]>
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> | |
11fdf7f2 | 21 | #include <string.h> |
7c673cae FG |
22 | #include <sys/stat.h> |
23 | #include <sys/param.h> | |
24 | #include <fcntl.h> | |
25 | #include <sys/file.h> | |
26 | #include <sys/utsname.h> | |
27 | #include <sys/uio.h> | |
28 | ||
29 | #include <boost/lexical_cast.hpp> | |
30 | #include <boost/fusion/include/std_pair.hpp> | |
31 | ||
32 | #if defined(__FreeBSD__) | |
33 | #define XATTR_CREATE 0x1 | |
34 | #define XATTR_REPLACE 0x2 | |
35 | #else | |
36 | #include <sys/xattr.h> | |
37 | #endif | |
38 | ||
39 | #if defined(__linux__) | |
40 | #include <linux/falloc.h> | |
41 | #endif | |
42 | ||
43 | #include <sys/statvfs.h> | |
44 | ||
45 | #include "common/config.h" | |
46 | #include "common/version.h" | |
47 | ||
11fdf7f2 TL |
48 | #include "mon/MonClient.h" |
49 | ||
50 | #include "messages/MClientCaps.h" | |
51 | #include "messages/MClientLease.h" | |
52 | #include "messages/MClientQuota.h" | |
53 | #include "messages/MClientReclaim.h" | |
54 | #include "messages/MClientReclaimReply.h" | |
7c673cae | 55 | #include "messages/MClientReconnect.h" |
11fdf7f2 | 56 | #include "messages/MClientReply.h" |
7c673cae FG |
57 | #include "messages/MClientRequest.h" |
58 | #include "messages/MClientRequestForward.h" | |
11fdf7f2 | 59 | #include "messages/MClientSession.h" |
7c673cae FG |
60 | #include "messages/MClientSnap.h" |
61 | #include "messages/MCommandReply.h" | |
7c673cae FG |
62 | #include "messages/MFSMap.h" |
63 | #include "messages/MFSMapUser.h" | |
11fdf7f2 TL |
64 | #include "messages/MMDSMap.h" |
65 | #include "messages/MOSDMap.h" | |
7c673cae FG |
66 | |
67 | #include "mds/flock.h" | |
11fdf7f2 | 68 | #include "mds/cephfs_features.h" |
7c673cae FG |
69 | #include "osd/OSDMap.h" |
70 | #include "osdc/Filer.h" | |
71 | ||
72 | #include "common/Cond.h" | |
7c673cae FG |
73 | #include "common/perf_counters.h" |
74 | #include "common/admin_socket.h" | |
75 | #include "common/errno.h" | |
76 | #include "include/str_list.h" | |
77 | ||
78 | #define dout_subsys ceph_subsys_client | |
79 | ||
80 | #include "include/lru.h" | |
81 | #include "include/compat.h" | |
82 | #include "include/stringify.h" | |
83 | ||
84 | #include "Client.h" | |
85 | #include "Inode.h" | |
86 | #include "Dentry.h" | |
b32b8144 | 87 | #include "Delegation.h" |
7c673cae FG |
88 | #include "Dir.h" |
89 | #include "ClientSnapRealm.h" | |
90 | #include "Fh.h" | |
91 | #include "MetaSession.h" | |
92 | #include "MetaRequest.h" | |
93 | #include "ObjecterWriteback.h" | |
94 | #include "posix_acl.h" | |
95 | ||
11fdf7f2 | 96 | #include "include/ceph_assert.h" |
7c673cae FG |
97 | #include "include/stat.h" |
98 | ||
99 | #include "include/cephfs/ceph_statx.h" | |
100 | ||
101 | #if HAVE_GETGROUPLIST | |
102 | #include <grp.h> | |
103 | #include <pwd.h> | |
104 | #include <unistd.h> | |
105 | #endif | |
106 | ||
107 | #undef dout_prefix | |
108 | #define dout_prefix *_dout << "client." << whoami << " " | |
109 | ||
110 | #define tout(cct) if (!cct->_conf->client_trace.empty()) traceout | |
111 | ||
112 | // FreeBSD fails to define this | |
113 | #ifndef O_DSYNC | |
114 | #define O_DSYNC 0x0 | |
115 | #endif | |
116 | // Darwin fails to define this | |
117 | #ifndef O_RSYNC | |
118 | #define O_RSYNC 0x0 | |
119 | #endif | |
120 | ||
121 | #ifndef O_DIRECT | |
122 | #define O_DIRECT 0x0 | |
123 | #endif | |
124 | ||
125 | #define DEBUG_GETATTR_CAPS (CEPH_CAP_XATTR_SHARED) | |
126 | ||
127 | void client_flush_set_callback(void *p, ObjectCacher::ObjectSet *oset) | |
128 | { | |
129 | Client *client = static_cast<Client*>(p); | |
130 | client->flush_set_callback(oset); | |
131 | } | |
132 | ||
133 | ||
134 | // ------------- | |
135 | ||
136 | Client::CommandHook::CommandHook(Client *client) : | |
137 | m_client(client) | |
138 | { | |
139 | } | |
140 | ||
9f95a23c TL |
141 | int Client::CommandHook::call( |
142 | std::string_view command, | |
143 | const cmdmap_t& cmdmap, | |
144 | Formatter *f, | |
145 | std::ostream& errss, | |
146 | bufferlist& out) | |
7c673cae | 147 | { |
7c673cae | 148 | f->open_object_section("result"); |
9f95a23c TL |
149 | { |
150 | std::lock_guard l{m_client->client_lock}; | |
151 | if (command == "mds_requests") | |
152 | m_client->dump_mds_requests(f); | |
153 | else if (command == "mds_sessions") | |
154 | m_client->dump_mds_sessions(f); | |
155 | else if (command == "dump_cache") | |
156 | m_client->dump_cache(f); | |
157 | else if (command == "kick_stale_sessions") | |
158 | m_client->_kick_stale_sessions(); | |
159 | else if (command == "status") | |
160 | m_client->dump_status(f); | |
161 | else | |
162 | ceph_abort_msg("bad command registered"); | |
163 | } | |
7c673cae | 164 | f->close_section(); |
9f95a23c | 165 | return 0; |
7c673cae FG |
166 | } |
167 | ||
168 | ||
169 | // ------------- | |
170 | ||
171 | dir_result_t::dir_result_t(Inode *in, const UserPerm& perms) | |
172 | : inode(in), offset(0), next_offset(2), | |
173 | release_count(0), ordered_count(0), cache_index(0), start_shared_gen(0), | |
174 | perms(perms) | |
175 | { } | |
176 | ||
177 | void Client::_reset_faked_inos() | |
178 | { | |
179 | ino_t start = 1024; | |
180 | free_faked_inos.clear(); | |
181 | free_faked_inos.insert(start, (uint32_t)-1 - start + 1); | |
182 | last_used_faked_ino = 0; | |
11fdf7f2 | 183 | last_used_faked_root = 0; |
7c673cae FG |
184 | _use_faked_inos = sizeof(ino_t) < 8 || cct->_conf->client_use_faked_inos; |
185 | } | |
186 | ||
187 | void Client::_assign_faked_ino(Inode *in) | |
188 | { | |
11fdf7f2 TL |
189 | if (0 == last_used_faked_ino) |
190 | last_used_faked_ino = last_used_faked_ino + 2048; // start(1024)~2048 reserved for _assign_faked_root | |
7c673cae FG |
191 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
192 | if (it == free_faked_inos.end() && last_used_faked_ino > 0) { | |
11fdf7f2 | 193 | last_used_faked_ino = 2048; |
7c673cae FG |
194 | it = free_faked_inos.lower_bound(last_used_faked_ino + 1); |
195 | } | |
11fdf7f2 | 196 | ceph_assert(it != free_faked_inos.end()); |
7c673cae | 197 | if (last_used_faked_ino < it.get_start()) { |
11fdf7f2 | 198 | ceph_assert(it.get_len() > 0); |
7c673cae FG |
199 | last_used_faked_ino = it.get_start(); |
200 | } else { | |
201 | ++last_used_faked_ino; | |
11fdf7f2 | 202 | ceph_assert(it.get_start() + it.get_len() > last_used_faked_ino); |
7c673cae FG |
203 | } |
204 | in->faked_ino = last_used_faked_ino; | |
205 | free_faked_inos.erase(in->faked_ino); | |
206 | faked_ino_map[in->faked_ino] = in->vino(); | |
207 | } | |
208 | ||
11fdf7f2 TL |
209 | /* |
210 | * In the faked mode, if you export multiple subdirectories, | |
211 | * you will see that the inode numbers of the exported subdirectories | |
212 | * are the same. so we distinguish the mount point by reserving | |
213 | * the "fake ids" between "1024~2048" and combining the last | |
214 | * 10bits(0x3ff) of the "root inodes". | |
215 | */ | |
216 | void Client::_assign_faked_root(Inode *in) | |
217 | { | |
218 | interval_set<ino_t>::const_iterator it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
219 | if (it == free_faked_inos.end() && last_used_faked_root > 0) { | |
220 | last_used_faked_root = 0; | |
221 | it = free_faked_inos.lower_bound(last_used_faked_root + 1); | |
222 | } | |
223 | assert(it != free_faked_inos.end()); | |
224 | vinodeno_t inode_info = in->vino(); | |
225 | uint64_t inode_num = (uint64_t)inode_info.ino; | |
226 | ldout(cct, 10) << "inode_num " << inode_num << "inode_num & 0x3ff=" << (inode_num & 0x3ff)<< dendl; | |
227 | last_used_faked_root = it.get_start() + (inode_num & 0x3ff); // 0x3ff mask and get_start will not exceed 2048 | |
228 | assert(it.get_start() + it.get_len() > last_used_faked_root); | |
229 | ||
230 | in->faked_ino = last_used_faked_root; | |
231 | free_faked_inos.erase(in->faked_ino); | |
232 | faked_ino_map[in->faked_ino] = in->vino(); | |
233 | } | |
234 | ||
7c673cae FG |
235 | void Client::_release_faked_ino(Inode *in) |
236 | { | |
237 | free_faked_inos.insert(in->faked_ino); | |
238 | faked_ino_map.erase(in->faked_ino); | |
239 | } | |
240 | ||
241 | vinodeno_t Client::_map_faked_ino(ino_t ino) | |
242 | { | |
243 | vinodeno_t vino; | |
244 | if (ino == 1) | |
245 | vino = root->vino(); | |
246 | else if (faked_ino_map.count(ino)) | |
247 | vino = faked_ino_map[ino]; | |
248 | else | |
249 | vino = vinodeno_t(0, CEPH_NOSNAP); | |
11fdf7f2 | 250 | ldout(cct, 10) << __func__ << " " << ino << " -> " << vino << dendl; |
7c673cae FG |
251 | return vino; |
252 | } | |
253 | ||
254 | vinodeno_t Client::map_faked_ino(ino_t ino) | |
255 | { | |
11fdf7f2 | 256 | std::lock_guard lock(client_lock); |
7c673cae FG |
257 | return _map_faked_ino(ino); |
258 | } | |
259 | ||
260 | // cons/des | |
261 | ||
262 | Client::Client(Messenger *m, MonClient *mc, Objecter *objecter_) | |
263 | : Dispatcher(m->cct), | |
7c673cae | 264 | timer(m->cct, client_lock), |
11fdf7f2 TL |
265 | messenger(m), |
266 | monclient(mc), | |
267 | objecter(objecter_), | |
268 | whoami(mc->get_global_id()), | |
7c673cae FG |
269 | async_ino_invalidator(m->cct), |
270 | async_dentry_invalidator(m->cct), | |
271 | interrupt_finisher(m->cct), | |
272 | remount_finisher(m->cct), | |
273 | objecter_finisher(m->cct), | |
11fdf7f2 TL |
274 | m_command_hook(this), |
275 | fscid(0) | |
7c673cae FG |
276 | { |
277 | _reset_faked_inos(); | |
7c673cae | 278 | |
7c673cae FG |
279 | user_id = cct->_conf->client_mount_uid; |
280 | group_id = cct->_conf->client_mount_gid; | |
92f5a8d4 TL |
281 | fuse_default_permissions = cct->_conf.get_val<bool>( |
282 | "fuse_default_permissions"); | |
7c673cae | 283 | |
7c673cae FG |
284 | if (cct->_conf->client_acl_type == "posix_acl") |
285 | acl_type = POSIX_ACL; | |
286 | ||
7c673cae FG |
287 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
288 | ||
289 | // file handles | |
290 | free_fd_set.insert(10, 1<<30); | |
291 | ||
292 | mdsmap.reset(new MDSMap); | |
293 | ||
294 | // osd interfaces | |
295 | writeback_handler.reset(new ObjecterWriteback(objecter, &objecter_finisher, | |
296 | &client_lock)); | |
297 | objectcacher.reset(new ObjectCacher(cct, "libcephfs", *writeback_handler, client_lock, | |
298 | client_flush_set_callback, // all commit callback | |
299 | (void*)this, | |
300 | cct->_conf->client_oc_size, | |
301 | cct->_conf->client_oc_max_objects, | |
302 | cct->_conf->client_oc_max_dirty, | |
303 | cct->_conf->client_oc_target_dirty, | |
304 | cct->_conf->client_oc_max_dirty_age, | |
305 | true)); | |
306 | objecter_finisher.start(); | |
307 | filer.reset(new Filer(objecter, &objecter_finisher)); | |
31f18b77 | 308 | objecter->enable_blacklist_events(); |
7c673cae FG |
309 | } |
310 | ||
311 | ||
312 | Client::~Client() | |
313 | { | |
9f95a23c | 314 | ceph_assert(ceph_mutex_is_not_locked(client_lock)); |
7c673cae | 315 | |
31f18b77 FG |
316 | // It is necessary to hold client_lock, because any inode destruction |
317 | // may call into ObjectCacher, which asserts that it's lock (which is | |
318 | // client_lock) is held. | |
9f95a23c | 319 | std::lock_guard l{client_lock}; |
7c673cae FG |
320 | tear_down_cache(); |
321 | } | |
322 | ||
323 | void Client::tear_down_cache() | |
324 | { | |
325 | // fd's | |
326 | for (ceph::unordered_map<int, Fh*>::iterator it = fd_map.begin(); | |
327 | it != fd_map.end(); | |
328 | ++it) { | |
329 | Fh *fh = it->second; | |
11fdf7f2 | 330 | ldout(cct, 1) << __func__ << " forcing close of fh " << it->first << " ino " << fh->inode->ino << dendl; |
7c673cae FG |
331 | _release_fh(fh); |
332 | } | |
333 | fd_map.clear(); | |
334 | ||
335 | while (!opened_dirs.empty()) { | |
336 | dir_result_t *dirp = *opened_dirs.begin(); | |
11fdf7f2 | 337 | ldout(cct, 1) << __func__ << " forcing close of dir " << dirp << " ino " << dirp->inode->ino << dendl; |
7c673cae FG |
338 | _closedir(dirp); |
339 | } | |
340 | ||
341 | // caps! | |
342 | // *** FIXME *** | |
343 | ||
344 | // empty lru | |
7c673cae | 345 | trim_cache(); |
11fdf7f2 | 346 | ceph_assert(lru.lru_get_size() == 0); |
7c673cae FG |
347 | |
348 | // close root ino | |
11fdf7f2 | 349 | ceph_assert(inode_map.size() <= 1 + root_parents.size()); |
7c673cae FG |
350 | if (root && inode_map.size() == 1 + root_parents.size()) { |
351 | delete root; | |
352 | root = 0; | |
353 | root_ancestor = 0; | |
354 | while (!root_parents.empty()) | |
355 | root_parents.erase(root_parents.begin()); | |
356 | inode_map.clear(); | |
357 | _reset_faked_inos(); | |
358 | } | |
359 | ||
11fdf7f2 | 360 | ceph_assert(inode_map.empty()); |
7c673cae FG |
361 | } |
362 | ||
363 | inodeno_t Client::get_root_ino() | |
364 | { | |
11fdf7f2 | 365 | std::lock_guard l(client_lock); |
7c673cae FG |
366 | if (use_faked_inos()) |
367 | return root->faked_ino; | |
368 | else | |
369 | return root->ino; | |
370 | } | |
371 | ||
372 | Inode *Client::get_root() | |
373 | { | |
11fdf7f2 | 374 | std::lock_guard l(client_lock); |
7c673cae FG |
375 | root->ll_get(); |
376 | return root; | |
377 | } | |
378 | ||
379 | ||
380 | // debug crapola | |
381 | ||
382 | void Client::dump_inode(Formatter *f, Inode *in, set<Inode*>& did, bool disconnected) | |
383 | { | |
384 | filepath path; | |
385 | in->make_long_path(path); | |
386 | ldout(cct, 1) << "dump_inode: " | |
387 | << (disconnected ? "DISCONNECTED ":"") | |
388 | << "inode " << in->ino | |
389 | << " " << path | |
390 | << " ref " << in->get_num_ref() | |
391 | << *in << dendl; | |
392 | ||
393 | if (f) { | |
394 | f->open_object_section("inode"); | |
395 | f->dump_stream("path") << path; | |
396 | if (disconnected) | |
397 | f->dump_int("disconnected", 1); | |
398 | in->dump(f); | |
399 | f->close_section(); | |
400 | } | |
401 | ||
402 | did.insert(in); | |
403 | if (in->dir) { | |
404 | ldout(cct, 1) << " dir " << in->dir << " size " << in->dir->dentries.size() << dendl; | |
405 | for (ceph::unordered_map<string, Dentry*>::iterator it = in->dir->dentries.begin(); | |
406 | it != in->dir->dentries.end(); | |
407 | ++it) { | |
408 | ldout(cct, 1) << " " << in->ino << " dn " << it->first << " " << it->second << " ref " << it->second->ref << dendl; | |
409 | if (f) { | |
410 | f->open_object_section("dentry"); | |
411 | it->second->dump(f); | |
412 | f->close_section(); | |
413 | } | |
414 | if (it->second->inode) | |
415 | dump_inode(f, it->second->inode.get(), did, false); | |
416 | } | |
417 | } | |
418 | } | |
419 | ||
420 | void Client::dump_cache(Formatter *f) | |
421 | { | |
422 | set<Inode*> did; | |
423 | ||
11fdf7f2 | 424 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
425 | |
426 | if (f) | |
427 | f->open_array_section("cache"); | |
428 | ||
429 | if (root) | |
430 | dump_inode(f, root, did, true); | |
431 | ||
432 | // make a second pass to catch anything disconnected | |
433 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
434 | it != inode_map.end(); | |
435 | ++it) { | |
436 | if (did.count(it->second)) | |
437 | continue; | |
438 | dump_inode(f, it->second, did, true); | |
439 | } | |
440 | ||
441 | if (f) | |
442 | f->close_section(); | |
443 | } | |
444 | ||
445 | void Client::dump_status(Formatter *f) | |
446 | { | |
9f95a23c | 447 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
448 | |
449 | ldout(cct, 1) << __func__ << dendl; | |
450 | ||
451 | const epoch_t osd_epoch | |
452 | = objecter->with_osdmap(std::mem_fn(&OSDMap::get_epoch)); | |
453 | ||
454 | if (f) { | |
455 | f->open_object_section("metadata"); | |
456 | for (const auto& kv : metadata) | |
457 | f->dump_string(kv.first.c_str(), kv.second); | |
458 | f->close_section(); | |
459 | ||
460 | f->dump_int("dentry_count", lru.lru_get_size()); | |
461 | f->dump_int("dentry_pinned_count", lru.lru_get_num_pinned()); | |
462 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 463 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 | 464 | f->dump_object("inst", inst); |
11fdf7f2 TL |
465 | f->dump_object("addr", inst.addr); |
466 | f->dump_stream("inst_str") << inst.name << " " << inst.addr.get_legacy_str(); | |
467 | f->dump_string("addr_str", inst.addr.get_legacy_str()); | |
7c673cae FG |
468 | f->dump_int("inode_count", inode_map.size()); |
469 | f->dump_int("mds_epoch", mdsmap->get_epoch()); | |
470 | f->dump_int("osd_epoch", osd_epoch); | |
471 | f->dump_int("osd_epoch_barrier", cap_epoch_barrier); | |
f64942e4 | 472 | f->dump_bool("blacklisted", blacklisted); |
7c673cae FG |
473 | } |
474 | } | |
475 | ||
476 | int Client::init() | |
477 | { | |
478 | timer.init(); | |
479 | objectcacher->start(); | |
9f95a23c TL |
480 | { |
481 | std::lock_guard l{client_lock}; | |
482 | ceph_assert(!initialized); | |
483 | messenger->add_dispatcher_tail(this); | |
484 | } | |
7c673cae FG |
485 | _finish_init(); |
486 | return 0; | |
487 | } | |
488 | ||
489 | void Client::_finish_init() | |
490 | { | |
9f95a23c TL |
491 | { |
492 | std::lock_guard l{client_lock}; | |
493 | // logger | |
494 | PerfCountersBuilder plb(cct, "client", l_c_first, l_c_last); | |
495 | plb.add_time_avg(l_c_reply, "reply", "Latency of receiving a reply on metadata request"); | |
496 | plb.add_time_avg(l_c_lat, "lat", "Latency of processing a metadata request"); | |
497 | plb.add_time_avg(l_c_wrlat, "wrlat", "Latency of a file data write operation"); | |
498 | plb.add_time_avg(l_c_read, "rdlat", "Latency of a file data read operation"); | |
499 | plb.add_time_avg(l_c_fsync, "fsync", "Latency of a file sync operation"); | |
500 | logger.reset(plb.create_perf_counters()); | |
501 | cct->get_perfcounters_collection()->add(logger.get()); | |
502 | } | |
7c673cae | 503 | |
11fdf7f2 | 504 | cct->_conf.add_observer(this); |
7c673cae FG |
505 | |
506 | AdminSocket* admin_socket = cct->get_admin_socket(); | |
507 | int ret = admin_socket->register_command("mds_requests", | |
7c673cae FG |
508 | &m_command_hook, |
509 | "show in-progress mds requests"); | |
510 | if (ret < 0) { | |
511 | lderr(cct) << "error registering admin socket command: " | |
512 | << cpp_strerror(-ret) << dendl; | |
513 | } | |
514 | ret = admin_socket->register_command("mds_sessions", | |
7c673cae FG |
515 | &m_command_hook, |
516 | "show mds session state"); | |
517 | if (ret < 0) { | |
518 | lderr(cct) << "error registering admin socket command: " | |
519 | << cpp_strerror(-ret) << dendl; | |
520 | } | |
521 | ret = admin_socket->register_command("dump_cache", | |
7c673cae FG |
522 | &m_command_hook, |
523 | "show in-memory metadata cache contents"); | |
524 | if (ret < 0) { | |
525 | lderr(cct) << "error registering admin socket command: " | |
526 | << cpp_strerror(-ret) << dendl; | |
527 | } | |
528 | ret = admin_socket->register_command("kick_stale_sessions", | |
7c673cae FG |
529 | &m_command_hook, |
530 | "kick sessions that were remote reset"); | |
531 | if (ret < 0) { | |
532 | lderr(cct) << "error registering admin socket command: " | |
533 | << cpp_strerror(-ret) << dendl; | |
534 | } | |
535 | ret = admin_socket->register_command("status", | |
7c673cae FG |
536 | &m_command_hook, |
537 | "show overall client status"); | |
538 | if (ret < 0) { | |
539 | lderr(cct) << "error registering admin socket command: " | |
540 | << cpp_strerror(-ret) << dendl; | |
541 | } | |
542 | ||
9f95a23c | 543 | std::lock_guard l{client_lock}; |
7c673cae | 544 | initialized = true; |
7c673cae FG |
545 | } |
546 | ||
547 | void Client::shutdown() | |
548 | { | |
11fdf7f2 | 549 | ldout(cct, 1) << __func__ << dendl; |
7c673cae FG |
550 | |
551 | // If we were not mounted, but were being used for sending | |
552 | // MDS commands, we may have sessions that need closing. | |
9f95a23c TL |
553 | { |
554 | std::lock_guard l{client_lock}; | |
555 | _close_sessions(); | |
556 | } | |
11fdf7f2 | 557 | cct->_conf.remove_observer(this); |
7c673cae | 558 | |
11fdf7f2 | 559 | cct->get_admin_socket()->unregister_commands(&m_command_hook); |
7c673cae FG |
560 | |
561 | if (ino_invalidate_cb) { | |
562 | ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl; | |
563 | async_ino_invalidator.wait_for_empty(); | |
564 | async_ino_invalidator.stop(); | |
565 | } | |
566 | ||
567 | if (dentry_invalidate_cb) { | |
568 | ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl; | |
569 | async_dentry_invalidator.wait_for_empty(); | |
570 | async_dentry_invalidator.stop(); | |
571 | } | |
572 | ||
573 | if (switch_interrupt_cb) { | |
574 | ldout(cct, 10) << "shutdown stopping interrupt finisher" << dendl; | |
575 | interrupt_finisher.wait_for_empty(); | |
576 | interrupt_finisher.stop(); | |
577 | } | |
578 | ||
579 | if (remount_cb) { | |
580 | ldout(cct, 10) << "shutdown stopping remount finisher" << dendl; | |
581 | remount_finisher.wait_for_empty(); | |
582 | remount_finisher.stop(); | |
583 | } | |
584 | ||
585 | objectcacher->stop(); // outside of client_lock! this does a join. | |
9f95a23c TL |
586 | { |
587 | std::lock_guard l{client_lock}; | |
588 | ceph_assert(initialized); | |
589 | initialized = false; | |
590 | timer.shutdown(); | |
591 | } | |
7c673cae FG |
592 | objecter_finisher.wait_for_empty(); |
593 | objecter_finisher.stop(); | |
594 | ||
595 | if (logger) { | |
596 | cct->get_perfcounters_collection()->remove(logger.get()); | |
597 | logger.reset(); | |
598 | } | |
599 | } | |
600 | ||
601 | ||
602 | // =================== | |
603 | // metadata cache stuff | |
604 | ||
605 | void Client::trim_cache(bool trim_kernel_dcache) | |
606 | { | |
181888fb FG |
607 | uint64_t max = cct->_conf->client_cache_size; |
608 | ldout(cct, 20) << "trim_cache size " << lru.lru_get_size() << " max " << max << dendl; | |
7c673cae FG |
609 | unsigned last = 0; |
610 | while (lru.lru_get_size() != last) { | |
611 | last = lru.lru_get_size(); | |
612 | ||
181888fb | 613 | if (!unmounting && lru.lru_get_size() <= max) break; |
7c673cae FG |
614 | |
615 | // trim! | |
31f18b77 | 616 | Dentry *dn = static_cast<Dentry*>(lru.lru_get_next_expire()); |
7c673cae FG |
617 | if (!dn) |
618 | break; // done | |
619 | ||
620 | trim_dentry(dn); | |
621 | } | |
622 | ||
181888fb | 623 | if (trim_kernel_dcache && lru.lru_get_size() > max) |
7c673cae FG |
624 | _invalidate_kernel_dcache(); |
625 | ||
626 | // hose root? | |
627 | if (lru.lru_get_size() == 0 && root && root->get_num_ref() == 0 && inode_map.size() == 1 + root_parents.size()) { | |
628 | ldout(cct, 15) << "trim_cache trimmed root " << root << dendl; | |
629 | delete root; | |
630 | root = 0; | |
631 | root_ancestor = 0; | |
632 | while (!root_parents.empty()) | |
633 | root_parents.erase(root_parents.begin()); | |
634 | inode_map.clear(); | |
635 | _reset_faked_inos(); | |
636 | } | |
637 | } | |
638 | ||
639 | void Client::trim_cache_for_reconnect(MetaSession *s) | |
640 | { | |
641 | mds_rank_t mds = s->mds_num; | |
11fdf7f2 | 642 | ldout(cct, 20) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
643 | |
644 | int trimmed = 0; | |
645 | list<Dentry*> skipped; | |
646 | while (lru.lru_get_size() > 0) { | |
647 | Dentry *dn = static_cast<Dentry*>(lru.lru_expire()); | |
648 | if (!dn) | |
649 | break; | |
650 | ||
651 | if ((dn->inode && dn->inode->caps.count(mds)) || | |
652 | dn->dir->parent_inode->caps.count(mds)) { | |
653 | trim_dentry(dn); | |
654 | trimmed++; | |
655 | } else | |
656 | skipped.push_back(dn); | |
657 | } | |
658 | ||
659 | for(list<Dentry*>::iterator p = skipped.begin(); p != skipped.end(); ++p) | |
660 | lru.lru_insert_mid(*p); | |
661 | ||
11fdf7f2 | 662 | ldout(cct, 20) << __func__ << " mds." << mds |
7c673cae FG |
663 | << " trimmed " << trimmed << " dentries" << dendl; |
664 | ||
665 | if (s->caps.size() > 0) | |
666 | _invalidate_kernel_dcache(); | |
667 | } | |
668 | ||
669 | void Client::trim_dentry(Dentry *dn) | |
670 | { | |
671 | ldout(cct, 15) << "trim_dentry unlinking dn " << dn->name | |
11fdf7f2 TL |
672 | << " in dir " |
673 | << std::hex << dn->dir->parent_inode->ino << std::dec | |
7c673cae FG |
674 | << dendl; |
675 | if (dn->inode) { | |
676 | Inode *diri = dn->dir->parent_inode; | |
677 | diri->dir_release_count++; | |
678 | clear_dir_complete_and_ordered(diri, true); | |
679 | } | |
680 | unlink(dn, false, false); // drop dir, drop dentry | |
681 | } | |
682 | ||
683 | ||
1adf2230 AA |
684 | void Client::update_inode_file_size(Inode *in, int issued, uint64_t size, |
685 | uint64_t truncate_seq, uint64_t truncate_size) | |
7c673cae | 686 | { |
7c673cae FG |
687 | uint64_t prior_size = in->size; |
688 | ||
7c673cae FG |
689 | if (truncate_seq > in->truncate_seq || |
690 | (truncate_seq == in->truncate_seq && size > in->size)) { | |
691 | ldout(cct, 10) << "size " << in->size << " -> " << size << dendl; | |
692 | in->size = size; | |
693 | in->reported_size = size; | |
694 | if (truncate_seq != in->truncate_seq) { | |
695 | ldout(cct, 10) << "truncate_seq " << in->truncate_seq << " -> " | |
696 | << truncate_seq << dendl; | |
697 | in->truncate_seq = truncate_seq; | |
698 | in->oset.truncate_seq = truncate_seq; | |
699 | ||
700 | // truncate cached file data | |
701 | if (prior_size > size) { | |
702 | _invalidate_inode_cache(in, truncate_size, prior_size - truncate_size); | |
703 | } | |
704 | } | |
705 | ||
706 | // truncate inline data | |
707 | if (in->inline_version < CEPH_INLINE_NONE) { | |
708 | uint32_t len = in->inline_data.length(); | |
709 | if (size < len) | |
710 | in->inline_data.splice(size, len - size); | |
711 | } | |
712 | } | |
713 | if (truncate_seq >= in->truncate_seq && | |
714 | in->truncate_size != truncate_size) { | |
715 | if (in->is_file()) { | |
716 | ldout(cct, 10) << "truncate_size " << in->truncate_size << " -> " | |
717 | << truncate_size << dendl; | |
718 | in->truncate_size = truncate_size; | |
719 | in->oset.truncate_size = truncate_size; | |
720 | } else { | |
721 | ldout(cct, 0) << "Hmmm, truncate_seq && truncate_size changed on non-file inode!" << dendl; | |
722 | } | |
723 | } | |
1adf2230 AA |
724 | } |
725 | ||
726 | void Client::update_inode_file_time(Inode *in, int issued, uint64_t time_warp_seq, | |
727 | utime_t ctime, utime_t mtime, utime_t atime) | |
728 | { | |
729 | ldout(cct, 10) << __func__ << " " << *in << " " << ccap_string(issued) | |
730 | << " ctime " << ctime << " mtime " << mtime << dendl; | |
731 | ||
732 | if (time_warp_seq > in->time_warp_seq) | |
733 | ldout(cct, 10) << " mds time_warp_seq " << time_warp_seq | |
734 | << " is higher than local time_warp_seq " | |
735 | << in->time_warp_seq << dendl; | |
736 | ||
737 | int warn = false; | |
7c673cae FG |
738 | // be careful with size, mtime, atime |
739 | if (issued & (CEPH_CAP_FILE_EXCL| | |
740 | CEPH_CAP_FILE_WR| | |
741 | CEPH_CAP_FILE_BUFFER| | |
742 | CEPH_CAP_AUTH_EXCL| | |
743 | CEPH_CAP_XATTR_EXCL)) { | |
744 | ldout(cct, 30) << "Yay have enough caps to look at our times" << dendl; | |
745 | if (ctime > in->ctime) | |
746 | in->ctime = ctime; | |
747 | if (time_warp_seq > in->time_warp_seq) { | |
7c673cae FG |
748 | //the mds updated times, so take those! |
749 | in->mtime = mtime; | |
750 | in->atime = atime; | |
751 | in->time_warp_seq = time_warp_seq; | |
752 | } else if (time_warp_seq == in->time_warp_seq) { | |
753 | //take max times | |
754 | if (mtime > in->mtime) | |
755 | in->mtime = mtime; | |
756 | if (atime > in->atime) | |
757 | in->atime = atime; | |
758 | } else if (issued & CEPH_CAP_FILE_EXCL) { | |
759 | //ignore mds values as we have a higher seq | |
760 | } else warn = true; | |
761 | } else { | |
762 | ldout(cct, 30) << "Don't have enough caps, just taking mds' time values" << dendl; | |
763 | if (time_warp_seq >= in->time_warp_seq) { | |
764 | in->ctime = ctime; | |
765 | in->mtime = mtime; | |
766 | in->atime = atime; | |
767 | in->time_warp_seq = time_warp_seq; | |
768 | } else warn = true; | |
769 | } | |
770 | if (warn) { | |
771 | ldout(cct, 0) << "WARNING: " << *in << " mds time_warp_seq " | |
772 | << time_warp_seq << " is lower than local time_warp_seq " | |
773 | << in->time_warp_seq | |
774 | << dendl; | |
775 | } | |
776 | } | |
777 | ||
778 | void Client::_fragmap_remove_non_leaves(Inode *in) | |
779 | { | |
780 | for (map<frag_t,int>::iterator p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
781 | if (!in->dirfragtree.is_leaf(p->first)) | |
782 | in->fragmap.erase(p++); | |
783 | else | |
784 | ++p; | |
785 | } | |
786 | ||
787 | void Client::_fragmap_remove_stopped_mds(Inode *in, mds_rank_t mds) | |
788 | { | |
789 | for (auto p = in->fragmap.begin(); p != in->fragmap.end(); ) | |
790 | if (p->second == mds) | |
791 | in->fragmap.erase(p++); | |
792 | else | |
793 | ++p; | |
794 | } | |
795 | ||
796 | Inode * Client::add_update_inode(InodeStat *st, utime_t from, | |
797 | MetaSession *session, | |
798 | const UserPerm& request_perms) | |
799 | { | |
800 | Inode *in; | |
801 | bool was_new = false; | |
802 | if (inode_map.count(st->vino)) { | |
803 | in = inode_map[st->vino]; | |
11fdf7f2 | 804 | ldout(cct, 12) << __func__ << " had " << *in << " caps " << ccap_string(st->cap.caps) << dendl; |
7c673cae FG |
805 | } else { |
806 | in = new Inode(this, st->vino, &st->layout); | |
807 | inode_map[st->vino] = in; | |
808 | ||
809 | if (use_faked_inos()) | |
810 | _assign_faked_ino(in); | |
811 | ||
812 | if (!root) { | |
813 | root = in; | |
11fdf7f2 TL |
814 | if (use_faked_inos()) |
815 | _assign_faked_root(root); | |
7c673cae FG |
816 | root_ancestor = in; |
817 | cwd = root; | |
818 | } else if (!mounted) { | |
819 | root_parents[root_ancestor] = in; | |
820 | root_ancestor = in; | |
821 | } | |
822 | ||
823 | // immutable bits | |
824 | in->ino = st->vino.ino; | |
825 | in->snapid = st->vino.snapid; | |
826 | in->mode = st->mode & S_IFMT; | |
827 | was_new = true; | |
828 | } | |
829 | ||
830 | in->rdev = st->rdev; | |
831 | if (in->is_symlink()) | |
832 | in->symlink = st->symlink; | |
833 | ||
7c673cae | 834 | // only update inode if mds info is strictly newer, or it is the same and projected (odd). |
1adf2230 AA |
835 | bool new_version = false; |
836 | if (in->version == 0 || | |
837 | ((st->cap.flags & CEPH_CAP_FLAG_AUTH) && | |
838 | (in->version & ~1) < st->version)) | |
839 | new_version = true; | |
7c673cae | 840 | |
1adf2230 AA |
841 | int issued; |
842 | in->caps_issued(&issued); | |
843 | issued |= in->caps_dirty(); | |
844 | int new_issued = ~issued & (int)st->cap.caps; | |
7c673cae | 845 | |
1adf2230 AA |
846 | if ((new_version || (new_issued & CEPH_CAP_AUTH_SHARED)) && |
847 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
848 | in->mode = st->mode; | |
849 | in->uid = st->uid; | |
850 | in->gid = st->gid; | |
851 | in->btime = st->btime; | |
81eedcae | 852 | in->snap_btime = st->snap_btime; |
1adf2230 | 853 | } |
7c673cae | 854 | |
1adf2230 AA |
855 | if ((new_version || (new_issued & CEPH_CAP_LINK_SHARED)) && |
856 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
857 | in->nlink = st->nlink; | |
858 | } | |
7c673cae | 859 | |
1adf2230 AA |
860 | if (new_version || (new_issued & CEPH_CAP_ANY_RD)) { |
861 | update_inode_file_time(in, issued, st->time_warp_seq, | |
862 | st->ctime, st->mtime, st->atime); | |
863 | } | |
7c673cae | 864 | |
1adf2230 AA |
865 | if (new_version || |
866 | (new_issued & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR))) { | |
7c673cae | 867 | in->layout = st->layout; |
1adf2230 AA |
868 | update_inode_file_size(in, issued, st->size, st->truncate_seq, st->truncate_size); |
869 | } | |
7c673cae | 870 | |
1adf2230 AA |
871 | if (in->is_dir()) { |
872 | if (new_version || (new_issued & CEPH_CAP_FILE_SHARED)) { | |
873 | in->dirstat = st->dirstat; | |
874 | } | |
875 | // dir_layout/rstat/quota are not tracked by capability, update them only if | |
876 | // the inode stat is from auth mds | |
877 | if (new_version || (st->cap.flags & CEPH_CAP_FLAG_AUTH)) { | |
7c673cae FG |
878 | in->dir_layout = st->dir_layout; |
879 | ldout(cct, 20) << " dir hash is " << (int)in->dir_layout.dl_dir_hash << dendl; | |
1adf2230 AA |
880 | in->rstat = st->rstat; |
881 | in->quota = st->quota; | |
11fdf7f2 | 882 | in->dir_pin = st->dir_pin; |
1adf2230 AA |
883 | } |
884 | // move me if/when version reflects fragtree changes. | |
885 | if (in->dirfragtree != st->dirfragtree) { | |
886 | in->dirfragtree = st->dirfragtree; | |
887 | _fragmap_remove_non_leaves(in); | |
7c673cae | 888 | } |
7c673cae FG |
889 | } |
890 | ||
891 | if ((in->xattr_version == 0 || !(issued & CEPH_CAP_XATTR_EXCL)) && | |
892 | st->xattrbl.length() && | |
893 | st->xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
894 | auto p = st->xattrbl.cbegin(); |
895 | decode(in->xattrs, p); | |
7c673cae FG |
896 | in->xattr_version = st->xattr_version; |
897 | } | |
898 | ||
1adf2230 AA |
899 | if (st->inline_version > in->inline_version) { |
900 | in->inline_data = st->inline_data; | |
901 | in->inline_version = st->inline_version; | |
7c673cae FG |
902 | } |
903 | ||
1adf2230 AA |
904 | /* always take a newer change attr */ |
905 | if (st->change_attr > in->change_attr) | |
906 | in->change_attr = st->change_attr; | |
907 | ||
908 | if (st->version > in->version) | |
909 | in->version = st->version; | |
910 | ||
911 | if (was_new) | |
912 | ldout(cct, 12) << __func__ << " adding " << *in << " caps " << ccap_string(st->cap.caps) << dendl; | |
913 | ||
914 | if (!st->cap.caps) | |
915 | return in; // as with readdir returning indoes in different snaprealms (no caps!) | |
916 | ||
7c673cae | 917 | if (in->snapid == CEPH_NOSNAP) { |
a8e16298 TL |
918 | add_update_cap(in, session, st->cap.cap_id, st->cap.caps, st->cap.wanted, |
919 | st->cap.seq, st->cap.mseq, inodeno_t(st->cap.realm), | |
920 | st->cap.flags, request_perms); | |
28e407b8 | 921 | if (in->auth_cap && in->auth_cap->session == session) { |
7c673cae | 922 | in->max_size = st->max_size; |
28e407b8 AA |
923 | in->rstat = st->rstat; |
924 | } | |
7c673cae | 925 | |
1adf2230 AA |
926 | // setting I_COMPLETE needs to happen after adding the cap |
927 | if (in->is_dir() && | |
928 | (st->cap.caps & CEPH_CAP_FILE_SHARED) && | |
929 | (issued & CEPH_CAP_FILE_EXCL) == 0 && | |
930 | in->dirstat.nfiles == 0 && | |
931 | in->dirstat.nsubdirs == 0) { | |
932 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on empty dir " << *in << dendl; | |
933 | in->flags |= I_COMPLETE | I_DIR_ORDERED; | |
934 | if (in->dir) { | |
935 | ldout(cct, 10) << " dir is open on empty dir " << in->ino << " with " | |
936 | << in->dir->dentries.size() << " entries, marking all dentries null" << dendl; | |
937 | in->dir->readdir_cache.clear(); | |
938 | for (const auto& p : in->dir->dentries) { | |
939 | unlink(p.second, true, true); // keep dir, keep dentry | |
940 | } | |
941 | if (in->dir->dentries.empty()) | |
942 | close_dir(in->dir); | |
7c673cae | 943 | } |
7c673cae | 944 | } |
1adf2230 AA |
945 | } else { |
946 | in->snap_caps |= st->cap.caps; | |
7c673cae FG |
947 | } |
948 | ||
949 | return in; | |
950 | } | |
951 | ||
952 | ||
953 | /* | |
954 | * insert_dentry_inode - insert + link a single dentry + inode into the metadata cache. | |
955 | */ | |
956 | Dentry *Client::insert_dentry_inode(Dir *dir, const string& dname, LeaseStat *dlease, | |
957 | Inode *in, utime_t from, MetaSession *session, | |
958 | Dentry *old_dentry) | |
959 | { | |
960 | Dentry *dn = NULL; | |
961 | if (dir->dentries.count(dname)) | |
962 | dn = dir->dentries[dname]; | |
963 | ||
11fdf7f2 | 964 | ldout(cct, 12) << __func__ << " '" << dname << "' vino " << in->vino() |
7c673cae FG |
965 | << " in dir " << dir->parent_inode->vino() << " dn " << dn |
966 | << dendl; | |
967 | ||
968 | if (dn && dn->inode) { | |
969 | if (dn->inode->vino() == in->vino()) { | |
970 | touch_dn(dn); | |
971 | ldout(cct, 12) << " had dentry " << dname | |
972 | << " with correct vino " << dn->inode->vino() | |
973 | << dendl; | |
974 | } else { | |
975 | ldout(cct, 12) << " had dentry " << dname | |
976 | << " with WRONG vino " << dn->inode->vino() | |
977 | << dendl; | |
978 | unlink(dn, true, true); // keep dir, keep dentry | |
979 | } | |
980 | } | |
981 | ||
982 | if (!dn || !dn->inode) { | |
983 | InodeRef tmp_ref(in); | |
984 | if (old_dentry) { | |
985 | if (old_dentry->dir != dir) { | |
986 | Inode *old_diri = old_dentry->dir->parent_inode; | |
987 | old_diri->dir_ordered_count++; | |
988 | clear_dir_complete_and_ordered(old_diri, false); | |
989 | } | |
990 | unlink(old_dentry, dir == old_dentry->dir, false); // drop dentry, keep dir open if its the same dir | |
991 | } | |
992 | Inode *diri = dir->parent_inode; | |
993 | diri->dir_ordered_count++; | |
994 | clear_dir_complete_and_ordered(diri, false); | |
995 | dn = link(dir, dname, in, dn); | |
996 | } | |
997 | ||
998 | update_dentry_lease(dn, dlease, from, session); | |
999 | return dn; | |
1000 | } | |
1001 | ||
1002 | void Client::update_dentry_lease(Dentry *dn, LeaseStat *dlease, utime_t from, MetaSession *session) | |
1003 | { | |
1004 | utime_t dttl = from; | |
1005 | dttl += (float)dlease->duration_ms / 1000.0; | |
1006 | ||
11fdf7f2 | 1007 | ceph_assert(dn); |
7c673cae | 1008 | |
9f95a23c | 1009 | if (dlease->mask & CEPH_LEASE_VALID) { |
7c673cae FG |
1010 | if (dttl > dn->lease_ttl) { |
1011 | ldout(cct, 10) << "got dentry lease on " << dn->name | |
1012 | << " dur " << dlease->duration_ms << "ms ttl " << dttl << dendl; | |
1013 | dn->lease_ttl = dttl; | |
1014 | dn->lease_mds = session->mds_num; | |
1015 | dn->lease_seq = dlease->seq; | |
1016 | dn->lease_gen = session->cap_gen; | |
1017 | } | |
1018 | } | |
1019 | dn->cap_shared_gen = dn->dir->parent_inode->shared_gen; | |
1020 | } | |
1021 | ||
1022 | ||
1023 | /* | |
1024 | * update MDS location cache for a single inode | |
1025 | */ | |
1026 | void Client::update_dir_dist(Inode *in, DirStat *dst) | |
1027 | { | |
1028 | // auth | |
1029 | ldout(cct, 20) << "got dirfrag map for " << in->ino << " frag " << dst->frag << " to mds " << dst->auth << dendl; | |
1030 | if (dst->auth >= 0) { | |
1031 | in->fragmap[dst->frag] = dst->auth; | |
1032 | } else { | |
1033 | in->fragmap.erase(dst->frag); | |
1034 | } | |
1035 | if (!in->dirfragtree.is_leaf(dst->frag)) { | |
1036 | in->dirfragtree.force_to_leaf(cct, dst->frag); | |
1037 | _fragmap_remove_non_leaves(in); | |
1038 | } | |
1039 | ||
1040 | // replicated | |
1041 | in->dir_replicated = !dst->dist.empty(); // FIXME that's just one frag! | |
7c673cae FG |
1042 | } |
1043 | ||
1044 | void Client::clear_dir_complete_and_ordered(Inode *diri, bool complete) | |
1045 | { | |
1046 | if (diri->flags & I_COMPLETE) { | |
1047 | if (complete) { | |
1048 | ldout(cct, 10) << " clearing (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
1049 | diri->flags &= ~(I_COMPLETE | I_DIR_ORDERED); | |
1050 | } else { | |
1051 | if (diri->flags & I_DIR_ORDERED) { | |
1052 | ldout(cct, 10) << " clearing I_DIR_ORDERED on " << *diri << dendl; | |
1053 | diri->flags &= ~I_DIR_ORDERED; | |
1054 | } | |
1055 | } | |
1056 | if (diri->dir) | |
1057 | diri->dir->readdir_cache.clear(); | |
1058 | } | |
1059 | } | |
1060 | ||
1061 | /* | |
1062 | * insert results from readdir or lssnap into the metadata cache. | |
1063 | */ | |
1064 | void Client::insert_readdir_results(MetaRequest *request, MetaSession *session, Inode *diri) { | |
1065 | ||
11fdf7f2 | 1066 | auto& reply = request->reply; |
7c673cae | 1067 | ConnectionRef con = request->reply->get_connection(); |
11fdf7f2 TL |
1068 | uint64_t features; |
1069 | if(session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1070 | features = (uint64_t)-1; | |
1071 | } | |
1072 | else { | |
1073 | features = con->get_features(); | |
1074 | } | |
7c673cae FG |
1075 | |
1076 | dir_result_t *dirp = request->dirp; | |
11fdf7f2 | 1077 | ceph_assert(dirp); |
7c673cae FG |
1078 | |
1079 | // the extra buffer list is only set for readdir and lssnap replies | |
11fdf7f2 | 1080 | auto p = reply->get_extra_bl().cbegin(); |
7c673cae FG |
1081 | if (!p.end()) { |
1082 | // snapdir? | |
1083 | if (request->head.op == CEPH_MDS_OP_LSSNAP) { | |
11fdf7f2 | 1084 | ceph_assert(diri); |
7c673cae FG |
1085 | diri = open_snapdir(diri); |
1086 | } | |
1087 | ||
1088 | // only open dir if we're actually adding stuff to it! | |
1089 | Dir *dir = diri->open_dir(); | |
11fdf7f2 | 1090 | ceph_assert(dir); |
7c673cae FG |
1091 | |
1092 | // dirstat | |
11fdf7f2 | 1093 | DirStat dst(p, features); |
7c673cae FG |
1094 | __u32 numdn; |
1095 | __u16 flags; | |
11fdf7f2 TL |
1096 | decode(numdn, p); |
1097 | decode(flags, p); | |
7c673cae FG |
1098 | |
1099 | bool end = ((unsigned)flags & CEPH_READDIR_FRAG_END); | |
1100 | bool hash_order = ((unsigned)flags & CEPH_READDIR_HASH_ORDER); | |
1101 | ||
1102 | frag_t fg = (unsigned)request->head.args.readdir.frag; | |
1103 | unsigned readdir_offset = dirp->next_offset; | |
1104 | string readdir_start = dirp->last_name; | |
11fdf7f2 | 1105 | ceph_assert(!readdir_start.empty() || readdir_offset == 2); |
7c673cae FG |
1106 | |
1107 | unsigned last_hash = 0; | |
1108 | if (hash_order) { | |
1109 | if (!readdir_start.empty()) { | |
1110 | last_hash = ceph_frag_value(diri->hash_dentry_name(readdir_start)); | |
1111 | } else if (flags & CEPH_READDIR_OFFSET_HASH) { | |
1112 | /* mds understands offset_hash */ | |
1113 | last_hash = (unsigned)request->head.args.readdir.offset_hash; | |
1114 | } | |
1115 | } | |
1116 | ||
1117 | if (fg != dst.frag) { | |
1118 | ldout(cct, 10) << "insert_trace got new frag " << fg << " -> " << dst.frag << dendl; | |
1119 | fg = dst.frag; | |
1120 | if (!hash_order) { | |
1121 | readdir_offset = 2; | |
1122 | readdir_start.clear(); | |
1123 | dirp->offset = dir_result_t::make_fpos(fg, readdir_offset, false); | |
1124 | } | |
1125 | } | |
1126 | ||
1127 | ldout(cct, 10) << __func__ << " " << numdn << " readdir items, end=" << end | |
1128 | << ", hash_order=" << hash_order | |
1129 | << ", readdir_start " << readdir_start | |
1130 | << ", last_hash " << last_hash | |
1131 | << ", next_offset " << readdir_offset << dendl; | |
1132 | ||
1133 | if (diri->snapid != CEPH_SNAPDIR && | |
1134 | fg.is_leftmost() && readdir_offset == 2 && | |
1135 | !(hash_order && last_hash)) { | |
1136 | dirp->release_count = diri->dir_release_count; | |
1137 | dirp->ordered_count = diri->dir_ordered_count; | |
1138 | dirp->start_shared_gen = diri->shared_gen; | |
1139 | dirp->cache_index = 0; | |
1140 | } | |
1141 | ||
1142 | dirp->buffer_frag = fg; | |
1143 | ||
1144 | _readdir_drop_dirp_buffer(dirp); | |
1145 | dirp->buffer.reserve(numdn); | |
1146 | ||
1147 | string dname; | |
1148 | LeaseStat dlease; | |
1149 | for (unsigned i=0; i<numdn; i++) { | |
11fdf7f2 TL |
1150 | decode(dname, p); |
1151 | dlease.decode(p, features); | |
7c673cae FG |
1152 | InodeStat ist(p, features); |
1153 | ||
1154 | ldout(cct, 15) << "" << i << ": '" << dname << "'" << dendl; | |
1155 | ||
1156 | Inode *in = add_update_inode(&ist, request->sent_stamp, session, | |
1157 | request->perms); | |
1158 | Dentry *dn; | |
1159 | if (diri->dir->dentries.count(dname)) { | |
1160 | Dentry *olddn = diri->dir->dentries[dname]; | |
1161 | if (olddn->inode != in) { | |
1162 | // replace incorrect dentry | |
1163 | unlink(olddn, true, true); // keep dir, dentry | |
1164 | dn = link(dir, dname, in, olddn); | |
11fdf7f2 | 1165 | ceph_assert(dn == olddn); |
7c673cae FG |
1166 | } else { |
1167 | // keep existing dn | |
1168 | dn = olddn; | |
1169 | touch_dn(dn); | |
1170 | } | |
1171 | } else { | |
1172 | // new dn | |
1173 | dn = link(dir, dname, in, NULL); | |
1174 | } | |
1175 | ||
1176 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1177 | if (hash_order) { | |
1178 | unsigned hash = ceph_frag_value(diri->hash_dentry_name(dname)); | |
1179 | if (hash != last_hash) | |
1180 | readdir_offset = 2; | |
1181 | last_hash = hash; | |
1182 | dn->offset = dir_result_t::make_fpos(hash, readdir_offset++, true); | |
1183 | } else { | |
1184 | dn->offset = dir_result_t::make_fpos(fg, readdir_offset++, false); | |
1185 | } | |
1186 | // add to readdir cache | |
1187 | if (dirp->release_count == diri->dir_release_count && | |
1188 | dirp->ordered_count == diri->dir_ordered_count && | |
1189 | dirp->start_shared_gen == diri->shared_gen) { | |
1190 | if (dirp->cache_index == dir->readdir_cache.size()) { | |
1191 | if (i == 0) { | |
11fdf7f2 | 1192 | ceph_assert(!dirp->inode->is_complete_and_ordered()); |
7c673cae FG |
1193 | dir->readdir_cache.reserve(dirp->cache_index + numdn); |
1194 | } | |
1195 | dir->readdir_cache.push_back(dn); | |
1196 | } else if (dirp->cache_index < dir->readdir_cache.size()) { | |
1197 | if (dirp->inode->is_complete_and_ordered()) | |
11fdf7f2 | 1198 | ceph_assert(dir->readdir_cache[dirp->cache_index] == dn); |
7c673cae FG |
1199 | else |
1200 | dir->readdir_cache[dirp->cache_index] = dn; | |
1201 | } else { | |
11fdf7f2 | 1202 | ceph_abort_msg("unexpected readdir buffer idx"); |
7c673cae FG |
1203 | } |
1204 | dirp->cache_index++; | |
1205 | } | |
1206 | // add to cached result list | |
1207 | dirp->buffer.push_back(dir_result_t::dentry(dn->offset, dname, in)); | |
1208 | ldout(cct, 15) << __func__ << " " << hex << dn->offset << dec << ": '" << dname << "' -> " << in->ino << dendl; | |
1209 | } | |
1210 | ||
1211 | if (numdn > 0) | |
1212 | dirp->last_name = dname; | |
1213 | if (end) | |
1214 | dirp->next_offset = 2; | |
1215 | else | |
1216 | dirp->next_offset = readdir_offset; | |
1217 | ||
1218 | if (dir->is_empty()) | |
1219 | close_dir(dir); | |
1220 | } | |
1221 | } | |
1222 | ||
1223 | /** insert_trace | |
1224 | * | |
1225 | * insert a trace from a MDS reply into the cache. | |
1226 | */ | |
1227 | Inode* Client::insert_trace(MetaRequest *request, MetaSession *session) | |
1228 | { | |
11fdf7f2 | 1229 | auto& reply = request->reply; |
7c673cae FG |
1230 | int op = request->get_op(); |
1231 | ||
1232 | ldout(cct, 10) << "insert_trace from " << request->sent_stamp << " mds." << session->mds_num | |
1233 | << " is_target=" << (int)reply->head.is_target | |
1234 | << " is_dentry=" << (int)reply->head.is_dentry | |
1235 | << dendl; | |
1236 | ||
11fdf7f2 | 1237 | auto p = reply->get_trace_bl().cbegin(); |
7c673cae FG |
1238 | if (request->got_unsafe) { |
1239 | ldout(cct, 10) << "insert_trace -- already got unsafe; ignoring" << dendl; | |
11fdf7f2 | 1240 | ceph_assert(p.end()); |
7c673cae FG |
1241 | return NULL; |
1242 | } | |
1243 | ||
1244 | if (p.end()) { | |
1245 | ldout(cct, 10) << "insert_trace -- no trace" << dendl; | |
1246 | ||
1247 | Dentry *d = request->dentry(); | |
1248 | if (d) { | |
1249 | Inode *diri = d->dir->parent_inode; | |
1250 | diri->dir_release_count++; | |
1251 | clear_dir_complete_and_ordered(diri, true); | |
1252 | } | |
1253 | ||
1254 | if (d && reply->get_result() == 0) { | |
1255 | if (op == CEPH_MDS_OP_RENAME) { | |
1256 | // rename | |
1257 | Dentry *od = request->old_dentry(); | |
1258 | ldout(cct, 10) << " unlinking rename src dn " << od << " for traceless reply" << dendl; | |
11fdf7f2 | 1259 | ceph_assert(od); |
7c673cae FG |
1260 | unlink(od, true, true); // keep dir, dentry |
1261 | } else if (op == CEPH_MDS_OP_RMDIR || | |
1262 | op == CEPH_MDS_OP_UNLINK) { | |
1263 | // unlink, rmdir | |
1264 | ldout(cct, 10) << " unlinking unlink/rmdir dn " << d << " for traceless reply" << dendl; | |
1265 | unlink(d, true, true); // keep dir, dentry | |
1266 | } | |
1267 | } | |
1268 | return NULL; | |
1269 | } | |
1270 | ||
1271 | ConnectionRef con = request->reply->get_connection(); | |
11fdf7f2 TL |
1272 | uint64_t features; |
1273 | if (session->mds_features.test(CEPHFS_FEATURE_REPLY_ENCODING)) { | |
1274 | features = (uint64_t)-1; | |
1275 | } | |
1276 | else { | |
1277 | features = con->get_features(); | |
1278 | } | |
7c673cae FG |
1279 | ldout(cct, 10) << " features 0x" << hex << features << dec << dendl; |
1280 | ||
1281 | // snap trace | |
1282 | SnapRealm *realm = NULL; | |
1283 | if (reply->snapbl.length()) | |
1284 | update_snap_trace(reply->snapbl, &realm); | |
1285 | ||
1286 | ldout(cct, 10) << " hrm " | |
1287 | << " is_target=" << (int)reply->head.is_target | |
1288 | << " is_dentry=" << (int)reply->head.is_dentry | |
1289 | << dendl; | |
1290 | ||
1291 | InodeStat dirst; | |
1292 | DirStat dst; | |
1293 | string dname; | |
1294 | LeaseStat dlease; | |
1295 | InodeStat ist; | |
1296 | ||
1297 | if (reply->head.is_dentry) { | |
1298 | dirst.decode(p, features); | |
11fdf7f2 TL |
1299 | dst.decode(p, features); |
1300 | decode(dname, p); | |
1301 | dlease.decode(p, features); | |
7c673cae FG |
1302 | } |
1303 | ||
1304 | Inode *in = 0; | |
1305 | if (reply->head.is_target) { | |
1306 | ist.decode(p, features); | |
1307 | if (cct->_conf->client_debug_getattr_caps) { | |
1308 | unsigned wanted = 0; | |
1309 | if (op == CEPH_MDS_OP_GETATTR || op == CEPH_MDS_OP_LOOKUP) | |
1310 | wanted = request->head.args.getattr.mask; | |
1311 | else if (op == CEPH_MDS_OP_OPEN || op == CEPH_MDS_OP_CREATE) | |
1312 | wanted = request->head.args.open.mask; | |
1313 | ||
1314 | if ((wanted & CEPH_CAP_XATTR_SHARED) && | |
1315 | !(ist.xattr_version > 0 && ist.xattrbl.length() > 0)) | |
11fdf7f2 | 1316 | ceph_abort_msg("MDS reply does not contain xattrs"); |
7c673cae FG |
1317 | } |
1318 | ||
1319 | in = add_update_inode(&ist, request->sent_stamp, session, | |
1320 | request->perms); | |
1321 | } | |
1322 | ||
1323 | Inode *diri = NULL; | |
1324 | if (reply->head.is_dentry) { | |
1325 | diri = add_update_inode(&dirst, request->sent_stamp, session, | |
1326 | request->perms); | |
1327 | update_dir_dist(diri, &dst); // dir stat info is attached to .. | |
1328 | ||
1329 | if (in) { | |
1330 | Dir *dir = diri->open_dir(); | |
1331 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session, | |
1332 | (op == CEPH_MDS_OP_RENAME) ? request->old_dentry() : NULL); | |
1333 | } else { | |
1334 | Dentry *dn = NULL; | |
1335 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1336 | dn = diri->dir->dentries[dname]; | |
1337 | if (dn->inode) { | |
1338 | diri->dir_ordered_count++; | |
1339 | clear_dir_complete_and_ordered(diri, false); | |
1340 | unlink(dn, true, true); // keep dir, dentry | |
1341 | } | |
1342 | } | |
1343 | if (dlease.duration_ms > 0) { | |
1344 | if (!dn) { | |
1345 | Dir *dir = diri->open_dir(); | |
1346 | dn = link(dir, dname, NULL, NULL); | |
1347 | } | |
1348 | update_dentry_lease(dn, &dlease, request->sent_stamp, session); | |
1349 | } | |
1350 | } | |
1351 | } else if (op == CEPH_MDS_OP_LOOKUPSNAP || | |
1352 | op == CEPH_MDS_OP_MKSNAP) { | |
1353 | ldout(cct, 10) << " faking snap lookup weirdness" << dendl; | |
1354 | // fake it for snap lookup | |
1355 | vinodeno_t vino = ist.vino; | |
1356 | vino.snapid = CEPH_SNAPDIR; | |
11fdf7f2 | 1357 | ceph_assert(inode_map.count(vino)); |
7c673cae FG |
1358 | diri = inode_map[vino]; |
1359 | ||
1360 | string dname = request->path.last_dentry(); | |
1361 | ||
1362 | LeaseStat dlease; | |
1363 | dlease.duration_ms = 0; | |
1364 | ||
1365 | if (in) { | |
1366 | Dir *dir = diri->open_dir(); | |
1367 | insert_dentry_inode(dir, dname, &dlease, in, request->sent_stamp, session); | |
1368 | } else { | |
1369 | if (diri->dir && diri->dir->dentries.count(dname)) { | |
1370 | Dentry *dn = diri->dir->dentries[dname]; | |
1371 | if (dn->inode) | |
1372 | unlink(dn, true, true); // keep dir, dentry | |
1373 | } | |
1374 | } | |
1375 | } | |
1376 | ||
1377 | if (in) { | |
1378 | if (op == CEPH_MDS_OP_READDIR || | |
1379 | op == CEPH_MDS_OP_LSSNAP) { | |
1380 | insert_readdir_results(request, session, in); | |
1381 | } else if (op == CEPH_MDS_OP_LOOKUPNAME) { | |
1382 | // hack: return parent inode instead | |
1383 | in = diri; | |
1384 | } | |
1385 | ||
1386 | if (request->dentry() == NULL && in != request->inode()) { | |
1387 | // pin the target inode if its parent dentry is not pinned | |
1388 | request->set_other_inode(in); | |
1389 | } | |
1390 | } | |
1391 | ||
1392 | if (realm) | |
1393 | put_snap_realm(realm); | |
1394 | ||
1395 | request->target = in; | |
1396 | return in; | |
1397 | } | |
1398 | ||
1399 | // ------- | |
1400 | ||
1401 | mds_rank_t Client::choose_target_mds(MetaRequest *req, Inode** phash_diri) | |
1402 | { | |
1403 | mds_rank_t mds = MDS_RANK_NONE; | |
1404 | __u32 hash = 0; | |
1405 | bool is_hash = false; | |
1406 | ||
1407 | Inode *in = NULL; | |
1408 | Dentry *de = NULL; | |
7c673cae FG |
1409 | |
1410 | if (req->resend_mds >= 0) { | |
1411 | mds = req->resend_mds; | |
1412 | req->resend_mds = -1; | |
11fdf7f2 | 1413 | ldout(cct, 10) << __func__ << " resend_mds specified as mds." << mds << dendl; |
7c673cae FG |
1414 | goto out; |
1415 | } | |
1416 | ||
1417 | if (cct->_conf->client_use_random_mds) | |
1418 | goto random_mds; | |
1419 | ||
1420 | in = req->inode(); | |
1421 | de = req->dentry(); | |
1422 | if (in) { | |
11fdf7f2 | 1423 | ldout(cct, 20) << __func__ << " starting with req->inode " << *in << dendl; |
7c673cae FG |
1424 | if (req->path.depth()) { |
1425 | hash = in->hash_dentry_name(req->path[0]); | |
11fdf7f2 | 1426 | ldout(cct, 20) << __func__ << " inode dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1427 | << " on " << req->path[0] |
1428 | << " => " << hash << dendl; | |
1429 | is_hash = true; | |
1430 | } | |
1431 | } else if (de) { | |
1432 | if (de->inode) { | |
1433 | in = de->inode.get(); | |
11fdf7f2 | 1434 | ldout(cct, 20) << __func__ << " starting with req->dentry inode " << *in << dendl; |
7c673cae FG |
1435 | } else { |
1436 | in = de->dir->parent_inode; | |
1437 | hash = in->hash_dentry_name(de->name); | |
11fdf7f2 | 1438 | ldout(cct, 20) << __func__ << " dentry dir hash is " << (int)in->dir_layout.dl_dir_hash |
7c673cae FG |
1439 | << " on " << de->name |
1440 | << " => " << hash << dendl; | |
1441 | is_hash = true; | |
1442 | } | |
1443 | } | |
1444 | if (in) { | |
1445 | if (in->snapid != CEPH_NOSNAP) { | |
11fdf7f2 | 1446 | ldout(cct, 10) << __func__ << " " << *in << " is snapped, using nonsnap parent" << dendl; |
7c673cae FG |
1447 | while (in->snapid != CEPH_NOSNAP) { |
1448 | if (in->snapid == CEPH_SNAPDIR) | |
1449 | in = in->snapdir_parent.get(); | |
11fdf7f2 | 1450 | else if (!in->dentries.empty()) |
7c673cae FG |
1451 | /* In most cases there will only be one dentry, so getting it |
1452 | * will be the correct action. If there are multiple hard links, | |
1453 | * I think the MDS should be able to redirect as needed*/ | |
1454 | in = in->get_first_parent()->dir->parent_inode; | |
1455 | else { | |
1456 | ldout(cct, 10) << "got unlinked inode, can't look at parent" << dendl; | |
1457 | break; | |
1458 | } | |
1459 | } | |
1460 | is_hash = false; | |
1461 | } | |
1462 | ||
11fdf7f2 | 1463 | ldout(cct, 20) << __func__ << " " << *in << " is_hash=" << is_hash |
7c673cae FG |
1464 | << " hash=" << hash << dendl; |
1465 | ||
1466 | if (is_hash && S_ISDIR(in->mode) && !in->fragmap.empty()) { | |
1467 | frag_t fg = in->dirfragtree[hash]; | |
1468 | if (in->fragmap.count(fg)) { | |
1469 | mds = in->fragmap[fg]; | |
1470 | if (phash_diri) | |
1471 | *phash_diri = in; | |
91327a77 AA |
1472 | } else if (in->auth_cap) { |
1473 | mds = in->auth_cap->session->mds_num; | |
1474 | } | |
1475 | if (mds >= 0) { | |
11fdf7f2 | 1476 | ldout(cct, 10) << __func__ << " from dirfragtree hash" << dendl; |
7c673cae FG |
1477 | goto out; |
1478 | } | |
1479 | } | |
1480 | ||
11fdf7f2 TL |
1481 | if (in->auth_cap && req->auth_is_best()) { |
1482 | mds = in->auth_cap->session->mds_num; | |
1483 | } else if (!in->caps.empty()) { | |
1484 | mds = in->caps.begin()->second.session->mds_num; | |
1485 | } else { | |
7c673cae | 1486 | goto random_mds; |
11fdf7f2 TL |
1487 | } |
1488 | ldout(cct, 10) << __func__ << " from caps on inode " << *in << dendl; | |
7c673cae FG |
1489 | |
1490 | goto out; | |
1491 | } | |
1492 | ||
1493 | random_mds: | |
1494 | if (mds < 0) { | |
1495 | mds = _get_random_up_mds(); | |
1496 | ldout(cct, 10) << "did not get mds through better means, so chose random mds " << mds << dendl; | |
1497 | } | |
1498 | ||
1499 | out: | |
1500 | ldout(cct, 20) << "mds is " << mds << dendl; | |
1501 | return mds; | |
1502 | } | |
1503 | ||
1504 | ||
1505 | void Client::connect_mds_targets(mds_rank_t mds) | |
1506 | { | |
11fdf7f2 TL |
1507 | ldout(cct, 10) << __func__ << " for mds." << mds << dendl; |
1508 | ceph_assert(mds_sessions.count(mds)); | |
7c673cae FG |
1509 | const MDSMap::mds_info_t& info = mdsmap->get_mds_info(mds); |
1510 | for (set<mds_rank_t>::const_iterator q = info.export_targets.begin(); | |
1511 | q != info.export_targets.end(); | |
1512 | ++q) { | |
1513 | if (mds_sessions.count(*q) == 0 && | |
1514 | mdsmap->is_clientreplay_or_active_or_stopping(*q)) { | |
1515 | ldout(cct, 10) << "check_mds_sessions opening mds." << mds | |
1516 | << " export target mds." << *q << dendl; | |
1517 | _open_mds_session(*q); | |
1518 | } | |
1519 | } | |
1520 | } | |
1521 | ||
1522 | void Client::dump_mds_sessions(Formatter *f) | |
1523 | { | |
1524 | f->dump_int("id", get_nodeid().v); | |
11fdf7f2 | 1525 | entity_inst_t inst(messenger->get_myname(), messenger->get_myaddr_legacy()); |
1adf2230 AA |
1526 | f->dump_object("inst", inst); |
1527 | f->dump_stream("inst_str") << inst; | |
1528 | f->dump_stream("addr_str") << inst.addr; | |
7c673cae | 1529 | f->open_array_section("sessions"); |
11fdf7f2 | 1530 | for (const auto &p : mds_sessions) { |
7c673cae | 1531 | f->open_object_section("session"); |
11fdf7f2 | 1532 | p.second.dump(f); |
7c673cae FG |
1533 | f->close_section(); |
1534 | } | |
1535 | f->close_section(); | |
1536 | f->dump_int("mdsmap_epoch", mdsmap->get_epoch()); | |
1537 | } | |
1538 | void Client::dump_mds_requests(Formatter *f) | |
1539 | { | |
1540 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
1541 | p != mds_requests.end(); | |
1542 | ++p) { | |
1543 | f->open_object_section("request"); | |
1544 | p->second->dump(f); | |
1545 | f->close_section(); | |
1546 | } | |
1547 | } | |
1548 | ||
9f95a23c | 1549 | int Client::verify_reply_trace(int r, MetaSession *session, |
11fdf7f2 | 1550 | MetaRequest *request, const MConstRef<MClientReply>& reply, |
7c673cae FG |
1551 | InodeRef *ptarget, bool *pcreated, |
1552 | const UserPerm& perms) | |
1553 | { | |
1554 | // check whether this request actually did the create, and set created flag | |
1555 | bufferlist extra_bl; | |
1556 | inodeno_t created_ino; | |
1557 | bool got_created_ino = false; | |
1558 | ceph::unordered_map<vinodeno_t, Inode*>::iterator p; | |
1559 | ||
11fdf7f2 | 1560 | extra_bl = reply->get_extra_bl(); |
7c673cae | 1561 | if (extra_bl.length() >= 8) { |
9f95a23c TL |
1562 | if (session->mds_features.test(CEPHFS_FEATURE_DELEG_INO)) { |
1563 | struct openc_response_t ocres; | |
1564 | ||
1565 | decode(ocres, extra_bl); | |
1566 | created_ino = ocres.created_ino; | |
1567 | /* | |
1568 | * The userland cephfs client doesn't have a way to do an async create | |
1569 | * (yet), so just discard delegated_inos for now. Eventually we should | |
1570 | * store them and use them in create calls, even if they are synchronous, | |
1571 | * if only for testing purposes. | |
1572 | */ | |
1573 | ldout(cct, 10) << "delegated_inos: " << ocres.delegated_inos << dendl; | |
1574 | } else { | |
1575 | // u64 containing number of created ino | |
1576 | decode(created_ino, extra_bl); | |
1577 | } | |
7c673cae | 1578 | ldout(cct, 10) << "make_request created ino " << created_ino << dendl; |
9f95a23c | 1579 | got_created_ino = true; |
7c673cae FG |
1580 | } |
1581 | ||
1582 | if (pcreated) | |
1583 | *pcreated = got_created_ino; | |
1584 | ||
1585 | if (request->target) { | |
1586 | *ptarget = request->target; | |
1587 | ldout(cct, 20) << "make_request target is " << *ptarget->get() << dendl; | |
1588 | } else { | |
1589 | if (got_created_ino && (p = inode_map.find(vinodeno_t(created_ino, CEPH_NOSNAP))) != inode_map.end()) { | |
1590 | (*ptarget) = p->second; | |
1591 | ldout(cct, 20) << "make_request created, target is " << *ptarget->get() << dendl; | |
1592 | } else { | |
1593 | // we got a traceless reply, and need to look up what we just | |
1594 | // created. for now, do this by name. someday, do this by the | |
1595 | // ino... which we know! FIXME. | |
1596 | InodeRef target; | |
1597 | Dentry *d = request->dentry(); | |
1598 | if (d) { | |
1599 | if (d->dir) { | |
1600 | ldout(cct, 10) << "make_request got traceless reply, looking up #" | |
1601 | << d->dir->parent_inode->ino << "/" << d->name | |
1602 | << " got_ino " << got_created_ino | |
1603 | << " ino " << created_ino | |
1604 | << dendl; | |
1605 | r = _do_lookup(d->dir->parent_inode, d->name, request->regetattr_mask, | |
1606 | &target, perms); | |
1607 | } else { | |
1608 | // if the dentry is not linked, just do our best. see #5021. | |
11fdf7f2 | 1609 | ceph_abort_msg("how did this happen? i want logs!"); |
7c673cae FG |
1610 | } |
1611 | } else { | |
1612 | Inode *in = request->inode(); | |
1613 | ldout(cct, 10) << "make_request got traceless reply, forcing getattr on #" | |
1614 | << in->ino << dendl; | |
1615 | r = _getattr(in, request->regetattr_mask, perms, true); | |
1616 | target = in; | |
1617 | } | |
1618 | if (r >= 0) { | |
1619 | // verify ino returned in reply and trace_dist are the same | |
1620 | if (got_created_ino && | |
1621 | created_ino.val != target->ino.val) { | |
1622 | ldout(cct, 5) << "create got ino " << created_ino << " but then failed on lookup; EINTR?" << dendl; | |
1623 | r = -EINTR; | |
1624 | } | |
1625 | if (ptarget) | |
1626 | ptarget->swap(target); | |
1627 | } | |
1628 | } | |
1629 | } | |
1630 | ||
1631 | return r; | |
1632 | } | |
1633 | ||
1634 | ||
1635 | /** | |
1636 | * make a request | |
1637 | * | |
1638 | * Blocking helper to make an MDS request. | |
1639 | * | |
1640 | * If the ptarget flag is set, behavior changes slightly: the caller | |
1641 | * expects to get a pointer to the inode we are creating or operating | |
1642 | * on. As a result, we will follow up any traceless mutation reply | |
1643 | * with a getattr or lookup to transparently handle a traceless reply | |
1644 | * from the MDS (as when the MDS restarts and the client has to replay | |
1645 | * a request). | |
1646 | * | |
1647 | * @param request the MetaRequest to execute | |
1648 | * @param perms The user uid/gid to execute as (eventually, full group lists?) | |
1649 | * @param ptarget [optional] address to store a pointer to the target inode we want to create or operate on | |
1650 | * @param pcreated [optional; required if ptarget] where to store a bool of whether our create atomically created a file | |
1651 | * @param use_mds [optional] prefer a specific mds (-1 for default) | |
1652 | * @param pdirbl [optional; disallowed if ptarget] where to pass extra reply payload to the caller | |
1653 | */ | |
1654 | int Client::make_request(MetaRequest *request, | |
1655 | const UserPerm& perms, | |
1656 | InodeRef *ptarget, bool *pcreated, | |
1657 | mds_rank_t use_mds, | |
1658 | bufferlist *pdirbl) | |
1659 | { | |
1660 | int r = 0; | |
1661 | ||
1662 | // assign a unique tid | |
1663 | ceph_tid_t tid = ++last_tid; | |
1664 | request->set_tid(tid); | |
1665 | ||
1666 | // and timestamp | |
1667 | request->op_stamp = ceph_clock_now(); | |
1668 | ||
1669 | // make note | |
1670 | mds_requests[tid] = request->get(); | |
1671 | if (oldest_tid == 0 && request->get_op() != CEPH_MDS_OP_SETFILELOCK) | |
1672 | oldest_tid = tid; | |
1673 | ||
1674 | request->set_caller_perms(perms); | |
1675 | ||
1676 | if (cct->_conf->client_inject_fixed_oldest_tid) { | |
1677 | ldout(cct, 20) << __func__ << " injecting fixed oldest_client_tid(1)" << dendl; | |
1678 | request->set_oldest_client_tid(1); | |
1679 | } else { | |
1680 | request->set_oldest_client_tid(oldest_tid); | |
1681 | } | |
1682 | ||
1683 | // hack target mds? | |
1684 | if (use_mds >= 0) | |
1685 | request->resend_mds = use_mds; | |
1686 | ||
9f95a23c | 1687 | MetaSession *session = NULL; |
7c673cae FG |
1688 | while (1) { |
1689 | if (request->aborted()) | |
1690 | break; | |
1691 | ||
31f18b77 FG |
1692 | if (blacklisted) { |
1693 | request->abort(-EBLACKLISTED); | |
1694 | break; | |
1695 | } | |
1696 | ||
7c673cae | 1697 | // set up wait cond |
9f95a23c | 1698 | ceph::condition_variable caller_cond; |
7c673cae FG |
1699 | request->caller_cond = &caller_cond; |
1700 | ||
1701 | // choose mds | |
1702 | Inode *hash_diri = NULL; | |
1703 | mds_rank_t mds = choose_target_mds(request, &hash_diri); | |
1704 | int mds_state = (mds == MDS_RANK_NONE) ? MDSMap::STATE_NULL : mdsmap->get_state(mds); | |
1705 | if (mds_state != MDSMap::STATE_ACTIVE && mds_state != MDSMap::STATE_STOPPING) { | |
1706 | if (mds_state == MDSMap::STATE_NULL && mds >= mdsmap->get_max_mds()) { | |
1707 | if (hash_diri) { | |
1708 | ldout(cct, 10) << " target mds." << mds << " has stopped, remove it from fragmap" << dendl; | |
1709 | _fragmap_remove_stopped_mds(hash_diri, mds); | |
1710 | } else { | |
1711 | ldout(cct, 10) << " target mds." << mds << " has stopped, trying a random mds" << dendl; | |
1712 | request->resend_mds = _get_random_up_mds(); | |
1713 | } | |
1714 | } else { | |
1715 | ldout(cct, 10) << " target mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
1716 | wait_on_list(waiting_for_mdsmap); | |
1717 | } | |
1718 | continue; | |
1719 | } | |
1720 | ||
1721 | // open a session? | |
7c673cae FG |
1722 | if (!have_open_session(mds)) { |
1723 | session = _get_or_open_mds_session(mds); | |
1724 | ||
1725 | // wait | |
1726 | if (session->state == MetaSession::STATE_OPENING) { | |
1727 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
1728 | wait_on_context_list(session->waiting_for_open); | |
1729 | // Abort requests on REJECT from MDS | |
1730 | if (rejected_by_mds.count(mds)) { | |
1731 | request->abort(-EPERM); | |
1732 | break; | |
1733 | } | |
1734 | continue; | |
1735 | } | |
1736 | ||
1737 | if (!have_open_session(mds)) | |
1738 | continue; | |
1739 | } else { | |
11fdf7f2 | 1740 | session = &mds_sessions.at(mds); |
7c673cae FG |
1741 | } |
1742 | ||
1743 | // send request. | |
1744 | send_request(request, session); | |
1745 | ||
1746 | // wait for signal | |
1747 | ldout(cct, 20) << "awaiting reply|forward|kick on " << &caller_cond << dendl; | |
1748 | request->kick = false; | |
9f95a23c TL |
1749 | std::unique_lock l{client_lock, std::adopt_lock}; |
1750 | caller_cond.wait(l, [request] { | |
1751 | return (request->reply || // reply | |
1752 | request->resend_mds >= 0 || // forward | |
1753 | request->kick); | |
1754 | }); | |
1755 | l.release(); | |
1756 | request->caller_cond = nullptr; | |
7c673cae FG |
1757 | |
1758 | // did we get a reply? | |
1759 | if (request->reply) | |
1760 | break; | |
1761 | } | |
1762 | ||
1763 | if (!request->reply) { | |
11fdf7f2 TL |
1764 | ceph_assert(request->aborted()); |
1765 | ceph_assert(!request->got_unsafe); | |
7c673cae FG |
1766 | r = request->get_abort_code(); |
1767 | request->item.remove_myself(); | |
1768 | unregister_request(request); | |
11fdf7f2 | 1769 | put_request(request); |
7c673cae FG |
1770 | return r; |
1771 | } | |
1772 | ||
1773 | // got it! | |
11fdf7f2 | 1774 | auto reply = std::move(request->reply); |
7c673cae FG |
1775 | r = reply->get_result(); |
1776 | if (r >= 0) | |
1777 | request->success = true; | |
1778 | ||
1779 | // kick dispatcher (we've got it!) | |
11fdf7f2 | 1780 | ceph_assert(request->dispatch_cond); |
9f95a23c | 1781 | request->dispatch_cond->notify_all(); |
7c673cae FG |
1782 | ldout(cct, 20) << "sendrecv kickback on tid " << tid << " " << request->dispatch_cond << dendl; |
1783 | request->dispatch_cond = 0; | |
1784 | ||
1785 | if (r >= 0 && ptarget) | |
9f95a23c | 1786 | r = verify_reply_trace(r, session, request, reply, ptarget, pcreated, perms); |
7c673cae FG |
1787 | |
1788 | if (pdirbl) | |
11fdf7f2 | 1789 | *pdirbl = reply->get_extra_bl(); |
7c673cae FG |
1790 | |
1791 | // -- log times -- | |
1792 | utime_t lat = ceph_clock_now(); | |
1793 | lat -= request->sent_stamp; | |
1794 | ldout(cct, 20) << "lat " << lat << dendl; | |
1795 | logger->tinc(l_c_lat, lat); | |
1796 | logger->tinc(l_c_reply, lat); | |
1797 | ||
1798 | put_request(request); | |
7c673cae FG |
1799 | return r; |
1800 | } | |
1801 | ||
1802 | void Client::unregister_request(MetaRequest *req) | |
1803 | { | |
1804 | mds_requests.erase(req->tid); | |
1805 | if (req->tid == oldest_tid) { | |
1806 | map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.upper_bound(oldest_tid); | |
1807 | while (true) { | |
1808 | if (p == mds_requests.end()) { | |
1809 | oldest_tid = 0; | |
1810 | break; | |
1811 | } | |
1812 | if (p->second->get_op() != CEPH_MDS_OP_SETFILELOCK) { | |
1813 | oldest_tid = p->first; | |
1814 | break; | |
1815 | } | |
1816 | ++p; | |
1817 | } | |
1818 | } | |
1819 | put_request(req); | |
1820 | } | |
1821 | ||
1822 | void Client::put_request(MetaRequest *request) | |
1823 | { | |
1824 | if (request->_put()) { | |
1825 | int op = -1; | |
1826 | if (request->success) | |
1827 | op = request->get_op(); | |
1828 | InodeRef other_in; | |
1829 | request->take_other_inode(&other_in); | |
1830 | delete request; | |
1831 | ||
1832 | if (other_in && | |
1833 | (op == CEPH_MDS_OP_RMDIR || | |
1834 | op == CEPH_MDS_OP_RENAME || | |
1835 | op == CEPH_MDS_OP_RMSNAP)) { | |
1836 | _try_to_trim_inode(other_in.get(), false); | |
1837 | } | |
1838 | } | |
1839 | } | |
1840 | ||
1841 | int Client::encode_inode_release(Inode *in, MetaRequest *req, | |
1842 | mds_rank_t mds, int drop, | |
1843 | int unless, int force) | |
1844 | { | |
11fdf7f2 | 1845 | ldout(cct, 20) << __func__ << " enter(in:" << *in << ", req:" << req |
7c673cae FG |
1846 | << " mds:" << mds << ", drop:" << drop << ", unless:" << unless |
1847 | << ", have:" << ", force:" << force << ")" << dendl; | |
1848 | int released = 0; | |
11fdf7f2 TL |
1849 | auto it = in->caps.find(mds); |
1850 | if (it != in->caps.end()) { | |
1851 | Cap &cap = it->second; | |
7c673cae | 1852 | drop &= ~(in->dirty_caps | get_caps_used(in)); |
11fdf7f2 TL |
1853 | if ((drop & cap.issued) && |
1854 | !(unless & cap.issued)) { | |
1855 | ldout(cct, 25) << "Dropping caps. Initial " << ccap_string(cap.issued) << dendl; | |
1856 | cap.issued &= ~drop; | |
1857 | cap.implemented &= ~drop; | |
7c673cae | 1858 | released = 1; |
11fdf7f2 | 1859 | ldout(cct, 25) << "Now have: " << ccap_string(cap.issued) << dendl; |
7c673cae FG |
1860 | } else { |
1861 | released = force; | |
1862 | } | |
1863 | if (released) { | |
1864 | ceph_mds_request_release rel; | |
1865 | rel.ino = in->ino; | |
11fdf7f2 TL |
1866 | rel.cap_id = cap.cap_id; |
1867 | rel.seq = cap.seq; | |
1868 | rel.issue_seq = cap.issue_seq; | |
1869 | rel.mseq = cap.mseq; | |
1870 | rel.caps = cap.implemented; | |
1871 | rel.wanted = cap.wanted; | |
7c673cae FG |
1872 | rel.dname_len = 0; |
1873 | rel.dname_seq = 0; | |
1874 | req->cap_releases.push_back(MClientRequest::Release(rel,"")); | |
1875 | } | |
1876 | } | |
11fdf7f2 | 1877 | ldout(cct, 25) << __func__ << " exit(in:" << *in << ") released:" |
7c673cae FG |
1878 | << released << dendl; |
1879 | return released; | |
1880 | } | |
1881 | ||
1882 | void Client::encode_dentry_release(Dentry *dn, MetaRequest *req, | |
1883 | mds_rank_t mds, int drop, int unless) | |
1884 | { | |
11fdf7f2 | 1885 | ldout(cct, 20) << __func__ << " enter(dn:" |
7c673cae FG |
1886 | << dn << ")" << dendl; |
1887 | int released = 0; | |
1888 | if (dn->dir) | |
1889 | released = encode_inode_release(dn->dir->parent_inode, req, | |
1890 | mds, drop, unless, 1); | |
1891 | if (released && dn->lease_mds == mds) { | |
1892 | ldout(cct, 25) << "preemptively releasing dn to mds" << dendl; | |
11fdf7f2 | 1893 | auto& rel = req->cap_releases.back(); |
7c673cae FG |
1894 | rel.item.dname_len = dn->name.length(); |
1895 | rel.item.dname_seq = dn->lease_seq; | |
1896 | rel.dname = dn->name; | |
1897 | } | |
11fdf7f2 | 1898 | ldout(cct, 25) << __func__ << " exit(dn:" |
7c673cae FG |
1899 | << dn << ")" << dendl; |
1900 | } | |
1901 | ||
1902 | ||
1903 | /* | |
1904 | * This requires the MClientRequest *request member to be set. | |
1905 | * It will error out horribly without one. | |
1906 | * Additionally, if you set any *drop member, you'd better have | |
1907 | * set the corresponding dentry! | |
1908 | */ | |
1909 | void Client::encode_cap_releases(MetaRequest *req, mds_rank_t mds) | |
1910 | { | |
11fdf7f2 | 1911 | ldout(cct, 20) << __func__ << " enter (req: " |
7c673cae FG |
1912 | << req << ", mds: " << mds << ")" << dendl; |
1913 | if (req->inode_drop && req->inode()) | |
1914 | encode_inode_release(req->inode(), req, | |
1915 | mds, req->inode_drop, | |
1916 | req->inode_unless); | |
1917 | ||
1918 | if (req->old_inode_drop && req->old_inode()) | |
1919 | encode_inode_release(req->old_inode(), req, | |
1920 | mds, req->old_inode_drop, | |
1921 | req->old_inode_unless); | |
1922 | if (req->other_inode_drop && req->other_inode()) | |
1923 | encode_inode_release(req->other_inode(), req, | |
1924 | mds, req->other_inode_drop, | |
1925 | req->other_inode_unless); | |
1926 | ||
1927 | if (req->dentry_drop && req->dentry()) | |
1928 | encode_dentry_release(req->dentry(), req, | |
1929 | mds, req->dentry_drop, | |
1930 | req->dentry_unless); | |
1931 | ||
1932 | if (req->old_dentry_drop && req->old_dentry()) | |
1933 | encode_dentry_release(req->old_dentry(), req, | |
1934 | mds, req->old_dentry_drop, | |
1935 | req->old_dentry_unless); | |
11fdf7f2 | 1936 | ldout(cct, 25) << __func__ << " exit (req: " |
7c673cae FG |
1937 | << req << ", mds " << mds <<dendl; |
1938 | } | |
1939 | ||
1940 | bool Client::have_open_session(mds_rank_t mds) | |
1941 | { | |
11fdf7f2 TL |
1942 | const auto &it = mds_sessions.find(mds); |
1943 | return it != mds_sessions.end() && | |
1944 | (it->second.state == MetaSession::STATE_OPEN || | |
1945 | it->second.state == MetaSession::STATE_STALE); | |
7c673cae FG |
1946 | } |
1947 | ||
1948 | MetaSession *Client::_get_mds_session(mds_rank_t mds, Connection *con) | |
1949 | { | |
11fdf7f2 TL |
1950 | const auto &it = mds_sessions.find(mds); |
1951 | if (it == mds_sessions.end() || it->second.con != con) { | |
7c673cae | 1952 | return NULL; |
11fdf7f2 TL |
1953 | } else { |
1954 | return &it->second; | |
1955 | } | |
7c673cae FG |
1956 | } |
1957 | ||
1958 | MetaSession *Client::_get_or_open_mds_session(mds_rank_t mds) | |
1959 | { | |
11fdf7f2 TL |
1960 | auto it = mds_sessions.find(mds); |
1961 | return it == mds_sessions.end() ? _open_mds_session(mds) : &it->second; | |
7c673cae FG |
1962 | } |
1963 | ||
1964 | /** | |
1965 | * Populate a map of strings with client-identifying metadata, | |
1966 | * such as the hostname. Call this once at initialization. | |
1967 | */ | |
1968 | void Client::populate_metadata(const std::string &mount_root) | |
1969 | { | |
1970 | // Hostname | |
1971 | struct utsname u; | |
1972 | int r = uname(&u); | |
1973 | if (r >= 0) { | |
1974 | metadata["hostname"] = u.nodename; | |
1975 | ldout(cct, 20) << __func__ << " read hostname '" << u.nodename << "'" << dendl; | |
1976 | } else { | |
1977 | ldout(cct, 1) << __func__ << " failed to read hostname (" << cpp_strerror(r) << ")" << dendl; | |
1978 | } | |
1979 | ||
1980 | metadata["pid"] = stringify(getpid()); | |
1981 | ||
1982 | // Ceph entity id (the '0' in "client.0") | |
1983 | metadata["entity_id"] = cct->_conf->name.get_id(); | |
1984 | ||
1985 | // Our mount position | |
1986 | if (!mount_root.empty()) { | |
1987 | metadata["root"] = mount_root; | |
1988 | } | |
1989 | ||
1990 | // Ceph version | |
1991 | metadata["ceph_version"] = pretty_version_to_str(); | |
1992 | metadata["ceph_sha1"] = git_version_to_str(); | |
1993 | ||
1994 | // Apply any metadata from the user's configured overrides | |
1995 | std::vector<std::string> tokens; | |
1996 | get_str_vec(cct->_conf->client_metadata, ",", tokens); | |
1997 | for (const auto &i : tokens) { | |
1998 | auto eqpos = i.find("="); | |
1999 | // Throw out anything that isn't of the form "<str>=<str>" | |
2000 | if (eqpos == 0 || eqpos == std::string::npos || eqpos == i.size()) { | |
2001 | lderr(cct) << "Invalid metadata keyval pair: '" << i << "'" << dendl; | |
2002 | continue; | |
2003 | } | |
2004 | metadata[i.substr(0, eqpos)] = i.substr(eqpos + 1); | |
2005 | } | |
2006 | } | |
2007 | ||
2008 | /** | |
2009 | * Optionally add or override client metadata fields. | |
2010 | */ | |
2011 | void Client::update_metadata(std::string const &k, std::string const &v) | |
2012 | { | |
11fdf7f2 TL |
2013 | std::lock_guard l(client_lock); |
2014 | ceph_assert(initialized); | |
7c673cae | 2015 | |
11fdf7f2 TL |
2016 | auto it = metadata.find(k); |
2017 | if (it != metadata.end()) { | |
7c673cae | 2018 | ldout(cct, 1) << __func__ << " warning, overriding metadata field '" << k |
11fdf7f2 | 2019 | << "' from '" << it->second << "' to '" << v << "'" << dendl; |
7c673cae FG |
2020 | } |
2021 | ||
2022 | metadata[k] = v; | |
2023 | } | |
2024 | ||
2025 | MetaSession *Client::_open_mds_session(mds_rank_t mds) | |
2026 | { | |
11fdf7f2 TL |
2027 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
2028 | auto addrs = mdsmap->get_addrs(mds); | |
2029 | auto em = mds_sessions.emplace(std::piecewise_construct, | |
2030 | std::forward_as_tuple(mds), | |
2031 | std::forward_as_tuple(mds, messenger->connect_to_mds(addrs), addrs)); | |
2032 | ceph_assert(em.second); /* not already present */ | |
2033 | MetaSession *session = &em.first->second; | |
7c673cae FG |
2034 | |
2035 | // Maybe skip sending a request to open if this MDS daemon | |
2036 | // has previously sent us a REJECT. | |
2037 | if (rejected_by_mds.count(mds)) { | |
11fdf7f2 TL |
2038 | if (rejected_by_mds[mds] == session->addrs) { |
2039 | ldout(cct, 4) << __func__ << " mds." << mds << " skipping " | |
7c673cae FG |
2040 | "because we were rejected" << dendl; |
2041 | return session; | |
2042 | } else { | |
11fdf7f2 | 2043 | ldout(cct, 4) << __func__ << " mds." << mds << " old inst " |
7c673cae FG |
2044 | "rejected us, trying with new inst" << dendl; |
2045 | rejected_by_mds.erase(mds); | |
2046 | } | |
2047 | } | |
2048 | ||
9f95a23c | 2049 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_OPEN); |
11fdf7f2 TL |
2050 | m->metadata = metadata; |
2051 | m->supported_features = feature_bitset_t(CEPHFS_FEATURES_CLIENT_SUPPORTED); | |
2052 | session->con->send_message2(std::move(m)); | |
7c673cae FG |
2053 | return session; |
2054 | } | |
2055 | ||
2056 | void Client::_close_mds_session(MetaSession *s) | |
2057 | { | |
11fdf7f2 | 2058 | ldout(cct, 2) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
7c673cae | 2059 | s->state = MetaSession::STATE_CLOSING; |
9f95a23c | 2060 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
2061 | } |
2062 | ||
2063 | void Client::_closed_mds_session(MetaSession *s) | |
2064 | { | |
11fdf7f2 | 2065 | ldout(cct, 5) << __func__ << " mds." << s->mds_num << " seq " << s->seq << dendl; |
7c673cae FG |
2066 | s->state = MetaSession::STATE_CLOSED; |
2067 | s->con->mark_down(); | |
2068 | signal_context_list(s->waiting_for_open); | |
9f95a23c | 2069 | mount_cond.notify_all(); |
7c673cae FG |
2070 | remove_session_caps(s); |
2071 | kick_requests_closed(s); | |
2072 | mds_sessions.erase(s->mds_num); | |
7c673cae FG |
2073 | } |
2074 | ||
11fdf7f2 | 2075 | void Client::handle_client_session(const MConstRef<MClientSession>& m) |
7c673cae FG |
2076 | { |
2077 | mds_rank_t from = mds_rank_t(m->get_source().num()); | |
11fdf7f2 | 2078 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << from << dendl; |
7c673cae FG |
2079 | |
2080 | MetaSession *session = _get_mds_session(from, m->get_connection().get()); | |
2081 | if (!session) { | |
2082 | ldout(cct, 10) << " discarding session message from sessionless mds " << m->get_source_inst() << dendl; | |
7c673cae FG |
2083 | return; |
2084 | } | |
2085 | ||
2086 | switch (m->get_op()) { | |
2087 | case CEPH_SESSION_OPEN: | |
11fdf7f2 TL |
2088 | { |
2089 | feature_bitset_t missing_features(CEPHFS_FEATURES_CLIENT_REQUIRED); | |
2090 | missing_features -= m->supported_features; | |
2091 | if (!missing_features.empty()) { | |
2092 | lderr(cct) << "mds." << from << " lacks required features '" | |
2093 | << missing_features << "', closing session " << dendl; | |
2094 | rejected_by_mds[session->mds_num] = session->addrs; | |
2095 | _close_mds_session(session); | |
2096 | _closed_mds_session(session); | |
2097 | break; | |
2098 | } | |
2099 | session->mds_features = std::move(m->supported_features); | |
2100 | ||
2101 | renew_caps(session); | |
2102 | session->state = MetaSession::STATE_OPEN; | |
2103 | if (unmounting) | |
9f95a23c | 2104 | mount_cond.notify_all(); |
11fdf7f2 TL |
2105 | else |
2106 | connect_mds_targets(from); | |
2107 | signal_context_list(session->waiting_for_open); | |
2108 | break; | |
2109 | } | |
7c673cae FG |
2110 | |
2111 | case CEPH_SESSION_CLOSE: | |
2112 | _closed_mds_session(session); | |
2113 | break; | |
2114 | ||
2115 | case CEPH_SESSION_RENEWCAPS: | |
2116 | if (session->cap_renew_seq == m->get_seq()) { | |
a8e16298 | 2117 | bool was_stale = ceph_clock_now() >= session->cap_ttl; |
7c673cae FG |
2118 | session->cap_ttl = |
2119 | session->last_cap_renew_request + mdsmap->get_session_timeout(); | |
a8e16298 TL |
2120 | if (was_stale) |
2121 | wake_up_session_caps(session, false); | |
7c673cae FG |
2122 | } |
2123 | break; | |
2124 | ||
2125 | case CEPH_SESSION_STALE: | |
28e407b8 AA |
2126 | // invalidate session caps/leases |
2127 | session->cap_gen++; | |
2128 | session->cap_ttl = ceph_clock_now(); | |
2129 | session->cap_ttl -= 1; | |
7c673cae FG |
2130 | renew_caps(session); |
2131 | break; | |
2132 | ||
2133 | case CEPH_SESSION_RECALL_STATE: | |
2134 | trim_caps(session, m->get_max_caps()); | |
2135 | break; | |
2136 | ||
2137 | case CEPH_SESSION_FLUSHMSG: | |
a8e16298 | 2138 | /* flush cap release */ |
11fdf7f2 TL |
2139 | if (auto& m = session->release; m) { |
2140 | session->con->send_message2(std::move(m)); | |
a8e16298 | 2141 | } |
9f95a23c | 2142 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_FLUSHMSG_ACK, m->get_seq())); |
7c673cae FG |
2143 | break; |
2144 | ||
2145 | case CEPH_SESSION_FORCE_RO: | |
2146 | force_session_readonly(session); | |
2147 | break; | |
2148 | ||
2149 | case CEPH_SESSION_REJECT: | |
11fdf7f2 TL |
2150 | { |
2151 | std::string_view error_str; | |
2152 | auto it = m->metadata.find("error_string"); | |
2153 | if (it != m->metadata.end()) | |
2154 | error_str = it->second; | |
2155 | else | |
2156 | error_str = "unknown error"; | |
2157 | lderr(cct) << "mds." << from << " rejected us (" << error_str << ")" << dendl; | |
7c673cae | 2158 | |
11fdf7f2 TL |
2159 | rejected_by_mds[session->mds_num] = session->addrs; |
2160 | _closed_mds_session(session); | |
2161 | } | |
7c673cae FG |
2162 | break; |
2163 | ||
2164 | default: | |
2165 | ceph_abort(); | |
2166 | } | |
7c673cae FG |
2167 | } |
2168 | ||
2169 | bool Client::_any_stale_sessions() const | |
2170 | { | |
9f95a23c | 2171 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae | 2172 | |
11fdf7f2 TL |
2173 | for (const auto &p : mds_sessions) { |
2174 | if (p.second.state == MetaSession::STATE_STALE) { | |
7c673cae FG |
2175 | return true; |
2176 | } | |
2177 | } | |
2178 | ||
2179 | return false; | |
2180 | } | |
2181 | ||
2182 | void Client::_kick_stale_sessions() | |
2183 | { | |
11fdf7f2 | 2184 | ldout(cct, 1) << __func__ << dendl; |
7c673cae | 2185 | |
11fdf7f2 TL |
2186 | for (auto it = mds_sessions.begin(); it != mds_sessions.end(); ) { |
2187 | MetaSession &s = it->second; | |
2188 | ++it; | |
2189 | if (s.state == MetaSession::STATE_STALE) | |
2190 | _closed_mds_session(&s); | |
7c673cae FG |
2191 | } |
2192 | } | |
2193 | ||
2194 | void Client::send_request(MetaRequest *request, MetaSession *session, | |
2195 | bool drop_cap_releases) | |
2196 | { | |
2197 | // make the request | |
2198 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2199 | ldout(cct, 10) << __func__ << " rebuilding request " << request->get_tid() |
7c673cae | 2200 | << " for mds." << mds << dendl; |
11fdf7f2 | 2201 | auto r = build_client_request(request); |
7c673cae FG |
2202 | if (request->dentry()) { |
2203 | r->set_dentry_wanted(); | |
2204 | } | |
2205 | if (request->got_unsafe) { | |
2206 | r->set_replayed_op(); | |
2207 | if (request->target) | |
2208 | r->head.ino = request->target->ino; | |
2209 | } else { | |
2210 | encode_cap_releases(request, mds); | |
2211 | if (drop_cap_releases) // we haven't send cap reconnect yet, drop cap releases | |
2212 | request->cap_releases.clear(); | |
2213 | else | |
2214 | r->releases.swap(request->cap_releases); | |
2215 | } | |
2216 | r->set_mdsmap_epoch(mdsmap->get_epoch()); | |
2217 | if (r->head.op == CEPH_MDS_OP_SETXATTR) { | |
2218 | objecter->with_osdmap([r](const OSDMap& o) { | |
2219 | r->set_osdmap_epoch(o.get_epoch()); | |
2220 | }); | |
2221 | } | |
2222 | ||
2223 | if (request->mds == -1) { | |
2224 | request->sent_stamp = ceph_clock_now(); | |
11fdf7f2 | 2225 | ldout(cct, 20) << __func__ << " set sent_stamp to " << request->sent_stamp << dendl; |
7c673cae FG |
2226 | } |
2227 | request->mds = mds; | |
2228 | ||
2229 | Inode *in = request->inode(); | |
11fdf7f2 TL |
2230 | if (in) { |
2231 | auto it = in->caps.find(mds); | |
2232 | if (it != in->caps.end()) { | |
2233 | request->sent_on_mseq = it->second.mseq; | |
2234 | } | |
2235 | } | |
7c673cae FG |
2236 | |
2237 | session->requests.push_back(&request->item); | |
2238 | ||
11fdf7f2 TL |
2239 | ldout(cct, 10) << __func__ << " " << *r << " to mds." << mds << dendl; |
2240 | session->con->send_message2(std::move(r)); | |
7c673cae FG |
2241 | } |
2242 | ||
9f95a23c | 2243 | ref_t<MClientRequest> Client::build_client_request(MetaRequest *request) |
7c673cae | 2244 | { |
9f95a23c | 2245 | auto req = make_message<MClientRequest>(request->get_op()); |
7c673cae FG |
2246 | req->set_tid(request->tid); |
2247 | req->set_stamp(request->op_stamp); | |
2248 | memcpy(&req->head, &request->head, sizeof(ceph_mds_request_head)); | |
2249 | ||
2250 | // if the filepath's haven't been set, set them! | |
2251 | if (request->path.empty()) { | |
2252 | Inode *in = request->inode(); | |
2253 | Dentry *de = request->dentry(); | |
2254 | if (in) | |
2255 | in->make_nosnap_relative_path(request->path); | |
2256 | else if (de) { | |
2257 | if (de->inode) | |
2258 | de->inode->make_nosnap_relative_path(request->path); | |
2259 | else if (de->dir) { | |
2260 | de->dir->parent_inode->make_nosnap_relative_path(request->path); | |
2261 | request->path.push_dentry(de->name); | |
2262 | } | |
2263 | else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2264 | << " No path, inode, or appropriately-endowed dentry given!" | |
2265 | << dendl; | |
2266 | } else ldout(cct, 1) << "Warning -- unable to construct a filepath!" | |
2267 | << " No path, inode, or dentry given!" | |
2268 | << dendl; | |
2269 | } | |
2270 | req->set_filepath(request->get_filepath()); | |
2271 | req->set_filepath2(request->get_filepath2()); | |
2272 | req->set_data(request->data); | |
2273 | req->set_retry_attempt(request->retry_attempt++); | |
2274 | req->head.num_fwd = request->num_fwd; | |
2275 | const gid_t *_gids; | |
2276 | int gid_count = request->perms.get_gids(&_gids); | |
2277 | req->set_gid_list(gid_count, _gids); | |
2278 | return req; | |
2279 | } | |
2280 | ||
2281 | ||
2282 | ||
11fdf7f2 | 2283 | void Client::handle_client_request_forward(const MConstRef<MClientRequestForward>& fwd) |
7c673cae FG |
2284 | { |
2285 | mds_rank_t mds = mds_rank_t(fwd->get_source().num()); | |
2286 | MetaSession *session = _get_mds_session(mds, fwd->get_connection().get()); | |
2287 | if (!session) { | |
7c673cae FG |
2288 | return; |
2289 | } | |
2290 | ceph_tid_t tid = fwd->get_tid(); | |
2291 | ||
2292 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2293 | ldout(cct, 10) << __func__ << " no pending request on tid " << tid << dendl; |
7c673cae FG |
2294 | return; |
2295 | } | |
2296 | ||
2297 | MetaRequest *request = mds_requests[tid]; | |
11fdf7f2 | 2298 | ceph_assert(request); |
7c673cae FG |
2299 | |
2300 | // reset retry counter | |
2301 | request->retry_attempt = 0; | |
2302 | ||
2303 | // request not forwarded, or dest mds has no session. | |
2304 | // resend. | |
11fdf7f2 | 2305 | ldout(cct, 10) << __func__ << " tid " << tid |
7c673cae FG |
2306 | << " fwd " << fwd->get_num_fwd() |
2307 | << " to mds." << fwd->get_dest_mds() | |
2308 | << ", resending to " << fwd->get_dest_mds() | |
2309 | << dendl; | |
2310 | ||
2311 | request->mds = -1; | |
2312 | request->item.remove_myself(); | |
2313 | request->num_fwd = fwd->get_num_fwd(); | |
2314 | request->resend_mds = fwd->get_dest_mds(); | |
9f95a23c | 2315 | request->caller_cond->notify_all(); |
7c673cae FG |
2316 | } |
2317 | ||
2318 | bool Client::is_dir_operation(MetaRequest *req) | |
2319 | { | |
2320 | int op = req->get_op(); | |
2321 | if (op == CEPH_MDS_OP_MKNOD || op == CEPH_MDS_OP_LINK || | |
2322 | op == CEPH_MDS_OP_UNLINK || op == CEPH_MDS_OP_RENAME || | |
2323 | op == CEPH_MDS_OP_MKDIR || op == CEPH_MDS_OP_RMDIR || | |
2324 | op == CEPH_MDS_OP_SYMLINK || op == CEPH_MDS_OP_CREATE) | |
2325 | return true; | |
2326 | return false; | |
2327 | } | |
2328 | ||
11fdf7f2 | 2329 | void Client::handle_client_reply(const MConstRef<MClientReply>& reply) |
7c673cae FG |
2330 | { |
2331 | mds_rank_t mds_num = mds_rank_t(reply->get_source().num()); | |
2332 | MetaSession *session = _get_mds_session(mds_num, reply->get_connection().get()); | |
2333 | if (!session) { | |
7c673cae FG |
2334 | return; |
2335 | } | |
2336 | ||
2337 | ceph_tid_t tid = reply->get_tid(); | |
2338 | bool is_safe = reply->is_safe(); | |
2339 | ||
2340 | if (mds_requests.count(tid) == 0) { | |
11fdf7f2 | 2341 | lderr(cct) << __func__ << " no pending request on tid " << tid |
7c673cae | 2342 | << " safe is:" << is_safe << dendl; |
7c673cae FG |
2343 | return; |
2344 | } | |
2345 | MetaRequest *request = mds_requests.at(tid); | |
2346 | ||
11fdf7f2 | 2347 | ldout(cct, 20) << __func__ << " got a reply. Safe:" << is_safe |
7c673cae FG |
2348 | << " tid " << tid << dendl; |
2349 | ||
2350 | if (request->got_unsafe && !is_safe) { | |
2351 | //duplicate response | |
2352 | ldout(cct, 0) << "got a duplicate reply on tid " << tid << " from mds " | |
2353 | << mds_num << " safe:" << is_safe << dendl; | |
7c673cae FG |
2354 | return; |
2355 | } | |
2356 | ||
2357 | if (-ESTALE == reply->get_result()) { // see if we can get to proper MDS | |
2358 | ldout(cct, 20) << "got ESTALE on tid " << request->tid | |
2359 | << " from mds." << request->mds << dendl; | |
2360 | request->send_to_auth = true; | |
2361 | request->resend_mds = choose_target_mds(request); | |
2362 | Inode *in = request->inode(); | |
11fdf7f2 | 2363 | std::map<mds_rank_t, Cap>::const_iterator it; |
7c673cae FG |
2364 | if (request->resend_mds >= 0 && |
2365 | request->resend_mds == request->mds && | |
2366 | (in == NULL || | |
11fdf7f2 TL |
2367 | (it = in->caps.find(request->resend_mds)) != in->caps.end() || |
2368 | request->sent_on_mseq == it->second.mseq)) { | |
2369 | ldout(cct, 20) << "have to return ESTALE" << dendl; | |
7c673cae | 2370 | } else { |
9f95a23c | 2371 | request->caller_cond->notify_all(); |
7c673cae FG |
2372 | return; |
2373 | } | |
7c673cae FG |
2374 | } |
2375 | ||
11fdf7f2 | 2376 | ceph_assert(!request->reply); |
7c673cae FG |
2377 | request->reply = reply; |
2378 | insert_trace(request, session); | |
2379 | ||
2380 | // Handle unsafe reply | |
2381 | if (!is_safe) { | |
2382 | request->got_unsafe = true; | |
2383 | session->unsafe_requests.push_back(&request->unsafe_item); | |
2384 | if (is_dir_operation(request)) { | |
2385 | Inode *dir = request->inode(); | |
11fdf7f2 | 2386 | ceph_assert(dir); |
7c673cae FG |
2387 | dir->unsafe_ops.push_back(&request->unsafe_dir_item); |
2388 | } | |
2389 | if (request->target) { | |
2390 | InodeRef &in = request->target; | |
2391 | in->unsafe_ops.push_back(&request->unsafe_target_item); | |
2392 | } | |
2393 | } | |
2394 | ||
2395 | // Only signal the caller once (on the first reply): | |
2396 | // Either its an unsafe reply, or its a safe reply and no unsafe reply was sent. | |
2397 | if (!is_safe || !request->got_unsafe) { | |
9f95a23c | 2398 | ceph::condition_variable cond; |
7c673cae FG |
2399 | request->dispatch_cond = &cond; |
2400 | ||
2401 | // wake up waiter | |
11fdf7f2 | 2402 | ldout(cct, 20) << __func__ << " signalling caller " << (void*)request->caller_cond << dendl; |
9f95a23c | 2403 | request->caller_cond->notify_all(); |
7c673cae FG |
2404 | |
2405 | // wake for kick back | |
9f95a23c TL |
2406 | std::unique_lock l{client_lock, std::adopt_lock}; |
2407 | cond.wait(l, [tid, request, &cond, this] { | |
2408 | if (request->dispatch_cond) { | |
2409 | ldout(cct, 20) << "handle_client_reply awaiting kickback on tid " | |
2410 | << tid << " " << &cond << dendl; | |
2411 | } | |
2412 | return !request->dispatch_cond; | |
2413 | }); | |
2414 | l.release(); | |
7c673cae FG |
2415 | } |
2416 | ||
2417 | if (is_safe) { | |
2418 | // the filesystem change is committed to disk | |
2419 | // we're done, clean up | |
2420 | if (request->got_unsafe) { | |
2421 | request->unsafe_item.remove_myself(); | |
2422 | request->unsafe_dir_item.remove_myself(); | |
2423 | request->unsafe_target_item.remove_myself(); | |
2424 | signal_cond_list(request->waitfor_safe); | |
2425 | } | |
2426 | request->item.remove_myself(); | |
2427 | unregister_request(request); | |
2428 | } | |
2429 | if (unmounting) | |
9f95a23c | 2430 | mount_cond.notify_all(); |
7c673cae FG |
2431 | } |
2432 | ||
2433 | void Client::_handle_full_flag(int64_t pool) | |
2434 | { | |
2435 | ldout(cct, 1) << __func__ << ": FULL: cancelling outstanding operations " | |
2436 | << "on " << pool << dendl; | |
2437 | // Cancel all outstanding ops in this pool with -ENOSPC: it is necessary | |
2438 | // to do this rather than blocking, because otherwise when we fill up we | |
2439 | // potentially lock caps forever on files with dirty pages, and we need | |
2440 | // to be able to release those caps to the MDS so that it can delete files | |
2441 | // and free up space. | |
2442 | epoch_t cancelled_epoch = objecter->op_cancel_writes(-ENOSPC, pool); | |
2443 | ||
2444 | // For all inodes with layouts in this pool and a pending flush write op | |
2445 | // (i.e. one of the ones we will cancel), we've got to purge_set their data | |
2446 | // from ObjectCacher so that it doesn't re-issue the write in response to | |
2447 | // the ENOSPC error. | |
2448 | // Fortunately since we're cancelling everything in a given pool, we don't | |
2449 | // need to know which ops belong to which ObjectSet, we can just blow all | |
2450 | // the un-flushed cached data away and mark any dirty inodes' async_err | |
2451 | // field with -ENOSPC as long as we're sure all the ops we cancelled were | |
2452 | // affecting this pool, and all the objectsets we're purging were also | |
2453 | // in this pool. | |
2454 | for (unordered_map<vinodeno_t,Inode*>::iterator i = inode_map.begin(); | |
2455 | i != inode_map.end(); ++i) | |
2456 | { | |
2457 | Inode *inode = i->second; | |
2458 | if (inode->oset.dirty_or_tx | |
2459 | && (pool == -1 || inode->layout.pool_id == pool)) { | |
2460 | ldout(cct, 4) << __func__ << ": FULL: inode 0x" << std::hex << i->first << std::dec | |
2461 | << " has dirty objects, purging and setting ENOSPC" << dendl; | |
2462 | objectcacher->purge_set(&inode->oset); | |
2463 | inode->set_async_err(-ENOSPC); | |
2464 | } | |
2465 | } | |
2466 | ||
2467 | if (cancelled_epoch != (epoch_t)-1) { | |
2468 | set_cap_epoch_barrier(cancelled_epoch); | |
2469 | } | |
2470 | } | |
2471 | ||
11fdf7f2 | 2472 | void Client::handle_osd_map(const MConstRef<MOSDMap>& m) |
7c673cae | 2473 | { |
31f18b77 FG |
2474 | std::set<entity_addr_t> new_blacklists; |
2475 | objecter->consume_blacklist_events(&new_blacklists); | |
2476 | ||
11fdf7f2 TL |
2477 | const auto myaddrs = messenger->get_myaddrs(); |
2478 | bool new_blacklist = false; | |
2479 | bool prenautilus = objecter->with_osdmap( | |
2480 | [&](const OSDMap& o) { | |
9f95a23c | 2481 | return o.require_osd_release < ceph_release_t::nautilus; |
11fdf7f2 TL |
2482 | }); |
2483 | if (!blacklisted) { | |
2484 | for (auto a : myaddrs.v) { | |
2485 | // blacklist entries are always TYPE_ANY for nautilus+ | |
2486 | a.set_type(entity_addr_t::TYPE_ANY); | |
2487 | if (new_blacklists.count(a)) { | |
2488 | new_blacklist = true; | |
2489 | break; | |
2490 | } | |
2491 | if (prenautilus) { | |
2492 | // ...except pre-nautilus, they were TYPE_LEGACY | |
2493 | a.set_type(entity_addr_t::TYPE_LEGACY); | |
2494 | if (new_blacklists.count(a)) { | |
2495 | new_blacklist = true; | |
2496 | break; | |
2497 | } | |
2498 | } | |
2499 | } | |
2500 | } | |
2501 | if (new_blacklist) { | |
31f18b77 FG |
2502 | auto epoch = objecter->with_osdmap([](const OSDMap &o){ |
2503 | return o.get_epoch(); | |
2504 | }); | |
2505 | lderr(cct) << "I was blacklisted at osd epoch " << epoch << dendl; | |
2506 | blacklisted = true; | |
31f18b77 | 2507 | |
11fdf7f2 | 2508 | _abort_mds_sessions(-EBLACKLISTED); |
31f18b77 FG |
2509 | |
2510 | // Since we know all our OSD ops will fail, cancel them all preemtively, | |
2511 | // so that on an unhealthy cluster we can umount promptly even if e.g. | |
2512 | // some PGs were inaccessible. | |
2513 | objecter->op_cancel_writes(-EBLACKLISTED); | |
2514 | ||
2515 | } else if (blacklisted) { | |
2516 | // Handle case where we were blacklisted but no longer are | |
11fdf7f2 TL |
2517 | blacklisted = objecter->with_osdmap([myaddrs](const OSDMap &o){ |
2518 | return o.is_blacklisted(myaddrs);}); | |
31f18b77 FG |
2519 | } |
2520 | ||
f64942e4 AA |
2521 | // Always subscribe to next osdmap for blacklisted client |
2522 | // until this client is not blacklisted. | |
2523 | if (blacklisted) { | |
2524 | objecter->maybe_request_map(); | |
2525 | } | |
2526 | ||
7c673cae FG |
2527 | if (objecter->osdmap_full_flag()) { |
2528 | _handle_full_flag(-1); | |
2529 | } else { | |
2530 | // Accumulate local list of full pools so that I can drop | |
2531 | // the objecter lock before re-entering objecter in | |
2532 | // cancel_writes | |
2533 | std::vector<int64_t> full_pools; | |
2534 | ||
2535 | objecter->with_osdmap([&full_pools](const OSDMap &o) { | |
2536 | for (const auto& kv : o.get_pools()) { | |
2537 | if (kv.second.has_flag(pg_pool_t::FLAG_FULL)) { | |
2538 | full_pools.push_back(kv.first); | |
2539 | } | |
2540 | } | |
2541 | }); | |
2542 | ||
2543 | for (auto p : full_pools) | |
2544 | _handle_full_flag(p); | |
2545 | ||
2546 | // Subscribe to subsequent maps to watch for the full flag going | |
2547 | // away. For the global full flag objecter does this for us, but | |
2548 | // it pays no attention to the per-pool full flag so in this branch | |
2549 | // we do it ourselves. | |
2550 | if (!full_pools.empty()) { | |
2551 | objecter->maybe_request_map(); | |
2552 | } | |
2553 | } | |
7c673cae FG |
2554 | } |
2555 | ||
2556 | ||
2557 | // ------------------------ | |
2558 | // incoming messages | |
2559 | ||
2560 | ||
11fdf7f2 | 2561 | bool Client::ms_dispatch2(const MessageRef &m) |
7c673cae | 2562 | { |
11fdf7f2 | 2563 | std::lock_guard l(client_lock); |
7c673cae FG |
2564 | if (!initialized) { |
2565 | ldout(cct, 10) << "inactive, discarding " << *m << dendl; | |
7c673cae FG |
2566 | return true; |
2567 | } | |
2568 | ||
2569 | switch (m->get_type()) { | |
2570 | // mounting and mds sessions | |
2571 | case CEPH_MSG_MDS_MAP: | |
9f95a23c | 2572 | handle_mds_map(ref_cast<MMDSMap>(m)); |
7c673cae FG |
2573 | break; |
2574 | case CEPH_MSG_FS_MAP: | |
9f95a23c | 2575 | handle_fs_map(ref_cast<MFSMap>(m)); |
7c673cae FG |
2576 | break; |
2577 | case CEPH_MSG_FS_MAP_USER: | |
9f95a23c | 2578 | handle_fs_map_user(ref_cast<MFSMapUser>(m)); |
7c673cae FG |
2579 | break; |
2580 | case CEPH_MSG_CLIENT_SESSION: | |
9f95a23c | 2581 | handle_client_session(ref_cast<MClientSession>(m)); |
7c673cae FG |
2582 | break; |
2583 | ||
2584 | case CEPH_MSG_OSD_MAP: | |
9f95a23c | 2585 | handle_osd_map(ref_cast<MOSDMap>(m)); |
7c673cae FG |
2586 | break; |
2587 | ||
2588 | // requests | |
2589 | case CEPH_MSG_CLIENT_REQUEST_FORWARD: | |
9f95a23c | 2590 | handle_client_request_forward(ref_cast<MClientRequestForward>(m)); |
7c673cae FG |
2591 | break; |
2592 | case CEPH_MSG_CLIENT_REPLY: | |
9f95a23c | 2593 | handle_client_reply(ref_cast<MClientReply>(m)); |
11fdf7f2 TL |
2594 | break; |
2595 | ||
2596 | // reclaim reply | |
2597 | case CEPH_MSG_CLIENT_RECLAIM_REPLY: | |
9f95a23c | 2598 | handle_client_reclaim_reply(ref_cast<MClientReclaimReply>(m)); |
7c673cae FG |
2599 | break; |
2600 | ||
2601 | case CEPH_MSG_CLIENT_SNAP: | |
9f95a23c | 2602 | handle_snap(ref_cast<MClientSnap>(m)); |
7c673cae FG |
2603 | break; |
2604 | case CEPH_MSG_CLIENT_CAPS: | |
9f95a23c | 2605 | handle_caps(ref_cast<MClientCaps>(m)); |
7c673cae FG |
2606 | break; |
2607 | case CEPH_MSG_CLIENT_LEASE: | |
9f95a23c | 2608 | handle_lease(ref_cast<MClientLease>(m)); |
7c673cae FG |
2609 | break; |
2610 | case MSG_COMMAND_REPLY: | |
2611 | if (m->get_source().type() == CEPH_ENTITY_TYPE_MDS) { | |
9f95a23c | 2612 | handle_command_reply(ref_cast<MCommandReply>(m)); |
7c673cae FG |
2613 | } else { |
2614 | return false; | |
2615 | } | |
2616 | break; | |
2617 | case CEPH_MSG_CLIENT_QUOTA: | |
9f95a23c | 2618 | handle_quota(ref_cast<MClientQuota>(m)); |
7c673cae FG |
2619 | break; |
2620 | ||
2621 | default: | |
2622 | return false; | |
2623 | } | |
2624 | ||
2625 | // unmounting? | |
2626 | if (unmounting) { | |
2627 | ldout(cct, 10) << "unmounting: trim pass, size was " << lru.lru_get_size() | |
2628 | << "+" << inode_map.size() << dendl; | |
2629 | long unsigned size = lru.lru_get_size() + inode_map.size(); | |
2630 | trim_cache(); | |
2631 | if (size < lru.lru_get_size() + inode_map.size()) { | |
2632 | ldout(cct, 10) << "unmounting: trim pass, cache shrank, poking unmount()" << dendl; | |
9f95a23c | 2633 | mount_cond.notify_all(); |
7c673cae FG |
2634 | } else { |
2635 | ldout(cct, 10) << "unmounting: trim pass, size still " << lru.lru_get_size() | |
2636 | << "+" << inode_map.size() << dendl; | |
2637 | } | |
2638 | } | |
2639 | ||
2640 | return true; | |
2641 | } | |
2642 | ||
11fdf7f2 | 2643 | void Client::handle_fs_map(const MConstRef<MFSMap>& m) |
7c673cae FG |
2644 | { |
2645 | fsmap.reset(new FSMap(m->get_fsmap())); | |
7c673cae FG |
2646 | |
2647 | signal_cond_list(waiting_for_fsmap); | |
2648 | ||
2649 | monclient->sub_got("fsmap", fsmap->get_epoch()); | |
2650 | } | |
2651 | ||
11fdf7f2 | 2652 | void Client::handle_fs_map_user(const MConstRef<MFSMapUser>& m) |
7c673cae FG |
2653 | { |
2654 | fsmap_user.reset(new FSMapUser); | |
2655 | *fsmap_user = m->get_fsmap(); | |
7c673cae FG |
2656 | |
2657 | monclient->sub_got("fsmap.user", fsmap_user->get_epoch()); | |
2658 | signal_cond_list(waiting_for_fsmap); | |
2659 | } | |
2660 | ||
11fdf7f2 | 2661 | void Client::handle_mds_map(const MConstRef<MMDSMap>& m) |
7c673cae | 2662 | { |
f64942e4 | 2663 | mds_gid_t old_inc, new_inc; |
7c673cae | 2664 | if (m->get_epoch() <= mdsmap->get_epoch()) { |
11fdf7f2 | 2665 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() |
7c673cae FG |
2666 | << " is identical to or older than our " |
2667 | << mdsmap->get_epoch() << dendl; | |
7c673cae | 2668 | return; |
f64942e4 | 2669 | } |
7c673cae | 2670 | |
11fdf7f2 | 2671 | ldout(cct, 1) << __func__ << " epoch " << m->get_epoch() << dendl; |
7c673cae FG |
2672 | |
2673 | std::unique_ptr<MDSMap> oldmap(new MDSMap); | |
2674 | oldmap.swap(mdsmap); | |
2675 | ||
2676 | mdsmap->decode(m->get_encoded()); | |
2677 | ||
2678 | // Cancel any commands for missing or laggy GIDs | |
2679 | std::list<ceph_tid_t> cancel_ops; | |
2680 | auto &commands = command_table.get_commands(); | |
2681 | for (const auto &i : commands) { | |
2682 | auto &op = i.second; | |
2683 | const mds_gid_t op_mds_gid = op.mds_gid; | |
2684 | if (mdsmap->is_dne_gid(op_mds_gid) || mdsmap->is_laggy_gid(op_mds_gid)) { | |
2685 | ldout(cct, 1) << __func__ << ": cancelling command op " << i.first << dendl; | |
2686 | cancel_ops.push_back(i.first); | |
2687 | if (op.outs) { | |
2688 | std::ostringstream ss; | |
2689 | ss << "MDS " << op_mds_gid << " went away"; | |
2690 | *(op.outs) = ss.str(); | |
2691 | } | |
2692 | op.con->mark_down(); | |
2693 | if (op.on_finish) { | |
2694 | op.on_finish->complete(-ETIMEDOUT); | |
2695 | } | |
2696 | } | |
2697 | } | |
2698 | ||
2699 | for (std::list<ceph_tid_t>::iterator i = cancel_ops.begin(); | |
2700 | i != cancel_ops.end(); ++i) { | |
2701 | command_table.erase(*i); | |
2702 | } | |
2703 | ||
2704 | // reset session | |
11fdf7f2 | 2705 | for (auto p = mds_sessions.begin(); p != mds_sessions.end(); ) { |
7c673cae | 2706 | mds_rank_t mds = p->first; |
11fdf7f2 | 2707 | MetaSession *session = &p->second; |
7c673cae FG |
2708 | ++p; |
2709 | ||
2710 | int oldstate = oldmap->get_state(mds); | |
2711 | int newstate = mdsmap->get_state(mds); | |
2712 | if (!mdsmap->is_up(mds)) { | |
2713 | session->con->mark_down(); | |
11fdf7f2 | 2714 | } else if (mdsmap->get_addrs(mds) != session->addrs) { |
f64942e4 AA |
2715 | old_inc = oldmap->get_incarnation(mds); |
2716 | new_inc = mdsmap->get_incarnation(mds); | |
2717 | if (old_inc != new_inc) { | |
2718 | ldout(cct, 1) << "mds incarnation changed from " | |
2719 | << old_inc << " to " << new_inc << dendl; | |
2720 | oldstate = MDSMap::STATE_NULL; | |
2721 | } | |
7c673cae | 2722 | session->con->mark_down(); |
11fdf7f2 | 2723 | session->addrs = mdsmap->get_addrs(mds); |
7c673cae FG |
2724 | // When new MDS starts to take over, notify kernel to trim unused entries |
2725 | // in its dcache/icache. Hopefully, the kernel will release some unused | |
2726 | // inodes before the new MDS enters reconnect state. | |
2727 | trim_cache_for_reconnect(session); | |
2728 | } else if (oldstate == newstate) | |
2729 | continue; // no change | |
2730 | ||
2731 | session->mds_state = newstate; | |
2732 | if (newstate == MDSMap::STATE_RECONNECT) { | |
11fdf7f2 | 2733 | session->con = messenger->connect_to_mds(session->addrs); |
7c673cae | 2734 | send_reconnect(session); |
81eedcae TL |
2735 | } else if (newstate > MDSMap::STATE_RECONNECT) { |
2736 | if (oldstate < MDSMap::STATE_RECONNECT) { | |
2737 | ldout(cct, 1) << "we may miss the MDSMap::RECONNECT, close mds session ... " << dendl; | |
2738 | _closed_mds_session(session); | |
2739 | continue; | |
2740 | } | |
2741 | if (newstate >= MDSMap::STATE_ACTIVE) { | |
2742 | if (oldstate < MDSMap::STATE_ACTIVE) { | |
2743 | // kick new requests | |
2744 | kick_requests(session); | |
2745 | kick_flushing_caps(session); | |
2746 | signal_context_list(session->waiting_for_open); | |
2747 | wake_up_session_caps(session, true); | |
2748 | } | |
2749 | connect_mds_targets(mds); | |
7c673cae | 2750 | } |
7c673cae FG |
2751 | } else if (newstate == MDSMap::STATE_NULL && |
2752 | mds >= mdsmap->get_max_mds()) { | |
2753 | _closed_mds_session(session); | |
2754 | } | |
2755 | } | |
2756 | ||
2757 | // kick any waiting threads | |
2758 | signal_cond_list(waiting_for_mdsmap); | |
2759 | ||
7c673cae FG |
2760 | monclient->sub_got("mdsmap", mdsmap->get_epoch()); |
2761 | } | |
2762 | ||
2763 | void Client::send_reconnect(MetaSession *session) | |
2764 | { | |
2765 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 2766 | ldout(cct, 10) << __func__ << " to mds." << mds << dendl; |
7c673cae FG |
2767 | |
2768 | // trim unused caps to reduce MDS's cache rejoin time | |
2769 | trim_cache_for_reconnect(session); | |
2770 | ||
2771 | session->readonly = false; | |
2772 | ||
11fdf7f2 | 2773 | session->release.reset(); |
7c673cae FG |
2774 | |
2775 | // reset my cap seq number | |
2776 | session->seq = 0; | |
2777 | //connect to the mds' offload targets | |
2778 | connect_mds_targets(mds); | |
2779 | //make sure unsafe requests get saved | |
2780 | resend_unsafe_requests(session); | |
2781 | ||
11fdf7f2 TL |
2782 | early_kick_flushing_caps(session); |
2783 | ||
9f95a23c | 2784 | auto m = make_message<MClientReconnect>(); |
11fdf7f2 | 2785 | bool allow_multi = session->mds_features.test(CEPHFS_FEATURE_MULTI_RECONNECT); |
7c673cae FG |
2786 | |
2787 | // i have an open session. | |
2788 | ceph::unordered_set<inodeno_t> did_snaprealm; | |
2789 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator p = inode_map.begin(); | |
2790 | p != inode_map.end(); | |
2791 | ++p) { | |
2792 | Inode *in = p->second; | |
11fdf7f2 TL |
2793 | auto it = in->caps.find(mds); |
2794 | if (it != in->caps.end()) { | |
2795 | if (allow_multi && | |
9f95a23c TL |
2796 | m->get_approx_size() >= |
2797 | static_cast<size_t>((std::numeric_limits<int>::max() >> 1))) { | |
11fdf7f2 TL |
2798 | m->mark_more(); |
2799 | session->con->send_message2(std::move(m)); | |
2800 | ||
9f95a23c | 2801 | m = make_message<MClientReconnect>(); |
11fdf7f2 TL |
2802 | } |
2803 | ||
2804 | Cap &cap = it->second; | |
7c673cae | 2805 | ldout(cct, 10) << " caps on " << p->first |
11fdf7f2 | 2806 | << " " << ccap_string(cap.issued) |
7c673cae FG |
2807 | << " wants " << ccap_string(in->caps_wanted()) |
2808 | << dendl; | |
2809 | filepath path; | |
2810 | in->make_long_path(path); | |
2811 | ldout(cct, 10) << " path " << path << dendl; | |
2812 | ||
2813 | bufferlist flockbl; | |
2814 | _encode_filelocks(in, flockbl); | |
2815 | ||
11fdf7f2 TL |
2816 | cap.seq = 0; // reset seq. |
2817 | cap.issue_seq = 0; // reset seq. | |
2818 | cap.mseq = 0; // reset seq. | |
2819 | // cap gen should catch up with session cap_gen | |
2820 | if (cap.gen < session->cap_gen) { | |
2821 | cap.gen = session->cap_gen; | |
2822 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
2823 | } else { | |
2824 | cap.issued = cap.implemented; | |
2825 | } | |
7c673cae FG |
2826 | snapid_t snap_follows = 0; |
2827 | if (!in->cap_snaps.empty()) | |
2828 | snap_follows = in->cap_snaps.begin()->first; | |
2829 | ||
2830 | m->add_cap(p->first.ino, | |
11fdf7f2 | 2831 | cap.cap_id, |
7c673cae FG |
2832 | path.get_ino(), path.get_path(), // ino |
2833 | in->caps_wanted(), // wanted | |
11fdf7f2 | 2834 | cap.issued, // issued |
7c673cae FG |
2835 | in->snaprealm->ino, |
2836 | snap_follows, | |
2837 | flockbl); | |
2838 | ||
2839 | if (did_snaprealm.count(in->snaprealm->ino) == 0) { | |
2840 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
2841 | m->add_snaprealm(in->snaprealm->ino, in->snaprealm->seq, in->snaprealm->parent); | |
2842 | did_snaprealm.insert(in->snaprealm->ino); | |
2843 | } | |
2844 | } | |
2845 | } | |
2846 | ||
11fdf7f2 TL |
2847 | if (!allow_multi) |
2848 | m->set_encoding_version(0); // use connection features to choose encoding | |
2849 | session->con->send_message2(std::move(m)); | |
7c673cae | 2850 | |
9f95a23c | 2851 | mount_cond.notify_all(); |
11fdf7f2 TL |
2852 | |
2853 | if (session->reclaim_state == MetaSession::RECLAIMING) | |
2854 | signal_cond_list(waiting_for_reclaim); | |
7c673cae FG |
2855 | } |
2856 | ||
2857 | ||
2858 | void Client::kick_requests(MetaSession *session) | |
2859 | { | |
11fdf7f2 | 2860 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
2861 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
2862 | p != mds_requests.end(); | |
2863 | ++p) { | |
31f18b77 FG |
2864 | MetaRequest *req = p->second; |
2865 | if (req->got_unsafe) | |
2866 | continue; | |
2867 | if (req->aborted()) { | |
2868 | if (req->caller_cond) { | |
2869 | req->kick = true; | |
9f95a23c | 2870 | req->caller_cond->notify_all(); |
31f18b77 | 2871 | } |
7c673cae | 2872 | continue; |
31f18b77 FG |
2873 | } |
2874 | if (req->retry_attempt > 0) | |
7c673cae | 2875 | continue; // new requests only |
31f18b77 | 2876 | if (req->mds == session->mds_num) { |
7c673cae FG |
2877 | send_request(p->second, session); |
2878 | } | |
2879 | } | |
2880 | } | |
2881 | ||
2882 | void Client::resend_unsafe_requests(MetaSession *session) | |
2883 | { | |
2884 | for (xlist<MetaRequest*>::iterator iter = session->unsafe_requests.begin(); | |
2885 | !iter.end(); | |
2886 | ++iter) | |
2887 | send_request(*iter, session); | |
2888 | ||
2889 | // also re-send old requests when MDS enters reconnect stage. So that MDS can | |
2890 | // process completed requests in clientreplay stage. | |
2891 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); | |
2892 | p != mds_requests.end(); | |
2893 | ++p) { | |
2894 | MetaRequest *req = p->second; | |
2895 | if (req->got_unsafe) | |
2896 | continue; | |
31f18b77 FG |
2897 | if (req->aborted()) |
2898 | continue; | |
7c673cae FG |
2899 | if (req->retry_attempt == 0) |
2900 | continue; // old requests only | |
2901 | if (req->mds == session->mds_num) | |
2902 | send_request(req, session, true); | |
2903 | } | |
2904 | } | |
2905 | ||
2906 | void Client::wait_unsafe_requests() | |
2907 | { | |
2908 | list<MetaRequest*> last_unsafe_reqs; | |
11fdf7f2 TL |
2909 | for (const auto &p : mds_sessions) { |
2910 | const MetaSession &s = p.second; | |
2911 | if (!s.unsafe_requests.empty()) { | |
2912 | MetaRequest *req = s.unsafe_requests.back(); | |
7c673cae FG |
2913 | req->get(); |
2914 | last_unsafe_reqs.push_back(req); | |
2915 | } | |
2916 | } | |
2917 | ||
2918 | for (list<MetaRequest*>::iterator p = last_unsafe_reqs.begin(); | |
2919 | p != last_unsafe_reqs.end(); | |
2920 | ++p) { | |
2921 | MetaRequest *req = *p; | |
2922 | if (req->unsafe_item.is_on_list()) | |
2923 | wait_on_list(req->waitfor_safe); | |
2924 | put_request(req); | |
2925 | } | |
2926 | } | |
2927 | ||
2928 | void Client::kick_requests_closed(MetaSession *session) | |
2929 | { | |
11fdf7f2 | 2930 | ldout(cct, 10) << __func__ << " for mds." << session->mds_num << dendl; |
7c673cae FG |
2931 | for (map<ceph_tid_t, MetaRequest*>::iterator p = mds_requests.begin(); |
2932 | p != mds_requests.end(); ) { | |
2933 | MetaRequest *req = p->second; | |
2934 | ++p; | |
2935 | if (req->mds == session->mds_num) { | |
2936 | if (req->caller_cond) { | |
2937 | req->kick = true; | |
9f95a23c | 2938 | req->caller_cond->notify_all(); |
7c673cae FG |
2939 | } |
2940 | req->item.remove_myself(); | |
2941 | if (req->got_unsafe) { | |
11fdf7f2 | 2942 | lderr(cct) << __func__ << " removing unsafe request " << req->get_tid() << dendl; |
7c673cae | 2943 | req->unsafe_item.remove_myself(); |
eafe8130 TL |
2944 | if (is_dir_operation(req)) { |
2945 | Inode *dir = req->inode(); | |
2946 | assert(dir); | |
2947 | dir->set_async_err(-EIO); | |
2948 | lderr(cct) << "kick_requests_closed drop req of inode(dir) : " | |
2949 | << dir->ino << " " << req->get_tid() << dendl; | |
2950 | req->unsafe_dir_item.remove_myself(); | |
2951 | } | |
2952 | if (req->target) { | |
2953 | InodeRef &in = req->target; | |
2954 | in->set_async_err(-EIO); | |
2955 | lderr(cct) << "kick_requests_closed drop req of inode : " | |
2956 | << in->ino << " " << req->get_tid() << dendl; | |
2957 | req->unsafe_target_item.remove_myself(); | |
2958 | } | |
7c673cae FG |
2959 | signal_cond_list(req->waitfor_safe); |
2960 | unregister_request(req); | |
2961 | } | |
2962 | } | |
2963 | } | |
11fdf7f2 TL |
2964 | ceph_assert(session->requests.empty()); |
2965 | ceph_assert(session->unsafe_requests.empty()); | |
7c673cae FG |
2966 | } |
2967 | ||
2968 | ||
2969 | ||
2970 | ||
2971 | /************ | |
2972 | * leases | |
2973 | */ | |
2974 | ||
2975 | void Client::got_mds_push(MetaSession *s) | |
2976 | { | |
2977 | s->seq++; | |
2978 | ldout(cct, 10) << " mds." << s->mds_num << " seq now " << s->seq << dendl; | |
2979 | if (s->state == MetaSession::STATE_CLOSING) { | |
9f95a23c | 2980 | s->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_CLOSE, s->seq)); |
7c673cae FG |
2981 | } |
2982 | } | |
2983 | ||
11fdf7f2 | 2984 | void Client::handle_lease(const MConstRef<MClientLease>& m) |
7c673cae | 2985 | { |
11fdf7f2 | 2986 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae | 2987 | |
11fdf7f2 | 2988 | ceph_assert(m->get_action() == CEPH_MDS_LEASE_REVOKE); |
7c673cae FG |
2989 | |
2990 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
2991 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
2992 | if (!session) { | |
7c673cae FG |
2993 | return; |
2994 | } | |
2995 | ||
2996 | got_mds_push(session); | |
2997 | ||
2998 | ceph_seq_t seq = m->get_seq(); | |
2999 | ||
3000 | Inode *in; | |
3001 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); | |
3002 | if (inode_map.count(vino) == 0) { | |
3003 | ldout(cct, 10) << " don't have vino " << vino << dendl; | |
3004 | goto revoke; | |
3005 | } | |
3006 | in = inode_map[vino]; | |
3007 | ||
9f95a23c | 3008 | if (m->get_mask() & CEPH_LEASE_VALID) { |
7c673cae FG |
3009 | if (!in->dir || in->dir->dentries.count(m->dname) == 0) { |
3010 | ldout(cct, 10) << " don't have dir|dentry " << m->get_ino() << "/" << m->dname <<dendl; | |
3011 | goto revoke; | |
3012 | } | |
3013 | Dentry *dn = in->dir->dentries[m->dname]; | |
3014 | ldout(cct, 10) << " revoked DN lease on " << dn << dendl; | |
3015 | dn->lease_mds = -1; | |
3016 | } | |
3017 | ||
3018 | revoke: | |
11fdf7f2 | 3019 | { |
9f95a23c TL |
3020 | auto reply = make_message<MClientLease>(CEPH_MDS_LEASE_RELEASE, seq, |
3021 | m->get_mask(), m->get_ino(), | |
3022 | m->get_first(), m->get_last(), m->dname); | |
11fdf7f2 TL |
3023 | m->get_connection()->send_message2(std::move(reply)); |
3024 | } | |
7c673cae FG |
3025 | } |
3026 | ||
3027 | void Client::put_inode(Inode *in, int n) | |
3028 | { | |
11fdf7f2 | 3029 | ldout(cct, 10) << __func__ << " on " << *in << dendl; |
7c673cae FG |
3030 | int left = in->_put(n); |
3031 | if (left == 0) { | |
3032 | // release any caps | |
3033 | remove_all_caps(in); | |
3034 | ||
11fdf7f2 | 3035 | ldout(cct, 10) << __func__ << " deleting " << *in << dendl; |
7c673cae | 3036 | bool unclean = objectcacher->release_set(&in->oset); |
11fdf7f2 | 3037 | ceph_assert(!unclean); |
7c673cae FG |
3038 | inode_map.erase(in->vino()); |
3039 | if (use_faked_inos()) | |
3040 | _release_faked_ino(in); | |
3041 | ||
3042 | if (in == root) { | |
3043 | root = 0; | |
3044 | root_ancestor = 0; | |
3045 | while (!root_parents.empty()) | |
3046 | root_parents.erase(root_parents.begin()); | |
3047 | } | |
3048 | ||
3049 | delete in; | |
3050 | } | |
3051 | } | |
3052 | ||
3053 | void Client::close_dir(Dir *dir) | |
3054 | { | |
3055 | Inode *in = dir->parent_inode; | |
11fdf7f2 TL |
3056 | ldout(cct, 15) << __func__ << " dir " << dir << " on " << in << dendl; |
3057 | ceph_assert(dir->is_empty()); | |
3058 | ceph_assert(in->dir == dir); | |
3059 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked | |
3060 | if (!in->dentries.empty()) | |
7c673cae FG |
3061 | in->get_first_parent()->put(); // unpin dentry |
3062 | ||
3063 | delete in->dir; | |
3064 | in->dir = 0; | |
3065 | put_inode(in); // unpin inode | |
3066 | } | |
3067 | ||
3068 | /** | |
3069 | * Don't call this with in==NULL, use get_or_create for that | |
3070 | * leave dn set to default NULL unless you're trying to add | |
3071 | * a new inode to a pre-created Dentry | |
3072 | */ | |
3073 | Dentry* Client::link(Dir *dir, const string& name, Inode *in, Dentry *dn) | |
3074 | { | |
3075 | if (!dn) { | |
3076 | // create a new Dentry | |
11fdf7f2 TL |
3077 | dn = new Dentry(dir, name); |
3078 | ||
7c673cae FG |
3079 | lru.lru_insert_mid(dn); // mid or top? |
3080 | ||
3081 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in | |
3082 | << " dn " << dn << " (new dn)" << dendl; | |
3083 | } else { | |
11fdf7f2 | 3084 | ceph_assert(!dn->inode); |
7c673cae FG |
3085 | ldout(cct, 15) << "link dir " << dir->parent_inode << " '" << name << "' to inode " << in |
3086 | << " dn " << dn << " (old dn)" << dendl; | |
3087 | } | |
3088 | ||
3089 | if (in) { // link to inode | |
11fdf7f2 | 3090 | InodeRef tmp_ref; |
7c673cae | 3091 | // only one parent for directories! |
11fdf7f2 TL |
3092 | if (in->is_dir() && !in->dentries.empty()) { |
3093 | tmp_ref = in; // prevent unlink below from freeing the inode. | |
7c673cae | 3094 | Dentry *olddn = in->get_first_parent(); |
11fdf7f2 | 3095 | ceph_assert(olddn->dir != dir || olddn->name != name); |
7c673cae FG |
3096 | Inode *old_diri = olddn->dir->parent_inode; |
3097 | old_diri->dir_release_count++; | |
3098 | clear_dir_complete_and_ordered(old_diri, true); | |
3099 | unlink(olddn, true, true); // keep dir, dentry | |
3100 | } | |
3101 | ||
11fdf7f2 TL |
3102 | dn->link(in); |
3103 | ldout(cct, 20) << "link inode " << in << " parents now " << in->dentries << dendl; | |
7c673cae FG |
3104 | } |
3105 | ||
3106 | return dn; | |
3107 | } | |
3108 | ||
3109 | void Client::unlink(Dentry *dn, bool keepdir, bool keepdentry) | |
3110 | { | |
11fdf7f2 | 3111 | InodeRef in(dn->inode); |
7c673cae FG |
3112 | ldout(cct, 15) << "unlink dir " << dn->dir->parent_inode << " '" << dn->name << "' dn " << dn |
3113 | << " inode " << dn->inode << dendl; | |
3114 | ||
3115 | // unlink from inode | |
11fdf7f2 TL |
3116 | if (dn->inode) { |
3117 | dn->unlink(); | |
3118 | ldout(cct, 20) << "unlink inode " << in << " parents now " << in->dentries << dendl; | |
7c673cae FG |
3119 | } |
3120 | ||
3121 | if (keepdentry) { | |
3122 | dn->lease_mds = -1; | |
3123 | } else { | |
3124 | ldout(cct, 15) << "unlink removing '" << dn->name << "' dn " << dn << dendl; | |
3125 | ||
3126 | // unlink from dir | |
11fdf7f2 TL |
3127 | Dir *dir = dn->dir; |
3128 | dn->detach(); | |
7c673cae FG |
3129 | |
3130 | // delete den | |
3131 | lru.lru_remove(dn); | |
3132 | dn->put(); | |
11fdf7f2 TL |
3133 | |
3134 | if (dir->is_empty() && !keepdir) | |
3135 | close_dir(dir); | |
7c673cae FG |
3136 | } |
3137 | } | |
3138 | ||
3139 | /** | |
3140 | * For asynchronous flushes, check for errors from the IO and | |
3141 | * update the inode if necessary | |
3142 | */ | |
3143 | class C_Client_FlushComplete : public Context { | |
3144 | private: | |
3145 | Client *client; | |
3146 | InodeRef inode; | |
3147 | public: | |
3148 | C_Client_FlushComplete(Client *c, Inode *in) : client(c), inode(in) { } | |
3149 | void finish(int r) override { | |
9f95a23c | 3150 | ceph_assert(ceph_mutex_is_locked_by_me(client->client_lock)); |
7c673cae FG |
3151 | if (r != 0) { |
3152 | client_t const whoami = client->whoami; // For the benefit of ldout prefix | |
3153 | ldout(client->cct, 1) << "I/O error from flush on inode " << inode | |
3154 | << " 0x" << std::hex << inode->ino << std::dec | |
3155 | << ": " << r << "(" << cpp_strerror(r) << ")" << dendl; | |
3156 | inode->set_async_err(r); | |
3157 | } | |
3158 | } | |
3159 | }; | |
3160 | ||
3161 | ||
3162 | /**** | |
3163 | * caps | |
3164 | */ | |
3165 | ||
3166 | void Client::get_cap_ref(Inode *in, int cap) | |
3167 | { | |
3168 | if ((cap & CEPH_CAP_FILE_BUFFER) && | |
3169 | in->cap_refs[CEPH_CAP_FILE_BUFFER] == 0) { | |
11fdf7f2 | 3170 | ldout(cct, 5) << __func__ << " got first FILE_BUFFER ref on " << *in << dendl; |
7c673cae FG |
3171 | in->get(); |
3172 | } | |
3173 | if ((cap & CEPH_CAP_FILE_CACHE) && | |
3174 | in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
11fdf7f2 | 3175 | ldout(cct, 5) << __func__ << " got first FILE_CACHE ref on " << *in << dendl; |
7c673cae FG |
3176 | in->get(); |
3177 | } | |
3178 | in->get_cap_ref(cap); | |
3179 | } | |
3180 | ||
3181 | void Client::put_cap_ref(Inode *in, int cap) | |
3182 | { | |
3183 | int last = in->put_cap_ref(cap); | |
3184 | if (last) { | |
3185 | int put_nref = 0; | |
3186 | int drop = last & ~in->caps_issued(); | |
3187 | if (in->snapid == CEPH_NOSNAP) { | |
3188 | if ((last & CEPH_CAP_FILE_WR) && | |
3189 | !in->cap_snaps.empty() && | |
3190 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3191 | ldout(cct, 10) << __func__ << " finishing pending cap_snap on " << *in << dendl; |
7c673cae FG |
3192 | in->cap_snaps.rbegin()->second.writing = 0; |
3193 | finish_cap_snap(in, in->cap_snaps.rbegin()->second, get_caps_used(in)); | |
3194 | signal_cond_list(in->waitfor_caps); // wake up blocked sync writers | |
3195 | } | |
3196 | if (last & CEPH_CAP_FILE_BUFFER) { | |
3197 | for (auto &p : in->cap_snaps) | |
3198 | p.second.dirty_data = 0; | |
3199 | signal_cond_list(in->waitfor_commit); | |
11fdf7f2 | 3200 | ldout(cct, 5) << __func__ << " dropped last FILE_BUFFER ref on " << *in << dendl; |
7c673cae FG |
3201 | ++put_nref; |
3202 | } | |
3203 | } | |
3204 | if (last & CEPH_CAP_FILE_CACHE) { | |
11fdf7f2 | 3205 | ldout(cct, 5) << __func__ << " dropped last FILE_CACHE ref on " << *in << dendl; |
7c673cae FG |
3206 | ++put_nref; |
3207 | } | |
3208 | if (drop) | |
3209 | check_caps(in, 0); | |
3210 | if (put_nref) | |
3211 | put_inode(in, put_nref); | |
3212 | } | |
3213 | } | |
3214 | ||
3215 | int Client::get_caps(Inode *in, int need, int want, int *phave, loff_t endoff) | |
3216 | { | |
3217 | int r = check_pool_perm(in, need); | |
3218 | if (r < 0) | |
3219 | return r; | |
3220 | ||
3221 | while (1) { | |
3222 | int file_wanted = in->caps_file_wanted(); | |
3223 | if ((file_wanted & need) != need) { | |
3224 | ldout(cct, 10) << "get_caps " << *in << " need " << ccap_string(need) | |
3225 | << " file_wanted " << ccap_string(file_wanted) << ", EBADF " | |
3226 | << dendl; | |
3227 | return -EBADF; | |
3228 | } | |
3229 | ||
3230 | int implemented; | |
3231 | int have = in->caps_issued(&implemented); | |
3232 | ||
3233 | bool waitfor_caps = false; | |
3234 | bool waitfor_commit = false; | |
3235 | ||
3236 | if (have & need & CEPH_CAP_FILE_WR) { | |
3237 | if (endoff > 0 && | |
3238 | (endoff >= (loff_t)in->max_size || | |
3239 | endoff > (loff_t)(in->size << 1)) && | |
3240 | endoff > (loff_t)in->wanted_max_size) { | |
3241 | ldout(cct, 10) << "wanted_max_size " << in->wanted_max_size << " -> " << endoff << dendl; | |
3242 | in->wanted_max_size = endoff; | |
3243 | check_caps(in, 0); | |
3244 | } | |
3245 | ||
3246 | if (endoff >= 0 && endoff > (loff_t)in->max_size) { | |
3247 | ldout(cct, 10) << "waiting on max_size, endoff " << endoff << " max_size " << in->max_size << " on " << *in << dendl; | |
3248 | waitfor_caps = true; | |
3249 | } | |
3250 | if (!in->cap_snaps.empty()) { | |
3251 | if (in->cap_snaps.rbegin()->second.writing) { | |
3252 | ldout(cct, 10) << "waiting on cap_snap write to complete" << dendl; | |
3253 | waitfor_caps = true; | |
3254 | } | |
3255 | for (auto &p : in->cap_snaps) { | |
3256 | if (p.second.dirty_data) { | |
3257 | waitfor_commit = true; | |
3258 | break; | |
3259 | } | |
3260 | } | |
3261 | if (waitfor_commit) { | |
3262 | _flush(in, new C_Client_FlushComplete(this, in)); | |
3263 | ldout(cct, 10) << "waiting for WRBUFFER to get dropped" << dendl; | |
3264 | } | |
3265 | } | |
3266 | } | |
3267 | ||
3268 | if (!waitfor_caps && !waitfor_commit) { | |
3269 | if ((have & need) == need) { | |
7c673cae FG |
3270 | int revoking = implemented & ~have; |
3271 | ldout(cct, 10) << "get_caps " << *in << " have " << ccap_string(have) | |
3272 | << " need " << ccap_string(need) << " want " << ccap_string(want) | |
c07f9fc5 | 3273 | << " revoking " << ccap_string(revoking) |
7c673cae | 3274 | << dendl; |
c07f9fc5 | 3275 | if ((revoking & want) == 0) { |
7c673cae FG |
3276 | *phave = need | (have & want); |
3277 | in->get_cap_ref(need); | |
3278 | return 0; | |
3279 | } | |
3280 | } | |
3281 | ldout(cct, 10) << "waiting for caps " << *in << " need " << ccap_string(need) << " want " << ccap_string(want) << dendl; | |
3282 | waitfor_caps = true; | |
3283 | } | |
3284 | ||
3285 | if ((need & CEPH_CAP_FILE_WR) && in->auth_cap && | |
3286 | in->auth_cap->session->readonly) | |
3287 | return -EROFS; | |
3288 | ||
3289 | if (in->flags & I_CAP_DROPPED) { | |
3290 | int mds_wanted = in->caps_mds_wanted(); | |
3291 | if ((mds_wanted & need) != need) { | |
3292 | int ret = _renew_caps(in); | |
3293 | if (ret < 0) | |
3294 | return ret; | |
3295 | continue; | |
3296 | } | |
a8e16298 | 3297 | if (!(file_wanted & ~mds_wanted)) |
7c673cae | 3298 | in->flags &= ~I_CAP_DROPPED; |
7c673cae FG |
3299 | } |
3300 | ||
3301 | if (waitfor_caps) | |
3302 | wait_on_list(in->waitfor_caps); | |
3303 | else if (waitfor_commit) | |
3304 | wait_on_list(in->waitfor_commit); | |
3305 | } | |
3306 | } | |
3307 | ||
3308 | int Client::get_caps_used(Inode *in) | |
3309 | { | |
3310 | unsigned used = in->caps_used(); | |
3311 | if (!(used & CEPH_CAP_FILE_CACHE) && | |
3312 | !objectcacher->set_is_empty(&in->oset)) | |
3313 | used |= CEPH_CAP_FILE_CACHE; | |
3314 | return used; | |
3315 | } | |
3316 | ||
3317 | void Client::cap_delay_requeue(Inode *in) | |
3318 | { | |
11fdf7f2 | 3319 | ldout(cct, 10) << __func__ << " on " << *in << dendl; |
7c673cae FG |
3320 | in->hold_caps_until = ceph_clock_now(); |
3321 | in->hold_caps_until += cct->_conf->client_caps_release_delay; | |
28e407b8 | 3322 | delayed_list.push_back(&in->delay_cap_item); |
7c673cae FG |
3323 | } |
3324 | ||
3325 | void Client::send_cap(Inode *in, MetaSession *session, Cap *cap, | |
eafe8130 | 3326 | int flags, int used, int want, int retain, |
7c673cae FG |
3327 | int flush, ceph_tid_t flush_tid) |
3328 | { | |
3329 | int held = cap->issued | cap->implemented; | |
3330 | int revoking = cap->implemented & ~cap->issued; | |
3331 | retain &= ~revoking; | |
3332 | int dropping = cap->issued & ~retain; | |
3333 | int op = CEPH_CAP_OP_UPDATE; | |
3334 | ||
11fdf7f2 | 3335 | ldout(cct, 10) << __func__ << " " << *in |
7c673cae | 3336 | << " mds." << session->mds_num << " seq " << cap->seq |
7c673cae FG |
3337 | << " used " << ccap_string(used) |
3338 | << " want " << ccap_string(want) | |
3339 | << " flush " << ccap_string(flush) | |
3340 | << " retain " << ccap_string(retain) | |
3341 | << " held "<< ccap_string(held) | |
3342 | << " revoking " << ccap_string(revoking) | |
3343 | << " dropping " << ccap_string(dropping) | |
3344 | << dendl; | |
3345 | ||
3346 | if (cct->_conf->client_inject_release_failure && revoking) { | |
3347 | const int would_have_issued = cap->issued & retain; | |
3348 | const int would_have_implemented = cap->implemented & (cap->issued | used); | |
3349 | // Simulated bug: | |
3350 | // - tell the server we think issued is whatever they issued plus whatever we implemented | |
3351 | // - leave what we have implemented in place | |
3352 | ldout(cct, 20) << __func__ << " injecting failure to release caps" << dendl; | |
3353 | cap->issued = cap->issued | cap->implemented; | |
3354 | ||
3355 | // Make an exception for revoking xattr caps: we are injecting | |
3356 | // failure to release other caps, but allow xattr because client | |
3357 | // will block on xattr ops if it can't release these to MDS (#9800) | |
3358 | const int xattr_mask = CEPH_CAP_XATTR_SHARED | CEPH_CAP_XATTR_EXCL; | |
3359 | cap->issued ^= xattr_mask & revoking; | |
3360 | cap->implemented ^= xattr_mask & revoking; | |
3361 | ||
3362 | ldout(cct, 20) << __func__ << " issued " << ccap_string(cap->issued) << " vs " << ccap_string(would_have_issued) << dendl; | |
3363 | ldout(cct, 20) << __func__ << " implemented " << ccap_string(cap->implemented) << " vs " << ccap_string(would_have_implemented) << dendl; | |
3364 | } else { | |
3365 | // Normal behaviour | |
3366 | cap->issued &= retain; | |
3367 | cap->implemented &= cap->issued | used; | |
3368 | } | |
3369 | ||
3370 | snapid_t follows = 0; | |
3371 | ||
3372 | if (flush) | |
3373 | follows = in->snaprealm->get_snap_context().seq; | |
3374 | ||
9f95a23c | 3375 | auto m = make_message<MClientCaps>(op, |
7c673cae FG |
3376 | in->ino, |
3377 | 0, | |
3378 | cap->cap_id, cap->seq, | |
3379 | cap->implemented, | |
3380 | want, | |
3381 | flush, | |
3382 | cap->mseq, | |
3383 | cap_epoch_barrier); | |
3384 | m->caller_uid = in->cap_dirtier_uid; | |
3385 | m->caller_gid = in->cap_dirtier_gid; | |
3386 | ||
3387 | m->head.issue_seq = cap->issue_seq; | |
3388 | m->set_tid(flush_tid); | |
3389 | ||
3390 | m->head.uid = in->uid; | |
3391 | m->head.gid = in->gid; | |
3392 | m->head.mode = in->mode; | |
3393 | ||
3394 | m->head.nlink = in->nlink; | |
3395 | ||
3396 | if (flush & CEPH_CAP_XATTR_EXCL) { | |
11fdf7f2 | 3397 | encode(in->xattrs, m->xattrbl); |
7c673cae FG |
3398 | m->head.xattr_version = in->xattr_version; |
3399 | } | |
3400 | ||
3401 | m->size = in->size; | |
3402 | m->max_size = in->max_size; | |
3403 | m->truncate_seq = in->truncate_seq; | |
3404 | m->truncate_size = in->truncate_size; | |
3405 | m->mtime = in->mtime; | |
3406 | m->atime = in->atime; | |
3407 | m->ctime = in->ctime; | |
3408 | m->btime = in->btime; | |
3409 | m->time_warp_seq = in->time_warp_seq; | |
3410 | m->change_attr = in->change_attr; | |
eafe8130 TL |
3411 | |
3412 | if (!(flags & MClientCaps::FLAG_PENDING_CAPSNAP) && | |
3413 | !in->cap_snaps.empty() && | |
3414 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3415 | flags |= MClientCaps::FLAG_PENDING_CAPSNAP; | |
3416 | m->flags = flags; | |
3417 | ||
7c673cae FG |
3418 | if (flush & CEPH_CAP_FILE_WR) { |
3419 | m->inline_version = in->inline_version; | |
3420 | m->inline_data = in->inline_data; | |
3421 | } | |
3422 | ||
3423 | in->reported_size = in->size; | |
3424 | m->set_snap_follows(follows); | |
3425 | cap->wanted = want; | |
3426 | if (cap == in->auth_cap) { | |
3427 | m->set_max_size(in->wanted_max_size); | |
3428 | in->requested_max_size = in->wanted_max_size; | |
3429 | ldout(cct, 15) << "auth cap, setting max_size = " << in->requested_max_size << dendl; | |
3430 | } | |
3431 | ||
3432 | if (!session->flushing_caps_tids.empty()) | |
3433 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3434 | ||
11fdf7f2 | 3435 | session->con->send_message2(std::move(m)); |
7c673cae FG |
3436 | } |
3437 | ||
31f18b77 FG |
3438 | static bool is_max_size_approaching(Inode *in) |
3439 | { | |
3440 | /* mds will adjust max size according to the reported size */ | |
3441 | if (in->flushing_caps & CEPH_CAP_FILE_WR) | |
3442 | return false; | |
3443 | if (in->size >= in->max_size) | |
3444 | return true; | |
3445 | /* half of previous max_size increment has been used */ | |
3446 | if (in->max_size > in->reported_size && | |
3447 | (in->size << 1) >= in->max_size + in->reported_size) | |
3448 | return true; | |
3449 | return false; | |
3450 | } | |
7c673cae | 3451 | |
11fdf7f2 TL |
3452 | static int adjust_caps_used_for_lazyio(int used, int issued, int implemented) |
3453 | { | |
3454 | if (!(used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER))) | |
3455 | return used; | |
3456 | if (!(implemented & CEPH_CAP_FILE_LAZYIO)) | |
3457 | return used; | |
3458 | ||
3459 | if (issued & CEPH_CAP_FILE_LAZYIO) { | |
3460 | if (!(issued & CEPH_CAP_FILE_CACHE)) { | |
3461 | used &= ~CEPH_CAP_FILE_CACHE; | |
3462 | used |= CEPH_CAP_FILE_LAZYIO; | |
3463 | } | |
3464 | if (!(issued & CEPH_CAP_FILE_BUFFER)) { | |
3465 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3466 | used |= CEPH_CAP_FILE_LAZYIO; | |
3467 | } | |
3468 | } else { | |
3469 | if (!(implemented & CEPH_CAP_FILE_CACHE)) { | |
3470 | used &= ~CEPH_CAP_FILE_CACHE; | |
3471 | used |= CEPH_CAP_FILE_LAZYIO; | |
3472 | } | |
3473 | if (!(implemented & CEPH_CAP_FILE_BUFFER)) { | |
3474 | used &= ~CEPH_CAP_FILE_BUFFER; | |
3475 | used |= CEPH_CAP_FILE_LAZYIO; | |
3476 | } | |
3477 | } | |
3478 | return used; | |
3479 | } | |
3480 | ||
7c673cae FG |
3481 | /** |
3482 | * check_caps | |
3483 | * | |
3484 | * Examine currently used and wanted versus held caps. Release, flush or ack | |
3485 | * revoked caps to the MDS as appropriate. | |
3486 | * | |
3487 | * @param in the inode to check | |
3488 | * @param flags flags to apply to cap check | |
3489 | */ | |
3490 | void Client::check_caps(Inode *in, unsigned flags) | |
3491 | { | |
3492 | unsigned wanted = in->caps_wanted(); | |
3493 | unsigned used = get_caps_used(in); | |
3494 | unsigned cap_used; | |
3495 | ||
7c673cae FG |
3496 | int implemented; |
3497 | int issued = in->caps_issued(&implemented); | |
3498 | int revoking = implemented & ~issued; | |
3499 | ||
11fdf7f2 TL |
3500 | int orig_used = used; |
3501 | used = adjust_caps_used_for_lazyio(used, issued, implemented); | |
3502 | ||
7c673cae | 3503 | int retain = wanted | used | CEPH_CAP_PIN; |
a8e16298 TL |
3504 | if (!unmounting && in->nlink > 0) { |
3505 | if (wanted) { | |
7c673cae | 3506 | retain |= CEPH_CAP_ANY; |
a8e16298 TL |
3507 | } else if (in->is_dir() && |
3508 | (issued & CEPH_CAP_FILE_SHARED) && | |
3509 | (in->flags & I_COMPLETE)) { | |
3510 | // we do this here because we don't want to drop to Fs (and then | |
3511 | // drop the Fs if we do a create!) if that alone makes us send lookups | |
3512 | // to the MDS. Doing it in in->caps_wanted() has knock-on effects elsewhere | |
3513 | wanted = CEPH_CAP_ANY_SHARED | CEPH_CAP_FILE_EXCL; | |
3514 | retain |= wanted; | |
3515 | } else { | |
7c673cae | 3516 | retain |= CEPH_CAP_ANY_SHARED; |
a8e16298 TL |
3517 | // keep RD only if we didn't have the file open RW, |
3518 | // because then the mds would revoke it anyway to | |
3519 | // journal max_size=0. | |
3520 | if (in->max_size == 0) | |
3521 | retain |= CEPH_CAP_ANY_RD; | |
3522 | } | |
7c673cae FG |
3523 | } |
3524 | ||
11fdf7f2 | 3525 | ldout(cct, 10) << __func__ << " on " << *in |
7c673cae FG |
3526 | << " wanted " << ccap_string(wanted) |
3527 | << " used " << ccap_string(used) | |
3528 | << " issued " << ccap_string(issued) | |
3529 | << " revoking " << ccap_string(revoking) | |
3530 | << " flags=" << flags | |
3531 | << dendl; | |
3532 | ||
3533 | if (in->snapid != CEPH_NOSNAP) | |
3534 | return; //snap caps last forever, can't write | |
3535 | ||
3536 | if (in->caps.empty()) | |
3537 | return; // guard if at end of func | |
3538 | ||
11fdf7f2 TL |
3539 | if (!(orig_used & CEPH_CAP_FILE_BUFFER) && |
3540 | (revoking & used & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
94b18763 | 3541 | if (_release(in)) |
11fdf7f2 | 3542 | used &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
94b18763 | 3543 | } |
7c673cae | 3544 | |
7c673cae | 3545 | |
11fdf7f2 TL |
3546 | for (auto &p : in->caps) { |
3547 | mds_rank_t mds = p.first; | |
3548 | Cap &cap = p.second; | |
7c673cae | 3549 | |
11fdf7f2 | 3550 | MetaSession *session = &mds_sessions.at(mds); |
7c673cae FG |
3551 | |
3552 | cap_used = used; | |
11fdf7f2 | 3553 | if (in->auth_cap && &cap != in->auth_cap) |
7c673cae FG |
3554 | cap_used &= ~in->auth_cap->issued; |
3555 | ||
11fdf7f2 | 3556 | revoking = cap.implemented & ~cap.issued; |
7c673cae FG |
3557 | |
3558 | ldout(cct, 10) << " cap mds." << mds | |
11fdf7f2 TL |
3559 | << " issued " << ccap_string(cap.issued) |
3560 | << " implemented " << ccap_string(cap.implemented) | |
7c673cae FG |
3561 | << " revoking " << ccap_string(revoking) << dendl; |
3562 | ||
3563 | if (in->wanted_max_size > in->max_size && | |
3564 | in->wanted_max_size > in->requested_max_size && | |
11fdf7f2 | 3565 | &cap == in->auth_cap) |
7c673cae FG |
3566 | goto ack; |
3567 | ||
3568 | /* approaching file_max? */ | |
11fdf7f2 TL |
3569 | if ((cap.issued & CEPH_CAP_FILE_WR) && |
3570 | &cap == in->auth_cap && | |
31f18b77 | 3571 | is_max_size_approaching(in)) { |
7c673cae | 3572 | ldout(cct, 10) << "size " << in->size << " approaching max_size " << in->max_size |
31f18b77 | 3573 | << ", reported " << in->reported_size << dendl; |
7c673cae FG |
3574 | goto ack; |
3575 | } | |
3576 | ||
3577 | /* completed revocation? */ | |
3578 | if (revoking && (revoking & cap_used) == 0) { | |
11fdf7f2 | 3579 | ldout(cct, 10) << "completed revocation of " << ccap_string(cap.implemented & ~cap.issued) << dendl; |
7c673cae FG |
3580 | goto ack; |
3581 | } | |
3582 | ||
3583 | /* want more caps from mds? */ | |
11fdf7f2 | 3584 | if (wanted & ~(cap.wanted | cap.issued)) |
7c673cae FG |
3585 | goto ack; |
3586 | ||
3587 | if (!revoking && unmounting && (cap_used == 0)) | |
3588 | goto ack; | |
3589 | ||
11fdf7f2 | 3590 | if ((cap.issued & ~retain) == 0 && // and we don't have anything we wouldn't like |
a8e16298 | 3591 | !in->dirty_caps) // and we have no dirty caps |
7c673cae FG |
3592 | continue; |
3593 | ||
11fdf7f2 | 3594 | if (!(flags & CHECK_CAPS_NODELAY)) { |
7c673cae | 3595 | ldout(cct, 10) << "delaying cap release" << dendl; |
11fdf7f2 | 3596 | cap_delay_requeue(in); |
7c673cae FG |
3597 | continue; |
3598 | } | |
3599 | ||
3600 | ack: | |
eafe8130 TL |
3601 | if (&cap == in->auth_cap) { |
3602 | if (in->flags & I_KICK_FLUSH) { | |
3603 | ldout(cct, 20) << " reflushing caps (check_caps) on " << *in | |
3604 | << " to mds." << mds << dendl; | |
3605 | kick_flushing_caps(in, session); | |
3606 | } | |
3607 | if (!in->cap_snaps.empty() && | |
3608 | in->cap_snaps.rbegin()->second.flush_tid == 0) | |
3609 | flush_snaps(in); | |
7c673cae FG |
3610 | } |
3611 | ||
3612 | int flushing; | |
3613 | ceph_tid_t flush_tid; | |
11fdf7f2 | 3614 | if (in->auth_cap == &cap && in->dirty_caps) { |
7c673cae FG |
3615 | flushing = mark_caps_flushing(in, &flush_tid); |
3616 | } else { | |
3617 | flushing = 0; | |
3618 | flush_tid = 0; | |
3619 | } | |
3620 | ||
eafe8130 TL |
3621 | int msg_flags = (flags & CHECK_CAPS_SYNCHRONOUS) ? MClientCaps::FLAG_SYNC : 0; |
3622 | send_cap(in, session, &cap, msg_flags, cap_used, wanted, retain, | |
3623 | flushing, flush_tid); | |
7c673cae FG |
3624 | } |
3625 | } | |
3626 | ||
3627 | ||
3628 | void Client::queue_cap_snap(Inode *in, SnapContext& old_snapc) | |
3629 | { | |
3630 | int used = get_caps_used(in); | |
3631 | int dirty = in->caps_dirty(); | |
11fdf7f2 | 3632 | ldout(cct, 10) << __func__ << " " << *in << " snapc " << old_snapc << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3633 | |
3634 | if (in->cap_snaps.size() && | |
3635 | in->cap_snaps.rbegin()->second.writing) { | |
11fdf7f2 | 3636 | ldout(cct, 10) << __func__ << " already have pending cap_snap on " << *in << dendl; |
7c673cae FG |
3637 | return; |
3638 | } else if (in->caps_dirty() || | |
3639 | (used & CEPH_CAP_FILE_WR) || | |
3640 | (dirty & CEPH_CAP_ANY_WR)) { | |
3641 | const auto &capsnapem = in->cap_snaps.emplace(std::piecewise_construct, std::make_tuple(old_snapc.seq), std::make_tuple(in)); | |
11fdf7f2 | 3642 | ceph_assert(capsnapem.second); /* element inserted */ |
7c673cae FG |
3643 | CapSnap &capsnap = capsnapem.first->second; |
3644 | capsnap.context = old_snapc; | |
3645 | capsnap.issued = in->caps_issued(); | |
3646 | capsnap.dirty = in->caps_dirty(); | |
3647 | ||
3648 | capsnap.dirty_data = (used & CEPH_CAP_FILE_BUFFER); | |
3649 | ||
3650 | capsnap.uid = in->uid; | |
3651 | capsnap.gid = in->gid; | |
3652 | capsnap.mode = in->mode; | |
3653 | capsnap.btime = in->btime; | |
3654 | capsnap.xattrs = in->xattrs; | |
3655 | capsnap.xattr_version = in->xattr_version; | |
11fdf7f2 TL |
3656 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; |
3657 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
7c673cae FG |
3658 | |
3659 | if (used & CEPH_CAP_FILE_WR) { | |
11fdf7f2 | 3660 | ldout(cct, 10) << __func__ << " WR used on " << *in << dendl; |
7c673cae FG |
3661 | capsnap.writing = 1; |
3662 | } else { | |
3663 | finish_cap_snap(in, capsnap, used); | |
3664 | } | |
3665 | } else { | |
11fdf7f2 | 3666 | ldout(cct, 10) << __func__ << " not dirty|writing on " << *in << dendl; |
7c673cae FG |
3667 | } |
3668 | } | |
3669 | ||
3670 | void Client::finish_cap_snap(Inode *in, CapSnap &capsnap, int used) | |
3671 | { | |
11fdf7f2 | 3672 | ldout(cct, 10) << __func__ << " " << *in << " capsnap " << (void *)&capsnap << " used " << ccap_string(used) << dendl; |
7c673cae FG |
3673 | capsnap.size = in->size; |
3674 | capsnap.mtime = in->mtime; | |
3675 | capsnap.atime = in->atime; | |
3676 | capsnap.ctime = in->ctime; | |
3677 | capsnap.time_warp_seq = in->time_warp_seq; | |
3678 | capsnap.change_attr = in->change_attr; | |
7c673cae FG |
3679 | capsnap.dirty |= in->caps_dirty(); |
3680 | ||
11fdf7f2 TL |
3681 | /* Only reset it if it wasn't set before */ |
3682 | if (capsnap.cap_dirtier_uid == -1) { | |
3683 | capsnap.cap_dirtier_uid = in->cap_dirtier_uid; | |
3684 | capsnap.cap_dirtier_gid = in->cap_dirtier_gid; | |
3685 | } | |
3686 | ||
7c673cae FG |
3687 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { |
3688 | capsnap.inline_data = in->inline_data; | |
3689 | capsnap.inline_version = in->inline_version; | |
3690 | } | |
3691 | ||
3692 | if (used & CEPH_CAP_FILE_BUFFER) { | |
11fdf7f2 | 3693 | ldout(cct, 10) << __func__ << " " << *in << " cap_snap " << &capsnap << " used " << used |
7c673cae FG |
3694 | << " WRBUFFER, delaying" << dendl; |
3695 | } else { | |
3696 | capsnap.dirty_data = 0; | |
3697 | flush_snaps(in); | |
3698 | } | |
3699 | } | |
3700 | ||
3701 | void Client::_flushed_cap_snap(Inode *in, snapid_t seq) | |
3702 | { | |
11fdf7f2 | 3703 | ldout(cct, 10) << __func__ << " seq " << seq << " on " << *in << dendl; |
7c673cae FG |
3704 | in->cap_snaps.at(seq).dirty_data = 0; |
3705 | flush_snaps(in); | |
3706 | } | |
3707 | ||
eafe8130 TL |
3708 | void Client::send_flush_snap(Inode *in, MetaSession *session, |
3709 | snapid_t follows, CapSnap& capsnap) | |
3710 | { | |
9f95a23c TL |
3711 | auto m = make_message<MClientCaps>(CEPH_CAP_OP_FLUSHSNAP, |
3712 | in->ino, in->snaprealm->ino, 0, | |
3713 | in->auth_cap->mseq, cap_epoch_barrier); | |
eafe8130 TL |
3714 | m->caller_uid = capsnap.cap_dirtier_uid; |
3715 | m->caller_gid = capsnap.cap_dirtier_gid; | |
3716 | ||
3717 | m->set_client_tid(capsnap.flush_tid); | |
3718 | m->head.snap_follows = follows; | |
3719 | ||
3720 | m->head.caps = capsnap.issued; | |
3721 | m->head.dirty = capsnap.dirty; | |
3722 | ||
3723 | m->head.uid = capsnap.uid; | |
3724 | m->head.gid = capsnap.gid; | |
3725 | m->head.mode = capsnap.mode; | |
3726 | m->btime = capsnap.btime; | |
3727 | ||
3728 | m->size = capsnap.size; | |
3729 | ||
3730 | m->head.xattr_version = capsnap.xattr_version; | |
3731 | encode(capsnap.xattrs, m->xattrbl); | |
3732 | ||
3733 | m->ctime = capsnap.ctime; | |
3734 | m->btime = capsnap.btime; | |
3735 | m->mtime = capsnap.mtime; | |
3736 | m->atime = capsnap.atime; | |
3737 | m->time_warp_seq = capsnap.time_warp_seq; | |
3738 | m->change_attr = capsnap.change_attr; | |
3739 | ||
3740 | if (capsnap.dirty & CEPH_CAP_FILE_WR) { | |
3741 | m->inline_version = in->inline_version; | |
3742 | m->inline_data = in->inline_data; | |
3743 | } | |
3744 | ||
3745 | ceph_assert(!session->flushing_caps_tids.empty()); | |
3746 | m->set_oldest_flush_tid(*session->flushing_caps_tids.begin()); | |
3747 | ||
3748 | session->con->send_message2(std::move(m)); | |
3749 | } | |
3750 | ||
3751 | void Client::flush_snaps(Inode *in) | |
7c673cae | 3752 | { |
eafe8130 | 3753 | ldout(cct, 10) << "flush_snaps on " << *in << dendl; |
11fdf7f2 | 3754 | ceph_assert(in->cap_snaps.size()); |
7c673cae FG |
3755 | |
3756 | // pick auth mds | |
11fdf7f2 | 3757 | ceph_assert(in->auth_cap); |
7c673cae | 3758 | MetaSession *session = in->auth_cap->session; |
7c673cae FG |
3759 | |
3760 | for (auto &p : in->cap_snaps) { | |
3761 | CapSnap &capsnap = p.second; | |
eafe8130 TL |
3762 | // only do new flush |
3763 | if (capsnap.flush_tid > 0) | |
3764 | continue; | |
7c673cae FG |
3765 | |
3766 | ldout(cct, 10) << "flush_snaps mds." << session->mds_num | |
3767 | << " follows " << p.first | |
3768 | << " size " << capsnap.size | |
3769 | << " mtime " << capsnap.mtime | |
3770 | << " dirty_data=" << capsnap.dirty_data | |
3771 | << " writing=" << capsnap.writing | |
3772 | << " on " << *in << dendl; | |
3773 | if (capsnap.dirty_data || capsnap.writing) | |
eafe8130 | 3774 | break; |
7c673cae | 3775 | |
eafe8130 TL |
3776 | capsnap.flush_tid = ++last_flush_tid; |
3777 | session->flushing_caps_tids.insert(capsnap.flush_tid); | |
3778 | in->flushing_cap_tids[capsnap.flush_tid] = 0; | |
3779 | if (!in->flushing_cap_item.is_on_list()) | |
3780 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
7c673cae | 3781 | |
eafe8130 | 3782 | send_flush_snap(in, session, p.first, capsnap); |
7c673cae FG |
3783 | } |
3784 | } | |
3785 | ||
9f95a23c | 3786 | void Client::wait_on_list(list<ceph::condition_variable*>& ls) |
7c673cae | 3787 | { |
9f95a23c | 3788 | ceph::condition_variable cond; |
7c673cae | 3789 | ls.push_back(&cond); |
9f95a23c TL |
3790 | std::unique_lock l{client_lock, std::adopt_lock}; |
3791 | cond.wait(l); | |
3792 | l.release(); | |
7c673cae FG |
3793 | ls.remove(&cond); |
3794 | } | |
3795 | ||
9f95a23c | 3796 | void Client::signal_cond_list(list<ceph::condition_variable*>& ls) |
7c673cae | 3797 | { |
9f95a23c TL |
3798 | for (auto cond : ls) { |
3799 | cond->notify_all(); | |
3800 | } | |
7c673cae FG |
3801 | } |
3802 | ||
3803 | void Client::wait_on_context_list(list<Context*>& ls) | |
3804 | { | |
9f95a23c | 3805 | ceph::condition_variable cond; |
7c673cae FG |
3806 | bool done = false; |
3807 | int r; | |
9f95a23c TL |
3808 | ls.push_back(new C_Cond(cond, &done, &r)); |
3809 | std::unique_lock l{client_lock, std::adopt_lock}; | |
3810 | cond.wait(l, [&done] { return done;}); | |
3811 | l.release(); | |
7c673cae FG |
3812 | } |
3813 | ||
3814 | void Client::signal_context_list(list<Context*>& ls) | |
3815 | { | |
3816 | while (!ls.empty()) { | |
3817 | ls.front()->complete(0); | |
3818 | ls.pop_front(); | |
3819 | } | |
3820 | } | |
3821 | ||
a8e16298 | 3822 | void Client::wake_up_session_caps(MetaSession *s, bool reconnect) |
7c673cae | 3823 | { |
11fdf7f2 TL |
3824 | for (const auto &cap : s->caps) { |
3825 | auto &in = cap->inode; | |
a8e16298 | 3826 | if (reconnect) { |
11fdf7f2 TL |
3827 | in.requested_max_size = 0; |
3828 | in.wanted_max_size = 0; | |
a8e16298 TL |
3829 | } else { |
3830 | if (cap->gen < s->cap_gen) { | |
3831 | // mds did not re-issue stale cap. | |
3832 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
3833 | // make sure mds knows what we want. | |
11fdf7f2 TL |
3834 | if (in.caps_file_wanted() & ~cap->wanted) |
3835 | in.flags |= I_CAP_DROPPED; | |
a8e16298 TL |
3836 | } |
3837 | } | |
11fdf7f2 | 3838 | signal_cond_list(in.waitfor_caps); |
7c673cae FG |
3839 | } |
3840 | } | |
3841 | ||
3842 | ||
3843 | // flush dirty data (from objectcache) | |
3844 | ||
3845 | class C_Client_CacheInvalidate : public Context { | |
3846 | private: | |
3847 | Client *client; | |
3848 | vinodeno_t ino; | |
3849 | int64_t offset, length; | |
3850 | public: | |
3851 | C_Client_CacheInvalidate(Client *c, Inode *in, int64_t off, int64_t len) : | |
3852 | client(c), offset(off), length(len) { | |
3853 | if (client->use_faked_inos()) | |
3854 | ino = vinodeno_t(in->faked_ino, CEPH_NOSNAP); | |
3855 | else | |
3856 | ino = in->vino(); | |
3857 | } | |
3858 | void finish(int r) override { | |
3859 | // _async_invalidate takes the lock when it needs to, call this back from outside of lock. | |
9f95a23c | 3860 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
3861 | client->_async_invalidate(ino, offset, length); |
3862 | } | |
3863 | }; | |
3864 | ||
3865 | void Client::_async_invalidate(vinodeno_t ino, int64_t off, int64_t len) | |
3866 | { | |
3867 | if (unmounting) | |
3868 | return; | |
11fdf7f2 | 3869 | ldout(cct, 10) << __func__ << " " << ino << " " << off << "~" << len << dendl; |
7c673cae FG |
3870 | ino_invalidate_cb(callback_handle, ino, off, len); |
3871 | } | |
3872 | ||
3873 | void Client::_schedule_invalidate_callback(Inode *in, int64_t off, int64_t len) { | |
3874 | ||
3875 | if (ino_invalidate_cb) | |
3876 | // we queue the invalidate, which calls the callback and decrements the ref | |
3877 | async_ino_invalidator.queue(new C_Client_CacheInvalidate(this, in, off, len)); | |
3878 | } | |
3879 | ||
3880 | void Client::_invalidate_inode_cache(Inode *in) | |
3881 | { | |
11fdf7f2 | 3882 | ldout(cct, 10) << __func__ << " " << *in << dendl; |
7c673cae FG |
3883 | |
3884 | // invalidate our userspace inode cache | |
94b18763 | 3885 | if (cct->_conf->client_oc) { |
7c673cae | 3886 | objectcacher->release_set(&in->oset); |
94b18763 FG |
3887 | if (!objectcacher->set_is_empty(&in->oset)) |
3888 | lderr(cct) << "failed to invalidate cache for " << *in << dendl; | |
3889 | } | |
7c673cae FG |
3890 | |
3891 | _schedule_invalidate_callback(in, 0, 0); | |
3892 | } | |
3893 | ||
3894 | void Client::_invalidate_inode_cache(Inode *in, int64_t off, int64_t len) | |
3895 | { | |
11fdf7f2 | 3896 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
3897 | |
3898 | // invalidate our userspace inode cache | |
3899 | if (cct->_conf->client_oc) { | |
3900 | vector<ObjectExtent> ls; | |
3901 | Striper::file_to_extents(cct, in->ino, &in->layout, off, len, in->truncate_size, ls); | |
28e407b8 | 3902 | objectcacher->discard_writeback(&in->oset, ls, nullptr); |
7c673cae FG |
3903 | } |
3904 | ||
3905 | _schedule_invalidate_callback(in, off, len); | |
3906 | } | |
3907 | ||
3908 | bool Client::_release(Inode *in) | |
3909 | { | |
3910 | ldout(cct, 20) << "_release " << *in << dendl; | |
3911 | if (in->cap_refs[CEPH_CAP_FILE_CACHE] == 0) { | |
3912 | _invalidate_inode_cache(in); | |
3913 | return true; | |
3914 | } | |
3915 | return false; | |
3916 | } | |
3917 | ||
3918 | bool Client::_flush(Inode *in, Context *onfinish) | |
3919 | { | |
3920 | ldout(cct, 10) << "_flush " << *in << dendl; | |
3921 | ||
3922 | if (!in->oset.dirty_or_tx) { | |
3923 | ldout(cct, 10) << " nothing to flush" << dendl; | |
3924 | onfinish->complete(0); | |
3925 | return true; | |
3926 | } | |
3927 | ||
3928 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
1adf2230 | 3929 | ldout(cct, 8) << __func__ << ": FULL, purging for ENOSPC" << dendl; |
7c673cae FG |
3930 | objectcacher->purge_set(&in->oset); |
3931 | if (onfinish) { | |
3932 | onfinish->complete(-ENOSPC); | |
3933 | } | |
3934 | return true; | |
3935 | } | |
3936 | ||
3937 | return objectcacher->flush_set(&in->oset, onfinish); | |
3938 | } | |
3939 | ||
3940 | void Client::_flush_range(Inode *in, int64_t offset, uint64_t size) | |
3941 | { | |
9f95a23c | 3942 | ceph_assert(ceph_mutex_is_locked(client_lock)); |
7c673cae FG |
3943 | if (!in->oset.dirty_or_tx) { |
3944 | ldout(cct, 10) << " nothing to flush" << dendl; | |
3945 | return; | |
3946 | } | |
3947 | ||
11fdf7f2 | 3948 | C_SaferCond onflush("Client::_flush_range flock"); |
7c673cae | 3949 | bool ret = objectcacher->file_flush(&in->oset, &in->layout, in->snaprealm->get_snap_context(), |
11fdf7f2 | 3950 | offset, size, &onflush); |
7c673cae FG |
3951 | if (!ret) { |
3952 | // wait for flush | |
9f95a23c | 3953 | client_lock.unlock(); |
11fdf7f2 | 3954 | onflush.wait(); |
9f95a23c | 3955 | client_lock.lock(); |
7c673cae FG |
3956 | } |
3957 | } | |
3958 | ||
3959 | void Client::flush_set_callback(ObjectCacher::ObjectSet *oset) | |
3960 | { | |
11fdf7f2 | 3961 | // std::lock_guard l(client_lock); |
9f95a23c | 3962 | ceph_assert(ceph_mutex_is_locked(client_lock)); // will be called via dispatch() -> objecter -> ... |
7c673cae | 3963 | Inode *in = static_cast<Inode *>(oset->parent); |
11fdf7f2 | 3964 | ceph_assert(in); |
7c673cae FG |
3965 | _flushed(in); |
3966 | } | |
3967 | ||
3968 | void Client::_flushed(Inode *in) | |
3969 | { | |
3970 | ldout(cct, 10) << "_flushed " << *in << dendl; | |
3971 | ||
3972 | put_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
3973 | } | |
3974 | ||
3975 | ||
3976 | ||
3977 | // checks common to add_update_cap, handle_cap_grant | |
11fdf7f2 | 3978 | void Client::check_cap_issue(Inode *in, unsigned issued) |
7c673cae FG |
3979 | { |
3980 | unsigned had = in->caps_issued(); | |
3981 | ||
3982 | if ((issued & CEPH_CAP_FILE_CACHE) && | |
3983 | !(had & CEPH_CAP_FILE_CACHE)) | |
3984 | in->cache_gen++; | |
3985 | ||
3986 | if ((issued & CEPH_CAP_FILE_SHARED) && | |
3987 | !(had & CEPH_CAP_FILE_SHARED)) { | |
3988 | in->shared_gen++; | |
3989 | ||
3990 | if (in->is_dir()) | |
3991 | clear_dir_complete_and_ordered(in, true); | |
3992 | } | |
3993 | } | |
3994 | ||
3995 | void Client::add_update_cap(Inode *in, MetaSession *mds_session, uint64_t cap_id, | |
a8e16298 TL |
3996 | unsigned issued, unsigned wanted, unsigned seq, unsigned mseq, |
3997 | inodeno_t realm, int flags, const UserPerm& cap_perms) | |
7c673cae | 3998 | { |
11fdf7f2 TL |
3999 | if (!in->is_any_caps()) { |
4000 | ceph_assert(in->snaprealm == 0); | |
4001 | in->snaprealm = get_snap_realm(realm); | |
4002 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4003 | ldout(cct, 15) << __func__ << " first one, opened snaprealm " << in->snaprealm << dendl; | |
4004 | } else { | |
4005 | ceph_assert(in->snaprealm); | |
4006 | if ((flags & CEPH_CAP_FLAG_AUTH) && | |
4007 | realm != inodeno_t(-1) && in->snaprealm->ino != realm) { | |
4008 | in->snaprealm_item.remove_myself(); | |
4009 | auto oldrealm = in->snaprealm; | |
4010 | in->snaprealm = get_snap_realm(realm); | |
4011 | in->snaprealm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4012 | put_snap_realm(oldrealm); | |
4013 | } | |
4014 | } | |
4015 | ||
7c673cae | 4016 | mds_rank_t mds = mds_session->mds_num; |
11fdf7f2 TL |
4017 | const auto &capem = in->caps.emplace(std::piecewise_construct, std::forward_as_tuple(mds), std::forward_as_tuple(*in, mds_session)); |
4018 | Cap &cap = capem.first->second; | |
4019 | if (!capem.second) { | |
4020 | if (cap.gen < mds_session->cap_gen) | |
4021 | cap.issued = cap.implemented = CEPH_CAP_PIN; | |
7c673cae FG |
4022 | |
4023 | /* | |
4024 | * auth mds of the inode changed. we received the cap export | |
4025 | * message, but still haven't received the cap import message. | |
4026 | * handle_cap_export() updated the new auth MDS' cap. | |
4027 | * | |
4028 | * "ceph_seq_cmp(seq, cap->seq) <= 0" means we are processing | |
4029 | * a message that was send before the cap import message. So | |
4030 | * don't remove caps. | |
4031 | */ | |
11fdf7f2 | 4032 | if (ceph_seq_cmp(seq, cap.seq) <= 0) { |
92f5a8d4 TL |
4033 | if (&cap != in->auth_cap) |
4034 | ldout(cct, 0) << "WARNING: " << "inode " << *in << " caps on mds." << mds << " != auth_cap." << dendl; | |
4035 | ||
11fdf7f2 TL |
4036 | ceph_assert(cap.cap_id == cap_id); |
4037 | seq = cap.seq; | |
4038 | mseq = cap.mseq; | |
4039 | issued |= cap.issued; | |
7c673cae FG |
4040 | flags |= CEPH_CAP_FLAG_AUTH; |
4041 | } | |
7c673cae FG |
4042 | } |
4043 | ||
11fdf7f2 | 4044 | check_cap_issue(in, issued); |
7c673cae FG |
4045 | |
4046 | if (flags & CEPH_CAP_FLAG_AUTH) { | |
11fdf7f2 | 4047 | if (in->auth_cap != &cap && |
7c673cae FG |
4048 | (!in->auth_cap || ceph_seq_cmp(in->auth_cap->mseq, mseq) < 0)) { |
4049 | if (in->auth_cap && in->flushing_cap_item.is_on_list()) { | |
11fdf7f2 | 4050 | ldout(cct, 10) << __func__ << " changing auth cap: " |
7c673cae FG |
4051 | << "add myself to new auth MDS' flushing caps list" << dendl; |
4052 | adjust_session_flushing_caps(in, in->auth_cap->session, mds_session); | |
4053 | } | |
11fdf7f2 | 4054 | in->auth_cap = ∩ |
7c673cae FG |
4055 | } |
4056 | } | |
4057 | ||
11fdf7f2 TL |
4058 | unsigned old_caps = cap.issued; |
4059 | cap.cap_id = cap_id; | |
4060 | cap.issued = issued; | |
4061 | cap.implemented |= issued; | |
4062 | if (ceph_seq_cmp(mseq, cap.mseq) > 0) | |
4063 | cap.wanted = wanted; | |
a8e16298 | 4064 | else |
11fdf7f2 TL |
4065 | cap.wanted |= wanted; |
4066 | cap.seq = seq; | |
4067 | cap.issue_seq = seq; | |
4068 | cap.mseq = mseq; | |
4069 | cap.gen = mds_session->cap_gen; | |
4070 | cap.latest_perms = cap_perms; | |
4071 | ldout(cct, 10) << __func__ << " issued " << ccap_string(old_caps) << " -> " << ccap_string(cap.issued) | |
4072 | << " from mds." << mds | |
4073 | << " on " << *in | |
4074 | << dendl; | |
4075 | ||
4076 | if ((issued & ~old_caps) && in->auth_cap == &cap) { | |
7c673cae | 4077 | // non-auth MDS is revoking the newly grant caps ? |
11fdf7f2 TL |
4078 | for (auto &p : in->caps) { |
4079 | if (&p.second == &cap) | |
7c673cae | 4080 | continue; |
11fdf7f2 | 4081 | if (p.second.implemented & ~p.second.issued & issued) { |
7c673cae FG |
4082 | check_caps(in, CHECK_CAPS_NODELAY); |
4083 | break; | |
4084 | } | |
4085 | } | |
4086 | } | |
4087 | ||
4088 | if (issued & ~old_caps) | |
4089 | signal_cond_list(in->waitfor_caps); | |
4090 | } | |
4091 | ||
4092 | void Client::remove_cap(Cap *cap, bool queue_release) | |
4093 | { | |
11fdf7f2 | 4094 | auto &in = cap->inode; |
7c673cae FG |
4095 | MetaSession *session = cap->session; |
4096 | mds_rank_t mds = cap->session->mds_num; | |
4097 | ||
11fdf7f2 | 4098 | ldout(cct, 10) << __func__ << " mds." << mds << " on " << in << dendl; |
7c673cae FG |
4099 | |
4100 | if (queue_release) { | |
4101 | session->enqueue_cap_release( | |
11fdf7f2 | 4102 | in.ino, |
7c673cae FG |
4103 | cap->cap_id, |
4104 | cap->issue_seq, | |
4105 | cap->mseq, | |
4106 | cap_epoch_barrier); | |
4107 | } | |
4108 | ||
11fdf7f2 TL |
4109 | if (in.auth_cap == cap) { |
4110 | if (in.flushing_cap_item.is_on_list()) { | |
7c673cae | 4111 | ldout(cct, 10) << " removing myself from flushing_cap list" << dendl; |
11fdf7f2 | 4112 | in.flushing_cap_item.remove_myself(); |
7c673cae | 4113 | } |
11fdf7f2 | 4114 | in.auth_cap = NULL; |
7c673cae | 4115 | } |
11fdf7f2 TL |
4116 | size_t n = in.caps.erase(mds); |
4117 | ceph_assert(n == 1); | |
7c673cae FG |
4118 | cap = nullptr; |
4119 | ||
11fdf7f2 TL |
4120 | if (!in.is_any_caps()) { |
4121 | ldout(cct, 15) << __func__ << " last one, closing snaprealm " << in.snaprealm << dendl; | |
4122 | in.snaprealm_item.remove_myself(); | |
4123 | put_snap_realm(in.snaprealm); | |
4124 | in.snaprealm = 0; | |
7c673cae FG |
4125 | } |
4126 | } | |
4127 | ||
4128 | void Client::remove_all_caps(Inode *in) | |
4129 | { | |
4130 | while (!in->caps.empty()) | |
11fdf7f2 | 4131 | remove_cap(&in->caps.begin()->second, true); |
7c673cae FG |
4132 | } |
4133 | ||
4134 | void Client::remove_session_caps(MetaSession *s) | |
4135 | { | |
11fdf7f2 | 4136 | ldout(cct, 10) << __func__ << " mds." << s->mds_num << dendl; |
7c673cae FG |
4137 | |
4138 | while (s->caps.size()) { | |
4139 | Cap *cap = *s->caps.begin(); | |
11fdf7f2 | 4140 | InodeRef in(&cap->inode); |
eafe8130 | 4141 | bool dirty_caps = false; |
7c673cae | 4142 | if (in->auth_cap == cap) { |
7c673cae FG |
4143 | dirty_caps = in->dirty_caps | in->flushing_caps; |
4144 | in->wanted_max_size = 0; | |
4145 | in->requested_max_size = 0; | |
7c673cae | 4146 | } |
a8e16298 TL |
4147 | if (cap->wanted | cap->issued) |
4148 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 4149 | remove_cap(cap, false); |
eafe8130 | 4150 | in->cap_snaps.clear(); |
7c673cae | 4151 | if (dirty_caps) { |
11fdf7f2 | 4152 | lderr(cct) << __func__ << " still has dirty|flushing caps on " << *in << dendl; |
7c673cae FG |
4153 | if (in->flushing_caps) { |
4154 | num_flushing_caps--; | |
4155 | in->flushing_cap_tids.clear(); | |
4156 | } | |
4157 | in->flushing_caps = 0; | |
28e407b8 | 4158 | in->mark_caps_clean(); |
11fdf7f2 | 4159 | put_inode(in.get()); |
7c673cae | 4160 | } |
a8e16298 | 4161 | signal_cond_list(in->waitfor_caps); |
7c673cae FG |
4162 | } |
4163 | s->flushing_caps_tids.clear(); | |
9f95a23c | 4164 | sync_cond.notify_all(); |
7c673cae FG |
4165 | } |
4166 | ||
91327a77 | 4167 | int Client::_do_remount(bool retry_on_error) |
b32b8144 | 4168 | { |
11fdf7f2 | 4169 | uint64_t max_retries = g_conf().get_val<uint64_t>("mds_max_retries_on_remount_failure"); |
91327a77 | 4170 | |
b32b8144 FG |
4171 | errno = 0; |
4172 | int r = remount_cb(callback_handle); | |
91327a77 AA |
4173 | if (r == 0) { |
4174 | retries_on_invalidate = 0; | |
4175 | } else { | |
b32b8144 FG |
4176 | int e = errno; |
4177 | client_t whoami = get_nodeid(); | |
4178 | if (r == -1) { | |
4179 | lderr(cct) << | |
4180 | "failed to remount (to trim kernel dentries): " | |
4181 | "errno = " << e << " (" << strerror(e) << ")" << dendl; | |
4182 | } else { | |
4183 | lderr(cct) << | |
4184 | "failed to remount (to trim kernel dentries): " | |
4185 | "return code = " << r << dendl; | |
4186 | } | |
91327a77 | 4187 | bool should_abort = |
11fdf7f2 TL |
4188 | (cct->_conf.get_val<bool>("client_die_on_failed_remount") || |
4189 | cct->_conf.get_val<bool>("client_die_on_failed_dentry_invalidate")) && | |
91327a77 | 4190 | !(retry_on_error && (++retries_on_invalidate < max_retries)); |
b32b8144 FG |
4191 | if (should_abort && !unmounting) { |
4192 | lderr(cct) << "failed to remount for kernel dentry trimming; quitting!" << dendl; | |
4193 | ceph_abort(); | |
4194 | } | |
4195 | } | |
4196 | return r; | |
4197 | } | |
4198 | ||
7c673cae FG |
4199 | class C_Client_Remount : public Context { |
4200 | private: | |
4201 | Client *client; | |
4202 | public: | |
4203 | explicit C_Client_Remount(Client *c) : client(c) {} | |
4204 | void finish(int r) override { | |
11fdf7f2 | 4205 | ceph_assert(r == 0); |
91327a77 | 4206 | client->_do_remount(true); |
7c673cae FG |
4207 | } |
4208 | }; | |
4209 | ||
4210 | void Client::_invalidate_kernel_dcache() | |
4211 | { | |
4212 | if (unmounting) | |
4213 | return; | |
94b18763 FG |
4214 | if (can_invalidate_dentries) { |
4215 | if (dentry_invalidate_cb && root->dir) { | |
4216 | for (ceph::unordered_map<string, Dentry*>::iterator p = root->dir->dentries.begin(); | |
4217 | p != root->dir->dentries.end(); | |
4218 | ++p) { | |
4219 | if (p->second->inode) | |
4220 | _schedule_invalidate_dentry_callback(p->second, false); | |
4221 | } | |
7c673cae FG |
4222 | } |
4223 | } else if (remount_cb) { | |
4224 | // Hacky: | |
4225 | // when remounting a file system, linux kernel trims all unused dentries in the fs | |
4226 | remount_finisher.queue(new C_Client_Remount(this)); | |
4227 | } | |
4228 | } | |
4229 | ||
91327a77 AA |
4230 | void Client::_trim_negative_child_dentries(InodeRef& in) |
4231 | { | |
4232 | if (!in->is_dir()) | |
4233 | return; | |
4234 | ||
4235 | Dir* dir = in->dir; | |
4236 | if (dir && dir->dentries.size() == dir->num_null_dentries) { | |
4237 | for (auto p = dir->dentries.begin(); p != dir->dentries.end(); ) { | |
4238 | Dentry *dn = p->second; | |
4239 | ++p; | |
11fdf7f2 | 4240 | ceph_assert(!dn->inode); |
91327a77 AA |
4241 | if (dn->lru_is_expireable()) |
4242 | unlink(dn, true, false); // keep dir, drop dentry | |
4243 | } | |
4244 | if (dir->dentries.empty()) { | |
4245 | close_dir(dir); | |
4246 | } | |
4247 | } | |
4248 | ||
4249 | if (in->flags & I_SNAPDIR_OPEN) { | |
4250 | InodeRef snapdir = open_snapdir(in.get()); | |
4251 | _trim_negative_child_dentries(snapdir); | |
4252 | } | |
4253 | } | |
4254 | ||
28e407b8 | 4255 | void Client::trim_caps(MetaSession *s, uint64_t max) |
7c673cae FG |
4256 | { |
4257 | mds_rank_t mds = s->mds_num; | |
28e407b8 | 4258 | size_t caps_size = s->caps.size(); |
11fdf7f2 | 4259 | ldout(cct, 10) << __func__ << " mds." << mds << " max " << max |
7c673cae FG |
4260 | << " caps " << caps_size << dendl; |
4261 | ||
28e407b8 AA |
4262 | uint64_t trimmed = 0; |
4263 | auto p = s->caps.begin(); | |
4264 | std::set<Dentry *> to_trim; /* this avoids caps other than the one we're | |
4265 | * looking at from getting deleted during traversal. */ | |
7c673cae FG |
4266 | while ((caps_size - trimmed) > max && !p.end()) { |
4267 | Cap *cap = *p; | |
11fdf7f2 | 4268 | InodeRef in(&cap->inode); |
7c673cae FG |
4269 | |
4270 | // Increment p early because it will be invalidated if cap | |
4271 | // is deleted inside remove_cap | |
4272 | ++p; | |
4273 | ||
4274 | if (in->caps.size() > 1 && cap != in->auth_cap) { | |
4275 | int mine = cap->issued | cap->implemented; | |
4276 | int oissued = in->auth_cap ? in->auth_cap->issued : 0; | |
4277 | // disposable non-auth cap | |
b32b8144 | 4278 | if (!(get_caps_used(in.get()) & ~oissued & mine)) { |
7c673cae | 4279 | ldout(cct, 20) << " removing unused, unneeded non-auth cap on " << *in << dendl; |
28e407b8 | 4280 | cap = (remove_cap(cap, true), nullptr); |
7c673cae FG |
4281 | trimmed++; |
4282 | } | |
4283 | } else { | |
4284 | ldout(cct, 20) << " trying to trim dentries for " << *in << dendl; | |
91327a77 | 4285 | _trim_negative_child_dentries(in); |
7c673cae | 4286 | bool all = true; |
11fdf7f2 TL |
4287 | auto q = in->dentries.begin(); |
4288 | while (q != in->dentries.end()) { | |
4289 | Dentry *dn = *q; | |
4290 | ++q; | |
7c673cae FG |
4291 | if (dn->lru_is_expireable()) { |
4292 | if (can_invalidate_dentries && | |
4293 | dn->dir->parent_inode->ino == MDS_INO_ROOT) { | |
4294 | // Only issue one of these per DN for inodes in root: handle | |
4295 | // others more efficiently by calling for root-child DNs at | |
4296 | // the end of this function. | |
4297 | _schedule_invalidate_dentry_callback(dn, true); | |
4298 | } | |
28e407b8 AA |
4299 | ldout(cct, 20) << " queueing dentry for trimming: " << dn->name << dendl; |
4300 | to_trim.insert(dn); | |
7c673cae FG |
4301 | } else { |
4302 | ldout(cct, 20) << " not expirable: " << dn->name << dendl; | |
4303 | all = false; | |
4304 | } | |
4305 | } | |
4306 | if (all && in->ino != MDS_INO_ROOT) { | |
4307 | ldout(cct, 20) << __func__ << " counting as trimmed: " << *in << dendl; | |
4308 | trimmed++; | |
4309 | } | |
4310 | } | |
4311 | } | |
28e407b8 AA |
4312 | ldout(cct, 20) << " trimming queued dentries: " << dendl; |
4313 | for (const auto &dn : to_trim) { | |
4314 | trim_dentry(dn); | |
4315 | } | |
4316 | to_trim.clear(); | |
7c673cae | 4317 | |
b32b8144 | 4318 | caps_size = s->caps.size(); |
11fdf7f2 | 4319 | if (caps_size > (size_t)max) |
7c673cae FG |
4320 | _invalidate_kernel_dcache(); |
4321 | } | |
4322 | ||
4323 | void Client::force_session_readonly(MetaSession *s) | |
4324 | { | |
4325 | s->readonly = true; | |
4326 | for (xlist<Cap*>::iterator p = s->caps.begin(); !p.end(); ++p) { | |
11fdf7f2 TL |
4327 | auto &in = (*p)->inode; |
4328 | if (in.caps_wanted() & CEPH_CAP_FILE_WR) | |
4329 | signal_cond_list(in.waitfor_caps); | |
7c673cae FG |
4330 | } |
4331 | } | |
4332 | ||
7c673cae FG |
4333 | int Client::mark_caps_flushing(Inode *in, ceph_tid_t* ptid) |
4334 | { | |
4335 | MetaSession *session = in->auth_cap->session; | |
4336 | ||
4337 | int flushing = in->dirty_caps; | |
11fdf7f2 | 4338 | ceph_assert(flushing); |
7c673cae FG |
4339 | |
4340 | ceph_tid_t flush_tid = ++last_flush_tid; | |
4341 | in->flushing_cap_tids[flush_tid] = flushing; | |
4342 | ||
4343 | if (!in->flushing_caps) { | |
11fdf7f2 | 4344 | ldout(cct, 10) << __func__ << " " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4345 | num_flushing_caps++; |
4346 | } else { | |
11fdf7f2 | 4347 | ldout(cct, 10) << __func__ << " (more) " << ccap_string(flushing) << " " << *in << dendl; |
7c673cae FG |
4348 | } |
4349 | ||
4350 | in->flushing_caps |= flushing; | |
28e407b8 | 4351 | in->mark_caps_clean(); |
7c673cae FG |
4352 | |
4353 | if (!in->flushing_cap_item.is_on_list()) | |
4354 | session->flushing_caps.push_back(&in->flushing_cap_item); | |
4355 | session->flushing_caps_tids.insert(flush_tid); | |
4356 | ||
4357 | *ptid = flush_tid; | |
4358 | return flushing; | |
4359 | } | |
4360 | ||
4361 | void Client::adjust_session_flushing_caps(Inode *in, MetaSession *old_s, MetaSession *new_s) | |
4362 | { | |
4363 | for (auto &p : in->cap_snaps) { | |
4364 | CapSnap &capsnap = p.second; | |
4365 | if (capsnap.flush_tid > 0) { | |
4366 | old_s->flushing_caps_tids.erase(capsnap.flush_tid); | |
4367 | new_s->flushing_caps_tids.insert(capsnap.flush_tid); | |
4368 | } | |
4369 | } | |
4370 | for (map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
4371 | it != in->flushing_cap_tids.end(); | |
4372 | ++it) { | |
4373 | old_s->flushing_caps_tids.erase(it->first); | |
4374 | new_s->flushing_caps_tids.insert(it->first); | |
4375 | } | |
4376 | new_s->flushing_caps.push_back(&in->flushing_cap_item); | |
4377 | } | |
4378 | ||
4379 | /* | |
4380 | * Flush all caps back to the MDS. Because the callers generally wait on the | |
4381 | * result of this function (syncfs and umount cases), we set | |
4382 | * CHECK_CAPS_SYNCHRONOUS on the last check_caps call. | |
4383 | */ | |
4384 | void Client::flush_caps_sync() | |
4385 | { | |
4386 | ldout(cct, 10) << __func__ << dendl; | |
28e407b8 | 4387 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
4388 | while (!p.end()) { |
4389 | unsigned flags = CHECK_CAPS_NODELAY; | |
4390 | Inode *in = *p; | |
4391 | ||
4392 | ++p; | |
28e407b8 AA |
4393 | delayed_list.pop_front(); |
4394 | if (p.end() && dirty_list.empty()) | |
7c673cae FG |
4395 | flags |= CHECK_CAPS_SYNCHRONOUS; |
4396 | check_caps(in, flags); | |
4397 | } | |
4398 | ||
4399 | // other caps, too | |
28e407b8 | 4400 | p = dirty_list.begin(); |
7c673cae FG |
4401 | while (!p.end()) { |
4402 | unsigned flags = CHECK_CAPS_NODELAY; | |
4403 | Inode *in = *p; | |
4404 | ||
4405 | ++p; | |
4406 | if (p.end()) | |
4407 | flags |= CHECK_CAPS_SYNCHRONOUS; | |
4408 | check_caps(in, flags); | |
4409 | } | |
4410 | } | |
4411 | ||
7c673cae FG |
4412 | void Client::wait_sync_caps(Inode *in, ceph_tid_t want) |
4413 | { | |
4414 | while (in->flushing_caps) { | |
4415 | map<ceph_tid_t, int>::iterator it = in->flushing_cap_tids.begin(); | |
11fdf7f2 | 4416 | ceph_assert(it != in->flushing_cap_tids.end()); |
7c673cae FG |
4417 | if (it->first > want) |
4418 | break; | |
11fdf7f2 | 4419 | ldout(cct, 10) << __func__ << " on " << *in << " flushing " |
7c673cae FG |
4420 | << ccap_string(it->second) << " want " << want |
4421 | << " last " << it->first << dendl; | |
4422 | wait_on_list(in->waitfor_caps); | |
4423 | } | |
4424 | } | |
4425 | ||
4426 | void Client::wait_sync_caps(ceph_tid_t want) | |
4427 | { | |
4428 | retry: | |
11fdf7f2 | 4429 | ldout(cct, 10) << __func__ << " want " << want << " (last is " << last_flush_tid << ", " |
7c673cae | 4430 | << num_flushing_caps << " total flushing)" << dendl; |
11fdf7f2 TL |
4431 | for (auto &p : mds_sessions) { |
4432 | MetaSession *s = &p.second; | |
7c673cae FG |
4433 | if (s->flushing_caps_tids.empty()) |
4434 | continue; | |
4435 | ceph_tid_t oldest_tid = *s->flushing_caps_tids.begin(); | |
4436 | if (oldest_tid <= want) { | |
11fdf7f2 | 4437 | ldout(cct, 10) << " waiting on mds." << p.first << " tid " << oldest_tid |
7c673cae | 4438 | << " (want " << want << ")" << dendl; |
9f95a23c TL |
4439 | std::unique_lock l{client_lock, std::adopt_lock}; |
4440 | sync_cond.wait(l); | |
4441 | l.release(); | |
7c673cae FG |
4442 | goto retry; |
4443 | } | |
4444 | } | |
4445 | } | |
4446 | ||
eafe8130 TL |
4447 | void Client::kick_flushing_caps(Inode *in, MetaSession *session) |
4448 | { | |
4449 | in->flags &= ~I_KICK_FLUSH; | |
4450 | ||
4451 | Cap *cap = in->auth_cap; | |
4452 | ceph_assert(cap->session == session); | |
4453 | ||
4454 | ceph_tid_t last_snap_flush = 0; | |
4455 | for (auto p = in->flushing_cap_tids.rbegin(); | |
4456 | p != in->flushing_cap_tids.rend(); | |
4457 | ++p) { | |
4458 | if (!p->second) { | |
4459 | last_snap_flush = p->first; | |
4460 | break; | |
4461 | } | |
4462 | } | |
4463 | ||
4464 | int wanted = in->caps_wanted(); | |
4465 | int used = get_caps_used(in) | in->caps_dirty(); | |
4466 | auto it = in->cap_snaps.begin(); | |
4467 | for (auto& p : in->flushing_cap_tids) { | |
4468 | if (p.second) { | |
4469 | int msg_flags = p.first < last_snap_flush ? MClientCaps::FLAG_PENDING_CAPSNAP : 0; | |
4470 | send_cap(in, session, cap, msg_flags, used, wanted, (cap->issued | cap->implemented), | |
4471 | p.second, p.first); | |
4472 | } else { | |
4473 | ceph_assert(it != in->cap_snaps.end()); | |
4474 | ceph_assert(it->second.flush_tid == p.first); | |
4475 | send_flush_snap(in, session, it->first, it->second); | |
4476 | ++it; | |
4477 | } | |
4478 | } | |
4479 | } | |
4480 | ||
7c673cae FG |
4481 | void Client::kick_flushing_caps(MetaSession *session) |
4482 | { | |
4483 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 4484 | ldout(cct, 10) << __func__ << " mds." << mds << dendl; |
7c673cae FG |
4485 | |
4486 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { | |
4487 | Inode *in = *p; | |
eafe8130 TL |
4488 | if (in->flags & I_KICK_FLUSH) { |
4489 | ldout(cct, 20) << " reflushing caps on " << *in << " to mds." << mds << dendl; | |
4490 | kick_flushing_caps(in, session); | |
4491 | } | |
7c673cae | 4492 | } |
7c673cae FG |
4493 | } |
4494 | ||
4495 | void Client::early_kick_flushing_caps(MetaSession *session) | |
4496 | { | |
7c673cae FG |
4497 | for (xlist<Inode*>::iterator p = session->flushing_caps.begin(); !p.end(); ++p) { |
4498 | Inode *in = *p; | |
11fdf7f2 TL |
4499 | Cap *cap = in->auth_cap; |
4500 | ceph_assert(cap); | |
7c673cae FG |
4501 | |
4502 | // if flushing caps were revoked, we re-send the cap flush in client reconnect | |
4503 | // stage. This guarantees that MDS processes the cap flush message before issuing | |
4504 | // the flushing caps to other client. | |
eafe8130 TL |
4505 | if ((in->flushing_caps & in->auth_cap->issued) == in->flushing_caps) { |
4506 | in->flags |= I_KICK_FLUSH; | |
7c673cae | 4507 | continue; |
eafe8130 | 4508 | } |
7c673cae FG |
4509 | |
4510 | ldout(cct, 20) << " reflushing caps (early_kick) on " << *in | |
4511 | << " to mds." << session->mds_num << dendl; | |
11fdf7f2 TL |
4512 | // send_reconnect() also will reset these sequence numbers. make sure |
4513 | // sequence numbers in cap flush message match later reconnect message. | |
4514 | cap->seq = 0; | |
4515 | cap->issue_seq = 0; | |
4516 | cap->mseq = 0; | |
4517 | cap->issued = cap->implemented; | |
4518 | ||
eafe8130 | 4519 | kick_flushing_caps(in, session); |
7c673cae FG |
4520 | } |
4521 | } | |
4522 | ||
7c673cae FG |
4523 | void SnapRealm::build_snap_context() |
4524 | { | |
4525 | set<snapid_t> snaps; | |
4526 | snapid_t max_seq = seq; | |
4527 | ||
4528 | // start with prior_parents? | |
4529 | for (unsigned i=0; i<prior_parent_snaps.size(); i++) | |
4530 | snaps.insert(prior_parent_snaps[i]); | |
4531 | ||
4532 | // current parent's snaps | |
4533 | if (pparent) { | |
4534 | const SnapContext& psnapc = pparent->get_snap_context(); | |
4535 | for (unsigned i=0; i<psnapc.snaps.size(); i++) | |
4536 | if (psnapc.snaps[i] >= parent_since) | |
4537 | snaps.insert(psnapc.snaps[i]); | |
4538 | if (psnapc.seq > max_seq) | |
4539 | max_seq = psnapc.seq; | |
4540 | } | |
4541 | ||
4542 | // my snaps | |
4543 | for (unsigned i=0; i<my_snaps.size(); i++) | |
4544 | snaps.insert(my_snaps[i]); | |
4545 | ||
4546 | // ok! | |
4547 | cached_snap_context.seq = max_seq; | |
4548 | cached_snap_context.snaps.resize(0); | |
4549 | cached_snap_context.snaps.reserve(snaps.size()); | |
4550 | for (set<snapid_t>::reverse_iterator p = snaps.rbegin(); p != snaps.rend(); ++p) | |
4551 | cached_snap_context.snaps.push_back(*p); | |
4552 | } | |
4553 | ||
4554 | void Client::invalidate_snaprealm_and_children(SnapRealm *realm) | |
4555 | { | |
4556 | list<SnapRealm*> q; | |
4557 | q.push_back(realm); | |
4558 | ||
4559 | while (!q.empty()) { | |
4560 | realm = q.front(); | |
4561 | q.pop_front(); | |
4562 | ||
11fdf7f2 | 4563 | ldout(cct, 10) << __func__ << " " << *realm << dendl; |
7c673cae FG |
4564 | realm->invalidate_cache(); |
4565 | ||
4566 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4567 | p != realm->pchildren.end(); | |
4568 | ++p) | |
4569 | q.push_back(*p); | |
4570 | } | |
4571 | } | |
4572 | ||
4573 | SnapRealm *Client::get_snap_realm(inodeno_t r) | |
4574 | { | |
4575 | SnapRealm *realm = snap_realms[r]; | |
4576 | if (!realm) | |
4577 | snap_realms[r] = realm = new SnapRealm(r); | |
11fdf7f2 | 4578 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4579 | realm->nref++; |
4580 | return realm; | |
4581 | } | |
4582 | ||
4583 | SnapRealm *Client::get_snap_realm_maybe(inodeno_t r) | |
4584 | { | |
4585 | if (snap_realms.count(r) == 0) { | |
11fdf7f2 | 4586 | ldout(cct, 20) << __func__ << " " << r << " fail" << dendl; |
7c673cae FG |
4587 | return NULL; |
4588 | } | |
4589 | SnapRealm *realm = snap_realms[r]; | |
11fdf7f2 | 4590 | ldout(cct, 20) << __func__ << " " << r << " " << realm << " " << realm->nref << " -> " << (realm->nref + 1) << dendl; |
7c673cae FG |
4591 | realm->nref++; |
4592 | return realm; | |
4593 | } | |
4594 | ||
4595 | void Client::put_snap_realm(SnapRealm *realm) | |
4596 | { | |
11fdf7f2 | 4597 | ldout(cct, 20) << __func__ << " " << realm->ino << " " << realm |
7c673cae FG |
4598 | << " " << realm->nref << " -> " << (realm->nref - 1) << dendl; |
4599 | if (--realm->nref == 0) { | |
4600 | snap_realms.erase(realm->ino); | |
4601 | if (realm->pparent) { | |
4602 | realm->pparent->pchildren.erase(realm); | |
4603 | put_snap_realm(realm->pparent); | |
4604 | } | |
4605 | delete realm; | |
4606 | } | |
4607 | } | |
4608 | ||
4609 | bool Client::adjust_realm_parent(SnapRealm *realm, inodeno_t parent) | |
4610 | { | |
4611 | if (realm->parent != parent) { | |
11fdf7f2 | 4612 | ldout(cct, 10) << __func__ << " " << *realm |
7c673cae FG |
4613 | << " " << realm->parent << " -> " << parent << dendl; |
4614 | realm->parent = parent; | |
4615 | if (realm->pparent) { | |
4616 | realm->pparent->pchildren.erase(realm); | |
4617 | put_snap_realm(realm->pparent); | |
4618 | } | |
4619 | realm->pparent = get_snap_realm(parent); | |
4620 | realm->pparent->pchildren.insert(realm); | |
4621 | return true; | |
4622 | } | |
4623 | return false; | |
4624 | } | |
4625 | ||
4626 | static bool has_new_snaps(const SnapContext& old_snapc, | |
4627 | const SnapContext& new_snapc) | |
4628 | { | |
4629 | return !new_snapc.snaps.empty() && new_snapc.snaps[0] > old_snapc.seq; | |
4630 | } | |
4631 | ||
4632 | ||
11fdf7f2 | 4633 | void Client::update_snap_trace(const bufferlist& bl, SnapRealm **realm_ret, bool flush) |
7c673cae FG |
4634 | { |
4635 | SnapRealm *first_realm = NULL; | |
11fdf7f2 | 4636 | ldout(cct, 10) << __func__ << " len " << bl.length() << dendl; |
7c673cae FG |
4637 | |
4638 | map<SnapRealm*, SnapContext> dirty_realms; | |
4639 | ||
11fdf7f2 | 4640 | auto p = bl.cbegin(); |
7c673cae FG |
4641 | while (!p.end()) { |
4642 | SnapRealmInfo info; | |
11fdf7f2 | 4643 | decode(info, p); |
7c673cae FG |
4644 | SnapRealm *realm = get_snap_realm(info.ino()); |
4645 | ||
4646 | bool invalidate = false; | |
4647 | ||
4648 | if (info.seq() > realm->seq) { | |
11fdf7f2 | 4649 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() << " > " << realm->seq |
7c673cae FG |
4650 | << dendl; |
4651 | ||
4652 | if (flush) { | |
4653 | // writeback any dirty caps _before_ updating snap list (i.e. with old snap info) | |
4654 | // flush me + children | |
4655 | list<SnapRealm*> q; | |
4656 | q.push_back(realm); | |
4657 | while (!q.empty()) { | |
4658 | SnapRealm *realm = q.front(); | |
4659 | q.pop_front(); | |
4660 | ||
4661 | for (set<SnapRealm*>::iterator p = realm->pchildren.begin(); | |
4662 | p != realm->pchildren.end(); | |
4663 | ++p) | |
4664 | q.push_back(*p); | |
4665 | ||
4666 | if (dirty_realms.count(realm) == 0) { | |
4667 | realm->nref++; | |
4668 | dirty_realms[realm] = realm->get_snap_context(); | |
4669 | } | |
4670 | } | |
4671 | } | |
4672 | ||
4673 | // update | |
4674 | realm->seq = info.seq(); | |
4675 | realm->created = info.created(); | |
4676 | realm->parent_since = info.parent_since(); | |
4677 | realm->prior_parent_snaps = info.prior_parent_snaps; | |
4678 | realm->my_snaps = info.my_snaps; | |
4679 | invalidate = true; | |
4680 | } | |
4681 | ||
4682 | // _always_ verify parent | |
4683 | if (adjust_realm_parent(realm, info.parent())) | |
4684 | invalidate = true; | |
4685 | ||
4686 | if (invalidate) { | |
4687 | invalidate_snaprealm_and_children(realm); | |
11fdf7f2 | 4688 | ldout(cct, 15) << __func__ << " " << *realm << " self|parent updated" << dendl; |
7c673cae FG |
4689 | ldout(cct, 15) << " snapc " << realm->get_snap_context() << dendl; |
4690 | } else { | |
11fdf7f2 | 4691 | ldout(cct, 10) << __func__ << " " << *realm << " seq " << info.seq() |
7c673cae FG |
4692 | << " <= " << realm->seq << " and same parent, SKIPPING" << dendl; |
4693 | } | |
4694 | ||
4695 | if (!first_realm) | |
4696 | first_realm = realm; | |
4697 | else | |
4698 | put_snap_realm(realm); | |
4699 | } | |
4700 | ||
4701 | for (map<SnapRealm*, SnapContext>::iterator q = dirty_realms.begin(); | |
4702 | q != dirty_realms.end(); | |
4703 | ++q) { | |
4704 | SnapRealm *realm = q->first; | |
4705 | // if there are new snaps ? | |
4706 | if (has_new_snaps(q->second, realm->get_snap_context())) { | |
4707 | ldout(cct, 10) << " flushing caps on " << *realm << dendl; | |
4708 | xlist<Inode*>::iterator r = realm->inodes_with_caps.begin(); | |
4709 | while (!r.end()) { | |
4710 | Inode *in = *r; | |
4711 | ++r; | |
4712 | queue_cap_snap(in, q->second); | |
4713 | } | |
4714 | } else { | |
4715 | ldout(cct, 10) << " no new snap on " << *realm << dendl; | |
4716 | } | |
4717 | put_snap_realm(realm); | |
4718 | } | |
4719 | ||
4720 | if (realm_ret) | |
4721 | *realm_ret = first_realm; | |
4722 | else | |
4723 | put_snap_realm(first_realm); | |
4724 | } | |
4725 | ||
11fdf7f2 | 4726 | void Client::handle_snap(const MConstRef<MClientSnap>& m) |
7c673cae | 4727 | { |
11fdf7f2 | 4728 | ldout(cct, 10) << __func__ << " " << *m << dendl; |
7c673cae FG |
4729 | mds_rank_t mds = mds_rank_t(m->get_source().num()); |
4730 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
4731 | if (!session) { | |
7c673cae FG |
4732 | return; |
4733 | } | |
4734 | ||
4735 | got_mds_push(session); | |
4736 | ||
4737 | map<Inode*, SnapContext> to_move; | |
4738 | SnapRealm *realm = 0; | |
4739 | ||
4740 | if (m->head.op == CEPH_SNAP_OP_SPLIT) { | |
11fdf7f2 | 4741 | ceph_assert(m->head.split); |
7c673cae | 4742 | SnapRealmInfo info; |
11fdf7f2 TL |
4743 | auto p = m->bl.cbegin(); |
4744 | decode(info, p); | |
4745 | ceph_assert(info.ino() == m->head.split); | |
7c673cae FG |
4746 | |
4747 | // flush, then move, ino's. | |
4748 | realm = get_snap_realm(info.ino()); | |
4749 | ldout(cct, 10) << " splitting off " << *realm << dendl; | |
11fdf7f2 TL |
4750 | for (auto& ino : m->split_inos) { |
4751 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
7c673cae FG |
4752 | if (inode_map.count(vino)) { |
4753 | Inode *in = inode_map[vino]; | |
4754 | if (!in->snaprealm || in->snaprealm == realm) | |
4755 | continue; | |
4756 | if (in->snaprealm->created > info.created()) { | |
4757 | ldout(cct, 10) << " NOT moving " << *in << " from _newer_ realm " | |
4758 | << *in->snaprealm << dendl; | |
4759 | continue; | |
4760 | } | |
4761 | ldout(cct, 10) << " moving " << *in << " from " << *in->snaprealm << dendl; | |
4762 | ||
4763 | ||
4764 | in->snaprealm_item.remove_myself(); | |
4765 | to_move[in] = in->snaprealm->get_snap_context(); | |
4766 | put_snap_realm(in->snaprealm); | |
4767 | } | |
4768 | } | |
4769 | ||
4770 | // move child snaprealms, too | |
11fdf7f2 TL |
4771 | for (auto& child_realm : m->split_realms) { |
4772 | ldout(cct, 10) << "adjusting snaprealm " << child_realm << " parent" << dendl; | |
4773 | SnapRealm *child = get_snap_realm_maybe(child_realm); | |
7c673cae FG |
4774 | if (!child) |
4775 | continue; | |
4776 | adjust_realm_parent(child, realm->ino); | |
4777 | put_snap_realm(child); | |
4778 | } | |
4779 | } | |
4780 | ||
4781 | update_snap_trace(m->bl, NULL, m->head.op != CEPH_SNAP_OP_DESTROY); | |
4782 | ||
4783 | if (realm) { | |
4784 | for (auto p = to_move.begin(); p != to_move.end(); ++p) { | |
4785 | Inode *in = p->first; | |
4786 | in->snaprealm = realm; | |
4787 | realm->inodes_with_caps.push_back(&in->snaprealm_item); | |
4788 | realm->nref++; | |
4789 | // queue for snap writeback | |
4790 | if (has_new_snaps(p->second, realm->get_snap_context())) | |
4791 | queue_cap_snap(in, p->second); | |
4792 | } | |
4793 | put_snap_realm(realm); | |
4794 | } | |
7c673cae FG |
4795 | } |
4796 | ||
11fdf7f2 | 4797 | void Client::handle_quota(const MConstRef<MClientQuota>& m) |
7c673cae FG |
4798 | { |
4799 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
4800 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
4801 | if (!session) { | |
7c673cae FG |
4802 | return; |
4803 | } | |
4804 | ||
4805 | got_mds_push(session); | |
4806 | ||
11fdf7f2 | 4807 | ldout(cct, 10) << __func__ << " " << *m << " from mds." << mds << dendl; |
7c673cae FG |
4808 | |
4809 | vinodeno_t vino(m->ino, CEPH_NOSNAP); | |
4810 | if (inode_map.count(vino)) { | |
4811 | Inode *in = NULL; | |
4812 | in = inode_map[vino]; | |
4813 | ||
4814 | if (in) { | |
4815 | in->quota = m->quota; | |
4816 | in->rstat = m->rstat; | |
4817 | } | |
4818 | } | |
7c673cae FG |
4819 | } |
4820 | ||
11fdf7f2 | 4821 | void Client::handle_caps(const MConstRef<MClientCaps>& m) |
7c673cae FG |
4822 | { |
4823 | mds_rank_t mds = mds_rank_t(m->get_source().num()); | |
4824 | MetaSession *session = _get_mds_session(mds, m->get_connection().get()); | |
4825 | if (!session) { | |
7c673cae FG |
4826 | return; |
4827 | } | |
4828 | ||
4829 | if (m->osd_epoch_barrier && !objecter->have_map(m->osd_epoch_barrier)) { | |
4830 | // Pause RADOS operations until we see the required epoch | |
4831 | objecter->set_epoch_barrier(m->osd_epoch_barrier); | |
4832 | } | |
4833 | ||
4834 | if (m->osd_epoch_barrier > cap_epoch_barrier) { | |
4835 | // Record the barrier so that we will transmit it to MDS when releasing | |
4836 | set_cap_epoch_barrier(m->osd_epoch_barrier); | |
4837 | } | |
4838 | ||
4839 | got_mds_push(session); | |
4840 | ||
11fdf7f2 | 4841 | Inode *in; |
7c673cae | 4842 | vinodeno_t vino(m->get_ino(), CEPH_NOSNAP); |
11fdf7f2 TL |
4843 | if (auto it = inode_map.find(vino); it != inode_map.end()) { |
4844 | in = it->second; | |
4845 | } else { | |
7c673cae | 4846 | if (m->get_op() == CEPH_CAP_OP_IMPORT) { |
11fdf7f2 | 4847 | ldout(cct, 5) << __func__ << " don't have vino " << vino << " on IMPORT, immediately releasing" << dendl; |
7c673cae FG |
4848 | session->enqueue_cap_release( |
4849 | m->get_ino(), | |
4850 | m->get_cap_id(), | |
4851 | m->get_seq(), | |
4852 | m->get_mseq(), | |
4853 | cap_epoch_barrier); | |
4854 | } else { | |
11fdf7f2 | 4855 | ldout(cct, 5) << __func__ << " don't have vino " << vino << ", dropping" << dendl; |
7c673cae | 4856 | } |
7c673cae FG |
4857 | |
4858 | // in case the mds is waiting on e.g. a revocation | |
4859 | flush_cap_releases(); | |
4860 | return; | |
4861 | } | |
4862 | ||
4863 | switch (m->get_op()) { | |
11fdf7f2 TL |
4864 | case CEPH_CAP_OP_EXPORT: return handle_cap_export(session, in, m); |
4865 | case CEPH_CAP_OP_FLUSHSNAP_ACK: return handle_cap_flushsnap_ack(session, in, m); | |
4866 | case CEPH_CAP_OP_IMPORT: /* no return */ handle_cap_import(session, in, m); | |
7c673cae FG |
4867 | } |
4868 | ||
11fdf7f2 TL |
4869 | if (auto it = in->caps.find(mds); it != in->caps.end()) { |
4870 | Cap &cap = in->caps.at(mds); | |
7c673cae | 4871 | |
11fdf7f2 TL |
4872 | switch (m->get_op()) { |
4873 | case CEPH_CAP_OP_TRUNC: return handle_cap_trunc(session, in, m); | |
4874 | case CEPH_CAP_OP_IMPORT: | |
4875 | case CEPH_CAP_OP_REVOKE: | |
4876 | case CEPH_CAP_OP_GRANT: return handle_cap_grant(session, in, &cap, m); | |
4877 | case CEPH_CAP_OP_FLUSH_ACK: return handle_cap_flush_ack(session, in, &cap, m); | |
4878 | } | |
4879 | } else { | |
4880 | ldout(cct, 5) << __func__ << " don't have " << *in << " cap on mds." << mds << dendl; | |
4881 | return; | |
7c673cae FG |
4882 | } |
4883 | } | |
4884 | ||
11fdf7f2 | 4885 | void Client::handle_cap_import(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
4886 | { |
4887 | mds_rank_t mds = session->mds_num; | |
4888 | ||
11fdf7f2 | 4889 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
4890 | << " IMPORT from mds." << mds << dendl; |
4891 | ||
4892 | const mds_rank_t peer_mds = mds_rank_t(m->peer.mds); | |
4893 | Cap *cap = NULL; | |
4894 | UserPerm cap_perms; | |
11fdf7f2 TL |
4895 | if (auto it = in->caps.find(peer_mds); m->peer.cap_id && it != in->caps.end()) { |
4896 | cap = &it->second; | |
4897 | cap_perms = cap->latest_perms; | |
7c673cae FG |
4898 | } |
4899 | ||
4900 | // add/update it | |
4901 | SnapRealm *realm = NULL; | |
4902 | update_snap_trace(m->snapbl, &realm); | |
4903 | ||
4904 | add_update_cap(in, session, m->get_cap_id(), | |
a8e16298 TL |
4905 | m->get_caps(), m->get_wanted(), m->get_seq(), m->get_mseq(), |
4906 | m->get_realm(), CEPH_CAP_FLAG_AUTH, cap_perms); | |
7c673cae FG |
4907 | |
4908 | if (cap && cap->cap_id == m->peer.cap_id) { | |
4909 | remove_cap(cap, (m->peer.flags & CEPH_CAP_FLAG_RELEASE)); | |
4910 | } | |
4911 | ||
4912 | if (realm) | |
4913 | put_snap_realm(realm); | |
4914 | ||
eafe8130 | 4915 | if (in->auth_cap && in->auth_cap->session == session) { |
7c673cae | 4916 | // reflush any/all caps (if we are now the auth_cap) |
eafe8130 | 4917 | kick_flushing_caps(in, session); |
7c673cae FG |
4918 | } |
4919 | } | |
4920 | ||
11fdf7f2 | 4921 | void Client::handle_cap_export(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
4922 | { |
4923 | mds_rank_t mds = session->mds_num; | |
4924 | ||
11fdf7f2 | 4925 | ldout(cct, 5) << __func__ << " ino " << m->get_ino() << " mseq " << m->get_mseq() |
7c673cae FG |
4926 | << " EXPORT from mds." << mds << dendl; |
4927 | ||
11fdf7f2 TL |
4928 | auto it = in->caps.find(mds); |
4929 | if (it != in->caps.end()) { | |
4930 | Cap &cap = it->second; | |
4931 | if (cap.cap_id == m->get_cap_id()) { | |
4932 | if (m->peer.cap_id) { | |
4933 | const auto peer_mds = mds_rank_t(m->peer.mds); | |
4934 | MetaSession *tsession = _get_or_open_mds_session(peer_mds); | |
4935 | auto it = in->caps.find(peer_mds); | |
4936 | if (it != in->caps.end()) { | |
4937 | Cap &tcap = it->second; | |
4938 | if (tcap.cap_id == m->peer.cap_id && | |
4939 | ceph_seq_cmp(tcap.seq, m->peer.seq) < 0) { | |
4940 | tcap.cap_id = m->peer.cap_id; | |
4941 | tcap.seq = m->peer.seq - 1; | |
4942 | tcap.issue_seq = tcap.seq; | |
4943 | tcap.issued |= cap.issued; | |
4944 | tcap.implemented |= cap.issued; | |
4945 | if (&cap == in->auth_cap) | |
4946 | in->auth_cap = &tcap; | |
4947 | if (in->auth_cap == &tcap && in->flushing_cap_item.is_on_list()) | |
4948 | adjust_session_flushing_caps(in, session, tsession); | |
4949 | } | |
4950 | } else { | |
4951 | add_update_cap(in, tsession, m->peer.cap_id, cap.issued, 0, | |
4952 | m->peer.seq - 1, m->peer.mseq, (uint64_t)-1, | |
4953 | &cap == in->auth_cap ? CEPH_CAP_FLAG_AUTH : 0, | |
4954 | cap.latest_perms); | |
4955 | } | |
7c673cae | 4956 | } else { |
11fdf7f2 TL |
4957 | if (cap.wanted | cap.issued) |
4958 | in->flags |= I_CAP_DROPPED; | |
7c673cae | 4959 | } |
7c673cae | 4960 | |
11fdf7f2 TL |
4961 | remove_cap(&cap, false); |
4962 | } | |
7c673cae | 4963 | } |
7c673cae FG |
4964 | } |
4965 | ||
11fdf7f2 | 4966 | void Client::handle_cap_trunc(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae FG |
4967 | { |
4968 | mds_rank_t mds = session->mds_num; | |
11fdf7f2 | 4969 | ceph_assert(in->caps.count(mds)); |
7c673cae | 4970 | |
11fdf7f2 | 4971 | ldout(cct, 10) << __func__ << " on ino " << *in |
7c673cae FG |
4972 | << " size " << in->size << " -> " << m->get_size() |
4973 | << dendl; | |
4974 | ||
1adf2230 AA |
4975 | int issued; |
4976 | in->caps_issued(&issued); | |
4977 | issued |= in->caps_dirty(); | |
4978 | update_inode_file_size(in, issued, m->get_size(), | |
4979 | m->get_truncate_seq(), m->get_truncate_size()); | |
7c673cae FG |
4980 | } |
4981 | ||
11fdf7f2 | 4982 | void Client::handle_cap_flush_ack(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
4983 | { |
4984 | ceph_tid_t flush_ack_tid = m->get_client_tid(); | |
4985 | int dirty = m->get_dirty(); | |
4986 | int cleaned = 0; | |
4987 | int flushed = 0; | |
4988 | ||
11fdf7f2 TL |
4989 | auto it = in->flushing_cap_tids.begin(); |
4990 | if (it->first < flush_ack_tid) { | |
4991 | ldout(cct, 0) << __func__ << " mds." << session->mds_num | |
4992 | << " got unexpected flush ack tid " << flush_ack_tid | |
4993 | << " expected is " << it->first << dendl; | |
4994 | } | |
4995 | for (; it != in->flushing_cap_tids.end(); ) { | |
eafe8130 TL |
4996 | if (!it->second) { |
4997 | // cap snap | |
4998 | ++it; | |
4999 | continue; | |
5000 | } | |
7c673cae FG |
5001 | if (it->first == flush_ack_tid) |
5002 | cleaned = it->second; | |
5003 | if (it->first <= flush_ack_tid) { | |
5004 | session->flushing_caps_tids.erase(it->first); | |
5005 | in->flushing_cap_tids.erase(it++); | |
5006 | ++flushed; | |
5007 | continue; | |
5008 | } | |
5009 | cleaned &= ~it->second; | |
5010 | if (!cleaned) | |
5011 | break; | |
5012 | ++it; | |
5013 | } | |
5014 | ||
11fdf7f2 | 5015 | ldout(cct, 5) << __func__ << " mds." << session->mds_num |
7c673cae FG |
5016 | << " cleaned " << ccap_string(cleaned) << " on " << *in |
5017 | << " with " << ccap_string(dirty) << dendl; | |
5018 | ||
5019 | if (flushed) { | |
5020 | signal_cond_list(in->waitfor_caps); | |
5021 | if (session->flushing_caps_tids.empty() || | |
5022 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5023 | sync_cond.notify_all(); |
7c673cae FG |
5024 | } |
5025 | ||
5026 | if (!dirty) { | |
5027 | in->cap_dirtier_uid = -1; | |
5028 | in->cap_dirtier_gid = -1; | |
5029 | } | |
5030 | ||
5031 | if (!cleaned) { | |
5032 | ldout(cct, 10) << " tid " << m->get_client_tid() << " != any cap bit tids" << dendl; | |
5033 | } else { | |
5034 | if (in->flushing_caps) { | |
5035 | ldout(cct, 5) << " flushing_caps " << ccap_string(in->flushing_caps) | |
5036 | << " -> " << ccap_string(in->flushing_caps & ~cleaned) << dendl; | |
5037 | in->flushing_caps &= ~cleaned; | |
5038 | if (in->flushing_caps == 0) { | |
5039 | ldout(cct, 10) << " " << *in << " !flushing" << dendl; | |
5040 | num_flushing_caps--; | |
eafe8130 | 5041 | if (in->flushing_cap_tids.empty()) |
7c673cae FG |
5042 | in->flushing_cap_item.remove_myself(); |
5043 | } | |
5044 | if (!in->caps_dirty()) | |
5045 | put_inode(in); | |
5046 | } | |
5047 | } | |
7c673cae FG |
5048 | } |
5049 | ||
5050 | ||
11fdf7f2 | 5051 | void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, const MConstRef<MClientCaps>& m) |
7c673cae | 5052 | { |
eafe8130 | 5053 | ceph_tid_t flush_ack_tid = m->get_client_tid(); |
7c673cae | 5054 | mds_rank_t mds = session->mds_num; |
11fdf7f2 | 5055 | ceph_assert(in->caps.count(mds)); |
7c673cae FG |
5056 | snapid_t follows = m->get_snap_follows(); |
5057 | ||
11fdf7f2 TL |
5058 | if (auto it = in->cap_snaps.find(follows); it != in->cap_snaps.end()) { |
5059 | auto& capsnap = it->second; | |
eafe8130 TL |
5060 | if (flush_ack_tid != capsnap.flush_tid) { |
5061 | ldout(cct, 10) << " tid " << flush_ack_tid << " != " << capsnap.flush_tid << dendl; | |
7c673cae | 5062 | } else { |
eafe8130 | 5063 | InodeRef tmp_ref(in); |
11fdf7f2 | 5064 | ldout(cct, 5) << __func__ << " mds." << mds << " flushed snap follows " << follows |
7c673cae | 5065 | << " on " << *in << dendl; |
7c673cae | 5066 | session->flushing_caps_tids.erase(capsnap.flush_tid); |
eafe8130 TL |
5067 | in->flushing_cap_tids.erase(capsnap.flush_tid); |
5068 | if (in->flushing_caps == 0 && in->flushing_cap_tids.empty()) | |
5069 | in->flushing_cap_item.remove_myself(); | |
11fdf7f2 | 5070 | in->cap_snaps.erase(it); |
eafe8130 TL |
5071 | |
5072 | signal_cond_list(in->waitfor_caps); | |
5073 | if (session->flushing_caps_tids.empty() || | |
5074 | *session->flushing_caps_tids.begin() > flush_ack_tid) | |
9f95a23c | 5075 | sync_cond.notify_all(); |
7c673cae FG |
5076 | } |
5077 | } else { | |
11fdf7f2 | 5078 | ldout(cct, 5) << __func__ << " DUP(?) mds." << mds << " flushed snap follows " << follows |
7c673cae FG |
5079 | << " on " << *in << dendl; |
5080 | // we may not have it if we send multiple FLUSHSNAP requests and (got multiple FLUSHEDSNAPs back) | |
5081 | } | |
7c673cae FG |
5082 | } |
5083 | ||
5084 | class C_Client_DentryInvalidate : public Context { | |
5085 | private: | |
5086 | Client *client; | |
5087 | vinodeno_t dirino; | |
5088 | vinodeno_t ino; | |
5089 | string name; | |
5090 | public: | |
5091 | C_Client_DentryInvalidate(Client *c, Dentry *dn, bool del) : | |
5092 | client(c), name(dn->name) { | |
5093 | if (client->use_faked_inos()) { | |
5094 | dirino.ino = dn->dir->parent_inode->faked_ino; | |
5095 | if (del) | |
5096 | ino.ino = dn->inode->faked_ino; | |
5097 | } else { | |
5098 | dirino = dn->dir->parent_inode->vino(); | |
5099 | if (del) | |
5100 | ino = dn->inode->vino(); | |
5101 | } | |
5102 | if (!del) | |
5103 | ino.ino = inodeno_t(); | |
5104 | } | |
5105 | void finish(int r) override { | |
5106 | // _async_dentry_invalidate is responsible for its own locking | |
9f95a23c | 5107 | ceph_assert(ceph_mutex_is_not_locked_by_me(client->client_lock)); |
7c673cae FG |
5108 | client->_async_dentry_invalidate(dirino, ino, name); |
5109 | } | |
5110 | }; | |
5111 | ||
5112 | void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name) | |
5113 | { | |
5114 | if (unmounting) | |
5115 | return; | |
11fdf7f2 | 5116 | ldout(cct, 10) << __func__ << " '" << name << "' ino " << ino |
7c673cae FG |
5117 | << " in dir " << dirino << dendl; |
5118 | dentry_invalidate_cb(callback_handle, dirino, ino, name); | |
5119 | } | |
5120 | ||
5121 | void Client::_schedule_invalidate_dentry_callback(Dentry *dn, bool del) | |
5122 | { | |
5123 | if (dentry_invalidate_cb && dn->inode->ll_ref > 0) | |
5124 | async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn, del)); | |
5125 | } | |
5126 | ||
5127 | void Client::_try_to_trim_inode(Inode *in, bool sched_inval) | |
5128 | { | |
5129 | int ref = in->get_num_ref(); | |
494da23a | 5130 | ldout(cct, 5) << __func__ << " in " << *in <<dendl; |
7c673cae FG |
5131 | |
5132 | if (in->dir && !in->dir->dentries.empty()) { | |
5133 | for (auto p = in->dir->dentries.begin(); | |
5134 | p != in->dir->dentries.end(); ) { | |
5135 | Dentry *dn = p->second; | |
5136 | ++p; | |
5137 | /* rmsnap removes whole subtree, need trim inodes recursively. | |
5138 | * we don't need to invalidate dentries recursively. because | |
5139 | * invalidating a directory dentry effectively invalidate | |
5140 | * whole subtree */ | |
5141 | if (in->snapid != CEPH_NOSNAP && dn->inode && dn->inode->is_dir()) | |
5142 | _try_to_trim_inode(dn->inode.get(), false); | |
5143 | ||
5144 | if (dn->lru_is_expireable()) | |
5145 | unlink(dn, true, false); // keep dir, drop dentry | |
5146 | } | |
5147 | if (in->dir->dentries.empty()) { | |
5148 | close_dir(in->dir); | |
5149 | --ref; | |
5150 | } | |
5151 | } | |
5152 | ||
5153 | if (ref > 0 && (in->flags & I_SNAPDIR_OPEN)) { | |
5154 | InodeRef snapdir = open_snapdir(in); | |
5155 | _try_to_trim_inode(snapdir.get(), false); | |
5156 | --ref; | |
5157 | } | |
5158 | ||
494da23a | 5159 | if (ref > 0) { |
11fdf7f2 TL |
5160 | auto q = in->dentries.begin(); |
5161 | while (q != in->dentries.end()) { | |
5162 | Dentry *dn = *q; | |
5163 | ++q; | |
494da23a TL |
5164 | if( in->ll_ref > 0 && sched_inval) { |
5165 | // FIXME: we play lots of unlink/link tricks when handling MDS replies, | |
5166 | // so in->dentries doesn't always reflect the state of kernel's dcache. | |
5167 | _schedule_invalidate_dentry_callback(dn, true); | |
5168 | } | |
7c673cae FG |
5169 | unlink(dn, true, true); |
5170 | } | |
5171 | } | |
5172 | } | |
5173 | ||
11fdf7f2 | 5174 | void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, const MConstRef<MClientCaps>& m) |
7c673cae FG |
5175 | { |
5176 | mds_rank_t mds = session->mds_num; | |
5177 | int used = get_caps_used(in); | |
5178 | int wanted = in->caps_wanted(); | |
5179 | ||
a8e16298 TL |
5180 | const unsigned new_caps = m->get_caps(); |
5181 | const bool was_stale = session->cap_gen > cap->gen; | |
11fdf7f2 | 5182 | ldout(cct, 5) << __func__ << " on in " << m->get_ino() |
7c673cae FG |
5183 | << " mds." << mds << " seq " << m->get_seq() |
5184 | << " caps now " << ccap_string(new_caps) | |
a8e16298 | 5185 | << " was " << ccap_string(cap->issued) |
92f5a8d4 | 5186 | << (was_stale ? " (stale)" : "") << dendl; |
a8e16298 TL |
5187 | |
5188 | if (was_stale) | |
5189 | cap->issued = cap->implemented = CEPH_CAP_PIN; | |
7c673cae | 5190 | cap->seq = m->get_seq(); |
28e407b8 | 5191 | cap->gen = session->cap_gen; |
7c673cae | 5192 | |
11fdf7f2 | 5193 | check_cap_issue(in, new_caps); |
a8e16298 | 5194 | |
7c673cae | 5195 | // update inode |
1adf2230 AA |
5196 | int issued; |
5197 | in->caps_issued(&issued); | |
5198 | issued |= in->caps_dirty(); | |
7c673cae | 5199 | |
1adf2230 AA |
5200 | if ((new_caps & CEPH_CAP_AUTH_SHARED) && |
5201 | !(issued & CEPH_CAP_AUTH_EXCL)) { | |
7c673cae FG |
5202 | in->mode = m->head.mode; |
5203 | in->uid = m->head.uid; | |
5204 | in->gid = m->head.gid; | |
5205 | in->btime = m->btime; | |
5206 | } | |
5207 | bool deleted_inode = false; | |
1adf2230 AA |
5208 | if ((new_caps & CEPH_CAP_LINK_SHARED) && |
5209 | !(issued & CEPH_CAP_LINK_EXCL)) { | |
7c673cae FG |
5210 | in->nlink = m->head.nlink; |
5211 | if (in->nlink == 0 && | |
5212 | (new_caps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL))) | |
5213 | deleted_inode = true; | |
5214 | } | |
1adf2230 | 5215 | if (!(issued & CEPH_CAP_XATTR_EXCL) && |
7c673cae FG |
5216 | m->xattrbl.length() && |
5217 | m->head.xattr_version > in->xattr_version) { | |
11fdf7f2 TL |
5218 | auto p = m->xattrbl.cbegin(); |
5219 | decode(in->xattrs, p); | |
7c673cae FG |
5220 | in->xattr_version = m->head.xattr_version; |
5221 | } | |
28e407b8 AA |
5222 | |
5223 | if ((new_caps & CEPH_CAP_FILE_SHARED) && m->dirstat_is_valid()) { | |
5224 | in->dirstat.nfiles = m->get_nfiles(); | |
5225 | in->dirstat.nsubdirs = m->get_nsubdirs(); | |
5226 | } | |
5227 | ||
1adf2230 AA |
5228 | if (new_caps & CEPH_CAP_ANY_RD) { |
5229 | update_inode_file_time(in, issued, m->get_time_warp_seq(), | |
5230 | m->get_ctime(), m->get_mtime(), m->get_atime()); | |
5231 | } | |
5232 | ||
5233 | if (new_caps & (CEPH_CAP_ANY_FILE_RD | CEPH_CAP_ANY_FILE_WR)) { | |
5234 | in->layout = m->get_layout(); | |
5235 | update_inode_file_size(in, issued, m->get_size(), | |
5236 | m->get_truncate_seq(), m->get_truncate_size()); | |
5237 | } | |
5238 | ||
5239 | if (m->inline_version > in->inline_version) { | |
5240 | in->inline_data = m->inline_data; | |
5241 | in->inline_version = m->inline_version; | |
5242 | } | |
5243 | ||
5244 | /* always take a newer change attr */ | |
5245 | if (m->get_change_attr() > in->change_attr) | |
5246 | in->change_attr = m->get_change_attr(); | |
7c673cae FG |
5247 | |
5248 | // max_size | |
5249 | if (cap == in->auth_cap && | |
1adf2230 AA |
5250 | (new_caps & CEPH_CAP_ANY_FILE_WR) && |
5251 | (m->get_max_size() != in->max_size)) { | |
7c673cae FG |
5252 | ldout(cct, 10) << "max_size " << in->max_size << " -> " << m->get_max_size() << dendl; |
5253 | in->max_size = m->get_max_size(); | |
5254 | if (in->max_size > in->wanted_max_size) { | |
5255 | in->wanted_max_size = 0; | |
5256 | in->requested_max_size = 0; | |
5257 | } | |
5258 | } | |
5259 | ||
5260 | bool check = false; | |
a8e16298 TL |
5261 | if ((was_stale || m->get_op() == CEPH_CAP_OP_IMPORT) && |
5262 | (wanted & ~(cap->wanted | new_caps))) { | |
5263 | // If mds is importing cap, prior cap messages that update 'wanted' | |
5264 | // may get dropped by mds (migrate seq mismatch). | |
5265 | // | |
5266 | // We don't send cap message to update 'wanted' if what we want are | |
5267 | // already issued. If mds revokes caps, cap message that releases caps | |
5268 | // also tells mds what we want. But if caps got revoked by mds forcedly | |
5269 | // (session stale). We may haven't told mds what we want. | |
7c673cae | 5270 | check = true; |
a8e16298 | 5271 | } |
7c673cae | 5272 | |
7c673cae FG |
5273 | |
5274 | // update caps | |
a8e16298 | 5275 | auto revoked = cap->issued & ~new_caps; |
b32b8144 FG |
5276 | if (revoked) { |
5277 | ldout(cct, 10) << " revocation of " << ccap_string(revoked) << dendl; | |
7c673cae FG |
5278 | cap->issued = new_caps; |
5279 | cap->implemented |= new_caps; | |
5280 | ||
b32b8144 FG |
5281 | // recall delegations if we're losing caps necessary for them |
5282 | if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_RD)) | |
5283 | in->recall_deleg(false); | |
5284 | else if (revoked & ceph_deleg_caps_for_type(CEPH_DELEGATION_WR)) | |
5285 | in->recall_deleg(true); | |
5286 | ||
11fdf7f2 TL |
5287 | used = adjust_caps_used_for_lazyio(used, cap->issued, cap->implemented); |
5288 | if ((used & revoked & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO)) && | |
28e407b8 | 5289 | !_flush(in, new C_Client_FlushComplete(this, in))) { |
7c673cae | 5290 | // waitin' for flush |
11fdf7f2 | 5291 | } else if (used & revoked & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO)) { |
7c673cae FG |
5292 | if (_release(in)) |
5293 | check = true; | |
5294 | } else { | |
5295 | cap->wanted = 0; // don't let check_caps skip sending a response to MDS | |
5296 | check = true; | |
5297 | } | |
a8e16298 TL |
5298 | } else if (cap->issued == new_caps) { |
5299 | ldout(cct, 10) << " caps unchanged at " << ccap_string(cap->issued) << dendl; | |
7c673cae | 5300 | } else { |
a8e16298 | 5301 | ldout(cct, 10) << " grant, new caps are " << ccap_string(new_caps & ~cap->issued) << dendl; |
7c673cae FG |
5302 | cap->issued = new_caps; |
5303 | cap->implemented |= new_caps; | |
5304 | ||
5305 | if (cap == in->auth_cap) { | |
5306 | // non-auth MDS is revoking the newly grant caps ? | |
11fdf7f2 TL |
5307 | for (const auto &p : in->caps) { |
5308 | if (&p.second == cap) | |
7c673cae | 5309 | continue; |
11fdf7f2 | 5310 | if (p.second.implemented & ~p.second.issued & new_caps) { |
7c673cae FG |
5311 | check = true; |
5312 | break; | |
5313 | } | |
5314 | } | |
5315 | } | |
5316 | } | |
5317 | ||
5318 | if (check) | |
5319 | check_caps(in, 0); | |
5320 | ||
5321 | // wake up waiters | |
5322 | if (new_caps) | |
5323 | signal_cond_list(in->waitfor_caps); | |
5324 | ||
5325 | // may drop inode's last ref | |
5326 | if (deleted_inode) | |
5327 | _try_to_trim_inode(in, true); | |
7c673cae FG |
5328 | } |
5329 | ||
7c673cae FG |
5330 | int Client::inode_permission(Inode *in, const UserPerm& perms, unsigned want) |
5331 | { | |
5332 | if (perms.uid() == 0) | |
5333 | return 0; | |
5334 | ||
5335 | if (perms.uid() != in->uid && (in->mode & S_IRWXG)) { | |
5336 | int ret = _posix_acl_permission(in, perms, want); | |
5337 | if (ret != -EAGAIN) | |
5338 | return ret; | |
5339 | } | |
5340 | ||
5341 | // check permissions before doing anything else | |
5342 | if (!in->check_mode(perms, want)) | |
5343 | return -EACCES; | |
5344 | return 0; | |
5345 | } | |
5346 | ||
5347 | int Client::xattr_permission(Inode *in, const char *name, unsigned want, | |
5348 | const UserPerm& perms) | |
5349 | { | |
5350 | int r = _getattr_for_perm(in, perms); | |
5351 | if (r < 0) | |
5352 | goto out; | |
5353 | ||
5354 | r = 0; | |
5355 | if (strncmp(name, "system.", 7) == 0) { | |
5356 | if ((want & MAY_WRITE) && (perms.uid() != 0 && perms.uid() != in->uid)) | |
5357 | r = -EPERM; | |
5358 | } else { | |
5359 | r = inode_permission(in, perms, want); | |
5360 | } | |
5361 | out: | |
1adf2230 | 5362 | ldout(cct, 5) << __func__ << " " << in << " = " << r << dendl; |
7c673cae FG |
5363 | return r; |
5364 | } | |
5365 | ||
5366 | ostream& operator<<(ostream &out, const UserPerm& perm) { | |
5367 | out << "UserPerm(uid: " << perm.uid() << ", gid: " << perm.gid() << ")"; | |
5368 | return out; | |
5369 | } | |
5370 | ||
5371 | int Client::may_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
5372 | const UserPerm& perms) | |
5373 | { | |
181888fb | 5374 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5375 | int r = _getattr_for_perm(in, perms); |
5376 | if (r < 0) | |
5377 | goto out; | |
5378 | ||
5379 | if (mask & CEPH_SETATTR_SIZE) { | |
5380 | r = inode_permission(in, perms, MAY_WRITE); | |
5381 | if (r < 0) | |
5382 | goto out; | |
5383 | } | |
5384 | ||
5385 | r = -EPERM; | |
5386 | if (mask & CEPH_SETATTR_UID) { | |
5387 | if (perms.uid() != 0 && (perms.uid() != in->uid || stx->stx_uid != in->uid)) | |
5388 | goto out; | |
5389 | } | |
5390 | if (mask & CEPH_SETATTR_GID) { | |
5391 | if (perms.uid() != 0 && (perms.uid() != in->uid || | |
5392 | (!perms.gid_in_groups(stx->stx_gid) && stx->stx_gid != in->gid))) | |
5393 | goto out; | |
5394 | } | |
5395 | ||
5396 | if (mask & CEPH_SETATTR_MODE) { | |
5397 | if (perms.uid() != 0 && perms.uid() != in->uid) | |
5398 | goto out; | |
5399 | ||
5400 | gid_t i_gid = (mask & CEPH_SETATTR_GID) ? stx->stx_gid : in->gid; | |
5401 | if (perms.uid() != 0 && !perms.gid_in_groups(i_gid)) | |
5402 | stx->stx_mode &= ~S_ISGID; | |
5403 | } | |
5404 | ||
5405 | if (mask & (CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME | | |
5406 | CEPH_SETATTR_MTIME | CEPH_SETATTR_ATIME)) { | |
5407 | if (perms.uid() != 0 && perms.uid() != in->uid) { | |
5408 | int check_mask = CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME; | |
5409 | if (!(mask & CEPH_SETATTR_MTIME_NOW)) | |
5410 | check_mask |= CEPH_SETATTR_MTIME; | |
5411 | if (!(mask & CEPH_SETATTR_ATIME_NOW)) | |
5412 | check_mask |= CEPH_SETATTR_ATIME; | |
5413 | if (check_mask & mask) { | |
5414 | goto out; | |
5415 | } else { | |
5416 | r = inode_permission(in, perms, MAY_WRITE); | |
5417 | if (r < 0) | |
5418 | goto out; | |
5419 | } | |
5420 | } | |
5421 | } | |
5422 | r = 0; | |
5423 | out: | |
5424 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5425 | return r; | |
5426 | } | |
5427 | ||
5428 | int Client::may_open(Inode *in, int flags, const UserPerm& perms) | |
5429 | { | |
181888fb | 5430 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5431 | unsigned want = 0; |
5432 | ||
5433 | if ((flags & O_ACCMODE) == O_WRONLY) | |
5434 | want = MAY_WRITE; | |
5435 | else if ((flags & O_ACCMODE) == O_RDWR) | |
5436 | want = MAY_READ | MAY_WRITE; | |
5437 | else if ((flags & O_ACCMODE) == O_RDONLY) | |
5438 | want = MAY_READ; | |
5439 | if (flags & O_TRUNC) | |
5440 | want |= MAY_WRITE; | |
5441 | ||
5442 | int r = 0; | |
5443 | switch (in->mode & S_IFMT) { | |
5444 | case S_IFLNK: | |
5445 | r = -ELOOP; | |
5446 | goto out; | |
5447 | case S_IFDIR: | |
5448 | if (want & MAY_WRITE) { | |
5449 | r = -EISDIR; | |
5450 | goto out; | |
5451 | } | |
5452 | break; | |
5453 | } | |
5454 | ||
5455 | r = _getattr_for_perm(in, perms); | |
5456 | if (r < 0) | |
5457 | goto out; | |
5458 | ||
5459 | r = inode_permission(in, perms, want); | |
5460 | out: | |
5461 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5462 | return r; | |
5463 | } | |
5464 | ||
5465 | int Client::may_lookup(Inode *dir, const UserPerm& perms) | |
5466 | { | |
181888fb | 5467 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5468 | int r = _getattr_for_perm(dir, perms); |
5469 | if (r < 0) | |
5470 | goto out; | |
5471 | ||
5472 | r = inode_permission(dir, perms, MAY_EXEC); | |
5473 | out: | |
5474 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5475 | return r; | |
5476 | } | |
5477 | ||
5478 | int Client::may_create(Inode *dir, const UserPerm& perms) | |
5479 | { | |
181888fb | 5480 | ldout(cct, 20) << __func__ << " " << *dir << "; " << perms << dendl; |
7c673cae FG |
5481 | int r = _getattr_for_perm(dir, perms); |
5482 | if (r < 0) | |
5483 | goto out; | |
5484 | ||
5485 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5486 | out: | |
5487 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5488 | return r; | |
5489 | } | |
5490 | ||
5491 | int Client::may_delete(Inode *dir, const char *name, const UserPerm& perms) | |
5492 | { | |
181888fb | 5493 | ldout(cct, 20) << __func__ << " " << *dir << "; " << "; name " << name << "; " << perms << dendl; |
7c673cae FG |
5494 | int r = _getattr_for_perm(dir, perms); |
5495 | if (r < 0) | |
5496 | goto out; | |
5497 | ||
5498 | r = inode_permission(dir, perms, MAY_EXEC | MAY_WRITE); | |
5499 | if (r < 0) | |
5500 | goto out; | |
5501 | ||
5502 | /* 'name == NULL' means rmsnap */ | |
5503 | if (perms.uid() != 0 && name && (dir->mode & S_ISVTX)) { | |
5504 | InodeRef otherin; | |
5505 | r = _lookup(dir, name, CEPH_CAP_AUTH_SHARED, &otherin, perms); | |
5506 | if (r < 0) | |
5507 | goto out; | |
5508 | if (dir->uid != perms.uid() && otherin->uid != perms.uid()) | |
5509 | r = -EPERM; | |
5510 | } | |
5511 | out: | |
5512 | ldout(cct, 3) << __func__ << " " << dir << " = " << r << dendl; | |
5513 | return r; | |
5514 | } | |
5515 | ||
5516 | int Client::may_hardlink(Inode *in, const UserPerm& perms) | |
5517 | { | |
181888fb | 5518 | ldout(cct, 20) << __func__ << " " << *in << "; " << perms << dendl; |
7c673cae FG |
5519 | int r = _getattr_for_perm(in, perms); |
5520 | if (r < 0) | |
5521 | goto out; | |
5522 | ||
5523 | if (perms.uid() == 0 || perms.uid() == in->uid) { | |
5524 | r = 0; | |
5525 | goto out; | |
5526 | } | |
5527 | ||
5528 | r = -EPERM; | |
5529 | if (!S_ISREG(in->mode)) | |
5530 | goto out; | |
5531 | ||
5532 | if (in->mode & S_ISUID) | |
5533 | goto out; | |
5534 | ||
5535 | if ((in->mode & (S_ISGID | S_IXGRP)) == (S_ISGID | S_IXGRP)) | |
5536 | goto out; | |
5537 | ||
5538 | r = inode_permission(in, perms, MAY_READ | MAY_WRITE); | |
5539 | out: | |
5540 | ldout(cct, 3) << __func__ << " " << in << " = " << r << dendl; | |
5541 | return r; | |
5542 | } | |
5543 | ||
5544 | int Client::_getattr_for_perm(Inode *in, const UserPerm& perms) | |
5545 | { | |
5546 | int mask = CEPH_STAT_CAP_MODE; | |
5547 | bool force = false; | |
5548 | if (acl_type != NO_ACL) { | |
5549 | mask |= CEPH_STAT_CAP_XATTR; | |
5550 | force = in->xattr_version == 0; | |
5551 | } | |
5552 | return _getattr(in, mask, perms, force); | |
5553 | } | |
5554 | ||
5555 | vinodeno_t Client::_get_vino(Inode *in) | |
5556 | { | |
5557 | /* The caller must hold the client lock */ | |
5558 | return vinodeno_t(in->ino, in->snapid); | |
5559 | } | |
5560 | ||
7c673cae FG |
5561 | /** |
5562 | * Resolve an MDS spec to a list of MDS daemon GIDs. | |
5563 | * | |
5564 | * The spec is a string representing a GID, rank, filesystem:rank, or name/id. | |
5565 | * It may be '*' in which case it matches all GIDs. | |
5566 | * | |
5567 | * If no error is returned, the `targets` vector will be populated with at least | |
5568 | * one MDS. | |
5569 | */ | |
5570 | int Client::resolve_mds( | |
5571 | const std::string &mds_spec, | |
5572 | std::vector<mds_gid_t> *targets) | |
5573 | { | |
11fdf7f2 TL |
5574 | ceph_assert(fsmap); |
5575 | ceph_assert(targets != nullptr); | |
7c673cae FG |
5576 | |
5577 | mds_role_t role; | |
5578 | std::stringstream ss; | |
5579 | int role_r = fsmap->parse_role(mds_spec, &role, ss); | |
5580 | if (role_r == 0) { | |
5581 | // We got a role, resolve it to a GID | |
5582 | ldout(cct, 10) << __func__ << ": resolved '" << mds_spec << "' to role '" | |
5583 | << role << "'" << dendl; | |
5584 | targets->push_back( | |
5585 | fsmap->get_filesystem(role.fscid)->mds_map.get_info(role.rank).global_id); | |
5586 | return 0; | |
5587 | } | |
5588 | ||
5589 | std::string strtol_err; | |
5590 | long long rank_or_gid = strict_strtoll(mds_spec.c_str(), 10, &strtol_err); | |
5591 | if (strtol_err.empty()) { | |
5592 | // It is a possible GID | |
5593 | const mds_gid_t mds_gid = mds_gid_t(rank_or_gid); | |
5594 | if (fsmap->gid_exists(mds_gid)) { | |
5595 | ldout(cct, 10) << __func__ << ": validated GID " << mds_gid << dendl; | |
5596 | targets->push_back(mds_gid); | |
5597 | } else { | |
5598 | lderr(cct) << __func__ << ": GID " << mds_gid << " not in MDS map" | |
5599 | << dendl; | |
5600 | return -ENOENT; | |
5601 | } | |
5602 | } else if (mds_spec == "*") { | |
5603 | // It is a wildcard: use all MDSs | |
5604 | const auto mds_info = fsmap->get_mds_info(); | |
5605 | ||
5606 | if (mds_info.empty()) { | |
5607 | lderr(cct) << __func__ << ": * passed but no MDS daemons found" << dendl; | |
5608 | return -ENOENT; | |
5609 | } | |
5610 | ||
5611 | for (const auto i : mds_info) { | |
5612 | targets->push_back(i.first); | |
5613 | } | |
5614 | } else { | |
5615 | // It did not parse as an integer, it is not a wildcard, it must be a name | |
5616 | const mds_gid_t mds_gid = fsmap->find_mds_gid_by_name(mds_spec); | |
5617 | if (mds_gid == 0) { | |
5618 | lderr(cct) << "MDS ID '" << mds_spec << "' not found" << dendl; | |
5619 | ||
5620 | lderr(cct) << "FSMap: " << *fsmap << dendl; | |
5621 | ||
5622 | return -ENOENT; | |
5623 | } else { | |
5624 | ldout(cct, 10) << __func__ << ": resolved ID '" << mds_spec | |
5625 | << "' to GID " << mds_gid << dendl; | |
5626 | targets->push_back(mds_gid); | |
5627 | } | |
5628 | } | |
5629 | ||
5630 | return 0; | |
5631 | } | |
5632 | ||
5633 | ||
5634 | /** | |
5635 | * Authenticate with mon and establish global ID | |
5636 | */ | |
5637 | int Client::authenticate() | |
5638 | { | |
9f95a23c | 5639 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
5640 | |
5641 | if (monclient->is_authenticated()) { | |
5642 | return 0; | |
5643 | } | |
5644 | ||
9f95a23c | 5645 | client_lock.unlock(); |
7c673cae | 5646 | int r = monclient->authenticate(cct->_conf->client_mount_timeout); |
9f95a23c | 5647 | client_lock.lock(); |
7c673cae FG |
5648 | if (r < 0) { |
5649 | return r; | |
5650 | } | |
5651 | ||
5652 | whoami = monclient->get_global_id(); | |
5653 | messenger->set_myname(entity_name_t::CLIENT(whoami.v)); | |
5654 | ||
5655 | return 0; | |
5656 | } | |
5657 | ||
5658 | int Client::fetch_fsmap(bool user) | |
5659 | { | |
5660 | int r; | |
5661 | // Retrieve FSMap to enable looking up daemon addresses. We need FSMap | |
5662 | // rather than MDSMap because no one MDSMap contains all the daemons, and | |
5663 | // a `tell` can address any daemon. | |
5664 | version_t fsmap_latest; | |
5665 | do { | |
5666 | C_SaferCond cond; | |
5667 | monclient->get_version("fsmap", &fsmap_latest, NULL, &cond); | |
9f95a23c | 5668 | client_lock.unlock(); |
7c673cae | 5669 | r = cond.wait(); |
9f95a23c | 5670 | client_lock.lock(); |
7c673cae FG |
5671 | } while (r == -EAGAIN); |
5672 | ||
5673 | if (r < 0) { | |
5674 | lderr(cct) << "Failed to learn FSMap version: " << cpp_strerror(r) << dendl; | |
5675 | return r; | |
5676 | } | |
5677 | ||
5678 | ldout(cct, 10) << __func__ << " learned FSMap version " << fsmap_latest << dendl; | |
5679 | ||
5680 | if (user) { | |
5681 | if (!fsmap_user || fsmap_user->get_epoch() < fsmap_latest) { | |
5682 | monclient->sub_want("fsmap.user", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5683 | monclient->renew_subs(); | |
5684 | wait_on_list(waiting_for_fsmap); | |
5685 | } | |
11fdf7f2 TL |
5686 | ceph_assert(fsmap_user); |
5687 | ceph_assert(fsmap_user->get_epoch() >= fsmap_latest); | |
7c673cae FG |
5688 | } else { |
5689 | if (!fsmap || fsmap->get_epoch() < fsmap_latest) { | |
5690 | monclient->sub_want("fsmap", fsmap_latest, CEPH_SUBSCRIBE_ONETIME); | |
5691 | monclient->renew_subs(); | |
5692 | wait_on_list(waiting_for_fsmap); | |
5693 | } | |
11fdf7f2 TL |
5694 | ceph_assert(fsmap); |
5695 | ceph_assert(fsmap->get_epoch() >= fsmap_latest); | |
7c673cae FG |
5696 | } |
5697 | ldout(cct, 10) << __func__ << " finished waiting for FSMap version " | |
5698 | << fsmap_latest << dendl; | |
5699 | return 0; | |
5700 | } | |
5701 | ||
5702 | /** | |
5703 | * | |
5704 | * @mds_spec one of ID, rank, GID, "*" | |
5705 | * | |
5706 | */ | |
5707 | int Client::mds_command( | |
5708 | const std::string &mds_spec, | |
5709 | const vector<string>& cmd, | |
5710 | const bufferlist& inbl, | |
5711 | bufferlist *outbl, | |
5712 | string *outs, | |
5713 | Context *onfinish) | |
5714 | { | |
11fdf7f2 | 5715 | std::lock_guard lock(client_lock); |
7c673cae | 5716 | |
181888fb FG |
5717 | if (!initialized) |
5718 | return -ENOTCONN; | |
7c673cae FG |
5719 | |
5720 | int r; | |
5721 | r = authenticate(); | |
5722 | if (r < 0) { | |
5723 | return r; | |
5724 | } | |
5725 | ||
5726 | r = fetch_fsmap(false); | |
5727 | if (r < 0) { | |
5728 | return r; | |
5729 | } | |
5730 | ||
5731 | // Look up MDS target(s) of the command | |
5732 | std::vector<mds_gid_t> targets; | |
5733 | r = resolve_mds(mds_spec, &targets); | |
5734 | if (r < 0) { | |
5735 | return r; | |
5736 | } | |
5737 | ||
5738 | // If daemons are laggy, we won't send them commands. If all | |
5739 | // are laggy then we fail. | |
5740 | std::vector<mds_gid_t> non_laggy; | |
5741 | for (const auto gid : targets) { | |
5742 | const auto info = fsmap->get_info_gid(gid); | |
5743 | if (!info.laggy()) { | |
5744 | non_laggy.push_back(gid); | |
5745 | } | |
5746 | } | |
5747 | if (non_laggy.size() == 0) { | |
5748 | *outs = "All targeted MDS daemons are laggy"; | |
5749 | return -ENOENT; | |
5750 | } | |
5751 | ||
5752 | if (metadata.empty()) { | |
5753 | // We are called on an unmounted client, so metadata | |
5754 | // won't be initialized yet. | |
5755 | populate_metadata(""); | |
5756 | } | |
5757 | ||
5758 | // Send commands to targets | |
5759 | C_GatherBuilder gather(cct, onfinish); | |
5760 | for (const auto target_gid : non_laggy) { | |
5761 | const auto info = fsmap->get_info_gid(target_gid); | |
5762 | ||
5763 | // Open a connection to the target MDS | |
11fdf7f2 | 5764 | ConnectionRef conn = messenger->connect_to_mds(info.get_addrs()); |
7c673cae FG |
5765 | |
5766 | // Generate MDSCommandOp state | |
5767 | auto &op = command_table.start_command(); | |
5768 | ||
5769 | op.on_finish = gather.new_sub(); | |
5770 | op.cmd = cmd; | |
5771 | op.outbl = outbl; | |
5772 | op.outs = outs; | |
5773 | op.inbl = inbl; | |
5774 | op.mds_gid = target_gid; | |
5775 | op.con = conn; | |
5776 | ||
5777 | ldout(cct, 4) << __func__ << ": new command op to " << target_gid | |
5778 | << " tid=" << op.tid << cmd << dendl; | |
5779 | ||
5780 | // Construct and send MCommand | |
11fdf7f2 TL |
5781 | auto m = op.get_message(monclient->get_fsid()); |
5782 | conn->send_message2(std::move(m)); | |
7c673cae FG |
5783 | } |
5784 | gather.activate(); | |
5785 | ||
5786 | return 0; | |
5787 | } | |
5788 | ||
11fdf7f2 | 5789 | void Client::handle_command_reply(const MConstRef<MCommandReply>& m) |
7c673cae FG |
5790 | { |
5791 | ceph_tid_t const tid = m->get_tid(); | |
5792 | ||
5793 | ldout(cct, 10) << __func__ << ": tid=" << m->get_tid() << dendl; | |
5794 | ||
5795 | if (!command_table.exists(tid)) { | |
5796 | ldout(cct, 1) << __func__ << ": unknown tid " << tid << ", dropping" << dendl; | |
7c673cae FG |
5797 | return; |
5798 | } | |
5799 | ||
5800 | auto &op = command_table.get_command(tid); | |
5801 | if (op.outbl) { | |
11fdf7f2 | 5802 | *op.outbl = m->get_data(); |
7c673cae FG |
5803 | } |
5804 | if (op.outs) { | |
5805 | *op.outs = m->rs; | |
5806 | } | |
5807 | ||
5808 | if (op.on_finish) { | |
5809 | op.on_finish->complete(m->r); | |
5810 | } | |
5811 | ||
5812 | command_table.erase(tid); | |
7c673cae FG |
5813 | } |
5814 | ||
5815 | // ------------------- | |
5816 | // MOUNT | |
5817 | ||
11fdf7f2 | 5818 | int Client::subscribe_mdsmap(const std::string &fs_name) |
7c673cae | 5819 | { |
7c673cae FG |
5820 | int r = authenticate(); |
5821 | if (r < 0) { | |
5822 | lderr(cct) << "authentication failed: " << cpp_strerror(r) << dendl; | |
5823 | return r; | |
5824 | } | |
5825 | ||
11fdf7f2 TL |
5826 | std::string resolved_fs_name; |
5827 | if (fs_name.empty()) { | |
9f95a23c TL |
5828 | resolved_fs_name = cct->_conf.get_val<std::string>("client_fs"); |
5829 | if (resolved_fs_name.empty()) | |
5830 | // Try the backwards compatibility fs name option | |
5831 | resolved_fs_name = cct->_conf.get_val<std::string>("client_mds_namespace"); | |
11fdf7f2 TL |
5832 | } else { |
5833 | resolved_fs_name = fs_name; | |
5834 | } | |
5835 | ||
7c673cae | 5836 | std::string want = "mdsmap"; |
11fdf7f2 | 5837 | if (!resolved_fs_name.empty()) { |
7c673cae FG |
5838 | r = fetch_fsmap(true); |
5839 | if (r < 0) | |
5840 | return r; | |
11fdf7f2 TL |
5841 | fscid = fsmap_user->get_fs_cid(resolved_fs_name); |
5842 | if (fscid == FS_CLUSTER_ID_NONE) { | |
7c673cae | 5843 | return -ENOENT; |
11fdf7f2 | 5844 | } |
7c673cae FG |
5845 | |
5846 | std::ostringstream oss; | |
11fdf7f2 | 5847 | oss << want << "." << fscid; |
7c673cae FG |
5848 | want = oss.str(); |
5849 | } | |
5850 | ldout(cct, 10) << "Subscribing to map '" << want << "'" << dendl; | |
5851 | ||
5852 | monclient->sub_want(want, 0, 0); | |
5853 | monclient->renew_subs(); | |
5854 | ||
11fdf7f2 TL |
5855 | return 0; |
5856 | } | |
5857 | ||
5858 | int Client::mount(const std::string &mount_root, const UserPerm& perms, | |
5859 | bool require_mds, const std::string &fs_name) | |
5860 | { | |
5861 | std::lock_guard lock(client_lock); | |
5862 | ||
5863 | if (mounted) { | |
5864 | ldout(cct, 5) << "already mounted" << dendl; | |
5865 | return 0; | |
5866 | } | |
5867 | ||
5868 | unmounting = false; | |
5869 | ||
5870 | int r = subscribe_mdsmap(fs_name); | |
5871 | if (r < 0) { | |
5872 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
5873 | return r; | |
5874 | } | |
5875 | ||
7c673cae FG |
5876 | tick(); // start tick |
5877 | ||
5878 | if (require_mds) { | |
5879 | while (1) { | |
5880 | auto availability = mdsmap->is_cluster_available(); | |
5881 | if (availability == MDSMap::STUCK_UNAVAILABLE) { | |
5882 | // Error out | |
5883 | ldout(cct, 10) << "mds cluster unavailable: epoch=" << mdsmap->get_epoch() << dendl; | |
5884 | return CEPH_FUSE_NO_MDS_UP; | |
5885 | } else if (availability == MDSMap::AVAILABLE) { | |
5886 | // Continue to mount | |
5887 | break; | |
5888 | } else if (availability == MDSMap::TRANSIENT_UNAVAILABLE) { | |
5889 | // Else, wait. MDSMonitor will update the map to bring | |
5890 | // us to a conclusion eventually. | |
5891 | wait_on_list(waiting_for_mdsmap); | |
5892 | } else { | |
5893 | // Unexpected value! | |
5894 | ceph_abort(); | |
5895 | } | |
5896 | } | |
5897 | } | |
5898 | ||
5899 | populate_metadata(mount_root.empty() ? "/" : mount_root); | |
5900 | ||
5901 | filepath fp(CEPH_INO_ROOT); | |
5902 | if (!mount_root.empty()) { | |
5903 | fp = filepath(mount_root.c_str()); | |
5904 | } | |
5905 | while (true) { | |
5906 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
5907 | req->set_filepath(fp); | |
5908 | req->head.args.getattr.mask = CEPH_STAT_CAP_INODE_ALL; | |
5909 | int res = make_request(req, perms); | |
5910 | if (res < 0) { | |
5911 | if (res == -EACCES && root) { | |
5912 | ldout(cct, 1) << __func__ << " EACCES on parent of mount point; quotas may not work" << dendl; | |
5913 | break; | |
5914 | } | |
5915 | return res; | |
5916 | } | |
5917 | ||
5918 | if (fp.depth()) | |
5919 | fp.pop_dentry(); | |
5920 | else | |
5921 | break; | |
5922 | } | |
5923 | ||
11fdf7f2 | 5924 | ceph_assert(root); |
7c673cae FG |
5925 | _ll_get(root); |
5926 | ||
5927 | mounted = true; | |
5928 | ||
5929 | // trace? | |
5930 | if (!cct->_conf->client_trace.empty()) { | |
5931 | traceout.open(cct->_conf->client_trace.c_str()); | |
5932 | if (traceout.is_open()) { | |
5933 | ldout(cct, 1) << "opened trace file '" << cct->_conf->client_trace << "'" << dendl; | |
5934 | } else { | |
5935 | ldout(cct, 1) << "FAILED to open trace file '" << cct->_conf->client_trace << "'" << dendl; | |
5936 | } | |
5937 | } | |
5938 | ||
5939 | /* | |
5940 | ldout(cct, 3) << "op: // client trace data structs" << dendl; | |
5941 | ldout(cct, 3) << "op: struct stat st;" << dendl; | |
5942 | ldout(cct, 3) << "op: struct utimbuf utim;" << dendl; | |
5943 | ldout(cct, 3) << "op: int readlinkbuf_len = 1000;" << dendl; | |
5944 | ldout(cct, 3) << "op: char readlinkbuf[readlinkbuf_len];" << dendl; | |
5945 | ldout(cct, 3) << "op: map<string, inode_t*> dir_contents;" << dendl; | |
5946 | ldout(cct, 3) << "op: map<int, int> open_files;" << dendl; | |
5947 | ldout(cct, 3) << "op: int fd;" << dendl; | |
5948 | */ | |
5949 | return 0; | |
5950 | } | |
5951 | ||
5952 | // UNMOUNT | |
5953 | ||
5954 | void Client::_close_sessions() | |
5955 | { | |
5956 | while (!mds_sessions.empty()) { | |
5957 | // send session closes! | |
11fdf7f2 TL |
5958 | for (auto &p : mds_sessions) { |
5959 | if (p.second.state != MetaSession::STATE_CLOSING) { | |
5960 | _close_mds_session(&p.second); | |
7c673cae FG |
5961 | } |
5962 | } | |
5963 | ||
5964 | // wait for sessions to close | |
5965 | ldout(cct, 2) << "waiting for " << mds_sessions.size() << " mds sessions to close" << dendl; | |
9f95a23c TL |
5966 | std::unique_lock l{client_lock, std::adopt_lock}; |
5967 | mount_cond.wait(l); | |
5968 | l.release(); | |
7c673cae FG |
5969 | } |
5970 | } | |
5971 | ||
31f18b77 FG |
5972 | void Client::flush_mdlog_sync() |
5973 | { | |
5974 | if (mds_requests.empty()) | |
5975 | return; | |
11fdf7f2 TL |
5976 | for (auto &p : mds_sessions) { |
5977 | flush_mdlog(&p.second); | |
31f18b77 FG |
5978 | } |
5979 | } | |
5980 | ||
5981 | void Client::flush_mdlog(MetaSession *session) | |
5982 | { | |
5983 | // Only send this to Luminous or newer MDS daemons, older daemons | |
5984 | // will crash if they see an unknown CEPH_SESSION_* value in this msg. | |
5985 | const uint64_t features = session->con->get_features(); | |
5986 | if (HAVE_FEATURE(features, SERVER_LUMINOUS)) { | |
9f95a23c | 5987 | auto m = make_message<MClientSession>(CEPH_SESSION_REQUEST_FLUSH_MDLOG); |
11fdf7f2 | 5988 | session->con->send_message2(std::move(m)); |
31f18b77 FG |
5989 | } |
5990 | } | |
5991 | ||
5992 | ||
11fdf7f2 TL |
5993 | void Client::_abort_mds_sessions(int err) |
5994 | { | |
5995 | for (auto p = mds_requests.begin(); p != mds_requests.end(); ) { | |
5996 | auto req = p->second; | |
5997 | ++p; | |
5998 | // unsafe requests will be removed during close session below. | |
5999 | if (req->got_unsafe) | |
6000 | continue; | |
6001 | ||
6002 | req->abort(err); | |
6003 | if (req->caller_cond) { | |
6004 | req->kick = true; | |
9f95a23c | 6005 | req->caller_cond->notify_all(); |
11fdf7f2 TL |
6006 | } |
6007 | } | |
6008 | ||
6009 | // Process aborts on any requests that were on this waitlist. | |
6010 | // Any requests that were on a waiting_for_open session waitlist | |
6011 | // will get kicked during close session below. | |
6012 | signal_cond_list(waiting_for_mdsmap); | |
6013 | ||
6014 | // Force-close all sessions | |
6015 | while(!mds_sessions.empty()) { | |
6016 | auto& session = mds_sessions.begin()->second; | |
6017 | _closed_mds_session(&session); | |
6018 | } | |
6019 | } | |
6020 | ||
6021 | void Client::_unmount(bool abort) | |
7c673cae | 6022 | { |
9f95a23c | 6023 | std::unique_lock lock{client_lock, std::adopt_lock}; |
181888fb FG |
6024 | if (unmounting) |
6025 | return; | |
7c673cae | 6026 | |
11fdf7f2 TL |
6027 | if (abort || blacklisted) { |
6028 | ldout(cct, 2) << "unmounting (" << (abort ? "abort)" : "blacklisted)") << dendl; | |
6029 | } else { | |
6030 | ldout(cct, 2) << "unmounting" << dendl; | |
6031 | } | |
7c673cae FG |
6032 | unmounting = true; |
6033 | ||
b32b8144 FG |
6034 | deleg_timeout = 0; |
6035 | ||
11fdf7f2 TL |
6036 | if (abort) { |
6037 | // Abort all mds sessions | |
6038 | _abort_mds_sessions(-ENOTCONN); | |
6039 | ||
6040 | objecter->op_cancel_writes(-ENOTCONN); | |
6041 | } else { | |
6042 | // flush the mdlog for pending requests, if any | |
6043 | flush_mdlog_sync(); | |
6044 | } | |
6045 | ||
9f95a23c TL |
6046 | mount_cond.wait(lock, [this] { |
6047 | if (!mds_requests.empty()) { | |
6048 | ldout(cct, 10) << "waiting on " << mds_requests.size() << " requests" | |
6049 | << dendl; | |
6050 | } | |
6051 | return mds_requests.empty(); | |
6052 | }); | |
7c673cae FG |
6053 | if (tick_event) |
6054 | timer.cancel_event(tick_event); | |
6055 | tick_event = 0; | |
6056 | ||
6057 | cwd.reset(); | |
6058 | ||
6059 | // clean up any unclosed files | |
6060 | while (!fd_map.empty()) { | |
6061 | Fh *fh = fd_map.begin()->second; | |
6062 | fd_map.erase(fd_map.begin()); | |
6063 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *fh->inode << dendl; | |
6064 | _release_fh(fh); | |
6065 | } | |
6066 | ||
6067 | while (!ll_unclosed_fh_set.empty()) { | |
6068 | set<Fh*>::iterator it = ll_unclosed_fh_set.begin(); | |
6069 | Fh *fh = *it; | |
6070 | ll_unclosed_fh_set.erase(fh); | |
6071 | ldout(cct, 0) << " destroyed lost open file " << fh << " on " << *(fh->inode) << dendl; | |
6072 | _release_fh(fh); | |
6073 | } | |
6074 | ||
6075 | while (!opened_dirs.empty()) { | |
6076 | dir_result_t *dirp = *opened_dirs.begin(); | |
6077 | ldout(cct, 0) << " destroyed lost open dir " << dirp << " on " << *dirp->inode << dendl; | |
6078 | _closedir(dirp); | |
6079 | } | |
6080 | ||
6081 | _ll_drop_pins(); | |
6082 | ||
9f95a23c TL |
6083 | mount_cond.wait(lock, [this] { |
6084 | if (unsafe_sync_write > 0) { | |
6085 | ldout(cct, 0) << unsafe_sync_write << " unsafe_sync_writes, waiting" | |
6086 | << dendl; | |
6087 | } | |
6088 | return unsafe_sync_write <= 0; | |
6089 | }); | |
7c673cae FG |
6090 | |
6091 | if (cct->_conf->client_oc) { | |
6092 | // flush/release all buffered data | |
11fdf7f2 TL |
6093 | std::list<InodeRef> anchor; |
6094 | for (auto& p : inode_map) { | |
6095 | Inode *in = p.second; | |
7c673cae | 6096 | if (!in) { |
11fdf7f2 TL |
6097 | ldout(cct, 0) << "null inode_map entry ino " << p.first << dendl; |
6098 | ceph_assert(in); | |
7c673cae | 6099 | } |
11fdf7f2 TL |
6100 | |
6101 | // prevent inode from getting freed | |
6102 | anchor.emplace_back(in); | |
6103 | ||
6104 | if (abort || blacklisted) { | |
6105 | objectcacher->purge_set(&in->oset); | |
6106 | } else if (!in->caps.empty()) { | |
7c673cae FG |
6107 | _release(in); |
6108 | _flush(in, new C_Client_FlushComplete(this, in)); | |
6109 | } | |
6110 | } | |
6111 | } | |
6112 | ||
11fdf7f2 TL |
6113 | if (abort || blacklisted) { |
6114 | for (auto p = dirty_list.begin(); !p.end(); ) { | |
6115 | Inode *in = *p; | |
6116 | ++p; | |
6117 | if (in->dirty_caps) { | |
6118 | ldout(cct, 0) << " drop dirty caps on " << *in << dendl; | |
6119 | in->mark_caps_clean(); | |
6120 | put_inode(in); | |
6121 | } | |
6122 | } | |
6123 | } else { | |
6124 | flush_caps_sync(); | |
6125 | wait_sync_caps(last_flush_tid); | |
6126 | } | |
7c673cae FG |
6127 | |
6128 | // empty lru cache | |
7c673cae FG |
6129 | trim_cache(); |
6130 | ||
6131 | while (lru.lru_get_size() > 0 || | |
6132 | !inode_map.empty()) { | |
6133 | ldout(cct, 2) << "cache still has " << lru.lru_get_size() | |
6134 | << "+" << inode_map.size() << " items" | |
6135 | << ", waiting (for caps to release?)" | |
6136 | << dendl; | |
9f95a23c TL |
6137 | if (auto r = mount_cond.wait_for(lock, ceph::make_timespan(5)); |
6138 | r == std::cv_status::timeout) { | |
7c673cae FG |
6139 | dump_cache(NULL); |
6140 | } | |
6141 | } | |
11fdf7f2 TL |
6142 | ceph_assert(lru.lru_get_size() == 0); |
6143 | ceph_assert(inode_map.empty()); | |
7c673cae FG |
6144 | |
6145 | // stop tracing | |
6146 | if (!cct->_conf->client_trace.empty()) { | |
6147 | ldout(cct, 1) << "closing trace file '" << cct->_conf->client_trace << "'" << dendl; | |
6148 | traceout.close(); | |
6149 | } | |
6150 | ||
6151 | _close_sessions(); | |
6152 | ||
6153 | mounted = false; | |
6154 | ||
9f95a23c | 6155 | lock.release(); |
7c673cae FG |
6156 | ldout(cct, 2) << "unmounted." << dendl; |
6157 | } | |
6158 | ||
b32b8144 FG |
6159 | void Client::unmount() |
6160 | { | |
11fdf7f2 TL |
6161 | std::lock_guard lock(client_lock); |
6162 | _unmount(false); | |
6163 | } | |
6164 | ||
6165 | void Client::abort_conn() | |
6166 | { | |
6167 | std::lock_guard lock(client_lock); | |
6168 | _unmount(true); | |
b32b8144 FG |
6169 | } |
6170 | ||
7c673cae FG |
6171 | void Client::flush_cap_releases() |
6172 | { | |
6173 | // send any cap releases | |
11fdf7f2 TL |
6174 | for (auto &p : mds_sessions) { |
6175 | auto &session = p.second; | |
6176 | if (session.release && mdsmap->is_clientreplay_or_active_or_stopping( | |
6177 | p.first)) { | |
7c673cae FG |
6178 | if (cct->_conf->client_inject_release_failure) { |
6179 | ldout(cct, 20) << __func__ << " injecting failure to send cap release message" << dendl; | |
7c673cae | 6180 | } else { |
11fdf7f2 | 6181 | session.con->send_message2(std::move(session.release)); |
7c673cae | 6182 | } |
11fdf7f2 | 6183 | session.release.reset(); |
7c673cae FG |
6184 | } |
6185 | } | |
6186 | } | |
6187 | ||
6188 | void Client::tick() | |
6189 | { | |
6190 | if (cct->_conf->client_debug_inject_tick_delay > 0) { | |
6191 | sleep(cct->_conf->client_debug_inject_tick_delay); | |
11fdf7f2 TL |
6192 | ceph_assert(0 == cct->_conf.set_val("client_debug_inject_tick_delay", "0")); |
6193 | cct->_conf.apply_changes(nullptr); | |
7c673cae FG |
6194 | } |
6195 | ||
6196 | ldout(cct, 21) << "tick" << dendl; | |
3efd9988 FG |
6197 | tick_event = timer.add_event_after( |
6198 | cct->_conf->client_tick_interval, | |
9f95a23c | 6199 | new LambdaContext([this](int) { |
3efd9988 | 6200 | // Called back via Timer, which takes client_lock for us |
9f95a23c | 6201 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
3efd9988 FG |
6202 | tick(); |
6203 | })); | |
7c673cae FG |
6204 | utime_t now = ceph_clock_now(); |
6205 | ||
6206 | if (!mounted && !mds_requests.empty()) { | |
6207 | MetaRequest *req = mds_requests.begin()->second; | |
6208 | if (req->op_stamp + cct->_conf->client_mount_timeout < now) { | |
6209 | req->abort(-ETIMEDOUT); | |
6210 | if (req->caller_cond) { | |
6211 | req->kick = true; | |
9f95a23c | 6212 | req->caller_cond->notify_all(); |
7c673cae FG |
6213 | } |
6214 | signal_cond_list(waiting_for_mdsmap); | |
11fdf7f2 TL |
6215 | for (auto &p : mds_sessions) { |
6216 | signal_context_list(p.second.waiting_for_open); | |
6217 | } | |
7c673cae FG |
6218 | } |
6219 | } | |
6220 | ||
6221 | if (mdsmap->get_epoch()) { | |
6222 | // renew caps? | |
6223 | utime_t el = now - last_cap_renew; | |
6224 | if (el > mdsmap->get_session_timeout() / 3.0) | |
6225 | renew_caps(); | |
6226 | ||
6227 | flush_cap_releases(); | |
6228 | } | |
6229 | ||
6230 | // delayed caps | |
28e407b8 | 6231 | xlist<Inode*>::iterator p = delayed_list.begin(); |
7c673cae FG |
6232 | while (!p.end()) { |
6233 | Inode *in = *p; | |
6234 | ++p; | |
6235 | if (in->hold_caps_until > now) | |
6236 | break; | |
28e407b8 | 6237 | delayed_list.pop_front(); |
7c673cae FG |
6238 | check_caps(in, CHECK_CAPS_NODELAY); |
6239 | } | |
6240 | ||
6241 | trim_cache(true); | |
6242 | } | |
6243 | ||
6244 | void Client::renew_caps() | |
6245 | { | |
6246 | ldout(cct, 10) << "renew_caps()" << dendl; | |
6247 | last_cap_renew = ceph_clock_now(); | |
6248 | ||
11fdf7f2 TL |
6249 | for (auto &p : mds_sessions) { |
6250 | ldout(cct, 15) << "renew_caps requesting from mds." << p.first << dendl; | |
6251 | if (mdsmap->get_state(p.first) >= MDSMap::STATE_REJOIN) | |
6252 | renew_caps(&p.second); | |
7c673cae FG |
6253 | } |
6254 | } | |
6255 | ||
6256 | void Client::renew_caps(MetaSession *session) | |
6257 | { | |
6258 | ldout(cct, 10) << "renew_caps mds." << session->mds_num << dendl; | |
6259 | session->last_cap_renew_request = ceph_clock_now(); | |
6260 | uint64_t seq = ++session->cap_renew_seq; | |
9f95a23c | 6261 | session->con->send_message2(make_message<MClientSession>(CEPH_SESSION_REQUEST_RENEWCAPS, seq)); |
7c673cae FG |
6262 | } |
6263 | ||
6264 | ||
6265 | // =============================================================== | |
6266 | // high level (POSIXy) interface | |
6267 | ||
6268 | int Client::_do_lookup(Inode *dir, const string& name, int mask, | |
6269 | InodeRef *target, const UserPerm& perms) | |
6270 | { | |
6271 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_LOOKUPSNAP : CEPH_MDS_OP_LOOKUP; | |
6272 | MetaRequest *req = new MetaRequest(op); | |
6273 | filepath path; | |
6274 | dir->make_nosnap_relative_path(path); | |
6275 | path.push_dentry(name); | |
6276 | req->set_filepath(path); | |
6277 | req->set_inode(dir); | |
6278 | if (cct->_conf->client_debug_getattr_caps && op == CEPH_MDS_OP_LOOKUP) | |
6279 | mask |= DEBUG_GETATTR_CAPS; | |
6280 | req->head.args.getattr.mask = mask; | |
6281 | ||
11fdf7f2 | 6282 | ldout(cct, 10) << __func__ << " on " << path << dendl; |
7c673cae FG |
6283 | |
6284 | int r = make_request(req, perms, target); | |
11fdf7f2 | 6285 | ldout(cct, 10) << __func__ << " res is " << r << dendl; |
7c673cae FG |
6286 | return r; |
6287 | } | |
6288 | ||
6289 | int Client::_lookup(Inode *dir, const string& dname, int mask, InodeRef *target, | |
6290 | const UserPerm& perms) | |
6291 | { | |
6292 | int r = 0; | |
6293 | Dentry *dn = NULL; | |
6294 | ||
7c673cae | 6295 | if (dname == "..") { |
11fdf7f2 TL |
6296 | if (dir->dentries.empty()) { |
6297 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); | |
6298 | filepath path(dir->ino); | |
6299 | req->set_filepath(path); | |
6300 | ||
6301 | InodeRef tmptarget; | |
6302 | int r = make_request(req, perms, &tmptarget, NULL, rand() % mdsmap->get_num_in_mds()); | |
6303 | ||
6304 | if (r == 0) { | |
6305 | Inode *tempino = tmptarget.get(); | |
6306 | _ll_get(tempino); | |
6307 | *target = tempino; | |
6308 | ldout(cct, 8) << __func__ << " found target " << (*target)->ino << dendl; | |
6309 | } else { | |
6310 | *target = dir; | |
6311 | } | |
6312 | } | |
7c673cae FG |
6313 | else |
6314 | *target = dir->get_first_parent()->dir->parent_inode; //dirs can't be hard-linked | |
6315 | goto done; | |
6316 | } | |
6317 | ||
6318 | if (dname == ".") { | |
6319 | *target = dir; | |
6320 | goto done; | |
6321 | } | |
6322 | ||
11fdf7f2 TL |
6323 | if (!dir->is_dir()) { |
6324 | r = -ENOTDIR; | |
6325 | goto done; | |
6326 | } | |
6327 | ||
7c673cae FG |
6328 | if (dname.length() > NAME_MAX) { |
6329 | r = -ENAMETOOLONG; | |
6330 | goto done; | |
6331 | } | |
6332 | ||
6333 | if (dname == cct->_conf->client_snapdir && | |
6334 | dir->snapid == CEPH_NOSNAP) { | |
6335 | *target = open_snapdir(dir); | |
6336 | goto done; | |
6337 | } | |
6338 | ||
6339 | if (dir->dir && | |
6340 | dir->dir->dentries.count(dname)) { | |
6341 | dn = dir->dir->dentries[dname]; | |
6342 | ||
11fdf7f2 | 6343 | ldout(cct, 20) << __func__ << " have dn " << dname << " mds." << dn->lease_mds << " ttl " << dn->lease_ttl |
7c673cae FG |
6344 | << " seq " << dn->lease_seq |
6345 | << dendl; | |
6346 | ||
94b18763 | 6347 | if (!dn->inode || dn->inode->caps_issued_mask(mask, true)) { |
7c673cae FG |
6348 | // is dn lease valid? |
6349 | utime_t now = ceph_clock_now(); | |
6350 | if (dn->lease_mds >= 0 && | |
6351 | dn->lease_ttl > now && | |
6352 | mds_sessions.count(dn->lease_mds)) { | |
11fdf7f2 TL |
6353 | MetaSession &s = mds_sessions.at(dn->lease_mds); |
6354 | if (s.cap_ttl > now && | |
6355 | s.cap_gen == dn->lease_gen) { | |
7c673cae FG |
6356 | // touch this mds's dir cap too, even though we don't _explicitly_ use it here, to |
6357 | // make trim_caps() behave. | |
6358 | dir->try_touch_cap(dn->lease_mds); | |
6359 | goto hit_dn; | |
6360 | } | |
11fdf7f2 | 6361 | ldout(cct, 20) << " bad lease, cap_ttl " << s.cap_ttl << ", cap_gen " << s.cap_gen |
7c673cae FG |
6362 | << " vs lease_gen " << dn->lease_gen << dendl; |
6363 | } | |
92f5a8d4 | 6364 | // dir shared caps? |
94b18763 | 6365 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae | 6366 | if (dn->cap_shared_gen == dir->shared_gen && |
94b18763 | 6367 | (!dn->inode || dn->inode->caps_issued_mask(mask, true))) |
7c673cae FG |
6368 | goto hit_dn; |
6369 | if (!dn->inode && (dir->flags & I_COMPLETE)) { | |
11fdf7f2 | 6370 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " |
7c673cae FG |
6371 | << *dir << " dn '" << dname << "'" << dendl; |
6372 | return -ENOENT; | |
6373 | } | |
6374 | } | |
6375 | } else { | |
6376 | ldout(cct, 20) << " no cap on " << dn->inode->vino() << dendl; | |
6377 | } | |
6378 | } else { | |
6379 | // can we conclude ENOENT locally? | |
94b18763 | 6380 | if (dir->caps_issued_mask(CEPH_CAP_FILE_SHARED, true) && |
7c673cae | 6381 | (dir->flags & I_COMPLETE)) { |
11fdf7f2 | 6382 | ldout(cct, 10) << __func__ << " concluded ENOENT locally for " << *dir << " dn '" << dname << "'" << dendl; |
7c673cae FG |
6383 | return -ENOENT; |
6384 | } | |
6385 | } | |
6386 | ||
6387 | r = _do_lookup(dir, dname, mask, target, perms); | |
6388 | goto done; | |
6389 | ||
6390 | hit_dn: | |
6391 | if (dn->inode) { | |
6392 | *target = dn->inode; | |
6393 | } else { | |
6394 | r = -ENOENT; | |
6395 | } | |
6396 | touch_dn(dn); | |
6397 | ||
6398 | done: | |
6399 | if (r < 0) | |
11fdf7f2 | 6400 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << r << dendl; |
7c673cae | 6401 | else |
11fdf7f2 | 6402 | ldout(cct, 10) << __func__ << " " << *dir << " " << dname << " = " << **target << dendl; |
7c673cae FG |
6403 | return r; |
6404 | } | |
6405 | ||
6406 | int Client::get_or_create(Inode *dir, const char* name, | |
6407 | Dentry **pdn, bool expect_null) | |
6408 | { | |
6409 | // lookup | |
11fdf7f2 | 6410 | ldout(cct, 20) << __func__ << " " << *dir << " name " << name << dendl; |
7c673cae FG |
6411 | dir->open_dir(); |
6412 | if (dir->dir->dentries.count(name)) { | |
6413 | Dentry *dn = dir->dir->dentries[name]; | |
6414 | ||
6415 | // is dn lease valid? | |
6416 | utime_t now = ceph_clock_now(); | |
6417 | if (dn->inode && | |
6418 | dn->lease_mds >= 0 && | |
6419 | dn->lease_ttl > now && | |
6420 | mds_sessions.count(dn->lease_mds)) { | |
11fdf7f2 TL |
6421 | MetaSession &s = mds_sessions.at(dn->lease_mds); |
6422 | if (s.cap_ttl > now && | |
6423 | s.cap_gen == dn->lease_gen) { | |
7c673cae FG |
6424 | if (expect_null) |
6425 | return -EEXIST; | |
6426 | } | |
6427 | } | |
6428 | *pdn = dn; | |
6429 | } else { | |
6430 | // otherwise link up a new one | |
6431 | *pdn = link(dir->dir, name, NULL, NULL); | |
6432 | } | |
6433 | ||
6434 | // success | |
6435 | return 0; | |
6436 | } | |
6437 | ||
6438 | int Client::path_walk(const filepath& origpath, InodeRef *end, | |
6439 | const UserPerm& perms, bool followsym, int mask) | |
6440 | { | |
6441 | filepath path = origpath; | |
6442 | InodeRef cur; | |
6443 | if (origpath.absolute()) | |
6444 | cur = root; | |
6445 | else | |
6446 | cur = cwd; | |
11fdf7f2 | 6447 | ceph_assert(cur); |
7c673cae | 6448 | |
11fdf7f2 | 6449 | ldout(cct, 10) << __func__ << " " << path << dendl; |
7c673cae FG |
6450 | |
6451 | int symlinks = 0; | |
6452 | ||
6453 | unsigned i=0; | |
6454 | while (i < path.depth() && cur) { | |
6455 | int caps = 0; | |
6456 | const string &dname = path[i]; | |
6457 | ldout(cct, 10) << " " << i << " " << *cur << " " << dname << dendl; | |
6458 | ldout(cct, 20) << " (path is " << path << ")" << dendl; | |
6459 | InodeRef next; | |
6460 | if (cct->_conf->client_permissions) { | |
6461 | int r = may_lookup(cur.get(), perms); | |
6462 | if (r < 0) | |
6463 | return r; | |
6464 | caps = CEPH_CAP_AUTH_SHARED; | |
6465 | } | |
6466 | ||
6467 | /* Get extra requested caps on the last component */ | |
6468 | if (i == (path.depth() - 1)) | |
6469 | caps |= mask; | |
6470 | int r = _lookup(cur.get(), dname, caps, &next, perms); | |
6471 | if (r < 0) | |
6472 | return r; | |
6473 | // only follow trailing symlink if followsym. always follow | |
6474 | // 'directory' symlinks. | |
6475 | if (next && next->is_symlink()) { | |
6476 | symlinks++; | |
6477 | ldout(cct, 20) << " symlink count " << symlinks << ", value is '" << next->symlink << "'" << dendl; | |
6478 | if (symlinks > MAXSYMLINKS) { | |
6479 | return -ELOOP; | |
6480 | } | |
6481 | ||
6482 | if (i < path.depth() - 1) { | |
6483 | // dir symlink | |
6484 | // replace consumed components of path with symlink dir target | |
6485 | filepath resolved(next->symlink.c_str()); | |
6486 | resolved.append(path.postfixpath(i + 1)); | |
6487 | path = resolved; | |
6488 | i = 0; | |
6489 | if (next->symlink[0] == '/') { | |
6490 | cur = root; | |
6491 | } | |
6492 | continue; | |
6493 | } else if (followsym) { | |
6494 | if (next->symlink[0] == '/') { | |
6495 | path = next->symlink.c_str(); | |
6496 | i = 0; | |
6497 | // reset position | |
6498 | cur = root; | |
6499 | } else { | |
6500 | filepath more(next->symlink.c_str()); | |
6501 | // we need to remove the symlink component from off of the path | |
6502 | // before adding the target that the symlink points to. remain | |
6503 | // at the same position in the path. | |
6504 | path.pop_dentry(); | |
6505 | path.append(more); | |
6506 | } | |
6507 | continue; | |
6508 | } | |
6509 | } | |
6510 | cur.swap(next); | |
6511 | i++; | |
6512 | } | |
6513 | if (!cur) | |
6514 | return -ENOENT; | |
6515 | if (end) | |
6516 | end->swap(cur); | |
6517 | return 0; | |
6518 | } | |
6519 | ||
6520 | ||
6521 | // namespace ops | |
6522 | ||
6523 | int Client::link(const char *relexisting, const char *relpath, const UserPerm& perm) | |
6524 | { | |
11fdf7f2 | 6525 | std::lock_guard lock(client_lock); |
7c673cae FG |
6526 | tout(cct) << "link" << std::endl; |
6527 | tout(cct) << relexisting << std::endl; | |
6528 | tout(cct) << relpath << std::endl; | |
6529 | ||
181888fb FG |
6530 | if (unmounting) |
6531 | return -ENOTCONN; | |
6532 | ||
7c673cae FG |
6533 | filepath existing(relexisting); |
6534 | ||
6535 | InodeRef in, dir; | |
6536 | int r = path_walk(existing, &in, perm, true); | |
6537 | if (r < 0) | |
6538 | return r; | |
6539 | if (std::string(relpath) == "/") { | |
6540 | r = -EEXIST; | |
6541 | return r; | |
6542 | } | |
6543 | filepath path(relpath); | |
6544 | string name = path.last_dentry(); | |
6545 | path.pop_dentry(); | |
6546 | ||
6547 | r = path_walk(path, &dir, perm, true); | |
6548 | if (r < 0) | |
6549 | return r; | |
6550 | if (cct->_conf->client_permissions) { | |
6551 | if (S_ISDIR(in->mode)) { | |
6552 | r = -EPERM; | |
6553 | return r; | |
6554 | } | |
6555 | r = may_hardlink(in.get(), perm); | |
6556 | if (r < 0) | |
6557 | return r; | |
6558 | r = may_create(dir.get(), perm); | |
6559 | if (r < 0) | |
6560 | return r; | |
6561 | } | |
6562 | r = _link(in.get(), dir.get(), name.c_str(), perm); | |
6563 | return r; | |
6564 | } | |
6565 | ||
6566 | int Client::unlink(const char *relpath, const UserPerm& perm) | |
6567 | { | |
11fdf7f2 TL |
6568 | std::lock_guard lock(client_lock); |
6569 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6570 | tout(cct) << relpath << std::endl; |
6571 | ||
181888fb FG |
6572 | if (unmounting) |
6573 | return -ENOTCONN; | |
6574 | ||
7c673cae FG |
6575 | if (std::string(relpath) == "/") |
6576 | return -EISDIR; | |
6577 | ||
6578 | filepath path(relpath); | |
6579 | string name = path.last_dentry(); | |
6580 | path.pop_dentry(); | |
6581 | InodeRef dir; | |
6582 | int r = path_walk(path, &dir, perm); | |
6583 | if (r < 0) | |
6584 | return r; | |
6585 | if (cct->_conf->client_permissions) { | |
6586 | r = may_delete(dir.get(), name.c_str(), perm); | |
6587 | if (r < 0) | |
6588 | return r; | |
6589 | } | |
6590 | return _unlink(dir.get(), name.c_str(), perm); | |
6591 | } | |
6592 | ||
6593 | int Client::rename(const char *relfrom, const char *relto, const UserPerm& perm) | |
6594 | { | |
11fdf7f2 TL |
6595 | std::lock_guard lock(client_lock); |
6596 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6597 | tout(cct) << relfrom << std::endl; |
6598 | tout(cct) << relto << std::endl; | |
6599 | ||
181888fb FG |
6600 | if (unmounting) |
6601 | return -ENOTCONN; | |
6602 | ||
7c673cae FG |
6603 | if (std::string(relfrom) == "/" || std::string(relto) == "/") |
6604 | return -EBUSY; | |
6605 | ||
6606 | filepath from(relfrom); | |
6607 | filepath to(relto); | |
6608 | string fromname = from.last_dentry(); | |
6609 | from.pop_dentry(); | |
6610 | string toname = to.last_dentry(); | |
6611 | to.pop_dentry(); | |
6612 | ||
6613 | InodeRef fromdir, todir; | |
6614 | int r = path_walk(from, &fromdir, perm); | |
6615 | if (r < 0) | |
6616 | goto out; | |
6617 | r = path_walk(to, &todir, perm); | |
6618 | if (r < 0) | |
6619 | goto out; | |
6620 | ||
6621 | if (cct->_conf->client_permissions) { | |
6622 | int r = may_delete(fromdir.get(), fromname.c_str(), perm); | |
6623 | if (r < 0) | |
6624 | return r; | |
6625 | r = may_delete(todir.get(), toname.c_str(), perm); | |
6626 | if (r < 0 && r != -ENOENT) | |
6627 | return r; | |
6628 | } | |
6629 | r = _rename(fromdir.get(), fromname.c_str(), todir.get(), toname.c_str(), perm); | |
6630 | out: | |
6631 | return r; | |
6632 | } | |
6633 | ||
6634 | // dirs | |
6635 | ||
6636 | int Client::mkdir(const char *relpath, mode_t mode, const UserPerm& perm) | |
6637 | { | |
11fdf7f2 TL |
6638 | std::lock_guard lock(client_lock); |
6639 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6640 | tout(cct) << relpath << std::endl; |
6641 | tout(cct) << mode << std::endl; | |
11fdf7f2 | 6642 | ldout(cct, 10) << __func__ << ": " << relpath << dendl; |
7c673cae | 6643 | |
181888fb FG |
6644 | if (unmounting) |
6645 | return -ENOTCONN; | |
6646 | ||
7c673cae FG |
6647 | if (std::string(relpath) == "/") |
6648 | return -EEXIST; | |
6649 | ||
6650 | filepath path(relpath); | |
6651 | string name = path.last_dentry(); | |
6652 | path.pop_dentry(); | |
6653 | InodeRef dir; | |
6654 | int r = path_walk(path, &dir, perm); | |
6655 | if (r < 0) | |
6656 | return r; | |
6657 | if (cct->_conf->client_permissions) { | |
6658 | r = may_create(dir.get(), perm); | |
6659 | if (r < 0) | |
6660 | return r; | |
6661 | } | |
6662 | return _mkdir(dir.get(), name.c_str(), mode, perm); | |
6663 | } | |
6664 | ||
6665 | int Client::mkdirs(const char *relpath, mode_t mode, const UserPerm& perms) | |
6666 | { | |
11fdf7f2 | 6667 | std::lock_guard lock(client_lock); |
7c673cae | 6668 | ldout(cct, 10) << "Client::mkdirs " << relpath << dendl; |
11fdf7f2 | 6669 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
6670 | tout(cct) << relpath << std::endl; |
6671 | tout(cct) << mode << std::endl; | |
6672 | ||
181888fb FG |
6673 | if (unmounting) |
6674 | return -ENOTCONN; | |
6675 | ||
7c673cae FG |
6676 | //get through existing parts of path |
6677 | filepath path(relpath); | |
6678 | unsigned int i; | |
6679 | int r = 0, caps = 0; | |
6680 | InodeRef cur, next; | |
6681 | cur = cwd; | |
6682 | for (i=0; i<path.depth(); ++i) { | |
6683 | if (cct->_conf->client_permissions) { | |
6684 | r = may_lookup(cur.get(), perms); | |
6685 | if (r < 0) | |
6686 | break; | |
6687 | caps = CEPH_CAP_AUTH_SHARED; | |
6688 | } | |
6689 | r = _lookup(cur.get(), path[i].c_str(), caps, &next, perms); | |
6690 | if (r < 0) | |
6691 | break; | |
6692 | cur.swap(next); | |
6693 | } | |
7c673cae | 6694 | if (r!=-ENOENT) return r; |
11fdf7f2 | 6695 | ldout(cct, 20) << __func__ << " got through " << i << " directories on path " << relpath << dendl; |
7c673cae FG |
6696 | //make new directory at each level |
6697 | for (; i<path.depth(); ++i) { | |
6698 | if (cct->_conf->client_permissions) { | |
6699 | r = may_create(cur.get(), perms); | |
6700 | if (r < 0) | |
6701 | return r; | |
6702 | } | |
6703 | //make new dir | |
6704 | r = _mkdir(cur.get(), path[i].c_str(), mode, perms, &next); | |
c07f9fc5 | 6705 | |
7c673cae | 6706 | //check proper creation/existence |
c07f9fc5 FG |
6707 | if(-EEXIST == r && i < path.depth() - 1) { |
6708 | r = _lookup(cur.get(), path[i].c_str(), CEPH_CAP_AUTH_SHARED, &next, perms); | |
6709 | } | |
6710 | if (r < 0) | |
6711 | return r; | |
7c673cae FG |
6712 | //move to new dir and continue |
6713 | cur.swap(next); | |
11fdf7f2 | 6714 | ldout(cct, 20) << __func__ << ": successfully created directory " |
7c673cae FG |
6715 | << filepath(cur->ino).get_path() << dendl; |
6716 | } | |
6717 | return 0; | |
6718 | } | |
6719 | ||
6720 | int Client::rmdir(const char *relpath, const UserPerm& perms) | |
6721 | { | |
11fdf7f2 TL |
6722 | std::lock_guard lock(client_lock); |
6723 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6724 | tout(cct) << relpath << std::endl; |
6725 | ||
181888fb FG |
6726 | if (unmounting) |
6727 | return -ENOTCONN; | |
6728 | ||
7c673cae FG |
6729 | if (std::string(relpath) == "/") |
6730 | return -EBUSY; | |
6731 | ||
6732 | filepath path(relpath); | |
6733 | string name = path.last_dentry(); | |
6734 | path.pop_dentry(); | |
6735 | InodeRef dir; | |
6736 | int r = path_walk(path, &dir, perms); | |
6737 | if (r < 0) | |
6738 | return r; | |
6739 | if (cct->_conf->client_permissions) { | |
6740 | int r = may_delete(dir.get(), name.c_str(), perms); | |
6741 | if (r < 0) | |
6742 | return r; | |
6743 | } | |
6744 | return _rmdir(dir.get(), name.c_str(), perms); | |
6745 | } | |
6746 | ||
6747 | int Client::mknod(const char *relpath, mode_t mode, const UserPerm& perms, dev_t rdev) | |
6748 | { | |
11fdf7f2 TL |
6749 | std::lock_guard lock(client_lock); |
6750 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6751 | tout(cct) << relpath << std::endl; |
6752 | tout(cct) << mode << std::endl; | |
6753 | tout(cct) << rdev << std::endl; | |
6754 | ||
181888fb FG |
6755 | if (unmounting) |
6756 | return -ENOTCONN; | |
6757 | ||
7c673cae FG |
6758 | if (std::string(relpath) == "/") |
6759 | return -EEXIST; | |
6760 | ||
6761 | filepath path(relpath); | |
6762 | string name = path.last_dentry(); | |
6763 | path.pop_dentry(); | |
6764 | InodeRef dir; | |
6765 | int r = path_walk(path, &dir, perms); | |
6766 | if (r < 0) | |
6767 | return r; | |
6768 | if (cct->_conf->client_permissions) { | |
6769 | int r = may_create(dir.get(), perms); | |
6770 | if (r < 0) | |
6771 | return r; | |
6772 | } | |
6773 | return _mknod(dir.get(), name.c_str(), mode, rdev, perms); | |
6774 | } | |
6775 | ||
6776 | // symlinks | |
6777 | ||
6778 | int Client::symlink(const char *target, const char *relpath, const UserPerm& perms) | |
6779 | { | |
11fdf7f2 TL |
6780 | std::lock_guard lock(client_lock); |
6781 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6782 | tout(cct) << target << std::endl; |
6783 | tout(cct) << relpath << std::endl; | |
6784 | ||
181888fb FG |
6785 | if (unmounting) |
6786 | return -ENOTCONN; | |
6787 | ||
7c673cae FG |
6788 | if (std::string(relpath) == "/") |
6789 | return -EEXIST; | |
6790 | ||
6791 | filepath path(relpath); | |
6792 | string name = path.last_dentry(); | |
6793 | path.pop_dentry(); | |
6794 | InodeRef dir; | |
6795 | int r = path_walk(path, &dir, perms); | |
6796 | if (r < 0) | |
6797 | return r; | |
6798 | if (cct->_conf->client_permissions) { | |
6799 | int r = may_create(dir.get(), perms); | |
6800 | if (r < 0) | |
6801 | return r; | |
6802 | } | |
6803 | return _symlink(dir.get(), name.c_str(), target, perms); | |
6804 | } | |
6805 | ||
6806 | int Client::readlink(const char *relpath, char *buf, loff_t size, const UserPerm& perms) | |
6807 | { | |
11fdf7f2 TL |
6808 | std::lock_guard lock(client_lock); |
6809 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
6810 | tout(cct) << relpath << std::endl; |
6811 | ||
181888fb FG |
6812 | if (unmounting) |
6813 | return -ENOTCONN; | |
6814 | ||
7c673cae FG |
6815 | filepath path(relpath); |
6816 | InodeRef in; | |
6817 | int r = path_walk(path, &in, perms, false); | |
6818 | if (r < 0) | |
6819 | return r; | |
6820 | ||
6821 | return _readlink(in.get(), buf, size); | |
6822 | } | |
6823 | ||
6824 | int Client::_readlink(Inode *in, char *buf, size_t size) | |
6825 | { | |
6826 | if (!in->is_symlink()) | |
6827 | return -EINVAL; | |
6828 | ||
6829 | // copy into buf (at most size bytes) | |
6830 | int r = in->symlink.length(); | |
6831 | if (r > (int)size) | |
6832 | r = size; | |
6833 | memcpy(buf, in->symlink.c_str(), r); | |
6834 | return r; | |
6835 | } | |
6836 | ||
6837 | ||
6838 | // inode stuff | |
6839 | ||
6840 | int Client::_getattr(Inode *in, int mask, const UserPerm& perms, bool force) | |
6841 | { | |
94b18763 | 6842 | bool yes = in->caps_issued_mask(mask, true); |
7c673cae | 6843 | |
11fdf7f2 | 6844 | ldout(cct, 10) << __func__ << " mask " << ccap_string(mask) << " issued=" << yes << dendl; |
7c673cae FG |
6845 | if (yes && !force) |
6846 | return 0; | |
6847 | ||
6848 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_GETATTR); | |
6849 | filepath path; | |
6850 | in->make_nosnap_relative_path(path); | |
6851 | req->set_filepath(path); | |
6852 | req->set_inode(in); | |
6853 | req->head.args.getattr.mask = mask; | |
6854 | ||
6855 | int res = make_request(req, perms); | |
11fdf7f2 | 6856 | ldout(cct, 10) << __func__ << " result=" << res << dendl; |
7c673cae FG |
6857 | return res; |
6858 | } | |
6859 | ||
6860 | int Client::_do_setattr(Inode *in, struct ceph_statx *stx, int mask, | |
6861 | const UserPerm& perms, InodeRef *inp) | |
6862 | { | |
6863 | int issued = in->caps_issued(); | |
6864 | ||
11fdf7f2 | 6865 | ldout(cct, 10) << __func__ << " mask " << mask << " issued " << |
7c673cae FG |
6866 | ccap_string(issued) << dendl; |
6867 | ||
6868 | if (in->snapid != CEPH_NOSNAP) { | |
6869 | return -EROFS; | |
6870 | } | |
6871 | if ((mask & CEPH_SETATTR_SIZE) && | |
6872 | (unsigned long)stx->stx_size > in->size && | |
6873 | is_quota_bytes_exceeded(in, (unsigned long)stx->stx_size - in->size, | |
6874 | perms)) { | |
6875 | return -EDQUOT; | |
6876 | } | |
6877 | ||
6878 | // make the change locally? | |
6879 | if ((in->cap_dirtier_uid >= 0 && perms.uid() != in->cap_dirtier_uid) || | |
6880 | (in->cap_dirtier_gid >= 0 && perms.gid() != in->cap_dirtier_gid)) { | |
6881 | ldout(cct, 10) << __func__ << " caller " << perms.uid() << ":" << perms.gid() | |
6882 | << " != cap dirtier " << in->cap_dirtier_uid << ":" | |
6883 | << in->cap_dirtier_gid << ", forcing sync setattr" | |
6884 | << dendl; | |
6885 | /* | |
6886 | * This works because we implicitly flush the caps as part of the | |
6887 | * request, so the cap update check will happen with the writeback | |
6888 | * cap context, and then the setattr check will happen with the | |
6889 | * caller's context. | |
6890 | * | |
6891 | * In reality this pattern is likely pretty rare (different users | |
6892 | * setattr'ing the same file). If that turns out not to be the | |
6893 | * case later, we can build a more complex pipelined cap writeback | |
6894 | * infrastructure... | |
6895 | */ | |
6896 | if (!mask) | |
6897 | mask |= CEPH_SETATTR_CTIME; | |
6898 | goto force_request; | |
6899 | } | |
6900 | ||
6901 | if (!mask) { | |
6902 | // caller just needs us to bump the ctime | |
6903 | in->ctime = ceph_clock_now(); | |
6904 | in->cap_dirtier_uid = perms.uid(); | |
6905 | in->cap_dirtier_gid = perms.gid(); | |
6906 | if (issued & CEPH_CAP_AUTH_EXCL) | |
28e407b8 | 6907 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae | 6908 | else if (issued & CEPH_CAP_FILE_EXCL) |
28e407b8 | 6909 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae | 6910 | else if (issued & CEPH_CAP_XATTR_EXCL) |
28e407b8 | 6911 | in->mark_caps_dirty(CEPH_CAP_XATTR_EXCL); |
7c673cae FG |
6912 | else |
6913 | mask |= CEPH_SETATTR_CTIME; | |
6914 | } | |
6915 | ||
6916 | if (in->caps_issued_mask(CEPH_CAP_AUTH_EXCL)) { | |
6917 | bool kill_sguid = mask & (CEPH_SETATTR_SIZE|CEPH_SETATTR_KILL_SGUID); | |
6918 | ||
6919 | mask &= ~CEPH_SETATTR_KILL_SGUID; | |
6920 | ||
6921 | if (mask & CEPH_SETATTR_UID) { | |
6922 | in->ctime = ceph_clock_now(); | |
6923 | in->cap_dirtier_uid = perms.uid(); | |
6924 | in->cap_dirtier_gid = perms.gid(); | |
6925 | in->uid = stx->stx_uid; | |
28e407b8 | 6926 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
6927 | mask &= ~CEPH_SETATTR_UID; |
6928 | kill_sguid = true; | |
6929 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
6930 | } | |
6931 | if (mask & CEPH_SETATTR_GID) { | |
6932 | in->ctime = ceph_clock_now(); | |
6933 | in->cap_dirtier_uid = perms.uid(); | |
6934 | in->cap_dirtier_gid = perms.gid(); | |
6935 | in->gid = stx->stx_gid; | |
28e407b8 | 6936 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
6937 | mask &= ~CEPH_SETATTR_GID; |
6938 | kill_sguid = true; | |
6939 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
6940 | } | |
6941 | ||
6942 | if (mask & CEPH_SETATTR_MODE) { | |
6943 | in->ctime = ceph_clock_now(); | |
6944 | in->cap_dirtier_uid = perms.uid(); | |
6945 | in->cap_dirtier_gid = perms.gid(); | |
6946 | in->mode = (in->mode & ~07777) | (stx->stx_mode & 07777); | |
28e407b8 | 6947 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
6948 | mask &= ~CEPH_SETATTR_MODE; |
6949 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
181888fb | 6950 | } else if (kill_sguid && S_ISREG(in->mode) && (in->mode & (S_IXUSR|S_IXGRP|S_IXOTH))) { |
7c673cae | 6951 | /* Must squash the any setuid/setgid bits with an ownership change */ |
181888fb | 6952 | in->mode &= ~(S_ISUID|S_ISGID); |
28e407b8 | 6953 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
6954 | } |
6955 | ||
6956 | if (mask & CEPH_SETATTR_BTIME) { | |
6957 | in->ctime = ceph_clock_now(); | |
6958 | in->cap_dirtier_uid = perms.uid(); | |
6959 | in->cap_dirtier_gid = perms.gid(); | |
6960 | in->btime = utime_t(stx->stx_btime); | |
28e407b8 | 6961 | in->mark_caps_dirty(CEPH_CAP_AUTH_EXCL); |
7c673cae FG |
6962 | mask &= ~CEPH_SETATTR_BTIME; |
6963 | ldout(cct,10) << "changing btime to " << in->btime << dendl; | |
6964 | } | |
6965 | } else if (mask & CEPH_SETATTR_SIZE) { | |
6966 | /* If we don't have Ax, then we must ask the server to clear them on truncate */ | |
6967 | mask |= CEPH_SETATTR_KILL_SGUID; | |
6968 | } | |
6969 | ||
6970 | if (in->caps_issued_mask(CEPH_CAP_FILE_EXCL)) { | |
6971 | if (mask & (CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME)) { | |
6972 | if (mask & CEPH_SETATTR_MTIME) | |
6973 | in->mtime = utime_t(stx->stx_mtime); | |
6974 | if (mask & CEPH_SETATTR_ATIME) | |
6975 | in->atime = utime_t(stx->stx_atime); | |
6976 | in->ctime = ceph_clock_now(); | |
6977 | in->cap_dirtier_uid = perms.uid(); | |
6978 | in->cap_dirtier_gid = perms.gid(); | |
6979 | in->time_warp_seq++; | |
28e407b8 | 6980 | in->mark_caps_dirty(CEPH_CAP_FILE_EXCL); |
7c673cae FG |
6981 | mask &= ~(CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME); |
6982 | } | |
6983 | } | |
6984 | if (!mask) { | |
6985 | in->change_attr++; | |
6986 | return 0; | |
6987 | } | |
6988 | ||
6989 | force_request: | |
6990 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETATTR); | |
6991 | ||
6992 | filepath path; | |
6993 | ||
6994 | in->make_nosnap_relative_path(path); | |
6995 | req->set_filepath(path); | |
6996 | req->set_inode(in); | |
6997 | ||
6998 | if (mask & CEPH_SETATTR_KILL_SGUID) { | |
6999 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7000 | } | |
7001 | if (mask & CEPH_SETATTR_MODE) { | |
7002 | req->head.args.setattr.mode = stx->stx_mode; | |
7003 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7004 | ldout(cct,10) << "changing mode to " << stx->stx_mode << dendl; | |
7005 | } | |
7006 | if (mask & CEPH_SETATTR_UID) { | |
7007 | req->head.args.setattr.uid = stx->stx_uid; | |
7008 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7009 | ldout(cct,10) << "changing uid to " << stx->stx_uid << dendl; | |
7010 | } | |
7011 | if (mask & CEPH_SETATTR_GID) { | |
7012 | req->head.args.setattr.gid = stx->stx_gid; | |
7013 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7014 | ldout(cct,10) << "changing gid to " << stx->stx_gid << dendl; | |
7015 | } | |
7016 | if (mask & CEPH_SETATTR_BTIME) { | |
7017 | req->head.args.setattr.btime = utime_t(stx->stx_btime); | |
7018 | req->inode_drop |= CEPH_CAP_AUTH_SHARED; | |
7019 | } | |
7020 | if (mask & CEPH_SETATTR_MTIME) { | |
7021 | req->head.args.setattr.mtime = utime_t(stx->stx_mtime); | |
94b18763 | 7022 | req->inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | |
7c673cae FG |
7023 | CEPH_CAP_FILE_WR; |
7024 | } | |
7025 | if (mask & CEPH_SETATTR_ATIME) { | |
7026 | req->head.args.setattr.atime = utime_t(stx->stx_atime); | |
7027 | req->inode_drop |= CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_RD | | |
7028 | CEPH_CAP_FILE_WR; | |
7029 | } | |
7030 | if (mask & CEPH_SETATTR_SIZE) { | |
7031 | if ((unsigned long)stx->stx_size < mdsmap->get_max_filesize()) { | |
7032 | req->head.args.setattr.size = stx->stx_size; | |
7033 | ldout(cct,10) << "changing size to " << stx->stx_size << dendl; | |
7034 | } else { //too big! | |
7035 | put_request(req); | |
7036 | ldout(cct,10) << "unable to set size to " << stx->stx_size << ". Too large!" << dendl; | |
7037 | return -EFBIG; | |
7038 | } | |
94b18763 | 7039 | req->inode_drop |= CEPH_CAP_FILE_SHARED | CEPH_CAP_FILE_RD | |
7c673cae FG |
7040 | CEPH_CAP_FILE_WR; |
7041 | } | |
7042 | req->head.args.setattr.mask = mask; | |
7043 | ||
7044 | req->regetattr_mask = mask; | |
7045 | ||
7046 | int res = make_request(req, perms, inp); | |
7047 | ldout(cct, 10) << "_setattr result=" << res << dendl; | |
7048 | return res; | |
7049 | } | |
7050 | ||
7051 | /* Note that we only care about attrs that setattr cares about */ | |
7052 | void Client::stat_to_statx(struct stat *st, struct ceph_statx *stx) | |
7053 | { | |
7054 | stx->stx_size = st->st_size; | |
7055 | stx->stx_mode = st->st_mode; | |
7056 | stx->stx_uid = st->st_uid; | |
7057 | stx->stx_gid = st->st_gid; | |
11fdf7f2 TL |
7058 | #ifdef __APPLE__ |
7059 | stx->stx_mtime = st->st_mtimespec; | |
7060 | stx->stx_atime = st->st_atimespec; | |
7061 | #else | |
7c673cae FG |
7062 | stx->stx_mtime = st->st_mtim; |
7063 | stx->stx_atime = st->st_atim; | |
11fdf7f2 | 7064 | #endif |
7c673cae FG |
7065 | } |
7066 | ||
7067 | int Client::__setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
7068 | const UserPerm& perms, InodeRef *inp) | |
7069 | { | |
7070 | int ret = _do_setattr(in, stx, mask, perms, inp); | |
7071 | if (ret < 0) | |
7072 | return ret; | |
7073 | if (mask & CEPH_SETATTR_MODE) | |
7074 | ret = _posix_acl_chmod(in, stx->stx_mode, perms); | |
7075 | return ret; | |
7076 | } | |
7077 | ||
7078 | int Client::_setattrx(InodeRef &in, struct ceph_statx *stx, int mask, | |
7079 | const UserPerm& perms) | |
7080 | { | |
7081 | mask &= (CEPH_SETATTR_MODE | CEPH_SETATTR_UID | | |
7082 | CEPH_SETATTR_GID | CEPH_SETATTR_MTIME | | |
7083 | CEPH_SETATTR_ATIME | CEPH_SETATTR_SIZE | | |
7084 | CEPH_SETATTR_CTIME | CEPH_SETATTR_BTIME); | |
7085 | if (cct->_conf->client_permissions) { | |
7086 | int r = may_setattr(in.get(), stx, mask, perms); | |
7087 | if (r < 0) | |
7088 | return r; | |
7089 | } | |
7090 | return __setattrx(in.get(), stx, mask, perms); | |
7091 | } | |
7092 | ||
7093 | int Client::_setattr(InodeRef &in, struct stat *attr, int mask, | |
7094 | const UserPerm& perms) | |
7095 | { | |
7096 | struct ceph_statx stx; | |
7097 | ||
7098 | stat_to_statx(attr, &stx); | |
7099 | mask &= ~CEPH_SETATTR_BTIME; | |
181888fb FG |
7100 | |
7101 | if ((mask & CEPH_SETATTR_UID) && attr->st_uid == static_cast<uid_t>(-1)) { | |
7102 | mask &= ~CEPH_SETATTR_UID; | |
7103 | } | |
7104 | if ((mask & CEPH_SETATTR_GID) && attr->st_gid == static_cast<uid_t>(-1)) { | |
7105 | mask &= ~CEPH_SETATTR_GID; | |
7106 | } | |
7107 | ||
7c673cae FG |
7108 | return _setattrx(in, &stx, mask, perms); |
7109 | } | |
7110 | ||
7111 | int Client::setattr(const char *relpath, struct stat *attr, int mask, | |
7112 | const UserPerm& perms) | |
7113 | { | |
11fdf7f2 TL |
7114 | std::lock_guard lock(client_lock); |
7115 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7116 | tout(cct) << relpath << std::endl; |
7117 | tout(cct) << mask << std::endl; | |
7118 | ||
181888fb FG |
7119 | if (unmounting) |
7120 | return -ENOTCONN; | |
7121 | ||
7c673cae FG |
7122 | filepath path(relpath); |
7123 | InodeRef in; | |
7124 | int r = path_walk(path, &in, perms); | |
7125 | if (r < 0) | |
7126 | return r; | |
7127 | return _setattr(in, attr, mask, perms); | |
7128 | } | |
7129 | ||
7130 | int Client::setattrx(const char *relpath, struct ceph_statx *stx, int mask, | |
7131 | const UserPerm& perms, int flags) | |
7132 | { | |
11fdf7f2 TL |
7133 | std::lock_guard lock(client_lock); |
7134 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7135 | tout(cct) << relpath << std::endl; |
7136 | tout(cct) << mask << std::endl; | |
7137 | ||
181888fb FG |
7138 | if (unmounting) |
7139 | return -ENOTCONN; | |
7140 | ||
7c673cae FG |
7141 | filepath path(relpath); |
7142 | InodeRef in; | |
7143 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW)); | |
7144 | if (r < 0) | |
7145 | return r; | |
7146 | return _setattrx(in, stx, mask, perms); | |
7147 | } | |
7148 | ||
7149 | int Client::fsetattr(int fd, struct stat *attr, int mask, const UserPerm& perms) | |
7150 | { | |
11fdf7f2 TL |
7151 | std::lock_guard lock(client_lock); |
7152 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7153 | tout(cct) << fd << std::endl; |
7154 | tout(cct) << mask << std::endl; | |
7155 | ||
181888fb FG |
7156 | if (unmounting) |
7157 | return -ENOTCONN; | |
7158 | ||
7c673cae FG |
7159 | Fh *f = get_filehandle(fd); |
7160 | if (!f) | |
7161 | return -EBADF; | |
7162 | #if defined(__linux__) && defined(O_PATH) | |
7163 | if (f->flags & O_PATH) | |
7164 | return -EBADF; | |
7165 | #endif | |
7166 | return _setattr(f->inode, attr, mask, perms); | |
7167 | } | |
7168 | ||
7169 | int Client::fsetattrx(int fd, struct ceph_statx *stx, int mask, const UserPerm& perms) | |
7170 | { | |
11fdf7f2 TL |
7171 | std::lock_guard lock(client_lock); |
7172 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7173 | tout(cct) << fd << std::endl; |
7174 | tout(cct) << mask << std::endl; | |
7175 | ||
181888fb FG |
7176 | if (unmounting) |
7177 | return -ENOTCONN; | |
7178 | ||
7c673cae FG |
7179 | Fh *f = get_filehandle(fd); |
7180 | if (!f) | |
7181 | return -EBADF; | |
7182 | #if defined(__linux__) && defined(O_PATH) | |
7183 | if (f->flags & O_PATH) | |
7184 | return -EBADF; | |
7185 | #endif | |
7186 | return _setattrx(f->inode, stx, mask, perms); | |
7187 | } | |
7188 | ||
7189 | int Client::stat(const char *relpath, struct stat *stbuf, const UserPerm& perms, | |
7190 | frag_info_t *dirstat, int mask) | |
7191 | { | |
11fdf7f2 TL |
7192 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
7193 | std::lock_guard lock(client_lock); | |
7c673cae FG |
7194 | tout(cct) << "stat" << std::endl; |
7195 | tout(cct) << relpath << std::endl; | |
181888fb FG |
7196 | |
7197 | if (unmounting) | |
7198 | return -ENOTCONN; | |
7199 | ||
7c673cae FG |
7200 | filepath path(relpath); |
7201 | InodeRef in; | |
7202 | int r = path_walk(path, &in, perms, true, mask); | |
7203 | if (r < 0) | |
7204 | return r; | |
7205 | r = _getattr(in, mask, perms); | |
7206 | if (r < 0) { | |
11fdf7f2 | 7207 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7208 | return r; |
7209 | } | |
7210 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7211 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7212 | return r; |
7213 | } | |
7214 | ||
7215 | unsigned Client::statx_to_mask(unsigned int flags, unsigned int want) | |
7216 | { | |
7217 | unsigned mask = 0; | |
7218 | ||
7219 | /* if NO_ATTR_SYNC is set, then we don't need any -- just use what's in cache */ | |
7220 | if (flags & AT_NO_ATTR_SYNC) | |
7221 | goto out; | |
7222 | ||
7223 | /* Always set PIN to distinguish from AT_NO_ATTR_SYNC case */ | |
7224 | mask |= CEPH_CAP_PIN; | |
7225 | if (want & (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7226 | mask |= CEPH_CAP_AUTH_SHARED; | |
7227 | if (want & (CEPH_STATX_NLINK|CEPH_STATX_CTIME|CEPH_STATX_VERSION)) | |
7228 | mask |= CEPH_CAP_LINK_SHARED; | |
7229 | if (want & (CEPH_STATX_ATIME|CEPH_STATX_MTIME|CEPH_STATX_CTIME|CEPH_STATX_SIZE|CEPH_STATX_BLOCKS|CEPH_STATX_VERSION)) | |
7230 | mask |= CEPH_CAP_FILE_SHARED; | |
7231 | if (want & (CEPH_STATX_VERSION|CEPH_STATX_CTIME)) | |
7232 | mask |= CEPH_CAP_XATTR_SHARED; | |
7233 | out: | |
7234 | return mask; | |
7235 | } | |
7236 | ||
7237 | int Client::statx(const char *relpath, struct ceph_statx *stx, | |
7238 | const UserPerm& perms, | |
7239 | unsigned int want, unsigned int flags) | |
7240 | { | |
11fdf7f2 TL |
7241 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " want " << want << ")" << dendl; |
7242 | std::lock_guard lock(client_lock); | |
7c673cae FG |
7243 | tout(cct) << "statx" << std::endl; |
7244 | tout(cct) << relpath << std::endl; | |
181888fb FG |
7245 | |
7246 | if (unmounting) | |
7247 | return -ENOTCONN; | |
7248 | ||
7c673cae FG |
7249 | filepath path(relpath); |
7250 | InodeRef in; | |
7251 | ||
7252 | unsigned mask = statx_to_mask(flags, want); | |
7253 | ||
7254 | int r = path_walk(path, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); | |
7255 | if (r < 0) | |
7256 | return r; | |
7257 | ||
7258 | r = _getattr(in, mask, perms); | |
7259 | if (r < 0) { | |
11fdf7f2 | 7260 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7261 | return r; |
7262 | } | |
7263 | ||
7264 | fill_statx(in, mask, stx); | |
11fdf7f2 | 7265 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << stx->stx_mask << ")" << dendl; |
7c673cae FG |
7266 | return r; |
7267 | } | |
7268 | ||
7269 | int Client::lstat(const char *relpath, struct stat *stbuf, | |
7270 | const UserPerm& perms, frag_info_t *dirstat, int mask) | |
7271 | { | |
11fdf7f2 TL |
7272 | ldout(cct, 3) << __func__ << " enter (relpath " << relpath << " mask " << mask << ")" << dendl; |
7273 | std::lock_guard lock(client_lock); | |
7274 | tout(cct) << __func__ << std::endl; | |
7c673cae | 7275 | tout(cct) << relpath << std::endl; |
181888fb FG |
7276 | |
7277 | if (unmounting) | |
7278 | return -ENOTCONN; | |
7279 | ||
7c673cae FG |
7280 | filepath path(relpath); |
7281 | InodeRef in; | |
7282 | // don't follow symlinks | |
7283 | int r = path_walk(path, &in, perms, false, mask); | |
7284 | if (r < 0) | |
7285 | return r; | |
7286 | r = _getattr(in, mask, perms); | |
7287 | if (r < 0) { | |
11fdf7f2 | 7288 | ldout(cct, 3) << __func__ << " exit on error!" << dendl; |
7c673cae FG |
7289 | return r; |
7290 | } | |
7291 | fill_stat(in, stbuf, dirstat); | |
11fdf7f2 | 7292 | ldout(cct, 3) << __func__ << " exit (relpath " << relpath << " mask " << mask << ")" << dendl; |
7c673cae FG |
7293 | return r; |
7294 | } | |
7295 | ||
7296 | int Client::fill_stat(Inode *in, struct stat *st, frag_info_t *dirstat, nest_info_t *rstat) | |
7297 | { | |
11fdf7f2 | 7298 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
7299 | << " mode 0" << oct << in->mode << dec |
7300 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7301 | memset(st, 0, sizeof(struct stat)); | |
7302 | if (use_faked_inos()) | |
7303 | st->st_ino = in->faked_ino; | |
7304 | else | |
7305 | st->st_ino = in->ino; | |
7306 | st->st_dev = in->snapid; | |
7307 | st->st_mode = in->mode; | |
7308 | st->st_rdev = in->rdev; | |
28e407b8 AA |
7309 | if (in->is_dir()) { |
7310 | switch (in->nlink) { | |
7311 | case 0: | |
7312 | st->st_nlink = 0; /* dir is unlinked */ | |
7313 | break; | |
7314 | case 1: | |
7315 | st->st_nlink = 1 /* parent dentry */ | |
7316 | + 1 /* <dir>/. */ | |
7317 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
7318 | break; | |
7319 | default: | |
7320 | ceph_abort(); | |
7321 | } | |
7322 | } else { | |
7323 | st->st_nlink = in->nlink; | |
7324 | } | |
7c673cae FG |
7325 | st->st_uid = in->uid; |
7326 | st->st_gid = in->gid; | |
7327 | if (in->ctime > in->mtime) { | |
7328 | stat_set_ctime_sec(st, in->ctime.sec()); | |
7329 | stat_set_ctime_nsec(st, in->ctime.nsec()); | |
7330 | } else { | |
7331 | stat_set_ctime_sec(st, in->mtime.sec()); | |
7332 | stat_set_ctime_nsec(st, in->mtime.nsec()); | |
7333 | } | |
7334 | stat_set_atime_sec(st, in->atime.sec()); | |
7335 | stat_set_atime_nsec(st, in->atime.nsec()); | |
7336 | stat_set_mtime_sec(st, in->mtime.sec()); | |
7337 | stat_set_mtime_nsec(st, in->mtime.nsec()); | |
7338 | if (in->is_dir()) { | |
7339 | if (cct->_conf->client_dirsize_rbytes) | |
7340 | st->st_size = in->rstat.rbytes; | |
7341 | else | |
7342 | st->st_size = in->dirstat.size(); | |
7343 | st->st_blocks = 1; | |
7344 | } else { | |
7345 | st->st_size = in->size; | |
7346 | st->st_blocks = (in->size + 511) >> 9; | |
7347 | } | |
11fdf7f2 | 7348 | st->st_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
7c673cae FG |
7349 | |
7350 | if (dirstat) | |
7351 | *dirstat = in->dirstat; | |
7352 | if (rstat) | |
7353 | *rstat = in->rstat; | |
7354 | ||
7355 | return in->caps_issued(); | |
7356 | } | |
7357 | ||
7358 | void Client::fill_statx(Inode *in, unsigned int mask, struct ceph_statx *stx) | |
7359 | { | |
11fdf7f2 | 7360 | ldout(cct, 10) << __func__ << " on " << in->ino << " snap/dev" << in->snapid |
7c673cae FG |
7361 | << " mode 0" << oct << in->mode << dec |
7362 | << " mtime " << in->mtime << " ctime " << in->ctime << dendl; | |
7363 | memset(stx, 0, sizeof(struct ceph_statx)); | |
7364 | ||
7365 | /* | |
7366 | * If mask is 0, then the caller set AT_NO_ATTR_SYNC. Reset the mask | |
7367 | * so that all bits are set. | |
7368 | */ | |
7369 | if (!mask) | |
7370 | mask = ~0; | |
7371 | ||
7372 | /* These are always considered to be available */ | |
7373 | stx->stx_dev = in->snapid; | |
11fdf7f2 | 7374 | stx->stx_blksize = std::max<uint32_t>(in->layout.stripe_unit, 4096); |
7c673cae FG |
7375 | |
7376 | /* Type bits are always set, even when CEPH_STATX_MODE is not */ | |
7377 | stx->stx_mode = S_IFMT & in->mode; | |
7378 | stx->stx_ino = use_faked_inos() ? in->faked_ino : (ino_t)in->ino; | |
7379 | stx->stx_rdev = in->rdev; | |
7380 | stx->stx_mask |= (CEPH_STATX_INO|CEPH_STATX_RDEV); | |
7381 | ||
7382 | if (mask & CEPH_CAP_AUTH_SHARED) { | |
7383 | stx->stx_uid = in->uid; | |
7384 | stx->stx_gid = in->gid; | |
7385 | stx->stx_mode = in->mode; | |
7386 | in->btime.to_timespec(&stx->stx_btime); | |
7387 | stx->stx_mask |= (CEPH_STATX_MODE|CEPH_STATX_UID|CEPH_STATX_GID|CEPH_STATX_BTIME); | |
7388 | } | |
7389 | ||
7390 | if (mask & CEPH_CAP_LINK_SHARED) { | |
28e407b8 AA |
7391 | if (in->is_dir()) { |
7392 | switch (in->nlink) { | |
7393 | case 0: | |
7394 | stx->stx_nlink = 0; /* dir is unlinked */ | |
7395 | break; | |
7396 | case 1: | |
7397 | stx->stx_nlink = 1 /* parent dentry */ | |
7398 | + 1 /* <dir>/. */ | |
7399 | + in->dirstat.nsubdirs; /* include <dir>/. self-reference */ | |
7400 | break; | |
7401 | default: | |
7402 | ceph_abort(); | |
7403 | } | |
7404 | } else { | |
7405 | stx->stx_nlink = in->nlink; | |
7406 | } | |
7c673cae FG |
7407 | stx->stx_mask |= CEPH_STATX_NLINK; |
7408 | } | |
7409 | ||
7410 | if (mask & CEPH_CAP_FILE_SHARED) { | |
7411 | ||
7412 | in->atime.to_timespec(&stx->stx_atime); | |
7413 | in->mtime.to_timespec(&stx->stx_mtime); | |
7414 | ||
7415 | if (in->is_dir()) { | |
7416 | if (cct->_conf->client_dirsize_rbytes) | |
7417 | stx->stx_size = in->rstat.rbytes; | |
7418 | else | |
7419 | stx->stx_size = in->dirstat.size(); | |
7420 | stx->stx_blocks = 1; | |
7421 | } else { | |
7422 | stx->stx_size = in->size; | |
7423 | stx->stx_blocks = (in->size + 511) >> 9; | |
7424 | } | |
7425 | stx->stx_mask |= (CEPH_STATX_ATIME|CEPH_STATX_MTIME| | |
7426 | CEPH_STATX_SIZE|CEPH_STATX_BLOCKS); | |
7427 | } | |
7428 | ||
7429 | /* Change time and change_attr both require all shared caps to view */ | |
7430 | if ((mask & CEPH_STAT_CAP_INODE_ALL) == CEPH_STAT_CAP_INODE_ALL) { | |
7431 | stx->stx_version = in->change_attr; | |
7432 | if (in->ctime > in->mtime) | |
7433 | in->ctime.to_timespec(&stx->stx_ctime); | |
7434 | else | |
7435 | in->mtime.to_timespec(&stx->stx_ctime); | |
7436 | stx->stx_mask |= (CEPH_STATX_CTIME|CEPH_STATX_VERSION); | |
7437 | } | |
7438 | ||
7439 | } | |
7440 | ||
7441 | void Client::touch_dn(Dentry *dn) | |
7442 | { | |
7443 | lru.lru_touch(dn); | |
7444 | } | |
7445 | ||
7446 | int Client::chmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
7447 | { | |
11fdf7f2 TL |
7448 | std::lock_guard lock(client_lock); |
7449 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7450 | tout(cct) << relpath << std::endl; |
7451 | tout(cct) << mode << std::endl; | |
181888fb FG |
7452 | |
7453 | if (unmounting) | |
7454 | return -ENOTCONN; | |
7455 | ||
7c673cae FG |
7456 | filepath path(relpath); |
7457 | InodeRef in; | |
7458 | int r = path_walk(path, &in, perms); | |
7459 | if (r < 0) | |
7460 | return r; | |
7461 | struct stat attr; | |
7462 | attr.st_mode = mode; | |
7463 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
7464 | } | |
7465 | ||
7466 | int Client::fchmod(int fd, mode_t mode, const UserPerm& perms) | |
7467 | { | |
11fdf7f2 TL |
7468 | std::lock_guard lock(client_lock); |
7469 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7470 | tout(cct) << fd << std::endl; |
7471 | tout(cct) << mode << std::endl; | |
181888fb FG |
7472 | |
7473 | if (unmounting) | |
7474 | return -ENOTCONN; | |
7475 | ||
7c673cae FG |
7476 | Fh *f = get_filehandle(fd); |
7477 | if (!f) | |
7478 | return -EBADF; | |
7479 | #if defined(__linux__) && defined(O_PATH) | |
7480 | if (f->flags & O_PATH) | |
7481 | return -EBADF; | |
7482 | #endif | |
7483 | struct stat attr; | |
7484 | attr.st_mode = mode; | |
7485 | return _setattr(f->inode, &attr, CEPH_SETATTR_MODE, perms); | |
7486 | } | |
7487 | ||
7488 | int Client::lchmod(const char *relpath, mode_t mode, const UserPerm& perms) | |
7489 | { | |
11fdf7f2 TL |
7490 | std::lock_guard lock(client_lock); |
7491 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7492 | tout(cct) << relpath << std::endl; |
7493 | tout(cct) << mode << std::endl; | |
181888fb FG |
7494 | |
7495 | if (unmounting) | |
7496 | return -ENOTCONN; | |
7497 | ||
7c673cae FG |
7498 | filepath path(relpath); |
7499 | InodeRef in; | |
7500 | // don't follow symlinks | |
7501 | int r = path_walk(path, &in, perms, false); | |
7502 | if (r < 0) | |
7503 | return r; | |
7504 | struct stat attr; | |
7505 | attr.st_mode = mode; | |
7506 | return _setattr(in, &attr, CEPH_SETATTR_MODE, perms); | |
7507 | } | |
7508 | ||
7509 | int Client::chown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
7510 | const UserPerm& perms) | |
7511 | { | |
11fdf7f2 TL |
7512 | std::lock_guard lock(client_lock); |
7513 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7514 | tout(cct) << relpath << std::endl; |
7515 | tout(cct) << new_uid << std::endl; | |
7516 | tout(cct) << new_gid << std::endl; | |
181888fb FG |
7517 | |
7518 | if (unmounting) | |
7519 | return -ENOTCONN; | |
7520 | ||
7c673cae FG |
7521 | filepath path(relpath); |
7522 | InodeRef in; | |
7523 | int r = path_walk(path, &in, perms); | |
7524 | if (r < 0) | |
7525 | return r; | |
7526 | struct stat attr; | |
7527 | attr.st_uid = new_uid; | |
7528 | attr.st_gid = new_gid; | |
181888fb | 7529 | return _setattr(in, &attr, CEPH_SETATTR_UID|CEPH_SETATTR_GID, perms); |
7c673cae FG |
7530 | } |
7531 | ||
7532 | int Client::fchown(int fd, uid_t new_uid, gid_t new_gid, const UserPerm& perms) | |
7533 | { | |
11fdf7f2 TL |
7534 | std::lock_guard lock(client_lock); |
7535 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7536 | tout(cct) << fd << std::endl; |
7537 | tout(cct) << new_uid << std::endl; | |
7538 | tout(cct) << new_gid << std::endl; | |
181888fb FG |
7539 | |
7540 | if (unmounting) | |
7541 | return -ENOTCONN; | |
7542 | ||
7c673cae FG |
7543 | Fh *f = get_filehandle(fd); |
7544 | if (!f) | |
7545 | return -EBADF; | |
7546 | #if defined(__linux__) && defined(O_PATH) | |
7547 | if (f->flags & O_PATH) | |
7548 | return -EBADF; | |
7549 | #endif | |
7550 | struct stat attr; | |
7551 | attr.st_uid = new_uid; | |
7552 | attr.st_gid = new_gid; | |
7553 | int mask = 0; | |
7554 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
7555 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
7556 | return _setattr(f->inode, &attr, mask, perms); | |
7557 | } | |
7558 | ||
7559 | int Client::lchown(const char *relpath, uid_t new_uid, gid_t new_gid, | |
7560 | const UserPerm& perms) | |
7561 | { | |
11fdf7f2 TL |
7562 | std::lock_guard lock(client_lock); |
7563 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7564 | tout(cct) << relpath << std::endl; |
7565 | tout(cct) << new_uid << std::endl; | |
7566 | tout(cct) << new_gid << std::endl; | |
181888fb FG |
7567 | |
7568 | if (unmounting) | |
7569 | return -ENOTCONN; | |
7570 | ||
7c673cae FG |
7571 | filepath path(relpath); |
7572 | InodeRef in; | |
7573 | // don't follow symlinks | |
7574 | int r = path_walk(path, &in, perms, false); | |
7575 | if (r < 0) | |
7576 | return r; | |
7577 | struct stat attr; | |
7578 | attr.st_uid = new_uid; | |
7579 | attr.st_gid = new_gid; | |
7580 | int mask = 0; | |
7581 | if (new_uid != static_cast<uid_t>(-1)) mask |= CEPH_SETATTR_UID; | |
7582 | if (new_gid != static_cast<gid_t>(-1)) mask |= CEPH_SETATTR_GID; | |
7583 | return _setattr(in, &attr, mask, perms); | |
7584 | } | |
7585 | ||
11fdf7f2 TL |
7586 | static void attr_set_atime_and_mtime(struct stat *attr, |
7587 | const utime_t &atime, | |
7588 | const utime_t &mtime) | |
7589 | { | |
7590 | stat_set_atime_sec(attr, atime.tv.tv_sec); | |
7591 | stat_set_atime_nsec(attr, atime.tv.tv_nsec); | |
7592 | stat_set_mtime_sec(attr, mtime.tv.tv_sec); | |
7593 | stat_set_mtime_nsec(attr, mtime.tv.tv_nsec); | |
7594 | } | |
7595 | ||
7596 | // for [l]utime() invoke the timeval variant as the timespec | |
7597 | // variant are not yet implemented. for futime[s](), invoke | |
7598 | // the timespec variant. | |
7c673cae FG |
7599 | int Client::utime(const char *relpath, struct utimbuf *buf, |
7600 | const UserPerm& perms) | |
7601 | { | |
11fdf7f2 TL |
7602 | struct timeval tv[2]; |
7603 | tv[0].tv_sec = buf->actime; | |
7604 | tv[0].tv_usec = 0; | |
7605 | tv[1].tv_sec = buf->modtime; | |
7606 | tv[1].tv_usec = 0; | |
7607 | ||
7608 | return utimes(relpath, tv, perms); | |
7609 | } | |
7610 | ||
7611 | int Client::lutime(const char *relpath, struct utimbuf *buf, | |
7612 | const UserPerm& perms) | |
7613 | { | |
7614 | struct timeval tv[2]; | |
7615 | tv[0].tv_sec = buf->actime; | |
7616 | tv[0].tv_usec = 0; | |
7617 | tv[1].tv_sec = buf->modtime; | |
7618 | tv[1].tv_usec = 0; | |
7619 | ||
7620 | return lutimes(relpath, tv, perms); | |
7621 | } | |
7622 | ||
7623 | int Client::futime(int fd, struct utimbuf *buf, const UserPerm& perms) | |
7624 | { | |
7625 | struct timespec ts[2]; | |
7626 | ts[0].tv_sec = buf->actime; | |
7627 | ts[0].tv_nsec = 0; | |
7628 | ts[1].tv_sec = buf->modtime; | |
7629 | ts[1].tv_nsec = 0; | |
7630 | ||
7631 | return futimens(fd, ts, perms); | |
7632 | } | |
7633 | ||
7634 | int Client::utimes(const char *relpath, struct timeval times[2], | |
7635 | const UserPerm& perms) | |
7636 | { | |
7637 | std::lock_guard lock(client_lock); | |
7638 | tout(cct) << __func__ << std::endl; | |
7c673cae | 7639 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
7640 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
7641 | << std::endl; | |
7642 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
7643 | << std::endl; | |
181888fb FG |
7644 | |
7645 | if (unmounting) | |
7646 | return -ENOTCONN; | |
7647 | ||
7c673cae FG |
7648 | filepath path(relpath); |
7649 | InodeRef in; | |
7650 | int r = path_walk(path, &in, perms); | |
7651 | if (r < 0) | |
7652 | return r; | |
7653 | struct stat attr; | |
11fdf7f2 TL |
7654 | utime_t atime(times[0]); |
7655 | utime_t mtime(times[1]); | |
7656 | ||
7657 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
7658 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
7659 | } | |
7660 | ||
11fdf7f2 TL |
7661 | int Client::lutimes(const char *relpath, struct timeval times[2], |
7662 | const UserPerm& perms) | |
7c673cae | 7663 | { |
11fdf7f2 TL |
7664 | std::lock_guard lock(client_lock); |
7665 | tout(cct) << __func__ << std::endl; | |
7c673cae | 7666 | tout(cct) << relpath << std::endl; |
11fdf7f2 TL |
7667 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_usec |
7668 | << std::endl; | |
7669 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_usec | |
7670 | << std::endl; | |
181888fb FG |
7671 | |
7672 | if (unmounting) | |
7673 | return -ENOTCONN; | |
7674 | ||
7c673cae FG |
7675 | filepath path(relpath); |
7676 | InodeRef in; | |
7c673cae FG |
7677 | int r = path_walk(path, &in, perms, false); |
7678 | if (r < 0) | |
7679 | return r; | |
7680 | struct stat attr; | |
11fdf7f2 TL |
7681 | utime_t atime(times[0]); |
7682 | utime_t mtime(times[1]); | |
7683 | ||
7684 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7c673cae FG |
7685 | return _setattr(in, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); |
7686 | } | |
7687 | ||
11fdf7f2 TL |
7688 | int Client::futimes(int fd, struct timeval times[2], const UserPerm& perms) |
7689 | { | |
7690 | struct timespec ts[2]; | |
7691 | ts[0].tv_sec = times[0].tv_sec; | |
7692 | ts[0].tv_nsec = times[0].tv_usec * 1000; | |
7693 | ts[1].tv_sec = times[1].tv_sec; | |
7694 | ts[1].tv_nsec = times[1].tv_usec * 1000; | |
7695 | ||
7696 | return futimens(fd, ts, perms); | |
7697 | } | |
7698 | ||
7699 | int Client::futimens(int fd, struct timespec times[2], const UserPerm& perms) | |
7700 | { | |
7701 | std::lock_guard lock(client_lock); | |
7702 | tout(cct) << __func__ << std::endl; | |
7703 | tout(cct) << fd << std::endl; | |
7704 | tout(cct) << "atime: " << times[0].tv_sec << "." << times[0].tv_nsec | |
7705 | << std::endl; | |
7706 | tout(cct) << "mtime: " << times[1].tv_sec << "." << times[1].tv_nsec | |
7707 | << std::endl; | |
7708 | ||
7709 | if (unmounting) | |
7710 | return -ENOTCONN; | |
7711 | ||
7712 | Fh *f = get_filehandle(fd); | |
7713 | if (!f) | |
7714 | return -EBADF; | |
7715 | #if defined(__linux__) && defined(O_PATH) | |
7716 | if (f->flags & O_PATH) | |
7717 | return -EBADF; | |
7718 | #endif | |
7719 | struct stat attr; | |
7720 | utime_t atime(times[0]); | |
7721 | utime_t mtime(times[1]); | |
7722 | ||
7723 | attr_set_atime_and_mtime(&attr, atime, mtime); | |
7724 | return _setattr(f->inode, &attr, CEPH_SETATTR_MTIME|CEPH_SETATTR_ATIME, perms); | |
7725 | } | |
7726 | ||
7c673cae FG |
7727 | int Client::flock(int fd, int operation, uint64_t owner) |
7728 | { | |
11fdf7f2 TL |
7729 | std::lock_guard lock(client_lock); |
7730 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7731 | tout(cct) << fd << std::endl; |
7732 | tout(cct) << operation << std::endl; | |
7733 | tout(cct) << owner << std::endl; | |
181888fb FG |
7734 | |
7735 | if (unmounting) | |
7736 | return -ENOTCONN; | |
7737 | ||
7c673cae FG |
7738 | Fh *f = get_filehandle(fd); |
7739 | if (!f) | |
7740 | return -EBADF; | |
7741 | ||
7742 | return _flock(f, operation, owner); | |
7743 | } | |
7744 | ||
7745 | int Client::opendir(const char *relpath, dir_result_t **dirpp, const UserPerm& perms) | |
7746 | { | |
11fdf7f2 TL |
7747 | std::lock_guard lock(client_lock); |
7748 | tout(cct) << __func__ << std::endl; | |
7c673cae | 7749 | tout(cct) << relpath << std::endl; |
181888fb FG |
7750 | |
7751 | if (unmounting) | |
7752 | return -ENOTCONN; | |
7753 | ||
7c673cae FG |
7754 | filepath path(relpath); |
7755 | InodeRef in; | |
7756 | int r = path_walk(path, &in, perms, true); | |
7757 | if (r < 0) | |
7758 | return r; | |
7759 | if (cct->_conf->client_permissions) { | |
7760 | int r = may_open(in.get(), O_RDONLY, perms); | |
7761 | if (r < 0) | |
7762 | return r; | |
7763 | } | |
7764 | r = _opendir(in.get(), dirpp, perms); | |
7765 | /* if ENOTDIR, dirpp will be an uninitialized point and it's very dangerous to access its value */ | |
7766 | if (r != -ENOTDIR) | |
7767 | tout(cct) << (unsigned long)*dirpp << std::endl; | |
7768 | return r; | |
7769 | } | |
7770 | ||
7771 | int Client::_opendir(Inode *in, dir_result_t **dirpp, const UserPerm& perms) | |
7772 | { | |
7773 | if (!in->is_dir()) | |
7774 | return -ENOTDIR; | |
7775 | *dirpp = new dir_result_t(in, perms); | |
7776 | opened_dirs.insert(*dirpp); | |
11fdf7f2 | 7777 | ldout(cct, 8) << __func__ << "(" << in->ino << ") = " << 0 << " (" << *dirpp << ")" << dendl; |
7c673cae FG |
7778 | return 0; |
7779 | } | |
7780 | ||
7781 | ||
7782 | int Client::closedir(dir_result_t *dir) | |
7783 | { | |
11fdf7f2 TL |
7784 | std::lock_guard lock(client_lock); |
7785 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
7786 | tout(cct) << (unsigned long)dir << std::endl; |
7787 | ||
11fdf7f2 | 7788 | ldout(cct, 3) << __func__ << "(" << dir << ") = 0" << dendl; |
7c673cae FG |
7789 | _closedir(dir); |
7790 | return 0; | |
7791 | } | |
7792 | ||
7793 | void Client::_closedir(dir_result_t *dirp) | |
7794 | { | |
11fdf7f2 | 7795 | ldout(cct, 10) << __func__ << "(" << dirp << ")" << dendl; |
7c673cae | 7796 | if (dirp->inode) { |
11fdf7f2 | 7797 | ldout(cct, 10) << __func__ << " detaching inode " << dirp->inode << dendl; |
7c673cae FG |
7798 | dirp->inode.reset(); |
7799 | } | |
7800 | _readdir_drop_dirp_buffer(dirp); | |
7801 | opened_dirs.erase(dirp); | |
7802 | delete dirp; | |
7803 | } | |
7804 | ||
7805 | void Client::rewinddir(dir_result_t *dirp) | |
7806 | { | |
11fdf7f2 TL |
7807 | std::lock_guard lock(client_lock); |
7808 | ldout(cct, 3) << __func__ << "(" << dirp << ")" << dendl; | |
181888fb FG |
7809 | |
7810 | if (unmounting) | |
7811 | return; | |
7812 | ||
7c673cae FG |
7813 | dir_result_t *d = static_cast<dir_result_t*>(dirp); |
7814 | _readdir_drop_dirp_buffer(d); | |
7815 | d->reset(); | |
7816 | } | |
7817 | ||
7818 | loff_t Client::telldir(dir_result_t *dirp) | |
7819 | { | |
7820 | dir_result_t *d = static_cast<dir_result_t*>(dirp); | |
11fdf7f2 | 7821 | ldout(cct, 3) << __func__ << "(" << dirp << ") = " << d->offset << dendl; |
7c673cae FG |
7822 | return d->offset; |
7823 | } | |
7824 | ||
7825 | void Client::seekdir(dir_result_t *dirp, loff_t offset) | |
7826 | { | |
11fdf7f2 | 7827 | std::lock_guard lock(client_lock); |
7c673cae | 7828 | |
11fdf7f2 | 7829 | ldout(cct, 3) << __func__ << "(" << dirp << ", " << offset << ")" << dendl; |
7c673cae | 7830 | |
181888fb FG |
7831 | if (unmounting) |
7832 | return; | |
7833 | ||
7c673cae FG |
7834 | if (offset == dirp->offset) |
7835 | return; | |
7836 | ||
7837 | if (offset > dirp->offset) | |
7838 | dirp->release_count = 0; // bump if we do a forward seek | |
7839 | else | |
7840 | dirp->ordered_count = 0; // disable filling readdir cache | |
7841 | ||
7842 | if (dirp->hash_order()) { | |
7843 | if (dirp->offset > offset) { | |
7844 | _readdir_drop_dirp_buffer(dirp); | |
7845 | dirp->reset(); | |
7846 | } | |
7847 | } else { | |
7848 | if (offset == 0 || | |
7849 | dirp->buffer_frag != frag_t(dir_result_t::fpos_high(offset)) || | |
7850 | dirp->offset_low() > dir_result_t::fpos_low(offset)) { | |
7851 | _readdir_drop_dirp_buffer(dirp); | |
7852 | dirp->reset(); | |
7853 | } | |
7854 | } | |
7855 | ||
7856 | dirp->offset = offset; | |
7857 | } | |
7858 | ||
7859 | ||
7860 | //struct dirent { | |
7861 | // ino_t d_ino; /* inode number */ | |
7862 | // off_t d_off; /* offset to the next dirent */ | |
7863 | // unsigned short d_reclen; /* length of this record */ | |
7864 | // unsigned char d_type; /* type of file */ | |
7865 | // char d_name[256]; /* filename */ | |
7866 | //}; | |
7867 | void Client::fill_dirent(struct dirent *de, const char *name, int type, uint64_t ino, loff_t next_off) | |
7868 | { | |
7869 | strncpy(de->d_name, name, 255); | |
7870 | de->d_name[255] = '\0'; | |
7871 | #ifndef __CYGWIN__ | |
7872 | de->d_ino = ino; | |
11fdf7f2 | 7873 | #if !defined(__APPLE__) && !defined(__FreeBSD__) |
7c673cae FG |
7874 | de->d_off = next_off; |
7875 | #endif | |
7876 | de->d_reclen = 1; | |
7877 | de->d_type = IFTODT(type); | |
11fdf7f2 | 7878 | ldout(cct, 10) << __func__ << " '" << de->d_name << "' -> " << inodeno_t(de->d_ino) |
7c673cae FG |
7879 | << " type " << (int)de->d_type << " w/ next_off " << hex << next_off << dec << dendl; |
7880 | #endif | |
7881 | } | |
7882 | ||
7883 | void Client::_readdir_next_frag(dir_result_t *dirp) | |
7884 | { | |
7885 | frag_t fg = dirp->buffer_frag; | |
7886 | ||
7887 | if (fg.is_rightmost()) { | |
11fdf7f2 | 7888 | ldout(cct, 10) << __func__ << " advance from " << fg << " to END" << dendl; |
7c673cae FG |
7889 | dirp->set_end(); |
7890 | return; | |
7891 | } | |
7892 | ||
7893 | // advance | |
7894 | fg = fg.next(); | |
11fdf7f2 | 7895 | ldout(cct, 10) << __func__ << " advance from " << dirp->buffer_frag << " to " << fg << dendl; |
7c673cae FG |
7896 | |
7897 | if (dirp->hash_order()) { | |
7898 | // keep last_name | |
7899 | int64_t new_offset = dir_result_t::make_fpos(fg.value(), 2, true); | |
7900 | if (dirp->offset < new_offset) // don't decrease offset | |
7901 | dirp->offset = new_offset; | |
7902 | } else { | |
7903 | dirp->last_name.clear(); | |
7904 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); | |
7905 | _readdir_rechoose_frag(dirp); | |
7906 | } | |
7907 | } | |
7908 | ||
7909 | void Client::_readdir_rechoose_frag(dir_result_t *dirp) | |
7910 | { | |
11fdf7f2 | 7911 | ceph_assert(dirp->inode); |
7c673cae FG |
7912 | |
7913 | if (dirp->hash_order()) | |
7914 | return; | |
7915 | ||
7916 | frag_t cur = frag_t(dirp->offset_high()); | |
7917 | frag_t fg = dirp->inode->dirfragtree[cur.value()]; | |
7918 | if (fg != cur) { | |
11fdf7f2 | 7919 | ldout(cct, 10) << __func__ << " frag " << cur << " maps to " << fg << dendl; |
7c673cae FG |
7920 | dirp->offset = dir_result_t::make_fpos(fg, 2, false); |
7921 | dirp->last_name.clear(); | |
7922 | dirp->next_offset = 2; | |
7923 | } | |
7924 | } | |
7925 | ||
7926 | void Client::_readdir_drop_dirp_buffer(dir_result_t *dirp) | |
7927 | { | |
11fdf7f2 | 7928 | ldout(cct, 10) << __func__ << " " << dirp << dendl; |
7c673cae FG |
7929 | dirp->buffer.clear(); |
7930 | } | |
7931 | ||
7932 | int Client::_readdir_get_frag(dir_result_t *dirp) | |
7933 | { | |
11fdf7f2 TL |
7934 | ceph_assert(dirp); |
7935 | ceph_assert(dirp->inode); | |
7c673cae FG |
7936 | |
7937 | // get the current frag. | |
7938 | frag_t fg; | |
7939 | if (dirp->hash_order()) | |
7940 | fg = dirp->inode->dirfragtree[dirp->offset_high()]; | |
7941 | else | |
7942 | fg = frag_t(dirp->offset_high()); | |
7943 | ||
11fdf7f2 | 7944 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " fg " << fg |
7c673cae FG |
7945 | << " offset " << hex << dirp->offset << dec << dendl; |
7946 | ||
7947 | int op = CEPH_MDS_OP_READDIR; | |
7948 | if (dirp->inode && dirp->inode->snapid == CEPH_SNAPDIR) | |
7949 | op = CEPH_MDS_OP_LSSNAP; | |
7950 | ||
7951 | InodeRef& diri = dirp->inode; | |
7952 | ||
7953 | MetaRequest *req = new MetaRequest(op); | |
7954 | filepath path; | |
7955 | diri->make_nosnap_relative_path(path); | |
7956 | req->set_filepath(path); | |
7957 | req->set_inode(diri.get()); | |
7958 | req->head.args.readdir.frag = fg; | |
7959 | req->head.args.readdir.flags = CEPH_READDIR_REPLY_BITFLAGS; | |
7960 | if (dirp->last_name.length()) { | |
94b18763 | 7961 | req->path2.set_path(dirp->last_name); |
7c673cae FG |
7962 | } else if (dirp->hash_order()) { |
7963 | req->head.args.readdir.offset_hash = dirp->offset_high(); | |
7964 | } | |
7965 | req->dirp = dirp; | |
7966 | ||
7967 | bufferlist dirbl; | |
7968 | int res = make_request(req, dirp->perms, NULL, NULL, -1, &dirbl); | |
7969 | ||
7970 | if (res == -EAGAIN) { | |
11fdf7f2 | 7971 | ldout(cct, 10) << __func__ << " got EAGAIN, retrying" << dendl; |
7c673cae FG |
7972 | _readdir_rechoose_frag(dirp); |
7973 | return _readdir_get_frag(dirp); | |
7974 | } | |
7975 | ||
7976 | if (res == 0) { | |
11fdf7f2 | 7977 | ldout(cct, 10) << __func__ << " " << dirp << " got frag " << dirp->buffer_frag |
7c673cae FG |
7978 | << " size " << dirp->buffer.size() << dendl; |
7979 | } else { | |
11fdf7f2 | 7980 | ldout(cct, 10) << __func__ << " got error " << res << ", setting end flag" << dendl; |
7c673cae FG |
7981 | dirp->set_end(); |
7982 | } | |
7983 | ||
7984 | return res; | |
7985 | } | |
7986 | ||
7987 | struct dentry_off_lt { | |
7988 | bool operator()(const Dentry* dn, int64_t off) const { | |
7989 | return dir_result_t::fpos_cmp(dn->offset, off) < 0; | |
7990 | } | |
7991 | }; | |
7992 | ||
7993 | int Client::_readdir_cache_cb(dir_result_t *dirp, add_dirent_cb_t cb, void *p, | |
7994 | int caps, bool getref) | |
7995 | { | |
9f95a23c | 7996 | ceph_assert(ceph_mutex_is_locked(client_lock)); |
11fdf7f2 | 7997 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino |
7c673cae FG |
7998 | << " last_name " << dirp->last_name << " offset " << hex << dirp->offset << dec |
7999 | << dendl; | |
8000 | Dir *dir = dirp->inode->dir; | |
8001 | ||
8002 | if (!dir) { | |
8003 | ldout(cct, 10) << " dir is empty" << dendl; | |
8004 | dirp->set_end(); | |
8005 | return 0; | |
8006 | } | |
8007 | ||
8008 | vector<Dentry*>::iterator pd = std::lower_bound(dir->readdir_cache.begin(), | |
8009 | dir->readdir_cache.end(), | |
8010 | dirp->offset, dentry_off_lt()); | |
8011 | ||
8012 | string dn_name; | |
8013 | while (true) { | |
8014 | if (!dirp->inode->is_complete_and_ordered()) | |
8015 | return -EAGAIN; | |
8016 | if (pd == dir->readdir_cache.end()) | |
8017 | break; | |
8018 | Dentry *dn = *pd; | |
8019 | if (dn->inode == NULL) { | |
8020 | ldout(cct, 15) << " skipping null '" << dn->name << "'" << dendl; | |
8021 | ++pd; | |
8022 | continue; | |
8023 | } | |
8024 | if (dn->cap_shared_gen != dir->parent_inode->shared_gen) { | |
8025 | ldout(cct, 15) << " skipping mismatch shared gen '" << dn->name << "'" << dendl; | |
8026 | ++pd; | |
8027 | continue; | |
8028 | } | |
8029 | ||
92f5a8d4 | 8030 | int idx = pd - dir->readdir_cache.begin(); |
7c673cae FG |
8031 | int r = _getattr(dn->inode, caps, dirp->perms); |
8032 | if (r < 0) | |
8033 | return r; | |
92f5a8d4 TL |
8034 | |
8035 | // the content of readdir_cache may change after _getattr(), so pd may be invalid iterator | |
8036 | pd = dir->readdir_cache.begin() + idx; | |
8037 | if (pd >= dir->readdir_cache.end() || *pd != dn) | |
8038 | return -EAGAIN; | |
7c673cae FG |
8039 | |
8040 | struct ceph_statx stx; | |
8041 | struct dirent de; | |
8042 | fill_statx(dn->inode, caps, &stx); | |
8043 | ||
8044 | uint64_t next_off = dn->offset + 1; | |
eafe8130 | 8045 | fill_dirent(&de, dn->name.c_str(), stx.stx_mode, stx.stx_ino, next_off); |
7c673cae FG |
8046 | ++pd; |
8047 | if (pd == dir->readdir_cache.end()) | |
8048 | next_off = dir_result_t::END; | |
8049 | ||
8050 | Inode *in = NULL; | |
7c673cae FG |
8051 | if (getref) { |
8052 | in = dn->inode.get(); | |
8053 | _ll_get(in); | |
8054 | } | |
8055 | ||
8056 | dn_name = dn->name; // fill in name while we have lock | |
8057 | ||
9f95a23c | 8058 | client_lock.unlock(); |
7c673cae | 8059 | r = cb(p, &de, &stx, next_off, in); // _next_ offset |
9f95a23c | 8060 | client_lock.lock(); |
7c673cae FG |
8061 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << dn->offset << dec |
8062 | << " = " << r << dendl; | |
8063 | if (r < 0) { | |
8064 | return r; | |
8065 | } | |
8066 | ||
8067 | dirp->offset = next_off; | |
8068 | if (dirp->at_end()) | |
8069 | dirp->next_offset = 2; | |
8070 | else | |
8071 | dirp->next_offset = dirp->offset_low(); | |
8072 | dirp->last_name = dn_name; // we successfully returned this one; update! | |
28e407b8 | 8073 | dirp->release_count = 0; // last_name no longer match cache index |
7c673cae FG |
8074 | if (r > 0) |
8075 | return r; | |
8076 | } | |
8077 | ||
11fdf7f2 | 8078 | ldout(cct, 10) << __func__ << " " << dirp << " on " << dirp->inode->ino << " at end" << dendl; |
7c673cae FG |
8079 | dirp->set_end(); |
8080 | return 0; | |
8081 | } | |
8082 | ||
8083 | int Client::readdir_r_cb(dir_result_t *d, add_dirent_cb_t cb, void *p, | |
8084 | unsigned want, unsigned flags, bool getref) | |
8085 | { | |
8086 | int caps = statx_to_mask(flags, want); | |
8087 | ||
11fdf7f2 | 8088 | std::lock_guard lock(client_lock); |
7c673cae | 8089 | |
181888fb FG |
8090 | if (unmounting) |
8091 | return -ENOTCONN; | |
8092 | ||
7c673cae FG |
8093 | dir_result_t *dirp = static_cast<dir_result_t*>(d); |
8094 | ||
11fdf7f2 | 8095 | ldout(cct, 10) << __func__ << " " << *dirp->inode << " offset " << hex << dirp->offset |
7c673cae FG |
8096 | << dec << " at_end=" << dirp->at_end() |
8097 | << " hash_order=" << dirp->hash_order() << dendl; | |
8098 | ||
8099 | struct dirent de; | |
8100 | struct ceph_statx stx; | |
8101 | memset(&de, 0, sizeof(de)); | |
8102 | memset(&stx, 0, sizeof(stx)); | |
8103 | ||
8104 | InodeRef& diri = dirp->inode; | |
8105 | ||
8106 | if (dirp->at_end()) | |
8107 | return 0; | |
8108 | ||
8109 | if (dirp->offset == 0) { | |
8110 | ldout(cct, 15) << " including ." << dendl; | |
11fdf7f2 | 8111 | ceph_assert(diri->dentries.size() < 2); // can't have multiple hard-links to a dir |
7c673cae FG |
8112 | uint64_t next_off = 1; |
8113 | ||
8114 | int r; | |
8115 | r = _getattr(diri, caps, dirp->perms); | |
8116 | if (r < 0) | |
8117 | return r; | |
8118 | ||
8119 | fill_statx(diri, caps, &stx); | |
8120 | fill_dirent(&de, ".", S_IFDIR, stx.stx_ino, next_off); | |
8121 | ||
8122 | Inode *inode = NULL; | |
8123 | if (getref) { | |
8124 | inode = diri.get(); | |
8125 | _ll_get(inode); | |
8126 | } | |
8127 | ||
9f95a23c | 8128 | client_lock.unlock(); |
7c673cae | 8129 | r = cb(p, &de, &stx, next_off, inode); |
9f95a23c | 8130 | client_lock.lock(); |
7c673cae FG |
8131 | if (r < 0) |
8132 | return r; | |
8133 | ||
8134 | dirp->offset = next_off; | |
8135 | if (r > 0) | |
8136 | return r; | |
8137 | } | |
8138 | if (dirp->offset == 1) { | |
8139 | ldout(cct, 15) << " including .." << dendl; | |
8140 | uint64_t next_off = 2; | |
8141 | InodeRef in; | |
11fdf7f2 | 8142 | if (diri->dentries.empty()) |
7c673cae FG |
8143 | in = diri; |
8144 | else | |
94b18763 | 8145 | in = diri->get_first_parent()->dir->parent_inode; |
7c673cae FG |
8146 | |
8147 | int r; | |
94b18763 | 8148 | r = _getattr(in, caps, dirp->perms); |
7c673cae FG |
8149 | if (r < 0) |
8150 | return r; | |
8151 | ||
8152 | fill_statx(in, caps, &stx); | |
8153 | fill_dirent(&de, "..", S_IFDIR, stx.stx_ino, next_off); | |
8154 | ||
8155 | Inode *inode = NULL; | |
8156 | if (getref) { | |
8157 | inode = in.get(); | |
8158 | _ll_get(inode); | |
8159 | } | |
8160 | ||
9f95a23c | 8161 | client_lock.unlock(); |
7c673cae | 8162 | r = cb(p, &de, &stx, next_off, inode); |
9f95a23c | 8163 | client_lock.lock(); |
7c673cae FG |
8164 | if (r < 0) |
8165 | return r; | |
8166 | ||
8167 | dirp->offset = next_off; | |
8168 | if (r > 0) | |
8169 | return r; | |
8170 | } | |
8171 | ||
8172 | // can we read from our cache? | |
8173 | ldout(cct, 10) << "offset " << hex << dirp->offset << dec | |
8174 | << " snapid " << dirp->inode->snapid << " (complete && ordered) " | |
8175 | << dirp->inode->is_complete_and_ordered() | |
8176 | << " issued " << ccap_string(dirp->inode->caps_issued()) | |
8177 | << dendl; | |
8178 | if (dirp->inode->snapid != CEPH_SNAPDIR && | |
8179 | dirp->inode->is_complete_and_ordered() && | |
94b18763 | 8180 | dirp->inode->caps_issued_mask(CEPH_CAP_FILE_SHARED, true)) { |
7c673cae FG |
8181 | int err = _readdir_cache_cb(dirp, cb, p, caps, getref); |
8182 | if (err != -EAGAIN) | |
8183 | return err; | |
8184 | } | |
8185 | ||
8186 | while (1) { | |
8187 | if (dirp->at_end()) | |
8188 | return 0; | |
8189 | ||
8190 | bool check_caps = true; | |
8191 | if (!dirp->is_cached()) { | |
8192 | int r = _readdir_get_frag(dirp); | |
8193 | if (r) | |
8194 | return r; | |
8195 | // _readdir_get_frag () may updates dirp->offset if the replied dirfrag is | |
8196 | // different than the requested one. (our dirfragtree was outdated) | |
8197 | check_caps = false; | |
8198 | } | |
8199 | frag_t fg = dirp->buffer_frag; | |
8200 | ||
8201 | ldout(cct, 10) << "frag " << fg << " buffer size " << dirp->buffer.size() | |
8202 | << " offset " << hex << dirp->offset << dendl; | |
8203 | ||
8204 | for (auto it = std::lower_bound(dirp->buffer.begin(), dirp->buffer.end(), | |
8205 | dirp->offset, dir_result_t::dentry_off_lt()); | |
8206 | it != dirp->buffer.end(); | |
8207 | ++it) { | |
8208 | dir_result_t::dentry &entry = *it; | |
8209 | ||
8210 | uint64_t next_off = entry.offset + 1; | |
8211 | ||
8212 | int r; | |
8213 | if (check_caps) { | |
8214 | r = _getattr(entry.inode, caps, dirp->perms); | |
8215 | if (r < 0) | |
8216 | return r; | |
8217 | } | |
8218 | ||
8219 | fill_statx(entry.inode, caps, &stx); | |
8220 | fill_dirent(&de, entry.name.c_str(), stx.stx_mode, stx.stx_ino, next_off); | |
8221 | ||
8222 | Inode *inode = NULL; | |
8223 | if (getref) { | |
8224 | inode = entry.inode.get(); | |
8225 | _ll_get(inode); | |
8226 | } | |
8227 | ||
9f95a23c | 8228 | client_lock.unlock(); |
7c673cae | 8229 | r = cb(p, &de, &stx, next_off, inode); // _next_ offset |
9f95a23c | 8230 | client_lock.lock(); |
7c673cae FG |
8231 | |
8232 | ldout(cct, 15) << " de " << de.d_name << " off " << hex << next_off - 1 << dec | |
8233 | << " = " << r << dendl; | |
8234 | if (r < 0) | |
8235 | return r; | |
8236 | ||
8237 | dirp->offset = next_off; | |
8238 | if (r > 0) | |
8239 | return r; | |
8240 | } | |
8241 | ||
8242 | if (dirp->next_offset > 2) { | |
8243 | ldout(cct, 10) << " fetching next chunk of this frag" << dendl; | |
8244 | _readdir_drop_dirp_buffer(dirp); | |
8245 | continue; // more! | |
8246 | } | |
8247 | ||
8248 | if (!fg.is_rightmost()) { | |
8249 | // next frag! | |
8250 | _readdir_next_frag(dirp); | |
8251 | continue; | |
8252 | } | |
8253 | ||
8254 | if (diri->shared_gen == dirp->start_shared_gen && | |
8255 | diri->dir_release_count == dirp->release_count) { | |
8256 | if (diri->dir_ordered_count == dirp->ordered_count) { | |
8257 | ldout(cct, 10) << " marking (I_COMPLETE|I_DIR_ORDERED) on " << *diri << dendl; | |
8258 | if (diri->dir) { | |
11fdf7f2 | 8259 | ceph_assert(diri->dir->readdir_cache.size() >= dirp->cache_index); |
7c673cae FG |
8260 | diri->dir->readdir_cache.resize(dirp->cache_index); |
8261 | } | |
8262 | diri->flags |= I_COMPLETE | I_DIR_ORDERED; | |
8263 | } else { | |
8264 | ldout(cct, 10) << " marking I_COMPLETE on " << *diri << dendl; | |
8265 | diri->flags |= I_COMPLETE; | |
8266 | } | |
8267 | } | |
8268 | ||
8269 | dirp->set_end(); | |
8270 | return 0; | |
8271 | } | |
8272 | ceph_abort(); | |
8273 | return 0; | |
8274 | } | |
8275 | ||
8276 | ||
8277 | int Client::readdir_r(dir_result_t *d, struct dirent *de) | |
8278 | { | |
8279 | return readdirplus_r(d, de, 0, 0, 0, NULL); | |
8280 | } | |
8281 | ||
8282 | /* | |
8283 | * readdirplus_r | |
8284 | * | |
8285 | * returns | |
8286 | * 1 if we got a dirent | |
8287 | * 0 for end of directory | |
8288 | * <0 on error | |
8289 | */ | |
8290 | ||
8291 | struct single_readdir { | |
8292 | struct dirent *de; | |
8293 | struct ceph_statx *stx; | |
8294 | Inode *inode; | |
8295 | bool full; | |
8296 | }; | |
8297 | ||
8298 | static int _readdir_single_dirent_cb(void *p, struct dirent *de, | |
8299 | struct ceph_statx *stx, off_t off, | |
8300 | Inode *in) | |
8301 | { | |
8302 | single_readdir *c = static_cast<single_readdir *>(p); | |
8303 | ||
8304 | if (c->full) | |
8305 | return -1; // already filled this dirent | |
8306 | ||
8307 | *c->de = *de; | |
8308 | if (c->stx) | |
8309 | *c->stx = *stx; | |
8310 | c->inode = in; | |
8311 | c->full = true; | |
8312 | return 1; | |
8313 | } | |
8314 | ||
8315 | struct dirent *Client::readdir(dir_result_t *d) | |
8316 | { | |
8317 | int ret; | |
8318 | static struct dirent de; | |
8319 | single_readdir sr; | |
8320 | sr.de = &de; | |
8321 | sr.stx = NULL; | |
8322 | sr.inode = NULL; | |
8323 | sr.full = false; | |
8324 | ||
8325 | // our callback fills the dirent and sets sr.full=true on first | |
8326 | // call, and returns -1 the second time around. | |
8327 | ret = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr); | |
8328 | if (ret < -1) { | |
8329 | errno = -ret; // this sucks. | |
8330 | return (dirent *) NULL; | |
8331 | } | |
8332 | if (sr.full) { | |
8333 | return &de; | |
8334 | } | |
8335 | return (dirent *) NULL; | |
8336 | } | |
8337 | ||
8338 | int Client::readdirplus_r(dir_result_t *d, struct dirent *de, | |
8339 | struct ceph_statx *stx, unsigned want, | |
8340 | unsigned flags, Inode **out) | |
8341 | { | |
8342 | single_readdir sr; | |
8343 | sr.de = de; | |
8344 | sr.stx = stx; | |
8345 | sr.inode = NULL; | |
8346 | sr.full = false; | |
8347 | ||
8348 | // our callback fills the dirent and sets sr.full=true on first | |
8349 | // call, and returns -1 the second time around. | |
8350 | int r = readdir_r_cb(d, _readdir_single_dirent_cb, (void *)&sr, want, flags, out); | |
8351 | if (r < -1) | |
8352 | return r; | |
8353 | if (out) | |
8354 | *out = sr.inode; | |
8355 | if (sr.full) | |
8356 | return 1; | |
8357 | return 0; | |
8358 | } | |
8359 | ||
8360 | ||
8361 | /* getdents */ | |
8362 | struct getdents_result { | |
8363 | char *buf; | |
8364 | int buflen; | |
8365 | int pos; | |
8366 | bool fullent; | |
8367 | }; | |
8368 | ||
8369 | static int _readdir_getdent_cb(void *p, struct dirent *de, | |
8370 | struct ceph_statx *stx, off_t off, Inode *in) | |
8371 | { | |
8372 | struct getdents_result *c = static_cast<getdents_result *>(p); | |
8373 | ||
8374 | int dlen; | |
8375 | if (c->fullent) | |
8376 | dlen = sizeof(*de); | |
8377 | else | |
8378 | dlen = strlen(de->d_name) + 1; | |
8379 | ||
8380 | if (c->pos + dlen > c->buflen) | |
8381 | return -1; // doesn't fit | |
8382 | ||
8383 | if (c->fullent) { | |
8384 | memcpy(c->buf + c->pos, de, sizeof(*de)); | |
8385 | } else { | |
8386 | memcpy(c->buf + c->pos, de->d_name, dlen); | |
8387 | } | |
8388 | c->pos += dlen; | |
8389 | return 0; | |
8390 | } | |
8391 | ||
8392 | int Client::_getdents(dir_result_t *dir, char *buf, int buflen, bool fullent) | |
8393 | { | |
8394 | getdents_result gr; | |
8395 | gr.buf = buf; | |
8396 | gr.buflen = buflen; | |
8397 | gr.fullent = fullent; | |
8398 | gr.pos = 0; | |
8399 | ||
8400 | int r = readdir_r_cb(dir, _readdir_getdent_cb, (void *)&gr); | |
8401 | ||
8402 | if (r < 0) { // some error | |
8403 | if (r == -1) { // buffer ran out of space | |
8404 | if (gr.pos) { // but we got some entries already! | |
8405 | return gr.pos; | |
8406 | } // or we need a larger buffer | |
8407 | return -ERANGE; | |
8408 | } else { // actual error, return it | |
8409 | return r; | |
8410 | } | |
8411 | } | |
8412 | return gr.pos; | |
8413 | } | |
8414 | ||
8415 | ||
8416 | /* getdir */ | |
8417 | struct getdir_result { | |
8418 | list<string> *contents; | |
8419 | int num; | |
8420 | }; | |
8421 | ||
8422 | static int _getdir_cb(void *p, struct dirent *de, struct ceph_statx *stx, off_t off, Inode *in) | |
8423 | { | |
8424 | getdir_result *r = static_cast<getdir_result *>(p); | |
8425 | ||
8426 | r->contents->push_back(de->d_name); | |
8427 | r->num++; | |
8428 | return 0; | |
8429 | } | |
8430 | ||
8431 | int Client::getdir(const char *relpath, list<string>& contents, | |
8432 | const UserPerm& perms) | |
8433 | { | |
8434 | ldout(cct, 3) << "getdir(" << relpath << ")" << dendl; | |
8435 | { | |
11fdf7f2 | 8436 | std::lock_guard lock(client_lock); |
7c673cae FG |
8437 | tout(cct) << "getdir" << std::endl; |
8438 | tout(cct) << relpath << std::endl; | |
8439 | } | |
8440 | ||
8441 | dir_result_t *d; | |
8442 | int r = opendir(relpath, &d, perms); | |
8443 | if (r < 0) | |
8444 | return r; | |
8445 | ||
8446 | getdir_result gr; | |
8447 | gr.contents = &contents; | |
8448 | gr.num = 0; | |
8449 | r = readdir_r_cb(d, _getdir_cb, (void *)&gr); | |
8450 | ||
8451 | closedir(d); | |
8452 | ||
8453 | if (r < 0) | |
8454 | return r; | |
8455 | return gr.num; | |
8456 | } | |
8457 | ||
8458 | ||
8459 | /****** file i/o **********/ | |
8460 | int Client::open(const char *relpath, int flags, const UserPerm& perms, | |
8461 | mode_t mode, int stripe_unit, int stripe_count, | |
8462 | int object_size, const char *data_pool) | |
8463 | { | |
8464 | ldout(cct, 3) << "open enter(" << relpath << ", " << ceph_flags_sys2wire(flags) << "," << mode << ")" << dendl; | |
11fdf7f2 | 8465 | std::lock_guard lock(client_lock); |
7c673cae FG |
8466 | tout(cct) << "open" << std::endl; |
8467 | tout(cct) << relpath << std::endl; | |
8468 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
8469 | ||
181888fb FG |
8470 | if (unmounting) |
8471 | return -ENOTCONN; | |
8472 | ||
7c673cae FG |
8473 | Fh *fh = NULL; |
8474 | ||
8475 | #if defined(__linux__) && defined(O_PATH) | |
8476 | /* When the O_PATH is being specified, others flags than O_DIRECTORY | |
8477 | * and O_NOFOLLOW are ignored. Please refer do_entry_open() function | |
8478 | * in kernel (fs/open.c). */ | |
8479 | if (flags & O_PATH) | |
8480 | flags &= O_DIRECTORY | O_NOFOLLOW | O_PATH; | |
8481 | #endif | |
8482 | ||
8483 | filepath path(relpath); | |
8484 | InodeRef in; | |
8485 | bool created = false; | |
8486 | /* O_CREATE with O_EXCL enforces O_NOFOLLOW. */ | |
8487 | bool followsym = !((flags & O_NOFOLLOW) || ((flags & O_CREAT) && (flags & O_EXCL))); | |
8488 | int r = path_walk(path, &in, perms, followsym, ceph_caps_for_mode(mode)); | |
8489 | ||
8490 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
8491 | return -EEXIST; | |
8492 | ||
8493 | #if defined(__linux__) && defined(O_PATH) | |
8494 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW) && !(flags & O_PATH)) | |
8495 | #else | |
8496 | if (r == 0 && in->is_symlink() && (flags & O_NOFOLLOW)) | |
8497 | #endif | |
8498 | return -ELOOP; | |
8499 | ||
8500 | if (r == -ENOENT && (flags & O_CREAT)) { | |
8501 | filepath dirpath = path; | |
8502 | string dname = dirpath.last_dentry(); | |
8503 | dirpath.pop_dentry(); | |
8504 | InodeRef dir; | |
8505 | r = path_walk(dirpath, &dir, perms, true, | |
8506 | cct->_conf->client_permissions ? CEPH_CAP_AUTH_SHARED : 0); | |
8507 | if (r < 0) | |
8508 | goto out; | |
8509 | if (cct->_conf->client_permissions) { | |
8510 | r = may_create(dir.get(), perms); | |
8511 | if (r < 0) | |
8512 | goto out; | |
8513 | } | |
8514 | r = _create(dir.get(), dname.c_str(), flags, mode, &in, &fh, stripe_unit, | |
8515 | stripe_count, object_size, data_pool, &created, perms); | |
8516 | } | |
8517 | if (r < 0) | |
8518 | goto out; | |
8519 | ||
8520 | if (!created) { | |
8521 | // posix says we can only check permissions of existing files | |
8522 | if (cct->_conf->client_permissions) { | |
8523 | r = may_open(in.get(), flags, perms); | |
8524 | if (r < 0) | |
8525 | goto out; | |
8526 | } | |
8527 | } | |
8528 | ||
8529 | if (!fh) | |
8530 | r = _open(in.get(), flags, mode, &fh, perms); | |
8531 | if (r >= 0) { | |
8532 | // allocate a integer file descriptor | |
11fdf7f2 | 8533 | ceph_assert(fh); |
7c673cae | 8534 | r = get_fd(); |
11fdf7f2 | 8535 | ceph_assert(fd_map.count(r) == 0); |
7c673cae FG |
8536 | fd_map[r] = fh; |
8537 | } | |
8538 | ||
8539 | out: | |
8540 | tout(cct) << r << std::endl; | |
8541 | ldout(cct, 3) << "open exit(" << path << ", " << ceph_flags_sys2wire(flags) << ") = " << r << dendl; | |
8542 | return r; | |
8543 | } | |
8544 | ||
8545 | int Client::open(const char *relpath, int flags, const UserPerm& perms, mode_t mode) | |
8546 | { | |
8547 | /* Use default file striping parameters */ | |
8548 | return open(relpath, flags, perms, mode, 0, 0, 0, NULL); | |
8549 | } | |
8550 | ||
8551 | int Client::lookup_hash(inodeno_t ino, inodeno_t dirino, const char *name, | |
8552 | const UserPerm& perms) | |
8553 | { | |
11fdf7f2 TL |
8554 | std::lock_guard lock(client_lock); |
8555 | ldout(cct, 3) << __func__ << " enter(" << ino << ", #" << dirino << "/" << name << ")" << dendl; | |
7c673cae | 8556 | |
181888fb FG |
8557 | if (unmounting) |
8558 | return -ENOTCONN; | |
8559 | ||
7c673cae FG |
8560 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPHASH); |
8561 | filepath path(ino); | |
8562 | req->set_filepath(path); | |
8563 | ||
8564 | uint32_t h = ceph_str_hash(CEPH_STR_HASH_RJENKINS, name, strlen(name)); | |
8565 | char f[30]; | |
8566 | sprintf(f, "%u", h); | |
8567 | filepath path2(dirino); | |
8568 | path2.push_dentry(string(f)); | |
8569 | req->set_filepath2(path2); | |
8570 | ||
8571 | int r = make_request(req, perms, NULL, NULL, | |
8572 | rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 8573 | ldout(cct, 3) << __func__ << " exit(" << ino << ", #" << dirino << "/" << name << ") = " << r << dendl; |
7c673cae FG |
8574 | return r; |
8575 | } | |
8576 | ||
8577 | ||
8578 | /** | |
8579 | * Load inode into local cache. | |
8580 | * | |
8581 | * If inode pointer is non-NULL, and take a reference on | |
8582 | * the resulting Inode object in one operation, so that caller | |
8583 | * can safely assume inode will still be there after return. | |
8584 | */ | |
1adf2230 | 8585 | int Client::_lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) |
7c673cae | 8586 | { |
11fdf7f2 | 8587 | ldout(cct, 8) << __func__ << " enter(" << ino << ")" << dendl; |
7c673cae | 8588 | |
181888fb FG |
8589 | if (unmounting) |
8590 | return -ENOTCONN; | |
8591 | ||
7c673cae FG |
8592 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPINO); |
8593 | filepath path(ino); | |
8594 | req->set_filepath(path); | |
8595 | ||
8596 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
8597 | if (r == 0 && inode != NULL) { | |
8598 | vinodeno_t vino(ino, CEPH_NOSNAP); | |
8599 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
11fdf7f2 | 8600 | ceph_assert(p != inode_map.end()); |
7c673cae FG |
8601 | *inode = p->second; |
8602 | _ll_get(*inode); | |
8603 | } | |
11fdf7f2 | 8604 | ldout(cct, 8) << __func__ << " exit(" << ino << ") = " << r << dendl; |
7c673cae FG |
8605 | return r; |
8606 | } | |
8607 | ||
1adf2230 AA |
8608 | int Client::lookup_ino(inodeno_t ino, const UserPerm& perms, Inode **inode) |
8609 | { | |
11fdf7f2 | 8610 | std::lock_guard lock(client_lock); |
1adf2230 AA |
8611 | return _lookup_ino(ino, perms, inode); |
8612 | } | |
7c673cae FG |
8613 | |
8614 | /** | |
8615 | * Find the parent inode of `ino` and insert it into | |
8616 | * our cache. Conditionally also set `parent` to a referenced | |
8617 | * Inode* if caller provides non-NULL value. | |
8618 | */ | |
1adf2230 | 8619 | int Client::_lookup_parent(Inode *ino, const UserPerm& perms, Inode **parent) |
7c673cae | 8620 | { |
11fdf7f2 | 8621 | ldout(cct, 8) << __func__ << " enter(" << ino->ino << ")" << dendl; |
7c673cae | 8622 | |
7c673cae FG |
8623 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPPARENT); |
8624 | filepath path(ino->ino); | |
8625 | req->set_filepath(path); | |
8626 | ||
8627 | InodeRef target; | |
8628 | int r = make_request(req, perms, &target, NULL, rand() % mdsmap->get_num_in_mds()); | |
8629 | // Give caller a reference to the parent ino if they provided a pointer. | |
8630 | if (parent != NULL) { | |
8631 | if (r == 0) { | |
8632 | *parent = target.get(); | |
8633 | _ll_get(*parent); | |
11fdf7f2 | 8634 | ldout(cct, 8) << __func__ << " found parent " << (*parent)->ino << dendl; |
7c673cae FG |
8635 | } else { |
8636 | *parent = NULL; | |
8637 | } | |
8638 | } | |
11fdf7f2 | 8639 | ldout(cct, 8) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
8640 | return r; |
8641 | } | |
8642 | ||
7c673cae FG |
8643 | /** |
8644 | * Populate the parent dentry for `ino`, provided it is | |
8645 | * a child of `parent`. | |
8646 | */ | |
1adf2230 | 8647 | int Client::_lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
7c673cae | 8648 | { |
11fdf7f2 TL |
8649 | ceph_assert(parent->is_dir()); |
8650 | ldout(cct, 3) << __func__ << " enter(" << ino->ino << ")" << dendl; | |
7c673cae | 8651 | |
181888fb FG |
8652 | if (unmounting) |
8653 | return -ENOTCONN; | |
8654 | ||
7c673cae FG |
8655 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
8656 | req->set_filepath2(filepath(parent->ino)); | |
8657 | req->set_filepath(filepath(ino->ino)); | |
8658 | req->set_inode(ino); | |
8659 | ||
8660 | int r = make_request(req, perms, NULL, NULL, rand() % mdsmap->get_num_in_mds()); | |
11fdf7f2 | 8661 | ldout(cct, 3) << __func__ << " exit(" << ino->ino << ") = " << r << dendl; |
7c673cae FG |
8662 | return r; |
8663 | } | |
8664 | ||
1adf2230 AA |
8665 | int Client::lookup_name(Inode *ino, Inode *parent, const UserPerm& perms) |
8666 | { | |
11fdf7f2 | 8667 | std::lock_guard lock(client_lock); |
1adf2230 AA |
8668 | return _lookup_name(ino, parent, perms); |
8669 | } | |
7c673cae | 8670 | |
11fdf7f2 | 8671 | Fh *Client::_create_fh(Inode *in, int flags, int cmode, const UserPerm& perms) |
7c673cae | 8672 | { |
11fdf7f2 TL |
8673 | ceph_assert(in); |
8674 | Fh *f = new Fh(in, flags, cmode, perms); | |
7c673cae | 8675 | |
11fdf7f2 | 8676 | ldout(cct, 10) << __func__ << " " << in->ino << " mode " << cmode << dendl; |
7c673cae FG |
8677 | |
8678 | if (in->snapid != CEPH_NOSNAP) { | |
8679 | in->snap_cap_refs++; | |
8680 | ldout(cct, 5) << "open success, fh is " << f << " combined IMMUTABLE SNAP caps " | |
8681 | << ccap_string(in->caps_issued()) << dendl; | |
8682 | } | |
8683 | ||
11fdf7f2 | 8684 | const auto& conf = cct->_conf; |
7c673cae FG |
8685 | f->readahead.set_trigger_requests(1); |
8686 | f->readahead.set_min_readahead_size(conf->client_readahead_min); | |
8687 | uint64_t max_readahead = Readahead::NO_LIMIT; | |
8688 | if (conf->client_readahead_max_bytes) { | |
11fdf7f2 | 8689 | max_readahead = std::min(max_readahead, (uint64_t)conf->client_readahead_max_bytes); |
7c673cae FG |
8690 | } |
8691 | if (conf->client_readahead_max_periods) { | |
11fdf7f2 | 8692 | max_readahead = std::min(max_readahead, in->layout.get_period()*(uint64_t)conf->client_readahead_max_periods); |
7c673cae FG |
8693 | } |
8694 | f->readahead.set_max_readahead_size(max_readahead); | |
8695 | vector<uint64_t> alignments; | |
8696 | alignments.push_back(in->layout.get_period()); | |
8697 | alignments.push_back(in->layout.stripe_unit); | |
8698 | f->readahead.set_alignments(alignments); | |
8699 | ||
8700 | return f; | |
8701 | } | |
8702 | ||
8703 | int Client::_release_fh(Fh *f) | |
8704 | { | |
8705 | //ldout(cct, 3) << "op: client->close(open_files[ " << fh << " ]);" << dendl; | |
8706 | //ldout(cct, 3) << "op: open_files.erase( " << fh << " );" << dendl; | |
8707 | Inode *in = f->inode.get(); | |
11fdf7f2 | 8708 | ldout(cct, 8) << __func__ << " " << f << " mode " << f->mode << " on " << *in << dendl; |
7c673cae | 8709 | |
b32b8144 FG |
8710 | in->unset_deleg(f); |
8711 | ||
7c673cae FG |
8712 | if (in->snapid == CEPH_NOSNAP) { |
8713 | if (in->put_open_ref(f->mode)) { | |
8714 | _flush(in, new C_Client_FlushComplete(this, in)); | |
8715 | check_caps(in, 0); | |
8716 | } | |
8717 | } else { | |
11fdf7f2 | 8718 | ceph_assert(in->snap_cap_refs > 0); |
7c673cae FG |
8719 | in->snap_cap_refs--; |
8720 | } | |
8721 | ||
8722 | _release_filelocks(f); | |
8723 | ||
8724 | // Finally, read any async err (i.e. from flushes) | |
8725 | int err = f->take_async_err(); | |
8726 | if (err != 0) { | |
11fdf7f2 | 8727 | ldout(cct, 1) << __func__ << " " << f << " on inode " << *in << " caught async_err = " |
7c673cae FG |
8728 | << cpp_strerror(err) << dendl; |
8729 | } else { | |
11fdf7f2 | 8730 | ldout(cct, 10) << __func__ << " " << f << " on inode " << *in << " no async_err state" << dendl; |
7c673cae FG |
8731 | } |
8732 | ||
8733 | _put_fh(f); | |
8734 | ||
8735 | return err; | |
8736 | } | |
8737 | ||
8738 | void Client::_put_fh(Fh *f) | |
8739 | { | |
8740 | int left = f->put(); | |
8741 | if (!left) { | |
8742 | delete f; | |
8743 | } | |
8744 | } | |
8745 | ||
8746 | int Client::_open(Inode *in, int flags, mode_t mode, Fh **fhp, | |
8747 | const UserPerm& perms) | |
8748 | { | |
8749 | if (in->snapid != CEPH_NOSNAP && | |
8750 | (flags & (O_WRONLY | O_RDWR | O_CREAT | O_TRUNC | O_APPEND))) { | |
8751 | return -EROFS; | |
8752 | } | |
8753 | ||
8754 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
8755 | int cflags = ceph_flags_sys2wire(flags); |
8756 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
8757 | cflags |= CEPH_O_LAZY; | |
8758 | ||
8759 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
8760 | int want = ceph_caps_for_mode(cmode); |
8761 | int result = 0; | |
8762 | ||
8763 | in->get_open_ref(cmode); // make note of pending open, since it effects _wanted_ caps. | |
8764 | ||
b32b8144 | 8765 | if ((flags & O_TRUNC) == 0 && in->caps_issued_mask(want)) { |
7c673cae FG |
8766 | // update wanted? |
8767 | check_caps(in, CHECK_CAPS_NODELAY); | |
8768 | } else { | |
b32b8144 | 8769 | |
7c673cae FG |
8770 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); |
8771 | filepath path; | |
8772 | in->make_nosnap_relative_path(path); | |
8773 | req->set_filepath(path); | |
11fdf7f2 | 8774 | req->head.args.open.flags = cflags & ~CEPH_O_CREAT; |
7c673cae FG |
8775 | req->head.args.open.mode = mode; |
8776 | req->head.args.open.pool = -1; | |
8777 | if (cct->_conf->client_debug_getattr_caps) | |
8778 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
8779 | else | |
8780 | req->head.args.open.mask = 0; | |
8781 | req->head.args.open.old_size = in->size; // for O_TRUNC | |
8782 | req->set_inode(in); | |
8783 | result = make_request(req, perms); | |
b32b8144 FG |
8784 | |
8785 | /* | |
8786 | * NFS expects that delegations will be broken on a conflicting open, | |
8787 | * not just when there is actual conflicting access to the file. SMB leases | |
8788 | * and oplocks also have similar semantics. | |
8789 | * | |
8790 | * Ensure that clients that have delegations enabled will wait on minimal | |
8791 | * caps during open, just to ensure that other clients holding delegations | |
8792 | * return theirs first. | |
8793 | */ | |
8794 | if (deleg_timeout && result == 0) { | |
8795 | int need = 0, have; | |
8796 | ||
8797 | if (cmode & CEPH_FILE_MODE_WR) | |
8798 | need |= CEPH_CAP_FILE_WR; | |
8799 | if (cmode & CEPH_FILE_MODE_RD) | |
8800 | need |= CEPH_CAP_FILE_RD; | |
8801 | ||
8802 | result = get_caps(in, need, want, &have, -1); | |
8803 | if (result < 0) { | |
1adf2230 | 8804 | ldout(cct, 8) << "Unable to get caps after open of inode " << *in << |
b32b8144 FG |
8805 | " . Denying open: " << |
8806 | cpp_strerror(result) << dendl; | |
8807 | in->put_open_ref(cmode); | |
8808 | } else { | |
8809 | put_cap_ref(in, need); | |
8810 | } | |
8811 | } | |
7c673cae FG |
8812 | } |
8813 | ||
8814 | // success? | |
8815 | if (result >= 0) { | |
8816 | if (fhp) | |
8817 | *fhp = _create_fh(in, flags, cmode, perms); | |
8818 | } else { | |
8819 | in->put_open_ref(cmode); | |
8820 | } | |
8821 | ||
8822 | trim_cache(); | |
8823 | ||
8824 | return result; | |
8825 | } | |
8826 | ||
8827 | int Client::_renew_caps(Inode *in) | |
8828 | { | |
8829 | int wanted = in->caps_file_wanted(); | |
8830 | if (in->is_any_caps() && | |
8831 | ((wanted & CEPH_CAP_ANY_WR) == 0 || in->auth_cap)) { | |
8832 | check_caps(in, CHECK_CAPS_NODELAY); | |
8833 | return 0; | |
8834 | } | |
8835 | ||
8836 | int flags = 0; | |
8837 | if ((wanted & CEPH_CAP_FILE_RD) && (wanted & CEPH_CAP_FILE_WR)) | |
8838 | flags = O_RDWR; | |
8839 | else if (wanted & CEPH_CAP_FILE_RD) | |
8840 | flags = O_RDONLY; | |
8841 | else if (wanted & CEPH_CAP_FILE_WR) | |
8842 | flags = O_WRONLY; | |
8843 | ||
8844 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_OPEN); | |
8845 | filepath path; | |
8846 | in->make_nosnap_relative_path(path); | |
8847 | req->set_filepath(path); | |
8848 | req->head.args.open.flags = flags; | |
8849 | req->head.args.open.pool = -1; | |
8850 | if (cct->_conf->client_debug_getattr_caps) | |
8851 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
8852 | else | |
8853 | req->head.args.open.mask = 0; | |
8854 | req->set_inode(in); | |
8855 | ||
8856 | // duplicate in case Cap goes away; not sure if that race is a concern? | |
8857 | const UserPerm *pperm = in->get_best_perms(); | |
8858 | UserPerm perms; | |
8859 | if (pperm != NULL) | |
8860 | perms = *pperm; | |
8861 | int ret = make_request(req, perms); | |
8862 | return ret; | |
8863 | } | |
8864 | ||
8865 | int Client::close(int fd) | |
8866 | { | |
8867 | ldout(cct, 3) << "close enter(" << fd << ")" << dendl; | |
11fdf7f2 | 8868 | std::lock_guard lock(client_lock); |
7c673cae FG |
8869 | tout(cct) << "close" << std::endl; |
8870 | tout(cct) << fd << std::endl; | |
8871 | ||
181888fb FG |
8872 | if (unmounting) |
8873 | return -ENOTCONN; | |
8874 | ||
7c673cae FG |
8875 | Fh *fh = get_filehandle(fd); |
8876 | if (!fh) | |
8877 | return -EBADF; | |
8878 | int err = _release_fh(fh); | |
8879 | fd_map.erase(fd); | |
8880 | put_fd(fd); | |
8881 | ldout(cct, 3) << "close exit(" << fd << ")" << dendl; | |
8882 | return err; | |
8883 | } | |
8884 | ||
8885 | ||
8886 | // ------------ | |
8887 | // read, write | |
8888 | ||
8889 | loff_t Client::lseek(int fd, loff_t offset, int whence) | |
8890 | { | |
11fdf7f2 | 8891 | std::lock_guard lock(client_lock); |
7c673cae FG |
8892 | tout(cct) << "lseek" << std::endl; |
8893 | tout(cct) << fd << std::endl; | |
8894 | tout(cct) << offset << std::endl; | |
8895 | tout(cct) << whence << std::endl; | |
8896 | ||
181888fb FG |
8897 | if (unmounting) |
8898 | return -ENOTCONN; | |
8899 | ||
7c673cae FG |
8900 | Fh *f = get_filehandle(fd); |
8901 | if (!f) | |
8902 | return -EBADF; | |
8903 | #if defined(__linux__) && defined(O_PATH) | |
8904 | if (f->flags & O_PATH) | |
8905 | return -EBADF; | |
8906 | #endif | |
8907 | return _lseek(f, offset, whence); | |
8908 | } | |
8909 | ||
8910 | loff_t Client::_lseek(Fh *f, loff_t offset, int whence) | |
8911 | { | |
8912 | Inode *in = f->inode.get(); | |
9f95a23c | 8913 | bool whence_check = false; |
11fdf7f2 | 8914 | loff_t pos = -1; |
7c673cae | 8915 | |
9f95a23c TL |
8916 | switch (whence) { |
8917 | case SEEK_END: | |
8918 | whence_check = true; | |
8919 | break; | |
8920 | ||
8921 | #ifdef SEEK_DATA | |
8922 | case SEEK_DATA: | |
8923 | whence_check = true; | |
8924 | break; | |
8925 | #endif | |
8926 | ||
8927 | #ifdef SEEK_HOLE | |
8928 | case SEEK_HOLE: | |
8929 | whence_check = true; | |
8930 | break; | |
8931 | #endif | |
8932 | } | |
8933 | ||
8934 | if (whence_check) { | |
8935 | int r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
8936 | if (r < 0) | |
92f5a8d4 | 8937 | return r; |
92f5a8d4 TL |
8938 | } |
8939 | ||
7c673cae FG |
8940 | switch (whence) { |
8941 | case SEEK_SET: | |
11fdf7f2 | 8942 | pos = offset; |
7c673cae FG |
8943 | break; |
8944 | ||
8945 | case SEEK_CUR: | |
92f5a8d4 | 8946 | pos = f->pos + offset; |
7c673cae FG |
8947 | break; |
8948 | ||
8949 | case SEEK_END: | |
11fdf7f2 | 8950 | pos = in->size + offset; |
7c673cae FG |
8951 | break; |
8952 | ||
9f95a23c | 8953 | #ifdef SEEK_DATA |
92f5a8d4 | 8954 | case SEEK_DATA: |
9f95a23c TL |
8955 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
8956 | return -ENXIO; | |
92f5a8d4 TL |
8957 | pos = offset; |
8958 | break; | |
9f95a23c | 8959 | #endif |
92f5a8d4 | 8960 | |
9f95a23c | 8961 | #ifdef SEEK_HOLE |
92f5a8d4 | 8962 | case SEEK_HOLE: |
9f95a23c TL |
8963 | if (offset < 0 || static_cast<uint64_t>(offset) >= in->size) |
8964 | return -ENXIO; | |
8965 | pos = in->size; | |
92f5a8d4 | 8966 | break; |
9f95a23c | 8967 | #endif |
92f5a8d4 | 8968 | |
7c673cae | 8969 | default: |
92f5a8d4 TL |
8970 | ldout(cct, 1) << __func__ << ": invalid whence value " << whence << dendl; |
8971 | return -EINVAL; | |
7c673cae FG |
8972 | } |
8973 | ||
11fdf7f2 TL |
8974 | if (pos < 0) { |
8975 | return -EINVAL; | |
8976 | } else { | |
8977 | f->pos = pos; | |
8978 | } | |
8979 | ||
1adf2230 | 8980 | ldout(cct, 8) << "_lseek(" << f << ", " << offset << ", " << whence << ") = " << f->pos << dendl; |
7c673cae FG |
8981 | return f->pos; |
8982 | } | |
8983 | ||
8984 | ||
8985 | void Client::lock_fh_pos(Fh *f) | |
8986 | { | |
11fdf7f2 | 8987 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae FG |
8988 | |
8989 | if (f->pos_locked || !f->pos_waiters.empty()) { | |
9f95a23c | 8990 | ceph::condition_variable cond; |
7c673cae | 8991 | f->pos_waiters.push_back(&cond); |
11fdf7f2 | 8992 | ldout(cct, 10) << __func__ << " BLOCKING on " << f << dendl; |
9f95a23c TL |
8993 | std::unique_lock l{client_lock, std::adopt_lock}; |
8994 | cond.wait(l, [f, me=&cond] { | |
8995 | return !f->pos_locked && f->pos_waiters.front() == me; | |
8996 | }); | |
8997 | l.release(); | |
11fdf7f2 TL |
8998 | ldout(cct, 10) << __func__ << " UNBLOCKING on " << f << dendl; |
8999 | ceph_assert(f->pos_waiters.front() == &cond); | |
7c673cae FG |
9000 | f->pos_waiters.pop_front(); |
9001 | } | |
9002 | ||
9003 | f->pos_locked = true; | |
9004 | } | |
9005 | ||
9006 | void Client::unlock_fh_pos(Fh *f) | |
9007 | { | |
11fdf7f2 | 9008 | ldout(cct, 10) << __func__ << " " << f << dendl; |
7c673cae FG |
9009 | f->pos_locked = false; |
9010 | } | |
9011 | ||
9012 | int Client::uninline_data(Inode *in, Context *onfinish) | |
9013 | { | |
9014 | if (!in->inline_data.length()) { | |
9015 | onfinish->complete(0); | |
9016 | return 0; | |
9017 | } | |
9018 | ||
9019 | char oid_buf[32]; | |
9020 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (long long unsigned)in->ino); | |
9021 | object_t oid = oid_buf; | |
9022 | ||
9023 | ObjectOperation create_ops; | |
9024 | create_ops.create(false); | |
9025 | ||
9026 | objecter->mutate(oid, | |
9027 | OSDMap::file_to_object_locator(in->layout), | |
9028 | create_ops, | |
9029 | in->snaprealm->get_snap_context(), | |
9030 | ceph::real_clock::now(), | |
9031 | 0, | |
9032 | NULL); | |
9033 | ||
9034 | bufferlist inline_version_bl; | |
11fdf7f2 | 9035 | encode(in->inline_version, inline_version_bl); |
7c673cae FG |
9036 | |
9037 | ObjectOperation uninline_ops; | |
9038 | uninline_ops.cmpxattr("inline_version", | |
9039 | CEPH_OSD_CMPXATTR_OP_GT, | |
9040 | CEPH_OSD_CMPXATTR_MODE_U64, | |
9041 | inline_version_bl); | |
9042 | bufferlist inline_data = in->inline_data; | |
9043 | uninline_ops.write(0, inline_data, in->truncate_size, in->truncate_seq); | |
9044 | uninline_ops.setxattr("inline_version", stringify(in->inline_version)); | |
9045 | ||
9046 | objecter->mutate(oid, | |
9047 | OSDMap::file_to_object_locator(in->layout), | |
9048 | uninline_ops, | |
9049 | in->snaprealm->get_snap_context(), | |
9050 | ceph::real_clock::now(), | |
9051 | 0, | |
9052 | onfinish); | |
9053 | ||
9054 | return 0; | |
9055 | } | |
9056 | ||
9057 | // | |
9058 | ||
9059 | // blocking osd interface | |
9060 | ||
9061 | int Client::read(int fd, char *buf, loff_t size, loff_t offset) | |
9062 | { | |
11fdf7f2 | 9063 | std::lock_guard lock(client_lock); |
7c673cae FG |
9064 | tout(cct) << "read" << std::endl; |
9065 | tout(cct) << fd << std::endl; | |
9066 | tout(cct) << size << std::endl; | |
9067 | tout(cct) << offset << std::endl; | |
9068 | ||
181888fb FG |
9069 | if (unmounting) |
9070 | return -ENOTCONN; | |
9071 | ||
7c673cae FG |
9072 | Fh *f = get_filehandle(fd); |
9073 | if (!f) | |
9074 | return -EBADF; | |
9075 | #if defined(__linux__) && defined(O_PATH) | |
9076 | if (f->flags & O_PATH) | |
9077 | return -EBADF; | |
9078 | #endif | |
9079 | bufferlist bl; | |
11fdf7f2 TL |
9080 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
9081 | size = std::min(size, (loff_t)INT_MAX); | |
7c673cae FG |
9082 | int r = _read(f, offset, size, &bl); |
9083 | ldout(cct, 3) << "read(" << fd << ", " << (void*)buf << ", " << size << ", " << offset << ") = " << r << dendl; | |
9084 | if (r >= 0) { | |
9f95a23c | 9085 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
9086 | r = bl.length(); |
9087 | } | |
9088 | return r; | |
9089 | } | |
9090 | ||
9091 | int Client::preadv(int fd, const struct iovec *iov, int iovcnt, loff_t offset) | |
9092 | { | |
9093 | if (iovcnt < 0) | |
9094 | return -EINVAL; | |
9095 | return _preadv_pwritev(fd, iov, iovcnt, offset, false); | |
9096 | } | |
9097 | ||
11fdf7f2 | 9098 | int64_t Client::_read(Fh *f, int64_t offset, uint64_t size, bufferlist *bl) |
7c673cae | 9099 | { |
11fdf7f2 TL |
9100 | int want, have = 0; |
9101 | bool movepos = false; | |
9102 | std::unique_ptr<C_SaferCond> onuninline; | |
9103 | int64_t r = 0; | |
9104 | const auto& conf = cct->_conf; | |
7c673cae | 9105 | Inode *in = f->inode.get(); |
11fdf7f2 TL |
9106 | utime_t lat; |
9107 | utime_t start = ceph_clock_now(); | |
7c673cae FG |
9108 | |
9109 | if ((f->mode & CEPH_FILE_MODE_RD) == 0) | |
9110 | return -EBADF; | |
9111 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; | |
9112 | ||
7c673cae FG |
9113 | if (offset < 0) { |
9114 | lock_fh_pos(f); | |
9115 | offset = f->pos; | |
9116 | movepos = true; | |
9117 | } | |
9118 | loff_t start_pos = offset; | |
9119 | ||
9120 | if (in->inline_version == 0) { | |
11fdf7f2 | 9121 | r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); |
c07f9fc5 | 9122 | if (r < 0) { |
11fdf7f2 | 9123 | goto done; |
c07f9fc5 | 9124 | } |
11fdf7f2 | 9125 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
9126 | } |
9127 | ||
9128 | retry: | |
11fdf7f2 TL |
9129 | if (f->mode & CEPH_FILE_MODE_LAZY) |
9130 | want = CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO; | |
9131 | else | |
9132 | want = CEPH_CAP_FILE_CACHE; | |
9133 | r = get_caps(in, CEPH_CAP_FILE_RD, want, &have, -1); | |
c07f9fc5 | 9134 | if (r < 0) { |
11fdf7f2 | 9135 | goto done; |
c07f9fc5 | 9136 | } |
7c673cae | 9137 | if (f->flags & O_DIRECT) |
11fdf7f2 | 9138 | have &= ~(CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
9139 | |
9140 | if (in->inline_version < CEPH_INLINE_NONE) { | |
9141 | if (!(have & CEPH_CAP_FILE_CACHE)) { | |
11fdf7f2 TL |
9142 | onuninline.reset(new C_SaferCond("Client::_read_uninline_data flock")); |
9143 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
9144 | } else { |
9145 | uint32_t len = in->inline_data.length(); | |
7c673cae FG |
9146 | uint64_t endoff = offset + size; |
9147 | if (endoff > in->size) | |
9148 | endoff = in->size; | |
9149 | ||
9150 | if (offset < len) { | |
9151 | if (endoff <= len) { | |
9152 | bl->substr_of(in->inline_data, offset, endoff - offset); | |
9153 | } else { | |
9154 | bl->substr_of(in->inline_data, offset, len - offset); | |
9155 | bl->append_zero(endoff - len); | |
9156 | } | |
11fdf7f2 | 9157 | r = endoff - offset; |
7c673cae FG |
9158 | } else if ((uint64_t)offset < endoff) { |
9159 | bl->append_zero(endoff - offset); | |
11fdf7f2 TL |
9160 | r = endoff - offset; |
9161 | } else { | |
9162 | r = 0; | |
7c673cae | 9163 | } |
7c673cae FG |
9164 | goto success; |
9165 | } | |
9166 | } | |
9167 | ||
9168 | if (!conf->client_debug_force_sync_read && | |
11fdf7f2 TL |
9169 | conf->client_oc && |
9170 | (have & (CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
9171 | |
9172 | if (f->flags & O_RSYNC) { | |
9173 | _flush_range(in, offset, size); | |
9174 | } | |
9175 | r = _read_async(f, offset, size, bl); | |
9176 | if (r < 0) | |
9177 | goto done; | |
9178 | } else { | |
9179 | if (f->flags & O_DIRECT) | |
9180 | _flush_range(in, offset, size); | |
9181 | ||
9182 | bool checkeof = false; | |
9183 | r = _read_sync(f, offset, size, bl, &checkeof); | |
9184 | if (r < 0) | |
9185 | goto done; | |
9186 | if (checkeof) { | |
9187 | offset += r; | |
9188 | size -= r; | |
9189 | ||
9190 | put_cap_ref(in, CEPH_CAP_FILE_RD); | |
9191 | have = 0; | |
9192 | // reverify size | |
9193 | r = _getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
9194 | if (r < 0) | |
9195 | goto done; | |
9196 | ||
9197 | // eof? short read. | |
9198 | if ((uint64_t)offset < in->size) | |
9199 | goto retry; | |
9200 | } | |
9201 | } | |
9202 | ||
9203 | success: | |
11fdf7f2 | 9204 | ceph_assert(r >= 0); |
7c673cae FG |
9205 | if (movepos) { |
9206 | // adjust fd pos | |
11fdf7f2 | 9207 | f->pos = start_pos + r; |
7c673cae | 9208 | } |
11fdf7f2 TL |
9209 | |
9210 | lat = ceph_clock_now(); | |
9211 | lat -= start; | |
9212 | logger->tinc(l_c_read, lat); | |
7c673cae FG |
9213 | |
9214 | done: | |
9215 | // done! | |
11fdf7f2 | 9216 | |
7c673cae | 9217 | if (onuninline) { |
9f95a23c | 9218 | client_lock.unlock(); |
11fdf7f2 | 9219 | int ret = onuninline->wait(); |
9f95a23c | 9220 | client_lock.lock(); |
11fdf7f2 | 9221 | if (ret >= 0 || ret == -ECANCELED) { |
7c673cae FG |
9222 | in->inline_data.clear(); |
9223 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 9224 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
9225 | check_caps(in, 0); |
9226 | } else | |
11fdf7f2 | 9227 | r = ret; |
7c673cae | 9228 | } |
11fdf7f2 | 9229 | if (have) { |
7c673cae | 9230 | put_cap_ref(in, CEPH_CAP_FILE_RD); |
11fdf7f2 TL |
9231 | } |
9232 | if (movepos) { | |
9233 | unlock_fh_pos(f); | |
9234 | } | |
9235 | return r; | |
7c673cae FG |
9236 | } |
9237 | ||
9238 | Client::C_Readahead::C_Readahead(Client *c, Fh *f) : | |
9239 | client(c), f(f) { | |
9240 | f->get(); | |
9241 | f->readahead.inc_pending(); | |
9242 | } | |
9243 | ||
9244 | Client::C_Readahead::~C_Readahead() { | |
9245 | f->readahead.dec_pending(); | |
9246 | client->_put_fh(f); | |
9247 | } | |
9248 | ||
9249 | void Client::C_Readahead::finish(int r) { | |
9250 | lgeneric_subdout(client->cct, client, 20) << "client." << client->get_nodeid() << " " << "C_Readahead on " << f->inode << dendl; | |
9251 | client->put_cap_ref(f->inode.get(), CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
9252 | } | |
9253 | ||
9254 | int Client::_read_async(Fh *f, uint64_t off, uint64_t len, bufferlist *bl) | |
9255 | { | |
11fdf7f2 | 9256 | const auto& conf = cct->_conf; |
7c673cae FG |
9257 | Inode *in = f->inode.get(); |
9258 | ||
11fdf7f2 | 9259 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae FG |
9260 | |
9261 | // trim read based on file size? | |
9262 | if (off >= in->size) | |
9263 | return 0; | |
9264 | if (len == 0) | |
9265 | return 0; | |
9266 | if (off + len > in->size) { | |
9267 | len = in->size - off; | |
9268 | } | |
9269 | ||
9270 | ldout(cct, 10) << " min_bytes=" << f->readahead.get_min_readahead_size() | |
9271 | << " max_bytes=" << f->readahead.get_max_readahead_size() | |
9272 | << " max_periods=" << conf->client_readahead_max_periods << dendl; | |
9273 | ||
9274 | // read (and possibly block) | |
11fdf7f2 TL |
9275 | int r = 0; |
9276 | C_SaferCond onfinish("Client::_read_async flock"); | |
7c673cae | 9277 | r = objectcacher->file_read(&in->oset, &in->layout, in->snapid, |
11fdf7f2 | 9278 | off, len, bl, 0, &onfinish); |
7c673cae FG |
9279 | if (r == 0) { |
9280 | get_cap_ref(in, CEPH_CAP_FILE_CACHE); | |
9f95a23c | 9281 | client_lock.unlock(); |
11fdf7f2 | 9282 | r = onfinish.wait(); |
9f95a23c | 9283 | client_lock.lock(); |
7c673cae | 9284 | put_cap_ref(in, CEPH_CAP_FILE_CACHE); |
7c673cae FG |
9285 | } |
9286 | ||
9287 | if(f->readahead.get_min_readahead_size() > 0) { | |
9288 | pair<uint64_t, uint64_t> readahead_extent = f->readahead.update(off, len, in->size); | |
9289 | if (readahead_extent.second > 0) { | |
9290 | ldout(cct, 20) << "readahead " << readahead_extent.first << "~" << readahead_extent.second | |
9291 | << " (caller wants " << off << "~" << len << ")" << dendl; | |
9292 | Context *onfinish2 = new C_Readahead(this, f); | |
9293 | int r2 = objectcacher->file_read(&in->oset, &in->layout, in->snapid, | |
9294 | readahead_extent.first, readahead_extent.second, | |
9295 | NULL, 0, onfinish2); | |
9296 | if (r2 == 0) { | |
9297 | ldout(cct, 20) << "readahead initiated, c " << onfinish2 << dendl; | |
9298 | get_cap_ref(in, CEPH_CAP_FILE_RD | CEPH_CAP_FILE_CACHE); | |
9299 | } else { | |
9300 | ldout(cct, 20) << "readahead was no-op, already cached" << dendl; | |
9301 | delete onfinish2; | |
9302 | } | |
9303 | } | |
9304 | } | |
9305 | ||
9306 | return r; | |
9307 | } | |
9308 | ||
9309 | int Client::_read_sync(Fh *f, uint64_t off, uint64_t len, bufferlist *bl, | |
9310 | bool *checkeof) | |
9311 | { | |
9312 | Inode *in = f->inode.get(); | |
9313 | uint64_t pos = off; | |
9314 | int left = len; | |
9315 | int read = 0; | |
9316 | ||
11fdf7f2 | 9317 | ldout(cct, 10) << __func__ << " " << *in << " " << off << "~" << len << dendl; |
7c673cae | 9318 | |
7c673cae | 9319 | while (left > 0) { |
11fdf7f2 | 9320 | C_SaferCond onfinish("Client::_read_sync flock"); |
7c673cae FG |
9321 | bufferlist tbl; |
9322 | ||
9323 | int wanted = left; | |
9324 | filer->read_trunc(in->ino, &in->layout, in->snapid, | |
9325 | pos, left, &tbl, 0, | |
9326 | in->truncate_size, in->truncate_seq, | |
11fdf7f2 | 9327 | &onfinish); |
9f95a23c | 9328 | client_lock.unlock(); |
11fdf7f2 | 9329 | int r = onfinish.wait(); |
9f95a23c | 9330 | client_lock.lock(); |
7c673cae FG |
9331 | |
9332 | // if we get ENOENT from OSD, assume 0 bytes returned | |
9333 | if (r == -ENOENT) | |
9334 | r = 0; | |
9335 | if (r < 0) | |
9336 | return r; | |
9337 | if (tbl.length()) { | |
9338 | r = tbl.length(); | |
9339 | ||
9340 | read += r; | |
9341 | pos += r; | |
9342 | left -= r; | |
9343 | bl->claim_append(tbl); | |
9344 | } | |
9345 | // short read? | |
9346 | if (r >= 0 && r < wanted) { | |
9347 | if (pos < in->size) { | |
9348 | // zero up to known EOF | |
9349 | int64_t some = in->size - pos; | |
9350 | if (some > left) | |
9351 | some = left; | |
11fdf7f2 TL |
9352 | auto z = buffer::ptr_node::create(some); |
9353 | z->zero(); | |
9354 | bl->push_back(std::move(z)); | |
7c673cae FG |
9355 | read += some; |
9356 | pos += some; | |
9357 | left -= some; | |
9358 | if (left == 0) | |
9359 | return read; | |
9360 | } | |
9361 | ||
9362 | *checkeof = true; | |
9363 | return read; | |
9364 | } | |
9365 | } | |
9366 | return read; | |
9367 | } | |
9368 | ||
9369 | ||
9370 | /* | |
9371 | * we keep count of uncommitted sync writes on the inode, so that | |
9372 | * fsync can DDRT. | |
9373 | */ | |
9374 | void Client::_sync_write_commit(Inode *in) | |
9375 | { | |
11fdf7f2 | 9376 | ceph_assert(unsafe_sync_write > 0); |
7c673cae FG |
9377 | unsafe_sync_write--; |
9378 | ||
9379 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9380 | ||
11fdf7f2 | 9381 | ldout(cct, 15) << __func__ << " unsafe_sync_write = " << unsafe_sync_write << dendl; |
7c673cae | 9382 | if (unsafe_sync_write == 0 && unmounting) { |
11fdf7f2 | 9383 | ldout(cct, 10) << __func__ << " -- no more unsafe writes, unmount can proceed" << dendl; |
9f95a23c | 9384 | mount_cond.notify_all(); |
7c673cae FG |
9385 | } |
9386 | } | |
9387 | ||
9388 | int Client::write(int fd, const char *buf, loff_t size, loff_t offset) | |
9389 | { | |
11fdf7f2 | 9390 | std::lock_guard lock(client_lock); |
7c673cae FG |
9391 | tout(cct) << "write" << std::endl; |
9392 | tout(cct) << fd << std::endl; | |
9393 | tout(cct) << size << std::endl; | |
9394 | tout(cct) << offset << std::endl; | |
9395 | ||
181888fb FG |
9396 | if (unmounting) |
9397 | return -ENOTCONN; | |
9398 | ||
7c673cae FG |
9399 | Fh *fh = get_filehandle(fd); |
9400 | if (!fh) | |
9401 | return -EBADF; | |
9402 | #if defined(__linux__) && defined(O_PATH) | |
9403 | if (fh->flags & O_PATH) | |
9404 | return -EBADF; | |
9405 | #endif | |
11fdf7f2 TL |
9406 | /* We can't return bytes written larger than INT_MAX, clamp size to that */ |
9407 | size = std::min(size, (loff_t)INT_MAX); | |
9408 | int r = _write(fh, offset, size, buf, NULL, false); | |
7c673cae FG |
9409 | ldout(cct, 3) << "write(" << fd << ", \"...\", " << size << ", " << offset << ") = " << r << dendl; |
9410 | return r; | |
9411 | } | |
9412 | ||
9413 | int Client::pwritev(int fd, const struct iovec *iov, int iovcnt, int64_t offset) | |
9414 | { | |
9415 | if (iovcnt < 0) | |
9416 | return -EINVAL; | |
9417 | return _preadv_pwritev(fd, iov, iovcnt, offset, true); | |
9418 | } | |
9419 | ||
11fdf7f2 TL |
9420 | int64_t Client::_preadv_pwritev_locked(Fh *fh, const struct iovec *iov, |
9421 | unsigned iovcnt, int64_t offset, bool write, | |
9422 | bool clamp_to_int) | |
7c673cae | 9423 | { |
7c673cae FG |
9424 | #if defined(__linux__) && defined(O_PATH) |
9425 | if (fh->flags & O_PATH) | |
9426 | return -EBADF; | |
9427 | #endif | |
9428 | loff_t totallen = 0; | |
9429 | for (unsigned i = 0; i < iovcnt; i++) { | |
9430 | totallen += iov[i].iov_len; | |
9431 | } | |
11fdf7f2 TL |
9432 | |
9433 | /* | |
9434 | * Some of the API functions take 64-bit size values, but only return | |
9435 | * 32-bit signed integers. Clamp the I/O sizes in those functions so that | |
9436 | * we don't do I/Os larger than the values we can return. | |
9437 | */ | |
9438 | if (clamp_to_int) { | |
9439 | totallen = std::min(totallen, (loff_t)INT_MAX); | |
9440 | } | |
7c673cae | 9441 | if (write) { |
11fdf7f2 TL |
9442 | int64_t w = _write(fh, offset, totallen, NULL, iov, iovcnt); |
9443 | ldout(cct, 3) << "pwritev(" << fh << ", \"...\", " << totallen << ", " << offset << ") = " << w << dendl; | |
7c673cae FG |
9444 | return w; |
9445 | } else { | |
9446 | bufferlist bl; | |
11fdf7f2 TL |
9447 | int64_t r = _read(fh, offset, totallen, &bl); |
9448 | ldout(cct, 3) << "preadv(" << fh << ", " << offset << ") = " << r << dendl; | |
7c673cae FG |
9449 | if (r <= 0) |
9450 | return r; | |
9451 | ||
9f95a23c | 9452 | auto iter = bl.cbegin(); |
7c673cae FG |
9453 | for (unsigned j = 0, resid = r; j < iovcnt && resid > 0; j++) { |
9454 | /* | |
9455 | * This piece of code aims to handle the case that bufferlist does not have enough data | |
9456 | * to fill in the iov | |
9457 | */ | |
9f95a23c TL |
9458 | const auto round_size = std::min<unsigned>(resid, iov[j].iov_len); |
9459 | iter.copy(round_size, reinterpret_cast<char*>(iov[j].iov_base)); | |
9460 | resid -= round_size; | |
9461 | /* iter is self-updating */ | |
7c673cae FG |
9462 | } |
9463 | return r; | |
9464 | } | |
9465 | } | |
9466 | ||
11fdf7f2 TL |
9467 | int Client::_preadv_pwritev(int fd, const struct iovec *iov, unsigned iovcnt, int64_t offset, bool write) |
9468 | { | |
9469 | std::lock_guard lock(client_lock); | |
9470 | tout(cct) << fd << std::endl; | |
9471 | tout(cct) << offset << std::endl; | |
9472 | ||
9473 | if (unmounting) | |
9474 | return -ENOTCONN; | |
9475 | ||
9476 | Fh *fh = get_filehandle(fd); | |
9477 | if (!fh) | |
9478 | return -EBADF; | |
9479 | return _preadv_pwritev_locked(fh, iov, iovcnt, offset, write, true); | |
9480 | } | |
9481 | ||
9482 | int64_t Client::_write(Fh *f, int64_t offset, uint64_t size, const char *buf, | |
9483 | const struct iovec *iov, int iovcnt) | |
7c673cae | 9484 | { |
f64942e4 AA |
9485 | uint64_t fpos = 0; |
9486 | ||
7c673cae FG |
9487 | if ((uint64_t)(offset+size) > mdsmap->get_max_filesize()) //too large! |
9488 | return -EFBIG; | |
9489 | ||
9490 | //ldout(cct, 7) << "write fh " << fh << " size " << size << " offset " << offset << dendl; | |
9491 | Inode *in = f->inode.get(); | |
9492 | ||
9493 | if (objecter->osdmap_pool_full(in->layout.pool_id)) { | |
9494 | return -ENOSPC; | |
9495 | } | |
9496 | ||
11fdf7f2 | 9497 | ceph_assert(in->snapid == CEPH_NOSNAP); |
7c673cae FG |
9498 | |
9499 | // was Fh opened as writeable? | |
9500 | if ((f->mode & CEPH_FILE_MODE_WR) == 0) | |
9501 | return -EBADF; | |
9502 | ||
7c673cae FG |
9503 | // use/adjust fd pos? |
9504 | if (offset < 0) { | |
9505 | lock_fh_pos(f); | |
9506 | /* | |
9507 | * FIXME: this is racy in that we may block _after_ this point waiting for caps, and size may | |
9508 | * change out from under us. | |
9509 | */ | |
9510 | if (f->flags & O_APPEND) { | |
9f95a23c | 9511 | auto r = _lseek(f, 0, SEEK_END); |
7c673cae FG |
9512 | if (r < 0) { |
9513 | unlock_fh_pos(f); | |
9514 | return r; | |
9515 | } | |
9516 | } | |
9517 | offset = f->pos; | |
f64942e4 | 9518 | fpos = offset+size; |
7c673cae FG |
9519 | unlock_fh_pos(f); |
9520 | } | |
9521 | ||
11fdf7f2 TL |
9522 | // check quota |
9523 | uint64_t endoff = offset + size; | |
9524 | if (endoff > in->size && is_quota_bytes_exceeded(in, endoff - in->size, | |
9525 | f->actor_perms)) { | |
9526 | return -EDQUOT; | |
9527 | } | |
9528 | ||
7c673cae FG |
9529 | //bool lazy = f->mode == CEPH_FILE_MODE_LAZY; |
9530 | ||
9531 | ldout(cct, 10) << "cur file size is " << in->size << dendl; | |
9532 | ||
9533 | // time it. | |
9534 | utime_t start = ceph_clock_now(); | |
9535 | ||
9536 | if (in->inline_version == 0) { | |
9537 | int r = _getattr(in, CEPH_STAT_CAP_INLINE_DATA, f->actor_perms, true); | |
9538 | if (r < 0) | |
9539 | return r; | |
11fdf7f2 | 9540 | ceph_assert(in->inline_version > 0); |
7c673cae FG |
9541 | } |
9542 | ||
9543 | // copy into fresh buffer (since our write may be resub, async) | |
9544 | bufferlist bl; | |
9545 | if (buf) { | |
9546 | if (size > 0) | |
9547 | bl.append(buf, size); | |
9548 | } else if (iov){ | |
9549 | for (int i = 0; i < iovcnt; i++) { | |
9550 | if (iov[i].iov_len > 0) { | |
9551 | bl.append((const char *)iov[i].iov_base, iov[i].iov_len); | |
9552 | } | |
9553 | } | |
9554 | } | |
9555 | ||
9556 | utime_t lat; | |
9557 | uint64_t totalwritten; | |
11fdf7f2 TL |
9558 | int want, have; |
9559 | if (f->mode & CEPH_FILE_MODE_LAZY) | |
9560 | want = CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO; | |
9561 | else | |
9562 | want = CEPH_CAP_FILE_BUFFER; | |
9563 | int r = get_caps(in, CEPH_CAP_FILE_WR|CEPH_CAP_AUTH_SHARED, want, &have, endoff); | |
7c673cae FG |
9564 | if (r < 0) |
9565 | return r; | |
9566 | ||
9567 | /* clear the setuid/setgid bits, if any */ | |
181888fb | 9568 | if (unlikely(in->mode & (S_ISUID|S_ISGID)) && size > 0) { |
7c673cae FG |
9569 | struct ceph_statx stx = { 0 }; |
9570 | ||
9571 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
9572 | r = __setattrx(in, &stx, CEPH_SETATTR_KILL_SGUID, f->actor_perms); | |
9573 | if (r < 0) | |
9574 | return r; | |
9575 | } else { | |
9576 | put_cap_ref(in, CEPH_CAP_AUTH_SHARED); | |
9577 | } | |
9578 | ||
9579 | if (f->flags & O_DIRECT) | |
11fdf7f2 | 9580 | have &= ~(CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO); |
7c673cae FG |
9581 | |
9582 | ldout(cct, 10) << " snaprealm " << *in->snaprealm << dendl; | |
9583 | ||
11fdf7f2 TL |
9584 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
9585 | ||
7c673cae FG |
9586 | if (in->inline_version < CEPH_INLINE_NONE) { |
9587 | if (endoff > cct->_conf->client_max_inline_size || | |
9588 | endoff > CEPH_INLINE_MAX_SIZE || | |
9589 | !(have & CEPH_CAP_FILE_BUFFER)) { | |
11fdf7f2 TL |
9590 | onuninline.reset(new C_SaferCond("Client::_write_uninline_data flock")); |
9591 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
9592 | } else { |
9593 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9594 | ||
9595 | uint32_t len = in->inline_data.length(); | |
9596 | ||
9597 | if (endoff < len) | |
9f95a23c | 9598 | in->inline_data.begin(endoff).copy(len - endoff, bl); // XXX |
7c673cae FG |
9599 | |
9600 | if (offset < len) | |
9601 | in->inline_data.splice(offset, len - offset); | |
9602 | else if (offset > len) | |
9603 | in->inline_data.append_zero(offset - len); | |
9604 | ||
9605 | in->inline_data.append(bl); | |
9606 | in->inline_version++; | |
9607 | ||
9608 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9609 | ||
9610 | goto success; | |
9611 | } | |
9612 | } | |
9613 | ||
11fdf7f2 TL |
9614 | if (cct->_conf->client_oc && |
9615 | (have & (CEPH_CAP_FILE_BUFFER | CEPH_CAP_FILE_LAZYIO))) { | |
7c673cae FG |
9616 | // do buffered write |
9617 | if (!in->oset.dirty_or_tx) | |
9618 | get_cap_ref(in, CEPH_CAP_FILE_CACHE | CEPH_CAP_FILE_BUFFER); | |
9619 | ||
9620 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9621 | ||
9622 | // async, caching, non-blocking. | |
9623 | r = objectcacher->file_write(&in->oset, &in->layout, | |
9624 | in->snaprealm->get_snap_context(), | |
9625 | offset, size, bl, ceph::real_clock::now(), | |
9626 | 0); | |
9627 | put_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
9628 | ||
9629 | if (r < 0) | |
9630 | goto done; | |
9631 | ||
9632 | // flush cached write if O_SYNC is set on file fh | |
9633 | // O_DSYNC == O_SYNC on linux < 2.6.33 | |
9634 | // O_SYNC = __O_SYNC | O_DSYNC on linux >= 2.6.33 | |
9635 | if ((f->flags & O_SYNC) || (f->flags & O_DSYNC)) { | |
9636 | _flush_range(in, offset, size); | |
9637 | } | |
9638 | } else { | |
9639 | if (f->flags & O_DIRECT) | |
9640 | _flush_range(in, offset, size); | |
9641 | ||
9642 | // simple, non-atomic sync write | |
11fdf7f2 | 9643 | C_SaferCond onfinish("Client::_write flock"); |
7c673cae FG |
9644 | unsafe_sync_write++; |
9645 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); // released by onsafe callback | |
9646 | ||
9647 | filer->write_trunc(in->ino, &in->layout, in->snaprealm->get_snap_context(), | |
9648 | offset, size, bl, ceph::real_clock::now(), 0, | |
9649 | in->truncate_size, in->truncate_seq, | |
11fdf7f2 | 9650 | &onfinish); |
9f95a23c | 9651 | client_lock.unlock(); |
11fdf7f2 | 9652 | onfinish.wait(); |
9f95a23c | 9653 | client_lock.lock(); |
7c673cae FG |
9654 | _sync_write_commit(in); |
9655 | } | |
9656 | ||
9657 | // if we get here, write was successful, update client metadata | |
9658 | success: | |
9659 | // time | |
9660 | lat = ceph_clock_now(); | |
9661 | lat -= start; | |
9662 | logger->tinc(l_c_wrlat, lat); | |
9663 | ||
f64942e4 AA |
9664 | if (fpos) { |
9665 | lock_fh_pos(f); | |
9666 | f->pos = fpos; | |
9667 | unlock_fh_pos(f); | |
9668 | } | |
7c673cae | 9669 | totalwritten = size; |
11fdf7f2 | 9670 | r = (int64_t)totalwritten; |
7c673cae FG |
9671 | |
9672 | // extend file? | |
9673 | if (totalwritten + offset > in->size) { | |
9674 | in->size = totalwritten + offset; | |
28e407b8 | 9675 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 9676 | |
11fdf7f2 | 9677 | if (is_quota_bytes_approaching(in, f->actor_perms)) { |
7c673cae | 9678 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
9679 | } else if (is_max_size_approaching(in)) { |
9680 | check_caps(in, 0); | |
7c673cae FG |
9681 | } |
9682 | ||
9683 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", extending file size" << dendl; | |
9684 | } else { | |
9685 | ldout(cct, 7) << "wrote to " << totalwritten+offset << ", leaving file size at " << in->size << dendl; | |
9686 | } | |
9687 | ||
9688 | // mtime | |
91327a77 | 9689 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 9690 | in->change_attr++; |
28e407b8 | 9691 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
9692 | |
9693 | done: | |
9694 | ||
11fdf7f2 | 9695 | if (nullptr != onuninline) { |
9f95a23c | 9696 | client_lock.unlock(); |
11fdf7f2 | 9697 | int uninline_ret = onuninline->wait(); |
9f95a23c | 9698 | client_lock.lock(); |
7c673cae FG |
9699 | |
9700 | if (uninline_ret >= 0 || uninline_ret == -ECANCELED) { | |
9701 | in->inline_data.clear(); | |
9702 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 9703 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
9704 | check_caps(in, 0); |
9705 | } else | |
9706 | r = uninline_ret; | |
9707 | } | |
9708 | ||
9709 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
9710 | return r; | |
9711 | } | |
9712 | ||
9713 | int Client::_flush(Fh *f) | |
9714 | { | |
9715 | Inode *in = f->inode.get(); | |
9716 | int err = f->take_async_err(); | |
9717 | if (err != 0) { | |
9718 | ldout(cct, 1) << __func__ << ": " << f << " on inode " << *in << " caught async_err = " | |
9719 | << cpp_strerror(err) << dendl; | |
9720 | } else { | |
9721 | ldout(cct, 10) << __func__ << ": " << f << " on inode " << *in << " no async_err state" << dendl; | |
9722 | } | |
9723 | ||
9724 | return err; | |
9725 | } | |
9726 | ||
9727 | int Client::truncate(const char *relpath, loff_t length, const UserPerm& perms) | |
9728 | { | |
9729 | struct ceph_statx stx; | |
9730 | stx.stx_size = length; | |
9731 | return setattrx(relpath, &stx, CEPH_SETATTR_SIZE, perms); | |
9732 | } | |
9733 | ||
9734 | int Client::ftruncate(int fd, loff_t length, const UserPerm& perms) | |
9735 | { | |
11fdf7f2 TL |
9736 | std::lock_guard lock(client_lock); |
9737 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
9738 | tout(cct) << fd << std::endl; |
9739 | tout(cct) << length << std::endl; | |
9740 | ||
181888fb FG |
9741 | if (unmounting) |
9742 | return -ENOTCONN; | |
9743 | ||
7c673cae FG |
9744 | Fh *f = get_filehandle(fd); |
9745 | if (!f) | |
9746 | return -EBADF; | |
9747 | #if defined(__linux__) && defined(O_PATH) | |
9748 | if (f->flags & O_PATH) | |
9749 | return -EBADF; | |
9750 | #endif | |
9751 | struct stat attr; | |
9752 | attr.st_size = length; | |
9753 | return _setattr(f->inode, &attr, CEPH_SETATTR_SIZE, perms); | |
9754 | } | |
9755 | ||
9756 | int Client::fsync(int fd, bool syncdataonly) | |
9757 | { | |
11fdf7f2 | 9758 | std::lock_guard lock(client_lock); |
7c673cae FG |
9759 | tout(cct) << "fsync" << std::endl; |
9760 | tout(cct) << fd << std::endl; | |
9761 | tout(cct) << syncdataonly << std::endl; | |
9762 | ||
181888fb FG |
9763 | if (unmounting) |
9764 | return -ENOTCONN; | |
9765 | ||
7c673cae FG |
9766 | Fh *f = get_filehandle(fd); |
9767 | if (!f) | |
9768 | return -EBADF; | |
9769 | #if defined(__linux__) && defined(O_PATH) | |
9770 | if (f->flags & O_PATH) | |
9771 | return -EBADF; | |
9772 | #endif | |
9773 | int r = _fsync(f, syncdataonly); | |
9774 | if (r == 0) { | |
9775 | // The IOs in this fsync were okay, but maybe something happened | |
9776 | // in the background that we shoudl be reporting? | |
9777 | r = f->take_async_err(); | |
1adf2230 | 9778 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly |
7c673cae FG |
9779 | << ") = 0, async_err = " << r << dendl; |
9780 | } else { | |
9781 | // Assume that an error we encountered during fsync, even reported | |
9782 | // synchronously, would also have applied the error to the Fh, and we | |
9783 | // should clear it here to avoid returning the same error again on next | |
9784 | // call. | |
1adf2230 | 9785 | ldout(cct, 5) << "fsync(" << fd << ", " << syncdataonly << ") = " |
7c673cae FG |
9786 | << r << dendl; |
9787 | f->take_async_err(); | |
9788 | } | |
9789 | return r; | |
9790 | } | |
9791 | ||
9792 | int Client::_fsync(Inode *in, bool syncdataonly) | |
9793 | { | |
9794 | int r = 0; | |
11fdf7f2 | 9795 | std::unique_ptr<C_SaferCond> object_cacher_completion = nullptr; |
7c673cae FG |
9796 | ceph_tid_t flush_tid = 0; |
9797 | InodeRef tmp_ref; | |
11fdf7f2 TL |
9798 | utime_t lat; |
9799 | utime_t start = ceph_clock_now(); | |
7c673cae | 9800 | |
1adf2230 | 9801 | ldout(cct, 8) << "_fsync on " << *in << " " << (syncdataonly ? "(dataonly)":"(data+metadata)") << dendl; |
7c673cae FG |
9802 | |
9803 | if (cct->_conf->client_oc) { | |
11fdf7f2 TL |
9804 | object_cacher_completion.reset(new C_SaferCond("Client::_fsync::lock")); |
9805 | tmp_ref = in; // take a reference; C_SaferCond doesn't and _flush won't either | |
9806 | _flush(in, object_cacher_completion.get()); | |
7c673cae FG |
9807 | ldout(cct, 15) << "using return-valued form of _fsync" << dendl; |
9808 | } | |
9809 | ||
9810 | if (!syncdataonly && in->dirty_caps) { | |
9811 | check_caps(in, CHECK_CAPS_NODELAY|CHECK_CAPS_SYNCHRONOUS); | |
9812 | if (in->flushing_caps) | |
9813 | flush_tid = last_flush_tid; | |
9814 | } else ldout(cct, 10) << "no metadata needs to commit" << dendl; | |
9815 | ||
9816 | if (!syncdataonly && !in->unsafe_ops.empty()) { | |
28e407b8 AA |
9817 | flush_mdlog_sync(); |
9818 | ||
7c673cae FG |
9819 | MetaRequest *req = in->unsafe_ops.back(); |
9820 | ldout(cct, 15) << "waiting on unsafe requests, last tid " << req->get_tid() << dendl; | |
9821 | ||
9822 | req->get(); | |
9823 | wait_on_list(req->waitfor_safe); | |
9824 | put_request(req); | |
9825 | } | |
9826 | ||
11fdf7f2 | 9827 | if (nullptr != object_cacher_completion) { // wait on a real reply instead of guessing |
9f95a23c | 9828 | client_lock.unlock(); |
7c673cae | 9829 | ldout(cct, 15) << "waiting on data to flush" << dendl; |
11fdf7f2 | 9830 | r = object_cacher_completion->wait(); |
9f95a23c | 9831 | client_lock.lock(); |
7c673cae FG |
9832 | ldout(cct, 15) << "got " << r << " from flush writeback" << dendl; |
9833 | } else { | |
9834 | // FIXME: this can starve | |
9835 | while (in->cap_refs[CEPH_CAP_FILE_BUFFER] > 0) { | |
9836 | ldout(cct, 10) << "ino " << in->ino << " has " << in->cap_refs[CEPH_CAP_FILE_BUFFER] | |
9837 | << " uncommitted, waiting" << dendl; | |
9838 | wait_on_list(in->waitfor_commit); | |
9839 | } | |
9840 | } | |
9841 | ||
9842 | if (!r) { | |
9843 | if (flush_tid > 0) | |
9844 | wait_sync_caps(in, flush_tid); | |
9845 | ||
9846 | ldout(cct, 10) << "ino " << in->ino << " has no uncommitted writes" << dendl; | |
9847 | } else { | |
1adf2230 | 9848 | ldout(cct, 8) << "ino " << in->ino << " failed to commit to disk! " |
7c673cae FG |
9849 | << cpp_strerror(-r) << dendl; |
9850 | } | |
11fdf7f2 TL |
9851 | |
9852 | lat = ceph_clock_now(); | |
9853 | lat -= start; | |
9854 | logger->tinc(l_c_fsync, lat); | |
7c673cae FG |
9855 | |
9856 | return r; | |
9857 | } | |
9858 | ||
9859 | int Client::_fsync(Fh *f, bool syncdataonly) | |
9860 | { | |
1adf2230 | 9861 | ldout(cct, 8) << "_fsync(" << f << ", " << (syncdataonly ? "dataonly)":"data+metadata)") << dendl; |
7c673cae FG |
9862 | return _fsync(f->inode.get(), syncdataonly); |
9863 | } | |
9864 | ||
9865 | int Client::fstat(int fd, struct stat *stbuf, const UserPerm& perms, int mask) | |
9866 | { | |
11fdf7f2 | 9867 | std::lock_guard lock(client_lock); |
7c673cae FG |
9868 | tout(cct) << "fstat mask " << hex << mask << dec << std::endl; |
9869 | tout(cct) << fd << std::endl; | |
9870 | ||
181888fb FG |
9871 | if (unmounting) |
9872 | return -ENOTCONN; | |
9873 | ||
7c673cae FG |
9874 | Fh *f = get_filehandle(fd); |
9875 | if (!f) | |
9876 | return -EBADF; | |
9877 | int r = _getattr(f->inode, mask, perms); | |
9878 | if (r < 0) | |
9879 | return r; | |
9880 | fill_stat(f->inode, stbuf, NULL); | |
1adf2230 | 9881 | ldout(cct, 5) << "fstat(" << fd << ", " << stbuf << ") = " << r << dendl; |
7c673cae FG |
9882 | return r; |
9883 | } | |
9884 | ||
9885 | int Client::fstatx(int fd, struct ceph_statx *stx, const UserPerm& perms, | |
9886 | unsigned int want, unsigned int flags) | |
9887 | { | |
11fdf7f2 | 9888 | std::lock_guard lock(client_lock); |
7c673cae FG |
9889 | tout(cct) << "fstatx flags " << hex << flags << " want " << want << dec << std::endl; |
9890 | tout(cct) << fd << std::endl; | |
9891 | ||
181888fb FG |
9892 | if (unmounting) |
9893 | return -ENOTCONN; | |
9894 | ||
7c673cae FG |
9895 | Fh *f = get_filehandle(fd); |
9896 | if (!f) | |
9897 | return -EBADF; | |
9898 | ||
9899 | unsigned mask = statx_to_mask(flags, want); | |
9900 | ||
9901 | int r = 0; | |
94b18763 | 9902 | if (mask && !f->inode->caps_issued_mask(mask, true)) { |
7c673cae FG |
9903 | r = _getattr(f->inode, mask, perms); |
9904 | if (r < 0) { | |
9905 | ldout(cct, 3) << "fstatx exit on error!" << dendl; | |
9906 | return r; | |
9907 | } | |
9908 | } | |
9909 | ||
9910 | fill_statx(f->inode, mask, stx); | |
9911 | ldout(cct, 3) << "fstatx(" << fd << ", " << stx << ") = " << r << dendl; | |
9912 | return r; | |
9913 | } | |
9914 | ||
9915 | // not written yet, but i want to link! | |
9916 | ||
9917 | int Client::chdir(const char *relpath, std::string &new_cwd, | |
9918 | const UserPerm& perms) | |
9919 | { | |
11fdf7f2 | 9920 | std::lock_guard lock(client_lock); |
7c673cae FG |
9921 | tout(cct) << "chdir" << std::endl; |
9922 | tout(cct) << relpath << std::endl; | |
181888fb FG |
9923 | |
9924 | if (unmounting) | |
9925 | return -ENOTCONN; | |
9926 | ||
7c673cae FG |
9927 | filepath path(relpath); |
9928 | InodeRef in; | |
9929 | int r = path_walk(path, &in, perms); | |
9930 | if (r < 0) | |
9931 | return r; | |
92f5a8d4 TL |
9932 | |
9933 | if (!(in.get()->is_dir())) | |
9934 | return -ENOTDIR; | |
9935 | ||
7c673cae FG |
9936 | if (cwd != in) |
9937 | cwd.swap(in); | |
9938 | ldout(cct, 3) << "chdir(" << relpath << ") cwd now " << cwd->ino << dendl; | |
9939 | ||
b5b8bbf5 | 9940 | _getcwd(new_cwd, perms); |
7c673cae FG |
9941 | return 0; |
9942 | } | |
9943 | ||
b5b8bbf5 | 9944 | void Client::_getcwd(string& dir, const UserPerm& perms) |
7c673cae FG |
9945 | { |
9946 | filepath path; | |
11fdf7f2 | 9947 | ldout(cct, 10) << __func__ << " " << *cwd << dendl; |
7c673cae FG |
9948 | |
9949 | Inode *in = cwd.get(); | |
9950 | while (in != root) { | |
11fdf7f2 | 9951 | ceph_assert(in->dentries.size() < 2); // dirs can't be hard-linked |
7c673cae FG |
9952 | |
9953 | // A cwd or ancester is unlinked | |
11fdf7f2 | 9954 | if (in->dentries.empty()) { |
7c673cae FG |
9955 | return; |
9956 | } | |
9957 | ||
9958 | Dentry *dn = in->get_first_parent(); | |
9959 | ||
9960 | ||
9961 | if (!dn) { | |
9962 | // look it up | |
11fdf7f2 | 9963 | ldout(cct, 10) << __func__ << " looking up parent for " << *in << dendl; |
7c673cae FG |
9964 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LOOKUPNAME); |
9965 | filepath path(in->ino); | |
9966 | req->set_filepath(path); | |
9967 | req->set_inode(in); | |
9968 | int res = make_request(req, perms); | |
9969 | if (res < 0) | |
9970 | break; | |
9971 | ||
9972 | // start over | |
9973 | path = filepath(); | |
9974 | in = cwd.get(); | |
9975 | continue; | |
9976 | } | |
9977 | path.push_front_dentry(dn->name); | |
9978 | in = dn->dir->parent_inode; | |
9979 | } | |
9980 | dir = "/"; | |
9981 | dir += path.get_path(); | |
9982 | } | |
9983 | ||
b5b8bbf5 FG |
9984 | void Client::getcwd(string& dir, const UserPerm& perms) |
9985 | { | |
11fdf7f2 | 9986 | std::lock_guard l(client_lock); |
181888fb FG |
9987 | if (!unmounting) |
9988 | _getcwd(dir, perms); | |
b5b8bbf5 FG |
9989 | } |
9990 | ||
7c673cae FG |
9991 | int Client::statfs(const char *path, struct statvfs *stbuf, |
9992 | const UserPerm& perms) | |
9993 | { | |
11fdf7f2 TL |
9994 | std::lock_guard l(client_lock); |
9995 | tout(cct) << __func__ << std::endl; | |
91327a77 | 9996 | unsigned long int total_files_on_fs; |
7c673cae | 9997 | |
181888fb FG |
9998 | if (unmounting) |
9999 | return -ENOTCONN; | |
10000 | ||
7c673cae FG |
10001 | ceph_statfs stats; |
10002 | C_SaferCond cond; | |
d2e6a577 FG |
10003 | |
10004 | const vector<int64_t> &data_pools = mdsmap->get_data_pools(); | |
10005 | if (data_pools.size() == 1) { | |
10006 | objecter->get_fs_stats(stats, data_pools[0], &cond); | |
10007 | } else { | |
10008 | objecter->get_fs_stats(stats, boost::optional<int64_t>(), &cond); | |
10009 | } | |
7c673cae | 10010 | |
9f95a23c | 10011 | client_lock.unlock(); |
7c673cae | 10012 | int rval = cond.wait(); |
91327a77 AA |
10013 | assert(root); |
10014 | total_files_on_fs = root->rstat.rfiles + root->rstat.rsubdirs; | |
9f95a23c | 10015 | client_lock.lock(); |
7c673cae FG |
10016 | |
10017 | if (rval < 0) { | |
10018 | ldout(cct, 1) << "underlying call to statfs returned error: " | |
10019 | << cpp_strerror(rval) | |
10020 | << dendl; | |
10021 | return rval; | |
10022 | } | |
10023 | ||
10024 | memset(stbuf, 0, sizeof(*stbuf)); | |
10025 | ||
10026 | /* | |
10027 | * we're going to set a block size of 4MB so we can represent larger | |
10028 | * FSes without overflowing. Additionally convert the space | |
10029 | * measurements from KB to bytes while making them in terms of | |
10030 | * blocks. We use 4MB only because it is big enough, and because it | |
10031 | * actually *is* the (ceph) default block size. | |
10032 | */ | |
10033 | const int CEPH_BLOCK_SHIFT = 22; | |
10034 | stbuf->f_frsize = 1 << CEPH_BLOCK_SHIFT; | |
10035 | stbuf->f_bsize = 1 << CEPH_BLOCK_SHIFT; | |
91327a77 AA |
10036 | stbuf->f_files = total_files_on_fs; |
10037 | stbuf->f_ffree = 0; | |
7c673cae FG |
10038 | stbuf->f_favail = -1; |
10039 | stbuf->f_fsid = -1; // ?? | |
10040 | stbuf->f_flag = 0; // ?? | |
10041 | stbuf->f_namemax = NAME_MAX; | |
10042 | ||
10043 | // Usually quota_root will == root_ancestor, but if the mount root has no | |
10044 | // quota but we can see a parent of it that does have a quota, we'll | |
10045 | // respect that one instead. | |
11fdf7f2 | 10046 | ceph_assert(root != nullptr); |
7c673cae FG |
10047 | Inode *quota_root = root->quota.is_enable() ? root : get_quota_root(root, perms); |
10048 | ||
10049 | // get_quota_root should always give us something | |
10050 | // because client quotas are always enabled | |
11fdf7f2 | 10051 | ceph_assert(quota_root != nullptr); |
7c673cae FG |
10052 | |
10053 | if (quota_root && cct->_conf->client_quota_df && quota_root->quota.max_bytes) { | |
10054 | ||
10055 | // Skip the getattr if any sessions are stale, as we don't want to | |
10056 | // block `df` if this client has e.g. been evicted, or if the MDS cluster | |
10057 | // is unhealthy. | |
10058 | if (!_any_stale_sessions()) { | |
10059 | int r = _getattr(quota_root, 0, perms, true); | |
10060 | if (r != 0) { | |
10061 | // Ignore return value: error getting latest inode metadata is not a good | |
10062 | // reason to break "df". | |
10063 | lderr(cct) << "Error in getattr on quota root 0x" | |
10064 | << std::hex << quota_root->ino << std::dec | |
10065 | << " statfs result may be outdated" << dendl; | |
10066 | } | |
10067 | } | |
10068 | ||
10069 | // Special case: if there is a size quota set on the Inode acting | |
10070 | // as the root for this client mount, then report the quota status | |
10071 | // as the filesystem statistics. | |
10072 | const fsblkcnt_t total = quota_root->quota.max_bytes >> CEPH_BLOCK_SHIFT; | |
10073 | const fsblkcnt_t used = quota_root->rstat.rbytes >> CEPH_BLOCK_SHIFT; | |
31f18b77 FG |
10074 | // It is possible for a quota to be exceeded: arithmetic here must |
10075 | // handle case where used > total. | |
10076 | const fsblkcnt_t free = total > used ? total - used : 0; | |
7c673cae FG |
10077 | |
10078 | stbuf->f_blocks = total; | |
10079 | stbuf->f_bfree = free; | |
10080 | stbuf->f_bavail = free; | |
10081 | } else { | |
d2e6a577 | 10082 | // General case: report the cluster statistics returned from RADOS. Because |
7c673cae FG |
10083 | // multiple pools may be used without one filesystem namespace via |
10084 | // layouts, this is the most correct thing we can do. | |
10085 | stbuf->f_blocks = stats.kb >> (CEPH_BLOCK_SHIFT - 10); | |
10086 | stbuf->f_bfree = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
10087 | stbuf->f_bavail = stats.kb_avail >> (CEPH_BLOCK_SHIFT - 10); | |
10088 | } | |
10089 | ||
10090 | return rval; | |
10091 | } | |
10092 | ||
10093 | int Client::_do_filelock(Inode *in, Fh *fh, int lock_type, int op, int sleep, | |
10094 | struct flock *fl, uint64_t owner, bool removing) | |
10095 | { | |
11fdf7f2 | 10096 | ldout(cct, 10) << __func__ << " ino " << in->ino |
7c673cae FG |
10097 | << (lock_type == CEPH_LOCK_FCNTL ? " fcntl" : " flock") |
10098 | << " type " << fl->l_type << " owner " << owner | |
10099 | << " " << fl->l_start << "~" << fl->l_len << dendl; | |
10100 | ||
10101 | int lock_cmd; | |
10102 | if (F_RDLCK == fl->l_type) | |
10103 | lock_cmd = CEPH_LOCK_SHARED; | |
10104 | else if (F_WRLCK == fl->l_type) | |
10105 | lock_cmd = CEPH_LOCK_EXCL; | |
10106 | else if (F_UNLCK == fl->l_type) | |
10107 | lock_cmd = CEPH_LOCK_UNLOCK; | |
10108 | else | |
10109 | return -EIO; | |
10110 | ||
10111 | if (op != CEPH_MDS_OP_SETFILELOCK || lock_cmd == CEPH_LOCK_UNLOCK) | |
10112 | sleep = 0; | |
10113 | ||
10114 | /* | |
10115 | * Set the most significant bit, so that MDS knows the 'owner' | |
10116 | * is sufficient to identify the owner of lock. (old code uses | |
10117 | * both 'owner' and 'pid') | |
10118 | */ | |
10119 | owner |= (1ULL << 63); | |
10120 | ||
10121 | MetaRequest *req = new MetaRequest(op); | |
10122 | filepath path; | |
10123 | in->make_nosnap_relative_path(path); | |
10124 | req->set_filepath(path); | |
10125 | req->set_inode(in); | |
10126 | ||
10127 | req->head.args.filelock_change.rule = lock_type; | |
10128 | req->head.args.filelock_change.type = lock_cmd; | |
10129 | req->head.args.filelock_change.owner = owner; | |
10130 | req->head.args.filelock_change.pid = fl->l_pid; | |
10131 | req->head.args.filelock_change.start = fl->l_start; | |
10132 | req->head.args.filelock_change.length = fl->l_len; | |
10133 | req->head.args.filelock_change.wait = sleep; | |
10134 | ||
10135 | int ret; | |
10136 | bufferlist bl; | |
10137 | ||
10138 | if (sleep && switch_interrupt_cb) { | |
10139 | // enable interrupt | |
10140 | switch_interrupt_cb(callback_handle, req->get()); | |
10141 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
7c673cae FG |
10142 | // disable interrupt |
10143 | switch_interrupt_cb(callback_handle, NULL); | |
31f18b77 FG |
10144 | if (ret == 0 && req->aborted()) { |
10145 | // effect of this lock request has been revoked by the 'lock intr' request | |
10146 | ret = req->get_abort_code(); | |
10147 | } | |
7c673cae FG |
10148 | put_request(req); |
10149 | } else { | |
10150 | ret = make_request(req, fh->actor_perms, NULL, NULL, -1, &bl); | |
10151 | } | |
10152 | ||
10153 | if (ret == 0) { | |
10154 | if (op == CEPH_MDS_OP_GETFILELOCK) { | |
10155 | ceph_filelock filelock; | |
11fdf7f2 TL |
10156 | auto p = bl.cbegin(); |
10157 | decode(filelock, p); | |
7c673cae FG |
10158 | |
10159 | if (CEPH_LOCK_SHARED == filelock.type) | |
10160 | fl->l_type = F_RDLCK; | |
10161 | else if (CEPH_LOCK_EXCL == filelock.type) | |
10162 | fl->l_type = F_WRLCK; | |
10163 | else | |
10164 | fl->l_type = F_UNLCK; | |
10165 | ||
10166 | fl->l_whence = SEEK_SET; | |
10167 | fl->l_start = filelock.start; | |
10168 | fl->l_len = filelock.length; | |
10169 | fl->l_pid = filelock.pid; | |
10170 | } else if (op == CEPH_MDS_OP_SETFILELOCK) { | |
10171 | ceph_lock_state_t *lock_state; | |
10172 | if (lock_type == CEPH_LOCK_FCNTL) { | |
10173 | if (!in->fcntl_locks) | |
11fdf7f2 TL |
10174 | in->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
10175 | lock_state = in->fcntl_locks.get(); | |
7c673cae FG |
10176 | } else if (lock_type == CEPH_LOCK_FLOCK) { |
10177 | if (!in->flock_locks) | |
11fdf7f2 TL |
10178 | in->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
10179 | lock_state = in->flock_locks.get(); | |
7c673cae FG |
10180 | } else { |
10181 | ceph_abort(); | |
10182 | return -EINVAL; | |
10183 | } | |
10184 | _update_lock_state(fl, owner, lock_state); | |
10185 | ||
10186 | if (!removing) { | |
10187 | if (lock_type == CEPH_LOCK_FCNTL) { | |
10188 | if (!fh->fcntl_locks) | |
11fdf7f2 TL |
10189 | fh->fcntl_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FCNTL)); |
10190 | lock_state = fh->fcntl_locks.get(); | |
7c673cae FG |
10191 | } else { |
10192 | if (!fh->flock_locks) | |
11fdf7f2 TL |
10193 | fh->flock_locks.reset(new ceph_lock_state_t(cct, CEPH_LOCK_FLOCK)); |
10194 | lock_state = fh->flock_locks.get(); | |
7c673cae FG |
10195 | } |
10196 | _update_lock_state(fl, owner, lock_state); | |
10197 | } | |
10198 | } else | |
10199 | ceph_abort(); | |
10200 | } | |
10201 | return ret; | |
10202 | } | |
10203 | ||
10204 | int Client::_interrupt_filelock(MetaRequest *req) | |
10205 | { | |
31f18b77 FG |
10206 | // Set abort code, but do not kick. The abort code prevents the request |
10207 | // from being re-sent. | |
10208 | req->abort(-EINTR); | |
10209 | if (req->mds < 0) | |
10210 | return 0; // haven't sent the request | |
10211 | ||
7c673cae FG |
10212 | Inode *in = req->inode(); |
10213 | ||
10214 | int lock_type; | |
10215 | if (req->head.args.filelock_change.rule == CEPH_LOCK_FLOCK) | |
10216 | lock_type = CEPH_LOCK_FLOCK_INTR; | |
10217 | else if (req->head.args.filelock_change.rule == CEPH_LOCK_FCNTL) | |
10218 | lock_type = CEPH_LOCK_FCNTL_INTR; | |
10219 | else { | |
10220 | ceph_abort(); | |
10221 | return -EINVAL; | |
10222 | } | |
10223 | ||
10224 | MetaRequest *intr_req = new MetaRequest(CEPH_MDS_OP_SETFILELOCK); | |
10225 | filepath path; | |
10226 | in->make_nosnap_relative_path(path); | |
10227 | intr_req->set_filepath(path); | |
10228 | intr_req->set_inode(in); | |
10229 | intr_req->head.args.filelock_change = req->head.args.filelock_change; | |
10230 | intr_req->head.args.filelock_change.rule = lock_type; | |
10231 | intr_req->head.args.filelock_change.type = CEPH_LOCK_UNLOCK; | |
10232 | ||
10233 | UserPerm perms(req->get_uid(), req->get_gid()); | |
10234 | return make_request(intr_req, perms, NULL, NULL, -1); | |
10235 | } | |
10236 | ||
10237 | void Client::_encode_filelocks(Inode *in, bufferlist& bl) | |
10238 | { | |
10239 | if (!in->fcntl_locks && !in->flock_locks) | |
10240 | return; | |
10241 | ||
10242 | unsigned nr_fcntl_locks = in->fcntl_locks ? in->fcntl_locks->held_locks.size() : 0; | |
11fdf7f2 | 10243 | encode(nr_fcntl_locks, bl); |
7c673cae | 10244 | if (nr_fcntl_locks) { |
11fdf7f2 | 10245 | auto &lock_state = in->fcntl_locks; |
7c673cae FG |
10246 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); |
10247 | p != lock_state->held_locks.end(); | |
10248 | ++p) | |
11fdf7f2 | 10249 | encode(p->second, bl); |
7c673cae FG |
10250 | } |
10251 | ||
10252 | unsigned nr_flock_locks = in->flock_locks ? in->flock_locks->held_locks.size() : 0; | |
11fdf7f2 | 10253 | encode(nr_flock_locks, bl); |
7c673cae | 10254 | if (nr_flock_locks) { |
11fdf7f2 | 10255 | auto &lock_state = in->flock_locks; |
7c673cae FG |
10256 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); |
10257 | p != lock_state->held_locks.end(); | |
10258 | ++p) | |
11fdf7f2 | 10259 | encode(p->second, bl); |
7c673cae FG |
10260 | } |
10261 | ||
11fdf7f2 | 10262 | ldout(cct, 10) << __func__ << " ino " << in->ino << ", " << nr_fcntl_locks |
7c673cae FG |
10263 | << " fcntl locks, " << nr_flock_locks << " flock locks" << dendl; |
10264 | } | |
10265 | ||
10266 | void Client::_release_filelocks(Fh *fh) | |
10267 | { | |
10268 | if (!fh->fcntl_locks && !fh->flock_locks) | |
10269 | return; | |
10270 | ||
10271 | Inode *in = fh->inode.get(); | |
11fdf7f2 | 10272 | ldout(cct, 10) << __func__ << " " << fh << " ino " << in->ino << dendl; |
7c673cae FG |
10273 | |
10274 | list<pair<int, ceph_filelock> > to_release; | |
10275 | ||
10276 | if (fh->fcntl_locks) { | |
11fdf7f2 | 10277 | auto &lock_state = fh->fcntl_locks; |
7c673cae FG |
10278 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); |
10279 | p != lock_state->held_locks.end(); | |
10280 | ++p) | |
10281 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FCNTL, p->second)); | |
11fdf7f2 | 10282 | lock_state.reset(); |
7c673cae FG |
10283 | } |
10284 | if (fh->flock_locks) { | |
11fdf7f2 | 10285 | auto &lock_state = fh->flock_locks; |
7c673cae FG |
10286 | for(multimap<uint64_t, ceph_filelock>::iterator p = lock_state->held_locks.begin(); |
10287 | p != lock_state->held_locks.end(); | |
10288 | ++p) | |
10289 | to_release.push_back(pair<int, ceph_filelock>(CEPH_LOCK_FLOCK, p->second)); | |
11fdf7f2 | 10290 | lock_state.reset(); |
7c673cae FG |
10291 | } |
10292 | ||
10293 | if (to_release.empty()) | |
10294 | return; | |
10295 | ||
11fdf7f2 TL |
10296 | // mds has already released filelocks if session was closed. |
10297 | if (in->caps.empty()) | |
10298 | return; | |
10299 | ||
7c673cae FG |
10300 | struct flock fl; |
10301 | memset(&fl, 0, sizeof(fl)); | |
10302 | fl.l_whence = SEEK_SET; | |
10303 | fl.l_type = F_UNLCK; | |
10304 | ||
10305 | for (list<pair<int, ceph_filelock> >::iterator p = to_release.begin(); | |
10306 | p != to_release.end(); | |
10307 | ++p) { | |
10308 | fl.l_start = p->second.start; | |
10309 | fl.l_len = p->second.length; | |
10310 | fl.l_pid = p->second.pid; | |
10311 | _do_filelock(in, fh, p->first, CEPH_MDS_OP_SETFILELOCK, 0, &fl, | |
10312 | p->second.owner, true); | |
10313 | } | |
10314 | } | |
10315 | ||
10316 | void Client::_update_lock_state(struct flock *fl, uint64_t owner, | |
10317 | ceph_lock_state_t *lock_state) | |
10318 | { | |
10319 | int lock_cmd; | |
10320 | if (F_RDLCK == fl->l_type) | |
10321 | lock_cmd = CEPH_LOCK_SHARED; | |
10322 | else if (F_WRLCK == fl->l_type) | |
10323 | lock_cmd = CEPH_LOCK_EXCL; | |
10324 | else | |
10325 | lock_cmd = CEPH_LOCK_UNLOCK;; | |
10326 | ||
10327 | ceph_filelock filelock; | |
10328 | filelock.start = fl->l_start; | |
10329 | filelock.length = fl->l_len; | |
10330 | filelock.client = 0; | |
10331 | // see comment in _do_filelock() | |
10332 | filelock.owner = owner | (1ULL << 63); | |
10333 | filelock.pid = fl->l_pid; | |
10334 | filelock.type = lock_cmd; | |
10335 | ||
10336 | if (filelock.type == CEPH_LOCK_UNLOCK) { | |
10337 | list<ceph_filelock> activated_locks; | |
10338 | lock_state->remove_lock(filelock, activated_locks); | |
10339 | } else { | |
10340 | bool r = lock_state->add_lock(filelock, false, false, NULL); | |
11fdf7f2 | 10341 | ceph_assert(r); |
7c673cae FG |
10342 | } |
10343 | } | |
10344 | ||
10345 | int Client::_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
10346 | { | |
10347 | Inode *in = fh->inode.get(); | |
10348 | ldout(cct, 10) << "_getlk " << fh << " ino " << in->ino << dendl; | |
10349 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_GETFILELOCK, 0, fl, owner); | |
10350 | return ret; | |
10351 | } | |
10352 | ||
10353 | int Client::_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
10354 | { | |
10355 | Inode *in = fh->inode.get(); | |
10356 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << dendl; | |
10357 | int ret = _do_filelock(in, fh, CEPH_LOCK_FCNTL, CEPH_MDS_OP_SETFILELOCK, sleep, fl, owner); | |
10358 | ldout(cct, 10) << "_setlk " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
10359 | return ret; | |
10360 | } | |
10361 | ||
10362 | int Client::_flock(Fh *fh, int cmd, uint64_t owner) | |
10363 | { | |
10364 | Inode *in = fh->inode.get(); | |
10365 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << dendl; | |
10366 | ||
10367 | int sleep = !(cmd & LOCK_NB); | |
10368 | cmd &= ~LOCK_NB; | |
10369 | ||
10370 | int type; | |
10371 | switch (cmd) { | |
10372 | case LOCK_SH: | |
10373 | type = F_RDLCK; | |
10374 | break; | |
10375 | case LOCK_EX: | |
10376 | type = F_WRLCK; | |
10377 | break; | |
10378 | case LOCK_UN: | |
10379 | type = F_UNLCK; | |
10380 | break; | |
10381 | default: | |
10382 | return -EINVAL; | |
10383 | } | |
10384 | ||
10385 | struct flock fl; | |
10386 | memset(&fl, 0, sizeof(fl)); | |
10387 | fl.l_type = type; | |
10388 | fl.l_whence = SEEK_SET; | |
10389 | ||
10390 | int ret = _do_filelock(in, fh, CEPH_LOCK_FLOCK, CEPH_MDS_OP_SETFILELOCK, sleep, &fl, owner); | |
10391 | ldout(cct, 10) << "_flock " << fh << " ino " << in->ino << " result=" << ret << dendl; | |
10392 | return ret; | |
10393 | } | |
10394 | ||
10395 | int Client::ll_statfs(Inode *in, struct statvfs *stbuf, const UserPerm& perms) | |
10396 | { | |
10397 | /* Since the only thing this does is wrap a call to statfs, and | |
10398 | statfs takes a lock, it doesn't seem we have a need to split it | |
10399 | out. */ | |
10400 | return statfs(0, stbuf, perms); | |
10401 | } | |
10402 | ||
10403 | void Client::ll_register_callbacks(struct client_callback_args *args) | |
10404 | { | |
10405 | if (!args) | |
10406 | return; | |
11fdf7f2 TL |
10407 | std::lock_guard l(client_lock); |
10408 | ldout(cct, 10) << __func__ << " cb " << args->handle | |
7c673cae FG |
10409 | << " invalidate_ino_cb " << args->ino_cb |
10410 | << " invalidate_dentry_cb " << args->dentry_cb | |
7c673cae FG |
10411 | << " switch_interrupt_cb " << args->switch_intr_cb |
10412 | << " remount_cb " << args->remount_cb | |
10413 | << dendl; | |
10414 | callback_handle = args->handle; | |
10415 | if (args->ino_cb) { | |
10416 | ino_invalidate_cb = args->ino_cb; | |
10417 | async_ino_invalidator.start(); | |
10418 | } | |
10419 | if (args->dentry_cb) { | |
10420 | dentry_invalidate_cb = args->dentry_cb; | |
10421 | async_dentry_invalidator.start(); | |
10422 | } | |
10423 | if (args->switch_intr_cb) { | |
10424 | switch_interrupt_cb = args->switch_intr_cb; | |
10425 | interrupt_finisher.start(); | |
10426 | } | |
10427 | if (args->remount_cb) { | |
10428 | remount_cb = args->remount_cb; | |
10429 | remount_finisher.start(); | |
10430 | } | |
7c673cae FG |
10431 | umask_cb = args->umask_cb; |
10432 | } | |
10433 | ||
10434 | int Client::test_dentry_handling(bool can_invalidate) | |
10435 | { | |
10436 | int r = 0; | |
10437 | ||
10438 | can_invalidate_dentries = can_invalidate; | |
10439 | ||
10440 | if (can_invalidate_dentries) { | |
11fdf7f2 | 10441 | ceph_assert(dentry_invalidate_cb); |
7c673cae | 10442 | ldout(cct, 1) << "using dentry_invalidate_cb" << dendl; |
b32b8144 | 10443 | r = 0; |
11fdf7f2 TL |
10444 | } else { |
10445 | ceph_assert(remount_cb); | |
7c673cae | 10446 | ldout(cct, 1) << "using remount_cb" << dendl; |
91327a77 | 10447 | r = _do_remount(false); |
b32b8144 | 10448 | } |
11fdf7f2 | 10449 | |
7c673cae FG |
10450 | return r; |
10451 | } | |
10452 | ||
10453 | int Client::_sync_fs() | |
10454 | { | |
11fdf7f2 | 10455 | ldout(cct, 10) << __func__ << dendl; |
7c673cae FG |
10456 | |
10457 | // flush file data | |
11fdf7f2 TL |
10458 | std::unique_ptr<C_SaferCond> cond = nullptr; |
10459 | if (cct->_conf->client_oc) { | |
10460 | cond.reset(new C_SaferCond("Client::_sync_fs:lock")); | |
10461 | objectcacher->flush_all(cond.get()); | |
10462 | } | |
7c673cae FG |
10463 | |
10464 | // flush caps | |
10465 | flush_caps_sync(); | |
10466 | ceph_tid_t flush_tid = last_flush_tid; | |
10467 | ||
10468 | // wait for unsafe mds requests | |
10469 | wait_unsafe_requests(); | |
10470 | ||
10471 | wait_sync_caps(flush_tid); | |
10472 | ||
11fdf7f2 | 10473 | if (nullptr != cond) { |
9f95a23c | 10474 | client_lock.unlock(); |
11fdf7f2 TL |
10475 | ldout(cct, 15) << __func__ << " waiting on data to flush" << dendl; |
10476 | cond->wait(); | |
10477 | ldout(cct, 15) << __func__ << " flush finished" << dendl; | |
9f95a23c | 10478 | client_lock.lock(); |
7c673cae FG |
10479 | } |
10480 | ||
10481 | return 0; | |
10482 | } | |
10483 | ||
10484 | int Client::sync_fs() | |
10485 | { | |
11fdf7f2 | 10486 | std::lock_guard l(client_lock); |
181888fb FG |
10487 | |
10488 | if (unmounting) | |
10489 | return -ENOTCONN; | |
10490 | ||
7c673cae FG |
10491 | return _sync_fs(); |
10492 | } | |
10493 | ||
10494 | int64_t Client::drop_caches() | |
10495 | { | |
11fdf7f2 | 10496 | std::lock_guard l(client_lock); |
7c673cae FG |
10497 | return objectcacher->release_all(); |
10498 | } | |
10499 | ||
11fdf7f2 TL |
10500 | int Client::_lazyio(Fh *fh, int enable) |
10501 | { | |
10502 | Inode *in = fh->inode.get(); | |
10503 | ldout(cct, 20) << __func__ << " " << *in << " " << !!enable << dendl; | |
10504 | ||
10505 | if (!!(fh->mode & CEPH_FILE_MODE_LAZY) == !!enable) | |
10506 | return 0; | |
10507 | ||
10508 | int orig_mode = fh->mode; | |
10509 | if (enable) { | |
10510 | fh->mode |= CEPH_FILE_MODE_LAZY; | |
10511 | in->get_open_ref(fh->mode); | |
10512 | in->put_open_ref(orig_mode); | |
10513 | check_caps(in, CHECK_CAPS_NODELAY); | |
10514 | } else { | |
10515 | fh->mode &= ~CEPH_FILE_MODE_LAZY; | |
10516 | in->get_open_ref(fh->mode); | |
10517 | in->put_open_ref(orig_mode); | |
10518 | check_caps(in, 0); | |
10519 | } | |
10520 | ||
10521 | return 0; | |
10522 | } | |
10523 | ||
10524 | int Client::lazyio(int fd, int enable) | |
10525 | { | |
10526 | std::lock_guard l(client_lock); | |
10527 | Fh *f = get_filehandle(fd); | |
10528 | if (!f) | |
10529 | return -EBADF; | |
10530 | ||
10531 | return _lazyio(f, enable); | |
10532 | } | |
10533 | ||
10534 | int Client::ll_lazyio(Fh *fh, int enable) | |
10535 | { | |
10536 | std::lock_guard lock(client_lock); | |
10537 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << !!enable << dendl; | |
10538 | tout(cct) << __func__ << std::endl; | |
10539 | ||
10540 | return _lazyio(fh, enable); | |
10541 | } | |
7c673cae | 10542 | |
92f5a8d4 | 10543 | int Client::lazyio_propagate(int fd, loff_t offset, size_t count) |
7c673cae | 10544 | { |
11fdf7f2 | 10545 | std::lock_guard l(client_lock); |
92f5a8d4 | 10546 | ldout(cct, 3) << "op: client->lazyio_propagate(" << fd |
7c673cae FG |
10547 | << ", " << offset << ", " << count << ")" << dendl; |
10548 | ||
10549 | Fh *f = get_filehandle(fd); | |
10550 | if (!f) | |
10551 | return -EBADF; | |
10552 | ||
10553 | // for now | |
10554 | _fsync(f, true); | |
10555 | ||
10556 | return 0; | |
10557 | } | |
10558 | ||
10559 | int Client::lazyio_synchronize(int fd, loff_t offset, size_t count) | |
10560 | { | |
11fdf7f2 | 10561 | std::lock_guard l(client_lock); |
7c673cae FG |
10562 | ldout(cct, 3) << "op: client->lazyio_synchronize(" << fd |
10563 | << ", " << offset << ", " << count << ")" << dendl; | |
10564 | ||
10565 | Fh *f = get_filehandle(fd); | |
10566 | if (!f) | |
10567 | return -EBADF; | |
10568 | Inode *in = f->inode.get(); | |
10569 | ||
10570 | _fsync(f, true); | |
92f5a8d4 TL |
10571 | if (_release(in)) { |
10572 | int r =_getattr(in, CEPH_STAT_CAP_SIZE, f->actor_perms); | |
10573 | if (r < 0) | |
10574 | return r; | |
10575 | } | |
7c673cae FG |
10576 | return 0; |
10577 | } | |
10578 | ||
10579 | ||
10580 | // ============================= | |
10581 | // snaps | |
10582 | ||
10583 | int Client::mksnap(const char *relpath, const char *name, const UserPerm& perm) | |
10584 | { | |
11fdf7f2 | 10585 | std::lock_guard l(client_lock); |
181888fb FG |
10586 | |
10587 | if (unmounting) | |
10588 | return -ENOTCONN; | |
10589 | ||
7c673cae FG |
10590 | filepath path(relpath); |
10591 | InodeRef in; | |
10592 | int r = path_walk(path, &in, perm); | |
10593 | if (r < 0) | |
10594 | return r; | |
10595 | if (cct->_conf->client_permissions) { | |
10596 | r = may_create(in.get(), perm); | |
10597 | if (r < 0) | |
10598 | return r; | |
10599 | } | |
10600 | Inode *snapdir = open_snapdir(in.get()); | |
10601 | return _mkdir(snapdir, name, 0, perm); | |
10602 | } | |
181888fb | 10603 | |
7c673cae FG |
10604 | int Client::rmsnap(const char *relpath, const char *name, const UserPerm& perms) |
10605 | { | |
11fdf7f2 | 10606 | std::lock_guard l(client_lock); |
181888fb FG |
10607 | |
10608 | if (unmounting) | |
10609 | return -ENOTCONN; | |
10610 | ||
7c673cae FG |
10611 | filepath path(relpath); |
10612 | InodeRef in; | |
10613 | int r = path_walk(path, &in, perms); | |
10614 | if (r < 0) | |
10615 | return r; | |
10616 | if (cct->_conf->client_permissions) { | |
10617 | r = may_delete(in.get(), NULL, perms); | |
10618 | if (r < 0) | |
10619 | return r; | |
10620 | } | |
10621 | Inode *snapdir = open_snapdir(in.get()); | |
10622 | return _rmdir(snapdir, name, perms); | |
10623 | } | |
10624 | ||
10625 | // ============================= | |
10626 | // expose caps | |
10627 | ||
10628 | int Client::get_caps_issued(int fd) { | |
10629 | ||
11fdf7f2 | 10630 | std::lock_guard lock(client_lock); |
7c673cae | 10631 | |
181888fb FG |
10632 | if (unmounting) |
10633 | return -ENOTCONN; | |
10634 | ||
7c673cae FG |
10635 | Fh *f = get_filehandle(fd); |
10636 | if (!f) | |
10637 | return -EBADF; | |
10638 | ||
10639 | return f->inode->caps_issued(); | |
10640 | } | |
10641 | ||
10642 | int Client::get_caps_issued(const char *path, const UserPerm& perms) | |
10643 | { | |
11fdf7f2 | 10644 | std::lock_guard lock(client_lock); |
181888fb FG |
10645 | |
10646 | if (unmounting) | |
10647 | return -ENOTCONN; | |
10648 | ||
7c673cae FG |
10649 | filepath p(path); |
10650 | InodeRef in; | |
10651 | int r = path_walk(p, &in, perms, true); | |
10652 | if (r < 0) | |
10653 | return r; | |
10654 | return in->caps_issued(); | |
10655 | } | |
10656 | ||
10657 | // ========================================= | |
10658 | // low level | |
10659 | ||
10660 | Inode *Client::open_snapdir(Inode *diri) | |
10661 | { | |
10662 | Inode *in; | |
10663 | vinodeno_t vino(diri->ino, CEPH_SNAPDIR); | |
10664 | if (!inode_map.count(vino)) { | |
10665 | in = new Inode(this, vino, &diri->layout); | |
10666 | ||
10667 | in->ino = diri->ino; | |
10668 | in->snapid = CEPH_SNAPDIR; | |
10669 | in->mode = diri->mode; | |
10670 | in->uid = diri->uid; | |
10671 | in->gid = diri->gid; | |
494da23a | 10672 | in->nlink = 1; |
7c673cae FG |
10673 | in->mtime = diri->mtime; |
10674 | in->ctime = diri->ctime; | |
10675 | in->btime = diri->btime; | |
10676 | in->size = diri->size; | |
10677 | in->change_attr = diri->change_attr; | |
10678 | ||
10679 | in->dirfragtree.clear(); | |
10680 | in->snapdir_parent = diri; | |
10681 | diri->flags |= I_SNAPDIR_OPEN; | |
10682 | inode_map[vino] = in; | |
10683 | if (use_faked_inos()) | |
10684 | _assign_faked_ino(in); | |
10685 | ldout(cct, 10) << "open_snapdir created snapshot inode " << *in << dendl; | |
10686 | } else { | |
10687 | in = inode_map[vino]; | |
10688 | ldout(cct, 10) << "open_snapdir had snapshot inode " << *in << dendl; | |
10689 | } | |
10690 | return in; | |
10691 | } | |
10692 | ||
10693 | int Client::ll_lookup(Inode *parent, const char *name, struct stat *attr, | |
10694 | Inode **out, const UserPerm& perms) | |
10695 | { | |
11fdf7f2 | 10696 | std::lock_guard lock(client_lock); |
31f18b77 | 10697 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 TL |
10698 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
10699 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
10700 | tout(cct) << name << std::endl; |
10701 | ||
181888fb FG |
10702 | if (unmounting) |
10703 | return -ENOTCONN; | |
10704 | ||
7c673cae | 10705 | int r = 0; |
11fdf7f2 TL |
10706 | if (!fuse_default_permissions) { |
10707 | if (strcmp(name, ".") && strcmp(name, "..")) { | |
10708 | r = may_lookup(parent, perms); | |
10709 | if (r < 0) | |
10710 | return r; | |
10711 | } | |
7c673cae FG |
10712 | } |
10713 | ||
10714 | string dname(name); | |
10715 | InodeRef in; | |
10716 | ||
10717 | r = _lookup(parent, dname, CEPH_STAT_CAP_INODE_ALL, &in, perms); | |
10718 | if (r < 0) { | |
10719 | attr->st_ino = 0; | |
10720 | goto out; | |
10721 | } | |
10722 | ||
11fdf7f2 | 10723 | ceph_assert(in); |
7c673cae FG |
10724 | fill_stat(in, attr); |
10725 | _ll_get(in.get()); | |
10726 | ||
10727 | out: | |
11fdf7f2 | 10728 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
10729 | << " -> " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; |
10730 | tout(cct) << attr->st_ino << std::endl; | |
10731 | *out = in.get(); | |
10732 | return r; | |
10733 | } | |
10734 | ||
1adf2230 AA |
10735 | int Client::ll_lookup_inode( |
10736 | struct inodeno_t ino, | |
10737 | const UserPerm& perms, | |
10738 | Inode **inode) | |
10739 | { | |
81eedcae | 10740 | ceph_assert(inode != NULL); |
11fdf7f2 | 10741 | std::lock_guard lock(client_lock); |
1adf2230 AA |
10742 | ldout(cct, 3) << "ll_lookup_inode " << ino << dendl; |
10743 | ||
81eedcae TL |
10744 | if (unmounting) |
10745 | return -ENOTCONN; | |
10746 | ||
1adf2230 AA |
10747 | // Num1: get inode and *inode |
10748 | int r = _lookup_ino(ino, perms, inode); | |
81eedcae | 10749 | if (r) |
1adf2230 | 10750 | return r; |
81eedcae | 10751 | |
11fdf7f2 | 10752 | ceph_assert(*inode != NULL); |
1adf2230 | 10753 | |
81eedcae TL |
10754 | if (!(*inode)->dentries.empty()) { |
10755 | ldout(cct, 8) << __func__ << " dentry already present" << dendl; | |
10756 | return 0; | |
10757 | } | |
10758 | ||
10759 | if ((*inode)->is_root()) { | |
10760 | ldout(cct, 8) << "ino is root, no parent" << dendl; | |
10761 | return 0; | |
10762 | } | |
10763 | ||
1adf2230 AA |
10764 | // Num2: Request the parent inode, so that we can look up the name |
10765 | Inode *parent; | |
10766 | r = _lookup_parent(*inode, perms, &parent); | |
81eedcae | 10767 | if (r) { |
1adf2230 AA |
10768 | _ll_forget(*inode, 1); |
10769 | return r; | |
1adf2230 | 10770 | } |
81eedcae | 10771 | |
11fdf7f2 | 10772 | ceph_assert(parent != NULL); |
1adf2230 AA |
10773 | |
10774 | // Num3: Finally, get the name (dentry) of the requested inode | |
10775 | r = _lookup_name(*inode, parent, perms); | |
10776 | if (r) { | |
10777 | // Unexpected error | |
10778 | _ll_forget(parent, 1); | |
10779 | _ll_forget(*inode, 1); | |
10780 | return r; | |
10781 | } | |
10782 | ||
10783 | _ll_forget(parent, 1); | |
10784 | return 0; | |
10785 | } | |
10786 | ||
7c673cae FG |
10787 | int Client::ll_lookupx(Inode *parent, const char *name, Inode **out, |
10788 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
10789 | const UserPerm& perms) | |
10790 | { | |
11fdf7f2 | 10791 | std::lock_guard lock(client_lock); |
31f18b77 | 10792 | vinodeno_t vparent = _get_vino(parent); |
11fdf7f2 | 10793 | ldout(cct, 3) << __func__ << " " << vparent << " " << name << dendl; |
7c673cae FG |
10794 | tout(cct) << "ll_lookupx" << std::endl; |
10795 | tout(cct) << name << std::endl; | |
10796 | ||
181888fb FG |
10797 | if (unmounting) |
10798 | return -ENOTCONN; | |
10799 | ||
7c673cae | 10800 | int r = 0; |
11fdf7f2 | 10801 | if (!fuse_default_permissions) { |
7c673cae FG |
10802 | r = may_lookup(parent, perms); |
10803 | if (r < 0) | |
10804 | return r; | |
10805 | } | |
10806 | ||
10807 | string dname(name); | |
10808 | InodeRef in; | |
10809 | ||
10810 | unsigned mask = statx_to_mask(flags, want); | |
10811 | r = _lookup(parent, dname, mask, &in, perms); | |
10812 | if (r < 0) { | |
10813 | stx->stx_ino = 0; | |
10814 | stx->stx_mask = 0; | |
10815 | } else { | |
11fdf7f2 | 10816 | ceph_assert(in); |
7c673cae FG |
10817 | fill_statx(in, mask, stx); |
10818 | _ll_get(in.get()); | |
10819 | } | |
10820 | ||
11fdf7f2 | 10821 | ldout(cct, 3) << __func__ << " " << vparent << " " << name |
7c673cae FG |
10822 | << " -> " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; |
10823 | tout(cct) << stx->stx_ino << std::endl; | |
10824 | *out = in.get(); | |
10825 | return r; | |
10826 | } | |
10827 | ||
10828 | int Client::ll_walk(const char* name, Inode **out, struct ceph_statx *stx, | |
10829 | unsigned int want, unsigned int flags, const UserPerm& perms) | |
10830 | { | |
11fdf7f2 | 10831 | std::lock_guard lock(client_lock); |
181888fb FG |
10832 | |
10833 | if (unmounting) | |
10834 | return -ENOTCONN; | |
10835 | ||
7c673cae FG |
10836 | filepath fp(name, 0); |
10837 | InodeRef in; | |
10838 | int rc; | |
10839 | unsigned mask = statx_to_mask(flags, want); | |
10840 | ||
11fdf7f2 TL |
10841 | ldout(cct, 3) << __func__ << " " << name << dendl; |
10842 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
10843 | tout(cct) << name << std::endl; |
10844 | ||
10845 | rc = path_walk(fp, &in, perms, !(flags & AT_SYMLINK_NOFOLLOW), mask); | |
10846 | if (rc < 0) { | |
10847 | /* zero out mask, just in case... */ | |
10848 | stx->stx_mask = 0; | |
10849 | stx->stx_ino = 0; | |
10850 | *out = NULL; | |
10851 | return rc; | |
10852 | } else { | |
11fdf7f2 | 10853 | ceph_assert(in); |
7c673cae FG |
10854 | fill_statx(in, mask, stx); |
10855 | _ll_get(in.get()); | |
10856 | *out = in.get(); | |
10857 | return 0; | |
10858 | } | |
10859 | } | |
10860 | ||
10861 | void Client::_ll_get(Inode *in) | |
10862 | { | |
10863 | if (in->ll_ref == 0) { | |
10864 | in->get(); | |
11fdf7f2 TL |
10865 | if (in->is_dir() && !in->dentries.empty()) { |
10866 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
10867 | in->get_first_parent()->get(); // pin dentry |
10868 | } | |
11fdf7f2 TL |
10869 | if (in->snapid != CEPH_NOSNAP) |
10870 | ll_snap_ref[in->snapid]++; | |
7c673cae FG |
10871 | } |
10872 | in->ll_get(); | |
11fdf7f2 | 10873 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " -> " << in->ll_ref << dendl; |
7c673cae FG |
10874 | } |
10875 | ||
494da23a | 10876 | int Client::_ll_put(Inode *in, uint64_t num) |
7c673cae FG |
10877 | { |
10878 | in->ll_put(num); | |
11fdf7f2 | 10879 | ldout(cct, 20) << __func__ << " " << in << " " << in->ino << " " << num << " -> " << in->ll_ref << dendl; |
7c673cae | 10880 | if (in->ll_ref == 0) { |
11fdf7f2 TL |
10881 | if (in->is_dir() && !in->dentries.empty()) { |
10882 | ceph_assert(in->dentries.size() == 1); // dirs can't be hard-linked | |
7c673cae FG |
10883 | in->get_first_parent()->put(); // unpin dentry |
10884 | } | |
11fdf7f2 TL |
10885 | if (in->snapid != CEPH_NOSNAP) { |
10886 | auto p = ll_snap_ref.find(in->snapid); | |
10887 | ceph_assert(p != ll_snap_ref.end()); | |
10888 | ceph_assert(p->second > 0); | |
10889 | if (--p->second == 0) | |
10890 | ll_snap_ref.erase(p); | |
10891 | } | |
7c673cae FG |
10892 | put_inode(in); |
10893 | return 0; | |
10894 | } else { | |
10895 | return in->ll_ref; | |
10896 | } | |
10897 | } | |
10898 | ||
10899 | void Client::_ll_drop_pins() | |
10900 | { | |
11fdf7f2 | 10901 | ldout(cct, 10) << __func__ << dendl; |
1adf2230 | 10902 | std::set<InodeRef> to_be_put; //this set will be deconstructed item by item when exit |
7c673cae FG |
10903 | ceph::unordered_map<vinodeno_t, Inode*>::iterator next; |
10904 | for (ceph::unordered_map<vinodeno_t, Inode*>::iterator it = inode_map.begin(); | |
10905 | it != inode_map.end(); | |
10906 | it = next) { | |
10907 | Inode *in = it->second; | |
10908 | next = it; | |
10909 | ++next; | |
1adf2230 AA |
10910 | if (in->ll_ref){ |
10911 | to_be_put.insert(in); | |
7c673cae | 10912 | _ll_put(in, in->ll_ref); |
1adf2230 | 10913 | } |
7c673cae FG |
10914 | } |
10915 | } | |
10916 | ||
494da23a | 10917 | bool Client::_ll_forget(Inode *in, uint64_t count) |
7c673cae | 10918 | { |
11fdf7f2 | 10919 | inodeno_t ino = in->ino; |
7c673cae | 10920 | |
11fdf7f2 TL |
10921 | ldout(cct, 8) << __func__ << " " << ino << " " << count << dendl; |
10922 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
10923 | tout(cct) << ino.val << std::endl; |
10924 | tout(cct) << count << std::endl; | |
10925 | ||
181888fb FG |
10926 | // Ignore forget if we're no longer mounted |
10927 | if (unmounting) | |
10928 | return true; | |
10929 | ||
7c673cae FG |
10930 | if (ino == 1) return true; // ignore forget on root. |
10931 | ||
10932 | bool last = false; | |
10933 | if (in->ll_ref < count) { | |
10934 | ldout(cct, 1) << "WARNING: ll_forget on " << ino << " " << count | |
10935 | << ", which only has ll_ref=" << in->ll_ref << dendl; | |
10936 | _ll_put(in, in->ll_ref); | |
10937 | last = true; | |
10938 | } else { | |
10939 | if (_ll_put(in, count) == 0) | |
10940 | last = true; | |
10941 | } | |
10942 | ||
10943 | return last; | |
10944 | } | |
10945 | ||
494da23a | 10946 | bool Client::ll_forget(Inode *in, uint64_t count) |
1adf2230 | 10947 | { |
11fdf7f2 | 10948 | std::lock_guard lock(client_lock); |
1adf2230 AA |
10949 | return _ll_forget(in, count); |
10950 | } | |
10951 | ||
7c673cae FG |
10952 | bool Client::ll_put(Inode *in) |
10953 | { | |
10954 | /* ll_forget already takes the lock */ | |
10955 | return ll_forget(in, 1); | |
10956 | } | |
10957 | ||
11fdf7f2 TL |
10958 | int Client::ll_get_snap_ref(snapid_t snap) |
10959 | { | |
10960 | std::lock_guard lock(client_lock); | |
10961 | auto p = ll_snap_ref.find(snap); | |
10962 | if (p != ll_snap_ref.end()) | |
10963 | return p->second; | |
10964 | return 0; | |
10965 | } | |
10966 | ||
7c673cae FG |
10967 | snapid_t Client::ll_get_snapid(Inode *in) |
10968 | { | |
11fdf7f2 | 10969 | std::lock_guard lock(client_lock); |
7c673cae FG |
10970 | return in->snapid; |
10971 | } | |
10972 | ||
10973 | Inode *Client::ll_get_inode(ino_t ino) | |
10974 | { | |
11fdf7f2 | 10975 | std::lock_guard lock(client_lock); |
181888fb FG |
10976 | |
10977 | if (unmounting) | |
10978 | return NULL; | |
10979 | ||
7c673cae FG |
10980 | vinodeno_t vino = _map_faked_ino(ino); |
10981 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); | |
10982 | if (p == inode_map.end()) | |
10983 | return NULL; | |
10984 | Inode *in = p->second; | |
10985 | _ll_get(in); | |
10986 | return in; | |
10987 | } | |
10988 | ||
10989 | Inode *Client::ll_get_inode(vinodeno_t vino) | |
10990 | { | |
11fdf7f2 | 10991 | std::lock_guard lock(client_lock); |
181888fb FG |
10992 | |
10993 | if (unmounting) | |
10994 | return NULL; | |
10995 | ||
7c673cae FG |
10996 | unordered_map<vinodeno_t,Inode*>::iterator p = inode_map.find(vino); |
10997 | if (p == inode_map.end()) | |
10998 | return NULL; | |
10999 | Inode *in = p->second; | |
11000 | _ll_get(in); | |
11001 | return in; | |
11002 | } | |
11003 | ||
11004 | int Client::_ll_getattr(Inode *in, int caps, const UserPerm& perms) | |
11005 | { | |
11006 | vinodeno_t vino = _get_vino(in); | |
11007 | ||
11fdf7f2 TL |
11008 | ldout(cct, 8) << __func__ << " " << vino << dendl; |
11009 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11010 | tout(cct) << vino.ino.val << std::endl; |
11011 | ||
11012 | if (vino.snapid < CEPH_NOSNAP) | |
11013 | return 0; | |
11014 | else | |
11015 | return _getattr(in, caps, perms); | |
11016 | } | |
11017 | ||
11018 | int Client::ll_getattr(Inode *in, struct stat *attr, const UserPerm& perms) | |
11019 | { | |
11fdf7f2 | 11020 | std::lock_guard lock(client_lock); |
7c673cae | 11021 | |
181888fb FG |
11022 | if (unmounting) |
11023 | return -ENOTCONN; | |
11024 | ||
7c673cae FG |
11025 | int res = _ll_getattr(in, CEPH_STAT_CAP_INODE_ALL, perms); |
11026 | ||
11027 | if (res == 0) | |
11028 | fill_stat(in, attr); | |
11fdf7f2 | 11029 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11030 | return res; |
11031 | } | |
11032 | ||
11033 | int Client::ll_getattrx(Inode *in, struct ceph_statx *stx, unsigned int want, | |
11034 | unsigned int flags, const UserPerm& perms) | |
11035 | { | |
11fdf7f2 | 11036 | std::lock_guard lock(client_lock); |
7c673cae | 11037 | |
181888fb FG |
11038 | if (unmounting) |
11039 | return -ENOTCONN; | |
11040 | ||
7c673cae FG |
11041 | int res = 0; |
11042 | unsigned mask = statx_to_mask(flags, want); | |
11043 | ||
94b18763 | 11044 | if (mask && !in->caps_issued_mask(mask, true)) |
7c673cae FG |
11045 | res = _ll_getattr(in, mask, perms); |
11046 | ||
11047 | if (res == 0) | |
11048 | fill_statx(in, mask, stx); | |
11fdf7f2 | 11049 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11050 | return res; |
11051 | } | |
11052 | ||
11053 | int Client::_ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
11054 | const UserPerm& perms, InodeRef *inp) | |
11055 | { | |
11056 | vinodeno_t vino = _get_vino(in); | |
11057 | ||
11fdf7f2 | 11058 | ldout(cct, 8) << __func__ << " " << vino << " mask " << hex << mask << dec |
7c673cae | 11059 | << dendl; |
11fdf7f2 | 11060 | tout(cct) << __func__ << std::endl; |
7c673cae FG |
11061 | tout(cct) << vino.ino.val << std::endl; |
11062 | tout(cct) << stx->stx_mode << std::endl; | |
11063 | tout(cct) << stx->stx_uid << std::endl; | |
11064 | tout(cct) << stx->stx_gid << std::endl; | |
11065 | tout(cct) << stx->stx_size << std::endl; | |
11066 | tout(cct) << stx->stx_mtime << std::endl; | |
11067 | tout(cct) << stx->stx_atime << std::endl; | |
11068 | tout(cct) << stx->stx_btime << std::endl; | |
11069 | tout(cct) << mask << std::endl; | |
11070 | ||
11fdf7f2 | 11071 | if (!fuse_default_permissions) { |
7c673cae FG |
11072 | int res = may_setattr(in, stx, mask, perms); |
11073 | if (res < 0) | |
11074 | return res; | |
11075 | } | |
11076 | ||
11077 | mask &= ~(CEPH_SETATTR_MTIME_NOW | CEPH_SETATTR_ATIME_NOW); | |
11078 | ||
11079 | return __setattrx(in, stx, mask, perms, inp); | |
11080 | } | |
11081 | ||
11082 | int Client::ll_setattrx(Inode *in, struct ceph_statx *stx, int mask, | |
11083 | const UserPerm& perms) | |
11084 | { | |
11fdf7f2 | 11085 | std::lock_guard lock(client_lock); |
181888fb FG |
11086 | |
11087 | if (unmounting) | |
11088 | return -ENOTCONN; | |
11089 | ||
7c673cae FG |
11090 | InodeRef target(in); |
11091 | int res = _ll_setattrx(in, stx, mask, perms, &target); | |
11092 | if (res == 0) { | |
11fdf7f2 | 11093 | ceph_assert(in == target.get()); |
7c673cae FG |
11094 | fill_statx(in, in->caps_issued(), stx); |
11095 | } | |
11096 | ||
11fdf7f2 | 11097 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11098 | return res; |
11099 | } | |
11100 | ||
11101 | int Client::ll_setattr(Inode *in, struct stat *attr, int mask, | |
11102 | const UserPerm& perms) | |
11103 | { | |
11104 | struct ceph_statx stx; | |
11105 | stat_to_statx(attr, &stx); | |
11106 | ||
11fdf7f2 | 11107 | std::lock_guard lock(client_lock); |
181888fb FG |
11108 | |
11109 | if (unmounting) | |
11110 | return -ENOTCONN; | |
11111 | ||
7c673cae FG |
11112 | InodeRef target(in); |
11113 | int res = _ll_setattrx(in, &stx, mask, perms, &target); | |
11114 | if (res == 0) { | |
11fdf7f2 | 11115 | ceph_assert(in == target.get()); |
7c673cae FG |
11116 | fill_stat(in, attr); |
11117 | } | |
11118 | ||
11fdf7f2 | 11119 | ldout(cct, 3) << __func__ << " " << _get_vino(in) << " = " << res << dendl; |
7c673cae FG |
11120 | return res; |
11121 | } | |
11122 | ||
11123 | ||
11124 | // ---------- | |
11125 | // xattrs | |
11126 | ||
11127 | int Client::getxattr(const char *path, const char *name, void *value, size_t size, | |
11128 | const UserPerm& perms) | |
11129 | { | |
11fdf7f2 | 11130 | std::lock_guard lock(client_lock); |
181888fb FG |
11131 | |
11132 | if (unmounting) | |
11133 | return -ENOTCONN; | |
11134 | ||
7c673cae FG |
11135 | InodeRef in; |
11136 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
11137 | if (r < 0) | |
11138 | return r; | |
11139 | return _getxattr(in, name, value, size, perms); | |
11140 | } | |
11141 | ||
11142 | int Client::lgetxattr(const char *path, const char *name, void *value, size_t size, | |
11143 | const UserPerm& perms) | |
11144 | { | |
11fdf7f2 | 11145 | std::lock_guard lock(client_lock); |
181888fb FG |
11146 | |
11147 | if (unmounting) | |
11148 | return -ENOTCONN; | |
11149 | ||
7c673cae FG |
11150 | InodeRef in; |
11151 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
11152 | if (r < 0) | |
11153 | return r; | |
11154 | return _getxattr(in, name, value, size, perms); | |
11155 | } | |
11156 | ||
11157 | int Client::fgetxattr(int fd, const char *name, void *value, size_t size, | |
11158 | const UserPerm& perms) | |
11159 | { | |
11fdf7f2 | 11160 | std::lock_guard lock(client_lock); |
181888fb FG |
11161 | |
11162 | if (unmounting) | |
11163 | return -ENOTCONN; | |
11164 | ||
7c673cae FG |
11165 | Fh *f = get_filehandle(fd); |
11166 | if (!f) | |
11167 | return -EBADF; | |
11168 | return _getxattr(f->inode, name, value, size, perms); | |
11169 | } | |
11170 | ||
11171 | int Client::listxattr(const char *path, char *list, size_t size, | |
11172 | const UserPerm& perms) | |
11173 | { | |
11fdf7f2 | 11174 | std::lock_guard lock(client_lock); |
181888fb FG |
11175 | |
11176 | if (unmounting) | |
11177 | return -ENOTCONN; | |
11178 | ||
7c673cae FG |
11179 | InodeRef in; |
11180 | int r = Client::path_walk(path, &in, perms, true, CEPH_STAT_CAP_XATTR); | |
11181 | if (r < 0) | |
11182 | return r; | |
11183 | return Client::_listxattr(in.get(), list, size, perms); | |
11184 | } | |
11185 | ||
11186 | int Client::llistxattr(const char *path, char *list, size_t size, | |
11187 | const UserPerm& perms) | |
11188 | { | |
11fdf7f2 | 11189 | std::lock_guard lock(client_lock); |
181888fb FG |
11190 | |
11191 | if (unmounting) | |
11192 | return -ENOTCONN; | |
11193 | ||
7c673cae FG |
11194 | InodeRef in; |
11195 | int r = Client::path_walk(path, &in, perms, false, CEPH_STAT_CAP_XATTR); | |
11196 | if (r < 0) | |
11197 | return r; | |
11198 | return Client::_listxattr(in.get(), list, size, perms); | |
11199 | } | |
11200 | ||
11201 | int Client::flistxattr(int fd, char *list, size_t size, const UserPerm& perms) | |
11202 | { | |
11fdf7f2 | 11203 | std::lock_guard lock(client_lock); |
181888fb FG |
11204 | |
11205 | if (unmounting) | |
11206 | return -ENOTCONN; | |
11207 | ||
7c673cae FG |
11208 | Fh *f = get_filehandle(fd); |
11209 | if (!f) | |
11210 | return -EBADF; | |
11211 | return Client::_listxattr(f->inode.get(), list, size, perms); | |
11212 | } | |
11213 | ||
11214 | int Client::removexattr(const char *path, const char *name, | |
11215 | const UserPerm& perms) | |
11216 | { | |
11fdf7f2 | 11217 | std::lock_guard lock(client_lock); |
181888fb FG |
11218 | |
11219 | if (unmounting) | |
11220 | return -ENOTCONN; | |
11221 | ||
7c673cae FG |
11222 | InodeRef in; |
11223 | int r = Client::path_walk(path, &in, perms, true); | |
11224 | if (r < 0) | |
11225 | return r; | |
11226 | return _removexattr(in, name, perms); | |
11227 | } | |
11228 | ||
11229 | int Client::lremovexattr(const char *path, const char *name, | |
11230 | const UserPerm& perms) | |
11231 | { | |
11fdf7f2 | 11232 | std::lock_guard lock(client_lock); |
181888fb FG |
11233 | |
11234 | if (unmounting) | |
11235 | return -ENOTCONN; | |
11236 | ||
7c673cae FG |
11237 | InodeRef in; |
11238 | int r = Client::path_walk(path, &in, perms, false); | |
11239 | if (r < 0) | |
11240 | return r; | |
11241 | return _removexattr(in, name, perms); | |
11242 | } | |
11243 | ||
11244 | int Client::fremovexattr(int fd, const char *name, const UserPerm& perms) | |
11245 | { | |
11fdf7f2 | 11246 | std::lock_guard lock(client_lock); |
181888fb FG |
11247 | |
11248 | if (unmounting) | |
11249 | return -ENOTCONN; | |
11250 | ||
7c673cae FG |
11251 | Fh *f = get_filehandle(fd); |
11252 | if (!f) | |
11253 | return -EBADF; | |
11254 | return _removexattr(f->inode, name, perms); | |
11255 | } | |
11256 | ||
11257 | int Client::setxattr(const char *path, const char *name, const void *value, | |
11258 | size_t size, int flags, const UserPerm& perms) | |
11259 | { | |
11260 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
11261 | ||
11fdf7f2 | 11262 | std::lock_guard lock(client_lock); |
181888fb FG |
11263 | |
11264 | if (unmounting) | |
11265 | return -ENOTCONN; | |
11266 | ||
7c673cae FG |
11267 | InodeRef in; |
11268 | int r = Client::path_walk(path, &in, perms, true); | |
11269 | if (r < 0) | |
11270 | return r; | |
11271 | return _setxattr(in, name, value, size, flags, perms); | |
11272 | } | |
11273 | ||
11274 | int Client::lsetxattr(const char *path, const char *name, const void *value, | |
11275 | size_t size, int flags, const UserPerm& perms) | |
11276 | { | |
11277 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
11278 | ||
11fdf7f2 | 11279 | std::lock_guard lock(client_lock); |
181888fb FG |
11280 | |
11281 | if (unmounting) | |
11282 | return -ENOTCONN; | |
11283 | ||
7c673cae FG |
11284 | InodeRef in; |
11285 | int r = Client::path_walk(path, &in, perms, false); | |
11286 | if (r < 0) | |
11287 | return r; | |
11288 | return _setxattr(in, name, value, size, flags, perms); | |
11289 | } | |
11290 | ||
11291 | int Client::fsetxattr(int fd, const char *name, const void *value, size_t size, | |
11292 | int flags, const UserPerm& perms) | |
11293 | { | |
11294 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
11295 | ||
11fdf7f2 | 11296 | std::lock_guard lock(client_lock); |
181888fb FG |
11297 | |
11298 | if (unmounting) | |
11299 | return -ENOTCONN; | |
11300 | ||
7c673cae FG |
11301 | Fh *f = get_filehandle(fd); |
11302 | if (!f) | |
11303 | return -EBADF; | |
11304 | return _setxattr(f->inode, name, value, size, flags, perms); | |
11305 | } | |
11306 | ||
11307 | int Client::_getxattr(Inode *in, const char *name, void *value, size_t size, | |
11308 | const UserPerm& perms) | |
11309 | { | |
11310 | int r; | |
11311 | ||
11312 | const VXattr *vxattr = _match_vxattr(in, name); | |
11313 | if (vxattr) { | |
11314 | r = -ENODATA; | |
11315 | ||
11316 | // Do a force getattr to get the latest quota before returning | |
11317 | // a value to userspace. | |
28e407b8 AA |
11318 | int flags = 0; |
11319 | if (vxattr->flags & VXATTR_RSTAT) { | |
11320 | flags |= CEPH_STAT_RSTAT; | |
11321 | } | |
11322 | r = _getattr(in, flags, perms, true); | |
7c673cae FG |
11323 | if (r != 0) { |
11324 | // Error from getattr! | |
11325 | return r; | |
11326 | } | |
11327 | ||
11328 | // call pointer-to-member function | |
11329 | char buf[256]; | |
11330 | if (!(vxattr->exists_cb && !(this->*(vxattr->exists_cb))(in))) { | |
11331 | r = (this->*(vxattr->getxattr_cb))(in, buf, sizeof(buf)); | |
11332 | } else { | |
11333 | r = -ENODATA; | |
11334 | } | |
11335 | ||
11336 | if (size != 0) { | |
11337 | if (r > (int)size) { | |
11338 | r = -ERANGE; | |
11339 | } else if (r > 0) { | |
11340 | memcpy(value, buf, r); | |
11341 | } | |
11342 | } | |
11343 | goto out; | |
11344 | } | |
11345 | ||
11346 | if (acl_type == NO_ACL && !strncmp(name, "system.", 7)) { | |
11347 | r = -EOPNOTSUPP; | |
11348 | goto out; | |
11349 | } | |
11350 | ||
11351 | r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
11352 | if (r == 0) { | |
11353 | string n(name); | |
11354 | r = -ENODATA; | |
11355 | if (in->xattrs.count(n)) { | |
11356 | r = in->xattrs[n].length(); | |
11357 | if (r > 0 && size != 0) { | |
11358 | if (size >= (unsigned)r) | |
11359 | memcpy(value, in->xattrs[n].c_str(), r); | |
11360 | else | |
11361 | r = -ERANGE; | |
11362 | } | |
11363 | } | |
11364 | } | |
11365 | out: | |
1adf2230 | 11366 | ldout(cct, 8) << "_getxattr(" << in->ino << ", \"" << name << "\", " << size << ") = " << r << dendl; |
7c673cae FG |
11367 | return r; |
11368 | } | |
11369 | ||
11370 | int Client::_getxattr(InodeRef &in, const char *name, void *value, size_t size, | |
11371 | const UserPerm& perms) | |
11372 | { | |
11373 | if (cct->_conf->client_permissions) { | |
11374 | int r = xattr_permission(in.get(), name, MAY_READ, perms); | |
11375 | if (r < 0) | |
11376 | return r; | |
11377 | } | |
11378 | return _getxattr(in.get(), name, value, size, perms); | |
11379 | } | |
11380 | ||
11381 | int Client::ll_getxattr(Inode *in, const char *name, void *value, | |
11382 | size_t size, const UserPerm& perms) | |
11383 | { | |
11fdf7f2 | 11384 | std::lock_guard lock(client_lock); |
7c673cae | 11385 | |
181888fb FG |
11386 | if (unmounting) |
11387 | return -ENOTCONN; | |
11388 | ||
7c673cae FG |
11389 | vinodeno_t vino = _get_vino(in); |
11390 | ||
11fdf7f2 TL |
11391 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
11392 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11393 | tout(cct) << vino.ino.val << std::endl; |
11394 | tout(cct) << name << std::endl; | |
11395 | ||
11fdf7f2 | 11396 | if (!fuse_default_permissions) { |
7c673cae FG |
11397 | int r = xattr_permission(in, name, MAY_READ, perms); |
11398 | if (r < 0) | |
11399 | return r; | |
11400 | } | |
11401 | ||
11402 | return _getxattr(in, name, value, size, perms); | |
11403 | } | |
11404 | ||
11405 | int Client::_listxattr(Inode *in, char *name, size_t size, | |
11406 | const UserPerm& perms) | |
11407 | { | |
81eedcae | 11408 | bool len_only = (size == 0); |
7c673cae | 11409 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); |
81eedcae TL |
11410 | if (r != 0) { |
11411 | goto out; | |
11412 | } | |
7c673cae | 11413 | |
81eedcae TL |
11414 | r = 0; |
11415 | for (const auto& p : in->xattrs) { | |
11416 | size_t this_len = p.first.length() + 1; | |
11417 | r += this_len; | |
11418 | if (len_only) | |
11419 | continue; | |
7c673cae | 11420 | |
81eedcae TL |
11421 | if (this_len > size) { |
11422 | r = -ERANGE; | |
11423 | goto out; | |
11424 | } | |
11425 | ||
11426 | memcpy(name, p.first.c_str(), this_len); | |
11427 | name += this_len; | |
11428 | size -= this_len; | |
11429 | } | |
81eedcae | 11430 | out: |
11fdf7f2 | 11431 | ldout(cct, 8) << __func__ << "(" << in->ino << ", " << size << ") = " << r << dendl; |
7c673cae FG |
11432 | return r; |
11433 | } | |
11434 | ||
11435 | int Client::ll_listxattr(Inode *in, char *names, size_t size, | |
11436 | const UserPerm& perms) | |
11437 | { | |
11fdf7f2 | 11438 | std::lock_guard lock(client_lock); |
7c673cae | 11439 | |
181888fb FG |
11440 | if (unmounting) |
11441 | return -ENOTCONN; | |
11442 | ||
7c673cae FG |
11443 | vinodeno_t vino = _get_vino(in); |
11444 | ||
11fdf7f2 TL |
11445 | ldout(cct, 3) << __func__ << " " << vino << " size " << size << dendl; |
11446 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11447 | tout(cct) << vino.ino.val << std::endl; |
11448 | tout(cct) << size << std::endl; | |
11449 | ||
11450 | return _listxattr(in, names, size, perms); | |
11451 | } | |
11452 | ||
11453 | int Client::_do_setxattr(Inode *in, const char *name, const void *value, | |
11454 | size_t size, int flags, const UserPerm& perms) | |
11455 | { | |
11456 | ||
11457 | int xattr_flags = 0; | |
11458 | if (!value) | |
11459 | xattr_flags |= CEPH_XATTR_REMOVE; | |
11460 | if (flags & XATTR_CREATE) | |
11461 | xattr_flags |= CEPH_XATTR_CREATE; | |
11462 | if (flags & XATTR_REPLACE) | |
11463 | xattr_flags |= CEPH_XATTR_REPLACE; | |
11464 | ||
11465 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SETXATTR); | |
11466 | filepath path; | |
11467 | in->make_nosnap_relative_path(path); | |
11468 | req->set_filepath(path); | |
11469 | req->set_string2(name); | |
11470 | req->set_inode(in); | |
11471 | req->head.args.setxattr.flags = xattr_flags; | |
11472 | ||
11473 | bufferlist bl; | |
11fdf7f2 | 11474 | assert (value || size == 0); |
7c673cae FG |
11475 | bl.append((const char*)value, size); |
11476 | req->set_data(bl); | |
11477 | ||
11478 | int res = make_request(req, perms); | |
11479 | ||
11480 | trim_cache(); | |
11fdf7f2 | 11481 | ldout(cct, 3) << __func__ << "(" << in->ino << ", \"" << name << "\") = " << |
7c673cae FG |
11482 | res << dendl; |
11483 | return res; | |
11484 | } | |
11485 | ||
11486 | int Client::_setxattr(Inode *in, const char *name, const void *value, | |
11487 | size_t size, int flags, const UserPerm& perms) | |
11488 | { | |
11489 | if (in->snapid != CEPH_NOSNAP) { | |
11490 | return -EROFS; | |
11491 | } | |
11492 | ||
11493 | bool posix_acl_xattr = false; | |
11494 | if (acl_type == POSIX_ACL) | |
11495 | posix_acl_xattr = !strncmp(name, "system.", 7); | |
11496 | ||
11497 | if (strncmp(name, "user.", 5) && | |
11498 | strncmp(name, "security.", 9) && | |
11499 | strncmp(name, "trusted.", 8) && | |
11500 | strncmp(name, "ceph.", 5) && | |
11501 | !posix_acl_xattr) | |
11502 | return -EOPNOTSUPP; | |
11503 | ||
11fdf7f2 TL |
11504 | bool check_realm = false; |
11505 | ||
7c673cae FG |
11506 | if (posix_acl_xattr) { |
11507 | if (!strcmp(name, ACL_EA_ACCESS)) { | |
11508 | mode_t new_mode = in->mode; | |
11509 | if (value) { | |
11510 | int ret = posix_acl_equiv_mode(value, size, &new_mode); | |
11511 | if (ret < 0) | |
11512 | return ret; | |
11513 | if (ret == 0) { | |
11514 | value = NULL; | |
11515 | size = 0; | |
11516 | } | |
11517 | if (new_mode != in->mode) { | |
11518 | struct ceph_statx stx; | |
11519 | stx.stx_mode = new_mode; | |
11520 | ret = _do_setattr(in, &stx, CEPH_SETATTR_MODE, perms, NULL); | |
11521 | if (ret < 0) | |
11522 | return ret; | |
11523 | } | |
11524 | } | |
11525 | } else if (!strcmp(name, ACL_EA_DEFAULT)) { | |
11526 | if (value) { | |
11527 | if (!S_ISDIR(in->mode)) | |
11528 | return -EACCES; | |
11529 | int ret = posix_acl_check(value, size); | |
11530 | if (ret < 0) | |
11531 | return -EINVAL; | |
11532 | if (ret == 0) { | |
11533 | value = NULL; | |
11534 | size = 0; | |
11535 | } | |
11536 | } | |
11537 | } else { | |
11538 | return -EOPNOTSUPP; | |
11539 | } | |
11540 | } else { | |
11541 | const VXattr *vxattr = _match_vxattr(in, name); | |
11fdf7f2 TL |
11542 | if (vxattr) { |
11543 | if (vxattr->readonly) | |
11544 | return -EOPNOTSUPP; | |
11545 | if (vxattr->name.compare(0, 10, "ceph.quota") == 0 && value) | |
11546 | check_realm = true; | |
11547 | } | |
7c673cae FG |
11548 | } |
11549 | ||
11fdf7f2 TL |
11550 | int ret = _do_setxattr(in, name, value, size, flags, perms); |
11551 | if (ret >= 0 && check_realm) { | |
11552 | // check if snaprealm was created for quota inode | |
11553 | if (in->quota.is_enable() && | |
11554 | !(in->snaprealm && in->snaprealm->ino == in->ino)) | |
11555 | ret = -EOPNOTSUPP; | |
11556 | } | |
11557 | ||
11558 | return ret; | |
7c673cae FG |
11559 | } |
11560 | ||
11561 | int Client::_setxattr(InodeRef &in, const char *name, const void *value, | |
11562 | size_t size, int flags, const UserPerm& perms) | |
11563 | { | |
11564 | if (cct->_conf->client_permissions) { | |
11565 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
11566 | if (r < 0) | |
11567 | return r; | |
11568 | } | |
11569 | return _setxattr(in.get(), name, value, size, flags, perms); | |
11570 | } | |
11571 | ||
11572 | int Client::_setxattr_check_data_pool(string& name, string& value, const OSDMap *osdmap) | |
11573 | { | |
11574 | string tmp; | |
11575 | if (name == "layout") { | |
11576 | string::iterator begin = value.begin(); | |
11577 | string::iterator end = value.end(); | |
11578 | keys_and_values<string::iterator> p; // create instance of parser | |
11579 | std::map<string, string> m; // map to receive results | |
11580 | if (!qi::parse(begin, end, p, m)) { // returns true if successful | |
11581 | return -EINVAL; | |
11582 | } | |
11583 | if (begin != end) | |
11584 | return -EINVAL; | |
11585 | for (map<string,string>::iterator q = m.begin(); q != m.end(); ++q) { | |
11586 | if (q->first == "pool") { | |
11587 | tmp = q->second; | |
11588 | break; | |
11589 | } | |
11590 | } | |
11591 | } else if (name == "layout.pool") { | |
11592 | tmp = value; | |
11593 | } | |
11594 | ||
11595 | if (tmp.length()) { | |
11596 | int64_t pool; | |
11597 | try { | |
11598 | pool = boost::lexical_cast<unsigned>(tmp); | |
11599 | if (!osdmap->have_pg_pool(pool)) | |
11600 | return -ENOENT; | |
11601 | } catch (boost::bad_lexical_cast const&) { | |
11602 | pool = osdmap->lookup_pg_pool_name(tmp); | |
11603 | if (pool < 0) { | |
11604 | return -ENOENT; | |
11605 | } | |
11606 | } | |
11607 | } | |
11608 | ||
11609 | return 0; | |
11610 | } | |
11611 | ||
11612 | void Client::_setxattr_maybe_wait_for_osdmap(const char *name, const void *value, size_t size) | |
11613 | { | |
11614 | // For setting pool of layout, MetaRequest need osdmap epoch. | |
11615 | // There is a race which create a new data pool but client and mds both don't have. | |
11616 | // Make client got the latest osdmap which make mds quickly judge whether get newer osdmap. | |
11617 | if (strcmp(name, "ceph.file.layout.pool") == 0 || strcmp(name, "ceph.dir.layout.pool") == 0 || | |
11618 | strcmp(name, "ceph.file.layout") == 0 || strcmp(name, "ceph.dir.layout") == 0) { | |
11619 | string rest(strstr(name, "layout")); | |
11620 | string v((const char*)value, size); | |
11621 | int r = objecter->with_osdmap([&](const OSDMap& o) { | |
11622 | return _setxattr_check_data_pool(rest, v, &o); | |
11623 | }); | |
11624 | ||
11625 | if (r == -ENOENT) { | |
11626 | C_SaferCond ctx; | |
11627 | objecter->wait_for_latest_osdmap(&ctx); | |
11628 | ctx.wait(); | |
11629 | } | |
11630 | } | |
11631 | } | |
11632 | ||
11633 | int Client::ll_setxattr(Inode *in, const char *name, const void *value, | |
11634 | size_t size, int flags, const UserPerm& perms) | |
11635 | { | |
11636 | _setxattr_maybe_wait_for_osdmap(name, value, size); | |
11637 | ||
11fdf7f2 | 11638 | std::lock_guard lock(client_lock); |
7c673cae | 11639 | |
181888fb FG |
11640 | if (unmounting) |
11641 | return -ENOTCONN; | |
11642 | ||
7c673cae FG |
11643 | vinodeno_t vino = _get_vino(in); |
11644 | ||
11fdf7f2 TL |
11645 | ldout(cct, 3) << __func__ << " " << vino << " " << name << " size " << size << dendl; |
11646 | tout(cct) << __func__ << std::endl; | |
7c673cae FG |
11647 | tout(cct) << vino.ino.val << std::endl; |
11648 | tout(cct) << name << std::endl; | |
11649 | ||
11fdf7f2 | 11650 | if (!fuse_default_permissions) { |
7c673cae FG |
11651 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
11652 | if (r < 0) | |
11653 | return r; | |
11654 | } | |
11655 | return _setxattr(in, name, value, size, flags, perms); | |
11656 | } | |
11657 | ||
11658 | int Client::_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
11659 | { | |
11660 | if (in->snapid != CEPH_NOSNAP) { | |
11661 | return -EROFS; | |
11662 | } | |
11663 | ||
11664 | // same xattrs supported by kernel client | |
11665 | if (strncmp(name, "user.", 5) && | |
11666 | strncmp(name, "system.", 7) && | |
11667 | strncmp(name, "security.", 9) && | |
11668 | strncmp(name, "trusted.", 8) && | |
11669 | strncmp(name, "ceph.", 5)) | |
11670 | return -EOPNOTSUPP; | |
11671 | ||
11672 | const VXattr *vxattr = _match_vxattr(in, name); | |
11673 | if (vxattr && vxattr->readonly) | |
11674 | return -EOPNOTSUPP; | |
11675 | ||
11676 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_RMXATTR); | |
11677 | filepath path; | |
11678 | in->make_nosnap_relative_path(path); | |
11679 | req->set_filepath(path); | |
11680 | req->set_filepath2(name); | |
11681 | req->set_inode(in); | |
11682 | ||
11683 | int res = make_request(req, perms); | |
11684 | ||
11685 | trim_cache(); | |
1adf2230 | 11686 | ldout(cct, 8) << "_removexattr(" << in->ino << ", \"" << name << "\") = " << res << dendl; |
7c673cae FG |
11687 | return res; |
11688 | } | |
11689 | ||
11690 | int Client::_removexattr(InodeRef &in, const char *name, const UserPerm& perms) | |
11691 | { | |
11692 | if (cct->_conf->client_permissions) { | |
11693 | int r = xattr_permission(in.get(), name, MAY_WRITE, perms); | |
11694 | if (r < 0) | |
11695 | return r; | |
11696 | } | |
11697 | return _removexattr(in.get(), name, perms); | |
11698 | } | |
11699 | ||
11700 | int Client::ll_removexattr(Inode *in, const char *name, const UserPerm& perms) | |
11701 | { | |
11fdf7f2 | 11702 | std::lock_guard lock(client_lock); |
7c673cae | 11703 | |
181888fb FG |
11704 | if (unmounting) |
11705 | return -ENOTCONN; | |
11706 | ||
7c673cae FG |
11707 | vinodeno_t vino = _get_vino(in); |
11708 | ||
11709 | ldout(cct, 3) << "ll_removexattr " << vino << " " << name << dendl; | |
11710 | tout(cct) << "ll_removexattr" << std::endl; | |
11711 | tout(cct) << vino.ino.val << std::endl; | |
11712 | tout(cct) << name << std::endl; | |
11713 | ||
11fdf7f2 | 11714 | if (!fuse_default_permissions) { |
7c673cae FG |
11715 | int r = xattr_permission(in, name, MAY_WRITE, perms); |
11716 | if (r < 0) | |
11717 | return r; | |
11718 | } | |
11719 | ||
11720 | return _removexattr(in, name, perms); | |
11721 | } | |
11722 | ||
11723 | bool Client::_vxattrcb_quota_exists(Inode *in) | |
11724 | { | |
11fdf7f2 TL |
11725 | return in->quota.is_enable() && |
11726 | in->snaprealm && in->snaprealm->ino == in->ino; | |
7c673cae FG |
11727 | } |
11728 | size_t Client::_vxattrcb_quota(Inode *in, char *val, size_t size) | |
11729 | { | |
11730 | return snprintf(val, size, | |
11731 | "max_bytes=%lld max_files=%lld", | |
11732 | (long long int)in->quota.max_bytes, | |
11733 | (long long int)in->quota.max_files); | |
11734 | } | |
11735 | size_t Client::_vxattrcb_quota_max_bytes(Inode *in, char *val, size_t size) | |
11736 | { | |
11737 | return snprintf(val, size, "%lld", (long long int)in->quota.max_bytes); | |
11738 | } | |
11739 | size_t Client::_vxattrcb_quota_max_files(Inode *in, char *val, size_t size) | |
11740 | { | |
11741 | return snprintf(val, size, "%lld", (long long int)in->quota.max_files); | |
11742 | } | |
11743 | ||
11744 | bool Client::_vxattrcb_layout_exists(Inode *in) | |
11745 | { | |
11746 | return in->layout != file_layout_t(); | |
11747 | } | |
11748 | size_t Client::_vxattrcb_layout(Inode *in, char *val, size_t size) | |
11749 | { | |
11750 | int r = snprintf(val, size, | |
11fdf7f2 | 11751 | "stripe_unit=%llu stripe_count=%llu object_size=%llu pool=", |
7c673cae FG |
11752 | (unsigned long long)in->layout.stripe_unit, |
11753 | (unsigned long long)in->layout.stripe_count, | |
11754 | (unsigned long long)in->layout.object_size); | |
11755 | objecter->with_osdmap([&](const OSDMap& o) { | |
11756 | if (o.have_pg_pool(in->layout.pool_id)) | |
11757 | r += snprintf(val + r, size - r, "%s", | |
11758 | o.get_pool_name(in->layout.pool_id).c_str()); | |
11759 | else | |
11760 | r += snprintf(val + r, size - r, "%" PRIu64, | |
11761 | (uint64_t)in->layout.pool_id); | |
11762 | }); | |
11763 | if (in->layout.pool_ns.length()) | |
11764 | r += snprintf(val + r, size - r, " pool_namespace=%s", | |
11765 | in->layout.pool_ns.c_str()); | |
11766 | return r; | |
11767 | } | |
11768 | size_t Client::_vxattrcb_layout_stripe_unit(Inode *in, char *val, size_t size) | |
11769 | { | |
11fdf7f2 | 11770 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_unit); |
7c673cae FG |
11771 | } |
11772 | size_t Client::_vxattrcb_layout_stripe_count(Inode *in, char *val, size_t size) | |
11773 | { | |
11fdf7f2 | 11774 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.stripe_count); |
7c673cae FG |
11775 | } |
11776 | size_t Client::_vxattrcb_layout_object_size(Inode *in, char *val, size_t size) | |
11777 | { | |
11fdf7f2 | 11778 | return snprintf(val, size, "%llu", (unsigned long long)in->layout.object_size); |
7c673cae FG |
11779 | } |
11780 | size_t Client::_vxattrcb_layout_pool(Inode *in, char *val, size_t size) | |
11781 | { | |
11782 | size_t r; | |
11783 | objecter->with_osdmap([&](const OSDMap& o) { | |
11784 | if (o.have_pg_pool(in->layout.pool_id)) | |
11785 | r = snprintf(val, size, "%s", o.get_pool_name( | |
11786 | in->layout.pool_id).c_str()); | |
11787 | else | |
11788 | r = snprintf(val, size, "%" PRIu64, (uint64_t)in->layout.pool_id); | |
11789 | }); | |
11790 | return r; | |
11791 | } | |
11792 | size_t Client::_vxattrcb_layout_pool_namespace(Inode *in, char *val, size_t size) | |
11793 | { | |
11794 | return snprintf(val, size, "%s", in->layout.pool_ns.c_str()); | |
11795 | } | |
11796 | size_t Client::_vxattrcb_dir_entries(Inode *in, char *val, size_t size) | |
11797 | { | |
11fdf7f2 | 11798 | return snprintf(val, size, "%llu", (unsigned long long)(in->dirstat.nfiles + in->dirstat.nsubdirs)); |
7c673cae FG |
11799 | } |
11800 | size_t Client::_vxattrcb_dir_files(Inode *in, char *val, size_t size) | |
11801 | { | |
11fdf7f2 | 11802 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nfiles); |
7c673cae FG |
11803 | } |
11804 | size_t Client::_vxattrcb_dir_subdirs(Inode *in, char *val, size_t size) | |
11805 | { | |
11fdf7f2 | 11806 | return snprintf(val, size, "%llu", (unsigned long long)in->dirstat.nsubdirs); |
7c673cae FG |
11807 | } |
11808 | size_t Client::_vxattrcb_dir_rentries(Inode *in, char *val, size_t size) | |
11809 | { | |
11fdf7f2 | 11810 | return snprintf(val, size, "%llu", (unsigned long long)(in->rstat.rfiles + in->rstat.rsubdirs)); |
7c673cae FG |
11811 | } |
11812 | size_t Client::_vxattrcb_dir_rfiles(Inode *in, char *val, size_t size) | |
11813 | { | |
11fdf7f2 | 11814 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rfiles); |
7c673cae FG |
11815 | } |
11816 | size_t Client::_vxattrcb_dir_rsubdirs(Inode *in, char *val, size_t size) | |
11817 | { | |
11fdf7f2 | 11818 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rsubdirs); |
7c673cae FG |
11819 | } |
11820 | size_t Client::_vxattrcb_dir_rbytes(Inode *in, char *val, size_t size) | |
11821 | { | |
11fdf7f2 | 11822 | return snprintf(val, size, "%llu", (unsigned long long)in->rstat.rbytes); |
7c673cae FG |
11823 | } |
11824 | size_t Client::_vxattrcb_dir_rctime(Inode *in, char *val, size_t size) | |
11825 | { | |
81eedcae | 11826 | return snprintf(val, size, "%ld.%09ld", (long)in->rstat.rctime.sec(), |
7c673cae FG |
11827 | (long)in->rstat.rctime.nsec()); |
11828 | } | |
11fdf7f2 TL |
11829 | bool Client::_vxattrcb_dir_pin_exists(Inode *in) |
11830 | { | |
11831 | return in->dir_pin != -ENODATA; | |
11832 | } | |
11833 | size_t Client::_vxattrcb_dir_pin(Inode *in, char *val, size_t size) | |
11834 | { | |
11835 | return snprintf(val, size, "%ld", (long)in->dir_pin); | |
11836 | } | |
7c673cae | 11837 | |
81eedcae TL |
11838 | bool Client::_vxattrcb_snap_btime_exists(Inode *in) |
11839 | { | |
11840 | return !in->snap_btime.is_zero(); | |
11841 | } | |
11842 | ||
11843 | size_t Client::_vxattrcb_snap_btime(Inode *in, char *val, size_t size) | |
11844 | { | |
11845 | return snprintf(val, size, "%llu.%09lu", | |
11846 | (long long unsigned)in->snap_btime.sec(), | |
11847 | (long unsigned)in->snap_btime.nsec()); | |
11848 | } | |
11849 | ||
7c673cae FG |
11850 | #define CEPH_XATTR_NAME(_type, _name) "ceph." #_type "." #_name |
11851 | #define CEPH_XATTR_NAME2(_type, _name, _name2) "ceph." #_type "." #_name "." #_name2 | |
11852 | ||
11853 | #define XATTR_NAME_CEPH(_type, _name) \ | |
11854 | { \ | |
11855 | name: CEPH_XATTR_NAME(_type, _name), \ | |
11856 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
11857 | readonly: true, \ | |
7c673cae | 11858 | exists_cb: NULL, \ |
28e407b8 AA |
11859 | flags: 0, \ |
11860 | } | |
11861 | #define XATTR_NAME_CEPH2(_type, _name, _flags) \ | |
11862 | { \ | |
11863 | name: CEPH_XATTR_NAME(_type, _name), \ | |
11864 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
11865 | readonly: true, \ | |
28e407b8 AA |
11866 | exists_cb: NULL, \ |
11867 | flags: _flags, \ | |
7c673cae FG |
11868 | } |
11869 | #define XATTR_LAYOUT_FIELD(_type, _name, _field) \ | |
11870 | { \ | |
11871 | name: CEPH_XATTR_NAME2(_type, _name, _field), \ | |
11872 | getxattr_cb: &Client::_vxattrcb_ ## _name ## _ ## _field, \ | |
11873 | readonly: false, \ | |
7c673cae | 11874 | exists_cb: &Client::_vxattrcb_layout_exists, \ |
28e407b8 | 11875 | flags: 0, \ |
7c673cae FG |
11876 | } |
11877 | #define XATTR_QUOTA_FIELD(_type, _name) \ | |
11878 | { \ | |
11879 | name: CEPH_XATTR_NAME(_type, _name), \ | |
11880 | getxattr_cb: &Client::_vxattrcb_ ## _type ## _ ## _name, \ | |
11881 | readonly: false, \ | |
7c673cae | 11882 | exists_cb: &Client::_vxattrcb_quota_exists, \ |
28e407b8 | 11883 | flags: 0, \ |
7c673cae FG |
11884 | } |
11885 | ||
11886 | const Client::VXattr Client::_dir_vxattrs[] = { | |
11887 | { | |
11888 | name: "ceph.dir.layout", | |
11889 | getxattr_cb: &Client::_vxattrcb_layout, | |
11890 | readonly: false, | |
7c673cae | 11891 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 11892 | flags: 0, |
7c673cae FG |
11893 | }, |
11894 | XATTR_LAYOUT_FIELD(dir, layout, stripe_unit), | |
11895 | XATTR_LAYOUT_FIELD(dir, layout, stripe_count), | |
11896 | XATTR_LAYOUT_FIELD(dir, layout, object_size), | |
11897 | XATTR_LAYOUT_FIELD(dir, layout, pool), | |
11898 | XATTR_LAYOUT_FIELD(dir, layout, pool_namespace), | |
11899 | XATTR_NAME_CEPH(dir, entries), | |
11900 | XATTR_NAME_CEPH(dir, files), | |
11901 | XATTR_NAME_CEPH(dir, subdirs), | |
28e407b8 AA |
11902 | XATTR_NAME_CEPH2(dir, rentries, VXATTR_RSTAT), |
11903 | XATTR_NAME_CEPH2(dir, rfiles, VXATTR_RSTAT), | |
11904 | XATTR_NAME_CEPH2(dir, rsubdirs, VXATTR_RSTAT), | |
11905 | XATTR_NAME_CEPH2(dir, rbytes, VXATTR_RSTAT), | |
11906 | XATTR_NAME_CEPH2(dir, rctime, VXATTR_RSTAT), | |
7c673cae FG |
11907 | { |
11908 | name: "ceph.quota", | |
11909 | getxattr_cb: &Client::_vxattrcb_quota, | |
11910 | readonly: false, | |
7c673cae | 11911 | exists_cb: &Client::_vxattrcb_quota_exists, |
28e407b8 | 11912 | flags: 0, |
7c673cae FG |
11913 | }, |
11914 | XATTR_QUOTA_FIELD(quota, max_bytes), | |
11915 | XATTR_QUOTA_FIELD(quota, max_files), | |
11fdf7f2 TL |
11916 | { |
11917 | name: "ceph.dir.pin", | |
11918 | getxattr_cb: &Client::_vxattrcb_dir_pin, | |
11919 | readonly: false, | |
11fdf7f2 TL |
11920 | exists_cb: &Client::_vxattrcb_dir_pin_exists, |
11921 | flags: 0, | |
11922 | }, | |
81eedcae TL |
11923 | { |
11924 | name: "ceph.snap.btime", | |
11925 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
11926 | readonly: true, | |
81eedcae TL |
11927 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
11928 | flags: 0, | |
11929 | }, | |
7c673cae FG |
11930 | { name: "" } /* Required table terminator */ |
11931 | }; | |
11932 | ||
11933 | const Client::VXattr Client::_file_vxattrs[] = { | |
11934 | { | |
11935 | name: "ceph.file.layout", | |
11936 | getxattr_cb: &Client::_vxattrcb_layout, | |
11937 | readonly: false, | |
7c673cae | 11938 | exists_cb: &Client::_vxattrcb_layout_exists, |
28e407b8 | 11939 | flags: 0, |
7c673cae FG |
11940 | }, |
11941 | XATTR_LAYOUT_FIELD(file, layout, stripe_unit), | |
11942 | XATTR_LAYOUT_FIELD(file, layout, stripe_count), | |
11943 | XATTR_LAYOUT_FIELD(file, layout, object_size), | |
11944 | XATTR_LAYOUT_FIELD(file, layout, pool), | |
11945 | XATTR_LAYOUT_FIELD(file, layout, pool_namespace), | |
81eedcae TL |
11946 | { |
11947 | name: "ceph.snap.btime", | |
11948 | getxattr_cb: &Client::_vxattrcb_snap_btime, | |
11949 | readonly: true, | |
81eedcae TL |
11950 | exists_cb: &Client::_vxattrcb_snap_btime_exists, |
11951 | flags: 0, | |
11952 | }, | |
7c673cae FG |
11953 | { name: "" } /* Required table terminator */ |
11954 | }; | |
11955 | ||
11956 | const Client::VXattr *Client::_get_vxattrs(Inode *in) | |
11957 | { | |
11958 | if (in->is_dir()) | |
11959 | return _dir_vxattrs; | |
11960 | else if (in->is_file()) | |
11961 | return _file_vxattrs; | |
11962 | return NULL; | |
11963 | } | |
11964 | ||
11965 | const Client::VXattr *Client::_match_vxattr(Inode *in, const char *name) | |
11966 | { | |
11967 | if (strncmp(name, "ceph.", 5) == 0) { | |
11968 | const VXattr *vxattr = _get_vxattrs(in); | |
11969 | if (vxattr) { | |
11970 | while (!vxattr->name.empty()) { | |
11971 | if (vxattr->name == name) | |
11972 | return vxattr; | |
11973 | vxattr++; | |
11974 | } | |
11975 | } | |
11976 | } | |
11977 | return NULL; | |
11978 | } | |
11979 | ||
7c673cae FG |
11980 | int Client::ll_readlink(Inode *in, char *buf, size_t buflen, const UserPerm& perms) |
11981 | { | |
11fdf7f2 | 11982 | std::lock_guard lock(client_lock); |
7c673cae | 11983 | |
181888fb FG |
11984 | if (unmounting) |
11985 | return -ENOTCONN; | |
11986 | ||
7c673cae FG |
11987 | vinodeno_t vino = _get_vino(in); |
11988 | ||
11989 | ldout(cct, 3) << "ll_readlink " << vino << dendl; | |
11990 | tout(cct) << "ll_readlink" << std::endl; | |
11991 | tout(cct) << vino.ino.val << std::endl; | |
11992 | ||
11fdf7f2 TL |
11993 | for (auto dn : in->dentries) { |
11994 | touch_dn(dn); | |
7c673cae FG |
11995 | } |
11996 | ||
11997 | int r = _readlink(in, buf, buflen); // FIXME: no permission checking! | |
11998 | ldout(cct, 3) << "ll_readlink " << vino << " = " << r << dendl; | |
11999 | return r; | |
12000 | } | |
12001 | ||
12002 | int Client::_mknod(Inode *dir, const char *name, mode_t mode, dev_t rdev, | |
12003 | const UserPerm& perms, InodeRef *inp) | |
12004 | { | |
1adf2230 | 12005 | ldout(cct, 8) << "_mknod(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
12006 | << mode << dec << ", " << rdev << ", uid " << perms.uid() |
12007 | << ", gid " << perms.gid() << ")" << dendl; | |
12008 | ||
12009 | if (strlen(name) > NAME_MAX) | |
12010 | return -ENAMETOOLONG; | |
12011 | ||
12012 | if (dir->snapid != CEPH_NOSNAP) { | |
12013 | return -EROFS; | |
12014 | } | |
12015 | if (is_quota_files_exceeded(dir, perms)) { | |
12016 | return -EDQUOT; | |
12017 | } | |
12018 | ||
12019 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_MKNOD); | |
12020 | ||
12021 | filepath path; | |
12022 | dir->make_nosnap_relative_path(path); | |
12023 | path.push_dentry(name); | |
12024 | req->set_filepath(path); | |
12025 | req->set_inode(dir); | |
12026 | req->head.args.mknod.rdev = rdev; | |
12027 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12028 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12029 | ||
12030 | bufferlist xattrs_bl; | |
12031 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
12032 | if (res < 0) | |
12033 | goto fail; | |
12034 | req->head.args.mknod.mode = mode; | |
12035 | if (xattrs_bl.length() > 0) | |
12036 | req->set_data(xattrs_bl); | |
12037 | ||
12038 | Dentry *de; | |
12039 | res = get_or_create(dir, name, &de); | |
12040 | if (res < 0) | |
12041 | goto fail; | |
12042 | req->set_dentry(de); | |
12043 | ||
12044 | res = make_request(req, perms, inp); | |
12045 | ||
12046 | trim_cache(); | |
12047 | ||
1adf2230 | 12048 | ldout(cct, 8) << "mknod(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
12049 | return res; |
12050 | ||
12051 | fail: | |
12052 | put_request(req); | |
12053 | return res; | |
12054 | } | |
12055 | ||
12056 | int Client::ll_mknod(Inode *parent, const char *name, mode_t mode, | |
12057 | dev_t rdev, struct stat *attr, Inode **out, | |
12058 | const UserPerm& perms) | |
12059 | { | |
11fdf7f2 | 12060 | std::lock_guard lock(client_lock); |
7c673cae | 12061 | |
181888fb FG |
12062 | if (unmounting) |
12063 | return -ENOTCONN; | |
12064 | ||
7c673cae FG |
12065 | vinodeno_t vparent = _get_vino(parent); |
12066 | ||
12067 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name << dendl; | |
12068 | tout(cct) << "ll_mknod" << std::endl; | |
12069 | tout(cct) << vparent.ino.val << std::endl; | |
12070 | tout(cct) << name << std::endl; | |
12071 | tout(cct) << mode << std::endl; | |
12072 | tout(cct) << rdev << std::endl; | |
12073 | ||
11fdf7f2 | 12074 | if (!fuse_default_permissions) { |
7c673cae FG |
12075 | int r = may_create(parent, perms); |
12076 | if (r < 0) | |
12077 | return r; | |
12078 | } | |
12079 | ||
12080 | InodeRef in; | |
12081 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
12082 | if (r == 0) { | |
12083 | fill_stat(in, attr); | |
12084 | _ll_get(in.get()); | |
12085 | } | |
12086 | tout(cct) << attr->st_ino << std::endl; | |
12087 | ldout(cct, 3) << "ll_mknod " << vparent << " " << name | |
12088 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
12089 | *out = in.get(); | |
12090 | return r; | |
12091 | } | |
12092 | ||
12093 | int Client::ll_mknodx(Inode *parent, const char *name, mode_t mode, | |
12094 | dev_t rdev, Inode **out, | |
12095 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
12096 | const UserPerm& perms) | |
12097 | { | |
12098 | unsigned caps = statx_to_mask(flags, want); | |
11fdf7f2 | 12099 | std::lock_guard lock(client_lock); |
7c673cae | 12100 | |
181888fb FG |
12101 | if (unmounting) |
12102 | return -ENOTCONN; | |
12103 | ||
7c673cae FG |
12104 | vinodeno_t vparent = _get_vino(parent); |
12105 | ||
12106 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name << dendl; | |
12107 | tout(cct) << "ll_mknodx" << std::endl; | |
12108 | tout(cct) << vparent.ino.val << std::endl; | |
12109 | tout(cct) << name << std::endl; | |
12110 | tout(cct) << mode << std::endl; | |
12111 | tout(cct) << rdev << std::endl; | |
12112 | ||
11fdf7f2 | 12113 | if (!fuse_default_permissions) { |
7c673cae FG |
12114 | int r = may_create(parent, perms); |
12115 | if (r < 0) | |
12116 | return r; | |
12117 | } | |
12118 | ||
12119 | InodeRef in; | |
12120 | int r = _mknod(parent, name, mode, rdev, perms, &in); | |
12121 | if (r == 0) { | |
12122 | fill_statx(in, caps, stx); | |
12123 | _ll_get(in.get()); | |
12124 | } | |
12125 | tout(cct) << stx->stx_ino << std::endl; | |
12126 | ldout(cct, 3) << "ll_mknodx " << vparent << " " << name | |
12127 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
12128 | *out = in.get(); | |
12129 | return r; | |
12130 | } | |
12131 | ||
12132 | int Client::_create(Inode *dir, const char *name, int flags, mode_t mode, | |
12133 | InodeRef *inp, Fh **fhp, int stripe_unit, int stripe_count, | |
12134 | int object_size, const char *data_pool, bool *created, | |
12135 | const UserPerm& perms) | |
12136 | { | |
1adf2230 | 12137 | ldout(cct, 8) << "_create(" << dir->ino << " " << name << ", 0" << oct << |
7c673cae FG |
12138 | mode << dec << ")" << dendl; |
12139 | ||
12140 | if (strlen(name) > NAME_MAX) | |
12141 | return -ENAMETOOLONG; | |
12142 | if (dir->snapid != CEPH_NOSNAP) { | |
12143 | return -EROFS; | |
12144 | } | |
12145 | if (is_quota_files_exceeded(dir, perms)) { | |
12146 | return -EDQUOT; | |
12147 | } | |
12148 | ||
12149 | // use normalized flags to generate cmode | |
11fdf7f2 TL |
12150 | int cflags = ceph_flags_sys2wire(flags); |
12151 | if (cct->_conf.get_val<bool>("client_force_lazyio")) | |
12152 | cflags |= CEPH_O_LAZY; | |
12153 | ||
12154 | int cmode = ceph_flags_to_mode(cflags); | |
7c673cae FG |
12155 | |
12156 | int64_t pool_id = -1; | |
12157 | if (data_pool && *data_pool) { | |
12158 | pool_id = objecter->with_osdmap( | |
12159 | std::mem_fn(&OSDMap::lookup_pg_pool_name), data_pool); | |
12160 | if (pool_id < 0) | |
12161 | return -EINVAL; | |
12162 | if (pool_id > 0xffffffffll) | |
12163 | return -ERANGE; // bummer! | |
12164 | } | |
12165 | ||
12166 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_CREATE); | |
12167 | ||
12168 | filepath path; | |
12169 | dir->make_nosnap_relative_path(path); | |
12170 | path.push_dentry(name); | |
12171 | req->set_filepath(path); | |
12172 | req->set_inode(dir); | |
11fdf7f2 | 12173 | req->head.args.open.flags = cflags | CEPH_O_CREAT; |
7c673cae FG |
12174 | |
12175 | req->head.args.open.stripe_unit = stripe_unit; | |
12176 | req->head.args.open.stripe_count = stripe_count; | |
12177 | req->head.args.open.object_size = object_size; | |
12178 | if (cct->_conf->client_debug_getattr_caps) | |
12179 | req->head.args.open.mask = DEBUG_GETATTR_CAPS; | |
12180 | else | |
12181 | req->head.args.open.mask = 0; | |
12182 | req->head.args.open.pool = pool_id; | |
12183 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12184 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12185 | ||
12186 | mode |= S_IFREG; | |
12187 | bufferlist xattrs_bl; | |
12188 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perms); | |
12189 | if (res < 0) | |
12190 | goto fail; | |
12191 | req->head.args.open.mode = mode; | |
12192 | if (xattrs_bl.length() > 0) | |
12193 | req->set_data(xattrs_bl); | |
12194 | ||
12195 | Dentry *de; | |
12196 | res = get_or_create(dir, name, &de); | |
12197 | if (res < 0) | |
12198 | goto fail; | |
12199 | req->set_dentry(de); | |
12200 | ||
12201 | res = make_request(req, perms, inp, created); | |
12202 | if (res < 0) { | |
12203 | goto reply_error; | |
12204 | } | |
12205 | ||
12206 | /* If the caller passed a value in fhp, do the open */ | |
12207 | if(fhp) { | |
12208 | (*inp)->get_open_ref(cmode); | |
12209 | *fhp = _create_fh(inp->get(), flags, cmode, perms); | |
12210 | } | |
12211 | ||
12212 | reply_error: | |
12213 | trim_cache(); | |
12214 | ||
1adf2230 | 12215 | ldout(cct, 8) << "create(" << path << ", 0" << oct << mode << dec |
7c673cae FG |
12216 | << " layout " << stripe_unit |
12217 | << ' ' << stripe_count | |
12218 | << ' ' << object_size | |
12219 | <<") = " << res << dendl; | |
12220 | return res; | |
12221 | ||
12222 | fail: | |
12223 | put_request(req); | |
12224 | return res; | |
12225 | } | |
12226 | ||
12227 | ||
12228 | int Client::_mkdir(Inode *dir, const char *name, mode_t mode, const UserPerm& perm, | |
12229 | InodeRef *inp) | |
12230 | { | |
1adf2230 | 12231 | ldout(cct, 8) << "_mkdir(" << dir->ino << " " << name << ", 0" << oct |
7c673cae FG |
12232 | << mode << dec << ", uid " << perm.uid() |
12233 | << ", gid " << perm.gid() << ")" << dendl; | |
12234 | ||
12235 | if (strlen(name) > NAME_MAX) | |
12236 | return -ENAMETOOLONG; | |
12237 | ||
12238 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
12239 | return -EROFS; | |
12240 | } | |
12241 | if (is_quota_files_exceeded(dir, perm)) { | |
12242 | return -EDQUOT; | |
12243 | } | |
12244 | MetaRequest *req = new MetaRequest(dir->snapid == CEPH_SNAPDIR ? | |
12245 | CEPH_MDS_OP_MKSNAP : CEPH_MDS_OP_MKDIR); | |
12246 | ||
12247 | filepath path; | |
12248 | dir->make_nosnap_relative_path(path); | |
12249 | path.push_dentry(name); | |
12250 | req->set_filepath(path); | |
12251 | req->set_inode(dir); | |
12252 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12253 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12254 | ||
12255 | mode |= S_IFDIR; | |
12256 | bufferlist xattrs_bl; | |
12257 | int res = _posix_acl_create(dir, &mode, xattrs_bl, perm); | |
12258 | if (res < 0) | |
12259 | goto fail; | |
12260 | req->head.args.mkdir.mode = mode; | |
12261 | if (xattrs_bl.length() > 0) | |
12262 | req->set_data(xattrs_bl); | |
12263 | ||
12264 | Dentry *de; | |
12265 | res = get_or_create(dir, name, &de); | |
12266 | if (res < 0) | |
12267 | goto fail; | |
12268 | req->set_dentry(de); | |
12269 | ||
12270 | ldout(cct, 10) << "_mkdir: making request" << dendl; | |
12271 | res = make_request(req, perm, inp); | |
12272 | ldout(cct, 10) << "_mkdir result is " << res << dendl; | |
12273 | ||
12274 | trim_cache(); | |
12275 | ||
1adf2230 | 12276 | ldout(cct, 8) << "_mkdir(" << path << ", 0" << oct << mode << dec << ") = " << res << dendl; |
7c673cae FG |
12277 | return res; |
12278 | ||
12279 | fail: | |
12280 | put_request(req); | |
12281 | return res; | |
12282 | } | |
12283 | ||
12284 | int Client::ll_mkdir(Inode *parent, const char *name, mode_t mode, | |
12285 | struct stat *attr, Inode **out, const UserPerm& perm) | |
12286 | { | |
11fdf7f2 | 12287 | std::lock_guard lock(client_lock); |
7c673cae | 12288 | |
181888fb FG |
12289 | if (unmounting) |
12290 | return -ENOTCONN; | |
12291 | ||
7c673cae FG |
12292 | vinodeno_t vparent = _get_vino(parent); |
12293 | ||
12294 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name << dendl; | |
12295 | tout(cct) << "ll_mkdir" << std::endl; | |
12296 | tout(cct) << vparent.ino.val << std::endl; | |
12297 | tout(cct) << name << std::endl; | |
12298 | tout(cct) << mode << std::endl; | |
12299 | ||
11fdf7f2 | 12300 | if (!fuse_default_permissions) { |
7c673cae FG |
12301 | int r = may_create(parent, perm); |
12302 | if (r < 0) | |
12303 | return r; | |
12304 | } | |
12305 | ||
12306 | InodeRef in; | |
12307 | int r = _mkdir(parent, name, mode, perm, &in); | |
12308 | if (r == 0) { | |
12309 | fill_stat(in, attr); | |
12310 | _ll_get(in.get()); | |
12311 | } | |
12312 | tout(cct) << attr->st_ino << std::endl; | |
12313 | ldout(cct, 3) << "ll_mkdir " << vparent << " " << name | |
12314 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
12315 | *out = in.get(); | |
12316 | return r; | |
12317 | } | |
12318 | ||
12319 | int Client::ll_mkdirx(Inode *parent, const char *name, mode_t mode, Inode **out, | |
12320 | struct ceph_statx *stx, unsigned want, unsigned flags, | |
12321 | const UserPerm& perms) | |
12322 | { | |
11fdf7f2 | 12323 | std::lock_guard lock(client_lock); |
7c673cae | 12324 | |
181888fb FG |
12325 | if (unmounting) |
12326 | return -ENOTCONN; | |
12327 | ||
7c673cae FG |
12328 | vinodeno_t vparent = _get_vino(parent); |
12329 | ||
12330 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name << dendl; | |
12331 | tout(cct) << "ll_mkdirx" << std::endl; | |
12332 | tout(cct) << vparent.ino.val << std::endl; | |
12333 | tout(cct) << name << std::endl; | |
12334 | tout(cct) << mode << std::endl; | |
12335 | ||
11fdf7f2 | 12336 | if (!fuse_default_permissions) { |
7c673cae FG |
12337 | int r = may_create(parent, perms); |
12338 | if (r < 0) | |
12339 | return r; | |
12340 | } | |
12341 | ||
12342 | InodeRef in; | |
12343 | int r = _mkdir(parent, name, mode, perms, &in); | |
12344 | if (r == 0) { | |
12345 | fill_statx(in, statx_to_mask(flags, want), stx); | |
12346 | _ll_get(in.get()); | |
12347 | } else { | |
12348 | stx->stx_ino = 0; | |
12349 | stx->stx_mask = 0; | |
12350 | } | |
12351 | tout(cct) << stx->stx_ino << std::endl; | |
12352 | ldout(cct, 3) << "ll_mkdirx " << vparent << " " << name | |
12353 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
12354 | *out = in.get(); | |
12355 | return r; | |
12356 | } | |
12357 | ||
12358 | int Client::_symlink(Inode *dir, const char *name, const char *target, | |
12359 | const UserPerm& perms, InodeRef *inp) | |
12360 | { | |
1adf2230 | 12361 | ldout(cct, 8) << "_symlink(" << dir->ino << " " << name << ", " << target |
7c673cae FG |
12362 | << ", uid " << perms.uid() << ", gid " << perms.gid() << ")" |
12363 | << dendl; | |
12364 | ||
12365 | if (strlen(name) > NAME_MAX) | |
12366 | return -ENAMETOOLONG; | |
12367 | ||
12368 | if (dir->snapid != CEPH_NOSNAP) { | |
12369 | return -EROFS; | |
12370 | } | |
12371 | if (is_quota_files_exceeded(dir, perms)) { | |
12372 | return -EDQUOT; | |
12373 | } | |
12374 | ||
12375 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_SYMLINK); | |
12376 | ||
12377 | filepath path; | |
12378 | dir->make_nosnap_relative_path(path); | |
12379 | path.push_dentry(name); | |
12380 | req->set_filepath(path); | |
12381 | req->set_inode(dir); | |
12382 | req->set_string2(target); | |
12383 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12384 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12385 | ||
12386 | Dentry *de; | |
12387 | int res = get_or_create(dir, name, &de); | |
12388 | if (res < 0) | |
12389 | goto fail; | |
12390 | req->set_dentry(de); | |
12391 | ||
12392 | res = make_request(req, perms, inp); | |
12393 | ||
12394 | trim_cache(); | |
1adf2230 | 12395 | ldout(cct, 8) << "_symlink(\"" << path << "\", \"" << target << "\") = " << |
7c673cae FG |
12396 | res << dendl; |
12397 | return res; | |
12398 | ||
12399 | fail: | |
12400 | put_request(req); | |
12401 | return res; | |
12402 | } | |
12403 | ||
12404 | int Client::ll_symlink(Inode *parent, const char *name, const char *value, | |
12405 | struct stat *attr, Inode **out, const UserPerm& perms) | |
12406 | { | |
11fdf7f2 | 12407 | std::lock_guard lock(client_lock); |
7c673cae | 12408 | |
181888fb FG |
12409 | if (unmounting) |
12410 | return -ENOTCONN; | |
12411 | ||
7c673cae FG |
12412 | vinodeno_t vparent = _get_vino(parent); |
12413 | ||
12414 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name << " -> " << value | |
12415 | << dendl; | |
12416 | tout(cct) << "ll_symlink" << std::endl; | |
12417 | tout(cct) << vparent.ino.val << std::endl; | |
12418 | tout(cct) << name << std::endl; | |
12419 | tout(cct) << value << std::endl; | |
12420 | ||
11fdf7f2 | 12421 | if (!fuse_default_permissions) { |
7c673cae FG |
12422 | int r = may_create(parent, perms); |
12423 | if (r < 0) | |
12424 | return r; | |
12425 | } | |
12426 | ||
12427 | InodeRef in; | |
12428 | int r = _symlink(parent, name, value, perms, &in); | |
12429 | if (r == 0) { | |
12430 | fill_stat(in, attr); | |
12431 | _ll_get(in.get()); | |
12432 | } | |
12433 | tout(cct) << attr->st_ino << std::endl; | |
12434 | ldout(cct, 3) << "ll_symlink " << vparent << " " << name | |
12435 | << " = " << r << " (" << hex << attr->st_ino << dec << ")" << dendl; | |
12436 | *out = in.get(); | |
12437 | return r; | |
12438 | } | |
12439 | ||
12440 | int Client::ll_symlinkx(Inode *parent, const char *name, const char *value, | |
12441 | Inode **out, struct ceph_statx *stx, unsigned want, | |
12442 | unsigned flags, const UserPerm& perms) | |
12443 | { | |
11fdf7f2 | 12444 | std::lock_guard lock(client_lock); |
7c673cae | 12445 | |
181888fb FG |
12446 | if (unmounting) |
12447 | return -ENOTCONN; | |
12448 | ||
7c673cae FG |
12449 | vinodeno_t vparent = _get_vino(parent); |
12450 | ||
12451 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name << " -> " << value | |
12452 | << dendl; | |
12453 | tout(cct) << "ll_symlinkx" << std::endl; | |
12454 | tout(cct) << vparent.ino.val << std::endl; | |
12455 | tout(cct) << name << std::endl; | |
12456 | tout(cct) << value << std::endl; | |
12457 | ||
11fdf7f2 | 12458 | if (!fuse_default_permissions) { |
7c673cae FG |
12459 | int r = may_create(parent, perms); |
12460 | if (r < 0) | |
12461 | return r; | |
12462 | } | |
12463 | ||
12464 | InodeRef in; | |
12465 | int r = _symlink(parent, name, value, perms, &in); | |
12466 | if (r == 0) { | |
12467 | fill_statx(in, statx_to_mask(flags, want), stx); | |
12468 | _ll_get(in.get()); | |
12469 | } | |
12470 | tout(cct) << stx->stx_ino << std::endl; | |
12471 | ldout(cct, 3) << "ll_symlinkx " << vparent << " " << name | |
12472 | << " = " << r << " (" << hex << stx->stx_ino << dec << ")" << dendl; | |
12473 | *out = in.get(); | |
12474 | return r; | |
12475 | } | |
12476 | ||
12477 | int Client::_unlink(Inode *dir, const char *name, const UserPerm& perm) | |
12478 | { | |
1adf2230 | 12479 | ldout(cct, 8) << "_unlink(" << dir->ino << " " << name |
7c673cae FG |
12480 | << " uid " << perm.uid() << " gid " << perm.gid() |
12481 | << ")" << dendl; | |
12482 | ||
12483 | if (dir->snapid != CEPH_NOSNAP) { | |
12484 | return -EROFS; | |
12485 | } | |
12486 | ||
12487 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_UNLINK); | |
12488 | ||
12489 | filepath path; | |
12490 | dir->make_nosnap_relative_path(path); | |
12491 | path.push_dentry(name); | |
12492 | req->set_filepath(path); | |
12493 | ||
12494 | InodeRef otherin; | |
b32b8144 | 12495 | Inode *in; |
7c673cae | 12496 | Dentry *de; |
b32b8144 | 12497 | |
7c673cae FG |
12498 | int res = get_or_create(dir, name, &de); |
12499 | if (res < 0) | |
12500 | goto fail; | |
12501 | req->set_dentry(de); | |
12502 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12503 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12504 | ||
12505 | res = _lookup(dir, name, 0, &otherin, perm); | |
12506 | if (res < 0) | |
12507 | goto fail; | |
b32b8144 FG |
12508 | |
12509 | in = otherin.get(); | |
12510 | req->set_other_inode(in); | |
12511 | in->break_all_delegs(); | |
7c673cae FG |
12512 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
12513 | ||
12514 | req->set_inode(dir); | |
12515 | ||
12516 | res = make_request(req, perm); | |
12517 | ||
12518 | trim_cache(); | |
1adf2230 | 12519 | ldout(cct, 8) << "unlink(" << path << ") = " << res << dendl; |
7c673cae FG |
12520 | return res; |
12521 | ||
12522 | fail: | |
12523 | put_request(req); | |
12524 | return res; | |
12525 | } | |
12526 | ||
12527 | int Client::ll_unlink(Inode *in, const char *name, const UserPerm& perm) | |
12528 | { | |
11fdf7f2 | 12529 | std::lock_guard lock(client_lock); |
7c673cae | 12530 | |
181888fb FG |
12531 | if (unmounting) |
12532 | return -ENOTCONN; | |
12533 | ||
7c673cae FG |
12534 | vinodeno_t vino = _get_vino(in); |
12535 | ||
12536 | ldout(cct, 3) << "ll_unlink " << vino << " " << name << dendl; | |
12537 | tout(cct) << "ll_unlink" << std::endl; | |
12538 | tout(cct) << vino.ino.val << std::endl; | |
12539 | tout(cct) << name << std::endl; | |
12540 | ||
11fdf7f2 | 12541 | if (!fuse_default_permissions) { |
7c673cae FG |
12542 | int r = may_delete(in, name, perm); |
12543 | if (r < 0) | |
12544 | return r; | |
12545 | } | |
12546 | return _unlink(in, name, perm); | |
12547 | } | |
12548 | ||
12549 | int Client::_rmdir(Inode *dir, const char *name, const UserPerm& perms) | |
12550 | { | |
1adf2230 | 12551 | ldout(cct, 8) << "_rmdir(" << dir->ino << " " << name << " uid " |
7c673cae FG |
12552 | << perms.uid() << " gid " << perms.gid() << ")" << dendl; |
12553 | ||
12554 | if (dir->snapid != CEPH_NOSNAP && dir->snapid != CEPH_SNAPDIR) { | |
12555 | return -EROFS; | |
12556 | } | |
b32b8144 FG |
12557 | |
12558 | int op = dir->snapid == CEPH_SNAPDIR ? CEPH_MDS_OP_RMSNAP : CEPH_MDS_OP_RMDIR; | |
12559 | MetaRequest *req = new MetaRequest(op); | |
7c673cae FG |
12560 | filepath path; |
12561 | dir->make_nosnap_relative_path(path); | |
12562 | path.push_dentry(name); | |
12563 | req->set_filepath(path); | |
11fdf7f2 | 12564 | req->set_inode(dir); |
7c673cae FG |
12565 | |
12566 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12567 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12568 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; | |
12569 | ||
12570 | InodeRef in; | |
12571 | ||
12572 | Dentry *de; | |
12573 | int res = get_or_create(dir, name, &de); | |
12574 | if (res < 0) | |
12575 | goto fail; | |
b32b8144 FG |
12576 | if (op == CEPH_MDS_OP_RMDIR) |
12577 | req->set_dentry(de); | |
12578 | else | |
12579 | de->get(); | |
12580 | ||
7c673cae FG |
12581 | res = _lookup(dir, name, 0, &in, perms); |
12582 | if (res < 0) | |
12583 | goto fail; | |
11fdf7f2 TL |
12584 | |
12585 | if (op == CEPH_MDS_OP_RMSNAP) { | |
7c673cae | 12586 | unlink(de, true, true); |
b32b8144 | 12587 | de->put(); |
7c673cae | 12588 | } |
11fdf7f2 | 12589 | req->set_other_inode(in.get()); |
7c673cae FG |
12590 | |
12591 | res = make_request(req, perms); | |
12592 | ||
12593 | trim_cache(); | |
1adf2230 | 12594 | ldout(cct, 8) << "rmdir(" << path << ") = " << res << dendl; |
7c673cae FG |
12595 | return res; |
12596 | ||
12597 | fail: | |
12598 | put_request(req); | |
12599 | return res; | |
12600 | } | |
12601 | ||
12602 | int Client::ll_rmdir(Inode *in, const char *name, const UserPerm& perms) | |
12603 | { | |
11fdf7f2 | 12604 | std::lock_guard lock(client_lock); |
7c673cae | 12605 | |
181888fb FG |
12606 | if (unmounting) |
12607 | return -ENOTCONN; | |
12608 | ||
7c673cae FG |
12609 | vinodeno_t vino = _get_vino(in); |
12610 | ||
12611 | ldout(cct, 3) << "ll_rmdir " << vino << " " << name << dendl; | |
12612 | tout(cct) << "ll_rmdir" << std::endl; | |
12613 | tout(cct) << vino.ino.val << std::endl; | |
12614 | tout(cct) << name << std::endl; | |
12615 | ||
11fdf7f2 | 12616 | if (!fuse_default_permissions) { |
7c673cae FG |
12617 | int r = may_delete(in, name, perms); |
12618 | if (r < 0) | |
12619 | return r; | |
12620 | } | |
12621 | ||
12622 | return _rmdir(in, name, perms); | |
12623 | } | |
12624 | ||
12625 | int Client::_rename(Inode *fromdir, const char *fromname, Inode *todir, const char *toname, const UserPerm& perm) | |
12626 | { | |
1adf2230 | 12627 | ldout(cct, 8) << "_rename(" << fromdir->ino << " " << fromname << " to " |
7c673cae FG |
12628 | << todir->ino << " " << toname |
12629 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" | |
12630 | << dendl; | |
12631 | ||
12632 | if (fromdir->snapid != todir->snapid) | |
12633 | return -EXDEV; | |
12634 | ||
12635 | int op = CEPH_MDS_OP_RENAME; | |
12636 | if (fromdir->snapid != CEPH_NOSNAP) { | |
12637 | if (fromdir == todir && fromdir->snapid == CEPH_SNAPDIR) | |
12638 | op = CEPH_MDS_OP_RENAMESNAP; | |
12639 | else | |
12640 | return -EROFS; | |
12641 | } | |
7c673cae FG |
12642 | |
12643 | InodeRef target; | |
12644 | MetaRequest *req = new MetaRequest(op); | |
12645 | ||
12646 | filepath from; | |
12647 | fromdir->make_nosnap_relative_path(from); | |
12648 | from.push_dentry(fromname); | |
12649 | filepath to; | |
12650 | todir->make_nosnap_relative_path(to); | |
12651 | to.push_dentry(toname); | |
12652 | req->set_filepath(to); | |
12653 | req->set_filepath2(from); | |
12654 | ||
12655 | Dentry *oldde; | |
12656 | int res = get_or_create(fromdir, fromname, &oldde); | |
12657 | if (res < 0) | |
12658 | goto fail; | |
12659 | Dentry *de; | |
12660 | res = get_or_create(todir, toname, &de); | |
12661 | if (res < 0) | |
12662 | goto fail; | |
12663 | ||
12664 | if (op == CEPH_MDS_OP_RENAME) { | |
12665 | req->set_old_dentry(oldde); | |
12666 | req->old_dentry_drop = CEPH_CAP_FILE_SHARED; | |
12667 | req->old_dentry_unless = CEPH_CAP_FILE_EXCL; | |
12668 | ||
12669 | req->set_dentry(de); | |
12670 | req->dentry_drop = CEPH_CAP_FILE_SHARED; | |
12671 | req->dentry_unless = CEPH_CAP_FILE_EXCL; | |
12672 | ||
12673 | InodeRef oldin, otherin; | |
9f95a23c TL |
12674 | Inode *fromdir_root = nullptr; |
12675 | Inode *todir_root = nullptr; | |
12676 | int mask = 0; | |
12677 | bool quota_check = false; | |
12678 | if (fromdir != todir) { | |
12679 | fromdir_root = | |
12680 | fromdir->quota.is_enable() ? fromdir : get_quota_root(fromdir, perm); | |
12681 | todir_root = | |
12682 | todir->quota.is_enable() ? todir : get_quota_root(todir, perm); | |
12683 | ||
12684 | if (todir_root->quota.is_enable() && fromdir_root != todir_root) { | |
12685 | // use CEPH_STAT_RSTAT mask to force send getattr or lookup request | |
12686 | // to auth MDS to get latest rstat for todir_root and source dir | |
12687 | // even if their dentry caches and inode caps are satisfied. | |
12688 | res = _getattr(todir_root, CEPH_STAT_RSTAT, perm, true); | |
12689 | if (res < 0) | |
12690 | goto fail; | |
12691 | ||
12692 | quota_check = true; | |
12693 | if (oldde->inode && oldde->inode->is_dir()) { | |
12694 | mask |= CEPH_STAT_RSTAT; | |
12695 | } | |
12696 | } | |
12697 | } | |
12698 | ||
12699 | res = _lookup(fromdir, fromname, mask, &oldin, perm); | |
7c673cae FG |
12700 | if (res < 0) |
12701 | goto fail; | |
b32b8144 FG |
12702 | |
12703 | Inode *oldinode = oldin.get(); | |
12704 | oldinode->break_all_delegs(); | |
12705 | req->set_old_inode(oldinode); | |
7c673cae FG |
12706 | req->old_inode_drop = CEPH_CAP_LINK_SHARED; |
12707 | ||
9f95a23c TL |
12708 | if (quota_check) { |
12709 | int64_t old_bytes, old_files; | |
12710 | if (oldinode->is_dir()) { | |
12711 | old_bytes = oldinode->rstat.rbytes; | |
12712 | old_files = oldinode->rstat.rsize(); | |
12713 | } else { | |
12714 | old_bytes = oldinode->size; | |
12715 | old_files = 1; | |
12716 | } | |
12717 | ||
12718 | bool quota_exceed = false; | |
12719 | if (todir_root && todir_root->quota.max_bytes && | |
12720 | (old_bytes + todir_root->rstat.rbytes) >= todir_root->quota.max_bytes) { | |
12721 | ldout(cct, 10) << "_rename (" << oldinode->ino << " bytes=" | |
12722 | << old_bytes << ") to (" << todir->ino | |
12723 | << ") will exceed quota on " << *todir_root << dendl; | |
12724 | quota_exceed = true; | |
12725 | } | |
12726 | ||
12727 | if (todir_root && todir_root->quota.max_files && | |
12728 | (old_files + todir_root->rstat.rsize()) >= todir_root->quota.max_files) { | |
12729 | ldout(cct, 10) << "_rename (" << oldinode->ino << " files=" | |
12730 | << old_files << ") to (" << todir->ino | |
12731 | << ") will exceed quota on " << *todir_root << dendl; | |
12732 | quota_exceed = true; | |
12733 | } | |
12734 | ||
12735 | if (quota_exceed) { | |
12736 | res = (oldinode->is_dir()) ? -EXDEV : -EDQUOT; | |
12737 | goto fail; | |
12738 | } | |
12739 | } | |
12740 | ||
7c673cae | 12741 | res = _lookup(todir, toname, 0, &otherin, perm); |
b32b8144 FG |
12742 | switch (res) { |
12743 | case 0: | |
12744 | { | |
12745 | Inode *in = otherin.get(); | |
12746 | req->set_other_inode(in); | |
12747 | in->break_all_delegs(); | |
12748 | } | |
7c673cae | 12749 | req->other_inode_drop = CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL; |
b32b8144 FG |
12750 | break; |
12751 | case -ENOENT: | |
12752 | break; | |
12753 | default: | |
12754 | goto fail; | |
7c673cae FG |
12755 | } |
12756 | ||
12757 | req->set_inode(todir); | |
12758 | } else { | |
12759 | // renamesnap reply contains no tracedn, so we need to invalidate | |
12760 | // dentry manually | |
12761 | unlink(oldde, true, true); | |
12762 | unlink(de, true, true); | |
11fdf7f2 TL |
12763 | |
12764 | req->set_inode(todir); | |
7c673cae FG |
12765 | } |
12766 | ||
12767 | res = make_request(req, perm, &target); | |
12768 | ldout(cct, 10) << "rename result is " << res << dendl; | |
12769 | ||
12770 | // renamed item from our cache | |
12771 | ||
12772 | trim_cache(); | |
1adf2230 | 12773 | ldout(cct, 8) << "_rename(" << from << ", " << to << ") = " << res << dendl; |
7c673cae FG |
12774 | return res; |
12775 | ||
12776 | fail: | |
12777 | put_request(req); | |
12778 | return res; | |
12779 | } | |
12780 | ||
12781 | int Client::ll_rename(Inode *parent, const char *name, Inode *newparent, | |
12782 | const char *newname, const UserPerm& perm) | |
12783 | { | |
11fdf7f2 | 12784 | std::lock_guard lock(client_lock); |
7c673cae | 12785 | |
181888fb FG |
12786 | if (unmounting) |
12787 | return -ENOTCONN; | |
12788 | ||
7c673cae FG |
12789 | vinodeno_t vparent = _get_vino(parent); |
12790 | vinodeno_t vnewparent = _get_vino(newparent); | |
12791 | ||
12792 | ldout(cct, 3) << "ll_rename " << vparent << " " << name << " to " | |
12793 | << vnewparent << " " << newname << dendl; | |
12794 | tout(cct) << "ll_rename" << std::endl; | |
12795 | tout(cct) << vparent.ino.val << std::endl; | |
12796 | tout(cct) << name << std::endl; | |
12797 | tout(cct) << vnewparent.ino.val << std::endl; | |
12798 | tout(cct) << newname << std::endl; | |
12799 | ||
11fdf7f2 | 12800 | if (!fuse_default_permissions) { |
7c673cae FG |
12801 | int r = may_delete(parent, name, perm); |
12802 | if (r < 0) | |
12803 | return r; | |
12804 | r = may_delete(newparent, newname, perm); | |
12805 | if (r < 0 && r != -ENOENT) | |
12806 | return r; | |
12807 | } | |
12808 | ||
12809 | return _rename(parent, name, newparent, newname, perm); | |
12810 | } | |
12811 | ||
12812 | int Client::_link(Inode *in, Inode *dir, const char *newname, const UserPerm& perm, InodeRef *inp) | |
12813 | { | |
1adf2230 | 12814 | ldout(cct, 8) << "_link(" << in->ino << " to " << dir->ino << " " << newname |
7c673cae FG |
12815 | << " uid " << perm.uid() << " gid " << perm.gid() << ")" << dendl; |
12816 | ||
12817 | if (strlen(newname) > NAME_MAX) | |
12818 | return -ENAMETOOLONG; | |
12819 | ||
12820 | if (in->snapid != CEPH_NOSNAP || dir->snapid != CEPH_NOSNAP) { | |
12821 | return -EROFS; | |
12822 | } | |
12823 | if (is_quota_files_exceeded(dir, perm)) { | |
12824 | return -EDQUOT; | |
12825 | } | |
12826 | ||
b32b8144 | 12827 | in->break_all_delegs(); |
7c673cae FG |
12828 | MetaRequest *req = new MetaRequest(CEPH_MDS_OP_LINK); |
12829 | ||
12830 | filepath path(newname, dir->ino); | |
12831 | req->set_filepath(path); | |
12832 | filepath existing(in->ino); | |
12833 | req->set_filepath2(existing); | |
12834 | ||
12835 | req->set_inode(dir); | |
12836 | req->inode_drop = CEPH_CAP_FILE_SHARED; | |
12837 | req->inode_unless = CEPH_CAP_FILE_EXCL; | |
12838 | ||
12839 | Dentry *de; | |
12840 | int res = get_or_create(dir, newname, &de); | |
12841 | if (res < 0) | |
12842 | goto fail; | |
12843 | req->set_dentry(de); | |
12844 | ||
12845 | res = make_request(req, perm, inp); | |
12846 | ldout(cct, 10) << "link result is " << res << dendl; | |
12847 | ||
12848 | trim_cache(); | |
1adf2230 | 12849 | ldout(cct, 8) << "link(" << existing << ", " << path << ") = " << res << dendl; |
7c673cae FG |
12850 | return res; |
12851 | ||
12852 | fail: | |
12853 | put_request(req); | |
12854 | return res; | |
12855 | } | |
12856 | ||
12857 | int Client::ll_link(Inode *in, Inode *newparent, const char *newname, | |
12858 | const UserPerm& perm) | |
12859 | { | |
11fdf7f2 | 12860 | std::lock_guard lock(client_lock); |
7c673cae | 12861 | |
181888fb FG |
12862 | if (unmounting) |
12863 | return -ENOTCONN; | |
12864 | ||
7c673cae FG |
12865 | vinodeno_t vino = _get_vino(in); |
12866 | vinodeno_t vnewparent = _get_vino(newparent); | |
12867 | ||
31f18b77 | 12868 | ldout(cct, 3) << "ll_link " << vino << " to " << vnewparent << " " << |
7c673cae FG |
12869 | newname << dendl; |
12870 | tout(cct) << "ll_link" << std::endl; | |
12871 | tout(cct) << vino.ino.val << std::endl; | |
12872 | tout(cct) << vnewparent << std::endl; | |
12873 | tout(cct) << newname << std::endl; | |
12874 | ||
7c673cae FG |
12875 | InodeRef target; |
12876 | ||
11fdf7f2 | 12877 | if (!fuse_default_permissions) { |
7c673cae FG |
12878 | if (S_ISDIR(in->mode)) |
12879 | return -EPERM; | |
12880 | ||
11fdf7f2 | 12881 | int r = may_hardlink(in, perm); |
7c673cae FG |
12882 | if (r < 0) |
12883 | return r; | |
12884 | ||
12885 | r = may_create(newparent, perm); | |
12886 | if (r < 0) | |
12887 | return r; | |
12888 | } | |
12889 | ||
12890 | return _link(in, newparent, newname, perm, &target); | |
12891 | } | |
12892 | ||
12893 | int Client::ll_num_osds(void) | |
12894 | { | |
11fdf7f2 | 12895 | std::lock_guard lock(client_lock); |
7c673cae FG |
12896 | return objecter->with_osdmap(std::mem_fn(&OSDMap::get_num_osds)); |
12897 | } | |
12898 | ||
12899 | int Client::ll_osdaddr(int osd, uint32_t *addr) | |
12900 | { | |
11fdf7f2 | 12901 | std::lock_guard lock(client_lock); |
181888fb | 12902 | |
7c673cae FG |
12903 | entity_addr_t g; |
12904 | bool exists = objecter->with_osdmap([&](const OSDMap& o) { | |
12905 | if (!o.exists(osd)) | |
12906 | return false; | |
11fdf7f2 | 12907 | g = o.get_addrs(osd).front(); |
7c673cae FG |
12908 | return true; |
12909 | }); | |
12910 | if (!exists) | |
12911 | return -1; | |
12912 | uint32_t nb_addr = (g.in4_addr()).sin_addr.s_addr; | |
12913 | *addr = ntohl(nb_addr); | |
12914 | return 0; | |
12915 | } | |
181888fb | 12916 | |
7c673cae FG |
12917 | uint32_t Client::ll_stripe_unit(Inode *in) |
12918 | { | |
11fdf7f2 | 12919 | std::lock_guard lock(client_lock); |
7c673cae FG |
12920 | return in->layout.stripe_unit; |
12921 | } | |
12922 | ||
12923 | uint64_t Client::ll_snap_seq(Inode *in) | |
12924 | { | |
11fdf7f2 | 12925 | std::lock_guard lock(client_lock); |
7c673cae FG |
12926 | return in->snaprealm->seq; |
12927 | } | |
12928 | ||
12929 | int Client::ll_file_layout(Inode *in, file_layout_t *layout) | |
12930 | { | |
11fdf7f2 | 12931 | std::lock_guard lock(client_lock); |
7c673cae FG |
12932 | *layout = in->layout; |
12933 | return 0; | |
12934 | } | |
12935 | ||
12936 | int Client::ll_file_layout(Fh *fh, file_layout_t *layout) | |
12937 | { | |
12938 | return ll_file_layout(fh->inode.get(), layout); | |
12939 | } | |
12940 | ||
12941 | /* Currently we cannot take advantage of redundancy in reads, since we | |
12942 | would have to go through all possible placement groups (a | |
12943 | potentially quite large number determined by a hash), and use CRUSH | |
12944 | to calculate the appropriate set of OSDs for each placement group, | |
12945 | then index into that. An array with one entry per OSD is much more | |
12946 | tractable and works for demonstration purposes. */ | |
12947 | ||
12948 | int Client::ll_get_stripe_osd(Inode *in, uint64_t blockno, | |
12949 | file_layout_t* layout) | |
12950 | { | |
11fdf7f2 | 12951 | std::lock_guard lock(client_lock); |
181888fb | 12952 | |
28e407b8 | 12953 | inodeno_t ino = in->ino; |
7c673cae FG |
12954 | uint32_t object_size = layout->object_size; |
12955 | uint32_t su = layout->stripe_unit; | |
12956 | uint32_t stripe_count = layout->stripe_count; | |
12957 | uint64_t stripes_per_object = object_size / su; | |
11fdf7f2 | 12958 | uint64_t stripeno = 0, stripepos = 0; |
7c673cae | 12959 | |
11fdf7f2 TL |
12960 | if(stripe_count) { |
12961 | stripeno = blockno / stripe_count; // which horizontal stripe (Y) | |
12962 | stripepos = blockno % stripe_count; // which object in the object set (X) | |
12963 | } | |
7c673cae FG |
12964 | uint64_t objectsetno = stripeno / stripes_per_object; // which object set |
12965 | uint64_t objectno = objectsetno * stripe_count + stripepos; // object id | |
12966 | ||
12967 | object_t oid = file_object_t(ino, objectno); | |
12968 | return objecter->with_osdmap([&](const OSDMap& o) { | |
12969 | ceph_object_layout olayout = | |
12970 | o.file_to_object_layout(oid, *layout); | |
12971 | pg_t pg = (pg_t)olayout.ol_pgid; | |
12972 | vector<int> osds; | |
12973 | int primary; | |
12974 | o.pg_to_acting_osds(pg, &osds, &primary); | |
12975 | return primary; | |
12976 | }); | |
12977 | } | |
12978 | ||
12979 | /* Return the offset of the block, internal to the object */ | |
12980 | ||
12981 | uint64_t Client::ll_get_internal_offset(Inode *in, uint64_t blockno) | |
12982 | { | |
11fdf7f2 | 12983 | std::lock_guard lock(client_lock); |
7c673cae FG |
12984 | file_layout_t *layout=&(in->layout); |
12985 | uint32_t object_size = layout->object_size; | |
12986 | uint32_t su = layout->stripe_unit; | |
12987 | uint64_t stripes_per_object = object_size / su; | |
12988 | ||
12989 | return (blockno % stripes_per_object) * su; | |
12990 | } | |
12991 | ||
12992 | int Client::ll_opendir(Inode *in, int flags, dir_result_t** dirpp, | |
12993 | const UserPerm& perms) | |
12994 | { | |
11fdf7f2 | 12995 | std::lock_guard lock(client_lock); |
7c673cae | 12996 | |
181888fb FG |
12997 | if (unmounting) |
12998 | return -ENOTCONN; | |
12999 | ||
7c673cae FG |
13000 | vinodeno_t vino = _get_vino(in); |
13001 | ||
13002 | ldout(cct, 3) << "ll_opendir " << vino << dendl; | |
13003 | tout(cct) << "ll_opendir" << std::endl; | |
13004 | tout(cct) << vino.ino.val << std::endl; | |
13005 | ||
11fdf7f2 | 13006 | if (!fuse_default_permissions) { |
7c673cae FG |
13007 | int r = may_open(in, flags, perms); |
13008 | if (r < 0) | |
13009 | return r; | |
13010 | } | |
13011 | ||
13012 | int r = _opendir(in, dirpp, perms); | |
13013 | tout(cct) << (unsigned long)*dirpp << std::endl; | |
13014 | ||
13015 | ldout(cct, 3) << "ll_opendir " << vino << " = " << r << " (" << *dirpp << ")" | |
13016 | << dendl; | |
13017 | return r; | |
13018 | } | |
13019 | ||
13020 | int Client::ll_releasedir(dir_result_t *dirp) | |
13021 | { | |
11fdf7f2 | 13022 | std::lock_guard lock(client_lock); |
7c673cae FG |
13023 | ldout(cct, 3) << "ll_releasedir " << dirp << dendl; |
13024 | tout(cct) << "ll_releasedir" << std::endl; | |
13025 | tout(cct) << (unsigned long)dirp << std::endl; | |
181888fb FG |
13026 | |
13027 | if (unmounting) | |
13028 | return -ENOTCONN; | |
13029 | ||
7c673cae FG |
13030 | _closedir(dirp); |
13031 | return 0; | |
13032 | } | |
13033 | ||
13034 | int Client::ll_fsyncdir(dir_result_t *dirp) | |
13035 | { | |
11fdf7f2 | 13036 | std::lock_guard lock(client_lock); |
7c673cae FG |
13037 | ldout(cct, 3) << "ll_fsyncdir " << dirp << dendl; |
13038 | tout(cct) << "ll_fsyncdir" << std::endl; | |
13039 | tout(cct) << (unsigned long)dirp << std::endl; | |
13040 | ||
181888fb FG |
13041 | if (unmounting) |
13042 | return -ENOTCONN; | |
13043 | ||
7c673cae FG |
13044 | return _fsync(dirp->inode.get(), false); |
13045 | } | |
13046 | ||
13047 | int Client::ll_open(Inode *in, int flags, Fh **fhp, const UserPerm& perms) | |
13048 | { | |
11fdf7f2 | 13049 | ceph_assert(!(flags & O_CREAT)); |
7c673cae | 13050 | |
11fdf7f2 | 13051 | std::lock_guard lock(client_lock); |
7c673cae | 13052 | |
181888fb FG |
13053 | if (unmounting) |
13054 | return -ENOTCONN; | |
13055 | ||
7c673cae FG |
13056 | vinodeno_t vino = _get_vino(in); |
13057 | ||
13058 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << dendl; | |
13059 | tout(cct) << "ll_open" << std::endl; | |
13060 | tout(cct) << vino.ino.val << std::endl; | |
13061 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
13062 | ||
13063 | int r; | |
11fdf7f2 | 13064 | if (!fuse_default_permissions) { |
7c673cae FG |
13065 | r = may_open(in, flags, perms); |
13066 | if (r < 0) | |
13067 | goto out; | |
13068 | } | |
13069 | ||
13070 | r = _open(in, flags, 0, fhp /* may be NULL */, perms); | |
13071 | ||
13072 | out: | |
13073 | Fh *fhptr = fhp ? *fhp : NULL; | |
13074 | if (fhptr) { | |
13075 | ll_unclosed_fh_set.insert(fhptr); | |
13076 | } | |
13077 | tout(cct) << (unsigned long)fhptr << std::endl; | |
13078 | ldout(cct, 3) << "ll_open " << vino << " " << ceph_flags_sys2wire(flags) << | |
13079 | " = " << r << " (" << fhptr << ")" << dendl; | |
13080 | return r; | |
13081 | } | |
13082 | ||
13083 | int Client::_ll_create(Inode *parent, const char *name, mode_t mode, | |
13084 | int flags, InodeRef *in, int caps, Fh **fhp, | |
13085 | const UserPerm& perms) | |
13086 | { | |
13087 | *fhp = NULL; | |
13088 | ||
13089 | vinodeno_t vparent = _get_vino(parent); | |
13090 | ||
1adf2230 | 13091 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
13092 | mode << dec << " " << ceph_flags_sys2wire(flags) << ", uid " << perms.uid() |
13093 | << ", gid " << perms.gid() << dendl; | |
13094 | tout(cct) << "ll_create" << std::endl; | |
13095 | tout(cct) << vparent.ino.val << std::endl; | |
13096 | tout(cct) << name << std::endl; | |
13097 | tout(cct) << mode << std::endl; | |
13098 | tout(cct) << ceph_flags_sys2wire(flags) << std::endl; | |
13099 | ||
13100 | bool created = false; | |
13101 | int r = _lookup(parent, name, caps, in, perms); | |
13102 | ||
13103 | if (r == 0 && (flags & O_CREAT) && (flags & O_EXCL)) | |
13104 | return -EEXIST; | |
13105 | ||
13106 | if (r == -ENOENT && (flags & O_CREAT)) { | |
11fdf7f2 | 13107 | if (!fuse_default_permissions) { |
7c673cae FG |
13108 | r = may_create(parent, perms); |
13109 | if (r < 0) | |
13110 | goto out; | |
13111 | } | |
13112 | r = _create(parent, name, flags, mode, in, fhp, 0, 0, 0, NULL, &created, | |
13113 | perms); | |
13114 | if (r < 0) | |
13115 | goto out; | |
13116 | } | |
13117 | ||
13118 | if (r < 0) | |
13119 | goto out; | |
13120 | ||
11fdf7f2 | 13121 | ceph_assert(*in); |
7c673cae FG |
13122 | |
13123 | ldout(cct, 20) << "_ll_create created = " << created << dendl; | |
13124 | if (!created) { | |
11fdf7f2 | 13125 | if (!fuse_default_permissions) { |
7c673cae FG |
13126 | r = may_open(in->get(), flags, perms); |
13127 | if (r < 0) { | |
13128 | if (*fhp) { | |
13129 | int release_r = _release_fh(*fhp); | |
11fdf7f2 | 13130 | ceph_assert(release_r == 0); // during create, no async data ops should have happened |
7c673cae FG |
13131 | } |
13132 | goto out; | |
13133 | } | |
13134 | } | |
13135 | if (*fhp == NULL) { | |
13136 | r = _open(in->get(), flags, mode, fhp, perms); | |
13137 | if (r < 0) | |
13138 | goto out; | |
13139 | } | |
13140 | } | |
13141 | ||
13142 | out: | |
13143 | if (*fhp) { | |
13144 | ll_unclosed_fh_set.insert(*fhp); | |
13145 | } | |
13146 | ||
13147 | ino_t ino = 0; | |
13148 | if (r >= 0) { | |
13149 | Inode *inode = in->get(); | |
13150 | if (use_faked_inos()) | |
13151 | ino = inode->faked_ino; | |
13152 | else | |
13153 | ino = inode->ino; | |
13154 | } | |
13155 | ||
13156 | tout(cct) << (unsigned long)*fhp << std::endl; | |
13157 | tout(cct) << ino << std::endl; | |
1adf2230 | 13158 | ldout(cct, 8) << "_ll_create " << vparent << " " << name << " 0" << oct << |
7c673cae FG |
13159 | mode << dec << " " << ceph_flags_sys2wire(flags) << " = " << r << " (" << |
13160 | *fhp << " " << hex << ino << dec << ")" << dendl; | |
13161 | ||
13162 | return r; | |
13163 | } | |
13164 | ||
13165 | int Client::ll_create(Inode *parent, const char *name, mode_t mode, | |
13166 | int flags, struct stat *attr, Inode **outp, Fh **fhp, | |
13167 | const UserPerm& perms) | |
13168 | { | |
11fdf7f2 | 13169 | std::lock_guard lock(client_lock); |
7c673cae FG |
13170 | InodeRef in; |
13171 | ||
181888fb FG |
13172 | if (unmounting) |
13173 | return -ENOTCONN; | |
13174 | ||
7c673cae FG |
13175 | int r = _ll_create(parent, name, mode, flags, &in, CEPH_STAT_CAP_INODE_ALL, |
13176 | fhp, perms); | |
13177 | if (r >= 0) { | |
11fdf7f2 | 13178 | ceph_assert(in); |
7c673cae FG |
13179 | |
13180 | // passing an Inode in outp requires an additional ref | |
13181 | if (outp) { | |
13182 | _ll_get(in.get()); | |
13183 | *outp = in.get(); | |
13184 | } | |
13185 | fill_stat(in, attr); | |
13186 | } else { | |
13187 | attr->st_ino = 0; | |
13188 | } | |
13189 | ||
13190 | return r; | |
13191 | } | |
13192 | ||
13193 | int Client::ll_createx(Inode *parent, const char *name, mode_t mode, | |
13194 | int oflags, Inode **outp, Fh **fhp, | |
13195 | struct ceph_statx *stx, unsigned want, unsigned lflags, | |
13196 | const UserPerm& perms) | |
13197 | { | |
13198 | unsigned caps = statx_to_mask(lflags, want); | |
11fdf7f2 | 13199 | std::lock_guard lock(client_lock); |
7c673cae FG |
13200 | InodeRef in; |
13201 | ||
181888fb FG |
13202 | if (unmounting) |
13203 | return -ENOTCONN; | |
7c673cae FG |
13204 | |
13205 | int r = _ll_create(parent, name, mode, oflags, &in, caps, fhp, perms); | |
13206 | if (r >= 0) { | |
11fdf7f2 | 13207 | ceph_assert(in); |
7c673cae FG |
13208 | |
13209 | // passing an Inode in outp requires an additional ref | |
13210 | if (outp) { | |
13211 | _ll_get(in.get()); | |
13212 | *outp = in.get(); | |
13213 | } | |
13214 | fill_statx(in, caps, stx); | |
13215 | } else { | |
13216 | stx->stx_ino = 0; | |
13217 | stx->stx_mask = 0; | |
13218 | } | |
13219 | ||
13220 | return r; | |
13221 | } | |
13222 | ||
13223 | loff_t Client::ll_lseek(Fh *fh, loff_t offset, int whence) | |
13224 | { | |
11fdf7f2 | 13225 | std::lock_guard lock(client_lock); |
7c673cae FG |
13226 | tout(cct) << "ll_lseek" << std::endl; |
13227 | tout(cct) << offset << std::endl; | |
13228 | tout(cct) << whence << std::endl; | |
13229 | ||
181888fb FG |
13230 | if (unmounting) |
13231 | return -ENOTCONN; | |
13232 | ||
7c673cae FG |
13233 | return _lseek(fh, offset, whence); |
13234 | } | |
13235 | ||
13236 | int Client::ll_read(Fh *fh, loff_t off, loff_t len, bufferlist *bl) | |
13237 | { | |
11fdf7f2 | 13238 | std::lock_guard lock(client_lock); |
7c673cae FG |
13239 | ldout(cct, 3) << "ll_read " << fh << " " << fh->inode->ino << " " << " " << off << "~" << len << dendl; |
13240 | tout(cct) << "ll_read" << std::endl; | |
13241 | tout(cct) << (unsigned long)fh << std::endl; | |
13242 | tout(cct) << off << std::endl; | |
13243 | tout(cct) << len << std::endl; | |
13244 | ||
181888fb FG |
13245 | if (unmounting) |
13246 | return -ENOTCONN; | |
13247 | ||
11fdf7f2 TL |
13248 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
13249 | len = std::min(len, (loff_t)INT_MAX); | |
7c673cae FG |
13250 | return _read(fh, off, len, bl); |
13251 | } | |
13252 | ||
13253 | int Client::ll_read_block(Inode *in, uint64_t blockid, | |
13254 | char *buf, | |
13255 | uint64_t offset, | |
13256 | uint64_t length, | |
13257 | file_layout_t* layout) | |
13258 | { | |
11fdf7f2 | 13259 | std::lock_guard lock(client_lock); |
181888fb FG |
13260 | |
13261 | if (unmounting) | |
13262 | return -ENOTCONN; | |
13263 | ||
b32b8144 | 13264 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
13265 | object_t oid = file_object_t(vino.ino, blockid); |
13266 | C_SaferCond onfinish; | |
13267 | bufferlist bl; | |
13268 | ||
13269 | objecter->read(oid, | |
13270 | object_locator_t(layout->pool_id), | |
13271 | offset, | |
13272 | length, | |
13273 | vino.snapid, | |
13274 | &bl, | |
13275 | CEPH_OSD_FLAG_READ, | |
13276 | &onfinish); | |
13277 | ||
9f95a23c | 13278 | client_lock.unlock(); |
7c673cae | 13279 | int r = onfinish.wait(); |
9f95a23c | 13280 | client_lock.lock(); |
7c673cae FG |
13281 | |
13282 | if (r >= 0) { | |
9f95a23c | 13283 | bl.begin().copy(bl.length(), buf); |
7c673cae FG |
13284 | r = bl.length(); |
13285 | } | |
13286 | ||
13287 | return r; | |
13288 | } | |
13289 | ||
13290 | /* It appears that the OSD doesn't return success unless the entire | |
13291 | buffer was written, return the write length on success. */ | |
13292 | ||
13293 | int Client::ll_write_block(Inode *in, uint64_t blockid, | |
13294 | char* buf, uint64_t offset, | |
13295 | uint64_t length, file_layout_t* layout, | |
13296 | uint64_t snapseq, uint32_t sync) | |
13297 | { | |
7c673cae | 13298 | vinodeno_t vino = ll_get_vino(in); |
7c673cae | 13299 | int r = 0; |
11fdf7f2 TL |
13300 | std::unique_ptr<C_SaferCond> onsafe = nullptr; |
13301 | ||
7c673cae FG |
13302 | if (length == 0) { |
13303 | return -EINVAL; | |
13304 | } | |
13305 | if (true || sync) { | |
13306 | /* if write is stable, the epilogue is waiting on | |
13307 | * flock */ | |
11fdf7f2 | 13308 | onsafe.reset(new C_SaferCond("Client::ll_write_block flock")); |
7c673cae FG |
13309 | } |
13310 | object_t oid = file_object_t(vino.ino, blockid); | |
13311 | SnapContext fakesnap; | |
11fdf7f2 TL |
13312 | ceph::bufferlist bl; |
13313 | if (length > 0) { | |
13314 | bl.push_back(buffer::copy(buf, length)); | |
13315 | } | |
7c673cae FG |
13316 | |
13317 | ldout(cct, 1) << "ll_block_write for " << vino.ino << "." << blockid | |
13318 | << dendl; | |
13319 | ||
13320 | fakesnap.seq = snapseq; | |
13321 | ||
13322 | /* lock just in time */ | |
9f95a23c | 13323 | client_lock.lock(); |
181888fb | 13324 | if (unmounting) { |
9f95a23c | 13325 | client_lock.unlock(); |
181888fb FG |
13326 | return -ENOTCONN; |
13327 | } | |
7c673cae FG |
13328 | |
13329 | objecter->write(oid, | |
13330 | object_locator_t(layout->pool_id), | |
13331 | offset, | |
13332 | length, | |
13333 | fakesnap, | |
13334 | bl, | |
13335 | ceph::real_clock::now(), | |
13336 | 0, | |
11fdf7f2 | 13337 | onsafe.get()); |
7c673cae | 13338 | |
9f95a23c | 13339 | client_lock.unlock(); |
11fdf7f2 TL |
13340 | if (nullptr != onsafe) { |
13341 | r = onsafe->wait(); | |
7c673cae FG |
13342 | } |
13343 | ||
13344 | if (r < 0) { | |
13345 | return r; | |
13346 | } else { | |
13347 | return length; | |
13348 | } | |
13349 | } | |
13350 | ||
13351 | int Client::ll_commit_blocks(Inode *in, | |
13352 | uint64_t offset, | |
13353 | uint64_t length) | |
13354 | { | |
11fdf7f2 | 13355 | std::lock_guard lock(client_lock); |
7c673cae FG |
13356 | /* |
13357 | BarrierContext *bctx; | |
b32b8144 | 13358 | vinodeno_t vino = _get_vino(in); |
7c673cae FG |
13359 | uint64_t ino = vino.ino; |
13360 | ||
13361 | ldout(cct, 1) << "ll_commit_blocks for " << vino.ino << " from " | |
13362 | << offset << " to " << length << dendl; | |
13363 | ||
13364 | if (length == 0) { | |
13365 | return -EINVAL; | |
13366 | } | |
13367 | ||
13368 | map<uint64_t, BarrierContext*>::iterator p = barriers.find(ino); | |
13369 | if (p != barriers.end()) { | |
13370 | barrier_interval civ(offset, offset + length); | |
13371 | p->second->commit_barrier(civ); | |
13372 | } | |
13373 | */ | |
13374 | return 0; | |
13375 | } | |
13376 | ||
13377 | int Client::ll_write(Fh *fh, loff_t off, loff_t len, const char *data) | |
13378 | { | |
11fdf7f2 | 13379 | std::lock_guard lock(client_lock); |
7c673cae FG |
13380 | ldout(cct, 3) << "ll_write " << fh << " " << fh->inode->ino << " " << off << |
13381 | "~" << len << dendl; | |
13382 | tout(cct) << "ll_write" << std::endl; | |
13383 | tout(cct) << (unsigned long)fh << std::endl; | |
13384 | tout(cct) << off << std::endl; | |
13385 | tout(cct) << len << std::endl; | |
13386 | ||
181888fb FG |
13387 | if (unmounting) |
13388 | return -ENOTCONN; | |
13389 | ||
11fdf7f2 TL |
13390 | /* We can't return bytes written larger than INT_MAX, clamp len to that */ |
13391 | len = std::min(len, (loff_t)INT_MAX); | |
7c673cae FG |
13392 | int r = _write(fh, off, len, data, NULL, 0); |
13393 | ldout(cct, 3) << "ll_write " << fh << " " << off << "~" << len << " = " << r | |
13394 | << dendl; | |
13395 | return r; | |
13396 | } | |
13397 | ||
11fdf7f2 TL |
13398 | int64_t Client::ll_writev(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) |
13399 | { | |
13400 | std::lock_guard lock(client_lock); | |
13401 | if (unmounting) | |
13402 | return -ENOTCONN; | |
13403 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, true, false); | |
13404 | } | |
13405 | ||
13406 | int64_t Client::ll_readv(struct Fh *fh, const struct iovec *iov, int iovcnt, int64_t off) | |
13407 | { | |
13408 | std::lock_guard lock(client_lock); | |
13409 | if (unmounting) | |
13410 | return -ENOTCONN; | |
13411 | return _preadv_pwritev_locked(fh, iov, iovcnt, off, false, false); | |
13412 | } | |
13413 | ||
7c673cae FG |
13414 | int Client::ll_flush(Fh *fh) |
13415 | { | |
11fdf7f2 | 13416 | std::lock_guard lock(client_lock); |
7c673cae FG |
13417 | ldout(cct, 3) << "ll_flush " << fh << " " << fh->inode->ino << " " << dendl; |
13418 | tout(cct) << "ll_flush" << std::endl; | |
13419 | tout(cct) << (unsigned long)fh << std::endl; | |
13420 | ||
181888fb FG |
13421 | if (unmounting) |
13422 | return -ENOTCONN; | |
13423 | ||
7c673cae FG |
13424 | return _flush(fh); |
13425 | } | |
13426 | ||
13427 | int Client::ll_fsync(Fh *fh, bool syncdataonly) | |
13428 | { | |
11fdf7f2 | 13429 | std::lock_guard lock(client_lock); |
7c673cae FG |
13430 | ldout(cct, 3) << "ll_fsync " << fh << " " << fh->inode->ino << " " << dendl; |
13431 | tout(cct) << "ll_fsync" << std::endl; | |
13432 | tout(cct) << (unsigned long)fh << std::endl; | |
13433 | ||
181888fb FG |
13434 | if (unmounting) |
13435 | return -ENOTCONN; | |
13436 | ||
7c673cae FG |
13437 | int r = _fsync(fh, syncdataonly); |
13438 | if (r) { | |
13439 | // If we're returning an error, clear it from the FH | |
13440 | fh->take_async_err(); | |
13441 | } | |
13442 | return r; | |
13443 | } | |
13444 | ||
28e407b8 AA |
13445 | int Client::ll_sync_inode(Inode *in, bool syncdataonly) |
13446 | { | |
11fdf7f2 | 13447 | std::lock_guard lock(client_lock); |
28e407b8 AA |
13448 | ldout(cct, 3) << "ll_sync_inode " << *in << " " << dendl; |
13449 | tout(cct) << "ll_sync_inode" << std::endl; | |
13450 | tout(cct) << (unsigned long)in << std::endl; | |
13451 | ||
13452 | if (unmounting) | |
13453 | return -ENOTCONN; | |
13454 | ||
13455 | return _fsync(in, syncdataonly); | |
13456 | } | |
13457 | ||
7c673cae FG |
13458 | #ifdef FALLOC_FL_PUNCH_HOLE |
13459 | ||
13460 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) | |
13461 | { | |
13462 | if (offset < 0 || length <= 0) | |
13463 | return -EINVAL; | |
13464 | ||
13465 | if (mode & ~(FALLOC_FL_KEEP_SIZE | FALLOC_FL_PUNCH_HOLE)) | |
13466 | return -EOPNOTSUPP; | |
13467 | ||
13468 | if ((mode & FALLOC_FL_PUNCH_HOLE) && !(mode & FALLOC_FL_KEEP_SIZE)) | |
13469 | return -EOPNOTSUPP; | |
13470 | ||
13471 | Inode *in = fh->inode.get(); | |
13472 | ||
13473 | if (objecter->osdmap_pool_full(in->layout.pool_id) && | |
13474 | !(mode & FALLOC_FL_PUNCH_HOLE)) { | |
13475 | return -ENOSPC; | |
13476 | } | |
13477 | ||
13478 | if (in->snapid != CEPH_NOSNAP) | |
13479 | return -EROFS; | |
13480 | ||
13481 | if ((fh->mode & CEPH_FILE_MODE_WR) == 0) | |
13482 | return -EBADF; | |
13483 | ||
13484 | uint64_t size = offset + length; | |
13485 | if (!(mode & (FALLOC_FL_PUNCH_HOLE | FALLOC_FL_KEEP_SIZE)) && | |
13486 | size > in->size && | |
11fdf7f2 | 13487 | is_quota_bytes_exceeded(in, size - in->size, fh->actor_perms)) { |
7c673cae FG |
13488 | return -EDQUOT; |
13489 | } | |
13490 | ||
13491 | int have; | |
13492 | int r = get_caps(in, CEPH_CAP_FILE_WR, CEPH_CAP_FILE_BUFFER, &have, -1); | |
13493 | if (r < 0) | |
13494 | return r; | |
13495 | ||
11fdf7f2 | 13496 | std::unique_ptr<C_SaferCond> onuninline = nullptr; |
7c673cae FG |
13497 | if (mode & FALLOC_FL_PUNCH_HOLE) { |
13498 | if (in->inline_version < CEPH_INLINE_NONE && | |
13499 | (have & CEPH_CAP_FILE_BUFFER)) { | |
13500 | bufferlist bl; | |
9f95a23c | 13501 | auto inline_iter = in->inline_data.cbegin(); |
7c673cae FG |
13502 | int len = in->inline_data.length(); |
13503 | if (offset < len) { | |
13504 | if (offset > 0) | |
9f95a23c | 13505 | inline_iter.copy(offset, bl); |
7c673cae FG |
13506 | int size = length; |
13507 | if (offset + size > len) | |
13508 | size = len - offset; | |
13509 | if (size > 0) | |
13510 | bl.append_zero(size); | |
9f95a23c TL |
13511 | if (offset + size < len) { |
13512 | inline_iter += size; | |
13513 | inline_iter.copy(len - offset - size, bl); | |
13514 | } | |
7c673cae FG |
13515 | in->inline_data = bl; |
13516 | in->inline_version++; | |
13517 | } | |
91327a77 | 13518 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 13519 | in->change_attr++; |
28e407b8 | 13520 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
13521 | } else { |
13522 | if (in->inline_version < CEPH_INLINE_NONE) { | |
11fdf7f2 TL |
13523 | onuninline.reset(new C_SaferCond("Client::_fallocate_uninline_data flock")); |
13524 | uninline_data(in, onuninline.get()); | |
7c673cae FG |
13525 | } |
13526 | ||
11fdf7f2 | 13527 | C_SaferCond onfinish("Client::_punch_hole flock"); |
7c673cae FG |
13528 | |
13529 | unsafe_sync_write++; | |
13530 | get_cap_ref(in, CEPH_CAP_FILE_BUFFER); | |
13531 | ||
13532 | _invalidate_inode_cache(in, offset, length); | |
13533 | filer->zero(in->ino, &in->layout, | |
13534 | in->snaprealm->get_snap_context(), | |
13535 | offset, length, | |
13536 | ceph::real_clock::now(), | |
11fdf7f2 | 13537 | 0, true, &onfinish); |
91327a77 | 13538 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 13539 | in->change_attr++; |
28e407b8 | 13540 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 13541 | |
9f95a23c | 13542 | client_lock.unlock(); |
11fdf7f2 | 13543 | onfinish.wait(); |
9f95a23c | 13544 | client_lock.lock(); |
7c673cae FG |
13545 | _sync_write_commit(in); |
13546 | } | |
13547 | } else if (!(mode & FALLOC_FL_KEEP_SIZE)) { | |
13548 | uint64_t size = offset + length; | |
13549 | if (size > in->size) { | |
13550 | in->size = size; | |
91327a77 | 13551 | in->mtime = in->ctime = ceph_clock_now(); |
7c673cae | 13552 | in->change_attr++; |
28e407b8 | 13553 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae | 13554 | |
11fdf7f2 | 13555 | if (is_quota_bytes_approaching(in, fh->actor_perms)) { |
7c673cae | 13556 | check_caps(in, CHECK_CAPS_NODELAY); |
31f18b77 FG |
13557 | } else if (is_max_size_approaching(in)) { |
13558 | check_caps(in, 0); | |
7c673cae FG |
13559 | } |
13560 | } | |
13561 | } | |
13562 | ||
11fdf7f2 | 13563 | if (nullptr != onuninline) { |
9f95a23c | 13564 | client_lock.unlock(); |
11fdf7f2 | 13565 | int ret = onuninline->wait(); |
9f95a23c | 13566 | client_lock.lock(); |
7c673cae | 13567 | |
11fdf7f2 | 13568 | if (ret >= 0 || ret == -ECANCELED) { |
7c673cae FG |
13569 | in->inline_data.clear(); |
13570 | in->inline_version = CEPH_INLINE_NONE; | |
28e407b8 | 13571 | in->mark_caps_dirty(CEPH_CAP_FILE_WR); |
7c673cae FG |
13572 | check_caps(in, 0); |
13573 | } else | |
11fdf7f2 | 13574 | r = ret; |
7c673cae FG |
13575 | } |
13576 | ||
13577 | put_cap_ref(in, CEPH_CAP_FILE_WR); | |
13578 | return r; | |
13579 | } | |
13580 | #else | |
13581 | ||
13582 | int Client::_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) | |
13583 | { | |
13584 | return -EOPNOTSUPP; | |
13585 | } | |
13586 | ||
13587 | #endif | |
13588 | ||
13589 | ||
11fdf7f2 | 13590 | int Client::ll_fallocate(Fh *fh, int mode, int64_t offset, int64_t length) |
7c673cae | 13591 | { |
11fdf7f2 TL |
13592 | std::lock_guard lock(client_lock); |
13593 | ldout(cct, 3) << __func__ << " " << fh << " " << fh->inode->ino << " " << dendl; | |
13594 | tout(cct) << __func__ << " " << mode << " " << offset << " " << length << std::endl; | |
7c673cae FG |
13595 | tout(cct) << (unsigned long)fh << std::endl; |
13596 | ||
181888fb FG |
13597 | if (unmounting) |
13598 | return -ENOTCONN; | |
13599 | ||
7c673cae FG |
13600 | return _fallocate(fh, mode, offset, length); |
13601 | } | |
13602 | ||
13603 | int Client::fallocate(int fd, int mode, loff_t offset, loff_t length) | |
13604 | { | |
11fdf7f2 TL |
13605 | std::lock_guard lock(client_lock); |
13606 | tout(cct) << __func__ << " " << " " << fd << mode << " " << offset << " " << length << std::endl; | |
7c673cae | 13607 | |
181888fb FG |
13608 | if (unmounting) |
13609 | return -ENOTCONN; | |
13610 | ||
7c673cae FG |
13611 | Fh *fh = get_filehandle(fd); |
13612 | if (!fh) | |
13613 | return -EBADF; | |
13614 | #if defined(__linux__) && defined(O_PATH) | |
13615 | if (fh->flags & O_PATH) | |
13616 | return -EBADF; | |
13617 | #endif | |
13618 | return _fallocate(fh, mode, offset, length); | |
13619 | } | |
13620 | ||
13621 | int Client::ll_release(Fh *fh) | |
13622 | { | |
11fdf7f2 | 13623 | std::lock_guard lock(client_lock); |
91327a77 AA |
13624 | |
13625 | if (unmounting) | |
13626 | return -ENOTCONN; | |
13627 | ||
11fdf7f2 | 13628 | ldout(cct, 3) << __func__ << " (fh)" << fh << " " << fh->inode->ino << " " << |
7c673cae | 13629 | dendl; |
11fdf7f2 | 13630 | tout(cct) << __func__ << " (fh)" << std::endl; |
7c673cae FG |
13631 | tout(cct) << (unsigned long)fh << std::endl; |
13632 | ||
13633 | if (ll_unclosed_fh_set.count(fh)) | |
13634 | ll_unclosed_fh_set.erase(fh); | |
13635 | return _release_fh(fh); | |
13636 | } | |
13637 | ||
13638 | int Client::ll_getlk(Fh *fh, struct flock *fl, uint64_t owner) | |
13639 | { | |
11fdf7f2 | 13640 | std::lock_guard lock(client_lock); |
7c673cae FG |
13641 | |
13642 | ldout(cct, 3) << "ll_getlk (fh)" << fh << " " << fh->inode->ino << dendl; | |
13643 | tout(cct) << "ll_getk (fh)" << (unsigned long)fh << std::endl; | |
13644 | ||
181888fb FG |
13645 | if (unmounting) |
13646 | return -ENOTCONN; | |
13647 | ||
7c673cae FG |
13648 | return _getlk(fh, fl, owner); |
13649 | } | |
13650 | ||
13651 | int Client::ll_setlk(Fh *fh, struct flock *fl, uint64_t owner, int sleep) | |
13652 | { | |
11fdf7f2 | 13653 | std::lock_guard lock(client_lock); |
7c673cae | 13654 | |
11fdf7f2 TL |
13655 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
13656 | tout(cct) << __func__ << " (fh)" << (unsigned long)fh << std::endl; | |
7c673cae | 13657 | |
181888fb FG |
13658 | if (unmounting) |
13659 | return -ENOTCONN; | |
13660 | ||
7c673cae FG |
13661 | return _setlk(fh, fl, owner, sleep); |
13662 | } | |
13663 | ||
13664 | int Client::ll_flock(Fh *fh, int cmd, uint64_t owner) | |
13665 | { | |
11fdf7f2 | 13666 | std::lock_guard lock(client_lock); |
7c673cae | 13667 | |
11fdf7f2 TL |
13668 | ldout(cct, 3) << __func__ << " (fh) " << fh << " " << fh->inode->ino << dendl; |
13669 | tout(cct) << __func__ << " (fh)" << (unsigned long)fh << std::endl; | |
7c673cae | 13670 | |
181888fb FG |
13671 | if (unmounting) |
13672 | return -ENOTCONN; | |
13673 | ||
7c673cae FG |
13674 | return _flock(fh, cmd, owner); |
13675 | } | |
13676 | ||
b32b8144 FG |
13677 | int Client::set_deleg_timeout(uint32_t timeout) |
13678 | { | |
11fdf7f2 | 13679 | std::lock_guard lock(client_lock); |
b32b8144 FG |
13680 | |
13681 | /* | |
13682 | * The whole point is to prevent blacklisting so we must time out the | |
13683 | * delegation before the session autoclose timeout kicks in. | |
13684 | */ | |
13685 | if (timeout >= mdsmap->get_session_autoclose()) | |
13686 | return -EINVAL; | |
13687 | ||
13688 | deleg_timeout = timeout; | |
13689 | return 0; | |
13690 | } | |
13691 | ||
13692 | int Client::ll_delegation(Fh *fh, unsigned cmd, ceph_deleg_cb_t cb, void *priv) | |
13693 | { | |
13694 | int ret = -EINVAL; | |
13695 | ||
11fdf7f2 | 13696 | std::lock_guard lock(client_lock); |
b32b8144 FG |
13697 | |
13698 | if (!mounted) | |
13699 | return -ENOTCONN; | |
13700 | ||
13701 | Inode *inode = fh->inode.get(); | |
13702 | ||
13703 | switch(cmd) { | |
13704 | case CEPH_DELEGATION_NONE: | |
13705 | inode->unset_deleg(fh); | |
13706 | ret = 0; | |
13707 | break; | |
13708 | default: | |
13709 | try { | |
13710 | ret = inode->set_deleg(fh, cmd, cb, priv); | |
11fdf7f2 | 13711 | } catch (std::bad_alloc&) { |
b32b8144 FG |
13712 | ret = -ENOMEM; |
13713 | } | |
13714 | break; | |
13715 | } | |
13716 | return ret; | |
13717 | } | |
13718 | ||
7c673cae FG |
13719 | class C_Client_RequestInterrupt : public Context { |
13720 | private: | |
13721 | Client *client; | |
13722 | MetaRequest *req; | |
13723 | public: | |
13724 | C_Client_RequestInterrupt(Client *c, MetaRequest *r) : client(c), req(r) { | |
13725 | req->get(); | |
13726 | } | |
13727 | void finish(int r) override { | |
11fdf7f2 TL |
13728 | std::lock_guard l(client->client_lock); |
13729 | ceph_assert(req->head.op == CEPH_MDS_OP_SETFILELOCK); | |
7c673cae FG |
13730 | client->_interrupt_filelock(req); |
13731 | client->put_request(req); | |
13732 | } | |
13733 | }; | |
13734 | ||
13735 | void Client::ll_interrupt(void *d) | |
13736 | { | |
13737 | MetaRequest *req = static_cast<MetaRequest*>(d); | |
11fdf7f2 TL |
13738 | ldout(cct, 3) << __func__ << " tid " << req->get_tid() << dendl; |
13739 | tout(cct) << __func__ << " tid " << req->get_tid() << std::endl; | |
7c673cae FG |
13740 | interrupt_finisher.queue(new C_Client_RequestInterrupt(this, req)); |
13741 | } | |
13742 | ||
13743 | // ========================================= | |
13744 | // layout | |
13745 | ||
13746 | // expose file layouts | |
13747 | ||
13748 | int Client::describe_layout(const char *relpath, file_layout_t *lp, | |
13749 | const UserPerm& perms) | |
13750 | { | |
11fdf7f2 | 13751 | std::lock_guard lock(client_lock); |
7c673cae | 13752 | |
181888fb FG |
13753 | if (unmounting) |
13754 | return -ENOTCONN; | |
13755 | ||
7c673cae FG |
13756 | filepath path(relpath); |
13757 | InodeRef in; | |
13758 | int r = path_walk(path, &in, perms); | |
13759 | if (r < 0) | |
13760 | return r; | |
13761 | ||
13762 | *lp = in->layout; | |
13763 | ||
11fdf7f2 | 13764 | ldout(cct, 3) << __func__ << "(" << relpath << ") = 0" << dendl; |
7c673cae FG |
13765 | return 0; |
13766 | } | |
13767 | ||
13768 | int Client::fdescribe_layout(int fd, file_layout_t *lp) | |
13769 | { | |
11fdf7f2 | 13770 | std::lock_guard lock(client_lock); |
7c673cae | 13771 | |
181888fb FG |
13772 | if (unmounting) |
13773 | return -ENOTCONN; | |
13774 | ||
7c673cae FG |
13775 | Fh *f = get_filehandle(fd); |
13776 | if (!f) | |
13777 | return -EBADF; | |
13778 | Inode *in = f->inode.get(); | |
13779 | ||
13780 | *lp = in->layout; | |
13781 | ||
11fdf7f2 | 13782 | ldout(cct, 3) << __func__ << "(" << fd << ") = 0" << dendl; |
7c673cae FG |
13783 | return 0; |
13784 | } | |
13785 | ||
d2e6a577 FG |
13786 | int64_t Client::get_default_pool_id() |
13787 | { | |
11fdf7f2 | 13788 | std::lock_guard lock(client_lock); |
181888fb FG |
13789 | |
13790 | if (unmounting) | |
13791 | return -ENOTCONN; | |
13792 | ||
d2e6a577 FG |
13793 | /* first data pool is the default */ |
13794 | return mdsmap->get_first_data_pool(); | |
13795 | } | |
7c673cae FG |
13796 | |
13797 | // expose osdmap | |
13798 | ||
13799 | int64_t Client::get_pool_id(const char *pool_name) | |
13800 | { | |
11fdf7f2 | 13801 | std::lock_guard lock(client_lock); |
181888fb FG |
13802 | |
13803 | if (unmounting) | |
13804 | return -ENOTCONN; | |
13805 | ||
7c673cae FG |
13806 | return objecter->with_osdmap(std::mem_fn(&OSDMap::lookup_pg_pool_name), |
13807 | pool_name); | |
13808 | } | |
13809 | ||
13810 | string Client::get_pool_name(int64_t pool) | |
13811 | { | |
11fdf7f2 | 13812 | std::lock_guard lock(client_lock); |
181888fb FG |
13813 | |
13814 | if (unmounting) | |
13815 | return string(); | |
13816 | ||
7c673cae FG |
13817 | return objecter->with_osdmap([pool](const OSDMap& o) { |
13818 | return o.have_pg_pool(pool) ? o.get_pool_name(pool) : string(); | |
13819 | }); | |
13820 | } | |
13821 | ||
13822 | int Client::get_pool_replication(int64_t pool) | |
13823 | { | |
11fdf7f2 | 13824 | std::lock_guard lock(client_lock); |
181888fb FG |
13825 | |
13826 | if (unmounting) | |
13827 | return -ENOTCONN; | |
13828 | ||
7c673cae FG |
13829 | return objecter->with_osdmap([pool](const OSDMap& o) { |
13830 | return o.have_pg_pool(pool) ? o.get_pg_pool(pool)->get_size() : -ENOENT; | |
13831 | }); | |
13832 | } | |
13833 | ||
13834 | int Client::get_file_extent_osds(int fd, loff_t off, loff_t *len, vector<int>& osds) | |
13835 | { | |
11fdf7f2 | 13836 | std::lock_guard lock(client_lock); |
7c673cae | 13837 | |
181888fb FG |
13838 | if (unmounting) |
13839 | return -ENOTCONN; | |
13840 | ||
7c673cae FG |
13841 | Fh *f = get_filehandle(fd); |
13842 | if (!f) | |
13843 | return -EBADF; | |
13844 | Inode *in = f->inode.get(); | |
13845 | ||
13846 | vector<ObjectExtent> extents; | |
13847 | Striper::file_to_extents(cct, in->ino, &in->layout, off, 1, in->truncate_size, extents); | |
11fdf7f2 | 13848 | ceph_assert(extents.size() == 1); |
7c673cae FG |
13849 | |
13850 | objecter->with_osdmap([&](const OSDMap& o) { | |
13851 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
13852 | o.pg_to_acting_osds(pg, osds); | |
13853 | }); | |
13854 | ||
13855 | if (osds.empty()) | |
13856 | return -EINVAL; | |
13857 | ||
13858 | /* | |
13859 | * Return the remainder of the extent (stripe unit) | |
13860 | * | |
13861 | * If length = 1 is passed to Striper::file_to_extents we get a single | |
13862 | * extent back, but its length is one so we still need to compute the length | |
13863 | * to the end of the stripe unit. | |
13864 | * | |
13865 | * If length = su then we may get 1 or 2 objects back in the extents vector | |
13866 | * which would have to be examined. Even then, the offsets are local to the | |
13867 | * object, so matching up to the file offset is extra work. | |
13868 | * | |
13869 | * It seems simpler to stick with length = 1 and manually compute the | |
13870 | * remainder. | |
13871 | */ | |
13872 | if (len) { | |
13873 | uint64_t su = in->layout.stripe_unit; | |
13874 | *len = su - (off % su); | |
13875 | } | |
13876 | ||
13877 | return 0; | |
13878 | } | |
13879 | ||
13880 | int Client::get_osd_crush_location(int id, vector<pair<string, string> >& path) | |
13881 | { | |
11fdf7f2 | 13882 | std::lock_guard lock(client_lock); |
181888fb FG |
13883 | |
13884 | if (unmounting) | |
13885 | return -ENOTCONN; | |
13886 | ||
7c673cae FG |
13887 | if (id < 0) |
13888 | return -EINVAL; | |
13889 | return objecter->with_osdmap([&](const OSDMap& o) { | |
13890 | return o.crush->get_full_location_ordered(id, path); | |
13891 | }); | |
13892 | } | |
13893 | ||
13894 | int Client::get_file_stripe_address(int fd, loff_t offset, | |
13895 | vector<entity_addr_t>& address) | |
13896 | { | |
11fdf7f2 | 13897 | std::lock_guard lock(client_lock); |
7c673cae | 13898 | |
181888fb FG |
13899 | if (unmounting) |
13900 | return -ENOTCONN; | |
13901 | ||
7c673cae FG |
13902 | Fh *f = get_filehandle(fd); |
13903 | if (!f) | |
13904 | return -EBADF; | |
13905 | Inode *in = f->inode.get(); | |
13906 | ||
13907 | // which object? | |
13908 | vector<ObjectExtent> extents; | |
13909 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, 1, | |
13910 | in->truncate_size, extents); | |
11fdf7f2 | 13911 | ceph_assert(extents.size() == 1); |
7c673cae FG |
13912 | |
13913 | // now we have the object and its 'layout' | |
13914 | return objecter->with_osdmap([&](const OSDMap& o) { | |
13915 | pg_t pg = o.object_locator_to_pg(extents[0].oid, extents[0].oloc); | |
13916 | vector<int> osds; | |
13917 | o.pg_to_acting_osds(pg, osds); | |
13918 | if (osds.empty()) | |
13919 | return -EINVAL; | |
13920 | for (unsigned i = 0; i < osds.size(); i++) { | |
11fdf7f2 | 13921 | entity_addr_t addr = o.get_addrs(osds[i]).front(); |
7c673cae FG |
13922 | address.push_back(addr); |
13923 | } | |
13924 | return 0; | |
13925 | }); | |
13926 | } | |
13927 | ||
13928 | int Client::get_osd_addr(int osd, entity_addr_t& addr) | |
13929 | { | |
11fdf7f2 | 13930 | std::lock_guard lock(client_lock); |
181888fb FG |
13931 | |
13932 | if (unmounting) | |
13933 | return -ENOTCONN; | |
13934 | ||
7c673cae FG |
13935 | return objecter->with_osdmap([&](const OSDMap& o) { |
13936 | if (!o.exists(osd)) | |
13937 | return -ENOENT; | |
13938 | ||
11fdf7f2 | 13939 | addr = o.get_addrs(osd).front(); |
7c673cae FG |
13940 | return 0; |
13941 | }); | |
13942 | } | |
13943 | ||
13944 | int Client::enumerate_layout(int fd, vector<ObjectExtent>& result, | |
13945 | loff_t length, loff_t offset) | |
13946 | { | |
11fdf7f2 | 13947 | std::lock_guard lock(client_lock); |
7c673cae | 13948 | |
181888fb FG |
13949 | if (unmounting) |
13950 | return -ENOTCONN; | |
13951 | ||
7c673cae FG |
13952 | Fh *f = get_filehandle(fd); |
13953 | if (!f) | |
13954 | return -EBADF; | |
13955 | Inode *in = f->inode.get(); | |
13956 | ||
13957 | // map to a list of extents | |
13958 | Striper::file_to_extents(cct, in->ino, &in->layout, offset, length, in->truncate_size, result); | |
13959 | ||
11fdf7f2 | 13960 | ldout(cct, 3) << __func__ << "(" << fd << ", " << length << ", " << offset << ") = 0" << dendl; |
7c673cae FG |
13961 | return 0; |
13962 | } | |
13963 | ||
13964 | ||
b32b8144 | 13965 | /* find an osd with the same ip. -ENXIO if none. */ |
7c673cae FG |
13966 | int Client::get_local_osd() |
13967 | { | |
11fdf7f2 | 13968 | std::lock_guard lock(client_lock); |
181888fb FG |
13969 | |
13970 | if (unmounting) | |
13971 | return -ENOTCONN; | |
13972 | ||
7c673cae FG |
13973 | objecter->with_osdmap([this](const OSDMap& o) { |
13974 | if (o.get_epoch() != local_osd_epoch) { | |
11fdf7f2 | 13975 | local_osd = o.find_osd_on_ip(messenger->get_myaddrs().front()); |
7c673cae FG |
13976 | local_osd_epoch = o.get_epoch(); |
13977 | } | |
13978 | }); | |
13979 | return local_osd; | |
13980 | } | |
13981 | ||
13982 | ||
13983 | ||
13984 | ||
13985 | ||
13986 | ||
13987 | // =============================== | |
13988 | ||
13989 | void Client::ms_handle_connect(Connection *con) | |
13990 | { | |
11fdf7f2 | 13991 | ldout(cct, 10) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
13992 | } |
13993 | ||
13994 | bool Client::ms_handle_reset(Connection *con) | |
13995 | { | |
11fdf7f2 | 13996 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
13997 | return false; |
13998 | } | |
13999 | ||
14000 | void Client::ms_handle_remote_reset(Connection *con) | |
14001 | { | |
11fdf7f2 TL |
14002 | ldout(cct, 0) << __func__ << " on " << con->get_peer_addr() << dendl; |
14003 | std::lock_guard l(client_lock); | |
7c673cae FG |
14004 | switch (con->get_peer_type()) { |
14005 | case CEPH_ENTITY_TYPE_MDS: | |
14006 | { | |
14007 | // kludge to figure out which mds this is; fixme with a Connection* state | |
14008 | mds_rank_t mds = MDS_RANK_NONE; | |
14009 | MetaSession *s = NULL; | |
11fdf7f2 TL |
14010 | for (auto &p : mds_sessions) { |
14011 | if (mdsmap->get_addrs(p.first) == con->get_peer_addrs()) { | |
14012 | mds = p.first; | |
14013 | s = &p.second; | |
7c673cae FG |
14014 | } |
14015 | } | |
14016 | if (mds >= 0) { | |
d2e6a577 | 14017 | assert (s != NULL); |
7c673cae FG |
14018 | switch (s->state) { |
14019 | case MetaSession::STATE_CLOSING: | |
14020 | ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl; | |
14021 | _closed_mds_session(s); | |
14022 | break; | |
14023 | ||
14024 | case MetaSession::STATE_OPENING: | |
14025 | { | |
14026 | ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl; | |
14027 | list<Context*> waiters; | |
14028 | waiters.swap(s->waiting_for_open); | |
14029 | _closed_mds_session(s); | |
14030 | MetaSession *news = _get_or_open_mds_session(mds); | |
14031 | news->waiting_for_open.swap(waiters); | |
14032 | } | |
14033 | break; | |
14034 | ||
14035 | case MetaSession::STATE_OPEN: | |
14036 | { | |
28e407b8 | 14037 | objecter->maybe_request_map(); /* to check if we are blacklisted */ |
11fdf7f2 | 14038 | const auto& conf = cct->_conf; |
7c673cae FG |
14039 | if (conf->client_reconnect_stale) { |
14040 | ldout(cct, 1) << "reset from mds we were open; close mds session for reconnect" << dendl; | |
14041 | _closed_mds_session(s); | |
14042 | } else { | |
14043 | ldout(cct, 1) << "reset from mds we were open; mark session as stale" << dendl; | |
14044 | s->state = MetaSession::STATE_STALE; | |
14045 | } | |
14046 | } | |
14047 | break; | |
14048 | ||
14049 | case MetaSession::STATE_NEW: | |
14050 | case MetaSession::STATE_CLOSED: | |
14051 | default: | |
14052 | break; | |
14053 | } | |
14054 | } | |
14055 | } | |
14056 | break; | |
14057 | } | |
14058 | } | |
14059 | ||
14060 | bool Client::ms_handle_refused(Connection *con) | |
14061 | { | |
11fdf7f2 | 14062 | ldout(cct, 1) << __func__ << " on " << con->get_peer_addr() << dendl; |
7c673cae FG |
14063 | return false; |
14064 | } | |
14065 | ||
7c673cae FG |
14066 | Inode *Client::get_quota_root(Inode *in, const UserPerm& perms) |
14067 | { | |
11fdf7f2 TL |
14068 | Inode *quota_in = root_ancestor; |
14069 | SnapRealm *realm = in->snaprealm; | |
14070 | while (realm) { | |
14071 | ldout(cct, 10) << __func__ << " realm " << realm->ino << dendl; | |
14072 | if (realm->ino != in->ino) { | |
14073 | auto p = inode_map.find(vinodeno_t(realm->ino, CEPH_NOSNAP)); | |
14074 | if (p == inode_map.end()) | |
14075 | break; | |
7c673cae | 14076 | |
11fdf7f2 TL |
14077 | if (p->second->quota.is_enable()) { |
14078 | quota_in = p->second; | |
14079 | break; | |
7c673cae | 14080 | } |
7c673cae | 14081 | } |
11fdf7f2 | 14082 | realm = realm->pparent; |
7c673cae | 14083 | } |
11fdf7f2 TL |
14084 | ldout(cct, 10) << __func__ << " " << in->vino() << " -> " << quota_in->vino() << dendl; |
14085 | return quota_in; | |
7c673cae FG |
14086 | } |
14087 | ||
14088 | /** | |
14089 | * Traverse quota ancestors of the Inode, return true | |
14090 | * if any of them passes the passed function | |
14091 | */ | |
14092 | bool Client::check_quota_condition(Inode *in, const UserPerm& perms, | |
14093 | std::function<bool (const Inode &in)> test) | |
14094 | { | |
14095 | while (true) { | |
11fdf7f2 | 14096 | ceph_assert(in != NULL); |
7c673cae FG |
14097 | if (test(*in)) { |
14098 | return true; | |
14099 | } | |
14100 | ||
14101 | if (in == root_ancestor) { | |
14102 | // We're done traversing, drop out | |
14103 | return false; | |
14104 | } else { | |
14105 | // Continue up the tree | |
14106 | in = get_quota_root(in, perms); | |
14107 | } | |
14108 | } | |
14109 | ||
14110 | return false; | |
14111 | } | |
14112 | ||
14113 | bool Client::is_quota_files_exceeded(Inode *in, const UserPerm& perms) | |
14114 | { | |
14115 | return check_quota_condition(in, perms, | |
14116 | [](const Inode &in) { | |
14117 | return in.quota.max_files && in.rstat.rsize() >= in.quota.max_files; | |
14118 | }); | |
14119 | } | |
14120 | ||
14121 | bool Client::is_quota_bytes_exceeded(Inode *in, int64_t new_bytes, | |
11fdf7f2 | 14122 | const UserPerm& perms) |
7c673cae FG |
14123 | { |
14124 | return check_quota_condition(in, perms, | |
11fdf7f2 | 14125 | [&new_bytes](const Inode &in) { |
7c673cae FG |
14126 | return in.quota.max_bytes && (in.rstat.rbytes + new_bytes) |
14127 | > in.quota.max_bytes; | |
14128 | }); | |
14129 | } | |
14130 | ||
11fdf7f2 | 14131 | bool Client::is_quota_bytes_approaching(Inode *in, const UserPerm& perms) |
7c673cae | 14132 | { |
9f95a23c TL |
14133 | ceph_assert(in->size >= in->reported_size); |
14134 | const uint64_t size = in->size - in->reported_size; | |
11fdf7f2 | 14135 | return check_quota_condition(in, perms, |
9f95a23c | 14136 | [&size](const Inode &in) { |
11fdf7f2 TL |
14137 | if (in.quota.max_bytes) { |
14138 | if (in.rstat.rbytes >= in.quota.max_bytes) { | |
14139 | return true; | |
14140 | } | |
14141 | ||
11fdf7f2 | 14142 | const uint64_t space = in.quota.max_bytes - in.rstat.rbytes; |
11fdf7f2 TL |
14143 | return (space >> 4) < size; |
14144 | } else { | |
14145 | return false; | |
14146 | } | |
14147 | }); | |
7c673cae FG |
14148 | } |
14149 | ||
14150 | enum { | |
14151 | POOL_CHECKED = 1, | |
14152 | POOL_CHECKING = 2, | |
14153 | POOL_READ = 4, | |
14154 | POOL_WRITE = 8, | |
14155 | }; | |
14156 | ||
14157 | int Client::check_pool_perm(Inode *in, int need) | |
14158 | { | |
14159 | if (!cct->_conf->client_check_pool_perm) | |
14160 | return 0; | |
14161 | ||
14162 | int64_t pool_id = in->layout.pool_id; | |
14163 | std::string pool_ns = in->layout.pool_ns; | |
14164 | std::pair<int64_t, std::string> perm_key(pool_id, pool_ns); | |
14165 | int have = 0; | |
14166 | while (true) { | |
14167 | auto it = pool_perms.find(perm_key); | |
14168 | if (it == pool_perms.end()) | |
14169 | break; | |
14170 | if (it->second == POOL_CHECKING) { | |
14171 | // avoid concurrent checkings | |
14172 | wait_on_list(waiting_for_pool_perm); | |
14173 | } else { | |
14174 | have = it->second; | |
11fdf7f2 | 14175 | ceph_assert(have & POOL_CHECKED); |
7c673cae FG |
14176 | break; |
14177 | } | |
14178 | } | |
14179 | ||
14180 | if (!have) { | |
14181 | if (in->snapid != CEPH_NOSNAP) { | |
14182 | // pool permission check needs to write to the first object. But for snapshot, | |
14183 | // head of the first object may have alread been deleted. To avoid creating | |
14184 | // orphan object, skip the check for now. | |
14185 | return 0; | |
14186 | } | |
14187 | ||
14188 | pool_perms[perm_key] = POOL_CHECKING; | |
14189 | ||
14190 | char oid_buf[32]; | |
14191 | snprintf(oid_buf, sizeof(oid_buf), "%llx.00000000", (unsigned long long)in->ino); | |
14192 | object_t oid = oid_buf; | |
14193 | ||
14194 | SnapContext nullsnapc; | |
14195 | ||
14196 | C_SaferCond rd_cond; | |
14197 | ObjectOperation rd_op; | |
14198 | rd_op.stat(NULL, (ceph::real_time*)nullptr, NULL); | |
14199 | ||
14200 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), rd_op, | |
14201 | nullsnapc, ceph::real_clock::now(), 0, &rd_cond); | |
14202 | ||
14203 | C_SaferCond wr_cond; | |
14204 | ObjectOperation wr_op; | |
14205 | wr_op.create(true); | |
14206 | ||
14207 | objecter->mutate(oid, OSDMap::file_to_object_locator(in->layout), wr_op, | |
14208 | nullsnapc, ceph::real_clock::now(), 0, &wr_cond); | |
14209 | ||
9f95a23c | 14210 | client_lock.unlock(); |
7c673cae FG |
14211 | int rd_ret = rd_cond.wait(); |
14212 | int wr_ret = wr_cond.wait(); | |
9f95a23c | 14213 | client_lock.lock(); |
7c673cae FG |
14214 | |
14215 | bool errored = false; | |
14216 | ||
14217 | if (rd_ret == 0 || rd_ret == -ENOENT) | |
14218 | have |= POOL_READ; | |
14219 | else if (rd_ret != -EPERM) { | |
11fdf7f2 | 14220 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
14221 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
14222 | errored = true; | |
14223 | } | |
14224 | ||
14225 | if (wr_ret == 0 || wr_ret == -EEXIST) | |
14226 | have |= POOL_WRITE; | |
14227 | else if (wr_ret != -EPERM) { | |
11fdf7f2 | 14228 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
14229 | << " rd_err = " << rd_ret << " wr_err = " << wr_ret << dendl; |
14230 | errored = true; | |
14231 | } | |
14232 | ||
14233 | if (errored) { | |
14234 | // Indeterminate: erase CHECKING state so that subsequent calls re-check. | |
14235 | // Raise EIO because actual error code might be misleading for | |
14236 | // userspace filesystem user. | |
14237 | pool_perms.erase(perm_key); | |
14238 | signal_cond_list(waiting_for_pool_perm); | |
14239 | return -EIO; | |
14240 | } | |
14241 | ||
14242 | pool_perms[perm_key] = have | POOL_CHECKED; | |
14243 | signal_cond_list(waiting_for_pool_perm); | |
14244 | } | |
14245 | ||
14246 | if ((need & CEPH_CAP_FILE_RD) && !(have & POOL_READ)) { | |
11fdf7f2 | 14247 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
14248 | << " need " << ccap_string(need) << ", but no read perm" << dendl; |
14249 | return -EPERM; | |
14250 | } | |
14251 | if ((need & CEPH_CAP_FILE_WR) && !(have & POOL_WRITE)) { | |
11fdf7f2 | 14252 | ldout(cct, 10) << __func__ << " on pool " << pool_id << " ns " << pool_ns |
7c673cae FG |
14253 | << " need " << ccap_string(need) << ", but no write perm" << dendl; |
14254 | return -EPERM; | |
14255 | } | |
14256 | ||
14257 | return 0; | |
14258 | } | |
14259 | ||
14260 | int Client::_posix_acl_permission(Inode *in, const UserPerm& perms, unsigned want) | |
14261 | { | |
14262 | if (acl_type == POSIX_ACL) { | |
14263 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
14264 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
14265 | ||
14266 | return posix_acl_permits(access_acl, in->uid, in->gid, perms, want); | |
14267 | } | |
14268 | } | |
14269 | return -EAGAIN; | |
14270 | } | |
14271 | ||
14272 | int Client::_posix_acl_chmod(Inode *in, mode_t mode, const UserPerm& perms) | |
14273 | { | |
14274 | if (acl_type == NO_ACL) | |
14275 | return 0; | |
14276 | ||
14277 | int r = _getattr(in, CEPH_STAT_CAP_XATTR, perms, in->xattr_version == 0); | |
14278 | if (r < 0) | |
14279 | goto out; | |
14280 | ||
14281 | if (acl_type == POSIX_ACL) { | |
14282 | if (in->xattrs.count(ACL_EA_ACCESS)) { | |
14283 | const bufferptr& access_acl = in->xattrs[ACL_EA_ACCESS]; | |
14284 | bufferptr acl(access_acl.c_str(), access_acl.length()); | |
14285 | r = posix_acl_access_chmod(acl, mode); | |
14286 | if (r < 0) | |
14287 | goto out; | |
14288 | r = _do_setxattr(in, ACL_EA_ACCESS, acl.c_str(), acl.length(), 0, perms); | |
14289 | } else { | |
14290 | r = 0; | |
14291 | } | |
14292 | } | |
14293 | out: | |
14294 | ldout(cct, 10) << __func__ << " ino " << in->ino << " result=" << r << dendl; | |
14295 | return r; | |
14296 | } | |
14297 | ||
14298 | int Client::_posix_acl_create(Inode *dir, mode_t *mode, bufferlist& xattrs_bl, | |
14299 | const UserPerm& perms) | |
14300 | { | |
14301 | if (acl_type == NO_ACL) | |
14302 | return 0; | |
14303 | ||
14304 | if (S_ISLNK(*mode)) | |
14305 | return 0; | |
14306 | ||
14307 | int r = _getattr(dir, CEPH_STAT_CAP_XATTR, perms, dir->xattr_version == 0); | |
14308 | if (r < 0) | |
14309 | goto out; | |
14310 | ||
14311 | if (acl_type == POSIX_ACL) { | |
14312 | if (dir->xattrs.count(ACL_EA_DEFAULT)) { | |
14313 | map<string, bufferptr> xattrs; | |
14314 | ||
14315 | const bufferptr& default_acl = dir->xattrs[ACL_EA_DEFAULT]; | |
14316 | bufferptr acl(default_acl.c_str(), default_acl.length()); | |
14317 | r = posix_acl_inherit_mode(acl, mode); | |
14318 | if (r < 0) | |
14319 | goto out; | |
14320 | ||
14321 | if (r > 0) { | |
14322 | r = posix_acl_equiv_mode(acl.c_str(), acl.length(), mode); | |
14323 | if (r < 0) | |
14324 | goto out; | |
14325 | if (r > 0) | |
14326 | xattrs[ACL_EA_ACCESS] = acl; | |
14327 | } | |
14328 | ||
14329 | if (S_ISDIR(*mode)) | |
14330 | xattrs[ACL_EA_DEFAULT] = dir->xattrs[ACL_EA_DEFAULT]; | |
14331 | ||
14332 | r = xattrs.size(); | |
14333 | if (r > 0) | |
11fdf7f2 | 14334 | encode(xattrs, xattrs_bl); |
7c673cae FG |
14335 | } else { |
14336 | if (umask_cb) | |
14337 | *mode &= ~umask_cb(callback_handle); | |
14338 | r = 0; | |
14339 | } | |
14340 | } | |
14341 | out: | |
14342 | ldout(cct, 10) << __func__ << " dir ino " << dir->ino << " result=" << r << dendl; | |
14343 | return r; | |
14344 | } | |
14345 | ||
14346 | void Client::set_filer_flags(int flags) | |
14347 | { | |
11fdf7f2 TL |
14348 | std::lock_guard l(client_lock); |
14349 | ceph_assert(flags == 0 || | |
7c673cae FG |
14350 | flags == CEPH_OSD_FLAG_LOCALIZE_READS); |
14351 | objecter->add_global_op_flags(flags); | |
14352 | } | |
14353 | ||
14354 | void Client::clear_filer_flags(int flags) | |
14355 | { | |
11fdf7f2 TL |
14356 | std::lock_guard l(client_lock); |
14357 | ceph_assert(flags == CEPH_OSD_FLAG_LOCALIZE_READS); | |
7c673cae FG |
14358 | objecter->clear_global_op_flag(flags); |
14359 | } | |
14360 | ||
11fdf7f2 TL |
14361 | // called before mount |
14362 | void Client::set_uuid(const std::string& uuid) | |
14363 | { | |
14364 | std::lock_guard l(client_lock); | |
14365 | assert(initialized); | |
14366 | assert(!uuid.empty()); | |
14367 | ||
14368 | metadata["uuid"] = uuid; | |
14369 | _close_sessions(); | |
14370 | } | |
14371 | ||
14372 | // called before mount. 0 means infinite | |
14373 | void Client::set_session_timeout(unsigned timeout) | |
14374 | { | |
14375 | std::lock_guard l(client_lock); | |
14376 | assert(initialized); | |
14377 | ||
14378 | metadata["timeout"] = stringify(timeout); | |
14379 | } | |
14380 | ||
14381 | // called before mount | |
14382 | int Client::start_reclaim(const std::string& uuid, unsigned flags, | |
14383 | const std::string& fs_name) | |
14384 | { | |
14385 | std::lock_guard l(client_lock); | |
14386 | if (!initialized) | |
14387 | return -ENOTCONN; | |
14388 | ||
14389 | if (uuid.empty()) | |
14390 | return -EINVAL; | |
14391 | ||
14392 | { | |
14393 | auto it = metadata.find("uuid"); | |
14394 | if (it != metadata.end() && it->second == uuid) | |
14395 | return -EINVAL; | |
14396 | } | |
14397 | ||
14398 | int r = subscribe_mdsmap(fs_name); | |
14399 | if (r < 0) { | |
14400 | lderr(cct) << "mdsmap subscription failed: " << cpp_strerror(r) << dendl; | |
14401 | return r; | |
14402 | } | |
14403 | ||
14404 | if (metadata.empty()) | |
14405 | populate_metadata(""); | |
14406 | ||
14407 | while (mdsmap->get_epoch() == 0) | |
14408 | wait_on_list(waiting_for_mdsmap); | |
14409 | ||
14410 | reclaim_errno = 0; | |
14411 | for (unsigned mds = 0; mds < mdsmap->get_num_in_mds(); ) { | |
14412 | if (!mdsmap->is_up(mds)) { | |
14413 | ldout(cct, 10) << "mds." << mds << " not active, waiting for new mdsmap" << dendl; | |
14414 | wait_on_list(waiting_for_mdsmap); | |
14415 | continue; | |
14416 | } | |
14417 | ||
14418 | MetaSession *session; | |
14419 | if (!have_open_session(mds)) { | |
14420 | session = _get_or_open_mds_session(mds); | |
14421 | if (session->state != MetaSession::STATE_OPENING) { | |
14422 | // umounting? | |
14423 | return -EINVAL; | |
14424 | } | |
14425 | ldout(cct, 10) << "waiting for session to mds." << mds << " to open" << dendl; | |
14426 | wait_on_context_list(session->waiting_for_open); | |
14427 | if (rejected_by_mds.count(mds)) | |
14428 | return -EPERM; | |
14429 | continue; | |
14430 | } | |
14431 | ||
14432 | session = &mds_sessions.at(mds); | |
14433 | if (!session->mds_features.test(CEPHFS_FEATURE_RECLAIM_CLIENT)) | |
14434 | return -EOPNOTSUPP; | |
14435 | ||
14436 | if (session->reclaim_state == MetaSession::RECLAIM_NULL || | |
14437 | session->reclaim_state == MetaSession::RECLAIMING) { | |
14438 | session->reclaim_state = MetaSession::RECLAIMING; | |
9f95a23c | 14439 | auto m = make_message<MClientReclaim>(uuid, flags); |
11fdf7f2 TL |
14440 | session->con->send_message2(std::move(m)); |
14441 | wait_on_list(waiting_for_reclaim); | |
14442 | } else if (session->reclaim_state == MetaSession::RECLAIM_FAIL) { | |
14443 | return reclaim_errno ? : -ENOTRECOVERABLE; | |
14444 | } else { | |
14445 | mds++; | |
14446 | } | |
14447 | } | |
14448 | ||
14449 | // didn't find target session in any mds | |
14450 | if (reclaim_target_addrs.empty()) { | |
14451 | if (flags & CEPH_RECLAIM_RESET) | |
14452 | return -ENOENT; | |
14453 | return -ENOTRECOVERABLE; | |
14454 | } | |
14455 | ||
14456 | if (flags & CEPH_RECLAIM_RESET) | |
14457 | return 0; | |
14458 | ||
14459 | // use blacklist to check if target session was killed | |
14460 | // (config option mds_session_blacklist_on_evict needs to be true) | |
14461 | C_SaferCond cond; | |
14462 | if (!objecter->wait_for_map(reclaim_osd_epoch, &cond)) { | |
14463 | ldout(cct, 10) << __func__ << ": waiting for OSD epoch " << reclaim_osd_epoch << dendl; | |
9f95a23c | 14464 | client_lock.unlock(); |
11fdf7f2 | 14465 | cond.wait(); |
9f95a23c | 14466 | client_lock.lock(); |
11fdf7f2 TL |
14467 | } |
14468 | ||
14469 | bool blacklisted = objecter->with_osdmap( | |
14470 | [this](const OSDMap &osd_map) -> bool { | |
14471 | return osd_map.is_blacklisted(reclaim_target_addrs); | |
14472 | }); | |
14473 | if (blacklisted) | |
14474 | return -ENOTRECOVERABLE; | |
14475 | ||
14476 | metadata["reclaiming_uuid"] = uuid; | |
14477 | return 0; | |
14478 | } | |
14479 | ||
14480 | void Client::finish_reclaim() | |
14481 | { | |
14482 | auto it = metadata.find("reclaiming_uuid"); | |
14483 | if (it == metadata.end()) { | |
14484 | for (auto &p : mds_sessions) | |
14485 | p.second.reclaim_state = MetaSession::RECLAIM_NULL; | |
14486 | return; | |
14487 | } | |
14488 | ||
14489 | for (auto &p : mds_sessions) { | |
14490 | p.second.reclaim_state = MetaSession::RECLAIM_NULL; | |
9f95a23c | 14491 | auto m = make_message<MClientReclaim>("", MClientReclaim::FLAG_FINISH); |
11fdf7f2 TL |
14492 | p.second.con->send_message2(std::move(m)); |
14493 | } | |
14494 | ||
14495 | metadata["uuid"] = it->second; | |
14496 | metadata.erase(it); | |
14497 | } | |
14498 | ||
14499 | void Client::handle_client_reclaim_reply(const MConstRef<MClientReclaimReply>& reply) | |
14500 | { | |
14501 | mds_rank_t from = mds_rank_t(reply->get_source().num()); | |
14502 | ldout(cct, 10) << __func__ << " " << *reply << " from mds." << from << dendl; | |
14503 | ||
14504 | MetaSession *session = _get_mds_session(from, reply->get_connection().get()); | |
14505 | if (!session) { | |
14506 | ldout(cct, 10) << " discarding reclaim reply from sessionless mds." << from << dendl; | |
14507 | return; | |
14508 | } | |
14509 | ||
14510 | if (reply->get_result() >= 0) { | |
14511 | session->reclaim_state = MetaSession::RECLAIM_OK; | |
14512 | if (reply->get_epoch() > reclaim_osd_epoch) | |
14513 | reclaim_osd_epoch = reply->get_epoch(); | |
14514 | if (!reply->get_addrs().empty()) | |
14515 | reclaim_target_addrs = reply->get_addrs(); | |
14516 | } else { | |
14517 | session->reclaim_state = MetaSession::RECLAIM_FAIL; | |
14518 | reclaim_errno = reply->get_result(); | |
14519 | } | |
14520 | ||
14521 | signal_cond_list(waiting_for_reclaim); | |
14522 | } | |
14523 | ||
7c673cae FG |
14524 | /** |
14525 | * This is included in cap release messages, to cause | |
14526 | * the MDS to wait until this OSD map epoch. It is necessary | |
14527 | * in corner cases where we cancel RADOS ops, so that | |
14528 | * nobody else tries to do IO to the same objects in | |
14529 | * the same epoch as the cancelled ops. | |
14530 | */ | |
14531 | void Client::set_cap_epoch_barrier(epoch_t e) | |
14532 | { | |
14533 | ldout(cct, 5) << __func__ << " epoch = " << e << dendl; | |
14534 | cap_epoch_barrier = e; | |
14535 | } | |
14536 | ||
14537 | const char** Client::get_tracked_conf_keys() const | |
14538 | { | |
14539 | static const char* keys[] = { | |
14540 | "client_cache_size", | |
14541 | "client_cache_mid", | |
14542 | "client_acl_type", | |
b32b8144 FG |
14543 | "client_deleg_timeout", |
14544 | "client_deleg_break_on_open", | |
7c673cae FG |
14545 | NULL |
14546 | }; | |
14547 | return keys; | |
14548 | } | |
14549 | ||
11fdf7f2 | 14550 | void Client::handle_conf_change(const ConfigProxy& conf, |
7c673cae FG |
14551 | const std::set <std::string> &changed) |
14552 | { | |
11fdf7f2 | 14553 | std::lock_guard lock(client_lock); |
7c673cae | 14554 | |
181888fb | 14555 | if (changed.count("client_cache_mid")) { |
7c673cae FG |
14556 | lru.lru_set_midpoint(cct->_conf->client_cache_mid); |
14557 | } | |
14558 | if (changed.count("client_acl_type")) { | |
14559 | acl_type = NO_ACL; | |
14560 | if (cct->_conf->client_acl_type == "posix_acl") | |
14561 | acl_type = POSIX_ACL; | |
14562 | } | |
14563 | } | |
14564 | ||
7c673cae FG |
14565 | void intrusive_ptr_add_ref(Inode *in) |
14566 | { | |
14567 | in->get(); | |
14568 | } | |
14569 | ||
14570 | void intrusive_ptr_release(Inode *in) | |
14571 | { | |
14572 | in->client->put_inode(in); | |
14573 | } | |
14574 | ||
14575 | mds_rank_t Client::_get_random_up_mds() const | |
14576 | { | |
9f95a23c | 14577 | ceph_assert(ceph_mutex_is_locked_by_me(client_lock)); |
7c673cae FG |
14578 | |
14579 | std::set<mds_rank_t> up; | |
14580 | mdsmap->get_up_mds_set(up); | |
14581 | ||
14582 | if (up.empty()) | |
14583 | return MDS_RANK_NONE; | |
14584 | std::set<mds_rank_t>::const_iterator p = up.begin(); | |
14585 | for (int n = rand() % up.size(); n; n--) | |
14586 | ++p; | |
14587 | return *p; | |
14588 | } | |
14589 | ||
14590 | ||
14591 | StandaloneClient::StandaloneClient(Messenger *m, MonClient *mc) | |
14592 | : Client(m, mc, new Objecter(m->cct, m, mc, NULL, 0, 0)) | |
14593 | { | |
14594 | monclient->set_messenger(m); | |
14595 | objecter->set_client_incarnation(0); | |
14596 | } | |
14597 | ||
14598 | StandaloneClient::~StandaloneClient() | |
14599 | { | |
14600 | delete objecter; | |
14601 | objecter = nullptr; | |
14602 | } | |
14603 | ||
14604 | int StandaloneClient::init() | |
14605 | { | |
14606 | timer.init(); | |
14607 | objectcacher->start(); | |
14608 | objecter->init(); | |
14609 | ||
9f95a23c | 14610 | client_lock.lock(); |
11fdf7f2 | 14611 | ceph_assert(!is_initialized()); |
7c673cae FG |
14612 | |
14613 | messenger->add_dispatcher_tail(objecter); | |
14614 | messenger->add_dispatcher_tail(this); | |
14615 | ||
14616 | monclient->set_want_keys(CEPH_ENTITY_TYPE_MDS | CEPH_ENTITY_TYPE_OSD); | |
14617 | int r = monclient->init(); | |
14618 | if (r < 0) { | |
14619 | // need to do cleanup because we're in an intermediate init state | |
14620 | timer.shutdown(); | |
9f95a23c | 14621 | client_lock.unlock(); |
7c673cae FG |
14622 | objecter->shutdown(); |
14623 | objectcacher->stop(); | |
14624 | monclient->shutdown(); | |
14625 | return r; | |
14626 | } | |
14627 | objecter->start(); | |
14628 | ||
9f95a23c | 14629 | client_lock.unlock(); |
7c673cae FG |
14630 | _finish_init(); |
14631 | ||
14632 | return 0; | |
14633 | } | |
14634 | ||
14635 | void StandaloneClient::shutdown() | |
14636 | { | |
14637 | Client::shutdown(); | |
14638 | objecter->shutdown(); | |
14639 | monclient->shutdown(); | |
14640 | } |