]>
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 | ||
15 | ||
16 | // unix-ey fs stuff | |
17 | #include <unistd.h> | |
18 | #include <sys/types.h> | |
19 | #include <time.h> | |
20 | #include <utime.h> | |
21 | #include <sys/stat.h> | |
22 | #include <sys/param.h> | |
23 | #include <fcntl.h> | |
24 | #include <sys/file.h> | |
25 | #include <sys/utsname.h> | |
26 | #include <sys/uio.h> | |
27 | ||
28 | #include <boost/lexical_cast.hpp> | |
29 | #include <boost/fusion/include/std_pair.hpp> | |
30 | ||
31 | #if defined(__FreeBSD__) | |
32 | #define XATTR_CREATE 0x1 | |
33 | #define XATTR_REPLACE 0x2 | |
34 | #else | |
35 | #include <sys/xattr.h> | |
36 | #endif | |
37 | ||
38 | #if defined(__linux__) | |
39 | #include <linux/falloc.h> | |
40 | #endif | |
41 | ||
42 | #include <sys/statvfs.h> | |
43 | ||
44 | #include "common/config.h" | |
45 | #include "common/version.h" | |
46 | ||
47 | // ceph stuff | |
48 | #include "messages/MClientSession.h" | |
49 | #include "messages/MClientReconnect.h" | |
50 | #include "messages/MClientRequest.h" | |
51 | #include "messages/MClientRequestForward.h" | |
52 | #include "messages/MClientReply.h" | |
53 | #include "messages/MClientCaps.h" | |
54 | #include "messages/MClientLease.h" | |
55 | #include "messages/MClientSnap.h" | |
56 | #include "messages/MCommandReply.h" | |
57 | #include "messages/MOSDMap.h" | |
58 | #include "messages/MClientQuota.h" | |
59 | #include "messages/MClientCapRelease.h" | |
60 | #include "messages/MMDSMap.h" | |
61 | #include "messages/MFSMap.h" | |
62 | #include "messages/MFSMapUser.h" | |
63 | ||
64 | #include "mon/MonClient.h" | |
65 | ||
66 | #include "mds/flock.h" | |
67 | #include "osd/OSDMap.h" | |
68 | #include "osdc/Filer.h" | |
69 | ||
70 | #include "common/Cond.h" | |
71 | #include "common/Mutex.h" | |
72 | #include "common/perf_counters.h" | |
73 | #include "common/admin_socket.h" | |
74 | #include "common/errno.h" | |
75 | #include "include/str_list.h" | |
76 | ||
77 | #define dout_subsys ceph_subsys_client | |
78 | ||
79 | #include "include/lru.h" | |
80 | #include "include/compat.h" | |
81 | #include "include/stringify.h" | |
82 | ||
83 | #include "Client.h" | |
84 | #include "Inode.h" | |
85 | #include "Dentry.h" | |
b32b8144 | 86 | #include "Delegation.h" |
7c673cae FG |
87 | #include "Dir.h" |
88 | #include "ClientSnapRealm.h" | |
89 | #include "Fh.h" | |
90 | #include "MetaSession.h" | |
91 | #include "MetaRequest.h" | |
92 | #include "ObjecterWriteback.h" | |
93 | #include "posix_acl.h" | |
94 | ||
95 | #include "include/assert.h" | |
96 | #include "include/stat.h" | |
97 | ||
98 | #include "include/cephfs/ceph_statx.h" | |
99 | ||
100 | #if HAVE_GETGROUPLIST | |
101 | #include <grp.h> | |
102 | #include <pwd.h> | |
103 | #include <unistd.h> | |
104 | #endif | |
105 | ||
106 | #undef dout_prefix | |
107 | #define dout_prefix *_dout << "client." << whoami << " " | |
108 | ||
109 | #define tout(cct) if (!cct->_conf->client_trace.empty()) traceout | |
110 | ||
111 | // FreeBSD fails to define this | |
112 | #ifndef O_DSYNC | |
113 | #define O_DSYNC 0x0 | |
114 | #endif | |
115 | // Darwin fails to define this | |
116 | #ifndef O_RSYNC | |
117 | #define O_RSYNC 0x0 | |
118 | #endif | |
119 | ||
120 | #ifndef O_DIRECT | |
121 | #define O_DIRECT 0x0 | |
122 | #endif | |
123 | ||
124 | #define DEBUG_GETATTR_CAPS (CEPH_CAP_XATTR_SHARED) | |
125 | ||
126 | void client_flush_set_callback(void *p, ObjectCacher::ObjectSet *oset) | |
127 | { | |
128 | Client *client = static_cast<Client*>(p); | |
129 | client->flush_set_callback(oset); | |
130 | } | |
131 | ||
132 | ||
133 | // ------------- | |
134 | ||
135 | Client::CommandHook::CommandHook(Client *client) : | |
136 | m_client(client) | |
137 | { | |
138 | } | |
139 | ||
140 | bool Client::CommandHook::call(std::string command, cmdmap_t& cmdmap, | |
141 | std::string format, bufferlist& out) | |
142 | { | |
143 | Formatter *f = Formatter::create(format); | |
144 | f->open_object_section("result"); | |
145 | m_client->client_lock.Lock(); | |
146 | if (command == "mds_requests") | |
147 | m_client->dump_mds_requests(f); | |
148 | else if (command == "mds_sessions") | |
149 | m_client->dump_mds_sessions(f); | |
150 | else if (command == "dump_cache") | |
151 | m_client->dump_cache(f); | |
152 | else if (command == "kick_stale_sessions") | |
153 | m_client->_kick_stale_sessions(); | |
154 | else if (command == "status") | |
155 | m_client->dump_status(f); | |
156 | else | |
157 | assert(0 == "bad command registered"); | |
158 | m_client->client_lock.Unlock(); | |
159 | f->close_section(); | |
160 | f->flush(out); | |
161 | delete f; | |
162 | return true; | |
163 | } | |
164 | ||
165 | ||
166 | // ------------- | |
167 | ||
168 | dir_result_t::dir_result_t(Inode *in, const UserPerm& perms) | |
169 | : inode(in), offset(0), next_offset(2), | |
170 | release_count(0), ordered_count(0), cache_index(0), start_shared_gen(0), | |
171 | perms(perms) | |
172 | { } | |
173 | ||
174 | void Client::_reset_faked_inos() | |
175 | { | |
176 | ino_t start = 1024; | |
177 | free_faked_inos.clear(); | |
178 | free_faked_inos.insert(start, (uint32_t)-1 - start + 1); | |
179 | last_used_faked_ino = 0; | |
180 | _use_faked_inos = sizeof(ino_t) < 8 || cct->_conf->client_use_faked_inos; | |
181 | } | |
182 | ||
183 | void Client::_assign_faked_ino(Inode *in) | |
184 | { | |
185 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_ino + 1); | |
186 | if (it == free_faked_inos.end() && last_used_faked_ino > 0) { | |
187 | last_used_faked_ino = 0; | |
188 | it = free_faked_inos.lower_bound(last_used_faked_ino + 1); | |
189 | } | |
190 | assert(it != free_faked_inos.end()); | |
191 | if (last_used_faked_ino < it.get_start()) { | |
192 | assert(it.get_len() > 0); | |
193 | last_used_faked_ino = it.get_start(); | |
194 | } else { | |
195 | ++last_used_faked_ino; | |
196 | assert(it.get_start() + it.get_len() > last_used_faked_ino); | |
197 | } | |
198 | in->faked_ino = last_used_faked_ino; | |
199 | free_faked_inos.erase(in->faked_ino); | |
200 | faked_ino_map[in->faked_ino] = in->vino(); | |
201 | } | |
202 | ||
203 | void Client::_release_faked_ino(Inode *in) | |
204 | { | |
205 | free_faked_inos.insert(in->faked_ino); | |
206 | faked_ino_map.erase(in->faked_ino); | |
207 | } | |
208 | ||
209 | vinodeno_t Client::_map_faked_ino(ino_t ino) | |
210 | { | |
211 | vinodeno_t vino; | |
212 | if (ino == 1) | |
213 | vino = root->vino(); | |
214 | else if (faked_ino_map.count(ino)) | |
215 | vino = faked_ino_map[ino]; | |
216 | else | |
217 | vino = vinodeno_t(0, CEPH_NOSNAP); | |
218 | ldout(cct, 10) << "map_faked_ino " << ino << " -> " << vino << dendl; | |
219 | return vino; | |
220 | } | |
221 | ||
222 | vinodeno_t Client::map_faked_ino(ino_t ino) | |
223 | { | |
224 | Mutex::Locker lock(client_lock); | |
225 | return _map_faked_ino(ino); | |
226 | } | |
227 | ||
228 | // cons/des | |
229 | ||
230 | Client::Client(Messenger *m, MonClient *mc, Objecter *objecter_) | |
231 | : Dispatcher(m->cct), | |
232 | m_command_hook(this), | |
233 | timer(m->cct, client_lock), | |
234 | callback_handle(NULL), | |
235 | switch_interrupt_cb(NULL), | |
236 | remount_cb(NULL), | |
237 | ino_invalidate_cb(NULL), | |
238 | dentry_invalidate_cb(NULL), | |
239 | getgroups_cb(NULL), | |
240 | umask_cb(NULL), | |
241 | can_invalidate_dentries(false), | |
7c673cae FG |
242 | async_ino_invalidator(m->cct), |
243 | async_dentry_invalidator(m->cct), | |
244 | interrupt_finisher(m->cct), | |
245 | remount_finisher(m->cct), | |
246 | objecter_finisher(m->cct), | |
247 | tick_event(NULL), | |
248 | messenger(m), monclient(mc), | |
249 | objecter(objecter_), | |
250 | whoami(mc->get_global_id()), cap_epoch_barrier(0), | |
251 | last_tid(0), oldest_tid(0), last_flush_tid(1), | |
252 | initialized(false), | |
31f18b77 | 253 | mounted(false), unmounting(false), blacklisted(false), |
b32b8144 | 254 | local_osd(-ENXIO), local_osd_epoch(0), |
7c673cae | 255 | unsafe_sync_write(0), |
b32b8144 FG |
256 | client_lock("Client::client_lock"), |
257 | deleg_timeout(0) | |
7c673cae FG |
258 | { |
259 | _reset_faked_inos(); | |
260 | // | |
261 | root = 0; | |
262 | ||
263 | num_flushing_caps = 0; | |
264 | ||
265 | _dir_vxattrs_name_size = _vxattrs_calcu_name_size(_dir_vxattrs); | |
266 | _file_vxattrs_name_size = _vxattrs_calcu_name_size(_file_vxattrs); | |
267 | ||
268 | user_id = cct->_conf->client_mount_uid; | |
269 | group_id = cct->_conf->client_mount_gid; | |
270 | ||
271 | acl_type = NO_ACL; | |
272 | if (cct->_conf->client_acl_type == "posix_acl") | |
273 | acl_type = POSIX_ACL; | |
274 | ||
7c673cae FG |
275 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
276 | ||
277 | // file handles | |
278 | free_fd_set.insert(10, 1<<30); | |
279 | ||
280 | mdsmap.reset(new MDSMap); | |
281 | ||
282 | // osd interfaces | |
283 | writeback_handler.reset(new ObjecterWriteback(objecter, &objecter_finisher, | |
284 | &client_lock)); | |
285 | objectcacher.reset(new ObjectCacher(cct, "libcephfs", *writeback_handler, client_lock, | |
286 | client_flush_set_callback, // all commit callback | |
287 | (void*)this, | |
288 | cct->_conf->client_oc_size, | |
289 | cct->_conf->client_oc_max_objects, | |
290 | cct->_conf->client_oc_max_dirty, | |
291 | cct->_conf->client_oc_target_dirty, | |
292 | cct->_conf->client_oc_max_dirty_age, | |
293 | true)); | |
294 | objecter_finisher.start(); | |
295 | filer.reset(new Filer(objecter, &objecter_finisher)); | |
31f18b77 | 296 | objecter->enable_blacklist_events(); |
7c673cae FG |
297 | } |
298 | ||
299 | ||
300 | Client::~Client() | |
301 | { | |
302 | assert(!client_lock.is_locked()); | |
303 | ||
31f18b77 FG |
304 | // It is necessary to hold client_lock, because any inode destruction |
305 | // may call into ObjectCacher, which asserts that it's lock (which is | |
306 | // client_lock) is held. | |
307 | client_lock.Lock(); | |
7c673cae | 308 | tear_down_cache(); |
31f18b77 | 309 | client_lock.Unlock(); |
7c673cae FG |
310 | } |
311 | ||
312 | void Client::tear_down_cache() | |
313 | { | |
314 | // fd's | |
315 | for (ceph::unordered_map<int, Fh*>::iterator it = fd_map.begin(); | |
316 | it != fd_map.end(); | |
317 | ++it) { | |
318 | Fh *fh = it->second; | |
319 | ldout(cct, 1) << "tear_down_cache forcing close of fh " << it->first << " ino " << fh->inode->ino << dendl; | |
320 | _release_fh(fh); | |
321 | } | |
322 | fd_map.clear(); | |
323 | ||
324 | while (!opened_dirs.empty()) { | |
325 | dir_result_t *dirp = *opened_dirs.begin(); | |
326 | ldout(cct, 1) << "tear_down_cache forcing close of dir " << dirp << " ino " << dirp->inode->ino << dendl; | |
327 | _closedir(dirp); | |
328 | } | |
329 | ||
330 | // caps! | |
331 | // *** FIXME *** | |
332 | ||
333 | // empty lru | |
7c673cae FG |
334 | trim_cache(); |
335 | assert(lru.lru_get_size() == 0); | |
336 | ||
337 | // close root ino | |
338 | assert(inode_map.size() <= 1 + root_parents.size()); | |
339 | if (root && inode_map.size() == 1 + root_parents.size()) { | |
340 | delete root; | |
341 | root = 0; | |
342 | root_ancestor = 0; | |
343 | while (!root_parents.empty()) | |
344 | root_parents.erase(root_parents.begin()); | |
345 | inode_map.clear(); | |
346 | _reset_faked_inos(); | |
347 | } | |
348 | ||
349 | assert(inode_map.empty()); | |
350 | } | |
351 | ||
352 | inodeno_t Client::get_root_ino() | |
353 | { | |
354 | Mutex::Locker l(client_lock); | |
355 | if (use_faked_inos()) | |
356 | return root->faked_ino; | |
357 | else | |
358 | return root->ino; | |
359 | } | |
360 | ||
361 | Inode *Client::get_root() | |
362 | { | |
363 | Mutex::Locker l(client_lock); | |
364 | root->ll_get(); | |
365 | return root; | |
366 | } | |
367 | ||
368 | ||
369 | // debug crapola | |
370 | ||
371 | void Client::dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected) | |
372 | { | |
373 | filepath path; | |
374 | in->make_long_path(path); | |
375 | ldout(cct, 1) << "dump_inode: " | |
376 | << (disconnected ? "DISCONNECTED ":"") | |
377 | << "inode " << in->ino | |
378 | << " " << path | |
379 | << " ref " << in->get_num_ref() | |
380 | << *in << dendl; | |
381 | ||
382 | if (f) { | |
383 | f->open_object_section("inode"); | |
384 | f->dump_stream("path") << path; | |
385 | if (disconnected) | |
386 | f->dump_int("disconnected", 1); | |
387 | in->dump(f); | |
388 | f->close_section(); | |
389 | } | |
390 | ||
391 | did.insert(in); | |
392 | if (in->dir) { | |
393 | ldout(cct, 1) << " dir " << in->dir << " size " << in->dir->dentries.size() << dendl; | |
394 | for (ceph::unordered_map<string, Dentry*>::iterator it = in->dir->dentries.begin(); | |
395 | it != in->dir->dentries.end(); | |
396 | ++it) { | |
397 | ldout(cct, 1) << " " << in->ino << " dn " << it->first << " " << it->second << " ref " << it->second->ref << dendl; | |
398 | if (f) { | |
399 | f->open_object_section("dentry"); | |
400 | it->second->dump(f); | |
401 | f->close_section(); | |
402 | } | |
403 | if (it->second->inode) | |
404 | dump_inode(f, it->second->inode.get(), did, false); | |
405 | } | |
406 | } | |
407 | } | |
408 | ||
409 | void Client::dump_cache(Formatter *f) | |
410 | { | |
411 | set<Inode*> did; | |
412 | ||
413 | ldout(cct, 1) << "dump_cache" << dendl; | |
414 | ||
415 | if (f) | |
416 | f->open_array_section("cache"); | |
417 | ||
418 | if (root) | |
419 | dump_inode(f, root, did, true); | |
420 | ||
421 | // make a second pass to catch anything disconnected | |
422 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
423 | it != inode_map.end(); | |
424 | ++it) { | |
425 | if (did.count(it->second)) | |
426 | continue; | |
427 | dump_inode(f, it->second, did, true); | |
428 | } | |
429 | ||
430 | if (f) | |
431 | f->close_section(); | |
432 | } | |
433 | ||
434 | void Client::dump_status(Formatter *f) | |
435 | { | |
436 | assert(client_lock.is_locked_by_me()); | |
437 | ||
438 | ldout(cct, 1) << __func__ << dendl; | |
439 | ||
440 | const epoch_t osd_epoch | |
441 | = objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
442 | ||
443 | if (f) { | |
444 | f->open_object_section("metadata"); | |
445 | for (const auto& kv : metadata) | |
446 | f->dump_string(kv.first.c_str(), kv.second); | |
447 | f->close_section(); | |
448 | ||
449 | f->dump_int("dentry_count", lru.lru_get_size()); | |
450 | f->dump_int("dentry_pinned_count", lru.lru_get_num_pinned()); | |
451 | f->dump_int("id", get_nodeid().v); | |
452 | f->dump_int("inode_count", inode_map.size()); | |
453 | f->dump_int("mds_epoch", mdsmap->get_epoch()); | |
454 | f->dump_int("osd_epoch", osd_epoch); | |
455 | f->dump_int("osd_epoch_barrier", cap_epoch_barrier); | |
456 | } | |
457 | } | |
458 | ||
459 | int Client::init() | |
460 | { | |
461 | timer.init(); | |
462 | objectcacher->start(); | |
463 | ||
464 | client_lock.Lock(); | |
465 | assert(!initialized); | |
466 | ||
467 | messenger->add_dispatcher_tail(this); | |
468 | client_lock.Unlock(); | |
469 | ||
470 | _finish_init(); | |
471 | return 0; | |
472 | } | |
473 | ||
474 | void Client::_finish_init() | |
475 | { | |
476 | client_lock.Lock(); | |
477 | // logger | |
478 | PerfCountersBuilder plb(cct, "client", l_c_first, l_c_last); | |
479 | plb.add_time_avg(l_c_reply, "reply", "Latency of receiving a reply on metadata request"); | |
480 | plb.add_time_avg(l_c_lat, "lat", "Latency of processing a metadata request"); | |
481 | plb.add_time_avg(l_c_wrlat, "wrlat", "Latency of a file data write operation"); | |
482 | logger.reset(plb.create_perf_counters()); | |
483 | cct->get_perfcounters_collection()->add(logger.get()); | |
484 | ||
485 | client_lock.Unlock(); | |
486 | ||
487 | cct->_conf->add_observer(this); | |
488 | ||
489 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
490 | int ret = admin_socket->register_command("mds_requests", | |
491 | "mds_requests", | |
492 | &m_command_hook, | |
493 | "show in-progress mds requests"); | |
494 | if (ret < 0) { | |
495 | lderr(cct) << "error registering admin socket command: " | |
496 | << cpp_strerror(-ret) << dendl; | |
497 | } | |
498 | ret = admin_socket->register_command("mds_sessions", | |
499 | "mds_sessions", | |
500 | &m_command_hook, | |
501 | "show mds session state"); | |
502 | if (ret < 0) { | |
503 | lderr(cct) << "error registering admin socket command: " | |
504 | << cpp_strerror(-ret) << dendl; | |
505 | } | |
506 | ret = admin_socket->register_command("dump_cache", | |
507 | "dump_cache", | |
508 | &m_command_hook, | |
509 | "show in-memory metadata cache contents"); | |
510 | if (ret < 0) { | |
511 | lderr(cct) << "error registering admin socket command: " | |
512 | << cpp_strerror(-ret) << dendl; | |
513 | } | |
514 | ret = admin_socket->register_command("kick_stale_sessions", | |
515 | "kick_stale_sessions", | |
516 | &m_command_hook, | |
517 | "kick sessions that were remote reset"); | |
518 | if (ret < 0) { | |
519 | lderr(cct) << "error registering admin socket command: " | |
520 | << cpp_strerror(-ret) << dendl; | |
521 | } | |
522 | ret = admin_socket->register_command("status", | |
523 | "status", | |
524 | &m_command_hook, | |
525 | "show overall client status"); | |
526 | if (ret < 0) { | |
527 | lderr(cct) << "error registering admin socket command: " | |
528 | << cpp_strerror(-ret) << dendl; | |
529 | } | |
530 | ||
531 | client_lock.Lock(); | |
532 | initialized = true; | |
533 | client_lock.Unlock(); | |
534 | } | |
535 | ||
536 | void Client::shutdown() | |
537 | { | |
538 | ldout(cct, 1) << "shutdown" << dendl; | |
539 | ||
540 | // If we were not mounted, but were being used for sending | |
541 | // MDS commands, we may have sessions that need closing. | |
542 | client_lock.Lock(); | |
543 | _close_sessions(); | |
544 | client_lock.Unlock(); | |
545 | ||
546 | cct->_conf->remove_observer(this); | |
547 | ||
548 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
549 | admin_socket->unregister_command("mds_requests"); | |
550 | admin_socket->unregister_command("mds_sessions"); | |
551 | admin_socket->unregister_command("dump_cache"); | |
552 | admin_socket->unregister_command("kick_stale_sessions"); | |
553 | admin_socket->unregister_command("status"); | |
554 | ||
555 | if (ino_invalidate_cb) { | |
556 | ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl; | |
557 | async_ino_invalidator.wait_for_empty(); | |
558 | async_ino_invalidator.stop(); | |
559 | } | |
560 | ||
561 | if (dentry_invalidate_cb) { | |
562 | ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl; | |
563 | async_dentry_invalidator.wait_for_empty(); | |
564 | async_dentry_invalidator.stop(); | |
565 | } | |
566 | ||
567 | if (switch_interrupt_cb) { | |
568 | ldout(cct, 10) << "shutdown stopping interrupt finisher" << dendl; | |
569 | interrupt_finisher.wait_for_empty(); | |
570 | interrupt_finisher.stop(); | |
571 | } | |
572 | ||
573 | if (remount_cb) { | |
574 | ldout(cct, 10) << "shutdown stopping remount finisher" << dendl; | |
575 | remount_finisher.wait_for_empty(); | |
576 | remount_finisher.stop(); | |
577 | } | |
578 | ||
579 | objectcacher->stop(); // outside of client_lock! this does a join. | |
580 | ||
581 | client_lock.Lock(); | |
582 | assert(initialized); | |
583 | initialized = false; | |
584 | timer.shutdown(); | |
585 | client_lock.Unlock(); | |
586 | ||
587 | objecter_finisher.wait_for_empty(); | |
588 | objecter_finisher.stop(); | |
589 | ||
590 | if (logger) { | |
591 | cct->get_perfcounters_collection()->remove(logger.get()); | |
592 | logger.reset(); | |
593 | } | |
594 | } | |
595 | ||
596 | ||
597 | // =================== | |
598 | // metadata cache stuff | |
599 | ||
600 | void Client::trim_cache(bool trim_kernel_dcache) | |
601 | { | |
181888fb FG |
602 | uint64_t max = cct->_conf->client_cache_size; |
603 | ldout(cct, 20) << "trim_cache size " << lru.lru_get_size() << " max " << max << dendl; | |
7c673cae FG |
604 | unsigned last = 0; |
605 | while (lru.lru_get_size() != last) { | |
606 | last = lru.lru_get_size(); | |
607 | ||
181888fb | 608 | if (!unmounting && lru.lru_get_size() <= max) break; |
7c673cae FG |
609 | |
610 | // trim! | |
31f18b77 | 611 | Dentry *dn = static_cast<Dentry*>(lru.lru_get_next_expire()); |
7c673cae FG |
612 | if (!dn) |
613 | break; // done | |
614 | ||
615 | trim_dentry(dn); | |
616 | } | |
617 | ||
181888fb | 618 | if (trim_kernel_dcache && lru.lru_get_size() > max) |
7c673cae FG |
619 | _invalidate_kernel_dcache(); |
620 | ||
621 | // hose root? | |
622 | if (lru.lru_get_size() == 0 && root && root->get_num_ref() == 0 && inode_map.size() == 1 + root_parents.size()) { | |
623 | ldout(cct, 15) << "trim_cache trimmed root " << root << dendl; | |
624 | delete root; | |
625 | root = 0; | |
626 | root_ancestor = 0; | |
627 | while (!root_parents.empty()) | |
628 | root_parents.erase(root_parents.begin()); | |
629 | inode_map.clear(); | |
630 | _reset_faked_inos(); | |
631 | } | |
632 | } | |
633 | ||
634 | void Client::trim_cache_for_reconnect(MetaSession *s) | |
635 | { | |
636 | mds_rank_t mds = s->mds_num; | |
637 | ldout(cct, 20) << "trim_cache_for_reconnect mds." << mds << dendl; | |
638 | ||
639 | int trimmed = 0; | |
640 | list<Dentry*> skipped; | |
641 | while (lru.lru_get_size() > 0) { | |
642 | Dentry *dn = static_cast<Dentry*>(lru.lru_expire()); | |
643 | if (!dn) | |
644 | break; | |
645 | ||
646 | if ((dn->inode && dn->inode->caps.count(mds)) || | |
647 | dn->dir->parent_inode->caps.count(mds)) { | |
648 | trim_dentry(dn); | |
649 | trimmed++; | |
650 | } else | |
651 | skipped.push_back(dn); | |
652 | } | |
653 | ||
654 | for(list<Dentry*>::iterator p = skipped.begin(); p != skipped.end(); ++p) | |
655 | lru.lru_insert_mid(*p); | |
656 | ||
657 | ldout(cct, 20) << "trim_cache_for_reconnect mds." << mds | |
658 | << " trimmed " << trimmed << " dentries" << dendl; | |
659 | ||
660 | if (s->caps.size() > 0) | |
661 | _invalidate_kernel_dcache(); | |
662 | } | |
663 | ||
664 | void Client::trim_dentry(Dentry *dn) | |
665 | { | |
666 | ldout(cct, 15) << "trim_dentry unlinking dn " << dn->name | |
667 | << " in dir " << hex << dn->dir->parent_inode->ino | |
668 | << dendl; | |
669 | if (dn->inode) { | |
670 | Inode *diri = dn->dir->parent_inode; | |
671 | diri->dir_release_count++; | |
672 | clear_dir_complete_and_ordered(diri, true); | |
673 | } | |
674 | unlink(dn, false, false); // drop dir, drop dentry | |
675 | } | |
676 | ||
677 | ||
678 | void Client::update_inode_file_bits(Inode *in, | |
679 | uint64_t truncate_seq, uint64_t truncate_size, | |
680 | uint64_t size, uint64_t change_attr, | |
681 | uint64_t time_warp_seq, utime_t ctime, | |
682 | utime_t mtime, | |
683 | utime_t atime, | |
684 | version_t inline_version, | |
685 | bufferlist& inline_data, | |
686 | int issued) | |
687 | { | |
688 | bool warn = false; | |
689 | ldout(cct, 10) << "update_inode_file_bits " << *in << " " << ccap_string(issued) | |
690 | << " mtime " << mtime << dendl; | |
691 | ldout(cct, 25) << "truncate_seq: mds " << truncate_seq << " local " | |
692 | << in->truncate_seq << " time_warp_seq: mds " << time_warp_seq | |
693 | << " local " << in->time_warp_seq << dendl; | |
694 | uint64_t prior_size = in->size; | |
695 | ||
696 | if (inline_version > in->inline_version) { | |
697 | in->inline_data = inline_data; | |
698 | in->inline_version = inline_version; | |
699 | } | |
700 | ||
701 | /* always take a newer change attr */ | |
702 | if (change_attr > in->change_attr) | |
703 | in->change_attr = change_attr; | |
704 | ||
705 | if (truncate_seq > in->truncate_seq || | |
706 | (truncate_seq == in->truncate_seq && size > in->size)) { | |
707 | ldout(cct, 10) << "size " << in->size << " -> " << size << dendl; | |
708 | in->size = size; | |
709 | in->reported_size = size; | |
710 | if (truncate_seq != in->truncate_seq) { | |
711 | ldout(cct, 10) << "truncate_seq " << in->truncate_seq << " -> " | |
712 | << truncate_seq << dendl; | |
713 | in->truncate_seq = truncate_seq; | |
714 | in->oset.truncate_seq = truncate_seq; | |
715 | ||
716 | // truncate cached file data | |
717 | if (prior_size > size) { | |
718 | _invalidate_inode_cache(in, truncate_size, prior_size - truncate_size); | |
719 | } | |
720 | } | |
721 | ||
722 | // truncate inline data | |
723 | if (in->inline_version < CEPH_INLINE_NONE) { | |
724 | uint32_t len = in->inline_data.length(); | |
725 | if (size < len) | |
726 | in->inline_data.splice(size, len - size); | |
727 | } | |
728 | } | |
729 | if (truncate_seq >= in->truncate_seq && | |
730 | in->truncate_size != truncate_size) { | |
731 | if (in->is_file()) { | |
732 | ldout(cct, 10) << "truncate_size " << in->truncate_size << " -> " | |
733 | << truncate_size << dendl; | |
734 | in->truncate_size = truncate_size; | |
735 | in->oset.truncate_size = truncate_size; | |
736 | } else { | |
737 | ldout(cct, 0) << "Hmmm, truncate_seq && truncate_size changed on non-file inode!" << dendl; | |
738 | } | |
739 | } | |
740 | ||
741 | // be careful with size, mtime, atime | |
742 | if (issued & (CEPH_CAP_FILE_EXCL| | |
743 | CEPH_CAP_FILE_WR| | |
744 | CEPH_CAP_FILE_BUFFER| | |
745 | CEPH_CAP_AUTH_EXCL| | |
746 | CEPH_CAP_XATTR_EXCL)) { | |
747 | ldout(cct, 30) << "Yay have enough caps to look at our times" << dendl; | |
748 | if (ctime > in->ctime) | |
749 | in->ctime = ctime; | |
750 | if (time_warp_seq > in->time_warp_seq) { | |
751 | ldout(cct, 10) << "mds time_warp_seq " << time_warp_seq << " on inode " << *in | |
752 | << " is higher than local time_warp_seq " | |
753 | << in->time_warp_seq << dendl; | |
754 | //the mds updated times, so take those! | |
755 | in->mtime = mtime; | |
756 | in->atime = atime; | |
757 | in->time_warp_seq = time_warp_seq; | |
758 | } else if (time_warp_seq == in->time_warp_seq) { | |
759 | //take max times | |
760 | if (mtime > in->mtime) | |
761 | in->mtime = mtime; | |
762 | if (atime > in->atime) | |
763 | in->atime = atime; | |
764 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
765 | //ignore mds values as we have a higher seq | |
766 | } else warn = true; | |
767 | } else { | |
768 | ldout(cct, 30) << "Don't have enough caps, just taking mds' time values" << dendl; | |
769 | if (time_warp_seq >= in->time_warp_seq) { | |
770 | in->ctime = ctime; | |
771 | in->mtime = mtime; | |
772 | in->atime = atime; | |
773 | in->time_warp_seq = time_warp_seq; | |
774 | } else warn = true; | |
775 | } | |
776 | if (warn) { | |
777 | ldout(cct, 0) << "WARNING: " << *in << " mds time_warp_seq " | |
778 | << time_warp_seq << " is lower than local time_warp_seq " | |
779 | << in->time_warp_seq | |
780 | << dendl; | |
781 | } | |
782 | } | |
783 | ||
784 | void Client::_fragmap_remove_non_leaves(Inode *in) | |
785 | { | |
786 | for (map<frag_t,int>::iterator p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
787 | if (!in->dirfragtree.is_leaf(p->first)) | |
788 | in->fragmap.erase(p++); | |
789 | else | |
790 | ++p; | |
791 | } | |
792 | ||
793 | void Client::_fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds) | |
794 | { | |
795 | for (auto p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
796 | if (p->second == mds) | |
797 | in->fragmap.erase(p++); | |
798 | else | |
799 | ++p; | |
800 | } | |
801 | ||
802 | Inode * Client::add_update_inode(InodeStat *st, utime_t from, | |
803 | MetaSession *session, | |
804 | const UserPerm& request_perms) | |
805 | { | |
806 | Inode *in; | |
807 | bool was_new = false; | |
808 | if (inode_map.count(st->vino)) { | |
809 | in = inode_map[st->vino]; | |
810 | ldout(cct, 12) << "add_update_inode had " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
811 | } else { | |
812 | in = new Inode(this, st->vino, &st->layout); | |
813 | inode_map[st->vino] = in; | |
814 | ||
815 | if (use_faked_inos()) | |
816 | _assign_faked_ino(in); | |
817 | ||
818 | if (!root) { | |
819 | root = in; | |
820 | root_ancestor = in; | |
821 | cwd = root; | |
822 | } else if (!mounted) { | |
823 | root_parents[root_ancestor] = in; | |
824 | root_ancestor = in; | |
825 | } | |
826 | ||
827 | // immutable bits | |
828 | in->ino = st->vino.ino; | |
829 | in->snapid = st->vino.snapid; | |
830 | in->mode = st->mode & S_IFMT; | |
831 | was_new = true; | |
832 | } | |
833 | ||
834 | in->rdev = st->rdev; | |
835 | if (in->is_symlink()) | |
836 | in->symlink = st->symlink; | |
837 | ||
838 | if (was_new) | |
839 | ldout(cct, 12) << "add_update_inode adding " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
840 | ||
841 | if (!st->cap.caps) | |
842 | return in; // as with readdir returning indoes in different snaprealms (no caps!) | |
843 | ||
844 | // only update inode if mds info is strictly newer, or it is the same and projected (odd). | |
845 | bool updating_inode = false; | |
846 | int issued = 0; | |
847 | if (st->version == 0 || | |
848 | (in->version & ~1) < st->version) { | |
849 | updating_inode = true; | |
850 | ||
851 | int implemented = 0; | |
852 | issued = in->caps_issued(&implemented) | in->caps_dirty(); | |
853 | issued |= implemented; | |
854 | ||
855 | in->version = st->version; | |
856 | ||
857 | if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
858 | in->mode = st->mode; | |
859 | in->uid = st->uid; | |
860 | in->gid = st->gid; | |
861 | in->btime = st->btime; | |
862 | } | |
863 | ||
864 | if ((issued & CEPH_CAP_LINK_EXCL) == 0) { | |
865 | in->nlink = st->nlink; | |
866 | } | |
867 | ||
868 | in->dirstat = st->dirstat; | |
869 | in->rstat = st->rstat; | |
870 | in->quota = st->quota; | |
871 | in->layout = st->layout; | |
872 | ||
873 | if (in->is_dir()) { | |
874 | in->dir_layout = st->dir_layout; | |
875 | ldout(cct, 20) << " dir hash is " << (int)in->dir_layout.dl_dir_hash << dendl; | |
876 | } | |
877 | ||
878 | update_inode_file_bits(in, st->truncate_seq, st->truncate_size, st->size, | |
879 | st->change_attr, st->time_warp_seq, st->ctime, | |
880 | st->mtime, st->atime, st->inline_version, | |
881 | st->inline_data, issued); | |
882 | } else if (st->inline_version > in->inline_version) { | |
883 | in->inline_data = st->inline_data; | |
884 | in->inline_version = st->inline_version; | |
885 | } | |
886 | ||
887 | if ((in->xattr_version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && | |
888 | st->xattrbl.length() && | |
889 | st->xattr_version > in->xattr_version) { | |
890 | bufferlist::iterator p = st->xattrbl.begin(); | |
891 | ::decode(in->xattrs, p); | |
892 | in->xattr_version = st->xattr_version; | |
893 | } | |
894 | ||
895 | // move me if/when version reflects fragtree changes. | |
896 | if (in->dirfragtree != st->dirfragtree) { | |
897 | in->dirfragtree = st->dirfragtree; | |
898 | _fragmap_remove_non_leaves(in); | |
899 | } | |
900 | ||
901 | if (in->snapid == CEPH_NOSNAP) { | |
902 | add_update_cap(in, session, st->cap.cap_id, st->cap.caps, st->cap.seq, | |
903 | st->cap.mseq, inodeno_t(st->cap.realm), st->cap.flags, | |
904 | request_perms); | |
905 | if (in->auth_cap && in->auth_cap->session == session) | |
906 | in->max_size = st->max_size; | |
907 | } else | |
908 | in->snap_caps |= st->cap.caps; | |
909 | ||
910 | // setting I_COMPLETE needs to happen after adding the cap | |
911 | if (updating_inode && | |
912 | in->is_dir() && | |
913 | (st->cap.caps & CEPH_CAP_FILE_SHARED) && | |
914 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
915 | in->dirstat.nfiles == 0 && | |
916 | in->dirstat.nsubdirs == 0) { | |
917 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on empty dir " << *in << dendl; | |
918 | in->flags |= I_COMPLETE | I_DIR_ORDERED; | |
919 | if (in->dir) { | |
920 | ldout(cct, 10) << " dir is open on empty dir " << in->ino << " with " | |
921 | << in->dir->dentries.size() << " entries, marking all dentries null" << dendl; | |
922 | in->dir->readdir_cache.clear(); | |
923 | for (auto p = in->dir->dentries.begin(); | |
924 | p != in->dir->dentries.end(); | |
925 | ++p) { | |
926 | unlink(p->second, true, true); // keep dir, keep dentry | |
927 | } | |
928 | if (in->dir->dentries.empty()) | |
929 | close_dir(in->dir); | |
930 | } | |
931 | } | |
932 | ||
933 | return in; | |
934 | } | |
935 | ||
936 | ||
937 | /* | |
938 | * insert_dentry_inode - insert + link a single dentry + inode into the metadata cache. | |
939 | */ | |
940 | Dentry *Client::insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
941 | Inode *in, utime_t from, MetaSession *session, | |
942 | Dentry *old_dentry) | |
943 | { | |
944 | Dentry *dn = NULL; | |
945 | if (dir->dentries.count(dname)) | |
946 | dn = dir->dentries[dname]; | |
947 | ||
948 | ldout(cct, 12) << "insert_dentry_inode '" << dname << "' vino " << in->vino() | |
949 | << " in dir " << dir->parent_inode->vino() << " dn " << dn | |
950 | << dendl; | |
951 | ||
952 | if (dn && dn->inode) { | |
953 | if (dn->inode->vino() == in->vino()) { | |
954 | touch_dn(dn); | |
955 | ldout(cct, 12) << " had dentry " << dname | |
956 | << " with correct vino " << dn->inode->vino() | |
957 | << dendl; | |
958 | } else { | |
959 | ldout(cct, 12) << " had dentry " << dname | |
960 | << " with WRONG vino " << dn->inode->vino() | |
961 | << dendl; | |
962 | unlink(dn, true, true); // keep dir, keep dentry | |
963 | } | |
964 | } | |
965 | ||
966 | if (!dn || !dn->inode) { | |
967 | InodeRef tmp_ref(in); | |
968 | if (old_dentry) { | |
969 | if (old_dentry->dir != dir) { | |
970 | Inode *old_diri = old_dentry->dir->parent_inode; | |
971 | old_diri->dir_ordered_count++; | |
972 | clear_dir_complete_and_ordered(old_diri, false); | |
973 | } | |
974 | unlink(old_dentry, dir == old_dentry->dir, false); // drop dentry, keep dir open if its the same dir | |
975 | } | |
976 | Inode *diri = dir->parent_inode; | |
977 | diri->dir_ordered_count++; | |
978 | clear_dir_complete_and_ordered(diri, false); | |
979 | dn = link(dir, dname, in, dn); | |
980 | } | |
981 | ||
982 | update_dentry_lease(dn, dlease, from, session); | |
983 | return dn; | |
984 | } | |
985 | ||
986 | void Client::update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session) | |
987 | { | |
988 | utime_t dttl = from; | |
989 | dttl += (float)dlease->duration_ms / 1000.0; | |
990 | ||
991 | assert(dn); | |
992 | ||
993 | if (dlease->mask & CEPH_LOCK_DN) { | |
994 | if (dttl > dn->lease_ttl) { | |
995 | ldout(cct, 10) << "got dentry lease on " << dn->name | |
996 | << " dur " << dlease->duration_ms << "ms ttl " << dttl << dendl; | |
997 | dn->lease_ttl = dttl; | |
998 | dn->lease_mds = session->mds_num; | |
999 | dn->lease_seq = dlease->seq; | |
1000 | dn->lease_gen = session->cap_gen; | |
1001 | } | |
1002 | } | |
1003 | dn->cap_shared_gen = dn->dir->parent_inode->shared_gen; | |
1004 | } | |
1005 | ||
1006 | ||
1007 | /* | |
1008 | * update MDS location cache for a single inode | |
1009 | */ | |
1010 | void Client::update_dir_dist(Inode *in, DirStat *dst) | |
1011 | { | |
1012 | // auth | |
1013 | ldout(cct, 20) << "got dirfrag map for " << in->ino << " frag " << dst->frag << " to mds " << dst->auth << dendl; | |
1014 | if (dst->auth >= 0) { | |
1015 | in->fragmap[dst->frag] = dst->auth; | |
1016 | } else { | |
1017 | in->fragmap.erase(dst->frag); | |
1018 | } | |
1019 | if (!in->dirfragtree.is_leaf(dst->frag)) { | |
1020 | in->dirfragtree.force_to_leaf(cct, dst->frag); | |
1021 | _fragmap_remove_non_leaves(in); | |
1022 | } | |
1023 | ||
1024 | // replicated | |
1025 | in->dir_replicated = !dst->dist.empty(); // FIXME that's just one frag! | |
1026 | ||
1027 | // dist | |
1028 | /* | |
1029 | if (!st->dirfrag_dist.empty()) { // FIXME | |
1030 | set<int> dist = st->dirfrag_dist.begin()->second; | |
1031 | if (dist.empty() && !in->dir_contacts.empty()) | |
1032 | ldout(cct, 9) << "lost dist spec for " << in->ino | |
1033 | << " " << dist << dendl; | |
1034 | if (!dist.empty() && in->dir_contacts.empty()) | |
1035 | ldout(cct, 9) << "got dist spec for " << in->ino | |
1036 | << " " << dist << dendl; | |
1037 | in->dir_contacts = dist; | |
1038 | } | |
1039 | */ | |
1040 | } | |
1041 | ||
1042 | void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete) | |
1043 | { | |
1044 | if (diri->flags & I_COMPLETE) { | |
1045 | if (complete) { | |
1046 | ldout(cct, 10) << " clearing (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
1047 | diri->flags &= ~(I_COMPLETE | I_DIR_ORDERED); | |
1048 | } else { | |
1049 | if (diri->flags & I_DIR_ORDERED) { | |
1050 | ldout(cct, 10) << " clearing I_DIR_ORDERED on " << *diri << dendl; | |
1051 | diri->flags &= ~I_DIR_ORDERED; | |
1052 | } | |
1053 | } | |
1054 | if (diri->dir) | |
1055 | diri->dir->readdir_cache.clear(); | |
1056 | } | |
1057 | } | |
1058 | ||
1059 | /* | |
1060 | * insert results from readdir or lssnap into the metadata cache. | |
1061 | */ | |
1062 | void Client::insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri) { | |
1063 | ||
1064 | MClientReply *reply = request->reply; | |
1065 | ConnectionRef con = request->reply->get_connection(); | |
1066 | uint64_t features = con->get_features(); | |
1067 | ||
1068 | dir_result_t *dirp = request->dirp; | |
1069 | assert(dirp); | |
1070 | ||
1071 | // the extra buffer list is only set for readdir and lssnap replies | |
1072 | bufferlist::iterator p = reply->get_extra_bl().begin(); | |
1073 | if (!p.end()) { | |
1074 | // snapdir? | |
1075 | if (request->head.op == CEPH_MDS_OP_LSSNAP) { | |
1076 | assert(diri); | |
1077 | diri = open_snapdir(diri); | |
1078 | } | |
1079 | ||
1080 | // only open dir if we're actually adding stuff to it! | |
1081 | Dir *dir = diri->open_dir(); | |
1082 | assert(dir); | |
1083 | ||
1084 | // dirstat | |
1085 | DirStat dst(p); | |
1086 | __u32 numdn; | |
1087 | __u16 flags; | |
1088 | ::decode(numdn, p); | |
1089 | ::decode(flags, p); | |
1090 | ||
1091 | bool end = ((unsigned)flags & CEPH_READDIR_FRAG_END); | |
1092 | bool hash_order = ((unsigned)flags & CEPH_READDIR_HASH_ORDER); | |
1093 | ||
1094 | frag_t fg = (unsigned)request->head.args.readdir.frag; | |
1095 | unsigned readdir_offset = dirp->next_offset; | |
1096 | string readdir_start = dirp->last_name; | |
1097 | assert(!readdir_start.empty() || readdir_offset == 2); | |
1098 | ||
1099 | unsigned last_hash = 0; | |
1100 | if (hash_order) { | |
1101 | if (!readdir_start.empty()) { | |
1102 | last_hash = ceph_frag_value(diri->hash_dentry_name(readdir_start)); | |
1103 | } else if (flags & CEPH_READDIR_OFFSET_HASH) { | |
1104 | /* mds understands offset_hash */ | |
1105 | last_hash = (unsigned)request->head.args.readdir.offset_hash; | |
1106 | } | |
1107 | } | |
1108 | ||
1109 | if (fg != dst.frag) { | |
1110 | ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl; | |
1111 | fg = dst.frag; | |
1112 | if (!hash_order) { | |
1113 | readdir_offset = 2; | |
1114 | readdir_start.clear(); | |
1115 | dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false); | |
1116 | } | |
1117 | } | |
1118 | ||
1119 | ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end | |
1120 | << ", hash_order=" << hash_order | |
1121 | << ", readdir_start " << readdir_start | |
1122 | << ", last_hash " << last_hash | |
1123 | << ", next_offset " << readdir_offset << dendl; | |
1124 | ||
1125 | if (diri->snapid != CEPH_SNAPDIR && | |
1126 | fg.is_leftmost() && readdir_offset == 2 && | |
1127 | !(hash_order && last_hash)) { | |
1128 | dirp->release_count = diri->dir_release_count; | |
1129 | dirp->ordered_count = diri->dir_ordered_count; | |
1130 | dirp->start_shared_gen = diri->shared_gen; | |
1131 | dirp->cache_index = 0; | |
1132 | } | |
1133 | ||
1134 | dirp->buffer_frag = fg; | |
1135 | ||
1136 | _readdir_drop_dirp_buffer(dirp); | |
1137 | dirp->buffer.reserve(numdn); | |
1138 | ||
1139 | string dname; | |
1140 | LeaseStat dlease; | |
1141 | for (unsigned i=0; i<numdn; i++) { | |
1142 | ::decode(dname, p); | |
1143 | ::decode(dlease, p); | |
1144 | InodeStat ist(p, features); | |
1145 | ||
1146 | ldout(cct, 15) << "" << i << ": '" << dname << "'" << dendl; | |
1147 | ||
1148 | Inode *in = add_update_inode(&ist, request->sent_stamp, session, | |
1149 | request->perms); | |
1150 | Dentry *dn; | |
1151 | if (diri->dir->dentries.count(dname)) { | |
1152 | Dentry *olddn = diri->dir->dentries[dname]; | |
1153 | if (olddn->inode != in) { | |
1154 | // replace incorrect dentry | |
1155 | unlink(olddn, true, true); // keep dir, dentry | |
1156 | dn = link(dir, dname, in, olddn); | |
1157 | assert(dn == olddn); | |
1158 | } else { | |
1159 | // keep existing dn | |
1160 | dn = olddn; | |
1161 | touch_dn(dn); | |
1162 | } | |
1163 | } else { | |
1164 | // new dn | |
1165 | dn = link(dir, dname, in, NULL); | |
1166 | } | |
1167 | ||
1168 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1169 | if (hash_order) { | |
1170 | unsigned hash = ceph_frag_value(diri->hash_dentry_name(dname)); | |
1171 | if (hash != last_hash) | |
1172 | readdir_offset = 2; | |
1173 | last_hash = hash; | |
1174 | dn->offset = dir_result_t::make_fpos(hash, readdir_offset++, true); | |
1175 | } else { | |
1176 | dn->offset = dir_result_t::make_fpos(fg, readdir_offset++, false); | |
1177 | } | |
1178 | // add to readdir cache | |
1179 | if (dirp->release_count == diri->dir_release_count && | |
1180 | dirp->ordered_count == diri->dir_ordered_count && | |
1181 | dirp->start_shared_gen == diri->shared_gen) { | |
1182 | if (dirp->cache_index == dir->readdir_cache.size()) { | |
1183 | if (i == 0) { | |
1184 | assert(!dirp->inode->is_complete_and_ordered()); | |
1185 | dir->readdir_cache.reserve(dirp->cache_index + numdn); | |
1186 | } | |
1187 | dir->readdir_cache.push_back(dn); | |
1188 | } else if (dirp->cache_index < dir->readdir_cache.size()) { | |
1189 | if (dirp->inode->is_complete_and_ordered()) | |
1190 | assert(dir->readdir_cache[dirp->cache_index] == dn); | |
1191 | else | |
1192 | dir->readdir_cache[dirp->cache_index] = dn; | |
1193 | } else { | |
1194 | assert(0 == "unexpected readdir buffer idx"); | |
1195 | } | |
1196 | dirp->cache_index++; | |
1197 | } | |
1198 | // add to cached result list | |
1199 | dirp->buffer.push_back(dir_result_t::dentry(dn->offset, dname, in)); | |
1200 | ldout(cct, 15) << __func__ << " " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl; | |
1201 | } | |
1202 | ||
1203 | if (numdn > 0) | |
1204 | dirp->last_name = dname; | |
1205 | if (end) | |
1206 | dirp->next_offset = 2; | |
1207 | else | |
1208 | dirp->next_offset = readdir_offset; | |
1209 | ||
1210 | if (dir->is_empty()) | |
1211 | close_dir(dir); | |
1212 | } | |
1213 | } | |
1214 | ||
1215 | /** insert_trace | |
1216 | * | |
1217 | * insert a trace from a MDS reply into the cache. | |
1218 | */ | |
1219 | Inode* Client::insert_trace(MetaRequest *request, MetaSession *session) | |
1220 | { | |
1221 | MClientReply *reply = request->reply; | |
1222 | int op = request->get_op(); | |
1223 | ||
1224 | ldout(cct, 10) << "insert_trace from " << request->sent_stamp << " mds." << session->mds_num | |
1225 | << " is_target=" << (int)reply->head.is_target | |
1226 | << " is_dentry=" << (int)reply->head.is_dentry | |
1227 | << dendl; | |
1228 | ||
1229 | bufferlist::iterator p = reply->get_trace_bl().begin(); | |
1230 | if (request->got_unsafe) { | |
1231 | ldout(cct, 10) << "insert_trace -- already got unsafe; ignoring" << dendl; | |
1232 | assert(p.end()); | |
1233 | return NULL; | |
1234 | } | |
1235 | ||
1236 | if (p.end()) { | |
1237 | ldout(cct, 10) << "insert_trace -- no trace" << dendl; | |
1238 | ||
1239 | Dentry *d = request->dentry(); | |
1240 | if (d) { | |
1241 | Inode *diri = d->dir->parent_inode; | |
1242 | diri->dir_release_count++; | |
1243 | clear_dir_complete_and_ordered(diri, true); | |
1244 | } | |
1245 | ||
1246 | if (d && reply->get_result() == 0) { | |
1247 | if (op == CEPH_MDS_OP_RENAME) { | |
1248 | // rename | |
1249 | Dentry *od = request->old_dentry(); | |
1250 | ldout(cct, 10) << " unlinking rename src dn " << od << " for traceless reply" << dendl; | |
1251 | assert(od); | |
1252 | unlink(od, true, true); // keep dir, dentry | |
1253 | } else if (op == CEPH_MDS_OP_RMDIR || | |
1254 | op == CEPH_MDS_OP_UNLINK) { | |
1255 | // unlink, rmdir | |
1256 | ldout(cct, 10) << " unlinking unlink/rmdir dn " << d << " for traceless reply" << dendl; | |
1257 | unlink(d, true, true); // keep dir, dentry | |
1258 | } | |
1259 | } | |
1260 | return NULL; | |
1261 | } | |
1262 | ||
1263 | ConnectionRef con = request->reply->get_connection(); | |
1264 | uint64_t features = con->get_features(); | |
1265 | ldout(cct, 10) << " features 0x" << hex << features << dec << dendl; | |
1266 | ||
1267 | // snap trace | |
1268 | SnapRealm *realm = NULL; | |
1269 | if (reply->snapbl.length()) | |
1270 | update_snap_trace(reply->snapbl, &realm); | |
1271 | ||
1272 | ldout(cct, 10) << " hrm " | |
1273 | << " is_target=" << (int)reply->head.is_target | |
1274 | << " is_dentry=" << (int)reply->head.is_dentry | |
1275 | << dendl; | |
1276 | ||
1277 | InodeStat dirst; | |
1278 | DirStat dst; | |
1279 | string dname; | |
1280 | LeaseStat dlease; | |
1281 | InodeStat ist; | |
1282 | ||
1283 | if (reply->head.is_dentry) { | |
1284 | dirst.decode(p, features); | |
1285 | dst.decode(p); | |
1286 | ::decode(dname, p); | |
1287 | ::decode(dlease, p); | |
1288 | } | |
1289 | ||
1290 | Inode *in = 0; | |
1291 | if (reply->head.is_target) { | |
1292 | ist.decode(p, features); | |
1293 | if (cct->_conf->client_debug_getattr_caps) { | |
1294 | unsigned wanted = 0; | |
1295 | if (op == CEPH_MDS_OP_GETATTR || op == CEPH_MDS_OP_LOOKUP) | |
1296 | wanted = request->head.args.getattr.mask; | |
1297 | else if (op == CEPH_MDS_OP_OPEN || op == CEPH_MDS_OP_CREATE) | |
1298 | wanted = request->head.args.open.mask; | |
1299 | ||
1300 | if ((wanted & CEPH_CAP_XATTR_SHARED) && | |
1301 | !(ist.xattr_version > 0 && ist.xattrbl.length() > 0)) | |
1302 | assert(0 == "MDS reply does not contain xattrs"); | |
1303 | } | |
1304 | ||
1305 | in = add_update_inode(&ist, request->sent_stamp, session, | |
1306 | request->perms); | |
1307 | } | |
1308 | ||
1309 | Inode *diri = NULL; | |
1310 | if (reply->head.is_dentry) { | |
1311 | diri = add_update_inode(&dirst, request->sent_stamp, session, | |
1312 | request->perms); | |
1313 | update_dir_dist(diri, &dst); // dir stat info is attached to .. | |
1314 | ||
1315 | if (in) { | |
1316 | Dir *dir = diri->open_dir(); | |
1317 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session, | |
1318 | (op == CEPH_MDS_OP_RENAME) ? request->old_dentry() : NULL); | |
1319 | } else { | |
1320 | Dentry *dn = NULL; | |
1321 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1322 | dn = diri->dir->dentries[dname]; | |
1323 | if (dn->inode) { | |
1324 | diri->dir_ordered_count++; | |
1325 | clear_dir_complete_and_ordered(diri, false); | |
1326 | unlink(dn, true, true); // keep dir, dentry | |
1327 | } | |
1328 | } | |
1329 | if (dlease.duration_ms > 0) { | |
1330 | if (!dn) { | |
1331 | Dir *dir = diri->open_dir(); | |
1332 | dn = link(dir, dname, NULL, NULL); | |
1333 | } | |
1334 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1335 | } | |
1336 | } | |
1337 | } else if (op == CEPH_MDS_OP_LOOKUPSNAP || | |
1338 | op == CEPH_MDS_OP_MKSNAP) { | |
1339 | ldout(cct, 10) << " faking snap lookup weirdness" << dendl; | |
1340 | // fake it for snap lookup | |
1341 | vinodeno_t vino = ist.vino; | |
1342 | vino.snapid = CEPH_SNAPDIR; | |
1343 | assert(inode_map.count(vino)); | |
1344 | diri = inode_map[vino]; | |
1345 | ||
1346 | string dname = request->path.last_dentry(); | |
1347 | ||
1348 | LeaseStat dlease; | |
1349 | dlease.duration_ms = 0; | |
1350 | ||
1351 | if (in) { | |
1352 | Dir *dir = diri->open_dir(); | |
1353 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session); | |
1354 | } else { | |
1355 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1356 | Dentry *dn = diri->dir->dentries[dname]; | |
1357 | if (dn->inode) | |
1358 | unlink(dn, true, true); // keep dir, dentry | |
1359 | } | |
1360 | } | |
1361 | } | |
1362 | ||
1363 | if (in) { | |
1364 | if (op == CEPH_MDS_OP_READDIR || | |
1365 | op == CEPH_MDS_OP_LSSNAP) { | |
1366 | insert_readdir_results(request, session, in); | |
1367 | } else if (op == CEPH_MDS_OP_LOOKUPNAME) { | |
1368 | // hack: return parent inode instead | |
1369 | in = diri; | |
1370 | } | |
1371 | ||
1372 | if (request->dentry() == NULL && in != request->inode()) { | |
1373 | // pin the target inode if its parent dentry is not pinned | |
1374 | request->set_other_inode(in); | |
1375 | } | |
1376 | } | |
1377 | ||
1378 | if (realm) | |
1379 | put_snap_realm(realm); | |
1380 | ||
1381 | request->target = in; | |
1382 | return in; | |
1383 | } | |
1384 | ||
1385 | // ------- | |
1386 | ||
1387 | mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri) | |
1388 | { | |
1389 | mds_rank_t mds = MDS_RANK_NONE; | |
1390 | __u32 hash = 0; | |
1391 | bool is_hash = false; | |
1392 | ||
1393 | Inode *in = NULL; | |
1394 | Dentry *de = NULL; | |
1395 | Cap *cap = NULL; | |
1396 | ||
1397 | if (req->resend_mds >= 0) { | |
1398 | mds = req->resend_mds; | |
1399 | req->resend_mds = -1; | |
1400 | ldout(cct, 10) << "choose_target_mds resend_mds specified as mds." << mds << dendl; | |
1401 | goto out; | |
1402 | } | |
1403 | ||
1404 | if (cct->_conf->client_use_random_mds) | |
1405 | goto random_mds; | |
1406 | ||
1407 | in = req->inode(); | |
1408 | de = req->dentry(); | |
1409 | if (in) { | |
1410 | ldout(cct, 20) << "choose_target_mds starting with req->inode " << *in << dendl; | |
1411 | if (req->path.depth()) { | |
1412 | hash = in->hash_dentry_name(req->path[0]); | |
1413 | ldout(cct, 20) << "choose_target_mds inode dir hash is " << (int)in->dir_layout.dl_dir_hash | |
1414 | << " on " << req->path[0] | |
1415 | << " => " << hash << dendl; | |
1416 | is_hash = true; | |
1417 | } | |
1418 | } else if (de) { | |
1419 | if (de->inode) { | |
1420 | in = de->inode.get(); | |
1421 | ldout(cct, 20) << "choose_target_mds starting with req->dentry inode " << *in << dendl; | |
1422 | } else { | |
1423 | in = de->dir->parent_inode; | |
1424 | hash = in->hash_dentry_name(de->name); | |
1425 | ldout(cct, 20) << "choose_target_mds dentry dir hash is " << (int)in->dir_layout.dl_dir_hash | |
1426 | << " on " << de->name | |
1427 | << " => " << hash << dendl; | |
1428 | is_hash = true; | |
1429 | } | |
1430 | } | |
1431 | if (in) { | |
1432 | if (in->snapid != CEPH_NOSNAP) { | |
1433 | ldout(cct, 10) << "choose_target_mds " << *in << " is snapped, using nonsnap parent" << dendl; | |
1434 | while (in->snapid != CEPH_NOSNAP) { | |
1435 | if (in->snapid == CEPH_SNAPDIR) | |
1436 | in = in->snapdir_parent.get(); | |
1437 | else if (!in->dn_set.empty()) | |
1438 | /* In most cases there will only be one dentry, so getting it | |
1439 | * will be the correct action. If there are multiple hard links, | |
1440 | * I think the MDS should be able to redirect as needed*/ | |
1441 | in = in->get_first_parent()->dir->parent_inode; | |
1442 | else { | |
1443 | ldout(cct, 10) << "got unlinked inode, can't look at parent" << dendl; | |
1444 | break; | |
1445 | } | |
1446 | } | |
1447 | is_hash = false; | |
1448 | } | |
1449 | ||
1450 | ldout(cct, 20) << "choose_target_mds " << *in << " is_hash=" << is_hash | |
1451 | << " hash=" << hash << dendl; | |
1452 | ||
1453 | if (is_hash && S_ISDIR(in->mode) && !in->fragmap.empty()) { | |
1454 | frag_t fg = in->dirfragtree[hash]; | |
1455 | if (in->fragmap.count(fg)) { | |
1456 | mds = in->fragmap[fg]; | |
1457 | if (phash_diri) | |
1458 | *phash_diri = in; | |
1459 | ldout(cct, 10) << "choose_target_mds from dirfragtree hash" << dendl; | |
1460 | goto out; | |
1461 | } | |
1462 | } | |
1463 | ||
1464 | if (req->auth_is_best()) | |
1465 | cap = in->auth_cap; | |
1466 | if (!cap && !in->caps.empty()) | |
1467 | cap = in->caps.begin()->second; | |
1468 | if (!cap) | |
1469 | goto random_mds; | |
1470 | mds = cap->session->mds_num; | |
1471 | ldout(cct, 10) << "choose_target_mds from caps on inode " << *in << dendl; | |
1472 | ||
1473 | goto out; | |
1474 | } | |
1475 | ||
1476 | random_mds: | |
1477 | if (mds < 0) { | |
1478 | mds = _get_random_up_mds(); | |
1479 | ldout(cct, 10) << "did not get mds through better means, so chose random mds " << mds << dendl; | |
1480 | } | |
1481 | ||
1482 | out: | |
1483 | ldout(cct, 20) << "mds is " << mds << dendl; | |
1484 | return mds; | |
1485 | } | |
1486 | ||
1487 | ||
1488 | void Client::connect_mds_targets(mds_rank_t mds) | |
1489 | { | |
1490 | ldout(cct, 10) << "connect_mds_targets for mds." << mds << dendl; | |
1491 | assert(mds_sessions.count(mds)); | |
1492 | const MDSMap::mds_info_t& info = mdsmap->get_mds_info(mds); | |
1493 | for (set<mds_rank_t>::const_iterator q = info.export_targets.begin(); | |
1494 | q != info.export_targets.end(); | |
1495 | ++q) { | |
1496 | if (mds_sessions.count(*q) == 0 && | |
1497 | mdsmap->is_clientreplay_or_active_or_stopping(*q)) { | |
1498 | ldout(cct, 10) << "check_mds_sessions opening mds." << mds | |
1499 | << " export target mds." << *q << dendl; | |
1500 | _open_mds_session(*q); | |
1501 | } | |
1502 | } | |
1503 | } | |
1504 | ||
1505 | void Client::dump_mds_sessions(Formatter *f) | |
1506 | { | |
1507 | f->dump_int("id", get_nodeid().v); | |
1508 | f->open_array_section("sessions"); | |
1509 | for (map<mds_rank_t,MetaSession*>::const_iterator p = mds_sessions.begin(); p != mds_sessions.end(); ++p) { | |
1510 | f->open_object_section("session"); | |
1511 | p->second->dump(f); | |
1512 | f->close_section(); | |
1513 | } | |
1514 | f->close_section(); | |
1515 | f->dump_int("mdsmap_epoch", mdsmap->get_epoch()); | |
1516 | } | |
1517 | void Client::dump_mds_requests(Formatter *f) | |
1518 | { | |
1519 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
1520 | p != mds_requests.end(); | |
1521 | ++p) { | |
1522 | f->open_object_section("request"); | |
1523 | p->second->dump(f); | |
1524 | f->close_section(); | |
1525 | } | |
1526 | } | |
1527 | ||
1528 | int Client::verify_reply_trace(int r, | |
1529 | MetaRequest *request, MClientReply *reply, | |
1530 | InodeRef *ptarget, bool *pcreated, | |
1531 | const UserPerm& perms) | |
1532 | { | |
1533 | // check whether this request actually did the create, and set created flag | |
1534 | bufferlist extra_bl; | |
1535 | inodeno_t created_ino; | |
1536 | bool got_created_ino = false; | |
1537 | ceph::unordered_map<vinodeno_t, Inode*>::iterator p; | |
1538 | ||
1539 | extra_bl.claim(reply->get_extra_bl()); | |
1540 | if (extra_bl.length() >= 8) { | |
1541 | // if the extra bufferlist has a buffer, we assume its the created inode | |
1542 | // and that this request to create succeeded in actually creating | |
1543 | // the inode (won the race with other create requests) | |
1544 | ::decode(created_ino, extra_bl); | |
1545 | got_created_ino = true; | |
1546 | ldout(cct, 10) << "make_request created ino " << created_ino << dendl; | |
1547 | } | |
1548 | ||
1549 | if (pcreated) | |
1550 | *pcreated = got_created_ino; | |
1551 | ||
1552 | if (request->target) { | |
1553 | *ptarget = request->target; | |
1554 | ldout(cct, 20) << "make_request target is " << *ptarget->get() << dendl; | |
1555 | } else { | |
1556 | if (got_created_ino && (p = inode_map.find(vinodeno_t(created_ino, CEPH_NOSNAP))) != inode_map.end()) { | |
1557 | (*ptarget) = p->second; | |
1558 | ldout(cct, 20) << "make_request created, target is " << *ptarget->get() << dendl; | |
1559 | } else { | |
1560 | // we got a traceless reply, and need to look up what we just | |
1561 | // created. for now, do this by name. someday, do this by the | |
1562 | // ino... which we know! FIXME. | |
1563 | InodeRef target; | |
1564 | Dentry *d = request->dentry(); | |
1565 | if (d) { | |
1566 | if (d->dir) { | |
1567 | ldout(cct, 10) << "make_request got traceless reply, looking up #" | |
1568 | << d->dir->parent_inode->ino << "/" << d->name | |
1569 | << " got_ino " << got_created_ino | |
1570 | << " ino " << created_ino | |
1571 | << dendl; | |
1572 | r = _do_lookup(d->dir->parent_inode, d->name, request->regetattr_mask, | |
1573 | &target, perms); | |
1574 | } else { | |
1575 | // if the dentry is not linked, just do our best. see #5021. | |
1576 | assert(0 == "how did this happen? i want logs!"); | |
1577 | } | |
1578 | } else { | |
1579 | Inode *in = request->inode(); | |
1580 | ldout(cct, 10) << "make_request got traceless reply, forcing getattr on #" | |
1581 | << in->ino << dendl; | |
1582 | r = _getattr(in, request->regetattr_mask, perms, true); | |
1583 | target = in; | |
1584 | } | |
1585 | if (r >= 0) { | |
1586 | // verify ino returned in reply and trace_dist are the same | |
1587 | if (got_created_ino && | |
1588 | created_ino.val != target->ino.val) { | |
1589 | ldout(cct, 5) << "create got ino " << created_ino << " but then failed on lookup; EINTR?" << dendl; | |
1590 | r = -EINTR; | |
1591 | } | |
1592 | if (ptarget) | |
1593 | ptarget->swap(target); | |
1594 | } | |
1595 | } | |
1596 | } | |
1597 | ||
1598 | return r; | |
1599 | } | |
1600 | ||
1601 | ||
1602 | /** | |
1603 | * make a request | |
1604 | * | |
1605 | * Blocking helper to make an MDS request. | |
1606 | * | |
1607 | * If the ptarget flag is set, behavior changes slightly: the caller | |
1608 | * expects to get a pointer to the inode we are creating or operating | |
1609 | * on. As a result, we will follow up any traceless mutation reply | |
1610 | * with a getattr or lookup to transparently handle a traceless reply | |
1611 | * from the MDS (as when the MDS restarts and the client has to replay | |
1612 | * a request). | |
1613 | * | |
1614 | * @param request the MetaRequest to execute | |
1615 | * @param perms The user uid/gid to execute as (eventually, full group lists?) | |
1616 | * @param ptarget [optional] address to store a pointer to the target inode we want to create or operate on | |
1617 | * @param pcreated [optional; required if ptarget] where to store a bool of whether our create atomically created a file | |
1618 | * @param use_mds [optional] prefer a specific mds (-1 for default) | |
1619 | * @param pdirbl [optional; disallowed if ptarget] where to pass extra reply payload to the caller | |
1620 | */ | |
1621 | int Client::make_request(MetaRequest *request, | |
1622 | const UserPerm& perms, | |
1623 | InodeRef *ptarget, bool *pcreated, | |
1624 | mds_rank_t use_mds, | |
1625 | bufferlist *pdirbl) | |
1626 | { | |
1627 | int r = 0; | |
1628 | ||
1629 | // assign a unique tid | |
1630 | ceph_tid_t tid = ++last_tid; | |
1631 | request->set_tid(tid); | |
1632 | ||
1633 | // and timestamp | |
1634 | request->op_stamp = ceph_clock_now(); | |
1635 | ||
1636 | // make note | |
1637 | mds_requests[tid] = request->get(); | |
1638 | if (oldest_tid == 0 && request->get_op() != CEPH_MDS_OP_SETFILELOCK) | |
1639 | oldest_tid = tid; | |
1640 | ||
1641 | request->set_caller_perms(perms); | |
1642 | ||
1643 | if (cct->_conf->client_inject_fixed_oldest_tid) { | |
1644 | ldout(cct, 20) << __func__ << " injecting fixed oldest_client_tid(1)" << dendl; | |
1645 | request->set_oldest_client_tid(1); | |
1646 | } else { | |
1647 | request->set_oldest_client_tid(oldest_tid); | |
1648 | } | |
1649 | ||
1650 | // hack target mds? | |
1651 | if (use_mds >= 0) | |
1652 | request->resend_mds = use_mds; | |
1653 | ||
1654 | while (1) { | |
1655 | if (request->aborted()) | |
1656 | break; | |
1657 | ||
31f18b77 FG |
1658 | if (blacklisted) { |
1659 | request->abort(-EBLACKLISTED); | |
1660 | break; | |
1661 | } | |
1662 | ||
7c673cae FG |
1663 | // set up wait cond |
1664 | Cond caller_cond; | |
1665 | request->caller_cond = &caller_cond; | |
1666 | ||
1667 | // choose mds | |
1668 | Inode *hash_diri = NULL; | |
1669 | mds_rank_t mds = choose_target_mds(request, &hash_diri); | |
1670 | int mds_state = (mds == MDS_RANK_NONE) ? MDSMap::STATE_NULL : mdsmap->get_state(mds); | |
1671 | if (mds_state != MDSMap::STATE_ACTIVE && mds_state != MDSMap::STATE_STOPPING) { | |
1672 | if (mds_state == MDSMap::STATE_NULL && mds >= mdsmap->get_max_mds()) { | |
1673 | if (hash_diri) { | |
1674 | ldout(cct, 10) << " target mds." << mds << " has stopped, remove it from fragmap" << dendl; | |
1675 | _fragmap_remove_stopped_mds(hash_diri, mds); | |
1676 | } else { | |
1677 | ldout(cct, 10) << " target mds." << mds << " has stopped, trying a random mds" << dendl; | |
1678 | request->resend_mds = _get_random_up_mds(); | |
1679 | } | |
1680 | } else { | |
1681 | ldout(cct, 10) << " target mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
1682 | wait_on_list(waiting_for_mdsmap); | |
1683 | } | |
1684 | continue; | |
1685 | } | |
1686 | ||
1687 | // open a session? | |
1688 | MetaSession *session = NULL; | |
1689 | if (!have_open_session(mds)) { | |
1690 | session = _get_or_open_mds_session(mds); | |
1691 | ||
1692 | // wait | |
1693 | if (session->state == MetaSession::STATE_OPENING) { | |
1694 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
1695 | wait_on_context_list(session->waiting_for_open); | |
1696 | // Abort requests on REJECT from MDS | |
1697 | if (rejected_by_mds.count(mds)) { | |
1698 | request->abort(-EPERM); | |
1699 | break; | |
1700 | } | |
1701 | continue; | |
1702 | } | |
1703 | ||
1704 | if (!have_open_session(mds)) | |
1705 | continue; | |
1706 | } else { | |
1707 | session = mds_sessions[mds]; | |
1708 | } | |
1709 | ||
1710 | // send request. | |
1711 | send_request(request, session); | |
1712 | ||
1713 | // wait for signal | |
1714 | ldout(cct, 20) << "awaiting reply|forward|kick on " << &caller_cond << dendl; | |
1715 | request->kick = false; | |
1716 | while (!request->reply && // reply | |
1717 | request->resend_mds < 0 && // forward | |
1718 | !request->kick) | |
1719 | caller_cond.Wait(client_lock); | |
1720 | request->caller_cond = NULL; | |
1721 | ||
1722 | // did we get a reply? | |
1723 | if (request->reply) | |
1724 | break; | |
1725 | } | |
1726 | ||
1727 | if (!request->reply) { | |
1728 | assert(request->aborted()); | |
1729 | assert(!request->got_unsafe); | |
1730 | r = request->get_abort_code(); | |
1731 | request->item.remove_myself(); | |
1732 | unregister_request(request); | |
1733 | put_request(request); // ours | |
1734 | return r; | |
1735 | } | |
1736 | ||
1737 | // got it! | |
1738 | MClientReply *reply = request->reply; | |
1739 | request->reply = NULL; | |
1740 | r = reply->get_result(); | |
1741 | if (r >= 0) | |
1742 | request->success = true; | |
1743 | ||
1744 | // kick dispatcher (we've got it!) | |
1745 | assert(request->dispatch_cond); | |
1746 | request->dispatch_cond->Signal(); | |
1747 | ldout(cct, 20) << "sendrecv kickback on tid " << tid << " " << request->dispatch_cond << dendl; | |
1748 | request->dispatch_cond = 0; | |
1749 | ||
1750 | if (r >= 0 && ptarget) | |
1751 | r = verify_reply_trace(r, request, reply, ptarget, pcreated, perms); | |
1752 | ||
1753 | if (pdirbl) | |
1754 | pdirbl->claim(reply->get_extra_bl()); | |
1755 | ||
1756 | // -- log times -- | |
1757 | utime_t lat = ceph_clock_now(); | |
1758 | lat -= request->sent_stamp; | |
1759 | ldout(cct, 20) << "lat " << lat << dendl; | |
1760 | logger->tinc(l_c_lat, lat); | |
1761 | logger->tinc(l_c_reply, lat); | |
1762 | ||
1763 | put_request(request); | |
1764 | ||
1765 | reply->put(); | |
1766 | return r; | |
1767 | } | |
1768 | ||
1769 | void Client::unregister_request(MetaRequest *req) | |
1770 | { | |
1771 | mds_requests.erase(req->tid); | |
1772 | if (req->tid == oldest_tid) { | |
1773 | map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.upper_bound(oldest_tid); | |
1774 | while (true) { | |
1775 | if (p == mds_requests.end()) { | |
1776 | oldest_tid = 0; | |
1777 | break; | |
1778 | } | |
1779 | if (p->second->get_op() != CEPH_MDS_OP_SETFILELOCK) { | |
1780 | oldest_tid = p->first; | |
1781 | break; | |
1782 | } | |
1783 | ++p; | |
1784 | } | |
1785 | } | |
1786 | put_request(req); | |
1787 | } | |
1788 | ||
1789 | void Client::put_request(MetaRequest *request) | |
1790 | { | |
1791 | if (request->_put()) { | |
1792 | int op = -1; | |
1793 | if (request->success) | |
1794 | op = request->get_op(); | |
1795 | InodeRef other_in; | |
1796 | request->take_other_inode(&other_in); | |
1797 | delete request; | |
1798 | ||
1799 | if (other_in && | |
1800 | (op == CEPH_MDS_OP_RMDIR || | |
1801 | op == CEPH_MDS_OP_RENAME || | |
1802 | op == CEPH_MDS_OP_RMSNAP)) { | |
1803 | _try_to_trim_inode(other_in.get(), false); | |
1804 | } | |
1805 | } | |
1806 | } | |
1807 | ||
1808 | int Client::encode_inode_release(Inode *in, MetaRequest *req, | |
1809 | mds_rank_t mds, int drop, | |
1810 | int unless, int force) | |
1811 | { | |
1812 | ldout(cct, 20) << "encode_inode_release enter(in:" << *in << ", req:" << req | |
1813 | << " mds:" << mds << ", drop:" << drop << ", unless:" << unless | |
1814 | << ", have:" << ", force:" << force << ")" << dendl; | |
1815 | int released = 0; | |
1816 | if (in->caps.count(mds)) { | |
1817 | Cap *caps = in->caps[mds]; | |
1818 | drop &= ~(in->dirty_caps | get_caps_used(in)); | |
1819 | if ((drop & caps->issued) && | |
1820 | !(unless & caps->issued)) { | |
1821 | ldout(cct, 25) << "Dropping caps. Initial " << ccap_string(caps->issued) << dendl; | |
1822 | caps->issued &= ~drop; | |
1823 | caps->implemented &= ~drop; | |
1824 | released = 1; | |
1825 | ldout(cct, 25) << "Now have: " << ccap_string(caps->issued) << dendl; | |
1826 | } else { | |
1827 | released = force; | |
1828 | } | |
1829 | if (released) { | |
1830 | ceph_mds_request_release rel; | |
1831 | rel.ino = in->ino; | |
1832 | rel.cap_id = caps->cap_id; | |
1833 | rel.seq = caps->seq; | |
1834 | rel.issue_seq = caps->issue_seq; | |
1835 | rel.mseq = caps->mseq; | |
1836 | rel.caps = caps->implemented; | |
1837 | rel.wanted = caps->wanted; | |
1838 | rel.dname_len = 0; | |
1839 | rel.dname_seq = 0; | |
1840 | req->cap_releases.push_back(MClientRequest::Release(rel,"")); | |
1841 | } | |
1842 | } | |
1843 | ldout(cct, 25) << "encode_inode_release exit(in:" << *in << ") released:" | |
1844 | << released << dendl; | |
1845 | return released; | |
1846 | } | |
1847 | ||
1848 | void Client::encode_dentry_release(Dentry *dn, MetaRequest *req, | |
1849 | mds_rank_t mds, int drop, int unless) | |
1850 | { | |
1851 | ldout(cct, 20) << "encode_dentry_release enter(dn:" | |
1852 | << dn << ")" << dendl; | |
1853 | int released = 0; | |
1854 | if (dn->dir) | |
1855 | released = encode_inode_release(dn->dir->parent_inode, req, | |
1856 | mds, drop, unless, 1); | |
1857 | if (released && dn->lease_mds == mds) { | |
1858 | ldout(cct, 25) << "preemptively releasing dn to mds" << dendl; | |
1859 | MClientRequest::Release& rel = req->cap_releases.back(); | |
1860 | rel.item.dname_len = dn->name.length(); | |
1861 | rel.item.dname_seq = dn->lease_seq; | |
1862 | rel.dname = dn->name; | |
1863 | } | |
1864 | ldout(cct, 25) << "encode_dentry_release exit(dn:" | |
1865 | << dn << ")" << dendl; | |
1866 | } | |
1867 | ||
1868 | ||
1869 | /* | |
1870 | * This requires the MClientRequest *request member to be set. | |
1871 | * It will error out horribly without one. | |
1872 | * Additionally, if you set any *drop member, you'd better have | |
1873 | * set the corresponding dentry! | |
1874 | */ | |
1875 | void Client::encode_cap_releases(MetaRequest *req, mds_rank_t mds) | |
1876 | { | |
1877 | ldout(cct, 20) << "encode_cap_releases enter (req: " | |
1878 | << req << ", mds: " << mds << ")" << dendl; | |
1879 | if (req->inode_drop && req->inode()) | |
1880 | encode_inode_release(req->inode(), req, | |
1881 | mds, req->inode_drop, | |
1882 | req->inode_unless); | |
1883 | ||
1884 | if (req->old_inode_drop && req->old_inode()) | |
1885 | encode_inode_release(req->old_inode(), req, | |
1886 | mds, req->old_inode_drop, | |
1887 | req->old_inode_unless); | |
1888 | if (req->other_inode_drop && req->other_inode()) | |
1889 | encode_inode_release(req->other_inode(), req, | |
1890 | mds, req->other_inode_drop, | |
1891 | req->other_inode_unless); | |
1892 | ||
1893 | if (req->dentry_drop && req->dentry()) | |
1894 | encode_dentry_release(req->dentry(), req, | |
1895 | mds, req->dentry_drop, | |
1896 | req->dentry_unless); | |
1897 | ||
1898 | if (req->old_dentry_drop && req->old_dentry()) | |
1899 | encode_dentry_release(req->old_dentry(), req, | |
1900 | mds, req->old_dentry_drop, | |
1901 | req->old_dentry_unless); | |
1902 | ldout(cct, 25) << "encode_cap_releases exit (req: " | |
1903 | << req << ", mds " << mds <<dendl; | |
1904 | } | |
1905 | ||
1906 | bool Client::have_open_session(mds_rank_t mds) | |
1907 | { | |
1908 | return | |
1909 | mds_sessions.count(mds) && | |
1910 | (mds_sessions[mds]->state == MetaSession::STATE_OPEN || | |
1911 | mds_sessions[mds]->state == MetaSession::STATE_STALE); | |
1912 | } | |
1913 | ||
1914 | MetaSession *Client::_get_mds_session(mds_rank_t mds, Connection *con) | |
1915 | { | |
1916 | if (mds_sessions.count(mds) == 0) | |
1917 | return NULL; | |
1918 | MetaSession *s = mds_sessions[mds]; | |
1919 | if (s->con != con) | |
1920 | return NULL; | |
1921 | return s; | |
1922 | } | |
1923 | ||
1924 | MetaSession *Client::_get_or_open_mds_session(mds_rank_t mds) | |
1925 | { | |
1926 | if (mds_sessions.count(mds)) | |
1927 | return mds_sessions[mds]; | |
1928 | return _open_mds_session(mds); | |
1929 | } | |
1930 | ||
1931 | /** | |
1932 | * Populate a map of strings with client-identifying metadata, | |
1933 | * such as the hostname. Call this once at initialization. | |
1934 | */ | |
1935 | void Client::populate_metadata(const std::string &mount_root) | |
1936 | { | |
1937 | // Hostname | |
1938 | struct utsname u; | |
1939 | int r = uname(&u); | |
1940 | if (r >= 0) { | |
1941 | metadata["hostname"] = u.nodename; | |
1942 | ldout(cct, 20) << __func__ << " read hostname '" << u.nodename << "'" << dendl; | |
1943 | } else { | |
1944 | ldout(cct, 1) << __func__ << " failed to read hostname (" << cpp_strerror(r) << ")" << dendl; | |
1945 | } | |
1946 | ||
1947 | metadata["pid"] = stringify(getpid()); | |
1948 | ||
1949 | // Ceph entity id (the '0' in "client.0") | |
1950 | metadata["entity_id"] = cct->_conf->name.get_id(); | |
1951 | ||
1952 | // Our mount position | |
1953 | if (!mount_root.empty()) { | |
1954 | metadata["root"] = mount_root; | |
1955 | } | |
1956 | ||
1957 | // Ceph version | |
1958 | metadata["ceph_version"] = pretty_version_to_str(); | |
1959 | metadata["ceph_sha1"] = git_version_to_str(); | |
1960 | ||
1961 | // Apply any metadata from the user's configured overrides | |
1962 | std::vector<std::string> tokens; | |
1963 | get_str_vec(cct->_conf->client_metadata, ",", tokens); | |
1964 | for (const auto &i : tokens) { | |
1965 | auto eqpos = i.find("="); | |
1966 | // Throw out anything that isn't of the form "<str>=<str>" | |
1967 | if (eqpos == 0 || eqpos == std::string::npos || eqpos == i.size()) { | |
1968 | lderr(cct) << "Invalid metadata keyval pair: '" << i << "'" << dendl; | |
1969 | continue; | |
1970 | } | |
1971 | metadata[i.substr(0, eqpos)] = i.substr(eqpos + 1); | |
1972 | } | |
1973 | } | |
1974 | ||
1975 | /** | |
1976 | * Optionally add or override client metadata fields. | |
1977 | */ | |
1978 | void Client::update_metadata(std::string const &k, std::string const &v) | |
1979 | { | |
1980 | Mutex::Locker l(client_lock); | |
1981 | assert(initialized); | |
1982 | ||
1983 | if (metadata.count(k)) { | |
1984 | ldout(cct, 1) << __func__ << " warning, overriding metadata field '" << k | |
1985 | << "' from '" << metadata[k] << "' to '" << v << "'" << dendl; | |
1986 | } | |
1987 | ||
1988 | metadata[k] = v; | |
1989 | } | |
1990 | ||
1991 | MetaSession *Client::_open_mds_session(mds_rank_t mds) | |
1992 | { | |
1993 | ldout(cct, 10) << "_open_mds_session mds." << mds << dendl; | |
1994 | assert(mds_sessions.count(mds) == 0); | |
1995 | MetaSession *session = new MetaSession; | |
1996 | session->mds_num = mds; | |
1997 | session->seq = 0; | |
1998 | session->inst = mdsmap->get_inst(mds); | |
1999 | session->con = messenger->get_connection(session->inst); | |
2000 | session->state = MetaSession::STATE_OPENING; | |
2001 | session->mds_state = MDSMap::STATE_NULL; | |
2002 | mds_sessions[mds] = session; | |
2003 | ||
2004 | // Maybe skip sending a request to open if this MDS daemon | |
2005 | // has previously sent us a REJECT. | |
2006 | if (rejected_by_mds.count(mds)) { | |
2007 | if (rejected_by_mds[mds] == session->inst) { | |
2008 | ldout(cct, 4) << "_open_mds_session mds." << mds << " skipping " | |
2009 | "because we were rejected" << dendl; | |
2010 | return session; | |
2011 | } else { | |
2012 | ldout(cct, 4) << "_open_mds_session mds." << mds << " old inst " | |
2013 | "rejected us, trying with new inst" << dendl; | |
2014 | rejected_by_mds.erase(mds); | |
2015 | } | |
2016 | } | |
2017 | ||
2018 | MClientSession *m = new MClientSession(CEPH_SESSION_REQUEST_OPEN); | |
2019 | m->client_meta = metadata; | |
2020 | session->con->send_message(m); | |
2021 | return session; | |
2022 | } | |
2023 | ||
2024 | void Client::_close_mds_session(MetaSession *s) | |
2025 | { | |
2026 | ldout(cct, 2) << "_close_mds_session mds." << s->mds_num << " seq " << s->seq << dendl; | |
2027 | s->state = MetaSession::STATE_CLOSING; | |
2028 | s->con->send_message(new MClientSession(CEPH_SESSION_REQUEST_CLOSE, s->seq)); | |
2029 | } | |
2030 | ||
2031 | void Client::_closed_mds_session(MetaSession *s) | |
2032 | { | |
2033 | s->state = MetaSession::STATE_CLOSED; | |
2034 | s->con->mark_down(); | |
2035 | signal_context_list(s->waiting_for_open); | |
2036 | mount_cond.Signal(); | |
2037 | remove_session_caps(s); | |
2038 | kick_requests_closed(s); | |
2039 | mds_sessions.erase(s->mds_num); | |
2040 | delete s; | |
2041 | } | |
2042 | ||
2043 | void Client::handle_client_session(MClientSession *m) | |
2044 | { | |
2045 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
2046 | ldout(cct, 10) << "handle_client_session " << *m << " from mds." << from << dendl; | |
2047 | ||
2048 | MetaSession *session = _get_mds_session(from, m->get_connection().get()); | |
2049 | if (!session) { | |
2050 | ldout(cct, 10) << " discarding session message from sessionless mds " << m->get_source_inst() << dendl; | |
2051 | m->put(); | |
2052 | return; | |
2053 | } | |
2054 | ||
2055 | switch (m->get_op()) { | |
2056 | case CEPH_SESSION_OPEN: | |
2057 | renew_caps(session); | |
2058 | session->state = MetaSession::STATE_OPEN; | |
2059 | if (unmounting) | |
2060 | mount_cond.Signal(); | |
2061 | else | |
2062 | connect_mds_targets(from); | |
2063 | signal_context_list(session->waiting_for_open); | |
2064 | break; | |
2065 | ||
2066 | case CEPH_SESSION_CLOSE: | |
2067 | _closed_mds_session(session); | |
2068 | break; | |
2069 | ||
2070 | case CEPH_SESSION_RENEWCAPS: | |
2071 | if (session->cap_renew_seq == m->get_seq()) { | |
2072 | session->cap_ttl = | |
2073 | session->last_cap_renew_request + mdsmap->get_session_timeout(); | |
2074 | wake_inode_waiters(session); | |
2075 | } | |
2076 | break; | |
2077 | ||
2078 | case CEPH_SESSION_STALE: | |
2079 | renew_caps(session); | |
2080 | break; | |
2081 | ||
2082 | case CEPH_SESSION_RECALL_STATE: | |
2083 | trim_caps(session, m->get_max_caps()); | |
2084 | break; | |
2085 | ||
2086 | case CEPH_SESSION_FLUSHMSG: | |
2087 | session->con->send_message(new MClientSession(CEPH_SESSION_FLUSHMSG_ACK, m->get_seq())); | |
2088 | break; | |
2089 | ||
2090 | case CEPH_SESSION_FORCE_RO: | |
2091 | force_session_readonly(session); | |
2092 | break; | |
2093 | ||
2094 | case CEPH_SESSION_REJECT: | |
2095 | rejected_by_mds[session->mds_num] = session->inst; | |
2096 | _closed_mds_session(session); | |
2097 | ||
2098 | break; | |
2099 | ||
2100 | default: | |
2101 | ceph_abort(); | |
2102 | } | |
2103 | ||
2104 | m->put(); | |
2105 | } | |
2106 | ||
2107 | bool Client::_any_stale_sessions() const | |
2108 | { | |
2109 | assert(client_lock.is_locked_by_me()); | |
2110 | ||
2111 | for (const auto &i : mds_sessions) { | |
2112 | if (i.second->state == MetaSession::STATE_STALE) { | |
2113 | return true; | |
2114 | } | |
2115 | } | |
2116 | ||
2117 | return false; | |
2118 | } | |
2119 | ||
2120 | void Client::_kick_stale_sessions() | |
2121 | { | |
2122 | ldout(cct, 1) << "kick_stale_sessions" << dendl; | |
2123 | ||
2124 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
2125 | p != mds_sessions.end(); ) { | |
2126 | MetaSession *s = p->second; | |
2127 | ++p; | |
2128 | if (s->state == MetaSession::STATE_STALE) | |
2129 | _closed_mds_session(s); | |
2130 | } | |
2131 | } | |
2132 | ||
2133 | void Client::send_request(MetaRequest *request, MetaSession *session, | |
2134 | bool drop_cap_releases) | |
2135 | { | |
2136 | // make the request | |
2137 | mds_rank_t mds = session->mds_num; | |
2138 | ldout(cct, 10) << "send_request rebuilding request " << request->get_tid() | |
2139 | << " for mds." << mds << dendl; | |
2140 | MClientRequest *r = build_client_request(request); | |
2141 | if (request->dentry()) { | |
2142 | r->set_dentry_wanted(); | |
2143 | } | |
2144 | if (request->got_unsafe) { | |
2145 | r->set_replayed_op(); | |
2146 | if (request->target) | |
2147 | r->head.ino = request->target->ino; | |
2148 | } else { | |
2149 | encode_cap_releases(request, mds); | |
2150 | if (drop_cap_releases) // we haven't send cap reconnect yet, drop cap releases | |
2151 | request->cap_releases.clear(); | |
2152 | else | |
2153 | r->releases.swap(request->cap_releases); | |
2154 | } | |
2155 | r->set_mdsmap_epoch(mdsmap->get_epoch()); | |
2156 | if (r->head.op == CEPH_MDS_OP_SETXATTR) { | |
2157 | objecter->with_osdmap([r](const OSDMap& o) { | |
2158 | r->set_osdmap_epoch(o.get_epoch()); | |
2159 | }); | |
2160 | } | |
2161 | ||
2162 | if (request->mds == -1) { | |
2163 | request->sent_stamp = ceph_clock_now(); | |
2164 | ldout(cct, 20) << "send_request set sent_stamp to " << request->sent_stamp << dendl; | |
2165 | } | |
2166 | request->mds = mds; | |
2167 | ||
2168 | Inode *in = request->inode(); | |
2169 | if (in && in->caps.count(mds)) | |
2170 | request->sent_on_mseq = in->caps[mds]->mseq; | |
2171 | ||
2172 | session->requests.push_back(&request->item); | |
2173 | ||
2174 | ldout(cct, 10) << "send_request " << *r << " to mds." << mds << dendl; | |
2175 | session->con->send_message(r); | |
2176 | } | |
2177 | ||
2178 | MClientRequest* Client::build_client_request(MetaRequest *request) | |
2179 | { | |
2180 | MClientRequest *req = new MClientRequest(request->get_op()); | |
2181 | req->set_tid(request->tid); | |
2182 | req->set_stamp(request->op_stamp); | |
2183 | memcpy(&req->head, &request->head, sizeof(ceph_mds_request_head)); | |
2184 | ||
2185 | // if the filepath's haven't been set, set them! | |
2186 | if (request->path.empty()) { | |
2187 | Inode *in = request->inode(); | |
2188 | Dentry *de = request->dentry(); | |
2189 | if (in) | |
2190 | in->make_nosnap_relative_path(request->path); | |
2191 | else if (de) { | |
2192 | if (de->inode) | |
2193 | de->inode->make_nosnap_relative_path(request->path); | |
2194 | else if (de->dir) { | |
2195 | de->dir->parent_inode->make_nosnap_relative_path(request->path); | |
2196 | request->path.push_dentry(de->name); | |
2197 | } | |
2198 | else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2199 | << " No path, inode, or appropriately-endowed dentry given!" | |
2200 | << dendl; | |
2201 | } else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2202 | << " No path, inode, or dentry given!" | |
2203 | << dendl; | |
2204 | } | |
2205 | req->set_filepath(request->get_filepath()); | |
2206 | req->set_filepath2(request->get_filepath2()); | |
2207 | req->set_data(request->data); | |
2208 | req->set_retry_attempt(request->retry_attempt++); | |
2209 | req->head.num_fwd = request->num_fwd; | |
2210 | const gid_t *_gids; | |
2211 | int gid_count = request->perms.get_gids(&_gids); | |
2212 | req->set_gid_list(gid_count, _gids); | |
2213 | return req; | |
2214 | } | |
2215 | ||
2216 | ||
2217 | ||
2218 | void Client::handle_client_request_forward(MClientRequestForward *fwd) | |
2219 | { | |
2220 | mds_rank_t mds = mds_rank_t(fwd->get_source().num()); | |
2221 | MetaSession *session = _get_mds_session(mds, fwd->get_connection().get()); | |
2222 | if (!session) { | |
2223 | fwd->put(); | |
2224 | return; | |
2225 | } | |
2226 | ceph_tid_t tid = fwd->get_tid(); | |
2227 | ||
2228 | if (mds_requests.count(tid) == 0) { | |
2229 | ldout(cct, 10) << "handle_client_request_forward no pending request on tid " << tid << dendl; | |
2230 | fwd->put(); | |
2231 | return; | |
2232 | } | |
2233 | ||
2234 | MetaRequest *request = mds_requests[tid]; | |
2235 | assert(request); | |
2236 | ||
2237 | // reset retry counter | |
2238 | request->retry_attempt = 0; | |
2239 | ||
2240 | // request not forwarded, or dest mds has no session. | |
2241 | // resend. | |
2242 | ldout(cct, 10) << "handle_client_request tid " << tid | |
2243 | << " fwd " << fwd->get_num_fwd() | |
2244 | << " to mds." << fwd->get_dest_mds() | |
2245 | << ", resending to " << fwd->get_dest_mds() | |
2246 | << dendl; | |
2247 | ||
2248 | request->mds = -1; | |
2249 | request->item.remove_myself(); | |
2250 | request->num_fwd = fwd->get_num_fwd(); | |
2251 | request->resend_mds = fwd->get_dest_mds(); | |
2252 | request->caller_cond->Signal(); | |
2253 | ||
2254 | fwd->put(); | |
2255 | } | |
2256 | ||
2257 | bool Client::is_dir_operation(MetaRequest *req) | |
2258 | { | |
2259 | int op = req->get_op(); | |
2260 | if (op == CEPH_MDS_OP_MKNOD || op == CEPH_MDS_OP_LINK || | |
2261 | op == CEPH_MDS_OP_UNLINK || op == CEPH_MDS_OP_RENAME || | |
2262 | op == CEPH_MDS_OP_MKDIR || op == CEPH_MDS_OP_RMDIR || | |
2263 | op == CEPH_MDS_OP_SYMLINK || op == CEPH_MDS_OP_CREATE) | |
2264 | return true; | |
2265 | return false; | |
2266 | } | |
2267 | ||
2268 | void Client::handle_client_reply(MClientReply *reply) | |
2269 | { | |
2270 | mds_rank_t mds_num = mds_rank_t(reply->get_source().num()); | |
2271 | MetaSession *session = _get_mds_session(mds_num, reply->get_connection().get()); | |
2272 | if (!session) { | |
2273 | reply->put(); | |
2274 | return; | |
2275 | } | |
2276 | ||
2277 | ceph_tid_t tid = reply->get_tid(); | |
2278 | bool is_safe = reply->is_safe(); | |
2279 | ||
2280 | if (mds_requests.count(tid) == 0) { | |
2281 | lderr(cct) << "handle_client_reply no pending request on tid " << tid | |
2282 | << " safe is:" << is_safe << dendl; | |
2283 | reply->put(); | |
2284 | return; | |
2285 | } | |
2286 | MetaRequest *request = mds_requests.at(tid); | |
2287 | ||
2288 | ldout(cct, 20) << "handle_client_reply got a reply. Safe:" << is_safe | |
2289 | << " tid " << tid << dendl; | |
2290 | ||
2291 | if (request->got_unsafe && !is_safe) { | |
2292 | //duplicate response | |
2293 | ldout(cct, 0) << "got a duplicate reply on tid " << tid << " from mds " | |
2294 | << mds_num << " safe:" << is_safe << dendl; | |
2295 | reply->put(); | |
2296 | return; | |
2297 | } | |
2298 | ||
2299 | if (-ESTALE == reply->get_result()) { // see if we can get to proper MDS | |
2300 | ldout(cct, 20) << "got ESTALE on tid " << request->tid | |
2301 | << " from mds." << request->mds << dendl; | |
2302 | request->send_to_auth = true; | |
2303 | request->resend_mds = choose_target_mds(request); | |
2304 | Inode *in = request->inode(); | |
2305 | if (request->resend_mds >= 0 && | |
2306 | request->resend_mds == request->mds && | |
2307 | (in == NULL || | |
2308 | in->caps.count(request->resend_mds) == 0 || | |
2309 | request->sent_on_mseq == in->caps[request->resend_mds]->mseq)) { | |
2310 | // have to return ESTALE | |
2311 | } else { | |
2312 | request->caller_cond->Signal(); | |
2313 | reply->put(); | |
2314 | return; | |
2315 | } | |
2316 | ldout(cct, 20) << "have to return ESTALE" << dendl; | |
2317 | } | |
2318 | ||
2319 | assert(request->reply == NULL); | |
2320 | request->reply = reply; | |
2321 | insert_trace(request, session); | |
2322 | ||
2323 | // Handle unsafe reply | |
2324 | if (!is_safe) { | |
2325 | request->got_unsafe = true; | |
2326 | session->unsafe_requests.push_back(&request->unsafe_item); | |
2327 | if (is_dir_operation(request)) { | |
2328 | Inode *dir = request->inode(); | |
2329 | assert(dir); | |
2330 | dir->unsafe_ops.push_back(&request->unsafe_dir_item); | |
2331 | } | |
2332 | if (request->target) { | |
2333 | InodeRef &in = request->target; | |
2334 | in->unsafe_ops.push_back(&request->unsafe_target_item); | |
2335 | } | |
2336 | } | |
2337 | ||
2338 | // Only signal the caller once (on the first reply): | |
2339 | // Either its an unsafe reply, or its a safe reply and no unsafe reply was sent. | |
2340 | if (!is_safe || !request->got_unsafe) { | |
2341 | Cond cond; | |
2342 | request->dispatch_cond = &cond; | |
2343 | ||
2344 | // wake up waiter | |
2345 | ldout(cct, 20) << "handle_client_reply signalling caller " << (void*)request->caller_cond << dendl; | |
2346 | request->caller_cond->Signal(); | |
2347 | ||
2348 | // wake for kick back | |
2349 | while (request->dispatch_cond) { | |
2350 | ldout(cct, 20) << "handle_client_reply awaiting kickback on tid " << tid << " " << &cond << dendl; | |
2351 | cond.Wait(client_lock); | |
2352 | } | |
2353 | } | |
2354 | ||
2355 | if (is_safe) { | |
2356 | // the filesystem change is committed to disk | |
2357 | // we're done, clean up | |
2358 | if (request->got_unsafe) { | |
2359 | request->unsafe_item.remove_myself(); | |
2360 | request->unsafe_dir_item.remove_myself(); | |
2361 | request->unsafe_target_item.remove_myself(); | |
2362 | signal_cond_list(request->waitfor_safe); | |
2363 | } | |
2364 | request->item.remove_myself(); | |
2365 | unregister_request(request); | |
2366 | } | |
2367 | if (unmounting) | |
2368 | mount_cond.Signal(); | |
2369 | } | |
2370 | ||
2371 | void Client::_handle_full_flag(int64_t pool) | |
2372 | { | |
2373 | ldout(cct, 1) << __func__ << ": FULL: cancelling outstanding operations " | |
2374 | << "on " << pool << dendl; | |
2375 | // Cancel all outstanding ops in this pool with -ENOSPC: it is necessary | |
2376 | // to do this rather than blocking, because otherwise when we fill up we | |
2377 | // potentially lock caps forever on files with dirty pages, and we need | |
2378 | // to be able to release those caps to the MDS so that it can delete files | |
2379 | // and free up space. | |
2380 | epoch_t cancelled_epoch = objecter->op_cancel_writes(-ENOSPC, pool); | |
2381 | ||
2382 | // For all inodes with layouts in this pool and a pending flush write op | |
2383 | // (i.e. one of the ones we will cancel), we've got to purge_set their data | |
2384 | // from ObjectCacher so that it doesn't re-issue the write in response to | |
2385 | // the ENOSPC error. | |
2386 | // Fortunately since we're cancelling everything in a given pool, we don't | |
2387 | // need to know which ops belong to which ObjectSet, we can just blow all | |
2388 | // the un-flushed cached data away and mark any dirty inodes' async_err | |
2389 | // field with -ENOSPC as long as we're sure all the ops we cancelled were | |
2390 | // affecting this pool, and all the objectsets we're purging were also | |
2391 | // in this pool. | |
2392 | for (unordered_map<vinodeno_t,Inode*>::iterator i = inode_map.begin(); | |
2393 | i != inode_map.end(); ++i) | |
2394 | { | |
2395 | Inode *inode = i->second; | |
2396 | if (inode->oset.dirty_or_tx | |
2397 | && (pool == -1 || inode->layout.pool_id == pool)) { | |
2398 | ldout(cct, 4) << __func__ << ": FULL: inode 0x" << std::hex << i->first << std::dec | |
2399 | << " has dirty objects, purging and setting ENOSPC" << dendl; | |
2400 | objectcacher->purge_set(&inode->oset); | |
2401 | inode->set_async_err(-ENOSPC); | |
2402 | } | |
2403 | } | |
2404 | ||
2405 | if (cancelled_epoch != (epoch_t)-1) { | |
2406 | set_cap_epoch_barrier(cancelled_epoch); | |
2407 | } | |
2408 | } | |
2409 | ||
2410 | void Client::handle_osd_map(MOSDMap *m) | |
2411 | { | |
31f18b77 FG |
2412 | std::set<entity_addr_t> new_blacklists; |
2413 | objecter->consume_blacklist_events(&new_blacklists); | |
2414 | ||
2415 | const auto myaddr = messenger->get_myaddr(); | |
2416 | if (!blacklisted && new_blacklists.count(myaddr)) { | |
2417 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ | |
2418 | return o.get_epoch(); | |
2419 | }); | |
2420 | lderr(cct) << "I was blacklisted at osd epoch " << epoch << dendl; | |
2421 | blacklisted = true; | |
2422 | for (std::map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
2423 | p != mds_requests.end(); ) { | |
2424 | auto req = p->second; | |
2425 | ++p; | |
2426 | req->abort(-EBLACKLISTED); | |
2427 | if (req->caller_cond) { | |
2428 | req->kick = true; | |
2429 | req->caller_cond->Signal(); | |
2430 | } | |
2431 | } | |
2432 | ||
2433 | // Progress aborts on any requests that were on this waitlist. Any | |
2434 | // requests that were on a waiting_for_open session waitlist | |
2435 | // will get kicked during close session below. | |
2436 | signal_cond_list(waiting_for_mdsmap); | |
2437 | ||
2438 | // Force-close all sessions: assume this is not abandoning any state | |
2439 | // on the MDS side because the MDS will have seen the blacklist too. | |
2440 | while(!mds_sessions.empty()) { | |
2441 | auto i = mds_sessions.begin(); | |
2442 | auto session = i->second; | |
2443 | _closed_mds_session(session); | |
2444 | } | |
2445 | ||
2446 | // Since we know all our OSD ops will fail, cancel them all preemtively, | |
2447 | // so that on an unhealthy cluster we can umount promptly even if e.g. | |
2448 | // some PGs were inaccessible. | |
2449 | objecter->op_cancel_writes(-EBLACKLISTED); | |
2450 | ||
2451 | } else if (blacklisted) { | |
2452 | // Handle case where we were blacklisted but no longer are | |
2453 | blacklisted = objecter->with_osdmap([myaddr](const OSDMap &o){ | |
2454 | return o.is_blacklisted(myaddr);}); | |
2455 | } | |
2456 | ||
7c673cae FG |
2457 | if (objecter->osdmap_full_flag()) { |
2458 | _handle_full_flag(-1); | |
2459 | } else { | |
2460 | // Accumulate local list of full pools so that I can drop | |
2461 | // the objecter lock before re-entering objecter in | |
2462 | // cancel_writes | |
2463 | std::vector<int64_t> full_pools; | |
2464 | ||
2465 | objecter->with_osdmap([&full_pools](const OSDMap &o) { | |
2466 | for (const auto& kv : o.get_pools()) { | |
2467 | if (kv.second.has_flag(pg_pool_t::FLAG_FULL)) { | |
2468 | full_pools.push_back(kv.first); | |
2469 | } | |
2470 | } | |
2471 | }); | |
2472 | ||
2473 | for (auto p : full_pools) | |
2474 | _handle_full_flag(p); | |
2475 | ||
2476 | // Subscribe to subsequent maps to watch for the full flag going | |
2477 | // away. For the global full flag objecter does this for us, but | |
2478 | // it pays no attention to the per-pool full flag so in this branch | |
2479 | // we do it ourselves. | |
2480 | if (!full_pools.empty()) { | |
2481 | objecter->maybe_request_map(); | |
2482 | } | |
2483 | } | |
2484 | ||
2485 | m->put(); | |
2486 | } | |
2487 | ||
2488 | ||
2489 | // ------------------------ | |
2490 | // incoming messages | |
2491 | ||
2492 | ||
2493 | bool Client::ms_dispatch(Message *m) | |
2494 | { | |
2495 | Mutex::Locker l(client_lock); | |
2496 | if (!initialized) { | |
2497 | ldout(cct, 10) << "inactive, discarding " << *m << dendl; | |
2498 | m->put(); | |
2499 | return true; | |
2500 | } | |
2501 | ||
2502 | switch (m->get_type()) { | |
2503 | // mounting and mds sessions | |
2504 | case CEPH_MSG_MDS_MAP: | |
2505 | handle_mds_map(static_cast<MMDSMap*>(m)); | |
2506 | break; | |
2507 | case CEPH_MSG_FS_MAP: | |
2508 | handle_fs_map(static_cast<MFSMap*>(m)); | |
2509 | break; | |
2510 | case CEPH_MSG_FS_MAP_USER: | |
2511 | handle_fs_map_user(static_cast<MFSMapUser*>(m)); | |
2512 | break; | |
2513 | case CEPH_MSG_CLIENT_SESSION: | |
2514 | handle_client_session(static_cast<MClientSession*>(m)); | |
2515 | break; | |
2516 | ||
2517 | case CEPH_MSG_OSD_MAP: | |
2518 | handle_osd_map(static_cast<MOSDMap*>(m)); | |
2519 | break; | |
2520 | ||
2521 | // requests | |
2522 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
2523 | handle_client_request_forward(static_cast<MClientRequestForward*>(m)); | |
2524 | break; | |
2525 | case CEPH_MSG_CLIENT_REPLY: | |
2526 | handle_client_reply(static_cast<MClientReply*>(m)); | |
2527 | break; | |
2528 | ||
2529 | case CEPH_MSG_CLIENT_SNAP: | |
2530 | handle_snap(static_cast<MClientSnap*>(m)); | |
2531 | break; | |
2532 | case CEPH_MSG_CLIENT_CAPS: | |
2533 | handle_caps(static_cast<MClientCaps*>(m)); | |
2534 | break; | |
2535 | case CEPH_MSG_CLIENT_LEASE: | |
2536 | handle_lease(static_cast<MClientLease*>(m)); | |
2537 | break; | |
2538 | case MSG_COMMAND_REPLY: | |
2539 | if (m->get_source().type() == CEPH_ENTITY_TYPE_MDS) { | |
2540 | handle_command_reply(static_cast<MCommandReply*>(m)); | |
2541 | } else { | |
2542 | return false; | |
2543 | } | |
2544 | break; | |
2545 | case CEPH_MSG_CLIENT_QUOTA: | |
2546 | handle_quota(static_cast<MClientQuota*>(m)); | |
2547 | break; | |
2548 | ||
2549 | default: | |
2550 | return false; | |
2551 | } | |
2552 | ||
2553 | // unmounting? | |
2554 | if (unmounting) { | |
2555 | ldout(cct, 10) << "unmounting: trim pass, size was " << lru.lru_get_size() | |
2556 | << "+" << inode_map.size() << dendl; | |
2557 | long unsigned size = lru.lru_get_size() + inode_map.size(); | |
2558 | trim_cache(); | |
2559 | if (size < lru.lru_get_size() + inode_map.size()) { | |
2560 | ldout(cct, 10) << "unmounting: trim pass, cache shrank, poking unmount()" << dendl; | |
2561 | mount_cond.Signal(); | |
2562 | } else { | |
2563 | ldout(cct, 10) << "unmounting: trim pass, size still " << lru.lru_get_size() | |
2564 | << "+" << inode_map.size() << dendl; | |
2565 | } | |
2566 | } | |
2567 | ||
2568 | return true; | |
2569 | } | |
2570 | ||
2571 | void Client::handle_fs_map(MFSMap *m) | |
2572 | { | |
2573 | fsmap.reset(new FSMap(m->get_fsmap())); | |
2574 | m->put(); | |
2575 | ||
2576 | signal_cond_list(waiting_for_fsmap); | |
2577 | ||
2578 | monclient->sub_got("fsmap", fsmap->get_epoch()); | |
2579 | } | |
2580 | ||
2581 | void Client::handle_fs_map_user(MFSMapUser *m) | |
2582 | { | |
2583 | fsmap_user.reset(new FSMapUser); | |
2584 | *fsmap_user = m->get_fsmap(); | |
2585 | m->put(); | |
2586 | ||
2587 | monclient->sub_got("fsmap.user", fsmap_user->get_epoch()); | |
2588 | signal_cond_list(waiting_for_fsmap); | |
2589 | } | |
2590 | ||
2591 | void Client::handle_mds_map(MMDSMap* m) | |
2592 | { | |
2593 | if (m->get_epoch() <= mdsmap->get_epoch()) { | |
2594 | ldout(cct, 1) << "handle_mds_map epoch " << m->get_epoch() | |
2595 | << " is identical to or older than our " | |
2596 | << mdsmap->get_epoch() << dendl; | |
2597 | m->put(); | |
2598 | return; | |
2599 | } | |
2600 | ||
2601 | ldout(cct, 1) << "handle_mds_map epoch " << m->get_epoch() << dendl; | |
2602 | ||
2603 | std::unique_ptr<MDSMap> oldmap(new MDSMap); | |
2604 | oldmap.swap(mdsmap); | |
2605 | ||
2606 | mdsmap->decode(m->get_encoded()); | |
2607 | ||
2608 | // Cancel any commands for missing or laggy GIDs | |
2609 | std::list<ceph_tid_t> cancel_ops; | |
2610 | auto &commands = command_table.get_commands(); | |
2611 | for (const auto &i : commands) { | |
2612 | auto &op = i.second; | |
2613 | const mds_gid_t op_mds_gid = op.mds_gid; | |
2614 | if (mdsmap->is_dne_gid(op_mds_gid) || mdsmap->is_laggy_gid(op_mds_gid)) { | |
2615 | ldout(cct, 1) << __func__ << ": cancelling command op " << i.first << dendl; | |
2616 | cancel_ops.push_back(i.first); | |
2617 | if (op.outs) { | |
2618 | std::ostringstream ss; | |
2619 | ss << "MDS " << op_mds_gid << " went away"; | |
2620 | *(op.outs) = ss.str(); | |
2621 | } | |
2622 | op.con->mark_down(); | |
2623 | if (op.on_finish) { | |
2624 | op.on_finish->complete(-ETIMEDOUT); | |
2625 | } | |
2626 | } | |
2627 | } | |
2628 | ||
2629 | for (std::list<ceph_tid_t>::iterator i = cancel_ops.begin(); | |
2630 | i != cancel_ops.end(); ++i) { | |
2631 | command_table.erase(*i); | |
2632 | } | |
2633 | ||
2634 | // reset session | |
2635 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
2636 | p != mds_sessions.end(); ) { | |
2637 | mds_rank_t mds = p->first; | |
2638 | MetaSession *session = p->second; | |
2639 | ++p; | |
2640 | ||
2641 | int oldstate = oldmap->get_state(mds); | |
2642 | int newstate = mdsmap->get_state(mds); | |
2643 | if (!mdsmap->is_up(mds)) { | |
2644 | session->con->mark_down(); | |
2645 | } else if (mdsmap->get_inst(mds) != session->inst) { | |
2646 | session->con->mark_down(); | |
2647 | session->inst = mdsmap->get_inst(mds); | |
2648 | // When new MDS starts to take over, notify kernel to trim unused entries | |
2649 | // in its dcache/icache. Hopefully, the kernel will release some unused | |
2650 | // inodes before the new MDS enters reconnect state. | |
2651 | trim_cache_for_reconnect(session); | |
2652 | } else if (oldstate == newstate) | |
2653 | continue; // no change | |
2654 | ||
2655 | session->mds_state = newstate; | |
2656 | if (newstate == MDSMap::STATE_RECONNECT) { | |
2657 | session->con = messenger->get_connection(session->inst); | |
2658 | send_reconnect(session); | |
2659 | } else if (newstate >= MDSMap::STATE_ACTIVE) { | |
2660 | if (oldstate < MDSMap::STATE_ACTIVE) { | |
2661 | // kick new requests | |
2662 | kick_requests(session); | |
2663 | kick_flushing_caps(session); | |
2664 | signal_context_list(session->waiting_for_open); | |
2665 | kick_maxsize_requests(session); | |
2666 | wake_inode_waiters(session); | |
2667 | } | |
2668 | connect_mds_targets(mds); | |
2669 | } else if (newstate == MDSMap::STATE_NULL && | |
2670 | mds >= mdsmap->get_max_mds()) { | |
2671 | _closed_mds_session(session); | |
2672 | } | |
2673 | } | |
2674 | ||
2675 | // kick any waiting threads | |
2676 | signal_cond_list(waiting_for_mdsmap); | |
2677 | ||
2678 | m->put(); | |
2679 | ||
2680 | monclient->sub_got("mdsmap", mdsmap->get_epoch()); | |
2681 | } | |
2682 | ||
2683 | void Client::send_reconnect(MetaSession *session) | |
2684 | { | |
2685 | mds_rank_t mds = session->mds_num; | |
2686 | ldout(cct, 10) << "send_reconnect to mds." << mds << dendl; | |
2687 | ||
2688 | // trim unused caps to reduce MDS's cache rejoin time | |
2689 | trim_cache_for_reconnect(session); | |
2690 | ||
2691 | session->readonly = false; | |
2692 | ||
2693 | if (session->release) { | |
2694 | session->release->put(); | |
2695 | session->release = NULL; | |
2696 | } | |
2697 | ||
2698 | // reset my cap seq number | |
2699 | session->seq = 0; | |
2700 | //connect to the mds' offload targets | |
2701 | connect_mds_targets(mds); | |
2702 | //make sure unsafe requests get saved | |
2703 | resend_unsafe_requests(session); | |
2704 | ||
2705 | MClientReconnect *m = new MClientReconnect; | |
2706 | ||
2707 | // i have an open session. | |
2708 | ceph::unordered_set<inodeno_t> did_snaprealm; | |
2709 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
2710 | p != inode_map.end(); | |
2711 | ++p) { | |
2712 | Inode *in = p->second; | |
2713 | if (in->caps.count(mds)) { | |
2714 | ldout(cct, 10) << " caps on " << p->first | |
2715 | << " " << ccap_string(in->caps[mds]->issued) | |
2716 | << " wants " << ccap_string(in->caps_wanted()) | |
2717 | << dendl; | |
2718 | filepath path; | |
2719 | in->make_long_path(path); | |
2720 | ldout(cct, 10) << " path " << path << dendl; | |
2721 | ||
2722 | bufferlist flockbl; | |
2723 | _encode_filelocks(in, flockbl); | |
2724 | ||
2725 | Cap *cap = in->caps[mds]; | |
2726 | cap->seq = 0; // reset seq. | |
2727 | cap->issue_seq = 0; // reset seq. | |
2728 | cap->mseq = 0; // reset seq. | |
2729 | cap->issued = cap->implemented; | |
2730 | ||
2731 | snapid_t snap_follows = 0; | |
2732 | if (!in->cap_snaps.empty()) | |
2733 | snap_follows = in->cap_snaps.begin()->first; | |
2734 | ||
2735 | m->add_cap(p->first.ino, | |
2736 | cap->cap_id, | |
2737 | path.get_ino(), path.get_path(), // ino | |
2738 | in->caps_wanted(), // wanted | |
2739 | cap->issued, // issued | |
2740 | in->snaprealm->ino, | |
2741 | snap_follows, | |
2742 | flockbl); | |
2743 | ||
2744 | if (did_snaprealm.count(in->snaprealm->ino) == 0) { | |
2745 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
2746 | m->add_snaprealm(in->snaprealm->ino, in->snaprealm->seq, in->snaprealm->parent); | |
2747 | did_snaprealm.insert(in->snaprealm->ino); | |
2748 | } | |
2749 | } | |
2750 | } | |
2751 | ||
2752 | early_kick_flushing_caps(session); | |
2753 | ||
2754 | session->con->send_message(m); | |
2755 | ||
2756 | mount_cond.Signal(); | |
2757 | } | |
2758 | ||
2759 | ||
2760 | void Client::kick_requests(MetaSession *session) | |
2761 | { | |
2762 | ldout(cct, 10) << "kick_requests for mds." << session->mds_num << dendl; | |
2763 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
2764 | p != mds_requests.end(); | |
2765 | ++p) { | |
31f18b77 FG |
2766 | MetaRequest *req = p->second; |
2767 | if (req->got_unsafe) | |
2768 | continue; | |
2769 | if (req->aborted()) { | |
2770 | if (req->caller_cond) { | |
2771 | req->kick = true; | |
2772 | req->caller_cond->Signal(); | |
2773 | } | |
7c673cae | 2774 | continue; |
31f18b77 FG |
2775 | } |
2776 | if (req->retry_attempt > 0) | |
7c673cae | 2777 | continue; // new requests only |
31f18b77 | 2778 | if (req->mds == session->mds_num) { |
7c673cae FG |
2779 | send_request(p->second, session); |
2780 | } | |
2781 | } | |
2782 | } | |
2783 | ||
2784 | void Client::resend_unsafe_requests(MetaSession *session) | |
2785 | { | |
2786 | for (xlist<MetaRequest*>::iterator iter = session->unsafe_requests.begin(); | |
2787 | !iter.end(); | |
2788 | ++iter) | |
2789 | send_request(*iter, session); | |
2790 | ||
2791 | // also re-send old requests when MDS enters reconnect stage. So that MDS can | |
2792 | // process completed requests in clientreplay stage. | |
2793 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
2794 | p != mds_requests.end(); | |
2795 | ++p) { | |
2796 | MetaRequest *req = p->second; | |
2797 | if (req->got_unsafe) | |
2798 | continue; | |
31f18b77 FG |
2799 | if (req->aborted()) |
2800 | continue; | |
7c673cae FG |
2801 | if (req->retry_attempt == 0) |
2802 | continue; // old requests only | |
2803 | if (req->mds == session->mds_num) | |
2804 | send_request(req, session, true); | |
2805 | } | |
2806 | } | |
2807 | ||
2808 | void Client::wait_unsafe_requests() | |
2809 | { | |
2810 | list<MetaRequest*> last_unsafe_reqs; | |
2811 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
2812 | p != mds_sessions.end(); | |
2813 | ++p) { | |
2814 | MetaSession *s = p->second; | |
2815 | if (!s->unsafe_requests.empty()) { | |
2816 | MetaRequest *req = s->unsafe_requests.back(); | |
2817 | req->get(); | |
2818 | last_unsafe_reqs.push_back(req); | |
2819 | } | |
2820 | } | |
2821 | ||
2822 | for (list<MetaRequest*>::iterator p = last_unsafe_reqs.begin(); | |
2823 | p != last_unsafe_reqs.end(); | |
2824 | ++p) { | |
2825 | MetaRequest *req = *p; | |
2826 | if (req->unsafe_item.is_on_list()) | |
2827 | wait_on_list(req->waitfor_safe); | |
2828 | put_request(req); | |
2829 | } | |
2830 | } | |
2831 | ||
2832 | void Client::kick_requests_closed(MetaSession *session) | |
2833 | { | |
2834 | ldout(cct, 10) << "kick_requests_closed for mds." << session->mds_num << dendl; | |
2835 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
2836 | p != mds_requests.end(); ) { | |
2837 | MetaRequest *req = p->second; | |
2838 | ++p; | |
2839 | if (req->mds == session->mds_num) { | |
2840 | if (req->caller_cond) { | |
2841 | req->kick = true; | |
2842 | req->caller_cond->Signal(); | |
2843 | } | |
2844 | req->item.remove_myself(); | |
2845 | if (req->got_unsafe) { | |
2846 | lderr(cct) << "kick_requests_closed removing unsafe request " << req->get_tid() << dendl; | |
2847 | req->unsafe_item.remove_myself(); | |
2848 | req->unsafe_dir_item.remove_myself(); | |
2849 | req->unsafe_target_item.remove_myself(); | |
2850 | signal_cond_list(req->waitfor_safe); | |
2851 | unregister_request(req); | |
2852 | } | |
2853 | } | |
2854 | } | |
2855 | assert(session->requests.empty()); | |
2856 | assert(session->unsafe_requests.empty()); | |
2857 | } | |
2858 | ||
2859 | ||
2860 | ||
2861 | ||
2862 | /************ | |
2863 | * leases | |
2864 | */ | |
2865 | ||
2866 | void Client::got_mds_push(MetaSession *s) | |
2867 | { | |
2868 | s->seq++; | |
2869 | ldout(cct, 10) << " mds." << s->mds_num << " seq now " << s->seq << dendl; | |
2870 | if (s->state == MetaSession::STATE_CLOSING) { | |
2871 | s->con->send_message(new MClientSession(CEPH_SESSION_REQUEST_CLOSE, s->seq)); | |
2872 | } | |
2873 | } | |
2874 | ||
2875 | void Client::handle_lease(MClientLease *m) | |
2876 | { | |
2877 | ldout(cct, 10) << "handle_lease " << *m << dendl; | |
2878 | ||
2879 | assert(m->get_action() == CEPH_MDS_LEASE_REVOKE); | |
2880 | ||
2881 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
2882 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
2883 | if (!session) { | |
2884 | m->put(); | |
2885 | return; | |
2886 | } | |
2887 | ||
2888 | got_mds_push(session); | |
2889 | ||
2890 | ceph_seq_t seq = m->get_seq(); | |
2891 | ||
2892 | Inode *in; | |
2893 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
2894 | if (inode_map.count(vino) == 0) { | |
2895 | ldout(cct, 10) << " don't have vino " << vino << dendl; | |
2896 | goto revoke; | |
2897 | } | |
2898 | in = inode_map[vino]; | |
2899 | ||
2900 | if (m->get_mask() & CEPH_LOCK_DN) { | |
2901 | if (!in->dir || in->dir->dentries.count(m->dname) == 0) { | |
2902 | ldout(cct, 10) << " don't have dir|dentry " << m->get_ino() << "/" << m->dname <<dendl; | |
2903 | goto revoke; | |
2904 | } | |
2905 | Dentry *dn = in->dir->dentries[m->dname]; | |
2906 | ldout(cct, 10) << " revoked DN lease on " << dn << dendl; | |
2907 | dn->lease_mds = -1; | |
2908 | } | |
2909 | ||
2910 | revoke: | |
2911 | m->get_connection()->send_message( | |
2912 | new MClientLease( | |
2913 | CEPH_MDS_LEASE_RELEASE, seq, | |
2914 | m->get_mask(), m->get_ino(), m->get_first(), m->get_last(), m->dname)); | |
2915 | m->put(); | |
2916 | } | |
2917 | ||
2918 | void Client::put_inode(Inode *in, int n) | |
2919 | { | |
2920 | ldout(cct, 10) << "put_inode on " << *in << dendl; | |
2921 | int left = in->_put(n); | |
2922 | if (left == 0) { | |
2923 | // release any caps | |
2924 | remove_all_caps(in); | |
2925 | ||
2926 | ldout(cct, 10) << "put_inode deleting " << *in << dendl; | |
2927 | bool unclean = objectcacher->release_set(&in->oset); | |
2928 | assert(!unclean); | |
2929 | inode_map.erase(in->vino()); | |
2930 | if (use_faked_inos()) | |
2931 | _release_faked_ino(in); | |
2932 | ||
2933 | if (in == root) { | |
2934 | root = 0; | |
2935 | root_ancestor = 0; | |
2936 | while (!root_parents.empty()) | |
2937 | root_parents.erase(root_parents.begin()); | |
2938 | } | |
2939 | ||
2940 | delete in; | |
2941 | } | |
2942 | } | |
2943 | ||
2944 | void Client::close_dir(Dir *dir) | |
2945 | { | |
2946 | Inode *in = dir->parent_inode; | |
2947 | ldout(cct, 15) << "close_dir dir " << dir << " on " << in << dendl; | |
2948 | assert(dir->is_empty()); | |
2949 | assert(in->dir == dir); | |
2950 | assert(in->dn_set.size() < 2); // dirs can't be hard-linked | |
2951 | if (!in->dn_set.empty()) | |
2952 | in->get_first_parent()->put(); // unpin dentry | |
2953 | ||
2954 | delete in->dir; | |
2955 | in->dir = 0; | |
2956 | put_inode(in); // unpin inode | |
2957 | } | |
2958 | ||
2959 | /** | |
2960 | * Don't call this with in==NULL, use get_or_create for that | |
2961 | * leave dn set to default NULL unless you're trying to add | |
2962 | * a new inode to a pre-created Dentry | |
2963 | */ | |
2964 | Dentry* Client::link(Dir *dir, const string& name, Inode *in, Dentry *dn) | |
2965 | { | |
2966 | if (!dn) { | |
2967 | // create a new Dentry | |
2968 | dn = new Dentry; | |
2969 | dn->name = name; | |
2970 | ||
2971 | // link to dir | |
2972 | dn->dir = dir; | |
2973 | dir->dentries[dn->name] = dn; | |
2974 | lru.lru_insert_mid(dn); // mid or top? | |
2975 | ||
2976 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
2977 | << " dn " << dn << " (new dn)" << dendl; | |
2978 | } else { | |
2979 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
2980 | << " dn " << dn << " (old dn)" << dendl; | |
2981 | } | |
2982 | ||
2983 | if (in) { // link to inode | |
2984 | dn->inode = in; | |
2985 | if (in->is_dir()) { | |
2986 | if (in->dir) | |
2987 | dn->get(); // dir -> dn pin | |
2988 | if (in->ll_ref) | |
2989 | dn->get(); // ll_ref -> dn pin | |
2990 | } | |
2991 | ||
2992 | assert(in->dn_set.count(dn) == 0); | |
2993 | ||
2994 | // only one parent for directories! | |
2995 | if (in->is_dir() && !in->dn_set.empty()) { | |
2996 | Dentry *olddn = in->get_first_parent(); | |
2997 | assert(olddn->dir != dir || olddn->name != name); | |
2998 | Inode *old_diri = olddn->dir->parent_inode; | |
2999 | old_diri->dir_release_count++; | |
3000 | clear_dir_complete_and_ordered(old_diri, true); | |
3001 | unlink(olddn, true, true); // keep dir, dentry | |
3002 | } | |
3003 | ||
3004 | in->dn_set.insert(dn); | |
3005 | ||
3006 | ldout(cct, 20) << "link inode " << in << " parents now " << in->dn_set << dendl; | |
3007 | } | |
3008 | ||
3009 | return dn; | |
3010 | } | |
3011 | ||
3012 | void Client::unlink(Dentry *dn, bool keepdir, bool keepdentry) | |
3013 | { | |
3014 | InodeRef in; | |
3015 | in.swap(dn->inode); | |
3016 | ldout(cct, 15) << "unlink dir " << dn->dir->parent_inode << " '" << dn->name << "' dn " << dn | |
3017 | << " inode " << dn->inode << dendl; | |
3018 | ||
3019 | // unlink from inode | |
3020 | if (in) { | |
3021 | if (in->is_dir()) { | |
3022 | if (in->dir) | |
3023 | dn->put(); // dir -> dn pin | |
3024 | if (in->ll_ref) | |
3025 | dn->put(); // ll_ref -> dn pin | |
3026 | } | |
3027 | dn->inode = 0; | |
3028 | assert(in->dn_set.count(dn)); | |
3029 | in->dn_set.erase(dn); | |
3030 | ldout(cct, 20) << "unlink inode " << in << " parents now " << in->dn_set << dendl; | |
3031 | } | |
3032 | ||
3033 | if (keepdentry) { | |
3034 | dn->lease_mds = -1; | |
3035 | } else { | |
3036 | ldout(cct, 15) << "unlink removing '" << dn->name << "' dn " << dn << dendl; | |
3037 | ||
3038 | // unlink from dir | |
3039 | dn->dir->dentries.erase(dn->name); | |
3040 | if (dn->dir->is_empty() && !keepdir) | |
3041 | close_dir(dn->dir); | |
3042 | dn->dir = 0; | |
3043 | ||
3044 | // delete den | |
3045 | lru.lru_remove(dn); | |
3046 | dn->put(); | |
3047 | } | |
3048 | } | |
3049 | ||
3050 | /** | |
3051 | * For asynchronous flushes, check for errors from the IO and | |
3052 | * update the inode if necessary | |
3053 | */ | |
3054 | class C_Client_FlushComplete : public Context { | |
3055 | private: | |
3056 | Client *client; | |
3057 | InodeRef inode; | |
3058 | public: | |
3059 | C_Client_FlushComplete(Client *c, Inode *in) : client(c), inode(in) { } | |
3060 | void finish(int r) override { | |
3061 | assert(client->client_lock.is_locked_by_me()); | |
3062 | if (r != 0) { | |
3063 | client_t const whoami = client->whoami; // For the benefit of ldout prefix | |
3064 | ldout(client->cct, 1) << "I/O error from flush on inode " << inode | |
3065 | << " 0x" << std::hex << inode->ino << std::dec | |
3066 | << ": " << r << "(" << cpp_strerror(r) << ")" << dendl; | |
3067 | inode->set_async_err(r); | |
3068 | } | |
3069 | } | |
3070 | }; | |
3071 | ||
3072 | ||
3073 | /**** | |
3074 | * caps | |
3075 | */ | |
3076 | ||
3077 | void Client::get_cap_ref(Inode *in, int cap) | |
3078 | { | |
3079 | if ((cap & CEPH_CAP_FILE_BUFFER) && | |
3080 | in->cap_refs[CEPH_CAP_FILE_BUFFER] == 0) { | |
3081 | ldout(cct, 5) << "get_cap_ref got first FILE_BUFFER ref on " << *in << dendl; | |
3082 | in->get(); | |
3083 | } | |
3084 | if ((cap & CEPH_CAP_FILE_CACHE) && | |
3085 | in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
3086 | ldout(cct, 5) << "get_cap_ref got first FILE_CACHE ref on " << *in << dendl; | |
3087 | in->get(); | |
3088 | } | |
3089 | in->get_cap_ref(cap); | |
3090 | } | |
3091 | ||
3092 | void Client::put_cap_ref(Inode *in, int cap) | |
3093 | { | |
3094 | int last = in->put_cap_ref(cap); | |
3095 | if (last) { | |
3096 | int put_nref = 0; | |
3097 | int drop = last & ~in->caps_issued(); | |
3098 | if (in->snapid == CEPH_NOSNAP) { | |
3099 | if ((last & CEPH_CAP_FILE_WR) && | |
3100 | !in->cap_snaps.empty() && | |
3101 | in->cap_snaps.rbegin()->second.writing) { | |
3102 | ldout(cct, 10) << "put_cap_ref finishing pending cap_snap on " << *in << dendl; | |
3103 | in->cap_snaps.rbegin()->second.writing = 0; | |
3104 | finish_cap_snap(in, in->cap_snaps.rbegin()->second, get_caps_used(in)); | |
3105 | signal_cond_list(in->waitfor_caps); // wake up blocked sync writers | |
3106 | } | |
3107 | if (last & CEPH_CAP_FILE_BUFFER) { | |
3108 | for (auto &p : in->cap_snaps) | |
3109 | p.second.dirty_data = 0; | |
3110 | signal_cond_list(in->waitfor_commit); | |
3111 | ldout(cct, 5) << "put_cap_ref dropped last FILE_BUFFER ref on " << *in << dendl; | |
3112 | ++put_nref; | |
3113 | } | |
3114 | } | |
3115 | if (last & CEPH_CAP_FILE_CACHE) { | |
3116 | ldout(cct, 5) << "put_cap_ref dropped last FILE_CACHE ref on " << *in << dendl; | |
3117 | ++put_nref; | |
3118 | } | |
3119 | if (drop) | |
3120 | check_caps(in, 0); | |
3121 | if (put_nref) | |
3122 | put_inode(in, put_nref); | |
3123 | } | |
3124 | } | |
3125 | ||
3126 | int Client::get_caps(Inode *in, int need, int want, int *phave, loff_t endoff) | |
3127 | { | |
3128 | int r = check_pool_perm(in, need); | |
3129 | if (r < 0) | |
3130 | return r; | |
3131 | ||
3132 | while (1) { | |
3133 | int file_wanted = in->caps_file_wanted(); | |
3134 | if ((file_wanted & need) != need) { | |
3135 | ldout(cct, 10) << "get_caps " << *in << " need " << ccap_string(need) | |
3136 | << " file_wanted " << ccap_string(file_wanted) << ", EBADF " | |
3137 | << dendl; | |
3138 | return -EBADF; | |
3139 | } | |
3140 | ||
3141 | int implemented; | |
3142 | int have = in->caps_issued(&implemented); | |
3143 | ||
3144 | bool waitfor_caps = false; | |
3145 | bool waitfor_commit = false; | |
3146 | ||
3147 | if (have & need & CEPH_CAP_FILE_WR) { | |
3148 | if (endoff > 0 && | |
3149 | (endoff >= (loff_t)in->max_size || | |
3150 | endoff > (loff_t)(in->size << 1)) && | |
3151 | endoff > (loff_t)in->wanted_max_size) { | |
3152 | ldout(cct, 10) << "wanted_max_size " << in->wanted_max_size << " -> " << endoff << dendl; | |
3153 | in->wanted_max_size = endoff; | |
3154 | check_caps(in, 0); | |
3155 | } | |
3156 | ||
3157 | if (endoff >= 0 && endoff > (loff_t)in->max_size) { | |
3158 | ldout(cct, 10) << "waiting on max_size, endoff " << endoff << " max_size " << in->max_size << " on " << *in << dendl; | |
3159 | waitfor_caps = true; | |
3160 | } | |
3161 | if (!in->cap_snaps.empty()) { | |
3162 | if (in->cap_snaps.rbegin()->second.writing) { | |
3163 | ldout(cct, 10) << "waiting on cap_snap write to complete" << dendl; | |
3164 | waitfor_caps = true; | |
3165 | } | |
3166 | for (auto &p : in->cap_snaps) { | |
3167 | if (p.second.dirty_data) { | |
3168 | waitfor_commit = true; | |
3169 | break; | |
3170 | } | |
3171 | } | |
3172 | if (waitfor_commit) { | |
3173 | _flush(in, new C_Client_FlushComplete(this, in)); | |
3174 | ldout(cct, 10) << "waiting for WRBUFFER to get dropped" << dendl; | |
3175 | } | |
3176 | } | |
3177 | } | |
3178 | ||
3179 | if (!waitfor_caps && !waitfor_commit) { | |
3180 | if ((have & need) == need) { | |
7c673cae FG |
3181 | int revoking = implemented & ~have; |
3182 | ldout(cct, 10) << "get_caps " << *in << " have " << ccap_string(have) | |
3183 | << " need " << ccap_string(need) << " want " << ccap_string(want) | |
c07f9fc5 | 3184 | << " revoking " << ccap_string(revoking) |
7c673cae | 3185 | << dendl; |
c07f9fc5 | 3186 | if ((revoking & want) == 0) { |
7c673cae FG |
3187 | *phave = need | (have & want); |
3188 | in->get_cap_ref(need); | |
3189 | return 0; | |
3190 | } | |
3191 | } | |
3192 | ldout(cct, 10) << "waiting for caps " << *in << " need " << ccap_string(need) << " want " << ccap_string(want) << dendl; | |
3193 | waitfor_caps = true; | |
3194 | } | |
3195 | ||
3196 | if ((need & CEPH_CAP_FILE_WR) && in->auth_cap && | |
3197 | in->auth_cap->session->readonly) | |
3198 | return -EROFS; | |
3199 | ||
3200 | if (in->flags & I_CAP_DROPPED) { | |
3201 | int mds_wanted = in->caps_mds_wanted(); | |
3202 | if ((mds_wanted & need) != need) { | |
3203 | int ret = _renew_caps(in); | |
3204 | if (ret < 0) | |
3205 | return ret; | |
3206 | continue; | |
3207 | } | |
3208 | if ((mds_wanted & file_wanted) == | |
3209 | (file_wanted & (CEPH_CAP_FILE_RD | CEPH_CAP_FILE_WR))) { | |
3210 | in->flags &= ~I_CAP_DROPPED; | |
3211 | } | |
3212 | } | |
3213 | ||
3214 | if (waitfor_caps) | |
3215 | wait_on_list(in->waitfor_caps); | |
3216 | else if (waitfor_commit) | |
3217 | wait_on_list(in->waitfor_commit); | |
3218 | } | |
3219 | } | |
3220 | ||
3221 | int Client::get_caps_used(Inode *in) | |
3222 | { | |
3223 | unsigned used = in->caps_used(); | |
3224 | if (!(used & CEPH_CAP_FILE_CACHE) && | |
3225 | !objectcacher->set_is_empty(&in->oset)) | |
3226 | used |= CEPH_CAP_FILE_CACHE; | |
3227 | return used; | |
3228 | } | |
3229 | ||
3230 | void Client::cap_delay_requeue(Inode *in) | |
3231 | { | |
3232 | ldout(cct, 10) << "cap_delay_requeue on " << *in << dendl; | |
3233 | in->hold_caps_until = ceph_clock_now(); | |
3234 | in->hold_caps_until += cct->_conf->client_caps_release_delay; | |
3235 | delayed_caps.push_back(&in->cap_item); | |
3236 | } | |
3237 | ||
3238 | void Client::send_cap(Inode *in, MetaSession *session, Cap *cap, | |
3239 | bool sync, int used, int want, int retain, | |
3240 | int flush, ceph_tid_t flush_tid) | |
3241 | { | |
3242 | int held = cap->issued | cap->implemented; | |
3243 | int revoking = cap->implemented & ~cap->issued; | |
3244 | retain &= ~revoking; | |
3245 | int dropping = cap->issued & ~retain; | |
3246 | int op = CEPH_CAP_OP_UPDATE; | |
3247 | ||
3248 | ldout(cct, 10) << "send_cap " << *in | |
3249 | << " mds." << session->mds_num << " seq " << cap->seq | |
3250 | << (sync ? " sync " : " async ") | |
3251 | << " used " << ccap_string(used) | |
3252 | << " want " << ccap_string(want) | |
3253 | << " flush " << ccap_string(flush) | |
3254 | << " retain " << ccap_string(retain) | |
3255 | << " held "<< ccap_string(held) | |
3256 | << " revoking " << ccap_string(revoking) | |
3257 | << " dropping " << ccap_string(dropping) | |
3258 | << dendl; | |
3259 | ||
3260 | if (cct->_conf->client_inject_release_failure && revoking) { | |
3261 | const int would_have_issued = cap->issued & retain; | |
3262 | const int would_have_implemented = cap->implemented & (cap->issued | used); | |
3263 | // Simulated bug: | |
3264 | // - tell the server we think issued is whatever they issued plus whatever we implemented | |
3265 | // - leave what we have implemented in place | |
3266 | ldout(cct, 20) << __func__ << " injecting failure to release caps" << dendl; | |
3267 | cap->issued = cap->issued | cap->implemented; | |
3268 | ||
3269 | // Make an exception for revoking xattr caps: we are injecting | |
3270 | // failure to release other caps, but allow xattr because client | |
3271 | // will block on xattr ops if it can't release these to MDS (#9800) | |
3272 | const int xattr_mask = CEPH_CAP_XATTR_SHARED | CEPH_CAP_XATTR_EXCL; | |
3273 | cap->issued ^= xattr_mask & revoking; | |
3274 | cap->implemented ^= xattr_mask & revoking; | |
3275 | ||
3276 | ldout(cct, 20) << __func__ << " issued " << ccap_string(cap->issued) << " vs " << ccap_string(would_have_issued) << dendl; | |
3277 | ldout(cct, 20) << __func__ << " implemented " << ccap_string(cap->implemented) << " vs " << ccap_string(would_have_implemented) << dendl; | |
3278 | } else { | |
3279 | // Normal behaviour | |
3280 | cap->issued &= retain; | |
3281 | cap->implemented &= cap->issued | used; | |
3282 | } | |
3283 | ||
3284 | snapid_t follows = 0; | |
3285 | ||
3286 | if (flush) | |
3287 | follows = in->snaprealm->get_snap_context().seq; | |
3288 | ||
3289 | MClientCaps *m = new MClientCaps(op, | |
3290 | in->ino, | |
3291 | 0, | |
3292 | cap->cap_id, cap->seq, | |
3293 | cap->implemented, | |
3294 | want, | |
3295 | flush, | |
3296 | cap->mseq, | |
3297 | cap_epoch_barrier); | |
3298 | m->caller_uid = in->cap_dirtier_uid; | |
3299 | m->caller_gid = in->cap_dirtier_gid; | |
3300 | ||
3301 | m->head.issue_seq = cap->issue_seq; | |
3302 | m->set_tid(flush_tid); | |
3303 | ||
3304 | m->head.uid = in->uid; | |
3305 | m->head.gid = in->gid; | |
3306 | m->head.mode = in->mode; | |
3307 | ||
3308 | m->head.nlink = in->nlink; | |
3309 | ||
3310 | if (flush & CEPH_CAP_XATTR_EXCL) { | |
3311 | ::encode(in->xattrs, m->xattrbl); | |
3312 | m->head.xattr_version = in->xattr_version; | |
3313 | } | |
3314 | ||
3315 | m->size = in->size; | |
3316 | m->max_size = in->max_size; | |
3317 | m->truncate_seq = in->truncate_seq; | |
3318 | m->truncate_size = in->truncate_size; | |
3319 | m->mtime = in->mtime; | |
3320 | m->atime = in->atime; | |
3321 | m->ctime = in->ctime; | |
3322 | m->btime = in->btime; | |
3323 | m->time_warp_seq = in->time_warp_seq; | |
3324 | m->change_attr = in->change_attr; | |
3325 | if (sync) | |
3326 | m->flags |= CLIENT_CAPS_SYNC; | |
3327 | ||
3328 | if (flush & CEPH_CAP_FILE_WR) { | |
3329 | m->inline_version = in->inline_version; | |
3330 | m->inline_data = in->inline_data; | |
3331 | } | |
3332 | ||
3333 | in->reported_size = in->size; | |
3334 | m->set_snap_follows(follows); | |
3335 | cap->wanted = want; | |
3336 | if (cap == in->auth_cap) { | |
3337 | m->set_max_size(in->wanted_max_size); | |
3338 | in->requested_max_size = in->wanted_max_size; | |
3339 | ldout(cct, 15) << "auth cap, setting max_size = " << in->requested_max_size << dendl; | |
3340 | } | |
3341 | ||
3342 | if (!session->flushing_caps_tids.empty()) | |
3343 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3344 | ||
3345 | session->con->send_message(m); | |
3346 | } | |
3347 | ||
31f18b77 FG |
3348 | static bool is_max_size_approaching(Inode *in) |
3349 | { | |
3350 | /* mds will adjust max size according to the reported size */ | |
3351 | if (in->flushing_caps & CEPH_CAP_FILE_WR) | |
3352 | return false; | |
3353 | if (in->size >= in->max_size) | |
3354 | return true; | |
3355 | /* half of previous max_size increment has been used */ | |
3356 | if (in->max_size > in->reported_size && | |
3357 | (in->size << 1) >= in->max_size + in->reported_size) | |
3358 | return true; | |
3359 | return false; | |
3360 | } | |
7c673cae FG |
3361 | |
3362 | /** | |
3363 | * check_caps | |
3364 | * | |
3365 | * Examine currently used and wanted versus held caps. Release, flush or ack | |
3366 | * revoked caps to the MDS as appropriate. | |
3367 | * | |
3368 | * @param in the inode to check | |
3369 | * @param flags flags to apply to cap check | |
3370 | */ | |
3371 | void Client::check_caps(Inode *in, unsigned flags) | |
3372 | { | |
3373 | unsigned wanted = in->caps_wanted(); | |
3374 | unsigned used = get_caps_used(in); | |
3375 | unsigned cap_used; | |
3376 | ||
3377 | if (in->is_dir() && (in->flags & I_COMPLETE)) { | |
3378 | // we do this here because we don't want to drop to Fs (and then | |
3379 | // drop the Fs if we do a create!) if that alone makes us send lookups | |
3380 | // to the MDS. Doing it in in->caps_wanted() has knock-on effects elsewhere | |
3381 | wanted |= CEPH_CAP_FILE_EXCL; | |
3382 | } | |
3383 | ||
3384 | int implemented; | |
3385 | int issued = in->caps_issued(&implemented); | |
3386 | int revoking = implemented & ~issued; | |
3387 | ||
3388 | int retain = wanted | used | CEPH_CAP_PIN; | |
3389 | if (!unmounting) { | |
3390 | if (wanted) | |
3391 | retain |= CEPH_CAP_ANY; | |
3392 | else | |
3393 | retain |= CEPH_CAP_ANY_SHARED; | |
3394 | } | |
3395 | ||
3396 | ldout(cct, 10) << "check_caps on " << *in | |
3397 | << " wanted " << ccap_string(wanted) | |
3398 | << " used " << ccap_string(used) | |
3399 | << " issued " << ccap_string(issued) | |
3400 | << " revoking " << ccap_string(revoking) | |
3401 | << " flags=" << flags | |
3402 | << dendl; | |
3403 | ||
3404 | if (in->snapid != CEPH_NOSNAP) | |
3405 | return; //snap caps last forever, can't write | |
3406 | ||
3407 | if (in->caps.empty()) | |
3408 | return; // guard if at end of func | |
3409 | ||
3410 | if ((revoking & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) && | |
3411 | (used & CEPH_CAP_FILE_CACHE) && !(used & CEPH_CAP_FILE_BUFFER)) | |
3412 | _release(in); | |
3413 | ||
3414 | if (!in->cap_snaps.empty()) | |
3415 | flush_snaps(in); | |
3416 | ||
3417 | if (flags & CHECK_CAPS_NODELAY) | |
3418 | in->hold_caps_until = utime_t(); | |
3419 | else | |
3420 | cap_delay_requeue(in); | |
3421 | ||
3422 | utime_t now = ceph_clock_now(); | |
3423 | ||
3424 | map<mds_rank_t, Cap*>::iterator it = in->caps.begin(); | |
3425 | while (it != in->caps.end()) { | |
3426 | mds_rank_t mds = it->first; | |
3427 | Cap *cap = it->second; | |
3428 | ++it; | |
3429 | ||
3430 | MetaSession *session = mds_sessions[mds]; | |
3431 | assert(session); | |
3432 | ||
3433 | cap_used = used; | |
3434 | if (in->auth_cap && cap != in->auth_cap) | |
3435 | cap_used &= ~in->auth_cap->issued; | |
3436 | ||
3437 | revoking = cap->implemented & ~cap->issued; | |
3438 | ||
3439 | ldout(cct, 10) << " cap mds." << mds | |
3440 | << " issued " << ccap_string(cap->issued) | |
3441 | << " implemented " << ccap_string(cap->implemented) | |
3442 | << " revoking " << ccap_string(revoking) << dendl; | |
3443 | ||
3444 | if (in->wanted_max_size > in->max_size && | |
3445 | in->wanted_max_size > in->requested_max_size && | |
3446 | cap == in->auth_cap) | |
3447 | goto ack; | |
3448 | ||
3449 | /* approaching file_max? */ | |
3450 | if ((cap->issued & CEPH_CAP_FILE_WR) && | |
31f18b77 FG |
3451 | cap == in->auth_cap && |
3452 | is_max_size_approaching(in)) { | |
7c673cae | 3453 | ldout(cct, 10) << "size " << in->size << " approaching max_size " << in->max_size |
31f18b77 | 3454 | << ", reported " << in->reported_size << dendl; |
7c673cae FG |
3455 | goto ack; |
3456 | } | |
3457 | ||
3458 | /* completed revocation? */ | |
3459 | if (revoking && (revoking & cap_used) == 0) { | |
3460 | ldout(cct, 10) << "completed revocation of " << ccap_string(cap->implemented & ~cap->issued) << dendl; | |
3461 | goto ack; | |
3462 | } | |
3463 | ||
3464 | /* want more caps from mds? */ | |
3465 | if (wanted & ~(cap->wanted | cap->issued)) | |
3466 | goto ack; | |
3467 | ||
3468 | if (!revoking && unmounting && (cap_used == 0)) | |
3469 | goto ack; | |
3470 | ||
3471 | if (wanted == cap->wanted && // mds knows what we want. | |
3472 | ((cap->issued & ~retain) == 0) &&// and we don't have anything we wouldn't like | |
3473 | !in->dirty_caps) // and we have no dirty caps | |
3474 | continue; | |
3475 | ||
3476 | if (now < in->hold_caps_until) { | |
3477 | ldout(cct, 10) << "delaying cap release" << dendl; | |
3478 | continue; | |
3479 | } | |
3480 | ||
3481 | ack: | |
3482 | // re-send old cap/snapcap flushes first. | |
3483 | if (session->mds_state >= MDSMap::STATE_RECONNECT && | |
3484 | session->mds_state < MDSMap::STATE_ACTIVE && | |
3485 | session->early_flushing_caps.count(in) == 0) { | |
3486 | ldout(cct, 20) << " reflushing caps (check_caps) on " << *in | |
3487 | << " to mds." << session->mds_num << dendl; | |
3488 | session->early_flushing_caps.insert(in); | |
3489 | if (in->cap_snaps.size()) | |
3490 | flush_snaps(in, true); | |
3491 | if (in->flushing_caps) | |
3492 | flush_caps(in, session, flags & CHECK_CAPS_SYNCHRONOUS); | |
3493 | } | |
3494 | ||
3495 | int flushing; | |
3496 | ceph_tid_t flush_tid; | |
3497 | if (in->auth_cap == cap && in->dirty_caps) { | |
3498 | flushing = mark_caps_flushing(in, &flush_tid); | |
3499 | } else { | |
3500 | flushing = 0; | |
3501 | flush_tid = 0; | |
3502 | } | |
3503 | ||
3504 | send_cap(in, session, cap, flags & CHECK_CAPS_SYNCHRONOUS, cap_used, wanted, | |
3505 | retain, flushing, flush_tid); | |
3506 | } | |
3507 | } | |
3508 | ||
3509 | ||
3510 | void Client::queue_cap_snap(Inode *in, SnapContext& old_snapc) | |
3511 | { | |
3512 | int used = get_caps_used(in); | |
3513 | int dirty = in->caps_dirty(); | |
3514 | ldout(cct, 10) << "queue_cap_snap " << *in << " snapc " << old_snapc << " used " << ccap_string(used) << dendl; | |
3515 | ||
3516 | if (in->cap_snaps.size() && | |
3517 | in->cap_snaps.rbegin()->second.writing) { | |
3518 | ldout(cct, 10) << "queue_cap_snap already have pending cap_snap on " << *in << dendl; | |
3519 | return; | |
3520 | } else if (in->caps_dirty() || | |
3521 | (used & CEPH_CAP_FILE_WR) || | |
3522 | (dirty & CEPH_CAP_ANY_WR)) { | |
3523 | const auto &capsnapem = in->cap_snaps.emplace(std::piecewise_construct, std::make_tuple(old_snapc.seq), std::make_tuple(in)); | |
3524 | assert(capsnapem.second == true); /* element inserted */ | |
3525 | CapSnap &capsnap = capsnapem.first->second; | |
3526 | capsnap.context = old_snapc; | |
3527 | capsnap.issued = in->caps_issued(); | |
3528 | capsnap.dirty = in->caps_dirty(); | |
3529 | ||
3530 | capsnap.dirty_data = (used & CEPH_CAP_FILE_BUFFER); | |
3531 | ||
3532 | capsnap.uid = in->uid; | |
3533 | capsnap.gid = in->gid; | |
3534 | capsnap.mode = in->mode; | |
3535 | capsnap.btime = in->btime; | |
3536 | capsnap.xattrs = in->xattrs; | |
3537 | capsnap.xattr_version = in->xattr_version; | |
3538 | ||
3539 | if (used & CEPH_CAP_FILE_WR) { | |
3540 | ldout(cct, 10) << "queue_cap_snap WR used on " << *in << dendl; | |
3541 | capsnap.writing = 1; | |
3542 | } else { | |
3543 | finish_cap_snap(in, capsnap, used); | |
3544 | } | |
3545 | } else { | |
3546 | ldout(cct, 10) << "queue_cap_snap not dirty|writing on " << *in << dendl; | |
3547 | } | |
3548 | } | |
3549 | ||
3550 | void Client::finish_cap_snap(Inode *in, CapSnap &capsnap, int used) | |
3551 | { | |
3552 | ldout(cct, 10) << "finish_cap_snap " << *in << " capsnap " << (void *)&capsnap << " used " << ccap_string(used) << dendl; | |
3553 | capsnap.size = in->size; | |
3554 | capsnap.mtime = in->mtime; | |
3555 | capsnap.atime = in->atime; | |
3556 | capsnap.ctime = in->ctime; | |
3557 | capsnap.time_warp_seq = in->time_warp_seq; | |
3558 | capsnap.change_attr = in->change_attr; | |
3559 | ||
3560 | capsnap.dirty |= in->caps_dirty(); | |
3561 | ||
3562 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
3563 | capsnap.inline_data = in->inline_data; | |
3564 | capsnap.inline_version = in->inline_version; | |
3565 | } | |
3566 | ||
3567 | if (used & CEPH_CAP_FILE_BUFFER) { | |
3568 | ldout(cct, 10) << "finish_cap_snap " << *in << " cap_snap " << &capsnap << " used " << used | |
3569 | << " WRBUFFER, delaying" << dendl; | |
3570 | } else { | |
3571 | capsnap.dirty_data = 0; | |
3572 | flush_snaps(in); | |
3573 | } | |
3574 | } | |
3575 | ||
3576 | void Client::_flushed_cap_snap(Inode *in, snapid_t seq) | |
3577 | { | |
3578 | ldout(cct, 10) << "_flushed_cap_snap seq " << seq << " on " << *in << dendl; | |
3579 | in->cap_snaps.at(seq).dirty_data = 0; | |
3580 | flush_snaps(in); | |
3581 | } | |
3582 | ||
3583 | void Client::flush_snaps(Inode *in, bool all_again) | |
3584 | { | |
3585 | ldout(cct, 10) << "flush_snaps on " << *in << " all_again " << all_again << dendl; | |
3586 | assert(in->cap_snaps.size()); | |
3587 | ||
3588 | // pick auth mds | |
3589 | assert(in->auth_cap); | |
3590 | MetaSession *session = in->auth_cap->session; | |
3591 | int mseq = in->auth_cap->mseq; | |
3592 | ||
3593 | for (auto &p : in->cap_snaps) { | |
3594 | CapSnap &capsnap = p.second; | |
3595 | if (!all_again) { | |
3596 | // only flush once per session | |
3597 | if (capsnap.flush_tid > 0) | |
3598 | continue; | |
3599 | } | |
3600 | ||
3601 | ldout(cct, 10) << "flush_snaps mds." << session->mds_num | |
3602 | << " follows " << p.first | |
3603 | << " size " << capsnap.size | |
3604 | << " mtime " << capsnap.mtime | |
3605 | << " dirty_data=" << capsnap.dirty_data | |
3606 | << " writing=" << capsnap.writing | |
3607 | << " on " << *in << dendl; | |
3608 | if (capsnap.dirty_data || capsnap.writing) | |
3609 | continue; | |
3610 | ||
3611 | if (capsnap.flush_tid == 0) { | |
3612 | capsnap.flush_tid = ++last_flush_tid; | |
3613 | if (!in->flushing_cap_item.is_on_list()) | |
3614 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
3615 | session->flushing_caps_tids.insert(capsnap.flush_tid); | |
3616 | } | |
3617 | ||
3618 | MClientCaps *m = new MClientCaps(CEPH_CAP_OP_FLUSHSNAP, in->ino, in->snaprealm->ino, 0, mseq, | |
3619 | cap_epoch_barrier); | |
3620 | if (user_id >= 0) | |
3621 | m->caller_uid = user_id; | |
3622 | if (group_id >= 0) | |
3623 | m->caller_gid = group_id; | |
3624 | ||
3625 | m->set_client_tid(capsnap.flush_tid); | |
3626 | m->head.snap_follows = p.first; | |
3627 | ||
3628 | m->head.caps = capsnap.issued; | |
3629 | m->head.dirty = capsnap.dirty; | |
3630 | ||
3631 | m->head.uid = capsnap.uid; | |
3632 | m->head.gid = capsnap.gid; | |
3633 | m->head.mode = capsnap.mode; | |
3634 | m->btime = capsnap.btime; | |
3635 | ||
3636 | m->size = capsnap.size; | |
3637 | ||
3638 | m->head.xattr_version = capsnap.xattr_version; | |
3639 | ::encode(capsnap.xattrs, m->xattrbl); | |
3640 | ||
3641 | m->ctime = capsnap.ctime; | |
3642 | m->btime = capsnap.btime; | |
3643 | m->mtime = capsnap.mtime; | |
3644 | m->atime = capsnap.atime; | |
3645 | m->time_warp_seq = capsnap.time_warp_seq; | |
3646 | m->change_attr = capsnap.change_attr; | |
3647 | ||
3648 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
3649 | m->inline_version = in->inline_version; | |
3650 | m->inline_data = in->inline_data; | |
3651 | } | |
3652 | ||
3653 | assert(!session->flushing_caps_tids.empty()); | |
3654 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3655 | ||
3656 | session->con->send_message(m); | |
3657 | } | |
3658 | } | |
3659 | ||
3660 | ||
3661 | ||
3662 | void Client::wait_on_list(list<Cond*>& ls) | |
3663 | { | |
3664 | Cond cond; | |
3665 | ls.push_back(&cond); | |
3666 | cond.Wait(client_lock); | |
3667 | ls.remove(&cond); | |
3668 | } | |
3669 | ||
3670 | void Client::signal_cond_list(list<Cond*>& ls) | |
3671 | { | |
3672 | for (list<Cond*>::iterator it = ls.begin(); it != ls.end(); ++it) | |
3673 | (*it)->Signal(); | |
3674 | } | |
3675 | ||
3676 | void Client::wait_on_context_list(list<Context*>& ls) | |
3677 | { | |
3678 | Cond cond; | |
3679 | bool done = false; | |
3680 | int r; | |
3681 | ls.push_back(new C_Cond(&cond, &done, &r)); | |
3682 | while (!done) | |
3683 | cond.Wait(client_lock); | |
3684 | } | |
3685 | ||
3686 | void Client::signal_context_list(list<Context*>& ls) | |
3687 | { | |
3688 | while (!ls.empty()) { | |
3689 | ls.front()->complete(0); | |
3690 | ls.pop_front(); | |
3691 | } | |
3692 | } | |
3693 | ||
3694 | void Client::wake_inode_waiters(MetaSession *s) | |
3695 | { | |
3696 | xlist<Cap*>::iterator iter = s->caps.begin(); | |
3697 | while (!iter.end()){ | |
3698 | signal_cond_list((*iter)->inode->waitfor_caps); | |
3699 | ++iter; | |
3700 | } | |
3701 | } | |
3702 | ||
3703 | ||
3704 | // flush dirty data (from objectcache) | |
3705 | ||
3706 | class C_Client_CacheInvalidate : public Context { | |
3707 | private: | |
3708 | Client *client; | |
3709 | vinodeno_t ino; | |
3710 | int64_t offset, length; | |
3711 | public: | |
3712 | C_Client_CacheInvalidate(Client *c, Inode *in, int64_t off, int64_t len) : | |
3713 | client(c), offset(off), length(len) { | |
3714 | if (client->use_faked_inos()) | |
3715 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
3716 | else | |
3717 | ino = in->vino(); | |
3718 | } | |
3719 | void finish(int r) override { | |
3720 | // _async_invalidate takes the lock when it needs to, call this back from outside of lock. | |
3721 | assert(!client->client_lock.is_locked_by_me()); | |
3722 | client->_async_invalidate(ino, offset, length); | |
3723 | } | |
3724 | }; | |
3725 | ||
3726 | void Client::_async_invalidate(vinodeno_t ino, int64_t off, int64_t len) | |
3727 | { | |
3728 | if (unmounting) | |
3729 | return; | |
3730 | ldout(cct, 10) << "_async_invalidate " << ino << " " << off << "~" << len << dendl; | |
3731 | ino_invalidate_cb(callback_handle, ino, off, len); | |
3732 | } | |
3733 | ||
3734 | void Client::_schedule_invalidate_callback(Inode *in, int64_t off, int64_t len) { | |
3735 | ||
3736 | if (ino_invalidate_cb) | |
3737 | // we queue the invalidate, which calls the callback and decrements the ref | |
3738 | async_ino_invalidator.queue(new C_Client_CacheInvalidate(this, in, off, len)); | |
3739 | } | |
3740 | ||
3741 | void Client::_invalidate_inode_cache(Inode *in) | |
3742 | { | |
3743 | ldout(cct, 10) << "_invalidate_inode_cache " << *in << dendl; | |
3744 | ||
3745 | // invalidate our userspace inode cache | |
3746 | if (cct->_conf->client_oc) | |
3747 | objectcacher->release_set(&in->oset); | |
3748 | ||
3749 | _schedule_invalidate_callback(in, 0, 0); | |
3750 | } | |
3751 | ||
3752 | void Client::_invalidate_inode_cache(Inode *in, int64_t off, int64_t len) | |
3753 | { | |
3754 | ldout(cct, 10) << "_invalidate_inode_cache " << *in << " " << off << "~" << len << dendl; | |
3755 | ||
3756 | // invalidate our userspace inode cache | |
3757 | if (cct->_conf->client_oc) { | |
3758 | vector<ObjectExtent> ls; | |
3759 | Striper::file_to_extents(cct, in->ino, &in->layout, off, len, in->truncate_size, ls); | |
3760 | objectcacher->discard_set(&in->oset, ls); | |
3761 | } | |
3762 | ||
3763 | _schedule_invalidate_callback(in, off, len); | |
3764 | } | |
3765 | ||
3766 | bool Client::_release(Inode *in) | |
3767 | { | |
3768 | ldout(cct, 20) << "_release " << *in << dendl; | |
3769 | if (in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
3770 | _invalidate_inode_cache(in); | |
3771 | return true; | |
3772 | } | |
3773 | return false; | |
3774 | } | |
3775 | ||
3776 | bool Client::_flush(Inode *in, Context *onfinish) | |
3777 | { | |
3778 | ldout(cct, 10) << "_flush " << *in << dendl; | |
3779 | ||
3780 | if (!in->oset.dirty_or_tx) { | |
3781 | ldout(cct, 10) << " nothing to flush" << dendl; | |
3782 | onfinish->complete(0); | |
3783 | return true; | |
3784 | } | |
3785 | ||
3786 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
3787 | ldout(cct, 1) << __func__ << ": FULL, purging for ENOSPC" << dendl; | |
3788 | objectcacher->purge_set(&in->oset); | |
3789 | if (onfinish) { | |
3790 | onfinish->complete(-ENOSPC); | |
3791 | } | |
3792 | return true; | |
3793 | } | |
3794 | ||
3795 | return objectcacher->flush_set(&in->oset, onfinish); | |
3796 | } | |
3797 | ||
3798 | void Client::_flush_range(Inode *in, int64_t offset, uint64_t size) | |
3799 | { | |
3800 | assert(client_lock.is_locked()); | |
3801 | if (!in->oset.dirty_or_tx) { | |
3802 | ldout(cct, 10) << " nothing to flush" << dendl; | |
3803 | return; | |
3804 | } | |
3805 | ||
3806 | Mutex flock("Client::_flush_range flock"); | |
3807 | Cond cond; | |
3808 | bool safe = false; | |
3809 | Context *onflush = new C_SafeCond(&flock, &cond, &safe); | |
3810 | bool ret = objectcacher->file_flush(&in->oset, &in->layout, in->snaprealm->get_snap_context(), | |
3811 | offset, size, onflush); | |
3812 | if (!ret) { | |
3813 | // wait for flush | |
3814 | client_lock.Unlock(); | |
3815 | flock.Lock(); | |
3816 | while (!safe) | |
3817 | cond.Wait(flock); | |
3818 | flock.Unlock(); | |
3819 | client_lock.Lock(); | |
3820 | } | |
3821 | } | |
3822 | ||
3823 | void Client::flush_set_callback(ObjectCacher::ObjectSet *oset) | |
3824 | { | |
3825 | // Mutex::Locker l(client_lock); | |
3826 | assert(client_lock.is_locked()); // will be called via dispatch() -> objecter -> ... | |
3827 | Inode *in = static_cast<Inode *>(oset->parent); | |
3828 | assert(in); | |
3829 | _flushed(in); | |
3830 | } | |
3831 | ||
3832 | void Client::_flushed(Inode *in) | |
3833 | { | |
3834 | ldout(cct, 10) << "_flushed " << *in << dendl; | |
3835 | ||
3836 | put_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
3837 | } | |
3838 | ||
3839 | ||
3840 | ||
3841 | // checks common to add_update_cap, handle_cap_grant | |
3842 | void Client::check_cap_issue(Inode *in, Cap *cap, unsigned issued) | |
3843 | { | |
3844 | unsigned had = in->caps_issued(); | |
3845 | ||
3846 | if ((issued & CEPH_CAP_FILE_CACHE) && | |
3847 | !(had & CEPH_CAP_FILE_CACHE)) | |
3848 | in->cache_gen++; | |
3849 | ||
3850 | if ((issued & CEPH_CAP_FILE_SHARED) && | |
3851 | !(had & CEPH_CAP_FILE_SHARED)) { | |
3852 | in->shared_gen++; | |
3853 | ||
3854 | if (in->is_dir()) | |
3855 | clear_dir_complete_and_ordered(in, true); | |
3856 | } | |
3857 | } | |
3858 | ||
3859 | void Client::add_update_cap(Inode *in, MetaSession *mds_session, uint64_t cap_id, | |
3860 | unsigned issued, unsigned seq, unsigned mseq, inodeno_t realm, | |
3861 | int flags, const UserPerm& cap_perms) | |
3862 | { | |
3863 | Cap *cap = 0; | |
3864 | mds_rank_t mds = mds_session->mds_num; | |
3865 | if (in->caps.count(mds)) { | |
3866 | cap = in->caps[mds]; | |
3867 | ||
3868 | /* | |
3869 | * auth mds of the inode changed. we received the cap export | |
3870 | * message, but still haven't received the cap import message. | |
3871 | * handle_cap_export() updated the new auth MDS' cap. | |
3872 | * | |
3873 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
3874 | * a message that was send before the cap import message. So | |
3875 | * don't remove caps. | |
3876 | */ | |
3877 | if (ceph_seq_cmp(seq, cap->seq) <= 0) { | |
3878 | assert(cap == in->auth_cap); | |
3879 | assert(cap->cap_id == cap_id); | |
3880 | seq = cap->seq; | |
3881 | mseq = cap->mseq; | |
3882 | issued |= cap->issued; | |
3883 | flags |= CEPH_CAP_FLAG_AUTH; | |
3884 | } | |
3885 | } else { | |
3886 | mds_session->num_caps++; | |
3887 | if (!in->is_any_caps()) { | |
3888 | assert(in->snaprealm == 0); | |
3889 | in->snaprealm = get_snap_realm(realm); | |
3890 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
3891 | ldout(cct, 15) << "add_update_cap first one, opened snaprealm " << in->snaprealm << dendl; | |
3892 | } | |
3893 | in->caps[mds] = cap = new Cap; | |
3894 | ||
3895 | mds_session->caps.push_back(&cap->cap_item); | |
3896 | cap->session = mds_session; | |
3897 | cap->inode = in; | |
3898 | cap->gen = mds_session->cap_gen; | |
3899 | cap_list.push_back(&in->cap_item); | |
3900 | } | |
3901 | ||
3902 | check_cap_issue(in, cap, issued); | |
3903 | ||
3904 | if (flags & CEPH_CAP_FLAG_AUTH) { | |
3905 | if (in->auth_cap != cap && | |
3906 | (!in->auth_cap || ceph_seq_cmp(in->auth_cap->mseq, mseq) < 0)) { | |
3907 | if (in->auth_cap && in->flushing_cap_item.is_on_list()) { | |
3908 | ldout(cct, 10) << "add_update_cap changing auth cap: " | |
3909 | << "add myself to new auth MDS' flushing caps list" << dendl; | |
3910 | adjust_session_flushing_caps(in, in->auth_cap->session, mds_session); | |
3911 | } | |
3912 | in->auth_cap = cap; | |
3913 | } | |
3914 | } | |
3915 | ||
3916 | unsigned old_caps = cap->issued; | |
3917 | cap->cap_id = cap_id; | |
3918 | cap->issued |= issued; | |
3919 | cap->implemented |= issued; | |
3920 | cap->seq = seq; | |
3921 | cap->issue_seq = seq; | |
3922 | cap->mseq = mseq; | |
3923 | cap->latest_perms = cap_perms; | |
3924 | ldout(cct, 10) << "add_update_cap issued " << ccap_string(old_caps) << " -> " << ccap_string(cap->issued) | |
3925 | << " from mds." << mds | |
3926 | << " on " << *in | |
3927 | << dendl; | |
3928 | ||
3929 | if ((issued & ~old_caps) && in->auth_cap == cap) { | |
3930 | // non-auth MDS is revoking the newly grant caps ? | |
3931 | for (map<mds_rank_t,Cap*>::iterator it = in->caps.begin(); it != in->caps.end(); ++it) { | |
3932 | if (it->second == cap) | |
3933 | continue; | |
3934 | if (it->second->implemented & ~it->second->issued & issued) { | |
3935 | check_caps(in, CHECK_CAPS_NODELAY); | |
3936 | break; | |
3937 | } | |
3938 | } | |
3939 | } | |
3940 | ||
3941 | if (issued & ~old_caps) | |
3942 | signal_cond_list(in->waitfor_caps); | |
3943 | } | |
3944 | ||
3945 | void Client::remove_cap(Cap *cap, bool queue_release) | |
3946 | { | |
3947 | Inode *in = cap->inode; | |
3948 | MetaSession *session = cap->session; | |
3949 | mds_rank_t mds = cap->session->mds_num; | |
3950 | ||
3951 | ldout(cct, 10) << "remove_cap mds." << mds << " on " << *in << dendl; | |
3952 | ||
3953 | if (queue_release) { | |
3954 | session->enqueue_cap_release( | |
3955 | in->ino, | |
3956 | cap->cap_id, | |
3957 | cap->issue_seq, | |
3958 | cap->mseq, | |
3959 | cap_epoch_barrier); | |
3960 | } | |
3961 | ||
3962 | if (in->auth_cap == cap) { | |
3963 | if (in->flushing_cap_item.is_on_list()) { | |
3964 | ldout(cct, 10) << " removing myself from flushing_cap list" << dendl; | |
3965 | in->flushing_cap_item.remove_myself(); | |
3966 | } | |
3967 | in->auth_cap = NULL; | |
3968 | } | |
3969 | assert(in->caps.count(mds)); | |
3970 | in->caps.erase(mds); | |
3971 | ||
3972 | cap->cap_item.remove_myself(); | |
3973 | delete cap; | |
3974 | cap = nullptr; | |
3975 | ||
3976 | if (!in->is_any_caps()) { | |
3977 | ldout(cct, 15) << "remove_cap last one, closing snaprealm " << in->snaprealm << dendl; | |
3978 | in->snaprealm_item.remove_myself(); | |
3979 | put_snap_realm(in->snaprealm); | |
3980 | in->snaprealm = 0; | |
3981 | } | |
3982 | } | |
3983 | ||
3984 | void Client::remove_all_caps(Inode *in) | |
3985 | { | |
3986 | while (!in->caps.empty()) | |
3987 | remove_cap(in->caps.begin()->second, true); | |
3988 | } | |
3989 | ||
3990 | void Client::remove_session_caps(MetaSession *s) | |
3991 | { | |
3992 | ldout(cct, 10) << "remove_session_caps mds." << s->mds_num << dendl; | |
3993 | ||
3994 | while (s->caps.size()) { | |
3995 | Cap *cap = *s->caps.begin(); | |
3996 | Inode *in = cap->inode; | |
3997 | bool dirty_caps = false, cap_snaps = false; | |
3998 | if (in->auth_cap == cap) { | |
3999 | cap_snaps = !in->cap_snaps.empty(); | |
4000 | dirty_caps = in->dirty_caps | in->flushing_caps; | |
4001 | in->wanted_max_size = 0; | |
4002 | in->requested_max_size = 0; | |
4003 | in->flags |= I_CAP_DROPPED; | |
4004 | } | |
4005 | remove_cap(cap, false); | |
4006 | signal_cond_list(in->waitfor_caps); | |
4007 | if (cap_snaps) { | |
4008 | InodeRef tmp_ref(in); | |
4009 | in->cap_snaps.clear(); | |
4010 | } | |
4011 | if (dirty_caps) { | |
4012 | lderr(cct) << "remove_session_caps still has dirty|flushing caps on " << *in << dendl; | |
4013 | if (in->flushing_caps) { | |
4014 | num_flushing_caps--; | |
4015 | in->flushing_cap_tids.clear(); | |
4016 | } | |
4017 | in->flushing_caps = 0; | |
4018 | in->dirty_caps = 0; | |
4019 | put_inode(in); | |
4020 | } | |
4021 | } | |
4022 | s->flushing_caps_tids.clear(); | |
4023 | sync_cond.Signal(); | |
4024 | } | |
4025 | ||
b32b8144 FG |
4026 | int Client::_do_remount(void) |
4027 | { | |
4028 | errno = 0; | |
4029 | int r = remount_cb(callback_handle); | |
4030 | if (r != 0) { | |
4031 | int e = errno; | |
4032 | client_t whoami = get_nodeid(); | |
4033 | if (r == -1) { | |
4034 | lderr(cct) << | |
4035 | "failed to remount (to trim kernel dentries): " | |
4036 | "errno = " << e << " (" << strerror(e) << ")" << dendl; | |
4037 | } else { | |
4038 | lderr(cct) << | |
4039 | "failed to remount (to trim kernel dentries): " | |
4040 | "return code = " << r << dendl; | |
4041 | } | |
4042 | bool should_abort = cct->_conf->get_val<bool>("client_die_on_failed_remount") || | |
4043 | cct->_conf->get_val<bool>("client_die_on_failed_dentry_invalidate"); | |
4044 | if (should_abort && !unmounting) { | |
4045 | lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl; | |
4046 | ceph_abort(); | |
4047 | } | |
4048 | } | |
4049 | return r; | |
4050 | } | |
4051 | ||
7c673cae FG |
4052 | class C_Client_Remount : public Context { |
4053 | private: | |
4054 | Client *client; | |
4055 | public: | |
4056 | explicit C_Client_Remount(Client *c) : client(c) {} | |
4057 | void finish(int r) override { | |
b32b8144 FG |
4058 | assert(r == 0); |
4059 | client->_do_remount(); | |
7c673cae FG |
4060 | } |
4061 | }; | |
4062 | ||
4063 | void Client::_invalidate_kernel_dcache() | |
4064 | { | |
4065 | if (unmounting) | |
4066 | return; | |
4067 | if (can_invalidate_dentries && dentry_invalidate_cb && root->dir) { | |
4068 | for (ceph::unordered_map<string, Dentry*>::iterator p = root->dir->dentries.begin(); | |
4069 | p != root->dir->dentries.end(); | |
4070 | ++p) { | |
4071 | if (p->second->inode) | |
4072 | _schedule_invalidate_dentry_callback(p->second, false); | |
4073 | } | |
4074 | } else if (remount_cb) { | |
4075 | // Hacky: | |
4076 | // when remounting a file system, linux kernel trims all unused dentries in the fs | |
4077 | remount_finisher.queue(new C_Client_Remount(this)); | |
4078 | } | |
4079 | } | |
4080 | ||
4081 | void Client::trim_caps(MetaSession *s, int max) | |
4082 | { | |
4083 | mds_rank_t mds = s->mds_num; | |
4084 | int caps_size = s->caps.size(); | |
4085 | ldout(cct, 10) << "trim_caps mds." << mds << " max " << max | |
4086 | << " caps " << caps_size << dendl; | |
4087 | ||
4088 | int trimmed = 0; | |
4089 | xlist<Cap*>::iterator p = s->caps.begin(); | |
b32b8144 | 4090 | std::set<InodeRef> anchor; /* prevent put_inode from deleting all caps during traversal */ |
7c673cae FG |
4091 | while ((caps_size - trimmed) > max && !p.end()) { |
4092 | Cap *cap = *p; | |
b32b8144 | 4093 | InodeRef in(cap->inode); |
7c673cae FG |
4094 | |
4095 | // Increment p early because it will be invalidated if cap | |
4096 | // is deleted inside remove_cap | |
4097 | ++p; | |
4098 | ||
4099 | if (in->caps.size() > 1 && cap != in->auth_cap) { | |
4100 | int mine = cap->issued | cap->implemented; | |
4101 | int oissued = in->auth_cap ? in->auth_cap->issued : 0; | |
4102 | // disposable non-auth cap | |
b32b8144 | 4103 | if (!(get_caps_used(in.get()) & ~oissued & mine)) { |
7c673cae FG |
4104 | ldout(cct, 20) << " removing unused, unneeded non-auth cap on " << *in << dendl; |
4105 | remove_cap(cap, true); | |
b32b8144 | 4106 | /* N.B. no need to push onto anchor, as we are only removing one cap */ |
7c673cae FG |
4107 | trimmed++; |
4108 | } | |
4109 | } else { | |
4110 | ldout(cct, 20) << " trying to trim dentries for " << *in << dendl; | |
4111 | bool all = true; | |
4112 | set<Dentry*>::iterator q = in->dn_set.begin(); | |
7c673cae FG |
4113 | while (q != in->dn_set.end()) { |
4114 | Dentry *dn = *q++; | |
4115 | if (dn->lru_is_expireable()) { | |
4116 | if (can_invalidate_dentries && | |
4117 | dn->dir->parent_inode->ino == MDS_INO_ROOT) { | |
4118 | // Only issue one of these per DN for inodes in root: handle | |
4119 | // others more efficiently by calling for root-child DNs at | |
4120 | // the end of this function. | |
4121 | _schedule_invalidate_dentry_callback(dn, true); | |
4122 | } | |
b32b8144 FG |
4123 | ldout(cct, 20) << " anchoring inode: " << in->ino << dendl; |
4124 | anchor.insert(in); | |
7c673cae FG |
4125 | trim_dentry(dn); |
4126 | } else { | |
4127 | ldout(cct, 20) << " not expirable: " << dn->name << dendl; | |
4128 | all = false; | |
4129 | } | |
4130 | } | |
4131 | if (all && in->ino != MDS_INO_ROOT) { | |
4132 | ldout(cct, 20) << __func__ << " counting as trimmed: " << *in << dendl; | |
4133 | trimmed++; | |
4134 | } | |
4135 | } | |
4136 | } | |
b32b8144 FG |
4137 | ldout(cct, 20) << " clearing anchored inodes" << dendl; |
4138 | anchor.clear(); | |
7c673cae | 4139 | |
b32b8144 FG |
4140 | caps_size = s->caps.size(); |
4141 | if (caps_size > max) | |
7c673cae FG |
4142 | _invalidate_kernel_dcache(); |
4143 | } | |
4144 | ||
4145 | void Client::force_session_readonly(MetaSession *s) | |
4146 | { | |
4147 | s->readonly = true; | |
4148 | for (xlist<Cap*>::iterator p = s->caps.begin(); !p.end(); ++p) { | |
4149 | Inode *in = (*p)->inode; | |
4150 | if (in->caps_wanted() & CEPH_CAP_FILE_WR) | |
4151 | signal_cond_list(in->waitfor_caps); | |
4152 | } | |
4153 | } | |
4154 | ||
4155 | void Client::mark_caps_dirty(Inode *in, int caps) | |
4156 | { | |
4157 | ldout(cct, 10) << "mark_caps_dirty " << *in << " " << ccap_string(in->dirty_caps) << " -> " | |
4158 | << ccap_string(in->dirty_caps | caps) << dendl; | |
4159 | if (caps && !in->caps_dirty()) | |
4160 | in->get(); | |
4161 | in->dirty_caps |= caps; | |
4162 | } | |
4163 | ||
4164 | int Client::mark_caps_flushing(Inode *in, ceph_tid_t* ptid) | |
4165 | { | |
4166 | MetaSession *session = in->auth_cap->session; | |
4167 | ||
4168 | int flushing = in->dirty_caps; | |
4169 | assert(flushing); | |
4170 | ||
4171 | ceph_tid_t flush_tid = ++last_flush_tid; | |
4172 | in->flushing_cap_tids[flush_tid] = flushing; | |
4173 | ||
4174 | if (!in->flushing_caps) { | |
4175 | ldout(cct, 10) << "mark_caps_flushing " << ccap_string(flushing) << " " << *in << dendl; | |
4176 | num_flushing_caps++; | |
4177 | } else { | |
4178 | ldout(cct, 10) << "mark_caps_flushing (more) " << ccap_string(flushing) << " " << *in << dendl; | |
4179 | } | |
4180 | ||
4181 | in->flushing_caps |= flushing; | |
4182 | in->dirty_caps = 0; | |
4183 | ||
4184 | if (!in->flushing_cap_item.is_on_list()) | |
4185 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
4186 | session->flushing_caps_tids.insert(flush_tid); | |
4187 | ||
4188 | *ptid = flush_tid; | |
4189 | return flushing; | |
4190 | } | |
4191 | ||
4192 | void Client::adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s) | |
4193 | { | |
4194 | for (auto &p : in->cap_snaps) { | |
4195 | CapSnap &capsnap = p.second; | |
4196 | if (capsnap.flush_tid > 0) { | |
4197 | old_s->flushing_caps_tids.erase(capsnap.flush_tid); | |
4198 | new_s->flushing_caps_tids.insert(capsnap.flush_tid); | |
4199 | } | |
4200 | } | |
4201 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4202 | it != in->flushing_cap_tids.end(); | |
4203 | ++it) { | |
4204 | old_s->flushing_caps_tids.erase(it->first); | |
4205 | new_s->flushing_caps_tids.insert(it->first); | |
4206 | } | |
4207 | new_s->flushing_caps.push_back(&in->flushing_cap_item); | |
4208 | } | |
4209 | ||
4210 | /* | |
4211 | * Flush all caps back to the MDS. Because the callers generally wait on the | |
4212 | * result of this function (syncfs and umount cases), we set | |
4213 | * CHECK_CAPS_SYNCHRONOUS on the last check_caps call. | |
4214 | */ | |
4215 | void Client::flush_caps_sync() | |
4216 | { | |
4217 | ldout(cct, 10) << __func__ << dendl; | |
4218 | xlist<Inode*>::iterator p = delayed_caps.begin(); | |
4219 | while (!p.end()) { | |
4220 | unsigned flags = CHECK_CAPS_NODELAY; | |
4221 | Inode *in = *p; | |
4222 | ||
4223 | ++p; | |
4224 | delayed_caps.pop_front(); | |
4225 | if (p.end() && cap_list.empty()) | |
4226 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4227 | check_caps(in, flags); | |
4228 | } | |
4229 | ||
4230 | // other caps, too | |
4231 | p = cap_list.begin(); | |
4232 | while (!p.end()) { | |
4233 | unsigned flags = CHECK_CAPS_NODELAY; | |
4234 | Inode *in = *p; | |
4235 | ||
4236 | ++p; | |
4237 | if (p.end()) | |
4238 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4239 | check_caps(in, flags); | |
4240 | } | |
4241 | } | |
4242 | ||
4243 | void Client::flush_caps(Inode *in, MetaSession *session, bool sync) | |
4244 | { | |
4245 | ldout(cct, 10) << "flush_caps " << in << " mds." << session->mds_num << dendl; | |
4246 | Cap *cap = in->auth_cap; | |
4247 | assert(cap->session == session); | |
4248 | ||
4249 | for (map<ceph_tid_t,int>::iterator p = in->flushing_cap_tids.begin(); | |
4250 | p != in->flushing_cap_tids.end(); | |
4251 | ++p) { | |
4252 | bool req_sync = false; | |
4253 | ||
4254 | /* If this is a synchronous request, then flush the journal on last one */ | |
4255 | if (sync && (p->first == in->flushing_cap_tids.rbegin()->first)) | |
4256 | req_sync = true; | |
4257 | ||
4258 | send_cap(in, session, cap, req_sync, | |
4259 | (get_caps_used(in) | in->caps_dirty()), | |
4260 | in->caps_wanted(), (cap->issued | cap->implemented), | |
4261 | p->second, p->first); | |
4262 | } | |
4263 | } | |
4264 | ||
4265 | void Client::wait_sync_caps(Inode *in, ceph_tid_t want) | |
4266 | { | |
4267 | while (in->flushing_caps) { | |
4268 | map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4269 | assert(it != in->flushing_cap_tids.end()); | |
4270 | if (it->first > want) | |
4271 | break; | |
4272 | ldout(cct, 10) << "wait_sync_caps on " << *in << " flushing " | |
4273 | << ccap_string(it->second) << " want " << want | |
4274 | << " last " << it->first << dendl; | |
4275 | wait_on_list(in->waitfor_caps); | |
4276 | } | |
4277 | } | |
4278 | ||
4279 | void Client::wait_sync_caps(ceph_tid_t want) | |
4280 | { | |
4281 | retry: | |
4282 | ldout(cct, 10) << "wait_sync_caps want " << want << " (last is " << last_flush_tid << ", " | |
4283 | << num_flushing_caps << " total flushing)" << dendl; | |
4284 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
4285 | p != mds_sessions.end(); | |
4286 | ++p) { | |
4287 | MetaSession *s = p->second; | |
4288 | if (s->flushing_caps_tids.empty()) | |
4289 | continue; | |
4290 | ceph_tid_t oldest_tid = *s->flushing_caps_tids.begin(); | |
4291 | if (oldest_tid <= want) { | |
4292 | ldout(cct, 10) << " waiting on mds." << p->first << " tid " << oldest_tid | |
4293 | << " (want " << want << ")" << dendl; | |
4294 | sync_cond.Wait(client_lock); | |
4295 | goto retry; | |
4296 | } | |
4297 | } | |
4298 | } | |
4299 | ||
4300 | void Client::kick_flushing_caps(MetaSession *session) | |
4301 | { | |
4302 | mds_rank_t mds = session->mds_num; | |
4303 | ldout(cct, 10) << "kick_flushing_caps mds." << mds << dendl; | |
4304 | ||
4305 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4306 | Inode *in = *p; | |
4307 | if (session->early_flushing_caps.count(in)) | |
4308 | continue; | |
4309 | ldout(cct, 20) << " reflushing caps on " << *in << " to mds." << mds << dendl; | |
4310 | if (in->cap_snaps.size()) | |
4311 | flush_snaps(in, true); | |
4312 | if (in->flushing_caps) | |
4313 | flush_caps(in, session); | |
4314 | } | |
4315 | ||
4316 | session->early_flushing_caps.clear(); | |
4317 | } | |
4318 | ||
4319 | void Client::early_kick_flushing_caps(MetaSession *session) | |
4320 | { | |
4321 | session->early_flushing_caps.clear(); | |
4322 | ||
4323 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4324 | Inode *in = *p; | |
4325 | assert(in->auth_cap); | |
4326 | ||
4327 | // if flushing caps were revoked, we re-send the cap flush in client reconnect | |
4328 | // stage. This guarantees that MDS processes the cap flush message before issuing | |
4329 | // the flushing caps to other client. | |
4330 | if ((in->flushing_caps & in->auth_cap->issued) == in->flushing_caps) | |
4331 | continue; | |
4332 | ||
4333 | ldout(cct, 20) << " reflushing caps (early_kick) on " << *in | |
4334 | << " to mds." << session->mds_num << dendl; | |
4335 | ||
4336 | session->early_flushing_caps.insert(in); | |
4337 | ||
4338 | if (in->cap_snaps.size()) | |
4339 | flush_snaps(in, true); | |
4340 | if (in->flushing_caps) | |
4341 | flush_caps(in, session); | |
4342 | ||
4343 | } | |
4344 | } | |
4345 | ||
4346 | void Client::kick_maxsize_requests(MetaSession *session) | |
4347 | { | |
4348 | xlist<Cap*>::iterator iter = session->caps.begin(); | |
4349 | while (!iter.end()){ | |
4350 | (*iter)->inode->requested_max_size = 0; | |
4351 | (*iter)->inode->wanted_max_size = 0; | |
4352 | signal_cond_list((*iter)->inode->waitfor_caps); | |
4353 | ++iter; | |
4354 | } | |
4355 | } | |
4356 | ||
4357 | void SnapRealm::build_snap_context() | |
4358 | { | |
4359 | set<snapid_t> snaps; | |
4360 | snapid_t max_seq = seq; | |
4361 | ||
4362 | // start with prior_parents? | |
4363 | for (unsigned i=0; i<prior_parent_snaps.size(); i++) | |
4364 | snaps.insert(prior_parent_snaps[i]); | |
4365 | ||
4366 | // current parent's snaps | |
4367 | if (pparent) { | |
4368 | const SnapContext& psnapc = pparent->get_snap_context(); | |
4369 | for (unsigned i=0; i<psnapc.snaps.size(); i++) | |
4370 | if (psnapc.snaps[i] >= parent_since) | |
4371 | snaps.insert(psnapc.snaps[i]); | |
4372 | if (psnapc.seq > max_seq) | |
4373 | max_seq = psnapc.seq; | |
4374 | } | |
4375 | ||
4376 | // my snaps | |
4377 | for (unsigned i=0; i<my_snaps.size(); i++) | |
4378 | snaps.insert(my_snaps[i]); | |
4379 | ||
4380 | // ok! | |
4381 | cached_snap_context.seq = max_seq; | |
4382 | cached_snap_context.snaps.resize(0); | |
4383 | cached_snap_context.snaps.reserve(snaps.size()); | |
4384 | for (set<snapid_t>::reverse_iterator p = snaps.rbegin(); p != snaps.rend(); ++p) | |
4385 | cached_snap_context.snaps.push_back(*p); | |
4386 | } | |
4387 | ||
4388 | void Client::invalidate_snaprealm_and_children(SnapRealm *realm) | |
4389 | { | |
4390 | list<SnapRealm*> q; | |
4391 | q.push_back(realm); | |
4392 | ||
4393 | while (!q.empty()) { | |
4394 | realm = q.front(); | |
4395 | q.pop_front(); | |
4396 | ||
4397 | ldout(cct, 10) << "invalidate_snaprealm_and_children " << *realm << dendl; | |
4398 | realm->invalidate_cache(); | |
4399 | ||
4400 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4401 | p != realm->pchildren.end(); | |
4402 | ++p) | |
4403 | q.push_back(*p); | |
4404 | } | |
4405 | } | |
4406 | ||
4407 | SnapRealm *Client::get_snap_realm(inodeno_t r) | |
4408 | { | |
4409 | SnapRealm *realm = snap_realms[r]; | |
4410 | if (!realm) | |
4411 | snap_realms[r] = realm = new SnapRealm(r); | |
4412 | ldout(cct, 20) << "get_snap_realm " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; | |
4413 | realm->nref++; | |
4414 | return realm; | |
4415 | } | |
4416 | ||
4417 | SnapRealm *Client::get_snap_realm_maybe(inodeno_t r) | |
4418 | { | |
4419 | if (snap_realms.count(r) == 0) { | |
4420 | ldout(cct, 20) << "get_snap_realm_maybe " << r << " fail" << dendl; | |
4421 | return NULL; | |
4422 | } | |
4423 | SnapRealm *realm = snap_realms[r]; | |
4424 | ldout(cct, 20) << "get_snap_realm_maybe " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; | |
4425 | realm->nref++; | |
4426 | return realm; | |
4427 | } | |
4428 | ||
4429 | void Client::put_snap_realm(SnapRealm *realm) | |
4430 | { | |
4431 | ldout(cct, 20) << "put_snap_realm " << realm->ino << " " << realm | |
4432 | << " " << realm->nref << " -> " << (realm->nref - 1) << dendl; | |
4433 | if (--realm->nref == 0) { | |
4434 | snap_realms.erase(realm->ino); | |
4435 | if (realm->pparent) { | |
4436 | realm->pparent->pchildren.erase(realm); | |
4437 | put_snap_realm(realm->pparent); | |
4438 | } | |
4439 | delete realm; | |
4440 | } | |
4441 | } | |
4442 | ||
4443 | bool Client::adjust_realm_parent(SnapRealm *realm, inodeno_t parent) | |
4444 | { | |
4445 | if (realm->parent != parent) { | |
4446 | ldout(cct, 10) << "adjust_realm_parent " << *realm | |
4447 | << " " << realm->parent << " -> " << parent << dendl; | |
4448 | realm->parent = parent; | |
4449 | if (realm->pparent) { | |
4450 | realm->pparent->pchildren.erase(realm); | |
4451 | put_snap_realm(realm->pparent); | |
4452 | } | |
4453 | realm->pparent = get_snap_realm(parent); | |
4454 | realm->pparent->pchildren.insert(realm); | |
4455 | return true; | |
4456 | } | |
4457 | return false; | |
4458 | } | |
4459 | ||
4460 | static bool has_new_snaps(const SnapContext& old_snapc, | |
4461 | const SnapContext& new_snapc) | |
4462 | { | |
4463 | return !new_snapc.snaps.empty() && new_snapc.snaps[0] > old_snapc.seq; | |
4464 | } | |
4465 | ||
4466 | ||
4467 | void Client::update_snap_trace(bufferlist& bl, SnapRealm **realm_ret, bool flush) | |
4468 | { | |
4469 | SnapRealm *first_realm = NULL; | |
4470 | ldout(cct, 10) << "update_snap_trace len " << bl.length() << dendl; | |
4471 | ||
4472 | map<SnapRealm*, SnapContext> dirty_realms; | |
4473 | ||
4474 | bufferlist::iterator p = bl.begin(); | |
4475 | while (!p.end()) { | |
4476 | SnapRealmInfo info; | |
4477 | ::decode(info, p); | |
4478 | SnapRealm *realm = get_snap_realm(info.ino()); | |
4479 | ||
4480 | bool invalidate = false; | |
4481 | ||
4482 | if (info.seq() > realm->seq) { | |
4483 | ldout(cct, 10) << "update_snap_trace " << *realm << " seq " << info.seq() << " > " << realm->seq | |
4484 | << dendl; | |
4485 | ||
4486 | if (flush) { | |
4487 | // writeback any dirty caps _before_ updating snap list (i.e. with old snap info) | |
4488 | // flush me + children | |
4489 | list<SnapRealm*> q; | |
4490 | q.push_back(realm); | |
4491 | while (!q.empty()) { | |
4492 | SnapRealm *realm = q.front(); | |
4493 | q.pop_front(); | |
4494 | ||
4495 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4496 | p != realm->pchildren.end(); | |
4497 | ++p) | |
4498 | q.push_back(*p); | |
4499 | ||
4500 | if (dirty_realms.count(realm) == 0) { | |
4501 | realm->nref++; | |
4502 | dirty_realms[realm] = realm->get_snap_context(); | |
4503 | } | |
4504 | } | |
4505 | } | |
4506 | ||
4507 | // update | |
4508 | realm->seq = info.seq(); | |
4509 | realm->created = info.created(); | |
4510 | realm->parent_since = info.parent_since(); | |
4511 | realm->prior_parent_snaps = info.prior_parent_snaps; | |
4512 | realm->my_snaps = info.my_snaps; | |
4513 | invalidate = true; | |
4514 | } | |
4515 | ||
4516 | // _always_ verify parent | |
4517 | if (adjust_realm_parent(realm, info.parent())) | |
4518 | invalidate = true; | |
4519 | ||
4520 | if (invalidate) { | |
4521 | invalidate_snaprealm_and_children(realm); | |
4522 | ldout(cct, 15) << "update_snap_trace " << *realm << " self|parent updated" << dendl; | |
4523 | ldout(cct, 15) << " snapc " << realm->get_snap_context() << dendl; | |
4524 | } else { | |
4525 | ldout(cct, 10) << "update_snap_trace " << *realm << " seq " << info.seq() | |
4526 | << " <= " << realm->seq << " and same parent, SKIPPING" << dendl; | |
4527 | } | |
4528 | ||
4529 | if (!first_realm) | |
4530 | first_realm = realm; | |
4531 | else | |
4532 | put_snap_realm(realm); | |
4533 | } | |
4534 | ||
4535 | for (map<SnapRealm*, SnapContext>::iterator q = dirty_realms.begin(); | |
4536 | q != dirty_realms.end(); | |
4537 | ++q) { | |
4538 | SnapRealm *realm = q->first; | |
4539 | // if there are new snaps ? | |
4540 | if (has_new_snaps(q->second, realm->get_snap_context())) { | |
4541 | ldout(cct, 10) << " flushing caps on " << *realm << dendl; | |
4542 | xlist<Inode*>::iterator r = realm->inodes_with_caps.begin(); | |
4543 | while (!r.end()) { | |
4544 | Inode *in = *r; | |
4545 | ++r; | |
4546 | queue_cap_snap(in, q->second); | |
4547 | } | |
4548 | } else { | |
4549 | ldout(cct, 10) << " no new snap on " << *realm << dendl; | |
4550 | } | |
4551 | put_snap_realm(realm); | |
4552 | } | |
4553 | ||
4554 | if (realm_ret) | |
4555 | *realm_ret = first_realm; | |
4556 | else | |
4557 | put_snap_realm(first_realm); | |
4558 | } | |
4559 | ||
4560 | void Client::handle_snap(MClientSnap *m) | |
4561 | { | |
4562 | ldout(cct, 10) << "handle_snap " << *m << dendl; | |
4563 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
4564 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
4565 | if (!session) { | |
4566 | m->put(); | |
4567 | return; | |
4568 | } | |
4569 | ||
4570 | got_mds_push(session); | |
4571 | ||
4572 | map<Inode*, SnapContext> to_move; | |
4573 | SnapRealm *realm = 0; | |
4574 | ||
4575 | if (m->head.op == CEPH_SNAP_OP_SPLIT) { | |
4576 | assert(m->head.split); | |
4577 | SnapRealmInfo info; | |
4578 | bufferlist::iterator p = m->bl.begin(); | |
4579 | ::decode(info, p); | |
4580 | assert(info.ino() == m->head.split); | |
4581 | ||
4582 | // flush, then move, ino's. | |
4583 | realm = get_snap_realm(info.ino()); | |
4584 | ldout(cct, 10) << " splitting off " << *realm << dendl; | |
4585 | for (vector<inodeno_t>::iterator p = m->split_inos.begin(); | |
4586 | p != m->split_inos.end(); | |
4587 | ++p) { | |
4588 | vinodeno_t vino(*p, CEPH_NOSNAP); | |
4589 | if (inode_map.count(vino)) { | |
4590 | Inode *in = inode_map[vino]; | |
4591 | if (!in->snaprealm || in->snaprealm == realm) | |
4592 | continue; | |
4593 | if (in->snaprealm->created > info.created()) { | |
4594 | ldout(cct, 10) << " NOT moving " << *in << " from _newer_ realm " | |
4595 | << *in->snaprealm << dendl; | |
4596 | continue; | |
4597 | } | |
4598 | ldout(cct, 10) << " moving " << *in << " from " << *in->snaprealm << dendl; | |
4599 | ||
4600 | ||
4601 | in->snaprealm_item.remove_myself(); | |
4602 | to_move[in] = in->snaprealm->get_snap_context(); | |
4603 | put_snap_realm(in->snaprealm); | |
4604 | } | |
4605 | } | |
4606 | ||
4607 | // move child snaprealms, too | |
4608 | for (vector<inodeno_t>::iterator p = m->split_realms.begin(); | |
4609 | p != m->split_realms.end(); | |
4610 | ++p) { | |
4611 | ldout(cct, 10) << "adjusting snaprealm " << *p << " parent" << dendl; | |
4612 | SnapRealm *child = get_snap_realm_maybe(*p); | |
4613 | if (!child) | |
4614 | continue; | |
4615 | adjust_realm_parent(child, realm->ino); | |
4616 | put_snap_realm(child); | |
4617 | } | |
4618 | } | |
4619 | ||
4620 | update_snap_trace(m->bl, NULL, m->head.op != CEPH_SNAP_OP_DESTROY); | |
4621 | ||
4622 | if (realm) { | |
4623 | for (auto p = to_move.begin(); p != to_move.end(); ++p) { | |
4624 | Inode *in = p->first; | |
4625 | in->snaprealm = realm; | |
4626 | realm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4627 | realm->nref++; | |
4628 | // queue for snap writeback | |
4629 | if (has_new_snaps(p->second, realm->get_snap_context())) | |
4630 | queue_cap_snap(in, p->second); | |
4631 | } | |
4632 | put_snap_realm(realm); | |
4633 | } | |
4634 | ||
4635 | m->put(); | |
4636 | } | |
4637 | ||
4638 | void Client::handle_quota(MClientQuota *m) | |
4639 | { | |
4640 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
4641 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
4642 | if (!session) { | |
4643 | m->put(); | |
4644 | return; | |
4645 | } | |
4646 | ||
4647 | got_mds_push(session); | |
4648 | ||
4649 | ldout(cct, 10) << "handle_quota " << *m << " from mds." << mds << dendl; | |
4650 | ||
4651 | vinodeno_t vino(m->ino, CEPH_NOSNAP); | |
4652 | if (inode_map.count(vino)) { | |
4653 | Inode *in = NULL; | |
4654 | in = inode_map[vino]; | |
4655 | ||
4656 | if (in) { | |
4657 | in->quota = m->quota; | |
4658 | in->rstat = m->rstat; | |
4659 | } | |
4660 | } | |
4661 | ||
4662 | m->put(); | |
4663 | } | |
4664 | ||
4665 | void Client::handle_caps(MClientCaps *m) | |
4666 | { | |
4667 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
4668 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
4669 | if (!session) { | |
4670 | m->put(); | |
4671 | return; | |
4672 | } | |
4673 | ||
4674 | if (m->osd_epoch_barrier && !objecter->have_map(m->osd_epoch_barrier)) { | |
4675 | // Pause RADOS operations until we see the required epoch | |
4676 | objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
4677 | } | |
4678 | ||
4679 | if (m->osd_epoch_barrier > cap_epoch_barrier) { | |
4680 | // Record the barrier so that we will transmit it to MDS when releasing | |
4681 | set_cap_epoch_barrier(m->osd_epoch_barrier); | |
4682 | } | |
4683 | ||
4684 | got_mds_push(session); | |
4685 | ||
4686 | m->clear_payload(); // for if/when we send back to MDS | |
4687 | ||
4688 | Inode *in = 0; | |
4689 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
4690 | if (inode_map.count(vino)) | |
4691 | in = inode_map[vino]; | |
4692 | if (!in) { | |
4693 | if (m->get_op() == CEPH_CAP_OP_IMPORT) { | |
4694 | ldout(cct, 5) << "handle_caps don't have vino " << vino << " on IMPORT, immediately releasing" << dendl; | |
4695 | session->enqueue_cap_release( | |
4696 | m->get_ino(), | |
4697 | m->get_cap_id(), | |
4698 | m->get_seq(), | |
4699 | m->get_mseq(), | |
4700 | cap_epoch_barrier); | |
4701 | } else { | |
4702 | ldout(cct, 5) << "handle_caps don't have vino " << vino << ", dropping" << dendl; | |
4703 | } | |
4704 | m->put(); | |
4705 | ||
4706 | // in case the mds is waiting on e.g. a revocation | |
4707 | flush_cap_releases(); | |
4708 | return; | |
4709 | } | |
4710 | ||
4711 | switch (m->get_op()) { | |
4712 | case CEPH_CAP_OP_EXPORT: | |
4713 | return handle_cap_export(session, in, m); | |
4714 | case CEPH_CAP_OP_FLUSHSNAP_ACK: | |
4715 | return handle_cap_flushsnap_ack(session, in, m); | |
4716 | case CEPH_CAP_OP_IMPORT: | |
4717 | handle_cap_import(session, in, m); | |
4718 | } | |
4719 | ||
4720 | if (in->caps.count(mds) == 0) { | |
4721 | ldout(cct, 5) << "handle_caps don't have " << *in << " cap on mds." << mds << dendl; | |
4722 | m->put(); | |
4723 | return; | |
4724 | } | |
4725 | ||
4726 | Cap *cap = in->caps[mds]; | |
4727 | ||
4728 | switch (m->get_op()) { | |
4729 | case CEPH_CAP_OP_TRUNC: return handle_cap_trunc(session, in, m); | |
4730 | case CEPH_CAP_OP_IMPORT: | |
4731 | case CEPH_CAP_OP_REVOKE: | |
4732 | case CEPH_CAP_OP_GRANT: return handle_cap_grant(session, in, cap, m); | |
4733 | case CEPH_CAP_OP_FLUSH_ACK: return handle_cap_flush_ack(session, in, cap, m); | |
4734 | default: | |
4735 | m->put(); | |
4736 | } | |
4737 | } | |
4738 | ||
4739 | void Client::handle_cap_import(MetaSession *session, Inode *in, MClientCaps *m) | |
4740 | { | |
4741 | mds_rank_t mds = session->mds_num; | |
4742 | ||
4743 | ldout(cct, 5) << "handle_cap_import ino " << m->get_ino() << " mseq " << m->get_mseq() | |
4744 | << " IMPORT from mds." << mds << dendl; | |
4745 | ||
4746 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
4747 | Cap *cap = NULL; | |
4748 | UserPerm cap_perms; | |
4749 | if (m->peer.cap_id && in->caps.count(peer_mds)) { | |
4750 | cap = in->caps[peer_mds]; | |
4751 | if (cap) { | |
4752 | cap_perms = cap->latest_perms; | |
4753 | } | |
4754 | } | |
4755 | ||
4756 | // add/update it | |
4757 | SnapRealm *realm = NULL; | |
4758 | update_snap_trace(m->snapbl, &realm); | |
4759 | ||
4760 | add_update_cap(in, session, m->get_cap_id(), | |
4761 | m->get_caps(), m->get_seq(), m->get_mseq(), m->get_realm(), | |
4762 | CEPH_CAP_FLAG_AUTH, cap_perms); | |
4763 | ||
4764 | if (cap && cap->cap_id == m->peer.cap_id) { | |
4765 | remove_cap(cap, (m->peer.flags & CEPH_CAP_FLAG_RELEASE)); | |
4766 | } | |
4767 | ||
4768 | if (realm) | |
4769 | put_snap_realm(realm); | |
4770 | ||
4771 | if (in->auth_cap && in->auth_cap->session->mds_num == mds) { | |
4772 | // reflush any/all caps (if we are now the auth_cap) | |
4773 | if (in->cap_snaps.size()) | |
4774 | flush_snaps(in, true); | |
4775 | if (in->flushing_caps) | |
4776 | flush_caps(in, session); | |
4777 | } | |
4778 | } | |
4779 | ||
4780 | void Client::handle_cap_export(MetaSession *session, Inode *in, MClientCaps *m) | |
4781 | { | |
4782 | mds_rank_t mds = session->mds_num; | |
4783 | ||
4784 | ldout(cct, 5) << "handle_cap_export ino " << m->get_ino() << " mseq " << m->get_mseq() | |
4785 | << " EXPORT from mds." << mds << dendl; | |
4786 | ||
4787 | Cap *cap = NULL; | |
4788 | if (in->caps.count(mds)) | |
4789 | cap = in->caps[mds]; | |
4790 | ||
4791 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
4792 | ||
4793 | if (cap && cap->cap_id == m->get_cap_id()) { | |
4794 | if (m->peer.cap_id) { | |
4795 | MetaSession *tsession = _get_or_open_mds_session(peer_mds); | |
4796 | if (in->caps.count(peer_mds)) { | |
4797 | Cap *tcap = in->caps[peer_mds]; | |
181888fb | 4798 | if (tcap->cap_id == m->peer.cap_id && |
7c673cae FG |
4799 | ceph_seq_cmp(tcap->seq, m->peer.seq) < 0) { |
4800 | tcap->cap_id = m->peer.cap_id; | |
4801 | tcap->seq = m->peer.seq - 1; | |
4802 | tcap->issue_seq = tcap->seq; | |
4803 | tcap->mseq = m->peer.mseq; | |
4804 | tcap->issued |= cap->issued; | |
4805 | tcap->implemented |= cap->issued; | |
4806 | if (cap == in->auth_cap) | |
4807 | in->auth_cap = tcap; | |
4808 | if (in->auth_cap == tcap && in->flushing_cap_item.is_on_list()) | |
4809 | adjust_session_flushing_caps(in, session, tsession); | |
4810 | } | |
4811 | } else { | |
4812 | add_update_cap(in, tsession, m->peer.cap_id, cap->issued, | |
4813 | m->peer.seq - 1, m->peer.mseq, (uint64_t)-1, | |
4814 | cap == in->auth_cap ? CEPH_CAP_FLAG_AUTH : 0, | |
4815 | cap->latest_perms); | |
4816 | } | |
4817 | } else { | |
4818 | if (cap == in->auth_cap) | |
4819 | in->flags |= I_CAP_DROPPED; | |
4820 | } | |
4821 | ||
4822 | remove_cap(cap, false); | |
4823 | } | |
4824 | ||
4825 | m->put(); | |
4826 | } | |
4827 | ||
4828 | void Client::handle_cap_trunc(MetaSession *session, Inode *in, MClientCaps *m) | |
4829 | { | |
4830 | mds_rank_t mds = session->mds_num; | |
4831 | assert(in->caps[mds]); | |
4832 | ||
4833 | ldout(cct, 10) << "handle_cap_trunc on ino " << *in | |
4834 | << " size " << in->size << " -> " << m->get_size() | |
4835 | << dendl; | |
4836 | ||
4837 | int implemented = 0; | |
4838 | int issued = in->caps_issued(&implemented) | in->caps_dirty(); | |
4839 | issued |= implemented; | |
4840 | update_inode_file_bits(in, m->get_truncate_seq(), m->get_truncate_size(), | |
4841 | m->get_size(), m->get_change_attr(), m->get_time_warp_seq(), | |
4842 | m->get_ctime(), m->get_mtime(), m->get_atime(), | |
4843 | m->inline_version, m->inline_data, issued); | |
4844 | m->put(); | |
4845 | } | |
4846 | ||
4847 | void Client::handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, MClientCaps *m) | |
4848 | { | |
4849 | ceph_tid_t flush_ack_tid = m->get_client_tid(); | |
4850 | int dirty = m->get_dirty(); | |
4851 | int cleaned = 0; | |
4852 | int flushed = 0; | |
4853 | ||
4854 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4855 | it != in->flushing_cap_tids.end(); ) { | |
4856 | if (it->first == flush_ack_tid) | |
4857 | cleaned = it->second; | |
4858 | if (it->first <= flush_ack_tid) { | |
4859 | session->flushing_caps_tids.erase(it->first); | |
4860 | in->flushing_cap_tids.erase(it++); | |
4861 | ++flushed; | |
4862 | continue; | |
4863 | } | |
4864 | cleaned &= ~it->second; | |
4865 | if (!cleaned) | |
4866 | break; | |
4867 | ++it; | |
4868 | } | |
4869 | ||
4870 | ldout(cct, 5) << "handle_cap_flush_ack mds." << session->mds_num | |
4871 | << " cleaned " << ccap_string(cleaned) << " on " << *in | |
4872 | << " with " << ccap_string(dirty) << dendl; | |
4873 | ||
4874 | if (flushed) { | |
4875 | signal_cond_list(in->waitfor_caps); | |
4876 | if (session->flushing_caps_tids.empty() || | |
4877 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
4878 | sync_cond.Signal(); | |
4879 | } | |
4880 | ||
4881 | if (!dirty) { | |
4882 | in->cap_dirtier_uid = -1; | |
4883 | in->cap_dirtier_gid = -1; | |
4884 | } | |
4885 | ||
4886 | if (!cleaned) { | |
4887 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != any cap bit tids" << dendl; | |
4888 | } else { | |
4889 | if (in->flushing_caps) { | |
4890 | ldout(cct, 5) << " flushing_caps " << ccap_string(in->flushing_caps) | |
4891 | << " -> " << ccap_string(in->flushing_caps & ~cleaned) << dendl; | |
4892 | in->flushing_caps &= ~cleaned; | |
4893 | if (in->flushing_caps == 0) { | |
4894 | ldout(cct, 10) << " " << *in << " !flushing" << dendl; | |
4895 | num_flushing_caps--; | |
4896 | if (in->cap_snaps.empty()) | |
4897 | in->flushing_cap_item.remove_myself(); | |
4898 | } | |
4899 | if (!in->caps_dirty()) | |
4900 | put_inode(in); | |
4901 | } | |
4902 | } | |
4903 | ||
4904 | m->put(); | |
4905 | } | |
4906 | ||
4907 | ||
4908 | void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, MClientCaps *m) | |
4909 | { | |
4910 | mds_rank_t mds = session->mds_num; | |
4911 | assert(in->caps[mds]); | |
4912 | snapid_t follows = m->get_snap_follows(); | |
4913 | ||
4914 | if (in->cap_snaps.count(follows)) { | |
4915 | CapSnap &capsnap = in->cap_snaps.at(follows); | |
4916 | if (m->get_client_tid() != capsnap.flush_tid) { | |
4917 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != " << capsnap.flush_tid << dendl; | |
4918 | } else { | |
4919 | ldout(cct, 5) << "handle_cap_flushedsnap mds." << mds << " flushed snap follows " << follows | |
4920 | << " on " << *in << dendl; | |
4921 | InodeRef tmp_ref; | |
4922 | if (in->get_num_ref() == 1) | |
4923 | tmp_ref = in; // make sure inode not get freed while erasing item from in->cap_snaps | |
4924 | if (in->flushing_caps == 0 && in->cap_snaps.empty()) | |
4925 | in->flushing_cap_item.remove_myself(); | |
4926 | session->flushing_caps_tids.erase(capsnap.flush_tid); | |
4927 | in->cap_snaps.erase(follows); | |
4928 | } | |
4929 | } else { | |
4930 | ldout(cct, 5) << "handle_cap_flushedsnap DUP(?) mds." << mds << " flushed snap follows " << follows | |
4931 | << " on " << *in << dendl; | |
4932 | // we may not have it if we send multiple FLUSHSNAP requests and (got multiple FLUSHEDSNAPs back) | |
4933 | } | |
4934 | ||
4935 | m->put(); | |
4936 | } | |
4937 | ||
4938 | class C_Client_DentryInvalidate : public Context { | |
4939 | private: | |
4940 | Client *client; | |
4941 | vinodeno_t dirino; | |
4942 | vinodeno_t ino; | |
4943 | string name; | |
4944 | public: | |
4945 | C_Client_DentryInvalidate(Client *c, Dentry *dn, bool del) : | |
4946 | client(c), name(dn->name) { | |
4947 | if (client->use_faked_inos()) { | |
4948 | dirino.ino = dn->dir->parent_inode->faked_ino; | |
4949 | if (del) | |
4950 | ino.ino = dn->inode->faked_ino; | |
4951 | } else { | |
4952 | dirino = dn->dir->parent_inode->vino(); | |
4953 | if (del) | |
4954 | ino = dn->inode->vino(); | |
4955 | } | |
4956 | if (!del) | |
4957 | ino.ino = inodeno_t(); | |
4958 | } | |
4959 | void finish(int r) override { | |
4960 | // _async_dentry_invalidate is responsible for its own locking | |
4961 | assert(!client->client_lock.is_locked_by_me()); | |
4962 | client->_async_dentry_invalidate(dirino, ino, name); | |
4963 | } | |
4964 | }; | |
4965 | ||
4966 | void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name) | |
4967 | { | |
4968 | if (unmounting) | |
4969 | return; | |
4970 | ldout(cct, 10) << "_async_dentry_invalidate '" << name << "' ino " << ino | |
4971 | << " in dir " << dirino << dendl; | |
4972 | dentry_invalidate_cb(callback_handle, dirino, ino, name); | |
4973 | } | |
4974 | ||
4975 | void Client::_schedule_invalidate_dentry_callback(Dentry *dn, bool del) | |
4976 | { | |
4977 | if (dentry_invalidate_cb && dn->inode->ll_ref > 0) | |
4978 | async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn, del)); | |
4979 | } | |
4980 | ||
4981 | void Client::_try_to_trim_inode(Inode *in, bool sched_inval) | |
4982 | { | |
4983 | int ref = in->get_num_ref(); | |
4984 | ||
4985 | if (in->dir && !in->dir->dentries.empty()) { | |
4986 | for (auto p = in->dir->dentries.begin(); | |
4987 | p != in->dir->dentries.end(); ) { | |
4988 | Dentry *dn = p->second; | |
4989 | ++p; | |
4990 | /* rmsnap removes whole subtree, need trim inodes recursively. | |
4991 | * we don't need to invalidate dentries recursively. because | |
4992 | * invalidating a directory dentry effectively invalidate | |
4993 | * whole subtree */ | |
4994 | if (in->snapid != CEPH_NOSNAP && dn->inode && dn->inode->is_dir()) | |
4995 | _try_to_trim_inode(dn->inode.get(), false); | |
4996 | ||
4997 | if (dn->lru_is_expireable()) | |
4998 | unlink(dn, true, false); // keep dir, drop dentry | |
4999 | } | |
5000 | if (in->dir->dentries.empty()) { | |
5001 | close_dir(in->dir); | |
5002 | --ref; | |
5003 | } | |
5004 | } | |
5005 | ||
5006 | if (ref > 0 && (in->flags & I_SNAPDIR_OPEN)) { | |
5007 | InodeRef snapdir = open_snapdir(in); | |
5008 | _try_to_trim_inode(snapdir.get(), false); | |
5009 | --ref; | |
5010 | } | |
5011 | ||
5012 | if (ref > 0 && in->ll_ref > 0 && sched_inval) { | |
5013 | set<Dentry*>::iterator q = in->dn_set.begin(); | |
5014 | while (q != in->dn_set.end()) { | |
5015 | Dentry *dn = *q++; | |
5016 | // FIXME: we play lots of unlink/link tricks when handling MDS replies, | |
5017 | // so in->dn_set doesn't always reflect the state of kernel's dcache. | |
5018 | _schedule_invalidate_dentry_callback(dn, true); | |
5019 | unlink(dn, true, true); | |
5020 | } | |
5021 | } | |
5022 | } | |
5023 | ||
5024 | void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, MClientCaps *m) | |
5025 | { | |
5026 | mds_rank_t mds = session->mds_num; | |
5027 | int used = get_caps_used(in); | |
5028 | int wanted = in->caps_wanted(); | |
5029 | ||
5030 | const int old_caps = cap->issued; | |
5031 | const int new_caps = m->get_caps(); | |
5032 | ldout(cct, 5) << "handle_cap_grant on in " << m->get_ino() | |
5033 | << " mds." << mds << " seq " << m->get_seq() | |
5034 | << " caps now " << ccap_string(new_caps) | |
5035 | << " was " << ccap_string(old_caps) << dendl; | |
5036 | cap->seq = m->get_seq(); | |
5037 | ||
5038 | in->layout = m->get_layout(); | |
5039 | ||
5040 | // update inode | |
5041 | int implemented = 0; | |
5042 | int issued = in->caps_issued(&implemented) | in->caps_dirty(); | |
5043 | issued |= implemented; | |
5044 | ||
5045 | if ((issued & CEPH_CAP_AUTH_EXCL) == 0) { | |
5046 | in->mode = m->head.mode; | |
5047 | in->uid = m->head.uid; | |
5048 | in->gid = m->head.gid; | |
5049 | in->btime = m->btime; | |
5050 | } | |
5051 | bool deleted_inode = false; | |
5052 | if ((issued & CEPH_CAP_LINK_EXCL) == 0) { | |
5053 | in->nlink = m->head.nlink; | |
5054 | if (in->nlink == 0 && | |
5055 | (new_caps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) | |
5056 | deleted_inode = true; | |
5057 | } | |
5058 | if ((issued & CEPH_CAP_XATTR_EXCL) == 0 && | |
5059 | m->xattrbl.length() && | |
5060 | m->head.xattr_version > in->xattr_version) { | |
5061 | bufferlist::iterator p = m->xattrbl.begin(); | |
5062 | ::decode(in->xattrs, p); | |
5063 | in->xattr_version = m->head.xattr_version; | |
5064 | } | |
5065 | update_inode_file_bits(in, m->get_truncate_seq(), m->get_truncate_size(), m->get_size(), | |
5066 | m->get_change_attr(), m->get_time_warp_seq(), m->get_ctime(), | |
5067 | m->get_mtime(), m->get_atime(), | |
5068 | m->inline_version, m->inline_data, issued); | |
5069 | ||
5070 | // max_size | |
5071 | if (cap == in->auth_cap && | |
5072 | m->get_max_size() != in->max_size) { | |
5073 | ldout(cct, 10) << "max_size " << in->max_size << " -> " << m->get_max_size() << dendl; | |
5074 | in->max_size = m->get_max_size(); | |
5075 | if (in->max_size > in->wanted_max_size) { | |
5076 | in->wanted_max_size = 0; | |
5077 | in->requested_max_size = 0; | |
5078 | } | |
5079 | } | |
5080 | ||
5081 | bool check = false; | |
5082 | if (m->get_op() == CEPH_CAP_OP_IMPORT && m->get_wanted() != wanted) | |
5083 | check = true; | |
5084 | ||
5085 | check_cap_issue(in, cap, new_caps); | |
5086 | ||
5087 | // update caps | |
b32b8144 FG |
5088 | int revoked = old_caps & ~new_caps; |
5089 | if (revoked) { | |
5090 | ldout(cct, 10) << " revocation of " << ccap_string(revoked) << dendl; | |
7c673cae FG |
5091 | cap->issued = new_caps; |
5092 | cap->implemented |= new_caps; | |
5093 | ||
b32b8144 FG |
5094 | // recall delegations if we're losing caps necessary for them |
5095 | if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_RD)) | |
5096 | in->recall_deleg(false); | |
5097 | else if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_WR)) | |
5098 | in->recall_deleg(true); | |
5099 | ||
7c673cae FG |
5100 | if (((used & ~new_caps) & CEPH_CAP_FILE_BUFFER) |
5101 | && !_flush(in, new C_Client_FlushComplete(this, in))) { | |
5102 | // waitin' for flush | |
5103 | } else if ((old_caps & ~new_caps) & CEPH_CAP_FILE_CACHE) { | |
5104 | if (_release(in)) | |
5105 | check = true; | |
5106 | } else { | |
5107 | cap->wanted = 0; // don't let check_caps skip sending a response to MDS | |
5108 | check = true; | |
5109 | } | |
5110 | ||
5111 | } else if (old_caps == new_caps) { | |
5112 | ldout(cct, 10) << " caps unchanged at " << ccap_string(old_caps) << dendl; | |
5113 | } else { | |
5114 | ldout(cct, 10) << " grant, new caps are " << ccap_string(new_caps & ~old_caps) << dendl; | |
5115 | cap->issued = new_caps; | |
5116 | cap->implemented |= new_caps; | |
5117 | ||
5118 | if (cap == in->auth_cap) { | |
5119 | // non-auth MDS is revoking the newly grant caps ? | |
5120 | for (map<mds_rank_t, Cap*>::iterator it = in->caps.begin(); it != in->caps.end(); ++it) { | |
5121 | if (it->second == cap) | |
5122 | continue; | |
5123 | if (it->second->implemented & ~it->second->issued & new_caps) { | |
5124 | check = true; | |
5125 | break; | |
5126 | } | |
5127 | } | |
5128 | } | |
5129 | } | |
5130 | ||
5131 | if (check) | |
5132 | check_caps(in, 0); | |
5133 | ||
5134 | // wake up waiters | |
5135 | if (new_caps) | |
5136 | signal_cond_list(in->waitfor_caps); | |
5137 | ||
5138 | // may drop inode's last ref | |
5139 | if (deleted_inode) | |
5140 | _try_to_trim_inode(in, true); | |
5141 | ||
5142 | m->put(); | |
5143 | } | |
5144 | ||
5145 | int Client::_getgrouplist(gid_t** sgids, uid_t uid, gid_t gid) | |
5146 | { | |
5147 | // cppcheck-suppress variableScope | |
5148 | int sgid_count; | |
5149 | gid_t *sgid_buf; | |
5150 | ||
5151 | if (getgroups_cb) { | |
5152 | sgid_count = getgroups_cb(callback_handle, &sgid_buf); | |
5153 | if (sgid_count > 0) { | |
5154 | *sgids = sgid_buf; | |
5155 | return sgid_count; | |
5156 | } | |
5157 | } | |
5158 | ||
5159 | #if HAVE_GETGROUPLIST | |
5160 | struct passwd *pw; | |
5161 | pw = getpwuid(uid); | |
5162 | if (pw == NULL) { | |
5163 | ldout(cct, 3) << "getting user entry failed" << dendl; | |
5164 | return -errno; | |
5165 | } | |
5166 | //use PAM to get the group list | |
5167 | // initial number of group entries, defaults to posix standard of 16 | |
5168 | // PAM implementations may provide more than 16 groups.... | |
5169 | sgid_count = 16; | |
5170 | sgid_buf = (gid_t*)malloc(sgid_count * sizeof(gid_t)); | |
5171 | if (sgid_buf == NULL) { | |
5172 | ldout(cct, 3) << "allocating group memory failed" << dendl; | |
5173 | return -ENOMEM; | |
5174 | } | |
5175 | ||
5176 | while (1) { | |
5177 | #if defined(__APPLE__) | |
5178 | if (getgrouplist(pw->pw_name, gid, (int*)sgid_buf, &sgid_count) == -1) { | |
5179 | #else | |
5180 | if (getgrouplist(pw->pw_name, gid, sgid_buf, &sgid_count) == -1) { | |
5181 | #endif | |
5182 | // we need to resize the group list and try again | |
5183 | void *_realloc = NULL; | |
5184 | if ((_realloc = realloc(sgid_buf, sgid_count * sizeof(gid_t))) == NULL) { | |
5185 | ldout(cct, 3) << "allocating group memory failed" << dendl; | |
5186 | free(sgid_buf); | |
5187 | return -ENOMEM; | |
5188 | } | |
5189 | sgid_buf = (gid_t*)_realloc; | |
5190 | continue; | |
5191 | } | |
5192 | // list was successfully retrieved | |
5193 | break; | |
5194 | } | |
5195 | *sgids = sgid_buf; | |
5196 | return sgid_count; | |
5197 | #else | |
5198 | return 0; | |
5199 | #endif | |
5200 | } | |
5201 | ||
5202 | int Client::inode_permission(Inode *in, const UserPerm& perms, unsigned want) | |
5203 | { | |
5204 | if (perms.uid() == 0) | |
5205 | return 0; | |
5206 | ||
5207 | if (perms.uid() != in->uid && (in->mode & S_IRWXG)) { | |
5208 | int ret = _posix_acl_permission(in, perms, want); | |
5209 | if (ret != -EAGAIN) | |
5210 | return ret; | |
5211 | } | |
5212 | ||
5213 | // check permissions before doing anything else | |
5214 | if (!in->check_mode(perms, want)) | |
5215 | return -EACCES; | |
5216 | return 0; | |
5217 | } | |
5218 | ||
5219 | int Client::xattr_permission(Inode *in, const char *name, unsigned want, | |
5220 | const UserPerm& perms) | |
5221 | { | |
5222 | int r = _getattr_for_perm(in, perms); | |
5223 | if (r < 0) | |
5224 | goto out; | |
5225 | ||
5226 | r = 0; | |
5227 | if (strncmp(name, "system.", 7) == 0) { | |
5228 | if ((want & MAY_WRITE) && (perms.uid() != 0 && perms.uid() != in->uid)) | |
5229 | r = -EPERM; | |
5230 | } else { | |
5231 | r = inode_permission(in, perms, want); | |
5232 | } | |
5233 | out: | |
5234 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5235 | return r; | |
5236 | } | |
5237 | ||
5238 | ostream& operator<<(ostream &out, const UserPerm& perm) { | |
5239 | out << "UserPerm(uid: " << perm.uid() << ", gid: " << perm.gid() << ")"; | |
5240 | return out; | |
5241 | } | |
5242 | ||
5243 | int Client::may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
5244 | const UserPerm& perms) | |
5245 | { | |
181888fb | 5246 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5247 | int r = _getattr_for_perm(in, perms); |
5248 | if (r < 0) | |
5249 | goto out; | |
5250 | ||
5251 | if (mask & CEPH_SETATTR_SIZE) { | |
5252 | r = inode_permission(in, perms, MAY_WRITE); | |
5253 | if (r < 0) | |
5254 | goto out; | |
5255 | } | |
5256 | ||
5257 | r = -EPERM; | |
5258 | if (mask & CEPH_SETATTR_UID) { | |
5259 | if (perms.uid() != 0 && (perms.uid() != in->uid || stx->stx_uid != in->uid)) | |
5260 | goto out; | |
5261 | } | |
5262 | if (mask & CEPH_SETATTR_GID) { | |
5263 | if (perms.uid() != 0 && (perms.uid() != in->uid || | |
5264 | (!perms.gid_in_groups(stx->stx_gid) && stx->stx_gid != in->gid))) | |
5265 | goto out; | |
5266 | } | |
5267 | ||
5268 | if (mask & CEPH_SETATTR_MODE) { | |
5269 | if (perms.uid() != 0 && perms.uid() != in->uid) | |
5270 | goto out; | |
5271 | ||
5272 | gid_t i_gid = (mask & CEPH_SETATTR_GID) ? stx->stx_gid : in->gid; | |
5273 | if (perms.uid() != 0 && !perms.gid_in_groups(i_gid)) | |
5274 | stx->stx_mode &= ~S_ISGID; | |
5275 | } | |
5276 | ||
5277 | if (mask & (CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME | | |
5278 | CEPH_SETATTR_MTIME | CEPH_SETATTR_ATIME)) { | |
5279 | if (perms.uid() != 0 && perms.uid() != in->uid) { | |
5280 | int check_mask = CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME; | |
5281 | if (!(mask & CEPH_SETATTR_MTIME_NOW)) | |
5282 | check_mask |= CEPH_SETATTR_MTIME; | |
5283 | if (!(mask & CEPH_SETATTR_ATIME_NOW)) | |
5284 | check_mask |= CEPH_SETATTR_ATIME; | |
5285 | if (check_mask & mask) { | |
5286 | goto out; | |
5287 | } else { | |
5288 | r = inode_permission(in, perms, MAY_WRITE); | |
5289 | if (r < 0) | |
5290 | goto out; | |
5291 | } | |
5292 | } | |
5293 | } | |
5294 | r = 0; | |
5295 | out: | |
5296 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5297 | return r; | |
5298 | } | |
5299 | ||
5300 | int Client::may_open(Inode *in, int flags, const UserPerm& perms) | |
5301 | { | |
181888fb | 5302 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5303 | unsigned want = 0; |
5304 | ||
5305 | if ((flags & O_ACCMODE) == O_WRONLY) | |
5306 | want = MAY_WRITE; | |
5307 | else if ((flags & O_ACCMODE) == O_RDWR) | |
5308 | want = MAY_READ | MAY_WRITE; | |
5309 | else if ((flags & O_ACCMODE) == O_RDONLY) | |
5310 | want = MAY_READ; | |
5311 | if (flags & O_TRUNC) | |
5312 | want |= MAY_WRITE; | |
5313 | ||
5314 | int r = 0; | |
5315 | switch (in->mode & S_IFMT) { | |
5316 | case S_IFLNK: | |
5317 | r = -ELOOP; | |
5318 | goto out; | |
5319 | case S_IFDIR: | |
5320 | if (want & MAY_WRITE) { | |
5321 | r = -EISDIR; | |
5322 | goto out; | |
5323 | } | |
5324 | break; | |
5325 | } | |
5326 | ||
5327 | r = _getattr_for_perm(in, perms); | |
5328 | if (r < 0) | |
5329 | goto out; | |
5330 | ||
5331 | r = inode_permission(in, perms, want); | |
5332 | out: | |
5333 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5334 | return r; | |
5335 | } | |
5336 | ||
5337 | int Client::may_lookup(Inode *dir, const UserPerm& perms) | |
5338 | { | |
181888fb | 5339 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5340 | int r = _getattr_for_perm(dir, perms); |
5341 | if (r < 0) | |
5342 | goto out; | |
5343 | ||
5344 | r = inode_permission(dir, perms, MAY_EXEC); | |
5345 | out: | |
5346 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5347 | return r; | |
5348 | } | |
5349 | ||
5350 | int Client::may_create(Inode *dir, const UserPerm& perms) | |
5351 | { | |
181888fb | 5352 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5353 | int r = _getattr_for_perm(dir, perms); |
5354 | if (r < 0) | |
5355 | goto out; | |
5356 | ||
5357 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5358 | out: | |
5359 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5360 | return r; | |
5361 | } | |
5362 | ||
5363 | int Client::may_delete(Inode *dir, const char *name, const UserPerm& perms) | |
5364 | { | |
181888fb | 5365 | ldout(cct, 20) << __func__ << " " << *dir << "; " << "; name " << name << "; " << perms << dendl; |
7c673cae FG |
5366 | int r = _getattr_for_perm(dir, perms); |
5367 | if (r < 0) | |
5368 | goto out; | |
5369 | ||
5370 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5371 | if (r < 0) | |
5372 | goto out; | |
5373 | ||
5374 | /* 'name == NULL' means rmsnap */ | |
5375 | if (perms.uid() != 0 && name && (dir->mode & S_ISVTX)) { | |
5376 | InodeRef otherin; | |
5377 | r = _lookup(dir, name, CEPH_CAP_AUTH_SHARED, &otherin, perms); | |
5378 | if (r < 0) | |
5379 | goto out; | |
5380 | if (dir->uid != perms.uid() && otherin->uid != perms.uid()) | |
5381 | r = -EPERM; | |
5382 | } | |
5383 | out: | |
5384 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5385 | return r; | |
5386 | } | |
5387 | ||
5388 | int Client::may_hardlink(Inode *in, const UserPerm& perms) | |
5389 | { | |
181888fb | 5390 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5391 | int r = _getattr_for_perm(in, perms); |
5392 | if (r < 0) | |
5393 | goto out; | |
5394 | ||
5395 | if (perms.uid() == 0 || perms.uid() == in->uid) { | |
5396 | r = 0; | |
5397 | goto out; | |
5398 | } | |
5399 | ||
5400 | r = -EPERM; | |
5401 | if (!S_ISREG(in->mode)) | |
5402 | goto out; | |
5403 | ||
5404 | if (in->mode & S_ISUID) | |
5405 | goto out; | |
5406 | ||
5407 | if ((in->mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
5408 | goto out; | |
5409 | ||
5410 | r = inode_permission(in, perms, MAY_READ | MAY_WRITE); | |
5411 | out: | |
5412 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5413 | return r; | |
5414 | } | |
5415 | ||
5416 | int Client::_getattr_for_perm(Inode *in, const UserPerm& perms) | |
5417 | { | |
5418 | int mask = CEPH_STAT_CAP_MODE; | |
5419 | bool force = false; | |
5420 | if (acl_type != NO_ACL) { | |
5421 | mask |= CEPH_STAT_CAP_XATTR; | |
5422 | force = in->xattr_version == 0; | |
5423 | } | |
5424 | return _getattr(in, mask, perms, force); | |
5425 | } | |
5426 | ||
5427 | vinodeno_t Client::_get_vino(Inode *in) | |
5428 | { | |
5429 | /* The caller must hold the client lock */ | |
5430 | return vinodeno_t(in->ino, in->snapid); | |
5431 | } | |
5432 | ||
5433 | inodeno_t Client::_get_inodeno(Inode *in) | |
5434 | { | |
5435 | /* The caller must hold the client lock */ | |
5436 | return in->ino; | |
5437 | } | |
5438 | ||
5439 | ||
5440 | /** | |
5441 | * Resolve an MDS spec to a list of MDS daemon GIDs. | |
5442 | * | |
5443 | * The spec is a string representing a GID, rank, filesystem:rank, or name/id. | |
5444 | * It may be '*' in which case it matches all GIDs. | |
5445 | * | |
5446 | * If no error is returned, the `targets` vector will be populated with at least | |
5447 | * one MDS. | |
5448 | */ | |
5449 | int Client::resolve_mds( | |
5450 | const std::string &mds_spec, | |
5451 | std::vector<mds_gid_t> *targets) | |
5452 | { | |
5453 | assert(fsmap); | |
5454 | assert(targets != nullptr); | |
5455 | ||
5456 | mds_role_t role; | |
5457 | std::stringstream ss; | |
5458 | int role_r = fsmap->parse_role(mds_spec, &role, ss); | |
5459 | if (role_r == 0) { | |
5460 | // We got a role, resolve it to a GID | |
5461 | ldout(cct, 10) << __func__ << ": resolved '" << mds_spec << "' to role '" | |
5462 | << role << "'" << dendl; | |
5463 | targets->push_back( | |
5464 | fsmap->get_filesystem(role.fscid)->mds_map.get_info(role.rank).global_id); | |
5465 | return 0; | |
5466 | } | |
5467 | ||
5468 | std::string strtol_err; | |
5469 | long long rank_or_gid = strict_strtoll(mds_spec.c_str(), 10, &strtol_err); | |
5470 | if (strtol_err.empty()) { | |
5471 | // It is a possible GID | |
5472 | const mds_gid_t mds_gid = mds_gid_t(rank_or_gid); | |
5473 | if (fsmap->gid_exists(mds_gid)) { | |
5474 | ldout(cct, 10) << __func__ << ": validated GID " << mds_gid << dendl; | |
5475 | targets->push_back(mds_gid); | |
5476 | } else { | |
5477 | lderr(cct) << __func__ << ": GID " << mds_gid << " not in MDS map" | |
5478 | << dendl; | |
5479 | return -ENOENT; | |
5480 | } | |
5481 | } else if (mds_spec == "*") { | |
5482 | // It is a wildcard: use all MDSs | |
5483 | const auto mds_info = fsmap->get_mds_info(); | |
5484 | ||
5485 | if (mds_info.empty()) { | |
5486 | lderr(cct) << __func__ << ": * passed but no MDS daemons found" << dendl; | |
5487 | return -ENOENT; | |
5488 | } | |
5489 | ||
5490 | for (const auto i : mds_info) { | |
5491 | targets->push_back(i.first); | |
5492 | } | |
5493 | } else { | |
5494 | // It did not parse as an integer, it is not a wildcard, it must be a name | |
5495 | const mds_gid_t mds_gid = fsmap->find_mds_gid_by_name(mds_spec); | |
5496 | if (mds_gid == 0) { | |
5497 | lderr(cct) << "MDS ID '" << mds_spec << "' not found" << dendl; | |
5498 | ||
5499 | lderr(cct) << "FSMap: " << *fsmap << dendl; | |
5500 | ||
5501 | return -ENOENT; | |
5502 | } else { | |
5503 | ldout(cct, 10) << __func__ << ": resolved ID '" << mds_spec | |
5504 | << "' to GID " << mds_gid << dendl; | |
5505 | targets->push_back(mds_gid); | |
5506 | } | |
5507 | } | |
5508 | ||
5509 | return 0; | |
5510 | } | |
5511 | ||
5512 | ||
5513 | /** | |
5514 | * Authenticate with mon and establish global ID | |
5515 | */ | |
5516 | int Client::authenticate() | |
5517 | { | |
5518 | assert(client_lock.is_locked_by_me()); | |
5519 | ||
5520 | if (monclient->is_authenticated()) { | |
5521 | return 0; | |
5522 | } | |
5523 | ||
5524 | client_lock.Unlock(); | |
5525 | int r = monclient->authenticate(cct->_conf->client_mount_timeout); | |
5526 | client_lock.Lock(); | |
5527 | if (r < 0) { | |
5528 | return r; | |
5529 | } | |
5530 | ||
5531 | whoami = monclient->get_global_id(); | |
5532 | messenger->set_myname(entity_name_t::CLIENT(whoami.v)); | |
5533 | ||
5534 | return 0; | |
5535 | } | |
5536 | ||
5537 | int Client::fetch_fsmap(bool user) | |
5538 | { | |
5539 | int r; | |
5540 | // Retrieve FSMap to enable looking up daemon addresses. We need FSMap | |
5541 | // rather than MDSMap because no one MDSMap contains all the daemons, and | |
5542 | // a `tell` can address any daemon. | |
5543 | version_t fsmap_latest; | |
5544 | do { | |
5545 | C_SaferCond cond; | |
5546 | monclient->get_version("fsmap", &fsmap_latest, NULL, &cond); | |
5547 | client_lock.Unlock(); | |
5548 | r = cond.wait(); | |
5549 | client_lock.Lock(); | |
5550 | } while (r == -EAGAIN); | |
5551 | ||
5552 | if (r < 0) { | |
5553 | lderr(cct) << "Failed to learn FSMap version: " << cpp_strerror(r) << dendl; | |
5554 | return r; | |
5555 | } | |
5556 | ||
5557 | ldout(cct, 10) << __func__ << " learned FSMap version " << fsmap_latest << dendl; | |
5558 | ||
5559 | if (user) { | |
5560 | if (!fsmap_user || fsmap_user->get_epoch() < fsmap_latest) { | |
5561 | monclient->sub_want("fsmap.user", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5562 | monclient->renew_subs(); | |
5563 | wait_on_list(waiting_for_fsmap); | |
5564 | } | |
5565 | assert(fsmap_user); | |
5566 | assert(fsmap_user->get_epoch() >= fsmap_latest); | |
5567 | } else { | |
5568 | if (!fsmap || fsmap->get_epoch() < fsmap_latest) { | |
5569 | monclient->sub_want("fsmap", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5570 | monclient->renew_subs(); | |
5571 | wait_on_list(waiting_for_fsmap); | |
5572 | } | |
5573 | assert(fsmap); | |
5574 | assert(fsmap->get_epoch() >= fsmap_latest); | |
5575 | } | |
5576 | ldout(cct, 10) << __func__ << " finished waiting for FSMap version " | |
5577 | << fsmap_latest << dendl; | |
5578 | return 0; | |
5579 | } | |
5580 | ||
5581 | /** | |
5582 | * | |
5583 | * @mds_spec one of ID, rank, GID, "*" | |
5584 | * | |
5585 | */ | |
5586 | int Client::mds_command( | |
5587 | const std::string &mds_spec, | |
5588 | const vector<string>& cmd, | |
5589 | const bufferlist& inbl, | |
5590 | bufferlist *outbl, | |
5591 | string *outs, | |
5592 | Context *onfinish) | |
5593 | { | |
5594 | Mutex::Locker lock(client_lock); | |
5595 | ||
181888fb FG |
5596 | if (!initialized) |
5597 | return -ENOTCONN; | |
7c673cae FG |
5598 | |
5599 | int r; | |
5600 | r = authenticate(); | |
5601 | if (r < 0) { | |
5602 | return r; | |
5603 | } | |
5604 | ||
5605 | r = fetch_fsmap(false); | |
5606 | if (r < 0) { | |
5607 | return r; | |
5608 | } | |
5609 | ||
5610 | // Look up MDS target(s) of the command | |
5611 | std::vector<mds_gid_t> targets; | |
5612 | r = resolve_mds(mds_spec, &targets); | |
5613 | if (r < 0) { | |
5614 | return r; | |
5615 | } | |
5616 | ||
5617 | // If daemons are laggy, we won't send them commands. If all | |
5618 | // are laggy then we fail. | |
5619 | std::vector<mds_gid_t> non_laggy; | |
5620 | for (const auto gid : targets) { | |
5621 | const auto info = fsmap->get_info_gid(gid); | |
5622 | if (!info.laggy()) { | |
5623 | non_laggy.push_back(gid); | |
5624 | } | |
5625 | } | |
5626 | if (non_laggy.size() == 0) { | |
5627 | *outs = "All targeted MDS daemons are laggy"; | |
5628 | return -ENOENT; | |
5629 | } | |
5630 | ||
5631 | if (metadata.empty()) { | |
5632 | // We are called on an unmounted client, so metadata | |
5633 | // won't be initialized yet. | |
5634 | populate_metadata(""); | |
5635 | } | |
5636 | ||
5637 | // Send commands to targets | |
5638 | C_GatherBuilder gather(cct, onfinish); | |
5639 | for (const auto target_gid : non_laggy) { | |
5640 | const auto info = fsmap->get_info_gid(target_gid); | |
5641 | ||
5642 | // Open a connection to the target MDS | |
5643 | entity_inst_t inst = info.get_inst(); | |
5644 | ConnectionRef conn = messenger->get_connection(inst); | |
5645 | ||
5646 | // Generate MDSCommandOp state | |
5647 | auto &op = command_table.start_command(); | |
5648 | ||
5649 | op.on_finish = gather.new_sub(); | |
5650 | op.cmd = cmd; | |
5651 | op.outbl = outbl; | |
5652 | op.outs = outs; | |
5653 | op.inbl = inbl; | |
5654 | op.mds_gid = target_gid; | |
5655 | op.con = conn; | |
5656 | ||
5657 | ldout(cct, 4) << __func__ << ": new command op to " << target_gid | |
5658 | << " tid=" << op.tid << cmd << dendl; | |
5659 | ||
5660 | // Construct and send MCommand | |
5661 | MCommand *m = op.get_message(monclient->get_fsid()); | |
5662 | conn->send_message(m); | |
5663 | } | |
5664 | gather.activate(); | |
5665 | ||
5666 | return 0; | |
5667 | } | |
5668 | ||
5669 | void Client::handle_command_reply(MCommandReply *m) | |
5670 | { | |
5671 | ceph_tid_t const tid = m->get_tid(); | |
5672 | ||
5673 | ldout(cct, 10) << __func__ << ": tid=" << m->get_tid() << dendl; | |
5674 | ||
5675 | if (!command_table.exists(tid)) { | |
5676 | ldout(cct, 1) << __func__ << ": unknown tid " << tid << ", dropping" << dendl; | |
5677 | m->put(); | |
5678 | return; | |
5679 | } | |
5680 | ||
5681 | auto &op = command_table.get_command(tid); | |
5682 | if (op.outbl) { | |
5683 | op.outbl->claim(m->get_data()); | |
5684 | } | |
5685 | if (op.outs) { | |
5686 | *op.outs = m->rs; | |
5687 | } | |
5688 | ||
5689 | if (op.on_finish) { | |
5690 | op.on_finish->complete(m->r); | |
5691 | } | |
5692 | ||
5693 | command_table.erase(tid); | |
5694 | ||
5695 | m->put(); | |
5696 | } | |
5697 | ||
5698 | // ------------------- | |
5699 | // MOUNT | |
5700 | ||
5701 | int Client::mount(const std::string &mount_root, const UserPerm& perms, | |
5702 | bool require_mds) | |
5703 | { | |
5704 | Mutex::Locker lock(client_lock); | |
5705 | ||
5706 | if (mounted) { | |
5707 | ldout(cct, 5) << "already mounted" << dendl; | |
5708 | return 0; | |
5709 | } | |
5710 | ||
b32b8144 FG |
5711 | unmounting = false; |
5712 | ||
7c673cae FG |
5713 | int r = authenticate(); |
5714 | if (r < 0) { | |
5715 | lderr(cct) << "authentication failed: " << cpp_strerror(r) << dendl; | |
5716 | return r; | |
5717 | } | |
5718 | ||
5719 | std::string want = "mdsmap"; | |
5720 | const auto &mds_ns = cct->_conf->client_mds_namespace; | |
5721 | if (!mds_ns.empty()) { | |
5722 | r = fetch_fsmap(true); | |
5723 | if (r < 0) | |
5724 | return r; | |
5725 | fs_cluster_id_t cid = fsmap_user->get_fs_cid(mds_ns); | |
5726 | if (cid == FS_CLUSTER_ID_NONE) | |
5727 | return -ENOENT; | |
5728 | ||
5729 | std::ostringstream oss; | |
5730 | oss << want << "." << cid; | |
5731 | want = oss.str(); | |
5732 | } | |
5733 | ldout(cct, 10) << "Subscribing to map '" << want << "'" << dendl; | |
5734 | ||
5735 | monclient->sub_want(want, 0, 0); | |
5736 | monclient->renew_subs(); | |
5737 | ||
5738 | tick(); // start tick | |
5739 | ||
5740 | if (require_mds) { | |
5741 | while (1) { | |
5742 | auto availability = mdsmap->is_cluster_available(); | |
5743 | if (availability == MDSMap::STUCK_UNAVAILABLE) { | |
5744 | // Error out | |
5745 | ldout(cct, 10) << "mds cluster unavailable: epoch=" << mdsmap->get_epoch() << dendl; | |
5746 | return CEPH_FUSE_NO_MDS_UP; | |
5747 | } else if (availability == MDSMap::AVAILABLE) { | |
5748 | // Continue to mount | |
5749 | break; | |
5750 | } else if (availability == MDSMap::TRANSIENT_UNAVAILABLE) { | |
5751 | // Else, wait. MDSMonitor will update the map to bring | |
5752 | // us to a conclusion eventually. | |
5753 | wait_on_list(waiting_for_mdsmap); | |
5754 | } else { | |
5755 | // Unexpected value! | |
5756 | ceph_abort(); | |
5757 | } | |
5758 | } | |
5759 | } | |
5760 | ||
5761 | populate_metadata(mount_root.empty() ? "/" : mount_root); | |
5762 | ||
5763 | filepath fp(CEPH_INO_ROOT); | |
5764 | if (!mount_root.empty()) { | |
5765 | fp = filepath(mount_root.c_str()); | |
5766 | } | |
5767 | while (true) { | |
5768 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
5769 | req->set_filepath(fp); | |
5770 | req->head.args.getattr.mask = CEPH_STAT_CAP_INODE_ALL; | |
5771 | int res = make_request(req, perms); | |
5772 | if (res < 0) { | |
5773 | if (res == -EACCES && root) { | |
5774 | ldout(cct, 1) << __func__ << " EACCES on parent of mount point; quotas may not work" << dendl; | |
5775 | break; | |
5776 | } | |
5777 | return res; | |
5778 | } | |
5779 | ||
5780 | if (fp.depth()) | |
5781 | fp.pop_dentry(); | |
5782 | else | |
5783 | break; | |
5784 | } | |
5785 | ||
5786 | assert(root); | |
5787 | _ll_get(root); | |
5788 | ||
5789 | mounted = true; | |
5790 | ||
5791 | // trace? | |
5792 | if (!cct->_conf->client_trace.empty()) { | |
5793 | traceout.open(cct->_conf->client_trace.c_str()); | |
5794 | if (traceout.is_open()) { | |
5795 | ldout(cct, 1) << "opened trace file '" << cct->_conf->client_trace << "'" << dendl; | |
5796 | } else { | |
5797 | ldout(cct, 1) << "FAILED to open trace file '" << cct->_conf->client_trace << "'" << dendl; | |
5798 | } | |
5799 | } | |
5800 | ||
5801 | /* | |
5802 | ldout(cct, 3) << "op: // client trace data structs" << dendl; | |
5803 | ldout(cct, 3) << "op: struct stat st;" << dendl; | |
5804 | ldout(cct, 3) << "op: struct utimbuf utim;" << dendl; | |
5805 | ldout(cct, 3) << "op: int readlinkbuf_len = 1000;" << dendl; | |
5806 | ldout(cct, 3) << "op: char readlinkbuf[readlinkbuf_len];" << dendl; | |
5807 | ldout(cct, 3) << "op: map<string, inode_t*> dir_contents;" << dendl; | |
5808 | ldout(cct, 3) << "op: map<int, int> open_files;" << dendl; | |
5809 | ldout(cct, 3) << "op: int fd;" << dendl; | |
5810 | */ | |
5811 | return 0; | |
5812 | } | |
5813 | ||
5814 | // UNMOUNT | |
5815 | ||
5816 | void Client::_close_sessions() | |
5817 | { | |
5818 | while (!mds_sessions.empty()) { | |
5819 | // send session closes! | |
5820 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
5821 | p != mds_sessions.end(); | |
5822 | ++p) { | |
5823 | if (p->second->state != MetaSession::STATE_CLOSING) { | |
5824 | _close_mds_session(p->second); | |
5825 | } | |
5826 | } | |
5827 | ||
5828 | // wait for sessions to close | |
5829 | ldout(cct, 2) << "waiting for " << mds_sessions.size() << " mds sessions to close" << dendl; | |
5830 | mount_cond.Wait(client_lock); | |
5831 | } | |
5832 | } | |
5833 | ||
31f18b77 FG |
5834 | void Client::flush_mdlog_sync() |
5835 | { | |
5836 | if (mds_requests.empty()) | |
5837 | return; | |
5838 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
5839 | p != mds_sessions.end(); | |
5840 | ++p) { | |
5841 | MetaSession *s = p->second; | |
5842 | flush_mdlog(s); | |
5843 | } | |
5844 | } | |
5845 | ||
5846 | void Client::flush_mdlog(MetaSession *session) | |
5847 | { | |
5848 | // Only send this to Luminous or newer MDS daemons, older daemons | |
5849 | // will crash if they see an unknown CEPH_SESSION_* value in this msg. | |
5850 | const uint64_t features = session->con->get_features(); | |
5851 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
5852 | MClientSession *m = new MClientSession(CEPH_SESSION_REQUEST_FLUSH_MDLOG); | |
5853 | session->con->send_message(m); | |
5854 | } | |
5855 | } | |
5856 | ||
5857 | ||
b32b8144 | 5858 | void Client::_unmount() |
7c673cae | 5859 | { |
181888fb FG |
5860 | if (unmounting) |
5861 | return; | |
7c673cae FG |
5862 | |
5863 | ldout(cct, 2) << "unmounting" << dendl; | |
5864 | unmounting = true; | |
5865 | ||
b32b8144 FG |
5866 | deleg_timeout = 0; |
5867 | ||
31f18b77 | 5868 | flush_mdlog_sync(); // flush the mdlog for pending requests, if any |
7c673cae FG |
5869 | while (!mds_requests.empty()) { |
5870 | ldout(cct, 10) << "waiting on " << mds_requests.size() << " requests" << dendl; | |
5871 | mount_cond.Wait(client_lock); | |
5872 | } | |
5873 | ||
5874 | if (tick_event) | |
5875 | timer.cancel_event(tick_event); | |
5876 | tick_event = 0; | |
5877 | ||
5878 | cwd.reset(); | |
5879 | ||
5880 | // clean up any unclosed files | |
5881 | while (!fd_map.empty()) { | |
5882 | Fh *fh = fd_map.begin()->second; | |
5883 | fd_map.erase(fd_map.begin()); | |
5884 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *fh->inode << dendl; | |
5885 | _release_fh(fh); | |
5886 | } | |
5887 | ||
5888 | while (!ll_unclosed_fh_set.empty()) { | |
5889 | set<Fh*>::iterator it = ll_unclosed_fh_set.begin(); | |
5890 | Fh *fh = *it; | |
5891 | ll_unclosed_fh_set.erase(fh); | |
5892 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *(fh->inode) << dendl; | |
5893 | _release_fh(fh); | |
5894 | } | |
5895 | ||
5896 | while (!opened_dirs.empty()) { | |
5897 | dir_result_t *dirp = *opened_dirs.begin(); | |
5898 | ldout(cct, 0) << " destroyed lost open dir " << dirp << " on " << *dirp->inode << dendl; | |
5899 | _closedir(dirp); | |
5900 | } | |
5901 | ||
5902 | _ll_drop_pins(); | |
5903 | ||
31f18b77 FG |
5904 | if (blacklisted) { |
5905 | ldout(cct, 0) << " skipping clean shutdown, we are blacklisted" << dendl; | |
5906 | ||
5907 | if (cct->_conf->client_oc) { | |
5908 | // Purge all cached data so that ObjectCacher doesn't get hung up | |
5909 | // trying to flush it. ObjectCacher's behaviour on EBLACKLISTED | |
5910 | // is to just leave things marked dirty | |
5911 | // (http://tracker.ceph.com/issues/9105) | |
5912 | for (const auto &i : inode_map) { | |
5913 | objectcacher->purge_set(&(i.second->oset)); | |
5914 | } | |
5915 | } | |
5916 | ||
5917 | mounted = false; | |
5918 | return; | |
5919 | } | |
5920 | ||
7c673cae FG |
5921 | while (unsafe_sync_write > 0) { |
5922 | ldout(cct, 0) << unsafe_sync_write << " unsafe_sync_writes, waiting" << dendl; | |
5923 | mount_cond.Wait(client_lock); | |
5924 | } | |
5925 | ||
5926 | if (cct->_conf->client_oc) { | |
5927 | // flush/release all buffered data | |
5928 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; | |
5929 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
5930 | p != inode_map.end(); | |
5931 | p = next) { | |
5932 | next = p; | |
5933 | ++next; | |
5934 | Inode *in = p->second; | |
5935 | if (!in) { | |
5936 | ldout(cct, 0) << "null inode_map entry ino " << p->first << dendl; | |
5937 | assert(in); | |
5938 | } | |
5939 | if (!in->caps.empty()) { | |
5940 | InodeRef tmp_ref(in); | |
5941 | _release(in); | |
5942 | _flush(in, new C_Client_FlushComplete(this, in)); | |
5943 | } | |
5944 | } | |
5945 | } | |
5946 | ||
5947 | flush_caps_sync(); | |
5948 | wait_sync_caps(last_flush_tid); | |
5949 | ||
5950 | // empty lru cache | |
7c673cae FG |
5951 | trim_cache(); |
5952 | ||
5953 | while (lru.lru_get_size() > 0 || | |
5954 | !inode_map.empty()) { | |
5955 | ldout(cct, 2) << "cache still has " << lru.lru_get_size() | |
5956 | << "+" << inode_map.size() << " items" | |
5957 | << ", waiting (for caps to release?)" | |
5958 | << dendl; | |
5959 | utime_t until = ceph_clock_now() + utime_t(5, 0); | |
5960 | int r = mount_cond.WaitUntil(client_lock, until); | |
5961 | if (r == ETIMEDOUT) { | |
5962 | dump_cache(NULL); | |
5963 | } | |
5964 | } | |
5965 | assert(lru.lru_get_size() == 0); | |
5966 | assert(inode_map.empty()); | |
5967 | ||
5968 | // stop tracing | |
5969 | if (!cct->_conf->client_trace.empty()) { | |
5970 | ldout(cct, 1) << "closing trace file '" << cct->_conf->client_trace << "'" << dendl; | |
5971 | traceout.close(); | |
5972 | } | |
5973 | ||
5974 | _close_sessions(); | |
5975 | ||
5976 | mounted = false; | |
5977 | ||
5978 | ldout(cct, 2) << "unmounted." << dendl; | |
5979 | } | |
5980 | ||
b32b8144 FG |
5981 | void Client::unmount() |
5982 | { | |
5983 | Mutex::Locker lock(client_lock); | |
5984 | _unmount(); | |
5985 | } | |
5986 | ||
7c673cae FG |
5987 | void Client::flush_cap_releases() |
5988 | { | |
5989 | // send any cap releases | |
5990 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
5991 | p != mds_sessions.end(); | |
5992 | ++p) { | |
5993 | if (p->second->release && mdsmap->is_clientreplay_or_active_or_stopping( | |
5994 | p->first)) { | |
5995 | if (cct->_conf->client_inject_release_failure) { | |
5996 | ldout(cct, 20) << __func__ << " injecting failure to send cap release message" << dendl; | |
5997 | p->second->release->put(); | |
5998 | } else { | |
5999 | p->second->con->send_message(p->second->release); | |
6000 | } | |
6001 | p->second->release = 0; | |
6002 | } | |
6003 | } | |
6004 | } | |
6005 | ||
6006 | void Client::tick() | |
6007 | { | |
6008 | if (cct->_conf->client_debug_inject_tick_delay > 0) { | |
6009 | sleep(cct->_conf->client_debug_inject_tick_delay); | |
6010 | assert(0 == cct->_conf->set_val("client_debug_inject_tick_delay", "0")); | |
6011 | cct->_conf->apply_changes(NULL); | |
6012 | } | |
6013 | ||
6014 | ldout(cct, 21) << "tick" << dendl; | |
3efd9988 FG |
6015 | tick_event = timer.add_event_after( |
6016 | cct->_conf->client_tick_interval, | |
6017 | new FunctionContext([this](int) { | |
6018 | // Called back via Timer, which takes client_lock for us | |
6019 | assert(client_lock.is_locked_by_me()); | |
6020 | tick(); | |
6021 | })); | |
7c673cae FG |
6022 | utime_t now = ceph_clock_now(); |
6023 | ||
6024 | if (!mounted && !mds_requests.empty()) { | |
6025 | MetaRequest *req = mds_requests.begin()->second; | |
6026 | if (req->op_stamp + cct->_conf->client_mount_timeout < now) { | |
6027 | req->abort(-ETIMEDOUT); | |
6028 | if (req->caller_cond) { | |
6029 | req->kick = true; | |
6030 | req->caller_cond->Signal(); | |
6031 | } | |
6032 | signal_cond_list(waiting_for_mdsmap); | |
6033 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
6034 | p != mds_sessions.end(); | |
6035 | ++p) | |
6036 | signal_context_list(p->second->waiting_for_open); | |
6037 | } | |
6038 | } | |
6039 | ||
6040 | if (mdsmap->get_epoch()) { | |
6041 | // renew caps? | |
6042 | utime_t el = now - last_cap_renew; | |
6043 | if (el > mdsmap->get_session_timeout() / 3.0) | |
6044 | renew_caps(); | |
6045 | ||
6046 | flush_cap_releases(); | |
6047 | } | |
6048 | ||
6049 | // delayed caps | |
6050 | xlist<Inode*>::iterator p = delayed_caps.begin(); | |
6051 | while (!p.end()) { | |
6052 | Inode *in = *p; | |
6053 | ++p; | |
6054 | if (in->hold_caps_until > now) | |
6055 | break; | |
6056 | delayed_caps.pop_front(); | |
6057 | cap_list.push_back(&in->cap_item); | |
6058 | check_caps(in, CHECK_CAPS_NODELAY); | |
6059 | } | |
6060 | ||
6061 | trim_cache(true); | |
6062 | } | |
6063 | ||
6064 | void Client::renew_caps() | |
6065 | { | |
6066 | ldout(cct, 10) << "renew_caps()" << dendl; | |
6067 | last_cap_renew = ceph_clock_now(); | |
6068 | ||
6069 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
6070 | p != mds_sessions.end(); | |
6071 | ++p) { | |
6072 | ldout(cct, 15) << "renew_caps requesting from mds." << p->first << dendl; | |
6073 | if (mdsmap->get_state(p->first) >= MDSMap::STATE_REJOIN) | |
6074 | renew_caps(p->second); | |
6075 | } | |
6076 | } | |
6077 | ||
6078 | void Client::renew_caps(MetaSession *session) | |
6079 | { | |
6080 | ldout(cct, 10) << "renew_caps mds." << session->mds_num << dendl; | |
6081 | session->last_cap_renew_request = ceph_clock_now(); | |
6082 | uint64_t seq = ++session->cap_renew_seq; | |
6083 | session->con->send_message(new MClientSession(CEPH_SESSION_REQUEST_RENEWCAPS, seq)); | |
6084 | } | |
6085 | ||
6086 | ||
6087 | // =============================================================== | |
6088 | // high level (POSIXy) interface | |
6089 | ||
6090 | int Client::_do_lookup(Inode *dir, const string& name, int mask, | |
6091 | InodeRef *target, const UserPerm& perms) | |
6092 | { | |
6093 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
6094 | MetaRequest *req = new MetaRequest(op); | |
6095 | filepath path; | |
6096 | dir->make_nosnap_relative_path(path); | |
6097 | path.push_dentry(name); | |
6098 | req->set_filepath(path); | |
6099 | req->set_inode(dir); | |
6100 | if (cct->_conf->client_debug_getattr_caps && op == CEPH_MDS_OP_LOOKUP) | |
6101 | mask |= DEBUG_GETATTR_CAPS; | |
6102 | req->head.args.getattr.mask = mask; | |
6103 | ||
6104 | ldout(cct, 10) << "_do_lookup on " << path << dendl; | |
6105 | ||
6106 | int r = make_request(req, perms, target); | |
6107 | ldout(cct, 10) << "_do_lookup res is " << r << dendl; | |
6108 | return r; | |
6109 | } | |
6110 | ||
6111 | int Client::_lookup(Inode *dir, const string& dname, int mask, InodeRef *target, | |
6112 | const UserPerm& perms) | |
6113 | { | |
6114 | int r = 0; | |
6115 | Dentry *dn = NULL; | |
6116 | ||
6117 | if (!dir->is_dir()) { | |
6118 | r = -ENOTDIR; | |
6119 | goto done; | |
6120 | } | |
6121 | ||
6122 | if (dname == "..") { | |
6123 | if (dir->dn_set.empty()) | |
6124 | *target = dir; | |
6125 | else | |
6126 | *target = dir->get_first_parent()->dir->parent_inode; //dirs can't be hard-linked | |
6127 | goto done; | |
6128 | } | |
6129 | ||
6130 | if (dname == ".") { | |
6131 | *target = dir; | |
6132 | goto done; | |
6133 | } | |
6134 | ||
6135 | if (dname.length() > NAME_MAX) { | |
6136 | r = -ENAMETOOLONG; | |
6137 | goto done; | |
6138 | } | |
6139 | ||
6140 | if (dname == cct->_conf->client_snapdir && | |
6141 | dir->snapid == CEPH_NOSNAP) { | |
6142 | *target = open_snapdir(dir); | |
6143 | goto done; | |
6144 | } | |
6145 | ||
6146 | if (dir->dir && | |
6147 | dir->dir->dentries.count(dname)) { | |
6148 | dn = dir->dir->dentries[dname]; | |
6149 | ||
6150 | ldout(cct, 20) << "_lookup have dn " << dname << " mds." << dn->lease_mds << " ttl " << dn->lease_ttl | |
6151 | << " seq " << dn->lease_seq | |
6152 | << dendl; | |
6153 | ||
6154 | if (!dn->inode || dn->inode->caps_issued_mask(mask)) { | |
6155 | // is dn lease valid? | |
6156 | utime_t now = ceph_clock_now(); | |
6157 | if (dn->lease_mds >= 0 && | |
6158 | dn->lease_ttl > now && | |
6159 | mds_sessions.count(dn->lease_mds)) { | |
6160 | MetaSession *s = mds_sessions[dn->lease_mds]; | |
6161 | if (s->cap_ttl > now && | |
6162 | s->cap_gen == dn->lease_gen) { | |
6163 | // touch this mds's dir cap too, even though we don't _explicitly_ use it here, to | |
6164 | // make trim_caps() behave. | |
6165 | dir->try_touch_cap(dn->lease_mds); | |
6166 | goto hit_dn; | |
6167 | } | |
6168 | ldout(cct, 20) << " bad lease, cap_ttl " << s->cap_ttl << ", cap_gen " << s->cap_gen | |
6169 | << " vs lease_gen " << dn->lease_gen << dendl; | |
6170 | } | |
6171 | // dir lease? | |
6172 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED)) { | |
6173 | if (dn->cap_shared_gen == dir->shared_gen && | |
6174 | (!dn->inode || dn->inode->caps_issued_mask(mask))) | |
6175 | goto hit_dn; | |
6176 | if (!dn->inode && (dir->flags & I_COMPLETE)) { | |
6177 | ldout(cct, 10) << "_lookup concluded ENOENT locally for " | |
6178 | << *dir << " dn '" << dname << "'" << dendl; | |
6179 | return -ENOENT; | |
6180 | } | |
6181 | } | |
6182 | } else { | |
6183 | ldout(cct, 20) << " no cap on " << dn->inode->vino() << dendl; | |
6184 | } | |
6185 | } else { | |
6186 | // can we conclude ENOENT locally? | |
6187 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED) && | |
6188 | (dir->flags & I_COMPLETE)) { | |
6189 | ldout(cct, 10) << "_lookup concluded ENOENT locally for " << *dir << " dn '" << dname << "'" << dendl; | |
6190 | return -ENOENT; | |
6191 | } | |
6192 | } | |
6193 | ||
6194 | r = _do_lookup(dir, dname, mask, target, perms); | |
6195 | goto done; | |
6196 | ||
6197 | hit_dn: | |
6198 | if (dn->inode) { | |
6199 | *target = dn->inode; | |
6200 | } else { | |
6201 | r = -ENOENT; | |
6202 | } | |
6203 | touch_dn(dn); | |
6204 | ||
6205 | done: | |
6206 | if (r < 0) | |
6207 | ldout(cct, 10) << "_lookup " << *dir << " " << dname << " = " << r << dendl; | |
6208 | else | |
6209 | ldout(cct, 10) << "_lookup " << *dir << " " << dname << " = " << **target << dendl; | |
6210 | return r; | |
6211 | } | |
6212 | ||
6213 | int Client::get_or_create(Inode *dir, const char* name, | |
6214 | Dentry **pdn, bool expect_null) | |
6215 | { | |
6216 | // lookup | |
6217 | ldout(cct, 20) << "get_or_create " << *dir << " name " << name << dendl; | |
6218 | dir->open_dir(); | |
6219 | if (dir->dir->dentries.count(name)) { | |
6220 | Dentry *dn = dir->dir->dentries[name]; | |
6221 | ||
6222 | // is dn lease valid? | |
6223 | utime_t now = ceph_clock_now(); | |
6224 | if (dn->inode && | |
6225 | dn->lease_mds >= 0 && | |
6226 | dn->lease_ttl > now && | |
6227 | mds_sessions.count(dn->lease_mds)) { | |
6228 | MetaSession *s = mds_sessions[dn->lease_mds]; | |
6229 | if (s->cap_ttl > now && | |
6230 | s->cap_gen == dn->lease_gen) { | |
6231 | if (expect_null) | |
6232 | return -EEXIST; | |
6233 | } | |
6234 | } | |
6235 | *pdn = dn; | |
6236 | } else { | |
6237 | // otherwise link up a new one | |
6238 | *pdn = link(dir->dir, name, NULL, NULL); | |
6239 | } | |
6240 | ||
6241 | // success | |
6242 | return 0; | |
6243 | } | |
6244 | ||
6245 | int Client::path_walk(const filepath& origpath, InodeRef *end, | |
6246 | const UserPerm& perms, bool followsym, int mask) | |
6247 | { | |
6248 | filepath path = origpath; | |
6249 | InodeRef cur; | |
6250 | if (origpath.absolute()) | |
6251 | cur = root; | |
6252 | else | |
6253 | cur = cwd; | |
6254 | assert(cur); | |
6255 | ||
6256 | ldout(cct, 10) << "path_walk " << path << dendl; | |
6257 | ||
6258 | int symlinks = 0; | |
6259 | ||
6260 | unsigned i=0; | |
6261 | while (i < path.depth() && cur) { | |
6262 | int caps = 0; | |
6263 | const string &dname = path[i]; | |
6264 | ldout(cct, 10) << " " << i << " " << *cur << " " << dname << dendl; | |
6265 | ldout(cct, 20) << " (path is " << path << ")" << dendl; | |
6266 | InodeRef next; | |
6267 | if (cct->_conf->client_permissions) { | |
6268 | int r = may_lookup(cur.get(), perms); | |
6269 | if (r < 0) | |
6270 | return r; | |
6271 | caps = CEPH_CAP_AUTH_SHARED; | |
6272 | } | |
6273 | ||
6274 | /* Get extra requested caps on the last component */ | |
6275 | if (i == (path.depth() - 1)) | |
6276 | caps |= mask; | |
6277 | int r = _lookup(cur.get(), dname, caps, &next, perms); | |
6278 | if (r < 0) | |
6279 | return r; | |
6280 | // only follow trailing symlink if followsym. always follow | |
6281 | // 'directory' symlinks. | |
6282 | if (next && next->is_symlink()) { | |
6283 | symlinks++; | |
6284 | ldout(cct, 20) << " symlink count " << symlinks << ", value is '" << next->symlink << "'" << dendl; | |
6285 | if (symlinks > MAXSYMLINKS) { | |
6286 | return -ELOOP; | |
6287 | } | |
6288 | ||
6289 | if (i < path.depth() - 1) { | |
6290 | // dir symlink | |
6291 | // replace consumed components of path with symlink dir target | |
6292 | filepath resolved(next->symlink.c_str()); | |
6293 | resolved.append(path.postfixpath(i + 1)); | |
6294 | path = resolved; | |
6295 | i = 0; | |
6296 | if (next->symlink[0] == '/') { | |
6297 | cur = root; | |
6298 | } | |
6299 | continue; | |
6300 | } else if (followsym) { | |
6301 | if (next->symlink[0] == '/') { | |
6302 | path = next->symlink.c_str(); | |
6303 | i = 0; | |
6304 | // reset position | |
6305 | cur = root; | |
6306 | } else { | |
6307 | filepath more(next->symlink.c_str()); | |
6308 | // we need to remove the symlink component from off of the path | |
6309 | // before adding the target that the symlink points to. remain | |
6310 | // at the same position in the path. | |
6311 | path.pop_dentry(); | |
6312 | path.append(more); | |
6313 | } | |
6314 | continue; | |
6315 | } | |
6316 | } | |
6317 | cur.swap(next); | |
6318 | i++; | |
6319 | } | |
6320 | if (!cur) | |
6321 | return -ENOENT; | |
6322 | if (end) | |
6323 | end->swap(cur); | |
6324 | return 0; | |
6325 | } | |
6326 | ||
6327 | ||
6328 | // namespace ops | |
6329 | ||
6330 | int Client::link(const char *relexisting, const char *relpath, const UserPerm& perm) | |
6331 | { | |
6332 | Mutex::Locker lock(client_lock); | |
6333 | tout(cct) << "link" << std::endl; | |
6334 | tout(cct) << relexisting << std::endl; | |
6335 | tout(cct) << relpath << std::endl; | |
6336 | ||
181888fb FG |
6337 | if (unmounting) |
6338 | return -ENOTCONN; | |
6339 | ||
7c673cae FG |
6340 | filepath existing(relexisting); |
6341 | ||
6342 | InodeRef in, dir; | |
6343 | int r = path_walk(existing, &in, perm, true); | |
6344 | if (r < 0) | |
6345 | return r; | |
6346 | if (std::string(relpath) == "/") { | |
6347 | r = -EEXIST; | |
6348 | return r; | |
6349 | } | |
6350 | filepath path(relpath); | |
6351 | string name = path.last_dentry(); | |
6352 | path.pop_dentry(); | |
6353 | ||
6354 | r = path_walk(path, &dir, perm, true); | |
6355 | if (r < 0) | |
6356 | return r; | |
6357 | if (cct->_conf->client_permissions) { | |
6358 | if (S_ISDIR(in->mode)) { | |
6359 | r = -EPERM; | |
6360 | return r; | |
6361 | } | |
6362 | r = may_hardlink(in.get(), perm); | |
6363 | if (r < 0) | |
6364 | return r; | |
6365 | r = may_create(dir.get(), perm); | |
6366 | if (r < 0) | |
6367 | return r; | |
6368 | } | |
6369 | r = _link(in.get(), dir.get(), name.c_str(), perm); | |
6370 | return r; | |
6371 | } | |
6372 | ||
6373 | int Client::unlink(const char *relpath, const UserPerm& perm) | |
6374 | { | |
6375 | Mutex::Locker lock(client_lock); | |
6376 | tout(cct) << "unlink" << std::endl; | |
6377 | tout(cct) << relpath << std::endl; | |
6378 | ||
181888fb FG |
6379 | if (unmounting) |
6380 | return -ENOTCONN; | |
6381 | ||
7c673cae FG |
6382 | if (std::string(relpath) == "/") |
6383 | return -EISDIR; | |
6384 | ||
6385 | filepath path(relpath); | |
6386 | string name = path.last_dentry(); | |
6387 | path.pop_dentry(); | |
6388 | InodeRef dir; | |
6389 | int r = path_walk(path, &dir, perm); | |
6390 | if (r < 0) | |
6391 | return r; | |
6392 | if (cct->_conf->client_permissions) { | |
6393 | r = may_delete(dir.get(), name.c_str(), perm); | |
6394 | if (r < 0) | |
6395 | return r; | |
6396 | } | |
6397 | return _unlink(dir.get(), name.c_str(), perm); | |
6398 | } | |
6399 | ||
6400 | int Client::rename(const char *relfrom, const char *relto, const UserPerm& perm) | |
6401 | { | |
6402 | Mutex::Locker lock(client_lock); | |
6403 | tout(cct) << "rename" << std::endl; | |
6404 | tout(cct) << relfrom << std::endl; | |
6405 | tout(cct) << relto << std::endl; | |
6406 | ||
181888fb FG |
6407 | if (unmounting) |
6408 | return -ENOTCONN; | |
6409 | ||
7c673cae FG |
6410 | if (std::string(relfrom) == "/" || std::string(relto) == "/") |
6411 | return -EBUSY; | |
6412 | ||
6413 | filepath from(relfrom); | |
6414 | filepath to(relto); | |
6415 | string fromname = from.last_dentry(); | |
6416 | from.pop_dentry(); | |
6417 | string toname = to.last_dentry(); | |
6418 | to.pop_dentry(); | |
6419 | ||
6420 | InodeRef fromdir, todir; | |
6421 | int r = path_walk(from, &fromdir, perm); | |
6422 | if (r < 0) | |
6423 | goto out; | |
6424 | r = path_walk(to, &todir, perm); | |
6425 | if (r < 0) | |
6426 | goto out; | |
6427 | ||
6428 | if (cct->_conf->client_permissions) { | |
6429 | int r = may_delete(fromdir.get(), fromname.c_str(), perm); | |
6430 | if (r < 0) | |
6431 | return r; | |
6432 | r = may_delete(todir.get(), toname.c_str(), perm); | |
6433 | if (r < 0 && r != -ENOENT) | |
6434 | return r; | |
6435 | } | |
6436 | r = _rename(fromdir.get(), fromname.c_str(), todir.get(), toname.c_str(), perm); | |
6437 | out: | |
6438 | return r; | |
6439 | } | |
6440 | ||
6441 | // dirs | |
6442 | ||
6443 | int Client::mkdir(const char *relpath, mode_t mode, const UserPerm& perm) | |
6444 | { | |
6445 | Mutex::Locker lock(client_lock); | |
6446 | tout(cct) << "mkdir" << std::endl; | |
6447 | tout(cct) << relpath << std::endl; | |
6448 | tout(cct) << mode << std::endl; | |
6449 | ldout(cct, 10) << "mkdir: " << relpath << dendl; | |
6450 | ||
181888fb FG |
6451 | if (unmounting) |
6452 | return -ENOTCONN; | |
6453 | ||
7c673cae FG |
6454 | if (std::string(relpath) == "/") |
6455 | return -EEXIST; | |
6456 | ||
6457 | filepath path(relpath); | |
6458 | string name = path.last_dentry(); | |
6459 | path.pop_dentry(); | |
6460 | InodeRef dir; | |
6461 | int r = path_walk(path, &dir, perm); | |
6462 | if (r < 0) | |
6463 | return r; | |
6464 | if (cct->_conf->client_permissions) { | |
6465 | r = may_create(dir.get(), perm); | |
6466 | if (r < 0) | |
6467 | return r; | |
6468 | } | |
6469 | return _mkdir(dir.get(), name.c_str(), mode, perm); | |
6470 | } | |
6471 | ||
6472 | int Client::mkdirs(const char *relpath, mode_t mode, const UserPerm& perms) | |
6473 | { | |
6474 | Mutex::Locker lock(client_lock); | |
6475 | ldout(cct, 10) << "Client::mkdirs " << relpath << dendl; | |
6476 | tout(cct) << "mkdirs" << std::endl; | |
6477 | tout(cct) << relpath << std::endl; | |
6478 | tout(cct) << mode << std::endl; | |
6479 | ||
181888fb FG |
6480 | if (unmounting) |
6481 | return -ENOTCONN; | |
6482 | ||
7c673cae FG |
6483 | //get through existing parts of path |
6484 | filepath path(relpath); | |
6485 | unsigned int i; | |
6486 | int r = 0, caps = 0; | |
6487 | InodeRef cur, next; | |
6488 | cur = cwd; | |
6489 | for (i=0; i<path.depth(); ++i) { | |
6490 | if (cct->_conf->client_permissions) { | |
6491 | r = may_lookup(cur.get(), perms); | |
6492 | if (r < 0) | |
6493 | break; | |
6494 | caps = CEPH_CAP_AUTH_SHARED; | |
6495 | } | |
6496 | r = _lookup(cur.get(), path[i].c_str(), caps, &next, perms); | |
6497 | if (r < 0) | |
6498 | break; | |
6499 | cur.swap(next); | |
6500 | } | |
6501 | //check that we have work left to do | |
6502 | if (i==path.depth()) return -EEXIST; | |
6503 | if (r!=-ENOENT) return r; | |
6504 | ldout(cct, 20) << "mkdirs got through " << i << " directories on path " << relpath << dendl; | |
6505 | //make new directory at each level | |
6506 | for (; i<path.depth(); ++i) { | |
6507 | if (cct->_conf->client_permissions) { | |
6508 | r = may_create(cur.get(), perms); | |
6509 | if (r < 0) | |
6510 | return r; | |
6511 | } | |
6512 | //make new dir | |
6513 | r = _mkdir(cur.get(), path[i].c_str(), mode, perms, &next); | |
c07f9fc5 | 6514 | |
7c673cae | 6515 | //check proper creation/existence |
c07f9fc5 FG |
6516 | if(-EEXIST == r && i < path.depth() - 1) { |
6517 | r = _lookup(cur.get(), path[i].c_str(), CEPH_CAP_AUTH_SHARED, &next, perms); | |
6518 | } | |
6519 | if (r < 0) | |
6520 | return r; | |
7c673cae FG |
6521 | //move to new dir and continue |
6522 | cur.swap(next); | |
6523 | ldout(cct, 20) << "mkdirs: successfully created directory " | |
6524 | << filepath(cur->ino).get_path() << dendl; | |
6525 | } | |
6526 | return 0; | |
6527 | } | |
6528 | ||
6529 | int Client::rmdir(const char *relpath, const UserPerm& perms) | |
6530 | { | |
6531 | Mutex::Locker lock(client_lock); | |
6532 | tout(cct) << "rmdir" << std::endl; | |
6533 | tout(cct) << relpath << std::endl; | |
6534 | ||
181888fb FG |
6535 | if (unmounting) |
6536 | return -ENOTCONN; | |
6537 | ||
7c673cae FG |
6538 | if (std::string(relpath) == "/") |
6539 | return -EBUSY; | |
6540 | ||
6541 | filepath path(relpath); | |
6542 | string name = path.last_dentry(); | |
6543 | path.pop_dentry(); | |
6544 | InodeRef dir; | |
6545 | int r = path_walk(path, &dir, perms); | |
6546 | if (r < 0) | |
6547 | return r; | |
6548 | if (cct->_conf->client_permissions) { | |
6549 | int r = may_delete(dir.get(), name.c_str(), perms); | |
6550 | if (r < 0) | |
6551 | return r; | |
6552 | } | |
6553 | return _rmdir(dir.get(), name.c_str(), perms); | |
6554 | } | |
6555 | ||
6556 | int Client::mknod(const char *relpath, mode_t mode, const UserPerm& perms, dev_t rdev) | |
6557 | { | |
6558 | Mutex::Locker lock(client_lock); | |
6559 | tout(cct) << "mknod" << std::endl; | |
6560 | tout(cct) << relpath << std::endl; | |
6561 | tout(cct) << mode << std::endl; | |
6562 | tout(cct) << rdev << std::endl; | |
6563 | ||
181888fb FG |
6564 | if (unmounting) |
6565 | return -ENOTCONN; | |
6566 | ||
7c673cae FG |
6567 | if (std::string(relpath) == "/") |
6568 | return -EEXIST; | |
6569 | ||
6570 | filepath path(relpath); | |
6571 | string name = path.last_dentry(); | |
6572 | path.pop_dentry(); | |
6573 | InodeRef dir; | |
6574 | int r = path_walk(path, &dir, perms); | |
6575 | if (r < 0) | |
6576 | return r; | |
6577 | if (cct->_conf->client_permissions) { | |
6578 | int r = may_create(dir.get(), perms); | |
6579 | if (r < 0) | |
6580 | return r; | |
6581 | } | |
6582 | return _mknod(dir.get(), name.c_str(), mode, rdev, perms); | |
6583 | } | |
6584 | ||
6585 | // symlinks | |
6586 | ||
6587 | int Client::symlink(const char *target, const char *relpath, const UserPerm& perms) | |
6588 | { | |
6589 | Mutex::Locker lock(client_lock); | |
6590 | tout(cct) << "symlink" << std::endl; | |
6591 | tout(cct) << target << std::endl; | |
6592 | tout(cct) << relpath << std::endl; | |
6593 | ||
181888fb FG |
6594 | if (unmounting) |
6595 | return -ENOTCONN; | |
6596 | ||
7c673cae FG |
6597 | if (std::string(relpath) == "/") |
6598 | return -EEXIST; | |
6599 | ||
6600 | filepath path(relpath); | |
6601 | string name = path.last_dentry(); | |
6602 | path.pop_dentry(); | |
6603 | InodeRef dir; | |
6604 | int r = path_walk(path, &dir, perms); | |
6605 | if (r < 0) | |
6606 | return r; | |
6607 | if (cct->_conf->client_permissions) { | |
6608 | int r = may_create(dir.get(), perms); | |
6609 | if (r < 0) | |
6610 | return r; | |
6611 | } | |
6612 | return _symlink(dir.get(), name.c_str(), target, perms); | |
6613 | } | |
6614 | ||
6615 | int Client::readlink(const char *relpath, char *buf, loff_t size, const UserPerm& perms) | |
6616 | { | |
6617 | Mutex::Locker lock(client_lock); | |
6618 | tout(cct) << "readlink" << std::endl; | |
6619 | tout(cct) << relpath << std::endl; | |
6620 | ||
181888fb FG |
6621 | if (unmounting) |
6622 | return -ENOTCONN; | |
6623 | ||
7c673cae FG |
6624 | filepath path(relpath); |
6625 | InodeRef in; | |
6626 | int r = path_walk(path, &in, perms, false); | |
6627 | if (r < 0) | |
6628 | return r; | |
6629 | ||
6630 | return _readlink(in.get(), buf, size); | |
6631 | } | |
6632 | ||
6633 | int Client::_readlink(Inode *in, char *buf, size_t size) | |
6634 | { | |
6635 | if (!in->is_symlink()) | |
6636 | return -EINVAL; | |
6637 | ||
6638 | // copy into buf (at most size bytes) | |
6639 | int r = in->symlink.length(); | |
6640 | if (r > (int)size) | |
6641 | r = size; | |
6642 | memcpy(buf, in->symlink.c_str(), r); | |
6643 | return r; | |
6644 | } | |
6645 | ||
6646 | ||
6647 | // inode stuff | |
6648 | ||
6649 | int Client::_getattr(Inode *in, int mask, const UserPerm& perms, bool force) | |
6650 | { | |
6651 | bool yes = in->caps_issued_mask(mask); | |
6652 | ||
6653 | ldout(cct, 10) << "_getattr mask " << ccap_string(mask) << " issued=" << yes << dendl; | |
6654 | if (yes && !force) | |
6655 | return 0; | |
6656 | ||
6657 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
6658 | filepath path; | |
6659 | in->make_nosnap_relative_path(path); | |
6660 | req->set_filepath(path); | |
6661 | req->set_inode(in); | |
6662 | req->head.args.getattr.mask = mask; | |
6663 | ||
6664 | int res = make_request(req, perms); | |
6665 | ldout(cct, 10) << "_getattr result=" << res << dendl; | |
6666 | return res; | |
6667 | } | |
6668 | ||
6669 | int Client::_do_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
6670 | const UserPerm& perms, InodeRef *inp) | |
6671 | { | |
6672 | int issued = in->caps_issued(); | |
6673 | ||
6674 | ldout(cct, 10) << "_setattr mask " << mask << " issued " << | |
6675 | ccap_string(issued) << dendl; | |
6676 | ||
6677 | if (in->snapid != CEPH_NOSNAP) { | |
6678 | return -EROFS; | |
6679 | } | |
6680 | if ((mask & CEPH_SETATTR_SIZE) && | |
6681 | (unsigned long)stx->stx_size > in->size && | |
6682 | is_quota_bytes_exceeded(in, (unsigned long)stx->stx_size - in->size, | |
6683 | perms)) { | |
6684 | return -EDQUOT; | |
6685 | } | |
6686 | ||
6687 | // make the change locally? | |
6688 | if ((in->cap_dirtier_uid >= 0 && perms.uid() != in->cap_dirtier_uid) || | |
6689 | (in->cap_dirtier_gid >= 0 && perms.gid() != in->cap_dirtier_gid)) { | |
6690 | ldout(cct, 10) << __func__ << " caller " << perms.uid() << ":" << perms.gid() | |
6691 | << " != cap dirtier " << in->cap_dirtier_uid << ":" | |
6692 | << in->cap_dirtier_gid << ", forcing sync setattr" | |
6693 | << dendl; | |
6694 | /* | |
6695 | * This works because we implicitly flush the caps as part of the | |
6696 | * request, so the cap update check will happen with the writeback | |
6697 | * cap context, and then the setattr check will happen with the | |
6698 | * caller's context. | |
6699 | * | |
6700 | * In reality this pattern is likely pretty rare (different users | |
6701 | * setattr'ing the same file). If that turns out not to be the | |
6702 | * case later, we can build a more complex pipelined cap writeback | |
6703 | * infrastructure... | |
6704 | */ | |
6705 | if (!mask) | |
6706 | mask |= CEPH_SETATTR_CTIME; | |
6707 | goto force_request; | |
6708 | } | |
6709 | ||
6710 | if (!mask) { | |
6711 | // caller just needs us to bump the ctime | |
6712 | in->ctime = ceph_clock_now(); | |
6713 | in->cap_dirtier_uid = perms.uid(); | |
6714 | in->cap_dirtier_gid = perms.gid(); | |
6715 | if (issued & CEPH_CAP_AUTH_EXCL) | |
6716 | mark_caps_dirty(in, CEPH_CAP_AUTH_EXCL); | |
6717 | else if (issued & CEPH_CAP_FILE_EXCL) | |
6718 | mark_caps_dirty(in, CEPH_CAP_FILE_EXCL); | |
6719 | else if (issued & CEPH_CAP_XATTR_EXCL) | |
6720 | mark_caps_dirty(in, CEPH_CAP_XATTR_EXCL); | |
6721 | else | |
6722 | mask |= CEPH_SETATTR_CTIME; | |
6723 | } | |
6724 | ||
6725 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
6726 | bool kill_sguid = mask & (CEPH_SETATTR_SIZE|CEPH_SETATTR_KILL_SGUID); | |
6727 | ||
6728 | mask &= ~CEPH_SETATTR_KILL_SGUID; | |
6729 | ||
6730 | if (mask & CEPH_SETATTR_UID) { | |
6731 | in->ctime = ceph_clock_now(); | |
6732 | in->cap_dirtier_uid = perms.uid(); | |
6733 | in->cap_dirtier_gid = perms.gid(); | |
6734 | in->uid = stx->stx_uid; | |
6735 | mark_caps_dirty(in, CEPH_CAP_AUTH_EXCL); | |
6736 | mask &= ~CEPH_SETATTR_UID; | |
6737 | kill_sguid = true; | |
6738 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
6739 | } | |
6740 | if (mask & CEPH_SETATTR_GID) { | |
6741 | in->ctime = ceph_clock_now(); | |
6742 | in->cap_dirtier_uid = perms.uid(); | |
6743 | in->cap_dirtier_gid = perms.gid(); | |
6744 | in->gid = stx->stx_gid; | |
6745 | mark_caps_dirty(in, CEPH_CAP_AUTH_EXCL); | |
6746 | mask &= ~CEPH_SETATTR_GID; | |
6747 | kill_sguid = true; | |
6748 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
6749 | } | |
6750 | ||
6751 | if (mask & CEPH_SETATTR_MODE) { | |
6752 | in->ctime = ceph_clock_now(); | |
6753 | in->cap_dirtier_uid = perms.uid(); | |
6754 | in->cap_dirtier_gid = perms.gid(); | |
6755 | in->mode = (in->mode & ~07777) | (stx->stx_mode & 07777); | |
6756 | mark_caps_dirty(in, CEPH_CAP_AUTH_EXCL); | |
6757 | mask &= ~CEPH_SETATTR_MODE; | |
6758 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
181888fb | 6759 | } else if (kill_sguid && S_ISREG(in->mode) && (in->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) { |
7c673cae | 6760 | /* Must squash the any setuid/setgid bits with an ownership change */ |
181888fb | 6761 | in->mode &= ~(S_ISUID|S_ISGID); |
7c673cae FG |
6762 | mark_caps_dirty(in, CEPH_CAP_AUTH_EXCL); |
6763 | } | |
6764 | ||
6765 | if (mask & CEPH_SETATTR_BTIME) { | |
6766 | in->ctime = ceph_clock_now(); | |
6767 | in->cap_dirtier_uid = perms.uid(); | |
6768 | in->cap_dirtier_gid = perms.gid(); | |
6769 | in->btime = utime_t(stx->stx_btime); | |
6770 | mark_caps_dirty(in, CEPH_CAP_AUTH_EXCL); | |
6771 | mask &= ~CEPH_SETATTR_BTIME; | |
6772 | ldout(cct,10) << "changing btime to " << in->btime << dendl; | |
6773 | } | |
6774 | } else if (mask & CEPH_SETATTR_SIZE) { | |
6775 | /* If we don't have Ax, then we must ask the server to clear them on truncate */ | |
6776 | mask |= CEPH_SETATTR_KILL_SGUID; | |
6777 | } | |
6778 | ||
6779 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
6780 | if (mask & (CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME)) { | |
6781 | if (mask & CEPH_SETATTR_MTIME) | |
6782 | in->mtime = utime_t(stx->stx_mtime); | |
6783 | if (mask & CEPH_SETATTR_ATIME) | |
6784 | in->atime = utime_t(stx->stx_atime); | |
6785 | in->ctime = ceph_clock_now(); | |
6786 | in->cap_dirtier_uid = perms.uid(); | |
6787 | in->cap_dirtier_gid = perms.gid(); | |
6788 | in->time_warp_seq++; | |
6789 | mark_caps_dirty(in, CEPH_CAP_FILE_EXCL); | |
6790 | mask &= ~(CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME); | |
6791 | } | |
6792 | } | |
6793 | if (!mask) { | |
6794 | in->change_attr++; | |
6795 | return 0; | |
6796 | } | |
6797 | ||
6798 | force_request: | |
6799 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETATTR); | |
6800 | ||
6801 | filepath path; | |
6802 | ||
6803 | in->make_nosnap_relative_path(path); | |
6804 | req->set_filepath(path); | |
6805 | req->set_inode(in); | |
6806 | ||
6807 | if (mask & CEPH_SETATTR_KILL_SGUID) { | |
6808 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
6809 | } | |
6810 | if (mask & CEPH_SETATTR_MODE) { | |
6811 | req->head.args.setattr.mode = stx->stx_mode; | |
6812 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
6813 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
6814 | } | |
6815 | if (mask & CEPH_SETATTR_UID) { | |
6816 | req->head.args.setattr.uid = stx->stx_uid; | |
6817 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
6818 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
6819 | } | |
6820 | if (mask & CEPH_SETATTR_GID) { | |
6821 | req->head.args.setattr.gid = stx->stx_gid; | |
6822 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
6823 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
6824 | } | |
6825 | if (mask & CEPH_SETATTR_BTIME) { | |
6826 | req->head.args.setattr.btime = utime_t(stx->stx_btime); | |
6827 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
6828 | } | |
6829 | if (mask & CEPH_SETATTR_MTIME) { | |
6830 | req->head.args.setattr.mtime = utime_t(stx->stx_mtime); | |
6831 | req->inode_drop |= CEPH_CAP_AUTH_SHARED | CEPH_CAP_FILE_RD | | |
6832 | CEPH_CAP_FILE_WR; | |
6833 | } | |
6834 | if (mask & CEPH_SETATTR_ATIME) { | |
6835 | req->head.args.setattr.atime = utime_t(stx->stx_atime); | |
6836 | req->inode_drop |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
6837 | CEPH_CAP_FILE_WR; | |
6838 | } | |
6839 | if (mask & CEPH_SETATTR_SIZE) { | |
6840 | if ((unsigned long)stx->stx_size < mdsmap->get_max_filesize()) { | |
6841 | req->head.args.setattr.size = stx->stx_size; | |
6842 | ldout(cct,10) << "changing size to " << stx->stx_size << dendl; | |
6843 | } else { //too big! | |
6844 | put_request(req); | |
6845 | ldout(cct,10) << "unable to set size to " << stx->stx_size << ". Too large!" << dendl; | |
6846 | return -EFBIG; | |
6847 | } | |
6848 | req->inode_drop |= CEPH_CAP_AUTH_SHARED | CEPH_CAP_FILE_RD | | |
6849 | CEPH_CAP_FILE_WR; | |
6850 | } | |
6851 | req->head.args.setattr.mask = mask; | |
6852 | ||
6853 | req->regetattr_mask = mask; | |
6854 | ||
6855 | int res = make_request(req, perms, inp); | |
6856 | ldout(cct, 10) << "_setattr result=" << res << dendl; | |
6857 | return res; | |
6858 | } | |
6859 | ||
6860 | /* Note that we only care about attrs that setattr cares about */ | |
6861 | void Client::stat_to_statx(struct stat *st, struct ceph_statx *stx) | |
6862 | { | |
6863 | stx->stx_size = st->st_size; | |
6864 | stx->stx_mode = st->st_mode; | |
6865 | stx->stx_uid = st->st_uid; | |
6866 | stx->stx_gid = st->st_gid; | |
6867 | stx->stx_mtime = st->st_mtim; | |
6868 | stx->stx_atime = st->st_atim; | |
6869 | } | |
6870 | ||
6871 | int Client::__setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
6872 | const UserPerm& perms, InodeRef *inp) | |
6873 | { | |
6874 | int ret = _do_setattr(in, stx, mask, perms, inp); | |
6875 | if (ret < 0) | |
6876 | return ret; | |
6877 | if (mask & CEPH_SETATTR_MODE) | |
6878 | ret = _posix_acl_chmod(in, stx->stx_mode, perms); | |
6879 | return ret; | |
6880 | } | |
6881 | ||
6882 | int Client::_setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
6883 | const UserPerm& perms) | |
6884 | { | |
6885 | mask &= (CEPH_SETATTR_MODE | CEPH_SETATTR_UID | | |
6886 | CEPH_SETATTR_GID | CEPH_SETATTR_MTIME | | |
6887 | CEPH_SETATTR_ATIME | CEPH_SETATTR_SIZE | | |
6888 | CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME); | |
6889 | if (cct->_conf->client_permissions) { | |
6890 | int r = may_setattr(in.get(), stx, mask, perms); | |
6891 | if (r < 0) | |
6892 | return r; | |
6893 | } | |
6894 | return __setattrx(in.get(), stx, mask, perms); | |
6895 | } | |
6896 | ||
6897 | int Client::_setattr(InodeRef &in, struct stat *attr, int mask, | |
6898 | const UserPerm& perms) | |
6899 | { | |
6900 | struct ceph_statx stx; | |
6901 | ||
6902 | stat_to_statx(attr, &stx); | |
6903 | mask &= ~CEPH_SETATTR_BTIME; | |
181888fb FG |
6904 | |
6905 | if ((mask & CEPH_SETATTR_UID) && attr->st_uid == static_cast<uid_t>(-1)) { | |
6906 | mask &= ~CEPH_SETATTR_UID; | |
6907 | } | |
6908 | if ((mask & CEPH_SETATTR_GID) && attr->st_gid == static_cast<uid_t>(-1)) { | |
6909 | mask &= ~CEPH_SETATTR_GID; | |
6910 | } | |
6911 | ||
7c673cae FG |
6912 | return _setattrx(in, &stx, mask, perms); |
6913 | } | |
6914 | ||
6915 | int Client::setattr(const char *relpath, struct stat *attr, int mask, | |
6916 | const UserPerm& perms) | |
6917 | { | |
6918 | Mutex::Locker lock(client_lock); | |
6919 | tout(cct) << "setattr" << std::endl; | |
6920 | tout(cct) << relpath << std::endl; | |
6921 | tout(cct) << mask << std::endl; | |
6922 | ||
181888fb FG |
6923 | if (unmounting) |
6924 | return -ENOTCONN; | |
6925 | ||
7c673cae FG |
6926 | filepath path(relpath); |
6927 | InodeRef in; | |
6928 | int r = path_walk(path, &in, perms); | |
6929 | if (r < 0) | |
6930 | return r; | |
6931 | return _setattr(in, attr, mask, perms); | |
6932 | } | |
6933 | ||
6934 | int Client::setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
6935 | const UserPerm& perms, int flags) | |
6936 | { | |
6937 | Mutex::Locker lock(client_lock); | |
6938 | tout(cct) << "setattrx" << std::endl; | |
6939 | tout(cct) << relpath << std::endl; | |
6940 | tout(cct) << mask << std::endl; | |
6941 | ||
181888fb FG |
6942 | if (unmounting) |
6943 | return -ENOTCONN; | |
6944 | ||
7c673cae FG |
6945 | filepath path(relpath); |
6946 | InodeRef in; | |
6947 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW)); | |
6948 | if (r < 0) | |
6949 | return r; | |
6950 | return _setattrx(in, stx, mask, perms); | |
6951 | } | |
6952 | ||
6953 | int Client::fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms) | |
6954 | { | |
6955 | Mutex::Locker lock(client_lock); | |
6956 | tout(cct) << "fsetattr" << std::endl; | |
6957 | tout(cct) << fd << std::endl; | |
6958 | tout(cct) << mask << std::endl; | |
6959 | ||
181888fb FG |
6960 | if (unmounting) |
6961 | return -ENOTCONN; | |
6962 | ||
7c673cae FG |
6963 | Fh *f = get_filehandle(fd); |
6964 | if (!f) | |
6965 | return -EBADF; | |
6966 | #if defined(__linux__) && defined(O_PATH) | |
6967 | if (f->flags & O_PATH) | |
6968 | return -EBADF; | |
6969 | #endif | |
6970 | return _setattr(f->inode, attr, mask, perms); | |
6971 | } | |
6972 | ||
6973 | int Client::fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms) | |
6974 | { | |
6975 | Mutex::Locker lock(client_lock); | |
6976 | tout(cct) << "fsetattr" << std::endl; | |
6977 | tout(cct) << fd << std::endl; | |
6978 | tout(cct) << mask << std::endl; | |
6979 | ||
181888fb FG |
6980 | if (unmounting) |
6981 | return -ENOTCONN; | |
6982 | ||
7c673cae FG |
6983 | Fh *f = get_filehandle(fd); |
6984 | if (!f) | |
6985 | return -EBADF; | |
6986 | #if defined(__linux__) && defined(O_PATH) | |
6987 | if (f->flags & O_PATH) | |
6988 | return -EBADF; | |
6989 | #endif | |
6990 | return _setattrx(f->inode, stx, mask, perms); | |
6991 | } | |
6992 | ||
6993 | int Client::stat(const char *relpath, struct stat *stbuf, const UserPerm& perms, | |
6994 | frag_info_t *dirstat, int mask) | |
6995 | { | |
6996 | ldout(cct, 3) << "stat enter (relpath " << relpath << " mask " << mask << ")" << dendl; | |
6997 | Mutex::Locker lock(client_lock); | |
6998 | tout(cct) << "stat" << std::endl; | |
6999 | tout(cct) << relpath << std::endl; | |
181888fb FG |
7000 | |
7001 | if (unmounting) | |
7002 | return -ENOTCONN; | |
7003 | ||
7c673cae FG |
7004 | filepath path(relpath); |
7005 | InodeRef in; | |
7006 | int r = path_walk(path, &in, perms, true, mask); | |
7007 | if (r < 0) | |
7008 | return r; | |
7009 | r = _getattr(in, mask, perms); | |
7010 | if (r < 0) { | |
7011 | ldout(cct, 3) << "stat exit on error!" << dendl; | |
7012 | return r; | |
7013 | } | |
7014 | fill_stat(in, stbuf, dirstat); | |
7015 | ldout(cct, 3) << "stat exit (relpath " << relpath << " mask " << mask << ")" << dendl; | |
7016 | return r; | |
7017 | } | |
7018 | ||
7019 | unsigned Client::statx_to_mask(unsigned int flags, unsigned int want) | |
7020 | { | |
7021 | unsigned mask = 0; | |
7022 | ||
7023 | /* if NO_ATTR_SYNC is set, then we don't need any -- just use what's in cache */ | |
7024 | if (flags & AT_NO_ATTR_SYNC) | |
7025 | goto out; | |
7026 | ||
7027 | /* Always set PIN to distinguish from AT_NO_ATTR_SYNC case */ | |
7028 | mask |= CEPH_CAP_PIN; | |
7029 | if (want & (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7030 | mask |= CEPH_CAP_AUTH_SHARED; | |
7031 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7032 | mask |= CEPH_CAP_LINK_SHARED; | |
7033 | if (want & (CEPH_STATX_ATIME|CEPH_STATX_MTIME|CEPH_STATX_CTIME|CEPH_STATX_SIZE|CEPH_STATX_BLOCKS|CEPH_STATX_VERSION)) | |
7034 | mask |= CEPH_CAP_FILE_SHARED; | |
7035 | if (want & (CEPH_STATX_VERSION|CEPH_STATX_CTIME)) | |
7036 | mask |= CEPH_CAP_XATTR_SHARED; | |
7037 | out: | |
7038 | return mask; | |
7039 | } | |
7040 | ||
7041 | int Client::statx(const char *relpath, struct ceph_statx *stx, | |
7042 | const UserPerm& perms, | |
7043 | unsigned int want, unsigned int flags) | |
7044 | { | |
7045 | ldout(cct, 3) << "statx enter (relpath " << relpath << " want " << want << ")" << dendl; | |
7046 | Mutex::Locker lock(client_lock); | |
7047 | tout(cct) << "statx" << std::endl; | |
7048 | tout(cct) << relpath << std::endl; | |
181888fb FG |
7049 | |
7050 | if (unmounting) | |
7051 | return -ENOTCONN; | |
7052 | ||
7c673cae FG |
7053 | filepath path(relpath); |
7054 | InodeRef in; | |
7055 | ||
7056 | unsigned mask = statx_to_mask(flags, want); | |
7057 | ||
7058 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); | |
7059 | if (r < 0) | |
7060 | return r; | |
7061 | ||
7062 | r = _getattr(in, mask, perms); | |
7063 | if (r < 0) { | |
7064 | ldout(cct, 3) << "statx exit on error!" << dendl; | |
7065 | return r; | |
7066 | } | |
7067 | ||
7068 | fill_statx(in, mask, stx); | |
7069 | ldout(cct, 3) << "statx exit (relpath " << relpath << " mask " << stx->stx_mask << ")" << dendl; | |
7070 | return r; | |
7071 | } | |
7072 | ||
7073 | int Client::lstat(const char *relpath, struct stat *stbuf, | |
7074 | const UserPerm& perms, frag_info_t *dirstat, int mask) | |
7075 | { | |
7076 | ldout(cct, 3) << "lstat enter (relpath " << relpath << " mask " << mask << ")" << dendl; | |
7077 | Mutex::Locker lock(client_lock); | |
7078 | tout(cct) << "lstat" << std::endl; | |
7079 | tout(cct) << relpath << std::endl; | |
181888fb FG |
7080 | |
7081 | if (unmounting) | |
7082 | return -ENOTCONN; | |
7083 | ||
7c673cae FG |
7084 | filepath path(relpath); |
7085 | InodeRef in; | |
7086 | // don't follow symlinks | |
7087 | int r = path_walk(path, &in, perms, false, mask); | |
7088 | if (r < 0) | |
7089 | return r; | |
7090 | r = _getattr(in, mask, perms); | |
7091 | if (r < 0) { | |
7092 | ldout(cct, 3) << "lstat exit on error!" << dendl; | |
7093 | return r; | |
7094 | } | |
7095 | fill_stat(in, stbuf, dirstat); | |
7096 | ldout(cct, 3) << "lstat exit (relpath " << relpath << " mask " << mask << ")" << dendl; | |
7097 | return r; | |
7098 | } | |
7099 | ||
7100 | int Client::fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat, nest_info_t *rstat) | |
7101 | { | |
7102 | ldout(cct, 10) << "fill_stat on " << in->ino << " snap/dev" << in->snapid | |
7103 | << " mode 0" << oct << in->mode << dec | |
7104 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7105 | memset(st, 0, sizeof(struct stat)); | |
7106 | if (use_faked_inos()) | |
7107 | st->st_ino = in->faked_ino; | |
7108 | else | |
7109 | st->st_ino = in->ino; | |
7110 | st->st_dev = in->snapid; | |
7111 | st->st_mode = in->mode; | |
7112 | st->st_rdev = in->rdev; | |
7113 | st->st_nlink = in->nlink; | |
7114 | st->st_uid = in->uid; | |
7115 | st->st_gid = in->gid; | |
7116 | if (in->ctime > in->mtime) { | |
7117 | stat_set_ctime_sec(st, in->ctime.sec()); | |
7118 | stat_set_ctime_nsec(st, in->ctime.nsec()); | |
7119 | } else { | |
7120 | stat_set_ctime_sec(st, in->mtime.sec()); | |
7121 | stat_set_ctime_nsec(st, in->mtime.nsec()); | |
7122 | } | |
7123 | stat_set_atime_sec(st, in->atime.sec()); | |
7124 | stat_set_atime_nsec(st, in->atime.nsec()); | |
7125 | stat_set_mtime_sec(st, in->mtime.sec()); | |
7126 | stat_set_mtime_nsec(st, in->mtime.nsec()); | |
7127 | if (in->is_dir()) { | |
7128 | if (cct->_conf->client_dirsize_rbytes) | |
7129 | st->st_size = in->rstat.rbytes; | |
7130 | else | |
7131 | st->st_size = in->dirstat.size(); | |
7132 | st->st_blocks = 1; | |
7133 | } else { | |
7134 | st->st_size = in->size; | |
7135 | st->st_blocks = (in->size + 511) >> 9; | |
7136 | } | |
7137 | st->st_blksize = MAX(in->layout.stripe_unit, 4096); | |
7138 | ||
7139 | if (dirstat) | |
7140 | *dirstat = in->dirstat; | |
7141 | if (rstat) | |
7142 | *rstat = in->rstat; | |
7143 | ||
7144 | return in->caps_issued(); | |
7145 | } | |
7146 | ||
7147 | void Client::fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx) | |
7148 | { | |
7149 | ldout(cct, 10) << "fill_statx on " << in->ino << " snap/dev" << in->snapid | |
7150 | << " mode 0" << oct << in->mode << dec | |
7151 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7152 | memset(stx, 0, sizeof(struct ceph_statx)); | |
7153 | ||
7154 | /* | |
7155 | * If mask is 0, then the caller set AT_NO_ATTR_SYNC. Reset the mask | |
7156 | * so that all bits are set. | |
7157 | */ | |
7158 | if (!mask) | |
7159 | mask = ~0; | |
7160 | ||
7161 | /* These are always considered to be available */ | |
7162 | stx->stx_dev = in->snapid; | |
7163 | stx->stx_blksize = MAX(in->layout.stripe_unit, 4096); | |
7164 | ||
7165 | /* Type bits are always set, even when CEPH_STATX_MODE is not */ | |
7166 | stx->stx_mode = S_IFMT & in->mode; | |
7167 | stx->stx_ino = use_faked_inos() ? in->faked_ino : (ino_t)in->ino; | |
7168 | stx->stx_rdev = in->rdev; | |
7169 | stx->stx_mask |= (CEPH_STATX_INO|CEPH_STATX_RDEV); | |
7170 | ||
7171 | if (mask & CEPH_CAP_AUTH_SHARED) { | |
7172 | stx->stx_uid = in->uid; | |
7173 | stx->stx_gid = in->gid; | |
7174 | stx->stx_mode = in->mode; | |
7175 | in->btime.to_timespec(&stx->stx_btime); | |
7176 | stx->stx_mask |= (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME); | |
7177 | } | |
7178 | ||
7179 | if (mask & CEPH_CAP_LINK_SHARED) { | |
7180 | stx->stx_nlink = in->nlink; | |
7181 | stx->stx_mask |= CEPH_STATX_NLINK; | |
7182 | } | |
7183 | ||
7184 | if (mask & CEPH_CAP_FILE_SHARED) { | |
7185 | ||
7186 | in->atime.to_timespec(&stx->stx_atime); | |
7187 | in->mtime.to_timespec(&stx->stx_mtime); | |
7188 | ||
7189 | if (in->is_dir()) { | |
7190 | if (cct->_conf->client_dirsize_rbytes) | |
7191 | stx->stx_size = in->rstat.rbytes; | |
7192 | else | |
7193 | stx->stx_size = in->dirstat.size(); | |
7194 | stx->stx_blocks = 1; | |
7195 | } else { | |
7196 | stx->stx_size = in->size; | |
7197 | stx->stx_blocks = (in->size + 511) >> 9; | |
7198 | } | |
7199 | stx->stx_mask |= (CEPH_STATX_ATIME|CEPH_STATX_MTIME| | |
7200 | CEPH_STATX_SIZE|CEPH_STATX_BLOCKS); | |
7201 | } | |
7202 | ||
7203 | /* Change time and change_attr both require all shared caps to view */ | |
7204 | if ((mask & CEPH_STAT_CAP_INODE_ALL) == CEPH_STAT_CAP_INODE_ALL) { | |
7205 | stx->stx_version = in->change_attr; | |
7206 | if (in->ctime > in->mtime) | |
7207 | in->ctime.to_timespec(&stx->stx_ctime); | |
7208 | else | |
7209 | in->mtime.to_timespec(&stx->stx_ctime); | |
7210 | stx->stx_mask |= (CEPH_STATX_CTIME|CEPH_STATX_VERSION); | |
7211 | } | |
7212 | ||
7213 | } | |
7214 | ||
7215 | void Client::touch_dn(Dentry *dn) | |
7216 | { | |
7217 | lru.lru_touch(dn); | |
7218 | } | |
7219 | ||
7220 | int Client::chmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
7221 | { | |
7222 | Mutex::Locker lock(client_lock); | |
7223 | tout(cct) << "chmod" << std::endl; | |
7224 | tout(cct) << relpath << std::endl; | |
7225 | tout(cct) << mode << std::endl; | |
181888fb FG |
7226 | |
7227 | if (unmounting) | |
7228 | return -ENOTCONN; | |
7229 | ||
7c673cae FG |
7230 | filepath path(relpath); |
7231 | InodeRef in; | |
7232 | int r = path_walk(path, &in, perms); | |
7233 | if (r < 0) | |
7234 | return r; | |
7235 | struct stat attr; | |
7236 | attr.st_mode = mode; | |
7237 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
7238 | } | |
7239 | ||
7240 | int Client::fchmod(int fd, mode_t mode, const UserPerm& perms) | |
7241 | { | |
7242 | Mutex::Locker lock(client_lock); | |
7243 | tout(cct) << "fchmod" << std::endl; | |
7244 | tout(cct) << fd << std::endl; | |
7245 | tout(cct) << mode << std::endl; | |
181888fb FG |
7246 | |
7247 | if (unmounting) | |
7248 | return -ENOTCONN; | |
7249 | ||
7c673cae FG |
7250 | Fh *f = get_filehandle(fd); |
7251 | if (!f) | |
7252 | return -EBADF; | |
7253 | #if defined(__linux__) && defined(O_PATH) | |
7254 | if (f->flags & O_PATH) | |
7255 | return -EBADF; | |
7256 | #endif | |
7257 | struct stat attr; | |
7258 | attr.st_mode = mode; | |
7259 | return _setattr(f->inode, &attr, CEPH_SETATTR_MODE, perms); | |
7260 | } | |
7261 | ||
7262 | int Client::lchmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
7263 | { | |
7264 | Mutex::Locker lock(client_lock); | |
7265 | tout(cct) << "lchmod" << std::endl; | |
7266 | tout(cct) << relpath << std::endl; | |
7267 | tout(cct) << mode << std::endl; | |
181888fb FG |
7268 | |
7269 | if (unmounting) | |
7270 | return -ENOTCONN; | |
7271 | ||
7c673cae FG |
7272 | filepath path(relpath); |
7273 | InodeRef in; | |
7274 | // don't follow symlinks | |
7275 | int r = path_walk(path, &in, perms, false); | |
7276 | if (r < 0) | |
7277 | return r; | |
7278 | struct stat attr; | |
7279 | attr.st_mode = mode; | |
7280 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
7281 | } | |
7282 | ||
7283 | int Client::chown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
7284 | const UserPerm& perms) | |
7285 | { | |
7286 | Mutex::Locker lock(client_lock); | |
7287 | tout(cct) << "chown" << std::endl; | |
7288 | tout(cct) << relpath << std::endl; | |
7289 | tout(cct) << new_uid << std::endl; | |
7290 | tout(cct) << new_gid << std::endl; | |
181888fb FG |
7291 | |
7292 | if (unmounting) | |
7293 | return -ENOTCONN; | |
7294 | ||
7c673cae FG |
7295 | filepath path(relpath); |
7296 | InodeRef in; | |
7297 | int r = path_walk(path, &in, perms); | |
7298 | if (r < 0) | |
7299 | return r; | |
7300 | struct stat attr; | |
7301 | attr.st_uid = new_uid; | |
7302 | attr.st_gid = new_gid; | |
181888fb | 7303 | return _setattr(in, &attr, CEPH_SETATTR_UID|CEPH_SETATTR_GID, perms); |
7c673cae FG |
7304 | } |
7305 | ||
7306 | int Client::fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms) | |
7307 | { | |
7308 | Mutex::Locker lock(client_lock); | |
7309 | tout(cct) << "fchown" << std::endl; | |
7310 | tout(cct) << fd << std::endl; | |
7311 | tout(cct) << new_uid << std::endl; | |
7312 | tout(cct) << new_gid << std::endl; | |
181888fb FG |
7313 | |
7314 | if (unmounting) | |
7315 | return -ENOTCONN; | |
7316 | ||
7c673cae FG |
7317 | Fh *f = get_filehandle(fd); |
7318 | if (!f) | |
7319 | return -EBADF; | |
7320 | #if defined(__linux__) && defined(O_PATH) | |
7321 | if (f->flags & O_PATH) | |
7322 | return -EBADF; | |
7323 | #endif | |
7324 | struct stat attr; | |
7325 | attr.st_uid = new_uid; | |
7326 | attr.st_gid = new_gid; | |
7327 | int mask = 0; | |
7328 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
7329 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
7330 | return _setattr(f->inode, &attr, mask, perms); | |
7331 | } | |
7332 | ||
7333 | int Client::lchown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
7334 | const UserPerm& perms) | |
7335 | { | |
7336 | Mutex::Locker lock(client_lock); | |
7337 | tout(cct) << "lchown" << std::endl; | |
7338 | tout(cct) << relpath << std::endl; | |
7339 | tout(cct) << new_uid << std::endl; | |
7340 | tout(cct) << new_gid << std::endl; | |
181888fb FG |
7341 | |
7342 | if (unmounting) | |
7343 | return -ENOTCONN; | |
7344 | ||
7c673cae FG |
7345 | filepath path(relpath); |
7346 | InodeRef in; | |
7347 | // don't follow symlinks | |
7348 | int r = path_walk(path, &in, perms, false); | |
7349 | if (r < 0) | |
7350 | return r; | |
7351 | struct stat attr; | |
7352 | attr.st_uid = new_uid; | |
7353 | attr.st_gid = new_gid; | |
7354 | int mask = 0; | |
7355 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
7356 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
7357 | return _setattr(in, &attr, mask, perms); | |
7358 | } | |
7359 | ||
7360 | int Client::utime(const char *relpath, struct utimbuf *buf, | |
7361 | const UserPerm& perms) | |
7362 | { | |
7363 | Mutex::Locker lock(client_lock); | |
7364 | tout(cct) << "utime" << std::endl; | |
7365 | tout(cct) << relpath << std::endl; | |
7366 | tout(cct) << buf->modtime << std::endl; | |
7367 | tout(cct) << buf->actime << std::endl; | |
181888fb FG |
7368 | |
7369 | if (unmounting) | |
7370 | return -ENOTCONN; | |
7371 | ||
7c673cae FG |
7372 | filepath path(relpath); |
7373 | InodeRef in; | |
7374 | int r = path_walk(path, &in, perms); | |
7375 | if (r < 0) | |
7376 | return r; | |
7377 | struct stat attr; | |
7378 | stat_set_mtime_sec(&attr, buf->modtime); | |
7379 | stat_set_mtime_nsec(&attr, 0); | |
7380 | stat_set_atime_sec(&attr, buf->actime); | |
7381 | stat_set_atime_nsec(&attr, 0); | |
7382 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
7383 | } | |
7384 | ||
7385 | int Client::lutime(const char *relpath, struct utimbuf *buf, | |
7386 | const UserPerm& perms) | |
7387 | { | |
7388 | Mutex::Locker lock(client_lock); | |
7389 | tout(cct) << "lutime" << std::endl; | |
7390 | tout(cct) << relpath << std::endl; | |
7391 | tout(cct) << buf->modtime << std::endl; | |
7392 | tout(cct) << buf->actime << std::endl; | |
181888fb FG |
7393 | |
7394 | if (unmounting) | |
7395 | return -ENOTCONN; | |
7396 | ||
7c673cae FG |
7397 | filepath path(relpath); |
7398 | InodeRef in; | |
7399 | // don't follow symlinks | |
7400 | int r = path_walk(path, &in, perms, false); | |
7401 | if (r < 0) | |
7402 | return r; | |
7403 | struct stat attr; | |
7404 | stat_set_mtime_sec(&attr, buf->modtime); | |
7405 | stat_set_mtime_nsec(&attr, 0); | |
7406 | stat_set_atime_sec(&attr, buf->actime); | |
7407 | stat_set_atime_nsec(&attr, 0); | |
7408 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
7409 | } | |
7410 | ||
7411 | int Client::flock(int fd, int operation, uint64_t owner) | |
7412 | { | |
7413 | Mutex::Locker lock(client_lock); | |
7414 | tout(cct) << "flock" << std::endl; | |
7415 | tout(cct) << fd << std::endl; | |
7416 | tout(cct) << operation << std::endl; | |
7417 | tout(cct) << owner << std::endl; | |
181888fb FG |
7418 | |
7419 | if (unmounting) | |
7420 | return -ENOTCONN; | |
7421 | ||
7c673cae FG |
7422 | Fh *f = get_filehandle(fd); |
7423 | if (!f) | |
7424 | return -EBADF; | |
7425 | ||
7426 | return _flock(f, operation, owner); | |
7427 | } | |
7428 | ||
7429 | int Client::opendir(const char *relpath, dir_result_t **dirpp, const UserPerm& perms) | |
7430 | { | |
7431 | Mutex::Locker lock(client_lock); | |
7432 | tout(cct) << "opendir" << std::endl; | |
7433 | tout(cct) << relpath << std::endl; | |
181888fb FG |
7434 | |
7435 | if (unmounting) | |
7436 | return -ENOTCONN; | |
7437 | ||
7c673cae FG |
7438 | filepath path(relpath); |
7439 | InodeRef in; | |
7440 | int r = path_walk(path, &in, perms, true); | |
7441 | if (r < 0) | |
7442 | return r; | |
7443 | if (cct->_conf->client_permissions) { | |
7444 | int r = may_open(in.get(), O_RDONLY, perms); | |
7445 | if (r < 0) | |
7446 | return r; | |
7447 | } | |
7448 | r = _opendir(in.get(), dirpp, perms); | |
7449 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
7450 | if (r != -ENOTDIR) | |
7451 | tout(cct) << (unsigned long)*dirpp << std::endl; | |
7452 | return r; | |
7453 | } | |
7454 | ||
7455 | int Client::_opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms) | |
7456 | { | |
7457 | if (!in->is_dir()) | |
7458 | return -ENOTDIR; | |
7459 | *dirpp = new dir_result_t(in, perms); | |
7460 | opened_dirs.insert(*dirpp); | |
7461 | ldout(cct, 3) << "_opendir(" << in->ino << ") = " << 0 << " (" << *dirpp << ")" << dendl; | |
7462 | return 0; | |
7463 | } | |
7464 | ||
7465 | ||
7466 | int Client::closedir(dir_result_t *dir) | |
7467 | { | |
7468 | Mutex::Locker lock(client_lock); | |
7469 | tout(cct) << "closedir" << std::endl; | |
7470 | tout(cct) << (unsigned long)dir << std::endl; | |
7471 | ||
7472 | ldout(cct, 3) << "closedir(" << dir << ") = 0" << dendl; | |
7473 | _closedir(dir); | |
7474 | return 0; | |
7475 | } | |
7476 | ||
7477 | void Client::_closedir(dir_result_t *dirp) | |
7478 | { | |
7479 | ldout(cct, 10) << "_closedir(" << dirp << ")" << dendl; | |
7480 | if (dirp->inode) { | |
7481 | ldout(cct, 10) << "_closedir detaching inode " << dirp->inode << dendl; | |
7482 | dirp->inode.reset(); | |
7483 | } | |
7484 | _readdir_drop_dirp_buffer(dirp); | |
7485 | opened_dirs.erase(dirp); | |
7486 | delete dirp; | |
7487 | } | |
7488 | ||
7489 | void Client::rewinddir(dir_result_t *dirp) | |
7490 | { | |
7491 | Mutex::Locker lock(client_lock); | |
7c673cae | 7492 | ldout(cct, 3) << "rewinddir(" << dirp << ")" << dendl; |
181888fb FG |
7493 | |
7494 | if (unmounting) | |
7495 | return; | |
7496 | ||
7c673cae FG |
7497 | dir_result_t *d = static_cast<dir_result_t*>(dirp); |
7498 | _readdir_drop_dirp_buffer(d); | |
7499 | d->reset(); | |
7500 | } | |
7501 | ||
7502 | loff_t Client::telldir(dir_result_t *dirp) | |
7503 | { | |
7504 | dir_result_t *d = static_cast<dir_result_t*>(dirp); | |
7505 | ldout(cct, 3) << "telldir(" << dirp << ") = " << d->offset << dendl; | |
7506 | return d->offset; | |
7507 | } | |
7508 | ||
7509 | void Client::seekdir(dir_result_t *dirp, loff_t offset) | |
7510 | { | |
7511 | Mutex::Locker lock(client_lock); | |
7512 | ||
7513 | ldout(cct, 3) << "seekdir(" << dirp << ", " << offset << ")" << dendl; | |
7514 | ||
181888fb FG |
7515 | if (unmounting) |
7516 | return; | |
7517 | ||
7c673cae FG |
7518 | if (offset == dirp->offset) |
7519 | return; | |
7520 | ||
7521 | if (offset > dirp->offset) | |
7522 | dirp->release_count = 0; // bump if we do a forward seek | |
7523 | else | |
7524 | dirp->ordered_count = 0; // disable filling readdir cache | |
7525 | ||
7526 | if (dirp->hash_order()) { | |
7527 | if (dirp->offset > offset) { | |
7528 | _readdir_drop_dirp_buffer(dirp); | |
7529 | dirp->reset(); | |
7530 | } | |
7531 | } else { | |
7532 | if (offset == 0 || | |
7533 | dirp->buffer_frag != frag_t(dir_result_t::fpos_high(offset)) || | |
7534 | dirp->offset_low() > dir_result_t::fpos_low(offset)) { | |
7535 | _readdir_drop_dirp_buffer(dirp); | |
7536 | dirp->reset(); | |
7537 | } | |
7538 | } | |
7539 | ||
7540 | dirp->offset = offset; | |
7541 | } | |
7542 | ||
7543 | ||
7544 | //struct dirent { | |
7545 | // ino_t d_ino; /* inode number */ | |
7546 | // off_t d_off; /* offset to the next dirent */ | |
7547 | // unsigned short d_reclen; /* length of this record */ | |
7548 | // unsigned char d_type; /* type of file */ | |
7549 | // char d_name[256]; /* filename */ | |
7550 | //}; | |
7551 | void Client::fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off) | |
7552 | { | |
7553 | strncpy(de->d_name, name, 255); | |
7554 | de->d_name[255] = '\0'; | |
7555 | #ifndef __CYGWIN__ | |
7556 | de->d_ino = ino; | |
7557 | #if !defined(DARWIN) && !defined(__FreeBSD__) | |
7558 | de->d_off = next_off; | |
7559 | #endif | |
7560 | de->d_reclen = 1; | |
7561 | de->d_type = IFTODT(type); | |
7562 | ldout(cct, 10) << "fill_dirent '" << de->d_name << "' -> " << inodeno_t(de->d_ino) | |
7563 | << " type " << (int)de->d_type << " w/ next_off " << hex << next_off << dec << dendl; | |
7564 | #endif | |
7565 | } | |
7566 | ||
7567 | void Client::_readdir_next_frag(dir_result_t *dirp) | |
7568 | { | |
7569 | frag_t fg = dirp->buffer_frag; | |
7570 | ||
7571 | if (fg.is_rightmost()) { | |
7572 | ldout(cct, 10) << "_readdir_next_frag advance from " << fg << " to END" << dendl; | |
7573 | dirp->set_end(); | |
7574 | return; | |
7575 | } | |
7576 | ||
7577 | // advance | |
7578 | fg = fg.next(); | |
7579 | ldout(cct, 10) << "_readdir_next_frag advance from " << dirp->buffer_frag << " to " << fg << dendl; | |
7580 | ||
7581 | if (dirp->hash_order()) { | |
7582 | // keep last_name | |
7583 | int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true); | |
7584 | if (dirp->offset < new_offset) // don't decrease offset | |
7585 | dirp->offset = new_offset; | |
7586 | } else { | |
7587 | dirp->last_name.clear(); | |
7588 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
7589 | _readdir_rechoose_frag(dirp); | |
7590 | } | |
7591 | } | |
7592 | ||
7593 | void Client::_readdir_rechoose_frag(dir_result_t *dirp) | |
7594 | { | |
7595 | assert(dirp->inode); | |
7596 | ||
7597 | if (dirp->hash_order()) | |
7598 | return; | |
7599 | ||
7600 | frag_t cur = frag_t(dirp->offset_high()); | |
7601 | frag_t fg = dirp->inode->dirfragtree[cur.value()]; | |
7602 | if (fg != cur) { | |
7603 | ldout(cct, 10) << "_readdir_rechoose_frag frag " << cur << " maps to " << fg << dendl; | |
7604 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
7605 | dirp->last_name.clear(); | |
7606 | dirp->next_offset = 2; | |
7607 | } | |
7608 | } | |
7609 | ||
7610 | void Client::_readdir_drop_dirp_buffer(dir_result_t *dirp) | |
7611 | { | |
7612 | ldout(cct, 10) << "_readdir_drop_dirp_buffer " << dirp << dendl; | |
7613 | dirp->buffer.clear(); | |
7614 | } | |
7615 | ||
7616 | int Client::_readdir_get_frag(dir_result_t *dirp) | |
7617 | { | |
7618 | assert(dirp); | |
7619 | assert(dirp->inode); | |
7620 | ||
7621 | // get the current frag. | |
7622 | frag_t fg; | |
7623 | if (dirp->hash_order()) | |
7624 | fg = dirp->inode->dirfragtree[dirp->offset_high()]; | |
7625 | else | |
7626 | fg = frag_t(dirp->offset_high()); | |
7627 | ||
7628 | ldout(cct, 10) << "_readdir_get_frag " << dirp << " on " << dirp->inode->ino << " fg " << fg | |
7629 | << " offset " << hex << dirp->offset << dec << dendl; | |
7630 | ||
7631 | int op = CEPH_MDS_OP_READDIR; | |
7632 | if (dirp->inode && dirp->inode->snapid == CEPH_SNAPDIR) | |
7633 | op = CEPH_MDS_OP_LSSNAP; | |
7634 | ||
7635 | InodeRef& diri = dirp->inode; | |
7636 | ||
7637 | MetaRequest *req = new MetaRequest(op); | |
7638 | filepath path; | |
7639 | diri->make_nosnap_relative_path(path); | |
7640 | req->set_filepath(path); | |
7641 | req->set_inode(diri.get()); | |
7642 | req->head.args.readdir.frag = fg; | |
7643 | req->head.args.readdir.flags = CEPH_READDIR_REPLY_BITFLAGS; | |
7644 | if (dirp->last_name.length()) { | |
7645 | req->path2.set_path(dirp->last_name.c_str()); | |
7646 | } else if (dirp->hash_order()) { | |
7647 | req->head.args.readdir.offset_hash = dirp->offset_high(); | |
7648 | } | |
7649 | req->dirp = dirp; | |
7650 | ||
7651 | bufferlist dirbl; | |
7652 | int res = make_request(req, dirp->perms, NULL, NULL, -1, &dirbl); | |
7653 | ||
7654 | if (res == -EAGAIN) { | |
7655 | ldout(cct, 10) << "_readdir_get_frag got EAGAIN, retrying" << dendl; | |
7656 | _readdir_rechoose_frag(dirp); | |
7657 | return _readdir_get_frag(dirp); | |
7658 | } | |
7659 | ||
7660 | if (res == 0) { | |
7661 | ldout(cct, 10) << "_readdir_get_frag " << dirp << " got frag " << dirp->buffer_frag | |
7662 | << " size " << dirp->buffer.size() << dendl; | |
7663 | } else { | |
7664 | ldout(cct, 10) << "_readdir_get_frag got error " << res << ", setting end flag" << dendl; | |
7665 | dirp->set_end(); | |
7666 | } | |
7667 | ||
7668 | return res; | |
7669 | } | |
7670 | ||
7671 | struct dentry_off_lt { | |
7672 | bool operator()(const Dentry* dn, int64_t off) const { | |
7673 | return dir_result_t::fpos_cmp(dn->offset, off) < 0; | |
7674 | } | |
7675 | }; | |
7676 | ||
7677 | int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
7678 | int caps, bool getref) | |
7679 | { | |
7680 | assert(client_lock.is_locked()); | |
7681 | ldout(cct, 10) << "_readdir_cache_cb " << dirp << " on " << dirp->inode->ino | |
7682 | << " last_name " << dirp->last_name << " offset " << hex << dirp->offset << dec | |
7683 | << dendl; | |
7684 | Dir *dir = dirp->inode->dir; | |
7685 | ||
7686 | if (!dir) { | |
7687 | ldout(cct, 10) << " dir is empty" << dendl; | |
7688 | dirp->set_end(); | |
7689 | return 0; | |
7690 | } | |
7691 | ||
7692 | vector<Dentry*>::iterator pd = std::lower_bound(dir->readdir_cache.begin(), | |
7693 | dir->readdir_cache.end(), | |
7694 | dirp->offset, dentry_off_lt()); | |
7695 | ||
7696 | string dn_name; | |
7697 | while (true) { | |
7698 | if (!dirp->inode->is_complete_and_ordered()) | |
7699 | return -EAGAIN; | |
7700 | if (pd == dir->readdir_cache.end()) | |
7701 | break; | |
7702 | Dentry *dn = *pd; | |
7703 | if (dn->inode == NULL) { | |
7704 | ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl; | |
7705 | ++pd; | |
7706 | continue; | |
7707 | } | |
7708 | if (dn->cap_shared_gen != dir->parent_inode->shared_gen) { | |
7709 | ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl; | |
7710 | ++pd; | |
7711 | continue; | |
7712 | } | |
7713 | ||
7714 | int r = _getattr(dn->inode, caps, dirp->perms); | |
7715 | if (r < 0) | |
7716 | return r; | |
7717 | ||
7718 | struct ceph_statx stx; | |
7719 | struct dirent de; | |
7720 | fill_statx(dn->inode, caps, &stx); | |
7721 | ||
7722 | uint64_t next_off = dn->offset + 1; | |
7723 | ++pd; | |
7724 | if (pd == dir->readdir_cache.end()) | |
7725 | next_off = dir_result_t::END; | |
7726 | ||
7727 | Inode *in = NULL; | |
7728 | fill_dirent(&de, dn->name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
7729 | if (getref) { | |
7730 | in = dn->inode.get(); | |
7731 | _ll_get(in); | |
7732 | } | |
7733 | ||
7734 | dn_name = dn->name; // fill in name while we have lock | |
7735 | ||
7736 | client_lock.Unlock(); | |
7737 | r = cb(p, &de, &stx, next_off, in); // _next_ offset | |
7738 | client_lock.Lock(); | |
7739 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << dn->offset << dec | |
7740 | << " = " << r << dendl; | |
7741 | if (r < 0) { | |
7742 | return r; | |
7743 | } | |
7744 | ||
7745 | dirp->offset = next_off; | |
7746 | if (dirp->at_end()) | |
7747 | dirp->next_offset = 2; | |
7748 | else | |
7749 | dirp->next_offset = dirp->offset_low(); | |
7750 | dirp->last_name = dn_name; // we successfully returned this one; update! | |
7751 | if (r > 0) | |
7752 | return r; | |
7753 | } | |
7754 | ||
7755 | ldout(cct, 10) << "_readdir_cache_cb " << dirp << " on " << dirp->inode->ino << " at end" << dendl; | |
7756 | dirp->set_end(); | |
7757 | return 0; | |
7758 | } | |
7759 | ||
7760 | int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p, | |
7761 | unsigned want, unsigned flags, bool getref) | |
7762 | { | |
7763 | int caps = statx_to_mask(flags, want); | |
7764 | ||
7765 | Mutex::Locker lock(client_lock); | |
7766 | ||
181888fb FG |
7767 | if (unmounting) |
7768 | return -ENOTCONN; | |
7769 | ||
7c673cae FG |
7770 | dir_result_t *dirp = static_cast<dir_result_t*>(d); |
7771 | ||
7772 | ldout(cct, 10) << "readdir_r_cb " << *dirp->inode << " offset " << hex << dirp->offset | |
7773 | << dec << " at_end=" << dirp->at_end() | |
7774 | << " hash_order=" << dirp->hash_order() << dendl; | |
7775 | ||
7776 | struct dirent de; | |
7777 | struct ceph_statx stx; | |
7778 | memset(&de, 0, sizeof(de)); | |
7779 | memset(&stx, 0, sizeof(stx)); | |
7780 | ||
7781 | InodeRef& diri = dirp->inode; | |
7782 | ||
7783 | if (dirp->at_end()) | |
7784 | return 0; | |
7785 | ||
7786 | if (dirp->offset == 0) { | |
7787 | ldout(cct, 15) << " including ." << dendl; | |
7788 | assert(diri->dn_set.size() < 2); // can't have multiple hard-links to a dir | |
7789 | uint64_t next_off = 1; | |
7790 | ||
7791 | int r; | |
7792 | r = _getattr(diri, caps, dirp->perms); | |
7793 | if (r < 0) | |
7794 | return r; | |
7795 | ||
7796 | fill_statx(diri, caps, &stx); | |
7797 | fill_dirent(&de, ".", S_IFDIR, stx.stx_ino, next_off); | |
7798 | ||
7799 | Inode *inode = NULL; | |
7800 | if (getref) { | |
7801 | inode = diri.get(); | |
7802 | _ll_get(inode); | |
7803 | } | |
7804 | ||
7805 | client_lock.Unlock(); | |
7806 | r = cb(p, &de, &stx, next_off, inode); | |
7807 | client_lock.Lock(); | |
7808 | if (r < 0) | |
7809 | return r; | |
7810 | ||
7811 | dirp->offset = next_off; | |
7812 | if (r > 0) | |
7813 | return r; | |
7814 | } | |
7815 | if (dirp->offset == 1) { | |
7816 | ldout(cct, 15) << " including .." << dendl; | |
7817 | uint64_t next_off = 2; | |
7818 | InodeRef in; | |
7819 | if (diri->dn_set.empty()) | |
7820 | in = diri; | |
7821 | else | |
7822 | in = diri->get_first_parent()->inode; | |
7823 | ||
7824 | int r; | |
7825 | r = _getattr(diri, caps, dirp->perms); | |
7826 | if (r < 0) | |
7827 | return r; | |
7828 | ||
7829 | fill_statx(in, caps, &stx); | |
7830 | fill_dirent(&de, "..", S_IFDIR, stx.stx_ino, next_off); | |
7831 | ||
7832 | Inode *inode = NULL; | |
7833 | if (getref) { | |
7834 | inode = in.get(); | |
7835 | _ll_get(inode); | |
7836 | } | |
7837 | ||
7838 | client_lock.Unlock(); | |
7839 | r = cb(p, &de, &stx, next_off, inode); | |
7840 | client_lock.Lock(); | |
7841 | if (r < 0) | |
7842 | return r; | |
7843 | ||
7844 | dirp->offset = next_off; | |
7845 | if (r > 0) | |
7846 | return r; | |
7847 | } | |
7848 | ||
7849 | // can we read from our cache? | |
7850 | ldout(cct, 10) << "offset " << hex << dirp->offset << dec | |
7851 | << " snapid " << dirp->inode->snapid << " (complete && ordered) " | |
7852 | << dirp->inode->is_complete_and_ordered() | |
7853 | << " issued " << ccap_string(dirp->inode->caps_issued()) | |
7854 | << dendl; | |
7855 | if (dirp->inode->snapid != CEPH_SNAPDIR && | |
7856 | dirp->inode->is_complete_and_ordered() && | |
7857 | dirp->inode->caps_issued_mask(CEPH_CAP_FILE_SHARED)) { | |
7858 | int err = _readdir_cache_cb(dirp, cb, p, caps, getref); | |
7859 | if (err != -EAGAIN) | |
7860 | return err; | |
7861 | } | |
7862 | ||
7863 | while (1) { | |
7864 | if (dirp->at_end()) | |
7865 | return 0; | |
7866 | ||
7867 | bool check_caps = true; | |
7868 | if (!dirp->is_cached()) { | |
7869 | int r = _readdir_get_frag(dirp); | |
7870 | if (r) | |
7871 | return r; | |
7872 | // _readdir_get_frag () may updates dirp->offset if the replied dirfrag is | |
7873 | // different than the requested one. (our dirfragtree was outdated) | |
7874 | check_caps = false; | |
7875 | } | |
7876 | frag_t fg = dirp->buffer_frag; | |
7877 | ||
7878 | ldout(cct, 10) << "frag " << fg << " buffer size " << dirp->buffer.size() | |
7879 | << " offset " << hex << dirp->offset << dendl; | |
7880 | ||
7881 | for (auto it = std::lower_bound(dirp->buffer.begin(), dirp->buffer.end(), | |
7882 | dirp->offset, dir_result_t::dentry_off_lt()); | |
7883 | it != dirp->buffer.end(); | |
7884 | ++it) { | |
7885 | dir_result_t::dentry &entry = *it; | |
7886 | ||
7887 | uint64_t next_off = entry.offset + 1; | |
7888 | ||
7889 | int r; | |
7890 | if (check_caps) { | |
7891 | r = _getattr(entry.inode, caps, dirp->perms); | |
7892 | if (r < 0) | |
7893 | return r; | |
7894 | } | |
7895 | ||
7896 | fill_statx(entry.inode, caps, &stx); | |
7897 | fill_dirent(&de, entry.name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
7898 | ||
7899 | Inode *inode = NULL; | |
7900 | if (getref) { | |
7901 | inode = entry.inode.get(); | |
7902 | _ll_get(inode); | |
7903 | } | |
7904 | ||
7905 | client_lock.Unlock(); | |
7906 | r = cb(p, &de, &stx, next_off, inode); // _next_ offset | |
7907 | client_lock.Lock(); | |
7908 | ||
7909 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << next_off - 1 << dec | |
7910 | << " = " << r << dendl; | |
7911 | if (r < 0) | |
7912 | return r; | |
7913 | ||
7914 | dirp->offset = next_off; | |
7915 | if (r > 0) | |
7916 | return r; | |
7917 | } | |
7918 | ||
7919 | if (dirp->next_offset > 2) { | |
7920 | ldout(cct, 10) << " fetching next chunk of this frag" << dendl; | |
7921 | _readdir_drop_dirp_buffer(dirp); | |
7922 | continue; // more! | |
7923 | } | |
7924 | ||
7925 | if (!fg.is_rightmost()) { | |
7926 | // next frag! | |
7927 | _readdir_next_frag(dirp); | |
7928 | continue; | |
7929 | } | |
7930 | ||
7931 | if (diri->shared_gen == dirp->start_shared_gen && | |
7932 | diri->dir_release_count == dirp->release_count) { | |
7933 | if (diri->dir_ordered_count == dirp->ordered_count) { | |
7934 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
7935 | if (diri->dir) { | |
7936 | assert(diri->dir->readdir_cache.size() >= dirp->cache_index); | |
7937 | diri->dir->readdir_cache.resize(dirp->cache_index); | |
7938 | } | |
7939 | diri->flags |= I_COMPLETE | I_DIR_ORDERED; | |
7940 | } else { | |
7941 | ldout(cct, 10) << " marking I_COMPLETE on " << *diri << dendl; | |
7942 | diri->flags |= I_COMPLETE; | |
7943 | } | |
7944 | } | |
7945 | ||
7946 | dirp->set_end(); | |
7947 | return 0; | |
7948 | } | |
7949 | ceph_abort(); | |
7950 | return 0; | |
7951 | } | |
7952 | ||
7953 | ||
7954 | int Client::readdir_r(dir_result_t *d, struct dirent *de) | |
7955 | { | |
7956 | return readdirplus_r(d, de, 0, 0, 0, NULL); | |
7957 | } | |
7958 | ||
7959 | /* | |
7960 | * readdirplus_r | |
7961 | * | |
7962 | * returns | |
7963 | * 1 if we got a dirent | |
7964 | * 0 for end of directory | |
7965 | * <0 on error | |
7966 | */ | |
7967 | ||
7968 | struct single_readdir { | |
7969 | struct dirent *de; | |
7970 | struct ceph_statx *stx; | |
7971 | Inode *inode; | |
7972 | bool full; | |
7973 | }; | |
7974 | ||
7975 | static int _readdir_single_dirent_cb(void *p, struct dirent *de, | |
7976 | struct ceph_statx *stx, off_t off, | |
7977 | Inode *in) | |
7978 | { | |
7979 | single_readdir *c = static_cast<single_readdir *>(p); | |
7980 | ||
7981 | if (c->full) | |
7982 | return -1; // already filled this dirent | |
7983 | ||
7984 | *c->de = *de; | |
7985 | if (c->stx) | |
7986 | *c->stx = *stx; | |
7987 | c->inode = in; | |
7988 | c->full = true; | |
7989 | return 1; | |
7990 | } | |
7991 | ||
7992 | struct dirent *Client::readdir(dir_result_t *d) | |
7993 | { | |
7994 | int ret; | |
7995 | static struct dirent de; | |
7996 | single_readdir sr; | |
7997 | sr.de = &de; | |
7998 | sr.stx = NULL; | |
7999 | sr.inode = NULL; | |
8000 | sr.full = false; | |
8001 | ||
8002 | // our callback fills the dirent and sets sr.full=true on first | |
8003 | // call, and returns -1 the second time around. | |
8004 | ret = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr); | |
8005 | if (ret < -1) { | |
8006 | errno = -ret; // this sucks. | |
8007 | return (dirent *) NULL; | |
8008 | } | |
8009 | if (sr.full) { | |
8010 | return &de; | |
8011 | } | |
8012 | return (dirent *) NULL; | |
8013 | } | |
8014 | ||
8015 | int Client::readdirplus_r(dir_result_t *d, struct dirent *de, | |
8016 | struct ceph_statx *stx, unsigned want, | |
8017 | unsigned flags, Inode **out) | |
8018 | { | |
8019 | single_readdir sr; | |
8020 | sr.de = de; | |
8021 | sr.stx = stx; | |
8022 | sr.inode = NULL; | |
8023 | sr.full = false; | |
8024 | ||
8025 | // our callback fills the dirent and sets sr.full=true on first | |
8026 | // call, and returns -1 the second time around. | |
8027 | int r = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr, want, flags, out); | |
8028 | if (r < -1) | |
8029 | return r; | |
8030 | if (out) | |
8031 | *out = sr.inode; | |
8032 | if (sr.full) | |
8033 | return 1; | |
8034 | return 0; | |
8035 | } | |
8036 | ||
8037 | ||
8038 | /* getdents */ | |
8039 | struct getdents_result { | |
8040 | char *buf; | |
8041 | int buflen; | |
8042 | int pos; | |
8043 | bool fullent; | |
8044 | }; | |
8045 | ||
8046 | static int _readdir_getdent_cb(void *p, struct dirent *de, | |
8047 | struct ceph_statx *stx, off_t off, Inode *in) | |
8048 | { | |
8049 | struct getdents_result *c = static_cast<getdents_result *>(p); | |
8050 | ||
8051 | int dlen; | |
8052 | if (c->fullent) | |
8053 | dlen = sizeof(*de); | |
8054 | else | |
8055 | dlen = strlen(de->d_name) + 1; | |
8056 | ||
8057 | if (c->pos + dlen > c->buflen) | |
8058 | return -1; // doesn't fit | |
8059 | ||
8060 | if (c->fullent) { | |
8061 | memcpy(c->buf + c->pos, de, sizeof(*de)); | |
8062 | } else { | |
8063 | memcpy(c->buf + c->pos, de->d_name, dlen); | |
8064 | } | |
8065 | c->pos += dlen; | |
8066 | return 0; | |
8067 | } | |
8068 | ||
8069 | int Client::_getdents(dir_result_t *dir, char *buf, int buflen, bool fullent) | |
8070 | { | |
8071 | getdents_result gr; | |
8072 | gr.buf = buf; | |
8073 | gr.buflen = buflen; | |
8074 | gr.fullent = fullent; | |
8075 | gr.pos = 0; | |
8076 | ||
8077 | int r = readdir_r_cb(dir, _readdir_getdent_cb, (void *)&gr); | |
8078 | ||
8079 | if (r < 0) { // some error | |
8080 | if (r == -1) { // buffer ran out of space | |
8081 | if (gr.pos) { // but we got some entries already! | |
8082 | return gr.pos; | |
8083 | } // or we need a larger buffer | |
8084 | return -ERANGE; | |
8085 | } else { // actual error, return it | |
8086 | return r; | |
8087 | } | |
8088 | } | |
8089 | return gr.pos; | |
8090 | } | |
8091 | ||
8092 | ||
8093 | /* getdir */ | |
8094 | struct getdir_result { | |
8095 | list<string> *contents; | |
8096 | int num; | |
8097 | }; | |
8098 | ||
8099 | static int _getdir_cb(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in) | |
8100 | { | |
8101 | getdir_result *r = static_cast<getdir_result *>(p); | |
8102 | ||
8103 | r->contents->push_back(de->d_name); | |
8104 | r->num++; | |
8105 | return 0; | |
8106 | } | |
8107 | ||
8108 | int Client::getdir(const char *relpath, list<string>& contents, | |
8109 | const UserPerm& perms) | |
8110 | { | |
8111 | ldout(cct, 3) << "getdir(" << relpath << ")" << dendl; | |
8112 | { | |
8113 | Mutex::Locker lock(client_lock); | |
8114 | tout(cct) << "getdir" << std::endl; | |
8115 | tout(cct) << relpath << std::endl; | |
8116 | } | |
8117 | ||
8118 | dir_result_t *d; | |
8119 | int r = opendir(relpath, &d, perms); | |
8120 | if (r < 0) | |
8121 | return r; | |
8122 | ||
8123 | getdir_result gr; | |
8124 | gr.contents = &contents; | |
8125 | gr.num = 0; | |
8126 | r = readdir_r_cb(d, _getdir_cb, (void *)&gr); | |
8127 | ||
8128 | closedir(d); | |
8129 | ||
8130 | if (r < 0) | |
8131 | return r; | |
8132 | return gr.num; | |
8133 | } | |
8134 | ||
8135 | ||
8136 | /****** file i/o **********/ | |
8137 | int Client::open(const char *relpath, int flags, const UserPerm& perms, | |
8138 | mode_t mode, int stripe_unit, int stripe_count, | |
8139 | int object_size, const char *data_pool) | |
8140 | { | |
8141 | ldout(cct, 3) << "open enter(" << relpath << ", " << ceph_flags_sys2wire(flags) << "," << mode << ")" << dendl; | |
8142 | Mutex::Locker lock(client_lock); | |
8143 | tout(cct) << "open" << std::endl; | |
8144 | tout(cct) << relpath << std::endl; | |
8145 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
8146 | ||
181888fb FG |
8147 | if (unmounting) |
8148 | return -ENOTCONN; | |
8149 | ||
7c673cae FG |
8150 | Fh *fh = NULL; |
8151 | ||
8152 | #if defined(__linux__) && defined(O_PATH) | |
8153 | /* When the O_PATH is being specified, others flags than O_DIRECTORY | |
8154 | * and O_NOFOLLOW are ignored. Please refer do_entry_open() function | |
8155 | * in kernel (fs/open.c). */ | |
8156 | if (flags & O_PATH) | |
8157 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
8158 | #endif | |
8159 | ||
8160 | filepath path(relpath); | |
8161 | InodeRef in; | |
8162 | bool created = false; | |
8163 | /* O_CREATE with O_EXCL enforces O_NOFOLLOW. */ | |
8164 | bool followsym = !((flags & O_NOFOLLOW) || ((flags & O_CREAT) && (flags & O_EXCL))); | |
8165 | int r = path_walk(path, &in, perms, followsym, ceph_caps_for_mode(mode)); | |
8166 | ||
8167 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
8168 | return -EEXIST; | |
8169 | ||
8170 | #if defined(__linux__) && defined(O_PATH) | |
8171 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW) && !(flags & O_PATH)) | |
8172 | #else | |
8173 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW)) | |
8174 | #endif | |
8175 | return -ELOOP; | |
8176 | ||
8177 | if (r == -ENOENT && (flags & O_CREAT)) { | |
8178 | filepath dirpath = path; | |
8179 | string dname = dirpath.last_dentry(); | |
8180 | dirpath.pop_dentry(); | |
8181 | InodeRef dir; | |
8182 | r = path_walk(dirpath, &dir, perms, true, | |
8183 | cct->_conf->client_permissions ? CEPH_CAP_AUTH_SHARED : 0); | |
8184 | if (r < 0) | |
8185 | goto out; | |
8186 | if (cct->_conf->client_permissions) { | |
8187 | r = may_create(dir.get(), perms); | |
8188 | if (r < 0) | |
8189 | goto out; | |
8190 | } | |
8191 | r = _create(dir.get(), dname.c_str(), flags, mode, &in, &fh, stripe_unit, | |
8192 | stripe_count, object_size, data_pool, &created, perms); | |
8193 | } | |
8194 | if (r < 0) | |
8195 | goto out; | |
8196 | ||
8197 | if (!created) { | |
8198 | // posix says we can only check permissions of existing files | |
8199 | if (cct->_conf->client_permissions) { | |
8200 | r = may_open(in.get(), flags, perms); | |
8201 | if (r < 0) | |
8202 | goto out; | |
8203 | } | |
8204 | } | |
8205 | ||
8206 | if (!fh) | |
8207 | r = _open(in.get(), flags, mode, &fh, perms); | |
8208 | if (r >= 0) { | |
8209 | // allocate a integer file descriptor | |
8210 | assert(fh); | |
8211 | r = get_fd(); | |
8212 | assert(fd_map.count(r) == 0); | |
8213 | fd_map[r] = fh; | |
8214 | } | |
8215 | ||
8216 | out: | |
8217 | tout(cct) << r << std::endl; | |
8218 | ldout(cct, 3) << "open exit(" << path << ", " << ceph_flags_sys2wire(flags) << ") = " << r << dendl; | |
8219 | return r; | |
8220 | } | |
8221 | ||
8222 | int Client::open(const char *relpath, int flags, const UserPerm& perms, mode_t mode) | |
8223 | { | |
8224 | /* Use default file striping parameters */ | |
8225 | return open(relpath, flags, perms, mode, 0, 0, 0, NULL); | |
8226 | } | |
8227 | ||
8228 | int Client::lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, | |
8229 | const UserPerm& perms) | |
8230 | { | |
8231 | Mutex::Locker lock(client_lock); | |
8232 | ldout(cct, 3) << "lookup_hash enter(" << ino << ", #" << dirino << "/" << name << ")" << dendl; | |
8233 | ||
181888fb FG |
8234 | if (unmounting) |
8235 | return -ENOTCONN; | |
8236 | ||
7c673cae FG |
8237 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPHASH); |
8238 | filepath path(ino); | |
8239 | req->set_filepath(path); | |
8240 | ||
8241 | uint32_t h = ceph_str_hash(CEPH_STR_HASH_RJENKINS, name, strlen(name)); | |
8242 | char f[30]; | |
8243 | sprintf(f, "%u", h); | |
8244 | filepath path2(dirino); | |
8245 | path2.push_dentry(string(f)); | |
8246 | req->set_filepath2(path2); | |
8247 | ||
8248 | int r = make_request(req, perms, NULL, NULL, | |
8249 | rand() % mdsmap->get_num_in_mds()); | |
8250 | ldout(cct, 3) << "lookup_hash exit(" << ino << ", #" << dirino << "/" << name << ") = " << r << dendl; | |
8251 | return r; | |
8252 | } | |
8253 | ||
8254 | ||
8255 | /** | |
8256 | * Load inode into local cache. | |
8257 | * | |
8258 | * If inode pointer is non-NULL, and take a reference on | |
8259 | * the resulting Inode object in one operation, so that caller | |
8260 | * can safely assume inode will still be there after return. | |
8261 | */ | |
8262 | int Client::lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) | |
8263 | { | |
8264 | Mutex::Locker lock(client_lock); | |
8265 | ldout(cct, 3) << "lookup_ino enter(" << ino << ")" << dendl; | |
8266 | ||
181888fb FG |
8267 | if (unmounting) |
8268 | return -ENOTCONN; | |
8269 | ||
7c673cae FG |
8270 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPINO); |
8271 | filepath path(ino); | |
8272 | req->set_filepath(path); | |
8273 | ||
8274 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
8275 | if (r == 0 && inode != NULL) { | |
8276 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
8277 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
8278 | assert(p != inode_map.end()); | |
8279 | *inode = p->second; | |
8280 | _ll_get(*inode); | |
8281 | } | |
8282 | ldout(cct, 3) << "lookup_ino exit(" << ino << ") = " << r << dendl; | |
8283 | return r; | |
8284 | } | |
8285 | ||
8286 | ||
8287 | ||
8288 | /** | |
8289 | * Find the parent inode of `ino` and insert it into | |
8290 | * our cache. Conditionally also set `parent` to a referenced | |
8291 | * Inode* if caller provides non-NULL value. | |
8292 | */ | |
8293 | int Client::lookup_parent(Inode *ino, const UserPerm& perms, Inode **parent) | |
8294 | { | |
8295 | Mutex::Locker lock(client_lock); | |
8296 | ldout(cct, 3) << "lookup_parent enter(" << ino->ino << ")" << dendl; | |
8297 | ||
181888fb FG |
8298 | if (unmounting) |
8299 | return -ENOTCONN; | |
8300 | ||
7c673cae FG |
8301 | if (!ino->dn_set.empty()) { |
8302 | // if we exposed the parent here, we'd need to check permissions, | |
8303 | // but right now we just rely on the MDS doing so in make_request | |
8304 | ldout(cct, 3) << "lookup_parent dentry already present" << dendl; | |
8305 | return 0; | |
8306 | } | |
8307 | ||
8308 | if (ino->is_root()) { | |
8309 | *parent = NULL; | |
8310 | ldout(cct, 3) << "ino is root, no parent" << dendl; | |
8311 | return -EINVAL; | |
8312 | } | |
8313 | ||
8314 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); | |
8315 | filepath path(ino->ino); | |
8316 | req->set_filepath(path); | |
8317 | ||
8318 | InodeRef target; | |
8319 | int r = make_request(req, perms, &target, NULL, rand() % mdsmap->get_num_in_mds()); | |
8320 | // Give caller a reference to the parent ino if they provided a pointer. | |
8321 | if (parent != NULL) { | |
8322 | if (r == 0) { | |
8323 | *parent = target.get(); | |
8324 | _ll_get(*parent); | |
8325 | ldout(cct, 3) << "lookup_parent found parent " << (*parent)->ino << dendl; | |
8326 | } else { | |
8327 | *parent = NULL; | |
8328 | } | |
8329 | } | |
8330 | ldout(cct, 3) << "lookup_parent exit(" << ino->ino << ") = " << r << dendl; | |
8331 | return r; | |
8332 | } | |
8333 | ||
8334 | ||
8335 | /** | |
8336 | * Populate the parent dentry for `ino`, provided it is | |
8337 | * a child of `parent`. | |
8338 | */ | |
8339 | int Client::lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) | |
8340 | { | |
8341 | assert(parent->is_dir()); | |
8342 | ||
8343 | Mutex::Locker lock(client_lock); | |
8344 | ldout(cct, 3) << "lookup_name enter(" << ino->ino << ")" << dendl; | |
8345 | ||
181888fb FG |
8346 | if (unmounting) |
8347 | return -ENOTCONN; | |
8348 | ||
7c673cae FG |
8349 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
8350 | req->set_filepath2(filepath(parent->ino)); | |
8351 | req->set_filepath(filepath(ino->ino)); | |
8352 | req->set_inode(ino); | |
8353 | ||
8354 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
8355 | ldout(cct, 3) << "lookup_name exit(" << ino->ino << ") = " << r << dendl; | |
8356 | return r; | |
8357 | } | |
8358 | ||
8359 | ||
8360 | Fh *Client::_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms) | |
8361 | { | |
8362 | assert(in); | |
8363 | Fh *f = new Fh(in); | |
8364 | f->mode = cmode; | |
8365 | f->flags = flags; | |
8366 | ||
8367 | // inode | |
8368 | f->actor_perms = perms; | |
8369 | ||
8370 | ldout(cct, 10) << "_create_fh " << in->ino << " mode " << cmode << dendl; | |
8371 | ||
8372 | if (in->snapid != CEPH_NOSNAP) { | |
8373 | in->snap_cap_refs++; | |
8374 | ldout(cct, 5) << "open success, fh is " << f << " combined IMMUTABLE SNAP caps " | |
8375 | << ccap_string(in->caps_issued()) << dendl; | |
8376 | } | |
8377 | ||
8378 | const md_config_t *conf = cct->_conf; | |
8379 | f->readahead.set_trigger_requests(1); | |
8380 | f->readahead.set_min_readahead_size(conf->client_readahead_min); | |
8381 | uint64_t max_readahead = Readahead::NO_LIMIT; | |
8382 | if (conf->client_readahead_max_bytes) { | |
8383 | max_readahead = MIN(max_readahead, (uint64_t)conf->client_readahead_max_bytes); | |
8384 | } | |
8385 | if (conf->client_readahead_max_periods) { | |
8386 | max_readahead = MIN(max_readahead, in->layout.get_period()*(uint64_t)conf->client_readahead_max_periods); | |
8387 | } | |
8388 | f->readahead.set_max_readahead_size(max_readahead); | |
8389 | vector<uint64_t> alignments; | |
8390 | alignments.push_back(in->layout.get_period()); | |
8391 | alignments.push_back(in->layout.stripe_unit); | |
8392 | f->readahead.set_alignments(alignments); | |
8393 | ||
8394 | return f; | |
8395 | } | |
8396 | ||
8397 | int Client::_release_fh(Fh *f) | |
8398 | { | |
8399 | //ldout(cct, 3) << "op: client->close(open_files[ " << fh << " ]);" << dendl; | |
8400 | //ldout(cct, 3) << "op: open_files.erase( " << fh << " );" << dendl; | |
8401 | Inode *in = f->inode.get(); | |
8402 | ldout(cct, 5) << "_release_fh " << f << " mode " << f->mode << " on " << *in << dendl; | |
8403 | ||
b32b8144 FG |
8404 | in->unset_deleg(f); |
8405 | ||
7c673cae FG |
8406 | if (in->snapid == CEPH_NOSNAP) { |
8407 | if (in->put_open_ref(f->mode)) { | |
8408 | _flush(in, new C_Client_FlushComplete(this, in)); | |
8409 | check_caps(in, 0); | |
8410 | } | |
8411 | } else { | |
8412 | assert(in->snap_cap_refs > 0); | |
8413 | in->snap_cap_refs--; | |
8414 | } | |
8415 | ||
8416 | _release_filelocks(f); | |
8417 | ||
8418 | // Finally, read any async err (i.e. from flushes) | |
8419 | int err = f->take_async_err(); | |
8420 | if (err != 0) { | |
8421 | ldout(cct, 1) << "_release_fh " << f << " on inode " << *in << " caught async_err = " | |
8422 | << cpp_strerror(err) << dendl; | |
8423 | } else { | |
8424 | ldout(cct, 10) << "_release_fh " << f << " on inode " << *in << " no async_err state" << dendl; | |
8425 | } | |
8426 | ||
8427 | _put_fh(f); | |
8428 | ||
8429 | return err; | |
8430 | } | |
8431 | ||
8432 | void Client::_put_fh(Fh *f) | |
8433 | { | |
8434 | int left = f->put(); | |
8435 | if (!left) { | |
8436 | delete f; | |
8437 | } | |
8438 | } | |
8439 | ||
8440 | int Client::_open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
8441 | const UserPerm& perms) | |
8442 | { | |
8443 | if (in->snapid != CEPH_NOSNAP && | |
8444 | (flags & (O_WRONLY | O_RDWR | O_CREAT | O_TRUNC | O_APPEND))) { | |
8445 | return -EROFS; | |
8446 | } | |
8447 | ||
8448 | // use normalized flags to generate cmode | |
8449 | int cmode = ceph_flags_to_mode(ceph_flags_sys2wire(flags)); | |
8450 | if (cmode < 0) | |
8451 | return -EINVAL; | |
8452 | int want = ceph_caps_for_mode(cmode); | |
8453 | int result = 0; | |
8454 | ||
8455 | in->get_open_ref(cmode); // make note of pending open, since it effects _wanted_ caps. | |
8456 | ||
b32b8144 | 8457 | if ((flags & O_TRUNC) == 0 && in->caps_issued_mask(want)) { |
7c673cae FG |
8458 | // update wanted? |
8459 | check_caps(in, CHECK_CAPS_NODELAY); | |
8460 | } else { | |
b32b8144 | 8461 | |
7c673cae FG |
8462 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); |
8463 | filepath path; | |
8464 | in->make_nosnap_relative_path(path); | |
8465 | req->set_filepath(path); | |
8466 | req->head.args.open.flags = ceph_flags_sys2wire(flags & ~O_CREAT); | |
8467 | req->head.args.open.mode = mode; | |
8468 | req->head.args.open.pool = -1; | |
8469 | if (cct->_conf->client_debug_getattr_caps) | |
8470 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
8471 | else | |
8472 | req->head.args.open.mask = 0; | |
8473 | req->head.args.open.old_size = in->size; // for O_TRUNC | |
8474 | req->set_inode(in); | |
8475 | result = make_request(req, perms); | |
b32b8144 FG |
8476 | |
8477 | /* | |
8478 | * NFS expects that delegations will be broken on a conflicting open, | |
8479 | * not just when there is actual conflicting access to the file. SMB leases | |
8480 | * and oplocks also have similar semantics. | |
8481 | * | |
8482 | * Ensure that clients that have delegations enabled will wait on minimal | |
8483 | * caps during open, just to ensure that other clients holding delegations | |
8484 | * return theirs first. | |
8485 | */ | |
8486 | if (deleg_timeout && result == 0) { | |
8487 | int need = 0, have; | |
8488 | ||
8489 | if (cmode & CEPH_FILE_MODE_WR) | |
8490 | need |= CEPH_CAP_FILE_WR; | |
8491 | if (cmode & CEPH_FILE_MODE_RD) | |
8492 | need |= CEPH_CAP_FILE_RD; | |
8493 | ||
8494 | result = get_caps(in, need, want, &have, -1); | |
8495 | if (result < 0) { | |
8496 | ldout(cct, 1) << "Unable to get caps after open of inode " << *in << | |
8497 | " . Denying open: " << | |
8498 | cpp_strerror(result) << dendl; | |
8499 | in->put_open_ref(cmode); | |
8500 | } else { | |
8501 | put_cap_ref(in, need); | |
8502 | } | |
8503 | } | |
7c673cae FG |
8504 | } |
8505 | ||
8506 | // success? | |
8507 | if (result >= 0) { | |
8508 | if (fhp) | |
8509 | *fhp = _create_fh(in, flags, cmode, perms); | |
8510 | } else { | |
8511 | in->put_open_ref(cmode); | |
8512 | } | |
8513 | ||
8514 | trim_cache(); | |
8515 | ||
8516 | return result; | |
8517 | } | |
8518 | ||
8519 | int Client::_renew_caps(Inode *in) | |
8520 | { | |
8521 | int wanted = in->caps_file_wanted(); | |
8522 | if (in->is_any_caps() && | |
8523 | ((wanted & CEPH_CAP_ANY_WR) == 0 || in->auth_cap)) { | |
8524 | check_caps(in, CHECK_CAPS_NODELAY); | |
8525 | return 0; | |
8526 | } | |
8527 | ||
8528 | int flags = 0; | |
8529 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
8530 | flags = O_RDWR; | |
8531 | else if (wanted & CEPH_CAP_FILE_RD) | |
8532 | flags = O_RDONLY; | |
8533 | else if (wanted & CEPH_CAP_FILE_WR) | |
8534 | flags = O_WRONLY; | |
8535 | ||
8536 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); | |
8537 | filepath path; | |
8538 | in->make_nosnap_relative_path(path); | |
8539 | req->set_filepath(path); | |
8540 | req->head.args.open.flags = flags; | |
8541 | req->head.args.open.pool = -1; | |
8542 | if (cct->_conf->client_debug_getattr_caps) | |
8543 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
8544 | else | |
8545 | req->head.args.open.mask = 0; | |
8546 | req->set_inode(in); | |
8547 | ||
8548 | // duplicate in case Cap goes away; not sure if that race is a concern? | |
8549 | const UserPerm *pperm = in->get_best_perms(); | |
8550 | UserPerm perms; | |
8551 | if (pperm != NULL) | |
8552 | perms = *pperm; | |
8553 | int ret = make_request(req, perms); | |
8554 | return ret; | |
8555 | } | |
8556 | ||
8557 | int Client::close(int fd) | |
8558 | { | |
8559 | ldout(cct, 3) << "close enter(" << fd << ")" << dendl; | |
8560 | Mutex::Locker lock(client_lock); | |
8561 | tout(cct) << "close" << std::endl; | |
8562 | tout(cct) << fd << std::endl; | |
8563 | ||
181888fb FG |
8564 | if (unmounting) |
8565 | return -ENOTCONN; | |
8566 | ||
7c673cae FG |
8567 | Fh *fh = get_filehandle(fd); |
8568 | if (!fh) | |
8569 | return -EBADF; | |
8570 | int err = _release_fh(fh); | |
8571 | fd_map.erase(fd); | |
8572 | put_fd(fd); | |
8573 | ldout(cct, 3) << "close exit(" << fd << ")" << dendl; | |
8574 | return err; | |
8575 | } | |
8576 | ||
8577 | ||
8578 | // ------------ | |
8579 | // read, write | |
8580 | ||
8581 | loff_t Client::lseek(int fd, loff_t offset, int whence) | |
8582 | { | |
8583 | Mutex::Locker lock(client_lock); | |
8584 | tout(cct) << "lseek" << std::endl; | |
8585 | tout(cct) << fd << std::endl; | |
8586 | tout(cct) << offset << std::endl; | |
8587 | tout(cct) << whence << std::endl; | |
8588 | ||
181888fb FG |
8589 | if (unmounting) |
8590 | return -ENOTCONN; | |
8591 | ||
7c673cae FG |
8592 | Fh *f = get_filehandle(fd); |
8593 | if (!f) | |
8594 | return -EBADF; | |
8595 | #if defined(__linux__) && defined(O_PATH) | |
8596 | if (f->flags & O_PATH) | |
8597 | return -EBADF; | |
8598 | #endif | |
8599 | return _lseek(f, offset, whence); | |
8600 | } | |
8601 | ||
8602 | loff_t Client::_lseek(Fh *f, loff_t offset, int whence) | |
8603 | { | |
8604 | Inode *in = f->inode.get(); | |
8605 | int r; | |
8606 | ||
8607 | switch (whence) { | |
8608 | case SEEK_SET: | |
8609 | f->pos = offset; | |
8610 | break; | |
8611 | ||
8612 | case SEEK_CUR: | |
8613 | f->pos += offset; | |
8614 | break; | |
8615 | ||
8616 | case SEEK_END: | |
8617 | r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
8618 | if (r < 0) | |
8619 | return r; | |
8620 | f->pos = in->size + offset; | |
8621 | break; | |
8622 | ||
8623 | default: | |
8624 | ceph_abort(); | |
8625 | } | |
8626 | ||
8627 | ldout(cct, 3) << "_lseek(" << f << ", " << offset << ", " << whence << ") = " << f->pos << dendl; | |
8628 | return f->pos; | |
8629 | } | |
8630 | ||
8631 | ||
8632 | void Client::lock_fh_pos(Fh *f) | |
8633 | { | |
8634 | ldout(cct, 10) << "lock_fh_pos " << f << dendl; | |
8635 | ||
8636 | if (f->pos_locked || !f->pos_waiters.empty()) { | |
8637 | Cond cond; | |
8638 | f->pos_waiters.push_back(&cond); | |
8639 | ldout(cct, 10) << "lock_fh_pos BLOCKING on " << f << dendl; | |
8640 | while (f->pos_locked || f->pos_waiters.front() != &cond) | |
8641 | cond.Wait(client_lock); | |
8642 | ldout(cct, 10) << "lock_fh_pos UNBLOCKING on " << f << dendl; | |
8643 | assert(f->pos_waiters.front() == &cond); | |
8644 | f->pos_waiters.pop_front(); | |
8645 | } | |
8646 | ||
8647 | f->pos_locked = true; | |
8648 | } | |
8649 | ||
8650 | void Client::unlock_fh_pos(Fh *f) | |
8651 | { | |
8652 | ldout(cct, 10) << "unlock_fh_pos " << f << dendl; | |
8653 | f->pos_locked = false; | |
8654 | } | |
8655 | ||
8656 | int Client::uninline_data(Inode *in, Context *onfinish) | |
8657 | { | |
8658 | if (!in->inline_data.length()) { | |
8659 | onfinish->complete(0); | |
8660 | return 0; | |
8661 | } | |
8662 | ||
8663 | char oid_buf[32]; | |
8664 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (long long unsigned)in->ino); | |
8665 | object_t oid = oid_buf; | |
8666 | ||
8667 | ObjectOperation create_ops; | |
8668 | create_ops.create(false); | |
8669 | ||
8670 | objecter->mutate(oid, | |
8671 | OSDMap::file_to_object_locator(in->layout), | |
8672 | create_ops, | |
8673 | in->snaprealm->get_snap_context(), | |
8674 | ceph::real_clock::now(), | |
8675 | 0, | |
8676 | NULL); | |
8677 | ||
8678 | bufferlist inline_version_bl; | |
8679 | ::encode(in->inline_version, inline_version_bl); | |
8680 | ||
8681 | ObjectOperation uninline_ops; | |
8682 | uninline_ops.cmpxattr("inline_version", | |
8683 | CEPH_OSD_CMPXATTR_OP_GT, | |
8684 | CEPH_OSD_CMPXATTR_MODE_U64, | |
8685 | inline_version_bl); | |
8686 | bufferlist inline_data = in->inline_data; | |
8687 | uninline_ops.write(0, inline_data, in->truncate_size, in->truncate_seq); | |
8688 | uninline_ops.setxattr("inline_version", stringify(in->inline_version)); | |
8689 | ||
8690 | objecter->mutate(oid, | |
8691 | OSDMap::file_to_object_locator(in->layout), | |
8692 | uninline_ops, | |
8693 | in->snaprealm->get_snap_context(), | |
8694 | ceph::real_clock::now(), | |
8695 | 0, | |
8696 | onfinish); | |
8697 | ||
8698 | return 0; | |
8699 | } | |
8700 | ||
8701 | // | |
8702 | ||
8703 | // blocking osd interface | |
8704 | ||
8705 | int Client::read(int fd, char *buf, loff_t size, loff_t offset) | |
8706 | { | |
8707 | Mutex::Locker lock(client_lock); | |
8708 | tout(cct) << "read" << std::endl; | |
8709 | tout(cct) << fd << std::endl; | |
8710 | tout(cct) << size << std::endl; | |
8711 | tout(cct) << offset << std::endl; | |
8712 | ||
181888fb FG |
8713 | if (unmounting) |
8714 | return -ENOTCONN; | |
8715 | ||
7c673cae FG |
8716 | Fh *f = get_filehandle(fd); |
8717 | if (!f) | |
8718 | return -EBADF; | |
8719 | #if defined(__linux__) && defined(O_PATH) | |
8720 | if (f->flags & O_PATH) | |
8721 | return -EBADF; | |
8722 | #endif | |
8723 | bufferlist bl; | |
8724 | int r = _read(f, offset, size, &bl); | |
8725 | ldout(cct, 3) << "read(" << fd << ", " << (void*)buf << ", " << size << ", " << offset << ") = " << r << dendl; | |
8726 | if (r >= 0) { | |
8727 | bl.copy(0, bl.length(), buf); | |
8728 | r = bl.length(); | |
8729 | } | |
8730 | return r; | |
8731 | } | |
8732 | ||
8733 | int Client::preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset) | |
8734 | { | |
8735 | if (iovcnt < 0) | |
8736 | return -EINVAL; | |
8737 | return _preadv_pwritev(fd, iov, iovcnt, offset, false); | |
8738 | } | |
8739 | ||
8740 | int Client::_read(Fh *f, int64_t offset, uint64_t size, bufferlist *bl) | |
8741 | { | |
8742 | const md_config_t *conf = cct->_conf; | |
8743 | Inode *in = f->inode.get(); | |
8744 | ||
8745 | if ((f->mode & CEPH_FILE_MODE_RD) == 0) | |
8746 | return -EBADF; | |
8747 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; | |
8748 | ||
8749 | bool movepos = false; | |
8750 | if (offset < 0) { | |
8751 | lock_fh_pos(f); | |
8752 | offset = f->pos; | |
8753 | movepos = true; | |
8754 | } | |
8755 | loff_t start_pos = offset; | |
8756 | ||
8757 | if (in->inline_version == 0) { | |
8758 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
c07f9fc5 FG |
8759 | if (r < 0) { |
8760 | if (movepos) | |
8761 | unlock_fh_pos(f); | |
7c673cae | 8762 | return r; |
c07f9fc5 | 8763 | } |
7c673cae FG |
8764 | assert(in->inline_version > 0); |
8765 | } | |
8766 | ||
8767 | retry: | |
8768 | int have; | |
8769 | int r = get_caps(in, CEPH_CAP_FILE_RD, CEPH_CAP_FILE_CACHE, &have, -1); | |
c07f9fc5 FG |
8770 | if (r < 0) { |
8771 | if (movepos) | |
8772 | unlock_fh_pos(f); | |
7c673cae | 8773 | return r; |
c07f9fc5 | 8774 | } |
7c673cae FG |
8775 | if (f->flags & O_DIRECT) |
8776 | have &= ~CEPH_CAP_FILE_CACHE; | |
8777 | ||
8778 | Mutex uninline_flock("Client::_read_uninline_data flock"); | |
8779 | Cond uninline_cond; | |
8780 | bool uninline_done = false; | |
8781 | int uninline_ret = 0; | |
8782 | Context *onuninline = NULL; | |
8783 | ||
8784 | if (in->inline_version < CEPH_INLINE_NONE) { | |
8785 | if (!(have & CEPH_CAP_FILE_CACHE)) { | |
8786 | onuninline = new C_SafeCond(&uninline_flock, | |
8787 | &uninline_cond, | |
8788 | &uninline_done, | |
8789 | &uninline_ret); | |
8790 | uninline_data(in, onuninline); | |
8791 | } else { | |
8792 | uint32_t len = in->inline_data.length(); | |
8793 | ||
8794 | uint64_t endoff = offset + size; | |
8795 | if (endoff > in->size) | |
8796 | endoff = in->size; | |
8797 | ||
8798 | if (offset < len) { | |
8799 | if (endoff <= len) { | |
8800 | bl->substr_of(in->inline_data, offset, endoff - offset); | |
8801 | } else { | |
8802 | bl->substr_of(in->inline_data, offset, len - offset); | |
8803 | bl->append_zero(endoff - len); | |
8804 | } | |
8805 | } else if ((uint64_t)offset < endoff) { | |
8806 | bl->append_zero(endoff - offset); | |
8807 | } | |
8808 | ||
8809 | goto success; | |
8810 | } | |
8811 | } | |
8812 | ||
8813 | if (!conf->client_debug_force_sync_read && | |
8814 | (conf->client_oc && (have & CEPH_CAP_FILE_CACHE))) { | |
8815 | ||
8816 | if (f->flags & O_RSYNC) { | |
8817 | _flush_range(in, offset, size); | |
8818 | } | |
8819 | r = _read_async(f, offset, size, bl); | |
8820 | if (r < 0) | |
8821 | goto done; | |
8822 | } else { | |
8823 | if (f->flags & O_DIRECT) | |
8824 | _flush_range(in, offset, size); | |
8825 | ||
8826 | bool checkeof = false; | |
8827 | r = _read_sync(f, offset, size, bl, &checkeof); | |
8828 | if (r < 0) | |
8829 | goto done; | |
8830 | if (checkeof) { | |
8831 | offset += r; | |
8832 | size -= r; | |
8833 | ||
8834 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
8835 | have = 0; | |
8836 | // reverify size | |
8837 | r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
8838 | if (r < 0) | |
8839 | goto done; | |
8840 | ||
8841 | // eof? short read. | |
8842 | if ((uint64_t)offset < in->size) | |
8843 | goto retry; | |
8844 | } | |
8845 | } | |
8846 | ||
8847 | success: | |
8848 | if (movepos) { | |
8849 | // adjust fd pos | |
8850 | f->pos = start_pos + bl->length(); | |
8851 | unlock_fh_pos(f); | |
8852 | } | |
8853 | ||
8854 | done: | |
8855 | // done! | |
8856 | ||
8857 | if (onuninline) { | |
8858 | client_lock.Unlock(); | |
8859 | uninline_flock.Lock(); | |
8860 | while (!uninline_done) | |
8861 | uninline_cond.Wait(uninline_flock); | |
8862 | uninline_flock.Unlock(); | |
8863 | client_lock.Lock(); | |
8864 | ||
8865 | if (uninline_ret >= 0 || uninline_ret == -ECANCELED) { | |
8866 | in->inline_data.clear(); | |
8867 | in->inline_version = CEPH_INLINE_NONE; | |
8868 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
8869 | check_caps(in, 0); | |
8870 | } else | |
8871 | r = uninline_ret; | |
8872 | } | |
8873 | ||
8874 | if (have) | |
8875 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
c07f9fc5 FG |
8876 | if (r < 0) { |
8877 | if (movepos) | |
8878 | unlock_fh_pos(f); | |
8879 | return r; | |
8880 | } else | |
8881 | return bl->length(); | |
7c673cae FG |
8882 | } |
8883 | ||
8884 | Client::C_Readahead::C_Readahead(Client *c, Fh *f) : | |
8885 | client(c), f(f) { | |
8886 | f->get(); | |
8887 | f->readahead.inc_pending(); | |
8888 | } | |
8889 | ||
8890 | Client::C_Readahead::~C_Readahead() { | |
8891 | f->readahead.dec_pending(); | |
8892 | client->_put_fh(f); | |
8893 | } | |
8894 | ||
8895 | void Client::C_Readahead::finish(int r) { | |
8896 | lgeneric_subdout(client->cct, client, 20) << "client." << client->get_nodeid() << " " << "C_Readahead on " << f->inode << dendl; | |
8897 | client->put_cap_ref(f->inode.get(), CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
8898 | } | |
8899 | ||
8900 | int Client::_read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl) | |
8901 | { | |
8902 | const md_config_t *conf = cct->_conf; | |
8903 | Inode *in = f->inode.get(); | |
8904 | ||
8905 | ldout(cct, 10) << "_read_async " << *in << " " << off << "~" << len << dendl; | |
8906 | ||
8907 | // trim read based on file size? | |
8908 | if (off >= in->size) | |
8909 | return 0; | |
8910 | if (len == 0) | |
8911 | return 0; | |
8912 | if (off + len > in->size) { | |
8913 | len = in->size - off; | |
8914 | } | |
8915 | ||
8916 | ldout(cct, 10) << " min_bytes=" << f->readahead.get_min_readahead_size() | |
8917 | << " max_bytes=" << f->readahead.get_max_readahead_size() | |
8918 | << " max_periods=" << conf->client_readahead_max_periods << dendl; | |
8919 | ||
8920 | // read (and possibly block) | |
8921 | int r, rvalue = 0; | |
8922 | Mutex flock("Client::_read_async flock"); | |
8923 | Cond cond; | |
8924 | bool done = false; | |
8925 | Context *onfinish = new C_SafeCond(&flock, &cond, &done, &rvalue); | |
8926 | r = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
8927 | off, len, bl, 0, onfinish); | |
8928 | if (r == 0) { | |
8929 | get_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
8930 | client_lock.Unlock(); | |
8931 | flock.Lock(); | |
8932 | while (!done) | |
8933 | cond.Wait(flock); | |
8934 | flock.Unlock(); | |
8935 | client_lock.Lock(); | |
8936 | put_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
8937 | r = rvalue; | |
8938 | } else { | |
8939 | // it was cached. | |
8940 | delete onfinish; | |
8941 | } | |
8942 | ||
8943 | if(f->readahead.get_min_readahead_size() > 0) { | |
8944 | pair<uint64_t, uint64_t> readahead_extent = f->readahead.update(off, len, in->size); | |
8945 | if (readahead_extent.second > 0) { | |
8946 | ldout(cct, 20) << "readahead " << readahead_extent.first << "~" << readahead_extent.second | |
8947 | << " (caller wants " << off << "~" << len << ")" << dendl; | |
8948 | Context *onfinish2 = new C_Readahead(this, f); | |
8949 | int r2 = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
8950 | readahead_extent.first, readahead_extent.second, | |
8951 | NULL, 0, onfinish2); | |
8952 | if (r2 == 0) { | |
8953 | ldout(cct, 20) << "readahead initiated, c " << onfinish2 << dendl; | |
8954 | get_cap_ref(in, CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
8955 | } else { | |
8956 | ldout(cct, 20) << "readahead was no-op, already cached" << dendl; | |
8957 | delete onfinish2; | |
8958 | } | |
8959 | } | |
8960 | } | |
8961 | ||
8962 | return r; | |
8963 | } | |
8964 | ||
8965 | int Client::_read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, | |
8966 | bool *checkeof) | |
8967 | { | |
8968 | Inode *in = f->inode.get(); | |
8969 | uint64_t pos = off; | |
8970 | int left = len; | |
8971 | int read = 0; | |
8972 | ||
8973 | ldout(cct, 10) << "_read_sync " << *in << " " << off << "~" << len << dendl; | |
8974 | ||
8975 | Mutex flock("Client::_read_sync flock"); | |
8976 | Cond cond; | |
8977 | while (left > 0) { | |
8978 | int r = 0; | |
8979 | bool done = false; | |
8980 | Context *onfinish = new C_SafeCond(&flock, &cond, &done, &r); | |
8981 | bufferlist tbl; | |
8982 | ||
8983 | int wanted = left; | |
8984 | filer->read_trunc(in->ino, &in->layout, in->snapid, | |
8985 | pos, left, &tbl, 0, | |
8986 | in->truncate_size, in->truncate_seq, | |
8987 | onfinish); | |
8988 | client_lock.Unlock(); | |
8989 | flock.Lock(); | |
8990 | while (!done) | |
8991 | cond.Wait(flock); | |
8992 | flock.Unlock(); | |
8993 | client_lock.Lock(); | |
8994 | ||
8995 | // if we get ENOENT from OSD, assume 0 bytes returned | |
8996 | if (r == -ENOENT) | |
8997 | r = 0; | |
8998 | if (r < 0) | |
8999 | return r; | |
9000 | if (tbl.length()) { | |
9001 | r = tbl.length(); | |
9002 | ||
9003 | read += r; | |
9004 | pos += r; | |
9005 | left -= r; | |
9006 | bl->claim_append(tbl); | |
9007 | } | |
9008 | // short read? | |
9009 | if (r >= 0 && r < wanted) { | |
9010 | if (pos < in->size) { | |
9011 | // zero up to known EOF | |
9012 | int64_t some = in->size - pos; | |
9013 | if (some > left) | |
9014 | some = left; | |
9015 | bufferptr z(some); | |
9016 | z.zero(); | |
9017 | bl->push_back(z); | |
9018 | read += some; | |
9019 | pos += some; | |
9020 | left -= some; | |
9021 | if (left == 0) | |
9022 | return read; | |
9023 | } | |
9024 | ||
9025 | *checkeof = true; | |
9026 | return read; | |
9027 | } | |
9028 | } | |
9029 | return read; | |
9030 | } | |
9031 | ||
9032 | ||
9033 | /* | |
9034 | * we keep count of uncommitted sync writes on the inode, so that | |
9035 | * fsync can DDRT. | |
9036 | */ | |
9037 | void Client::_sync_write_commit(Inode *in) | |
9038 | { | |
9039 | assert(unsafe_sync_write > 0); | |
9040 | unsafe_sync_write--; | |
9041 | ||
9042 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9043 | ||
9044 | ldout(cct, 15) << "sync_write_commit unsafe_sync_write = " << unsafe_sync_write << dendl; | |
9045 | if (unsafe_sync_write == 0 && unmounting) { | |
9046 | ldout(cct, 10) << "sync_write_commit -- no more unsafe writes, unmount can proceed" << dendl; | |
9047 | mount_cond.Signal(); | |
9048 | } | |
9049 | } | |
9050 | ||
9051 | int Client::write(int fd, const char *buf, loff_t size, loff_t offset) | |
9052 | { | |
9053 | Mutex::Locker lock(client_lock); | |
9054 | tout(cct) << "write" << std::endl; | |
9055 | tout(cct) << fd << std::endl; | |
9056 | tout(cct) << size << std::endl; | |
9057 | tout(cct) << offset << std::endl; | |
9058 | ||
181888fb FG |
9059 | if (unmounting) |
9060 | return -ENOTCONN; | |
9061 | ||
7c673cae FG |
9062 | Fh *fh = get_filehandle(fd); |
9063 | if (!fh) | |
9064 | return -EBADF; | |
9065 | #if defined(__linux__) && defined(O_PATH) | |
9066 | if (fh->flags & O_PATH) | |
9067 | return -EBADF; | |
9068 | #endif | |
9069 | int r = _write(fh, offset, size, buf, NULL, 0); | |
9070 | ldout(cct, 3) << "write(" << fd << ", \"...\", " << size << ", " << offset << ") = " << r << dendl; | |
9071 | return r; | |
9072 | } | |
9073 | ||
9074 | int Client::pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset) | |
9075 | { | |
9076 | if (iovcnt < 0) | |
9077 | return -EINVAL; | |
9078 | return _preadv_pwritev(fd, iov, iovcnt, offset, true); | |
9079 | } | |
9080 | ||
9081 | int Client::_preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write) | |
9082 | { | |
9083 | Mutex::Locker lock(client_lock); | |
9084 | tout(cct) << fd << std::endl; | |
9085 | tout(cct) << offset << std::endl; | |
9086 | ||
181888fb FG |
9087 | if (unmounting) |
9088 | return -ENOTCONN; | |
9089 | ||
7c673cae FG |
9090 | Fh *fh = get_filehandle(fd); |
9091 | if (!fh) | |
9092 | return -EBADF; | |
9093 | #if defined(__linux__) && defined(O_PATH) | |
9094 | if (fh->flags & O_PATH) | |
9095 | return -EBADF; | |
9096 | #endif | |
9097 | loff_t totallen = 0; | |
9098 | for (unsigned i = 0; i < iovcnt; i++) { | |
9099 | totallen += iov[i].iov_len; | |
9100 | } | |
9101 | if (write) { | |
9102 | int w = _write(fh, offset, totallen, NULL, iov, iovcnt); | |
9103 | ldout(cct, 3) << "pwritev(" << fd << ", \"...\", " << totallen << ", " << offset << ") = " << w << dendl; | |
9104 | return w; | |
9105 | } else { | |
9106 | bufferlist bl; | |
9107 | int r = _read(fh, offset, totallen, &bl); | |
9108 | ldout(cct, 3) << "preadv(" << fd << ", " << offset << ") = " << r << dendl; | |
9109 | if (r <= 0) | |
9110 | return r; | |
9111 | ||
9112 | int bufoff = 0; | |
9113 | for (unsigned j = 0, resid = r; j < iovcnt && resid > 0; j++) { | |
9114 | /* | |
9115 | * This piece of code aims to handle the case that bufferlist does not have enough data | |
9116 | * to fill in the iov | |
9117 | */ | |
9118 | if (resid < iov[j].iov_len) { | |
9119 | bl.copy(bufoff, resid, (char *)iov[j].iov_base); | |
9120 | break; | |
9121 | } else { | |
9122 | bl.copy(bufoff, iov[j].iov_len, (char *)iov[j].iov_base); | |
9123 | } | |
9124 | resid -= iov[j].iov_len; | |
9125 | bufoff += iov[j].iov_len; | |
9126 | } | |
9127 | return r; | |
9128 | } | |
9129 | } | |
9130 | ||
9131 | int Client::_write(Fh *f, int64_t offset, uint64_t size, const char *buf, | |
9132 | const struct iovec *iov, int iovcnt) | |
9133 | { | |
9134 | if ((uint64_t)(offset+size) > mdsmap->get_max_filesize()) //too large! | |
9135 | return -EFBIG; | |
9136 | ||
9137 | //ldout(cct, 7) << "write fh " << fh << " size " << size << " offset " << offset << dendl; | |
9138 | Inode *in = f->inode.get(); | |
9139 | ||
9140 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
9141 | return -ENOSPC; | |
9142 | } | |
9143 | ||
9144 | assert(in->snapid == CEPH_NOSNAP); | |
9145 | ||
9146 | // was Fh opened as writeable? | |
9147 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) | |
9148 | return -EBADF; | |
9149 | ||
9150 | // check quota | |
9151 | uint64_t endoff = offset + size; | |
9152 | if (endoff > in->size && is_quota_bytes_exceeded(in, endoff - in->size, | |
9153 | f->actor_perms)) { | |
9154 | return -EDQUOT; | |
9155 | } | |
9156 | ||
9157 | // use/adjust fd pos? | |
9158 | if (offset < 0) { | |
9159 | lock_fh_pos(f); | |
9160 | /* | |
9161 | * FIXME: this is racy in that we may block _after_ this point waiting for caps, and size may | |
9162 | * change out from under us. | |
9163 | */ | |
9164 | if (f->flags & O_APPEND) { | |
9165 | int r = _lseek(f, 0, SEEK_END); | |
9166 | if (r < 0) { | |
9167 | unlock_fh_pos(f); | |
9168 | return r; | |
9169 | } | |
9170 | } | |
9171 | offset = f->pos; | |
9172 | f->pos = offset+size; | |
9173 | unlock_fh_pos(f); | |
9174 | } | |
9175 | ||
9176 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; | |
9177 | ||
9178 | ldout(cct, 10) << "cur file size is " << in->size << dendl; | |
9179 | ||
9180 | // time it. | |
9181 | utime_t start = ceph_clock_now(); | |
9182 | ||
9183 | if (in->inline_version == 0) { | |
9184 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
9185 | if (r < 0) | |
9186 | return r; | |
9187 | assert(in->inline_version > 0); | |
9188 | } | |
9189 | ||
9190 | // copy into fresh buffer (since our write may be resub, async) | |
9191 | bufferlist bl; | |
9192 | if (buf) { | |
9193 | if (size > 0) | |
9194 | bl.append(buf, size); | |
9195 | } else if (iov){ | |
9196 | for (int i = 0; i < iovcnt; i++) { | |
9197 | if (iov[i].iov_len > 0) { | |
9198 | bl.append((const char *)iov[i].iov_base, iov[i].iov_len); | |
9199 | } | |
9200 | } | |
9201 | } | |
9202 | ||
9203 | utime_t lat; | |
9204 | uint64_t totalwritten; | |
9205 | int have; | |
9206 | int r = get_caps(in, CEPH_CAP_FILE_WR|CEPH_CAP_AUTH_SHARED, | |
9207 | CEPH_CAP_FILE_BUFFER, &have, endoff); | |
9208 | if (r < 0) | |
9209 | return r; | |
9210 | ||
9211 | /* clear the setuid/setgid bits, if any */ | |
181888fb | 9212 | if (unlikely(in->mode & (S_ISUID|S_ISGID)) && size > 0) { |
7c673cae FG |
9213 | struct ceph_statx stx = { 0 }; |
9214 | ||
9215 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
9216 | r = __setattrx(in, &stx, CEPH_SETATTR_KILL_SGUID, f->actor_perms); | |
9217 | if (r < 0) | |
9218 | return r; | |
9219 | } else { | |
9220 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
9221 | } | |
9222 | ||
9223 | if (f->flags & O_DIRECT) | |
9224 | have &= ~CEPH_CAP_FILE_BUFFER; | |
9225 | ||
9226 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
9227 | ||
9228 | Mutex uninline_flock("Client::_write_uninline_data flock"); | |
9229 | Cond uninline_cond; | |
9230 | bool uninline_done = false; | |
9231 | int uninline_ret = 0; | |
9232 | Context *onuninline = NULL; | |
9233 | ||
9234 | if (in->inline_version < CEPH_INLINE_NONE) { | |
9235 | if (endoff > cct->_conf->client_max_inline_size || | |
9236 | endoff > CEPH_INLINE_MAX_SIZE || | |
9237 | !(have & CEPH_CAP_FILE_BUFFER)) { | |
9238 | onuninline = new C_SafeCond(&uninline_flock, | |
9239 | &uninline_cond, | |
9240 | &uninline_done, | |
9241 | &uninline_ret); | |
9242 | uninline_data(in, onuninline); | |
9243 | } else { | |
9244 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9245 | ||
9246 | uint32_t len = in->inline_data.length(); | |
9247 | ||
9248 | if (endoff < len) | |
9249 | in->inline_data.copy(endoff, len - endoff, bl); | |
9250 | ||
9251 | if (offset < len) | |
9252 | in->inline_data.splice(offset, len - offset); | |
9253 | else if (offset > len) | |
9254 | in->inline_data.append_zero(offset - len); | |
9255 | ||
9256 | in->inline_data.append(bl); | |
9257 | in->inline_version++; | |
9258 | ||
9259 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9260 | ||
9261 | goto success; | |
9262 | } | |
9263 | } | |
9264 | ||
9265 | if (cct->_conf->client_oc && (have & CEPH_CAP_FILE_BUFFER)) { | |
9266 | // do buffered write | |
9267 | if (!in->oset.dirty_or_tx) | |
9268 | get_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
9269 | ||
9270 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9271 | ||
9272 | // async, caching, non-blocking. | |
9273 | r = objectcacher->file_write(&in->oset, &in->layout, | |
9274 | in->snaprealm->get_snap_context(), | |
9275 | offset, size, bl, ceph::real_clock::now(), | |
9276 | 0); | |
9277 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9278 | ||
9279 | if (r < 0) | |
9280 | goto done; | |
9281 | ||
9282 | // flush cached write if O_SYNC is set on file fh | |
9283 | // O_DSYNC == O_SYNC on linux < 2.6.33 | |
9284 | // O_SYNC = __O_SYNC | O_DSYNC on linux >= 2.6.33 | |
9285 | if ((f->flags & O_SYNC) || (f->flags & O_DSYNC)) { | |
9286 | _flush_range(in, offset, size); | |
9287 | } | |
9288 | } else { | |
9289 | if (f->flags & O_DIRECT) | |
9290 | _flush_range(in, offset, size); | |
9291 | ||
9292 | // simple, non-atomic sync write | |
9293 | Mutex flock("Client::_write flock"); | |
9294 | Cond cond; | |
9295 | bool done = false; | |
9296 | Context *onfinish = new C_SafeCond(&flock, &cond, &done); | |
9297 | ||
9298 | unsafe_sync_write++; | |
9299 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); // released by onsafe callback | |
9300 | ||
9301 | filer->write_trunc(in->ino, &in->layout, in->snaprealm->get_snap_context(), | |
9302 | offset, size, bl, ceph::real_clock::now(), 0, | |
9303 | in->truncate_size, in->truncate_seq, | |
9304 | onfinish); | |
9305 | client_lock.Unlock(); | |
9306 | flock.Lock(); | |
9307 | ||
9308 | while (!done) | |
9309 | cond.Wait(flock); | |
9310 | flock.Unlock(); | |
9311 | client_lock.Lock(); | |
9312 | _sync_write_commit(in); | |
9313 | } | |
9314 | ||
9315 | // if we get here, write was successful, update client metadata | |
9316 | success: | |
9317 | // time | |
9318 | lat = ceph_clock_now(); | |
9319 | lat -= start; | |
9320 | logger->tinc(l_c_wrlat, lat); | |
9321 | ||
9322 | totalwritten = size; | |
9323 | r = (int)totalwritten; | |
9324 | ||
9325 | // extend file? | |
9326 | if (totalwritten + offset > in->size) { | |
9327 | in->size = totalwritten + offset; | |
9328 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
9329 | ||
9330 | if (is_quota_bytes_approaching(in, f->actor_perms)) { | |
9331 | check_caps(in, CHECK_CAPS_NODELAY); | |
31f18b77 FG |
9332 | } else if (is_max_size_approaching(in)) { |
9333 | check_caps(in, 0); | |
7c673cae FG |
9334 | } |
9335 | ||
9336 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", extending file size" << dendl; | |
9337 | } else { | |
9338 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", leaving file size at " << in->size << dendl; | |
9339 | } | |
9340 | ||
9341 | // mtime | |
9342 | in->mtime = ceph_clock_now(); | |
9343 | in->change_attr++; | |
9344 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
9345 | ||
9346 | done: | |
9347 | ||
9348 | if (onuninline) { | |
9349 | client_lock.Unlock(); | |
9350 | uninline_flock.Lock(); | |
9351 | while (!uninline_done) | |
9352 | uninline_cond.Wait(uninline_flock); | |
9353 | uninline_flock.Unlock(); | |
9354 | client_lock.Lock(); | |
9355 | ||
9356 | if (uninline_ret >= 0 || uninline_ret == -ECANCELED) { | |
9357 | in->inline_data.clear(); | |
9358 | in->inline_version = CEPH_INLINE_NONE; | |
9359 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
9360 | check_caps(in, 0); | |
9361 | } else | |
9362 | r = uninline_ret; | |
9363 | } | |
9364 | ||
9365 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
9366 | return r; | |
9367 | } | |
9368 | ||
9369 | int Client::_flush(Fh *f) | |
9370 | { | |
9371 | Inode *in = f->inode.get(); | |
9372 | int err = f->take_async_err(); | |
9373 | if (err != 0) { | |
9374 | ldout(cct, 1) << __func__ << ": " << f << " on inode " << *in << " caught async_err = " | |
9375 | << cpp_strerror(err) << dendl; | |
9376 | } else { | |
9377 | ldout(cct, 10) << __func__ << ": " << f << " on inode " << *in << " no async_err state" << dendl; | |
9378 | } | |
9379 | ||
9380 | return err; | |
9381 | } | |
9382 | ||
9383 | int Client::truncate(const char *relpath, loff_t length, const UserPerm& perms) | |
9384 | { | |
9385 | struct ceph_statx stx; | |
9386 | stx.stx_size = length; | |
9387 | return setattrx(relpath, &stx, CEPH_SETATTR_SIZE, perms); | |
9388 | } | |
9389 | ||
9390 | int Client::ftruncate(int fd, loff_t length, const UserPerm& perms) | |
9391 | { | |
9392 | Mutex::Locker lock(client_lock); | |
9393 | tout(cct) << "ftruncate" << std::endl; | |
9394 | tout(cct) << fd << std::endl; | |
9395 | tout(cct) << length << std::endl; | |
9396 | ||
181888fb FG |
9397 | if (unmounting) |
9398 | return -ENOTCONN; | |
9399 | ||
7c673cae FG |
9400 | Fh *f = get_filehandle(fd); |
9401 | if (!f) | |
9402 | return -EBADF; | |
9403 | #if defined(__linux__) && defined(O_PATH) | |
9404 | if (f->flags & O_PATH) | |
9405 | return -EBADF; | |
9406 | #endif | |
9407 | struct stat attr; | |
9408 | attr.st_size = length; | |
9409 | return _setattr(f->inode, &attr, CEPH_SETATTR_SIZE, perms); | |
9410 | } | |
9411 | ||
9412 | int Client::fsync(int fd, bool syncdataonly) | |
9413 | { | |
9414 | Mutex::Locker lock(client_lock); | |
9415 | tout(cct) << "fsync" << std::endl; | |
9416 | tout(cct) << fd << std::endl; | |
9417 | tout(cct) << syncdataonly << std::endl; | |
9418 | ||
181888fb FG |
9419 | if (unmounting) |
9420 | return -ENOTCONN; | |
9421 | ||
7c673cae FG |
9422 | Fh *f = get_filehandle(fd); |
9423 | if (!f) | |
9424 | return -EBADF; | |
9425 | #if defined(__linux__) && defined(O_PATH) | |
9426 | if (f->flags & O_PATH) | |
9427 | return -EBADF; | |
9428 | #endif | |
9429 | int r = _fsync(f, syncdataonly); | |
9430 | if (r == 0) { | |
9431 | // The IOs in this fsync were okay, but maybe something happened | |
9432 | // in the background that we shoudl be reporting? | |
9433 | r = f->take_async_err(); | |
9434 | ldout(cct, 3) << "fsync(" << fd << ", " << syncdataonly | |
9435 | << ") = 0, async_err = " << r << dendl; | |
9436 | } else { | |
9437 | // Assume that an error we encountered during fsync, even reported | |
9438 | // synchronously, would also have applied the error to the Fh, and we | |
9439 | // should clear it here to avoid returning the same error again on next | |
9440 | // call. | |
9441 | ldout(cct, 3) << "fsync(" << fd << ", " << syncdataonly << ") = " | |
9442 | << r << dendl; | |
9443 | f->take_async_err(); | |
9444 | } | |
9445 | return r; | |
9446 | } | |
9447 | ||
9448 | int Client::_fsync(Inode *in, bool syncdataonly) | |
9449 | { | |
9450 | int r = 0; | |
9451 | Mutex lock("Client::_fsync::lock"); | |
9452 | Cond cond; | |
9453 | bool done = false; | |
9454 | C_SafeCond *object_cacher_completion = NULL; | |
9455 | ceph_tid_t flush_tid = 0; | |
9456 | InodeRef tmp_ref; | |
9457 | ||
9458 | ldout(cct, 3) << "_fsync on " << *in << " " << (syncdataonly ? "(dataonly)":"(data+metadata)") << dendl; | |
9459 | ||
9460 | if (cct->_conf->client_oc) { | |
9461 | object_cacher_completion = new C_SafeCond(&lock, &cond, &done, &r); | |
9462 | tmp_ref = in; // take a reference; C_SafeCond doesn't and _flush won't either | |
9463 | _flush(in, object_cacher_completion); | |
9464 | ldout(cct, 15) << "using return-valued form of _fsync" << dendl; | |
9465 | } | |
9466 | ||
9467 | if (!syncdataonly && in->dirty_caps) { | |
9468 | check_caps(in, CHECK_CAPS_NODELAY|CHECK_CAPS_SYNCHRONOUS); | |
9469 | if (in->flushing_caps) | |
9470 | flush_tid = last_flush_tid; | |
9471 | } else ldout(cct, 10) << "no metadata needs to commit" << dendl; | |
9472 | ||
9473 | if (!syncdataonly && !in->unsafe_ops.empty()) { | |
9474 | MetaRequest *req = in->unsafe_ops.back(); | |
9475 | ldout(cct, 15) << "waiting on unsafe requests, last tid " << req->get_tid() << dendl; | |
9476 | ||
9477 | req->get(); | |
9478 | wait_on_list(req->waitfor_safe); | |
9479 | put_request(req); | |
9480 | } | |
9481 | ||
9482 | if (object_cacher_completion) { // wait on a real reply instead of guessing | |
9483 | client_lock.Unlock(); | |
9484 | lock.Lock(); | |
9485 | ldout(cct, 15) << "waiting on data to flush" << dendl; | |
9486 | while (!done) | |
9487 | cond.Wait(lock); | |
9488 | lock.Unlock(); | |
9489 | client_lock.Lock(); | |
9490 | ldout(cct, 15) << "got " << r << " from flush writeback" << dendl; | |
9491 | } else { | |
9492 | // FIXME: this can starve | |
9493 | while (in->cap_refs[CEPH_CAP_FILE_BUFFER] > 0) { | |
9494 | ldout(cct, 10) << "ino " << in->ino << " has " << in->cap_refs[CEPH_CAP_FILE_BUFFER] | |
9495 | << " uncommitted, waiting" << dendl; | |
9496 | wait_on_list(in->waitfor_commit); | |
9497 | } | |
9498 | } | |
9499 | ||
9500 | if (!r) { | |
9501 | if (flush_tid > 0) | |
9502 | wait_sync_caps(in, flush_tid); | |
9503 | ||
9504 | ldout(cct, 10) << "ino " << in->ino << " has no uncommitted writes" << dendl; | |
9505 | } else { | |
9506 | ldout(cct, 1) << "ino " << in->ino << " failed to commit to disk! " | |
9507 | << cpp_strerror(-r) << dendl; | |
9508 | } | |
9509 | ||
9510 | return r; | |
9511 | } | |
9512 | ||
9513 | int Client::_fsync(Fh *f, bool syncdataonly) | |
9514 | { | |
9515 | ldout(cct, 3) << "_fsync(" << f << ", " << (syncdataonly ? "dataonly)":"data+metadata)") << dendl; | |
9516 | return _fsync(f->inode.get(), syncdataonly); | |
9517 | } | |
9518 | ||
9519 | int Client::fstat(int fd, struct stat *stbuf, const UserPerm& perms, int mask) | |
9520 | { | |
9521 | Mutex::Locker lock(client_lock); | |
9522 | tout(cct) << "fstat mask " << hex << mask << dec << std::endl; | |
9523 | tout(cct) << fd << std::endl; | |
9524 | ||
181888fb FG |
9525 | if (unmounting) |
9526 | return -ENOTCONN; | |
9527 | ||
7c673cae FG |
9528 | Fh *f = get_filehandle(fd); |
9529 | if (!f) | |
9530 | return -EBADF; | |
9531 | int r = _getattr(f->inode, mask, perms); | |
9532 | if (r < 0) | |
9533 | return r; | |
9534 | fill_stat(f->inode, stbuf, NULL); | |
9535 | ldout(cct, 3) << "fstat(" << fd << ", " << stbuf << ") = " << r << dendl; | |
9536 | return r; | |
9537 | } | |
9538 | ||
9539 | int Client::fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
9540 | unsigned int want, unsigned int flags) | |
9541 | { | |
9542 | Mutex::Locker lock(client_lock); | |
9543 | tout(cct) << "fstatx flags " << hex << flags << " want " << want << dec << std::endl; | |
9544 | tout(cct) << fd << std::endl; | |
9545 | ||
181888fb FG |
9546 | if (unmounting) |
9547 | return -ENOTCONN; | |
9548 | ||
7c673cae FG |
9549 | Fh *f = get_filehandle(fd); |
9550 | if (!f) | |
9551 | return -EBADF; | |
9552 | ||
9553 | unsigned mask = statx_to_mask(flags, want); | |
9554 | ||
9555 | int r = 0; | |
9556 | if (mask && !f->inode->caps_issued_mask(mask)) { | |
9557 | r = _getattr(f->inode, mask, perms); | |
9558 | if (r < 0) { | |
9559 | ldout(cct, 3) << "fstatx exit on error!" << dendl; | |
9560 | return r; | |
9561 | } | |
9562 | } | |
9563 | ||
9564 | fill_statx(f->inode, mask, stx); | |
9565 | ldout(cct, 3) << "fstatx(" << fd << ", " << stx << ") = " << r << dendl; | |
9566 | return r; | |
9567 | } | |
9568 | ||
9569 | // not written yet, but i want to link! | |
9570 | ||
9571 | int Client::chdir(const char *relpath, std::string &new_cwd, | |
9572 | const UserPerm& perms) | |
9573 | { | |
9574 | Mutex::Locker lock(client_lock); | |
9575 | tout(cct) << "chdir" << std::endl; | |
9576 | tout(cct) << relpath << std::endl; | |
181888fb FG |
9577 | |
9578 | if (unmounting) | |
9579 | return -ENOTCONN; | |
9580 | ||
7c673cae FG |
9581 | filepath path(relpath); |
9582 | InodeRef in; | |
9583 | int r = path_walk(path, &in, perms); | |
9584 | if (r < 0) | |
9585 | return r; | |
9586 | if (cwd != in) | |
9587 | cwd.swap(in); | |
9588 | ldout(cct, 3) << "chdir(" << relpath << ") cwd now " << cwd->ino << dendl; | |
9589 | ||
b5b8bbf5 | 9590 | _getcwd(new_cwd, perms); |
7c673cae FG |
9591 | return 0; |
9592 | } | |
9593 | ||
b5b8bbf5 | 9594 | void Client::_getcwd(string& dir, const UserPerm& perms) |
7c673cae FG |
9595 | { |
9596 | filepath path; | |
9597 | ldout(cct, 10) << "getcwd " << *cwd << dendl; | |
9598 | ||
9599 | Inode *in = cwd.get(); | |
9600 | while (in != root) { | |
9601 | assert(in->dn_set.size() < 2); // dirs can't be hard-linked | |
9602 | ||
9603 | // A cwd or ancester is unlinked | |
9604 | if (in->dn_set.empty()) { | |
9605 | return; | |
9606 | } | |
9607 | ||
9608 | Dentry *dn = in->get_first_parent(); | |
9609 | ||
9610 | ||
9611 | if (!dn) { | |
9612 | // look it up | |
9613 | ldout(cct, 10) << "getcwd looking up parent for " << *in << dendl; | |
9614 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); | |
9615 | filepath path(in->ino); | |
9616 | req->set_filepath(path); | |
9617 | req->set_inode(in); | |
9618 | int res = make_request(req, perms); | |
9619 | if (res < 0) | |
9620 | break; | |
9621 | ||
9622 | // start over | |
9623 | path = filepath(); | |
9624 | in = cwd.get(); | |
9625 | continue; | |
9626 | } | |
9627 | path.push_front_dentry(dn->name); | |
9628 | in = dn->dir->parent_inode; | |
9629 | } | |
9630 | dir = "/"; | |
9631 | dir += path.get_path(); | |
9632 | } | |
9633 | ||
b5b8bbf5 FG |
9634 | void Client::getcwd(string& dir, const UserPerm& perms) |
9635 | { | |
9636 | Mutex::Locker l(client_lock); | |
181888fb FG |
9637 | if (!unmounting) |
9638 | _getcwd(dir, perms); | |
b5b8bbf5 FG |
9639 | } |
9640 | ||
7c673cae FG |
9641 | int Client::statfs(const char *path, struct statvfs *stbuf, |
9642 | const UserPerm& perms) | |
9643 | { | |
9644 | Mutex::Locker l(client_lock); | |
9645 | tout(cct) << "statfs" << std::endl; | |
9646 | ||
181888fb FG |
9647 | if (unmounting) |
9648 | return -ENOTCONN; | |
9649 | ||
7c673cae FG |
9650 | ceph_statfs stats; |
9651 | C_SaferCond cond; | |
d2e6a577 FG |
9652 | |
9653 | const vector<int64_t> &data_pools = mdsmap->get_data_pools(); | |
9654 | if (data_pools.size() == 1) { | |
9655 | objecter->get_fs_stats(stats, data_pools[0], &cond); | |
9656 | } else { | |
9657 | objecter->get_fs_stats(stats, boost::optional<int64_t>(), &cond); | |
9658 | } | |
7c673cae FG |
9659 | |
9660 | client_lock.Unlock(); | |
9661 | int rval = cond.wait(); | |
9662 | client_lock.Lock(); | |
9663 | ||
9664 | if (rval < 0) { | |
9665 | ldout(cct, 1) << "underlying call to statfs returned error: " | |
9666 | << cpp_strerror(rval) | |
9667 | << dendl; | |
9668 | return rval; | |
9669 | } | |
9670 | ||
9671 | memset(stbuf, 0, sizeof(*stbuf)); | |
9672 | ||
9673 | /* | |
9674 | * we're going to set a block size of 4MB so we can represent larger | |
9675 | * FSes without overflowing. Additionally convert the space | |
9676 | * measurements from KB to bytes while making them in terms of | |
9677 | * blocks. We use 4MB only because it is big enough, and because it | |
9678 | * actually *is* the (ceph) default block size. | |
9679 | */ | |
9680 | const int CEPH_BLOCK_SHIFT = 22; | |
9681 | stbuf->f_frsize = 1 << CEPH_BLOCK_SHIFT; | |
9682 | stbuf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
9683 | stbuf->f_files = stats.num_objects; | |
9684 | stbuf->f_ffree = -1; | |
9685 | stbuf->f_favail = -1; | |
9686 | stbuf->f_fsid = -1; // ?? | |
9687 | stbuf->f_flag = 0; // ?? | |
9688 | stbuf->f_namemax = NAME_MAX; | |
9689 | ||
9690 | // Usually quota_root will == root_ancestor, but if the mount root has no | |
9691 | // quota but we can see a parent of it that does have a quota, we'll | |
9692 | // respect that one instead. | |
9693 | assert(root != nullptr); | |
9694 | Inode *quota_root = root->quota.is_enable() ? root : get_quota_root(root, perms); | |
9695 | ||
9696 | // get_quota_root should always give us something | |
9697 | // because client quotas are always enabled | |
9698 | assert(quota_root != nullptr); | |
9699 | ||
9700 | if (quota_root && cct->_conf->client_quota_df && quota_root->quota.max_bytes) { | |
9701 | ||
9702 | // Skip the getattr if any sessions are stale, as we don't want to | |
9703 | // block `df` if this client has e.g. been evicted, or if the MDS cluster | |
9704 | // is unhealthy. | |
9705 | if (!_any_stale_sessions()) { | |
9706 | int r = _getattr(quota_root, 0, perms, true); | |
9707 | if (r != 0) { | |
9708 | // Ignore return value: error getting latest inode metadata is not a good | |
9709 | // reason to break "df". | |
9710 | lderr(cct) << "Error in getattr on quota root 0x" | |
9711 | << std::hex << quota_root->ino << std::dec | |
9712 | << " statfs result may be outdated" << dendl; | |
9713 | } | |
9714 | } | |
9715 | ||
9716 | // Special case: if there is a size quota set on the Inode acting | |
9717 | // as the root for this client mount, then report the quota status | |
9718 | // as the filesystem statistics. | |
9719 | const fsblkcnt_t total = quota_root->quota.max_bytes >> CEPH_BLOCK_SHIFT; | |
9720 | const fsblkcnt_t used = quota_root->rstat.rbytes >> CEPH_BLOCK_SHIFT; | |
31f18b77 FG |
9721 | // It is possible for a quota to be exceeded: arithmetic here must |
9722 | // handle case where used > total. | |
9723 | const fsblkcnt_t free = total > used ? total - used : 0; | |
7c673cae FG |
9724 | |
9725 | stbuf->f_blocks = total; | |
9726 | stbuf->f_bfree = free; | |
9727 | stbuf->f_bavail = free; | |
9728 | } else { | |
d2e6a577 | 9729 | // General case: report the cluster statistics returned from RADOS. Because |
7c673cae FG |
9730 | // multiple pools may be used without one filesystem namespace via |
9731 | // layouts, this is the most correct thing we can do. | |
9732 | stbuf->f_blocks = stats.kb >> (CEPH_BLOCK_SHIFT - 10); | |
9733 | stbuf->f_bfree = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
9734 | stbuf->f_bavail = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
9735 | } | |
9736 | ||
9737 | return rval; | |
9738 | } | |
9739 | ||
9740 | int Client::_do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
9741 | struct flock *fl, uint64_t owner, bool removing) | |
9742 | { | |
9743 | ldout(cct, 10) << "_do_filelock ino " << in->ino | |
9744 | << (lock_type == CEPH_LOCK_FCNTL ? " fcntl" : " flock") | |
9745 | << " type " << fl->l_type << " owner " << owner | |
9746 | << " " << fl->l_start << "~" << fl->l_len << dendl; | |
9747 | ||
9748 | int lock_cmd; | |
9749 | if (F_RDLCK == fl->l_type) | |
9750 | lock_cmd = CEPH_LOCK_SHARED; | |
9751 | else if (F_WRLCK == fl->l_type) | |
9752 | lock_cmd = CEPH_LOCK_EXCL; | |
9753 | else if (F_UNLCK == fl->l_type) | |
9754 | lock_cmd = CEPH_LOCK_UNLOCK; | |
9755 | else | |
9756 | return -EIO; | |
9757 | ||
9758 | if (op != CEPH_MDS_OP_SETFILELOCK || lock_cmd == CEPH_LOCK_UNLOCK) | |
9759 | sleep = 0; | |
9760 | ||
9761 | /* | |
9762 | * Set the most significant bit, so that MDS knows the 'owner' | |
9763 | * is sufficient to identify the owner of lock. (old code uses | |
9764 | * both 'owner' and 'pid') | |
9765 | */ | |
9766 | owner |= (1ULL << 63); | |
9767 | ||
9768 | MetaRequest *req = new MetaRequest(op); | |
9769 | filepath path; | |
9770 | in->make_nosnap_relative_path(path); | |
9771 | req->set_filepath(path); | |
9772 | req->set_inode(in); | |
9773 | ||
9774 | req->head.args.filelock_change.rule = lock_type; | |
9775 | req->head.args.filelock_change.type = lock_cmd; | |
9776 | req->head.args.filelock_change.owner = owner; | |
9777 | req->head.args.filelock_change.pid = fl->l_pid; | |
9778 | req->head.args.filelock_change.start = fl->l_start; | |
9779 | req->head.args.filelock_change.length = fl->l_len; | |
9780 | req->head.args.filelock_change.wait = sleep; | |
9781 | ||
9782 | int ret; | |
9783 | bufferlist bl; | |
9784 | ||
9785 | if (sleep && switch_interrupt_cb) { | |
9786 | // enable interrupt | |
9787 | switch_interrupt_cb(callback_handle, req->get()); | |
9788 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
7c673cae FG |
9789 | // disable interrupt |
9790 | switch_interrupt_cb(callback_handle, NULL); | |
31f18b77 FG |
9791 | if (ret == 0 && req->aborted()) { |
9792 | // effect of this lock request has been revoked by the 'lock intr' request | |
9793 | ret = req->get_abort_code(); | |
9794 | } | |
7c673cae FG |
9795 | put_request(req); |
9796 | } else { | |
9797 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
9798 | } | |
9799 | ||
9800 | if (ret == 0) { | |
9801 | if (op == CEPH_MDS_OP_GETFILELOCK) { | |
9802 | ceph_filelock filelock; | |
9803 | bufferlist::iterator p = bl.begin(); | |
9804 | ::decode(filelock, p); | |
9805 | ||
9806 | if (CEPH_LOCK_SHARED == filelock.type) | |
9807 | fl->l_type = F_RDLCK; | |
9808 | else if (CEPH_LOCK_EXCL == filelock.type) | |
9809 | fl->l_type = F_WRLCK; | |
9810 | else | |
9811 | fl->l_type = F_UNLCK; | |
9812 | ||
9813 | fl->l_whence = SEEK_SET; | |
9814 | fl->l_start = filelock.start; | |
9815 | fl->l_len = filelock.length; | |
9816 | fl->l_pid = filelock.pid; | |
9817 | } else if (op == CEPH_MDS_OP_SETFILELOCK) { | |
9818 | ceph_lock_state_t *lock_state; | |
9819 | if (lock_type == CEPH_LOCK_FCNTL) { | |
9820 | if (!in->fcntl_locks) | |
9821 | in->fcntl_locks = new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL); | |
9822 | lock_state = in->fcntl_locks; | |
9823 | } else if (lock_type == CEPH_LOCK_FLOCK) { | |
9824 | if (!in->flock_locks) | |
9825 | in->flock_locks = new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK); | |
9826 | lock_state = in->flock_locks; | |
9827 | } else { | |
9828 | ceph_abort(); | |
9829 | return -EINVAL; | |
9830 | } | |
9831 | _update_lock_state(fl, owner, lock_state); | |
9832 | ||
9833 | if (!removing) { | |
9834 | if (lock_type == CEPH_LOCK_FCNTL) { | |
9835 | if (!fh->fcntl_locks) | |
9836 | fh->fcntl_locks = new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL); | |
9837 | lock_state = fh->fcntl_locks; | |
9838 | } else { | |
9839 | if (!fh->flock_locks) | |
9840 | fh->flock_locks = new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK); | |
9841 | lock_state = fh->flock_locks; | |
9842 | } | |
9843 | _update_lock_state(fl, owner, lock_state); | |
9844 | } | |
9845 | } else | |
9846 | ceph_abort(); | |
9847 | } | |
9848 | return ret; | |
9849 | } | |
9850 | ||
9851 | int Client::_interrupt_filelock(MetaRequest *req) | |
9852 | { | |
31f18b77 FG |
9853 | // Set abort code, but do not kick. The abort code prevents the request |
9854 | // from being re-sent. | |
9855 | req->abort(-EINTR); | |
9856 | if (req->mds < 0) | |
9857 | return 0; // haven't sent the request | |
9858 | ||
7c673cae FG |
9859 | Inode *in = req->inode(); |
9860 | ||
9861 | int lock_type; | |
9862 | if (req->head.args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
9863 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
9864 | else if (req->head.args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
9865 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
9866 | else { | |
9867 | ceph_abort(); | |
9868 | return -EINVAL; | |
9869 | } | |
9870 | ||
9871 | MetaRequest *intr_req = new MetaRequest(CEPH_MDS_OP_SETFILELOCK); | |
9872 | filepath path; | |
9873 | in->make_nosnap_relative_path(path); | |
9874 | intr_req->set_filepath(path); | |
9875 | intr_req->set_inode(in); | |
9876 | intr_req->head.args.filelock_change = req->head.args.filelock_change; | |
9877 | intr_req->head.args.filelock_change.rule = lock_type; | |
9878 | intr_req->head.args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
9879 | ||
9880 | UserPerm perms(req->get_uid(), req->get_gid()); | |
9881 | return make_request(intr_req, perms, NULL, NULL, -1); | |
9882 | } | |
9883 | ||
9884 | void Client::_encode_filelocks(Inode *in, bufferlist& bl) | |
9885 | { | |
9886 | if (!in->fcntl_locks && !in->flock_locks) | |
9887 | return; | |
9888 | ||
9889 | unsigned nr_fcntl_locks = in->fcntl_locks ? in->fcntl_locks->held_locks.size() : 0; | |
9890 | ::encode(nr_fcntl_locks, bl); | |
9891 | if (nr_fcntl_locks) { | |
9892 | ceph_lock_state_t* lock_state = in->fcntl_locks; | |
9893 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); | |
9894 | p != lock_state->held_locks.end(); | |
9895 | ++p) | |
9896 | ::encode(p->second, bl); | |
9897 | } | |
9898 | ||
9899 | unsigned nr_flock_locks = in->flock_locks ? in->flock_locks->held_locks.size() : 0; | |
9900 | ::encode(nr_flock_locks, bl); | |
9901 | if (nr_flock_locks) { | |
9902 | ceph_lock_state_t* lock_state = in->flock_locks; | |
9903 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); | |
9904 | p != lock_state->held_locks.end(); | |
9905 | ++p) | |
9906 | ::encode(p->second, bl); | |
9907 | } | |
9908 | ||
9909 | ldout(cct, 10) << "_encode_filelocks ino " << in->ino << ", " << nr_fcntl_locks | |
9910 | << " fcntl locks, " << nr_flock_locks << " flock locks" << dendl; | |
9911 | } | |
9912 | ||
9913 | void Client::_release_filelocks(Fh *fh) | |
9914 | { | |
9915 | if (!fh->fcntl_locks && !fh->flock_locks) | |
9916 | return; | |
9917 | ||
9918 | Inode *in = fh->inode.get(); | |
9919 | ldout(cct, 10) << "_release_filelocks " << fh << " ino " << in->ino << dendl; | |
9920 | ||
9921 | list<pair<int, ceph_filelock> > to_release; | |
9922 | ||
9923 | if (fh->fcntl_locks) { | |
9924 | ceph_lock_state_t* lock_state = fh->fcntl_locks; | |
9925 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); | |
9926 | p != lock_state->held_locks.end(); | |
9927 | ++p) | |
9928 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FCNTL, p->second)); | |
9929 | delete fh->fcntl_locks; | |
9930 | } | |
9931 | if (fh->flock_locks) { | |
9932 | ceph_lock_state_t* lock_state = fh->flock_locks; | |
9933 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); | |
9934 | p != lock_state->held_locks.end(); | |
9935 | ++p) | |
9936 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FLOCK, p->second)); | |
9937 | delete fh->flock_locks; | |
9938 | } | |
9939 | ||
9940 | if (to_release.empty()) | |
9941 | return; | |
9942 | ||
9943 | struct flock fl; | |
9944 | memset(&fl, 0, sizeof(fl)); | |
9945 | fl.l_whence = SEEK_SET; | |
9946 | fl.l_type = F_UNLCK; | |
9947 | ||
9948 | for (list<pair<int, ceph_filelock> >::iterator p = to_release.begin(); | |
9949 | p != to_release.end(); | |
9950 | ++p) { | |
9951 | fl.l_start = p->second.start; | |
9952 | fl.l_len = p->second.length; | |
9953 | fl.l_pid = p->second.pid; | |
9954 | _do_filelock(in, fh, p->first, CEPH_MDS_OP_SETFILELOCK, 0, &fl, | |
9955 | p->second.owner, true); | |
9956 | } | |
9957 | } | |
9958 | ||
9959 | void Client::_update_lock_state(struct flock *fl, uint64_t owner, | |
9960 | ceph_lock_state_t *lock_state) | |
9961 | { | |
9962 | int lock_cmd; | |
9963 | if (F_RDLCK == fl->l_type) | |
9964 | lock_cmd = CEPH_LOCK_SHARED; | |
9965 | else if (F_WRLCK == fl->l_type) | |
9966 | lock_cmd = CEPH_LOCK_EXCL; | |
9967 | else | |
9968 | lock_cmd = CEPH_LOCK_UNLOCK;; | |
9969 | ||
9970 | ceph_filelock filelock; | |
9971 | filelock.start = fl->l_start; | |
9972 | filelock.length = fl->l_len; | |
9973 | filelock.client = 0; | |
9974 | // see comment in _do_filelock() | |
9975 | filelock.owner = owner | (1ULL << 63); | |
9976 | filelock.pid = fl->l_pid; | |
9977 | filelock.type = lock_cmd; | |
9978 | ||
9979 | if (filelock.type == CEPH_LOCK_UNLOCK) { | |
9980 | list<ceph_filelock> activated_locks; | |
9981 | lock_state->remove_lock(filelock, activated_locks); | |
9982 | } else { | |
9983 | bool r = lock_state->add_lock(filelock, false, false, NULL); | |
9984 | assert(r); | |
9985 | } | |
9986 | } | |
9987 | ||
9988 | int Client::_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
9989 | { | |
9990 | Inode *in = fh->inode.get(); | |
9991 | ldout(cct, 10) << "_getlk " << fh << " ino " << in->ino << dendl; | |
9992 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_GETFILELOCK, 0, fl, owner); | |
9993 | return ret; | |
9994 | } | |
9995 | ||
9996 | int Client::_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
9997 | { | |
9998 | Inode *in = fh->inode.get(); | |
9999 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << dendl; | |
10000 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_SETFILELOCK, sleep, fl, owner); | |
10001 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
10002 | return ret; | |
10003 | } | |
10004 | ||
10005 | int Client::_flock(Fh *fh, int cmd, uint64_t owner) | |
10006 | { | |
10007 | Inode *in = fh->inode.get(); | |
10008 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << dendl; | |
10009 | ||
10010 | int sleep = !(cmd & LOCK_NB); | |
10011 | cmd &= ~LOCK_NB; | |
10012 | ||
10013 | int type; | |
10014 | switch (cmd) { | |
10015 | case LOCK_SH: | |
10016 | type = F_RDLCK; | |
10017 | break; | |
10018 | case LOCK_EX: | |
10019 | type = F_WRLCK; | |
10020 | break; | |
10021 | case LOCK_UN: | |
10022 | type = F_UNLCK; | |
10023 | break; | |
10024 | default: | |
10025 | return -EINVAL; | |
10026 | } | |
10027 | ||
10028 | struct flock fl; | |
10029 | memset(&fl, 0, sizeof(fl)); | |
10030 | fl.l_type = type; | |
10031 | fl.l_whence = SEEK_SET; | |
10032 | ||
10033 | int ret = _do_filelock(in, fh, CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, sleep, &fl, owner); | |
10034 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
10035 | return ret; | |
10036 | } | |
10037 | ||
10038 | int Client::ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms) | |
10039 | { | |
10040 | /* Since the only thing this does is wrap a call to statfs, and | |
10041 | statfs takes a lock, it doesn't seem we have a need to split it | |
10042 | out. */ | |
10043 | return statfs(0, stbuf, perms); | |
10044 | } | |
10045 | ||
10046 | void Client::ll_register_callbacks(struct client_callback_args *args) | |
10047 | { | |
10048 | if (!args) | |
10049 | return; | |
10050 | Mutex::Locker l(client_lock); | |
10051 | ldout(cct, 10) << "ll_register_callbacks cb " << args->handle | |
10052 | << " invalidate_ino_cb " << args->ino_cb | |
10053 | << " invalidate_dentry_cb " << args->dentry_cb | |
10054 | << " getgroups_cb" << args->getgroups_cb | |
10055 | << " switch_interrupt_cb " << args->switch_intr_cb | |
10056 | << " remount_cb " << args->remount_cb | |
10057 | << dendl; | |
10058 | callback_handle = args->handle; | |
10059 | if (args->ino_cb) { | |
10060 | ino_invalidate_cb = args->ino_cb; | |
10061 | async_ino_invalidator.start(); | |
10062 | } | |
10063 | if (args->dentry_cb) { | |
10064 | dentry_invalidate_cb = args->dentry_cb; | |
10065 | async_dentry_invalidator.start(); | |
10066 | } | |
10067 | if (args->switch_intr_cb) { | |
10068 | switch_interrupt_cb = args->switch_intr_cb; | |
10069 | interrupt_finisher.start(); | |
10070 | } | |
10071 | if (args->remount_cb) { | |
10072 | remount_cb = args->remount_cb; | |
10073 | remount_finisher.start(); | |
10074 | } | |
10075 | getgroups_cb = args->getgroups_cb; | |
10076 | umask_cb = args->umask_cb; | |
10077 | } | |
10078 | ||
10079 | int Client::test_dentry_handling(bool can_invalidate) | |
10080 | { | |
10081 | int r = 0; | |
10082 | ||
10083 | can_invalidate_dentries = can_invalidate; | |
10084 | ||
10085 | if (can_invalidate_dentries) { | |
10086 | assert(dentry_invalidate_cb); | |
10087 | ldout(cct, 1) << "using dentry_invalidate_cb" << dendl; | |
b32b8144 | 10088 | r = 0; |
7c673cae FG |
10089 | } else if (remount_cb) { |
10090 | ldout(cct, 1) << "using remount_cb" << dendl; | |
b32b8144 FG |
10091 | r = _do_remount(); |
10092 | } | |
10093 | if (r) { | |
10094 | bool should_abort = cct->_conf->get_val<bool>("client_die_on_failed_dentry_invalidate"); | |
10095 | if (should_abort) { | |
10096 | lderr(cct) << "no method to invalidate kernel dentry cache; quitting!" << dendl; | |
7c673cae | 10097 | ceph_abort(); |
b32b8144 FG |
10098 | } else { |
10099 | lderr(cct) << "no method to invalidate kernel dentry cache; expect issues!" << dendl; | |
10100 | } | |
7c673cae FG |
10101 | } |
10102 | return r; | |
10103 | } | |
10104 | ||
10105 | int Client::_sync_fs() | |
10106 | { | |
10107 | ldout(cct, 10) << "_sync_fs" << dendl; | |
10108 | ||
10109 | // flush file data | |
10110 | Mutex lock("Client::_fsync::lock"); | |
10111 | Cond cond; | |
10112 | bool flush_done = false; | |
10113 | if (cct->_conf->client_oc) | |
10114 | objectcacher->flush_all(new C_SafeCond(&lock, &cond, &flush_done)); | |
10115 | else | |
10116 | flush_done = true; | |
10117 | ||
10118 | // flush caps | |
10119 | flush_caps_sync(); | |
10120 | ceph_tid_t flush_tid = last_flush_tid; | |
10121 | ||
10122 | // wait for unsafe mds requests | |
10123 | wait_unsafe_requests(); | |
10124 | ||
10125 | wait_sync_caps(flush_tid); | |
10126 | ||
10127 | if (!flush_done) { | |
10128 | client_lock.Unlock(); | |
10129 | lock.Lock(); | |
10130 | ldout(cct, 15) << "waiting on data to flush" << dendl; | |
10131 | while (!flush_done) | |
10132 | cond.Wait(lock); | |
10133 | lock.Unlock(); | |
10134 | client_lock.Lock(); | |
10135 | } | |
10136 | ||
10137 | return 0; | |
10138 | } | |
10139 | ||
10140 | int Client::sync_fs() | |
10141 | { | |
10142 | Mutex::Locker l(client_lock); | |
181888fb FG |
10143 | |
10144 | if (unmounting) | |
10145 | return -ENOTCONN; | |
10146 | ||
7c673cae FG |
10147 | return _sync_fs(); |
10148 | } | |
10149 | ||
10150 | int64_t Client::drop_caches() | |
10151 | { | |
10152 | Mutex::Locker l(client_lock); | |
10153 | return objectcacher->release_all(); | |
10154 | } | |
10155 | ||
10156 | ||
10157 | int Client::lazyio_propogate(int fd, loff_t offset, size_t count) | |
10158 | { | |
10159 | Mutex::Locker l(client_lock); | |
10160 | ldout(cct, 3) << "op: client->lazyio_propogate(" << fd | |
10161 | << ", " << offset << ", " << count << ")" << dendl; | |
10162 | ||
10163 | Fh *f = get_filehandle(fd); | |
10164 | if (!f) | |
10165 | return -EBADF; | |
10166 | ||
10167 | // for now | |
10168 | _fsync(f, true); | |
10169 | ||
10170 | return 0; | |
10171 | } | |
10172 | ||
10173 | int Client::lazyio_synchronize(int fd, loff_t offset, size_t count) | |
10174 | { | |
10175 | Mutex::Locker l(client_lock); | |
10176 | ldout(cct, 3) << "op: client->lazyio_synchronize(" << fd | |
10177 | << ", " << offset << ", " << count << ")" << dendl; | |
10178 | ||
10179 | Fh *f = get_filehandle(fd); | |
10180 | if (!f) | |
10181 | return -EBADF; | |
10182 | Inode *in = f->inode.get(); | |
10183 | ||
10184 | _fsync(f, true); | |
10185 | if (_release(in)) | |
10186 | check_caps(in, 0); | |
10187 | return 0; | |
10188 | } | |
10189 | ||
10190 | ||
10191 | // ============================= | |
10192 | // snaps | |
10193 | ||
10194 | int Client::mksnap(const char *relpath, const char *name, const UserPerm& perm) | |
10195 | { | |
10196 | Mutex::Locker l(client_lock); | |
181888fb FG |
10197 | |
10198 | if (unmounting) | |
10199 | return -ENOTCONN; | |
10200 | ||
7c673cae FG |
10201 | filepath path(relpath); |
10202 | InodeRef in; | |
10203 | int r = path_walk(path, &in, perm); | |
10204 | if (r < 0) | |
10205 | return r; | |
10206 | if (cct->_conf->client_permissions) { | |
10207 | r = may_create(in.get(), perm); | |
10208 | if (r < 0) | |
10209 | return r; | |
10210 | } | |
10211 | Inode *snapdir = open_snapdir(in.get()); | |
10212 | return _mkdir(snapdir, name, 0, perm); | |
10213 | } | |
181888fb | 10214 | |
7c673cae FG |
10215 | int Client::rmsnap(const char *relpath, const char *name, const UserPerm& perms) |
10216 | { | |
10217 | Mutex::Locker l(client_lock); | |
181888fb FG |
10218 | |
10219 | if (unmounting) | |
10220 | return -ENOTCONN; | |
10221 | ||
7c673cae FG |
10222 | filepath path(relpath); |
10223 | InodeRef in; | |
10224 | int r = path_walk(path, &in, perms); | |
10225 | if (r < 0) | |
10226 | return r; | |
10227 | if (cct->_conf->client_permissions) { | |
10228 | r = may_delete(in.get(), NULL, perms); | |
10229 | if (r < 0) | |
10230 | return r; | |
10231 | } | |
10232 | Inode *snapdir = open_snapdir(in.get()); | |
10233 | return _rmdir(snapdir, name, perms); | |
10234 | } | |
10235 | ||
10236 | // ============================= | |
10237 | // expose caps | |
10238 | ||
10239 | int Client::get_caps_issued(int fd) { | |
10240 | ||
10241 | Mutex::Locker lock(client_lock); | |
10242 | ||
181888fb FG |
10243 | if (unmounting) |
10244 | return -ENOTCONN; | |
10245 | ||
7c673cae FG |
10246 | Fh *f = get_filehandle(fd); |
10247 | if (!f) | |
10248 | return -EBADF; | |
10249 | ||
10250 | return f->inode->caps_issued(); | |
10251 | } | |
10252 | ||
10253 | int Client::get_caps_issued(const char *path, const UserPerm& perms) | |
10254 | { | |
10255 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10256 | |
10257 | if (unmounting) | |
10258 | return -ENOTCONN; | |
10259 | ||
7c673cae FG |
10260 | filepath p(path); |
10261 | InodeRef in; | |
10262 | int r = path_walk(p, &in, perms, true); | |
10263 | if (r < 0) | |
10264 | return r; | |
10265 | return in->caps_issued(); | |
10266 | } | |
10267 | ||
10268 | // ========================================= | |
10269 | // low level | |
10270 | ||
10271 | Inode *Client::open_snapdir(Inode *diri) | |
10272 | { | |
10273 | Inode *in; | |
10274 | vinodeno_t vino(diri->ino, CEPH_SNAPDIR); | |
10275 | if (!inode_map.count(vino)) { | |
10276 | in = new Inode(this, vino, &diri->layout); | |
10277 | ||
10278 | in->ino = diri->ino; | |
10279 | in->snapid = CEPH_SNAPDIR; | |
10280 | in->mode = diri->mode; | |
10281 | in->uid = diri->uid; | |
10282 | in->gid = diri->gid; | |
10283 | in->mtime = diri->mtime; | |
10284 | in->ctime = diri->ctime; | |
10285 | in->btime = diri->btime; | |
10286 | in->size = diri->size; | |
10287 | in->change_attr = diri->change_attr; | |
10288 | ||
10289 | in->dirfragtree.clear(); | |
10290 | in->snapdir_parent = diri; | |
10291 | diri->flags |= I_SNAPDIR_OPEN; | |
10292 | inode_map[vino] = in; | |
10293 | if (use_faked_inos()) | |
10294 | _assign_faked_ino(in); | |
10295 | ldout(cct, 10) << "open_snapdir created snapshot inode " << *in << dendl; | |
10296 | } else { | |
10297 | in = inode_map[vino]; | |
10298 | ldout(cct, 10) << "open_snapdir had snapshot inode " << *in << dendl; | |
10299 | } | |
10300 | return in; | |
10301 | } | |
10302 | ||
10303 | int Client::ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
10304 | Inode **out, const UserPerm& perms) | |
10305 | { | |
10306 | Mutex::Locker lock(client_lock); | |
31f18b77 FG |
10307 | vinodeno_t vparent = _get_vino(parent); |
10308 | ldout(cct, 3) << "ll_lookup " << vparent << " " << name << dendl; | |
7c673cae FG |
10309 | tout(cct) << "ll_lookup" << std::endl; |
10310 | tout(cct) << name << std::endl; | |
10311 | ||
181888fb FG |
10312 | if (unmounting) |
10313 | return -ENOTCONN; | |
10314 | ||
7c673cae FG |
10315 | int r = 0; |
10316 | if (!cct->_conf->fuse_default_permissions) { | |
10317 | r = may_lookup(parent, perms); | |
10318 | if (r < 0) | |
10319 | return r; | |
10320 | } | |
10321 | ||
10322 | string dname(name); | |
10323 | InodeRef in; | |
10324 | ||
10325 | r = _lookup(parent, dname, CEPH_STAT_CAP_INODE_ALL, &in, perms); | |
10326 | if (r < 0) { | |
10327 | attr->st_ino = 0; | |
10328 | goto out; | |
10329 | } | |
10330 | ||
10331 | assert(in); | |
10332 | fill_stat(in, attr); | |
10333 | _ll_get(in.get()); | |
10334 | ||
10335 | out: | |
31f18b77 | 10336 | ldout(cct, 3) << "ll_lookup " << vparent << " " << name |
7c673cae FG |
10337 | << " -> " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; |
10338 | tout(cct) << attr->st_ino << std::endl; | |
10339 | *out = in.get(); | |
10340 | return r; | |
10341 | } | |
10342 | ||
10343 | int Client::ll_lookupx(Inode *parent, const char *name, Inode **out, | |
10344 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
10345 | const UserPerm& perms) | |
10346 | { | |
10347 | Mutex::Locker lock(client_lock); | |
31f18b77 FG |
10348 | vinodeno_t vparent = _get_vino(parent); |
10349 | ldout(cct, 3) << "ll_lookupx " << vparent << " " << name << dendl; | |
7c673cae FG |
10350 | tout(cct) << "ll_lookupx" << std::endl; |
10351 | tout(cct) << name << std::endl; | |
10352 | ||
181888fb FG |
10353 | if (unmounting) |
10354 | return -ENOTCONN; | |
10355 | ||
7c673cae FG |
10356 | int r = 0; |
10357 | if (!cct->_conf->fuse_default_permissions) { | |
10358 | r = may_lookup(parent, perms); | |
10359 | if (r < 0) | |
10360 | return r; | |
10361 | } | |
10362 | ||
10363 | string dname(name); | |
10364 | InodeRef in; | |
10365 | ||
10366 | unsigned mask = statx_to_mask(flags, want); | |
10367 | r = _lookup(parent, dname, mask, &in, perms); | |
10368 | if (r < 0) { | |
10369 | stx->stx_ino = 0; | |
10370 | stx->stx_mask = 0; | |
10371 | } else { | |
10372 | assert(in); | |
10373 | fill_statx(in, mask, stx); | |
10374 | _ll_get(in.get()); | |
10375 | } | |
10376 | ||
31f18b77 | 10377 | ldout(cct, 3) << "ll_lookupx " << vparent << " " << name |
7c673cae FG |
10378 | << " -> " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; |
10379 | tout(cct) << stx->stx_ino << std::endl; | |
10380 | *out = in.get(); | |
10381 | return r; | |
10382 | } | |
10383 | ||
10384 | int Client::ll_walk(const char* name, Inode **out, struct ceph_statx *stx, | |
10385 | unsigned int want, unsigned int flags, const UserPerm& perms) | |
10386 | { | |
10387 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10388 | |
10389 | if (unmounting) | |
10390 | return -ENOTCONN; | |
10391 | ||
7c673cae FG |
10392 | filepath fp(name, 0); |
10393 | InodeRef in; | |
10394 | int rc; | |
10395 | unsigned mask = statx_to_mask(flags, want); | |
10396 | ||
10397 | ldout(cct, 3) << "ll_walk" << name << dendl; | |
10398 | tout(cct) << "ll_walk" << std::endl; | |
10399 | tout(cct) << name << std::endl; | |
10400 | ||
10401 | rc = path_walk(fp, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); | |
10402 | if (rc < 0) { | |
10403 | /* zero out mask, just in case... */ | |
10404 | stx->stx_mask = 0; | |
10405 | stx->stx_ino = 0; | |
10406 | *out = NULL; | |
10407 | return rc; | |
10408 | } else { | |
10409 | assert(in); | |
10410 | fill_statx(in, mask, stx); | |
10411 | _ll_get(in.get()); | |
10412 | *out = in.get(); | |
10413 | return 0; | |
10414 | } | |
10415 | } | |
10416 | ||
10417 | void Client::_ll_get(Inode *in) | |
10418 | { | |
10419 | if (in->ll_ref == 0) { | |
10420 | in->get(); | |
10421 | if (in->is_dir() && !in->dn_set.empty()) { | |
10422 | assert(in->dn_set.size() == 1); // dirs can't be hard-linked | |
10423 | in->get_first_parent()->get(); // pin dentry | |
10424 | } | |
10425 | } | |
10426 | in->ll_get(); | |
10427 | ldout(cct, 20) << "_ll_get " << in << " " << in->ino << " -> " << in->ll_ref << dendl; | |
10428 | } | |
10429 | ||
10430 | int Client::_ll_put(Inode *in, int num) | |
10431 | { | |
10432 | in->ll_put(num); | |
10433 | ldout(cct, 20) << "_ll_put " << in << " " << in->ino << " " << num << " -> " << in->ll_ref << dendl; | |
10434 | if (in->ll_ref == 0) { | |
10435 | if (in->is_dir() && !in->dn_set.empty()) { | |
10436 | assert(in->dn_set.size() == 1); // dirs can't be hard-linked | |
10437 | in->get_first_parent()->put(); // unpin dentry | |
10438 | } | |
10439 | put_inode(in); | |
10440 | return 0; | |
10441 | } else { | |
10442 | return in->ll_ref; | |
10443 | } | |
10444 | } | |
10445 | ||
10446 | void Client::_ll_drop_pins() | |
10447 | { | |
10448 | ldout(cct, 10) << "_ll_drop_pins" << dendl; | |
10449 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; | |
10450 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
10451 | it != inode_map.end(); | |
10452 | it = next) { | |
10453 | Inode *in = it->second; | |
10454 | next = it; | |
10455 | ++next; | |
10456 | if (in->ll_ref) | |
10457 | _ll_put(in, in->ll_ref); | |
10458 | } | |
10459 | } | |
10460 | ||
10461 | bool Client::ll_forget(Inode *in, int count) | |
10462 | { | |
10463 | Mutex::Locker lock(client_lock); | |
10464 | inodeno_t ino = _get_inodeno(in); | |
10465 | ||
10466 | ldout(cct, 3) << "ll_forget " << ino << " " << count << dendl; | |
10467 | tout(cct) << "ll_forget" << std::endl; | |
10468 | tout(cct) << ino.val << std::endl; | |
10469 | tout(cct) << count << std::endl; | |
10470 | ||
181888fb FG |
10471 | // Ignore forget if we're no longer mounted |
10472 | if (unmounting) | |
10473 | return true; | |
10474 | ||
7c673cae FG |
10475 | if (ino == 1) return true; // ignore forget on root. |
10476 | ||
10477 | bool last = false; | |
10478 | if (in->ll_ref < count) { | |
10479 | ldout(cct, 1) << "WARNING: ll_forget on " << ino << " " << count | |
10480 | << ", which only has ll_ref=" << in->ll_ref << dendl; | |
10481 | _ll_put(in, in->ll_ref); | |
10482 | last = true; | |
10483 | } else { | |
10484 | if (_ll_put(in, count) == 0) | |
10485 | last = true; | |
10486 | } | |
10487 | ||
10488 | return last; | |
10489 | } | |
10490 | ||
10491 | bool Client::ll_put(Inode *in) | |
10492 | { | |
10493 | /* ll_forget already takes the lock */ | |
10494 | return ll_forget(in, 1); | |
10495 | } | |
10496 | ||
10497 | snapid_t Client::ll_get_snapid(Inode *in) | |
10498 | { | |
10499 | Mutex::Locker lock(client_lock); | |
10500 | return in->snapid; | |
10501 | } | |
10502 | ||
10503 | Inode *Client::ll_get_inode(ino_t ino) | |
10504 | { | |
10505 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10506 | |
10507 | if (unmounting) | |
10508 | return NULL; | |
10509 | ||
7c673cae FG |
10510 | vinodeno_t vino = _map_faked_ino(ino); |
10511 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
10512 | if (p == inode_map.end()) | |
10513 | return NULL; | |
10514 | Inode *in = p->second; | |
10515 | _ll_get(in); | |
10516 | return in; | |
10517 | } | |
10518 | ||
10519 | Inode *Client::ll_get_inode(vinodeno_t vino) | |
10520 | { | |
10521 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10522 | |
10523 | if (unmounting) | |
10524 | return NULL; | |
10525 | ||
7c673cae FG |
10526 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
10527 | if (p == inode_map.end()) | |
10528 | return NULL; | |
10529 | Inode *in = p->second; | |
10530 | _ll_get(in); | |
10531 | return in; | |
10532 | } | |
10533 | ||
10534 | int Client::_ll_getattr(Inode *in, int caps, const UserPerm& perms) | |
10535 | { | |
10536 | vinodeno_t vino = _get_vino(in); | |
10537 | ||
10538 | ldout(cct, 3) << "ll_getattr " << vino << dendl; | |
10539 | tout(cct) << "ll_getattr" << std::endl; | |
10540 | tout(cct) << vino.ino.val << std::endl; | |
10541 | ||
10542 | if (vino.snapid < CEPH_NOSNAP) | |
10543 | return 0; | |
10544 | else | |
10545 | return _getattr(in, caps, perms); | |
10546 | } | |
10547 | ||
10548 | int Client::ll_getattr(Inode *in, struct stat *attr, const UserPerm& perms) | |
10549 | { | |
10550 | Mutex::Locker lock(client_lock); | |
10551 | ||
181888fb FG |
10552 | if (unmounting) |
10553 | return -ENOTCONN; | |
10554 | ||
7c673cae FG |
10555 | int res = _ll_getattr(in, CEPH_STAT_CAP_INODE_ALL, perms); |
10556 | ||
10557 | if (res == 0) | |
10558 | fill_stat(in, attr); | |
10559 | ldout(cct, 3) << "ll_getattr " << _get_vino(in) << " = " << res << dendl; | |
10560 | return res; | |
10561 | } | |
10562 | ||
10563 | int Client::ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
10564 | unsigned int flags, const UserPerm& perms) | |
10565 | { | |
10566 | Mutex::Locker lock(client_lock); | |
10567 | ||
181888fb FG |
10568 | if (unmounting) |
10569 | return -ENOTCONN; | |
10570 | ||
7c673cae FG |
10571 | int res = 0; |
10572 | unsigned mask = statx_to_mask(flags, want); | |
10573 | ||
10574 | if (mask && !in->caps_issued_mask(mask)) | |
10575 | res = _ll_getattr(in, mask, perms); | |
10576 | ||
10577 | if (res == 0) | |
10578 | fill_statx(in, mask, stx); | |
10579 | ldout(cct, 3) << "ll_getattrx " << _get_vino(in) << " = " << res << dendl; | |
10580 | return res; | |
10581 | } | |
10582 | ||
10583 | int Client::_ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
10584 | const UserPerm& perms, InodeRef *inp) | |
10585 | { | |
10586 | vinodeno_t vino = _get_vino(in); | |
10587 | ||
10588 | ldout(cct, 3) << "ll_setattrx " << vino << " mask " << hex << mask << dec | |
10589 | << dendl; | |
10590 | tout(cct) << "ll_setattrx" << std::endl; | |
10591 | tout(cct) << vino.ino.val << std::endl; | |
10592 | tout(cct) << stx->stx_mode << std::endl; | |
10593 | tout(cct) << stx->stx_uid << std::endl; | |
10594 | tout(cct) << stx->stx_gid << std::endl; | |
10595 | tout(cct) << stx->stx_size << std::endl; | |
10596 | tout(cct) << stx->stx_mtime << std::endl; | |
10597 | tout(cct) << stx->stx_atime << std::endl; | |
10598 | tout(cct) << stx->stx_btime << std::endl; | |
10599 | tout(cct) << mask << std::endl; | |
10600 | ||
10601 | if (!cct->_conf->fuse_default_permissions) { | |
10602 | int res = may_setattr(in, stx, mask, perms); | |
10603 | if (res < 0) | |
10604 | return res; | |
10605 | } | |
10606 | ||
10607 | mask &= ~(CEPH_SETATTR_MTIME_NOW | CEPH_SETATTR_ATIME_NOW); | |
10608 | ||
10609 | return __setattrx(in, stx, mask, perms, inp); | |
10610 | } | |
10611 | ||
10612 | int Client::ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
10613 | const UserPerm& perms) | |
10614 | { | |
10615 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10616 | |
10617 | if (unmounting) | |
10618 | return -ENOTCONN; | |
10619 | ||
7c673cae FG |
10620 | InodeRef target(in); |
10621 | int res = _ll_setattrx(in, stx, mask, perms, &target); | |
10622 | if (res == 0) { | |
10623 | assert(in == target.get()); | |
10624 | fill_statx(in, in->caps_issued(), stx); | |
10625 | } | |
10626 | ||
10627 | ldout(cct, 3) << "ll_setattrx " << _get_vino(in) << " = " << res << dendl; | |
10628 | return res; | |
10629 | } | |
10630 | ||
10631 | int Client::ll_setattr(Inode *in, struct stat *attr, int mask, | |
10632 | const UserPerm& perms) | |
10633 | { | |
10634 | struct ceph_statx stx; | |
10635 | stat_to_statx(attr, &stx); | |
10636 | ||
10637 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10638 | |
10639 | if (unmounting) | |
10640 | return -ENOTCONN; | |
10641 | ||
7c673cae FG |
10642 | InodeRef target(in); |
10643 | int res = _ll_setattrx(in, &stx, mask, perms, &target); | |
10644 | if (res == 0) { | |
10645 | assert(in == target.get()); | |
10646 | fill_stat(in, attr); | |
10647 | } | |
10648 | ||
10649 | ldout(cct, 3) << "ll_setattr " << _get_vino(in) << " = " << res << dendl; | |
10650 | return res; | |
10651 | } | |
10652 | ||
10653 | ||
10654 | // ---------- | |
10655 | // xattrs | |
10656 | ||
10657 | int Client::getxattr(const char *path, const char *name, void *value, size_t size, | |
10658 | const UserPerm& perms) | |
10659 | { | |
10660 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10661 | |
10662 | if (unmounting) | |
10663 | return -ENOTCONN; | |
10664 | ||
7c673cae FG |
10665 | InodeRef in; |
10666 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
10667 | if (r < 0) | |
10668 | return r; | |
10669 | return _getxattr(in, name, value, size, perms); | |
10670 | } | |
10671 | ||
10672 | int Client::lgetxattr(const char *path, const char *name, void *value, size_t size, | |
10673 | const UserPerm& perms) | |
10674 | { | |
10675 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10676 | |
10677 | if (unmounting) | |
10678 | return -ENOTCONN; | |
10679 | ||
7c673cae FG |
10680 | InodeRef in; |
10681 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
10682 | if (r < 0) | |
10683 | return r; | |
10684 | return _getxattr(in, name, value, size, perms); | |
10685 | } | |
10686 | ||
10687 | int Client::fgetxattr(int fd, const char *name, void *value, size_t size, | |
10688 | const UserPerm& perms) | |
10689 | { | |
10690 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10691 | |
10692 | if (unmounting) | |
10693 | return -ENOTCONN; | |
10694 | ||
7c673cae FG |
10695 | Fh *f = get_filehandle(fd); |
10696 | if (!f) | |
10697 | return -EBADF; | |
10698 | return _getxattr(f->inode, name, value, size, perms); | |
10699 | } | |
10700 | ||
10701 | int Client::listxattr(const char *path, char *list, size_t size, | |
10702 | const UserPerm& perms) | |
10703 | { | |
10704 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10705 | |
10706 | if (unmounting) | |
10707 | return -ENOTCONN; | |
10708 | ||
7c673cae FG |
10709 | InodeRef in; |
10710 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
10711 | if (r < 0) | |
10712 | return r; | |
10713 | return Client::_listxattr(in.get(), list, size, perms); | |
10714 | } | |
10715 | ||
10716 | int Client::llistxattr(const char *path, char *list, size_t size, | |
10717 | const UserPerm& perms) | |
10718 | { | |
10719 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10720 | |
10721 | if (unmounting) | |
10722 | return -ENOTCONN; | |
10723 | ||
7c673cae FG |
10724 | InodeRef in; |
10725 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
10726 | if (r < 0) | |
10727 | return r; | |
10728 | return Client::_listxattr(in.get(), list, size, perms); | |
10729 | } | |
10730 | ||
10731 | int Client::flistxattr(int fd, char *list, size_t size, const UserPerm& perms) | |
10732 | { | |
10733 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10734 | |
10735 | if (unmounting) | |
10736 | return -ENOTCONN; | |
10737 | ||
7c673cae FG |
10738 | Fh *f = get_filehandle(fd); |
10739 | if (!f) | |
10740 | return -EBADF; | |
10741 | return Client::_listxattr(f->inode.get(), list, size, perms); | |
10742 | } | |
10743 | ||
10744 | int Client::removexattr(const char *path, const char *name, | |
10745 | const UserPerm& perms) | |
10746 | { | |
10747 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10748 | |
10749 | if (unmounting) | |
10750 | return -ENOTCONN; | |
10751 | ||
7c673cae FG |
10752 | InodeRef in; |
10753 | int r = Client::path_walk(path, &in, perms, true); | |
10754 | if (r < 0) | |
10755 | return r; | |
10756 | return _removexattr(in, name, perms); | |
10757 | } | |
10758 | ||
10759 | int Client::lremovexattr(const char *path, const char *name, | |
10760 | const UserPerm& perms) | |
10761 | { | |
10762 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10763 | |
10764 | if (unmounting) | |
10765 | return -ENOTCONN; | |
10766 | ||
7c673cae FG |
10767 | InodeRef in; |
10768 | int r = Client::path_walk(path, &in, perms, false); | |
10769 | if (r < 0) | |
10770 | return r; | |
10771 | return _removexattr(in, name, perms); | |
10772 | } | |
10773 | ||
10774 | int Client::fremovexattr(int fd, const char *name, const UserPerm& perms) | |
10775 | { | |
10776 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10777 | |
10778 | if (unmounting) | |
10779 | return -ENOTCONN; | |
10780 | ||
7c673cae FG |
10781 | Fh *f = get_filehandle(fd); |
10782 | if (!f) | |
10783 | return -EBADF; | |
10784 | return _removexattr(f->inode, name, perms); | |
10785 | } | |
10786 | ||
10787 | int Client::setxattr(const char *path, const char *name, const void *value, | |
10788 | size_t size, int flags, const UserPerm& perms) | |
10789 | { | |
10790 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
10791 | ||
10792 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10793 | |
10794 | if (unmounting) | |
10795 | return -ENOTCONN; | |
10796 | ||
7c673cae FG |
10797 | InodeRef in; |
10798 | int r = Client::path_walk(path, &in, perms, true); | |
10799 | if (r < 0) | |
10800 | return r; | |
10801 | return _setxattr(in, name, value, size, flags, perms); | |
10802 | } | |
10803 | ||
10804 | int Client::lsetxattr(const char *path, const char *name, const void *value, | |
10805 | size_t size, int flags, const UserPerm& perms) | |
10806 | { | |
10807 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
10808 | ||
10809 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10810 | |
10811 | if (unmounting) | |
10812 | return -ENOTCONN; | |
10813 | ||
7c673cae FG |
10814 | InodeRef in; |
10815 | int r = Client::path_walk(path, &in, perms, false); | |
10816 | if (r < 0) | |
10817 | return r; | |
10818 | return _setxattr(in, name, value, size, flags, perms); | |
10819 | } | |
10820 | ||
10821 | int Client::fsetxattr(int fd, const char *name, const void *value, size_t size, | |
10822 | int flags, const UserPerm& perms) | |
10823 | { | |
10824 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
10825 | ||
10826 | Mutex::Locker lock(client_lock); | |
181888fb FG |
10827 | |
10828 | if (unmounting) | |
10829 | return -ENOTCONN; | |
10830 | ||
7c673cae FG |
10831 | Fh *f = get_filehandle(fd); |
10832 | if (!f) | |
10833 | return -EBADF; | |
10834 | return _setxattr(f->inode, name, value, size, flags, perms); | |
10835 | } | |
10836 | ||
10837 | int Client::_getxattr(Inode *in, const char *name, void *value, size_t size, | |
10838 | const UserPerm& perms) | |
10839 | { | |
10840 | int r; | |
10841 | ||
10842 | const VXattr *vxattr = _match_vxattr(in, name); | |
10843 | if (vxattr) { | |
10844 | r = -ENODATA; | |
10845 | ||
10846 | // Do a force getattr to get the latest quota before returning | |
10847 | // a value to userspace. | |
10848 | r = _getattr(in, 0, perms, true); | |
10849 | if (r != 0) { | |
10850 | // Error from getattr! | |
10851 | return r; | |
10852 | } | |
10853 | ||
10854 | // call pointer-to-member function | |
10855 | char buf[256]; | |
10856 | if (!(vxattr->exists_cb && !(this->*(vxattr->exists_cb))(in))) { | |
10857 | r = (this->*(vxattr->getxattr_cb))(in, buf, sizeof(buf)); | |
10858 | } else { | |
10859 | r = -ENODATA; | |
10860 | } | |
10861 | ||
10862 | if (size != 0) { | |
10863 | if (r > (int)size) { | |
10864 | r = -ERANGE; | |
10865 | } else if (r > 0) { | |
10866 | memcpy(value, buf, r); | |
10867 | } | |
10868 | } | |
10869 | goto out; | |
10870 | } | |
10871 | ||
10872 | if (acl_type == NO_ACL && !strncmp(name, "system.", 7)) { | |
10873 | r = -EOPNOTSUPP; | |
10874 | goto out; | |
10875 | } | |
10876 | ||
10877 | r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
10878 | if (r == 0) { | |
10879 | string n(name); | |
10880 | r = -ENODATA; | |
10881 | if (in->xattrs.count(n)) { | |
10882 | r = in->xattrs[n].length(); | |
10883 | if (r > 0 && size != 0) { | |
10884 | if (size >= (unsigned)r) | |
10885 | memcpy(value, in->xattrs[n].c_str(), r); | |
10886 | else | |
10887 | r = -ERANGE; | |
10888 | } | |
10889 | } | |
10890 | } | |
10891 | out: | |
10892 | ldout(cct, 3) << "_getxattr(" << in->ino << ", \"" << name << "\", " << size << ") = " << r << dendl; | |
10893 | return r; | |
10894 | } | |
10895 | ||
10896 | int Client::_getxattr(InodeRef &in, const char *name, void *value, size_t size, | |
10897 | const UserPerm& perms) | |
10898 | { | |
10899 | if (cct->_conf->client_permissions) { | |
10900 | int r = xattr_permission(in.get(), name, MAY_READ, perms); | |
10901 | if (r < 0) | |
10902 | return r; | |
10903 | } | |
10904 | return _getxattr(in.get(), name, value, size, perms); | |
10905 | } | |
10906 | ||
10907 | int Client::ll_getxattr(Inode *in, const char *name, void *value, | |
10908 | size_t size, const UserPerm& perms) | |
10909 | { | |
10910 | Mutex::Locker lock(client_lock); | |
10911 | ||
181888fb FG |
10912 | if (unmounting) |
10913 | return -ENOTCONN; | |
10914 | ||
7c673cae FG |
10915 | vinodeno_t vino = _get_vino(in); |
10916 | ||
10917 | ldout(cct, 3) << "ll_getxattr " << vino << " " << name << " size " << size << dendl; | |
10918 | tout(cct) << "ll_getxattr" << std::endl; | |
10919 | tout(cct) << vino.ino.val << std::endl; | |
10920 | tout(cct) << name << std::endl; | |
10921 | ||
10922 | if (!cct->_conf->fuse_default_permissions) { | |
10923 | int r = xattr_permission(in, name, MAY_READ, perms); | |
10924 | if (r < 0) | |
10925 | return r; | |
10926 | } | |
10927 | ||
10928 | return _getxattr(in, name, value, size, perms); | |
10929 | } | |
10930 | ||
10931 | int Client::_listxattr(Inode *in, char *name, size_t size, | |
10932 | const UserPerm& perms) | |
10933 | { | |
10934 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
10935 | if (r == 0) { | |
10936 | for (map<string,bufferptr>::iterator p = in->xattrs.begin(); | |
10937 | p != in->xattrs.end(); | |
10938 | ++p) | |
10939 | r += p->first.length() + 1; | |
10940 | ||
10941 | const VXattr *vxattrs = _get_vxattrs(in); | |
10942 | r += _vxattrs_name_size(vxattrs); | |
10943 | ||
10944 | if (size != 0) { | |
10945 | if (size >= (unsigned)r) { | |
10946 | for (map<string,bufferptr>::iterator p = in->xattrs.begin(); | |
10947 | p != in->xattrs.end(); | |
10948 | ++p) { | |
10949 | memcpy(name, p->first.c_str(), p->first.length()); | |
10950 | name += p->first.length(); | |
10951 | *name = '\0'; | |
10952 | name++; | |
10953 | } | |
10954 | if (vxattrs) { | |
10955 | for (int i = 0; !vxattrs[i].name.empty(); i++) { | |
10956 | const VXattr& vxattr = vxattrs[i]; | |
10957 | if (vxattr.hidden) | |
10958 | continue; | |
10959 | // call pointer-to-member function | |
10960 | if(vxattr.exists_cb && !(this->*(vxattr.exists_cb))(in)) | |
10961 | continue; | |
10962 | memcpy(name, vxattr.name.c_str(), vxattr.name.length()); | |
10963 | name += vxattr.name.length(); | |
10964 | *name = '\0'; | |
10965 | name++; | |
10966 | } | |
10967 | } | |
10968 | } else | |
10969 | r = -ERANGE; | |
10970 | } | |
10971 | } | |
10972 | ldout(cct, 3) << "_listxattr(" << in->ino << ", " << size << ") = " << r << dendl; | |
10973 | return r; | |
10974 | } | |
10975 | ||
10976 | int Client::ll_listxattr(Inode *in, char *names, size_t size, | |
10977 | const UserPerm& perms) | |
10978 | { | |
10979 | Mutex::Locker lock(client_lock); | |
10980 | ||
181888fb FG |
10981 | if (unmounting) |
10982 | return -ENOTCONN; | |
10983 | ||
7c673cae FG |
10984 | vinodeno_t vino = _get_vino(in); |
10985 | ||
10986 | ldout(cct, 3) << "ll_listxattr " << vino << " size " << size << dendl; | |
10987 | tout(cct) << "ll_listxattr" << std::endl; | |
10988 | tout(cct) << vino.ino.val << std::endl; | |
10989 | tout(cct) << size << std::endl; | |
10990 | ||
10991 | return _listxattr(in, names, size, perms); | |
10992 | } | |
10993 | ||
10994 | int Client::_do_setxattr(Inode *in, const char *name, const void *value, | |
10995 | size_t size, int flags, const UserPerm& perms) | |
10996 | { | |
10997 | ||
10998 | int xattr_flags = 0; | |
10999 | if (!value) | |
11000 | xattr_flags |= CEPH_XATTR_REMOVE; | |
11001 | if (flags & XATTR_CREATE) | |
11002 | xattr_flags |= CEPH_XATTR_CREATE; | |
11003 | if (flags & XATTR_REPLACE) | |
11004 | xattr_flags |= CEPH_XATTR_REPLACE; | |
11005 | ||
11006 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETXATTR); | |
11007 | filepath path; | |
11008 | in->make_nosnap_relative_path(path); | |
11009 | req->set_filepath(path); | |
11010 | req->set_string2(name); | |
11011 | req->set_inode(in); | |
11012 | req->head.args.setxattr.flags = xattr_flags; | |
11013 | ||
11014 | bufferlist bl; | |
11015 | bl.append((const char*)value, size); | |
11016 | req->set_data(bl); | |
11017 | ||
11018 | int res = make_request(req, perms); | |
11019 | ||
11020 | trim_cache(); | |
11021 | ldout(cct, 3) << "_setxattr(" << in->ino << ", \"" << name << "\") = " << | |
11022 | res << dendl; | |
11023 | return res; | |
11024 | } | |
11025 | ||
11026 | int Client::_setxattr(Inode *in, const char *name, const void *value, | |
11027 | size_t size, int flags, const UserPerm& perms) | |
11028 | { | |
11029 | if (in->snapid != CEPH_NOSNAP) { | |
11030 | return -EROFS; | |
11031 | } | |
11032 | ||
11033 | bool posix_acl_xattr = false; | |
11034 | if (acl_type == POSIX_ACL) | |
11035 | posix_acl_xattr = !strncmp(name, "system.", 7); | |
11036 | ||
11037 | if (strncmp(name, "user.", 5) && | |
11038 | strncmp(name, "security.", 9) && | |
11039 | strncmp(name, "trusted.", 8) && | |
11040 | strncmp(name, "ceph.", 5) && | |
11041 | !posix_acl_xattr) | |
11042 | return -EOPNOTSUPP; | |
11043 | ||
11044 | if (posix_acl_xattr) { | |
11045 | if (!strcmp(name, ACL_EA_ACCESS)) { | |
11046 | mode_t new_mode = in->mode; | |
11047 | if (value) { | |
11048 | int ret = posix_acl_equiv_mode(value, size, &new_mode); | |
11049 | if (ret < 0) | |
11050 | return ret; | |
11051 | if (ret == 0) { | |
11052 | value = NULL; | |
11053 | size = 0; | |
11054 | } | |
11055 | if (new_mode != in->mode) { | |
11056 | struct ceph_statx stx; | |
11057 | stx.stx_mode = new_mode; | |
11058 | ret = _do_setattr(in, &stx, CEPH_SETATTR_MODE, perms, NULL); | |
11059 | if (ret < 0) | |
11060 | return ret; | |
11061 | } | |
11062 | } | |
11063 | } else if (!strcmp(name, ACL_EA_DEFAULT)) { | |
11064 | if (value) { | |
11065 | if (!S_ISDIR(in->mode)) | |
11066 | return -EACCES; | |
11067 | int ret = posix_acl_check(value, size); | |
11068 | if (ret < 0) | |
11069 | return -EINVAL; | |
11070 | if (ret == 0) { | |
11071 | value = NULL; | |
11072 | size = 0; | |
11073 | } | |
11074 | } | |
11075 | } else { | |
11076 | return -EOPNOTSUPP; | |
11077 | } | |
11078 | } else { | |
11079 | const VXattr *vxattr = _match_vxattr(in, name); | |
11080 | if (vxattr && vxattr->readonly) | |
11081 | return -EOPNOTSUPP; | |
11082 | } | |
11083 | ||
11084 | return _do_setxattr(in, name, value, size, flags, perms); | |
11085 | } | |
11086 | ||
11087 | int Client::_setxattr(InodeRef &in, const char *name, const void *value, | |
11088 | size_t size, int flags, const UserPerm& perms) | |
11089 | { | |
11090 | if (cct->_conf->client_permissions) { | |
11091 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
11092 | if (r < 0) | |
11093 | return r; | |
11094 | } | |
11095 | return _setxattr(in.get(), name, value, size, flags, perms); | |
11096 | } | |
11097 | ||
11098 | int Client::_setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap) | |
11099 | { | |
11100 | string tmp; | |
11101 | if (name == "layout") { | |
11102 | string::iterator begin = value.begin(); | |
11103 | string::iterator end = value.end(); | |
11104 | keys_and_values<string::iterator> p; // create instance of parser | |
11105 | std::map<string, string> m; // map to receive results | |
11106 | if (!qi::parse(begin, end, p, m)) { // returns true if successful | |
11107 | return -EINVAL; | |
11108 | } | |
11109 | if (begin != end) | |
11110 | return -EINVAL; | |
11111 | for (map<string,string>::iterator q = m.begin(); q != m.end(); ++q) { | |
11112 | if (q->first == "pool") { | |
11113 | tmp = q->second; | |
11114 | break; | |
11115 | } | |
11116 | } | |
11117 | } else if (name == "layout.pool") { | |
11118 | tmp = value; | |
11119 | } | |
11120 | ||
11121 | if (tmp.length()) { | |
11122 | int64_t pool; | |
11123 | try { | |
11124 | pool = boost::lexical_cast<unsigned>(tmp); | |
11125 | if (!osdmap->have_pg_pool(pool)) | |
11126 | return -ENOENT; | |
11127 | } catch (boost::bad_lexical_cast const&) { | |
11128 | pool = osdmap->lookup_pg_pool_name(tmp); | |
11129 | if (pool < 0) { | |
11130 | return -ENOENT; | |
11131 | } | |
11132 | } | |
11133 | } | |
11134 | ||
11135 | return 0; | |
11136 | } | |
11137 | ||
11138 | void Client::_setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t size) | |
11139 | { | |
11140 | // For setting pool of layout, MetaRequest need osdmap epoch. | |
11141 | // There is a race which create a new data pool but client and mds both don't have. | |
11142 | // Make client got the latest osdmap which make mds quickly judge whether get newer osdmap. | |
11143 | if (strcmp(name, "ceph.file.layout.pool") == 0 || strcmp(name, "ceph.dir.layout.pool") == 0 || | |
11144 | strcmp(name, "ceph.file.layout") == 0 || strcmp(name, "ceph.dir.layout") == 0) { | |
11145 | string rest(strstr(name, "layout")); | |
11146 | string v((const char*)value, size); | |
11147 | int r = objecter->with_osdmap([&](const OSDMap& o) { | |
11148 | return _setxattr_check_data_pool(rest, v, &o); | |
11149 | }); | |
11150 | ||
11151 | if (r == -ENOENT) { | |
11152 | C_SaferCond ctx; | |
11153 | objecter->wait_for_latest_osdmap(&ctx); | |
11154 | ctx.wait(); | |
11155 | } | |
11156 | } | |
11157 | } | |
11158 | ||
11159 | int Client::ll_setxattr(Inode *in, const char *name, const void *value, | |
11160 | size_t size, int flags, const UserPerm& perms) | |
11161 | { | |
11162 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
11163 | ||
11164 | Mutex::Locker lock(client_lock); | |
11165 | ||
181888fb FG |
11166 | if (unmounting) |
11167 | return -ENOTCONN; | |
11168 | ||
7c673cae FG |
11169 | vinodeno_t vino = _get_vino(in); |
11170 | ||
11171 | ldout(cct, 3) << "ll_setxattr " << vino << " " << name << " size " << size << dendl; | |
11172 | tout(cct) << "ll_setxattr" << std::endl; | |
11173 | tout(cct) << vino.ino.val << std::endl; | |
11174 | tout(cct) << name << std::endl; | |
11175 | ||
11176 | if (!cct->_conf->fuse_default_permissions) { | |
11177 | int r = xattr_permission(in, name, MAY_WRITE, perms); | |
11178 | if (r < 0) | |
11179 | return r; | |
11180 | } | |
11181 | return _setxattr(in, name, value, size, flags, perms); | |
11182 | } | |
11183 | ||
11184 | int Client::_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
11185 | { | |
11186 | if (in->snapid != CEPH_NOSNAP) { | |
11187 | return -EROFS; | |
11188 | } | |
11189 | ||
11190 | // same xattrs supported by kernel client | |
11191 | if (strncmp(name, "user.", 5) && | |
11192 | strncmp(name, "system.", 7) && | |
11193 | strncmp(name, "security.", 9) && | |
11194 | strncmp(name, "trusted.", 8) && | |
11195 | strncmp(name, "ceph.", 5)) | |
11196 | return -EOPNOTSUPP; | |
11197 | ||
11198 | const VXattr *vxattr = _match_vxattr(in, name); | |
11199 | if (vxattr && vxattr->readonly) | |
11200 | return -EOPNOTSUPP; | |
11201 | ||
11202 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_RMXATTR); | |
11203 | filepath path; | |
11204 | in->make_nosnap_relative_path(path); | |
11205 | req->set_filepath(path); | |
11206 | req->set_filepath2(name); | |
11207 | req->set_inode(in); | |
11208 | ||
11209 | int res = make_request(req, perms); | |
11210 | ||
11211 | trim_cache(); | |
11212 | ldout(cct, 3) << "_removexattr(" << in->ino << ", \"" << name << "\") = " << res << dendl; | |
11213 | return res; | |
11214 | } | |
11215 | ||
11216 | int Client::_removexattr(InodeRef &in, const char *name, const UserPerm& perms) | |
11217 | { | |
11218 | if (cct->_conf->client_permissions) { | |
11219 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
11220 | if (r < 0) | |
11221 | return r; | |
11222 | } | |
11223 | return _removexattr(in.get(), name, perms); | |
11224 | } | |
11225 | ||
11226 | int Client::ll_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
11227 | { | |
11228 | Mutex::Locker lock(client_lock); | |
11229 | ||
181888fb FG |
11230 | if (unmounting) |
11231 | return -ENOTCONN; | |
11232 | ||
7c673cae FG |
11233 | vinodeno_t vino = _get_vino(in); |
11234 | ||
11235 | ldout(cct, 3) << "ll_removexattr " << vino << " " << name << dendl; | |
11236 | tout(cct) << "ll_removexattr" << std::endl; | |
11237 | tout(cct) << vino.ino.val << std::endl; | |
11238 | tout(cct) << name << std::endl; | |
11239 | ||
11240 | if (!cct->_conf->fuse_default_permissions) { | |
11241 | int r = xattr_permission(in, name, MAY_WRITE, perms); | |
11242 | if (r < 0) | |
11243 | return r; | |
11244 | } | |
11245 | ||
11246 | return _removexattr(in, name, perms); | |
11247 | } | |
11248 | ||
11249 | bool Client::_vxattrcb_quota_exists(Inode *in) | |
11250 | { | |
11251 | return in->quota.is_enable(); | |
11252 | } | |
11253 | size_t Client::_vxattrcb_quota(Inode *in, char *val, size_t size) | |
11254 | { | |
11255 | return snprintf(val, size, | |
11256 | "max_bytes=%lld max_files=%lld", | |
11257 | (long long int)in->quota.max_bytes, | |
11258 | (long long int)in->quota.max_files); | |
11259 | } | |
11260 | size_t Client::_vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size) | |
11261 | { | |
11262 | return snprintf(val, size, "%lld", (long long int)in->quota.max_bytes); | |
11263 | } | |
11264 | size_t Client::_vxattrcb_quota_max_files(Inode *in, char *val, size_t size) | |
11265 | { | |
11266 | return snprintf(val, size, "%lld", (long long int)in->quota.max_files); | |
11267 | } | |
11268 | ||
11269 | bool Client::_vxattrcb_layout_exists(Inode *in) | |
11270 | { | |
11271 | return in->layout != file_layout_t(); | |
11272 | } | |
11273 | size_t Client::_vxattrcb_layout(Inode *in, char *val, size_t size) | |
11274 | { | |
11275 | int r = snprintf(val, size, | |
11276 | "stripe_unit=%lld stripe_count=%lld object_size=%lld pool=", | |
11277 | (unsigned long long)in->layout.stripe_unit, | |
11278 | (unsigned long long)in->layout.stripe_count, | |
11279 | (unsigned long long)in->layout.object_size); | |
11280 | objecter->with_osdmap([&](const OSDMap& o) { | |
11281 | if (o.have_pg_pool(in->layout.pool_id)) | |
11282 | r += snprintf(val + r, size - r, "%s", | |
11283 | o.get_pool_name(in->layout.pool_id).c_str()); | |
11284 | else | |
11285 | r += snprintf(val + r, size - r, "%" PRIu64, | |
11286 | (uint64_t)in->layout.pool_id); | |
11287 | }); | |
11288 | if (in->layout.pool_ns.length()) | |
11289 | r += snprintf(val + r, size - r, " pool_namespace=%s", | |
11290 | in->layout.pool_ns.c_str()); | |
11291 | return r; | |
11292 | } | |
11293 | size_t Client::_vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size) | |
11294 | { | |
11295 | return snprintf(val, size, "%lld", (unsigned long long)in->layout.stripe_unit); | |
11296 | } | |
11297 | size_t Client::_vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size) | |
11298 | { | |
11299 | return snprintf(val, size, "%lld", (unsigned long long)in->layout.stripe_count); | |
11300 | } | |
11301 | size_t Client::_vxattrcb_layout_object_size(Inode *in, char *val, size_t size) | |
11302 | { | |
11303 | return snprintf(val, size, "%lld", (unsigned long long)in->layout.object_size); | |
11304 | } | |
11305 | size_t Client::_vxattrcb_layout_pool(Inode *in, char *val, size_t size) | |
11306 | { | |
11307 | size_t r; | |
11308 | objecter->with_osdmap([&](const OSDMap& o) { | |
11309 | if (o.have_pg_pool(in->layout.pool_id)) | |
11310 | r = snprintf(val, size, "%s", o.get_pool_name( | |
11311 | in->layout.pool_id).c_str()); | |
11312 | else | |
11313 | r = snprintf(val, size, "%" PRIu64, (uint64_t)in->layout.pool_id); | |
11314 | }); | |
11315 | return r; | |
11316 | } | |
11317 | size_t Client::_vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size) | |
11318 | { | |
11319 | return snprintf(val, size, "%s", in->layout.pool_ns.c_str()); | |
11320 | } | |
11321 | size_t Client::_vxattrcb_dir_entries(Inode *in, char *val, size_t size) | |
11322 | { | |
11323 | return snprintf(val, size, "%lld", (unsigned long long)(in->dirstat.nfiles + in->dirstat.nsubdirs)); | |
11324 | } | |
11325 | size_t Client::_vxattrcb_dir_files(Inode *in, char *val, size_t size) | |
11326 | { | |
11327 | return snprintf(val, size, "%lld", (unsigned long long)in->dirstat.nfiles); | |
11328 | } | |
11329 | size_t Client::_vxattrcb_dir_subdirs(Inode *in, char *val, size_t size) | |
11330 | { | |
11331 | return snprintf(val, size, "%lld", (unsigned long long)in->dirstat.nsubdirs); | |
11332 | } | |
11333 | size_t Client::_vxattrcb_dir_rentries(Inode *in, char *val, size_t size) | |
11334 | { | |
11335 | return snprintf(val, size, "%lld", (unsigned long long)(in->rstat.rfiles + in->rstat.rsubdirs)); | |
11336 | } | |
11337 | size_t Client::_vxattrcb_dir_rfiles(Inode *in, char *val, size_t size) | |
11338 | { | |
11339 | return snprintf(val, size, "%lld", (unsigned long long)in->rstat.rfiles); | |
11340 | } | |
11341 | size_t Client::_vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size) | |
11342 | { | |
11343 | return snprintf(val, size, "%lld", (unsigned long long)in->rstat.rsubdirs); | |
11344 | } | |
11345 | size_t Client::_vxattrcb_dir_rbytes(Inode *in, char *val, size_t size) | |
11346 | { | |
11347 | return snprintf(val, size, "%lld", (unsigned long long)in->rstat.rbytes); | |
11348 | } | |
11349 | size_t Client::_vxattrcb_dir_rctime(Inode *in, char *val, size_t size) | |
11350 | { | |
11351 | return snprintf(val, size, "%ld.09%ld", (long)in->rstat.rctime.sec(), | |
11352 | (long)in->rstat.rctime.nsec()); | |
11353 | } | |
11354 | ||
11355 | #define CEPH_XATTR_NAME(_type, _name) "ceph." #_type "." #_name | |
11356 | #define CEPH_XATTR_NAME2(_type, _name, _name2) "ceph." #_type "." #_name "." #_name2 | |
11357 | ||
11358 | #define XATTR_NAME_CEPH(_type, _name) \ | |
11359 | { \ | |
11360 | name: CEPH_XATTR_NAME(_type, _name), \ | |
11361 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
11362 | readonly: true, \ | |
11363 | hidden: false, \ | |
11364 | exists_cb: NULL, \ | |
11365 | } | |
11366 | #define XATTR_LAYOUT_FIELD(_type, _name, _field) \ | |
11367 | { \ | |
11368 | name: CEPH_XATTR_NAME2(_type, _name, _field), \ | |
11369 | getxattr_cb: &Client::_vxattrcb_ ## _name ## _ ## _field, \ | |
11370 | readonly: false, \ | |
11371 | hidden: true, \ | |
11372 | exists_cb: &Client::_vxattrcb_layout_exists, \ | |
11373 | } | |
11374 | #define XATTR_QUOTA_FIELD(_type, _name) \ | |
11375 | { \ | |
11376 | name: CEPH_XATTR_NAME(_type, _name), \ | |
11377 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
11378 | readonly: false, \ | |
11379 | hidden: true, \ | |
11380 | exists_cb: &Client::_vxattrcb_quota_exists, \ | |
11381 | } | |
11382 | ||
11383 | const Client::VXattr Client::_dir_vxattrs[] = { | |
11384 | { | |
11385 | name: "ceph.dir.layout", | |
11386 | getxattr_cb: &Client::_vxattrcb_layout, | |
11387 | readonly: false, | |
11388 | hidden: true, | |
11389 | exists_cb: &Client::_vxattrcb_layout_exists, | |
11390 | }, | |
11391 | XATTR_LAYOUT_FIELD(dir, layout, stripe_unit), | |
11392 | XATTR_LAYOUT_FIELD(dir, layout, stripe_count), | |
11393 | XATTR_LAYOUT_FIELD(dir, layout, object_size), | |
11394 | XATTR_LAYOUT_FIELD(dir, layout, pool), | |
11395 | XATTR_LAYOUT_FIELD(dir, layout, pool_namespace), | |
11396 | XATTR_NAME_CEPH(dir, entries), | |
11397 | XATTR_NAME_CEPH(dir, files), | |
11398 | XATTR_NAME_CEPH(dir, subdirs), | |
11399 | XATTR_NAME_CEPH(dir, rentries), | |
11400 | XATTR_NAME_CEPH(dir, rfiles), | |
11401 | XATTR_NAME_CEPH(dir, rsubdirs), | |
11402 | XATTR_NAME_CEPH(dir, rbytes), | |
11403 | XATTR_NAME_CEPH(dir, rctime), | |
11404 | { | |
11405 | name: "ceph.quota", | |
11406 | getxattr_cb: &Client::_vxattrcb_quota, | |
11407 | readonly: false, | |
11408 | hidden: true, | |
11409 | exists_cb: &Client::_vxattrcb_quota_exists, | |
11410 | }, | |
11411 | XATTR_QUOTA_FIELD(quota, max_bytes), | |
11412 | XATTR_QUOTA_FIELD(quota, max_files), | |
11413 | { name: "" } /* Required table terminator */ | |
11414 | }; | |
11415 | ||
11416 | const Client::VXattr Client::_file_vxattrs[] = { | |
11417 | { | |
11418 | name: "ceph.file.layout", | |
11419 | getxattr_cb: &Client::_vxattrcb_layout, | |
11420 | readonly: false, | |
11421 | hidden: true, | |
11422 | exists_cb: &Client::_vxattrcb_layout_exists, | |
11423 | }, | |
11424 | XATTR_LAYOUT_FIELD(file, layout, stripe_unit), | |
11425 | XATTR_LAYOUT_FIELD(file, layout, stripe_count), | |
11426 | XATTR_LAYOUT_FIELD(file, layout, object_size), | |
11427 | XATTR_LAYOUT_FIELD(file, layout, pool), | |
11428 | XATTR_LAYOUT_FIELD(file, layout, pool_namespace), | |
11429 | { name: "" } /* Required table terminator */ | |
11430 | }; | |
11431 | ||
11432 | const Client::VXattr *Client::_get_vxattrs(Inode *in) | |
11433 | { | |
11434 | if (in->is_dir()) | |
11435 | return _dir_vxattrs; | |
11436 | else if (in->is_file()) | |
11437 | return _file_vxattrs; | |
11438 | return NULL; | |
11439 | } | |
11440 | ||
11441 | const Client::VXattr *Client::_match_vxattr(Inode *in, const char *name) | |
11442 | { | |
11443 | if (strncmp(name, "ceph.", 5) == 0) { | |
11444 | const VXattr *vxattr = _get_vxattrs(in); | |
11445 | if (vxattr) { | |
11446 | while (!vxattr->name.empty()) { | |
11447 | if (vxattr->name == name) | |
11448 | return vxattr; | |
11449 | vxattr++; | |
11450 | } | |
11451 | } | |
11452 | } | |
11453 | return NULL; | |
11454 | } | |
11455 | ||
11456 | size_t Client::_vxattrs_calcu_name_size(const VXattr *vxattr) | |
11457 | { | |
11458 | size_t len = 0; | |
11459 | while (!vxattr->name.empty()) { | |
11460 | if (!vxattr->hidden) | |
11461 | len += vxattr->name.length() + 1; | |
11462 | vxattr++; | |
11463 | } | |
11464 | return len; | |
11465 | } | |
11466 | ||
11467 | int Client::ll_readlink(Inode *in, char *buf, size_t buflen, const UserPerm& perms) | |
11468 | { | |
11469 | Mutex::Locker lock(client_lock); | |
11470 | ||
181888fb FG |
11471 | if (unmounting) |
11472 | return -ENOTCONN; | |
11473 | ||
7c673cae FG |
11474 | vinodeno_t vino = _get_vino(in); |
11475 | ||
11476 | ldout(cct, 3) << "ll_readlink " << vino << dendl; | |
11477 | tout(cct) << "ll_readlink" << std::endl; | |
11478 | tout(cct) << vino.ino.val << std::endl; | |
11479 | ||
11480 | set<Dentry*>::iterator dn = in->dn_set.begin(); | |
11481 | while (dn != in->dn_set.end()) { | |
11482 | touch_dn(*dn); | |
11483 | ++dn; | |
11484 | } | |
11485 | ||
11486 | int r = _readlink(in, buf, buflen); // FIXME: no permission checking! | |
11487 | ldout(cct, 3) << "ll_readlink " << vino << " = " << r << dendl; | |
11488 | return r; | |
11489 | } | |
11490 | ||
11491 | int Client::_mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
11492 | const UserPerm& perms, InodeRef *inp) | |
11493 | { | |
11494 | ldout(cct, 3) << "_mknod(" << dir->ino << " " << name << ", 0" << oct | |
11495 | << mode << dec << ", " << rdev << ", uid " << perms.uid() | |
11496 | << ", gid " << perms.gid() << ")" << dendl; | |
11497 | ||
11498 | if (strlen(name) > NAME_MAX) | |
11499 | return -ENAMETOOLONG; | |
11500 | ||
11501 | if (dir->snapid != CEPH_NOSNAP) { | |
11502 | return -EROFS; | |
11503 | } | |
11504 | if (is_quota_files_exceeded(dir, perms)) { | |
11505 | return -EDQUOT; | |
11506 | } | |
11507 | ||
11508 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_MKNOD); | |
11509 | ||
11510 | filepath path; | |
11511 | dir->make_nosnap_relative_path(path); | |
11512 | path.push_dentry(name); | |
11513 | req->set_filepath(path); | |
11514 | req->set_inode(dir); | |
11515 | req->head.args.mknod.rdev = rdev; | |
11516 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
11517 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
11518 | ||
11519 | bufferlist xattrs_bl; | |
11520 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
11521 | if (res < 0) | |
11522 | goto fail; | |
11523 | req->head.args.mknod.mode = mode; | |
11524 | if (xattrs_bl.length() > 0) | |
11525 | req->set_data(xattrs_bl); | |
11526 | ||
11527 | Dentry *de; | |
11528 | res = get_or_create(dir, name, &de); | |
11529 | if (res < 0) | |
11530 | goto fail; | |
11531 | req->set_dentry(de); | |
11532 | ||
11533 | res = make_request(req, perms, inp); | |
11534 | ||
11535 | trim_cache(); | |
11536 | ||
11537 | ldout(cct, 3) << "mknod(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; | |
11538 | return res; | |
11539 | ||
11540 | fail: | |
11541 | put_request(req); | |
11542 | return res; | |
11543 | } | |
11544 | ||
11545 | int Client::ll_mknod(Inode *parent, const char *name, mode_t mode, | |
11546 | dev_t rdev, struct stat *attr, Inode **out, | |
11547 | const UserPerm& perms) | |
11548 | { | |
11549 | Mutex::Locker lock(client_lock); | |
11550 | ||
181888fb FG |
11551 | if (unmounting) |
11552 | return -ENOTCONN; | |
11553 | ||
7c673cae FG |
11554 | vinodeno_t vparent = _get_vino(parent); |
11555 | ||
11556 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name << dendl; | |
11557 | tout(cct) << "ll_mknod" << std::endl; | |
11558 | tout(cct) << vparent.ino.val << std::endl; | |
11559 | tout(cct) << name << std::endl; | |
11560 | tout(cct) << mode << std::endl; | |
11561 | tout(cct) << rdev << std::endl; | |
11562 | ||
11563 | if (!cct->_conf->fuse_default_permissions) { | |
11564 | int r = may_create(parent, perms); | |
11565 | if (r < 0) | |
11566 | return r; | |
11567 | } | |
11568 | ||
11569 | InodeRef in; | |
11570 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
11571 | if (r == 0) { | |
11572 | fill_stat(in, attr); | |
11573 | _ll_get(in.get()); | |
11574 | } | |
11575 | tout(cct) << attr->st_ino << std::endl; | |
11576 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name | |
11577 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
11578 | *out = in.get(); | |
11579 | return r; | |
11580 | } | |
11581 | ||
11582 | int Client::ll_mknodx(Inode *parent, const char *name, mode_t mode, | |
11583 | dev_t rdev, Inode **out, | |
11584 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
11585 | const UserPerm& perms) | |
11586 | { | |
11587 | unsigned caps = statx_to_mask(flags, want); | |
11588 | Mutex::Locker lock(client_lock); | |
11589 | ||
181888fb FG |
11590 | if (unmounting) |
11591 | return -ENOTCONN; | |
11592 | ||
7c673cae FG |
11593 | vinodeno_t vparent = _get_vino(parent); |
11594 | ||
11595 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name << dendl; | |
11596 | tout(cct) << "ll_mknodx" << std::endl; | |
11597 | tout(cct) << vparent.ino.val << std::endl; | |
11598 | tout(cct) << name << std::endl; | |
11599 | tout(cct) << mode << std::endl; | |
11600 | tout(cct) << rdev << std::endl; | |
11601 | ||
11602 | if (!cct->_conf->fuse_default_permissions) { | |
11603 | int r = may_create(parent, perms); | |
11604 | if (r < 0) | |
11605 | return r; | |
11606 | } | |
11607 | ||
11608 | InodeRef in; | |
11609 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
11610 | if (r == 0) { | |
11611 | fill_statx(in, caps, stx); | |
11612 | _ll_get(in.get()); | |
11613 | } | |
11614 | tout(cct) << stx->stx_ino << std::endl; | |
11615 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name | |
11616 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
11617 | *out = in.get(); | |
11618 | return r; | |
11619 | } | |
11620 | ||
11621 | int Client::_create(Inode *dir, const char *name, int flags, mode_t mode, | |
11622 | InodeRef *inp, Fh **fhp, int stripe_unit, int stripe_count, | |
11623 | int object_size, const char *data_pool, bool *created, | |
11624 | const UserPerm& perms) | |
11625 | { | |
11626 | ldout(cct, 3) << "_create(" << dir->ino << " " << name << ", 0" << oct << | |
11627 | mode << dec << ")" << dendl; | |
11628 | ||
11629 | if (strlen(name) > NAME_MAX) | |
11630 | return -ENAMETOOLONG; | |
11631 | if (dir->snapid != CEPH_NOSNAP) { | |
11632 | return -EROFS; | |
11633 | } | |
11634 | if (is_quota_files_exceeded(dir, perms)) { | |
11635 | return -EDQUOT; | |
11636 | } | |
11637 | ||
11638 | // use normalized flags to generate cmode | |
11639 | int cmode = ceph_flags_to_mode(ceph_flags_sys2wire(flags)); | |
11640 | if (cmode < 0) | |
11641 | return -EINVAL; | |
11642 | ||
11643 | int64_t pool_id = -1; | |
11644 | if (data_pool && *data_pool) { | |
11645 | pool_id = objecter->with_osdmap( | |
11646 | std::mem_fn(&OSDMap::lookup_pg_pool_name), data_pool); | |
11647 | if (pool_id < 0) | |
11648 | return -EINVAL; | |
11649 | if (pool_id > 0xffffffffll) | |
11650 | return -ERANGE; // bummer! | |
11651 | } | |
11652 | ||
11653 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_CREATE); | |
11654 | ||
11655 | filepath path; | |
11656 | dir->make_nosnap_relative_path(path); | |
11657 | path.push_dentry(name); | |
11658 | req->set_filepath(path); | |
11659 | req->set_inode(dir); | |
11660 | req->head.args.open.flags = ceph_flags_sys2wire(flags | O_CREAT); | |
11661 | ||
11662 | req->head.args.open.stripe_unit = stripe_unit; | |
11663 | req->head.args.open.stripe_count = stripe_count; | |
11664 | req->head.args.open.object_size = object_size; | |
11665 | if (cct->_conf->client_debug_getattr_caps) | |
11666 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
11667 | else | |
11668 | req->head.args.open.mask = 0; | |
11669 | req->head.args.open.pool = pool_id; | |
11670 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
11671 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
11672 | ||
11673 | mode |= S_IFREG; | |
11674 | bufferlist xattrs_bl; | |
11675 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
11676 | if (res < 0) | |
11677 | goto fail; | |
11678 | req->head.args.open.mode = mode; | |
11679 | if (xattrs_bl.length() > 0) | |
11680 | req->set_data(xattrs_bl); | |
11681 | ||
11682 | Dentry *de; | |
11683 | res = get_or_create(dir, name, &de); | |
11684 | if (res < 0) | |
11685 | goto fail; | |
11686 | req->set_dentry(de); | |
11687 | ||
11688 | res = make_request(req, perms, inp, created); | |
11689 | if (res < 0) { | |
11690 | goto reply_error; | |
11691 | } | |
11692 | ||
11693 | /* If the caller passed a value in fhp, do the open */ | |
11694 | if(fhp) { | |
11695 | (*inp)->get_open_ref(cmode); | |
11696 | *fhp = _create_fh(inp->get(), flags, cmode, perms); | |
11697 | } | |
11698 | ||
11699 | reply_error: | |
11700 | trim_cache(); | |
11701 | ||
11702 | ldout(cct, 3) << "create(" << path << ", 0" << oct << mode << dec | |
11703 | << " layout " << stripe_unit | |
11704 | << ' ' << stripe_count | |
11705 | << ' ' << object_size | |
11706 | <<") = " << res << dendl; | |
11707 | return res; | |
11708 | ||
11709 | fail: | |
11710 | put_request(req); | |
11711 | return res; | |
11712 | } | |
11713 | ||
11714 | ||
11715 | int Client::_mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, | |
11716 | InodeRef *inp) | |
11717 | { | |
11718 | ldout(cct, 3) << "_mkdir(" << dir->ino << " " << name << ", 0" << oct | |
11719 | << mode << dec << ", uid " << perm.uid() | |
11720 | << ", gid " << perm.gid() << ")" << dendl; | |
11721 | ||
11722 | if (strlen(name) > NAME_MAX) | |
11723 | return -ENAMETOOLONG; | |
11724 | ||
11725 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
11726 | return -EROFS; | |
11727 | } | |
11728 | if (is_quota_files_exceeded(dir, perm)) { | |
11729 | return -EDQUOT; | |
11730 | } | |
11731 | MetaRequest *req = new MetaRequest(dir->snapid == CEPH_SNAPDIR ? | |
11732 | CEPH_MDS_OP_MKSNAP : CEPH_MDS_OP_MKDIR); | |
11733 | ||
11734 | filepath path; | |
11735 | dir->make_nosnap_relative_path(path); | |
11736 | path.push_dentry(name); | |
11737 | req->set_filepath(path); | |
11738 | req->set_inode(dir); | |
11739 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
11740 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
11741 | ||
11742 | mode |= S_IFDIR; | |
11743 | bufferlist xattrs_bl; | |
11744 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perm); | |
11745 | if (res < 0) | |
11746 | goto fail; | |
11747 | req->head.args.mkdir.mode = mode; | |
11748 | if (xattrs_bl.length() > 0) | |
11749 | req->set_data(xattrs_bl); | |
11750 | ||
11751 | Dentry *de; | |
11752 | res = get_or_create(dir, name, &de); | |
11753 | if (res < 0) | |
11754 | goto fail; | |
11755 | req->set_dentry(de); | |
11756 | ||
11757 | ldout(cct, 10) << "_mkdir: making request" << dendl; | |
11758 | res = make_request(req, perm, inp); | |
11759 | ldout(cct, 10) << "_mkdir result is " << res << dendl; | |
11760 | ||
11761 | trim_cache(); | |
11762 | ||
11763 | ldout(cct, 3) << "_mkdir(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; | |
11764 | return res; | |
11765 | ||
11766 | fail: | |
11767 | put_request(req); | |
11768 | return res; | |
11769 | } | |
11770 | ||
11771 | int Client::ll_mkdir(Inode *parent, const char *name, mode_t mode, | |
11772 | struct stat *attr, Inode **out, const UserPerm& perm) | |
11773 | { | |
11774 | Mutex::Locker lock(client_lock); | |
11775 | ||
181888fb FG |
11776 | if (unmounting) |
11777 | return -ENOTCONN; | |
11778 | ||
7c673cae FG |
11779 | vinodeno_t vparent = _get_vino(parent); |
11780 | ||
11781 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name << dendl; | |
11782 | tout(cct) << "ll_mkdir" << std::endl; | |
11783 | tout(cct) << vparent.ino.val << std::endl; | |
11784 | tout(cct) << name << std::endl; | |
11785 | tout(cct) << mode << std::endl; | |
11786 | ||
11787 | if (!cct->_conf->fuse_default_permissions) { | |
11788 | int r = may_create(parent, perm); | |
11789 | if (r < 0) | |
11790 | return r; | |
11791 | } | |
11792 | ||
11793 | InodeRef in; | |
11794 | int r = _mkdir(parent, name, mode, perm, &in); | |
11795 | if (r == 0) { | |
11796 | fill_stat(in, attr); | |
11797 | _ll_get(in.get()); | |
11798 | } | |
11799 | tout(cct) << attr->st_ino << std::endl; | |
11800 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name | |
11801 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
11802 | *out = in.get(); | |
11803 | return r; | |
11804 | } | |
11805 | ||
11806 | int Client::ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
11807 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
11808 | const UserPerm& perms) | |
11809 | { | |
11810 | Mutex::Locker lock(client_lock); | |
11811 | ||
181888fb FG |
11812 | if (unmounting) |
11813 | return -ENOTCONN; | |
11814 | ||
7c673cae FG |
11815 | vinodeno_t vparent = _get_vino(parent); |
11816 | ||
11817 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name << dendl; | |
11818 | tout(cct) << "ll_mkdirx" << std::endl; | |
11819 | tout(cct) << vparent.ino.val << std::endl; | |
11820 | tout(cct) << name << std::endl; | |
11821 | tout(cct) << mode << std::endl; | |
11822 | ||
11823 | if (!cct->_conf->fuse_default_permissions) { | |
11824 | int r = may_create(parent, perms); | |
11825 | if (r < 0) | |
11826 | return r; | |
11827 | } | |
11828 | ||
11829 | InodeRef in; | |
11830 | int r = _mkdir(parent, name, mode, perms, &in); | |
11831 | if (r == 0) { | |
11832 | fill_statx(in, statx_to_mask(flags, want), stx); | |
11833 | _ll_get(in.get()); | |
11834 | } else { | |
11835 | stx->stx_ino = 0; | |
11836 | stx->stx_mask = 0; | |
11837 | } | |
11838 | tout(cct) << stx->stx_ino << std::endl; | |
11839 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name | |
11840 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
11841 | *out = in.get(); | |
11842 | return r; | |
11843 | } | |
11844 | ||
11845 | int Client::_symlink(Inode *dir, const char *name, const char *target, | |
11846 | const UserPerm& perms, InodeRef *inp) | |
11847 | { | |
11848 | ldout(cct, 3) << "_symlink(" << dir->ino << " " << name << ", " << target | |
11849 | << ", uid " << perms.uid() << ", gid " << perms.gid() << ")" | |
11850 | << dendl; | |
11851 | ||
11852 | if (strlen(name) > NAME_MAX) | |
11853 | return -ENAMETOOLONG; | |
11854 | ||
11855 | if (dir->snapid != CEPH_NOSNAP) { | |
11856 | return -EROFS; | |
11857 | } | |
11858 | if (is_quota_files_exceeded(dir, perms)) { | |
11859 | return -EDQUOT; | |
11860 | } | |
11861 | ||
11862 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SYMLINK); | |
11863 | ||
11864 | filepath path; | |
11865 | dir->make_nosnap_relative_path(path); | |
11866 | path.push_dentry(name); | |
11867 | req->set_filepath(path); | |
11868 | req->set_inode(dir); | |
11869 | req->set_string2(target); | |
11870 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
11871 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
11872 | ||
11873 | Dentry *de; | |
11874 | int res = get_or_create(dir, name, &de); | |
11875 | if (res < 0) | |
11876 | goto fail; | |
11877 | req->set_dentry(de); | |
11878 | ||
11879 | res = make_request(req, perms, inp); | |
11880 | ||
11881 | trim_cache(); | |
11882 | ldout(cct, 3) << "_symlink(\"" << path << "\", \"" << target << "\") = " << | |
11883 | res << dendl; | |
11884 | return res; | |
11885 | ||
11886 | fail: | |
11887 | put_request(req); | |
11888 | return res; | |
11889 | } | |
11890 | ||
11891 | int Client::ll_symlink(Inode *parent, const char *name, const char *value, | |
11892 | struct stat *attr, Inode **out, const UserPerm& perms) | |
11893 | { | |
11894 | Mutex::Locker lock(client_lock); | |
11895 | ||
181888fb FG |
11896 | if (unmounting) |
11897 | return -ENOTCONN; | |
11898 | ||
7c673cae FG |
11899 | vinodeno_t vparent = _get_vino(parent); |
11900 | ||
11901 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name << " -> " << value | |
11902 | << dendl; | |
11903 | tout(cct) << "ll_symlink" << std::endl; | |
11904 | tout(cct) << vparent.ino.val << std::endl; | |
11905 | tout(cct) << name << std::endl; | |
11906 | tout(cct) << value << std::endl; | |
11907 | ||
11908 | if (!cct->_conf->fuse_default_permissions) { | |
11909 | int r = may_create(parent, perms); | |
11910 | if (r < 0) | |
11911 | return r; | |
11912 | } | |
11913 | ||
11914 | InodeRef in; | |
11915 | int r = _symlink(parent, name, value, perms, &in); | |
11916 | if (r == 0) { | |
11917 | fill_stat(in, attr); | |
11918 | _ll_get(in.get()); | |
11919 | } | |
11920 | tout(cct) << attr->st_ino << std::endl; | |
11921 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name | |
11922 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
11923 | *out = in.get(); | |
11924 | return r; | |
11925 | } | |
11926 | ||
11927 | int Client::ll_symlinkx(Inode *parent, const char *name, const char *value, | |
11928 | Inode **out, struct ceph_statx *stx, unsigned want, | |
11929 | unsigned flags, const UserPerm& perms) | |
11930 | { | |
11931 | Mutex::Locker lock(client_lock); | |
11932 | ||
181888fb FG |
11933 | if (unmounting) |
11934 | return -ENOTCONN; | |
11935 | ||
7c673cae FG |
11936 | vinodeno_t vparent = _get_vino(parent); |
11937 | ||
11938 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name << " -> " << value | |
11939 | << dendl; | |
11940 | tout(cct) << "ll_symlinkx" << std::endl; | |
11941 | tout(cct) << vparent.ino.val << std::endl; | |
11942 | tout(cct) << name << std::endl; | |
11943 | tout(cct) << value << std::endl; | |
11944 | ||
11945 | if (!cct->_conf->fuse_default_permissions) { | |
11946 | int r = may_create(parent, perms); | |
11947 | if (r < 0) | |
11948 | return r; | |
11949 | } | |
11950 | ||
11951 | InodeRef in; | |
11952 | int r = _symlink(parent, name, value, perms, &in); | |
11953 | if (r == 0) { | |
11954 | fill_statx(in, statx_to_mask(flags, want), stx); | |
11955 | _ll_get(in.get()); | |
11956 | } | |
11957 | tout(cct) << stx->stx_ino << std::endl; | |
11958 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name | |
11959 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
11960 | *out = in.get(); | |
11961 | return r; | |
11962 | } | |
11963 | ||
11964 | int Client::_unlink(Inode *dir, const char *name, const UserPerm& perm) | |
11965 | { | |
11966 | ldout(cct, 3) << "_unlink(" << dir->ino << " " << name | |
11967 | << " uid " << perm.uid() << " gid " << perm.gid() | |
11968 | << ")" << dendl; | |
11969 | ||
11970 | if (dir->snapid != CEPH_NOSNAP) { | |
11971 | return -EROFS; | |
11972 | } | |
11973 | ||
11974 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_UNLINK); | |
11975 | ||
11976 | filepath path; | |
11977 | dir->make_nosnap_relative_path(path); | |
11978 | path.push_dentry(name); | |
11979 | req->set_filepath(path); | |
11980 | ||
11981 | InodeRef otherin; | |
b32b8144 | 11982 | Inode *in; |
7c673cae | 11983 | Dentry *de; |
b32b8144 | 11984 | |
7c673cae FG |
11985 | int res = get_or_create(dir, name, &de); |
11986 | if (res < 0) | |
11987 | goto fail; | |
11988 | req->set_dentry(de); | |
11989 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
11990 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
11991 | ||
11992 | res = _lookup(dir, name, 0, &otherin, perm); | |
11993 | if (res < 0) | |
11994 | goto fail; | |
b32b8144 FG |
11995 | |
11996 | in = otherin.get(); | |
11997 | req->set_other_inode(in); | |
11998 | in->break_all_delegs(); | |
7c673cae FG |
11999 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
12000 | ||
12001 | req->set_inode(dir); | |
12002 | ||
12003 | res = make_request(req, perm); | |
12004 | ||
12005 | trim_cache(); | |
12006 | ldout(cct, 3) << "unlink(" << path << ") = " << res << dendl; | |
12007 | return res; | |
12008 | ||
12009 | fail: | |
12010 | put_request(req); | |
12011 | return res; | |
12012 | } | |
12013 | ||
12014 | int Client::ll_unlink(Inode *in, const char *name, const UserPerm& perm) | |
12015 | { | |
12016 | Mutex::Locker lock(client_lock); | |
12017 | ||
181888fb FG |
12018 | if (unmounting) |
12019 | return -ENOTCONN; | |
12020 | ||
7c673cae FG |
12021 | vinodeno_t vino = _get_vino(in); |
12022 | ||
12023 | ldout(cct, 3) << "ll_unlink " << vino << " " << name << dendl; | |
12024 | tout(cct) << "ll_unlink" << std::endl; | |
12025 | tout(cct) << vino.ino.val << std::endl; | |
12026 | tout(cct) << name << std::endl; | |
12027 | ||
12028 | if (!cct->_conf->fuse_default_permissions) { | |
12029 | int r = may_delete(in, name, perm); | |
12030 | if (r < 0) | |
12031 | return r; | |
12032 | } | |
12033 | return _unlink(in, name, perm); | |
12034 | } | |
12035 | ||
12036 | int Client::_rmdir(Inode *dir, const char *name, const UserPerm& perms) | |
12037 | { | |
12038 | ldout(cct, 3) << "_rmdir(" << dir->ino << " " << name << " uid " | |
12039 | << perms.uid() << " gid " << perms.gid() << ")" << dendl; | |
12040 | ||
12041 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
12042 | return -EROFS; | |
12043 | } | |
b32b8144 FG |
12044 | |
12045 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_RMSNAP : CEPH_MDS_OP_RMDIR; | |
12046 | MetaRequest *req = new MetaRequest(op); | |
7c673cae FG |
12047 | filepath path; |
12048 | dir->make_nosnap_relative_path(path); | |
12049 | path.push_dentry(name); | |
12050 | req->set_filepath(path); | |
12051 | ||
12052 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12053 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12054 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
12055 | ||
12056 | InodeRef in; | |
12057 | ||
12058 | Dentry *de; | |
12059 | int res = get_or_create(dir, name, &de); | |
12060 | if (res < 0) | |
12061 | goto fail; | |
b32b8144 FG |
12062 | if (op == CEPH_MDS_OP_RMDIR) |
12063 | req->set_dentry(de); | |
12064 | else | |
12065 | de->get(); | |
12066 | ||
7c673cae FG |
12067 | res = _lookup(dir, name, 0, &in, perms); |
12068 | if (res < 0) | |
12069 | goto fail; | |
b32b8144 | 12070 | if (op == CEPH_MDS_OP_RMDIR) { |
7c673cae | 12071 | req->set_inode(dir); |
7c673cae FG |
12072 | req->set_other_inode(in.get()); |
12073 | } else { | |
12074 | unlink(de, true, true); | |
b32b8144 | 12075 | de->put(); |
7c673cae FG |
12076 | req->set_other_inode(in.get()); |
12077 | } | |
12078 | ||
12079 | res = make_request(req, perms); | |
12080 | ||
12081 | trim_cache(); | |
12082 | ldout(cct, 3) << "rmdir(" << path << ") = " << res << dendl; | |
12083 | return res; | |
12084 | ||
12085 | fail: | |
12086 | put_request(req); | |
12087 | return res; | |
12088 | } | |
12089 | ||
12090 | int Client::ll_rmdir(Inode *in, const char *name, const UserPerm& perms) | |
12091 | { | |
12092 | Mutex::Locker lock(client_lock); | |
12093 | ||
181888fb FG |
12094 | if (unmounting) |
12095 | return -ENOTCONN; | |
12096 | ||
7c673cae FG |
12097 | vinodeno_t vino = _get_vino(in); |
12098 | ||
12099 | ldout(cct, 3) << "ll_rmdir " << vino << " " << name << dendl; | |
12100 | tout(cct) << "ll_rmdir" << std::endl; | |
12101 | tout(cct) << vino.ino.val << std::endl; | |
12102 | tout(cct) << name << std::endl; | |
12103 | ||
12104 | if (!cct->_conf->fuse_default_permissions) { | |
12105 | int r = may_delete(in, name, perms); | |
12106 | if (r < 0) | |
12107 | return r; | |
12108 | } | |
12109 | ||
12110 | return _rmdir(in, name, perms); | |
12111 | } | |
12112 | ||
12113 | int Client::_rename(Inode *fromdir, const char *fromname, Inode *todir, const char *toname, const UserPerm& perm) | |
12114 | { | |
12115 | ldout(cct, 3) << "_rename(" << fromdir->ino << " " << fromname << " to " | |
12116 | << todir->ino << " " << toname | |
12117 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" | |
12118 | << dendl; | |
12119 | ||
12120 | if (fromdir->snapid != todir->snapid) | |
12121 | return -EXDEV; | |
12122 | ||
12123 | int op = CEPH_MDS_OP_RENAME; | |
12124 | if (fromdir->snapid != CEPH_NOSNAP) { | |
12125 | if (fromdir == todir && fromdir->snapid == CEPH_SNAPDIR) | |
12126 | op = CEPH_MDS_OP_RENAMESNAP; | |
12127 | else | |
12128 | return -EROFS; | |
12129 | } | |
12130 | if (fromdir != todir) { | |
12131 | Inode *fromdir_root = | |
12132 | fromdir->quota.is_enable() ? fromdir : get_quota_root(fromdir, perm); | |
12133 | Inode *todir_root = | |
12134 | todir->quota.is_enable() ? todir : get_quota_root(todir, perm); | |
12135 | if (fromdir_root != todir_root) { | |
12136 | return -EXDEV; | |
12137 | } | |
12138 | } | |
12139 | ||
12140 | InodeRef target; | |
12141 | MetaRequest *req = new MetaRequest(op); | |
12142 | ||
12143 | filepath from; | |
12144 | fromdir->make_nosnap_relative_path(from); | |
12145 | from.push_dentry(fromname); | |
12146 | filepath to; | |
12147 | todir->make_nosnap_relative_path(to); | |
12148 | to.push_dentry(toname); | |
12149 | req->set_filepath(to); | |
12150 | req->set_filepath2(from); | |
12151 | ||
12152 | Dentry *oldde; | |
12153 | int res = get_or_create(fromdir, fromname, &oldde); | |
12154 | if (res < 0) | |
12155 | goto fail; | |
12156 | Dentry *de; | |
12157 | res = get_or_create(todir, toname, &de); | |
12158 | if (res < 0) | |
12159 | goto fail; | |
12160 | ||
12161 | if (op == CEPH_MDS_OP_RENAME) { | |
12162 | req->set_old_dentry(oldde); | |
12163 | req->old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
12164 | req->old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
12165 | ||
12166 | req->set_dentry(de); | |
12167 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12168 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12169 | ||
12170 | InodeRef oldin, otherin; | |
12171 | res = _lookup(fromdir, fromname, 0, &oldin, perm); | |
12172 | if (res < 0) | |
12173 | goto fail; | |
b32b8144 FG |
12174 | |
12175 | Inode *oldinode = oldin.get(); | |
12176 | oldinode->break_all_delegs(); | |
12177 | req->set_old_inode(oldinode); | |
7c673cae FG |
12178 | req->old_inode_drop = CEPH_CAP_LINK_SHARED; |
12179 | ||
12180 | res = _lookup(todir, toname, 0, &otherin, perm); | |
b32b8144 FG |
12181 | switch (res) { |
12182 | case 0: | |
12183 | { | |
12184 | Inode *in = otherin.get(); | |
12185 | req->set_other_inode(in); | |
12186 | in->break_all_delegs(); | |
12187 | } | |
7c673cae | 12188 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
b32b8144 FG |
12189 | break; |
12190 | case -ENOENT: | |
12191 | break; | |
12192 | default: | |
12193 | goto fail; | |
7c673cae FG |
12194 | } |
12195 | ||
12196 | req->set_inode(todir); | |
12197 | } else { | |
12198 | // renamesnap reply contains no tracedn, so we need to invalidate | |
12199 | // dentry manually | |
12200 | unlink(oldde, true, true); | |
12201 | unlink(de, true, true); | |
12202 | } | |
12203 | ||
12204 | res = make_request(req, perm, &target); | |
12205 | ldout(cct, 10) << "rename result is " << res << dendl; | |
12206 | ||
12207 | // renamed item from our cache | |
12208 | ||
12209 | trim_cache(); | |
12210 | ldout(cct, 3) << "_rename(" << from << ", " << to << ") = " << res << dendl; | |
12211 | return res; | |
12212 | ||
12213 | fail: | |
12214 | put_request(req); | |
12215 | return res; | |
12216 | } | |
12217 | ||
12218 | int Client::ll_rename(Inode *parent, const char *name, Inode *newparent, | |
12219 | const char *newname, const UserPerm& perm) | |
12220 | { | |
12221 | Mutex::Locker lock(client_lock); | |
12222 | ||
181888fb FG |
12223 | if (unmounting) |
12224 | return -ENOTCONN; | |
12225 | ||
7c673cae FG |
12226 | vinodeno_t vparent = _get_vino(parent); |
12227 | vinodeno_t vnewparent = _get_vino(newparent); | |
12228 | ||
12229 | ldout(cct, 3) << "ll_rename " << vparent << " " << name << " to " | |
12230 | << vnewparent << " " << newname << dendl; | |
12231 | tout(cct) << "ll_rename" << std::endl; | |
12232 | tout(cct) << vparent.ino.val << std::endl; | |
12233 | tout(cct) << name << std::endl; | |
12234 | tout(cct) << vnewparent.ino.val << std::endl; | |
12235 | tout(cct) << newname << std::endl; | |
12236 | ||
12237 | if (!cct->_conf->fuse_default_permissions) { | |
12238 | int r = may_delete(parent, name, perm); | |
12239 | if (r < 0) | |
12240 | return r; | |
12241 | r = may_delete(newparent, newname, perm); | |
12242 | if (r < 0 && r != -ENOENT) | |
12243 | return r; | |
12244 | } | |
12245 | ||
12246 | return _rename(parent, name, newparent, newname, perm); | |
12247 | } | |
12248 | ||
12249 | int Client::_link(Inode *in, Inode *dir, const char *newname, const UserPerm& perm, InodeRef *inp) | |
12250 | { | |
12251 | ldout(cct, 3) << "_link(" << in->ino << " to " << dir->ino << " " << newname | |
12252 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" << dendl; | |
12253 | ||
12254 | if (strlen(newname) > NAME_MAX) | |
12255 | return -ENAMETOOLONG; | |
12256 | ||
12257 | if (in->snapid != CEPH_NOSNAP || dir->snapid != CEPH_NOSNAP) { | |
12258 | return -EROFS; | |
12259 | } | |
12260 | if (is_quota_files_exceeded(dir, perm)) { | |
12261 | return -EDQUOT; | |
12262 | } | |
12263 | ||
b32b8144 | 12264 | in->break_all_delegs(); |
7c673cae FG |
12265 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LINK); |
12266 | ||
12267 | filepath path(newname, dir->ino); | |
12268 | req->set_filepath(path); | |
12269 | filepath existing(in->ino); | |
12270 | req->set_filepath2(existing); | |
12271 | ||
12272 | req->set_inode(dir); | |
12273 | req->inode_drop = CEPH_CAP_FILE_SHARED; | |
12274 | req->inode_unless = CEPH_CAP_FILE_EXCL; | |
12275 | ||
12276 | Dentry *de; | |
12277 | int res = get_or_create(dir, newname, &de); | |
12278 | if (res < 0) | |
12279 | goto fail; | |
12280 | req->set_dentry(de); | |
12281 | ||
12282 | res = make_request(req, perm, inp); | |
12283 | ldout(cct, 10) << "link result is " << res << dendl; | |
12284 | ||
12285 | trim_cache(); | |
12286 | ldout(cct, 3) << "link(" << existing << ", " << path << ") = " << res << dendl; | |
12287 | return res; | |
12288 | ||
12289 | fail: | |
12290 | put_request(req); | |
12291 | return res; | |
12292 | } | |
12293 | ||
12294 | int Client::ll_link(Inode *in, Inode *newparent, const char *newname, | |
12295 | const UserPerm& perm) | |
12296 | { | |
12297 | Mutex::Locker lock(client_lock); | |
12298 | ||
181888fb FG |
12299 | if (unmounting) |
12300 | return -ENOTCONN; | |
12301 | ||
7c673cae FG |
12302 | vinodeno_t vino = _get_vino(in); |
12303 | vinodeno_t vnewparent = _get_vino(newparent); | |
12304 | ||
31f18b77 | 12305 | ldout(cct, 3) << "ll_link " << vino << " to " << vnewparent << " " << |
7c673cae FG |
12306 | newname << dendl; |
12307 | tout(cct) << "ll_link" << std::endl; | |
12308 | tout(cct) << vino.ino.val << std::endl; | |
12309 | tout(cct) << vnewparent << std::endl; | |
12310 | tout(cct) << newname << std::endl; | |
12311 | ||
12312 | int r = 0; | |
12313 | InodeRef target; | |
12314 | ||
12315 | if (!cct->_conf->fuse_default_permissions) { | |
12316 | if (S_ISDIR(in->mode)) | |
12317 | return -EPERM; | |
12318 | ||
12319 | r = may_hardlink(in, perm); | |
12320 | if (r < 0) | |
12321 | return r; | |
12322 | ||
12323 | r = may_create(newparent, perm); | |
12324 | if (r < 0) | |
12325 | return r; | |
12326 | } | |
12327 | ||
12328 | return _link(in, newparent, newname, perm, &target); | |
12329 | } | |
12330 | ||
12331 | int Client::ll_num_osds(void) | |
12332 | { | |
12333 | Mutex::Locker lock(client_lock); | |
12334 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_num_osds)); | |
12335 | } | |
12336 | ||
12337 | int Client::ll_osdaddr(int osd, uint32_t *addr) | |
12338 | { | |
12339 | Mutex::Locker lock(client_lock); | |
181888fb | 12340 | |
7c673cae FG |
12341 | entity_addr_t g; |
12342 | bool exists = objecter->with_osdmap([&](const OSDMap& o) { | |
12343 | if (!o.exists(osd)) | |
12344 | return false; | |
12345 | g = o.get_addr(osd); | |
12346 | return true; | |
12347 | }); | |
12348 | if (!exists) | |
12349 | return -1; | |
12350 | uint32_t nb_addr = (g.in4_addr()).sin_addr.s_addr; | |
12351 | *addr = ntohl(nb_addr); | |
12352 | return 0; | |
12353 | } | |
181888fb | 12354 | |
7c673cae FG |
12355 | uint32_t Client::ll_stripe_unit(Inode *in) |
12356 | { | |
12357 | Mutex::Locker lock(client_lock); | |
12358 | return in->layout.stripe_unit; | |
12359 | } | |
12360 | ||
12361 | uint64_t Client::ll_snap_seq(Inode *in) | |
12362 | { | |
12363 | Mutex::Locker lock(client_lock); | |
12364 | return in->snaprealm->seq; | |
12365 | } | |
12366 | ||
12367 | int Client::ll_file_layout(Inode *in, file_layout_t *layout) | |
12368 | { | |
12369 | Mutex::Locker lock(client_lock); | |
12370 | *layout = in->layout; | |
12371 | return 0; | |
12372 | } | |
12373 | ||
12374 | int Client::ll_file_layout(Fh *fh, file_layout_t *layout) | |
12375 | { | |
12376 | return ll_file_layout(fh->inode.get(), layout); | |
12377 | } | |
12378 | ||
12379 | /* Currently we cannot take advantage of redundancy in reads, since we | |
12380 | would have to go through all possible placement groups (a | |
12381 | potentially quite large number determined by a hash), and use CRUSH | |
12382 | to calculate the appropriate set of OSDs for each placement group, | |
12383 | then index into that. An array with one entry per OSD is much more | |
12384 | tractable and works for demonstration purposes. */ | |
12385 | ||
12386 | int Client::ll_get_stripe_osd(Inode *in, uint64_t blockno, | |
12387 | file_layout_t* layout) | |
12388 | { | |
12389 | Mutex::Locker lock(client_lock); | |
181888fb | 12390 | |
7c673cae FG |
12391 | inodeno_t ino = ll_get_inodeno(in); |
12392 | uint32_t object_size = layout->object_size; | |
12393 | uint32_t su = layout->stripe_unit; | |
12394 | uint32_t stripe_count = layout->stripe_count; | |
12395 | uint64_t stripes_per_object = object_size / su; | |
12396 | ||
12397 | uint64_t stripeno = blockno / stripe_count; // which horizontal stripe (Y) | |
12398 | uint64_t stripepos = blockno % stripe_count; // which object in the object set (X) | |
12399 | uint64_t objectsetno = stripeno / stripes_per_object; // which object set | |
12400 | uint64_t objectno = objectsetno * stripe_count + stripepos; // object id | |
12401 | ||
12402 | object_t oid = file_object_t(ino, objectno); | |
12403 | return objecter->with_osdmap([&](const OSDMap& o) { | |
12404 | ceph_object_layout olayout = | |
12405 | o.file_to_object_layout(oid, *layout); | |
12406 | pg_t pg = (pg_t)olayout.ol_pgid; | |
12407 | vector<int> osds; | |
12408 | int primary; | |
12409 | o.pg_to_acting_osds(pg, &osds, &primary); | |
12410 | return primary; | |
12411 | }); | |
12412 | } | |
12413 | ||
12414 | /* Return the offset of the block, internal to the object */ | |
12415 | ||
12416 | uint64_t Client::ll_get_internal_offset(Inode *in, uint64_t blockno) | |
12417 | { | |
12418 | Mutex::Locker lock(client_lock); | |
12419 | file_layout_t *layout=&(in->layout); | |
12420 | uint32_t object_size = layout->object_size; | |
12421 | uint32_t su = layout->stripe_unit; | |
12422 | uint64_t stripes_per_object = object_size / su; | |
12423 | ||
12424 | return (blockno % stripes_per_object) * su; | |
12425 | } | |
12426 | ||
12427 | int Client::ll_opendir(Inode *in, int flags, dir_result_t** dirpp, | |
12428 | const UserPerm& perms) | |
12429 | { | |
12430 | Mutex::Locker lock(client_lock); | |
12431 | ||
181888fb FG |
12432 | if (unmounting) |
12433 | return -ENOTCONN; | |
12434 | ||
7c673cae FG |
12435 | vinodeno_t vino = _get_vino(in); |
12436 | ||
12437 | ldout(cct, 3) << "ll_opendir " << vino << dendl; | |
12438 | tout(cct) << "ll_opendir" << std::endl; | |
12439 | tout(cct) << vino.ino.val << std::endl; | |
12440 | ||
12441 | if (!cct->_conf->fuse_default_permissions) { | |
12442 | int r = may_open(in, flags, perms); | |
12443 | if (r < 0) | |
12444 | return r; | |
12445 | } | |
12446 | ||
12447 | int r = _opendir(in, dirpp, perms); | |
12448 | tout(cct) << (unsigned long)*dirpp << std::endl; | |
12449 | ||
12450 | ldout(cct, 3) << "ll_opendir " << vino << " = " << r << " (" << *dirpp << ")" | |
12451 | << dendl; | |
12452 | return r; | |
12453 | } | |
12454 | ||
12455 | int Client::ll_releasedir(dir_result_t *dirp) | |
12456 | { | |
12457 | Mutex::Locker lock(client_lock); | |
12458 | ldout(cct, 3) << "ll_releasedir " << dirp << dendl; | |
12459 | tout(cct) << "ll_releasedir" << std::endl; | |
12460 | tout(cct) << (unsigned long)dirp << std::endl; | |
181888fb FG |
12461 | |
12462 | if (unmounting) | |
12463 | return -ENOTCONN; | |
12464 | ||
7c673cae FG |
12465 | _closedir(dirp); |
12466 | return 0; | |
12467 | } | |
12468 | ||
12469 | int Client::ll_fsyncdir(dir_result_t *dirp) | |
12470 | { | |
12471 | Mutex::Locker lock(client_lock); | |
12472 | ldout(cct, 3) << "ll_fsyncdir " << dirp << dendl; | |
12473 | tout(cct) << "ll_fsyncdir" << std::endl; | |
12474 | tout(cct) << (unsigned long)dirp << std::endl; | |
12475 | ||
181888fb FG |
12476 | if (unmounting) |
12477 | return -ENOTCONN; | |
12478 | ||
7c673cae FG |
12479 | return _fsync(dirp->inode.get(), false); |
12480 | } | |
12481 | ||
12482 | int Client::ll_open(Inode *in, int flags, Fh **fhp, const UserPerm& perms) | |
12483 | { | |
12484 | assert(!(flags & O_CREAT)); | |
12485 | ||
12486 | Mutex::Locker lock(client_lock); | |
12487 | ||
181888fb FG |
12488 | if (unmounting) |
12489 | return -ENOTCONN; | |
12490 | ||
7c673cae FG |
12491 | vinodeno_t vino = _get_vino(in); |
12492 | ||
12493 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << dendl; | |
12494 | tout(cct) << "ll_open" << std::endl; | |
12495 | tout(cct) << vino.ino.val << std::endl; | |
12496 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
12497 | ||
12498 | int r; | |
12499 | if (!cct->_conf->fuse_default_permissions) { | |
12500 | r = may_open(in, flags, perms); | |
12501 | if (r < 0) | |
12502 | goto out; | |
12503 | } | |
12504 | ||
12505 | r = _open(in, flags, 0, fhp /* may be NULL */, perms); | |
12506 | ||
12507 | out: | |
12508 | Fh *fhptr = fhp ? *fhp : NULL; | |
12509 | if (fhptr) { | |
12510 | ll_unclosed_fh_set.insert(fhptr); | |
12511 | } | |
12512 | tout(cct) << (unsigned long)fhptr << std::endl; | |
12513 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << | |
12514 | " = " << r << " (" << fhptr << ")" << dendl; | |
12515 | return r; | |
12516 | } | |
12517 | ||
12518 | int Client::_ll_create(Inode *parent, const char *name, mode_t mode, | |
12519 | int flags, InodeRef *in, int caps, Fh **fhp, | |
12520 | const UserPerm& perms) | |
12521 | { | |
12522 | *fhp = NULL; | |
12523 | ||
12524 | vinodeno_t vparent = _get_vino(parent); | |
12525 | ||
12526 | ldout(cct, 3) << "_ll_create " << vparent << " " << name << " 0" << oct << | |
12527 | mode << dec << " " << ceph_flags_sys2wire(flags) << ", uid " << perms.uid() | |
12528 | << ", gid " << perms.gid() << dendl; | |
12529 | tout(cct) << "ll_create" << std::endl; | |
12530 | tout(cct) << vparent.ino.val << std::endl; | |
12531 | tout(cct) << name << std::endl; | |
12532 | tout(cct) << mode << std::endl; | |
12533 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
12534 | ||
12535 | bool created = false; | |
12536 | int r = _lookup(parent, name, caps, in, perms); | |
12537 | ||
12538 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
12539 | return -EEXIST; | |
12540 | ||
12541 | if (r == -ENOENT && (flags & O_CREAT)) { | |
12542 | if (!cct->_conf->fuse_default_permissions) { | |
12543 | r = may_create(parent, perms); | |
12544 | if (r < 0) | |
12545 | goto out; | |
12546 | } | |
12547 | r = _create(parent, name, flags, mode, in, fhp, 0, 0, 0, NULL, &created, | |
12548 | perms); | |
12549 | if (r < 0) | |
12550 | goto out; | |
12551 | } | |
12552 | ||
12553 | if (r < 0) | |
12554 | goto out; | |
12555 | ||
12556 | assert(*in); | |
12557 | ||
12558 | ldout(cct, 20) << "_ll_create created = " << created << dendl; | |
12559 | if (!created) { | |
12560 | if (!cct->_conf->fuse_default_permissions) { | |
12561 | r = may_open(in->get(), flags, perms); | |
12562 | if (r < 0) { | |
12563 | if (*fhp) { | |
12564 | int release_r = _release_fh(*fhp); | |
12565 | assert(release_r == 0); // during create, no async data ops should have happened | |
12566 | } | |
12567 | goto out; | |
12568 | } | |
12569 | } | |
12570 | if (*fhp == NULL) { | |
12571 | r = _open(in->get(), flags, mode, fhp, perms); | |
12572 | if (r < 0) | |
12573 | goto out; | |
12574 | } | |
12575 | } | |
12576 | ||
12577 | out: | |
12578 | if (*fhp) { | |
12579 | ll_unclosed_fh_set.insert(*fhp); | |
12580 | } | |
12581 | ||
12582 | ino_t ino = 0; | |
12583 | if (r >= 0) { | |
12584 | Inode *inode = in->get(); | |
12585 | if (use_faked_inos()) | |
12586 | ino = inode->faked_ino; | |
12587 | else | |
12588 | ino = inode->ino; | |
12589 | } | |
12590 | ||
12591 | tout(cct) << (unsigned long)*fhp << std::endl; | |
12592 | tout(cct) << ino << std::endl; | |
31f18b77 | 12593 | ldout(cct, 3) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
12594 | mode << dec << " " << ceph_flags_sys2wire(flags) << " = " << r << " (" << |
12595 | *fhp << " " << hex << ino << dec << ")" << dendl; | |
12596 | ||
12597 | return r; | |
12598 | } | |
12599 | ||
12600 | int Client::ll_create(Inode *parent, const char *name, mode_t mode, | |
12601 | int flags, struct stat *attr, Inode **outp, Fh **fhp, | |
12602 | const UserPerm& perms) | |
12603 | { | |
12604 | Mutex::Locker lock(client_lock); | |
12605 | InodeRef in; | |
12606 | ||
181888fb FG |
12607 | if (unmounting) |
12608 | return -ENOTCONN; | |
12609 | ||
7c673cae FG |
12610 | int r = _ll_create(parent, name, mode, flags, &in, CEPH_STAT_CAP_INODE_ALL, |
12611 | fhp, perms); | |
12612 | if (r >= 0) { | |
12613 | assert(in); | |
12614 | ||
12615 | // passing an Inode in outp requires an additional ref | |
12616 | if (outp) { | |
12617 | _ll_get(in.get()); | |
12618 | *outp = in.get(); | |
12619 | } | |
12620 | fill_stat(in, attr); | |
12621 | } else { | |
12622 | attr->st_ino = 0; | |
12623 | } | |
12624 | ||
12625 | return r; | |
12626 | } | |
12627 | ||
12628 | int Client::ll_createx(Inode *parent, const char *name, mode_t mode, | |
12629 | int oflags, Inode **outp, Fh **fhp, | |
12630 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
12631 | const UserPerm& perms) | |
12632 | { | |
12633 | unsigned caps = statx_to_mask(lflags, want); | |
12634 | Mutex::Locker lock(client_lock); | |
12635 | InodeRef in; | |
12636 | ||
181888fb FG |
12637 | if (unmounting) |
12638 | return -ENOTCONN; | |
7c673cae FG |
12639 | |
12640 | int r = _ll_create(parent, name, mode, oflags, &in, caps, fhp, perms); | |
12641 | if (r >= 0) { | |
12642 | assert(in); | |
12643 | ||
12644 | // passing an Inode in outp requires an additional ref | |
12645 | if (outp) { | |
12646 | _ll_get(in.get()); | |
12647 | *outp = in.get(); | |
12648 | } | |
12649 | fill_statx(in, caps, stx); | |
12650 | } else { | |
12651 | stx->stx_ino = 0; | |
12652 | stx->stx_mask = 0; | |
12653 | } | |
12654 | ||
12655 | return r; | |
12656 | } | |
12657 | ||
12658 | loff_t Client::ll_lseek(Fh *fh, loff_t offset, int whence) | |
12659 | { | |
12660 | Mutex::Locker lock(client_lock); | |
12661 | tout(cct) << "ll_lseek" << std::endl; | |
12662 | tout(cct) << offset << std::endl; | |
12663 | tout(cct) << whence << std::endl; | |
12664 | ||
181888fb FG |
12665 | if (unmounting) |
12666 | return -ENOTCONN; | |
12667 | ||
7c673cae FG |
12668 | return _lseek(fh, offset, whence); |
12669 | } | |
12670 | ||
12671 | int Client::ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl) | |
12672 | { | |
12673 | Mutex::Locker lock(client_lock); | |
12674 | ldout(cct, 3) << "ll_read " << fh << " " << fh->inode->ino << " " << " " << off << "~" << len << dendl; | |
12675 | tout(cct) << "ll_read" << std::endl; | |
12676 | tout(cct) << (unsigned long)fh << std::endl; | |
12677 | tout(cct) << off << std::endl; | |
12678 | tout(cct) << len << std::endl; | |
12679 | ||
181888fb FG |
12680 | if (unmounting) |
12681 | return -ENOTCONN; | |
12682 | ||
7c673cae FG |
12683 | return _read(fh, off, len, bl); |
12684 | } | |
12685 | ||
12686 | int Client::ll_read_block(Inode *in, uint64_t blockid, | |
12687 | char *buf, | |
12688 | uint64_t offset, | |
12689 | uint64_t length, | |
12690 | file_layout_t* layout) | |
12691 | { | |
12692 | Mutex::Locker lock(client_lock); | |
181888fb FG |
12693 | |
12694 | if (unmounting) | |
12695 | return -ENOTCONN; | |
12696 | ||
b32b8144 | 12697 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
12698 | object_t oid = file_object_t(vino.ino, blockid); |
12699 | C_SaferCond onfinish; | |
12700 | bufferlist bl; | |
12701 | ||
12702 | objecter->read(oid, | |
12703 | object_locator_t(layout->pool_id), | |
12704 | offset, | |
12705 | length, | |
12706 | vino.snapid, | |
12707 | &bl, | |
12708 | CEPH_OSD_FLAG_READ, | |
12709 | &onfinish); | |
12710 | ||
12711 | client_lock.Unlock(); | |
12712 | int r = onfinish.wait(); | |
12713 | client_lock.Lock(); | |
12714 | ||
12715 | if (r >= 0) { | |
12716 | bl.copy(0, bl.length(), buf); | |
12717 | r = bl.length(); | |
12718 | } | |
12719 | ||
12720 | return r; | |
12721 | } | |
12722 | ||
12723 | /* It appears that the OSD doesn't return success unless the entire | |
12724 | buffer was written, return the write length on success. */ | |
12725 | ||
12726 | int Client::ll_write_block(Inode *in, uint64_t blockid, | |
12727 | char* buf, uint64_t offset, | |
12728 | uint64_t length, file_layout_t* layout, | |
12729 | uint64_t snapseq, uint32_t sync) | |
12730 | { | |
12731 | Mutex flock("Client::ll_write_block flock"); | |
12732 | vinodeno_t vino = ll_get_vino(in); | |
12733 | Cond cond; | |
12734 | bool done; | |
12735 | int r = 0; | |
181888fb | 12736 | Context *onsafe = nullptr; |
7c673cae FG |
12737 | |
12738 | if (length == 0) { | |
12739 | return -EINVAL; | |
12740 | } | |
12741 | if (true || sync) { | |
12742 | /* if write is stable, the epilogue is waiting on | |
12743 | * flock */ | |
12744 | onsafe = new C_SafeCond(&flock, &cond, &done, &r); | |
12745 | done = false; | |
12746 | } else { | |
12747 | /* if write is unstable, we just place a barrier for | |
12748 | * future commits to wait on */ | |
12749 | /*onsafe = new C_Block_Sync(this, vino.ino, | |
12750 | barrier_interval(offset, offset + length), &r); | |
12751 | */ | |
12752 | done = true; | |
12753 | } | |
12754 | object_t oid = file_object_t(vino.ino, blockid); | |
12755 | SnapContext fakesnap; | |
12756 | bufferptr bp; | |
12757 | if (length > 0) bp = buffer::copy(buf, length); | |
12758 | bufferlist bl; | |
12759 | bl.push_back(bp); | |
12760 | ||
12761 | ldout(cct, 1) << "ll_block_write for " << vino.ino << "." << blockid | |
12762 | << dendl; | |
12763 | ||
12764 | fakesnap.seq = snapseq; | |
12765 | ||
12766 | /* lock just in time */ | |
12767 | client_lock.Lock(); | |
181888fb FG |
12768 | if (unmounting) { |
12769 | client_lock.Unlock(); | |
12770 | delete onsafe; | |
12771 | return -ENOTCONN; | |
12772 | } | |
7c673cae FG |
12773 | |
12774 | objecter->write(oid, | |
12775 | object_locator_t(layout->pool_id), | |
12776 | offset, | |
12777 | length, | |
12778 | fakesnap, | |
12779 | bl, | |
12780 | ceph::real_clock::now(), | |
12781 | 0, | |
12782 | onsafe); | |
12783 | ||
12784 | client_lock.Unlock(); | |
12785 | if (!done /* also !sync */) { | |
12786 | flock.Lock(); | |
12787 | while (! done) | |
12788 | cond.Wait(flock); | |
12789 | flock.Unlock(); | |
12790 | } | |
12791 | ||
12792 | if (r < 0) { | |
12793 | return r; | |
12794 | } else { | |
12795 | return length; | |
12796 | } | |
12797 | } | |
12798 | ||
12799 | int Client::ll_commit_blocks(Inode *in, | |
12800 | uint64_t offset, | |
12801 | uint64_t length) | |
12802 | { | |
12803 | Mutex::Locker lock(client_lock); | |
12804 | /* | |
12805 | BarrierContext *bctx; | |
b32b8144 | 12806 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
12807 | uint64_t ino = vino.ino; |
12808 | ||
12809 | ldout(cct, 1) << "ll_commit_blocks for " << vino.ino << " from " | |
12810 | << offset << " to " << length << dendl; | |
12811 | ||
12812 | if (length == 0) { | |
12813 | return -EINVAL; | |
12814 | } | |
12815 | ||
12816 | map<uint64_t, BarrierContext*>::iterator p = barriers.find(ino); | |
12817 | if (p != barriers.end()) { | |
12818 | barrier_interval civ(offset, offset + length); | |
12819 | p->second->commit_barrier(civ); | |
12820 | } | |
12821 | */ | |
12822 | return 0; | |
12823 | } | |
12824 | ||
12825 | int Client::ll_write(Fh *fh, loff_t off, loff_t len, const char *data) | |
12826 | { | |
12827 | Mutex::Locker lock(client_lock); | |
12828 | ldout(cct, 3) << "ll_write " << fh << " " << fh->inode->ino << " " << off << | |
12829 | "~" << len << dendl; | |
12830 | tout(cct) << "ll_write" << std::endl; | |
12831 | tout(cct) << (unsigned long)fh << std::endl; | |
12832 | tout(cct) << off << std::endl; | |
12833 | tout(cct) << len << std::endl; | |
12834 | ||
181888fb FG |
12835 | if (unmounting) |
12836 | return -ENOTCONN; | |
12837 | ||
7c673cae FG |
12838 | int r = _write(fh, off, len, data, NULL, 0); |
12839 | ldout(cct, 3) << "ll_write " << fh << " " << off << "~" << len << " = " << r | |
12840 | << dendl; | |
12841 | return r; | |
12842 | } | |
12843 | ||
12844 | int Client::ll_flush(Fh *fh) | |
12845 | { | |
12846 | Mutex::Locker lock(client_lock); | |
12847 | ldout(cct, 3) << "ll_flush " << fh << " " << fh->inode->ino << " " << dendl; | |
12848 | tout(cct) << "ll_flush" << std::endl; | |
12849 | tout(cct) << (unsigned long)fh << std::endl; | |
12850 | ||
181888fb FG |
12851 | if (unmounting) |
12852 | return -ENOTCONN; | |
12853 | ||
7c673cae FG |
12854 | return _flush(fh); |
12855 | } | |
12856 | ||
12857 | int Client::ll_fsync(Fh *fh, bool syncdataonly) | |
12858 | { | |
12859 | Mutex::Locker lock(client_lock); | |
12860 | ldout(cct, 3) << "ll_fsync " << fh << " " << fh->inode->ino << " " << dendl; | |
12861 | tout(cct) << "ll_fsync" << std::endl; | |
12862 | tout(cct) << (unsigned long)fh << std::endl; | |
12863 | ||
181888fb FG |
12864 | if (unmounting) |
12865 | return -ENOTCONN; | |
12866 | ||
7c673cae FG |
12867 | int r = _fsync(fh, syncdataonly); |
12868 | if (r) { | |
12869 | // If we're returning an error, clear it from the FH | |
12870 | fh->take_async_err(); | |
12871 | } | |
12872 | return r; | |
12873 | } | |
12874 | ||
12875 | #ifdef FALLOC_FL_PUNCH_HOLE | |
12876 | ||
12877 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) | |
12878 | { | |
12879 | if (offset < 0 || length <= 0) | |
12880 | return -EINVAL; | |
12881 | ||
12882 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
12883 | return -EOPNOTSUPP; | |
12884 | ||
12885 | if ((mode & FALLOC_FL_PUNCH_HOLE) && !(mode & FALLOC_FL_KEEP_SIZE)) | |
12886 | return -EOPNOTSUPP; | |
12887 | ||
12888 | Inode *in = fh->inode.get(); | |
12889 | ||
12890 | if (objecter->osdmap_pool_full(in->layout.pool_id) && | |
12891 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
12892 | return -ENOSPC; | |
12893 | } | |
12894 | ||
12895 | if (in->snapid != CEPH_NOSNAP) | |
12896 | return -EROFS; | |
12897 | ||
12898 | if ((fh->mode & CEPH_FILE_MODE_WR) == 0) | |
12899 | return -EBADF; | |
12900 | ||
12901 | uint64_t size = offset + length; | |
12902 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
12903 | size > in->size && | |
12904 | is_quota_bytes_exceeded(in, size - in->size, fh->actor_perms)) { | |
12905 | return -EDQUOT; | |
12906 | } | |
12907 | ||
12908 | int have; | |
12909 | int r = get_caps(in, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER, &have, -1); | |
12910 | if (r < 0) | |
12911 | return r; | |
12912 | ||
12913 | Mutex uninline_flock("Client::_fallocate_uninline_data flock"); | |
12914 | Cond uninline_cond; | |
12915 | bool uninline_done = false; | |
12916 | int uninline_ret = 0; | |
12917 | Context *onuninline = NULL; | |
12918 | ||
12919 | if (mode & FALLOC_FL_PUNCH_HOLE) { | |
12920 | if (in->inline_version < CEPH_INLINE_NONE && | |
12921 | (have & CEPH_CAP_FILE_BUFFER)) { | |
12922 | bufferlist bl; | |
12923 | int len = in->inline_data.length(); | |
12924 | if (offset < len) { | |
12925 | if (offset > 0) | |
12926 | in->inline_data.copy(0, offset, bl); | |
12927 | int size = length; | |
12928 | if (offset + size > len) | |
12929 | size = len - offset; | |
12930 | if (size > 0) | |
12931 | bl.append_zero(size); | |
12932 | if (offset + size < len) | |
12933 | in->inline_data.copy(offset + size, len - offset - size, bl); | |
12934 | in->inline_data = bl; | |
12935 | in->inline_version++; | |
12936 | } | |
12937 | in->mtime = ceph_clock_now(); | |
12938 | in->change_attr++; | |
12939 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
12940 | } else { | |
12941 | if (in->inline_version < CEPH_INLINE_NONE) { | |
12942 | onuninline = new C_SafeCond(&uninline_flock, | |
12943 | &uninline_cond, | |
12944 | &uninline_done, | |
12945 | &uninline_ret); | |
12946 | uninline_data(in, onuninline); | |
12947 | } | |
12948 | ||
12949 | Mutex flock("Client::_punch_hole flock"); | |
12950 | Cond cond; | |
12951 | bool done = false; | |
12952 | Context *onfinish = new C_SafeCond(&flock, &cond, &done); | |
12953 | ||
12954 | unsafe_sync_write++; | |
12955 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
12956 | ||
12957 | _invalidate_inode_cache(in, offset, length); | |
12958 | filer->zero(in->ino, &in->layout, | |
12959 | in->snaprealm->get_snap_context(), | |
12960 | offset, length, | |
12961 | ceph::real_clock::now(), | |
12962 | 0, true, onfinish); | |
12963 | in->mtime = ceph_clock_now(); | |
12964 | in->change_attr++; | |
12965 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
12966 | ||
12967 | client_lock.Unlock(); | |
12968 | flock.Lock(); | |
12969 | while (!done) | |
12970 | cond.Wait(flock); | |
12971 | flock.Unlock(); | |
12972 | client_lock.Lock(); | |
12973 | _sync_write_commit(in); | |
12974 | } | |
12975 | } else if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
12976 | uint64_t size = offset + length; | |
12977 | if (size > in->size) { | |
12978 | in->size = size; | |
12979 | in->mtime = ceph_clock_now(); | |
12980 | in->change_attr++; | |
12981 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
12982 | ||
12983 | if (is_quota_bytes_approaching(in, fh->actor_perms)) { | |
12984 | check_caps(in, CHECK_CAPS_NODELAY); | |
31f18b77 FG |
12985 | } else if (is_max_size_approaching(in)) { |
12986 | check_caps(in, 0); | |
7c673cae FG |
12987 | } |
12988 | } | |
12989 | } | |
12990 | ||
12991 | if (onuninline) { | |
12992 | client_lock.Unlock(); | |
12993 | uninline_flock.Lock(); | |
12994 | while (!uninline_done) | |
12995 | uninline_cond.Wait(uninline_flock); | |
12996 | uninline_flock.Unlock(); | |
12997 | client_lock.Lock(); | |
12998 | ||
12999 | if (uninline_ret >= 0 || uninline_ret == -ECANCELED) { | |
13000 | in->inline_data.clear(); | |
13001 | in->inline_version = CEPH_INLINE_NONE; | |
13002 | mark_caps_dirty(in, CEPH_CAP_FILE_WR); | |
13003 | check_caps(in, 0); | |
13004 | } else | |
13005 | r = uninline_ret; | |
13006 | } | |
13007 | ||
13008 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
13009 | return r; | |
13010 | } | |
13011 | #else | |
13012 | ||
13013 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) | |
13014 | { | |
13015 | return -EOPNOTSUPP; | |
13016 | } | |
13017 | ||
13018 | #endif | |
13019 | ||
13020 | ||
13021 | int Client::ll_fallocate(Fh *fh, int mode, loff_t offset, loff_t length) | |
13022 | { | |
13023 | Mutex::Locker lock(client_lock); | |
13024 | ldout(cct, 3) << "ll_fallocate " << fh << " " << fh->inode->ino << " " << dendl; | |
13025 | tout(cct) << "ll_fallocate " << mode << " " << offset << " " << length << std::endl; | |
13026 | tout(cct) << (unsigned long)fh << std::endl; | |
13027 | ||
181888fb FG |
13028 | if (unmounting) |
13029 | return -ENOTCONN; | |
13030 | ||
7c673cae FG |
13031 | return _fallocate(fh, mode, offset, length); |
13032 | } | |
13033 | ||
13034 | int Client::fallocate(int fd, int mode, loff_t offset, loff_t length) | |
13035 | { | |
13036 | Mutex::Locker lock(client_lock); | |
13037 | tout(cct) << "fallocate " << " " << fd << mode << " " << offset << " " << length << std::endl; | |
13038 | ||
181888fb FG |
13039 | if (unmounting) |
13040 | return -ENOTCONN; | |
13041 | ||
7c673cae FG |
13042 | Fh *fh = get_filehandle(fd); |
13043 | if (!fh) | |
13044 | return -EBADF; | |
13045 | #if defined(__linux__) && defined(O_PATH) | |
13046 | if (fh->flags & O_PATH) | |
13047 | return -EBADF; | |
13048 | #endif | |
13049 | return _fallocate(fh, mode, offset, length); | |
13050 | } | |
13051 | ||
13052 | int Client::ll_release(Fh *fh) | |
13053 | { | |
13054 | Mutex::Locker lock(client_lock); | |
13055 | ldout(cct, 3) << "ll_release (fh)" << fh << " " << fh->inode->ino << " " << | |
13056 | dendl; | |
13057 | tout(cct) << "ll_release (fh)" << std::endl; | |
13058 | tout(cct) << (unsigned long)fh << std::endl; | |
13059 | ||
181888fb FG |
13060 | if (unmounting) |
13061 | return -ENOTCONN; | |
13062 | ||
7c673cae FG |
13063 | if (ll_unclosed_fh_set.count(fh)) |
13064 | ll_unclosed_fh_set.erase(fh); | |
13065 | return _release_fh(fh); | |
13066 | } | |
13067 | ||
13068 | int Client::ll_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
13069 | { | |
13070 | Mutex::Locker lock(client_lock); | |
13071 | ||
13072 | ldout(cct, 3) << "ll_getlk (fh)" << fh << " " << fh->inode->ino << dendl; | |
13073 | tout(cct) << "ll_getk (fh)" << (unsigned long)fh << std::endl; | |
13074 | ||
181888fb FG |
13075 | if (unmounting) |
13076 | return -ENOTCONN; | |
13077 | ||
7c673cae FG |
13078 | return _getlk(fh, fl, owner); |
13079 | } | |
13080 | ||
13081 | int Client::ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
13082 | { | |
13083 | Mutex::Locker lock(client_lock); | |
13084 | ||
13085 | ldout(cct, 3) << "ll_setlk (fh) " << fh << " " << fh->inode->ino << dendl; | |
13086 | tout(cct) << "ll_setk (fh)" << (unsigned long)fh << std::endl; | |
13087 | ||
181888fb FG |
13088 | if (unmounting) |
13089 | return -ENOTCONN; | |
13090 | ||
7c673cae FG |
13091 | return _setlk(fh, fl, owner, sleep); |
13092 | } | |
13093 | ||
13094 | int Client::ll_flock(Fh *fh, int cmd, uint64_t owner) | |
13095 | { | |
13096 | Mutex::Locker lock(client_lock); | |
13097 | ||
13098 | ldout(cct, 3) << "ll_flock (fh) " << fh << " " << fh->inode->ino << dendl; | |
13099 | tout(cct) << "ll_flock (fh)" << (unsigned long)fh << std::endl; | |
13100 | ||
181888fb FG |
13101 | if (unmounting) |
13102 | return -ENOTCONN; | |
13103 | ||
7c673cae FG |
13104 | return _flock(fh, cmd, owner); |
13105 | } | |
13106 | ||
b32b8144 FG |
13107 | int Client::set_deleg_timeout(uint32_t timeout) |
13108 | { | |
13109 | Mutex::Locker lock(client_lock); | |
13110 | ||
13111 | /* | |
13112 | * The whole point is to prevent blacklisting so we must time out the | |
13113 | * delegation before the session autoclose timeout kicks in. | |
13114 | */ | |
13115 | if (timeout >= mdsmap->get_session_autoclose()) | |
13116 | return -EINVAL; | |
13117 | ||
13118 | deleg_timeout = timeout; | |
13119 | return 0; | |
13120 | } | |
13121 | ||
13122 | int Client::ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv) | |
13123 | { | |
13124 | int ret = -EINVAL; | |
13125 | ||
13126 | Mutex::Locker lock(client_lock); | |
13127 | ||
13128 | if (!mounted) | |
13129 | return -ENOTCONN; | |
13130 | ||
13131 | Inode *inode = fh->inode.get(); | |
13132 | ||
13133 | switch(cmd) { | |
13134 | case CEPH_DELEGATION_NONE: | |
13135 | inode->unset_deleg(fh); | |
13136 | ret = 0; | |
13137 | break; | |
13138 | default: | |
13139 | try { | |
13140 | ret = inode->set_deleg(fh, cmd, cb, priv); | |
13141 | } catch (std::bad_alloc) { | |
13142 | ret = -ENOMEM; | |
13143 | } | |
13144 | break; | |
13145 | } | |
13146 | return ret; | |
13147 | } | |
13148 | ||
7c673cae FG |
13149 | class C_Client_RequestInterrupt : public Context { |
13150 | private: | |
13151 | Client *client; | |
13152 | MetaRequest *req; | |
13153 | public: | |
13154 | C_Client_RequestInterrupt(Client *c, MetaRequest *r) : client(c), req(r) { | |
13155 | req->get(); | |
13156 | } | |
13157 | void finish(int r) override { | |
13158 | Mutex::Locker l(client->client_lock); | |
13159 | assert(req->head.op == CEPH_MDS_OP_SETFILELOCK); | |
13160 | client->_interrupt_filelock(req); | |
13161 | client->put_request(req); | |
13162 | } | |
13163 | }; | |
13164 | ||
13165 | void Client::ll_interrupt(void *d) | |
13166 | { | |
13167 | MetaRequest *req = static_cast<MetaRequest*>(d); | |
13168 | ldout(cct, 3) << "ll_interrupt tid " << req->get_tid() << dendl; | |
13169 | tout(cct) << "ll_interrupt tid " << req->get_tid() << std::endl; | |
13170 | interrupt_finisher.queue(new C_Client_RequestInterrupt(this, req)); | |
13171 | } | |
13172 | ||
13173 | // ========================================= | |
13174 | // layout | |
13175 | ||
13176 | // expose file layouts | |
13177 | ||
13178 | int Client::describe_layout(const char *relpath, file_layout_t *lp, | |
13179 | const UserPerm& perms) | |
13180 | { | |
13181 | Mutex::Locker lock(client_lock); | |
13182 | ||
181888fb FG |
13183 | if (unmounting) |
13184 | return -ENOTCONN; | |
13185 | ||
7c673cae FG |
13186 | filepath path(relpath); |
13187 | InodeRef in; | |
13188 | int r = path_walk(path, &in, perms); | |
13189 | if (r < 0) | |
13190 | return r; | |
13191 | ||
13192 | *lp = in->layout; | |
13193 | ||
13194 | ldout(cct, 3) << "describe_layout(" << relpath << ") = 0" << dendl; | |
13195 | return 0; | |
13196 | } | |
13197 | ||
13198 | int Client::fdescribe_layout(int fd, file_layout_t *lp) | |
13199 | { | |
13200 | Mutex::Locker lock(client_lock); | |
13201 | ||
181888fb FG |
13202 | if (unmounting) |
13203 | return -ENOTCONN; | |
13204 | ||
7c673cae FG |
13205 | Fh *f = get_filehandle(fd); |
13206 | if (!f) | |
13207 | return -EBADF; | |
13208 | Inode *in = f->inode.get(); | |
13209 | ||
13210 | *lp = in->layout; | |
13211 | ||
13212 | ldout(cct, 3) << "fdescribe_layout(" << fd << ") = 0" << dendl; | |
13213 | return 0; | |
13214 | } | |
13215 | ||
d2e6a577 FG |
13216 | int64_t Client::get_default_pool_id() |
13217 | { | |
13218 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13219 | |
13220 | if (unmounting) | |
13221 | return -ENOTCONN; | |
13222 | ||
d2e6a577 FG |
13223 | /* first data pool is the default */ |
13224 | return mdsmap->get_first_data_pool(); | |
13225 | } | |
7c673cae FG |
13226 | |
13227 | // expose osdmap | |
13228 | ||
13229 | int64_t Client::get_pool_id(const char *pool_name) | |
13230 | { | |
13231 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13232 | |
13233 | if (unmounting) | |
13234 | return -ENOTCONN; | |
13235 | ||
7c673cae FG |
13236 | return objecter->with_osdmap(std::mem_fn(&OSDMap::lookup_pg_pool_name), |
13237 | pool_name); | |
13238 | } | |
13239 | ||
13240 | string Client::get_pool_name(int64_t pool) | |
13241 | { | |
13242 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13243 | |
13244 | if (unmounting) | |
13245 | return string(); | |
13246 | ||
7c673cae FG |
13247 | return objecter->with_osdmap([pool](const OSDMap& o) { |
13248 | return o.have_pg_pool(pool) ? o.get_pool_name(pool) : string(); | |
13249 | }); | |
13250 | } | |
13251 | ||
13252 | int Client::get_pool_replication(int64_t pool) | |
13253 | { | |
13254 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13255 | |
13256 | if (unmounting) | |
13257 | return -ENOTCONN; | |
13258 | ||
7c673cae FG |
13259 | return objecter->with_osdmap([pool](const OSDMap& o) { |
13260 | return o.have_pg_pool(pool) ? o.get_pg_pool(pool)->get_size() : -ENOENT; | |
13261 | }); | |
13262 | } | |
13263 | ||
13264 | int Client::get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds) | |
13265 | { | |
13266 | Mutex::Locker lock(client_lock); | |
13267 | ||
181888fb FG |
13268 | if (unmounting) |
13269 | return -ENOTCONN; | |
13270 | ||
7c673cae FG |
13271 | Fh *f = get_filehandle(fd); |
13272 | if (!f) | |
13273 | return -EBADF; | |
13274 | Inode *in = f->inode.get(); | |
13275 | ||
13276 | vector<ObjectExtent> extents; | |
13277 | Striper::file_to_extents(cct, in->ino, &in->layout, off, 1, in->truncate_size, extents); | |
13278 | assert(extents.size() == 1); | |
13279 | ||
13280 | objecter->with_osdmap([&](const OSDMap& o) { | |
13281 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
13282 | o.pg_to_acting_osds(pg, osds); | |
13283 | }); | |
13284 | ||
13285 | if (osds.empty()) | |
13286 | return -EINVAL; | |
13287 | ||
13288 | /* | |
13289 | * Return the remainder of the extent (stripe unit) | |
13290 | * | |
13291 | * If length = 1 is passed to Striper::file_to_extents we get a single | |
13292 | * extent back, but its length is one so we still need to compute the length | |
13293 | * to the end of the stripe unit. | |
13294 | * | |
13295 | * If length = su then we may get 1 or 2 objects back in the extents vector | |
13296 | * which would have to be examined. Even then, the offsets are local to the | |
13297 | * object, so matching up to the file offset is extra work. | |
13298 | * | |
13299 | * It seems simpler to stick with length = 1 and manually compute the | |
13300 | * remainder. | |
13301 | */ | |
13302 | if (len) { | |
13303 | uint64_t su = in->layout.stripe_unit; | |
13304 | *len = su - (off % su); | |
13305 | } | |
13306 | ||
13307 | return 0; | |
13308 | } | |
13309 | ||
13310 | int Client::get_osd_crush_location(int id, vector<pair<string, string> >& path) | |
13311 | { | |
13312 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13313 | |
13314 | if (unmounting) | |
13315 | return -ENOTCONN; | |
13316 | ||
7c673cae FG |
13317 | if (id < 0) |
13318 | return -EINVAL; | |
13319 | return objecter->with_osdmap([&](const OSDMap& o) { | |
13320 | return o.crush->get_full_location_ordered(id, path); | |
13321 | }); | |
13322 | } | |
13323 | ||
13324 | int Client::get_file_stripe_address(int fd, loff_t offset, | |
13325 | vector<entity_addr_t>& address) | |
13326 | { | |
13327 | Mutex::Locker lock(client_lock); | |
13328 | ||
181888fb FG |
13329 | if (unmounting) |
13330 | return -ENOTCONN; | |
13331 | ||
7c673cae FG |
13332 | Fh *f = get_filehandle(fd); |
13333 | if (!f) | |
13334 | return -EBADF; | |
13335 | Inode *in = f->inode.get(); | |
13336 | ||
13337 | // which object? | |
13338 | vector<ObjectExtent> extents; | |
13339 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, 1, | |
13340 | in->truncate_size, extents); | |
13341 | assert(extents.size() == 1); | |
13342 | ||
13343 | // now we have the object and its 'layout' | |
13344 | return objecter->with_osdmap([&](const OSDMap& o) { | |
13345 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
13346 | vector<int> osds; | |
13347 | o.pg_to_acting_osds(pg, osds); | |
13348 | if (osds.empty()) | |
13349 | return -EINVAL; | |
13350 | for (unsigned i = 0; i < osds.size(); i++) { | |
13351 | entity_addr_t addr = o.get_addr(osds[i]); | |
13352 | address.push_back(addr); | |
13353 | } | |
13354 | return 0; | |
13355 | }); | |
13356 | } | |
13357 | ||
13358 | int Client::get_osd_addr(int osd, entity_addr_t& addr) | |
13359 | { | |
13360 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13361 | |
13362 | if (unmounting) | |
13363 | return -ENOTCONN; | |
13364 | ||
7c673cae FG |
13365 | return objecter->with_osdmap([&](const OSDMap& o) { |
13366 | if (!o.exists(osd)) | |
13367 | return -ENOENT; | |
13368 | ||
13369 | addr = o.get_addr(osd); | |
13370 | return 0; | |
13371 | }); | |
13372 | } | |
13373 | ||
13374 | int Client::enumerate_layout(int fd, vector<ObjectExtent>& result, | |
13375 | loff_t length, loff_t offset) | |
13376 | { | |
13377 | Mutex::Locker lock(client_lock); | |
13378 | ||
181888fb FG |
13379 | if (unmounting) |
13380 | return -ENOTCONN; | |
13381 | ||
7c673cae FG |
13382 | Fh *f = get_filehandle(fd); |
13383 | if (!f) | |
13384 | return -EBADF; | |
13385 | Inode *in = f->inode.get(); | |
13386 | ||
13387 | // map to a list of extents | |
13388 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, length, in->truncate_size, result); | |
13389 | ||
13390 | ldout(cct, 3) << "enumerate_layout(" << fd << ", " << length << ", " << offset << ") = 0" << dendl; | |
13391 | return 0; | |
13392 | } | |
13393 | ||
13394 | ||
b32b8144 | 13395 | /* find an osd with the same ip. -ENXIO if none. */ |
7c673cae FG |
13396 | int Client::get_local_osd() |
13397 | { | |
13398 | Mutex::Locker lock(client_lock); | |
181888fb FG |
13399 | |
13400 | if (unmounting) | |
13401 | return -ENOTCONN; | |
13402 | ||
7c673cae FG |
13403 | objecter->with_osdmap([this](const OSDMap& o) { |
13404 | if (o.get_epoch() != local_osd_epoch) { | |
13405 | local_osd = o.find_osd_on_ip(messenger->get_myaddr()); | |
13406 | local_osd_epoch = o.get_epoch(); | |
13407 | } | |
13408 | }); | |
13409 | return local_osd; | |
13410 | } | |
13411 | ||
13412 | ||
13413 | ||
13414 | ||
13415 | ||
13416 | ||
13417 | // =============================== | |
13418 | ||
13419 | void Client::ms_handle_connect(Connection *con) | |
13420 | { | |
13421 | ldout(cct, 10) << "ms_handle_connect on " << con->get_peer_addr() << dendl; | |
13422 | } | |
13423 | ||
13424 | bool Client::ms_handle_reset(Connection *con) | |
13425 | { | |
13426 | ldout(cct, 0) << "ms_handle_reset on " << con->get_peer_addr() << dendl; | |
13427 | return false; | |
13428 | } | |
13429 | ||
13430 | void Client::ms_handle_remote_reset(Connection *con) | |
13431 | { | |
13432 | ldout(cct, 0) << "ms_handle_remote_reset on " << con->get_peer_addr() << dendl; | |
13433 | Mutex::Locker l(client_lock); | |
13434 | switch (con->get_peer_type()) { | |
13435 | case CEPH_ENTITY_TYPE_MDS: | |
13436 | { | |
13437 | // kludge to figure out which mds this is; fixme with a Connection* state | |
13438 | mds_rank_t mds = MDS_RANK_NONE; | |
13439 | MetaSession *s = NULL; | |
13440 | for (map<mds_rank_t,MetaSession*>::iterator p = mds_sessions.begin(); | |
13441 | p != mds_sessions.end(); | |
13442 | ++p) { | |
13443 | if (mdsmap->get_addr(p->first) == con->get_peer_addr()) { | |
13444 | mds = p->first; | |
13445 | s = p->second; | |
13446 | } | |
13447 | } | |
13448 | if (mds >= 0) { | |
d2e6a577 | 13449 | assert (s != NULL); |
7c673cae FG |
13450 | switch (s->state) { |
13451 | case MetaSession::STATE_CLOSING: | |
13452 | ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl; | |
13453 | _closed_mds_session(s); | |
13454 | break; | |
13455 | ||
13456 | case MetaSession::STATE_OPENING: | |
13457 | { | |
13458 | ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl; | |
13459 | list<Context*> waiters; | |
13460 | waiters.swap(s->waiting_for_open); | |
13461 | _closed_mds_session(s); | |
13462 | MetaSession *news = _get_or_open_mds_session(mds); | |
13463 | news->waiting_for_open.swap(waiters); | |
13464 | } | |
13465 | break; | |
13466 | ||
13467 | case MetaSession::STATE_OPEN: | |
13468 | { | |
13469 | const md_config_t *conf = cct->_conf; | |
13470 | if (conf->client_reconnect_stale) { | |
13471 | ldout(cct, 1) << "reset from mds we were open; close mds session for reconnect" << dendl; | |
13472 | _closed_mds_session(s); | |
13473 | } else { | |
13474 | ldout(cct, 1) << "reset from mds we were open; mark session as stale" << dendl; | |
13475 | s->state = MetaSession::STATE_STALE; | |
13476 | } | |
13477 | } | |
13478 | break; | |
13479 | ||
13480 | case MetaSession::STATE_NEW: | |
13481 | case MetaSession::STATE_CLOSED: | |
13482 | default: | |
13483 | break; | |
13484 | } | |
13485 | } | |
13486 | } | |
13487 | break; | |
13488 | } | |
13489 | } | |
13490 | ||
13491 | bool Client::ms_handle_refused(Connection *con) | |
13492 | { | |
13493 | ldout(cct, 1) << "ms_handle_refused on " << con->get_peer_addr() << dendl; | |
13494 | return false; | |
13495 | } | |
13496 | ||
13497 | bool Client::ms_get_authorizer(int dest_type, AuthAuthorizer **authorizer, bool force_new) | |
13498 | { | |
13499 | if (dest_type == CEPH_ENTITY_TYPE_MON) | |
13500 | return true; | |
13501 | *authorizer = monclient->build_authorizer(dest_type); | |
13502 | return true; | |
13503 | } | |
13504 | ||
13505 | Inode *Client::get_quota_root(Inode *in, const UserPerm& perms) | |
13506 | { | |
13507 | Inode *cur = in; | |
13508 | utime_t now = ceph_clock_now(); | |
13509 | ||
13510 | while (cur) { | |
13511 | if (cur != in && cur->quota.is_enable()) | |
13512 | break; | |
13513 | ||
13514 | Inode *parent_in = NULL; | |
13515 | if (!cur->dn_set.empty()) { | |
13516 | for (auto p = cur->dn_set.begin(); p != cur->dn_set.end(); ++p) { | |
13517 | Dentry *dn = *p; | |
13518 | if (dn->lease_mds >= 0 && | |
13519 | dn->lease_ttl > now && | |
13520 | mds_sessions.count(dn->lease_mds)) { | |
13521 | parent_in = dn->dir->parent_inode; | |
13522 | } else { | |
13523 | Inode *diri = dn->dir->parent_inode; | |
13524 | if (diri->caps_issued_mask(CEPH_CAP_FILE_SHARED) && | |
13525 | diri->shared_gen == dn->cap_shared_gen) { | |
13526 | parent_in = dn->dir->parent_inode; | |
13527 | } | |
13528 | } | |
13529 | if (parent_in) | |
13530 | break; | |
13531 | } | |
13532 | } else if (root_parents.count(cur)) { | |
13533 | parent_in = root_parents[cur].get(); | |
13534 | } | |
13535 | ||
13536 | if (parent_in) { | |
13537 | cur = parent_in; | |
13538 | continue; | |
13539 | } | |
13540 | ||
13541 | if (cur == root_ancestor) | |
13542 | break; | |
13543 | ||
181888fb FG |
13544 | // deleted inode |
13545 | if (cur->nlink == 0) { | |
13546 | cur = root_ancestor; | |
13547 | break; | |
13548 | } | |
13549 | ||
7c673cae FG |
13550 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
13551 | filepath path(cur->ino); | |
13552 | req->set_filepath(path); | |
13553 | req->set_inode(cur); | |
13554 | ||
13555 | InodeRef parent_ref; | |
13556 | int ret = make_request(req, perms, &parent_ref); | |
13557 | if (ret < 0) { | |
13558 | ldout(cct, 1) << __func__ << " " << in->vino() | |
13559 | << " failed to find parent of " << cur->vino() | |
13560 | << " err " << ret << dendl; | |
13561 | // FIXME: what to do? | |
13562 | cur = root_ancestor; | |
13563 | break; | |
13564 | } | |
13565 | ||
13566 | now = ceph_clock_now(); | |
13567 | if (cur == in) | |
13568 | cur = parent_ref.get(); | |
13569 | else | |
13570 | cur = in; // start over | |
13571 | } | |
13572 | ||
13573 | ldout(cct, 10) << __func__ << " " << in->vino() << " -> " << cur->vino() << dendl; | |
13574 | return cur; | |
13575 | } | |
13576 | ||
13577 | /** | |
13578 | * Traverse quota ancestors of the Inode, return true | |
13579 | * if any of them passes the passed function | |
13580 | */ | |
13581 | bool Client::check_quota_condition(Inode *in, const UserPerm& perms, | |
13582 | std::function<bool (const Inode &in)> test) | |
13583 | { | |
13584 | while (true) { | |
13585 | assert(in != NULL); | |
13586 | if (test(*in)) { | |
13587 | return true; | |
13588 | } | |
13589 | ||
13590 | if (in == root_ancestor) { | |
13591 | // We're done traversing, drop out | |
13592 | return false; | |
13593 | } else { | |
13594 | // Continue up the tree | |
13595 | in = get_quota_root(in, perms); | |
13596 | } | |
13597 | } | |
13598 | ||
13599 | return false; | |
13600 | } | |
13601 | ||
13602 | bool Client::is_quota_files_exceeded(Inode *in, const UserPerm& perms) | |
13603 | { | |
13604 | return check_quota_condition(in, perms, | |
13605 | [](const Inode &in) { | |
13606 | return in.quota.max_files && in.rstat.rsize() >= in.quota.max_files; | |
13607 | }); | |
13608 | } | |
13609 | ||
13610 | bool Client::is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
13611 | const UserPerm& perms) | |
13612 | { | |
13613 | return check_quota_condition(in, perms, | |
13614 | [&new_bytes](const Inode &in) { | |
13615 | return in.quota.max_bytes && (in.rstat.rbytes + new_bytes) | |
13616 | > in.quota.max_bytes; | |
13617 | }); | |
13618 | } | |
13619 | ||
13620 | bool Client::is_quota_bytes_approaching(Inode *in, const UserPerm& perms) | |
13621 | { | |
13622 | return check_quota_condition(in, perms, | |
13623 | [](const Inode &in) { | |
13624 | if (in.quota.max_bytes) { | |
13625 | if (in.rstat.rbytes >= in.quota.max_bytes) { | |
13626 | return true; | |
13627 | } | |
13628 | ||
13629 | assert(in.size >= in.reported_size); | |
13630 | const uint64_t space = in.quota.max_bytes - in.rstat.rbytes; | |
13631 | const uint64_t size = in.size - in.reported_size; | |
13632 | return (space >> 4) < size; | |
13633 | } else { | |
13634 | return false; | |
13635 | } | |
13636 | }); | |
13637 | } | |
13638 | ||
13639 | enum { | |
13640 | POOL_CHECKED = 1, | |
13641 | POOL_CHECKING = 2, | |
13642 | POOL_READ = 4, | |
13643 | POOL_WRITE = 8, | |
13644 | }; | |
13645 | ||
13646 | int Client::check_pool_perm(Inode *in, int need) | |
13647 | { | |
13648 | if (!cct->_conf->client_check_pool_perm) | |
13649 | return 0; | |
13650 | ||
13651 | int64_t pool_id = in->layout.pool_id; | |
13652 | std::string pool_ns = in->layout.pool_ns; | |
13653 | std::pair<int64_t, std::string> perm_key(pool_id, pool_ns); | |
13654 | int have = 0; | |
13655 | while (true) { | |
13656 | auto it = pool_perms.find(perm_key); | |
13657 | if (it == pool_perms.end()) | |
13658 | break; | |
13659 | if (it->second == POOL_CHECKING) { | |
13660 | // avoid concurrent checkings | |
13661 | wait_on_list(waiting_for_pool_perm); | |
13662 | } else { | |
13663 | have = it->second; | |
13664 | assert(have & POOL_CHECKED); | |
13665 | break; | |
13666 | } | |
13667 | } | |
13668 | ||
13669 | if (!have) { | |
13670 | if (in->snapid != CEPH_NOSNAP) { | |
13671 | // pool permission check needs to write to the first object. But for snapshot, | |
13672 | // head of the first object may have alread been deleted. To avoid creating | |
13673 | // orphan object, skip the check for now. | |
13674 | return 0; | |
13675 | } | |
13676 | ||
13677 | pool_perms[perm_key] = POOL_CHECKING; | |
13678 | ||
13679 | char oid_buf[32]; | |
13680 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (unsigned long long)in->ino); | |
13681 | object_t oid = oid_buf; | |
13682 | ||
13683 | SnapContext nullsnapc; | |
13684 | ||
13685 | C_SaferCond rd_cond; | |
13686 | ObjectOperation rd_op; | |
13687 | rd_op.stat(NULL, (ceph::real_time*)nullptr, NULL); | |
13688 | ||
13689 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), rd_op, | |
13690 | nullsnapc, ceph::real_clock::now(), 0, &rd_cond); | |
13691 | ||
13692 | C_SaferCond wr_cond; | |
13693 | ObjectOperation wr_op; | |
13694 | wr_op.create(true); | |
13695 | ||
13696 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), wr_op, | |
13697 | nullsnapc, ceph::real_clock::now(), 0, &wr_cond); | |
13698 | ||
13699 | client_lock.Unlock(); | |
13700 | int rd_ret = rd_cond.wait(); | |
13701 | int wr_ret = wr_cond.wait(); | |
13702 | client_lock.Lock(); | |
13703 | ||
13704 | bool errored = false; | |
13705 | ||
13706 | if (rd_ret == 0 || rd_ret == -ENOENT) | |
13707 | have |= POOL_READ; | |
13708 | else if (rd_ret != -EPERM) { | |
13709 | ldout(cct, 10) << "check_pool_perm on pool " << pool_id << " ns " << pool_ns | |
13710 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; | |
13711 | errored = true; | |
13712 | } | |
13713 | ||
13714 | if (wr_ret == 0 || wr_ret == -EEXIST) | |
13715 | have |= POOL_WRITE; | |
13716 | else if (wr_ret != -EPERM) { | |
13717 | ldout(cct, 10) << "check_pool_perm on pool " << pool_id << " ns " << pool_ns | |
13718 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; | |
13719 | errored = true; | |
13720 | } | |
13721 | ||
13722 | if (errored) { | |
13723 | // Indeterminate: erase CHECKING state so that subsequent calls re-check. | |
13724 | // Raise EIO because actual error code might be misleading for | |
13725 | // userspace filesystem user. | |
13726 | pool_perms.erase(perm_key); | |
13727 | signal_cond_list(waiting_for_pool_perm); | |
13728 | return -EIO; | |
13729 | } | |
13730 | ||
13731 | pool_perms[perm_key] = have | POOL_CHECKED; | |
13732 | signal_cond_list(waiting_for_pool_perm); | |
13733 | } | |
13734 | ||
13735 | if ((need & CEPH_CAP_FILE_RD) && !(have & POOL_READ)) { | |
13736 | ldout(cct, 10) << "check_pool_perm on pool " << pool_id << " ns " << pool_ns | |
13737 | << " need " << ccap_string(need) << ", but no read perm" << dendl; | |
13738 | return -EPERM; | |
13739 | } | |
13740 | if ((need & CEPH_CAP_FILE_WR) && !(have & POOL_WRITE)) { | |
13741 | ldout(cct, 10) << "check_pool_perm on pool " << pool_id << " ns " << pool_ns | |
13742 | << " need " << ccap_string(need) << ", but no write perm" << dendl; | |
13743 | return -EPERM; | |
13744 | } | |
13745 | ||
13746 | return 0; | |
13747 | } | |
13748 | ||
13749 | int Client::_posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want) | |
13750 | { | |
13751 | if (acl_type == POSIX_ACL) { | |
13752 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
13753 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
13754 | ||
13755 | return posix_acl_permits(access_acl, in->uid, in->gid, perms, want); | |
13756 | } | |
13757 | } | |
13758 | return -EAGAIN; | |
13759 | } | |
13760 | ||
13761 | int Client::_posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms) | |
13762 | { | |
13763 | if (acl_type == NO_ACL) | |
13764 | return 0; | |
13765 | ||
13766 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
13767 | if (r < 0) | |
13768 | goto out; | |
13769 | ||
13770 | if (acl_type == POSIX_ACL) { | |
13771 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
13772 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
13773 | bufferptr acl(access_acl.c_str(), access_acl.length()); | |
13774 | r = posix_acl_access_chmod(acl, mode); | |
13775 | if (r < 0) | |
13776 | goto out; | |
13777 | r = _do_setxattr(in, ACL_EA_ACCESS, acl.c_str(), acl.length(), 0, perms); | |
13778 | } else { | |
13779 | r = 0; | |
13780 | } | |
13781 | } | |
13782 | out: | |
13783 | ldout(cct, 10) << __func__ << " ino " << in->ino << " result=" << r << dendl; | |
13784 | return r; | |
13785 | } | |
13786 | ||
13787 | int Client::_posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
13788 | const UserPerm& perms) | |
13789 | { | |
13790 | if (acl_type == NO_ACL) | |
13791 | return 0; | |
13792 | ||
13793 | if (S_ISLNK(*mode)) | |
13794 | return 0; | |
13795 | ||
13796 | int r = _getattr(dir, CEPH_STAT_CAP_XATTR, perms, dir->xattr_version == 0); | |
13797 | if (r < 0) | |
13798 | goto out; | |
13799 | ||
13800 | if (acl_type == POSIX_ACL) { | |
13801 | if (dir->xattrs.count(ACL_EA_DEFAULT)) { | |
13802 | map<string, bufferptr> xattrs; | |
13803 | ||
13804 | const bufferptr& default_acl = dir->xattrs[ACL_EA_DEFAULT]; | |
13805 | bufferptr acl(default_acl.c_str(), default_acl.length()); | |
13806 | r = posix_acl_inherit_mode(acl, mode); | |
13807 | if (r < 0) | |
13808 | goto out; | |
13809 | ||
13810 | if (r > 0) { | |
13811 | r = posix_acl_equiv_mode(acl.c_str(), acl.length(), mode); | |
13812 | if (r < 0) | |
13813 | goto out; | |
13814 | if (r > 0) | |
13815 | xattrs[ACL_EA_ACCESS] = acl; | |
13816 | } | |
13817 | ||
13818 | if (S_ISDIR(*mode)) | |
13819 | xattrs[ACL_EA_DEFAULT] = dir->xattrs[ACL_EA_DEFAULT]; | |
13820 | ||
13821 | r = xattrs.size(); | |
13822 | if (r > 0) | |
13823 | ::encode(xattrs, xattrs_bl); | |
13824 | } else { | |
13825 | if (umask_cb) | |
13826 | *mode &= ~umask_cb(callback_handle); | |
13827 | r = 0; | |
13828 | } | |
13829 | } | |
13830 | out: | |
13831 | ldout(cct, 10) << __func__ << " dir ino " << dir->ino << " result=" << r << dendl; | |
13832 | return r; | |
13833 | } | |
13834 | ||
13835 | void Client::set_filer_flags(int flags) | |
13836 | { | |
13837 | Mutex::Locker l(client_lock); | |
13838 | assert(flags == 0 || | |
13839 | flags == CEPH_OSD_FLAG_LOCALIZE_READS); | |
13840 | objecter->add_global_op_flags(flags); | |
13841 | } | |
13842 | ||
13843 | void Client::clear_filer_flags(int flags) | |
13844 | { | |
13845 | Mutex::Locker l(client_lock); | |
13846 | assert(flags == CEPH_OSD_FLAG_LOCALIZE_READS); | |
13847 | objecter->clear_global_op_flag(flags); | |
13848 | } | |
13849 | ||
13850 | /** | |
13851 | * This is included in cap release messages, to cause | |
13852 | * the MDS to wait until this OSD map epoch. It is necessary | |
13853 | * in corner cases where we cancel RADOS ops, so that | |
13854 | * nobody else tries to do IO to the same objects in | |
13855 | * the same epoch as the cancelled ops. | |
13856 | */ | |
13857 | void Client::set_cap_epoch_barrier(epoch_t e) | |
13858 | { | |
13859 | ldout(cct, 5) << __func__ << " epoch = " << e << dendl; | |
13860 | cap_epoch_barrier = e; | |
13861 | } | |
13862 | ||
13863 | const char** Client::get_tracked_conf_keys() const | |
13864 | { | |
13865 | static const char* keys[] = { | |
13866 | "client_cache_size", | |
13867 | "client_cache_mid", | |
13868 | "client_acl_type", | |
b32b8144 FG |
13869 | "client_deleg_timeout", |
13870 | "client_deleg_break_on_open", | |
7c673cae FG |
13871 | NULL |
13872 | }; | |
13873 | return keys; | |
13874 | } | |
13875 | ||
13876 | void Client::handle_conf_change(const struct md_config_t *conf, | |
13877 | const std::set <std::string> &changed) | |
13878 | { | |
13879 | Mutex::Locker lock(client_lock); | |
13880 | ||
181888fb | 13881 | if (changed.count("client_cache_mid")) { |
7c673cae FG |
13882 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
13883 | } | |
13884 | if (changed.count("client_acl_type")) { | |
13885 | acl_type = NO_ACL; | |
13886 | if (cct->_conf->client_acl_type == "posix_acl") | |
13887 | acl_type = POSIX_ACL; | |
13888 | } | |
13889 | } | |
13890 | ||
13891 | void Client::init_groups(UserPerm *perms) | |
13892 | { | |
13893 | gid_t *sgids; | |
13894 | int count = _getgrouplist(&sgids, perms->uid(), perms->gid()); | |
13895 | perms->init_gids(sgids, count); | |
13896 | } | |
13897 | ||
13898 | void intrusive_ptr_add_ref(Inode *in) | |
13899 | { | |
13900 | in->get(); | |
13901 | } | |
13902 | ||
13903 | void intrusive_ptr_release(Inode *in) | |
13904 | { | |
13905 | in->client->put_inode(in); | |
13906 | } | |
13907 | ||
13908 | mds_rank_t Client::_get_random_up_mds() const | |
13909 | { | |
13910 | assert(client_lock.is_locked_by_me()); | |
13911 | ||
13912 | std::set<mds_rank_t> up; | |
13913 | mdsmap->get_up_mds_set(up); | |
13914 | ||
13915 | if (up.empty()) | |
13916 | return MDS_RANK_NONE; | |
13917 | std::set<mds_rank_t>::const_iterator p = up.begin(); | |
13918 | for (int n = rand() % up.size(); n; n--) | |
13919 | ++p; | |
13920 | return *p; | |
13921 | } | |
13922 | ||
13923 | ||
13924 | StandaloneClient::StandaloneClient(Messenger *m, MonClient *mc) | |
13925 | : Client(m, mc, new Objecter(m->cct, m, mc, NULL, 0, 0)) | |
13926 | { | |
13927 | monclient->set_messenger(m); | |
13928 | objecter->set_client_incarnation(0); | |
13929 | } | |
13930 | ||
13931 | StandaloneClient::~StandaloneClient() | |
13932 | { | |
13933 | delete objecter; | |
13934 | objecter = nullptr; | |
13935 | } | |
13936 | ||
13937 | int StandaloneClient::init() | |
13938 | { | |
13939 | timer.init(); | |
13940 | objectcacher->start(); | |
13941 | objecter->init(); | |
13942 | ||
13943 | client_lock.Lock(); | |
13944 | assert(!initialized); | |
13945 | ||
13946 | messenger->add_dispatcher_tail(objecter); | |
13947 | messenger->add_dispatcher_tail(this); | |
13948 | ||
13949 | monclient->set_want_keys(CEPH_ENTITY_TYPE_MDS | CEPH_ENTITY_TYPE_OSD); | |
13950 | int r = monclient->init(); | |
13951 | if (r < 0) { | |
13952 | // need to do cleanup because we're in an intermediate init state | |
13953 | timer.shutdown(); | |
13954 | client_lock.Unlock(); | |
13955 | objecter->shutdown(); | |
13956 | objectcacher->stop(); | |
13957 | monclient->shutdown(); | |
13958 | return r; | |
13959 | } | |
13960 | objecter->start(); | |
13961 | ||
13962 | client_lock.Unlock(); | |
13963 | _finish_init(); | |
13964 | ||
13965 | return 0; | |
13966 | } | |
13967 | ||
13968 | void StandaloneClient::shutdown() | |
13969 | { | |
13970 | Client::shutdown(); | |
13971 | objecter->shutdown(); | |
13972 | monclient->shutdown(); | |
13973 | } |