "osd." << osd << " does not exist. "; | |
9227 | continue; | |
9228 | } | |
7c673cae | 9229 | |
224ce89b WB |
9230 | ostringstream oss; |
9231 | oss << "osd." << osd; | |
9232 | string name = oss.str(); | |
7c673cae | 9233 | |
3a9019d9 FG |
9234 | if (newcrush.get_max_devices() < osd + 1) { |
9235 | newcrush.set_max_devices(osd + 1); | |
9236 | } | |
224ce89b WB |
9237 | string action; |
9238 | if (newcrush.item_exists(osd)) { | |
9239 | action = "updating"; | |
9240 | } else { | |
9241 | action = "creating"; | |
9242 | newcrush.set_item_name(osd, name); | |
9243 | } | |
7c673cae | 9244 | |
224ce89b WB |
9245 | dout(5) << action << " crush item id " << osd << " name '" << name |
9246 | << "' device_class '" << device_class << "'" | |
9247 | << dendl; | |
9248 | err = newcrush.update_device_class(osd, device_class, name, &ss); | |
9249 | if (err < 0) { | |
9250 | goto reply; | |
9251 | } | |
9252 | if (err == 0 && !_have_pending_crush()) { | |
9253 | if (!stop) { | |
9254 | // for single osd only, wildcard makes too much noise | |
9255 | ss << "set-device-class item id " << osd << " name '" << name | |
11fdf7f2 | 9256 | << "' device_class '" << device_class << "': no change. "; |
224ce89b WB |
9257 | } |
9258 | } else { | |
9259 | updated.insert(osd); | |
9260 | } | |
9261 | } | |
7c673cae FG |
9262 | } |
9263 | ||
224ce89b WB |
9264 | if (!updated.empty()) { |
9265 | pending_inc.crush.clear(); | |
9266 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9267 | ss << "set osd(s) " << updated << " to class '" << device_class << "'"; | |
9268 | getline(ss, rs); | |
9269 | wait_for_finished_proposal(op, | |
9270 | new Monitor::C_Command(mon,op, 0, rs, get_last_committed() + 1)); | |
9271 | return true; | |
9272 | } | |
7c673cae | 9273 | |
c07f9fc5 FG |
9274 | } else if (prefix == "osd crush rm-device-class") { |
9275 | bool stop = false; | |
9276 | vector<string> idvec; | |
11fdf7f2 | 9277 | cmd_getval(cct, cmdmap, "ids", idvec); |
c07f9fc5 FG |
9278 | CrushWrapper newcrush; |
9279 | _get_pending_crush(newcrush); | |
9280 | set<int> updated; | |
9281 | ||
9282 | for (unsigned j = 0; j < idvec.size() && !stop; j++) { | |
9283 | set<int> osds; | |
9284 | ||
9285 | // wildcard? | |
9286 | if (j == 0 && | |
9287 | (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) { | |
9288 | osdmap.get_all_osds(osds); | |
9289 | stop = true; | |
9290 | } else { | |
9291 | // try traditional single osd way | |
9292 | long osd = parse_osd_id(idvec[j].c_str(), &ss); | |
9293 | if (osd < 0) { | |
9294 | // ss has reason for failure | |
9295 | ss << ", unable to parse osd id:\"" << idvec[j] << "\". "; | |
9296 | err = -EINVAL; | |
9297 | goto reply; | |
9298 | } | |
9299 | osds.insert(osd); | |
9300 | } | |
9301 | ||
9302 | for (auto &osd : osds) { | |
9303 | if (!osdmap.exists(osd)) { | |
9304 | ss << "osd." << osd << " does not exist. "; | |
9305 | continue; | |
9306 | } | |
9307 | ||
9308 | auto class_name = newcrush.get_item_class(osd); | |
c07f9fc5 FG |
9309 | if (!class_name) { |
9310 | ss << "osd." << osd << " belongs to no class, "; | |
9311 | continue; | |
9312 | } | |
9313 | // note that we do not verify if class_is_in_use here | |
9314 | // in case the device is misclassified and user wants | |
9315 | // to overridely reset... | |
9316 | ||
11fdf7f2 | 9317 | err = newcrush.remove_device_class(cct, osd, &ss); |
c07f9fc5 FG |
9318 | if (err < 0) { |
9319 | // ss has reason for failure | |
9320 | goto reply; | |
9321 | } | |
9322 | updated.insert(osd); | |
9323 | } | |
9324 | } | |
9325 | ||
9326 | if (!updated.empty()) { | |
9327 | pending_inc.crush.clear(); | |
9328 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9329 | ss << "done removing class of osd(s): " << updated; | |
9330 | getline(ss, rs); | |
9331 | wait_for_finished_proposal(op, | |
9332 | new Monitor::C_Command(mon,op, 0, rs, get_last_committed() + 1)); | |
9333 | return true; | |
9334 | } | |
11fdf7f2 TL |
9335 | } else if (prefix == "osd crush class create") { |
9336 | string device_class; | |
9337 | if (!cmd_getval(g_ceph_context, cmdmap, "class", device_class)) { | |
9338 | err = -EINVAL; // no value! | |
9339 | goto reply; | |
9340 | } | |
9341 | if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) { | |
9342 | ss << "you must complete the upgrade and 'ceph osd require-osd-release " | |
9343 | << "luminous' before using crush device classes"; | |
9344 | err = -EPERM; | |
9345 | goto reply; | |
9346 | } | |
9347 | if (!_have_pending_crush() && | |
9348 | _get_stable_crush().class_exists(device_class)) { | |
9349 | ss << "class '" << device_class << "' already exists"; | |
9350 | goto reply; | |
9351 | } | |
9352 | CrushWrapper newcrush; | |
9353 | _get_pending_crush(newcrush); | |
9354 | if (newcrush.class_exists(device_class)) { | |
9355 | ss << "class '" << device_class << "' already exists"; | |
9356 | goto update; | |
9357 | } | |
9358 | int class_id = newcrush.get_or_create_class_id(device_class); | |
9359 | pending_inc.crush.clear(); | |
9360 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9361 | ss << "created class " << device_class << " with id " << class_id | |
9362 | << " to crush map"; | |
9363 | goto update; | |
9364 | } else if (prefix == "osd crush class rm") { | |
9365 | string device_class; | |
9366 | if (!cmd_getval(g_ceph_context, cmdmap, "class", device_class)) { | |
9367 | err = -EINVAL; // no value! | |
9368 | goto reply; | |
9369 | } | |
9370 | if (osdmap.require_osd_release < CEPH_RELEASE_LUMINOUS) { | |
9371 | ss << "you must complete the upgrade and 'ceph osd require-osd-release " | |
9372 | << "luminous' before using crush device classes"; | |
9373 | err = -EPERM; | |
9374 | goto reply; | |
9375 | } | |
9376 | ||
9377 | if (!osdmap.crush->class_exists(device_class)) { | |
9378 | err = 0; | |
9379 | goto reply; | |
9380 | } | |
9381 | ||
9382 | CrushWrapper newcrush; | |
9383 | _get_pending_crush(newcrush); | |
9384 | if (!newcrush.class_exists(device_class)) { | |
9385 | err = 0; // make command idempotent | |
9386 | goto wait; | |
9387 | } | |
9388 | int class_id = newcrush.get_class_id(device_class); | |
9389 | stringstream ts; | |
9390 | if (newcrush.class_is_in_use(class_id, &ts)) { | |
9391 | err = -EBUSY; | |
9392 | ss << "class '" << device_class << "' " << ts.str(); | |
9393 | goto reply; | |
9394 | } | |
9395 | ||
9396 | // check if class is used by any erasure-code-profiles | |
9397 | mempool::osdmap::map<string,map<string,string>> old_ec_profiles = | |
9398 | osdmap.get_erasure_code_profiles(); | |
9399 | auto ec_profiles = pending_inc.get_erasure_code_profiles(); | |
9400 | #ifdef HAVE_STDLIB_MAP_SPLICING | |
9401 | ec_profiles.merge(old_ec_profiles); | |
9402 | #else | |
9403 | ec_profiles.insert(make_move_iterator(begin(old_ec_profiles)), | |
9404 | make_move_iterator(end(old_ec_profiles))); | |
9405 | #endif | |
9406 | list<string> referenced_by; | |
9407 | for (auto &i: ec_profiles) { | |
9408 | for (auto &j: i.second) { | |
9409 | if ("crush-device-class" == j.first && device_class == j.second) { | |
9410 | referenced_by.push_back(i.first); | |
9411 | } | |
9412 | } | |
9413 | } | |
9414 | if (!referenced_by.empty()) { | |
9415 | err = -EBUSY; | |
9416 | ss << "class '" << device_class | |
9417 | << "' is still referenced by erasure-code-profile(s): " << referenced_by; | |
9418 | goto reply; | |
9419 | } | |
9420 | ||
9421 | set<int> osds; | |
9422 | newcrush.get_devices_by_class(device_class, &osds); | |
9423 | for (auto& p: osds) { | |
9424 | err = newcrush.remove_device_class(g_ceph_context, p, &ss); | |
9425 | if (err < 0) { | |
9426 | // ss has reason for failure | |
9427 | goto reply; | |
9428 | } | |
9429 | } | |
9430 | ||
9431 | if (osds.empty()) { | |
9432 | // empty class, remove directly | |
9433 | err = newcrush.remove_class_name(device_class); | |
9434 | if (err < 0) { | |
9435 | ss << "class '" << device_class << "' cannot be removed '" | |
9436 | << cpp_strerror(err) << "'"; | |
9437 | goto reply; | |
9438 | } | |
9439 | } | |
9440 | ||
9441 | pending_inc.crush.clear(); | |
9442 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9443 | ss << "removed class " << device_class << " with id " << class_id | |
9444 | << " from crush map"; | |
9445 | goto update; | |
35e4c445 FG |
9446 | } else if (prefix == "osd crush class rename") { |
9447 | string srcname, dstname; | |
11fdf7f2 | 9448 | if (!cmd_getval(cct, cmdmap, "srcname", srcname)) { |
35e4c445 FG |
9449 | err = -EINVAL; |
9450 | goto reply; | |
9451 | } | |
11fdf7f2 | 9452 | if (!cmd_getval(cct, cmdmap, "dstname", dstname)) { |
35e4c445 FG |
9453 | err = -EINVAL; |
9454 | goto reply; | |
9455 | } | |
9456 | ||
9457 | CrushWrapper newcrush; | |
9458 | _get_pending_crush(newcrush); | |
181888fb FG |
9459 | if (!newcrush.class_exists(srcname) && newcrush.class_exists(dstname)) { |
9460 | // suppose this is a replay and return success | |
9461 | // so command is idempotent | |
9462 | ss << "already renamed to '" << dstname << "'"; | |
9463 | err = 0; | |
35e4c445 FG |
9464 | goto reply; |
9465 | } | |
c07f9fc5 | 9466 | |
35e4c445 FG |
9467 | err = newcrush.rename_class(srcname, dstname); |
9468 | if (err < 0) { | |
9469 | ss << "fail to rename '" << srcname << "' to '" << dstname << "' : " | |
9470 | << cpp_strerror(err); | |
9471 | goto reply; | |
9472 | } | |
9473 | ||
9474 | pending_inc.crush.clear(); | |
9475 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9476 | ss << "rename class '" << srcname << "' to '" << dstname << "'"; | |
9477 | goto update; | |
7c673cae FG |
9478 | } else if (prefix == "osd crush add-bucket") { |
9479 | // os crush add-bucket <name> <type> | |
9480 | string name, typestr; | |
11fdf7f2 TL |
9481 | vector<string> argvec; |
9482 | cmd_getval(cct, cmdmap, "name", name); | |
9483 | cmd_getval(cct, cmdmap, "type", typestr); | |
9484 | cmd_getval(cct, cmdmap, "args", argvec); | |
9485 | map<string,string> loc; | |
9486 | if (!argvec.empty()) { | |
9487 | CrushWrapper::parse_loc_map(argvec, &loc); | |
9488 | dout(0) << "will create and move bucket '" << name | |
9489 | << "' to location " << loc << dendl; | |
9490 | } | |
7c673cae FG |
9491 | |
9492 | if (!_have_pending_crush() && | |
9493 | _get_stable_crush().name_exists(name)) { | |
9494 | ss << "bucket '" << name << "' already exists"; | |
9495 | goto reply; | |
9496 | } | |
9497 | ||
9498 | CrushWrapper newcrush; | |
9499 | _get_pending_crush(newcrush); | |
9500 | ||
9501 | if (newcrush.name_exists(name)) { | |
9502 | ss << "bucket '" << name << "' already exists"; | |
9503 | goto update; | |
9504 | } | |
9505 | int type = newcrush.get_type_id(typestr); | |
9506 | if (type < 0) { | |
9507 | ss << "type '" << typestr << "' does not exist"; | |
9508 | err = -EINVAL; | |
9509 | goto reply; | |
9510 | } | |
9511 | if (type == 0) { | |
9512 | ss << "type '" << typestr << "' is for devices, not buckets"; | |
9513 | err = -EINVAL; | |
9514 | goto reply; | |
9515 | } | |
9516 | int bucketno; | |
9517 | err = newcrush.add_bucket(0, 0, | |
9518 | CRUSH_HASH_DEFAULT, type, 0, NULL, | |
9519 | NULL, &bucketno); | |
9520 | if (err < 0) { | |
9521 | ss << "add_bucket error: '" << cpp_strerror(err) << "'"; | |
9522 | goto reply; | |
9523 | } | |
9524 | err = newcrush.set_item_name(bucketno, name); | |
9525 | if (err < 0) { | |
9526 | ss << "error setting bucket name to '" << name << "'"; | |
9527 | goto reply; | |
9528 | } | |
9529 | ||
11fdf7f2 TL |
9530 | if (!loc.empty()) { |
9531 | if (!newcrush.check_item_loc(cct, bucketno, loc, | |
9532 | (int *)NULL)) { | |
9533 | err = newcrush.move_bucket(cct, bucketno, loc); | |
9534 | if (err < 0) { | |
9535 | ss << "error moving bucket '" << name << "' to location " << loc; | |
9536 | goto reply; | |
9537 | } | |
9538 | } else { | |
9539 | ss << "no need to move item id " << bucketno << " name '" << name | |
9540 | << "' to location " << loc << " in crush map"; | |
9541 | } | |
9542 | } | |
9543 | ||
7c673cae FG |
9544 | pending_inc.crush.clear(); |
9545 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
11fdf7f2 TL |
9546 | if (loc.empty()) { |
9547 | ss << "added bucket " << name << " type " << typestr | |
9548 | << " to crush map"; | |
9549 | } else { | |
9550 | ss << "added bucket " << name << " type " << typestr | |
9551 | << " to location " << loc; | |
9552 | } | |
7c673cae FG |
9553 | goto update; |
9554 | } else if (prefix == "osd crush rename-bucket") { | |
9555 | string srcname, dstname; | |
11fdf7f2 TL |
9556 | cmd_getval(cct, cmdmap, "srcname", srcname); |
9557 | cmd_getval(cct, cmdmap, "dstname", dstname); | |
7c673cae FG |
9558 | |
9559 | err = crush_rename_bucket(srcname, dstname, &ss); | |
9560 | if (err == -EALREADY) // equivalent to success for idempotency | |
9561 | err = 0; | |
9562 | if (err) | |
9563 | goto reply; | |
9564 | else | |
9565 | goto update; | |
c07f9fc5 FG |
9566 | } else if (prefix == "osd crush weight-set create" || |
9567 | prefix == "osd crush weight-set create-compat") { | |
9568 | CrushWrapper newcrush; | |
9569 | _get_pending_crush(newcrush); | |
9570 | int64_t pool; | |
9571 | int positions; | |
9572 | if (newcrush.has_non_straw2_buckets()) { | |
9573 | ss << "crush map contains one or more bucket(s) that are not straw2"; | |
224ce89b WB |
9574 | err = -EPERM; |
9575 | goto reply; | |
9576 | } | |
c07f9fc5 FG |
9577 | if (prefix == "osd crush weight-set create") { |
9578 | if (osdmap.require_min_compat_client > 0 && | |
9579 | osdmap.require_min_compat_client < CEPH_RELEASE_LUMINOUS) { | |
9580 | ss << "require_min_compat_client " | |
9581 | << ceph_release_name(osdmap.require_min_compat_client) | |
9582 | << " < luminous, which is required for per-pool weight-sets. " | |
9583 | << "Try 'ceph osd set-require-min-compat-client luminous' " | |
9584 | << "before using the new interface"; | |
9585 | err = -EPERM; | |
9586 | goto reply; | |
9587 | } | |
9588 | string poolname, mode; | |
11fdf7f2 | 9589 | cmd_getval(cct, cmdmap, "pool", poolname); |
c07f9fc5 FG |
9590 | pool = osdmap.lookup_pg_pool_name(poolname.c_str()); |
9591 | if (pool < 0) { | |
9592 | ss << "pool '" << poolname << "' not found"; | |
9593 | err = -ENOENT; | |
9594 | goto reply; | |
9595 | } | |
11fdf7f2 | 9596 | cmd_getval(cct, cmdmap, "mode", mode); |
c07f9fc5 FG |
9597 | if (mode != "flat" && mode != "positional") { |
9598 | ss << "unrecognized weight-set mode '" << mode << "'"; | |
9599 | err = -EINVAL; | |
9600 | goto reply; | |
9601 | } | |
9602 | positions = mode == "flat" ? 1 : osdmap.get_pg_pool(pool)->get_size(); | |
9603 | } else { | |
9604 | pool = CrushWrapper::DEFAULT_CHOOSE_ARGS; | |
9605 | positions = 1; | |
224ce89b | 9606 | } |
11fdf7f2 TL |
9607 | if (!newcrush.create_choose_args(pool, positions)) { |
9608 | if (pool == CrushWrapper::DEFAULT_CHOOSE_ARGS) { | |
9609 | ss << "compat weight-set already created"; | |
9610 | } else { | |
9611 | ss << "weight-set for pool '" << osdmap.get_pool_name(pool) | |
9612 | << "' already created"; | |
9613 | } | |
9614 | goto reply; | |
9615 | } | |
c07f9fc5 FG |
9616 | pending_inc.crush.clear(); |
9617 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9618 | goto update; | |
224ce89b | 9619 | |
c07f9fc5 FG |
9620 | } else if (prefix == "osd crush weight-set rm" || |
9621 | prefix == "osd crush weight-set rm-compat") { | |
224ce89b WB |
9622 | CrushWrapper newcrush; |
9623 | _get_pending_crush(newcrush); | |
c07f9fc5 FG |
9624 | int64_t pool; |
9625 | if (prefix == "osd crush weight-set rm") { | |
9626 | string poolname; | |
11fdf7f2 | 9627 | cmd_getval(cct, cmdmap, "pool", poolname); |
c07f9fc5 FG |
9628 | pool = osdmap.lookup_pg_pool_name(poolname.c_str()); |
9629 | if (pool < 0) { | |
9630 | ss << "pool '" << poolname << "' not found"; | |
9631 | err = -ENOENT; | |
9632 | goto reply; | |
9633 | } | |
9634 | } else { | |
9635 | pool = CrushWrapper::DEFAULT_CHOOSE_ARGS; | |
224ce89b | 9636 | } |
c07f9fc5 FG |
9637 | newcrush.rm_choose_args(pool); |
9638 | pending_inc.crush.clear(); | |
9639 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9640 | goto update; | |
224ce89b | 9641 | |
c07f9fc5 FG |
9642 | } else if (prefix == "osd crush weight-set reweight" || |
9643 | prefix == "osd crush weight-set reweight-compat") { | |
9644 | string poolname, item; | |
9645 | vector<double> weight; | |
11fdf7f2 TL |
9646 | cmd_getval(cct, cmdmap, "pool", poolname); |
9647 | cmd_getval(cct, cmdmap, "item", item); | |
9648 | cmd_getval(cct, cmdmap, "weight", weight); | |
c07f9fc5 FG |
9649 | CrushWrapper newcrush; |
9650 | _get_pending_crush(newcrush); | |
9651 | int64_t pool; | |
9652 | if (prefix == "osd crush weight-set reweight") { | |
9653 | pool = osdmap.lookup_pg_pool_name(poolname.c_str()); | |
9654 | if (pool < 0) { | |
9655 | ss << "pool '" << poolname << "' not found"; | |
9656 | err = -ENOENT; | |
9657 | goto reply; | |
9658 | } | |
9659 | if (!newcrush.have_choose_args(pool)) { | |
9660 | ss << "no weight-set for pool '" << poolname << "'"; | |
9661 | err = -ENOENT; | |
9662 | goto reply; | |
9663 | } | |
9664 | auto arg_map = newcrush.choose_args_get(pool); | |
9665 | int positions = newcrush.get_choose_args_positions(arg_map); | |
9666 | if (weight.size() != (size_t)positions) { | |
9667 | ss << "must specify exact " << positions << " weight values"; | |
9668 | err = -EINVAL; | |
9669 | goto reply; | |
9670 | } | |
9671 | } else { | |
9672 | pool = CrushWrapper::DEFAULT_CHOOSE_ARGS; | |
9673 | if (!newcrush.have_choose_args(pool)) { | |
9674 | ss << "no backward-compatible weight-set"; | |
9675 | err = -ENOENT; | |
9676 | goto reply; | |
9677 | } | |
224ce89b | 9678 | } |
c07f9fc5 FG |
9679 | if (!newcrush.name_exists(item)) { |
9680 | ss << "item '" << item << "' does not exist"; | |
9681 | err = -ENOENT; | |
224ce89b WB |
9682 | goto reply; |
9683 | } | |
c07f9fc5 | 9684 | err = newcrush.choose_args_adjust_item_weightf( |
11fdf7f2 | 9685 | cct, |
c07f9fc5 FG |
9686 | newcrush.choose_args_get(pool), |
9687 | newcrush.get_item_id(item), | |
9688 | weight, | |
9689 | &ss); | |
224ce89b | 9690 | if (err < 0) { |
224ce89b WB |
9691 | goto reply; |
9692 | } | |
c07f9fc5 | 9693 | err = 0; |
224ce89b WB |
9694 | pending_inc.crush.clear(); |
9695 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
224ce89b | 9696 | goto update; |
7c673cae FG |
9697 | } else if (osdid_present && |
9698 | (prefix == "osd crush set" || prefix == "osd crush add")) { | |
9699 | // <OsdName> is 'osd.<id>' or '<id>', passed as int64_t id | |
9700 | // osd crush set <OsdName> <weight> <loc1> [<loc2> ...] | |
9701 | // osd crush add <OsdName> <weight> <loc1> [<loc2> ...] | |
9702 | ||
9703 | if (!osdmap.exists(osdid)) { | |
9704 | err = -ENOENT; | |
11fdf7f2 TL |
9705 | ss << osd_name |
9706 | << " does not exist. Create it before updating the crush map"; | |
7c673cae FG |
9707 | goto reply; |
9708 | } | |
9709 | ||
9710 | double weight; | |
11fdf7f2 | 9711 | if (!cmd_getval(cct, cmdmap, "weight", weight)) { |
7c673cae | 9712 | ss << "unable to parse weight value '" |
11fdf7f2 | 9713 | << cmd_vartype_stringify(cmdmap.at("weight")) << "'"; |
7c673cae FG |
9714 | err = -EINVAL; |
9715 | goto reply; | |
9716 | } | |
9717 | ||
9718 | string args; | |
9719 | vector<string> argvec; | |
11fdf7f2 | 9720 | cmd_getval(cct, cmdmap, "args", argvec); |
7c673cae FG |
9721 | map<string,string> loc; |
9722 | CrushWrapper::parse_loc_map(argvec, &loc); | |
9723 | ||
9724 | if (prefix == "osd crush set" | |
9725 | && !_get_stable_crush().item_exists(osdid)) { | |
9726 | err = -ENOENT; | |
11fdf7f2 | 9727 | ss << "unable to set item id " << osdid << " name '" << osd_name |
7c673cae FG |
9728 | << "' weight " << weight << " at location " << loc |
9729 | << ": does not exist"; | |
9730 | goto reply; | |
9731 | } | |
9732 | ||
9733 | dout(5) << "adding/updating crush item id " << osdid << " name '" | |
11fdf7f2 | 9734 | << osd_name << "' weight " << weight << " at location " |
7c673cae FG |
9735 | << loc << dendl; |
9736 | CrushWrapper newcrush; | |
9737 | _get_pending_crush(newcrush); | |
9738 | ||
9739 | string action; | |
9740 | if (prefix == "osd crush set" || | |
11fdf7f2 | 9741 | newcrush.check_item_loc(cct, osdid, loc, (int *)NULL)) { |
7c673cae | 9742 | action = "set"; |
11fdf7f2 | 9743 | err = newcrush.update_item(cct, osdid, weight, osd_name, loc); |
7c673cae FG |
9744 | } else { |
9745 | action = "add"; | |
11fdf7f2 | 9746 | err = newcrush.insert_item(cct, osdid, weight, osd_name, loc); |
7c673cae FG |
9747 | if (err == 0) |
9748 | err = 1; | |
9749 | } | |
9750 | ||
9751 | if (err < 0) | |
9752 | goto reply; | |
9753 | ||
9754 | if (err == 0 && !_have_pending_crush()) { | |
11fdf7f2 TL |
9755 | ss << action << " item id " << osdid << " name '" << osd_name |
9756 | << "' weight " << weight << " at location " << loc << ": no change"; | |
7c673cae FG |
9757 | goto reply; |
9758 | } | |
9759 | ||
9760 | pending_inc.crush.clear(); | |
9761 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
11fdf7f2 TL |
9762 | ss << action << " item id " << osdid << " name '" << osd_name << "' weight " |
9763 | << weight << " at location " << loc << " to crush map"; | |
7c673cae FG |
9764 | getline(ss, rs); |
9765 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
9766 | get_last_committed() + 1)); | |
9767 | return true; | |
9768 | ||
9769 | } else if (prefix == "osd crush create-or-move") { | |
9770 | do { | |
9771 | // osd crush create-or-move <OsdName> <initial_weight> <loc1> [<loc2> ...] | |
9772 | if (!osdmap.exists(osdid)) { | |
9773 | err = -ENOENT; | |
11fdf7f2 TL |
9774 | ss << osd_name |
9775 | << " does not exist. create it before updating the crush map"; | |
7c673cae FG |
9776 | goto reply; |
9777 | } | |
9778 | ||
9779 | double weight; | |
11fdf7f2 | 9780 | if (!cmd_getval(cct, cmdmap, "weight", weight)) { |
7c673cae | 9781 | ss << "unable to parse weight value '" |
11fdf7f2 | 9782 | << cmd_vartype_stringify(cmdmap.at("weight")) << "'"; |
7c673cae FG |
9783 | err = -EINVAL; |
9784 | goto reply; | |
9785 | } | |
9786 | ||
9787 | string args; | |
9788 | vector<string> argvec; | |
11fdf7f2 | 9789 | cmd_getval(cct, cmdmap, "args", argvec); |
7c673cae FG |
9790 | map<string,string> loc; |
9791 | CrushWrapper::parse_loc_map(argvec, &loc); | |
9792 | ||
11fdf7f2 TL |
9793 | dout(0) << "create-or-move crush item name '" << osd_name |
9794 | << "' initial_weight " << weight << " at location " << loc | |
9795 | << dendl; | |
7c673cae FG |
9796 | |
9797 | CrushWrapper newcrush; | |
9798 | _get_pending_crush(newcrush); | |
9799 | ||
11fdf7f2 TL |
9800 | err = newcrush.create_or_move_item(cct, osdid, weight, osd_name, loc, |
9801 | g_conf()->osd_crush_update_weight_set); | |
7c673cae | 9802 | if (err == 0) { |
11fdf7f2 TL |
9803 | ss << "create-or-move updated item name '" << osd_name |
9804 | << "' weight " << weight | |
7c673cae FG |
9805 | << " at location " << loc << " to crush map"; |
9806 | break; | |
9807 | } | |
9808 | if (err > 0) { | |
9809 | pending_inc.crush.clear(); | |
9810 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
11fdf7f2 TL |
9811 | ss << "create-or-move updating item name '" << osd_name |
9812 | << "' weight " << weight | |
7c673cae FG |
9813 | << " at location " << loc << " to crush map"; |
9814 | getline(ss, rs); | |
9815 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
9816 | get_last_committed() + 1)); | |
9817 | return true; | |
9818 | } | |
9819 | } while (false); | |
9820 | ||
9821 | } else if (prefix == "osd crush move") { | |
9822 | do { | |
9823 | // osd crush move <name> <loc1> [<loc2> ...] | |
11fdf7f2 | 9824 | string name; |
7c673cae | 9825 | vector<string> argvec; |
11fdf7f2 TL |
9826 | cmd_getval(cct, cmdmap, "name", name); |
9827 | cmd_getval(cct, cmdmap, "args", argvec); | |
7c673cae FG |
9828 | map<string,string> loc; |
9829 | CrushWrapper::parse_loc_map(argvec, &loc); | |
9830 | ||
9831 | dout(0) << "moving crush item name '" << name << "' to location " << loc << dendl; | |
9832 | CrushWrapper newcrush; | |
9833 | _get_pending_crush(newcrush); | |
9834 | ||
9835 | if (!newcrush.name_exists(name)) { | |
9836 | err = -ENOENT; | |
9837 | ss << "item " << name << " does not exist"; | |
9838 | break; | |
9839 | } | |
9840 | int id = newcrush.get_item_id(name); | |
9841 | ||
11fdf7f2 | 9842 | if (!newcrush.check_item_loc(cct, id, loc, (int *)NULL)) { |
7c673cae | 9843 | if (id >= 0) { |
11fdf7f2 TL |
9844 | err = newcrush.create_or_move_item( |
9845 | cct, id, 0, name, loc, | |
9846 | g_conf()->osd_crush_update_weight_set); | |
7c673cae | 9847 | } else { |
11fdf7f2 | 9848 | err = newcrush.move_bucket(cct, id, loc); |
7c673cae FG |
9849 | } |
9850 | if (err >= 0) { | |
9851 | ss << "moved item id " << id << " name '" << name << "' to location " << loc << " in crush map"; | |
9852 | pending_inc.crush.clear(); | |
9853 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9854 | getline(ss, rs); | |
9855 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
9856 | get_last_committed() + 1)); | |
9857 | return true; | |
9858 | } | |
9859 | } else { | |
9860 | ss << "no need to move item id " << id << " name '" << name << "' to location " << loc << " in crush map"; | |
9861 | err = 0; | |
9862 | } | |
9863 | } while (false); | |
31f18b77 | 9864 | } else if (prefix == "osd crush swap-bucket") { |
11fdf7f2 TL |
9865 | string source, dest; |
9866 | cmd_getval(cct, cmdmap, "source", source); | |
9867 | cmd_getval(cct, cmdmap, "dest", dest); | |
9868 | ||
9869 | bool force = false; | |
9870 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", force); | |
9871 | ||
31f18b77 FG |
9872 | CrushWrapper newcrush; |
9873 | _get_pending_crush(newcrush); | |
9874 | if (!newcrush.name_exists(source)) { | |
9875 | ss << "source item " << source << " does not exist"; | |
9876 | err = -ENOENT; | |
9877 | goto reply; | |
9878 | } | |
9879 | if (!newcrush.name_exists(dest)) { | |
9880 | ss << "dest item " << dest << " does not exist"; | |
9881 | err = -ENOENT; | |
9882 | goto reply; | |
9883 | } | |
9884 | int sid = newcrush.get_item_id(source); | |
9885 | int did = newcrush.get_item_id(dest); | |
9886 | int sparent; | |
11fdf7f2 | 9887 | if (newcrush.get_immediate_parent_id(sid, &sparent) == 0 && !force) { |
31f18b77 FG |
9888 | ss << "source item " << source << " is not an orphan bucket; pass --yes-i-really-mean-it to proceed anyway"; |
9889 | err = -EPERM; | |
9890 | goto reply; | |
9891 | } | |
9892 | if (newcrush.get_bucket_alg(sid) != newcrush.get_bucket_alg(did) && | |
11fdf7f2 | 9893 | !force) { |
31f18b77 FG |
9894 | ss << "source bucket alg " << crush_alg_name(newcrush.get_bucket_alg(sid)) << " != " |
9895 | << "dest bucket alg " << crush_alg_name(newcrush.get_bucket_alg(did)) | |
9896 | << "; pass --yes-i-really-mean-it to proceed anyway"; | |
9897 | err = -EPERM; | |
9898 | goto reply; | |
9899 | } | |
11fdf7f2 | 9900 | int r = newcrush.swap_bucket(cct, sid, did); |
31f18b77 FG |
9901 | if (r < 0) { |
9902 | ss << "failed to swap bucket contents: " << cpp_strerror(r); | |
224ce89b | 9903 | err = r; |
31f18b77 FG |
9904 | goto reply; |
9905 | } | |
9906 | ss << "swapped bucket of " << source << " to " << dest; | |
9907 | pending_inc.crush.clear(); | |
9908 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9909 | wait_for_finished_proposal(op, | |
9910 | new Monitor::C_Command(mon, op, err, ss.str(), | |
9911 | get_last_committed() + 1)); | |
9912 | return true; | |
9913 | } else if (prefix == "osd crush link") { | |
9914 | // osd crush link <name> <loc1> [<loc2> ...] | |
9915 | string name; | |
11fdf7f2 | 9916 | cmd_getval(cct, cmdmap, "name", name); |
31f18b77 | 9917 | vector<string> argvec; |
11fdf7f2 | 9918 | cmd_getval(cct, cmdmap, "args", argvec); |
31f18b77 FG |
9919 | map<string,string> loc; |
9920 | CrushWrapper::parse_loc_map(argvec, &loc); | |
9921 | ||
9922 | // Need an explicit check for name_exists because get_item_id returns | |
9923 | // 0 on unfound. | |
9924 | int id = osdmap.crush->get_item_id(name); | |
7c673cae FG |
9925 | if (!osdmap.crush->name_exists(name)) { |
9926 | err = -ENOENT; | |
9927 | ss << "item " << name << " does not exist"; | |
9928 | goto reply; | |
9929 | } else { | |
9930 | dout(5) << "resolved crush name '" << name << "' to id " << id << dendl; | |
9931 | } | |
11fdf7f2 | 9932 | if (osdmap.crush->check_item_loc(cct, id, loc, (int*) NULL)) { |
7c673cae FG |
9933 | ss << "no need to move item id " << id << " name '" << name |
9934 | << "' to location " << loc << " in crush map"; | |
9935 | err = 0; | |
9936 | goto reply; | |
9937 | } | |
9938 | ||
9939 | dout(5) << "linking crush item name '" << name << "' at location " << loc << dendl; | |
9940 | CrushWrapper newcrush; | |
9941 | _get_pending_crush(newcrush); | |
9942 | ||
9943 | if (!newcrush.name_exists(name)) { | |
9944 | err = -ENOENT; | |
9945 | ss << "item " << name << " does not exist"; | |
9946 | goto reply; | |
9947 | } else { | |
9948 | int id = newcrush.get_item_id(name); | |
11fdf7f2 TL |
9949 | if (!newcrush.check_item_loc(cct, id, loc, (int *)NULL)) { |
9950 | err = newcrush.link_bucket(cct, id, loc); | |
7c673cae FG |
9951 | if (err >= 0) { |
9952 | ss << "linked item id " << id << " name '" << name | |
9953 | << "' to location " << loc << " in crush map"; | |
9954 | pending_inc.crush.clear(); | |
9955 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
9956 | } else { | |
9957 | ss << "cannot link item id " << id << " name '" << name | |
9958 | << "' to location " << loc; | |
9959 | goto reply; | |
9960 | } | |
9961 | } else { | |
9962 | ss << "no need to move item id " << id << " name '" << name | |
9963 | << "' to location " << loc << " in crush map"; | |
9964 | err = 0; | |
9965 | } | |
9966 | } | |
9967 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, ss.str(), | |
9968 | get_last_committed() + 1)); | |
9969 | return true; | |
9970 | } else if (prefix == "osd crush rm" || | |
9971 | prefix == "osd crush remove" || | |
9972 | prefix == "osd crush unlink") { | |
9973 | do { | |
9974 | // osd crush rm <id> [ancestor] | |
9975 | CrushWrapper newcrush; | |
9976 | _get_pending_crush(newcrush); | |
9977 | ||
9978 | string name; | |
11fdf7f2 | 9979 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae FG |
9980 | |
9981 | if (!osdmap.crush->name_exists(name)) { | |
9982 | err = 0; | |
9983 | ss << "device '" << name << "' does not appear in the crush map"; | |
9984 | break; | |
9985 | } | |
9986 | if (!newcrush.name_exists(name)) { | |
9987 | err = 0; | |
9988 | ss << "device '" << name << "' does not appear in the crush map"; | |
9989 | getline(ss, rs); | |
9990 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
9991 | get_last_committed() + 1)); | |
9992 | return true; | |
9993 | } | |
9994 | int id = newcrush.get_item_id(name); | |
31f18b77 FG |
9995 | int ancestor = 0; |
9996 | ||
7c673cae FG |
9997 | bool unlink_only = prefix == "osd crush unlink"; |
9998 | string ancestor_str; | |
11fdf7f2 | 9999 | if (cmd_getval(cct, cmdmap, "ancestor", ancestor_str)) { |
7c673cae FG |
10000 | if (!newcrush.name_exists(ancestor_str)) { |
10001 | err = -ENOENT; | |
10002 | ss << "ancestor item '" << ancestor_str | |
10003 | << "' does not appear in the crush map"; | |
10004 | break; | |
10005 | } | |
31f18b77 | 10006 | ancestor = newcrush.get_item_id(ancestor_str); |
7c673cae | 10007 | } |
31f18b77 FG |
10008 | |
10009 | err = prepare_command_osd_crush_remove( | |
10010 | newcrush, | |
10011 | id, ancestor, | |
10012 | (ancestor < 0), unlink_only); | |
10013 | ||
7c673cae FG |
10014 | if (err == -ENOENT) { |
10015 | ss << "item " << id << " does not appear in that position"; | |
10016 | err = 0; | |
10017 | break; | |
10018 | } | |
10019 | if (err == 0) { | |
81eedcae TL |
10020 | if (!unlink_only) |
10021 | pending_inc.new_crush_node_flags[id] = 0; | |
7c673cae FG |
10022 | ss << "removed item id " << id << " name '" << name << "' from crush map"; |
10023 | getline(ss, rs); | |
10024 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10025 | get_last_committed() + 1)); | |
10026 | return true; | |
10027 | } | |
10028 | } while (false); | |
10029 | ||
10030 | } else if (prefix == "osd crush reweight-all") { | |
7c673cae FG |
10031 | CrushWrapper newcrush; |
10032 | _get_pending_crush(newcrush); | |
10033 | ||
11fdf7f2 | 10034 | newcrush.reweight(cct); |
7c673cae FG |
10035 | pending_inc.crush.clear(); |
10036 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10037 | ss << "reweighted crush hierarchy"; | |
10038 | getline(ss, rs); | |
10039 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10040 | get_last_committed() + 1)); | |
10041 | return true; | |
10042 | } else if (prefix == "osd crush reweight") { | |
10043 | // osd crush reweight <name> <weight> | |
10044 | CrushWrapper newcrush; | |
10045 | _get_pending_crush(newcrush); | |
10046 | ||
10047 | string name; | |
11fdf7f2 | 10048 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae FG |
10049 | if (!newcrush.name_exists(name)) { |
10050 | err = -ENOENT; | |
10051 | ss << "device '" << name << "' does not appear in the crush map"; | |
10052 | goto reply; | |
10053 | } | |
10054 | ||
10055 | int id = newcrush.get_item_id(name); | |
10056 | if (id < 0) { | |
10057 | ss << "device '" << name << "' is not a leaf in the crush map"; | |
10058 | err = -EINVAL; | |
10059 | goto reply; | |
10060 | } | |
10061 | double w; | |
11fdf7f2 | 10062 | if (!cmd_getval(cct, cmdmap, "weight", w)) { |
7c673cae | 10063 | ss << "unable to parse weight value '" |
11fdf7f2 | 10064 | << cmd_vartype_stringify(cmdmap.at("weight")) << "'"; |
7c673cae FG |
10065 | err = -EINVAL; |
10066 | goto reply; | |
10067 | } | |
10068 | ||
11fdf7f2 TL |
10069 | err = newcrush.adjust_item_weightf(cct, id, w, |
10070 | g_conf()->osd_crush_update_weight_set); | |
7c673cae FG |
10071 | if (err < 0) |
10072 | goto reply; | |
10073 | pending_inc.crush.clear(); | |
10074 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10075 | ss << "reweighted item id " << id << " name '" << name << "' to " << w | |
10076 | << " in crush map"; | |
10077 | getline(ss, rs); | |
10078 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10079 | get_last_committed() + 1)); | |
10080 | return true; | |
10081 | } else if (prefix == "osd crush reweight-subtree") { | |
10082 | // osd crush reweight <name> <weight> | |
10083 | CrushWrapper newcrush; | |
10084 | _get_pending_crush(newcrush); | |
10085 | ||
10086 | string name; | |
11fdf7f2 | 10087 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae FG |
10088 | if (!newcrush.name_exists(name)) { |
10089 | err = -ENOENT; | |
10090 | ss << "device '" << name << "' does not appear in the crush map"; | |
10091 | goto reply; | |
10092 | } | |
10093 | ||
10094 | int id = newcrush.get_item_id(name); | |
10095 | if (id >= 0) { | |
10096 | ss << "device '" << name << "' is not a subtree in the crush map"; | |
10097 | err = -EINVAL; | |
10098 | goto reply; | |
10099 | } | |
10100 | double w; | |
11fdf7f2 | 10101 | if (!cmd_getval(cct, cmdmap, "weight", w)) { |
7c673cae | 10102 | ss << "unable to parse weight value '" |
11fdf7f2 | 10103 | << cmd_vartype_stringify(cmdmap.at("weight")) << "'"; |
7c673cae FG |
10104 | err = -EINVAL; |
10105 | goto reply; | |
10106 | } | |
10107 | ||
11fdf7f2 TL |
10108 | err = newcrush.adjust_subtree_weightf(cct, id, w, |
10109 | g_conf()->osd_crush_update_weight_set); | |
7c673cae FG |
10110 | if (err < 0) |
10111 | goto reply; | |
10112 | pending_inc.crush.clear(); | |
10113 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10114 | ss << "reweighted subtree id " << id << " name '" << name << "' to " << w | |
10115 | << " in crush map"; | |
10116 | getline(ss, rs); | |
10117 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10118 | get_last_committed() + 1)); | |
10119 | return true; | |
10120 | } else if (prefix == "osd crush tunables") { | |
10121 | CrushWrapper newcrush; | |
10122 | _get_pending_crush(newcrush); | |
10123 | ||
10124 | err = 0; | |
10125 | string profile; | |
11fdf7f2 | 10126 | cmd_getval(cct, cmdmap, "profile", profile); |
7c673cae FG |
10127 | if (profile == "legacy" || profile == "argonaut") { |
10128 | newcrush.set_tunables_legacy(); | |
10129 | } else if (profile == "bobtail") { | |
10130 | newcrush.set_tunables_bobtail(); | |
10131 | } else if (profile == "firefly") { | |
10132 | newcrush.set_tunables_firefly(); | |
10133 | } else if (profile == "hammer") { | |
10134 | newcrush.set_tunables_hammer(); | |
10135 | } else if (profile == "jewel") { | |
10136 | newcrush.set_tunables_jewel(); | |
10137 | } else if (profile == "optimal") { | |
10138 | newcrush.set_tunables_optimal(); | |
10139 | } else if (profile == "default") { | |
10140 | newcrush.set_tunables_default(); | |
10141 | } else { | |
10142 | ss << "unrecognized profile '" << profile << "'"; | |
10143 | err = -EINVAL; | |
10144 | goto reply; | |
10145 | } | |
10146 | ||
10147 | if (!validate_crush_against_features(&newcrush, ss)) { | |
10148 | err = -EINVAL; | |
10149 | goto reply; | |
10150 | } | |
10151 | ||
10152 | pending_inc.crush.clear(); | |
10153 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10154 | ss << "adjusted tunables profile to " << profile; | |
10155 | getline(ss, rs); | |
10156 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10157 | get_last_committed() + 1)); | |
10158 | return true; | |
10159 | } else if (prefix == "osd crush set-tunable") { | |
10160 | CrushWrapper newcrush; | |
10161 | _get_pending_crush(newcrush); | |
10162 | ||
10163 | err = 0; | |
10164 | string tunable; | |
11fdf7f2 | 10165 | cmd_getval(cct, cmdmap, "tunable", tunable); |
7c673cae FG |
10166 | |
10167 | int64_t value = -1; | |
11fdf7f2 | 10168 | if (!cmd_getval(cct, cmdmap, "value", value)) { |
7c673cae | 10169 | err = -EINVAL; |
11fdf7f2 TL |
10170 | ss << "failed to parse integer value " |
10171 | << cmd_vartype_stringify(cmdmap.at("value")); | |
7c673cae FG |
10172 | goto reply; |
10173 | } | |
10174 | ||
10175 | if (tunable == "straw_calc_version") { | |
224ce89b | 10176 | if (value != 0 && value != 1) { |
7c673cae FG |
10177 | ss << "value must be 0 or 1; got " << value; |
10178 | err = -EINVAL; | |
10179 | goto reply; | |
10180 | } | |
10181 | newcrush.set_straw_calc_version(value); | |
10182 | } else { | |
10183 | ss << "unrecognized tunable '" << tunable << "'"; | |
10184 | err = -EINVAL; | |
10185 | goto reply; | |
10186 | } | |
10187 | ||
10188 | if (!validate_crush_against_features(&newcrush, ss)) { | |
10189 | err = -EINVAL; | |
10190 | goto reply; | |
10191 | } | |
10192 | ||
10193 | pending_inc.crush.clear(); | |
10194 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10195 | ss << "adjusted tunable " << tunable << " to " << value; | |
10196 | getline(ss, rs); | |
10197 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10198 | get_last_committed() + 1)); | |
10199 | return true; | |
10200 | ||
10201 | } else if (prefix == "osd crush rule create-simple") { | |
10202 | string name, root, type, mode; | |
11fdf7f2 TL |
10203 | cmd_getval(cct, cmdmap, "name", name); |
10204 | cmd_getval(cct, cmdmap, "root", root); | |
10205 | cmd_getval(cct, cmdmap, "type", type); | |
10206 | cmd_getval(cct, cmdmap, "mode", mode); | |
7c673cae FG |
10207 | if (mode == "") |
10208 | mode = "firstn"; | |
10209 | ||
10210 | if (osdmap.crush->rule_exists(name)) { | |
31f18b77 FG |
10211 | // The name is uniquely associated to a ruleid and the rule it contains |
10212 | // From the user point of view, the rule is more meaningfull. | |
10213 | ss << "rule " << name << " already exists"; | |
7c673cae FG |
10214 | err = 0; |
10215 | goto reply; | |
10216 | } | |
10217 | ||
10218 | CrushWrapper newcrush; | |
10219 | _get_pending_crush(newcrush); | |
10220 | ||
10221 | if (newcrush.rule_exists(name)) { | |
31f18b77 FG |
10222 | // The name is uniquely associated to a ruleid and the rule it contains |
10223 | // From the user point of view, the rule is more meaningfull. | |
10224 | ss << "rule " << name << " already exists"; | |
7c673cae FG |
10225 | err = 0; |
10226 | } else { | |
224ce89b | 10227 | int ruleno = newcrush.add_simple_rule(name, root, type, "", mode, |
7c673cae FG |
10228 | pg_pool_t::TYPE_REPLICATED, &ss); |
10229 | if (ruleno < 0) { | |
10230 | err = ruleno; | |
10231 | goto reply; | |
10232 | } | |
10233 | ||
10234 | pending_inc.crush.clear(); | |
10235 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10236 | } | |
10237 | getline(ss, rs); | |
10238 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10239 | get_last_committed() + 1)); | |
10240 | return true; | |
10241 | ||
224ce89b WB |
10242 | } else if (prefix == "osd crush rule create-replicated") { |
10243 | string name, root, type, device_class; | |
11fdf7f2 TL |
10244 | cmd_getval(cct, cmdmap, "name", name); |
10245 | cmd_getval(cct, cmdmap, "root", root); | |
10246 | cmd_getval(cct, cmdmap, "type", type); | |
10247 | cmd_getval(cct, cmdmap, "class", device_class); | |
224ce89b WB |
10248 | |
10249 | if (osdmap.crush->rule_exists(name)) { | |
10250 | // The name is uniquely associated to a ruleid and the rule it contains | |
10251 | // From the user point of view, the rule is more meaningfull. | |
10252 | ss << "rule " << name << " already exists"; | |
10253 | err = 0; | |
10254 | goto reply; | |
10255 | } | |
10256 | ||
10257 | CrushWrapper newcrush; | |
10258 | _get_pending_crush(newcrush); | |
10259 | ||
10260 | if (newcrush.rule_exists(name)) { | |
10261 | // The name is uniquely associated to a ruleid and the rule it contains | |
10262 | // From the user point of view, the rule is more meaningfull. | |
10263 | ss << "rule " << name << " already exists"; | |
10264 | err = 0; | |
10265 | } else { | |
10266 | int ruleno = newcrush.add_simple_rule( | |
10267 | name, root, type, device_class, | |
10268 | "firstn", pg_pool_t::TYPE_REPLICATED, &ss); | |
10269 | if (ruleno < 0) { | |
10270 | err = ruleno; | |
10271 | goto reply; | |
10272 | } | |
10273 | ||
10274 | pending_inc.crush.clear(); | |
10275 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10276 | } | |
10277 | getline(ss, rs); | |
10278 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10279 | get_last_committed() + 1)); | |
10280 | return true; | |
10281 | ||
7c673cae FG |
10282 | } else if (prefix == "osd erasure-code-profile rm") { |
10283 | string name; | |
11fdf7f2 | 10284 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae FG |
10285 | |
10286 | if (erasure_code_profile_in_use(pending_inc.new_pools, name, &ss)) | |
10287 | goto wait; | |
10288 | ||
10289 | if (erasure_code_profile_in_use(osdmap.pools, name, &ss)) { | |
10290 | err = -EBUSY; | |
10291 | goto reply; | |
10292 | } | |
10293 | ||
10294 | if (osdmap.has_erasure_code_profile(name) || | |
10295 | pending_inc.new_erasure_code_profiles.count(name)) { | |
10296 | if (osdmap.has_erasure_code_profile(name)) { | |
10297 | pending_inc.old_erasure_code_profiles.push_back(name); | |
10298 | } else { | |
10299 | dout(20) << "erasure code profile rm " << name << ": creation canceled" << dendl; | |
10300 | pending_inc.new_erasure_code_profiles.erase(name); | |
10301 | } | |
10302 | ||
10303 | getline(ss, rs); | |
10304 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10305 | get_last_committed() + 1)); | |
10306 | return true; | |
10307 | } else { | |
10308 | ss << "erasure-code-profile " << name << " does not exist"; | |
10309 | err = 0; | |
10310 | goto reply; | |
10311 | } | |
10312 | ||
10313 | } else if (prefix == "osd erasure-code-profile set") { | |
10314 | string name; | |
11fdf7f2 | 10315 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae | 10316 | vector<string> profile; |
11fdf7f2 TL |
10317 | cmd_getval(cct, cmdmap, "profile", profile); |
10318 | ||
10319 | bool force = false; | |
10320 | cmd_getval(cct, cmdmap, "force", force); | |
10321 | ||
7c673cae FG |
10322 | map<string,string> profile_map; |
10323 | err = parse_erasure_code_profile(profile, &profile_map, &ss); | |
10324 | if (err) | |
10325 | goto reply; | |
10326 | if (profile_map.find("plugin") == profile_map.end()) { | |
10327 | ss << "erasure-code-profile " << profile_map | |
10328 | << " must contain a plugin entry" << std::endl; | |
10329 | err = -EINVAL; | |
10330 | goto reply; | |
10331 | } | |
10332 | string plugin = profile_map["plugin"]; | |
10333 | ||
10334 | if (pending_inc.has_erasure_code_profile(name)) { | |
10335 | dout(20) << "erasure code profile " << name << " try again" << dendl; | |
10336 | goto wait; | |
10337 | } else { | |
7c673cae FG |
10338 | err = normalize_profile(name, profile_map, force, &ss); |
10339 | if (err) | |
10340 | goto reply; | |
10341 | ||
10342 | if (osdmap.has_erasure_code_profile(name)) { | |
10343 | ErasureCodeProfile existing_profile_map = | |
10344 | osdmap.get_erasure_code_profile(name); | |
10345 | err = normalize_profile(name, existing_profile_map, force, &ss); | |
10346 | if (err) | |
10347 | goto reply; | |
10348 | ||
10349 | if (existing_profile_map == profile_map) { | |
10350 | err = 0; | |
10351 | goto reply; | |
10352 | } | |
10353 | if (!force) { | |
10354 | err = -EPERM; | |
10355 | ss << "will not override erasure code profile " << name | |
10356 | << " because the existing profile " | |
10357 | << existing_profile_map | |
10358 | << " is different from the proposed profile " | |
10359 | << profile_map; | |
10360 | goto reply; | |
10361 | } | |
10362 | } | |
10363 | ||
10364 | dout(20) << "erasure code profile set " << name << "=" | |
10365 | << profile_map << dendl; | |
10366 | pending_inc.set_erasure_code_profile(name, profile_map); | |
10367 | } | |
10368 | ||
10369 | getline(ss, rs); | |
10370 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10371 | get_last_committed() + 1)); | |
10372 | return true; | |
10373 | ||
10374 | } else if (prefix == "osd crush rule create-erasure") { | |
10375 | err = check_cluster_features(CEPH_FEATURE_CRUSH_V2, ss); | |
10376 | if (err == -EAGAIN) | |
10377 | goto wait; | |
10378 | if (err) | |
10379 | goto reply; | |
10380 | string name, poolstr; | |
11fdf7f2 | 10381 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae | 10382 | string profile; |
11fdf7f2 | 10383 | cmd_getval(cct, cmdmap, "profile", profile); |
7c673cae FG |
10384 | if (profile == "") |
10385 | profile = "default"; | |
10386 | if (profile == "default") { | |
10387 | if (!osdmap.has_erasure_code_profile(profile)) { | |
10388 | if (pending_inc.has_erasure_code_profile(profile)) { | |
10389 | dout(20) << "erasure code profile " << profile << " already pending" << dendl; | |
10390 | goto wait; | |
10391 | } | |
10392 | ||
10393 | map<string,string> profile_map; | |
11fdf7f2 | 10394 | err = osdmap.get_erasure_code_profile_default(cct, |
7c673cae FG |
10395 | profile_map, |
10396 | &ss); | |
10397 | if (err) | |
10398 | goto reply; | |
10399 | err = normalize_profile(name, profile_map, true, &ss); | |
10400 | if (err) | |
10401 | goto reply; | |
10402 | dout(20) << "erasure code profile set " << profile << "=" | |
10403 | << profile_map << dendl; | |
10404 | pending_inc.set_erasure_code_profile(profile, profile_map); | |
10405 | goto wait; | |
10406 | } | |
10407 | } | |
10408 | ||
31f18b77 FG |
10409 | int rule; |
10410 | err = crush_rule_create_erasure(name, profile, &rule, &ss); | |
7c673cae FG |
10411 | if (err < 0) { |
10412 | switch(err) { | |
10413 | case -EEXIST: // return immediately | |
10414 | ss << "rule " << name << " already exists"; | |
10415 | err = 0; | |
10416 | goto reply; | |
10417 | break; | |
10418 | case -EALREADY: // wait for pending to be proposed | |
10419 | ss << "rule " << name << " already exists"; | |
10420 | err = 0; | |
10421 | break; | |
10422 | default: // non recoverable error | |
10423 | goto reply; | |
10424 | break; | |
10425 | } | |
10426 | } else { | |
31f18b77 | 10427 | ss << "created rule " << name << " at " << rule; |
7c673cae FG |
10428 | } |
10429 | ||
10430 | getline(ss, rs); | |
10431 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10432 | get_last_committed() + 1)); | |
10433 | return true; | |
10434 | ||
10435 | } else if (prefix == "osd crush rule rm") { | |
10436 | string name; | |
11fdf7f2 | 10437 | cmd_getval(cct, cmdmap, "name", name); |
7c673cae FG |
10438 | |
10439 | if (!osdmap.crush->rule_exists(name)) { | |
10440 | ss << "rule " << name << " does not exist"; | |
10441 | err = 0; | |
10442 | goto reply; | |
10443 | } | |
10444 | ||
10445 | CrushWrapper newcrush; | |
10446 | _get_pending_crush(newcrush); | |
10447 | ||
10448 | if (!newcrush.rule_exists(name)) { | |
10449 | ss << "rule " << name << " does not exist"; | |
10450 | err = 0; | |
10451 | } else { | |
10452 | int ruleno = newcrush.get_rule_id(name); | |
11fdf7f2 | 10453 | ceph_assert(ruleno >= 0); |
7c673cae FG |
10454 | |
10455 | // make sure it is not in use. | |
10456 | // FIXME: this is ok in some situations, but let's not bother with that | |
10457 | // complexity now. | |
10458 | int ruleset = newcrush.get_rule_mask_ruleset(ruleno); | |
3efd9988 | 10459 | if (osdmap.crush_rule_in_use(ruleset)) { |
7c673cae FG |
10460 | ss << "crush ruleset " << name << " " << ruleset << " is in use"; |
10461 | err = -EBUSY; | |
10462 | goto reply; | |
10463 | } | |
10464 | ||
10465 | err = newcrush.remove_rule(ruleno); | |
10466 | if (err < 0) { | |
10467 | goto reply; | |
10468 | } | |
10469 | ||
10470 | pending_inc.crush.clear(); | |
10471 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10472 | } | |
10473 | getline(ss, rs); | |
10474 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10475 | get_last_committed() + 1)); | |
10476 | return true; | |
10477 | ||
b5b8bbf5 FG |
10478 | } else if (prefix == "osd crush rule rename") { |
10479 | string srcname; | |
10480 | string dstname; | |
11fdf7f2 TL |
10481 | cmd_getval(cct, cmdmap, "srcname", srcname); |
10482 | cmd_getval(cct, cmdmap, "dstname", dstname); | |
b5b8bbf5 FG |
10483 | if (srcname.empty() || dstname.empty()) { |
10484 | ss << "must specify both source rule name and destination rule name"; | |
10485 | err = -EINVAL; | |
10486 | goto reply; | |
10487 | } | |
10488 | if (srcname == dstname) { | |
10489 | ss << "destination rule name is equal to source rule name"; | |
10490 | err = 0; | |
10491 | goto reply; | |
10492 | } | |
10493 | ||
10494 | CrushWrapper newcrush; | |
10495 | _get_pending_crush(newcrush); | |
181888fb FG |
10496 | if (!newcrush.rule_exists(srcname) && newcrush.rule_exists(dstname)) { |
10497 | // srcname does not exist and dstname already exists | |
10498 | // suppose this is a replay and return success | |
10499 | // (so this command is idempotent) | |
10500 | ss << "already renamed to '" << dstname << "'"; | |
10501 | err = 0; | |
10502 | goto reply; | |
10503 | } | |
10504 | ||
b5b8bbf5 FG |
10505 | err = newcrush.rename_rule(srcname, dstname, &ss); |
10506 | if (err < 0) { | |
10507 | // ss has reason for failure | |
10508 | goto reply; | |
10509 | } | |
10510 | pending_inc.crush.clear(); | |
10511 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
10512 | getline(ss, rs); | |
10513 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10514 | get_last_committed() + 1)); | |
10515 | return true; | |
10516 | ||
7c673cae FG |
10517 | } else if (prefix == "osd setmaxosd") { |
10518 | int64_t newmax; | |
11fdf7f2 | 10519 | if (!cmd_getval(cct, cmdmap, "newmax", newmax)) { |
7c673cae | 10520 | ss << "unable to parse 'newmax' value '" |
11fdf7f2 | 10521 | << cmd_vartype_stringify(cmdmap.at("newmax")) << "'"; |
7c673cae FG |
10522 | err = -EINVAL; |
10523 | goto reply; | |
10524 | } | |
10525 | ||
11fdf7f2 | 10526 | if (newmax > g_conf()->mon_max_osd) { |
7c673cae FG |
10527 | err = -ERANGE; |
10528 | ss << "cannot set max_osd to " << newmax << " which is > conf.mon_max_osd (" | |
11fdf7f2 | 10529 | << g_conf()->mon_max_osd << ")"; |
7c673cae FG |
10530 | goto reply; |
10531 | } | |
10532 | ||
10533 | // Don't allow shrinking OSD number as this will cause data loss | |
10534 | // and may cause kernel crashes. | |
10535 | // Note: setmaxosd sets the maximum OSD number and not the number of OSDs | |
10536 | if (newmax < osdmap.get_max_osd()) { | |
10537 | // Check if the OSDs exist between current max and new value. | |
10538 | // If there are any OSDs exist, then don't allow shrinking number | |
10539 | // of OSDs. | |
10540 | for (int i = newmax; i < osdmap.get_max_osd(); i++) { | |
10541 | if (osdmap.exists(i)) { | |
10542 | err = -EBUSY; | |
10543 | ss << "cannot shrink max_osd to " << newmax | |
10544 | << " because osd." << i << " (and possibly others) still in use"; | |
10545 | goto reply; | |
10546 | } | |
10547 | } | |
10548 | } | |
10549 | ||
10550 | pending_inc.new_max_osd = newmax; | |
10551 | ss << "set new max_osd = " << pending_inc.new_max_osd; | |
10552 | getline(ss, rs); | |
10553 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10554 | get_last_committed() + 1)); | |
10555 | return true; | |
10556 | ||
10557 | } else if (prefix == "osd set-full-ratio" || | |
10558 | prefix == "osd set-backfillfull-ratio" || | |
10559 | prefix == "osd set-nearfull-ratio") { | |
7c673cae | 10560 | double n; |
11fdf7f2 | 10561 | if (!cmd_getval(cct, cmdmap, "ratio", n)) { |
7c673cae | 10562 | ss << "unable to parse 'ratio' value '" |
11fdf7f2 | 10563 | << cmd_vartype_stringify(cmdmap.at("ratio")) << "'"; |
7c673cae FG |
10564 | err = -EINVAL; |
10565 | goto reply; | |
10566 | } | |
10567 | if (prefix == "osd set-full-ratio") | |
10568 | pending_inc.new_full_ratio = n; | |
10569 | else if (prefix == "osd set-backfillfull-ratio") | |
10570 | pending_inc.new_backfillfull_ratio = n; | |
10571 | else if (prefix == "osd set-nearfull-ratio") | |
10572 | pending_inc.new_nearfull_ratio = n; | |
10573 | ss << prefix << " " << n; | |
10574 | getline(ss, rs); | |
10575 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10576 | get_last_committed() + 1)); | |
10577 | return true; | |
10578 | } else if (prefix == "osd set-require-min-compat-client") { | |
7c673cae | 10579 | string v; |
11fdf7f2 | 10580 | cmd_getval(cct, cmdmap, "version", v); |
31f18b77 FG |
10581 | int vno = ceph_release_from_name(v.c_str()); |
10582 | if (vno <= 0) { | |
7c673cae FG |
10583 | ss << "version " << v << " is not recognized"; |
10584 | err = -EINVAL; | |
10585 | goto reply; | |
10586 | } | |
10587 | OSDMap newmap; | |
10588 | newmap.deepish_copy_from(osdmap); | |
10589 | newmap.apply_incremental(pending_inc); | |
31f18b77 FG |
10590 | newmap.require_min_compat_client = vno; |
10591 | auto mvno = newmap.get_min_compat_client(); | |
10592 | if (vno < mvno) { | |
10593 | ss << "osdmap current utilizes features that require " | |
10594 | << ceph_release_name(mvno) | |
10595 | << "; cannot set require_min_compat_client below that to " | |
10596 | << ceph_release_name(vno); | |
7c673cae FG |
10597 | err = -EPERM; |
10598 | goto reply; | |
10599 | } | |
11fdf7f2 TL |
10600 | bool sure = false; |
10601 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", sure); | |
10602 | if (!sure) { | |
31f18b77 FG |
10603 | FeatureMap m; |
10604 | mon->get_combined_feature_map(&m); | |
10605 | uint64_t features = ceph_release_features(vno); | |
10606 | bool first = true; | |
10607 | bool ok = true; | |
10608 | for (int type : { | |
10609 | CEPH_ENTITY_TYPE_CLIENT, | |
10610 | CEPH_ENTITY_TYPE_MDS, | |
10611 | CEPH_ENTITY_TYPE_MGR }) { | |
10612 | auto p = m.m.find(type); | |
10613 | if (p == m.m.end()) { | |
10614 | continue; | |
10615 | } | |
10616 | for (auto& q : p->second) { | |
10617 | uint64_t missing = ~q.first & features; | |
10618 | if (missing) { | |
10619 | if (first) { | |
10620 | ss << "cannot set require_min_compat_client to " << v << ": "; | |
10621 | } else { | |
10622 | ss << "; "; | |
10623 | } | |
10624 | first = false; | |
10625 | ss << q.second << " connected " << ceph_entity_type_name(type) | |
10626 | << "(s) look like " << ceph_release_name( | |
10627 | ceph_release_from_features(q.first)) | |
10628 | << " (missing 0x" << std::hex << missing << std::dec << ")"; | |
10629 | ok = false; | |
10630 | } | |
10631 | } | |
10632 | } | |
10633 | if (!ok) { | |
10634 | ss << "; add --yes-i-really-mean-it to do it anyway"; | |
10635 | err = -EPERM; | |
10636 | goto reply; | |
10637 | } | |
10638 | } | |
10639 | ss << "set require_min_compat_client to " << ceph_release_name(vno); | |
10640 | pending_inc.new_require_min_compat_client = vno; | |
7c673cae FG |
10641 | getline(ss, rs); |
10642 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
10643 | get_last_committed() + 1)); | |
10644 | return true; | |
10645 | } else if (prefix == "osd pause") { | |
10646 | return prepare_set_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR); | |
10647 | ||
10648 | } else if (prefix == "osd unpause") { | |
10649 | return prepare_unset_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR); | |
10650 | ||
10651 | } else if (prefix == "osd set") { | |
11fdf7f2 TL |
10652 | bool sure = false; |
10653 | cmd_getval(g_ceph_context, cmdmap, "yes_i_really_mean_it", sure); | |
10654 | ||
7c673cae | 10655 | string key; |
11fdf7f2 | 10656 | cmd_getval(cct, cmdmap, "key", key); |
7c673cae FG |
10657 | if (key == "full") |
10658 | return prepare_set_flag(op, CEPH_OSDMAP_FULL); | |
10659 | else if (key == "pause") | |
10660 | return prepare_set_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR); | |
10661 | else if (key == "noup") | |
10662 | return prepare_set_flag(op, CEPH_OSDMAP_NOUP); | |
10663 | else if (key == "nodown") | |
10664 | return prepare_set_flag(op, CEPH_OSDMAP_NODOWN); | |
10665 | else if (key == "noout") | |
10666 | return prepare_set_flag(op, CEPH_OSDMAP_NOOUT); | |
10667 | else if (key == "noin") | |
10668 | return prepare_set_flag(op, CEPH_OSDMAP_NOIN); | |
10669 | else if (key == "nobackfill") | |
10670 | return prepare_set_flag(op, CEPH_OSDMAP_NOBACKFILL); | |
10671 | else if (key == "norebalance") | |
10672 | return prepare_set_flag(op, CEPH_OSDMAP_NOREBALANCE); | |
10673 | else if (key == "norecover") | |
10674 | return prepare_set_flag(op, CEPH_OSDMAP_NORECOVER); | |
10675 | else if (key == "noscrub") | |
10676 | return prepare_set_flag(op, CEPH_OSDMAP_NOSCRUB); | |
10677 | else if (key == "nodeep-scrub") | |
10678 | return prepare_set_flag(op, CEPH_OSDMAP_NODEEP_SCRUB); | |
10679 | else if (key == "notieragent") | |
10680 | return prepare_set_flag(op, CEPH_OSDMAP_NOTIERAGENT); | |
11fdf7f2 TL |
10681 | else if (key == "nosnaptrim") |
10682 | return prepare_set_flag(op, CEPH_OSDMAP_NOSNAPTRIM); | |
10683 | else if (key == "pglog_hardlimit") { | |
10684 | if (!osdmap.get_num_up_osds() && !sure) { | |
f64942e4 AA |
10685 | ss << "Not advisable to continue since no OSDs are up. Pass " |
10686 | << "--yes-i-really-mean-it if you really wish to continue."; | |
10687 | err = -EPERM; | |
10688 | goto reply; | |
10689 | } | |
10690 | // The release check here is required because for OSD_PGLOG_HARDLIMIT, | |
10691 | // we are reusing a jewel feature bit that was retired in luminous. | |
10692 | if (osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS && | |
10693 | (HAVE_FEATURE(osdmap.get_up_osd_features(), OSD_PGLOG_HARDLIMIT) | |
11fdf7f2 | 10694 | || sure)) { |
f64942e4 AA |
10695 | return prepare_set_flag(op, CEPH_OSDMAP_PGLOG_HARDLIMIT); |
10696 | } else { | |
10697 | ss << "not all up OSDs have OSD_PGLOG_HARDLIMIT feature"; | |
10698 | err = -EPERM; | |
10699 | goto reply; | |
10700 | } | |
7c673cae FG |
10701 | } else { |
10702 | ss << "unrecognized flag '" << key << "'"; | |
10703 | err = -EINVAL; | |
10704 | } | |
10705 | ||
10706 | } else if (prefix == "osd unset") { | |
10707 | string key; | |
11fdf7f2 | 10708 | cmd_getval(cct, cmdmap, "key", key); |
7c673cae FG |
10709 | if (key == "full") |
10710 | return prepare_unset_flag(op, CEPH_OSDMAP_FULL); | |
10711 | else if (key == "pause") | |
10712 | return prepare_unset_flag(op, CEPH_OSDMAP_PAUSERD | CEPH_OSDMAP_PAUSEWR); | |
10713 | else if (key == "noup") | |
10714 | return prepare_unset_flag(op, CEPH_OSDMAP_NOUP); | |
10715 | else if (key == "nodown") | |
10716 | return prepare_unset_flag(op, CEPH_OSDMAP_NODOWN); | |
10717 | else if (key == "noout") | |
10718 | return prepare_unset_flag(op, CEPH_OSDMAP_NOOUT); | |
10719 | else if (key == "noin") | |
10720 | return prepare_unset_flag(op, CEPH_OSDMAP_NOIN); | |
10721 | else if (key == "nobackfill") | |
10722 | return prepare_unset_flag(op, CEPH_OSDMAP_NOBACKFILL); | |
10723 | else if (key == "norebalance") | |
10724 | return prepare_unset_flag(op, CEPH_OSDMAP_NOREBALANCE); | |
10725 | else if (key == "norecover") | |
10726 | return prepare_unset_flag(op, CEPH_OSDMAP_NORECOVER); | |
10727 | else if (key == "noscrub") | |
10728 | return prepare_unset_flag(op, CEPH_OSDMAP_NOSCRUB); | |
10729 | else if (key == "nodeep-scrub") | |
10730 | return prepare_unset_flag(op, CEPH_OSDMAP_NODEEP_SCRUB); | |
10731 | else if (key == "notieragent") | |
10732 | return prepare_unset_flag(op, CEPH_OSDMAP_NOTIERAGENT); | |
11fdf7f2 TL |
10733 | else if (key == "nosnaptrim") |
10734 | return prepare_unset_flag(op, CEPH_OSDMAP_NOSNAPTRIM); | |
224ce89b | 10735 | else { |
7c673cae FG |
10736 | ss << "unrecognized flag '" << key << "'"; |
10737 | err = -EINVAL; | |
10738 | } | |
10739 | ||
31f18b77 FG |
10740 | } else if (prefix == "osd require-osd-release") { |
10741 | string release; | |
11fdf7f2 TL |
10742 | cmd_getval(cct, cmdmap, "release", release); |
10743 | bool sure = false; | |
10744 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", sure); | |
31f18b77 FG |
10745 | int rel = ceph_release_from_name(release.c_str()); |
10746 | if (rel <= 0) { | |
10747 | ss << "unrecognized release " << release; | |
10748 | err = -EINVAL; | |
10749 | goto reply; | |
10750 | } | |
d2e6a577 FG |
10751 | if (rel == osdmap.require_osd_release) { |
10752 | // idempotent | |
10753 | err = 0; | |
10754 | goto reply; | |
10755 | } | |
11fdf7f2 TL |
10756 | ceph_assert(osdmap.require_osd_release >= CEPH_RELEASE_LUMINOUS); |
10757 | if (!osdmap.get_num_up_osds() && !sure) { | |
10758 | ss << "Not advisable to continue since no OSDs are up. Pass " | |
10759 | << "--yes-i-really-mean-it if you really wish to continue."; | |
10760 | err = -EPERM; | |
10761 | goto reply; | |
10762 | } | |
10763 | if (rel == CEPH_RELEASE_MIMIC) { | |
10764 | if (!mon->monmap->get_required_features().contains_all( | |
10765 | ceph::features::mon::FEATURE_MIMIC)) { | |
10766 | ss << "not all mons are mimic"; | |
10767 | err = -EPERM; | |
10768 | goto reply; | |
10769 | } | |
10770 | if ((!HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_MIMIC)) | |
10771 | && !sure) { | |
10772 | ss << "not all up OSDs have CEPH_FEATURE_SERVER_MIMIC feature"; | |
10773 | err = -EPERM; | |
10774 | goto reply; | |
10775 | } | |
10776 | } else if (rel == CEPH_RELEASE_NAUTILUS) { | |
10777 | if (!mon->monmap->get_required_features().contains_all( | |
10778 | ceph::features::mon::FEATURE_NAUTILUS)) { | |
10779 | ss << "not all mons are nautilus"; | |
10780 | err = -EPERM; | |
10781 | goto reply; | |
10782 | } | |
10783 | if ((!HAVE_FEATURE(osdmap.get_up_osd_features(), SERVER_NAUTILUS)) | |
10784 | && !sure) { | |
10785 | ss << "not all up OSDs have CEPH_FEATURE_SERVER_NAUTILUS feature"; | |
31f18b77 FG |
10786 | err = -EPERM; |
10787 | goto reply; | |
10788 | } | |
10789 | } else { | |
10790 | ss << "not supported for this release yet"; | |
10791 | err = -EPERM; | |
10792 | goto reply; | |
10793 | } | |
10794 | if (rel < osdmap.require_osd_release) { | |
10795 | ss << "require_osd_release cannot be lowered once it has been set"; | |
10796 | err = -EPERM; | |
10797 | goto reply; | |
10798 | } | |
10799 | pending_inc.new_require_osd_release = rel; | |
10800 | goto update; | |
7c673cae FG |
10801 | } else if (prefix == "osd down" || |
10802 | prefix == "osd out" || | |
10803 | prefix == "osd in" || | |
10804 | prefix == "osd rm") { | |
10805 | ||
10806 | bool any = false; | |
31f18b77 FG |
10807 | bool stop = false; |
10808 | bool verbose = true; | |
7c673cae FG |
10809 | |
10810 | vector<string> idvec; | |
11fdf7f2 | 10811 | cmd_getval(cct, cmdmap, "ids", idvec); |
31f18b77 FG |
10812 | for (unsigned j = 0; j < idvec.size() && !stop; j++) { |
10813 | set<int> osds; | |
10814 | ||
10815 | // wildcard? | |
10816 | if (j == 0 && | |
10817 | (idvec[0] == "any" || idvec[0] == "all" || idvec[0] == "*")) { | |
10818 | if (prefix == "osd in") { | |
10819 | // touch out osds only | |
81eedcae | 10820 | osdmap.get_out_existing_osds(osds); |
31f18b77 FG |
10821 | } else { |
10822 | osdmap.get_all_osds(osds); | |
10823 | } | |
10824 | stop = true; | |
10825 | verbose = false; // so the output is less noisy. | |
10826 | } else { | |
10827 | long osd = parse_osd_id(idvec[j].c_str(), &ss); | |
10828 | if (osd < 0) { | |
10829 | ss << "invalid osd id" << osd; | |
10830 | err = -EINVAL; | |
10831 | continue; | |
10832 | } else if (!osdmap.exists(osd)) { | |
10833 | ss << "osd." << osd << " does not exist. "; | |
10834 | continue; | |
10835 | } | |
10836 | ||
10837 | osds.insert(osd); | |
7c673cae | 10838 | } |
31f18b77 FG |
10839 | |
10840 | for (auto &osd : osds) { | |
10841 | if (prefix == "osd down") { | |
10842 | if (osdmap.is_down(osd)) { | |
10843 | if (verbose) | |
10844 | ss << "osd." << osd << " is already down. "; | |
10845 | } else { | |
10846 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_UP); | |
10847 | ss << "marked down osd." << osd << ". "; | |
10848 | any = true; | |
10849 | } | |
10850 | } else if (prefix == "osd out") { | |
10851 | if (osdmap.is_out(osd)) { | |
10852 | if (verbose) | |
10853 | ss << "osd." << osd << " is already out. "; | |
10854 | } else { | |
10855 | pending_inc.new_weight[osd] = CEPH_OSD_OUT; | |
10856 | if (osdmap.osd_weight[osd]) { | |
10857 | if (pending_inc.new_xinfo.count(osd) == 0) { | |
10858 | pending_inc.new_xinfo[osd] = osdmap.osd_xinfo[osd]; | |
10859 | } | |
10860 | pending_inc.new_xinfo[osd].old_weight = osdmap.osd_weight[osd]; | |
7c673cae | 10861 | } |
31f18b77 | 10862 | ss << "marked out osd." << osd << ". "; |
224ce89b WB |
10863 | std::ostringstream msg; |
10864 | msg << "Client " << op->get_session()->entity_name | |
10865 | << " marked osd." << osd << " out"; | |
10866 | if (osdmap.is_up(osd)) { | |
10867 | msg << ", while it was still marked up"; | |
10868 | } else { | |
3efd9988 FG |
10869 | auto period = ceph_clock_now() - down_pending_out[osd]; |
10870 | msg << ", after it was down for " << int(period.sec()) | |
224ce89b WB |
10871 | << " seconds"; |
10872 | } | |
10873 | ||
10874 | mon->clog->info() << msg.str(); | |
31f18b77 | 10875 | any = true; |
7c673cae | 10876 | } |
31f18b77 FG |
10877 | } else if (prefix == "osd in") { |
10878 | if (osdmap.is_in(osd)) { | |
10879 | if (verbose) | |
10880 | ss << "osd." << osd << " is already in. "; | |
10881 | } else { | |
10882 | if (osdmap.osd_xinfo[osd].old_weight > 0) { | |
10883 | pending_inc.new_weight[osd] = osdmap.osd_xinfo[osd].old_weight; | |
10884 | if (pending_inc.new_xinfo.count(osd) == 0) { | |
10885 | pending_inc.new_xinfo[osd] = osdmap.osd_xinfo[osd]; | |
10886 | } | |
10887 | pending_inc.new_xinfo[osd].old_weight = 0; | |
10888 | } else { | |
10889 | pending_inc.new_weight[osd] = CEPH_OSD_IN; | |
7c673cae | 10890 | } |
31f18b77 FG |
10891 | ss << "marked in osd." << osd << ". "; |
10892 | any = true; | |
10893 | } | |
10894 | } else if (prefix == "osd rm") { | |
10895 | err = prepare_command_osd_remove(osd); | |
10896 | ||
10897 | if (err == -EBUSY) { | |
10898 | if (any) | |
10899 | ss << ", "; | |
10900 | ss << "osd." << osd << " is still up; must be down before removal. "; | |
7c673cae | 10901 | } else { |
11fdf7f2 | 10902 | ceph_assert(err == 0); |
31f18b77 FG |
10903 | if (any) { |
10904 | ss << ", osd." << osd; | |
10905 | } else { | |
10906 | ss << "removed osd." << osd; | |
10907 | } | |
10908 | any = true; | |
7c673cae | 10909 | } |
31f18b77 FG |
10910 | } |
10911 | } | |
10912 | } | |
10913 | if (any) { | |
10914 | getline(ss, rs); | |
10915 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, rs, | |
10916 | get_last_committed() + 1)); | |
10917 | return true; | |
10918 | } | |
81eedcae TL |
10919 | } else if (prefix == "osd set-group" || |
10920 | prefix == "osd unset-group" || | |
10921 | prefix == "osd add-noup" || | |
31f18b77 FG |
10922 | prefix == "osd add-nodown" || |
10923 | prefix == "osd add-noin" || | |
81eedcae TL |
10924 | prefix == "osd add-noout" || |
10925 | prefix == "osd rm-noup" || | |
10926 | prefix == "osd rm-nodown" || | |
10927 | prefix == "osd rm-noin" || | |
10928 | prefix == "osd rm-noout") { | |
10929 | bool do_set = prefix == "osd set-group" || | |
10930 | prefix.find("add") != string::npos; | |
10931 | string flag_str; | |
10932 | unsigned flags = 0; | |
10933 | vector<string> who; | |
10934 | if (prefix == "osd set-group" || prefix == "osd unset-group") { | |
10935 | cmd_getval(cct, cmdmap, "flags", flag_str); | |
10936 | cmd_getval(cct, cmdmap, "who", who); | |
10937 | vector<string> raw_flags; | |
10938 | boost::split(raw_flags, flag_str, boost::is_any_of(",")); | |
10939 | for (auto& f : raw_flags) { | |
10940 | if (f == "noup") | |
10941 | flags |= CEPH_OSD_NOUP; | |
10942 | else if (f == "nodown") | |
10943 | flags |= CEPH_OSD_NODOWN; | |
10944 | else if (f == "noin") | |
10945 | flags |= CEPH_OSD_NOIN; | |
10946 | else if (f == "noout") | |
10947 | flags |= CEPH_OSD_NOOUT; | |
10948 | else { | |
10949 | ss << "unrecognized flag '" << f << "', must be one of " | |
10950 | << "{noup,nodown,noin,noout}"; | |
10951 | err = -EINVAL; | |
10952 | goto reply; | |
10953 | } | |
10954 | } | |
31f18b77 | 10955 | } else { |
81eedcae TL |
10956 | cmd_getval(cct, cmdmap, "ids", who); |
10957 | if (prefix.find("noup") != string::npos) | |
10958 | flags = CEPH_OSD_NOUP; | |
10959 | else if (prefix.find("nodown") != string::npos) | |
10960 | flags = CEPH_OSD_NODOWN; | |
10961 | else if (prefix.find("noin") != string::npos) | |
10962 | flags = CEPH_OSD_NOIN; | |
10963 | else if (prefix.find("noout") != string::npos) | |
10964 | flags = CEPH_OSD_NOOUT; | |
10965 | else | |
10966 | ceph_assert(0 == "Unreachable!"); | |
31f18b77 | 10967 | } |
81eedcae TL |
10968 | if (flags == 0) { |
10969 | ss << "must specify flag(s) {noup,nodwon,noin,noout} to set/unset"; | |
10970 | err = -EINVAL; | |
10971 | goto reply; | |
10972 | } | |
10973 | if (who.empty()) { | |
10974 | ss << "must specify at least one or more targets to set/unset"; | |
10975 | err = -EINVAL; | |
10976 | goto reply; | |
10977 | } | |
10978 | set<int> osds; | |
10979 | set<int> crush_nodes; | |
10980 | set<int> device_classes; | |
10981 | for (auto& w : who) { | |
10982 | if (w == "any" || w == "all" || w == "*") { | |
31f18b77 | 10983 | osdmap.get_all_osds(osds); |
81eedcae | 10984 | break; |
31f18b77 | 10985 | } |
81eedcae TL |
10986 | std::stringstream ts; |
10987 | if (auto osd = parse_osd_id(w.c_str(), &ts); osd >= 0) { | |
10988 | osds.insert(osd); | |
10989 | } else if (osdmap.crush->name_exists(w)) { | |
10990 | crush_nodes.insert(osdmap.crush->get_item_id(w)); | |
10991 | } else if (osdmap.crush->class_exists(w)) { | |
10992 | device_classes.insert(osdmap.crush->get_class_id(w)); | |
10993 | } else { | |
10994 | ss << "unable to parse osd id or crush node or device class: " | |
10995 | << "\"" << w << "\". "; | |
7c673cae FG |
10996 | } |
10997 | } | |
81eedcae TL |
10998 | if (osds.empty() && crush_nodes.empty() && device_classes.empty()) { |
10999 | // ss has reason for failure | |
11000 | err = -EINVAL; | |
11001 | goto reply; | |
31f18b77 | 11002 | } |
31f18b77 | 11003 | bool any = false; |
81eedcae TL |
11004 | for (auto osd : osds) { |
11005 | if (!osdmap.exists(osd)) { | |
11006 | ss << "osd." << osd << " does not exist. "; | |
11007 | continue; | |
11008 | } | |
11009 | if (do_set) { | |
11010 | if (flags & CEPH_OSD_NOUP) { | |
11011 | any |= osdmap.is_noup_by_osd(osd) ? | |
11012 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOUP) : | |
11013 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOUP); | |
31f18b77 | 11014 | } |
81eedcae TL |
11015 | if (flags & CEPH_OSD_NODOWN) { |
11016 | any |= osdmap.is_nodown_by_osd(osd) ? | |
11017 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NODOWN) : | |
11018 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NODOWN); | |
11019 | } | |
11020 | if (flags & CEPH_OSD_NOIN) { | |
11021 | any |= osdmap.is_noin_by_osd(osd) ? | |
11022 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOIN) : | |
11023 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOIN); | |
11024 | } | |
11025 | if (flags & CEPH_OSD_NOOUT) { | |
11026 | any |= osdmap.is_noout_by_osd(osd) ? | |
11027 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOOUT) : | |
11028 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOOUT); | |
31f18b77 | 11029 | } |
31f18b77 | 11030 | } else { |
81eedcae TL |
11031 | if (flags & CEPH_OSD_NOUP) { |
11032 | any |= osdmap.is_noup_by_osd(osd) ? | |
11033 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOUP) : | |
11034 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOUP); | |
31f18b77 | 11035 | } |
81eedcae TL |
11036 | if (flags & CEPH_OSD_NODOWN) { |
11037 | any |= osdmap.is_nodown_by_osd(osd) ? | |
11038 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NODOWN) : | |
11039 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NODOWN); | |
31f18b77 | 11040 | } |
81eedcae TL |
11041 | if (flags & CEPH_OSD_NOIN) { |
11042 | any |= osdmap.is_noin_by_osd(osd) ? | |
11043 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOIN) : | |
11044 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOIN); | |
11045 | } | |
11046 | if (flags & CEPH_OSD_NOOUT) { | |
11047 | any |= osdmap.is_noout_by_osd(osd) ? | |
11048 | pending_inc.pending_osd_state_set(osd, CEPH_OSD_NOOUT) : | |
11049 | pending_inc.pending_osd_state_clear(osd, CEPH_OSD_NOOUT); | |
31f18b77 FG |
11050 | } |
11051 | } | |
11052 | } | |
81eedcae TL |
11053 | for (auto& id : crush_nodes) { |
11054 | auto old_flags = osdmap.get_crush_node_flags(id); | |
11055 | auto& pending_flags = pending_inc.new_crush_node_flags[id]; | |
11056 | pending_flags |= old_flags; // adopt existing flags first! | |
11057 | if (do_set) { | |
11058 | pending_flags |= flags; | |
11059 | } else { | |
11060 | pending_flags &= ~flags; | |
11061 | } | |
11062 | any = true; | |
11063 | } | |
11064 | for (auto& id : device_classes) { | |
11065 | auto old_flags = osdmap.get_device_class_flags(id); | |
11066 | auto& pending_flags = pending_inc.new_device_class_flags[id]; | |
11067 | pending_flags |= old_flags; | |
11068 | if (do_set) { | |
11069 | pending_flags |= flags; | |
11070 | } else { | |
11071 | pending_flags &= ~flags; | |
11072 | } | |
11073 | any = true; | |
11074 | } | |
31f18b77 FG |
11075 | if (any) { |
11076 | getline(ss, rs); | |
11077 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, err, rs, | |
11078 | get_last_committed() + 1)); | |
7c673cae FG |
11079 | return true; |
11080 | } | |
11081 | } else if (prefix == "osd pg-temp") { | |
11082 | string pgidstr; | |
11fdf7f2 | 11083 | if (!cmd_getval(cct, cmdmap, "pgid", pgidstr)) { |
7c673cae | 11084 | ss << "unable to parse 'pgid' value '" |
11fdf7f2 | 11085 | << cmd_vartype_stringify(cmdmap.at("pgid")) << "'"; |
7c673cae FG |
11086 | err = -EINVAL; |
11087 | goto reply; | |
11088 | } | |
11089 | pg_t pgid; | |
11090 | if (!pgid.parse(pgidstr.c_str())) { | |
11091 | ss << "invalid pgid '" << pgidstr << "'"; | |
11092 | err = -EINVAL; | |
11093 | goto reply; | |
11094 | } | |
11095 | if (!osdmap.pg_exists(pgid)) { | |
11096 | ss << "pg " << pgid << " does not exist"; | |
11097 | err = -ENOENT; | |
11098 | goto reply; | |
11099 | } | |
11100 | if (pending_inc.new_pg_temp.count(pgid)) { | |
11101 | dout(10) << __func__ << " waiting for pending update on " << pgid << dendl; | |
11102 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
11103 | return true; | |
11104 | } | |
11105 | ||
11106 | vector<int64_t> id_vec; | |
11107 | vector<int32_t> new_pg_temp; | |
11fdf7f2 TL |
11108 | cmd_getval(cct, cmdmap, "id", id_vec); |
11109 | if (id_vec.empty()) { | |
11110 | pending_inc.new_pg_temp[pgid] = mempool::osdmap::vector<int>(); | |
11111 | ss << "done cleaning up pg_temp of " << pgid; | |
11112 | goto update; | |
7c673cae FG |
11113 | } |
11114 | for (auto osd : id_vec) { | |
11115 | if (!osdmap.exists(osd)) { | |
11116 | ss << "osd." << osd << " does not exist"; | |
11117 | err = -ENOENT; | |
11118 | goto reply; | |
11119 | } | |
11120 | new_pg_temp.push_back(osd); | |
11121 | } | |
11122 | ||
224ce89b WB |
11123 | int pool_min_size = osdmap.get_pg_pool_min_size(pgid); |
11124 | if ((int)new_pg_temp.size() < pool_min_size) { | |
11125 | ss << "num of osds (" << new_pg_temp.size() <<") < pool min size (" | |
11126 | << pool_min_size << ")"; | |
11127 | err = -EINVAL; | |
11128 | goto reply; | |
11129 | } | |
11130 | ||
11131 | int pool_size = osdmap.get_pg_pool_size(pgid); | |
11132 | if ((int)new_pg_temp.size() > pool_size) { | |
11133 | ss << "num of osds (" << new_pg_temp.size() <<") > pool size (" | |
11134 | << pool_size << ")"; | |
11135 | err = -EINVAL; | |
11136 | goto reply; | |
11137 | } | |
11138 | ||
7c673cae FG |
11139 | pending_inc.new_pg_temp[pgid] = mempool::osdmap::vector<int>( |
11140 | new_pg_temp.begin(), new_pg_temp.end()); | |
11141 | ss << "set " << pgid << " pg_temp mapping to " << new_pg_temp; | |
11142 | goto update; | |
11143 | } else if (prefix == "osd primary-temp") { | |
11144 | string pgidstr; | |
11fdf7f2 | 11145 | if (!cmd_getval(cct, cmdmap, "pgid", pgidstr)) { |
7c673cae | 11146 | ss << "unable to parse 'pgid' value '" |
11fdf7f2 | 11147 | << cmd_vartype_stringify(cmdmap.at("pgid")) << "'"; |
7c673cae FG |
11148 | err = -EINVAL; |
11149 | goto reply; | |
11150 | } | |
11151 | pg_t pgid; | |
11152 | if (!pgid.parse(pgidstr.c_str())) { | |
11153 | ss << "invalid pgid '" << pgidstr << "'"; | |
11154 | err = -EINVAL; | |
11155 | goto reply; | |
11156 | } | |
11157 | if (!osdmap.pg_exists(pgid)) { | |
11158 | ss << "pg " << pgid << " does not exist"; | |
11159 | err = -ENOENT; | |
11160 | goto reply; | |
11161 | } | |
11162 | ||
11163 | int64_t osd; | |
11fdf7f2 | 11164 | if (!cmd_getval(cct, cmdmap, "id", osd)) { |
7c673cae | 11165 | ss << "unable to parse 'id' value '" |
11fdf7f2 | 11166 | << cmd_vartype_stringify(cmdmap.at("id")) << "'"; |
7c673cae FG |
11167 | err = -EINVAL; |
11168 | goto reply; | |
11169 | } | |
11170 | if (osd != -1 && !osdmap.exists(osd)) { | |
11171 | ss << "osd." << osd << " does not exist"; | |
11172 | err = -ENOENT; | |
11173 | goto reply; | |
11174 | } | |
11175 | ||
31f18b77 FG |
11176 | if (osdmap.require_min_compat_client > 0 && |
11177 | osdmap.require_min_compat_client < CEPH_RELEASE_FIREFLY) { | |
11178 | ss << "require_min_compat_client " | |
11179 | << ceph_release_name(osdmap.require_min_compat_client) | |
7c673cae FG |
11180 | << " < firefly, which is required for primary-temp"; |
11181 | err = -EPERM; | |
11182 | goto reply; | |
7c673cae FG |
11183 | } |
11184 | ||
11185 | pending_inc.new_primary_temp[pgid] = osd; | |
11186 | ss << "set " << pgid << " primary_temp mapping to " << osd; | |
11187 | goto update; | |
11fdf7f2 TL |
11188 | } else if (prefix == "pg repeer") { |
11189 | pg_t pgid; | |
11190 | string pgidstr; | |
11191 | cmd_getval(cct, cmdmap, "pgid", pgidstr); | |
11192 | if (!pgid.parse(pgidstr.c_str())) { | |
11193 | ss << "invalid pgid '" << pgidstr << "'"; | |
11194 | err = -EINVAL; | |
11195 | goto reply; | |
11196 | } | |
11197 | if (!osdmap.pg_exists(pgid)) { | |
11198 | ss << "pg '" << pgidstr << "' does not exist"; | |
11199 | err = -ENOENT; | |
11200 | goto reply; | |
11201 | } | |
11202 | vector<int> acting; | |
11203 | int primary; | |
11204 | osdmap.pg_to_acting_osds(pgid, &acting, &primary); | |
11205 | if (primary < 0) { | |
11206 | err = -EAGAIN; | |
11207 | ss << "pg currently has no primary"; | |
11208 | goto reply; | |
11209 | } | |
11210 | if (acting.size() > 1) { | |
11211 | // map to just primary; it will map back to what it wants | |
11212 | pending_inc.new_pg_temp[pgid] = { primary }; | |
11213 | } else { | |
11214 | // hmm, pick another arbitrary osd to induce a change. Note | |
11215 | // that this won't work if there is only one suitable OSD in the cluster. | |
11216 | int i; | |
11217 | bool done = false; | |
11218 | for (i = 0; i < osdmap.get_max_osd(); ++i) { | |
11219 | if (i == primary || !osdmap.is_up(i) || !osdmap.exists(i)) { | |
11220 | continue; | |
11221 | } | |
11222 | pending_inc.new_pg_temp[pgid] = { primary, i }; | |
11223 | done = true; | |
11224 | break; | |
11225 | } | |
11226 | if (!done) { | |
11227 | err = -EAGAIN; | |
11228 | ss << "not enough up OSDs in the cluster to force repeer"; | |
11229 | goto reply; | |
11230 | } | |
11231 | } | |
11232 | goto update; | |
224ce89b WB |
11233 | } else if (prefix == "osd pg-upmap" || |
11234 | prefix == "osd rm-pg-upmap" || | |
11235 | prefix == "osd pg-upmap-items" || | |
11236 | prefix == "osd rm-pg-upmap-items") { | |
31f18b77 FG |
11237 | if (osdmap.require_min_compat_client < CEPH_RELEASE_LUMINOUS) { |
11238 | ss << "min_compat_client " | |
11239 | << ceph_release_name(osdmap.require_min_compat_client) | |
224ce89b WB |
11240 | << " < luminous, which is required for pg-upmap. " |
11241 | << "Try 'ceph osd set-require-min-compat-client luminous' " | |
11242 | << "before using the new interface"; | |
7c673cae FG |
11243 | err = -EPERM; |
11244 | goto reply; | |
11245 | } | |
11246 | err = check_cluster_features(CEPH_FEATUREMASK_OSDMAP_PG_UPMAP, ss); | |
11247 | if (err == -EAGAIN) | |
11248 | goto wait; | |
11249 | if (err < 0) | |
11250 | goto reply; | |
11251 | string pgidstr; | |
11fdf7f2 | 11252 | if (!cmd_getval(cct, cmdmap, "pgid", pgidstr)) { |
7c673cae | 11253 | ss << "unable to parse 'pgid' value '" |
11fdf7f2 | 11254 | << cmd_vartype_stringify(cmdmap.at("pgid")) << "'"; |
7c673cae FG |
11255 | err = -EINVAL; |
11256 | goto reply; | |
11257 | } | |
11258 | pg_t pgid; | |
11259 | if (!pgid.parse(pgidstr.c_str())) { | |
11260 | ss << "invalid pgid '" << pgidstr << "'"; | |
11261 | err = -EINVAL; | |
11262 | goto reply; | |
11263 | } | |
11264 | if (!osdmap.pg_exists(pgid)) { | |
11265 | ss << "pg " << pgid << " does not exist"; | |
11266 | err = -ENOENT; | |
11267 | goto reply; | |
11268 | } | |
94b18763 FG |
11269 | if (pending_inc.old_pools.count(pgid.pool())) { |
11270 | ss << "pool of " << pgid << " is pending removal"; | |
11271 | err = -ENOENT; | |
11272 | getline(ss, rs); | |
11273 | wait_for_finished_proposal(op, | |
11274 | new Monitor::C_Command(mon, op, err, rs, get_last_committed() + 1)); | |
11275 | return true; | |
11276 | } | |
224ce89b WB |
11277 | |
11278 | enum { | |
11279 | OP_PG_UPMAP, | |
11280 | OP_RM_PG_UPMAP, | |
11281 | OP_PG_UPMAP_ITEMS, | |
11282 | OP_RM_PG_UPMAP_ITEMS, | |
11283 | } option; | |
11284 | ||
11285 | if (prefix == "osd pg-upmap") { | |
11286 | option = OP_PG_UPMAP; | |
11287 | } else if (prefix == "osd rm-pg-upmap") { | |
11288 | option = OP_RM_PG_UPMAP; | |
11289 | } else if (prefix == "osd pg-upmap-items") { | |
11290 | option = OP_PG_UPMAP_ITEMS; | |
11291 | } else { | |
11292 | option = OP_RM_PG_UPMAP_ITEMS; | |
7c673cae | 11293 | } |
224ce89b WB |
11294 | |
11295 | // check pending upmap changes | |
11296 | switch (option) { | |
11297 | case OP_PG_UPMAP: // fall through | |
11298 | case OP_RM_PG_UPMAP: | |
11299 | if (pending_inc.new_pg_upmap.count(pgid) || | |
11300 | pending_inc.old_pg_upmap.count(pgid)) { | |
11301 | dout(10) << __func__ << " waiting for pending update on " | |
11302 | << pgid << dendl; | |
11303 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
11304 | return true; | |
7c673cae | 11305 | } |
224ce89b | 11306 | break; |
7c673cae | 11307 | |
224ce89b WB |
11308 | case OP_PG_UPMAP_ITEMS: // fall through |
11309 | case OP_RM_PG_UPMAP_ITEMS: | |
11310 | if (pending_inc.new_pg_upmap_items.count(pgid) || | |
11311 | pending_inc.old_pg_upmap_items.count(pgid)) { | |
11312 | dout(10) << __func__ << " waiting for pending update on " | |
11313 | << pgid << dendl; | |
11314 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
11315 | return true; | |
11316 | } | |
11317 | break; | |
7c673cae | 11318 | |
224ce89b | 11319 | default: |
11fdf7f2 | 11320 | ceph_abort_msg("invalid option"); |
7c673cae | 11321 | } |
224ce89b WB |
11322 | |
11323 | switch (option) { | |
11324 | case OP_PG_UPMAP: | |
11325 | { | |
11326 | vector<int64_t> id_vec; | |
11fdf7f2 | 11327 | if (!cmd_getval(cct, cmdmap, "id", id_vec)) { |
224ce89b | 11328 | ss << "unable to parse 'id' value(s) '" |
11fdf7f2 | 11329 | << cmd_vartype_stringify(cmdmap.at("id")) << "'"; |
224ce89b WB |
11330 | err = -EINVAL; |
11331 | goto reply; | |
11332 | } | |
11333 | ||
11334 | int pool_min_size = osdmap.get_pg_pool_min_size(pgid); | |
11335 | if ((int)id_vec.size() < pool_min_size) { | |
11336 | ss << "num of osds (" << id_vec.size() <<") < pool min size (" | |
11337 | << pool_min_size << ")"; | |
11338 | err = -EINVAL; | |
11339 | goto reply; | |
11340 | } | |
11341 | ||
11342 | int pool_size = osdmap.get_pg_pool_size(pgid); | |
11343 | if ((int)id_vec.size() > pool_size) { | |
11344 | ss << "num of osds (" << id_vec.size() <<") > pool size (" | |
11345 | << pool_size << ")"; | |
11346 | err = -EINVAL; | |
11347 | goto reply; | |
11348 | } | |
11349 | ||
11350 | vector<int32_t> new_pg_upmap; | |
11351 | for (auto osd : id_vec) { | |
11352 | if (osd != CRUSH_ITEM_NONE && !osdmap.exists(osd)) { | |
11353 | ss << "osd." << osd << " does not exist"; | |
11354 | err = -ENOENT; | |
11355 | goto reply; | |
11356 | } | |
11357 | auto it = std::find(new_pg_upmap.begin(), new_pg_upmap.end(), osd); | |
11358 | if (it != new_pg_upmap.end()) { | |
11359 | ss << "osd." << osd << " already exists, "; | |
11360 | continue; | |
11361 | } | |
11362 | new_pg_upmap.push_back(osd); | |
11363 | } | |
11364 | ||
11365 | if (new_pg_upmap.empty()) { | |
11366 | ss << "no valid upmap items(pairs) is specified"; | |
11367 | err = -EINVAL; | |
11368 | goto reply; | |
11369 | } | |
11370 | ||
11371 | pending_inc.new_pg_upmap[pgid] = mempool::osdmap::vector<int32_t>( | |
11372 | new_pg_upmap.begin(), new_pg_upmap.end()); | |
11373 | ss << "set " << pgid << " pg_upmap mapping to " << new_pg_upmap; | |
7c673cae | 11374 | } |
224ce89b WB |
11375 | break; |
11376 | ||
11377 | case OP_RM_PG_UPMAP: | |
11378 | { | |
11379 | pending_inc.old_pg_upmap.insert(pgid); | |
11380 | ss << "clear " << pgid << " pg_upmap mapping"; | |
7c673cae | 11381 | } |
224ce89b | 11382 | break; |
7c673cae | 11383 | |
224ce89b WB |
11384 | case OP_PG_UPMAP_ITEMS: |
11385 | { | |
11386 | vector<int64_t> id_vec; | |
11fdf7f2 | 11387 | if (!cmd_getval(cct, cmdmap, "id", id_vec)) { |
224ce89b | 11388 | ss << "unable to parse 'id' value(s) '" |
11fdf7f2 | 11389 | << cmd_vartype_stringify(cmdmap.at("id")) << "'"; |
224ce89b WB |
11390 | err = -EINVAL; |
11391 | goto reply; | |
11392 | } | |
11393 | ||
11394 | if (id_vec.size() % 2) { | |
11395 | ss << "you must specify pairs of osd ids to be remapped"; | |
11396 | err = -EINVAL; | |
11397 | goto reply; | |
11398 | } | |
11399 | ||
11400 | int pool_size = osdmap.get_pg_pool_size(pgid); | |
11401 | if ((int)(id_vec.size() / 2) > pool_size) { | |
11402 | ss << "num of osd pairs (" << id_vec.size() / 2 <<") > pool size (" | |
11403 | << pool_size << ")"; | |
11404 | err = -EINVAL; | |
11405 | goto reply; | |
11406 | } | |
11407 | ||
11408 | vector<pair<int32_t,int32_t>> new_pg_upmap_items; | |
11409 | ostringstream items; | |
11410 | items << "["; | |
11411 | for (auto p = id_vec.begin(); p != id_vec.end(); ++p) { | |
11412 | int from = *p++; | |
11413 | int to = *p; | |
11414 | if (from == to) { | |
11415 | ss << "from osd." << from << " == to osd." << to << ", "; | |
11416 | continue; | |
11417 | } | |
11418 | if (!osdmap.exists(from)) { | |
11419 | ss << "osd." << from << " does not exist"; | |
11420 | err = -ENOENT; | |
11421 | goto reply; | |
11422 | } | |
11423 | if (to != CRUSH_ITEM_NONE && !osdmap.exists(to)) { | |
11424 | ss << "osd." << to << " does not exist"; | |
11425 | err = -ENOENT; | |
11426 | goto reply; | |
11427 | } | |
c07f9fc5 FG |
11428 | pair<int32_t,int32_t> entry = make_pair(from, to); |
11429 | auto it = std::find(new_pg_upmap_items.begin(), | |
11430 | new_pg_upmap_items.end(), entry); | |
11431 | if (it != new_pg_upmap_items.end()) { | |
11432 | ss << "osd." << from << " -> osd." << to << " already exists, "; | |
11433 | continue; | |
11434 | } | |
11435 | new_pg_upmap_items.push_back(entry); | |
224ce89b WB |
11436 | items << from << "->" << to << ","; |
11437 | } | |
11438 | string out(items.str()); | |
11439 | out.resize(out.size() - 1); // drop last ',' | |
11440 | out += "]"; | |
11441 | ||
11442 | if (new_pg_upmap_items.empty()) { | |
11443 | ss << "no valid upmap items(pairs) is specified"; | |
11444 | err = -EINVAL; | |
11445 | goto reply; | |
11446 | } | |
11447 | ||
11448 | pending_inc.new_pg_upmap_items[pgid] = | |
11449 | mempool::osdmap::vector<pair<int32_t,int32_t>>( | |
11450 | new_pg_upmap_items.begin(), new_pg_upmap_items.end()); | |
11451 | ss << "set " << pgid << " pg_upmap_items mapping to " << out; | |
11452 | } | |
11453 | break; | |
11454 | ||
11455 | case OP_RM_PG_UPMAP_ITEMS: | |
11456 | { | |
11457 | pending_inc.old_pg_upmap_items.insert(pgid); | |
11458 | ss << "clear " << pgid << " pg_upmap_items mapping"; | |
11459 | } | |
11460 | break; | |
11461 | ||
11462 | default: | |
11fdf7f2 | 11463 | ceph_abort_msg("invalid option"); |
7c673cae FG |
11464 | } |
11465 | ||
7c673cae FG |
11466 | goto update; |
11467 | } else if (prefix == "osd primary-affinity") { | |
11468 | int64_t id; | |
11fdf7f2 | 11469 | if (!cmd_getval(cct, cmdmap, "id", id)) { |
7c673cae | 11470 | ss << "invalid osd id value '" |
11fdf7f2 | 11471 | << cmd_vartype_stringify(cmdmap.at("id")) << "'"; |
7c673cae FG |
11472 | err = -EINVAL; |
11473 | goto reply; | |
11474 | } | |
11475 | double w; | |
11fdf7f2 | 11476 | if (!cmd_getval(cct, cmdmap, "weight", w)) { |
7c673cae | 11477 | ss << "unable to parse 'weight' value '" |
11fdf7f2 | 11478 | << cmd_vartype_stringify(cmdmap.at("weight")) << "'"; |
7c673cae FG |
11479 | err = -EINVAL; |
11480 | goto reply; | |
11481 | } | |
11482 | long ww = (int)((double)CEPH_OSD_MAX_PRIMARY_AFFINITY*w); | |
11483 | if (ww < 0L) { | |
11484 | ss << "weight must be >= 0"; | |
11485 | err = -EINVAL; | |
11486 | goto reply; | |
11487 | } | |
31f18b77 FG |
11488 | if (osdmap.require_min_compat_client > 0 && |
11489 | osdmap.require_min_compat_client < CEPH_RELEASE_FIREFLY) { | |
11490 | ss << "require_min_compat_client " | |
11491 | << ceph_release_name(osdmap.require_min_compat_client) | |
7c673cae FG |
11492 | << " < firefly, which is required for primary-affinity"; |
11493 | err = -EPERM; | |
11494 | goto reply; | |
7c673cae | 11495 | } |
7c673cae FG |
11496 | if (osdmap.exists(id)) { |
11497 | pending_inc.new_primary_affinity[id] = ww; | |
11498 | ss << "set osd." << id << " primary-affinity to " << w << " (" << ios::hex << ww << ios::dec << ")"; | |
11499 | getline(ss, rs); | |
11500 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11501 | get_last_committed() + 1)); | |
11502 | return true; | |
11503 | } else { | |
11504 | ss << "osd." << id << " does not exist"; | |
11505 | err = -ENOENT; | |
11506 | goto reply; | |
11507 | } | |
11508 | } else if (prefix == "osd reweight") { | |
11509 | int64_t id; | |
11fdf7f2 | 11510 | if (!cmd_getval(cct, cmdmap, "id", id)) { |
7c673cae | 11511 | ss << "unable to parse osd id value '" |
11fdf7f2 | 11512 | << cmd_vartype_stringify(cmdmap.at("id")) << "'"; |
7c673cae FG |
11513 | err = -EINVAL; |
11514 | goto reply; | |
11515 | } | |
11516 | double w; | |
11fdf7f2 | 11517 | if (!cmd_getval(cct, cmdmap, "weight", w)) { |
7c673cae | 11518 | ss << "unable to parse weight value '" |
11fdf7f2 | 11519 | << cmd_vartype_stringify(cmdmap.at("weight")) << "'"; |
7c673cae FG |
11520 | err = -EINVAL; |
11521 | goto reply; | |
11522 | } | |
11523 | long ww = (int)((double)CEPH_OSD_IN*w); | |
11524 | if (ww < 0L) { | |
11525 | ss << "weight must be >= 0"; | |
11526 | err = -EINVAL; | |
11527 | goto reply; | |
11528 | } | |
11529 | if (osdmap.exists(id)) { | |
11530 | pending_inc.new_weight[id] = ww; | |
11531 | ss << "reweighted osd." << id << " to " << w << " (" << std::hex << ww << std::dec << ")"; | |
11532 | getline(ss, rs); | |
11533 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11534 | get_last_committed() + 1)); | |
11535 | return true; | |
11536 | } else { | |
11537 | ss << "osd." << id << " does not exist"; | |
11538 | err = -ENOENT; | |
11539 | goto reply; | |
11540 | } | |
11541 | } else if (prefix == "osd reweightn") { | |
11542 | map<int32_t, uint32_t> weights; | |
11fdf7f2 | 11543 | err = parse_reweights(cct, cmdmap, osdmap, &weights); |
7c673cae FG |
11544 | if (err) { |
11545 | ss << "unable to parse 'weights' value '" | |
11fdf7f2 | 11546 | << cmd_vartype_stringify(cmdmap.at("weights")) << "'"; |
7c673cae FG |
11547 | goto reply; |
11548 | } | |
11549 | pending_inc.new_weight.insert(weights.begin(), weights.end()); | |
11550 | wait_for_finished_proposal( | |
11551 | op, | |
11552 | new Monitor::C_Command(mon, op, 0, rs, rdata, get_last_committed() + 1)); | |
224ce89b | 11553 | return true; |
7c673cae FG |
11554 | } else if (prefix == "osd lost") { |
11555 | int64_t id; | |
11fdf7f2 | 11556 | if (!cmd_getval(cct, cmdmap, "id", id)) { |
7c673cae | 11557 | ss << "unable to parse osd id value '" |
11fdf7f2 | 11558 | << cmd_vartype_stringify(cmdmap.at("id")) << "'"; |
7c673cae FG |
11559 | err = -EINVAL; |
11560 | goto reply; | |
11561 | } | |
11fdf7f2 TL |
11562 | bool sure = false; |
11563 | cmd_getval(g_ceph_context, cmdmap, "yes_i_really_mean_it", sure); | |
11564 | if (!sure) { | |
7c673cae FG |
11565 | ss << "are you SURE? this might mean real, permanent data loss. pass " |
11566 | "--yes-i-really-mean-it if you really do."; | |
11567 | err = -EPERM; | |
11568 | goto reply; | |
11569 | } else if (!osdmap.exists(id)) { | |
11570 | ss << "osd." << id << " does not exist"; | |
11571 | err = -ENOENT; | |
11572 | goto reply; | |
11573 | } else if (!osdmap.is_down(id)) { | |
11574 | ss << "osd." << id << " is not down"; | |
11575 | err = -EBUSY; | |
11576 | goto reply; | |
11577 | } else { | |
11578 | epoch_t e = osdmap.get_info(id).down_at; | |
11579 | pending_inc.new_lost[id] = e; | |
11580 | ss << "marked osd lost in epoch " << e; | |
11581 | getline(ss, rs); | |
11582 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11583 | get_last_committed() + 1)); | |
11584 | return true; | |
11585 | } | |
11586 | ||
11fdf7f2 TL |
11587 | } else if (prefix == "osd destroy-actual" || |
11588 | prefix == "osd purge-actual" || | |
11589 | prefix == "osd purge-new") { | |
31f18b77 FG |
11590 | /* Destroying an OSD means that we don't expect to further make use of |
11591 | * the OSDs data (which may even become unreadable after this operation), | |
11592 | * and that we are okay with scrubbing all its cephx keys and config-key | |
11593 | * data (which may include lockbox keys, thus rendering the osd's data | |
11594 | * unreadable). | |
11595 | * | |
11596 | * The OSD will not be removed. Instead, we will mark it as destroyed, | |
11597 | * such that a subsequent call to `create` will not reuse the osd id. | |
11598 | * This will play into being able to recreate the OSD, at the same | |
11599 | * crush location, with minimal data movement. | |
11600 | */ | |
11601 | ||
11602 | // make sure authmon is writeable. | |
11603 | if (!mon->authmon()->is_writeable()) { | |
11604 | dout(10) << __func__ << " waiting for auth mon to be writeable for " | |
11605 | << "osd destroy" << dendl; | |
11606 | mon->authmon()->wait_for_writeable(op, new C_RetryMessage(this, op)); | |
11607 | return false; | |
11608 | } | |
11609 | ||
11610 | int64_t id; | |
11fdf7f2 TL |
11611 | if (!cmd_getval(cct, cmdmap, "id", id)) { |
11612 | auto p = cmdmap.find("id"); | |
11613 | if (p == cmdmap.end()) { | |
11614 | ss << "no osd id specified"; | |
11615 | } else { | |
11616 | ss << "unable to parse osd id value '" | |
11617 | << cmd_vartype_stringify(cmdmap.at("id")) << ""; | |
11618 | } | |
31f18b77 FG |
11619 | err = -EINVAL; |
11620 | goto reply; | |
11621 | } | |
11622 | ||
11fdf7f2 | 11623 | bool is_destroy = (prefix == "osd destroy-actual"); |
31f18b77 | 11624 | if (!is_destroy) { |
11fdf7f2 TL |
11625 | ceph_assert("osd purge-actual" == prefix || |
11626 | "osd purge-new" == prefix); | |
31f18b77 FG |
11627 | } |
11628 | ||
11fdf7f2 TL |
11629 | bool sure = false; |
11630 | cmd_getval(g_ceph_context, cmdmap, "yes_i_really_mean_it", sure); | |
11631 | if (!sure) { | |
11632 | ss << "Are you SURE? Did you verify with 'ceph osd safe-to-destroy'? " | |
11633 | << "This will mean real, permanent data loss, as well " | |
11634 | << "as deletion of cephx and lockbox keys. " | |
11635 | << "Pass --yes-i-really-mean-it if you really do."; | |
31f18b77 FG |
11636 | err = -EPERM; |
11637 | goto reply; | |
d2e6a577 | 11638 | } else if (!osdmap.exists(id)) { |
31f18b77 | 11639 | ss << "osd." << id << " does not exist"; |
d2e6a577 | 11640 | err = 0; // idempotent |
31f18b77 FG |
11641 | goto reply; |
11642 | } else if (osdmap.is_up(id)) { | |
11643 | ss << "osd." << id << " is not `down`."; | |
11644 | err = -EBUSY; | |
11645 | goto reply; | |
11646 | } else if (is_destroy && osdmap.is_destroyed(id)) { | |
11647 | ss << "destroyed osd." << id; | |
11648 | err = 0; | |
11649 | goto reply; | |
11650 | } | |
11651 | ||
11fdf7f2 TL |
11652 | if (prefix == "osd purge-new" && |
11653 | (osdmap.get_state(id) & CEPH_OSD_NEW) == 0) { | |
11654 | ss << "osd." << id << " is not new"; | |
11655 | err = -EPERM; | |
11656 | goto reply; | |
11657 | } | |
11658 | ||
31f18b77 FG |
11659 | bool goto_reply = false; |
11660 | ||
11661 | paxos->plug(); | |
11662 | if (is_destroy) { | |
11663 | err = prepare_command_osd_destroy(id, ss); | |
11664 | // we checked above that it should exist. | |
11fdf7f2 | 11665 | ceph_assert(err != -ENOENT); |
31f18b77 FG |
11666 | } else { |
11667 | err = prepare_command_osd_purge(id, ss); | |
11668 | if (err == -ENOENT) { | |
11669 | err = 0; | |
11670 | ss << "osd." << id << " does not exist."; | |
11671 | goto_reply = true; | |
11672 | } | |
11673 | } | |
11674 | paxos->unplug(); | |
11675 | ||
11676 | if (err < 0 || goto_reply) { | |
11677 | goto reply; | |
11678 | } | |
11679 | ||
11680 | if (is_destroy) { | |
11681 | ss << "destroyed osd." << id; | |
11682 | } else { | |
11683 | ss << "purged osd." << id; | |
11684 | } | |
11685 | ||
11686 | getline(ss, rs); | |
11687 | wait_for_finished_proposal(op, | |
11688 | new Monitor::C_Command(mon, op, 0, rs, get_last_committed() + 1)); | |
11689 | force_immediate_propose(); | |
11690 | return true; | |
11691 | ||
11692 | } else if (prefix == "osd new") { | |
11693 | ||
11694 | // make sure authmon is writeable. | |
11695 | if (!mon->authmon()->is_writeable()) { | |
11696 | dout(10) << __func__ << " waiting for auth mon to be writeable for " | |
224ce89b | 11697 | << "osd new" << dendl; |
31f18b77 FG |
11698 | mon->authmon()->wait_for_writeable(op, new C_RetryMessage(this, op)); |
11699 | return false; | |
11700 | } | |
11701 | ||
3a9019d9 | 11702 | map<string,string> param_map; |
31f18b77 FG |
11703 | |
11704 | bufferlist bl = m->get_data(); | |
3a9019d9 FG |
11705 | string param_json = bl.to_str(); |
11706 | dout(20) << __func__ << " osd new json = " << param_json << dendl; | |
31f18b77 | 11707 | |
3a9019d9 | 11708 | err = get_json_str_map(param_json, ss, ¶m_map); |
31f18b77 FG |
11709 | if (err < 0) |
11710 | goto reply; | |
11711 | ||
3a9019d9 | 11712 | dout(20) << __func__ << " osd new params " << param_map << dendl; |
31f18b77 FG |
11713 | |
11714 | paxos->plug(); | |
3a9019d9 | 11715 | err = prepare_command_osd_new(op, cmdmap, param_map, ss, f.get()); |
31f18b77 FG |
11716 | paxos->unplug(); |
11717 | ||
11718 | if (err < 0) { | |
11719 | goto reply; | |
11720 | } | |
11721 | ||
11722 | if (f) { | |
11723 | f->flush(rdata); | |
11724 | } else { | |
11725 | rdata.append(ss); | |
11726 | } | |
11727 | ||
11728 | if (err == EEXIST) { | |
11729 | // idempotent operation | |
11730 | err = 0; | |
11731 | goto reply; | |
11732 | } | |
11733 | ||
11734 | wait_for_finished_proposal(op, | |
11735 | new Monitor::C_Command(mon, op, 0, rs, rdata, | |
11736 | get_last_committed() + 1)); | |
11737 | force_immediate_propose(); | |
11738 | return true; | |
11739 | ||
7c673cae | 11740 | } else if (prefix == "osd create") { |
7c673cae FG |
11741 | |
11742 | // optional id provided? | |
31f18b77 | 11743 | int64_t id = -1, cmd_id = -1; |
11fdf7f2 | 11744 | if (cmd_getval(cct, cmdmap, "id", cmd_id)) { |
31f18b77 FG |
11745 | if (cmd_id < 0) { |
11746 | ss << "invalid osd id value '" << cmd_id << "'"; | |
7c673cae FG |
11747 | err = -EINVAL; |
11748 | goto reply; | |
11749 | } | |
31f18b77 | 11750 | dout(10) << " osd create got id " << cmd_id << dendl; |
7c673cae FG |
11751 | } |
11752 | ||
7c673cae FG |
11753 | uuid_d uuid; |
11754 | string uuidstr; | |
11fdf7f2 | 11755 | if (cmd_getval(cct, cmdmap, "uuid", uuidstr)) { |
7c673cae | 11756 | if (!uuid.parse(uuidstr.c_str())) { |
31f18b77 FG |
11757 | ss << "invalid uuid value '" << uuidstr << "'"; |
11758 | err = -EINVAL; | |
11759 | goto reply; | |
7c673cae | 11760 | } |
31f18b77 FG |
11761 | // we only care about the id if we also have the uuid, to |
11762 | // ensure the operation's idempotency. | |
11763 | id = cmd_id; | |
7c673cae FG |
11764 | } |
11765 | ||
31f18b77 FG |
11766 | int32_t new_id = -1; |
11767 | err = prepare_command_osd_create(id, uuid, &new_id, ss); | |
11768 | if (err < 0) { | |
11769 | if (err == -EAGAIN) { | |
11770 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
11771 | return true; | |
11772 | } | |
11773 | // a check has failed; reply to the user. | |
11774 | goto reply; | |
11775 | ||
11776 | } else if (err == EEXIST) { | |
11777 | // this is an idempotent operation; we can go ahead and reply. | |
11778 | if (f) { | |
11779 | f->open_object_section("created_osd"); | |
11780 | f->dump_int("osdid", new_id); | |
11781 | f->close_section(); | |
11782 | f->flush(rdata); | |
11783 | } else { | |
11784 | ss << new_id; | |
11785 | rdata.append(ss); | |
7c673cae | 11786 | } |
31f18b77 FG |
11787 | err = 0; |
11788 | goto reply; | |
7c673cae FG |
11789 | } |
11790 | ||
3a9019d9 FG |
11791 | string empty_device_class; |
11792 | do_osd_create(id, uuid, empty_device_class, &new_id); | |
31f18b77 | 11793 | |
7c673cae FG |
11794 | if (f) { |
11795 | f->open_object_section("created_osd"); | |
31f18b77 | 11796 | f->dump_int("osdid", new_id); |
7c673cae FG |
11797 | f->close_section(); |
11798 | f->flush(rdata); | |
11799 | } else { | |
31f18b77 | 11800 | ss << new_id; |
7c673cae FG |
11801 | rdata.append(ss); |
11802 | } | |
31f18b77 FG |
11803 | wait_for_finished_proposal(op, |
11804 | new Monitor::C_Command(mon, op, 0, rs, rdata, | |
11805 | get_last_committed() + 1)); | |
7c673cae FG |
11806 | return true; |
11807 | ||
11808 | } else if (prefix == "osd blacklist clear") { | |
11809 | pending_inc.new_blacklist.clear(); | |
11810 | std::list<std::pair<entity_addr_t,utime_t > > blacklist; | |
11811 | osdmap.get_blacklist(&blacklist); | |
11812 | for (const auto &entry : blacklist) { | |
11813 | pending_inc.old_blacklist.push_back(entry.first); | |
11814 | } | |
11815 | ss << " removed all blacklist entries"; | |
11816 | getline(ss, rs); | |
11817 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11818 | get_last_committed() + 1)); | |
11819 | return true; | |
11820 | } else if (prefix == "osd blacklist") { | |
11821 | string addrstr; | |
11fdf7f2 | 11822 | cmd_getval(cct, cmdmap, "addr", addrstr); |
7c673cae FG |
11823 | entity_addr_t addr; |
11824 | if (!addr.parse(addrstr.c_str(), 0)) { | |
11825 | ss << "unable to parse address " << addrstr; | |
11826 | err = -EINVAL; | |
11827 | goto reply; | |
11828 | } | |
11829 | else { | |
11fdf7f2 TL |
11830 | if (osdmap.require_osd_release >= CEPH_RELEASE_NAUTILUS) { |
11831 | // always blacklist type ANY | |
11832 | addr.set_type(entity_addr_t::TYPE_ANY); | |
11833 | } else { | |
11834 | addr.set_type(entity_addr_t::TYPE_LEGACY); | |
11835 | } | |
11836 | ||
7c673cae | 11837 | string blacklistop; |
11fdf7f2 | 11838 | cmd_getval(cct, cmdmap, "blacklistop", blacklistop); |
7c673cae FG |
11839 | if (blacklistop == "add") { |
11840 | utime_t expires = ceph_clock_now(); | |
11841 | double d; | |
11842 | // default one hour | |
11fdf7f2 TL |
11843 | cmd_getval(cct, cmdmap, "expire", d, |
11844 | g_conf()->mon_osd_blacklist_default_expire); | |
7c673cae FG |
11845 | expires += d; |
11846 | ||
11847 | pending_inc.new_blacklist[addr] = expires; | |
224ce89b WB |
11848 | |
11849 | { | |
11850 | // cancel any pending un-blacklisting request too | |
11851 | auto it = std::find(pending_inc.old_blacklist.begin(), | |
11852 | pending_inc.old_blacklist.end(), addr); | |
11853 | if (it != pending_inc.old_blacklist.end()) { | |
11854 | pending_inc.old_blacklist.erase(it); | |
11855 | } | |
11856 | } | |
11857 | ||
7c673cae FG |
11858 | ss << "blacklisting " << addr << " until " << expires << " (" << d << " sec)"; |
11859 | getline(ss, rs); | |
11860 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11861 | get_last_committed() + 1)); | |
11862 | return true; | |
11863 | } else if (blacklistop == "rm") { | |
11864 | if (osdmap.is_blacklisted(addr) || | |
11865 | pending_inc.new_blacklist.count(addr)) { | |
11866 | if (osdmap.is_blacklisted(addr)) | |
11867 | pending_inc.old_blacklist.push_back(addr); | |
11868 | else | |
11869 | pending_inc.new_blacklist.erase(addr); | |
11870 | ss << "un-blacklisting " << addr; | |
11871 | getline(ss, rs); | |
11872 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11873 | get_last_committed() + 1)); | |
11874 | return true; | |
11875 | } | |
11876 | ss << addr << " isn't blacklisted"; | |
11877 | err = 0; | |
11878 | goto reply; | |
11879 | } | |
11880 | } | |
11881 | } else if (prefix == "osd pool mksnap") { | |
11882 | string poolstr; | |
11fdf7f2 | 11883 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
11884 | int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str()); |
11885 | if (pool < 0) { | |
11886 | ss << "unrecognized pool '" << poolstr << "'"; | |
11887 | err = -ENOENT; | |
11888 | goto reply; | |
11889 | } | |
11890 | string snapname; | |
11fdf7f2 | 11891 | cmd_getval(cct, cmdmap, "snap", snapname); |
7c673cae FG |
11892 | const pg_pool_t *p = osdmap.get_pg_pool(pool); |
11893 | if (p->is_unmanaged_snaps_mode()) { | |
11894 | ss << "pool " << poolstr << " is in unmanaged snaps mode"; | |
11895 | err = -EINVAL; | |
11896 | goto reply; | |
11897 | } else if (p->snap_exists(snapname.c_str())) { | |
11898 | ss << "pool " << poolstr << " snap " << snapname << " already exists"; | |
11899 | err = 0; | |
11900 | goto reply; | |
11901 | } else if (p->is_tier()) { | |
11902 | ss << "pool " << poolstr << " is a cache tier"; | |
11903 | err = -EINVAL; | |
11904 | goto reply; | |
11905 | } | |
11906 | pg_pool_t *pp = 0; | |
11907 | if (pending_inc.new_pools.count(pool)) | |
11908 | pp = &pending_inc.new_pools[pool]; | |
11909 | if (!pp) { | |
11910 | pp = &pending_inc.new_pools[pool]; | |
11911 | *pp = *p; | |
11912 | } | |
11913 | if (pp->snap_exists(snapname.c_str())) { | |
11914 | ss << "pool " << poolstr << " snap " << snapname << " already exists"; | |
11915 | } else { | |
11916 | pp->add_snap(snapname.c_str(), ceph_clock_now()); | |
11917 | pp->set_snap_epoch(pending_inc.epoch); | |
11918 | ss << "created pool " << poolstr << " snap " << snapname; | |
11919 | } | |
11920 | getline(ss, rs); | |
11921 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11922 | get_last_committed() + 1)); | |
11923 | return true; | |
11924 | } else if (prefix == "osd pool rmsnap") { | |
11925 | string poolstr; | |
11fdf7f2 | 11926 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
11927 | int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str()); |
11928 | if (pool < 0) { | |
11929 | ss << "unrecognized pool '" << poolstr << "'"; | |
11930 | err = -ENOENT; | |
11931 | goto reply; | |
11932 | } | |
11933 | string snapname; | |
11fdf7f2 | 11934 | cmd_getval(cct, cmdmap, "snap", snapname); |
7c673cae FG |
11935 | const pg_pool_t *p = osdmap.get_pg_pool(pool); |
11936 | if (p->is_unmanaged_snaps_mode()) { | |
11937 | ss << "pool " << poolstr << " is in unmanaged snaps mode"; | |
11938 | err = -EINVAL; | |
11939 | goto reply; | |
11940 | } else if (!p->snap_exists(snapname.c_str())) { | |
11941 | ss << "pool " << poolstr << " snap " << snapname << " does not exist"; | |
11942 | err = 0; | |
11943 | goto reply; | |
11944 | } | |
11945 | pg_pool_t *pp = 0; | |
11946 | if (pending_inc.new_pools.count(pool)) | |
11947 | pp = &pending_inc.new_pools[pool]; | |
11948 | if (!pp) { | |
11949 | pp = &pending_inc.new_pools[pool]; | |
11950 | *pp = *p; | |
11951 | } | |
11952 | snapid_t sn = pp->snap_exists(snapname.c_str()); | |
11953 | if (sn) { | |
11954 | pp->remove_snap(sn); | |
11955 | pp->set_snap_epoch(pending_inc.epoch); | |
11956 | ss << "removed pool " << poolstr << " snap " << snapname; | |
11957 | } else { | |
11958 | ss << "already removed pool " << poolstr << " snap " << snapname; | |
11959 | } | |
11960 | getline(ss, rs); | |
11961 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
11962 | get_last_committed() + 1)); | |
11963 | return true; | |
11964 | } else if (prefix == "osd pool create") { | |
11fdf7f2 | 11965 | int64_t pg_num, pg_num_min; |
7c673cae | 11966 | int64_t pgp_num; |
11fdf7f2 TL |
11967 | cmd_getval(cct, cmdmap, "pg_num", pg_num, int64_t(0)); |
11968 | cmd_getval(cct, cmdmap, "pgp_num", pgp_num, pg_num); | |
11969 | cmd_getval(cct, cmdmap, "pg_num_min", pg_num_min, int64_t(0)); | |
7c673cae FG |
11970 | |
11971 | string pool_type_str; | |
11fdf7f2 | 11972 | cmd_getval(cct, cmdmap, "pool_type", pool_type_str); |
7c673cae | 11973 | if (pool_type_str.empty()) |
11fdf7f2 | 11974 | pool_type_str = g_conf().get_val<string>("osd_pool_default_type"); |
7c673cae FG |
11975 | |
11976 | string poolstr; | |
11fdf7f2 | 11977 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
11978 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
11979 | if (pool_id >= 0) { | |
11980 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11981 | if (pool_type_str != p->get_type_name()) { | |
11982 | ss << "pool '" << poolstr << "' cannot change to type " << pool_type_str; | |
11983 | err = -EINVAL; | |
11984 | } else { | |
11985 | ss << "pool '" << poolstr << "' already exists"; | |
11986 | err = 0; | |
11987 | } | |
11988 | goto reply; | |
11989 | } | |
11990 | ||
11991 | int pool_type; | |
11992 | if (pool_type_str == "replicated") { | |
11993 | pool_type = pg_pool_t::TYPE_REPLICATED; | |
11994 | } else if (pool_type_str == "erasure") { | |
7c673cae FG |
11995 | pool_type = pg_pool_t::TYPE_ERASURE; |
11996 | } else { | |
11997 | ss << "unknown pool type '" << pool_type_str << "'"; | |
11998 | err = -EINVAL; | |
11999 | goto reply; | |
12000 | } | |
12001 | ||
31f18b77 | 12002 | bool implicit_rule_creation = false; |
94b18763 | 12003 | int64_t expected_num_objects = 0; |
31f18b77 | 12004 | string rule_name; |
11fdf7f2 | 12005 | cmd_getval(cct, cmdmap, "rule", rule_name); |
7c673cae | 12006 | string erasure_code_profile; |
11fdf7f2 | 12007 | cmd_getval(cct, cmdmap, "erasure_code_profile", erasure_code_profile); |
7c673cae FG |
12008 | |
12009 | if (pool_type == pg_pool_t::TYPE_ERASURE) { | |
12010 | if (erasure_code_profile == "") | |
12011 | erasure_code_profile = "default"; | |
12012 | //handle the erasure code profile | |
12013 | if (erasure_code_profile == "default") { | |
12014 | if (!osdmap.has_erasure_code_profile(erasure_code_profile)) { | |
12015 | if (pending_inc.has_erasure_code_profile(erasure_code_profile)) { | |
12016 | dout(20) << "erasure code profile " << erasure_code_profile << " already pending" << dendl; | |
12017 | goto wait; | |
12018 | } | |
12019 | ||
12020 | map<string,string> profile_map; | |
11fdf7f2 | 12021 | err = osdmap.get_erasure_code_profile_default(cct, |
7c673cae FG |
12022 | profile_map, |
12023 | &ss); | |
12024 | if (err) | |
12025 | goto reply; | |
12026 | dout(20) << "erasure code profile " << erasure_code_profile << " set" << dendl; | |
12027 | pending_inc.set_erasure_code_profile(erasure_code_profile, profile_map); | |
12028 | goto wait; | |
12029 | } | |
12030 | } | |
31f18b77 FG |
12031 | if (rule_name == "") { |
12032 | implicit_rule_creation = true; | |
7c673cae | 12033 | if (erasure_code_profile == "default") { |
31f18b77 | 12034 | rule_name = "erasure-code"; |
7c673cae | 12035 | } else { |
31f18b77 | 12036 | dout(1) << "implicitly use rule named after the pool: " |
7c673cae | 12037 | << poolstr << dendl; |
31f18b77 | 12038 | rule_name = poolstr; |
7c673cae FG |
12039 | } |
12040 | } | |
11fdf7f2 | 12041 | cmd_getval(g_ceph_context, cmdmap, "expected_num_objects", |
94b18763 | 12042 | expected_num_objects, int64_t(0)); |
7c673cae | 12043 | } else { |
31f18b77 | 12044 | //NOTE:for replicated pool,cmd_map will put rule_name to erasure_code_profile field |
94b18763 FG |
12045 | // and put expected_num_objects to rule field |
12046 | if (erasure_code_profile != "") { // cmd is from CLI | |
12047 | if (rule_name != "") { | |
12048 | string interr; | |
12049 | expected_num_objects = strict_strtoll(rule_name.c_str(), 10, &interr); | |
12050 | if (interr.length()) { | |
12051 | ss << "error parsing integer value '" << rule_name << "': " << interr; | |
12052 | err = -EINVAL; | |
12053 | goto reply; | |
12054 | } | |
12055 | } | |
12056 | rule_name = erasure_code_profile; | |
12057 | } else { // cmd is well-formed | |
11fdf7f2 | 12058 | cmd_getval(g_ceph_context, cmdmap, "expected_num_objects", |
94b18763 FG |
12059 | expected_num_objects, int64_t(0)); |
12060 | } | |
7c673cae FG |
12061 | } |
12062 | ||
31f18b77 FG |
12063 | if (!implicit_rule_creation && rule_name != "") { |
12064 | int rule; | |
12065 | err = get_crush_rule(rule_name, &rule, &ss); | |
7c673cae FG |
12066 | if (err == -EAGAIN) { |
12067 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12068 | return true; | |
12069 | } | |
12070 | if (err) | |
12071 | goto reply; | |
12072 | } | |
12073 | ||
7c673cae FG |
12074 | if (expected_num_objects < 0) { |
12075 | ss << "'expected_num_objects' must be non-negative"; | |
12076 | err = -EINVAL; | |
12077 | goto reply; | |
12078 | } | |
12079 | ||
91327a77 AA |
12080 | if (expected_num_objects > 0 && |
12081 | cct->_conf->osd_objectstore == "filestore" && | |
12082 | cct->_conf->filestore_merge_threshold > 0) { | |
12083 | ss << "'expected_num_objects' requires 'filestore_merge_threshold < 0'"; | |
12084 | err = -EINVAL; | |
12085 | goto reply; | |
12086 | } | |
12087 | ||
12088 | if (expected_num_objects == 0 && | |
12089 | cct->_conf->osd_objectstore == "filestore" && | |
12090 | cct->_conf->filestore_merge_threshold < 0) { | |
12091 | int osds = osdmap.get_num_osds(); | |
12092 | if (osds && (pg_num >= 1024 || pg_num / osds >= 100)) { | |
12093 | ss << "For better initial performance on pools expected to store a " | |
12094 | << "large number of objects, consider supplying the " | |
12095 | << "expected_num_objects parameter when creating the pool.\n"; | |
12096 | } | |
12097 | } | |
12098 | ||
7c673cae | 12099 | int64_t fast_read_param; |
11fdf7f2 | 12100 | cmd_getval(cct, cmdmap, "fast_read", fast_read_param, int64_t(-1)); |
7c673cae FG |
12101 | FastReadType fast_read = FAST_READ_DEFAULT; |
12102 | if (fast_read_param == 0) | |
12103 | fast_read = FAST_READ_OFF; | |
12104 | else if (fast_read_param > 0) | |
12105 | fast_read = FAST_READ_ON; | |
11fdf7f2 TL |
12106 | |
12107 | int64_t repl_size = 0; | |
12108 | cmd_getval(cct, cmdmap, "size", repl_size); | |
12109 | int64_t target_size_bytes = 0; | |
12110 | double target_size_ratio = 0.0; | |
12111 | cmd_getval(cct, cmdmap, "target_size_bytes", target_size_bytes); | |
12112 | cmd_getval(cct, cmdmap, "target_size_ratio", target_size_ratio); | |
12113 | ||
12114 | err = prepare_new_pool(poolstr, | |
7c673cae | 12115 | -1, // default crush rule |
31f18b77 | 12116 | rule_name, |
11fdf7f2 TL |
12117 | pg_num, pgp_num, pg_num_min, |
12118 | repl_size, target_size_bytes, target_size_ratio, | |
7c673cae FG |
12119 | erasure_code_profile, pool_type, |
12120 | (uint64_t)expected_num_objects, | |
12121 | fast_read, | |
12122 | &ss); | |
12123 | if (err < 0) { | |
12124 | switch(err) { | |
12125 | case -EEXIST: | |
12126 | ss << "pool '" << poolstr << "' already exists"; | |
12127 | break; | |
12128 | case -EAGAIN: | |
12129 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12130 | return true; | |
12131 | case -ERANGE: | |
12132 | goto reply; | |
12133 | default: | |
12134 | goto reply; | |
12135 | break; | |
12136 | } | |
12137 | } else { | |
12138 | ss << "pool '" << poolstr << "' created"; | |
12139 | } | |
12140 | getline(ss, rs); | |
12141 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
12142 | get_last_committed() + 1)); | |
12143 | return true; | |
12144 | ||
12145 | } else if (prefix == "osd pool delete" || | |
12146 | prefix == "osd pool rm") { | |
12147 | // osd pool delete/rm <poolname> <poolname again> --yes-i-really-really-mean-it | |
12148 | string poolstr, poolstr2, sure; | |
11fdf7f2 TL |
12149 | cmd_getval(cct, cmdmap, "pool", poolstr); |
12150 | cmd_getval(cct, cmdmap, "pool2", poolstr2); | |
7c673cae FG |
12151 | int64_t pool = osdmap.lookup_pg_pool_name(poolstr.c_str()); |
12152 | if (pool < 0) { | |
12153 | ss << "pool '" << poolstr << "' does not exist"; | |
12154 | err = 0; | |
12155 | goto reply; | |
12156 | } | |
12157 | ||
11fdf7f2 TL |
12158 | bool force_no_fake = false; |
12159 | cmd_getval(cct, cmdmap, "yes_i_really_really_mean_it", force_no_fake); | |
12160 | bool force = false; | |
12161 | cmd_getval(cct, cmdmap, "yes_i_really_really_mean_it_not_faking", force); | |
7c673cae | 12162 | if (poolstr2 != poolstr || |
11fdf7f2 | 12163 | (!force && !force_no_fake)) { |
7c673cae FG |
12164 | ss << "WARNING: this will *PERMANENTLY DESTROY* all data stored in pool " << poolstr |
12165 | << ". If you are *ABSOLUTELY CERTAIN* that is what you want, pass the pool name *twice*, " | |
12166 | << "followed by --yes-i-really-really-mean-it."; | |
12167 | err = -EPERM; | |
12168 | goto reply; | |
12169 | } | |
12170 | err = _prepare_remove_pool(pool, &ss, force_no_fake); | |
12171 | if (err == -EAGAIN) { | |
12172 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12173 | return true; | |
12174 | } | |
12175 | if (err < 0) | |
12176 | goto reply; | |
12177 | goto update; | |
12178 | } else if (prefix == "osd pool rename") { | |
12179 | string srcpoolstr, destpoolstr; | |
11fdf7f2 TL |
12180 | cmd_getval(cct, cmdmap, "srcpool", srcpoolstr); |
12181 | cmd_getval(cct, cmdmap, "destpool", destpoolstr); | |
7c673cae FG |
12182 | int64_t pool_src = osdmap.lookup_pg_pool_name(srcpoolstr.c_str()); |
12183 | int64_t pool_dst = osdmap.lookup_pg_pool_name(destpoolstr.c_str()); | |
12184 | ||
12185 | if (pool_src < 0) { | |
12186 | if (pool_dst >= 0) { | |
12187 | // src pool doesn't exist, dst pool does exist: to ensure idempotency | |
12188 | // of operations, assume this rename succeeded, as it is not changing | |
12189 | // the current state. Make sure we output something understandable | |
12190 | // for whoever is issuing the command, if they are paying attention, | |
12191 | // in case it was not intentional; or to avoid a "wtf?" and a bug | |
12192 | // report in case it was intentional, while expecting a failure. | |
12193 | ss << "pool '" << srcpoolstr << "' does not exist; pool '" | |
12194 | << destpoolstr << "' does -- assuming successful rename"; | |
12195 | err = 0; | |
12196 | } else { | |
12197 | ss << "unrecognized pool '" << srcpoolstr << "'"; | |
12198 | err = -ENOENT; | |
12199 | } | |
12200 | goto reply; | |
12201 | } else if (pool_dst >= 0) { | |
12202 | // source pool exists and so does the destination pool | |
12203 | ss << "pool '" << destpoolstr << "' already exists"; | |
12204 | err = -EEXIST; | |
12205 | goto reply; | |
12206 | } | |
12207 | ||
12208 | int ret = _prepare_rename_pool(pool_src, destpoolstr); | |
12209 | if (ret == 0) { | |
12210 | ss << "pool '" << srcpoolstr << "' renamed to '" << destpoolstr << "'"; | |
12211 | } else { | |
12212 | ss << "failed to rename pool '" << srcpoolstr << "' to '" << destpoolstr << "': " | |
12213 | << cpp_strerror(ret); | |
12214 | } | |
12215 | getline(ss, rs); | |
12216 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, ret, rs, | |
12217 | get_last_committed() + 1)); | |
12218 | return true; | |
12219 | ||
12220 | } else if (prefix == "osd pool set") { | |
12221 | err = prepare_command_pool_set(cmdmap, ss); | |
12222 | if (err == -EAGAIN) | |
12223 | goto wait; | |
12224 | if (err < 0) | |
12225 | goto reply; | |
12226 | ||
12227 | getline(ss, rs); | |
12228 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
12229 | get_last_committed() + 1)); | |
12230 | return true; | |
12231 | } else if (prefix == "osd tier add") { | |
12232 | err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss); | |
12233 | if (err == -EAGAIN) | |
12234 | goto wait; | |
12235 | if (err) | |
12236 | goto reply; | |
12237 | string poolstr; | |
11fdf7f2 | 12238 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12239 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12240 | if (pool_id < 0) { | |
12241 | ss << "unrecognized pool '" << poolstr << "'"; | |
12242 | err = -ENOENT; | |
12243 | goto reply; | |
12244 | } | |
12245 | string tierpoolstr; | |
11fdf7f2 | 12246 | cmd_getval(cct, cmdmap, "tierpool", tierpoolstr); |
7c673cae FG |
12247 | int64_t tierpool_id = osdmap.lookup_pg_pool_name(tierpoolstr); |
12248 | if (tierpool_id < 0) { | |
12249 | ss << "unrecognized pool '" << tierpoolstr << "'"; | |
12250 | err = -ENOENT; | |
12251 | goto reply; | |
12252 | } | |
12253 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11fdf7f2 | 12254 | ceph_assert(p); |
7c673cae | 12255 | const pg_pool_t *tp = osdmap.get_pg_pool(tierpool_id); |
11fdf7f2 | 12256 | ceph_assert(tp); |
7c673cae FG |
12257 | |
12258 | if (!_check_become_tier(tierpool_id, tp, pool_id, p, &err, &ss)) { | |
12259 | goto reply; | |
12260 | } | |
12261 | ||
12262 | // make sure new tier is empty | |
12263 | string force_nonempty; | |
11fdf7f2 TL |
12264 | cmd_getval(cct, cmdmap, "force_nonempty", force_nonempty); |
12265 | const pool_stat_t *pstats = mon->mgrstatmon()->get_pool_stat(tierpool_id); | |
31f18b77 | 12266 | if (pstats && pstats->stats.sum.num_objects != 0 && |
7c673cae FG |
12267 | force_nonempty != "--force-nonempty") { |
12268 | ss << "tier pool '" << tierpoolstr << "' is not empty; --force-nonempty to force"; | |
12269 | err = -ENOTEMPTY; | |
12270 | goto reply; | |
12271 | } | |
11fdf7f2 | 12272 | if (tp->is_erasure()) { |
7c673cae FG |
12273 | ss << "tier pool '" << tierpoolstr |
12274 | << "' is an ec pool, which cannot be a tier"; | |
12275 | err = -ENOTSUP; | |
12276 | goto reply; | |
12277 | } | |
12278 | if ((!tp->removed_snaps.empty() || !tp->snaps.empty()) && | |
12279 | ((force_nonempty != "--force-nonempty") || | |
11fdf7f2 | 12280 | (!g_conf()->mon_debug_unsafe_allow_tier_with_nonempty_snaps))) { |
7c673cae FG |
12281 | ss << "tier pool '" << tierpoolstr << "' has snapshot state; it cannot be added as a tier without breaking the pool"; |
12282 | err = -ENOTEMPTY; | |
12283 | goto reply; | |
12284 | } | |
12285 | // go | |
12286 | pg_pool_t *np = pending_inc.get_new_pool(pool_id, p); | |
12287 | pg_pool_t *ntp = pending_inc.get_new_pool(tierpool_id, tp); | |
12288 | if (np->tiers.count(tierpool_id) || ntp->is_tier()) { | |
12289 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12290 | return true; | |
12291 | } | |
12292 | np->tiers.insert(tierpool_id); | |
12293 | np->set_snap_epoch(pending_inc.epoch); // tier will update to our snap info | |
12294 | ntp->tier_of = pool_id; | |
12295 | ss << "pool '" << tierpoolstr << "' is now (or already was) a tier of '" << poolstr << "'"; | |
12296 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
12297 | get_last_committed() + 1)); | |
12298 | return true; | |
12299 | } else if (prefix == "osd tier remove" || | |
12300 | prefix == "osd tier rm") { | |
12301 | string poolstr; | |
11fdf7f2 | 12302 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12303 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12304 | if (pool_id < 0) { | |
12305 | ss << "unrecognized pool '" << poolstr << "'"; | |
12306 | err = -ENOENT; | |
12307 | goto reply; | |
12308 | } | |
12309 | string tierpoolstr; | |
11fdf7f2 | 12310 | cmd_getval(cct, cmdmap, "tierpool", tierpoolstr); |
7c673cae FG |
12311 | int64_t tierpool_id = osdmap.lookup_pg_pool_name(tierpoolstr); |
12312 | if (tierpool_id < 0) { | |
12313 | ss << "unrecognized pool '" << tierpoolstr << "'"; | |
12314 | err = -ENOENT; | |
12315 | goto reply; | |
12316 | } | |
12317 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11fdf7f2 | 12318 | ceph_assert(p); |
7c673cae | 12319 | const pg_pool_t *tp = osdmap.get_pg_pool(tierpool_id); |
11fdf7f2 | 12320 | ceph_assert(tp); |
7c673cae FG |
12321 | |
12322 | if (!_check_remove_tier(pool_id, p, tp, &err, &ss)) { | |
12323 | goto reply; | |
12324 | } | |
12325 | ||
12326 | if (p->tiers.count(tierpool_id) == 0) { | |
12327 | ss << "pool '" << tierpoolstr << "' is now (or already was) not a tier of '" << poolstr << "'"; | |
12328 | err = 0; | |
12329 | goto reply; | |
12330 | } | |
12331 | if (tp->tier_of != pool_id) { | |
12332 | ss << "tier pool '" << tierpoolstr << "' is a tier of '" | |
12333 | << osdmap.get_pool_name(tp->tier_of) << "': " | |
12334 | // be scary about it; this is an inconsistency and bells must go off | |
12335 | << "THIS SHOULD NOT HAVE HAPPENED AT ALL"; | |
12336 | err = -EINVAL; | |
12337 | goto reply; | |
12338 | } | |
12339 | if (p->read_tier == tierpool_id) { | |
12340 | ss << "tier pool '" << tierpoolstr << "' is the overlay for '" << poolstr << "'; please remove-overlay first"; | |
12341 | err = -EBUSY; | |
12342 | goto reply; | |
12343 | } | |
12344 | // go | |
12345 | pg_pool_t *np = pending_inc.get_new_pool(pool_id, p); | |
12346 | pg_pool_t *ntp = pending_inc.get_new_pool(tierpool_id, tp); | |
12347 | if (np->tiers.count(tierpool_id) == 0 || | |
12348 | ntp->tier_of != pool_id || | |
12349 | np->read_tier == tierpool_id) { | |
12350 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12351 | return true; | |
12352 | } | |
12353 | np->tiers.erase(tierpool_id); | |
12354 | ntp->clear_tier(); | |
12355 | ss << "pool '" << tierpoolstr << "' is now (or already was) not a tier of '" << poolstr << "'"; | |
12356 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
12357 | get_last_committed() + 1)); | |
12358 | return true; | |
12359 | } else if (prefix == "osd tier set-overlay") { | |
12360 | err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss); | |
12361 | if (err == -EAGAIN) | |
12362 | goto wait; | |
12363 | if (err) | |
12364 | goto reply; | |
12365 | string poolstr; | |
11fdf7f2 | 12366 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12367 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12368 | if (pool_id < 0) { | |
12369 | ss << "unrecognized pool '" << poolstr << "'"; | |
12370 | err = -ENOENT; | |
12371 | goto reply; | |
12372 | } | |
12373 | string overlaypoolstr; | |
11fdf7f2 | 12374 | cmd_getval(cct, cmdmap, "overlaypool", overlaypoolstr); |
7c673cae FG |
12375 | int64_t overlaypool_id = osdmap.lookup_pg_pool_name(overlaypoolstr); |
12376 | if (overlaypool_id < 0) { | |
12377 | ss << "unrecognized pool '" << overlaypoolstr << "'"; | |
12378 | err = -ENOENT; | |
12379 | goto reply; | |
12380 | } | |
12381 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11fdf7f2 | 12382 | ceph_assert(p); |
7c673cae | 12383 | const pg_pool_t *overlay_p = osdmap.get_pg_pool(overlaypool_id); |
11fdf7f2 | 12384 | ceph_assert(overlay_p); |
7c673cae FG |
12385 | if (p->tiers.count(overlaypool_id) == 0) { |
12386 | ss << "tier pool '" << overlaypoolstr << "' is not a tier of '" << poolstr << "'"; | |
12387 | err = -EINVAL; | |
12388 | goto reply; | |
12389 | } | |
12390 | if (p->read_tier == overlaypool_id) { | |
12391 | err = 0; | |
12392 | ss << "overlay for '" << poolstr << "' is now (or already was) '" << overlaypoolstr << "'"; | |
12393 | goto reply; | |
12394 | } | |
12395 | if (p->has_read_tier()) { | |
12396 | ss << "pool '" << poolstr << "' has overlay '" | |
12397 | << osdmap.get_pool_name(p->read_tier) | |
12398 | << "'; please remove-overlay first"; | |
12399 | err = -EINVAL; | |
12400 | goto reply; | |
12401 | } | |
12402 | ||
12403 | // go | |
12404 | pg_pool_t *np = pending_inc.get_new_pool(pool_id, p); | |
12405 | np->read_tier = overlaypool_id; | |
12406 | np->write_tier = overlaypool_id; | |
12407 | np->set_last_force_op_resend(pending_inc.epoch); | |
12408 | pg_pool_t *noverlay_p = pending_inc.get_new_pool(overlaypool_id, overlay_p); | |
12409 | noverlay_p->set_last_force_op_resend(pending_inc.epoch); | |
12410 | ss << "overlay for '" << poolstr << "' is now (or already was) '" << overlaypoolstr << "'"; | |
12411 | if (overlay_p->cache_mode == pg_pool_t::CACHEMODE_NONE) | |
12412 | ss <<" (WARNING: overlay pool cache_mode is still NONE)"; | |
12413 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
12414 | get_last_committed() + 1)); | |
12415 | return true; | |
12416 | } else if (prefix == "osd tier remove-overlay" || | |
12417 | prefix == "osd tier rm-overlay") { | |
12418 | string poolstr; | |
11fdf7f2 | 12419 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12420 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12421 | if (pool_id < 0) { | |
12422 | ss << "unrecognized pool '" << poolstr << "'"; | |
12423 | err = -ENOENT; | |
12424 | goto reply; | |
12425 | } | |
12426 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11fdf7f2 | 12427 | ceph_assert(p); |
7c673cae FG |
12428 | if (!p->has_read_tier()) { |
12429 | err = 0; | |
12430 | ss << "there is now (or already was) no overlay for '" << poolstr << "'"; | |
12431 | goto reply; | |
12432 | } | |
12433 | ||
12434 | if (!_check_remove_tier(pool_id, p, NULL, &err, &ss)) { | |
12435 | goto reply; | |
12436 | } | |
12437 | ||
12438 | // go | |
12439 | pg_pool_t *np = pending_inc.get_new_pool(pool_id, p); | |
12440 | if (np->has_read_tier()) { | |
12441 | const pg_pool_t *op = osdmap.get_pg_pool(np->read_tier); | |
12442 | pg_pool_t *nop = pending_inc.get_new_pool(np->read_tier,op); | |
12443 | nop->set_last_force_op_resend(pending_inc.epoch); | |
12444 | } | |
12445 | if (np->has_write_tier()) { | |
12446 | const pg_pool_t *op = osdmap.get_pg_pool(np->write_tier); | |
12447 | pg_pool_t *nop = pending_inc.get_new_pool(np->write_tier, op); | |
12448 | nop->set_last_force_op_resend(pending_inc.epoch); | |
12449 | } | |
12450 | np->clear_read_tier(); | |
12451 | np->clear_write_tier(); | |
12452 | np->set_last_force_op_resend(pending_inc.epoch); | |
12453 | ss << "there is now (or already was) no overlay for '" << poolstr << "'"; | |
12454 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
12455 | get_last_committed() + 1)); | |
12456 | return true; | |
12457 | } else if (prefix == "osd tier cache-mode") { | |
12458 | err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss); | |
12459 | if (err == -EAGAIN) | |
12460 | goto wait; | |
12461 | if (err) | |
12462 | goto reply; | |
12463 | string poolstr; | |
11fdf7f2 | 12464 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12465 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12466 | if (pool_id < 0) { | |
12467 | ss << "unrecognized pool '" << poolstr << "'"; | |
12468 | err = -ENOENT; | |
12469 | goto reply; | |
12470 | } | |
12471 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11fdf7f2 | 12472 | ceph_assert(p); |
7c673cae FG |
12473 | if (!p->is_tier()) { |
12474 | ss << "pool '" << poolstr << "' is not a tier"; | |
12475 | err = -EINVAL; | |
12476 | goto reply; | |
12477 | } | |
12478 | string modestr; | |
11fdf7f2 | 12479 | cmd_getval(cct, cmdmap, "mode", modestr); |
7c673cae FG |
12480 | pg_pool_t::cache_mode_t mode = pg_pool_t::get_cache_mode_from_str(modestr); |
12481 | if (mode < 0) { | |
12482 | ss << "'" << modestr << "' is not a valid cache mode"; | |
12483 | err = -EINVAL; | |
12484 | goto reply; | |
12485 | } | |
12486 | ||
11fdf7f2 TL |
12487 | bool sure = false; |
12488 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", sure); | |
12489 | ||
7c673cae FG |
12490 | if ((mode != pg_pool_t::CACHEMODE_WRITEBACK && |
12491 | mode != pg_pool_t::CACHEMODE_NONE && | |
12492 | mode != pg_pool_t::CACHEMODE_PROXY && | |
12493 | mode != pg_pool_t::CACHEMODE_READPROXY) && | |
11fdf7f2 | 12494 | !sure) { |
7c673cae FG |
12495 | ss << "'" << modestr << "' is not a well-supported cache mode and may " |
12496 | << "corrupt your data. pass --yes-i-really-mean-it to force."; | |
12497 | err = -EPERM; | |
12498 | goto reply; | |
12499 | } | |
12500 | ||
12501 | // pool already has this cache-mode set and there are no pending changes | |
12502 | if (p->cache_mode == mode && | |
12503 | (pending_inc.new_pools.count(pool_id) == 0 || | |
12504 | pending_inc.new_pools[pool_id].cache_mode == p->cache_mode)) { | |
12505 | ss << "set cache-mode for pool '" << poolstr << "'" | |
12506 | << " to " << pg_pool_t::get_cache_mode_name(mode); | |
12507 | err = 0; | |
12508 | goto reply; | |
12509 | } | |
12510 | ||
12511 | /* Mode description: | |
12512 | * | |
12513 | * none: No cache-mode defined | |
12514 | * forward: Forward all reads and writes to base pool | |
12515 | * writeback: Cache writes, promote reads from base pool | |
12516 | * readonly: Forward writes to base pool | |
12517 | * readforward: Writes are in writeback mode, Reads are in forward mode | |
12518 | * proxy: Proxy all reads and writes to base pool | |
12519 | * readproxy: Writes are in writeback mode, Reads are in proxy mode | |
12520 | * | |
12521 | * Hence, these are the allowed transitions: | |
12522 | * | |
12523 | * none -> any | |
12524 | * forward -> proxy || readforward || readproxy || writeback || any IF num_objects_dirty == 0 | |
12525 | * proxy -> forward || readforward || readproxy || writeback || any IF num_objects_dirty == 0 | |
12526 | * readforward -> forward || proxy || readproxy || writeback || any IF num_objects_dirty == 0 | |
12527 | * readproxy -> forward || proxy || readforward || writeback || any IF num_objects_dirty == 0 | |
12528 | * writeback -> readforward || readproxy || forward || proxy | |
12529 | * readonly -> any | |
12530 | */ | |
12531 | ||
12532 | // We check if the transition is valid against the current pool mode, as | |
12533 | // it is the only committed state thus far. We will blantly squash | |
12534 | // whatever mode is on the pending state. | |
12535 | ||
12536 | if (p->cache_mode == pg_pool_t::CACHEMODE_WRITEBACK && | |
12537 | (mode != pg_pool_t::CACHEMODE_FORWARD && | |
12538 | mode != pg_pool_t::CACHEMODE_PROXY && | |
12539 | mode != pg_pool_t::CACHEMODE_READFORWARD && | |
12540 | mode != pg_pool_t::CACHEMODE_READPROXY)) { | |
12541 | ss << "unable to set cache-mode '" << pg_pool_t::get_cache_mode_name(mode) | |
12542 | << "' on a '" << pg_pool_t::get_cache_mode_name(p->cache_mode) | |
12543 | << "' pool; only '" | |
12544 | << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_FORWARD) | |
12545 | << "','" | |
12546 | << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_PROXY) | |
12547 | << "','" | |
12548 | << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_READFORWARD) | |
12549 | << "','" | |
12550 | << pg_pool_t::get_cache_mode_name(pg_pool_t::CACHEMODE_READPROXY) | |
12551 | << "' allowed."; | |
12552 | err = -EINVAL; | |
12553 | goto reply; | |
12554 | } | |
12555 | if ((p->cache_mode == pg_pool_t::CACHEMODE_READFORWARD && | |
12556 | (mode != pg_pool_t::CACHEMODE_WRITEBACK && | |
12557 | mode != pg_pool_t::CACHEMODE_FORWARD && | |
12558 | mode != pg_pool_t::CACHEMODE_PROXY && | |
12559 | mode != pg_pool_t::CACHEMODE_READPROXY)) || | |
12560 | ||
12561 | (p->cache_mode == pg_pool_t::CACHEMODE_READPROXY && | |
12562 | (mode != pg_pool_t::CACHEMODE_WRITEBACK && | |
12563 | mode != pg_pool_t::CACHEMODE_FORWARD && | |
12564 | mode != pg_pool_t::CACHEMODE_READFORWARD && | |
12565 | mode != pg_pool_t::CACHEMODE_PROXY)) || | |
12566 | ||
12567 | (p->cache_mode == pg_pool_t::CACHEMODE_PROXY && | |
12568 | (mode != pg_pool_t::CACHEMODE_WRITEBACK && | |
12569 | mode != pg_pool_t::CACHEMODE_FORWARD && | |
12570 | mode != pg_pool_t::CACHEMODE_READFORWARD && | |
12571 | mode != pg_pool_t::CACHEMODE_READPROXY)) || | |
12572 | ||
12573 | (p->cache_mode == pg_pool_t::CACHEMODE_FORWARD && | |
12574 | (mode != pg_pool_t::CACHEMODE_WRITEBACK && | |
12575 | mode != pg_pool_t::CACHEMODE_READFORWARD && | |
12576 | mode != pg_pool_t::CACHEMODE_PROXY && | |
12577 | mode != pg_pool_t::CACHEMODE_READPROXY))) { | |
12578 | ||
31f18b77 | 12579 | const pool_stat_t* pstats = |
11fdf7f2 | 12580 | mon->mgrstatmon()->get_pool_stat(pool_id); |
7c673cae | 12581 | |
31f18b77 | 12582 | if (pstats && pstats->stats.sum.num_objects_dirty > 0) { |
7c673cae FG |
12583 | ss << "unable to set cache-mode '" |
12584 | << pg_pool_t::get_cache_mode_name(mode) << "' on pool '" << poolstr | |
12585 | << "': dirty objects found"; | |
12586 | err = -EBUSY; | |
12587 | goto reply; | |
12588 | } | |
12589 | } | |
12590 | // go | |
12591 | pg_pool_t *np = pending_inc.get_new_pool(pool_id, p); | |
12592 | np->cache_mode = mode; | |
12593 | // set this both when moving to and from cache_mode NONE. this is to | |
12594 | // capture legacy pools that were set up before this flag existed. | |
12595 | np->flags |= pg_pool_t::FLAG_INCOMPLETE_CLONES; | |
12596 | ss << "set cache-mode for pool '" << poolstr | |
12597 | << "' to " << pg_pool_t::get_cache_mode_name(mode); | |
12598 | if (mode == pg_pool_t::CACHEMODE_NONE) { | |
12599 | const pg_pool_t *base_pool = osdmap.get_pg_pool(np->tier_of); | |
11fdf7f2 | 12600 | ceph_assert(base_pool); |
7c673cae FG |
12601 | if (base_pool->read_tier == pool_id || |
12602 | base_pool->write_tier == pool_id) | |
12603 | ss <<" (WARNING: pool is still configured as read or write tier)"; | |
12604 | } | |
12605 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
12606 | get_last_committed() + 1)); | |
12607 | return true; | |
12608 | } else if (prefix == "osd tier add-cache") { | |
12609 | err = check_cluster_features(CEPH_FEATURE_OSD_CACHEPOOL, ss); | |
12610 | if (err == -EAGAIN) | |
12611 | goto wait; | |
12612 | if (err) | |
12613 | goto reply; | |
12614 | string poolstr; | |
11fdf7f2 | 12615 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12616 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12617 | if (pool_id < 0) { | |
12618 | ss << "unrecognized pool '" << poolstr << "'"; | |
12619 | err = -ENOENT; | |
12620 | goto reply; | |
12621 | } | |
12622 | string tierpoolstr; | |
11fdf7f2 | 12623 | cmd_getval(cct, cmdmap, "tierpool", tierpoolstr); |
7c673cae FG |
12624 | int64_t tierpool_id = osdmap.lookup_pg_pool_name(tierpoolstr); |
12625 | if (tierpool_id < 0) { | |
12626 | ss << "unrecognized pool '" << tierpoolstr << "'"; | |
12627 | err = -ENOENT; | |
12628 | goto reply; | |
12629 | } | |
12630 | const pg_pool_t *p = osdmap.get_pg_pool(pool_id); | |
11fdf7f2 | 12631 | ceph_assert(p); |
7c673cae | 12632 | const pg_pool_t *tp = osdmap.get_pg_pool(tierpool_id); |
11fdf7f2 | 12633 | ceph_assert(tp); |
7c673cae FG |
12634 | |
12635 | if (!_check_become_tier(tierpool_id, tp, pool_id, p, &err, &ss)) { | |
12636 | goto reply; | |
12637 | } | |
12638 | ||
12639 | int64_t size = 0; | |
11fdf7f2 | 12640 | if (!cmd_getval(cct, cmdmap, "size", size)) { |
7c673cae | 12641 | ss << "unable to parse 'size' value '" |
11fdf7f2 | 12642 | << cmd_vartype_stringify(cmdmap.at("size")) << "'"; |
7c673cae FG |
12643 | err = -EINVAL; |
12644 | goto reply; | |
12645 | } | |
12646 | // make sure new tier is empty | |
31f18b77 | 12647 | const pool_stat_t *pstats = |
11fdf7f2 | 12648 | mon->mgrstatmon()->get_pool_stat(tierpool_id); |
31f18b77 | 12649 | if (pstats && pstats->stats.sum.num_objects != 0) { |
7c673cae FG |
12650 | ss << "tier pool '" << tierpoolstr << "' is not empty"; |
12651 | err = -ENOTEMPTY; | |
12652 | goto reply; | |
12653 | } | |
11fdf7f2 | 12654 | auto& modestr = g_conf().get_val<string>("osd_tier_default_cache_mode"); |
7c673cae FG |
12655 | pg_pool_t::cache_mode_t mode = pg_pool_t::get_cache_mode_from_str(modestr); |
12656 | if (mode < 0) { | |
12657 | ss << "osd tier cache default mode '" << modestr << "' is not a valid cache mode"; | |
12658 | err = -EINVAL; | |
12659 | goto reply; | |
12660 | } | |
12661 | HitSet::Params hsp; | |
11fdf7f2 TL |
12662 | auto& cache_hit_set_type = |
12663 | g_conf().get_val<string>("osd_tier_default_cache_hit_set_type"); | |
12664 | if (cache_hit_set_type == "bloom") { | |
7c673cae | 12665 | BloomHitSet::Params *bsp = new BloomHitSet::Params; |
11fdf7f2 | 12666 | bsp->set_fpp(g_conf().get_val<double>("osd_pool_default_hit_set_bloom_fpp")); |
7c673cae | 12667 | hsp = HitSet::Params(bsp); |
11fdf7f2 | 12668 | } else if (cache_hit_set_type == "explicit_hash") { |
7c673cae | 12669 | hsp = HitSet::Params(new ExplicitHashHitSet::Params); |
11fdf7f2 | 12670 | } else if (cache_hit_set_type == "explicit_object") { |
7c673cae FG |
12671 | hsp = HitSet::Params(new ExplicitObjectHitSet::Params); |
12672 | } else { | |
11fdf7f2 TL |
12673 | ss << "osd tier cache default hit set type '" |
12674 | << cache_hit_set_type << "' is not a known type"; | |
7c673cae FG |
12675 | err = -EINVAL; |
12676 | goto reply; | |
12677 | } | |
12678 | // go | |
12679 | pg_pool_t *np = pending_inc.get_new_pool(pool_id, p); | |
12680 | pg_pool_t *ntp = pending_inc.get_new_pool(tierpool_id, tp); | |
12681 | if (np->tiers.count(tierpool_id) || ntp->is_tier()) { | |
12682 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12683 | return true; | |
12684 | } | |
12685 | np->tiers.insert(tierpool_id); | |
12686 | np->read_tier = np->write_tier = tierpool_id; | |
12687 | np->set_snap_epoch(pending_inc.epoch); // tier will update to our snap info | |
12688 | np->set_last_force_op_resend(pending_inc.epoch); | |
12689 | ntp->set_last_force_op_resend(pending_inc.epoch); | |
12690 | ntp->tier_of = pool_id; | |
12691 | ntp->cache_mode = mode; | |
11fdf7f2 TL |
12692 | ntp->hit_set_count = g_conf().get_val<uint64_t>("osd_tier_default_cache_hit_set_count"); |
12693 | ntp->hit_set_period = g_conf().get_val<uint64_t>("osd_tier_default_cache_hit_set_period"); | |
12694 | ntp->min_read_recency_for_promote = g_conf().get_val<uint64_t>("osd_tier_default_cache_min_read_recency_for_promote"); | |
12695 | ntp->min_write_recency_for_promote = g_conf().get_val<uint64_t>("osd_tier_default_cache_min_write_recency_for_promote"); | |
12696 | ntp->hit_set_grade_decay_rate = g_conf().get_val<uint64_t>("osd_tier_default_cache_hit_set_grade_decay_rate"); | |
12697 | ntp->hit_set_search_last_n = g_conf().get_val<uint64_t>("osd_tier_default_cache_hit_set_search_last_n"); | |
7c673cae FG |
12698 | ntp->hit_set_params = hsp; |
12699 | ntp->target_max_bytes = size; | |
12700 | ss << "pool '" << tierpoolstr << "' is now (or already was) a cache tier of '" << poolstr << "'"; | |
12701 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, ss.str(), | |
12702 | get_last_committed() + 1)); | |
12703 | return true; | |
12704 | } else if (prefix == "osd pool set-quota") { | |
12705 | string poolstr; | |
11fdf7f2 | 12706 | cmd_getval(cct, cmdmap, "pool", poolstr); |
7c673cae FG |
12707 | int64_t pool_id = osdmap.lookup_pg_pool_name(poolstr); |
12708 | if (pool_id < 0) { | |
12709 | ss << "unrecognized pool '" << poolstr << "'"; | |
12710 | err = -ENOENT; | |
12711 | goto reply; | |
12712 | } | |
12713 | ||
12714 | string field; | |
11fdf7f2 | 12715 | cmd_getval(cct, cmdmap, "field", field); |
7c673cae FG |
12716 | if (field != "max_objects" && field != "max_bytes") { |
12717 | ss << "unrecognized field '" << field << "'; should be 'max_bytes' or 'max_objects'"; | |
12718 | err = -EINVAL; | |
12719 | goto reply; | |
12720 | } | |
12721 | ||
12722 | // val could contain unit designations, so we treat as a string | |
12723 | string val; | |
11fdf7f2 | 12724 | cmd_getval(cct, cmdmap, "val", val); |
1adf2230 AA |
12725 | string tss; |
12726 | int64_t value; | |
12727 | if (field == "max_objects") { | |
12728 | value = strict_sistrtoll(val.c_str(), &tss); | |
12729 | } else if (field == "max_bytes") { | |
12730 | value = strict_iecstrtoll(val.c_str(), &tss); | |
12731 | } else { | |
11fdf7f2 | 12732 | ceph_abort_msg("unrecognized option"); |
1adf2230 AA |
12733 | } |
12734 | if (!tss.empty()) { | |
12735 | ss << "error parsing value '" << val << "': " << tss; | |
12736 | err = -EINVAL; | |
7c673cae FG |
12737 | goto reply; |
12738 | } | |
12739 | ||
12740 | pg_pool_t *pi = pending_inc.get_new_pool(pool_id, osdmap.get_pg_pool(pool_id)); | |
12741 | if (field == "max_objects") { | |
12742 | pi->quota_max_objects = value; | |
12743 | } else if (field == "max_bytes") { | |
12744 | pi->quota_max_bytes = value; | |
12745 | } else { | |
11fdf7f2 | 12746 | ceph_abort_msg("unrecognized option"); |
7c673cae FG |
12747 | } |
12748 | ss << "set-quota " << field << " = " << value << " for pool " << poolstr; | |
12749 | rs = ss.str(); | |
12750 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
12751 | get_last_committed() + 1)); | |
12752 | return true; | |
c07f9fc5 FG |
12753 | } else if (prefix == "osd pool application enable" || |
12754 | prefix == "osd pool application disable" || | |
12755 | prefix == "osd pool application set" || | |
12756 | prefix == "osd pool application rm") { | |
12757 | err = prepare_command_pool_application(prefix, cmdmap, ss); | |
11fdf7f2 | 12758 | if (err == -EAGAIN) { |
c07f9fc5 | 12759 | goto wait; |
11fdf7f2 | 12760 | } else if (err < 0) { |
7c673cae | 12761 | goto reply; |
7c673cae | 12762 | } else { |
11fdf7f2 | 12763 | goto update; |
7c673cae | 12764 | } |
c07f9fc5 FG |
12765 | } else if (prefix == "osd force-create-pg") { |
12766 | pg_t pgid; | |
12767 | string pgidstr; | |
11fdf7f2 | 12768 | cmd_getval(cct, cmdmap, "pgid", pgidstr); |
c07f9fc5 FG |
12769 | if (!pgid.parse(pgidstr.c_str())) { |
12770 | ss << "invalid pgid '" << pgidstr << "'"; | |
12771 | err = -EINVAL; | |
12772 | goto reply; | |
12773 | } | |
94b18763 FG |
12774 | if (!osdmap.pg_exists(pgid)) { |
12775 | ss << "pg " << pgid << " should not exist"; | |
12776 | err = -ENOENT; | |
12777 | goto reply; | |
12778 | } | |
11fdf7f2 TL |
12779 | bool sure = false; |
12780 | cmd_getval(cct, cmdmap, "yes_i_really_mean_it", sure); | |
12781 | if (!sure) { | |
12782 | ss << "This command will recreate a lost (as in data lost) PG with data in it, such " | |
12783 | << "that the cluster will give up ever trying to recover the lost data. Do this " | |
12784 | << "only if you are certain that all copies of the PG are in fact lost and you are " | |
12785 | << "willing to accept that the data is permanently destroyed. Pass " | |
12786 | << "--yes-i-really-mean-it to proceed."; | |
12787 | err = -EPERM; | |
12788 | goto reply; | |
12789 | } | |
c07f9fc5 FG |
12790 | bool creating_now; |
12791 | { | |
12792 | std::lock_guard<std::mutex> l(creating_pgs_lock); | |
12793 | auto emplaced = creating_pgs.pgs.emplace(pgid, | |
12794 | make_pair(osdmap.get_epoch(), | |
12795 | ceph_clock_now())); | |
12796 | creating_now = emplaced.second; | |
12797 | } | |
12798 | if (creating_now) { | |
12799 | ss << "pg " << pgidstr << " now creating, ok"; | |
11fdf7f2 TL |
12800 | // set the pool's CREATING flag so that (1) the osd won't ignore our |
12801 | // create message and (2) we won't propose any future pg_num changes | |
12802 | // until after the PG has been instantiated. | |
12803 | if (pending_inc.new_pools.count(pgid.pool()) == 0) { | |
12804 | pending_inc.new_pools[pgid.pool()] = *osdmap.get_pg_pool(pgid.pool()); | |
12805 | } | |
12806 | pending_inc.new_pools[pgid.pool()].flags |= pg_pool_t::FLAG_CREATING; | |
c07f9fc5 FG |
12807 | err = 0; |
12808 | goto update; | |
12809 | } else { | |
12810 | ss << "pg " << pgid << " already creating"; | |
12811 | err = 0; | |
12812 | goto reply; | |
12813 | } | |
7c673cae FG |
12814 | } else { |
12815 | err = -EINVAL; | |
12816 | } | |
12817 | ||
12818 | reply: | |
12819 | getline(ss, rs); | |
12820 | if (err < 0 && rs.length() == 0) | |
12821 | rs = cpp_strerror(err); | |
12822 | mon->reply_command(op, err, rs, rdata, get_last_committed()); | |
12823 | return ret; | |
12824 | ||
12825 | update: | |
12826 | getline(ss, rs); | |
12827 | wait_for_finished_proposal(op, new Monitor::C_Command(mon, op, 0, rs, | |
12828 | get_last_committed() + 1)); | |
12829 | return true; | |
12830 | ||
12831 | wait: | |
12832 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
12833 | return true; | |
12834 | } | |
12835 | ||
28e407b8 | 12836 | bool OSDMonitor::enforce_pool_op_caps(MonOpRequestRef op) |
7c673cae FG |
12837 | { |
12838 | op->mark_osdmon_event(__func__); | |
28e407b8 | 12839 | |
7c673cae | 12840 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); |
11fdf7f2 | 12841 | MonSession *session = op->get_session(); |
28e407b8 AA |
12842 | if (!session) { |
12843 | _pool_op_reply(op, -EPERM, osdmap.get_epoch()); | |
12844 | return true; | |
12845 | } | |
12846 | ||
12847 | switch (m->op) { | |
12848 | case POOL_OP_CREATE_UNMANAGED_SNAP: | |
12849 | case POOL_OP_DELETE_UNMANAGED_SNAP: | |
12850 | { | |
12851 | const std::string* pool_name = nullptr; | |
12852 | const pg_pool_t *pg_pool = osdmap.get_pg_pool(m->pool); | |
12853 | if (pg_pool != nullptr) { | |
12854 | pool_name = &osdmap.get_pool_name(m->pool); | |
12855 | } | |
12856 | ||
12857 | if (!is_unmanaged_snap_op_permitted(cct, mon->key_server, | |
12858 | session->entity_name, session->caps, | |
11fdf7f2 | 12859 | session->get_peer_socket_addr(), |
28e407b8 AA |
12860 | pool_name)) { |
12861 | dout(0) << "got unmanaged-snap pool op from entity with insufficient " | |
12862 | << "privileges. message: " << *m << std::endl | |
12863 | << "caps: " << session->caps << dendl; | |
12864 | _pool_op_reply(op, -EPERM, osdmap.get_epoch()); | |
12865 | return true; | |
12866 | } | |
12867 | } | |
12868 | break; | |
12869 | default: | |
12870 | if (!session->is_capable("osd", MON_CAP_W)) { | |
12871 | dout(0) << "got pool op from entity with insufficient privileges. " | |
12872 | << "message: " << *m << std::endl | |
12873 | << "caps: " << session->caps << dendl; | |
12874 | _pool_op_reply(op, -EPERM, osdmap.get_epoch()); | |
12875 | return true; | |
12876 | } | |
12877 | break; | |
12878 | } | |
12879 | ||
12880 | return false; | |
12881 | } | |
12882 | ||
12883 | bool OSDMonitor::preprocess_pool_op(MonOpRequestRef op) | |
12884 | { | |
12885 | op->mark_osdmon_event(__func__); | |
12886 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); | |
12887 | ||
12888 | if (enforce_pool_op_caps(op)) { | |
12889 | return true; | |
12890 | } | |
12891 | ||
7c673cae FG |
12892 | if (m->fsid != mon->monmap->fsid) { |
12893 | dout(0) << __func__ << " drop message on fsid " << m->fsid | |
12894 | << " != " << mon->monmap->fsid << " for " << *m << dendl; | |
12895 | _pool_op_reply(op, -EINVAL, osdmap.get_epoch()); | |
12896 | return true; | |
12897 | } | |
12898 | ||
12899 | if (m->op == POOL_OP_CREATE) | |
12900 | return preprocess_pool_op_create(op); | |
12901 | ||
11fdf7f2 TL |
12902 | const pg_pool_t *p = osdmap.get_pg_pool(m->pool); |
12903 | if (p == nullptr) { | |
7c673cae | 12904 | dout(10) << "attempt to operate on non-existent pool id " << m->pool << dendl; |
11fdf7f2 TL |
12905 | if (m->op == POOL_OP_DELETE) { |
12906 | _pool_op_reply(op, 0, osdmap.get_epoch()); | |
12907 | } else { | |
12908 | _pool_op_reply(op, -ENOENT, osdmap.get_epoch()); | |
12909 | } | |
7c673cae FG |
12910 | return true; |
12911 | } | |
12912 | ||
12913 | // check if the snap and snapname exist | |
12914 | bool snap_exists = false; | |
7c673cae FG |
12915 | if (p->snap_exists(m->name.c_str())) |
12916 | snap_exists = true; | |
12917 | ||
12918 | switch (m->op) { | |
12919 | case POOL_OP_CREATE_SNAP: | |
12920 | if (p->is_unmanaged_snaps_mode() || p->is_tier()) { | |
12921 | _pool_op_reply(op, -EINVAL, osdmap.get_epoch()); | |
12922 | return true; | |
12923 | } | |
12924 | if (snap_exists) { | |
12925 | _pool_op_reply(op, 0, osdmap.get_epoch()); | |
12926 | return true; | |
12927 | } | |
12928 | return false; | |
12929 | case POOL_OP_CREATE_UNMANAGED_SNAP: | |
12930 | if (p->is_pool_snaps_mode()) { | |
12931 | _pool_op_reply(op, -EINVAL, osdmap.get_epoch()); | |
12932 | return true; | |
12933 | } | |
12934 | return false; | |
12935 | case POOL_OP_DELETE_SNAP: | |
12936 | if (p->is_unmanaged_snaps_mode()) { | |
12937 | _pool_op_reply(op, -EINVAL, osdmap.get_epoch()); | |
12938 | return true; | |
12939 | } | |
12940 | if (!snap_exists) { | |
12941 | _pool_op_reply(op, 0, osdmap.get_epoch()); | |
12942 | return true; | |
12943 | } | |
12944 | return false; | |
12945 | case POOL_OP_DELETE_UNMANAGED_SNAP: | |
12946 | if (p->is_pool_snaps_mode()) { | |
12947 | _pool_op_reply(op, -EINVAL, osdmap.get_epoch()); | |
12948 | return true; | |
12949 | } | |
12950 | if (p->is_removed_snap(m->snapid)) { | |
12951 | _pool_op_reply(op, 0, osdmap.get_epoch()); | |
12952 | return true; | |
12953 | } | |
12954 | return false; | |
12955 | case POOL_OP_DELETE: | |
12956 | if (osdmap.lookup_pg_pool_name(m->name.c_str()) >= 0) { | |
12957 | _pool_op_reply(op, 0, osdmap.get_epoch()); | |
12958 | return true; | |
12959 | } | |
12960 | return false; | |
12961 | case POOL_OP_AUID_CHANGE: | |
12962 | return false; | |
12963 | default: | |
12964 | ceph_abort(); | |
12965 | break; | |
12966 | } | |
12967 | ||
12968 | return false; | |
12969 | } | |
12970 | ||
12971 | bool OSDMonitor::preprocess_pool_op_create(MonOpRequestRef op) | |
12972 | { | |
12973 | op->mark_osdmon_event(__func__); | |
12974 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); | |
7c673cae FG |
12975 | int64_t pool = osdmap.lookup_pg_pool_name(m->name.c_str()); |
12976 | if (pool >= 0) { | |
12977 | _pool_op_reply(op, 0, osdmap.get_epoch()); | |
12978 | return true; | |
12979 | } | |
12980 | ||
12981 | return false; | |
12982 | } | |
12983 | ||
12984 | bool OSDMonitor::prepare_pool_op(MonOpRequestRef op) | |
12985 | { | |
12986 | op->mark_osdmon_event(__func__); | |
12987 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); | |
12988 | dout(10) << "prepare_pool_op " << *m << dendl; | |
12989 | if (m->op == POOL_OP_CREATE) { | |
12990 | return prepare_pool_op_create(op); | |
12991 | } else if (m->op == POOL_OP_DELETE) { | |
12992 | return prepare_pool_op_delete(op); | |
12993 | } | |
12994 | ||
12995 | int ret = 0; | |
12996 | bool changed = false; | |
12997 | ||
12998 | if (!osdmap.have_pg_pool(m->pool)) { | |
12999 | _pool_op_reply(op, -ENOENT, osdmap.get_epoch()); | |
13000 | return false; | |
13001 | } | |
13002 | ||
13003 | const pg_pool_t *pool = osdmap.get_pg_pool(m->pool); | |
13004 | ||
13005 | switch (m->op) { | |
13006 | case POOL_OP_CREATE_SNAP: | |
13007 | if (pool->is_tier()) { | |
13008 | ret = -EINVAL; | |
13009 | _pool_op_reply(op, ret, osdmap.get_epoch()); | |
13010 | return false; | |
13011 | } // else, fall through | |
13012 | case POOL_OP_DELETE_SNAP: | |
13013 | if (!pool->is_unmanaged_snaps_mode()) { | |
13014 | bool snap_exists = pool->snap_exists(m->name.c_str()); | |
13015 | if ((m->op == POOL_OP_CREATE_SNAP && snap_exists) | |
13016 | || (m->op == POOL_OP_DELETE_SNAP && !snap_exists)) { | |
13017 | ret = 0; | |
13018 | } else { | |
13019 | break; | |
13020 | } | |
13021 | } else { | |
13022 | ret = -EINVAL; | |
13023 | } | |
13024 | _pool_op_reply(op, ret, osdmap.get_epoch()); | |
13025 | return false; | |
13026 | ||
13027 | case POOL_OP_DELETE_UNMANAGED_SNAP: | |
13028 | // we won't allow removal of an unmanaged snapshot from a pool | |
13029 | // not in unmanaged snaps mode. | |
13030 | if (!pool->is_unmanaged_snaps_mode()) { | |
13031 | _pool_op_reply(op, -ENOTSUP, osdmap.get_epoch()); | |
13032 | return false; | |
13033 | } | |
13034 | /* fall-thru */ | |
13035 | case POOL_OP_CREATE_UNMANAGED_SNAP: | |
13036 | // but we will allow creating an unmanaged snapshot on any pool | |
13037 | // as long as it is not in 'pool' snaps mode. | |
13038 | if (pool->is_pool_snaps_mode()) { | |
13039 | _pool_op_reply(op, -EINVAL, osdmap.get_epoch()); | |
13040 | return false; | |
13041 | } | |
13042 | } | |
13043 | ||
13044 | // projected pool info | |
13045 | pg_pool_t pp; | |
13046 | if (pending_inc.new_pools.count(m->pool)) | |
13047 | pp = pending_inc.new_pools[m->pool]; | |
13048 | else | |
13049 | pp = *osdmap.get_pg_pool(m->pool); | |
13050 | ||
13051 | bufferlist reply_data; | |
13052 | ||
13053 | // pool snaps vs unmanaged snaps are mutually exclusive | |
13054 | switch (m->op) { | |
13055 | case POOL_OP_CREATE_SNAP: | |
13056 | case POOL_OP_DELETE_SNAP: | |
13057 | if (pp.is_unmanaged_snaps_mode()) { | |
13058 | ret = -EINVAL; | |
13059 | goto out; | |
13060 | } | |
13061 | break; | |
13062 | ||
13063 | case POOL_OP_CREATE_UNMANAGED_SNAP: | |
13064 | case POOL_OP_DELETE_UNMANAGED_SNAP: | |
13065 | if (pp.is_pool_snaps_mode()) { | |
13066 | ret = -EINVAL; | |
13067 | goto out; | |
13068 | } | |
13069 | } | |
13070 | ||
13071 | switch (m->op) { | |
13072 | case POOL_OP_CREATE_SNAP: | |
13073 | if (!pp.snap_exists(m->name.c_str())) { | |
13074 | pp.add_snap(m->name.c_str(), ceph_clock_now()); | |
11fdf7f2 TL |
13075 | dout(10) << "create snap in pool " << m->pool << " " << m->name |
13076 | << " seq " << pp.get_snap_epoch() << dendl; | |
7c673cae FG |
13077 | changed = true; |
13078 | } | |
13079 | break; | |
13080 | ||
13081 | case POOL_OP_DELETE_SNAP: | |
13082 | { | |
13083 | snapid_t s = pp.snap_exists(m->name.c_str()); | |
13084 | if (s) { | |
13085 | pp.remove_snap(s); | |
11fdf7f2 | 13086 | pending_inc.new_removed_snaps[m->pool].insert(s); |
7c673cae FG |
13087 | changed = true; |
13088 | } | |
13089 | } | |
13090 | break; | |
13091 | ||
13092 | case POOL_OP_CREATE_UNMANAGED_SNAP: | |
13093 | { | |
13094 | uint64_t snapid; | |
13095 | pp.add_unmanaged_snap(snapid); | |
11fdf7f2 | 13096 | encode(snapid, reply_data); |
7c673cae FG |
13097 | changed = true; |
13098 | } | |
13099 | break; | |
13100 | ||
13101 | case POOL_OP_DELETE_UNMANAGED_SNAP: | |
13102 | if (!pp.is_removed_snap(m->snapid)) { | |
28e407b8 AA |
13103 | if (m->snapid > pp.get_snap_seq()) { |
13104 | _pool_op_reply(op, -ENOENT, osdmap.get_epoch()); | |
13105 | return false; | |
13106 | } | |
7c673cae | 13107 | pp.remove_unmanaged_snap(m->snapid); |
11fdf7f2 | 13108 | pending_inc.new_removed_snaps[m->pool].insert(m->snapid); |
7c673cae FG |
13109 | changed = true; |
13110 | } | |
13111 | break; | |
13112 | ||
13113 | case POOL_OP_AUID_CHANGE: | |
11fdf7f2 TL |
13114 | _pool_op_reply(op, -EOPNOTSUPP, osdmap.get_epoch()); |
13115 | return false; | |
7c673cae FG |
13116 | |
13117 | default: | |
13118 | ceph_abort(); | |
13119 | break; | |
13120 | } | |
13121 | ||
13122 | if (changed) { | |
13123 | pp.set_snap_epoch(pending_inc.epoch); | |
13124 | pending_inc.new_pools[m->pool] = pp; | |
13125 | } | |
13126 | ||
13127 | out: | |
13128 | wait_for_finished_proposal(op, new OSDMonitor::C_PoolOp(this, op, ret, pending_inc.epoch, &reply_data)); | |
13129 | return true; | |
13130 | } | |
13131 | ||
13132 | bool OSDMonitor::prepare_pool_op_create(MonOpRequestRef op) | |
13133 | { | |
13134 | op->mark_osdmon_event(__func__); | |
13135 | int err = prepare_new_pool(op); | |
13136 | wait_for_finished_proposal(op, new OSDMonitor::C_PoolOp(this, op, err, pending_inc.epoch)); | |
13137 | return true; | |
13138 | } | |
13139 | ||
13140 | int OSDMonitor::_check_remove_pool(int64_t pool_id, const pg_pool_t& pool, | |
13141 | ostream *ss) | |
13142 | { | |
13143 | const string& poolstr = osdmap.get_pool_name(pool_id); | |
13144 | ||
13145 | // If the Pool is in use by CephFS, refuse to delete it | |
28e407b8 | 13146 | FSMap const &pending_fsmap = mon->mdsmon()->get_pending_fsmap(); |
7c673cae FG |
13147 | if (pending_fsmap.pool_in_use(pool_id)) { |
13148 | *ss << "pool '" << poolstr << "' is in use by CephFS"; | |
13149 | return -EBUSY; | |
13150 | } | |
13151 | ||
13152 | if (pool.tier_of >= 0) { | |
13153 | *ss << "pool '" << poolstr << "' is a tier of '" | |
13154 | << osdmap.get_pool_name(pool.tier_of) << "'"; | |
13155 | return -EBUSY; | |
13156 | } | |
13157 | if (!pool.tiers.empty()) { | |
13158 | *ss << "pool '" << poolstr << "' has tiers"; | |
13159 | for(auto tier : pool.tiers) { | |
13160 | *ss << " " << osdmap.get_pool_name(tier); | |
13161 | } | |
13162 | return -EBUSY; | |
13163 | } | |
13164 | ||
11fdf7f2 | 13165 | if (!g_conf()->mon_allow_pool_delete) { |
7c673cae FG |
13166 | *ss << "pool deletion is disabled; you must first set the mon_allow_pool_delete config option to true before you can destroy a pool"; |
13167 | return -EPERM; | |
13168 | } | |
13169 | ||
13170 | if (pool.has_flag(pg_pool_t::FLAG_NODELETE)) { | |
13171 | *ss << "pool deletion is disabled; you must unset nodelete flag for the pool first"; | |
13172 | return -EPERM; | |
13173 | } | |
13174 | ||
13175 | *ss << "pool '" << poolstr << "' removed"; | |
13176 | return 0; | |
13177 | } | |
13178 | ||
13179 | /** | |
13180 | * Check if it is safe to add a tier to a base pool | |
13181 | * | |
13182 | * @return | |
13183 | * True if the operation should proceed, false if we should abort here | |
13184 | * (abort doesn't necessarily mean error, could be idempotency) | |
13185 | */ | |
13186 | bool OSDMonitor::_check_become_tier( | |
13187 | const int64_t tier_pool_id, const pg_pool_t *tier_pool, | |
13188 | const int64_t base_pool_id, const pg_pool_t *base_pool, | |
13189 | int *err, | |
13190 | ostream *ss) const | |
13191 | { | |
13192 | const std::string &tier_pool_name = osdmap.get_pool_name(tier_pool_id); | |
13193 | const std::string &base_pool_name = osdmap.get_pool_name(base_pool_id); | |
13194 | ||
28e407b8 | 13195 | const FSMap &pending_fsmap = mon->mdsmon()->get_pending_fsmap(); |
7c673cae FG |
13196 | if (pending_fsmap.pool_in_use(tier_pool_id)) { |
13197 | *ss << "pool '" << tier_pool_name << "' is in use by CephFS"; | |
13198 | *err = -EBUSY; | |
13199 | return false; | |
13200 | } | |
13201 | ||
13202 | if (base_pool->tiers.count(tier_pool_id)) { | |
11fdf7f2 | 13203 | ceph_assert(tier_pool->tier_of == base_pool_id); |
7c673cae FG |
13204 | *err = 0; |
13205 | *ss << "pool '" << tier_pool_name << "' is now (or already was) a tier of '" | |
13206 | << base_pool_name << "'"; | |
13207 | return false; | |
13208 | } | |
13209 | ||
13210 | if (base_pool->is_tier()) { | |
13211 | *ss << "pool '" << base_pool_name << "' is already a tier of '" | |
13212 | << osdmap.get_pool_name(base_pool->tier_of) << "', " | |
13213 | << "multiple tiers are not yet supported."; | |
13214 | *err = -EINVAL; | |
13215 | return false; | |
13216 | } | |
13217 | ||
13218 | if (tier_pool->has_tiers()) { | |
13219 | *ss << "pool '" << tier_pool_name << "' has following tier(s) already:"; | |
13220 | for (set<uint64_t>::iterator it = tier_pool->tiers.begin(); | |
13221 | it != tier_pool->tiers.end(); ++it) | |
13222 | *ss << "'" << osdmap.get_pool_name(*it) << "',"; | |
13223 | *ss << " multiple tiers are not yet supported."; | |
13224 | *err = -EINVAL; | |
13225 | return false; | |
13226 | } | |
13227 | ||
13228 | if (tier_pool->is_tier()) { | |
13229 | *ss << "tier pool '" << tier_pool_name << "' is already a tier of '" | |
13230 | << osdmap.get_pool_name(tier_pool->tier_of) << "'"; | |
13231 | *err = -EINVAL; | |
13232 | return false; | |
13233 | } | |
13234 | ||
13235 | *err = 0; | |
13236 | return true; | |
13237 | } | |
13238 | ||
13239 | ||
13240 | /** | |
13241 | * Check if it is safe to remove a tier from this base pool | |
13242 | * | |
13243 | * @return | |
13244 | * True if the operation should proceed, false if we should abort here | |
13245 | * (abort doesn't necessarily mean error, could be idempotency) | |
13246 | */ | |
13247 | bool OSDMonitor::_check_remove_tier( | |
13248 | const int64_t base_pool_id, const pg_pool_t *base_pool, | |
13249 | const pg_pool_t *tier_pool, | |
13250 | int *err, ostream *ss) const | |
13251 | { | |
13252 | const std::string &base_pool_name = osdmap.get_pool_name(base_pool_id); | |
13253 | ||
13254 | // Apply CephFS-specific checks | |
28e407b8 | 13255 | const FSMap &pending_fsmap = mon->mdsmon()->get_pending_fsmap(); |
7c673cae | 13256 | if (pending_fsmap.pool_in_use(base_pool_id)) { |
94b18763 FG |
13257 | if (base_pool->is_erasure() && !base_pool->allows_ecoverwrites()) { |
13258 | // If the underlying pool is erasure coded and does not allow EC | |
13259 | // overwrites, we can't permit the removal of the replicated tier that | |
13260 | // CephFS relies on to access it | |
13261 | *ss << "pool '" << base_pool_name << | |
13262 | "' does not allow EC overwrites and is in use by CephFS" | |
13263 | " via its tier"; | |
7c673cae FG |
13264 | *err = -EBUSY; |
13265 | return false; | |
13266 | } | |
13267 | ||
13268 | if (tier_pool && tier_pool->cache_mode == pg_pool_t::CACHEMODE_WRITEBACK) { | |
13269 | *ss << "pool '" << base_pool_name << "' is in use by CephFS, and this " | |
13270 | "tier is still in use as a writeback cache. Change the cache " | |
13271 | "mode and flush the cache before removing it"; | |
13272 | *err = -EBUSY; | |
13273 | return false; | |
13274 | } | |
13275 | } | |
13276 | ||
13277 | *err = 0; | |
13278 | return true; | |
13279 | } | |
13280 | ||
13281 | int OSDMonitor::_prepare_remove_pool( | |
13282 | int64_t pool, ostream *ss, bool no_fake) | |
13283 | { | |
224ce89b | 13284 | dout(10) << __func__ << " " << pool << dendl; |
7c673cae FG |
13285 | const pg_pool_t *p = osdmap.get_pg_pool(pool); |
13286 | int r = _check_remove_pool(pool, *p, ss); | |
13287 | if (r < 0) | |
13288 | return r; | |
13289 | ||
13290 | auto new_pool = pending_inc.new_pools.find(pool); | |
13291 | if (new_pool != pending_inc.new_pools.end()) { | |
13292 | // if there is a problem with the pending info, wait and retry | |
13293 | // this op. | |
13294 | const auto& p = new_pool->second; | |
13295 | int r = _check_remove_pool(pool, p, ss); | |
13296 | if (r < 0) | |
13297 | return -EAGAIN; | |
13298 | } | |
13299 | ||
13300 | if (pending_inc.old_pools.count(pool)) { | |
224ce89b | 13301 | dout(10) << __func__ << " " << pool << " already pending removal" |
7c673cae FG |
13302 | << dendl; |
13303 | return 0; | |
13304 | } | |
13305 | ||
11fdf7f2 | 13306 | if (g_conf()->mon_fake_pool_delete && !no_fake) { |
7c673cae FG |
13307 | string old_name = osdmap.get_pool_name(pool); |
13308 | string new_name = old_name + "." + stringify(pool) + ".DELETED"; | |
13309 | dout(1) << __func__ << " faking pool deletion: renaming " << pool << " " | |
13310 | << old_name << " -> " << new_name << dendl; | |
13311 | pending_inc.new_pool_names[pool] = new_name; | |
13312 | return 0; | |
13313 | } | |
13314 | ||
13315 | // remove | |
13316 | pending_inc.old_pools.insert(pool); | |
13317 | ||
224ce89b | 13318 | // remove any pg_temp mappings for this pool |
7c673cae FG |
13319 | for (auto p = osdmap.pg_temp->begin(); |
13320 | p != osdmap.pg_temp->end(); | |
13321 | ++p) { | |
11fdf7f2 | 13322 | if (p->first.pool() == pool) { |
224ce89b | 13323 | dout(10) << __func__ << " " << pool << " removing obsolete pg_temp " |
7c673cae FG |
13324 | << p->first << dendl; |
13325 | pending_inc.new_pg_temp[p->first].clear(); | |
13326 | } | |
13327 | } | |
224ce89b | 13328 | // remove any primary_temp mappings for this pool |
7c673cae FG |
13329 | for (auto p = osdmap.primary_temp->begin(); |
13330 | p != osdmap.primary_temp->end(); | |
13331 | ++p) { | |
11fdf7f2 | 13332 | if (p->first.pool() == pool) { |
224ce89b | 13333 | dout(10) << __func__ << " " << pool |
7c673cae FG |
13334 | << " removing obsolete primary_temp" << p->first << dendl; |
13335 | pending_inc.new_primary_temp[p->first] = -1; | |
13336 | } | |
13337 | } | |
224ce89b WB |
13338 | // remove any pg_upmap mappings for this pool |
13339 | for (auto& p : osdmap.pg_upmap) { | |
11fdf7f2 | 13340 | if (p.first.pool() == pool) { |
224ce89b WB |
13341 | dout(10) << __func__ << " " << pool |
13342 | << " removing obsolete pg_upmap " | |
13343 | << p.first << dendl; | |
13344 | pending_inc.old_pg_upmap.insert(p.first); | |
13345 | } | |
13346 | } | |
94b18763 FG |
13347 | // remove any pending pg_upmap mappings for this pool |
13348 | { | |
13349 | auto it = pending_inc.new_pg_upmap.begin(); | |
13350 | while (it != pending_inc.new_pg_upmap.end()) { | |
11fdf7f2 | 13351 | if (it->first.pool() == pool) { |
94b18763 FG |
13352 | dout(10) << __func__ << " " << pool |
13353 | << " removing pending pg_upmap " | |
13354 | << it->first << dendl; | |
13355 | it = pending_inc.new_pg_upmap.erase(it); | |
13356 | } else { | |
13357 | it++; | |
13358 | } | |
13359 | } | |
13360 | } | |
224ce89b WB |
13361 | // remove any pg_upmap_items mappings for this pool |
13362 | for (auto& p : osdmap.pg_upmap_items) { | |
11fdf7f2 | 13363 | if (p.first.pool() == pool) { |
224ce89b WB |
13364 | dout(10) << __func__ << " " << pool |
13365 | << " removing obsolete pg_upmap_items " << p.first | |
13366 | << dendl; | |
13367 | pending_inc.old_pg_upmap_items.insert(p.first); | |
13368 | } | |
13369 | } | |
94b18763 FG |
13370 | // remove any pending pg_upmap mappings for this pool |
13371 | { | |
13372 | auto it = pending_inc.new_pg_upmap_items.begin(); | |
13373 | while (it != pending_inc.new_pg_upmap_items.end()) { | |
11fdf7f2 | 13374 | if (it->first.pool() == pool) { |
94b18763 FG |
13375 | dout(10) << __func__ << " " << pool |
13376 | << " removing pending pg_upmap_items " | |
13377 | << it->first << dendl; | |
13378 | it = pending_inc.new_pg_upmap_items.erase(it); | |
13379 | } else { | |
13380 | it++; | |
13381 | } | |
13382 | } | |
13383 | } | |
35e4c445 FG |
13384 | |
13385 | // remove any choose_args for this pool | |
13386 | CrushWrapper newcrush; | |
13387 | _get_pending_crush(newcrush); | |
13388 | if (newcrush.have_choose_args(pool)) { | |
13389 | dout(10) << __func__ << " removing choose_args for pool " << pool << dendl; | |
13390 | newcrush.rm_choose_args(pool); | |
13391 | pending_inc.crush.clear(); | |
13392 | newcrush.encode(pending_inc.crush, mon->get_quorum_con_features()); | |
13393 | } | |
7c673cae FG |
13394 | return 0; |
13395 | } | |
13396 | ||
13397 | int OSDMonitor::_prepare_rename_pool(int64_t pool, string newname) | |
13398 | { | |
13399 | dout(10) << "_prepare_rename_pool " << pool << dendl; | |
13400 | if (pending_inc.old_pools.count(pool)) { | |
13401 | dout(10) << "_prepare_rename_pool " << pool << " pending removal" << dendl; | |
13402 | return -ENOENT; | |
13403 | } | |
13404 | for (map<int64_t,string>::iterator p = pending_inc.new_pool_names.begin(); | |
13405 | p != pending_inc.new_pool_names.end(); | |
13406 | ++p) { | |
13407 | if (p->second == newname && p->first != pool) { | |
13408 | return -EEXIST; | |
13409 | } | |
13410 | } | |
13411 | ||
13412 | pending_inc.new_pool_names[pool] = newname; | |
13413 | return 0; | |
13414 | } | |
13415 | ||
13416 | bool OSDMonitor::prepare_pool_op_delete(MonOpRequestRef op) | |
13417 | { | |
13418 | op->mark_osdmon_event(__func__); | |
13419 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); | |
13420 | ostringstream ss; | |
13421 | int ret = _prepare_remove_pool(m->pool, &ss, false); | |
13422 | if (ret == -EAGAIN) { | |
13423 | wait_for_finished_proposal(op, new C_RetryMessage(this, op)); | |
13424 | return true; | |
13425 | } | |
13426 | if (ret < 0) | |
13427 | dout(10) << __func__ << " got " << ret << " " << ss.str() << dendl; | |
13428 | wait_for_finished_proposal(op, new OSDMonitor::C_PoolOp(this, op, ret, | |
13429 | pending_inc.epoch)); | |
13430 | return true; | |
13431 | } | |
13432 | ||
13433 | void OSDMonitor::_pool_op_reply(MonOpRequestRef op, | |
13434 | int ret, epoch_t epoch, bufferlist *blp) | |
13435 | { | |
13436 | op->mark_osdmon_event(__func__); | |
13437 | MPoolOp *m = static_cast<MPoolOp*>(op->get_req()); | |
13438 | dout(20) << "_pool_op_reply " << ret << dendl; | |
13439 | MPoolOpReply *reply = new MPoolOpReply(m->fsid, m->get_tid(), | |
13440 | ret, epoch, get_last_committed(), blp); | |
13441 | mon->send_reply(op, reply); | |
13442 | } | |
81eedcae TL |
13443 | |
13444 | void OSDMonitor::convert_pool_priorities(void) | |
13445 | { | |
13446 | pool_opts_t::key_t key = pool_opts_t::get_opt_desc("recovery_priority").key; | |
13447 | int64_t max_prio = 0; | |
13448 | int64_t min_prio = 0; | |
13449 | for (const auto &i : osdmap.get_pools()) { | |
13450 | const auto &pool = i.second; | |
13451 | ||
13452 | if (pool.opts.is_set(key)) { | |
13453 | int64_t prio; | |
13454 | pool.opts.get(key, &prio); | |
13455 | if (prio > max_prio) | |
13456 | max_prio = prio; | |
13457 | if (prio < min_prio) | |
13458 | min_prio = prio; | |
13459 | } | |
13460 | } | |
13461 | if (max_prio <= OSD_POOL_PRIORITY_MAX && min_prio >= OSD_POOL_PRIORITY_MIN) { | |
13462 | dout(20) << __func__ << " nothing to fix" << dendl; | |
13463 | return; | |
13464 | } | |
13465 | // Current pool priorities exceeds new maximum | |
13466 | for (const auto &i : osdmap.get_pools()) { | |
13467 | const auto pool_id = i.first; | |
13468 | pg_pool_t pool = i.second; | |
13469 | ||
13470 | int64_t prio = 0; | |
13471 | pool.opts.get(key, &prio); | |
13472 | int64_t n; | |
13473 | ||
13474 | if (prio > 0 && max_prio > OSD_POOL_PRIORITY_MAX) { // Likely scenario | |
13475 | // Scaled priority range 0 to OSD_POOL_PRIORITY_MAX | |
13476 | n = (float)prio / max_prio * OSD_POOL_PRIORITY_MAX; | |
13477 | } else if (prio < 0 && min_prio < OSD_POOL_PRIORITY_MIN) { | |
13478 | // Scaled priority range OSD_POOL_PRIORITY_MIN to 0 | |
13479 | n = (float)prio / min_prio * OSD_POOL_PRIORITY_MIN; | |
13480 | } else { | |
13481 | continue; | |
13482 | } | |
13483 | if (n == 0) { | |
13484 | pool.opts.unset(key); | |
13485 | } else { | |
13486 | pool.opts.set(key, static_cast<int64_t>(n)); | |
13487 | } | |
13488 | dout(10) << __func__ << " pool " << pool_id | |
13489 | << " recovery_priority adjusted " | |
13490 | << prio << " to " << n << dendl; | |
13491 | pool.last_change = pending_inc.epoch; | |
13492 | pending_inc.new_pools[pool_id] = pool; | |
13493 | } | |
13494 | } |