]>
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 | * | |
8 | * This is free software; you can redistribute it and/or | |
9 | * modify it under the terms of the GNU Lesser General Public | |
10 | * License version 2.1, as published by the Free Software | |
11 | * Foundation. See file COPYING. | |
12 | * | |
13 | */ | |
14 | ||
11fdf7f2 | 15 | #include <string_view> |
7c673cae FG |
16 | |
17 | #include "MDSRank.h" | |
18 | #include "MDCache.h" | |
19 | #include "Locker.h" | |
28e407b8 | 20 | #include "MDBalancer.h" |
91327a77 | 21 | #include "Migrator.h" |
7c673cae FG |
22 | #include "CInode.h" |
23 | #include "CDir.h" | |
24 | #include "CDentry.h" | |
25 | #include "Mutation.h" | |
26 | #include "MDSContext.h" | |
27 | ||
28 | #include "MDLog.h" | |
29 | #include "MDSMap.h" | |
30 | ||
31 | #include "events/EUpdate.h" | |
32 | #include "events/EOpen.h" | |
33 | ||
34 | #include "msg/Messenger.h" | |
35 | #include "osdc/Objecter.h" | |
36 | ||
37 | #include "messages/MInodeFileCaps.h" | |
38 | #include "messages/MLock.h" | |
39 | #include "messages/MClientLease.h" | |
40 | #include "messages/MClientReply.h" | |
41 | #include "messages/MClientCaps.h" | |
42 | #include "messages/MClientCapRelease.h" | |
43 | ||
44 | #include "messages/MMDSSlaveRequest.h" | |
45 | ||
46 | #include <errno.h> | |
47 | ||
48 | #include "common/config.h" | |
49 | ||
50 | ||
51 | #define dout_subsys ceph_subsys_mds | |
52 | #undef dout_prefix | |
53 | #define dout_context g_ceph_context | |
54 | #define dout_prefix _prefix(_dout, mds) | |
55 | static ostream& _prefix(std::ostream *_dout, MDSRank *mds) { | |
56 | return *_dout << "mds." << mds->get_nodeid() << ".locker "; | |
57 | } | |
58 | ||
59 | ||
11fdf7f2 | 60 | class LockerContext : public MDSContext { |
7c673cae FG |
61 | protected: |
62 | Locker *locker; | |
63 | MDSRank *get_mds() override | |
64 | { | |
65 | return locker->mds; | |
66 | } | |
67 | ||
68 | public: | |
69 | explicit LockerContext(Locker *locker_) : locker(locker_) { | |
11fdf7f2 | 70 | ceph_assert(locker != NULL); |
7c673cae FG |
71 | } |
72 | }; | |
73 | ||
74 | class LockerLogContext : public MDSLogContextBase { | |
75 | protected: | |
76 | Locker *locker; | |
77 | MDSRank *get_mds() override | |
78 | { | |
79 | return locker->mds; | |
80 | } | |
81 | ||
82 | public: | |
83 | explicit LockerLogContext(Locker *locker_) : locker(locker_) { | |
11fdf7f2 | 84 | ceph_assert(locker != NULL); |
7c673cae FG |
85 | } |
86 | }; | |
87 | ||
11fdf7f2 TL |
88 | Locker::Locker(MDSRank *m, MDCache *c) : |
89 | mds(m), mdcache(c), need_snapflush_inodes(member_offset(CInode, item_caps)) {} | |
90 | ||
91 | ||
92 | void Locker::dispatch(const Message::const_ref &m) | |
7c673cae FG |
93 | { |
94 | ||
95 | switch (m->get_type()) { | |
7c673cae FG |
96 | // inter-mds locking |
97 | case MSG_MDS_LOCK: | |
11fdf7f2 | 98 | handle_lock(MLock::msgref_cast(m)); |
7c673cae FG |
99 | break; |
100 | // inter-mds caps | |
101 | case MSG_MDS_INODEFILECAPS: | |
11fdf7f2 | 102 | handle_inode_file_caps(MInodeFileCaps::msgref_cast(m)); |
7c673cae | 103 | break; |
7c673cae FG |
104 | // client sync |
105 | case CEPH_MSG_CLIENT_CAPS: | |
11fdf7f2 | 106 | handle_client_caps(MClientCaps::msgref_cast(m)); |
7c673cae FG |
107 | break; |
108 | case CEPH_MSG_CLIENT_CAPRELEASE: | |
11fdf7f2 | 109 | handle_client_cap_release(MClientCapRelease::msgref_cast(m)); |
7c673cae FG |
110 | break; |
111 | case CEPH_MSG_CLIENT_LEASE: | |
11fdf7f2 | 112 | handle_client_lease(MClientLease::msgref_cast(m)); |
7c673cae | 113 | break; |
7c673cae FG |
114 | default: |
115 | derr << "locker unknown message " << m->get_type() << dendl; | |
11fdf7f2 | 116 | ceph_abort_msg("locker unknown message"); |
7c673cae FG |
117 | } |
118 | } | |
119 | ||
120 | void Locker::tick() | |
121 | { | |
122 | scatter_tick(); | |
123 | caps_tick(); | |
124 | } | |
125 | ||
126 | /* | |
127 | * locks vs rejoin | |
128 | * | |
129 | * | |
130 | * | |
131 | */ | |
132 | ||
133 | void Locker::send_lock_message(SimpleLock *lock, int msg) | |
134 | { | |
181888fb | 135 | for (const auto &it : lock->get_parent()->get_replicas()) { |
7c673cae | 136 | if (mds->is_cluster_degraded() && |
181888fb | 137 | mds->mdsmap->get_state(it.first) < MDSMap::STATE_REJOIN) |
7c673cae | 138 | continue; |
11fdf7f2 | 139 | auto m = MLock::create(lock, msg, mds->get_nodeid()); |
181888fb | 140 | mds->send_message_mds(m, it.first); |
7c673cae FG |
141 | } |
142 | } | |
143 | ||
144 | void Locker::send_lock_message(SimpleLock *lock, int msg, const bufferlist &data) | |
145 | { | |
181888fb | 146 | for (const auto &it : lock->get_parent()->get_replicas()) { |
7c673cae | 147 | if (mds->is_cluster_degraded() && |
181888fb | 148 | mds->mdsmap->get_state(it.first) < MDSMap::STATE_REJOIN) |
7c673cae | 149 | continue; |
11fdf7f2 | 150 | auto m = MLock::create(lock, msg, mds->get_nodeid()); |
7c673cae | 151 | m->set_data(data); |
181888fb | 152 | mds->send_message_mds(m, it.first); |
7c673cae FG |
153 | } |
154 | } | |
155 | ||
156 | ||
157 | ||
158 | ||
11fdf7f2 | 159 | void Locker::include_snap_rdlocks(CInode *in, MutationImpl::LockOpVec& lov) |
7c673cae FG |
160 | { |
161 | // rdlock ancestor snaps | |
162 | CInode *t = in; | |
7c673cae FG |
163 | while (t->get_projected_parent_dn()) { |
164 | t = t->get_projected_parent_dn()->get_dir()->get_inode(); | |
11fdf7f2 | 165 | lov.add_rdlock(&t->snaplock); |
7c673cae | 166 | } |
11fdf7f2 | 167 | lov.add_rdlock(&in->snaplock); |
7c673cae FG |
168 | } |
169 | ||
11fdf7f2 | 170 | void Locker::include_snap_rdlocks_wlayout(CInode *in, MutationImpl::LockOpVec& lov, |
7c673cae FG |
171 | file_layout_t **layout) |
172 | { | |
173 | //rdlock ancestor snaps | |
174 | CInode *t = in; | |
11fdf7f2 TL |
175 | lov.add_rdlock(&in->snaplock); |
176 | lov.add_rdlock(&in->policylock); | |
7c673cae FG |
177 | bool found_layout = false; |
178 | while (t) { | |
11fdf7f2 | 179 | lov.add_rdlock(&t->snaplock); |
7c673cae | 180 | if (!found_layout) { |
11fdf7f2 | 181 | lov.add_rdlock(&t->policylock); |
7c673cae FG |
182 | if (t->get_projected_inode()->has_layout()) { |
183 | *layout = &t->get_projected_inode()->layout; | |
184 | found_layout = true; | |
185 | } | |
186 | } | |
187 | if (t->get_projected_parent_dn() && | |
188 | t->get_projected_parent_dn()->get_dir()) | |
189 | t = t->get_projected_parent_dn()->get_dir()->get_inode(); | |
190 | else t = NULL; | |
191 | } | |
192 | } | |
193 | ||
194 | struct MarkEventOnDestruct { | |
195 | MDRequestRef& mdr; | |
11fdf7f2 | 196 | std::string_view message; |
7c673cae | 197 | bool mark_event; |
11fdf7f2 TL |
198 | MarkEventOnDestruct(MDRequestRef& _mdr, std::string_view _message) : |
199 | mdr(_mdr), | |
200 | message(_message), | |
201 | mark_event(true) {} | |
7c673cae FG |
202 | ~MarkEventOnDestruct() { |
203 | if (mark_event) | |
204 | mdr->mark_event(message); | |
205 | } | |
206 | }; | |
207 | ||
208 | /* If this function returns false, the mdr has been placed | |
209 | * on the appropriate wait list */ | |
210 | bool Locker::acquire_locks(MDRequestRef& mdr, | |
11fdf7f2 | 211 | MutationImpl::LockOpVec& lov, |
7c673cae FG |
212 | CInode *auth_pin_freeze, |
213 | bool auth_pin_nonblock) | |
214 | { | |
215 | if (mdr->done_locking && | |
216 | !mdr->is_slave()) { // not on slaves! master requests locks piecemeal. | |
217 | dout(10) << "acquire_locks " << *mdr << " - done locking" << dendl; | |
218 | return true; // at least we had better be! | |
219 | } | |
220 | dout(10) << "acquire_locks " << *mdr << dendl; | |
221 | ||
222 | MarkEventOnDestruct marker(mdr, "failed to acquire_locks"); | |
223 | ||
224 | client_t client = mdr->get_client(); | |
225 | ||
11fdf7f2 | 226 | set<MDSCacheObject*> mustpin; // items to authpin |
7c673cae FG |
227 | |
228 | // xlocks | |
11fdf7f2 TL |
229 | for (int i = 0, size = lov.size(); i < size; ++i) { |
230 | auto& p = lov[i]; | |
231 | SimpleLock *lock = p.lock; | |
232 | MDSCacheObject *object = lock->get_parent(); | |
233 | ||
234 | if (p.is_xlock()) { | |
235 | if ((lock->get_type() == CEPH_LOCK_ISNAP || | |
236 | lock->get_type() == CEPH_LOCK_IPOLICY) && | |
237 | mds->is_cluster_degraded() && | |
238 | mdr->is_master() && | |
239 | !mdr->is_queued_for_replay()) { | |
240 | // waiting for recovering mds, to guarantee replayed requests and mksnap/setlayout | |
241 | // get processed in proper order. | |
242 | bool wait = false; | |
243 | if (object->is_auth()) { | |
244 | if (!mdr->locks.count(lock)) { | |
245 | set<mds_rank_t> ls; | |
246 | object->list_replicas(ls); | |
247 | for (auto m : ls) { | |
248 | if (mds->mdsmap->get_state(m) < MDSMap::STATE_ACTIVE) { | |
249 | wait = true; | |
250 | break; | |
251 | } | |
b32b8144 FG |
252 | } |
253 | } | |
11fdf7f2 TL |
254 | } else { |
255 | // if the lock is the latest locked one, it's possible that slave mds got the lock | |
256 | // while there are recovering mds. | |
257 | if (!mdr->locks.count(lock) || lock == *mdr->locks.rbegin()) | |
258 | wait = true; | |
259 | } | |
260 | if (wait) { | |
261 | dout(10) << " must xlock " << *lock << " " << *object | |
262 | << ", waiting for cluster recovered" << dendl; | |
263 | mds->locker->drop_locks(mdr.get(), NULL); | |
264 | mdr->drop_local_auth_pins(); | |
265 | mds->wait_for_cluster_recovered(new C_MDS_RetryRequest(mdcache, mdr)); | |
266 | return false; | |
b32b8144 | 267 | } |
b32b8144 | 268 | } |
7c673cae | 269 | |
11fdf7f2 | 270 | dout(20) << " must xlock " << *lock << " " << *object << dendl; |
7c673cae | 271 | |
11fdf7f2 | 272 | mustpin.insert(object); |
7c673cae | 273 | |
11fdf7f2 TL |
274 | // augment xlock with a versionlock? |
275 | if (lock->get_type() == CEPH_LOCK_DN) { | |
276 | CDentry *dn = static_cast<CDentry*>(object); | |
277 | if (!dn->is_auth()) | |
278 | continue; | |
279 | if (mdr->is_master()) { | |
280 | // master. wrlock versionlock so we can pipeline dentry updates to journal. | |
281 | lov.add_wrlock(&dn->versionlock); | |
282 | } else { | |
283 | // slave. exclusively lock the dentry version (i.e. block other journal updates). | |
284 | // this makes rollback safe. | |
285 | lov.add_xlock(&dn->versionlock); | |
286 | } | |
7c673cae | 287 | } |
11fdf7f2 TL |
288 | if (lock->get_type() > CEPH_LOCK_IVERSION) { |
289 | // inode version lock? | |
290 | CInode *in = static_cast<CInode*>(object); | |
291 | if (!in->is_auth()) | |
292 | continue; | |
293 | if (mdr->is_master()) { | |
294 | // master. wrlock versionlock so we can pipeline inode updates to journal. | |
295 | lov.add_wrlock(&in->versionlock); | |
296 | } else { | |
297 | // slave. exclusively lock the inode version (i.e. block other journal updates). | |
298 | // this makes rollback safe. | |
299 | lov.add_xlock(&in->versionlock); | |
300 | } | |
7c673cae | 301 | } |
11fdf7f2 TL |
302 | } else if (p.is_wrlock()) { |
303 | dout(20) << " must wrlock " << *lock << " " << *object << dendl; | |
304 | if (object->is_auth()) { | |
305 | mustpin.insert(object); | |
306 | } else if (!object->is_auth() && | |
307 | !lock->can_wrlock(client) && // we might have to request a scatter | |
308 | !mdr->is_slave()) { // if we are slave (remote_wrlock), the master already authpinned | |
309 | dout(15) << " will also auth_pin " << *object | |
310 | << " in case we need to request a scatter" << dendl; | |
311 | mustpin.insert(object); | |
312 | } | |
313 | } else if (p.is_remote_wrlock()) { | |
314 | dout(20) << " must remote_wrlock on mds." << p.wrlock_target << " " | |
315 | << *lock << " " << *object << dendl; | |
7c673cae | 316 | mustpin.insert(object); |
11fdf7f2 TL |
317 | } else if (p.is_rdlock()) { |
318 | ||
319 | dout(20) << " must rdlock " << *lock << " " << *object << dendl; | |
320 | if (object->is_auth()) { | |
321 | mustpin.insert(object); | |
322 | } else if (!object->is_auth() && | |
323 | !lock->can_rdlock(client)) { // we might have to request an rdlock | |
324 | dout(15) << " will also auth_pin " << *object | |
325 | << " in case we need to request a rdlock" << dendl; | |
326 | mustpin.insert(object); | |
327 | } | |
328 | } else { | |
329 | ceph_assert(0 == "locker unknown lock operation"); | |
7c673cae FG |
330 | } |
331 | } | |
332 | ||
11fdf7f2 | 333 | lov.sort_and_merge(); |
7c673cae FG |
334 | |
335 | // AUTH PINS | |
336 | map<mds_rank_t, set<MDSCacheObject*> > mustpin_remote; // mds -> (object set) | |
337 | ||
338 | // can i auth pin them all now? | |
339 | marker.message = "failed to authpin local pins"; | |
11fdf7f2 TL |
340 | for (const auto &p : mustpin) { |
341 | MDSCacheObject *object = p; | |
7c673cae FG |
342 | |
343 | dout(10) << " must authpin " << *object << dendl; | |
344 | ||
345 | if (mdr->is_auth_pinned(object)) { | |
346 | if (object != (MDSCacheObject*)auth_pin_freeze) | |
347 | continue; | |
348 | if (mdr->more()->is_remote_frozen_authpin) { | |
349 | if (mdr->more()->rename_inode == auth_pin_freeze) | |
350 | continue; | |
351 | // unfreeze auth pin for the wrong inode | |
352 | mustpin_remote[mdr->more()->rename_inode->authority().first].size(); | |
353 | } | |
354 | } | |
355 | ||
356 | if (!object->is_auth()) { | |
357 | if (!mdr->locks.empty()) | |
358 | drop_locks(mdr.get()); | |
359 | if (object->is_ambiguous_auth()) { | |
360 | // wait | |
91327a77 | 361 | marker.message = "waiting for single auth, object is being migrated"; |
7c673cae FG |
362 | dout(10) << " ambiguous auth, waiting to authpin " << *object << dendl; |
363 | object->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, new C_MDS_RetryRequest(mdcache, mdr)); | |
364 | mdr->drop_local_auth_pins(); | |
365 | return false; | |
366 | } | |
367 | mustpin_remote[object->authority().first].insert(object); | |
368 | continue; | |
369 | } | |
91327a77 AA |
370 | int err = 0; |
371 | if (!object->can_auth_pin(&err)) { | |
7c673cae FG |
372 | // wait |
373 | drop_locks(mdr.get()); | |
374 | mdr->drop_local_auth_pins(); | |
375 | if (auth_pin_nonblock) { | |
376 | dout(10) << " can't auth_pin (freezing?) " << *object << ", nonblocking" << dendl; | |
377 | mdr->aborted = true; | |
378 | return false; | |
379 | } | |
91327a77 AA |
380 | if (err == MDSCacheObject::ERR_EXPORTING_TREE) { |
381 | marker.message = "failed to authpin, subtree is being exported"; | |
382 | } else if (err == MDSCacheObject::ERR_FRAGMENTING_DIR) { | |
383 | marker.message = "failed to authpin, dir is being fragmented"; | |
384 | } else if (err == MDSCacheObject::ERR_EXPORTING_INODE) { | |
385 | marker.message = "failed to authpin, inode is being exported"; | |
386 | } | |
7c673cae FG |
387 | dout(10) << " can't auth_pin (freezing?), waiting to authpin " << *object << dendl; |
388 | object->add_waiter(MDSCacheObject::WAIT_UNFREEZE, new C_MDS_RetryRequest(mdcache, mdr)); | |
224ce89b WB |
389 | |
390 | if (!mdr->remote_auth_pins.empty()) | |
391 | notify_freeze_waiter(object); | |
392 | ||
7c673cae FG |
393 | return false; |
394 | } | |
395 | } | |
396 | ||
397 | // ok, grab local auth pins | |
11fdf7f2 TL |
398 | for (const auto& p : mustpin) { |
399 | MDSCacheObject *object = p; | |
7c673cae FG |
400 | if (mdr->is_auth_pinned(object)) { |
401 | dout(10) << " already auth_pinned " << *object << dendl; | |
402 | } else if (object->is_auth()) { | |
403 | dout(10) << " auth_pinning " << *object << dendl; | |
404 | mdr->auth_pin(object); | |
405 | } | |
406 | } | |
407 | ||
408 | // request remote auth_pins | |
409 | if (!mustpin_remote.empty()) { | |
410 | marker.message = "requesting remote authpins"; | |
11fdf7f2 TL |
411 | for (const auto& p : mdr->remote_auth_pins) { |
412 | if (mustpin.count(p.first)) { | |
413 | ceph_assert(p.second == p.first->authority().first); | |
414 | map<mds_rank_t, set<MDSCacheObject*> >::iterator q = mustpin_remote.find(p.second); | |
7c673cae | 415 | if (q != mustpin_remote.end()) |
11fdf7f2 | 416 | q->second.insert(p.first); |
7c673cae FG |
417 | } |
418 | } | |
419 | for (map<mds_rank_t, set<MDSCacheObject*> >::iterator p = mustpin_remote.begin(); | |
420 | p != mustpin_remote.end(); | |
421 | ++p) { | |
422 | dout(10) << "requesting remote auth_pins from mds." << p->first << dendl; | |
423 | ||
424 | // wait for active auth | |
425 | if (mds->is_cluster_degraded() && | |
426 | !mds->mdsmap->is_clientreplay_or_active_or_stopping(p->first)) { | |
427 | dout(10) << " mds." << p->first << " is not active" << dendl; | |
428 | if (mdr->more()->waiting_on_slave.empty()) | |
429 | mds->wait_for_active_peer(p->first, new C_MDS_RetryRequest(mdcache, mdr)); | |
430 | return false; | |
431 | } | |
432 | ||
11fdf7f2 | 433 | auto req = MMDSSlaveRequest::create(mdr->reqid, mdr->attempt, MMDSSlaveRequest::OP_AUTHPIN); |
7c673cae FG |
434 | for (set<MDSCacheObject*>::iterator q = p->second.begin(); |
435 | q != p->second.end(); | |
436 | ++q) { | |
437 | dout(10) << " req remote auth_pin of " << **q << dendl; | |
438 | MDSCacheObjectInfo info; | |
439 | (*q)->set_object_info(info); | |
440 | req->get_authpins().push_back(info); | |
441 | if (*q == auth_pin_freeze) | |
442 | (*q)->set_object_info(req->get_authpin_freeze()); | |
443 | mdr->pin(*q); | |
444 | } | |
445 | if (auth_pin_nonblock) | |
446 | req->mark_nonblock(); | |
447 | mds->send_message_mds(req, p->first); | |
448 | ||
449 | // put in waiting list | |
11fdf7f2 | 450 | ceph_assert(mdr->more()->waiting_on_slave.count(p->first) == 0); |
7c673cae FG |
451 | mdr->more()->waiting_on_slave.insert(p->first); |
452 | } | |
453 | return false; | |
454 | } | |
455 | ||
456 | // caps i'll need to issue | |
457 | set<CInode*> issue_set; | |
458 | bool result = false; | |
459 | ||
460 | // acquire locks. | |
461 | // make sure they match currently acquired locks. | |
11fdf7f2 TL |
462 | auto existing = mdr->locks.begin(); |
463 | for (const auto& p : lov) { | |
464 | bool need_wrlock = p.is_wrlock(); | |
465 | bool need_remote_wrlock = p.is_remote_wrlock(); | |
7c673cae FG |
466 | |
467 | // already locked? | |
11fdf7f2 | 468 | if (existing != mdr->locks.end() && existing->lock == p.lock) { |
7c673cae | 469 | // right kind? |
11fdf7f2 TL |
470 | auto it = existing++; |
471 | auto have = *it; // don't reference | |
472 | ||
473 | if (have.is_xlock() && p.is_xlock()) { | |
474 | dout(10) << " already xlocked " << *have.lock << " " << *have.lock->get_parent() << dendl; | |
7c673cae FG |
475 | continue; |
476 | } | |
11fdf7f2 TL |
477 | |
478 | if (have.is_remote_wrlock() && | |
479 | (!need_remote_wrlock || have.wrlock_target != p.wrlock_target)) { | |
480 | dout(10) << " unlocking remote_wrlock on wrong mds." << have.wrlock_target | |
481 | << " " << *have.lock << " " << *have.lock->get_parent() << dendl; | |
482 | remote_wrlock_finish(it, mdr.get()); | |
483 | have.clear_remote_wrlock(); | |
7c673cae | 484 | } |
11fdf7f2 | 485 | |
7c673cae | 486 | if (need_wrlock || need_remote_wrlock) { |
11fdf7f2 TL |
487 | if (need_wrlock == have.is_wrlock() && |
488 | need_remote_wrlock == have.is_remote_wrlock()) { | |
7c673cae | 489 | if (need_wrlock) |
11fdf7f2 | 490 | dout(10) << " already wrlocked " << *have.lock << " " << *have.lock->get_parent() << dendl; |
7c673cae | 491 | if (need_remote_wrlock) |
11fdf7f2 | 492 | dout(10) << " already remote_wrlocked " << *have.lock << " " << *have.lock->get_parent() << dendl; |
7c673cae FG |
493 | continue; |
494 | } | |
11fdf7f2 TL |
495 | |
496 | if (have.is_wrlock()) { | |
497 | if (!need_wrlock) | |
498 | dout(10) << " unlocking extra " << *have.lock << " " << *have.lock->get_parent() << dendl; | |
499 | else if (need_remote_wrlock) // acquire remote_wrlock first | |
500 | dout(10) << " unlocking out-of-order " << *have.lock << " " << *have.lock->get_parent() << dendl; | |
501 | bool need_issue = false; | |
502 | wrlock_finish(it, mdr.get(), &need_issue); | |
503 | if (need_issue) | |
504 | issue_set.insert(static_cast<CInode*>(have.lock->get_parent())); | |
505 | } | |
506 | } else if (have.is_rdlock() && p.is_rdlock()) { | |
507 | dout(10) << " already rdlocked " << *have.lock << " " << *have.lock->get_parent() << dendl; | |
7c673cae FG |
508 | continue; |
509 | } | |
510 | } | |
511 | ||
512 | // hose any stray locks | |
7c673cae | 513 | while (existing != mdr->locks.end()) { |
11fdf7f2 TL |
514 | auto it = existing++; |
515 | auto stray = *it; // don't reference | |
516 | dout(10) << " unlocking out-of-order " << *stray.lock << " " << *stray.lock->get_parent() << dendl; | |
7c673cae | 517 | bool need_issue = false; |
11fdf7f2 TL |
518 | if (stray.is_xlock()) { |
519 | xlock_finish(it, mdr.get(), &need_issue); | |
520 | } else if (stray.is_rdlock()) { | |
521 | rdlock_finish(it, mdr.get(), &need_issue); | |
7c673cae FG |
522 | } else { |
523 | // may have acquired both wrlock and remore wrlock | |
11fdf7f2 TL |
524 | if (stray.is_wrlock()) |
525 | wrlock_finish(it, mdr.get(), &need_issue); | |
526 | if (stray.is_remote_wrlock()) | |
527 | remote_wrlock_finish(it, mdr.get()); | |
7c673cae FG |
528 | } |
529 | if (need_issue) | |
11fdf7f2 | 530 | issue_set.insert(static_cast<CInode*>(stray.lock->get_parent())); |
7c673cae FG |
531 | } |
532 | ||
533 | // lock | |
11fdf7f2 | 534 | if (mdr->locking && p.lock != mdr->locking) { |
7c673cae FG |
535 | cancel_locking(mdr.get(), &issue_set); |
536 | } | |
11fdf7f2 | 537 | if (p.is_xlock()) { |
7c673cae | 538 | marker.message = "failed to xlock, waiting"; |
11fdf7f2 | 539 | if (!xlock_start(p.lock, mdr)) |
7c673cae | 540 | goto out; |
11fdf7f2 | 541 | dout(10) << " got xlock on " << *p.lock << " " << *p.lock->get_parent() << dendl; |
7c673cae | 542 | } else if (need_wrlock || need_remote_wrlock) { |
11fdf7f2 | 543 | if (need_remote_wrlock && !mdr->is_remote_wrlocked(p)) { |
7c673cae | 544 | marker.message = "waiting for remote wrlocks"; |
11fdf7f2 | 545 | remote_wrlock_start(p, p.wrlock_target, mdr); |
7c673cae FG |
546 | goto out; |
547 | } | |
11fdf7f2 | 548 | if (need_wrlock) { |
7c673cae | 549 | marker.message = "failed to wrlock, waiting"; |
11fdf7f2 | 550 | if (need_remote_wrlock && !p.lock->can_wrlock(mdr->get_client())) { |
7c673cae FG |
551 | marker.message = "failed to wrlock, dropping remote wrlock and waiting"; |
552 | // can't take the wrlock because the scatter lock is gathering. need to | |
553 | // release the remote wrlock, so that the gathering process can finish. | |
11fdf7f2 TL |
554 | auto it = mdr->locks.end(); |
555 | ++it; | |
556 | remote_wrlock_finish(it, mdr.get()); | |
557 | remote_wrlock_start(p, p.wrlock_target, mdr); | |
7c673cae FG |
558 | goto out; |
559 | } | |
560 | // nowait if we have already gotten remote wrlock | |
11fdf7f2 | 561 | if (!wrlock_start(p, mdr, need_remote_wrlock)) |
7c673cae | 562 | goto out; |
11fdf7f2 | 563 | dout(10) << " got wrlock on " << *p.lock << " " << *p.lock->get_parent() << dendl; |
7c673cae FG |
564 | } |
565 | } else { | |
11fdf7f2 TL |
566 | ceph_assert(mdr->is_master()); |
567 | if (p.lock->needs_recover()) { | |
b32b8144 FG |
568 | if (mds->is_cluster_degraded()) { |
569 | if (!mdr->is_queued_for_replay()) { | |
570 | // see comments in SimpleLock::set_state_rejoin() and | |
571 | // ScatterLock::encode_state_for_rejoin() | |
572 | drop_locks(mdr.get()); | |
573 | mds->wait_for_cluster_recovered(new C_MDS_RetryRequest(mdcache, mdr)); | |
11fdf7f2 | 574 | dout(10) << " rejoin recovering " << *p.lock << " " << *p.lock->get_parent() |
b32b8144 FG |
575 | << ", waiting for cluster recovered" << dendl; |
576 | marker.message = "rejoin recovering lock, waiting for cluster recovered"; | |
577 | return false; | |
7c673cae | 578 | } |
b32b8144 | 579 | } else { |
11fdf7f2 | 580 | p.lock->clear_need_recover(); |
7c673cae FG |
581 | } |
582 | } | |
583 | ||
584 | marker.message = "failed to rdlock, waiting"; | |
11fdf7f2 | 585 | if (!rdlock_start(p, mdr)) |
7c673cae | 586 | goto out; |
11fdf7f2 | 587 | dout(10) << " got rdlock on " << *p.lock << " " << *p.lock->get_parent() << dendl; |
7c673cae FG |
588 | } |
589 | } | |
590 | ||
591 | // any extra unneeded locks? | |
592 | while (existing != mdr->locks.end()) { | |
11fdf7f2 TL |
593 | auto it = existing++; |
594 | auto stray = *it; | |
595 | dout(10) << " unlocking extra " << *stray.lock << " " << *stray.lock->get_parent() << dendl; | |
7c673cae | 596 | bool need_issue = false; |
11fdf7f2 TL |
597 | if (stray.is_xlock()) { |
598 | xlock_finish(it, mdr.get(), &need_issue); | |
599 | } else if (stray.is_rdlock()) { | |
600 | rdlock_finish(it, mdr.get(), &need_issue); | |
7c673cae FG |
601 | } else { |
602 | // may have acquired both wrlock and remore wrlock | |
11fdf7f2 TL |
603 | if (stray.is_wrlock()) |
604 | wrlock_finish(it, mdr.get(), &need_issue); | |
605 | if (stray.is_remote_wrlock()) | |
606 | remote_wrlock_finish(it, mdr.get()); | |
7c673cae FG |
607 | } |
608 | if (need_issue) | |
11fdf7f2 | 609 | issue_set.insert(static_cast<CInode*>(stray.lock->get_parent())); |
7c673cae FG |
610 | } |
611 | ||
612 | mdr->done_locking = true; | |
613 | mdr->set_mds_stamp(ceph_clock_now()); | |
614 | result = true; | |
615 | marker.message = "acquired locks"; | |
616 | ||
617 | out: | |
618 | issue_caps_set(issue_set); | |
619 | return result; | |
620 | } | |
621 | ||
224ce89b WB |
622 | void Locker::notify_freeze_waiter(MDSCacheObject *o) |
623 | { | |
624 | CDir *dir = NULL; | |
625 | if (CInode *in = dynamic_cast<CInode*>(o)) { | |
626 | if (!in->is_root()) | |
627 | dir = in->get_parent_dir(); | |
628 | } else if (CDentry *dn = dynamic_cast<CDentry*>(o)) { | |
629 | dir = dn->get_dir(); | |
630 | } else { | |
631 | dir = dynamic_cast<CDir*>(o); | |
11fdf7f2 | 632 | ceph_assert(dir); |
224ce89b WB |
633 | } |
634 | if (dir) { | |
635 | if (dir->is_freezing_dir()) | |
636 | mdcache->fragment_freeze_inc_num_waiters(dir); | |
637 | if (dir->is_freezing_tree()) { | |
638 | while (!dir->is_freezing_tree_root()) | |
639 | dir = dir->get_parent_dir(); | |
640 | mdcache->migrator->export_freeze_inc_num_waiters(dir); | |
641 | } | |
642 | } | |
643 | } | |
7c673cae FG |
644 | |
645 | void Locker::set_xlocks_done(MutationImpl *mut, bool skip_dentry) | |
646 | { | |
11fdf7f2 TL |
647 | for (const auto &p : mut->locks) { |
648 | if (!p.is_xlock()) | |
649 | continue; | |
650 | MDSCacheObject *obj = p.lock->get_parent(); | |
651 | ceph_assert(obj->is_auth()); | |
7c673cae | 652 | if (skip_dentry && |
11fdf7f2 | 653 | (p.lock->get_type() == CEPH_LOCK_DN || p.lock->get_type() == CEPH_LOCK_DVERSION)) |
7c673cae | 654 | continue; |
11fdf7f2 TL |
655 | dout(10) << "set_xlocks_done on " << *p.lock << " " << *obj << dendl; |
656 | p.lock->set_xlock_done(); | |
7c673cae FG |
657 | } |
658 | } | |
659 | ||
11fdf7f2 TL |
660 | void Locker::_drop_locks(MutationImpl *mut, set<CInode*> *pneed_issue, |
661 | bool drop_rdlocks) | |
7c673cae FG |
662 | { |
663 | set<mds_rank_t> slaves; | |
664 | ||
11fdf7f2 TL |
665 | for (auto it = mut->locks.begin(); it != mut->locks.end(); ) { |
666 | SimpleLock *lock = it->lock; | |
667 | MDSCacheObject *obj = lock->get_parent(); | |
668 | ||
669 | if (it->is_xlock()) { | |
670 | if (obj->is_auth()) { | |
671 | bool ni = false; | |
672 | xlock_finish(it++, mut, &ni); | |
673 | if (ni) | |
674 | pneed_issue->insert(static_cast<CInode*>(obj)); | |
675 | } else { | |
676 | ceph_assert(lock->get_sm()->can_remote_xlock); | |
677 | slaves.insert(obj->authority().first); | |
678 | lock->put_xlock(); | |
679 | mut->locks.erase(it++); | |
680 | } | |
681 | } else if (it->is_wrlock() || it->is_remote_wrlock()) { | |
682 | if (it->is_remote_wrlock()) { | |
683 | slaves.insert(it->wrlock_target); | |
684 | it->clear_remote_wrlock(); | |
685 | } | |
686 | if (it->is_wrlock()) { | |
687 | bool ni = false; | |
688 | wrlock_finish(it++, mut, &ni); | |
689 | if (ni) | |
690 | pneed_issue->insert(static_cast<CInode*>(obj)); | |
691 | } else { | |
692 | mut->locks.erase(it++); | |
693 | } | |
694 | } else if (drop_rdlocks && it->is_rdlock()) { | |
695 | bool ni = false; | |
696 | rdlock_finish(it++, mut, &ni); | |
697 | if (ni) | |
698 | pneed_issue->insert(static_cast<CInode*>(obj)); | |
699 | } else { | |
700 | ++it; | |
7c673cae | 701 | } |
7c673cae FG |
702 | } |
703 | ||
704 | for (set<mds_rank_t>::iterator p = slaves.begin(); p != slaves.end(); ++p) { | |
705 | if (!mds->is_cluster_degraded() || | |
706 | mds->mdsmap->get_state(*p) >= MDSMap::STATE_REJOIN) { | |
707 | dout(10) << "_drop_non_rdlocks dropping remote locks on mds." << *p << dendl; | |
11fdf7f2 | 708 | auto slavereq = MMDSSlaveRequest::create(mut->reqid, mut->attempt, MMDSSlaveRequest::OP_DROPLOCKS); |
7c673cae FG |
709 | mds->send_message_mds(slavereq, *p); |
710 | } | |
711 | } | |
712 | } | |
713 | ||
714 | void Locker::cancel_locking(MutationImpl *mut, set<CInode*> *pneed_issue) | |
715 | { | |
716 | SimpleLock *lock = mut->locking; | |
11fdf7f2 | 717 | ceph_assert(lock); |
7c673cae FG |
718 | dout(10) << "cancel_locking " << *lock << " on " << *mut << dendl; |
719 | ||
720 | if (lock->get_parent()->is_auth()) { | |
721 | bool need_issue = false; | |
722 | if (lock->get_state() == LOCK_PREXLOCK) { | |
723 | _finish_xlock(lock, -1, &need_issue); | |
a8e16298 | 724 | } else if (lock->get_state() == LOCK_LOCK_XLOCK) { |
7c673cae FG |
725 | lock->set_state(LOCK_XLOCKDONE); |
726 | eval_gather(lock, true, &need_issue); | |
727 | } | |
728 | if (need_issue) | |
729 | pneed_issue->insert(static_cast<CInode *>(lock->get_parent())); | |
730 | } | |
731 | mut->finish_locking(lock); | |
732 | } | |
733 | ||
734 | void Locker::drop_locks(MutationImpl *mut, set<CInode*> *pneed_issue) | |
735 | { | |
736 | // leftover locks | |
737 | set<CInode*> my_need_issue; | |
738 | if (!pneed_issue) | |
739 | pneed_issue = &my_need_issue; | |
740 | ||
741 | if (mut->locking) | |
742 | cancel_locking(mut, pneed_issue); | |
11fdf7f2 | 743 | _drop_locks(mut, pneed_issue, true); |
7c673cae FG |
744 | |
745 | if (pneed_issue == &my_need_issue) | |
746 | issue_caps_set(*pneed_issue); | |
747 | mut->done_locking = false; | |
748 | } | |
749 | ||
750 | void Locker::drop_non_rdlocks(MutationImpl *mut, set<CInode*> *pneed_issue) | |
751 | { | |
752 | set<CInode*> my_need_issue; | |
753 | if (!pneed_issue) | |
754 | pneed_issue = &my_need_issue; | |
755 | ||
11fdf7f2 | 756 | _drop_locks(mut, pneed_issue, false); |
7c673cae FG |
757 | |
758 | if (pneed_issue == &my_need_issue) | |
759 | issue_caps_set(*pneed_issue); | |
760 | } | |
761 | ||
b32b8144 | 762 | void Locker::drop_rdlocks_for_early_reply(MutationImpl *mut) |
7c673cae | 763 | { |
b32b8144 | 764 | set<CInode*> need_issue; |
7c673cae | 765 | |
11fdf7f2 TL |
766 | for (auto it = mut->locks.begin(); it != mut->locks.end(); ) { |
767 | if (!it->is_rdlock()) { | |
768 | ++it; | |
769 | continue; | |
770 | } | |
771 | SimpleLock *lock = it->lock; | |
b32b8144 FG |
772 | // make later mksnap/setlayout (at other mds) wait for this unsafe request |
773 | if (lock->get_type() == CEPH_LOCK_ISNAP || | |
11fdf7f2 TL |
774 | lock->get_type() == CEPH_LOCK_IPOLICY) { |
775 | ++it; | |
b32b8144 | 776 | continue; |
11fdf7f2 | 777 | } |
b32b8144 | 778 | bool ni = false; |
11fdf7f2 | 779 | rdlock_finish(it++, mut, &ni); |
b32b8144 FG |
780 | if (ni) |
781 | need_issue.insert(static_cast<CInode*>(lock->get_parent())); | |
782 | } | |
7c673cae | 783 | |
b32b8144 | 784 | issue_caps_set(need_issue); |
7c673cae FG |
785 | } |
786 | ||
a8e16298 TL |
787 | void Locker::drop_locks_for_fragment_unfreeze(MutationImpl *mut) |
788 | { | |
789 | set<CInode*> need_issue; | |
790 | ||
791 | for (auto it = mut->locks.begin(); it != mut->locks.end(); ) { | |
11fdf7f2 | 792 | SimpleLock *lock = it->lock; |
a8e16298 | 793 | if (lock->get_type() == CEPH_LOCK_IDFT) { |
11fdf7f2 | 794 | ++it; |
a8e16298 TL |
795 | continue; |
796 | } | |
797 | bool ni = false; | |
11fdf7f2 | 798 | wrlock_finish(it++, mut, &ni); |
a8e16298 TL |
799 | if (ni) |
800 | need_issue.insert(static_cast<CInode*>(lock->get_parent())); | |
801 | } | |
802 | issue_caps_set(need_issue); | |
803 | } | |
804 | ||
7c673cae FG |
805 | // generics |
806 | ||
11fdf7f2 | 807 | void Locker::eval_gather(SimpleLock *lock, bool first, bool *pneed_issue, MDSContext::vec *pfinishers) |
7c673cae FG |
808 | { |
809 | dout(10) << "eval_gather " << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 | 810 | ceph_assert(!lock->is_stable()); |
7c673cae FG |
811 | |
812 | int next = lock->get_next_state(); | |
813 | ||
814 | CInode *in = 0; | |
815 | bool caps = lock->get_cap_shift(); | |
816 | if (lock->get_type() != CEPH_LOCK_DN) | |
817 | in = static_cast<CInode *>(lock->get_parent()); | |
818 | ||
819 | bool need_issue = false; | |
820 | ||
821 | int loner_issued = 0, other_issued = 0, xlocker_issued = 0; | |
11fdf7f2 | 822 | ceph_assert(!caps || in != NULL); |
7c673cae FG |
823 | if (caps && in->is_head()) { |
824 | in->get_caps_issued(&loner_issued, &other_issued, &xlocker_issued, | |
825 | lock->get_cap_shift(), lock->get_cap_mask()); | |
826 | dout(10) << " next state is " << lock->get_state_name(next) | |
827 | << " issued/allows loner " << gcap_string(loner_issued) | |
828 | << "/" << gcap_string(lock->gcaps_allowed(CAP_LONER, next)) | |
829 | << " xlocker " << gcap_string(xlocker_issued) | |
830 | << "/" << gcap_string(lock->gcaps_allowed(CAP_XLOCKER, next)) | |
831 | << " other " << gcap_string(other_issued) | |
832 | << "/" << gcap_string(lock->gcaps_allowed(CAP_ANY, next)) | |
833 | << dendl; | |
834 | ||
835 | if (first && ((~lock->gcaps_allowed(CAP_ANY, next) & other_issued) || | |
836 | (~lock->gcaps_allowed(CAP_LONER, next) & loner_issued) || | |
837 | (~lock->gcaps_allowed(CAP_XLOCKER, next) & xlocker_issued))) | |
838 | need_issue = true; | |
839 | } | |
840 | ||
841 | #define IS_TRUE_AND_LT_AUTH(x, auth) (x && ((auth && x <= AUTH) || (!auth && x < AUTH))) | |
842 | bool auth = lock->get_parent()->is_auth(); | |
843 | if (!lock->is_gathering() && | |
844 | (IS_TRUE_AND_LT_AUTH(lock->get_sm()->states[next].can_rdlock, auth) || !lock->is_rdlocked()) && | |
845 | (IS_TRUE_AND_LT_AUTH(lock->get_sm()->states[next].can_wrlock, auth) || !lock->is_wrlocked()) && | |
846 | (IS_TRUE_AND_LT_AUTH(lock->get_sm()->states[next].can_xlock, auth) || !lock->is_xlocked()) && | |
847 | (IS_TRUE_AND_LT_AUTH(lock->get_sm()->states[next].can_lease, auth) || !lock->is_leased()) && | |
848 | !(lock->get_parent()->is_auth() && lock->is_flushing()) && // i.e. wait for scatter_writebehind! | |
849 | (!caps || ((~lock->gcaps_allowed(CAP_ANY, next) & other_issued) == 0 && | |
850 | (~lock->gcaps_allowed(CAP_LONER, next) & loner_issued) == 0 && | |
851 | (~lock->gcaps_allowed(CAP_XLOCKER, next) & xlocker_issued) == 0)) && | |
852 | lock->get_state() != LOCK_SYNC_MIX2 && // these states need an explicit trigger from the auth mds | |
853 | lock->get_state() != LOCK_MIX_SYNC2 | |
854 | ) { | |
855 | dout(7) << "eval_gather finished gather on " << *lock | |
856 | << " on " << *lock->get_parent() << dendl; | |
857 | ||
858 | if (lock->get_sm() == &sm_filelock) { | |
11fdf7f2 | 859 | ceph_assert(in); |
7c673cae FG |
860 | if (in->state_test(CInode::STATE_RECOVERING)) { |
861 | dout(7) << "eval_gather finished gather, but still recovering" << dendl; | |
862 | return; | |
863 | } else if (in->state_test(CInode::STATE_NEEDSRECOVER)) { | |
864 | dout(7) << "eval_gather finished gather, but need to recover" << dendl; | |
865 | mds->mdcache->queue_file_recover(in); | |
866 | mds->mdcache->do_file_recover(); | |
867 | return; | |
868 | } | |
869 | } | |
870 | ||
871 | if (!lock->get_parent()->is_auth()) { | |
872 | // replica: tell auth | |
873 | mds_rank_t auth = lock->get_parent()->authority().first; | |
874 | ||
875 | if (lock->get_parent()->is_rejoining() && | |
876 | mds->mdsmap->get_state(auth) == MDSMap::STATE_REJOIN) { | |
877 | dout(7) << "eval_gather finished gather, but still rejoining " | |
878 | << *lock->get_parent() << dendl; | |
879 | return; | |
880 | } | |
881 | ||
882 | if (!mds->is_cluster_degraded() || | |
883 | mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) { | |
884 | switch (lock->get_state()) { | |
885 | case LOCK_SYNC_LOCK: | |
11fdf7f2 | 886 | mds->send_message_mds(MLock::create(lock, LOCK_AC_LOCKACK, mds->get_nodeid()), auth); |
7c673cae FG |
887 | break; |
888 | ||
889 | case LOCK_MIX_SYNC: | |
890 | { | |
11fdf7f2 | 891 | auto reply = MLock::create(lock, LOCK_AC_SYNCACK, mds->get_nodeid()); |
7c673cae FG |
892 | lock->encode_locked_state(reply->get_data()); |
893 | mds->send_message_mds(reply, auth); | |
894 | next = LOCK_MIX_SYNC2; | |
895 | (static_cast<ScatterLock *>(lock))->start_flush(); | |
896 | } | |
897 | break; | |
898 | ||
899 | case LOCK_MIX_SYNC2: | |
900 | (static_cast<ScatterLock *>(lock))->finish_flush(); | |
901 | (static_cast<ScatterLock *>(lock))->clear_flushed(); | |
902 | ||
903 | case LOCK_SYNC_MIX2: | |
904 | // do nothing, we already acked | |
905 | break; | |
906 | ||
907 | case LOCK_SYNC_MIX: | |
908 | { | |
11fdf7f2 | 909 | auto reply = MLock::create(lock, LOCK_AC_MIXACK, mds->get_nodeid()); |
7c673cae FG |
910 | mds->send_message_mds(reply, auth); |
911 | next = LOCK_SYNC_MIX2; | |
912 | } | |
913 | break; | |
914 | ||
915 | case LOCK_MIX_LOCK: | |
916 | { | |
917 | bufferlist data; | |
918 | lock->encode_locked_state(data); | |
11fdf7f2 | 919 | mds->send_message_mds(MLock::create(lock, LOCK_AC_LOCKACK, mds->get_nodeid(), data), auth); |
7c673cae FG |
920 | (static_cast<ScatterLock *>(lock))->start_flush(); |
921 | // we'll get an AC_LOCKFLUSHED to complete | |
922 | } | |
923 | break; | |
924 | ||
925 | default: | |
926 | ceph_abort(); | |
927 | } | |
928 | } | |
929 | } else { | |
930 | // auth | |
931 | ||
932 | // once the first (local) stage of mix->lock gather complete we can | |
933 | // gather from replicas | |
934 | if (lock->get_state() == LOCK_MIX_LOCK && | |
935 | lock->get_parent()->is_replicated()) { | |
936 | dout(10) << " finished (local) gather for mix->lock, now gathering from replicas" << dendl; | |
937 | send_lock_message(lock, LOCK_AC_LOCK); | |
938 | lock->init_gather(); | |
939 | lock->set_state(LOCK_MIX_LOCK2); | |
940 | return; | |
941 | } | |
942 | ||
943 | if (lock->is_dirty() && !lock->is_flushed()) { | |
944 | scatter_writebehind(static_cast<ScatterLock *>(lock)); | |
945 | mds->mdlog->flush(); | |
946 | return; | |
947 | } | |
948 | lock->clear_flushed(); | |
949 | ||
950 | switch (lock->get_state()) { | |
951 | // to mixed | |
952 | case LOCK_TSYN_MIX: | |
953 | case LOCK_SYNC_MIX: | |
954 | case LOCK_EXCL_MIX: | |
b32b8144 | 955 | case LOCK_XSYN_MIX: |
7c673cae FG |
956 | in->start_scatter(static_cast<ScatterLock *>(lock)); |
957 | if (lock->get_parent()->is_replicated()) { | |
958 | bufferlist softdata; | |
959 | lock->encode_locked_state(softdata); | |
960 | send_lock_message(lock, LOCK_AC_MIX, softdata); | |
961 | } | |
962 | (static_cast<ScatterLock *>(lock))->clear_scatter_wanted(); | |
963 | break; | |
964 | ||
965 | case LOCK_XLOCK: | |
966 | case LOCK_XLOCKDONE: | |
967 | if (next != LOCK_SYNC) | |
968 | break; | |
969 | // fall-thru | |
970 | ||
971 | // to sync | |
972 | case LOCK_EXCL_SYNC: | |
973 | case LOCK_LOCK_SYNC: | |
974 | case LOCK_MIX_SYNC: | |
975 | case LOCK_XSYN_SYNC: | |
976 | if (lock->get_parent()->is_replicated()) { | |
977 | bufferlist softdata; | |
978 | lock->encode_locked_state(softdata); | |
979 | send_lock_message(lock, LOCK_AC_SYNC, softdata); | |
980 | } | |
981 | break; | |
982 | } | |
983 | ||
984 | } | |
985 | ||
986 | lock->set_state(next); | |
987 | ||
988 | if (lock->get_parent()->is_auth() && | |
989 | lock->is_stable()) | |
990 | lock->get_parent()->auth_unpin(lock); | |
991 | ||
992 | // drop loner before doing waiters | |
993 | if (caps && | |
994 | in->is_head() && | |
995 | in->is_auth() && | |
996 | in->get_wanted_loner() != in->get_loner()) { | |
997 | dout(10) << " trying to drop loner" << dendl; | |
998 | if (in->try_drop_loner()) { | |
999 | dout(10) << " dropped loner" << dendl; | |
1000 | need_issue = true; | |
1001 | } | |
1002 | } | |
1003 | ||
1004 | if (pfinishers) | |
1005 | lock->take_waiting(SimpleLock::WAIT_STABLE|SimpleLock::WAIT_WR|SimpleLock::WAIT_RD|SimpleLock::WAIT_XLOCK, | |
1006 | *pfinishers); | |
1007 | else | |
1008 | lock->finish_waiters(SimpleLock::WAIT_STABLE|SimpleLock::WAIT_WR|SimpleLock::WAIT_RD|SimpleLock::WAIT_XLOCK); | |
1009 | ||
1010 | if (caps && in->is_head()) | |
1011 | need_issue = true; | |
1012 | ||
1013 | if (lock->get_parent()->is_auth() && | |
1014 | lock->is_stable()) | |
1015 | try_eval(lock, &need_issue); | |
1016 | } | |
1017 | ||
1018 | if (need_issue) { | |
1019 | if (pneed_issue) | |
1020 | *pneed_issue = true; | |
1021 | else if (in->is_head()) | |
1022 | issue_caps(in); | |
1023 | } | |
1024 | ||
1025 | } | |
1026 | ||
1027 | bool Locker::eval(CInode *in, int mask, bool caps_imported) | |
1028 | { | |
1029 | bool need_issue = caps_imported; | |
11fdf7f2 | 1030 | MDSContext::vec finishers; |
7c673cae FG |
1031 | |
1032 | dout(10) << "eval " << mask << " " << *in << dendl; | |
1033 | ||
1034 | // choose loner? | |
1035 | if (in->is_auth() && in->is_head()) { | |
b32b8144 FG |
1036 | client_t orig_loner = in->get_loner(); |
1037 | if (in->choose_ideal_loner()) { | |
1038 | dout(10) << "eval set loner: client." << orig_loner << " -> client." << in->get_loner() << dendl; | |
1039 | need_issue = true; | |
1040 | mask = -1; | |
1041 | } else if (in->get_wanted_loner() != in->get_loner()) { | |
1042 | dout(10) << "eval want loner: client." << in->get_wanted_loner() << " but failed to set it" << dendl; | |
1043 | mask = -1; | |
1044 | } | |
7c673cae FG |
1045 | } |
1046 | ||
1047 | retry: | |
1048 | if (mask & CEPH_LOCK_IFILE) | |
1049 | eval_any(&in->filelock, &need_issue, &finishers, caps_imported); | |
1050 | if (mask & CEPH_LOCK_IAUTH) | |
1051 | eval_any(&in->authlock, &need_issue, &finishers, caps_imported); | |
1052 | if (mask & CEPH_LOCK_ILINK) | |
1053 | eval_any(&in->linklock, &need_issue, &finishers, caps_imported); | |
1054 | if (mask & CEPH_LOCK_IXATTR) | |
1055 | eval_any(&in->xattrlock, &need_issue, &finishers, caps_imported); | |
1056 | if (mask & CEPH_LOCK_INEST) | |
1057 | eval_any(&in->nestlock, &need_issue, &finishers, caps_imported); | |
1058 | if (mask & CEPH_LOCK_IFLOCK) | |
1059 | eval_any(&in->flocklock, &need_issue, &finishers, caps_imported); | |
1060 | if (mask & CEPH_LOCK_IPOLICY) | |
1061 | eval_any(&in->policylock, &need_issue, &finishers, caps_imported); | |
1062 | ||
1063 | // drop loner? | |
1064 | if (in->is_auth() && in->is_head() && in->get_wanted_loner() != in->get_loner()) { | |
7c673cae | 1065 | if (in->try_drop_loner()) { |
7c673cae | 1066 | need_issue = true; |
7c673cae | 1067 | if (in->get_wanted_loner() >= 0) { |
b32b8144 FG |
1068 | dout(10) << "eval end set loner to client." << in->get_loner() << dendl; |
1069 | bool ok = in->try_set_loner(); | |
11fdf7f2 | 1070 | ceph_assert(ok); |
b32b8144 FG |
1071 | mask = -1; |
1072 | goto retry; | |
7c673cae FG |
1073 | } |
1074 | } | |
1075 | } | |
1076 | ||
1077 | finish_contexts(g_ceph_context, finishers); | |
1078 | ||
1079 | if (need_issue && in->is_head()) | |
1080 | issue_caps(in); | |
1081 | ||
1082 | dout(10) << "eval done" << dendl; | |
1083 | return need_issue; | |
1084 | } | |
1085 | ||
1086 | class C_Locker_Eval : public LockerContext { | |
1087 | MDSCacheObject *p; | |
1088 | int mask; | |
1089 | public: | |
1090 | C_Locker_Eval(Locker *l, MDSCacheObject *pp, int m) : LockerContext(l), p(pp), mask(m) { | |
1091 | // We are used as an MDSCacheObject waiter, so should | |
1092 | // only be invoked by someone already holding the big lock. | |
11fdf7f2 | 1093 | ceph_assert(locker->mds->mds_lock.is_locked_by_me()); |
7c673cae FG |
1094 | p->get(MDSCacheObject::PIN_PTRWAITER); |
1095 | } | |
1096 | void finish(int r) override { | |
1097 | locker->try_eval(p, mask); | |
1098 | p->put(MDSCacheObject::PIN_PTRWAITER); | |
1099 | } | |
1100 | }; | |
1101 | ||
1102 | void Locker::try_eval(MDSCacheObject *p, int mask) | |
1103 | { | |
1104 | // unstable and ambiguous auth? | |
1105 | if (p->is_ambiguous_auth()) { | |
1106 | dout(7) << "try_eval ambiguous auth, waiting on " << *p << dendl; | |
1107 | p->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, new C_Locker_Eval(this, p, mask)); | |
1108 | return; | |
1109 | } | |
1110 | ||
1111 | if (p->is_auth() && p->is_frozen()) { | |
1112 | dout(7) << "try_eval frozen, waiting on " << *p << dendl; | |
1113 | p->add_waiter(MDSCacheObject::WAIT_UNFREEZE, new C_Locker_Eval(this, p, mask)); | |
1114 | return; | |
1115 | } | |
1116 | ||
1117 | if (mask & CEPH_LOCK_DN) { | |
11fdf7f2 | 1118 | ceph_assert(mask == CEPH_LOCK_DN); |
7c673cae FG |
1119 | bool need_issue = false; // ignore this, no caps on dentries |
1120 | CDentry *dn = static_cast<CDentry *>(p); | |
1121 | eval_any(&dn->lock, &need_issue); | |
1122 | } else { | |
1123 | CInode *in = static_cast<CInode *>(p); | |
1124 | eval(in, mask); | |
1125 | } | |
1126 | } | |
1127 | ||
1128 | void Locker::try_eval(SimpleLock *lock, bool *pneed_issue) | |
1129 | { | |
1130 | MDSCacheObject *p = lock->get_parent(); | |
1131 | ||
1132 | // unstable and ambiguous auth? | |
1133 | if (p->is_ambiguous_auth()) { | |
1134 | dout(7) << "try_eval " << *lock << " ambiguousauth, waiting on " << *p << dendl; | |
1135 | p->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, new C_Locker_Eval(this, p, lock->get_type())); | |
1136 | return; | |
1137 | } | |
1138 | ||
1139 | if (!p->is_auth()) { | |
1140 | dout(7) << "try_eval " << *lock << " not auth for " << *p << dendl; | |
1141 | return; | |
1142 | } | |
1143 | ||
1144 | if (p->is_frozen()) { | |
1145 | dout(7) << "try_eval " << *lock << " frozen, waiting on " << *p << dendl; | |
1146 | p->add_waiter(MDSCacheObject::WAIT_UNFREEZE, new C_Locker_Eval(this, p, lock->get_type())); | |
1147 | return; | |
1148 | } | |
1149 | ||
1150 | /* | |
1151 | * We could have a situation like: | |
1152 | * | |
1153 | * - mds A authpins item on mds B | |
1154 | * - mds B starts to freeze tree containing item | |
1155 | * - mds A tries wrlock_start on A, sends REQSCATTER to B | |
1156 | * - mds B lock is unstable, sets scatter_wanted | |
1157 | * - mds B lock stabilizes, calls try_eval. | |
1158 | * | |
1159 | * We can defer while freezing without causing a deadlock. Honor | |
1160 | * scatter_wanted flag here. This will never get deferred by the | |
1161 | * checks above due to the auth_pin held by the master. | |
1162 | */ | |
1163 | if (lock->is_scatterlock()) { | |
1164 | ScatterLock *slock = static_cast<ScatterLock *>(lock); | |
1165 | if (slock->get_scatter_wanted() && | |
1166 | slock->get_state() != LOCK_MIX) { | |
1167 | scatter_mix(slock, pneed_issue); | |
1168 | if (!lock->is_stable()) | |
1169 | return; | |
1170 | } else if (slock->get_unscatter_wanted() && | |
1171 | slock->get_state() != LOCK_LOCK) { | |
1172 | simple_lock(slock, pneed_issue); | |
1173 | if (!lock->is_stable()) { | |
1174 | return; | |
1175 | } | |
1176 | } | |
1177 | } | |
1178 | ||
11fdf7f2 TL |
1179 | if (lock->get_type() != CEPH_LOCK_DN && |
1180 | lock->get_type() != CEPH_LOCK_ISNAP && | |
1181 | p->is_freezing()) { | |
7c673cae FG |
1182 | dout(7) << "try_eval " << *lock << " freezing, waiting on " << *p << dendl; |
1183 | p->add_waiter(MDSCacheObject::WAIT_UNFREEZE, new C_Locker_Eval(this, p, lock->get_type())); | |
1184 | return; | |
1185 | } | |
1186 | ||
1187 | eval(lock, pneed_issue); | |
1188 | } | |
1189 | ||
1190 | void Locker::eval_cap_gather(CInode *in, set<CInode*> *issue_set) | |
1191 | { | |
1192 | bool need_issue = false; | |
11fdf7f2 | 1193 | MDSContext::vec finishers; |
7c673cae FG |
1194 | |
1195 | // kick locks now | |
1196 | if (!in->filelock.is_stable()) | |
1197 | eval_gather(&in->filelock, false, &need_issue, &finishers); | |
1198 | if (!in->authlock.is_stable()) | |
1199 | eval_gather(&in->authlock, false, &need_issue, &finishers); | |
1200 | if (!in->linklock.is_stable()) | |
1201 | eval_gather(&in->linklock, false, &need_issue, &finishers); | |
1202 | if (!in->xattrlock.is_stable()) | |
1203 | eval_gather(&in->xattrlock, false, &need_issue, &finishers); | |
1204 | ||
1205 | if (need_issue && in->is_head()) { | |
1206 | if (issue_set) | |
1207 | issue_set->insert(in); | |
1208 | else | |
1209 | issue_caps(in); | |
1210 | } | |
1211 | ||
1212 | finish_contexts(g_ceph_context, finishers); | |
1213 | } | |
1214 | ||
1215 | void Locker::eval_scatter_gathers(CInode *in) | |
1216 | { | |
1217 | bool need_issue = false; | |
11fdf7f2 | 1218 | MDSContext::vec finishers; |
7c673cae FG |
1219 | |
1220 | dout(10) << "eval_scatter_gathers " << *in << dendl; | |
1221 | ||
1222 | // kick locks now | |
1223 | if (!in->filelock.is_stable()) | |
1224 | eval_gather(&in->filelock, false, &need_issue, &finishers); | |
1225 | if (!in->nestlock.is_stable()) | |
1226 | eval_gather(&in->nestlock, false, &need_issue, &finishers); | |
1227 | if (!in->dirfragtreelock.is_stable()) | |
1228 | eval_gather(&in->dirfragtreelock, false, &need_issue, &finishers); | |
1229 | ||
1230 | if (need_issue && in->is_head()) | |
1231 | issue_caps(in); | |
1232 | ||
1233 | finish_contexts(g_ceph_context, finishers); | |
1234 | } | |
1235 | ||
1236 | void Locker::eval(SimpleLock *lock, bool *need_issue) | |
1237 | { | |
1238 | switch (lock->get_type()) { | |
1239 | case CEPH_LOCK_IFILE: | |
1240 | return file_eval(static_cast<ScatterLock*>(lock), need_issue); | |
1241 | case CEPH_LOCK_IDFT: | |
1242 | case CEPH_LOCK_INEST: | |
1243 | return scatter_eval(static_cast<ScatterLock*>(lock), need_issue); | |
1244 | default: | |
1245 | return simple_eval(lock, need_issue); | |
1246 | } | |
1247 | } | |
1248 | ||
1249 | ||
1250 | // ------------------ | |
1251 | // rdlock | |
1252 | ||
1253 | bool Locker::_rdlock_kick(SimpleLock *lock, bool as_anon) | |
1254 | { | |
1255 | // kick the lock | |
1256 | if (lock->is_stable()) { | |
1257 | if (lock->get_parent()->is_auth()) { | |
1258 | if (lock->get_sm() == &sm_scatterlock) { | |
1259 | // not until tempsync is fully implemented | |
1260 | //if (lock->get_parent()->is_replicated()) | |
1261 | //scatter_tempsync((ScatterLock*)lock); | |
1262 | //else | |
1263 | simple_sync(lock); | |
1264 | } else if (lock->get_sm() == &sm_filelock) { | |
1265 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
1266 | if (lock->get_state() == LOCK_EXCL && | |
1267 | in->get_target_loner() >= 0 && | |
1268 | !in->is_dir() && !as_anon) // as_anon => caller wants SYNC, not XSYN | |
1269 | file_xsyn(lock); | |
1270 | else | |
1271 | simple_sync(lock); | |
1272 | } else | |
1273 | simple_sync(lock); | |
1274 | return true; | |
1275 | } else { | |
1276 | // request rdlock state change from auth | |
1277 | mds_rank_t auth = lock->get_parent()->authority().first; | |
1278 | if (!mds->is_cluster_degraded() || | |
1279 | mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) { | |
1280 | dout(10) << "requesting rdlock from auth on " | |
1281 | << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 | 1282 | mds->send_message_mds(MLock::create(lock, LOCK_AC_REQRDLOCK, mds->get_nodeid()), auth); |
7c673cae FG |
1283 | } |
1284 | return false; | |
1285 | } | |
1286 | } | |
1287 | if (lock->get_type() == CEPH_LOCK_IFILE) { | |
1288 | CInode *in = static_cast<CInode *>(lock->get_parent()); | |
1289 | if (in->state_test(CInode::STATE_RECOVERING)) { | |
1290 | mds->mdcache->recovery_queue.prioritize(in); | |
1291 | } | |
1292 | } | |
1293 | ||
1294 | return false; | |
1295 | } | |
1296 | ||
11fdf7f2 | 1297 | bool Locker::rdlock_try(SimpleLock *lock, client_t client, MDSContext *con) |
7c673cae FG |
1298 | { |
1299 | dout(7) << "rdlock_try on " << *lock << " on " << *lock->get_parent() << dendl; | |
1300 | ||
1301 | // can read? grab ref. | |
1302 | if (lock->can_rdlock(client)) | |
1303 | return true; | |
1304 | ||
1305 | _rdlock_kick(lock, false); | |
1306 | ||
1307 | if (lock->can_rdlock(client)) | |
1308 | return true; | |
1309 | ||
1310 | // wait! | |
1311 | if (con) { | |
1312 | dout(7) << "rdlock_try waiting on " << *lock << " on " << *lock->get_parent() << dendl; | |
1313 | lock->add_waiter(SimpleLock::WAIT_STABLE|SimpleLock::WAIT_RD, con); | |
1314 | } | |
1315 | return false; | |
1316 | } | |
1317 | ||
1318 | bool Locker::rdlock_start(SimpleLock *lock, MDRequestRef& mut, bool as_anon) | |
1319 | { | |
1320 | dout(7) << "rdlock_start on " << *lock << " on " << *lock->get_parent() << dendl; | |
1321 | ||
1322 | // client may be allowed to rdlock the same item it has xlocked. | |
1323 | // UNLESS someone passes in as_anon, or we're reading snapped version here. | |
1324 | if (mut->snapid != CEPH_NOSNAP) | |
1325 | as_anon = true; | |
1326 | client_t client = as_anon ? -1 : mut->get_client(); | |
1327 | ||
1328 | CInode *in = 0; | |
1329 | if (lock->get_type() != CEPH_LOCK_DN) | |
1330 | in = static_cast<CInode *>(lock->get_parent()); | |
1331 | ||
1332 | /* | |
1333 | if (!lock->get_parent()->is_auth() && | |
1334 | lock->fw_rdlock_to_auth()) { | |
1335 | mdcache->request_forward(mut, lock->get_parent()->authority().first); | |
1336 | return false; | |
1337 | } | |
1338 | */ | |
1339 | ||
1340 | while (1) { | |
1341 | // can read? grab ref. | |
1342 | if (lock->can_rdlock(client)) { | |
1343 | lock->get_rdlock(); | |
11fdf7f2 | 1344 | mut->locks.emplace_hint(mut->locks.end(), lock, MutationImpl::LockOp::RDLOCK); |
7c673cae FG |
1345 | return true; |
1346 | } | |
1347 | ||
1348 | // hmm, wait a second. | |
1349 | if (in && !in->is_head() && in->is_auth() && | |
1350 | lock->get_state() == LOCK_SNAP_SYNC) { | |
1351 | // okay, we actually need to kick the head's lock to get ourselves synced up. | |
1352 | CInode *head = mdcache->get_inode(in->ino()); | |
11fdf7f2 | 1353 | ceph_assert(head); |
c07f9fc5 FG |
1354 | SimpleLock *hlock = head->get_lock(CEPH_LOCK_IFILE); |
1355 | if (hlock->get_state() == LOCK_SYNC) | |
1356 | hlock = head->get_lock(lock->get_type()); | |
1357 | ||
7c673cae FG |
1358 | if (hlock->get_state() != LOCK_SYNC) { |
1359 | dout(10) << "rdlock_start trying head inode " << *head << dendl; | |
c07f9fc5 | 1360 | if (!rdlock_start(hlock, mut, true)) // ** as_anon, no rdlock on EXCL ** |
7c673cae FG |
1361 | return false; |
1362 | // oh, check our lock again then | |
1363 | } | |
1364 | } | |
1365 | ||
1366 | if (!_rdlock_kick(lock, as_anon)) | |
1367 | break; | |
1368 | } | |
1369 | ||
1370 | // wait! | |
1371 | int wait_on; | |
1372 | if (lock->get_parent()->is_auth() && lock->is_stable()) | |
1373 | wait_on = SimpleLock::WAIT_RD; | |
1374 | else | |
1375 | wait_on = SimpleLock::WAIT_STABLE; // REQRDLOCK is ignored if lock is unstable, so we need to retry. | |
1376 | dout(7) << "rdlock_start waiting on " << *lock << " on " << *lock->get_parent() << dendl; | |
1377 | lock->add_waiter(wait_on, new C_MDS_RetryRequest(mdcache, mut)); | |
1378 | nudge_log(lock); | |
1379 | return false; | |
1380 | } | |
1381 | ||
1382 | void Locker::nudge_log(SimpleLock *lock) | |
1383 | { | |
1384 | dout(10) << "nudge_log " << *lock << " on " << *lock->get_parent() << dendl; | |
1385 | if (lock->get_parent()->is_auth() && lock->is_unstable_and_locked()) // as with xlockdone, or cap flush | |
1386 | mds->mdlog->flush(); | |
1387 | } | |
1388 | ||
11fdf7f2 | 1389 | void Locker::rdlock_finish(const MutationImpl::lock_iterator& it, MutationImpl *mut, bool *pneed_issue) |
7c673cae | 1390 | { |
11fdf7f2 TL |
1391 | ceph_assert(it->is_rdlock()); |
1392 | SimpleLock *lock = it->lock; | |
7c673cae FG |
1393 | // drop ref |
1394 | lock->put_rdlock(); | |
11fdf7f2 TL |
1395 | if (mut) |
1396 | mut->locks.erase(it); | |
7c673cae FG |
1397 | |
1398 | dout(7) << "rdlock_finish on " << *lock << " on " << *lock->get_parent() << dendl; | |
1399 | ||
1400 | // last one? | |
1401 | if (!lock->is_rdlocked()) { | |
1402 | if (!lock->is_stable()) | |
1403 | eval_gather(lock, false, pneed_issue); | |
1404 | else if (lock->get_parent()->is_auth()) | |
1405 | try_eval(lock, pneed_issue); | |
1406 | } | |
1407 | } | |
1408 | ||
1409 | ||
11fdf7f2 | 1410 | bool Locker::can_rdlock_set(MutationImpl::LockOpVec& lov) |
7c673cae | 1411 | { |
11fdf7f2 TL |
1412 | dout(10) << "can_rdlock_set " << dendl; |
1413 | for (const auto& p : lov) { | |
1414 | ceph_assert(p.is_rdlock()); | |
1415 | if (!p.lock->can_rdlock(-1)) { | |
1416 | dout(10) << "can_rdlock_set can't rdlock " << *p << " on " << *p.lock->get_parent() << dendl; | |
7c673cae FG |
1417 | return false; |
1418 | } | |
11fdf7f2 | 1419 | } |
7c673cae FG |
1420 | return true; |
1421 | } | |
1422 | ||
7c673cae | 1423 | |
11fdf7f2 | 1424 | void Locker::rdlock_take_set(MutationImpl::LockOpVec& lov, MutationRef& mut) |
7c673cae | 1425 | { |
11fdf7f2 TL |
1426 | dout(10) << "rdlock_take_set " << dendl; |
1427 | for (const auto& p : lov) { | |
1428 | ceph_assert(p.is_rdlock()); | |
1429 | p.lock->get_rdlock(); | |
1430 | mut->locks.emplace(p.lock, MutationImpl::LockOp::RDLOCK); | |
7c673cae FG |
1431 | } |
1432 | } | |
1433 | ||
1434 | // ------------------ | |
1435 | // wrlock | |
1436 | ||
1437 | void Locker::wrlock_force(SimpleLock *lock, MutationRef& mut) | |
1438 | { | |
1439 | if (lock->get_type() == CEPH_LOCK_IVERSION || | |
1440 | lock->get_type() == CEPH_LOCK_DVERSION) | |
1441 | return local_wrlock_grab(static_cast<LocalLock*>(lock), mut); | |
1442 | ||
1443 | dout(7) << "wrlock_force on " << *lock | |
1444 | << " on " << *lock->get_parent() << dendl; | |
1445 | lock->get_wrlock(true); | |
11fdf7f2 | 1446 | mut->locks.emplace(lock, MutationImpl::LockOp::WRLOCK); |
7c673cae FG |
1447 | } |
1448 | ||
1449 | bool Locker::wrlock_start(SimpleLock *lock, MDRequestRef& mut, bool nowait) | |
1450 | { | |
1451 | if (lock->get_type() == CEPH_LOCK_IVERSION || | |
1452 | lock->get_type() == CEPH_LOCK_DVERSION) | |
1453 | return local_wrlock_start(static_cast<LocalLock*>(lock), mut); | |
1454 | ||
1455 | dout(10) << "wrlock_start " << *lock << " on " << *lock->get_parent() << dendl; | |
1456 | ||
1457 | CInode *in = static_cast<CInode *>(lock->get_parent()); | |
1458 | client_t client = mut->get_client(); | |
1459 | bool want_scatter = !nowait && lock->get_parent()->is_auth() && | |
1460 | (in->has_subtree_or_exporting_dirfrag() || | |
1461 | static_cast<ScatterLock*>(lock)->get_scatter_wanted()); | |
1462 | ||
1463 | while (1) { | |
1464 | // wrlock? | |
1465 | if (lock->can_wrlock(client) && | |
1466 | (!want_scatter || lock->get_state() == LOCK_MIX)) { | |
1467 | lock->get_wrlock(); | |
11fdf7f2 TL |
1468 | auto it = mut->locks.emplace_hint(mut->locks.end(), lock, MutationImpl::LockOp::WRLOCK); |
1469 | it->flags |= MutationImpl::LockOp::WRLOCK; // may already remote_wrlocked | |
7c673cae FG |
1470 | return true; |
1471 | } | |
1472 | ||
1473 | if (lock->get_type() == CEPH_LOCK_IFILE && | |
1474 | in->state_test(CInode::STATE_RECOVERING)) { | |
1475 | mds->mdcache->recovery_queue.prioritize(in); | |
1476 | } | |
1477 | ||
1478 | if (!lock->is_stable()) | |
1479 | break; | |
1480 | ||
1481 | if (in->is_auth()) { | |
1482 | // don't do nested lock state change if we have dirty scatterdata and | |
1483 | // may scatter_writebehind or start_scatter, because nowait==true implies | |
1484 | // that the caller already has a log entry open! | |
1485 | if (nowait && lock->is_dirty()) | |
1486 | return false; | |
1487 | ||
1488 | if (want_scatter) | |
1489 | scatter_mix(static_cast<ScatterLock*>(lock)); | |
1490 | else | |
1491 | simple_lock(lock); | |
1492 | ||
1493 | if (nowait && !lock->can_wrlock(client)) | |
1494 | return false; | |
1495 | ||
1496 | } else { | |
1497 | // replica. | |
1498 | // auth should be auth_pinned (see acquire_locks wrlock weird mustpin case). | |
1499 | mds_rank_t auth = lock->get_parent()->authority().first; | |
1500 | if (!mds->is_cluster_degraded() || | |
1501 | mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) { | |
1502 | dout(10) << "requesting scatter from auth on " | |
1503 | << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 | 1504 | mds->send_message_mds(MLock::create(lock, LOCK_AC_REQSCATTER, mds->get_nodeid()), auth); |
7c673cae FG |
1505 | } |
1506 | break; | |
1507 | } | |
1508 | } | |
1509 | ||
1510 | if (!nowait) { | |
1511 | dout(7) << "wrlock_start waiting on " << *lock << " on " << *lock->get_parent() << dendl; | |
1512 | lock->add_waiter(SimpleLock::WAIT_STABLE, new C_MDS_RetryRequest(mdcache, mut)); | |
1513 | nudge_log(lock); | |
1514 | } | |
1515 | ||
1516 | return false; | |
1517 | } | |
1518 | ||
11fdf7f2 | 1519 | void Locker::wrlock_finish(const MutationImpl::lock_iterator& it, MutationImpl *mut, bool *pneed_issue) |
7c673cae | 1520 | { |
11fdf7f2 TL |
1521 | ceph_assert(it->is_wrlock()); |
1522 | SimpleLock* lock = it->lock; | |
1523 | ||
7c673cae FG |
1524 | if (lock->get_type() == CEPH_LOCK_IVERSION || |
1525 | lock->get_type() == CEPH_LOCK_DVERSION) | |
11fdf7f2 | 1526 | return local_wrlock_finish(it, mut); |
7c673cae FG |
1527 | |
1528 | dout(7) << "wrlock_finish on " << *lock << " on " << *lock->get_parent() << dendl; | |
1529 | lock->put_wrlock(); | |
11fdf7f2 TL |
1530 | |
1531 | if (it->is_remote_wrlock()) | |
1532 | it->clear_wrlock(); | |
1533 | else | |
1534 | mut->locks.erase(it); | |
7c673cae FG |
1535 | |
1536 | if (!lock->is_wrlocked()) { | |
1537 | if (!lock->is_stable()) | |
1538 | eval_gather(lock, false, pneed_issue); | |
1539 | else if (lock->get_parent()->is_auth()) | |
1540 | try_eval(lock, pneed_issue); | |
1541 | } | |
1542 | } | |
1543 | ||
1544 | ||
1545 | // remote wrlock | |
1546 | ||
1547 | void Locker::remote_wrlock_start(SimpleLock *lock, mds_rank_t target, MDRequestRef& mut) | |
1548 | { | |
1549 | dout(7) << "remote_wrlock_start mds." << target << " on " << *lock << " on " << *lock->get_parent() << dendl; | |
1550 | ||
1551 | // wait for active target | |
1552 | if (mds->is_cluster_degraded() && | |
1553 | !mds->mdsmap->is_clientreplay_or_active_or_stopping(target)) { | |
1554 | dout(7) << " mds." << target << " is not active" << dendl; | |
1555 | if (mut->more()->waiting_on_slave.empty()) | |
1556 | mds->wait_for_active_peer(target, new C_MDS_RetryRequest(mdcache, mut)); | |
1557 | return; | |
1558 | } | |
1559 | ||
1560 | // send lock request | |
1561 | mut->start_locking(lock, target); | |
1562 | mut->more()->slaves.insert(target); | |
11fdf7f2 | 1563 | auto r = MMDSSlaveRequest::create(mut->reqid, mut->attempt, MMDSSlaveRequest::OP_WRLOCK); |
7c673cae FG |
1564 | r->set_lock_type(lock->get_type()); |
1565 | lock->get_parent()->set_object_info(r->get_object_info()); | |
1566 | mds->send_message_mds(r, target); | |
1567 | ||
11fdf7f2 | 1568 | ceph_assert(mut->more()->waiting_on_slave.count(target) == 0); |
7c673cae FG |
1569 | mut->more()->waiting_on_slave.insert(target); |
1570 | } | |
1571 | ||
11fdf7f2 | 1572 | void Locker::remote_wrlock_finish(const MutationImpl::lock_iterator& it, MutationImpl *mut) |
7c673cae | 1573 | { |
11fdf7f2 TL |
1574 | ceph_assert(it->is_remote_wrlock()); |
1575 | SimpleLock *lock = it->lock; | |
1576 | mds_rank_t target = it->wrlock_target; | |
1577 | ||
1578 | if (it->is_wrlock()) | |
1579 | it->clear_remote_wrlock(); | |
1580 | else | |
1581 | mut->locks.erase(it); | |
7c673cae FG |
1582 | |
1583 | dout(7) << "remote_wrlock_finish releasing remote wrlock on mds." << target | |
1584 | << " " << *lock->get_parent() << dendl; | |
1585 | if (!mds->is_cluster_degraded() || | |
1586 | mds->mdsmap->get_state(target) >= MDSMap::STATE_REJOIN) { | |
11fdf7f2 | 1587 | auto slavereq = MMDSSlaveRequest::create(mut->reqid, mut->attempt, MMDSSlaveRequest::OP_UNWRLOCK); |
7c673cae FG |
1588 | slavereq->set_lock_type(lock->get_type()); |
1589 | lock->get_parent()->set_object_info(slavereq->get_object_info()); | |
1590 | mds->send_message_mds(slavereq, target); | |
1591 | } | |
1592 | } | |
1593 | ||
1594 | ||
1595 | // ------------------ | |
1596 | // xlock | |
1597 | ||
1598 | bool Locker::xlock_start(SimpleLock *lock, MDRequestRef& mut) | |
1599 | { | |
1600 | if (lock->get_type() == CEPH_LOCK_IVERSION || | |
1601 | lock->get_type() == CEPH_LOCK_DVERSION) | |
1602 | return local_xlock_start(static_cast<LocalLock*>(lock), mut); | |
1603 | ||
1604 | dout(7) << "xlock_start on " << *lock << " on " << *lock->get_parent() << dendl; | |
1605 | client_t client = mut->get_client(); | |
1606 | ||
a8e16298 TL |
1607 | CInode *in = nullptr; |
1608 | if (lock->get_cap_shift()) | |
1609 | in = static_cast<CInode *>(lock->get_parent()); | |
1610 | ||
7c673cae FG |
1611 | // auth? |
1612 | if (lock->get_parent()->is_auth()) { | |
1613 | // auth | |
1614 | while (1) { | |
a8e16298 TL |
1615 | if (lock->can_xlock(client) && |
1616 | !(lock->get_state() == LOCK_LOCK_XLOCK && // client is not xlocker or | |
1617 | in && in->issued_caps_need_gather(lock))) { // xlocker does not hold shared cap | |
7c673cae FG |
1618 | lock->set_state(LOCK_XLOCK); |
1619 | lock->get_xlock(mut, client); | |
11fdf7f2 | 1620 | mut->locks.emplace_hint(mut->locks.end(), lock, MutationImpl::LockOp::XLOCK); |
7c673cae FG |
1621 | mut->finish_locking(lock); |
1622 | return true; | |
1623 | } | |
1624 | ||
a8e16298 TL |
1625 | if (lock->get_type() == CEPH_LOCK_IFILE && |
1626 | in->state_test(CInode::STATE_RECOVERING)) { | |
1627 | mds->mdcache->recovery_queue.prioritize(in); | |
7c673cae FG |
1628 | } |
1629 | ||
1630 | if (!lock->is_stable() && (lock->get_state() != LOCK_XLOCKDONE || | |
1631 | lock->get_xlock_by_client() != client || | |
1632 | lock->is_waiter_for(SimpleLock::WAIT_STABLE))) | |
1633 | break; | |
1634 | ||
1635 | if (lock->get_state() == LOCK_LOCK || lock->get_state() == LOCK_XLOCKDONE) { | |
1636 | mut->start_locking(lock); | |
1637 | simple_xlock(lock); | |
1638 | } else { | |
1639 | simple_lock(lock); | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | lock->add_waiter(SimpleLock::WAIT_WR|SimpleLock::WAIT_STABLE, new C_MDS_RetryRequest(mdcache, mut)); | |
1644 | nudge_log(lock); | |
1645 | return false; | |
1646 | } else { | |
1647 | // replica | |
11fdf7f2 TL |
1648 | ceph_assert(lock->get_sm()->can_remote_xlock); |
1649 | ceph_assert(!mut->slave_request); | |
7c673cae FG |
1650 | |
1651 | // wait for single auth | |
1652 | if (lock->get_parent()->is_ambiguous_auth()) { | |
1653 | lock->get_parent()->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, | |
1654 | new C_MDS_RetryRequest(mdcache, mut)); | |
1655 | return false; | |
1656 | } | |
1657 | ||
1658 | // wait for active auth | |
1659 | mds_rank_t auth = lock->get_parent()->authority().first; | |
1660 | if (mds->is_cluster_degraded() && | |
1661 | !mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) { | |
1662 | dout(7) << " mds." << auth << " is not active" << dendl; | |
1663 | if (mut->more()->waiting_on_slave.empty()) | |
1664 | mds->wait_for_active_peer(auth, new C_MDS_RetryRequest(mdcache, mut)); | |
1665 | return false; | |
1666 | } | |
1667 | ||
1668 | // send lock request | |
1669 | mut->more()->slaves.insert(auth); | |
1670 | mut->start_locking(lock, auth); | |
11fdf7f2 | 1671 | auto r = MMDSSlaveRequest::create(mut->reqid, mut->attempt, MMDSSlaveRequest::OP_XLOCK); |
7c673cae FG |
1672 | r->set_lock_type(lock->get_type()); |
1673 | lock->get_parent()->set_object_info(r->get_object_info()); | |
1674 | mds->send_message_mds(r, auth); | |
1675 | ||
11fdf7f2 | 1676 | ceph_assert(mut->more()->waiting_on_slave.count(auth) == 0); |
7c673cae FG |
1677 | mut->more()->waiting_on_slave.insert(auth); |
1678 | ||
1679 | return false; | |
1680 | } | |
1681 | } | |
1682 | ||
1683 | void Locker::_finish_xlock(SimpleLock *lock, client_t xlocker, bool *pneed_issue) | |
1684 | { | |
11fdf7f2 TL |
1685 | ceph_assert(!lock->is_stable()); |
1686 | if (lock->get_type() != CEPH_LOCK_DN && | |
1687 | lock->get_type() != CEPH_LOCK_ISNAP && | |
1688 | lock->get_num_rdlocks() == 0 && | |
7c673cae | 1689 | lock->get_num_wrlocks() == 0 && |
b32b8144 | 1690 | !lock->is_leased() && |
11fdf7f2 | 1691 | lock->get_state() != LOCK_XLOCKSNAP) { |
7c673cae FG |
1692 | CInode *in = static_cast<CInode*>(lock->get_parent()); |
1693 | client_t loner = in->get_target_loner(); | |
1694 | if (loner >= 0 && (xlocker < 0 || xlocker == loner)) { | |
1695 | lock->set_state(LOCK_EXCL); | |
1696 | lock->get_parent()->auth_unpin(lock); | |
1697 | lock->finish_waiters(SimpleLock::WAIT_STABLE|SimpleLock::WAIT_WR|SimpleLock::WAIT_RD); | |
1698 | if (lock->get_cap_shift()) | |
1699 | *pneed_issue = true; | |
1700 | if (lock->get_parent()->is_auth() && | |
1701 | lock->is_stable()) | |
1702 | try_eval(lock, pneed_issue); | |
1703 | return; | |
1704 | } | |
1705 | } | |
1706 | // the xlocker may have CEPH_CAP_GSHARED, need to revoke it if next state is LOCK_LOCK | |
1707 | eval_gather(lock, lock->get_state() != LOCK_XLOCKSNAP, pneed_issue); | |
1708 | } | |
1709 | ||
11fdf7f2 | 1710 | void Locker::xlock_finish(const MutationImpl::lock_iterator& it, MutationImpl *mut, bool *pneed_issue) |
7c673cae | 1711 | { |
11fdf7f2 TL |
1712 | ceph_assert(it->is_xlock()); |
1713 | SimpleLock *lock = it->lock; | |
1714 | ||
7c673cae FG |
1715 | if (lock->get_type() == CEPH_LOCK_IVERSION || |
1716 | lock->get_type() == CEPH_LOCK_DVERSION) | |
11fdf7f2 | 1717 | return local_xlock_finish(it, mut); |
7c673cae FG |
1718 | |
1719 | dout(10) << "xlock_finish on " << *lock << " " << *lock->get_parent() << dendl; | |
1720 | ||
1721 | client_t xlocker = lock->get_xlock_by_client(); | |
1722 | ||
1723 | // drop ref | |
1724 | lock->put_xlock(); | |
11fdf7f2 TL |
1725 | ceph_assert(mut); |
1726 | mut->locks.erase(it); | |
7c673cae FG |
1727 | |
1728 | bool do_issue = false; | |
1729 | ||
1730 | // remote xlock? | |
1731 | if (!lock->get_parent()->is_auth()) { | |
11fdf7f2 | 1732 | ceph_assert(lock->get_sm()->can_remote_xlock); |
7c673cae FG |
1733 | |
1734 | // tell auth | |
1735 | dout(7) << "xlock_finish releasing remote xlock on " << *lock->get_parent() << dendl; | |
1736 | mds_rank_t auth = lock->get_parent()->authority().first; | |
1737 | if (!mds->is_cluster_degraded() || | |
1738 | mds->mdsmap->get_state(auth) >= MDSMap::STATE_REJOIN) { | |
11fdf7f2 | 1739 | auto slavereq = MMDSSlaveRequest::create(mut->reqid, mut->attempt, MMDSSlaveRequest::OP_UNXLOCK); |
7c673cae FG |
1740 | slavereq->set_lock_type(lock->get_type()); |
1741 | lock->get_parent()->set_object_info(slavereq->get_object_info()); | |
1742 | mds->send_message_mds(slavereq, auth); | |
1743 | } | |
1744 | // others waiting? | |
1745 | lock->finish_waiters(SimpleLock::WAIT_STABLE | | |
1746 | SimpleLock::WAIT_WR | | |
1747 | SimpleLock::WAIT_RD, 0); | |
1748 | } else { | |
a8e16298 TL |
1749 | if (lock->get_num_xlocks() == 0 && |
1750 | lock->get_state() != LOCK_LOCK_XLOCK) { // no one is taking xlock | |
7c673cae FG |
1751 | _finish_xlock(lock, xlocker, &do_issue); |
1752 | } | |
1753 | } | |
1754 | ||
1755 | if (do_issue) { | |
1756 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
1757 | if (in->is_head()) { | |
1758 | if (pneed_issue) | |
1759 | *pneed_issue = true; | |
1760 | else | |
1761 | issue_caps(in); | |
1762 | } | |
1763 | } | |
1764 | } | |
1765 | ||
11fdf7f2 | 1766 | void Locker::xlock_export(const MutationImpl::lock_iterator& it, MutationImpl *mut) |
7c673cae | 1767 | { |
11fdf7f2 TL |
1768 | ceph_assert(it->is_xlock()); |
1769 | SimpleLock *lock = it->lock; | |
7c673cae FG |
1770 | dout(10) << "xlock_export on " << *lock << " " << *lock->get_parent() << dendl; |
1771 | ||
1772 | lock->put_xlock(); | |
11fdf7f2 | 1773 | mut->locks.erase(it); |
7c673cae FG |
1774 | |
1775 | MDSCacheObject *p = lock->get_parent(); | |
11fdf7f2 | 1776 | ceph_assert(p->state_test(CInode::STATE_AMBIGUOUSAUTH)); // we are exporting this (inode) |
7c673cae FG |
1777 | |
1778 | if (!lock->is_stable()) | |
1779 | lock->get_parent()->auth_unpin(lock); | |
1780 | ||
1781 | lock->set_state(LOCK_LOCK); | |
1782 | } | |
1783 | ||
1784 | void Locker::xlock_import(SimpleLock *lock) | |
1785 | { | |
1786 | dout(10) << "xlock_import on " << *lock << " " << *lock->get_parent() << dendl; | |
1787 | lock->get_parent()->auth_pin(lock); | |
1788 | } | |
1789 | ||
1790 | ||
1791 | ||
1792 | // file i/o ----------------------------------------- | |
1793 | ||
1794 | version_t Locker::issue_file_data_version(CInode *in) | |
1795 | { | |
1796 | dout(7) << "issue_file_data_version on " << *in << dendl; | |
1797 | return in->inode.file_data_version; | |
1798 | } | |
1799 | ||
1800 | class C_Locker_FileUpdate_finish : public LockerLogContext { | |
1801 | CInode *in; | |
1802 | MutationRef mut; | |
11fdf7f2 | 1803 | unsigned flags; |
7c673cae | 1804 | client_t client; |
11fdf7f2 | 1805 | MClientCaps::ref ack; |
7c673cae | 1806 | public: |
11fdf7f2 TL |
1807 | C_Locker_FileUpdate_finish(Locker *l, CInode *i, MutationRef& m, unsigned f, |
1808 | const MClientCaps::ref &ack, client_t c=-1) | |
1809 | : LockerLogContext(l), in(i), mut(m), flags(f), client(c), ack(ack) { | |
7c673cae FG |
1810 | in->get(CInode::PIN_PTRWAITER); |
1811 | } | |
1812 | void finish(int r) override { | |
11fdf7f2 | 1813 | locker->file_update_finish(in, mut, flags, client, ack); |
7c673cae FG |
1814 | in->put(CInode::PIN_PTRWAITER); |
1815 | } | |
1816 | }; | |
1817 | ||
11fdf7f2 TL |
1818 | enum { |
1819 | UPDATE_SHAREMAX = 1, | |
1820 | UPDATE_NEEDSISSUE = 2, | |
1821 | UPDATE_SNAPFLUSH = 4, | |
1822 | }; | |
1823 | ||
1824 | void Locker::file_update_finish(CInode *in, MutationRef& mut, unsigned flags, | |
1825 | client_t client, const MClientCaps::ref &ack) | |
7c673cae FG |
1826 | { |
1827 | dout(10) << "file_update_finish on " << *in << dendl; | |
1828 | in->pop_and_dirty_projected_inode(mut->ls); | |
1829 | ||
1830 | mut->apply(); | |
11fdf7f2 | 1831 | |
7c673cae FG |
1832 | if (ack) { |
1833 | Session *session = mds->get_session(client); | |
1834 | if (session) { | |
1835 | // "oldest flush tid" > 0 means client uses unique TID for each flush | |
1836 | if (ack->get_oldest_flush_tid() > 0) | |
11fdf7f2 | 1837 | session->add_completed_flush(ack->get_client_tid()); |
7c673cae FG |
1838 | mds->send_message_client_counted(ack, session); |
1839 | } else { | |
1840 | dout(10) << " no session for client." << client << " " << *ack << dendl; | |
7c673cae FG |
1841 | } |
1842 | } | |
1843 | ||
1844 | set<CInode*> need_issue; | |
1845 | drop_locks(mut.get(), &need_issue); | |
1846 | ||
11fdf7f2 TL |
1847 | if (in->is_head()) { |
1848 | if ((flags & UPDATE_NEEDSISSUE) && need_issue.count(in) == 0) { | |
1849 | Capability *cap = in->get_client_cap(client); | |
1850 | if (cap && (cap->wanted() & ~cap->pending())) | |
1851 | issue_caps(in, cap); | |
1852 | } | |
1853 | ||
1854 | if ((flags & UPDATE_SHAREMAX) && in->is_auth() && | |
1855 | (in->filelock.gcaps_allowed(CAP_LONER) & (CEPH_CAP_GWR|CEPH_CAP_GBUFFER))) | |
1856 | share_inode_max_size(in); | |
1857 | ||
1858 | } else if ((flags & UPDATE_SNAPFLUSH) && !in->client_snap_caps.empty()) { | |
7c673cae FG |
1859 | dout(10) << " client_snap_caps " << in->client_snap_caps << dendl; |
1860 | // check for snap writeback completion | |
1861 | bool gather = false; | |
94b18763 | 1862 | auto p = in->client_snap_caps.begin(); |
7c673cae | 1863 | while (p != in->client_snap_caps.end()) { |
11fdf7f2 TL |
1864 | auto q = p->second.find(client); |
1865 | if (q != p->second.end()) { | |
1866 | SimpleLock *lock = in->get_lock(p->first); | |
1867 | ceph_assert(lock); | |
1868 | dout(10) << " completing client_snap_caps for " << ccap_string(p->first) | |
1869 | << " lock " << *lock << " on " << *in << dendl; | |
1870 | lock->put_wrlock(); | |
1871 | ||
1872 | p->second.erase(q); | |
1873 | if (p->second.empty()) { | |
1874 | gather = true; | |
1875 | in->client_snap_caps.erase(p++); | |
1876 | } else | |
1877 | ++p; | |
1878 | } | |
7c673cae FG |
1879 | } |
1880 | if (gather) { | |
11fdf7f2 | 1881 | if (in->client_snap_caps.empty()) { |
7c673cae | 1882 | in->item_open_file.remove_myself(); |
11fdf7f2 TL |
1883 | in->item_caps.remove_myself(); |
1884 | } | |
7c673cae FG |
1885 | eval_cap_gather(in, &need_issue); |
1886 | } | |
7c673cae FG |
1887 | } |
1888 | issue_caps_set(need_issue); | |
1889 | ||
11fdf7f2 | 1890 | mds->balancer->hit_inode(in, META_POP_IWR); |
28e407b8 | 1891 | |
7c673cae FG |
1892 | // auth unpin after issuing caps |
1893 | mut->cleanup(); | |
1894 | } | |
1895 | ||
1896 | Capability* Locker::issue_new_caps(CInode *in, | |
1897 | int mode, | |
1898 | Session *session, | |
1899 | SnapRealm *realm, | |
1900 | bool is_replay) | |
1901 | { | |
1902 | dout(7) << "issue_new_caps for mode " << mode << " on " << *in << dendl; | |
1903 | bool is_new; | |
1904 | ||
1905 | // if replay, try to reconnect cap, and otherwise do nothing. | |
a8e16298 TL |
1906 | if (is_replay) |
1907 | return mds->mdcache->try_reconnect_cap(in, session); | |
1908 | ||
7c673cae FG |
1909 | |
1910 | // my needs | |
11fdf7f2 TL |
1911 | ceph_assert(session->info.inst.name.is_client()); |
1912 | client_t my_client = session->get_client(); | |
7c673cae FG |
1913 | int my_want = ceph_caps_for_mode(mode); |
1914 | ||
1915 | // register a capability | |
1916 | Capability *cap = in->get_client_cap(my_client); | |
1917 | if (!cap) { | |
1918 | // new cap | |
1919 | cap = in->add_client_cap(my_client, session, realm); | |
1920 | cap->set_wanted(my_want); | |
1921 | cap->mark_new(); | |
1922 | cap->inc_suppress(); // suppress file cap messages for new cap (we'll bundle with the open() reply) | |
1923 | is_new = true; | |
1924 | } else { | |
1925 | is_new = false; | |
1926 | // make sure it wants sufficient caps | |
1927 | if (my_want & ~cap->wanted()) { | |
1928 | // augment wanted caps for this client | |
1929 | cap->set_wanted(cap->wanted() | my_want); | |
1930 | } | |
1931 | } | |
1932 | ||
1933 | if (in->is_auth()) { | |
1934 | // [auth] twiddle mode? | |
1935 | eval(in, CEPH_CAP_LOCKS); | |
1936 | ||
1937 | if (_need_flush_mdlog(in, my_want)) | |
1938 | mds->mdlog->flush(); | |
1939 | ||
1940 | } else { | |
1941 | // [replica] tell auth about any new caps wanted | |
1942 | request_inode_file_caps(in); | |
1943 | } | |
1944 | ||
1945 | // issue caps (pot. incl new one) | |
1946 | //issue_caps(in); // note: _eval above may have done this already... | |
1947 | ||
1948 | // re-issue whatever we can | |
1949 | //cap->issue(cap->pending()); | |
1950 | ||
1951 | if (is_new) | |
1952 | cap->dec_suppress(); | |
1953 | ||
1954 | return cap; | |
1955 | } | |
1956 | ||
1957 | ||
1958 | void Locker::issue_caps_set(set<CInode*>& inset) | |
1959 | { | |
1960 | for (set<CInode*>::iterator p = inset.begin(); p != inset.end(); ++p) | |
1961 | issue_caps(*p); | |
1962 | } | |
1963 | ||
1964 | bool Locker::issue_caps(CInode *in, Capability *only_cap) | |
1965 | { | |
1966 | // allowed caps are determined by the lock mode. | |
1967 | int all_allowed = in->get_caps_allowed_by_type(CAP_ANY); | |
1968 | int loner_allowed = in->get_caps_allowed_by_type(CAP_LONER); | |
1969 | int xlocker_allowed = in->get_caps_allowed_by_type(CAP_XLOCKER); | |
1970 | ||
1971 | client_t loner = in->get_loner(); | |
1972 | if (loner >= 0) { | |
1973 | dout(7) << "issue_caps loner client." << loner | |
1974 | << " allowed=" << ccap_string(loner_allowed) | |
1975 | << ", xlocker allowed=" << ccap_string(xlocker_allowed) | |
1976 | << ", others allowed=" << ccap_string(all_allowed) | |
1977 | << " on " << *in << dendl; | |
1978 | } else { | |
1979 | dout(7) << "issue_caps allowed=" << ccap_string(all_allowed) | |
1980 | << ", xlocker allowed=" << ccap_string(xlocker_allowed) | |
1981 | << " on " << *in << dendl; | |
1982 | } | |
1983 | ||
11fdf7f2 | 1984 | ceph_assert(in->is_head()); |
7c673cae FG |
1985 | |
1986 | // count conflicts with | |
1987 | int nissued = 0; | |
1988 | ||
1989 | // client caps | |
11fdf7f2 | 1990 | map<client_t, Capability>::iterator it; |
7c673cae FG |
1991 | if (only_cap) |
1992 | it = in->client_caps.find(only_cap->get_client()); | |
1993 | else | |
1994 | it = in->client_caps.begin(); | |
1995 | for (; it != in->client_caps.end(); ++it) { | |
11fdf7f2 | 1996 | Capability *cap = &it->second; |
7c673cae FG |
1997 | if (cap->is_stale()) |
1998 | continue; | |
1999 | ||
2000 | // do not issue _new_ bits when size|mtime is projected | |
2001 | int allowed; | |
2002 | if (loner == it->first) | |
2003 | allowed = loner_allowed; | |
2004 | else | |
2005 | allowed = all_allowed; | |
2006 | ||
2007 | // add in any xlocker-only caps (for locks this client is the xlocker for) | |
2008 | allowed |= xlocker_allowed & in->get_xlocker_mask(it->first); | |
2009 | ||
11fdf7f2 TL |
2010 | if ((in->inode.inline_data.version != CEPH_INLINE_NONE && |
2011 | cap->is_noinline()) || | |
2012 | (!in->inode.layout.pool_ns.empty() && | |
2013 | cap->is_nopoolns())) | |
7c673cae FG |
2014 | allowed &= ~(CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR); |
2015 | ||
2016 | int pending = cap->pending(); | |
2017 | int wanted = cap->wanted(); | |
2018 | ||
2019 | dout(20) << " client." << it->first | |
2020 | << " pending " << ccap_string(pending) | |
2021 | << " allowed " << ccap_string(allowed) | |
2022 | << " wanted " << ccap_string(wanted) | |
2023 | << dendl; | |
2024 | ||
2025 | if (!(pending & ~allowed)) { | |
2026 | // skip if suppress or new, and not revocation | |
2027 | if (cap->is_new() || cap->is_suppress()) { | |
2028 | dout(20) << " !revoke and new|suppressed, skipping client." << it->first << dendl; | |
2029 | continue; | |
2030 | } | |
2031 | } | |
2032 | ||
2033 | // notify clients about deleted inode, to make sure they release caps ASAP. | |
2034 | if (in->inode.nlink == 0) | |
2035 | wanted |= CEPH_CAP_LINK_SHARED; | |
2036 | ||
2037 | // are there caps that the client _wants_ and can have, but aren't pending? | |
2038 | // or do we need to revoke? | |
2039 | if (((wanted & allowed) & ~pending) || // missing wanted+allowed caps | |
2040 | (pending & ~allowed)) { // need to revoke ~allowed caps. | |
2041 | // issue | |
2042 | nissued++; | |
2043 | ||
2044 | // include caps that clients generally like, while we're at it. | |
2045 | int likes = in->get_caps_liked(); | |
2046 | int before = pending; | |
2047 | long seq; | |
2048 | if (pending & ~allowed) | |
2049 | seq = cap->issue((wanted|likes) & allowed & pending); // if revoking, don't issue anything new. | |
2050 | else | |
2051 | seq = cap->issue((wanted|likes) & allowed); | |
2052 | int after = cap->pending(); | |
2053 | ||
2054 | if (cap->is_new()) { | |
2055 | // haven't send caps to client yet | |
2056 | if (before & ~after) | |
2057 | cap->confirm_receipt(seq, after); | |
2058 | } else { | |
2059 | dout(7) << " sending MClientCaps to client." << it->first | |
2060 | << " seq " << cap->get_last_seq() | |
2061 | << " new pending " << ccap_string(after) << " was " << ccap_string(before) | |
2062 | << dendl; | |
2063 | ||
2064 | int op = (before & ~after) ? CEPH_CAP_OP_REVOKE : CEPH_CAP_OP_GRANT; | |
2065 | if (op == CEPH_CAP_OP_REVOKE) { | |
2066 | revoking_caps.push_back(&cap->item_revoking_caps); | |
2067 | revoking_caps_by_client[cap->get_client()].push_back(&cap->item_client_revoking_caps); | |
2068 | cap->set_last_revoke_stamp(ceph_clock_now()); | |
2069 | cap->reset_num_revoke_warnings(); | |
2070 | } | |
2071 | ||
11fdf7f2 TL |
2072 | auto m = MClientCaps::create(op, in->ino(), |
2073 | in->find_snaprealm()->inode->ino(), | |
2074 | cap->get_cap_id(), | |
2075 | cap->get_last_seq(), | |
2076 | after, wanted, 0, | |
2077 | cap->get_mseq(), | |
2078 | mds->get_osd_epoch_barrier()); | |
7c673cae FG |
2079 | in->encode_cap_message(m, cap); |
2080 | ||
11fdf7f2 | 2081 | mds->send_message_client_counted(m, cap->get_session()); |
7c673cae FG |
2082 | } |
2083 | } | |
2084 | ||
2085 | if (only_cap) | |
2086 | break; | |
2087 | } | |
2088 | ||
2089 | return (nissued == 0); // true if no re-issued, no callbacks | |
2090 | } | |
2091 | ||
2092 | void Locker::issue_truncate(CInode *in) | |
2093 | { | |
2094 | dout(7) << "issue_truncate on " << *in << dendl; | |
2095 | ||
11fdf7f2 TL |
2096 | for (auto &p : in->client_caps) { |
2097 | Capability *cap = &p.second; | |
2098 | auto m = MClientCaps::create(CEPH_CAP_OP_TRUNC, | |
2099 | in->ino(), | |
2100 | in->find_snaprealm()->inode->ino(), | |
2101 | cap->get_cap_id(), cap->get_last_seq(), | |
2102 | cap->pending(), cap->wanted(), 0, | |
2103 | cap->get_mseq(), | |
2104 | mds->get_osd_epoch_barrier()); | |
7c673cae | 2105 | in->encode_cap_message(m, cap); |
11fdf7f2 | 2106 | mds->send_message_client_counted(m, p.first); |
7c673cae FG |
2107 | } |
2108 | ||
2109 | // should we increase max_size? | |
2110 | if (in->is_auth() && in->is_file()) | |
2111 | check_inode_max_size(in); | |
2112 | } | |
2113 | ||
a8e16298 | 2114 | void Locker::revoke_stale_caps(Session *session) |
7c673cae | 2115 | { |
a8e16298 | 2116 | dout(10) << "revoke_stale_caps for " << session->info.inst.name << dendl; |
7c673cae | 2117 | |
a8e16298 TL |
2118 | std::vector<CInode*> to_eval; |
2119 | ||
2120 | for (auto p = session->caps.begin(); !p.end(); ) { | |
2121 | Capability *cap = *p; | |
2122 | ++p; | |
2123 | if (!cap->is_notable()) { | |
2124 | // the rest ones are not being revoked and don't have writeable range | |
2125 | // and don't want exclusive caps or want file read/write. They don't | |
2126 | // need recover, they don't affect eval_gather()/try_eval() | |
2127 | break; | |
2128 | } | |
2129 | ||
2130 | int issued = cap->issued(); | |
2131 | if (!(issued & ~CEPH_CAP_PIN)) | |
2132 | continue; | |
2133 | ||
2134 | CInode *in = cap->get_inode(); | |
7c673cae FG |
2135 | dout(10) << " revoking " << ccap_string(issued) << " on " << *in << dendl; |
2136 | cap->revoke(); | |
2137 | ||
2138 | if (in->is_auth() && | |
2139 | in->inode.client_ranges.count(cap->get_client())) | |
2140 | in->state_set(CInode::STATE_NEEDSRECOVER); | |
2141 | ||
a8e16298 TL |
2142 | // eval lock/inode may finish contexts, which may modify other cap's position |
2143 | // in the session->caps. | |
2144 | to_eval.push_back(in); | |
7c673cae | 2145 | } |
7c673cae | 2146 | |
a8e16298 TL |
2147 | // invalidate the rest |
2148 | session->inc_cap_gen(); | |
7c673cae | 2149 | |
a8e16298 TL |
2150 | for (auto in : to_eval) { |
2151 | if (in->state_test(CInode::STATE_EXPORTINGCAPS)) | |
2152 | continue; | |
2153 | ||
2154 | if (!in->filelock.is_stable()) | |
2155 | eval_gather(&in->filelock); | |
2156 | if (!in->linklock.is_stable()) | |
2157 | eval_gather(&in->linklock); | |
2158 | if (!in->authlock.is_stable()) | |
2159 | eval_gather(&in->authlock); | |
2160 | if (!in->xattrlock.is_stable()) | |
2161 | eval_gather(&in->xattrlock); | |
2162 | ||
2163 | if (in->is_auth()) | |
2164 | try_eval(in, CEPH_CAP_LOCKS); | |
2165 | else | |
2166 | request_inode_file_caps(in); | |
7c673cae FG |
2167 | } |
2168 | } | |
2169 | ||
2170 | void Locker::resume_stale_caps(Session *session) | |
2171 | { | |
2172 | dout(10) << "resume_stale_caps for " << session->info.inst.name << dendl; | |
2173 | ||
11fdf7f2 | 2174 | bool lazy = session->info.has_feature(CEPHFS_FEATURE_LAZY_CAP_WANTED); |
a8e16298 | 2175 | for (xlist<Capability*>::iterator p = session->caps.begin(); !p.end(); ) { |
7c673cae | 2176 | Capability *cap = *p; |
a8e16298 | 2177 | ++p; |
11fdf7f2 | 2178 | if (lazy && !cap->is_notable()) |
a8e16298 TL |
2179 | break; // see revoke_stale_caps() |
2180 | ||
7c673cae | 2181 | CInode *in = cap->get_inode(); |
a8e16298 TL |
2182 | ceph_assert(in->is_head()); |
2183 | dout(10) << " clearing stale flag on " << *in << dendl; | |
7c673cae | 2184 | |
a8e16298 TL |
2185 | if (in->state_test(CInode::STATE_EXPORTINGCAPS)) { |
2186 | // if export succeeds, the cap will be removed. if export fails, | |
2187 | // we need to re-issue the cap if it's not stale. | |
2188 | in->state_set(CInode::STATE_EVALSTALECAPS); | |
2189 | continue; | |
7c673cae | 2190 | } |
a8e16298 TL |
2191 | |
2192 | if (!in->is_auth() || !eval(in, CEPH_CAP_LOCKS)) | |
2193 | issue_caps(in, cap); | |
7c673cae FG |
2194 | } |
2195 | } | |
2196 | ||
2197 | void Locker::remove_stale_leases(Session *session) | |
2198 | { | |
2199 | dout(10) << "remove_stale_leases for " << session->info.inst.name << dendl; | |
2200 | xlist<ClientLease*>::iterator p = session->leases.begin(); | |
2201 | while (!p.end()) { | |
2202 | ClientLease *l = *p; | |
2203 | ++p; | |
2204 | CDentry *parent = static_cast<CDentry*>(l->parent); | |
2205 | dout(15) << " removing lease on " << *parent << dendl; | |
2206 | parent->remove_client_lease(l, this); | |
2207 | } | |
2208 | } | |
2209 | ||
2210 | ||
2211 | class C_MDL_RequestInodeFileCaps : public LockerContext { | |
2212 | CInode *in; | |
2213 | public: | |
2214 | C_MDL_RequestInodeFileCaps(Locker *l, CInode *i) : LockerContext(l), in(i) { | |
2215 | in->get(CInode::PIN_PTRWAITER); | |
2216 | } | |
2217 | void finish(int r) override { | |
2218 | if (!in->is_auth()) | |
2219 | locker->request_inode_file_caps(in); | |
2220 | in->put(CInode::PIN_PTRWAITER); | |
2221 | } | |
2222 | }; | |
2223 | ||
2224 | void Locker::request_inode_file_caps(CInode *in) | |
2225 | { | |
11fdf7f2 | 2226 | ceph_assert(!in->is_auth()); |
7c673cae | 2227 | |
f64942e4 | 2228 | int wanted = in->get_caps_wanted() & in->get_caps_allowed_ever() & ~CEPH_CAP_PIN; |
7c673cae FG |
2229 | if (wanted != in->replica_caps_wanted) { |
2230 | // wait for single auth | |
2231 | if (in->is_ambiguous_auth()) { | |
2232 | in->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, | |
2233 | new C_MDL_RequestInodeFileCaps(this, in)); | |
2234 | return; | |
2235 | } | |
2236 | ||
2237 | mds_rank_t auth = in->authority().first; | |
2238 | if (mds->is_cluster_degraded() && | |
2239 | mds->mdsmap->get_state(auth) == MDSMap::STATE_REJOIN) { | |
2240 | mds->wait_for_active_peer(auth, new C_MDL_RequestInodeFileCaps(this, in)); | |
2241 | return; | |
2242 | } | |
2243 | ||
2244 | dout(7) << "request_inode_file_caps " << ccap_string(wanted) | |
2245 | << " was " << ccap_string(in->replica_caps_wanted) | |
2246 | << " on " << *in << " to mds." << auth << dendl; | |
2247 | ||
2248 | in->replica_caps_wanted = wanted; | |
2249 | ||
2250 | if (!mds->is_cluster_degraded() || | |
2251 | mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) | |
11fdf7f2 | 2252 | mds->send_message_mds(MInodeFileCaps::create(in->ino(), in->replica_caps_wanted), auth); |
7c673cae FG |
2253 | } |
2254 | } | |
2255 | ||
11fdf7f2 | 2256 | void Locker::handle_inode_file_caps(const MInodeFileCaps::const_ref &m) |
7c673cae FG |
2257 | { |
2258 | // nobody should be talking to us during recovery. | |
a8e16298 TL |
2259 | if (mds->get_state() < MDSMap::STATE_CLIENTREPLAY) { |
2260 | if (mds->get_want_state() >= MDSMap::STATE_CLIENTREPLAY) { | |
2261 | mds->wait_for_replay(new C_MDS_RetryMessage(mds, m)); | |
2262 | return; | |
2263 | } | |
11fdf7f2 | 2264 | ceph_abort_msg("got unexpected message during recovery"); |
a8e16298 | 2265 | } |
7c673cae FG |
2266 | |
2267 | // ok | |
2268 | CInode *in = mdcache->get_inode(m->get_ino()); | |
2269 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
2270 | ||
11fdf7f2 TL |
2271 | ceph_assert(in); |
2272 | ceph_assert(in->is_auth()); | |
7c673cae FG |
2273 | |
2274 | dout(7) << "handle_inode_file_caps replica mds." << from << " wants caps " << ccap_string(m->get_caps()) << " on " << *in << dendl; | |
2275 | ||
11fdf7f2 | 2276 | in->set_mds_caps_wanted(from, m->get_caps()); |
7c673cae FG |
2277 | |
2278 | try_eval(in, CEPH_CAP_LOCKS); | |
7c673cae FG |
2279 | } |
2280 | ||
2281 | ||
2282 | class C_MDL_CheckMaxSize : public LockerContext { | |
2283 | CInode *in; | |
2284 | uint64_t new_max_size; | |
2285 | uint64_t newsize; | |
2286 | utime_t mtime; | |
2287 | ||
2288 | public: | |
2289 | C_MDL_CheckMaxSize(Locker *l, CInode *i, uint64_t _new_max_size, | |
2290 | uint64_t _newsize, utime_t _mtime) : | |
2291 | LockerContext(l), in(i), | |
2292 | new_max_size(_new_max_size), newsize(_newsize), mtime(_mtime) | |
2293 | { | |
2294 | in->get(CInode::PIN_PTRWAITER); | |
2295 | } | |
2296 | void finish(int r) override { | |
2297 | if (in->is_auth()) | |
2298 | locker->check_inode_max_size(in, false, new_max_size, newsize, mtime); | |
2299 | in->put(CInode::PIN_PTRWAITER); | |
2300 | } | |
2301 | }; | |
2302 | ||
94b18763 | 2303 | uint64_t Locker::calc_new_max_size(CInode::mempool_inode *pi, uint64_t size) |
31f18b77 FG |
2304 | { |
2305 | uint64_t new_max = (size + 1) << 1; | |
11fdf7f2 | 2306 | uint64_t max_inc = g_conf()->mds_client_writeable_range_max_inc_objs; |
31f18b77 | 2307 | if (max_inc > 0) { |
b32b8144 | 2308 | max_inc *= pi->layout.object_size; |
11fdf7f2 | 2309 | new_max = std::min(new_max, size + max_inc); |
31f18b77 | 2310 | } |
11fdf7f2 | 2311 | return round_up_to(new_max, pi->get_layout_size_increment()); |
31f18b77 | 2312 | } |
7c673cae | 2313 | |
a8e16298 | 2314 | void Locker::calc_new_client_ranges(CInode *in, uint64_t size, bool update, |
94b18763 | 2315 | CInode::mempool_inode::client_range_map *new_ranges, |
7c673cae FG |
2316 | bool *max_increased) |
2317 | { | |
94b18763 | 2318 | auto latest = in->get_projected_inode(); |
7c673cae | 2319 | uint64_t ms; |
a8e16298 | 2320 | if (latest->has_layout()) { |
31f18b77 | 2321 | ms = calc_new_max_size(latest, size); |
7c673cae FG |
2322 | } else { |
2323 | // Layout-less directories like ~mds0/, have zero size | |
2324 | ms = 0; | |
2325 | } | |
2326 | ||
2327 | // increase ranges as appropriate. | |
2328 | // shrink to 0 if no WR|BUFFER caps issued. | |
11fdf7f2 TL |
2329 | for (auto &p : in->client_caps) { |
2330 | if ((p.second.issued() | p.second.wanted()) & CEPH_CAP_ANY_FILE_WR) { | |
2331 | client_writeable_range_t& nr = (*new_ranges)[p.first]; | |
7c673cae | 2332 | nr.range.first = 0; |
11fdf7f2 TL |
2333 | if (latest->client_ranges.count(p.first)) { |
2334 | client_writeable_range_t& oldr = latest->client_ranges[p.first]; | |
7c673cae FG |
2335 | if (ms > oldr.range.last) |
2336 | *max_increased = true; | |
11fdf7f2 | 2337 | nr.range.last = std::max(ms, oldr.range.last); |
7c673cae FG |
2338 | nr.follows = oldr.follows; |
2339 | } else { | |
31f18b77 | 2340 | *max_increased = true; |
7c673cae FG |
2341 | nr.range.last = ms; |
2342 | nr.follows = in->first - 1; | |
2343 | } | |
a8e16298 | 2344 | if (update) |
11fdf7f2 | 2345 | p.second.mark_clientwriteable(); |
a8e16298 TL |
2346 | } else { |
2347 | if (update) | |
11fdf7f2 | 2348 | p.second.clear_clientwriteable(); |
7c673cae FG |
2349 | } |
2350 | } | |
2351 | } | |
2352 | ||
2353 | bool Locker::check_inode_max_size(CInode *in, bool force_wrlock, | |
2354 | uint64_t new_max_size, uint64_t new_size, | |
2355 | utime_t new_mtime) | |
2356 | { | |
11fdf7f2 TL |
2357 | ceph_assert(in->is_auth()); |
2358 | ceph_assert(in->is_file()); | |
7c673cae | 2359 | |
94b18763 FG |
2360 | CInode::mempool_inode *latest = in->get_projected_inode(); |
2361 | CInode::mempool_inode::client_range_map new_ranges; | |
7c673cae FG |
2362 | uint64_t size = latest->size; |
2363 | bool update_size = new_size > 0; | |
2364 | bool update_max = false; | |
2365 | bool max_increased = false; | |
2366 | ||
2367 | if (update_size) { | |
11fdf7f2 TL |
2368 | new_size = size = std::max(size, new_size); |
2369 | new_mtime = std::max(new_mtime, latest->mtime); | |
7c673cae FG |
2370 | if (latest->size == new_size && latest->mtime == new_mtime) |
2371 | update_size = false; | |
2372 | } | |
2373 | ||
a8e16298 TL |
2374 | int can_update = 1; |
2375 | if (in->is_frozen()) { | |
2376 | can_update = -1; | |
2377 | } else if (!force_wrlock && !in->filelock.can_wrlock(in->get_loner())) { | |
2378 | // lock? | |
2379 | if (in->filelock.is_stable()) { | |
2380 | if (in->get_target_loner() >= 0) | |
2381 | file_excl(&in->filelock); | |
2382 | else | |
2383 | simple_lock(&in->filelock); | |
2384 | } | |
2385 | if (!in->filelock.can_wrlock(in->get_loner())) | |
2386 | can_update = -2; | |
2387 | } | |
2388 | ||
2389 | calc_new_client_ranges(in, std::max(new_max_size, size), can_update > 0, | |
2390 | &new_ranges, &max_increased); | |
7c673cae FG |
2391 | |
2392 | if (max_increased || latest->client_ranges != new_ranges) | |
2393 | update_max = true; | |
2394 | ||
2395 | if (!update_size && !update_max) { | |
2396 | dout(20) << "check_inode_max_size no-op on " << *in << dendl; | |
2397 | return false; | |
2398 | } | |
2399 | ||
2400 | dout(10) << "check_inode_max_size new_ranges " << new_ranges | |
2401 | << " update_size " << update_size | |
2402 | << " on " << *in << dendl; | |
2403 | ||
a8e16298 TL |
2404 | if (can_update < 0) { |
2405 | auto cms = new C_MDL_CheckMaxSize(this, in, new_max_size, new_size, new_mtime); | |
2406 | if (can_update == -1) { | |
2407 | dout(10) << "check_inode_max_size frozen, waiting on " << *in << dendl; | |
2408 | in->add_waiter(CInode::WAIT_UNFREEZE, cms); | |
2409 | } else { | |
7c673cae FG |
2410 | in->filelock.add_waiter(SimpleLock::WAIT_STABLE, cms); |
2411 | dout(10) << "check_inode_max_size can't wrlock, waiting on " << *in << dendl; | |
7c673cae | 2412 | } |
a8e16298 | 2413 | return false; |
7c673cae FG |
2414 | } |
2415 | ||
2416 | MutationRef mut(new MutationImpl()); | |
2417 | mut->ls = mds->mdlog->get_current_segment(); | |
2418 | ||
94b18763 FG |
2419 | auto &pi = in->project_inode(); |
2420 | pi.inode.version = in->pre_dirty(); | |
7c673cae FG |
2421 | |
2422 | if (update_max) { | |
94b18763 FG |
2423 | dout(10) << "check_inode_max_size client_ranges " << pi.inode.client_ranges << " -> " << new_ranges << dendl; |
2424 | pi.inode.client_ranges = new_ranges; | |
7c673cae FG |
2425 | } |
2426 | ||
2427 | if (update_size) { | |
94b18763 FG |
2428 | dout(10) << "check_inode_max_size size " << pi.inode.size << " -> " << new_size << dendl; |
2429 | pi.inode.size = new_size; | |
2430 | pi.inode.rstat.rbytes = new_size; | |
2431 | dout(10) << "check_inode_max_size mtime " << pi.inode.mtime << " -> " << new_mtime << dendl; | |
2432 | pi.inode.mtime = new_mtime; | |
91327a77 AA |
2433 | if (new_mtime > pi.inode.ctime) { |
2434 | pi.inode.ctime = new_mtime; | |
2435 | if (new_mtime > pi.inode.rstat.rctime) | |
2436 | pi.inode.rstat.rctime = new_mtime; | |
2437 | } | |
7c673cae FG |
2438 | } |
2439 | ||
2440 | // use EOpen if the file is still open; otherwise, use EUpdate. | |
2441 | // this is just an optimization to push open files forward into | |
2442 | // newer log segments. | |
2443 | LogEvent *le; | |
2444 | EMetaBlob *metablob; | |
2445 | if (in->is_any_caps_wanted() && in->last == CEPH_NOSNAP) { | |
2446 | EOpen *eo = new EOpen(mds->mdlog); | |
2447 | eo->add_ino(in->ino()); | |
2448 | metablob = &eo->metablob; | |
2449 | le = eo; | |
7c673cae FG |
2450 | } else { |
2451 | EUpdate *eu = new EUpdate(mds->mdlog, "check_inode_max_size"); | |
2452 | metablob = &eu->metablob; | |
2453 | le = eu; | |
2454 | } | |
2455 | mds->mdlog->start_entry(le); | |
2456 | if (update_size) { // FIXME if/when we do max_size nested accounting | |
2457 | mdcache->predirty_journal_parents(mut, metablob, in, 0, PREDIRTY_PRIMARY); | |
2458 | // no cow, here! | |
2459 | CDentry *parent = in->get_projected_parent_dn(); | |
2460 | metablob->add_primary_dentry(parent, in, true); | |
2461 | } else { | |
2462 | metablob->add_dir_context(in->get_projected_parent_dn()->get_dir()); | |
2463 | mdcache->journal_dirty_inode(mut.get(), metablob, in); | |
2464 | } | |
11fdf7f2 TL |
2465 | mds->mdlog->submit_entry(le, new C_Locker_FileUpdate_finish(this, in, mut, |
2466 | UPDATE_SHAREMAX, MClientCaps::ref())); | |
7c673cae FG |
2467 | wrlock_force(&in->filelock, mut); // wrlock for duration of journal |
2468 | mut->auth_pin(in); | |
2469 | ||
2470 | // make max_size _increase_ timely | |
2471 | if (max_increased) | |
2472 | mds->mdlog->flush(); | |
2473 | ||
2474 | return true; | |
2475 | } | |
2476 | ||
2477 | ||
2478 | void Locker::share_inode_max_size(CInode *in, Capability *only_cap) | |
2479 | { | |
2480 | /* | |
2481 | * only share if currently issued a WR cap. if client doesn't have it, | |
2482 | * file_max doesn't matter, and the client will get it if/when they get | |
2483 | * the cap later. | |
2484 | */ | |
2485 | dout(10) << "share_inode_max_size on " << *in << dendl; | |
11fdf7f2 | 2486 | map<client_t, Capability>::iterator it; |
7c673cae FG |
2487 | if (only_cap) |
2488 | it = in->client_caps.find(only_cap->get_client()); | |
2489 | else | |
2490 | it = in->client_caps.begin(); | |
2491 | for (; it != in->client_caps.end(); ++it) { | |
2492 | const client_t client = it->first; | |
11fdf7f2 | 2493 | Capability *cap = &it->second; |
7c673cae FG |
2494 | if (cap->is_suppress()) |
2495 | continue; | |
2496 | if (cap->pending() & (CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER)) { | |
2497 | dout(10) << "share_inode_max_size with client." << client << dendl; | |
2498 | cap->inc_last_seq(); | |
11fdf7f2 TL |
2499 | auto m = MClientCaps::create(CEPH_CAP_OP_GRANT, |
2500 | in->ino(), | |
2501 | in->find_snaprealm()->inode->ino(), | |
2502 | cap->get_cap_id(), | |
2503 | cap->get_last_seq(), | |
2504 | cap->pending(), | |
2505 | cap->wanted(), 0, | |
2506 | cap->get_mseq(), | |
2507 | mds->get_osd_epoch_barrier()); | |
7c673cae FG |
2508 | in->encode_cap_message(m, cap); |
2509 | mds->send_message_client_counted(m, client); | |
2510 | } | |
2511 | if (only_cap) | |
2512 | break; | |
2513 | } | |
2514 | } | |
2515 | ||
2516 | bool Locker::_need_flush_mdlog(CInode *in, int wanted) | |
2517 | { | |
2518 | /* flush log if caps are wanted by client but corresponding lock is unstable and locked by | |
2519 | * pending mutations. */ | |
2520 | if (((wanted & (CEPH_CAP_FILE_RD|CEPH_CAP_FILE_WR|CEPH_CAP_FILE_SHARED|CEPH_CAP_FILE_EXCL)) && | |
2521 | in->filelock.is_unstable_and_locked()) || | |
2522 | ((wanted & (CEPH_CAP_AUTH_SHARED|CEPH_CAP_AUTH_EXCL)) && | |
2523 | in->authlock.is_unstable_and_locked()) || | |
2524 | ((wanted & (CEPH_CAP_LINK_SHARED|CEPH_CAP_LINK_EXCL)) && | |
2525 | in->linklock.is_unstable_and_locked()) || | |
2526 | ((wanted & (CEPH_CAP_XATTR_SHARED|CEPH_CAP_XATTR_EXCL)) && | |
2527 | in->xattrlock.is_unstable_and_locked())) | |
2528 | return true; | |
2529 | return false; | |
2530 | } | |
2531 | ||
2532 | void Locker::adjust_cap_wanted(Capability *cap, int wanted, int issue_seq) | |
2533 | { | |
2534 | if (ceph_seq_cmp(issue_seq, cap->get_last_issue()) == 0) { | |
2535 | dout(10) << " wanted " << ccap_string(cap->wanted()) | |
2536 | << " -> " << ccap_string(wanted) << dendl; | |
2537 | cap->set_wanted(wanted); | |
2538 | } else if (wanted & ~cap->wanted()) { | |
2539 | dout(10) << " wanted " << ccap_string(cap->wanted()) | |
2540 | << " -> " << ccap_string(wanted) | |
2541 | << " (added caps even though we had seq mismatch!)" << dendl; | |
2542 | cap->set_wanted(wanted | cap->wanted()); | |
2543 | } else { | |
2544 | dout(10) << " NOT changing wanted " << ccap_string(cap->wanted()) | |
2545 | << " -> " << ccap_string(wanted) | |
2546 | << " (issue_seq " << issue_seq << " != last_issue " | |
2547 | << cap->get_last_issue() << ")" << dendl; | |
2548 | return; | |
2549 | } | |
2550 | ||
2551 | CInode *cur = cap->get_inode(); | |
2552 | if (!cur->is_auth()) { | |
2553 | request_inode_file_caps(cur); | |
2554 | return; | |
2555 | } | |
2556 | ||
11fdf7f2 | 2557 | if (cap->wanted()) { |
7c673cae FG |
2558 | if (cur->state_test(CInode::STATE_RECOVERING) && |
2559 | (cap->wanted() & (CEPH_CAP_FILE_RD | | |
2560 | CEPH_CAP_FILE_WR))) { | |
2561 | mds->mdcache->recovery_queue.prioritize(cur); | |
2562 | } | |
2563 | ||
11fdf7f2 TL |
2564 | if (mdcache->open_file_table.should_log_open(cur)) { |
2565 | ceph_assert(cur->last == CEPH_NOSNAP); | |
7c673cae FG |
2566 | EOpen *le = new EOpen(mds->mdlog); |
2567 | mds->mdlog->start_entry(le); | |
2568 | le->add_clean_inode(cur); | |
7c673cae FG |
2569 | mds->mdlog->submit_entry(le); |
2570 | } | |
2571 | } | |
2572 | } | |
2573 | ||
11fdf7f2 TL |
2574 | void Locker::snapflush_nudge(CInode *in) |
2575 | { | |
2576 | ceph_assert(in->last != CEPH_NOSNAP); | |
2577 | if (in->client_snap_caps.empty()) | |
2578 | return; | |
2579 | ||
2580 | CInode *head = mdcache->get_inode(in->ino()); | |
2581 | // head inode gets unpinned when snapflush starts. It might get trimmed | |
2582 | // before snapflush finishes. | |
2583 | if (!head) | |
2584 | return; | |
7c673cae | 2585 | |
11fdf7f2 TL |
2586 | ceph_assert(head->is_auth()); |
2587 | if (head->client_need_snapflush.empty()) | |
2588 | return; | |
2589 | ||
2590 | SimpleLock *hlock = head->get_lock(CEPH_LOCK_IFILE); | |
2591 | if (hlock->get_state() == LOCK_SYNC || !hlock->is_stable()) { | |
2592 | hlock = NULL; | |
2593 | for (int i = 0; i < num_cinode_locks; i++) { | |
2594 | SimpleLock *lock = head->get_lock(cinode_lock_info[i].lock); | |
2595 | if (lock->get_state() != LOCK_SYNC && lock->is_stable()) { | |
2596 | hlock = lock; | |
2597 | break; | |
2598 | } | |
2599 | } | |
2600 | } | |
2601 | if (hlock) { | |
2602 | _rdlock_kick(hlock, true); | |
2603 | } else { | |
2604 | // also, requeue, in case of unstable lock | |
2605 | need_snapflush_inodes.push_back(&in->item_caps); | |
2606 | } | |
2607 | } | |
2608 | ||
2609 | void Locker::mark_need_snapflush_inode(CInode *in) | |
2610 | { | |
2611 | ceph_assert(in->last != CEPH_NOSNAP); | |
2612 | if (!in->item_caps.is_on_list()) { | |
2613 | need_snapflush_inodes.push_back(&in->item_caps); | |
2614 | utime_t now = ceph_clock_now(); | |
2615 | in->last_dirstat_prop = now; | |
2616 | dout(10) << "mark_need_snapflush_inode " << *in << " - added at " << now << dendl; | |
2617 | } | |
2618 | } | |
7c673cae | 2619 | |
c07f9fc5 | 2620 | void Locker::_do_null_snapflush(CInode *head_in, client_t client, snapid_t last) |
7c673cae FG |
2621 | { |
2622 | dout(10) << "_do_null_snapflush client." << client << " on " << *head_in << dendl; | |
c07f9fc5 FG |
2623 | for (auto p = head_in->client_need_snapflush.begin(); |
2624 | p != head_in->client_need_snapflush.end() && p->first < last; ) { | |
7c673cae | 2625 | snapid_t snapid = p->first; |
94b18763 | 2626 | auto &clients = p->second; |
7c673cae FG |
2627 | ++p; // be careful, q loop below depends on this |
2628 | ||
2629 | if (clients.count(client)) { | |
2630 | dout(10) << " doing async NULL snapflush on " << snapid << " from client." << client << dendl; | |
b32b8144 | 2631 | CInode *sin = mdcache->pick_inode_snap(head_in, snapid - 1); |
11fdf7f2 TL |
2632 | ceph_assert(sin); |
2633 | ceph_assert(sin->first <= snapid); | |
2634 | _do_snap_update(sin, snapid, 0, sin->first - 1, client, MClientCaps::ref(), MClientCaps::ref()); | |
7c673cae FG |
2635 | head_in->remove_need_snapflush(sin, snapid, client); |
2636 | } | |
2637 | } | |
2638 | } | |
2639 | ||
2640 | ||
2641 | bool Locker::should_defer_client_cap_frozen(CInode *in) | |
2642 | { | |
2643 | /* | |
2644 | * This policy needs to be AT LEAST as permissive as allowing a client request | |
2645 | * to go forward, or else a client request can release something, the release | |
2646 | * gets deferred, but the request gets processed and deadlocks because when the | |
2647 | * caps can't get revoked. | |
2648 | * | |
2649 | * Currently, a request wait if anything locked is freezing (can't | |
2650 | * auth_pin), which would avoid any deadlock with cap release. Thus @in | |
2651 | * _MUST_ be in the lock/auth_pin set. | |
2652 | * | |
2653 | * auth_pins==0 implies no unstable lock and not auth pinnned by | |
2654 | * client request, otherwise continue even it's freezing. | |
2655 | */ | |
2656 | return (in->is_freezing() && in->get_num_auth_pins() == 0) || in->is_frozen(); | |
2657 | } | |
2658 | ||
11fdf7f2 | 2659 | void Locker::handle_client_caps(const MClientCaps::const_ref &m) |
7c673cae | 2660 | { |
7c673cae | 2661 | client_t client = m->get_source().num(); |
7c673cae | 2662 | snapid_t follows = m->get_snap_follows(); |
11fdf7f2 TL |
2663 | auto op = m->get_op(); |
2664 | auto dirty = m->get_dirty(); | |
7c673cae | 2665 | dout(7) << "handle_client_caps " |
7c673cae FG |
2666 | << " on " << m->get_ino() |
2667 | << " tid " << m->get_client_tid() << " follows " << follows | |
11fdf7f2 TL |
2668 | << " op " << ceph_cap_op_name(op) |
2669 | << " flags 0x" << std::hex << m->flags << std::dec << dendl; | |
7c673cae | 2670 | |
94b18763 | 2671 | Session *session = mds->get_session(m); |
7c673cae FG |
2672 | if (!mds->is_clientreplay() && !mds->is_active() && !mds->is_stopping()) { |
2673 | if (!session) { | |
2674 | dout(5) << " no session, dropping " << *m << dendl; | |
7c673cae FG |
2675 | return; |
2676 | } | |
2677 | if (session->is_closed() || | |
2678 | session->is_closing() || | |
2679 | session->is_killing()) { | |
2680 | dout(7) << " session closed|closing|killing, dropping " << *m << dendl; | |
7c673cae FG |
2681 | return; |
2682 | } | |
11fdf7f2 TL |
2683 | if ((mds->is_reconnect() || mds->get_want_state() == MDSMap::STATE_RECONNECT) && |
2684 | dirty && m->get_client_tid() > 0 && | |
7c673cae | 2685 | !session->have_completed_flush(m->get_client_tid())) { |
11fdf7f2 TL |
2686 | mdcache->set_reconnected_dirty_caps(client, m->get_ino(), dirty, |
2687 | op == CEPH_CAP_OP_FLUSHSNAP); | |
7c673cae FG |
2688 | } |
2689 | mds->wait_for_replay(new C_MDS_RetryMessage(mds, m)); | |
2690 | return; | |
2691 | } | |
2692 | ||
2693 | if (m->get_client_tid() > 0 && session && | |
2694 | session->have_completed_flush(m->get_client_tid())) { | |
2695 | dout(7) << "handle_client_caps already flushed tid " << m->get_client_tid() | |
2696 | << " for client." << client << dendl; | |
11fdf7f2 TL |
2697 | MClientCaps::ref ack; |
2698 | if (op == CEPH_CAP_OP_FLUSHSNAP) { | |
2699 | ack = MClientCaps::create(CEPH_CAP_OP_FLUSHSNAP_ACK, m->get_ino(), 0, 0, 0, 0, 0, dirty, 0, mds->get_osd_epoch_barrier()); | |
7c673cae | 2700 | } else { |
11fdf7f2 | 2701 | ack = MClientCaps::create(CEPH_CAP_OP_FLUSH_ACK, m->get_ino(), 0, m->get_cap_id(), m->get_seq(), m->get_caps(), 0, dirty, 0, mds->get_osd_epoch_barrier()); |
7c673cae FG |
2702 | } |
2703 | ack->set_snap_follows(follows); | |
2704 | ack->set_client_tid(m->get_client_tid()); | |
2705 | mds->send_message_client_counted(ack, m->get_connection()); | |
11fdf7f2 | 2706 | if (op == CEPH_CAP_OP_FLUSHSNAP) { |
7c673cae FG |
2707 | return; |
2708 | } else { | |
2709 | // fall-thru because the message may release some caps | |
11fdf7f2 TL |
2710 | dirty = false; |
2711 | op = CEPH_CAP_OP_UPDATE; | |
7c673cae FG |
2712 | } |
2713 | } | |
2714 | ||
2715 | // "oldest flush tid" > 0 means client uses unique TID for each flush | |
2716 | if (m->get_oldest_flush_tid() > 0 && session) { | |
2717 | if (session->trim_completed_flushes(m->get_oldest_flush_tid())) { | |
2718 | mds->mdlog->get_current_segment()->touched_sessions.insert(session->info.inst.name); | |
2719 | ||
2720 | if (session->get_num_trim_flushes_warnings() > 0 && | |
11fdf7f2 | 2721 | session->get_num_completed_flushes() * 2 < g_conf()->mds_max_completed_flushes) |
7c673cae FG |
2722 | session->reset_num_trim_flushes_warnings(); |
2723 | } else { | |
2724 | if (session->get_num_completed_flushes() >= | |
11fdf7f2 | 2725 | (g_conf()->mds_max_completed_flushes << session->get_num_trim_flushes_warnings())) { |
7c673cae FG |
2726 | session->inc_num_trim_flushes_warnings(); |
2727 | stringstream ss; | |
2728 | ss << "client." << session->get_client() << " does not advance its oldest_flush_tid (" | |
2729 | << m->get_oldest_flush_tid() << "), " | |
2730 | << session->get_num_completed_flushes() | |
2731 | << " completed flushes recorded in session"; | |
2732 | mds->clog->warn() << ss.str(); | |
2733 | dout(20) << __func__ << " " << ss.str() << dendl; | |
2734 | } | |
2735 | } | |
2736 | } | |
2737 | ||
2738 | CInode *head_in = mdcache->get_inode(m->get_ino()); | |
2739 | if (!head_in) { | |
2740 | if (mds->is_clientreplay()) { | |
2741 | dout(7) << "handle_client_caps on unknown ino " << m->get_ino() | |
2742 | << ", will try again after replayed client requests" << dendl; | |
2743 | mdcache->wait_replay_cap_reconnect(m->get_ino(), new C_MDS_RetryMessage(mds, m)); | |
2744 | return; | |
2745 | } | |
1adf2230 AA |
2746 | |
2747 | /* | |
2748 | * "handle_client_caps on unknown ino xxx” is normal after migrating a subtree | |
2749 | * Sequence of events that cause this are: | |
2750 | * - client sends caps message to mds.a | |
2751 | * - mds finishes subtree migration, send cap export to client | |
2752 | * - mds trim its cache | |
2753 | * - mds receives cap messages from client | |
2754 | */ | |
2755 | dout(7) << "handle_client_caps on unknown ino " << m->get_ino() << ", dropping" << dendl; | |
7c673cae FG |
2756 | return; |
2757 | } | |
2758 | ||
2759 | if (m->osd_epoch_barrier && !mds->objecter->have_map(m->osd_epoch_barrier)) { | |
2760 | // Pause RADOS operations until we see the required epoch | |
2761 | mds->objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
2762 | } | |
2763 | ||
2764 | if (mds->get_osd_epoch_barrier() < m->osd_epoch_barrier) { | |
2765 | // Record the barrier so that we will retransmit it to clients | |
2766 | mds->set_osd_epoch_barrier(m->osd_epoch_barrier); | |
2767 | } | |
2768 | ||
b32b8144 | 2769 | dout(10) << " head inode " << *head_in << dendl; |
7c673cae FG |
2770 | |
2771 | Capability *cap = 0; | |
b32b8144 | 2772 | cap = head_in->get_client_cap(client); |
7c673cae | 2773 | if (!cap) { |
b32b8144 | 2774 | dout(7) << "handle_client_caps no cap for client." << client << " on " << *head_in << dendl; |
7c673cae FG |
2775 | return; |
2776 | } | |
11fdf7f2 | 2777 | ceph_assert(cap); |
7c673cae FG |
2778 | |
2779 | // freezing|frozen? | |
b32b8144 FG |
2780 | if (should_defer_client_cap_frozen(head_in)) { |
2781 | dout(7) << "handle_client_caps freezing|frozen on " << *head_in << dendl; | |
2782 | head_in->add_waiter(CInode::WAIT_UNFREEZE, new C_MDS_RetryMessage(mds, m)); | |
7c673cae FG |
2783 | return; |
2784 | } | |
2785 | if (ceph_seq_cmp(m->get_mseq(), cap->get_mseq()) < 0) { | |
2786 | dout(7) << "handle_client_caps mseq " << m->get_mseq() << " < " << cap->get_mseq() | |
2787 | << ", dropping" << dendl; | |
7c673cae FG |
2788 | return; |
2789 | } | |
2790 | ||
11fdf7f2 | 2791 | bool need_unpin = false; |
7c673cae FG |
2792 | |
2793 | // flushsnap? | |
2794 | if (op == CEPH_CAP_OP_FLUSHSNAP) { | |
b32b8144 FG |
2795 | if (!head_in->is_auth()) { |
2796 | dout(7) << " not auth, ignoring flushsnap on " << *head_in << dendl; | |
7c673cae FG |
2797 | goto out; |
2798 | } | |
2799 | ||
b32b8144 | 2800 | SnapRealm *realm = head_in->find_snaprealm(); |
7c673cae FG |
2801 | snapid_t snap = realm->get_snap_following(follows); |
2802 | dout(10) << " flushsnap follows " << follows << " -> snap " << snap << dendl; | |
2803 | ||
11fdf7f2 TL |
2804 | auto p = head_in->client_need_snapflush.begin(); |
2805 | if (p != head_in->client_need_snapflush.end() && p->first < snap) { | |
2806 | head_in->auth_pin(this); // prevent subtree frozen | |
2807 | need_unpin = true; | |
2808 | _do_null_snapflush(head_in, client, snap); | |
2809 | } | |
2810 | ||
b32b8144 FG |
2811 | CInode *in = head_in; |
2812 | if (snap != CEPH_NOSNAP) { | |
2813 | in = mdcache->pick_inode_snap(head_in, snap - 1); | |
2814 | if (in != head_in) | |
2815 | dout(10) << " snapped inode " << *in << dendl; | |
2816 | } | |
2817 | ||
7c673cae FG |
2818 | // we can prepare the ack now, since this FLUSHEDSNAP is independent of any |
2819 | // other cap ops. (except possibly duplicate FLUSHSNAP requests, but worst | |
2820 | // case we get a dup response, so whatever.) | |
11fdf7f2 TL |
2821 | MClientCaps::ref ack; |
2822 | if (dirty) { | |
2823 | ack = MClientCaps::create(CEPH_CAP_OP_FLUSHSNAP_ACK, in->ino(), 0, 0, 0, 0, 0, dirty, 0, mds->get_osd_epoch_barrier()); | |
7c673cae FG |
2824 | ack->set_snap_follows(follows); |
2825 | ack->set_client_tid(m->get_client_tid()); | |
2826 | ack->set_oldest_flush_tid(m->get_oldest_flush_tid()); | |
2827 | } | |
2828 | ||
2829 | if (in == head_in || | |
2830 | (head_in->client_need_snapflush.count(snap) && | |
2831 | head_in->client_need_snapflush[snap].count(client))) { | |
2832 | dout(7) << " flushsnap snap " << snap | |
2833 | << " client." << client << " on " << *in << dendl; | |
2834 | ||
2835 | // this cap now follows a later snap (i.e. the one initiating this flush, or later) | |
2836 | if (in == head_in) | |
2837 | cap->client_follows = snap < CEPH_NOSNAP ? snap : realm->get_newest_seq(); | |
2838 | ||
11fdf7f2 | 2839 | _do_snap_update(in, snap, dirty, follows, client, m, ack); |
7c673cae FG |
2840 | |
2841 | if (in != head_in) | |
2842 | head_in->remove_need_snapflush(in, snap, client); | |
7c673cae FG |
2843 | } else { |
2844 | dout(7) << " not expecting flushsnap " << snap << " from client." << client << " on " << *in << dendl; | |
2845 | if (ack) | |
2846 | mds->send_message_client_counted(ack, m->get_connection()); | |
2847 | } | |
2848 | goto out; | |
2849 | } | |
2850 | ||
2851 | if (cap->get_cap_id() != m->get_cap_id()) { | |
2852 | dout(7) << " ignoring client capid " << m->get_cap_id() << " != my " << cap->get_cap_id() << dendl; | |
2853 | } else { | |
b32b8144 FG |
2854 | CInode *in = head_in; |
2855 | if (follows > 0) { | |
2856 | in = mdcache->pick_inode_snap(head_in, follows); | |
2857 | // intermediate snap inodes | |
2858 | while (in != head_in) { | |
11fdf7f2 TL |
2859 | ceph_assert(in->last != CEPH_NOSNAP); |
2860 | if (in->is_auth() && dirty) { | |
b32b8144 | 2861 | dout(10) << " updating intermediate snapped inode " << *in << dendl; |
11fdf7f2 | 2862 | _do_cap_update(in, NULL, dirty, follows, m, MClientCaps::ref()); |
b32b8144 FG |
2863 | } |
2864 | in = mdcache->pick_inode_snap(head_in, in->last); | |
7c673cae | 2865 | } |
7c673cae FG |
2866 | } |
2867 | ||
2868 | // head inode, and cap | |
11fdf7f2 | 2869 | MClientCaps::ref ack; |
7c673cae FG |
2870 | |
2871 | int caps = m->get_caps(); | |
2872 | if (caps & ~cap->issued()) { | |
2873 | dout(10) << " confirming not issued caps " << ccap_string(caps & ~cap->issued()) << dendl; | |
2874 | caps &= cap->issued(); | |
2875 | } | |
2876 | ||
2877 | cap->confirm_receipt(m->get_seq(), caps); | |
2878 | dout(10) << " follows " << follows | |
2879 | << " retains " << ccap_string(m->get_caps()) | |
11fdf7f2 | 2880 | << " dirty " << ccap_string(dirty) |
7c673cae FG |
2881 | << " on " << *in << dendl; |
2882 | ||
2883 | ||
2884 | // missing/skipped snapflush? | |
2885 | // The client MAY send a snapflush if it is issued WR/EXCL caps, but | |
2886 | // presently only does so when it has actual dirty metadata. But, we | |
2887 | // set up the need_snapflush stuff based on the issued caps. | |
2888 | // We can infer that the client WONT send a FLUSHSNAP once they have | |
2889 | // released all WR/EXCL caps (the FLUSHSNAP always comes before the cap | |
2890 | // update/release). | |
2891 | if (!head_in->client_need_snapflush.empty()) { | |
11fdf7f2 TL |
2892 | if (!(cap->issued() & CEPH_CAP_ANY_FILE_WR) && |
2893 | !(m->flags & MClientCaps::FLAG_PENDING_CAPSNAP)) { | |
2894 | head_in->auth_pin(this); // prevent subtree frozen | |
2895 | need_unpin = true; | |
7c673cae FG |
2896 | _do_null_snapflush(head_in, client); |
2897 | } else { | |
2898 | dout(10) << " revocation in progress, not making any conclusions about null snapflushes" << dendl; | |
2899 | } | |
2900 | } | |
11fdf7f2 TL |
2901 | |
2902 | bool need_snapflush = cap->need_snapflush(); | |
2903 | if (dirty && in->is_auth()) { | |
2904 | dout(7) << " flush client." << client << " dirty " << ccap_string(dirty) | |
7c673cae | 2905 | << " seq " << m->get_seq() << " on " << *in << dendl; |
11fdf7f2 TL |
2906 | ack = MClientCaps::create(CEPH_CAP_OP_FLUSH_ACK, in->ino(), 0, cap->get_cap_id(), m->get_seq(), |
2907 | m->get_caps(), 0, dirty, 0, mds->get_osd_epoch_barrier()); | |
7c673cae FG |
2908 | ack->set_client_tid(m->get_client_tid()); |
2909 | ack->set_oldest_flush_tid(m->get_oldest_flush_tid()); | |
11fdf7f2 TL |
2910 | |
2911 | // client flushes and releases caps at the same time. make sure MDCache::cow_inode() | |
2912 | // properly setup CInode::client_need_snapflush | |
2913 | if ((dirty & ~cap->issued()) && !need_snapflush) | |
2914 | cap->mark_needsnapflush(); | |
7c673cae FG |
2915 | } |
2916 | ||
2917 | // filter wanted based on what we could ever give out (given auth/replica status) | |
11fdf7f2 | 2918 | bool need_flush = m->flags & MClientCaps::FLAG_SYNC; |
f64942e4 | 2919 | int new_wanted = m->get_wanted(); |
7c673cae | 2920 | if (new_wanted != cap->wanted()) { |
f64942e4 | 2921 | if (!need_flush && in->is_auth() && (new_wanted & ~cap->pending())) { |
7c673cae FG |
2922 | // exapnding caps. make sure we aren't waiting for a log flush |
2923 | need_flush = _need_flush_mdlog(head_in, new_wanted & ~cap->pending()); | |
2924 | } | |
2925 | ||
2926 | adjust_cap_wanted(cap, new_wanted, m->get_issue_seq()); | |
2927 | } | |
11fdf7f2 TL |
2928 | |
2929 | bool updated = in->is_auth() && | |
2930 | _do_cap_update(in, cap, dirty, follows, m, ack, &need_flush); | |
2931 | ||
2932 | if (cap->need_snapflush() && | |
2933 | (!need_snapflush || !(m->flags & MClientCaps::FLAG_PENDING_CAPSNAP))) | |
2934 | cap->clear_needsnapflush(); | |
2935 | ||
2936 | if (updated) { | |
7c673cae FG |
2937 | eval(in, CEPH_CAP_LOCKS); |
2938 | ||
2939 | if (!need_flush && (cap->wanted() & ~cap->pending())) | |
2940 | need_flush = _need_flush_mdlog(in, cap->wanted() & ~cap->pending()); | |
2941 | } else { | |
2942 | // no update, ack now. | |
2943 | if (ack) | |
2944 | mds->send_message_client_counted(ack, m->get_connection()); | |
2945 | ||
2946 | bool did_issue = eval(in, CEPH_CAP_LOCKS); | |
2947 | if (!did_issue && (cap->wanted() & ~cap->pending())) | |
2948 | issue_caps(in, cap); | |
2949 | ||
2950 | if (cap->get_last_seq() == 0 && | |
2951 | (cap->pending() & (CEPH_CAP_FILE_WR|CEPH_CAP_FILE_BUFFER))) { | |
7c673cae FG |
2952 | share_inode_max_size(in, cap); |
2953 | } | |
2954 | } | |
2955 | ||
2956 | if (need_flush) | |
2957 | mds->mdlog->flush(); | |
2958 | } | |
2959 | ||
2960 | out: | |
11fdf7f2 TL |
2961 | if (need_unpin) |
2962 | head_in->auth_unpin(this); | |
7c673cae FG |
2963 | } |
2964 | ||
2965 | ||
2966 | class C_Locker_RetryRequestCapRelease : public LockerContext { | |
2967 | client_t client; | |
2968 | ceph_mds_request_release item; | |
2969 | public: | |
2970 | C_Locker_RetryRequestCapRelease(Locker *l, client_t c, const ceph_mds_request_release& it) : | |
2971 | LockerContext(l), client(c), item(it) { } | |
2972 | void finish(int r) override { | |
2973 | string dname; | |
2974 | MDRequestRef null_ref; | |
2975 | locker->process_request_cap_release(null_ref, client, item, dname); | |
2976 | } | |
2977 | }; | |
2978 | ||
2979 | void Locker::process_request_cap_release(MDRequestRef& mdr, client_t client, const ceph_mds_request_release& item, | |
11fdf7f2 | 2980 | std::string_view dname) |
7c673cae FG |
2981 | { |
2982 | inodeno_t ino = (uint64_t)item.ino; | |
2983 | uint64_t cap_id = item.cap_id; | |
2984 | int caps = item.caps; | |
2985 | int wanted = item.wanted; | |
2986 | int seq = item.seq; | |
2987 | int issue_seq = item.issue_seq; | |
2988 | int mseq = item.mseq; | |
2989 | ||
2990 | CInode *in = mdcache->get_inode(ino); | |
2991 | if (!in) | |
2992 | return; | |
2993 | ||
2994 | if (dname.length()) { | |
2995 | frag_t fg = in->pick_dirfrag(dname); | |
2996 | CDir *dir = in->get_dirfrag(fg); | |
2997 | if (dir) { | |
2998 | CDentry *dn = dir->lookup(dname); | |
2999 | if (dn) { | |
3000 | ClientLease *l = dn->get_client_lease(client); | |
3001 | if (l) { | |
3002 | dout(10) << "process_cap_release removing lease on " << *dn << dendl; | |
3003 | dn->remove_client_lease(l, this); | |
3004 | } else { | |
3005 | dout(7) << "process_cap_release client." << client | |
3006 | << " doesn't have lease on " << *dn << dendl; | |
3007 | } | |
3008 | } else { | |
3009 | dout(7) << "process_cap_release client." << client << " released lease on dn " | |
3010 | << dir->dirfrag() << "/" << dname << " which dne" << dendl; | |
3011 | } | |
3012 | } | |
3013 | } | |
3014 | ||
3015 | Capability *cap = in->get_client_cap(client); | |
3016 | if (!cap) | |
3017 | return; | |
3018 | ||
3019 | dout(10) << "process_cap_release client." << client << " " << ccap_string(caps) << " on " << *in | |
3020 | << (mdr ? "" : " (DEFERRED, no mdr)") | |
3021 | << dendl; | |
3022 | ||
3023 | if (ceph_seq_cmp(mseq, cap->get_mseq()) < 0) { | |
3024 | dout(7) << " mseq " << mseq << " < " << cap->get_mseq() << ", dropping" << dendl; | |
3025 | return; | |
3026 | } | |
3027 | ||
3028 | if (cap->get_cap_id() != cap_id) { | |
3029 | dout(7) << " cap_id " << cap_id << " != " << cap->get_cap_id() << ", dropping" << dendl; | |
3030 | return; | |
3031 | } | |
3032 | ||
3033 | if (should_defer_client_cap_frozen(in)) { | |
3034 | dout(7) << " frozen, deferring" << dendl; | |
3035 | in->add_waiter(CInode::WAIT_UNFREEZE, new C_Locker_RetryRequestCapRelease(this, client, item)); | |
3036 | return; | |
3037 | } | |
3038 | ||
3039 | if (caps & ~cap->issued()) { | |
3040 | dout(10) << " confirming not issued caps " << ccap_string(caps & ~cap->issued()) << dendl; | |
3041 | caps &= cap->issued(); | |
3042 | } | |
3043 | cap->confirm_receipt(seq, caps); | |
3044 | ||
3045 | if (!in->client_need_snapflush.empty() && | |
3046 | (cap->issued() & CEPH_CAP_ANY_FILE_WR) == 0) { | |
3047 | _do_null_snapflush(in, client); | |
3048 | } | |
3049 | ||
3050 | adjust_cap_wanted(cap, wanted, issue_seq); | |
3051 | ||
3052 | if (mdr) | |
3053 | cap->inc_suppress(); | |
3054 | eval(in, CEPH_CAP_LOCKS); | |
3055 | if (mdr) | |
3056 | cap->dec_suppress(); | |
3057 | ||
3058 | // take note; we may need to reissue on this cap later | |
3059 | if (mdr) | |
3060 | mdr->cap_releases[in->vino()] = cap->get_last_seq(); | |
3061 | } | |
3062 | ||
3063 | class C_Locker_RetryKickIssueCaps : public LockerContext { | |
3064 | CInode *in; | |
3065 | client_t client; | |
3066 | ceph_seq_t seq; | |
3067 | public: | |
3068 | C_Locker_RetryKickIssueCaps(Locker *l, CInode *i, client_t c, ceph_seq_t s) : | |
3069 | LockerContext(l), in(i), client(c), seq(s) { | |
3070 | in->get(CInode::PIN_PTRWAITER); | |
3071 | } | |
3072 | void finish(int r) override { | |
3073 | locker->kick_issue_caps(in, client, seq); | |
3074 | in->put(CInode::PIN_PTRWAITER); | |
3075 | } | |
3076 | }; | |
3077 | ||
3078 | void Locker::kick_issue_caps(CInode *in, client_t client, ceph_seq_t seq) | |
3079 | { | |
3080 | Capability *cap = in->get_client_cap(client); | |
a8e16298 | 3081 | if (!cap || cap->get_last_seq() != seq) |
7c673cae FG |
3082 | return; |
3083 | if (in->is_frozen()) { | |
3084 | dout(10) << "kick_issue_caps waiting for unfreeze on " << *in << dendl; | |
3085 | in->add_waiter(CInode::WAIT_UNFREEZE, | |
3086 | new C_Locker_RetryKickIssueCaps(this, in, client, seq)); | |
3087 | return; | |
3088 | } | |
3089 | dout(10) << "kick_issue_caps released at current seq " << seq | |
3090 | << ", reissuing" << dendl; | |
3091 | issue_caps(in, cap); | |
3092 | } | |
3093 | ||
3094 | void Locker::kick_cap_releases(MDRequestRef& mdr) | |
3095 | { | |
3096 | client_t client = mdr->get_client(); | |
3097 | for (map<vinodeno_t,ceph_seq_t>::iterator p = mdr->cap_releases.begin(); | |
3098 | p != mdr->cap_releases.end(); | |
3099 | ++p) { | |
3100 | CInode *in = mdcache->get_inode(p->first); | |
3101 | if (!in) | |
3102 | continue; | |
3103 | kick_issue_caps(in, client, p->second); | |
3104 | } | |
3105 | } | |
3106 | ||
7c673cae FG |
3107 | /** |
3108 | * m and ack might be NULL, so don't dereference them unless dirty != 0 | |
3109 | */ | |
11fdf7f2 | 3110 | void Locker::_do_snap_update(CInode *in, snapid_t snap, int dirty, snapid_t follows, client_t client, const MClientCaps::const_ref &m, const MClientCaps::ref &ack) |
7c673cae FG |
3111 | { |
3112 | dout(10) << "_do_snap_update dirty " << ccap_string(dirty) | |
3113 | << " follows " << follows << " snap " << snap | |
3114 | << " on " << *in << dendl; | |
3115 | ||
3116 | if (snap == CEPH_NOSNAP) { | |
3117 | // hmm, i guess snap was already deleted? just ack! | |
3118 | dout(10) << " wow, the snap following " << follows | |
3119 | << " was already deleted. nothing to record, just ack." << dendl; | |
3120 | if (ack) | |
3121 | mds->send_message_client_counted(ack, m->get_connection()); | |
3122 | return; | |
3123 | } | |
3124 | ||
3125 | EUpdate *le = new EUpdate(mds->mdlog, "snap flush"); | |
3126 | mds->mdlog->start_entry(le); | |
3127 | MutationRef mut = new MutationImpl(); | |
3128 | mut->ls = mds->mdlog->get_current_segment(); | |
3129 | ||
3130 | // normal metadata updates that we can apply to the head as well. | |
3131 | ||
3132 | // update xattrs? | |
94b18763 FG |
3133 | CInode::mempool_xattr_map *px = nullptr; |
3134 | bool xattrs = (dirty & CEPH_CAP_XATTR_EXCL) && | |
3135 | m->xattrbl.length() && | |
3136 | m->head.xattr_version > in->get_projected_inode()->xattr_version; | |
3137 | ||
3138 | CInode::mempool_old_inode *oi = 0; | |
7c673cae FG |
3139 | if (in->is_multiversion()) { |
3140 | oi = in->pick_old_inode(snap); | |
3141 | } | |
3142 | ||
94b18763 | 3143 | CInode::mempool_inode *i; |
7c673cae FG |
3144 | if (oi) { |
3145 | dout(10) << " writing into old inode" << dendl; | |
94b18763 FG |
3146 | auto &pi = in->project_inode(); |
3147 | pi.inode.version = in->pre_dirty(); | |
7c673cae FG |
3148 | if (snap > oi->first) |
3149 | in->split_old_inode(snap); | |
94b18763 | 3150 | i = &oi->inode; |
7c673cae FG |
3151 | if (xattrs) |
3152 | px = &oi->xattrs; | |
3153 | } else { | |
94b18763 FG |
3154 | auto &pi = in->project_inode(xattrs); |
3155 | pi.inode.version = in->pre_dirty(); | |
3156 | i = &pi.inode; | |
7c673cae | 3157 | if (xattrs) |
94b18763 | 3158 | px = pi.xattrs.get(); |
7c673cae FG |
3159 | } |
3160 | ||
94b18763 | 3161 | _update_cap_fields(in, dirty, m, i); |
7c673cae FG |
3162 | |
3163 | // xattr | |
94b18763 FG |
3164 | if (xattrs) { |
3165 | dout(7) << " xattrs v" << i->xattr_version << " -> " << m->head.xattr_version | |
7c673cae | 3166 | << " len " << m->xattrbl.length() << dendl; |
94b18763 | 3167 | i->xattr_version = m->head.xattr_version; |
11fdf7f2 TL |
3168 | auto p = m->xattrbl.cbegin(); |
3169 | decode(*px, p); | |
7c673cae FG |
3170 | } |
3171 | ||
94b18763 FG |
3172 | { |
3173 | auto it = i->client_ranges.find(client); | |
3174 | if (it != i->client_ranges.end()) { | |
3175 | if (in->last == snap) { | |
3176 | dout(10) << " removing client_range entirely" << dendl; | |
3177 | i->client_ranges.erase(it); | |
3178 | } else { | |
3179 | dout(10) << " client_range now follows " << snap << dendl; | |
3180 | it->second.follows = snap; | |
3181 | } | |
7c673cae FG |
3182 | } |
3183 | } | |
3184 | ||
3185 | mut->auth_pin(in); | |
3186 | mdcache->predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY, 0, follows); | |
3187 | mdcache->journal_dirty_inode(mut.get(), &le->metablob, in, follows); | |
3188 | ||
3189 | // "oldest flush tid" > 0 means client uses unique TID for each flush | |
3190 | if (ack && ack->get_oldest_flush_tid() > 0) | |
3191 | le->metablob.add_client_flush(metareqid_t(m->get_source(), ack->get_client_tid()), | |
3192 | ack->get_oldest_flush_tid()); | |
3193 | ||
11fdf7f2 TL |
3194 | mds->mdlog->submit_entry(le, new C_Locker_FileUpdate_finish(this, in, mut, UPDATE_SNAPFLUSH, |
3195 | ack, client)); | |
7c673cae FG |
3196 | } |
3197 | ||
11fdf7f2 | 3198 | void Locker::_update_cap_fields(CInode *in, int dirty, const MClientCaps::const_ref &m, CInode::mempool_inode *pi) |
7c673cae FG |
3199 | { |
3200 | if (dirty == 0) | |
3201 | return; | |
3202 | ||
3203 | /* m must be valid if there are dirty caps */ | |
11fdf7f2 | 3204 | ceph_assert(m); |
7c673cae FG |
3205 | uint64_t features = m->get_connection()->get_features(); |
3206 | ||
3207 | if (m->get_ctime() > pi->ctime) { | |
3208 | dout(7) << " ctime " << pi->ctime << " -> " << m->get_ctime() | |
3209 | << " for " << *in << dendl; | |
91327a77 AA |
3210 | pi->ctime = m->get_ctime(); |
3211 | if (m->get_ctime() > pi->rstat.rctime) | |
3212 | pi->rstat.rctime = m->get_ctime(); | |
7c673cae FG |
3213 | } |
3214 | ||
3215 | if ((features & CEPH_FEATURE_FS_CHANGE_ATTR) && | |
3216 | m->get_change_attr() > pi->change_attr) { | |
3217 | dout(7) << " change_attr " << pi->change_attr << " -> " << m->get_change_attr() | |
3218 | << " for " << *in << dendl; | |
3219 | pi->change_attr = m->get_change_attr(); | |
3220 | } | |
3221 | ||
3222 | // file | |
3223 | if (dirty & (CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR)) { | |
3224 | utime_t atime = m->get_atime(); | |
3225 | utime_t mtime = m->get_mtime(); | |
3226 | uint64_t size = m->get_size(); | |
3227 | version_t inline_version = m->inline_version; | |
3228 | ||
3229 | if (((dirty & CEPH_CAP_FILE_WR) && mtime > pi->mtime) || | |
3230 | ((dirty & CEPH_CAP_FILE_EXCL) && mtime != pi->mtime)) { | |
3231 | dout(7) << " mtime " << pi->mtime << " -> " << mtime | |
3232 | << " for " << *in << dendl; | |
3233 | pi->mtime = mtime; | |
91327a77 AA |
3234 | if (mtime > pi->rstat.rctime) |
3235 | pi->rstat.rctime = mtime; | |
7c673cae FG |
3236 | } |
3237 | if (in->inode.is_file() && // ONLY if regular file | |
3238 | size > pi->size) { | |
3239 | dout(7) << " size " << pi->size << " -> " << size | |
3240 | << " for " << *in << dendl; | |
3241 | pi->size = size; | |
3242 | pi->rstat.rbytes = size; | |
3243 | } | |
3244 | if (in->inode.is_file() && | |
3245 | (dirty & CEPH_CAP_FILE_WR) && | |
3246 | inline_version > pi->inline_data.version) { | |
3247 | pi->inline_data.version = inline_version; | |
3248 | if (inline_version != CEPH_INLINE_NONE && m->inline_data.length() > 0) | |
3249 | pi->inline_data.get_data() = m->inline_data; | |
3250 | else | |
3251 | pi->inline_data.free_data(); | |
3252 | } | |
3253 | if ((dirty & CEPH_CAP_FILE_EXCL) && atime != pi->atime) { | |
3254 | dout(7) << " atime " << pi->atime << " -> " << atime | |
3255 | << " for " << *in << dendl; | |
3256 | pi->atime = atime; | |
3257 | } | |
3258 | if ((dirty & CEPH_CAP_FILE_EXCL) && | |
3259 | ceph_seq_cmp(pi->time_warp_seq, m->get_time_warp_seq()) < 0) { | |
3260 | dout(7) << " time_warp_seq " << pi->time_warp_seq << " -> " << m->get_time_warp_seq() | |
3261 | << " for " << *in << dendl; | |
3262 | pi->time_warp_seq = m->get_time_warp_seq(); | |
3263 | } | |
3264 | } | |
3265 | // auth | |
3266 | if (dirty & CEPH_CAP_AUTH_EXCL) { | |
3267 | if (m->head.uid != pi->uid) { | |
3268 | dout(7) << " uid " << pi->uid | |
3269 | << " -> " << m->head.uid | |
3270 | << " for " << *in << dendl; | |
3271 | pi->uid = m->head.uid; | |
3272 | } | |
3273 | if (m->head.gid != pi->gid) { | |
3274 | dout(7) << " gid " << pi->gid | |
3275 | << " -> " << m->head.gid | |
3276 | << " for " << *in << dendl; | |
3277 | pi->gid = m->head.gid; | |
3278 | } | |
3279 | if (m->head.mode != pi->mode) { | |
3280 | dout(7) << " mode " << oct << pi->mode | |
3281 | << " -> " << m->head.mode << dec | |
3282 | << " for " << *in << dendl; | |
3283 | pi->mode = m->head.mode; | |
3284 | } | |
3285 | if ((features & CEPH_FEATURE_FS_BTIME) && m->get_btime() != pi->btime) { | |
3286 | dout(7) << " btime " << oct << pi->btime | |
3287 | << " -> " << m->get_btime() << dec | |
3288 | << " for " << *in << dendl; | |
3289 | pi->btime = m->get_btime(); | |
3290 | } | |
3291 | } | |
3292 | } | |
3293 | ||
3294 | /* | |
3295 | * update inode based on cap flush|flushsnap|wanted. | |
3296 | * adjust max_size, if needed. | |
3297 | * if we update, return true; otherwise, false (no updated needed). | |
3298 | */ | |
3299 | bool Locker::_do_cap_update(CInode *in, Capability *cap, | |
3300 | int dirty, snapid_t follows, | |
11fdf7f2 | 3301 | const MClientCaps::const_ref &m, const MClientCaps::ref &ack, |
7c673cae FG |
3302 | bool *need_flush) |
3303 | { | |
3304 | dout(10) << "_do_cap_update dirty " << ccap_string(dirty) | |
3305 | << " issued " << ccap_string(cap ? cap->issued() : 0) | |
3306 | << " wanted " << ccap_string(cap ? cap->wanted() : 0) | |
3307 | << " on " << *in << dendl; | |
11fdf7f2 | 3308 | ceph_assert(in->is_auth()); |
7c673cae | 3309 | client_t client = m->get_source().num(); |
94b18763 | 3310 | CInode::mempool_inode *latest = in->get_projected_inode(); |
7c673cae FG |
3311 | |
3312 | // increase or zero max_size? | |
3313 | uint64_t size = m->get_size(); | |
3314 | bool change_max = false; | |
3315 | uint64_t old_max = latest->client_ranges.count(client) ? latest->client_ranges[client].range.last : 0; | |
3316 | uint64_t new_max = old_max; | |
3317 | ||
3318 | if (in->is_file()) { | |
3319 | bool forced_change_max = false; | |
3320 | dout(20) << "inode is file" << dendl; | |
3321 | if (cap && ((cap->issued() | cap->wanted()) & CEPH_CAP_ANY_FILE_WR)) { | |
3322 | dout(20) << "client has write caps; m->get_max_size=" | |
3323 | << m->get_max_size() << "; old_max=" << old_max << dendl; | |
3324 | if (m->get_max_size() > new_max) { | |
3325 | dout(10) << "client requests file_max " << m->get_max_size() | |
3326 | << " > max " << old_max << dendl; | |
3327 | change_max = true; | |
3328 | forced_change_max = true; | |
31f18b77 | 3329 | new_max = calc_new_max_size(latest, m->get_max_size()); |
7c673cae | 3330 | } else { |
31f18b77 | 3331 | new_max = calc_new_max_size(latest, size); |
7c673cae FG |
3332 | |
3333 | if (new_max > old_max) | |
3334 | change_max = true; | |
3335 | else | |
3336 | new_max = old_max; | |
3337 | } | |
3338 | } else { | |
3339 | if (old_max) { | |
3340 | change_max = true; | |
3341 | new_max = 0; | |
3342 | } | |
3343 | } | |
3344 | ||
3345 | if (in->last == CEPH_NOSNAP && | |
3346 | change_max && | |
3347 | !in->filelock.can_wrlock(client) && | |
3348 | !in->filelock.can_force_wrlock(client)) { | |
3349 | dout(10) << " i want to change file_max, but lock won't allow it (yet)" << dendl; | |
3350 | if (in->filelock.is_stable()) { | |
3351 | bool need_issue = false; | |
3352 | if (cap) | |
3353 | cap->inc_suppress(); | |
11fdf7f2 | 3354 | if (in->get_mds_caps_wanted().empty() && |
7c673cae FG |
3355 | (in->get_loner() >= 0 || (in->get_wanted_loner() >= 0 && in->try_set_loner()))) { |
3356 | if (in->filelock.get_state() != LOCK_EXCL) | |
3357 | file_excl(&in->filelock, &need_issue); | |
3358 | } else | |
3359 | simple_lock(&in->filelock, &need_issue); | |
3360 | if (need_issue) | |
3361 | issue_caps(in); | |
3362 | if (cap) | |
3363 | cap->dec_suppress(); | |
3364 | } | |
3365 | if (!in->filelock.can_wrlock(client) && | |
3366 | !in->filelock.can_force_wrlock(client)) { | |
3367 | C_MDL_CheckMaxSize *cms = new C_MDL_CheckMaxSize(this, in, | |
3368 | forced_change_max ? new_max : 0, | |
3369 | 0, utime_t()); | |
3370 | ||
3371 | in->filelock.add_waiter(SimpleLock::WAIT_STABLE, cms); | |
3372 | change_max = false; | |
3373 | } | |
3374 | } | |
3375 | } | |
3376 | ||
3377 | if (m->flockbl.length()) { | |
3378 | int32_t num_locks; | |
11fdf7f2 TL |
3379 | auto bli = m->flockbl.cbegin(); |
3380 | decode(num_locks, bli); | |
7c673cae FG |
3381 | for ( int i=0; i < num_locks; ++i) { |
3382 | ceph_filelock decoded_lock; | |
11fdf7f2 | 3383 | decode(decoded_lock, bli); |
7c673cae FG |
3384 | in->get_fcntl_lock_state()->held_locks. |
3385 | insert(pair<uint64_t, ceph_filelock>(decoded_lock.start, decoded_lock)); | |
3386 | ++in->get_fcntl_lock_state()->client_held_lock_counts[(client_t)(decoded_lock.client)]; | |
3387 | } | |
11fdf7f2 | 3388 | decode(num_locks, bli); |
7c673cae FG |
3389 | for ( int i=0; i < num_locks; ++i) { |
3390 | ceph_filelock decoded_lock; | |
11fdf7f2 | 3391 | decode(decoded_lock, bli); |
7c673cae FG |
3392 | in->get_flock_lock_state()->held_locks. |
3393 | insert(pair<uint64_t, ceph_filelock>(decoded_lock.start, decoded_lock)); | |
3394 | ++in->get_flock_lock_state()->client_held_lock_counts[(client_t)(decoded_lock.client)]; | |
3395 | } | |
3396 | } | |
3397 | ||
3398 | if (!dirty && !change_max) | |
3399 | return false; | |
3400 | ||
94b18763 | 3401 | Session *session = mds->get_session(m); |
7c673cae FG |
3402 | if (session->check_access(in, MAY_WRITE, |
3403 | m->caller_uid, m->caller_gid, NULL, 0, 0) < 0) { | |
7c673cae FG |
3404 | dout(10) << "check_access failed, dropping cap update on " << *in << dendl; |
3405 | return false; | |
3406 | } | |
7c673cae FG |
3407 | |
3408 | // do the update. | |
3409 | EUpdate *le = new EUpdate(mds->mdlog, "cap update"); | |
3410 | mds->mdlog->start_entry(le); | |
3411 | ||
94b18763 FG |
3412 | bool xattr = (dirty & CEPH_CAP_XATTR_EXCL) && |
3413 | m->xattrbl.length() && | |
3414 | m->head.xattr_version > in->get_projected_inode()->xattr_version; | |
7c673cae | 3415 | |
94b18763 FG |
3416 | auto &pi = in->project_inode(xattr); |
3417 | pi.inode.version = in->pre_dirty(); | |
7c673cae FG |
3418 | |
3419 | MutationRef mut(new MutationImpl()); | |
3420 | mut->ls = mds->mdlog->get_current_segment(); | |
3421 | ||
94b18763 | 3422 | _update_cap_fields(in, dirty, m, &pi.inode); |
7c673cae FG |
3423 | |
3424 | if (change_max) { | |
3425 | dout(7) << " max_size " << old_max << " -> " << new_max | |
3426 | << " for " << *in << dendl; | |
3427 | if (new_max) { | |
94b18763 FG |
3428 | auto &cr = pi.inode.client_ranges[client]; |
3429 | cr.range.first = 0; | |
3430 | cr.range.last = new_max; | |
3431 | cr.follows = in->first - 1; | |
a8e16298 TL |
3432 | if (cap) |
3433 | cap->mark_clientwriteable(); | |
3434 | } else { | |
94b18763 | 3435 | pi.inode.client_ranges.erase(client); |
a8e16298 TL |
3436 | if (cap) |
3437 | cap->clear_clientwriteable(); | |
3438 | } | |
7c673cae FG |
3439 | } |
3440 | ||
3441 | if (change_max || (dirty & (CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR))) | |
3442 | wrlock_force(&in->filelock, mut); // wrlock for duration of journal | |
3443 | ||
3444 | // auth | |
3445 | if (dirty & CEPH_CAP_AUTH_EXCL) | |
3446 | wrlock_force(&in->authlock, mut); | |
3447 | ||
94b18763 FG |
3448 | // xattrs update? |
3449 | if (xattr) { | |
3450 | dout(7) << " xattrs v" << pi.inode.xattr_version << " -> " << m->head.xattr_version << dendl; | |
3451 | pi.inode.xattr_version = m->head.xattr_version; | |
11fdf7f2 TL |
3452 | auto p = m->xattrbl.cbegin(); |
3453 | decode(*pi.xattrs, p); | |
7c673cae FG |
3454 | wrlock_force(&in->xattrlock, mut); |
3455 | } | |
3456 | ||
3457 | mut->auth_pin(in); | |
3458 | mdcache->predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY, 0, follows); | |
3459 | mdcache->journal_dirty_inode(mut.get(), &le->metablob, in, follows); | |
3460 | ||
3461 | // "oldest flush tid" > 0 means client uses unique TID for each flush | |
3462 | if (ack && ack->get_oldest_flush_tid() > 0) | |
3463 | le->metablob.add_client_flush(metareqid_t(m->get_source(), ack->get_client_tid()), | |
3464 | ack->get_oldest_flush_tid()); | |
3465 | ||
11fdf7f2 TL |
3466 | unsigned update_flags = 0; |
3467 | if (change_max) | |
3468 | update_flags |= UPDATE_SHAREMAX; | |
3469 | if (cap) | |
3470 | update_flags |= UPDATE_NEEDSISSUE; | |
3471 | mds->mdlog->submit_entry(le, new C_Locker_FileUpdate_finish(this, in, mut, update_flags, | |
3472 | ack, client)); | |
7c673cae FG |
3473 | if (need_flush && !*need_flush && |
3474 | ((change_max && new_max) || // max INCREASE | |
3475 | _need_flush_mdlog(in, dirty))) | |
3476 | *need_flush = true; | |
3477 | ||
3478 | return true; | |
3479 | } | |
3480 | ||
11fdf7f2 | 3481 | void Locker::handle_client_cap_release(const MClientCapRelease::const_ref &m) |
7c673cae FG |
3482 | { |
3483 | client_t client = m->get_source().num(); | |
3484 | dout(10) << "handle_client_cap_release " << *m << dendl; | |
3485 | ||
3486 | if (!mds->is_clientreplay() && !mds->is_active() && !mds->is_stopping()) { | |
3487 | mds->wait_for_replay(new C_MDS_RetryMessage(mds, m)); | |
3488 | return; | |
3489 | } | |
3490 | ||
3491 | if (m->osd_epoch_barrier && !mds->objecter->have_map(m->osd_epoch_barrier)) { | |
3492 | // Pause RADOS operations until we see the required epoch | |
3493 | mds->objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
3494 | } | |
3495 | ||
3496 | if (mds->get_osd_epoch_barrier() < m->osd_epoch_barrier) { | |
3497 | // Record the barrier so that we will retransmit it to clients | |
3498 | mds->set_osd_epoch_barrier(m->osd_epoch_barrier); | |
3499 | } | |
3500 | ||
94b18763 | 3501 | Session *session = mds->get_session(m); |
7c673cae | 3502 | |
11fdf7f2 TL |
3503 | for (const auto &cap : m->caps) { |
3504 | _do_cap_release(client, inodeno_t((uint64_t)cap.ino) , cap.cap_id, cap.migrate_seq, cap.seq); | |
7c673cae FG |
3505 | } |
3506 | ||
3507 | if (session) { | |
3508 | session->notify_cap_release(m->caps.size()); | |
3509 | } | |
7c673cae FG |
3510 | } |
3511 | ||
3512 | class C_Locker_RetryCapRelease : public LockerContext { | |
3513 | client_t client; | |
3514 | inodeno_t ino; | |
3515 | uint64_t cap_id; | |
3516 | ceph_seq_t migrate_seq; | |
3517 | ceph_seq_t issue_seq; | |
3518 | public: | |
3519 | C_Locker_RetryCapRelease(Locker *l, client_t c, inodeno_t i, uint64_t id, | |
3520 | ceph_seq_t mseq, ceph_seq_t seq) : | |
3521 | LockerContext(l), client(c), ino(i), cap_id(id), migrate_seq(mseq), issue_seq(seq) {} | |
3522 | void finish(int r) override { | |
3523 | locker->_do_cap_release(client, ino, cap_id, migrate_seq, issue_seq); | |
3524 | } | |
3525 | }; | |
3526 | ||
3527 | void Locker::_do_cap_release(client_t client, inodeno_t ino, uint64_t cap_id, | |
3528 | ceph_seq_t mseq, ceph_seq_t seq) | |
3529 | { | |
3530 | CInode *in = mdcache->get_inode(ino); | |
3531 | if (!in) { | |
3532 | dout(7) << "_do_cap_release missing ino " << ino << dendl; | |
3533 | return; | |
3534 | } | |
3535 | Capability *cap = in->get_client_cap(client); | |
3536 | if (!cap) { | |
3537 | dout(7) << "_do_cap_release no cap for client" << client << " on "<< *in << dendl; | |
3538 | return; | |
3539 | } | |
3540 | ||
3541 | dout(7) << "_do_cap_release for client." << client << " on "<< *in << dendl; | |
3542 | if (cap->get_cap_id() != cap_id) { | |
3543 | dout(7) << " capid " << cap_id << " != " << cap->get_cap_id() << ", ignore" << dendl; | |
3544 | return; | |
3545 | } | |
3546 | if (ceph_seq_cmp(mseq, cap->get_mseq()) < 0) { | |
3547 | dout(7) << " mseq " << mseq << " < " << cap->get_mseq() << ", ignore" << dendl; | |
3548 | return; | |
3549 | } | |
3550 | if (should_defer_client_cap_frozen(in)) { | |
3551 | dout(7) << " freezing|frozen, deferring" << dendl; | |
3552 | in->add_waiter(CInode::WAIT_UNFREEZE, | |
3553 | new C_Locker_RetryCapRelease(this, client, ino, cap_id, mseq, seq)); | |
3554 | return; | |
3555 | } | |
3556 | if (seq != cap->get_last_issue()) { | |
3557 | dout(7) << " issue_seq " << seq << " != " << cap->get_last_issue() << dendl; | |
3558 | // clean out any old revoke history | |
3559 | cap->clean_revoke_from(seq); | |
3560 | eval_cap_gather(in); | |
3561 | return; | |
3562 | } | |
a8e16298 | 3563 | remove_client_cap(in, cap); |
7c673cae FG |
3564 | } |
3565 | ||
a8e16298 | 3566 | void Locker::remove_client_cap(CInode *in, Capability *cap) |
7c673cae | 3567 | { |
a8e16298 | 3568 | client_t client = cap->get_client(); |
7c673cae FG |
3569 | // clean out any pending snapflush state |
3570 | if (!in->client_need_snapflush.empty()) | |
3571 | _do_null_snapflush(in, client); | |
3572 | ||
a8e16298 | 3573 | bool notable = cap->is_notable(); |
7c673cae | 3574 | in->remove_client_cap(client); |
a8e16298 TL |
3575 | if (!notable) |
3576 | return; | |
7c673cae FG |
3577 | |
3578 | if (in->is_auth()) { | |
3579 | // make sure we clear out the client byte range | |
3580 | if (in->get_projected_inode()->client_ranges.count(client) && | |
3581 | !(in->inode.nlink == 0 && !in->is_any_caps())) // unless it's unlink + stray | |
3582 | check_inode_max_size(in); | |
3583 | } else { | |
3584 | request_inode_file_caps(in); | |
3585 | } | |
3586 | ||
3587 | try_eval(in, CEPH_CAP_LOCKS); | |
3588 | } | |
3589 | ||
3590 | ||
3591 | /** | |
3592 | * Return true if any currently revoking caps exceed the | |
f64942e4 | 3593 | * session_timeout threshold. |
7c673cae | 3594 | */ |
91327a77 AA |
3595 | bool Locker::any_late_revoking_caps(xlist<Capability*> const &revoking, |
3596 | double timeout) const | |
7c673cae FG |
3597 | { |
3598 | xlist<Capability*>::const_iterator p = revoking.begin(); | |
3599 | if (p.end()) { | |
3600 | // No revoking caps at the moment | |
3601 | return false; | |
3602 | } else { | |
3603 | utime_t now = ceph_clock_now(); | |
3604 | utime_t age = now - (*p)->get_last_revoke_stamp(); | |
91327a77 | 3605 | if (age <= timeout) { |
7c673cae FG |
3606 | return false; |
3607 | } else { | |
3608 | return true; | |
3609 | } | |
3610 | } | |
3611 | } | |
3612 | ||
91327a77 AA |
3613 | void Locker::get_late_revoking_clients(std::list<client_t> *result, |
3614 | double timeout) const | |
7c673cae | 3615 | { |
91327a77 | 3616 | if (!any_late_revoking_caps(revoking_caps, timeout)) { |
7c673cae FG |
3617 | // Fast path: no misbehaving clients, execute in O(1) |
3618 | return; | |
3619 | } | |
3620 | ||
3621 | // Slow path: execute in O(N_clients) | |
91327a77 AA |
3622 | for (auto &p : revoking_caps_by_client) { |
3623 | if (any_late_revoking_caps(p.second, timeout)) { | |
3624 | result->push_back(p.first); | |
7c673cae FG |
3625 | } |
3626 | } | |
3627 | } | |
3628 | ||
3629 | // Hard-code instead of surfacing a config settings because this is | |
3630 | // really a hack that should go away at some point when we have better | |
3631 | // inspection tools for getting at detailed cap state (#7316) | |
3632 | #define MAX_WARN_CAPS 100 | |
3633 | ||
3634 | void Locker::caps_tick() | |
3635 | { | |
3636 | utime_t now = ceph_clock_now(); | |
3637 | ||
11fdf7f2 TL |
3638 | if (!need_snapflush_inodes.empty()) { |
3639 | // snap inodes that needs flush are auth pinned, they affect | |
3640 | // subtree/difrarg freeze. | |
3641 | utime_t cutoff = now; | |
3642 | cutoff -= g_conf()->mds_freeze_tree_timeout / 3; | |
3643 | ||
3644 | CInode *last = need_snapflush_inodes.back(); | |
3645 | while (!need_snapflush_inodes.empty()) { | |
3646 | CInode *in = need_snapflush_inodes.front(); | |
3647 | if (in->last_dirstat_prop >= cutoff) | |
3648 | break; | |
3649 | in->item_caps.remove_myself(); | |
3650 | snapflush_nudge(in); | |
3651 | if (in == last) | |
3652 | break; | |
3653 | } | |
3654 | } | |
3655 | ||
7c673cae FG |
3656 | dout(20) << __func__ << " " << revoking_caps.size() << " revoking caps" << dendl; |
3657 | ||
11fdf7f2 TL |
3658 | now = ceph_clock_now(); |
3659 | int n = 0; | |
7c673cae FG |
3660 | for (xlist<Capability*>::iterator p = revoking_caps.begin(); !p.end(); ++p) { |
3661 | Capability *cap = *p; | |
3662 | ||
3663 | utime_t age = now - cap->get_last_revoke_stamp(); | |
11fdf7f2 | 3664 | dout(20) << __func__ << " age = " << age << " client." << cap->get_client() << "." << cap->get_inode()->ino() << dendl; |
f64942e4 AA |
3665 | if (age <= mds->mdsmap->get_session_timeout()) { |
3666 | dout(20) << __func__ << " age below timeout " << mds->mdsmap->get_session_timeout() << dendl; | |
7c673cae FG |
3667 | break; |
3668 | } else { | |
11fdf7f2 TL |
3669 | ++n; |
3670 | if (n > MAX_WARN_CAPS) { | |
7c673cae FG |
3671 | dout(1) << __func__ << " more than " << MAX_WARN_CAPS << " caps are late" |
3672 | << "revoking, ignoring subsequent caps" << dendl; | |
3673 | break; | |
3674 | } | |
3675 | } | |
3676 | // exponential backoff of warning intervals | |
f64942e4 | 3677 | if (age > mds->mdsmap->get_session_timeout() * (1 << cap->get_num_revoke_warnings())) { |
7c673cae FG |
3678 | cap->inc_num_revoke_warnings(); |
3679 | stringstream ss; | |
3680 | ss << "client." << cap->get_client() << " isn't responding to mclientcaps(revoke), ino " | |
3681 | << cap->get_inode()->ino() << " pending " << ccap_string(cap->pending()) | |
3682 | << " issued " << ccap_string(cap->issued()) << ", sent " << age << " seconds ago"; | |
3683 | mds->clog->warn() << ss.str(); | |
3684 | dout(20) << __func__ << " " << ss.str() << dendl; | |
3685 | } else { | |
11fdf7f2 | 3686 | dout(20) << __func__ << " silencing log message (backoff) for " << "client." << cap->get_client() << "." << cap->get_inode()->ino() << dendl; |
7c673cae FG |
3687 | } |
3688 | } | |
3689 | } | |
3690 | ||
3691 | ||
11fdf7f2 | 3692 | void Locker::handle_client_lease(const MClientLease::const_ref &m) |
7c673cae FG |
3693 | { |
3694 | dout(10) << "handle_client_lease " << *m << dendl; | |
3695 | ||
11fdf7f2 | 3696 | ceph_assert(m->get_source().is_client()); |
7c673cae FG |
3697 | client_t client = m->get_source().num(); |
3698 | ||
3699 | CInode *in = mdcache->get_inode(m->get_ino(), m->get_last()); | |
3700 | if (!in) { | |
3701 | dout(7) << "handle_client_lease don't have ino " << m->get_ino() << "." << m->get_last() << dendl; | |
7c673cae FG |
3702 | return; |
3703 | } | |
3704 | CDentry *dn = 0; | |
3705 | ||
3706 | frag_t fg = in->pick_dirfrag(m->dname); | |
3707 | CDir *dir = in->get_dirfrag(fg); | |
3708 | if (dir) | |
3709 | dn = dir->lookup(m->dname); | |
3710 | if (!dn) { | |
3711 | dout(7) << "handle_client_lease don't have dn " << m->get_ino() << " " << m->dname << dendl; | |
7c673cae FG |
3712 | return; |
3713 | } | |
3714 | dout(10) << " on " << *dn << dendl; | |
3715 | ||
3716 | // replica and lock | |
3717 | ClientLease *l = dn->get_client_lease(client); | |
3718 | if (!l) { | |
3719 | dout(7) << "handle_client_lease didn't have lease for client." << client << " of " << *dn << dendl; | |
7c673cae FG |
3720 | return; |
3721 | } | |
3722 | ||
3723 | switch (m->get_action()) { | |
3724 | case CEPH_MDS_LEASE_REVOKE_ACK: | |
3725 | case CEPH_MDS_LEASE_RELEASE: | |
3726 | if (l->seq != m->get_seq()) { | |
3727 | dout(7) << "handle_client_lease release - seq " << l->seq << " != provided " << m->get_seq() << dendl; | |
3728 | } else { | |
3729 | dout(7) << "handle_client_lease client." << client | |
3730 | << " on " << *dn << dendl; | |
3731 | dn->remove_client_lease(l, this); | |
3732 | } | |
7c673cae FG |
3733 | break; |
3734 | ||
3735 | case CEPH_MDS_LEASE_RENEW: | |
3736 | { | |
3737 | dout(7) << "handle_client_lease client." << client << " renew on " << *dn | |
3738 | << (!dn->lock.can_lease(client)?", revoking lease":"") << dendl; | |
3739 | if (dn->lock.can_lease(client)) { | |
11fdf7f2 | 3740 | auto reply = MClientLease::create(*m); |
7c673cae | 3741 | int pool = 1; // fixme.. do something smart! |
11fdf7f2 TL |
3742 | reply->h.duration_ms = (int)(1000 * mdcache->client_lease_durations[pool]); |
3743 | reply->h.seq = ++l->seq; | |
3744 | reply->clear_payload(); | |
7c673cae FG |
3745 | |
3746 | utime_t now = ceph_clock_now(); | |
3747 | now += mdcache->client_lease_durations[pool]; | |
3748 | mdcache->touch_client_lease(l, pool, now); | |
3749 | ||
11fdf7f2 | 3750 | mds->send_message_client_counted(reply, m->get_connection()); |
7c673cae FG |
3751 | } |
3752 | } | |
3753 | break; | |
3754 | ||
3755 | default: | |
3756 | ceph_abort(); // implement me | |
3757 | break; | |
3758 | } | |
3759 | } | |
3760 | ||
3761 | ||
3762 | void Locker::issue_client_lease(CDentry *dn, client_t client, | |
3763 | bufferlist &bl, utime_t now, Session *session) | |
3764 | { | |
3765 | CInode *diri = dn->get_dir()->get_inode(); | |
3766 | if (!diri->is_stray() && // do not issue dn leases in stray dir! | |
3767 | ((!diri->filelock.can_lease(client) && | |
3768 | (diri->get_client_cap_pending(client) & (CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_EXCL)) == 0)) && | |
3769 | dn->lock.can_lease(client)) { | |
3770 | int pool = 1; // fixme.. do something smart! | |
3771 | // issue a dentry lease | |
3772 | ClientLease *l = dn->add_client_lease(client, session); | |
3773 | session->touch_lease(l); | |
3774 | ||
3775 | now += mdcache->client_lease_durations[pool]; | |
3776 | mdcache->touch_client_lease(l, pool, now); | |
3777 | ||
11fdf7f2 TL |
3778 | LeaseStat lstat; |
3779 | lstat.mask = 1 | CEPH_LOCK_DN; // old and new bit values | |
3780 | lstat.duration_ms = (uint32_t)(1000 * mdcache->client_lease_durations[pool]); | |
3781 | lstat.seq = ++l->seq; | |
3782 | encode_lease(bl, session->info, lstat); | |
3783 | dout(20) << "issue_client_lease seq " << lstat.seq << " dur " << lstat.duration_ms << "ms " | |
7c673cae FG |
3784 | << " on " << *dn << dendl; |
3785 | } else { | |
3786 | // null lease | |
11fdf7f2 TL |
3787 | LeaseStat lstat; |
3788 | encode_lease(bl, session->info, lstat); | |
7c673cae FG |
3789 | dout(20) << "issue_client_lease no/null lease on " << *dn << dendl; |
3790 | } | |
3791 | } | |
3792 | ||
3793 | ||
3794 | void Locker::revoke_client_leases(SimpleLock *lock) | |
3795 | { | |
3796 | int n = 0; | |
3797 | CDentry *dn = static_cast<CDentry*>(lock->get_parent()); | |
3798 | for (map<client_t, ClientLease*>::iterator p = dn->client_lease_map.begin(); | |
3799 | p != dn->client_lease_map.end(); | |
3800 | ++p) { | |
3801 | ClientLease *l = p->second; | |
3802 | ||
3803 | n++; | |
11fdf7f2 | 3804 | ceph_assert(lock->get_type() == CEPH_LOCK_DN); |
7c673cae FG |
3805 | |
3806 | CDentry *dn = static_cast<CDentry*>(lock->get_parent()); | |
3807 | int mask = 1 | CEPH_LOCK_DN; // old and new bits | |
3808 | ||
3809 | // i should also revoke the dir ICONTENT lease, if they have it! | |
3810 | CInode *diri = dn->get_dir()->get_inode(); | |
11fdf7f2 TL |
3811 | auto lease = MClientLease::create(CEPH_MDS_LEASE_REVOKE, l->seq, mask, diri->ino(), diri->first, CEPH_NOSNAP, dn->get_name()); |
3812 | mds->send_message_client_counted(lease, l->client); | |
7c673cae | 3813 | } |
7c673cae FG |
3814 | } |
3815 | ||
11fdf7f2 TL |
3816 | void Locker::encode_lease(bufferlist& bl, const session_info_t& info, |
3817 | const LeaseStat& ls) | |
3818 | { | |
3819 | if (info.has_feature(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
3820 | ENCODE_START(1, 1, bl); | |
3821 | encode(ls.mask, bl); | |
3822 | encode(ls.duration_ms, bl); | |
3823 | encode(ls.seq, bl); | |
3824 | ENCODE_FINISH(bl); | |
3825 | } | |
3826 | else { | |
3827 | encode(ls.mask, bl); | |
3828 | encode(ls.duration_ms, bl); | |
3829 | encode(ls.seq, bl); | |
3830 | } | |
3831 | } | |
7c673cae FG |
3832 | |
3833 | // locks ---------------------------------------------------------------- | |
3834 | ||
11fdf7f2 | 3835 | SimpleLock *Locker::get_lock(int lock_type, const MDSCacheObjectInfo &info) |
7c673cae FG |
3836 | { |
3837 | switch (lock_type) { | |
3838 | case CEPH_LOCK_DN: | |
3839 | { | |
3840 | // be careful; info.dirfrag may have incorrect frag; recalculate based on dname. | |
3841 | CInode *diri = mdcache->get_inode(info.dirfrag.ino); | |
3842 | frag_t fg; | |
3843 | CDir *dir = 0; | |
3844 | CDentry *dn = 0; | |
3845 | if (diri) { | |
3846 | fg = diri->pick_dirfrag(info.dname); | |
3847 | dir = diri->get_dirfrag(fg); | |
3848 | if (dir) | |
3849 | dn = dir->lookup(info.dname, info.snapid); | |
3850 | } | |
3851 | if (!dn) { | |
3852 | dout(7) << "get_lock don't have dn " << info.dirfrag.ino << " " << info.dname << dendl; | |
3853 | return 0; | |
3854 | } | |
3855 | return &dn->lock; | |
3856 | } | |
3857 | ||
3858 | case CEPH_LOCK_IAUTH: | |
3859 | case CEPH_LOCK_ILINK: | |
3860 | case CEPH_LOCK_IDFT: | |
3861 | case CEPH_LOCK_IFILE: | |
3862 | case CEPH_LOCK_INEST: | |
3863 | case CEPH_LOCK_IXATTR: | |
3864 | case CEPH_LOCK_ISNAP: | |
3865 | case CEPH_LOCK_IFLOCK: | |
3866 | case CEPH_LOCK_IPOLICY: | |
3867 | { | |
3868 | CInode *in = mdcache->get_inode(info.ino, info.snapid); | |
3869 | if (!in) { | |
3870 | dout(7) << "get_lock don't have ino " << info.ino << dendl; | |
3871 | return 0; | |
3872 | } | |
3873 | switch (lock_type) { | |
3874 | case CEPH_LOCK_IAUTH: return &in->authlock; | |
3875 | case CEPH_LOCK_ILINK: return &in->linklock; | |
3876 | case CEPH_LOCK_IDFT: return &in->dirfragtreelock; | |
3877 | case CEPH_LOCK_IFILE: return &in->filelock; | |
3878 | case CEPH_LOCK_INEST: return &in->nestlock; | |
3879 | case CEPH_LOCK_IXATTR: return &in->xattrlock; | |
3880 | case CEPH_LOCK_ISNAP: return &in->snaplock; | |
3881 | case CEPH_LOCK_IFLOCK: return &in->flocklock; | |
3882 | case CEPH_LOCK_IPOLICY: return &in->policylock; | |
3883 | } | |
3884 | } | |
3885 | ||
3886 | default: | |
3887 | dout(7) << "get_lock don't know lock_type " << lock_type << dendl; | |
3888 | ceph_abort(); | |
3889 | break; | |
3890 | } | |
3891 | ||
3892 | return 0; | |
3893 | } | |
3894 | ||
11fdf7f2 | 3895 | void Locker::handle_lock(const MLock::const_ref &m) |
7c673cae FG |
3896 | { |
3897 | // nobody should be talking to us during recovery. | |
11fdf7f2 | 3898 | ceph_assert(mds->is_rejoin() || mds->is_clientreplay() || mds->is_active() || mds->is_stopping()); |
7c673cae FG |
3899 | |
3900 | SimpleLock *lock = get_lock(m->get_lock_type(), m->get_object_info()); | |
3901 | if (!lock) { | |
3902 | dout(10) << "don't have object " << m->get_object_info() << ", must have trimmed, dropping" << dendl; | |
7c673cae FG |
3903 | return; |
3904 | } | |
3905 | ||
3906 | switch (lock->get_type()) { | |
3907 | case CEPH_LOCK_DN: | |
3908 | case CEPH_LOCK_IAUTH: | |
3909 | case CEPH_LOCK_ILINK: | |
3910 | case CEPH_LOCK_ISNAP: | |
3911 | case CEPH_LOCK_IXATTR: | |
3912 | case CEPH_LOCK_IFLOCK: | |
3913 | case CEPH_LOCK_IPOLICY: | |
3914 | handle_simple_lock(lock, m); | |
3915 | break; | |
3916 | ||
3917 | case CEPH_LOCK_IDFT: | |
3918 | case CEPH_LOCK_INEST: | |
3919 | //handle_scatter_lock((ScatterLock*)lock, m); | |
3920 | //break; | |
3921 | ||
3922 | case CEPH_LOCK_IFILE: | |
3923 | handle_file_lock(static_cast<ScatterLock*>(lock), m); | |
3924 | break; | |
3925 | ||
3926 | default: | |
3927 | dout(7) << "handle_lock got otype " << m->get_lock_type() << dendl; | |
3928 | ceph_abort(); | |
3929 | break; | |
3930 | } | |
3931 | } | |
3932 | ||
3933 | ||
3934 | ||
3935 | ||
3936 | ||
3937 | // ========================================================================== | |
3938 | // simple lock | |
3939 | ||
3940 | /** This function may take a reference to m if it needs one, but does | |
3941 | * not put references. */ | |
11fdf7f2 | 3942 | void Locker::handle_reqrdlock(SimpleLock *lock, const MLock::const_ref &m) |
7c673cae FG |
3943 | { |
3944 | MDSCacheObject *parent = lock->get_parent(); | |
3945 | if (parent->is_auth() && | |
3946 | lock->get_state() != LOCK_SYNC && | |
3947 | !parent->is_frozen()) { | |
3948 | dout(7) << "handle_reqrdlock got rdlock request on " << *lock | |
3949 | << " on " << *parent << dendl; | |
11fdf7f2 | 3950 | ceph_assert(parent->is_auth()); // replica auth pinned if they're doing this! |
7c673cae FG |
3951 | if (lock->is_stable()) { |
3952 | simple_sync(lock); | |
3953 | } else { | |
3954 | dout(7) << "handle_reqrdlock delaying request until lock is stable" << dendl; | |
3955 | lock->add_waiter(SimpleLock::WAIT_STABLE | MDSCacheObject::WAIT_UNFREEZE, | |
11fdf7f2 | 3956 | new C_MDS_RetryMessage(mds, m)); |
7c673cae FG |
3957 | } |
3958 | } else { | |
3959 | dout(7) << "handle_reqrdlock dropping rdlock request on " << *lock | |
3960 | << " on " << *parent << dendl; | |
3961 | // replica should retry | |
3962 | } | |
3963 | } | |
3964 | ||
11fdf7f2 | 3965 | void Locker::handle_simple_lock(SimpleLock *lock, const MLock::const_ref &m) |
7c673cae FG |
3966 | { |
3967 | int from = m->get_asker(); | |
3968 | ||
3969 | dout(10) << "handle_simple_lock " << *m | |
3970 | << " on " << *lock << " " << *lock->get_parent() << dendl; | |
3971 | ||
3972 | if (mds->is_rejoin()) { | |
3973 | if (lock->get_parent()->is_rejoining()) { | |
3974 | dout(7) << "handle_simple_lock still rejoining " << *lock->get_parent() | |
3975 | << ", dropping " << *m << dendl; | |
7c673cae FG |
3976 | return; |
3977 | } | |
3978 | } | |
3979 | ||
3980 | switch (m->get_action()) { | |
3981 | // -- replica -- | |
3982 | case LOCK_AC_SYNC: | |
11fdf7f2 | 3983 | ceph_assert(lock->get_state() == LOCK_LOCK); |
7c673cae FG |
3984 | lock->decode_locked_state(m->get_data()); |
3985 | lock->set_state(LOCK_SYNC); | |
3986 | lock->finish_waiters(SimpleLock::WAIT_RD|SimpleLock::WAIT_STABLE); | |
3987 | break; | |
3988 | ||
3989 | case LOCK_AC_LOCK: | |
11fdf7f2 | 3990 | ceph_assert(lock->get_state() == LOCK_SYNC); |
7c673cae FG |
3991 | lock->set_state(LOCK_SYNC_LOCK); |
3992 | if (lock->is_leased()) | |
3993 | revoke_client_leases(lock); | |
3994 | eval_gather(lock, true); | |
31f18b77 FG |
3995 | if (lock->is_unstable_and_locked()) |
3996 | mds->mdlog->flush(); | |
7c673cae FG |
3997 | break; |
3998 | ||
3999 | ||
4000 | // -- auth -- | |
4001 | case LOCK_AC_LOCKACK: | |
11fdf7f2 | 4002 | ceph_assert(lock->get_state() == LOCK_SYNC_LOCK || |
7c673cae | 4003 | lock->get_state() == LOCK_SYNC_EXCL); |
11fdf7f2 | 4004 | ceph_assert(lock->is_gathering(from)); |
7c673cae FG |
4005 | lock->remove_gather(from); |
4006 | ||
4007 | if (lock->is_gathering()) { | |
4008 | dout(7) << "handle_simple_lock " << *lock << " on " << *lock->get_parent() << " from " << from | |
4009 | << ", still gathering " << lock->get_gather_set() << dendl; | |
4010 | } else { | |
4011 | dout(7) << "handle_simple_lock " << *lock << " on " << *lock->get_parent() << " from " << from | |
4012 | << ", last one" << dendl; | |
4013 | eval_gather(lock); | |
4014 | } | |
4015 | break; | |
4016 | ||
4017 | case LOCK_AC_REQRDLOCK: | |
4018 | handle_reqrdlock(lock, m); | |
4019 | break; | |
4020 | ||
4021 | } | |
7c673cae FG |
4022 | } |
4023 | ||
4024 | /* unused, currently. | |
4025 | ||
4026 | class C_Locker_SimpleEval : public Context { | |
4027 | Locker *locker; | |
4028 | SimpleLock *lock; | |
4029 | public: | |
4030 | C_Locker_SimpleEval(Locker *l, SimpleLock *lk) : locker(l), lock(lk) {} | |
4031 | void finish(int r) { | |
4032 | locker->try_simple_eval(lock); | |
4033 | } | |
4034 | }; | |
4035 | ||
4036 | void Locker::try_simple_eval(SimpleLock *lock) | |
4037 | { | |
4038 | // unstable and ambiguous auth? | |
4039 | if (!lock->is_stable() && | |
4040 | lock->get_parent()->is_ambiguous_auth()) { | |
4041 | dout(7) << "simple_eval not stable and ambiguous auth, waiting on " << *lock->get_parent() << dendl; | |
4042 | //if (!lock->get_parent()->is_waiter(MDSCacheObject::WAIT_SINGLEAUTH)) | |
4043 | lock->get_parent()->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, new C_Locker_SimpleEval(this, lock)); | |
4044 | return; | |
4045 | } | |
4046 | ||
4047 | if (!lock->get_parent()->is_auth()) { | |
4048 | dout(7) << "try_simple_eval not auth for " << *lock->get_parent() << dendl; | |
4049 | return; | |
4050 | } | |
4051 | ||
4052 | if (!lock->get_parent()->can_auth_pin()) { | |
4053 | dout(7) << "try_simple_eval can't auth_pin, waiting on " << *lock->get_parent() << dendl; | |
4054 | //if (!lock->get_parent()->is_waiter(MDSCacheObject::WAIT_SINGLEAUTH)) | |
4055 | lock->get_parent()->add_waiter(MDSCacheObject::WAIT_UNFREEZE, new C_Locker_SimpleEval(this, lock)); | |
4056 | return; | |
4057 | } | |
4058 | ||
4059 | if (lock->is_stable()) | |
4060 | simple_eval(lock); | |
4061 | } | |
4062 | */ | |
4063 | ||
4064 | ||
4065 | void Locker::simple_eval(SimpleLock *lock, bool *need_issue) | |
4066 | { | |
4067 | dout(10) << "simple_eval " << *lock << " on " << *lock->get_parent() << dendl; | |
4068 | ||
11fdf7f2 TL |
4069 | ceph_assert(lock->get_parent()->is_auth()); |
4070 | ceph_assert(lock->is_stable()); | |
7c673cae FG |
4071 | |
4072 | if (lock->get_parent()->is_freezing_or_frozen()) { | |
11fdf7f2 TL |
4073 | // dentry/snap lock in unreadable state can block path traverse |
4074 | if ((lock->get_type() != CEPH_LOCK_DN && | |
4075 | lock->get_type() != CEPH_LOCK_ISNAP) || | |
7c673cae | 4076 | lock->get_state() == LOCK_SYNC || |
11fdf7f2 | 4077 | lock->get_parent()->is_frozen()) |
7c673cae FG |
4078 | return; |
4079 | } | |
4080 | ||
4081 | if (mdcache->is_readonly()) { | |
4082 | if (lock->get_state() != LOCK_SYNC) { | |
4083 | dout(10) << "simple_eval read-only FS, syncing " << *lock << " on " << *lock->get_parent() << dendl; | |
4084 | simple_sync(lock, need_issue); | |
4085 | } | |
4086 | return; | |
4087 | } | |
4088 | ||
4089 | CInode *in = 0; | |
4090 | int wanted = 0; | |
11fdf7f2 | 4091 | if (lock->get_cap_shift()) { |
7c673cae FG |
4092 | in = static_cast<CInode*>(lock->get_parent()); |
4093 | in->get_caps_wanted(&wanted, NULL, lock->get_cap_shift()); | |
4094 | } | |
4095 | ||
4096 | // -> excl? | |
4097 | if (lock->get_state() != LOCK_EXCL && | |
4098 | in && in->get_target_loner() >= 0 && | |
4099 | (wanted & CEPH_CAP_GEXCL)) { | |
4100 | dout(7) << "simple_eval stable, going to excl " << *lock | |
4101 | << " on " << *lock->get_parent() << dendl; | |
4102 | simple_excl(lock, need_issue); | |
4103 | } | |
4104 | ||
4105 | // stable -> sync? | |
4106 | else if (lock->get_state() != LOCK_SYNC && | |
4107 | !lock->is_wrlocked() && | |
4108 | ((!(wanted & CEPH_CAP_GEXCL) && !lock->is_waiter_for(SimpleLock::WAIT_WR)) || | |
4109 | (lock->get_state() == LOCK_EXCL && in && in->get_target_loner() < 0))) { | |
4110 | dout(7) << "simple_eval stable, syncing " << *lock | |
4111 | << " on " << *lock->get_parent() << dendl; | |
4112 | simple_sync(lock, need_issue); | |
4113 | } | |
4114 | } | |
4115 | ||
4116 | ||
4117 | // mid | |
4118 | ||
4119 | bool Locker::simple_sync(SimpleLock *lock, bool *need_issue) | |
4120 | { | |
4121 | dout(7) << "simple_sync on " << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 TL |
4122 | ceph_assert(lock->get_parent()->is_auth()); |
4123 | ceph_assert(lock->is_stable()); | |
7c673cae FG |
4124 | |
4125 | CInode *in = 0; | |
4126 | if (lock->get_cap_shift()) | |
4127 | in = static_cast<CInode *>(lock->get_parent()); | |
4128 | ||
4129 | int old_state = lock->get_state(); | |
4130 | ||
4131 | if (old_state != LOCK_TSYN) { | |
4132 | ||
4133 | switch (lock->get_state()) { | |
4134 | case LOCK_MIX: lock->set_state(LOCK_MIX_SYNC); break; | |
4135 | case LOCK_LOCK: lock->set_state(LOCK_LOCK_SYNC); break; | |
4136 | case LOCK_XSYN: lock->set_state(LOCK_XSYN_SYNC); break; | |
4137 | case LOCK_EXCL: lock->set_state(LOCK_EXCL_SYNC); break; | |
4138 | default: ceph_abort(); | |
4139 | } | |
4140 | ||
4141 | int gather = 0; | |
4142 | if (lock->is_wrlocked()) | |
4143 | gather++; | |
4144 | ||
4145 | if (lock->get_parent()->is_replicated() && old_state == LOCK_MIX) { | |
4146 | send_lock_message(lock, LOCK_AC_SYNC); | |
4147 | lock->init_gather(); | |
4148 | gather++; | |
4149 | } | |
4150 | ||
4151 | if (in && in->is_head()) { | |
4152 | if (in->issued_caps_need_gather(lock)) { | |
4153 | if (need_issue) | |
4154 | *need_issue = true; | |
4155 | else | |
4156 | issue_caps(in); | |
4157 | gather++; | |
4158 | } | |
4159 | } | |
4160 | ||
4161 | bool need_recover = false; | |
4162 | if (lock->get_type() == CEPH_LOCK_IFILE) { | |
11fdf7f2 | 4163 | ceph_assert(in); |
7c673cae FG |
4164 | if (in->state_test(CInode::STATE_NEEDSRECOVER)) { |
4165 | mds->mdcache->queue_file_recover(in); | |
4166 | need_recover = true; | |
4167 | gather++; | |
4168 | } | |
4169 | } | |
4170 | ||
4171 | if (!gather && lock->is_dirty()) { | |
4172 | lock->get_parent()->auth_pin(lock); | |
4173 | scatter_writebehind(static_cast<ScatterLock*>(lock)); | |
4174 | mds->mdlog->flush(); | |
4175 | return false; | |
4176 | } | |
4177 | ||
4178 | if (gather) { | |
4179 | lock->get_parent()->auth_pin(lock); | |
4180 | if (need_recover) | |
4181 | mds->mdcache->do_file_recover(); | |
4182 | return false; | |
4183 | } | |
4184 | } | |
4185 | ||
4186 | if (lock->get_parent()->is_replicated()) { // FIXME | |
4187 | bufferlist data; | |
4188 | lock->encode_locked_state(data); | |
4189 | send_lock_message(lock, LOCK_AC_SYNC, data); | |
4190 | } | |
4191 | lock->set_state(LOCK_SYNC); | |
4192 | lock->finish_waiters(SimpleLock::WAIT_RD|SimpleLock::WAIT_STABLE); | |
4193 | if (in && in->is_head()) { | |
4194 | if (need_issue) | |
4195 | *need_issue = true; | |
4196 | else | |
4197 | issue_caps(in); | |
4198 | } | |
4199 | return true; | |
4200 | } | |
4201 | ||
4202 | void Locker::simple_excl(SimpleLock *lock, bool *need_issue) | |
4203 | { | |
4204 | dout(7) << "simple_excl on " << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 TL |
4205 | ceph_assert(lock->get_parent()->is_auth()); |
4206 | ceph_assert(lock->is_stable()); | |
7c673cae FG |
4207 | |
4208 | CInode *in = 0; | |
4209 | if (lock->get_cap_shift()) | |
4210 | in = static_cast<CInode *>(lock->get_parent()); | |
4211 | ||
4212 | switch (lock->get_state()) { | |
4213 | case LOCK_LOCK: lock->set_state(LOCK_LOCK_EXCL); break; | |
4214 | case LOCK_SYNC: lock->set_state(LOCK_SYNC_EXCL); break; | |
4215 | case LOCK_XSYN: lock->set_state(LOCK_XSYN_EXCL); break; | |
4216 | default: ceph_abort(); | |
4217 | } | |
4218 | ||
4219 | int gather = 0; | |
4220 | if (lock->is_rdlocked()) | |
4221 | gather++; | |
4222 | if (lock->is_wrlocked()) | |
4223 | gather++; | |
4224 | ||
4225 | if (lock->get_parent()->is_replicated() && | |
4226 | lock->get_state() != LOCK_LOCK_EXCL && | |
4227 | lock->get_state() != LOCK_XSYN_EXCL) { | |
4228 | send_lock_message(lock, LOCK_AC_LOCK); | |
4229 | lock->init_gather(); | |
4230 | gather++; | |
4231 | } | |
4232 | ||
4233 | if (in && in->is_head()) { | |
4234 | if (in->issued_caps_need_gather(lock)) { | |
4235 | if (need_issue) | |
4236 | *need_issue = true; | |
4237 | else | |
4238 | issue_caps(in); | |
4239 | gather++; | |
4240 | } | |
4241 | } | |
4242 | ||
4243 | if (gather) { | |
4244 | lock->get_parent()->auth_pin(lock); | |
4245 | } else { | |
4246 | lock->set_state(LOCK_EXCL); | |
4247 | lock->finish_waiters(SimpleLock::WAIT_WR|SimpleLock::WAIT_STABLE); | |
4248 | if (in) { | |
4249 | if (need_issue) | |
4250 | *need_issue = true; | |
4251 | else | |
4252 | issue_caps(in); | |
4253 | } | |
4254 | } | |
4255 | } | |
4256 | ||
4257 | void Locker::simple_lock(SimpleLock *lock, bool *need_issue) | |
4258 | { | |
4259 | dout(7) << "simple_lock on " << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 TL |
4260 | ceph_assert(lock->get_parent()->is_auth()); |
4261 | ceph_assert(lock->is_stable()); | |
4262 | ceph_assert(lock->get_state() != LOCK_LOCK); | |
7c673cae FG |
4263 | |
4264 | CInode *in = 0; | |
4265 | if (lock->get_cap_shift()) | |
4266 | in = static_cast<CInode *>(lock->get_parent()); | |
4267 | ||
4268 | int old_state = lock->get_state(); | |
4269 | ||
4270 | switch (lock->get_state()) { | |
4271 | case LOCK_SYNC: lock->set_state(LOCK_SYNC_LOCK); break; | |
b32b8144 | 4272 | case LOCK_XSYN: lock->set_state(LOCK_XSYN_LOCK); break; |
7c673cae FG |
4273 | case LOCK_EXCL: lock->set_state(LOCK_EXCL_LOCK); break; |
4274 | case LOCK_MIX: lock->set_state(LOCK_MIX_LOCK); | |
4275 | (static_cast<ScatterLock *>(lock))->clear_unscatter_wanted(); | |
4276 | break; | |
4277 | case LOCK_TSYN: lock->set_state(LOCK_TSYN_LOCK); break; | |
4278 | default: ceph_abort(); | |
4279 | } | |
4280 | ||
4281 | int gather = 0; | |
4282 | if (lock->is_leased()) { | |
4283 | gather++; | |
4284 | revoke_client_leases(lock); | |
4285 | } | |
4286 | if (lock->is_rdlocked()) | |
4287 | gather++; | |
4288 | if (in && in->is_head()) { | |
4289 | if (in->issued_caps_need_gather(lock)) { | |
4290 | if (need_issue) | |
4291 | *need_issue = true; | |
4292 | else | |
4293 | issue_caps(in); | |
4294 | gather++; | |
4295 | } | |
4296 | } | |
4297 | ||
4298 | bool need_recover = false; | |
4299 | if (lock->get_type() == CEPH_LOCK_IFILE) { | |
11fdf7f2 | 4300 | ceph_assert(in); |
7c673cae FG |
4301 | if(in->state_test(CInode::STATE_NEEDSRECOVER)) { |
4302 | mds->mdcache->queue_file_recover(in); | |
4303 | need_recover = true; | |
4304 | gather++; | |
4305 | } | |
4306 | } | |
4307 | ||
4308 | if (lock->get_parent()->is_replicated() && | |
4309 | lock->get_state() == LOCK_MIX_LOCK && | |
4310 | gather) { | |
4311 | dout(10) << " doing local stage of mix->lock gather before gathering from replicas" << dendl; | |
4312 | } else { | |
4313 | // move to second stage of gather now, so we don't send the lock action later. | |
4314 | if (lock->get_state() == LOCK_MIX_LOCK) | |
4315 | lock->set_state(LOCK_MIX_LOCK2); | |
4316 | ||
4317 | if (lock->get_parent()->is_replicated() && | |
4318 | lock->get_sm()->states[old_state].replica_state != LOCK_LOCK) { // replica may already be LOCK | |
4319 | gather++; | |
4320 | send_lock_message(lock, LOCK_AC_LOCK); | |
4321 | lock->init_gather(); | |
4322 | } | |
4323 | } | |
4324 | ||
4325 | if (!gather && lock->is_dirty()) { | |
4326 | lock->get_parent()->auth_pin(lock); | |
4327 | scatter_writebehind(static_cast<ScatterLock*>(lock)); | |
4328 | mds->mdlog->flush(); | |
4329 | return; | |
4330 | } | |
4331 | ||
4332 | if (gather) { | |
4333 | lock->get_parent()->auth_pin(lock); | |
4334 | if (need_recover) | |
4335 | mds->mdcache->do_file_recover(); | |
4336 | } else { | |
4337 | lock->set_state(LOCK_LOCK); | |
4338 | lock->finish_waiters(ScatterLock::WAIT_XLOCK|ScatterLock::WAIT_WR|ScatterLock::WAIT_STABLE); | |
4339 | } | |
4340 | } | |
4341 | ||
4342 | ||
4343 | void Locker::simple_xlock(SimpleLock *lock) | |
4344 | { | |
4345 | dout(7) << "simple_xlock on " << *lock << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 | 4346 | ceph_assert(lock->get_parent()->is_auth()); |
7c673cae | 4347 | //assert(lock->is_stable()); |
11fdf7f2 | 4348 | ceph_assert(lock->get_state() != LOCK_XLOCK); |
7c673cae FG |
4349 | |
4350 | CInode *in = 0; | |
4351 | if (lock->get_cap_shift()) | |
4352 | in = static_cast<CInode *>(lock->get_parent()); | |
4353 | ||
4354 | if (lock->is_stable()) | |
4355 | lock->get_parent()->auth_pin(lock); | |
4356 | ||
4357 | switch (lock->get_state()) { | |
4358 | case LOCK_LOCK: | |
4359 | case LOCK_XLOCKDONE: lock->set_state(LOCK_LOCK_XLOCK); break; | |
4360 | default: ceph_abort(); | |
4361 | } | |
4362 | ||
4363 | int gather = 0; | |
4364 | if (lock->is_rdlocked()) | |
4365 | gather++; | |
4366 | if (lock->is_wrlocked()) | |
4367 | gather++; | |
4368 | ||
4369 | if (in && in->is_head()) { | |
4370 | if (in->issued_caps_need_gather(lock)) { | |
4371 | issue_caps(in); | |
4372 | gather++; | |
4373 | } | |
4374 | } | |
4375 | ||
4376 | if (!gather) { | |
4377 | lock->set_state(LOCK_PREXLOCK); | |
4378 | //assert("shouldn't be called if we are already xlockable" == 0); | |
4379 | } | |
4380 | } | |
4381 | ||
4382 | ||
4383 | ||
4384 | ||
4385 | ||
4386 | // ========================================================================== | |
4387 | // scatter lock | |
4388 | ||
4389 | /* | |
4390 | ||
4391 | Some notes on scatterlocks. | |
4392 | ||
4393 | - The scatter/gather is driven by the inode lock. The scatter always | |
4394 | brings in the latest metadata from the fragments. | |
4395 | ||
4396 | - When in a scattered/MIX state, fragments are only allowed to | |
4397 | update/be written to if the accounted stat matches the inode's | |
4398 | current version. | |
4399 | ||
4400 | - That means, on gather, we _only_ assimilate diffs for frag metadata | |
4401 | that match the current version, because those are the only ones | |
4402 | written during this scatter/gather cycle. (Others didn't permit | |
4403 | it.) We increment the version and journal this to disk. | |
4404 | ||
4405 | - When possible, we also simultaneously update our local frag | |
4406 | accounted stats to match. | |
4407 | ||
4408 | - On scatter, the new inode info is broadcast to frags, both local | |
4409 | and remote. If possible (auth and !frozen), the dirfrag auth | |
4410 | should update the accounted state (if it isn't already up to date). | |
4411 | Note that this may occur on both the local inode auth node and | |
4412 | inode replicas, so there are two potential paths. If it is NOT | |
4413 | possible, they need to mark_stale to prevent any possible writes. | |
4414 | ||
4415 | - A scatter can be to MIX (potentially writeable) or to SYNC (read | |
4416 | only). Both are opportunities to update the frag accounted stats, | |
4417 | even though only the MIX case is affected by a stale dirfrag. | |
4418 | ||
4419 | - Because many scatter/gather cycles can potentially go by without a | |
4420 | frag being able to update its accounted stats (due to being frozen | |
4421 | by exports/refragments in progress), the frag may have (even very) | |
4422 | old stat versions. That's fine. If when we do want to update it, | |
4423 | we can update accounted_* and the version first. | |
4424 | ||
4425 | */ | |
4426 | ||
4427 | class C_Locker_ScatterWB : public LockerLogContext { | |
4428 | ScatterLock *lock; | |
4429 | MutationRef mut; | |
4430 | public: | |
4431 | C_Locker_ScatterWB(Locker *l, ScatterLock *sl, MutationRef& m) : | |
4432 | LockerLogContext(l), lock(sl), mut(m) {} | |
4433 | void finish(int r) override { | |
4434 | locker->scatter_writebehind_finish(lock, mut); | |
4435 | } | |
4436 | }; | |
4437 | ||
4438 | void Locker::scatter_writebehind(ScatterLock *lock) | |
4439 | { | |
4440 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
4441 | dout(10) << "scatter_writebehind " << in->inode.mtime << " on " << *lock << " on " << *in << dendl; | |
4442 | ||
4443 | // journal | |
4444 | MutationRef mut(new MutationImpl()); | |
4445 | mut->ls = mds->mdlog->get_current_segment(); | |
4446 | ||
4447 | // forcefully take a wrlock | |
4448 | lock->get_wrlock(true); | |
11fdf7f2 | 4449 | mut->locks.emplace(lock, MutationImpl::LockOp::WRLOCK); |
7c673cae FG |
4450 | |
4451 | in->pre_cow_old_inode(); // avoid cow mayhem | |
4452 | ||
94b18763 FG |
4453 | auto &pi = in->project_inode(); |
4454 | pi.inode.version = in->pre_dirty(); | |
7c673cae FG |
4455 | |
4456 | in->finish_scatter_gather_update(lock->get_type()); | |
4457 | lock->start_flush(); | |
4458 | ||
4459 | EUpdate *le = new EUpdate(mds->mdlog, "scatter_writebehind"); | |
4460 | mds->mdlog->start_entry(le); | |
4461 | ||
4462 | mdcache->predirty_journal_parents(mut, &le->metablob, in, 0, PREDIRTY_PRIMARY); | |
4463 | mdcache->journal_dirty_inode(mut.get(), &le->metablob, in); | |
4464 | ||
4465 | in->finish_scatter_gather_update_accounted(lock->get_type(), mut, &le->metablob); | |
4466 | ||
4467 | mds->mdlog->submit_entry(le, new C_Locker_ScatterWB(this, lock, mut)); | |
4468 | } | |
4469 | ||
4470 | void Locker::scatter_writebehind_finish(ScatterLock *lock, MutationRef& mut) | |
4471 | { | |
4472 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
4473 | dout(10) << "scatter_writebehind_finish on " << *lock << " on " << *in << dendl; | |
4474 | in->pop_and_dirty_projected_inode(mut->ls); | |
4475 | ||
4476 | lock->finish_flush(); | |
4477 | ||
4478 | // if replicas may have flushed in a mix->lock state, send another | |
4479 | // message so they can finish_flush(). | |
4480 | if (in->is_replicated()) { | |
4481 | switch (lock->get_state()) { | |
4482 | case LOCK_MIX_LOCK: | |
4483 | case LOCK_MIX_LOCK2: | |
4484 | case LOCK_MIX_EXCL: | |
4485 | case LOCK_MIX_TSYN: | |
4486 | send_lock_message(lock, LOCK_AC_LOCKFLUSHED); | |
4487 | } | |
4488 | } | |
4489 | ||
4490 | mut->apply(); | |
4491 | drop_locks(mut.get()); | |
4492 | mut->cleanup(); | |
4493 | ||
4494 | if (lock->is_stable()) | |
4495 | lock->finish_waiters(ScatterLock::WAIT_STABLE); | |
4496 | ||
4497 | //scatter_eval_gather(lock); | |
4498 | } | |
4499 | ||
4500 | void Locker::scatter_eval(ScatterLock *lock, bool *need_issue) | |
4501 | { | |
4502 | dout(10) << "scatter_eval " << *lock << " on " << *lock->get_parent() << dendl; | |
4503 | ||
11fdf7f2 TL |
4504 | ceph_assert(lock->get_parent()->is_auth()); |
4505 | ceph_assert(lock->is_stable()); | |
7c673cae FG |
4506 | |
4507 | if (lock->get_parent()->is_freezing_or_frozen()) { | |
4508 | dout(20) << " freezing|frozen" << dendl; | |
4509 | return; | |
4510 | } | |
4511 | ||
4512 | if (mdcache->is_readonly()) { | |
4513 | if (lock->get_state() != LOCK_SYNC) { | |
4514 | dout(10) << "scatter_eval read-only FS, syncing " << *lock << " on " << *lock->get_parent() << dendl; | |
4515 | simple_sync(lock, need_issue); | |
4516 | } | |
4517 | return; | |
4518 | } | |
4519 | ||
4520 | if (!lock->is_rdlocked() && | |
4521 | lock->get_state() != LOCK_MIX && | |
4522 | lock->get_scatter_wanted()) { | |
4523 | dout(10) << "scatter_eval scatter_wanted, bump to mix " << *lock | |
4524 | << " on " << *lock->get_parent() << dendl; | |
4525 | scatter_mix(lock, need_issue); | |
4526 | return; | |
4527 | } | |
4528 | ||
4529 | if (lock->get_type() == CEPH_LOCK_INEST) { | |
4530 | // in general, we want to keep INEST writable at all times. | |
4531 | if (!lock->is_rdlocked()) { | |
4532 | if (lock->get_parent()->is_replicated()) { | |
4533 | if (lock->get_state() != LOCK_MIX) | |
4534 | scatter_mix(lock, need_issue); | |
4535 | } else { | |
4536 | if (lock->get_state() != LOCK_LOCK) | |
4537 | simple_lock(lock, need_issue); | |
4538 | } | |
4539 | } | |
4540 | return; | |
4541 | } | |
4542 | ||
4543 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
4544 | if (!in->has_subtree_or_exporting_dirfrag() || in->is_base()) { | |
4545 | // i _should_ be sync. | |
4546 | if (!lock->is_wrlocked() && | |
4547 | lock->get_state() != LOCK_SYNC) { | |
4548 | dout(10) << "scatter_eval no wrlocks|xlocks, not subtree root inode, syncing" << dendl; | |
4549 | simple_sync(lock, need_issue); | |
4550 | } | |
4551 | } | |
4552 | } | |
4553 | ||
4554 | ||
4555 | /* | |
4556 | * mark a scatterlock to indicate that the dir fnode has some dirty data | |
4557 | */ | |
4558 | void Locker::mark_updated_scatterlock(ScatterLock *lock) | |
4559 | { | |
4560 | lock->mark_dirty(); | |
4561 | if (lock->get_updated_item()->is_on_list()) { | |
4562 | dout(10) << "mark_updated_scatterlock " << *lock | |
4563 | << " - already on list since " << lock->get_update_stamp() << dendl; | |
4564 | } else { | |
4565 | updated_scatterlocks.push_back(lock->get_updated_item()); | |
4566 | utime_t now = ceph_clock_now(); | |
4567 | lock->set_update_stamp(now); | |
4568 | dout(10) << "mark_updated_scatterlock " << *lock | |
4569 | << " - added at " << now << dendl; | |
4570 | } | |
4571 | } | |
4572 | ||
4573 | /* | |
4574 | * this is called by scatter_tick and LogSegment::try_to_trim() when | |
4575 | * trying to flush dirty scattered data (i.e. updated fnode) back to | |
4576 | * the inode. | |
4577 | * | |
4578 | * we need to lock|scatter in order to push fnode changes into the | |
4579 | * inode.dirstat. | |
4580 | */ | |
11fdf7f2 | 4581 | void Locker::scatter_nudge(ScatterLock *lock, MDSContext *c, bool forcelockchange) |
7c673cae FG |
4582 | { |
4583 | CInode *p = static_cast<CInode *>(lock->get_parent()); | |
4584 | ||
4585 | if (p->is_frozen() || p->is_freezing()) { | |
4586 | dout(10) << "scatter_nudge waiting for unfreeze on " << *p << dendl; | |
4587 | if (c) | |
4588 | p->add_waiter(MDSCacheObject::WAIT_UNFREEZE, c); | |
b32b8144 | 4589 | else if (lock->is_dirty()) |
7c673cae FG |
4590 | // just requeue. not ideal.. starvation prone.. |
4591 | updated_scatterlocks.push_back(lock->get_updated_item()); | |
4592 | return; | |
4593 | } | |
4594 | ||
4595 | if (p->is_ambiguous_auth()) { | |
4596 | dout(10) << "scatter_nudge waiting for single auth on " << *p << dendl; | |
4597 | if (c) | |
4598 | p->add_waiter(MDSCacheObject::WAIT_SINGLEAUTH, c); | |
b32b8144 | 4599 | else if (lock->is_dirty()) |
7c673cae FG |
4600 | // just requeue. not ideal.. starvation prone.. |
4601 | updated_scatterlocks.push_back(lock->get_updated_item()); | |
4602 | return; | |
4603 | } | |
4604 | ||
4605 | if (p->is_auth()) { | |
4606 | int count = 0; | |
4607 | while (true) { | |
4608 | if (lock->is_stable()) { | |
4609 | // can we do it now? | |
4610 | // (only if we're not replicated.. if we are, we really do need | |
4611 | // to nudge the lock state!) | |
4612 | /* | |
4613 | actually, even if we're not replicated, we can't stay in MIX, because another mds | |
4614 | could discover and replicate us at any time. if that happens while we're flushing, | |
4615 | they end up in MIX but their inode has the old scatterstat version. | |
4616 | ||
4617 | if (!forcelockchange && !lock->get_parent()->is_replicated() && lock->can_wrlock(-1)) { | |
4618 | dout(10) << "scatter_nudge auth, propagating " << *lock << " on " << *p << dendl; | |
4619 | scatter_writebehind(lock); | |
4620 | if (c) | |
4621 | lock->add_waiter(SimpleLock::WAIT_STABLE, c); | |
4622 | return; | |
4623 | } | |
4624 | */ | |
4625 | ||
4626 | if (mdcache->is_readonly()) { | |
4627 | if (lock->get_state() != LOCK_SYNC) { | |
4628 | dout(10) << "scatter_nudge auth, read-only FS, syncing " << *lock << " on " << *p << dendl; | |
4629 | simple_sync(static_cast<ScatterLock*>(lock)); | |
4630 | } | |
4631 | break; | |
4632 | } | |
4633 | ||
4634 | // adjust lock state | |
4635 | dout(10) << "scatter_nudge auth, scatter/unscattering " << *lock << " on " << *p << dendl; | |
4636 | switch (lock->get_type()) { | |
4637 | case CEPH_LOCK_IFILE: | |
4638 | if (p->is_replicated() && lock->get_state() != LOCK_MIX) | |
4639 | scatter_mix(static_cast<ScatterLock*>(lock)); | |
4640 | else if (lock->get_state() != LOCK_LOCK) | |
4641 | simple_lock(static_cast<ScatterLock*>(lock)); | |
4642 | else | |
4643 | simple_sync(static_cast<ScatterLock*>(lock)); | |
4644 | break; | |
4645 | ||
4646 | case CEPH_LOCK_IDFT: | |
4647 | case CEPH_LOCK_INEST: | |
4648 | if (p->is_replicated() && lock->get_state() != LOCK_MIX) | |
4649 | scatter_mix(lock); | |
4650 | else if (lock->get_state() != LOCK_LOCK) | |
4651 | simple_lock(lock); | |
4652 | else | |
4653 | simple_sync(lock); | |
4654 | break; | |
4655 | default: | |
4656 | ceph_abort(); | |
4657 | } | |
4658 | ++count; | |
4659 | if (lock->is_stable() && count == 2) { | |
4660 | dout(10) << "scatter_nudge oh, stable after two cycles." << dendl; | |
4661 | // this should only realy happen when called via | |
4662 | // handle_file_lock due to AC_NUDGE, because the rest of the | |
4663 | // time we are replicated or have dirty data and won't get | |
4664 | // called. bailing here avoids an infinite loop. | |
11fdf7f2 | 4665 | ceph_assert(!c); |
7c673cae FG |
4666 | break; |
4667 | } | |
4668 | } else { | |
4669 | dout(10) << "scatter_nudge auth, waiting for stable " << *lock << " on " << *p << dendl; | |
4670 | if (c) | |
4671 | lock->add_waiter(SimpleLock::WAIT_STABLE, c); | |
4672 | return; | |
4673 | } | |
4674 | } | |
4675 | } else { | |
4676 | dout(10) << "scatter_nudge replica, requesting scatter/unscatter of " | |
4677 | << *lock << " on " << *p << dendl; | |
4678 | // request unscatter? | |
4679 | mds_rank_t auth = lock->get_parent()->authority().first; | |
11fdf7f2 TL |
4680 | if (!mds->is_cluster_degraded() || mds->mdsmap->is_clientreplay_or_active_or_stopping(auth)) { |
4681 | mds->send_message_mds(MLock::create(lock, LOCK_AC_NUDGE, mds->get_nodeid()), auth); | |
4682 | } | |
7c673cae FG |
4683 | |
4684 | // wait... | |
4685 | if (c) | |
4686 | lock->add_waiter(SimpleLock::WAIT_STABLE, c); | |
4687 | ||
4688 | // also, requeue, in case we had wrong auth or something | |
b32b8144 FG |
4689 | if (lock->is_dirty()) |
4690 | updated_scatterlocks.push_back(lock->get_updated_item()); | |
7c673cae FG |
4691 | } |
4692 | } | |
4693 | ||
4694 | void Locker::scatter_tick() | |
4695 | { | |
4696 | dout(10) << "scatter_tick" << dendl; | |
4697 | ||
4698 | // updated | |
4699 | utime_t now = ceph_clock_now(); | |
4700 | int n = updated_scatterlocks.size(); | |
4701 | while (!updated_scatterlocks.empty()) { | |
4702 | ScatterLock *lock = updated_scatterlocks.front(); | |
4703 | ||
4704 | if (n-- == 0) break; // scatter_nudge() may requeue; avoid looping | |
4705 | ||
4706 | if (!lock->is_dirty()) { | |
4707 | updated_scatterlocks.pop_front(); | |
4708 | dout(10) << " removing from updated_scatterlocks " | |
4709 | << *lock << " " << *lock->get_parent() << dendl; | |
4710 | continue; | |
4711 | } | |
11fdf7f2 | 4712 | if (now - lock->get_update_stamp() < g_conf()->mds_scatter_nudge_interval) |
7c673cae FG |
4713 | break; |
4714 | updated_scatterlocks.pop_front(); | |
4715 | scatter_nudge(lock, 0); | |
4716 | } | |
4717 | mds->mdlog->flush(); | |
4718 | } | |
4719 | ||
4720 | ||
4721 | void Locker::scatter_tempsync(ScatterLock *lock, bool *need_issue) | |
4722 | { | |
4723 | dout(10) << "scatter_tempsync " << *lock | |
4724 | << " on " << *lock->get_parent() << dendl; | |
11fdf7f2 TL |
4725 | ceph_assert(lock->get_parent()->is_auth()); |
4726 | ceph_assert(lock->is_stable()); | |
7c673cae | 4727 | |
11fdf7f2 | 4728 | ceph_abort_msg("not fully implemented, at least not for filelock"); |
7c673cae FG |
4729 | |
4730 | CInode *in = static_cast<CInode *>(lock->get_parent()); | |
4731 | ||
4732 | switch (lock->get_state()) { | |
4733 | case LOCK_SYNC: ceph_abort(); // this shouldn't happen | |
4734 | case LOCK_LOCK: lock->set_state(LOCK_LOCK_TSYN); break; | |
4735 | case LOCK_MIX: lock->set_state(LOCK_MIX_TSYN); break; | |
4736 | default: ceph_abort(); | |
4737 | } | |
4738 | ||
4739 | int gather = 0; | |
4740 | if (lock->is_wrlocked()) | |
4741 | gather++; | |
4742 | ||
4743 | if (lock->get_cap_shift() && | |
4744 | in->is_head() && | |
4745 | in->issued_caps_need_gather(lock)) { | |
4746 | if (need_issue) | |
4747 | *need_issue = true; | |
4748 | else | |
4749 | issue_caps(in); | |
4750 | gather++; | |
4751 | } | |
4752 | ||
4753 | if (lock->get_state() == LOCK_MIX_TSYN && | |
4754 | in->is_replicated()) { | |
4755 | lock->init_gather(); | |
4756 | send_lock_message(lock, LOCK_AC_LOCK); | |
4757 | gather++; | |
4758 | } | |
4759 | ||
4760 | if (gather) { | |
4761 | in->auth_pin(lock); | |
4762 | } else { | |
4763 | // do tempsync | |
4764 | lock->set_state(LOCK_TSYN); | |
4765 | lock->finish_waiters(ScatterLock::WAIT_RD|ScatterLock::WAIT_STABLE); | |
4766 | if (lock->get_cap_shift()) { | |
4767 | if (need_issue) | |
4768 | *need_issue = true; | |
4769 | else | |
4770 | issue_caps(in); | |
4771 | } | |
4772 | } | |
4773 | } | |
4774 | ||
4775 | ||
4776 | ||
4777 | // ========================================================================== | |
4778 | // local lock | |
4779 | ||
4780 | void Locker::local_wrlock_grab(LocalLock *lock, MutationRef& mut) | |
4781 | { | |
4782 | dout(7) << "local_wrlock_grab on " << *lock | |
4783 | << " on " << *lock->get_parent() << dendl; | |
4784 | ||
11fdf7f2 TL |
4785 | ceph_assert(lock->get_parent()->is_auth()); |
4786 | ceph_assert(lock->can_wrlock()); | |
7c673cae | 4787 | lock->get_wrlock(mut->get_client()); |
11fdf7f2 TL |
4788 | |
4789 | auto ret = mut->locks.emplace(lock, MutationImpl::LockOp::WRLOCK); | |
4790 | ceph_assert(ret.second); | |
7c673cae FG |
4791 | } |
4792 | ||
4793 | bool Locker::local_wrlock_start(LocalLock *lock, MDRequestRef& mut) | |
4794 | { | |
4795 | dout(7) << "local_wrlock_start on " << *lock | |
4796 | << " on " << *lock->get_parent() << dendl; | |
4797 | ||
11fdf7f2 | 4798 | ceph_assert(lock->get_parent()->is_auth()); |
7c673cae | 4799 | if (lock->can_wrlock()) { |
7c673cae | 4800 | lock->get_wrlock(mut->get_client()); |
11fdf7f2 TL |
4801 | auto it = mut->locks.emplace_hint(mut->locks.end(), lock, MutationImpl::LockOp::WRLOCK); |
4802 | ceph_assert(it->is_wrlock()); | |
7c673cae FG |
4803 | return true; |
4804 | } else { | |
4805 | lock->add_waiter(SimpleLock::WAIT_WR|SimpleLock::WAIT_STABLE, new C_MDS_RetryRequest(mdcache, mut)); | |
4806 | return false; | |
4807 | } | |
4808 | } | |
4809 | ||
11fdf7f2 | 4810 | void Locker::local_wrlock_finish(const MutationImpl::lock_iterator& it, MutationImpl *mut) |
7c673cae | 4811 | { |
11fdf7f2 TL |
4812 | ceph_assert(it->is_wrlock()); |
4813 | LocalLock *lock = static_cast<LocalLock*>(it->lock); | |
7c673cae FG |
4814 | dout(7) << "local_wrlock_finish on " << *lock |
4815 | << " on " << *lock->get_parent() << dendl; | |
4816 | lock->put_wrlock(); | |
11fdf7f2 | 4817 | mut->locks.erase(it); |
7c673cae FG |
4818 | if (lock->get_num_wrlocks() == 0) { |
4819 | lock->finish_waiters(SimpleLock::WAIT_STABLE | | |
4820 | SimpleLock::WAIT_WR | | |
4821 | SimpleLock::WAIT_RD); | |
4822 | } | |
4823 | } | |
4824 | ||
4825 | bool Locker::local_xlock_start(LocalLock *lock, MDRequestRef& mut) | |
4826 | { | |
4827 | dout(7) << "local_xlock_start on " << *lock | |
4828 | << " on " << *lock->get_parent() << dendl; | |
4829 | ||
11fdf7f2 | 4830 | ceph_assert(lock->get_parent()->is_auth()); |
7c673cae FG |
4831 | if (!lock->can_xlock_local()) { |
4832 | lock->add_waiter(SimpleLock::WAIT_WR|SimpleLock::WAIT_STABLE, new C_MDS_RetryRequest(mdcache, mut)); | |
4833 | return false; | |
4834 | } | |
4835 | ||
4836 | lock->get_xlock(mut, mut->get_client()); | |
11fdf7f2 | 4837 | mut->locks.emplace_hint(mut->locks.end(), lock, MutationImpl::LockOp::XLOCK); |
7c673cae FG |
4838 | return true; |
4839 | } | |
4840 | ||
11fdf7f2 | 4841 | void Locker::local_xlock_finish(const MutationImpl::lock_iterator& it, MutationImpl *mut) |
7c673cae | 4842 | { |
11fdf7f2 TL |
4843 | ceph_assert(it->is_xlock()); |
4844 | LocalLock *lock = static_cast<LocalLock*>(it->lock); | |
7c673cae FG |
4845 | dout(7) << "local_xlock_finish on " << *lock |
4846 | << " on " << *lock->get_parent() << dendl; | |
4847 | lock->put_xlock(); | |
11fdf7f2 | 4848 | mut->locks.erase(it); |
7c673cae FG |
4849 | |
4850 | lock->finish_waiters(SimpleLock::WAIT_STABLE | | |
4851 | SimpleLock::WAIT_WR | | |
4852 | SimpleLock::WAIT_RD); | |
4853 | } | |
4854 | ||
4855 | ||
4856 | ||
4857 | // ========================================================================== | |
4858 | // file lock | |
4859 | ||
4860 | ||
4861 | void Locker::file_eval(ScatterLock *lock, bool *need_issue) | |
4862 | { | |
4863 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
4864 | int loner_wanted, other_wanted; | |
4865 | int wanted = in->get_caps_wanted(&loner_wanted, &other_wanted, CEPH_CAP_SFILE); | |
4866 | dout(7) << "file_eval wanted=" << gcap_string(wanted) | |
4867 | << " loner_wanted=" << gcap_string(loner_wanted) | |
4868 | << " other_wanted=" << gcap_string(other_wanted) | |
4869 | << " filelock=" << *lock << " on " << *lock->get_parent() | |
4870 | << dendl; | |
4871 | ||
11fdf7f2 TL |
4872 | ceph_assert(lock->get_parent()->is_auth()); |
4873 | ceph_assert(lock->is_stable()); | |
7c673cae FG |
4874 | |
4875 | if (lock->get_parent()->is_freezing_or_frozen()) | |
4876 | return; | |
4877 | ||
4878 | if (mdcache->is_readonly()) { | |
4879 | if (lock->get_state() != LOCK_SYNC) { | |
4880 | dout(10) << "file_eval read-only FS, syncing " << *lock << " on " << *lock->get_parent() << dendl; | |
4881 | simple_sync(lock, need_issue); | |
4882 | } | |
4883 | return; | |
4884 | } | |
4885 | ||
4886 | // excl -> *? | |
4887 | if (lock->get_state() == LOCK_EXCL) { | |
4888 | dout(20) << " is excl" << dendl; | |
4889 | int loner_issued, other_issued, xlocker_issued; | |
4890 | in->get_caps_issued(&loner_issued, &other_issued, &xlocker_issued, CEPH_CAP_SFILE); | |
4891 | dout(7) << "file_eval loner_issued=" << gcap_string(loner_issued) | |
4892 | << " other_issued=" << gcap_string(other_issued) | |
4893 | << " xlocker_issued=" << gcap_string(xlocker_issued) | |
4894 | << dendl; | |
4895 | if (!((loner_wanted|loner_issued) & (CEPH_CAP_GEXCL|CEPH_CAP_GWR|CEPH_CAP_GBUFFER)) || | |
4896 | (other_wanted & (CEPH_CAP_GEXCL|CEPH_CAP_GWR|CEPH_CAP_GRD)) || | |
4897 | (in->inode.is_dir() && in->multiple_nonstale_caps())) { // FIXME.. :/ | |
4898 | dout(20) << " should lose it" << dendl; | |
4899 | // we should lose it. | |
4900 | // loner other want | |
4901 | // R R SYNC | |
4902 | // R R|W MIX | |
4903 | // R W MIX | |
4904 | // R|W R MIX | |
4905 | // R|W R|W MIX | |
4906 | // R|W W MIX | |
4907 | // W R MIX | |
4908 | // W R|W MIX | |
4909 | // W W MIX | |
4910 | // -> any writer means MIX; RD doesn't matter. | |
4911 | if (((other_wanted|loner_wanted) & CEPH_CAP_GWR) || | |
4912 | lock->is_waiter_for(SimpleLock::WAIT_WR)) | |
4913 | scatter_mix(lock, need_issue); | |
4914 | else if (!lock->is_wrlocked()) // let excl wrlocks drain first | |
4915 | simple_sync(lock, need_issue); | |
4916 | else | |
4917 | dout(10) << " waiting for wrlock to drain" << dendl; | |
4918 | } | |
4919 | } | |
4920 | ||
4921 | // * -> excl? | |
4922 | else if (lock->get_state() != LOCK_EXCL && | |
4923 | !lock->is_rdlocked() && | |
4924 | //!lock->is_waiter_for(SimpleLock::WAIT_WR) && | |
4925 | ((wanted & (CEPH_CAP_GWR|CEPH_CAP_GBUFFER)) || | |
4926 | (in->inode.is_dir() && !in->has_subtree_or_exporting_dirfrag())) && | |
4927 | in->get_target_loner() >= 0) { | |
4928 | dout(7) << "file_eval stable, bump to loner " << *lock | |
4929 | << " on " << *lock->get_parent() << dendl; | |
4930 | file_excl(lock, need_issue); | |
4931 | } | |
4932 | ||
4933 | // * -> mixed? | |
4934 | else if (lock->get_state() != LOCK_MIX && | |
4935 | !lock->is_rdlocked() && | |
4936 | //!lock->is_waiter_for(SimpleLock::WAIT_WR) && | |
4937 | (lock->get_scatter_wanted() || | |
b32b8144 | 4938 | (in->get_target_loner() < 0 && (wanted & CEPH_CAP_GWR)))) { |
7c673cae FG |
4939 | dout(7) << "file_eval stable, bump to mixed " << *lock |
4940 | << " on " << *lock->get_parent() << dendl; | |
4941 | scatter_mix(lock, need_issue); | |
4942 | } | |
4943 | ||
4944 | // * -> sync? | |
4945 | else if (lock->get_state() != LOCK_SYNC && | |
4946 | !lock->is_wrlocked() && // drain wrlocks first! | |
4947 | !lock->is_waiter_for(SimpleLock::WAIT_WR) && | |
f64942e4 | 4948 | !(wanted & CEPH_CAP_GWR) && |
7c673cae FG |
4949 | !((lock->get_state() == LOCK_MIX) && |
4950 | in->is_dir() && in->has_subtree_or_exporting_dirfrag()) // if we are a delegation point, stay where we are | |
4951 | //((wanted & CEPH_CAP_RD) || | |
4952 | //in->is_replicated() || | |
b32b8144 | 4953 | //lock->is_leased() || |
7c673cae FG |
4954 | //(!loner && lock->get_state() == LOCK_EXCL)) && |
4955 | ) { | |
4956 | dout(7) << "file_eval stable, bump to sync " << *lock | |
4957 | << " on " << *lock->get_parent() << dendl; | |
4958 | simple_sync(lock, need_issue); | |
4959 | } | |
4960 | } | |
4961 | ||
4962 | ||
4963 | ||
4964 | void Locker::scatter_mix(ScatterLock *lock, bool *need_issue) | |
4965 | { | |
4966 | dout(7) << "scatter_mix " << *lock << " on " << *lock->get_parent() << dendl; | |
4967 | ||
4968 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
11fdf7f2 TL |
4969 | ceph_assert(in->is_auth()); |
4970 | ceph_assert(lock->is_stable()); | |
7c673cae FG |
4971 | |
4972 | if (lock->get_state() == LOCK_LOCK) { | |
4973 | in->start_scatter(lock); | |
4974 | if (in->is_replicated()) { | |
4975 | // data | |
4976 | bufferlist softdata; | |
4977 | lock->encode_locked_state(softdata); | |
4978 | ||
4979 | // bcast to replicas | |
4980 | send_lock_message(lock, LOCK_AC_MIX, softdata); | |
4981 | } | |
4982 | ||
4983 | // change lock | |
4984 | lock->set_state(LOCK_MIX); | |
4985 | lock->clear_scatter_wanted(); | |
4986 | if (lock->get_cap_shift()) { | |
4987 | if (need_issue) | |
4988 | *need_issue = true; | |
4989 | else | |
4990 | issue_caps(in); | |
4991 | } | |
4992 | } else { | |
4993 | // gather? | |
4994 | switch (lock->get_state()) { | |
4995 | case LOCK_SYNC: lock->set_state(LOCK_SYNC_MIX); break; | |
7c673cae | 4996 | case LOCK_EXCL: lock->set_state(LOCK_EXCL_MIX); break; |
b32b8144 | 4997 | case LOCK_XSYN: lock->set_state(LOCK_XSYN_MIX); break; |
7c673cae FG |
4998 | case LOCK_TSYN: lock->set_state(LOCK_TSYN_MIX); break; |
4999 | default: ceph_abort(); | |
5000 | } | |
5001 | ||
5002 | int gather = 0; | |
5003 | if (lock->is_rdlocked()) | |
5004 | gather++; | |
5005 | if (in->is_replicated()) { | |
b32b8144 | 5006 | if (lock->get_state() == LOCK_SYNC_MIX) { // for the rest states, replicas are already LOCK |
7c673cae FG |
5007 | send_lock_message(lock, LOCK_AC_MIX); |
5008 | lock->init_gather(); | |
5009 | gather++; | |
5010 | } | |
5011 | } | |
5012 | if (lock->is_leased()) { | |
5013 | revoke_client_leases(lock); | |
5014 | gather++; | |
5015 | } | |
5016 | if (lock->get_cap_shift() && | |
5017 | in->is_head() && | |
5018 | in->issued_caps_need_gather(lock)) { | |
5019 | if (need_issue) | |
5020 | *need_issue = true; | |
5021 | else | |
5022 | issue_caps(in); | |
5023 | gather++; | |
5024 | } | |
5025 | bool need_recover = false; | |
5026 | if (in->state_test(CInode::STATE_NEEDSRECOVER)) { | |
5027 | mds->mdcache->queue_file_recover(in); | |
5028 | need_recover = true; | |
5029 | gather++; | |
5030 | } | |
5031 | ||
5032 | if (gather) { | |
5033 | lock->get_parent()->auth_pin(lock); | |
5034 | if (need_recover) | |
5035 | mds->mdcache->do_file_recover(); | |
5036 | } else { | |
5037 | in->start_scatter(lock); | |
5038 | lock->set_state(LOCK_MIX); | |
5039 | lock->clear_scatter_wanted(); | |
5040 | if (in->is_replicated()) { | |
5041 | bufferlist softdata; | |
5042 | lock->encode_locked_state(softdata); | |
5043 | send_lock_message(lock, LOCK_AC_MIX, softdata); | |
5044 | } | |
5045 | if (lock->get_cap_shift()) { | |
5046 | if (need_issue) | |
5047 | *need_issue = true; | |
5048 | else | |
5049 | issue_caps(in); | |
5050 | } | |
5051 | } | |
5052 | } | |
5053 | } | |
5054 | ||
5055 | ||
5056 | void Locker::file_excl(ScatterLock *lock, bool *need_issue) | |
5057 | { | |
5058 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
5059 | dout(7) << "file_excl " << *lock << " on " << *lock->get_parent() << dendl; | |
5060 | ||
11fdf7f2 TL |
5061 | ceph_assert(in->is_auth()); |
5062 | ceph_assert(lock->is_stable()); | |
7c673cae | 5063 | |
11fdf7f2 | 5064 | ceph_assert((in->get_loner() >= 0 && in->get_mds_caps_wanted().empty()) || |
7c673cae FG |
5065 | (lock->get_state() == LOCK_XSYN)); // must do xsyn -> excl -> <anything else> |
5066 | ||
5067 | switch (lock->get_state()) { | |
5068 | case LOCK_SYNC: lock->set_state(LOCK_SYNC_EXCL); break; | |
5069 | case LOCK_MIX: lock->set_state(LOCK_MIX_EXCL); break; | |
5070 | case LOCK_LOCK: lock->set_state(LOCK_LOCK_EXCL); break; | |
5071 | case LOCK_XSYN: lock->set_state(LOCK_XSYN_EXCL); break; | |
5072 | default: ceph_abort(); | |
5073 | } | |
5074 | int gather = 0; | |
5075 | ||
5076 | if (lock->is_rdlocked()) | |
5077 | gather++; | |
5078 | if (lock->is_wrlocked()) | |
5079 | gather++; | |
5080 | ||
5081 | if (in->is_replicated() && | |
5082 | lock->get_state() != LOCK_LOCK_EXCL && | |
5083 | lock->get_state() != LOCK_XSYN_EXCL) { // if we were lock, replicas are already lock. | |
5084 | send_lock_message(lock, LOCK_AC_LOCK); | |
5085 | lock->init_gather(); | |
5086 | gather++; | |
5087 | } | |
5088 | if (lock->is_leased()) { | |
5089 | revoke_client_leases(lock); | |
5090 | gather++; | |
5091 | } | |
5092 | if (in->is_head() && | |
5093 | in->issued_caps_need_gather(lock)) { | |
5094 | if (need_issue) | |
5095 | *need_issue = true; | |
5096 | else | |
5097 | issue_caps(in); | |
5098 | gather++; | |
5099 | } | |
5100 | bool need_recover = false; | |
5101 | if (in->state_test(CInode::STATE_NEEDSRECOVER)) { | |
5102 | mds->mdcache->queue_file_recover(in); | |
5103 | need_recover = true; | |
5104 | gather++; | |
5105 | } | |
5106 | ||
5107 | if (gather) { | |
5108 | lock->get_parent()->auth_pin(lock); | |
5109 | if (need_recover) | |
5110 | mds->mdcache->do_file_recover(); | |
5111 | } else { | |
5112 | lock->set_state(LOCK_EXCL); | |
5113 | if (need_issue) | |
5114 | *need_issue = true; | |
5115 | else | |
5116 | issue_caps(in); | |
5117 | } | |
5118 | } | |
5119 | ||
5120 | void Locker::file_xsyn(SimpleLock *lock, bool *need_issue) | |
5121 | { | |
5122 | dout(7) << "file_xsyn on " << *lock << " on " << *lock->get_parent() << dendl; | |
5123 | CInode *in = static_cast<CInode *>(lock->get_parent()); | |
11fdf7f2 TL |
5124 | ceph_assert(in->is_auth()); |
5125 | ceph_assert(in->get_loner() >= 0 && in->get_mds_caps_wanted().empty()); | |
7c673cae FG |
5126 | |
5127 | switch (lock->get_state()) { | |
5128 | case LOCK_EXCL: lock->set_state(LOCK_EXCL_XSYN); break; | |
5129 | default: ceph_abort(); | |
5130 | } | |
5131 | ||
5132 | int gather = 0; | |
5133 | if (lock->is_wrlocked()) | |
5134 | gather++; | |
5135 | ||
5136 | if (in->is_head() && | |
5137 | in->issued_caps_need_gather(lock)) { | |
5138 | if (need_issue) | |
5139 | *need_issue = true; | |
5140 | else | |
5141 | issue_caps(in); | |
5142 | gather++; | |
5143 | } | |
5144 | ||
5145 | if (gather) { | |
5146 | lock->get_parent()->auth_pin(lock); | |
5147 | } else { | |
5148 | lock->set_state(LOCK_XSYN); | |
5149 | lock->finish_waiters(SimpleLock::WAIT_RD|SimpleLock::WAIT_STABLE); | |
5150 | if (need_issue) | |
5151 | *need_issue = true; | |
5152 | else | |
5153 | issue_caps(in); | |
5154 | } | |
5155 | } | |
5156 | ||
5157 | void Locker::file_recover(ScatterLock *lock) | |
5158 | { | |
5159 | CInode *in = static_cast<CInode *>(lock->get_parent()); | |
5160 | dout(7) << "file_recover " << *lock << " on " << *in << dendl; | |
5161 | ||
11fdf7f2 | 5162 | ceph_assert(in->is_auth()); |
7c673cae | 5163 | //assert(lock->is_stable()); |
11fdf7f2 | 5164 | ceph_assert(lock->get_state() == LOCK_PRE_SCAN); // only called from MDCache::start_files_to_recover() |
7c673cae FG |
5165 | |
5166 | int gather = 0; | |
5167 | ||
5168 | /* | |
5169 | if (in->is_replicated() | |
5170 | lock->get_sm()->states[oldstate].replica_state != LOCK_LOCK) { | |
5171 | send_lock_message(lock, LOCK_AC_LOCK); | |
5172 | lock->init_gather(); | |
5173 | gather++; | |
5174 | } | |
5175 | */ | |
5176 | if (in->is_head() && | |
5177 | in->issued_caps_need_gather(lock)) { | |
5178 | issue_caps(in); | |
5179 | gather++; | |
5180 | } | |
5181 | ||
5182 | lock->set_state(LOCK_SCAN); | |
5183 | if (gather) | |
5184 | in->state_set(CInode::STATE_NEEDSRECOVER); | |
5185 | else | |
5186 | mds->mdcache->queue_file_recover(in); | |
5187 | } | |
5188 | ||
5189 | ||
5190 | // messenger | |
11fdf7f2 | 5191 | void Locker::handle_file_lock(ScatterLock *lock, const MLock::const_ref &m) |
7c673cae FG |
5192 | { |
5193 | CInode *in = static_cast<CInode*>(lock->get_parent()); | |
5194 | int from = m->get_asker(); | |
5195 | ||
5196 | if (mds->is_rejoin()) { | |
5197 | if (in->is_rejoining()) { | |
5198 | dout(7) << "handle_file_lock still rejoining " << *in | |
5199 | << ", dropping " << *m << dendl; | |
7c673cae FG |
5200 | return; |
5201 | } | |
5202 | } | |
5203 | ||
11fdf7f2 | 5204 | dout(7) << "handle_file_lock a=" << lock->get_lock_action_name(m->get_action()) |
7c673cae FG |
5205 | << " on " << *lock |
5206 | << " from mds." << from << " " | |
5207 | << *in << dendl; | |
5208 | ||
5209 | bool caps = lock->get_cap_shift(); | |
5210 | ||
5211 | switch (m->get_action()) { | |
5212 | // -- replica -- | |
5213 | case LOCK_AC_SYNC: | |
11fdf7f2 | 5214 | ceph_assert(lock->get_state() == LOCK_LOCK || |
7c673cae FG |
5215 | lock->get_state() == LOCK_MIX || |
5216 | lock->get_state() == LOCK_MIX_SYNC2); | |
5217 | ||
5218 | if (lock->get_state() == LOCK_MIX) { | |
5219 | lock->set_state(LOCK_MIX_SYNC); | |
5220 | eval_gather(lock, true); | |
31f18b77 FG |
5221 | if (lock->is_unstable_and_locked()) |
5222 | mds->mdlog->flush(); | |
7c673cae FG |
5223 | break; |
5224 | } | |
5225 | ||
5226 | (static_cast<ScatterLock *>(lock))->finish_flush(); | |
5227 | (static_cast<ScatterLock *>(lock))->clear_flushed(); | |
5228 | ||
5229 | // ok | |
5230 | lock->decode_locked_state(m->get_data()); | |
5231 | lock->set_state(LOCK_SYNC); | |
5232 | ||
5233 | lock->get_rdlock(); | |
5234 | if (caps) | |
5235 | issue_caps(in); | |
5236 | lock->finish_waiters(SimpleLock::WAIT_RD|SimpleLock::WAIT_STABLE); | |
5237 | lock->put_rdlock(); | |
5238 | break; | |
5239 | ||
5240 | case LOCK_AC_LOCK: | |
5241 | switch (lock->get_state()) { | |
5242 | case LOCK_SYNC: lock->set_state(LOCK_SYNC_LOCK); break; | |
5243 | case LOCK_MIX: lock->set_state(LOCK_MIX_LOCK); break; | |
5244 | default: ceph_abort(); | |
5245 | } | |
5246 | ||
5247 | eval_gather(lock, true); | |
31f18b77 FG |
5248 | if (lock->is_unstable_and_locked()) |
5249 | mds->mdlog->flush(); | |
5250 | ||
7c673cae FG |
5251 | break; |
5252 | ||
5253 | case LOCK_AC_LOCKFLUSHED: | |
5254 | (static_cast<ScatterLock *>(lock))->finish_flush(); | |
5255 | (static_cast<ScatterLock *>(lock))->clear_flushed(); | |
5256 | // wake up scatter_nudge waiters | |
5257 | if (lock->is_stable()) | |
5258 | lock->finish_waiters(SimpleLock::WAIT_STABLE); | |
5259 | break; | |
5260 | ||
5261 | case LOCK_AC_MIX: | |
11fdf7f2 | 5262 | ceph_assert(lock->get_state() == LOCK_SYNC || |
7c673cae FG |
5263 | lock->get_state() == LOCK_LOCK || |
5264 | lock->get_state() == LOCK_SYNC_MIX2); | |
5265 | ||
5266 | if (lock->get_state() == LOCK_SYNC) { | |
5267 | // MIXED | |
5268 | lock->set_state(LOCK_SYNC_MIX); | |
5269 | eval_gather(lock, true); | |
31f18b77 FG |
5270 | if (lock->is_unstable_and_locked()) |
5271 | mds->mdlog->flush(); | |
7c673cae FG |
5272 | break; |
5273 | } | |
c07f9fc5 | 5274 | |
7c673cae | 5275 | // ok |
7c673cae | 5276 | lock->set_state(LOCK_MIX); |
c07f9fc5 | 5277 | lock->decode_locked_state(m->get_data()); |
7c673cae FG |
5278 | |
5279 | if (caps) | |
5280 | issue_caps(in); | |
5281 | ||
5282 | lock->finish_waiters(SimpleLock::WAIT_WR|SimpleLock::WAIT_STABLE); | |
5283 | break; | |
5284 | ||
5285 | ||
5286 | // -- auth -- | |
5287 | case LOCK_AC_LOCKACK: | |
11fdf7f2 | 5288 | ceph_assert(lock->get_state() == LOCK_SYNC_LOCK || |
7c673cae FG |
5289 | lock->get_state() == LOCK_MIX_LOCK || |
5290 | lock->get_state() == LOCK_MIX_LOCK2 || | |
5291 | lock->get_state() == LOCK_MIX_EXCL || | |
5292 | lock->get_state() == LOCK_SYNC_EXCL || | |
5293 | lock->get_state() == LOCK_SYNC_MIX || | |
5294 | lock->get_state() == LOCK_MIX_TSYN); | |
11fdf7f2 | 5295 | ceph_assert(lock->is_gathering(from)); |
7c673cae FG |
5296 | lock->remove_gather(from); |
5297 | ||
5298 | if (lock->get_state() == LOCK_MIX_LOCK || | |
5299 | lock->get_state() == LOCK_MIX_LOCK2 || | |
5300 | lock->get_state() == LOCK_MIX_EXCL || | |
5301 | lock->get_state() == LOCK_MIX_TSYN) { | |
5302 | lock->decode_locked_state(m->get_data()); | |
5303 | // replica is waiting for AC_LOCKFLUSHED, eval_gather() should not | |
5304 | // delay calling scatter_writebehind(). | |
5305 | lock->clear_flushed(); | |
5306 | } | |
5307 | ||
5308 | if (lock->is_gathering()) { | |
5309 | dout(7) << "handle_file_lock " << *in << " from " << from | |
5310 | << ", still gathering " << lock->get_gather_set() << dendl; | |
5311 | } else { | |
5312 | dout(7) << "handle_file_lock " << *in << " from " << from | |
5313 | << ", last one" << dendl; | |
5314 | eval_gather(lock); | |
5315 | } | |
5316 | break; | |
5317 | ||
5318 | case LOCK_AC_SYNCACK: | |
11fdf7f2 TL |
5319 | ceph_assert(lock->get_state() == LOCK_MIX_SYNC); |
5320 | ceph_assert(lock->is_gathering(from)); | |
7c673cae FG |
5321 | lock->remove_gather(from); |
5322 | ||
5323 | lock->decode_locked_state(m->get_data()); | |
5324 | ||
5325 | if (lock->is_gathering()) { | |
5326 | dout(7) << "handle_file_lock " << *in << " from " << from | |
5327 | << ", still gathering " << lock->get_gather_set() << dendl; | |
5328 | } else { | |
5329 | dout(7) << "handle_file_lock " << *in << " from " << from | |
5330 | << ", last one" << dendl; | |
5331 | eval_gather(lock); | |
5332 | } | |
5333 | break; | |
5334 | ||
5335 | case LOCK_AC_MIXACK: | |
11fdf7f2 TL |
5336 | ceph_assert(lock->get_state() == LOCK_SYNC_MIX); |
5337 | ceph_assert(lock->is_gathering(from)); | |
7c673cae FG |
5338 | lock->remove_gather(from); |
5339 | ||
5340 | if (lock->is_gathering()) { | |
5341 | dout(7) << "handle_file_lock " << *in << " from " << from | |
5342 | << ", still gathering " << lock->get_gather_set() << dendl; | |
5343 | } else { | |
5344 | dout(7) << "handle_file_lock " << *in << " from " << from | |
5345 | << ", last one" << dendl; | |
5346 | eval_gather(lock); | |
5347 | } | |
5348 | break; | |
5349 | ||
5350 | ||
5351 | // requests.... | |
5352 | case LOCK_AC_REQSCATTER: | |
5353 | if (lock->is_stable()) { | |
5354 | /* NOTE: we can do this _even_ if !can_auth_pin (i.e. freezing) | |
5355 | * because the replica should be holding an auth_pin if they're | |
5356 | * doing this (and thus, we are freezing, not frozen, and indefinite | |
5357 | * starvation isn't an issue). | |
5358 | */ | |
5359 | dout(7) << "handle_file_lock got scatter request on " << *lock | |
5360 | << " on " << *lock->get_parent() << dendl; | |
5361 | if (lock->get_state() != LOCK_MIX) // i.e., the reqscatter didn't race with an actual mix/scatter | |
5362 | scatter_mix(lock); | |
5363 | } else { | |
5364 | dout(7) << "handle_file_lock got scatter request, !stable, marking scatter_wanted on " << *lock | |
5365 | << " on " << *lock->get_parent() << dendl; | |
5366 | lock->set_scatter_wanted(); | |
5367 | } | |
5368 | break; | |
5369 | ||
5370 | case LOCK_AC_REQUNSCATTER: | |
5371 | if (lock->is_stable()) { | |
5372 | /* NOTE: we can do this _even_ if !can_auth_pin (i.e. freezing) | |
5373 | * because the replica should be holding an auth_pin if they're | |
5374 | * doing this (and thus, we are freezing, not frozen, and indefinite | |
5375 | * starvation isn't an issue). | |
5376 | */ | |
5377 | dout(7) << "handle_file_lock got unscatter request on " << *lock | |
5378 | << " on " << *lock->get_parent() << dendl; | |
5379 | if (lock->get_state() == LOCK_MIX) // i.e., the reqscatter didn't race with an actual mix/scatter | |
5380 | simple_lock(lock); // FIXME tempsync? | |
5381 | } else { | |
5382 | dout(7) << "handle_file_lock ignoring unscatter request on " << *lock | |
5383 | << " on " << *lock->get_parent() << dendl; | |
5384 | lock->set_unscatter_wanted(); | |
5385 | } | |
5386 | break; | |
5387 | ||
5388 | case LOCK_AC_REQRDLOCK: | |
5389 | handle_reqrdlock(lock, m); | |
5390 | break; | |
5391 | ||
5392 | case LOCK_AC_NUDGE: | |
5393 | if (!lock->get_parent()->is_auth()) { | |
5394 | dout(7) << "handle_file_lock IGNORING nudge on non-auth " << *lock | |
5395 | << " on " << *lock->get_parent() << dendl; | |
5396 | } else if (!lock->get_parent()->is_replicated()) { | |
5397 | dout(7) << "handle_file_lock IGNORING nudge on non-replicated " << *lock | |
5398 | << " on " << *lock->get_parent() << dendl; | |
5399 | } else { | |
5400 | dout(7) << "handle_file_lock trying nudge on " << *lock | |
5401 | << " on " << *lock->get_parent() << dendl; | |
5402 | scatter_nudge(lock, 0, true); | |
5403 | mds->mdlog->flush(); | |
5404 | } | |
5405 | break; | |
5406 | ||
5407 | default: | |
5408 | ceph_abort(); | |
5409 | } | |
7c673cae | 5410 | } |